Loading...
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 "extent_map.h"
15#include "disk-io.h"
16#include "transaction.h"
17#include "print-tree.h"
18#include "volumes.h"
19#include "async-thread.h"
20#include "dev-replace.h"
21#include "sysfs.h"
22#include "zoned.h"
23#include "block-group.h"
24#include "fs.h"
25#include "accessors.h"
26#include "scrub.h"
27
28/*
29 * Device replace overview
30 *
31 * [Objective]
32 * To copy all extents (both new and on-disk) from source device to target
33 * device, while still keeping the filesystem read-write.
34 *
35 * [Method]
36 * There are two main methods involved:
37 *
38 * - Write duplication
39 *
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.
42 *
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
47 *
48 * - Copy existing extents
49 *
50 * This happens by re-using scrub facility, as scrub also iterates through
51 * existing extents from commit root.
52 *
53 * Location: scrub_write_block_to_dev_replace() from
54 * scrub_block_complete()
55 * Content: Data/meta from commit root.
56 *
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
59 * for NOCOW writes.
60 *
61 * After replace is done, the finishing part is done by swapping the target and
62 * source devices.
63 *
64 * Location: btrfs_dev_replace_update_device_in_mapping_tree() from
65 * btrfs_dev_replace_finishing()
66 */
67
68static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
69 int scrub_ret);
70static int btrfs_dev_replace_kthread(void *data);
71
72int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
73{
74 struct btrfs_dev_lookup_args args = { .devid = BTRFS_DEV_REPLACE_DEVID };
75 struct btrfs_key key;
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;
79 int slot;
80 int ret = 0;
81 struct btrfs_path *path = NULL;
82 int item_size;
83 struct btrfs_dev_replace_item *ptr;
84 u64 src_devid;
85
86 if (!dev_root)
87 return 0;
88
89 path = btrfs_alloc_path();
90 if (!path) {
91 ret = -ENOMEM;
92 goto out;
93 }
94
95 key.objectid = 0;
96 key.type = BTRFS_DEV_REPLACE_KEY;
97 key.offset = 0;
98 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
99 if (ret) {
100no_valid_dev_replace_entry_found:
101 /*
102 * We don't have a replace item or it's corrupted. If there is
103 * a replace target, fail the mount.
104 */
105 if (btrfs_find_device(fs_info->fs_devices, &args)) {
106 btrfs_err(fs_info,
107 "found replace target device without a valid replace item");
108 ret = -EUCLEAN;
109 goto out;
110 }
111 ret = 0;
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;
128 goto out;
129 }
130 slot = path->slots[0];
131 eb = path->nodes[0];
132 item_size = btrfs_item_size(eb, slot);
133 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
134
135 if (item_size != sizeof(struct btrfs_dev_replace_item)) {
136 btrfs_warn(fs_info,
137 "dev_replace entry found has unexpected size, ignore entry");
138 goto no_valid_dev_replace_entry_found;
139 }
140
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;
157
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:
163 /*
164 * We don't have an active replace item but if there is a
165 * replace target, fail the mount.
166 */
167 if (btrfs_find_device(fs_info->fs_devices, &args)) {
168 btrfs_err(fs_info,
169"replace without active item, run 'device scan --forget' on the target device");
170 ret = -EUCLEAN;
171 } else {
172 dev_replace->srcdev = NULL;
173 dev_replace->tgtdev = NULL;
174 }
175 break;
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);
181
182 /*
183 * allow 'btrfs dev replace_cancel' if src/tgt device is
184 * missing
185 */
186 if (!dev_replace->srcdev &&
187 !btrfs_test_opt(fs_info, DEGRADED)) {
188 ret = -EIO;
189 btrfs_warn(fs_info,
190 "cannot mount because device replace operation is ongoing and");
191 btrfs_warn(fs_info,
192 "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
193 src_devid);
194 }
195 if (!dev_replace->tgtdev &&
196 !btrfs_test_opt(fs_info, DEGRADED)) {
197 ret = -EIO;
198 btrfs_warn(fs_info,
199 "cannot mount because device replace operation is ongoing and");
200 btrfs_warn(fs_info,
201 "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
202 BTRFS_DEV_REPLACE_DEVID);
203 }
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;
216 }
217 set_bit(BTRFS_DEV_STATE_REPLACE_TGT,
218 &dev_replace->tgtdev->dev_state);
219
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);
227 }
228 break;
229 }
230
231out:
232 btrfs_free_path(path);
233 return ret;
234}
235
236/*
237 * Initialize a new device for device replace target from a given source dev
238 * and path.
239 *
240 * Return 0 and new device in @device_out, otherwise return < 0
241 */
242static 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)
246{
247 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
248 struct btrfs_device *device;
249 struct bdev_handle *bdev_handle;
250 struct block_device *bdev;
251 u64 devid = BTRFS_DEV_REPLACE_DEVID;
252 int ret = 0;
253
254 *device_out = NULL;
255 if (srcdev->fs_devices->seeding) {
256 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
257 return -EINVAL;
258 }
259
260 bdev_handle = bdev_open_by_path(device_path, BLK_OPEN_WRITE,
261 fs_info->bdev_holder, NULL);
262 if (IS_ERR(bdev_handle)) {
263 btrfs_err(fs_info, "target device %s is invalid!", device_path);
264 return PTR_ERR(bdev_handle);
265 }
266 bdev = bdev_handle->bdev;
267
268 if (!btrfs_check_device_zone_type(fs_info, bdev)) {
269 btrfs_err(fs_info,
270 "dev-replace: zoned type of target device mismatch with filesystem");
271 ret = -EINVAL;
272 goto error;
273 }
274
275 sync_blockdev(bdev);
276
277 list_for_each_entry(device, &fs_devices->devices, dev_list) {
278 if (device->bdev == bdev) {
279 btrfs_err(fs_info,
280 "target device is in the filesystem!");
281 ret = -EEXIST;
282 goto error;
283 }
284 }
285
286
287 if (bdev_nr_bytes(bdev) < btrfs_device_get_total_bytes(srcdev)) {
288 btrfs_err(fs_info,
289 "target device is smaller than source device!");
290 ret = -EINVAL;
291 goto error;
292 }
293
294
295 device = btrfs_alloc_device(NULL, &devid, NULL, device_path);
296 if (IS_ERR(device)) {
297 ret = PTR_ERR(device);
298 goto error;
299 }
300
301 ret = lookup_bdev(device_path, &device->devt);
302 if (ret)
303 goto error;
304
305 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
306 device->generation = 0;
307 device->io_width = fs_info->sectorsize;
308 device->io_align = fs_info->sectorsize;
309 device->sector_size = fs_info->sectorsize;
310 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
311 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
312 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
313 device->commit_total_bytes = srcdev->commit_total_bytes;
314 device->commit_bytes_used = device->bytes_used;
315 device->fs_info = fs_info;
316 device->bdev = bdev;
317 device->bdev_handle = bdev_handle;
318 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
319 set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
320 device->dev_stats_valid = 1;
321 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
322 device->fs_devices = fs_devices;
323
324 ret = btrfs_get_dev_zone_info(device, false);
325 if (ret)
326 goto error;
327
328 mutex_lock(&fs_devices->device_list_mutex);
329 list_add(&device->dev_list, &fs_devices->devices);
330 fs_devices->num_devices++;
331 fs_devices->open_devices++;
332 mutex_unlock(&fs_devices->device_list_mutex);
333
334 *device_out = device;
335 return 0;
336
337error:
338 bdev_release(bdev_handle);
339 return ret;
340}
341
342/*
343 * called from commit_transaction. Writes changed device replace state to
344 * disk.
345 */
346int btrfs_run_dev_replace(struct btrfs_trans_handle *trans)
347{
348 struct btrfs_fs_info *fs_info = trans->fs_info;
349 int ret;
350 struct btrfs_root *dev_root = fs_info->dev_root;
351 struct btrfs_path *path;
352 struct btrfs_key key;
353 struct extent_buffer *eb;
354 struct btrfs_dev_replace_item *ptr;
355 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
356
357 down_read(&dev_replace->rwsem);
358 if (!dev_replace->is_valid ||
359 !dev_replace->item_needs_writeback) {
360 up_read(&dev_replace->rwsem);
361 return 0;
362 }
363 up_read(&dev_replace->rwsem);
364
365 key.objectid = 0;
366 key.type = BTRFS_DEV_REPLACE_KEY;
367 key.offset = 0;
368
369 path = btrfs_alloc_path();
370 if (!path) {
371 ret = -ENOMEM;
372 goto out;
373 }
374 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
375 if (ret < 0) {
376 btrfs_warn(fs_info,
377 "error %d while searching for dev_replace item!",
378 ret);
379 goto out;
380 }
381
382 if (ret == 0 &&
383 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
384 /*
385 * need to delete old one and insert a new one.
