xref: /openbmc/linux/fs/btrfs/dev-replace.c (revision 6dfcd296)
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(dev_root->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(dev_root->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_root *root, char *tgtdev_name,
308 				u64 srcdevid, char *srcdev_name, int read_src)
309 {
310 	struct btrfs_trans_handle *trans;
311 	struct btrfs_fs_info *fs_info = root->fs_info;
312 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
313 	int ret;
314 	struct btrfs_device *tgt_device = NULL;
315 	struct btrfs_device *src_device = NULL;
316 
317 	/* the disk copy procedure reuses the scrub code */
318 	mutex_lock(&fs_info->volume_mutex);
319 	ret = btrfs_find_device_by_devspec(root, srcdevid,
320 					    srcdev_name, &src_device);
321 	if (ret) {
322 		mutex_unlock(&fs_info->volume_mutex);
323 		return ret;
324 	}
325 
326 	ret = btrfs_init_dev_replace_tgtdev(root, tgtdev_name,
327 					    src_device, &tgt_device);
328 	mutex_unlock(&fs_info->volume_mutex);
329 	if (ret)
330 		return ret;
331 
332 	/*
333 	 * Here we commit the transaction to make sure commit_total_bytes
334 	 * of all the devices are updated.
335 	 */
336 	trans = btrfs_attach_transaction(root);
337 	if (!IS_ERR(trans)) {
338 		ret = btrfs_commit_transaction(trans, root);
339 		if (ret)
340 			return ret;
341 	} else if (PTR_ERR(trans) != -ENOENT) {
342 		return PTR_ERR(trans);
343 	}
344 
345 	btrfs_dev_replace_lock(dev_replace, 1);
346 	switch (dev_replace->replace_state) {
347 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
348 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
349 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
350 		break;
351 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
352 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
353 		ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
354 		goto leave;
355 	}
356 
357 	dev_replace->cont_reading_from_srcdev_mode = read_src;
358 	WARN_ON(!src_device);
359 	dev_replace->srcdev = src_device;
360 	WARN_ON(!tgt_device);
361 	dev_replace->tgtdev = tgt_device;
362 
363 	btrfs_info_in_rcu(fs_info,
364 		      "dev_replace from %s (devid %llu) to %s started",
365 		      src_device->missing ? "<missing disk>" :
366 		        rcu_str_deref(src_device->name),
367 		      src_device->devid,
368 		      rcu_str_deref(tgt_device->name));
369 
370 	/*
371 	 * from now on, the writes to the srcdev are all duplicated to
372 	 * go to the tgtdev as well (refer to btrfs_map_block()).
373 	 */
374 	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
375 	dev_replace->time_started = get_seconds();
376 	dev_replace->cursor_left = 0;
377 	dev_replace->committed_cursor_left = 0;
378 	dev_replace->cursor_left_last_write_of_item = 0;
379 	dev_replace->cursor_right = 0;
380 	dev_replace->is_valid = 1;
381 	dev_replace->item_needs_writeback = 1;
382 	atomic64_set(&dev_replace->num_write_errors, 0);
383 	atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
384 	btrfs_dev_replace_unlock(dev_replace, 1);
385 
386 	ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device);
387 	if (ret)
388 		btrfs_err(fs_info, "kobj add dev failed %d", ret);
389 
390 	btrfs_wait_ordered_roots(root->fs_info, -1, 0, (u64)-1);
391 
392 	/* force writing the updated state information to disk */
393 	trans = btrfs_start_transaction(root, 0);
394 	if (IS_ERR(trans)) {
395 		ret = PTR_ERR(trans);
396 		btrfs_dev_replace_lock(dev_replace, 1);
397 		goto leave;
398 	}
399 
400 	ret = btrfs_commit_transaction(trans, root);
401 	WARN_ON(ret);
402 
403 	/* the disk copy procedure reuses the scrub code */
404 	ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
405 			      btrfs_device_get_total_bytes(src_device),
406 			      &dev_replace->scrub_progress, 0, 1);
407 
408 	ret = btrfs_dev_replace_finishing(fs_info, ret);
409 	if (ret == -EINPROGRESS) {
410 		ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
411 	} else {
412 		WARN_ON(ret);
413 	}
414 
415 	return ret;
416 
417 leave:
418 	dev_replace->srcdev = NULL;
419 	dev_replace->tgtdev = NULL;
420 	btrfs_dev_replace_unlock(dev_replace, 1);
421 	btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
422 	return ret;
423 }
424 
425 int btrfs_dev_replace_by_ioctl(struct btrfs_root *root,
426 			    struct btrfs_ioctl_dev_replace_args *args)
427 {
428 	int ret;
429 
430 	switch (args->start.cont_reading_from_srcdev_mode) {
431 	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
432 	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
433 		break;
434 	default:
435 		return -EINVAL;
436 	}
437 
438 	if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
439 	    args->start.tgtdev_name[0] == '\0')
440 		return -EINVAL;
441 
442 	ret = btrfs_dev_replace_start(root, args->start.tgtdev_name,
443 					args->start.srcdevid,
444 					args->start.srcdev_name,
445 					args->start.cont_reading_from_srcdev_mode);
446 	args->result = ret;
447 	/* don't warn if EINPROGRESS, someone else might be running scrub */
448 	if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS)
449 		ret = 0;
450 
451 	return ret;
452 }
453 
454 /*
455  * blocked until all in-flight bios operations are finished.
