Merge tag 'for-4.12/dm-fixes-2' of...

Merge tag 'for-4.12/dm-fixes-2' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm

Pull device mapper fixes from Mike Snitzer:

 - a couple DM thin provisioning fixes

 - a few request-based DM and DM multipath fixes for issues that were
   made when merging Christoph's changes with Bart's changes for 4.12

 - a DM bufio unsigned overflow fix

 - a couple pure fixes for the DM cache target.

 - various very small tweaks to the DM cache target that enable
   considerable speed improvements in the face of continuous IO. Given
   that the cache target was significantly reworked for 4.12 I see no
   reason to sit on these advances until 4.13 considering the favorable
   results associated with such minimalist tweaks.

* tag 'for-4.12/dm-fixes-2' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
  dm cache: handle kmalloc failure allocating background_tracker struct
  dm bufio: make the parameter "retain_bytes" unsigned long
  dm mpath: multipath_clone_and_map must not return -EIO
  dm mpath: don't return -EIO from dm_report_EIO
  dm rq: add a missing break to map_request
  dm space map disk: fix some book keeping in the disk space map
  dm thin metadata: call precommit before saving the roots
  dm cache policy smq: don't do any writebacks unless IDLE
  dm cache: simplify the IDLE vs BUSY state calculation
  dm cache: track all IO to the cache rather than just the origin device's IO
  dm cache policy smq: stop preemptively demoting blocks
  dm cache policy smq: put newly promoted entries at the top of the multiqueue
  dm cache policy smq: be more aggressive about triggering a writeback
  dm cache policy smq: only demote entries in bottom half of the clean multiqueue
  dm cache: fix incorrect 'idle_time' reset in IO tracker

drivers/md/dm-bufio.c

@@ -218,7 +218,7 @@ static DEFINE_SPINLOCK(param_spinlock);
  * Buffers are freed after this timeout
  */
 static unsigned dm_bufio_max_age = DM_BUFIO_DEFAULT_AGE_SECS;
-static unsigned dm_bufio_retain_bytes = DM_BUFIO_DEFAULT_RETAIN_BYTES;
+static unsigned long dm_bufio_retain_bytes = DM_BUFIO_DEFAULT_RETAIN_BYTES;
 
 static unsigned long dm_bufio_peak_allocated;
 static unsigned long dm_bufio_allocated_kmem_cache;
@@ -1558,10 +1558,10 @@ static bool __try_evict_buffer(struct dm_buffer *b, gfp_t gfp)
 	return true;
 }
 
-static unsigned get_retain_buffers(struct dm_bufio_client *c)
+static unsigned long get_retain_buffers(struct dm_bufio_client *c)
 {
-        unsigned retain_bytes = ACCESS_ONCE(dm_bufio_retain_bytes);
-        return retain_bytes / c->block_size;
+        unsigned long retain_bytes = ACCESS_ONCE(dm_bufio_retain_bytes);
+        return retain_bytes >> (c->sectors_per_block_bits + SECTOR_SHIFT);
 }
 
 static unsigned long __scan(struct dm_bufio_client *c, unsigned long nr_to_scan,
@@ -1571,7 +1571,7 @@ static unsigned long __scan(struct dm_bufio_client *c, unsigned long nr_to_scan,
 	struct dm_buffer *b, *tmp;
 	unsigned long freed = 0;
 	unsigned long count = nr_to_scan;
-	unsigned retain_target = get_retain_buffers(c);
+	unsigned long retain_target = get_retain_buffers(c);
 
 	for (l = 0; l < LIST_SIZE; l++) {
 		list_for_each_entry_safe_reverse(b, tmp, &c->lru[l], lru_list) {
@@ -1794,8 +1794,8 @@ static bool older_than(struct dm_buffer *b, unsigned long age_hz)
 static void __evict_old_buffers(struct dm_bufio_client *c, unsigned long age_hz)
 {
 	struct dm_buffer *b, *tmp;
-	unsigned retain_target = get_retain_buffers(c);
-	unsigned count;
+	unsigned long retain_target = get_retain_buffers(c);
+	unsigned long count;
 	LIST_HEAD(write_list);
 
 	dm_bufio_lock(c);
@@ -1955,7 +1955,7 @@ MODULE_PARM_DESC(max_cache_size_bytes, "Size of metadata cache");
 module_param_named(max_age_seconds, dm_bufio_max_age, uint, S_IRUGO | S_IWUSR);
 MODULE_PARM_DESC(max_age_seconds, "Max age of a buffer in seconds");
 
