xref: /openbmc/linux/fs/btrfs/dev-replace.c (revision ca460cc2)
1 /*
2  * Copyright (C) STRATO AG 2012.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/kthread.h>
27 #include <linux/math64.h>
28 #include <asm/div64.h>
29 #include "ctree.h"
30 #include "extent_map.h"
31 #include "disk-io.h"
32 #include "transaction.h"
33 #include "print-tree.h"
34 #include "volumes.h"
35 #include "async-thread.h"
36 #include "check-integrity.h"
37 #include "rcu-string.h"
38 #include "dev-replace.h"
39 #include "sysfs.h"
40 
41 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
42 				       int scrub_ret);
43 static void btrfs_dev_replace_update_device_in_mapping_tree(
44 						struct btrfs_fs_info *fs_info,
45 						struct btrfs_device *srcdev,
46 						struct btrfs_device *tgtdev);
47 static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid,
48 					 char *srcdev_name,
49 					 struct btrfs_device **device);
50 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info);
51 static int btrfs_dev_replace_kthread(void *data);
52 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info);
53 
54 
55 int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
56 {
57 	struct btrfs_key key;
58 	struct btrfs_root *dev_root = fs_info->dev_root;
59 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
60 	struct extent_buffer *eb;
61 	int slot;
62 	int ret = 0;
63 	struct btrfs_path *path = NULL;
64 	int item_size;
65 	struct btrfs_dev_replace_item *ptr;
66 	u64 src_devid;
67 
68 	path = btrfs_alloc_path();
69 	if (!path) {
70 		ret = -ENOMEM;
71 		goto out;
72 	}
73 
74 	key.objectid = 0;
75 	key.type = BTRFS_DEV_REPLACE_KEY;
76 	key.offset = 0;
77 	ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
78 	if (ret) {
79 no_valid_dev_replace_entry_found:
80 		ret = 0;
81 		dev_replace->replace_state =
82 			BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED;
83 		dev_replace->cont_reading_from_srcdev_mode =
84 		    BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
85 		dev_replace->replace_state = 0;
86 		dev_replace->time_started = 0;
87 		dev_replace->time_stopped = 0;
88 		atomic64_set(&dev_replace->num_write_errors, 0);
89 		atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
90 		dev_replace->cursor_left = 0;
91 		dev_replace->committed_cursor_left = 0;
92 		dev_replace->cursor_left_last_write_of_item = 0;
93 		dev_replace->cursor_right = 0;
94 		dev_replace->srcdev = NULL;
95 		dev_replace->tgtdev = NULL;
96 		dev_replace->is_valid = 0;
97 		dev_replace->item_needs_writeback = 0;
98 		goto out;
99 	}
100 	slot = path->slots[0];
101 	eb = path->nodes[0];
102 	item_size = btrfs_item_size_nr(eb, slot);
103 	ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
104 
105 	if (item_size != sizeof(struct btrfs_dev_replace_item)) {
106 		btrfs_warn(fs_info,
107 			"dev_replace entry found has unexpected size, ignore entry");
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 			btrfs_warn(fs_info,
151 			   "cannot mount because device replace operation is ongoing and");
152 			btrfs_warn(fs_info,
153 			   "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
154 			   src_devid);
155 		}
156 		if (!dev_replace->tgtdev &&
157 		    !btrfs_test_opt(dev_root, DEGRADED)) {
158 			ret = -EIO;
159 			btrfs_warn(fs_info,
160 			   "cannot mount because device replace operation is ongoing and");
161 			btrfs_warn(fs_info,
162 			   "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
163 				BTRFS_DEV_REPLACE_DEVID);
164 		}
165 		if (dev_replace->tgtdev) {
166 			if (dev_replace->srcdev) {
167 				dev_replace->tgtdev->total_bytes =
168 					dev_replace->srcdev->total_bytes;
169 				dev_replace->tgtdev->disk_total_bytes =
170 					dev_replace->srcdev->disk_total_bytes;
171 				dev_replace->tgtdev->commit_total_bytes =
172 					dev_replace->srcdev->commit_total_bytes;
173 				dev_replace->tgtdev->bytes_used =
174 					dev_replace->srcdev->bytes_used;
175 				dev_replace->tgtdev->commit_bytes_used =
176 					dev_replace->srcdev->commit_bytes_used;
177 			}
178 			dev_replace->tgtdev->is_tgtdev_for_dev_replace = 1;
179 			btrfs_init_dev_replace_tgtdev_for_resume(fs_info,
180 				dev_replace->tgtdev);
181 		}
182 		break;
183 	}
184 
185 out:
186 	if (path)
187 		btrfs_free_path(path);
188 	return ret;
189 }
190 
191 /*
192  * called from commit_transaction. Writes changed device replace state to
193  * disk.
