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/*
2 * Copyright (C) STRATO AG 2012. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18#include <linux/sched.h>
19#include <linux/bio.h>
20#include <linux/slab.h>
21#include <linux/buffer_head.h>
22#include <linux/blkdev.h>
23#include <linux/random.h>
24#include <linux/iocontext.h>
25#include <linux/capability.h>
26#include <linux/kthread.h>
27#include <linux/math64.h>
28#include <asm/div64.h>
29#include "ctree.h"
30#include "extent_map.h"
31#include "disk-io.h"
32#include "transaction.h"
33#include "print-tree.h"
34#include "volumes.h"
35#include "async-thread.h"
36#include "check-integrity.h"
37#include "rcu-string.h"
38#include "dev-replace.h"
39#include "sysfs.h"
40
41static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
42 int scrub_ret);
43static void btrfs_dev_replace_update_device_in_mapping_tree(
44 struct btrfs_fs_info *fs_info,
45 struct btrfs_device *srcdev,
46 struct btrfs_device *tgtdev);
47static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info);
48static int btrfs_dev_replace_kthread(void *data);
49static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info);
50
51
52int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
53{
54 struct btrfs_key key;
55 struct btrfs_root *dev_root = fs_info->dev_root;
56 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
57 struct extent_buffer *eb;
58 int slot;
59 int ret = 0;
60 struct btrfs_path *path = NULL;
61 int item_size;
62 struct btrfs_dev_replace_item *ptr;
63 u64 src_devid;
64
65 path = btrfs_alloc_path();
66 if (!path) {
67 ret = -ENOMEM;
68 goto out;
69 }
70
71 key.objectid = 0;
72 key.type = BTRFS_DEV_REPLACE_KEY;
73 key.offset = 0;
74 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
75 if (ret) {
76no_valid_dev_replace_entry_found:
77 ret = 0;
78 dev_replace->replace_state =
79 BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED;
80 dev_replace->cont_reading_from_srcdev_mode =
81 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
82 dev_replace->replace_state = 0;
83 dev_replace->time_started = 0;
84 dev_replace->time_stopped = 0;
85 atomic64_set(&dev_replace->num_write_errors, 0);
86 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
87 dev_replace->cursor_left = 0;
88 dev_replace->committed_cursor_left = 0;
89 dev_replace->cursor_left_last_write_of_item = 0;
90 dev_replace->cursor_right = 0;
91 dev_replace->srcdev = NULL;
92 dev_replace->tgtdev = NULL;
93 dev_replace->is_valid = 0;
94 dev_replace->item_needs_writeback = 0;
95 goto out;
96 }
97 slot = path->slots[0];
98 eb = path->nodes[0];
99 item_size = btrfs_item_size_nr(eb, slot);
100 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
101
102 if (item_size != sizeof(struct btrfs_dev_replace_item)) {
103 btrfs_warn(fs_info,
104 "dev_replace entry found has unexpected size, ignore entry");
105 goto no_valid_dev_replace_entry_found;
106 }
107
108 src_devid = btrfs_dev_replace_src_devid(eb, ptr);
109 dev_replace->cont_reading_from_srcdev_mode =
110 btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
111 dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
112 dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
113 dev_replace->time_stopped =
114 btrfs_dev_replace_time_stopped(eb, ptr);
115 atomic64_set(&dev_replace->num_write_errors,
116 btrfs_dev_replace_num_write_errors(eb, ptr));
117 atomic64_set(&dev_replace->num_uncorrectable_read_errors,
118 btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
119 dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
120 dev_replace->committed_cursor_left = dev_replace->cursor_left;
121 dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
