1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2021 Western Digital Corporation or its affiliates.
6 #include <linux/blkdev.h>
8 #include <linux/sched/mm.h>
9 #include <linux/slab.h>
13 #define DM_MSG_PREFIX "zone"
15 #define DM_ZONE_INVALID_WP_OFST UINT_MAX
18 * For internal zone reports bypassing the top BIO submission path.
20 static 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)
24 struct gendisk *disk = md->disk;
26 struct dm_report_zones_args args = {
27 .next_sector = sector,
33 struct dm_target *tgt;
35 tgt = dm_table_find_target(t, args.next_sector);
36 if (WARN_ON_ONCE(!tgt->type->report_zones))
40 ret = tgt->type->report_zones(tgt, &args,
41 nr_zones - args.zone_idx);
44 } while (args.zone_idx < nr_zones &&
45 args.next_sector < get_capacity(disk));
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().
55 int dm_blk_report_zones(struct gendisk *disk, sector_t sector,
56 unsigned int nr_zones, report_zones_cb cb, void *data)
58 struct mapped_device *md = disk->private_data;
62 if (dm_suspended_md(md))
65 map = dm_get_live_table(md, &srcu_idx);
69 ret = dm_blk_do_report_zones(md, map, sector, nr_zones, cb, data);
71 dm_put_live_table(md, srcu_idx);
76 static int dm_report_zones_cb(struct blk_zone *zone, unsigned int idx,
79 struct dm_report_zones_args *args = data;
80 sector_t sector_diff = args->tgt->begin - args->start;
83 * Ignore zones beyond the target range.
85 if (zone->start >= args->start + args->tgt->len)
89 * Remap the start sector and write pointer position of the zone
90 * to match its position in the target range.
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;
99 zone->wp += sector_diff;
102 args->next_sector = zone->start + zone->len;
103 return args->orig_cb(zone, args->zone_idx++, args->orig_data);
107 * Helper for drivers of zoned targets to implement struct target_type
108 * report_zones operation.
110 int dm_report_zones(struct block_device *bdev, sector_t start, sector_t sector,
111 struct dm_report_zones_args *args, unsigned int nr_zones)
114 * Set the target mapping start sector first so that
115 * dm_report_zones_cb() can correctly remap zone information.
119 return blkdev_report_zones(bdev, sector, nr_zones,
120 dm_report_zones_cb, args);
122 EXPORT_SYMBOL_GPL(dm_report_zones);
124 bool dm_is_zone_write(struct mapped_device *md, struct bio *bio)
126 struct request_queue *q = md->queue;
128 if (!blk_queue_is_zoned(q))
131 switch (bio_op(bio)) {
132 case REQ_OP_WRITE_ZEROES:
134 return !op_is_flush(bio->bi_opf) && bio_sectors(bio);
140 void dm_cleanup_zoned_dev(struct mapped_device *md)
142 struct request_queue *q = md->queue;
145 kfree(q->conv_zones_bitmap);
146 q->conv_zones_bitmap = NULL;
147 kfree(q->seq_zones_wlock);
148 q->seq_zones_wlock = NULL;
151 kvfree(md->zwp_offset);
152 md->zwp_offset = NULL;
156 static unsigned int dm_get_zone_wp_offset(struct blk_zone *zone)
158 switch (zone->cond) {
159 case BLK_ZONE_COND_IMP_OPEN:
160 case BLK_ZONE_COND_EXP_OPEN:
161 case BLK_ZONE_COND_CLOSED:
162 return zone->wp - zone->start;
163 case BLK_ZONE_COND_FULL:
165 case BLK_ZONE_COND_EMPTY:
166 case BLK_ZONE_COND_NOT_WP:
167 case BLK_ZONE_COND_OFFLINE:
168 case BLK_ZONE_COND_READONLY:
171 * Conventional, offline and read-only zones do not have a valid
172 * write pointer. Use 0 as for an empty zone.
