[BLOCK] reimplement handling of barrier request

Reimplement handling of barrier requests.

* Flexible handling to deal with various capabilities of
  target devices.
* Retry support for falling back.
* Tagged queues which don't support ordered tag can do ordered.

Signed-off-by: Tejun Heo <htejun@gmail.com>
Signed-off-by: Jens Axboe <axboe@suse.de>

block/elevator.c

@@ -304,15 +304,7 @@ void elv_requeue_request(request_queue_t *q, struct request *rq)
 
 	rq->flags &= ~REQ_STARTED;
 
-	/*
+	__elv_add_request(q, rq, ELEVATOR_INSERT_REQUEUE, 0);
-	 * if this is the flush, requeue the original instead and drop the flush
-	 */
-	if (rq->flags & REQ_BAR_FLUSH) {
-		clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags);
-		rq = rq->end_io_data;
-	}
-
-	__elv_add_request(q, rq, ELEVATOR_INSERT_FRONT, 0);
 }
 
 static void elv_drain_elevator(request_queue_t *q)
@@ -332,7 +324,18 @@ static void elv_drain_elevator(request_queue_t *q)
 void __elv_add_request(request_queue_t *q, struct request *rq, int where,
 		       int plug)
 {
+	struct list_head *pos;
+	unsigned ordseq;
+
+	if (q->ordcolor)
+		rq->flags |= REQ_ORDERED_COLOR;
+
 	if (rq->flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) {
+		/*
+		 * toggle ordered color
+		 */
+		q->ordcolor ^= 1;
+
 		/*
 		 * barriers implicitly indicate back insertion
 		 */
@@ -393,6 +396,30 @@ void __elv_add_request(request_queue_t *q, struct request *rq, int where,
 		q->elevator->ops->elevator_add_req_fn(q, rq);
 		break;
 
+	case ELEVATOR_INSERT_REQUEUE:
+		/*
+		 * If ordered flush isn't in progress, we do front
+		 * insertion; otherwise, requests should be requeued
+		 * in ordseq order.
+		 */
+		rq->flags |= REQ_SOFTBARRIER;
+
+		if (q->ordseq == 0) {
+			list_add(&rq->queuelist, &q->queue_head);
+			break;
+		}
+
+		ordseq = blk_ordered_req_seq(rq);
+
+		list_for_each(pos, &q->queue_head) {
+			struct request *pos_rq = list_entry_rq(pos);
+			if (ordseq <= blk_ordered_req_seq(pos_rq))
+				break;
+		}
+
+		list_add_tail(&rq->queuelist, pos);
+		break;
+
 	default:
 		printk(KERN_ERR "%s: bad insertion point %d\n",
 		       __FUNCTION__, where);
@@ -422,25 +449,16 @@ static inline struct request *__elv_next_request(request_queue_t *q)
 {
 	struct request *rq;
 
-	if (unlikely(list_empty(&q->queue_head) &&
+	while (1) {
-		     !q->elevator->ops->elevator_dispatch_fn(q, 0)))
+		while (!list_empty(&q->queue_head)) {
-		return NULL;
+			rq = list_entry_rq(q->queue_head.next);
+			if (blk_do_ordered(q, &rq))
+				return rq;
+		}
 
-	rq = list_entry_rq(q->queue_head.next);
+		if (!q->elevator->ops->elevator_dispatch_fn(q, 0))
-
+			return NULL;
-	/*
-	 * if this is a barrier write and the device has to issue a
-	 * flush sequence to support it, check how far we are
-	 */
-	if (blk_fs_request(rq) && blk_barrier_rq(rq)) {
-		BUG_ON(q->ordered == QUEUE_ORDERED_NONE);
-
-		if (q->ordered == QUEUE_ORDERED_FLUSH &&
-		    !blk_barrier_preflush(rq))
-			rq = blk_start_pre_flush(q, rq);
 	}
-
-	return rq;
 }
 