122 dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
123 dev_replace->is_valid = 1;
124
125 dev_replace->item_needs_writeback = 0;
126 switch (dev_replace->replace_state) {
127 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
128 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
129 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
130 dev_replace->srcdev = NULL;
131 dev_replace->tgtdev = NULL;
132 break;
133 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
134 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
135 dev_replace->srcdev = btrfs_find_device(fs_info, src_devid,
136 NULL, NULL);
137 dev_replace->tgtdev = btrfs_find_device(fs_info,
138 BTRFS_DEV_REPLACE_DEVID,
139 NULL, NULL);
140 /*
141 * allow 'btrfs dev replace_cancel' if src/tgt device is
142 * missing
143 */
144 if (!dev_replace->srcdev &&
145 !btrfs_test_opt(fs_info, DEGRADED)) {
146 ret = -EIO;
147 btrfs_warn(fs_info,
148 "cannot mount because device replace operation is ongoing and");
149 btrfs_warn(fs_info,
150 "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
151 src_devid);
152 }
153 if (!dev_replace->tgtdev &&
154 !btrfs_test_opt(fs_info, DEGRADED)) {
155 ret = -EIO;
156 btrfs_warn(fs_info,
157 "cannot mount because device replace operation is ongoing and");
158 btrfs_warn(fs_info,
159 "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
160 BTRFS_DEV_REPLACE_DEVID);
161 }
162 if (dev_replace->tgtdev) {
163 if (dev_replace->srcdev) {
164 dev_replace->tgtdev->total_bytes =
165 dev_replace->srcdev->total_bytes;
166 dev_replace->tgtdev->disk_total_bytes =
167 dev_replace->srcdev->disk_total_bytes;
168 dev_replace->tgtdev->commit_total_bytes =
169 dev_replace->srcdev->commit_total_bytes;
170 dev_replace->tgtdev->bytes_used =
171 dev_replace->srcdev->bytes_used;
172 dev_replace->tgtdev->commit_bytes_used =
173 dev_replace->srcdev->commit_bytes_used;
174 }
175 dev_replace->tgtdev->is_tgtdev_for_dev_replace = 1;
176 btrfs_init_dev_replace_tgtdev_for_resume(fs_info,
177 dev_replace->tgtdev);
178 }
179 break;
180 }
181
182out:
183 btrfs_free_path(path);
184 return ret;
185}
186
187/*
188 * called from commit_transaction. Writes changed device replace state to
189 * disk.
190 */
191int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
192 struct btrfs_fs_info *fs_info)
193{
194 int ret;
195 struct btrfs_root *dev_root = fs_info->dev_root;
196 struct btrfs_path *path;
197 struct btrfs_key key;
198 struct extent_buffer *eb;
199 struct btrfs_dev_replace_item *ptr;
200 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
201
202 btrfs_dev_replace_lock(dev_replace, 0);
203 if (!dev_replace->is_valid ||
204 !dev_replace->item_needs_writeback) {
205 btrfs_dev_replace_unlock(dev_replace, 0);
206 return 0;
207 }
208 btrfs_dev_replace_unlock(dev_replace, 0);
209
210 key.objectid = 0;
211 key.type = BTRFS_DEV_REPLACE_KEY;
212 key.offset = 0;
213
214 path = btrfs_alloc_path();
215 if (!path) {
216 ret = -ENOMEM;
217 goto out;
218 }
219 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
220 if (ret < 0) {
221 btrfs_warn(fs_info,
222 "error %d while searching for dev_replace item!",
223 ret);
224 goto out;
225 }
226
227 if (ret == 0 &&
228 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
229 /*
230 * need to delete old one and insert a new one.
231 * Since no attempt is made to recover any old state, if the
232 * dev_replace state is 'running', the data on the target
233 * drive is lost.
234 * It would be possible to recover the state: just make sure
235 * that the beginning of the item is never changed and always
236 * contains all the essential information. Then read this
237 * minimal set of information and use it as a base for the
238 * new state.
