xref: /openbmc/linux/fs/btrfs/dev-replace.c (revision 8e9356c6)
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 
40 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
41 				       int scrub_ret);
42 static void btrfs_dev_replace_update_device_in_mapping_tree(
43 						struct btrfs_fs_info *fs_info,
44 						struct btrfs_device *srcdev,
45 						struct btrfs_device *tgtdev);
46 static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid,
47 					 char *srcdev_name,
48 					 struct btrfs_device **device);
49 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info);
50 static int btrfs_dev_replace_kthread(void *data);
51 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info);
52 
53 
54 int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
55 {
56 	struct btrfs_key key;
57 	struct btrfs_root *dev_root = fs_info->dev_root;
58 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
59 	struct extent_buffer *eb;
60 	int slot;
61 	int ret = 0;
62 	struct btrfs_path *path = NULL;
63 	int item_size;
64 	struct btrfs_dev_replace_item *ptr;
65 	u64 src_devid;
66 
67 	path = btrfs_alloc_path();
68 	if (!path) {
69 		ret = -ENOMEM;
70 		goto out;
71 	}
72 
73 	key.objectid = 0;
74 	key.type = BTRFS_DEV_REPLACE_KEY;
75 	key.offset = 0;
76 	ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
77 	if (ret) {
78 no_valid_dev_replace_entry_found:
79 		ret = 0;
80 		dev_replace->replace_state =
81 			BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED;
82 		dev_replace->cont_reading_from_srcdev_mode =
83 		    BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
84 		dev_replace->replace_state = 0;
85 		dev_replace->time_started = 0;
86 		dev_replace->time_stopped = 0;
87 		atomic64_set(&dev_replace->num_write_errors, 0);
88 		atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
89 		dev_replace->cursor_left = 0;
90 		dev_replace->committed_cursor_left = 0;
91 		dev_replace->cursor_left_last_write_of_item = 0;
92 		dev_replace->cursor_right = 0;
93 		dev_replace->srcdev = NULL;
94 		dev_replace->tgtdev = NULL;
95 		dev_replace->is_valid = 0;
96 		dev_replace->item_needs_writeback = 0;
97 		goto out;
98 	}
99 	slot = path->slots[0];
100 	eb = path->nodes[0];
101 	item_size = btrfs_item_size_nr(eb, slot);
102 	ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
103 
104 	if (item_size != sizeof(struct btrfs_dev_replace_item)) {
105 		btrfs_warn(fs_info,
106 			"dev_replace entry found has unexpected size, ignore entry");
107 		goto no_valid_dev_replace_entry_found;
108 	}
109 
110 	src_devid = btrfs_dev_replace_src_devid(eb, ptr);
111 	dev_replace->cont_reading_from_srcdev_mode =
112 		btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
113 	dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
114 	dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
115 	dev_replace->time_stopped =
116 		btrfs_dev_replace_time_stopped(eb, ptr);
117 	atomic64_set(&dev_replace->num_write_errors,
118 		     btrfs_dev_replace_num_write_errors(eb, ptr));
119 	atomic64_set(&dev_replace->num_uncorrectable_read_errors,
120 		     btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
121 	dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
122 	dev_replace->committed_cursor_left = dev_replace->cursor_left;
123 	dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
124 	dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
125 	dev_replace->is_valid = 1;
126 
127 	dev_replace->item_needs_writeback = 0;
128 	switch (dev_replace->replace_state) {
129 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
