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);
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 = kcalloc(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);
542 /* Validate superblock log */
543 nr_zones = BTRFS_NR_SB_LOG_ZONES;
544 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
547 int sb_pos = BTRFS_NR_SB_LOG_ZONES * i;
549 sb_zone = sb_zone_number(zone_info->zone_size_shift, i);
550 if (sb_zone + 1 >= zone_info->nr_zones)
553 ret = btrfs_get_dev_zones(device,
554 zone_start_physical(sb_zone, zone_info),
555 &zone_info->sb_zones[sb_pos],
560 if (nr_zones != BTRFS_NR_SB_LOG_ZONES) {
561 btrfs_err_in_rcu(device->fs_info,
562 "zoned: failed to read super block log zone info at devid %llu zone %u",
563 device->devid, sb_zone);
569 * If zones[0] is conventional, always use the beginning of the
570 * zone to record superblock. No need to validate in that case.
572 if (zone_info->sb_zones[BTRFS_NR_SB_LOG_ZONES * i].type ==
573 BLK_ZONE_TYPE_CONVENTIONAL)
576 ret = sb_write_pointer(device->bdev,
577 &zone_info->sb_zones[sb_pos], &sb_wp);
578 if (ret != -ENOENT && ret) {
579 btrfs_err_in_rcu(device->fs_info,
580 "zoned: super block log zone corrupted devid %llu zone %u",
581 device->devid, sb_zone);
590 switch (bdev_zoned_model(bdev)) {
592 model = "host-managed zoned";
596 model = "host-aware zoned";
601 emulated = "emulated ";
605 btrfs_err_in_rcu(fs_info, "zoned: unsupported model %d on %s",
606 bdev_zoned_model(bdev),
607 rcu_str_deref(device->name));
609 goto out_free_zone_info;
612 btrfs_info_in_rcu(fs_info,
613 "%s block device %s, %u %szones of %llu bytes",
614 model, rcu_str_deref(device->name), zone_info->nr_zones,
615 emulated, zone_info->zone_size);
622 btrfs_destroy_dev_zone_info(device);
627 void btrfs_destroy_dev_zone_info(struct btrfs_device *device)
629 struct btrfs_zoned_device_info *zone_info = device->zone_info;
634 bitmap_free(zone_info->active_zones);
635 bitmap_free(zone_info->seq_zones);
636 bitmap_free(zone_info->empty_zones);
637 vfree(zone_info->zone_cache);
639 device->zone_info = NULL;
642 int btrfs_get_dev_zone(struct btrfs_device *device, u64 pos,
643 struct blk_zone *zone)
645 unsigned int nr_zones = 1;
648 ret = btrfs_get_dev_zones(device, pos, zone, &nr_zones);
649 if (ret != 0 || !nr_zones)
650 return ret ? ret : -EIO;
655 static int btrfs_check_for_zoned_device(struct btrfs_fs_info *fs_info)
657 struct btrfs_device *device;
659 list_for_each_entry(device, &fs_info->fs_devices->devices, dev_list) {
661 bdev_zoned_model(device->bdev) == BLK_ZONED_HM) {
663 "zoned: mode not enabled but zoned device found: %pg",
672 int btrfs_check_zoned_mode(struct btrfs_fs_info *fs_info)
674 struct btrfs_device *device;
676 u64 max_zone_append_size = 0;
680 * Host-Managed devices can't be used without the ZONED flag. With the
681 * ZONED all devices can be used, using zone emulation if required.
683 if (!btrfs_fs_incompat(fs_info, ZONED))
684 return btrfs_check_for_zoned_device(fs_info);
686 list_for_each_entry(device, &fs_info->fs_devices->devices, dev_list) {
687 struct btrfs_zoned_device_info *zone_info = device->zone_info;
693 zone_size = zone_info->zone_size;
694 } else if (zone_info->zone_size != zone_size) {
696 "zoned: unequal block device zone sizes: have %llu found %llu",
697 zone_info->zone_size, zone_size);
700 if (!max_zone_append_size ||
701 (zone_info->max_zone_append_size &&
702 zone_info->max_zone_append_size < max_zone_append_size))
703 max_zone_append_size = zone_info->max_zone_append_size;
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);
718 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
719 btrfs_err(fs_info, "zoned: mixed block groups not supported");
723 fs_info->zone_size = zone_size;
724 fs_info->max_zone_append_size = ALIGN_DOWN(max_zone_append_size,
725 fs_info->sectorsize);
726 fs_info->fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_ZONED;
727 if (fs_info->max_zone_append_size < fs_info->max_extent_size)
728 fs_info->max_extent_size = fs_info->max_zone_append_size;
731 * Check mount options here, because we might change fs_info->zoned
732 * from fs_info->zone_size.
734 ret = btrfs_check_mountopts_zoned(fs_info);
738 btrfs_info(fs_info, "zoned mode enabled with zone size %llu", zone_size);
742 int btrfs_check_mountopts_zoned(struct btrfs_fs_info *info)
744 if (!btrfs_is_zoned(info))
748 * Space cache writing is not COWed. Disable that to avoid write errors
749 * in sequential zones.
751 if (btrfs_test_opt(info, SPACE_CACHE)) {
752 btrfs_err(info, "zoned: space cache v1 is not supported");
756 if (btrfs_test_opt(info, NODATACOW)) {
757 btrfs_err(info, "zoned: NODATACOW not supported");
764 static int sb_log_location(struct block_device *bdev, struct blk_zone *zones,
765 int rw, u64 *bytenr_ret)
770 if (zones[0].type == BLK_ZONE_TYPE_CONVENTIONAL) {
771 *bytenr_ret = zones[0].start << SECTOR_SHIFT;
775 ret = sb_write_pointer(bdev, zones, &wp);
776 if (ret != -ENOENT && ret < 0)
780 struct blk_zone *reset = NULL;
782 if (wp == zones[0].start << SECTOR_SHIFT)
784 else if (wp == zones[1].start << SECTOR_SHIFT)
787 if (reset && reset->cond != BLK_ZONE_COND_EMPTY) {
788 ASSERT(sb_zone_is_full(reset));
790 ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
791 reset->start, reset->len,
796 reset->cond = BLK_ZONE_COND_EMPTY;
797 reset->wp = reset->start;
799 } else if (ret != -ENOENT) {
801 * For READ, we want the previous one. Move write pointer to
802 * the end of a zone, if it is at the head of a zone.
