xref: /openbmc/linux/fs/btrfs/dev-replace.c (revision f125e2d4)
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 
25 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
26 				       int scrub_ret);
27 static void btrfs_dev_replace_update_device_in_mapping_tree(
28 						struct btrfs_fs_info *fs_info,
29 						struct btrfs_device *srcdev,
30 						struct btrfs_device *tgtdev);
31 static int btrfs_dev_replace_kthread(void *data);
32 
33 int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
34 {
35 	struct btrfs_key key;
36 	struct btrfs_root *dev_root = fs_info->dev_root;
37 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
38 	struct extent_buffer *eb;
39 	int slot;
40 	int ret = 0;
41 	struct btrfs_path *path = NULL;
42 	int item_size;
43 	struct btrfs_dev_replace_item *ptr;
44 	u64 src_devid;
45 
46 	path = btrfs_alloc_path();
47 	if (!path) {
48 		ret = -ENOMEM;
49 		goto out;
50 	}
51 
52 	key.objectid = 0;
53 	key.type = BTRFS_DEV_REPLACE_KEY;
54 	key.offset = 0;
55 	ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
56 	if (ret) {
57 no_valid_dev_replace_entry_found:
58 		ret = 0;
59 		dev_replace->replace_state =
60 			BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
61 		dev_replace->cont_reading_from_srcdev_mode =
62 		    BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
63 		dev_replace->time_started = 0;
64 		dev_replace->time_stopped = 0;
65 		atomic64_set(&dev_replace->num_write_errors, 0);
66 		atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
67 		dev_replace->cursor_left = 0;
68 		dev_replace->committed_cursor_left = 0;
69 		dev_replace->cursor_left_last_write_of_item = 0;
70 		dev_replace->cursor_right = 0;
71 		dev_replace->srcdev = NULL;
72 		dev_replace->tgtdev = NULL;
73 		dev_replace->is_valid = 0;
74 		dev_replace->item_needs_writeback = 0;
75 		goto out;
76 	}
77 	slot = path->slots[0];
78 	eb = path->nodes[0];
79 	item_size = btrfs_item_size_nr(eb, slot);
80 	ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
81 
82 	if (item_size != sizeof(struct btrfs_dev_replace_item)) {
83 		btrfs_warn(fs_info,
84 			"dev_replace entry found has unexpected size, ignore entry");
85 		goto no_valid_dev_replace_entry_found;
86 	}
87 
88 	src_devid = btrfs_dev_replace_src_devid(eb, ptr);
89 	dev_replace->cont_reading_from_srcdev_mode =
90 		btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
91 	dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
92 	dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
93 	dev_replace->time_stopped =
94 		btrfs_dev_replace_time_stopped(eb, ptr);
95 	atomic64_set(&dev_replace->num_write_errors,
96 		     btrfs_dev_replace_num_write_errors(eb, ptr));
97 	atomic64_set(&dev_replace->num_uncorrectable_read_errors,
98 		     btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
99 	dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
100 	dev_replace->committed_cursor_left = dev_replace->cursor_left;
101 	dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
102 	dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
103 	dev_replace->is_valid = 1;
104 
105 	dev_replace->item_needs_writeback = 0;
106 	switch (dev_replace->replace_state) {
107 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
108 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
109 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
110 		dev_replace->srcdev = NULL;
111 		dev_replace->tgtdev = NULL;
112 		break;
113 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
114 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
115 		dev_replace->srcdev = btrfs_find_device(fs_info->fs_devices,
116 						src_devid, NULL, NULL, true);
117 		dev_replace->tgtdev = btrfs_find_device(fs_info->fs_devices,
118 							BTRFS_DEV_REPLACE_DEVID,
119 							NULL, NULL, true);
120 		/*
121 		 * allow 'btrfs dev replace_cancel' if src/tgt device is
122 		 * missing
123 		 */
124 		if (!dev_replace->srcdev &&
125 		    !btrfs_test_opt(fs_info, DEGRADED)) {
126 			ret = -EIO;
127 			btrfs_warn(fs_info,
128 			   "cannot mount because device replace operation is ongoing and");
129 			btrfs_warn(fs_info,
130 			   "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
131 			   src_devid);
132 		}
133 		if (!dev_replace->tgtdev &&
134 		    !btrfs_test_opt(fs_info, DEGRADED)) {
135 			ret = -EIO;
136 			btrfs_warn(fs_info,
137 			   "cannot mount because device replace operation is ongoing and");
138 			btrfs_warn(fs_info,