 struct request *elv_next_request(request_queue_t *q)
@@ -593,7 +611,21 @@ void elv_completed_request(request_queue_t *q, struct request *rq)
 	 * request is released from the driver, io must be done
 	 */
 	if (blk_account_rq(rq)) {
+		struct request *first_rq = list_entry_rq(q->queue_head.next);
+
 		q->in_flight--;
+
+		/*
+		 * Check if the queue is waiting for fs requests to be
+		 * drained for flush sequence.
+		 */
+		if (q->ordseq && q->in_flight == 0 &&
+		    blk_ordered_cur_seq(q) == QUEUE_ORDSEQ_DRAIN &&
+		    blk_ordered_req_seq(first_rq) > QUEUE_ORDSEQ_DRAIN) {
+			blk_ordered_complete_seq(q, QUEUE_ORDSEQ_DRAIN, 0);
+			q->request_fn(q);
+		}
+
 		if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn)
 			e->ops->elevator_completed_req_fn(q, rq);
 	}

block/ll_rw_blk.c

@@ -290,8 +290,8 @@ static inline void rq_init(request_queue_t *q, struct request *rq)
 
 /**
  * blk_queue_ordered - does this queue support ordered writes
- * @q:     the request queue
+ * @q:        the request queue
- * @flag:  see below
+ * @ordered:  one of QUEUE_ORDERED_*
  *
  * Description:
  *   For journalled file systems, doing ordered writes on a commit
@@ -300,28 +300,30 @@ static inline void rq_init(request_queue_t *q, struct request *rq)
  *   feature should call this function and indicate so.
  *
  **/
-void blk_queue_ordered(request_queue_t *q, int flag)
+int blk_queue_ordered(request_queue_t *q, unsigned ordered,
+		      prepare_flush_fn *prepare_flush_fn)
 {
-	switch (flag) {
+	if (ordered & (QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_POSTFLUSH) &&
-		case QUEUE_ORDERED_NONE:
+	    prepare_flush_fn == NULL) {
-			if (q->flush_rq)
+		printk(KERN_ERR "blk_queue_ordered: prepare_flush_fn required\n");
-				kmem_cache_free(request_cachep, q->flush_rq);
+		return -EINVAL;
-			q->flush_rq = NULL;
-			q->ordered = flag;
-			break;
-		case QUEUE_ORDERED_TAG:
-			q->ordered = flag;
-			break;
-		case QUEUE_ORDERED_FLUSH:
-			q->ordered = flag;
-			if (!q->flush_rq)
-				q->flush_rq = kmem_cache_alloc(request_cachep,
-								GFP_KERNEL);
-			break;
-		default:
-			printk("blk_queue_ordered: bad value %d\n", flag);
-			break;
 	}
+
+	if (ordered != QUEUE_ORDERED_NONE &&
+	    ordered != QUEUE_ORDERED_DRAIN &&
+	    ordered != QUEUE_ORDERED_DRAIN_FLUSH &&
+	    ordered != QUEUE_ORDERED_DRAIN_FUA &&
+	    ordered != QUEUE_ORDERED_TAG &&
+	    ordered != QUEUE_ORDERED_TAG_FLUSH &&
+	    ordered != QUEUE_ORDERED_TAG_FUA) {
+		printk(KERN_ERR "blk_queue_ordered: bad value %d\n", ordered);
+		return -EINVAL;
+	}
+
+	q->next_ordered = ordered;
+	q->prepare_flush_fn = prepare_flush_fn;
+
+	return 0;
 }
 
 EXPORT_SYMBOL(blk_queue_ordered);
@@ -346,167 +348,265 @@ EXPORT_SYMBOL(blk_queue_issue_flush_fn);
 /*
  * Cache flushing for ordered writes handling
  */
-static void blk_pre_flush_end_io(struct request *flush_rq, int error)
+inline unsigned blk_ordered_cur_seq(request_queue_t *q)
+{
+	if (!q->ordseq)
+		return 0;
+	return 1 << ffz(q->ordseq);
+}
+
+unsigned blk_ordered_req_seq(struct request *rq)
 {
-	struct request *rq = flush_rq->end_io_data;
 	request_queue_t *q = rq->q;
 
-	elv_completed_request(q, flush_rq);
+	BUG_ON(q->ordseq == 0);
 
-	rq->flags |= REQ_BAR_PREFLUSH;
+	if (rq == &q->pre_flush_rq)
+		return QUEUE_ORDSEQ_PREFLUSH;
+	if (rq == &q->bar_rq)
+		return QUEUE_ORDSEQ_BAR;
+	if (rq == &q->post_flush_rq)
+		return QUEUE_ORDSEQ_POSTFLUSH;
 