806 if (wp == zones[0].start << SECTOR_SHIFT)
807 zone_end = zones[1].start + zones[1].capacity;
808 else if (wp == zones[1].start << SECTOR_SHIFT)
809 zone_end = zones[0].start + zones[0].capacity;
811 wp = ALIGN_DOWN(zone_end << SECTOR_SHIFT,
812 BTRFS_SUPER_INFO_SIZE);
814 wp -= BTRFS_SUPER_INFO_SIZE;
822 int btrfs_sb_log_location_bdev(struct block_device *bdev, int mirror, int rw,
825 struct blk_zone zones[BTRFS_NR_SB_LOG_ZONES];
826 sector_t zone_sectors;
829 u8 zone_sectors_shift;
833 if (!bdev_is_zoned(bdev)) {
834 *bytenr_ret = btrfs_sb_offset(mirror);
838 ASSERT(rw == READ || rw == WRITE);
840 zone_sectors = bdev_zone_sectors(bdev);
841 if (!is_power_of_2(zone_sectors))
843 zone_sectors_shift = ilog2(zone_sectors);
844 nr_sectors = bdev_nr_sectors(bdev);
845 nr_zones = nr_sectors >> zone_sectors_shift;
847 sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
848 if (sb_zone + 1 >= nr_zones)
851 ret = blkdev_report_zones(bdev, zone_start_sector(sb_zone, bdev),
852 BTRFS_NR_SB_LOG_ZONES, copy_zone_info_cb,
856 if (ret != BTRFS_NR_SB_LOG_ZONES)
859 return sb_log_location(bdev, zones, rw, bytenr_ret);
862 int btrfs_sb_log_location(struct btrfs_device *device, int mirror, int rw,
865 struct btrfs_zoned_device_info *zinfo = device->zone_info;
869 * For a zoned filesystem on a non-zoned block device, use the same
870 * super block locations as regular filesystem. Doing so, the super
871 * block can always be retrieved and the zoned flag of the volume
872 * detected from the super block information.
874 if (!bdev_is_zoned(device->bdev)) {
875 *bytenr_ret = btrfs_sb_offset(mirror);
879 zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
880 if (zone_num + 1 >= zinfo->nr_zones)
883 return sb_log_location(device->bdev,
884 &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror],
888 static inline bool is_sb_log_zone(struct btrfs_zoned_device_info *zinfo,
896 zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
897 if (zone_num + 1 >= zinfo->nr_zones)
900 if (!test_bit(zone_num, zinfo->seq_zones))
906 int btrfs_advance_sb_log(struct btrfs_device *device, int mirror)
908 struct btrfs_zoned_device_info *zinfo = device->zone_info;
909 struct blk_zone *zone;
912 if (!is_sb_log_zone(zinfo, mirror))
915 zone = &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror];
916 for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) {
917 /* Advance the next zone */
918 if (zone->cond == BLK_ZONE_COND_FULL) {
923 if (zone->cond == BLK_ZONE_COND_EMPTY)
924 zone->cond = BLK_ZONE_COND_IMP_OPEN;
926 zone->wp += SUPER_INFO_SECTORS;
928 if (sb_zone_is_full(zone)) {
930 * No room left to write new superblock. Since
931 * superblock is written with REQ_SYNC, it is safe to
932 * finish the zone now.
934 * If the write pointer is exactly at the capacity,
935 * explicit ZONE_FINISH is not necessary.
937 if (zone->wp != zone->start + zone->capacity) {
940 ret = blkdev_zone_mgmt(device->bdev,
941 REQ_OP_ZONE_FINISH, zone->start,
942 zone->len, GFP_NOFS);
947 zone->wp = zone->start + zone->len;
948 zone->cond = BLK_ZONE_COND_FULL;
953 /* All the zones are FULL. Should not reach here. */
958 int btrfs_reset_sb_log_zones(struct block_device *bdev, int mirror)
960 sector_t zone_sectors;
962 u8 zone_sectors_shift;
966 zone_sectors = bdev_zone_sectors(bdev);
967 zone_sectors_shift = ilog2(zone_sectors);
968 nr_sectors = bdev_nr_sectors(bdev);
969 nr_zones = nr_sectors >> zone_sectors_shift;
971 sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
972 if (sb_zone + 1 >= nr_zones)
975 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
976 zone_start_sector(sb_zone, bdev),
977 zone_sectors * BTRFS_NR_SB_LOG_ZONES, GFP_NOFS);
981 * btrfs_find_allocatable_zones - find allocatable zones within a given region
983 * @device: the device to allocate a region on
984 * @hole_start: the position of the hole to allocate the region
985 * @num_bytes: size of wanted region
986 * @hole_end: the end of the hole
987 * @return: position of allocatable zones
989 * Allocatable region should not contain any superblock locations.
