linux-toradex.git/security/apparmor/include, branch master

apparmor: fix profile leak when a policy load fails

2026-10-29T17:44:18+00:00

The error paths of aa_replace_profiles() and aa_unpack() do not break
the refcount cycle between label and proxy. They just aa_put_profile()
the new profiles, which drops the label refcount from 2 to 1, and the
proxy keeps the profile alive. Every failed atomic policy load
(e.g. "profile can not be replaced", or a later profile in the blob
failing to unpack) leaks every new profile in the set, along with its
proxy, rules and its loaddata reference.

The leak is usually not visible from userspace, because the raw_data
directory for a failed load is not created, but it becomes visible
when the blob is deduplicated onto an existing loaddata, which then
keeps its policy/raw_data/ directory forever.

Add aa_put_unpublished_profile(), which drops the profile's own proxy
before putting the profile, and use it on those error paths. The
proxy of an unpacked profile is always freshly allocated and not
referenced by anything else before the profile is published, so
putting it is safe.

Fixes: 637f688dc3dc ("apparmor: switch from profiles to using labels on contexts")
Cc: stable@vger.kernel.org
Signed-off-by: Georgia Garcia 
Signed-off-by: John Johansen

apparmor: fix a use-after-free in aa_lookupn_ns()

2026-10-29T15:30:53+00:00

aa_lookupn_ns() takes a reference on the namespace it finds with
aa_get_ns(), which is kref_get() on the reference count that the
namespace is released through, without checking whether the namespace
is still in use.  A concurrent removal of a ":"-prefixed namespace
looks the same namespace up and only then takes the parent lock to
remove it, so the last reference can be dropped between the lookup and
the aa_get_ns() call: the reference count then reports the increment on
a zero count and the kernel classifies the following use of the object
as a use-after-free.  A user with CAP_MAC_ADMIN can trigger this by
writing a ":" removal and a ":/" removal to
/sys/kernel/security/apparmor/.remove concurrently.

refcount_t: addition on 0; use-after-free.
WARNING: lib/refcount.c:25 at refcount_warn_saturate+0xc0/0xe0, CPU#6
Call Trace:
 
 aa_lookupn_ns+0xa9/0xc0
 aa_remove_profiles+0x4fb/0x730
 ? __kvmalloc_node_noprof+0x219/0x5c0
 ? __pfx_aa_remove_profiles+0x10/0x10
 ? aa_loaddata_alloc+0x65/0x1f0
 ? _copy_from_user+0x2d/0x80
 ? aa_loaddata_alloc+0x111/0x1f0
 profile_remove+0x2ee/0x4a0
 vfs_write+0x21e/0xcf0
 ? rcu_core+0x29c/0x1740
 ? timerqueue_linked_add+0x1f4/0x3d0
 ? __pfx_vfs_write+0x10/0x10
 ? mutex_lock+0x81/0xe0
 ? __pfx_mutex_lock+0x10/0x10
 ? __pfx_rcu_core+0x10/0x10
 ? __pfx_hrtimer_update_next_event+0x10/0x10
 ? mutex_unlock+0x7b/0xd0
 ? fdget_pos+0x24d/0x4b0
 ksys_write+0xf7/0x1c0
 ? __pfx_ksys_write+0x10/0x10
 ? restore_fpregs_from_fpstate+0x37/0xc0
 do_syscall_64+0xf9/0x540
 entry_SYSCALL_64_after_hwframe+0x77/0x7f
 

Fix by rejecting a namespace whose reference count is zero.

Fixes: 3664268f19ea0 ("apparmor: add namespace lookup fns()")
Cc: stable@vger.kernel.org
Signed-off-by: Binbin Deng <18983559317@163.com>
Signed-off-by: John Johansen

apparmor: fix table_size() integer overflow on 32 bit

2026-10-29T15:30:53+00:00

table_size() checks ALIGN(sizeof(struct table_header) + len * el_size, 8)
in size_t, where @len is td_lolen, a u32 user-provided value from the
policy blob. On 32 bit kernels both operations can overflow.

Triggering this bug allows to:
 - overflow a 16 byte allocation with up to 4GiB of policy data
 - store the table header at address ZERO_SIZE_PTR (kvzalloc(0))

However, this bug requires CAP_MAC_ADMIN, only affects 32 bit kernels,
and the copy runs until the machine is destroyed, so it gives no
controllable primitive and is therefore not a security issue.

Compute the size with the checked helpers and return 0 when the result is
not representable, so the callers can reject the table. A valid table is
never 0 bytes, the header alone is 12.

64 bit kernels cannot overflow (these structs are 32 bit), and are hence
not affected.

