xref: /openbmc/linux/fs/btrfs/dev-replace.c (revision 96de2506)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) STRATO AG 2012.  All rights reserved.
4  */
5 
6 #include <linux/sched.h>
7 #include <linux/bio.h>
8 #include <linux/slab.h>
9 #include <linux/blkdev.h>
10 #include <linux/kthread.h>
11 #include <linux/math64.h>
12 #include "ctree.h"
13 #include "extent_map.h"
14 #include "disk-io.h"
15 #include "transaction.h"
16 #include "print-tree.h"
17 #include "volumes.h"
18 #include "async-thread.h"
19 #include "check-integrity.h"
20 #include "rcu-string.h"
21 #include "dev-replace.h"
22 #include "sysfs.h"
23 
24 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
25 				       int scrub_ret);
26 static void btrfs_dev_replace_update_device_in_mapping_tree(
27 						struct btrfs_fs_info *fs_info,
28 						struct btrfs_device *srcdev,
29 						struct btrfs_device *tgtdev);
30 static int btrfs_dev_replace_kthread(void *data);
31 
32 int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
33 {
34 	struct btrfs_key key;
35 	struct btrfs_root *dev_root = fs_info->dev_root;
36 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
37 	struct extent_buffer *eb;
38 	int slot;
39 	int ret = 0;
40 	struct btrfs_path *path = NULL;
41 	int item_size;
42 	struct btrfs_dev_replace_item *ptr;
43 	u64 src_devid;
44 
45 	path = btrfs_alloc_path();
46 	if (!path) {
47 		ret = -ENOMEM;
48 		goto out;
49 	}
50 
51 	key.objectid = 0;
52 	key.type = BTRFS_DEV_REPLACE_KEY;
53 	key.offset = 0;
54 	ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
55 	if (ret) {
56 no_valid_dev_replace_entry_found:
57 		ret = 0;
58 		dev_replace->replace_state =
59 			BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED;
60 		dev_replace->cont_reading_from_srcdev_mode =
61 		    BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
62 		dev_replace->replace_state = 0;
63 		dev_replace->time_started = 0;
64 		dev_replace->time_stopped = 0;
65 		atomic64_set(&dev_replace->num_write_errors, 0);
66 		atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
67 		dev_replace->cursor_left = 0;
68 		dev_replace->committed_cursor_left = 0;
69 		dev_replace->cursor_left_last_write_of_item = 0;
70 		dev_replace->cursor_right = 0;
71 		dev_replace->srcdev = NULL;
72 		dev_replace->tgtdev = NULL;
73 		dev_replace->is_valid = 0;
74 		dev_replace->item_needs_writeback = 0;
75 		goto out;
76 	}
77 	slot = path->slots[0];
78 	eb = path->nodes[0];
79 	item_size = btrfs_item_size_nr(eb, slot);
80 	ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
81 
82 	if (item_size != sizeof(struct btrfs_dev_replace_item)) {
83 		btrfs_warn(fs_info,
84 			"dev_replace entry found has unexpected size, ignore entry");
85 		goto no_valid_dev_replace_entry_found;
86 	}
87 
88 	src_devid = btrfs_dev_replace_src_devid(eb, ptr);
89 	dev_replace->cont_reading_from_srcdev_mode =
90 		btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
91 	dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
92 	dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
93 	dev_replace->time_stopped =
94 		btrfs_dev_replace_time_stopped(eb, ptr);
95 	atomic64_set(&dev_replace->num_write_errors,
96 		     btrfs_dev_replace_num_write_errors(eb, ptr));
97 	atomic64_set(&dev_replace->num_uncorrectable_read_errors,
98 		     btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
99 	dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
100 	dev_replace->committed_cursor_left = dev_replace->cursor_left;
101 	dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
102 	dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
103 	dev_replace->is_valid = 1;
104 
105 	dev_replace->item_needs_writeback = 0;
106 	switch (dev_replace->replace_state) {
107 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
108 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
109 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
110 		dev_replace->srcdev = NULL;
111 		dev_replace->tgtdev = NULL;
112 		break;
113 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
114 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
115 		dev_replace->srcdev = btrfs_find_device(fs_info, src_devid,
116 							NULL, NULL);
117 		dev_replace->tgtdev = btrfs_find_device(fs_info,
118 							BTRFS_DEV_REPLACE_DEVID,
119 							NULL, NULL);
120 		/*
121 		 * allow 'btrfs dev replace_cancel' if src/tgt device is
122 		 * missing
123 		 */
124 		if (!dev_replace->srcdev &&
125 		    !btrfs_test_opt(fs_info, DEGRADED)) {
126 			ret = -EIO;
127 			btrfs_warn(fs_info,
128 			   "cannot mount because device replace operation is ongoing and");
129 			btrfs_warn(fs_info,
130 			   "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
131 			   src_devid);
