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>
10 #include "rcu-string.h"
12 #include "block-group.h"
13 #include "transaction.h"
14 #include "dev-replace.h"
15 #include "space-info.h"
17 /* Maximum number of zones to report per blkdev_report_zones() call */
18 #define BTRFS_REPORT_NR_ZONES 4096
19 /* Invalid allocation pointer value for missing devices */
20 #define WP_MISSING_DEV ((u64)-1)
21 /* Pseudo write pointer value for conventional zone */
22 #define WP_CONVENTIONAL ((u64)-2)
25 * Location of the first zone of superblock logging zone pairs.
27 * - primary superblock: 0B (zone 0)
28 * - first copy: 512G (zone starting at that offset)
29 * - second copy: 4T (zone starting at that offset)
31 #define BTRFS_SB_LOG_PRIMARY_OFFSET (0ULL)
32 #define BTRFS_SB_LOG_FIRST_OFFSET (512ULL * SZ_1G)
33 #define BTRFS_SB_LOG_SECOND_OFFSET (4096ULL * SZ_1G)
35 #define BTRFS_SB_LOG_FIRST_SHIFT const_ilog2(BTRFS_SB_LOG_FIRST_OFFSET)
36 #define BTRFS_SB_LOG_SECOND_SHIFT const_ilog2(BTRFS_SB_LOG_SECOND_OFFSET)
38 /* Number of superblock log zones */
39 #define BTRFS_NR_SB_LOG_ZONES 2
42 * Maximum supported zone size. Currently, SMR disks have a zone size of
43 * 256MiB, and we are expecting ZNS drives to be in the 1-4GiB range. We do not
44 * expect the zone size to become larger than 8GiB in the near future.
46 #define BTRFS_MAX_ZONE_SIZE SZ_8G
48 static int copy_zone_info_cb(struct blk_zone *zone, unsigned int idx, void *data)
50 struct blk_zone *zones = data;
52 memcpy(&zones[idx], zone, sizeof(*zone));
57 static int sb_write_pointer(struct block_device *bdev, struct blk_zone *zones,
60 bool empty[BTRFS_NR_SB_LOG_ZONES];
61 bool full[BTRFS_NR_SB_LOG_ZONES];
64 ASSERT(zones[0].type != BLK_ZONE_TYPE_CONVENTIONAL &&
65 zones[1].type != BLK_ZONE_TYPE_CONVENTIONAL);
67 empty[0] = (zones[0].cond == BLK_ZONE_COND_EMPTY);
68 empty[1] = (zones[1].cond == BLK_ZONE_COND_EMPTY);
69 full[0] = (zones[0].cond == BLK_ZONE_COND_FULL);
70 full[1] = (zones[1].cond == BLK_ZONE_COND_FULL);
73 * Possible states of log buffer zones
75 * Empty[0] In use[0] Full[0]
81 * *: Special case, no superblock is written
82 * 0: Use write pointer of zones[0]
83 * 1: Use write pointer of zones[1]
84 * C: Compare super blcoks from zones[0] and zones[1], use the latest
85 * one determined by generation
89 if (empty[0] && empty[1]) {
90 /* Special case to distinguish no superblock to read */
91 *wp_ret = zones[0].start << SECTOR_SHIFT;
93 } else if (full[0] && full[1]) {
94 /* Compare two super blocks */
95 struct address_space *mapping = bdev->bd_inode->i_mapping;
96 struct page *page[BTRFS_NR_SB_LOG_ZONES];
97 struct btrfs_super_block *super[BTRFS_NR_SB_LOG_ZONES];
100 for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) {
103 bytenr = ((zones[i].start + zones[i].len)
104 << SECTOR_SHIFT) - BTRFS_SUPER_INFO_SIZE;
106 page[i] = read_cache_page_gfp(mapping,
107 bytenr >> PAGE_SHIFT, GFP_NOFS);
108 if (IS_ERR(page[i])) {
110 btrfs_release_disk_super(super[0]);
111 return PTR_ERR(page[i]);
113 super[i] = page_address(page[i]);
116 if (super[0]->generation > super[1]->generation)
117 sector = zones[1].start;
119 sector = zones[0].start;
121 for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++)
122 btrfs_release_disk_super(super[i]);
123 } else if (!full[0] && (empty[1] || full[1])) {
124 sector = zones[0].wp;
125 } else if (full[0]) {
126 sector = zones[1].wp;
130 *wp_ret = sector << SECTOR_SHIFT;
135 * Get the first zone number of the superblock mirror
137 static inline u32 sb_zone_number(int shift, int mirror)
141 ASSERT(mirror < BTRFS_SUPER_MIRROR_MAX);
143 case 0: zone = 0; break;
144 case 1: zone = 1ULL << (BTRFS_SB_LOG_FIRST_SHIFT - shift); break;
145 case 2: zone = 1ULL << (BTRFS_SB_LOG_SECOND_SHIFT - shift); break;
148 ASSERT(zone <= U32_MAX);
154 * Emulate blkdev_report_zones() for a non-zoned device. It slices up the block
155 * device into static sized chunks and fake a conventional zone on each of
