xref: /openbmc/linux/fs/btrfs/dev-replace.c (revision 5104d265)
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 "compat.h"
30 #include "ctree.h"
31 #include "extent_map.h"
32 #include "disk-io.h"
33 #include "transaction.h"
34 #include "print-tree.h"
35 #include "volumes.h"
36 #include "async-thread.h"
37 #include "check-integrity.h"
38 #include "rcu-string.h"
39 #include "dev-replace.h"
40 
41 static u64 btrfs_get_seconds_since_1970(void);
42 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
43 				       int scrub_ret);
44 static void btrfs_dev_replace_update_device_in_mapping_tree(
45 						struct btrfs_fs_info *fs_info,
46 						struct btrfs_device *srcdev,
47 						struct btrfs_device *tgtdev);
48 static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid,
49 					 char *srcdev_name,
50 					 struct btrfs_device **device);
51 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info);
52 static int btrfs_dev_replace_kthread(void *data);
53 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info);
54 
55 
56 int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
57 {
58 	struct btrfs_key key;
59 	struct btrfs_root *dev_root = fs_info->dev_root;
60 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
61 	struct extent_buffer *eb;
62 	int slot;
63 	int ret = 0;
64 	struct btrfs_path *path = NULL;
65 	int item_size;
66 	struct btrfs_dev_replace_item *ptr;
67 	u64 src_devid;
68 
69 	path = btrfs_alloc_path();
70 	if (!path) {
71 		ret = -ENOMEM;
72 		goto out;
73 	}
74 
75 	key.objectid = 0;
76 	key.type = BTRFS_DEV_REPLACE_KEY;
77 	key.offset = 0;
78 	ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
79 	if (ret) {
80 no_valid_dev_replace_entry_found:
81 		ret = 0;
82 		dev_replace->replace_state =
83 			BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED;
84 		dev_replace->cont_reading_from_srcdev_mode =
85 		    BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
86 		dev_replace->replace_state = 0;
87 		dev_replace->time_started = 0;
88 		dev_replace->time_stopped = 0;
89 		atomic64_set(&dev_replace->num_write_errors, 0);
90 		atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
91 		dev_replace->cursor_left = 0;
92 		dev_replace->committed_cursor_left = 0;
93 		dev_replace->cursor_left_last_write_of_item = 0;
94 		dev_replace->cursor_right = 0;
95 		dev_replace->srcdev = NULL;
96 		dev_replace->tgtdev = NULL;
97 		dev_replace->is_valid = 0;
98 		dev_replace->item_needs_writeback = 0;
99 		goto out;
100 	}
101 	slot = path->slots[0];
102 	eb = path->nodes[0];
103 	item_size = btrfs_item_size_nr(eb, slot);
104 	ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
105 
106 	if (item_size != sizeof(struct btrfs_dev_replace_item)) {
107 		pr_warn("btrfs: dev_replace entry found has unexpected size, ignore entry\n");
108 		goto no_valid_dev_replace_entry_found;
109 	}
110 
111 	src_devid = btrfs_dev_replace_src_devid(eb, ptr);
112 	dev_replace->cont_reading_from_srcdev_mode =
113 		btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
114 	dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
115 	dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
116 	dev_replace->time_stopped =
117 		btrfs_dev_replace_time_stopped(eb, ptr);
118 	atomic64_set(&dev_replace->num_write_errors,
119 		     btrfs_dev_replace_num_write_errors(eb, ptr));
120 	atomic64_set(&dev_replace->num_uncorrectable_read_errors,
121 		     btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
122 	dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
123 	dev_replace->committed_cursor_left = dev_replace->cursor_left;
124 	dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
125 	dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
