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++) {
124 bytenr = ((zones[i].start + zones[i].len)
125 << SECTOR_SHIFT) - BTRFS_SUPER_INFO_SIZE;
127 page[i] = read_cache_page_gfp(mapping,
128 bytenr >> PAGE_SHIFT, GFP_NOFS);
129 if (IS_ERR(page[i])) {
131 btrfs_release_disk_super(super[0]);
132 return PTR_ERR(page[i]);
134 super[i] = page_address(page[i]);
137 if (super[0]->generation > super[1]->generation)
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);
418 if (!IS_ALIGNED(nr_sectors, zone_sectors))
419 zone_info->nr_zones++;
421 max_active_zones = bdev_max_active_zones(bdev);
422 if (max_active_zones && max_active_zones < BTRFS_MIN_ACTIVE_ZONES) {
423 btrfs_err_in_rcu(fs_info,
424 "zoned: %s: max active zones %u is too small, need at least %u active zones",
425 rcu_str_deref(device->name), max_active_zones,
426 BTRFS_MIN_ACTIVE_ZONES);
430 zone_info->max_active_zones = max_active_zones;
432 zone_info->seq_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
433 if (!zone_info->seq_zones) {
438 zone_info->empty_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
439 if (!zone_info->empty_zones) {
444 zone_info->active_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
445 if (!zone_info->active_zones) {
450 zones = kcalloc(BTRFS_REPORT_NR_ZONES, sizeof(struct blk_zone), GFP_KERNEL);
457 * Enable zone cache only for a zoned device. On a non-zoned device, we
458 * fill the zone info with emulated CONVENTIONAL zones, so no need to
461 if (populate_cache && bdev_is_zoned(device->bdev)) {
462 zone_info->zone_cache = vzalloc(sizeof(struct blk_zone) *
463 zone_info->nr_zones);
464 if (!zone_info->zone_cache) {
465 btrfs_err_in_rcu(device->fs_info,
466 "zoned: failed to allocate zone cache for %s",
467 rcu_str_deref(device->name));
475 while (sector < nr_sectors) {
476 nr_zones = BTRFS_REPORT_NR_ZONES;
477 ret = btrfs_get_dev_zones(device, sector << SECTOR_SHIFT, zones,
482 for (i = 0; i < nr_zones; i++) {
483 if (zones[i].type == BLK_ZONE_TYPE_SEQWRITE_REQ)
484 __set_bit(nreported, zone_info->seq_zones);
485 switch (zones[i].cond) {
486 case BLK_ZONE_COND_EMPTY:
487 __set_bit(nreported, zone_info->empty_zones);
489 case BLK_ZONE_COND_IMP_OPEN:
490 case BLK_ZONE_COND_EXP_OPEN:
491 case BLK_ZONE_COND_CLOSED:
492 __set_bit(nreported, zone_info->active_zones);
498 sector = zones[nr_zones - 1].start + zones[nr_zones - 1].len;
501 if (nreported != zone_info->nr_zones) {
502 btrfs_err_in_rcu(device->fs_info,
503 "inconsistent number of zones on %s (%u/%u)",
504 rcu_str_deref(device->name), nreported,
505 zone_info->nr_zones);
510 if (max_active_zones) {
511 if (nactive > max_active_zones) {
512 btrfs_err_in_rcu(device->fs_info,
513 "zoned: %u active zones on %s exceeds max_active_zones %u",
514 nactive, rcu_str_deref(device->name),
519 atomic_set(&zone_info->active_zones_left,
520 max_active_zones - nactive);
523 /* Validate superblock log */
524 nr_zones = BTRFS_NR_SB_LOG_ZONES;
525 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
528 int sb_pos = BTRFS_NR_SB_LOG_ZONES * i;
530 sb_zone = sb_zone_number(zone_info->zone_size_shift, i);
531 if (sb_zone + 1 >= zone_info->nr_zones)
534 ret = btrfs_get_dev_zones(device,
535 zone_start_physical(sb_zone, zone_info),
536 &zone_info->sb_zones[sb_pos],
541 if (nr_zones != BTRFS_NR_SB_LOG_ZONES) {
542 btrfs_err_in_rcu(device->fs_info,
543 "zoned: failed to read super block log zone info at devid %llu zone %u",
544 device->devid, sb_zone);
550 * If zones[0] is conventional, always use the beginning of the
551 * zone to record superblock. No need to validate in that case.
553 if (zone_info->sb_zones[BTRFS_NR_SB_LOG_ZONES * i].type ==
554 BLK_ZONE_TYPE_CONVENTIONAL)
557 ret = sb_write_pointer(device->bdev,
558 &zone_info->sb_zones[sb_pos], &sb_wp);
559 if (ret != -ENOENT && ret) {
560 btrfs_err_in_rcu(device->fs_info,
561 "zoned: super block log zone corrupted devid %llu zone %u",
562 device->devid, sb_zone);
571 switch (bdev_zoned_model(bdev)) {
573 model = "host-managed zoned";
577 model = "host-aware zoned";
582 emulated = "emulated ";
586 btrfs_err_in_rcu(fs_info, "zoned: unsupported model %d on %s",
587 bdev_zoned_model(bdev),
588 rcu_str_deref(device->name));
590 goto out_free_zone_info;
593 btrfs_info_in_rcu(fs_info,
594 "%s block device %s, %u %szones of %llu bytes",
595 model, rcu_str_deref(device->name), zone_info->nr_zones,
596 emulated, zone_info->zone_size);
603 btrfs_destroy_dev_zone_info(device);
608 void btrfs_destroy_dev_zone_info(struct btrfs_device *device)
610 struct btrfs_zoned_device_info *zone_info = device->zone_info;
615 bitmap_free(zone_info->active_zones);
616 bitmap_free(zone_info->seq_zones);
617 bitmap_free(zone_info->empty_zones);
618 vfree(zone_info->zone_cache);
620 device->zone_info = NULL;
623 int btrfs_get_dev_zone(struct btrfs_device *device, u64 pos,
624 struct blk_zone *zone)
626 unsigned int nr_zones = 1;
629 ret = btrfs_get_dev_zones(device, pos, zone, &nr_zones);
630 if (ret != 0 || !nr_zones)
631 return ret ? ret : -EIO;
636 int btrfs_check_zoned_mode(struct btrfs_fs_info *fs_info)
638 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
639 struct btrfs_device *device;
640 u64 zoned_devices = 0;
643 const bool incompat_zoned = btrfs_fs_incompat(fs_info, ZONED);
646 /* Count zoned devices */
647 list_for_each_entry(device, &fs_devices->devices, dev_list) {
648 enum blk_zoned_model model;
653 model = bdev_zoned_model(device->bdev);
655 * A Host-Managed zoned device must be used as a zoned device.