139 			   "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
140 				BTRFS_DEV_REPLACE_DEVID);
141 		}
142 		if (dev_replace->tgtdev) {
143 			if (dev_replace->srcdev) {
144 				dev_replace->tgtdev->total_bytes =
145 					dev_replace->srcdev->total_bytes;
146 				dev_replace->tgtdev->disk_total_bytes =
147 					dev_replace->srcdev->disk_total_bytes;
148 				dev_replace->tgtdev->commit_total_bytes =
149 					dev_replace->srcdev->commit_total_bytes;
150 				dev_replace->tgtdev->bytes_used =
151 					dev_replace->srcdev->bytes_used;
152 				dev_replace->tgtdev->commit_bytes_used =
153 					dev_replace->srcdev->commit_bytes_used;
154 			}
155 			set_bit(BTRFS_DEV_STATE_REPLACE_TGT,
156 				&dev_replace->tgtdev->dev_state);
157 
158 			WARN_ON(fs_info->fs_devices->rw_devices == 0);
159 			dev_replace->tgtdev->io_width = fs_info->sectorsize;
160 			dev_replace->tgtdev->io_align = fs_info->sectorsize;
161 			dev_replace->tgtdev->sector_size = fs_info->sectorsize;
162 			dev_replace->tgtdev->fs_info = fs_info;
163 			set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
164 				&dev_replace->tgtdev->dev_state);
165 		}
166 		break;
167 	}
168 
169 out:
170 	btrfs_free_path(path);
171 	return ret;
172 }
173 
174 /*
175  * Initialize a new device for device replace target from a given source dev
176  * and path.
177  *
178  * Return 0 and new device in @device_out, otherwise return < 0
179  */
180 static int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
181 				  const char *device_path,
182 				  struct btrfs_device *srcdev,
183 				  struct btrfs_device **device_out)
184 {
185 	struct btrfs_device *device;
186 	struct block_device *bdev;
187 	struct list_head *devices;
188 	struct rcu_string *name;
189 	u64 devid = BTRFS_DEV_REPLACE_DEVID;
190 	int ret = 0;
191 
192 	*device_out = NULL;
193 	if (fs_info->fs_devices->seeding) {
194 		btrfs_err(fs_info, "the filesystem is a seed filesystem!");
195 		return -EINVAL;
196 	}
197 
198 	bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
199 				  fs_info->bdev_holder);
200 	if (IS_ERR(bdev)) {
201 		btrfs_err(fs_info, "target device %s is invalid!", device_path);
202 		return PTR_ERR(bdev);
203 	}
204 
205 	sync_blockdev(bdev);
206 
207 	devices = &fs_info->fs_devices->devices;
208 	list_for_each_entry(device, devices, dev_list) {
209 		if (device->bdev == bdev) {
210 			btrfs_err(fs_info,
211 				  "target device is in the filesystem!");
212 			ret = -EEXIST;
213 			goto error;
214 		}
215 	}
216 
217 
218 	if (i_size_read(bdev->bd_inode) <
219 	    btrfs_device_get_total_bytes(srcdev)) {
220 		btrfs_err(fs_info,
221 			  "target device is smaller than source device!");
222 		ret = -EINVAL;
223 		goto error;
224 	}
225 
226 
227 	device = btrfs_alloc_device(NULL, &devid, NULL);
228 	if (IS_ERR(device)) {
229 		ret = PTR_ERR(device);
230 		goto error;
231 	}
232 
233 	name = rcu_string_strdup(device_path, GFP_KERNEL);
234 	if (!name) {
235 		btrfs_free_device(device);
236 		ret = -ENOMEM;
237 		goto error;
238 	}
239 	rcu_assign_pointer(device->name, name);
240 
241 	set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
242 	device->generation = 0;
243 	device->io_width = fs_info->sectorsize;
244 	device->io_align = fs_info->sectorsize;
245 	device->sector_size = fs_info->sectorsize;
246 	device->total_bytes = btrfs_device_get_total_bytes(srcdev);
247 	device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
248 	device->bytes_used = btrfs_device_get_bytes_used(srcdev);
249 	device->commit_total_bytes = srcdev->commit_total_bytes;
250 	device->commit_bytes_used = device->bytes_used;
251 	device->fs_info = fs_info;
252 	device->bdev = bdev;
253 	set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
254 	set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
255 	device->mode = FMODE_EXCL;
256 	device->dev_stats_valid = 1;
257 	set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
258 	device->fs_devices = fs_info->fs_devices;
259 
260 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
261 	list_add(&device->dev_list, &fs_info->fs_devices->devices);
262 	fs_info->fs_devices->num_devices++;
263 	fs_info->fs_devices->open_devices++;
264 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
265 
266 	*device_out = device;
267 	return 0;
268 
269 error:
270 	blkdev_put(bdev, FMODE_EXCL);
271 	return ret;
272 }
273 
274 /*
275  * called from commit_transaction. Writes changed device replace state to
276  * disk.
