1 // SPDX-License-Identifier: GPL-2.0
3 #include <linux/bitops.h>
4 #include <linux/slab.h>
5 #include <linux/blkdev.h>
6 #include <linux/sched/mm.h>
7 #include <linux/atomic.h>
8 #include <linux/vmalloc.h>
12 #include "rcu-string.h"
14 #include "block-group.h"
15 #include "transaction.h"
16 #include "dev-replace.h"
17 #include "space-info.h"
19 /* Maximum number of zones to report per blkdev_report_zones() call */
20 #define BTRFS_REPORT_NR_ZONES 4096
21 /* Invalid allocation pointer value for missing devices */
22 #define WP_MISSING_DEV ((u64)-1)
23 /* Pseudo write pointer value for conventional zone */
24 #define WP_CONVENTIONAL ((u64)-2)
27 * Location of the first zone of superblock logging zone pairs.
29 * - primary superblock: 0B (zone 0)
30 * - first copy: 512G (zone starting at that offset)
31 * - second copy: 4T (zone starting at that offset)
33 #define BTRFS_SB_LOG_PRIMARY_OFFSET (0ULL)
34 #define BTRFS_SB_LOG_FIRST_OFFSET (512ULL * SZ_1G)
35 #define BTRFS_SB_LOG_SECOND_OFFSET (4096ULL * SZ_1G)
37 #define BTRFS_SB_LOG_FIRST_SHIFT const_ilog2(BTRFS_SB_LOG_FIRST_OFFSET)
38 #define BTRFS_SB_LOG_SECOND_SHIFT const_ilog2(BTRFS_SB_LOG_SECOND_OFFSET)
40 /* Number of superblock log zones */
41 #define BTRFS_NR_SB_LOG_ZONES 2
44 * Minimum of active zones we need:
46 * - BTRFS_SUPER_MIRROR_MAX zones for superblock mirrors
47 * - 3 zones to ensure at least one zone per SYSTEM, META and DATA block group
48 * - 1 zone for tree-log dedicated block group
49 * - 1 zone for relocation
51 #define BTRFS_MIN_ACTIVE_ZONES (BTRFS_SUPER_MIRROR_MAX + 5)
54 * Minimum / maximum supported zone size. Currently, SMR disks have a zone
55 * size of 256MiB, and we are expecting ZNS drives to be in the 1-4GiB range.
56 * We do not expect the zone size to become larger than 8GiB or smaller than
57 * 4MiB in the near future.
59 #define BTRFS_MAX_ZONE_SIZE SZ_8G
60 #define BTRFS_MIN_ZONE_SIZE SZ_4M
62 #define SUPER_INFO_SECTORS ((u64)BTRFS_SUPER_INFO_SIZE >> SECTOR_SHIFT)
64 static inline bool sb_zone_is_full(const struct blk_zone *zone)
66 return (zone->cond == BLK_ZONE_COND_FULL) ||
67 (zone->wp + SUPER_INFO_SECTORS > zone->start + zone->capacity);
70 static int copy_zone_info_cb(struct blk_zone *zone, unsigned int idx, void *data)
72 struct blk_zone *zones = data;
74 memcpy(&zones[idx], zone, sizeof(*zone));
79 static int sb_write_pointer(struct block_device *bdev, struct blk_zone *zones,
82 bool empty[BTRFS_NR_SB_LOG_ZONES];
83 bool full[BTRFS_NR_SB_LOG_ZONES];
87 for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) {
88 ASSERT(zones[i].type != BLK_ZONE_TYPE_CONVENTIONAL);
89 empty[i] = (zones[i].cond == BLK_ZONE_COND_EMPTY);
90 full[i] = sb_zone_is_full(&zones[i]);
94 * Possible states of log buffer zones
96 * Empty[0] In use[0] Full[0]
102 * *: Special case, no superblock is written
103 * 0: Use write pointer of zones[0]
104 * 1: Use write pointer of zones[1]
105 * C: Compare super blocks from zones[0] and zones[1], use the latest
106 * one determined by generation
110 if (empty[0] && empty[1]) {
111 /* Special case to distinguish no superblock to read */
112 *wp_ret = zones[0].start << SECTOR_SHIFT;
114 } else if (full[0] && full[1]) {
115 /* Compare two super blocks */
116 struct address_space *mapping = bdev->bd_inode->i_mapping;
117 struct page *page[BTRFS_NR_SB_LOG_ZONES];
118 struct btrfs_super_block *super[BTRFS_NR_SB_LOG_ZONES];
121 for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) {
122 u64 zone_end = (zones[i].start + zones[i].capacity) << SECTOR_SHIFT;
123 u64 bytenr = ALIGN_DOWN(zone_end, BTRFS_SUPER_INFO_SIZE) -
124 BTRFS_SUPER_INFO_SIZE;
126 page[i] = read_cache_page_gfp(mapping,
127 bytenr >> PAGE_SHIFT, GFP_NOFS);
128 if (IS_ERR(page[i])) {
130 btrfs_release_disk_super(super[0]);
131 return PTR_ERR(page[i]);
133 super[i] = page_address(page[i]);
136 if (btrfs_super_generation(super[0]) >
137 btrfs_super_generation(super[1]))
138 sector = zones[1].start;
140 sector = zones[0].start;
142 for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++)
143 btrfs_release_disk_super(super[i]);
144 } else if (!full[0] && (empty[1] || full[1])) {
145 sector = zones[0].wp;
146 } else if (full[0]) {
147 sector = zones[1].wp;
151 *wp_ret = sector << SECTOR_SHIFT;
156 * Get the first zone number of the superblock mirror
158 static inline u32 sb_zone_number(int shift, int mirror)
162 ASSERT(mirror < BTRFS_SUPER_MIRROR_MAX);
164 case 0: zone = 0; break;
165 case 1: zone = 1ULL << (BTRFS_SB_LOG_FIRST_SHIFT - shift); break;
166 case 2: zone = 1ULL << (BTRFS_SB_LOG_SECOND_SHIFT - shift); break;
169 ASSERT(zone <= U32_MAX);
174 static inline sector_t zone_start_sector(u32 zone_number,
175 struct block_device *bdev)
177 return (sector_t)zone_number << ilog2(bdev_zone_sectors(bdev));
180 static inline u64 zone_start_physical(u32 zone_number,
181 struct btrfs_zoned_device_info *zone_info)
183 return (u64)zone_number << zone_info->zone_size_shift;
187 * Emulate blkdev_report_zones() for a non-zoned device. It slices up the block
188 * device into static sized chunks and fake a conventional zone on each of
191 static int emulate_report_zones(struct btrfs_device *device, u64 pos,
192 struct blk_zone *zones, unsigned int nr_zones)
194 const sector_t zone_sectors = device->fs_info->zone_size >> SECTOR_SHIFT;
195 sector_t bdev_size = bdev_nr_sectors(device->bdev);
198 pos >>= SECTOR_SHIFT;
199 for (i = 0; i < nr_zones; i++) {
200 zones[i].start = i * zone_sectors + pos;
201 zones[i].len = zone_sectors;
202 zones[i].capacity = zone_sectors;
203 zones[i].wp = zones[i].start + zone_sectors;
204 zones[i].type = BLK_ZONE_TYPE_CONVENTIONAL;
205 zones[i].cond = BLK_ZONE_COND_NOT_WP;
207 if (zones[i].wp >= bdev_size) {
216 static int btrfs_get_dev_zones(struct btrfs_device *device, u64 pos,
217 struct blk_zone *zones, unsigned int *nr_zones)
219 struct btrfs_zoned_device_info *zinfo = device->zone_info;
226 if (!bdev_is_zoned(device->bdev)) {
227 ret = emulate_report_zones(device, pos, zones, *nr_zones);
233 if (zinfo->zone_cache) {
236 ASSERT(IS_ALIGNED(pos, zinfo->zone_size));
237 zno = pos >> zinfo->zone_size_shift;
239 * We cannot report zones beyond the zone end. So, it is OK to
240 * cap *nr_zones to at the end.
