xref: /linux/fs/btrfs/dev-replace.c (revision 83bce9c2baa51e439480a713119a73d3c8b61083)
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 u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info);
48 static int btrfs_dev_replace_kthread(void *data);
49 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info);
50 
51 
52 int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
53 {
54 	struct btrfs_key key;
55 	struct btrfs_root *dev_root = fs_info->dev_root;
56 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
57 	struct extent_buffer *eb;
58 	int slot;
59 	int ret = 0;
60 	struct btrfs_path *path = NULL;
61 	int item_size;
62 	struct btrfs_dev_replace_item *ptr;
63 	u64 src_devid;
64 
65 	path = btrfs_alloc_path();
66 	if (!path) {
67 		ret = -ENOMEM;
68 		goto out;
69 	}
70 
71 	key.objectid = 0;
72 	key.type = BTRFS_DEV_REPLACE_KEY;
73 	key.offset = 0;
74 	ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
75 	if (ret) {
76 no_valid_dev_replace_entry_found:
77 		ret = 0;
78 		dev_replace->replace_state =
79 			BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED;
80 		dev_replace->cont_reading_from_srcdev_mode =
81 		    BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
82 		dev_replace->replace_state = 0;
83 		dev_replace->time_started = 0;
84 		dev_replace->time_stopped = 0;
85 		atomic64_set(&dev_replace->num_write_errors, 0);
86 		atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
87 		dev_replace->cursor_left = 0;
88 		dev_replace->committed_cursor_left = 0;
89 		dev_replace->cursor_left_last_write_of_item = 0;
90 		dev_replace->cursor_right = 0;
91 		dev_replace->srcdev = NULL;
92 		dev_replace->tgtdev = NULL;
93 		dev_replace->is_valid = 0;
94 		dev_replace->item_needs_writeback = 0;
95 		goto out;
96 	}
97 	slot = path->slots[0];
98 	eb = path->nodes[0];
99 	item_size = btrfs_item_size_nr(eb, slot);
100 	ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
101 
102 	if (item_size != sizeof(struct btrfs_dev_replace_item)) {
103 		btrfs_warn(fs_info,
104 			"dev_replace entry found has unexpected size, ignore entry");
105 		goto no_valid_dev_replace_entry_found;
106 	}
107 
108 	src_devid = btrfs_dev_replace_src_devid(eb, ptr);
109 	dev_replace->cont_reading_from_srcdev_mode =
110 		btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
111 	dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
112 	dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
113 	dev_replace->time_stopped =
114 		btrfs_dev_replace_time_stopped(eb, ptr);
115 	atomic64_set(&dev_replace->num_write_errors,
116 		     btrfs_dev_replace_num_write_errors(eb, ptr));
117 	atomic64_set(&dev_replace->num_uncorrectable_read_errors,
118 		     btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
119 	dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
120 	dev_replace->committed_cursor_left = dev_replace->cursor_left;
121 	dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
122 	dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
123 	dev_replace->is_valid = 1;
124 
125 	dev_replace->item_needs_writeback = 0;
126 	switch (dev_replace->replace_state) {
127 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
128 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
129 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
130 		dev_replace->srcdev = NULL;
131 		dev_replace->tgtdev = NULL;
132 		break;
133 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
134 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
135 		dev_replace->srcdev = btrfs_find_device(fs_info, src_devid,
136 							NULL, NULL);
137 		dev_replace->tgtdev = btrfs_find_device(fs_info,
138 							BTRFS_DEV_REPLACE_DEVID,
139 							NULL, NULL);
140 		/*
141 		 * allow 'btrfs dev replace_cancel' if src/tgt device is
142 		 * missing
143 		 */
144 		if (!dev_replace->srcdev &&
145 		    !btrfs_test_opt(fs_info, DEGRADED)) {
146 			ret = -EIO;
147 			btrfs_warn(fs_info,
148 			   "cannot mount because device replace operation is ongoing and");
149 			btrfs_warn(fs_info,
150 			   "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
151 			   src_devid);
152 		}
153 		if (!dev_replace->tgtdev &&
154 		    !btrfs_test_opt(fs_info, DEGRADED)) {
155 			ret = -EIO;
156 			btrfs_warn(fs_info,
157 			   "cannot mount because device replace operation is ongoing and");
158 			btrfs_warn(fs_info,
159 			   "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
160 				BTRFS_DEV_REPLACE_DEVID);
161 		}
162 		if (dev_replace->tgtdev) {
163 			if (dev_replace->srcdev) {
164 				dev_replace->tgtdev->total_bytes =
165 					dev_replace->srcdev->total_bytes;
166 				dev_replace->tgtdev->disk_total_bytes =
167 					dev_replace->srcdev->disk_total_bytes;
168 				dev_replace->tgtdev->commit_total_bytes =
169 					dev_replace->srcdev->commit_total_bytes;
170 				dev_replace->tgtdev->bytes_used =
171 					dev_replace->srcdev->bytes_used;
172 				dev_replace->tgtdev->commit_bytes_used =
173 					dev_replace->srcdev->commit_bytes_used;
174 			}
175 			dev_replace->tgtdev->is_tgtdev_for_dev_replace = 1;
176 			btrfs_init_dev_replace_tgtdev_for_resume(fs_info,
177 				dev_replace->tgtdev);
178 		}
179 		break;
180 	}
181 
182 out:
183 	btrfs_free_path(path);
184 	return ret;
185 }
186 
187 /*
188  * called from commit_transaction. Writes changed device replace state to
189  * disk.
