dso__decompress_kmodule_path() unconditionally calls close(fd) on the
return value of decompress_kmodule(). When decompression fails or the
DSO is not compressed, decompress_kmodule() returns -1. close(-1)
fails with EBADF and clobbers errno, which callers up the chain
(dso__get_filename → __open_dso) depend on for error propagation.
Guard the close() call with fd >= 0 so only valid file descriptors are
closed.
Fixes: 42b3fa6708 ("perf tools: Introduce dso__decompress_kmodule_{fd,path}")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Reviewed-by: Ian Rogers <irogers@google.com>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
__open_dso() computes fd = -errno when dso__get_filename() returns NULL.
Some failure paths in dso__get_filename() (e.g. binary type mismatch)
return NULL without making a syscall, leaving errno at 0 from a prior
successful call. fd = -0 = 0, which is stdin — subsequent code treats
it as a valid file descriptor.
Fall back to ENOENT when errno is 0, ensuring fd is always negative on
failure.
The forced ENOENT stays in errno for the callers that check it after a
negative fd. It must not misdirect the try_to_open_dso() fallback
loop, though: dso__get_filename()'s chroot fallback used to accept a
stale ENOENT even when stat() succeeded on a non-regular file (e.g. a
directory). Re-stat() there and only take the chroot path when
stat() actually failed with ENOENT [sashiko-bot review of PATCH 1/5].
Fixes: eba5102d2f ("perf tools: Add global list of opened dso objects")
Reported-by: sashiko-bot <sashiko-bot@kernel.org>
Reviewed-by: Ian Rogers <irogers@google.com>
Assisted-by: Claude:claude-opus-4.6
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Signed-off-by: Namhyung Kim <namhyung@kernel.org>
sched_class->balance() is gone from sched_ext and what balance_one() does is
run dispatch to produce something pickable. Update the balance-era names to
dispatch terms:
- balance_one() -> dispatch_one()
- SCX_RQ_IN_BALANCE -> SCX_RQ_IN_DISPATCH
No BPF scheduler reads the flag. The enum autogen headers gain the new name
with the old entry retained like other removed enumerators, zero-filling at
load time. No functional changes.
Signed-off-by: Tejun Heo <tj@kernel.org>
The fixup demoting a keep verdict when @prev is not on ext_sched_class
guarded against the rq-level SCX_RQ_BAL_KEEP flag going stale back when
balancing and picking were separate operations.
The verdict now travels in the return value, created and consumed in one
invocation against the @prev it evaluated, and every keep decision tests
SCX_TASK_QUEUED under the rq lock, which implies ext_sched_class as a class
switch dequeues first. Drop the fixup along with dispatch_core_pick()'s
copy.
Signed-off-by: Tejun Heo <tj@kernel.org>
kick_sync_wait_bal_cb() assumes it runs on the rq's CPU from the
__schedule() tail: the snapshots it compares against live in that CPU's
percpu area and the busy-wait runs with the rq lock dropped and IRQs
enabled.
However, dispatch can now drop the rq lock while the callback sits queued,
and rq lock takers in that window (the sched class change paths, the scx
task iterator) flush pending balance callbacks on release, running the
callback on a foreign CPU. Such a run compares against unrelated snapshots
and can deadlock when the executing CPU is itself a wait target.
Bail on a foreign CPU and leave the wait state alone. The wait only observes
progress that the resched kicks already guarantee and the rq's next wait
picks up the stale cpus_to_sync bits.
Fixes: 4c95380701 ("sched/ext: Fold balance_scx() into pick_task_scx()")
Cc: stable@vger.kernel.org # v6.19+
Signed-off-by: Tejun Heo <tj@kernel.org>
kernel/sched/ext/arena.c uses the gen_pool allocator, which is built only
when GENERIC_ALLOCATOR is set. SCHED_CLASS_EXT doesn't select it, so on
configs where nothing else does, the build fails to link:
build_policy.o: undefined reference to `gen_pool_create'
build_policy.o: undefined reference to `gen_pool_for_each_chunk'
build_policy.o: undefined reference to `gen_pool_destroy'
Fixes: 9eca087deb ("sched_ext: Sub-allocator over kernel-claimed BPF arena pages")
Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202608151315.tvN3X0Oq-lkp@intel.com/
Closes: https://lore.kernel.org/oe-kbuild-all/202608151632.3p91bTQj-lkp@intel.com/
Signed-off-by: Tejun Heo <tj@kernel.org>
fcg_dispatch() true-ups the current cgroup's cvtime when its slice
expires or its DSQ runs empty while the slice is still active:
__sync_fetch_and_add(&cgc->cvtime_delta,
(cpuc->cur_at + cgrp_slice_ns - now) *
FCG_HWEIGHT_ONE / (cgc->hweight ?: 1));
The true-up should be actual minus charged: on CNS_EXPIRE, the overrun
(now - cur_at - cgrp_slice_ns) should be added; on CNS_EMPTY, the
unused portion of the slice should be subtracted. The expression above
has the sign inverted, and in the CNS_EXPIRE case now is already past
cur_at + cgrp_slice_ns, so the u64 subtraction wraps. The
multiplication preserves the two's complement encoding but the
unsigned division by hweight destroys it, adding roughly 2^64/hweight
per expiry instead of a small correction.
Under saturation the hweight budget clamp in cgrp_cap_budget() masks
most of the garbage, so the weight distribution barely moves, but the
accounting is broken all the same. Compute the delta as a signed value
and use fetch_and_add()/fetch_and_sub() so that the dividends stay
positive, as BPF division is unsigned.