132 		}
133 		if (!dev_replace->tgtdev &&
134 		    !btrfs_test_opt(fs_info, DEGRADED)) {
135 			ret = -EIO;
136 			btrfs_warn(fs_info,
137 			   "cannot mount because device replace operation is ongoing and");
138 			btrfs_warn(fs_info,
139 			   "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
140 				BTRFS_DEV_REPLACE_DEVID);
141 		}
142 		if (dev_replace->tgtdev) {
143 			if (dev_replace->srcdev) {
144 				dev_replace->tgtdev->total_bytes =
145 					dev_replace->srcdev->total_bytes;
146 				dev_replace->tgtdev->disk_total_bytes =
147 					dev_replace->srcdev->disk_total_bytes;
148 				dev_replace->tgtdev->commit_total_bytes =
149 					dev_replace->srcdev->commit_total_bytes;
150 				dev_replace->tgtdev->bytes_used =
151 					dev_replace->srcdev->bytes_used;
152 				dev_replace->tgtdev->commit_bytes_used =
153 					dev_replace->srcdev->commit_bytes_used;
154 			}
155 			set_bit(BTRFS_DEV_STATE_REPLACE_TGT,
156 				&dev_replace->tgtdev->dev_state);
157 
158 			WARN_ON(fs_info->fs_devices->rw_devices == 0);
159 			dev_replace->tgtdev->io_width = fs_info->sectorsize;
160 			dev_replace->tgtdev->io_align = fs_info->sectorsize;
161 			dev_replace->tgtdev->sector_size = fs_info->sectorsize;
162 			dev_replace->tgtdev->fs_info = fs_info;
163 			set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
164 				&dev_replace->tgtdev->dev_state);
165 		}
166 		break;
167 	}
168 
169 out:
170 	btrfs_free_path(path);
171 	return ret;
172 }
173 
174 /*
175  * Initialize a new device for device replace target from a given source dev
176  * and path.
177  *
178  * Return 0 and new device in @device_out, otherwise return < 0
179  */
180 static int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
181 				  const char *device_path,
182 				  struct btrfs_device *srcdev,
183 				  struct btrfs_device **device_out)
184 {
185 	struct btrfs_device *device;
186 	struct block_device *bdev;
187 	struct list_head *devices;
188 	struct rcu_string *name;
189 	u64 devid = BTRFS_DEV_REPLACE_DEVID;
190 	int ret = 0;
191 
192 	*device_out = NULL;
193 	if (fs_info->fs_devices->seeding) {
194 		btrfs_err(fs_info, "the filesystem is a seed filesystem!");
195 		return -EINVAL;
196 	}
197 
198 	bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
199 				  fs_info->bdev_holder);
200 	if (IS_ERR(bdev)) {
201 		btrfs_err(fs_info, "target device %s is invalid!", device_path);
202 		return PTR_ERR(bdev);
203 	}
204 
205 	filemap_write_and_wait(bdev->bd_inode->i_mapping);
206 
207 	devices = &fs_info->fs_devices->devices;
208 	list_for_each_entry(device, devices, dev_list) {
209 		if (device->bdev == bdev) {
210 			btrfs_err(fs_info,
211 				  "target device is in the filesystem!");
212 			ret = -EEXIST;
213 			goto error;
214 		}
215 	}
216 
217 
218 	if (i_size_read(bdev->bd_inode) <
219 	    btrfs_device_get_total_bytes(srcdev)) {
220 		btrfs_err(fs_info,
221 			  "target device is smaller than source device!");
222 		ret = -EINVAL;
223 		goto error;
224 	}
225 
226 
227 	device = btrfs_alloc_device(NULL, &devid, NULL);
228 	if (IS_ERR(device)) {
229 		ret = PTR_ERR(device);
230 		goto error;
231 	}
232 
233 	name = rcu_string_strdup(device_path, GFP_KERNEL);
234 	if (!name) {
235 		btrfs_free_device(device);
236 		ret = -ENOMEM;
237 		goto error;
238 	}
239 	rcu_assign_pointer(device->name, name);
240 
241 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
242 	set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
243 	device->generation = 0;
244 	device->io_width = fs_info->sectorsize;
245 	device->io_align = fs_info->sectorsize;
246 	device->sector_size = fs_info->sectorsize;
247 	device->total_bytes = btrfs_device_get_total_bytes(srcdev);
248 	device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
249 	device->bytes_used = btrfs_device_get_bytes_used(srcdev);
250 	device->commit_total_bytes = srcdev->commit_total_bytes;
251 	device->commit_bytes_used = device->bytes_used;
252 	device->fs_info = fs_info;
253 	device->bdev = bdev;
254 	set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
255 	set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
256 	device->mode = FMODE_EXCL;
257 	device->dev_stats_valid = 1;
258 	set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
259 	device->fs_devices = fs_info->fs_devices;
260 	list_add(&device->dev_list, &fs_info->fs_devices->devices);
261 	fs_info->fs_devices->num_devices++;
262 	fs_info->fs_devices->open_devices++;
263 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
264 
265 	*device_out = device;
266 	return 0;
267 
268 error:
269 	blkdev_put(bdev, FMODE_EXCL);
270 	return ret;
271 }
272 
273 /*
274  * called from commit_transaction. Writes changed device replace state to
275  * disk.
