[PATCH] can: j1939: j1939_sk_bind(): fix j1939_ecu leak when re-bind failed

[PATCH] can: j1939: j1939_sk_bind(): fix j1939_ecu leak when re-bind failed

[Tetsuo Handa]

syzbot is reporting "struct j1939_ecu" refcount leak, which occurs when
netdev_hold() is called during ECU creation but the corresponding
netdev_put() is never executed because the parent "struct j1939_ecu"
object is leaked.

  unregister_netdevice: waiting for vxcan1 to become free. Usage count = 3
  ref_tracker: netdev@ffff8880710f0700 has 1/2 users at
       __netdev_tracker_alloc include/linux/netdevice.h:4496 [inline]
       netdev_hold include/linux/netdevice.h:4525 [inline]
       j1939_ecu_create_locked+0x1c9/0x400 net/can/j1939/bus.c:159
       j1939_local_ecu_get+0xeb/0x220 net/can/j1939/bus.c:293
       j1939_sk_bind+0x70a/0xc60 net/can/j1939/socket.c:529
       __sys_bind_socket net/socket.c:1920 [inline]
       __sys_bind+0x2e3/0x410 net/socket.c:1951
       __do_sys_bind net/socket.c:1956 [inline]
       __se_sys_bind net/socket.c:1954 [inline]
       __x64_sys_bind+0x7a/0x90 net/socket.c:1954
       do_syscall_x64 arch/x86/entry/syscall_64.c:63 [inline]
       do_syscall_64+0x174/0x580 arch/x86/entry/syscall_64.c:94
       entry_SYSCALL_64_after_hwframe+0x77/0x7f

  ref_tracker: netdev@ffff8880710f0700 has 1/2 users at
       __netdev_tracker_alloc include/linux/netdevice.h:4496 [inline]
       netdev_hold include/linux/netdevice.h:4525 [inline]
       j1939_priv_create net/can/j1939/main.c:140 [inline]
       j1939_netdev_start+0x387/0xb20 net/can/j1939/main.c:268
       j1939_sk_bind+0x946/0xc60 net/can/j1939/socket.c:506
       __sys_bind_socket net/socket.c:1920 [inline]
       __sys_bind+0x2e3/0x410 net/socket.c:1951
       __do_sys_bind net/socket.c:1956 [inline]
       __se_sys_bind net/socket.c:1954 [inline]
       __x64_sys_bind+0x7a/0x90 net/socket.c:1954
       do_syscall_x64 arch/x86/entry/syscall_64.c:63 [inline]
       do_syscall_64+0x174/0x580 arch/x86/entry/syscall_64.c:94
       entry_SYSCALL_64_after_hwframe+0x77/0x7f

The root cause lies in the error handling of j1939_sk_bind() during a
re-bind operation (binding an already bound socket to the same interface).
Currently, the function prematurely drops the old ECU references by calling
j1939_local_ecu_put() before verifying whether the new configuration can be
successfully acquired via j1939_local_ecu_get().

If j1939_local_ecu_get() subsequently fails, the function unconditionally
calls j1939_netdev_stop() and clears jsk->priv. This leaves the socket in a
half-broken state where the old ECU's refcount has already been decremented
incompletely, but the socket destruct pathway (j1939_sk_sock_destruct) can
no longer perform proper cleanup because jsk->priv is NULL. As a result,
the old "struct j1939_ecu" remains orphaned on the priv->ecus list,
permanently leaking both the ECU object and the net_device reference held
inside it.

Fix this by deferring the removal and release of the old ECU references
until after j1939_local_ecu_get() has successfully acquired the new
resources. As a side effect of this change, the socket's state no longer
changes when the re-bind operation failed.

Reported-by: syzbot+e2af46126e0644cbebdd@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=e2af46126e0644cbebdd
Analyzed-by: AI Mode in Google Search (no mail address)
Fixes: f214744c8a27 ("can: j1939: j1939_sk_bind(): call j1939_priv_put() immediately when j1939_local_ecu_get() failed")
Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
---
 net/can/j1939/socket.c | 37 ++++++++++++++++++++++---------------
 1 file changed, 22 insertions(+), 15 deletions(-)

diff --git a/net/can/j1939/socket.c b/net/can/j1939/socket.c
index 50a598ef5fd4..24efb25c58c3 100644
--- a/net/can/j1939/socket.c
+++ b/net/can/j1939/socket.c
@@ -450,6 +450,7 @@ static int j1939_sk_bind(struct socket *sock, struct sockaddr_unsized *uaddr, in
 	struct sock *sk;
 	struct net *net;
 	int ret = 0;
+	bool was_bound;
 
 	ret = j1939_sk_sanity_check(addr, len);
 	if (ret)
@@ -462,7 +463,8 @@ static int j1939_sk_bind(struct socket *sock, struct sockaddr_unsized *uaddr, in
 	net = sock_net(sk);
 
 	/* Already bound to an interface? */
-	if (jsk->state & J1939_SOCK_BOUND) {
+	was_bound = (jsk->state & J1939_SOCK_BOUND);
+	if (was_bound) {
 		/* A re-bind() to a different interface is not
 		 * supported.
 		 */
@@ -470,10 +472,6 @@ static int j1939_sk_bind(struct socket *sock, struct sockaddr_unsized *uaddr, in
 			ret = -EINVAL;
 			goto out_release_sock;
 		}
-
-		/* drop old references */
-		j1939_jsk_del(priv, jsk);
-		j1939_local_ecu_put(priv, jsk->addr.src_name, jsk->addr.sa);
 	} else {
 		struct can_ml_priv *can_ml;
 		struct net_device *ndev;
@@ -519,22 +517,31 @@ static int j1939_sk_bind(struct socket *sock, struct sockaddr_unsized *uaddr, in
 		jsk->priv = priv;
 	}
 
