Merge branch 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/shli/md

Pull MD updates from Shaohua Li:
"Some small fixes for MD:

- fix raid5-cache potential problems if raid5 cache isn't fully
recovered

- fix a wait-within-wait warning in raid1/10

- make raid5-PPL support disks with writeback cache enabled"

* 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/shli/md:
raid5-ppl: PPL support for disks with write-back cache enabled
md/r5cache: print more info of log recovery
md/raid1,raid10: silence warning about wait-within-wait
md: introduce new personality funciton start()

Documentation/md/raid5-ppl.txt

···
 PPL is available for md version-1 metadata and external (specifically IMSM)
 metadata arrays. It can be enabled using mdadm option --consistency-policy=ppl.
 
-Currently, volatile write-back cache should be disabled on all member drives
-when using PPL. Otherwise it cannot guarantee consistency in case of power
-failure.
+There is a limitation of maximum 64 disks in the array for PPL. It allows to
+keep data structures and implementation simple. RAID5 arrays with so many disks
+are not likely due to high risk of multiple disks failure. Such restriction
+should not be a real life limitation.

drivers/md/dm-raid.c

···
 		goto bad;
 	}
 
+	r = md_start(&rs->md);
+
+	if (r) {
+		ti->error = "Failed to start raid array";
+		mddev_unlock(&rs->md);
+		goto bad_md_start;
+	}
+
 	rs->callbacks.congested_fn = raid_is_congested;
 	dm_table_add_target_callbacks(ti->table, &rs->callbacks);
 
···
 	mddev_unlock(&rs->md);
 	return 0;
 
+bad_md_start:
 bad_journal_mode_set:
 bad_stripe_cache:
 bad_check_reshape:

drivers/md/md.c

···
 	return NULL;
 }
 
-static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
+struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev)
 {
 	struct md_rdev *rdev;
 
···
 
 	return NULL;
 }
+EXPORT_SYMBOL_GPL(md_find_rdev_rcu);
 
 static struct md_personality *find_pers(int level, char *clevel)
 {
···
 	if (start_readonly && mddev->ro == 0)
 		mddev->ro = 2; /* read-only, but switch on first write */
 
-	/*
-	 * NOTE: some pers->run(), for example r5l_recovery_log(), wakes
-	 * up mddev->thread. It is important to initialize critical
-	 * resources for mddev->thread BEFORE calling pers->run().
-	 */
 	err = pers->run(mddev);
 	if (err)
 		pr_warn("md: pers->run() failed ...\n");
···
 	if (mddev_is_clustered(mddev))
 		md_allow_write(mddev);
 
+	/* run start up tasks that require md_thread */
+	md_start(mddev);
+
 	md_wakeup_thread(mddev->thread);
 	md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
 
···
 out:
 	return err;
 }
+
+int md_start(struct mddev *mddev)
+{
+	int ret = 0;
+
+	if (mddev->pers->start) {
+		set_bit(MD_RECOVERY_WAIT, &mddev->recovery);
+		md_wakeup_thread(mddev->thread);
+		ret = mddev->pers->start(mddev);
+		clear_bit(MD_RECOVERY_WAIT, &mddev->recovery);
+		md_wakeup_thread(mddev->sync_thread);
+	}
+	return ret;
+}
+EXPORT_SYMBOL_GPL(md_start);
 
 static int restart_array(struct mddev *mddev)
 {
···
 		return -ENODEV;
 
 	rcu_read_lock();
-	rdev = find_rdev_rcu(mddev, dev);
+	rdev = md_find_rdev_rcu(mddev, dev);
 	if (!rdev)
 		err =  -ENODEV;
 	else {
···
 	int ret;
 