-	if (!flush_rq->errors)
+	if ((rq->flags & REQ_ORDERED_COLOR) ==
-		elv_requeue_request(q, rq);
+	    (q->orig_bar_rq->flags & REQ_ORDERED_COLOR))
-	else {
+		return QUEUE_ORDSEQ_DRAIN;
-		q->end_flush_fn(q, flush_rq);
+	else
-		clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags);
+		return QUEUE_ORDSEQ_DONE;
-		q->request_fn(q);
-	}
 }
 
-static void blk_post_flush_end_io(struct request *flush_rq, int error)
+void blk_ordered_complete_seq(request_queue_t *q, unsigned seq, int error)
 {
-	struct request *rq = flush_rq->end_io_data;
+	struct request *rq;
-	request_queue_t *q = rq->q;
+	int uptodate;
 
-	elv_completed_request(q, flush_rq);
+	if (error && !q->orderr)
+		q->orderr = error;
 
-	rq->flags |= REQ_BAR_POSTFLUSH;
+	BUG_ON(q->ordseq & seq);
+	q->ordseq |= seq;
 
-	q->end_flush_fn(q, flush_rq);
+	if (blk_ordered_cur_seq(q) != QUEUE_ORDSEQ_DONE)
-	clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags);
+		return;
-	q->request_fn(q);
-}
-
-struct request *blk_start_pre_flush(request_queue_t *q, struct request *rq)
-{
-	struct request *flush_rq = q->flush_rq;
-
-	BUG_ON(!blk_barrier_rq(rq));
-
-	if (test_and_set_bit(QUEUE_FLAG_FLUSH, &q->queue_flags))
-		return NULL;
-
-	rq_init(q, flush_rq);
-	flush_rq->elevator_private = NULL;
-	flush_rq->flags = REQ_BAR_FLUSH;
-	flush_rq->rq_disk = rq->rq_disk;
-	flush_rq->rl = NULL;
 
 	/*
-	 * prepare_flush returns 0 if no flush is needed, just mark both
+	 * Okay, sequence complete.
-	 * pre and post flush as done in that case
 	 */
-	if (!q->prepare_flush_fn(q, flush_rq)) {
+	rq = q->orig_bar_rq;
-		rq->flags |= REQ_BAR_PREFLUSH | REQ_BAR_POSTFLUSH;
+	uptodate = q->orderr ? q->orderr : 1;
-		clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags);
+
-		return rq;
+	q->ordseq = 0;
+
+	end_that_request_first(rq, uptodate, rq->hard_nr_sectors);
+	end_that_request_last(rq, uptodate);
+}
+
+static void pre_flush_end_io(struct request *rq, int error)
+{
+	elv_completed_request(rq->q, rq);
+	blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_PREFLUSH, error);
+}
+
+static void bar_end_io(struct request *rq, int error)
+{
+	elv_completed_request(rq->q, rq);
+	blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_BAR, error);
+}
+
+static void post_flush_end_io(struct request *rq, int error)
+{
+	elv_completed_request(rq->q, rq);
+	blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_POSTFLUSH, error);
+}
+
+static void queue_flush(request_queue_t *q, unsigned which)
+{
+	struct request *rq;
+	rq_end_io_fn *end_io;
+
+	if (which == QUEUE_ORDERED_PREFLUSH) {
+		rq = &q->pre_flush_rq;
+		end_io = pre_flush_end_io;
+	} else {
+		rq = &q->post_flush_rq;
+		end_io = post_flush_end_io;
 	}
 
+	rq_init(q, rq);
+	rq->flags = REQ_HARDBARRIER;
+	rq->elevator_private = NULL;
+	rq->rq_disk = q->bar_rq.rq_disk;
+	rq->rl = NULL;
+	rq->end_io = end_io;
+	q->prepare_flush_fn(q, rq);
+
+	__elv_add_request(q, rq, ELEVATOR_INSERT_FRONT, 0);
+}
+
+static inline struct request *start_ordered(request_queue_t *q,
+					    struct request *rq)
+{
+	q->bi_size = 0;
+	q->orderr = 0;
+	q->ordered = q->next_ordered;
+	q->ordseq |= QUEUE_ORDSEQ_STARTED;
+
 	/*
-	 * some drivers dequeue requests right away, some only after io
+	 * Prep proxy barrier request.
-	 * completion. make sure the request is dequeued.
 	 */
-	if (!list_empty(&rq->queuelist))
+	blkdev_dequeue_request(rq);
-		blkdev_dequeue_request(rq);
+	q->orig_bar_rq = rq;
+	rq = &q->bar_rq;
+	rq_init(q, rq);
+	rq->flags = bio_data_dir(q->orig_bar_rq->bio);
+	rq->flags |= q->ordered & QUEUE_ORDERED_FUA ? REQ_FUA : 0;
+	rq->elevator_private = NULL;
+	rq->rl = NULL;
+	init_request_from_bio(rq, q->orig_bar_rq->bio);
+	rq->end_io = bar_end_io;
 
