xref: /openbmc/linux/drivers/md/dm-zoned-target.c (revision a44e4f3a)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2017 Western Digital Corporation or its affiliates.
4  *
5  * This file is released under the GPL.
6  */
7 
8 #include "dm-zoned.h"
9 
10 #include <linux/module.h>
11 
12 #define	DM_MSG_PREFIX		"zoned"
13 
14 #define DMZ_MIN_BIOS		8192
15 
16 /*
17  * Zone BIO context.
18  */
19 struct dmz_bioctx {
20 	struct dmz_target	*target;
21 	struct dm_zone		*zone;
22 	struct bio		*bio;
23 	refcount_t		ref;
24 };
25 
26 /*
27  * Chunk work descriptor.
28  */
29 struct dm_chunk_work {
30 	struct work_struct	work;
31 	refcount_t		refcount;
32 	struct dmz_target	*target;
33 	unsigned int		chunk;
34 	struct bio_list		bio_list;
35 };
36 
37 /*
38  * Target descriptor.
39  */
40 struct dmz_target {
41 	struct dm_dev		*ddev;
42 
43 	unsigned long		flags;
44 
45 	/* Zoned block device information */
46 	struct dmz_dev		*dev;
47 
48 	/* For metadata handling */
49 	struct dmz_metadata     *metadata;
50 
51 	/* For reclaim */
52 	struct dmz_reclaim	*reclaim;
53 
54 	/* For chunk work */
55 	struct radix_tree_root	chunk_rxtree;
56 	struct workqueue_struct *chunk_wq;
57 	struct mutex		chunk_lock;
58 
59 	/* For cloned BIOs to zones */
60 	struct bio_set		bio_set;
61 
62 	/* For flush */
63 	spinlock_t		flush_lock;
64 	struct bio_list		flush_list;
65 	struct delayed_work	flush_work;
66 	struct workqueue_struct *flush_wq;
67 };
68 
69 /*
70  * Flush intervals (seconds).
71  */
72 #define DMZ_FLUSH_PERIOD	(10 * HZ)
73 
74 /*
75  * Target BIO completion.
76  */
77 static inline void dmz_bio_endio(struct bio *bio, blk_status_t status)
78 {
79 	struct dmz_bioctx *bioctx = dm_per_bio_data(bio, sizeof(struct dmz_bioctx));
80 
81 	if (status != BLK_STS_OK && bio->bi_status == BLK_STS_OK)
82 		bio->bi_status = status;
83 
84 	if (refcount_dec_and_test(&bioctx->ref)) {
85 		struct dm_zone *zone = bioctx->zone;
86 
87 		if (zone) {
88 			if (bio->bi_status != BLK_STS_OK &&
89 			    bio_op(bio) == REQ_OP_WRITE &&
90 			    dmz_is_seq(zone))
91 				set_bit(DMZ_SEQ_WRITE_ERR, &zone->flags);
92 			dmz_deactivate_zone(zone);
93 		}
94 		bio_endio(bio);
95 	}
96 }
97 
98 /*
99  * Completion callback for an internally cloned target BIO. This terminates the
100  * target BIO when there are no more references to its context.
101  */
102 static void dmz_clone_endio(struct bio *clone)
103 {
104 	struct dmz_bioctx *bioctx = clone->bi_private;
105 	blk_status_t status = clone->bi_status;
106 
107 	bio_put(clone);
108 	dmz_bio_endio(bioctx->bio, status);
109 }
110 
111 /*
112  * Issue a clone of a target BIO. The clone may only partially process the
113  * original target BIO.
114  */
115 static int dmz_submit_bio(struct dmz_target *dmz, struct dm_zone *zone,
116 			  struct bio *bio, sector_t chunk_block,
117 			  unsigned int nr_blocks)
118 {
119 	struct dmz_bioctx *bioctx = dm_per_bio_data(bio, sizeof(struct dmz_bioctx));
120 	struct bio *clone;
121 
122 	clone = bio_clone_fast(bio, GFP_NOIO, &dmz->bio_set);
123 	if (!clone)
124 		return -ENOMEM;
125 
126 	bio_set_dev(clone, dmz->dev->bdev);
127 	clone->bi_iter.bi_sector =
128 		dmz_start_sect(dmz->metadata, zone) + dmz_blk2sect(chunk_block);
129 	clone->bi_iter.bi_size = dmz_blk2sect(nr_blocks) << SECTOR_SHIFT;
130 	clone->bi_end_io = dmz_clone_endio;
131 	clone->bi_private = bioctx;
132 
133 	bio_advance(bio, clone->bi_iter.bi_size);
134 
135 	refcount_inc(&bioctx->ref);
136 	generic_make_request(clone);
137 
138 	if (bio_op(bio) == REQ_OP_WRITE && dmz_is_seq(zone))
139 		zone->wp_block += nr_blocks;
140 
141 	return 0;
142 }
143 
144 /*
145  * Zero out pages of discarded blocks accessed by a read BIO.