656 * A Host-Aware zoned device and a non-zoned devices can be
657 * treated as a zoned device, if ZONED flag is enabled in the
660 if (model == BLK_ZONED_HM ||
661 (model == BLK_ZONED_HA && incompat_zoned) ||
662 (model == BLK_ZONED_NONE && incompat_zoned)) {
663 struct btrfs_zoned_device_info *zone_info;
665 zone_info = device->zone_info;
668 zone_size = zone_info->zone_size;
669 } else if (zone_info->zone_size != zone_size) {
671 "zoned: unequal block device zone sizes: have %llu found %llu",
672 device->zone_info->zone_size,
681 if (!zoned_devices && !incompat_zoned)
684 if (!zoned_devices && incompat_zoned) {
685 /* No zoned block device found on ZONED filesystem */
687 "zoned: no zoned devices found on a zoned filesystem");
692 if (zoned_devices && !incompat_zoned) {
694 "zoned: mode not enabled but zoned device found");
699 if (zoned_devices != nr_devices) {
701 "zoned: cannot mix zoned and regular devices");
707 * stripe_size is always aligned to BTRFS_STRIPE_LEN in
708 * btrfs_create_chunk(). Since we want stripe_len == zone_size,
709 * check the alignment here.
711 if (!IS_ALIGNED(zone_size, BTRFS_STRIPE_LEN)) {
713 "zoned: zone size %llu not aligned to stripe %u",
714 zone_size, BTRFS_STRIPE_LEN);
719 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
720 btrfs_err(fs_info, "zoned: mixed block groups not supported");
725 fs_info->zone_size = zone_size;
726 fs_info->fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_ZONED;
729 * Check mount options here, because we might change fs_info->zoned
730 * from fs_info->zone_size.
732 ret = btrfs_check_mountopts_zoned(fs_info);
736 btrfs_info(fs_info, "zoned mode enabled with zone size %llu", zone_size);
741 int btrfs_check_mountopts_zoned(struct btrfs_fs_info *info)
743 if (!btrfs_is_zoned(info))
747 * Space cache writing is not COWed. Disable that to avoid write errors
748 * in sequential zones.
750 if (btrfs_test_opt(info, SPACE_CACHE)) {
751 btrfs_err(info, "zoned: space cache v1 is not supported");
755 if (btrfs_test_opt(info, NODATACOW)) {
756 btrfs_err(info, "zoned: NODATACOW not supported");
763 static int sb_log_location(struct block_device *bdev, struct blk_zone *zones,
764 int rw, u64 *bytenr_ret)
769 if (zones[0].type == BLK_ZONE_TYPE_CONVENTIONAL) {
770 *bytenr_ret = zones[0].start << SECTOR_SHIFT;
774 ret = sb_write_pointer(bdev, zones, &wp);
775 if (ret != -ENOENT && ret < 0)
779 struct blk_zone *reset = NULL;
781 if (wp == zones[0].start << SECTOR_SHIFT)
783 else if (wp == zones[1].start << SECTOR_SHIFT)
786 if (reset && reset->cond != BLK_ZONE_COND_EMPTY) {
787 ASSERT(sb_zone_is_full(reset));
789 ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
790 reset->start, reset->len,
795 reset->cond = BLK_ZONE_COND_EMPTY;
796 reset->wp = reset->start;
798 } else if (ret != -ENOENT) {
800 * For READ, we want the previous one. Move write pointer to
801 * the end of a zone, if it is at the head of a zone.
805 if (wp == zones[0].start << SECTOR_SHIFT)
806 zone_end = zones[1].start + zones[1].capacity;
807 else if (wp == zones[1].start << SECTOR_SHIFT)
808 zone_end = zones[0].start + zones[0].capacity;
810 wp = ALIGN_DOWN(zone_end << SECTOR_SHIFT,
811 BTRFS_SUPER_INFO_SIZE);
813 wp -= BTRFS_SUPER_INFO_SIZE;
821 int btrfs_sb_log_location_bdev(struct block_device *bdev, int mirror, int rw,
824 struct blk_zone zones[BTRFS_NR_SB_LOG_ZONES];
825 sector_t zone_sectors;
828 u8 zone_sectors_shift;
832 if (!bdev_is_zoned(bdev)) {
833 *bytenr_ret = btrfs_sb_offset(mirror);
837 ASSERT(rw == READ || rw == WRITE);
839 zone_sectors = bdev_zone_sectors(bdev);
840 if (!is_power_of_2(zone_sectors))
842 zone_sectors_shift = ilog2(zone_sectors);
843 nr_sectors = bdev_nr_sectors(bdev);
844 nr_zones = nr_sectors >> zone_sectors_shift;
846 sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
847 if (sb_zone + 1 >= nr_zones)
850 ret = blkdev_report_zones(bdev, zone_start_sector(sb_zone, bdev),
851 BTRFS_NR_SB_LOG_ZONES, copy_zone_info_cb,
855 if (ret != BTRFS_NR_SB_LOG_ZONES)
858 return sb_log_location(bdev, zones, rw, bytenr_ret);
861 int btrfs_sb_log_location(struct btrfs_device *device, int mirror, int rw,
864 struct btrfs_zoned_device_info *zinfo = device->zone_info;
868 * For a zoned filesystem on a non-zoned block device, use the same
869 * super block locations as regular filesystem. Doing so, the super
870 * block can always be retrieved and the zoned flag of the volume
871 * detected from the super block information.