239 */
240 ret = btrfs_del_item(trans, dev_root, path);
241 if (ret != 0) {
242 btrfs_warn(fs_info,
243 "delete too small dev_replace item failed %d!",
244 ret);
245 goto out;
246 }
247 ret = 1;
248 }
249
250 if (ret == 1) {
251 /* need to insert a new item */
252 btrfs_release_path(path);
253 ret = btrfs_insert_empty_item(trans, dev_root, path,
254 &key, sizeof(*ptr));
255 if (ret < 0) {
256 btrfs_warn(fs_info,
257 "insert dev_replace item failed %d!", ret);
258 goto out;
259 }
260 }
261
262 eb = path->nodes[0];
263 ptr = btrfs_item_ptr(eb, path->slots[0],
264 struct btrfs_dev_replace_item);
265
266 btrfs_dev_replace_lock(dev_replace, 1);
267 if (dev_replace->srcdev)
268 btrfs_set_dev_replace_src_devid(eb, ptr,
269 dev_replace->srcdev->devid);
270 else
271 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
272 btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
273 dev_replace->cont_reading_from_srcdev_mode);
274 btrfs_set_dev_replace_replace_state(eb, ptr,
275 dev_replace->replace_state);
276 btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
277 btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
278 btrfs_set_dev_replace_num_write_errors(eb, ptr,
279 atomic64_read(&dev_replace->num_write_errors));
280 btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
281 atomic64_read(&dev_replace->num_uncorrectable_read_errors));
282 dev_replace->cursor_left_last_write_of_item =
283 dev_replace->cursor_left;
284 btrfs_set_dev_replace_cursor_left(eb, ptr,
285 dev_replace->cursor_left_last_write_of_item);
286 btrfs_set_dev_replace_cursor_right(eb, ptr,
287 dev_replace->cursor_right);
288 dev_replace->item_needs_writeback = 0;
289 btrfs_dev_replace_unlock(dev_replace, 1);
290
291 btrfs_mark_buffer_dirty(eb);
292
293out:
294 btrfs_free_path(path);
295
296 return ret;
297}
298
299void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
300{
301 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
302
303 dev_replace->committed_cursor_left =
304 dev_replace->cursor_left_last_write_of_item;
305}
306
307int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info, char *tgtdev_name,
308 u64 srcdevid, char *srcdev_name, int read_src)
309{
310 struct btrfs_root *root = fs_info->dev_root;
311 struct btrfs_trans_handle *trans;
312 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
313 int ret;
314 struct btrfs_device *tgt_device = NULL;
315 struct btrfs_device *src_device = NULL;
316
317 /* the disk copy procedure reuses the scrub code */
318 mutex_lock(&fs_info->volume_mutex);
319 ret = btrfs_find_device_by_devspec(fs_info, srcdevid,
320 srcdev_name, &src_device);
321 if (ret) {
322 mutex_unlock(&fs_info->volume_mutex);
323 return ret;
324 }
325
326 ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
327 src_device, &tgt_device);
328 mutex_unlock(&fs_info->volume_mutex);
329 if (ret)
330 return ret;
331
332 /*
333 * Here we commit the transaction to make sure commit_total_bytes
334 * of all the devices are updated.
335 */
336 trans = btrfs_attach_transaction(root);
337 if (!IS_ERR(trans)) {
338 ret = btrfs_commit_transaction(trans);
339 if (ret)
340 return ret;
341 } else if (PTR_ERR(trans) != -ENOENT) {
342 return PTR_ERR(trans);
343 }
344
345 btrfs_dev_replace_lock(dev_replace, 1);
346 switch (dev_replace->replace_state) {
347 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
348 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
349 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
350 break;
351 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
352 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
353 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
354 goto leave;
355 }
356
357 dev_replace->cont_reading_from_srcdev_mode = read_src;
358 WARN_ON(!src_device);
359 dev_replace->srcdev = src_device;
360 WARN_ON(!tgt_device);
361 dev_replace->tgtdev = tgt_device;
362
363 btrfs_info_in_rcu(fs_info,
364 "dev_replace from %s (devid %llu) to %s started",
365 src_device->missing ? "<missing disk>" :
366 rcu_str_deref(src_device->name),
367 src_device->devid,
368 rcu_str_deref(tgt_device->name));
369
370 /*
371 * from now on, the writes to the srcdev are all duplicated to
372 * go to the tgtdev as well (refer to btrfs_map_block()).