991 u64 btrfs_find_allocatable_zones(struct btrfs_device *device, u64 hole_start,
992 u64 hole_end, u64 num_bytes)
994 struct btrfs_zoned_device_info *zinfo = device->zone_info;
995 const u8 shift = zinfo->zone_size_shift;
996 u64 nzones = num_bytes >> shift;
997 u64 pos = hole_start;
1002 ASSERT(IS_ALIGNED(hole_start, zinfo->zone_size));
1003 ASSERT(IS_ALIGNED(num_bytes, zinfo->zone_size));
1005 while (pos < hole_end) {
1006 begin = pos >> shift;
1007 end = begin + nzones;
1009 if (end > zinfo->nr_zones)
1012 /* Check if zones in the region are all empty */
1013 if (btrfs_dev_is_sequential(device, pos) &&
1014 find_next_zero_bit(zinfo->empty_zones, end, begin) != end) {
1015 pos += zinfo->zone_size;
1020 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1024 sb_zone = sb_zone_number(shift, i);
1025 if (!(end <= sb_zone ||
1026 sb_zone + BTRFS_NR_SB_LOG_ZONES <= begin)) {
1028 pos = zone_start_physical(
1029 sb_zone + BTRFS_NR_SB_LOG_ZONES, zinfo);
1033 /* We also need to exclude regular superblock positions */
1034 sb_pos = btrfs_sb_offset(i);
1035 if (!(pos + num_bytes <= sb_pos ||
1036 sb_pos + BTRFS_SUPER_INFO_SIZE <= pos)) {
1038 pos = ALIGN(sb_pos + BTRFS_SUPER_INFO_SIZE,
1050 static bool btrfs_dev_set_active_zone(struct btrfs_device *device, u64 pos)
1052 struct btrfs_zoned_device_info *zone_info = device->zone_info;
1053 unsigned int zno = (pos >> zone_info->zone_size_shift);
1055 /* We can use any number of zones */
1056 if (zone_info->max_active_zones == 0)
1059 if (!test_bit(zno, zone_info->active_zones)) {
1060 /* Active zone left? */
1061 if (atomic_dec_if_positive(&zone_info->active_zones_left) < 0)
1063 if (test_and_set_bit(zno, zone_info->active_zones)) {
1064 /* Someone already set the bit */
1065 atomic_inc(&zone_info->active_zones_left);
1072 static void btrfs_dev_clear_active_zone(struct btrfs_device *device, u64 pos)
1074 struct btrfs_zoned_device_info *zone_info = device->zone_info;
1075 unsigned int zno = (pos >> zone_info->zone_size_shift);
1077 /* We can use any number of zones */
1078 if (zone_info->max_active_zones == 0)
1081 if (test_and_clear_bit(zno, zone_info->active_zones))
1082 atomic_inc(&zone_info->active_zones_left);
1085 int btrfs_reset_device_zone(struct btrfs_device *device, u64 physical,
1086 u64 length, u64 *bytes)
1091 ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_RESET,
1092 physical >> SECTOR_SHIFT, length >> SECTOR_SHIFT,
1099 btrfs_dev_set_zone_empty(device, physical);
1100 btrfs_dev_clear_active_zone(device, physical);
1101 physical += device->zone_info->zone_size;
1102 length -= device->zone_info->zone_size;
1108 int btrfs_ensure_empty_zones(struct btrfs_device *device, u64 start, u64 size)
1110 struct btrfs_zoned_device_info *zinfo = device->zone_info;
1111 const u8 shift = zinfo->zone_size_shift;
1112 unsigned long begin = start >> shift;
1113 unsigned long end = (start + size) >> shift;
1117 ASSERT(IS_ALIGNED(start, zinfo->zone_size));
1118 ASSERT(IS_ALIGNED(size, zinfo->zone_size));
1120 if (end > zinfo->nr_zones)
1123 /* All the zones are conventional */
1124 if (find_next_bit(zinfo->seq_zones, begin, end) == end)
1127 /* All the zones are sequential and empty */
1128 if (find_next_zero_bit(zinfo->seq_zones, begin, end) == end &&
1129 find_next_zero_bit(zinfo->empty_zones, begin, end) == end)
1132 for (pos = start; pos < start + size; pos += zinfo->zone_size) {
1135 if (!btrfs_dev_is_sequential(device, pos) ||
1136 btrfs_dev_is_empty_zone(device, pos))
1139 /* Free regions should be empty */
1142 "zoned: resetting device %s (devid %llu) zone %llu for allocation",
1143 rcu_str_deref(device->name), device->devid, pos >> shift);
1146 ret = btrfs_reset_device_zone(device, pos, zinfo->zone_size,
1156 * Calculate an allocation pointer from the extent allocation information
1157 * for a block group consist of conventional zones. It is pointed to the
1158 * end of the highest addressed extent in the block group as an allocation
1161 static int calculate_alloc_pointer(struct btrfs_block_group *cache,
1162 u64 *offset_ret, bool new)
1164 struct btrfs_fs_info *fs_info = cache->fs_info;
1165 struct btrfs_root *root;
1166 struct btrfs_path *path;
1167 struct btrfs_key key;
1168 struct btrfs_key found_key;
1173 * Avoid tree lookups for a new block group, there's no use for it.
1174 * It must always be 0.
1176 * Also, we have a lock chain of extent buffer lock -> chunk mutex.
1177 * For new a block group, this function is called from
1178 * btrfs_make_block_group() which is already taking the chunk mutex.
1179 * Thus, we cannot call calculate_alloc_pointer() which takes extent
1180 * buffer locks to avoid deadlock.