-	flush_rq->end_io_data = rq;
+	/*
-	flush_rq->end_io = blk_pre_flush_end_io;
+	 * Queue ordered sequence.  As we stack them at the head, we
+	 * need to queue in reverse order.  Note that we rely on that
+	 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
+	 * request gets inbetween ordered sequence.
+	 */
+	if (q->ordered & QUEUE_ORDERED_POSTFLUSH)
+		queue_flush(q, QUEUE_ORDERED_POSTFLUSH);
+	else
+		q->ordseq |= QUEUE_ORDSEQ_POSTFLUSH;
 
-	__elv_add_request(q, flush_rq, ELEVATOR_INSERT_FRONT, 0);
+	__elv_add_request(q, rq, ELEVATOR_INSERT_FRONT, 0);
-	return flush_rq;
+
+	if (q->ordered & QUEUE_ORDERED_PREFLUSH) {
+		queue_flush(q, QUEUE_ORDERED_PREFLUSH);
+		rq = &q->pre_flush_rq;
+	} else
+		q->ordseq |= QUEUE_ORDSEQ_PREFLUSH;
+
+	if ((q->ordered & QUEUE_ORDERED_TAG) || q->in_flight == 0)
+		q->ordseq |= QUEUE_ORDSEQ_DRAIN;
+	else
+		rq = NULL;
+
+	return rq;
 }
 
-static void blk_start_post_flush(request_queue_t *q, struct request *rq)
+int blk_do_ordered(request_queue_t *q, struct request **rqp)
 {
-	struct request *flush_rq = q->flush_rq;
+	struct request *rq = *rqp, *allowed_rq;
+	int is_barrier = blk_fs_request(rq) && blk_barrier_rq(rq);
 
-	BUG_ON(!blk_barrier_rq(rq));
+	if (!q->ordseq) {
+		if (!is_barrier)
+			return 1;
 
-	rq_init(q, flush_rq);
+		if (q->next_ordered != QUEUE_ORDERED_NONE) {
-	flush_rq->elevator_private = NULL;
+			*rqp = start_ordered(q, rq);
-	flush_rq->flags = REQ_BAR_FLUSH;
+			return 1;
-	flush_rq->rq_disk = rq->rq_disk;
+		} else {
-	flush_rq->rl = NULL;
+			/*
-
+			 * This can happen when the queue switches to
-	if (q->prepare_flush_fn(q, flush_rq)) {
+			 * ORDERED_NONE while this request is on it.
-		flush_rq->end_io_data = rq;
+			 */
-		flush_rq->end_io = blk_post_flush_end_io;
+			blkdev_dequeue_request(rq);
-
+			end_that_request_first(rq, -EOPNOTSUPP,
-		__elv_add_request(q, flush_rq, ELEVATOR_INSERT_FRONT, 0);
+					       rq->hard_nr_sectors);
-		q->request_fn(q);
+			end_that_request_last(rq, -EOPNOTSUPP);
+			*rqp = NULL;
+			return 0;
+		}
 	}
-}
 
