Linux Audio

Check our new training course

Yocto distribution development and maintenance

Need a Yocto distribution for your embedded project?
Loading...
v6.8
  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * Copyright (C) 2021 Western Digital Corporation or its affiliates.
  4 */
  5
  6#include <linux/blkdev.h>
  7#include <linux/mm.h>
  8#include <linux/sched/mm.h>
  9#include <linux/slab.h>
 10#include <linux/bitmap.h>
 11
 12#include "dm-core.h"
 13
 14#define DM_MSG_PREFIX "zone"
 15
 16#define DM_ZONE_INVALID_WP_OFST		UINT_MAX
 17
 18/*
 19 * For internal zone reports bypassing the top BIO submission path.
 20 */
 21static int dm_blk_do_report_zones(struct mapped_device *md, struct dm_table *t,
 22				  sector_t sector, unsigned int nr_zones,
 23				  report_zones_cb cb, void *data)
 24{
 25	struct gendisk *disk = md->disk;
 26	int ret;
 27	struct dm_report_zones_args args = {
 28		.next_sector = sector,
 29		.orig_data = data,
 30		.orig_cb = cb,
 31	};
 32
 33	do {
 34		struct dm_target *tgt;
 35
 36		tgt = dm_table_find_target(t, args.next_sector);
 37		if (WARN_ON_ONCE(!tgt->type->report_zones))
 38			return -EIO;
 39
 40		args.tgt = tgt;
 41		ret = tgt->type->report_zones(tgt, &args,
 42					      nr_zones - args.zone_idx);
 43		if (ret < 0)
 44			return ret;
 45	} while (args.zone_idx < nr_zones &&
 46		 args.next_sector < get_capacity(disk));
 47
 48	return args.zone_idx;
 49}
 50
 51/*
 52 * User facing dm device block device report zone operation. This calls the
 53 * report_zones operation for each target of a device table. This operation is
 54 * generally implemented by targets using dm_report_zones().
 55 */
 56int dm_blk_report_zones(struct gendisk *disk, sector_t sector,
 57			unsigned int nr_zones, report_zones_cb cb, void *data)
 58{
 59	struct mapped_device *md = disk->private_data;
 60	struct dm_table *map;
 61	int srcu_idx, ret;
 62
 63	if (dm_suspended_md(md))
 64		return -EAGAIN;
 65
 66	map = dm_get_live_table(md, &srcu_idx);
 67	if (!map)
 68		return -EIO;
 69
 70	ret = dm_blk_do_report_zones(md, map, sector, nr_zones, cb, data);
 71
 72	dm_put_live_table(md, srcu_idx);
 73
 74	return ret;
 75}
 76
 77static int dm_report_zones_cb(struct blk_zone *zone, unsigned int idx,
 78			      void *data)
 79{
 80	struct dm_report_zones_args *args = data;
 81	sector_t sector_diff = args->tgt->begin - args->start;
 82
 83	/*
 84	 * Ignore zones beyond the target range.
 85	 */
 86	if (zone->start >= args->start + args->tgt->len)
 87		return 0;
 88
 89	/*
 90	 * Remap the start sector and write pointer position of the zone
 91	 * to match its position in the target range.
 92	 */
 93	zone->start += sector_diff;
 94	if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) {
 95		if (zone->cond == BLK_ZONE_COND_FULL)
 96			zone->wp = zone->start + zone->len;
 97		else if (zone->cond == BLK_ZONE_COND_EMPTY)
 98			zone->wp = zone->start;
 99		else
100			zone->wp += sector_diff;
101	}
102
103	args->next_sector = zone->start + zone->len;
104	return args->orig_cb(zone, args->zone_idx++, args->orig_data);
105}
106
107/*
108 * Helper for drivers of zoned targets to implement struct target_type
109 * report_zones operation.
110 */
111int dm_report_zones(struct block_device *bdev, sector_t start, sector_t sector,
112		    struct dm_report_zones_args *args, unsigned int nr_zones)
113{
114	/*
115	 * Set the target mapping start sector first so that
116	 * dm_report_zones_cb() can correctly remap zone information.
117	 */
118	args->start = start;
119
120	return blkdev_report_zones(bdev, sector, nr_zones,
121				   dm_report_zones_cb, args);
122}
123EXPORT_SYMBOL_GPL(dm_report_zones);
124
125bool dm_is_zone_write(struct mapped_device *md, struct bio *bio)
126{
127	struct request_queue *q = md->queue;
128
129	if (!blk_queue_is_zoned(q))
130		return false;
131
132	switch (bio_op(bio)) {
133	case REQ_OP_WRITE_ZEROES:
 
134	case REQ_OP_WRITE:
135		return !op_is_flush(bio->bi_opf) && bio_sectors(bio);
136	default:
137		return false;
138	}
139}
140
141void dm_cleanup_zoned_dev(struct mapped_device *md)
142{
143	if (md->disk) {
144		bitmap_free(md->disk->conv_zones_bitmap);
145		md->disk->conv_zones_bitmap = NULL;
146		bitmap_free(md->disk->seq_zones_wlock);
147		md->disk->seq_zones_wlock = NULL;
 
 
148	}
149
150	kvfree(md->zwp_offset);
151	md->zwp_offset = NULL;
152	md->nr_zones = 0;
153}
154
155static unsigned int dm_get_zone_wp_offset(struct blk_zone *zone)
156{
157	switch (zone->cond) {
158	case BLK_ZONE_COND_IMP_OPEN:
159	case BLK_ZONE_COND_EXP_OPEN:
160	case BLK_ZONE_COND_CLOSED:
161		return zone->wp - zone->start;
162	case BLK_ZONE_COND_FULL:
163		return zone->len;
164	case BLK_ZONE_COND_EMPTY:
165	case BLK_ZONE_COND_NOT_WP:
166	case BLK_ZONE_COND_OFFLINE:
167	case BLK_ZONE_COND_READONLY:
168	default:
169		/*
170		 * Conventional, offline and read-only zones do not have a valid
171		 * write pointer. Use 0 as for an empty zone.
172		 */
173		return 0;
174	}
175}
176
177static int dm_zone_revalidate_cb(struct blk_zone *zone, unsigned int idx,
178				 void *data)
179{
180	struct mapped_device *md = data;
181	struct gendisk *disk = md->disk;
182
183	switch (zone->type) {
184	case BLK_ZONE_TYPE_CONVENTIONAL:
185		if (!disk->conv_zones_bitmap) {
186			disk->conv_zones_bitmap = bitmap_zalloc(disk->nr_zones,
187								GFP_NOIO);
188			if (!disk->conv_zones_bitmap)
 
