xref: /linux/fs/btrfs/dev-replace.c (revision 3f0a50f345f78183f6e9b39c2f45ca5dcaa511ca)
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 #include "zoned.h"
25 #include "block-group.h"
26 
27 /*
28  * Device replace overview
29  *
30  * [Objective]
31  * To copy all extents (both new and on-disk) from source device to target
32  * device, while still keeping the filesystem read-write.
33  *
34  * [Method]
35  * There are two main methods involved:
36  *
37  * - Write duplication
38  *
39  *   All new writes will be written to both target and source devices, so even
40  *   if replace gets canceled, sources device still contains up-to-date data.
41  *
42  *   Location:		handle_ops_on_dev_replace() from __btrfs_map_block()
43  *   Start:		btrfs_dev_replace_start()
44  *   End:		btrfs_dev_replace_finishing()
45  *   Content:		Latest data/metadata
46  *
47  * - Copy existing extents
48  *
49  *   This happens by re-using scrub facility, as scrub also iterates through
50  *   existing extents from commit root.
51  *
52  *   Location:		scrub_write_block_to_dev_replace() from
53  *   			scrub_block_complete()
54  *   Content:		Data/meta from commit root.
55  *
56  * Due to the content difference, we need to avoid nocow write when dev-replace
57  * is happening.  This is done by marking the block group read-only and waiting
58  * for NOCOW writes.
59  *
60  * After replace is done, the finishing part is done by swapping the target and
61  * source devices.
62  *
63  *   Location:		btrfs_dev_replace_update_device_in_mapping_tree() from
64  *   			btrfs_dev_replace_finishing()
65  */
66 
67 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
68 				       int scrub_ret);
69 static int btrfs_dev_replace_kthread(void *data);
70 
71 int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
72 {
73 	struct btrfs_dev_lookup_args args = { .devid = BTRFS_DEV_REPLACE_DEVID };
74 	struct btrfs_key key;
75 	struct btrfs_root *dev_root = fs_info->dev_root;
76 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
77 	struct extent_buffer *eb;
78 	int slot;
79 	int ret = 0;
80 	struct btrfs_path *path = NULL;
81 	int item_size;
82 	struct btrfs_dev_replace_item *ptr;
83 	u64 src_devid;
84 
85 	if (!dev_root)
86 		return 0;
87 
88 	path = btrfs_alloc_path();
89 	if (!path) {
90 		ret = -ENOMEM;
91 		goto out;
92 	}
93 
94 	key.objectid = 0;
95 	key.type = BTRFS_DEV_REPLACE_KEY;
96 	key.offset = 0;
97 	ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
98 	if (ret) {
99 no_valid_dev_replace_entry_found:
100 		/*
101 		 * We don't have a replace item or it's corrupted.  If there is
102 		 * a replace target, fail the mount.
103 		 */
104 		if (btrfs_find_device(fs_info->fs_devices, &args)) {
105 			btrfs_err(fs_info,
106 			"found replace target device without a valid replace item");
107 			ret = -EUCLEAN;
108 			goto out;
109 		}
110 		ret = 0;
111 		dev_replace->replace_state =
112 			BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
113 		dev_replace->cont_reading_from_srcdev_mode =
114 		    BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
115 		dev_replace->time_started = 0;
116 		dev_replace->time_stopped = 0;
117 		atomic64_set(&dev_replace->num_write_errors, 0);
118 		atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
119 		dev_replace->cursor_left = 0;
120 		dev_replace->committed_cursor_left = 0;
121 		dev_replace->cursor_left_last_write_of_item = 0;
122 		dev_replace->cursor_right = 0;
123 		dev_replace->srcdev = NULL;
124 		dev_replace->tgtdev = NULL;
125 		dev_replace->is_valid = 0;
126 		dev_replace->item_needs_writeback = 0;
127 		goto out;
128 	}
129 	slot = path->slots[0];
130 	eb = path->nodes[0];
131 	item_size = btrfs_item_size(eb, slot);
132 	ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
133 
134 	if (item_size != sizeof(struct btrfs_dev_replace_item)) {
135 		btrfs_warn(fs_info,
136 			"dev_replace entry found has unexpected size, ignore entry");
137 		goto no_valid_dev_replace_entry_found;
138 	}
139 
140 	src_devid = btrfs_dev_replace_src_devid(eb, ptr);
141 	dev_replace->cont_reading_from_srcdev_mode =
142 		btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
143 	dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
144 	dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
145 	dev_replace->time_stopped =
146 		btrfs_dev_replace_time_stopped(eb, ptr);
147 	atomic64_set(&dev_replace->num_write_errors,
148 		     btrfs_dev_replace_num_write_errors(eb, ptr));
149 	atomic64_set(&dev_replace->num_uncorrectable_read_errors,
150 		     btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
151 	dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
152 	dev_replace->committed_cursor_left = dev_replace->cursor_left;
153 	dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
154 	dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
155 	dev_replace->is_valid = 1;
156 
157 	dev_replace->item_needs_writeback = 0;
158 	switch (dev_replace->replace_state) {
159 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
160 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
161 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
162 		/*
163 		 * We don't have an active replace item but if there is a
164 		 * replace target, fail the mount.