242 *nr_zones = min_t(u32, *nr_zones, zinfo->nr_zones - zno);
244 for (i = 0; i < *nr_zones; i++) {
245 struct blk_zone *zone_info;
247 zone_info = &zinfo->zone_cache[zno + i];
252 if (i == *nr_zones) {
253 /* Cache hit on all the zones */
254 memcpy(zones, zinfo->zone_cache + zno,
255 sizeof(*zinfo->zone_cache) * *nr_zones);
260 ret = blkdev_report_zones(device->bdev, pos >> SECTOR_SHIFT, *nr_zones,
261 copy_zone_info_cb, zones);
263 btrfs_err_in_rcu(device->fs_info,
264 "zoned: failed to read zone %llu on %s (devid %llu)",
265 pos, rcu_str_deref(device->name),
274 if (zinfo->zone_cache)
275 memcpy(zinfo->zone_cache + zno, zones,
276 sizeof(*zinfo->zone_cache) * *nr_zones);
281 /* The emulated zone size is determined from the size of device extent */
282 static int calculate_emulated_zone_size(struct btrfs_fs_info *fs_info)
284 struct btrfs_path *path;
285 struct btrfs_root *root = fs_info->dev_root;
286 struct btrfs_key key;
287 struct extent_buffer *leaf;
288 struct btrfs_dev_extent *dext;
292 key.type = BTRFS_DEV_EXTENT_KEY;
295 path = btrfs_alloc_path();
299 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
303 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
304 ret = btrfs_next_leaf(root, path);
307 /* No dev extents at all? Not good */
314 leaf = path->nodes[0];
315 dext = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent);
316 fs_info->zone_size = btrfs_dev_extent_length(leaf, dext);
320 btrfs_free_path(path);
325 int btrfs_get_dev_zone_info_all_devices(struct btrfs_fs_info *fs_info)
327 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
328 struct btrfs_device *device;
331 /* fs_info->zone_size might not set yet. Use the incomapt flag here. */
332 if (!btrfs_fs_incompat(fs_info, ZONED))
335 mutex_lock(&fs_devices->device_list_mutex);
336 list_for_each_entry(device, &fs_devices->devices, dev_list) {
337 /* We can skip reading of zone info for missing devices */
341 ret = btrfs_get_dev_zone_info(device, true);
345 mutex_unlock(&fs_devices->device_list_mutex);
350 int btrfs_get_dev_zone_info(struct btrfs_device *device, bool populate_cache)
352 struct btrfs_fs_info *fs_info = device->fs_info;
353 struct btrfs_zoned_device_info *zone_info = NULL;
354 struct block_device *bdev = device->bdev;
355 unsigned int max_active_zones;
356 unsigned int nactive;
359 struct blk_zone *zones = NULL;
360 unsigned int i, nreported = 0, nr_zones;
361 sector_t zone_sectors;
362 char *model, *emulated;
366 * Cannot use btrfs_is_zoned here, since fs_info::zone_size might not
369 if (!btrfs_fs_incompat(fs_info, ZONED))
372 if (device->zone_info)
375 zone_info = kzalloc(sizeof(*zone_info), GFP_KERNEL);
379 device->zone_info = zone_info;
381 if (!bdev_is_zoned(bdev)) {
382 if (!fs_info->zone_size) {
383 ret = calculate_emulated_zone_size(fs_info);
388 ASSERT(fs_info->zone_size);
389 zone_sectors = fs_info->zone_size >> SECTOR_SHIFT;
391 zone_sectors = bdev_zone_sectors(bdev);
394 /* Check if it's power of 2 (see is_power_of_2) */
395 ASSERT(zone_sectors != 0 && (zone_sectors & (zone_sectors - 1)) == 0);
396 zone_info->zone_size = zone_sectors << SECTOR_SHIFT;
398 /* We reject devices with a zone size larger than 8GB */
399 if (zone_info->zone_size > BTRFS_MAX_ZONE_SIZE) {
400 btrfs_err_in_rcu(fs_info,
401 "zoned: %s: zone size %llu larger than supported maximum %llu",
402 rcu_str_deref(device->name),
403 zone_info->zone_size, BTRFS_MAX_ZONE_SIZE);
406 } else if (zone_info->zone_size < BTRFS_MIN_ZONE_SIZE) {
407 btrfs_err_in_rcu(fs_info,
408 "zoned: %s: zone size %llu smaller than supported minimum %u",
409 rcu_str_deref(device->name),
410 zone_info->zone_size, BTRFS_MIN_ZONE_SIZE);
415 nr_sectors = bdev_nr_sectors(bdev);
416 zone_info->zone_size_shift = ilog2(zone_info->zone_size);
417 zone_info->nr_zones = nr_sectors >> ilog2(zone_sectors);
419 * We limit max_zone_append_size also by max_segments *
420 * PAGE_SIZE. Technically, we can have multiple pages per segment. But,
421 * since btrfs adds the pages one by one to a bio, and btrfs cannot
422 * increase the metadata reservation even if it increases the number of
423 * extents, it is safe to stick with the limit.
425 * With the zoned emulation, we can have non-zoned device on the zoned
426 * mode. In this case, we don't have a valid max zone append size. So,
427 * use max_segments * PAGE_SIZE as the pseudo max_zone_append_size.
429 if (bdev_is_zoned(bdev)) {
430 zone_info->max_zone_append_size = min_t(u64,
431 (u64)bdev_max_zone_append_sectors(bdev) << SECTOR_SHIFT,
432 (u64)bdev_max_segments(bdev) << PAGE_SHIFT);
434 zone_info->max_zone_append_size =
435 (u64)bdev_max_segments(bdev) << PAGE_SHIFT;
437 if (!IS_ALIGNED(nr_sectors, zone_sectors))
438 zone_info->nr_zones++;
440 max_active_zones = bdev_max_active_zones(bdev);
441 if (max_active_zones && max_active_zones < BTRFS_MIN_ACTIVE_ZONES) {
442 btrfs_err_in_rcu(fs_info,
443 "zoned: %s: max active zones %u is too small, need at least %u active zones",
444 rcu_str_deref(device->name), max_active_zones,
445 BTRFS_MIN_ACTIVE_ZONES);
449 zone_info->max_active_zones = max_active_zones;
451 zone_info->seq_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
452 if (!zone_info->seq_zones) {
457 zone_info->empty_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
458 if (!zone_info->empty_zones) {
463 zone_info->active_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
464 if (!zone_info->active_zones) {
469 zones = kvcalloc(BTRFS_REPORT_NR_ZONES, sizeof(struct blk_zone), GFP_KERNEL);
476 * Enable zone cache only for a zoned device. On a non-zoned device, we
477 * fill the zone info with emulated CONVENTIONAL zones, so no need to
480 if (populate_cache && bdev_is_zoned(device->bdev)) {
481 zone_info->zone_cache = vzalloc(sizeof(struct blk_zone) *
482 zone_info->nr_zones);
483 if (!zone_info->zone_cache) {
484 btrfs_err_in_rcu(device->fs_info,
485 "zoned: failed to allocate zone cache for %s",
486 rcu_str_deref(device->name));
494 while (sector < nr_sectors) {
495 nr_zones = BTRFS_REPORT_NR_ZONES;
496 ret = btrfs_get_dev_zones(device, sector << SECTOR_SHIFT, zones,
501 for (i = 0; i < nr_zones; i++) {
502 if (zones[i].type == BLK_ZONE_TYPE_SEQWRITE_REQ)
503 __set_bit(nreported, zone_info->seq_zones);
504 switch (zones[i].cond) {
505 case BLK_ZONE_COND_EMPTY:
506 __set_bit(nreported, zone_info->empty_zones);
508 case BLK_ZONE_COND_IMP_OPEN:
509 case BLK_ZONE_COND_EXP_OPEN:
510 case BLK_ZONE_COND_CLOSED:
511 __set_bit(nreported, zone_info->active_zones);
517 sector = zones[nr_zones - 1].start + zones[nr_zones - 1].len;
520 if (nreported != zone_info->nr_zones) {
521 btrfs_err_in_rcu(device->fs_info,
522 "inconsistent number of zones on %s (%u/%u)",
523 rcu_str_deref(device->name), nreported,
524 zone_info->nr_zones);
529 if (max_active_zones) {
530 if (nactive > max_active_zones) {
531 btrfs_err_in_rcu(device->fs_info,
532 "zoned: %u active zones on %s exceeds max_active_zones %u",
533 nactive, rcu_str_deref(device->name),
538 atomic_set(&zone_info->active_zones_left,
539 max_active_zones - nactive);
540 set_bit(BTRFS_FS_ACTIVE_ZONE_TRACKING, &fs_info->flags);
543 /* Validate superblock log */
544 nr_zones = BTRFS_NR_SB_LOG_ZONES;
545 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
548 int sb_pos = BTRFS_NR_SB_LOG_ZONES * i;
550 sb_zone = sb_zone_number(zone_info->zone_size_shift, i);
551 if (sb_zone + 1 >= zone_info->nr_zones)
554 ret = btrfs_get_dev_zones(device,
555 zone_start_physical(sb_zone, zone_info),
556 &zone_info->sb_zones[sb_pos],
561 if (nr_zones != BTRFS_NR_SB_LOG_ZONES) {
562 btrfs_err_in_rcu(device->fs_info,
563 "zoned: failed to read super block log zone info at devid %llu zone %u",
564 device->devid, sb_zone);
570 * If zones[0] is conventional, always use the beginning of the
571 * zone to record superblock. No need to validate in that case.