386 * Since no attempt is made to recover any old state, if the
387 * dev_replace state is 'running', the data on the target
388 * drive is lost.
389 * It would be possible to recover the state: just make sure
390 * that the beginning of the item is never changed and always
391 * contains all the essential information. Then read this
392 * minimal set of information and use it as a base for the
393 * new state.
394 */
395 ret = btrfs_del_item(trans, dev_root, path);
396 if (ret != 0) {
397 btrfs_warn(fs_info,
398 "delete too small dev_replace item failed %d!",
399 ret);
400 goto out;
401 }
402 ret = 1;
403 }
404
405 if (ret == 1) {
406 /* need to insert a new item */
407 btrfs_release_path(path);
408 ret = btrfs_insert_empty_item(trans, dev_root, path,
409 &key, sizeof(*ptr));
410 if (ret < 0) {
411 btrfs_warn(fs_info,
412 "insert dev_replace item failed %d!", ret);
413 goto out;
414 }
415 }
416
417 eb = path->nodes[0];
418 ptr = btrfs_item_ptr(eb, path->slots[0],
419 struct btrfs_dev_replace_item);
420
421 down_write(&dev_replace->rwsem);
422 if (dev_replace->srcdev)
423 btrfs_set_dev_replace_src_devid(eb, ptr,
424 dev_replace->srcdev->devid);
425 else
426 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
427 btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
428 dev_replace->cont_reading_from_srcdev_mode);
429 btrfs_set_dev_replace_replace_state(eb, ptr,
430 dev_replace->replace_state);
431 btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
432 btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
433 btrfs_set_dev_replace_num_write_errors(eb, ptr,
434 atomic64_read(&dev_replace->num_write_errors));
435 btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
436 atomic64_read(&dev_replace->num_uncorrectable_read_errors));
437 dev_replace->cursor_left_last_write_of_item =
438 dev_replace->cursor_left;
439 btrfs_set_dev_replace_cursor_left(eb, ptr,
440 dev_replace->cursor_left_last_write_of_item);
441 btrfs_set_dev_replace_cursor_right(eb, ptr,
442 dev_replace->cursor_right);
443 dev_replace->item_needs_writeback = 0;
444 up_write(&dev_replace->rwsem);
445
446 btrfs_mark_buffer_dirty(trans, eb);
447
448out:
449 btrfs_free_path(path);
450
451 return ret;
452}
453
454static int mark_block_group_to_copy(struct btrfs_fs_info *fs_info,
455 struct btrfs_device *src_dev)
456{
457 struct btrfs_path *path;
458 struct btrfs_key key;
459 struct btrfs_key found_key;
460 struct btrfs_root *root = fs_info->dev_root;
461 struct btrfs_dev_extent *dev_extent = NULL;
462 struct btrfs_block_group *cache;
463 struct btrfs_trans_handle *trans;
464 int iter_ret = 0;
465 int ret = 0;
466 u64 chunk_offset;
467
468 /* Do not use "to_copy" on non zoned filesystem for now */
469 if (!btrfs_is_zoned(fs_info))
470 return 0;
471
472 mutex_lock(&fs_info->chunk_mutex);
473
474 /* Ensure we don't have pending new block group */
475 spin_lock(&fs_info->trans_lock);
476 while (fs_info->running_transaction &&
477 !list_empty(&fs_info->running_transaction->dev_update_list)) {
478 spin_unlock(&fs_info->trans_lock);
479 mutex_unlock(&fs_info->chunk_mutex);
480 trans = btrfs_attach_transaction(root);
481 if (IS_ERR(trans)) {
482 ret = PTR_ERR(trans);
483 mutex_lock(&fs_info->chunk_mutex);
484 if (ret == -ENOENT) {
485 spin_lock(&fs_info->trans_lock);
486 continue;
487 } else {
488 goto unlock;
489 }
490 }
491
492 ret = btrfs_commit_transaction(trans);
493 mutex_lock(&fs_info->chunk_mutex);
494 if (ret)
495 goto unlock;
496
497 spin_lock(&fs_info->trans_lock);
498 }
499 spin_unlock(&fs_info->trans_lock);
500
501 path = btrfs_alloc_path();
502 if (!path) {
503 ret = -ENOMEM;
504 goto unlock;
505 }
506
507 path->reada = READA_FORWARD;
508 path->search_commit_root = 1;
509 path->skip_locking = 1;
510
511 key.objectid = src_dev->devid;
512 key.type = BTRFS_DEV_EXTENT_KEY;
513 key.offset = 0;
514
515 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) {
516 struct extent_buffer *leaf = path->nodes[0];
517
518 if (found_key.objectid != src_dev->devid)
519 break;
520
521 if (found_key.type != BTRFS_DEV_EXTENT_KEY)
522 break;
523
524 if (found_key.offset < key.offset)
525 break;
526
527 dev_extent = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent);
528
529 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dev_extent);
530
531 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
532 if (!cache)
533 continue;
534
535 set_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags);
536 btrfs_put_block_group(cache);
537 }
538 if (iter_ret < 0)
539 ret = iter_ret;
540
541 btrfs_free_path(path);
542unlock:
543 mutex_unlock(&fs_info->chunk_mutex);
544
545 return ret;
546}
547
548bool btrfs_finish_block_group_to_copy(struct btrfs_device *srcdev,
549 struct btrfs_block_group *cache,
550 u64 physical)
551{
552 struct btrfs_fs_info *fs_info = cache->fs_info;
553 struct btrfs_chunk_map *map;
554 u64 chunk_offset = cache->start;
555 int num_extents, cur_extent;
556 int i;
557
558 /* Do not use "to_copy" on non zoned filesystem for now */
559 if (!btrfs_is_zoned(fs_info))
560 return true;
561
562 spin_lock(&cache->lock);
563 if (test_bit(BLOCK_GROUP_FLAG_REMOVED, &cache->runtime_flags)) {
564 spin_unlock(&cache->lock);
565 return true;
566 }
567 spin_unlock(&cache->lock);
568
569 map = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
570 ASSERT(!IS_ERR(map));
571
572 num_extents = 0;
573 cur_extent = 0;
574 for (i = 0; i < map->num_stripes; i++) {
575 /* We have more device extent to copy */
576 if (srcdev != map->stripes[i].dev)
577 continue;
578
579 num_extents++;
580 if (physical == map->stripes[i].physical)
581 cur_extent = i;
582 }
583
584 btrfs_free_chunk_map(map);
585
586 if (num_extents > 1 && cur_extent < num_extents - 1) {
587 /*
588 * Has more stripes on this device. Keep this block group
589 * readonly until we finish all the stripes.
590 */
591 return false;
592 }
593
594 /* Last stripe on this device */
595 clear_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags);
596
597 return true;
598}
599
600static int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
601 const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
602 int read_src)
603{
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;
607 int ret;
608 struct btrfs_device *tgt_device = NULL;
609 struct btrfs_device *src_device = NULL;
610
611 src_device = btrfs_find_device_by_devspec(fs_info, srcdevid,
612 srcdev_name);
613 if (IS_ERR(src_device))
614 return PTR_ERR(src_device);
615
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);
620 return -ETXTBSY;
621 }
622
623 /*
624 * Here we commit the transaction to make sure commit_total_bytes
625 * of all the devices are updated.
626 */
627 trans = btrfs_attach_transaction(root);
628 if (!IS_ERR(trans)) {
629 ret = btrfs_commit_transaction(trans);
630 if (ret)
631 return ret;
632 } else if (PTR_ERR(trans) != -ENOENT) {
633 return PTR_ERR(trans);
634 }
635
636 ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
637 src_device, &tgt_device);
638 if (ret)
639 return ret;
640
641 ret = mark_block_group_to_copy(fs_info, src_device);
642 if (ret)
643 return ret;
644
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:
650 break;
651 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
652 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
653 ASSERT(0);
654 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
655 up_write(&dev_replace->rwsem);
656 goto leave;
657 }
658
659 dev_replace->cont_reading_from_srcdev_mode = read_src;
660 dev_replace->srcdev = src_device;
661 dev_replace->tgtdev = tgt_device;
662
663 btrfs_info_in_rcu(fs_info,
664 "dev_replace from %s (devid %llu) to %s started",
665 btrfs_dev_name(src_device),
666 src_device->devid,
667 btrfs_dev_name(tgt_device));
668
669 /*
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()).
