xref: /openbmc/linux/fs/btrfs/dev-replace.c (revision b96fc2f3)
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 	btrfs_free_path(path);
187 	return ret;
188 }
189 
190 /*
191  * called from commit_transaction. Writes changed device replace state to
192  * disk.
193  */
194 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
195 			  struct btrfs_fs_info *fs_info)
196 {
197 	int ret;
198 	struct btrfs_root *dev_root = fs_info->dev_root;
199 	struct btrfs_path *path;
200 	struct btrfs_key key;
201 	struct extent_buffer *eb;
202 	struct btrfs_dev_replace_item *ptr;
203 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
204 
205 	btrfs_dev_replace_lock(dev_replace);
206 	if (!dev_replace->is_valid ||
207 	    !dev_replace->item_needs_writeback) {
208 		btrfs_dev_replace_unlock(dev_replace);
209 		return 0;
210 	}
211 	btrfs_dev_replace_unlock(dev_replace);
212 
213 	key.objectid = 0;
214 	key.type = BTRFS_DEV_REPLACE_KEY;
215 	key.offset = 0;
216 
217 	path = btrfs_alloc_path();
218 	if (!path) {
219 		ret = -ENOMEM;
220 		goto out;
221 	}
222 	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
223 	if (ret < 0) {
224 		btrfs_warn(fs_info, "error %d while searching for dev_replace item!",
225 			ret);
226 		goto out;
227 	}
228 
229 	if (ret == 0 &&
230 	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
231 		/*
232 		 * need to delete old one and insert a new one.
233 		 * Since no attempt is made to recover any old state, if the
234 		 * dev_replace state is 'running', the data on the target
235 		 * drive is lost.
236 		 * It would be possible to recover the state: just make sure
237 		 * that the beginning of the item is never changed and always
238 		 * contains all the essential information. Then read this
239 		 * minimal set of information and use it as a base for the
240 		 * new state.
241 		 */
242 		ret = btrfs_del_item(trans, dev_root, path);
243 		if (ret != 0) {
244 			btrfs_warn(fs_info, "delete too small dev_replace item failed %d!",
245 				ret);
246 			goto out;
247 		}
248 		ret = 1;
249 	}
250 
251 	if (ret == 1) {
252 		/* need to insert a new item */
253 		btrfs_release_path(path);
254 		ret = btrfs_insert_empty_item(trans, dev_root, path,
255 					      &key, sizeof(*ptr));
256 		if (ret < 0) {
257 			btrfs_warn(fs_info, "insert dev_replace item failed %d!",
258 				ret);
259 			goto out;
260 		}
261 	}
262 
263 	eb = path->nodes[0];
264 	ptr = btrfs_item_ptr(eb, path->slots[0],
265 			     struct btrfs_dev_replace_item);
266 
267 	btrfs_dev_replace_lock(dev_replace);
268 	if (dev_replace->srcdev)
269 		btrfs_set_dev_replace_src_devid(eb, ptr,
270 			dev_replace->srcdev->devid);
271 	else
272 		btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
273 	btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
274 		dev_replace->cont_reading_from_srcdev_mode);
275 	btrfs_set_dev_replace_replace_state(eb, ptr,
276 		dev_replace->replace_state);
277 	btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
278 	btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
279 	btrfs_set_dev_replace_num_write_errors(eb, ptr,
280 		atomic64_read(&dev_replace->num_write_errors));
281 	btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
282 		atomic64_read(&dev_replace->num_uncorrectable_read_errors));
283 	dev_replace->cursor_left_last_write_of_item =
284 		dev_replace->cursor_left;
285 	btrfs_set_dev_replace_cursor_left(eb, ptr,
286 		dev_replace->cursor_left_last_write_of_item);
287 	btrfs_set_dev_replace_cursor_right(eb, ptr,
288 		dev_replace->cursor_right);
289 	dev_replace->item_needs_writeback = 0;
290 	btrfs_dev_replace_unlock(dev_replace);
291 
292 	btrfs_mark_buffer_dirty(eb);
293 
294 out:
295 	btrfs_free_path(path);
296 
297 	return ret;
298 }
299 
300 void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
301 {
302 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
303 
304 	dev_replace->committed_cursor_left =
305 		dev_replace->cursor_left_last_write_of_item;
306 }
307 
308 int btrfs_dev_replace_start(struct btrfs_root *root,
309 			    struct btrfs_ioctl_dev_replace_args *args)
310 {
311 	struct btrfs_trans_handle *trans;
312 	struct btrfs_fs_info *fs_info = root->fs_info;
313 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
314 	int ret;
315 	struct btrfs_device *tgt_device = NULL;
316 	struct btrfs_device *src_device = NULL;
317 
318 	switch (args->start.cont_reading_from_srcdev_mode) {
319 	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
320 	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
321 		break;
322 	default:
323 		return -EINVAL;
324 	}
325 
326 	if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
327 	    args->start.tgtdev_name[0] == '\0')
328 		return -EINVAL;
329 
330 	/*
331 	 * Here we commit the transaction to make sure commit_total_bytes
332 	 * of all the devices are updated.