189				return -ENOMEM;
190		}
191		set_bit(idx, disk->conv_zones_bitmap);
192		break;
193	case BLK_ZONE_TYPE_SEQWRITE_REQ:
194	case BLK_ZONE_TYPE_SEQWRITE_PREF:
195		if (!disk->seq_zones_wlock) {
196			disk->seq_zones_wlock = bitmap_zalloc(disk->nr_zones,
197							      GFP_NOIO);
198			if (!disk->seq_zones_wlock)
 
199				return -ENOMEM;
200		}
201		if (!md->zwp_offset) {
202			md->zwp_offset =
203				kvcalloc(disk->nr_zones, sizeof(unsigned int),
204					 GFP_KERNEL);
205			if (!md->zwp_offset)
206				return -ENOMEM;
207		}
208		md->zwp_offset[idx] = dm_get_zone_wp_offset(zone);
209
210		break;
211	default:
212		DMERR("Invalid zone type 0x%x at sectors %llu",
213		      (int)zone->type, zone->start);
214		return -ENODEV;
215	}
216
217	return 0;
218}
219
220/*
221 * Revalidate the zones of a mapped device to initialize resource necessary
222 * for zone append emulation. Note that we cannot simply use the block layer
223 * blk_revalidate_disk_zones() function here as the mapped device is suspended
224 * (this is called from __bind() context).
225 */
226static int dm_revalidate_zones(struct mapped_device *md, struct dm_table *t)
227{
228	struct gendisk *disk = md->disk;
229	unsigned int noio_flag;
230	int ret;
231
232	/*
233	 * Check if something changed. If yes, cleanup the current resources
234	 * and reallocate everything.
235	 */
236	if (!disk->nr_zones || disk->nr_zones != md->nr_zones)
237		dm_cleanup_zoned_dev(md);
238	if (md->nr_zones)
239		return 0;
240
241	/*
242	 * Scan all zones to initialize everything. Ensure that all vmalloc
243	 * operations in this context are done as if GFP_NOIO was specified.
244	 */
245	noio_flag = memalloc_noio_save();
246	ret = dm_blk_do_report_zones(md, t, 0, disk->nr_zones,
247				     dm_zone_revalidate_cb, md);
248	memalloc_noio_restore(noio_flag);
249	if (ret < 0)
250		goto err;
251	if (ret != disk->nr_zones) {
252		ret = -EIO;
253		goto err;
254	}
255
256	md->nr_zones = disk->nr_zones;
257
258	return 0;
259
260err:
261	DMERR("Revalidate zones failed %d", ret);
262	dm_cleanup_zoned_dev(md);
263	return ret;
264}
265
266static int device_not_zone_append_capable(struct dm_target *ti,
267					  struct dm_dev *dev, sector_t start,
268					  sector_t len, void *data)
269{
270	return !bdev_is_zoned(dev->bdev);
271}
272
273static bool dm_table_supports_zone_append(struct dm_table *t)
274{
275	for (unsigned int i = 0; i < t->num_targets; i++) {
276		struct dm_target *ti = dm_table_get_target(t, i);
 
 
 
277
278		if (ti->emulate_zone_append)
279			return false;
280
281		if (!ti->type->iterate_devices ||
282		    ti->type->iterate_devices(ti, device_not_zone_append_capable, NULL))
283			return false;
284	}
285
286	return true;
287}
288
289int dm_set_zones_restrictions(struct dm_table *t, struct request_queue *q)
290{
291	struct mapped_device *md = t->md;
292
293	/*
294	 * For a zoned target, the number of zones should be updated for the
295	 * correct value to be exposed in sysfs queue/nr_zones.
296	 */
297	WARN_ON_ONCE(queue_is_mq(q));
298	md->disk->nr_zones = bdev_nr_zones(md->disk->part0);
299
300	/* Check if zone append is natively supported */
301	if (dm_table_supports_zone_append(t)) {
302		clear_bit(DMF_EMULATE_ZONE_APPEND, &md->flags);
303		dm_cleanup_zoned_dev(md);
304		return 0;
305	}
306
307	/*
308	 * Mark the mapped device as needing zone append emulation and
309	 * initialize the emulation resources once the capacity is set.
310	 */
311	set_bit(DMF_EMULATE_ZONE_APPEND, &md->flags);
312	if (!get_capacity(md->disk))
313		return 0;
314
315	return dm_revalidate_zones(md, t);
316}
317
318static int dm_update_zone_wp_offset_cb(struct blk_zone *zone, unsigned int idx,
319				       void *data)
320{
321	unsigned int *wp_offset = data;
322
323	*wp_offset = dm_get_zone_wp_offset(zone);
324
325	return 0;
326}
327
328static int dm_update_zone_wp_offset(struct mapped_device *md, unsigned int zno,
329				    unsigned int *wp_ofst)
330{
331	sector_t sector = zno * bdev_zone_sectors(md->disk->part0);
332	unsigned int noio_flag;
333	struct dm_table *t;
334	int srcu_idx, ret;
335
336	t = dm_get_live_table(md, &srcu_idx);
337	if (!t)
338		return -EIO;
339
340	/*
341	 * Ensure that all memory allocations in this context are done as if
342	 * GFP_NOIO was specified.
343	 */
344	noio_flag = memalloc_noio_save();
345	ret = dm_blk_do_report_zones(md, t, sector, 1,
346				     dm_update_zone_wp_offset_cb, wp_ofst);
347	memalloc_noio_restore(noio_flag);
348
349	dm_put_live_table(md, srcu_idx);
350
351	if (ret != 1)
352		return -EIO;
353
354	return 0;
355}
356
357struct orig_bio_details {
358	enum req_op op;
359	unsigned int nr_sectors;
360};
361
362/*
363 * First phase of BIO mapping for targets with zone append emulation:
364 * check all BIO that change a zone writer pointer and change zone
365 * append operations into regular write operations.
366 */
367static bool dm_zone_map_bio_begin(struct mapped_device *md,
368				  unsigned int zno, struct bio *clone)
369{
370	sector_t zsectors = bdev_zone_sectors(md->disk->part0);
 