126 	dev_replace->is_valid = 1;
127 
128 	dev_replace->item_needs_writeback = 0;
129 	switch (dev_replace->replace_state) {
130 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
131 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
132 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
133 		dev_replace->srcdev = NULL;
134 		dev_replace->tgtdev = NULL;
135 		break;
136 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
137 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
138 		dev_replace->srcdev = btrfs_find_device(fs_info, src_devid,
139 							NULL, NULL);
140 		dev_replace->tgtdev = btrfs_find_device(fs_info,
141 							BTRFS_DEV_REPLACE_DEVID,
142 							NULL, NULL);
143 		/*
144 		 * allow 'btrfs dev replace_cancel' if src/tgt device is
145 		 * missing
146 		 */
147 		if (!dev_replace->srcdev &&
148 		    !btrfs_test_opt(dev_root, DEGRADED)) {
149 			ret = -EIO;
150 			pr_warn("btrfs: cannot mount because device replace operation is ongoing and\n" "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?\n",
151 				(unsigned long long)src_devid);
152 		}
153 		if (!dev_replace->tgtdev &&
154 		    !btrfs_test_opt(dev_root, DEGRADED)) {
155 			ret = -EIO;
156 			pr_warn("btrfs: cannot mount because device replace operation is ongoing and\n" "tgtdev (devid %llu) is missing, need to run btrfs dev scan?\n",
157 				(unsigned long long)BTRFS_DEV_REPLACE_DEVID);
158 		}
159 		if (dev_replace->tgtdev) {
160 			if (dev_replace->srcdev) {
161 				dev_replace->tgtdev->total_bytes =
162 					dev_replace->srcdev->total_bytes;
163 				dev_replace->tgtdev->disk_total_bytes =
164 					dev_replace->srcdev->disk_total_bytes;
165 				dev_replace->tgtdev->bytes_used =
166 					dev_replace->srcdev->bytes_used;
167 			}
168 			dev_replace->tgtdev->is_tgtdev_for_dev_replace = 1;
169 			btrfs_init_dev_replace_tgtdev_for_resume(fs_info,
170 				dev_replace->tgtdev);
171 		}
172 		break;
173 	}
174 
175 out:
176 	if (path)
177 		btrfs_free_path(path);
178 	return ret;
179 }
180 
181 /*
182  * called from commit_transaction. Writes changed device replace state to
183  * disk.
184  */
185 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
186 			  struct btrfs_fs_info *fs_info)
187 {
188 	int ret;
189 	struct btrfs_root *dev_root = fs_info->dev_root;
190 	struct btrfs_path *path;
191 	struct btrfs_key key;
192 	struct extent_buffer *eb;
193 	struct btrfs_dev_replace_item *ptr;
194 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
195 
196 	btrfs_dev_replace_lock(dev_replace);
197 	if (!dev_replace->is_valid ||
198 	    !dev_replace->item_needs_writeback) {
199 		btrfs_dev_replace_unlock(dev_replace);
200 		return 0;
201 	}
202 	btrfs_dev_replace_unlock(dev_replace);
203 
204 	key.objectid = 0;
205 	key.type = BTRFS_DEV_REPLACE_KEY;
206 	key.offset = 0;
207 
208 	path = btrfs_alloc_path();
209 	if (!path) {
210 		ret = -ENOMEM;
211 		goto out;
212 	}
213 	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
214 	if (ret < 0) {
215 		pr_warn("btrfs: error %d while searching for dev_replace item!\n",
216 			ret);
217 		goto out;
218 	}
219 
220 	if (ret == 0 &&
221 	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
222 		/*
223 		 * need to delete old one and insert a new one.
224 		 * Since no attempt is made to recover any old state, if the
225 		 * dev_replace state is 'running', the data on the target
226 		 * drive is lost.
227 		 * It would be possible to recover the state: just make sure
228 		 * that the beginning of the item is never changed and always
229 		 * contains all the essential information. Then read this
230 		 * minimal set of information and use it as a base for the
231 		 * new state.
