Skip to contentSkip to conversation

vsock: use sock_error() to consume sk_err after a failed connect

PATCH In review [email protected] 2 participants 4 messages Aug 4 – Aug 6

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH v4 1/1 vsock: use sock_error() to consume sk_err after a failed connect

From: Nguyen Dinh Phi <[email protected]>

Syzbot reported an issue which can be reproduced with these steps:

r0 = socket(AF_VSOCK, SOCK_STREAM, 0)
bind(r0, {VMADDR_CID_ANY, PORT})
connect(r0, {VMADDR_CID_LOCAL, PORT}) -> -1, EPROTO (self-connect)
listen(r0, backlog) -> 0
r1 = socket(AF_VSOCK, SOCK_STREAM, 0)
connect(r1, {VMADDR_CID_LOCAL, PORT}) -> 0
accept(r0) -> -1, EPROTO (stale sk_err)

Basically, it creates a socket (r0) and triggers a self-connect after binding it. This self-connect fails with EPROTO because it loops back to r0 while the socket is still in the TCP_SYN_SENT state, causing it to be incorrectly dispatched to the connecting-client path. The unexpected packet type encountered there sets sk_err to EPROTO.

After that, it invokes a listen() call on the same socket. This listen() call succeeds because the kernel's listening path never inspects or clears sk_err. Then, a new socket (r1) is created as a normal client and connects to r0. However, vsock_accept() rejects this incoming connection because the listener's sk_err still holds the EPROTO error from the earlier failed self-connect.

This rejection causes the child socket created for r1's connection to never be freed on virtio or hyperv transports; only the VMCI transport implements pending_work to revisit and clean up a rejected socket.

Fix the issue in blocking connect() by using sock_error() to read the sk_err to prevent the rejection branch from occurring in this scenario.

sock_error() atomically reads and clears sk_err, ensuring the error is consumed when vsock_connect() returns and cannot affect subsequent operations on the same socket. This matches the established pattern used by other protocol connect() implementations in the network stack like __inet_stream_connect(), tipc_wait_for_connect()...

For non-blocking connection, vsock_connect_timeout() may set sk->sk_err after vsock_connect() has returned. To handle it, we also remove the sk_err checks from vsock_accept(). Nothing in vsock sets sk_err on a listening socket, so accept() has no reason to inspect it at all.

---
V2: Add reproducer steps to commit message.
V3: Fix truncated title and add annotations to reproducer steps.
V4: Remove sk_err checks from vsock_accept()

 net/vmw_vsock/af_vsock.c | 13 ++++---------
 1 file changed, 4 insertions(+), 9 deletions(-)
net/vmw_vsock/af_vsock.c +4 -9
index 622dbd046799..594fe27d2ebe 100644
@@ -1847,12 +1847,10 @@ static int vsock_connect(struct socket *sock, struct sockaddr_unsized *addr,
 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 	}
 
-	if (sk->sk_err) {
-		err = -sk->sk_err;
+	err = sock_error(sk);
+	if (err) {
 		sk->sk_state = TCP_CLOSE;
 		sock->state = SS_UNCONNECTED;
-	} else {
-		err = 0;
 	}
 
 out_wait:
@@ -1893,7 +1891,7 @@ static int vsock_accept(struct socket *sock, struct socket *newsock,
 	timeout = sock_rcvtimeo(listener, arg->flags & O_NONBLOCK);
 
 	while ((connected = vsock_dequeue_accept(listener)) == NULL &&
-	       listener->sk_err == 0 && timeout != 0) {
+	       timeout != 0) {
 		prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
 		release_sock(listener);
 		timeout = schedule_timeout(timeout);
@@ -1906,11 +1904,8 @@ static int vsock_accept(struct socket *sock, struct socket *newsock,
 		}
 	}
 
-	if (listener->sk_err) {
-		err = -listener->sk_err;
-	} else if (!connected) {
+	if (!connected)
 		err = -EAGAIN;
-	}
 
 	if (connected) {
 		sk_acceptq_removed(listener);
Signature · 1 line
2.43.0
#1 lore raw
Quoted 88 lines · On Tue, Aug 04, 2026 at 09:52:36PM +0800, [email protected] wrote:

