xref: /openbmc/linux/fs/btrfs/dev-replace.c (revision 8bd1369b)
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/buffer_head.h>
10 #include <linux/blkdev.h>
11 #include <linux/random.h>
12 #include <linux/iocontext.h>
13 #include <linux/capability.h>
14 #include <linux/kthread.h>
15 #include <linux/math64.h>
16 #include <asm/div64.h>
17 #include "ctree.h"
18 #include "extent_map.h"
19 #include "disk-io.h"
20 #include "transaction.h"
21 #include "print-tree.h"
22 #include "volumes.h"
23 #include "async-thread.h"
24 #include "check-integrity.h"
25 #include "rcu-string.h"
26 #include "dev-replace.h"
27 #include "sysfs.h"
28 
29 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
30 				       int scrub_ret);
31 static void btrfs_dev_replace_update_device_in_mapping_tree(
32 						struct btrfs_fs_info *fs_info,
33 						struct btrfs_device *srcdev,
34 						struct btrfs_device *tgtdev);
35 static int btrfs_dev_replace_kthread(void *data);
36 
37 int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
38 {
39 	struct btrfs_key key;
40 	struct btrfs_root *dev_root = fs_info->dev_root;
41 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
42 	struct extent_buffer *eb;
43 	int slot;
44 	int ret = 0;
45 	struct btrfs_path *path = NULL;
46 	int item_size;
47 	struct btrfs_dev_replace_item *ptr;
48 	u64 src_devid;
49 
50 	path = btrfs_alloc_path();
51 	if (!path) {
52 		ret = -ENOMEM;
53 		goto out;
54 	}
55 
56 	key.objectid = 0;
57 	key.type = BTRFS_DEV_REPLACE_KEY;
58 	key.offset = 0;
59 	ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
60 	if (ret) {
61 no_valid_dev_replace_entry_found:
62 		ret = 0;
63 		dev_replace->replace_state =
64 			BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED;
65 		dev_replace->cont_reading_from_srcdev_mode =
66 		    BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
67 		dev_replace->replace_state = 0;
68 		dev_replace->time_started = 0;
69 		dev_replace->time_stopped = 0;
70 		atomic64_set(&dev_replace->num_write_errors, 0);
71 		atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
72 		dev_replace->cursor_left = 0;
73 		dev_replace->committed_cursor_left = 0;
74 		dev_replace->cursor_left_last_write_of_item = 0;
75 		dev_replace->cursor_right = 0;
76 		dev_replace->srcdev = NULL;
77 		dev_replace->tgtdev = NULL;
78 		dev_replace->is_valid = 0;
79 		dev_replace->item_needs_writeback = 0;
80 		goto out;
81 	}
82 	slot = path->slots[0];
83 	eb = path->nodes[0];
84 	item_size = btrfs_item_size_nr(eb, slot);
85 	ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
86 
87 	if (item_size != sizeof(struct btrfs_dev_replace_item)) {
88 		btrfs_warn(fs_info,
89 			"dev_replace entry found has unexpected size, ignore entry");
90 		goto no_valid_dev_replace_entry_found;
91 	}
92 
93 	src_devid = btrfs_dev_replace_src_devid(eb, ptr);
94 	dev_replace->cont_reading_from_srcdev_mode =
95 		btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
96 	dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
97 	dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
98 	dev_replace->time_stopped =
99 		btrfs_dev_replace_time_stopped(eb, ptr);
100 	atomic64_set(&dev_replace->num_write_errors,
101 		     btrfs_dev_replace_num_write_errors(eb, ptr));
102 	atomic64_set(&dev_replace->num_uncorrectable_read_errors,
103 		     btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
104 	dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
105 	dev_replace->committed_cursor_left = dev_replace->cursor_left;
106 	dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
107 	dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
108 	dev_replace->is_valid = 1;
109 
110 	dev_replace->item_needs_writeback = 0;
111 	switch (dev_replace->replace_state) {
112 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
113 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
114 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
115 		dev_replace->srcdev = NULL;
116 		dev_replace->tgtdev = NULL;
117 		break;
118 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
119 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
120 		dev_replace->srcdev = btrfs_find_device(fs_info, src_devid,
121 							NULL, NULL);
122 		dev_replace->tgtdev = btrfs_find_device(fs_info,
123 							BTRFS_DEV_REPLACE_DEVID,
124 							NULL, NULL);
125 		/*
126 		 * allow 'btrfs dev replace_cancel' if src/tgt device is
127 		 * missing
128 		 */
129 		if (!dev_replace->srcdev &&
130 		    !btrfs_test_opt(fs_info, DEGRADED)) {
131 			ret = -EIO;
132 			btrfs_warn(fs_info,
133 			   "cannot mount because device replace operation is ongoing and");
134 			btrfs_warn(fs_info,
135 			   "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
136 			   src_devid);
