Skip to contentSkip to conversation

af_unix: fix holding spinlock in oob handling

PATCH In review Rao Shoaib 3 participants 8 messages Aug 11, 2021 – Aug 13, 2021

Subsystems (3)
lore
Share & tools
lore

Rao Shoaib

author
#0 lore raw

PATCH 1/1 af_unix: fix holding spinlock in oob handling

From: Rao Shoaib <[email protected]>

syzkaller found that OOB code was holding spinlock while calling a function in which it could sleep.

Reported-by [email protected]
Fixes 314001f0bf92 ("af_unix: Add OOB support")
---
 net/unix/af_unix.c | 36 ++++++++++++++++++++++++------------
 1 file changed, 24 insertions(+), 12 deletions(-)
net/unix/af_unix.c +24 -12
index 00d8b08cdbe1..a626e52c629a 100644
@@ -2362,19 +2362,37 @@ static int unix_stream_recv_urg(struct unix_stream_read_state *state)
 	struct sock *sk = sock->sk;
 	struct unix_sock *u = unix_sk(sk);
 	int chunk = 1;
+	struct sk_buff *oob_skb;
 
-	if (sock_flag(sk, SOCK_URGINLINE) || !u->oob_skb)
+	mutex_lock(&u->iolock);
+	unix_state_lock(sk);
+
+	if (sock_flag(sk, SOCK_URGINLINE) || !u->oob_skb) {
+		unix_state_unlock(sk);
+		mutex_unlock(&u->iolock);
 		return -EINVAL;
+	}
 
-	chunk = state->recv_actor(u->oob_skb, 0, chunk, state);
-	if (chunk < 0)
-		return -EFAULT;
+	oob_skb = u->oob_skb;
 
 	if (!(state->flags & MSG_PEEK)) {
-		UNIXCB(u->oob_skb).consumed += 1;
-		kfree_skb(u->oob_skb);
 		u->oob_skb = NULL;
 	}
+
+	unix_state_unlock(sk);
+
+	chunk = state->recv_actor(oob_skb, 0, chunk, state);
+
+	if (!(state->flags & MSG_PEEK)) {
+		UNIXCB(oob_skb).consumed += 1;
+		kfree_skb(oob_skb);
+	}
+
+	mutex_unlock(&u->iolock);
+
+	if (chunk < 0)
+		return -EFAULT;
+
 	state->msg->msg_flags |= MSG_OOB;
 	return 1;
 }
@@ -2434,13 +2452,7 @@ static int unix_stream_read_generic(struct unix_stream_read_state *state,
 	if (unlikely(flags & MSG_OOB)) {
 		err = -EOPNOTSUPP;
 #if IS_ENABLED(CONFIG_AF_UNIX_OOB)
-		mutex_lock(&u->iolock);
-		unix_state_lock(sk);
-
 		err = unix_stream_recv_urg(state);
-
-		unix_state_unlock(sk);
-		mutex_unlock(&u->iolock);
 #endif
 		goto out;
 	}
Signature · 1 line
2.27.0
#1 lore raw
Quoted 6 lines · Rao Shoaib

From: Rao Shoaib <[email protected]>

syzkaller found that OOB code was holding spinlock while calling a function in which it could sleep.

Quoted 6 lines
Reported-by [email protected]
Fixes 314001f0bf92 ("af_unix: Add OOB support")

---

Please do not add these empty lines.

Fixes ...
Reported-by ...
Signed-off-by ...

Also you might take a look at queue_oob()

1) Setting skb->len tp 1 should not be needed, skb_put() already does that
2) After unix_state_lock(other); we probably need to check status of the other socket.
3) Some skb_free() calls should have been consume_skb()

net/unix/af_unix.c +0 -0
index ec02e70a549b42f6c102253508c48426a13f7bc4..0c27e2976f9d234ca3bb131731375bc51a056846

100644

net/unix/af_unix.c +0 -0
@@ -1908,7 +1908,6 @@ static int queue_oob(struct socket *sock, struct

msghdr *msg, struct sock *other

                return err;

        skb_put(skb, 1);
-       skb->len = 1;
        err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, 1);

        if (err) {
@@ -1917,11 +1916,17 @@ static int queue_oob(struct socket *sock,
struct msghdr *msg, struct sock *other
        }

        unix_state_lock(other);
+       if (sock_flag(other, SOCK_DEAD) ||
+           (other->sk_shutdown & RCV_SHUTDOWN)) {
+               unix_state_unlock(other);
+               kfree_skb(skb);
+               return -EPIPE;
+       }
        maybe_add_creds(skb, sock, other);
        skb_get(skb);

        if (ousk->oob_skb)

- kfree_skb(ousk->oob_skb);
+ consume_skb(ousk->oob_skb);

        ousk->oob_skb = skb;
author
#2 lore raw

Hi Eric,

Thanks for your review I will take care of the comments.

Shoaib

Quoted 57 lines · Eric Dumazet
Quoted 11 lines · Rao Shoaib

From: Rao Shoaib <[email protected]>

syzkaller found that OOB code was holding spinlock while calling a function in which it could sleep.