-static inline int blk_check_end_barrier(request_queue_t *q, struct request *rq,
+	if (q->ordered & QUEUE_ORDERED_TAG) {
-					int sectors)
+		if (is_barrier && rq != &q->bar_rq)
-{
+			*rqp = NULL;
-	if (sectors > rq->nr_sectors)
+		return 1;
-		sectors = rq->nr_sectors;
-
-	rq->nr_sectors -= sectors;
-	return rq->nr_sectors;
-}
-
-static int __blk_complete_barrier_rq(request_queue_t *q, struct request *rq,
-				     int sectors, int queue_locked)
-{
-	if (q->ordered != QUEUE_ORDERED_FLUSH)
-		return 0;
-	if (!blk_fs_request(rq) || !blk_barrier_rq(rq))
-		return 0;
-	if (blk_barrier_postflush(rq))
-		return 0;
-
-	if (!blk_check_end_barrier(q, rq, sectors)) {
-		unsigned long flags = 0;
-
-		if (!queue_locked)
-			spin_lock_irqsave(q->queue_lock, flags);
-
-		blk_start_post_flush(q, rq);
-
-		if (!queue_locked)
-			spin_unlock_irqrestore(q->queue_lock, flags);
 	}
 
+	switch (blk_ordered_cur_seq(q)) {
+	case QUEUE_ORDSEQ_PREFLUSH:
+		allowed_rq = &q->pre_flush_rq;
+		break;
+	case QUEUE_ORDSEQ_BAR:
+		allowed_rq = &q->bar_rq;
+		break;
+	case QUEUE_ORDSEQ_POSTFLUSH:
+		allowed_rq = &q->post_flush_rq;
+		break;
+	default:
+		allowed_rq = NULL;
+		break;
+	}
+
+	if (rq != allowed_rq &&
+	    (blk_fs_request(rq) || rq == &q->pre_flush_rq ||
+	     rq == &q->post_flush_rq))
+		*rqp = NULL;
+
 	return 1;
 }
 
-/**
+static int flush_dry_bio_endio(struct bio *bio, unsigned int bytes, int error)
- * blk_complete_barrier_rq - complete possible barrier request
- * @q:  the request queue for the device
- * @rq:  the request
- * @sectors:  number of sectors to complete
- *
- * Description:
- *   Used in driver end_io handling to determine whether to postpone
- *   completion of a barrier request until a post flush has been done. This
- *   is the unlocked variant, used if the caller doesn't already hold the
- *   queue lock.
- **/
-int blk_complete_barrier_rq(request_queue_t *q, struct request *rq, int sectors)
 {
-	return __blk_complete_barrier_rq(q, rq, sectors, 0);
+	request_queue_t *q = bio->bi_private;
-}
+	struct bio_vec *bvec;
-EXPORT_SYMBOL(blk_complete_barrier_rq);
+	int i;
 
-/**
+	/*
- * blk_complete_barrier_rq_locked - complete possible barrier request
+	 * This is dry run, restore bio_sector and size.  We'll finish
- * @q:  the request queue for the device
+	 * this request again with the original bi_end_io after an
- * @rq:  the request
+	 * error occurs or post flush is complete.
- * @sectors:  number of sectors to complete
+	 */
- *
+	q->bi_size += bytes;
- * Description:
+
- *   See blk_complete_barrier_rq(). This variant must be used if the caller
+	if (bio->bi_size)
- *   holds the queue lock.
+		return 1;
- **/
+
-int blk_complete_barrier_rq_locked(request_queue_t *q, struct request *rq,
+	/* Rewind bvec's */
-				   int sectors)
+	bio->bi_idx = 0;
-{
+	bio_for_each_segment(bvec, bio, i) {
-	return __blk_complete_barrier_rq(q, rq, sectors, 1);
+		bvec->bv_len += bvec->bv_offset;
+		bvec->bv_offset = 0;
+	}
+
+	/* Reset bio */
+	set_bit(BIO_UPTODATE, &bio->bi_flags);
+	bio->bi_size = q->bi_size;
+	bio->bi_sector -= (q->bi_size >> 9);
+	q->bi_size = 0;
+
+	return 0;
+}
+
+static inline int ordered_bio_endio(struct request *rq, struct bio *bio,
+				    unsigned int nbytes, int error)
+{
+	request_queue_t *q = rq->q;
+	bio_end_io_t *endio;
+	void *private;
+
+	if (&q->bar_rq != rq)
+		return 0;
+
+	/*
+	 * Okay, this is the barrier request in progress, dry finish it.
+	 */
+	if (error && !q->orderr)
+		q->orderr = error;
+
+	endio = bio->bi_end_io;
+	private = bio->bi_private;
+	bio->bi_end_io = flush_dry_bio_endio;
+	bio->bi_private = q;
+
+	bio_endio(bio, nbytes, error);
+
+	bio->bi_end_io = endio;
+	bio->bi_private = private;
+
+	return 1;
 }
-EXPORT_SYMBOL(blk_complete_barrier_rq_locked);
 