165 		 */
166 		if (btrfs_find_device(fs_info->fs_devices, &args)) {
167 			btrfs_err(fs_info,
168 			"replace devid present without an active replace item");
169 			ret = -EUCLEAN;
170 		} else {
171 			dev_replace->srcdev = NULL;
172 			dev_replace->tgtdev = NULL;
173 		}
174 		break;
175 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
176 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
177 		dev_replace->tgtdev = btrfs_find_device(fs_info->fs_devices, &args);
178 		args.devid = src_devid;
179 		dev_replace->srcdev = btrfs_find_device(fs_info->fs_devices, &args);
180 
181 		/*
182 		 * allow 'btrfs dev replace_cancel' if src/tgt device is
183 		 * missing
184 		 */
185 		if (!dev_replace->srcdev &&
186 		    !btrfs_test_opt(fs_info, DEGRADED)) {
187 			ret = -EIO;
188 			btrfs_warn(fs_info,
189 			   "cannot mount because device replace operation is ongoing and");
190 			btrfs_warn(fs_info,
191 			   "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
192 			   src_devid);
193 		}
194 		if (!dev_replace->tgtdev &&
195 		    !btrfs_test_opt(fs_info, DEGRADED)) {
196 			ret = -EIO;
197 			btrfs_warn(fs_info,
198 			   "cannot mount because device replace operation is ongoing and");
199 			btrfs_warn(fs_info,
200 			   "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
201 				BTRFS_DEV_REPLACE_DEVID);
202 		}
203 		if (dev_replace->tgtdev) {
204 			if (dev_replace->srcdev) {
205 				dev_replace->tgtdev->total_bytes =
206 					dev_replace->srcdev->total_bytes;
207 				dev_replace->tgtdev->disk_total_bytes =
208 					dev_replace->srcdev->disk_total_bytes;
209 				dev_replace->tgtdev->commit_total_bytes =
210 					dev_replace->srcdev->commit_total_bytes;
211 				dev_replace->tgtdev->bytes_used =
212 					dev_replace->srcdev->bytes_used;
213 				dev_replace->tgtdev->commit_bytes_used =
214 					dev_replace->srcdev->commit_bytes_used;
215 			}
216 			set_bit(BTRFS_DEV_STATE_REPLACE_TGT,
217 				&dev_replace->tgtdev->dev_state);
218 
219 			WARN_ON(fs_info->fs_devices->rw_devices == 0);
220 			dev_replace->tgtdev->io_width = fs_info->sectorsize;
221 			dev_replace->tgtdev->io_align = fs_info->sectorsize;
222 			dev_replace->tgtdev->sector_size = fs_info->sectorsize;
223 			dev_replace->tgtdev->fs_info = fs_info;
224 			set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
225 				&dev_replace->tgtdev->dev_state);
226 		}
227 		break;
228 	}
229 
230 out:
231 	btrfs_free_path(path);
232 	return ret;
233 }
234 
235 /*
236  * Initialize a new device for device replace target from a given source dev
237  * and path.
238  *
239  * Return 0 and new device in @device_out, otherwise return < 0
240  */
241 static int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
242 				  const char *device_path,
243 				  struct btrfs_device *srcdev,
244 				  struct btrfs_device **device_out)
245 {
246 	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
247 	struct btrfs_device *device;
248 	struct block_device *bdev;
249 	struct rcu_string *name;
250 	u64 devid = BTRFS_DEV_REPLACE_DEVID;
251 	int ret = 0;
252 
253 	*device_out = NULL;
254 	if (srcdev->fs_devices->seeding) {
255 		btrfs_err(fs_info, "the filesystem is a seed filesystem!");
256 		return -EINVAL;
257 	}
258 
259 	bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
260 				  fs_info->bdev_holder);
261 	if (IS_ERR(bdev)) {
262 		btrfs_err(fs_info, "target device %s is invalid!", device_path);
263 		return PTR_ERR(bdev);
264 	}
265 
266 	if (!btrfs_check_device_zone_type(fs_info, bdev)) {
267 		btrfs_err(fs_info,
268 		"dev-replace: zoned type of target device mismatch with filesystem");
269 		ret = -EINVAL;
270 		goto error;
271 	}
272 
273 	sync_blockdev(bdev);
274 
275 	list_for_each_entry(device, &fs_devices->devices, dev_list) {
276 		if (device->bdev == bdev) {
277 			btrfs_err(fs_info,
278 				  "target device is in the filesystem!");
279 			ret = -EEXIST;
280 			goto error;
281 		}
282 	}
283 
284 
285 	if (bdev_nr_bytes(bdev) < btrfs_device_get_total_bytes(srcdev)) {
286 		btrfs_err(fs_info,
287 			  "target device is smaller than source device!");
288 		ret = -EINVAL;
289 		goto error;
290 	}
291 
292 
293 	device = btrfs_alloc_device(NULL, &devid, NULL);
294 	if (IS_ERR(device)) {
295 		ret = PTR_ERR(device);
296 		goto error;
297 	}
298 
299 	name = rcu_string_strdup(device_path, GFP_KERNEL);
300 	if (!name) {
301 		btrfs_free_device(device);
302 		ret = -ENOMEM;
303 		goto error;
304 	}
305 	rcu_assign_pointer(device->name, name);
306 	ret = lookup_bdev(device_path, &device->devt);
307 	if (ret)
308 		goto error;
309 
310 	set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
311 	device->generation = 0;
312 	device->io_width = fs_info->sectorsize;
313 	device->io_align = fs_info->sectorsize;
314 	device->sector_size = fs_info->sectorsize;
315 	device->total_bytes = btrfs_device_get_total_bytes(srcdev);
316 	device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
317 	device->bytes_used = btrfs_device_get_bytes_used(srcdev);
318 	device->commit_total_bytes = srcdev->commit_total_bytes;
319 	device->commit_bytes_used = device->bytes_used;
320 	device->fs_info = fs_info;
321 	device->bdev = bdev;
322 	set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
323 	set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
324 	device->mode = FMODE_EXCL;
325 	device->dev_stats_valid = 1;
326 	set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
327 	device->fs_devices = fs_devices;
328 
329 	ret = btrfs_get_dev_zone_info(device, false);
330 	if (ret)
331 		goto error;
332 
333 	mutex_lock(&fs_devices->device_list_mutex);
334 	list_add(&device->dev_list, &fs_devices->devices);
335 	fs_devices->num_devices++;
336 	fs_devices->open_devices++;
337 	mutex_unlock(&fs_devices->device_list_mutex);
338 
339 	*device_out = device;
340 	return 0;
341 
342 error:
343 	blkdev_put(bdev, FMODE_EXCL);
344 	return ret;
345 }
346 
347 /*
348  * called from commit_transaction. Writes changed device replace state to
349  * disk.