Instrumented the true-up and ran a saturated three-leaf cgroup tree on
a 4-CPU VM: without the fix, each expiry added ~5e15 (2^64/hweight
territory) to cvtime_delta; with it, the corrections are back to
slice scale, with the overrun added and the unused portion subtracted.
Fixes: a4103eacc2 ("sched_ext: Add a cgroup scheduler which uses flattened hierarchy")
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Tao Cui <cuitao@kylinos.cn>
Signed-off-by: Tejun Heo <tj@kernel.org>
scx_bpf_dsq_reenq() queues a deferred reenq (dru) that runs from
run_deferred(), not ops.dispatch(). If the DSQ is destroyed before the dru
runs, process_deferred_reenq_users() sees dsq->id == SCX_DSQ_INVALID and
hits the BUG_ON. destroy_dsq() doesn't flush pending drus, so just skip.
tj: Read dsq->id once with READ_ONCE(). Reading it separately in the INVALID
check and the BUG_ON would leave a window where destroy_dsq() can
invalidate the id between the two reads and still trigger the BUG_ON.
Fixes: 84b1a0ea0b ("sched_ext: Implement scx_bpf_dsq_reenq() for user DSQs")
Cc: stable@vger.kernel.org # v7.1+
Signed-off-by: Tao Cui <cuitao@kylinos.cn>
Signed-off-by: Tejun Heo <tj@kernel.org>
libbpf strips the last ___flavor suffix when resolving kfunc externs, so
the bare ___v2 declaration resolves to scx_bpf_dsq_move_to_local, whose
BTF proto lacks @enq_flags. The extern never matches, bpf_ksym_exists()
returns false on every kernel that has the ___v2 kfunc, and the macro
falls back to ___v1, silently dropping @enq_flags.
Add the trailing ___compat suffix used by the other versioned externs in
this file (scx_bpf_dsq_insert___v2, scx_bpf_reenqueue_local___v2).
Any caller passing non-zero @enq_flags through the compat macro silently
loses them.
Fixes: 860683763e ("sched_ext: Add enq_flags to scx_bpf_dsq_move_to_local()")
Cc: stable@vger.kernel.org # v7.1+
Assisted-by: Z.ai:glm-5.2
Signed-off-by: fangqiurong <fangqiurong@kylinos.cn>
Signed-off-by: Tejun Heo <tj@kernel.org>
scx_bpf_events() always reads the root scheduler's event counters, so a
sub-scheduler program querying its own events silently gets the root's
instead and has no BPF-visible way to read its own (the per-scheduler sysfs
"events" file is the only interface). Resolve the scheduler from the calling
program with scx_prog_sched(). Unassociated programs follow the usual
scx_prog_sched() resolution: the root scheduler under a pre-sub-attach
compat root and zeroed counters otherwise.
Also fix up the malformed comment into proper kerneldoc.
Signed-off-by: Tejun Heo <tj@kernel.org>
scx_root_disable() invalidates each rq's clock before taking the rq lock.
scx_rq_clock_invalidate() is a plain read-modify-write of rq->scx.flags and
every other writer of the word runs under the rq lock, so the unlocked
update can race a concurrent flags update and lose one side's bits.
The invalidation doesn't matter in the first place. The cached clock is read
only by scx_bpf_now() from a loaded scheduler's BPF programs, nothing can
re-validate the clock while sched_ext is disabled as scx_rq_clock_update()
is gated on scx_enabled() too, and the usual rq lock cycles under the next
scheduler refresh or invalidate it before it's practically observable. Drop
the invalidation instead of fixing the locking.
v2: Description and comment updated - the invalidation is unnecessary rather
than subsumed by the rq lock cycle below.
Fixes: 3a9910b590 ("sched_ext: Implement scx_bpf_now()")
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Changwoo Min <changwoo@igalia.com>
Fix missing pci_disable_device() in probe and remove.
Use pcim_enable_device(), pcim_request_region(), devm_ioremap(),
devm_ioremap_uc() and devm_ioremap_wc() for the PCI path. Convert
aux_start to devm_request_mem_region(). Guard atyfb_remove() to
only unmap/release for non-PCI (Atari) devices. Keep iounmap for
sprite.addr outside the guard since it uses raw ioremap().
Signed-off-by: Shixiong Ou <oushixiong@kylinos.cn>
Signed-off-by: Helge Deller <deller@gmx.de>
Fix missing pci_disable_device() in probe and remove.
Use pcim_enable_device(), devm_request_mem_region(),
devm_ioremap() and devm_ioremap_wc() to replace manual resource
management. Remove all release_mem_region() and iounmap() calls.
Use devm_request_mem_region() instead of pcim_request_region()
because the requested sizes (16384 for MMIO, maxvram for FB) do
not match the full PCI BAR sizes.
Signed-off-by: Shixiong Ou <oushixiong@kylinos.cn>
Signed-off-by: Helge Deller <deller@gmx.de>
Fix missing pci_disable_device() in probe and remove.
Use pcim_enable_device(), pcim_request_all_regions(),
devm_ioremap() and devm_ioremap_wc() to replace manual resource
management. Remove all pci_release_regions() and iounmap() calls.
Merge failed_init label into failed_enable.
Signed-off-by: Shixiong Ou <oushixiong@kylinos.cn>
Signed-off-by: Helge Deller <deller@gmx.de>
Fix missing pci_disable_device() in probe and remove.