277  */
278 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans)
279 {
280 	struct btrfs_fs_info *fs_info = trans->fs_info;
281 	int ret;
282 	struct btrfs_root *dev_root = fs_info->dev_root;
283 	struct btrfs_path *path;
284 	struct btrfs_key key;
285 	struct extent_buffer *eb;
286 	struct btrfs_dev_replace_item *ptr;
287 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
288 
289 	down_read(&dev_replace->rwsem);
290 	if (!dev_replace->is_valid ||
291 	    !dev_replace->item_needs_writeback) {
292 		up_read(&dev_replace->rwsem);
293 		return 0;
294 	}
295 	up_read(&dev_replace->rwsem);
296 
297 	key.objectid = 0;
298 	key.type = BTRFS_DEV_REPLACE_KEY;
299 	key.offset = 0;
300 
301 	path = btrfs_alloc_path();
302 	if (!path) {
303 		ret = -ENOMEM;
304 		goto out;
305 	}
306 	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
307 	if (ret < 0) {
308 		btrfs_warn(fs_info,
309 			   "error %d while searching for dev_replace item!",
310 			   ret);
311 		goto out;
312 	}
313 
314 	if (ret == 0 &&
315 	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
316 		/*
317 		 * need to delete old one and insert a new one.
318 		 * Since no attempt is made to recover any old state, if the
319 		 * dev_replace state is 'running', the data on the target
320 		 * drive is lost.
321 		 * It would be possible to recover the state: just make sure
322 		 * that the beginning of the item is never changed and always
323 		 * contains all the essential information. Then read this
324 		 * minimal set of information and use it as a base for the
325 		 * new state.
326 		 */
327 		ret = btrfs_del_item(trans, dev_root, path);
328 		if (ret != 0) {
329 			btrfs_warn(fs_info,
330 				   "delete too small dev_replace item failed %d!",
331 				   ret);
332 			goto out;
333 		}
334 		ret = 1;
335 	}
336 
337 	if (ret == 1) {
338 		/* need to insert a new item */
339 		btrfs_release_path(path);
340 		ret = btrfs_insert_empty_item(trans, dev_root, path,
341 					      &key, sizeof(*ptr));
342 		if (ret < 0) {
343 			btrfs_warn(fs_info,
344 				   "insert dev_replace item failed %d!", ret);
345 			goto out;
346 		}
347 	}
348 
349 	eb = path->nodes[0];
350 	ptr = btrfs_item_ptr(eb, path->slots[0],
351 			     struct btrfs_dev_replace_item);
352 
353 	down_write(&dev_replace->rwsem);
354 	if (dev_replace->srcdev)
355 		btrfs_set_dev_replace_src_devid(eb, ptr,
356 			dev_replace->srcdev->devid);
357 	else
358 		btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
359 	btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
360 		dev_replace->cont_reading_from_srcdev_mode);
361 	btrfs_set_dev_replace_replace_state(eb, ptr,
362 		dev_replace->replace_state);
363 	btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
364 	btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
365 	btrfs_set_dev_replace_num_write_errors(eb, ptr,
366 		atomic64_read(&dev_replace->num_write_errors));
367 	btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
368 		atomic64_read(&dev_replace->num_uncorrectable_read_errors));
369 	dev_replace->cursor_left_last_write_of_item =
370 		dev_replace->cursor_left;
371 	btrfs_set_dev_replace_cursor_left(eb, ptr,
372 		dev_replace->cursor_left_last_write_of_item);
373 	btrfs_set_dev_replace_cursor_right(eb, ptr,
374 		dev_replace->cursor_right);
375 	dev_replace->item_needs_writeback = 0;
376 	up_write(&dev_replace->rwsem);
377 
378 	btrfs_mark_buffer_dirty(eb);
379 
380 out:
381 	btrfs_free_path(path);
382 
383 	return ret;
384 }
385 
386 static char* btrfs_dev_name(struct btrfs_device *device)
387 {
388 	if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
389 		return "<missing disk>";
390 	else
391 		return rcu_str_deref(device->name);
392 }
393 
394 static int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
395 		const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
396 		int read_src)
397 {
398 	struct btrfs_root *root = fs_info->dev_root;
399 	struct btrfs_trans_handle *trans;
400 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
401 	int ret;
402 	struct btrfs_device *tgt_device = NULL;
403 	struct btrfs_device *src_device = NULL;
404 
405 	src_device = btrfs_find_device_by_devspec(fs_info, srcdevid,
406 						  srcdev_name);
407 	if (IS_ERR(src_device))
408 		return PTR_ERR(src_device);
409 
410 	if (btrfs_pinned_by_swapfile(fs_info, src_device)) {
411 		btrfs_warn_in_rcu(fs_info,