194  */
195 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
196 			  struct btrfs_fs_info *fs_info)
197 {
198 	int ret;
199 	struct btrfs_root *dev_root = fs_info->dev_root;
200 	struct btrfs_path *path;
201 	struct btrfs_key key;
202 	struct extent_buffer *eb;
203 	struct btrfs_dev_replace_item *ptr;
204 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
205 
206 	btrfs_dev_replace_lock(dev_replace);
207 	if (!dev_replace->is_valid ||
208 	    !dev_replace->item_needs_writeback) {
209 		btrfs_dev_replace_unlock(dev_replace);
210 		return 0;
211 	}
212 	btrfs_dev_replace_unlock(dev_replace);
213 
214 	key.objectid = 0;
215 	key.type = BTRFS_DEV_REPLACE_KEY;
216 	key.offset = 0;
217 
218 	path = btrfs_alloc_path();
219 	if (!path) {
220 		ret = -ENOMEM;
221 		goto out;
222 	}
223 	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
224 	if (ret < 0) {
225 		btrfs_warn(fs_info, "error %d while searching for dev_replace item!",
226 			ret);
227 		goto out;
228 	}
229 
230 	if (ret == 0 &&
231 	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
232 		/*
233 		 * need to delete old one and insert a new one.
234 		 * Since no attempt is made to recover any old state, if the
235 		 * dev_replace state is 'running', the data on the target
236 		 * drive is lost.
237 		 * It would be possible to recover the state: just make sure
238 		 * that the beginning of the item is never changed and always
239 		 * contains all the essential information. Then read this
240 		 * minimal set of information and use it as a base for the
241 		 * new state.
242 		 */
243 		ret = btrfs_del_item(trans, dev_root, path);
244 		if (ret != 0) {
245 			btrfs_warn(fs_info, "delete too small dev_replace item failed %d!",
246 				ret);
247 			goto out;
248 		}
249 		ret = 1;
250 	}
251 
252 	if (ret == 1) {
253 		/* need to insert a new item */
254 		btrfs_release_path(path);
255 		ret = btrfs_insert_empty_item(trans, dev_root, path,
256 					      &key, sizeof(*ptr));
257 		if (ret < 0) {
258 			btrfs_warn(fs_info, "insert dev_replace item failed %d!",
259 				ret);
260 			goto out;
261 		}
262 	}
263 
264 	eb = path->nodes[0];
265 	ptr = btrfs_item_ptr(eb, path->slots[0],
266 			     struct btrfs_dev_replace_item);
267 
268 	btrfs_dev_replace_lock(dev_replace);
269 	if (dev_replace->srcdev)
270 		btrfs_set_dev_replace_src_devid(eb, ptr,
271 			dev_replace->srcdev->devid);
272 	else
273 		btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
274 	btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
275 		dev_replace->cont_reading_from_srcdev_mode);
276 	btrfs_set_dev_replace_replace_state(eb, ptr,
277 		dev_replace->replace_state);
278 	btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
279 	btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
280 	btrfs_set_dev_replace_num_write_errors(eb, ptr,
281 		atomic64_read(&dev_replace->num_write_errors));
282 	btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
283 		atomic64_read(&dev_replace->num_uncorrectable_read_errors));
284 	dev_replace->cursor_left_last_write_of_item =
285 		dev_replace->cursor_left;
286 	btrfs_set_dev_replace_cursor_left(eb, ptr,
287 		dev_replace->cursor_left_last_write_of_item);
288 	btrfs_set_dev_replace_cursor_right(eb, ptr,
289 		dev_replace->cursor_right);
290 	dev_replace->item_needs_writeback = 0;
291 	btrfs_dev_replace_unlock(dev_replace);
292 
293 	btrfs_mark_buffer_dirty(eb);
294 
295 out:
296 	btrfs_free_path(path);
297 
298 	return ret;
299 }
300 
301 void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
302 {
303 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
304 
305 	dev_replace->committed_cursor_left =
306 		dev_replace->cursor_left_last_write_of_item;
307 }
308 
309 int btrfs_dev_replace_start(struct btrfs_root *root,
310 			    struct btrfs_ioctl_dev_replace_args *args)
311 {
312 	struct btrfs_trans_handle *trans;
313 	struct btrfs_fs_info *fs_info = root->fs_info;
314 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
315 	int ret;
316 	struct btrfs_device *tgt_device = NULL;
317 	struct btrfs_device *src_device = NULL;
318 
319 	if (btrfs_fs_incompat(fs_info, RAID56)) {
320 		btrfs_warn(fs_info, "dev_replace cannot yet handle RAID5/RAID6");
321 		return -EOPNOTSUPP;
322 	}
323 
324 	switch (args->start.cont_reading_from_srcdev_mode) {
325 	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
326 	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
327 		break;
328 	default:
329 		return -EINVAL;
330 	}
331 
332 	if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
333 	    args->start.tgtdev_name[0] == '\0')
334 		return -EINVAL;
335 
336 	/*
337 	 * Here we commit the transaction to make sure commit_total_bytes
338 	 * of all the devices are updated.
