2019-05-19 13:08:20 +01:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2014-12-10 15:52:10 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2014-12-10 15:52:10 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2014-12-10 15:52:10 -08:00
|
|
|
|
2018-05-08 15:14:57 -03:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2014-12-10 15:52:08 -08:00
|
|
|
|
|
|
|
|
|
2026-07-21 16:13:45 +02:00
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
binfmt_misc: include what is used
The include list still reflects code that left this file years ago:
nothing here uses sched/mm.h, pagemap.h, namei.h, syscalls.h or
anything from fs/internal.h anymore, mount.h and the bm_fs_type
forward declaration lost their last user when the pinned bm_mnt
machinery was removed. Drop all of that and instead spell out the
headers the file actually relies on but so far pulled in
transitively: bitops, bits, bug, cleanup, cred, kstrtox, printk,
refcount, string and user_namespace. With that nothing needs the
kernel.h grab bag anymore, so it goes too, and the list is sorted
alphabetically.
Link: https://patch.msgid.link/20260710-work-binfmt_misc-locking-v3-23-a162f7cb58d6@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-10 11:33:24 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2008-07-23 21:29:15 -07:00
|
|
|
|
binfmt_misc: include what is used
The include list still reflects code that left this file years ago:
nothing here uses sched/mm.h, pagemap.h, namei.h, syscalls.h or
anything from fs/internal.h anymore, mount.h and the bm_fs_type
forward declaration lost their last user when the pinned bm_mnt
machinery was removed. Drop all of that and instead spell out the
headers the file actually relies on but so far pulled in
transitively: bitops, bits, bug, cleanup, cred, kstrtox, printk,
refcount, string and user_namespace. With that nothing needs the
kernel.h grab bag anymore, so it goes too, and the list is sorted
alphabetically.
Link: https://patch.msgid.link/20260710-work-binfmt_misc-locking-v3-23-a162f7cb58d6@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-10 11:33:24 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
binfmt_misc: include what is used
The include list still reflects code that left this file years ago:
nothing here uses sched/mm.h, pagemap.h, namei.h, syscalls.h or
anything from fs/internal.h anymore, mount.h and the bm_fs_type
forward declaration lost their last user when the pinned bm_mnt
machinery was removed. Drop all of that and instead spell out the
headers the file actually relies on but so far pulled in
transitively: bitops, bits, bug, cleanup, cred, kstrtox, printk,
refcount, string and user_namespace. With that nothing needs the
kernel.h grab bag anymore, so it goes too, and the list is sorted
alphabetically.
Link: https://patch.msgid.link/20260710-work-binfmt_misc-locking-v3-23-a162f7cb58d6@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-10 11:33:24 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-14 21:58:08 +02:00
|
|
|
|
binfmt_misc: include what is used
The include list still reflects code that left this file years ago:
nothing here uses sched/mm.h, pagemap.h, namei.h, syscalls.h or
anything from fs/internal.h anymore, mount.h and the bm_fs_type
forward declaration lost their last user when the pinned bm_mnt
machinery was removed. Drop all of that and instead spell out the
headers the file actually relies on but so far pulled in
transitively: bitops, bits, bug, cleanup, cred, kstrtox, printk,
refcount, string and user_namespace. With that nothing needs the
kernel.h grab bag anymore, so it goes too, and the list is sorted
alphabetically.
Link: https://patch.msgid.link/20260710-work-binfmt_misc-locking-v3-23-a162f7cb58d6@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-10 11:33:24 +02:00
|
|
|
|
|
|
|
|
|
2014-12-10 15:52:08 -08:00
|
|
|
|
binfmt_misc: include what is used
The include list still reflects code that left this file years ago:
nothing here uses sched/mm.h, pagemap.h, namei.h, syscalls.h or
anything from fs/internal.h anymore, mount.h and the bm_fs_type
forward declaration lost their last user when the pinned bm_mnt
machinery was removed. Drop all of that and instead spell out the
headers the file actually relies on but so far pulled in
transitively: bitops, bits, bug, cleanup, cred, kstrtox, printk,
refcount, string and user_namespace. With that nothing needs the
kernel.h grab bag anymore, so it goes too, and the list is sorted
alphabetically.
Link: https://patch.msgid.link/20260710-work-binfmt_misc-locking-v3-23-a162f7cb58d6@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-10 11:33:24 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2016-02-17 16:51:16 -08:00
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:08 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-21 16:13:52 +02:00
|
|
|
|
2026-07-21 16:14:00 +02:00
|
|
|
|
binfmt_misc: let a register string create an entry disabled
An entry is matchable as soon as it is registered. create_entry() sets
the enabled bit for every type and add_entry() links it straight into the
instance, so everything an entry needs has to fit in the write that
creates it.
Add a 'D' flag. The entry is created disabled and has to be enabled by
writing '1' to its entry file before it can match anything. That splits a
registration into create and activate, which a later patch uses to
configure an entry beyond what one register string can carry. It is
useful on its own too. Entries can be staged without dispatching the
moment they are written.
A staged entry stays out of the search list entirely. add_entry() only
hashes an entry that is born matchable, and the first '1' written to
the entry file hashes a staged one, which takes its place in the search
order at that point. The rcu insertion publishes the fully configured
entry, so the exec side keeps the plain enabled test it always had.
Removal cannot rely on the search list anymore. Whether an entry was
already removed is now decided by its dentry, '-1' to the status file
walks the directory instead of the list so staged entries do not
survive it, and a '1' through a file handle held across a removal
publishes nothing.
'D' is consumed at registration and not recorded. What matters afterwards
is whether the entry is enabled, and the entry file already reports that.
A 'B' entry's flags field had to be empty so far because every flag it
could name shaped the invocation, which a bpf handler picks per exec with
bpf_binprm_set_flags(). 'D' shapes the registration instead. So the rule
becomes what it always meant: a 'B' entry carries no invocation flags,
and 'D' composes.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-1-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:03 +02:00
|
|
|
|
2026-07-10 11:33:09 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
binfmt_misc: let a register string create an entry disabled
An entry is matchable as soon as it is registered. create_entry() sets
the enabled bit for every type and add_entry() links it straight into the
instance, so everything an entry needs has to fit in the write that
creates it.
Add a 'D' flag. The entry is created disabled and has to be enabled by
writing '1' to its entry file before it can match anything. That splits a
registration into create and activate, which a later patch uses to
configure an entry beyond what one register string can carry. It is
useful on its own too. Entries can be staged without dispatching the
moment they are written.
A staged entry stays out of the search list entirely. add_entry() only
hashes an entry that is born matchable, and the first '1' written to
the entry file hashes a staged one, which takes its place in the search
order at that point. The rcu insertion publishes the fully configured
entry, so the exec side keeps the plain enabled test it always had.
Removal cannot rely on the search list anymore. Whether an entry was
already removed is now decided by its dentry, '-1' to the status file
walks the directory instead of the list so staged entries do not
survive it, and a '1' through a file handle held across a removal
publishes nothing.
'D' is consumed at registration and not recorded. What matters afterwards
is whether the entry is enabled, and the entry file already reports that.
A 'B' entry's flags field had to be empty so far because every flag it
could name shaped the invocation, which a bpf handler picks per exec with
bpf_binprm_set_flags(). 'D' shapes the registration instead. So the rule
becomes what it always meant: a 'B' entry carries no invocation flags,
and 'D' composes.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-1-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:03 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-21 16:13:45 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-21 16:13:53 +02:00
|
|
|
|
2026-07-21 16:14:00 +02:00
|
|
|
|
binfmt_misc: let a register string create an entry disabled
An entry is matchable as soon as it is registered. create_entry() sets
the enabled bit for every type and add_entry() links it straight into the
instance, so everything an entry needs has to fit in the write that
creates it.
Add a 'D' flag. The entry is created disabled and has to be enabled by
writing '1' to its entry file before it can match anything. That splits a
registration into create and activate, which a later patch uses to
configure an entry beyond what one register string can carry. It is
useful on its own too. Entries can be staged without dispatching the
moment they are written.
A staged entry stays out of the search list entirely. add_entry() only
hashes an entry that is born matchable, and the first '1' written to
the entry file hashes a staged one, which takes its place in the search
order at that point. The rcu insertion publishes the fully configured
entry, so the exec side keeps the plain enabled test it always had.
Removal cannot rely on the search list anymore. Whether an entry was
already removed is now decided by its dentry, '-1' to the status file
walks the directory instead of the list so staged entries do not
survive it, and a '1' through a file handle held across a removal
publishes nothing.
'D' is consumed at registration and not recorded. What matters afterwards
is whether the entry is enabled, and the entry file already reports that.
A 'B' entry's flags field had to be empty so far because every flag it
could name shaped the invocation, which a bpf handler picks per exec with
bpf_binprm_set_flags(). 'D' shapes the registration instead. So the rule
becomes what it always meant: a 'B' entry carries no invocation flags,
and 'D' composes.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-1-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:03 +02:00
|
|
|
|
2026-07-21 16:13:45 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:10 +02:00
|
|
|
|
2026-07-10 11:33:05 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-10-03 16:15:58 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
|
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
2026-07-10 11:33:06 +02:00
|
|
|
|
2026-07-10 11:33:20 +02:00
|
|
|
|
2026-07-10 11:33:10 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
binfmt_misc: expand the register format limit to 1920 bytes
The current code places a 256 byte limit on the registration format.
This ends up being fairly limited when you try to do matching against a
binary format like ELF:
- the magic & mask formats cannot have any embedded NUL chars
(string_unescape_inplace halts at the first NUL)
- each escape sequence quadruples the size: \x00 is needed for NUL
- trying to match bytes at the start of the file as well as further
on leads to a lot of \x00 sequences in the mask
- magic & mask have to be the same length (when decoded)
- still need bytes for the other fields
- impossible!
Let's look at a concrete (and common) example: using QEMU to run MIPS
ELFs. The name field uses 11 bytes "qemu-mipsel". The interp uses 20
bytes "/usr/bin/qemu-mipsel". The type & flags takes up 4 bytes. We
need 7 bytes for the delimiter (usually ":"). We can skip offset. So
already we're down to 107 bytes to use with the magic/mask instead of
the real limit of 128 (BINPRM_BUF_SIZE). If people use shell code to
register (which they do the majority of the time), they're down to ~26
possible bytes since the escape sequence must be \x##.
The ELF format looks like (both 32 & 64 bit):
e_ident: 16 bytes
e_type: 2 bytes
e_machine: 2 bytes
Those 20 bytes are enough for most architectures because they have so few
formats in the first place, thus they can be uniquely identified. That
also means for shell users, since 20 is smaller than 26, they can sanely
register a handler.
But for some targets (like MIPS), we need to poke further. The ELF fields
continue on:
e_entry: 4 or 8 bytes
e_phoff: 4 or 8 bytes
e_shoff: 4 or 8 bytes
e_flags: 4 bytes
We only care about e_flags here as that includes the bits to identify
whether the ELF is O32/N32/N64. But now we have to consume another 16
bytes (for 32 bit ELFs) or 28 bytes (for 64 bit ELFs) just to match the
flags. If every byte is escaped, we send 288 more bytes to the kernel
((20 {e_ident,e_type,e_machine} + 12 {e_entry,e_phoff,e_shoff} + 4
{e_flags}) * 2 {mask,magic} * 4 {escape}) and we've clearly blown our
budget.
Even if we try to be clever and do the decoding ourselves (rather than
relying on the kernel to process \x##), we still can't hit the mark --
string_unescape_inplace treats mask & magic as C strings so NUL cannot
be embedded. That leaves us with having to pass \x00 for the 12/24
entry/phoff/shoff bytes (as those will be completely random addresses),
and that is a minimum requirement of 48/96 bytes for the mask alone.
Add up the rest and we blow through it (this is for 64 bit ELFs):
magic: 20 {e_ident,e_type,e_machine} + 24 {e_entry,e_phoff,e_shoff} +
4 {e_flags} = 48 # ^^ See note below.
mask: 20 {e_ident,e_type,e_machine} + 96 {e_entry,e_phoff,e_shoff} +
4 {e_flags} = 120
Remember above we had 107 left over, and now we're at 168. This is of
course the *best* case scenario -- you'll also want to have NUL bytes
in the magic & mask too to match literal zeros.
Note: the reason we can use 24 in the magic is that we can work off of the
fact that for bytes the mask would clobber, we can stuff any value into
magic that we want. So when mask is \x00, we don't need the magic to also
be \x00, it can be an unescaped raw byte like '!'. This lets us handle
more formats (barely) under the current 256 limit, but that's a pretty
tall hoop to force people to jump through.
With all that said, let's bump the limit from 256 bytes to 1920. This way
we support escaping every byte of the mask & magic field (which is 1024
bytes by themselves -- 128 * 4 * 2), and we leave plenty of room for other
fields. Like long paths to the interpreter (when you have source in your
/really/long/homedir/qemu/foo). Since the current code stuffs more than
one structure into the same buffer, we leave a bit of space to easily
round up to 2k. 1920 is just as arbitrary as 256 ;).
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-10-13 15:52:03 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:10 +02:00
|
|
|
|
binfmt_misc: expand the register format limit to 1920 bytes
The current code places a 256 byte limit on the registration format.
This ends up being fairly limited when you try to do matching against a
binary format like ELF:
- the magic & mask formats cannot have any embedded NUL chars
(string_unescape_inplace halts at the first NUL)
- each escape sequence quadruples the size: \x00 is needed for NUL
- trying to match bytes at the start of the file as well as further
on leads to a lot of \x00 sequences in the mask
- magic & mask have to be the same length (when decoded)
- still need bytes for the other fields
- impossible!
Let's look at a concrete (and common) example: using QEMU to run MIPS
ELFs. The name field uses 11 bytes "qemu-mipsel". The interp uses 20
bytes "/usr/bin/qemu-mipsel". The type & flags takes up 4 bytes. We
need 7 bytes for the delimiter (usually ":"). We can skip offset. So
already we're down to 107 bytes to use with the magic/mask instead of
the real limit of 128 (BINPRM_BUF_SIZE). If people use shell code to
register (which they do the majority of the time), they're down to ~26
possible bytes since the escape sequence must be \x##.