373 */
374 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
375 dev_replace->time_started = get_seconds();
376 dev_replace->cursor_left = 0;
377 dev_replace->committed_cursor_left = 0;
378 dev_replace->cursor_left_last_write_of_item = 0;
379 dev_replace->cursor_right = 0;
380 dev_replace->is_valid = 1;
381 dev_replace->item_needs_writeback = 1;
382 atomic64_set(&dev_replace->num_write_errors, 0);
383 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
384 btrfs_dev_replace_unlock(dev_replace, 1);
385
386 ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device);
387 if (ret)
388 btrfs_err(fs_info, "kobj add dev failed %d", ret);
389
390 btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
391
392 /* force writing the updated state information to disk */
393 trans = btrfs_start_transaction(root, 0);
394 if (IS_ERR(trans)) {
395 ret = PTR_ERR(trans);
396 btrfs_dev_replace_lock(dev_replace, 1);
397 goto leave;
398 }
399
400 ret = btrfs_commit_transaction(trans);
401 WARN_ON(ret);
402
403 /* the disk copy procedure reuses the scrub code */
404 ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
405 btrfs_device_get_total_bytes(src_device),
406 &dev_replace->scrub_progress, 0, 1);
407
408 ret = btrfs_dev_replace_finishing(fs_info, ret);
409 if (ret == -EINPROGRESS) {
410 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
411 } else {
412 WARN_ON(ret);
413 }
414
415 return ret;
416
417leave:
418 dev_replace->srcdev = NULL;
419 dev_replace->tgtdev = NULL;
420 btrfs_dev_replace_unlock(dev_replace, 1);
421 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
422 return ret;
423}
424
425int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
426 struct btrfs_ioctl_dev_replace_args *args)
427{
428 int ret;
429
430 switch (args->start.cont_reading_from_srcdev_mode) {
431 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
432 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
433 break;
434 default:
435 return -EINVAL;
436 }
437
438 if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
439 args->start.tgtdev_name[0] == '\0')
440 return -EINVAL;
441
442 ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name,
443 args->start.srcdevid,
444 args->start.srcdev_name,
445 args->start.cont_reading_from_srcdev_mode);
446 args->result = ret;
447 /* don't warn if EINPROGRESS, someone else might be running scrub */
448 if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS)
449 ret = 0;
450
451 return ret;
452}
453
454/*
455 * blocked until all in-flight bios operations are finished.
456 */
457static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
458{
459 set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
460 wait_event(fs_info->replace_wait, !percpu_counter_sum(
461 &fs_info->bio_counter));
462}
463
464/*
465 * we have removed target device, it is safe to allow new bios request.
466 */
467static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
468{
469 clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
470 wake_up(&fs_info->replace_wait);
471}
472
473static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
474 int scrub_ret)
475{
476 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
477 struct btrfs_device *tgt_device;
478 struct btrfs_device *src_device;
479 struct btrfs_root *root = fs_info->tree_root;
480 u8 uuid_tmp[BTRFS_UUID_SIZE];
481 struct btrfs_trans_handle *trans;
482 int ret = 0;
483
484 /* don't allow cancel or unmount to disturb the finishing procedure */
485 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
486
487 btrfs_dev_replace_lock(dev_replace, 0);
488 /* was the operation canceled, or is it finished? */
489 if (dev_replace->replace_state !=
490 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
491 btrfs_dev_replace_unlock(dev_replace, 0);
492 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
493 return 0;
494 }
495
496 tgt_device = dev_replace->tgtdev;
497 src_device = dev_replace->srcdev;
498 btrfs_dev_replace_unlock(dev_replace, 0);
499
500 /*
501 * flush all outstanding I/O and inode extent mappings before the
502 * copy operation is declared as being finished
503 */
504 ret = btrfs_start_delalloc_roots(fs_info, 0, -1);
505 if (ret) {
506 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
507 return ret;
508 }
509 btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
510
511 trans = btrfs_start_transaction(root, 0);
512 if (IS_ERR(trans)) {
513 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
514 return PTR_ERR(trans);
515 }
516 ret = btrfs_commit_transaction(trans);
517 WARN_ON(ret);
518
519 mutex_lock(&uuid_mutex);
520 /* keep away write_all_supers() during the finishing procedure */
521 mutex_lock(&fs_info->fs_devices->device_list_mutex);
522 mutex_lock(&fs_info->chunk_mutex);
523 btrfs_dev_replace_lock(dev_replace, 1);
524 dev_replace->replace_state =
525 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
526 : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
527 dev_replace->tgtdev = NULL;
528 dev_replace->srcdev = NULL;
529 dev_replace->time_stopped = get_seconds();