158 static int emulate_report_zones(struct btrfs_device *device, u64 pos,
159 struct blk_zone *zones, unsigned int nr_zones)
161 const sector_t zone_sectors = device->fs_info->zone_size >> SECTOR_SHIFT;
162 sector_t bdev_size = bdev_nr_sectors(device->bdev);
165 pos >>= SECTOR_SHIFT;
166 for (i = 0; i < nr_zones; i++) {
167 zones[i].start = i * zone_sectors + pos;
168 zones[i].len = zone_sectors;
169 zones[i].capacity = zone_sectors;
170 zones[i].wp = zones[i].start + zone_sectors;
171 zones[i].type = BLK_ZONE_TYPE_CONVENTIONAL;
172 zones[i].cond = BLK_ZONE_COND_NOT_WP;
174 if (zones[i].wp >= bdev_size) {
183 static int btrfs_get_dev_zones(struct btrfs_device *device, u64 pos,
184 struct blk_zone *zones, unsigned int *nr_zones)
191 if (!bdev_is_zoned(device->bdev)) {
192 ret = emulate_report_zones(device, pos, zones, *nr_zones);
197 ret = blkdev_report_zones(device->bdev, pos >> SECTOR_SHIFT, *nr_zones,
198 copy_zone_info_cb, zones);
200 btrfs_err_in_rcu(device->fs_info,
201 "zoned: failed to read zone %llu on %s (devid %llu)",
202 pos, rcu_str_deref(device->name),
213 /* The emulated zone size is determined from the size of device extent */
214 static int calculate_emulated_zone_size(struct btrfs_fs_info *fs_info)
216 struct btrfs_path *path;
217 struct btrfs_root *root = fs_info->dev_root;
218 struct btrfs_key key;
219 struct extent_buffer *leaf;
220 struct btrfs_dev_extent *dext;
224 key.type = BTRFS_DEV_EXTENT_KEY;
227 path = btrfs_alloc_path();
231 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
235 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
236 ret = btrfs_next_item(root, path);
239 /* No dev extents at all? Not good */
246 leaf = path->nodes[0];
247 dext = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent);
248 fs_info->zone_size = btrfs_dev_extent_length(leaf, dext);
252 btrfs_free_path(path);
257 int btrfs_get_dev_zone_info_all_devices(struct btrfs_fs_info *fs_info)
259 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
260 struct btrfs_device *device;
263 /* fs_info->zone_size might not set yet. Use the incomapt flag here. */
264 if (!btrfs_fs_incompat(fs_info, ZONED))
267 mutex_lock(&fs_devices->device_list_mutex);
268 list_for_each_entry(device, &fs_devices->devices, dev_list) {
269 /* We can skip reading of zone info for missing devices */
273 ret = btrfs_get_dev_zone_info(device);
277 mutex_unlock(&fs_devices->device_list_mutex);
282 int btrfs_get_dev_zone_info(struct btrfs_device *device)
284 struct btrfs_fs_info *fs_info = device->fs_info;
285 struct btrfs_zoned_device_info *zone_info = NULL;
286 struct block_device *bdev = device->bdev;
287 struct request_queue *queue = bdev_get_queue(bdev);
290 struct blk_zone *zones = NULL;
291 unsigned int i, nreported = 0, nr_zones;
292 sector_t zone_sectors;
293 char *model, *emulated;
297 * Cannot use btrfs_is_zoned here, since fs_info::zone_size might not
300 if (!btrfs_fs_incompat(fs_info, ZONED))
303 if (device->zone_info)
306 zone_info = kzalloc(sizeof(*zone_info), GFP_KERNEL);
310 if (!bdev_is_zoned(bdev)) {
311 if (!fs_info->zone_size) {
312 ret = calculate_emulated_zone_size(fs_info);
317 ASSERT(fs_info->zone_size);
318 zone_sectors = fs_info->zone_size >> SECTOR_SHIFT;
320 zone_sectors = bdev_zone_sectors(bdev);
323 /* Check if it's power of 2 (see is_power_of_2) */
324 ASSERT(zone_sectors != 0 && (zone_sectors & (zone_sectors - 1)) == 0);
325 zone_info->zone_size = zone_sectors << SECTOR_SHIFT;
327 /* We reject devices with a zone size larger than 8GB */
328 if (zone_info->zone_size > BTRFS_MAX_ZONE_SIZE) {
329 btrfs_err_in_rcu(fs_info,
330 "zoned: %s: zone size %llu larger than supported maximum %llu",
331 rcu_str_deref(device->name),
332 zone_info->zone_size, BTRFS_MAX_ZONE_SIZE);
337 nr_sectors = bdev_nr_sectors(bdev);
338 zone_info->zone_size_shift = ilog2(zone_info->zone_size);
339 zone_info->max_zone_append_size =
340 (u64)queue_max_zone_append_sectors(queue) << SECTOR_SHIFT;