Reported-by [email protected]
Fixes 314001f0bf92 ("af_unix: Add OOB support")

---

Please do not add these empty lines.

Fixes: ...
Reported-by: ...
Signed-off-by: ...
Also you might take a look at queue_oob()

1) Setting skb->len tp 1 should not be needed, skb_put() already does that
2) After unix_state_lock(other); we probably need to check status of the other socket.
3) Some skb_free() calls should have been consume_skb()

net/unix/af_unix.c +0 -0
index ec02e70a549b42f6c102253508c48426a13f7bc4..0c27e2976f9d234ca3bb131731375bc51a056846

100644

net/unix/af_unix.c +0 -0
@@ -1908,7 +1908,6 @@ static int queue_oob(struct socket *sock, struct

msghdr *msg, struct sock *other

                 return err;

         skb_put(skb, 1);
net/unix/af_unix.c +0 -1
-       skb->len = 1;
         err = skb_copy_datagram_from_iter(skb, 0, &msg->msg_iter, 1);
 
         if (err) {
@@ -1917,11 +1916,17 @@ static int queue_oob(struct socket *sock,

struct msghdr *msg, struct sock *other

         }

         unix_state_lock(other);
net/unix/af_unix.c +7 -1
+       if (sock_flag(other, SOCK_DEAD) ||
+           (other->sk_shutdown & RCV_SHUTDOWN)) {
+               unix_state_unlock(other);
+               kfree_skb(skb);
+               return -EPIPE;
+       }
         maybe_add_creds(skb, sock, other);
         skb_get(skb);
 
         if (ousk->oob_skb)
-               kfree_skb(ousk->oob_skb);
+               consume_skb(ousk->oob_skb);
 
         ousk->oob_skb = skb;
author
#3 lore raw
In reply to Eric Dumazet

Review

Quoted 3 lines · Eric Dumazet
         if (ousk->oob_skb)
(quoted diff) +1 -1
-               kfree_skb(ousk->oob_skb);
+               consume_skb(ousk->oob_skb);

Should I be using consume_skb(), as the skb is not being consumed, the ref count is decremented and if zero skb will be freed.

Shoaib

#4 lore raw
Quoted 10 lines · Shoaib Rao

Review

Quoted 3 lines · Eric Dumazet
         if (ousk->oob_skb)
(quoted diff) +1 -1
-               kfree_skb(ousk->oob_skb);
+               consume_skb(ousk->oob_skb);

Should I be using consume_skb(), as the skb is not being consumed, the ref count is decremented and if zero skb will be freed.

consume_skb() and kfree_skb() have the same ref count handling.

The difference is that kfree_skb() is used by convention when a packet is dropped

Admins can look closely at packet drops with drop_monitor, or :

perf record -a -g -e skb:kfree_skb sleep 10
perf report

In your case, the oob_skb is not really dropped. It is replaced by another one, it is part of the normal operation.

author
#5 lore raw
Quoted 14 lines · Eric Dumazet
Quoted 8 lines · Shoaib Rao

Review

Quoted 3 lines · Eric Dumazet
          if (ousk->oob_skb)
(quoted diff) +1 -1
-               kfree_skb(ousk->oob_skb);
+               consume_skb(ousk->oob_skb);

Should I be using consume_skb(), as the skb is not being consumed, the ref count is decremented and if zero skb will be freed.

consume_skb() and kfree_skb() have the same ref count handling.

The difference is that kfree_skb() is used by convention when a packet is dropped

Quoted 7 lines

Admins can look closely at packet drops with drop_monitor, or :

perf record -a -g -e skb:kfree_skb sleep 10
perf report

In your case, the oob_skb is not really dropped. It is replaced by another one, it is part of the normal operation.

Thanks a lot for the explanation. This was very helpful. In my case the skb may be dropped (oob was not read but the read has passed beyond oob, or could become part of normal data). Anyways, I will change it to use consume_skb().

Regards,

Shoaib

maintainer
#6 lore raw
In reply to Rao Shoaib
Quoted 10 lines · On Wed, 11 Aug 2021 15:06:52 -0700 Rao Shoaib wrote:

From: Rao Shoaib <[email protected]>

syzkaller found that OOB code was holding spinlock while calling a function in which it could sleep.

Reported-by [email protected]
Fixes 314001f0bf92 ("af_unix: Add OOB support")

IIUC issues pointed out by Eric are separate so I removed the spacing between the tags and applied, thanks!

author
#7 lore raw

OH, I just sent you a combined patch. I will resend it with just the nits pointed out by Eric.

Shoaib

Quoted 13 lines · Jakub Kicinski
Quoted 10 lines · On Wed, 11 Aug 2021 15:06:52 -0700 Rao Shoaib wrote:

From: Rao Shoaib <[email protected]>

syzkaller found that OOB code was holding spinlock while calling a function in which it could sleep.

Reported-by [email protected]
Fixes 314001f0bf92 ("af_unix: Add OOB support")

IIUC issues pointed out by Eric are separate so I removed the spacing between the tags and applied, thanks!