873 if (!bdev_is_zoned(device->bdev)) {
874 *bytenr_ret = btrfs_sb_offset(mirror);
878 zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
879 if (zone_num + 1 >= zinfo->nr_zones)
882 return sb_log_location(device->bdev,
883 &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror],
887 static inline bool is_sb_log_zone(struct btrfs_zoned_device_info *zinfo,
895 zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
896 if (zone_num + 1 >= zinfo->nr_zones)
899 if (!test_bit(zone_num, zinfo->seq_zones))
905 int btrfs_advance_sb_log(struct btrfs_device *device, int mirror)
907 struct btrfs_zoned_device_info *zinfo = device->zone_info;
908 struct blk_zone *zone;
911 if (!is_sb_log_zone(zinfo, mirror))
914 zone = &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror];
915 for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) {
916 /* Advance the next zone */
917 if (zone->cond == BLK_ZONE_COND_FULL) {
922 if (zone->cond == BLK_ZONE_COND_EMPTY)
923 zone->cond = BLK_ZONE_COND_IMP_OPEN;
925 zone->wp += SUPER_INFO_SECTORS;
927 if (sb_zone_is_full(zone)) {
929 * No room left to write new superblock. Since
930 * superblock is written with REQ_SYNC, it is safe to
931 * finish the zone now.
933 * If the write pointer is exactly at the capacity,
934 * explicit ZONE_FINISH is not necessary.
936 if (zone->wp != zone->start + zone->capacity) {
939 ret = blkdev_zone_mgmt(device->bdev,
940 REQ_OP_ZONE_FINISH, zone->start,
941 zone->len, GFP_NOFS);
946 zone->wp = zone->start + zone->len;
947 zone->cond = BLK_ZONE_COND_FULL;
952 /* All the zones are FULL. Should not reach here. */
957 int btrfs_reset_sb_log_zones(struct block_device *bdev, int mirror)
959 sector_t zone_sectors;
961 u8 zone_sectors_shift;
965 zone_sectors = bdev_zone_sectors(bdev);
966 zone_sectors_shift = ilog2(zone_sectors);
967 nr_sectors = bdev_nr_sectors(bdev);
968 nr_zones = nr_sectors >> zone_sectors_shift;
970 sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
971 if (sb_zone + 1 >= nr_zones)
974 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
975 zone_start_sector(sb_zone, bdev),
976 zone_sectors * BTRFS_NR_SB_LOG_ZONES, GFP_NOFS);
980 * btrfs_find_allocatable_zones - find allocatable zones within a given region
982 * @device: the device to allocate a region on
983 * @hole_start: the position of the hole to allocate the region
984 * @num_bytes: size of wanted region
985 * @hole_end: the end of the hole
986 * @return: position of allocatable zones
988 * Allocatable region should not contain any superblock locations.
990 u64 btrfs_find_allocatable_zones(struct btrfs_device *device, u64 hole_start,
991 u64 hole_end, u64 num_bytes)
993 struct btrfs_zoned_device_info *zinfo = device->zone_info;
994 const u8 shift = zinfo->zone_size_shift;
995 u64 nzones = num_bytes >> shift;
996 u64 pos = hole_start;
1001 ASSERT(IS_ALIGNED(hole_start, zinfo->zone_size));
1002 ASSERT(IS_ALIGNED(num_bytes, zinfo->zone_size));
1004 while (pos < hole_end) {
1005 begin = pos >> shift;
1006 end = begin + nzones;
1008 if (end > zinfo->nr_zones)
1011 /* Check if zones in the region are all empty */
1012 if (btrfs_dev_is_sequential(device, pos) &&
1013 find_next_zero_bit(zinfo->empty_zones, end, begin) != end) {
1014 pos += zinfo->zone_size;
1019 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1023 sb_zone = sb_zone_number(shift, i);
1024 if (!(end <= sb_zone ||
1025 sb_zone + BTRFS_NR_SB_LOG_ZONES <= begin)) {
1027 pos = zone_start_physical(
1028 sb_zone + BTRFS_NR_SB_LOG_ZONES, zinfo);
1032 /* We also need to exclude regular superblock positions */
1033 sb_pos = btrfs_sb_offset(i);
1034 if (!(pos + num_bytes <= sb_pos ||
1035 sb_pos + BTRFS_SUPER_INFO_SIZE <= pos)) {
1037 pos = ALIGN(sb_pos + BTRFS_SUPER_INFO_SIZE,
1049 static bool btrfs_dev_set_active_zone(struct btrfs_device *device, u64 pos)
1051 struct btrfs_zoned_device_info *zone_info = device->zone_info;
1052 unsigned int zno = (pos >> zone_info->zone_size_shift);
1054 /* We can use any number of zones */
1055 if (zone_info->max_active_zones == 0)
1058 if (!test_bit(zno, zone_info->active_zones)) {
1059 /* Active zone left? */
1060 if (atomic_dec_if_positive(&zone_info->active_zones_left) < 0)
1062 if (test_and_set_bit(zno, zone_info->active_zones)) {
1063 /* Someone already set the bit */
1064 atomic_inc(&zone_info->active_zones_left);
1071 static void btrfs_dev_clear_active_zone(struct btrfs_device *device, u64 pos)
1073 struct btrfs_zoned_device_info *zone_info = device->zone_info;
1074 unsigned int zno = (pos >> zone_info->zone_size_shift);
1076 /* We can use any number of zones */
1077 if (zone_info->max_active_zones == 0)
1080 if (test_and_clear_bit(zno, zone_info->active_zones))
1081 atomic_inc(&zone_info->active_zones_left);
1084 int btrfs_reset_device_zone(struct btrfs_device *device, u64 physical,
1085 u64 length, u64 *bytes)