-module_param_named(retain_bytes, dm_bufio_retain_bytes, uint, S_IRUGO | S_IWUSR);
+module_param_named(retain_bytes, dm_bufio_retain_bytes, ulong, S_IRUGO | S_IWUSR);
 MODULE_PARM_DESC(retain_bytes, "Try to keep at least this many bytes cached in memory");
 
 module_param_named(peak_allocated_bytes, dm_bufio_peak_allocated, ulong, S_IRUGO | S_IWUSR);

drivers/md/dm-cache-background-tracker.c

@@ -33,6 +33,11 @@ struct background_tracker *btracker_create(unsigned max_work)
 {
 	struct background_tracker *b = kmalloc(sizeof(*b), GFP_KERNEL);
 
+	if (!b) {
+		DMERR("couldn't create background_tracker");
+		return NULL;
+	}
+
 	b->max_work = max_work;
 	atomic_set(&b->pending_promotes, 0);
 	atomic_set(&b->pending_writebacks, 0);

drivers/md/dm-cache-policy-smq.c

@@ -1120,8 +1120,6 @@ static bool clean_target_met(struct smq_policy *mq, bool idle)
 	 * Cache entries may not be populated.  So we cannot rely on the
 	 * size of the clean queue.
 	 */
-	unsigned nr_clean;
-
 	if (idle) {
 		/*
 		 * We'd like to clean everything.
@@ -1129,18 +1127,16 @@ static bool clean_target_met(struct smq_policy *mq, bool idle)
 		return q_size(&mq->dirty) == 0u;
 	}
 
-	nr_clean = from_cblock(mq->cache_size) - q_size(&mq->dirty);
-	return (nr_clean + btracker_nr_writebacks_queued(mq->bg_work)) >=
-		percent_to_target(mq, CLEAN_TARGET);
+	/*
+	 * If we're busy we don't worry about cleaning at all.
+	 */
+	return true;
 }
 
-static bool free_target_met(struct smq_policy *mq, bool idle)
+static bool free_target_met(struct smq_policy *mq)
 {
 	unsigned nr_free;
 
-	if (!idle)
-		return true;
-
 	nr_free = from_cblock(mq->cache_size) - mq->cache_alloc.nr_allocated;
 	return (nr_free + btracker_nr_demotions_queued(mq->bg_work)) >=
 		percent_to_target(mq, FREE_TARGET);
@@ -1190,9 +1186,9 @@ static void queue_demotion(struct smq_policy *mq)
 	if (unlikely(WARN_ON_ONCE(!mq->migrations_allowed)))
 		return;
 
-	e = q_peek(&mq->clean, mq->clean.nr_levels, true);
+	e = q_peek(&mq->clean, mq->clean.nr_levels / 2, true);
 	if (!e) {
-		if (!clean_target_met(mq, false))
+		if (!clean_target_met(mq, true))
 			queue_writeback(mq);
 		return;
 	}
@@ -1220,7 +1216,7 @@ static void queue_promotion(struct smq_policy *mq, dm_oblock_t oblock,
 		 * We always claim to be 'idle' to ensure some demotions happen
 		 * with continuous loads.
 		 */
-		if (!free_target_met(mq, true))
+		if (!free_target_met(mq))
 			queue_demotion(mq);
 		return;
 	}
@@ -1421,14 +1417,10 @@ static int smq_get_background_work(struct dm_cache_policy *p, bool idle,
 	spin_lock_irqsave(&mq->lock, flags);
 	r = btracker_issue(mq->bg_work, result);
 	if (r == -ENODATA) {
-		/* find some writeback work to do */
-		if (mq->migrations_allowed && !free_target_met(mq, idle))
-			queue_demotion(mq);
-
-		else if (!clean_target_met(mq, idle))
+		if (!clean_target_met(mq, idle)) {
 			queue_writeback(mq);
-
-		r = btracker_issue(mq->bg_work, result);
+			r = btracker_issue(mq->bg_work, result);
+		}
 	}
 	spin_unlock_irqrestore(&mq->lock, flags);
 
