On Tue, Feb 04, 2025 at 01:38:50AM +0100, Michal Luczaj wrote:
On 2/3/25 10:57, syzbot wrote:
Hello,
syzbot found the following issue on:
HEAD commit: c2933b2befe2 Merge tag 'net-6.14-rc1' of git://git.kernel...
git tree: net-next
console output: https://syzkaller.appspot.com/x/log.txt?x=16f676b0580000
kernel config: https://syzkaller.appspot.com/x/.config?x=d033b14aeef39158
dashboard link: https://syzkaller.appspot.com/bug?extid=9d55b199192a4be7d02c
compiler: Debian clang version 15.0.6, GNU ld (GNU Binutils for Debian) 2.40
syz repro: https://syzkaller.appspot.com/x/repro.syz?x=13300b24580000
C reproducer: https://syzkaller.appspot.com/x/repro.c?x=12418518580000
Downloadable assets:
disk image: https://storage.googleapis.com/syzbot-assets/c7667ae12603/disk-c2933b2b.raw.xz
vmlinux: https://storage.googleapis.com/syzbot-assets/944ca63002c1/vmlinux-c2933b2b.xz
kernel image: https://storage.googleapis.com/syzbot-assets/30748115bf0b/bzImage-c2933b2b.xz
The issue was bisected to:
commit fcdd2242c0231032fc84e1404315c245ae56322a
Author: Michal Luczaj <mhal@xxxxxxx>
Date: Tue Jan 28 13:15:27 2025 +0000
vsock: Keep the binding until socket destruction
syzbot is correct (thanks), bisected commit introduced a regression.
sock_orphan(sk) is being called without taking into consideration that it
does `sk->sk_wq = NULL`. Later, if SO_LINGER is set, sk->sk_wq gets
dereferenced in virtio_transport_wait_close().
Repro, as shown by syzbot, is simply
from socket import *
lis = socket(AF_VSOCK, SOCK_STREAM)
lis.bind((1, 1234)) # VMADDR_CID_LOCAL
lis.listen()
s = socket(AF_VSOCK, SOCK_STREAM)
s.setsockopt(SOL_SOCKET, SO_LINGER, (1<<32) | 1)
s.connect(lis.getsockname())
s.close()
A way of fixing this is to put sock_orphan(sk) back where it was before the
breaking patch and instead explicitly flip just the SOCK_DEAD bit, i.e.
diff --git a/net/vmw_vsock/af_vsock.c b/net/vmw_vsock/af_vsock.c
index 075695173648..06250bb9afe2 100644
--- a/net/vmw_vsock/af_vsock.c
+++ b/net/vmw_vsock/af_vsock.c
@@ -824,13 +824,14 @@ static void __vsock_release(struct sock *sk, int level)
*/
lock_sock_nested(sk, level);
- sock_orphan(sk);
+ sock_set_flag(sk, SOCK_DEAD);
if (vsk->transport)
vsk->transport->release(vsk);
else if (sock_type_connectible(sk->sk_type))
vsock_remove_sock(vsk);
+ sock_orphan(sk);
sk->sk_shutdown = SHUTDOWN_MASK;
skb_queue_purge(&sk->sk_receive_queue);
I'm not sure this is the most elegant code (sock_orphan(sk) sets SOCK_DEAD
on a socket that is already SOCK_DEAD), but here it goes:
https://lore.kernel.org/netdev/20250204-vsock-linger-nullderef-v1-0-6eb1760fa93e@xxxxxxx/
What about the fix proposed here:
https://lore.kernel.org/lkml/20250203124959.114591-1-aha310510@xxxxxxxxx/
One more note: man socket(7) says lingering also happens on shutdown().
Should vsock follow that?
Good point, I think so.
IMHO we should handle both of them in af_vsock.c if it's possible, but
maybe we need a bit of refactoring.
Anyway, net-next material, right?
Stefano