130 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
131 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
132 		dev_replace->srcdev = NULL;
133 		dev_replace->tgtdev = NULL;
134 		break;
135 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
136 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
137 		dev_replace->srcdev = btrfs_find_device(fs_info, src_devid,
138 							NULL, NULL);
139 		dev_replace->tgtdev = btrfs_find_device(fs_info,
140 							BTRFS_DEV_REPLACE_DEVID,
141 							NULL, NULL);
142 		/*
143 		 * allow 'btrfs dev replace_cancel' if src/tgt device is
144 		 * missing
145 		 */
146 		if (!dev_replace->srcdev &&
147 		    !btrfs_test_opt(dev_root, DEGRADED)) {
148 			ret = -EIO;
149 			btrfs_warn(fs_info,
150 			   "cannot mount because device replace operation is ongoing and");
151 			btrfs_warn(fs_info,
152 			   "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
153 			   src_devid);
154 		}
155 		if (!dev_replace->tgtdev &&
156 		    !btrfs_test_opt(dev_root, DEGRADED)) {
157 			ret = -EIO;
158 			btrfs_warn(fs_info,
159 			   "cannot mount because device replace operation is ongoing and");
160 			btrfs_warn(fs_info,
161 			   "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
162 				BTRFS_DEV_REPLACE_DEVID);
163 		}
164 		if (dev_replace->tgtdev) {
165 			if (dev_replace->srcdev) {
166 				dev_replace->tgtdev->total_bytes =
167 					dev_replace->srcdev->total_bytes;
168 				dev_replace->tgtdev->disk_total_bytes =
169 					dev_replace->srcdev->disk_total_bytes;
170 				dev_replace->tgtdev->bytes_used =
171 					dev_replace->srcdev->bytes_used;
172 			}
173 			dev_replace->tgtdev->is_tgtdev_for_dev_replace = 1;
174 			btrfs_init_dev_replace_tgtdev_for_resume(fs_info,
175 				dev_replace->tgtdev);
176 		}
177 		break;
178 	}
179 
180 out:
181 	if (path)
182 		btrfs_free_path(path);
183 	return ret;
184 }
185 
186 /*
187  * called from commit_transaction. Writes changed device replace state to
188  * disk.
189  */
190 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
191 			  struct btrfs_fs_info *fs_info)
192 {
193 	int ret;
194 	struct btrfs_root *dev_root = fs_info->dev_root;
195 	struct btrfs_path *path;
196 	struct btrfs_key key;
197 	struct extent_buffer *eb;
198 	struct btrfs_dev_replace_item *ptr;
199 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
200 
201 	btrfs_dev_replace_lock(dev_replace);
202 	if (!dev_replace->is_valid ||
203 	    !dev_replace->item_needs_writeback) {
204 		btrfs_dev_replace_unlock(dev_replace);
205 		return 0;
206 	}
207 	btrfs_dev_replace_unlock(dev_replace);
208 
209 	key.objectid = 0;
210 	key.type = BTRFS_DEV_REPLACE_KEY;
211 	key.offset = 0;
212 
213 	path = btrfs_alloc_path();
214 	if (!path) {
215 		ret = -ENOMEM;
216 		goto out;
217 	}
218 	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
219 	if (ret < 0) {
220 		btrfs_warn(fs_info, "error %d while searching for dev_replace item!",
221 			ret);
222 		goto out;
223 	}
224 
225 	if (ret == 0 &&
226 	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
227 		/*
228 		 * need to delete old one and insert a new one.
229 		 * Since no attempt is made to recover any old state, if the
230 		 * dev_replace state is 'running', the data on the target
231 		 * drive is lost.
232 		 * It would be possible to recover the state: just make sure
233 		 * that the beginning of the item is never changed and always
234 		 * contains all the essential information. Then read this
235 		 * minimal set of information and use it as a base for the
236 		 * new state.