456  */
457 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
458 {
459 	set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
460 	wait_event(fs_info->replace_wait, !percpu_counter_sum(
461 		   &fs_info->bio_counter));
462 }
463 
464 /*
465  * we have removed target device, it is safe to allow new bios request.
466  */
467 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
468 {
469 	clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
470 	wake_up(&fs_info->replace_wait);
471 }
472 
473 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
474 				       int scrub_ret)
475 {
476 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
477 	struct btrfs_device *tgt_device;
478 	struct btrfs_device *src_device;
479 	struct btrfs_root *root = fs_info->tree_root;
480 	u8 uuid_tmp[BTRFS_UUID_SIZE];
481 	struct btrfs_trans_handle *trans;
482 	int ret = 0;
483 
484 	/* don't allow cancel or unmount to disturb the finishing procedure */
485 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
486 
487 	btrfs_dev_replace_lock(dev_replace, 0);
488 	/* was the operation canceled, or is it finished? */
489 	if (dev_replace->replace_state !=
490 	    BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
491 		btrfs_dev_replace_unlock(dev_replace, 0);
492 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
493 		return 0;
494 	}
495 
496 	tgt_device = dev_replace->tgtdev;
497 	src_device = dev_replace->srcdev;
498 	btrfs_dev_replace_unlock(dev_replace, 0);
499 
500 	/*
501 	 * flush all outstanding I/O and inode extent mappings before the
502 	 * copy operation is declared as being finished
503 	 */
504 	ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
505 	if (ret) {
506 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
507 		return ret;
508 	}
509 	btrfs_wait_ordered_roots(root->fs_info, -1, 0, (u64)-1);
510 
511 	trans = btrfs_start_transaction(root, 0);
512 	if (IS_ERR(trans)) {
513 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
514 		return PTR_ERR(trans);
515 	}
516 	ret = btrfs_commit_transaction(trans, root);
517 	WARN_ON(ret);
518 
519 	mutex_lock(&uuid_mutex);
520 	/* keep away write_all_supers() during the finishing procedure */
521 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
522 	mutex_lock(&root->fs_info->chunk_mutex);
523 	btrfs_dev_replace_lock(dev_replace, 1);
524 	dev_replace->replace_state =
525 		scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
526 			  : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
527 	dev_replace->tgtdev = NULL;
528 	dev_replace->srcdev = NULL;
529 	dev_replace->time_stopped = get_seconds();
530 	dev_replace->item_needs_writeback = 1;
531 
532 	/* replace old device with new one in mapping tree */
533 	if (!scrub_ret) {
534 		btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
535 								src_device,
536 								tgt_device);
537 	} else {
538 		btrfs_err_in_rcu(root->fs_info,
539 			      "btrfs_scrub_dev(%s, %llu, %s) failed %d",
540 			      src_device->missing ? "<missing disk>" :
541 			        rcu_str_deref(src_device->name),
542 			      src_device->devid,
543 			      rcu_str_deref(tgt_device->name), scrub_ret);
544 		btrfs_dev_replace_unlock(dev_replace, 1);
545 		mutex_unlock(&root->fs_info->chunk_mutex);
546 		mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
547 		mutex_unlock(&uuid_mutex);
548 		if (tgt_device)
549 			btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
550 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
551 
552 		return scrub_ret;
553 	}
554 
555 	btrfs_info_in_rcu(root->fs_info,
556 		      "dev_replace from %s (devid %llu) to %s finished",
557 		      src_device->missing ? "<missing disk>" :
558 		        rcu_str_deref(src_device->name),
559 		      src_device->devid,
560 		      rcu_str_deref(tgt_device->name));
561 	tgt_device->is_tgtdev_for_dev_replace = 0;
562 	tgt_device->devid = src_device->devid;
563 	src_device->devid = BTRFS_DEV_REPLACE_DEVID;
564 	memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
565 	memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
566 	memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
567 	btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
568 	btrfs_device_set_disk_total_bytes(tgt_device,
569 					  src_device->disk_total_bytes);
570 	btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
571 	ASSERT(list_empty(&src_device->resized_list));