350  */
351 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans)
352 {
353 	struct btrfs_fs_info *fs_info = trans->fs_info;
354 	int ret;
355 	struct btrfs_root *dev_root = fs_info->dev_root;
356 	struct btrfs_path *path;
357 	struct btrfs_key key;
358 	struct extent_buffer *eb;
359 	struct btrfs_dev_replace_item *ptr;
360 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
361 
362 	down_read(&dev_replace->rwsem);
363 	if (!dev_replace->is_valid ||
364 	    !dev_replace->item_needs_writeback) {
365 		up_read(&dev_replace->rwsem);
366 		return 0;
367 	}
368 	up_read(&dev_replace->rwsem);
369 
370 	key.objectid = 0;
371 	key.type = BTRFS_DEV_REPLACE_KEY;
372 	key.offset = 0;
373 
374 	path = btrfs_alloc_path();
375 	if (!path) {
376 		ret = -ENOMEM;
377 		goto out;
378 	}
379 	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
380 	if (ret < 0) {
381 		btrfs_warn(fs_info,
382 			   "error %d while searching for dev_replace item!",
383 			   ret);
384 		goto out;
385 	}
386 
387 	if (ret == 0 &&
388 	    btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
389 		/*
390 		 * need to delete old one and insert a new one.
391 		 * Since no attempt is made to recover any old state, if the
392 		 * dev_replace state is 'running', the data on the target
393 		 * drive is lost.
394 		 * It would be possible to recover the state: just make sure
395 		 * that the beginning of the item is never changed and always
396 		 * contains all the essential information. Then read this
397 		 * minimal set of information and use it as a base for the
398 		 * new state.
399 		 */
400 		ret = btrfs_del_item(trans, dev_root, path);
401 		if (ret != 0) {
402 			btrfs_warn(fs_info,
403 				   "delete too small dev_replace item failed %d!",
404 				   ret);
405 			goto out;
406 		}
407 		ret = 1;
408 	}
409 
410 	if (ret == 1) {
411 		/* need to insert a new item */
412 		btrfs_release_path(path);
413 		ret = btrfs_insert_empty_item(trans, dev_root, path,
414 					      &key, sizeof(*ptr));
415 		if (ret < 0) {
416 			btrfs_warn(fs_info,
417 				   "insert dev_replace item failed %d!", ret);
418 			goto out;
419 		}
420 	}
421 
422 	eb = path->nodes[0];
423 	ptr = btrfs_item_ptr(eb, path->slots[0],
424 			     struct btrfs_dev_replace_item);
425 
426 	down_write(&dev_replace->rwsem);
427 	if (dev_replace->srcdev)
428 		btrfs_set_dev_replace_src_devid(eb, ptr,
429 			dev_replace->srcdev->devid);
430 	else
431 		btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
432 	btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
433 		dev_replace->cont_reading_from_srcdev_mode);
434 	btrfs_set_dev_replace_replace_state(eb, ptr,
435 		dev_replace->replace_state);
436 	btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
437 	btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
438 	btrfs_set_dev_replace_num_write_errors(eb, ptr,
439 		atomic64_read(&dev_replace->num_write_errors));
440 	btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
441 		atomic64_read(&dev_replace->num_uncorrectable_read_errors));
442 	dev_replace->cursor_left_last_write_of_item =
443 		dev_replace->cursor_left;
444 	btrfs_set_dev_replace_cursor_left(eb, ptr,
445 		dev_replace->cursor_left_last_write_of_item);
446 	btrfs_set_dev_replace_cursor_right(eb, ptr,
447 		dev_replace->cursor_right);
448 	dev_replace->item_needs_writeback = 0;
449 	up_write(&dev_replace->rwsem);
450 
451 	btrfs_mark_buffer_dirty(eb);
452 
453 out:
454 	btrfs_free_path(path);
455 
456 	return ret;
457 }
458 
459 static char* btrfs_dev_name(struct btrfs_device *device)
460 {
461 	if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
462 		return "<missing disk>";
463 	else
464 		return rcu_str_deref(device->name);
465 }
466 
467 static int mark_block_group_to_copy(struct btrfs_fs_info *fs_info,
468 				    struct btrfs_device *src_dev)
469 {
470 	struct btrfs_path *path;
471 	struct btrfs_key key;
472 	struct btrfs_key found_key;
473 	struct btrfs_root *root = fs_info->dev_root;
474 	struct btrfs_dev_extent *dev_extent = NULL;
475 	struct btrfs_block_group *cache;
476 	struct btrfs_trans_handle *trans;
477 	int ret = 0;
478 	u64 chunk_offset;
479 
480 	/* Do not use "to_copy" on non zoned filesystem for now */
481 	if (!btrfs_is_zoned(fs_info))
482 		return 0;
483 
484 	mutex_lock(&fs_info->chunk_mutex);
485 
486 	/* Ensure we don't have pending new block group */
487 	spin_lock(&fs_info->trans_lock);
488 	while (fs_info->running_transaction &&
489 	       !list_empty(&fs_info->running_transaction->dev_update_list)) {
490 		spin_unlock(&fs_info->trans_lock);
491 		mutex_unlock(&fs_info->chunk_mutex);