Fixes: e06f75a6a2b4 ("AppArmor: dfa match engine")
Signed-off-by: Maxime Bélair 
Signed-off-by: John Johansen

apparmor: fix refcount leak in aa_get_newest_profile()

2026-10-29T15:28:03+00:00

The labels_profile() macro is evaluating its argument twice

  #define labels_profile(X) ((X)->vec[(X)->size - 1])

unfortunately aa_get_newest_profile() passes
   aa_get_newest_label(&p->label)
so every call to aa_get_newest_profile results in 2 refcounts
being taken with one being leaked.

There is no need for labels_profile() to be a macro. Convert it to
an inline fn, which ensures all uses are passing the correct struct
and removes the double evaluation.

Fixes: 637f688dc3dc ("apparmor: switch from profiles to using labels on contexts")
Cc: stable@vger.kernel.org
Reported-by: Georgia Garcia 
Signed-off-by: John Johansen

apparmor: fix deadlock in complain-mode change_hat

2026-08-16T06:39:38+00:00

The use of change_hat when in complain mode can cause a deadlock
when the hat doesn't exist and a new learning profile is created
for the missing profile. This is because change_hat() has taken
the lock to search the hat list and creating the new learning
profile needs to take the lock to add it to the list.

From the bug report:

Originally found in 7.0.0 in LTS ubuntu 26.04 with pam_apparmor + su
in complain mode set to change hats. Then verified in newest
available vanilla kernel I've compiled to see if still present:

7.2-rc7 vanilla -> affected

checked also some other kernels:
6.18.44 vanilla -> affected
6.12.95 with debian patches -> unaffected

On systems without bug (for example 6.12.95 debian) it just prints:

aa_change_hat rc=0

On systems with bug, the executable always hangs, prints nothing and
becomes unkillable. (And once stuck this way, it will cause any
further hat changes to also cause the changing process to get stuck)

Then in syslog you can find hint about cause:

kernel: INFO: task hat:3409 blocked for more than 483 seconds.
kernel: Not tainted 7.2.0-rc7 #1
kernel: "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
kernel: task:hat state:D stack:0 pid:3409 tgid:3409 ppid:2605 task_flags:0x400000 flags:0x00080800
kernel: Call Trace:
kernel:
kernel: __schedule+0x48f/0xfe0
kernel: schedule+0x27/0xa0
kernel: schedule_preempt_disabled+0x15/0x30
kernel: __mutex_lock.constprop.0+0x569/0xa10
kernel: aa_new_learning_profile+0x15f/0x210
kernel: build_change_hat+0x19f/0x3b0
kernel: change_hat.isra.0+0x5dd/0xd60
kernel: aa_change_hat+0x2f3/0x710
kernel: aa_setprocattr_changehat+0x121/0x1f0
kernel: do_setattr+0x28c/0x340
kernel: apparmor_setselfattr+0x20/0x50
kernel: security_setselfattr+0xf6/0x110
kernel: __x64_sys_lsm_set_self_attr+0x53/0x90
kernel: do_syscall_64+0xdd/0x5e0
kernel: ? __mod_memcg_lruvec_state+0xfd/0x260
kernel: ? lruvec_stat_mod_folio+0x8d/0xd0
kernel: ? __folio_mod_stat+0x2d/0x90
kernel: ? map_anon_folio_pte_nopf+0xd1/0x1f0
kernel: ? do_anonymous_page+0x184/0xa10
kernel: ? __handle_mm_fault+0x805/0x870
kernel: ? count_memcg_events+0xef/0x230
kernel: ? handle_mm_fault+0x1f0/0x2f0
kernel: ? do_user_addr_fault+0x2bb/0x7b0
kernel: ? do_syscall_64+0x94/0x5e0
kernel: ? exc_page_fault+0x75/0x160
kernel: entry_SYSCALL_64_after_hwframe+0x76/0x7e
kernel: RIP: 0033:0x7f815e134c8d
kernel: RSP: 002b:00007fff6df94ea8 EFLAGS: 00000246 ORIG_RAX: 00000000000001cc
kernel: RAX: ffffffffffffffda RBX: 0000556d8c81d040 RCX: 00007f815e134c8d
kernel: RDX: 0000000000000046 RSI: 0000556d8c81d040 RDI: 0000000000000064
kernel: RBP: 00007fff6df94ef0 R08: 00007f815e212ac8 R09: 000000000000000c
kernel: R10: 0000000000000000 R11: 0000000000000246 R12: 0000556d8c81d010
kernel: R13: 0000000000000026 R14: 0000000000000046 R15: 0000000000000064
kernel:
kernel: INFO: task hat:3409 is blocked on a mutex likely owned by task hat:3409.

To fix the issue, lift the locking out of the core of
aa_new_learning_profile(), introduce a wrapper function that takes the
lock where needed, and have build_change_hat() call the core function
that no longer takes the lock.