572 	tgt_device->commit_total_bytes = src_device->commit_total_bytes;
573 	tgt_device->commit_bytes_used = src_device->bytes_used;
574 
575 	btrfs_assign_next_active_device(fs_info, src_device, tgt_device);
576 
577 	list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
578 	fs_info->fs_devices->rw_devices++;
579 
580 	btrfs_dev_replace_unlock(dev_replace, 1);
581 
582 	btrfs_rm_dev_replace_blocked(fs_info);
583 
584 	btrfs_rm_dev_replace_remove_srcdev(fs_info, src_device);
585 
586 	btrfs_rm_dev_replace_unblocked(fs_info);
587 
588 	/*
589 	 * this is again a consistent state where no dev_replace procedure
590 	 * is running, the target device is part of the filesystem, the
591 	 * source device is not part of the filesystem anymore and its 1st
592 	 * superblock is scratched out so that it is no longer marked to
593 	 * belong to this filesystem.
594 	 */
595 	mutex_unlock(&root->fs_info->chunk_mutex);
596 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
597 	mutex_unlock(&uuid_mutex);
598 
599 	/* replace the sysfs entry */
600 	btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device);
601 	btrfs_rm_dev_replace_free_srcdev(fs_info, src_device);
602 
603 	/* write back the superblocks */
604 	trans = btrfs_start_transaction(root, 0);
605 	if (!IS_ERR(trans))
606 		btrfs_commit_transaction(trans, root);
607 
608 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
609 
610 	return 0;
611 }
612 
613 static void btrfs_dev_replace_update_device_in_mapping_tree(
614 						struct btrfs_fs_info *fs_info,
615 						struct btrfs_device *srcdev,
616 						struct btrfs_device *tgtdev)
617 {
618 	struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
619 	struct extent_map *em;
620 	struct map_lookup *map;
621 	u64 start = 0;
622 	int i;
623 
624 	write_lock(&em_tree->lock);
625 	do {
626 		em = lookup_extent_mapping(em_tree, start, (u64)-1);
627 		if (!em)
628 			break;
629 		map = em->map_lookup;
630 		for (i = 0; i < map->num_stripes; i++)
631 			if (srcdev == map->stripes[i].dev)
632 				map->stripes[i].dev = tgtdev;
633 		start = em->start + em->len;
634 		free_extent_map(em);
635 	} while (start);
636 	write_unlock(&em_tree->lock);
637 }
638 
639 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
640 			      struct btrfs_ioctl_dev_replace_args *args)
641 {
642 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
643 	struct btrfs_device *srcdev;
644 
645 	btrfs_dev_replace_lock(dev_replace, 0);
646 	/* even if !dev_replace_is_valid, the values are good enough for
647 	 * the replace_status ioctl */
648 	args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
649 	args->status.replace_state = dev_replace->replace_state;
650 	args->status.time_started = dev_replace->time_started;
651 	args->status.time_stopped = dev_replace->time_stopped;
652 	args->status.num_write_errors =
653 		atomic64_read(&dev_replace->num_write_errors);
654 	args->status.num_uncorrectable_read_errors =
655 		atomic64_read(&dev_replace->num_uncorrectable_read_errors);
656 	switch (dev_replace->replace_state) {
657 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
658 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
659 		args->status.progress_1000 = 0;
660 		break;
661 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
662 		args->status.progress_1000 = 1000;
663 		break;
664 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
665 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
666 		srcdev = dev_replace->srcdev;
667 		args->status.progress_1000 = div_u64(dev_replace->cursor_left,
668 			div_u64(btrfs_device_get_total_bytes(srcdev), 1000));
669 		break;
670 	}
671 	btrfs_dev_replace_unlock(dev_replace, 0);
672 }
673 
674 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info,
675 			     struct btrfs_ioctl_dev_replace_args *args)
676 {
677 	args->result = __btrfs_dev_replace_cancel(fs_info);
678 	return 0;
679 }
680 
681 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
682 {
683 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
684 	struct btrfs_device *tgt_device = NULL;
685 	struct btrfs_trans_handle *trans;
686 	struct btrfs_root *root = fs_info->tree_root;
687 	u64 result;
688 	int ret;
689 
690 	if (fs_info->sb->s_flags & MS_RDONLY)
691 		return -EROFS;
692 
693 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
694 	btrfs_dev_replace_lock(dev_replace, 1);