Use pcim_enable_device(), pcim_request_all_regions(),
devm_ioremap() and devm_ioremap_wc() to replace manual resource
management. Remove all pci_release_regions() and iounmap() calls.
Signed-off-by: Shixiong Ou <oushixiong@kylinos.cn>
Signed-off-by: Helge Deller <deller@gmx.de>
Fix missing pci_disable_device() in probe and remove.
Use pcim_enable_device(), pcim_request_region(), devm_ioremap()
and devm_ioremap_wc() to replace manual resource management.
Remove all release_mem_region() and iounmap() calls.
Signed-off-by: Shixiong Ou <oushixiong@kylinos.cn>
Signed-off-by: Helge Deller <deller@gmx.de>
Add LANDLOCK_RESTRICT_SELF_NO_NEW_PRIVS to the default flag setting.
Gate the flag on the ABI version, but do not expose any userspace
control over this flag as it has no practical effect on the resulting
sandbox.
Signed-off-by: Justin Suess <utilityemal77@gmail.com>
Link: https://patch.msgid.link/20260809154544.1253100-6-utilityemal77@gmail.com
Signed-off-by: Mickaël Salaün <mic@digikod.net>
Document setting no_new_privs with ruleset enforcement, following the
same compatibility section style as previous ABI additions.
Include a section explaining the tradeoffs of setting no_new_privs
through any means for privileged users of Landlock.
Signed-off-by: Justin Suess <utilityemal77@gmail.com>
Link: https://patch.msgid.link/20260809154544.1253100-5-utilityemal77@gmail.com
Signed-off-by: Mickaël Salaün <mic@digikod.net>
Check that a successful landlock_restrict_self(2) call with
LANDLOCK_RESTRICT_SELF_NO_NEW_PRIVS sets no_new_privs without a prior
prctl(2) call nor CAP_SYS_ADMIN, that a failed call from both an
invalid ruleset and hitting the layer maximum leaves the attribute
unchanged, and that LANDLOCK_RESTRICT_SELF_TSYNC extends it to sibling
threads. Also check that this flag requires a ruleset.
Turn the multi_threaded_success test into a multi_threaded fixture with
success, no_new_privs, and no_new_privs_max_layers variants to factor
out the threading code.
Finally, rename restrict_self_fd_logging_flags to
restrict_self_fd_flags, and restrict_self_logging_flags to
restrict_self_flags to indicate that non-logging flags are now tested.
Test coverage for security/landlock is 91.8% of 2373 lines according to
LLVM 22.
Signed-off-by: Justin Suess <utilityemal77@gmail.com>
Link: https://patch.msgid.link/20260809154544.1253100-4-utilityemal77@gmail.com
Signed-off-by: Mickaël Salaün <mic@digikod.net>
Add a landlock_restrict_self(2) flag to set the no_new_privs attribute
of the calling thread only after enforcement of the ruleset:
no_new_privs is set if and only if the call succeeds. This removes the
need for a prior prctl(2) PR_SET_NO_NEW_PRIVS call and guarantees that
a failed enforcement leaves the attribute unchanged.
Because no_new_privs is set by the call itself, the no_new_privs /
CAP_SYS_ADMIN requirement of landlock_restrict_self(2) is fulfilled by
construction, and the related EPERM check is skipped.
Unlike LANDLOCK_RESTRICT_SELF_LOG_SUBDOMAINS_OFF, this flag always
requires a valid ruleset: with a ruleset_fd of -1, such a call would be
nothing more than a Landlock-flavored prctl(2) PR_SET_NO_NEW_PRIVS, and
there is no valid use case for setting no_new_privs (possibly with
LANDLOCK_RESTRICT_SELF_TSYNC) without also enforcing Landlock
restrictions. Rejecting these calls also keeps the option of giving
them a meaning later.
The attribute is only set past the last point of failure, just before
committing the new credentials. When combined with
LANDLOCK_RESTRICT_SELF_TSYNC, no_new_privs is set on the sibling threads
as well, in their commit phase, with the same ordering.
Bump the Landlock ABI version to 11, and include the minimal related
test changes to keep the tests bisectable.
Cc: Mickaël Salaün <mic@digikod.net>
Signed-off-by: Justin Suess <utilityemal77@gmail.com>
Link: https://patch.msgid.link/20260809212459.2427878-1-utilityemal77@gmail.com
Signed-off-by: Mickaël Salaün <mic@digikod.net>
landlock_restrict_self(2) currently checks the no_new_privs /
CAP_SYS_ADMIN requirement before validating the flags argument. An
unprivileged caller without no_new_privs thus gets EPERM even when the
passed flags are invalid, hiding the EINVAL error.
Move the no_new_privs / CAP_SYS_ADMIN check just after the flags check
so that malformed calls consistently error out with EINVAL whatever the
caller's privileges, the same way seccomp(2) validates its flags before
checking no_new_privs.
Update the restrict_self_checks_ordering test accordingly.
Cc: Mickaël Salaün <mic@digikod.net>
Signed-off-by: Justin Suess <utilityemal77@gmail.com>
Link: https://patch.msgid.link/20260809154544.1253100-2-utilityemal77@gmail.com
Signed-off-by: Mickaël Salaün <mic@digikod.net>
The documentation embeds the canonical erratum documentation from the
header file, which is already a self-contained description of the issue.