276  */
277 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
278 			  struct btrfs_fs_info *fs_info)
279 {
280 	int ret;
281 	struct btrfs_root *dev_root = fs_info->dev_root;
282 	struct btrfs_path *path;
283 	struct btrfs_key key;
284 	struct extent_buffer *eb;
285 	struct btrfs_dev_replace_item *ptr;
286 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
287 
288 	btrfs_dev_replace_read_lock(dev_replace);
289 	if (!dev_replace->is_valid ||
290 	    !dev_replace->item_needs_writeback) {
291 		btrfs_dev_replace_read_unlock(dev_replace);
292 		return 0;
293 	}
294 	btrfs_dev_replace_read_unlock(dev_replace);
295 
296 	key.objectid = 0;
297 	key.type = BTRFS_DEV_REPLACE_KEY;
298 	key.offset = 0;
299 
300 	path = btrfs_alloc_path();
301 	if (!path) {
302 		ret = -ENOMEM;
303 		goto out;
304 	}
305 	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
306 	if (ret < 0) {
307 		btrfs_warn(fs_info,
308 			   "error %d while searching for dev_replace item!",
309 			   ret);
310 		goto out;
311 	}
312 
313 	if (ret == 0 &&
314 	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
315 		/*
316 		 * need to delete old one and insert a new one.
317 		 * Since no attempt is made to recover any old state, if the
318 		 * dev_replace state is 'running', the data on the target
319 		 * drive is lost.
320 		 * It would be possible to recover the state: just make sure
321 		 * that the beginning of the item is never changed and always
322 		 * contains all the essential information. Then read this
323 		 * minimal set of information and use it as a base for the
324 		 * new state.
325 		 */
326 		ret = btrfs_del_item(trans, dev_root, path);
327 		if (ret != 0) {
328 			btrfs_warn(fs_info,
329 				   "delete too small dev_replace item failed %d!",
330 				   ret);
331 			goto out;
332 		}
333 		ret = 1;
334 	}
335 
336 	if (ret == 1) {
337 		/* need to insert a new item */
338 		btrfs_release_path(path);
339 		ret = btrfs_insert_empty_item(trans, dev_root, path,
340 					      &key, sizeof(*ptr));
341 		if (ret < 0) {
342 			btrfs_warn(fs_info,
343 				   "insert dev_replace item failed %d!", ret);
344 			goto out;
345 		}
346 	}
347 
348 	eb = path->nodes[0];
349 	ptr = btrfs_item_ptr(eb, path->slots[0],
350 			     struct btrfs_dev_replace_item);
351 
352 	btrfs_dev_replace_write_lock(dev_replace);
353 	if (dev_replace->srcdev)
354 		btrfs_set_dev_replace_src_devid(eb, ptr,
355 			dev_replace->srcdev->devid);
356 	else
357 		btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
358 	btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
359 		dev_replace->cont_reading_from_srcdev_mode);
360 	btrfs_set_dev_replace_replace_state(eb, ptr,
361 		dev_replace->replace_state);
362 	btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
363 	btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
364 	btrfs_set_dev_replace_num_write_errors(eb, ptr,
365 		atomic64_read(&dev_replace->num_write_errors));
366 	btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
367 		atomic64_read(&dev_replace->num_uncorrectable_read_errors));
368 	dev_replace->cursor_left_last_write_of_item =
369 		dev_replace->cursor_left;
370 	btrfs_set_dev_replace_cursor_left(eb, ptr,
371 		dev_replace->cursor_left_last_write_of_item);
372 	btrfs_set_dev_replace_cursor_right(eb, ptr,
373 		dev_replace->cursor_right);
374 	dev_replace->item_needs_writeback = 0;
375 	btrfs_dev_replace_write_unlock(dev_replace);
376 
377 	btrfs_mark_buffer_dirty(eb);
378 
379 out:
380 	btrfs_free_path(path);
381 
382 	return ret;
383 }
384 
385 void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
386 {
387 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
388 
389 	dev_replace->committed_cursor_left =
390 		dev_replace->cursor_left_last_write_of_item;
391 }
392 
393 static char* btrfs_dev_name(struct btrfs_device *device)
394 {
395 	if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
396 		return "<missing disk>";
397 	else
398 		return rcu_str_deref(device->name);
399 }
400 
401 int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
402 		const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
403 		int read_src)
404 {
405 	struct btrfs_root *root = fs_info->dev_root;
406 	struct btrfs_trans_handle *trans;
407 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
408 	int ret;
409 	struct btrfs_device *tgt_device = NULL;