341 zone_info->nr_zones = nr_sectors >> ilog2(zone_sectors);
342 if (!IS_ALIGNED(nr_sectors, zone_sectors))
343 zone_info->nr_zones++;
345 if (bdev_is_zoned(bdev) && zone_info->max_zone_append_size == 0) {
346 btrfs_err(fs_info, "zoned: device %pg does not support zone append",
352 zone_info->seq_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
353 if (!zone_info->seq_zones) {
358 zone_info->empty_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
359 if (!zone_info->empty_zones) {
364 zones = kcalloc(BTRFS_REPORT_NR_ZONES, sizeof(struct blk_zone), GFP_KERNEL);
371 while (sector < nr_sectors) {
372 nr_zones = BTRFS_REPORT_NR_ZONES;
373 ret = btrfs_get_dev_zones(device, sector << SECTOR_SHIFT, zones,
378 for (i = 0; i < nr_zones; i++) {
379 if (zones[i].type == BLK_ZONE_TYPE_SEQWRITE_REQ)
380 __set_bit(nreported, zone_info->seq_zones);
381 if (zones[i].cond == BLK_ZONE_COND_EMPTY)
382 __set_bit(nreported, zone_info->empty_zones);
385 sector = zones[nr_zones - 1].start + zones[nr_zones - 1].len;
388 if (nreported != zone_info->nr_zones) {
389 btrfs_err_in_rcu(device->fs_info,
390 "inconsistent number of zones on %s (%u/%u)",
391 rcu_str_deref(device->name), nreported,
392 zone_info->nr_zones);
397 /* Validate superblock log */
398 nr_zones = BTRFS_NR_SB_LOG_ZONES;
399 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
402 int sb_pos = BTRFS_NR_SB_LOG_ZONES * i;
404 sb_zone = sb_zone_number(zone_info->zone_size_shift, i);
405 if (sb_zone + 1 >= zone_info->nr_zones)
408 sector = sb_zone << (zone_info->zone_size_shift - SECTOR_SHIFT);
409 ret = btrfs_get_dev_zones(device, sector << SECTOR_SHIFT,
410 &zone_info->sb_zones[sb_pos],
415 if (nr_zones != BTRFS_NR_SB_LOG_ZONES) {
416 btrfs_err_in_rcu(device->fs_info,
417 "zoned: failed to read super block log zone info at devid %llu zone %u",
418 device->devid, sb_zone);
424 * If zones[0] is conventional, always use the beggining of the
425 * zone to record superblock. No need to validate in that case.
427 if (zone_info->sb_zones[BTRFS_NR_SB_LOG_ZONES * i].type ==
428 BLK_ZONE_TYPE_CONVENTIONAL)
431 ret = sb_write_pointer(device->bdev,
432 &zone_info->sb_zones[sb_pos], &sb_wp);
433 if (ret != -ENOENT && ret) {
434 btrfs_err_in_rcu(device->fs_info,
435 "zoned: super block log zone corrupted devid %llu zone %u",
436 device->devid, sb_zone);
445 device->zone_info = zone_info;
447 switch (bdev_zoned_model(bdev)) {
449 model = "host-managed zoned";
453 model = "host-aware zoned";
458 emulated = "emulated ";
462 btrfs_err_in_rcu(fs_info, "zoned: unsupported model %d on %s",
463 bdev_zoned_model(bdev),
464 rcu_str_deref(device->name));
466 goto out_free_zone_info;
469 btrfs_info_in_rcu(fs_info,
470 "%s block device %s, %u %szones of %llu bytes",
471 model, rcu_str_deref(device->name), zone_info->nr_zones,
472 emulated, zone_info->zone_size);
479 bitmap_free(zone_info->empty_zones);
480 bitmap_free(zone_info->seq_zones);
482 device->zone_info = NULL;
487 void btrfs_destroy_dev_zone_info(struct btrfs_device *device)
489 struct btrfs_zoned_device_info *zone_info = device->zone_info;
494 bitmap_free(zone_info->seq_zones);
495 bitmap_free(zone_info->empty_zones);
497 device->zone_info = NULL;
500 int btrfs_get_dev_zone(struct btrfs_device *device, u64 pos,
501 struct blk_zone *zone)
503 unsigned int nr_zones = 1;
506 ret = btrfs_get_dev_zones(device, pos, zone, &nr_zones);
507 if (ret != 0 || !nr_zones)
508 return ret ? ret : -EIO;
513 int btrfs_check_zoned_mode(struct btrfs_fs_info *fs_info)
515 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
516 struct btrfs_device *device;
517 u64 zoned_devices = 0;
520 u64 max_zone_append_size = 0;
521 const bool incompat_zoned = btrfs_fs_incompat(fs_info, ZONED);
524 /* Count zoned devices */
525 list_for_each_entry(device, &fs_devices->devices, dev_list) {
526 enum blk_zoned_model model;
531 model = bdev_zoned_model(device->bdev);
533 * A Host-Managed zoned device must be used as a zoned device.