237 		 */
238 		ret = btrfs_del_item(trans, dev_root, path);
239 		if (ret != 0) {
240 			btrfs_warn(fs_info, "delete too small dev_replace item failed %d!",
241 				ret);
242 			goto out;
243 		}
244 		ret = 1;
245 	}
246 
247 	if (ret == 1) {
248 		/* need to insert a new item */
249 		btrfs_release_path(path);
250 		ret = btrfs_insert_empty_item(trans, dev_root, path,
251 					      &key, sizeof(*ptr));
252 		if (ret < 0) {
253 			btrfs_warn(fs_info, "insert dev_replace item failed %d!",
254 				ret);
255 			goto out;
256 		}
257 	}
258 
259 	eb = path->nodes[0];
260 	ptr = btrfs_item_ptr(eb, path->slots[0],
261 			     struct btrfs_dev_replace_item);
262 
263 	btrfs_dev_replace_lock(dev_replace);
264 	if (dev_replace->srcdev)
265 		btrfs_set_dev_replace_src_devid(eb, ptr,
266 			dev_replace->srcdev->devid);
267 	else
268 		btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
269 	btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
270 		dev_replace->cont_reading_from_srcdev_mode);
271 	btrfs_set_dev_replace_replace_state(eb, ptr,
272 		dev_replace->replace_state);
273 	btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
274 	btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
275 	btrfs_set_dev_replace_num_write_errors(eb, ptr,
276 		atomic64_read(&dev_replace->num_write_errors));
277 	btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
278 		atomic64_read(&dev_replace->num_uncorrectable_read_errors));
279 	dev_replace->cursor_left_last_write_of_item =
280 		dev_replace->cursor_left;
281 	btrfs_set_dev_replace_cursor_left(eb, ptr,
282 		dev_replace->cursor_left_last_write_of_item);
283 	btrfs_set_dev_replace_cursor_right(eb, ptr,
284 		dev_replace->cursor_right);
285 	dev_replace->item_needs_writeback = 0;
286 	btrfs_dev_replace_unlock(dev_replace);
287 
288 	btrfs_mark_buffer_dirty(eb);
289 
290 out:
291 	btrfs_free_path(path);
292 
293 	return ret;
294 }
295 
296 void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
297 {
298 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
299 
300 	dev_replace->committed_cursor_left =
301 		dev_replace->cursor_left_last_write_of_item;
302 }
303 
304 int btrfs_dev_replace_start(struct btrfs_root *root,
305 			    struct btrfs_ioctl_dev_replace_args *args)
306 {
307 	struct btrfs_trans_handle *trans;
308 	struct btrfs_fs_info *fs_info = root->fs_info;
309 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
310 	int ret;
311 	struct btrfs_device *tgt_device = NULL;
312 	struct btrfs_device *src_device = NULL;
313 
314 	if (btrfs_fs_incompat(fs_info, RAID56)) {
315 		btrfs_warn(fs_info, "dev_replace cannot yet handle RAID5/RAID6");
316 		return -EINVAL;
317 	}
318 
319 	switch (args->start.cont_reading_from_srcdev_mode) {
320 	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
321 	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
322 		break;
323 	default:
324 		return -EINVAL;
325 	}
326 
327 	if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
328 	    args->start.tgtdev_name[0] == '\0')
329 		return -EINVAL;
330 
331 	mutex_lock(&fs_info->volume_mutex);
332 	ret = btrfs_init_dev_replace_tgtdev(root, args->start.tgtdev_name,
333 					    &tgt_device);
334 	if (ret) {
335 		btrfs_err(fs_info, "target device %s is invalid!",
336 		       args->start.tgtdev_name);
337 		mutex_unlock(&fs_info->volume_mutex);
338 		return -EINVAL;
339 	}
340 
341 	ret = btrfs_dev_replace_find_srcdev(root, args->start.srcdevid,
342 					    args->start.srcdev_name,
343 					    &src_device);
344 	mutex_unlock(&fs_info->volume_mutex);
345 	if (ret) {
346 		ret = -EINVAL;
347 		goto leave_no_lock;
348 	}
349 
350 	if (tgt_device->total_bytes < src_device->total_bytes) {
351 		btrfs_err(fs_info, "target device is smaller than source device!");
352 		ret = -EINVAL;
353 		goto leave_no_lock;
354 	}
355 
356 	btrfs_dev_replace_lock(dev_replace);
357 	switch (dev_replace->replace_state) {
358 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
359 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
360 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
361 		break;
362 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
363 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
364 		args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
365 		goto leave;
366 	}
367 
368 	dev_replace->cont_reading_from_srcdev_mode =
369 		args->start.cont_reading_from_srcdev_mode;
370 	WARN_ON(!src_device);
371 	dev_replace->srcdev = src_device;
372 	WARN_ON(!tgt_device);
373 	dev_replace->tgtdev = tgt_device;
374 
375 	printk_in_rcu(KERN_INFO
376 		      "BTRFS: dev_replace from %s (devid %llu) to %s started\n",
377 		      src_device->missing ? "<missing disk>" :
378 		        rcu_str_deref(src_device->name),
379 		      src_device->devid,
380 		      rcu_str_deref(tgt_device->name));
381 
382 	tgt_device->total_bytes = src_device->total_bytes;
383 	tgt_device->disk_total_bytes = src_device->disk_total_bytes;
384 	tgt_device->bytes_used = src_device->bytes_used;
385 
386 	/*
387 	 * from now on, the writes to the srcdev are all duplicated to
388 	 * go to the tgtdev as well (refer to btrfs_map_block()).