333 	 */
334 	trans = btrfs_attach_transaction(root);
335 	if (!IS_ERR(trans)) {
336 		ret = btrfs_commit_transaction(trans, root);
337 		if (ret)
338 			return ret;
339 	} else if (PTR_ERR(trans) != -ENOENT) {
340 		return PTR_ERR(trans);
341 	}
342 
343 	/* the disk copy procedure reuses the scrub code */
344 	mutex_lock(&fs_info->volume_mutex);
345 	ret = btrfs_dev_replace_find_srcdev(root, args->start.srcdevid,
346 					    args->start.srcdev_name,
347 					    &src_device);
348 	if (ret) {
349 		mutex_unlock(&fs_info->volume_mutex);
350 		return ret;
351 	}
352 
353 	ret = btrfs_init_dev_replace_tgtdev(root, args->start.tgtdev_name,
354 					    src_device, &tgt_device);
355 	mutex_unlock(&fs_info->volume_mutex);
356 	if (ret)
357 		return ret;
358 
359 	btrfs_dev_replace_lock(dev_replace);
360 	switch (dev_replace->replace_state) {
361 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
362 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
363 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
364 		break;
365 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
366 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
367 		args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
368 		goto leave;
369 	}
370 
371 	dev_replace->cont_reading_from_srcdev_mode =
372 		args->start.cont_reading_from_srcdev_mode;
373 	WARN_ON(!src_device);
374 	dev_replace->srcdev = src_device;
375 	WARN_ON(!tgt_device);
376 	dev_replace->tgtdev = tgt_device;
377 
378 	ret = btrfs_kobj_add_device(tgt_device->fs_devices, tgt_device);
379 	if (ret)
380 		btrfs_err(root->fs_info, "kobj add dev failed %d\n", ret);
381 
382 	printk_in_rcu(KERN_INFO
383 		      "BTRFS: dev_replace from %s (devid %llu) to %s started\n",
384 		      src_device->missing ? "<missing disk>" :
385 		        rcu_str_deref(src_device->name),
386 		      src_device->devid,
387 		      rcu_str_deref(tgt_device->name));
388 
389 	/*
390 	 * from now on, the writes to the srcdev are all duplicated to
391 	 * go to the tgtdev as well (refer to btrfs_map_block()).
392 	 */
393 	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
394 	dev_replace->time_started = get_seconds();
395 	dev_replace->cursor_left = 0;
396 	dev_replace->committed_cursor_left = 0;
397 	dev_replace->cursor_left_last_write_of_item = 0;
398 	dev_replace->cursor_right = 0;
399 	dev_replace->is_valid = 1;
400 	dev_replace->item_needs_writeback = 1;
401 	args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
402 	btrfs_dev_replace_unlock(dev_replace);
403 
404 	btrfs_wait_ordered_roots(root->fs_info, -1);
405 
406 	/* force writing the updated state information to disk */
407 	trans = btrfs_start_transaction(root, 0);
408 	if (IS_ERR(trans)) {
409 		ret = PTR_ERR(trans);
410 		btrfs_dev_replace_lock(dev_replace);
411 		goto leave;
412 	}
413 
414 	ret = btrfs_commit_transaction(trans, root);
415 	WARN_ON(ret);
416 
417 	/* the disk copy procedure reuses the scrub code */
418 	ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
419 			      btrfs_device_get_total_bytes(src_device),
420 			      &dev_replace->scrub_progress, 0, 1);
421 
422 	ret = btrfs_dev_replace_finishing(root->fs_info, ret);
423 	/* don't warn if EINPROGRESS, someone else might be running scrub */
424 	if (ret == -EINPROGRESS) {
425 		args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
426 		ret = 0;
427 	} else {
428 		WARN_ON(ret);
429 	}
430 
431 	return ret;
432 
433 leave:
434 	dev_replace->srcdev = NULL;
435 	dev_replace->tgtdev = NULL;
436 	btrfs_dev_replace_unlock(dev_replace);
437 	btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
438 	return ret;
439 }
440 
441 /*
442  * blocked until all flighting bios are finished.