339 	 */
340 	trans = btrfs_attach_transaction(root);
341 	if (!IS_ERR(trans)) {
342 		ret = btrfs_commit_transaction(trans, root);
343 		if (ret)
344 			return ret;
345 	} else if (PTR_ERR(trans) != -ENOENT) {
346 		return PTR_ERR(trans);
347 	}
348 
349 	/* the disk copy procedure reuses the scrub code */
350 	mutex_lock(&fs_info->volume_mutex);
351 	ret = btrfs_dev_replace_find_srcdev(root, args->start.srcdevid,
352 					    args->start.srcdev_name,
353 					    &src_device);
354 	if (ret) {
355 		mutex_unlock(&fs_info->volume_mutex);
356 		return ret;
357 	}
358 
359 	ret = btrfs_init_dev_replace_tgtdev(root, args->start.tgtdev_name,
360 					    src_device, &tgt_device);
361 	mutex_unlock(&fs_info->volume_mutex);
362 	if (ret)
363 		return ret;
364 
365 	btrfs_dev_replace_lock(dev_replace);
366 	switch (dev_replace->replace_state) {
367 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
368 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
369 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
370 		break;
371 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
372 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
373 		args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
374 		goto leave;
375 	}
376 
377 	dev_replace->cont_reading_from_srcdev_mode =
378 		args->start.cont_reading_from_srcdev_mode;
379 	WARN_ON(!src_device);
380 	dev_replace->srcdev = src_device;
381 	WARN_ON(!tgt_device);
382 	dev_replace->tgtdev = tgt_device;
383 
384 	printk_in_rcu(KERN_INFO
385 		      "BTRFS: dev_replace from %s (devid %llu) to %s started\n",
386 		      src_device->missing ? "<missing disk>" :
387 		        rcu_str_deref(src_device->name),
388 		      src_device->devid,
389 		      rcu_str_deref(tgt_device->name));
390 
391 	/*
392 	 * from now on, the writes to the srcdev are all duplicated to
393 	 * go to the tgtdev as well (refer to btrfs_map_block()).
394 	 */
395 	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
396 	dev_replace->time_started = get_seconds();
397 	dev_replace->cursor_left = 0;
398 	dev_replace->committed_cursor_left = 0;
399 	dev_replace->cursor_left_last_write_of_item = 0;
400 	dev_replace->cursor_right = 0;
401 	dev_replace->is_valid = 1;
402 	dev_replace->item_needs_writeback = 1;
403 	args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
404 	btrfs_dev_replace_unlock(dev_replace);
405 
406 	btrfs_wait_ordered_roots(root->fs_info, -1);
407 
408 	/* force writing the updated state information to disk */
409 	trans = btrfs_start_transaction(root, 0);
410 	if (IS_ERR(trans)) {
411 		ret = PTR_ERR(trans);
412 		btrfs_dev_replace_lock(dev_replace);
413 		goto leave;
414 	}
415 
416 	ret = btrfs_commit_transaction(trans, root);
417 	WARN_ON(ret);
418 
419 	/* the disk copy procedure reuses the scrub code */
420 	ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
421 			      btrfs_device_get_total_bytes(src_device),
422 			      &dev_replace->scrub_progress, 0, 1);
423 
424 	ret = btrfs_dev_replace_finishing(root->fs_info, ret);
425 	WARN_ON(ret);
426 
427 	return 0;
428 
429 leave:
430 	dev_replace->srcdev = NULL;
431 	dev_replace->tgtdev = NULL;
432 	btrfs_dev_replace_unlock(dev_replace);
433 	btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
434 	return ret;
435 }
436 
437 /*
438  * blocked until all flighting bios are finished.