534 * A Host-Aware zoned device and a non-zoned devices can be
535 * treated as a zoned device, if ZONED flag is enabled in the
538 if (model == BLK_ZONED_HM ||
539 (model == BLK_ZONED_HA && incompat_zoned) ||
540 (model == BLK_ZONED_NONE && incompat_zoned)) {
541 struct btrfs_zoned_device_info *zone_info =
544 zone_info = device->zone_info;
547 zone_size = zone_info->zone_size;
548 } else if (zone_info->zone_size != zone_size) {
550 "zoned: unequal block device zone sizes: have %llu found %llu",
551 device->zone_info->zone_size,
556 if (!max_zone_append_size ||
557 (zone_info->max_zone_append_size &&
558 zone_info->max_zone_append_size < max_zone_append_size))
559 max_zone_append_size =
560 zone_info->max_zone_append_size;
565 if (!zoned_devices && !incompat_zoned)
568 if (!zoned_devices && incompat_zoned) {
569 /* No zoned block device found on ZONED filesystem */
571 "zoned: no zoned devices found on a zoned filesystem");
576 if (zoned_devices && !incompat_zoned) {
578 "zoned: mode not enabled but zoned device found");
583 if (zoned_devices != nr_devices) {
585 "zoned: cannot mix zoned and regular devices");
591 * stripe_size is always aligned to BTRFS_STRIPE_LEN in
592 * __btrfs_alloc_chunk(). Since we want stripe_len == zone_size,
593 * check the alignment here.
595 if (!IS_ALIGNED(zone_size, BTRFS_STRIPE_LEN)) {
597 "zoned: zone size %llu not aligned to stripe %u",
598 zone_size, BTRFS_STRIPE_LEN);
603 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
604 btrfs_err(fs_info, "zoned: mixed block groups not supported");
609 fs_info->zone_size = zone_size;
610 fs_info->max_zone_append_size = max_zone_append_size;
611 fs_info->fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_ZONED;
614 * Check mount options here, because we might change fs_info->zoned
615 * from fs_info->zone_size.
617 ret = btrfs_check_mountopts_zoned(fs_info);
621 btrfs_info(fs_info, "zoned mode enabled with zone size %llu", zone_size);
626 int btrfs_check_mountopts_zoned(struct btrfs_fs_info *info)
628 if (!btrfs_is_zoned(info))
632 * Space cache writing is not COWed. Disable that to avoid write errors
633 * in sequential zones.
635 if (btrfs_test_opt(info, SPACE_CACHE)) {
636 btrfs_err(info, "zoned: space cache v1 is not supported");
640 if (btrfs_test_opt(info, NODATACOW)) {
641 btrfs_err(info, "zoned: NODATACOW not supported");
648 static int sb_log_location(struct block_device *bdev, struct blk_zone *zones,
649 int rw, u64 *bytenr_ret)
654 if (zones[0].type == BLK_ZONE_TYPE_CONVENTIONAL) {
655 *bytenr_ret = zones[0].start << SECTOR_SHIFT;
659 ret = sb_write_pointer(bdev, zones, &wp);
660 if (ret != -ENOENT && ret < 0)
664 struct blk_zone *reset = NULL;
666 if (wp == zones[0].start << SECTOR_SHIFT)
668 else if (wp == zones[1].start << SECTOR_SHIFT)
671 if (reset && reset->cond != BLK_ZONE_COND_EMPTY) {
672 ASSERT(reset->cond == BLK_ZONE_COND_FULL);
674 ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
675 reset->start, reset->len,
680 reset->cond = BLK_ZONE_COND_EMPTY;
681 reset->wp = reset->start;
683 } else if (ret != -ENOENT) {
684 /* For READ, we want the precious one */
685 if (wp == zones[0].start << SECTOR_SHIFT)
686 wp = (zones[1].start + zones[1].len) << SECTOR_SHIFT;
687 wp -= BTRFS_SUPER_INFO_SIZE;
695 int btrfs_sb_log_location_bdev(struct block_device *bdev, int mirror, int rw,
698 struct blk_zone zones[BTRFS_NR_SB_LOG_ZONES];
699 sector_t zone_sectors;
702 u8 zone_sectors_shift;
706 if (!bdev_is_zoned(bdev)) {
707 *bytenr_ret = btrfs_sb_offset(mirror);
711 ASSERT(rw == READ || rw == WRITE);
713 zone_sectors = bdev_zone_sectors(bdev);
714 if (!is_power_of_2(zone_sectors))
716 zone_sectors_shift = ilog2(zone_sectors);
717 nr_sectors = bdev_nr_sectors(bdev);
718 nr_zones = nr_sectors >> zone_sectors_shift;
720 sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
721 if (sb_zone + 1 >= nr_zones)
724 ret = blkdev_report_zones(bdev, sb_zone << zone_sectors_shift,
725 BTRFS_NR_SB_LOG_ZONES, copy_zone_info_cb,
729 if (ret != BTRFS_NR_SB_LOG_ZONES)
732 return sb_log_location(bdev, zones, rw, bytenr_ret);
735 int btrfs_sb_log_location(struct btrfs_device *device, int mirror, int rw,
738 struct btrfs_zoned_device_info *zinfo = device->zone_info;
742 * For a zoned filesystem on a non-zoned block device, use the same
743 * super block locations as regular filesystem. Doing so, the super
744 * block can always be retrieved and the zoned flag of the volume
745 * detected from the super block information.