530 dev_replace->item_needs_writeback = 1;
531
532 /* replace old device with new one in mapping tree */
533 if (!scrub_ret) {
534 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
535 src_device,
536 tgt_device);
537 } else {
538 btrfs_err_in_rcu(fs_info,
539 "btrfs_scrub_dev(%s, %llu, %s) failed %d",
540 src_device->missing ? "<missing disk>" :
541 rcu_str_deref(src_device->name),
542 src_device->devid,
543 rcu_str_deref(tgt_device->name), scrub_ret);
544 btrfs_dev_replace_unlock(dev_replace, 1);
545 mutex_unlock(&fs_info->chunk_mutex);
546 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
547 mutex_unlock(&uuid_mutex);
548 if (tgt_device)
549 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
550 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
551
552 return scrub_ret;
553 }
554
555 btrfs_info_in_rcu(fs_info,
556 "dev_replace from %s (devid %llu) to %s finished",
557 src_device->missing ? "<missing disk>" :
558 rcu_str_deref(src_device->name),
559 src_device->devid,
560 rcu_str_deref(tgt_device->name));
561 tgt_device->is_tgtdev_for_dev_replace = 0;
562 tgt_device->devid = src_device->devid;
563 src_device->devid = BTRFS_DEV_REPLACE_DEVID;
564 memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
565 memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
566 memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
567 btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
568 btrfs_device_set_disk_total_bytes(tgt_device,
569 src_device->disk_total_bytes);
570 btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
571 ASSERT(list_empty(&src_device->resized_list));
572 tgt_device->commit_total_bytes = src_device->commit_total_bytes;
573 tgt_device->commit_bytes_used = src_device->bytes_used;
574
575 btrfs_assign_next_active_device(fs_info, src_device, tgt_device);
576
577 list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
578 fs_info->fs_devices->rw_devices++;
579
580 btrfs_dev_replace_unlock(dev_replace, 1);
581
582 btrfs_rm_dev_replace_blocked(fs_info);
583
584 btrfs_rm_dev_replace_remove_srcdev(fs_info, src_device);
585
586 btrfs_rm_dev_replace_unblocked(fs_info);
587
588 /*
589 * this is again a consistent state where no dev_replace procedure
590 * is running, the target device is part of the filesystem, the
591 * source device is not part of the filesystem anymore and its 1st
592 * superblock is scratched out so that it is no longer marked to
593 * belong to this filesystem.
594 */
595 mutex_unlock(&fs_info->chunk_mutex);
596 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
597 mutex_unlock(&uuid_mutex);
598
599 /* replace the sysfs entry */
600 btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device);
601 btrfs_rm_dev_replace_free_srcdev(fs_info, src_device);
602
603 /* write back the superblocks */
604 trans = btrfs_start_transaction(root, 0);
605 if (!IS_ERR(trans))
606 btrfs_commit_transaction(trans);
607
608 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
609
610 return 0;
611}
612
613static void btrfs_dev_replace_update_device_in_mapping_tree(
614 struct btrfs_fs_info *fs_info,
615 struct btrfs_device *srcdev,
616 struct btrfs_device *tgtdev)
617{
618 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
619 struct extent_map *em;
620 struct map_lookup *map;
621 u64 start = 0;
622 int i;
623
624 write_lock(&em_tree->lock);
625 do {
626 em = lookup_extent_mapping(em_tree, start, (u64)-1);
627 if (!em)
628 break;
629 map = em->map_lookup;
630 for (i = 0; i < map->num_stripes; i++)
631 if (srcdev == map->stripes[i].dev)
632 map->stripes[i].dev = tgtdev;
633 start = em->start + em->len;
634 free_extent_map(em);
635 } while (start);
636 write_unlock(&em_tree->lock);
637}
638
639void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
640 struct btrfs_ioctl_dev_replace_args *args)
641{
642 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
643 struct btrfs_device *srcdev;
644
645 btrfs_dev_replace_lock(dev_replace, 0);
646 /* even if !dev_replace_is_valid, the values are good enough for
647 * the replace_status ioctl */
648 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
649 args->status.replace_state = dev_replace->replace_state;
650 args->status.time_started = dev_replace->time_started;
651 args->status.time_stopped = dev_replace->time_stopped;
652 args->status.num_write_errors =
653 atomic64_read(&dev_replace->num_write_errors);
654 args->status.num_uncorrectable_read_errors =
655 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
656 switch (dev_replace->replace_state) {
657 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
658 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
659 args->status.progress_1000 = 0;
660 break;
661 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
662 args->status.progress_1000 = 1000;
663 break;
664 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
665 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
666 srcdev = dev_replace->srcdev;
667 args->status.progress_1000 = div_u64(dev_replace->cursor_left,