410 	struct btrfs_device *src_device = NULL;
411 
412 	ret = btrfs_find_device_by_devspec(fs_info, srcdevid,
413 					    srcdev_name, &src_device);
414 	if (ret)
415 		return ret;
416 
417 	ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
418 					    src_device, &tgt_device);
419 	if (ret)
420 		return ret;
421 
422 	/*
423 	 * Here we commit the transaction to make sure commit_total_bytes
424 	 * of all the devices are updated.
425 	 */
426 	trans = btrfs_attach_transaction(root);
427 	if (!IS_ERR(trans)) {
428 		ret = btrfs_commit_transaction(trans);
429 		if (ret)
430 			return ret;
431 	} else if (PTR_ERR(trans) != -ENOENT) {
432 		return PTR_ERR(trans);
433 	}
434 
435 	btrfs_dev_replace_write_lock(dev_replace);
436 	switch (dev_replace->replace_state) {
437 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
438 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
439 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
440 		break;
441 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
442 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
443 		ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
444 		goto leave;
445 	}
446 
447 	dev_replace->cont_reading_from_srcdev_mode = read_src;
448 	WARN_ON(!src_device);
449 	dev_replace->srcdev = src_device;
450 	dev_replace->tgtdev = tgt_device;
451 
452 	btrfs_info_in_rcu(fs_info,
453 		      "dev_replace from %s (devid %llu) to %s started",
454 		      btrfs_dev_name(src_device),
455 		      src_device->devid,
456 		      rcu_str_deref(tgt_device->name));
457 
458 	/*
459 	 * from now on, the writes to the srcdev are all duplicated to
460 	 * go to the tgtdev as well (refer to btrfs_map_block()).
461 	 */
462 	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
463 	dev_replace->time_started = ktime_get_real_seconds();
464 	dev_replace->cursor_left = 0;
465 	dev_replace->committed_cursor_left = 0;
466 	dev_replace->cursor_left_last_write_of_item = 0;
467 	dev_replace->cursor_right = 0;
468 	dev_replace->is_valid = 1;
469 	dev_replace->item_needs_writeback = 1;
470 	atomic64_set(&dev_replace->num_write_errors, 0);
471 	atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
472 	btrfs_dev_replace_write_unlock(dev_replace);
473 
474 	ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device);
475 	if (ret)
476 		btrfs_err(fs_info, "kobj add dev failed %d", ret);
477 
478 	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
479 
480 	/* force writing the updated state information to disk */
481 	trans = btrfs_start_transaction(root, 0);
482 	if (IS_ERR(trans)) {
483 		ret = PTR_ERR(trans);
484 		btrfs_dev_replace_write_lock(dev_replace);
485 		goto leave;
486 	}
487 
488 	ret = btrfs_commit_transaction(trans);
489 	WARN_ON(ret);
490 
491 	/* the disk copy procedure reuses the scrub code */
492 	ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
493 			      btrfs_device_get_total_bytes(src_device),
494 			      &dev_replace->scrub_progress, 0, 1);
495 
496 	ret = btrfs_dev_replace_finishing(fs_info, ret);
497 	if (ret == -EINPROGRESS) {
498 		ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
499 	} else {
500 		WARN_ON(ret);
501 	}
502 
503 	return ret;
504 
505 leave:
506 	dev_replace->srcdev = NULL;
507 	dev_replace->tgtdev = NULL;
508 	btrfs_dev_replace_write_unlock(dev_replace);
509 	btrfs_destroy_dev_replace_tgtdev(tgt_device);
510 	return ret;
511 }
512 
513 int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
514 			    struct btrfs_ioctl_dev_replace_args *args)
515 {
516 	int ret;
517 
518 	switch (args->start.cont_reading_from_srcdev_mode) {
519 	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
520 	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
521 		break;
522 	default:
523 		return -EINVAL;
524 	}
525 
526 	if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
527 	    args->start.tgtdev_name[0] == '\0')
528 		return -EINVAL;
529 
530 	ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name,
531 					args->start.srcdevid,
532 					args->start.srcdev_name,
533 					args->start.cont_reading_from_srcdev_mode);
534 	args->result = ret;
535 	/* don't warn if EINPROGRESS, someone else might be running scrub */
536 	if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS)
537 		ret = 0;
538 
539 	return ret;
540 }
541 
542 /*
543  * blocked until all in-flight bios operations are finished.