747 if (!bdev_is_zoned(device->bdev)) {
748 *bytenr_ret = btrfs_sb_offset(mirror);
752 zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
753 if (zone_num + 1 >= zinfo->nr_zones)
756 return sb_log_location(device->bdev,
757 &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror],
761 static inline bool is_sb_log_zone(struct btrfs_zoned_device_info *zinfo,
769 zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
770 if (zone_num + 1 >= zinfo->nr_zones)
773 if (!test_bit(zone_num, zinfo->seq_zones))
779 void btrfs_advance_sb_log(struct btrfs_device *device, int mirror)
781 struct btrfs_zoned_device_info *zinfo = device->zone_info;
782 struct blk_zone *zone;
784 if (!is_sb_log_zone(zinfo, mirror))
787 zone = &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror];
788 if (zone->cond != BLK_ZONE_COND_FULL) {
789 if (zone->cond == BLK_ZONE_COND_EMPTY)
790 zone->cond = BLK_ZONE_COND_IMP_OPEN;
792 zone->wp += (BTRFS_SUPER_INFO_SIZE >> SECTOR_SHIFT);
794 if (zone->wp == zone->start + zone->len)
795 zone->cond = BLK_ZONE_COND_FULL;
801 ASSERT(zone->cond != BLK_ZONE_COND_FULL);
802 if (zone->cond == BLK_ZONE_COND_EMPTY)
803 zone->cond = BLK_ZONE_COND_IMP_OPEN;
805 zone->wp += (BTRFS_SUPER_INFO_SIZE >> SECTOR_SHIFT);
807 if (zone->wp == zone->start + zone->len)
808 zone->cond = BLK_ZONE_COND_FULL;
811 int btrfs_reset_sb_log_zones(struct block_device *bdev, int mirror)
813 sector_t zone_sectors;
815 u8 zone_sectors_shift;
819 zone_sectors = bdev_zone_sectors(bdev);
820 zone_sectors_shift = ilog2(zone_sectors);
821 nr_sectors = bdev_nr_sectors(bdev);
822 nr_zones = nr_sectors >> zone_sectors_shift;
824 sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
825 if (sb_zone + 1 >= nr_zones)
828 return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
829 sb_zone << zone_sectors_shift,
830 zone_sectors * BTRFS_NR_SB_LOG_ZONES, GFP_NOFS);
834 * btrfs_find_allocatable_zones - find allocatable zones within a given region
836 * @device: the device to allocate a region on
837 * @hole_start: the position of the hole to allocate the region
838 * @num_bytes: size of wanted region
839 * @hole_end: the end of the hole
840 * @return: position of allocatable zones
842 * Allocatable region should not contain any superblock locations.
844 u64 btrfs_find_allocatable_zones(struct btrfs_device *device, u64 hole_start,
845 u64 hole_end, u64 num_bytes)
847 struct btrfs_zoned_device_info *zinfo = device->zone_info;
848 const u8 shift = zinfo->zone_size_shift;
849 u64 nzones = num_bytes >> shift;
850 u64 pos = hole_start;
855 ASSERT(IS_ALIGNED(hole_start, zinfo->zone_size));
856 ASSERT(IS_ALIGNED(num_bytes, zinfo->zone_size));
858 while (pos < hole_end) {
859 begin = pos >> shift;
860 end = begin + nzones;
862 if (end > zinfo->nr_zones)
865 /* Check if zones in the region are all empty */
866 if (btrfs_dev_is_sequential(device, pos) &&
867 find_next_zero_bit(zinfo->empty_zones, end, begin) != end) {
868 pos += zinfo->zone_size;
873 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
877 sb_zone = sb_zone_number(shift, i);
878 if (!(end <= sb_zone ||
879 sb_zone + BTRFS_NR_SB_LOG_ZONES <= begin)) {
881 pos = ((u64)sb_zone + BTRFS_NR_SB_LOG_ZONES) << shift;
885 /* We also need to exclude regular superblock positions */
886 sb_pos = btrfs_sb_offset(i);
887 if (!(pos + num_bytes <= sb_pos ||
888 sb_pos + BTRFS_SUPER_INFO_SIZE <= pos)) {
890 pos = ALIGN(sb_pos + BTRFS_SUPER_INFO_SIZE,
902 int btrfs_reset_device_zone(struct btrfs_device *device, u64 physical,
903 u64 length, u64 *bytes)
908 ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_RESET,
909 physical >> SECTOR_SHIFT, length >> SECTOR_SHIFT,
916 btrfs_dev_set_zone_empty(device, physical);
917 physical += device->zone_info->zone_size;
918 length -= device->zone_info->zone_size;
924 int btrfs_ensure_empty_zones(struct btrfs_device *device, u64 start, u64 size)
926 struct btrfs_zoned_device_info *zinfo = device->zone_info;
927 const u8 shift = zinfo->zone_size_shift;
928 unsigned long begin = start >> shift;
929 unsigned long end = (start + size) >> shift;
933 ASSERT(IS_ALIGNED(start, zinfo->zone_size));
934 ASSERT(IS_ALIGNED(size, zinfo->zone_size));
936 if (end > zinfo->nr_zones)
939 /* All the zones are conventional */
940 if (find_next_bit(zinfo->seq_zones, begin, end) == end)
943 /* All the zones are sequential and empty */
944 if (find_next_zero_bit(zinfo->seq_zones, begin, end) == end &&
945 find_next_zero_bit(zinfo->empty_zones, begin, end) == end)