The ELF format looks like (both 32 & 64 bit):
e_ident: 16 bytes
e_type: 2 bytes
e_machine: 2 bytes
Those 20 bytes are enough for most architectures because they have so few
formats in the first place, thus they can be uniquely identified. That
also means for shell users, since 20 is smaller than 26, they can sanely
register a handler.
But for some targets (like MIPS), we need to poke further. The ELF fields
continue on:
e_entry: 4 or 8 bytes
e_phoff: 4 or 8 bytes
e_shoff: 4 or 8 bytes
e_flags: 4 bytes
We only care about e_flags here as that includes the bits to identify
whether the ELF is O32/N32/N64. But now we have to consume another 16
bytes (for 32 bit ELFs) or 28 bytes (for 64 bit ELFs) just to match the
flags. If every byte is escaped, we send 288 more bytes to the kernel
((20 {e_ident,e_type,e_machine} + 12 {e_entry,e_phoff,e_shoff} + 4
{e_flags}) * 2 {mask,magic} * 4 {escape}) and we've clearly blown our
budget.
Even if we try to be clever and do the decoding ourselves (rather than
relying on the kernel to process \x##), we still can't hit the mark --
string_unescape_inplace treats mask & magic as C strings so NUL cannot
be embedded. That leaves us with having to pass \x00 for the 12/24
entry/phoff/shoff bytes (as those will be completely random addresses),
and that is a minimum requirement of 48/96 bytes for the mask alone.
Add up the rest and we blow through it (this is for 64 bit ELFs):
magic: 20 {e_ident,e_type,e_machine} + 24 {e_entry,e_phoff,e_shoff} +
4 {e_flags} = 48 # ^^ See note below.
mask: 20 {e_ident,e_type,e_machine} + 96 {e_entry,e_phoff,e_shoff} +
4 {e_flags} = 120
Remember above we had 107 left over, and now we're at 168. This is of
course the *best* case scenario -- you'll also want to have NUL bytes
in the magic & mask too to match literal zeros.
Note: the reason we can use 24 in the magic is that we can work off of the
fact that for bytes the mask would clobber, we can stuff any value into
magic that we want. So when mask is \x00, we don't need the magic to also
be \x00, it can be an unescaped raw byte like '!'. This lets us handle
more formats (barely) under the current 256 limit, but that's a pretty
tall hoop to force people to jump through.
With all that said, let's bump the limit from 256 bytes to 1920. This way
we support escaping every byte of the mask & magic field (which is 1024
bytes by themselves -- 128 * 4 * 2), and we leave plenty of room for other
fields. Like long paths to the interpreter (when you have source in your
/really/long/homedir/qemu/foo). Since the current code stuffs more than
one structure into the same buffer, we leave a bit of space to easily
round up to 2k. 1920 is just as arbitrary as 256 ;).
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-10-13 15:52:03 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:20 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-14 21:58:08 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:14 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
2026-07-14 21:58:08 +02:00
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:06 +02:00
|
|
|
|
2026-07-14 21:58:08 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:10 +02:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:14 +02:00
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:10 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2014-12-10 15:52:08 -08:00
|
|
|
|
2026-07-14 21:58:08 +02:00
|
|
|
|
|
|
|
|
|
binfmt_misc: let a register string create an entry disabled
An entry is matchable as soon as it is registered. create_entry() sets
the enabled bit for every type and add_entry() links it straight into the
instance, so everything an entry needs has to fit in the write that
creates it.
Add a 'D' flag. The entry is created disabled and has to be enabled by
writing '1' to its entry file before it can match anything. That splits a
registration into create and activate, which a later patch uses to
configure an entry beyond what one register string can carry. It is
useful on its own too. Entries can be staged without dispatching the
moment they are written.
A staged entry stays out of the search list entirely. add_entry() only
hashes an entry that is born matchable, and the first '1' written to
the entry file hashes a staged one, which takes its place in the search
order at that point. The rcu insertion publishes the fully configured
entry, so the exec side keeps the plain enabled test it always had.
Removal cannot rely on the search list anymore. Whether an entry was
already removed is now decided by its dentry, '-1' to the status file
walks the directory instead of the list so staged entries do not
survive it, and a '1' through a file handle held across a removal
publishes nothing.
'D' is consumed at registration and not recorded. What matters afterwards
is whether the entry is enabled, and the entry file already reports that.
A 'B' entry's flags field had to be empty so far because every flag it
could name shaped the invocation, which a bpf handler picks per exec with
bpf_binprm_set_flags(). 'D' shapes the registration instead. So the rule
becomes what it always meant: a 'B' entry carries no invocation flags,
and 'D' composes.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-1-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:03 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:08 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-14 21:58:08 +02:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-14 21:58:08 +02:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-14 21:58:08 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-14 21:58:08 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:10 +02:00
|
|
|
|
|
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
2026-07-14 21:58:08 +02:00
|
|
|
|
|
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
|
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
|
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-08-03 14:15:00 +02:00
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
|
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
|
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
2026-08-03 14:15:00 +02:00
|
|
|
|
|
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-08-03 14:15:00 +02:00
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-08-03 14:15:00 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
2026-09-02 15:31:14 -07:00
|
|
|
|
|
|
|
|
|
2026-08-03 14:15:00 +02:00
|
|
|
|
|
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
2026-08-03 14:15:00 +02:00
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-08-03 14:15:00 +02:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
|
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-14 21:58:08 +02:00
|
|
|
|
|
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:10 +02:00
|
|
|
|
|
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
2026-07-10 11:33:10 +02:00
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
|
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
2026-07-10 11:33:10 +02:00
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
2026-07-14 21:58:08 +02:00
|
|
|
|
|
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
2026-07-10 11:33:16 +02:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
2026-07-10 11:33:15 +02:00
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
2026-07-10 11:33:15 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:15 +02:00
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:15 +02:00
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
|
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
|
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-14 21:58:11 +02:00
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-14 21:58:11 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
|
|
|
|
|
2026-07-14 21:58:11 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
2026-07-14 21:58:11 +02:00
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
|
|
|
|
|
2026-07-21 16:13:46 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-21 16:13:54 +02:00
|
|
|
|
|
|
|
|
|
2026-07-21 16:14:01 +02:00
|
|
|
|
|
|
|
|
|
2026-07-21 16:13:46 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-21 16:13:47 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
2026-07-21 16:13:47 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
|
|
|
|
|
2026-07-21 16:13:47 +02:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-21 16:13:47 +02:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
2026-07-21 16:13:47 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-21 16:13:47 +02:00
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
2026-07-10 11:33:16 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-21 16:13:47 +02:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
2026-07-21 16:13:47 +02:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-21 16:13:47 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
syscalls: implement execveat() system call
This patchset adds execveat(2) for x86, and is derived from Meredydd
Luff's patch from Sept 2012 (https://lkml.org/lkml/2012/9/11/528).
The primary aim of adding an execveat syscall is to allow an
implementation of fexecve(3) that does not rely on the /proc filesystem,
at least for executables (rather than scripts). The current glibc version
of fexecve(3) is implemented via /proc, which causes problems in sandboxed
or otherwise restricted environments.
Given the desire for a /proc-free fexecve() implementation, HPA suggested
(https://lkml.org/lkml/2006/7/11/556) that an execveat(2) syscall would be
an appropriate generalization.
Also, having a new syscall means that it can take a flags argument without
back-compatibility concerns. The current implementation just defines the
AT_EMPTY_PATH and AT_SYMLINK_NOFOLLOW flags, but other flags could be
added in future -- for example, flags for new namespaces (as suggested at
https://lkml.org/lkml/2006/7/11/474).
Related history:
- https://lkml.org/lkml/2006/12/27/123 is an example of someone
realizing that fexecve() is likely to fail in a chroot environment.
- http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=514043 covered
documenting the /proc requirement of fexecve(3) in its manpage, to
"prevent other people from wasting their time".
- https://bugzilla.redhat.com/show_bug.cgi?id=241609 described a
problem where a process that did setuid() could not fexecve()
because it no longer had access to /proc/self/fd; this has since
been fixed.
This patch (of 4):
Add a new execveat(2) system call. execveat() is to execve() as openat()
is to open(): it takes a file descriptor that refers to a directory, and
resolves the filename relative to that.
In addition, if the filename is empty and AT_EMPTY_PATH is specified,
execveat() executes the file to which the file descriptor refers. This
replicates the functionality of fexecve(), which is a system call in other
UNIXen, but in Linux glibc it depends on opening "/proc/self/fd/<fd>" (and
so relies on /proc being mounted).
The filename fed to the executed program as argv[0] (or the name of the
script fed to a script interpreter) will be of the form "/dev/fd/<fd>"
(for an empty filename) or "/dev/fd/<fd>/<filename>", effectively
reflecting how the executable was found. This does however mean that
execution of a script in a /proc-less environment won't work; also, script
execution via an O_CLOEXEC file descriptor fails (as the file will not be
accessible after exec).
Based on patches by Meredydd Luff.
Signed-off-by: David Drysdale <drysdale@google.com>
Cc: Meredydd Luff <meredydd@senatehouse.org>
Cc: Shuah Khan <shuah.kh@samsung.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Rich Felker <dalias@aerifal.cx>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-12-12 16:57:29 -08:00
|
|
|
|
2026-07-21 16:13:47 +02:00
|
|
|
|
|
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
2026-07-21 16:13:47 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: let a bpf handler choose the invocation flags per exec
The 'P', 'C' and 'O' flags of a binfmt_misc entry - preserve argv[0],
compute credentials from the binary, and pass the binary as an open file
descriptor - are fixed at registration and apply to every binary the entry
matches. A bpf handler matches, selects the interpreter and reads the
binary per exec, so the flags should be its per-exec decision too: one
handler may match both setuid and non-setuid binaries, argv[0]-sensitive
ones and not.
Honor the flags the load program stages in bprm->bpf_flags through the
bpf_binprm_set_flags() kfunc: BPF_BINPRM_PRESERVE_ARGV0,
BPF_BINPRM_CREDENTIALS and BPF_BINPRM_EXECFD map to 'P', 'C' and 'O' and
keep the semantics of their static counterparts, credentials implying the
open file descriptor included.
Flags staged by a load program that then fails are dropped on the way out
so they cannot leak into a later handler's exec, and the argv[0] decision
acts on the entry's own choice instead of testing the accumulated
bprm->interp_flags bit, which an earlier chain level may have left set and
binfmt_misc never clears.
Since a 'B' entry's flags come from the program, it carries none in the
register string: 'P', 'C' and 'O' are rejected there alongside 'F', which
was already meaningless for it. load_misc_binary() takes the flags from
the entry for a static handler and from bprm->bpf_flags for a bpf one.
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-7-57b7529c002c@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:12 +02:00
|
|
|
|
2026-07-21 16:13:52 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: let a bpf handler choose the invocation flags per exec
The 'P', 'C' and 'O' flags of a binfmt_misc entry - preserve argv[0],
compute credentials from the binary, and pass the binary as an open file
descriptor - are fixed at registration and apply to every binary the entry
matches. A bpf handler matches, selects the interpreter and reads the
binary per exec, so the flags should be its per-exec decision too: one
handler may match both setuid and non-setuid binaries, argv[0]-sensitive
ones and not.
Honor the flags the load program stages in bprm->bpf_flags through the
bpf_binprm_set_flags() kfunc: BPF_BINPRM_PRESERVE_ARGV0,
BPF_BINPRM_CREDENTIALS and BPF_BINPRM_EXECFD map to 'P', 'C' and 'O' and
keep the semantics of their static counterparts, credentials implying the
open file descriptor included.
Flags staged by a load program that then fails are dropped on the way out
so they cannot leak into a later handler's exec, and the argv[0] decision
acts on the entry's own choice instead of testing the accumulated
bprm->interp_flags bit, which an earlier chain level may have left set and
binfmt_misc never clears.
Since a 'B' entry's flags come from the program, it carries none in the
register string: 'P', 'C' and 'O' are rejected there alongside 'F', which
was already meaningless for it. load_misc_binary() takes the flags from
the entry for a static handler and from bprm->bpf_flags for a bpf one.
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-7-57b7529c002c@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:12 +02:00
|
|
|
|
2026-07-21 16:13:46 +02:00
|
|
|
|
2020-01-28 14:25:39 +01:00
|
|
|
|
|
|
|
|
|
2007-07-19 01:48:16 -07:00
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:16 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2026-07-14 21:58:11 +02:00
|
|
|
|
2020-06-04 16:51:14 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:16 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2026-07-14 21:58:11 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:16 +02:00
|
|
|
|
2014-12-10 15:52:10 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-21 16:13:47 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-10-20 22:00:48 -04:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-21 16:13:47 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
2026-07-21 16:13:47 +02:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-21 16:14:00 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-21 16:13:47 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-21 16:13:47 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-21 16:13:47 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
syscalls: implement execveat() system call
This patchset adds execveat(2) for x86, and is derived from Meredydd
Luff's patch from Sept 2012 (https://lkml.org/lkml/2012/9/11/528).
The primary aim of adding an execveat syscall is to allow an
implementation of fexecve(3) that does not rely on the /proc filesystem,
at least for executables (rather than scripts). The current glibc version
of fexecve(3) is implemented via /proc, which causes problems in sandboxed
or otherwise restricted environments.
Given the desire for a /proc-free fexecve() implementation, HPA suggested
(https://lkml.org/lkml/2006/7/11/556) that an execveat(2) syscall would be
an appropriate generalization.
Also, having a new syscall means that it can take a flags argument without
back-compatibility concerns. The current implementation just defines the
AT_EMPTY_PATH and AT_SYMLINK_NOFOLLOW flags, but other flags could be
added in future -- for example, flags for new namespaces (as suggested at
https://lkml.org/lkml/2006/7/11/474).
Related history:
- https://lkml.org/lkml/2006/12/27/123 is an example of someone
realizing that fexecve() is likely to fail in a chroot environment.
- http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=514043 covered
documenting the /proc requirement of fexecve(3) in its manpage, to
"prevent other people from wasting their time".
- https://bugzilla.redhat.com/show_bug.cgi?id=241609 described a
problem where a process that did setuid() could not fexecve()
because it no longer had access to /proc/self/fd; this has since
been fixed.
This patch (of 4):
Add a new execveat(2) system call. execveat() is to execve() as openat()
is to open(): it takes a file descriptor that refers to a directory, and
resolves the filename relative to that.