672 */
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);
684
685 ret = btrfs_sysfs_add_device(tgt_device);
686 if (ret)
687 btrfs_err(fs_info, "kobj add dev failed %d", ret);
688
689 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
690
691 /*
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.
696 */
697 trans = btrfs_start_transaction(root, 1);
698 if (IS_ERR(trans)) {
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);
706 goto leave;
707 }
708
709 ret = btrfs_commit_transaction(trans);
710 WARN_ON(ret);
711
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);
716
717 ret = btrfs_dev_replace_finishing(fs_info, ret);
718 if (ret == -EINPROGRESS)
719 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
720
721 return ret;
722
723leave:
724 btrfs_destroy_dev_replace_tgtdev(tgt_device);
725 return ret;
726}
727
728static int btrfs_check_replace_dev_names(struct btrfs_ioctl_dev_replace_args *args)
729{
730 if (args->start.srcdevid == 0) {
731 if (memchr(args->start.srcdev_name, 0,
732 sizeof(args->start.srcdev_name)) == NULL)
733 return -ENAMETOOLONG;
734 } else {
735 args->start.srcdev_name[0] = 0;
736 }
737
738 if (memchr(args->start.tgtdev_name, 0,
739 sizeof(args->start.tgtdev_name)) == NULL)
740 return -ENAMETOOLONG;
741
742 return 0;
743}
744
745int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
746 struct btrfs_ioctl_dev_replace_args *args)
747{
748 int ret;
749
750 switch (args->start.cont_reading_from_srcdev_mode) {
751 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
752 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
753 break;
754 default:
755 return -EINVAL;
756 }
757 ret = btrfs_check_replace_dev_names(args);
758 if (ret < 0)
759 return ret;
760
761 ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name,
762 args->start.srcdevid,
763 args->start.srcdev_name,
764 args->start.cont_reading_from_srcdev_mode);
765 args->result = ret;
766 /* don't warn if EINPROGRESS, someone else might be running scrub */
767 if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS ||
768 ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR)
769 return 0;
770
771 return ret;
772}
773
774/*
775 * blocked until all in-flight bios operations are finished.
776 */
777static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
778{
779 set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
780 wait_event(fs_info->dev_replace.replace_wait, !percpu_counter_sum(
781 &fs_info->dev_replace.bio_counter));
782}
783
784/*
785 * we have removed target device, it is safe to allow new bios request.
786 */
787static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
788{
789 clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
790 wake_up(&fs_info->dev_replace.replace_wait);
791}
792
793/*
794 * When finishing the device replace, before swapping the source device with the
795 * target device we must update the chunk allocation state in the target device,
796 * as it is empty because replace works by directly copying the chunks and not
797 * through the normal chunk allocation path.
798 */
799static int btrfs_set_target_alloc_state(struct btrfs_device *srcdev,
800 struct btrfs_device *tgtdev)
801{
802 struct extent_state *cached_state = NULL;
803 u64 start = 0;
804 u64 found_start;
805 u64 found_end;
806 int ret = 0;
807
808 lockdep_assert_held(&srcdev->fs_info->chunk_mutex);
809
810 while (find_first_extent_bit(&srcdev->alloc_state, start,
811 &found_start, &found_end,
812 CHUNK_ALLOCATED, &cached_state)) {
813 ret = set_extent_bit(&tgtdev->alloc_state, found_start,
814 found_end, CHUNK_ALLOCATED, NULL);
815 if (ret)
816 break;
817 start = found_end + 1;
818 }
819
820 free_extent_state(cached_state);
821 return ret;
822}
823
824static void btrfs_dev_replace_update_device_in_mapping_tree(
825 struct btrfs_fs_info *fs_info,
826 struct btrfs_device *srcdev,
827 struct btrfs_device *tgtdev)
828{
829 u64 start = 0;
830 int i;
831
832 write_lock(&fs_info->mapping_tree_lock);
833 do {
834 struct btrfs_chunk_map *map;
835
836 map = btrfs_find_chunk_map_nolock(fs_info, start, U64_MAX);
837 if (!map)
838 break;
839 for (i = 0; i < map->num_stripes; i++)
840 if (srcdev == map->stripes[i].dev)
841 map->stripes[i].dev = tgtdev;
842 start = map->start + map->chunk_len;
843 btrfs_free_chunk_map(map);
844 } while (start);
845 write_unlock(&fs_info->mapping_tree_lock);
846}
847
848static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
849 int scrub_ret)
850{
851 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
852 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
853 struct btrfs_device *tgt_device;
854 struct btrfs_device *src_device;
855 struct btrfs_root *root = fs_info->tree_root;
856 u8 uuid_tmp[BTRFS_UUID_SIZE];
857 struct btrfs_trans_handle *trans;
858 int ret = 0;
859
860 /* don't allow cancel or unmount to disturb the finishing procedure */
861 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
862
863 down_read(&dev_replace->rwsem);
864 /* was the operation canceled, or is it finished? */
865 if (dev_replace->replace_state !=
866 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
867 up_read(&dev_replace->rwsem);
868 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
869 return 0;
870 }
871
872 tgt_device = dev_replace->tgtdev;
873 src_device = dev_replace->srcdev;
874 up_read(&dev_replace->rwsem);
875
876 /*
877 * flush all outstanding I/O and inode extent mappings before the
878 * copy operation is declared as being finished
879 */
880 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
881 if (ret) {
882 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
883 return ret;
884 }
885 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
886
887 /*
888 * We have to use this loop approach because at this point src_device
889 * has to be available for transaction commit to complete, yet new
890 * chunks shouldn't be allocated on the device.
891 */
892 while (1) {
893 trans = btrfs_start_transaction(root, 0);
894 if (IS_ERR(trans)) {
895 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
896 return PTR_ERR(trans);
897 }
898 ret = btrfs_commit_transaction(trans);
899 WARN_ON(ret);
900
901 /* Prevent write_all_supers() during the finishing procedure */
902 mutex_lock(&fs_devices->device_list_mutex);
903 /* Prevent new chunks being allocated on the source device */
904 mutex_lock(&fs_info->chunk_mutex);
905
906 if (!list_empty(&src_device->post_commit_list)) {
907 mutex_unlock(&fs_devices->device_list_mutex);
908 mutex_unlock(&fs_info->chunk_mutex);
909 } else {
910 break;
911 }
912 }
913
914 down_write(&dev_replace->rwsem);
915 dev_replace->replace_state =
916 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
917 : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
918 dev_replace->tgtdev = NULL;
919 dev_replace->srcdev = NULL;
920 dev_replace->time_stopped = ktime_get_real_seconds();
921 dev_replace->item_needs_writeback = 1;
922
923 /*
924 * Update allocation state in the new device and replace the old device
925 * with the new one in the mapping tree.
926 */
927 if (!scrub_ret) {
928 scrub_ret = btrfs_set_target_alloc_state(src_device, tgt_device);
929 if (scrub_ret)
930 goto error;
931 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
932 src_device,
933 tgt_device);
934 } else {
935 if (scrub_ret != -ECANCELED)
936 btrfs_err_in_rcu(fs_info,
937 "btrfs_scrub_dev(%s, %llu, %s) failed %d",
938 btrfs_dev_name(src_device),
939 src_device->devid,
940 btrfs_dev_name(tgt_device), scrub_ret);
941error:
942 up_write(&dev_replace->rwsem);
943 mutex_unlock(&fs_info->chunk_mutex);
944 mutex_unlock(&fs_devices->device_list_mutex);
945 btrfs_rm_dev_replace_blocked(fs_info);
946 if (tgt_device)
947 btrfs_destroy_dev_replace_tgtdev(tgt_device);
948 btrfs_rm_dev_replace_unblocked(fs_info);
949 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
950
951 return scrub_ret;
952 }
953
954 btrfs_info_in_rcu(fs_info,
955 "dev_replace from %s (devid %llu) to %s finished",
956 btrfs_dev_name(src_device),
957 src_device->devid,
958 btrfs_dev_name(tgt_device));
959 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state);
960 tgt_device->devid = src_device->devid;
961 src_device->devid = BTRFS_DEV_REPLACE_DEVID;
962 memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
963 memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
964 memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
965 btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
966 btrfs_device_set_disk_total_bytes(tgt_device,
967 src_device->disk_total_bytes);
968 btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
969 tgt_device->commit_bytes_used = src_device->bytes_used;
970
971 btrfs_assign_next_active_device(src_device, tgt_device);
972
973 list_add(&tgt_device->dev_alloc_list, &fs_devices->alloc_list);
974 fs_devices->rw_devices++;
975
976 up_write(&dev_replace->rwsem);
977 btrfs_rm_dev_replace_blocked(fs_info);
978
979 btrfs_rm_dev_replace_remove_srcdev(src_device);
980
981 btrfs_rm_dev_replace_unblocked(fs_info);
982
983 /*
984 * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will
985 * update on-disk dev stats value during commit transaction
986 */
987 atomic_inc(&tgt_device->dev_stats_ccnt);
988
989 /*
990 * this is again a consistent state where no dev_replace procedure
991 * is running, the target device is part of the filesystem, the
992 * source device is not part of the filesystem anymore and its 1st
993 * superblock is scratched out so that it is no longer marked to
994 * belong to this filesystem.