190  */
191 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
192 			  struct btrfs_fs_info *fs_info)
193 {
194 	int ret;
195 	struct btrfs_root *dev_root = fs_info->dev_root;
196 	struct btrfs_path *path;
197 	struct btrfs_key key;
198 	struct extent_buffer *eb;
199 	struct btrfs_dev_replace_item *ptr;
200 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
201 
202 	btrfs_dev_replace_lock(dev_replace, 0);
203 	if (!dev_replace->is_valid ||
204 	    !dev_replace->item_needs_writeback) {
205 		btrfs_dev_replace_unlock(dev_replace, 0);
206 		return 0;
207 	}
208 	btrfs_dev_replace_unlock(dev_replace, 0);
209 
210 	key.objectid = 0;
211 	key.type = BTRFS_DEV_REPLACE_KEY;
212 	key.offset = 0;
213 
214 	path = btrfs_alloc_path();
215 	if (!path) {
216 		ret = -ENOMEM;
217 		goto out;
218 	}
219 	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
220 	if (ret < 0) {
221 		btrfs_warn(fs_info,
222 			   "error %d while searching for dev_replace item!",
223 			   ret);
224 		goto out;
225 	}
226 
227 	if (ret == 0 &&
228 	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
229 		/*
230 		 * need to delete old one and insert a new one.
231 		 * Since no attempt is made to recover any old state, if the
232 		 * dev_replace state is 'running', the data on the target
233 		 * drive is lost.
234 		 * It would be possible to recover the state: just make sure
235 		 * that the beginning of the item is never changed and always
236 		 * contains all the essential information. Then read this
237 		 * minimal set of information and use it as a base for the
238 		 * new state.
239 		 */
240 		ret = btrfs_del_item(trans, dev_root, path);
241 		if (ret != 0) {
242 			btrfs_warn(fs_info,
243 				   "delete too small dev_replace item failed %d!",
244 				   ret);
245 			goto out;
246 		}
247 		ret = 1;
248 	}
249 
250 	if (ret == 1) {
251 		/* need to insert a new item */
252 		btrfs_release_path(path);
253 		ret = btrfs_insert_empty_item(trans, dev_root, path,
254 					      &key, sizeof(*ptr));
255 		if (ret < 0) {
256 			btrfs_warn(fs_info,
257 				   "insert dev_replace item failed %d!", ret);
258 			goto out;
259 		}
260 	}
261 
262 	eb = path->nodes[0];
263 	ptr = btrfs_item_ptr(eb, path->slots[0],
264 			     struct btrfs_dev_replace_item);
265 
266 	btrfs_dev_replace_lock(dev_replace, 1);
267 	if (dev_replace->srcdev)
268 		btrfs_set_dev_replace_src_devid(eb, ptr,
269 			dev_replace->srcdev->devid);
270 	else
271 		btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
272 	btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
273 		dev_replace->cont_reading_from_srcdev_mode);
274 	btrfs_set_dev_replace_replace_state(eb, ptr,
275 		dev_replace->replace_state);
276 	btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
277 	btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
278 	btrfs_set_dev_replace_num_write_errors(eb, ptr,
279 		atomic64_read(&dev_replace->num_write_errors));
280 	btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
281 		atomic64_read(&dev_replace->num_uncorrectable_read_errors));
282 	dev_replace->cursor_left_last_write_of_item =
283 		dev_replace->cursor_left;
284 	btrfs_set_dev_replace_cursor_left(eb, ptr,
285 		dev_replace->cursor_left_last_write_of_item);
286 	btrfs_set_dev_replace_cursor_right(eb, ptr,
287 		dev_replace->cursor_right);
288 	dev_replace->item_needs_writeback = 0;
289 	btrfs_dev_replace_unlock(dev_replace, 1);
290 
291 	btrfs_mark_buffer_dirty(eb);
292 
293 out:
294 	btrfs_free_path(path);
295 
296 	return ret;
297 }
298 
299 void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
300 {
301 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
302 
303 	dev_replace->committed_cursor_left =
304 		dev_replace->cursor_left_last_write_of_item;
305 }
306 
307 int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
308 		const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
309 		int read_src)
310 {
311 	struct btrfs_root *root = fs_info->dev_root;
312 	struct btrfs_trans_handle *trans;
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 	/* the disk copy procedure reuses the scrub code */
319 	mutex_lock(&fs_info->volume_mutex);
320 	ret = btrfs_find_device_by_devspec(fs_info, srcdevid,
321 					    srcdev_name, &src_device);
322 	if (ret) {
323 		mutex_unlock(&fs_info->volume_mutex);
324 		return ret;
