On 6/28/25 6:18 AM, Yu Kuai wrote: > Hi, > > 在 2025/06/27 19:04, Ming Lei 写道: >> I guess the patch in the following link may be simper, both two take >> similar approach: >> >> https://lore.kernel.org/linux-block/aFjbavzLAFO0Q7n1@fedora/ > > I this the above approach has concurrent problems if nbd_start_device > concurrent with nbd_start_device: > > t1: > nbd_start_device > lock > num_connections = 1 > unlock > t2: > nbd_add_socket > lock > config->num_connections++ > unlock > t3: > nbd_start_device > lock > num_connections = 2 > unlock > blk_mq_update_nr_hw_queues > > blk_mq_update_nr_hw_queues > //nr_hw_queues updated to 1 before failure > return -EINVAL > In the above case, yes I see that t1 would return -EINVAL (as config->num_connections doesn't match with num_connections) but then t3 would succeed to update nr_hw_queue (as both config->num_connections and num_connections set to 2 this time). Isn't it? If yes, then the above patch (from Ming) seems good. Thanks, --Nilay
