net/af_unix: fix a data-race in unix_dgram_sendmsg / unix_release_sock
authorEric Dumazet <edumazet@google.com>
Wed, 16 Jun 2021 14:47:15 +0000 (07:47 -0700)
committerDavid S. Miller <davem@davemloft.net>
Wed, 16 Jun 2021 19:51:55 +0000 (12:51 -0700)
While unix_may_send(sk, osk) is called while osk is locked, it appears
unix_release_sock() can overwrite unix_peer() after this lock has been
released, making KCSAN unhappy.

Changing unix_release_sock() to access/change unix_peer()
before lock is released should fix this issue.

BUG: KCSAN: data-race in unix_dgram_sendmsg / unix_release_sock

write to 0xffff88810465a338 of 8 bytes by task 20852 on cpu 1:
 unix_release_sock+0x4ed/0x6e0 net/unix/af_unix.c:558
 unix_release+0x2f/0x50 net/unix/af_unix.c:859
 __sock_release net/socket.c:599 [inline]
 sock_close+0x6c/0x150 net/socket.c:1258
 __fput+0x25b/0x4e0 fs/file_table.c:280
 ____fput+0x11/0x20 fs/file_table.c:313
 task_work_run+0xae/0x130 kernel/task_work.c:164
 tracehook_notify_resume include/linux/tracehook.h:189 [inline]
 exit_to_user_mode_loop kernel/entry/common.c:175 [inline]
 exit_to_user_mode_prepare+0x156/0x190 kernel/entry/common.c:209
 __syscall_exit_to_user_mode_work kernel/entry/common.c:291 [inline]
 syscall_exit_to_user_mode+0x20/0x40 kernel/entry/common.c:302
 do_syscall_64+0x56/0x90 arch/x86/entry/common.c:57
 entry_SYSCALL_64_after_hwframe+0x44/0xae

read to 0xffff88810465a338 of 8 bytes by task 20888 on cpu 0:
 unix_may_send net/unix/af_unix.c:189 [inline]
 unix_dgram_sendmsg+0x923/0x1610 net/unix/af_unix.c:1712
 sock_sendmsg_nosec net/socket.c:654 [inline]
 sock_sendmsg net/socket.c:674 [inline]
 ____sys_sendmsg+0x360/0x4d0 net/socket.c:2350
 ___sys_sendmsg net/socket.c:2404 [inline]
 __sys_sendmmsg+0x315/0x4b0 net/socket.c:2490
 __do_sys_sendmmsg net/socket.c:2519 [inline]
 __se_sys_sendmmsg net/socket.c:2516 [inline]
 __x64_sys_sendmmsg+0x53/0x60 net/socket.c:2516
 do_syscall_64+0x4a/0x90 arch/x86/entry/common.c:47
 entry_SYSCALL_64_after_hwframe+0x44/0xae

value changed: 0xffff888167905400 -> 0x0000000000000000

Reported by Kernel Concurrency Sanitizer on:
CPU: 0 PID: 20888 Comm: syz-executor.0 Not tainted 5.13.0-rc5-syzkaller #0
Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Reported-by: syzbot <syzkaller@googlegroups.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/unix/af_unix.c

index 5a31307ceb76df1cf699b161963b92e06300aa2b..5d1192ceb13973ee74cc0f208fb91a36adabc643 100644 (file)
@@ -535,12 +535,14 @@ static void unix_release_sock(struct sock *sk, int embrion)
        u->path.mnt = NULL;
        state = sk->sk_state;
        sk->sk_state = TCP_CLOSE;
+
+       skpair = unix_peer(sk);
+       unix_peer(sk) = NULL;
+
        unix_state_unlock(sk);
 
        wake_up_interruptible_all(&u->peer_wait);
 
-       skpair = unix_peer(sk);
-
        if (skpair != NULL) {
                if (sk->sk_type == SOCK_STREAM || sk->sk_type == SOCK_SEQPACKET) {
                        unix_state_lock(skpair);
@@ -555,7 +557,6 @@ static void unix_release_sock(struct sock *sk, int embrion)
 
                unix_dgram_peer_wake_disconnect(sk, skpair);
                sock_put(skpair); /* It may now die */
-               unix_peer(sk) = NULL;
        }
 
        /* Try to flush out this socket. Throw out buffers at least */