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 "check-integrity.h"
21 #include "rcu-string.h"
22 #include "dev-replace.h"
25 #include "block-group.h"
28 * Device replace overview
31 * To copy all extents (both new and on-disk) from source device to target
32 * device, while still keeping the filesystem read-write.
35 * There are two main methods involved:
39 * All new writes will be written to both target and source devices, so even
40 * if replace gets canceled, sources device still contans up-to-date data.
42 * Location: handle_ops_on_dev_replace() from __btrfs_map_block()
43 * Start: btrfs_dev_replace_start()
44 * End: btrfs_dev_replace_finishing()
45 * Content: Latest data/metadata
47 * - Copy existing extents
49 * This happens by re-using scrub facility, as scrub also iterates through
50 * existing extents from commit root.
52 * Location: scrub_write_block_to_dev_replace() from
53 * scrub_block_complete()
54 * Content: Data/meta from commit root.
56 * Due to the content difference, we need to avoid nocow write when dev-replace
57 * is happening. This is done by marking the block group read-only and waiting
60 * After replace is done, the finishing part is done by swapping the target and
63 * Location: btrfs_dev_replace_update_device_in_mapping_tree() from
64 * btrfs_dev_replace_finishing()
67 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
69 static int btrfs_dev_replace_kthread(void *data);
71 int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
74 struct btrfs_root *dev_root = fs_info->dev_root;
75 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
76 struct extent_buffer *eb;
79 struct btrfs_path *path = NULL;
81 struct btrfs_dev_replace_item *ptr;
87 path = btrfs_alloc_path();
94 key.type = BTRFS_DEV_REPLACE_KEY;
96 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
98 no_valid_dev_replace_entry_found:
100 * We don't have a replace item or it's corrupted. If there is
101 * a replace target, fail the mount.
103 if (btrfs_find_device(fs_info->fs_devices,
104 BTRFS_DEV_REPLACE_DEVID, NULL, NULL)) {
106 "found replace target device without a valid replace item");
111 dev_replace->replace_state =
112 BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
113 dev_replace->cont_reading_from_srcdev_mode =
114 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
115 dev_replace->time_started = 0;
116 dev_replace->time_stopped = 0;
117 atomic64_set(&dev_replace->num_write_errors, 0);
118 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
119 dev_replace->cursor_left = 0;
120 dev_replace->committed_cursor_left = 0;
121 dev_replace->cursor_left_last_write_of_item = 0;
122 dev_replace->cursor_right = 0;
123 dev_replace->srcdev = NULL;
124 dev_replace->tgtdev = NULL;
125 dev_replace->is_valid = 0;
126 dev_replace->item_needs_writeback = 0;
129 slot = path->slots[0];
131 item_size = btrfs_item_size_nr(eb, slot);
132 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
134 if (item_size != sizeof(struct btrfs_dev_replace_item)) {
136 "dev_replace entry found has unexpected size, ignore entry");
137 goto no_valid_dev_replace_entry_found;
140 src_devid = btrfs_dev_replace_src_devid(eb, ptr);
141 dev_replace->cont_reading_from_srcdev_mode =
142 btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
143 dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
144 dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
145 dev_replace->time_stopped =
146 btrfs_dev_replace_time_stopped(eb, ptr);
147 atomic64_set(&dev_replace->num_write_errors,
148 btrfs_dev_replace_num_write_errors(eb, ptr));
149 atomic64_set(&dev_replace->num_uncorrectable_read_errors,
150 btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
151 dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
152 dev_replace->committed_cursor_left = dev_replace->cursor_left;
153 dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
154 dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
155 dev_replace->is_valid = 1;
157 dev_replace->item_needs_writeback = 0;
158 switch (dev_replace->replace_state) {
159 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
160 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
161 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
163 * We don't have an active replace item but if there is a
164 * replace target, fail the mount.
166 if (btrfs_find_device(fs_info->fs_devices,
167 BTRFS_DEV_REPLACE_DEVID, NULL, NULL)) {
169 "replace devid present without an active replace item");
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->srcdev = btrfs_find_device(fs_info->fs_devices,
179 src_devid, NULL, NULL);
180 dev_replace->tgtdev = btrfs_find_device(fs_info->fs_devices,
181 BTRFS_DEV_REPLACE_DEVID,
184 * allow 'btrfs dev replace_cancel' if src/tgt device is
187 if (!dev_replace->srcdev &&
188 !btrfs_test_opt(fs_info, DEGRADED)) {
191 "cannot mount because device replace operation is ongoing and");
193 "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
196 if (!dev_replace->tgtdev &&
197 !btrfs_test_opt(fs_info, DEGRADED)) {
200 "cannot mount because device replace operation is ongoing and");
202 "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
203 BTRFS_DEV_REPLACE_DEVID);
205 if (dev_replace->tgtdev) {
206 if (dev_replace->srcdev) {
207 dev_replace->tgtdev->total_bytes =
208 dev_replace->srcdev->total_bytes;
209 dev_replace->tgtdev->disk_total_bytes =
210 dev_replace->srcdev->disk_total_bytes;
211 dev_replace->tgtdev->commit_total_bytes =
212 dev_replace->srcdev->commit_total_bytes;
213 dev_replace->tgtdev->bytes_used =
214 dev_replace->srcdev->bytes_used;
215 dev_replace->tgtdev->commit_bytes_used =
216 dev_replace->srcdev->commit_bytes_used;
218 set_bit(BTRFS_DEV_STATE_REPLACE_TGT,
219 &dev_replace->tgtdev->dev_state);
221 WARN_ON(fs_info->fs_devices->rw_devices == 0);
222 dev_replace->tgtdev->io_width = fs_info->sectorsize;
223 dev_replace->tgtdev->io_align = fs_info->sectorsize;
224 dev_replace->tgtdev->sector_size = fs_info->sectorsize;
225 dev_replace->tgtdev->fs_info = fs_info;
226 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
227 &dev_replace->tgtdev->dev_state);
233 btrfs_free_path(path);
238 * Initialize a new device for device replace target from a given source dev
241 * Return 0 and new device in @device_out, otherwise return < 0
243 static int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
244 const char *device_path,
245 struct btrfs_device *srcdev,
246 struct btrfs_device **device_out)
248 struct btrfs_device *device;
249 struct block_device *bdev;
250 struct rcu_string *name;
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 = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
261 fs_info->bdev_holder);
263 btrfs_err(fs_info, "target device %s is invalid!", device_path);
264 return PTR_ERR(bdev);
267 if (!btrfs_check_device_zone_type(fs_info, bdev)) {
269 "dev-replace: zoned type of target device mismatch with filesystem");
276 list_for_each_entry(device, &fs_info->fs_devices->devices, dev_list) {
277 if (device->bdev == bdev) {
279 "target device is in the filesystem!");
286 if (i_size_read(bdev->bd_inode) <
287 btrfs_device_get_total_bytes(srcdev)) {
289 "target device is smaller than source device!");
295 device = btrfs_alloc_device(NULL, &devid, NULL);
296 if (IS_ERR(device)) {
297 ret = PTR_ERR(device);
301 name = rcu_string_strdup(device_path, GFP_KERNEL);
303 btrfs_free_device(device);
307 rcu_assign_pointer(device->name, name);
309 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
310 device->generation = 0;
311 device->io_width = fs_info->sectorsize;
312 device->io_align = fs_info->sectorsize;
313 device->sector_size = fs_info->sectorsize;
314 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
315 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
316 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
317 device->commit_total_bytes = srcdev->commit_total_bytes;
318 device->commit_bytes_used = device->bytes_used;
319 device->fs_info = fs_info;
321 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
322 set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
323 device->mode = FMODE_EXCL;
324 device->dev_stats_valid = 1;
325 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
326 device->fs_devices = fs_info->fs_devices;
328 ret = btrfs_get_dev_zone_info(device);
332 mutex_lock(&fs_info->fs_devices->device_list_mutex);
333 list_add(&device->dev_list, &fs_info->fs_devices->devices);
334 fs_info->fs_devices->num_devices++;
335 fs_info->fs_devices->open_devices++;
336 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
338 *device_out = device;
342 blkdev_put(bdev, FMODE_EXCL);
347 * called from commit_transaction. Writes changed device replace state to
350 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans)
352 struct btrfs_fs_info *fs_info = trans->fs_info;
354 struct btrfs_root *dev_root = fs_info->dev_root;
355 struct btrfs_path *path;
356 struct btrfs_key key;
357 struct extent_buffer *eb;
358 struct btrfs_dev_replace_item *ptr;
359 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
361 down_read(&dev_replace->rwsem);
362 if (!dev_replace->is_valid ||
363 !dev_replace->item_needs_writeback) {
364 up_read(&dev_replace->rwsem);
367 up_read(&dev_replace->rwsem);
370 key.type = BTRFS_DEV_REPLACE_KEY;
373 path = btrfs_alloc_path();
378 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
381 "error %d while searching for dev_replace item!",
387 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
389 * need to delete old one and insert a new one.
390 * Since no attempt is made to recover any old state, if the
391 * dev_replace state is 'running', the data on the target
393 * It would be possible to recover the state: just make sure
394 * that the beginning of the item is never changed and always
395 * contains all the essential information. Then read this
396 * minimal set of information and use it as a base for the
399 ret = btrfs_del_item(trans, dev_root, path);
402 "delete too small dev_replace item failed %d!",
410 /* need to insert a new item */
411 btrfs_release_path(path);
412 ret = btrfs_insert_empty_item(trans, dev_root, path,
416 "insert dev_replace item failed %d!", ret);
422 ptr = btrfs_item_ptr(eb, path->slots[0],
423 struct btrfs_dev_replace_item);
425 down_write(&dev_replace->rwsem);
426 if (dev_replace->srcdev)
427 btrfs_set_dev_replace_src_devid(eb, ptr,
428 dev_replace->srcdev->devid);
430 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
431 btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
432 dev_replace->cont_reading_from_srcdev_mode);
433 btrfs_set_dev_replace_replace_state(eb, ptr,
434 dev_replace->replace_state);
435 btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
436 btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
437 btrfs_set_dev_replace_num_write_errors(eb, ptr,
438 atomic64_read(&dev_replace->num_write_errors));
439 btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
440 atomic64_read(&dev_replace->num_uncorrectable_read_errors));
441 dev_replace->cursor_left_last_write_of_item =
442 dev_replace->cursor_left;
443 btrfs_set_dev_replace_cursor_left(eb, ptr,
444 dev_replace->cursor_left_last_write_of_item);
445 btrfs_set_dev_replace_cursor_right(eb, ptr,
446 dev_replace->cursor_right);
447 dev_replace->item_needs_writeback = 0;
448 up_write(&dev_replace->rwsem);
450 btrfs_mark_buffer_dirty(eb);
453 btrfs_free_path(path);
458 static char* btrfs_dev_name(struct btrfs_device *device)
460 if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
461 return "<missing disk>";
463 return rcu_str_deref(device->name);
466 static int mark_block_group_to_copy(struct btrfs_fs_info *fs_info,
467 struct btrfs_device *src_dev)
469 struct btrfs_path *path;
470 struct btrfs_key key;
471 struct btrfs_key found_key;
472 struct btrfs_root *root = fs_info->dev_root;
473 struct btrfs_dev_extent *dev_extent = NULL;
474 struct btrfs_block_group *cache;
475 struct btrfs_trans_handle *trans;
479 /* Do not use "to_copy" on non zoned filesystem for now */
480 if (!btrfs_is_zoned(fs_info))
483 mutex_lock(&fs_info->chunk_mutex);
485 /* Ensure we don't have pending new block group */
486 spin_lock(&fs_info->trans_lock);
487 while (fs_info->running_transaction &&
488 !list_empty(&fs_info->running_transaction->dev_update_list)) {
489 spin_unlock(&fs_info->trans_lock);
490 mutex_unlock(&fs_info->chunk_mutex);
491 trans = btrfs_attach_transaction(root);
493 ret = PTR_ERR(trans);
494 mutex_lock(&fs_info->chunk_mutex);
495 if (ret == -ENOENT) {
496 spin_lock(&fs_info->trans_lock);
503 ret = btrfs_commit_transaction(trans);
504 mutex_lock(&fs_info->chunk_mutex);
508 spin_lock(&fs_info->trans_lock);
510 spin_unlock(&fs_info->trans_lock);
512 path = btrfs_alloc_path();
518 path->reada = READA_FORWARD;
519 path->search_commit_root = 1;
520 path->skip_locking = 1;
522 key.objectid = src_dev->devid;
523 key.type = BTRFS_DEV_EXTENT_KEY;
526 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
530 if (path->slots[0] >=
531 btrfs_header_nritems(path->nodes[0])) {
532 ret = btrfs_next_leaf(root, path);
545 struct extent_buffer *leaf = path->nodes[0];
546 int slot = path->slots[0];
548 btrfs_item_key_to_cpu(leaf, &found_key, slot);
550 if (found_key.objectid != src_dev->devid)
553 if (found_key.type != BTRFS_DEV_EXTENT_KEY)
556 if (found_key.offset < key.offset)
559 dev_extent = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
561 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dev_extent);
563 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
567 spin_lock(&cache->lock);
569 spin_unlock(&cache->lock);
571 btrfs_put_block_group(cache);
574 ret = btrfs_next_item(root, path);
583 btrfs_free_path(path);
585 mutex_unlock(&fs_info->chunk_mutex);
590 bool btrfs_finish_block_group_to_copy(struct btrfs_device *srcdev,
591 struct btrfs_block_group *cache,
594 struct btrfs_fs_info *fs_info = cache->fs_info;
595 struct extent_map *em;
596 struct map_lookup *map;
597 u64 chunk_offset = cache->start;
598 int num_extents, cur_extent;
601 /* Do not use "to_copy" on non zoned filesystem for now */
602 if (!btrfs_is_zoned(fs_info))
605 spin_lock(&cache->lock);
606 if (cache->removed) {
607 spin_unlock(&cache->lock);
610 spin_unlock(&cache->lock);
612 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
614 map = em->map_lookup;
616 num_extents = cur_extent = 0;
617 for (i = 0; i < map->num_stripes; i++) {
618 /* We have more device extent to copy */
619 if (srcdev != map->stripes[i].dev)
623 if (physical == map->stripes[i].physical)
629 if (num_extents > 1 && cur_extent < num_extents - 1) {
631 * Has more stripes on this device. Keep this block group
632 * readonly until we finish all the stripes.
637 /* Last stripe on this device */
638 spin_lock(&cache->lock);
640 spin_unlock(&cache->lock);
645 static int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
646 const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
649 struct btrfs_root *root = fs_info->dev_root;
650 struct btrfs_trans_handle *trans;
651 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
653 struct btrfs_device *tgt_device = NULL;
654 struct btrfs_device *src_device = NULL;
656 src_device = btrfs_find_device_by_devspec(fs_info, srcdevid,
658 if (IS_ERR(src_device))
659 return PTR_ERR(src_device);
661 if (btrfs_pinned_by_swapfile(fs_info, src_device)) {
662 btrfs_warn_in_rcu(fs_info,
663 "cannot replace device %s (devid %llu) due to active swapfile",
664 btrfs_dev_name(src_device), src_device->devid);
669 * Here we commit the transaction to make sure commit_total_bytes
670 * of all the devices are updated.
672 trans = btrfs_attach_transaction(root);
673 if (!IS_ERR(trans)) {
674 ret = btrfs_commit_transaction(trans);
677 } else if (PTR_ERR(trans) != -ENOENT) {
678 return PTR_ERR(trans);
681 ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
682 src_device, &tgt_device);
686 ret = mark_block_group_to_copy(fs_info, src_device);
690 down_write(&dev_replace->rwsem);
691 switch (dev_replace->replace_state) {
692 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
693 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
694 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
696 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
697 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
699 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
700 up_write(&dev_replace->rwsem);
704 dev_replace->cont_reading_from_srcdev_mode = read_src;
705 dev_replace->srcdev = src_device;
706 dev_replace->tgtdev = tgt_device;
708 btrfs_info_in_rcu(fs_info,
709 "dev_replace from %s (devid %llu) to %s started",
710 btrfs_dev_name(src_device),
712 rcu_str_deref(tgt_device->name));
715 * from now on, the writes to the srcdev are all duplicated to
716 * go to the tgtdev as well (refer to btrfs_map_block()).
718 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
719 dev_replace->time_started = ktime_get_real_seconds();
720 dev_replace->cursor_left = 0;
721 dev_replace->committed_cursor_left = 0;
722 dev_replace->cursor_left_last_write_of_item = 0;
723 dev_replace->cursor_right = 0;
724 dev_replace->is_valid = 1;
725 dev_replace->item_needs_writeback = 1;
726 atomic64_set(&dev_replace->num_write_errors, 0);
727 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
728 up_write(&dev_replace->rwsem);
730 ret = btrfs_sysfs_add_device(tgt_device);
732 btrfs_err(fs_info, "kobj add dev failed %d", ret);
734 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
736 /* Commit dev_replace state and reserve 1 item for it. */
737 trans = btrfs_start_transaction(root, 1);
739 ret = PTR_ERR(trans);
740 down_write(&dev_replace->rwsem);
741 dev_replace->replace_state =
742 BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
743 dev_replace->srcdev = NULL;
744 dev_replace->tgtdev = NULL;
745 up_write(&dev_replace->rwsem);
749 ret = btrfs_commit_transaction(trans);
752 /* the disk copy procedure reuses the scrub code */
753 ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
754 btrfs_device_get_total_bytes(src_device),
755 &dev_replace->scrub_progress, 0, 1);
757 ret = btrfs_dev_replace_finishing(fs_info, ret);
758 if (ret == -EINPROGRESS)
759 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
764 btrfs_destroy_dev_replace_tgtdev(tgt_device);
768 int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
769 struct btrfs_ioctl_dev_replace_args *args)
773 switch (args->start.cont_reading_from_srcdev_mode) {
774 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
775 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
781 if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
782 args->start.tgtdev_name[0] == '\0')
785 ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name,
786 args->start.srcdevid,
787 args->start.srcdev_name,
788 args->start.cont_reading_from_srcdev_mode);
790 /* don't warn if EINPROGRESS, someone else might be running scrub */
791 if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS ||
792 ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR)
799 * blocked until all in-flight bios operations are finished.
801 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
803 set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
804 wait_event(fs_info->dev_replace.replace_wait, !percpu_counter_sum(
805 &fs_info->dev_replace.bio_counter));
809 * we have removed target device, it is safe to allow new bios request.
811 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
813 clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
814 wake_up(&fs_info->dev_replace.replace_wait);
818 * When finishing the device replace, before swapping the source device with the
819 * target device we must update the chunk allocation state in the target device,
820 * as it is empty because replace works by directly copying the chunks and not
821 * through the normal chunk allocation path.
823 static int btrfs_set_target_alloc_state(struct btrfs_device *srcdev,
824 struct btrfs_device *tgtdev)
826 struct extent_state *cached_state = NULL;
832 lockdep_assert_held(&srcdev->fs_info->chunk_mutex);
834 while (!find_first_extent_bit(&srcdev->alloc_state, start,
835 &found_start, &found_end,
836 CHUNK_ALLOCATED, &cached_state)) {
837 ret = set_extent_bits(&tgtdev->alloc_state, found_start,
838 found_end, CHUNK_ALLOCATED);
841 start = found_end + 1;
844 free_extent_state(cached_state);
848 static void btrfs_dev_replace_update_device_in_mapping_tree(
849 struct btrfs_fs_info *fs_info,
850 struct btrfs_device *srcdev,
851 struct btrfs_device *tgtdev)
853 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
854 struct extent_map *em;
855 struct map_lookup *map;
859 write_lock(&em_tree->lock);
861 em = lookup_extent_mapping(em_tree, start, (u64)-1);
864 map = em->map_lookup;
865 for (i = 0; i < map->num_stripes; i++)
866 if (srcdev == map->stripes[i].dev)
867 map->stripes[i].dev = tgtdev;
868 start = em->start + em->len;
871 write_unlock(&em_tree->lock);
874 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
877 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
878 struct btrfs_device *tgt_device;
879 struct btrfs_device *src_device;
880 struct btrfs_root *root = fs_info->tree_root;
881 u8 uuid_tmp[BTRFS_UUID_SIZE];
882 struct btrfs_trans_handle *trans;
885 /* don't allow cancel or unmount to disturb the finishing procedure */
886 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
888 down_read(&dev_replace->rwsem);
889 /* was the operation canceled, or is it finished? */
890 if (dev_replace->replace_state !=
891 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
892 up_read(&dev_replace->rwsem);
893 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
897 tgt_device = dev_replace->tgtdev;
898 src_device = dev_replace->srcdev;
899 up_read(&dev_replace->rwsem);
902 * flush all outstanding I/O and inode extent mappings before the
903 * copy operation is declared as being finished
905 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
907 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
910 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
913 btrfs_reada_remove_dev(src_device);
916 * We have to use this loop approach because at this point src_device
917 * has to be available for transaction commit to complete, yet new
918 * chunks shouldn't be allocated on the device.
921 trans = btrfs_start_transaction(root, 0);
923 btrfs_reada_undo_remove_dev(src_device);
924 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
925 return PTR_ERR(trans);
927 ret = btrfs_commit_transaction(trans);
930 /* Prevent write_all_supers() during the finishing procedure */
931 mutex_lock(&fs_info->fs_devices->device_list_mutex);
932 /* Prevent new chunks being allocated on the source device */
933 mutex_lock(&fs_info->chunk_mutex);
935 if (!list_empty(&src_device->post_commit_list)) {
936 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
937 mutex_unlock(&fs_info->chunk_mutex);
943 down_write(&dev_replace->rwsem);
944 dev_replace->replace_state =
945 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
946 : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
947 dev_replace->tgtdev = NULL;
948 dev_replace->srcdev = NULL;
949 dev_replace->time_stopped = ktime_get_real_seconds();
950 dev_replace->item_needs_writeback = 1;
953 * Update allocation state in the new device and replace the old device
954 * with the new one in the mapping tree.
957 scrub_ret = btrfs_set_target_alloc_state(src_device, tgt_device);
960 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
964 if (scrub_ret != -ECANCELED)
965 btrfs_err_in_rcu(fs_info,
966 "btrfs_scrub_dev(%s, %llu, %s) failed %d",
967 btrfs_dev_name(src_device),
969 rcu_str_deref(tgt_device->name), scrub_ret);
971 up_write(&dev_replace->rwsem);
972 mutex_unlock(&fs_info->chunk_mutex);
973 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
974 btrfs_reada_undo_remove_dev(src_device);
975 btrfs_rm_dev_replace_blocked(fs_info);
977 btrfs_destroy_dev_replace_tgtdev(tgt_device);
978 btrfs_rm_dev_replace_unblocked(fs_info);
979 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
984 btrfs_info_in_rcu(fs_info,
985 "dev_replace from %s (devid %llu) to %s finished",
986 btrfs_dev_name(src_device),
988 rcu_str_deref(tgt_device->name));
989 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state);
990 tgt_device->devid = src_device->devid;
991 src_device->devid = BTRFS_DEV_REPLACE_DEVID;
992 memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
993 memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
994 memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
995 btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
996 btrfs_device_set_disk_total_bytes(tgt_device,
997 src_device->disk_total_bytes);
998 btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
999 tgt_device->commit_bytes_used = src_device->bytes_used;
1001 btrfs_assign_next_active_device(src_device, tgt_device);
1003 list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
1004 fs_info->fs_devices->rw_devices++;
1006 up_write(&dev_replace->rwsem);
1007 btrfs_rm_dev_replace_blocked(fs_info);
1009 btrfs_rm_dev_replace_remove_srcdev(src_device);
1011 btrfs_rm_dev_replace_unblocked(fs_info);
1014 * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will
1015 * update on-disk dev stats value during commit transaction
1017 atomic_inc(&tgt_device->dev_stats_ccnt);
1020 * this is again a consistent state where no dev_replace procedure
1021 * is running, the target device is part of the filesystem, the
1022 * source device is not part of the filesystem anymore and its 1st
1023 * superblock is scratched out so that it is no longer marked to
1024 * belong to this filesystem.
1026 mutex_unlock(&fs_info->chunk_mutex);
1027 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1029 /* replace the sysfs entry */
1030 btrfs_sysfs_remove_device(src_device);
1031 btrfs_sysfs_update_devid(tgt_device);
1032 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &src_device->dev_state))
1033 btrfs_scratch_superblocks(fs_info, src_device->bdev,
1034 src_device->name->str);
1036 /* write back the superblocks */
1037 trans = btrfs_start_transaction(root, 0);
1039 btrfs_commit_transaction(trans);
1041 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
1043 btrfs_rm_dev_replace_free_srcdev(src_device);
1049 * Read progress of device replace status according to the state and last
1050 * stored position. The value format is the same as for
1051 * btrfs_dev_replace::progress_1000
1053 static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info)
1055 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1058 switch (dev_replace->replace_state) {
1059 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1060 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1063 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1066 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1067 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1068 ret = div64_u64(dev_replace->cursor_left,
1069 div_u64(btrfs_device_get_total_bytes(
1070 dev_replace->srcdev), 1000));
1077 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
1078 struct btrfs_ioctl_dev_replace_args *args)
1080 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1082 down_read(&dev_replace->rwsem);
1083 /* even if !dev_replace_is_valid, the values are good enough for
1084 * the replace_status ioctl */
1085 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
1086 args->status.replace_state = dev_replace->replace_state;
1087 args->status.time_started = dev_replace->time_started;
1088 args->status.time_stopped = dev_replace->time_stopped;
1089 args->status.num_write_errors =
1090 atomic64_read(&dev_replace->num_write_errors);
1091 args->status.num_uncorrectable_read_errors =
1092 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
1093 args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
1094 up_read(&dev_replace->rwsem);
1097 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
1099 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1100 struct btrfs_device *tgt_device = NULL;
1101 struct btrfs_device *src_device = NULL;
1102 struct btrfs_trans_handle *trans;
1103 struct btrfs_root *root = fs_info->tree_root;
1107 if (sb_rdonly(fs_info->sb))
1110 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
1111 down_write(&dev_replace->rwsem);
1112 switch (dev_replace->replace_state) {
1113 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1114 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1115 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1116 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
1117 up_write(&dev_replace->rwsem);
1119 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1120 tgt_device = dev_replace->tgtdev;
1121 src_device = dev_replace->srcdev;
1122 up_write(&dev_replace->rwsem);
1123 ret = btrfs_scrub_cancel(fs_info);
1125 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
1127 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
1129 * btrfs_dev_replace_finishing() will handle the
1132 btrfs_info_in_rcu(fs_info,
1133 "dev_replace from %s (devid %llu) to %s canceled",
1134 btrfs_dev_name(src_device), src_device->devid,
1135 btrfs_dev_name(tgt_device));
1138 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1140 * Scrub doing the replace isn't running so we need to do the
1141 * cleanup step of btrfs_dev_replace_finishing() here
1143 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
1144 tgt_device = dev_replace->tgtdev;
1145 src_device = dev_replace->srcdev;
1146 dev_replace->tgtdev = NULL;
1147 dev_replace->srcdev = NULL;
1148 dev_replace->replace_state =
1149 BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
1150 dev_replace->time_stopped = ktime_get_real_seconds();
1151 dev_replace->item_needs_writeback = 1;
1153 up_write(&dev_replace->rwsem);
1155 /* Scrub for replace must not be running in suspended state */
1156 ret = btrfs_scrub_cancel(fs_info);
1157 ASSERT(ret != -ENOTCONN);
1159 trans = btrfs_start_transaction(root, 0);
1160 if (IS_ERR(trans)) {
1161 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
1162 return PTR_ERR(trans);
1164 ret = btrfs_commit_transaction(trans);
1167 btrfs_info_in_rcu(fs_info,
1168 "suspended dev_replace from %s (devid %llu) to %s canceled",
1169 btrfs_dev_name(src_device), src_device->devid,
1170 btrfs_dev_name(tgt_device));
1173 btrfs_destroy_dev_replace_tgtdev(tgt_device);
1176 up_write(&dev_replace->rwsem);
1180 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
1184 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
1186 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1188 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
1189 down_write(&dev_replace->rwsem);
1191 switch (dev_replace->replace_state) {
1192 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1193 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1194 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1195 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1197 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1198 dev_replace->replace_state =
1199 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
1200 dev_replace->time_stopped = ktime_get_real_seconds();
1201 dev_replace->item_needs_writeback = 1;
1202 btrfs_info(fs_info, "suspending dev_replace for unmount");
1206 up_write(&dev_replace->rwsem);
1207 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
1210 /* resume dev_replace procedure that was interrupted by unmount */
1211 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
1213 struct task_struct *task;
1214 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1216 down_write(&dev_replace->rwsem);
1218 switch (dev_replace->replace_state) {
1219 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1220 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1221 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1222 up_write(&dev_replace->rwsem);
1224 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1226 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1227 dev_replace->replace_state =
1228 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
1231 if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
1233 "cannot continue dev_replace, tgtdev is missing");
1235 "you may cancel the operation after 'mount -o degraded'");
1236 dev_replace->replace_state =
1237 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
1238 up_write(&dev_replace->rwsem);
1241 up_write(&dev_replace->rwsem);
1244 * This could collide with a paused balance, but the exclusive op logic
1245 * should never allow both to start and pause. We don't want to allow
1246 * dev-replace to start anyway.
1248 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
1249 down_write(&dev_replace->rwsem);
1250 dev_replace->replace_state =
1251 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
1252 up_write(&dev_replace->rwsem);
1254 "cannot resume dev-replace, other exclusive operation running");
1258 task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
1259 return PTR_ERR_OR_ZERO(task);
1262 static int btrfs_dev_replace_kthread(void *data)
1264 struct btrfs_fs_info *fs_info = data;
1265 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1269 progress = btrfs_dev_replace_progress(fs_info);
1270 progress = div_u64(progress, 10);
1271 btrfs_info_in_rcu(fs_info,
1272 "continuing dev_replace from %s (devid %llu) to target %s @%u%%",
1273 btrfs_dev_name(dev_replace->srcdev),
1274 dev_replace->srcdev->devid,
1275 btrfs_dev_name(dev_replace->tgtdev),
1276 (unsigned int)progress);
1278 ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
1279 dev_replace->committed_cursor_left,
1280 btrfs_device_get_total_bytes(dev_replace->srcdev),
1281 &dev_replace->scrub_progress, 0, 1);
1282 ret = btrfs_dev_replace_finishing(fs_info, ret);
1283 WARN_ON(ret && ret != -ECANCELED);
1285 btrfs_exclop_finish(fs_info);
1289 int __pure btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
1291 if (!dev_replace->is_valid)
1294 switch (dev_replace->replace_state) {
1295 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1296 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1297 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1299 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1300 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1302 * return true even if tgtdev is missing (this is
1303 * something that can happen if the dev_replace
1304 * procedure is suspended by an umount and then
1305 * the tgtdev is missing (or "btrfs dev scan") was
1306 * not called and the filesystem is remounted
1307 * in degraded state. This does not stop the
1308 * dev_replace procedure. It needs to be canceled
1309 * manually if the cancellation is wanted.
1316 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
1318 percpu_counter_inc(&fs_info->dev_replace.bio_counter);
1321 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
1323 percpu_counter_sub(&fs_info->dev_replace.bio_counter, amount);
1324 cond_wake_up_nomb(&fs_info->dev_replace.replace_wait);
1327 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
1330 percpu_counter_inc(&fs_info->dev_replace.bio_counter);
1331 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1332 &fs_info->fs_state)))
1335 btrfs_bio_counter_dec(fs_info);
1336 wait_event(fs_info->dev_replace.replace_wait,
1337 !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1338 &fs_info->fs_state));