From: Nguyen Dinh Phi <[email protected]>

Syzbot reported an issue which can be reproduced with these steps:

r0 = socket(AF_VSOCK, SOCK_STREAM, 0)
bind(r0, {VMADDR_CID_ANY, PORT})
connect(r0, {VMADDR_CID_LOCAL, PORT}) -> -1, EPROTO (self-connect)
listen(r0, backlog) -> 0
r1 = socket(AF_VSOCK, SOCK_STREAM, 0)
connect(r1, {VMADDR_CID_LOCAL, PORT}) -> 0
accept(r0) -> -1, EPROTO (stale sk_err)

Basically, it creates a socket (r0) and triggers a self-connect after binding it. This self-connect fails with EPROTO because it loops back to r0 while the socket is still in the TCP_SYN_SENT state, causing it to be incorrectly dispatched to the connecting-client path. The unexpected packet type encountered there sets sk_err to EPROTO.

After that, it invokes a listen() call on the same socket. This listen() call succeeds because the kernel's listening path never inspects or clears sk_err. Then, a new socket (r1) is created as a normal client and connects to r0. However, vsock_accept() rejects this incoming connection because the listener's sk_err still holds the EPROTO error from the earlier failed self-connect.

This rejection causes the child socket created for r1's connection to never be freed on virtio or hyperv transports; only the VMCI transport implements pending_work to revisit and clean up a rejected socket.

Fix the issue in blocking connect() by using sock_error() to read the sk_err to prevent the rejection branch from occurring in this scenario.

sock_error() atomically reads and clears sk_err, ensuring the error is consumed when vsock_connect() returns and cannot affect subsequent operations on the same socket. This matches the established pattern used by other protocol connect() implementations in the network stack like __inet_stream_connect(), tipc_wait_for_connect()...

For non-blocking connection, vsock_connect_timeout() may set sk->sk_err after vsock_connect() has returned. To handle it, we also remove the sk_err checks from vsock_accept(). Nothing in vsock sets sk_err on a listening socket, so accept() has no reason to inspect it at all.

---
V2: Add reproducer steps to commit message.
V3: Fix truncated title and add annotations to reproducer steps.
V4: Remove sk_err checks from vsock_accept()

net/vmw_vsock/af_vsock.c | 13 ++++---------

1 file changed, 4 insertions(+), 9 deletions(-)
net/vmw_vsock/af_vsock.c +2 -4
index 622dbd046799..594fe27d2ebe 100644
@@ -1847,12 +1847,10 @@ static int vsock_connect(struct socket *sock, struct sockaddr_unsized *addr,
 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 	}
 
-	if (sk->sk_err) {
-		err = -sk->sk_err;
+	err = sock_error(sk);
+	if (err) {
 		sk->sk_state = TCP_CLOSE;
 		sock->state = SS_UNCONNECTED;
-	} else {
-		err = 0;
 	}
 

out_wait:

net/vmw_vsock/af_vsock.c +1 -1
@@ -1893,7 +1891,7 @@ static int vsock_accept(struct socket *sock, struct socket *newsock,
 	timeout = sock_rcvtimeo(listener, arg->flags & O_NONBLOCK);
 
 	while ((connected = vsock_dequeue_accept(listener)) == NULL &&
-	       listener->sk_err == 0 && timeout != 0) {
+	       timeout != 0) {
 		prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE);
 		release_sock(listener);
 		timeout = schedule_timeout(timeout);
@@ -1906,11 +1904,8 @@ static int vsock_accept(struct socket *sock, struct socket *newsock,
 		}

net/vmw_vsock/af_vsock.c static int vsock_accept(struct socket *sock, struct socket *newsock, · line 1910

Quoted 10 lines
net/vmw_vsock/af_vsock.c +1 -4
 	}
 
-	if (listener->sk_err) {
-		err = -listener->sk_err;
-	} else if (!connected) {
+	if (!connected)
 		err = -EAGAIN;
-	}
 
 	if (connected) {

Can this become an } else { ?

Or just add a goto out when setting err = -EAGAIN.

net/vmw_vsock/af_vsock.c static int vsock_accept(struct socket *sock, struct socket *newsock, · line 1911

Quoted 1 line
net/vmw_vsock/af_vsock.c +0 -0
 		sk_acceptq_removed(listener);
		lock_sock_nested(connected, SINGLE_DEPTH_NESTING);
		vconnected = vsock_sk(connected);

		/* If the listener socket has received an error, then we should
		 * reject this socket and return.  Note that we simply mark the
		 * socket rejected, drop our reference, and let the cleanup
		 * function handle the cleanup; the fact that we found it in
		 * the listener's accept queue guarantees that the cleanup
		 * function hasn't run yet.
		 */
		if (err) {
			vconnected->rejected = true;
		} else {

Should we update this comment too and maybe remove the if (err) at all. With that change I guess rejected is never set at the end and maybe we can remove it at all from struct vsock_sock.