544  */
545 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
546 {
547 	set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
548 	wait_event(fs_info->replace_wait, !percpu_counter_sum(
549 		   &fs_info->bio_counter));
550 }
551 
552 /*
553  * we have removed target device, it is safe to allow new bios request.
554  */
555 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
556 {
557 	clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
558 	wake_up(&fs_info->replace_wait);
559 }
560 
561 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
562 				       int scrub_ret)
563 {
564 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
565 	struct btrfs_device *tgt_device;
566 	struct btrfs_device *src_device;
567 	struct btrfs_root *root = fs_info->tree_root;
568 	u8 uuid_tmp[BTRFS_UUID_SIZE];
569 	struct btrfs_trans_handle *trans;
570 	int ret = 0;
571 
572 	/* don't allow cancel or unmount to disturb the finishing procedure */
573 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
574 
575 	btrfs_dev_replace_read_lock(dev_replace);
576 	/* was the operation canceled, or is it finished? */
577 	if (dev_replace->replace_state !=
578 	    BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
579 		btrfs_dev_replace_read_unlock(dev_replace);
580 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
581 		return 0;
582 	}
583 
584 	tgt_device = dev_replace->tgtdev;
585 	src_device = dev_replace->srcdev;
586 	btrfs_dev_replace_read_unlock(dev_replace);
587 
588 	/*
589 	 * flush all outstanding I/O and inode extent mappings before the
590 	 * copy operation is declared as being finished
591 	 */
592 	ret = btrfs_start_delalloc_roots(fs_info, -1);
593 	if (ret) {
594 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
595 		return ret;
596 	}
597 	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
598 
599 	trans = btrfs_start_transaction(root, 0);
600 	if (IS_ERR(trans)) {
601 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
602 		return PTR_ERR(trans);
603 	}
604 	ret = btrfs_commit_transaction(trans);
605 	WARN_ON(ret);
606 
607 	/* keep away write_all_supers() during the finishing procedure */
608 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
609 	mutex_lock(&fs_info->chunk_mutex);
610 	btrfs_dev_replace_write_lock(dev_replace);
611 	dev_replace->replace_state =
612 		scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
613 			  : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
614 	dev_replace->tgtdev = NULL;
615 	dev_replace->srcdev = NULL;
616 	dev_replace->time_stopped = ktime_get_real_seconds();
617 	dev_replace->item_needs_writeback = 1;
618 
619 	/* replace old device with new one in mapping tree */
620 	if (!scrub_ret) {
621 		btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
622 								src_device,
623 								tgt_device);
624 	} else {
625 		btrfs_err_in_rcu(fs_info,
626 				 "btrfs_scrub_dev(%s, %llu, %s) failed %d",
627 				 btrfs_dev_name(src_device),
628 				 src_device->devid,
629 				 rcu_str_deref(tgt_device->name), scrub_ret);
630 		btrfs_dev_replace_write_unlock(dev_replace);
631 		mutex_unlock(&fs_info->chunk_mutex);
632 		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
633 		btrfs_rm_dev_replace_blocked(fs_info);
634 		if (tgt_device)
635 			btrfs_destroy_dev_replace_tgtdev(tgt_device);
636 		btrfs_rm_dev_replace_unblocked(fs_info);
637 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
638 
639 		return scrub_ret;
640 	}
641 
642 	btrfs_info_in_rcu(fs_info,
643 			  "dev_replace from %s (devid %llu) to %s finished",
644 			  btrfs_dev_name(src_device),
645 			  src_device->devid,
646 			  rcu_str_deref(tgt_device->name));
647 	clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state);
648 	tgt_device->devid = src_device->devid;
649 	src_device->devid = BTRFS_DEV_REPLACE_DEVID;
650 	memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
651 	memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
652 	memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
653 	btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
654 	btrfs_device_set_disk_total_bytes(tgt_device,
655 					  src_device->disk_total_bytes);
656 	btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
657 	ASSERT(list_empty(&src_device->resized_list));