573 if (zone_info->sb_zones[BTRFS_NR_SB_LOG_ZONES * i].type ==
574 BLK_ZONE_TYPE_CONVENTIONAL)
577 ret = sb_write_pointer(device->bdev,
578 &zone_info->sb_zones[sb_pos], &sb_wp);
579 if (ret != -ENOENT && ret) {
580 btrfs_err_in_rcu(device->fs_info,
581 "zoned: super block log zone corrupted devid %llu zone %u",
582 device->devid, sb_zone);
591 switch (bdev_zoned_model(bdev)) {
593 model = "host-managed zoned";
597 model = "host-aware zoned";
602 emulated = "emulated ";
606 btrfs_err_in_rcu(fs_info, "zoned: unsupported model %d on %s",
607 bdev_zoned_model(bdev),
608 rcu_str_deref(device->name));
610 goto out_free_zone_info;
613 btrfs_info_in_rcu(fs_info,
614 "%s block device %s, %u %szones of %llu bytes",
615 model, rcu_str_deref(device->name), zone_info->nr_zones,
616 emulated, zone_info->zone_size);
623 btrfs_destroy_dev_zone_info(device);
628 void btrfs_destroy_dev_zone_info(struct btrfs_device *device)
630 struct btrfs_zoned_device_info *zone_info = device->zone_info;
635 bitmap_free(zone_info->active_zones);
636 bitmap_free(zone_info->seq_zones);
637 bitmap_free(zone_info->empty_zones);
638 vfree(zone_info->zone_cache);
640 device->zone_info = NULL;
643 struct btrfs_zoned_device_info *btrfs_clone_dev_zone_info(struct btrfs_device *orig_dev)
645 struct btrfs_zoned_device_info *zone_info;
647 zone_info = kmemdup(orig_dev->zone_info, sizeof(*zone_info), GFP_KERNEL);
651 zone_info->seq_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
652 if (!zone_info->seq_zones)
655 bitmap_copy(zone_info->seq_zones, orig_dev->zone_info->seq_zones,
656 zone_info->nr_zones);
658 zone_info->empty_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
659 if (!zone_info->empty_zones)
662 bitmap_copy(zone_info->empty_zones, orig_dev->zone_info->empty_zones,
663 zone_info->nr_zones);
665 zone_info->active_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
666 if (!zone_info->active_zones)
669 bitmap_copy(zone_info->active_zones, orig_dev->zone_info->active_zones,
670 zone_info->nr_zones);
671 zone_info->zone_cache = NULL;
676 bitmap_free(zone_info->seq_zones);
677 bitmap_free(zone_info->empty_zones);
678 bitmap_free(zone_info->active_zones);
683 int btrfs_get_dev_zone(struct btrfs_device *device, u64 pos,
684 struct blk_zone *zone)
686 unsigned int nr_zones = 1;
689 ret = btrfs_get_dev_zones(device, pos, zone, &nr_zones);
690 if (ret != 0 || !nr_zones)
691 return ret ? ret : -EIO;
696 static int btrfs_check_for_zoned_device(struct btrfs_fs_info *fs_info)
698 struct btrfs_device *device;
700 list_for_each_entry(device, &fs_info->fs_devices->devices, dev_list) {
702 bdev_zoned_model(device->bdev) == BLK_ZONED_HM) {
704 "zoned: mode not enabled but zoned device found: %pg",
713 int btrfs_check_zoned_mode(struct btrfs_fs_info *fs_info)
715 struct btrfs_device *device;
717 u64 max_zone_append_size = 0;
721 * Host-Managed devices can't be used without the ZONED flag. With the
722 * ZONED all devices can be used, using zone emulation if required.
724 if (!btrfs_fs_incompat(fs_info, ZONED))
725 return btrfs_check_for_zoned_device(fs_info);
727 list_for_each_entry(device, &fs_info->fs_devices->devices, dev_list) {
728 struct btrfs_zoned_device_info *zone_info = device->zone_info;
734 zone_size = zone_info->zone_size;
735 } else if (zone_info->zone_size != zone_size) {
737 "zoned: unequal block device zone sizes: have %llu found %llu",
738 zone_info->zone_size, zone_size);
741 if (!max_zone_append_size ||
742 (zone_info->max_zone_append_size &&
743 zone_info->max_zone_append_size < max_zone_append_size))
744 max_zone_append_size = zone_info->max_zone_append_size;
748 * stripe_size is always aligned to BTRFS_STRIPE_LEN in
749 * btrfs_create_chunk(). Since we want stripe_len == zone_size,
750 * check the alignment here.
752 if (!IS_ALIGNED(zone_size, BTRFS_STRIPE_LEN)) {
754 "zoned: zone size %llu not aligned to stripe %u",
755 zone_size, BTRFS_STRIPE_LEN);
759 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
760 btrfs_err(fs_info, "zoned: mixed block groups not supported");
764 fs_info->zone_size = zone_size;
765 fs_info->max_zone_append_size = ALIGN_DOWN(max_zone_append_size,
766 fs_info->sectorsize);
767 fs_info->fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_ZONED;
768 if (fs_info->max_zone_append_size < fs_info->max_extent_size)
769 fs_info->max_extent_size = fs_info->max_zone_append_size;
772 * Check mount options here, because we might change fs_info->zoned
773 * from fs_info->zone_size.
775 ret = btrfs_check_mountopts_zoned(fs_info);
779 btrfs_info(fs_info, "zoned mode enabled with zone size %llu", zone_size);
783 int btrfs_check_mountopts_zoned(struct btrfs_fs_info *info)
785 if (!btrfs_is_zoned(info))
789 * Space cache writing is not COWed. Disable that to avoid write errors
790 * in sequential zones.
792 if (btrfs_test_opt(info, SPACE_CACHE)) {
793 btrfs_err(info, "zoned: space cache v1 is not supported");
797 if (btrfs_test_opt(info, NODATACOW)) {
798 btrfs_err(info, "zoned: NODATACOW not supported");
805 static int sb_log_location(struct block_device *bdev, struct blk_zone *zones,
806 int rw, u64 *bytenr_ret)
811 if (zones[0].type == BLK_ZONE_TYPE_CONVENTIONAL) {
812 *bytenr_ret = zones[0].start << SECTOR_SHIFT;
816 ret = sb_write_pointer(bdev, zones, &wp);
817 if (ret != -ENOENT && ret < 0)
821 struct blk_zone *reset = NULL;
823 if (wp == zones[0].start << SECTOR_SHIFT)
825 else if (wp == zones[1].start << SECTOR_SHIFT)
828 if (reset && reset->cond != BLK_ZONE_COND_EMPTY) {
829 ASSERT(sb_zone_is_full(reset));
831 ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
832 reset->start, reset->len,
837 reset->cond = BLK_ZONE_COND_EMPTY;
838 reset->wp = reset->start;
840 } else if (ret != -ENOENT) {
842 * For READ, we want the previous one. Move write pointer to
843 * the end of a zone, if it is at the head of a zone.