232 		 */
233 		ret = btrfs_del_item(trans, dev_root, path);
234 		if (ret != 0) {
235 			pr_warn("btrfs: delete too small dev_replace item failed %d!\n",
236 				ret);
237 			goto out;
238 		}
239 		ret = 1;
240 	}
241 
242 	if (ret == 1) {
243 		/* need to insert a new item */
244 		btrfs_release_path(path);
245 		ret = btrfs_insert_empty_item(trans, dev_root, path,
246 					      &key, sizeof(*ptr));
247 		if (ret < 0) {
248 			pr_warn("btrfs: insert dev_replace item failed %d!\n",
249 				ret);
250 			goto out;
251 		}
252 	}
253 
254 	eb = path->nodes[0];
255 	ptr = btrfs_item_ptr(eb, path->slots[0],
256 			     struct btrfs_dev_replace_item);
257 
258 	btrfs_dev_replace_lock(dev_replace);
259 	if (dev_replace->srcdev)
260 		btrfs_set_dev_replace_src_devid(eb, ptr,
261 			dev_replace->srcdev->devid);
262 	else
263 		btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
264 	btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
265 		dev_replace->cont_reading_from_srcdev_mode);
266 	btrfs_set_dev_replace_replace_state(eb, ptr,
267 		dev_replace->replace_state);
268 	btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
269 	btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
270 	btrfs_set_dev_replace_num_write_errors(eb, ptr,
271 		atomic64_read(&dev_replace->num_write_errors));
272 	btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
273 		atomic64_read(&dev_replace->num_uncorrectable_read_errors));
274 	dev_replace->cursor_left_last_write_of_item =
275 		dev_replace->cursor_left;
276 	btrfs_set_dev_replace_cursor_left(eb, ptr,
277 		dev_replace->cursor_left_last_write_of_item);
278 	btrfs_set_dev_replace_cursor_right(eb, ptr,
279 		dev_replace->cursor_right);
280 	dev_replace->item_needs_writeback = 0;
281 	btrfs_dev_replace_unlock(dev_replace);
282 
283 	btrfs_mark_buffer_dirty(eb);
284 
285 out:
286 	btrfs_free_path(path);
287 
288 	return ret;
289 }
290 
291 void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
292 {
293 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
294 
295 	dev_replace->committed_cursor_left =
296 		dev_replace->cursor_left_last_write_of_item;
297 }
298 
299 static u64 btrfs_get_seconds_since_1970(void)
300 {
301 	struct timespec t = CURRENT_TIME_SEC;
302 
303 	return t.tv_sec;
304 }
305 
306 int btrfs_dev_replace_start(struct btrfs_root *root,
307 			    struct btrfs_ioctl_dev_replace_args *args)
308 {
309 	struct btrfs_trans_handle *trans;
310 	struct btrfs_fs_info *fs_info = root->fs_info;
311 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
312 	int ret;
313 	struct btrfs_device *tgt_device = NULL;
314 	struct btrfs_device *src_device = NULL;
315 
316 	if (btrfs_fs_incompat(fs_info, RAID56)) {
317 		pr_warn("btrfs: dev_replace cannot yet handle RAID5/RAID6\n");
318 		return -EINVAL;
319 	}
320 
321 	switch (args->start.cont_reading_from_srcdev_mode) {
322 	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
323 	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
324 		break;
325 	default:
326 		return -EINVAL;
327 	}
328 
329 	if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
330 	    args->start.tgtdev_name[0] == '\0')
331 		return -EINVAL;
332 
333 	mutex_lock(&fs_info->volume_mutex);
334 	ret = btrfs_init_dev_replace_tgtdev(root, args->start.tgtdev_name,
335 					    &tgt_device);
336 	if (ret) {
337 		pr_err("btrfs: target device %s is invalid!\n",
338 		       args->start.tgtdev_name);
339 		mutex_unlock(&fs_info->volume_mutex);
340 		return -EINVAL;
341 	}
342 
343 	ret = btrfs_dev_replace_find_srcdev(root, args->start.srcdevid,
344 					    args->start.srcdev_name,
345 					    &src_device);
346 	mutex_unlock(&fs_info->volume_mutex);
347 	if (ret) {
348 		ret = -EINVAL;
349 		goto leave_no_lock;
350 	}
351 
352 	if (tgt_device->total_bytes < src_device->total_bytes) {
353 		pr_err("btrfs: target device is smaller than source device!\n");
354 		ret = -EINVAL;
355 		goto leave_no_lock;
356 	}
357 
358 	btrfs_dev_replace_lock(dev_replace);
359 	switch (dev_replace->replace_state) {
360 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
361 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
362 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
363 		break;
364 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
365 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
366 		args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
367 		goto leave;
368 	}
369 
370 	dev_replace->cont_reading_from_srcdev_mode =
371 		args->start.cont_reading_from_srcdev_mode;
372 	WARN_ON(!src_device);
373 	dev_replace->srcdev = src_device;
374 	WARN_ON(!tgt_device);
375 	dev_replace->tgtdev = tgt_device;
376 
377 	printk_in_rcu(KERN_INFO
378 		      "btrfs: dev_replace from %s (devid %llu) to %s) started\n",
379 		      src_device->missing ? "<missing disk>" :
380 		        rcu_str_deref(src_device->name),
381 		      src_device->devid,
382 		      rcu_str_deref(tgt_device->name));
383 
384 	tgt_device->total_bytes = src_device->total_bytes;
385 	tgt_device->disk_total_bytes = src_device->disk_total_bytes;
386 	tgt_device->bytes_used = src_device->bytes_used;
387 
388 	/*
389 	 * from now on, the writes to the srcdev are all duplicated to
390 	 * go to the tgtdev as well (refer to btrfs_map_block()).