668 div_u64(btrfs_device_get_total_bytes(srcdev), 1000));
669 break;
670 }
671 btrfs_dev_replace_unlock(dev_replace, 0);
672}
673
674int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info,
675 struct btrfs_ioctl_dev_replace_args *args)
676{
677 args->result = __btrfs_dev_replace_cancel(fs_info);
678 return 0;
679}
680
681static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
682{
683 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
684 struct btrfs_device *tgt_device = NULL;
685 struct btrfs_trans_handle *trans;
686 struct btrfs_root *root = fs_info->tree_root;
687 u64 result;
688 int ret;
689
690 if (fs_info->sb->s_flags & MS_RDONLY)
691 return -EROFS;
692
693 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
694 btrfs_dev_replace_lock(dev_replace, 1);
695 switch (dev_replace->replace_state) {
696 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
697 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
698 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
699 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
700 btrfs_dev_replace_unlock(dev_replace, 1);
701 goto leave;
702 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
703 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
704 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
705 tgt_device = dev_replace->tgtdev;
706 dev_replace->tgtdev = NULL;
707 dev_replace->srcdev = NULL;
708 break;
709 }
710 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
711 dev_replace->time_stopped = get_seconds();
712 dev_replace->item_needs_writeback = 1;
713 btrfs_dev_replace_unlock(dev_replace, 1);
714 btrfs_scrub_cancel(fs_info);
715
716 trans = btrfs_start_transaction(root, 0);
717 if (IS_ERR(trans)) {
718 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
719 return PTR_ERR(trans);
720 }
721 ret = btrfs_commit_transaction(trans);
722 WARN_ON(ret);
723 if (tgt_device)
724 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
725
726leave:
727 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
728 return result;
729}
730
731void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
732{
733 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
734
735 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
736 btrfs_dev_replace_lock(dev_replace, 1);
737 switch (dev_replace->replace_state) {
738 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
739 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
740 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
741 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
742 break;
743 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
744 dev_replace->replace_state =
745 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
746 dev_replace->time_stopped = get_seconds();
747 dev_replace->item_needs_writeback = 1;
748 btrfs_info(fs_info, "suspending dev_replace for unmount");
749 break;
750 }
751
752 btrfs_dev_replace_unlock(dev_replace, 1);
753 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
754}
755
756/* resume dev_replace procedure that was interrupted by unmount */
757int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
758{
759 struct task_struct *task;
760 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
761
762 btrfs_dev_replace_lock(dev_replace, 1);
763 switch (dev_replace->replace_state) {
764 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
765 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
766 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
767 btrfs_dev_replace_unlock(dev_replace, 1);
768 return 0;
769 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
770 break;
771 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
772 dev_replace->replace_state =
773 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
774 break;
775 }
776 if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
777 btrfs_info(fs_info,
778 "cannot continue dev_replace, tgtdev is missing");
779 btrfs_info(fs_info,
780 "you may cancel the operation after 'mount -o degraded'");
781 btrfs_dev_replace_unlock(dev_replace, 1);
782 return 0;
783 }
784 btrfs_dev_replace_unlock(dev_replace, 1);
785
786 WARN_ON(atomic_xchg(
787 &fs_info->mutually_exclusive_operation_running, 1));
788 task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
789 return PTR_ERR_OR_ZERO(task);
790}
791
792static int btrfs_dev_replace_kthread(void *data)
793{
794 struct btrfs_fs_info *fs_info = data;
795 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
796 struct btrfs_ioctl_dev_replace_args *status_args;
797 u64 progress;
798
799 status_args = kzalloc(sizeof(*status_args), GFP_KERNEL);
800 if (status_args) {
801 btrfs_dev_replace_status(fs_info, status_args);
802 progress = status_args->status.progress_1000;
803 kfree(status_args);
804 progress = div_u64(progress, 10);
805 btrfs_info_in_rcu(fs_info,
806 "continuing dev_replace from %s (devid %llu) to %s @%u%%",
807 dev_replace->srcdev->missing ? "<missing disk>" :
808 rcu_str_deref(dev_replace->srcdev->name),
809 dev_replace->srcdev->devid,
810 dev_replace->tgtdev ?