443  */
444 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
445 {
446 	set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
447 	wait_event(fs_info->replace_wait, !percpu_counter_sum(
448 		   &fs_info->bio_counter));
449 }
450 
451 /*
452  * we have removed target device, it is safe to allow new bios request.
453  */
454 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
455 {
456 	clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
457 	if (waitqueue_active(&fs_info->replace_wait))
458 		wake_up(&fs_info->replace_wait);
459 }
460 
461 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
462 				       int scrub_ret)
463 {
464 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
465 	struct btrfs_device *tgt_device;
466 	struct btrfs_device *src_device;
467 	struct btrfs_root *root = fs_info->tree_root;
468 	u8 uuid_tmp[BTRFS_UUID_SIZE];
469 	struct btrfs_trans_handle *trans;
470 	int ret = 0;
471 
472 	/* don't allow cancel or unmount to disturb the finishing procedure */
473 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
474 
475 	btrfs_dev_replace_lock(dev_replace);
476 	/* was the operation canceled, or is it finished? */
477 	if (dev_replace->replace_state !=
478 	    BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
479 		btrfs_dev_replace_unlock(dev_replace);
480 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
481 		return 0;
482 	}
483 
484 	tgt_device = dev_replace->tgtdev;
485 	src_device = dev_replace->srcdev;
486 	btrfs_dev_replace_unlock(dev_replace);
487 
488 	/*
489 	 * flush all outstanding I/O and inode extent mappings before the
490 	 * copy operation is declared as being finished
491 	 */
492 	ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
493 	if (ret) {
494 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
495 		return ret;
496 	}
497 	btrfs_wait_ordered_roots(root->fs_info, -1);
498 
499 	trans = btrfs_start_transaction(root, 0);
500 	if (IS_ERR(trans)) {
501 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
502 		return PTR_ERR(trans);
503 	}
504 	ret = btrfs_commit_transaction(trans, root);
505 	WARN_ON(ret);
506 
507 	mutex_lock(&uuid_mutex);
508 	/* keep away write_all_supers() during the finishing procedure */
509 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
510 	mutex_lock(&root->fs_info->chunk_mutex);
511 	btrfs_dev_replace_lock(dev_replace);
512 	dev_replace->replace_state =
513 		scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
514 			  : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
515 	dev_replace->tgtdev = NULL;
516 	dev_replace->srcdev = NULL;
517 	dev_replace->time_stopped = get_seconds();
518 	dev_replace->item_needs_writeback = 1;
519 
520 	/* replace old device with new one in mapping tree */
521 	if (!scrub_ret) {
522 		btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
523 								src_device,
524 								tgt_device);
525 	} else {
526 		printk_in_rcu(KERN_ERR
527 			      "BTRFS: btrfs_scrub_dev(%s, %llu, %s) failed %d\n",
528 			      src_device->missing ? "<missing disk>" :
529 			        rcu_str_deref(src_device->name),
530 			      src_device->devid,
531 			      rcu_str_deref(tgt_device->name), scrub_ret);
532 		btrfs_dev_replace_unlock(dev_replace);
533 		mutex_unlock(&root->fs_info->chunk_mutex);
534 		mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
535 		mutex_unlock(&uuid_mutex);
536 		if (tgt_device)
537 			btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
538 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
539 
540 		return scrub_ret;
541 	}
542 
543 	printk_in_rcu(KERN_INFO
544 		      "BTRFS: dev_replace from %s (devid %llu) to %s finished\n",
545 		      src_device->missing ? "<missing disk>" :
546 		        rcu_str_deref(src_device->name),
547 		      src_device->devid,
548 		      rcu_str_deref(tgt_device->name));
549 	tgt_device->is_tgtdev_for_dev_replace = 0;
550 	tgt_device->devid = src_device->devid;
551 	src_device->devid = BTRFS_DEV_REPLACE_DEVID;
552 	memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
553 	memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
554 	memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
555 	btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
556 	btrfs_device_set_disk_total_bytes(tgt_device,
557 					  src_device->disk_total_bytes);
558 	btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
559 	ASSERT(list_empty(&src_device->resized_list));
560 	tgt_device->commit_total_bytes = src_device->commit_total_bytes;
561 	tgt_device->commit_bytes_used = src_device->bytes_used;