658 	tgt_device->commit_total_bytes = src_device->commit_total_bytes;
659 	tgt_device->commit_bytes_used = src_device->bytes_used;
660 
661 	btrfs_assign_next_active_device(src_device, tgt_device);
662 
663 	list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
664 	fs_info->fs_devices->rw_devices++;
665 
666 	btrfs_dev_replace_write_unlock(dev_replace);
667 
668 	btrfs_rm_dev_replace_blocked(fs_info);
669 
670 	btrfs_rm_dev_replace_remove_srcdev(src_device);
671 
672 	btrfs_rm_dev_replace_unblocked(fs_info);
673 
674 	/*
675 	 * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will
676 	 * update on-disk dev stats value during commit transaction
677 	 */
678 	atomic_inc(&tgt_device->dev_stats_ccnt);
679 
680 	/*
681 	 * this is again a consistent state where no dev_replace procedure
682 	 * is running, the target device is part of the filesystem, the
683 	 * source device is not part of the filesystem anymore and its 1st
684 	 * superblock is scratched out so that it is no longer marked to
685 	 * belong to this filesystem.
686 	 */
687 	mutex_unlock(&fs_info->chunk_mutex);
688 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
689 
690 	/* replace the sysfs entry */
691 	btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device);
692 	btrfs_rm_dev_replace_free_srcdev(fs_info, src_device);
693 
694 	/* write back the superblocks */
695 	trans = btrfs_start_transaction(root, 0);
696 	if (!IS_ERR(trans))
697 		btrfs_commit_transaction(trans);
698 
699 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
700 
701 	return 0;
702 }
703 
704 static void btrfs_dev_replace_update_device_in_mapping_tree(
705 						struct btrfs_fs_info *fs_info,
706 						struct btrfs_device *srcdev,
707 						struct btrfs_device *tgtdev)
708 {
709 	struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
710 	struct extent_map *em;
711 	struct map_lookup *map;
712 	u64 start = 0;
713 	int i;
714 
715 	write_lock(&em_tree->lock);
716 	do {
717 		em = lookup_extent_mapping(em_tree, start, (u64)-1);
718 		if (!em)
719 			break;
720 		map = em->map_lookup;
721 		for (i = 0; i < map->num_stripes; i++)
722 			if (srcdev == map->stripes[i].dev)
723 				map->stripes[i].dev = tgtdev;
724 		start = em->start + em->len;
725 		free_extent_map(em);
726 	} while (start);
727 	write_unlock(&em_tree->lock);
728 }
729 
730 /*
731  * Read progress of device replace status according to the state and last
732  * stored position. The value format is the same as for
733  * btrfs_dev_replace::progress_1000
734  */
735 static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info)
736 {
737 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
738 	u64 ret = 0;
739 
740 	switch (dev_replace->replace_state) {
741 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
742 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
743 		ret = 0;
744 		break;
745 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
746 		ret = 1000;
747 		break;
748 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
749 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
750 		ret = div64_u64(dev_replace->cursor_left,
751 				div_u64(btrfs_device_get_total_bytes(
752 						dev_replace->srcdev), 1000));
753 		break;
754 	}
755 
756 	return ret;
757 }
758 
759 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
760 			      struct btrfs_ioctl_dev_replace_args *args)
761 {
762 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
763 
764 	btrfs_dev_replace_read_lock(dev_replace);
765 	/* even if !dev_replace_is_valid, the values are good enough for
766 	 * the replace_status ioctl */
767 	args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
768 	args->status.replace_state = dev_replace->replace_state;
769 	args->status.time_started = dev_replace->time_started;
770 	args->status.time_stopped = dev_replace->time_stopped;
771 	args->status.num_write_errors =
772 		atomic64_read(&dev_replace->num_write_errors);
773 	args->status.num_uncorrectable_read_errors =
774 		atomic64_read(&dev_replace->num_uncorrectable_read_errors);
775 	args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
776 	btrfs_dev_replace_read_unlock(dev_replace);
777 }
778 
779 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
780 {
781 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
