2018-10-05 16:40:00 -07:00
|
|
|
|
2018-01-30 21:55:03 +01:00
|
|
|
|
bpf tools: Introduce 'bpf' library and add bpf feature check
This is the first patch of libbpf. The goal of libbpf is to create a
standard way for accessing eBPF object files. This patch creates
'Makefile' and 'Build' for it, allows 'make' to build libbpf.a and
libbpf.so, 'make install' to put them into proper directories.
Most part of Makefile is borrowed from traceevent.
Before building, it checks the existence of libelf in Makefile, and deny
to build if not found. Instead of throwing an error if libelf not found,
the error raises in a phony target "elfdep". This design is to ensure
'make clean' still workable even if libelf is not found.
Because libbpf requires 'kern_version' field set for 'union bpf_attr'
(bpfdep" is used for that dependency), Kernel BPF API is also checked
by intruducing a new feature check 'bpf' into tools/build/feature,
which checks the existence and version of linux/bpf.h. When building
libbpf, it searches that file from include/uapi/linux in kernel source
tree (controlled by FEATURE_CHECK_CFLAGS-bpf). Since it searches kernel
source tree it reside, installing of newest kernel headers is not
required, except we are trying to port these files to an old kernel.
To avoid checking that file when perf building, the newly introduced
'bpf' feature check doesn't added into FEATURE_TESTS and
FEATURE_DISPLAY by default in tools/build/Makefile.feature, but added
into libbpf's specific.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kaixu Xia <xiakaixu@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Zefan Li <lizefan@huawei.com>
Bcc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1435716878-189507-4-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-07-01 02:13:51 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
tools lib bpf: Add BPF program pinning APIs
Add new APIs to pin a BPF program (or specific instances) to the
filesystem. The user can specify the path full path within a BPF
filesystem to pin the program.
bpf_program__pin_instance(prog, path, n) will pin the nth instance of
'prog' to the specified path.
bpf_program__pin(prog, path) will create the directory 'path' (if it
does not exist) and pin each instance within that directory. For
instance, path/0, path/1, path/2.
Committer notes:
- Add missing headers for mkdir()
- Check strdup() for failure
- Check snprintf >= size, not >, as == also means truncated, see 'man
snprintf', return value.
- Conditionally define BPF_FS_MAGIC, as it isn't in magic.h in older
systems and we're not yet having a tools/include/uapi/linux/magic.h
copy.
- Do not include linux/magic.h, not present in older distros.
Signed-off-by: Joe Stringer <joe@ovn.org>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: netdev@vger.kernel.org
Link: http://lkml.kernel.org/r/20170126212001.14103-2-joe@ovn.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-01-26 13:19:56 -08:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
bpf tools: Introduce 'bpf' library and add bpf feature check
This is the first patch of libbpf. The goal of libbpf is to create a
standard way for accessing eBPF object files. This patch creates
'Makefile' and 'Build' for it, allows 'make' to build libbpf.a and
libbpf.so, 'make install' to put them into proper directories.
Most part of Makefile is borrowed from traceevent.
Before building, it checks the existence of libelf in Makefile, and deny
to build if not found. Instead of throwing an error if libelf not found,
the error raises in a phony target "elfdep". This design is to ensure
'make clean' still workable even if libelf is not found.
Because libbpf requires 'kern_version' field set for 'union bpf_attr'
(bpfdep" is used for that dependency), Kernel BPF API is also checked
by intruducing a new feature check 'bpf' into tools/build/feature,
which checks the existence and version of linux/bpf.h. When building
libbpf, it searches that file from include/uapi/linux in kernel source
tree (controlled by FEATURE_CHECK_CFLAGS-bpf). Since it searches kernel
source tree it reside, installing of newest kernel headers is not
required, except we are trying to port these files to an old kernel.
To avoid checking that file when perf building, the newly introduced
'bpf' feature check doesn't added into FEATURE_TESTS and
FEATURE_DISPLAY by default in tools/build/Makefile.feature, but added
into libbpf's specific.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kaixu Xia <xiakaixu@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Zefan Li <lizefan@huawei.com>
Bcc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1435716878-189507-4-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-07-01 02:13:51 +00:00
|
|
|
|
|
|
|
|
|
2018-11-29 15:31:45 -08:00
|
|
|
|
2018-07-10 14:43:05 -07:00
|
|
|
|
2018-11-29 15:31:45 -08:00
|
|
|
|
bpf tools: Introduce 'bpf' library and add bpf feature check
This is the first patch of libbpf. The goal of libbpf is to create a
standard way for accessing eBPF object files. This patch creates
'Makefile' and 'Build' for it, allows 'make' to build libbpf.a and
libbpf.so, 'make install' to put them into proper directories.
Most part of Makefile is borrowed from traceevent.
Before building, it checks the existence of libelf in Makefile, and deny
to build if not found. Instead of throwing an error if libelf not found,
the error raises in a phony target "elfdep". This design is to ensure
'make clean' still workable even if libelf is not found.
Because libbpf requires 'kern_version' field set for 'union bpf_attr'
(bpfdep" is used for that dependency), Kernel BPF API is also checked
by intruducing a new feature check 'bpf' into tools/build/feature,
which checks the existence and version of linux/bpf.h. When building
libbpf, it searches that file from include/uapi/linux in kernel source
tree (controlled by FEATURE_CHECK_CFLAGS-bpf). Since it searches kernel
source tree it reside, installing of newest kernel headers is not
required, except we are trying to port these files to an old kernel.
To avoid checking that file when perf building, the newly introduced
'bpf' feature check doesn't added into FEATURE_TESTS and
FEATURE_DISPLAY by default in tools/build/Makefile.feature, but added
into libbpf's specific.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kaixu Xia <xiakaixu@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Zefan Li <lizefan@huawei.com>
Bcc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1435716878-189507-4-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-07-01 02:13:51 +00:00
|
|
|
|
2015-07-01 02:13:52 +00:00
|
|
|
|
|
|
|
|
|
tools lib bpf: Add BPF program pinning APIs
Add new APIs to pin a BPF program (or specific instances) to the
filesystem. The user can specify the path full path within a BPF
filesystem to pin the program.
bpf_program__pin_instance(prog, path, n) will pin the nth instance of
'prog' to the specified path.
bpf_program__pin(prog, path) will create the directory 'path' (if it
does not exist) and pin each instance within that directory. For
instance, path/0, path/1, path/2.
Committer notes:
- Add missing headers for mkdir()
- Check strdup() for failure
- Check snprintf >= size, not >, as == also means truncated, see 'man
snprintf', return value.
- Conditionally define BPF_FS_MAGIC, as it isn't in magic.h in older
systems and we're not yet having a tools/include/uapi/linux/magic.h
copy.
- Do not include linux/magic.h, not present in older distros.
Signed-off-by: Joe Stringer <joe@ovn.org>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: netdev@vger.kernel.org
Link: http://lkml.kernel.org/r/20170126212001.14103-2-joe@ovn.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-01-26 13:19:56 -08:00
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
2015-07-01 02:13:52 +00:00
|
|
|
|
bpf tools: Introduce 'bpf' library and add bpf feature check
This is the first patch of libbpf. The goal of libbpf is to create a
standard way for accessing eBPF object files. This patch creates
'Makefile' and 'Build' for it, allows 'make' to build libbpf.a and
libbpf.so, 'make install' to put them into proper directories.
Most part of Makefile is borrowed from traceevent.
Before building, it checks the existence of libelf in Makefile, and deny
to build if not found. Instead of throwing an error if libelf not found,
the error raises in a phony target "elfdep". This design is to ensure
'make clean' still workable even if libelf is not found.
Because libbpf requires 'kern_version' field set for 'union bpf_attr'
(bpfdep" is used for that dependency), Kernel BPF API is also checked
by intruducing a new feature check 'bpf' into tools/build/feature,
which checks the existence and version of linux/bpf.h. When building
libbpf, it searches that file from include/uapi/linux in kernel source
tree (controlled by FEATURE_CHECK_CFLAGS-bpf). Since it searches kernel
source tree it reside, installing of newest kernel headers is not
required, except we are trying to port these files to an old kernel.
To avoid checking that file when perf building, the newly introduced
'bpf' feature check doesn't added into FEATURE_TESTS and
FEATURE_DISPLAY by default in tools/build/Makefile.feature, but added
into libbpf's specific.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kaixu Xia <xiakaixu@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Zefan Li <lizefan@huawei.com>
Bcc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1435716878-189507-4-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-07-01 02:13:51 +00:00
|
|
|
|
2019-07-19 11:34:06 -03:00
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
bpf tools: Introduce 'bpf' library and add bpf feature check
This is the first patch of libbpf. The goal of libbpf is to create a
standard way for accessing eBPF object files. This patch creates
'Makefile' and 'Build' for it, allows 'make' to build libbpf.a and
libbpf.so, 'make install' to put them into proper directories.
Most part of Makefile is borrowed from traceevent.
Before building, it checks the existence of libelf in Makefile, and deny
to build if not found. Instead of throwing an error if libelf not found,
the error raises in a phony target "elfdep". This design is to ensure
'make clean' still workable even if libelf is not found.
Because libbpf requires 'kern_version' field set for 'union bpf_attr'
(bpfdep" is used for that dependency), Kernel BPF API is also checked
by intruducing a new feature check 'bpf' into tools/build/feature,
which checks the existence and version of linux/bpf.h. When building
libbpf, it searches that file from include/uapi/linux in kernel source
tree (controlled by FEATURE_CHECK_CFLAGS-bpf). Since it searches kernel
source tree it reside, installing of newest kernel headers is not
required, except we are trying to port these files to an old kernel.
To avoid checking that file when perf building, the newly introduced
'bpf' feature check doesn't added into FEATURE_TESTS and
FEATURE_DISPLAY by default in tools/build/Makefile.feature, but added
into libbpf's specific.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kaixu Xia <xiakaixu@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Zefan Li <lizefan@huawei.com>
Bcc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1435716878-189507-4-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-07-01 02:13:51 +00:00
|
|
|
|
2017-01-22 17:11:25 -08:00
|
|
|
|
2015-07-01 02:13:57 +00:00
|
|
|
|
bpf tools: Introduce 'bpf' library and add bpf feature check
This is the first patch of libbpf. The goal of libbpf is to create a
standard way for accessing eBPF object files. This patch creates
'Makefile' and 'Build' for it, allows 'make' to build libbpf.a and
libbpf.so, 'make install' to put them into proper directories.
Most part of Makefile is borrowed from traceevent.
Before building, it checks the existence of libelf in Makefile, and deny
to build if not found. Instead of throwing an error if libelf not found,
the error raises in a phony target "elfdep". This design is to ensure
'make clean' still workable even if libelf is not found.
Because libbpf requires 'kern_version' field set for 'union bpf_attr'
(bpfdep" is used for that dependency), Kernel BPF API is also checked
by intruducing a new feature check 'bpf' into tools/build/feature,
which checks the existence and version of linux/bpf.h. When building
libbpf, it searches that file from include/uapi/linux in kernel source
tree (controlled by FEATURE_CHECK_CFLAGS-bpf). Since it searches kernel
source tree it reside, installing of newest kernel headers is not
required, except we are trying to port these files to an old kernel.
To avoid checking that file when perf building, the newly introduced
'bpf' feature check doesn't added into FEATURE_TESTS and
FEATURE_DISPLAY by default in tools/build/Makefile.feature, but added
into libbpf's specific.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kaixu Xia <xiakaixu@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Zefan Li <lizefan@huawei.com>
Bcc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1435716878-189507-4-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-07-01 02:13:51 +00:00
|
|
|
|
2018-07-24 08:40:22 -07:00
|
|
|
|
2018-11-20 17:11:19 -08:00
|
|
|
|
2015-07-01 02:14:10 +00:00
|
|
|
|
tools lib bpf: Add BPF program pinning APIs
Add new APIs to pin a BPF program (or specific instances) to the
filesystem. The user can specify the path full path within a BPF
filesystem to pin the program.
bpf_program__pin_instance(prog, path, n) will pin the nth instance of
'prog' to the specified path.
bpf_program__pin(prog, path) will create the directory 'path' (if it
does not exist) and pin each instance within that directory. For
instance, path/0, path/1, path/2.
Committer notes:
- Add missing headers for mkdir()
- Check strdup() for failure
- Check snprintf >= size, not >, as == also means truncated, see 'man
snprintf', return value.
- Conditionally define BPF_FS_MAGIC, as it isn't in magic.h in older
systems and we're not yet having a tools/include/uapi/linux/magic.h
copy.
- Do not include linux/magic.h, not present in older distros.
Signed-off-by: Joe Stringer <joe@ovn.org>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: netdev@vger.kernel.org
Link: http://lkml.kernel.org/r/20170126212001.14103-2-joe@ovn.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-01-26 13:19:56 -08:00
|
|
|
|
2018-10-09 16:14:47 -07:00
|
|
|
|
2018-10-19 15:51:03 +02:00
|
|
|
|
2019-07-06 11:06:24 -07:00
|
|
|
|
2019-07-01 16:58:57 -07:00
|
|
|
|
2019-07-06 11:06:24 -07:00
|
|
|
|
tools lib bpf: Add BPF program pinning APIs
Add new APIs to pin a BPF program (or specific instances) to the
filesystem. The user can specify the path full path within a BPF
filesystem to pin the program.
bpf_program__pin_instance(prog, path, n) will pin the nth instance of
'prog' to the specified path.
bpf_program__pin(prog, path) will create the directory 'path' (if it
does not exist) and pin each instance within that directory. For
instance, path/0, path/1, path/2.
Committer notes:
- Add missing headers for mkdir()
- Check strdup() for failure
- Check snprintf >= size, not >, as == also means truncated, see 'man
snprintf', return value.
- Conditionally define BPF_FS_MAGIC, as it isn't in magic.h in older
systems and we're not yet having a tools/include/uapi/linux/magic.h
copy.
- Do not include linux/magic.h, not present in older distros.