371	unsigned int zwp_offset = READ_ONCE(md->zwp_offset[zno]);
372
373	/*
374	 * If the target zone is in an error state, recover by inspecting the
375	 * zone to get its current write pointer position. Note that since the
376	 * target zone is already locked, a BIO issuing context should never
377	 * see the zone write in the DM_ZONE_UPDATING_WP_OFST state.
378	 */
379	if (zwp_offset == DM_ZONE_INVALID_WP_OFST) {
380		if (dm_update_zone_wp_offset(md, zno, &zwp_offset))
381			return false;
382		WRITE_ONCE(md->zwp_offset[zno], zwp_offset);
383	}
384
385	switch (bio_op(clone)) {
386	case REQ_OP_ZONE_RESET:
387	case REQ_OP_ZONE_FINISH:
388		return true;
389	case REQ_OP_WRITE_ZEROES:
 
390	case REQ_OP_WRITE:
391		/* Writes must be aligned to the zone write pointer */
392		if ((clone->bi_iter.bi_sector & (zsectors - 1)) != zwp_offset)
393			return false;
394		break;
395	case REQ_OP_ZONE_APPEND:
396		/*
397		 * Change zone append operations into a non-mergeable regular
398		 * writes directed at the current write pointer position of the
399		 * target zone.
400		 */
401		clone->bi_opf = REQ_OP_WRITE | REQ_NOMERGE |
402			(clone->bi_opf & (~REQ_OP_MASK));
403		clone->bi_iter.bi_sector += zwp_offset;
 
404		break;
405	default:
406		DMWARN_LIMIT("Invalid BIO operation");
407		return false;
408	}
409
410	/* Cannot write to a full zone */
411	if (zwp_offset >= zsectors)
412		return false;
413
414	return true;
415}
416
417/*
418 * Second phase of BIO mapping for targets with zone append emulation:
419 * update the zone write pointer offset array to account for the additional
420 * data written to a zone. Note that at this point, the remapped clone BIO
421 * may already have completed, so we do not touch it.
422 */
423static blk_status_t dm_zone_map_bio_end(struct mapped_device *md, unsigned int zno,
424					struct orig_bio_details *orig_bio_details,
425					unsigned int nr_sectors)
426{
 
427	unsigned int zwp_offset = READ_ONCE(md->zwp_offset[zno]);
428
429	/* The clone BIO may already have been completed and failed */
430	if (zwp_offset == DM_ZONE_INVALID_WP_OFST)
431		return BLK_STS_IOERR;
432
433	/* Update the zone wp offset */
434	switch (orig_bio_details->op) {
435	case REQ_OP_ZONE_RESET:
436		WRITE_ONCE(md->zwp_offset[zno], 0);
437		return BLK_STS_OK;
438	case REQ_OP_ZONE_FINISH:
439		WRITE_ONCE(md->zwp_offset[zno],
440			   bdev_zone_sectors(md->disk->part0));
441		return BLK_STS_OK;
442	case REQ_OP_WRITE_ZEROES:
 
443	case REQ_OP_WRITE:
444		WRITE_ONCE(md->zwp_offset[zno], zwp_offset + nr_sectors);
445		return BLK_STS_OK;
446	case REQ_OP_ZONE_APPEND:
447		/*
448		 * Check that the target did not truncate the write operation
449		 * emulating a zone append.
450		 */
451		if (nr_sectors != orig_bio_details->nr_sectors) {
452			DMWARN_LIMIT("Truncated write for zone append");
453			return BLK_STS_IOERR;
454		}
455		WRITE_ONCE(md->zwp_offset[zno], zwp_offset + nr_sectors);
456		return BLK_STS_OK;
457	default:
458		DMWARN_LIMIT("Invalid BIO operation");
459		return BLK_STS_IOERR;
460	}
461}
462
463static inline void dm_zone_lock(struct gendisk *disk, unsigned int zno,
464				struct bio *clone)
465{
466	if (WARN_ON_ONCE(bio_flagged(clone, BIO_ZONE_WRITE_LOCKED)))
467		return;
468
469	wait_on_bit_lock_io(disk->seq_zones_wlock, zno, TASK_UNINTERRUPTIBLE);
470	bio_set_flag(clone, BIO_ZONE_WRITE_LOCKED);
471}
472
473static inline void dm_zone_unlock(struct gendisk *disk, unsigned int zno,
474				  struct bio *clone)
475{
476	if (!bio_flagged(clone, BIO_ZONE_WRITE_LOCKED))
477		return;
478
479	WARN_ON_ONCE(!test_bit(zno, disk->seq_zones_wlock));
480	clear_bit_unlock(zno, disk->seq_zones_wlock);
481	smp_mb__after_atomic();
482	wake_up_bit(disk->seq_zones_wlock, zno);
483
484	bio_clear_flag(clone, BIO_ZONE_WRITE_LOCKED);
485}
486
487static bool dm_need_zone_wp_tracking(struct bio *bio)
488{
489	/*
490	 * Special processing is not needed for operations that do not need the
491	 * zone write lock, that is, all operations that target conventional
492	 * zones and all operations that do not modify directly a sequential
493	 * zone write pointer.
494	 */
495	if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
496		return false;
497	switch (bio_op(bio)) {
498	case REQ_OP_WRITE_ZEROES:
 