 /**
  * blk_queue_bounce_limit - set bounce buffer limit for queue
@@ -1047,6 +1147,7 @@ static const char * const rq_flags[] = {
 	"REQ_SORTED",
 	"REQ_SOFTBARRIER",
 	"REQ_HARDBARRIER",
+	"REQ_FUA",
 	"REQ_CMD",
 	"REQ_NOMERGE",
 	"REQ_STARTED",
@@ -1066,6 +1167,7 @@ static const char * const rq_flags[] = {
 	"REQ_PM_SUSPEND",
 	"REQ_PM_RESUME",
 	"REQ_PM_SHUTDOWN",
+	"REQ_ORDERED_COLOR",
 };
 
 void blk_dump_rq_flags(struct request *rq, char *msg)
@@ -1643,8 +1745,6 @@ void blk_cleanup_queue(request_queue_t * q)
 	if (q->queue_tags)
 		__blk_queue_free_tags(q);
 
-	blk_queue_ordered(q, QUEUE_ORDERED_NONE);
-
 	kmem_cache_free(requestq_cachep, q);
 }
 
@@ -2714,7 +2814,7 @@ static int __make_request(request_queue_t *q, struct bio *bio)
 	spin_lock_prefetch(q->queue_lock);
 
 	barrier = bio_barrier(bio);
-	if (unlikely(barrier) && (q->ordered == QUEUE_ORDERED_NONE)) {
+	if (unlikely(barrier) && (q->next_ordered == QUEUE_ORDERED_NONE)) {
 		err = -EOPNOTSUPP;
 		goto end_io;
 	}
@@ -3075,7 +3175,8 @@ static int __end_that_request_first(struct request *req, int uptodate,
 		if (nr_bytes >= bio->bi_size) {
 			req->bio = bio->bi_next;
 			nbytes = bio->bi_size;
-			bio_endio(bio, nbytes, error);
+			if (!ordered_bio_endio(req, bio, nbytes, error))
+				bio_endio(bio, nbytes, error);
 			next_idx = 0;
 			bio_nbytes = 0;
 		} else {
@@ -3130,7 +3231,8 @@ static int __end_that_request_first(struct request *req, int uptodate,
 	 * if the request wasn't completed, update state
 	 */
 	if (bio_nbytes) {
-		bio_endio(bio, bio_nbytes, error);
+		if (!ordered_bio_endio(req, bio, bio_nbytes, error))
+			bio_endio(bio, bio_nbytes, error);
 		bio->bi_idx += next_idx;
 		bio_iovec(bio)->bv_offset += nr_bytes;
 		bio_iovec(bio)->bv_len -= nr_bytes;

include/linux/blkdev.h

@@ -207,6 +207,7 @@ enum rq_flag_bits {
 	__REQ_SORTED,		/* elevator knows about this request */
 	__REQ_SOFTBARRIER,	/* may not be passed by ioscheduler */
 	__REQ_HARDBARRIER,	/* may not be passed by drive either */
+	__REQ_FUA,		/* forced unit access */
 	__REQ_CMD,		/* is a regular fs rw request */
 	__REQ_NOMERGE,		/* don't touch this for merging */
 	__REQ_STARTED,		/* drive already may have started this one */
@@ -230,9 +231,7 @@ enum rq_flag_bits {
 	__REQ_PM_SUSPEND,	/* suspend request */
 	__REQ_PM_RESUME,	/* resume request */
 	__REQ_PM_SHUTDOWN,	/* shutdown request */
-	__REQ_BAR_PREFLUSH,	/* barrier pre-flush done */
+	__REQ_ORDERED_COLOR,	/* is before or after barrier */
-	__REQ_BAR_POSTFLUSH,	/* barrier post-flush */
-	__REQ_BAR_FLUSH,	/* rq is the flush request */
 	__REQ_NR_BITS,		/* stops here */
 };
 