948 for (pos = start; pos < start + size; pos += zinfo->zone_size) {
951 if (!btrfs_dev_is_sequential(device, pos) ||
952 btrfs_dev_is_empty_zone(device, pos))
955 /* Free regions should be empty */
958 "zoned: resetting device %s (devid %llu) zone %llu for allocation",
959 rcu_str_deref(device->name), device->devid, pos >> shift);
962 ret = btrfs_reset_device_zone(device, pos, zinfo->zone_size,
972 * Calculate an allocation pointer from the extent allocation information
973 * for a block group consist of conventional zones. It is pointed to the
974 * end of the highest addressed extent in the block group as an allocation
977 static int calculate_alloc_pointer(struct btrfs_block_group *cache,
980 struct btrfs_fs_info *fs_info = cache->fs_info;
981 struct btrfs_root *root = fs_info->extent_root;
982 struct btrfs_path *path;
983 struct btrfs_key key;
984 struct btrfs_key found_key;
988 path = btrfs_alloc_path();
992 key.objectid = cache->start + cache->length;
996 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
997 /* We should not find the exact match */
1003 ret = btrfs_previous_extent_item(root, path, cache->start);
1012 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
1014 if (found_key.type == BTRFS_EXTENT_ITEM_KEY)
1015 length = found_key.offset;
1017 length = fs_info->nodesize;
1019 if (!(found_key.objectid >= cache->start &&
1020 found_key.objectid + length <= cache->start + cache->length)) {
1024 *offset_ret = found_key.objectid + length - cache->start;
1028 btrfs_free_path(path);
1032 int btrfs_load_block_group_zone_info(struct btrfs_block_group *cache, bool new)
1034 struct btrfs_fs_info *fs_info = cache->fs_info;
1035 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
1036 struct extent_map *em;
1037 struct map_lookup *map;
1038 struct btrfs_device *device;
1039 u64 logical = cache->start;
1040 u64 length = cache->length;
1044 unsigned int nofs_flag;
1045 u64 *alloc_offsets = NULL;
1047 u32 num_sequential = 0, num_conventional = 0;
1049 if (!btrfs_is_zoned(fs_info))
1053 if (!IS_ALIGNED(length, fs_info->zone_size)) {
1055 "zoned: block group %llu len %llu unaligned to zone size %llu",
1056 logical, length, fs_info->zone_size);
1060 /* Get the chunk mapping */
1061 read_lock(&em_tree->lock);
1062 em = lookup_extent_mapping(em_tree, logical, length);
1063 read_unlock(&em_tree->lock);
1068 map = em->map_lookup;
1070 alloc_offsets = kcalloc(map->num_stripes, sizeof(*alloc_offsets), GFP_NOFS);
1071 if (!alloc_offsets) {
1072 free_extent_map(em);
1076 for (i = 0; i < map->num_stripes; i++) {
1078 struct blk_zone zone;
1079 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1080 int dev_replace_is_ongoing = 0;
1082 device = map->stripes[i].dev;
1083 physical = map->stripes[i].physical;
1085 if (device->bdev == NULL) {
1086 alloc_offsets[i] = WP_MISSING_DEV;
1090 is_sequential = btrfs_dev_is_sequential(device, physical);
1096 if (!is_sequential) {
1097 alloc_offsets[i] = WP_CONVENTIONAL;
1102 * This zone will be used for allocation, so mark this zone
1105 btrfs_dev_clear_zone_empty(device, physical);
1107 down_read(&dev_replace->rwsem);
1108 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
1109 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
1110 btrfs_dev_clear_zone_empty(dev_replace->tgtdev, physical);
1111 up_read(&dev_replace->rwsem);
1114 * The group is mapped to a sequential zone. Get the zone write
1115 * pointer to determine the allocation offset within the zone.
1117 WARN_ON(!IS_ALIGNED(physical, fs_info->zone_size));
1118 nofs_flag = memalloc_nofs_save();
1119 ret = btrfs_get_dev_zone(device, physical, &zone);
1120 memalloc_nofs_restore(nofs_flag);
1121 if (ret == -EIO || ret == -EOPNOTSUPP) {
1123 alloc_offsets[i] = WP_MISSING_DEV;
1129 switch (zone.cond) {
1130 case BLK_ZONE_COND_OFFLINE:
1131 case BLK_ZONE_COND_READONLY:
1133 "zoned: offline/readonly zone %llu on device %s (devid %llu)",
1134 physical >> device->zone_info->zone_size_shift,
1135 rcu_str_deref(device->name), device->devid);
1136 alloc_offsets[i] = WP_MISSING_DEV;
1138 case BLK_ZONE_COND_EMPTY:
1139 alloc_offsets[i] = 0;
1141 case BLK_ZONE_COND_FULL:
1142 alloc_offsets[i] = fs_info->zone_size;
1145 /* Partially used zone */
1147 ((zone.wp - zone.start) << SECTOR_SHIFT);
1152 if (num_sequential > 0)
1153 cache->seq_zone = true;
1155 if (num_conventional > 0) {
1157 * Avoid calling calculate_alloc_pointer() for new BG. It
1158 * is no use for new BG. It must be always 0.
