hamradio: defer ax25 kfree after unregister_netdev
authorLin Ma <linma@zju.edu.cn>
Mon, 8 Nov 2021 10:37:21 +0000 (18:37 +0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 29 Dec 2021 11:23:38 +0000 (12:23 +0100)
commitef5f7bfa19e3fc366f4c6d1a841ceaddf7a9f5d4
treeedb3a3484548979cd34dc09ff96cde95d3ab292e
parentdf8f79bcc2e431bb6e1f0a8202cd2b06fd429889
hamradio: defer ax25 kfree after unregister_netdev

commit 3e0588c291d6ce225f2b891753ca41d45ba42469 upstream.

There is a possible race condition (use-after-free) like below

 (USE)                       |  (FREE)
ax25_sendmsg                 |
 ax25_queue_xmit             |
  dev_queue_xmit             |
   __dev_queue_xmit          |
    __dev_xmit_skb           |
     sch_direct_xmit         | ...
      xmit_one               |
       netdev_start_xmit     | tty_ldisc_kill
        __netdev_start_xmit  |  mkiss_close
         ax_xmit             |   kfree
          ax_encaps          |
                             |

Even though there are two synchronization primitives before the kfree:
1. wait_for_completion(&ax->dead). This can prevent the race with
routines from mkiss_ioctl. However, it cannot stop the routine coming
from upper layer, i.e., the ax25_sendmsg.

2. netif_stop_queue(ax->dev). It seems that this line of code aims to
halt the transmit queue but it fails to stop the routine that already
being xmit.

This patch reorder the kfree after the unregister_netdev to avoid the
possible UAF as the unregister_netdev() is well synchronized and won't
return if there is a running routine.

Signed-off-by: Lin Ma <linma@zju.edu.cn>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/net/hamradio/mkiss.c