On 2019/4/24 23:38, Martin Wilck wrote:
On Wed, 2019-04-24 at 23:19 +0800, Yufen Yu wrote:After commit 396eaf21ee17 ("blk-mq: improve DM's blk-mq IO merging via blk_insert_cloned_request feedback"), map_request() will requeue the tio when issued clone request return BLK_STS_RESOURCE or BLK_STS_DEV_RESOURCE. Thus, if device drive status is error, a tio may be requeued multiple times until the return value is not DM_MAPIO_REQUEUE. That means type-start_iomay be called multiple tims, while type->end_io just be called when IO complete. In fact, even without the commit, setup_clone() fail also can make the tio requeue and miss call type->end_io. As servicer-time path selector for example, it selects path based on in_flight_size, which is increased by st_start_io() and decreased by st_end_io(). Missing call of end_io can lead to in_flight_size count error and let the selector make the wrong choice. In addition, queue-length path selector will also be affected. To fix the problem, we call type->end_io in ->release_clone_rq before tio requeue. It pass map_info to ->release_clone_rq() for requeue path, and pass NULL for the others path. Fixes: 396eaf21ee17 ("blk-mq: improve DM's blk-mq IO merging via blk_insert_cloned_request feedback") Cc: Ming Lei <[email protected]> Cc: Martin Wilck <[email protected]> Cc: Mikulas Patocka <[email protected]> Cc: Hou Tao <[email protected]> Signed-off-by: Yufen Yu <[email protected]> --- V2: - remove the new added interface ->end_stat() in V1 - use ->release_clone_rq() to call path selector ->end_io - add dm_rq_status to show clone status V1: - https://www.redhat.com/archives/dm-devel/2019-April/msg00148.html --- drivers/md/dm-mpath.c | 20 +++++++++++++++++++- drivers/md/dm-rq.c | 19 ++++++++++++++----- drivers/md/dm-target.c | 4 +++- include/linux/device-mapper.h | 13 ++++++++++++- 4 files changed, 48 insertions(+), 8 deletions(-) diff --git a/drivers/md/dm-mpath.c b/drivers/md/dm-mpath.c index 2ee5e357a0a7..e2f6e9c9363f 100644 --- a/drivers/md/dm-mpath.c +++ b/drivers/md/dm-mpath.c @@ -544,8 +544,26 @@ static int multipath_clone_and_map(struct dm_target *ti, struct request *rq, return DM_MAPIO_REMAPPED; }-static void multipath_release_clone(struct request *clone)+static void multipath_release_clone(struct request *clone, + union map_info *map_context + enum dm_rq_status clone_status) { + struct dm_mpath_io *mpio; + struct pgpath *pgpath; + + if (clone_status != DM_CLONE_RQ_OK && !map_context) {Should this be "... && map_context)", maybe ?
Yes, thanks for catching this! This is an obvious error.
+ mpio = get_mpio(map_context); + pgpath = mpio->pgpath; + + if (pgpath) { + struct path_selector *ps = &pgpath->pg->ps; + + if (ps->type->end_io) + ps->type->end_io(ps, &pgpath->path, + mpio->nr_bytes); + } + } + blk_put_request(clone); }diff --git a/drivers/md/dm-rq.c b/drivers/md/dm-rq.cindex b66745bd08bb..4cf309775d2c 100644 --- a/drivers/md/dm-rq.c +++ b/drivers/md/dm-rq.c @@ -168,7 +168,7 @@ static void dm_end_request(struct request *clone, blk_status_t error) struct request *rq = tio->orig;blk_rq_unprep_clone(clone);- tio->ti->type->release_clone_rq(clone); + tio->ti->type->release_clone_rq(clone, NULL, DM_CLONE_RQ_OK);rq_end_stats(md, rq);blk_mq_end_request(rq, error); @@ -201,7 +201,7 @@ static void dm_requeue_original_request(struct dm_rq_target_io *tio, bool delay_ rq_end_stats(md, rq); if (tio->clone) { blk_rq_unprep_clone(tio->clone); - tio->ti->type->release_clone_rq(tio->clone); + tio->ti->type->release_clone_rq(tio->clone, NULL, DM_CLONE_RQ_OK); }dm_mq_delay_requeue_request(rq, delay_ms);@@ -397,8 +397,12 @@ static int map_request(struct dm_rq_target_io *tio) break; case DM_MAPIO_REMAPPED: if (setup_clone(clone, rq, tio, GFP_ATOMIC)) { - /* -ENOMEM */ - ti->type->release_clone_rq(clone); + /* -ENOMEM + * For multipath, before requeue, we need to pass tio->info + * to release_clone_rq, avoiding missing call of path sector + * ->end_io. + */ + ti->type->release_clone_rq(clone, &tio->info, DM_CLONE_RQ_CANCAL); return DM_MAPIO_REQUEUE; }@@ -408,7 +412,12 @@ static int map_request(struct dm_rq_target_io*tio) ret = dm_dispatch_clone_request(clone, rq); if (ret == BLK_STS_RESOURCE || ret == BLK_STS_DEV_RESOURCE) { blk_rq_unprep_clone(clone); - tio->ti->type->release_clone_rq(clone); + /* + * For multipath, before requeue, we need to pass tio->info + * to release_clone_rq, avoiding missing call of path sector + * ->end_io. + */ + tio->ti->type->release_clone_rq(clone, &tio-info, DM_CLONE_RQ_CANCAL);tio->clone = NULL; return DM_MAPIO_REQUEUE; } diff --git a/drivers/md/dm-target.c b/drivers/md/dm-target.c index 314d17ca6466..fd0962a24a54 100644 --- a/drivers/md/dm-target.c +++ b/drivers/md/dm-target.c @@ -136,7 +136,9 @@ static int io_err_clone_and_map_rq(struct dm_target *ti, struct request *rq, return DM_MAPIO_KILL; }-static void io_err_release_clone_rq(struct request *clone)+static void io_err_release_clone_rq(struct request *clone, + union map_info *map_context, + enum dm_rq_status clone_status) { }diff --git a/include/linux/device-mapper.h b/include/linux/device-mapper.h index b0672756d056..9b9ab762a992 100644 --- a/include/linux/device-mapper.h +++ b/include/linux/device-mapper.h @@ -31,6 +31,15 @@ enum dm_queue_mode { DM_TYPE_NVME_BIO_BASED = 4, };+/*+ * clone request status + */ +enum dm_rq_status { + DM_CLONE_RQ_OK, + DM_CLONE_RQ_CANCAL, + DM_CLONE_RQ_MAX, +};"CANCAL" looks strange - did you mean "CANCEL"?
This is a spelling error.
Anyway, what do you need this new enum for? Couldn't you just pass the disposition (i.e. DM_MAPIO_REQUEUE), and use a different value (e.g. DM_ENDIO_DONE) at those call sites where end_io shouldn't be called? Regards, Martin .
Thanks for your suggestion. Passing DM_MAPIO_REQUEUE is a good idea. It indicates that request will be requeued and we need to call ->end_io. However, I am not sure if it is suitable for use DM_ENDIO_DONE. It is strange that we use DM_ENDIO_DONE in dm_requeue_original_request(), which is called in the case of DM_ENDIO_REQUEUE in dm_done(). Yufen Thanks a lot. -- dm-devel mailing list [email protected] https://www.redhat.com/mailman/listinfo/dm-devel