391 	 */
392 	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
393 	dev_replace->time_started = btrfs_get_seconds_since_1970();
394 	dev_replace->cursor_left = 0;
395 	dev_replace->committed_cursor_left = 0;
396 	dev_replace->cursor_left_last_write_of_item = 0;
397 	dev_replace->cursor_right = 0;
398 	dev_replace->is_valid = 1;
399 	dev_replace->item_needs_writeback = 1;
400 	args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
401 	btrfs_dev_replace_unlock(dev_replace);
402 
403 	btrfs_wait_all_ordered_extents(root->fs_info, 0);
404 
405 	/* force writing the updated state information to disk */
406 	trans = btrfs_start_transaction(root, 0);
407 	if (IS_ERR(trans)) {
408 		ret = PTR_ERR(trans);
409 		btrfs_dev_replace_lock(dev_replace);
410 		goto leave;
411 	}
412 
413 	ret = btrfs_commit_transaction(trans, root);
414 	WARN_ON(ret);
415 
416 	/* the disk copy procedure reuses the scrub code */
417 	ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
418 			      src_device->total_bytes,
419 			      &dev_replace->scrub_progress, 0, 1);
420 
421 	ret = btrfs_dev_replace_finishing(root->fs_info, ret);
422 	WARN_ON(ret);
423 
424 	return 0;
425 
426 leave:
427 	dev_replace->srcdev = NULL;
428 	dev_replace->tgtdev = NULL;
429 	btrfs_dev_replace_unlock(dev_replace);
430 leave_no_lock:
431 	if (tgt_device)
432 		btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
433 	return ret;
434 }
435 
436 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
437 				       int scrub_ret)
438 {
439 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
440 	struct btrfs_device *tgt_device;
441 	struct btrfs_device *src_device;
442 	struct btrfs_root *root = fs_info->tree_root;
443 	u8 uuid_tmp[BTRFS_UUID_SIZE];
444 	struct btrfs_trans_handle *trans;
445 	int ret = 0;
446 
447 	/* don't allow cancel or unmount to disturb the finishing procedure */
448 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
449 
450 	btrfs_dev_replace_lock(dev_replace);
451 	/* was the operation canceled, or is it finished? */
452 	if (dev_replace->replace_state !=
453 	    BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
454 		btrfs_dev_replace_unlock(dev_replace);
455 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
456 		return 0;
457 	}
458 
459 	tgt_device = dev_replace->tgtdev;
460 	src_device = dev_replace->srcdev;
461 	btrfs_dev_replace_unlock(dev_replace);
462 
463 	/* replace old device with new one in mapping tree */
464 	if (!scrub_ret)
465 		btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
466 								src_device,
467 								tgt_device);
468 
469 	/*
470 	 * flush all outstanding I/O and inode extent mappings before the
471 	 * copy operation is declared as being finished
472 	 */
473 	ret = btrfs_start_all_delalloc_inodes(root->fs_info, 0);
474 	if (ret) {
475 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
476 		return ret;
477 	}
478 	btrfs_wait_all_ordered_extents(root->fs_info, 0);
479 
480 	trans = btrfs_start_transaction(root, 0);
481 	if (IS_ERR(trans)) {
482 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
483 		return PTR_ERR(trans);
484 	}
485 	ret = btrfs_commit_transaction(trans, root);
486 	WARN_ON(ret);
487 
488 	/* keep away write_all_supers() during the finishing procedure */
489 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
490 	btrfs_dev_replace_lock(dev_replace);
491 	dev_replace->replace_state =
492 		scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
493 			  : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
494 	dev_replace->tgtdev = NULL;
495 	dev_replace->srcdev = NULL;
496 	dev_replace->time_stopped = btrfs_get_seconds_since_1970();
497 	dev_replace->item_needs_writeback = 1;
498 
499 	if (scrub_ret) {
500 		printk_in_rcu(KERN_ERR
501 			      "btrfs: btrfs_scrub_dev(%s, %llu, %s) failed %d\n",
502 			      src_device->missing ? "<missing disk>" :
503 			        rcu_str_deref(src_device->name),
504 			      src_device->devid,
505 			      rcu_str_deref(tgt_device->name), scrub_ret);
506 		btrfs_dev_replace_unlock(dev_replace);
507 		mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
508 		if (tgt_device)
509 			btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
510 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
511 
512 		return 0;
513 	}
514 
515 	printk_in_rcu(KERN_INFO
516 		      "btrfs: dev_replace from %s (devid %llu) to %s) finished\n",
517 		      src_device->missing ? "<missing disk>" :
518 		        rcu_str_deref(src_device->name),
519 		      src_device->devid,
520 		      rcu_str_deref(tgt_device->name));
521 	tgt_device->is_tgtdev_for_dev_replace = 0;