 	/* just incase thread restarts... */
-	if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
+	if (test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
+	    test_bit(MD_RECOVERY_WAIT, &mddev->recovery))
 		return;
 	if (mddev->ro) {/* never try to sync a read-only array */
 		set_bit(MD_RECOVERY_INTR, &mddev->recovery);

drivers/md/md.h

···
 	MD_RECOVERY_RESHAPE,	/* A reshape is happening */
 	MD_RECOVERY_FROZEN,	/* User request to abort, and not restart, any action */
 	MD_RECOVERY_ERROR,	/* sync-action interrupted because io-error */
+	MD_RECOVERY_WAIT,	/* waiting for pers->start() to finish */
 };
 
 static inline int __must_check mddev_lock(struct mddev *mddev)
···
 	struct list_head list;
 	struct module *owner;
 	bool (*make_request)(struct mddev *mddev, struct bio *bio);
+	/*
+	 * start up works that do NOT require md_thread. tasks that
+	 * requires md_thread should go into start()
+	 */
 	int (*run)(struct mddev *mddev);
+	/* start up works that require md threads */
+	int (*start)(struct mddev *mddev);
 	void (*free)(struct mddev *mddev, void *priv);
 	void (*status)(struct seq_file *seq, struct mddev *mddev);
 	/* error_handler must set ->faulty and clear ->in_sync
···
 
 extern void mddev_init(struct mddev *mddev);
 extern int md_run(struct mddev *mddev);
+extern int md_start(struct mddev *mddev);
 extern void md_stop(struct mddev *mddev);
 extern void md_stop_writes(struct mddev *mddev);
 extern int md_rdev_init(struct md_rdev *rdev);
···
 extern void md_update_sb(struct mddev *mddev, int force);
 extern void md_kick_rdev_from_array(struct md_rdev * rdev);
 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
+struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev);
 
 static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
 {

drivers/md/raid1.c

···
 		bio = bio_list_get(&conf->pending_bio_list);
 		conf->pending_count = 0;
 		spin_unlock_irq(&conf->device_lock);
+
+		/*
+		 * As this is called in a wait_event() loop (see freeze_array),
+		 * current->state might be TASK_UNINTERRUPTIBLE which will
+		 * cause a warning when we prepare to wait again.  As it is
+		 * rare that this path is taken, it is perfectly safe to force
+		 * us to go around the wait_event() loop again, so the warning
+		 * is a false-positive.  Silence the warning by resetting
+		 * thread state
+		 */
+		__set_current_state(TASK_RUNNING);
 		blk_start_plug(&plug);
 		flush_bio_list(conf, bio);
 		blk_finish_plug(&plug);

drivers/md/raid10.c

···
 		bio = bio_list_get(&conf->pending_bio_list);
 		conf->pending_count = 0;
 		spin_unlock_irq(&conf->device_lock);
+
+		/*
+		 * As this is called in a wait_event() loop (see freeze_array),
+		 * current->state might be TASK_UNINTERRUPTIBLE which will
+		 * cause a warning when we prepare to wait again.  As it is
+		 * rare that this path is taken, it is perfectly safe to force
+		 * us to go around the wait_event() loop again, so the warning
+		 * is a false-positive. Silence the warning by resetting
+		 * thread state
+		 */
+		__set_current_state(TASK_RUNNING);
+
 		blk_start_plug(&plug);
 		/* flush any pending bitmap writes to disk
 		 * before proceeding w/ I/O */

drivers/md/raid5-cache.c

···
 
 int r5l_handle_flush_request(struct r5l_log *log, struct bio *bio)
 {
-	if (!log)
-		return -ENODEV;
-
 	if (log->r5c_journal_mode == R5C_JOURNAL_MODE_WRITE_THROUGH) {
 		/*
 		 * in write through (journal only)
···
 void r5l_quiesce(struct r5l_log *log, int quiesce)
 {
 	struct mddev *mddev;
-	if (!log)
-		return;
 
 	if (quiesce) {
 		/* make sure r5l_write_super_and_discard_space exits */
···
 		raid5_release_stripe(sh);
 	}
 