847 if (wp == zones[0].start << SECTOR_SHIFT)
848 zone_end = zones[1].start + zones[1].capacity;
849 else if (wp == zones[1].start << SECTOR_SHIFT)
850 zone_end = zones[0].start + zones[0].capacity;
852 wp = ALIGN_DOWN(zone_end << SECTOR_SHIFT,
853 BTRFS_SUPER_INFO_SIZE);
855 wp -= BTRFS_SUPER_INFO_SIZE;
863 int btrfs_sb_log_location_bdev(struct block_device *bdev, int mirror, int rw,
866 struct blk_zone zones[BTRFS_NR_SB_LOG_ZONES];
867 sector_t zone_sectors;
870 u8 zone_sectors_shift;
874 if (!bdev_is_zoned(bdev)) {
875 *bytenr_ret = btrfs_sb_offset(mirror);
879 ASSERT(rw == READ || rw == WRITE);
881 zone_sectors = bdev_zone_sectors(bdev);
882 if (!is_power_of_2(zone_sectors))
884 zone_sectors_shift = ilog2(zone_sectors);
885 nr_sectors = bdev_nr_sectors(bdev);
886 nr_zones = nr_sectors >> zone_sectors_shift;
888 sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
889 if (sb_zone + 1 >= nr_zones)
892 ret = blkdev_report_zones(bdev, zone_start_sector(sb_zone, bdev),
893 BTRFS_NR_SB_LOG_ZONES, copy_zone_info_cb,
897 if (ret != BTRFS_NR_SB_LOG_ZONES)
900 return sb_log_location(bdev, zones, rw, bytenr_ret);
903 int btrfs_sb_log_location(struct btrfs_device *device, int mirror, int rw,
906 struct btrfs_zoned_device_info *zinfo = device->zone_info;
910 * For a zoned filesystem on a non-zoned block device, use the same
911 * super block locations as regular filesystem. Doing so, the super
912 * block can always be retrieved and the zoned flag of the volume
913 * detected from the super block information.
915 if (!bdev_is_zoned(device->bdev)) {
916 *bytenr_ret = btrfs_sb_offset(mirror);
920 zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
921 if (zone_num + 1 >= zinfo->nr_zones)
924 return sb_log_location(device->bdev,
925 &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror],
929 static inline bool is_sb_log_zone(struct btrfs_zoned_device_info *zinfo,
937 zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
938 if (zone_num + 1 >= zinfo->nr_zones)
941 if (!test_bit(zone_num, zinfo->seq_zones))
947 int btrfs_advance_sb_log(struct btrfs_device *device, int mirror)
949 struct btrfs_zoned_device_info *zinfo = device->zone_info;
950 struct blk_zone *zone;
953 if (!is_sb_log_zone(zinfo, mirror))
956 zone = &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror];
957 for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) {
958 /* Advance the next zone */
959 if (zone->cond == BLK_ZONE_COND_FULL) {
964 if (zone->cond == BLK_ZONE_COND_EMPTY)
965 zone->cond = BLK_ZONE_COND_IMP_OPEN;
967 zone->wp += SUPER_INFO_SECTORS;
969 if (sb_zone_is_full(zone)) {
971 * No room left to write new superblock. Since
972 * superblock is written with REQ_SYNC, it is safe to
973 * finish the zone now.
975 * If the write pointer is exactly at the capacity,
976 * explicit ZONE_FINISH is not necessary.
978 if (zone->wp != zone->start + zone->capacity) {
981 ret = blkdev_zone_mgmt(device->bdev,
982 REQ_OP_ZONE_FINISH, zone->start,
983 zone->len, GFP_NOFS);
988 zone->wp = zone->start + zone->len;
989 zone->cond = BLK_ZONE_COND_FULL;
994 /* All the zones are FULL. Should not reach here. */
999 int btrfs_reset_sb_log_zones(struct block_device *bdev, int mirror)
1001 sector_t zone_sectors;
1002 sector_t nr_sectors;
1003 u8 zone_sectors_shift;
1007 zone_sectors = bdev_zone_sectors(bdev);
1008 zone_sectors_shift = ilog2(zone_sectors);
1009 nr_sectors = bdev_nr_sectors(bdev);
1010 nr_zones = nr_sectors >> zone_sectors_shift;
1012 sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
1013 if (sb_zone + 1 >= nr_zones)
1016 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1017 zone_start_sector(sb_zone, bdev),
1018 zone_sectors * BTRFS_NR_SB_LOG_ZONES, GFP_NOFS);
1022 * btrfs_find_allocatable_zones - find allocatable zones within a given region
1024 * @device: the device to allocate a region on
1025 * @hole_start: the position of the hole to allocate the region
1026 * @num_bytes: size of wanted region
1027 * @hole_end: the end of the hole
1028 * @return: position of allocatable zones
1030 * Allocatable region should not contain any superblock locations.
1032 u64 btrfs_find_allocatable_zones(struct btrfs_device *device, u64 hole_start,
1033 u64 hole_end, u64 num_bytes)
1035 struct btrfs_zoned_device_info *zinfo = device->zone_info;
1036 const u8 shift = zinfo->zone_size_shift;
1037 u64 nzones = num_bytes >> shift;
1038 u64 pos = hole_start;
1043 ASSERT(IS_ALIGNED(hole_start, zinfo->zone_size));
1044 ASSERT(IS_ALIGNED(num_bytes, zinfo->zone_size));
1046 while (pos < hole_end) {
1047 begin = pos >> shift;
1048 end = begin + nzones;
1050 if (end > zinfo->nr_zones)
1053 /* Check if zones in the region are all empty */
1054 if (btrfs_dev_is_sequential(device, pos) &&
1055 find_next_zero_bit(zinfo->empty_zones, end, begin) != end) {
1056 pos += zinfo->zone_size;
1061 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1065 sb_zone = sb_zone_number(shift, i);
1066 if (!(end <= sb_zone ||
1067 sb_zone + BTRFS_NR_SB_LOG_ZONES <= begin)) {
1069 pos = zone_start_physical(
1070 sb_zone + BTRFS_NR_SB_LOG_ZONES, zinfo);
1074 /* We also need to exclude regular superblock positions */
1075 sb_pos = btrfs_sb_offset(i);
1076 if (!(pos + num_bytes <= sb_pos ||
1077 sb_pos + BTRFS_SUPER_INFO_SIZE <= pos)) {
1079 pos = ALIGN(sb_pos + BTRFS_SUPER_INFO_SIZE,
1091 static bool btrfs_dev_set_active_zone(struct btrfs_device *device, u64 pos)
1093 struct btrfs_zoned_device_info *zone_info = device->zone_info;
1094 unsigned int zno = (pos >> zone_info->zone_size_shift);
1096 /* We can use any number of zones */
1097 if (zone_info->max_active_zones == 0)
1100 if (!test_bit(zno, zone_info->active_zones)) {
1101 /* Active zone left? */
1102 if (atomic_dec_if_positive(&zone_info->active_zones_left) < 0)
1104 if (test_and_set_bit(zno, zone_info->active_zones)) {
1105 /* Someone already set the bit */
1106 atomic_inc(&zone_info->active_zones_left);
1113 static void btrfs_dev_clear_active_zone(struct btrfs_device *device, u64 pos)
1115 struct btrfs_zoned_device_info *zone_info = device->zone_info;
1116 unsigned int zno = (pos >> zone_info->zone_size_shift);
1118 /* We can use any number of zones */
1119 if (zone_info->max_active_zones == 0)
1122 if (test_and_clear_bit(zno, zone_info->active_zones))
1123 atomic_inc(&zone_info->active_zones_left);
1126 int btrfs_reset_device_zone(struct btrfs_device *device, u64 physical,
1127 u64 length, u64 *bytes)
1132 ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_RESET,
1133 physical >> SECTOR_SHIFT, length >> SECTOR_SHIFT,
1140 btrfs_dev_set_zone_empty(device, physical);
1141 btrfs_dev_clear_active_zone(device, physical);
1142 physical += device->zone_info->zone_size;
1143 length -= device->zone_info->zone_size;
1149 int btrfs_ensure_empty_zones(struct btrfs_device *device, u64 start, u64 size)
1151 struct btrfs_zoned_device_info *zinfo = device->zone_info;
1152 const u8 shift = zinfo->zone_size_shift;
1153 unsigned long begin = start >> shift;
1154 unsigned long end = (start + size) >> shift;
1158 ASSERT(IS_ALIGNED(start, zinfo->zone_size));
1159 ASSERT(IS_ALIGNED(size, zinfo->zone_size));
1161 if (end > zinfo->nr_zones)
1164 /* All the zones are conventional */
1165 if (find_next_bit(zinfo->seq_zones, end, begin) == end)
1168 /* All the zones are sequential and empty */
1169 if (find_next_zero_bit(zinfo->seq_zones, end, begin) == end &&
1170 find_next_zero_bit(zinfo->empty_zones, end, begin) == end)
1173 for (pos = start; pos < start + size; pos += zinfo->zone_size) {
1176 if (!btrfs_dev_is_sequential(device, pos) ||
1177 btrfs_dev_is_empty_zone(device, pos))
1180 /* Free regions should be empty */
1183 "zoned: resetting device %s (devid %llu) zone %llu for allocation",
1184 rcu_str_deref(device->name), device->devid, pos >> shift);
1187 ret = btrfs_reset_device_zone(device, pos, zinfo->zone_size,
1197 * Calculate an allocation pointer from the extent allocation information
1198 * for a block group consist of conventional zones. It is pointed to the
1199 * end of the highest addressed extent in the block group as an allocation
1202 static int calculate_alloc_pointer(struct btrfs_block_group *cache,
1203 u64 *offset_ret, bool new)
1205 struct btrfs_fs_info *fs_info = cache->fs_info;
1206 struct btrfs_root *root;
1207 struct btrfs_path *path;
1208 struct btrfs_key key;
1209 struct btrfs_key found_key;
1214 * Avoid tree lookups for a new block group, there's no use for it.