522 	tgt_device->devid = src_device->devid;
523 	src_device->devid = BTRFS_DEV_REPLACE_DEVID;
524 	tgt_device->bytes_used = src_device->bytes_used;
525 	memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
526 	memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
527 	memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
528 	tgt_device->total_bytes = src_device->total_bytes;
529 	tgt_device->disk_total_bytes = src_device->disk_total_bytes;
530 	tgt_device->bytes_used = src_device->bytes_used;
531 	if (fs_info->sb->s_bdev == src_device->bdev)
532 		fs_info->sb->s_bdev = tgt_device->bdev;
533 	if (fs_info->fs_devices->latest_bdev == src_device->bdev)
534 		fs_info->fs_devices->latest_bdev = tgt_device->bdev;
535 	list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
536 
537 	btrfs_rm_dev_replace_srcdev(fs_info, src_device);
538 	if (src_device->bdev) {
539 		/* zero out the old super */
540 		btrfs_scratch_superblock(src_device);
541 	}
542 	/*
543 	 * this is again a consistent state where no dev_replace procedure
544 	 * is running, the target device is part of the filesystem, the
545 	 * source device is not part of the filesystem anymore and its 1st
546 	 * superblock is scratched out so that it is no longer marked to
547 	 * belong to this filesystem.
548 	 */
549 	btrfs_dev_replace_unlock(dev_replace);
550 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
551 
552 	/* write back the superblocks */
553 	trans = btrfs_start_transaction(root, 0);
554 	if (!IS_ERR(trans))
555 		btrfs_commit_transaction(trans, root);
556 
557 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
558 
559 	return 0;
560 }
561 
562 static void btrfs_dev_replace_update_device_in_mapping_tree(
563 						struct btrfs_fs_info *fs_info,
564 						struct btrfs_device *srcdev,
565 						struct btrfs_device *tgtdev)
566 {
567 	struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
568 	struct extent_map *em;
569 	struct map_lookup *map;
570 	u64 start = 0;
571 	int i;
572 
573 	write_lock(&em_tree->lock);
574 	do {
575 		em = lookup_extent_mapping(em_tree, start, (u64)-1);
576 		if (!em)
577 			break;
578 		map = (struct map_lookup *)em->bdev;
579 		for (i = 0; i < map->num_stripes; i++)
580 			if (srcdev == map->stripes[i].dev)
581 				map->stripes[i].dev = tgtdev;
582 		start = em->start + em->len;
583 		free_extent_map(em);
584 	} while (start);
585 	write_unlock(&em_tree->lock);
586 }
587 
588 static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid,
589 					 char *srcdev_name,
590 					 struct btrfs_device **device)
591 {
592 	int ret;
593 
594 	if (srcdevid) {
595 		ret = 0;
596 		*device = btrfs_find_device(root->fs_info, srcdevid, NULL,
597 					    NULL);
598 		if (!*device)
599 			ret = -ENOENT;
600 	} else {
601 		ret = btrfs_find_device_missing_or_by_path(root, srcdev_name,
602 							   device);
603 	}
604 	return ret;
605 }
606 
607 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
608 			      struct btrfs_ioctl_dev_replace_args *args)
609 {
610 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
611 
612 	btrfs_dev_replace_lock(dev_replace);
613 	/* even if !dev_replace_is_valid, the values are good enough for
614 	 * the replace_status ioctl */
615 	args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
616 	args->status.replace_state = dev_replace->replace_state;
617 	args->status.time_started = dev_replace->time_started;
618 	args->status.time_stopped = dev_replace->time_stopped;
619 	args->status.num_write_errors =
620 		atomic64_read(&dev_replace->num_write_errors);
621 	args->status.num_uncorrectable_read_errors =
622 		atomic64_read(&dev_replace->num_uncorrectable_read_errors);
623 	switch (dev_replace->replace_state) {
624 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
625 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
626 		args->status.progress_1000 = 0;
627 		break;
628 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
629 		args->status.progress_1000 = 1000;
630 		break;
631 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
632 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
633 		args->status.progress_1000 = div64_u64(dev_replace->cursor_left,
634 			div64_u64(dev_replace->srcdev->total_bytes, 1000));
635 		break;
636 	}
637 	btrfs_dev_replace_unlock(dev_replace);
638 }
639 
640 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info,
641 			     struct btrfs_ioctl_dev_replace_args *args)
642 {
643 	args->result = __btrfs_dev_replace_cancel(fs_info);
644 	return 0;
645 }
646 
647 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
648 {
649 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
