[PATCH] ima: reject duplicate fsname= in a policy rule
From: Weidong Wang via B4 Relay
Date: Thu Nov 12 2026 - 02:03:29 EST
From: Weidong Wang <weidongw@xxxxxxxxxxxx>
ima_parse_rule() stores the fsname= value with kstrdup() but, unlike
fsmagic=, fs_subtype= and the other options, does not check whether the
rule already has one. A rule that names fsname= twice overwrites
entry->fsname and leaks the first string.
Reject the duplicate with -EINVAL, as fs_subtype= does. Such a rule used
to be accepted with the last fsname= taking effect; giving the option
twice was never meaningful, and the other options already reject it.
Writing the policy requires CAP_SYS_ADMIN, so the leak is limited to a
few bytes per malformed rule loaded by a privileged user.
The leak was found by an AI code review agent (an LLM-based bug-finding
tool developed by Zijie Zhao and the submitter) and confirmed on an
unmodified kernel, without error injection, on x86_64 under QEMU with
kmemleak: loading 64 rules with a duplicated fsname= left 64
unreferenced objects allocated by kstrdup() in ima_parse_rule() (inlined
into ima_parse_add_rule()); the same rules with a single fsname= left
none. With this patch the duplicate rules are rejected and kmemleak
reports nothing; rules with a single fsname= are still accepted.
The reproducer, fix and changelog were drafted with AI assistance
(Claude), then reviewed and tested by the submitter.
Fixes: f1b08bbcbdaf ("ima: define a new policy condition based on the filesystem name")
Cc: Zijie Zhao <zijie4@xxxxxxxxxxxx>
Signed-off-by: Weidong Wang <weidongw@xxxxxxxxxxxx>
---
Tested on v7.4-rc1 (05441d697f7f), x86_64, QEMU/KVM, kmemleak, booted
with slab_debug=U. 64 rules written to the IMA policy file at once:
trigger "dont_measure fsname=pocfsA<i> fsname=pocfsB<i>"
control "dont_measure fsname=pocfsA<i>"
trigger control
unpatched accepted; 64 unreferenced accepted; none
objects, all "pocfsA<i>"
patched write rejected; none accepted; none
Without slab_debug=U kmemleak reports none of these leaks on this
kernel: stale pointers in the per-CPU slab sheaves keep the leaked
strings "referenced". Writing the policy more than once needs
CONFIG_IMA_WRITE_POLICY=y, which the test kernel has. The code is
unchanged in mainline, linux-next (next-20261111) and next-integrity.
The reproducer is available on request.
---
security/integrity/ima/ima_policy.c | 5 +++++
1 file changed, 5 insertions(+)
diff --git a/security/integrity/ima/ima_policy.c b/security/integrity/ima/ima_policy.c
index 2dc1d93aa..f08ca00bd 100644
--- a/security/integrity/ima/ima_policy.c
+++ b/security/integrity/ima/ima_policy.c
@@ -1619,6 +1619,11 @@ static int ima_parse_rule(char *rule, struct ima_rule_entry *entry)
case Opt_fsname:
ima_log_string(ab, "fsname", args[0].from);
+ if (entry->fsname) {
+ result = -EINVAL;
+ break;
+ }
+
entry->fsname = kstrdup(args[0].from, GFP_KERNEL);
if (!entry->fsname) {
result = -ENOMEM;
---
base-commit: 05441d697f7f9a927264f4058cbbc6a03dfcac9d
change-id: 20261112-c3-n204-a82e17864b76
Best regards,
--
Weidong Wang <weidongw@xxxxxxxxxxxx>