Signed-off-by: Joe Stringer <joe@ovn.org>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: netdev@vger.kernel.org
Link: http://lkml.kernel.org/r/20170126212001.14103-2-joe@ovn.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-01-26 13:19:56 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-07-10 14:43:05 -07:00
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
bpf tools: Introduce 'bpf' library and add bpf feature check
This is the first patch of libbpf. The goal of libbpf is to create a
standard way for accessing eBPF object files. This patch creates
'Makefile' and 'Build' for it, allows 'make' to build libbpf.a and
libbpf.so, 'make install' to put them into proper directories.
Most part of Makefile is borrowed from traceevent.
Before building, it checks the existence of libelf in Makefile, and deny
to build if not found. Instead of throwing an error if libelf not found,
the error raises in a phony target "elfdep". This design is to ensure
'make clean' still workable even if libelf is not found.
Because libbpf requires 'kern_version' field set for 'union bpf_attr'
(bpfdep" is used for that dependency), Kernel BPF API is also checked
by intruducing a new feature check 'bpf' into tools/build/feature,
which checks the existence and version of linux/bpf.h. When building
libbpf, it searches that file from include/uapi/linux in kernel source
tree (controlled by FEATURE_CHECK_CFLAGS-bpf). Since it searches kernel
source tree it reside, installing of newest kernel headers is not
required, except we are trying to port these files to an old kernel.
To avoid checking that file when perf building, the newly introduced
'bpf' feature check doesn't added into FEATURE_TESTS and
FEATURE_DISPLAY by default in tools/build/Makefile.feature, but added
into libbpf's specific.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kaixu Xia <xiakaixu@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Zefan Li <lizefan@huawei.com>
Bcc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1435716878-189507-4-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-07-01 02:13:51 +00:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:04 +00:00
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
2018-09-14 16:47:14 -03:00
|
|
|
|
libbpf: detect supported kernel BTF features and sanitize BTF
Depending on used versions of libbpf, Clang, and kernel, it's possible to
have valid BPF object files with valid BTF information, that still won't
load successfully due to Clang emitting newer BTF features (e.g.,
BTF_KIND_FUNC, .BTF.ext's line_info/func_info, BTF_KIND_DATASEC, etc), that
are not yet supported by older kernel.
This patch adds detection of BTF features and sanitizes BPF object's BTF
by substituting various supported BTF kinds, which have compatible layout:
- BTF_KIND_FUNC -> BTF_KIND_TYPEDEF
- BTF_KIND_FUNC_PROTO -> BTF_KIND_ENUM
- BTF_KIND_VAR -> BTF_KIND_INT
- BTF_KIND_DATASEC -> BTF_KIND_STRUCT
Replacement is done in such a way as to preserve as much information as
possible (names, sizes, etc) where possible without violating kernel's
validation rules.
v2->v3:
- remove duplicate #defines from libbpf_util.h
v1->v2:
- add internal libbpf_internal.h w/ common stuff
- switch SK storage BTF to use new libbpf__probe_raw_btf()
Reported-by: Alexei Starovoitov <ast@fb.com>
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-05-10 14:13:15 -07:00
|
|
|
|
2015-07-01 02:13:52 +00:00
|
|
|
|
2016-07-18 06:01:08 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
tools lib bpf: Add BPF program pinning APIs
Add new APIs to pin a BPF program (or specific instances) to the
filesystem. The user can specify the path full path within a BPF
filesystem to pin the program.
bpf_program__pin_instance(prog, path, n) will pin the nth instance of
'prog' to the specified path.
bpf_program__pin(prog, path) will create the directory 'path' (if it
does not exist) and pin each instance within that directory. For
instance, path/0, path/1, path/2.
Committer notes:
- Add missing headers for mkdir()
- Check strdup() for failure
- Check snprintf >= size, not >, as == also means truncated, see 'man
snprintf', return value.
- Conditionally define BPF_FS_MAGIC, as it isn't in magic.h in older
systems and we're not yet having a tools/include/uapi/linux/magic.h
copy.
- Do not include linux/magic.h, not present in older distros.
Signed-off-by: Joe Stringer <joe@ovn.org>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: netdev@vger.kernel.org
Link: http://lkml.kernel.org/r/20170126212001.14103-2-joe@ovn.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-01-26 13:19:56 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-04-06 22:37:34 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:52 +00:00
|
|
|
|
|
|
|
|
|
2019-02-04 16:20:55 -08:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:52 +00:00
|
|
|
|
2019-02-01 16:14:17 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-02-04 16:20:55 -08:00
|
|
|
|
2015-07-01 02:13:52 +00:00
|
|
|
|
|
|
|
|
|
2019-02-04 16:20:55 -08:00
|
|
|
|
2015-07-01 02:13:52 +00:00
|
|
|
|
2019-02-01 16:14:17 -08:00
|
|
|
|
2015-07-01 02:13:52 +00:00
|
|
|
|
2019-02-01 16:14:17 -08:00
|
|
|
|
2015-07-01 02:13:52 +00:00
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
2019-02-01 16:14:14 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-02-01 16:14:17 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-02-01 16:14:14 -08:00
|
|
|
|
2019-02-01 16:14:17 -08:00
|
|
|
|
2019-02-01 16:14:14 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-03-11 22:30:38 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-20 17:11:19 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-04-24 00:45:56 +02:00
|
|
|
|
|
|
|
|
|
libbpf: detect supported kernel BTF features and sanitize BTF
Depending on used versions of libbpf, Clang, and kernel, it's possible to
have valid BPF object files with valid BTF information, that still won't
load successfully due to Clang emitting newer BTF features (e.g.,
BTF_KIND_FUNC, .BTF.ext's line_info/func_info, BTF_KIND_DATASEC, etc), that
are not yet supported by older kernel.
This patch adds detection of BTF features and sanitizes BPF object's BTF
by substituting various supported BTF kinds, which have compatible layout:
- BTF_KIND_FUNC -> BTF_KIND_TYPEDEF
- BTF_KIND_FUNC_PROTO -> BTF_KIND_ENUM
- BTF_KIND_VAR -> BTF_KIND_INT
- BTF_KIND_DATASEC -> BTF_KIND_STRUCT
Replacement is done in such a way as to preserve as much information as
possible (names, sizes, etc) where possible without violating kernel's
validation rules.
v2->v3:
- remove duplicate #defines from libbpf_util.h
v1->v2:
- add internal libbpf_internal.h w/ common stuff
- switch SK storage BTF to use new libbpf__probe_raw_btf()
Reported-by: Alexei Starovoitov <ast@fb.com>
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-05-10 14:13:15 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-20 17:11:19 -08:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-09-27 14:37:54 -07:00
|
|
|
|
2018-05-16 14:02:49 -07:00
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
2018-11-09 08:21:43 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
2016-07-13 10:44:01 +00:00
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
|
|
|
|
|
2019-04-01 21:27:47 -07:00
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
2015-11-16 12:10:09 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:08 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-30 15:08:01 -07:00
|
|
|
|
|
|
|
|
|
2018-11-19 15:29:16 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-12-07 16:42:29 -08:00
|
|
|
|
2018-11-20 17:11:19 -08:00
|
|
|
|
|
|
|
|
|
2018-12-07 16:42:31 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-24 23:25:19 +01:00
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
|
|
|
|
|
2015-11-27 08:47:36 +00:00
|
|
|
|
2019-06-17 12:26:54 -07:00
|
|
|
|
|
|
|
|
|
2018-05-16 14:02:49 -07:00
|
|
|
|
2018-11-20 20:55:56 -08:00
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
2018-07-24 08:40:21 -07:00
|
|
|
|
|
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:10 +00:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
2015-07-01 02:13:57 +00:00
|
|
|
|
2018-10-09 16:14:47 -07:00
|
|
|
|
2015-07-01 02:13:58 +00:00
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
|
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
2015-07-01 02:14:04 +00:00
|
|
|
|
2018-06-28 14:41:38 -07:00
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:54 +00:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
bpf tools: Collect symbol table from SHT_SYMTAB section
This patch collects symbols section. This section is useful when linking
BPF maps.
What 'bpf_map_xxx()' functions actually require are map's file
descriptors (and the internal verifier converts fds into pointers to
'struct bpf_map'), which we don't know when compiling. Therefore, we
should make compiler generate a 'ldr_64 r1, <imm>' instruction, and
fill the 'imm' field with the actual file descriptor when loading in
libbpf.
BPF programs should be written in this way:
struct bpf_map_def SEC("maps") my_map = {
.type = BPF_MAP_TYPE_HASH,
.key_size = sizeof(unsigned long),
.value_size = sizeof(unsigned long),
.max_entries = 1000000,
};
SEC("my_func=sys_write")
int my_func(void *ctx)
{
...
bpf_map_update_elem(&my_map, &key, &value, BPF_ANY);
...
}
Compiler should convert '&my_map' into a 'ldr_64, r1, <imm>'
instruction, where imm should be the address of 'my_map'. According to
the address, libbpf knows which map it actually referenced, and then
fills the imm field with the 'fd' of that map created by it.
However, since we never really 'link' the object file, the imm field is
only a record in relocation section. Therefore libbpf should do the
relocation:
1. In relocation section (type == SHT_REL), positions of each such
'ldr_64' instruction are recorded with a reference of an entry in
symbol table (SHT_SYMTAB);
2. From records in symbol table we can find the indics of map
variables.
Libbpf first record SHT_SYMTAB and positions of each instruction which
required bu such operation. Then create file descriptor. Finally, after
map creation complete, replace the imm field.
This is the first patch of BPF map related stuff. It records SHT_SYMTAB
into object's efile field for further use.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kaixu Xia <xiakaixu@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1435716878-189507-12-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-07-01 02:13:59 +00:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-12-08 02:25:30 +00:00
|
|
|
|
2015-07-01 02:14:01 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
perf bpf: Check relocation target section
Libbpf should check the target section before doing relocation to ensure
the relocation is correct. If not, a bug in LLVM causes an error. See
[1]. Also, if an incorrect BPF script uses both global variable and
map, global variable whould be treated as map and be relocated without
error.
This patch saves the id of the map section into obj->efile and compare
target section of a relocation symbol against it during relocation.
Previous patch introduces a test case about this problem. After this
patch:
# ~/perf test BPF
37: Test BPF filter :
37.1: Test basic BPF filtering : Ok
37.2: Test BPF prologue generation : Ok
37.3: Test BPF relocation checker : Ok
# perf test -v BPF
...
37.3: Test BPF relocation checker :
...
libbpf: loading object '[bpf_relocation_test]' from buffer
libbpf: section .strtab, size 126, link 0, flags 0, type=3
libbpf: section .text, size 0, link 0, flags 6, type=1
libbpf: section .data, size 0, link 0, flags 3, type=1
libbpf: section .bss, size 0, link 0, flags 3, type=8
libbpf: section func=sys_write, size 104, link 0, flags 6, type=1
libbpf: found program func=sys_write
libbpf: section .relfunc=sys_write, size 16, link 10, flags 0, type=9
libbpf: section maps, size 16, link 0, flags 3, type=1
libbpf: maps in [bpf_relocation_test]: 16 bytes
libbpf: section license, size 4, link 0, flags 3, type=1
libbpf: license of [bpf_relocation_test] is GPL
libbpf: section version, size 4, link 0, flags 3, type=1
libbpf: kernel version of [bpf_relocation_test] is 40400
libbpf: section .symtab, size 144, link 1, flags 0, type=2
libbpf: map 0 is "my_table"
libbpf: collecting relocating info for: 'func=sys_write'
libbpf: Program 'func=sys_write' contains non-map related relo data pointing to section 65522
bpf: failed to load buffer
Compile BPF program failed.
test child finished with 0
---- end ----
Test BPF filter subtest 2: Ok
[1] https://llvm.org/bugs/show_bug.cgi?id=26243
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Will Deacon <will.deacon@arm.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1453715801-7732-3-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-01-25 09:55:49 +00:00
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
2015-07-01 02:14:10 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-11-26 07:03:26 +00:00
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
2018-11-19 15:29:16 -08:00
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
2016-11-26 07:03:26 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-20 17:11:19 -08:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-10-02 13:35:39 -07:00
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
2015-11-16 12:10:09 +00:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-16 12:10:09 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-19 15:29:16 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-12-17 16:57:50 +09:00
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:08 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
2017-09-27 14:37:54 -07:00
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
2018-11-09 08:21:43 -08:00
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-09 08:21:43 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
2017-09-27 14:37:54 -07:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
2019-05-29 10:36:03 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-02-13 10:25:53 -08:00
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
2017-09-27 14:37:54 -07:00
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
2018-02-08 12:48:17 +01:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-09 08:21:43 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
|
|
|
|
|
2017-09-27 14:37:54 -07:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
|
|
|
|
|
2019-05-29 10:36:03 -07:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
2015-11-16 12:10:09 +00:00
|
|
|
|
|
|
|
|
|
2018-11-23 12:58:12 -08:00
|
|
|
|
2018-11-19 15:29:16 -08:00
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-09-27 14:37:54 -07:00
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-09-27 14:37:54 -07:00
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-20 17:11:19 -08:00
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-07-10 14:43:05 -07:00
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-09-27 14:37:54 -07:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:08 +00:00
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-09-27 14:37:54 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
2017-09-27 14:37:54 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-12-13 15:18:52 +00:00
|
|
|
|
|
|
|
|
|
2017-09-27 14:37:54 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-06-28 14:41:38 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-09-27 14:37:54 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-06-28 14:41:38 -07:00
|
|
|
|
2017-09-27 14:37:54 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:54 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
2019-05-29 10:36:11 -07:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:54 +00:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
2015-07-01 02:13:54 +00:00
|
|
|
|
2019-05-29 10:36:10 -07:00
|
|
|
|
2015-07-01 02:13:54 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
perf bpf: Check relocation target section
Libbpf should check the target section before doing relocation to ensure
the relocation is correct. If not, a bug in LLVM causes an error. See
[1]. Also, if an incorrect BPF script uses both global variable and
map, global variable whould be treated as map and be relocated without
error.
This patch saves the id of the map section into obj->efile and compare
target section of a relocation symbol against it during relocation.
