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 block_device *bdev;
250 u64 devid = BTRFS_DEV_REPLACE_DEVID;
254 if (srcdev->fs_devices->seeding) {
255 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
259 bdev = blkdev_get_by_path(device_path, BLK_OPEN_WRITE,
260 fs_info->bdev_holder, NULL);
262 btrfs_err(fs_info, "target device %s is invalid!", device_path);
263 return PTR_ERR(bdev);
266 if (!btrfs_check_device_zone_type(fs_info, bdev)) {
268 "dev-replace: zoned type of target device mismatch with filesystem");
275 list_for_each_entry(device, &fs_devices->devices, dev_list) {
276 if (device->bdev == bdev) {
278 "target device is in the filesystem!");
285 if (bdev_nr_bytes(bdev) < btrfs_device_get_total_bytes(srcdev)) {
287 "target device is smaller than source device!");
293 device = btrfs_alloc_device(NULL, &devid, NULL, device_path);
294 if (IS_ERR(device)) {
295 ret = PTR_ERR(device);
299 ret = lookup_bdev(device_path, &device->devt);
303 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
304 device->generation = 0;
305 device->io_width = fs_info->sectorsize;
306 device->io_align = fs_info->sectorsize;
307 device->sector_size = fs_info->sectorsize;
308 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
309 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
310 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
311 device->commit_total_bytes = srcdev->commit_total_bytes;
312 device->commit_bytes_used = device->bytes_used;
313 device->fs_info = fs_info;
315 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
316 set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
317 device->holder = fs_info->bdev_holder;
318 device->dev_stats_valid = 1;
319 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
320 device->fs_devices = fs_devices;
322 ret = btrfs_get_dev_zone_info(device, false);
326 mutex_lock(&fs_devices->device_list_mutex);
327 list_add(&device->dev_list, &fs_devices->devices);
328 fs_devices->num_devices++;
329 fs_devices->open_devices++;
330 mutex_unlock(&fs_devices->device_list_mutex);
332 *device_out = device;
336 blkdev_put(bdev, fs_info->bdev_holder);
341 * called from commit_transaction. Writes changed device replace state to
344 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans)
346 struct btrfs_fs_info *fs_info = trans->fs_info;
348 struct btrfs_root *dev_root = fs_info->dev_root;
349 struct btrfs_path *path;
350 struct btrfs_key key;
351 struct extent_buffer *eb;
352 struct btrfs_dev_replace_item *ptr;
353 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
355 down_read(&dev_replace->rwsem);
356 if (!dev_replace->is_valid ||
357 !dev_replace->item_needs_writeback) {
358 up_read(&dev_replace->rwsem);
361 up_read(&dev_replace->rwsem);
364 key.type = BTRFS_DEV_REPLACE_KEY;
367 path = btrfs_alloc_path();
372 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
375 "error %d while searching for dev_replace item!",
381 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
383 * need to delete old one and insert a new one.
384 * Since no attempt is made to recover any old state, if the
385 * dev_replace state is 'running', the data on the target
387 * It would be possible to recover the state: just make sure
388 * that the beginning of the item is never changed and always
389 * contains all the essential information. Then read this
390 * minimal set of information and use it as a base for the
393 ret = btrfs_del_item(trans, dev_root, path);
396 "delete too small dev_replace item failed %d!",
404 /* need to insert a new item */
405 btrfs_release_path(path);
406 ret = btrfs_insert_empty_item(trans, dev_root, path,
410 "insert dev_replace item failed %d!", ret);
416 ptr = btrfs_item_ptr(eb, path->slots[0],
417 struct btrfs_dev_replace_item);
419 down_write(&dev_replace->rwsem);
420 if (dev_replace->srcdev)
421 btrfs_set_dev_replace_src_devid(eb, ptr,
422 dev_replace->srcdev->devid);
424 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
425 btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
426 dev_replace->cont_reading_from_srcdev_mode);
427 btrfs_set_dev_replace_replace_state(eb, ptr,
428 dev_replace->replace_state);
429 btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
430 btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
431 btrfs_set_dev_replace_num_write_errors(eb, ptr,
432 atomic64_read(&dev_replace->num_write_errors));
433 btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
434 atomic64_read(&dev_replace->num_uncorrectable_read_errors));
435 dev_replace->cursor_left_last_write_of_item =
436 dev_replace->cursor_left;
437 btrfs_set_dev_replace_cursor_left(eb, ptr,
438 dev_replace->cursor_left_last_write_of_item);
439 btrfs_set_dev_replace_cursor_right(eb, ptr,
440 dev_replace->cursor_right);
441 dev_replace->item_needs_writeback = 0;
442 up_write(&dev_replace->rwsem);
444 btrfs_mark_buffer_dirty(trans, eb);
447 btrfs_free_path(path);
452 static int mark_block_group_to_copy(struct btrfs_fs_info *fs_info,
453 struct btrfs_device *src_dev)
455 struct btrfs_path *path;
456 struct btrfs_key key;
457 struct btrfs_key found_key;
458 struct btrfs_root *root = fs_info->dev_root;
459 struct btrfs_dev_extent *dev_extent = NULL;
460 struct btrfs_block_group *cache;
461 struct btrfs_trans_handle *trans;
466 /* Do not use "to_copy" on non zoned filesystem for now */
467 if (!btrfs_is_zoned(fs_info))
470 mutex_lock(&fs_info->chunk_mutex);
472 /* Ensure we don't have pending new block group */
473 spin_lock(&fs_info->trans_lock);
474 while (fs_info->running_transaction &&
475 !list_empty(&fs_info->running_transaction->dev_update_list)) {
476 spin_unlock(&fs_info->trans_lock);
477 mutex_unlock(&fs_info->chunk_mutex);
478 trans = btrfs_attach_transaction(root);
480 ret = PTR_ERR(trans);
481 mutex_lock(&fs_info->chunk_mutex);
482 if (ret == -ENOENT) {
483 spin_lock(&fs_info->trans_lock);
490 ret = btrfs_commit_transaction(trans);
491 mutex_lock(&fs_info->chunk_mutex);
495 spin_lock(&fs_info->trans_lock);
497 spin_unlock(&fs_info->trans_lock);
499 path = btrfs_alloc_path();
505 path->reada = READA_FORWARD;
506 path->search_commit_root = 1;
507 path->skip_locking = 1;
509 key.objectid = src_dev->devid;
510 key.type = BTRFS_DEV_EXTENT_KEY;
513 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) {
514 struct extent_buffer *leaf = path->nodes[0];
516 if (found_key.objectid != src_dev->devid)
519 if (found_key.type != BTRFS_DEV_EXTENT_KEY)
522 if (found_key.offset < key.offset)
525 dev_extent = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent);
527 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dev_extent);
529 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
533 set_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags);
534 btrfs_put_block_group(cache);
539 btrfs_free_path(path);
541 mutex_unlock(&fs_info->chunk_mutex);
546 bool btrfs_finish_block_group_to_copy(struct btrfs_device *srcdev,
547 struct btrfs_block_group *cache,
550 struct btrfs_fs_info *fs_info = cache->fs_info;
551 struct extent_map *em;
552 struct map_lookup *map;
553 u64 chunk_offset = cache->start;
554 int num_extents, cur_extent;
557 /* Do not use "to_copy" on non zoned filesystem for now */
558 if (!btrfs_is_zoned(fs_info))
561 spin_lock(&cache->lock);
562 if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &cache->runtime_flags)) {
563 spin_unlock(&cache->lock);
566 spin_unlock(&cache->lock);
568 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
570 map = em->map_lookup;
574 for (i = 0; i < map->num_stripes; i++) {
575 /* We have more device extent to copy */
576 if (srcdev != map->stripes[i].dev)
580 if (physical == map->stripes[i].physical)
586 if (num_extents > 1 && cur_extent < num_extents - 1) {
588 * Has more stripes on this device. Keep this block group
589 * readonly until we finish all the stripes.
594 /* Last stripe on this device */
595 clear_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags);
600 static int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
601 const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
604 struct btrfs_root *root = fs_info->dev_root;
605 struct btrfs_trans_handle *trans;
606 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
608 struct btrfs_device *tgt_device = NULL;
609 struct btrfs_device *src_device = NULL;
611 src_device = btrfs_find_device_by_devspec(fs_info, srcdevid,
613 if (IS_ERR(src_device))
614 return PTR_ERR(src_device);
616 if (btrfs_pinned_by_swapfile(fs_info, src_device)) {
617 btrfs_warn_in_rcu(fs_info,
618 "cannot replace device %s (devid %llu) due to active swapfile",
619 btrfs_dev_name(src_device), src_device->devid);
624 * Here we commit the transaction to make sure commit_total_bytes
625 * of all the devices are updated.
627 trans = btrfs_attach_transaction(root);
628 if (!IS_ERR(trans)) {
629 ret = btrfs_commit_transaction(trans);
632 } else if (PTR_ERR(trans) != -ENOENT) {
633 return PTR_ERR(trans);
636 ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
637 src_device, &tgt_device);
641 ret = mark_block_group_to_copy(fs_info, src_device);
645 down_write(&dev_replace->rwsem);
646 switch (dev_replace->replace_state) {
647 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
648 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
649 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
651 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
652 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
654 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
655 up_write(&dev_replace->rwsem);
659 dev_replace->cont_reading_from_srcdev_mode = read_src;
660 dev_replace->srcdev = src_device;
661 dev_replace->tgtdev = tgt_device;
663 btrfs_info_in_rcu(fs_info,
664 "dev_replace from %s (devid %llu) to %s started",
665 btrfs_dev_name(src_device),
667 btrfs_dev_name(tgt_device));
670 * from now on, the writes to the srcdev are all duplicated to
671 * go to the tgtdev as well (refer to btrfs_map_block()).
673 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
674 dev_replace->time_started = ktime_get_real_seconds();
675 dev_replace->cursor_left = 0;
676 dev_replace->committed_cursor_left = 0;
677 dev_replace->cursor_left_last_write_of_item = 0;
678 dev_replace->cursor_right = 0;
679 dev_replace->is_valid = 1;
680 dev_replace->item_needs_writeback = 1;
681 atomic64_set(&dev_replace->num_write_errors, 0);
682 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
683 up_write(&dev_replace->rwsem);
685 ret = btrfs_sysfs_add_device(tgt_device);
687 btrfs_err(fs_info, "kobj add dev failed %d", ret);
689 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
692 * Commit dev_replace state and reserve 1 item for it.
693 * This is crucial to ensure we won't miss copying extents for new block
694 * groups that are allocated after we started the device replace, and
695 * must be done after setting up the device replace state.
697 trans = btrfs_start_transaction(root, 1);
699 ret = PTR_ERR(trans);
700 down_write(&dev_replace->rwsem);
701 dev_replace->replace_state =
702 BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
703 dev_replace->srcdev = NULL;
704 dev_replace->tgtdev = NULL;
705 up_write(&dev_replace->rwsem);
709 ret = btrfs_commit_transaction(trans);
712 /* the disk copy procedure reuses the scrub code */
713 ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
714 btrfs_device_get_total_bytes(src_device),
715 &dev_replace->scrub_progress, 0, 1);
717 ret = btrfs_dev_replace_finishing(fs_info, ret);
718 if (ret == -EINPROGRESS)
719 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
724 btrfs_destroy_dev_replace_tgtdev(tgt_device);
728 int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
729 struct btrfs_ioctl_dev_replace_args *args)
733 switch (args->start.cont_reading_from_srcdev_mode) {
734 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
735 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
741 if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
742 args->start.tgtdev_name[0] == '\0')
745 ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name,
746 args->start.srcdevid,
747 args->start.srcdev_name,
748 args->start.cont_reading_from_srcdev_mode);
750 /* don't warn if EINPROGRESS, someone else might be running scrub */
751 if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS ||
752 ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR)
759 * blocked until all in-flight bios operations are finished.
761 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
763 set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
764 wait_event(fs_info->dev_replace.replace_wait, !percpu_counter_sum(
765 &fs_info->dev_replace.bio_counter));
769 * we have removed target device, it is safe to allow new bios request.
771 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
773 clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
774 wake_up(&fs_info->dev_replace.replace_wait);
778 * When finishing the device replace, before swapping the source device with the
779 * target device we must update the chunk allocation state in the target device,
780 * as it is empty because replace works by directly copying the chunks and not
781 * through the normal chunk allocation path.
783 static int btrfs_set_target_alloc_state(struct btrfs_device *srcdev,
784 struct btrfs_device *tgtdev)
786 struct extent_state *cached_state = NULL;
792 lockdep_assert_held(&srcdev->fs_info->chunk_mutex);
794 while (find_first_extent_bit(&srcdev->alloc_state, start,
795 &found_start, &found_end,
796 CHUNK_ALLOCATED, &cached_state)) {
797 ret = set_extent_bit(&tgtdev->alloc_state, found_start,
798 found_end, CHUNK_ALLOCATED, NULL);
801 start = found_end + 1;
804 free_extent_state(cached_state);
808 static void btrfs_dev_replace_update_device_in_mapping_tree(
809 struct btrfs_fs_info *fs_info,
810 struct btrfs_device *srcdev,
811 struct btrfs_device *tgtdev)
813 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
814 struct extent_map *em;
815 struct map_lookup *map;
819 write_lock(&em_tree->lock);
821 em = lookup_extent_mapping(em_tree, start, (u64)-1);
824 map = em->map_lookup;
825 for (i = 0; i < map->num_stripes; i++)
826 if (srcdev == map->stripes[i].dev)
827 map->stripes[i].dev = tgtdev;
828 start = em->start + em->len;
831 write_unlock(&em_tree->lock);
834 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
837 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
838 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
839 struct btrfs_device *tgt_device;
840 struct btrfs_device *src_device;
841 struct btrfs_root *root = fs_info->tree_root;
842 u8 uuid_tmp[BTRFS_UUID_SIZE];
843 struct btrfs_trans_handle *trans;
846 /* don't allow cancel or unmount to disturb the finishing procedure */
847 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
849 down_read(&dev_replace->rwsem);
850 /* was the operation canceled, or is it finished? */
851 if (dev_replace->replace_state !=
852 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
853 up_read(&dev_replace->rwsem);
854 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
858 tgt_device = dev_replace->tgtdev;
859 src_device = dev_replace->srcdev;
860 up_read(&dev_replace->rwsem);
863 * flush all outstanding I/O and inode extent mappings before the
864 * copy operation is declared as being finished
866 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
868 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
871 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
874 * We have to use this loop approach because at this point src_device
875 * has to be available for transaction commit to complete, yet new
876 * chunks shouldn't be allocated on the device.
879 trans = btrfs_start_transaction(root, 0);
881 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
882 return PTR_ERR(trans);
884 ret = btrfs_commit_transaction(trans);
887 /* Prevent write_all_supers() during the finishing procedure */
888 mutex_lock(&fs_devices->device_list_mutex);
889 /* Prevent new chunks being allocated on the source device */
890 mutex_lock(&fs_info->chunk_mutex);
892 if (!list_empty(&src_device->post_commit_list)) {
893 mutex_unlock(&fs_devices->device_list_mutex);
894 mutex_unlock(&fs_info->chunk_mutex);
900 down_write(&dev_replace->rwsem);
901 dev_replace->replace_state =
902 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
903 : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
904 dev_replace->tgtdev = NULL;
905 dev_replace->srcdev = NULL;
906 dev_replace->time_stopped = ktime_get_real_seconds();
907 dev_replace->item_needs_writeback = 1;
910 * Update allocation state in the new device and replace the old device
911 * with the new one in the mapping tree.
914 scrub_ret = btrfs_set_target_alloc_state(src_device, tgt_device);
917 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
921 if (scrub_ret != -ECANCELED)
922 btrfs_err_in_rcu(fs_info,
923 "btrfs_scrub_dev(%s, %llu, %s) failed %d",
924 btrfs_dev_name(src_device),
926 btrfs_dev_name(tgt_device), scrub_ret);
928 up_write(&dev_replace->rwsem);
929 mutex_unlock(&fs_info->chunk_mutex);
930 mutex_unlock(&fs_devices->device_list_mutex);
931 btrfs_rm_dev_replace_blocked(fs_info);
933 btrfs_destroy_dev_replace_tgtdev(tgt_device);
934 btrfs_rm_dev_replace_unblocked(fs_info);
935 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
940 btrfs_info_in_rcu(fs_info,
941 "dev_replace from %s (devid %llu) to %s finished",
942 btrfs_dev_name(src_device),
944 btrfs_dev_name(tgt_device));
945 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state);
946 tgt_device->devid = src_device->devid;
947 src_device->devid = BTRFS_DEV_REPLACE_DEVID;
948 memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
949 memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
950 memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
951 btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
952 btrfs_device_set_disk_total_bytes(tgt_device,
953 src_device->disk_total_bytes);
954 btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
955 tgt_device->commit_bytes_used = src_device->bytes_used;
957 btrfs_assign_next_active_device(src_device, tgt_device);
959 list_add(&tgt_device->dev_alloc_list, &fs_devices->alloc_list);
960 fs_devices->rw_devices++;
962 up_write(&dev_replace->rwsem);
963 btrfs_rm_dev_replace_blocked(fs_info);
965 btrfs_rm_dev_replace_remove_srcdev(src_device);
967 btrfs_rm_dev_replace_unblocked(fs_info);
970 * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will
971 * update on-disk dev stats value during commit transaction
973 atomic_inc(&tgt_device->dev_stats_ccnt);
976 * this is again a consistent state where no dev_replace procedure
977 * is running, the target device is part of the filesystem, the
978 * source device is not part of the filesystem anymore and its 1st
979 * superblock is scratched out so that it is no longer marked to
980 * belong to this filesystem.
982 mutex_unlock(&fs_info->chunk_mutex);
983 mutex_unlock(&fs_devices->device_list_mutex);
985 /* replace the sysfs entry */
986 btrfs_sysfs_remove_device(src_device);
987 btrfs_sysfs_update_devid(tgt_device);
988 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &src_device->dev_state))
989 btrfs_scratch_superblocks(fs_info, src_device->bdev,
990 src_device->name->str);
992 /* write back the superblocks */
993 trans = btrfs_start_transaction(root, 0);
995 btrfs_commit_transaction(trans);
997 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
999 btrfs_rm_dev_replace_free_srcdev(src_device);
1005 * Read progress of device replace status according to the state and last
1006 * stored position. The value format is the same as for
1007 * btrfs_dev_replace::progress_1000
1009 static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info)
1011 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1014 switch (dev_replace->replace_state) {
1015 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1016 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1019 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1022 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1023 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1024 ret = div64_u64(dev_replace->cursor_left,
1025 div_u64(btrfs_device_get_total_bytes(
1026 dev_replace->srcdev), 1000));
1033 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
1034 struct btrfs_ioctl_dev_replace_args *args)
1036 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1038 down_read(&dev_replace->rwsem);
1039 /* even if !dev_replace_is_valid, the values are good enough for
1040 * the replace_status ioctl */
1041 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
1042 args->status.replace_state = dev_replace->replace_state;
1043 args->status.time_started = dev_replace->time_started;
1044 args->status.time_stopped = dev_replace->time_stopped;
1045 args->status.num_write_errors =
1046 atomic64_read(&dev_replace->num_write_errors);
1047 args->status.num_uncorrectable_read_errors =
1048 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
1049 args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
1050 up_read(&dev_replace->rwsem);
1053 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
1055 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1056 struct btrfs_device *tgt_device = NULL;
1057 struct btrfs_device *src_device = NULL;
1058 struct btrfs_trans_handle *trans;
1059 struct btrfs_root *root = fs_info->tree_root;
1063 if (sb_rdonly(fs_info->sb))
1066 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
1067 down_write(&dev_replace->rwsem);
1068 switch (dev_replace->replace_state) {
1069 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1070 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1071 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1072 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
1073 up_write(&dev_replace->rwsem);
1075 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1076 tgt_device = dev_replace->tgtdev;
1077 src_device = dev_replace->srcdev;
1078 up_write(&dev_replace->rwsem);
1079 ret = btrfs_scrub_cancel(fs_info);
1081 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
1083 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
1085 * btrfs_dev_replace_finishing() will handle the
1088 btrfs_info_in_rcu(fs_info,
1089 "dev_replace from %s (devid %llu) to %s canceled",
1090 btrfs_dev_name(src_device), src_device->devid,
1091 btrfs_dev_name(tgt_device));
1094 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1096 * Scrub doing the replace isn't running so we need to do the
1097 * cleanup step of btrfs_dev_replace_finishing() here
1099 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
1100 tgt_device = dev_replace->tgtdev;
1101 src_device = dev_replace->srcdev;
1102 dev_replace->tgtdev = NULL;
1103 dev_replace->srcdev = NULL;
1104 dev_replace->replace_state =
1105 BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
1106 dev_replace->time_stopped = ktime_get_real_seconds();
1107 dev_replace->item_needs_writeback = 1;
1109 up_write(&dev_replace->rwsem);
1111 /* Scrub for replace must not be running in suspended state */
1112 btrfs_scrub_cancel(fs_info);
1114 trans = btrfs_start_transaction(root, 0);
1115 if (IS_ERR(trans)) {
1116 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
1117 return PTR_ERR(trans);
1119 ret = btrfs_commit_transaction(trans);
1122 btrfs_info_in_rcu(fs_info,
1123 "suspended dev_replace from %s (devid %llu) to %s canceled",
1124 btrfs_dev_name(src_device), src_device->devid,
1125 btrfs_dev_name(tgt_device));
1128 btrfs_destroy_dev_replace_tgtdev(tgt_device);
1131 up_write(&dev_replace->rwsem);
1135 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
1139 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
1141 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1143 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
1144 down_write(&dev_replace->rwsem);
1146 switch (dev_replace->replace_state) {
1147 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1148 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1149 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1150 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1152 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1153 dev_replace->replace_state =
1154 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
1155 dev_replace->time_stopped = ktime_get_real_seconds();
1156 dev_replace->item_needs_writeback = 1;
1157 btrfs_info(fs_info, "suspending dev_replace for unmount");
1161 up_write(&dev_replace->rwsem);
1162 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
1165 /* resume dev_replace procedure that was interrupted by unmount */
1166 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
1168 struct task_struct *task;
1169 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1171 down_write(&dev_replace->rwsem);
1173 switch (dev_replace->replace_state) {
1174 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1175 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1176 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1177 up_write(&dev_replace->rwsem);
1179 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1181 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1182 dev_replace->replace_state =
1183 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
1186 if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
1188 "cannot continue dev_replace, tgtdev is missing");
1190 "you may cancel the operation after 'mount -o degraded'");
1191 dev_replace->replace_state =
1192 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
1193 up_write(&dev_replace->rwsem);
1196 up_write(&dev_replace->rwsem);
1199 * This could collide with a paused balance, but the exclusive op logic
1200 * should never allow both to start and pause. We don't want to allow
1201 * dev-replace to start anyway.
1203 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
1204 down_write(&dev_replace->rwsem);
1205 dev_replace->replace_state =
1206 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
1207 up_write(&dev_replace->rwsem);
1209 "cannot resume dev-replace, other exclusive operation running");
1213 task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
1214 return PTR_ERR_OR_ZERO(task);
1217 static int btrfs_dev_replace_kthread(void *data)
1219 struct btrfs_fs_info *fs_info = data;
1220 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1224 progress = btrfs_dev_replace_progress(fs_info);
1225 progress = div_u64(progress, 10);
1226 btrfs_info_in_rcu(fs_info,
1227 "continuing dev_replace from %s (devid %llu) to target %s @%u%%",
1228 btrfs_dev_name(dev_replace->srcdev),
1229 dev_replace->srcdev->devid,
1230 btrfs_dev_name(dev_replace->tgtdev),
1231 (unsigned int)progress);
1233 ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
1234 dev_replace->committed_cursor_left,
1235 btrfs_device_get_total_bytes(dev_replace->srcdev),
1236 &dev_replace->scrub_progress, 0, 1);
1237 ret = btrfs_dev_replace_finishing(fs_info, ret);
1238 WARN_ON(ret && ret != -ECANCELED);
1240 btrfs_exclop_finish(fs_info);
1244 int __pure btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
1246 if (!dev_replace->is_valid)
1249 switch (dev_replace->replace_state) {
1250 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1251 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1252 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1254 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1255 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1257 * return true even if tgtdev is missing (this is
1258 * something that can happen if the dev_replace
1259 * procedure is suspended by an umount and then
1260 * the tgtdev is missing (or "btrfs dev scan") was
1261 * not called and the filesystem is remounted
1262 * in degraded state. This does not stop the
1263 * dev_replace procedure. It needs to be canceled
1264 * manually if the cancellation is wanted.
1271 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
1273 percpu_counter_sub(&fs_info->dev_replace.bio_counter, amount);
1274 cond_wake_up_nomb(&fs_info->dev_replace.replace_wait);
1277 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
1280 percpu_counter_inc(&fs_info->dev_replace.bio_counter);
1281 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1282 &fs_info->fs_state)))
1285 btrfs_bio_counter_dec(fs_info);
1286 wait_event(fs_info->dev_replace.replace_wait,
1287 !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1288 &fs_info->fs_state));