146  */
147 static void dmz_handle_read_zero(struct dmz_target *dmz, struct bio *bio,
148 				 sector_t chunk_block, unsigned int nr_blocks)
149 {
150 	unsigned int size = nr_blocks << DMZ_BLOCK_SHIFT;
151 
152 	/* Clear nr_blocks */
153 	swap(bio->bi_iter.bi_size, size);
154 	zero_fill_bio(bio);
155 	swap(bio->bi_iter.bi_size, size);
156 
157 	bio_advance(bio, size);
158 }
159 
160 /*
161  * Process a read BIO.
162  */
163 static int dmz_handle_read(struct dmz_target *dmz, struct dm_zone *zone,
164 			   struct bio *bio)
165 {
166 	sector_t chunk_block = dmz_chunk_block(dmz->dev, dmz_bio_block(bio));
167 	unsigned int nr_blocks = dmz_bio_blocks(bio);
168 	sector_t end_block = chunk_block + nr_blocks;
169 	struct dm_zone *rzone, *bzone;
170 	int ret;
171 
172 	/* Read into unmapped chunks need only zeroing the BIO buffer */
173 	if (!zone) {
174 		zero_fill_bio(bio);
175 		return 0;
176 	}
177 
178 	dmz_dev_debug(dmz->dev, "READ chunk %llu -> %s zone %u, block %llu, %u blocks",
179 		      (unsigned long long)dmz_bio_chunk(dmz->dev, bio),
180 		      (dmz_is_rnd(zone) ? "RND" : "SEQ"),
181 		      dmz_id(dmz->metadata, zone),
182 		      (unsigned long long)chunk_block, nr_blocks);
183 
184 	/* Check block validity to determine the read location */
185 	bzone = zone->bzone;
186 	while (chunk_block < end_block) {
187 		nr_blocks = 0;
188 		if (dmz_is_rnd(zone) || chunk_block < zone->wp_block) {
189 			/* Test block validity in the data zone */
190 			ret = dmz_block_valid(dmz->metadata, zone, chunk_block);
191 			if (ret < 0)
192 				return ret;
193 			if (ret > 0) {
194 				/* Read data zone blocks */
195 				nr_blocks = ret;
196 				rzone = zone;
197 			}
198 		}
199 
200 		/*
201 		 * No valid blocks found in the data zone.
202 		 * Check the buffer zone, if there is one.
203 		 */
204 		if (!nr_blocks && bzone) {
205 			ret = dmz_block_valid(dmz->metadata, bzone, chunk_block);
206 			if (ret < 0)
207 				return ret;
208 			if (ret > 0) {
209 				/* Read buffer zone blocks */
210 				nr_blocks = ret;
211 				rzone = bzone;
212 			}
213 		}
214 
215 		if (nr_blocks) {
216 			/* Valid blocks found: read them */
217 			nr_blocks = min_t(unsigned int, nr_blocks, end_block - chunk_block);
218 			ret = dmz_submit_bio(dmz, rzone, bio, chunk_block, nr_blocks);
219 			if (ret)
220 				return ret;
221 			chunk_block += nr_blocks;
222 		} else {
223 			/* No valid block: zeroout the current BIO block */
224 			dmz_handle_read_zero(dmz, bio, chunk_block, 1);
225 			chunk_block++;
226 		}
227 	}
228 
229 	return 0;
230 }
231 
232 /*
233  * Write blocks directly in a data zone, at the write pointer.
234  * If a buffer zone is assigned, invalidate the blocks written
235  * in place.
236  */
237 static int dmz_handle_direct_write(struct dmz_target *dmz,
238 				   struct dm_zone *zone, struct bio *bio,
239 				   sector_t chunk_block,
240 				   unsigned int nr_blocks)
241 {
242 	struct dmz_metadata *zmd = dmz->metadata;
243 	struct dm_zone *bzone = zone->bzone;
244 	int ret;
245 
246 	if (dmz_is_readonly(zone))
247 		return -EROFS;
248 
249 	/* Submit write */
250 	ret = dmz_submit_bio(dmz, zone, bio, chunk_block, nr_blocks);
251 	if (ret)
252 		return ret;
253 
254 	/*
255 	 * Validate the blocks in the data zone and invalidate
256 	 * in the buffer zone, if there is one.
