[PATCH v6 0/2] security: Expand task_setscheduler LSM hook
From: Aaron Tomlin
Date: Wed Nov 18 2026 - 23:07:30 EST
This series expands the task_setscheduler LSM hook to accept nullable
pointers to the requested sched_attr and cpumask alongside the target
task_struct. For sched_setaffinity(), passing the raw user-supplied
affinity mask allows LSMs such as the BPF LSM to audit or filter the
mask as requested by userspace before __sched_setaffinity() intersects
it with the task's allowed cpuset CPUs, which seccomp filters cannot
inspect behind user pointers and scheduler tracepoints only observe
post-intersection. Similarly, __sched_setscheduler() passes a local
sched_attr populated with the resolved scheduling policy and effective
SCHED_FLAG_RESET_ON_FORK state so parameter-only updates expose the
effective values rather than internal sentinels or caller defaults, and
cpuset task migrations and CLONE_INTO_CGROUP pass the destination
cpuset's effective_cpus mask (note that direct writes to cpuset.cpus
update task affinity via cpuset_update_tasks_cpumask() without invoking
this hook).
Because BPF_LSM_MAC trampolines place the previous return value in
context argument slot nr_args, expanding task_setscheduler from one
parameter to three shifts ret from slot 1 to slot 3; existing
two-argument BPF_PROG(name, struct task_struct *p, int ret) programs on
lsm/task_setscheduler will therefore be rejected by the verifier until
updated to the expanded signature. Patch 2 extends the BPF LSM
selftests to verify the hook's runtime arguments and -EPERM mediation
across sched_setscheduler(), sched_setparam(), sched_setattr() with
SCHED_FLAG_KEEP_POLICY, sched_setaffinity(), and
/proc/[pid]/timerslack_ns, alongside verifier checks for the __nullable
annotations.
Changes since v5:
- Replaced the second paragraph of Patch 1's commit message with the
concrete rationale for inspecting the user-requested affinity mask
and scheduling attributes prior to cpuset intersection, and clarified
the scope of affinity mediation relative to cpuset.cpus writes
(Chris Down, Peter Zijlstra)
- In __sched_setscheduler(), synchronised SCHED_FLAG_RESET_ON_FORK in
lsm_attr.sched_flags with the resolved reset_on_fork value so
parameter-only updates accurately reflect the task's effective
reset-on-fork state (Chris Down)
- Clarified the kerneldoc for security_task_setscheduler() to document
what @in_mask represents for sched_setaffinity(), cpuset cgroup
migrations, and CLONE_INTO_CGROUP (Chris Down)
- Added Patch 2 to extend the BPF LSM runtime and verifier selftests
for the expanded task_setscheduler hook
- Link to v5: https://lore.kernel.org/lkml/20261105172210.76706-1-atomlin@xxxxxxxxxxx/
Changes since v4:
- Rebased against v7.4-rc1
- In __sched_setscheduler(), populate a local copy of sched_attr with
the resolved scheduling policy before calling
security_task_setscheduler() so parameter-only updates (such as
sched_setparam() or sched_setattr() with SCHED_FLAG_KEEP_POLICY) pass
the effective policy rather than SETPARAM_POLICY (-1)
- Clarified the commit message to reflect that both the sched_attr and
cpumask parameters are introduced by this patch
- Link to v4: https://lore.kernel.org/all/20260720014156.459386-1-atomlin@xxxxxxxxxxx/
Changes since v3:
- Expanded the task_setscheduler LSM hook to accept the 'sched_attr'
payload, enabling BPF and other LSMs to inspect scheduling attributes
(Peter Zijlstra)
- Updated all call sites for security_task_setscheduler() to correctly
pass the attr structure when available, or NULL otherwise
(Peter Zijlstra)
- Added the necessary forward declaration for struct sched_attr to
include/linux/security.h
- Link to v3: https://lore.kernel.org/lkml/20260526142838.774711-1-atomlin@xxxxxxxxxxx/
Changes since v2:
- Dropped patch 1. This is to be addressed by the cgroup cpuset
maintainer (Waiman Long)
- Dropped patch 3. Will be submitted as a separate patch (Paul Moore)
- Link to v2: https://lore.kernel.org/lkml/20260509213803.968464-1-atomlin@xxxxxxxxxxx/
Changes since v1:
- Reordered the allocation and user-copy of new_mask in the MIPS
architecture's mipsmt_sys_sched_setaffinity() to occur before the
LSM hook is invoked. This ensures the security modules evaluate a fully
populated mask rather than uninitialised memory, while cleanly handling
error unwinding
- Updated cpuset_can_fork() to pass the destination cpuset's effective CPU
mask instead of NULL
- Link to v1: https://lore.kernel.org/lkml/20260509164847.939294-1-atomlin@xxxxxxxxxxx/
Aaron Tomlin (2):
security: Expand task_setscheduler LSM hook
selftests/bpf: Verify task_setscheduler LSM hook parameters
arch/mips/kernel/mips-mt-fpaff.c | 2 +-
fs/proc/base.c | 2 +-
include/linux/lsm_hook_defs.h | 4 +-
include/linux/security.h | 15 +-
kernel/cgroup/cpuset.c | 4 +-
kernel/sched/syscalls.c | 11 +-
security/commoncap.c | 9 +-
security/security.c | 21 ++-
security/selinux/hooks.c | 4 +-
security/smack/smack_lsm.c | 13 +-
.../selftests/bpf/prog_tests/test_lsm.c | 164 ++++++++++++++++++
tools/testing/selftests/bpf/progs/lsm.c | 51 ++++++
.../selftests/bpf/progs/verifier_lsm.c | 49 ++++++
13 files changed, 328 insertions(+), 21 deletions(-)
--
2.55.0