-	md_wakeup_thread(conf->mddev->thread);
 	/* reuse conf->wait_for_quiescent in recovery */
 	wait_event(conf->wait_for_quiescent,
 		   atomic_read(&conf->active_stripes) == 0);
···
 	ctx->seq += 10000;
 
 	if ((ctx->data_only_stripes == 0) && (ctx->data_parity_stripes == 0))
-		pr_debug("md/raid:%s: starting from clean shutdown\n",
+		pr_info("md/raid:%s: starting from clean shutdown\n",
 			 mdname(mddev));
 	else
-		pr_debug("md/raid:%s: recovering %d data-only stripes and %d data-parity stripes\n",
+		pr_info("md/raid:%s: recovering %d data-only stripes and %d data-parity stripes\n",
 			 mdname(mddev), ctx->data_only_stripes,
 			 ctx->data_parity_stripes);
 
···
 	return ret;
 }
 
+int r5l_start(struct r5l_log *log)
+{
+	int ret;
+
+	if (!log)
+		return 0;
+
+	ret = r5l_load_log(log);
+	if (ret) {
+		struct mddev *mddev = log->rdev->mddev;
+		struct r5conf *conf = mddev->private;
+
+		r5l_exit_log(conf);
+	}
+	return ret;
+}
+
 void r5c_update_on_rdev_error(struct mddev *mddev, struct md_rdev *rdev)
 {
 	struct r5conf *conf = mddev->private;
···
 
 	rcu_assign_pointer(conf->log, log);
 
-	if (r5l_load_log(log))
-		goto error;
-
 	set_bit(MD_HAS_JOURNAL, &conf->mddev->flags);
 	return 0;
 
-error:
 	rcu_assign_pointer(conf->log, NULL);
 	md_unregister_thread(&log->reclaim_thread);
 reclaim_thread:

drivers/md/raid5-log.h

···
 extern void r5c_update_on_rdev_error(struct mddev *mddev,
 				     struct md_rdev *rdev);
 extern bool r5c_big_stripe_cached(struct r5conf *conf, sector_t sect);
+extern int r5l_start(struct r5l_log *log);
 
 extern struct dma_async_tx_descriptor *
 ops_run_partial_parity(struct stripe_head *sh, struct raid5_percpu *percpu,
···
 extern void ppl_write_stripe_run(struct r5conf *conf);
 extern void ppl_stripe_write_finished(struct stripe_head *sh);
 extern int ppl_modify_log(struct r5conf *conf, struct md_rdev *rdev, bool add);
+extern void ppl_quiesce(struct r5conf *conf, int quiesce);
 
 static inline bool raid5_has_ppl(struct r5conf *conf)
 {
···
 		r5l_write_stripe_run(conf->log);
 	else if (raid5_has_ppl(conf))
 		ppl_write_stripe_run(conf);
+}
+
+static inline void log_flush_stripe_to_raid(struct r5conf *conf)
+{
+	if (conf->log)
+		r5l_flush_stripe_to_raid(conf->log);
+	else if (raid5_has_ppl(conf))
+		ppl_write_stripe_run(conf);
+}
+
+static inline int log_handle_flush_request(struct r5conf *conf, struct bio *bio)
+{
+	int ret = -ENODEV;
+
+	if (conf->log)
+		ret = r5l_handle_flush_request(conf->log, bio);
+	else if (raid5_has_ppl(conf))
+		ret = 0;
+
+	return ret;
+}
+
+static inline void log_quiesce(struct r5conf *conf, int quiesce)
+{
+	if (conf->log)
+		r5l_quiesce(conf->log, quiesce);
+	else if (raid5_has_ppl(conf))
+		ppl_quiesce(conf, quiesce);
 }
 
 static inline void log_exit(struct r5conf *conf)

drivers/md/raid5-ppl.c

···
  * (for a single member disk). New io_units are added to the end of the list
  * and the first io_unit is submitted, if it is not submitted already.
  * The current io_unit accepting new stripes is always at the end of the list.
+ *
+ * If write-back cache is enabled for any of the disks in the array, its data
+ * must be flushed before next io_unit is submitted.
  */
 