257 	 */
258 	ret = dmz_validate_blocks(zmd, zone, chunk_block, nr_blocks);
259 	if (ret == 0 && bzone)
260 		ret = dmz_invalidate_blocks(zmd, bzone, chunk_block, nr_blocks);
261 
262 	return ret;
263 }
264 
265 /*
266  * Write blocks in the buffer zone of @zone.
267  * If no buffer zone is assigned yet, get one.
268  * Called with @zone write locked.
269  */
270 static int dmz_handle_buffered_write(struct dmz_target *dmz,
271 				     struct dm_zone *zone, struct bio *bio,
272 				     sector_t chunk_block,
273 				     unsigned int nr_blocks)
274 {
275 	struct dmz_metadata *zmd = dmz->metadata;
276 	struct dm_zone *bzone;
277 	int ret;
278 
279 	/* Get the buffer zone. One will be allocated if needed */
280 	bzone = dmz_get_chunk_buffer(zmd, zone);
281 	if (IS_ERR(bzone))
282 		return PTR_ERR(bzone);
283 
284 	if (dmz_is_readonly(bzone))
285 		return -EROFS;
286 
287 	/* Submit write */
288 	ret = dmz_submit_bio(dmz, bzone, bio, chunk_block, nr_blocks);
289 	if (ret)
290 		return ret;
291 
292 	/*
293 	 * Validate the blocks in the buffer zone
294 	 * and invalidate in the data zone.
295 	 */
296 	ret = dmz_validate_blocks(zmd, bzone, chunk_block, nr_blocks);
297 	if (ret == 0 && chunk_block < zone->wp_block)
298 		ret = dmz_invalidate_blocks(zmd, zone, chunk_block, nr_blocks);
299 
300 	return ret;
301 }
302 
303 /*
304  * Process a write BIO.
305  */
306 static int dmz_handle_write(struct dmz_target *dmz, struct dm_zone *zone,
307 			    struct bio *bio)
308 {
309 	sector_t chunk_block = dmz_chunk_block(dmz->dev, dmz_bio_block(bio));
310 	unsigned int nr_blocks = dmz_bio_blocks(bio);
311 
312 	if (!zone)
313 		return -ENOSPC;
314 
315 	dmz_dev_debug(dmz->dev, "WRITE chunk %llu -> %s zone %u, block %llu, %u blocks",
316 		      (unsigned long long)dmz_bio_chunk(dmz->dev, bio),
317 		      (dmz_is_rnd(zone) ? "RND" : "SEQ"),
318 		      dmz_id(dmz->metadata, zone),
319 		      (unsigned long long)chunk_block, nr_blocks);
320 
321 	if (dmz_is_rnd(zone) || chunk_block == zone->wp_block) {
322 		/*
323 		 * zone is a random zone or it is a sequential zone
324 		 * and the BIO is aligned to the zone write pointer:
325 		 * direct write the zone.
326 		 */
327 		return dmz_handle_direct_write(dmz, zone, bio, chunk_block, nr_blocks);
328 	}
329 
330 	/*
331 	 * This is an unaligned write in a sequential zone:
332 	 * use buffered write.
333 	 */
334 	return dmz_handle_buffered_write(dmz, zone, bio, chunk_block, nr_blocks);
335 }
336 
337 /*
338  * Process a discard BIO.
339  */
340 static int dmz_handle_discard(struct dmz_target *dmz, struct dm_zone *zone,
341 			      struct bio *bio)
342 {
343 	struct dmz_metadata *zmd = dmz->metadata;
344 	sector_t block = dmz_bio_block(bio);
345 	unsigned int nr_blocks = dmz_bio_blocks(bio);
346 	sector_t chunk_block = dmz_chunk_block(dmz->dev, block);
347 	int ret = 0;
348 
349 	/* For unmapped chunks, there is nothing to do */
350 	if (!zone)
351 		return 0;
352 
353 	if (dmz_is_readonly(zone))
354 		return -EROFS;
355 
356 	dmz_dev_debug(dmz->dev, "DISCARD chunk %llu -> zone %u, block %llu, %u blocks",
357 		      (unsigned long long)dmz_bio_chunk(dmz->dev, bio),
358 		      dmz_id(zmd, zone),
359 		      (unsigned long long)chunk_block, nr_blocks);
360 
361 	/*
362 	 * Invalidate blocks in the data zone and its
363 	 * buffer zone if one is mapped.
364 	 */
365 	if (dmz_is_rnd(zone) || chunk_block < zone->wp_block)
366 		ret = dmz_invalidate_blocks(zmd, zone, chunk_block, nr_blocks);
367 	if (ret == 0 && zone->bzone)
368 		ret = dmz_invalidate_blocks(zmd, zone->bzone,
369 					    chunk_block, nr_blocks);
370 	return ret;
371 }
372 
373 /*
374  * Process a BIO.