389 	 */
390 	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
391 	dev_replace->time_started = get_seconds();
392 	dev_replace->cursor_left = 0;
393 	dev_replace->committed_cursor_left = 0;
394 	dev_replace->cursor_left_last_write_of_item = 0;
395 	dev_replace->cursor_right = 0;
396 	dev_replace->is_valid = 1;
397 	dev_replace->item_needs_writeback = 1;
398 	args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
399 	btrfs_dev_replace_unlock(dev_replace);
400 
401 	btrfs_wait_ordered_roots(root->fs_info, -1);
402 
403 	/* force writing the updated state information to disk */
404 	trans = btrfs_start_transaction(root, 0);
405 	if (IS_ERR(trans)) {
406 		ret = PTR_ERR(trans);
407 		btrfs_dev_replace_lock(dev_replace);
408 		goto leave;
409 	}
410 
411 	ret = btrfs_commit_transaction(trans, root);
412 	WARN_ON(ret);
413 
414 	/* the disk copy procedure reuses the scrub code */
415 	ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
416 			      src_device->total_bytes,
417 			      &dev_replace->scrub_progress, 0, 1);
418 
419 	ret = btrfs_dev_replace_finishing(root->fs_info, ret);
420 	WARN_ON(ret);
421 
422 	return 0;
423 
424 leave:
425 	dev_replace->srcdev = NULL;
426 	dev_replace->tgtdev = NULL;
427 	btrfs_dev_replace_unlock(dev_replace);
428 leave_no_lock:
429 	if (tgt_device)
430 		btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
431 	return ret;
432 }
433 
434 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
435 				       int scrub_ret)
436 {
437 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
438 	struct btrfs_device *tgt_device;
439 	struct btrfs_device *src_device;
440 	struct btrfs_root *root = fs_info->tree_root;
441 	u8 uuid_tmp[BTRFS_UUID_SIZE];
442 	struct btrfs_trans_handle *trans;
443 	int ret = 0;
444 
445 	/* don't allow cancel or unmount to disturb the finishing procedure */
446 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
447 
448 	btrfs_dev_replace_lock(dev_replace);
449 	/* was the operation canceled, or is it finished? */
450 	if (dev_replace->replace_state !=
451 	    BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
452 		btrfs_dev_replace_unlock(dev_replace);
453 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
454 		return 0;
455 	}
456 
457 	tgt_device = dev_replace->tgtdev;
458 	src_device = dev_replace->srcdev;
459 	btrfs_dev_replace_unlock(dev_replace);
460 
461 	/* replace old device with new one in mapping tree */
462 	if (!scrub_ret)
463 		btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
464 								src_device,
465 								tgt_device);
466 
467 	/*
468 	 * flush all outstanding I/O and inode extent mappings before the
469 	 * copy operation is declared as being finished
470 	 */
471 	ret = btrfs_start_delalloc_roots(root->fs_info, 0);
472 	if (ret) {
473 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
474 		return ret;
475 	}
476 	btrfs_wait_ordered_roots(root->fs_info, -1);
477 
478 	trans = btrfs_start_transaction(root, 0);
479 	if (IS_ERR(trans)) {
480 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
481 		return PTR_ERR(trans);
482 	}
483 	ret = btrfs_commit_transaction(trans, root);
484 	WARN_ON(ret);
485 
486 	/* keep away write_all_supers() during the finishing procedure */
487 	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
488 	btrfs_dev_replace_lock(dev_replace);
489 	dev_replace->replace_state =
490 		scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
491 			  : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
492 	dev_replace->tgtdev = NULL;
493 	dev_replace->srcdev = NULL;
494 	dev_replace->time_stopped = get_seconds();
495 	dev_replace->item_needs_writeback = 1;
496 
497 	if (scrub_ret) {
498 		printk_in_rcu(KERN_ERR
499 			      "BTRFS: btrfs_scrub_dev(%s, %llu, %s) failed %d\n",
500 			      src_device->missing ? "<missing disk>" :
501 			        rcu_str_deref(src_device->name),
502 			      src_device->devid,
503 			      rcu_str_deref(tgt_device->name), scrub_ret);
504 		btrfs_dev_replace_unlock(dev_replace);
505 		mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
506 		if (tgt_device)
507 			btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
508 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
509 
510 		return 0;
511 	}
512 
513 	printk_in_rcu(KERN_INFO
514 		      "BTRFS: dev_replace from %s (devid %llu) to %s) finished\n",
515 		      src_device->missing ? "<missing disk>" :
516 		        rcu_str_deref(src_device->name),
517 		      src_device->devid,
518 		      rcu_str_deref(tgt_device->name));
519 	tgt_device->is_tgtdev_for_dev_replace = 0;
520 	tgt_device->devid = src_device->devid;