492 		trans = btrfs_attach_transaction(root);
493 		if (IS_ERR(trans)) {
494 			ret = PTR_ERR(trans);
495 			mutex_lock(&fs_info->chunk_mutex);
496 			if (ret == -ENOENT) {
497 				spin_lock(&fs_info->trans_lock);
498 				continue;
499 			} else {
500 				goto unlock;
501 			}
502 		}
503 
504 		ret = btrfs_commit_transaction(trans);
505 		mutex_lock(&fs_info->chunk_mutex);
506 		if (ret)
507 			goto unlock;
508 
509 		spin_lock(&fs_info->trans_lock);
510 	}
511 	spin_unlock(&fs_info->trans_lock);
512 
513 	path = btrfs_alloc_path();
514 	if (!path) {
515 		ret = -ENOMEM;
516 		goto unlock;
517 	}
518 
519 	path->reada = READA_FORWARD;
520 	path->search_commit_root = 1;
521 	path->skip_locking = 1;
522 
523 	key.objectid = src_dev->devid;
524 	key.type = BTRFS_DEV_EXTENT_KEY;
525 	key.offset = 0;
526 
527 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
528 	if (ret < 0)
529 		goto free_path;
530 	if (ret > 0) {
531 		if (path->slots[0] >=
532 		    btrfs_header_nritems(path->nodes[0])) {
533 			ret = btrfs_next_leaf(root, path);
534 			if (ret < 0)
535 				goto free_path;
536 			if (ret > 0) {
537 				ret = 0;
538 				goto free_path;
539 			}
540 		} else {
541 			ret = 0;
542 		}
543 	}
544 
545 	while (1) {
546 		struct extent_buffer *leaf = path->nodes[0];
547 		int slot = path->slots[0];
548 
549 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
550 
551 		if (found_key.objectid != src_dev->devid)
552 			break;
553 
554 		if (found_key.type != BTRFS_DEV_EXTENT_KEY)
555 			break;
556 
557 		if (found_key.offset < key.offset)
558 			break;
559 
560 		dev_extent = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
561 
562 		chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dev_extent);
563 
564 		cache = btrfs_lookup_block_group(fs_info, chunk_offset);
565 		if (!cache)
566 			goto skip;
567 
568 		spin_lock(&cache->lock);
569 		cache->to_copy = 1;
570 		spin_unlock(&cache->lock);
571 
572 		btrfs_put_block_group(cache);
573 
574 skip:
575 		ret = btrfs_next_item(root, path);
576 		if (ret != 0) {
577 			if (ret > 0)
578 				ret = 0;
579 			break;
580 		}
581 	}
582 
583 free_path:
584 	btrfs_free_path(path);
585 unlock:
586 	mutex_unlock(&fs_info->chunk_mutex);
587 
588 	return ret;
589 }
590 
591 bool btrfs_finish_block_group_to_copy(struct btrfs_device *srcdev,
592 				      struct btrfs_block_group *cache,
593 				      u64 physical)
594 {
595 	struct btrfs_fs_info *fs_info = cache->fs_info;
596 	struct extent_map *em;
597 	struct map_lookup *map;
598 	u64 chunk_offset = cache->start;
599 	int num_extents, cur_extent;
600 	int i;
601 
602 	/* Do not use "to_copy" on non zoned filesystem for now */
603 	if (!btrfs_is_zoned(fs_info))
604 		return true;
605 
606 	spin_lock(&cache->lock);
607 	if (cache->removed) {
608 		spin_unlock(&cache->lock);
609 		return true;
610 	}
611 	spin_unlock(&cache->lock);
612 
613 	em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
614 	ASSERT(!IS_ERR(em));
615 	map = em->map_lookup;
616 
617 	num_extents = cur_extent = 0;
618 	for (i = 0; i < map->num_stripes; i++) {
619 		/* We have more device extent to copy */
620 		if (srcdev != map->stripes[i].dev)
621 			continue;
622 
623 		num_extents++;
624 		if (physical == map->stripes[i].physical)
625 			cur_extent = i;
626 	}
627 
628 	free_extent_map(em);
629 
630 	if (num_extents > 1 && cur_extent < num_extents - 1) {
631 		/*
632 		 * Has more stripes on this device. Keep this block group
633 		 * readonly until we finish all the stripes.
634 		 */
635 		return false;
636 	}
637 
638 	/* Last stripe on this device */
639 	spin_lock(&cache->lock);
640 	cache->to_copy = 0;
641 	spin_unlock(&cache->lock);
642 
643 	return true;
644 }
645 
646 static int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
647 		const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
648 		int read_src)
649 {
650 	struct btrfs_root *root = fs_info->dev_root;
651 	struct btrfs_trans_handle *trans;
652 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
653 	int ret;
654 	struct btrfs_device *tgt_device = NULL;
655 	struct btrfs_device *src_device = NULL;
656 
657 	src_device = btrfs_find_device_by_devspec(fs_info, srcdevid,
658 						  srcdev_name);
659 	if (IS_ERR(src_device))
660 		return PTR_ERR(src_device);
661 
662 	if (btrfs_pinned_by_swapfile(fs_info, src_device)) {
663 		btrfs_warn_in_rcu(fs_info,
664 	  "cannot replace device %s (devid %llu) due to active swapfile",
665 			btrfs_dev_name(src_device), src_device->devid);
666 		return -ETXTBSY;
667 	}
668 
669 	/*
670 	 * Here we commit the transaction to make sure commit_total_bytes
671 	 * of all the devices are updated.