@@ -1452,6 +1444,7 @@ static void __complete_background_work(struct smq_policy *mq,
 		clear_pending(mq, e);
 		if (success) {
 			e->oblock = work->oblock;
+			e->level = NR_CACHE_LEVELS - 1;
 			push(mq, e);
 			// h, q, a
 		} else {

drivers/md/dm-cache-target.c

@@ -94,6 +94,9 @@ static void iot_io_begin(struct io_tracker *iot, sector_t len)
 
 static void __iot_io_end(struct io_tracker *iot, sector_t len)
 {
+	if (!len)
+		return;
+
 	iot->in_flight -= len;
 	if (!iot->in_flight)
 		iot->idle_time = jiffies;
@@ -474,7 +477,7 @@ struct cache {
 	spinlock_t invalidation_lock;
 	struct list_head invalidation_requests;
 
-	struct io_tracker origin_tracker;
+	struct io_tracker tracker;
 
 	struct work_struct commit_ws;
 	struct batcher committer;
@@ -901,8 +904,7 @@ static dm_oblock_t get_bio_block(struct cache *cache, struct bio *bio)
 
 static bool accountable_bio(struct cache *cache, struct bio *bio)
 {
-	return ((bio->bi_bdev == cache->origin_dev->bdev) &&
-		bio_op(bio) != REQ_OP_DISCARD);
+	return bio_op(bio) != REQ_OP_DISCARD;
 }
 
 static void accounted_begin(struct cache *cache, struct bio *bio)
@@ -912,7 +914,7 @@ static void accounted_begin(struct cache *cache, struct bio *bio)
 
 	if (accountable_bio(cache, bio)) {
 		pb->len = bio_sectors(bio);
-		iot_io_begin(&cache->origin_tracker, pb->len);
+		iot_io_begin(&cache->tracker, pb->len);
 	}
 }
 
@@ -921,7 +923,7 @@ static void accounted_complete(struct cache *cache, struct bio *bio)
 	size_t pb_data_size = get_per_bio_data_size(cache);
 	struct per_bio_data *pb = get_per_bio_data(bio, pb_data_size);
 
-	iot_io_end(&cache->origin_tracker, pb->len);
+	iot_io_end(&cache->tracker, pb->len);
 }
 
 static void accounted_request(struct cache *cache, struct bio *bio)
@@ -1716,20 +1718,19 @@ static int invalidate_start(struct cache *cache, dm_cblock_t cblock,
 
 enum busy {
 	IDLE,
-	MODERATE,
 	BUSY
 };
 
 static enum busy spare_migration_bandwidth(struct cache *cache)
 {
-	bool idle = iot_idle_for(&cache->origin_tracker, HZ);
+	bool idle = iot_idle_for(&cache->tracker, HZ);
 	sector_t current_volume = (atomic_read(&cache->nr_io_migrations) + 1) *
 		cache->sectors_per_block;
 
-	if (current_volume <= cache->migration_threshold)
-		return idle ? IDLE : MODERATE;
+	if (idle && current_volume <= cache->migration_threshold)
+		return IDLE;
 	else
-		return idle ? MODERATE : BUSY;
+		return BUSY;
 }
 
 static void inc_hit_counter(struct cache *cache, struct bio *bio)
@@ -2045,8 +2046,6 @@ static void check_migrations(struct work_struct *ws)
 
 	for (;;) {
 		b = spare_migration_bandwidth(cache);
-		if (b == BUSY)
-			break;
 
 		r = policy_get_background_work(cache->policy, b == IDLE, &op);
 		if (r == -ENODATA)
@@ -2717,7 +2716,7 @@ static int cache_create(struct cache_args *ca, struct cache **result)
 
 	batcher_init(&cache->committer, commit_op, cache,
 		     issue_op, cache, cache->wq);
-	iot_init(&cache->origin_tracker);
+	iot_init(&cache->tracker);
 
 	init_rwsem(&cache->background_work_lock);
 	prevent_background_work(cache);
@@ -2941,7 +2940,7 @@ static void cache_postsuspend(struct dm_target *ti)
 
 	cancel_delayed_work(&cache->waker);
 	flush_workqueue(cache->wq);
-	WARN_ON(cache->origin_tracker.in_flight);
+	WARN_ON(cache->tracker.in_flight);
 
 	/*
 	 * If it's a flush suspend there won't be any deferred bios, so this

drivers/md/dm-mpath.c

@@ -447,7 +447,7 @@ static struct pgpath *choose_pgpath(struct multipath *m, size_t nr_bytes)
  * it has been invoked.
  */
 #define dm_report_EIO(m)						\
-({									\
+do {									\
 	struct mapped_device *md = dm_table_get_md((m)->ti->table);	\
 									\
 	pr_debug("%s: returning EIO; QIFNP = %d; SQIFNP = %d; DNFS = %d\n", \
@@ -455,8 +455,7 @@ static struct pgpath *choose_pgpath(struct multipath *m, size_t nr_bytes)
 		 test_bit(MPATHF_QUEUE_IF_NO_PATH, &(m)->flags),	\
 		 test_bit(MPATHF_SAVED_QUEUE_IF_NO_PATH, &(m)->flags),	\
 		 dm_noflush_suspending((m)->ti));			\
-	-EIO;								\
-})
+} while (0)
 