439  */
440 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
441 {
442 	s64 writers;
443 	DEFINE_WAIT(wait);
444 
445 	set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
446 	do {
447 		prepare_to_wait(&fs_info->replace_wait, &wait,
448 				TASK_UNINTERRUPTIBLE);
449 		writers = percpu_counter_sum(&fs_info->bio_counter);
450 		if (writers)
451 			schedule();
452 		finish_wait(&fs_info->replace_wait, &wait);
453 	} while (writers);
454 }
455 
456 /*
457  * we have removed target device, it is safe to allow new bios request.
458  */
459 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
460 {
461 	clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
462 	if (waitqueue_active(&fs_info->replace_wait))
463 		wake_up(&fs_info->replace_wait);
464 }
465 
466 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
467 				       int scrub_ret)
468 {
469 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
470 	struct btrfs_device *tgt_device;
471 	struct btrfs_device *src_device;
472 	struct btrfs_root *root = fs_info->tree_root;
473 	u8 uuid_tmp[BTRFS_UUID_SIZE];
474 	struct btrfs_trans_handle *trans;
475 	int ret = 0;
476 
477 	/* don't allow cancel or unmount to disturb the finishing procedure */
478 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
479 
480 	btrfs_dev_replace_lock(dev_replace);
481 	/* was the operation canceled, or is it finished? */
482 	if (dev_replace->replace_state !=
483 	    BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
484 		btrfs_dev_replace_unlock(dev_replace);
485 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
486 		return 0;
487 	}
488 
489 	tgt_device = dev_replace->tgtdev;
490 	src_device = dev_replace->srcdev;
491 	btrfs_dev_replace_unlock(dev_replace);
492 
493 	/*
494 	 * flush all outstanding I/O and inode extent mappings before the
495 	 * copy operation is declared as being finished
496 	 */
497 	ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
498 	if (ret) {
499 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
500 		return ret;
501 	}
502 	btrfs_wait_ordered_roots(root->fs_info, -1);
503 
504 	trans = btrfs_start_transaction(root, 0);
505 	if (IS_ERR(trans)) {
506 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
507 		return PTR_ERR(trans);
508 	}
509 	ret = btrfs_commit_transaction(trans, root);
510 	WARN_ON(ret);
511 
512 	mutex_lock(&uuid_mutex);
513 	/* keep away write_all_supers() during the finishing procedure */
514 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
515 	mutex_lock(&root->fs_info->chunk_mutex);
516 	btrfs_dev_replace_lock(dev_replace);
517 	dev_replace->replace_state =
518 		scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
519 			  : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
520 	dev_replace->tgtdev = NULL;
521 	dev_replace->srcdev = NULL;
522 	dev_replace->time_stopped = get_seconds();
523 	dev_replace->item_needs_writeback = 1;
524 
525 	/* replace old device with new one in mapping tree */
526 	if (!scrub_ret) {
527 		btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
528 								src_device,
529 								tgt_device);
530 	} else {
531 		printk_in_rcu(KERN_ERR
532 			      "BTRFS: btrfs_scrub_dev(%s, %llu, %s) failed %d\n",
533 			      src_device->missing ? "<missing disk>" :
534 			        rcu_str_deref(src_device->name),
535 			      src_device->devid,
536 			      rcu_str_deref(tgt_device->name), scrub_ret);
537 		btrfs_dev_replace_unlock(dev_replace);
538 		mutex_unlock(&root->fs_info->chunk_mutex);
539 		mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
540 		mutex_unlock(&uuid_mutex);
541 		if (tgt_device)
542 			btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
543 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
544 
545 		return 0;
546 	}
547 
548 	printk_in_rcu(KERN_INFO
549 		      "BTRFS: dev_replace from %s (devid %llu) to %s finished\n",
550 		      src_device->missing ? "<missing disk>" :
551 		        rcu_str_deref(src_device->name),
552 		      src_device->devid,
553 		      rcu_str_deref(tgt_device->name));
554 	tgt_device->is_tgtdev_for_dev_replace = 0;
555 	tgt_device->devid = src_device->devid;
556 	src_device->devid = BTRFS_DEV_REPLACE_DEVID;
557 	memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
558 	memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
559 	memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
560 	btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
561 	btrfs_device_set_disk_total_bytes(tgt_device,
562 					  src_device->disk_total_bytes);
563 	btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
564 	ASSERT(list_empty(&src_device->resized_list));
