nvme-tcp: fix possible UAF in nvme_tcp_poll

nvme_tcp_poll() may race with the send path error handler because
it may complete the request while it is actively being polled for
completion, resulting in a UAF panic [1]:

We should make sure to stop polling when we see an error when
trying to read from the socket. Hence make sure to propagate the
error so that the block layer breaks the polling cycle.

[1]:
--
[35665.692310] nvme nvme2: failed to send request -13
[35665.702265] nvme nvme2: unsupported pdu type (3)
[35665.702272] BUG: kernel NULL pointer dereference, address: 0000000000000000
[35665.702542] nvme nvme2: queue 1 receive failed:  -22
[35665.703209] #PF: supervisor write access in kernel mode
[35665.703213] #PF: error_code(0x0002) - not-present page
[35665.703214] PGD 8000003801cce067 P4D 8000003801cce067 PUD 37e6f79067 PMD 0
[35665.703220] Oops: 0002 [#1] SMP PTI
[35665.703658] nvme nvme2: starting error recovery
[35665.705809] Hardware name: Inspur aaabbb/YZMB-00882-104, BIOS 4.1.26 09/22/2022
[35665.705812] Workqueue: kblockd blk_mq_requeue_work
[35665.709172] RIP: 0010:_raw_spin_lock+0xc/0x30
[35665.715788] Call Trace:
[35665.716201]  <TASK>
[35665.716613]  ? show_trace_log_lvl+0x1c1/0x2d9
[35665.717049]  ? show_trace_log_lvl+0x1c1/0x2d9
[35665.717457]  ? blk_mq_request_bypass_insert+0x2c/0xb0
[35665.717950]  ? __die_body.cold+0x8/0xd
[35665.718361]  ? page_fault_oops+0xac/0x140
[35665.718749]  ? blk_mq_start_request+0x30/0xf0
[35665.719144]  ? nvme_tcp_queue_rq+0xc7/0x170 [nvme_tcp]
[35665.719547]  ? exc_page_fault+0x62/0x130
[35665.719938]  ? asm_exc_page_fault+0x22/0x30
[35665.720333]  ? _raw_spin_lock+0xc/0x30
[35665.720723]  blk_mq_request_bypass_insert+0x2c/0xb0
[35665.721101]  blk_mq_requeue_work+0xa5/0x180
[35665.721451]  process_one_work+0x1e8/0x390
[35665.721809]  worker_thread+0x53/0x3d0
[35665.722159]  ? process_one_work+0x390/0x390
[35665.722501]  kthread+0x124/0x150
[35665.722849]  ? set_kthread_struct+0x50/0x50
[35665.723182]  ret_from_fork+0x1f/0x30

Reported-by: Zhang Guanghui <[email protected]>
Signed-off-by: Sagi Grimberg <[email protected]>
Reviewed-by: Chaitanya Kulkarni <[email protected]>
Signed-off-by: Keith Busch <[email protected]>

drivers/nvme/host/tcp.c

@@ -2699,16 +2699,17 @@ static int nvme_tcp_poll(struct blk_mq_hw_ctx *hctx, struct io_comp_batch *iob)
 {
 	struct nvme_tcp_queue *queue = hctx->driver_data;
 	struct sock *sk = queue->sock->sk;
+	int ret;
 
 	if (!test_bit(NVME_TCP_Q_LIVE, &queue->flags))
 		return 0;
 
 	set_bit(NVME_TCP_Q_POLLING, &queue->flags);
 	if (sk_can_busy_loop(sk) && skb_queue_empty_lockless(&sk->sk_receive_queue))
 		sk_busy_loop(sk, true);
-	nvme_tcp_try_recv(queue);
+	ret = nvme_tcp_try_recv(queue);
 	clear_bit(NVME_TCP_Q_POLLING, &queue->flags);
-	return queue->nr_cqe;
+	return ret < 0 ? ret : queue->nr_cqe;
 }
 
 static int nvme_tcp_get_address(struct nvme_ctrl *ctrl, char *buf, int size)