1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) STRATO AG 2012. All rights reserved.
6 #include <linux/sched.h>
8 #include <linux/slab.h>
9 #include <linux/blkdev.h>
10 #include <linux/kthread.h>
11 #include <linux/math64.h>
14 #include "extent_map.h"
16 #include "transaction.h"
17 #include "print-tree.h"
19 #include "async-thread.h"
20 #include "dev-replace.h"
23 #include "block-group.h"
25 #include "accessors.h"
29 * Device replace overview
32 * To copy all extents (both new and on-disk) from source device to target
33 * device, while still keeping the filesystem read-write.
36 * There are two main methods involved:
40 * All new writes will be written to both target and source devices, so even
41 * if replace gets canceled, sources device still contains up-to-date data.
43 * Location: handle_ops_on_dev_replace() from btrfs_map_block()
44 * Start: btrfs_dev_replace_start()
45 * End: btrfs_dev_replace_finishing()
46 * Content: Latest data/metadata
48 * - Copy existing extents
50 * This happens by re-using scrub facility, as scrub also iterates through
51 * existing extents from commit root.
53 * Location: scrub_write_block_to_dev_replace() from
54 * scrub_block_complete()
55 * Content: Data/meta from commit root.
57 * Due to the content difference, we need to avoid nocow write when dev-replace
58 * is happening. This is done by marking the block group read-only and waiting
61 * After replace is done, the finishing part is done by swapping the target and
64 * Location: btrfs_dev_replace_update_device_in_mapping_tree() from
65 * btrfs_dev_replace_finishing()
68 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
70 static int btrfs_dev_replace_kthread(void *data);
72 int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
74 struct btrfs_dev_lookup_args args = { .devid = BTRFS_DEV_REPLACE_DEVID };
76 struct btrfs_root *dev_root = fs_info->dev_root;
77 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
78 struct extent_buffer *eb;
81 struct btrfs_path *path = NULL;
83 struct btrfs_dev_replace_item *ptr;
89 path = btrfs_alloc_path();
96 key.type = BTRFS_DEV_REPLACE_KEY;
98 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
100 no_valid_dev_replace_entry_found:
102 * We don't have a replace item or it's corrupted. If there is
103 * a replace target, fail the mount.
105 if (btrfs_find_device(fs_info->fs_devices, &args)) {
107 "found replace target device without a valid replace item");
112 dev_replace->replace_state =
113 BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
114 dev_replace->cont_reading_from_srcdev_mode =
115 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
116 dev_replace->time_started = 0;
117 dev_replace->time_stopped = 0;
118 atomic64_set(&dev_replace->num_write_errors, 0);
119 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
120 dev_replace->cursor_left = 0;
121 dev_replace->committed_cursor_left = 0;
122 dev_replace->cursor_left_last_write_of_item = 0;
123 dev_replace->cursor_right = 0;
124 dev_replace->srcdev = NULL;
125 dev_replace->tgtdev = NULL;
126 dev_replace->is_valid = 0;
127 dev_replace->item_needs_writeback = 0;
130 slot = path->slots[0];
132 item_size = btrfs_item_size(eb, slot);
133 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
135 if (item_size != sizeof(struct btrfs_dev_replace_item)) {
137 "dev_replace entry found has unexpected size, ignore entry");
138 goto no_valid_dev_replace_entry_found;
141 src_devid = btrfs_dev_replace_src_devid(eb, ptr);
142 dev_replace->cont_reading_from_srcdev_mode =
143 btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
144 dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
145 dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
146 dev_replace->time_stopped =
147 btrfs_dev_replace_time_stopped(eb, ptr);
148 atomic64_set(&dev_replace->num_write_errors,
149 btrfs_dev_replace_num_write_errors(eb, ptr));
150 atomic64_set(&dev_replace->num_uncorrectable_read_errors,
151 btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
152 dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
153 dev_replace->committed_cursor_left = dev_replace->cursor_left;
154 dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
155 dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
156 dev_replace->is_valid = 1;
158 dev_replace->item_needs_writeback = 0;
159 switch (dev_replace->replace_state) {
160 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
161 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
162 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
164 * We don't have an active replace item but if there is a
165 * replace target, fail the mount.