565 	tgt_device->commit_total_bytes = src_device->commit_total_bytes;
566 	tgt_device->commit_bytes_used = src_device->bytes_used;
567 	if (fs_info->sb->s_bdev == src_device->bdev)
568 		fs_info->sb->s_bdev = tgt_device->bdev;
569 	if (fs_info->fs_devices->latest_bdev == src_device->bdev)
570 		fs_info->fs_devices->latest_bdev = tgt_device->bdev;
571 	list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
572 	fs_info->fs_devices->rw_devices++;
573 
574 	/* replace the sysfs entry */
575 	btrfs_kobj_rm_device(fs_info, src_device);
576 	btrfs_kobj_add_device(fs_info, tgt_device);
577 
578 	btrfs_dev_replace_unlock(dev_replace);
579 
580 	btrfs_rm_dev_replace_blocked(fs_info);
581 
582 	btrfs_rm_dev_replace_srcdev(fs_info, src_device);
583 
584 	btrfs_rm_dev_replace_unblocked(fs_info);
585 
586 	/*
587 	 * this is again a consistent state where no dev_replace procedure
588 	 * is running, the target device is part of the filesystem, the
589 	 * source device is not part of the filesystem anymore and its 1st
590 	 * superblock is scratched out so that it is no longer marked to
591 	 * belong to this filesystem.
592 	 */
593 	mutex_unlock(&root->fs_info->chunk_mutex);
594 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
595 	mutex_unlock(&uuid_mutex);
596 
597 	/* write back the superblocks */
598 	trans = btrfs_start_transaction(root, 0);
599 	if (!IS_ERR(trans))
600 		btrfs_commit_transaction(trans, root);
601 
602 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
603 
604 	return 0;
605 }
606 
607 static void btrfs_dev_replace_update_device_in_mapping_tree(
608 						struct btrfs_fs_info *fs_info,
609 						struct btrfs_device *srcdev,
610 						struct btrfs_device *tgtdev)
611 {
612 	struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
613 	struct extent_map *em;
614 	struct map_lookup *map;
615 	u64 start = 0;
616 	int i;
617 
618 	write_lock(&em_tree->lock);
619 	do {
620 		em = lookup_extent_mapping(em_tree, start, (u64)-1);
621 		if (!em)
622 			break;
623 		map = (struct map_lookup *)em->bdev;
624 		for (i = 0; i < map->num_stripes; i++)
625 			if (srcdev == map->stripes[i].dev)
626 				map->stripes[i].dev = tgtdev;
627 		start = em->start + em->len;
628 		free_extent_map(em);
629 	} while (start);
630 	write_unlock(&em_tree->lock);
631 }
632 
633 static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid,
634 					 char *srcdev_name,
635 					 struct btrfs_device **device)
636 {
637 	int ret;
638 
639 	if (srcdevid) {
640 		ret = 0;
641 		*device = btrfs_find_device(root->fs_info, srcdevid, NULL,
642 					    NULL);
643 		if (!*device)
644 			ret = -ENOENT;
645 	} else {
646 		ret = btrfs_find_device_missing_or_by_path(root, srcdev_name,
647 							   device);
648 	}
649 	return ret;
650 }
651 
652 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
653 			      struct btrfs_ioctl_dev_replace_args *args)
654 {
655 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
656 	struct btrfs_device *srcdev;
657 
658 	btrfs_dev_replace_lock(dev_replace);
659 	/* even if !dev_replace_is_valid, the values are good enough for
660 	 * the replace_status ioctl */
661 	args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
662 	args->status.replace_state = dev_replace->replace_state;
663 	args->status.time_started = dev_replace->time_started;
664 	args->status.time_stopped = dev_replace->time_stopped;
665 	args->status.num_write_errors =
666 		atomic64_read(&dev_replace->num_write_errors);
667 	args->status.num_uncorrectable_read_errors =
668 		atomic64_read(&dev_replace->num_uncorrectable_read_errors);
669 	switch (dev_replace->replace_state) {
670 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
671 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
672 		args->status.progress_1000 = 0;
673 		break;
674 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
675 		args->status.progress_1000 = 1000;
676 		break;
677 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
678 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
679 		srcdev = dev_replace->srcdev;
680 		args->status.progress_1000 = div64_u64(dev_replace->cursor_left,
681 			div64_u64(btrfs_device_get_total_bytes(srcdev), 1000));
682 		break;
683 	}
684 	btrfs_dev_replace_unlock(dev_replace);
685 }
686 
687 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info,
688 			     struct btrfs_ioctl_dev_replace_args *args)
689 {
690 	args->result = __btrfs_dev_replace_cancel(fs_info);
691 	return 0;
692 }
693 
694 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
695 {
696 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
697 	struct btrfs_device *tgt_device = NULL;