Signed-off-by: Günther Noack <gnoack@google.com>
Link: https://patch.msgid.link/20260813093157.1436894-7-gnoack@google.com
[mic: Update the documentation date]
Signed-off-by: Mickaël Salaün <mic@digikod.net>
Even though OverlayFS uses vfs_rename() with RENAME_WHITEOUT on its
backing directories, and even though RENAME_WHITEOUT requires
LANDLOCK_ACCESS_FS_MAKE_REG, a process that renames non-regular files in
an OverlayFS can do so without having the LANDLOCK_ACCESS_FS_MAKE_REG
right in that location.
This works, and is supposed to work, because the changes to the backing
directories are done by OverlayFS, not by the originator task that did
the original rename() on the OverlayFS mount. Therefore, the changes
done to backing directories are not subject to the originator task's
credentials.
Test coverage for security/landlock is 91.8% of 2368 lines according to
LLVM 22.
Signed-off-by: Günther Noack <gnoack@google.com>
Link: https://patch.msgid.link/20260813093157.1436894-6-gnoack@google.com
[mic: Add test coverage, reflow commit message]
Signed-off-by: Mickaël Salaün <mic@digikod.net>
Add audit_layout1.make_whiteout: This test looks similar to
audit_layout1.make_char, but creates a whiteout object through mknod().
Since whiteout object creation is now guarded with
LANDLOCK_ACCESS_FS_MAKE_REG rather than LANDLOCK_ACCESS_FS_MAKE_CHAR, it
also needs to log the matching denial to audit.
Signed-off-by: Günther Noack <gnoack@google.com>
Link: https://patch.msgid.link/20260813093157.1436894-5-gnoack@google.com
Signed-off-by: Mickaël Salaün <mic@digikod.net>
Add tests to check that whiteout object creation is guarded by
LANDLOCK_ACCESS_FS_MAKE_REG, in the cases where these are created from
userspace:
* Conventional creation with mknod()
* Linking or renaming an existing whiteout object
* renameat2() with RENAME_WHITEOUT,
which creates a new whiteout object in the source location
* renameat2() with RENAME_EXCHANGE,
with one of the renamed objects being a whiteout object
Signed-off-by: Günther Noack <gnoack@google.com>
Link: https://patch.msgid.link/20260813093157.1436894-4-gnoack@google.com
[mic: Update commit message as requested]
Signed-off-by: Mickaël Salaün <mic@digikod.net>
Whiteout objects are used in the upper layer of an OverlayFS to
indicate that the file with this name does not exist in the unified
view, even if it is present in one of the lower layer file systems.
For the userspace implementations of OverlayFS (fuse-overlayfs),
whiteout objects can be created from userspace as well:
* mknod(2) with S_IFCHR and makedev(0, 0)
* renameat2(2) with RENAME_WHITEOUT,
creating the whiteout in the old place of the moved file.
This commit guards whiteout creation in both of these cases with
LANDLOCK_ACCESS_FS_MAKE_REG. Whiteout objects are *not* considered
character devices and are not bound to a driver.
LANDLOCK_ACCESS_FS_MAKE_REG describes the same permission class as a
whiteout object: creating one is the only S_IFCHR creation that the VFS
exempts from CAP_MKNOD, so it is as unprivileged as creating a regular
file, while LANDLOCK_ACCESS_FS_MAKE_CHAR and
LANDLOCK_ACCESS_FS_MAKE_BLOCK keep meaning the creation of devices that
expose a kernel interface [1].
For the mknod(2) case, introduce a Landlock erratum. The creation of
whiteout objects through mknod(2) was previously guarded using
LANDLOCK_ACCESS_FS_MAKE_CHAR, and it is now guarded using
LANDLOCK_ACCESS_FS_MAKE_REG.
For the renameat2(2) case, fix a bug: Before this commit, renameat2(2)
with RENAME_WHITEOUT would create a directory entry even when all
LANDLOCK_ACCESS_FS_MAKE_* rights were denied.
This does not affect normal renames within layered OverlayFS mounts:
When doing a regular rename() on a mounted fuse-overlayfs, it is the
fuse-overlayfs daemon that exercises renameat2() with RENAME_WHITEOUT,
and only the Landlock domain of that daemon is checked there.
Suggested-by: Christian Brauner <brauner@kernel.org>
Suggested-by: Mickaël Salaün <mic@digikod.net>
Cc: stable@vger.kernel.org
Fixes: cb2c7d1a17 ("landlock: Support filesystem access-control")
Depends-on: 49c9e09d96 ("landlock: Fix handling of disconnected directories")
Depends-on: fe72ce6710 ("landlock: Add errata documentation section")
Signed-off-by: Günther Noack <gnoack@google.com>
Link: https://patch.msgid.link/20260720.chow9ohYie5b@digikod.net [1]
Link: https://patch.msgid.link/20260813093157.1436894-3-gnoack@google.com
[mic: Record why LANDLOCK_ACCESS_FS_MAKE_REG is the matching right, and
add link(2) to the user doc]
Signed-off-by: Mickaël Salaün <mic@digikod.net>
By passing a (0, 0) device number, the audit test for
LANDLOCK_ACCESS_FS_MAKE_CHAR was accidentally creating a whiteout object
rather than a char device. In preparation to treating whiteout objects
differently, use an actual character device instead.