995 */
996 mutex_unlock(&fs_info->chunk_mutex);
997 mutex_unlock(&fs_devices->device_list_mutex);
998
999 /* replace the sysfs entry */
1000 btrfs_sysfs_remove_device(src_device);
1001 btrfs_sysfs_update_devid(tgt_device);
1002 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &src_device->dev_state))
1003 btrfs_scratch_superblocks(fs_info, src_device->bdev,
1004 src_device->name->str);
1005
1006 /* write back the superblocks */
1007 trans = btrfs_start_transaction(root, 0);
1008 if (!IS_ERR(trans))
1009 btrfs_commit_transaction(trans);
1010
1011 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
1012
1013 btrfs_rm_dev_replace_free_srcdev(src_device);
1014
1015 return 0;
1016}
1017
1018/*
1019 * Read progress of device replace status according to the state and last
1020 * stored position. The value format is the same as for
1021 * btrfs_dev_replace::progress_1000
1022 */
1023static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info)
1024{
1025 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1026 u64 ret = 0;
1027
1028 switch (dev_replace->replace_state) {
1029 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1030 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1031 ret = 0;
1032 break;
1033 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1034 ret = 1000;
1035 break;
1036 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1037 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1038 ret = div64_u64(dev_replace->cursor_left,
1039 div_u64(btrfs_device_get_total_bytes(
1040 dev_replace->srcdev), 1000));
1041 break;
1042 }
1043
1044 return ret;
1045}
1046
1047void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
1048 struct btrfs_ioctl_dev_replace_args *args)
1049{
1050 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1051
1052 down_read(&dev_replace->rwsem);
1053 /* even if !dev_replace_is_valid, the values are good enough for
1054 * the replace_status ioctl */
1055 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
1056 args->status.replace_state = dev_replace->replace_state;
1057 args->status.time_started = dev_replace->time_started;
1058 args->status.time_stopped = dev_replace->time_stopped;
1059 args->status.num_write_errors =
1060 atomic64_read(&dev_replace->num_write_errors);
1061 args->status.num_uncorrectable_read_errors =
1062 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
1063 args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
1064 up_read(&dev_replace->rwsem);
1065}
1066
1067int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
1068{
1069 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1070 struct btrfs_device *tgt_device = NULL;
1071 struct btrfs_device *src_device = NULL;
1072 struct btrfs_trans_handle *trans;
1073 struct btrfs_root *root = fs_info->tree_root;
1074 int result;
1075 int ret;
1076
1077 if (sb_rdonly(fs_info->sb))
1078 return -EROFS;
1079
1080 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
1081 down_write(&dev_replace->rwsem);
1082 switch (dev_replace->replace_state) {
1083 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1084 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1085 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1086 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
1087 up_write(&dev_replace->rwsem);
1088 break;
1089 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1090 tgt_device = dev_replace->tgtdev;
1091 src_device = dev_replace->srcdev;
1092 up_write(&dev_replace->rwsem);
1093 ret = btrfs_scrub_cancel(fs_info);
1094 if (ret < 0) {
1095 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
1096 } else {
1097 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
1098 /*
1099 * btrfs_dev_replace_finishing() will handle the
1100 * cleanup part
1101 */
1102 btrfs_info_in_rcu(fs_info,
1103 "dev_replace from %s (devid %llu) to %s canceled",
1104 btrfs_dev_name(src_device), src_device->devid,
1105 btrfs_dev_name(tgt_device));
1106 }
1107 break;
1108 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1109 /*
1110 * Scrub doing the replace isn't running so we need to do the
1111 * cleanup step of btrfs_dev_replace_finishing() here
1112 */
1113 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
1114 tgt_device = dev_replace->tgtdev;
1115 src_device = dev_replace->srcdev;
1116 dev_replace->tgtdev = NULL;
1117 dev_replace->srcdev = NULL;
1118 dev_replace->replace_state =
1119 BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
1120 dev_replace->time_stopped = ktime_get_real_seconds();
1121 dev_replace->item_needs_writeback = 1;
1122
1123 up_write(&dev_replace->rwsem);
1124
1125 /* Scrub for replace must not be running in suspended state */
1126 btrfs_scrub_cancel(fs_info);
1127
1128 trans = btrfs_start_transaction(root, 0);
1129 if (IS_ERR(trans)) {
1130 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
1131 return PTR_ERR(trans);
1132 }
1133 ret = btrfs_commit_transaction(trans);
1134 WARN_ON(ret);
1135
1136 btrfs_info_in_rcu(fs_info,
1137 "suspended dev_replace from %s (devid %llu) to %s canceled",
1138 btrfs_dev_name(src_device), src_device->devid,
1139 btrfs_dev_name(tgt_device));
1140
1141 if (tgt_device)
1142 btrfs_destroy_dev_replace_tgtdev(tgt_device);
1143 break;
1144 default:
1145 up_write(&dev_replace->rwsem);
1146 result = -EINVAL;
1147 }
1148
1149 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
1150 return result;
1151}
1152
1153void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
1154{
1155 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1156
1157 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
1158 down_write(&dev_replace->rwsem);
1159
1160 switch (dev_replace->replace_state) {
1161 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1162 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1163 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1164 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1165 break;
1166 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1167 dev_replace->replace_state =
1168 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
1169 dev_replace->time_stopped = ktime_get_real_seconds();
1170 dev_replace->item_needs_writeback = 1;
1171 btrfs_info(fs_info, "suspending dev_replace for unmount");
1172 break;
1173 }
1174
1175 up_write(&dev_replace->rwsem);
1176 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
1177}
1178
1179/* resume dev_replace procedure that was interrupted by unmount */
1180int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
1181{
1182 struct task_struct *task;
1183 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1184
1185 down_write(&dev_replace->rwsem);
1186
1187 switch (dev_replace->replace_state) {
1188 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1189 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1190 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1191 up_write(&dev_replace->rwsem);
1192 return 0;
1193 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1194 break;
1195 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1196 dev_replace->replace_state =
1197 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
1198 break;
1199 }
1200 if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
1201 btrfs_info(fs_info,
1202 "cannot continue dev_replace, tgtdev is missing");
1203 btrfs_info(fs_info,
1204 "you may cancel the operation after 'mount -o degraded'");
1205 dev_replace->replace_state =
1206 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
1207 up_write(&dev_replace->rwsem);
1208 return 0;
1209 }
1210 up_write(&dev_replace->rwsem);
1211
1212 /*
1213 * This could collide with a paused balance, but the exclusive op logic
1214 * should never allow both to start and pause. We don't want to allow
1215 * dev-replace to start anyway.