137 		}
138 		if (!dev_replace->tgtdev &&
139 		    !btrfs_test_opt(fs_info, DEGRADED)) {
140 			ret = -EIO;
141 			btrfs_warn(fs_info,
142 			   "cannot mount because device replace operation is ongoing and");
143 			btrfs_warn(fs_info,
144 			   "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
145 				BTRFS_DEV_REPLACE_DEVID);
146 		}
147 		if (dev_replace->tgtdev) {
148 			if (dev_replace->srcdev) {
149 				dev_replace->tgtdev->total_bytes =
150 					dev_replace->srcdev->total_bytes;
151 				dev_replace->tgtdev->disk_total_bytes =
152 					dev_replace->srcdev->disk_total_bytes;
153 				dev_replace->tgtdev->commit_total_bytes =
154 					dev_replace->srcdev->commit_total_bytes;
155 				dev_replace->tgtdev->bytes_used =
156 					dev_replace->srcdev->bytes_used;
157 				dev_replace->tgtdev->commit_bytes_used =
158 					dev_replace->srcdev->commit_bytes_used;
159 			}
160 			set_bit(BTRFS_DEV_STATE_REPLACE_TGT,
161 				&dev_replace->tgtdev->dev_state);
162 
163 			WARN_ON(fs_info->fs_devices->rw_devices == 0);
164 			dev_replace->tgtdev->io_width = fs_info->sectorsize;
165 			dev_replace->tgtdev->io_align = fs_info->sectorsize;
166 			dev_replace->tgtdev->sector_size = fs_info->sectorsize;
167 			dev_replace->tgtdev->fs_info = fs_info;
168 			set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
169 				&dev_replace->tgtdev->dev_state);
170 		}
171 		break;
172 	}
173 
174 out:
175 	btrfs_free_path(path);
176 	return ret;
177 }
178 
179 /*
180  * Initialize a new device for device replace target from a given source dev
181  * and path.
182  *
183  * Return 0 and new device in @device_out, otherwise return < 0
184  */
185 static int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
186 				  const char *device_path,
187 				  struct btrfs_device *srcdev,
188 				  struct btrfs_device **device_out)
189 {
190 	struct btrfs_device *device;
191 	struct block_device *bdev;
192 	struct list_head *devices;
193 	struct rcu_string *name;
194 	u64 devid = BTRFS_DEV_REPLACE_DEVID;
195 	int ret = 0;
196 
197 	*device_out = NULL;
198 	if (fs_info->fs_devices->seeding) {
199 		btrfs_err(fs_info, "the filesystem is a seed filesystem!");
200 		return -EINVAL;
201 	}
202 
203 	bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
204 				  fs_info->bdev_holder);
205 	if (IS_ERR(bdev)) {
206 		btrfs_err(fs_info, "target device %s is invalid!", device_path);
207 		return PTR_ERR(bdev);
208 	}
209 
210 	filemap_write_and_wait(bdev->bd_inode->i_mapping);
211 
212 	devices = &fs_info->fs_devices->devices;
213 	list_for_each_entry(device, devices, dev_list) {
214 		if (device->bdev == bdev) {
215 			btrfs_err(fs_info,
216 				  "target device is in the filesystem!");
217 			ret = -EEXIST;
218 			goto error;
219 		}
220 	}
221 
222 
223 	if (i_size_read(bdev->bd_inode) <
224 	    btrfs_device_get_total_bytes(srcdev)) {
225 		btrfs_err(fs_info,
226 			  "target device is smaller than source device!");
227 		ret = -EINVAL;
228 		goto error;
229 	}
230 
231 
232 	device = btrfs_alloc_device(NULL, &devid, NULL);
233 	if (IS_ERR(device)) {
234 		ret = PTR_ERR(device);
235 		goto error;
236 	}
237 
238 	name = rcu_string_strdup(device_path, GFP_KERNEL);
239 	if (!name) {
240 		btrfs_free_device(device);
241 		ret = -ENOMEM;
242 		goto error;
243 	}
244 	rcu_assign_pointer(device->name, name);
245 
246 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
247 	set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
248 	device->generation = 0;
249 	device->io_width = fs_info->sectorsize;
250 	device->io_align = fs_info->sectorsize;
251 	device->sector_size = fs_info->sectorsize;
252 	device->total_bytes = btrfs_device_get_total_bytes(srcdev);
253 	device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
254 	device->bytes_used = btrfs_device_get_bytes_used(srcdev);
255 	device->commit_total_bytes = srcdev->commit_total_bytes;
256 	device->commit_bytes_used = device->bytes_used;
257 	device->fs_info = fs_info;
258 	device->bdev = bdev;
259 	set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
260 	set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
261 	device->mode = FMODE_EXCL;
262 	device->dev_stats_valid = 1;
263 	set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
264 	device->fs_devices = fs_info->fs_devices;
265 	list_add(&device->dev_list, &fs_info->fs_devices->devices);
266 	fs_info->fs_devices->num_devices++;
267 	fs_info->fs_devices->open_devices++;
268 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
269 
270 	*device_out = device;
271 	return 0;
272 
273 error:
274 	blkdev_put(bdev, FMODE_EXCL);
275 	return ret;
276 }
277 
278 /*
279  * called from commit_transaction. Writes changed device replace state to
280  * disk.