1215 * It must always be 0.
1217 * Also, we have a lock chain of extent buffer lock -> chunk mutex.
1218 * For new a block group, this function is called from
1219 * btrfs_make_block_group() which is already taking the chunk mutex.
1220 * Thus, we cannot call calculate_alloc_pointer() which takes extent
1221 * buffer locks to avoid deadlock.
1228 path = btrfs_alloc_path();
1232 key.objectid = cache->start + cache->length;
1236 root = btrfs_extent_root(fs_info, key.objectid);
1237 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1238 /* We should not find the exact match */
1244 ret = btrfs_previous_extent_item(root, path, cache->start);
1253 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
1255 if (found_key.type == BTRFS_EXTENT_ITEM_KEY)
1256 length = found_key.offset;
1258 length = fs_info->nodesize;
1260 if (!(found_key.objectid >= cache->start &&
1261 found_key.objectid + length <= cache->start + cache->length)) {
1265 *offset_ret = found_key.objectid + length - cache->start;
1269 btrfs_free_path(path);
1273 int btrfs_load_block_group_zone_info(struct btrfs_block_group *cache, bool new)
1275 struct btrfs_fs_info *fs_info = cache->fs_info;
1276 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
1277 struct extent_map *em;
1278 struct map_lookup *map;
1279 struct btrfs_device *device;
1280 u64 logical = cache->start;
1281 u64 length = cache->length;
1284 unsigned int nofs_flag;
1285 u64 *alloc_offsets = NULL;
1287 u64 *physical = NULL;
1288 unsigned long *active = NULL;
1290 u32 num_sequential = 0, num_conventional = 0;
1292 if (!btrfs_is_zoned(fs_info))
1296 if (!IS_ALIGNED(length, fs_info->zone_size)) {
1298 "zoned: block group %llu len %llu unaligned to zone size %llu",
1299 logical, length, fs_info->zone_size);
1303 /* Get the chunk mapping */
1304 read_lock(&em_tree->lock);
1305 em = lookup_extent_mapping(em_tree, logical, length);
1306 read_unlock(&em_tree->lock);
1311 map = em->map_lookup;
1313 cache->physical_map = kmemdup(map, map_lookup_size(map->num_stripes), GFP_NOFS);
1314 if (!cache->physical_map) {
1319 alloc_offsets = kcalloc(map->num_stripes, sizeof(*alloc_offsets), GFP_NOFS);
1320 if (!alloc_offsets) {
1325 caps = kcalloc(map->num_stripes, sizeof(*caps), GFP_NOFS);
1331 physical = kcalloc(map->num_stripes, sizeof(*physical), GFP_NOFS);
1337 active = bitmap_zalloc(map->num_stripes, GFP_NOFS);
1343 for (i = 0; i < map->num_stripes; i++) {
1345 struct blk_zone zone;
1346 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1347 int dev_replace_is_ongoing = 0;
1349 device = map->stripes[i].dev;
1350 physical[i] = map->stripes[i].physical;
1352 if (device->bdev == NULL) {
1353 alloc_offsets[i] = WP_MISSING_DEV;
1357 is_sequential = btrfs_dev_is_sequential(device, physical[i]);
1364 * Consider a zone as active if we can allow any number of
1367 if (!device->zone_info->max_active_zones)
1368 __set_bit(i, active);
1370 if (!is_sequential) {
1371 alloc_offsets[i] = WP_CONVENTIONAL;
1376 * This zone will be used for allocation, so mark this zone
1379 btrfs_dev_clear_zone_empty(device, physical[i]);
1381 down_read(&dev_replace->rwsem);
1382 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
1383 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
1384 btrfs_dev_clear_zone_empty(dev_replace->tgtdev, physical[i]);
1385 up_read(&dev_replace->rwsem);
1388 * The group is mapped to a sequential zone. Get the zone write
1389 * pointer to determine the allocation offset within the zone.
1391 WARN_ON(!IS_ALIGNED(physical[i], fs_info->zone_size));
1392 nofs_flag = memalloc_nofs_save();
1393 ret = btrfs_get_dev_zone(device, physical[i], &zone);
1394 memalloc_nofs_restore(nofs_flag);
1395 if (ret == -EIO || ret == -EOPNOTSUPP) {
1397 alloc_offsets[i] = WP_MISSING_DEV;
1403 if (zone.type == BLK_ZONE_TYPE_CONVENTIONAL) {
1404 btrfs_err_in_rcu(fs_info,
1405 "zoned: unexpected conventional zone %llu on device %s (devid %llu)",
1406 zone.start << SECTOR_SHIFT,
1407 rcu_str_deref(device->name), device->devid);
1412 caps[i] = (zone.capacity << SECTOR_SHIFT);
1414 switch (zone.cond) {
1415 case BLK_ZONE_COND_OFFLINE:
1416 case BLK_ZONE_COND_READONLY:
1418 "zoned: offline/readonly zone %llu on device %s (devid %llu)",
1419 physical[i] >> device->zone_info->zone_size_shift,
1420 rcu_str_deref(device->name), device->devid);
1421 alloc_offsets[i] = WP_MISSING_DEV;
1423 case BLK_ZONE_COND_EMPTY:
1424 alloc_offsets[i] = 0;
1426 case BLK_ZONE_COND_FULL:
1427 alloc_offsets[i] = caps[i];
1430 /* Partially used zone */
1432 ((zone.wp - zone.start) << SECTOR_SHIFT);
1433 __set_bit(i, active);
1438 if (num_sequential > 0)
1439 set_bit(BLOCK_GROUP_FLAG_SEQUENTIAL_ZONE, &cache->runtime_flags);
1441 if (num_conventional > 0) {
1442 /* Zone capacity is always zone size in emulation */
1443 cache->zone_capacity = cache->length;
1444 ret = calculate_alloc_pointer(cache, &last_alloc, new);
1447 "zoned: failed to determine allocation offset of bg %llu",
1450 } else if (map->num_stripes == num_conventional) {
1451 cache->alloc_offset = last_alloc;
1452 set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &cache->runtime_flags);
1457 switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
1458 case 0: /* single */
1459 if (alloc_offsets[0] == WP_MISSING_DEV) {
1461 "zoned: cannot recover write pointer for zone %llu",
1466 cache->alloc_offset = alloc_offsets[0];
1467 cache->zone_capacity = caps[0];
1468 if (test_bit(0, active))
1469 set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &cache->runtime_flags);
1471 case BTRFS_BLOCK_GROUP_DUP:
1472 if (map->type & BTRFS_BLOCK_GROUP_DATA) {
1473 btrfs_err(fs_info, "zoned: profile DUP not yet supported on data bg");
1477 if (alloc_offsets[0] == WP_MISSING_DEV) {
1479 "zoned: cannot recover write pointer for zone %llu",
1484 if (alloc_offsets[1] == WP_MISSING_DEV) {
1486 "zoned: cannot recover write pointer for zone %llu",
1491 if (alloc_offsets[0] != alloc_offsets[1]) {
1493 "zoned: write pointer offset mismatch of zones in DUP profile");
1497 if (test_bit(0, active) != test_bit(1, active)) {
1498 if (!btrfs_zone_activate(cache)) {
1503 if (test_bit(0, active))
1504 set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE,