 #define PPL_SPACE_SIZE (128 * 1024)
···
 	struct kmem_cache *io_kc;
 	mempool_t *io_pool;
 	struct bio_set *bs;
+	struct bio_set *flush_bs;
 
 	/* used only for recovery */
 	int recovered_entries;
···
 	sector_t next_io_sector;
 	unsigned int entry_space;
 	bool use_multippl;
+	bool wb_cache_on;
+	unsigned long disk_flush_bitmap;
 };
 
 #define PPL_IO_INLINE_BVECS 32
···
 
 	struct list_head stripe_list;	/* stripes added to the io_unit */
 	atomic_t pending_stripes;	/* how many stripes not written to raid */
+	atomic_t pending_flushes;	/* how many disk flushes are in progress */
 
 	bool submitted;			/* true if write to log started */
 
···
 	INIT_LIST_HEAD(&io->log_sibling);
 	INIT_LIST_HEAD(&io->stripe_list);
 	atomic_set(&io->pending_stripes, 0);
+	atomic_set(&io->pending_flushes, 0);
 	bio_init(&io->bio, io->biovec, PPL_IO_INLINE_BVECS);
 
 	pplhdr = page_address(io->header_page);
···
 	if (log->use_multippl)
 		log->next_io_sector += (PPL_HEADER_SIZE + io->pp_size) >> 9;
 
+	WARN_ON(log->disk_flush_bitmap != 0);
+
 	list_for_each_entry(sh, &io->stripe_list, log_list) {
+		for (i = 0; i < sh->disks; i++) {
+			struct r5dev *dev = &sh->dev[i];
+
+			if ((ppl_conf->child_logs[i].wb_cache_on) &&
+			    (test_bit(R5_Wantwrite, &dev->flags))) {
+				set_bit(i, &log->disk_flush_bitmap);
+			}
+		}
+
 		/* entries for full stripe writes have no partial parity */
 		if (test_bit(STRIPE_FULL_WRITE, &sh->state))
 			continue;
···
 {
 	struct ppl_log *log = io->log;
 	struct ppl_conf *ppl_conf = log->ppl_conf;
+	struct r5conf *conf = ppl_conf->mddev->private;
 	unsigned long flags;
 
 	pr_debug("%s: seq: %llu\n", __func__, io->seq);
···
 	spin_unlock(&ppl_conf->no_mem_stripes_lock);
 