521 	src_device->devid = BTRFS_DEV_REPLACE_DEVID;
522 	tgt_device->bytes_used = src_device->bytes_used;
523 	memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
524 	memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
525 	memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
526 	tgt_device->total_bytes = src_device->total_bytes;
527 	tgt_device->disk_total_bytes = src_device->disk_total_bytes;
528 	tgt_device->bytes_used = src_device->bytes_used;
529 	if (fs_info->sb->s_bdev == src_device->bdev)
530 		fs_info->sb->s_bdev = tgt_device->bdev;
531 	if (fs_info->fs_devices->latest_bdev == src_device->bdev)
532 		fs_info->fs_devices->latest_bdev = tgt_device->bdev;
533 	list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
534 
535 	btrfs_rm_dev_replace_srcdev(fs_info, src_device);
536 
537 	/*
538 	 * this is again a consistent state where no dev_replace procedure
539 	 * is running, the target device is part of the filesystem, the
540 	 * source device is not part of the filesystem anymore and its 1st
541 	 * superblock is scratched out so that it is no longer marked to
542 	 * belong to this filesystem.
543 	 */
544 	btrfs_dev_replace_unlock(dev_replace);
545 	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
546 
547 	/* write back the superblocks */
548 	trans = btrfs_start_transaction(root, 0);
549 	if (!IS_ERR(trans))
550 		btrfs_commit_transaction(trans, root);
551 
552 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
553 
554 	return 0;
555 }
556 
557 static void btrfs_dev_replace_update_device_in_mapping_tree(
558 						struct btrfs_fs_info *fs_info,
559 						struct btrfs_device *srcdev,
560 						struct btrfs_device *tgtdev)
561 {
562 	struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
563 	struct extent_map *em;
564 	struct map_lookup *map;
565 	u64 start = 0;
566 	int i;
567 
568 	write_lock(&em_tree->lock);
569 	do {
570 		em = lookup_extent_mapping(em_tree, start, (u64)-1);
571 		if (!em)
572 			break;
573 		map = (struct map_lookup *)em->bdev;
574 		for (i = 0; i < map->num_stripes; i++)
575 			if (srcdev == map->stripes[i].dev)
576 				map->stripes[i].dev = tgtdev;
577 		start = em->start + em->len;
578 		free_extent_map(em);
579 	} while (start);
580 	write_unlock(&em_tree->lock);
581 }
582 
583 static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid,
584 					 char *srcdev_name,
585 					 struct btrfs_device **device)
586 {
587 	int ret;
588 
589 	if (srcdevid) {
590 		ret = 0;
591 		*device = btrfs_find_device(root->fs_info, srcdevid, NULL,
592 					    NULL);
593 		if (!*device)
594 			ret = -ENOENT;
595 	} else {
596 		ret = btrfs_find_device_missing_or_by_path(root, srcdev_name,
597 							   device);
598 	}
599 	return ret;
600 }
601 
602 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
603 			      struct btrfs_ioctl_dev_replace_args *args)
604 {
605 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
606 
607 	btrfs_dev_replace_lock(dev_replace);
608 	/* even if !dev_replace_is_valid, the values are good enough for
609 	 * the replace_status ioctl */
610 	args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
611 	args->status.replace_state = dev_replace->replace_state;
612 	args->status.time_started = dev_replace->time_started;
613 	args->status.time_stopped = dev_replace->time_stopped;
614 	args->status.num_write_errors =
615 		atomic64_read(&dev_replace->num_write_errors);
616 	args->status.num_uncorrectable_read_errors =
617 		atomic64_read(&dev_replace->num_uncorrectable_read_errors);
618 	switch (dev_replace->replace_state) {
619 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
620 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
621 		args->status.progress_1000 = 0;
622 		break;
623 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
624 		args->status.progress_1000 = 1000;
625 		break;
626 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
627 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
628 		args->status.progress_1000 = div64_u64(dev_replace->cursor_left,
629 			div64_u64(dev_replace->srcdev->total_bytes, 1000));
630 		break;
631 	}
632 	btrfs_dev_replace_unlock(dev_replace);
633 }
634 
635 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info,
636 			     struct btrfs_ioctl_dev_replace_args *args)
637 {
638 	args->result = __btrfs_dev_replace_cancel(fs_info);
639 	return 0;
640 }
641 
642 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
643 {
644 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
645 	struct btrfs_device *tgt_device = NULL;
646 	struct btrfs_trans_handle *trans;
