dm: always defer request allocation to the owner of the request_queue

DM already calls blk_mq_alloc_request on the request_queue of the
underlying device if it is a blk-mq device.  But now that we allow drivers
to allocate additional data and initialize it ahead of time we need to do
the same for all drivers.   Doing so and using the new cmd_size
infrastructure in the block layer greatly simplifies the dm-rq and mpath
code, and should also make arbitrary combinations of SQ and MQ devices
with SQ or MQ device mapper tables easily possible as a further step.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Jens Axboe <axboe@fb.com>

1 files changed, 56 insertions, 195 deletions

diff --git a/drivers/md/dm-rq.c b/drivers/md/dm-rq.c
index 3f12916f2424..8d0683474767 100644
--- a/drivers/md/dm-rq.c
+++ b/drivers/md/dm-rq.c
@@ -109,28 +109,6 @@ void dm_stop_queue(struct request_queue *q)
 		dm_mq_stop_queue(q);
 }
 
-static struct dm_rq_target_io *alloc_old_rq_tio(struct mapped_device *md,
-						gfp_t gfp_mask)
-{
-	return mempool_alloc(md->io_pool, gfp_mask);
-}
-
-static void free_old_rq_tio(struct dm_rq_target_io *tio)
-{
-	mempool_free(tio, tio->md->io_pool);
-}
-
-static struct request *alloc_old_clone_request(struct mapped_device *md,
-					       gfp_t gfp_mask)
-{
-	return mempool_alloc(md->rq_pool, gfp_mask);
-}
-
-static void free_old_clone_request(struct mapped_device *md, struct request *rq)
-{
-	mempool_free(rq, md->rq_pool);
-}
-
 /*
  * Partial completion handling for request-based dm
  */
@@ -185,7 +163,7 @@ static void end_clone_bio(struct bio *clone)
 
 static struct dm_rq_target_io *tio_from_request(struct request *rq)
 {
-	return (rq->q->mq_ops ? blk_mq_rq_to_pdu(rq) : rq->special);
+	return blk_mq_rq_to_pdu(rq);
 }
 
 static void rq_end_stats(struct mapped_device *md, struct request *orig)
@@ -233,31 +211,6 @@ static void rq_completed(struct mapped_device *md, int rw, bool run_queue)
 	dm_put(md);
 }
 
-static void free_rq_clone(struct request *clone)
-{
-	struct dm_rq_target_io *tio = clone->end_io_data;
-	struct mapped_device *md = tio->md;
-
-	blk_rq_unprep_clone(clone);
-
-	/*
-	 * It is possible for a clone_old_rq() allocated clone to
-	 * get passed in -- it may not yet have a request_queue.
-	 * This is known to occur if the error target replaces
-	 * a multipath target that has a request_fn queue stacked
-	 * on blk-mq queue(s).
-	 */
-	if (clone->q && clone->q->mq_ops)
-		/* stacked on blk-mq queue(s) */
-		tio->ti->type->release_clone_rq(clone);
-	else if (!md->queue->mq_ops)
-		/* request_fn queue stacked on request_fn queue(s) */
-		free_old_clone_request(md, clone);
-
-	if (!md->queue->mq_ops)
-		free_old_rq_tio(tio);
-}
-
 /*
  * Complete the clone and the original request.
  * Must be called without clone's queue lock held,
@@ -270,7 +223,9 @@ static void dm_end_request(struct request *clone, int error)
 	struct mapped_device *md = tio->md;
 	struct request *rq = tio->orig;
 
-	free_rq_clone(clone);
+	blk_rq_unprep_clone(clone);
+	tio->ti->type->release_clone_rq(clone);
+
 	rq_end_stats(md, rq);
 	if (!rq->q->mq_ops)
 		blk_end_request_all(rq, error);
@@ -279,22 +234,6 @@ static void dm_end_request(struct request *clone, int error)
 	rq_completed(md, rw, true);
 }
 
-static void dm_unprep_request(struct request *rq)
-{
-	struct dm_rq_target_io *tio = tio_from_request(rq);
-	struct request *clone = tio->clone;
-
-	if (!rq->q->mq_ops) {
-		rq->special = NULL;
-		rq->rq_flags &= ~RQF_DONTPREP;
-	}
-
-	if (clone)
-		free_rq_clone(clone);
-	else if (!tio->md->queue->mq_ops)
-		free_old_rq_tio(tio);
-}
-
 /*
  * Requeue the original request of a clone.
  */
@@ -333,7 +272,10 @@ static void dm_requeue_original_request(struct dm_rq_target_io *tio, bool delay_
 	int rw = rq_data_dir(rq);
 