499	case REQ_OP_WRITE:
500	case REQ_OP_ZONE_RESET:
501	case REQ_OP_ZONE_FINISH:
502	case REQ_OP_ZONE_APPEND:
503		return bio_zone_is_seq(bio);
504	default:
505		return false;
506	}
507}
508
509/*
510 * Special IO mapping for targets needing zone append emulation.
511 */
512int dm_zone_map_bio(struct dm_target_io *tio)
513{
514	struct dm_io *io = tio->io;
515	struct dm_target *ti = tio->ti;
516	struct mapped_device *md = io->md;
 
 
517	struct bio *clone = &tio->clone;
518	struct orig_bio_details orig_bio_details;
519	unsigned int zno;
520	blk_status_t sts;
521	int r;
522
523	/*
524	 * IOs that do not change a zone write pointer do not need
525	 * any additional special processing.
526	 */
527	if (!dm_need_zone_wp_tracking(clone))
528		return ti->type->map(ti, clone);
529
530	/* Lock the target zone */
531	zno = bio_zone_no(clone);
532	dm_zone_lock(md->disk, zno, clone);
533
534	orig_bio_details.nr_sectors = bio_sectors(clone);
535	orig_bio_details.op = bio_op(clone);
536
537	/*
538	 * Check that the bio and the target zone write pointer offset are
539	 * both valid, and if the bio is a zone append, remap it to a write.
540	 */
541	if (!dm_zone_map_bio_begin(md, zno, clone)) {
542		dm_zone_unlock(md->disk, zno, clone);
543		return DM_MAPIO_KILL;
544	}
545
 
 
 
 
 
 
 
546	/* Let the target do its work */
547	r = ti->type->map(ti, clone);
548	switch (r) {
549	case DM_MAPIO_SUBMITTED:
550		/*
551		 * The target submitted the clone BIO. The target zone will
552		 * be unlocked on completion of the clone.
553		 */
554		sts = dm_zone_map_bio_end(md, zno, &orig_bio_details,
555					  *tio->len_ptr);
556		break;
557	case DM_MAPIO_REMAPPED:
558		/*
559		 * The target only remapped the clone BIO. In case of error,
560		 * unlock the target zone here as the clone will not be
561		 * submitted.
562		 */
563		sts = dm_zone_map_bio_end(md, zno, &orig_bio_details,
564					  *tio->len_ptr);
565		if (sts != BLK_STS_OK)
566			dm_zone_unlock(md->disk, zno, clone);
567		break;
568	case DM_MAPIO_REQUEUE:
569	case DM_MAPIO_KILL:
570	default:
571		dm_zone_unlock(md->disk, zno, clone);
572		sts = BLK_STS_IOERR;
573		break;
574	}
575
 
 
 
576	if (sts != BLK_STS_OK)
577		return DM_MAPIO_KILL;
578
579	return r;
580}
581
582/*
583 * IO completion callback called from clone_endio().
584 */
585void dm_zone_endio(struct dm_io *io, struct bio *clone)
586{
587	struct mapped_device *md = io->md;
588	struct gendisk *disk = md->disk;
589	struct bio *orig_bio = io->orig_bio;
590	unsigned int zwp_offset;
591	unsigned int zno;
592
593	/*
594	 * For targets that do not emulate zone append, we only need to
595	 * handle native zone-append bios.
596	 */
597	if (!dm_emulate_zone_append(md)) {
598		/*
599		 * Get the offset within the zone of the written sector
600		 * and add that to the original bio sector position.
601		 */
602		if (clone->bi_status == BLK_STS_OK &&
603		    bio_op(clone) == REQ_OP_ZONE_APPEND) {
604			sector_t mask =
605				(sector_t)bdev_zone_sectors(disk->part0) - 1;
606
607			orig_bio->bi_iter.bi_sector +=
608				clone->bi_iter.bi_sector & mask;
609		}
610
611		return;
612	}
613
614	/*
615	 * For targets that do emulate zone append, if the clone BIO does not
616	 * own the target zone write lock, we have nothing to do.
617	 */
618	if (!bio_flagged(clone, BIO_ZONE_WRITE_LOCKED))
619		return;
620
621	zno = bio_zone_no(orig_bio);
622
623	if (clone->bi_status != BLK_STS_OK) {
624		/*
625		 * BIOs that modify a zone write pointer may leave the zone
626		 * in an unknown state in case of failure (e.g. the write
627		 * pointer was only partially advanced). In this case, set
628		 * the target zone write pointer as invalid unless it is
629		 * already being updated.
630		 */
631		WRITE_ONCE(md->zwp_offset[zno], DM_ZONE_INVALID_WP_OFST);
632	} else if (bio_op(orig_bio) == REQ_OP_ZONE_APPEND) {
633		/*
634		 * Get the written sector for zone append operation that were
635		 * emulated using regular write operations.
636		 */
637		zwp_offset = READ_ONCE(md->zwp_offset[zno]);
638		if (WARN_ON_ONCE(zwp_offset < bio_sectors(orig_bio)))
639			WRITE_ONCE(md->zwp_offset[zno],
640				   DM_ZONE_INVALID_WP_OFST);
641		else
642			orig_bio->bi_iter.bi_sector +=
643				zwp_offset - bio_sectors(orig_bio);
644	}
645
646	dm_zone_unlock(disk, zno, clone);
647}
v5.14.15
  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 * Copyright (C) 2021 Western Digital Corporation or its affiliates.
  4 */
  5
  6#include <linux/blkdev.h>
  7#include <linux/mm.h>
  8#include <linux/sched/mm.h>
  9#include <linux/slab.h>
 