Signed-off-by: Günther Noack <gnoack@google.com>
Link: https://patch.msgid.link/20260813093157.1436894-2-gnoack@google.com
Signed-off-by: Mickaël Salaün <mic@digikod.net>
Vineet Gupta says:
====================
bpf, x86: fix per-CPU address resolution into an extended register
The JIT resolves a per-CPU address with
add <dst>, gs:[this_cpu_off]
but builds the REX prefix with add_1mod(), which sets REX.B. The
destination is encoded in ModRM.reg, which REX.R extends, and the memory
operand is disp32 with no base, so REX.B does nothing and the high
register bit is dropped. Every extended destination therefore resolves
into whichever register shares the low three bits:
R5 -> RAX R7 -> RBP R8 -> RSI R9 -> RDI
The address is left unadjusted and an unrelated register is clobbered.
Patch 1 switches to add_2mod() so the bit goes through REX.R.
Clang reloads the address into R1 before each per-CPU access, so the
destination is never an extended register and the bug has been dormant
since v6.10. GCC keeps several per-CPU addresses live at once, which is
how it turned up: test_progs-bpf_gcc panics the kernel in
global_percpu_data/init, with the address of a .percpu variable in R5.
Patch 2 covers every register. A functional test only catches this if
the address happens to land in an extended register, so the test matches
the JITed add instead.
Changes in v3:
- Fold the five per-register programs into one that loads every
register, and drop the comment explaining the register choice
(Eduard Zingerman).
- Move the percpu_data declaration inside the arch guard, so other
targets no longer carry a .percpu section and an unused map (bpf-ci).
- Match the movabsq of each address as well as the add, so the matchers
stay on consecutive lines and the pair is checked to use the same
register.
- Restore the Reviewed-by on patch 1, dropped by mistake in v2.
Changes in v2:
- Add the selftest, patch 2/2 (Eduard Zingerman). It uses __jited()
rather than __xlated(): the xlated stream is identical for every
register, and the wrong prefix is only visible in the native encoding.
- No functional change to patch 1.
====================
Link: https://patch.msgid.link/20260814220254.3797467-1-vineet.gupta@linux.dev
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
An ld_imm64 of a per-CPU map value is followed by a mov_percpu_addr that
reuses the same register, so which register the address lands in decides
how the JIT encodes the add. Getting the REX prefix wrong there is
invisible to a functional test unless the address happens to land in an
extended register, which is why this went unnoticed.
Load a .percpu variable into every register in one program and match the
JITed add against the register each one must resolve into.
Signed-off-by: Vineet Gupta <vineet.gupta@linux.dev>
Link: https://patch.msgid.link/20260814220254.3797467-3-vineet.gupta@linux.dev
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
The destination of the per-CPU address MOV is encoded in ModRM.reg,
which is extended by REX.R, but the REX prefix is built with
add_1mod(), which sets REX.B. REX.B extends ModRM.rm and SIB.base, and
this instruction addresses memory as disp32 with no base, so the bit
has no effect at all and the high register bit is simply lost.
Every is_ereg() destination therefore resolves to the wrong register,
picking whichever one shares the low three bits:
R5 -> RAX R7 -> RBP R8 -> RSI R9 -> RDI
With BPF_REG_5, whose reg2hex is 0, the emitted
65 49 03 04 25 <off> add %gs:<off>,%rax
adds the per-CPU offset to RAX rather than R8. The destination keeps
the unadjusted address and RAX is clobbered, so the program goes on to
dereference a pointer that was never made per-CPU:
BUG: unable to handle page fault for address: 0000607e386a8894
RIP: bpf_prog_707837aafd2aa9ae_update_percpu_data+0x93/0xc9
Call Trace:
__bpf_prog_test_run_raw_tp+0x2dc/0x7d0
__flush_smp_call_function_queue+0x1e9/0xc80
Kernel panic - not syncing: Fatal exception in interrupt
R5 is the mildest of the four, aliasing a scratch register and faulting
at the store. R7 aliases RBP and would corrupt the frame pointer, R8
and R9 alias the argument registers.
Use add_2mod() so the register goes through REX.R, matching how
add_2reg() places it in ModRM.reg and how emit_priv_frame_ptr()
hardcodes 0x4c for the same instruction with R9. Encodings for the
non-extended registers are unchanged.
Problem showed up when trying to resurrect BPF_GCC CI (selftests built
with BPF_GCC).
This has gone unnoticed because clang reloads the address into R1
before each per-CPU access, so the destination is never an extended
register. GCC keeps several per-CPU addresses live at once, and
test_progs-bpf_gcc panics the kernel in global_percpu_data/init, where
the address of a .percpu variable ends up in R5.
Fixes: 7bdbf74463 ("bpf: add special internal-only MOV instruction to resolve per-CPU addrs")
Signed-off-by: Vineet Gupta <vineet.gupta@linux.dev>
Reviewed-by: Eduard Zingerman <eddyz87@gmail.com>
Cc: stable@vger.kernel.org
Link: https://patch.msgid.link/20260814220254.3797467-2-vineet.gupta@linux.dev
Signed-off-by: Eduard Zingerman <eddyz87@gmail.com>
When a function-like macro expands to an expression, that expression
doesn't need a semicolon after it. All uses have been verified to
have their own semicolons.
This was found using the following Coccinelle semantic patch:
@r@
identifier i : script:ocaml() { String.lowercase_ascii i = i };
expression e;
@@
*#define i(...) e;
Signed-off-by: Julia Lawall <Julia.Lawall@inria.fr>
Signed-off-by: Stafford Horne <shorne@gmail.com>
restore_sigcontext() copies the whole supervision register (SR) from the
signal frame and only clears SPR_SR_SM before the value is reloaded into
the hardware SR (through ESR and l.rfe) on the return to user space. All
other SR bits are left under user control.