325 	}
326 
327 	ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
328 					    src_device, &tgt_device);
329 	mutex_unlock(&fs_info->volume_mutex);
330 	if (ret)
331 		return ret;
332 
333 	/*
334 	 * Here we commit the transaction to make sure commit_total_bytes
335 	 * of all the devices are updated.
336 	 */
337 	trans = btrfs_attach_transaction(root);
338 	if (!IS_ERR(trans)) {
339 		ret = btrfs_commit_transaction(trans);
340 		if (ret)
341 			return ret;
342 	} else if (PTR_ERR(trans) != -ENOENT) {
343 		return PTR_ERR(trans);
344 	}
345 
346 	btrfs_dev_replace_lock(dev_replace, 1);
347 	switch (dev_replace->replace_state) {
348 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
349 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
350 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
351 		break;
352 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
353 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
354 		ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
355 		goto leave;
356 	}
357 
358 	dev_replace->cont_reading_from_srcdev_mode = read_src;
359 	WARN_ON(!src_device);
360 	dev_replace->srcdev = src_device;
361 	WARN_ON(!tgt_device);
362 	dev_replace->tgtdev = tgt_device;
363 
364 	btrfs_info_in_rcu(fs_info,
365 		      "dev_replace from %s (devid %llu) to %s started",
366 		      src_device->missing ? "<missing disk>" :
367 		        rcu_str_deref(src_device->name),
368 		      src_device->devid,
369 		      rcu_str_deref(tgt_device->name));
370 
371 	/*
372 	 * from now on, the writes to the srcdev are all duplicated to
373 	 * go to the tgtdev as well (refer to btrfs_map_block()).
374 	 */
375 	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
376 	dev_replace->time_started = get_seconds();
377 	dev_replace->cursor_left = 0;
378 	dev_replace->committed_cursor_left = 0;
379 	dev_replace->cursor_left_last_write_of_item = 0;
380 	dev_replace->cursor_right = 0;
381 	dev_replace->is_valid = 1;
382 	dev_replace->item_needs_writeback = 1;
383 	atomic64_set(&dev_replace->num_write_errors, 0);
384 	atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
385 	btrfs_dev_replace_unlock(dev_replace, 1);
386 
387 	ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device);
388 	if (ret)
389 		btrfs_err(fs_info, "kobj add dev failed %d", ret);
390 
391 	btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
392 
393 	/* force writing the updated state information to disk */
394 	trans = btrfs_start_transaction(root, 0);
395 	if (IS_ERR(trans)) {
396 		ret = PTR_ERR(trans);
397 		btrfs_dev_replace_lock(dev_replace, 1);
398 		goto leave;
399 	}
400 
401 	ret = btrfs_commit_transaction(trans);
402 	WARN_ON(ret);
403 
404 	/* the disk copy procedure reuses the scrub code */
405 	ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
406 			      btrfs_device_get_total_bytes(src_device),
407 			      &dev_replace->scrub_progress, 0, 1);
408 
409 	ret = btrfs_dev_replace_finishing(fs_info, ret);
410 	if (ret == -EINPROGRESS) {
411 		ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
412 	} else {
413 		WARN_ON(ret);
414 	}
415 
416 	return ret;
417 
418 leave:
419 	dev_replace->srcdev = NULL;
420 	dev_replace->tgtdev = NULL;
421 	btrfs_dev_replace_unlock(dev_replace, 1);
422 	btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
423 	return ret;
424 }
425 
426 int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
427 			    struct btrfs_ioctl_dev_replace_args *args)
428 {
429 	int ret;
430 
431 	switch (args->start.cont_reading_from_srcdev_mode) {
432 	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
433 	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
434 		break;
435 	default:
436 		return -EINVAL;
437 	}
438 
439 	if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
440 	    args->start.tgtdev_name[0] == '\0')
441 		return -EINVAL;
442 
443 	ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name,
444 					args->start.srcdevid,
445 					args->start.srcdev_name,
446 					args->start.cont_reading_from_srcdev_mode);
447 	args->result = ret;
448 	/* don't warn if EINPROGRESS, someone else might be running scrub */
449 	if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS)
450 		ret = 0;
451 
452 	return ret;
453 }
454 
455 /*
456  * blocked until all in-flight bios operations are finished.
