Re: linux-next: build failure after merge of the apparmor tree

[Kuniyuki Iwashima <kuniyu@amazon.com>]

From: Jakub Kicinski <kuba@kernel.org>
Date: Wed, 26 Mar 2025 04:26:55 -0700
> On Wed, 26 Mar 2025 15:01:48 +1100 Stephen Rothwell wrote:
> > After merging the apparmor tree, today's linux-next build (x86_64
> > allmodconfig) failed like this:
> > 
> > security/apparmor/af_unix.c: In function 'unix_state_double_lock':
> > security/apparmor/af_unix.c:627:17: error: implicit declaration of function 'unix_state_lock'; did you mean 'unix_state_double_lock'? [-Wimplicit-function-declaration]
> >   627 |                 unix_state_lock(sk1);
> >       |                 ^~~~~~~~~~~~~~~
> >       |                 unix_state_double_lock
> > security/apparmor/af_unix.c: In function 'unix_state_double_unlock':
> > security/apparmor/af_unix.c:642:17: error: implicit declaration of function 'unix_state_unlock'; did you mean 'unix_state_double_lock'? [-Wimplicit-function-declaration]
> >   642 |                 unix_state_unlock(sk1);
> >       |                 ^~~~~~~~~~~~~~~~~
> >       |                 unix_state_double_lock
> 
> Thanks Stephen! I'll pop this into the tree in a few hours,
> just giving Kuniyuki a bit more time to ack.

Thanks for catching this, Stephen !

The patch itself looks good, for the patch:

Reviewed-by: Kuniyuki Iwashima <kuniyu@amazon.com>


John:

I had a cursory look at this commit and the exact user of
unix_state_lock() is broken for SOCK_DGRAM.

https://web.git.kernel.org/pub/scm/linux/kernel/git/jj/linux-apparmor.git/commit/?h=apparmor-next&id=c05e705812d179f4b85aeacc34a555a42bc4f9ac

---8<---
+
+	/* TODO: update sock label with new task label */
+	unix_state_lock(sock->sk);
+	peer_sk = unix_peer(sock->sk);
+	if (peer_sk)
+		sock_hold(peer_sk);
+
+	is_sk_fs = is_unix_fs(sock->sk);
+	if (is_sk_fs && peer_sk)
+		sk_req = request;
+	if (sk_req)
+		error = unix_label_sock_perm(subj_cred, label, op, sk_req,
+					     sock);
+	unix_state_unlock(sock->sk);
+	if (!peer_sk)
+		return error;
+
+	unix_state_double_lock(sock->sk, peer_sk);

Here, unix_peer(sock->sk) could have been changed and must be
double checked.  See unix_dgram_sendmsg().

The patch seems to be written in 2022 and recently merged.
I'm not sure if it's reviewed by netdev folks at that time,
but please cc me and netdev next time for patches regarding
AF_UNIX.

Thanks!


+	if (!is_sk_fs && is_unix_fs(peer_sk)) {
+		last_error(error,
+			   unix_fs_perm(op, request, subj_cred, label,
+					unix_sk(peer_sk)));
+	} else if (!is_sk_fs) {
+		struct aa_sk_ctx *pctx = aa_sock(peer_sk);
+
+		last_error(error,
+			xcheck(aa_unix_peer_perm(subj_cred, label, op,
+						 MAY_READ | MAY_WRITE,
+						 sock->sk, peer_sk, NULL),
+			       aa_unix_peer_perm(file->f_cred, pctx->label, op,
+						 MAY_READ | MAY_WRITE,
+						 peer_sk, sock->sk, label)));
+	}
+	unix_state_double_unlock(sock->sk, peer_sk);
---8<---