647 	struct btrfs_root *root = fs_info->tree_root;
648 	u64 result;
649 	int ret;
650 
651 	if (fs_info->sb->s_flags & MS_RDONLY)
652 		return -EROFS;
653 
654 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
655 	btrfs_dev_replace_lock(dev_replace);
656 	switch (dev_replace->replace_state) {
657 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
658 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
659 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
660 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
661 		btrfs_dev_replace_unlock(dev_replace);
662 		goto leave;
663 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
664 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
665 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
666 		tgt_device = dev_replace->tgtdev;
667 		dev_replace->tgtdev = NULL;
668 		dev_replace->srcdev = NULL;
669 		break;
670 	}
671 	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
672 	dev_replace->time_stopped = get_seconds();
673 	dev_replace->item_needs_writeback = 1;
674 	btrfs_dev_replace_unlock(dev_replace);
675 	btrfs_scrub_cancel(fs_info);
676 
677 	trans = btrfs_start_transaction(root, 0);
678 	if (IS_ERR(trans)) {
679 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
680 		return PTR_ERR(trans);
681 	}
682 	ret = btrfs_commit_transaction(trans, root);
683 	WARN_ON(ret);
684 	if (tgt_device)
685 		btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
686 
687 leave:
688 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
689 	return result;
690 }
691 
692 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
693 {
694 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
695 
696 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
697 	btrfs_dev_replace_lock(dev_replace);
698 	switch (dev_replace->replace_state) {
699 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
700 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
701 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
702 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
703 		break;
704 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
705 		dev_replace->replace_state =
706 			BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
707 		dev_replace->time_stopped = get_seconds();
708 		dev_replace->item_needs_writeback = 1;
709 		btrfs_info(fs_info, "suspending dev_replace for unmount");
710 		break;
711 	}
712 
713 	btrfs_dev_replace_unlock(dev_replace);
714 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
715 }
716 
717 /* resume dev_replace procedure that was interrupted by unmount */
718 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
719 {
720 	struct task_struct *task;
721 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
722 
723 	btrfs_dev_replace_lock(dev_replace);
724 	switch (dev_replace->replace_state) {
725 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
726 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
727 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
728 		btrfs_dev_replace_unlock(dev_replace);
729 		return 0;
730 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
731 		break;
732 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
733 		dev_replace->replace_state =
734 			BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
735 		break;
736 	}
737 	if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
738 		btrfs_info(fs_info, "cannot continue dev_replace, tgtdev is missing");
739 		btrfs_info(fs_info,
740 			"you may cancel the operation after 'mount -o degraded'");
741 		btrfs_dev_replace_unlock(dev_replace);
742 		return 0;
743 	}
744 	btrfs_dev_replace_unlock(dev_replace);
745 
746 	WARN_ON(atomic_xchg(
747 		&fs_info->mutually_exclusive_operation_running, 1));
748 	task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
749 	return PTR_ERR_OR_ZERO(task);
750 }
751 
752 static int btrfs_dev_replace_kthread(void *data)
753 {
754 	struct btrfs_fs_info *fs_info = data;
755 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
756 	struct btrfs_ioctl_dev_replace_args *status_args;
757 	u64 progress;
758 
759 	status_args = kzalloc(sizeof(*status_args), GFP_NOFS);
760 	if (status_args) {
761 		btrfs_dev_replace_status(fs_info, status_args);
762 		progress = status_args->status.progress_1000;
763 		kfree(status_args);
764 		do_div(progress, 10);
765 		printk_in_rcu(KERN_INFO
766 			"BTRFS: continuing dev_replace from %s (devid %llu) to %s @%u%%\n",
767 			dev_replace->srcdev->missing ? "<missing disk>" :
768 			rcu_str_deref(dev_replace->srcdev->name),
769 			dev_replace->srcdev->devid,
770 			dev_replace->tgtdev ?