457  */
458 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
459 {
460 	set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
461 	wait_event(fs_info->replace_wait, !percpu_counter_sum(
462 		   &fs_info->bio_counter));
463 }
464 
465 /*
466  * we have removed target device, it is safe to allow new bios request.
467  */
468 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
469 {
470 	clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
471 	wake_up(&fs_info->replace_wait);
472 }
473 
474 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
475 				       int scrub_ret)
476 {
477 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
478 	struct btrfs_device *tgt_device;
479 	struct btrfs_device *src_device;
480 	struct btrfs_root *root = fs_info->tree_root;
481 	u8 uuid_tmp[BTRFS_UUID_SIZE];
482 	struct btrfs_trans_handle *trans;
483 	int ret = 0;
484 
485 	/* don't allow cancel or unmount to disturb the finishing procedure */
486 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
487 
488 	btrfs_dev_replace_lock(dev_replace, 0);
489 	/* was the operation canceled, or is it finished? */
490 	if (dev_replace->replace_state !=
491 	    BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
492 		btrfs_dev_replace_unlock(dev_replace, 0);
493 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
494 		return 0;
495 	}
496 
497 	tgt_device = dev_replace->tgtdev;
498 	src_device = dev_replace->srcdev;
499 	btrfs_dev_replace_unlock(dev_replace, 0);
500 
501 	/*
502 	 * flush all outstanding I/O and inode extent mappings before the
503 	 * copy operation is declared as being finished
504 	 */
505 	ret = btrfs_start_delalloc_roots(fs_info, 0, -1);
506 	if (ret) {
507 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
508 		return ret;
509 	}
510 	btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
511 
512 	trans = btrfs_start_transaction(root, 0);
513 	if (IS_ERR(trans)) {
514 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
515 		return PTR_ERR(trans);
516 	}
517 	ret = btrfs_commit_transaction(trans);
518 	WARN_ON(ret);
519 
520 	mutex_lock(&uuid_mutex);
521 	/* keep away write_all_supers() during the finishing procedure */
522 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
523 	mutex_lock(&fs_info->chunk_mutex);
524 	btrfs_dev_replace_lock(dev_replace, 1);
525 	dev_replace->replace_state =
526 		scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
527 			  : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
528 	dev_replace->tgtdev = NULL;
529 	dev_replace->srcdev = NULL;
530 	dev_replace->time_stopped = get_seconds();
531 	dev_replace->item_needs_writeback = 1;
532 
533 	/* replace old device with new one in mapping tree */
534 	if (!scrub_ret) {
535 		btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
536 								src_device,
537 								tgt_device);
538 	} else {
539 		btrfs_err_in_rcu(fs_info,
540 				 "btrfs_scrub_dev(%s, %llu, %s) failed %d",
541 				 src_device->missing ? "<missing disk>" :
542 				 rcu_str_deref(src_device->name),
543 				 src_device->devid,
544 				 rcu_str_deref(tgt_device->name), scrub_ret);
545 		btrfs_dev_replace_unlock(dev_replace, 1);
546 		mutex_unlock(&fs_info->chunk_mutex);
547 		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
548 		mutex_unlock(&uuid_mutex);
549 		if (tgt_device)
550 			btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
551 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
552 
553 		return scrub_ret;
554 	}
555 
556 	btrfs_info_in_rcu(fs_info,
557 			  "dev_replace from %s (devid %llu) to %s finished",
558 			  src_device->missing ? "<missing disk>" :
559 			  rcu_str_deref(src_device->name),
560 			  src_device->devid,
561 			  rcu_str_deref(tgt_device->name));
562 	tgt_device->is_tgtdev_for_dev_replace = 0;
563 	tgt_device->devid = src_device->devid;
564 	src_device->devid = BTRFS_DEV_REPLACE_DEVID;
565 	memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
566 	memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
567 	memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
568 	btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
569 	btrfs_device_set_disk_total_bytes(tgt_device,
570 					  src_device->disk_total_bytes);
571 	btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
572 	ASSERT(list_empty(&src_device->resized_list));