672 	 */
673 	trans = btrfs_attach_transaction(root);
674 	if (!IS_ERR(trans)) {
675 		ret = btrfs_commit_transaction(trans);
676 		if (ret)
677 			return ret;
678 	} else if (PTR_ERR(trans) != -ENOENT) {
679 		return PTR_ERR(trans);
680 	}
681 
682 	ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
683 					    src_device, &tgt_device);
684 	if (ret)
685 		return ret;
686 
687 	ret = mark_block_group_to_copy(fs_info, src_device);
688 	if (ret)
689 		return ret;
690 
691 	down_write(&dev_replace->rwsem);
692 	switch (dev_replace->replace_state) {
693 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
694 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
695 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
696 		break;
697 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
698 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
699 		ASSERT(0);
700 		ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
701 		up_write(&dev_replace->rwsem);
702 		goto leave;
703 	}
704 
705 	dev_replace->cont_reading_from_srcdev_mode = read_src;
706 	dev_replace->srcdev = src_device;
707 	dev_replace->tgtdev = tgt_device;
708 
709 	btrfs_info_in_rcu(fs_info,
710 		      "dev_replace from %s (devid %llu) to %s started",
711 		      btrfs_dev_name(src_device),
712 		      src_device->devid,
713 		      rcu_str_deref(tgt_device->name));
714 
715 	/*
716 	 * from now on, the writes to the srcdev are all duplicated to
717 	 * go to the tgtdev as well (refer to btrfs_map_block()).
718 	 */
719 	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
720 	dev_replace->time_started = ktime_get_real_seconds();
721 	dev_replace->cursor_left = 0;
722 	dev_replace->committed_cursor_left = 0;
723 	dev_replace->cursor_left_last_write_of_item = 0;
724 	dev_replace->cursor_right = 0;
725 	dev_replace->is_valid = 1;
726 	dev_replace->item_needs_writeback = 1;
727 	atomic64_set(&dev_replace->num_write_errors, 0);
728 	atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
729 	up_write(&dev_replace->rwsem);
730 
731 	ret = btrfs_sysfs_add_device(tgt_device);
732 	if (ret)
733 		btrfs_err(fs_info, "kobj add dev failed %d", ret);
734 
735 	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
736 
737 	/* Commit dev_replace state and reserve 1 item for it. */
738 	trans = btrfs_start_transaction(root, 1);
739 	if (IS_ERR(trans)) {
740 		ret = PTR_ERR(trans);
741 		down_write(&dev_replace->rwsem);
742 		dev_replace->replace_state =
743 			BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
744 		dev_replace->srcdev = NULL;
745 		dev_replace->tgtdev = NULL;
746 		up_write(&dev_replace->rwsem);
747 		goto leave;
748 	}
749 
750 	ret = btrfs_commit_transaction(trans);
751 	WARN_ON(ret);
752 
753 	/* the disk copy procedure reuses the scrub code */
754 	ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
755 			      btrfs_device_get_total_bytes(src_device),
756 			      &dev_replace->scrub_progress, 0, 1);
757 
758 	ret = btrfs_dev_replace_finishing(fs_info, ret);
759 	if (ret == -EINPROGRESS)
760 		ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
761 
762 	return ret;
763 
764 leave:
765 	btrfs_destroy_dev_replace_tgtdev(tgt_device);
766 	return ret;
767 }
768 
769 int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
770 			    struct btrfs_ioctl_dev_replace_args *args)
771 {
772 	int ret;
773 
774 	switch (args->start.cont_reading_from_srcdev_mode) {
775 	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
776 	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
777 		break;
778 	default:
779 		return -EINVAL;
780 	}
781 
782 	if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
783 	    args->start.tgtdev_name[0] == '\0')
784 		return -EINVAL;
785 
786 	ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name,
787 					args->start.srcdevid,
788 					args->start.srcdev_name,
789 					args->start.cont_reading_from_srcdev_mode);
790 	args->result = ret;
791 	/* don't warn if EINPROGRESS, someone else might be running scrub */
792 	if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS ||
793 	    ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR)
794 		return 0;
795 
796 	return ret;
797 }
798 
799 /*
800  * blocked until all in-flight bios operations are finished.
801  */
802 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
803 {
804 	set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
805 	wait_event(fs_info->dev_replace.replace_wait, !percpu_counter_sum(
806 		   &fs_info->dev_replace.bio_counter));
807 }
808 
809 /*
810  * we have removed target device, it is safe to allow new bios request.
811  */
812 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
813 {
814 	clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
815 	wake_up(&fs_info->dev_replace.replace_wait);
816 }
817 
818 /*
819  * When finishing the device replace, before swapping the source device with the
820  * target device we must update the chunk allocation state in the target device,
821  * as it is empty because replace works by directly copying the chunks and not
822  * through the normal chunk allocation path.