375  */
376 static void dmz_handle_bio(struct dmz_target *dmz, struct dm_chunk_work *cw,
377 			   struct bio *bio)
378 {
379 	struct dmz_bioctx *bioctx = dm_per_bio_data(bio, sizeof(struct dmz_bioctx));
380 	struct dmz_metadata *zmd = dmz->metadata;
381 	struct dm_zone *zone;
382 	int ret;
383 
384 	/*
385 	 * Write may trigger a zone allocation. So make sure the
386 	 * allocation can succeed.
387 	 */
388 	if (bio_op(bio) == REQ_OP_WRITE)
389 		dmz_schedule_reclaim(dmz->reclaim);
390 
391 	dmz_lock_metadata(zmd);
392 
393 	if (dmz->dev->flags & DMZ_BDEV_DYING) {
394 		ret = -EIO;
395 		goto out;
396 	}
397 
398 	/*
399 	 * Get the data zone mapping the chunk. There may be no
400 	 * mapping for read and discard. If a mapping is obtained,
401 	 + the zone returned will be set to active state.
402 	 */
403 	zone = dmz_get_chunk_mapping(zmd, dmz_bio_chunk(dmz->dev, bio),
404 				     bio_op(bio));
405 	if (IS_ERR(zone)) {
406 		ret = PTR_ERR(zone);
407 		goto out;
408 	}
409 
410 	/* Process the BIO */
411 	if (zone) {
412 		dmz_activate_zone(zone);
413 		bioctx->zone = zone;
414 	}
415 
416 	switch (bio_op(bio)) {
417 	case REQ_OP_READ:
418 		ret = dmz_handle_read(dmz, zone, bio);
419 		break;
420 	case REQ_OP_WRITE:
421 		ret = dmz_handle_write(dmz, zone, bio);
422 		break;
423 	case REQ_OP_DISCARD:
424 	case REQ_OP_WRITE_ZEROES:
425 		ret = dmz_handle_discard(dmz, zone, bio);
426 		break;
427 	default:
428 		dmz_dev_err(dmz->dev, "Unsupported BIO operation 0x%x",
429 			    bio_op(bio));
430 		ret = -EIO;
431 	}
432 
433 	/*
434 	 * Release the chunk mapping. This will check that the mapping
435 	 * is still valid, that is, that the zone used still has valid blocks.
436 	 */
437 	if (zone)
438 		dmz_put_chunk_mapping(zmd, zone);
439 out:
440 	dmz_bio_endio(bio, errno_to_blk_status(ret));
441 
442 	dmz_unlock_metadata(zmd);
443 }
444 
445 /*
446  * Increment a chunk reference counter.
447  */
448 static inline void dmz_get_chunk_work(struct dm_chunk_work *cw)
449 {
450 	refcount_inc(&cw->refcount);
451 }
452 
453 /*
454  * Decrement a chunk work reference count and
455  * free it if it becomes 0.
456  */
457 static void dmz_put_chunk_work(struct dm_chunk_work *cw)
458 {
459 	if (refcount_dec_and_test(&cw->refcount)) {
460 		WARN_ON(!bio_list_empty(&cw->bio_list));
461 		radix_tree_delete(&cw->target->chunk_rxtree, cw->chunk);
462 		kfree(cw);
463 	}
464 }
465 
466 /*
467  * Chunk BIO work function.
468  */
469 static void dmz_chunk_work(struct work_struct *work)
470 {
471 	struct dm_chunk_work *cw = container_of(work, struct dm_chunk_work, work);
472 	struct dmz_target *dmz = cw->target;
473 	struct bio *bio;
474 
475 	mutex_lock(&dmz->chunk_lock);
476 
477 	/* Process the chunk BIOs */
478 	while ((bio = bio_list_pop(&cw->bio_list))) {
479 		mutex_unlock(&dmz->chunk_lock);
480 		dmz_handle_bio(dmz, cw, bio);
481 		mutex_lock(&dmz->chunk_lock);
482 		dmz_put_chunk_work(cw);
483 	}
484 
485 	/* Queueing the work incremented the work refcount */
486 	dmz_put_chunk_work(cw);
487 
488 	mutex_unlock(&dmz->chunk_lock);
489 }
490 
491 /*
492  * Flush work.