167 if (btrfs_find_device(fs_info->fs_devices, &args)) {
169 "replace without active item, run 'device scan --forget' on the target device");
172 dev_replace->srcdev = NULL;
173 dev_replace->tgtdev = NULL;
176 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
177 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
178 dev_replace->tgtdev = btrfs_find_device(fs_info->fs_devices, &args);
179 args.devid = src_devid;
180 dev_replace->srcdev = btrfs_find_device(fs_info->fs_devices, &args);
183 * allow 'btrfs dev replace_cancel' if src/tgt device is
186 if (!dev_replace->srcdev &&
187 !btrfs_test_opt(fs_info, DEGRADED)) {
190 "cannot mount because device replace operation is ongoing and");
192 "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
195 if (!dev_replace->tgtdev &&
196 !btrfs_test_opt(fs_info, DEGRADED)) {
199 "cannot mount because device replace operation is ongoing and");
201 "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
202 BTRFS_DEV_REPLACE_DEVID);
204 if (dev_replace->tgtdev) {
205 if (dev_replace->srcdev) {
206 dev_replace->tgtdev->total_bytes =
207 dev_replace->srcdev->total_bytes;
208 dev_replace->tgtdev->disk_total_bytes =
209 dev_replace->srcdev->disk_total_bytes;
210 dev_replace->tgtdev->commit_total_bytes =
211 dev_replace->srcdev->commit_total_bytes;
212 dev_replace->tgtdev->bytes_used =
213 dev_replace->srcdev->bytes_used;
214 dev_replace->tgtdev->commit_bytes_used =
215 dev_replace->srcdev->commit_bytes_used;
217 set_bit(BTRFS_DEV_STATE_REPLACE_TGT,
218 &dev_replace->tgtdev->dev_state);
220 WARN_ON(fs_info->fs_devices->rw_devices == 0);
221 dev_replace->tgtdev->io_width = fs_info->sectorsize;
222 dev_replace->tgtdev->io_align = fs_info->sectorsize;
223 dev_replace->tgtdev->sector_size = fs_info->sectorsize;
224 dev_replace->tgtdev->fs_info = fs_info;
225 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
226 &dev_replace->tgtdev->dev_state);
232 btrfs_free_path(path);
237 * Initialize a new device for device replace target from a given source dev
240 * Return 0 and new device in @device_out, otherwise return < 0
242 static int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
243 const char *device_path,
244 struct btrfs_device *srcdev,
245 struct btrfs_device **device_out)
247 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
248 struct btrfs_device *device;
249 struct bdev_handle *bdev_handle;
250 struct block_device *bdev;
251 u64 devid = BTRFS_DEV_REPLACE_DEVID;
255 if (srcdev->fs_devices->seeding) {
256 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
260 bdev_handle = bdev_open_by_path(device_path, BLK_OPEN_WRITE,
261 fs_info->bdev_holder, NULL);
262 if (IS_ERR(bdev_handle)) {
263 btrfs_err(fs_info, "target device %s is invalid!", device_path);
264 return PTR_ERR(bdev_handle);
266 bdev = bdev_handle->bdev;
268 if (!btrfs_check_device_zone_type(fs_info, bdev)) {
270 "dev-replace: zoned type of target device mismatch with filesystem");
277 list_for_each_entry(device, &fs_devices->devices, dev_list) {
278 if (device->bdev == bdev) {
280 "target device is in the filesystem!");
287 if (bdev_nr_bytes(bdev) < btrfs_device_get_total_bytes(srcdev)) {
289 "target device is smaller than source device!");
295 device = btrfs_alloc_device(NULL, &devid, NULL, device_path);
296 if (IS_ERR(device)) {
297 ret = PTR_ERR(device);
301 ret = lookup_bdev(device_path, &device->devt);
305 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
306 device->generation = 0;
307 device->io_width = fs_info->sectorsize;
308 device->io_align = fs_info->sectorsize;
309 device->sector_size = fs_info->sectorsize;
310 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
311 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
312 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
313 device->commit_total_bytes = srcdev->commit_total_bytes;
314 device->commit_bytes_used = device->bytes_used;
315 device->fs_info = fs_info;
317 device->bdev_handle = bdev_handle;
318 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
319 set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
320 device->dev_stats_valid = 1;