562 	if (fs_info->sb->s_bdev == src_device->bdev)
563 		fs_info->sb->s_bdev = tgt_device->bdev;
564 	if (fs_info->fs_devices->latest_bdev == src_device->bdev)
565 		fs_info->fs_devices->latest_bdev = tgt_device->bdev;
566 	list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
567 	fs_info->fs_devices->rw_devices++;
568 
569 	btrfs_dev_replace_unlock(dev_replace);
570 
571 	btrfs_rm_dev_replace_blocked(fs_info);
572 
573 	btrfs_rm_dev_replace_remove_srcdev(fs_info, src_device);
574 
575 	btrfs_rm_dev_replace_unblocked(fs_info);
576 
577 	/*
578 	 * this is again a consistent state where no dev_replace procedure
579 	 * is running, the target device is part of the filesystem, the
580 	 * source device is not part of the filesystem anymore and its 1st
581 	 * superblock is scratched out so that it is no longer marked to
582 	 * belong to this filesystem.
583 	 */
584 	mutex_unlock(&root->fs_info->chunk_mutex);
585 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
586 	mutex_unlock(&uuid_mutex);
587 
588 	/* replace the sysfs entry */
589 	btrfs_kobj_rm_device(fs_info->fs_devices, src_device);
590 	btrfs_rm_dev_replace_free_srcdev(fs_info, src_device);
591 
592 	/* write back the superblocks */
593 	trans = btrfs_start_transaction(root, 0);
594 	if (!IS_ERR(trans))
595 		btrfs_commit_transaction(trans, root);
596 
597 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
598 
599 	return 0;
600 }
601 
602 static void btrfs_dev_replace_update_device_in_mapping_tree(
603 						struct btrfs_fs_info *fs_info,
604 						struct btrfs_device *srcdev,
605 						struct btrfs_device *tgtdev)
606 {
607 	struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
608 	struct extent_map *em;
609 	struct map_lookup *map;
610 	u64 start = 0;
611 	int i;
612 
613 	write_lock(&em_tree->lock);
614 	do {
615 		em = lookup_extent_mapping(em_tree, start, (u64)-1);
616 		if (!em)
617 			break;
618 		map = (struct map_lookup *)em->bdev;
619 		for (i = 0; i < map->num_stripes; i++)
620 			if (srcdev == map->stripes[i].dev)
621 				map->stripes[i].dev = tgtdev;
622 		start = em->start + em->len;
623 		free_extent_map(em);
624 	} while (start);
625 	write_unlock(&em_tree->lock);
626 }
627 
628 static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid,
629 					 char *srcdev_name,
630 					 struct btrfs_device **device)
631 {
632 	int ret;
633 
634 	if (srcdevid) {
635 		ret = 0;
636 		*device = btrfs_find_device(root->fs_info, srcdevid, NULL,
637 					    NULL);
638 		if (!*device)
639 			ret = -ENOENT;
640 	} else {
641 		ret = btrfs_find_device_missing_or_by_path(root, srcdev_name,
642 							   device);
643 	}
644 	return ret;
645 }
646 
647 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
648 			      struct btrfs_ioctl_dev_replace_args *args)
649 {
650 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
651 	struct btrfs_device *srcdev;
652 
653 	btrfs_dev_replace_lock(dev_replace);
654 	/* even if !dev_replace_is_valid, the values are good enough for
655 	 * the replace_status ioctl */
656 	args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
657 	args->status.replace_state = dev_replace->replace_state;
658 	args->status.time_started = dev_replace->time_started;
659 	args->status.time_stopped = dev_replace->time_stopped;
660 	args->status.num_write_errors =
661 		atomic64_read(&dev_replace->num_write_errors);
662 	args->status.num_uncorrectable_read_errors =
663 		atomic64_read(&dev_replace->num_uncorrectable_read_errors);
664 	switch (dev_replace->replace_state) {
665 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
666 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
667 		args->status.progress_1000 = 0;
668 		break;
669 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
670 		args->status.progress_1000 = 1000;
671 		break;
672 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
673 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
674 		srcdev = dev_replace->srcdev;
675 		args->status.progress_1000 = div_u64(dev_replace->cursor_left,
676 			div_u64(btrfs_device_get_total_bytes(srcdev), 1000));
677 		break;
678 	}
679 	btrfs_dev_replace_unlock(dev_replace);
680 }
681 
682 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info,
683 			     struct btrfs_ioctl_dev_replace_args *args)
684 {
685 	args->result = __btrfs_dev_replace_cancel(fs_info);
686 	return 0;
687 }
688 
689 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
690 {
691 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
692 	struct btrfs_device *tgt_device = NULL;
693 	struct btrfs_trans_handle *trans;
694 	struct btrfs_root *root = fs_info->tree_root;