281  */
282 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
283 			  struct btrfs_fs_info *fs_info)
284 {
285 	int ret;
286 	struct btrfs_root *dev_root = fs_info->dev_root;
287 	struct btrfs_path *path;
288 	struct btrfs_key key;
289 	struct extent_buffer *eb;
290 	struct btrfs_dev_replace_item *ptr;
291 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
292 
293 	btrfs_dev_replace_read_lock(dev_replace);
294 	if (!dev_replace->is_valid ||
295 	    !dev_replace->item_needs_writeback) {
296 		btrfs_dev_replace_read_unlock(dev_replace);
297 		return 0;
298 	}
299 	btrfs_dev_replace_read_unlock(dev_replace);
300 
301 	key.objectid = 0;
302 	key.type = BTRFS_DEV_REPLACE_KEY;
303 	key.offset = 0;
304 
305 	path = btrfs_alloc_path();
306 	if (!path) {
307 		ret = -ENOMEM;
308 		goto out;
309 	}
310 	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
311 	if (ret < 0) {
312 		btrfs_warn(fs_info,
313 			   "error %d while searching for dev_replace item!",
314 			   ret);
315 		goto out;
316 	}
317 
318 	if (ret == 0 &&
319 	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
320 		/*
321 		 * need to delete old one and insert a new one.
322 		 * Since no attempt is made to recover any old state, if the
323 		 * dev_replace state is 'running', the data on the target
324 		 * drive is lost.
325 		 * It would be possible to recover the state: just make sure
326 		 * that the beginning of the item is never changed and always
327 		 * contains all the essential information. Then read this
328 		 * minimal set of information and use it as a base for the
329 		 * new state.
330 		 */
331 		ret = btrfs_del_item(trans, dev_root, path);
332 		if (ret != 0) {
333 			btrfs_warn(fs_info,
334 				   "delete too small dev_replace item failed %d!",
335 				   ret);
336 			goto out;
337 		}
338 		ret = 1;
339 	}
340 
341 	if (ret == 1) {
342 		/* need to insert a new item */
343 		btrfs_release_path(path);
344 		ret = btrfs_insert_empty_item(trans, dev_root, path,
345 					      &key, sizeof(*ptr));
346 		if (ret < 0) {
347 			btrfs_warn(fs_info,
348 				   "insert dev_replace item failed %d!", ret);
349 			goto out;
350 		}
351 	}
352 
353 	eb = path->nodes[0];
354 	ptr = btrfs_item_ptr(eb, path->slots[0],
355 			     struct btrfs_dev_replace_item);
356 
357 	btrfs_dev_replace_write_lock(dev_replace);
358 	if (dev_replace->srcdev)
359 		btrfs_set_dev_replace_src_devid(eb, ptr,
360 			dev_replace->srcdev->devid);
361 	else
362 		btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
363 	btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
364 		dev_replace->cont_reading_from_srcdev_mode);
365 	btrfs_set_dev_replace_replace_state(eb, ptr,
366 		dev_replace->replace_state);
367 	btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
368 	btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
369 	btrfs_set_dev_replace_num_write_errors(eb, ptr,
370 		atomic64_read(&dev_replace->num_write_errors));
371 	btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
372 		atomic64_read(&dev_replace->num_uncorrectable_read_errors));
373 	dev_replace->cursor_left_last_write_of_item =
374 		dev_replace->cursor_left;
375 	btrfs_set_dev_replace_cursor_left(eb, ptr,
376 		dev_replace->cursor_left_last_write_of_item);
377 	btrfs_set_dev_replace_cursor_right(eb, ptr,
378 		dev_replace->cursor_right);
379 	dev_replace->item_needs_writeback = 0;
380 	btrfs_dev_replace_write_unlock(dev_replace);
381 
382 	btrfs_mark_buffer_dirty(eb);
383 
384 out:
385 	btrfs_free_path(path);
386 
387 	return ret;
388 }
389 
390 void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
391 {
392 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
393 
394 	dev_replace->committed_cursor_left =
395 		dev_replace->cursor_left_last_write_of_item;
396 }
397 
398 static char* btrfs_dev_name(struct btrfs_device *device)
399 {
400 	if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
401 		return "<missing disk>";
402 	else
403 		return rcu_str_deref(device->name);
404 }
405 
406 int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
407 		const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
408 		int read_src)
409 {
410 	struct btrfs_root *root = fs_info->dev_root;
411 	struct btrfs_trans_handle *trans;
412 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
413 	int ret;
414 	struct btrfs_device *tgt_device = NULL;