650 	struct btrfs_device *tgt_device = NULL;
651 	struct btrfs_trans_handle *trans;
652 	struct btrfs_root *root = fs_info->tree_root;
653 	u64 result;
654 	int ret;
655 
656 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
657 	btrfs_dev_replace_lock(dev_replace);
658 	switch (dev_replace->replace_state) {
659 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
660 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
661 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
662 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
663 		btrfs_dev_replace_unlock(dev_replace);
664 		goto leave;
665 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
666 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
667 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
668 		tgt_device = dev_replace->tgtdev;
669 		dev_replace->tgtdev = NULL;
670 		dev_replace->srcdev = NULL;
671 		break;
672 	}
673 	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
674 	dev_replace->time_stopped = btrfs_get_seconds_since_1970();
675 	dev_replace->item_needs_writeback = 1;
676 	btrfs_dev_replace_unlock(dev_replace);
677 	btrfs_scrub_cancel(fs_info);
678 
679 	trans = btrfs_start_transaction(root, 0);
680 	if (IS_ERR(trans)) {
681 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
682 		return PTR_ERR(trans);
683 	}
684 	ret = btrfs_commit_transaction(trans, root);
685 	WARN_ON(ret);
686 	if (tgt_device)
687 		btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
688 
689 leave:
690 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
691 	return result;
692 }
693 
694 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
695 {
696 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
697 
698 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
699 	btrfs_dev_replace_lock(dev_replace);
700 	switch (dev_replace->replace_state) {
701 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
702 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
703 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
704 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
705 		break;
706 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
707 		dev_replace->replace_state =
708 			BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
709 		dev_replace->time_stopped = btrfs_get_seconds_since_1970();
710 		dev_replace->item_needs_writeback = 1;
711 		pr_info("btrfs: suspending dev_replace for unmount\n");
712 		break;
713 	}
714 
715 	btrfs_dev_replace_unlock(dev_replace);
716 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
717 }
718 
719 /* resume dev_replace procedure that was interrupted by unmount */
720 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
721 {
722 	struct task_struct *task;
723 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
724 
725 	btrfs_dev_replace_lock(dev_replace);
726 	switch (dev_replace->replace_state) {
727 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
728 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
729 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
730 		btrfs_dev_replace_unlock(dev_replace);
731 		return 0;
732 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
733 		break;
734 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
735 		dev_replace->replace_state =
736 			BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
737 		break;
738 	}
739 	if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
740 		pr_info("btrfs: cannot continue dev_replace, tgtdev is missing\n"
741 			"btrfs: you may cancel the operation after 'mount -o degraded'\n");
742 		btrfs_dev_replace_unlock(dev_replace);
743 		return 0;
744 	}
745 	btrfs_dev_replace_unlock(dev_replace);
746 
747 	WARN_ON(atomic_xchg(
748 		&fs_info->mutually_exclusive_operation_running, 1));
749 	task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
750 	return PTR_RET(task);
751 }
752 
753 static int btrfs_dev_replace_kthread(void *data)
754 {
755 	struct btrfs_fs_info *fs_info = data;
756 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
757 	struct btrfs_ioctl_dev_replace_args *status_args;
758 	u64 progress;
759 
760 	status_args = kzalloc(sizeof(*status_args), GFP_NOFS);
761 	if (status_args) {
762 		btrfs_dev_replace_status(fs_info, status_args);
763 		progress = status_args->status.progress_1000;
764 		kfree(status_args);
765 		do_div(progress, 10);
766 		printk_in_rcu(KERN_INFO
767 			      "btrfs: continuing dev_replace from %s (devid %llu) to %s @%u%%\n",
768 			      dev_replace->srcdev->missing ? "<missing disk>" :
769 				rcu_str_deref(dev_replace->srcdev->name),
770 			      dev_replace->srcdev->devid,
771 			      dev_replace->tgtdev ?
