[Patch bpf-next] unix_bpf: fix a potential deadlock in unix_dgram_bpf_recvmsg()

From: Cong Wang <redacted>

As Eric noticed, __unix_dgram_recvmsg() may acquire u->iolock
too, so we have to release it before calling this function.

Fixes: 9825d866ce0d ("af_unix: Implement unix_dgram_bpf_recvmsg()")
Reported-by: Eric Dumazet <redacted>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jakub Sitnicki <jakub@cloudflare.com>
Cc: Lorenz Bauer <redacted>
Signed-off-by: Cong Wang <redacted>
---
 net/unix/unix_bpf.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/net/unix/unix_bpf.c b/net/unix/unix_bpf.c
index db0cda29fb2f..b07cb30e87b1 100644
--- a/net/unix/unix_bpf.c
+++ b/net/unix/unix_bpf.c

@@ -53,8 +53,9 @@ static int unix_dgram_bpf_recvmsg(struct sock *sk, struct msghdr *msg,
 	mutex_lock(&u->iolock);
 	if (!skb_queue_empty(&sk->sk_receive_queue) &&
 	    sk_psock_queue_empty(psock)) {
-		ret = __unix_dgram_recvmsg(sk, msg, len, flags);
-		goto out;
+		mutex_unlock(&u->iolock);
+		sk_psock_put(sk, psock);
+		return __unix_dgram_recvmsg(sk, msg, len, flags);
 	}
 
 msg_bytes_ready:

@@ -68,13 +69,13 @@ static int unix_dgram_bpf_recvmsg(struct sock *sk, struct msghdr *msg,
 		if (data) {
 			if (!sk_psock_queue_empty(psock))
 				goto msg_bytes_ready;
-			ret = __unix_dgram_recvmsg(sk, msg, len, flags);
-			goto out;
+			mutex_unlock(&u->iolock);
+			sk_psock_put(sk, psock);
+			return __unix_dgram_recvmsg(sk, msg, len, flags);
 		}
 		copied = -EAGAIN;
 	}
 	ret = copied;
-out:
 	mutex_unlock(&u->iolock);
 	sk_psock_put(sk, psock);
 	return ret;

-- 
2.27.0

Cong Wang wrote:

Is there a reason to grab the mutex_lock(u->iolock) above the
skb_queue_emptyaand sk_psock_queue_empty checks?

Could it be move here just above the msg_bytes_ready label?

On Tue, Jul 27, 2021 at 9:12 AM John Fastabend [off-list ref] wrote:

Is there a reason to grab the mutex_lock(u->iolock) above the
skb_queue_emptyaand sk_psock_queue_empty checks?

Could it be move here just above the msg_bytes_ready label?

The check of the receive queue is more accurate with lock.

Thanks.

On Fri, Jul 23, 2021 at 08:36 PM CEST, Cong Wang wrote:

Nit: Can be just `return copied`. `ret` became useless.

Acked-by: Jakub Sitnicki <jakub@cloudflare.com>

Jakub Sitnicki wrote:

On Fri, Jul 23, 2021 at 08:36 PM CEST, Cong Wang wrote:

quoted

From: Cong Wang <redacted>

As Eric noticed, __unix_dgram_recvmsg() may acquire u->iolock
too, so we have to release it before calling this function.

Fixes: 9825d866ce0d ("af_unix: Implement unix_dgram_bpf_recvmsg()")
Reported-by: Eric Dumazet <redacted>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jakub Sitnicki <jakub@cloudflare.com>
Cc: Lorenz Bauer <redacted>
Signed-off-by: Cong Wang <redacted>
---
 net/unix/unix_bpf.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/net/unix/unix_bpf.c b/net/unix/unix_bpf.c
index db0cda29fb2f..b07cb30e87b1 100644
--- a/net/unix/unix_bpf.c
+++ b/net/unix/unix_bpf.c