412 	  "cannot replace device %s (devid %llu) due to active swapfile",
413 			btrfs_dev_name(src_device), src_device->devid);
414 		return -ETXTBSY;
415 	}
416 
417 	/*
418 	 * Here we commit the transaction to make sure commit_total_bytes
419 	 * of all the devices are updated.
420 	 */
421 	trans = btrfs_attach_transaction(root);
422 	if (!IS_ERR(trans)) {
423 		ret = btrfs_commit_transaction(trans);
424 		if (ret)
425 			return ret;
426 	} else if (PTR_ERR(trans) != -ENOENT) {
427 		return PTR_ERR(trans);
428 	}
429 
430 	ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
431 					    src_device, &tgt_device);
432 	if (ret)
433 		return ret;
434 
435 	down_write(&dev_replace->rwsem);
436 	switch (dev_replace->replace_state) {
437 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
438 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
439 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
440 		break;
441 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
442 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
443 		ASSERT(0);
444 		ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
445 		up_write(&dev_replace->rwsem);
446 		goto leave;
447 	}
448 
449 	dev_replace->cont_reading_from_srcdev_mode = read_src;
450 	dev_replace->srcdev = src_device;
451 	dev_replace->tgtdev = tgt_device;
452 
453 	btrfs_info_in_rcu(fs_info,
454 		      "dev_replace from %s (devid %llu) to %s started",
455 		      btrfs_dev_name(src_device),
456 		      src_device->devid,
457 		      rcu_str_deref(tgt_device->name));
458 
459 	/*
460 	 * from now on, the writes to the srcdev are all duplicated to
461 	 * go to the tgtdev as well (refer to btrfs_map_block()).
462 	 */
463 	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
464 	dev_replace->time_started = ktime_get_real_seconds();
465 	dev_replace->cursor_left = 0;
466 	dev_replace->committed_cursor_left = 0;
467 	dev_replace->cursor_left_last_write_of_item = 0;
468 	dev_replace->cursor_right = 0;
469 	dev_replace->is_valid = 1;
470 	dev_replace->item_needs_writeback = 1;
471 	atomic64_set(&dev_replace->num_write_errors, 0);
472 	atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
473 	up_write(&dev_replace->rwsem);
474 
475 	ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device);
476 	if (ret)
477 		btrfs_err(fs_info, "kobj add dev failed %d", ret);
478 
479 	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
480 
481 	/* Commit dev_replace state and reserve 1 item for it. */
482 	trans = btrfs_start_transaction(root, 1);
483 	if (IS_ERR(trans)) {
484 		ret = PTR_ERR(trans);
485 		down_write(&dev_replace->rwsem);
486 		dev_replace->replace_state =
487 			BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
488 		dev_replace->srcdev = NULL;
489 		dev_replace->tgtdev = NULL;
490 		up_write(&dev_replace->rwsem);
491 		goto leave;
492 	}
493 
494 	ret = btrfs_commit_transaction(trans);
495 	WARN_ON(ret);
496 
497 	/* the disk copy procedure reuses the scrub code */
498 	ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
499 			      btrfs_device_get_total_bytes(src_device),
500 			      &dev_replace->scrub_progress, 0, 1);
501 
502 	ret = btrfs_dev_replace_finishing(fs_info, ret);
503 	if (ret == -EINPROGRESS)
504 		ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
505 
506 	return ret;
507 
508 leave:
509 	btrfs_destroy_dev_replace_tgtdev(tgt_device);
510 	return ret;
511 }
512 
513 int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
514 			    struct btrfs_ioctl_dev_replace_args *args)
515 {
516 	int ret;
517 
518 	switch (args->start.cont_reading_from_srcdev_mode) {
519 	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
520 	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
521 		break;
522 	default:
523 		return -EINVAL;
524 	}
525 
526 	if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
527 	    args->start.tgtdev_name[0] == '\0')
528 		return -EINVAL;
529 
530 	ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name,
531 					args->start.srcdevid,
532 					args->start.srcdev_name,
533 					args->start.cont_reading_from_srcdev_mode);
534 	args->result = ret;
535 	/* don't warn if EINPROGRESS, someone else might be running scrub */
536 	if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS ||
537 	    ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR)
538 		return 0;
539 
540 	return ret;
541 }
542 
543 /*
544  * blocked until all in-flight bios operations are finished.