772 				rcu_str_deref(dev_replace->tgtdev->name) :
773 				"<missing target disk>",
774 			      (unsigned int)progress);
775 	}
776 	btrfs_dev_replace_continue_on_mount(fs_info);
777 	atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
778 
779 	return 0;
780 }
781 
782 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info)
783 {
784 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
785 	int ret;
786 
787 	ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
788 			      dev_replace->committed_cursor_left,
789 			      dev_replace->srcdev->total_bytes,
790 			      &dev_replace->scrub_progress, 0, 1);
791 	ret = btrfs_dev_replace_finishing(fs_info, ret);
792 	WARN_ON(ret);
793 	return 0;
794 }
795 
796 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
797 {
798 	if (!dev_replace->is_valid)
799 		return 0;
800 
801 	switch (dev_replace->replace_state) {
802 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
803 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
804 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
805 		return 0;
806 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
807 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
808 		/*
809 		 * return true even if tgtdev is missing (this is
810 		 * something that can happen if the dev_replace
811 		 * procedure is suspended by an umount and then
812 		 * the tgtdev is missing (or "btrfs dev scan") was
813 		 * not called and the the filesystem is remounted
814 		 * in degraded state. This does not stop the
815 		 * dev_replace procedure. It needs to be canceled
816 		 * manually if the cancelation is wanted.
817 		 */
818 		break;
819 	}
820 	return 1;
821 }
822 
823 void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace)
824 {
825 	/* the beginning is just an optimization for the typical case */
826 	if (atomic_read(&dev_replace->nesting_level) == 0) {
827 acquire_lock:
828 		/* this is not a nested case where the same thread
829 		 * is trying to acqurire the same lock twice */
830 		mutex_lock(&dev_replace->lock);
831 		mutex_lock(&dev_replace->lock_management_lock);
832 		dev_replace->lock_owner = current->pid;
833 		atomic_inc(&dev_replace->nesting_level);
834 		mutex_unlock(&dev_replace->lock_management_lock);
835 		return;
836 	}
837 
838 	mutex_lock(&dev_replace->lock_management_lock);
839 	if (atomic_read(&dev_replace->nesting_level) > 0 &&
840 	    dev_replace->lock_owner == current->pid) {
841 		WARN_ON(!mutex_is_locked(&dev_replace->lock));
842 		atomic_inc(&dev_replace->nesting_level);
843 		mutex_unlock(&dev_replace->lock_management_lock);
844 		return;
845 	}
846 
847 	mutex_unlock(&dev_replace->lock_management_lock);
848 	goto acquire_lock;
849 }
850 
851 void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace)
852 {
853 	WARN_ON(!mutex_is_locked(&dev_replace->lock));
854 	mutex_lock(&dev_replace->lock_management_lock);
855 	WARN_ON(atomic_read(&dev_replace->nesting_level) < 1);
856 	WARN_ON(dev_replace->lock_owner != current->pid);
857 	atomic_dec(&dev_replace->nesting_level);
858 	if (atomic_read(&dev_replace->nesting_level) == 0) {
859 		dev_replace->lock_owner = 0;
860 		mutex_unlock(&dev_replace->lock_management_lock);
861 		mutex_unlock(&dev_replace->lock);
862 	} else {
863 		mutex_unlock(&dev_replace->lock_management_lock);
864 	}
865 }
866