On Sat, Mar 16, 2019 at 11:50:27PM -0700, Ming Lei wrote: > On Sat, Mar 16, 2019 at 12:15 AM Keith Busch <kbusch@xxxxxxxxxx> wrote: > > On Fri, Mar 15, 2019 at 01:57:38AM -0700, Jianchao Wang wrote: > > > + if (rq && rq->q == hctx->queue && > > > + rq->mq_hctx == hctx && > > > + (!iter_data->inflight || > > > + blk_mq_rq_state(rq) == MQ_RQ_IN_FLIGHT)) > > > + return iter_data->fn(rq, iter_data->data, reserved); > > > > There is still a window where the check may succeed, but the request is > > being assigned to a completely different request_queue. The callback > > then operates on a request it doesn't own. > > The situation above is only related with 'none' io scheduler. > > However, this patch doesn't change behavior for none given .rq[tag] is same with > .static_rq[tag], so I guess your concern may not be related with this patch. Right, it's not new, but queue iteration wasn't exported for drivers before this.