545  */
546 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
547 {
548 	set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
549 	wait_event(fs_info->dev_replace.replace_wait, !percpu_counter_sum(
550 		   &fs_info->dev_replace.bio_counter));
551 }
552 
553 /*
554  * we have removed target device, it is safe to allow new bios request.
555  */
556 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
557 {
558 	clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
559 	wake_up(&fs_info->dev_replace.replace_wait);
560 }
561 
562 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
563 				       int scrub_ret)
564 {
565 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
566 	struct btrfs_device *tgt_device;
567 	struct btrfs_device *src_device;
568 	struct btrfs_root *root = fs_info->tree_root;
569 	u8 uuid_tmp[BTRFS_UUID_SIZE];
570 	struct btrfs_trans_handle *trans;
571 	int ret = 0;
572 
573 	/* don't allow cancel or unmount to disturb the finishing procedure */
574 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
575 
576 	down_read(&dev_replace->rwsem);
577 	/* was the operation canceled, or is it finished? */
578 	if (dev_replace->replace_state !=
579 	    BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
580 		up_read(&dev_replace->rwsem);
581 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
582 		return 0;
583 	}
584 
585 	tgt_device = dev_replace->tgtdev;
586 	src_device = dev_replace->srcdev;
587 	up_read(&dev_replace->rwsem);
588 
589 	/*
590 	 * flush all outstanding I/O and inode extent mappings before the
591 	 * copy operation is declared as being finished
592 	 */
593 	ret = btrfs_start_delalloc_roots(fs_info, -1);
594 	if (ret) {
595 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
596 		return ret;
597 	}
598 	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
599 
600 	/*
601 	 * We have to use this loop approach because at this point src_device
602 	 * has to be available for transaction commit to complete, yet new
603 	 * chunks shouldn't be allocated on the device.
604 	 */
605 	while (1) {
606 		trans = btrfs_start_transaction(root, 0);
607 		if (IS_ERR(trans)) {
608 			mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
609 			return PTR_ERR(trans);
610 		}
611 		ret = btrfs_commit_transaction(trans);
612 		WARN_ON(ret);
613 
614 		/* Prevent write_all_supers() during the finishing procedure */
615 		mutex_lock(&fs_info->fs_devices->device_list_mutex);
616 		/* Prevent new chunks being allocated on the source device */
617 		mutex_lock(&fs_info->chunk_mutex);
618 
619 		if (!list_empty(&src_device->post_commit_list)) {
620 			mutex_unlock(&fs_info->fs_devices->device_list_mutex);
621 			mutex_unlock(&fs_info->chunk_mutex);
622 		} else {
623 			break;
624 		}
625 	}
626 
627 	down_write(&dev_replace->rwsem);
628 	dev_replace->replace_state =
629 		scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
630 			  : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
631 	dev_replace->tgtdev = NULL;
632 	dev_replace->srcdev = NULL;
633 	dev_replace->time_stopped = ktime_get_real_seconds();
634 	dev_replace->item_needs_writeback = 1;
635 
636 	/* replace old device with new one in mapping tree */
637 	if (!scrub_ret) {
638 		btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
639 								src_device,
640 								tgt_device);
641 	} else {
642 		if (scrub_ret != -ECANCELED)
643 			btrfs_err_in_rcu(fs_info,
644 				 "btrfs_scrub_dev(%s, %llu, %s) failed %d",
645 				 btrfs_dev_name(src_device),
646 				 src_device->devid,
647 				 rcu_str_deref(tgt_device->name), scrub_ret);
648 		up_write(&dev_replace->rwsem);
649 		mutex_unlock(&fs_info->chunk_mutex);
650 		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
651 		btrfs_rm_dev_replace_blocked(fs_info);
652 		if (tgt_device)
653 			btrfs_destroy_dev_replace_tgtdev(tgt_device);
654 		btrfs_rm_dev_replace_unblocked(fs_info);
655 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
656 
657 		return scrub_ret;
658 	}
659 
660 	btrfs_info_in_rcu(fs_info,
661 			  "dev_replace from %s (devid %llu) to %s finished",
662 			  btrfs_dev_name(src_device),
663 			  src_device->devid,
664 			  rcu_str_deref(tgt_device->name));
665 	clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state);
666 	tgt_device->devid = src_device->devid;
667 	src_device->devid = BTRFS_DEV_REPLACE_DEVID;
668 	memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
669 	memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
670 	memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
671 	btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
672 	btrfs_device_set_disk_total_bytes(tgt_device,
673 					  src_device->disk_total_bytes);
674 	btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