695 	switch (dev_replace->replace_state) {
696 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
697 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
698 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
699 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
700 		btrfs_dev_replace_unlock(dev_replace, 1);
701 		goto leave;
702 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
703 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
704 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
705 		tgt_device = dev_replace->tgtdev;
706 		dev_replace->tgtdev = NULL;
707 		dev_replace->srcdev = NULL;
708 		break;
709 	}
710 	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
711 	dev_replace->time_stopped = get_seconds();
712 	dev_replace->item_needs_writeback = 1;
713 	btrfs_dev_replace_unlock(dev_replace, 1);
714 	btrfs_scrub_cancel(fs_info);
715 
716 	trans = btrfs_start_transaction(root, 0);
717 	if (IS_ERR(trans)) {
718 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
719 		return PTR_ERR(trans);
720 	}
721 	ret = btrfs_commit_transaction(trans, root);
722 	WARN_ON(ret);
723 	if (tgt_device)
724 		btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
725 
726 leave:
727 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
728 	return result;
729 }
730 
731 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
732 {
733 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
734 
735 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
736 	btrfs_dev_replace_lock(dev_replace, 1);
737 	switch (dev_replace->replace_state) {
738 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
739 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
740 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
741 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
742 		break;
743 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
744 		dev_replace->replace_state =
745 			BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
746 		dev_replace->time_stopped = get_seconds();
747 		dev_replace->item_needs_writeback = 1;
748 		btrfs_info(fs_info, "suspending dev_replace for unmount");
749 		break;
750 	}
751 
752 	btrfs_dev_replace_unlock(dev_replace, 1);
753 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
754 }
755 
756 /* resume dev_replace procedure that was interrupted by unmount */
757 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
758 {
759 	struct task_struct *task;
760 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
761 
762 	btrfs_dev_replace_lock(dev_replace, 1);
763 	switch (dev_replace->replace_state) {
764 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
765 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
766 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
767 		btrfs_dev_replace_unlock(dev_replace, 1);
768 		return 0;
769 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
770 		break;
771 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
772 		dev_replace->replace_state =
773 			BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
774 		break;
775 	}
776 	if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
777 		btrfs_info(fs_info,
778 			   "cannot continue dev_replace, tgtdev is missing");
779 		btrfs_info(fs_info,
780 			   "you may cancel the operation after 'mount -o degraded'");
781 		btrfs_dev_replace_unlock(dev_replace, 1);
782 		return 0;
783 	}
784 	btrfs_dev_replace_unlock(dev_replace, 1);
785 
786 	WARN_ON(atomic_xchg(
787 		&fs_info->mutually_exclusive_operation_running, 1));
788 	task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
789 	return PTR_ERR_OR_ZERO(task);
790 }
791 
792 static int btrfs_dev_replace_kthread(void *data)
793 {
794 	struct btrfs_fs_info *fs_info = data;
795 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
796 	struct btrfs_ioctl_dev_replace_args *status_args;
797 	u64 progress;
798 
799 	status_args = kzalloc(sizeof(*status_args), GFP_KERNEL);
800 	if (status_args) {
801 		btrfs_dev_replace_status(fs_info, status_args);
802 		progress = status_args->status.progress_1000;
803 		kfree(status_args);
804 		progress = div_u64(progress, 10);
805 		btrfs_info_in_rcu(fs_info,
806 			"continuing dev_replace from %s (devid %llu) to %s @%u%%",
807 			dev_replace->srcdev->missing ? "<missing disk>" :
808 			rcu_str_deref(dev_replace->srcdev->name),
809 			dev_replace->srcdev->devid,
810 			dev_replace->tgtdev ?