Previous patch introduces a test case about this problem. After this
patch:
# ~/perf test BPF
37: Test BPF filter :
37.1: Test basic BPF filtering : Ok
37.2: Test BPF prologue generation : Ok
37.3: Test BPF relocation checker : Ok
# perf test -v BPF
...
37.3: Test BPF relocation checker :
...
libbpf: loading object '[bpf_relocation_test]' from buffer
libbpf: section .strtab, size 126, link 0, flags 0, type=3
libbpf: section .text, size 0, link 0, flags 6, type=1
libbpf: section .data, size 0, link 0, flags 3, type=1
libbpf: section .bss, size 0, link 0, flags 3, type=8
libbpf: section func=sys_write, size 104, link 0, flags 6, type=1
libbpf: found program func=sys_write
libbpf: section .relfunc=sys_write, size 16, link 10, flags 0, type=9
libbpf: section maps, size 16, link 0, flags 3, type=1
libbpf: maps in [bpf_relocation_test]: 16 bytes
libbpf: section license, size 4, link 0, flags 3, type=1
libbpf: license of [bpf_relocation_test] is GPL
libbpf: section version, size 4, link 0, flags 3, type=1
libbpf: kernel version of [bpf_relocation_test] is 40400
libbpf: section .symtab, size 144, link 1, flags 0, type=2
libbpf: map 0 is "my_table"
libbpf: collecting relocating info for: 'func=sys_write'
libbpf: Program 'func=sys_write' contains non-map related relo data pointing to section 65522
bpf: failed to load buffer
Compile BPF program failed.
test child finished with 0
---- end ----
Test BPF filter subtest 2: Ok
[1] https://llvm.org/bugs/show_bug.cgi?id=26243
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Will Deacon <will.deacon@arm.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1453715801-7732-3-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-01-25 09:55:49 +00:00
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:54 +00:00
|
|
|
|
2015-07-01 02:14:04 +00:00
|
|
|
|
2015-07-01 02:14:10 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
bpf tools: Collect symbol table from SHT_SYMTAB section
This patch collects symbols section. This section is useful when linking
BPF maps.
What 'bpf_map_xxx()' functions actually require are map's file
descriptors (and the internal verifier converts fds into pointers to
'struct bpf_map'), which we don't know when compiling. Therefore, we
should make compiler generate a 'ldr_64 r1, <imm>' instruction, and
fill the 'imm' field with the actual file descriptor when loading in
libbpf.
BPF programs should be written in this way:
struct bpf_map_def SEC("maps") my_map = {
.type = BPF_MAP_TYPE_HASH,
.key_size = sizeof(unsigned long),
.value_size = sizeof(unsigned long),
.max_entries = 1000000,
};
SEC("my_func=sys_write")
int my_func(void *ctx)
{
...
bpf_map_update_elem(&my_map, &key, &value, BPF_ANY);
...
}
Compiler should convert '&my_map' into a 'ldr_64, r1, <imm>'
instruction, where imm should be the address of 'my_map'. According to
the address, libbpf knows which map it actually referenced, and then
fills the imm field with the 'fd' of that map created by it.
However, since we never really 'link' the object file, the imm field is
only a record in relocation section. Therefore libbpf should do the
relocation:
1. In relocation section (type == SHT_REL), positions of each such
'ldr_64' instruction are recorded with a reference of an entry in
symbol table (SHT_SYMTAB);
2. From records in symbol table we can find the indics of map
variables.
Libbpf first record SHT_SYMTAB and positions of each instruction which
required bu such operation. Then create file descriptor. Finally, after
map creation complete, replace the imm field.
This is the first patch of BPF map related stuff. It records SHT_SYMTAB
into object's efile field for further use.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kaixu Xia <xiakaixu@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1435716878-189507-12-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-07-01 02:13:59 +00:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:01 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
2015-07-01 02:13:54 +00:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:54 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-29 10:36:04 -07:00
|
|
|
|
2018-07-30 10:53:23 +02:00
|
|
|
|
2019-05-29 10:36:04 -07:00
|
|
|
|
|
|
|
|
|
2018-07-30 10:53:23 +02:00
|
|
|
|
2019-05-29 10:36:04 -07:00
|
|
|
|
2015-07-01 02:13:54 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-29 10:36:10 -07:00
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-29 10:36:11 -07:00
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-29 10:36:11 -07:00
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-07-18 06:01:08 +00:00
|
|
|
|
2019-05-29 10:36:10 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-29 10:36:05 -07:00
|
|
|
|
2015-07-01 02:13:55 +00:00
|
|
|
|
2019-07-19 11:34:06 -03:00
|
|
|
|
2019-05-29 10:36:05 -07:00
|
|
|
|
|
|
|
|
|
2019-07-19 11:34:06 -03:00
|
|
|
|
2019-05-29 10:36:05 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2015-07-01 02:13:55 +00:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:57 +00:00
|
|
|
|
2019-05-29 10:36:11 -07:00
|
|
|
|
2015-07-01 02:13:57 +00:00
|
|
|
|
2019-05-29 10:36:11 -07:00
|
|
|
|
2015-07-01 02:13:57 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-29 10:36:11 -07:00
|
|
|
|
2015-07-01 02:13:57 +00:00
|
|
|
|
2018-10-09 16:14:47 -07:00
|
|
|
|
2015-07-01 02:13:57 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2015-07-01 02:13:57 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-29 10:36:11 -07:00
|
|
|
|
2015-07-01 02:13:57 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-11-15 04:05:47 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:54 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-11-15 04:05:47 +00:00
|
|
|
|
|
|
|
|
|
2018-11-20 20:55:56 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-04-09 23:20:14 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-26 13:38:37 +03:00
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
2019-06-17 12:26:54 -07:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:54 -07:00
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:54 -07:00
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-29 10:36:11 -07:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-04-16 11:47:17 -07:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:54 -07:00
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:54 -07:00
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:54 -07:00
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-11-15 04:05:47 +00:00
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-11-15 04:05:47 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:58 +00:00
|
|
|
|
|
|
|
|
|
2016-11-15 04:05:47 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
|
|
|
|
|
2016-11-15 04:05:47 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-10-05 10:41:57 -04:00
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
|
|
|
|
|
2019-04-24 00:45:57 +02:00
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-10-05 10:41:57 -04:00
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
|
|
|
|
|
2015-11-27 08:47:36 +00:00
|
|
|
|
|
|
|
|
|
2016-11-15 04:05:47 +00:00
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
2015-11-27 08:47:36 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
perf bpf: Check relocation target section
Libbpf should check the target section before doing relocation to ensure
the relocation is correct. If not, a bug in LLVM causes an error. See
[1]. Also, if an incorrect BPF script uses both global variable and
map, global variable whould be treated as map and be relocated without
error.
This patch saves the id of the map section into obj->efile and compare
target section of a relocation symbol against it during relocation.
Previous patch introduces a test case about this problem. After this
patch:
# ~/perf test BPF
37: Test BPF filter :
37.1: Test basic BPF filtering : Ok
37.2: Test BPF prologue generation : Ok
37.3: Test BPF relocation checker : Ok
# perf test -v BPF
...
37.3: Test BPF relocation checker :
...
libbpf: loading object '[bpf_relocation_test]' from buffer
libbpf: section .strtab, size 126, link 0, flags 0, type=3
libbpf: section .text, size 0, link 0, flags 6, type=1
libbpf: section .data, size 0, link 0, flags 3, type=1
libbpf: section .bss, size 0, link 0, flags 3, type=8
libbpf: section func=sys_write, size 104, link 0, flags 6, type=1
libbpf: found program func=sys_write
libbpf: section .relfunc=sys_write, size 16, link 10, flags 0, type=9
libbpf: section maps, size 16, link 0, flags 3, type=1
libbpf: maps in [bpf_relocation_test]: 16 bytes
libbpf: section license, size 4, link 0, flags 3, type=1
libbpf: license of [bpf_relocation_test] is GPL
libbpf: section version, size 4, link 0, flags 3, type=1
libbpf: kernel version of [bpf_relocation_test] is 40400
libbpf: section .symtab, size 144, link 1, flags 0, type=2
libbpf: map 0 is "my_table"
libbpf: collecting relocating info for: 'func=sys_write'
libbpf: Program 'func=sys_write' contains non-map related relo data pointing to section 65522
bpf: failed to load buffer
Compile BPF program failed.
test child finished with 0
---- end ----
Test BPF filter subtest 2: Ok
[1] https://llvm.org/bugs/show_bug.cgi?id=26243
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Will Deacon <will.deacon@arm.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1453715801-7732-3-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-01-25 09:55:49 +00:00
|
|
|
|
2015-11-27 08:47:36 +00:00
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-27 08:47:36 +00:00
|
|
|
|
2019-05-29 10:36:06 -07:00
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
2019-05-29 10:36:06 -07:00
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
2019-06-17 12:26:54 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-10-05 10:41:57 -04:00
|
|
|
|
2016-11-15 04:05:47 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-27 08:47:36 +00:00
|
|
|
|
2016-11-15 04:05:47 +00:00
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
|
|
|
|
|
2015-12-08 02:25:29 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
2016-11-15 04:05:47 +00:00
|
|
|
|
2017-10-05 10:41:57 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
2017-10-05 10:41:57 -04:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-10-15 11:19:55 -07:00
|
|
|
|
|
|
|
|
|
2017-10-05 10:41:57 -04:00
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
2017-10-05 10:41:57 -04:00
|
|
|
|
2015-11-27 08:47:36 +00:00
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
|
|
|
|
|
2016-11-15 04:05:47 +00:00
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-05 08:50:09 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
2019-07-05 08:50:09 -07:00
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
2019-07-05 08:50:09 -07:00
|
|
|
|
|
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
2019-07-05 08:50:09 -07:00
|
|
|
|
|
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-11-15 04:05:47 +00:00
|
|
|
|
2019-07-05 08:50:09 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
2019-04-24 00:45:56 +02:00
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-05 08:50:09 -07:00
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-05 08:50:09 -07:00
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-05 08:50:09 -07:00
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-05 08:50:09 -07:00
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-19 17:27:42 +01:00
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-19 17:27:42 +01:00
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-05 08:50:09 -07:00
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-19 17:27:42 +01:00
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-19 17:27:42 +01:00
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-04-24 00:45:56 +02:00
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
|
|
|
|
|
2015-11-27 08:47:36 +00:00
|
|
|
|
|
|
|
|
|
2018-02-08 12:48:32 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
libbpf: detect supported kernel BTF features and sanitize BTF
Depending on used versions of libbpf, Clang, and kernel, it's possible to
have valid BPF object files with valid BTF information, that still won't
load successfully due to Clang emitting newer BTF features (e.g.,
BTF_KIND_FUNC, .BTF.ext's line_info/func_info, BTF_KIND_DATASEC, etc), that
are not yet supported by older kernel.
This patch adds detection of BTF features and sanitizes BPF object's BTF
by substituting various supported BTF kinds, which have compatible layout:
- BTF_KIND_FUNC -> BTF_KIND_TYPEDEF
- BTF_KIND_FUNC_PROTO -> BTF_KIND_ENUM
- BTF_KIND_VAR -> BTF_KIND_INT
- BTF_KIND_DATASEC -> BTF_KIND_STRUCT
Replacement is done in such a way as to preserve as much information as
possible (names, sizes, etc) where possible without violating kernel's
validation rules.
v2->v3:
- remove duplicate #defines from libbpf_util.h
v1->v2:
- add internal libbpf_internal.h w/ common stuff
- switch SK storage BTF to use new libbpf__probe_raw_btf()
Reported-by: Alexei Starovoitov <ast@fb.com>
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-05-10 14:13:15 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-19 12:46:03 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
libbpf: detect supported kernel BTF features and sanitize BTF
Depending on used versions of libbpf, Clang, and kernel, it's possible to
have valid BPF object files with valid BTF information, that still won't
load successfully due to Clang emitting newer BTF features (e.g.,
BTF_KIND_FUNC, .BTF.ext's line_info/func_info, BTF_KIND_DATASEC, etc), that
are not yet supported by older kernel.
This patch adds detection of BTF features and sanitizes BPF object's BTF
by substituting various supported BTF kinds, which have compatible layout:
- BTF_KIND_FUNC -> BTF_KIND_TYPEDEF
- BTF_KIND_FUNC_PROTO -> BTF_KIND_ENUM
- BTF_KIND_VAR -> BTF_KIND_INT
- BTF_KIND_DATASEC -> BTF_KIND_STRUCT
Replacement is done in such a way as to preserve as much information as
possible (names, sizes, etc) where possible without violating kernel's
validation rules.
v2->v3:
- remove duplicate #defines from libbpf_util.h
v1->v2:
- add internal libbpf_internal.h w/ common stuff
- switch SK storage BTF to use new libbpf__probe_raw_btf()
Reported-by: Alexei Starovoitov <ast@fb.com>
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-05-10 14:13:15 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:55 -07:00
|
|
|
|
2019-06-17 12:26:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
2019-06-17 12:26:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:55 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-19 12:32:42 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:55 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-10-15 11:19:55 -07:00
|
|
|
|
2015-07-01 02:13:56 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-12-07 16:42:29 -08:00
|
|
|
|
2019-04-09 23:20:14 +02:00
|
|
|
|
2015-07-01 02:13:56 +00:00
|
|
|
|
perf bpf: Check relocation target section
Libbpf should check the target section before doing relocation to ensure
the relocation is correct. If not, a bug in LLVM causes an error. See
[1]. Also, if an incorrect BPF script uses both global variable and
map, global variable whould be treated as map and be relocated without
error.
This patch saves the id of the map section into obj->efile and compare
target section of a relocation symbol against it during relocation.
Previous patch introduces a test case about this problem. After this
patch:
# ~/perf test BPF
37: Test BPF filter :
37.1: Test basic BPF filtering : Ok
37.2: Test BPF prologue generation : Ok
37.3: Test BPF relocation checker : Ok
# perf test -v BPF
...