493  */
494 static void dmz_flush_work(struct work_struct *work)
495 {
496 	struct dmz_target *dmz = container_of(work, struct dmz_target, flush_work.work);
497 	struct bio *bio;
498 	int ret;
499 
500 	/* Flush dirty metadata blocks */
501 	ret = dmz_flush_metadata(dmz->metadata);
502 	if (ret)
503 		dmz_dev_debug(dmz->dev, "Metadata flush failed, rc=%d\n", ret);
504 
505 	/* Process queued flush requests */
506 	while (1) {
507 		spin_lock(&dmz->flush_lock);
508 		bio = bio_list_pop(&dmz->flush_list);
509 		spin_unlock(&dmz->flush_lock);
510 
511 		if (!bio)
512 			break;
513 
514 		dmz_bio_endio(bio, errno_to_blk_status(ret));
515 	}
516 
517 	queue_delayed_work(dmz->flush_wq, &dmz->flush_work, DMZ_FLUSH_PERIOD);
518 }
519 
520 /*
521  * Get a chunk work and start it to process a new BIO.
522  * If the BIO chunk has no work yet, create one.
523  */
524 static int dmz_queue_chunk_work(struct dmz_target *dmz, struct bio *bio)
525 {
526 	unsigned int chunk = dmz_bio_chunk(dmz->dev, bio);
527 	struct dm_chunk_work *cw;
528 	int ret = 0;
529 
530 	mutex_lock(&dmz->chunk_lock);
531 
532 	/* Get the BIO chunk work. If one is not active yet, create one */
533 	cw = radix_tree_lookup(&dmz->chunk_rxtree, chunk);
534 	if (!cw) {
535 
536 		/* Create a new chunk work */
537 		cw = kmalloc(sizeof(struct dm_chunk_work), GFP_NOIO);
538 		if (unlikely(!cw)) {
539 			ret = -ENOMEM;
540 			goto out;
541 		}
542 
543 		INIT_WORK(&cw->work, dmz_chunk_work);
544 		refcount_set(&cw->refcount, 0);
545 		cw->target = dmz;
546 		cw->chunk = chunk;
547 		bio_list_init(&cw->bio_list);
548 
549 		ret = radix_tree_insert(&dmz->chunk_rxtree, chunk, cw);
550 		if (unlikely(ret)) {
551 			kfree(cw);
552 			goto out;
553 		}
554 	}
555 
556 	bio_list_add(&cw->bio_list, bio);
557 	dmz_get_chunk_work(cw);
558 
559 	dmz_reclaim_bio_acc(dmz->reclaim);
560 	if (queue_work(dmz->chunk_wq, &cw->work))
561 		dmz_get_chunk_work(cw);
562 out:
563 	mutex_unlock(&dmz->chunk_lock);
564 	return ret;
565 }
566 
567 /*
568  * Check the backing device availability. If it's on the way out,
569  * start failing I/O. Reclaim and metadata components also call this
570  * function to cleanly abort operation in the event of such failure.
571  */
572 bool dmz_bdev_is_dying(struct dmz_dev *dmz_dev)
573 {
574 	struct gendisk *disk;
575 
576 	if (!(dmz_dev->flags & DMZ_BDEV_DYING)) {
577 		disk = dmz_dev->bdev->bd_disk;
578 		if (blk_queue_dying(bdev_get_queue(dmz_dev->bdev))) {
579 			dmz_dev_warn(dmz_dev, "Backing device queue dying");
580 			dmz_dev->flags |= DMZ_BDEV_DYING;
581 		} else if (disk->fops->check_events) {
582 			if (disk->fops->check_events(disk, 0) &
583 					DISK_EVENT_MEDIA_CHANGE) {
584 				dmz_dev_warn(dmz_dev, "Backing device offline");
585 				dmz_dev->flags |= DMZ_BDEV_DYING;
586 			}
587 		}
588 	}
589 
590 	return dmz_dev->flags & DMZ_BDEV_DYING;
591 }
592 
593 /*
594  * Process a new BIO.