1216 */
1217 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
1218 down_write(&dev_replace->rwsem);
1219 dev_replace->replace_state =
1220 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
1221 up_write(&dev_replace->rwsem);
1222 btrfs_info(fs_info,
1223 "cannot resume dev-replace, other exclusive operation running");
1224 return 0;
1225 }
1226
1227 task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
1228 return PTR_ERR_OR_ZERO(task);
1229}
1230
1231static int btrfs_dev_replace_kthread(void *data)
1232{
1233 struct btrfs_fs_info *fs_info = data;
1234 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1235 u64 progress;
1236 int ret;
1237
1238 progress = btrfs_dev_replace_progress(fs_info);
1239 progress = div_u64(progress, 10);
1240 btrfs_info_in_rcu(fs_info,
1241 "continuing dev_replace from %s (devid %llu) to target %s @%u%%",
1242 btrfs_dev_name(dev_replace->srcdev),
1243 dev_replace->srcdev->devid,
1244 btrfs_dev_name(dev_replace->tgtdev),
1245 (unsigned int)progress);
1246
1247 ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
1248 dev_replace->committed_cursor_left,
1249 btrfs_device_get_total_bytes(dev_replace->srcdev),
1250 &dev_replace->scrub_progress, 0, 1);
1251 ret = btrfs_dev_replace_finishing(fs_info, ret);
1252 WARN_ON(ret && ret != -ECANCELED);
1253
1254 btrfs_exclop_finish(fs_info);
1255 return 0;
1256}
1257
1258int __pure btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
1259{
1260 if (!dev_replace->is_valid)
1261 return 0;
1262
1263 switch (dev_replace->replace_state) {
1264 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1265 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1266 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1267 return 0;
1268 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1269 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1270 /*
1271 * return true even if tgtdev is missing (this is
1272 * something that can happen if the dev_replace
1273 * procedure is suspended by an umount and then
1274 * the tgtdev is missing (or "btrfs dev scan") was
1275 * not called and the filesystem is remounted
1276 * in degraded state. This does not stop the
1277 * dev_replace procedure. It needs to be canceled
1278 * manually if the cancellation is wanted.
1279 */
1280 break;
1281 }
1282 return 1;
1283}
1284
1285void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
1286{
1287 percpu_counter_sub(&fs_info->dev_replace.bio_counter, amount);
1288 cond_wake_up_nomb(&fs_info->dev_replace.replace_wait);
1289}
1290
1291void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
1292{
1293 while (1) {
1294 percpu_counter_inc(&fs_info->dev_replace.bio_counter);
1295 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1296 &fs_info->fs_state)))
1297 break;
1298
1299 btrfs_bio_counter_dec(fs_info);
1300 wait_event(fs_info->dev_replace.replace_wait,
1301 !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1302 &fs_info->fs_state));
1303 }
1304}
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 "extent_map.h"
15#include "disk-io.h"
16#include "transaction.h"
17#include "print-tree.h"
18#include "volumes.h"
19#include "async-thread.h"
20#include "check-integrity.h"
21#include "rcu-string.h"
22#include "dev-replace.h"
23#include "sysfs.h"
24
25static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
26 int scrub_ret);
27static void btrfs_dev_replace_update_device_in_mapping_tree(
28 struct btrfs_fs_info *fs_info,
29 struct btrfs_device *srcdev,
30 struct btrfs_device *tgtdev);
31static int btrfs_dev_replace_kthread(void *data);
32
33int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
34{
35 struct btrfs_key key;
36 struct btrfs_root *dev_root = fs_info->dev_root;
37 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
38 struct extent_buffer *eb;
39 int slot;
40 int ret = 0;
41 struct btrfs_path *path = NULL;
42 int item_size;
43 struct btrfs_dev_replace_item *ptr;
44 u64 src_devid;
45
46 path = btrfs_alloc_path();
47 if (!path) {
48 ret = -ENOMEM;
49 goto out;
50 }
51
52 key.objectid = 0;
53 key.type = BTRFS_DEV_REPLACE_KEY;
54 key.offset = 0;
55 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
56 if (ret) {
57no_valid_dev_replace_entry_found:
58 ret = 0;
59 dev_replace->replace_state =
60 BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
61 dev_replace->cont_reading_from_srcdev_mode =
62 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
63 dev_replace->time_started = 0;
64 dev_replace->time_stopped = 0;
65 atomic64_set(&dev_replace->num_write_errors, 0);
66 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
67 dev_replace->cursor_left = 0;
68 dev_replace->committed_cursor_left = 0;
69 dev_replace->cursor_left_last_write_of_item = 0;
70 dev_replace->cursor_right = 0;
71 dev_replace->srcdev = NULL;
72 dev_replace->tgtdev = NULL;
73 dev_replace->is_valid = 0;
74 dev_replace->item_needs_writeback = 0;
75 goto out;
76 }
77 slot = path->slots[0];
78 eb = path->nodes[0];
79 item_size = btrfs_item_size_nr(eb, slot);
80 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
81
82 if (item_size != sizeof(struct btrfs_dev_replace_item)) {
83 btrfs_warn(fs_info,
84 "dev_replace entry found has unexpected size, ignore entry");
85 goto no_valid_dev_replace_entry_found;
86 }
87
88 src_devid = btrfs_dev_replace_src_devid(eb, ptr);
89 dev_replace->cont_reading_from_srcdev_mode =
90 btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
91 dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
92 dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
93 dev_replace->time_stopped =
94 btrfs_dev_replace_time_stopped(eb, ptr);
95 atomic64_set(&dev_replace->num_write_errors,
96 btrfs_dev_replace_num_write_errors(eb, ptr));
97 atomic64_set(&dev_replace->num_uncorrectable_read_errors,
98 btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
99 dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
100 dev_replace->committed_cursor_left = dev_replace->cursor_left;
101 dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
102 dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
103 dev_replace->is_valid = 1;
104
105 dev_replace->item_needs_writeback = 0;
106 switch (dev_replace->replace_state) {
107 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
108 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
109 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
110 dev_replace->srcdev = NULL;
111 dev_replace->tgtdev = NULL;
112 break;
113 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
114 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
115 dev_replace->srcdev = btrfs_find_device(fs_info->fs_devices,
116 src_devid, NULL, NULL, true);
117 dev_replace->tgtdev = btrfs_find_device(fs_info->fs_devices,
118 BTRFS_DEV_REPLACE_DEVID,
119 NULL, NULL, true);
120 /*
121 * allow 'btrfs dev replace_cancel' if src/tgt device is
122 * missing
123 */
124 if (!dev_replace->srcdev &&
125 !btrfs_test_opt(fs_info, DEGRADED)) {
126 ret = -EIO;
127 btrfs_warn(fs_info,
128 "cannot mount because device replace operation is ongoing and");
129 btrfs_warn(fs_info,
130 "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
131 src_devid);
132 }
133 if (!dev_replace->tgtdev &&
134 !btrfs_test_opt(fs_info, DEGRADED)) {
135 ret = -EIO;
136 btrfs_warn(fs_info,
137 "cannot mount because device replace operation is ongoing and");
138 btrfs_warn(fs_info,
139 "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
140 BTRFS_DEV_REPLACE_DEVID);
141 }
142 if (dev_replace->tgtdev) {
143 if (dev_replace->srcdev) {
144 dev_replace->tgtdev->total_bytes =
145 dev_replace->srcdev->total_bytes;
146 dev_replace->tgtdev->disk_total_bytes =
147 dev_replace->srcdev->disk_total_bytes;
148 dev_replace->tgtdev->commit_total_bytes =
149 dev_replace->srcdev->commit_total_bytes;
150 dev_replace->tgtdev->bytes_used =
151 dev_replace->srcdev->bytes_used;
152 dev_replace->tgtdev->commit_bytes_used =
153 dev_replace->srcdev->commit_bytes_used;
154 }
155 set_bit(BTRFS_DEV_STATE_REPLACE_TGT,
156 &dev_replace->tgtdev->dev_state);
157
158 WARN_ON(fs_info->fs_devices->rw_devices == 0);
159 dev_replace->tgtdev->io_width = fs_info->sectorsize;
160 dev_replace->tgtdev->io_align = fs_info->sectorsize;
161 dev_replace->tgtdev->sector_size = fs_info->sectorsize;
162 dev_replace->tgtdev->fs_info = fs_info;
163 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
164 &dev_replace->tgtdev->dev_state);
165 }
166 break;
167 }
168
169out:
170 btrfs_free_path(path);
171 return ret;
172}
173
174/*
175 * Initialize a new device for device replace target from a given source dev
176 * and path.