+	/* get new references without dropping old references */
+	ret = j1939_local_ecu_get(priv, addr->can_addr.j1939.name, addr->can_addr.j1939.addr);
+	if (ret) {
+		/* nothing to undo if re-bind() failed */
+		if (!was_bound) {
+			j1939_netdev_stop(priv);
+			jsk->priv = NULL;
+			synchronize_rcu();
+			j1939_priv_put(priv);
+		}
+		goto out_release_sock;
+	}
+
+	/* drop old references after re-bind() succeeded */
+	if (was_bound) {
+		j1939_jsk_del(priv, jsk);
+		j1939_local_ecu_put(priv, jsk->addr.src_name, jsk->addr.sa);
+	}
+
 	/* set default transmit pgn */
 	if (j1939_pgn_is_valid(addr->can_addr.j1939.pgn))
 		jsk->pgn_rx_filter = addr->can_addr.j1939.pgn;
 	jsk->addr.src_name = addr->can_addr.j1939.name;
 	jsk->addr.sa = addr->can_addr.j1939.addr;
 
-	/* get new references */
-	ret = j1939_local_ecu_get(priv, jsk->addr.src_name, jsk->addr.sa);
-	if (ret) {
-		j1939_netdev_stop(priv);
-		jsk->priv = NULL;
-		synchronize_rcu();
-		j1939_priv_put(priv);
-		goto out_release_sock;
-	}
-
 	j1939_jsk_add(priv, jsk);
 
  out_release_sock: /* fall through */
-- 
2.55.0

Re: [PATCH] can: j1939: j1939_sk_bind(): fix j1939_ecu leak when re-bind failed

[sashiko-bot]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] Netdevice refcount leak due to a race condition where the unregister notifier misses a socket during bind or re-bind list manipulations.
--

--- Patch [1]: [PATCH] can: j1939: j1939_sk_bind(): fix j1939_ecu leak when re-bind failed ---
commit 472bfe7c306ed38083d747b95430b318ca86c9cf
Author: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>

can: j1939: j1939_sk_bind(): fix j1939_ecu leak when re-bind failed

This commit fixes an ECU and net_device reference leak in j1939_sk_bind().
It defers removing the old ECU references during a re-bind operation until
after the new resources have been successfully acquired.

Closes: https://syzkaller.appspot.com/bug?extid=e2af46126e0644cbebdd

> diff --git a/net/can/j1939/socket.c b/net/can/j1939/socket.c
> index 50a598ef5fd4a..24efb25c58c3f 100644
> --- a/net/can/j1939/socket.c
> +++ b/net/can/j1939/socket.c

[ ... ]

> @@ -519,22 +517,31 @@ static int j1939_sk_bind(struct socket *sock, struct sockaddr_unsized *uaddr, in
>  		jsk->priv = priv;
>  	}
>  
> +	/* get new references without dropping old references */
> +	ret = j1939_local_ecu_get(priv, addr->can_addr.j1939.name, addr->can_addr.j1939.addr);
> +	if (ret) {
> +		/* nothing to undo if re-bind() failed */
> +		if (!was_bound) {
> +			j1939_netdev_stop(priv);
> +			jsk->priv = NULL;
> +			synchronize_rcu();
> +			j1939_priv_put(priv);
> +		}
> +		goto out_release_sock;
> +	}
> +
> +	/* drop old references after re-bind() succeeded */
> +	if (was_bound) {
> +		j1939_jsk_del(priv, jsk);

[Severity: High]
This isn't a bug introduced by this patch, but could this create a race
condition with netdevice unregistration during bind and re-bind?

When a socket takes a reference to priv during an initial bind (when
jsk->priv = priv is set), or during a re-bind when it is temporarily
removed from the list via j1939_jsk_del() right here, the socket is holding a
netdevice reference but is absent from the priv->j1939_socks list.

If j1939_sk_netdev_event_unregister() executes concurrently during this
window, it searches for sockets to clean up by iterating that exact list:

net/can/j1939/socket.c:j1939_sk_netdev_event_unregister() {
    ...
    list_for_each_entry(jsk, &priv->j1939_socks, list) {
        ...
    }
}

Could the unregister notifier miss the socket completely while it is
temporarily unlinked, skipping the mandatory cleanup and resulting in
unregister_netdevice() hanging indefinitely in netdev_wait_allrefs()?

> +		j1939_local_ecu_put(priv, jsk->addr.src_name, jsk->addr.sa);
> +	}
> +
>  	/* set default transmit pgn */
>  	if (j1939_pgn_is_valid(addr->can_addr.j1939.pgn))
>  		jsk->pgn_rx_filter = addr->can_addr.j1939.pgn;
>  	jsk->addr.src_name = addr->can_addr.j1939.name;
>  	jsk->addr.sa = addr->can_addr.j1939.addr;
>  
> -	/* get new references */
> -	ret = j1939_local_ecu_get(priv, jsk->addr.src_name, jsk->addr.sa);
> -	if (ret) {
> -		j1939_netdev_stop(priv);
> -		jsk->priv = NULL;
> -		synchronize_rcu();
> -		j1939_priv_put(priv);
> -		goto out_release_sock;
> -	}
> -
>  	j1939_jsk_add(priv, jsk);
>  
>   out_release_sock: /* fall through */

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/8063b5c1-6171-4350-a7d1-6ccd079fb0ff@I-love.SAKURA.ne.jp?part=1