178 static int dm_zone_revalidate_cb(struct blk_zone *zone, unsigned int idx,
181 struct mapped_device *md = data;
182 struct request_queue *q = md->queue;
184 switch (zone->type) {
185 case BLK_ZONE_TYPE_CONVENTIONAL:
186 if (!q->conv_zones_bitmap) {
187 q->conv_zones_bitmap =
188 kcalloc(BITS_TO_LONGS(q->nr_zones),
189 sizeof(unsigned long), GFP_NOIO);
190 if (!q->conv_zones_bitmap)
193 set_bit(idx, q->conv_zones_bitmap);
195 case BLK_ZONE_TYPE_SEQWRITE_REQ:
196 case BLK_ZONE_TYPE_SEQWRITE_PREF:
197 if (!q->seq_zones_wlock) {
199 kcalloc(BITS_TO_LONGS(q->nr_zones),
200 sizeof(unsigned long), GFP_NOIO);
201 if (!q->seq_zones_wlock)
204 if (!md->zwp_offset) {
206 kvcalloc(q->nr_zones, sizeof(unsigned int),
211 md->zwp_offset[idx] = dm_get_zone_wp_offset(zone);
215 DMERR("Invalid zone type 0x%x at sectors %llu",
216 (int)zone->type, zone->start);
224 * Revalidate the zones of a mapped device to initialize resource necessary
225 * for zone append emulation. Note that we cannot simply use the block layer
226 * blk_revalidate_disk_zones() function here as the mapped device is suspended
227 * (this is called from __bind() context).
229 static int dm_revalidate_zones(struct mapped_device *md, struct dm_table *t)
231 struct request_queue *q = md->queue;
232 unsigned int noio_flag;
236 * Check if something changed. If yes, cleanup the current resources
237 * and reallocate everything.
239 if (!q->nr_zones || q->nr_zones != md->nr_zones)
240 dm_cleanup_zoned_dev(md);
245 * Scan all zones to initialize everything. Ensure that all vmalloc
246 * operations in this context are done as if GFP_NOIO was specified.
248 noio_flag = memalloc_noio_save();
249 ret = dm_blk_do_report_zones(md, t, 0, q->nr_zones,
250 dm_zone_revalidate_cb, md);
251 memalloc_noio_restore(noio_flag);
254 if (ret != q->nr_zones) {
259 md->nr_zones = q->nr_zones;
264 DMERR("Revalidate zones failed %d", ret);
265 dm_cleanup_zoned_dev(md);
269 static int device_not_zone_append_capable(struct dm_target *ti,
270 struct dm_dev *dev, sector_t start,
271 sector_t len, void *data)
273 return !blk_queue_is_zoned(bdev_get_queue(dev->bdev));
276 static bool dm_table_supports_zone_append(struct dm_table *t)
278 struct dm_target *ti;
281 for (i = 0; i < dm_table_get_num_targets(t); i++) {
282 ti = dm_table_get_target(t, i);
284 if (ti->emulate_zone_append)
287 if (!ti->type->iterate_devices ||
288 ti->type->iterate_devices(ti, device_not_zone_append_capable, NULL))
295 int dm_set_zones_restrictions(struct dm_table *t, struct request_queue *q)
297 struct mapped_device *md = t->md;
300 * For a zoned target, the number of zones should be updated for the
301 * correct value to be exposed in sysfs queue/nr_zones.
303 WARN_ON_ONCE(queue_is_mq(q));
304 q->nr_zones = blkdev_nr_zones(md->disk);
306 /* Check if zone append is natively supported */
307 if (dm_table_supports_zone_append(t)) {
308 clear_bit(DMF_EMULATE_ZONE_APPEND, &md->flags);
309 dm_cleanup_zoned_dev(md);
314 * Mark the mapped device as needing zone append emulation and
315 * initialize the emulation resources once the capacity is set.
317 set_bit(DMF_EMULATE_ZONE_APPEND, &md->flags);
318 if (!get_capacity(md->disk))
321 return dm_revalidate_zones(md, t);
324 static int dm_update_zone_wp_offset_cb(struct blk_zone *zone, unsigned int idx,
327 unsigned int *wp_offset = data;
329 *wp_offset = dm_get_zone_wp_offset(zone);
334 static int dm_update_zone_wp_offset(struct mapped_device *md, unsigned int zno,
335 unsigned int *wp_ofst)
337 sector_t sector = zno * blk_queue_zone_sectors(md->queue);
338 unsigned int noio_flag;
342 t = dm_get_live_table(md, &srcu_idx);
347 * Ensure that all memory allocations in this context are done as if
348 * GFP_NOIO was specified.
350 noio_flag = memalloc_noio_save();
351 ret = dm_blk_do_report_zones(md, t, sector, 1,
352 dm_update_zone_wp_offset_cb, wp_ofst);
353 memalloc_noio_restore(noio_flag);
355 dm_put_live_table(md, srcu_idx);
363 struct orig_bio_details {
365 unsigned int nr_sectors;
369 * First phase of BIO mapping for targets with zone append emulation:
370 * check all BIO that change a zone writer pointer and change zone
371 * append operations into regular write operations.