37.3: Test BPF relocation checker :
...
libbpf: loading object '[bpf_relocation_test]' from buffer
libbpf: section .strtab, size 126, link 0, flags 0, type=3
libbpf: section .text, size 0, link 0, flags 6, type=1
libbpf: section .data, size 0, link 0, flags 3, type=1
libbpf: section .bss, size 0, link 0, flags 3, type=8
libbpf: section func=sys_write, size 104, link 0, flags 6, type=1
libbpf: found program func=sys_write
libbpf: section .relfunc=sys_write, size 16, link 10, flags 0, type=9
libbpf: section maps, size 16, link 0, flags 3, type=1
libbpf: maps in [bpf_relocation_test]: 16 bytes
libbpf: section license, size 4, link 0, flags 3, type=1
libbpf: license of [bpf_relocation_test] is GPL
libbpf: section version, size 4, link 0, flags 3, type=1
libbpf: kernel version of [bpf_relocation_test] is 40400
libbpf: section .symtab, size 144, link 1, flags 0, type=2
libbpf: map 0 is "my_table"
libbpf: collecting relocating info for: 'func=sys_write'
libbpf: Program 'func=sys_write' contains non-map related relo data pointing to section 65522
bpf: failed to load buffer
Compile BPF program failed.
test child finished with 0
---- end ----
Test BPF filter subtest 2: Ok
[1] https://llvm.org/bugs/show_bug.cgi?id=26243
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Will Deacon <will.deacon@arm.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1453715801-7732-3-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-01-25 09:55:49 +00:00
|
|
|
|
2015-07-01 02:13:56 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-29 10:36:11 -07:00
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2015-07-01 02:13:56 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-02-08 12:48:17 +01:00
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:52 -07:00
|
|
|
|
2015-07-01 02:13:56 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-02-08 12:48:17 +01:00
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:52 -07:00
|
|
|
|
2015-07-01 02:13:56 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-02-08 12:48:17 +01:00
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:52 -07:00
|
|
|
|
2015-07-01 02:13:56 +00:00
|
|
|
|
2018-02-08 12:48:17 +01:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:56 +00:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:57 +00:00
|
|
|
|
2019-04-09 23:20:14 +02:00
|
|
|
|
2015-07-01 02:13:57 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:52 -07:00
|
|
|
|
|
|
|
|
|
2019-04-09 23:20:14 +02:00
|
|
|
|
2015-07-01 02:13:57 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:52 -07:00
|
|
|
|
|
|
|
|
|
2019-04-09 23:20:14 +02:00
|
|
|
|
perf bpf: Check relocation target section
Libbpf should check the target section before doing relocation to ensure
the relocation is correct. If not, a bug in LLVM causes an error. See
[1]. Also, if an incorrect BPF script uses both global variable and
map, global variable whould be treated as map and be relocated without
error.
This patch saves the id of the map section into obj->efile and compare
target section of a relocation symbol against it during relocation.
Previous patch introduces a test case about this problem. After this
patch:
# ~/perf test BPF
37: Test BPF filter :
37.1: Test basic BPF filtering : Ok
37.2: Test BPF prologue generation : Ok
37.3: Test BPF relocation checker : Ok
# perf test -v BPF
...
37.3: Test BPF relocation checker :
...
libbpf: loading object '[bpf_relocation_test]' from buffer
libbpf: section .strtab, size 126, link 0, flags 0, type=3
libbpf: section .text, size 0, link 0, flags 6, type=1
libbpf: section .data, size 0, link 0, flags 3, type=1
libbpf: section .bss, size 0, link 0, flags 3, type=8
libbpf: section func=sys_write, size 104, link 0, flags 6, type=1
libbpf: found program func=sys_write
libbpf: section .relfunc=sys_write, size 16, link 10, flags 0, type=9
libbpf: section maps, size 16, link 0, flags 3, type=1
libbpf: maps in [bpf_relocation_test]: 16 bytes
libbpf: section license, size 4, link 0, flags 3, type=1
libbpf: license of [bpf_relocation_test] is GPL
libbpf: section version, size 4, link 0, flags 3, type=1
libbpf: kernel version of [bpf_relocation_test] is 40400
libbpf: section .symtab, size 144, link 1, flags 0, type=2
libbpf: map 0 is "my_table"
libbpf: collecting relocating info for: 'func=sys_write'
libbpf: Program 'func=sys_write' contains non-map related relo data pointing to section 65522
bpf: failed to load buffer
Compile BPF program failed.
test child finished with 0
---- end ----
Test BPF filter subtest 2: Ok
[1] https://llvm.org/bugs/show_bug.cgi?id=26243
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Will Deacon <will.deacon@arm.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1453715801-7732-3-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-01-25 09:55:49 +00:00
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
2019-04-09 23:20:14 +02:00
|
|
|
|
|
|
|
|
|
2018-11-19 15:29:16 -08:00
|
|
|
|
2018-12-07 16:42:29 -08:00
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
bpf tools: Collect symbol table from SHT_SYMTAB section
This patch collects symbols section. This section is useful when linking
BPF maps.
What 'bpf_map_xxx()' functions actually require are map's file
descriptors (and the internal verifier converts fds into pointers to
'struct bpf_map'), which we don't know when compiling. Therefore, we
should make compiler generate a 'ldr_64 r1, <imm>' instruction, and
fill the 'imm' field with the actual file descriptor when loading in
libbpf.
BPF programs should be written in this way:
struct bpf_map_def SEC("maps") my_map = {
.type = BPF_MAP_TYPE_HASH,
.key_size = sizeof(unsigned long),
.value_size = sizeof(unsigned long),
.max_entries = 1000000,
};
SEC("my_func=sys_write")
int my_func(void *ctx)
{
...
bpf_map_update_elem(&my_map, &key, &value, BPF_ANY);
...
}
Compiler should convert '&my_map' into a 'ldr_64, r1, <imm>'
instruction, where imm should be the address of 'my_map'. According to
the address, libbpf knows which map it actually referenced, and then
fills the imm field with the 'fd' of that map created by it.
However, since we never really 'link' the object file, the imm field is
only a record in relocation section. Therefore libbpf should do the
relocation:
1. In relocation section (type == SHT_REL), positions of each such
'ldr_64' instruction are recorded with a reference of an entry in
symbol table (SHT_SYMTAB);
2. From records in symbol table we can find the indics of map
variables.
Libbpf first record SHT_SYMTAB and positions of each instruction which
required bu such operation. Then create file descriptor. Finally, after
map creation complete, replace the imm field.
This is the first patch of BPF map related stuff. It records SHT_SYMTAB
into object's efile field for further use.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kaixu Xia <xiakaixu@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1435716878-189507-12-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-07-01 02:13:59 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:52 -07:00
|
|
|
|
2015-12-08 02:25:30 +00:00
|
|
|
|
2019-06-17 12:26:52 -07:00
|
|
|
|
|
|
|
|
|
2019-04-09 23:20:12 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2019-04-09 23:20:12 +02:00
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:52 -07:00
|
|
|
|
2019-04-09 23:20:12 +02:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
2015-07-01 02:14:01 +00:00
|
|
|
|
2019-06-17 12:26:52 -07:00
|
|
|
|
2015-07-01 02:14:01 +00:00
|
|
|
|
2018-02-08 12:48:32 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:01 +00:00
|
|
|
|
2019-06-17 12:26:52 -07:00
|
|
|
|
2018-07-10 14:43:05 -07:00
|
|
|
|
2015-07-01 02:14:01 +00:00
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:52 -07:00
|
|
|
|
2015-07-01 02:14:01 +00:00
|
|
|
|
2019-06-17 12:26:52 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-02-08 12:48:17 +01:00
|
|
|
|
|
|
|
|
|
bpf tools: Collect symbol table from SHT_SYMTAB section
This patch collects symbols section. This section is useful when linking
BPF maps.
What 'bpf_map_xxx()' functions actually require are map's file
descriptors (and the internal verifier converts fds into pointers to
'struct bpf_map'), which we don't know when compiling. Therefore, we
should make compiler generate a 'ldr_64 r1, <imm>' instruction, and
fill the 'imm' field with the actual file descriptor when loading in
libbpf.
BPF programs should be written in this way:
struct bpf_map_def SEC("maps") my_map = {
.type = BPF_MAP_TYPE_HASH,
.key_size = sizeof(unsigned long),
.value_size = sizeof(unsigned long),
.max_entries = 1000000,
};
SEC("my_func=sys_write")
int my_func(void *ctx)
{
...
bpf_map_update_elem(&my_map, &key, &value, BPF_ANY);
...
}
Compiler should convert '&my_map' into a 'ldr_64, r1, <imm>'
instruction, where imm should be the address of 'my_map'. According to
the address, libbpf knows which map it actually referenced, and then
fills the imm field with the 'fd' of that map created by it.
However, since we never really 'link' the object file, the imm field is
only a record in relocation section. Therefore libbpf should do the
relocation:
1. In relocation section (type == SHT_REL), positions of each such
'ldr_64' instruction are recorded with a reference of an entry in
symbol table (SHT_SYMTAB);
2. From records in symbol table we can find the indics of map
variables.
Libbpf first record SHT_SYMTAB and positions of each instruction which
required bu such operation. Then create file descriptor. Finally, after
map creation complete, replace the imm field.
This is the first patch of BPF map related stuff. It records SHT_SYMTAB
into object's efile field for further use.
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@plumgrid.com>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: David Ahern <dsahern@gmail.com>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Kaixu Xia <xiakaixu@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Zefan Li <lizefan@huawei.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1435716878-189507-12-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2015-07-01 02:13:59 +00:00
|
|
|
|
2015-07-01 02:13:56 +00:00
|
|
|
|
2015-11-27 08:47:36 +00:00
|
|
|
|
2015-12-08 02:25:30 +00:00
|
|
|
|
|
|
|
|
|
2019-05-29 10:36:07 -07:00
|
|
|
|
2015-12-08 02:25:30 +00:00
|
|
|
|
2019-06-17 12:26:55 -07:00
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
2018-10-15 11:19:55 -07:00
|
|
|
|
2019-06-17 12:26:55 -07:00
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:53 -07:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:56 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-07-26 14:32:19 -07:00
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
|
|
|
|
|
2018-07-26 14:32:19 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
|
|
|
|
|
2019-05-29 10:36:11 -07:00
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
|
|
|
|
|
2019-05-29 10:36:11 -07:00
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
|
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
perf bpf: Check relocation target section
Libbpf should check the target section before doing relocation to ensure
the relocation is correct. If not, a bug in LLVM causes an error. See
[1]. Also, if an incorrect BPF script uses both global variable and
map, global variable whould be treated as map and be relocated without
error.
This patch saves the id of the map section into obj->efile and compare
target section of a relocation symbol against it during relocation.
Previous patch introduces a test case about this problem. After this
patch:
# ~/perf test BPF
37: Test BPF filter :
37.1: Test basic BPF filtering : Ok
37.2: Test BPF prologue generation : Ok
37.3: Test BPF relocation checker : Ok
# perf test -v BPF
...
37.3: Test BPF relocation checker :
...
libbpf: loading object '[bpf_relocation_test]' from buffer
libbpf: section .strtab, size 126, link 0, flags 0, type=3
libbpf: section .text, size 0, link 0, flags 6, type=1
libbpf: section .data, size 0, link 0, flags 3, type=1
libbpf: section .bss, size 0, link 0, flags 3, type=8
libbpf: section func=sys_write, size 104, link 0, flags 6, type=1
libbpf: found program func=sys_write
libbpf: section .relfunc=sys_write, size 16, link 10, flags 0, type=9
libbpf: section maps, size 16, link 0, flags 3, type=1
libbpf: maps in [bpf_relocation_test]: 16 bytes
libbpf: section license, size 4, link 0, flags 3, type=1
libbpf: license of [bpf_relocation_test] is GPL
libbpf: section version, size 4, link 0, flags 3, type=1
libbpf: kernel version of [bpf_relocation_test] is 40400
libbpf: section .symtab, size 144, link 1, flags 0, type=2
libbpf: map 0 is "my_table"
libbpf: collecting relocating info for: 'func=sys_write'
libbpf: Program 'func=sys_write' contains non-map related relo data pointing to section 65522
bpf: failed to load buffer
Compile BPF program failed.
test child finished with 0
---- end ----
Test BPF filter subtest 2: Ok
[1] https://llvm.org/bugs/show_bug.cgi?id=26243
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Tested-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: Brendan Gregg <brendan.d.gregg@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: He Kuang <hekuang@huawei.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Li Zefan <lizefan@huawei.com>
Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Will Deacon <will.deacon@arm.com>
Cc: pi3orama@163.com
Link: http://lkml.kernel.org/r/1453715801-7732-3-git-send-email-wangnan0@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2016-01-25 09:55:49 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-06-28 14:41:38 -07:00
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-04-24 00:45:56 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
2019-04-09 23:20:12 +02:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:54 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-04-09 23:20:12 +02:00
|
|
|
|
|
|
|
|
|
tools lib bpf: Fix map offsets in relocation
Commit 4708bbda5cb2 ("tools lib bpf: Fix maps resolution") attempted to
fix map resolution by identifying the number of symbols that point to
maps, and using this number to resolve each of the maps.
However, during relocation the original definition of the map size was
still in use. For up to two maps, the calculation was correct if there
was a small difference in size between the map definition in libbpf and
the one that the client library uses. However if the difference was
large, particularly if more than two maps were used in the BPF program,
the relocation would fail.
For example, when using a map definition with size 28, with three maps,
map relocation would count:
(sym_offset / sizeof(struct bpf_map_def) => map_idx)
(0 / 16 => 0), ie map_idx = 0
(28 / 16 => 1), ie map_idx = 1
(56 / 16 => 3), ie map_idx = 3
So, libbpf reports:
libbpf: bpf relocation: map_idx 3 large than 2
Fix map relocation by checking the exact offset of maps when doing
relocation.