1505 &cache->runtime_flags);
1507 cache->alloc_offset = alloc_offsets[0];
1508 cache->zone_capacity = min(caps[0], caps[1]);
1510 case BTRFS_BLOCK_GROUP_RAID1:
1511 case BTRFS_BLOCK_GROUP_RAID0:
1512 case BTRFS_BLOCK_GROUP_RAID10:
1513 case BTRFS_BLOCK_GROUP_RAID5:
1514 case BTRFS_BLOCK_GROUP_RAID6:
1515 /* non-single profiles are not supported yet */
1517 btrfs_err(fs_info, "zoned: profile %s not yet supported",
1518 btrfs_bg_type_to_raid_name(map->type));
1524 if (cache->alloc_offset > fs_info->zone_size) {
1526 "zoned: invalid write pointer %llu in block group %llu",
1527 cache->alloc_offset, cache->start);
1531 if (cache->alloc_offset > cache->zone_capacity) {
1533 "zoned: invalid write pointer %llu (larger than zone capacity %llu) in block group %llu",
1534 cache->alloc_offset, cache->zone_capacity,
1539 /* An extent is allocated after the write pointer */
1540 if (!ret && num_conventional && last_alloc > cache->alloc_offset) {
1542 "zoned: got wrong write pointer in BG %llu: %llu > %llu",
1543 logical, last_alloc, cache->alloc_offset);
1548 cache->meta_write_pointer = cache->alloc_offset + cache->start;
1549 if (test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &cache->runtime_flags)) {
1550 btrfs_get_block_group(cache);
1551 spin_lock(&fs_info->zone_active_bgs_lock);
1552 list_add_tail(&cache->active_bg_list,
1553 &fs_info->zone_active_bgs);
1554 spin_unlock(&fs_info->zone_active_bgs_lock);
1557 kfree(cache->physical_map);
1558 cache->physical_map = NULL;
1560 bitmap_free(active);
1563 kfree(alloc_offsets);
1564 free_extent_map(em);
1569 void btrfs_calc_zone_unusable(struct btrfs_block_group *cache)
1573 if (!btrfs_is_zoned(cache->fs_info))
1576 WARN_ON(cache->bytes_super != 0);
1577 unusable = (cache->alloc_offset - cache->used) +
1578 (cache->length - cache->zone_capacity);
1579 free = cache->zone_capacity - cache->alloc_offset;
1581 /* We only need ->free_space in ALLOC_SEQ block groups */
1582 cache->cached = BTRFS_CACHE_FINISHED;
1583 cache->free_space_ctl->free_space = free;
1584 cache->zone_unusable = unusable;
1587 void btrfs_redirty_list_add(struct btrfs_transaction *trans,
1588 struct extent_buffer *eb)
1590 struct btrfs_fs_info *fs_info = eb->fs_info;
1592 if (!btrfs_is_zoned(fs_info) ||
1593 btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN) ||
1594 !list_empty(&eb->release_list))
1597 memzero_extent_buffer(eb, 0, eb->len);
1598 set_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags);
1599 set_extent_buffer_dirty(eb);
1600 set_extent_bits_nowait(&trans->dirty_pages, eb->start,
1601 eb->start + eb->len - 1, EXTENT_DIRTY);
1603 spin_lock(&trans->releasing_ebs_lock);
1604 list_add_tail(&eb->release_list, &trans->releasing_ebs);
1605 spin_unlock(&trans->releasing_ebs_lock);
1606 atomic_inc(&eb->refs);
1609 void btrfs_free_redirty_list(struct btrfs_transaction *trans)
1611 spin_lock(&trans->releasing_ebs_lock);
1612 while (!list_empty(&trans->releasing_ebs)) {
1613 struct extent_buffer *eb;
1615 eb = list_first_entry(&trans->releasing_ebs,
1616 struct extent_buffer, release_list);
1617 list_del_init(&eb->release_list);
1618 free_extent_buffer(eb);
1620 spin_unlock(&trans->releasing_ebs_lock);
1623 bool btrfs_use_zone_append(struct btrfs_inode *inode, u64 start)
1625 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1626 struct btrfs_block_group *cache;
1629 if (!btrfs_is_zoned(fs_info))
1632 if (!is_data_inode(&inode->vfs_inode))
1636 * Using REQ_OP_ZONE_APPNED for relocation can break assumptions on the
1637 * extent layout the relocation code has.
1638 * Furthermore we have set aside own block-group from which only the
1639 * relocation "process" can allocate and make sure only one process at a
1640 * time can add pages to an extent that gets relocated, so it's safe to
1641 * use regular REQ_OP_WRITE for this special case.
1643 if (btrfs_is_data_reloc_root(inode->root))
1646 cache = btrfs_lookup_block_group(fs_info, start);
1651 ret = !!test_bit(BLOCK_GROUP_FLAG_SEQUENTIAL_ZONE, &cache->runtime_flags);
1652 btrfs_put_block_group(cache);
1657 void btrfs_record_physical_zoned(struct inode *inode, u64 file_offset,
1660 struct btrfs_ordered_extent *ordered;
1661 const u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
1663 if (bio_op(bio) != REQ_OP_ZONE_APPEND)
1666 ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode), file_offset);
1667 if (WARN_ON(!ordered))
1670 ordered->physical = physical;
1671 ordered->bdev = bio->bi_bdev;
1673 btrfs_put_ordered_extent(ordered);
1676 void btrfs_rewrite_logical_zoned(struct btrfs_ordered_extent *ordered)
1678 struct btrfs_inode *inode = BTRFS_I(ordered->inode);
1679 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1680 struct extent_map_tree *em_tree;
1681 struct extent_map *em;
1682 struct btrfs_ordered_sum *sum;
1683 u64 orig_logical = ordered->disk_bytenr;
1684 u64 *logical = NULL;
1687 /* Zoned devices should not have partitions. So, we can assume it is 0 */
1688 ASSERT(!bdev_is_partition(ordered->bdev));
1689 if (WARN_ON(!ordered->bdev))
1692 if (WARN_ON(btrfs_rmap_block(fs_info, orig_logical, ordered->bdev,
1693 ordered->physical, &logical, &nr,
1699 if (orig_logical == *logical)
1702 ordered->disk_bytenr = *logical;
1704 em_tree = &inode->extent_tree;
1705 write_lock(&em_tree->lock);
1706 em = search_extent_mapping(em_tree, ordered->file_offset,
1707 ordered->num_bytes);
1708 em->block_start = *logical;
1709 free_extent_map(em);
1710 write_unlock(&em_tree->lock);
1712 list_for_each_entry(sum, &ordered->list, list) {
1713 if (*logical < orig_logical)
1714 sum->bytenr -= orig_logical - *logical;
1716 sum->bytenr += *logical - orig_logical;
1723 bool btrfs_check_meta_write_pointer(struct btrfs_fs_info *fs_info,
1724 struct extent_buffer *eb,
1725 struct btrfs_block_group **cache_ret)
1727 struct btrfs_block_group *cache;
1730 if (!btrfs_is_zoned(fs_info))
1733 cache = btrfs_lookup_block_group(fs_info, eb->start);
1737 if (cache->meta_write_pointer != eb->start) {
1738 btrfs_put_block_group(cache);
1742 cache->meta_write_pointer = eb->start + eb->len;
1750 void btrfs_revert_meta_write_pointer(struct btrfs_block_group *cache,