595  */
596 static int dmz_map(struct dm_target *ti, struct bio *bio)
597 {
598 	struct dmz_target *dmz = ti->private;
599 	struct dmz_dev *dev = dmz->dev;
600 	struct dmz_bioctx *bioctx = dm_per_bio_data(bio, sizeof(struct dmz_bioctx));
601 	sector_t sector = bio->bi_iter.bi_sector;
602 	unsigned int nr_sectors = bio_sectors(bio);
603 	sector_t chunk_sector;
604 	int ret;
605 
606 	if (dmz_bdev_is_dying(dmz->dev))
607 		return DM_MAPIO_KILL;
608 
609 	dmz_dev_debug(dev, "BIO op %d sector %llu + %u => chunk %llu, block %llu, %u blocks",
610 		      bio_op(bio), (unsigned long long)sector, nr_sectors,
611 		      (unsigned long long)dmz_bio_chunk(dmz->dev, bio),
612 		      (unsigned long long)dmz_chunk_block(dmz->dev, dmz_bio_block(bio)),
613 		      (unsigned int)dmz_bio_blocks(bio));
614 
615 	bio_set_dev(bio, dev->bdev);
616 
617 	if (!nr_sectors && bio_op(bio) != REQ_OP_WRITE)
618 		return DM_MAPIO_REMAPPED;
619 
620 	/* The BIO should be block aligned */
621 	if ((nr_sectors & DMZ_BLOCK_SECTORS_MASK) || (sector & DMZ_BLOCK_SECTORS_MASK))
622 		return DM_MAPIO_KILL;
623 
624 	/* Initialize the BIO context */
625 	bioctx->target = dmz;
626 	bioctx->zone = NULL;
627 	bioctx->bio = bio;
628 	refcount_set(&bioctx->ref, 1);
629 
630 	/* Set the BIO pending in the flush list */
631 	if (!nr_sectors && bio_op(bio) == REQ_OP_WRITE) {
632 		spin_lock(&dmz->flush_lock);
633 		bio_list_add(&dmz->flush_list, bio);
634 		spin_unlock(&dmz->flush_lock);
635 		mod_delayed_work(dmz->flush_wq, &dmz->flush_work, 0);
636 		return DM_MAPIO_SUBMITTED;
637 	}
638 
639 	/* Split zone BIOs to fit entirely into a zone */
640 	chunk_sector = sector & (dev->zone_nr_sectors - 1);
641 	if (chunk_sector + nr_sectors > dev->zone_nr_sectors)
642 		dm_accept_partial_bio(bio, dev->zone_nr_sectors - chunk_sector);
643 
644 	/* Now ready to handle this BIO */
645 	ret = dmz_queue_chunk_work(dmz, bio);
646 	if (ret) {
647 		dmz_dev_debug(dmz->dev,
648 			      "BIO op %d, can't process chunk %llu, err %i\n",
649 			      bio_op(bio), (u64)dmz_bio_chunk(dmz->dev, bio),
650 			      ret);
651 		return DM_MAPIO_REQUEUE;
652 	}
653 
654 	return DM_MAPIO_SUBMITTED;
655 }
656 
657 /*
658  * Get zoned device information.
659  */
660 static int dmz_get_zoned_device(struct dm_target *ti, char *path)
661 {
662 	struct dmz_target *dmz = ti->private;
663 	struct request_queue *q;
664 	struct dmz_dev *dev;
665 	sector_t aligned_capacity;
666 	int ret;
667 
668 	/* Get the target device */
669 	ret = dm_get_device(ti, path, dm_table_get_mode(ti->table), &dmz->ddev);
670 	if (ret) {
671 		ti->error = "Get target device failed";
672 		dmz->ddev = NULL;
673 		return ret;
674 	}
675 
676 	dev = kzalloc(sizeof(struct dmz_dev), GFP_KERNEL);
677 	if (!dev) {
678 		ret = -ENOMEM;
679 		goto err;
680 	}
681 
682 	dev->bdev = dmz->ddev->bdev;
683 	(void)bdevname(dev->bdev, dev->name);
684 
685 	if (bdev_zoned_model(dev->bdev) == BLK_ZONED_NONE) {
686 		ti->error = "Not a zoned block device";
687 		ret = -EINVAL;
688 		goto err;
689 	}
690 
691 	q = bdev_get_queue(dev->bdev);
692 	dev->capacity = i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT;
693 	aligned_capacity = dev->capacity &
694 				~((sector_t)blk_queue_zone_sectors(q) - 1);
695 	if (ti->begin ||
696 	    ((ti->len != dev->capacity) && (ti->len != aligned_capacity))) {
697 		ti->error = "Partial mapping not supported";
698 		ret = -EINVAL;
699 		goto err;
700 	}
701 
702 	dev->zone_nr_sectors = blk_queue_zone_sectors(q);
703 	dev->zone_nr_sectors_shift = ilog2(dev->zone_nr_sectors);
704 
705 	dev->zone_nr_blocks = dmz_sect2blk(dev->zone_nr_sectors);
706 	dev->zone_nr_blocks_shift = ilog2(dev->zone_nr_blocks);
707 
708 	dev->nr_zones = blkdev_nr_zones(dev->bdev);
709 
710 	dmz->dev = dev;
711 
712 	return 0;
713 err:
714 	dm_put_device(ti, dmz->ddev);
715 	kfree(dev);
716 
717 	return ret;
718 }
719 
720 /*
721  * Cleanup zoned device information.
722  */
723 static void dmz_put_zoned_device(struct dm_target *ti)
724 {
725 	struct dmz_target *dmz = ti->private;
726 
727 	dm_put_device(ti, dmz->ddev);
728 	kfree(dmz->dev);
729 	dmz->dev = NULL;
730 }
731 
732 /*
733  * Setup target.