 	rq_end_stats(md, rq);
-	dm_unprep_request(rq);
+	if (tio->clone) {
+		blk_rq_unprep_clone(tio->clone);
+		tio->ti->type->release_clone_rq(tio->clone);
+	}
 
 	if (!rq->q->mq_ops)
 		dm_old_requeue_request(rq);
@@ -388,14 +330,11 @@ static void dm_softirq_done(struct request *rq)
 	if (!clone) {
 		rq_end_stats(tio->md, rq);
 		rw = rq_data_dir(rq);
-		if (!rq->q->mq_ops) {
+		if (!rq->q->mq_ops)
 			blk_end_request_all(rq, tio->error);
-			rq_completed(tio->md, rw, false);
-			free_old_rq_tio(tio);
-		} else {
+		else
 			blk_mq_end_request(rq, tio->error);
-			rq_completed(tio->md, rw, false);
-		}
+		rq_completed(tio->md, rw, false);
 		return;
 	}
 
@@ -439,16 +378,6 @@ static void end_clone_request(struct request *clone, int error)
 {
 	struct dm_rq_target_io *tio = clone->end_io_data;
 
-	if (!clone->q->mq_ops) {
-		/*
-		 * For just cleaning up the information of the queue in which
-		 * the clone was dispatched.
-		 * The clone is *NOT* freed actually here because it is alloced
-		 * from dm own mempool (RQF_ALLOCED isn't set).
-		 */
-		__blk_put_request(clone->q, clone);
-	}
-
 	/*
 	 * Actual request completion is done in a softirq context which doesn't
 	 * hold the clone's queue lock.  Otherwise, deadlock could occur because:
@@ -506,28 +435,6 @@ static int setup_clone(struct request *clone, struct request *rq,
 	return 0;
 }
 
-static struct request *clone_old_rq(struct request *rq, struct mapped_device *md,
-				    struct dm_rq_target_io *tio, gfp_t gfp_mask)
-{
-	/*
-	 * Create clone for use with .request_fn request_queue
-	 */
-	struct request *clone;
-
-	clone = alloc_old_clone_request(md, gfp_mask);
-	if (!clone)
-		return NULL;
-
-	blk_rq_init(NULL, clone);
-	if (setup_clone(clone, rq, tio, gfp_mask)) {
-		/* -ENOMEM */
-		free_old_clone_request(md, clone);
-		return NULL;
-	}
-
-	return clone;
-}
-
 static void map_tio_request(struct kthread_work *work);
 
 static void init_tio(struct dm_rq_target_io *tio, struct request *rq,
@@ -549,60 +456,6 @@ static void init_tio(struct dm_rq_target_io *tio, struct request *rq,
 		kthread_init_work(&tio->work, map_tio_request);
 }
 
-static struct dm_rq_target_io *dm_old_prep_tio(struct request *rq,
-					       struct mapped_device *md,
-					       gfp_t gfp_mask)
-{
-	struct dm_rq_target_io *tio;
-	int srcu_idx;
-	struct dm_table *table;
-
-	tio = alloc_old_rq_tio(md, gfp_mask);
-	if (!tio)
-		return NULL;
-
-	init_tio(tio, rq, md);
-
-	table = dm_get_live_table(md, &srcu_idx);
-	/*
-	 * Must clone a request if this .request_fn DM device
-	 * is stacked on .request_fn device(s).
-	 */
-	if (!dm_table_all_blk_mq_devices(table)) {
-		if (!clone_old_rq(rq, md, tio, gfp_mask)) {
-			dm_put_live_table(md, srcu_idx);
-			free_old_rq_tio(tio);
-			return NULL;
-		}
-	}
-	dm_put_live_table(md, srcu_idx);
-
-	return tio;
-}
-
-/*
- * Called with the queue lock held.
- */
-static int dm_old_prep_fn(struct request_queue *q, struct request *rq)
-{
-	struct mapped_device *md = q->queuedata;
-	struct dm_rq_target_io *tio;
-
-	if (unlikely(rq->special)) {
-		DMWARN("Already has something in rq->special.");
-		return BLKPREP_KILL;
-	}
-
-	tio = dm_old_prep_tio(rq, md, GFP_ATOMIC);
-	if (!tio)
-		return BLKPREP_DEFER;
-
-	rq->special = tio;
-	rq->rq_flags |= RQF_DONTPREP;
-
-	return BLKPREP_OK;
-}
-
 /*
  * Returns:
  * DM_MAPIO_*       : the request has been processed as indicated
@@ -617,31 +470,18 @@ static int map_request(struct dm_rq_target_io *tio)
 	struct request *rq = tio->orig;
 	struct request *clone = NULL;
 