321 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
322 device->fs_devices = fs_devices;
324 ret = btrfs_get_dev_zone_info(device, false);
328 mutex_lock(&fs_devices->device_list_mutex);
329 list_add(&device->dev_list, &fs_devices->devices);
330 fs_devices->num_devices++;
331 fs_devices->open_devices++;
332 mutex_unlock(&fs_devices->device_list_mutex);
334 *device_out = device;
338 bdev_release(bdev_handle);
343 * called from commit_transaction. Writes changed device replace state to
346 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans)
348 struct btrfs_fs_info *fs_info = trans->fs_info;
350 struct btrfs_root *dev_root = fs_info->dev_root;
351 struct btrfs_path *path;
352 struct btrfs_key key;
353 struct extent_buffer *eb;
354 struct btrfs_dev_replace_item *ptr;
355 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
357 down_read(&dev_replace->rwsem);
358 if (!dev_replace->is_valid ||
359 !dev_replace->item_needs_writeback) {
360 up_read(&dev_replace->rwsem);
363 up_read(&dev_replace->rwsem);
366 key.type = BTRFS_DEV_REPLACE_KEY;
369 path = btrfs_alloc_path();
374 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
377 "error %d while searching for dev_replace item!",
383 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
385 * need to delete old one and insert a new one.
386 * Since no attempt is made to recover any old state, if the
387 * dev_replace state is 'running', the data on the target
389 * It would be possible to recover the state: just make sure
390 * that the beginning of the item is never changed and always
391 * contains all the essential information. Then read this
392 * minimal set of information and use it as a base for the
395 ret = btrfs_del_item(trans, dev_root, path);
398 "delete too small dev_replace item failed %d!",
406 /* need to insert a new item */
407 btrfs_release_path(path);
408 ret = btrfs_insert_empty_item(trans, dev_root, path,
412 "insert dev_replace item failed %d!", ret);
418 ptr = btrfs_item_ptr(eb, path->slots[0],
419 struct btrfs_dev_replace_item);
421 down_write(&dev_replace->rwsem);
422 if (dev_replace->srcdev)
423 btrfs_set_dev_replace_src_devid(eb, ptr,
424 dev_replace->srcdev->devid);
426 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
427 btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
428 dev_replace->cont_reading_from_srcdev_mode);
429 btrfs_set_dev_replace_replace_state(eb, ptr,
430 dev_replace->replace_state);
431 btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
432 btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
433 btrfs_set_dev_replace_num_write_errors(eb, ptr,
434 atomic64_read(&dev_replace->num_write_errors));
435 btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
436 atomic64_read(&dev_replace->num_uncorrectable_read_errors));
437 dev_replace->cursor_left_last_write_of_item =
438 dev_replace->cursor_left;
439 btrfs_set_dev_replace_cursor_left(eb, ptr,
440 dev_replace->cursor_left_last_write_of_item);
441 btrfs_set_dev_replace_cursor_right(eb, ptr,
442 dev_replace->cursor_right);
443 dev_replace->item_needs_writeback = 0;
444 up_write(&dev_replace->rwsem);
446 btrfs_mark_buffer_dirty(trans, eb);
449 btrfs_free_path(path);
454 static int mark_block_group_to_copy(struct btrfs_fs_info *fs_info,
455 struct btrfs_device *src_dev)
457 struct btrfs_path *path;
458 struct btrfs_key key;
459 struct btrfs_key found_key;
460 struct btrfs_root *root = fs_info->dev_root;
461 struct btrfs_dev_extent *dev_extent = NULL;
462 struct btrfs_block_group *cache;
463 struct btrfs_trans_handle *trans;
468 /* Do not use "to_copy" on non zoned filesystem for now */
469 if (!btrfs_is_zoned(fs_info))
472 mutex_lock(&fs_info->chunk_mutex);
474 /* Ensure we don't have pending new block group */
475 spin_lock(&fs_info->trans_lock);
476 while (fs_info->running_transaction &&
477 !list_empty(&fs_info->running_transaction->dev_update_list)) {
478 spin_unlock(&fs_info->trans_lock);
479 mutex_unlock(&fs_info->chunk_mutex);
480 trans = btrfs_attach_transaction(root);
482 ret = PTR_ERR(trans);
483 mutex_lock(&fs_info->chunk_mutex);
484 if (ret == -ENOENT) {