675 	tgt_device->commit_bytes_used = src_device->bytes_used;
676 
677 	btrfs_assign_next_active_device(src_device, tgt_device);
678 
679 	list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
680 	fs_info->fs_devices->rw_devices++;
681 
682 	up_write(&dev_replace->rwsem);
683 	btrfs_rm_dev_replace_blocked(fs_info);
684 
685 	btrfs_rm_dev_replace_remove_srcdev(src_device);
686 
687 	btrfs_rm_dev_replace_unblocked(fs_info);
688 
689 	/*
690 	 * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will
691 	 * update on-disk dev stats value during commit transaction
692 	 */
693 	atomic_inc(&tgt_device->dev_stats_ccnt);
694 
695 	/*
696 	 * this is again a consistent state where no dev_replace procedure
697 	 * is running, the target device is part of the filesystem, the
698 	 * source device is not part of the filesystem anymore and its 1st
699 	 * superblock is scratched out so that it is no longer marked to
700 	 * belong to this filesystem.
701 	 */
702 	mutex_unlock(&fs_info->chunk_mutex);
703 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
704 
705 	/* replace the sysfs entry */
706 	btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device);
707 	btrfs_sysfs_update_devid(tgt_device);
708 	btrfs_rm_dev_replace_free_srcdev(src_device);
709 
710 	/* write back the superblocks */
711 	trans = btrfs_start_transaction(root, 0);
712 	if (!IS_ERR(trans))
713 		btrfs_commit_transaction(trans);
714 
715 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
716 
717 	return 0;
718 }
719 
720 static void btrfs_dev_replace_update_device_in_mapping_tree(
721 						struct btrfs_fs_info *fs_info,
722 						struct btrfs_device *srcdev,
723 						struct btrfs_device *tgtdev)
724 {
725 	struct extent_map_tree *em_tree = &fs_info->mapping_tree;
726 	struct extent_map *em;
727 	struct map_lookup *map;
728 	u64 start = 0;
729 	int i;
730 
731 	write_lock(&em_tree->lock);
732 	do {
733 		em = lookup_extent_mapping(em_tree, start, (u64)-1);
734 		if (!em)
735 			break;
736 		map = em->map_lookup;
737 		for (i = 0; i < map->num_stripes; i++)
738 			if (srcdev == map->stripes[i].dev)
739 				map->stripes[i].dev = tgtdev;
740 		start = em->start + em->len;
741 		free_extent_map(em);
742 	} while (start);
743 	write_unlock(&em_tree->lock);
744 }
745 
746 /*
747  * Read progress of device replace status according to the state and last
748  * stored position. The value format is the same as for
749  * btrfs_dev_replace::progress_1000
750  */
751 static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info)
752 {
753 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
754 	u64 ret = 0;
755 
756 	switch (dev_replace->replace_state) {
757 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
758 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
759 		ret = 0;
760 		break;
761 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
762 		ret = 1000;
763 		break;
764 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
765 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
766 		ret = div64_u64(dev_replace->cursor_left,
767 				div_u64(btrfs_device_get_total_bytes(
768 						dev_replace->srcdev), 1000));
769 		break;
770 	}
771 
772 	return ret;
773 }
774 
775 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
776 			      struct btrfs_ioctl_dev_replace_args *args)
777 {
778 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
779 
780 	down_read(&dev_replace->rwsem);
781 	/* even if !dev_replace_is_valid, the values are good enough for
782 	 * the replace_status ioctl */
783 	args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
784 	args->status.replace_state = dev_replace->replace_state;
785 	args->status.time_started = dev_replace->time_started;
786 	args->status.time_stopped = dev_replace->time_stopped;
787 	args->status.num_write_errors =
788 		atomic64_read(&dev_replace->num_write_errors);
789 	args->status.num_uncorrectable_read_errors =
790 		atomic64_read(&dev_replace->num_uncorrectable_read_errors);
791 	args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
792 	up_read(&dev_replace->rwsem);
793 }
794 
795 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
796 {
797 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
798 	struct btrfs_device *tgt_device = NULL;
799 	struct btrfs_device *src_device = NULL;
800 	struct btrfs_trans_handle *trans;
801 	struct btrfs_root *root = fs_info->tree_root;
802 	int result;
803 	int ret;
804 
805 	if (sb_rdonly(fs_info->sb))
806 		return -EROFS;
807 
808 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
809 	down_write(&dev_replace->rwsem);
810 	switch (dev_replace->replace_state) {
811 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
812 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