1090 ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_RESET,
1091 physical >> SECTOR_SHIFT, length >> SECTOR_SHIFT,
1098 btrfs_dev_set_zone_empty(device, physical);
1099 btrfs_dev_clear_active_zone(device, physical);
1100 physical += device->zone_info->zone_size;
1101 length -= device->zone_info->zone_size;
1107 int btrfs_ensure_empty_zones(struct btrfs_device *device, u64 start, u64 size)
1109 struct btrfs_zoned_device_info *zinfo = device->zone_info;
1110 const u8 shift = zinfo->zone_size_shift;
1111 unsigned long begin = start >> shift;
1112 unsigned long end = (start + size) >> shift;
1116 ASSERT(IS_ALIGNED(start, zinfo->zone_size));
1117 ASSERT(IS_ALIGNED(size, zinfo->zone_size));
1119 if (end > zinfo->nr_zones)
1122 /* All the zones are conventional */
1123 if (find_next_bit(zinfo->seq_zones, begin, end) == end)
1126 /* All the zones are sequential and empty */
1127 if (find_next_zero_bit(zinfo->seq_zones, begin, end) == end &&
1128 find_next_zero_bit(zinfo->empty_zones, begin, end) == end)
1131 for (pos = start; pos < start + size; pos += zinfo->zone_size) {
1134 if (!btrfs_dev_is_sequential(device, pos) ||
1135 btrfs_dev_is_empty_zone(device, pos))
1138 /* Free regions should be empty */
1141 "zoned: resetting device %s (devid %llu) zone %llu for allocation",
1142 rcu_str_deref(device->name), device->devid, pos >> shift);
1145 ret = btrfs_reset_device_zone(device, pos, zinfo->zone_size,
1155 * Calculate an allocation pointer from the extent allocation information
1156 * for a block group consist of conventional zones. It is pointed to the
1157 * end of the highest addressed extent in the block group as an allocation
1160 static int calculate_alloc_pointer(struct btrfs_block_group *cache,
1163 struct btrfs_fs_info *fs_info = cache->fs_info;
1164 struct btrfs_root *root;
1165 struct btrfs_path *path;
1166 struct btrfs_key key;
1167 struct btrfs_key found_key;
1171 path = btrfs_alloc_path();
1175 key.objectid = cache->start + cache->length;
1179 root = btrfs_extent_root(fs_info, key.objectid);
1180 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1181 /* We should not find the exact match */
1187 ret = btrfs_previous_extent_item(root, path, cache->start);
1196 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
1198 if (found_key.type == BTRFS_EXTENT_ITEM_KEY)
1199 length = found_key.offset;
1201 length = fs_info->nodesize;
1203 if (!(found_key.objectid >= cache->start &&
1204 found_key.objectid + length <= cache->start + cache->length)) {
1208 *offset_ret = found_key.objectid + length - cache->start;
1212 btrfs_free_path(path);
1216 int btrfs_load_block_group_zone_info(struct btrfs_block_group *cache, bool new)
1218 struct btrfs_fs_info *fs_info = cache->fs_info;
1219 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
1220 struct extent_map *em;
1221 struct map_lookup *map;
1222 struct btrfs_device *device;
1223 u64 logical = cache->start;
1224 u64 length = cache->length;
1227 unsigned int nofs_flag;
1228 u64 *alloc_offsets = NULL;
1230 u64 *physical = NULL;
1231 unsigned long *active = NULL;
1233 u32 num_sequential = 0, num_conventional = 0;
1235 if (!btrfs_is_zoned(fs_info))
1239 if (!IS_ALIGNED(length, fs_info->zone_size)) {
1241 "zoned: block group %llu len %llu unaligned to zone size %llu",
1242 logical, length, fs_info->zone_size);
1246 /* Get the chunk mapping */
1247 read_lock(&em_tree->lock);
1248 em = lookup_extent_mapping(em_tree, logical, length);
1249 read_unlock(&em_tree->lock);
1254 map = em->map_lookup;
1256 cache->physical_map = kmemdup(map, map_lookup_size(map->num_stripes), GFP_NOFS);
1257 if (!cache->physical_map) {
1262 alloc_offsets = kcalloc(map->num_stripes, sizeof(*alloc_offsets), GFP_NOFS);
1263 if (!alloc_offsets) {
1268 caps = kcalloc(map->num_stripes, sizeof(*caps), GFP_NOFS);
1274 physical = kcalloc(map->num_stripes, sizeof(*physical), GFP_NOFS);
1280 active = bitmap_zalloc(map->num_stripes, GFP_NOFS);
1286 for (i = 0; i < map->num_stripes; i++) {
1288 struct blk_zone zone;
1289 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1290 int dev_replace_is_ongoing = 0;
1292 device = map->stripes[i].dev;
1293 physical[i] = map->stripes[i].physical;
1295 if (device->bdev == NULL) {
1296 alloc_offsets[i] = WP_MISSING_DEV;
1300 is_sequential = btrfs_dev_is_sequential(device, physical[i]);
1306 if (!is_sequential) {
1307 alloc_offsets[i] = WP_CONVENTIONAL;
1312 * This zone will be used for allocation, so mark this zone
1315 btrfs_dev_clear_zone_empty(device, physical[i]);
1317 down_read(&dev_replace->rwsem);
1318 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
1319 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
1320 btrfs_dev_clear_zone_empty(dev_replace->tgtdev, physical[i]);
1321 up_read(&dev_replace->rwsem);
1324 * The group is mapped to a sequential zone. Get the zone write
1325 * pointer to determine the allocation offset within the zone.