485 spin_lock(&fs_info->trans_lock);
492 ret = btrfs_commit_transaction(trans);
493 mutex_lock(&fs_info->chunk_mutex);
497 spin_lock(&fs_info->trans_lock);
499 spin_unlock(&fs_info->trans_lock);
501 path = btrfs_alloc_path();
507 path->reada = READA_FORWARD;
508 path->search_commit_root = 1;
509 path->skip_locking = 1;
511 key.objectid = src_dev->devid;
512 key.type = BTRFS_DEV_EXTENT_KEY;
515 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) {
516 struct extent_buffer *leaf = path->nodes[0];
518 if (found_key.objectid != src_dev->devid)
521 if (found_key.type != BTRFS_DEV_EXTENT_KEY)
524 if (found_key.offset < key.offset)
527 dev_extent = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent);
529 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dev_extent);
531 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
535 set_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags);
536 btrfs_put_block_group(cache);
541 btrfs_free_path(path);
543 mutex_unlock(&fs_info->chunk_mutex);
548 bool btrfs_finish_block_group_to_copy(struct btrfs_device *srcdev,
549 struct btrfs_block_group *cache,
552 struct btrfs_fs_info *fs_info = cache->fs_info;
553 struct extent_map *em;
554 struct map_lookup *map;
555 u64 chunk_offset = cache->start;
556 int num_extents, cur_extent;
559 /* Do not use "to_copy" on non zoned filesystem for now */
560 if (!btrfs_is_zoned(fs_info))
563 spin_lock(&cache->lock);
564 if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &cache->runtime_flags)) {
565 spin_unlock(&cache->lock);
568 spin_unlock(&cache->lock);
570 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
572 map = em->map_lookup;
576 for (i = 0; i < map->num_stripes; i++) {
577 /* We have more device extent to copy */
578 if (srcdev != map->stripes[i].dev)
582 if (physical == map->stripes[i].physical)
588 if (num_extents > 1 && cur_extent < num_extents - 1) {
590 * Has more stripes on this device. Keep this block group
591 * readonly until we finish all the stripes.
596 /* Last stripe on this device */
597 clear_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags);
602 static int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
603 const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
606 struct btrfs_root *root = fs_info->dev_root;
607 struct btrfs_trans_handle *trans;
608 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
610 struct btrfs_device *tgt_device = NULL;
611 struct btrfs_device *src_device = NULL;
613 src_device = btrfs_find_device_by_devspec(fs_info, srcdevid,
615 if (IS_ERR(src_device))
616 return PTR_ERR(src_device);
618 if (btrfs_pinned_by_swapfile(fs_info, src_device)) {
619 btrfs_warn_in_rcu(fs_info,
620 "cannot replace device %s (devid %llu) due to active swapfile",
621 btrfs_dev_name(src_device), src_device->devid);
626 * Here we commit the transaction to make sure commit_total_bytes
627 * of all the devices are updated.
629 trans = btrfs_attach_transaction(root);
630 if (!IS_ERR(trans)) {
631 ret = btrfs_commit_transaction(trans);
634 } else if (PTR_ERR(trans) != -ENOENT) {
635 return PTR_ERR(trans);
638 ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
639 src_device, &tgt_device);
643 ret = mark_block_group_to_copy(fs_info, src_device);
647 down_write(&dev_replace->rwsem);
648 switch (dev_replace->replace_state) {
649 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
650 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
651 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
653 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
654 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
656 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
657 up_write(&dev_replace->rwsem);
661 dev_replace->cont_reading_from_srcdev_mode = read_src;
662 dev_replace->srcdev = src_device;
663 dev_replace->tgtdev = tgt_device;
665 btrfs_info_in_rcu(fs_info,
666 "dev_replace from %s (devid %llu) to %s started",
667 btrfs_dev_name(src_device),
669 btrfs_dev_name(tgt_device));
672 * from now on, the writes to the srcdev are all duplicated to
673 * go to the tgtdev as well (refer to btrfs_map_block()).
