The unpack_secmark() function currently uses kfree() to release memory
allocated for secmark structures and their labels. However, if a failure
occurs after partially parsing secmark, sensitive data may remain in
memory, posing a security risk.

To mitigate this, replace kfree() with kfree_sensitive() for freeing
secmark structures and their labels, aligning with the approach used
in free_ruleset().

I am submitting this as an RFC to seek freedback on whether this change
is appropriate and aligns with the subsystem's expectations. If
confirmed to be helpful, I will send a formal patch.

Signed-off-by: Zilin Guan <zilin@seu.edu.cn>
---
 security/apparmor/policy_unpack.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/security/apparmor/policy_unpack.c b/security/apparmor/policy_unpack.c
index 992b74c50..610e09c76 100644
--- a/security/apparmor/policy_unpack.c
+++ b/security/apparmor/policy_unpack.c
@@ -598,8 +598,8 @@ static bool unpack_secmark(struct aa_ext *e, struct aa_ruleset *rules)
 fail:
 	if (rules->secmark) {
 		for (i = 0; i < size; i++)
-			kfree(rules->secmark[i].label);
-		kfree(rules->secmark);
+			kfree_sensitive(rules->secmark[i].label);
+		kfree_sensitive(rules->secmark);
 		rules->secmark_count = 0;
 		rules->secmark = NULL;
 	}
-- 
2.34.1

On 4/17/25 21:52, Zilin Guan wrote:
> The unpack_secmark() function currently uses kfree() to release memory
> allocated for secmark structures and their labels. However, if a failure
> occurs after partially parsing secmark, sensitive data may remain in
> memory, posing a security risk.
> 
> To mitigate this, replace kfree() with kfree_sensitive() for freeing
> secmark structures and their labels, aligning with the approach used
> in free_ruleset().
> 
> I am submitting this as an RFC to seek freedback on whether this change
> is appropriate and aligns with the subsystem's expectations. If
> confirmed to be helpful, I will send a formal patch.
> 
> Signed-off-by: Zilin Guan <zilin@seu.edu.cn>

sorry I am super behind on the backlog, I will get this in to my tree today


Acked-by: John Johansen <john.johansen@canonical.com>

> ---
>   security/apparmor/policy_unpack.c | 4 ++--
>   1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/security/apparmor/policy_unpack.c b/security/apparmor/policy_unpack.c
> index 992b74c50..610e09c76 100644
> --- a/security/apparmor/policy_unpack.c
> +++ b/security/apparmor/policy_unpack.c
> @@ -598,8 +598,8 @@ static bool unpack_secmark(struct aa_ext *e, struct aa_ruleset *rules)
>   fail:
>   	if (rules->secmark) {
>   		for (i = 0; i < size; i++)
> -			kfree(rules->secmark[i].label);
> -		kfree(rules->secmark);
> +			kfree_sensitive(rules->secmark[i].label);
> +		kfree_sensitive(rules->secmark);
>   		rules->secmark_count = 0;
>   		rules->secmark = NULL;
>   	}

On Fri, Apr 18, 2025 at 04:52:50AM+0000, Zilin Guan wrote:
> To mitigate this, replace kfree() with kfree_sensitive() for freeing
> secmark structures and their labels, aligning with the approach used
> in free_ruleset().

To clarify, we propose using kfree_sensitive() for secmark structures and 
their labels because they are already freed with kfree_sensitive() in 
another error path, specifically in aa_free_profile() -> free_ruleset().

This change aligns both cleanup pathways, whether aborting early in 
unpack_secmark() or cleaning up later via aa_free_profile() -> 
free_ruleset(). It ensures that all secmark-related allocations are 
scrubbed before deallocation, mitigating any potential security risks.

Best Regards,
Zilin Guan