In addition, if the filename is empty and AT_EMPTY_PATH is specified,
execveat() executes the file to which the file descriptor refers. This
replicates the functionality of fexecve(), which is a system call in other
UNIXen, but in Linux glibc it depends on opening "/proc/self/fd/<fd>" (and
so relies on /proc being mounted).
The filename fed to the executed program as argv[0] (or the name of the
script fed to a script interpreter) will be of the form "/dev/fd/<fd>"
(for an empty filename) or "/dev/fd/<fd>/<filename>", effectively
reflecting how the executable was found. This does however mean that
execution of a script in a /proc-less environment won't work; also, script
execution via an O_CLOEXEC file descriptor fails (as the file will not be
accessible after exec).
Based on patches by Meredydd Luff.
Signed-off-by: David Drysdale <drysdale@google.com>
Cc: Meredydd Luff <meredydd@senatehouse.org>
Cc: Shuah Khan <shuah.kh@samsung.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Rich Felker <dalias@aerifal.cx>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-12-12 16:57:29 -08:00
|
|
|
|
2026-07-21 16:13:47 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-21 16:13:52 +02:00
|
|
|
|
2026-07-21 16:14:00 +02:00
|
|
|
|
|
|
|
|
|
2026-07-21 16:13:52 +02:00
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
|
|
|
|
|
2026-07-21 16:14:00 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
2026-07-21 16:14:00 +02:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
2026-07-21 16:14:00 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
2026-07-21 16:14:00 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-21 16:13:52 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
exec: do not leave bprm->interp on stack
If a series of scripts are executed, each triggering module loading via
unprintable bytes in the script header, kernel stack contents can leak
into the command line.
Normally execution of binfmt_script and binfmt_misc happens recursively.
However, when modules are enabled, and unprintable bytes exist in the
bprm->buf, execution will restart after attempting to load matching
binfmt modules. Unfortunately, the logic in binfmt_script and
binfmt_misc does not expect to get restarted. They leave bprm->interp
pointing to their local stack. This means on restart bprm->interp is
left pointing into unused stack memory which can then be copied into the
userspace argv areas.
After additional study, it seems that both recursion and restart remains
the desirable way to handle exec with scripts, misc, and modules. As
such, we need to protect the changes to interp.
This changes the logic to require allocation for any changes to the
bprm->interp. To avoid adding a new kmalloc to every exec, the default
value is left as-is. Only when passing through binfmt_script or
binfmt_misc does an allocation take place.
For a proof of concept, see DoTest.sh from:
http://www.halfdog.net/Security/2012/LinuxKernelBinfmtScriptStackDataDisclosure/
Signed-off-by: Kees Cook <keescook@chromium.org>
Cc: halfdog <me@halfdog.net>
Cc: P J P <ppandit@redhat.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-12-20 15:05:16 -08:00
|
|
|
|
2026-07-10 11:33:16 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
2014-12-10 15:52:10 -08:00
|
|
|
|
2026-07-10 11:33:16 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-21 16:13:52 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2020-05-18 18:43:20 -05:00
|
|
|
|
2026-07-21 16:13:46 +02:00
|
|
|
|
2026-07-20 14:36:49 +02:00
|
|
|
|
2026-07-21 16:13:46 +02:00
|
|
|
|
2020-05-29 22:00:54 -05:00
|
|
|
|
2026-07-10 11:33:16 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:23 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:23 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-21 16:13:45 +02:00
|
|
|
|
2026-07-10 11:33:19 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:19 +02:00
|
|
|
|
2026-07-21 16:13:45 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-21 16:13:45 +02:00
|
|
|
|
2026-07-10 11:33:19 +02:00
|
|
|
|
2026-07-21 16:13:45 +02:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2014-12-10 15:52:10 -08:00
|
|
|
|
2026-07-10 11:33:21 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-12-10 15:52:10 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:10 +02:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:21 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-12-10 15:52:08 -08:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
binfmt_misc: expand the register format limit to 1920 bytes
The current code places a 256 byte limit on the registration format.
This ends up being fairly limited when you try to do matching against a
binary format like ELF:
- the magic & mask formats cannot have any embedded NUL chars
(string_unescape_inplace halts at the first NUL)
- each escape sequence quadruples the size: \x00 is needed for NUL
- trying to match bytes at the start of the file as well as further
on leads to a lot of \x00 sequences in the mask
- magic & mask have to be the same length (when decoded)
- still need bytes for the other fields
- impossible!
Let's look at a concrete (and common) example: using QEMU to run MIPS
ELFs. The name field uses 11 bytes "qemu-mipsel". The interp uses 20
bytes "/usr/bin/qemu-mipsel". The type & flags takes up 4 bytes. We
need 7 bytes for the delimiter (usually ":"). We can skip offset. So
already we're down to 107 bytes to use with the magic/mask instead of
the real limit of 128 (BINPRM_BUF_SIZE). If people use shell code to
register (which they do the majority of the time), they're down to ~26
possible bytes since the escape sequence must be \x##.
The ELF format looks like (both 32 & 64 bit):
e_ident: 16 bytes
e_type: 2 bytes
e_machine: 2 bytes
Those 20 bytes are enough for most architectures because they have so few
formats in the first place, thus they can be uniquely identified. That
also means for shell users, since 20 is smaller than 26, they can sanely
register a handler.
But for some targets (like MIPS), we need to poke further. The ELF fields
continue on:
e_entry: 4 or 8 bytes
e_phoff: 4 or 8 bytes
e_shoff: 4 or 8 bytes
e_flags: 4 bytes
We only care about e_flags here as that includes the bits to identify
whether the ELF is O32/N32/N64. But now we have to consume another 16
bytes (for 32 bit ELFs) or 28 bytes (for 64 bit ELFs) just to match the
flags. If every byte is escaped, we send 288 more bytes to the kernel
((20 {e_ident,e_type,e_machine} + 12 {e_entry,e_phoff,e_shoff} + 4
{e_flags}) * 2 {mask,magic} * 4 {escape}) and we've clearly blown our
budget.
Even if we try to be clever and do the decoding ourselves (rather than
relying on the kernel to process \x##), we still can't hit the mark --
string_unescape_inplace treats mask & magic as C strings so NUL cannot
be embedded. That leaves us with having to pass \x00 for the 12/24
entry/phoff/shoff bytes (as those will be completely random addresses),
and that is a minimum requirement of 48/96 bytes for the mask alone.
Add up the rest and we blow through it (this is for 64 bit ELFs):
magic: 20 {e_ident,e_type,e_machine} + 24 {e_entry,e_phoff,e_shoff} +
4 {e_flags} = 48 # ^^ See note below.
mask: 20 {e_ident,e_type,e_machine} + 96 {e_entry,e_phoff,e_shoff} +
4 {e_flags} = 120
Remember above we had 107 left over, and now we're at 168. This is of
course the *best* case scenario -- you'll also want to have NUL bytes
in the magic & mask too to match literal zeros.
Note: the reason we can use 24 in the magic is that we can work off of the
fact that for bytes the mask would clobber, we can stuff any value into
magic that we want. So when mask is \x00, we don't need the magic to also
be \x00, it can be an unescaped raw byte like '!'. This lets us handle
more formats (barely) under the current 256 limit, but that's a pretty
tall hoop to force people to jump through.
With all that said, let's bump the limit from 256 bytes to 1920. This way
we support escaping every byte of the mask & magic field (which is 1024
bytes by themselves -- 128 * 4 * 2), and we leave plenty of room for other
fields. Like long paths to the interpreter (when you have source in your
/really/long/homedir/qemu/foo). Since the current code stuffs more than
one structure into the same buffer, we leave a bit of space to easily
round up to 2k. 1920 is just as arbitrary as 256 ;).
Signed-off-by: Mike Frysinger <vapier@gentoo.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-10-13 15:52:03 -07:00
|
|
|
|
2026-07-10 11:33:21 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-09-02 15:31:14 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:21 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:20 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:10 +02:00
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:21 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:23 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2014-12-10 15:52:08 -08:00
|
|
|
|
|
|
|
|
|
binfmt_misc: reject a flag character as the field delimiter
The registration string starts with a user chosen delimiter that
separates the individual fields. So that the field parsers terminate
even on a truncated string create_entry() pads the buffer with that
same delimiter:
memset(buf + count, del, 8);
Most fields are scanned for the delimiter with strchr()/scanarg() and
happily stop on the padding. The flags field is different: instead of
scanning for the delimiter check_special_flags() consumes the flag
characters 'P', 'O', 'C' and 'F' and stops at the first byte that is
none of them, relying on the trailing delimiter to end the scan.
If the delimiter is itself a flag character the padding no longer acts
as a terminator. The scan swallows all eight padding bytes and keeps
reading past the end of the allocation until it hits a byte that is
not a flag character. For example registering
PaPEPPxPPiP
with 'P' as the delimiter (name "a", type extension, magic "x",
interpreter "i", empty flags) leaves the flag scan running off the end
of the buffer. The registration is rejected in the end because the
parser does not stop exactly at buf + count, but only after the out of
bounds read has already happened. With an unlucky allocation layout the
scan can walk into an unmapped page; under KASAN it is reported as a
slab out of bounds read. binfmt_misc mounts are available to
unprivileged users in a user namespace so the read is reachable without
privileges.
Reject a delimiter that is one of the flag characters up front. Such a
registration was always rejected anyway, only after the out of bounds
read, so no valid registration string changes meaning.
Link: https://patch.msgid.link/20260710-work-binfmt_misc-locking-v3-3-a162f7cb58d6@kernel.org
Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Cc: stable@vger.kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-10 11:33:04 +02:00
|
|
|
|
2026-07-21 16:13:45 +02:00
|
|
|
|
2026-07-10 11:33:21 +02:00
|
|
|
|
binfmt_misc: reject a flag character as the field delimiter
The registration string starts with a user chosen delimiter that
separates the individual fields. So that the field parsers terminate
even on a truncated string create_entry() pads the buffer with that
same delimiter:
memset(buf + count, del, 8);
Most fields are scanned for the delimiter with strchr()/scanarg() and
happily stop on the padding. The flags field is different: instead of
scanning for the delimiter check_special_flags() consumes the flag
characters 'P', 'O', 'C' and 'F' and stops at the first byte that is
none of them, relying on the trailing delimiter to end the scan.
If the delimiter is itself a flag character the padding no longer acts
as a terminator. The scan swallows all eight padding bytes and keeps
reading past the end of the allocation until it hits a byte that is
not a flag character. For example registering
PaPEPPxPPiP
with 'P' as the delimiter (name "a", type extension, magic "x",
interpreter "i", empty flags) leaves the flag scan running off the end
of the buffer. The registration is rejected in the end because the
parser does not stop exactly at buf + count, but only after the out of
bounds read has already happened. With an unlucky allocation layout the
scan can walk into an unmapped page; under KASAN it is reported as a
slab out of bounds read. binfmt_misc mounts are available to
unprivileged users in a user namespace so the read is reachable without
privileges.
Reject a delimiter that is one of the flag characters up front. Such a
registration was always rejected anyway, only after the out of bounds
read, so no valid registration string changes meaning.
Link: https://patch.msgid.link/20260710-work-binfmt_misc-locking-v3-3-a162f7cb58d6@kernel.org
Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Cc: stable@vger.kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-10 11:33:04 +02:00
|
|
|
|
2014-12-10 15:52:08 -08:00
|
|
|
|
2026-07-10 11:33:20 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2014-12-10 15:52:08 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:21 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:21 +02:00
|
|
|
|
2014-12-10 15:52:08 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2014-12-10 15:52:08 -08:00
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:08 +02:00
|
|
|
|
2014-12-10 15:52:08 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:08 +02:00
|
|
|
|
2014-12-10 15:52:08 -08:00
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-12-10 15:52:08 -08:00
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:21 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:21 +02:00
|
|
|
|
2014-12-10 15:52:08 -08:00
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:21 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-12-10 15:52:08 -08:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:21 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:21 +02:00
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2014-12-10 15:52:08 -08:00
|
|
|
|
2014-12-10 15:52:10 -08:00
|
|
|
|
|
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
binfmt_misc: let a bpf handler choose the invocation flags per exec
The 'P', 'C' and 'O' flags of a binfmt_misc entry - preserve argv[0],
compute credentials from the binary, and pass the binary as an open file
descriptor - are fixed at registration and apply to every binary the entry
matches. A bpf handler matches, selects the interpreter and reads the
binary per exec, so the flags should be its per-exec decision too: one
handler may match both setuid and non-setuid binaries, argv[0]-sensitive
ones and not.
Honor the flags the load program stages in bprm->bpf_flags through the
bpf_binprm_set_flags() kfunc: BPF_BINPRM_PRESERVE_ARGV0,
BPF_BINPRM_CREDENTIALS and BPF_BINPRM_EXECFD map to 'P', 'C' and 'O' and
keep the semantics of their static counterparts, credentials implying the
open file descriptor included.
Flags staged by a load program that then fails are dropped on the way out
so they cannot leak into a later handler's exec, and the argv[0] decision
acts on the entry's own choice instead of testing the accumulated
bprm->interp_flags bit, which an earlier chain level may have left set and
binfmt_misc never clears.
Since a 'B' entry's flags come from the program, it carries none in the
register string: 'P', 'C' and 'O' are rejected there alongside 'F', which
was already meaningless for it. load_misc_binary() takes the flags from
the entry for a static handler and from bprm->bpf_flags for a bpf one.
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-7-57b7529c002c@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:12 +02:00
|
|
|
|
2026-07-21 16:13:45 +02:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
binfmt_misc: let a register string create an entry disabled
An entry is matchable as soon as it is registered. create_entry() sets
the enabled bit for every type and add_entry() links it straight into the
instance, so everything an entry needs has to fit in the write that
creates it.
Add a 'D' flag. The entry is created disabled and has to be enabled by
writing '1' to its entry file before it can match anything. That splits a
registration into create and activate, which a later patch uses to
configure an entry beyond what one register string can carry. It is
useful on its own too. Entries can be staged without dispatching the
moment they are written.
A staged entry stays out of the search list entirely. add_entry() only
hashes an entry that is born matchable, and the first '1' written to
the entry file hashes a staged one, which takes its place in the search
order at that point. The rcu insertion publishes the fully configured
entry, so the exec side keeps the plain enabled test it always had.