675 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
676 dev_replace->time_started = ktime_get_real_seconds();
677 dev_replace->cursor_left = 0;
678 dev_replace->committed_cursor_left = 0;
679 dev_replace->cursor_left_last_write_of_item = 0;
680 dev_replace->cursor_right = 0;
681 dev_replace->is_valid = 1;
682 dev_replace->item_needs_writeback = 1;
683 atomic64_set(&dev_replace->num_write_errors, 0);
684 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
685 up_write(&dev_replace->rwsem);
687 ret = btrfs_sysfs_add_device(tgt_device);
689 btrfs_err(fs_info, "kobj add dev failed %d", ret);
691 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
694 * Commit dev_replace state and reserve 1 item for it.
695 * This is crucial to ensure we won't miss copying extents for new block
696 * groups that are allocated after we started the device replace, and
697 * must be done after setting up the device replace state.
699 trans = btrfs_start_transaction(root, 1);
701 ret = PTR_ERR(trans);
702 down_write(&dev_replace->rwsem);
703 dev_replace->replace_state =
704 BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
705 dev_replace->srcdev = NULL;
706 dev_replace->tgtdev = NULL;
707 up_write(&dev_replace->rwsem);
711 ret = btrfs_commit_transaction(trans);
714 /* the disk copy procedure reuses the scrub code */
715 ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
716 btrfs_device_get_total_bytes(src_device),
717 &dev_replace->scrub_progress, 0, 1);
719 ret = btrfs_dev_replace_finishing(fs_info, ret);
720 if (ret == -EINPROGRESS)
721 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
726 btrfs_destroy_dev_replace_tgtdev(tgt_device);
730 int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
731 struct btrfs_ioctl_dev_replace_args *args)
735 switch (args->start.cont_reading_from_srcdev_mode) {
736 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
737 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
743 if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
744 args->start.tgtdev_name[0] == '\0')
747 ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name,
748 args->start.srcdevid,
749 args->start.srcdev_name,
750 args->start.cont_reading_from_srcdev_mode);
752 /* don't warn if EINPROGRESS, someone else might be running scrub */
753 if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS ||
754 ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR)
761 * blocked until all in-flight bios operations are finished.
763 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
765 set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
766 wait_event(fs_info->dev_replace.replace_wait, !percpu_counter_sum(
767 &fs_info->dev_replace.bio_counter));
771 * we have removed target device, it is safe to allow new bios request.
773 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
775 clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
776 wake_up(&fs_info->dev_replace.replace_wait);
780 * When finishing the device replace, before swapping the source device with the
781 * target device we must update the chunk allocation state in the target device,
782 * as it is empty because replace works by directly copying the chunks and not
783 * through the normal chunk allocation path.