An unprivileged task can thus return from a signal handler through a
crafted sigframe that clears SPR_SR_DME. With the data MMU disabled the
CPU performs no translation or protection on data accesses, so the task
gains read and write access to arbitrary physical memory, a local
privilege escalation. SPR_SR_IME, SPR_SR_SUMRA, SPR_SR_LEE, SPR_SR_EPH
and the cache-enable bits are exposed the same way. The ptrace GPR regset
already refuses any change to SR for exactly this reason.
Restore only the arithmetic flag bits (F, CY, OV) from the signal frame
and take every privileged control bit from the SR the kernel saved on
signal entry.
Verified with qemu-system-or1k -M or1k-sim: before this change an
unprivileged PoC clears SPR_SR_DME in rt_sigreturn and writes a marker to
physical address 0x03000000 (beyond the kernel's mem=32M); afterwards the
same PoC receives SIGSEGV and physical memory is unchanged.
Fixes: ac689eb7f9 ("OpenRISC: Signal handling")
Cc: stable@vger.kernel.org
Signed-off-by: Ali Ahmet Memis <ali@iusegentoo.com>
Signed-off-by: Stafford Horne <shorne@gmail.com>
- A couple of fixes to the sur40 touchscreen driver to correct
registration and teardown ordering, and to fix error path unwinding
when video device registration fails.
-----BEGIN PGP SIGNATURE-----
iHUEABYKAB0WIQST2eWILY88ieB2DOtAj56VGEWXnAUCan/qAgAKCRBAj56VGEWX
nIwLAQCryq8oWRzTY1H7QurNDJ89vSwdRJ8CxiLqQgJmWCzgzQD/Vr9SJmWacfn1
TEtDiqHoRSk2wOrZaSDRgsWTH2BOYQI=
=RCXW
-----END PGP SIGNATURE-----
Merge tag 'input-for-v7.2-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
Pull input fixes from Dmitry Torokhov:
- A couple of fixes to the sur40 touchscreen driver to correct
registration and teardown ordering, and to fix error path
unwinding when video device registration fails.
* tag 'input-for-v7.2-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input:
Input: sur40 - fix V4L error path cleanup
Input: sur40 - fix input device registration ordering
The while loop declares a second err variable that shadows the outer
one. When the crypt callback fails, the goto out path returns the outer
err, which still holds the -ENOMEM value assigned before the successful
allocation check. The real error from the cipher is discarded and the
caller sees -ENOMEM instead.
Drop the inner declaration so the callback error reaches the caller.
Verified with a test module that registers an lskcipher whose encrypt
callback fails with -EIO and calls it through a misaligned buffer.
An unpatched kernel returns -ENOMEM, a patched kernel returns -EIO.
Found with Clang's -Wshadow.
Fixes: 31865c4c4d ("crypto: skcipher - Add lskcipher")
Assisted-by: Claude:claude-fable-5
Signed-off-by: Karl Mehltretter <kmehltretter@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Use crypto_memneq() for constant-time comparison.
The CCM path in ocs-aes.c verifes the received authentication tag with
memcmp(), which returns early on the first mismatched byte. This leaks
valid-prefix length and allows for valid tag forgery which violates the
INT-CTXT guarantee of AEAD.
Assisted-by: gregkh_clanker_t1000
Signed-off-by: David C.C.M. Gall <david.ccm.gall@googlemail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Use crypto_memneq() for constant-time comparison.
The GCM path in keembay-ocs-aes-core.c verifes the received
authentication tag with memcmp(), which returns early on the first
mismatched byte. This leaks valid-prefix length and allows for valid
tag forgery which violates the INT-CTXT guarantee of AEAD.
Assisted-by: gregkh_clanker_t1000
Signed-off-by: David C.C.M. Gall <david.ccm.gall@googlemail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Use crypto_memneq() for a constant-time comparison.
sa_aead_dma_in_callback() compares the computed authentication tag
against the received tag with memcmp(), which short-circuits on the
first differing byte. An attacker who can submit decrypt requests and
observe completion latency could recover the expected tag byte by byte.
Valid tag forgery for AEAD breaks the INT-CTXT guarantee.
Assisted-by: gregkh_clanker_t1000
Signed-off-by: David C.C.M. Gall <david.ccm.gall@googlemail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Use a named initializer for the acpi_device_id fields which
makes the code more readable and consistent with how lists
are initialized in the rest of the kernel code base. Also
drop explicitly setting fields to 0 where it is redundant.
Signed-off-by: Pawel Zalewski (The Capable Hub) <pzalewski@thegoodpenguin.co.uk>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
The AAD buffer allocated in qce_aead_ccm_prepare_buf_assoclen()
can be smaller than the length later programmed into the DMA
scatterlist.
The allocation size is currently calculated as:
ALIGN(assoclen, 16) + MAX_CCM_ADATA_HEADER_LEN
while the DMA length is set to:
ALIGN(assoclen + adata_header_len, 16)
Since ALIGN() does not distribute over addition, the allocation
can be smaller than the DMA length. For example, when
assoclen = 32 and adata_header_len = 2:
allocation = ALIGN(32, 16) + 6 = 38
DMA length = ALIGN(32 + 2, 16) = 48
As a result, the QCE hardware can read beyond the allocated
buffer while computing the CBC-MAC over the associated data.
The extra bytes are folded into the authentication tag,
resulting in an incorrect tag and causing CCM self-test
failures such as:
alg: aead: ccm-aes-qce encryption test failed (wrong result)
on test vector 8
Fix the allocation by adding the maximum possible AAD header
length before alignment:
ALIGN(assoclen + MAX_CCM_ADATA_HEADER_LEN, 16)
This guarantees that the allocated buffer is large enough
for the fully padded AAD data for all supported header sizes.