813 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
814 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
815 		up_write(&dev_replace->rwsem);
816 		break;
817 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
818 		tgt_device = dev_replace->tgtdev;
819 		src_device = dev_replace->srcdev;
820 		up_write(&dev_replace->rwsem);
821 		ret = btrfs_scrub_cancel(fs_info);
822 		if (ret < 0) {
823 			result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
824 		} else {
825 			result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
826 			/*
827 			 * btrfs_dev_replace_finishing() will handle the
828 			 * cleanup part
829 			 */
830 			btrfs_info_in_rcu(fs_info,
831 				"dev_replace from %s (devid %llu) to %s canceled",
832 				btrfs_dev_name(src_device), src_device->devid,
833 				btrfs_dev_name(tgt_device));
834 		}
835 		break;
836 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
837 		/*
838 		 * Scrub doing the replace isn't running so we need to do the
839 		 * cleanup step of btrfs_dev_replace_finishing() here
840 		 */
841 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
842 		tgt_device = dev_replace->tgtdev;
843 		src_device = dev_replace->srcdev;
844 		dev_replace->tgtdev = NULL;
845 		dev_replace->srcdev = NULL;
846 		dev_replace->replace_state =
847 				BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
848 		dev_replace->time_stopped = ktime_get_real_seconds();
849 		dev_replace->item_needs_writeback = 1;
850 
851 		up_write(&dev_replace->rwsem);
852 
853 		/* Scrub for replace must not be running in suspended state */
854 		ret = btrfs_scrub_cancel(fs_info);
855 		ASSERT(ret != -ENOTCONN);
856 
857 		trans = btrfs_start_transaction(root, 0);
858 		if (IS_ERR(trans)) {
859 			mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
860 			return PTR_ERR(trans);
861 		}
862 		ret = btrfs_commit_transaction(trans);
863 		WARN_ON(ret);
864 
865 		btrfs_info_in_rcu(fs_info,
866 		"suspended dev_replace from %s (devid %llu) to %s canceled",
867 			btrfs_dev_name(src_device), src_device->devid,
868 			btrfs_dev_name(tgt_device));
869 
870 		if (tgt_device)
871 			btrfs_destroy_dev_replace_tgtdev(tgt_device);
872 		break;
873 	default:
874 		up_write(&dev_replace->rwsem);
875 		result = -EINVAL;
876 	}
877 
878 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
879 	return result;
880 }
881 
882 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
883 {
884 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
885 
886 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
887 	down_write(&dev_replace->rwsem);
888 
889 	switch (dev_replace->replace_state) {
890 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
891 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
892 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
893 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
894 		break;
895 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
896 		dev_replace->replace_state =
897 			BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
898 		dev_replace->time_stopped = ktime_get_real_seconds();
899 		dev_replace->item_needs_writeback = 1;
900 		btrfs_info(fs_info, "suspending dev_replace for unmount");
901 		break;
902 	}
903 
904 	up_write(&dev_replace->rwsem);
905 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
906 }
907 
908 /* resume dev_replace procedure that was interrupted by unmount */
909 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
910 {
911 	struct task_struct *task;
912 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
913 
914 	down_write(&dev_replace->rwsem);
915 
916 	switch (dev_replace->replace_state) {
917 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
918 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
919 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
920 		up_write(&dev_replace->rwsem);
921 		return 0;
922 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
923 		break;
924 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
925 		dev_replace->replace_state =
926 			BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
927 		break;
928 	}
929 	if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
930 		btrfs_info(fs_info,
931 			   "cannot continue dev_replace, tgtdev is missing");
932 		btrfs_info(fs_info,
933 			   "you may cancel the operation after 'mount -o degraded'");
934 		dev_replace->replace_state =
935 					BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
936 		up_write(&dev_replace->rwsem);
937 		return 0;
938 	}
939 	up_write(&dev_replace->rwsem);
940 
941 	/*
942 	 * This could collide with a paused balance, but the exclusive op logic
943 	 * should never allow both to start and pause. We don't want to allow
944 	 * dev-replace to start anyway.