177 *
178 * Return 0 and new device in @device_out, otherwise return < 0
179 */
180static int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
181 const char *device_path,
182 struct btrfs_device *srcdev,
183 struct btrfs_device **device_out)
184{
185 struct btrfs_device *device;
186 struct block_device *bdev;
187 struct list_head *devices;
188 struct rcu_string *name;
189 u64 devid = BTRFS_DEV_REPLACE_DEVID;
190 int ret = 0;
191
192 *device_out = NULL;
193 if (fs_info->fs_devices->seeding) {
194 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
195 return -EINVAL;
196 }
197
198 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
199 fs_info->bdev_holder);
200 if (IS_ERR(bdev)) {
201 btrfs_err(fs_info, "target device %s is invalid!", device_path);
202 return PTR_ERR(bdev);
203 }
204
205 sync_blockdev(bdev);
206
207 devices = &fs_info->fs_devices->devices;
208 list_for_each_entry(device, devices, dev_list) {
209 if (device->bdev == bdev) {
210 btrfs_err(fs_info,
211 "target device is in the filesystem!");
212 ret = -EEXIST;
213 goto error;
214 }
215 }
216
217
218 if (i_size_read(bdev->bd_inode) <
219 btrfs_device_get_total_bytes(srcdev)) {
220 btrfs_err(fs_info,
221 "target device is smaller than source device!");
222 ret = -EINVAL;
223 goto error;
224 }
225
226
227 device = btrfs_alloc_device(NULL, &devid, NULL);
228 if (IS_ERR(device)) {
229 ret = PTR_ERR(device);
230 goto error;
231 }
232
233 name = rcu_string_strdup(device_path, GFP_KERNEL);
234 if (!name) {
235 btrfs_free_device(device);
236 ret = -ENOMEM;
237 goto error;
238 }
239 rcu_assign_pointer(device->name, name);
240
241 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
242 device->generation = 0;
243 device->io_width = fs_info->sectorsize;
244 device->io_align = fs_info->sectorsize;
245 device->sector_size = fs_info->sectorsize;
246 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
247 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
248 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
249 device->commit_total_bytes = srcdev->commit_total_bytes;
250 device->commit_bytes_used = device->bytes_used;
251 device->fs_info = fs_info;
252 device->bdev = bdev;
253 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
254 set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
255 device->mode = FMODE_EXCL;
256 device->dev_stats_valid = 1;
257 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
258 device->fs_devices = fs_info->fs_devices;
259
260 mutex_lock(&fs_info->fs_devices->device_list_mutex);
261 list_add(&device->dev_list, &fs_info->fs_devices->devices);
262 fs_info->fs_devices->num_devices++;
263 fs_info->fs_devices->open_devices++;
264 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
265
266 *device_out = device;
267 return 0;
268
269error:
270 blkdev_put(bdev, FMODE_EXCL);
271 return ret;
272}
273
274/*
275 * called from commit_transaction. Writes changed device replace state to
276 * disk.
277 */
278int btrfs_run_dev_replace(struct btrfs_trans_handle *trans)
279{
280 struct btrfs_fs_info *fs_info = trans->fs_info;
281 int ret;
282 struct btrfs_root *dev_root = fs_info->dev_root;
283 struct btrfs_path *path;
284 struct btrfs_key key;
285 struct extent_buffer *eb;
286 struct btrfs_dev_replace_item *ptr;
287 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
288
289 down_read(&dev_replace->rwsem);
290 if (!dev_replace->is_valid ||
291 !dev_replace->item_needs_writeback) {
292 up_read(&dev_replace->rwsem);
293 return 0;
294 }
295 up_read(&dev_replace->rwsem);
296
297 key.objectid = 0;
298 key.type = BTRFS_DEV_REPLACE_KEY;
299 key.offset = 0;
300
301 path = btrfs_alloc_path();
302 if (!path) {
303 ret = -ENOMEM;
304 goto out;
305 }
306 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
307 if (ret < 0) {
308 btrfs_warn(fs_info,
309 "error %d while searching for dev_replace item!",
310 ret);
311 goto out;
312 }
313
314 if (ret == 0 &&
315 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
316 /*
317 * need to delete old one and insert a new one.
318 * Since no attempt is made to recover any old state, if the
319 * dev_replace state is 'running', the data on the target
320 * drive is lost.
321 * It would be possible to recover the state: just make sure
322 * that the beginning of the item is never changed and always
323 * contains all the essential information. Then read this
324 * minimal set of information and use it as a base for the
325 * new state.
326 */
327 ret = btrfs_del_item(trans, dev_root, path);
328 if (ret != 0) {
329 btrfs_warn(fs_info,
330 "delete too small dev_replace item failed %d!",
331 ret);
332 goto out;
333 }
334 ret = 1;
335 }
336
337 if (ret == 1) {
338 /* need to insert a new item */
339 btrfs_release_path(path);
340 ret = btrfs_insert_empty_item(trans, dev_root, path,
341 &key, sizeof(*ptr));
342 if (ret < 0) {
343 btrfs_warn(fs_info,
344 "insert dev_replace item failed %d!", ret);
345 goto out;
346 }
347 }
348
349 eb = path->nodes[0];
350 ptr = btrfs_item_ptr(eb, path->slots[0],
351 struct btrfs_dev_replace_item);
352
353 down_write(&dev_replace->rwsem);
354 if (dev_replace->srcdev)
355 btrfs_set_dev_replace_src_devid(eb, ptr,
356 dev_replace->srcdev->devid);
357 else
358 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
359 btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
360 dev_replace->cont_reading_from_srcdev_mode);
361 btrfs_set_dev_replace_replace_state(eb, ptr,
362 dev_replace->replace_state);
363 btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
364 btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
365 btrfs_set_dev_replace_num_write_errors(eb, ptr,
366 atomic64_read(&dev_replace->num_write_errors));
367 btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
368 atomic64_read(&dev_replace->num_uncorrectable_read_errors));
369 dev_replace->cursor_left_last_write_of_item =
370 dev_replace->cursor_left;
371 btrfs_set_dev_replace_cursor_left(eb, ptr,
372 dev_replace->cursor_left_last_write_of_item);
373 btrfs_set_dev_replace_cursor_right(eb, ptr,
374 dev_replace->cursor_right);
375 dev_replace->item_needs_writeback = 0;
376 up_write(&dev_replace->rwsem);
377
378 btrfs_mark_buffer_dirty(eb);
379
380out:
381 btrfs_free_path(path);
382
383 return ret;
384}
385
386static char* btrfs_dev_name(struct btrfs_device *device)
387{
388 if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
389 return "<missing disk>";
390 else
391 return rcu_str_deref(device->name);
392}
393
394static int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
395 const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
396 int read_src)
397{
398 struct btrfs_root *root = fs_info->dev_root;
399 struct btrfs_trans_handle *trans;
400 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
401 int ret;
402 struct btrfs_device *tgt_device = NULL;
403 struct btrfs_device *src_device = NULL;
404
405 src_device = btrfs_find_device_by_devspec(fs_info, srcdevid,
406 srcdev_name);
407 if (IS_ERR(src_device))
408 return PTR_ERR(src_device);
409
410 if (btrfs_pinned_by_swapfile(fs_info, src_device)) {
411 btrfs_warn_in_rcu(fs_info,
412 "cannot replace device %s (devid %llu) due to active swapfile",
413 btrfs_dev_name(src_device), src_device->devid);
414 return -ETXTBSY;
415 }
416
417 /*
418 * Here we commit the transaction to make sure commit_total_bytes
419 * of all the devices are updated.
420 */
421 trans = btrfs_attach_transaction(root);
422 if (!IS_ERR(trans)) {
423 ret = btrfs_commit_transaction(trans);
424 if (ret)
425 return ret;
426 } else if (PTR_ERR(trans) != -ENOENT) {
427 return PTR_ERR(trans);
428 }
429
430 ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
431 src_device, &tgt_device);
432 if (ret)
433 return ret;
434
435 down_write(&dev_replace->rwsem);
436 switch (dev_replace->replace_state) {
437 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
438 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
439 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
440 break;
441 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
442 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
443 ASSERT(0);
444 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
445 up_write(&dev_replace->rwsem);
446 goto leave;
447 }
448
449 dev_replace->cont_reading_from_srcdev_mode = read_src;
450 dev_replace->srcdev = src_device;
451 dev_replace->tgtdev = tgt_device;
452
453 btrfs_info_in_rcu(fs_info,
454 "dev_replace from %s (devid %llu) to %s started",
455 btrfs_dev_name(src_device),
456 src_device->devid,
457 rcu_str_deref(tgt_device->name));
458
459 /*
460 * from now on, the writes to the srcdev are all duplicated to
461 * go to the tgtdev as well (refer to btrfs_map_block()).