 	local_irq_restore(flags);
+
+	wake_up(&conf->wait_for_quiescent);
+}
+
+static void ppl_flush_endio(struct bio *bio)
+{
+	struct ppl_io_unit *io = bio->bi_private;
+	struct ppl_log *log = io->log;
+	struct ppl_conf *ppl_conf = log->ppl_conf;
+	struct r5conf *conf = ppl_conf->mddev->private;
+	char b[BDEVNAME_SIZE];
+
+	pr_debug("%s: dev: %s\n", __func__, bio_devname(bio, b));
+
+	if (bio->bi_status) {
+		struct md_rdev *rdev;
+
+		rcu_read_lock();
+		rdev = md_find_rdev_rcu(conf->mddev, bio_dev(bio));
+		if (rdev)
+			md_error(rdev->mddev, rdev);
+		rcu_read_unlock();
+	}
+
+	bio_put(bio);
+
+	if (atomic_dec_and_test(&io->pending_flushes)) {
+		ppl_io_unit_finished(io);
+		md_wakeup_thread(conf->mddev->thread);
+	}
+}
+
+static void ppl_do_flush(struct ppl_io_unit *io)
+{
+	struct ppl_log *log = io->log;
+	struct ppl_conf *ppl_conf = log->ppl_conf;
+	struct r5conf *conf = ppl_conf->mddev->private;
+	int raid_disks = conf->raid_disks;
+	int flushed_disks = 0;
+	int i;
+
+	atomic_set(&io->pending_flushes, raid_disks);
+
+	for_each_set_bit(i, &log->disk_flush_bitmap, raid_disks) {
+		struct md_rdev *rdev;
+		struct block_device *bdev = NULL;
+
+		rcu_read_lock();
+		rdev = rcu_dereference(conf->disks[i].rdev);
+		if (rdev && !test_bit(Faulty, &rdev->flags))
+			bdev = rdev->bdev;
+		rcu_read_unlock();
+
+		if (bdev) {
+			struct bio *bio;
+			char b[BDEVNAME_SIZE];
+
+			bio = bio_alloc_bioset(GFP_NOIO, 0, ppl_conf->flush_bs);
+			bio_set_dev(bio, bdev);
+			bio->bi_private = io;
+			bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
+			bio->bi_end_io = ppl_flush_endio;
+
+			pr_debug("%s: dev: %s\n", __func__,
+				 bio_devname(bio, b));
+
+			submit_bio(bio);
+			flushed_disks++;
+		}
+	}
+
+	log->disk_flush_bitmap = 0;
+
+	for (i = flushed_disks ; i < raid_disks; i++) {
+		if (atomic_dec_and_test(&io->pending_flushes))
+			ppl_io_unit_finished(io);
+	}
+}
+
+static inline bool ppl_no_io_unit_submitted(struct r5conf *conf,
+					    struct ppl_log *log)
+{
+	struct ppl_io_unit *io;
+
+	io = list_first_entry_or_null(&log->io_list, struct ppl_io_unit,
+				      log_sibling);
+
+	return !io || !io->submitted;
+}
+
+void ppl_quiesce(struct r5conf *conf, int quiesce)
+{
+	struct ppl_conf *ppl_conf = conf->log_private;
+	int i;
+
+	if (quiesce) {
+		for (i = 0; i < ppl_conf->count; i++) {
+			struct ppl_log *log = &ppl_conf->child_logs[i];
+
+			spin_lock_irq(&log->io_list_lock);
+			wait_event_lock_irq(conf->wait_for_quiescent,
+					    ppl_no_io_unit_submitted(conf, log),
+					    log->io_list_lock);
+			spin_unlock_irq(&log->io_list_lock);
+		}
+	}
 }
 
 void ppl_stripe_write_finished(struct stripe_head *sh)
···
 	io = sh->ppl_io;
 	sh->ppl_io = NULL;
 
-	if (io && atomic_dec_and_test(&io->pending_stripes))
-		ppl_io_unit_finished(io);
+	if (io && atomic_dec_and_test(&io->pending_stripes)) {
+		if (io->log->disk_flush_bitmap)
+			ppl_do_flush(io);
+		else
+			ppl_io_unit_finished(io);
+	}
 }
 
 static void ppl_xor(int size, struct page *page1, struct page *page2)
···
 
 	if (ppl_conf->bs)
 		bioset_free(ppl_conf->bs);
+	if (ppl_conf->flush_bs)
+		bioset_free(ppl_conf->flush_bs);
 	mempool_destroy(ppl_conf->io_pool);
 	kmem_cache_destroy(ppl_conf->io_kc);
 
···
 
 static void ppl_init_child_log(struct ppl_log *log, struct md_rdev *rdev)
 {
+	struct request_queue *q;
+
 	if ((rdev->ppl.size << 9) >= (PPL_SPACE_SIZE +
 				      PPL_HEADER_SIZE) * 2) {
 		log->use_multippl = true;
···
 				   PPL_HEADER_SIZE;
 	}
 	log->next_io_sector = rdev->ppl.sector;
+
+	q = bdev_get_queue(rdev->bdev);
+	if (test_bit(QUEUE_FLAG_WC, &q->queue_flags))
+		log->wb_cache_on = true;
 }
 
 int ppl_init_log(struct r5conf *conf)
···
 	struct ppl_conf *ppl_conf;
 	struct mddev *mddev = conf->mddev;
 	int ret = 0;
+	int max_disks;
 	int i;
-	bool need_cache_flush = false;
 
 	pr_debug("md/raid:%s: enabling distributed Partial Parity Log\n",
 		 mdname(conf->mddev));
···
 		return -EINVAL;
 	}
 