945 	 */
946 	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
947 		down_write(&dev_replace->rwsem);
948 		dev_replace->replace_state =
949 					BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
950 		up_write(&dev_replace->rwsem);
951 		btrfs_info(fs_info,
952 		"cannot resume dev-replace, other exclusive operation running");
953 		return 0;
954 	}
955 
956 	task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
957 	return PTR_ERR_OR_ZERO(task);
958 }
959 
960 static int btrfs_dev_replace_kthread(void *data)
961 {
962 	struct btrfs_fs_info *fs_info = data;
963 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
964 	u64 progress;
965 	int ret;
966 
967 	progress = btrfs_dev_replace_progress(fs_info);
968 	progress = div_u64(progress, 10);
969 	btrfs_info_in_rcu(fs_info,
970 		"continuing dev_replace from %s (devid %llu) to target %s @%u%%",
971 		btrfs_dev_name(dev_replace->srcdev),
972 		dev_replace->srcdev->devid,
973 		btrfs_dev_name(dev_replace->tgtdev),
974 		(unsigned int)progress);
975 
976 	ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
977 			      dev_replace->committed_cursor_left,
978 			      btrfs_device_get_total_bytes(dev_replace->srcdev),
979 			      &dev_replace->scrub_progress, 0, 1);
980 	ret = btrfs_dev_replace_finishing(fs_info, ret);
981 	WARN_ON(ret && ret != -ECANCELED);
982 
983 	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
984 	return 0;
985 }
986 
987 int __pure btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
988 {
989 	if (!dev_replace->is_valid)
990 		return 0;
991 
992 	switch (dev_replace->replace_state) {
993 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
994 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
995 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
996 		return 0;
997 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
998 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
999 		/*
1000 		 * return true even if tgtdev is missing (this is
1001 		 * something that can happen if the dev_replace
1002 		 * procedure is suspended by an umount and then
1003 		 * the tgtdev is missing (or "btrfs dev scan") was
1004 		 * not called and the filesystem is remounted
1005 		 * in degraded state. This does not stop the
1006 		 * dev_replace procedure. It needs to be canceled
1007 		 * manually if the cancellation is wanted.
1008 		 */
1009 		break;
1010 	}
1011 	return 1;
1012 }
1013 
1014 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
1015 {
1016 	percpu_counter_inc(&fs_info->dev_replace.bio_counter);
1017 }
1018 
1019 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
1020 {
1021 	percpu_counter_sub(&fs_info->dev_replace.bio_counter, amount);
1022 	cond_wake_up_nomb(&fs_info->dev_replace.replace_wait);
1023 }
1024 
1025 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
1026 {
1027 	while (1) {
1028 		percpu_counter_inc(&fs_info->dev_replace.bio_counter);
1029 		if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1030 				     &fs_info->fs_state)))
1031 			break;
1032 
1033 		btrfs_bio_counter_dec(fs_info);
1034 		wait_event(fs_info->dev_replace.replace_wait,
1035 			   !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1036 				     &fs_info->fs_state));
1037 	}
1038 }
1039