apparmor: fix use-after-free in rawdata dedup loop

commit 6f060496d0 upstream.

aa_replace_profiles() walks ns->rawdata_list to dedup the incoming
policy blob against entries already attached to existing profiles.
Per the kernel-doc on struct aa_loaddata, list membership does not
hold a reference: profiles hold pcount, and when the last pcount
drops, do_ploaddata_rmfs() is queued on a workqueue that takes
ns->lock and removes the entry. Between dropping the last pcount
and the workqueue running, an entry remains on the list with
pcount == 0.

aa_get_profile_loaddata() is an unconditional kref_get() on
pcount, so when the dedup loop hits such an entry, refcount
hardening reports

  refcount_t: addition on 0; use-after-free.

inside aa_replace_profiles(), and the poisoned counter then
trips "saturated" and "underflow" warnings on the subsequent
uses of the same loaddata.

Before commit a0b7091c4d ("apparmor: fix race on rawdata
dereference") the dedup path used a get_unless_zero-style helper
on a single counter, so the existing "if (tmp)" guard was
meaningful. The split-refcount refactor introduced
aa_get_profile_loaddata(), which has plain kref_get() semantics,
and the guard quietly became a no-op.

Introduce aa_get_profile_loaddata_not0(), matching the existing
_not0 convention used by aa_get_profile_not0(), and use it for
the rawdata_list dedup lookup so dying entries are skipped.

Reproduced on x86_64 with v7.1-rc5 in QEMU+KVM running Ubuntu
24.04 + stress-ng 0.17.06:

  stress-ng --apparmor 1 --klog-check --timeout 60s

Without this patch the three refcount_t warnings fire within a
few seconds. With it the same 60 s run is clean. Coverage is a
smoke-test only; a longer soak with CONFIG_KASAN, CONFIG_KCSAN
and CONFIG_PROVE_LOCKING would be welcome from anyone with the
cycles.

Fixes: a0b7091c4d ("apparmor: fix race on rawdata dereference")
Reported-by: Colin Ian King <colin.i.king@gmail.com>
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=221513
Cc: stable@vger.kernel.org
Signed-off-by: Ruslan Valiyev <linuxoid@gmail.com>
Signed-off-by: John Johansen <john.johansen@canonical.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

security/apparmor/include/policy_unpack.h

@@ -163,6 +163,25 @@ aa_get_profile_loaddata(struct aa_loaddata *data)
 	return data;
 }
 
+/**
+ * aa_get_profile_loaddata_not0 - get a profile reference count if not zero
+ * @data: reference to get a count on
+ *
+ * Like aa_get_profile_loaddata(), but safe to call on an entry that may
+ * be on a list (e.g. ns->rawdata_list) where the last pcount has already
+ * dropped and the deferred cleanup has not yet run.
+ *
+ * Returns: pointer to reference, or %NULL if @data is NULL or its
+ *          profile refcount has already reached zero.
+ */
+static inline struct aa_loaddata *
+aa_get_profile_loaddata_not0(struct aa_loaddata *data)
+{
+	if (data && kref_get_unless_zero(&data->pcount))
+		return data;
+	return NULL;
+}
+
 void __aa_loaddata_update(struct aa_loaddata *data, long revision);
 bool aa_rawdata_eq(struct aa_loaddata *l, struct aa_loaddata *r);
 void aa_loaddata_kref(struct kref *kref);

security/apparmor/policy.c

@@ -1175,8 +1175,12 @@ ssize_t aa_replace_profiles(struct aa_ns *policy_ns, struct aa_label *label,
 			if (aa_rawdata_eq(rawdata_ent, udata)) {
 				struct aa_loaddata *tmp;
 
-				tmp = aa_get_profile_loaddata(rawdata_ent);
-				/* check we didn't fail the race */
+				/*
+				 * Entries remain on rawdata_list with
+				 * pcount == 0 until do_ploaddata_rmfs()
+				 * runs; only take a live profile ref.
+				 */
+				tmp = aa_get_profile_loaddata_not0(rawdata_ent);
 				if (tmp) {
 					aa_put_profile_loaddata(udata);
 					udata = tmp;