573 	tgt_device->commit_total_bytes = src_device->commit_total_bytes;
574 	tgt_device->commit_bytes_used = src_device->bytes_used;
575 
576 	btrfs_assign_next_active_device(fs_info, src_device, tgt_device);
577 
578 	list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
579 	fs_info->fs_devices->rw_devices++;
580 
581 	btrfs_dev_replace_unlock(dev_replace, 1);
582 
583 	btrfs_rm_dev_replace_blocked(fs_info);
584 
585 	btrfs_rm_dev_replace_remove_srcdev(fs_info, src_device);
586 
587 	btrfs_rm_dev_replace_unblocked(fs_info);
588 
589 	/*
590 	 * this is again a consistent state where no dev_replace procedure
591 	 * is running, the target device is part of the filesystem, the
592 	 * source device is not part of the filesystem anymore and its 1st
593 	 * superblock is scratched out so that it is no longer marked to
594 	 * belong to this filesystem.
595 	 */
596 	mutex_unlock(&fs_info->chunk_mutex);
597 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
598 	mutex_unlock(&uuid_mutex);
599 
600 	/* replace the sysfs entry */
601 	btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device);
602 	btrfs_rm_dev_replace_free_srcdev(fs_info, src_device);
603 
604 	/* write back the superblocks */
605 	trans = btrfs_start_transaction(root, 0);
606 	if (!IS_ERR(trans))
607 		btrfs_commit_transaction(trans);
608 
609 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
610 
611 	return 0;
612 }
613 
614 static void btrfs_dev_replace_update_device_in_mapping_tree(
615 						struct btrfs_fs_info *fs_info,
616 						struct btrfs_device *srcdev,
617 						struct btrfs_device *tgtdev)
618 {
619 	struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
620 	struct extent_map *em;
621 	struct map_lookup *map;
622 	u64 start = 0;
623 	int i;
624 
625 	write_lock(&em_tree->lock);
626 	do {
627 		em = lookup_extent_mapping(em_tree, start, (u64)-1);
628 		if (!em)
629 			break;
630 		map = em->map_lookup;
631 		for (i = 0; i < map->num_stripes; i++)
632 			if (srcdev == map->stripes[i].dev)
633 				map->stripes[i].dev = tgtdev;
634 		start = em->start + em->len;
635 		free_extent_map(em);
636 	} while (start);
637 	write_unlock(&em_tree->lock);
638 }
639 
640 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
641 			      struct btrfs_ioctl_dev_replace_args *args)
642 {
643 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
644 	struct btrfs_device *srcdev;
645 
646 	btrfs_dev_replace_lock(dev_replace, 0);
647 	/* even if !dev_replace_is_valid, the values are good enough for
648 	 * the replace_status ioctl */
649 	args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
650 	args->status.replace_state = dev_replace->replace_state;
651 	args->status.time_started = dev_replace->time_started;
652 	args->status.time_stopped = dev_replace->time_stopped;
653 	args->status.num_write_errors =
654 		atomic64_read(&dev_replace->num_write_errors);
655 	args->status.num_uncorrectable_read_errors =
656 		atomic64_read(&dev_replace->num_uncorrectable_read_errors);
657 	switch (dev_replace->replace_state) {
658 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
659 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
660 		args->status.progress_1000 = 0;
661 		break;
662 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
663 		args->status.progress_1000 = 1000;
664 		break;
665 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
666 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
667 		srcdev = dev_replace->srcdev;
668 		args->status.progress_1000 = div_u64(dev_replace->cursor_left,
669 			div_u64(btrfs_device_get_total_bytes(srcdev), 1000));
670 		break;
671 	}
672 	btrfs_dev_replace_unlock(dev_replace, 0);
673 }
674 
675 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info,
676 			     struct btrfs_ioctl_dev_replace_args *args)
677 {
678 	args->result = __btrfs_dev_replace_cancel(fs_info);
679 	return 0;
680 }
681 
682 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
683 {
684 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
685 	struct btrfs_device *tgt_device = NULL;
686 	struct btrfs_trans_handle *trans;
687 	struct btrfs_root *root = fs_info->tree_root;
688 	u64 result;
689 	int ret;
690 
691 	if (fs_info->sb->s_flags & MS_RDONLY)
692 		return -EROFS;
693 
694 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
695 	btrfs_dev_replace_lock(dev_replace, 1);
696 	switch (dev_replace->replace_state) {
