Re: [syzbot] general protection fault in l2cap_chan_timeout (2)

[Luiz Augusto von Dentz]

Hi,

On Mon, May 31, 2021 at 2:04 AM Hillf Danton <hdanton@xxxxxxxx> wrote:
>
> On Mon, 31 May 2021 00:19:17 -0700
> > Hello,
> >
> > syzbot found the following issue on:
> >
> > HEAD commit:    ad9f25d3 Merge tag 'netfs-lib-fixes-20200525' of git://git..
> > git tree:       upstream
> > console output: https://syzkaller.appspot.com/x/log.txt?x=173d383dd00000
> > kernel config:  https://syzkaller.appspot.com/x/.config?x=266cda122a0b56c
> > dashboard link: https://syzkaller.appspot.com/bug?extid=008cdbf7a9044c2c2f99
> >
> > Unfortunately, I don't have any reproducer for this issue yet.
> >
> > IMPORTANT: if you fix the issue, please add the following tag to the commit:
> > Reported-by: syzbot+008cdbf7a9044c2c2f99@xxxxxxxxxxxxxxxxxxxxxxxxx
> >
> > general protection fault, probably for non-canonical address 0xdffffc000000005a: 0000 [#1] PREEMPT SMP KASAN
> > KASAN: null-ptr-deref in range [0x00000000000002d0-0x00000000000002d7]
> > CPU: 0 PID: 8 Comm: kworker/0:2 Not tainted 5.13.0-rc3-syzkaller #0
> > Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS Google 01/01/2011
> > Workqueue: events l2cap_chan_timeout
> > RIP: 0010:__mutex_lock_common kernel/locking/mutex.c:941 [inline]
> > RIP: 0010:__mutex_lock+0xf6/0x10c0 kernel/locking/mutex.c:1104
> > Code: d0 7c 08 84 d2 0f 85 cc 0c 00 00 8b 15 e3 55 5f 07 85 d2 75 29 48 8d 7d 60 48 b8 00 00 00 00 00 fc ff df 48 89 fa 48 c1 ea 03 <80> 3c 02 00 0f 85 db 0e 00 00 48 3b 6d 60 0f 85 5a 0a 00 00 bf 01
> > RSP: 0018:ffffc90000cd7b78 EFLAGS: 00010216
> > RAX: dffffc0000000000 RBX: 0000000000000000 RCX: 0000000000000001
> > RDX: 000000000000005a RSI: 0000000000000000 RDI: 00000000000002d0
> > RBP: 0000000000000270 R08: ffffffff880a40d9 R09: 0000000000000000
> > R10: ffffffff814b4be0 R11: 0000000000000000 R12: 0000000000000000
> > R13: dffffc0000000000 R14: ffff888072e47020 R15: ffff8880b9c34a40
> > FS:  0000000000000000(0000) GS:ffff8880b9c00000(0000) knlGS:0000000000000000
> > CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > CR2: 00007fcac15f1d58 CR3: 00000000628fa000 CR4: 0000000000350ef0
> > DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> > DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000600
> > Call Trace:
> >  l2cap_chan_timeout+0x69/0x2f0 net/bluetooth/l2cap_core.c:422
> >  process_one_work+0x98d/0x1600 kernel/workqueue.c:2276
> >  worker_thread+0x64c/0x1120 kernel/workqueue.c:2422
> >  kthread+0x3b1/0x4a0 kernel/kthread.c:313
> >  ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:294
>
> To fix the uaf reported, 1) releases connection through rcu 2) detects race
> under rcu lock in the delayed work callback.
>
> Note it is only for idea and thoughts are welcome if it makes sense to you.
>
> +++ x/net/bluetooth/l2cap_core.c
> @@ -414,12 +414,25 @@ static void l2cap_chan_timeout(struct wo
>  {
>         struct l2cap_chan *chan = container_of(work, struct l2cap_chan,
>                                                chan_timer.work);
> -       struct l2cap_conn *conn = chan->conn;
> +       struct l2cap_conn *conn;
>         int reason;
>
> +       rcu_read_lock();
> +       conn = chan->conn;
> +       if (conn && !kref_get_unless_zero(&conn->ref))
> +               conn = NULL;
> +       rcu_read_unlock();
> +
> +       if (!conn)
> +               goto put;
> +
>         BT_DBG("chan %p state %s", chan, state_to_string(chan->state));
>
>         mutex_lock(&conn->chan_lock);
> +
> +       if (!chan->conn)
> +               goto put;
> +
>         /* __set_chan_timer() calls l2cap_chan_hold(chan) while scheduling
>          * this work. No need to call l2cap_chan_hold(chan) here again.
>          */
> @@ -438,9 +451,13 @@ static void l2cap_chan_timeout(struct wo
>         chan->ops->close(chan);
>
>         l2cap_chan_unlock(chan);
> +put:
>         l2cap_chan_put(chan);
>
> +       if (!conn)
> +               return;
>         mutex_unlock(&conn->chan_lock);
> +       l2cap_conn_put(conn);
>  }
>
>  struct l2cap_chan *l2cap_chan_create(void)
> @@ -1915,12 +1932,19 @@ static void l2cap_conn_del(struct hci_co
>         l2cap_conn_put(conn);
>  }
>
> +static void l2cap_conn_rcu_fn(struct rcu_head *r)
> +{
> +       struct l2cap_conn *conn = container_of(r, struct l2cap_conn, rcu);
> +
> +       kfree(conn);
> +}
> +
>  static void l2cap_conn_free(struct kref *ref)
>  {
>         struct l2cap_conn *conn = container_of(ref, struct l2cap_conn, ref);
>
>         hci_conn_put(conn->hcon);
> -       kfree(conn);

Shouldn't we actually cancel the timeout work if the connection is
freed here? At least I don't see a valid reason to have a l2cap_chan
without l2cap_conn.

> +       call_rcu(&conn->rcu, l2cap_conn_rcu_fn);
>  }
>
>  struct l2cap_conn *l2cap_conn_get(struct l2cap_conn *conn)



-- 
Luiz Augusto von Dentz