785 static int btrfs_set_target_alloc_state(struct btrfs_device *srcdev,
786 struct btrfs_device *tgtdev)
788 struct extent_state *cached_state = NULL;
794 lockdep_assert_held(&srcdev->fs_info->chunk_mutex);
796 while (find_first_extent_bit(&srcdev->alloc_state, start,
797 &found_start, &found_end,
798 CHUNK_ALLOCATED, &cached_state)) {
799 ret = set_extent_bit(&tgtdev->alloc_state, found_start,
800 found_end, CHUNK_ALLOCATED, NULL);
803 start = found_end + 1;
806 free_extent_state(cached_state);
810 static void btrfs_dev_replace_update_device_in_mapping_tree(
811 struct btrfs_fs_info *fs_info,
812 struct btrfs_device *srcdev,
813 struct btrfs_device *tgtdev)
815 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
816 struct extent_map *em;
817 struct map_lookup *map;
821 write_lock(&em_tree->lock);
823 em = lookup_extent_mapping(em_tree, start, (u64)-1);
826 map = em->map_lookup;
827 for (i = 0; i < map->num_stripes; i++)
828 if (srcdev == map->stripes[i].dev)
829 map->stripes[i].dev = tgtdev;
830 start = em->start + em->len;
833 write_unlock(&em_tree->lock);
836 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
839 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
840 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
841 struct btrfs_device *tgt_device;
842 struct btrfs_device *src_device;
843 struct btrfs_root *root = fs_info->tree_root;
844 u8 uuid_tmp[BTRFS_UUID_SIZE];
845 struct btrfs_trans_handle *trans;
848 /* don't allow cancel or unmount to disturb the finishing procedure */
849 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
851 down_read(&dev_replace->rwsem);
852 /* was the operation canceled, or is it finished? */
853 if (dev_replace->replace_state !=
854 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
855 up_read(&dev_replace->rwsem);
856 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
860 tgt_device = dev_replace->tgtdev;
861 src_device = dev_replace->srcdev;
862 up_read(&dev_replace->rwsem);
865 * flush all outstanding I/O and inode extent mappings before the
866 * copy operation is declared as being finished
868 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
870 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
873 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
876 * We have to use this loop approach because at this point src_device
877 * has to be available for transaction commit to complete, yet new
878 * chunks shouldn't be allocated on the device.
881 trans = btrfs_start_transaction(root, 0);
883 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
884 return PTR_ERR(trans);
886 ret = btrfs_commit_transaction(trans);
889 /* Prevent write_all_supers() during the finishing procedure */
890 mutex_lock(&fs_devices->device_list_mutex);
891 /* Prevent new chunks being allocated on the source device */
892 mutex_lock(&fs_info->chunk_mutex);
894 if (!list_empty(&src_device->post_commit_list)) {
895 mutex_unlock(&fs_devices->device_list_mutex);
896 mutex_unlock(&fs_info->chunk_mutex);
902 down_write(&dev_replace->rwsem);
903 dev_replace->replace_state =
904 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
905 : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
906 dev_replace->tgtdev = NULL;
907 dev_replace->srcdev = NULL;
908 dev_replace->time_stopped = ktime_get_real_seconds();
909 dev_replace->item_needs_writeback = 1;
912 * Update allocation state in the new device and replace the old device
913 * with the new one in the mapping tree.
916 scrub_ret = btrfs_set_target_alloc_state(src_device, tgt_device);
919 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
923 if (scrub_ret != -ECANCELED)
924 btrfs_err_in_rcu(fs_info,
925 "btrfs_scrub_dev(%s, %llu, %s) failed %d",
926 btrfs_dev_name(src_device),
928 btrfs_dev_name(tgt_device), scrub_ret);
930 up_write(&dev_replace->rwsem);
931 mutex_unlock(&fs_info->chunk_mutex);
932 mutex_unlock(&fs_devices->device_list_mutex);
933 btrfs_rm_dev_replace_blocked(fs_info);
935 btrfs_destroy_dev_replace_tgtdev(tgt_device);
936 btrfs_rm_dev_replace_unblocked(fs_info);
937 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
942 btrfs_info_in_rcu(fs_info,
943 "dev_replace from %s (devid %llu) to %s finished",
944 btrfs_dev_name(src_device),
946 btrfs_dev_name(tgt_device));
947 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state);
948 tgt_device->devid = src_device->devid;
949 src_device->devid = BTRFS_DEV_REPLACE_DEVID;
950 memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
951 memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
952 memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
953 btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
954 btrfs_device_set_disk_total_bytes(tgt_device,
955 src_device->disk_total_bytes);
956 btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
957 tgt_device->commit_bytes_used = src_device->bytes_used;
959 btrfs_assign_next_active_device(src_device, tgt_device);
961 list_add(&tgt_device->dev_alloc_list, &fs_devices->alloc_list);
962 fs_devices->rw_devices++;
964 up_write(&dev_replace->rwsem);
965 btrfs_rm_dev_replace_blocked(fs_info);
967 btrfs_rm_dev_replace_remove_srcdev(src_device);
969 btrfs_rm_dev_replace_unblocked(fs_info);
972 * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will
973 * update on-disk dev stats value during commit transaction
975 atomic_inc(&tgt_device->dev_stats_ccnt);
978 * this is again a consistent state where no dev_replace procedure
979 * is running, the target device is part of the filesystem, the
980 * source device is not part of the filesystem anymore and its 1st
981 * superblock is scratched out so that it is no longer marked to
982 * belong to this filesystem.