 /*
  * Map cloned requests (request-based multipath)
@@ -481,7 +480,8 @@ static int multipath_clone_and_map(struct dm_target *ti, struct request *rq,
 	if (!pgpath) {
 		if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags))
 			return DM_MAPIO_DELAY_REQUEUE;
-		return dm_report_EIO(m);	/* Failed */
+		dm_report_EIO(m);	/* Failed */
+		return DM_MAPIO_KILL;
 	} else if (test_bit(MPATHF_QUEUE_IO, &m->flags) ||
 		   test_bit(MPATHF_PG_INIT_REQUIRED, &m->flags)) {
 		if (pg_init_all_paths(m))
@@ -558,7 +558,8 @@ static int __multipath_map_bio(struct multipath *m, struct bio *bio, struct dm_m
 	if (!pgpath) {
 		if (test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags))
 			return DM_MAPIO_REQUEUE;
-		return dm_report_EIO(m);
+		dm_report_EIO(m);
+		return -EIO;
 	}
 
 	mpio->pgpath = pgpath;
@@ -1493,7 +1494,7 @@ static int multipath_end_io(struct dm_target *ti, struct request *clone,
 		if (atomic_read(&m->nr_valid_paths) == 0 &&
 		    !test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) {
 			if (error == -EIO)
-				error = dm_report_EIO(m);
+				dm_report_EIO(m);
 			/* complete with the original error */
 			r = DM_ENDIO_DONE;
 		}
@@ -1524,8 +1525,10 @@ static int do_end_io_bio(struct multipath *m, struct bio *clone,
 		fail_path(mpio->pgpath);
 
 	if (atomic_read(&m->nr_valid_paths) == 0 &&
-	    !test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags))
-		return dm_report_EIO(m);
+	    !test_bit(MPATHF_QUEUE_IF_NO_PATH, &m->flags)) {
+		dm_report_EIO(m);
+		return -EIO;
+	}
 
 	/* Queue for the daemon to resubmit */
 	dm_bio_restore(get_bio_details_from_bio(clone), clone);

drivers/md/dm-rq.c

@@ -507,6 +507,7 @@ static int map_request(struct dm_rq_target_io *tio)
 	case DM_MAPIO_KILL:
 		/* The target wants to complete the I/O */
 		dm_kill_unmapped_request(rq, -EIO);
+		break;
 	default:
 		DMWARN("unimplemented target map return value: %d", r);
 		BUG();

drivers/md/dm-thin-metadata.c

@@ -484,11 +484,11 @@ static int __write_initial_superblock(struct dm_pool_metadata *pmd)
 	if (r < 0)
 		return r;
 
-	r = save_sm_roots(pmd);
+	r = dm_tm_pre_commit(pmd->tm);
 	if (r < 0)
 		return r;
 
-	r = dm_tm_pre_commit(pmd->tm);
+	r = save_sm_roots(pmd);
 	if (r < 0)
 		return r;
 

drivers/md/persistent-data/dm-space-map-disk.c

@@ -142,10 +142,23 @@ static int sm_disk_inc_block(struct dm_space_map *sm, dm_block_t b)
 
 static int sm_disk_dec_block(struct dm_space_map *sm, dm_block_t b)
 {
+	int r;
+	uint32_t old_count;
 	enum allocation_event ev;
 	struct sm_disk *smd = container_of(sm, struct sm_disk, sm);
 
-	return sm_ll_dec(&smd->ll, b, &ev);
+	r = sm_ll_dec(&smd->ll, b, &ev);
+	if (!r && (ev == SM_FREE)) {
+		/*
+		 * It's only free if it's also free in the last
+		 * transaction.
+		 */
+		r = sm_ll_lookup(&smd->old_ll, b, &old_count);
+		if (!r && !old_count)
+			smd->nr_allocated_this_transaction--;
+	}
+
+	return r;
 }
 
 static int sm_disk_new_block(struct dm_space_map *sm, dm_block_t *b)