698 	struct btrfs_trans_handle *trans;
699 	struct btrfs_root *root = fs_info->tree_root;
700 	u64 result;
701 	int ret;
702 
703 	if (fs_info->sb->s_flags & MS_RDONLY)
704 		return -EROFS;
705 
706 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
707 	btrfs_dev_replace_lock(dev_replace);
708 	switch (dev_replace->replace_state) {
709 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
710 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
711 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
712 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
713 		btrfs_dev_replace_unlock(dev_replace);
714 		goto leave;
715 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
716 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
717 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
718 		tgt_device = dev_replace->tgtdev;
719 		dev_replace->tgtdev = NULL;
720 		dev_replace->srcdev = NULL;
721 		break;
722 	}
723 	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
724 	dev_replace->time_stopped = get_seconds();
725 	dev_replace->item_needs_writeback = 1;
726 	btrfs_dev_replace_unlock(dev_replace);
727 	btrfs_scrub_cancel(fs_info);
728 
729 	trans = btrfs_start_transaction(root, 0);
730 	if (IS_ERR(trans)) {
731 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
732 		return PTR_ERR(trans);
733 	}
734 	ret = btrfs_commit_transaction(trans, root);
735 	WARN_ON(ret);
736 	if (tgt_device)
737 		btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
738 
739 leave:
740 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
741 	return result;
742 }
743 
744 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
745 {
746 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
747 
748 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
749 	btrfs_dev_replace_lock(dev_replace);
750 	switch (dev_replace->replace_state) {
751 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
752 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
753 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
754 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
755 		break;
756 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
757 		dev_replace->replace_state =
758 			BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
759 		dev_replace->time_stopped = get_seconds();
760 		dev_replace->item_needs_writeback = 1;
761 		btrfs_info(fs_info, "suspending dev_replace for unmount");
762 		break;
763 	}
764 
765 	btrfs_dev_replace_unlock(dev_replace);
766 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
767 }
768 
769 /* resume dev_replace procedure that was interrupted by unmount */
770 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
771 {
772 	struct task_struct *task;
773 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
774 
775 	btrfs_dev_replace_lock(dev_replace);
776 	switch (dev_replace->replace_state) {
777 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
778 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
779 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
780 		btrfs_dev_replace_unlock(dev_replace);
781 		return 0;
782 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
783 		break;
784 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
785 		dev_replace->replace_state =
786 			BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
787 		break;
788 	}
789 	if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
790 		btrfs_info(fs_info, "cannot continue dev_replace, tgtdev is missing");
791 		btrfs_info(fs_info,
792 			"you may cancel the operation after 'mount -o degraded'");
793 		btrfs_dev_replace_unlock(dev_replace);
794 		return 0;
795 	}
796 	btrfs_dev_replace_unlock(dev_replace);
797 
798 	WARN_ON(atomic_xchg(
799 		&fs_info->mutually_exclusive_operation_running, 1));
800 	task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
801 	return PTR_ERR_OR_ZERO(task);
802 }
803 
804 static int btrfs_dev_replace_kthread(void *data)
805 {
806 	struct btrfs_fs_info *fs_info = data;
807 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
808 	struct btrfs_ioctl_dev_replace_args *status_args;
809 	u64 progress;
810 
811 	status_args = kzalloc(sizeof(*status_args), GFP_NOFS);
812 	if (status_args) {
813 		btrfs_dev_replace_status(fs_info, status_args);
814 		progress = status_args->status.progress_1000;
815 		kfree(status_args);
816 		do_div(progress, 10);
817 		printk_in_rcu(KERN_INFO
818 			"BTRFS: continuing dev_replace from %s (devid %llu) to %s @%u%%\n",
819 			dev_replace->srcdev->missing ? "<missing disk>" :
820 			rcu_str_deref(dev_replace->srcdev->name),
821 			dev_replace->srcdev->devid,
822 			dev_replace->tgtdev ?