1187 path = btrfs_alloc_path();
1191 key.objectid = cache->start + cache->length;
1195 root = btrfs_extent_root(fs_info, key.objectid);
1196 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1197 /* We should not find the exact match */
1203 ret = btrfs_previous_extent_item(root, path, cache->start);
1212 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
1214 if (found_key.type == BTRFS_EXTENT_ITEM_KEY)
1215 length = found_key.offset;
1217 length = fs_info->nodesize;
1219 if (!(found_key.objectid >= cache->start &&
1220 found_key.objectid + length <= cache->start + cache->length)) {
1224 *offset_ret = found_key.objectid + length - cache->start;
1228 btrfs_free_path(path);
1232 int btrfs_load_block_group_zone_info(struct btrfs_block_group *cache, bool new)
1234 struct btrfs_fs_info *fs_info = cache->fs_info;
1235 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
1236 struct extent_map *em;
1237 struct map_lookup *map;
1238 struct btrfs_device *device;
1239 u64 logical = cache->start;
1240 u64 length = cache->length;
1243 unsigned int nofs_flag;
1244 u64 *alloc_offsets = NULL;
1246 u64 *physical = NULL;
1247 unsigned long *active = NULL;
1249 u32 num_sequential = 0, num_conventional = 0;
1251 if (!btrfs_is_zoned(fs_info))
1255 if (!IS_ALIGNED(length, fs_info->zone_size)) {
1257 "zoned: block group %llu len %llu unaligned to zone size %llu",
1258 logical, length, fs_info->zone_size);
1262 /* Get the chunk mapping */
1263 read_lock(&em_tree->lock);
1264 em = lookup_extent_mapping(em_tree, logical, length);
1265 read_unlock(&em_tree->lock);
1270 map = em->map_lookup;
1272 cache->physical_map = kmemdup(map, map_lookup_size(map->num_stripes), GFP_NOFS);
1273 if (!cache->physical_map) {
1278 alloc_offsets = kcalloc(map->num_stripes, sizeof(*alloc_offsets), GFP_NOFS);
1279 if (!alloc_offsets) {
1284 caps = kcalloc(map->num_stripes, sizeof(*caps), GFP_NOFS);
1290 physical = kcalloc(map->num_stripes, sizeof(*physical), GFP_NOFS);
1296 active = bitmap_zalloc(map->num_stripes, GFP_NOFS);
1302 for (i = 0; i < map->num_stripes; i++) {
1304 struct blk_zone zone;
1305 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1306 int dev_replace_is_ongoing = 0;
1308 device = map->stripes[i].dev;
1309 physical[i] = map->stripes[i].physical;
1311 if (device->bdev == NULL) {
1312 alloc_offsets[i] = WP_MISSING_DEV;
1316 is_sequential = btrfs_dev_is_sequential(device, physical[i]);
1323 * Consider a zone as active if we can allow any number of
1326 if (!device->zone_info->max_active_zones)
1327 __set_bit(i, active);
1329 if (!is_sequential) {
1330 alloc_offsets[i] = WP_CONVENTIONAL;
1335 * This zone will be used for allocation, so mark this zone
1338 btrfs_dev_clear_zone_empty(device, physical[i]);
1340 down_read(&dev_replace->rwsem);
1341 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
1342 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
1343 btrfs_dev_clear_zone_empty(dev_replace->tgtdev, physical[i]);
1344 up_read(&dev_replace->rwsem);
1347 * The group is mapped to a sequential zone. Get the zone write
1348 * pointer to determine the allocation offset within the zone.
1350 WARN_ON(!IS_ALIGNED(physical[i], fs_info->zone_size));
1351 nofs_flag = memalloc_nofs_save();
1352 ret = btrfs_get_dev_zone(device, physical[i], &zone);
1353 memalloc_nofs_restore(nofs_flag);
1354 if (ret == -EIO || ret == -EOPNOTSUPP) {
1356 alloc_offsets[i] = WP_MISSING_DEV;
1362 if (zone.type == BLK_ZONE_TYPE_CONVENTIONAL) {
1363 btrfs_err_in_rcu(fs_info,
1364 "zoned: unexpected conventional zone %llu on device %s (devid %llu)",
1365 zone.start << SECTOR_SHIFT,
1366 rcu_str_deref(device->name), device->devid);
1371 caps[i] = (zone.capacity << SECTOR_SHIFT);
1373 switch (zone.cond) {
1374 case BLK_ZONE_COND_OFFLINE:
1375 case BLK_ZONE_COND_READONLY:
1377 "zoned: offline/readonly zone %llu on device %s (devid %llu)",
1378 physical[i] >> device->zone_info->zone_size_shift,
1379 rcu_str_deref(device->name), device->devid);
1380 alloc_offsets[i] = WP_MISSING_DEV;
1382 case BLK_ZONE_COND_EMPTY:
1383 alloc_offsets[i] = 0;
1385 case BLK_ZONE_COND_FULL:
1386 alloc_offsets[i] = caps[i];
1389 /* Partially used zone */
1391 ((zone.wp - zone.start) << SECTOR_SHIFT);
1392 __set_bit(i, active);
1397 if (num_sequential > 0)
1398 cache->seq_zone = true;
1400 if (num_conventional > 0) {
1401 /* Zone capacity is always zone size in emulation */
1402 cache->zone_capacity = cache->length;
1403 ret = calculate_alloc_pointer(cache, &last_alloc, new);
1406 "zoned: failed to determine allocation offset of bg %llu",
1409 } else if (map->num_stripes == num_conventional) {
1410 cache->alloc_offset = last_alloc;
1411 set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &cache->runtime_flags);
1416 switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
1417 case 0: /* single */