984 mutex_unlock(&fs_info->chunk_mutex);
985 mutex_unlock(&fs_devices->device_list_mutex);
987 /* replace the sysfs entry */
988 btrfs_sysfs_remove_device(src_device);
989 btrfs_sysfs_update_devid(tgt_device);
990 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &src_device->dev_state))
991 btrfs_scratch_superblocks(fs_info, src_device->bdev,
992 src_device->name->str);
994 /* write back the superblocks */
995 trans = btrfs_start_transaction(root, 0);
997 btrfs_commit_transaction(trans);
999 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
1001 btrfs_rm_dev_replace_free_srcdev(src_device);
1007 * Read progress of device replace status according to the state and last
1008 * stored position. The value format is the same as for
1009 * btrfs_dev_replace::progress_1000
1011 static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info)
1013 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1016 switch (dev_replace->replace_state) {
1017 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1018 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1021 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1024 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1025 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1026 ret = div64_u64(dev_replace->cursor_left,
1027 div_u64(btrfs_device_get_total_bytes(
1028 dev_replace->srcdev), 1000));
1035 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
1036 struct btrfs_ioctl_dev_replace_args *args)
1038 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1040 down_read(&dev_replace->rwsem);
1041 /* even if !dev_replace_is_valid, the values are good enough for
1042 * the replace_status ioctl */
1043 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
1044 args->status.replace_state = dev_replace->replace_state;
1045 args->status.time_started = dev_replace->time_started;
1046 args->status.time_stopped = dev_replace->time_stopped;
1047 args->status.num_write_errors =
1048 atomic64_read(&dev_replace->num_write_errors);
1049 args->status.num_uncorrectable_read_errors =
1050 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
1051 args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
1052 up_read(&dev_replace->rwsem);
1055 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
1057 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1058 struct btrfs_device *tgt_device = NULL;
1059 struct btrfs_device *src_device = NULL;
1060 struct btrfs_trans_handle *trans;
1061 struct btrfs_root *root = fs_info->tree_root;
1065 if (sb_rdonly(fs_info->sb))
1068 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
1069 down_write(&dev_replace->rwsem);
1070 switch (dev_replace->replace_state) {
1071 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1072 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1073 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1074 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
1075 up_write(&dev_replace->rwsem);
1077 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1078 tgt_device = dev_replace->tgtdev;
1079 src_device = dev_replace->srcdev;
1080 up_write(&dev_replace->rwsem);
1081 ret = btrfs_scrub_cancel(fs_info);
1083 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
1085 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
1087 * btrfs_dev_replace_finishing() will handle the
1090 btrfs_info_in_rcu(fs_info,
1091 "dev_replace from %s (devid %llu) to %s canceled",
1092 btrfs_dev_name(src_device), src_device->devid,
1093 btrfs_dev_name(tgt_device));
1096 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1098 * Scrub doing the replace isn't running so we need to do the
1099 * cleanup step of btrfs_dev_replace_finishing() here
1101 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
1102 tgt_device = dev_replace->tgtdev;
1103 src_device = dev_replace->srcdev;
1104 dev_replace->tgtdev = NULL;
1105 dev_replace->srcdev = NULL;
1106 dev_replace->replace_state =
1107 BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
1108 dev_replace->time_stopped = ktime_get_real_seconds();
1109 dev_replace->item_needs_writeback = 1;
1111 up_write(&dev_replace->rwsem);
1113 /* Scrub for replace must not be running in suspended state */
1114 btrfs_scrub_cancel(fs_info);
1116 trans = btrfs_start_transaction(root, 0);
1117 if (IS_ERR(trans)) {
1118 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
1119 return PTR_ERR(trans);
1121 ret = btrfs_commit_transaction(trans);
1124 btrfs_info_in_rcu(fs_info,
1125 "suspended dev_replace from %s (devid %llu) to %s canceled",
1126 btrfs_dev_name(src_device), src_device->devid,
1127 btrfs_dev_name(tgt_device));
1130 btrfs_destroy_dev_replace_tgtdev(tgt_device);
1133 up_write(&dev_replace->rwsem);
1137 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