734  */
735 static int dmz_ctr(struct dm_target *ti, unsigned int argc, char **argv)
736 {
737 	struct dmz_target *dmz;
738 	struct dmz_dev *dev;
739 	int ret;
740 
741 	/* Check arguments */
742 	if (argc != 1) {
743 		ti->error = "Invalid argument count";
744 		return -EINVAL;
745 	}
746 
747 	/* Allocate and initialize the target descriptor */
748 	dmz = kzalloc(sizeof(struct dmz_target), GFP_KERNEL);
749 	if (!dmz) {
750 		ti->error = "Unable to allocate the zoned target descriptor";
751 		return -ENOMEM;
752 	}
753 	ti->private = dmz;
754 
755 	/* Get the target zoned block device */
756 	ret = dmz_get_zoned_device(ti, argv[0]);
757 	if (ret) {
758 		dmz->ddev = NULL;
759 		goto err;
760 	}
761 
762 	/* Initialize metadata */
763 	dev = dmz->dev;
764 	ret = dmz_ctr_metadata(dev, &dmz->metadata);
765 	if (ret) {
766 		ti->error = "Metadata initialization failed";
767 		goto err_dev;
768 	}
769 
770 	/* Set target (no write same support) */
771 	ti->max_io_len = dev->zone_nr_sectors << 9;
772 	ti->num_flush_bios = 1;
773 	ti->num_discard_bios = 1;
774 	ti->num_write_zeroes_bios = 1;
775 	ti->per_io_data_size = sizeof(struct dmz_bioctx);
776 	ti->flush_supported = true;
777 	ti->discards_supported = true;
778 
779 	/* The exposed capacity is the number of chunks that can be mapped */
780 	ti->len = (sector_t)dmz_nr_chunks(dmz->metadata) << dev->zone_nr_sectors_shift;
781 
782 	/* Zone BIO */
783 	ret = bioset_init(&dmz->bio_set, DMZ_MIN_BIOS, 0, 0);
784 	if (ret) {
785 		ti->error = "Create BIO set failed";
786 		goto err_meta;
787 	}
788 
789 	/* Chunk BIO work */
790 	mutex_init(&dmz->chunk_lock);
791 	INIT_RADIX_TREE(&dmz->chunk_rxtree, GFP_NOIO);
792 	dmz->chunk_wq = alloc_workqueue("dmz_cwq_%s", WQ_MEM_RECLAIM | WQ_UNBOUND,
793 					0, dev->name);
794 	if (!dmz->chunk_wq) {
795 		ti->error = "Create chunk workqueue failed";
796 		ret = -ENOMEM;
797 		goto err_bio;
798 	}
799 
800 	/* Flush work */
801 	spin_lock_init(&dmz->flush_lock);
802 	bio_list_init(&dmz->flush_list);
803 	INIT_DELAYED_WORK(&dmz->flush_work, dmz_flush_work);
804 	dmz->flush_wq = alloc_ordered_workqueue("dmz_fwq_%s", WQ_MEM_RECLAIM,
805 						dev->name);
806 	if (!dmz->flush_wq) {
807 		ti->error = "Create flush workqueue failed";
808 		ret = -ENOMEM;
809 		goto err_cwq;
810 	}
811 	mod_delayed_work(dmz->flush_wq, &dmz->flush_work, DMZ_FLUSH_PERIOD);
812 
813 	/* Initialize reclaim */
814 	ret = dmz_ctr_reclaim(dev, dmz->metadata, &dmz->reclaim);
815 	if (ret) {
816 		ti->error = "Zone reclaim initialization failed";
817 		goto err_fwq;
818 	}
819 
820 	dmz_dev_info(dev, "Target device: %llu 512-byte logical sectors (%llu blocks)",
821 		     (unsigned long long)ti->len,
822 		     (unsigned long long)dmz_sect2blk(ti->len));
823 
824 	return 0;
825 err_fwq:
826 	destroy_workqueue(dmz->flush_wq);
827 err_cwq:
828 	destroy_workqueue(dmz->chunk_wq);
829 err_bio:
830 	mutex_destroy(&dmz->chunk_lock);
831 	bioset_exit(&dmz->bio_set);
832 err_meta:
833 	dmz_dtr_metadata(dmz->metadata);
834 err_dev:
835 	dmz_put_zoned_device(ti);
836 err:
837 	kfree(dmz);
838 
839 	return ret;
840 }
841 
842 /*
843  * Cleanup target.