695 	u64 result;
696 	int ret;
697 
698 	if (fs_info->sb->s_flags & MS_RDONLY)
699 		return -EROFS;
700 
701 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
702 	btrfs_dev_replace_lock(dev_replace);
703 	switch (dev_replace->replace_state) {
704 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
705 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
706 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
707 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
708 		btrfs_dev_replace_unlock(dev_replace);
709 		goto leave;
710 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
711 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
712 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
713 		tgt_device = dev_replace->tgtdev;
714 		dev_replace->tgtdev = NULL;
715 		dev_replace->srcdev = NULL;
716 		break;
717 	}
718 	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
719 	dev_replace->time_stopped = get_seconds();
720 	dev_replace->item_needs_writeback = 1;
721 	btrfs_dev_replace_unlock(dev_replace);
722 	btrfs_scrub_cancel(fs_info);
723 
724 	trans = btrfs_start_transaction(root, 0);
725 	if (IS_ERR(trans)) {
726 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
727 		return PTR_ERR(trans);
728 	}
729 	ret = btrfs_commit_transaction(trans, root);
730 	WARN_ON(ret);
731 	if (tgt_device)
732 		btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
733 
734 leave:
735 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
736 	return result;
737 }
738 
739 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
740 {
741 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
742 
743 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
744 	btrfs_dev_replace_lock(dev_replace);
745 	switch (dev_replace->replace_state) {
746 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
747 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
748 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
749 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
750 		break;
751 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
752 		dev_replace->replace_state =
753 			BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
754 		dev_replace->time_stopped = get_seconds();
755 		dev_replace->item_needs_writeback = 1;
756 		btrfs_info(fs_info, "suspending dev_replace for unmount");
757 		break;
758 	}
759 
760 	btrfs_dev_replace_unlock(dev_replace);
761 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
762 }
763 
764 /* resume dev_replace procedure that was interrupted by unmount */
765 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
766 {
767 	struct task_struct *task;
768 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
769 
770 	btrfs_dev_replace_lock(dev_replace);
771 	switch (dev_replace->replace_state) {
772 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
773 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
774 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
775 		btrfs_dev_replace_unlock(dev_replace);
776 		return 0;
777 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
778 		break;
779 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
780 		dev_replace->replace_state =
781 			BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
782 		break;
783 	}
784 	if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
785 		btrfs_info(fs_info, "cannot continue dev_replace, tgtdev is missing");
786 		btrfs_info(fs_info,
787 			"you may cancel the operation after 'mount -o degraded'");
788 		btrfs_dev_replace_unlock(dev_replace);
789 		return 0;
790 	}
791 	btrfs_dev_replace_unlock(dev_replace);
792 
793 	WARN_ON(atomic_xchg(
794 		&fs_info->mutually_exclusive_operation_running, 1));
795 	task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
796 	return PTR_ERR_OR_ZERO(task);
797 }
798 
799 static int btrfs_dev_replace_kthread(void *data)
800 {
801 	struct btrfs_fs_info *fs_info = data;
802 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
803 	struct btrfs_ioctl_dev_replace_args *status_args;
804 	u64 progress;
805 
806 	status_args = kzalloc(sizeof(*status_args), GFP_NOFS);
807 	if (status_args) {
808 		btrfs_dev_replace_status(fs_info, status_args);
809 		progress = status_args->status.progress_1000;
810 		kfree(status_args);
811 		progress = div_u64(progress, 10);
812 		printk_in_rcu(KERN_INFO
813 			"BTRFS: continuing dev_replace from %s (devid %llu) to %s @%u%%\n",
814 			dev_replace->srcdev->missing ? "<missing disk>" :
815 			rcu_str_deref(dev_replace->srcdev->name),
816 			dev_replace->srcdev->devid,
817 			dev_replace->tgtdev ?