-	if (tio->clone) {
-		clone = tio->clone;
-		r = ti->type->map_rq(ti, clone, &tio->info);
-		if (r == DM_MAPIO_DELAY_REQUEUE)
-			return DM_MAPIO_REQUEUE; /* .request_fn requeue is always immediate */
-	} else {
-		r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone);
-		if (r < 0) {
-			/* The target wants to complete the I/O */
-			dm_kill_unmapped_request(rq, r);
-			return r;
-		}
-		if (r == DM_MAPIO_REMAPPED &&
-		    setup_clone(clone, rq, tio, GFP_ATOMIC)) {
-			/* -ENOMEM */
-			ti->type->release_clone_rq(clone);
-			return DM_MAPIO_REQUEUE;
-		}
-	}
-
+	r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone);
 	switch (r) {
 	case DM_MAPIO_SUBMITTED:
 		/* The target has taken the I/O to submit by itself later */
 		break;
 	case DM_MAPIO_REMAPPED:
+		if (setup_clone(clone, rq, tio, GFP_ATOMIC)) {
+			/* -ENOMEM */
+			ti->type->release_clone_rq(clone);
+			return DM_MAPIO_REQUEUE;
+		}
+
 		/* The target has remapped the I/O so dispatch it */
 		trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
 				     blk_rq_pos(rq));
@@ -700,6 +540,29 @@ static void dm_start_request(struct mapped_device *md, struct request *orig)
 	dm_get(md);
 }
 
+static int __dm_rq_init_rq(struct mapped_device *md, struct request *rq)
+{
+	struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
+
+	/*
+	 * Must initialize md member of tio, otherwise it won't
+	 * be available in dm_mq_queue_rq.
+	 */
+	tio->md = md;
+
+	if (md->init_tio_pdu) {
+		/* target-specific per-io data is immediately after the tio */
+		tio->info.ptr = tio + 1;
+	}
+
+	return 0;
+}
+
+static int dm_rq_init_rq(struct request_queue *q, struct request *rq, gfp_t gfp)
+{
+	return __dm_rq_init_rq(q->rq_alloc_data, rq);
+}
+
 static void map_tio_request(struct kthread_work *work)
 {
 	struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work);
@@ -794,6 +657,7 @@ static void dm_old_request_fn(struct request_queue *q)
 		dm_start_request(md, rq);
 
 		tio = tio_from_request(rq);
+		init_tio(tio, rq, md);
 		/* Establish tio->ti before queuing work (map_tio_request) */
 		tio->ti = ti;
 		kthread_queue_work(&md->kworker, &tio->work);
@@ -804,10 +668,22 @@ static void dm_old_request_fn(struct request_queue *q)
 /*
  * Fully initialize a .request_fn request-based queue.
  */
-int dm_old_init_request_queue(struct mapped_device *md)
+int dm_old_init_request_queue(struct mapped_device *md, struct dm_table *t)
 {
+	struct dm_target *immutable_tgt;
+
 	/* Fully initialize the queue */
+	md->queue->cmd_size = sizeof(struct dm_rq_target_io);
+	md->queue->rq_alloc_data = md;
 	md->queue->request_fn = dm_old_request_fn;
+	md->queue->init_rq_fn = dm_rq_init_rq;
+
+	immutable_tgt = dm_table_get_immutable_target(t);
+	if (immutable_tgt && immutable_tgt->per_io_data_size) {
+		/* any target-specific per-io data is immediately after the tio */
+		md->queue->cmd_size += immutable_tgt->per_io_data_size;
+		md->init_tio_pdu = true;
+	}
 	if (blk_init_allocated_queue(md->queue) < 0)
 		return -EINVAL;
 
@@ -816,7 +692,6 @@ int dm_old_init_request_queue(struct mapped_device *md)
 
 	dm_init_normal_md_queue(md);
 	blk_queue_softirq_done(md->queue, dm_softirq_done);
-	blk_queue_prep_rq(md->queue, dm_old_prep_fn);
 
 	/* Initialize the request-based DM worker thread */
 	kthread_init_worker(&md->kworker);
@@ -837,21 +712,7 @@ static int dm_mq_init_request(void *data, struct request *rq,
 		       unsigned int hctx_idx, unsigned int request_idx,
 		       unsigned int numa_node)
 {
-	struct mapped_device *md = data;
-	struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq);
-
-	/*
-	 * Must initialize md member of tio, otherwise it won't
-	 * be available in dm_mq_queue_rq.
-	 */
-	tio->md = md;
-
-	if (md->init_tio_pdu) {
-		/* target-specific per-io data is immediately after the tio */
-		tio->info.ptr = tio + 1;
-	}
-
-	return 0;
+	return __dm_rq_init_rq(data, rq);
 }
 
 static int dm_mq_queue_rq(struct blk_mq_hw_ctx *hctx,