1327 WARN_ON(!IS_ALIGNED(physical[i], fs_info->zone_size));
1328 nofs_flag = memalloc_nofs_save();
1329 ret = btrfs_get_dev_zone(device, physical[i], &zone);
1330 memalloc_nofs_restore(nofs_flag);
1331 if (ret == -EIO || ret == -EOPNOTSUPP) {
1333 alloc_offsets[i] = WP_MISSING_DEV;
1339 if (zone.type == BLK_ZONE_TYPE_CONVENTIONAL) {
1340 btrfs_err_in_rcu(fs_info,
1341 "zoned: unexpected conventional zone %llu on device %s (devid %llu)",
1342 zone.start << SECTOR_SHIFT,
1343 rcu_str_deref(device->name), device->devid);
1348 caps[i] = (zone.capacity << SECTOR_SHIFT);
1350 switch (zone.cond) {
1351 case BLK_ZONE_COND_OFFLINE:
1352 case BLK_ZONE_COND_READONLY:
1354 "zoned: offline/readonly zone %llu on device %s (devid %llu)",
1355 physical[i] >> device->zone_info->zone_size_shift,
1356 rcu_str_deref(device->name), device->devid);
1357 alloc_offsets[i] = WP_MISSING_DEV;
1359 case BLK_ZONE_COND_EMPTY:
1360 alloc_offsets[i] = 0;
1362 case BLK_ZONE_COND_FULL:
1363 alloc_offsets[i] = caps[i];
1366 /* Partially used zone */
1368 ((zone.wp - zone.start) << SECTOR_SHIFT);
1369 __set_bit(i, active);
1374 * Consider a zone as active if we can allow any number of
1377 if (!device->zone_info->max_active_zones)
1378 __set_bit(i, active);
1381 if (num_sequential > 0)
1382 cache->seq_zone = true;
1384 if (num_conventional > 0) {
1386 * Avoid calling calculate_alloc_pointer() for new BG. It
1387 * is no use for new BG. It must be always 0.
1389 * Also, we have a lock chain of extent buffer lock ->
1390 * chunk mutex. For new BG, this function is called from
1391 * btrfs_make_block_group() which is already taking the
1392 * chunk mutex. Thus, we cannot call
1393 * calculate_alloc_pointer() which takes extent buffer
1394 * locks to avoid deadlock.
1397 /* Zone capacity is always zone size in emulation */
1398 cache->zone_capacity = cache->length;
1400 cache->alloc_offset = 0;
1403 ret = calculate_alloc_pointer(cache, &last_alloc);
1404 if (ret || map->num_stripes == num_conventional) {
1406 cache->alloc_offset = last_alloc;
1409 "zoned: failed to determine allocation offset of bg %llu",
1415 switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
1416 case 0: /* single */
1417 if (alloc_offsets[0] == WP_MISSING_DEV) {
1419 "zoned: cannot recover write pointer for zone %llu",
1424 cache->alloc_offset = alloc_offsets[0];
1425 cache->zone_capacity = caps[0];
1426 cache->zone_is_active = test_bit(0, active);
1428 case BTRFS_BLOCK_GROUP_DUP:
1429 if (map->type & BTRFS_BLOCK_GROUP_DATA) {
1430 btrfs_err(fs_info, "zoned: profile DUP not yet supported on data bg");
1434 if (alloc_offsets[0] == WP_MISSING_DEV) {
1436 "zoned: cannot recover write pointer for zone %llu",
1441 if (alloc_offsets[1] == WP_MISSING_DEV) {
1443 "zoned: cannot recover write pointer for zone %llu",
1448 if (alloc_offsets[0] != alloc_offsets[1]) {
1450 "zoned: write pointer offset mismatch of zones in DUP profile");
1454 if (test_bit(0, active) != test_bit(1, active)) {
1455 if (!btrfs_zone_activate(cache)) {
1460 cache->zone_is_active = test_bit(0, active);
1462 cache->alloc_offset = alloc_offsets[0];
1463 cache->zone_capacity = min(caps[0], caps[1]);
1465 case BTRFS_BLOCK_GROUP_RAID1:
1466 case BTRFS_BLOCK_GROUP_RAID0:
1467 case BTRFS_BLOCK_GROUP_RAID10:
1468 case BTRFS_BLOCK_GROUP_RAID5:
1469 case BTRFS_BLOCK_GROUP_RAID6:
1470 /* non-single profiles are not supported yet */
1472 btrfs_err(fs_info, "zoned: profile %s not yet supported",
1473 btrfs_bg_type_to_raid_name(map->type));
1478 if (cache->zone_is_active) {
1479 btrfs_get_block_group(cache);
1480 spin_lock(&fs_info->zone_active_bgs_lock);
1481 list_add_tail(&cache->active_bg_list, &fs_info->zone_active_bgs);
1482 spin_unlock(&fs_info->zone_active_bgs_lock);
1486 if (cache->alloc_offset > fs_info->zone_size) {
1488 "zoned: invalid write pointer %llu in block group %llu",
1489 cache->alloc_offset, cache->start);
1493 if (cache->alloc_offset > cache->zone_capacity) {
1495 "zoned: invalid write pointer %llu (larger than zone capacity %llu) in block group %llu",
1496 cache->alloc_offset, cache->zone_capacity,
1501 /* An extent is allocated after the write pointer */
1502 if (!ret && num_conventional && last_alloc > cache->alloc_offset) {
1504 "zoned: got wrong write pointer in BG %llu: %llu > %llu",
1505 logical, last_alloc, cache->alloc_offset);
1510 cache->meta_write_pointer = cache->alloc_offset + cache->start;
1513 kfree(cache->physical_map);
1514 cache->physical_map = NULL;
1516 bitmap_free(active);
1519 kfree(alloc_offsets);
1520 free_extent_map(em);
1525 void btrfs_calc_zone_unusable(struct btrfs_block_group *cache)
1529 if (!btrfs_is_zoned(cache->fs_info))
1532 WARN_ON(cache->bytes_super != 0);
1533 unusable = (cache->alloc_offset - cache->used) +