Cc: stable@vger.kernel.org
Fixes: 9363efb418 ("crypto: qce - Add support for AEAD algorithms")
Signed-off-by: Md Sadre Alam <md.alam@oss.qualcomm.com>
Reviewed-by: Bartosz Golaszewski <bartosz.golaszewski@oss.qualcomm.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
On a hardware analytics error, decompress retries through the software
fallback, which writes req->dst with the CPU while it is still mapped
DMA_FROM_DEVICE. With SWIOTLB active the later dma_unmap_sg() copies the
stale bounce buffer over req->dst, corrupting the result.
Unmap before the fallback runs. The async path unmaps inline; the sync
path signals the retry with -EAGAIN so iaa_comp_adecompress() runs the
fallback after unmapping.
Fixes: 2ec6761df8 ("crypto: iaa - Add support for deflate-iaa compression algorithm")
Cc: stable@vger.kernel.org
Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Since commit e2c3b6b21c ("mm: zswap: use SG list decompression APIs
from zsmalloc"), zswap passes the raw zsmalloc SG list directly to
crypto drivers, so a compressed object spanning multiple pages reaches
IAA as a multi-entry source. Such requests currently fall back to
software decompression.
As IAA hardware requires a single DMA source buffer, linearize small
multi-entry sources into a pre-allocated bounce page and submit that to
the hardware instead of falling back to software. Keep the software
fallback only for multi-entry destinations. This recovers most of the
performance lost by using the software fallback.
Store the bounce-page state in the acomp request context alongside the
existing compression CRC, free it through a shared source-unmap helper,
and back the pages with a small module-wide mempool so the path remains
available in reclaim-driven callers.
Signed-off-by: Giovanni Cabiddu <giovanni.cabiddu@intel.com>
Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
When decompression falls back to deflate-generic after an analytics
error, the request no longer completes through IAA.
Move decompression byte accounting into the successful IAA completion
path in both the synchronous and asynchronous flows so decomp_bytes only
reflects bytes actually processed by IAA.
Signed-off-by: Giovanni Cabiddu <giovanni.cabiddu@intel.com>
Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Reviewed-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
IAA cannot process source or destination scatterlists with more than one
entry directly. Instead of failing these requests, route them through a
separate deflate acomp transform and keep the request alive in software.
The IAA driver has never handled multi-entry scatterlists, but the
limitation was latent until commit e2c3b6b21c ("mm: zswap: use SG list
decompression APIs from zsmalloc") made zswap pass the raw zsmalloc SG
list directly to crypto drivers, so objects spanning multiple pages now
reach IAA as multi-entry sources and would otherwise fail decompression.
Fallback to the generic DEFLATE implementation for scatterlists with
more than one entry. After the multi-entry cases fall back early,
simplify the DMA mapping path to a single scatterlist entry and fall
back on mapping failure as well.
Add counters to track the number of requests processed by the software
implementation on the compression direction.
Fixes: 2ec6761df8 ("crypto: iaa - Add support for deflate-iaa compression algorithm")
Fixes: e2c3b6b21c ("mm: zswap: use SG list decompression APIs from zsmalloc")
Cc: stable@vger.kernel.org
Signed-off-by: Giovanni Cabiddu <giovanni.cabiddu@intel.com>
Signed-off-by: Vinicius Costa Gomes <vinicius.gomes@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
In the CCM B0 block the message-length field Q spans L bytes, where
L (cl in the driver) is derived from the cipher IV flags byte as
c_ivin[0] + 1. set_aead_auth_iv() hardcoded writing only the last 2
bytes of a_ivin with cryptlen, implicitly assuming cl = 2.
When cl = 3 (a shorter nonce yielding a 3-byte length field) and the
packet is longer than 65535 bytes, cryptlen no longer fits in 2 bytes.
The dropped high byte made the auth IV built by the driver differ from
the one consumed by the hardware, so the software/hardware comparison
failed and the CCM request errored out.
Write the last cl bytes of a_ivin in a loop driven by the IV's CL
value, so the length-field width always matches the algorithm
configuration instead of assuming a fixed 2-byte field.
Fixes: c16a70c1f2 ("crypto: hisilicon/sec - add new algorithm mode for AEAD")
Signed-off-by: Zhushuai Yin <yinzhushuai@huawei.com>
Signed-off-by: Chenghai Huang <huangchenghai2@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Embed the single ring as a flexible array member in eip93_device
instead of allocating it separately. This simplifies the probe path
and uses struct_size() for a single allocation.
Assisted-by: opencode:big-pickle
Signed-off-by: Rosen Penev <rosenp@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
crypto_krb5_prepare_encryption() and crypto_krb5_prepare_checksum()
free the buffer holding the freshly derived keys with plain kfree(),
leaving the key material behind in the freed slab object.