1418 if (alloc_offsets[0] == WP_MISSING_DEV) {
1420 "zoned: cannot recover write pointer for zone %llu",
1425 cache->alloc_offset = alloc_offsets[0];
1426 cache->zone_capacity = caps[0];
1427 if (test_bit(0, active))
1428 set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &cache->runtime_flags);
1430 case BTRFS_BLOCK_GROUP_DUP:
1431 if (map->type & BTRFS_BLOCK_GROUP_DATA) {
1432 btrfs_err(fs_info, "zoned: profile DUP not yet supported on data bg");
1436 if (alloc_offsets[0] == WP_MISSING_DEV) {
1438 "zoned: cannot recover write pointer for zone %llu",
1443 if (alloc_offsets[1] == WP_MISSING_DEV) {
1445 "zoned: cannot recover write pointer for zone %llu",
1450 if (alloc_offsets[0] != alloc_offsets[1]) {
1452 "zoned: write pointer offset mismatch of zones in DUP profile");
1456 if (test_bit(0, active) != test_bit(1, active)) {
1457 if (!btrfs_zone_activate(cache)) {
1462 if (test_bit(0, active))
1463 set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE,
1464 &cache->runtime_flags);
1466 cache->alloc_offset = alloc_offsets[0];
1467 cache->zone_capacity = min(caps[0], caps[1]);
1469 case BTRFS_BLOCK_GROUP_RAID1:
1470 case BTRFS_BLOCK_GROUP_RAID0:
1471 case BTRFS_BLOCK_GROUP_RAID10:
1472 case BTRFS_BLOCK_GROUP_RAID5:
1473 case BTRFS_BLOCK_GROUP_RAID6:
1474 /* non-single profiles are not supported yet */
1476 btrfs_err(fs_info, "zoned: profile %s not yet supported",
1477 btrfs_bg_type_to_raid_name(map->type));
1483 if (cache->alloc_offset > fs_info->zone_size) {
1485 "zoned: invalid write pointer %llu in block group %llu",
1486 cache->alloc_offset, cache->start);
1490 if (cache->alloc_offset > cache->zone_capacity) {
1492 "zoned: invalid write pointer %llu (larger than zone capacity %llu) in block group %llu",
1493 cache->alloc_offset, cache->zone_capacity,
1498 /* An extent is allocated after the write pointer */
1499 if (!ret && num_conventional && last_alloc > cache->alloc_offset) {
1501 "zoned: got wrong write pointer in BG %llu: %llu > %llu",
1502 logical, last_alloc, cache->alloc_offset);
1507 cache->meta_write_pointer = cache->alloc_offset + cache->start;
1508 if (test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &cache->runtime_flags)) {
1509 btrfs_get_block_group(cache);
1510 spin_lock(&fs_info->zone_active_bgs_lock);
1511 list_add_tail(&cache->active_bg_list,
1512 &fs_info->zone_active_bgs);
1513 spin_unlock(&fs_info->zone_active_bgs_lock);
1516 kfree(cache->physical_map);
1517 cache->physical_map = NULL;
1519 bitmap_free(active);
1522 kfree(alloc_offsets);
1523 free_extent_map(em);
1528 void btrfs_calc_zone_unusable(struct btrfs_block_group *cache)
1532 if (!btrfs_is_zoned(cache->fs_info))
1535 WARN_ON(cache->bytes_super != 0);
1536 unusable = (cache->alloc_offset - cache->used) +
1537 (cache->length - cache->zone_capacity);
1538 free = cache->zone_capacity - cache->alloc_offset;
1540 /* We only need ->free_space in ALLOC_SEQ block groups */
1541 cache->cached = BTRFS_CACHE_FINISHED;
1542 cache->free_space_ctl->free_space = free;
1543 cache->zone_unusable = unusable;
1546 void btrfs_redirty_list_add(struct btrfs_transaction *trans,
1547 struct extent_buffer *eb)
1549 struct btrfs_fs_info *fs_info = eb->fs_info;
1551 if (!btrfs_is_zoned(fs_info) ||
1552 btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN) ||
1553 !list_empty(&eb->release_list))
1556 set_extent_buffer_dirty(eb);
1557 set_extent_bits_nowait(&trans->dirty_pages, eb->start,
1558 eb->start + eb->len - 1, EXTENT_DIRTY);
1559 memzero_extent_buffer(eb, 0, eb->len);
1560 set_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags);
1562 spin_lock(&trans->releasing_ebs_lock);
1563 list_add_tail(&eb->release_list, &trans->releasing_ebs);
1564 spin_unlock(&trans->releasing_ebs_lock);
1565 atomic_inc(&eb->refs);
1568 void btrfs_free_redirty_list(struct btrfs_transaction *trans)
1570 spin_lock(&trans->releasing_ebs_lock);
1571 while (!list_empty(&trans->releasing_ebs)) {
1572 struct extent_buffer *eb;
1574 eb = list_first_entry(&trans->releasing_ebs,
1575 struct extent_buffer, release_list);
1576 list_del_init(&eb->release_list);
1577 free_extent_buffer(eb);
1579 spin_unlock(&trans->releasing_ebs_lock);
1582 bool btrfs_use_zone_append(struct btrfs_inode *inode, u64 start)
1584 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1585 struct btrfs_block_group *cache;
1588 if (!btrfs_is_zoned(fs_info))
1591 if (!is_data_inode(&inode->vfs_inode))
1595 * Using REQ_OP_ZONE_APPNED for relocation can break assumptions on the
1596 * extent layout the relocation code has.
1597 * Furthermore we have set aside own block-group from which only the
1598 * relocation "process" can allocate and make sure only one process at a
1599 * time can add pages to an extent that gets relocated, so it's safe to
1600 * use regular REQ_OP_WRITE for this special case.