Removal cannot rely on the search list anymore. Whether an entry was
already removed is now decided by its dentry, '-1' to the status file
walks the directory instead of the list so staged entries do not
survive it, and a '1' through a file handle held across a removal
publishes nothing.
'D' is consumed at registration and not recorded. What matters afterwards
is whether the entry is enabled, and the entry file already reports that.
A 'B' entry's flags field had to be empty so far because every flag it
could name shaped the invocation, which a bpf handler picks per exec with
bpf_binprm_set_flags(). 'D' shapes the registration instead. So the rule
becomes what it always meant: a 'B' entry carries no invocation flags,
and 'D' composes.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-1-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:03 +02:00
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
binfmt_misc: let a register string create an entry disabled
An entry is matchable as soon as it is registered. create_entry() sets
the enabled bit for every type and add_entry() links it straight into the
instance, so everything an entry needs has to fit in the write that
creates it.
Add a 'D' flag. The entry is created disabled and has to be enabled by
writing '1' to its entry file before it can match anything. That splits a
registration into create and activate, which a later patch uses to
configure an entry beyond what one register string can carry. It is
useful on its own too. Entries can be staged without dispatching the
moment they are written.
A staged entry stays out of the search list entirely. add_entry() only
hashes an entry that is born matchable, and the first '1' written to
the entry file hashes a staged one, which takes its place in the search
order at that point. The rcu insertion publishes the fully configured
entry, so the exec side keeps the plain enabled test it always had.
Removal cannot rely on the search list anymore. Whether an entry was
already removed is now decided by its dentry, '-1' to the status file
walks the directory instead of the list so staged entries do not
survive it, and a '1' through a file handle held across a removal
publishes nothing.
'D' is consumed at registration and not recorded. What matters afterwards
is whether the entry is enabled, and the entry file already reports that.
A 'B' entry's flags field had to be empty so far because every flag it
could name shaped the invocation, which a bpf handler picks per exec with
bpf_binprm_set_flags(). 'D' shapes the registration instead. So the rule
becomes what it always meant: a 'B' entry carries no invocation flags,
and 'D' composes.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-1-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:03 +02:00
|
|
|
|
|
|
|
|
|
2026-07-21 16:13:45 +02:00
|
|
|
|
|
|
|
|
|
binfmt_misc: let a register string create an entry disabled
An entry is matchable as soon as it is registered. create_entry() sets
the enabled bit for every type and add_entry() links it straight into the
instance, so everything an entry needs has to fit in the write that
creates it.
Add a 'D' flag. The entry is created disabled and has to be enabled by
writing '1' to its entry file before it can match anything. That splits a
registration into create and activate, which a later patch uses to
configure an entry beyond what one register string can carry. It is
useful on its own too. Entries can be staged without dispatching the
moment they are written.
A staged entry stays out of the search list entirely. add_entry() only
hashes an entry that is born matchable, and the first '1' written to
the entry file hashes a staged one, which takes its place in the search
order at that point. The rcu insertion publishes the fully configured
entry, so the exec side keeps the plain enabled test it always had.
Removal cannot rely on the search list anymore. Whether an entry was
already removed is now decided by its dentry, '-1' to the status file
walks the directory instead of the list so staged entries do not
survive it, and a '1' through a file handle held across a removal
publishes nothing.
'D' is consumed at registration and not recorded. What matters afterwards
is whether the entry is enabled, and the entry file already reports that.
A 'B' entry's flags field had to be empty so far because every flag it
could name shaped the invocation, which a bpf handler picks per exec with
bpf_binprm_set_flags(). 'D' shapes the registration instead. So the rule
becomes what it always meant: a 'B' entry carries no invocation flags,
and 'D' composes.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-1-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:03 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
|
|
|
|
|
binfmt_misc: let a register string create an entry disabled
An entry is matchable as soon as it is registered. create_entry() sets
the enabled bit for every type and add_entry() links it straight into the
instance, so everything an entry needs has to fit in the write that
creates it.
Add a 'D' flag. The entry is created disabled and has to be enabled by
writing '1' to its entry file before it can match anything. That splits a
registration into create and activate, which a later patch uses to
configure an entry beyond what one register string can carry. It is
useful on its own too. Entries can be staged without dispatching the
moment they are written.
A staged entry stays out of the search list entirely. add_entry() only
hashes an entry that is born matchable, and the first '1' written to
the entry file hashes a staged one, which takes its place in the search
order at that point. The rcu insertion publishes the fully configured
entry, so the exec side keeps the plain enabled test it always had.
Removal cannot rely on the search list anymore. Whether an entry was
already removed is now decided by its dentry, '-1' to the status file
walks the directory instead of the list so staged entries do not
survive it, and a '1' through a file handle held across a removal
publishes nothing.
'D' is consumed at registration and not recorded. What matters afterwards
is whether the entry is enabled, and the entry file already reports that.
A 'B' entry's flags field had to be empty so far because every flag it
could name shaped the invocation, which a bpf handler picks per exec with
bpf_binprm_set_flags(). 'D' shapes the registration instead. So the rule
becomes what it always meant: a 'B' entry carries no invocation flags,
and 'D' composes.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-1-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:03 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-21 16:13:53 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-21 16:14:00 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
2014-12-10 15:52:10 -08:00
|
|
|
|
2026-07-21 16:13:43 +02:00
|
|
|
|
2026-07-21 16:14:00 +02:00
|
|
|
|
2026-07-21 16:13:43 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:21 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:17 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:17 +02:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
binfmt_misc: work around gcc-4.9 warning
gcc-4.9 on ARM gives us a mysterious warning about the binfmt_misc
parse_command function:
fs/binfmt_misc.c: In function 'parse_command.part.3':
fs/binfmt_misc.c:405:7: warning: array subscript is above array bounds [-Warray-bounds]
I've managed to trace this back to the ARM implementation of memset,
which is called from copy_from_user in case of a fault and which does
#define memset(p,v,n) \
({ \
void *__p = (p); size_t __n = n; \
if ((__n) != 0) { \
if (__builtin_constant_p((v)) && (v) == 0) \
__memzero((__p),(__n)); \
else \
memset((__p),(v),(__n)); \
} \
(__p); \
})
Apparently gcc gets confused by the check for "size != 0" and believes
that the size might be zero when it gets to the line that does "if
(s[count-1] == '\n')", so it would access data outside of the array.
gcc is clearly wrong here, since this condition was already checked
earlier in the function and the 'size' value can not change in the
meantime.
Fortunately, we can work around it and get rid of the warning by
rearranging the function to check for zero size after doing the
copy_from_user. It is still safe to pass a zero size into
copy_from_user, so it does not cause any side effects.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-10-13 15:52:08 -07:00
|
|
|
|
2026-07-10 11:33:17 +02:00
|
|
|
|
2014-12-10 15:52:10 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:17 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:17 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:17 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
binfmt_misc: let a register string create an entry disabled
An entry is matchable as soon as it is registered. create_entry() sets
the enabled bit for every type and add_entry() links it straight into the
instance, so everything an entry needs has to fit in the write that
creates it.
Add a 'D' flag. The entry is created disabled and has to be enabled by
writing '1' to its entry file before it can match anything. That splits a
registration into create and activate, which a later patch uses to
configure an entry beyond what one register string can carry. It is
useful on its own too. Entries can be staged without dispatching the
moment they are written.
A staged entry stays out of the search list entirely. add_entry() only
hashes an entry that is born matchable, and the first '1' written to
the entry file hashes a staged one, which takes its place in the search
order at that point. The rcu insertion publishes the fully configured
entry, so the exec side keeps the plain enabled test it always had.
Removal cannot rely on the search list anymore. Whether an entry was
already removed is now decided by its dentry, '-1' to the status file
walks the directory instead of the list so staged entries do not
survive it, and a '1' through a file handle held across a removal
publishes nothing.
'D' is consumed at registration and not recorded. What matters afterwards
is whether the entry is enabled, and the entry file already reports that.
A 'B' entry's flags field had to be empty so far because every flag it
could name shaped the invocation, which a bpf handler picks per exec with
bpf_binprm_set_flags(). 'D' shapes the registration instead. So the rule
becomes what it always meant: a 'B' entry carries no invocation flags,
and 'D' composes.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-1-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:03 +02:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
binfmt_misc: let a register string create an entry disabled
An entry is matchable as soon as it is registered. create_entry() sets
the enabled bit for every type and add_entry() links it straight into the
instance, so everything an entry needs has to fit in the write that
creates it.
Add a 'D' flag. The entry is created disabled and has to be enabled by
writing '1' to its entry file before it can match anything. That splits a
registration into create and activate, which a later patch uses to
configure an entry beyond what one register string can carry. It is
useful on its own too. Entries can be staged without dispatching the
moment they are written.
A staged entry stays out of the search list entirely. add_entry() only
hashes an entry that is born matchable, and the first '1' written to
the entry file hashes a staged one, which takes its place in the search
order at that point. The rcu insertion publishes the fully configured
entry, so the exec side keeps the plain enabled test it always had.
Removal cannot rely on the search list anymore. Whether an entry was
already removed is now decided by its dentry, '-1' to the status file
walks the directory instead of the list so staged entries do not
survive it, and a '1' through a file handle held across a removal
publishes nothing.
'D' is consumed at registration and not recorded. What matters afterwards
is whether the entry is enabled, and the entry file already reports that.
A 'B' entry's flags field had to be empty so far because every flag it
could name shaped the invocation, which a bpf handler picks per exec with
bpf_binprm_set_flags(). 'D' shapes the registration instead. So the rule
becomes what it always meant: a 'B' entry carries no invocation flags,
and 'D' composes.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-1-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:03 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:13 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:08 +02:00
|
|
|
|
2026-07-10 11:33:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:13 +02:00
|
|
|
|
2026-07-21 16:13:45 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:13 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:13 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:13 +02:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:13 +02:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:13 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:13 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:23 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2014-12-10 15:52:10 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2010-10-23 11:19:54 -04:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2023-10-04 14:52:38 -04:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:25 +02:00
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:05 +02:00
|
|
|
|
|
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
2010-06-07 14:34:48 -04:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:10 +02:00
|
|
|
|
2017-10-03 16:15:48 -07:00
|
|
|
|
2012-05-03 14:48:02 +02:00
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
|
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:06 +02:00
|
|
|
|
2026-07-10 11:33:05 +02:00
|
|
|
|
2026-07-10 11:33:06 +02:00
|
|
|
|
|
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:25 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-14 21:58:08 +02:00
|
|
|
|
2026-07-10 11:33:25 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:18 +02:00
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
2026-07-10 11:33:10 +02:00
|
|
|
|
|
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
2026-07-10 11:33:25 +02:00
|
|
|
|
2025-03-08 19:54:09 -05:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:25 +02:00
|
|
|
|
2026-07-10 11:33:18 +02:00
|
|
|
|
|
|
|
|
|
binfmt_misc: let a register string create an entry disabled
An entry is matchable as soon as it is registered. create_entry() sets
the enabled bit for every type and add_entry() links it straight into the
instance, so everything an entry needs has to fit in the write that
creates it.
Add a 'D' flag. The entry is created disabled and has to be enabled by
writing '1' to its entry file before it can match anything. That splits a
registration into create and activate, which a later patch uses to
configure an entry beyond what one register string can carry. It is
useful on its own too. Entries can be staged without dispatching the
moment they are written.
A staged entry stays out of the search list entirely. add_entry() only
hashes an entry that is born matchable, and the first '1' written to
the entry file hashes a staged one, which takes its place in the search
order at that point. The rcu insertion publishes the fully configured
entry, so the exec side keeps the plain enabled test it always had.
Removal cannot rely on the search list anymore. Whether an entry was
already removed is now decided by its dentry, '-1' to the status file
walks the directory instead of the list so staged entries do not
survive it, and a '1' through a file handle held across a removal
publishes nothing.
'D' is consumed at registration and not recorded. What matters afterwards
is whether the entry is enabled, and the entry file already reports that.
A 'B' entry's flags field had to be empty so far because every flag it
could name shaped the invocation, which a bpf handler picks per exec with
bpf_binprm_set_flags(). 'D' shapes the registration instead. So the rule
becomes what it always meant: a 'B' entry carries no invocation flags,
and 'D' composes.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-1-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:03 +02:00
|
|
|
|
2026-07-10 11:33:18 +02:00
|
|
|
|
|
|
|
|
|
binfmt_misc: let a register string create an entry disabled
An entry is matchable as soon as it is registered. create_entry() sets
the enabled bit for every type and add_entry() links it straight into the
instance, so everything an entry needs has to fit in the write that
creates it.
Add a 'D' flag. The entry is created disabled and has to be enabled by
writing '1' to its entry file before it can match anything. That splits a
registration into create and activate, which a later patch uses to
configure an entry beyond what one register string can carry. It is
useful on its own too. Entries can be staged without dispatching the
moment they are written.
A staged entry stays out of the search list entirely. add_entry() only
hashes an entry that is born matchable, and the first '1' written to
the entry file hashes a staged one, which takes its place in the search
order at that point. The rcu insertion publishes the fully configured
entry, so the exec side keeps the plain enabled test it always had.
Removal cannot rely on the search list anymore. Whether an entry was
already removed is now decided by its dentry, '-1' to the status file
walks the directory instead of the list so staged entries do not
survive it, and a '1' through a file handle held across a removal
publishes nothing.
'D' is consumed at registration and not recorded. What matters afterwards
is whether the entry is enabled, and the entry file already reports that.
A 'B' entry's flags field had to be empty so far because every flag it
could name shaped the invocation, which a bpf handler picks per exec with
bpf_binprm_set_flags(). 'D' shapes the registration instead. So the rule
becomes what it always meant: a 'B' entry carries no invocation flags,
and 'D' composes.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-1-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:03 +02:00
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:18 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: let a register string create an entry disabled
An entry is matchable as soon as it is registered. create_entry() sets
the enabled bit for every type and add_entry() links it straight into the
instance, so everything an entry needs has to fit in the write that
creates it.
Add a 'D' flag. The entry is created disabled and has to be enabled by
writing '1' to its entry file before it can match anything. That splits a
registration into create and activate, which a later patch uses to
configure an entry beyond what one register string can carry. It is
useful on its own too. Entries can be staged without dispatching the
moment they are written.
A staged entry stays out of the search list entirely. add_entry() only
hashes an entry that is born matchable, and the first '1' written to
the entry file hashes a staged one, which takes its place in the search
order at that point. The rcu insertion publishes the fully configured
entry, so the exec side keeps the plain enabled test it always had.