 10
 11#include "dm-core.h"
 12
 13#define DM_MSG_PREFIX "zone"
 14
 15#define DM_ZONE_INVALID_WP_OFST		UINT_MAX
 16
 17/*
 18 * For internal zone reports bypassing the top BIO submission path.
 19 */
 20static int dm_blk_do_report_zones(struct mapped_device *md, struct dm_table *t,
 21				  sector_t sector, unsigned int nr_zones,
 22				  report_zones_cb cb, void *data)
 23{
 24	struct gendisk *disk = md->disk;
 25	int ret;
 26	struct dm_report_zones_args args = {
 27		.next_sector = sector,
 28		.orig_data = data,
 29		.orig_cb = cb,
 30	};
 31
 32	do {
 33		struct dm_target *tgt;
 34
 35		tgt = dm_table_find_target(t, args.next_sector);
 36		if (WARN_ON_ONCE(!tgt->type->report_zones))
 37			return -EIO;
 38
 39		args.tgt = tgt;
 40		ret = tgt->type->report_zones(tgt, &args,
 41					      nr_zones - args.zone_idx);
 42		if (ret < 0)
 43			return ret;
 44	} while (args.zone_idx < nr_zones &&
 45		 args.next_sector < get_capacity(disk));
 46
 47	return args.zone_idx;
 48}
 49
 50/*
 51 * User facing dm device block device report zone operation. This calls the
 52 * report_zones operation for each target of a device table. This operation is
 53 * generally implemented by targets using dm_report_zones().
 54 */
 55int dm_blk_report_zones(struct gendisk *disk, sector_t sector,
 56			unsigned int nr_zones, report_zones_cb cb, void *data)
 57{
 58	struct mapped_device *md = disk->private_data;
 59	struct dm_table *map;
 60	int srcu_idx, ret;
 61
 62	if (dm_suspended_md(md))
 63		return -EAGAIN;
 64
 65	map = dm_get_live_table(md, &srcu_idx);
 66	if (!map)
 67		return -EIO;
 68
 69	ret = dm_blk_do_report_zones(md, map, sector, nr_zones, cb, data);
 70
 71	dm_put_live_table(md, srcu_idx);
 72
 73	return ret;
 74}
 75
 76static int dm_report_zones_cb(struct blk_zone *zone, unsigned int idx,
 77			      void *data)
 78{
 79	struct dm_report_zones_args *args = data;
 80	sector_t sector_diff = args->tgt->begin - args->start;
 81
 82	/*
 83	 * Ignore zones beyond the target range.
 84	 */
 85	if (zone->start >= args->start + args->tgt->len)
 86		return 0;
 87
 88	/*
 89	 * Remap the start sector and write pointer position of the zone
 90	 * to match its position in the target range.
 91	 */
 92	zone->start += sector_diff;
 93	if (zone->type != BLK_ZONE_TYPE_CONVENTIONAL) {
 94		if (zone->cond == BLK_ZONE_COND_FULL)
 95			zone->wp = zone->start + zone->len;
 96		else if (zone->cond == BLK_ZONE_COND_EMPTY)
 97			zone->wp = zone->start;
 98		else
 99			zone->wp += sector_diff;
100	}
101
102	args->next_sector = zone->start + zone->len;
103	return args->orig_cb(zone, args->zone_idx++, args->orig_data);
104}
105
106/*
107 * Helper for drivers of zoned targets to implement struct target_type
108 * report_zones operation.
109 */
110int dm_report_zones(struct block_device *bdev, sector_t start, sector_t sector,
111		    struct dm_report_zones_args *args, unsigned int nr_zones)
112{
113	/*
114	 * Set the target mapping start sector first so that
115	 * dm_report_zones_cb() can correctly remap zone information.
116	 */
117	args->start = start;
118
119	return blkdev_report_zones(bdev, sector, nr_zones,
120				   dm_report_zones_cb, args);
121}
122EXPORT_SYMBOL_GPL(dm_report_zones);
123
124bool dm_is_zone_write(struct mapped_device *md, struct bio *bio)
125{
126	struct request_queue *q = md->queue;
127
128	if (!blk_queue_is_zoned(q))
129		return false;
130
131	switch (bio_op(bio)) {
132	case REQ_OP_WRITE_ZEROES:
133	case REQ_OP_WRITE_SAME:
134	case REQ_OP_WRITE:
135		return !op_is_flush(bio->bi_opf) && bio_sectors(bio);
136	default:
137		return false;
138	}
139}
140
141void dm_cleanup_zoned_dev(struct mapped_device *md)
142{
143	struct request_queue *q = md->queue;
144
145	if (q) {
146		kfree(q->conv_zones_bitmap);
147		q->conv_zones_bitmap = NULL;
148		kfree(q->seq_zones_wlock);
149		q->seq_zones_wlock = NULL;
150	}
151
152	kvfree(md->zwp_offset);
153	md->zwp_offset = NULL;
154	md->nr_zones = 0;
155}
156
157static unsigned int dm_get_zone_wp_offset(struct blk_zone *zone)
158{
159	switch (zone->cond) {
160	case BLK_ZONE_COND_IMP_OPEN:
161	case BLK_ZONE_COND_EXP_OPEN:
162	case BLK_ZONE_COND_CLOSED:
163		return zone->wp - zone->start;
164	case BLK_ZONE_COND_FULL:
165		return zone->len;
166	case BLK_ZONE_COND_EMPTY:
167	case BLK_ZONE_COND_NOT_WP:
168	case BLK_ZONE_COND_OFFLINE:
169	case BLK_ZONE_COND_READONLY:
170	default:
171		/*
172		 * Conventional, offline and read-only zones do not have a valid
173		 * write pointer. Use 0 as for an empty zone.
174		 */
175		return 0;
176	}
177}
178
179static int dm_zone_revalidate_cb(struct blk_zone *zone, unsigned int idx,
180				 void *data)
181{
182	struct mapped_device *md = data;
183	struct request_queue *q = md->queue;
184
185	switch (zone->type) {
186	case BLK_ZONE_TYPE_CONVENTIONAL:
187		if (!q->conv_zones_bitmap) {
188			q->conv_zones_bitmap =
189				kcalloc(BITS_TO_LONGS(q->nr_zones),
190					sizeof(unsigned long), GFP_NOIO);
191			if (!q->conv_zones_bitmap)
192				return -ENOMEM;
193		}
194		set_bit(idx, q->conv_zones_bitmap);
195		break;
196	case BLK_ZONE_TYPE_SEQWRITE_REQ:
197	case BLK_ZONE_TYPE_SEQWRITE_PREF:
198		if (!q->seq_zones_wlock) {
199			q->seq_zones_wlock =
200				kcalloc(BITS_TO_LONGS(q->nr_zones),
201					sizeof(unsigned long), GFP_NOIO);
202			if (!q->seq_zones_wlock)
203				return -ENOMEM;
204		}
205		if (!md->zwp_offset) {
206			md->zwp_offset =
207				kvcalloc(q->nr_zones, sizeof(unsigned int),
208					 GFP_KERNEL);
209			if (!md->zwp_offset)
210				return -ENOMEM;
211		}
212		md->zwp_offset[idx] = dm_get_zone_wp_offset(zone);
213
214		break;
215	default:
216		DMERR("Invalid zone type 0x%x at sectors %llu",
217		      (int)zone->type, zone->start);
218		return -ENODEV;
219	}
220
221	return 0;
222}
223
224/*
225 * Revalidate the zones of a mapped device to initialize resource necessary
226 * for zone append emulation. Note that we cannot simply use the block layer
227 * blk_revalidate_disk_zones() function here as the mapped device is suspended
228 * (this is called from __bind() context).
229 */
230static int dm_revalidate_zones(struct mapped_device *md, struct dm_table *t)
231{
232	struct request_queue *q = md->queue;
233	unsigned int noio_flag;
234	int ret;
235
236	/*
237	 * Check if something changed. If yes, cleanup the current resources
238	 * and reallocate everything.
239	 */
240	if (!q->nr_zones || q->nr_zones != md->nr_zones)
241		dm_cleanup_zoned_dev(md);
242	if (md->nr_zones)
243		return 0;
244
245	/*
246	 * Scan all zones to initialize everything. Ensure that all vmalloc
247	 * operations in this context are done as if GFP_NOIO was specified.
248	 */
249	noio_flag = memalloc_noio_save();
250	ret = dm_blk_do_report_zones(md, t, 0, q->nr_zones,
251				     dm_zone_revalidate_cb, md);
252	memalloc_noio_restore(noio_flag);
253	if (ret < 0)
254		goto err;
255	if (ret != q->nr_zones) {
256		ret = -EIO;
257		goto err;
258	}
259
260	md->nr_zones = q->nr_zones;
261
262	return 0;
263
264err:
265	DMERR("Revalidate zones failed %d", ret);
266	dm_cleanup_zoned_dev(md);
267	return ret;
268}
269
270static int device_not_zone_append_capable(struct dm_target *ti,
271					  struct dm_dev *dev, sector_t start,
272					  sector_t len, void *data)
273{
274	return !blk_queue_is_zoned(bdev_get_queue(dev->bdev));
275}
276
277static bool dm_table_supports_zone_append(struct dm_table *t)
278{
279	struct dm_target *ti;
280	unsigned int i;
281
282	for (i = 0; i < dm_table_get_num_targets(t); i++) {
283		ti = dm_table_get_target(t, i);
284
285		if (ti->emulate_zone_append)
286			return false;
287
288		if (!ti->type->iterate_devices ||
289		    ti->type->iterate_devices(ti, device_not_zone_append_capable, NULL))
290			return false;
291	}
292
293	return true;
294}
295
296int dm_set_zones_restrictions(struct dm_table *t, struct request_queue *q)
297{
298	struct mapped_device *md = t->md;
299
300	/*
301	 * For a zoned target, the number of zones should be updated for the
302	 * correct value to be exposed in sysfs queue/nr_zones.
303	 */
304	WARN_ON_ONCE(queue_is_mq(q));
305	q->nr_zones = blkdev_nr_zones(md->disk);
306
307	/* Check if zone append is natively supported */
308	if (dm_table_supports_zone_append(t)) {
309		clear_bit(DMF_EMULATE_ZONE_APPEND, &md->flags);
310		dm_cleanup_zoned_dev(md);
311		return 0;
312	}
313
314	/*
315	 * Mark the mapped device as needing zone append emulation and
316	 * initialize the emulation resources once the capacity is set.
317	 */
318	set_bit(DMF_EMULATE_ZONE_APPEND, &md->flags);
319	if (!get_capacity(md->disk))
320		return 0;
321
322	return dm_revalidate_zones(md, t);
323}
324
325static int dm_update_zone_wp_offset_cb(struct blk_zone *zone, unsigned int idx,
326				       void *data)
327{
328	unsigned int *wp_offset = data;
329
330	*wp_offset = dm_get_zone_wp_offset(zone);
331
332	return 0;
333}
334
335static int dm_update_zone_wp_offset(struct mapped_device *md, unsigned int zno,
336				    unsigned int *wp_ofst)
337{
338	sector_t sector = zno * blk_queue_zone_sectors(md->queue);
339	unsigned int noio_flag;
340	struct dm_table *t;
341	int srcu_idx, ret;
342
343	t = dm_get_live_table(md, &srcu_idx);
344	if (!t)
345		return -EIO;
346
347	/*
348	 * Ensure that all memory allocations in this context are done as if
349	 * GFP_NOIO was specified.
350	 */
351	noio_flag = memalloc_noio_save();
352	ret = dm_blk_do_report_zones(md, t, sector, 1,
353				     dm_update_zone_wp_offset_cb, wp_ofst);
354	memalloc_noio_restore(noio_flag);
355
356	dm_put_live_table(md, srcu_idx);
357
358	if (ret != 1)
359		return -EIO;
360
361	return 0;
362}
363
 