811 rcu_str_deref(dev_replace->tgtdev->name) :
812 "<missing target disk>",
813 (unsigned int)progress);
814 }
815 btrfs_dev_replace_continue_on_mount(fs_info);
816 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
817
818 return 0;
819}
820
821static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info)
822{
823 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
824 int ret;
825
826 ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
827 dev_replace->committed_cursor_left,
828 btrfs_device_get_total_bytes(dev_replace->srcdev),
829 &dev_replace->scrub_progress, 0, 1);
830 ret = btrfs_dev_replace_finishing(fs_info, ret);
831 WARN_ON(ret);
832 return 0;
833}
834
835int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
836{
837 if (!dev_replace->is_valid)
838 return 0;
839
840 switch (dev_replace->replace_state) {
841 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
842 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
843 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
844 return 0;
845 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
846 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
847 /*
848 * return true even if tgtdev is missing (this is
849 * something that can happen if the dev_replace
850 * procedure is suspended by an umount and then
851 * the tgtdev is missing (or "btrfs dev scan") was
852 * not called and the the filesystem is remounted
853 * in degraded state. This does not stop the
854 * dev_replace procedure. It needs to be canceled
855 * manually if the cancellation is wanted.
856 */
857 break;
858 }
859 return 1;
860}
861
862void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace, int rw)
863{
864 if (rw == 1) {
865 /* write */
866again:
867 wait_event(dev_replace->read_lock_wq,
868 atomic_read(&dev_replace->blocking_readers) == 0);
869 write_lock(&dev_replace->lock);
870 if (atomic_read(&dev_replace->blocking_readers)) {
871 write_unlock(&dev_replace->lock);
872 goto again;
873 }
874 } else {
875 read_lock(&dev_replace->lock);
876 atomic_inc(&dev_replace->read_locks);
877 }
878}
879
880void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace, int rw)
881{
882 if (rw == 1) {
883 /* write */
884 ASSERT(atomic_read(&dev_replace->blocking_readers) == 0);
885 write_unlock(&dev_replace->lock);
886 } else {
887 ASSERT(atomic_read(&dev_replace->read_locks) > 0);
888 atomic_dec(&dev_replace->read_locks);
889 read_unlock(&dev_replace->lock);
890 }
891}
892
893/* inc blocking cnt and release read lock */
894void btrfs_dev_replace_set_lock_blocking(
895 struct btrfs_dev_replace *dev_replace)
896{
897 /* only set blocking for read lock */
898 ASSERT(atomic_read(&dev_replace->read_locks) > 0);
899 atomic_inc(&dev_replace->blocking_readers);
900 read_unlock(&dev_replace->lock);
901}
902
903/* acquire read lock and dec blocking cnt */
904void btrfs_dev_replace_clear_lock_blocking(
905 struct btrfs_dev_replace *dev_replace)
906{
907 /* only set blocking for read lock */
908 ASSERT(atomic_read(&dev_replace->read_locks) > 0);
909 ASSERT(atomic_read(&dev_replace->blocking_readers) > 0);
910 read_lock(&dev_replace->lock);
911 if (atomic_dec_and_test(&dev_replace->blocking_readers) &&
912 waitqueue_active(&dev_replace->read_lock_wq))
913 wake_up(&dev_replace->read_lock_wq);
914}
915
916void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
917{
918 percpu_counter_inc(&fs_info->bio_counter);
919}
920
921void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
922{
923 percpu_counter_sub(&fs_info->bio_counter, amount);
924
925 if (waitqueue_active(&fs_info->replace_wait))
926 wake_up(&fs_info->replace_wait);
927}
928
929void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
930{
931 while (1) {
932 percpu_counter_inc(&fs_info->bio_counter);
933 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
934 &fs_info->fs_state)))
935 break;
936
937 btrfs_bio_counter_dec(fs_info);
938 wait_event(fs_info->replace_wait,
939 !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
940 &fs_info->fs_state));
941 }
942}