415 	struct btrfs_device *src_device = NULL;
416 
417 	ret = btrfs_find_device_by_devspec(fs_info, srcdevid,
418 					    srcdev_name, &src_device);
419 	if (ret)
420 		return ret;
421 
422 	ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
423 					    src_device, &tgt_device);
424 	if (ret)
425 		return ret;
426 
427 	/*
428 	 * Here we commit the transaction to make sure commit_total_bytes
429 	 * of all the devices are updated.
430 	 */
431 	trans = btrfs_attach_transaction(root);
432 	if (!IS_ERR(trans)) {
433 		ret = btrfs_commit_transaction(trans);
434 		if (ret)
435 			return ret;
436 	} else if (PTR_ERR(trans) != -ENOENT) {
437 		return PTR_ERR(trans);
438 	}
439 
440 	btrfs_dev_replace_write_lock(dev_replace);
441 	switch (dev_replace->replace_state) {
442 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
443 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
444 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
445 		break;
446 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
447 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
448 		ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
449 		goto leave;
450 	}
451 
452 	dev_replace->cont_reading_from_srcdev_mode = read_src;
453 	WARN_ON(!src_device);
454 	dev_replace->srcdev = src_device;
455 	dev_replace->tgtdev = tgt_device;
456 
457 	btrfs_info_in_rcu(fs_info,
458 		      "dev_replace from %s (devid %llu) to %s started",
459 		      btrfs_dev_name(src_device),
460 		      src_device->devid,
461 		      rcu_str_deref(tgt_device->name));
462 
463 	/*
464 	 * from now on, the writes to the srcdev are all duplicated to
465 	 * go to the tgtdev as well (refer to btrfs_map_block()).
466 	 */
467 	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
468 	dev_replace->time_started = get_seconds();
469 	dev_replace->cursor_left = 0;
470 	dev_replace->committed_cursor_left = 0;
471 	dev_replace->cursor_left_last_write_of_item = 0;
472 	dev_replace->cursor_right = 0;
473 	dev_replace->is_valid = 1;
474 	dev_replace->item_needs_writeback = 1;
475 	atomic64_set(&dev_replace->num_write_errors, 0);
476 	atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
477 	btrfs_dev_replace_write_unlock(dev_replace);
478 
479 	ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device);
480 	if (ret)
481 		btrfs_err(fs_info, "kobj add dev failed %d", ret);
482 
483 	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
484 
485 	/* force writing the updated state information to disk */
486 	trans = btrfs_start_transaction(root, 0);
487 	if (IS_ERR(trans)) {
488 		ret = PTR_ERR(trans);
489 		btrfs_dev_replace_write_lock(dev_replace);
490 		goto leave;
491 	}
492 
493 	ret = btrfs_commit_transaction(trans);
494 	WARN_ON(ret);
495 
496 	/* the disk copy procedure reuses the scrub code */
497 	ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
498 			      btrfs_device_get_total_bytes(src_device),
499 			      &dev_replace->scrub_progress, 0, 1);
500 
501 	ret = btrfs_dev_replace_finishing(fs_info, ret);
502 	if (ret == -EINPROGRESS) {
503 		ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
504 	} else {
505 		WARN_ON(ret);
506 	}
507 
508 	return ret;
509 
510 leave:
511 	dev_replace->srcdev = NULL;
512 	dev_replace->tgtdev = NULL;
513 	btrfs_dev_replace_write_unlock(dev_replace);
514 	btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
515 	return ret;
516 }
517 
518 int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
519 			    struct btrfs_ioctl_dev_replace_args *args)
520 {
521 	int ret;
522 
523 	switch (args->start.cont_reading_from_srcdev_mode) {
524 	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
525 	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
526 		break;
527 	default:
528 		return -EINVAL;
529 	}
530 
531 	if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
532 	    args->start.tgtdev_name[0] == '\0')
533 		return -EINVAL;
534 
535 	ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name,
536 					args->start.srcdevid,
537 					args->start.srcdev_name,
538 					args->start.cont_reading_from_srcdev_mode);
539 	args->result = ret;
540 	/* don't warn if EINPROGRESS, someone else might be running scrub */
541 	if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS)
542 		ret = 0;
543 
544 	return ret;
545 }
546 
547 /*
548  * blocked until all in-flight bios operations are finished.