Signed-off-by: Joe Stringer <joe@ovn.org>
[Allow different map size in an object]
Signed-off-by: Wang Nan <wangnan0@huawei.com>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: netdev@vger.kernel.org
Fixes: 4708bbda5cb2 ("tools lib bpf: Fix maps resolution")
Link: http://lkml.kernel.org/r/20170123011128.26534-2-joe@ovn.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-01-22 17:11:22 -08:00
|
|
|
|
|
|
|
|
|
2019-04-09 23:20:12 +02:00
|
|
|
|
2019-05-29 10:36:10 -07:00
|
|
|
|
2019-04-09 23:20:12 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
2019-04-09 23:20:12 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
2019-02-04 11:00:58 -08:00
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-02-04 11:00:58 -08:00
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
2019-02-04 11:00:58 -08:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-07-10 14:43:06 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-02-14 15:01:42 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-20 17:11:19 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-04-24 00:45:56 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
libbpf: detect supported kernel BTF features and sanitize BTF
Depending on used versions of libbpf, Clang, and kernel, it's possible to
have valid BPF object files with valid BTF information, that still won't
load successfully due to Clang emitting newer BTF features (e.g.,
BTF_KIND_FUNC, .BTF.ext's line_info/func_info, BTF_KIND_DATASEC, etc), that
are not yet supported by older kernel.
This patch adds detection of BTF features and sanitizes BPF object's BTF
by substituting various supported BTF kinds, which have compatible layout:
- BTF_KIND_FUNC -> BTF_KIND_TYPEDEF
- BTF_KIND_FUNC_PROTO -> BTF_KIND_ENUM
- BTF_KIND_VAR -> BTF_KIND_INT
- BTF_KIND_DATASEC -> BTF_KIND_STRUCT
Replacement is done in such a way as to preserve as much information as
possible (names, sizes, etc) where possible without violating kernel's
validation rules.
v2->v3:
- remove duplicate #defines from libbpf_util.h
v1->v2:
- add internal libbpf_internal.h w/ common stuff
- switch SK storage BTF to use new libbpf__probe_raw_btf()
Reported-by: Alexei Starovoitov <ast@fb.com>
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-05-10 14:13:15 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-29 20:31:09 +02:00
|
|
|
|
libbpf: detect supported kernel BTF features and sanitize BTF
Depending on used versions of libbpf, Clang, and kernel, it's possible to
have valid BPF object files with valid BTF information, that still won't
load successfully due to Clang emitting newer BTF features (e.g.,
BTF_KIND_FUNC, .BTF.ext's line_info/func_info, BTF_KIND_DATASEC, etc), that
are not yet supported by older kernel.
This patch adds detection of BTF features and sanitizes BPF object's BTF
by substituting various supported BTF kinds, which have compatible layout:
- BTF_KIND_FUNC -> BTF_KIND_TYPEDEF
- BTF_KIND_FUNC_PROTO -> BTF_KIND_ENUM
- BTF_KIND_VAR -> BTF_KIND_INT
- BTF_KIND_DATASEC -> BTF_KIND_STRUCT
Replacement is done in such a way as to preserve as much information as
possible (names, sizes, etc) where possible without violating kernel's
validation rules.
v2->v3:
- remove duplicate #defines from libbpf_util.h
v1->v2:
- add internal libbpf_internal.h w/ common stuff
- switch SK storage BTF to use new libbpf__probe_raw_btf()
Reported-by: Alexei Starovoitov <ast@fb.com>
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-05-10 14:13:15 -07:00
|
|
|
|
2019-05-29 20:31:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
libbpf: detect supported kernel BTF features and sanitize BTF
Depending on used versions of libbpf, Clang, and kernel, it's possible to
have valid BPF object files with valid BTF information, that still won't
load successfully due to Clang emitting newer BTF features (e.g.,
BTF_KIND_FUNC, .BTF.ext's line_info/func_info, BTF_KIND_DATASEC, etc), that
are not yet supported by older kernel.
This patch adds detection of BTF features and sanitizes BPF object's BTF
by substituting various supported BTF kinds, which have compatible layout:
- BTF_KIND_FUNC -> BTF_KIND_TYPEDEF
- BTF_KIND_FUNC_PROTO -> BTF_KIND_ENUM
- BTF_KIND_VAR -> BTF_KIND_INT
- BTF_KIND_DATASEC -> BTF_KIND_STRUCT
Replacement is done in such a way as to preserve as much information as
possible (names, sizes, etc) where possible without violating kernel's
validation rules.
v2->v3:
- remove duplicate #defines from libbpf_util.h
v1->v2:
- add internal libbpf_internal.h w/ common stuff
- switch SK storage BTF to use new libbpf__probe_raw_btf()
Reported-by: Alexei Starovoitov <ast@fb.com>
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-05-10 14:13:15 -07:00
|
|
|
|
2019-05-29 20:31:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
libbpf: detect supported kernel BTF features and sanitize BTF
Depending on used versions of libbpf, Clang, and kernel, it's possible to
have valid BPF object files with valid BTF information, that still won't
load successfully due to Clang emitting newer BTF features (e.g.,
BTF_KIND_FUNC, .BTF.ext's line_info/func_info, BTF_KIND_DATASEC, etc), that
are not yet supported by older kernel.
This patch adds detection of BTF features and sanitizes BPF object's BTF
by substituting various supported BTF kinds, which have compatible layout:
- BTF_KIND_FUNC -> BTF_KIND_TYPEDEF
- BTF_KIND_FUNC_PROTO -> BTF_KIND_ENUM
- BTF_KIND_VAR -> BTF_KIND_INT
- BTF_KIND_DATASEC -> BTF_KIND_STRUCT
Replacement is done in such a way as to preserve as much information as
possible (names, sizes, etc) where possible without violating kernel's
validation rules.
v2->v3:
- remove duplicate #defines from libbpf_util.h
v1->v2:
- add internal libbpf_internal.h w/ common stuff
- switch SK storage BTF to use new libbpf__probe_raw_btf()
Reported-by: Alexei Starovoitov <ast@fb.com>
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-05-10 14:13:15 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-29 20:31:09 +02:00
|
|
|
|
libbpf: detect supported kernel BTF features and sanitize BTF
Depending on used versions of libbpf, Clang, and kernel, it's possible to
have valid BPF object files with valid BTF information, that still won't
load successfully due to Clang emitting newer BTF features (e.g.,
BTF_KIND_FUNC, .BTF.ext's line_info/func_info, BTF_KIND_DATASEC, etc), that
are not yet supported by older kernel.
This patch adds detection of BTF features and sanitizes BPF object's BTF
by substituting various supported BTF kinds, which have compatible layout:
- BTF_KIND_FUNC -> BTF_KIND_TYPEDEF
- BTF_KIND_FUNC_PROTO -> BTF_KIND_ENUM
- BTF_KIND_VAR -> BTF_KIND_INT
- BTF_KIND_DATASEC -> BTF_KIND_STRUCT
Replacement is done in such a way as to preserve as much information as
possible (names, sizes, etc) where possible without violating kernel's
validation rules.
v2->v3:
- remove duplicate #defines from libbpf_util.h
v1->v2:
- add internal libbpf_internal.h w/ common stuff
- switch SK storage BTF to use new libbpf__probe_raw_btf()
Reported-by: Alexei Starovoitov <ast@fb.com>
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-05-10 14:13:15 -07:00
|
|
|
|
2019-05-29 20:31:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
libbpf: detect supported kernel BTF features and sanitize BTF
Depending on used versions of libbpf, Clang, and kernel, it's possible to
have valid BPF object files with valid BTF information, that still won't
load successfully due to Clang emitting newer BTF features (e.g.,
BTF_KIND_FUNC, .BTF.ext's line_info/func_info, BTF_KIND_DATASEC, etc), that
are not yet supported by older kernel.
This patch adds detection of BTF features and sanitizes BPF object's BTF
by substituting various supported BTF kinds, which have compatible layout:
- BTF_KIND_FUNC -> BTF_KIND_TYPEDEF
- BTF_KIND_FUNC_PROTO -> BTF_KIND_ENUM
- BTF_KIND_VAR -> BTF_KIND_INT
- BTF_KIND_DATASEC -> BTF_KIND_STRUCT
Replacement is done in such a way as to preserve as much information as
possible (names, sizes, etc) where possible without violating kernel's
validation rules.
v2->v3:
- remove duplicate #defines from libbpf_util.h
v1->v2:
- add internal libbpf_internal.h w/ common stuff
- switch SK storage BTF to use new libbpf__probe_raw_btf()
Reported-by: Alexei Starovoitov <ast@fb.com>
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-05-10 14:13:15 -07:00
|
|
|
|
2019-05-29 20:31:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
libbpf: detect supported kernel BTF features and sanitize BTF
Depending on used versions of libbpf, Clang, and kernel, it's possible to
have valid BPF object files with valid BTF information, that still won't
load successfully due to Clang emitting newer BTF features (e.g.,
BTF_KIND_FUNC, .BTF.ext's line_info/func_info, BTF_KIND_DATASEC, etc), that
are not yet supported by older kernel.
This patch adds detection of BTF features and sanitizes BPF object's BTF
by substituting various supported BTF kinds, which have compatible layout:
- BTF_KIND_FUNC -> BTF_KIND_TYPEDEF
- BTF_KIND_FUNC_PROTO -> BTF_KIND_ENUM
- BTF_KIND_VAR -> BTF_KIND_INT
- BTF_KIND_DATASEC -> BTF_KIND_STRUCT
Replacement is done in such a way as to preserve as much information as
possible (names, sizes, etc) where possible without violating kernel's
validation rules.
v2->v3:
- remove duplicate #defines from libbpf_util.h
v1->v2:
- add internal libbpf_internal.h w/ common stuff
- switch SK storage BTF to use new libbpf__probe_raw_btf()
Reported-by: Alexei Starovoitov <ast@fb.com>
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-05-10 14:13:15 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-20 17:11:19 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-04-24 00:45:56 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
libbpf: detect supported kernel BTF features and sanitize BTF
Depending on used versions of libbpf, Clang, and kernel, it's possible to
have valid BPF object files with valid BTF information, that still won't
load successfully due to Clang emitting newer BTF features (e.g.,
BTF_KIND_FUNC, .BTF.ext's line_info/func_info, BTF_KIND_DATASEC, etc), that
are not yet supported by older kernel.
This patch adds detection of BTF features and sanitizes BPF object's BTF
by substituting various supported BTF kinds, which have compatible layout:
- BTF_KIND_FUNC -> BTF_KIND_TYPEDEF
- BTF_KIND_FUNC_PROTO -> BTF_KIND_ENUM
- BTF_KIND_VAR -> BTF_KIND_INT
- BTF_KIND_DATASEC -> BTF_KIND_STRUCT
Replacement is done in such a way as to preserve as much information as
possible (names, sizes, etc) where possible without violating kernel's
validation rules.
v2->v3:
- remove duplicate #defines from libbpf_util.h
v1->v2:
- add internal libbpf_internal.h w/ common stuff
- switch SK storage BTF to use new libbpf__probe_raw_btf()
Reported-by: Alexei Starovoitov <ast@fb.com>
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-05-10 14:13:15 -07:00
|
|
|
|
|
|
|
|
|
2019-04-24 00:45:56 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-14 20:38:49 -07:00
|
|
|
|
2019-04-24 00:45:56 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-20 17:11:19 -08:00
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:04 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
2019-07-06 11:06:25 -07:00
|
|
|
|
2015-07-01 02:14:04 +00:00
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
2015-07-01 02:14:04 +00:00
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
|
|
|
|
|
2018-07-30 10:53:23 +02:00
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
|
|
|
|
|
2018-07-10 14:43:06 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-20 17:11:20 -08:00
|
|
|
|
|
|
|
|
|
2018-05-16 14:02:49 -07:00
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-06 11:06:25 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-12 22:04:57 -07:00
|
|
|
|
2018-05-22 15:04:24 -07:00
|
|
|
|
|
|
|
|
|
2018-11-20 20:55:56 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
libbpf: allow specifying map definitions using BTF
This patch adds support for a new way to define BPF maps. It relies on
BTF to describe mandatory and optional attributes of a map, as well as
captures type information of key and value naturally. This eliminates
the need for BPF_ANNOTATE_KV_PAIR hack and ensures key/value sizes are
always in sync with the key/value type.
Relying on BTF, this approach allows for both forward and backward
compatibility w.r.t. extending supported map definition features. By
default, any unrecognized attributes are treated as an error, but it's
possible relax this using MAPS_RELAX_COMPAT flag. New attributes, added
in the future will need to be optional.
The outline of the new map definition (short, BTF-defined maps) is as follows:
1. All the maps should be defined in .maps ELF section. It's possible to
have both "legacy" map definitions in `maps` sections and BTF-defined
maps in .maps sections. Everything will still work transparently.
2. The map declaration and initialization is done through
a global/static variable of a struct type with few mandatory and
extra optional fields:
- type field is mandatory and specified type of BPF map;
- key/value fields are mandatory and capture key/value type/size information;
- max_entries attribute is optional; if max_entries is not specified or
initialized, it has to be provided in runtime through libbpf API
before loading bpf_object;
- map_flags is optional and if not defined, will be assumed to be 0.
3. Key/value fields should be **a pointer** to a type describing
key/value. The pointee type is assumed (and will be recorded as such
and used for size determination) to be a type describing key/value of
the map. This is done to save excessive amounts of space allocated in
corresponding ELF sections for key/value of big size.
4. As some maps disallow having BTF type ID associated with key/value,
it's possible to specify key/value size explicitly without
associating BTF type ID with it. Use key_size and value_size fields
to do that (see example below).
Here's an example of simple ARRAY map defintion:
struct my_value { int x, y, z; };
struct {
int type;
int max_entries;
int *key;
struct my_value *value;
} btf_map SEC(".maps") = {
.type = BPF_MAP_TYPE_ARRAY,
.max_entries = 16,
};
This will define BPF ARRAY map 'btf_map' with 16 elements. The key will
be of type int and thus key size will be 4 bytes. The value is struct
my_value of size 12 bytes. This map can be used from C code exactly the
same as with existing maps defined through struct bpf_map_def.