1602 if (btrfs_is_data_reloc_root(inode->root))
1605 cache = btrfs_lookup_block_group(fs_info, start);
1610 ret = cache->seq_zone;
1611 btrfs_put_block_group(cache);
1616 void btrfs_record_physical_zoned(struct inode *inode, u64 file_offset,
1619 struct btrfs_ordered_extent *ordered;
1620 const u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
1622 if (bio_op(bio) != REQ_OP_ZONE_APPEND)
1625 ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode), file_offset);
1626 if (WARN_ON(!ordered))
1629 ordered->physical = physical;
1630 ordered->bdev = bio->bi_bdev;
1632 btrfs_put_ordered_extent(ordered);
1635 void btrfs_rewrite_logical_zoned(struct btrfs_ordered_extent *ordered)
1637 struct btrfs_inode *inode = BTRFS_I(ordered->inode);
1638 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1639 struct extent_map_tree *em_tree;
1640 struct extent_map *em;
1641 struct btrfs_ordered_sum *sum;
1642 u64 orig_logical = ordered->disk_bytenr;
1643 u64 *logical = NULL;
1646 /* Zoned devices should not have partitions. So, we can assume it is 0 */
1647 ASSERT(!bdev_is_partition(ordered->bdev));
1648 if (WARN_ON(!ordered->bdev))
1651 if (WARN_ON(btrfs_rmap_block(fs_info, orig_logical, ordered->bdev,
1652 ordered->physical, &logical, &nr,
1658 if (orig_logical == *logical)
1661 ordered->disk_bytenr = *logical;
1663 em_tree = &inode->extent_tree;
1664 write_lock(&em_tree->lock);
1665 em = search_extent_mapping(em_tree, ordered->file_offset,
1666 ordered->num_bytes);
1667 em->block_start = *logical;
1668 free_extent_map(em);
1669 write_unlock(&em_tree->lock);
1671 list_for_each_entry(sum, &ordered->list, list) {
1672 if (*logical < orig_logical)
1673 sum->bytenr -= orig_logical - *logical;
1675 sum->bytenr += *logical - orig_logical;
1682 bool btrfs_check_meta_write_pointer(struct btrfs_fs_info *fs_info,
1683 struct extent_buffer *eb,
1684 struct btrfs_block_group **cache_ret)
1686 struct btrfs_block_group *cache;
1689 if (!btrfs_is_zoned(fs_info))
1692 cache = btrfs_lookup_block_group(fs_info, eb->start);
1696 if (cache->meta_write_pointer != eb->start) {
1697 btrfs_put_block_group(cache);
1701 cache->meta_write_pointer = eb->start + eb->len;
1709 void btrfs_revert_meta_write_pointer(struct btrfs_block_group *cache,
1710 struct extent_buffer *eb)
1712 if (!btrfs_is_zoned(eb->fs_info) || !cache)
1715 ASSERT(cache->meta_write_pointer == eb->start + eb->len);
1716 cache->meta_write_pointer = eb->start;
1719 int btrfs_zoned_issue_zeroout(struct btrfs_device *device, u64 physical, u64 length)
1721 if (!btrfs_dev_is_sequential(device, physical))
1724 return blkdev_issue_zeroout(device->bdev, physical >> SECTOR_SHIFT,
1725 length >> SECTOR_SHIFT, GFP_NOFS, 0);
1728 static int read_zone_info(struct btrfs_fs_info *fs_info, u64 logical,
1729 struct blk_zone *zone)
1731 struct btrfs_io_context *bioc = NULL;
1732 u64 mapped_length = PAGE_SIZE;
1733 unsigned int nofs_flag;
1737 ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS, logical,
1738 &mapped_length, &bioc);
1739 if (ret || !bioc || mapped_length < PAGE_SIZE) {
1744 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
1749 nofs_flag = memalloc_nofs_save();
1750 nmirrors = (int)bioc->num_stripes;
1751 for (i = 0; i < nmirrors; i++) {
1752 u64 physical = bioc->stripes[i].physical;
1753 struct btrfs_device *dev = bioc->stripes[i].dev;
1755 /* Missing device */
1759 ret = btrfs_get_dev_zone(dev, physical, zone);
1760 /* Failing device */
1761 if (ret == -EIO || ret == -EOPNOTSUPP)
1765 memalloc_nofs_restore(nofs_flag);
1767 btrfs_put_bioc(bioc);
1772 * Synchronize write pointer in a zone at @physical_start on @tgt_dev, by
1773 * filling zeros between @physical_pos to a write pointer of dev-replace
1776 int btrfs_sync_zone_write_pointer(struct btrfs_device *tgt_dev, u64 logical,
1777 u64 physical_start, u64 physical_pos)
1779 struct btrfs_fs_info *fs_info = tgt_dev->fs_info;
1780 struct blk_zone zone;
1785 if (!btrfs_dev_is_sequential(tgt_dev, physical_pos))
1788 ret = read_zone_info(fs_info, logical, &zone);
1792 wp = physical_start + ((zone.wp - zone.start) << SECTOR_SHIFT);
1794 if (physical_pos == wp)
1797 if (physical_pos > wp)
1800 length = wp - physical_pos;
1801 return btrfs_zoned_issue_zeroout(tgt_dev, physical_pos, length);
1804 struct btrfs_device *btrfs_zoned_get_device(struct btrfs_fs_info *fs_info,
1805 u64 logical, u64 length)
1807 struct btrfs_device *device;
1808 struct extent_map *em;
1809 struct map_lookup *map;
1811 em = btrfs_get_chunk_map(fs_info, logical, length);
1813 return ERR_CAST(em);
1815 map = em->map_lookup;
1816 /* We only support single profile for now */
1817 device = map->stripes[0].dev;
1819 free_extent_map(em);
1825 * Activate block group and underlying device zones
1827 * @block_group: the block group to activate
1829 * Return: true on success, false otherwise
1831 bool btrfs_zone_activate(struct btrfs_block_group *block_group)
1833 struct btrfs_fs_info *fs_info = block_group->fs_info;
1834 struct btrfs_space_info *space_info = block_group->space_info;
1835 struct map_lookup *map;
1836 struct btrfs_device *device;
1841 if (!btrfs_is_zoned(block_group->fs_info))
1844 map = block_group->physical_map;
1846 spin_lock(&space_info->lock);
1847 spin_lock(&block_group->lock);
1848 if (test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags)) {
1854 if (btrfs_zoned_bg_is_full(block_group)) {
1859 for (i = 0; i < map->num_stripes; i++) {
1860 device = map->stripes[i].dev;
1861 physical = map->stripes[i].physical;
1863 if (device->zone_info->max_active_zones == 0)
1866 if (!btrfs_dev_set_active_zone(device, physical)) {
1867 /* Cannot activate the zone */
1873 /* Successfully activated all the zones */
1874 set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags);
1875 space_info->active_total_bytes += block_group->length;
1876 spin_unlock(&block_group->lock);
1877 btrfs_try_granting_tickets(fs_info, space_info);
1878 spin_unlock(&space_info->lock);
1880 /* For the active block group list */
1881 btrfs_get_block_group(block_group);
1883 spin_lock(&fs_info->zone_active_bgs_lock);
1884 list_add_tail(&block_group->active_bg_list, &fs_info->zone_active_bgs);
1885 spin_unlock(&fs_info->zone_active_bgs_lock);
1890 spin_unlock(&block_group->lock);
1891 spin_unlock(&space_info->lock);
1895 static void wait_eb_writebacks(struct btrfs_block_group *block_group)
1897 struct btrfs_fs_info *fs_info = block_group->fs_info;
1898 const u64 end = block_group->start + block_group->length;
1899 struct radix_tree_iter iter;
1900 struct extent_buffer *eb;
1904 radix_tree_for_each_slot(slot, &fs_info->buffer_radix, &iter,
1905 block_group->start >> fs_info->sectorsize_bits) {
1906 eb = radix_tree_deref_slot(slot);
1909 if (radix_tree_deref_retry(eb)) {
1910 slot = radix_tree_iter_retry(&iter);
1914 if (eb->start < block_group->start)
1916 if (eb->start >= end)
1919 slot = radix_tree_iter_resume(slot, &iter);
1921 wait_on_extent_buffer_writeback(eb);
1927 static int do_zone_finish(struct btrfs_block_group *block_group, bool fully_written)
1929 struct btrfs_fs_info *fs_info = block_group->fs_info;
1930 struct map_lookup *map;
1931 const bool is_metadata = (block_group->flags &
1932 (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_SYSTEM));
1936 spin_lock(&block_group->lock);
1937 if (!test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags)) {
1938 spin_unlock(&block_group->lock);
1942 /* Check if we have unwritten allocated space */
1944 block_group->start + block_group->alloc_offset > block_group->meta_write_pointer) {
1945 spin_unlock(&block_group->lock);
1950 * If we are sure that the block group is full (= no more room left for
1951 * new allocation) and the IO for the last usable block is completed, we
1952 * don't need to wait for the other IOs. This holds because we ensure
1953 * the sequential IO submissions using the ZONE_APPEND command for data
1954 * and block_group->meta_write_pointer for metadata.