782 	struct btrfs_device *tgt_device = NULL;
783 	struct btrfs_device *src_device = NULL;
784 	struct btrfs_trans_handle *trans;
785 	struct btrfs_root *root = fs_info->tree_root;
786 	int result;
787 	int ret;
788 
789 	if (sb_rdonly(fs_info->sb))
790 		return -EROFS;
791 
792 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
793 	btrfs_dev_replace_write_lock(dev_replace);
794 	switch (dev_replace->replace_state) {
795 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
796 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
797 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
798 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
799 		btrfs_dev_replace_write_unlock(dev_replace);
800 		goto leave;
801 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
802 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
803 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
804 		tgt_device = dev_replace->tgtdev;
805 		src_device = dev_replace->srcdev;
806 		dev_replace->tgtdev = NULL;
807 		dev_replace->srcdev = NULL;
808 		break;
809 	}
810 	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
811 	dev_replace->time_stopped = ktime_get_real_seconds();
812 	dev_replace->item_needs_writeback = 1;
813 	btrfs_dev_replace_write_unlock(dev_replace);
814 	btrfs_scrub_cancel(fs_info);
815 
816 	trans = btrfs_start_transaction(root, 0);
817 	if (IS_ERR(trans)) {
818 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
819 		return PTR_ERR(trans);
820 	}
821 	ret = btrfs_commit_transaction(trans);
822 	WARN_ON(ret);
823 
824 	btrfs_info_in_rcu(fs_info,
825 		"dev_replace from %s (devid %llu) to %s canceled",
826 		btrfs_dev_name(src_device), src_device->devid,
827 		btrfs_dev_name(tgt_device));
828 
829 	if (tgt_device)
830 		btrfs_destroy_dev_replace_tgtdev(tgt_device);
831 
832 leave:
833 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
834 	return result;
835 }
836 
837 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
838 {
839 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
840 
841 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
842 	btrfs_dev_replace_write_lock(dev_replace);
843 	switch (dev_replace->replace_state) {
844 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
845 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
846 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
847 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
848 		break;
849 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
850 		dev_replace->replace_state =
851 			BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
852 		dev_replace->time_stopped = ktime_get_real_seconds();
853 		dev_replace->item_needs_writeback = 1;
854 		btrfs_info(fs_info, "suspending dev_replace for unmount");
855 		break;
856 	}
857 
858 	btrfs_dev_replace_write_unlock(dev_replace);
859 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
860 }
861 
862 /* resume dev_replace procedure that was interrupted by unmount */
863 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
864 {
865 	struct task_struct *task;
866 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
867 
868 	btrfs_dev_replace_write_lock(dev_replace);
869 	switch (dev_replace->replace_state) {
870 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
871 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
872 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
873 		btrfs_dev_replace_write_unlock(dev_replace);
874 		return 0;
875 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
876 		break;
877 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
878 		dev_replace->replace_state =
879 			BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
880 		break;
881 	}
882 	if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
883 		btrfs_info(fs_info,
884 			   "cannot continue dev_replace, tgtdev is missing");
885 		btrfs_info(fs_info,
886 			   "you may cancel the operation after 'mount -o degraded'");
887 		btrfs_dev_replace_write_unlock(dev_replace);
888 		return 0;
889 	}
890 	btrfs_dev_replace_write_unlock(dev_replace);
891 
892 	/*
893 	 * This could collide with a paused balance, but the exclusive op logic
894 	 * should never allow both to start and pause. We don't want to allow
895 	 * dev-replace to start anyway.