Here's an example of STACKMAP definition (which currently disallows BTF type
IDs for key/value):
struct {
__u32 type;
__u32 max_entries;
__u32 map_flags;
__u32 key_size;
__u32 value_size;
} stackmap SEC(".maps") = {
.type = BPF_MAP_TYPE_STACK_TRACE,
.max_entries = 128,
.map_flags = BPF_F_STACK_BUILD_ID,
.key_size = sizeof(__u32),
.value_size = PERF_MAX_STACK_DEPTH * sizeof(struct bpf_stack_build_id),
};
This approach is naturally extended to support map-in-map, by making a value
field to be another struct that describes inner map. This feature is not
implemented yet. It's also possible to incrementally add features like pinning
with full backwards and forward compatibility. Support for static
initialization of BPF_MAP_TYPE_PROG_ARRAY using pointers to BPF programs
is also on the roadmap.
Signed-off-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2019-06-17 12:26:56 -07:00
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
2018-05-22 15:04:24 -07:00
|
|
|
|
|
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-12 22:04:57 -07:00
|
|
|
|
|
|
|
|
|
2019-07-06 11:06:25 -07:00
|
|
|
|
|
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
2019-07-06 11:06:25 -07:00
|
|
|
|
2019-06-12 22:04:57 -07:00
|
|
|
|
2018-05-22 15:04:24 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:04 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-06 11:06:25 -07:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
2019-07-06 11:06:25 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:04 +00:00
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
2015-07-01 02:14:04 +00:00
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-29 10:36:10 -07:00
|
|
|
|
2015-07-01 02:14:04 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-12-07 16:42:29 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-29 10:36:11 -07:00
|
|
|
|
2018-12-07 16:42:29 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-29 10:36:11 -07:00
|
|
|
|
2018-12-07 16:42:29 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-12-07 16:42:31 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-12-07 16:42:29 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-19 15:29:16 -08:00
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-07-10 14:43:05 -07:00
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-19 15:29:16 -08:00
|
|
|
|
2018-12-07 16:42:29 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-19 15:29:16 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-02-20 01:00:07 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:05 +00:00
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
2015-07-01 02:14:05 +00:00
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
2015-07-01 02:14:05 +00:00
|
|
|
|
2018-11-19 15:29:16 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-12-07 16:42:29 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-19 15:29:16 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:05 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:05 +00:00
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:05 +00:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
2019-04-09 23:20:12 +02:00
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:05 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
2015-07-01 02:14:05 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-02-08 12:48:17 +01:00
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
|
|
|
|
|
2019-05-29 10:36:11 -07:00
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2015-07-01 02:14:02 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
2018-11-19 15:29:16 -08:00
|
|
|
|
2018-12-07 16:42:29 -08:00
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
2018-03-30 15:08:01 -07:00
|
|
|
|
2018-07-30 10:53:23 +02:00
|
|
|
|
2019-04-01 21:27:47 -07:00
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
2018-03-30 15:08:01 -07:00
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
2019-05-29 10:36:08 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-30 15:08:01 -07:00
|
|
|
|
2018-11-19 15:29:16 -08:00
|
|
|
|
|
|
|
|
|
2018-11-20 17:11:21 -08:00
|
|
|
|
|
|
|
|
|
2018-03-30 15:08:01 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-19 15:29:16 -08:00
|
|
|
|
2019-06-12 22:04:57 -07:00
|
|
|
|
2018-11-19 15:29:16 -08:00
|
|
|
|
|
|
|
|
|
2018-12-07 16:42:29 -08:00
|
|
|
|
2018-12-07 16:42:31 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-04-01 21:27:47 -07:00
|
|
|
|
2019-05-24 23:25:19 +01:00
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
2019-04-01 21:27:47 -07:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-04-01 21:27:47 -07:00
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
|
|
|
|
|
2019-04-01 21:27:47 -07:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-04-01 21:27:47 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2018-10-03 15:26:41 -07:00
|
|
|
|
2018-07-30 10:53:23 +02:00
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-30 15:08:01 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-07-13 10:44:02 +00:00
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2016-07-13 10:44:02 +00:00
|
|
|
|
2018-03-30 15:08:01 -07:00
|
|
|
|
2016-07-13 10:44:02 +00:00
|
|
|
|
|
|
|
|
|
2018-03-30 15:08:01 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-07-13 10:44:02 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2016-07-13 10:44:02 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-10-02 13:35:39 -07:00
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
2018-10-03 15:26:43 -07:00
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
2015-11-16 12:10:09 +00:00
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
2015-11-16 12:10:09 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
2015-11-16 12:10:09 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-19 15:29:16 -08:00
|
|
|
|
2018-12-07 16:42:29 -08:00
|
|
|
|
2015-11-16 12:10:09 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-02-13 10:25:53 -08:00
|
|
|
|
2015-11-16 12:10:09 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-19 15:29:16 -08:00
|
|
|
|
2015-11-16 12:10:09 +00:00
|
|
|
|
2018-12-07 16:42:29 -08:00
|
|
|
|
2015-11-16 12:10:09 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
|
|
|
|
|
2018-06-28 14:41:38 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
2019-05-24 11:36:47 +01:00
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-06-28 14:41:38 -07:00
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
2019-05-29 15:26:41 +01:00
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-05-10 10:24:42 -07:00
|
|
|
|
2015-07-01 02:13:57 +00:00
|
|
|
|
2018-05-10 10:24:42 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ipv6: sr: Add seg6local action End.BPF
This patch adds the End.BPF action to the LWT seg6local infrastructure.
This action works like any other seg6local End action, meaning that an IPv6
header with SRH is needed, whose DA has to be equal to the SID of the
action. It will also advance the SRH to the next segment, the BPF program
does not have to take care of this.
Since the BPF program may not be a source of instability in the kernel, it
is important to ensure that the integrity of the packet is maintained
before yielding it back to the IPv6 layer. The hook hence keeps track if
the SRH has been altered through the helpers, and re-validates its
content if needed with seg6_validate_srh. The state kept for validation is
stored in a per-CPU buffer. The BPF program is not allowed to directly
write into the packet, and only some fields of the SRH can be altered
through the helper bpf_lwt_seg6_store_bytes.
Performances profiling has shown that the SRH re-validation does not induce
a significant overhead. If the altered SRH is deemed as invalid, the packet
is dropped.
This validation is also done before executing any action through
bpf_lwt_seg6_action, and will not be performed again if the SRH is not
modified after calling the action.
The BPF program may return 3 types of return codes:
- BPF_OK: the End.BPF action will look up the next destination through
seg6_lookup_nexthop.
- BPF_REDIRECT: if an action has been executed through the
bpf_lwt_seg6_action helper, the BPF program should return this
value, as the skb's destination is already set and the default
lookup should not be performed.
- BPF_DROP : the packet will be dropped.
Signed-off-by: Mathieu Xhonneux <m.xhonneux@gmail.com>
Acked-by: David Lebrun <dlebrun@google.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-20 14:58:16 +01:00
|
|
|
|
2018-05-10 10:24:42 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-05-27 12:24:10 +01:00
|
|
|
|
2018-08-08 01:01:30 -07:00
|
|
|
|
2018-09-14 07:46:20 -07:00
|
|
|
|
2018-05-10 10:24:42 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-04-26 11:49:50 -07:00
|
|
|
|
2018-11-23 12:58:12 -08:00
|
|
|
|
2019-03-08 09:15:26 -08:00
|
|
|
|
2019-06-27 13:38:49 -07:00
|
|
|
|
2018-11-23 12:58:12 -08:00
|
|
|
|
|
|
|
|
|
2018-05-10 10:24:42 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:57 +00:00
|
|
|
|
|
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2015-07-01 02:13:57 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
2018-05-10 10:24:42 -07:00
|
|
|
|
2018-10-15 11:19:55 -07:00
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:54 +00:00
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
|
|
|
|
|
2019-04-24 00:45:56 +02:00
|
|
|
|
2018-10-15 11:19:55 -07:00
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2018-05-10 10:24:42 -07:00
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
2018-10-15 11:19:55 -07:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
2018-07-10 14:43:02 -07:00
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
2018-07-10 14:43:02 -07:00
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
2018-07-10 14:43:02 -07:00
|
|
|
|
2018-10-15 11:19:55 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-07-10 14:43:02 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:54 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-08-27 02:30:55 +00:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:54 +00:00
|
|
|
|
2015-08-27 02:30:55 +00:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:54 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-08-27 02:30:55 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-29 10:36:11 -07:00
|
|
|
|
2015-07-01 02:13:54 +00:00
|
|
|
|
2018-10-15 11:19:55 -07:00
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:04 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:04 +00:00
|
|
|
|
2015-07-01 02:14:07 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:04 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-24 11:36:47 +01:00
|
|
|
|
2015-07-01 02:14:04 +00:00
|
|
|
|
2019-05-24 11:36:47 +01:00
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
|
|
|
|
|
2019-05-24 11:36:47 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:04 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-24 11:36:47 +01:00
|
|
|
|
2015-07-01 02:14:04 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2015-07-01 02:14:04 +00:00
|
|
|
|
|
|
|
|
|
2019-05-24 11:36:47 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
tools lib bpf: Add BPF program pinning APIs
Add new APIs to pin a BPF program (or specific instances) to the
filesystem. The user can specify the path full path within a BPF
filesystem to pin the program.
bpf_program__pin_instance(prog, path, n) will pin the nth instance of
'prog' to the specified path.
bpf_program__pin(prog, path) will create the directory 'path' (if it
does not exist) and pin each instance within that directory. For
instance, path/0, path/1, path/2.
Committer notes:
- Add missing headers for mkdir()
- Check strdup() for failure
- Check snprintf >= size, not >, as == also means truncated, see 'man
snprintf', return value.
- Conditionally define BPF_FS_MAGIC, as it isn't in magic.h in older
systems and we're not yet having a tools/include/uapi/linux/magic.h
copy.
- Do not include linux/magic.h, not present in older distros.
Signed-off-by: Joe Stringer <joe@ovn.org>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: netdev@vger.kernel.org
Link: http://lkml.kernel.org/r/20170126212001.14103-2-joe@ovn.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-01-26 13:19:56 -08:00
|
|
|
|
|
|
|
|
|
2018-07-30 10:53:23 +02:00
|
|
|
|
tools lib bpf: Add BPF program pinning APIs
Add new APIs to pin a BPF program (or specific instances) to the
filesystem. The user can specify the path full path within a BPF
filesystem to pin the program.
bpf_program__pin_instance(prog, path, n) will pin the nth instance of
'prog' to the specified path.
bpf_program__pin(prog, path) will create the directory 'path' (if it
does not exist) and pin each instance within that directory. For
instance, path/0, path/1, path/2.
Committer notes:
- Add missing headers for mkdir()
- Check strdup() for failure
- Check snprintf >= size, not >, as == also means truncated, see 'man
snprintf', return value.
- Conditionally define BPF_FS_MAGIC, as it isn't in magic.h in older
systems and we're not yet having a tools/include/uapi/linux/magic.h
copy.
- Do not include linux/magic.h, not present in older distros.
Signed-off-by: Joe Stringer <joe@ovn.org>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: netdev@vger.kernel.org
Link: http://lkml.kernel.org/r/20170126212001.14103-2-joe@ovn.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-01-26 13:19:56 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-10-03 15:26:41 -07:00
|
|
|
|
2018-07-30 10:53:23 +02:00
|
|
|
|
tools lib bpf: Add BPF program pinning APIs
Add new APIs to pin a BPF program (or specific instances) to the
filesystem. The user can specify the path full path within a BPF
filesystem to pin the program.
bpf_program__pin_instance(prog, path, n) will pin the nth instance of
'prog' to the specified path.
bpf_program__pin(prog, path) will create the directory 'path' (if it
does not exist) and pin each instance within that directory. For
instance, path/0, path/1, path/2.
Committer notes:
- Add missing headers for mkdir()
- Check strdup() for failure
- Check snprintf >= size, not >, as == also means truncated, see 'man
snprintf', return value.
- Conditionally define BPF_FS_MAGIC, as it isn't in magic.h in older
systems and we're not yet having a tools/include/uapi/linux/magic.h
copy.
- Do not include linux/magic.h, not present in older distros.
Signed-off-by: Joe Stringer <joe@ovn.org>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: netdev@vger.kernel.org
Link: http://lkml.kernel.org/r/20170126212001.14103-2-joe@ovn.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-01-26 13:19:56 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-07-30 10:53:23 +02:00
|
|
|
|
tools lib bpf: Add BPF program pinning APIs
Add new APIs to pin a BPF program (or specific instances) to the
filesystem. The user can specify the path full path within a BPF
filesystem to pin the program.
bpf_program__pin_instance(prog, path, n) will pin the nth instance of
'prog' to the specified path.
bpf_program__pin(prog, path) will create the directory 'path' (if it
does not exist) and pin each instance within that directory. For
instance, path/0, path/1, path/2.
Committer notes:
- Add missing headers for mkdir()
- Check strdup() for failure
- Check snprintf >= size, not >, as == also means truncated, see 'man
snprintf', return value.
- Conditionally define BPF_FS_MAGIC, as it isn't in magic.h in older
systems and we're not yet having a tools/include/uapi/linux/magic.h
copy.
- Do not include linux/magic.h, not present in older distros.
Signed-off-by: Joe Stringer <joe@ovn.org>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: netdev@vger.kernel.org
Link: http://lkml.kernel.org/r/20170126212001.14103-2-joe@ovn.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-01-26 13:19:56 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-10-03 15:26:41 -07:00
|
|
|
|
2018-07-30 10:53:23 +02:00
|
|
|
|
tools lib bpf: Add BPF program pinning APIs
Add new APIs to pin a BPF program (or specific instances) to the
filesystem. The user can specify the path full path within a BPF
filesystem to pin the program.
bpf_program__pin_instance(prog, path, n) will pin the nth instance of
'prog' to the specified path.
bpf_program__pin(prog, path) will create the directory 'path' (if it
does not exist) and pin each instance within that directory. For
instance, path/0, path/1, path/2.