823  */
824 static int btrfs_set_target_alloc_state(struct btrfs_device *srcdev,
825 					struct btrfs_device *tgtdev)
826 {
827 	struct extent_state *cached_state = NULL;
828 	u64 start = 0;
829 	u64 found_start;
830 	u64 found_end;
831 	int ret = 0;
832 
833 	lockdep_assert_held(&srcdev->fs_info->chunk_mutex);
834 
835 	while (!find_first_extent_bit(&srcdev->alloc_state, start,
836 				      &found_start, &found_end,
837 				      CHUNK_ALLOCATED, &cached_state)) {
838 		ret = set_extent_bits(&tgtdev->alloc_state, found_start,
839 				      found_end, CHUNK_ALLOCATED);
840 		if (ret)
841 			break;
842 		start = found_end + 1;
843 	}
844 
845 	free_extent_state(cached_state);
846 	return ret;
847 }
848 
849 static void btrfs_dev_replace_update_device_in_mapping_tree(
850 						struct btrfs_fs_info *fs_info,
851 						struct btrfs_device *srcdev,
852 						struct btrfs_device *tgtdev)
853 {
854 	struct extent_map_tree *em_tree = &fs_info->mapping_tree;
855 	struct extent_map *em;
856 	struct map_lookup *map;
857 	u64 start = 0;
858 	int i;
859 
860 	write_lock(&em_tree->lock);
861 	do {
862 		em = lookup_extent_mapping(em_tree, start, (u64)-1);
863 		if (!em)
864 			break;
865 		map = em->map_lookup;
866 		for (i = 0; i < map->num_stripes; i++)
867 			if (srcdev == map->stripes[i].dev)
868 				map->stripes[i].dev = tgtdev;
869 		start = em->start + em->len;
870 		free_extent_map(em);
871 	} while (start);
872 	write_unlock(&em_tree->lock);
873 }
874 
875 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
876 				       int scrub_ret)
877 {
878 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
879 	struct btrfs_device *tgt_device;
880 	struct btrfs_device *src_device;
881 	struct btrfs_root *root = fs_info->tree_root;
882 	u8 uuid_tmp[BTRFS_UUID_SIZE];
883 	struct btrfs_trans_handle *trans;
884 	int ret = 0;
885 
886 	/* don't allow cancel or unmount to disturb the finishing procedure */
887 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
888 
889 	down_read(&dev_replace->rwsem);
890 	/* was the operation canceled, or is it finished? */
891 	if (dev_replace->replace_state !=
892 	    BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
893 		up_read(&dev_replace->rwsem);
894 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
895 		return 0;
896 	}
897 
898 	tgt_device = dev_replace->tgtdev;
899 	src_device = dev_replace->srcdev;
900 	up_read(&dev_replace->rwsem);
901 
902 	/*
903 	 * flush all outstanding I/O and inode extent mappings before the
904 	 * copy operation is declared as being finished
905 	 */
906 	ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
907 	if (ret) {
908 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
909 		return ret;
910 	}
911 	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
912 
913 	/*
914 	 * We have to use this loop approach because at this point src_device
915 	 * has to be available for transaction commit to complete, yet new
916 	 * chunks shouldn't be allocated on the device.
917 	 */
918 	while (1) {
919 		trans = btrfs_start_transaction(root, 0);
920 		if (IS_ERR(trans)) {
921 			mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
922 			return PTR_ERR(trans);
923 		}
924 		ret = btrfs_commit_transaction(trans);
925 		WARN_ON(ret);
926 
927 		/* Prevent write_all_supers() during the finishing procedure */
928 		mutex_lock(&fs_info->fs_devices->device_list_mutex);
929 		/* Prevent new chunks being allocated on the source device */
930 		mutex_lock(&fs_info->chunk_mutex);
931 
932 		if (!list_empty(&src_device->post_commit_list)) {
933 			mutex_unlock(&fs_info->fs_devices->device_list_mutex);
934 			mutex_unlock(&fs_info->chunk_mutex);
935 		} else {
936 			break;
937 		}
938 	}
939 
940 	down_write(&dev_replace->rwsem);
941 	dev_replace->replace_state =
942 		scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
943 			  : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
944 	dev_replace->tgtdev = NULL;
945 	dev_replace->srcdev = NULL;
946 	dev_replace->time_stopped = ktime_get_real_seconds();
947 	dev_replace->item_needs_writeback = 1;
948 
949 	/*
950 	 * Update allocation state in the new device and replace the old device
951 	 * with the new one in the mapping tree.
952 	 */
953 	if (!scrub_ret) {
954 		scrub_ret = btrfs_set_target_alloc_state(src_device, tgt_device);
955 		if (scrub_ret)
956 			goto error;
957 		btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
958 								src_device,
959 								tgt_device);
960 	} else {
961 		if (scrub_ret != -ECANCELED)
962 			btrfs_err_in_rcu(fs_info,
963 				 "btrfs_scrub_dev(%s, %llu, %s) failed %d",
964 				 btrfs_dev_name(src_device),
965 				 src_device->devid,
966 				 rcu_str_deref(tgt_device->name), scrub_ret);
967 error:
968 		up_write(&dev_replace->rwsem);
969 		mutex_unlock(&fs_info->chunk_mutex);
970 		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
971 		btrfs_rm_dev_replace_blocked(fs_info);
972 		if (tgt_device)
973 			btrfs_destroy_dev_replace_tgtdev(tgt_device);
974 		btrfs_rm_dev_replace_unblocked(fs_info);
975 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
976 
977 		return scrub_ret;
978 	}
979 
980 	btrfs_info_in_rcu(fs_info,
981 			  "dev_replace from %s (devid %llu) to %s finished",
982 			  btrfs_dev_name(src_device),
983 			  src_device->devid,
984 			  rcu_str_deref(tgt_device->name));
985 	clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state);
986 	tgt_device->devid = src_device->devid;
987 	src_device->devid = BTRFS_DEV_REPLACE_DEVID;
988 	memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
989 	memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
990 	memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
991 	btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
992 	btrfs_device_set_disk_total_bytes(tgt_device,
993 					  src_device->disk_total_bytes);
994 	btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
995 	tgt_device->commit_bytes_used = src_device->bytes_used;
996 
997 	btrfs_assign_next_active_device(src_device, tgt_device);
998 
999 	list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
1000 	fs_info->fs_devices->rw_devices++;
1001 
1002 	up_write(&dev_replace->rwsem);
1003 	btrfs_rm_dev_replace_blocked(fs_info);
1004 
1005 	btrfs_rm_dev_replace_remove_srcdev(src_device);
1006 
1007 	btrfs_rm_dev_replace_unblocked(fs_info);
1008 
1009 	/*
1010 	 * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will
1011 	 * update on-disk dev stats value during commit transaction
1012 	 */
1013 	atomic_inc(&tgt_device->dev_stats_ccnt);
1014 
1015 	/*
1016 	 * this is again a consistent state where no dev_replace procedure
1017 	 * is running, the target device is part of the filesystem, the
1018 	 * source device is not part of the filesystem anymore and its 1st
1019 	 * superblock is scratched out so that it is no longer marked to
1020 	 * belong to this filesystem.
