Hi Johan,
>> When trying to allocate skb for new PDU, l2cap_chan is unlocked so we
>> can sleep waiting for memory as otherwise there's possible deadlock as
>> fixed in e454c84464. However, in a6a5568c03 lock was moved from socket
>> to channel level and it's no longer safe to just unlock and lock again
>> without checking l2cap_chan state since channel can be disconnected
>> when lock is not held.
>>
>> This patch adds missing checks for l2cap_chan state when returning from
>> call which allocates skb.
>>
>> Scenario is easily reproducible by running rfcomm-tester in a loop.
>>
>> BUG: unable to handle kernel NULL pointer dereference at (null)
>> IP: [<ffffffffa0442169>] l2cap_do_send+0x29/0x120 [bluetooth]
>> PGD 0
>> Oops: 0000 [#1] SMP
>> Modules linked in:
>> CPU: 7 PID: 4038 Comm: krfcommd Not tainted 3.14.0-rc2+ #15
>> Hardware name: Dell Inc. OptiPlex 790/0HY9JP, BIOS A10 11/24/2011
>> task: ffff8802bdd731c0 ti: ffff8801ec986000 task.ti: ffff8801ec986000
>> RIP: 0010:[<ffffffffa0442169>] [<ffffffffa0442169>] l2cap_do_send+0x29/0x120
>> RSP: 0018:ffff8801ec987ad8 EFLAGS: 00010202
>> RAX: 0000000000000000 RBX: ffff8800c5796800 RCX: 0000000000000000
>> RDX: ffff880410e7a800 RSI: ffff8802b6c1da00 RDI: ffff8800c5796800
>> RBP: ffff8801ec987af8 R08: 00000000000000c0 R09: 0000000000000300
>> R10: 000000000000573b R11: 000000000000573a R12: ffff8802b6c1da00
>> R13: 0000000000000000 R14: ffff8802b6c1da00 R15: 0000000000000000
>> FS: 0000000000000000(0000) GS:ffff88042dce0000(0000) knlGS:0000000000000000
>> CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>> CR2: 0000000000000000 CR3: 000000041257c000 CR4: 00000000000407e0
>> Stack:
>> ffff8801ec987d78 ffff8800c5796800 ffff8801ec987d78 0000000000000000
>> ffff8801ec987ba8 ffffffffa0449e37 0000000000000004 ffff8801ec987af0
>> ffff8801ec987d40 0000000000000282 0000000000000000 ffffffff00000004
>> Call Trace:
>> [<ffffffffa0449e37>] l2cap_chan_send+0xaa7/0x1120 [bluetooth]
>> [<ffffffff81770100>] ? _raw_spin_unlock_bh+0x20/0x40
>> [<ffffffffa045188b>] l2cap_sock_sendmsg+0xcb/0x110 [bluetooth]
>> [<ffffffff81652b0f>] sock_sendmsg+0xaf/0xc0
>> [<ffffffff810a8381>] ? update_curr+0x141/0x200
>> [<ffffffff810a8961>] ? dequeue_entity+0x181/0x520
>> [<ffffffff81652b60>] kernel_sendmsg+0x40/0x60
>> [<ffffffffa04a8505>] rfcomm_send_frame+0x45/0x70 [rfcomm]
>> [<ffffffff810766f0>] ? internal_add_timer+0x20/0x50
>> [<ffffffffa04a8564>] rfcomm_send_cmd+0x34/0x60 [rfcomm]
>> [<ffffffffa04a8605>] rfcomm_send_disc+0x75/0xa0 [rfcomm]
>> [<ffffffffa04aacec>] rfcomm_run+0x8cc/0x1a30 [rfcomm]
>> [<ffffffffa04aa420>] ? rfcomm_check_accept+0xc0/0xc0 [rfcomm]
>> [<ffffffff8108e3a9>] kthread+0xc9/0xe0
>> [<ffffffff8108e2e0>] ? flush_kthread_worker+0xb0/0xb0
>> [<ffffffff817795fc>] ret_from_fork+0x7c/0xb0
>> [<ffffffff8108e2e0>] ? flush_kthread_worker+0xb0/0xb0
>> Code: 00 00 66 66 66 66 90 55 48 89 e5 48 83 ec 20 f6 05 d6 a3 02 00 04
>> RIP [<ffffffffa0442169>] l2cap_do_send+0x29/0x120 [bluetooth]
>> RSP <ffff8801ec987ad8>
>> CR2: 0000000000000000
>>
>> Signed-off-by: Andrzej Kaczmarek <andrzej.kaczmarek@xxxxxxxxx>
>> ---
>> net/bluetooth/l2cap_core.c | 10 ++++++++++
>> 1 file changed, 10 insertions(+)
>>
>> diff --git a/net/bluetooth/l2cap_core.c b/net/bluetooth/l2cap_core.c
>> index 6ace116..cfc7c69 100644
>> --- a/net/bluetooth/l2cap_core.c
>> +++ b/net/bluetooth/l2cap_core.c
>> @@ -2434,6 +2434,11 @@ int l2cap_chan_send(struct l2cap_chan *chan, struct msghdr *msg, size_t len,
>> if (IS_ERR(skb))
>> return PTR_ERR(skb);
>>
>> + if (chan->state != BT_CONNECTED) {
>> + kfree_skb(skb);
>> + return -ENOTCONN;
>> + }
>> +
>> l2cap_do_send(chan, skb);
>> return len;
>> }
>> @@ -2483,6 +2488,11 @@ int l2cap_chan_send(struct l2cap_chan *chan, struct msghdr *msg, size_t len,
>> if (IS_ERR(skb))
>> return PTR_ERR(skb);
>>
>> + if (chan->state != BT_CONNECTED) {
>> + kfree_skb(skb);
>> + return -ENOTCONN;
>> + }
>> +
>> l2cap_do_send(chan, skb);
>> err = len;
>> break;
>
> Acked-by: Johan Hedberg <johan.hedberg@xxxxxxxxx>
>
> However, I'd consider adding code comment explaining the lock was
> released and reacquired and the state needs to therefore be rechecked.
I agree. This needs comments in both cases. Especially since we are doing the same for the other check when using ERTM.
Regards
Marcel
--
To unsubscribe from this list: send the line "unsubscribe linux-bluetooth" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