818 			rcu_str_deref(dev_replace->tgtdev->name) :
819 			"<missing target disk>",
820 			(unsigned int)progress);
821 	}
822 	btrfs_dev_replace_continue_on_mount(fs_info);
823 	atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
824 
825 	return 0;
826 }
827 
828 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info)
829 {
830 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
831 	int ret;
832 
833 	ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
834 			      dev_replace->committed_cursor_left,
835 			      btrfs_device_get_total_bytes(dev_replace->srcdev),
836 			      &dev_replace->scrub_progress, 0, 1);
837 	ret = btrfs_dev_replace_finishing(fs_info, ret);
838 	WARN_ON(ret);
839 	return 0;
840 }
841 
842 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
843 {
844 	if (!dev_replace->is_valid)
845 		return 0;
846 
847 	switch (dev_replace->replace_state) {
848 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
849 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
850 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
851 		return 0;
852 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
853 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
854 		/*
855 		 * return true even if tgtdev is missing (this is
856 		 * something that can happen if the dev_replace
857 		 * procedure is suspended by an umount and then
858 		 * the tgtdev is missing (or "btrfs dev scan") was
859 		 * not called and the the filesystem is remounted
860 		 * in degraded state. This does not stop the
861 		 * dev_replace procedure. It needs to be canceled
862 		 * manually if the cancelation is wanted.
863 		 */
864 		break;
865 	}
866 	return 1;
867 }
868 
869 void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace)
870 {
871 	/* the beginning is just an optimization for the typical case */
872 	if (atomic_read(&dev_replace->nesting_level) == 0) {
873 acquire_lock:
874 		/* this is not a nested case where the same thread
875 		 * is trying to acqurire the same lock twice */
876 		mutex_lock(&dev_replace->lock);
877 		mutex_lock(&dev_replace->lock_management_lock);
878 		dev_replace->lock_owner = current->pid;
879 		atomic_inc(&dev_replace->nesting_level);
880 		mutex_unlock(&dev_replace->lock_management_lock);
881 		return;
882 	}
883 
884 	mutex_lock(&dev_replace->lock_management_lock);
885 	if (atomic_read(&dev_replace->nesting_level) > 0 &&
886 	    dev_replace->lock_owner == current->pid) {
887 		WARN_ON(!mutex_is_locked(&dev_replace->lock));
888 		atomic_inc(&dev_replace->nesting_level);
889 		mutex_unlock(&dev_replace->lock_management_lock);
890 		return;
891 	}
892 
893 	mutex_unlock(&dev_replace->lock_management_lock);
894 	goto acquire_lock;
895 }
896 
897 void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace)
898 {
899 	WARN_ON(!mutex_is_locked(&dev_replace->lock));
900 	mutex_lock(&dev_replace->lock_management_lock);
901 	WARN_ON(atomic_read(&dev_replace->nesting_level) < 1);
902 	WARN_ON(dev_replace->lock_owner != current->pid);
903 	atomic_dec(&dev_replace->nesting_level);
904 	if (atomic_read(&dev_replace->nesting_level) == 0) {
905 		dev_replace->lock_owner = 0;
906 		mutex_unlock(&dev_replace->lock_management_lock);
907 		mutex_unlock(&dev_replace->lock);
908 	} else {
909 		mutex_unlock(&dev_replace->lock_management_lock);
910 	}
911 }
912 
913 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
914 {
915 	percpu_counter_inc(&fs_info->bio_counter);
916 }
917 
918 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
919 {
920 	percpu_counter_sub(&fs_info->bio_counter, amount);
921 
922 	if (waitqueue_active(&fs_info->replace_wait))
923 		wake_up(&fs_info->replace_wait);
924 }
925 
926 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
927 {
928 	while (1) {
929 		percpu_counter_inc(&fs_info->bio_counter);
930 		if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
931 				     &fs_info->fs_state)))
932 			break;
933 
934 		btrfs_bio_counter_dec(fs_info);
935 		wait_event(fs_info->replace_wait,
936 			   !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
937 				     &fs_info->fs_state));
938 	}
939 }
940