@@ -241,6 +240,7 @@ enum rq_flag_bits {
 #define REQ_SORTED	(1 << __REQ_SORTED)
 #define REQ_SOFTBARRIER	(1 << __REQ_SOFTBARRIER)
 #define REQ_HARDBARRIER	(1 << __REQ_HARDBARRIER)
+#define REQ_FUA		(1 << __REQ_FUA)
 #define REQ_CMD		(1 << __REQ_CMD)
 #define REQ_NOMERGE	(1 << __REQ_NOMERGE)
 #define REQ_STARTED	(1 << __REQ_STARTED)
@@ -260,9 +260,7 @@ enum rq_flag_bits {
 #define REQ_PM_SUSPEND	(1 << __REQ_PM_SUSPEND)
 #define REQ_PM_RESUME	(1 << __REQ_PM_RESUME)
 #define REQ_PM_SHUTDOWN	(1 << __REQ_PM_SHUTDOWN)
-#define REQ_BAR_PREFLUSH	(1 << __REQ_BAR_PREFLUSH)
+#define REQ_ORDERED_COLOR	(1 << __REQ_ORDERED_COLOR)
-#define REQ_BAR_POSTFLUSH	(1 << __REQ_BAR_POSTFLUSH)
-#define REQ_BAR_FLUSH	(1 << __REQ_BAR_FLUSH)
 
 /*
  * State information carried for REQ_PM_SUSPEND and REQ_PM_RESUME
@@ -292,8 +290,7 @@ struct bio_vec;
 typedef int (merge_bvec_fn) (request_queue_t *, struct bio *, struct bio_vec *);
 typedef void (activity_fn) (void *data, int rw);
 typedef int (issue_flush_fn) (request_queue_t *, struct gendisk *, sector_t *);
-typedef int (prepare_flush_fn) (request_queue_t *, struct request *);
+typedef void (prepare_flush_fn) (request_queue_t *, struct request *);
-typedef void (end_flush_fn) (request_queue_t *, struct request *);
 
 enum blk_queue_state {
 	Queue_down,
@@ -335,7 +332,6 @@ struct request_queue
 	activity_fn		*activity_fn;
 	issue_flush_fn		*issue_flush_fn;
 	prepare_flush_fn	*prepare_flush_fn;
-	end_flush_fn		*end_flush_fn;
 
 	/*
 	 * Dispatch queue sorting
@@ -420,14 +416,11 @@ struct request_queue
 	/*
 	 * reserved for flush operations
 	 */
-	struct request		*flush_rq;
+	unsigned int		ordered, next_ordered, ordseq;
-	unsigned char		ordered;
+	int			orderr, ordcolor;
-};
+	struct request		pre_flush_rq, bar_rq, post_flush_rq;
-
+	struct request		*orig_bar_rq;
-enum {
+	unsigned int		bi_size;
-	QUEUE_ORDERED_NONE,
-	QUEUE_ORDERED_TAG,
-	QUEUE_ORDERED_FLUSH,
 };
 
 #define RQ_INACTIVE		(-1)
@@ -445,12 +438,51 @@ enum {
 #define QUEUE_FLAG_REENTER	6	/* Re-entrancy avoidance */
 #define QUEUE_FLAG_PLUGGED	7	/* queue is plugged */
 #define QUEUE_FLAG_ELVSWITCH	8	/* don't use elevator, just do FIFO */
-#define QUEUE_FLAG_FLUSH	9	/* doing barrier flush sequence */
+
+enum {
+	/*
+	 * Hardbarrier is supported with one of the following methods.
+	 *
+	 * NONE		: hardbarrier unsupported
+	 * DRAIN	: ordering by draining is enough
+	 * DRAIN_FLUSH	: ordering by draining w/ pre and post flushes
+	 * DRAIN_FUA	: ordering by draining w/ pre flush and FUA write
+	 * TAG		: ordering by tag is enough
+	 * TAG_FLUSH	: ordering by tag w/ pre and post flushes
+	 * TAG_FUA	: ordering by tag w/ pre flush and FUA write
+	 */
+	QUEUE_ORDERED_NONE	= 0x00,
+	QUEUE_ORDERED_DRAIN	= 0x01,
+	QUEUE_ORDERED_TAG	= 0x02,
+
+	QUEUE_ORDERED_PREFLUSH	= 0x10,
+	QUEUE_ORDERED_POSTFLUSH	= 0x20,
+	QUEUE_ORDERED_FUA	= 0x40,
+
+	QUEUE_ORDERED_DRAIN_FLUSH = QUEUE_ORDERED_DRAIN |
+			QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_POSTFLUSH,
+	QUEUE_ORDERED_DRAIN_FUA	= QUEUE_ORDERED_DRAIN |
+			QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_FUA,
+	QUEUE_ORDERED_TAG_FLUSH	= QUEUE_ORDERED_TAG |
+			QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_POSTFLUSH,
+	QUEUE_ORDERED_TAG_FUA	= QUEUE_ORDERED_TAG |
+			QUEUE_ORDERED_PREFLUSH | QUEUE_ORDERED_FUA,
+
+	/*
+	 * Ordered operation sequence
+	 */
+	QUEUE_ORDSEQ_STARTED	= 0x01,	/* flushing in progress */
+	QUEUE_ORDSEQ_DRAIN	= 0x02,	/* waiting for the queue to be drained */
+	QUEUE_ORDSEQ_PREFLUSH	= 0x04,	/* pre-flushing in progress */
+	QUEUE_ORDSEQ_BAR	= 0x08,	/* original barrier req in progress */
+	QUEUE_ORDSEQ_POSTFLUSH	= 0x10,	/* post-flushing in progress */
+	QUEUE_ORDSEQ_DONE	= 0x20,
+};
 