1021 	 */
1022 	mutex_unlock(&fs_info->chunk_mutex);
1023 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1024 
1025 	/* replace the sysfs entry */
1026 	btrfs_sysfs_remove_device(src_device);
1027 	btrfs_sysfs_update_devid(tgt_device);
1028 	if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &src_device->dev_state))
1029 		btrfs_scratch_superblocks(fs_info, src_device->bdev,
1030 					  src_device->name->str);
1031 
1032 	/* write back the superblocks */
1033 	trans = btrfs_start_transaction(root, 0);
1034 	if (!IS_ERR(trans))
1035 		btrfs_commit_transaction(trans);
1036 
1037 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
1038 
1039 	btrfs_rm_dev_replace_free_srcdev(src_device);
1040 
1041 	return 0;
1042 }
1043 
1044 /*
1045  * Read progress of device replace status according to the state and last
1046  * stored position. The value format is the same as for
1047  * btrfs_dev_replace::progress_1000
1048  */
1049 static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info)
1050 {
1051 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1052 	u64 ret = 0;
1053 
1054 	switch (dev_replace->replace_state) {
1055 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1056 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1057 		ret = 0;
1058 		break;
1059 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1060 		ret = 1000;
1061 		break;
1062 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1063 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1064 		ret = div64_u64(dev_replace->cursor_left,
1065 				div_u64(btrfs_device_get_total_bytes(
1066 						dev_replace->srcdev), 1000));
1067 		break;
1068 	}
1069 
1070 	return ret;
1071 }
1072 
1073 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
1074 			      struct btrfs_ioctl_dev_replace_args *args)
1075 {
1076 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1077 
1078 	down_read(&dev_replace->rwsem);
1079 	/* even if !dev_replace_is_valid, the values are good enough for
1080 	 * the replace_status ioctl */
1081 	args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
1082 	args->status.replace_state = dev_replace->replace_state;
1083 	args->status.time_started = dev_replace->time_started;
1084 	args->status.time_stopped = dev_replace->time_stopped;
1085 	args->status.num_write_errors =
1086 		atomic64_read(&dev_replace->num_write_errors);
1087 	args->status.num_uncorrectable_read_errors =
1088 		atomic64_read(&dev_replace->num_uncorrectable_read_errors);
1089 	args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
1090 	up_read(&dev_replace->rwsem);
1091 }
1092 
1093 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
1094 {
1095 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1096 	struct btrfs_device *tgt_device = NULL;
1097 	struct btrfs_device *src_device = NULL;
1098 	struct btrfs_trans_handle *trans;
1099 	struct btrfs_root *root = fs_info->tree_root;
1100 	int result;
1101 	int ret;
1102 
1103 	if (sb_rdonly(fs_info->sb))
1104 		return -EROFS;
1105 
1106 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
1107 	down_write(&dev_replace->rwsem);
1108 	switch (dev_replace->replace_state) {
1109 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1110 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1111 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1112 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
1113 		up_write(&dev_replace->rwsem);
1114 		break;
1115 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1116 		tgt_device = dev_replace->tgtdev;
1117 		src_device = dev_replace->srcdev;
1118 		up_write(&dev_replace->rwsem);
1119 		ret = btrfs_scrub_cancel(fs_info);
1120 		if (ret < 0) {
1121 			result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
1122 		} else {
1123 			result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
1124 			/*
1125 			 * btrfs_dev_replace_finishing() will handle the
1126 			 * cleanup part
1127 			 */
1128 			btrfs_info_in_rcu(fs_info,
1129 				"dev_replace from %s (devid %llu) to %s canceled",
1130 				btrfs_dev_name(src_device), src_device->devid,
1131 				btrfs_dev_name(tgt_device));
1132 		}
1133 		break;
1134 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1135 		/*
1136 		 * Scrub doing the replace isn't running so we need to do the
1137 		 * cleanup step of btrfs_dev_replace_finishing() here
1138 		 */
1139 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
1140 		tgt_device = dev_replace->tgtdev;
1141 		src_device = dev_replace->srcdev;
1142 		dev_replace->tgtdev = NULL;
1143 		dev_replace->srcdev = NULL;
1144 		dev_replace->replace_state =
1145 				BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
1146 		dev_replace->time_stopped = ktime_get_real_seconds();
1147 		dev_replace->item_needs_writeback = 1;
1148 
1149 		up_write(&dev_replace->rwsem);
1150 
1151 		/* Scrub for replace must not be running in suspended state */
1152 		ret = btrfs_scrub_cancel(fs_info);
1153 		ASSERT(ret != -ENOTCONN);
1154 
1155 		trans = btrfs_start_transaction(root, 0);
1156 		if (IS_ERR(trans)) {
1157 			mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
1158 			return PTR_ERR(trans);
1159 		}
1160 		ret = btrfs_commit_transaction(trans);
1161 		WARN_ON(ret);
1162 
1163 		btrfs_info_in_rcu(fs_info,
1164 		"suspended dev_replace from %s (devid %llu) to %s canceled",
1165 			btrfs_dev_name(src_device), src_device->devid,
1166 			btrfs_dev_name(tgt_device));
1167 
1168 		if (tgt_device)
1169 			btrfs_destroy_dev_replace_tgtdev(tgt_device);
1170 		break;
1171 	default:
1172 		up_write(&dev_replace->rwsem);
1173 		result = -EINVAL;
1174 	}
1175 