823 			rcu_str_deref(dev_replace->tgtdev->name) :
824 			"<missing target disk>",
825 			(unsigned int)progress);
826 	}
827 	btrfs_dev_replace_continue_on_mount(fs_info);
828 	atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
829 
830 	return 0;
831 }
832 
833 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info)
834 {
835 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
836 	int ret;
837 
838 	ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
839 			      dev_replace->committed_cursor_left,
840 			      btrfs_device_get_total_bytes(dev_replace->srcdev),
841 			      &dev_replace->scrub_progress, 0, 1);
842 	ret = btrfs_dev_replace_finishing(fs_info, ret);
843 	WARN_ON(ret);
844 	return 0;
845 }
846 
847 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
848 {
849 	if (!dev_replace->is_valid)
850 		return 0;
851 
852 	switch (dev_replace->replace_state) {
853 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
854 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
855 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
856 		return 0;
857 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
858 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
859 		/*
860 		 * return true even if tgtdev is missing (this is
861 		 * something that can happen if the dev_replace
862 		 * procedure is suspended by an umount and then
863 		 * the tgtdev is missing (or "btrfs dev scan") was
864 		 * not called and the the filesystem is remounted
865 		 * in degraded state. This does not stop the
866 		 * dev_replace procedure. It needs to be canceled
867 		 * manually if the cancelation is wanted.
868 		 */
869 		break;
870 	}
871 	return 1;
872 }
873 
874 void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace)
875 {
876 	/* the beginning is just an optimization for the typical case */
877 	if (atomic_read(&dev_replace->nesting_level) == 0) {
878 acquire_lock:
879 		/* this is not a nested case where the same thread
880 		 * is trying to acqurire the same lock twice */
881 		mutex_lock(&dev_replace->lock);
882 		mutex_lock(&dev_replace->lock_management_lock);
883 		dev_replace->lock_owner = current->pid;
884 		atomic_inc(&dev_replace->nesting_level);
885 		mutex_unlock(&dev_replace->lock_management_lock);
886 		return;
887 	}
888 
889 	mutex_lock(&dev_replace->lock_management_lock);
890 	if (atomic_read(&dev_replace->nesting_level) > 0 &&
891 	    dev_replace->lock_owner == current->pid) {
892 		WARN_ON(!mutex_is_locked(&dev_replace->lock));
893 		atomic_inc(&dev_replace->nesting_level);
894 		mutex_unlock(&dev_replace->lock_management_lock);
895 		return;
896 	}
897 
898 	mutex_unlock(&dev_replace->lock_management_lock);
899 	goto acquire_lock;
900 }
901 
902 void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace)
903 {
904 	WARN_ON(!mutex_is_locked(&dev_replace->lock));
905 	mutex_lock(&dev_replace->lock_management_lock);
906 	WARN_ON(atomic_read(&dev_replace->nesting_level) < 1);
907 	WARN_ON(dev_replace->lock_owner != current->pid);
908 	atomic_dec(&dev_replace->nesting_level);
909 	if (atomic_read(&dev_replace->nesting_level) == 0) {
910 		dev_replace->lock_owner = 0;
911 		mutex_unlock(&dev_replace->lock_management_lock);
912 		mutex_unlock(&dev_replace->lock);
913 	} else {
914 		mutex_unlock(&dev_replace->lock_management_lock);
915 	}
916 }
917 
918 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
919 {
920 	percpu_counter_inc(&fs_info->bio_counter);
921 }
922 
923 void btrfs_bio_counter_dec(struct btrfs_fs_info *fs_info)
924 {
925 	percpu_counter_dec(&fs_info->bio_counter);
926 
927 	if (waitqueue_active(&fs_info->replace_wait))
928 		wake_up(&fs_info->replace_wait);
929 }
930 
931 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
932 {
933 	DEFINE_WAIT(wait);
934 again:
935 	percpu_counter_inc(&fs_info->bio_counter);
936 	if (test_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state)) {
937 		btrfs_bio_counter_dec(fs_info);
938 		wait_event(fs_info->replace_wait,
939 			   !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
940 				     &fs_info->fs_state));
941 		goto again;
942 	}
943 
944 }
945