nvme-cli connect regression

[Sagi Grimberg]

On 4/17/25 13:12, Daniel Wagner wrote:
> On Thu, Apr 17, 2025 at 12:26:51AM +0300, Sagi Grimberg wrote:
>>> I think I can reproduce it. I have two connections and do a
>>>
>>>      while nvme read ....; do yes; done
>>>
>>> and block one of the connections, when the error handler kicks in the id
>>> ns command fails with EINTR.
>> What happens before Caleb's patch though?
> I am able to reproduce it with the distro kernel 6.13.9 on my laptop
> (real hardware) but not with one of my test setups (qemu) and a strip
> down configuration. Neither HEAD, v6.14 or v6.13.8 is returning EINTR.
>
> I will try the distro config in qemu as next.

Does this change make the issue go away?
diff --git a/drivers/nvme/host/tcp.c b/drivers/nvme/host/tcp.c
index 5041cbfd8272..87177ec75e49 100644
--- a/drivers/nvme/host/tcp.c
+++ b/drivers/nvme/host/tcp.c
@@ -1517,7 +1517,6 @@ static int nvme_tcp_init_connection(struct 
nvme_tcp_queue *queue)
                 msg.msg_control = cbuf;
                 msg.msg_controllen = sizeof(cbuf);
         }
-       msg.msg_flags = MSG_WAITALL;
         ret = kernel_recvmsg(queue->sock, &msg, &iov, 1,
                         iov.iov_len, msg.msg_flags);
         if (ret >= 0 && ret < sizeof(*icresp))