1751 struct extent_buffer *eb)
1753 if (!btrfs_is_zoned(eb->fs_info) || !cache)
1756 ASSERT(cache->meta_write_pointer == eb->start + eb->len);
1757 cache->meta_write_pointer = eb->start;
1760 int btrfs_zoned_issue_zeroout(struct btrfs_device *device, u64 physical, u64 length)
1762 if (!btrfs_dev_is_sequential(device, physical))
1765 return blkdev_issue_zeroout(device->bdev, physical >> SECTOR_SHIFT,
1766 length >> SECTOR_SHIFT, GFP_NOFS, 0);
1769 static int read_zone_info(struct btrfs_fs_info *fs_info, u64 logical,
1770 struct blk_zone *zone)
1772 struct btrfs_io_context *bioc = NULL;
1773 u64 mapped_length = PAGE_SIZE;
1774 unsigned int nofs_flag;
1778 ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS, logical,
1779 &mapped_length, &bioc);
1780 if (ret || !bioc || mapped_length < PAGE_SIZE) {
1785 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
1790 nofs_flag = memalloc_nofs_save();
1791 nmirrors = (int)bioc->num_stripes;
1792 for (i = 0; i < nmirrors; i++) {
1793 u64 physical = bioc->stripes[i].physical;
1794 struct btrfs_device *dev = bioc->stripes[i].dev;
1796 /* Missing device */
1800 ret = btrfs_get_dev_zone(dev, physical, zone);
1801 /* Failing device */
1802 if (ret == -EIO || ret == -EOPNOTSUPP)
1806 memalloc_nofs_restore(nofs_flag);
1808 btrfs_put_bioc(bioc);
1813 * Synchronize write pointer in a zone at @physical_start on @tgt_dev, by
1814 * filling zeros between @physical_pos to a write pointer of dev-replace
1817 int btrfs_sync_zone_write_pointer(struct btrfs_device *tgt_dev, u64 logical,
1818 u64 physical_start, u64 physical_pos)
1820 struct btrfs_fs_info *fs_info = tgt_dev->fs_info;
1821 struct blk_zone zone;
1826 if (!btrfs_dev_is_sequential(tgt_dev, physical_pos))
1829 ret = read_zone_info(fs_info, logical, &zone);
1833 wp = physical_start + ((zone.wp - zone.start) << SECTOR_SHIFT);
1835 if (physical_pos == wp)
1838 if (physical_pos > wp)
1841 length = wp - physical_pos;
1842 return btrfs_zoned_issue_zeroout(tgt_dev, physical_pos, length);
1845 struct btrfs_device *btrfs_zoned_get_device(struct btrfs_fs_info *fs_info,
1846 u64 logical, u64 length)
1848 struct btrfs_device *device;
1849 struct extent_map *em;
1850 struct map_lookup *map;
1852 em = btrfs_get_chunk_map(fs_info, logical, length);
1854 return ERR_CAST(em);
1856 map = em->map_lookup;
1857 /* We only support single profile for now */
1858 device = map->stripes[0].dev;
1860 free_extent_map(em);
1866 * Activate block group and underlying device zones
1868 * @block_group: the block group to activate
1870 * Return: true on success, false otherwise
1872 bool btrfs_zone_activate(struct btrfs_block_group *block_group)
1874 struct btrfs_fs_info *fs_info = block_group->fs_info;
1875 struct map_lookup *map;
1876 struct btrfs_device *device;
1881 if (!btrfs_is_zoned(block_group->fs_info))
1884 map = block_group->physical_map;
1886 spin_lock(&block_group->lock);
1887 if (test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags)) {
1893 if (btrfs_zoned_bg_is_full(block_group)) {
1898 for (i = 0; i < map->num_stripes; i++) {
1899 device = map->stripes[i].dev;
1900 physical = map->stripes[i].physical;
1902 if (device->zone_info->max_active_zones == 0)
1905 if (!btrfs_dev_set_active_zone(device, physical)) {
1906 /* Cannot activate the zone */
1912 /* Successfully activated all the zones */
1913 set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags);
1914 spin_unlock(&block_group->lock);
1916 /* For the active block group list */
1917 btrfs_get_block_group(block_group);
1919 spin_lock(&fs_info->zone_active_bgs_lock);
1920 list_add_tail(&block_group->active_bg_list, &fs_info->zone_active_bgs);
1921 spin_unlock(&fs_info->zone_active_bgs_lock);
1926 spin_unlock(&block_group->lock);
1930 static void wait_eb_writebacks(struct btrfs_block_group *block_group)
1932 struct btrfs_fs_info *fs_info = block_group->fs_info;
1933 const u64 end = block_group->start + block_group->length;
1934 struct radix_tree_iter iter;
1935 struct extent_buffer *eb;
1939 radix_tree_for_each_slot(slot, &fs_info->buffer_radix, &iter,
1940 block_group->start >> fs_info->sectorsize_bits) {
1941 eb = radix_tree_deref_slot(slot);
1944 if (radix_tree_deref_retry(eb)) {
1945 slot = radix_tree_iter_retry(&iter);
1949 if (eb->start < block_group->start)
1951 if (eb->start >= end)
1954 slot = radix_tree_iter_resume(slot, &iter);
1956 wait_on_extent_buffer_writeback(eb);
1962 static int do_zone_finish(struct btrfs_block_group *block_group, bool fully_written)
1964 struct btrfs_fs_info *fs_info = block_group->fs_info;
1965 struct map_lookup *map;
1966 const bool is_metadata = (block_group->flags &
1967 (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_SYSTEM));
1971 spin_lock(&block_group->lock);
1972 if (!test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags)) {
1973 spin_unlock(&block_group->lock);
1977 /* Check if we have unwritten allocated space */
1979 block_group->start + block_group->alloc_offset > block_group->meta_write_pointer) {
1980 spin_unlock(&block_group->lock);
1985 * If we are sure that the block group is full (= no more room left for
1986 * new allocation) and the IO for the last usable block is completed, we
1987 * don't need to wait for the other IOs. This holds because we ensure
1988 * the sequential IO submissions using the ZONE_APPEND command for data
1989 * and block_group->meta_write_pointer for metadata.
1991 if (!fully_written) {
1992 if (test_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags)) {
1993 spin_unlock(&block_group->lock);
1996 spin_unlock(&block_group->lock);
1998 ret = btrfs_inc_block_group_ro(block_group, false);
2002 /* Ensure all writes in this block group finish */
2003 btrfs_wait_block_group_reservations(block_group);
2004 /* No need to wait for NOCOW writers. Zoned mode does not allow that */
2005 btrfs_wait_ordered_roots(fs_info, U64_MAX, block_group->start,
2006 block_group->length);
2007 /* Wait for extent buffers to be written. */
2009 wait_eb_writebacks(block_group);
2011 spin_lock(&block_group->lock);
2014 * Bail out if someone already deactivated the block group, or
2015 * allocated space is left in the block group.