Removal cannot rely on the search list anymore. Whether an entry was
already removed is now decided by its dentry, '-1' to the status file
walks the directory instead of the list so staged entries do not
survive it, and a '1' through a file handle held across a removal
publishes nothing.
'D' is consumed at registration and not recorded. What matters afterwards
is whether the entry is enabled, and the entry file already reports that.
A 'B' entry's flags field had to be empty so far because every flag it
could name shaped the invocation, which a bpf handler picks per exec with
bpf_binprm_set_flags(). 'D' shapes the registration instead. So the rule
becomes what it always meant: a 'B' entry carries no invocation flags,
and 'D' composes.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-1-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:03 +02:00
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:18 +02:00
|
|
|
|
|
|
|
|
|
binfmt_misc: let a register string create an entry disabled
An entry is matchable as soon as it is registered. create_entry() sets
the enabled bit for every type and add_entry() links it straight into the
instance, so everything an entry needs has to fit in the write that
creates it.
Add a 'D' flag. The entry is created disabled and has to be enabled by
writing '1' to its entry file before it can match anything. That splits a
registration into create and activate, which a later patch uses to
configure an entry beyond what one register string can carry. It is
useful on its own too. Entries can be staged without dispatching the
moment they are written.
A staged entry stays out of the search list entirely. add_entry() only
hashes an entry that is born matchable, and the first '1' written to
the entry file hashes a staged one, which takes its place in the search
order at that point. The rcu insertion publishes the fully configured
entry, so the exec side keeps the plain enabled test it always had.
Removal cannot rely on the search list anymore. Whether an entry was
already removed is now decided by its dentry, '-1' to the status file
walks the directory instead of the list so staged entries do not
survive it, and a '1' through a file handle held across a removal
publishes nothing.
'D' is consumed at registration and not recorded. What matters afterwards
is whether the entry is enabled, and the entry file already reports that.
A 'B' entry's flags field had to be empty so far because every flag it
could name shaped the invocation, which a bpf handler picks per exec with
bpf_binprm_set_flags(). 'D' shapes the registration instead. So the rule
becomes what it always meant: a 'B' entry carries no invocation flags,
and 'D' composes.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-1-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:03 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:18 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:25 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:13 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:13 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:13 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2008-07-23 21:29:15 -07:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
2026-07-10 11:33:10 +02:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:17 +02:00
|
|
|
|
2026-07-10 11:33:08 +02:00
|
|
|
|
2014-12-10 15:52:10 -08:00
|
|
|
|
binfmt_misc: let a register string create an entry disabled
An entry is matchable as soon as it is registered. create_entry() sets
the enabled bit for every type and add_entry() links it straight into the
instance, so everything an entry needs has to fit in the write that
creates it.
Add a 'D' flag. The entry is created disabled and has to be enabled by
writing '1' to its entry file before it can match anything. That splits a
registration into create and activate, which a later patch uses to
configure an entry beyond what one register string can carry. It is
useful on its own too. Entries can be staged without dispatching the
moment they are written.
A staged entry stays out of the search list entirely. add_entry() only
hashes an entry that is born matchable, and the first '1' written to
the entry file hashes a staged one, which takes its place in the search
order at that point. The rcu insertion publishes the fully configured
entry, so the exec side keeps the plain enabled test it always had.
Removal cannot rely on the search list anymore. Whether an entry was
already removed is now decided by its dentry, '-1' to the status file
walks the directory instead of the list so staged entries do not
survive it, and a '1' through a file handle held across a removal
publishes nothing.
'D' is consumed at registration and not recorded. What matters afterwards
is whether the entry is enabled, and the entry file already reports that.
A 'B' entry's flags field had to be empty so far because every flag it
could name shaped the invocation, which a bpf handler picks per exec with
bpf_binprm_set_flags(). 'D' shapes the registration instead. So the rule
becomes what it always meant: a 'B' entry carries no invocation flags,
and 'D' composes.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-1-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:03 +02:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
binfmt_misc: let a register string create an entry disabled
An entry is matchable as soon as it is registered. create_entry() sets
the enabled bit for every type and add_entry() links it straight into the
instance, so everything an entry needs has to fit in the write that
creates it.
Add a 'D' flag. The entry is created disabled and has to be enabled by
writing '1' to its entry file before it can match anything. That splits a
registration into create and activate, which a later patch uses to
configure an entry beyond what one register string can carry. It is
useful on its own too. Entries can be staged without dispatching the
moment they are written.
A staged entry stays out of the search list entirely. add_entry() only
hashes an entry that is born matchable, and the first '1' written to
the entry file hashes a staged one, which takes its place in the search
order at that point. The rcu insertion publishes the fully configured
entry, so the exec side keeps the plain enabled test it always had.
Removal cannot rely on the search list anymore. Whether an entry was
already removed is now decided by its dentry, '-1' to the status file
walks the directory instead of the list so staged entries do not
survive it, and a '1' through a file handle held across a removal
publishes nothing.
'D' is consumed at registration and not recorded. What matters afterwards
is whether the entry is enabled, and the entry file already reports that.
A 'B' entry's flags field had to be empty so far because every flag it
could name shaped the invocation, which a bpf handler picks per exec with
bpf_binprm_set_flags(). 'D' shapes the registration instead. So the rule
becomes what it always meant: a 'B' entry carries no invocation flags,
and 'D' composes.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-1-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:03 +02:00
|
|
|
|
|
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-10-28 12:31:13 +02:00
|
|
|
|
binfmt_misc: let a register string create an entry disabled
An entry is matchable as soon as it is registered. create_entry() sets
the enabled bit for every type and add_entry() links it straight into the
instance, so everything an entry needs has to fit in the write that
creates it.
Add a 'D' flag. The entry is created disabled and has to be enabled by
writing '1' to its entry file before it can match anything. That splits a
registration into create and activate, which a later patch uses to
configure an entry beyond what one register string can carry. It is
useful on its own too. Entries can be staged without dispatching the
moment they are written.
A staged entry stays out of the search list entirely. add_entry() only
hashes an entry that is born matchable, and the first '1' written to
the entry file hashes a staged one, which takes its place in the search
order at that point. The rcu insertion publishes the fully configured
entry, so the exec side keeps the plain enabled test it always had.
Removal cannot rely on the search list anymore. Whether an entry was
already removed is now decided by its dentry, '-1' to the status file
walks the directory instead of the list so staged entries do not
survive it, and a '1' through a file handle held across a removal
publishes nothing.
'D' is consumed at registration and not recorded. What matters afterwards
is whether the entry is enabled, and the entry file already reports that.
A 'B' entry's flags field had to be empty so far because every flag it
could name shaped the invocation, which a bpf handler picks per exec with
bpf_binprm_set_flags(). 'D' shapes the registration instead. So the rule
becomes what it always meant: a 'B' entry carries no invocation flags,
and 'D' composes.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-1-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:03 +02:00
|
|
|
|
2026-07-10 11:33:08 +02:00
|
|
|
|
binfmt_misc: let a register string create an entry disabled
An entry is matchable as soon as it is registered. create_entry() sets
the enabled bit for every type and add_entry() links it straight into the
instance, so everything an entry needs has to fit in the write that
creates it.
Add a 'D' flag. The entry is created disabled and has to be enabled by
writing '1' to its entry file before it can match anything. That splits a
registration into create and activate, which a later patch uses to
configure an entry beyond what one register string can carry. It is
useful on its own too. Entries can be staged without dispatching the
moment they are written.
A staged entry stays out of the search list entirely. add_entry() only
hashes an entry that is born matchable, and the first '1' written to
the entry file hashes a staged one, which takes its place in the search
order at that point. The rcu insertion publishes the fully configured
entry, so the exec side keeps the plain enabled test it always had.
Removal cannot rely on the search list anymore. Whether an entry was
already removed is now decided by its dentry, '-1' to the status file
walks the directory instead of the list so staged entries do not
survive it, and a '1' through a file handle held across a removal
publishes nothing.
'D' is consumed at registration and not recorded. What matters afterwards
is whether the entry is enabled, and the entry file already reports that.
A 'B' entry's flags field had to be empty so far because every flag it
could name shaped the invocation, which a bpf handler picks per exec with
bpf_binprm_set_flags(). 'D' shapes the registration instead. So the rule
becomes what it always meant: a 'B' entry carries no invocation flags,
and 'D' composes.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-1-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:03 +02:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
binfmt_misc: let a register string create an entry disabled
An entry is matchable as soon as it is registered. create_entry() sets
the enabled bit for every type and add_entry() links it straight into the
instance, so everything an entry needs has to fit in the write that
creates it.
Add a 'D' flag. The entry is created disabled and has to be enabled by
writing '1' to its entry file before it can match anything. That splits a
registration into create and activate, which a later patch uses to
configure an entry beyond what one register string can carry. It is
useful on its own too. Entries can be staged without dispatching the
moment they are written.
A staged entry stays out of the search list entirely. add_entry() only
hashes an entry that is born matchable, and the first '1' written to
the entry file hashes a staged one, which takes its place in the search
order at that point. The rcu insertion publishes the fully configured
entry, so the exec side keeps the plain enabled test it always had.
Removal cannot rely on the search list anymore. Whether an entry was
already removed is now decided by its dentry, '-1' to the status file
walks the directory instead of the list so staged entries do not
survive it, and a '1' through a file handle held across a removal
publishes nothing.
'D' is consumed at registration and not recorded. What matters afterwards
is whether the entry is enabled, and the entry file already reports that.
A 'B' entry's flags field had to be empty so far because every flag it
could name shaped the invocation, which a bpf handler picks per exec with
bpf_binprm_set_flags(). 'D' shapes the registration instead. So the rule
becomes what it always meant: a 'B' entry carries no invocation flags,
and 'D' composes.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-1-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:03 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-12-10 15:52:10 -08:00
|
|
|
|
binfmt_misc: let a register string create an entry disabled
An entry is matchable as soon as it is registered. create_entry() sets
the enabled bit for every type and add_entry() links it straight into the
instance, so everything an entry needs has to fit in the write that
creates it.
Add a 'D' flag. The entry is created disabled and has to be enabled by
writing '1' to its entry file before it can match anything. That splits a
registration into create and activate, which a later patch uses to
configure an entry beyond what one register string can carry. It is
useful on its own too. Entries can be staged without dispatching the
moment they are written.
A staged entry stays out of the search list entirely. add_entry() only
hashes an entry that is born matchable, and the first '1' written to
the entry file hashes a staged one, which takes its place in the search
order at that point. The rcu insertion publishes the fully configured
entry, so the exec side keeps the plain enabled test it always had.
Removal cannot rely on the search list anymore. Whether an entry was
already removed is now decided by its dentry, '-1' to the status file
walks the directory instead of the list so staged entries do not
survive it, and a '1' through a file handle held across a removal
publishes nothing.
'D' is consumed at registration and not recorded. What matters afterwards
is whether the entry is enabled, and the entry file already reports that.
A 'B' entry's flags field had to be empty so far because every flag it
could name shaped the invocation, which a bpf handler picks per exec with
bpf_binprm_set_flags(). 'D' shapes the registration instead. So the rule
becomes what it always meant: a 'B' entry carries no invocation flags,
and 'D' composes.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-1-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:03 +02:00
|
|
|
|
2026-07-10 11:33:17 +02:00
|
|
|
|
2026-07-10 11:33:18 +02:00
|
|
|
|
2014-12-10 15:52:10 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2014-12-10 15:52:10 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2006-03-28 01:56:42 -08:00
|
|
|
|
2026-07-10 11:33:13 +02:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:13 +02:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-05-13 22:02:44 -06:00
|
|
|
|
2026-07-10 11:33:10 +02:00
|
|
|
|
2024-05-13 22:02:44 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: let a register string create an entry disabled
An entry is matchable as soon as it is registered. create_entry() sets
the enabled bit for every type and add_entry() links it straight into the
instance, so everything an entry needs has to fit in the write that
creates it.
Add a 'D' flag. The entry is created disabled and has to be enabled by
writing '1' to its entry file before it can match anything. That splits a
registration into create and activate, which a later patch uses to
configure an entry beyond what one register string can carry. It is
useful on its own too. Entries can be staged without dispatching the
moment they are written.
A staged entry stays out of the search list entirely. add_entry() only
hashes an entry that is born matchable, and the first '1' written to
the entry file hashes a staged one, which takes its place in the search
order at that point. The rcu insertion publishes the fully configured
entry, so the exec side keeps the plain enabled test it always had.
Removal cannot rely on the search list anymore. Whether an entry was
already removed is now decided by its dentry, '-1' to the status file
walks the directory instead of the list so staged entries do not
survive it, and a '1' through a file handle held across a removal
publishes nothing.
'D' is consumed at registration and not recorded. What matters afterwards
is whether the entry is enabled, and the entry file already reports that.
A 'B' entry's flags field had to be empty so far because every flag it
could name shaped the invocation, which a bpf handler picks per exec with
bpf_binprm_set_flags(). 'D' shapes the registration instead. So the rule
becomes what it always meant: a 'B' entry carries no invocation flags,
and 'D' composes.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-1-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:03 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-05-13 22:02:44 -06:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
2016-05-29 19:14:03 -04:00
|
|
|
|
2026-07-10 11:33:22 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: wire up bpf-backed 'B' entries
Activate a registered binfmt_misc_ops handler through the existing text
interface with the new 'B' entry type:
echo ':name:B::::<handler-name>:' > <binfmt_misc>/register
The offset, magic, and mask fields must be empty since the program does
the matching; the interpreter field carries the handler name since the
program supplies the interpreter. Reusing the register file keeps the
existing permission model intact: activating a handler requires the
same write access to a binfmt_misc instance as any other registration,
and the per user namespace instance semantics apply unchanged. A 'B'
entry in a container's own instance shadows the host's handlers just
like any other entry, and the privilege needed to shadow e.g. all ELF
binaries is the same as for a static 'M' entry matching \x7fELF today;
the only novelty is that matching becomes programmable.
The entry takes its own reference on the ops for its whole lifetime.
It is dropped from the SRCU callback that frees the entry rather than
synchronously on the final put: a walker may be asleep inside the
handler's match program while the entry's last reference goes away, so
the ops must stay callable until every walker has left the read
section - the same deferral the entry's own memory already gets. The
registration failure path, where the users refcount is not live yet,
drops it explicitly.