1160 * Also, we have a lock chain of extent buffer lock ->
1161 * chunk mutex. For new BG, this function is called from
1162 * btrfs_make_block_group() which is already taking the
1163 * chunk mutex. Thus, we cannot call
1164 * calculate_alloc_pointer() which takes extent buffer
1165 * locks to avoid deadlock.
1168 cache->alloc_offset = 0;
1171 ret = calculate_alloc_pointer(cache, &last_alloc);
1172 if (ret || map->num_stripes == num_conventional) {
1174 cache->alloc_offset = last_alloc;
1177 "zoned: failed to determine allocation offset of bg %llu",
1183 switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
1184 case 0: /* single */
1185 cache->alloc_offset = alloc_offsets[0];
1187 case BTRFS_BLOCK_GROUP_DUP:
1188 case BTRFS_BLOCK_GROUP_RAID1:
1189 case BTRFS_BLOCK_GROUP_RAID0:
1190 case BTRFS_BLOCK_GROUP_RAID10:
1191 case BTRFS_BLOCK_GROUP_RAID5:
1192 case BTRFS_BLOCK_GROUP_RAID6:
1193 /* non-single profiles are not supported yet */
1195 btrfs_err(fs_info, "zoned: profile %s not yet supported",
1196 btrfs_bg_type_to_raid_name(map->type));
1202 /* An extent is allocated after the write pointer */
1203 if (!ret && num_conventional && last_alloc > cache->alloc_offset) {
1205 "zoned: got wrong write pointer in BG %llu: %llu > %llu",
1206 logical, last_alloc, cache->alloc_offset);
1211 cache->meta_write_pointer = cache->alloc_offset + cache->start;
1213 kfree(alloc_offsets);
1214 free_extent_map(em);
1219 void btrfs_calc_zone_unusable(struct btrfs_block_group *cache)
1223 if (!btrfs_is_zoned(cache->fs_info))
1226 WARN_ON(cache->bytes_super != 0);
1227 unusable = cache->alloc_offset - cache->used;
1228 free = cache->length - cache->alloc_offset;
1230 /* We only need ->free_space in ALLOC_SEQ block groups */
1231 cache->last_byte_to_unpin = (u64)-1;
1232 cache->cached = BTRFS_CACHE_FINISHED;
1233 cache->free_space_ctl->free_space = free;
1234 cache->zone_unusable = unusable;
1236 /* Should not have any excluded extents. Just in case, though */
1237 btrfs_free_excluded_extents(cache);
1240 void btrfs_redirty_list_add(struct btrfs_transaction *trans,
1241 struct extent_buffer *eb)
1243 struct btrfs_fs_info *fs_info = eb->fs_info;
1245 if (!btrfs_is_zoned(fs_info) ||
1246 btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN) ||
1247 !list_empty(&eb->release_list))
1250 set_extent_buffer_dirty(eb);
1251 set_extent_bits_nowait(&trans->dirty_pages, eb->start,
1252 eb->start + eb->len - 1, EXTENT_DIRTY);
1253 memzero_extent_buffer(eb, 0, eb->len);
1254 set_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags);
1256 spin_lock(&trans->releasing_ebs_lock);
1257 list_add_tail(&eb->release_list, &trans->releasing_ebs);
1258 spin_unlock(&trans->releasing_ebs_lock);
1259 atomic_inc(&eb->refs);
1262 void btrfs_free_redirty_list(struct btrfs_transaction *trans)
1264 spin_lock(&trans->releasing_ebs_lock);
1265 while (!list_empty(&trans->releasing_ebs)) {
1266 struct extent_buffer *eb;
1268 eb = list_first_entry(&trans->releasing_ebs,
1269 struct extent_buffer, release_list);
1270 list_del_init(&eb->release_list);
1271 free_extent_buffer(eb);
1273 spin_unlock(&trans->releasing_ebs_lock);
1276 bool btrfs_use_zone_append(struct btrfs_inode *inode, struct extent_map *em)
1278 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1279 struct btrfs_block_group *cache;
1282 if (!btrfs_is_zoned(fs_info))
1285 if (!fs_info->max_zone_append_size)
1288 if (!is_data_inode(&inode->vfs_inode))
1291 cache = btrfs_lookup_block_group(fs_info, em->block_start);
1296 ret = cache->seq_zone;
1297 btrfs_put_block_group(cache);
1302 void btrfs_record_physical_zoned(struct inode *inode, u64 file_offset,
1305 struct btrfs_ordered_extent *ordered;
1306 const u64 physical = bio->bi_iter.bi_sector << SECTOR_SHIFT;
1308 if (bio_op(bio) != REQ_OP_ZONE_APPEND)
1311 ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode), file_offset);
1312 if (WARN_ON(!ordered))
1315 ordered->physical = physical;