549  */
550 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
551 {
552 	set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
553 	wait_event(fs_info->replace_wait, !percpu_counter_sum(
554 		   &fs_info->bio_counter));
555 }
556 
557 /*
558  * we have removed target device, it is safe to allow new bios request.
559  */
560 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
561 {
562 	clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
563 	wake_up(&fs_info->replace_wait);
564 }
565 
566 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
567 				       int scrub_ret)
568 {
569 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
570 	struct btrfs_device *tgt_device;
571 	struct btrfs_device *src_device;
572 	struct btrfs_root *root = fs_info->tree_root;
573 	u8 uuid_tmp[BTRFS_UUID_SIZE];
574 	struct btrfs_trans_handle *trans;
575 	int ret = 0;
576 
577 	/* don't allow cancel or unmount to disturb the finishing procedure */
578 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
579 
580 	btrfs_dev_replace_read_lock(dev_replace);
581 	/* was the operation canceled, or is it finished? */
582 	if (dev_replace->replace_state !=
583 	    BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
584 		btrfs_dev_replace_read_unlock(dev_replace);
585 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
586 		return 0;
587 	}
588 
589 	tgt_device = dev_replace->tgtdev;
590 	src_device = dev_replace->srcdev;
591 	btrfs_dev_replace_read_unlock(dev_replace);
592 
593 	/*
594 	 * flush all outstanding I/O and inode extent mappings before the
595 	 * copy operation is declared as being finished
596 	 */
597 	ret = btrfs_start_delalloc_roots(fs_info, -1);
598 	if (ret) {
599 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
600 		return ret;
601 	}
602 	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
603 
604 	trans = btrfs_start_transaction(root, 0);
605 	if (IS_ERR(trans)) {
606 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
607 		return PTR_ERR(trans);
608 	}
609 	ret = btrfs_commit_transaction(trans);
610 	WARN_ON(ret);
611 
612 	/* keep away write_all_supers() during the finishing procedure */
613 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
614 	mutex_lock(&fs_info->chunk_mutex);
615 	btrfs_dev_replace_write_lock(dev_replace);
616 	dev_replace->replace_state =
617 		scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
618 			  : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
619 	dev_replace->tgtdev = NULL;
620 	dev_replace->srcdev = NULL;
621 	dev_replace->time_stopped = get_seconds();
622 	dev_replace->item_needs_writeback = 1;
623 
624 	/* replace old device with new one in mapping tree */
625 	if (!scrub_ret) {
626 		btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
627 								src_device,
628 								tgt_device);
629 	} else {
630 		btrfs_err_in_rcu(fs_info,
631 				 "btrfs_scrub_dev(%s, %llu, %s) failed %d",
632 				 btrfs_dev_name(src_device),
633 				 src_device->devid,
634 				 rcu_str_deref(tgt_device->name), scrub_ret);
635 		btrfs_dev_replace_write_unlock(dev_replace);
636 		mutex_unlock(&fs_info->chunk_mutex);
637 		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
638 		btrfs_rm_dev_replace_blocked(fs_info);
639 		if (tgt_device)
640 			btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
641 		btrfs_rm_dev_replace_unblocked(fs_info);
642 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
643 
644 		return scrub_ret;
645 	}
646 
647 	btrfs_info_in_rcu(fs_info,
648 			  "dev_replace from %s (devid %llu) to %s finished",
649 			  btrfs_dev_name(src_device),
650 			  src_device->devid,
651 			  rcu_str_deref(tgt_device->name));
652 	clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state);
653 	tgt_device->devid = src_device->devid;
654 	src_device->devid = BTRFS_DEV_REPLACE_DEVID;
655 	memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
656 	memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
657 	memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
658 	btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
659 	btrfs_device_set_disk_total_bytes(tgt_device,
660 					  src_device->disk_total_bytes);
661 	btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
662 	ASSERT(list_empty(&src_device->resized_list));
663 	tgt_device->commit_total_bytes = src_device->commit_total_bytes;
664 	tgt_device->commit_bytes_used = src_device->bytes_used;
665 
666 	btrfs_assign_next_active_device(fs_info, src_device, tgt_device);
667 
668 	list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
669 	fs_info->fs_devices->rw_devices++;
670 
671 	btrfs_dev_replace_write_unlock(dev_replace);
672 
673 	btrfs_rm_dev_replace_blocked(fs_info);
674 
675 	btrfs_rm_dev_replace_remove_srcdev(fs_info, src_device);
676 
677 	btrfs_rm_dev_replace_unblocked(fs_info);
678 
679 	/*
680 	 * this is again a consistent state where no dev_replace procedure
681 	 * is running, the target device is part of the filesystem, the
682 	 * source device is not part of the filesystem anymore and its 1st
683 	 * superblock is scratched out so that it is no longer marked to
684 	 * belong to this filesystem.