1141 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
1143 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1145 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
1146 down_write(&dev_replace->rwsem);
1148 switch (dev_replace->replace_state) {
1149 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1150 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1151 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1152 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1154 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1155 dev_replace->replace_state =
1156 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
1157 dev_replace->time_stopped = ktime_get_real_seconds();
1158 dev_replace->item_needs_writeback = 1;
1159 btrfs_info(fs_info, "suspending dev_replace for unmount");
1163 up_write(&dev_replace->rwsem);
1164 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
1167 /* resume dev_replace procedure that was interrupted by unmount */
1168 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
1170 struct task_struct *task;
1171 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1173 down_write(&dev_replace->rwsem);
1175 switch (dev_replace->replace_state) {
1176 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1177 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1178 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1179 up_write(&dev_replace->rwsem);
1181 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1183 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1184 dev_replace->replace_state =
1185 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
1188 if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
1190 "cannot continue dev_replace, tgtdev is missing");
1192 "you may cancel the operation after 'mount -o degraded'");
1193 dev_replace->replace_state =
1194 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
1195 up_write(&dev_replace->rwsem);
1198 up_write(&dev_replace->rwsem);
1201 * This could collide with a paused balance, but the exclusive op logic
1202 * should never allow both to start and pause. We don't want to allow
1203 * dev-replace to start anyway.
1205 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
1206 down_write(&dev_replace->rwsem);
1207 dev_replace->replace_state =
1208 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
1209 up_write(&dev_replace->rwsem);
1211 "cannot resume dev-replace, other exclusive operation running");
1215 task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
1216 return PTR_ERR_OR_ZERO(task);
1219 static int btrfs_dev_replace_kthread(void *data)
1221 struct btrfs_fs_info *fs_info = data;
1222 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1226 progress = btrfs_dev_replace_progress(fs_info);
1227 progress = div_u64(progress, 10);
1228 btrfs_info_in_rcu(fs_info,
1229 "continuing dev_replace from %s (devid %llu) to target %s @%u%%",
1230 btrfs_dev_name(dev_replace->srcdev),
1231 dev_replace->srcdev->devid,
1232 btrfs_dev_name(dev_replace->tgtdev),
1233 (unsigned int)progress);
1235 ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
1236 dev_replace->committed_cursor_left,
1237 btrfs_device_get_total_bytes(dev_replace->srcdev),
1238 &dev_replace->scrub_progress, 0, 1);
1239 ret = btrfs_dev_replace_finishing(fs_info, ret);
1240 WARN_ON(ret && ret != -ECANCELED);
1242 btrfs_exclop_finish(fs_info);
1246 int __pure btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
1248 if (!dev_replace->is_valid)
1251 switch (dev_replace->replace_state) {
1252 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1253 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1254 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1256 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1257 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1259 * return true even if tgtdev is missing (this is
1260 * something that can happen if the dev_replace
1261 * procedure is suspended by an umount and then
1262 * the tgtdev is missing (or "btrfs dev scan") was
1263 * not called and the filesystem is remounted
1264 * in degraded state. This does not stop the
1265 * dev_replace procedure. It needs to be canceled
1266 * manually if the cancellation is wanted.
1273 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
1275 percpu_counter_sub(&fs_info->dev_replace.bio_counter, amount);
1276 cond_wake_up_nomb(&fs_info->dev_replace.replace_wait);
1279 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
1282 percpu_counter_inc(&fs_info->dev_replace.bio_counter);
1283 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1284 &fs_info->fs_state)))
1287 btrfs_bio_counter_dec(fs_info);
1288 wait_event(fs_info->dev_replace.replace_wait,
1289 !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1290 &fs_info->fs_state));