Committer notes:
- Add missing headers for mkdir()
- Check strdup() for failure
- Check snprintf >= size, not >, as == also means truncated, see 'man
snprintf', return value.
- Conditionally define BPF_FS_MAGIC, as it isn't in magic.h in older
systems and we're not yet having a tools/include/uapi/linux/magic.h
copy.
- Do not include linux/magic.h, not present in older distros.
Signed-off-by: Joe Stringer <joe@ovn.org>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: netdev@vger.kernel.org
Link: http://lkml.kernel.org/r/20170126212001.14103-2-joe@ovn.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-01-26 13:19:56 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
tools lib bpf: Add BPF program pinning APIs
Add new APIs to pin a BPF program (or specific instances) to the
filesystem. The user can specify the path full path within a BPF
filesystem to pin the program.
bpf_program__pin_instance(prog, path, n) will pin the nth instance of
'prog' to the specified path.
bpf_program__pin(prog, path) will create the directory 'path' (if it
does not exist) and pin each instance within that directory. For
instance, path/0, path/1, path/2.
Committer notes:
- Add missing headers for mkdir()
- Check strdup() for failure
- Check snprintf >= size, not >, as == also means truncated, see 'man
snprintf', return value.
- Conditionally define BPF_FS_MAGIC, as it isn't in magic.h in older
systems and we're not yet having a tools/include/uapi/linux/magic.h
copy.
- Do not include linux/magic.h, not present in older distros.
Signed-off-by: Joe Stringer <joe@ovn.org>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: netdev@vger.kernel.org
Link: http://lkml.kernel.org/r/20170126212001.14103-2-joe@ovn.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-01-26 13:19:56 -08:00
|
|
|
|
|
|
|
|
|
2018-07-30 10:53:23 +02:00
|
|
|
|
tools lib bpf: Add BPF program pinning APIs
Add new APIs to pin a BPF program (or specific instances) to the
filesystem. The user can specify the path full path within a BPF
filesystem to pin the program.
bpf_program__pin_instance(prog, path, n) will pin the nth instance of
'prog' to the specified path.
bpf_program__pin(prog, path) will create the directory 'path' (if it
does not exist) and pin each instance within that directory. For
instance, path/0, path/1, path/2.
Committer notes:
- Add missing headers for mkdir()
- Check strdup() for failure
- Check snprintf >= size, not >, as == also means truncated, see 'man
snprintf', return value.
- Conditionally define BPF_FS_MAGIC, as it isn't in magic.h in older
systems and we're not yet having a tools/include/uapi/linux/magic.h
copy.
- Do not include linux/magic.h, not present in older distros.
Signed-off-by: Joe Stringer <joe@ovn.org>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: netdev@vger.kernel.org
Link: http://lkml.kernel.org/r/20170126212001.14103-2-joe@ovn.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-01-26 13:19:56 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-07-30 10:53:23 +02:00
|
|
|
|
2018-10-03 15:26:41 -07:00
|
|
|
|
2018-07-30 10:53:23 +02:00
|
|
|
|
|
|
|
|
|
tools lib bpf: Add BPF program pinning APIs
Add new APIs to pin a BPF program (or specific instances) to the
filesystem. The user can specify the path full path within a BPF
filesystem to pin the program.
bpf_program__pin_instance(prog, path, n) will pin the nth instance of
'prog' to the specified path.
bpf_program__pin(prog, path) will create the directory 'path' (if it
does not exist) and pin each instance within that directory. For
instance, path/0, path/1, path/2.
Committer notes:
- Add missing headers for mkdir()
- Check strdup() for failure
- Check snprintf >= size, not >, as == also means truncated, see 'man
snprintf', return value.
- Conditionally define BPF_FS_MAGIC, as it isn't in magic.h in older
systems and we're not yet having a tools/include/uapi/linux/magic.h
copy.
- Do not include linux/magic.h, not present in older distros.
Signed-off-by: Joe Stringer <joe@ovn.org>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: netdev@vger.kernel.org
Link: http://lkml.kernel.org/r/20170126212001.14103-2-joe@ovn.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-01-26 13:19:56 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-09 08:21:42 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
tools lib bpf: Add BPF program pinning APIs
Add new APIs to pin a BPF program (or specific instances) to the
filesystem. The user can specify the path full path within a BPF
filesystem to pin the program.
bpf_program__pin_instance(prog, path, n) will pin the nth instance of
'prog' to the specified path.
bpf_program__pin(prog, path) will create the directory 'path' (if it
does not exist) and pin each instance within that directory. For
instance, path/0, path/1, path/2.
Committer notes:
- Add missing headers for mkdir()
- Check strdup() for failure
- Check snprintf >= size, not >, as == also means truncated, see 'man
snprintf', return value.
- Conditionally define BPF_FS_MAGIC, as it isn't in magic.h in older
systems and we're not yet having a tools/include/uapi/linux/magic.h
copy.
- Do not include linux/magic.h, not present in older distros.
Signed-off-by: Joe Stringer <joe@ovn.org>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: netdev@vger.kernel.org
Link: http://lkml.kernel.org/r/20170126212001.14103-2-joe@ovn.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-01-26 13:19:56 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-09 08:21:42 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
|
|
|
|
|
tools lib bpf: Add BPF program pinning APIs
Add new APIs to pin a BPF program (or specific instances) to the
filesystem. The user can specify the path full path within a BPF
filesystem to pin the program.
bpf_program__pin_instance(prog, path, n) will pin the nth instance of
'prog' to the specified path.
bpf_program__pin(prog, path) will create the directory 'path' (if it
does not exist) and pin each instance within that directory. For
instance, path/0, path/1, path/2.
Committer notes:
- Add missing headers for mkdir()
- Check strdup() for failure
- Check snprintf >= size, not >, as == also means truncated, see 'man
snprintf', return value.
- Conditionally define BPF_FS_MAGIC, as it isn't in magic.h in older
systems and we're not yet having a tools/include/uapi/linux/magic.h
copy.
- Do not include linux/magic.h, not present in older distros.
Signed-off-by: Joe Stringer <joe@ovn.org>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: netdev@vger.kernel.org
Link: http://lkml.kernel.org/r/20170126212001.14103-2-joe@ovn.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-01-26 13:19:56 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
tools lib bpf: Add BPF program pinning APIs
Add new APIs to pin a BPF program (or specific instances) to the
filesystem. The user can specify the path full path within a BPF
filesystem to pin the program.
bpf_program__pin_instance(prog, path, n) will pin the nth instance of
'prog' to the specified path.
bpf_program__pin(prog, path) will create the directory 'path' (if it
does not exist) and pin each instance within that directory. For
instance, path/0, path/1, path/2.
Committer notes:
- Add missing headers for mkdir()
- Check strdup() for failure
- Check snprintf >= size, not >, as == also means truncated, see 'man
snprintf', return value.
- Conditionally define BPF_FS_MAGIC, as it isn't in magic.h in older
systems and we're not yet having a tools/include/uapi/linux/magic.h
copy.
- Do not include linux/magic.h, not present in older distros.
Signed-off-by: Joe Stringer <joe@ovn.org>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: netdev@vger.kernel.org
Link: http://lkml.kernel.org/r/20170126212001.14103-2-joe@ovn.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-01-26 13:19:56 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
tools lib bpf: Add BPF program pinning APIs
Add new APIs to pin a BPF program (or specific instances) to the
filesystem. The user can specify the path full path within a BPF
filesystem to pin the program.
bpf_program__pin_instance(prog, path, n) will pin the nth instance of
'prog' to the specified path.
bpf_program__pin(prog, path) will create the directory 'path' (if it
does not exist) and pin each instance within that directory. For
instance, path/0, path/1, path/2.
Committer notes:
- Add missing headers for mkdir()
- Check strdup() for failure
- Check snprintf >= size, not >, as == also means truncated, see 'man
snprintf', return value.
- Conditionally define BPF_FS_MAGIC, as it isn't in magic.h in older
systems and we're not yet having a tools/include/uapi/linux/magic.h
copy.
- Do not include linux/magic.h, not present in older distros.
Signed-off-by: Joe Stringer <joe@ovn.org>
Cc: Alexei Starovoitov <ast@fb.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Wang Nan <wangnan0@huawei.com>
Cc: netdev@vger.kernel.org
Link: http://lkml.kernel.org/r/20170126212001.14103-2-joe@ovn.org
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
2017-01-26 13:19:56 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-01-26 13:19:57 -08:00
|
|
|
|
|
|
|
|
|
2018-07-30 10:53:23 +02:00
|
|
|
|
2017-01-26 13:19:57 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-10-03 15:26:41 -07:00
|
|
|
|
2018-07-30 10:53:23 +02:00
|
|
|
|
2017-01-26 13:19:57 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
2017-01-26 13:19:57 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-01-26 13:19:58 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-02-27 19:04:12 -08:00
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-02-27 19:04:12 -08:00
|
|
|
|
2017-01-26 13:19:58 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
2017-01-26 13:19:58 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-09 08:21:43 -08:00
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-09 08:21:43 -08:00
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-01-26 13:19:58 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-09 08:21:43 -08:00
|
|
|
|
2017-01-26 13:19:58 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
2017-01-26 13:19:58 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-11-26 07:03:26 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
2015-07-01 02:14:04 +00:00
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
2018-11-19 15:29:16 -08:00
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
2015-11-27 08:47:36 +00:00
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:00 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:10 +00:00
|
|
|
|
2015-07-01 02:13:53 +00:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:08 +00:00
|
|
|
|
2015-07-01 02:14:10 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
2015-08-27 02:30:55 +00:00
|
|
|
|
2016-06-03 12:22:51 -03:00
|
|
|
|
2015-08-27 02:30:55 +00:00
|
|
|
|
|
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
2015-11-06 13:49:38 +00:00
|
|
|
|
2016-06-03 12:22:51 -03:00
|
|
|
|
2015-11-06 13:49:38 +00:00
|
|
|
|
|
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
2019-02-14 15:01:43 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-11-26 07:03:26 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
2016-11-26 07:03:26 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-06-28 14:41:39 -07:00
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:08 +00:00
|
|
|
|
2018-11-12 15:44:53 -08:00
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
2015-07-01 02:14:08 +00:00
|
|
|
|
2018-11-12 15:44:53 -08:00
|
|
|
|
2015-07-01 02:14:08 +00:00
|
|
|
|
|
|
|
|
|
2018-11-12 15:44:53 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
2015-07-01 02:14:08 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-12 15:44:53 -08:00
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
2015-07-01 02:14:08 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-06-28 14:41:39 -07:00
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
2018-06-28 14:41:39 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-12 15:44:53 -08:00
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-12 15:44:53 -08:00
|
|
|
|
2018-06-28 14:41:39 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-06-03 12:38:21 -03:00
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:08 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
2015-07-01 02:14:08 +00:00
|
|
|
|
2016-06-03 12:36:39 -03:00
|
|
|
|
2015-07-01 02:14:08 +00:00
|
|
|
|
|
|
|
|
|
2018-06-28 14:41:37 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
2015-07-01 02:14:08 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-03 20:21:05 +09:00
|
|
|
|
2015-07-01 02:14:08 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-06 13:49:37 +00:00
|
|
|
|
2015-07-01 02:14:08 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
2015-07-01 02:14:08 +00:00
|
|
|
|
2015-11-16 12:10:09 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
2015-11-16 12:10:09 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-07-26 14:32:18 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-16 12:10:09 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-07-01 02:14:08 +00:00
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
2017-03-30 21:45:40 -07:00
|
|
|
|
2016-07-13 10:44:01 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
2016-07-13 10:44:01 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-07-13 10:44:01 +00:00
|
|
|
|
2017-01-22 17:11:24 -08:00
|
|
|
|
2017-01-22 17:11:23 -08:00
|
|
|
|
2017-01-22 17:11:24 -08:00
|
|
|
|
|
|
|
|
|
2017-01-22 17:11:23 -08:00
|
|
|
|
2018-04-17 10:28:46 -07:00
|
|
|
|
2017-01-22 17:11:24 -08:00
|
|
|
|
|
|
|
|
|
2016-07-13 10:44:01 +00:00
|
|
|
|
2018-04-23 14:30:38 -07:00
|
|
|
|
|
|
|
|
|
2018-03-30 15:08:01 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-10-31 12:57:18 -07:00
|
|
|
|
|
|
|
|
|
2018-03-30 15:08:01 -07:00
|
|
|
|
2018-09-26 15:24:53 -07:00
|
|
|
|
2018-10-31 12:57:18 -07:00
|
|
|
|
2018-03-30 15:08:01 -07:00
|
|
|
|
2018-09-26 15:24:53 -07:00
|
|
|
|
|
|
|
|
|
2018-10-31 12:57:18 -07:00
|
|
|
|
2018-04-17 10:28:45 -07:00
|
|
|
|
2018-09-26 15:24:53 -07:00
|
|
|
|
|
|
|
|
|
2018-10-31 12:57:18 -07:00
|
|
|
|
2018-09-26 15:24:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
selftests/bpf: Selftest for sys_bind hooks
Add selftest to work with bpf_sock_addr context from
`BPF_PROG_TYPE_CGROUP_SOCK_ADDR` programs.
Try to bind(2) on IP:port and apply:
* loads to make sure context can be read correctly, including narrow
loads (byte, half) for IP and full-size loads (word) for all fields;
* stores to those fields allowed by verifier.
All combination from IPv4/IPv6 and TCP/UDP are tested.
Both scenarios are tested:
* valid programs can be loaded and attached;
* invalid programs can be neither loaded nor attached.
Test passes when expected data can be read from context in the
BPF-program, and after the call to bind(2) socket is bound to IP:port
pair that was written by BPF-program to the context.
Example:
# ./test_sock_addr
Attached bind4 program.
Test case #1 (IPv4/TCP):
Requested: bind(192.168.1.254, 4040) ..
Actual: bind(127.0.0.1, 4444)
Test case #2 (IPv4/UDP):
Requested: bind(192.168.1.254, 4040) ..
