On Tue, 21 Oct 2025 at 12:48, Stefano Garzarella [off-list ref] wrote:

On Tue, 21 Oct 2025 at 10:27, Stefano Garzarella [off-list ref] wrote:

quoted

Hi Michal,

On Mon, Oct 20, 2025 at 05:02:56PM -0700, syzbot wrote:

quoted

Hello,

syzbot found the following issue on:

HEAD commit:    d9043c79ba68 Merge tag 'sched_urgent_for_v6.18_rc2' of git..
git tree:       upstream
console output: https://syzkaller.appspot.com/x/log.txt?x=130983cd980000
kernel config:  https://syzkaller.appspot.com/x/.config?x=f3e7b5a3627a90dd
dashboard link: https://syzkaller.appspot.com/bug?extid=10e35716f8e4929681fa
compiler:       gcc (Debian 12.2.0-14+deb12u1) 12.2.0, GNU ld (GNU Binutils for Debian) 2.40
syz repro:      https://syzkaller.appspot.com/x/repro.syz?x=17f0f52f980000
C reproducer:   https://syzkaller.appspot.com/x/repro.c?x=11ea9734580000

Downloadable assets:
disk image (non-bootable): https://storage.googleapis.com/syzbot-assets/d900f083ada3/non_bootable_disk-d9043c79.raw.xz
vmlinux: https://storage.googleapis.com/syzbot-assets/0546b6eaf1aa/vmlinux-d9043c79.xz
kernel image: https://storage.googleapis.com/syzbot-assets/81285b4ada51/bzImage-d9043c79.xz

IMPORTANT: if you fix the issue, please add the following tag to the commit:
Reported-by: syzbot+10e35716f8e4929681fa@syzkaller.appspotmail.com

======================================================
WARNING: possible circular locking dependency detected
syzkaller #0 Not tainted
------------------------------------------------------
syz.0.17/6098 is trying to acquire lock:
ffff8880363b8258 (sk_lock-AF_VSOCK){+.+.}-{0:0}, at: lock_sock include/net/sock.h:1679 [inline]
ffff8880363b8258 (sk_lock-AF_VSOCK){+.+.}-{0:0}, at: vsock_linger+0x25e/0x4d0 net/vmw_vsock/af_vsock.c:1066

Could this be related to our recent work on linger in vsock?

quoted

but task is already holding lock:
ffffffff906260a8 (vsock_register_mutex){+.+.}-{4:4}, at: vsock_assign_transport+0xf2/0x900 net/vmw_vsock/af_vsock.c:469

which lock already depends on the new lock.


the existing dependency chain (in reverse order) is:

-> #1 (vsock_register_mutex){+.+.}-{4:4}:
      __mutex_lock_common kernel/locking/mutex.c:598 [inline]
      __mutex_lock+0x193/0x1060 kernel/locking/mutex.c:760
      vsock_registered_transport_cid net/vmw_vsock/af_vsock.c:560 [inline]

Ah, no maybe this is related to commit 209fd720838a ("vsock:
Fix transport_{g2h,h2g} TOCTOU") where we added locking in
vsock_find_cid().

Maybe we can just move the checks on top of __vsock_bind() to the
caller. I mean:

        /* First ensure this socket isn't already bound. */
        if (vsock_addr_bound(&vsk->local_addr))
                return -EINVAL;

        /* Now bind to the provided address or select appropriate values if
         * none are provided (VMADDR_CID_ANY and VMADDR_PORT_ANY).  Note that
         * like AF_INET prevents binding to a non-local IP address (in most
         * cases), we only allow binding to a local CID.
         */
        if (addr->svm_cid != VMADDR_CID_ANY && !vsock_find_cid(addr->svm_cid))
                return -EADDRNOTAVAIL;

We have 2 callers: vsock_auto_bind() and vsock_bind().

vsock_auto_bind() is already checking if the socket is already bound,
if not is setting VMADDR_CID_ANY, so we can skip those checks.

In vsock_bind() we can do the checks before lock_sock(sk), at least the
checks on vm_addr, calling vsock_find_cid().

I'm preparing a patch to do this.

mmm, no, this is more related to vsock_linger() where sk_wait_event()
releases and locks again the sk_lock.
So, it should be related to commit 687aa0c5581b ("vsock: Fix
transport_* TOCTOU") where we take vsock_register_mutex in
vsock_assign_transport() while calling vsk->transport->release().

So, maybe we need to move the release and vsock_deassign_transport()
after unlocking vsock_register_mutex.

I implemented this here:
https://lore.kernel.org/netdev/20251021121718.137668-1-sgarzare@redhat.com/ (local)

sysbot successfully tested it.

Stefano