 
 
 
 
364/*
365 * First phase of BIO mapping for targets with zone append emulation:
366 * check all BIO that change a zone writer pointer and change zone
367 * append operations into regular write operations.
368 */
369static bool dm_zone_map_bio_begin(struct mapped_device *md,
370				  struct bio *orig_bio, struct bio *clone)
371{
372	sector_t zsectors = blk_queue_zone_sectors(md->queue);
373	unsigned int zno = bio_zone_no(orig_bio);
374	unsigned int zwp_offset = READ_ONCE(md->zwp_offset[zno]);
375
376	/*
377	 * If the target zone is in an error state, recover by inspecting the
378	 * zone to get its current write pointer position. Note that since the
379	 * target zone is already locked, a BIO issuing context should never
380	 * see the zone write in the DM_ZONE_UPDATING_WP_OFST state.
381	 */
382	if (zwp_offset == DM_ZONE_INVALID_WP_OFST) {
383		if (dm_update_zone_wp_offset(md, zno, &zwp_offset))
384			return false;
385		WRITE_ONCE(md->zwp_offset[zno], zwp_offset);
386	}
387
388	switch (bio_op(orig_bio)) {
389	case REQ_OP_ZONE_RESET:
390	case REQ_OP_ZONE_FINISH:
391		return true;
392	case REQ_OP_WRITE_ZEROES:
393	case REQ_OP_WRITE_SAME:
394	case REQ_OP_WRITE:
395		/* Writes must be aligned to the zone write pointer */
396		if ((clone->bi_iter.bi_sector & (zsectors - 1)) != zwp_offset)
397			return false;
398		break;
399	case REQ_OP_ZONE_APPEND:
400		/*
401		 * Change zone append operations into a non-mergeable regular
402		 * writes directed at the current write pointer position of the
403		 * target zone.
404		 */
405		clone->bi_opf = REQ_OP_WRITE | REQ_NOMERGE |
406			(orig_bio->bi_opf & (~REQ_OP_MASK));
407		clone->bi_iter.bi_sector =
408			orig_bio->bi_iter.bi_sector + zwp_offset;
409		break;
410	default:
411		DMWARN_LIMIT("Invalid BIO operation");
412		return false;
413	}
414
415	/* Cannot write to a full zone */
416	if (zwp_offset >= zsectors)
417		return false;
418
419	return true;
420}
421
422/*
423 * Second phase of BIO mapping for targets with zone append emulation:
424 * update the zone write pointer offset array to account for the additional
425 * data written to a zone. Note that at this point, the remapped clone BIO
426 * may already have completed, so we do not touch it.
427 */
428static blk_status_t dm_zone_map_bio_end(struct mapped_device *md,
429					struct bio *orig_bio,
430					unsigned int nr_sectors)
431{
432	unsigned int zno = bio_zone_no(orig_bio);
433	unsigned int zwp_offset = READ_ONCE(md->zwp_offset[zno]);
434
435	/* The clone BIO may already have been completed and failed */
436	if (zwp_offset == DM_ZONE_INVALID_WP_OFST)
437		return BLK_STS_IOERR;
438
439	/* Update the zone wp offset */
440	switch (bio_op(orig_bio)) {
441	case REQ_OP_ZONE_RESET:
442		WRITE_ONCE(md->zwp_offset[zno], 0);
443		return BLK_STS_OK;
444	case REQ_OP_ZONE_FINISH:
445		WRITE_ONCE(md->zwp_offset[zno],
446			   blk_queue_zone_sectors(md->queue));
447		return BLK_STS_OK;
448	case REQ_OP_WRITE_ZEROES:
449	case REQ_OP_WRITE_SAME:
450	case REQ_OP_WRITE:
451		WRITE_ONCE(md->zwp_offset[zno], zwp_offset + nr_sectors);
452		return BLK_STS_OK;
453	case REQ_OP_ZONE_APPEND:
454		/*
455		 * Check that the target did not truncate the write operation
456		 * emulating a zone append.
457		 */
458		if (nr_sectors != bio_sectors(orig_bio)) {
459			DMWARN_LIMIT("Truncated write for zone append");
460			return BLK_STS_IOERR;
461		}
462		WRITE_ONCE(md->zwp_offset[zno], zwp_offset + nr_sectors);
463		return BLK_STS_OK;
464	default:
465		DMWARN_LIMIT("Invalid BIO operation");
466		return BLK_STS_IOERR;
467	}
468}
469
470static inline void dm_zone_lock(struct request_queue *q,
471				unsigned int zno, struct bio *clone)
472{
473	if (WARN_ON_ONCE(bio_flagged(clone, BIO_ZONE_WRITE_LOCKED)))
474		return;
475
476	wait_on_bit_lock_io(q->seq_zones_wlock, zno, TASK_UNINTERRUPTIBLE);
477	bio_set_flag(clone, BIO_ZONE_WRITE_LOCKED);
478}
479
480static inline void dm_zone_unlock(struct request_queue *q,
481				  unsigned int zno, struct bio *clone)
482{
483	if (!bio_flagged(clone, BIO_ZONE_WRITE_LOCKED))
484		return;
485
486	WARN_ON_ONCE(!test_bit(zno, q->seq_zones_wlock));
487	clear_bit_unlock(zno, q->seq_zones_wlock);
488	smp_mb__after_atomic();
489	wake_up_bit(q->seq_zones_wlock, zno);
490
491	bio_clear_flag(clone, BIO_ZONE_WRITE_LOCKED);
492}
493
494static bool dm_need_zone_wp_tracking(struct bio *orig_bio)
495{
496	/*
497	 * Special processing is not needed for operations that do not need the
498	 * zone write lock, that is, all operations that target conventional
499	 * zones and all operations that do not modify directly a sequential
500	 * zone write pointer.
501	 */
502	if (op_is_flush(orig_bio->bi_opf) && !bio_sectors(orig_bio))
503		return false;
504	switch (bio_op(orig_bio)) {
505	case REQ_OP_WRITE_ZEROES:
506	case REQ_OP_WRITE_SAME:
507	case REQ_OP_WRITE:
508	case REQ_OP_ZONE_RESET:
509	case REQ_OP_ZONE_FINISH:
510	case REQ_OP_ZONE_APPEND:
511		return bio_zone_is_seq(orig_bio);
512	default:
513		return false;
514	}
515}
516
517/*
518 * Special IO mapping for targets needing zone append emulation.
519 */
520int dm_zone_map_bio(struct dm_target_io *tio)
521{
522	struct dm_io *io = tio->io;
523	struct dm_target *ti = tio->ti;
524	struct mapped_device *md = io->md;
525	struct request_queue *q = md->queue;
526	struct bio *orig_bio = io->orig_bio;
527	struct bio *clone = &tio->clone;
 