1534 (cache->length - cache->zone_capacity);
1535 free = cache->zone_capacity - cache->alloc_offset;
1537 /* We only need ->free_space in ALLOC_SEQ block groups */
1538 cache->last_byte_to_unpin = (u64)-1;
1539 cache->cached = BTRFS_CACHE_FINISHED;
1540 cache->free_space_ctl->free_space = free;
1541 cache->zone_unusable = unusable;
1544 void btrfs_redirty_list_add(struct btrfs_transaction *trans,
1545 struct extent_buffer *eb)
1547 struct btrfs_fs_info *fs_info = eb->fs_info;
1549 if (!btrfs_is_zoned(fs_info) ||
1550 btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN) ||
1551 !list_empty(&eb->release_list))
1554 set_extent_buffer_dirty(eb);
1555 set_extent_bits_nowait(&trans->dirty_pages, eb->start,
1556 eb->start + eb->len - 1, EXTENT_DIRTY);
1557 memzero_extent_buffer(eb, 0, eb->len);
1558 set_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags);
1560 spin_lock(&trans->releasing_ebs_lock);
1561 list_add_tail(&eb->release_list, &trans->releasing_ebs);
1562 spin_unlock(&trans->releasing_ebs_lock);
1563 atomic_inc(&eb->refs);
1566 void btrfs_free_redirty_list(struct btrfs_transaction *trans)
1568 spin_lock(&trans->releasing_ebs_lock);
1569 while (!list_empty(&trans->releasing_ebs)) {
1570 struct extent_buffer *eb;
1572 eb = list_first_entry(&trans->releasing_ebs,
1573 struct extent_buffer, release_list);
1574 list_del_init(&eb->release_list);
1575 free_extent_buffer(eb);
1577 spin_unlock(&trans->releasing_ebs_lock);
1580 bool btrfs_use_zone_append(struct btrfs_inode *inode, u64 start)
1582 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1583 struct btrfs_block_group *cache;
1586 if (!btrfs_is_zoned(fs_info))
1589 if (!is_data_inode(&inode->vfs_inode))
1593 * Using REQ_OP_ZONE_APPNED for relocation can break assumptions on the
1594 * extent layout the relocation code has.
1595 * Furthermore we have set aside own block-group from which only the
1596 * relocation "process" can allocate and make sure only one process at a
1597 * time can add pages to an extent that gets relocated, so it's safe to
1598 * use regular REQ_OP_WRITE for this special case.
1600 if (btrfs_is_data_reloc_root(inode->root))
1603 cache = btrfs_lookup_block_group(fs_info, start);
1608 ret = cache->seq_zone;
1609 btrfs_put_block_group(cache);
1614 void btrfs_record_physical_zoned(struct inode *inode, u64 file_offset,
1617 struct btrfs_ordered_extent *ordered;
1618 const u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
1620 if (bio_op(bio) != REQ_OP_ZONE_APPEND)
1623 ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode), file_offset);
1624 if (WARN_ON(!ordered))
1627 ordered->physical = physical;
1628 ordered->bdev = bio->bi_bdev;
1630 btrfs_put_ordered_extent(ordered);
1633 void btrfs_rewrite_logical_zoned(struct btrfs_ordered_extent *ordered)
1635 struct btrfs_inode *inode = BTRFS_I(ordered->inode);
1636 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1637 struct extent_map_tree *em_tree;
1638 struct extent_map *em;
1639 struct btrfs_ordered_sum *sum;
1640 u64 orig_logical = ordered->disk_bytenr;
1641 u64 *logical = NULL;
1644 /* Zoned devices should not have partitions. So, we can assume it is 0 */
1645 ASSERT(!bdev_is_partition(ordered->bdev));
1646 if (WARN_ON(!ordered->bdev))
1649 if (WARN_ON(btrfs_rmap_block(fs_info, orig_logical, ordered->bdev,
1650 ordered->physical, &logical, &nr,
1656 if (orig_logical == *logical)
1659 ordered->disk_bytenr = *logical;
1661 em_tree = &inode->extent_tree;
1662 write_lock(&em_tree->lock);
1663 em = search_extent_mapping(em_tree, ordered->file_offset,
1664 ordered->num_bytes);
1665 em->block_start = *logical;
1666 free_extent_map(em);
1667 write_unlock(&em_tree->lock);
1669 list_for_each_entry(sum, &ordered->list, list) {
1670 if (*logical < orig_logical)
1671 sum->bytenr -= orig_logical - *logical;
1673 sum->bytenr += *logical - orig_logical;
1680 bool btrfs_check_meta_write_pointer(struct btrfs_fs_info *fs_info,
1681 struct extent_buffer *eb,
1682 struct btrfs_block_group **cache_ret)
1684 struct btrfs_block_group *cache;
1687 if (!btrfs_is_zoned(fs_info))
1690 cache = btrfs_lookup_block_group(fs_info, eb->start);
1694 if (cache->meta_write_pointer != eb->start) {
1695 btrfs_put_block_group(cache);
1699 cache->meta_write_pointer = eb->start + eb->len;
1707 void btrfs_revert_meta_write_pointer(struct btrfs_block_group *cache,
1708 struct extent_buffer *eb)
1710 if (!btrfs_is_zoned(eb->fs_info) || !cache)
1713 ASSERT(cache->meta_write_pointer == eb->start + eb->len);
1714 cache->meta_write_pointer = eb->start;
1717 int btrfs_zoned_issue_zeroout(struct btrfs_device *device, u64 physical, u64 length)
1719 if (!btrfs_dev_is_sequential(device, physical))
1722 return blkdev_issue_zeroout(device->bdev, physical >> SECTOR_SHIFT,