373 static bool dm_zone_map_bio_begin(struct mapped_device *md,
374 unsigned int zno, struct bio *clone)
376 sector_t zsectors = blk_queue_zone_sectors(md->queue);
377 unsigned int zwp_offset = READ_ONCE(md->zwp_offset[zno]);
380 * If the target zone is in an error state, recover by inspecting the
381 * zone to get its current write pointer position. Note that since the
382 * target zone is already locked, a BIO issuing context should never
383 * see the zone write in the DM_ZONE_UPDATING_WP_OFST state.
385 if (zwp_offset == DM_ZONE_INVALID_WP_OFST) {
386 if (dm_update_zone_wp_offset(md, zno, &zwp_offset))
388 WRITE_ONCE(md->zwp_offset[zno], zwp_offset);
391 switch (bio_op(clone)) {
392 case REQ_OP_ZONE_RESET:
393 case REQ_OP_ZONE_FINISH:
395 case REQ_OP_WRITE_ZEROES:
397 /* Writes must be aligned to the zone write pointer */
398 if ((clone->bi_iter.bi_sector & (zsectors - 1)) != zwp_offset)
401 case REQ_OP_ZONE_APPEND:
403 * Change zone append operations into a non-mergeable regular
404 * writes directed at the current write pointer position of the
407 clone->bi_opf = REQ_OP_WRITE | REQ_NOMERGE |
408 (clone->bi_opf & (~REQ_OP_MASK));
409 clone->bi_iter.bi_sector += zwp_offset;
412 DMWARN_LIMIT("Invalid BIO operation");
416 /* Cannot write to a full zone */
417 if (zwp_offset >= zsectors)
424 * Second phase of BIO mapping for targets with zone append emulation:
425 * update the zone write pointer offset array to account for the additional
426 * data written to a zone. Note that at this point, the remapped clone BIO
427 * may already have completed, so we do not touch it.
429 static blk_status_t dm_zone_map_bio_end(struct mapped_device *md, unsigned int zno,
430 struct orig_bio_details *orig_bio_details,
431 unsigned int nr_sectors)
433 unsigned int zwp_offset = READ_ONCE(md->zwp_offset[zno]);
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;
439 /* Update the zone wp offset */
440 switch (orig_bio_details->op) {
441 case REQ_OP_ZONE_RESET:
442 WRITE_ONCE(md->zwp_offset[zno], 0);
444 case REQ_OP_ZONE_FINISH:
445 WRITE_ONCE(md->zwp_offset[zno],
446 blk_queue_zone_sectors(md->queue));
448 case REQ_OP_WRITE_ZEROES:
450 WRITE_ONCE(md->zwp_offset[zno], zwp_offset + nr_sectors);
452 case REQ_OP_ZONE_APPEND:
454 * Check that the target did not truncate the write operation
455 * emulating a zone append.
457 if (nr_sectors != orig_bio_details->nr_sectors) {
458 DMWARN_LIMIT("Truncated write for zone append");
459 return BLK_STS_IOERR;
461 WRITE_ONCE(md->zwp_offset[zno], zwp_offset + nr_sectors);
464 DMWARN_LIMIT("Invalid BIO operation");
465 return BLK_STS_IOERR;
469 static inline void dm_zone_lock(struct request_queue *q,
470 unsigned int zno, struct bio *clone)
472 if (WARN_ON_ONCE(bio_flagged(clone, BIO_ZONE_WRITE_LOCKED)))
475 wait_on_bit_lock_io(q->seq_zones_wlock, zno, TASK_UNINTERRUPTIBLE);
476 bio_set_flag(clone, BIO_ZONE_WRITE_LOCKED);
479 static inline void dm_zone_unlock(struct request_queue *q,
480 unsigned int zno, struct bio *clone)
482 if (!bio_flagged(clone, BIO_ZONE_WRITE_LOCKED))
485 WARN_ON_ONCE(!test_bit(zno, q->seq_zones_wlock));
486 clear_bit_unlock(zno, q->seq_zones_wlock);
487 smp_mb__after_atomic();
488 wake_up_bit(q->seq_zones_wlock, zno);
490 bio_clear_flag(clone, BIO_ZONE_WRITE_LOCKED);
493 static bool dm_need_zone_wp_tracking(struct bio *bio)
496 * Special processing is not needed for operations that do not need the
497 * zone write lock, that is, all operations that target conventional
498 * zones and all operations that do not modify directly a sequential
499 * zone write pointer.