Looking at commit d021c344051a ("VSOCK: Introduce VM Sockets") where rejected was introduced, I can't see any path where sk_err is set on a listener socket, so I guess that path was dead since the beginning.

So now I'm thinking if it's better to split in 2 patches (both with the same Fixes tag):
- Patch 1: "vsock: remove stale sk_err checks from vsock_accept()" Where we can also remove rejected since it's never set to true since the beginning
- Patch 2: "vsock: use sock_error() to consume sk_err after a failed connect"

WDYT?

Thanks,
Stefano

#2 lore raw
Quoted 101 lines · Stefano Garzarella
Quoted 100 lines · On Tue, Aug 04, 2026 at 09:52:36PM +0800, [email protected] wrote:

From: Nguyen Dinh Phi <[email protected]>

Syzbot reported an issue which can be reproduced with these steps:

  r0 = socket(AF_VSOCK, SOCK_STREAM, 0)
  bind(r0, {VMADDR_CID_ANY, PORT})
  connect(r0, {VMADDR_CID_LOCAL, PORT})   -> -1, EPROTO  (self-connect) listen(r0, backlog)                     -> 0
  r1 = socket(AF_VSOCK, SOCK_STREAM, 0)
  connect(r1, {VMADDR_CID_LOCAL, PORT})   -> 0
  accept(r0)                              -> -1, EPROTO  (stale sk_err)

Basically, it creates a socket (r0) and triggers a self-connect after binding it. This self-connect fails with EPROTO because it loops back to r0 while the socket is still in the TCP_SYN_SENT state, causing it to be incorrectly dispatched to the connecting-client path. The unexpected packet type encountered there sets sk_err to EPROTO.

After that, it invokes a listen() call on the same socket. This listen() call succeeds because the kernel's listening path never inspects or clears sk_err. Then, a new socket (r1) is created as a normal client and connects to r0. However, vsock_accept() rejects this incoming connection because the listener's sk_err still holds the EPROTO error from the earlier failed self-connect.

This rejection causes the child socket created for r1's connection to never be freed on virtio or hyperv transports; only the VMCI transport implements pending_work to revisit and clean up a rejected socket.

Fix the issue in blocking connect() by using sock_error() to read the sk_err to prevent the rejection branch from occurring in this scenario.

sock_error() atomically reads and clears sk_err, ensuring the error is consumed when vsock_connect() returns and cannot affect subsequent operations on the same socket. This matches the established pattern used by other protocol connect() implementations in the network stack like __inet_stream_connect(), tipc_wait_for_connect()...

For non-blocking connection, vsock_connect_timeout() may set sk->sk_err after vsock_connect() has returned. To handle it, we also remove the sk_err checks from vsock_accept(). Nothing in vsock sets sk_err on a listening socket, so accept() has no reason to inspect it at all.

---
V2: Add reproducer steps to commit message.
V3: Fix truncated title and add annotations to reproducer steps.
V4: Remove sk_err checks from vsock_accept()

net/vmw_vsock/af_vsock.c | 13 ++++---------

1 file changed, 4 insertions(+), 9 deletions(-)
net/vmw_vsock/af_vsock.c +2 -4
index 622dbd046799..594fe27d2ebe 100644
@@ -1847,12 +1847,10 @@ static int vsock_connect(struct socket *sock, 
 struct sockaddr_unsized *addr,
         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
     }
 
-    if (sk->sk_err) {
-        err = -sk->sk_err;
+    err = sock_error(sk);
+    if (err) {
         sk->sk_state = TCP_CLOSE;
         sock->state = SS_UNCONNECTED;
-    } else {
-        err = 0;
     }
 

out_wait:

net/vmw_vsock/af_vsock.c +0 -0
@@ -1893,7 +1891,7 @@ static int vsock_accept(struct socket *sock, 
 struct socket *newsock,
     timeout = sock_rcvtimeo(listener, arg->flags & O_NONBLOCK);
 

    while ((connected = vsock_dequeue_accept(listener)) == NULL && -           listener->sk_err == 0 && timeout != 0) {
+           timeout != 0) { prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE); release_sock(listener);
        timeout = schedule_timeout(timeout);

net/vmw_vsock/af_vsock.c +1 -4
@@ -1906,11 +1904,8 @@ static int vsock_accept(struct socket *sock, 
 struct socket *newsock,
         }
     }
 