1723 length >> SECTOR_SHIFT, GFP_NOFS, 0);
1726 static int read_zone_info(struct btrfs_fs_info *fs_info, u64 logical,
1727 struct blk_zone *zone)
1729 struct btrfs_io_context *bioc = NULL;
1730 u64 mapped_length = PAGE_SIZE;
1731 unsigned int nofs_flag;
1735 ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS, logical,
1736 &mapped_length, &bioc);
1737 if (ret || !bioc || mapped_length < PAGE_SIZE) {
1742 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
1747 nofs_flag = memalloc_nofs_save();
1748 nmirrors = (int)bioc->num_stripes;
1749 for (i = 0; i < nmirrors; i++) {
1750 u64 physical = bioc->stripes[i].physical;
1751 struct btrfs_device *dev = bioc->stripes[i].dev;
1753 /* Missing device */
1757 ret = btrfs_get_dev_zone(dev, physical, zone);
1758 /* Failing device */
1759 if (ret == -EIO || ret == -EOPNOTSUPP)
1763 memalloc_nofs_restore(nofs_flag);
1765 btrfs_put_bioc(bioc);
1770 * Synchronize write pointer in a zone at @physical_start on @tgt_dev, by
1771 * filling zeros between @physical_pos to a write pointer of dev-replace
1774 int btrfs_sync_zone_write_pointer(struct btrfs_device *tgt_dev, u64 logical,
1775 u64 physical_start, u64 physical_pos)
1777 struct btrfs_fs_info *fs_info = tgt_dev->fs_info;
1778 struct blk_zone zone;
1783 if (!btrfs_dev_is_sequential(tgt_dev, physical_pos))
1786 ret = read_zone_info(fs_info, logical, &zone);
1790 wp = physical_start + ((zone.wp - zone.start) << SECTOR_SHIFT);
1792 if (physical_pos == wp)
1795 if (physical_pos > wp)
1798 length = wp - physical_pos;
1799 return btrfs_zoned_issue_zeroout(tgt_dev, physical_pos, length);
1802 struct btrfs_device *btrfs_zoned_get_device(struct btrfs_fs_info *fs_info,
1803 u64 logical, u64 length)
1805 struct btrfs_device *device;
1806 struct extent_map *em;
1807 struct map_lookup *map;
1809 em = btrfs_get_chunk_map(fs_info, logical, length);
1811 return ERR_CAST(em);
1813 map = em->map_lookup;
1814 /* We only support single profile for now */
1815 device = map->stripes[0].dev;
1817 free_extent_map(em);
1823 * Activate block group and underlying device zones
1825 * @block_group: the block group to activate
1827 * Return: true on success, false otherwise
1829 bool btrfs_zone_activate(struct btrfs_block_group *block_group)
1831 struct btrfs_fs_info *fs_info = block_group->fs_info;
1832 struct map_lookup *map;
1833 struct btrfs_device *device;
1838 if (!btrfs_is_zoned(block_group->fs_info))
1841 map = block_group->physical_map;
1843 spin_lock(&block_group->lock);
1844 if (block_group->zone_is_active) {
1850 if (btrfs_zoned_bg_is_full(block_group)) {
1855 for (i = 0; i < map->num_stripes; i++) {
1856 device = map->stripes[i].dev;
1857 physical = map->stripes[i].physical;
1859 if (device->zone_info->max_active_zones == 0)
1862 if (!btrfs_dev_set_active_zone(device, physical)) {
1863 /* Cannot activate the zone */
1869 /* Successfully activated all the zones */
1870 block_group->zone_is_active = 1;
1871 spin_unlock(&block_group->lock);
1873 /* For the active block group list */
1874 btrfs_get_block_group(block_group);
1876 spin_lock(&fs_info->zone_active_bgs_lock);
1877 list_add_tail(&block_group->active_bg_list, &fs_info->zone_active_bgs);
1878 spin_unlock(&fs_info->zone_active_bgs_lock);
1883 spin_unlock(&block_group->lock);
1887 static int do_zone_finish(struct btrfs_block_group *block_group, bool fully_written)
1889 struct btrfs_fs_info *fs_info = block_group->fs_info;
1890 struct map_lookup *map;
1894 spin_lock(&block_group->lock);
1895 if (!block_group->zone_is_active) {
1896 spin_unlock(&block_group->lock);
1900 /* Check if we have unwritten allocated space */
1901 if ((block_group->flags &
1902 (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_SYSTEM)) &&
1903 block_group->start + block_group->alloc_offset > block_group->meta_write_pointer) {
1904 spin_unlock(&block_group->lock);
1909 * If we are sure that the block group is full (= no more room left for
1910 * new allocation) and the IO for the last usable block is completed, we
1911 * don't need to wait for the other IOs. This holds because we ensure
1912 * the sequential IO submissions using the ZONE_APPEND command for data
1913 * and block_group->meta_write_pointer for metadata.
1915 if (!fully_written) {
1916 spin_unlock(&block_group->lock);
1918 ret = btrfs_inc_block_group_ro(block_group, false);
1922 /* Ensure all writes in this block group finish */
1923 btrfs_wait_block_group_reservations(block_group);
1924 /* No need to wait for NOCOW writers. Zoned mode does not allow that */
1925 btrfs_wait_ordered_roots(fs_info, U64_MAX, block_group->start,
1926 block_group->length);
1928 spin_lock(&block_group->lock);
1931 * Bail out if someone already deactivated the block group, or
1932 * allocated space is left in the block group.