896 	 */
897 	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
898 		btrfs_info(fs_info,
899 		"cannot resume dev-replace, other exclusive operation running");
900 		return 0;
901 	}
902 
903 	task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
904 	return PTR_ERR_OR_ZERO(task);
905 }
906 
907 static int btrfs_dev_replace_kthread(void *data)
908 {
909 	struct btrfs_fs_info *fs_info = data;
910 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
911 	u64 progress;
912 	int ret;
913 
914 	progress = btrfs_dev_replace_progress(fs_info);
915 	progress = div_u64(progress, 10);
916 	btrfs_info_in_rcu(fs_info,
917 		"continuing dev_replace from %s (devid %llu) to target %s @%u%%",
918 		btrfs_dev_name(dev_replace->srcdev),
919 		dev_replace->srcdev->devid,
920 		btrfs_dev_name(dev_replace->tgtdev),
921 		(unsigned int)progress);
922 
923 	ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
924 			      dev_replace->committed_cursor_left,
925 			      btrfs_device_get_total_bytes(dev_replace->srcdev),
926 			      &dev_replace->scrub_progress, 0, 1);
927 	ret = btrfs_dev_replace_finishing(fs_info, ret);
928 	WARN_ON(ret);
929 
930 	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
931 	return 0;
932 }
933 
934 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
935 {
936 	if (!dev_replace->is_valid)
937 		return 0;
938 
939 	switch (dev_replace->replace_state) {
940 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
941 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
942 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
943 		return 0;
944 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
945 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
946 		/*
947 		 * return true even if tgtdev is missing (this is
948 		 * something that can happen if the dev_replace
949 		 * procedure is suspended by an umount and then
950 		 * the tgtdev is missing (or "btrfs dev scan") was
951 		 * not called and the the filesystem is remounted
952 		 * in degraded state. This does not stop the
953 		 * dev_replace procedure. It needs to be canceled
954 		 * manually if the cancellation is wanted.
955 		 */
956 		break;
957 	}
958 	return 1;
959 }
960 
961 void btrfs_dev_replace_read_lock(struct btrfs_dev_replace *dev_replace)
962 {
963 	read_lock(&dev_replace->lock);
964 	atomic_inc(&dev_replace->read_locks);
965 }
966 
967 void btrfs_dev_replace_read_unlock(struct btrfs_dev_replace *dev_replace)
968 {
969 	ASSERT(atomic_read(&dev_replace->read_locks) > 0);
970 	atomic_dec(&dev_replace->read_locks);
971 	read_unlock(&dev_replace->lock);
972 }
973 
974 void btrfs_dev_replace_write_lock(struct btrfs_dev_replace *dev_replace)
975 {
976 again:
977 	wait_event(dev_replace->read_lock_wq,
978 		   atomic_read(&dev_replace->blocking_readers) == 0);
979 	write_lock(&dev_replace->lock);
980 	if (atomic_read(&dev_replace->blocking_readers)) {
981 		write_unlock(&dev_replace->lock);
982 		goto again;
983 	}
984 }
985 
986 void btrfs_dev_replace_write_unlock(struct btrfs_dev_replace *dev_replace)
987 {
988 	ASSERT(atomic_read(&dev_replace->blocking_readers) == 0);
989 	write_unlock(&dev_replace->lock);
990 }
991 
992 /* inc blocking cnt and release read lock */
993 void btrfs_dev_replace_set_lock_blocking(
994 					struct btrfs_dev_replace *dev_replace)
995 {
996 	/* only set blocking for read lock */
997 	ASSERT(atomic_read(&dev_replace->read_locks) > 0);
998 	atomic_inc(&dev_replace->blocking_readers);
999 	read_unlock(&dev_replace->lock);
1000 }
1001 
1002 /* acquire read lock and dec blocking cnt */
1003 void btrfs_dev_replace_clear_lock_blocking(
1004 					struct btrfs_dev_replace *dev_replace)
1005 {
1006 	/* only set blocking for read lock */
1007 	ASSERT(atomic_read(&dev_replace->read_locks) > 0);
1008 	ASSERT(atomic_read(&dev_replace->blocking_readers) > 0);
1009 	read_lock(&dev_replace->lock);
1010 	/* Barrier implied by atomic_dec_and_test */
1011 	if (atomic_dec_and_test(&dev_replace->blocking_readers))
1012 		cond_wake_up_nomb(&dev_replace->read_lock_wq);
1013 }
1014 
1015 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
1016 {
1017 	percpu_counter_inc(&fs_info->bio_counter);
1018 }
1019 
1020 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
1021 {
1022 	percpu_counter_sub(&fs_info->bio_counter, amount);
1023 	cond_wake_up_nomb(&fs_info->replace_wait);
1024 }
1025 
1026 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
1027 {
1028 	while (1) {
1029 		percpu_counter_inc(&fs_info->bio_counter);
1030 		if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1031 				     &fs_info->fs_state)))
1032 			break;
1033 
1034 		btrfs_bio_counter_dec(fs_info);
1035 		wait_event(fs_info->replace_wait,
1036 			   !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1037 				     &fs_info->fs_state));
1038 	}
1039 }
1040