The match program runs from the lookup walk like magic and extension
matching and under the same rules: strict registration order, first
match wins. The walk became an SRCU read-side section in the previous
patch, so the program can sleep: it decides on the actual file
content - program headers beyond the prefetched bprm->buf, say - not
just on whatever happens to be resident in the page cache. A match
commits the exec to the handler. The sleepable load program then
selects the interpreter from load_misc_binary() by calling
bpf_binprm_set_interp() and returning zero; a failure fails the exec
instead of falling through to later entries. The walk is never left
and re-entered, so 'B' entries need no special semantics against
concurrent registration and removal whatsoever. -ENOEXEC keeps its
usual meaning and moves on to the remaining binary formats - a handler
whose load program discovers that it cannot serve the binary after all
hands it back to them - and so does returning zero without having
selected an interpreter; other program-supplied errors are clamped to
the errno range.
The 'F' flag is rejected for 'B' entries: it exists to pre-open a fixed
interpreter at registration time in the registrar's context, and a 'B'
entry has no fixed interpreter to pre-open.
'C' is accepted and behaves exactly as it does for a static entry. It
honors the suid bits of the matched binary while executing the
interpreter, which makes 'B' handlers usable for the setuid case, e.g.
a per-binary loader. This does not let the program's registrant widen
access: bprm_fill_uid() gates the credential transition on
vfsuid_has_mapping() in the caller's user namespace, so the interpreter
can only ever run as a uid that is mapped there, identical to a static
'C' entry. The computed path is opened with open_exec() under the
caller's credentials with the usual LSM and noexec checks, and the
programs run before the transition with the caller's credentials,
never elevated.
Link: https://lore.kernel.org/20260704211409.1978485-1-farid.m.zakaria@gmail.com
Link: https://patch.msgid.link/20260714-work-bpf-binfmt_misc-v2-4-57b7529c002c@kernel.org
Reviewed-by: Farid Zakaria <farid.m.zakaria@gmail.com>
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-14 21:58:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: fix possible deadlock in bm_register_write
There is a deadlock in bm_register_write:
First, in the begining of the function, a lock is taken on the binfmt_misc
root inode with inode_lock(d_inode(root)).
Then, if the user used the MISC_FMT_OPEN_FILE flag, the function will call
open_exec on the user-provided interpreter.
open_exec will call a path lookup, and if the path lookup process includes
the root of binfmt_misc, it will try to take a shared lock on its inode
again, but it is already locked, and the code will get stuck in a deadlock
To reproduce the bug:
$ echo ":iiiii:E::ii::/proc/sys/fs/binfmt_misc/bla:F" > /proc/sys/fs/binfmt_misc/register
backtrace of where the lock occurs (#5):
0 schedule () at ./arch/x86/include/asm/current.h:15
1 0xffffffff81b51237 in rwsem_down_read_slowpath (sem=0xffff888003b202e0, count=<optimized out>, state=state@entry=2) at kernel/locking/rwsem.c:992
2 0xffffffff81b5150a in __down_read_common (state=2, sem=<optimized out>) at kernel/locking/rwsem.c:1213
3 __down_read (sem=<optimized out>) at kernel/locking/rwsem.c:1222
4 down_read (sem=<optimized out>) at kernel/locking/rwsem.c:1355
5 0xffffffff811ee22a in inode_lock_shared (inode=<optimized out>) at ./include/linux/fs.h:783
6 open_last_lookups (op=0xffffc9000022fe34, file=0xffff888004098600, nd=0xffffc9000022fd10) at fs/namei.c:3177
7 path_openat (nd=nd@entry=0xffffc9000022fd10, op=op@entry=0xffffc9000022fe34, flags=flags@entry=65) at fs/namei.c:3366
8 0xffffffff811efe1c in do_filp_open (dfd=<optimized out>, pathname=pathname@entry=0xffff8880031b9000, op=op@entry=0xffffc9000022fe34) at fs/namei.c:3396
9 0xffffffff811e493f in do_open_execat (fd=fd@entry=-100, name=name@entry=0xffff8880031b9000, flags=<optimized out>, flags@entry=0) at fs/exec.c:913
10 0xffffffff811e4a92 in open_exec (name=<optimized out>) at fs/exec.c:948
11 0xffffffff8124aa84 in bm_register_write (file=<optimized out>, buffer=<optimized out>, count=19, ppos=<optimized out>) at fs/binfmt_misc.c:682
12 0xffffffff811decd2 in vfs_write (file=file@entry=0xffff888004098500, buf=buf@entry=0xa758d0 ":iiiii:E::ii::i:CF
", count=count@entry=19, pos=pos@entry=0xffffc9000022ff10) at fs/read_write.c:603
13 0xffffffff811defda in ksys_write (fd=<optimized out>, buf=0xa758d0 ":iiiii:E::ii::i:CF
", count=19) at fs/read_write.c:658
14 0xffffffff81b49813 in do_syscall_64 (nr=<optimized out>, regs=0xffffc9000022ff58) at arch/x86/entry/common.c:46
15 0xffffffff81c0007c in entry_SYSCALL_64 () at arch/x86/entry/entry_64.S:120
To solve the issue, the open_exec call is moved to before the write
lock is taken by bm_register_write
Link: https://lkml.kernel.org/r/20210228224414.95962-1-liorribak@gmail.com
Fixes: 948b701a607f1 ("binfmt_misc: add persistent opened binary handler for containers")
Signed-off-by: Lior Ribak <liorribak@gmail.com>
Acked-by: Helge Deller <deller@gmx.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-03-12 21:07:41 -08:00
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: fix possible deadlock in bm_register_write
There is a deadlock in bm_register_write:
First, in the begining of the function, a lock is taken on the binfmt_misc
root inode with inode_lock(d_inode(root)).
Then, if the user used the MISC_FMT_OPEN_FILE flag, the function will call
open_exec on the user-provided interpreter.
open_exec will call a path lookup, and if the path lookup process includes
the root of binfmt_misc, it will try to take a shared lock on its inode
again, but it is already locked, and the code will get stuck in a deadlock
To reproduce the bug:
$ echo ":iiiii:E::ii::/proc/sys/fs/binfmt_misc/bla:F" > /proc/sys/fs/binfmt_misc/register
backtrace of where the lock occurs (#5):
0 schedule () at ./arch/x86/include/asm/current.h:15
1 0xffffffff81b51237 in rwsem_down_read_slowpath (sem=0xffff888003b202e0, count=<optimized out>, state=state@entry=2) at kernel/locking/rwsem.c:992
2 0xffffffff81b5150a in __down_read_common (state=2, sem=<optimized out>) at kernel/locking/rwsem.c:1213
3 __down_read (sem=<optimized out>) at kernel/locking/rwsem.c:1222
4 down_read (sem=<optimized out>) at kernel/locking/rwsem.c:1355
5 0xffffffff811ee22a in inode_lock_shared (inode=<optimized out>) at ./include/linux/fs.h:783
6 open_last_lookups (op=0xffffc9000022fe34, file=0xffff888004098600, nd=0xffffc9000022fd10) at fs/namei.c:3177
7 path_openat (nd=nd@entry=0xffffc9000022fd10, op=op@entry=0xffffc9000022fe34, flags=flags@entry=65) at fs/namei.c:3366
8 0xffffffff811efe1c in do_filp_open (dfd=<optimized out>, pathname=pathname@entry=0xffff8880031b9000, op=op@entry=0xffffc9000022fe34) at fs/namei.c:3396
9 0xffffffff811e493f in do_open_execat (fd=fd@entry=-100, name=name@entry=0xffff8880031b9000, flags=<optimized out>, flags@entry=0) at fs/exec.c:913
10 0xffffffff811e4a92 in open_exec (name=<optimized out>) at fs/exec.c:948
11 0xffffffff8124aa84 in bm_register_write (file=<optimized out>, buffer=<optimized out>, count=19, ppos=<optimized out>) at fs/binfmt_misc.c:682
12 0xffffffff811decd2 in vfs_write (file=file@entry=0xffff888004098500, buf=buf@entry=0xa758d0 ":iiiii:E::ii::i:CF
", count=count@entry=19, pos=pos@entry=0xffffc9000022ff10) at fs/read_write.c:603
13 0xffffffff811defda in ksys_write (fd=<optimized out>, buf=0xa758d0 ":iiiii:E::ii::i:CF
", count=19) at fs/read_write.c:658
14 0xffffffff81b49813 in do_syscall_64 (nr=<optimized out>, regs=0xffffc9000022ff58) at arch/x86/entry/common.c:46
15 0xffffffff81c0007c in entry_SYSCALL_64 () at arch/x86/entry/entry_64.S:120
To solve the issue, the open_exec call is moved to before the write
lock is taken by bm_register_write
Link: https://lkml.kernel.org/r/20210228224414.95962-1-liorribak@gmail.com
Fixes: 948b701a607f1 ("binfmt_misc: add persistent opened binary handler for containers")
Signed-off-by: Lior Ribak <liorribak@gmail.com>
Acked-by: Helge Deller <deller@gmx.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-03-12 21:07:41 -08:00
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: fix possible deadlock in bm_register_write
There is a deadlock in bm_register_write:
First, in the begining of the function, a lock is taken on the binfmt_misc
root inode with inode_lock(d_inode(root)).
Then, if the user used the MISC_FMT_OPEN_FILE flag, the function will call
open_exec on the user-provided interpreter.
open_exec will call a path lookup, and if the path lookup process includes
the root of binfmt_misc, it will try to take a shared lock on its inode
again, but it is already locked, and the code will get stuck in a deadlock
To reproduce the bug:
$ echo ":iiiii:E::ii::/proc/sys/fs/binfmt_misc/bla:F" > /proc/sys/fs/binfmt_misc/register
backtrace of where the lock occurs (#5):
0 schedule () at ./arch/x86/include/asm/current.h:15
1 0xffffffff81b51237 in rwsem_down_read_slowpath (sem=0xffff888003b202e0, count=<optimized out>, state=state@entry=2) at kernel/locking/rwsem.c:992
2 0xffffffff81b5150a in __down_read_common (state=2, sem=<optimized out>) at kernel/locking/rwsem.c:1213
3 __down_read (sem=<optimized out>) at kernel/locking/rwsem.c:1222
4 down_read (sem=<optimized out>) at kernel/locking/rwsem.c:1355
5 0xffffffff811ee22a in inode_lock_shared (inode=<optimized out>) at ./include/linux/fs.h:783
6 open_last_lookups (op=0xffffc9000022fe34, file=0xffff888004098600, nd=0xffffc9000022fd10) at fs/namei.c:3177
7 path_openat (nd=nd@entry=0xffffc9000022fd10, op=op@entry=0xffffc9000022fe34, flags=flags@entry=65) at fs/namei.c:3366
8 0xffffffff811efe1c in do_filp_open (dfd=<optimized out>, pathname=pathname@entry=0xffff8880031b9000, op=op@entry=0xffffc9000022fe34) at fs/namei.c:3396
9 0xffffffff811e493f in do_open_execat (fd=fd@entry=-100, name=name@entry=0xffff8880031b9000, flags=<optimized out>, flags@entry=0) at fs/exec.c:913
10 0xffffffff811e4a92 in open_exec (name=<optimized out>) at fs/exec.c:948
11 0xffffffff8124aa84 in bm_register_write (file=<optimized out>, buffer=<optimized out>, count=19, ppos=<optimized out>) at fs/binfmt_misc.c:682
12 0xffffffff811decd2 in vfs_write (file=file@entry=0xffff888004098500, buf=buf@entry=0xa758d0 ":iiiii:E::ii::i:CF
", count=count@entry=19, pos=pos@entry=0xffffc9000022ff10) at fs/read_write.c:603
13 0xffffffff811defda in ksys_write (fd=<optimized out>, buf=0xa758d0 ":iiiii:E::ii::i:CF
", count=19) at fs/read_write.c:658
14 0xffffffff81b49813 in do_syscall_64 (nr=<optimized out>, regs=0xffffc9000022ff58) at arch/x86/entry/common.c:46
15 0xffffffff81c0007c in entry_SYSCALL_64 () at arch/x86/entry/entry_64.S:120
To solve the issue, the open_exec call is moved to before the write
lock is taken by bm_register_write
Link: https://lkml.kernel.org/r/20210228224414.95962-1-liorribak@gmail.com
Fixes: 948b701a607f1 ("binfmt_misc: add persistent opened binary handler for containers")
Signed-off-by: Lior Ribak <liorribak@gmail.com>
Acked-by: Helge Deller <deller@gmx.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-03-12 21:07:41 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-05-13 22:02:44 -06:00
|
|
|
|
2026-07-30 15:34:08 +02:00
|
|
|
|
2016-02-17 16:51:16 -08:00
|
|
|
|
2026-07-10 11:33:22 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2006-03-28 01:56:42 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
llseek: automatically add .llseek fop
All file_operations should get a .llseek operation so we can make
nonseekable_open the default for future file operations without a
.llseek pointer.
The three cases that we can automatically detect are no_llseek, seq_lseek
and default_llseek. For cases where we can we can automatically prove that
the file offset is always ignored, we use noop_llseek, which maintains
the current behavior of not returning an error from a seek.
New drivers should normally not use noop_llseek but instead use no_llseek
and call nonseekable_open at open time. Existing drivers can be converted
to do the same when the maintainer knows for certain that no user code
relies on calling seek on the device file.
The generated code is often incorrectly indented and right now contains
comments that clarify for each added line why a specific variant was
chosen. In the version that gets submitted upstream, the comments will
be gone and I will manually fix the indentation, because there does not
seem to be a way to do that using coccinelle.
Some amount of new code is currently sitting in linux-next that should get
the same modifications, which I will do at the end of the merge window.
Many thanks to Julia Lawall for helping me learn to write a semantic
patch that does all this.
===== begin semantic patch =====
// This adds an llseek= method to all file operations,
// as a preparation for making no_llseek the default.
//
// The rules are
// - use no_llseek explicitly if we do nonseekable_open
// - use seq_lseek for sequential files
// - use default_llseek if we know we access f_pos
// - use noop_llseek if we know we don't access f_pos,
// but we still want to allow users to call lseek
//
@ open1 exists @
identifier nested_open;
@@
nested_open(...)