1934 if (!block_group->zone_is_active) {
1935 spin_unlock(&block_group->lock);
1936 btrfs_dec_block_group_ro(block_group);
1940 if (block_group->reserved) {
1941 spin_unlock(&block_group->lock);
1942 btrfs_dec_block_group_ro(block_group);
1947 block_group->zone_is_active = 0;
1948 block_group->alloc_offset = block_group->zone_capacity;
1949 block_group->free_space_ctl->free_space = 0;
1950 btrfs_clear_treelog_bg(block_group);
1951 btrfs_clear_data_reloc_bg(block_group);
1952 spin_unlock(&block_group->lock);
1954 map = block_group->physical_map;
1955 for (i = 0; i < map->num_stripes; i++) {
1956 struct btrfs_device *device = map->stripes[i].dev;
1957 const u64 physical = map->stripes[i].physical;
1959 if (device->zone_info->max_active_zones == 0)
1962 ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_FINISH,
1963 physical >> SECTOR_SHIFT,
1964 device->zone_info->zone_size >> SECTOR_SHIFT,
1970 btrfs_dev_clear_active_zone(device, physical);
1974 btrfs_dec_block_group_ro(block_group);
1976 spin_lock(&fs_info->zone_active_bgs_lock);
1977 ASSERT(!list_empty(&block_group->active_bg_list));
1978 list_del_init(&block_group->active_bg_list);
1979 spin_unlock(&fs_info->zone_active_bgs_lock);
1981 /* For active_bg_list */
1982 btrfs_put_block_group(block_group);
1987 int btrfs_zone_finish(struct btrfs_block_group *block_group)
1989 if (!btrfs_is_zoned(block_group->fs_info))
1992 return do_zone_finish(block_group, false);
1995 bool btrfs_can_activate_zone(struct btrfs_fs_devices *fs_devices, u64 flags)
1997 struct btrfs_fs_info *fs_info = fs_devices->fs_info;
1998 struct btrfs_device *device;
2001 if (!btrfs_is_zoned(fs_info))
2004 /* Check if there is a device with active zones left */
2005 mutex_lock(&fs_info->chunk_mutex);
2006 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
2007 struct btrfs_zoned_device_info *zinfo = device->zone_info;
2012 if (!zinfo->max_active_zones ||
2013 atomic_read(&zinfo->active_zones_left)) {
2018 mutex_unlock(&fs_info->chunk_mutex);
2023 void btrfs_zone_finish_endio(struct btrfs_fs_info *fs_info, u64 logical, u64 length)
2025 struct btrfs_block_group *block_group;
2026 u64 min_alloc_bytes;
2028 if (!btrfs_is_zoned(fs_info))
2031 block_group = btrfs_lookup_block_group(fs_info, logical);
2032 ASSERT(block_group);
2034 /* No MIXED_BG on zoned btrfs. */
2035 if (block_group->flags & BTRFS_BLOCK_GROUP_DATA)
2036 min_alloc_bytes = fs_info->sectorsize;
2038 min_alloc_bytes = fs_info->nodesize;
2040 /* Bail out if we can allocate more data from this block group. */
2041 if (logical + length + min_alloc_bytes <=
2042 block_group->start + block_group->zone_capacity)
2045 do_zone_finish(block_group, true);
2048 btrfs_put_block_group(block_group);
2051 static void btrfs_zone_finish_endio_workfn(struct work_struct *work)
2053 struct btrfs_block_group *bg =
2054 container_of(work, struct btrfs_block_group, zone_finish_work);
2056 wait_on_extent_buffer_writeback(bg->last_eb);
2057 free_extent_buffer(bg->last_eb);
2058 btrfs_zone_finish_endio(bg->fs_info, bg->start, bg->length);
2059 btrfs_put_block_group(bg);
2062 void btrfs_schedule_zone_finish_bg(struct btrfs_block_group *bg,
2063 struct extent_buffer *eb)
2065 if (!bg->seq_zone || eb->start + eb->len * 2 <= bg->start + bg->zone_capacity)
2068 if (WARN_ON(bg->zone_finish_work.func == btrfs_zone_finish_endio_workfn)) {
2069 btrfs_err(bg->fs_info, "double scheduling of bg %llu zone finishing",
2075 btrfs_get_block_group(bg);
2076 atomic_inc(&eb->refs);
2078 INIT_WORK(&bg->zone_finish_work, btrfs_zone_finish_endio_workfn);
2079 queue_work(system_unbound_wq, &bg->zone_finish_work);
2082 void btrfs_clear_data_reloc_bg(struct btrfs_block_group *bg)
2084 struct btrfs_fs_info *fs_info = bg->fs_info;
2086 spin_lock(&fs_info->relocation_bg_lock);
2087 if (fs_info->data_reloc_bg == bg->start)
2088 fs_info->data_reloc_bg = 0;
2089 spin_unlock(&fs_info->relocation_bg_lock);
2092 void btrfs_free_zone_cache(struct btrfs_fs_info *fs_info)
2094 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2095 struct btrfs_device *device;
2097 if (!btrfs_is_zoned(fs_info))
2100 mutex_lock(&fs_devices->device_list_mutex);
2101 list_for_each_entry(device, &fs_devices->devices, dev_list) {
2102 if (device->zone_info) {
2103 vfree(device->zone_info->zone_cache);
2104 device->zone_info->zone_cache = NULL;
2107 mutex_unlock(&fs_devices->device_list_mutex);
2110 bool btrfs_zoned_should_reclaim(struct btrfs_fs_info *fs_info)
2112 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2113 struct btrfs_device *device;
2118 ASSERT(btrfs_is_zoned(fs_info));
2120 if (fs_info->bg_reclaim_threshold == 0)
2123 mutex_lock(&fs_devices->device_list_mutex);
2124 list_for_each_entry(device, &fs_devices->devices, dev_list) {
2128 total += device->disk_total_bytes;
2129 used += device->bytes_used;
2131 mutex_unlock(&fs_devices->device_list_mutex);
2133 factor = div64_u64(used * 100, total);
2134 return factor >= fs_info->bg_reclaim_threshold;
2137 void btrfs_zoned_release_data_reloc_bg(struct btrfs_fs_info *fs_info, u64 logical,
2140 struct btrfs_block_group *block_group;
2142 if (!btrfs_is_zoned(fs_info))
2145 block_group = btrfs_lookup_block_group(fs_info, logical);
2146 /* It should be called on a previous data relocation block group. */
2147 ASSERT(block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA));
2149 spin_lock(&block_group->lock);
2150 if (!block_group->zoned_data_reloc_ongoing)
2153 /* All relocation extents are written. */
2154 if (block_group->start + block_group->alloc_offset == logical + length) {
2155 /* Now, release this block group for further allocations. */
2156 block_group->zoned_data_reloc_ongoing = 0;
2160 spin_unlock(&block_group->lock);
2161 btrfs_put_block_group(block_group);