Actual: bind(127.0.0.1, 4444)
Attached bind6 program.
Test case #3 (IPv6/TCP):
Requested: bind(face:b00c:1234:5678::abcd, 6060) ..
Actual: bind(::1, 6666)
Test case #4 (IPv6/UDP):
Requested: bind(face:b00c:1234:5678::abcd, 6060) ..
Actual: bind(::1, 6666)
### SUCCESS
Signed-off-by: Andrey Ignatov <rdna@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-03-30 15:08:03 -07:00
|
|
|
|
2017-12-13 15:18:51 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-30 15:08:01 -07:00
|
|
|
|
2018-10-31 12:57:18 -07:00
|
|
|
|
2018-09-26 15:24:53 -07:00
|
|
|
|
2017-12-13 15:18:51 +00:00
|
|
|
|
2018-09-26 15:24:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-09-26 15:24:54 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-09-26 15:24:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-09-26 15:24:55 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-09-26 15:24:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-07 01:48:59 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-03-08 09:15:26 -08:00
|
|
|
|
|
|
|
|
|
2019-06-27 13:38:49 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-12-13 15:18:51 +00:00
|
|
|
|
|
|
|
|
|
2018-09-26 15:24:53 -07:00
|
|
|
|
2018-03-30 15:08:01 -07:00
|
|
|
|
2018-09-26 15:24:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-30 15:08:01 -07:00
|
|
|
|
2019-01-21 22:06:38 +09:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-07-10 14:42:59 -07:00
|
|
|
|
|
|
|
|
|
2017-12-13 15:18:51 +00:00
|
|
|
|
2019-01-21 22:06:38 +09:00
|
|
|
|
2017-12-13 15:18:51 +00:00
|
|
|
|
|
|
|
|
|
2018-07-10 14:42:59 -07:00
|
|
|
|
|
|
|
|
|
2017-12-13 15:18:51 +00:00
|
|
|
|
2018-07-10 14:42:59 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-01-21 22:06:38 +09:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-07-10 14:42:59 -07:00
|
|
|
|
|
|
|
|
|
2017-12-13 15:18:51 +00:00
|
|
|
|
2018-09-26 15:24:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-01-21 22:06:38 +09:00
|
|
|
|
2018-09-26 15:24:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-10-31 12:57:18 -07:00
|
|
|
|
2018-09-26 15:24:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-01-21 22:06:38 +09:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-09-26 15:24:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-07-10 14:42:59 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-12-13 15:18:51 +00:00
|
|
|
|
|
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
2016-06-03 12:15:52 -03:00
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
|
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
2016-06-02 14:21:06 -03:00
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
|
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
2015-11-27 08:47:36 +00:00
|
|
|
|
2016-06-02 11:02:05 -03:00
|
|
|
|
2015-11-27 08:47:36 +00:00
|
|
|
|
|
|
|
|
|
2018-07-24 08:40:21 -07:00
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
2018-05-22 15:04:24 -07:00
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
|
|
|
|
|
2018-07-24 08:40:21 -07:00
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
2018-05-22 15:04:24 -07:00
|
|
|
|
2018-04-18 15:56:05 -07:00
|
|
|
|
|
|
|
|
|
2016-06-03 12:38:21 -03:00
|
|
|
|
|
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
2016-06-02 10:51:59 -03:00
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
|
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
2018-07-10 14:43:01 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
bpf, libbpf: support global data/bss/rodata sections
This work adds BPF loader support for global data sections
to libbpf. This allows to write BPF programs in more natural
C-like way by being able to define global variables and const
data.
Back at LPC 2018 [0] we presented a first prototype which
implemented support for global data sections by extending BPF
syscall where union bpf_attr would get additional memory/size
pair for each section passed during prog load in order to later
add this base address into the ldimm64 instruction along with
the user provided offset when accessing a variable. Consensus
from LPC was that for proper upstream support, it would be
more desirable to use maps instead of bpf_attr extension as
this would allow for introspection of these sections as well
as potential live updates of their content. This work follows
this path by taking the following steps from loader side:
1) In bpf_object__elf_collect() step we pick up ".data",
".rodata", and ".bss" section information.
2) If present, in bpf_object__init_internal_map() we add
maps to the obj's map array that corresponds to each
of the present sections. Given section size and access
properties can differ, a single entry array map is
created with value size that is corresponding to the
ELF section size of .data, .bss or .rodata. These
internal maps are integrated into the normal map
handling of libbpf such that when user traverses all
obj maps, they can be differentiated from user-created
ones via bpf_map__is_internal(). In later steps when
we actually create these maps in the kernel via
bpf_object__create_maps(), then for .data and .rodata
sections their content is copied into the map through
bpf_map_update_elem(). For .bss this is not necessary
since array map is already zero-initialized by default.
Additionally, for .rodata the map is frozen as read-only
after setup, such that neither from program nor syscall
side writes would be possible.
3) In bpf_program__collect_reloc() step, we record the
corresponding map, insn index, and relocation type for
the global data.
4) And last but not least in the actual relocation step in
bpf_program__relocate(), we mark the ldimm64 instruction
with src_reg = BPF_PSEUDO_MAP_VALUE where in the first
imm field the map's file descriptor is stored as similarly
done as in BPF_PSEUDO_MAP_FD, and in the second imm field
(as ldimm64 is 2-insn wide) we store the access offset
into the section. Given these maps have only single element
ldimm64's off remains zero in both parts.
5) On kernel side, this special marked BPF_PSEUDO_MAP_VALUE
load will then store the actual target address in order
to have a 'map-lookup'-free access. That is, the actual
map value base address + offset. The destination register
in the verifier will then be marked as PTR_TO_MAP_VALUE,
containing the fixed offset as reg->off and backing BPF
map as reg->map_ptr. Meaning, it's treated as any other
normal map value from verification side, only with
efficient, direct value access instead of actual call to
map lookup helper as in the typical case.
Currently, only support for static global variables has been
added, and libbpf rejects non-static global variables from
loading. This can be lifted until we have proper semantics
for how BPF will treat multi-object BPF loads. From BTF side,
libbpf will set the value type id of the types corresponding
to the ".bss", ".data" and ".rodata" names which LLVM will
emit without the object name prefix. The key type will be
left as zero, thus making use of the key-less BTF option in
array maps.
Simple example dump of program using globals vars in each
section:
# bpftool prog
[...]
6784: sched_cls name load_static_dat tag a7e1291567277844 gpl
loaded_at 2019-03-11T15:39:34+0000 uid 0
xlated 1776B jited 993B memlock 4096B map_ids 2238,2237,2235,2236,2239,2240
# bpftool map show id 2237
2237: array name test_glo.bss flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2235
2235: array name test_glo.data flags 0x0
key 4B value 64B max_entries 1 memlock 4096B
# bpftool map show id 2236
2236: array name test_glo.rodata flags 0x80
key 4B value 96B max_entries 1 memlock 4096B
# bpftool prog dump xlated id 6784
int load_static_data(struct __sk_buff * skb):
; int load_static_data(struct __sk_buff *skb)
0: (b7) r6 = 0
; test_reloc(number, 0, &num0);
1: (63) *(u32 *)(r10 -4) = r6
2: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
3: (07) r2 += -4
; test_reloc(number, 0, &num0);
4: (18) r1 = map[id:2238]
6: (18) r3 = map[id:2237][0]+0 <-- direct addr in .bss area
8: (b7) r4 = 0
9: (85) call array_map_update_elem#100464
10: (b7) r1 = 1
; test_reloc(number, 1, &num1);
[...]
; test_reloc(string, 2, str2);
120: (18) r8 = map[id:2237][0]+16 <-- same here at offset +16
122: (18) r1 = map[id:2239]
124: (18) r3 = map[id:2237][0]+16
126: (b7) r4 = 0
127: (85) call array_map_update_elem#100464
128: (b7) r1 = 120
; str1[5] = 'x';
129: (73) *(u8 *)(r9 +5) = r1
; test_reloc(string, 3, str1);
130: (b7) r1 = 3
131: (63) *(u32 *)(r10 -4) = r1
132: (b7) r9 = 3
133: (bf) r2 = r10
; int load_static_data(struct __sk_buff *skb)
134: (07) r2 += -4
; test_reloc(string, 3, str1);
135: (18) r1 = map[id:2239]
137: (18) r3 = map[id:2235][0]+16 <-- direct addr in .data area
139: (b7) r4 = 0
140: (85) call array_map_update_elem#100464
141: (b7) r1 = 111
; __builtin_memcpy(&str2[2], "hello", sizeof("hello"));
142: (73) *(u8 *)(r8 +6) = r1 <-- further access based on .bss data
143: (b7) r1 = 108
144: (73) *(u8 *)(r8 +5) = r1
[...]
For Cilium use-case in particular, this enables migrating configuration
constants from Cilium daemon's generated header defines into global
data sections such that expensive runtime recompilations with LLVM can
be avoided altogether. Instead, the ELF file becomes effectively a
"template", meaning, it is compiled only once (!) and the Cilium daemon
will then rewrite relevant configuration data from the ELF's .data or
.rodata sections directly instead of recompiling the program. The
updated ELF is then loaded into the kernel and atomically replaces
the existing program in the networking datapath. More info in [0].
Based upon recent fix in LLVM, commit c0db6b6bd444 ("[BPF] Don't fail
for static variables").
[0] LPC 2018, BPF track, "ELF relocation for static data in BPF",
http://vger.kernel.org/lpc-bpf2018.html#session-3
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Andrii Nakryiko <andriin@fb.com>
Acked-by: Martin KaFai Lau <kafai@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
2019-04-09 23:20:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-06-28 14:41:37 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-20 20:55:56 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
|
|
|
|
|
2015-11-27 08:47:35 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-27 08:47:36 +00:00
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
2018-11-09 08:21:41 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-11-27 08:47:36 +00:00
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
2015-11-27 08:47:36 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-02-27 19:04:12 -08:00
|
|
|
|
2015-12-08 02:25:29 +00:00
|
|
|
|
2015-11-27 08:47:36 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-11-26 07:03:27 +00:00
|
|
|
|
2019-02-01 22:42:23 +01:00
|
|
|
|
2019-06-17 15:48:58 -07:00
|
|
|
|
2019-02-01 22:42:23 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-11-26 07:03:27 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-17 12:26:54 -07:00
|
|
|
|
2016-11-26 07:03:27 +00:00
|
|
|
|
2017-01-22 17:11:25 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-05-25 14:32:57 +05:30
|
|
|
|
2017-01-22 17:11:25 -08:00
|
|
|
|
2017-08-15 22:34:22 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-30 15:08:01 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-08-15 22:34:22 -07:00
|
|
|
|
2019-07-02 18:25:31 +08:00
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
2018-03-30 15:08:01 -07:00
|
|
|
|
|
|
|
|
|
2017-08-15 22:34:22 -07:00
|
|
|
|
2018-05-16 14:02:49 -07:00
|
|
|
|
2017-08-15 22:34:22 -07:00
|
|
|
|
|
|
|
|
|
2018-03-30 15:08:01 -07:00
|
|
|
|
|
|
|
|
|
2018-05-10 10:24:42 -07:00
|
|
|
|
|
|
|
|
|
2018-03-30 15:08:01 -07:00
|
|
|
|
2019-07-02 18:25:31 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-07-10 14:43:02 -07:00
|
|
|
|
2018-05-10 10:09:34 -07:00
|
|
|
|
2017-08-15 22:34:22 -07:00
|
|
|
|
|
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-30 15:08:01 -07:00
|
|
|
|
2018-05-16 14:02:49 -07:00
|
|
|
|
2018-03-30 15:08:01 -07:00
|
|
|
|
|
|
|
|
|
2018-07-10 14:42:59 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-03-30 15:08:01 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-04-01 21:27:47 -07:00
|
|
|
|
2019-05-24 23:25:19 +01:00
|
|
|
|
2018-09-03 08:30:07 +09:00
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-02-27 19:04:12 -08:00
|
|
|
|
2018-07-10 14:43:01 -07:00
|
|
|
|
|
|
|
|
|
2018-05-16 14:02:49 -07:00
|
|
|
|
|
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
|
|
|
|
|
2017-08-15 22:34:22 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-12-14 17:55:10 -08:00
|
|
|
|
2017-08-15 22:34:22 -07:00
|
|
|
|
|
|
|
|
|
2018-05-10 10:24:40 -07:00
|
|
|
|
2019-07-01 16:58:56 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-01 16:58:57 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-01 16:58:58 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-08 21:00:07 -07:00
|
|
|
|
|
|
|
|
|
2019-07-01 16:58:58 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-01 16:58:59 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-01 16:59:00 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-05-10 10:24:40 -07:00
|
|
|
|
2018-10-21 02:09:28 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-05-10 10:24:40 -07:00
|
|
|
|
2018-10-21 02:09:28 +02:00
|
|
|
|
2018-10-19 15:51:03 +02:00
|
|
|
|
2018-05-10 10:24:40 -07:00
|
|
|
|
2018-10-21 02:09:28 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-05-10 10:24:40 -07:00
|
|
|
|
2018-10-21 02:09:28 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-05-10 10:24:40 -07:00
|
|
|
|
2018-10-21 02:09:28 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-05-10 10:24:40 -07:00
|
|
|
|
2018-10-21 02:09:28 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-05-10 10:24:40 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-10-21 02:09:28 +02:00
|
|
|
|
2018-05-10 10:24:40 -07:00
|
|
|
|
|
|
|
|
|
2018-10-21 02:09:28 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-05-10 10:24:40 -07:00
|
|
|
|
|
|
|
|
|
2018-10-21 02:09:28 +02:00
|
|
|
|
|
|
|
|
|
2018-05-10 10:24:40 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-10-19 15:51:03 +02:00
|
|
|
|
2018-05-10 10:24:40 -07:00
|
|
|
|
|
|
|
|
|
2019-03-11 22:30:38 -07:00
|
|
|
|
2019-07-06 11:06:24 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-19 11:34:07 -03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-07-06 11:06:24 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-03-11 22:30:38 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-06-10 17:56:50 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|