1956 if (!fully_written) {
1957 spin_unlock(&block_group->lock);
1959 ret = btrfs_inc_block_group_ro(block_group, false);
1963 /* Ensure all writes in this block group finish */
1964 btrfs_wait_block_group_reservations(block_group);
1965 /* No need to wait for NOCOW writers. Zoned mode does not allow that */
1966 btrfs_wait_ordered_roots(fs_info, U64_MAX, block_group->start,
1967 block_group->length);
1968 /* Wait for extent buffers to be written. */
1970 wait_eb_writebacks(block_group);
1972 spin_lock(&block_group->lock);
1975 * Bail out if someone already deactivated the block group, or
1976 * allocated space is left in the block group.
1978 if (!test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE,
1979 &block_group->runtime_flags)) {
1980 spin_unlock(&block_group->lock);
1981 btrfs_dec_block_group_ro(block_group);
1985 if (block_group->reserved) {
1986 spin_unlock(&block_group->lock);
1987 btrfs_dec_block_group_ro(block_group);
1992 clear_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags);
1993 block_group->alloc_offset = block_group->zone_capacity;
1994 block_group->free_space_ctl->free_space = 0;
1995 btrfs_clear_treelog_bg(block_group);
1996 btrfs_clear_data_reloc_bg(block_group);
1997 spin_unlock(&block_group->lock);
1999 map = block_group->physical_map;
2000 for (i = 0; i < map->num_stripes; i++) {
2001 struct btrfs_device *device = map->stripes[i].dev;
2002 const u64 physical = map->stripes[i].physical;
2004 if (device->zone_info->max_active_zones == 0)
2007 ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_FINISH,
2008 physical >> SECTOR_SHIFT,
2009 device->zone_info->zone_size >> SECTOR_SHIFT,
2015 btrfs_dev_clear_active_zone(device, physical);
2019 btrfs_dec_block_group_ro(block_group);
2021 spin_lock(&fs_info->zone_active_bgs_lock);
2022 ASSERT(!list_empty(&block_group->active_bg_list));
2023 list_del_init(&block_group->active_bg_list);
2024 spin_unlock(&fs_info->zone_active_bgs_lock);
2026 /* For active_bg_list */
2027 btrfs_put_block_group(block_group);
2029 clear_and_wake_up_bit(BTRFS_FS_NEED_ZONE_FINISH, &fs_info->flags);
2034 int btrfs_zone_finish(struct btrfs_block_group *block_group)
2036 if (!btrfs_is_zoned(block_group->fs_info))
2039 return do_zone_finish(block_group, false);
2042 bool btrfs_can_activate_zone(struct btrfs_fs_devices *fs_devices, u64 flags)
2044 struct btrfs_fs_info *fs_info = fs_devices->fs_info;
2045 struct btrfs_device *device;
2048 if (!btrfs_is_zoned(fs_info))
2051 /* Check if there is a device with active zones left */
2052 mutex_lock(&fs_info->chunk_mutex);
2053 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
2054 struct btrfs_zoned_device_info *zinfo = device->zone_info;
2059 if (!zinfo->max_active_zones ||
2060 atomic_read(&zinfo->active_zones_left)) {
2065 mutex_unlock(&fs_info->chunk_mutex);
2068 set_bit(BTRFS_FS_NEED_ZONE_FINISH, &fs_info->flags);
2073 void btrfs_zone_finish_endio(struct btrfs_fs_info *fs_info, u64 logical, u64 length)
2075 struct btrfs_block_group *block_group;
2076 u64 min_alloc_bytes;
2078 if (!btrfs_is_zoned(fs_info))
2081 block_group = btrfs_lookup_block_group(fs_info, logical);
2082 ASSERT(block_group);
2084 /* No MIXED_BG on zoned btrfs. */
2085 if (block_group->flags & BTRFS_BLOCK_GROUP_DATA)
2086 min_alloc_bytes = fs_info->sectorsize;
2088 min_alloc_bytes = fs_info->nodesize;
2090 /* Bail out if we can allocate more data from this block group. */
2091 if (logical + length + min_alloc_bytes <=
2092 block_group->start + block_group->zone_capacity)
2095 do_zone_finish(block_group, true);
2098 btrfs_put_block_group(block_group);
2101 static void btrfs_zone_finish_endio_workfn(struct work_struct *work)
2103 struct btrfs_block_group *bg =
2104 container_of(work, struct btrfs_block_group, zone_finish_work);
2106 wait_on_extent_buffer_writeback(bg->last_eb);
2107 free_extent_buffer(bg->last_eb);
2108 btrfs_zone_finish_endio(bg->fs_info, bg->start, bg->length);
2109 btrfs_put_block_group(bg);
2112 void btrfs_schedule_zone_finish_bg(struct btrfs_block_group *bg,
2113 struct extent_buffer *eb)
2115 if (!bg->seq_zone || eb->start + eb->len * 2 <= bg->start + bg->zone_capacity)
2118 if (WARN_ON(bg->zone_finish_work.func == btrfs_zone_finish_endio_workfn)) {