1316 ordered->disk = bio->bi_bdev->bd_disk;
1317 ordered->partno = bio->bi_bdev->bd_partno;
1319 btrfs_put_ordered_extent(ordered);
1322 void btrfs_rewrite_logical_zoned(struct btrfs_ordered_extent *ordered)
1324 struct btrfs_inode *inode = BTRFS_I(ordered->inode);
1325 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1326 struct extent_map_tree *em_tree;
1327 struct extent_map *em;
1328 struct btrfs_ordered_sum *sum;
1329 struct block_device *bdev;
1330 u64 orig_logical = ordered->disk_bytenr;
1331 u64 *logical = NULL;
1334 /* Zoned devices should not have partitions. So, we can assume it is 0 */
1335 ASSERT(ordered->partno == 0);
1336 bdev = bdgrab(ordered->disk->part0);
1340 if (WARN_ON(btrfs_rmap_block(fs_info, orig_logical, bdev,
1341 ordered->physical, &logical, &nr,
1347 if (orig_logical == *logical)
1350 ordered->disk_bytenr = *logical;
1352 em_tree = &inode->extent_tree;
1353 write_lock(&em_tree->lock);
1354 em = search_extent_mapping(em_tree, ordered->file_offset,
1355 ordered->num_bytes);
1356 em->block_start = *logical;
1357 free_extent_map(em);
1358 write_unlock(&em_tree->lock);
1360 list_for_each_entry(sum, &ordered->list, list) {
1361 if (*logical < orig_logical)
1362 sum->bytenr -= orig_logical - *logical;
1364 sum->bytenr += *logical - orig_logical;
1372 bool btrfs_check_meta_write_pointer(struct btrfs_fs_info *fs_info,
1373 struct extent_buffer *eb,
1374 struct btrfs_block_group **cache_ret)
1376 struct btrfs_block_group *cache;
1379 if (!btrfs_is_zoned(fs_info))
1384 if (cache && (eb->start < cache->start ||
1385 cache->start + cache->length <= eb->start)) {
1386 btrfs_put_block_group(cache);
1392 cache = btrfs_lookup_block_group(fs_info, eb->start);
1395 if (cache->meta_write_pointer != eb->start) {
1396 btrfs_put_block_group(cache);
1400 cache->meta_write_pointer = eb->start + eb->len;
1409 void btrfs_revert_meta_write_pointer(struct btrfs_block_group *cache,
1410 struct extent_buffer *eb)
1412 if (!btrfs_is_zoned(eb->fs_info) || !cache)
1415 ASSERT(cache->meta_write_pointer == eb->start + eb->len);
1416 cache->meta_write_pointer = eb->start;
1419 int btrfs_zoned_issue_zeroout(struct btrfs_device *device, u64 physical, u64 length)
1421 if (!btrfs_dev_is_sequential(device, physical))
1424 return blkdev_issue_zeroout(device->bdev, physical >> SECTOR_SHIFT,
1425 length >> SECTOR_SHIFT, GFP_NOFS, 0);
1428 static int read_zone_info(struct btrfs_fs_info *fs_info, u64 logical,
1429 struct blk_zone *zone)
1431 struct btrfs_bio *bbio = NULL;
1432 u64 mapped_length = PAGE_SIZE;
1433 unsigned int nofs_flag;
1437 ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS, logical,
1438 &mapped_length, &bbio);
1439 if (ret || !bbio || mapped_length < PAGE_SIZE) {
1440 btrfs_put_bbio(bbio);
1444 if (bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK)
1447 nofs_flag = memalloc_nofs_save();
1448 nmirrors = (int)bbio->num_stripes;
1449 for (i = 0; i < nmirrors; i++) {
1450 u64 physical = bbio->stripes[i].physical;
1451 struct btrfs_device *dev = bbio->stripes[i].dev;
1453 /* Missing device */
1457 ret = btrfs_get_dev_zone(dev, physical, zone);
1458 /* Failing device */
1459 if (ret == -EIO || ret == -EOPNOTSUPP)
1463 memalloc_nofs_restore(nofs_flag);
1469 * Synchronize write pointer in a zone at @physical_start on @tgt_dev, by
1470 * filling zeros between @physical_pos to a write pointer of dev-replace
1473 int btrfs_sync_zone_write_pointer(struct btrfs_device *tgt_dev, u64 logical,
1474 u64 physical_start, u64 physical_pos)
1476 struct btrfs_fs_info *fs_info = tgt_dev->fs_info;
1477 struct blk_zone zone;
1482 if (!btrfs_dev_is_sequential(tgt_dev, physical_pos))
1485 ret = read_zone_info(fs_info, logical, &zone);
1489 wp = physical_start + ((zone.wp - zone.start) << SECTOR_SHIFT);
1491 if (physical_pos == wp)
1494 if (physical_pos > wp)
1497 length = wp - physical_pos;
1498 return btrfs_zoned_issue_zeroout(tgt_dev, physical_pos, length);