1176 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
1177 	return result;
1178 }
1179 
1180 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
1181 {
1182 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1183 
1184 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
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 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1192 		break;
1193 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1194 		dev_replace->replace_state =
1195 			BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
1196 		dev_replace->time_stopped = ktime_get_real_seconds();
1197 		dev_replace->item_needs_writeback = 1;
1198 		btrfs_info(fs_info, "suspending dev_replace for unmount");
1199 		break;
1200 	}
1201 
1202 	up_write(&dev_replace->rwsem);
1203 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
1204 }
1205 
1206 /* resume dev_replace procedure that was interrupted by unmount */
1207 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
1208 {
1209 	struct task_struct *task;
1210 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1211 
1212 	down_write(&dev_replace->rwsem);
1213 
1214 	switch (dev_replace->replace_state) {
1215 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1216 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1217 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1218 		up_write(&dev_replace->rwsem);
1219 		return 0;
1220 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1221 		break;
1222 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1223 		dev_replace->replace_state =
1224 			BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
1225 		break;
1226 	}
1227 	if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
1228 		btrfs_info(fs_info,
1229 			   "cannot continue dev_replace, tgtdev is missing");
1230 		btrfs_info(fs_info,
1231 			   "you may cancel the operation after 'mount -o degraded'");
1232 		dev_replace->replace_state =
1233 					BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
1234 		up_write(&dev_replace->rwsem);
1235 		return 0;
1236 	}
1237 	up_write(&dev_replace->rwsem);
1238 
1239 	/*
1240 	 * This could collide with a paused balance, but the exclusive op logic
1241 	 * should never allow both to start and pause. We don't want to allow
1242 	 * dev-replace to start anyway.
1243 	 */
1244 	if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_DEV_REPLACE)) {
1245 		down_write(&dev_replace->rwsem);
1246 		dev_replace->replace_state =
1247 					BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
1248 		up_write(&dev_replace->rwsem);
1249 		btrfs_info(fs_info,
1250 		"cannot resume dev-replace, other exclusive operation running");
1251 		return 0;
1252 	}
1253 
1254 	task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
1255 	return PTR_ERR_OR_ZERO(task);
1256 }
1257 
1258 static int btrfs_dev_replace_kthread(void *data)
1259 {
1260 	struct btrfs_fs_info *fs_info = data;
1261 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1262 	u64 progress;
1263 	int ret;
1264 
1265 	progress = btrfs_dev_replace_progress(fs_info);
1266 	progress = div_u64(progress, 10);
1267 	btrfs_info_in_rcu(fs_info,
1268 		"continuing dev_replace from %s (devid %llu) to target %s @%u%%",
1269 		btrfs_dev_name(dev_replace->srcdev),
1270 		dev_replace->srcdev->devid,
1271 		btrfs_dev_name(dev_replace->tgtdev),
1272 		(unsigned int)progress);
1273 
1274 	ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
1275 			      dev_replace->committed_cursor_left,
1276 			      btrfs_device_get_total_bytes(dev_replace->srcdev),
1277 			      &dev_replace->scrub_progress, 0, 1);
1278 	ret = btrfs_dev_replace_finishing(fs_info, ret);
1279 	WARN_ON(ret && ret != -ECANCELED);
1280 
1281 	btrfs_exclop_finish(fs_info);
1282 	return 0;
1283 }
1284 
1285 int __pure btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
1286 {
1287 	if (!dev_replace->is_valid)
1288 		return 0;
1289 
1290 	switch (dev_replace->replace_state) {
1291 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1292 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1293 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1294 		return 0;
1295 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1296 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1297 		/*
1298 		 * return true even if tgtdev is missing (this is
1299 		 * something that can happen if the dev_replace
1300 		 * procedure is suspended by an umount and then
1301 		 * the tgtdev is missing (or "btrfs dev scan") was
1302 		 * not called and the filesystem is remounted
1303 		 * in degraded state. This does not stop the
1304 		 * dev_replace procedure. It needs to be canceled
1305 		 * manually if the cancellation is wanted.
1306 		 */
1307 		break;
1308 	}
1309 	return 1;
1310 }
1311 
1312 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
1313 {
1314 	percpu_counter_inc(&fs_info->dev_replace.bio_counter);
1315 }
1316 
1317 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
1318 {
1319 	percpu_counter_sub(&fs_info->dev_replace.bio_counter, amount);
1320 	cond_wake_up_nomb(&fs_info->dev_replace.replace_wait);
1321 }
1322 
1323 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
1324 {
1325 	while (1) {
1326 		percpu_counter_inc(&fs_info->dev_replace.bio_counter);
1327 		if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1328 				     &fs_info->fs_state)))
1329 			break;
1330 
1331 		btrfs_bio_counter_dec(fs_info);
1332 		wait_event(fs_info->dev_replace.replace_wait,
1333 			   !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1334 				     &fs_info->fs_state));
1335 	}
1336 }
1337