{
<+...
nonseekable_open(...)
...+>
}
@ open exists@
identifier open_f;
identifier i, f;
identifier open1.nested_open;
@@
int open_f(struct inode *i, struct file *f)
{
<+...
(
nonseekable_open(...)
|
nested_open(...)
)
...+>
}
@ read disable optional_qualifier exists @
identifier read_f;
identifier f, p, s, off;
type ssize_t, size_t, loff_t;
expression E;
identifier func;
@@
ssize_t read_f(struct file *f, char *p, size_t s, loff_t *off)
{
<+...
(
*off = E
|
*off += E
|
func(..., off, ...)
|
E = *off
)
...+>
}
@ read_no_fpos disable optional_qualifier exists @
identifier read_f;
identifier f, p, s, off;
type ssize_t, size_t, loff_t;
@@
ssize_t read_f(struct file *f, char *p, size_t s, loff_t *off)
{
... when != off
}
@ write @
identifier write_f;
identifier f, p, s, off;
type ssize_t, size_t, loff_t;
expression E;
identifier func;
@@
ssize_t write_f(struct file *f, const char *p, size_t s, loff_t *off)
{
<+...
(
*off = E
|
*off += E
|
func(..., off, ...)
|
E = *off
)
...+>
}
@ write_no_fpos @
identifier write_f;
identifier f, p, s, off;
type ssize_t, size_t, loff_t;
@@
ssize_t write_f(struct file *f, const char *p, size_t s, loff_t *off)
{
... when != off
}
@ fops0 @
identifier fops;
@@
struct file_operations fops = {
...
};
@ has_llseek depends on fops0 @
identifier fops0.fops;
identifier llseek_f;
@@
struct file_operations fops = {
...
.llseek = llseek_f,
...
};
@ has_read depends on fops0 @
identifier fops0.fops;
identifier read_f;
@@
struct file_operations fops = {
...
.read = read_f,
...
};
@ has_write depends on fops0 @
identifier fops0.fops;
identifier write_f;
@@
struct file_operations fops = {
...
.write = write_f,
...
};
@ has_open depends on fops0 @
identifier fops0.fops;
identifier open_f;
@@
struct file_operations fops = {
...
.open = open_f,
...
};
// use no_llseek if we call nonseekable_open
////////////////////////////////////////////
@ nonseekable1 depends on !has_llseek && has_open @
identifier fops0.fops;
identifier nso ~= "nonseekable_open";
@@
struct file_operations fops = {
... .open = nso, ...
+.llseek = no_llseek, /* nonseekable */
};
@ nonseekable2 depends on !has_llseek @
identifier fops0.fops;
identifier open.open_f;
@@
struct file_operations fops = {
... .open = open_f, ...
+.llseek = no_llseek, /* open uses nonseekable */
};
// use seq_lseek for sequential files
/////////////////////////////////////
@ seq depends on !has_llseek @
identifier fops0.fops;
identifier sr ~= "seq_read";
@@
struct file_operations fops = {
... .read = sr, ...
+.llseek = seq_lseek, /* we have seq_read */
};
// use default_llseek if there is a readdir
///////////////////////////////////////////
@ fops1 depends on !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier readdir_e;
@@
// any other fop is used that changes pos
struct file_operations fops = {
... .readdir = readdir_e, ...
+.llseek = default_llseek, /* readdir is present */
};
// use default_llseek if at least one of read/write touches f_pos
/////////////////////////////////////////////////////////////////
@ fops2 depends on !fops1 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier read.read_f;
@@
// read fops use offset
struct file_operations fops = {
... .read = read_f, ...
+.llseek = default_llseek, /* read accesses f_pos */
};
@ fops3 depends on !fops1 && !fops2 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier write.write_f;
@@
// write fops use offset
struct file_operations fops = {
... .write = write_f, ...
+ .llseek = default_llseek, /* write accesses f_pos */
};
// Use noop_llseek if neither read nor write accesses f_pos
///////////////////////////////////////////////////////////
@ fops4 depends on !fops1 && !fops2 && !fops3 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier read_no_fpos.read_f;
identifier write_no_fpos.write_f;
@@
// write fops use offset
struct file_operations fops = {
...
.write = write_f,
.read = read_f,
...
+.llseek = noop_llseek, /* read and write both use no f_pos */
};
@ depends on has_write && !has_read && !fops1 && !fops2 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier write_no_fpos.write_f;
@@
struct file_operations fops = {
... .write = write_f, ...
+.llseek = noop_llseek, /* write uses no f_pos */
};
@ depends on has_read && !has_write && !fops1 && !fops2 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier read_no_fpos.read_f;
@@
struct file_operations fops = {
... .read = read_f, ...
+.llseek = noop_llseek, /* read uses no f_pos */
};
@ depends on !has_read && !has_write && !fops1 && !fops2 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
@@
struct file_operations fops = {
...
+.llseek = noop_llseek, /* no read or write fn */
};
===== End semantic patch =====
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Julia Lawall <julia@diku.dk>
Cc: Christoph Hellwig <hch@infradead.org>
2010-08-15 18:52:59 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
2026-07-10 11:33:23 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
2026-07-10 11:33:07 +02:00
|
|
|
|
2007-05-09 02:33:32 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2014-12-10 15:52:10 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
binfmt_misc: let a 'B' entry bind its interpreters
A 'B' entry's load program selects its interpreter by absolute path,
which open_exec() resolves at exec time in the mount namespace of whoever
runs the binary. The handler names an interpreter but does not get to say
which file that is. Whoever controls the filesystem view of the exec
decides that instead.
Static entries settled this long ago with 'F'. The interpreter is opened
at registration in the registrant's context and every exec runs a clone
of that file. Give a 'B' entry the same, for as many interpreters as it
needs.
An entry registered with 'D' cannot be matched yet, so it still belongs
to whoever is configuring it and can be given interpreters one write at a
time:
echo ':qemu:B::::qemu_user:D' > register
echo '+aarch64 /usr/bin/qemu-aarch64' > qemu
echo '+arm /usr/bin/qemu-arm' > qemu
echo 1 > qemu
Each path is opened by its write, with the credentials the entry file
was opened with, by the same helper that opens an 'F' interpreter. The
load program picks one per exec with bpf_binprm_select_interp() and the
entry hands out a clone of it. Nothing is resolved again, in any
namespace. The path is everything past the first space, so no
interpreter has to fit in a register string. An entry binds at most a
hundred interpreters (BINFMT_MISC_INTERP_MAX). Every binding pins a
struct file that no file descriptor accounts for, so RLIMIT_NOFILE does
not apply and some cap is needed. A hundred is plenty and raising it
later is cheap, lowering it is not.
Selection is by name so the register string and the program need not
agree on an order, and so the handler is not tied to where a distribution
puts its interpreters. A name is a single word of printable ASCII so the
entry file can report 'name path' lines. The interpreter runs under the
path it was registered under.
The entry file reads user memory once. bm_entry_write() copies the write
in and dispatches on the first byte, and parse_command() takes the copied
buffer. The status file has no binding to spell, so it keeps its own
small copy in read_command().
That moves the length cap ahead of the dispatch. A write to an entry file
longer than a binding can be is now refused with -E2BIG, and one from a
bad address reports -EFAULT, where the command parser used to report
-EINVAL for anything past three bytes.
Configurations of one instance are kept apart by the lock removal
already takes. Reading the set out of the entry file takes no lock.
Bindings are rcu-published and the open entry file pins the entry
together with everything it bound, so a reader either sees a whole node
or misses it. The interpreter is opened before the configuration lock
because resolving the path may walk this very filesystem, and only
after the command has been parsed and the name validated from the
copied buffer, so a write that can never bind opens nothing and the
errno reflects the actual failure.
Link: https://patch.msgid.link/20260730-work-binfmt_misc-preopen-v1-7-4a0b0da71f16@kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-30 15:34:09 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-07-10 11:33:17 +02:00
|
|
|
|
2026-07-10 11:33:07 +02:00
|
|
|
|
2014-12-10 15:52:10 -08:00
|
|
|
|
2026-07-10 11:33:17 +02:00
|
|
|
|
2026-07-10 11:33:07 +02:00
|
|
|
|
2014-12-10 15:52:10 -08:00
|
|
|
|
2026-07-10 11:33:17 +02:00
|
|
|
|
2026-07-10 11:33:18 +02:00
|
|
|
|
2014-12-10 15:52:10 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2014-12-10 15:52:10 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2006-03-28 01:56:42 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
llseek: automatically add .llseek fop
All file_operations should get a .llseek operation so we can make
nonseekable_open the default for future file operations without a
.llseek pointer.
The three cases that we can automatically detect are no_llseek, seq_lseek
and default_llseek. For cases where we can we can automatically prove that
the file offset is always ignored, we use noop_llseek, which maintains
the current behavior of not returning an error from a seek.
New drivers should normally not use noop_llseek but instead use no_llseek
and call nonseekable_open at open time. Existing drivers can be converted
to do the same when the maintainer knows for certain that no user code
relies on calling seek on the device file.
The generated code is often incorrectly indented and right now contains
comments that clarify for each added line why a specific variant was
chosen. In the version that gets submitted upstream, the comments will
be gone and I will manually fix the indentation, because there does not
seem to be a way to do that using coccinelle.
Some amount of new code is currently sitting in linux-next that should get
the same modifications, which I will do at the end of the merge window.
Many thanks to Julia Lawall for helping me learn to write a semantic
patch that does all this.
===== begin semantic patch =====
// This adds an llseek= method to all file operations,
// as a preparation for making no_llseek the default.
//
// The rules are
// - use no_llseek explicitly if we do nonseekable_open
// - use seq_lseek for sequential files
// - use default_llseek if we know we access f_pos
// - use noop_llseek if we know we don't access f_pos,
// but we still want to allow users to call lseek
//
@ open1 exists @
identifier nested_open;
@@
nested_open(...)
{
<+...
nonseekable_open(...)
...+>
}
@ open exists@
identifier open_f;
identifier i, f;
identifier open1.nested_open;
@@
int open_f(struct inode *i, struct file *f)
{
<+...
(
nonseekable_open(...)
|
nested_open(...)
)
...+>
}
@ read disable optional_qualifier exists @
identifier read_f;
identifier f, p, s, off;
type ssize_t, size_t, loff_t;
expression E;
identifier func;
@@
ssize_t read_f(struct file *f, char *p, size_t s, loff_t *off)
{
<+...
(
*off = E
|
*off += E
|
func(..., off, ...)
|
E = *off
)
...+>
}
@ read_no_fpos disable optional_qualifier exists @
identifier read_f;
identifier f, p, s, off;
type ssize_t, size_t, loff_t;
@@
ssize_t read_f(struct file *f, char *p, size_t s, loff_t *off)
{
... when != off
}
@ write @
identifier write_f;
identifier f, p, s, off;
type ssize_t, size_t, loff_t;
expression E;
identifier func;
@@
ssize_t write_f(struct file *f, const char *p, size_t s, loff_t *off)
{
<+...
(
*off = E
|
*off += E
|
func(..., off, ...)
|
E = *off
)
...+>
}
@ write_no_fpos @
identifier write_f;
identifier f, p, s, off;
type ssize_t, size_t, loff_t;
@@
ssize_t write_f(struct file *f, const char *p, size_t s, loff_t *off)
{
... when != off
}
@ fops0 @
identifier fops;
@@
struct file_operations fops = {
...
};
@ has_llseek depends on fops0 @
identifier fops0.fops;
identifier llseek_f;
@@
struct file_operations fops = {
...
.llseek = llseek_f,
...
};
@ has_read depends on fops0 @
identifier fops0.fops;
identifier read_f;
@@
struct file_operations fops = {
...
.read = read_f,
...
};
@ has_write depends on fops0 @
identifier fops0.fops;
identifier write_f;
@@
struct file_operations fops = {
...
.write = write_f,
...
};
@ has_open depends on fops0 @
identifier fops0.fops;
identifier open_f;
@@
struct file_operations fops = {
...
.open = open_f,
...
};
// use no_llseek if we call nonseekable_open
////////////////////////////////////////////
@ nonseekable1 depends on !has_llseek && has_open @
identifier fops0.fops;
identifier nso ~= "nonseekable_open";
@@
struct file_operations fops = {
... .open = nso, ...
+.llseek = no_llseek, /* nonseekable */
};
@ nonseekable2 depends on !has_llseek @
identifier fops0.fops;
identifier open.open_f;
@@
struct file_operations fops = {
... .open = open_f, ...
+.llseek = no_llseek, /* open uses nonseekable */
};
// use seq_lseek for sequential files
/////////////////////////////////////
@ seq depends on !has_llseek @
identifier fops0.fops;
identifier sr ~= "seq_read";
@@
struct file_operations fops = {
... .read = sr, ...
+.llseek = seq_lseek, /* we have seq_read */
};
// use default_llseek if there is a readdir
///////////////////////////////////////////
@ fops1 depends on !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier readdir_e;
@@
// any other fop is used that changes pos
struct file_operations fops = {
... .readdir = readdir_e, ...
+.llseek = default_llseek, /* readdir is present */
};
// use default_llseek if at least one of read/write touches f_pos
/////////////////////////////////////////////////////////////////
@ fops2 depends on !fops1 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier read.read_f;
@@
// read fops use offset
struct file_operations fops = {
... .read = read_f, ...
+.llseek = default_llseek, /* read accesses f_pos */
};
@ fops3 depends on !fops1 && !fops2 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier write.write_f;
@@
// write fops use offset
struct file_operations fops = {
... .write = write_f, ...
+ .llseek = default_llseek, /* write accesses f_pos */
};
// Use noop_llseek if neither read nor write accesses f_pos
///////////////////////////////////////////////////////////
@ fops4 depends on !fops1 && !fops2 && !fops3 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier read_no_fpos.read_f;
identifier write_no_fpos.write_f;
@@
// write fops use offset
struct file_operations fops = {
...
.write = write_f,
.read = read_f,
...
+.llseek = noop_llseek, /* read and write both use no f_pos */
};
@ depends on has_write && !has_read && !fops1 && !fops2 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier write_no_fpos.write_f;
@@
struct file_operations fops = {
... .write = write_f, ...