771 			rcu_str_deref(dev_replace->tgtdev->name) :
772 			"<missing target disk>",
773 			(unsigned int)progress);
774 	}
775 	btrfs_dev_replace_continue_on_mount(fs_info);
776 	atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
777 
778 	return 0;
779 }
780 
781 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info)
782 {
783 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
784 	int ret;
785 
786 	ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
787 			      dev_replace->committed_cursor_left,
788 			      dev_replace->srcdev->total_bytes,
789 			      &dev_replace->scrub_progress, 0, 1);
790 	ret = btrfs_dev_replace_finishing(fs_info, ret);
791 	WARN_ON(ret);
792 	return 0;
793 }
794 
795 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
796 {
797 	if (!dev_replace->is_valid)
798 		return 0;
799 
800 	switch (dev_replace->replace_state) {
801 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
802 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
803 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
804 		return 0;
805 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
806 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
807 		/*
808 		 * return true even if tgtdev is missing (this is
809 		 * something that can happen if the dev_replace
810 		 * procedure is suspended by an umount and then
811 		 * the tgtdev is missing (or "btrfs dev scan") was
812 		 * not called and the the filesystem is remounted
813 		 * in degraded state. This does not stop the
814 		 * dev_replace procedure. It needs to be canceled
815 		 * manually if the cancelation is wanted.
816 		 */
817 		break;
818 	}
819 	return 1;
820 }
821 
822 void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace)
823 {
824 	/* the beginning is just an optimization for the typical case */
825 	if (atomic_read(&dev_replace->nesting_level) == 0) {
826 acquire_lock:
827 		/* this is not a nested case where the same thread
828 		 * is trying to acqurire the same lock twice */
829 		mutex_lock(&dev_replace->lock);
830 		mutex_lock(&dev_replace->lock_management_lock);
831 		dev_replace->lock_owner = current->pid;
832 		atomic_inc(&dev_replace->nesting_level);
833 		mutex_unlock(&dev_replace->lock_management_lock);
834 		return;
835 	}
836 
837 	mutex_lock(&dev_replace->lock_management_lock);
838 	if (atomic_read(&dev_replace->nesting_level) > 0 &&
839 	    dev_replace->lock_owner == current->pid) {
840 		WARN_ON(!mutex_is_locked(&dev_replace->lock));
841 		atomic_inc(&dev_replace->nesting_level);
842 		mutex_unlock(&dev_replace->lock_management_lock);
843 		return;
844 	}
845 
846 	mutex_unlock(&dev_replace->lock_management_lock);
847 	goto acquire_lock;
848 }
849 
850 void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace)
851 {
852 	WARN_ON(!mutex_is_locked(&dev_replace->lock));
853 	mutex_lock(&dev_replace->lock_management_lock);
854 	WARN_ON(atomic_read(&dev_replace->nesting_level) < 1);
855 	WARN_ON(dev_replace->lock_owner != current->pid);
856 	atomic_dec(&dev_replace->nesting_level);
857 	if (atomic_read(&dev_replace->nesting_level) == 0) {
858 		dev_replace->lock_owner = 0;
859 		mutex_unlock(&dev_replace->lock_management_lock);
860 		mutex_unlock(&dev_replace->lock);
861 	} else {
862 		mutex_unlock(&dev_replace->lock_management_lock);
863 	}
864 }
865