462 */
463 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
464 dev_replace->time_started = ktime_get_real_seconds();
465 dev_replace->cursor_left = 0;
466 dev_replace->committed_cursor_left = 0;
467 dev_replace->cursor_left_last_write_of_item = 0;
468 dev_replace->cursor_right = 0;
469 dev_replace->is_valid = 1;
470 dev_replace->item_needs_writeback = 1;
471 atomic64_set(&dev_replace->num_write_errors, 0);
472 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
473 up_write(&dev_replace->rwsem);
474
475 ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device);
476 if (ret)
477 btrfs_err(fs_info, "kobj add dev failed %d", ret);
478
479 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
480
481 /* Commit dev_replace state and reserve 1 item for it. */
482 trans = btrfs_start_transaction(root, 1);
483 if (IS_ERR(trans)) {
484 ret = PTR_ERR(trans);
485 down_write(&dev_replace->rwsem);
486 dev_replace->replace_state =
487 BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
488 dev_replace->srcdev = NULL;
489 dev_replace->tgtdev = NULL;
490 up_write(&dev_replace->rwsem);
491 goto leave;
492 }
493
494 ret = btrfs_commit_transaction(trans);
495 WARN_ON(ret);
496
497 /* the disk copy procedure reuses the scrub code */
498 ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
499 btrfs_device_get_total_bytes(src_device),
500 &dev_replace->scrub_progress, 0, 1);
501
502 ret = btrfs_dev_replace_finishing(fs_info, ret);
503 if (ret == -EINPROGRESS) {
504 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
505 } else if (ret != -ECANCELED) {
506 WARN_ON(ret);
507 }
508
509 return ret;
510
511leave:
512 btrfs_destroy_dev_replace_tgtdev(tgt_device);
513 return ret;
514}
515
516int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
517 struct btrfs_ioctl_dev_replace_args *args)
518{
519 int ret;
520
521 switch (args->start.cont_reading_from_srcdev_mode) {
522 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
523 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
524 break;
525 default:
526 return -EINVAL;
527 }
528
529 if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
530 args->start.tgtdev_name[0] == '\0')
531 return -EINVAL;
532
533 ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name,
534 args->start.srcdevid,
535 args->start.srcdev_name,
536 args->start.cont_reading_from_srcdev_mode);
537 args->result = ret;
538 /* don't warn if EINPROGRESS, someone else might be running scrub */
539 if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS ||
540 ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR)
541 return 0;
542
543 return ret;
544}
545
546/*
547 * blocked until all in-flight bios operations are finished.
548 */
549static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
550{
551 set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
552 wait_event(fs_info->dev_replace.replace_wait, !percpu_counter_sum(
553 &fs_info->dev_replace.bio_counter));
554}
555
556/*
557 * we have removed target device, it is safe to allow new bios request.
558 */
559static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
560{
561 clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
562 wake_up(&fs_info->dev_replace.replace_wait);
563}
564
565static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
566 int scrub_ret)
567{
568 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
569 struct btrfs_device *tgt_device;
570 struct btrfs_device *src_device;
571 struct btrfs_root *root = fs_info->tree_root;
572 u8 uuid_tmp[BTRFS_UUID_SIZE];
573 struct btrfs_trans_handle *trans;
574 int ret = 0;
575
576 /* don't allow cancel or unmount to disturb the finishing procedure */
577 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
578
579 down_read(&dev_replace->rwsem);
580 /* was the operation canceled, or is it finished? */
581 if (dev_replace->replace_state !=
582 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
583 up_read(&dev_replace->rwsem);
584 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
585 return 0;
586 }
587
588 tgt_device = dev_replace->tgtdev;
589 src_device = dev_replace->srcdev;
590 up_read(&dev_replace->rwsem);
591
592 /*
593 * flush all outstanding I/O and inode extent mappings before the
594 * copy operation is declared as being finished
595 */
596 ret = btrfs_start_delalloc_roots(fs_info, -1);
597 if (ret) {
598 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
599 return ret;
600 }
601 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
602
603 /*
604 * We have to use this loop approach because at this point src_device
605 * has to be available for transaction commit to complete, yet new
606 * chunks shouldn't be allocated on the device.
607 */
608 while (1) {
609 trans = btrfs_start_transaction(root, 0);
610 if (IS_ERR(trans)) {
611 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
612 return PTR_ERR(trans);
613 }
614 ret = btrfs_commit_transaction(trans);
615 WARN_ON(ret);
616
617 /* Prevent write_all_supers() during the finishing procedure */
618 mutex_lock(&fs_info->fs_devices->device_list_mutex);
619 /* Prevent new chunks being allocated on the source device */
620 mutex_lock(&fs_info->chunk_mutex);
621
622 if (!list_empty(&src_device->post_commit_list)) {
623 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
624 mutex_unlock(&fs_info->chunk_mutex);
625 } else {
626 break;
627 }
628 }
629
630 down_write(&dev_replace->rwsem);
631 dev_replace->replace_state =
632 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
633 : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
634 dev_replace->tgtdev = NULL;
635 dev_replace->srcdev = NULL;
636 dev_replace->time_stopped = ktime_get_real_seconds();
637 dev_replace->item_needs_writeback = 1;
638
639 /* replace old device with new one in mapping tree */
640 if (!scrub_ret) {
641 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
642 src_device,
643 tgt_device);
644 } else {
645 if (scrub_ret != -ECANCELED)
646 btrfs_err_in_rcu(fs_info,
647 "btrfs_scrub_dev(%s, %llu, %s) failed %d",
648 btrfs_dev_name(src_device),
649 src_device->devid,
650 rcu_str_deref(tgt_device->name), scrub_ret);
651 up_write(&dev_replace->rwsem);
652 mutex_unlock(&fs_info->chunk_mutex);
653 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
654 btrfs_rm_dev_replace_blocked(fs_info);
655 if (tgt_device)
656 btrfs_destroy_dev_replace_tgtdev(tgt_device);
657 btrfs_rm_dev_replace_unblocked(fs_info);
658 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
659
660 return scrub_ret;
661 }
662
663 btrfs_info_in_rcu(fs_info,
664 "dev_replace from %s (devid %llu) to %s finished",
665 btrfs_dev_name(src_device),
666 src_device->devid,
667 rcu_str_deref(tgt_device->name));
668 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state);
669 tgt_device->devid = src_device->devid;
670 src_device->devid = BTRFS_DEV_REPLACE_DEVID;
671 memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
672 memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
673 memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
674 btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
675 btrfs_device_set_disk_total_bytes(tgt_device,
676 src_device->disk_total_bytes);
677 btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
678 tgt_device->commit_bytes_used = src_device->bytes_used;
679
680 btrfs_assign_next_active_device(src_device, tgt_device);
681
682 list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
683 fs_info->fs_devices->rw_devices++;
684
685 up_write(&dev_replace->rwsem);
686 btrfs_rm_dev_replace_blocked(fs_info);
687
688 btrfs_rm_dev_replace_remove_srcdev(src_device);
689
690 btrfs_rm_dev_replace_unblocked(fs_info);
691
692 /*
693 * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will
694 * update on-disk dev stats value during commit transaction
695 */
696 atomic_inc(&tgt_device->dev_stats_ccnt);
697
698 /*
699 * this is again a consistent state where no dev_replace procedure
700 * is running, the target device is part of the filesystem, the
701 * source device is not part of the filesystem anymore and its 1st
702 * superblock is scratched out so that it is no longer marked to
703 * belong to this filesystem.