In addition fix 4 other issues introduced by commit
32e92764d6f8d ("apparmor: grab ns lock and refresh when looking up changehat child profiles")
- aa_get_profile_rcu() was replaced-by: aa_get_profile without the
accompanying rcu_dereference_protected()
- an extra aa_get_label(label) was introduced at the start of
change_hat() without an accompanying aa_put_label() causing a
reference count leak.
- a reference count leak was introduced in the label_is_stale(label)
case, where the newest profile would be leaked instead of the
label passed to the function.
- a potential UAF when the lookup walks up the tree with new_ns != ns
the new label reference is put, and then used for the next lookup.
The mutex_lock, will block replacement, and removal in the locked
ns. However there are two cases where putting the reference can
result in the label being freed even with the lock held.

1. the label does not have a list reference (possible for temporary
or special profiles) in which case the put can trigger the
cleanup.
2. the new label reference is in a different namespace, which does
not have a lock held on it. This extends case 1 to also include
replacement, and removal that could be occurring in the namespace
new is in.

Reported-by: Martin Petricek
Link: https://lists.ubuntu.com/archives/apparmor/2026-August/014907.html
Fixes: 32e92764d6f8d ("apparmor: grab ns lock and refresh when looking up changehat child profiles")
Signed-off-by: John Johansen

apparmor: constify aa_label parameters on read-only query helpers

2026-08-11T05:49:43+00:00

Several label helpers only read from their struct aa_label * arguments:
they compare labels, test subset relationships, or check the mediation
bitmask, all via direct field/index access. Mark those parameters
const struct aa_label * to document intent and let the compiler enforce
that the label is not modified.

The converted functions are:
  - label_mediates(), label_mediates_safe()
  - aa_label_cmp() (and its vec_cmp() helper)
  - __aa_label_next_not_in_set(), aa_label_is_subset(),
    aa_label_is_unconfined_subset()
  - __aa_subj_label_is_cached()
  - aa_label_next_confined(), aa_label_next_in_merge()

These all access the label through direct indexing or manual iterators
rather than the label_for_each()/fn_for_each() macros, which are not
const-correct and so gate the majority of the remaining label consumers
(the print, match, and permission-check paths) from being constified.

No functional change.

Signed-off-by: John Johansen 
Assisted-by: Claude:claude-opus-4.8

apparmor: constify aa_dfa parameters on read-only compute paths

2026-08-11T05:49:43+00:00

Most uses of aa_dfa a read-only walking of the dfa. Have the compiler
enforce this.

Signed-off-by: John Johansen

apparmor: constify aa_profile parameters on read-only compute paths

2026-08-11T05:49:43+00:00

A number of functions take a struct aa_profile * argument that is only
ever read from: they compute DFA matches or apply the profile's mode
flags without modifying the profile, taking a reference on it, or
touching its embedded label. Mark those parameters const struct
aa_profile * to document intent and let the compiler enforce it.

The converted functions are the permission "compute" path plus a few
pure readers:
  - aa_apply_modes_to_perms(), aa_profile_match_label()
  - AUDIT_MODE()
  - aa_label_match() and its match_component()/label_compound_match()/
    label_components_match() helpers (label.c)
  - match_component()/label_compound_match()/label_components_match()/
    label_match()/change_profile_perms()/aa_xattrs_match() (domain.c)
  - match_iface()/match_addr_iface()/match_addr_iface_label()/
    skb_match_to_sk()/skb_match_to_cmd() (af_inet.c)
  - aa_profile_capget(), path_flags(), profile_query_cb()

The remaining aa_profile * parameters cannot be made const: the audit
path stores &profile->label into the owned, refcounted
apparmor_audit_data.subj_label/peer fields, and the domain/lifecycle
paths take references on the profile's embedded label
(aa_get_label()/aa_get_newest_label()/aa_get_profile()) or write
profile fields.

No functional change.

Signed-off-by: John Johansen 
Assisted-by: Claude:claude-opus-4.8

apparmor: constify aa_perms parameters that are read-only

2026-08-11T05:49:42+00:00

Several functions take a struct aa_perms * argument that is only ever
read from and never modified through the pointer. Mark those parameters
const struct aa_perms * to document intent and let the compiler enforce
that the permission set is not mutated.

The converted functions are:
  - aa_check_perms()
  - aa_do_perms()
  - do_perms() (af_inet)
  - match_label() (af_unix)
  - verify_perm()
  - aa_perms_accum() / aa_perms_accum_raw() (@addend only)

No functional change.

Signed-off-by: John Johansen 
Assisted-by: Claude:claude-opus-4.8

apparmor: refactory mount to use check_perms

2026-08-11T05:49:42+00:00

Move the mount permissions check to use the common backend
aa_check_perms() to check permissions. This will make it so caching,
audit, complain, logic can be handled consistently in a single place.

Signed-off-by: John Johansen