 #define blk_queue_plugged(q)	test_bit(QUEUE_FLAG_PLUGGED, &(q)->queue_flags)
 #define blk_queue_tagged(q)	test_bit(QUEUE_FLAG_QUEUED, &(q)->queue_flags)
 #define blk_queue_stopped(q)	test_bit(QUEUE_FLAG_STOPPED, &(q)->queue_flags)
-#define blk_queue_flushing(q)	test_bit(QUEUE_FLAG_FLUSH, &(q)->queue_flags)
+#define blk_queue_flushing(q)	((q)->ordseq)
 
 #define blk_fs_request(rq)	((rq)->flags & REQ_CMD)
 #define blk_pc_request(rq)	((rq)->flags & REQ_BLOCK_PC)
@@ -466,8 +498,7 @@ enum {
 
 #define blk_sorted_rq(rq)	((rq)->flags & REQ_SORTED)
 #define blk_barrier_rq(rq)	((rq)->flags & REQ_HARDBARRIER)
-#define blk_barrier_preflush(rq)	((rq)->flags & REQ_BAR_PREFLUSH)
+#define blk_fua_rq(rq)		((rq)->flags & REQ_FUA)
-#define blk_barrier_postflush(rq)	((rq)->flags & REQ_BAR_POSTFLUSH)
 
 #define list_entry_rq(ptr)	list_entry((ptr), struct request, queuelist)
 
@@ -665,11 +696,12 @@ extern void blk_queue_prep_rq(request_queue_t *, prep_rq_fn *pfn);
 extern void blk_queue_merge_bvec(request_queue_t *, merge_bvec_fn *);
 extern void blk_queue_dma_alignment(request_queue_t *, int);
 extern struct backing_dev_info *blk_get_backing_dev_info(struct block_device *bdev);
-extern void blk_queue_ordered(request_queue_t *, int);
+extern int blk_queue_ordered(request_queue_t *, unsigned, prepare_flush_fn *);
 extern void blk_queue_issue_flush_fn(request_queue_t *, issue_flush_fn *);
-extern struct request *blk_start_pre_flush(request_queue_t *,struct request *);
+extern int blk_do_ordered(request_queue_t *, struct request **);
-extern int blk_complete_barrier_rq(request_queue_t *, struct request *, int);
+extern unsigned blk_ordered_cur_seq(request_queue_t *);
-extern int blk_complete_barrier_rq_locked(request_queue_t *, struct request *, int);
+extern unsigned blk_ordered_req_seq(struct request *);
+extern void blk_ordered_complete_seq(request_queue_t *, unsigned, int);
 
 extern int blk_rq_map_sg(request_queue_t *, struct request *, struct scatterlist *);
 extern void blk_dump_rq_flags(struct request *, char *);

include/linux/elevator.h

@@ -130,6 +130,7 @@ extern int elv_try_last_merge(request_queue_t *, struct bio *);
 #define ELEVATOR_INSERT_FRONT	1
 #define ELEVATOR_INSERT_BACK	2
 #define ELEVATOR_INSERT_SORT	3
+#define ELEVATOR_INSERT_REQUEUE	4
 
 /*
  * return values from elevator_may_queue_fn