697 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
698 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
699 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
700 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
701 		btrfs_dev_replace_unlock(dev_replace, 1);
702 		goto leave;
703 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
704 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
705 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
706 		tgt_device = dev_replace->tgtdev;
707 		dev_replace->tgtdev = NULL;
708 		dev_replace->srcdev = NULL;
709 		break;
710 	}
711 	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
712 	dev_replace->time_stopped = get_seconds();
713 	dev_replace->item_needs_writeback = 1;
714 	btrfs_dev_replace_unlock(dev_replace, 1);
715 	btrfs_scrub_cancel(fs_info);
716 
717 	trans = btrfs_start_transaction(root, 0);
718 	if (IS_ERR(trans)) {
719 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
720 		return PTR_ERR(trans);
721 	}
722 	ret = btrfs_commit_transaction(trans);
723 	WARN_ON(ret);
724 	if (tgt_device)
725 		btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
726 
727 leave:
728 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
729 	return result;
730 }
731 
732 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
733 {
734 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
735 
736 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
737 	btrfs_dev_replace_lock(dev_replace, 1);
738 	switch (dev_replace->replace_state) {
739 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
740 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
741 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
742 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
743 		break;
744 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
745 		dev_replace->replace_state =
746 			BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
747 		dev_replace->time_stopped = get_seconds();
748 		dev_replace->item_needs_writeback = 1;
749 		btrfs_info(fs_info, "suspending dev_replace for unmount");
750 		break;
751 	}
752 
753 	btrfs_dev_replace_unlock(dev_replace, 1);
754 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
755 }
756 
757 /* resume dev_replace procedure that was interrupted by unmount */
758 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
759 {
760 	struct task_struct *task;
761 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
762 
763 	btrfs_dev_replace_lock(dev_replace, 1);
764 	switch (dev_replace->replace_state) {
765 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
766 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
767 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
768 		btrfs_dev_replace_unlock(dev_replace, 1);
769 		return 0;
770 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
771 		break;
772 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
773 		dev_replace->replace_state =
774 			BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
775 		break;
776 	}
777 	if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
778 		btrfs_info(fs_info,
779 			   "cannot continue dev_replace, tgtdev is missing");
780 		btrfs_info(fs_info,
781 			   "you may cancel the operation after 'mount -o degraded'");
782 		btrfs_dev_replace_unlock(dev_replace, 1);
783 		return 0;
784 	}
785 	btrfs_dev_replace_unlock(dev_replace, 1);
786 
787 	WARN_ON(atomic_xchg(
788 		&fs_info->mutually_exclusive_operation_running, 1));
789 	task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
790 	return PTR_ERR_OR_ZERO(task);
791 }
792 
793 static int btrfs_dev_replace_kthread(void *data)
794 {
795 	struct btrfs_fs_info *fs_info = data;
796 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
797 	struct btrfs_ioctl_dev_replace_args *status_args;
798 	u64 progress;
799 
800 	status_args = kzalloc(sizeof(*status_args), GFP_KERNEL);
801 	if (status_args) {
802 		btrfs_dev_replace_status(fs_info, status_args);
803 		progress = status_args->status.progress_1000;
804 		kfree(status_args);
805 		progress = div_u64(progress, 10);
806 		btrfs_info_in_rcu(fs_info,
807 			"continuing dev_replace from %s (devid %llu) to %s @%u%%",
808 			dev_replace->srcdev->missing ? "<missing disk>" :
809 			rcu_str_deref(dev_replace->srcdev->name),
810 			dev_replace->srcdev->devid,
811 			dev_replace->tgtdev ?
