Re: [PATCH 2/3] nbd: unify behavior in timeout no matter how many sockets is configured

[Josef Bacik <josef@toxicpanda.com>]

On 9/21/20 6:57 AM, Hou Pu wrote:
> When an nbd device is configured with multiple sockets, the request
> is requeued to an active socket in xmit_timeout, the original socket
> is closed if not.
>
> Some time, the backend nbd server hang, thus all sockets will be
> dropped and the nbd device is not usable. It would be better to have an
> option to wait for more time (just reset timer in nbd_xmit_timeout).
> Like what we do if we only have one socket. This patch allows it.
>
> Signed-off-by: Hou Pu <houpu@bytedance.com>
> ---
>   drivers/block/nbd.c | 7 ++-----
>   1 file changed, 2 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/block/nbd.c b/drivers/block/nbd.c
> index 538e9dcf5bf2..4c0bbb981cbc 100644
> --- a/drivers/block/nbd.c
> +++ b/drivers/block/nbd.c
> @@ -400,8 +400,7 @@ static enum blk_eh_timer_return nbd_xmit_timeout(struct request *req,
>   	    nbd->tag_set.timeout)
>   		goto error_out;
>   
> -	if (config->num_connections > 1 ||
> -	    (config->num_connections == 1 && nbd->tag_set.timeout)) {
> +	if (nbd->tag_set.timeout) {

So now if you don't set a .timeout and have num_connections > 1 we don't close 
the socket, so this won't work.  Thanks,

Josef