+	max_disks = FIELD_SIZEOF(struct ppl_log, disk_flush_bitmap) *
+		BITS_PER_BYTE;
+	if (conf->raid_disks > max_disks) {
+		pr_warn("md/raid:%s PPL doesn't support over %d disks in the array\n",
+			mdname(mddev), max_disks);
+		return -EINVAL;
+	}
+
 	ppl_conf = kzalloc(sizeof(struct ppl_conf), GFP_KERNEL);
 	if (!ppl_conf)
 		return -ENOMEM;
···
 
 	ppl_conf->bs = bioset_create(conf->raid_disks, 0, BIOSET_NEED_BVECS);
 	if (!ppl_conf->bs) {
+		ret = -ENOMEM;
+		goto err;
+	}
+
+	ppl_conf->flush_bs = bioset_create(conf->raid_disks, 0, 0);
+	if (!ppl_conf->flush_bs) {
 		ret = -ENOMEM;
 		goto err;
 	}
···
 		log->rdev = rdev;
 
 		if (rdev) {
-			struct request_queue *q;
-
 			ret = ppl_validate_rdev(rdev);
 			if (ret)
 				goto err;
 
-			q = bdev_get_queue(rdev->bdev);
-			if (test_bit(QUEUE_FLAG_WC, &q->queue_flags))
-				need_cache_flush = true;
 			ppl_init_child_log(log, rdev);
 		}
 	}
-
-	if (need_cache_flush)
-		pr_warn("md/raid:%s: Volatile write-back cache should be disabled on all member drives when using PPL!\n",
-			mdname(mddev));
 
 	/* load and possibly recover the logs from the member disks */
 	ret = ppl_load(ppl_conf);

drivers/md/raid5.c

···
 	bool do_flush = false;
 
 	if (unlikely(bi->bi_opf & REQ_PREFLUSH)) {
-		int ret = r5l_handle_flush_request(conf->log, bi);
+		int ret = log_handle_flush_request(conf, bi);
 
 		if (ret == 0)
 			return true;
···
 				break;
 		if (i == NR_STRIPE_HASH_LOCKS) {
 			spin_unlock_irq(&conf->device_lock);
-			r5l_flush_stripe_to_raid(conf->log);
+			log_flush_stripe_to_raid(conf);
 			spin_lock_irq(&conf->device_lock);
 			return batch_size;
 		}
···
 		wake_up(&conf->wait_for_overlap);
 		unlock_all_device_hash_locks_irq(conf);
 	}
-	r5l_quiesce(conf->log, quiesce);
+	log_quiesce(conf, quiesce);
 }
 
 static void *raid45_takeover_raid0(struct mddev *mddev, int level)
···
 	return err;
 }
 
+static int raid5_start(struct mddev *mddev)
+{
+	struct r5conf *conf = mddev->private;
+
+	return r5l_start(conf->log);
+}
+
 static struct md_personality raid6_personality =
 {
 	.name		= "raid6",
···
 	.owner		= THIS_MODULE,
 	.make_request	= raid5_make_request,
 	.run		= raid5_run,
+	.start		= raid5_start,
 	.free		= raid5_free,
 	.status		= raid5_status,
 	.error_handler	= raid5_error,
···
 	.owner		= THIS_MODULE,
 	.make_request	= raid5_make_request,
 	.run		= raid5_run,
+	.start		= raid5_start,
 	.free		= raid5_free,
 	.status		= raid5_status,
 	.error_handler	= raid5_error,
···
 	.owner		= THIS_MODULE,
 	.make_request	= raid5_make_request,
 	.run		= raid5_run,
+	.start		= raid5_start,
 	.free		= raid5_free,
 	.status		= raid5_status,
 	.error_handler	= raid5_error,