812 			rcu_str_deref(dev_replace->tgtdev->name) :
813 			"<missing target disk>",
814 			(unsigned int)progress);
815 	}
816 	btrfs_dev_replace_continue_on_mount(fs_info);
817 	atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
818 
819 	return 0;
820 }
821 
822 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info)
823 {
824 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
825 	int ret;
826 
827 	ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
828 			      dev_replace->committed_cursor_left,
829 			      btrfs_device_get_total_bytes(dev_replace->srcdev),
830 			      &dev_replace->scrub_progress, 0, 1);
831 	ret = btrfs_dev_replace_finishing(fs_info, ret);
832 	WARN_ON(ret);
833 	return 0;
834 }
835 
836 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
837 {
838 	if (!dev_replace->is_valid)
839 		return 0;
840 
841 	switch (dev_replace->replace_state) {
842 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
843 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
844 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
845 		return 0;
846 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
847 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
848 		/*
849 		 * return true even if tgtdev is missing (this is
850 		 * something that can happen if the dev_replace
851 		 * procedure is suspended by an umount and then
852 		 * the tgtdev is missing (or "btrfs dev scan") was
853 		 * not called and the the filesystem is remounted
854 		 * in degraded state. This does not stop the
855 		 * dev_replace procedure. It needs to be canceled
856 		 * manually if the cancellation is wanted.
857 		 */
858 		break;
859 	}
860 	return 1;
861 }
862 
863 void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace, int rw)
864 {
865 	if (rw == 1) {
866 		/* write */
867 again:
868 		wait_event(dev_replace->read_lock_wq,
869 			   atomic_read(&dev_replace->blocking_readers) == 0);
870 		write_lock(&dev_replace->lock);
871 		if (atomic_read(&dev_replace->blocking_readers)) {
872 			write_unlock(&dev_replace->lock);
873 			goto again;
874 		}
875 	} else {
876 		read_lock(&dev_replace->lock);
877 		atomic_inc(&dev_replace->read_locks);
878 	}
879 }
880 
881 void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace, int rw)
882 {
883 	if (rw == 1) {
884 		/* write */
885 		ASSERT(atomic_read(&dev_replace->blocking_readers) == 0);
886 		write_unlock(&dev_replace->lock);
887 	} else {
888 		ASSERT(atomic_read(&dev_replace->read_locks) > 0);
889 		atomic_dec(&dev_replace->read_locks);
890 		read_unlock(&dev_replace->lock);
891 	}
892 }
893 
894 /* inc blocking cnt and release read lock */
895 void btrfs_dev_replace_set_lock_blocking(
896 					struct btrfs_dev_replace *dev_replace)
897 {
898 	/* only set blocking for read lock */
899 	ASSERT(atomic_read(&dev_replace->read_locks) > 0);
900 	atomic_inc(&dev_replace->blocking_readers);
901 	read_unlock(&dev_replace->lock);
902 }
903 
904 /* acquire read lock and dec blocking cnt */
905 void btrfs_dev_replace_clear_lock_blocking(
906 					struct btrfs_dev_replace *dev_replace)
907 {
908 	/* only set blocking for read lock */
909 	ASSERT(atomic_read(&dev_replace->read_locks) > 0);
910 	ASSERT(atomic_read(&dev_replace->blocking_readers) > 0);
911 	read_lock(&dev_replace->lock);
912 	if (atomic_dec_and_test(&dev_replace->blocking_readers) &&
913 	    waitqueue_active(&dev_replace->read_lock_wq))
914 		wake_up(&dev_replace->read_lock_wq);
915 }
916 
917 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
918 {
919 	percpu_counter_inc(&fs_info->bio_counter);
920 }
921 
922 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
923 {
924 	percpu_counter_sub(&fs_info->bio_counter, amount);
925 
926 	if (waitqueue_active(&fs_info->replace_wait))
927 		wake_up(&fs_info->replace_wait);
928 }
929 
930 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
931 {
932 	while (1) {
933 		percpu_counter_inc(&fs_info->bio_counter);
934 		if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
935 				     &fs_info->fs_state)))
936 			break;
937 
938 		btrfs_bio_counter_dec(fs_info);
939 		wait_event(fs_info->replace_wait,
940 			   !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
941 				     &fs_info->fs_state));
942 	}
943 }
944