Fixes: 3936f02bf2 ("crypto/krb5: Implement Kerberos crypto core")
Cc: stable@vger.kernel.org
Signed-off-by: Jan Sebastian Götte <linux@jaseg.de>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
scheduler:
- revert fair scheduler patches due to regression
- mark fair as experimental
connector:
- fix OOB read in hdmi audio infoframe
log:
- fix divide by 0 if module param is set to 0
- fix OOB read on empty message
- fix infinite loop for too large scale
xe:
- Fix DPT Allocation paths
- Fixes around UM queue BO
- Order ring writes before ring tail updates
- Add termination on resume for PXP
- Document Sentinel and make CTX_TIMESTAMP read TOCTOU-safe
- Fix sync entry leak on OA config emit failure
- Check managed mutex initilization errors
- Fix min frequency setting
- Fix xe_device_probe error path
amdgpu:
- Bounds checking fix in CS IOCTL
- Bounds checking fix in GEM IOCTL
- Display fixes
- GPUVM fix
- ASPM fix
- UVD bounds checking fixes
- VCE 3 fix
- BT.2020 fixes
- NBIF 6.3.1 fix
- IP discovery fix
radeon:
- Runtime pm fix
amdxdna:
- skip attempting to populate unmapped pages
-----BEGIN PGP SIGNATURE-----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=rauR
-----END PGP SIGNATURE-----
Merge tag 'drm-fixes-2026-08-15' of https://gitlab.freedesktop.org/drm/kernel
Pull drm fixes from Dave Airlie:
"While this is large for rc8 time but also AI driven fixes is a lot of
it, we had a more traditional screw up, and a regression was just
found in the fair scheduling patches that went in back in rc1. This
reverts the fair scheduler back to an option and sets the default back
to what it should have been. We might have been a bit overly zealous
in switching over, but at least it feels more normal than the AI
driven fixes.
Apart from the scheduler, it's mostly amdgpu and xe fixes, with some
misc fixes to the log code and connector code.
scheduler:
- revert fair scheduler patches due to regression
- mark fair as experimental
connector:
- fix OOB read in hdmi audio infoframe
log:
- fix divide by 0 if module param is set to 0
- fix OOB read on empty message
- fix infinite loop for too large scale
xe:
- Fix DPT Allocation paths
- Fixes around UM queue BO
- Order ring writes before ring tail updates
- Add termination on resume for PXP
- Document Sentinel and make CTX_TIMESTAMP read TOCTOU-safe
- Fix sync entry leak on OA config emit failure
- Check managed mutex initilization errors
- Fix min frequency setting
- Fix xe_device_probe error path
amdgpu:
- Bounds checking fix in CS IOCTL
- Bounds checking fix in GEM IOCTL
- Display fixes
- GPUVM fix
- ASPM fix
- UVD bounds checking fixes
- VCE 3 fix
- BT.2020 fixes
- NBIF 6.3.1 fix
- IP discovery fix
radeon:
- Runtime pm fix
amdxdna:
- skip attempting to populate unmapped pages"
* tag 'drm-fixes-2026-08-15' of https://gitlab.freedesktop.org/drm/kernel: (51 commits)
drm/log: Fix infinite loop when scale is too large for display
drm/log: Fix out-of-bounds read on empty message length
drm/log: Fix division by zero when scale module parameter is 0
drm/xe: Fix xe_device_probe() failure
drm/xe: Fix a bug in pc_adjust_freq_bounds()
drm/xe/oa: Check managed mutex initialization errors
drm/xe/oa: Fix sync entry leak on OA config emit failure
drm/xe/lrc: document sentinel and make CTX_TIMESTAMP read TOCTOU-safe
drm/xe/pxp: add termination on resume
drm/xe: Order ring writes before ring tail updates
drm/xe/guc_ads: use uncached mapping for UM queue BO
drm/xe/guc_ads: allocate UM queues in VRAM on dGFX
drm/xe/guc_ads: allocate UM queues in a separate BO
drm/xe: Fix DPT allocation paths.
accel/amdxdna: Skip unmapped range in aie2_populate_range()
drm/amdgpu: Prefer default discovery offset
drm/amdgpu: Reject UVD message with invalid number of h265 refs
drm/amdgpu: fix nbif 6.3.1 l1 low power not functional
drm/amd/display: fix BT.2020 YCbCr output CSC matrices for DCE
drm/amd/display: fix BT.2020 YCbCr limited output CSC matrix
...
Changes:
Make expedited grace periods expedite normal RCU callbacks
Miscellaneous fixes:
* Improve diagnostic output with character task states.
* Mark accesses to inform KCSAN of concurrency design.
* Move from kmalloc() to kmalloc_obj().
* Documentation updates.
* Improve handling of RCU deferred quiescent states.
* Clean up unused function arguments and structure fields.
* Reduce show_rcu_gp_kthreads() stack space.
Tasks RCU updates:
* Clean up after SRCU re-implementation of Tasks Trace RCU.
* Mark accesses to inform KCSAN of concurrency design.
* Add ->lazy_timer status to diagnostic output.
* Remove an unnecessary memory barrier.
* Fix a data race, courtesy of KCSAN.
* Documentation updates.
* Convert cond_resched_tasks_rcu_qs() from macro to static inline
function.
SRCU updates:
* Add Rust helpers for SRCU.
* Avoid losing queued work at cleanup_srcu_struct() time.
Torture-test updates:
* Preparation work for immediate RCU priority deboosting.
* Test RCU readers from real interrupt handlers (as opposed to softirq).
* Simplify code through use of cpumask_next_wrap().
* Improve diagnostic output with character task states.
* Add rcutorture.nwriters parameter to allow lightweight stall testing,
and rcutorture.stall_only to make doing so easier.
* Test an RCU Tasks Trace grace period implying an RCU grace period.
* Make RCU Tasks Trace torturing track reader batches.
* Fix a data race, courtesy of KCSAN.
* Plug a shuffle_tmp_mask memory leak on kthread spawn failure.