-    if (listener->sk_err) {
-        err = -listener->sk_err;
-    } else if (!connected) {
+    if (!connected)
         err = -EAGAIN;
-    }
 
Quoted 6 lines

net/vmw_vsock/af_vsock.c static int vsock_accept(struct socket *sock, struct socket *newsock, · line 1910

Quoted 1 line
net/vmw_vsock/af_vsock.c +0 -0
     if (connected) {

Can this become an } else { ?

Or just add a goto out when setting err = -EAGAIN.

I will update it.

Quoted 17 lines
Quoted 1 line

        sk_acceptq_removed(listener);

        lock_sock_nested(connected, SINGLE_DEPTH_NESTING); vconnected = vsock_sk(connected);

        /* If the listener socket has received an error, then we should
         * reject this socket and return.  Note that we simply mark the
         * socket rejected, drop our reference, and let the cleanup
         * function handle the cleanup; the fact that we found it in
         * the listener's accept queue guarantees that the cleanup
         * function hasn't run yet.
         */
        if (err) {
            vconnected->rejected = true;
        } else {

Quoted 8 lines

Should we update this comment too and maybe remove the if (err) at all. With that change I guess rejected is never set at the end and maybe we can remove it at all from struct vsock_sock.

Looking at commit d021c344051a ("VSOCK: Introduce VM Sockets") where rejected was introduced, I can't see any path where sk_err is set on a listener socket, so I guess that path was dead since the beginning.

That seems true, let me verify it.

Quoted 10 lines

So now I'm thinking if it's better to split in 2 patches (both with the same Fixes tag):
- Patch 1: "vsock: remove stale sk_err checks from vsock_accept()" Where we can also remove rejected since it's never set to true since the beginning
- Patch 2: "vsock: use sock_error() to consume sk_err after a failed connect"

WDYT?

Yes, I felt the same when I was writing the commit message, but honestly I didn't know that I could split it into a series when sending the new version.

So, it may contain 3 patches, if the rejected flag can be removed from vsock_sock.

Thanks,
Phi

#3 lore raw
Quoted 147 lines · Nguyen Dinh Phi [SG]
Quoted 101 lines · Stefano Garzarella
Quoted 100 lines · On Tue, Aug 04, 2026 at 09:52:36PM +0800, [email protected] wrote:

From: Nguyen Dinh Phi <[email protected]>

Syzbot reported an issue which can be reproduced with these steps:

  r0 = socket(AF_VSOCK, SOCK_STREAM, 0)
  bind(r0, {VMADDR_CID_ANY, PORT})
  connect(r0, {VMADDR_CID_LOCAL, PORT})   -> -1, EPROTO  (self-connect) listen(r0, backlog)                     -> 0
  r1 = socket(AF_VSOCK, SOCK_STREAM, 0)
  connect(r1, {VMADDR_CID_LOCAL, PORT})   -> 0
  accept(r0)                              -> -1, EPROTO  (stale sk_err)

Basically, it creates a socket (r0) and triggers a self-connect after binding it. This self-connect fails with EPROTO because it loops back to r0 while the socket is still in the TCP_SYN_SENT state, causing it to be incorrectly dispatched to the connecting-client path. The unexpected packet type encountered there sets sk_err to EPROTO.

After that, it invokes a listen() call on the same socket. This listen() call succeeds because the kernel's listening path never inspects or clears sk_err. Then, a new socket (r1) is created as a normal client and connects to r0. However, vsock_accept() rejects this incoming connection because the listener's sk_err still holds the EPROTO error from the earlier failed self-connect.

This rejection causes the child socket created for r1's connection to never be freed on virtio or hyperv transports; only the VMCI transport implements pending_work to revisit and clean up a rejected socket.

