On Tue, Nov 15 2016 at 6:35pm -0500,
Bart Van Assche <bart.vanassche@xxxxxxxxxxx> wrote:
> If a single-queue dm device is stacked on top of multi-queue block
> devices and map_tio_request() is called while there are no paths then
> the request will be prepared for a single-queue path. If a path is
> added after a request was prepared and before __multipath_map() is
> called return DM_MAPIO_REQUEUE such that it gets unprepared and
> reprepared as a blk-mq request.
This patch makes little sense to me. There isn't a scenario that I'm
aware of that would allow the request_queue to transition between old
.request_fn and new blk-mq.
The dm-table code should prevent this.
Mike
> diff --git a/drivers/md/dm-mpath.c b/drivers/md/dm-mpath.c
> index 7559537..6b20349 100644
> --- a/drivers/md/dm-mpath.c
> +++ b/drivers/md/dm-mpath.c
> @@ -541,6 +541,7 @@ static int __multipath_map(struct dm_target *ti, struct request *clone,
> size_t nr_bytes = clone ? blk_rq_bytes(clone) : blk_rq_bytes(rq);
> struct pgpath *pgpath;
> struct block_device *bdev;
> + struct request_queue *q;
> struct dm_mpath_io *mpio;
>
> /* Do we need to select a new pgpath? */
> @@ -558,6 +559,18 @@ static int __multipath_map(struct dm_target *ti, struct request *clone,
> return r;
> }
>
> + bdev = pgpath->path.dev->bdev;
> + q = bdev_get_queue(bdev);
> +
> + /*
> + * When a request is prepared if there are no paths it may happen that
> + * the request was prepared for a single-queue path and that a
> + * multiqueue path is added before __multipath_map() is called. If
> + * that happens requeue to trigger unprepare and reprepare.
> + */
> + if ((clone && q->mq_ops) || (!clone && !q->mq_ops))
> + return r;
> +
> mpio = set_mpio(m, map_context);
> if (!mpio)
> /* ENOMEM, requeue */
> @@ -566,22 +579,20 @@ static int __multipath_map(struct dm_target *ti, struct request *clone,
> mpio->pgpath = pgpath;
> mpio->nr_bytes = nr_bytes;
>
> - bdev = pgpath->path.dev->bdev;
> -
> if (clone) {
> /*
> * Old request-based interface: allocated clone is passed in.
> * Used by: .request_fn stacked on .request_fn path(s).
> */
> - clone->q = bdev_get_queue(bdev);
> + clone->q = q;
> } else {
> /*
> * blk-mq request-based interface; used by both:
> * .request_fn stacked on blk-mq path(s) and
> * blk-mq stacked on blk-mq path(s).
> */
> - clone = blk_mq_alloc_request(bdev_get_queue(bdev),
> - rq_data_dir(rq), BLK_MQ_REQ_NOWAIT);
> + clone = blk_mq_alloc_request(q, rq_data_dir(rq),
> + BLK_MQ_REQ_NOWAIT);
> if (IS_ERR(clone)) {
> /* EBUSY, ENODEV or EWOULDBLOCK; requeue */
> clear_request_fn_mpio(m, map_context);
> --
> 2.10.1
>
> --
> dm-devel mailing list
> dm-devel@xxxxxxxxxx
> https://www.redhat.com/mailman/listinfo/dm-devel
--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel
