1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) STRATO AG 2012. All rights reserved.
4 */
5
6#include <linux/sched.h>
7#include <linux/bio.h>
8#include <linux/slab.h>
9#include <linux/blkdev.h>
10#include <linux/kthread.h>
11#include <linux/math64.h>
12#include "misc.h"
13#include "ctree.h"
14#include "disk-io.h"
15#include "transaction.h"
16#include "volumes.h"
17#include "async-thread.h"
18#include "dev-replace.h"
19#include "sysfs.h"
20#include "zoned.h"
21#include "block-group.h"
22#include "fs.h"
23#include "accessors.h"
24#include "scrub.h"
25
26/*
27 * Device replace overview
28 *
29 * [Objective]
30 * To copy all extents (both new and on-disk) from source device to target
31 * device, while still keeping the filesystem read-write.
32 *
33 * [Method]
34 * There are two main methods involved:
35 *
36 * - Write duplication
37 *
38 * All new writes will be written to both target and source devices, so even
39 * if replace gets canceled, sources device still contains up-to-date data.
40 *
41 * Location: handle_ops_on_dev_replace() from btrfs_map_block()
42 * Start: btrfs_dev_replace_start()
43 * End: btrfs_dev_replace_finishing()
44 * Content: Latest data/metadata
45 *
46 * - Copy existing extents
47 *
48 * This happens by re-using scrub facility, as scrub also iterates through
49 * existing extents from commit root.
50 *
51 * Location: scrub_write_block_to_dev_replace() from
52 * scrub_block_complete()
53 * Content: Data/meta from commit root.
54 *
55 * Due to the content difference, we need to avoid nocow write when dev-replace
56 * is happening. This is done by marking the block group read-only and waiting
57 * for NOCOW writes.
58 *
59 * After replace is done, the finishing part is done by swapping the target and
60 * source devices.
61 *
62 * Location: btrfs_dev_replace_update_device_in_mapping_tree() from
63 * btrfs_dev_replace_finishing()
64 */
65
66static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
67 int scrub_ret);
68static int btrfs_dev_replace_kthread(void *data);
69
70int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
71{
72 struct btrfs_dev_lookup_args args = { .devid = BTRFS_DEV_REPLACE_DEVID };
73 struct btrfs_key key;
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;
77 int slot;
78 int ret = 0;
79 struct btrfs_path *path = NULL;
80 int item_size;
81 struct btrfs_dev_replace_item *ptr;
82 u64 src_devid;
83
84 if (!dev_root)
85 return 0;
86
87 path = btrfs_alloc_path();
88 if (!path) {
89 ret = -ENOMEM;
90 goto out;
91 }
92
93 key.objectid = 0;
94 key.type = BTRFS_DEV_REPLACE_KEY;
95 key.offset = 0;
96 ret = btrfs_search_slot(NULL, root: dev_root, key: &key, p: path, ins_len: 0, cow: 0);
97 if (ret) {
98no_valid_dev_replace_entry_found:
99 /*
100 * We don't have a replace item or it's corrupted. If there is
101 * a replace target, fail the mount.
102 */
103 if (btrfs_find_device(fs_devices: fs_info->fs_devices, args: &args)) {
104 btrfs_err(fs_info,
105 "found replace target device without a valid replace item");
106 ret = -EUCLEAN;
107 goto out;
108 }
109 ret = 0;
110 dev_replace->replace_state =
111 BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
112 dev_replace->cont_reading_from_srcdev_mode =
113 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
114 dev_replace->time_started = 0;
115 dev_replace->time_stopped = 0;
116 atomic64_set(v: &dev_replace->num_write_errors, i: 0);
117 atomic64_set(v: &dev_replace->num_uncorrectable_read_errors, i: 0);
118 dev_replace->cursor_left = 0;
119 dev_replace->committed_cursor_left = 0;
120 dev_replace->cursor_left_last_write_of_item = 0;
121 dev_replace->cursor_right = 0;
122 dev_replace->srcdev = NULL;
123 dev_replace->tgtdev = NULL;
124 dev_replace->is_valid = 0;
125 dev_replace->item_needs_writeback = 0;
126 goto out;
127 }
128 slot = path->slots[0];
129 eb = path->nodes[0];
130 item_size = btrfs_item_size(eb, slot);
131 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
132
133 if (item_size != sizeof(struct btrfs_dev_replace_item)) {
134 btrfs_warn(fs_info,
135 "dev_replace entry found has unexpected size, ignore entry");
136 goto no_valid_dev_replace_entry_found;
137 }
138
139 src_devid = btrfs_dev_replace_src_devid(eb, s: ptr);
140 dev_replace->cont_reading_from_srcdev_mode =
141 btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, s: ptr);
142 dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, s: ptr);
143 dev_replace->time_started = btrfs_dev_replace_time_started(eb, s: ptr);
144 dev_replace->time_stopped =
145 btrfs_dev_replace_time_stopped(eb, s: ptr);
146 atomic64_set(v: &dev_replace->num_write_errors,
147 i: btrfs_dev_replace_num_write_errors(eb, s: ptr));
148 atomic64_set(v: &dev_replace->num_uncorrectable_read_errors,
149 i: btrfs_dev_replace_num_uncorrectable_read_errors(eb, s: ptr));
150 dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, s: ptr);
151 dev_replace->committed_cursor_left = dev_replace->cursor_left;
152 dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
153 dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, s: ptr);
154 dev_replace->is_valid = 1;
155
156 dev_replace->item_needs_writeback = 0;
157 switch (dev_replace->replace_state) {
158 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
159 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
160 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
161 /*
162 * We don't have an active replace item but if there is a
163 * replace target, fail the mount.
164 */
165 if (btrfs_find_device(fs_devices: fs_info->fs_devices, args: &args)) {
166 btrfs_err(fs_info,
167"replace without active item, run 'device scan --forget' on the target device");
168 ret = -EUCLEAN;
169 } else {
170 dev_replace->srcdev = NULL;
171 dev_replace->tgtdev = NULL;
172 }
173 break;
174 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
175 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
176 dev_replace->tgtdev = btrfs_find_device(fs_devices: fs_info->fs_devices, args: &args);
177 args.devid = src_devid;
178 dev_replace->srcdev = btrfs_find_device(fs_devices: fs_info->fs_devices, args: &args);
179
180 /*
181 * allow 'btrfs dev replace_cancel' if src/tgt device is
182 * missing
183 */
184 if (!dev_replace->srcdev &&
185 !btrfs_test_opt(fs_info, DEGRADED)) {
186 ret = -EIO;
187 btrfs_warn(fs_info,
188 "cannot mount because device replace operation is ongoing and");
189 btrfs_warn(fs_info,
190 "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
191 src_devid);
192 }
193 if (!dev_replace->tgtdev &&
194 !btrfs_test_opt(fs_info, DEGRADED)) {
195 ret = -EIO;
196 btrfs_warn(fs_info,
197 "cannot mount because device replace operation is ongoing and");
198 btrfs_warn(fs_info,
199 "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
200 BTRFS_DEV_REPLACE_DEVID);
201 }
202 if (dev_replace->tgtdev) {
203 if (dev_replace->srcdev) {
204 dev_replace->tgtdev->total_bytes =
205 dev_replace->srcdev->total_bytes;
206 dev_replace->tgtdev->disk_total_bytes =
207 dev_replace->srcdev->disk_total_bytes;
208 dev_replace->tgtdev->commit_total_bytes =
209 dev_replace->srcdev->commit_total_bytes;
210 dev_replace->tgtdev->bytes_used =
211 dev_replace->srcdev->bytes_used;
212 dev_replace->tgtdev->commit_bytes_used =
213 dev_replace->srcdev->commit_bytes_used;
214 }
215 set_bit(BTRFS_DEV_STATE_REPLACE_TGT,
216 addr: &dev_replace->tgtdev->dev_state);
217
218 WARN_ON(fs_info->fs_devices->rw_devices == 0);
219 dev_replace->tgtdev->io_width = fs_info->sectorsize;
220 dev_replace->tgtdev->io_align = fs_info->sectorsize;
221 dev_replace->tgtdev->sector_size = fs_info->sectorsize;
222 dev_replace->tgtdev->fs_info = fs_info;
223 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
224 addr: &dev_replace->tgtdev->dev_state);
225 }
226 break;
227 }
228
229out:
230 btrfs_free_path(p: path);
231 return ret;
232}
233
234/*
235 * Initialize a new device for device replace target from a given source dev
236 * and path.
237 *
238 * Return 0 and new device in @device_out, otherwise return < 0
239 */
240static int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
241 const char *device_path,
242 struct btrfs_device *srcdev,
243 struct btrfs_device **device_out)
244{
245 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
246 struct btrfs_device *device;
247 struct file *bdev_file;
248 struct block_device *bdev;
249 u64 devid = BTRFS_DEV_REPLACE_DEVID;
250 int ret = 0;
251
252 *device_out = NULL;
253 if (srcdev->fs_devices->seeding) {
254 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
255 return -EINVAL;
256 }
257
258 bdev_file = bdev_file_open_by_path(path: device_path, BLK_OPEN_WRITE,
259 holder: fs_info->bdev_holder, NULL);
260 if (IS_ERR(ptr: bdev_file)) {
261 btrfs_err(fs_info, "target device %s is invalid!", device_path);
262 return PTR_ERR(ptr: bdev_file);
263 }
264 bdev = file_bdev(bdev_file);
265
266 if (!btrfs_check_device_zone_type(fs_info, bdev)) {
267 btrfs_err(fs_info,
268 "dev-replace: zoned type of target device mismatch with filesystem");
269 ret = -EINVAL;
270 goto error;
271 }
272
273 sync_blockdev(bdev);
274
275 list_for_each_entry(device, &fs_devices->devices, dev_list) {
276 if (device->bdev == bdev) {
277 btrfs_err(fs_info,
278 "target device is in the filesystem!");
279 ret = -EEXIST;
280 goto error;
281 }
282 }
283
284
285 if (bdev_nr_bytes(bdev) < btrfs_device_get_total_bytes(dev: srcdev)) {
286 btrfs_err(fs_info,
287 "target device is smaller than source device!");
288 ret = -EINVAL;
289 goto error;
290 }
291
292
293 device = btrfs_alloc_device(NULL, devid: &devid, NULL, path: device_path);
294 if (IS_ERR(ptr: device)) {
295 ret = PTR_ERR(ptr: device);
296 goto error;
297 }
298
299 ret = lookup_bdev(pathname: device_path, dev: &device->devt);
300 if (ret)
301 goto error;
302
303 set_bit(BTRFS_DEV_STATE_WRITEABLE, addr: &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(dev: srcdev);
309 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(dev: srcdev);
310 device->bytes_used = btrfs_device_get_bytes_used(dev: srcdev);
311 device->commit_total_bytes = srcdev->commit_total_bytes;
312 device->commit_bytes_used = device->bytes_used;
313 device->fs_info = fs_info;
314 device->bdev = bdev;
315 device->bdev_file = bdev_file;
316 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, addr: &device->dev_state);
317 set_bit(BTRFS_DEV_STATE_REPLACE_TGT, addr: &device->dev_state);
318 device->dev_stats_valid = 1;
319 set_blocksize(bdev: device->bdev, BTRFS_BDEV_BLOCKSIZE);
320 device->fs_devices = fs_devices;
321
322 ret = btrfs_get_dev_zone_info(device, populate_cache: false);
323 if (ret)
324 goto error;
325
326 mutex_lock(&fs_devices->device_list_mutex);
327 list_add(new: &device->dev_list, head: &fs_devices->devices);
328 fs_devices->num_devices++;
329 fs_devices->open_devices++;
330 mutex_unlock(lock: &fs_devices->device_list_mutex);
331
332 *device_out = device;
333 return 0;
334
335error:
336 fput(bdev_file);
337 return ret;
338}
339
340/*
341 * called from commit_transaction. Writes changed device replace state to
342 * disk.
343 */
344int btrfs_run_dev_replace(struct btrfs_trans_handle *trans)
345{
346 struct btrfs_fs_info *fs_info = trans->fs_info;
347 int ret;
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;
354
355 down_read(sem: &dev_replace->rwsem);
356 if (!dev_replace->is_valid ||
357 !dev_replace->item_needs_writeback) {
358 up_read(sem: &dev_replace->rwsem);
359 return 0;
360 }
361 up_read(sem: &dev_replace->rwsem);
362
363 key.objectid = 0;
364 key.type = BTRFS_DEV_REPLACE_KEY;
365 key.offset = 0;
366
367 path = btrfs_alloc_path();
368 if (!path) {
369 ret = -ENOMEM;
370 goto out;
371 }
372 ret = btrfs_search_slot(trans, root: dev_root, key: &key, p: path, ins_len: -1, cow: 1);
373 if (ret < 0) {
374 btrfs_warn(fs_info,
375 "error %d while searching for dev_replace item!",
376 ret);
377 goto out;
378 }
379
380 if (ret == 0 &&
381 btrfs_item_size(eb: path->nodes[0], slot: path->slots[0]) < sizeof(*ptr)) {
382 /*
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
386 * drive is lost.
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
391 * new state.
392 */
393 ret = btrfs_del_item(trans, root: dev_root, path);
394 if (ret != 0) {
395 btrfs_warn(fs_info,
396 "delete too small dev_replace item failed %d!",
397 ret);
398 goto out;
399 }
400 ret = 1;
401 }
402
403 if (ret == 1) {
404 /* need to insert a new item */
405 btrfs_release_path(p: path);
406 ret = btrfs_insert_empty_item(trans, root: dev_root, path,
407 key: &key, data_size: sizeof(*ptr));
408 if (ret < 0) {
409 btrfs_warn(fs_info,
410 "insert dev_replace item failed %d!", ret);
411 goto out;
412 }
413 }
414
415 eb = path->nodes[0];
416 ptr = btrfs_item_ptr(eb, path->slots[0],
417 struct btrfs_dev_replace_item);
418
419 down_write(sem: &dev_replace->rwsem);
420 if (dev_replace->srcdev)
421 btrfs_set_dev_replace_src_devid(eb, s: ptr,
422 val: dev_replace->srcdev->devid);
423 else
424 btrfs_set_dev_replace_src_devid(eb, s: ptr, val: (u64)-1);
425 btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, s: ptr,
426 val: dev_replace->cont_reading_from_srcdev_mode);
427 btrfs_set_dev_replace_replace_state(eb, s: ptr,
428 val: dev_replace->replace_state);
429 btrfs_set_dev_replace_time_started(eb, s: ptr, val: dev_replace->time_started);
430 btrfs_set_dev_replace_time_stopped(eb, s: ptr, val: dev_replace->time_stopped);
431 btrfs_set_dev_replace_num_write_errors(eb, s: ptr,
432 val: atomic64_read(v: &dev_replace->num_write_errors));
433 btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, s: ptr,
434 val: atomic64_read(v: &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, s: ptr,
438 val: dev_replace->cursor_left_last_write_of_item);
439 btrfs_set_dev_replace_cursor_right(eb, s: ptr,
440 val: dev_replace->cursor_right);
441 dev_replace->item_needs_writeback = 0;
442 up_write(sem: &dev_replace->rwsem);
443
444 btrfs_mark_buffer_dirty(trans, buf: eb);
445
446out:
447 btrfs_free_path(p: path);
448
449 return ret;
450}
451
452static int mark_block_group_to_copy(struct btrfs_fs_info *fs_info,
453 struct btrfs_device *src_dev)
454{
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;
462 int iter_ret = 0;
463 int ret = 0;
464 u64 chunk_offset;
465
466 /* Do not use "to_copy" on non zoned filesystem for now */
467 if (!btrfs_is_zoned(fs_info))
468 return 0;
469
470 mutex_lock(&fs_info->chunk_mutex);
471
472 /* Ensure we don't have pending new block group */
473 spin_lock(lock: &fs_info->trans_lock);
474 while (fs_info->running_transaction &&
475 !list_empty(head: &fs_info->running_transaction->dev_update_list)) {
476 spin_unlock(lock: &fs_info->trans_lock);
477 mutex_unlock(lock: &fs_info->chunk_mutex);
478 trans = btrfs_attach_transaction(root);
479 if (IS_ERR(ptr: trans)) {
480 ret = PTR_ERR(ptr: trans);
481 mutex_lock(&fs_info->chunk_mutex);
482 if (ret == -ENOENT) {
483 spin_lock(lock: &fs_info->trans_lock);
484 continue;
485 } else {
486 goto unlock;
487 }
488 }
489
490 ret = btrfs_commit_transaction(trans);
491 mutex_lock(&fs_info->chunk_mutex);
492 if (ret)
493 goto unlock;
494
495 spin_lock(lock: &fs_info->trans_lock);
496 }
497 spin_unlock(lock: &fs_info->trans_lock);
498
499 path = btrfs_alloc_path();
500 if (!path) {
501 ret = -ENOMEM;
502 goto unlock;
503 }
504
505 path->reada = READA_FORWARD;
506 path->search_commit_root = 1;
507 path->skip_locking = 1;
508
509 key.objectid = src_dev->devid;
510 key.type = BTRFS_DEV_EXTENT_KEY;
511 key.offset = 0;
512
513 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) {
514 struct extent_buffer *leaf = path->nodes[0];
515
516 if (found_key.objectid != src_dev->devid)
517 break;
518
519 if (found_key.type != BTRFS_DEV_EXTENT_KEY)
520 break;
521
522 if (found_key.offset < key.offset)
523 break;
524
525 dev_extent = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent);
526
527 chunk_offset = btrfs_dev_extent_chunk_offset(eb: leaf, s: dev_extent);
528
529 cache = btrfs_lookup_block_group(info: fs_info, bytenr: chunk_offset);
530 if (!cache)
531 continue;
532
533 set_bit(nr: BLOCK_GROUP_FLAG_TO_COPY, addr: &cache->runtime_flags);
534 btrfs_put_block_group(cache);
535 }
536 if (iter_ret < 0)
537 ret = iter_ret;
538
539 btrfs_free_path(p: path);
540unlock:
541 mutex_unlock(lock: &fs_info->chunk_mutex);
542
543 return ret;
544}
545
546bool btrfs_finish_block_group_to_copy(struct btrfs_device *srcdev,
547 struct btrfs_block_group *cache,
548 u64 physical)
549{
550 struct btrfs_fs_info *fs_info = cache->fs_info;
551 struct btrfs_chunk_map *map;
552 u64 chunk_offset = cache->start;
553 int num_extents, cur_extent;
554 int i;
555
556 /* Do not use "to_copy" on non zoned filesystem for now */
557 if (!btrfs_is_zoned(fs_info))
558 return true;
559
560 spin_lock(lock: &cache->lock);
561 if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &cache->runtime_flags)) {
562 spin_unlock(lock: &cache->lock);
563 return true;
564 }
565 spin_unlock(lock: &cache->lock);
566
567 map = btrfs_get_chunk_map(fs_info, logical: chunk_offset, length: 1);
568 ASSERT(!IS_ERR(map));
569
570 num_extents = 0;
571 cur_extent = 0;
572 for (i = 0; i < map->num_stripes; i++) {
573 /* We have more device extent to copy */
574 if (srcdev != map->stripes[i].dev)
575 continue;
576
577 num_extents++;
578 if (physical == map->stripes[i].physical)
579 cur_extent = i;
580 }
581
582 btrfs_free_chunk_map(map);
583
584 if (num_extents > 1 && cur_extent < num_extents - 1) {
585 /*
586 * Has more stripes on this device. Keep this block group
587 * readonly until we finish all the stripes.
588 */
589 return false;
590 }
591
592 /* Last stripe on this device */
593 clear_bit(nr: BLOCK_GROUP_FLAG_TO_COPY, addr: &cache->runtime_flags);
594
595 return true;
596}
597
598static int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
599 const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
600 int read_src)
601{
602 struct btrfs_root *root = fs_info->dev_root;
603 struct btrfs_trans_handle *trans;
604 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
605 int ret;
606 struct btrfs_device *tgt_device = NULL;
607 struct btrfs_device *src_device = NULL;
608
609 src_device = btrfs_find_device_by_devspec(fs_info, devid: srcdevid,
610 devpath: srcdev_name);
611 if (IS_ERR(ptr: src_device))
612 return PTR_ERR(ptr: src_device);
613
614 if (btrfs_pinned_by_swapfile(fs_info, ptr: src_device)) {
615 btrfs_warn_in_rcu(fs_info,
616 "cannot replace device %s (devid %llu) due to active swapfile",
617 btrfs_dev_name(src_device), src_device->devid);
618 return -ETXTBSY;
619 }
620
621 /*
622 * Here we commit the transaction to make sure commit_total_bytes
623 * of all the devices are updated.
624 */
625 trans = btrfs_attach_transaction(root);
626 if (!IS_ERR(ptr: trans)) {
627 ret = btrfs_commit_transaction(trans);
628 if (ret)
629 return ret;
630 } else if (PTR_ERR(ptr: trans) != -ENOENT) {
631 return PTR_ERR(ptr: trans);
632 }
633
634 ret = btrfs_init_dev_replace_tgtdev(fs_info, device_path: tgtdev_name,
635 srcdev: src_device, device_out: &tgt_device);
636 if (ret)
637 return ret;
638
639 ret = mark_block_group_to_copy(fs_info, src_dev: src_device);
640 if (ret)
641 return ret;
642
643 down_write(sem: &dev_replace->rwsem);
644 switch (dev_replace->replace_state) {
645 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
646 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
647 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
648 break;
649 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
650 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
651 ASSERT(0);
652 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
653 up_write(sem: &dev_replace->rwsem);
654 goto leave;
655 }
656
657 dev_replace->cont_reading_from_srcdev_mode = read_src;
658 dev_replace->srcdev = src_device;
659 dev_replace->tgtdev = tgt_device;
660
661 btrfs_info_in_rcu(fs_info,
662 "dev_replace from %s (devid %llu) to %s started",
663 btrfs_dev_name(src_device),
664 src_device->devid,
665 btrfs_dev_name(tgt_device));
666
667 /*
668 * from now on, the writes to the srcdev are all duplicated to
669 * go to the tgtdev as well (refer to btrfs_map_block()).
670 */
671 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
672 dev_replace->time_started = ktime_get_real_seconds();
673 dev_replace->cursor_left = 0;
674 dev_replace->committed_cursor_left = 0;
675 dev_replace->cursor_left_last_write_of_item = 0;
676 dev_replace->cursor_right = 0;
677 dev_replace->is_valid = 1;
678 dev_replace->item_needs_writeback = 1;
679 atomic64_set(v: &dev_replace->num_write_errors, i: 0);
680 atomic64_set(v: &dev_replace->num_uncorrectable_read_errors, i: 0);
681 up_write(sem: &dev_replace->rwsem);
682
683 ret = btrfs_sysfs_add_device(device: tgt_device);
684 if (ret)
685 btrfs_err(fs_info, "kobj add dev failed %d", ret);
686
687 btrfs_wait_ordered_roots(fs_info, U64_MAX, range_start: 0, range_len: (u64)-1);
688
689 /*
690 * Commit dev_replace state and reserve 1 item for it.
691 * This is crucial to ensure we won't miss copying extents for new block
692 * groups that are allocated after we started the device replace, and
693 * must be done after setting up the device replace state.
694 */
695 trans = btrfs_start_transaction(root, num_items: 1);
696 if (IS_ERR(ptr: trans)) {
697 ret = PTR_ERR(ptr: trans);
698 down_write(sem: &dev_replace->rwsem);
699 dev_replace->replace_state =
700 BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
701 dev_replace->srcdev = NULL;
702 dev_replace->tgtdev = NULL;
703 up_write(sem: &dev_replace->rwsem);
704 goto leave;
705 }
706
707 ret = btrfs_commit_transaction(trans);
708 WARN_ON(ret);
709
710 /* the disk copy procedure reuses the scrub code */
711 ret = btrfs_scrub_dev(fs_info, devid: src_device->devid, start: 0,
712 end: btrfs_device_get_total_bytes(dev: src_device),
713 progress: &dev_replace->scrub_progress, readonly: 0, is_dev_replace: 1);
714
715 ret = btrfs_dev_replace_finishing(fs_info, scrub_ret: ret);
716 if (ret == -EINPROGRESS)
717 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
718
719 return ret;
720
721leave:
722 btrfs_destroy_dev_replace_tgtdev(tgtdev: tgt_device);
723 return ret;
724}
725
726static int btrfs_check_replace_dev_names(struct btrfs_ioctl_dev_replace_args *args)
727{
728 if (args->start.srcdevid == 0) {
729 if (memchr(p: args->start.srcdev_name, c: 0,
730 size: sizeof(args->start.srcdev_name)) == NULL)
731 return -ENAMETOOLONG;
732 } else {
733 args->start.srcdev_name[0] = 0;
734 }
735
736 if (memchr(p: args->start.tgtdev_name, c: 0,
737 size: sizeof(args->start.tgtdev_name)) == NULL)
738 return -ENAMETOOLONG;
739
740 return 0;
741}
742
743int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
744 struct btrfs_ioctl_dev_replace_args *args)
745{
746 int ret;
747
748 switch (args->start.cont_reading_from_srcdev_mode) {
749 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
750 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
751 break;
752 default:
753 return -EINVAL;
754 }
755 ret = btrfs_check_replace_dev_names(args);
756 if (ret < 0)
757 return ret;
758
759 ret = btrfs_dev_replace_start(fs_info, tgtdev_name: args->start.tgtdev_name,
760 srcdevid: args->start.srcdevid,
761 srcdev_name: args->start.srcdev_name,
762 read_src: args->start.cont_reading_from_srcdev_mode);
763 args->result = ret;
764 /* don't warn if EINPROGRESS, someone else might be running scrub */
765 if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS ||
766 ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR)
767 return 0;
768
769 return ret;
770}
771
772/*
773 * blocked until all in-flight bios operations are finished.
774 */
775static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
776{
777 set_bit(nr: BTRFS_FS_STATE_DEV_REPLACING, addr: &fs_info->fs_state);
778 wait_event(fs_info->dev_replace.replace_wait, !percpu_counter_sum(
779 &fs_info->dev_replace.bio_counter));
780}
781
782/*
783 * we have removed target device, it is safe to allow new bios request.
784 */
785static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
786{
787 clear_bit(nr: BTRFS_FS_STATE_DEV_REPLACING, addr: &fs_info->fs_state);
788 wake_up(&fs_info->dev_replace.replace_wait);
789}
790
791/*
792 * When finishing the device replace, before swapping the source device with the
793 * target device we must update the chunk allocation state in the target device,
794 * as it is empty because replace works by directly copying the chunks and not
795 * through the normal chunk allocation path.
796 */
797static int btrfs_set_target_alloc_state(struct btrfs_device *srcdev,
798 struct btrfs_device *tgtdev)
799{
800 struct extent_state *cached_state = NULL;
801 u64 start = 0;
802 u64 found_start;
803 u64 found_end;
804 int ret = 0;
805
806 lockdep_assert_held(&srcdev->fs_info->chunk_mutex);
807
808 while (find_first_extent_bit(tree: &srcdev->alloc_state, start,
809 start_ret: &found_start, end_ret: &found_end,
810 CHUNK_ALLOCATED, cached_state: &cached_state)) {
811 ret = set_extent_bit(tree: &tgtdev->alloc_state, start: found_start,
812 end: found_end, CHUNK_ALLOCATED, NULL);
813 if (ret)
814 break;
815 start = found_end + 1;
816 }
817
818 free_extent_state(state: cached_state);
819 return ret;
820}
821
822static void btrfs_dev_replace_update_device_in_mapping_tree(
823 struct btrfs_fs_info *fs_info,
824 struct btrfs_device *srcdev,
825 struct btrfs_device *tgtdev)
826{
827 u64 start = 0;
828 int i;
829
830 write_lock(&fs_info->mapping_tree_lock);
831 do {
832 struct btrfs_chunk_map *map;
833
834 map = btrfs_find_chunk_map_nolock(fs_info, logical: start, U64_MAX);
835 if (!map)
836 break;
837 for (i = 0; i < map->num_stripes; i++)
838 if (srcdev == map->stripes[i].dev)
839 map->stripes[i].dev = tgtdev;
840 start = map->start + map->chunk_len;
841 btrfs_free_chunk_map(map);
842 } while (start);
843 write_unlock(&fs_info->mapping_tree_lock);
844}
845
846static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
847 int scrub_ret)
848{
849 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
850 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
851 struct btrfs_device *tgt_device;
852 struct btrfs_device *src_device;
853 struct btrfs_root *root = fs_info->tree_root;
854 u8 uuid_tmp[BTRFS_UUID_SIZE];
855 struct btrfs_trans_handle *trans;
856 int ret = 0;
857
858 /* don't allow cancel or unmount to disturb the finishing procedure */
859 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
860
861 down_read(sem: &dev_replace->rwsem);
862 /* was the operation canceled, or is it finished? */
863 if (dev_replace->replace_state !=
864 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
865 up_read(sem: &dev_replace->rwsem);
866 mutex_unlock(lock: &dev_replace->lock_finishing_cancel_unmount);
867 return 0;
868 }
869
870 tgt_device = dev_replace->tgtdev;
871 src_device = dev_replace->srcdev;
872 up_read(sem: &dev_replace->rwsem);
873
874 /*
875 * flush all outstanding I/O and inode extent mappings before the
876 * copy operation is declared as being finished
877 */
878 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, in_reclaim_context: false);
879 if (ret) {
880 mutex_unlock(lock: &dev_replace->lock_finishing_cancel_unmount);
881 return ret;
882 }
883 btrfs_wait_ordered_roots(fs_info, U64_MAX, range_start: 0, range_len: (u64)-1);
884
885 /*
886 * We have to use this loop approach because at this point src_device
887 * has to be available for transaction commit to complete, yet new
888 * chunks shouldn't be allocated on the device.
889 */
890 while (1) {
891 trans = btrfs_start_transaction(root, num_items: 0);
892 if (IS_ERR(ptr: trans)) {
893 mutex_unlock(lock: &dev_replace->lock_finishing_cancel_unmount);
894 return PTR_ERR(ptr: trans);
895 }
896 ret = btrfs_commit_transaction(trans);
897 WARN_ON(ret);
898
899 /* Prevent write_all_supers() during the finishing procedure */
900 mutex_lock(&fs_devices->device_list_mutex);
901 /* Prevent new chunks being allocated on the source device */
902 mutex_lock(&fs_info->chunk_mutex);
903
904 if (!list_empty(head: &src_device->post_commit_list)) {
905 mutex_unlock(lock: &fs_devices->device_list_mutex);
906 mutex_unlock(lock: &fs_info->chunk_mutex);
907 } else {
908 break;
909 }
910 }
911
912 down_write(sem: &dev_replace->rwsem);
913 dev_replace->replace_state =
914 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
915 : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
916 dev_replace->tgtdev = NULL;
917 dev_replace->srcdev = NULL;
918 dev_replace->time_stopped = ktime_get_real_seconds();
919 dev_replace->item_needs_writeback = 1;
920
921 /*
922 * Update allocation state in the new device and replace the old device
923 * with the new one in the mapping tree.
924 */
925 if (!scrub_ret) {
926 scrub_ret = btrfs_set_target_alloc_state(srcdev: src_device, tgtdev: tgt_device);
927 if (scrub_ret)
928 goto error;
929 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
930 srcdev: src_device,
931 tgtdev: tgt_device);
932 } else {
933 if (scrub_ret != -ECANCELED)
934 btrfs_err_in_rcu(fs_info,
935 "btrfs_scrub_dev(%s, %llu, %s) failed %d",
936 btrfs_dev_name(src_device),
937 src_device->devid,
938 btrfs_dev_name(tgt_device), scrub_ret);
939error:
940 up_write(sem: &dev_replace->rwsem);
941 mutex_unlock(lock: &fs_info->chunk_mutex);
942 mutex_unlock(lock: &fs_devices->device_list_mutex);
943 btrfs_rm_dev_replace_blocked(fs_info);
944 if (tgt_device)
945 btrfs_destroy_dev_replace_tgtdev(tgtdev: tgt_device);
946 btrfs_rm_dev_replace_unblocked(fs_info);
947 mutex_unlock(lock: &dev_replace->lock_finishing_cancel_unmount);
948
949 return scrub_ret;
950 }
951
952 btrfs_info_in_rcu(fs_info,
953 "dev_replace from %s (devid %llu) to %s finished",
954 btrfs_dev_name(src_device),
955 src_device->devid,
956 btrfs_dev_name(tgt_device));
957 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, addr: &tgt_device->dev_state);
958 tgt_device->devid = src_device->devid;
959 src_device->devid = BTRFS_DEV_REPLACE_DEVID;
960 memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
961 memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
962 memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
963 btrfs_device_set_total_bytes(dev: tgt_device, size: src_device->total_bytes);
964 btrfs_device_set_disk_total_bytes(dev: tgt_device,
965 size: src_device->disk_total_bytes);
966 btrfs_device_set_bytes_used(dev: tgt_device, size: src_device->bytes_used);
967 tgt_device->commit_bytes_used = src_device->bytes_used;
968
969 btrfs_assign_next_active_device(device: src_device, this_dev: tgt_device);
970
971 list_add(new: &tgt_device->dev_alloc_list, head: &fs_devices->alloc_list);
972 fs_devices->rw_devices++;
973
974 up_write(sem: &dev_replace->rwsem);
975 btrfs_rm_dev_replace_blocked(fs_info);
976
977 btrfs_rm_dev_replace_remove_srcdev(srcdev: src_device);
978
979 btrfs_rm_dev_replace_unblocked(fs_info);
980
981 /*
982 * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will
983 * update on-disk dev stats value during commit transaction
984 */
985 atomic_inc(v: &tgt_device->dev_stats_ccnt);
986
987 /*
988 * this is again a consistent state where no dev_replace procedure
989 * is running, the target device is part of the filesystem, the
990 * source device is not part of the filesystem anymore and its 1st
991 * superblock is scratched out so that it is no longer marked to
992 * belong to this filesystem.
993 */
994 mutex_unlock(lock: &fs_info->chunk_mutex);
995 mutex_unlock(lock: &fs_devices->device_list_mutex);
996
997 /* replace the sysfs entry */
998 btrfs_sysfs_remove_device(device: src_device);
999 btrfs_sysfs_update_devid(device: tgt_device);
1000 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &src_device->dev_state))
1001 btrfs_scratch_superblocks(fs_info, device: src_device);
1002
1003 /* write back the superblocks */
1004 trans = btrfs_start_transaction(root, num_items: 0);
1005 if (!IS_ERR(ptr: trans))
1006 btrfs_commit_transaction(trans);
1007
1008 mutex_unlock(lock: &dev_replace->lock_finishing_cancel_unmount);
1009
1010 btrfs_rm_dev_replace_free_srcdev(srcdev: src_device);
1011
1012 return 0;
1013}
1014
1015/*
1016 * Read progress of device replace status according to the state and last
1017 * stored position. The value format is the same as for
1018 * btrfs_dev_replace::progress_1000
1019 */
1020static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info)
1021{
1022 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1023 u64 ret = 0;
1024
1025 switch (dev_replace->replace_state) {
1026 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1027 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1028 ret = 0;
1029 break;
1030 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1031 ret = 1000;
1032 break;
1033 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1034 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1035 ret = div64_u64(dividend: dev_replace->cursor_left,
1036 divisor: div_u64(dividend: btrfs_device_get_total_bytes(
1037 dev: dev_replace->srcdev), divisor: 1000));
1038 break;
1039 }
1040
1041 return ret;
1042}
1043
1044void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
1045 struct btrfs_ioctl_dev_replace_args *args)
1046{
1047 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1048
1049 down_read(sem: &dev_replace->rwsem);
1050 /* even if !dev_replace_is_valid, the values are good enough for
1051 * the replace_status ioctl */
1052 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
1053 args->status.replace_state = dev_replace->replace_state;
1054 args->status.time_started = dev_replace->time_started;
1055 args->status.time_stopped = dev_replace->time_stopped;
1056 args->status.num_write_errors =
1057 atomic64_read(v: &dev_replace->num_write_errors);
1058 args->status.num_uncorrectable_read_errors =
1059 atomic64_read(v: &dev_replace->num_uncorrectable_read_errors);
1060 args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
1061 up_read(sem: &dev_replace->rwsem);
1062}
1063
1064int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
1065{
1066 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1067 struct btrfs_device *tgt_device = NULL;
1068 struct btrfs_device *src_device = NULL;
1069 struct btrfs_trans_handle *trans;
1070 struct btrfs_root *root = fs_info->tree_root;
1071 int result;
1072 int ret;
1073
1074 if (sb_rdonly(sb: fs_info->sb))
1075 return -EROFS;
1076
1077 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
1078 down_write(sem: &dev_replace->rwsem);
1079 switch (dev_replace->replace_state) {
1080 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1081 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1082 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1083 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
1084 up_write(sem: &dev_replace->rwsem);
1085 break;
1086 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1087 tgt_device = dev_replace->tgtdev;
1088 src_device = dev_replace->srcdev;
1089 up_write(sem: &dev_replace->rwsem);
1090 ret = btrfs_scrub_cancel(info: fs_info);
1091 if (ret < 0) {
1092 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
1093 } else {
1094 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
1095 /*
1096 * btrfs_dev_replace_finishing() will handle the
1097 * cleanup part
1098 */
1099 btrfs_info_in_rcu(fs_info,
1100 "dev_replace from %s (devid %llu) to %s canceled",
1101 btrfs_dev_name(src_device), src_device->devid,
1102 btrfs_dev_name(tgt_device));
1103 }
1104 break;
1105 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1106 /*
1107 * Scrub doing the replace isn't running so we need to do the
1108 * cleanup step of btrfs_dev_replace_finishing() here
1109 */
1110 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
1111 tgt_device = dev_replace->tgtdev;
1112 src_device = dev_replace->srcdev;
1113 dev_replace->tgtdev = NULL;
1114 dev_replace->srcdev = NULL;
1115 dev_replace->replace_state =
1116 BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
1117 dev_replace->time_stopped = ktime_get_real_seconds();
1118 dev_replace->item_needs_writeback = 1;
1119
1120 up_write(sem: &dev_replace->rwsem);
1121
1122 /* Scrub for replace must not be running in suspended state */
1123 btrfs_scrub_cancel(info: fs_info);
1124
1125 trans = btrfs_start_transaction(root, num_items: 0);
1126 if (IS_ERR(ptr: trans)) {
1127 mutex_unlock(lock: &dev_replace->lock_finishing_cancel_unmount);
1128 return PTR_ERR(ptr: trans);
1129 }
1130 ret = btrfs_commit_transaction(trans);
1131 WARN_ON(ret);
1132
1133 btrfs_info_in_rcu(fs_info,
1134 "suspended dev_replace from %s (devid %llu) to %s canceled",
1135 btrfs_dev_name(src_device), src_device->devid,
1136 btrfs_dev_name(tgt_device));
1137
1138 if (tgt_device)
1139 btrfs_destroy_dev_replace_tgtdev(tgtdev: tgt_device);
1140 break;
1141 default:
1142 up_write(sem: &dev_replace->rwsem);
1143 result = -EINVAL;
1144 }
1145
1146 mutex_unlock(lock: &dev_replace->lock_finishing_cancel_unmount);
1147 return result;
1148}
1149
1150void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
1151{
1152 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1153
1154 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
1155 down_write(sem: &dev_replace->rwsem);
1156
1157 switch (dev_replace->replace_state) {
1158 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1159 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1160 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1161 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1162 break;
1163 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1164 dev_replace->replace_state =
1165 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
1166 dev_replace->time_stopped = ktime_get_real_seconds();
1167 dev_replace->item_needs_writeback = 1;
1168 btrfs_info(fs_info, "suspending dev_replace for unmount");
1169 break;
1170 }
1171
1172 up_write(sem: &dev_replace->rwsem);
1173 mutex_unlock(lock: &dev_replace->lock_finishing_cancel_unmount);
1174}
1175
1176/* resume dev_replace procedure that was interrupted by unmount */
1177int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
1178{
1179 struct task_struct *task;
1180 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1181
1182 down_write(sem: &dev_replace->rwsem);
1183
1184 switch (dev_replace->replace_state) {
1185 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1186 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1187 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1188 up_write(sem: &dev_replace->rwsem);
1189 return 0;
1190 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1191 break;
1192 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1193 dev_replace->replace_state =
1194 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
1195 break;
1196 }
1197 if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
1198 btrfs_info(fs_info,
1199 "cannot continue dev_replace, tgtdev is missing");
1200 btrfs_info(fs_info,
1201 "you may cancel the operation after 'mount -o degraded'");
1202 dev_replace->replace_state =
1203 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
1204 up_write(sem: &dev_replace->rwsem);
1205 return 0;
1206 }
1207 up_write(sem: &dev_replace->rwsem);
1208
1209 /*
1210 * This could collide with a paused balance, but the exclusive op logic
1211 * should never allow both to start and pause. We don't want to allow
1212 * dev-replace to start anyway.
1213 */
1214 if (!btrfs_exclop_start(fs_info, type: BTRFS_EXCLOP_DEV_REPLACE)) {
1215 down_write(sem: &dev_replace->rwsem);
1216 dev_replace->replace_state =
1217 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
1218 up_write(sem: &dev_replace->rwsem);
1219 btrfs_info(fs_info,
1220 "cannot resume dev-replace, other exclusive operation running");
1221 return 0;
1222 }
1223
1224 task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
1225 return PTR_ERR_OR_ZERO(ptr: task);
1226}
1227
1228static int btrfs_dev_replace_kthread(void *data)
1229{
1230 struct btrfs_fs_info *fs_info = data;
1231 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1232 u64 progress;
1233 int ret;
1234
1235 progress = btrfs_dev_replace_progress(fs_info);
1236 progress = div_u64(dividend: progress, divisor: 10);
1237 btrfs_info_in_rcu(fs_info,
1238 "continuing dev_replace from %s (devid %llu) to target %s @%u%%",
1239 btrfs_dev_name(dev_replace->srcdev),
1240 dev_replace->srcdev->devid,
1241 btrfs_dev_name(dev_replace->tgtdev),
1242 (unsigned int)progress);
1243
1244 ret = btrfs_scrub_dev(fs_info, devid: dev_replace->srcdev->devid,
1245 start: dev_replace->committed_cursor_left,
1246 end: btrfs_device_get_total_bytes(dev: dev_replace->srcdev),
1247 progress: &dev_replace->scrub_progress, readonly: 0, is_dev_replace: 1);
1248 ret = btrfs_dev_replace_finishing(fs_info, scrub_ret: ret);
1249 WARN_ON(ret && ret != -ECANCELED);
1250
1251 btrfs_exclop_finish(fs_info);
1252 return 0;
1253}
1254
1255int __pure btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
1256{
1257 if (!dev_replace->is_valid)
1258 return 0;
1259
1260 switch (dev_replace->replace_state) {
1261 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1262 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1263 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1264 return 0;
1265 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1266 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1267 /*
1268 * return true even if tgtdev is missing (this is
1269 * something that can happen if the dev_replace
1270 * procedure is suspended by an umount and then
1271 * the tgtdev is missing (or "btrfs dev scan") was
1272 * not called and the filesystem is remounted
1273 * in degraded state. This does not stop the
1274 * dev_replace procedure. It needs to be canceled
1275 * manually if the cancellation is wanted.
1276 */
1277 break;
1278 }
1279 return 1;
1280}
1281
1282void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
1283{
1284 percpu_counter_sub(fbc: &fs_info->dev_replace.bio_counter, amount);
1285 cond_wake_up_nomb(wq: &fs_info->dev_replace.replace_wait);
1286}
1287
1288void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
1289{
1290 while (1) {
1291 percpu_counter_inc(fbc: &fs_info->dev_replace.bio_counter);
1292 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1293 &fs_info->fs_state)))
1294 break;
1295
1296 btrfs_bio_counter_dec(fs_info);
1297 wait_event(fs_info->dev_replace.replace_wait,
1298 !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1299 &fs_info->fs_state));
1300 }
1301}
1302

source code of linux/fs/btrfs/dev-replace.c