2017 if (!test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE,
2018 &block_group->runtime_flags)) {
2019 spin_unlock(&block_group->lock);
2020 btrfs_dec_block_group_ro(block_group);
2024 if (block_group->reserved ||
2025 test_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC,
2026 &block_group->runtime_flags)) {
2027 spin_unlock(&block_group->lock);
2028 btrfs_dec_block_group_ro(block_group);
2033 clear_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags);
2034 block_group->alloc_offset = block_group->zone_capacity;
2035 block_group->free_space_ctl->free_space = 0;
2036 btrfs_clear_treelog_bg(block_group);
2037 btrfs_clear_data_reloc_bg(block_group);
2038 spin_unlock(&block_group->lock);
2040 map = block_group->physical_map;
2041 for (i = 0; i < map->num_stripes; i++) {
2042 struct btrfs_device *device = map->stripes[i].dev;
2043 const u64 physical = map->stripes[i].physical;
2045 if (device->zone_info->max_active_zones == 0)
2048 ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_FINISH,
2049 physical >> SECTOR_SHIFT,
2050 device->zone_info->zone_size >> SECTOR_SHIFT,
2056 btrfs_dev_clear_active_zone(device, physical);
2060 btrfs_dec_block_group_ro(block_group);
2062 spin_lock(&fs_info->zone_active_bgs_lock);
2063 ASSERT(!list_empty(&block_group->active_bg_list));
2064 list_del_init(&block_group->active_bg_list);
2065 spin_unlock(&fs_info->zone_active_bgs_lock);
2067 /* For active_bg_list */
2068 btrfs_put_block_group(block_group);
2070 clear_and_wake_up_bit(BTRFS_FS_NEED_ZONE_FINISH, &fs_info->flags);
2075 int btrfs_zone_finish(struct btrfs_block_group *block_group)
2077 if (!btrfs_is_zoned(block_group->fs_info))
2080 return do_zone_finish(block_group, false);
2083 bool btrfs_can_activate_zone(struct btrfs_fs_devices *fs_devices, u64 flags)
2085 struct btrfs_fs_info *fs_info = fs_devices->fs_info;
2086 struct btrfs_device *device;
2089 if (!btrfs_is_zoned(fs_info))
2092 /* Check if there is a device with active zones left */
2093 mutex_lock(&fs_info->chunk_mutex);
2094 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
2095 struct btrfs_zoned_device_info *zinfo = device->zone_info;
2100 if (!zinfo->max_active_zones) {
2105 switch (flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
2106 case 0: /* single */
2107 ret = (atomic_read(&zinfo->active_zones_left) >= 1);
2109 case BTRFS_BLOCK_GROUP_DUP:
2110 ret = (atomic_read(&zinfo->active_zones_left) >= 2);
2116 mutex_unlock(&fs_info->chunk_mutex);
2119 set_bit(BTRFS_FS_NEED_ZONE_FINISH, &fs_info->flags);
2124 void btrfs_zone_finish_endio(struct btrfs_fs_info *fs_info, u64 logical, u64 length)
2126 struct btrfs_block_group *block_group;
2127 u64 min_alloc_bytes;
2129 if (!btrfs_is_zoned(fs_info))
2132 block_group = btrfs_lookup_block_group(fs_info, logical);
2133 ASSERT(block_group);
2135 /* No MIXED_BG on zoned btrfs. */
2136 if (block_group->flags & BTRFS_BLOCK_GROUP_DATA)
2137 min_alloc_bytes = fs_info->sectorsize;
2139 min_alloc_bytes = fs_info->nodesize;
2141 /* Bail out if we can allocate more data from this block group. */
2142 if (logical + length + min_alloc_bytes <=
2143 block_group->start + block_group->zone_capacity)
2146 do_zone_finish(block_group, true);
2149 btrfs_put_block_group(block_group);
2152 static void btrfs_zone_finish_endio_workfn(struct work_struct *work)
2154 struct btrfs_block_group *bg =
2155 container_of(work, struct btrfs_block_group, zone_finish_work);
2157 wait_on_extent_buffer_writeback(bg->last_eb);
2158 free_extent_buffer(bg->last_eb);
2159 btrfs_zone_finish_endio(bg->fs_info, bg->start, bg->length);
2160 btrfs_put_block_group(bg);
2163 void btrfs_schedule_zone_finish_bg(struct btrfs_block_group *bg,
2164 struct extent_buffer *eb)
2166 if (!test_bit(BLOCK_GROUP_FLAG_SEQUENTIAL_ZONE, &bg->runtime_flags) ||
2167 eb->start + eb->len * 2 <= bg->start + bg->zone_capacity)
2170 if (WARN_ON(bg->zone_finish_work.func == btrfs_zone_finish_endio_workfn)) {
2171 btrfs_err(bg->fs_info, "double scheduling of bg %llu zone finishing",
2177 btrfs_get_block_group(bg);
2178 atomic_inc(&eb->refs);
2180 INIT_WORK(&bg->zone_finish_work, btrfs_zone_finish_endio_workfn);
2181 queue_work(system_unbound_wq, &bg->zone_finish_work);
2184 void btrfs_clear_data_reloc_bg(struct btrfs_block_group *bg)
2186 struct btrfs_fs_info *fs_info = bg->fs_info;
2188 spin_lock(&fs_info->relocation_bg_lock);
2189 if (fs_info->data_reloc_bg == bg->start)
2190 fs_info->data_reloc_bg = 0;
2191 spin_unlock(&fs_info->relocation_bg_lock);
2194 void btrfs_free_zone_cache(struct btrfs_fs_info *fs_info)
2196 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2197 struct btrfs_device *device;
2199 if (!btrfs_is_zoned(fs_info))
2202 mutex_lock(&fs_devices->device_list_mutex);
2203 list_for_each_entry(device, &fs_devices->devices, dev_list) {
2204 if (device->zone_info) {
2205 vfree(device->zone_info->zone_cache);
2206 device->zone_info->zone_cache = NULL;
2209 mutex_unlock(&fs_devices->device_list_mutex);
2212 bool btrfs_zoned_should_reclaim(struct btrfs_fs_info *fs_info)
2214 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2215 struct btrfs_device *device;
2220 ASSERT(btrfs_is_zoned(fs_info));
2222 if (fs_info->bg_reclaim_threshold == 0)
2225 mutex_lock(&fs_devices->device_list_mutex);
2226 list_for_each_entry(device, &fs_devices->devices, dev_list) {
2230 total += device->disk_total_bytes;
2231 used += device->bytes_used;
2233 mutex_unlock(&fs_devices->device_list_mutex);
2235 factor = div64_u64(used * 100, total);
2236 return factor >= fs_info->bg_reclaim_threshold;
2239 void btrfs_zoned_release_data_reloc_bg(struct btrfs_fs_info *fs_info, u64 logical,
2242 struct btrfs_block_group *block_group;
2244 if (!btrfs_is_zoned(fs_info))
2247 block_group = btrfs_lookup_block_group(fs_info, logical);
2248 /* It should be called on a previous data relocation block group. */
2249 ASSERT(block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA));
2251 spin_lock(&block_group->lock);
2252 if (!test_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags))
2255 /* All relocation extents are written. */
2256 if (block_group->start + block_group->alloc_offset == logical + length) {
2258 * Now, release this block group for further allocations and
2261 clear_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC,
2262 &block_group->runtime_flags);
2266 spin_unlock(&block_group->lock);
2267 btrfs_put_block_group(block_group);
2270 int btrfs_zone_finish_one_bg(struct btrfs_fs_info *fs_info)
2272 struct btrfs_block_group *block_group;
2273 struct btrfs_block_group *min_bg = NULL;
2274 u64 min_avail = U64_MAX;
2277 spin_lock(&fs_info->zone_active_bgs_lock);
2278 list_for_each_entry(block_group, &fs_info->zone_active_bgs,
2282 spin_lock(&block_group->lock);
2283 if (block_group->reserved || block_group->alloc_offset == 0 ||
2284 (block_group->flags & BTRFS_BLOCK_GROUP_SYSTEM) ||
2285 test_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags)) {
2286 spin_unlock(&block_group->lock);
2290 avail = block_group->zone_capacity - block_group->alloc_offset;
2291 if (min_avail > avail) {
2293 btrfs_put_block_group(min_bg);
2294 min_bg = block_group;
2296 btrfs_get_block_group(min_bg);
2298 spin_unlock(&block_group->lock);
2300 spin_unlock(&fs_info->zone_active_bgs_lock);
2305 ret = btrfs_zone_finish(min_bg);
2306 btrfs_put_block_group(min_bg);
2308 return ret < 0 ? ret : 1;
2311 int btrfs_zoned_activate_one_bg(struct btrfs_fs_info *fs_info,
2312 struct btrfs_space_info *space_info,
2315 struct btrfs_block_group *bg;
2318 if (!btrfs_is_zoned(fs_info) || (space_info->flags & BTRFS_BLOCK_GROUP_DATA))
2321 /* No more block groups to activate */
2322 if (space_info->active_total_bytes == space_info->total_bytes)
2327 bool need_finish = false;
2329 down_read(&space_info->groups_sem);
2330 for (index = 0; index < BTRFS_NR_RAID_TYPES; index++) {
2331 list_for_each_entry(bg, &space_info->block_groups[index],
2333 if (!spin_trylock(&bg->lock))
2335 if (btrfs_zoned_bg_is_full(bg) ||
2336 test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE,
2337 &bg->runtime_flags)) {
2338 spin_unlock(&bg->lock);
2341 spin_unlock(&bg->lock);
2343 if (btrfs_zone_activate(bg)) {
2344 up_read(&space_info->groups_sem);
2351 up_read(&space_info->groups_sem);
2353 if (!do_finish || !need_finish)
2356 ret = btrfs_zone_finish_one_bg(fs_info);