844  */
845 static void dmz_dtr(struct dm_target *ti)
846 {
847 	struct dmz_target *dmz = ti->private;
848 
849 	flush_workqueue(dmz->chunk_wq);
850 	destroy_workqueue(dmz->chunk_wq);
851 
852 	dmz_dtr_reclaim(dmz->reclaim);
853 
854 	cancel_delayed_work_sync(&dmz->flush_work);
855 	destroy_workqueue(dmz->flush_wq);
856 
857 	(void) dmz_flush_metadata(dmz->metadata);
858 
859 	dmz_dtr_metadata(dmz->metadata);
860 
861 	bioset_exit(&dmz->bio_set);
862 
863 	dmz_put_zoned_device(ti);
864 
865 	mutex_destroy(&dmz->chunk_lock);
866 
867 	kfree(dmz);
868 }
869 
870 /*
871  * Setup target request queue limits.
872  */
873 static void dmz_io_hints(struct dm_target *ti, struct queue_limits *limits)
874 {
875 	struct dmz_target *dmz = ti->private;
876 	unsigned int chunk_sectors = dmz->dev->zone_nr_sectors;
877 
878 	limits->logical_block_size = DMZ_BLOCK_SIZE;
879 	limits->physical_block_size = DMZ_BLOCK_SIZE;
880 
881 	blk_limits_io_min(limits, DMZ_BLOCK_SIZE);
882 	blk_limits_io_opt(limits, DMZ_BLOCK_SIZE);
883 
884 	limits->discard_alignment = DMZ_BLOCK_SIZE;
885 	limits->discard_granularity = DMZ_BLOCK_SIZE;
886 	limits->max_discard_sectors = chunk_sectors;
887 	limits->max_hw_discard_sectors = chunk_sectors;
888 	limits->max_write_zeroes_sectors = chunk_sectors;
889 
890 	/* FS hint to try to align to the device zone size */
891 	limits->chunk_sectors = chunk_sectors;
892 	limits->max_sectors = chunk_sectors;
893 
894 	/* We are exposing a drive-managed zoned block device */
895 	limits->zoned = BLK_ZONED_NONE;
896 }
897 
898 /*
899  * Pass on ioctl to the backend device.
900  */
901 static int dmz_prepare_ioctl(struct dm_target *ti, struct block_device **bdev)
902 {
903 	struct dmz_target *dmz = ti->private;
904 
905 	if (dmz_bdev_is_dying(dmz->dev))
906 		return -ENODEV;
907 
908 	*bdev = dmz->dev->bdev;
909 
910 	return 0;
911 }
912 
913 /*
914  * Stop works on suspend.
915  */
916 static void dmz_suspend(struct dm_target *ti)
917 {
918 	struct dmz_target *dmz = ti->private;
919 
920 	flush_workqueue(dmz->chunk_wq);
921 	dmz_suspend_reclaim(dmz->reclaim);
922 	cancel_delayed_work_sync(&dmz->flush_work);
923 }
924 
925 /*
926  * Restart works on resume or if suspend failed.
927  */
928 static void dmz_resume(struct dm_target *ti)
929 {
930 	struct dmz_target *dmz = ti->private;
931 
932 	queue_delayed_work(dmz->flush_wq, &dmz->flush_work, DMZ_FLUSH_PERIOD);
933 	dmz_resume_reclaim(dmz->reclaim);
934 }
935 
936 static int dmz_iterate_devices(struct dm_target *ti,
937 			       iterate_devices_callout_fn fn, void *data)
938 {
939 	struct dmz_target *dmz = ti->private;
940 	struct dmz_dev *dev = dmz->dev;
941 	sector_t capacity = dev->capacity & ~(dev->zone_nr_sectors - 1);
942 
943 	return fn(ti, dmz->ddev, 0, capacity, data);
944 }
945 
946 static struct target_type dmz_type = {
947 	.name		 = "zoned",
948 	.version	 = {1, 0, 0},
949 	.features	 = DM_TARGET_SINGLETON | DM_TARGET_ZONED_HM,
950 	.module		 = THIS_MODULE,
951 	.ctr		 = dmz_ctr,
952 	.dtr		 = dmz_dtr,
953 	.map		 = dmz_map,
954 	.io_hints	 = dmz_io_hints,
955 	.prepare_ioctl	 = dmz_prepare_ioctl,
956 	.postsuspend	 = dmz_suspend,
957 	.resume		 = dmz_resume,
958 	.iterate_devices = dmz_iterate_devices,
959 };
960 
961 static int __init dmz_init(void)
962 {
963 	return dm_register_target(&dmz_type);
964 }
965 
966 static void __exit dmz_exit(void)
967 {
968 	dm_unregister_target(&dmz_type);
969 }
970 
971 module_init(dmz_init);
972 module_exit(dmz_exit);
973 
974 MODULE_DESCRIPTION(DM_NAME " target for zoned block devices");
975 MODULE_AUTHOR("Damien Le Moal <damien.lemoal@wdc.com>");
976 MODULE_LICENSE("GPL");
977