528	unsigned int zno;
529	blk_status_t sts;
530	int r;
531
532	/*
533	 * IOs that do not change a zone write pointer do not need
534	 * any additional special processing.
535	 */
536	if (!dm_need_zone_wp_tracking(orig_bio))
537		return ti->type->map(ti, clone);
538
539	/* Lock the target zone */
540	zno = bio_zone_no(orig_bio);
541	dm_zone_lock(q, zno, clone);
 
 
 
542
543	/*
544	 * Check that the bio and the target zone write pointer offset are
545	 * both valid, and if the bio is a zone append, remap it to a write.
546	 */
547	if (!dm_zone_map_bio_begin(md, orig_bio, clone)) {
548		dm_zone_unlock(q, zno, clone);
549		return DM_MAPIO_KILL;
550	}
551
552	/*
553	 * The target map function may issue and complete the IO quickly.
554	 * Take an extra reference on the IO to make sure it does disappear
555	 * until we run dm_zone_map_bio_end().
556	 */
557	dm_io_inc_pending(io);
558
559	/* Let the target do its work */
560	r = ti->type->map(ti, clone);
561	switch (r) {
562	case DM_MAPIO_SUBMITTED:
563		/*
564		 * The target submitted the clone BIO. The target zone will
565		 * be unlocked on completion of the clone.
566		 */
567		sts = dm_zone_map_bio_end(md, orig_bio, *tio->len_ptr);
 
568		break;
569	case DM_MAPIO_REMAPPED:
570		/*
571		 * The target only remapped the clone BIO. In case of error,
572		 * unlock the target zone here as the clone will not be
573		 * submitted.
574		 */
575		sts = dm_zone_map_bio_end(md, orig_bio, *tio->len_ptr);
 
576		if (sts != BLK_STS_OK)
577			dm_zone_unlock(q, zno, clone);
578		break;
579	case DM_MAPIO_REQUEUE:
580	case DM_MAPIO_KILL:
581	default:
582		dm_zone_unlock(q, zno, clone);
583		sts = BLK_STS_IOERR;
584		break;
585	}
586
587	/* Drop the extra reference on the IO */
588	dm_io_dec_pending(io, sts);
589
590	if (sts != BLK_STS_OK)
591		return DM_MAPIO_KILL;
592
593	return r;
594}
595
596/*
597 * IO completion callback called from clone_endio().
598 */
599void dm_zone_endio(struct dm_io *io, struct bio *clone)
600{
601	struct mapped_device *md = io->md;
602	struct request_queue *q = md->queue;
603	struct bio *orig_bio = io->orig_bio;
604	unsigned int zwp_offset;
605	unsigned int zno;
606
607	/*
608	 * For targets that do not emulate zone append, we only need to
609	 * handle native zone-append bios.
610	 */
611	if (!dm_emulate_zone_append(md)) {
612		/*
613		 * Get the offset within the zone of the written sector
614		 * and add that to the original bio sector position.
615		 */
616		if (clone->bi_status == BLK_STS_OK &&
617		    bio_op(clone) == REQ_OP_ZONE_APPEND) {
618			sector_t mask = (sector_t)blk_queue_zone_sectors(q) - 1;
 
619
620			orig_bio->bi_iter.bi_sector +=
621				clone->bi_iter.bi_sector & mask;
622		}
623
624		return;
625	}
626
627	/*
628	 * For targets that do emulate zone append, if the clone BIO does not
629	 * own the target zone write lock, we have nothing to do.
630	 */
631	if (!bio_flagged(clone, BIO_ZONE_WRITE_LOCKED))
632		return;
633
634	zno = bio_zone_no(orig_bio);
635
636	if (clone->bi_status != BLK_STS_OK) {
637		/*
638		 * BIOs that modify a zone write pointer may leave the zone
639		 * in an unknown state in case of failure (e.g. the write
640		 * pointer was only partially advanced). In this case, set
641		 * the target zone write pointer as invalid unless it is
642		 * already being updated.
643		 */
644		WRITE_ONCE(md->zwp_offset[zno], DM_ZONE_INVALID_WP_OFST);
645	} else if (bio_op(orig_bio) == REQ_OP_ZONE_APPEND) {
646		/*
647		 * Get the written sector for zone append operation that were
648		 * emulated using regular write operations.
649		 */
650		zwp_offset = READ_ONCE(md->zwp_offset[zno]);
651		if (WARN_ON_ONCE(zwp_offset < bio_sectors(orig_bio)))
652			WRITE_ONCE(md->zwp_offset[zno],
653				   DM_ZONE_INVALID_WP_OFST);
654		else
655			orig_bio->bi_iter.bi_sector +=
656				zwp_offset - bio_sectors(orig_bio);
657	}
658
659	dm_zone_unlock(q, zno, clone);
660}