+.llseek = noop_llseek, /* write uses no f_pos */
};
@ depends on has_read && !has_write && !fops1 && !fops2 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
identifier read_no_fpos.read_f;
@@
struct file_operations fops = {
... .read = read_f, ...
+.llseek = noop_llseek, /* read uses no f_pos */
};
@ depends on !has_read && !has_write && !fops1 && !fops2 && !has_llseek && !nonseekable1 && !nonseekable2 && !seq @
identifier fops0.fops;
@@
struct file_operations fops = {
...
+.llseek = noop_llseek, /* no read or write fn */
};
===== End semantic patch =====
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Julia Lawall <julia@diku.dk>
Cc: Christoph Hellwig <hch@infradead.org>
2010-08-15 18:52:59 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:23 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2010-06-07 14:34:48 -04:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2019-03-25 16:38:28 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2014-12-10 15:52:10 -08:00
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
|
|
|
|
|
2017-03-25 21:15:37 -07:00
|
|
|
|
2007-05-08 00:32:31 -07:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2014-12-10 15:52:10 -08:00
|
|
|
|
2026-08-02 20:00:44 +02:00
|
|
|
|
|
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
|
|
|
|
|
2026-07-28 14:26:32 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-11-02 10:25:04 +01:00
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
|
|
|
|
|
2026-02-21 16:37:42 -08:00
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-10 11:33:05 +02:00
|
|
|
|
2026-07-10 11:33:06 +02:00
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
2026-07-10 11:33:15 +02:00
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: don't leak the user namespace when the mount fails
bm_get_tree() takes a reference to the user namespace and hands it to
get_tree_keyed() as the sget key. sget_fc() moves that reference into
sb->s_fs_info and clears fc->s_fs_info, so from that point on the
superblock owns it and bm_free() doesn't see it anymore.
The superblock drops it in ->put_super(). But generic_shutdown_super()
only calls ->put_super() from inside the if (sb->s_root) branch, so
nothing releases it when bm_fill_super() fails:
- The kzalloc_obj() failure leaves s_root NULL and the whole branch is
skipped.
- A simple_fill_super() failure in the file loop leaves s_root set, but
s_op still points at simple_super_operations, which has no
->put_super(). bm_fill_super() installs s_ops only once
simple_fill_super() returned success, and installing it earlier
wouldn't help either because simple_fill_super() overwrites s_op.
Either way vfs_get_super() calls deactivate_locked_super() and the
reference is gone for good. binfmt_misc mounts are available in a user
namespace and both the inode and the dentry cache are SLAB_ACCOUNT, so
an unprivileged caller under a tight memory cgroup can fail
simple_fill_super() on demand and leak one user namespace per attempt.
Drop the reference in ->kill_sb() instead, which runs unconditionally,
the same way nfsd and rpc_pipefs release their keyed s_fs_info.
That also stops ->put_super() from clearing s_fs_info while the
superblock is still on @fs_supers. generic_shutdown_super() leaves it
there on purpose so that sget_fc() keeps finding it until kill_sb() has
run, but a NULL s_fs_info makes test_keyed_super() miss it, so a
concurrent mount for the same user namespace skips the grab_super()
wait and creates a second superblock for a namespace that is still
being torn down.
Link: https://patch.msgid.link/20260728-work-binfmt_misc-usernsleak-v1-1-dbd8d5e626e7@kernel.org
Fixes: 21ca59b365c0 ("binfmt_misc: enable sandboxed mounts")
Cc: stable@vger.kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-28 15:48:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
2026-07-10 11:33:07 +02:00
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
2014-12-10 15:52:10 -08:00
|
|
|
|
2026-07-10 11:33:25 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-03-25 16:38:28 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
2019-03-25 16:38:28 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
2019-03-25 16:38:28 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
binfmt_misc: don't leak the user namespace when the mount fails
bm_get_tree() takes a reference to the user namespace and hands it to
get_tree_keyed() as the sget key. sget_fc() moves that reference into
sb->s_fs_info and clears fc->s_fs_info, so from that point on the
superblock owns it and bm_free() doesn't see it anymore.
The superblock drops it in ->put_super(). But generic_shutdown_super()
only calls ->put_super() from inside the if (sb->s_root) branch, so
nothing releases it when bm_fill_super() fails:
- The kzalloc_obj() failure leaves s_root NULL and the whole branch is
skipped.
- A simple_fill_super() failure in the file loop leaves s_root set, but
s_op still points at simple_super_operations, which has no
->put_super(). bm_fill_super() installs s_ops only once
simple_fill_super() returned success, and installing it earlier
wouldn't help either because simple_fill_super() overwrites s_op.
Either way vfs_get_super() calls deactivate_locked_super() and the
reference is gone for good. binfmt_misc mounts are available in a user
namespace and both the inode and the dentry cache are SLAB_ACCOUNT, so
an unprivileged caller under a tight memory cgroup can fail
simple_fill_super() on demand and leak one user namespace per attempt.
Drop the reference in ->kill_sb() instead, which runs unconditionally,
the same way nfsd and rpc_pipefs release their keyed s_fs_info.
That also stops ->put_super() from clearing s_fs_info while the
superblock is still on @fs_supers. generic_shutdown_super() leaves it
there on purpose so that sget_fc() keeps finding it until kill_sb() has
run, but a NULL s_fs_info makes test_keyed_super() miss it, so a
concurrent mount for the same user namespace skips the grab_super()
wait and creates a second superblock for a namespace that is still
being torn down.
Link: https://patch.msgid.link/20260728-work-binfmt_misc-usernsleak-v1-1-dbd8d5e626e7@kernel.org
Fixes: 21ca59b365c0 ("binfmt_misc: enable sandboxed mounts")
Cc: stable@vger.kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-28 15:48:10 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-03-25 16:38:28 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-03-25 16:38:28 +00:00
|
|
|
|
binfmt_misc: enable sandboxed mounts
Enable unprivileged sandboxes to create their own binfmt_misc mounts.
This is based on Laurent's work in [1] but has been significantly
reworked to fix various issues we identified in earlier versions.
While binfmt_misc can currently only be mounted in the initial user
namespace, binary types registered in this binfmt_misc instance are
available to all sandboxes (Either by having them installed in the
sandbox or by registering the binary type with the F flag causing the
interpreter to be opened right away). So binfmt_misc binary types are
already delegated to sandboxes implicitly.
However, while a sandbox has access to all registered binary types in
binfmt_misc a sandbox cannot currently register its own binary types
in binfmt_misc. This has prevented various use-cases some of which were
already outlined in [1] but we have a range of issues associated with
this (cf. [3]-[5] below which are just a small sample).
Extend binfmt_misc to be mountable in non-initial user namespaces.
Similar to other filesystem such as nfsd, mqueue, and sunrpc we use
keyed superblock management. The key determines whether we need to
create a new superblock or can reuse an already existing one. We use the
user namespace of the mount as key. This means a new binfmt_misc
superblock is created once per user namespace creation. Subsequent
mounts of binfmt_misc in the same user namespace will mount the same
binfmt_misc instance. We explicitly do not create a new binfmt_misc
superblock on every binfmt_misc mount as the semantics for
load_misc_binary() line up with the keying model. This also allows us to
retrieve the relevant binfmt_misc instance based on the caller's user
namespace which can be done in a simple (bounded to 32 levels) loop.
Similar to the current binfmt_misc semantics allowing access to the
binary types in the initial binfmt_misc instance we do allow sandboxes
access to their parent's binfmt_misc mounts if they do not have created
a separate binfmt_misc instance.
Overall, this will unblock the use-cases mentioned below and in general
will also allow to support and harden execution of another
architecture's binaries in tight sandboxes. For instance, using the
unshare binary it possible to start a chroot of another architecture and
configure the binfmt_misc interpreter without being root to run the
binaries in this chroot and without requiring the host to modify its
binary type handlers.
Henning had already posted a few experiments in the cover letter at [1].
But here's an additional example where an unprivileged container
registers qemu-user-static binary handlers for various binary types in
its separate binfmt_misc mount and is then seamlessly able to start
containers with a different architecture without affecting the host:
root [lxc monitor] /var/snap/lxd/common/lxd/containers f1
1000000 \_ /sbin/init
1000000 \_ /lib/systemd/systemd-journald
1000000 \_ /lib/systemd/systemd-udevd
1000100 \_ /lib/systemd/systemd-networkd
1000101 \_ /lib/systemd/systemd-resolved
1000000 \_ /usr/sbin/cron -f
1000103 \_ /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-activation --syslog-only
1000000 \_ /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1000104 \_ /usr/sbin/rsyslogd -n -iNONE
1000000 \_ /lib/systemd/systemd-logind
1000000 \_ /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1000107 \_ dnsmasq --conf-file=/dev/null -u lxc-dnsmasq --strict-order --bind-interfaces --pid-file=/run/lxc/dnsmasq.pid --liste
1000000 \_ [lxc monitor] /var/lib/lxc f1-s390x
1100000 \_ /usr/bin/qemu-s390x-static /sbin/init
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-journald
1100000 \_ /usr/bin/qemu-s390x-static /usr/sbin/cron -f
1100103 \_ /usr/bin/qemu-s390x-static /usr/bin/dbus-daemon --system --address=systemd: --nofork --nopidfile --systemd-ac
1100000 \_ /usr/bin/qemu-s390x-static /usr/bin/python3 /usr/bin/networkd-dispatcher --run-startup-triggers
1100104 \_ /usr/bin/qemu-s390x-static /usr/sbin/rsyslogd -n -iNONE
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-logind
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud console 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/0 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/1 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/2 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /sbin/agetty -o -p -- \u --noclear --keep-baud pts/3 115200,38400,9600 vt220
1100000 \_ /usr/bin/qemu-s390x-static /lib/systemd/systemd-udevd
[1]: https://lore.kernel.org/all/20191216091220.465626-1-laurent@vivier.eu
[2]: https://discuss.linuxcontainers.org/t/binfmt-misc-permission-denied
[3]: https://discuss.linuxcontainers.org/t/lxd-binfmt-support-for-qemu-static-interpreters
[4]: https://discuss.linuxcontainers.org/t/3-1-0-binfmt-support-service-in-unprivileged-guest-requires-write-access-on-hosts-proc-sys-fs-binfmt-misc
[5]: https://discuss.linuxcontainers.org/t/qemu-user-static-not-working-4-11
Link: https://lore.kernel.org/r/20191216091220.465626-2-laurent@vivier.eu (origin)
Link: https://lore.kernel.org/r/20211028103114.2849140-2-brauner@kernel.org (v1)
Cc: Sargun Dhillon <sargun@sargun.me>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Henning Schild <henning.schild@siemens.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Laurent Vivier <laurent@vivier.eu>
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Laurent Vivier <laurent@vivier.eu>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Kees Cook <keescook@chromium.org>
---
/* v2 */
- Serge Hallyn <serge@hallyn.com>:
- Use GFP_KERNEL_ACCOUNT for userspace triggered allocations when a
new binary type handler is registered.
- Christian Brauner <christian.brauner@ubuntu.com>:
- Switch authorship to me. I refused to do that earlier even though
Laurent said I should do so because I think it's genuinely bad form.
But by now I have changed so many things that it'd be unfair to
blame Laurent for any potential bugs in here.
- Add more comments that explain what's going on.
- Rename functions while changing them to better reflect what they are
doing to make the code easier to understand.
- In the first version when a specific binary type handler was removed
either through a write to the entry's file or all binary type
handlers were removed by a write to the binfmt_misc mount's status
file all cleanup work happened during inode eviction.
That includes removal of the relevant entries from entry list. While
that works fine I disliked that model after thinking about it for a
bit. Because it means that there was a window were someone has
already removed a or all binary handlers but they could still be
safely reached from load_misc_binary() when it has managed to take
the read_lock() on the entries list while inode eviction was already
happening. Again, that perfectly benign but it's cleaner to remove
the binary handler from the list immediately meaning that ones the
write to then entry's file or the binfmt_misc status file returns
the binary type cannot be executed anymore. That gives stronger
guarantees to the user.
2021-10-28 12:31:14 +02:00
|
|
|
|
binfmt_misc: don't leak the user namespace when the mount fails
bm_get_tree() takes a reference to the user namespace and hands it to
get_tree_keyed() as the sget key. sget_fc() moves that reference into
sb->s_fs_info and clears fc->s_fs_info, so from that point on the
superblock owns it and bm_free() doesn't see it anymore.
The superblock drops it in ->put_super(). But generic_shutdown_super()
only calls ->put_super() from inside the if (sb->s_root) branch, so
nothing releases it when bm_fill_super() fails:
- The kzalloc_obj() failure leaves s_root NULL and the whole branch is
skipped.
- A simple_fill_super() failure in the file loop leaves s_root set, but
s_op still points at simple_super_operations, which has no
->put_super(). bm_fill_super() installs s_ops only once
simple_fill_super() returned success, and installing it earlier
wouldn't help either because simple_fill_super() overwrites s_op.
Either way vfs_get_super() calls deactivate_locked_super() and the
reference is gone for good. binfmt_misc mounts are available in a user
namespace and both the inode and the dentry cache are SLAB_ACCOUNT, so
an unprivileged caller under a tight memory cgroup can fail
simple_fill_super() on demand and leak one user namespace per attempt.
Drop the reference in ->kill_sb() instead, which runs unconditionally,
the same way nfsd and rpc_pipefs release their keyed s_fs_info.
That also stops ->put_super() from clearing s_fs_info while the
superblock is still on @fs_supers. generic_shutdown_super() leaves it
there on purpose so that sget_fc() keeps finding it until kill_sb() has
run, but a NULL s_fs_info makes test_keyed_super() miss it, so a
concurrent mount for the same user namespace skips the grab_super()
wait and creates a second superblock for a namespace that is still
being torn down.
Link: https://patch.msgid.link/20260728-work-binfmt_misc-usernsleak-v1-1-dbd8d5e626e7@kernel.org
Fixes: 21ca59b365c0 ("binfmt_misc: enable sandboxed mounts")
Cc: stable@vger.kernel.org
Signed-off-by: Christian Brauner (Amutable) <brauner@kernel.org>
2026-07-28 15:48:10 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2013-03-02 19:39:14 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-03-17 03:05:16 -04:00
|
|
|
|
|
|
|
|
|
2022-02-09 08:49:20 +01:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-07-14 21:58:08 +02:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-05-27 11:57:52 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