@@ -53,8 +53,9 @@ static int unix_dgram_bpf_recvmsg(struct sock *sk, struct msghdr *msg,
 	mutex_lock(&u->iolock);
 	if (!skb_queue_empty(&sk->sk_receive_queue) &&
 	    sk_psock_queue_empty(psock)) {
-		ret = __unix_dgram_recvmsg(sk, msg, len, flags);
-		goto out;
+		mutex_unlock(&u->iolock);
+		sk_psock_put(sk, psock);
+		return __unix_dgram_recvmsg(sk, msg, len, flags);
 	}
 
 msg_bytes_ready:

@@ -68,13 +69,13 @@ static int unix_dgram_bpf_recvmsg(struct sock *sk, struct msghdr *msg,
 		if (data) {
 			if (!sk_psock_queue_empty(psock))
 				goto msg_bytes_ready;
-			ret = __unix_dgram_recvmsg(sk, msg, len, flags);
-			goto out;
+			mutex_unlock(&u->iolock);
+			sk_psock_put(sk, psock);
+			return __unix_dgram_recvmsg(sk, msg, len, flags);
 		}
 		copied = -EAGAIN;
 	}
 	ret = copied;
-out:
 	mutex_unlock(&u->iolock);
 	sk_psock_put(sk, psock);
 	return ret;

Nit: Can be just `return copied`. `ret` became useless.

Acked-by: Jakub Sitnicki <jakub@cloudflare.com>

Worth doing the small cleanup pointed out by Jakub but feel free to add
my ack.

Acked-by: John Fastabend <john.fastabend@gmail.com>

On Wed, Jul 28, 2021 at 11:53 AM John Fastabend
[off-list ref] wrote:

Jakub Sitnicki wrote:

quoted

On Fri, Jul 23, 2021 at 08:36 PM CEST, Cong Wang wrote:

quoted

From: Cong Wang <redacted>

As Eric noticed, __unix_dgram_recvmsg() may acquire u->iolock
too, so we have to release it before calling this function.

Fixes: 9825d866ce0d ("af_unix: Implement unix_dgram_bpf_recvmsg()")
Reported-by: Eric Dumazet <redacted>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Cc: Jakub Sitnicki <jakub@cloudflare.com>
Cc: Lorenz Bauer <redacted>
Signed-off-by: Cong Wang <redacted>
---
 net/unix/unix_bpf.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/net/unix/unix_bpf.c b/net/unix/unix_bpf.c
index db0cda29fb2f..b07cb30e87b1 100644
--- a/net/unix/unix_bpf.c
+++ b/net/unix/unix_bpf.c

@@ -53,8 +53,9 @@ static int unix_dgram_bpf_recvmsg(struct sock *sk, struct msghdr *msg,
    mutex_lock(&u->iolock);
    if (!skb_queue_empty(&sk->sk_receive_queue) &&
        sk_psock_queue_empty(psock)) {
-           ret = __unix_dgram_recvmsg(sk, msg, len, flags);
-           goto out;
+           mutex_unlock(&u->iolock);
+           sk_psock_put(sk, psock);
+           return __unix_dgram_recvmsg(sk, msg, len, flags);
    }

 msg_bytes_ready:

@@ -68,13 +69,13 @@ static int unix_dgram_bpf_recvmsg(struct sock *sk, struct msghdr *msg,
            if (data) {
                    if (!sk_psock_queue_empty(psock))
                            goto msg_bytes_ready;
-                   ret = __unix_dgram_recvmsg(sk, msg, len, flags);
-                   goto out;
+                   mutex_unlock(&u->iolock);
+                   sk_psock_put(sk, psock);
+                   return __unix_dgram_recvmsg(sk, msg, len, flags);
            }
            copied = -EAGAIN;
    }
    ret = copied;
-out:
    mutex_unlock(&u->iolock);
    sk_psock_put(sk, psock);
    return ret;

Nit: Can be just `return copied`. `ret` became useless.

Acked-by: Jakub Sitnicki <jakub@cloudflare.com>

Worth doing the small cleanup pointed out by Jakub but feel free to add
my ack.

I cleaned it up while applying. Applied to bpf-next, thanks.

Acked-by: John Fastabend <john.fastabend@gmail.com>