]> git.hungrycats.org Git - linux/commit
bluetooth/l2cap: sync sock recv cb and release
authorEdward Adam Davis <eadavis@qq.com>
Sat, 15 Jun 2024 01:45:54 +0000 (09:45 +0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 25 Jul 2024 07:53:37 +0000 (09:53 +0200)
commit3b732449b78183d17178db40be3a4401cf3cd629
treed60622879ab7ee014942ffb9d70767a3610ce2a3
parentf1db2c9acb6372d2cf575ddd3976b855928a2cb8
bluetooth/l2cap: sync sock recv cb and release

[ Upstream commit 89e856e124f9ae548572c56b1b70c2255705f8fe ]

The problem occurs between the system call to close the sock and hci_rx_work,
where the former releases the sock and the latter accesses it without lock protection.

           CPU0                       CPU1
           ----                       ----
           sock_close                 hci_rx_work
   l2cap_sock_release         hci_acldata_packet
   l2cap_sock_kill            l2cap_recv_frame
   sk_free                    l2cap_conless_channel
                              l2cap_sock_recv_cb

If hci_rx_work processes the data that needs to be received before the sock is
closed, then everything is normal; Otherwise, the work thread may access the
released sock when receiving data.

Add a chan mutex in the rx callback of the sock to achieve synchronization between
the sock release and recv cb.

Sock is dead, so set chan data to NULL, avoid others use invalid sock pointer.

Reported-and-tested-by: syzbot+b7f6f8c9303466e16c8a@syzkaller.appspotmail.com
Signed-off-by: Edward Adam Davis <eadavis@qq.com>
Signed-off-by: Luiz Augusto von Dentz <luiz.von.dentz@intel.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
net/bluetooth/l2cap_sock.c