Fix the issue in blocking connect() by using sock_error() to read the sk_err to prevent the rejection branch from occurring in this scenario.

sock_error() atomically reads and clears sk_err, ensuring the error is consumed when vsock_connect() returns and cannot affect subsequent operations on the same socket. This matches the established pattern used by other protocol connect() implementations in the network stack like __inet_stream_connect(), tipc_wait_for_connect()...

For non-blocking connection, vsock_connect_timeout() may set sk->sk_err after vsock_connect() has returned. To handle it, we also remove the sk_err checks from vsock_accept(). Nothing in vsock sets sk_err on a listening socket, so accept() has no reason to inspect it at all.

---
V2: Add reproducer steps to commit message.
V3: Fix truncated title and add annotations to reproducer steps.
V4: Remove sk_err checks from vsock_accept()

net/vmw_vsock/af_vsock.c | 13 ++++---------

1 file changed, 4 insertions(+), 9 deletions(-)
net/vmw_vsock/af_vsock.c +2 -4
index 622dbd046799..594fe27d2ebe 100644
@@ -1847,12 +1847,10 @@ static int vsock_connect(struct socket *sock, 
 struct sockaddr_unsized *addr,
         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
     }
 
-    if (sk->sk_err) {
-        err = -sk->sk_err;
+    err = sock_error(sk);
+    if (err) {
         sk->sk_state = TCP_CLOSE;
         sock->state = SS_UNCONNECTED;
-    } else {
-        err = 0;
     }
 

out_wait:

net/vmw_vsock/af_vsock.c +0 -0
@@ -1893,7 +1891,7 @@ static int vsock_accept(struct socket *sock, 
 struct socket *newsock,
     timeout = sock_rcvtimeo(listener, arg->flags & O_NONBLOCK);
 

    while ((connected = vsock_dequeue_accept(listener)) == NULL && -           listener->sk_err == 0 && timeout != 0) {
+           timeout != 0) { prepare_to_wait(sk_sleep(listener), &wait, TASK_INTERRUPTIBLE); release_sock(listener);
        timeout = schedule_timeout(timeout);

net/vmw_vsock/af_vsock.c +1 -4
@@ -1906,11 +1904,8 @@ static int vsock_accept(struct socket 
 *sock, struct socket *newsock,
         }
     }
 
-    if (listener->sk_err) {
-        err = -listener->sk_err;
-    } else if (!connected) {
+    if (!connected)
         err = -EAGAIN;
-    }
 
Quoted 6 lines

net/vmw_vsock/af_vsock.c static int vsock_accept(struct socket *sock, struct socket *newsock, · line 1910

Quoted 1 line
net/vmw_vsock/af_vsock.c +0 -0
     if (connected) {

Can this become an } else { ?

Or just add a goto out when setting err = -EAGAIN.

I will update it.

Quoted 17 lines
Quoted 1 line

        sk_acceptq_removed(listener);

        lock_sock_nested(connected, SINGLE_DEPTH_NESTING); vconnected = vsock_sk(connected);

        /* If the listener socket has received an error, then we should
         * reject this socket and return.  Note that we simply mark the
         * socket rejected, drop our reference, and let the cleanup
         * function handle the cleanup; the fact that we found it in
         * the listener's accept queue guarantees that the cleanup
         * function hasn't run yet.
         */
        if (err) {
            vconnected->rejected = true;
        } else {

Quoted 8 lines

Should we update this comment too and maybe remove the if (err) at all. With that change I guess rejected is never set at the end and maybe we can remove it at all from struct vsock_sock.

Looking at commit d021c344051a ("VSOCK: Introduce VM Sockets") where rejected was introduced, I can't see any path where sk_err is set on a listener socket, so I guess that path was dead since the beginning.

That seems true, let me verify it.

Quoted 8 lines

So now I'm thinking if it's better to split in 2 patches (both with the same Fixes tag):
- Patch 1: "vsock: remove stale sk_err checks from vsock_accept()" Where we can also remove rejected since it's never set to true since the beginning
- Patch 2: "vsock: use sock_error() to consume sk_err after a failed connect"

Quoted 7 lines
Quoted 2 lines

WDYT?

Yes, I felt the same when I was writing the commit message, but honestly I didn't know that I could split it into a series when sending the new version.

So, it may contain 3 patches, if the rejected flag can be removed from vsock_sock.

Yeah, or including the rejected flag removal in the patch 1 I suggested.

Up to you.

Thanks,
Stefano