2119 btrfs_err(bg->fs_info, "double scheduling of bg %llu zone finishing",
2125 btrfs_get_block_group(bg);
2126 atomic_inc(&eb->refs);
2128 INIT_WORK(&bg->zone_finish_work, btrfs_zone_finish_endio_workfn);
2129 queue_work(system_unbound_wq, &bg->zone_finish_work);
2132 void btrfs_clear_data_reloc_bg(struct btrfs_block_group *bg)
2134 struct btrfs_fs_info *fs_info = bg->fs_info;
2136 spin_lock(&fs_info->relocation_bg_lock);
2137 if (fs_info->data_reloc_bg == bg->start)
2138 fs_info->data_reloc_bg = 0;
2139 spin_unlock(&fs_info->relocation_bg_lock);
2142 void btrfs_free_zone_cache(struct btrfs_fs_info *fs_info)
2144 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2145 struct btrfs_device *device;
2147 if (!btrfs_is_zoned(fs_info))
2150 mutex_lock(&fs_devices->device_list_mutex);
2151 list_for_each_entry(device, &fs_devices->devices, dev_list) {
2152 if (device->zone_info) {
2153 vfree(device->zone_info->zone_cache);
2154 device->zone_info->zone_cache = NULL;
2157 mutex_unlock(&fs_devices->device_list_mutex);
2160 bool btrfs_zoned_should_reclaim(struct btrfs_fs_info *fs_info)
2162 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2163 struct btrfs_device *device;
2168 ASSERT(btrfs_is_zoned(fs_info));
2170 if (fs_info->bg_reclaim_threshold == 0)
2173 mutex_lock(&fs_devices->device_list_mutex);
2174 list_for_each_entry(device, &fs_devices->devices, dev_list) {
2178 total += device->disk_total_bytes;
2179 used += device->bytes_used;
2181 mutex_unlock(&fs_devices->device_list_mutex);
2183 factor = div64_u64(used * 100, total);
2184 return factor >= fs_info->bg_reclaim_threshold;
2187 void btrfs_zoned_release_data_reloc_bg(struct btrfs_fs_info *fs_info, u64 logical,
2190 struct btrfs_block_group *block_group;
2192 if (!btrfs_is_zoned(fs_info))
2195 block_group = btrfs_lookup_block_group(fs_info, logical);
2196 /* It should be called on a previous data relocation block group. */
2197 ASSERT(block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA));
2199 spin_lock(&block_group->lock);
2200 if (!test_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags))
2203 /* All relocation extents are written. */
2204 if (block_group->start + block_group->alloc_offset == logical + length) {
2205 /* Now, release this block group for further allocations. */
2206 clear_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC,
2207 &block_group->runtime_flags);
2211 spin_unlock(&block_group->lock);
2212 btrfs_put_block_group(block_group);
2215 int btrfs_zone_finish_one_bg(struct btrfs_fs_info *fs_info)
2217 struct btrfs_block_group *block_group;
2218 struct btrfs_block_group *min_bg = NULL;
2219 u64 min_avail = U64_MAX;
2222 spin_lock(&fs_info->zone_active_bgs_lock);
2223 list_for_each_entry(block_group, &fs_info->zone_active_bgs,
2227 spin_lock(&block_group->lock);
2228 if (block_group->reserved ||
2229 (block_group->flags & BTRFS_BLOCK_GROUP_SYSTEM)) {
2230 spin_unlock(&block_group->lock);
2234 avail = block_group->zone_capacity - block_group->alloc_offset;
2235 if (min_avail > avail) {
2237 btrfs_put_block_group(min_bg);
2238 min_bg = block_group;
2240 btrfs_get_block_group(min_bg);
2242 spin_unlock(&block_group->lock);
2244 spin_unlock(&fs_info->zone_active_bgs_lock);
2249 ret = btrfs_zone_finish(min_bg);
2250 btrfs_put_block_group(min_bg);
2252 return ret < 0 ? ret : 1;
2255 int btrfs_zoned_activate_one_bg(struct btrfs_fs_info *fs_info,
2256 struct btrfs_space_info *space_info,
2259 struct btrfs_block_group *bg;
2262 if (!btrfs_is_zoned(fs_info) || (space_info->flags & BTRFS_BLOCK_GROUP_DATA))
2265 /* No more block groups to activate */
2266 if (space_info->active_total_bytes == space_info->total_bytes)
2271 bool need_finish = false;
2273 down_read(&space_info->groups_sem);
2274 for (index = 0; index < BTRFS_NR_RAID_TYPES; index++) {
2275 list_for_each_entry(bg, &space_info->block_groups[index],
2277 if (!spin_trylock(&bg->lock))
2279 if (btrfs_zoned_bg_is_full(bg) ||
2280 test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE,
2281 &bg->runtime_flags)) {
2282 spin_unlock(&bg->lock);
2285 spin_unlock(&bg->lock);
2287 if (btrfs_zone_activate(bg)) {
2288 up_read(&space_info->groups_sem);
2295 up_read(&space_info->groups_sem);
2297 if (!do_finish || !need_finish)
2300 ret = btrfs_zone_finish_one_bg(fs_info);