501 if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
503 switch (bio_op(bio)) {
504 case REQ_OP_WRITE_ZEROES:
506 case REQ_OP_ZONE_RESET:
507 case REQ_OP_ZONE_FINISH:
508 case REQ_OP_ZONE_APPEND:
509 return bio_zone_is_seq(bio);
516 * Special IO mapping for targets needing zone append emulation.
518 int dm_zone_map_bio(struct dm_target_io *tio)
520 struct dm_io *io = tio->io;
521 struct dm_target *ti = tio->ti;
522 struct mapped_device *md = io->md;
523 struct request_queue *q = md->queue;
524 struct bio *clone = &tio->clone;
525 struct orig_bio_details orig_bio_details;
531 * IOs that do not change a zone write pointer do not need
532 * any additional special processing.
534 if (!dm_need_zone_wp_tracking(clone))
535 return ti->type->map(ti, clone);
537 /* Lock the target zone */
538 zno = bio_zone_no(clone);
539 dm_zone_lock(q, zno, clone);
541 orig_bio_details.nr_sectors = bio_sectors(clone);
542 orig_bio_details.op = bio_op(clone);
545 * Check that the bio and the target zone write pointer offset are
546 * both valid, and if the bio is a zone append, remap it to a write.
548 if (!dm_zone_map_bio_begin(md, zno, clone)) {
549 dm_zone_unlock(q, zno, clone);
550 return DM_MAPIO_KILL;
554 * The target map function may issue and complete the IO quickly.
555 * Take an extra reference on the IO to make sure it does disappear
556 * until we run dm_zone_map_bio_end().
558 dm_io_inc_pending(io);
560 /* Let the target do its work */
561 r = ti->type->map(ti, clone);
563 case DM_MAPIO_SUBMITTED:
565 * The target submitted the clone BIO. The target zone will
566 * be unlocked on completion of the clone.
568 sts = dm_zone_map_bio_end(md, zno, &orig_bio_details,
571 case DM_MAPIO_REMAPPED:
573 * The target only remapped the clone BIO. In case of error,
574 * unlock the target zone here as the clone will not be
577 sts = dm_zone_map_bio_end(md, zno, &orig_bio_details,
579 if (sts != BLK_STS_OK)
580 dm_zone_unlock(q, zno, clone);
582 case DM_MAPIO_REQUEUE:
585 dm_zone_unlock(q, zno, clone);
590 /* Drop the extra reference on the IO */
591 dm_io_dec_pending(io, sts);
593 if (sts != BLK_STS_OK)
594 return DM_MAPIO_KILL;
600 * IO completion callback called from clone_endio().
602 void dm_zone_endio(struct dm_io *io, struct bio *clone)
604 struct mapped_device *md = io->md;
605 struct request_queue *q = md->queue;
606 struct bio *orig_bio = io->orig_bio;
607 unsigned int zwp_offset;
611 * For targets that do not emulate zone append, we only need to
612 * handle native zone-append bios.
614 if (!dm_emulate_zone_append(md)) {
616 * Get the offset within the zone of the written sector
617 * and add that to the original bio sector position.
619 if (clone->bi_status == BLK_STS_OK &&
620 bio_op(clone) == REQ_OP_ZONE_APPEND) {
621 sector_t mask = (sector_t)blk_queue_zone_sectors(q) - 1;
623 orig_bio->bi_iter.bi_sector +=
624 clone->bi_iter.bi_sector & mask;
631 * For targets that do emulate zone append, if the clone BIO does not
632 * own the target zone write lock, we have nothing to do.
634 if (!bio_flagged(clone, BIO_ZONE_WRITE_LOCKED))
637 zno = bio_zone_no(orig_bio);
639 if (clone->bi_status != BLK_STS_OK) {
641 * BIOs that modify a zone write pointer may leave the zone
642 * in an unknown state in case of failure (e.g. the write
643 * pointer was only partially advanced). In this case, set
644 * the target zone write pointer as invalid unless it is
645 * already being updated.
647 WRITE_ONCE(md->zwp_offset[zno], DM_ZONE_INVALID_WP_OFST);
648 } else if (bio_op(orig_bio) == REQ_OP_ZONE_APPEND) {
650 * Get the written sector for zone append operation that were
651 * emulated using regular write operations.
653 zwp_offset = READ_ONCE(md->zwp_offset[zno]);
654 if (WARN_ON_ONCE(zwp_offset < bio_sectors(orig_bio)))
655 WRITE_ONCE(md->zwp_offset[zno],
656 DM_ZONE_INVALID_WP_OFST);
658 orig_bio->bi_iter.bi_sector +=
659 zwp_offset - bio_sectors(orig_bio);
662 dm_zone_unlock(q, zno, clone);