811 			rcu_str_deref(dev_replace->tgtdev->name) :
812 			"<missing target disk>",
813 			(unsigned int)progress);
814 	}
815 	btrfs_dev_replace_continue_on_mount(fs_info);
816 	atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
817 
818 	return 0;
819 }
820 
821 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info)
822 {
823 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
824 	int ret;
825 
826 	ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
827 			      dev_replace->committed_cursor_left,
828 			      btrfs_device_get_total_bytes(dev_replace->srcdev),
829 			      &dev_replace->scrub_progress, 0, 1);
830 	ret = btrfs_dev_replace_finishing(fs_info, ret);
831 	WARN_ON(ret);
832 	return 0;
833 }
834 
835 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
836 {
837 	if (!dev_replace->is_valid)
838 		return 0;
839 
840 	switch (dev_replace->replace_state) {
841 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
842 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
843 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
844 		return 0;
845 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
846 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
847 		/*
848 		 * return true even if tgtdev is missing (this is
849 		 * something that can happen if the dev_replace
850 		 * procedure is suspended by an umount and then
851 		 * the tgtdev is missing (or "btrfs dev scan") was
852 		 * not called and the the filesystem is remounted
853 		 * in degraded state. This does not stop the
854 		 * dev_replace procedure. It needs to be canceled
855 		 * manually if the cancellation is wanted.
856 		 */
857 		break;
858 	}
859 	return 1;
860 }
861 
862 void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace, int rw)
863 {
864 	if (rw == 1) {
865 		/* write */
866 again:
867 		wait_event(dev_replace->read_lock_wq,
868 			   atomic_read(&dev_replace->blocking_readers) == 0);
869 		write_lock(&dev_replace->lock);
870 		if (atomic_read(&dev_replace->blocking_readers)) {
871 			write_unlock(&dev_replace->lock);
872 			goto again;
873 		}
874 	} else {
875 		read_lock(&dev_replace->lock);
876 		atomic_inc(&dev_replace->read_locks);
877 	}
878 }
879 
880 void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace, int rw)
881 {
882 	if (rw == 1) {
883 		/* write */
884 		ASSERT(atomic_read(&dev_replace->blocking_readers) == 0);
885 		write_unlock(&dev_replace->lock);
886 	} else {
887 		ASSERT(atomic_read(&dev_replace->read_locks) > 0);
888 		atomic_dec(&dev_replace->read_locks);
889 		read_unlock(&dev_replace->lock);
890 	}
891 }
892 
893 /* inc blocking cnt and release read lock */
894 void btrfs_dev_replace_set_lock_blocking(
895 					struct btrfs_dev_replace *dev_replace)
896 {
897 	/* only set blocking for read lock */
898 	ASSERT(atomic_read(&dev_replace->read_locks) > 0);
899 	atomic_inc(&dev_replace->blocking_readers);
900 	read_unlock(&dev_replace->lock);
901 }
902 
903 /* acquire read lock and dec blocking cnt */
904 void btrfs_dev_replace_clear_lock_blocking(
905 					struct btrfs_dev_replace *dev_replace)
906 {
907 	/* only set blocking for read lock */
908 	ASSERT(atomic_read(&dev_replace->read_locks) > 0);
909 	ASSERT(atomic_read(&dev_replace->blocking_readers) > 0);
910 	read_lock(&dev_replace->lock);
911 	if (atomic_dec_and_test(&dev_replace->blocking_readers) &&
912 	    waitqueue_active(&dev_replace->read_lock_wq))
913 		wake_up(&dev_replace->read_lock_wq);
914 }
915 
916 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
917 {
918 	percpu_counter_inc(&fs_info->bio_counter);
919 }
920 
921 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
922 {
923 	percpu_counter_sub(&fs_info->bio_counter, amount);
924 
925 	if (waitqueue_active(&fs_info->replace_wait))
926 		wake_up(&fs_info->replace_wait);
927 }
928 
929 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
930 {
931 	while (1) {
932 		percpu_counter_inc(&fs_info->bio_counter);
933 		if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
934 				     &fs_info->fs_state)))
935 			break;
936 
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 	}
942 }
943