704 */
705 mutex_unlock(&fs_info->chunk_mutex);
706 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
707
708 /* replace the sysfs entry */
709 btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device);
710 btrfs_rm_dev_replace_free_srcdev(src_device);
711
712 /* write back the superblocks */
713 trans = btrfs_start_transaction(root, 0);
714 if (!IS_ERR(trans))
715 btrfs_commit_transaction(trans);
716
717 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
718
719 return 0;
720}
721
722static void btrfs_dev_replace_update_device_in_mapping_tree(
723 struct btrfs_fs_info *fs_info,
724 struct btrfs_device *srcdev,
725 struct btrfs_device *tgtdev)
726{
727 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
728 struct extent_map *em;
729 struct map_lookup *map;
730 u64 start = 0;
731 int i;
732
733 write_lock(&em_tree->lock);
734 do {
735 em = lookup_extent_mapping(em_tree, start, (u64)-1);
736 if (!em)
737 break;
738 map = em->map_lookup;
739 for (i = 0; i < map->num_stripes; i++)
740 if (srcdev == map->stripes[i].dev)
741 map->stripes[i].dev = tgtdev;
742 start = em->start + em->len;
743 free_extent_map(em);
744 } while (start);
745 write_unlock(&em_tree->lock);
746}
747
748/*
749 * Read progress of device replace status according to the state and last
750 * stored position. The value format is the same as for
751 * btrfs_dev_replace::progress_1000
752 */
753static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info)
754{
755 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
756 u64 ret = 0;
757
758 switch (dev_replace->replace_state) {
759 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
760 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
761 ret = 0;
762 break;
763 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
764 ret = 1000;
765 break;
766 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
767 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
768 ret = div64_u64(dev_replace->cursor_left,
769 div_u64(btrfs_device_get_total_bytes(
770 dev_replace->srcdev), 1000));
771 break;
772 }
773
774 return ret;
775}
776
777void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
778 struct btrfs_ioctl_dev_replace_args *args)
779{
780 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
781
782 down_read(&dev_replace->rwsem);
783 /* even if !dev_replace_is_valid, the values are good enough for
784 * the replace_status ioctl */
785 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
786 args->status.replace_state = dev_replace->replace_state;
787 args->status.time_started = dev_replace->time_started;
788 args->status.time_stopped = dev_replace->time_stopped;
789 args->status.num_write_errors =
790 atomic64_read(&dev_replace->num_write_errors);
791 args->status.num_uncorrectable_read_errors =
792 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
793 args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
794 up_read(&dev_replace->rwsem);
795}
796
797int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
798{
799 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
800 struct btrfs_device *tgt_device = NULL;
801 struct btrfs_device *src_device = NULL;
802 struct btrfs_trans_handle *trans;
803 struct btrfs_root *root = fs_info->tree_root;
804 int result;
805 int ret;
806
807 if (sb_rdonly(fs_info->sb))
808 return -EROFS;
809
810 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
811 down_write(&dev_replace->rwsem);
812 switch (dev_replace->replace_state) {
813 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
814 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
815 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
816 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
817 up_write(&dev_replace->rwsem);
818 break;
819 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
820 tgt_device = dev_replace->tgtdev;
821 src_device = dev_replace->srcdev;
822 up_write(&dev_replace->rwsem);
823 ret = btrfs_scrub_cancel(fs_info);
824 if (ret < 0) {
825 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
826 } else {
827 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
828 /*
829 * btrfs_dev_replace_finishing() will handle the
830 * cleanup part
831 */
832 btrfs_info_in_rcu(fs_info,
833 "dev_replace from %s (devid %llu) to %s canceled",
834 btrfs_dev_name(src_device), src_device->devid,
835 btrfs_dev_name(tgt_device));
836 }
837 break;
838 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
839 /*
840 * Scrub doing the replace isn't running so we need to do the
841 * cleanup step of btrfs_dev_replace_finishing() here
842 */
843 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
844 tgt_device = dev_replace->tgtdev;
845 src_device = dev_replace->srcdev;
846 dev_replace->tgtdev = NULL;
847 dev_replace->srcdev = NULL;
848 dev_replace->replace_state =
849 BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
850 dev_replace->time_stopped = ktime_get_real_seconds();
851 dev_replace->item_needs_writeback = 1;
852
853 up_write(&dev_replace->rwsem);
854
855 /* Scrub for replace must not be running in suspended state */
856 ret = btrfs_scrub_cancel(fs_info);
857 ASSERT(ret != -ENOTCONN);
858
859 trans = btrfs_start_transaction(root, 0);
860 if (IS_ERR(trans)) {
861 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
862 return PTR_ERR(trans);
863 }
864 ret = btrfs_commit_transaction(trans);
865 WARN_ON(ret);
866
867 btrfs_info_in_rcu(fs_info,
868 "suspended dev_replace from %s (devid %llu) to %s canceled",
869 btrfs_dev_name(src_device), src_device->devid,
870 btrfs_dev_name(tgt_device));
871
872 if (tgt_device)
873 btrfs_destroy_dev_replace_tgtdev(tgt_device);
874 break;
875 default:
876 up_write(&dev_replace->rwsem);
877 result = -EINVAL;
878 }
879
880 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
881 return result;
882}
883
884void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
885{
886 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
887
888 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
889 down_write(&dev_replace->rwsem);
890
891 switch (dev_replace->replace_state) {
892 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
893 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
894 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
895 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
896 break;
897 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
898 dev_replace->replace_state =
899 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
900 dev_replace->time_stopped = ktime_get_real_seconds();
901 dev_replace->item_needs_writeback = 1;
902 btrfs_info(fs_info, "suspending dev_replace for unmount");
903 break;
904 }
905
906 up_write(&dev_replace->rwsem);
907 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
908}
909
910/* resume dev_replace procedure that was interrupted by unmount */
911int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
912{
913 struct task_struct *task;
914 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
915
916 down_write(&dev_replace->rwsem);
917
918 switch (dev_replace->replace_state) {
919 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
920 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
921 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
922 up_write(&dev_replace->rwsem);
923 return 0;
924 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
925 break;
926 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
927 dev_replace->replace_state =
928 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
929 break;
930 }
931 if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
932 btrfs_info(fs_info,
933 "cannot continue dev_replace, tgtdev is missing");
934 btrfs_info(fs_info,
935 "you may cancel the operation after 'mount -o degraded'");
936 dev_replace->replace_state =
937 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
938 up_write(&dev_replace->rwsem);
939 return 0;
940 }
941 up_write(&dev_replace->rwsem);
942
943 /*
944 * This could collide with a paused balance, but the exclusive op logic
945 * should never allow both to start and pause. We don't want to allow
946 * dev-replace to start anyway.
947 */
948 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
949 down_write(&dev_replace->rwsem);
950 dev_replace->replace_state =
951 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
952 up_write(&dev_replace->rwsem);
953 btrfs_info(fs_info,
954 "cannot resume dev-replace, other exclusive operation running");
955 return 0;
956 }
957
958 task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
959 return PTR_ERR_OR_ZERO(task);
960}
961
962static int btrfs_dev_replace_kthread(void *data)
963{
964 struct btrfs_fs_info *fs_info = data;
965 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
966 u64 progress;
967 int ret;
968
969 progress = btrfs_dev_replace_progress(fs_info);
970 progress = div_u64(progress, 10);
971 btrfs_info_in_rcu(fs_info,
972 "continuing dev_replace from %s (devid %llu) to target %s @%u%%",
973 btrfs_dev_name(dev_replace->srcdev),
974 dev_replace->srcdev->devid,
975 btrfs_dev_name(dev_replace->tgtdev),
976 (unsigned int)progress);
977
978 ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
979 dev_replace->committed_cursor_left,
980 btrfs_device_get_total_bytes(dev_replace->srcdev),
981 &dev_replace->scrub_progress, 0, 1);
982 ret = btrfs_dev_replace_finishing(fs_info, ret);
983 WARN_ON(ret && ret != -ECANCELED);
984
985 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
986 return 0;
987}
988
989int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
990{
991 if (!dev_replace->is_valid)
992 return 0;
993
994 switch (dev_replace->replace_state) {
995 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
996 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
997 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
998 return 0;
999 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1000 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1001 /*
1002 * return true even if tgtdev is missing (this is
1003 * something that can happen if the dev_replace
1004 * procedure is suspended by an umount and then
1005 * the tgtdev is missing (or "btrfs dev scan") was
1006 * not called and the filesystem is remounted
1007 * in degraded state. This does not stop the
1008 * dev_replace procedure. It needs to be canceled
1009 * manually if the cancellation is wanted.
1010 */
1011 break;
1012 }
1013 return 1;
1014}
1015
1016void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
1017{
1018 percpu_counter_inc(&fs_info->dev_replace.bio_counter);
1019}
1020
1021void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
1022{
1023 percpu_counter_sub(&fs_info->dev_replace.bio_counter, amount);
1024 cond_wake_up_nomb(&fs_info->dev_replace.replace_wait);
1025}
1026
1027void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
1028{
1029 while (1) {
1030 percpu_counter_inc(&fs_info->dev_replace.bio_counter);
1031 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1032 &fs_info->fs_state)))
1033 break;
1034
1035 btrfs_bio_counter_dec(fs_info);
1036 wait_event(fs_info->dev_replace.replace_wait,
1037 !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1038 &fs_info->fs_state));
1039 }
1040}