685 	 */
686 	mutex_unlock(&fs_info->chunk_mutex);
687 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
688 
689 	/* replace the sysfs entry */
690 	btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device);
691 	btrfs_rm_dev_replace_free_srcdev(fs_info, src_device);
692 
693 	/* write back the superblocks */
694 	trans = btrfs_start_transaction(root, 0);
695 	if (!IS_ERR(trans))
696 		btrfs_commit_transaction(trans);
697 
698 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
699 
700 	return 0;
701 }
702 
703 static void btrfs_dev_replace_update_device_in_mapping_tree(
704 						struct btrfs_fs_info *fs_info,
705 						struct btrfs_device *srcdev,
706 						struct btrfs_device *tgtdev)
707 {
708 	struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
709 	struct extent_map *em;
710 	struct map_lookup *map;
711 	u64 start = 0;
712 	int i;
713 
714 	write_lock(&em_tree->lock);
715 	do {
716 		em = lookup_extent_mapping(em_tree, start, (u64)-1);
717 		if (!em)
718 			break;
719 		map = em->map_lookup;
720 		for (i = 0; i < map->num_stripes; i++)
721 			if (srcdev == map->stripes[i].dev)
722 				map->stripes[i].dev = tgtdev;
723 		start = em->start + em->len;
724 		free_extent_map(em);
725 	} while (start);
726 	write_unlock(&em_tree->lock);
727 }
728 
729 /*
730  * Read progress of device replace status according to the state and last
731  * stored position. The value format is the same as for
732  * btrfs_dev_replace::progress_1000
733  */
734 static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info)
735 {
736 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
737 	u64 ret = 0;
738 
739 	switch (dev_replace->replace_state) {
740 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
741 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
742 		ret = 0;
743 		break;
744 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
745 		ret = 1000;
746 		break;
747 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
748 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
749 		ret = div64_u64(dev_replace->cursor_left,
750 				div_u64(btrfs_device_get_total_bytes(
751 						dev_replace->srcdev), 1000));
752 		break;
753 	}
754 
755 	return ret;
756 }
757 
758 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
759 			      struct btrfs_ioctl_dev_replace_args *args)
760 {
761 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
762 
763 	btrfs_dev_replace_read_lock(dev_replace);
764 	/* even if !dev_replace_is_valid, the values are good enough for
765 	 * the replace_status ioctl */
766 	args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
767 	args->status.replace_state = dev_replace->replace_state;
768 	args->status.time_started = dev_replace->time_started;
769 	args->status.time_stopped = dev_replace->time_stopped;
770 	args->status.num_write_errors =
771 		atomic64_read(&dev_replace->num_write_errors);
772 	args->status.num_uncorrectable_read_errors =
773 		atomic64_read(&dev_replace->num_uncorrectable_read_errors);
774 	args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
775 	btrfs_dev_replace_read_unlock(dev_replace);
776 }
777 
778 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
779 {
780 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
781 	struct btrfs_device *tgt_device = NULL;
782 	struct btrfs_device *src_device = NULL;
783 	struct btrfs_trans_handle *trans;
784 	struct btrfs_root *root = fs_info->tree_root;
785 	int result;
786 	int ret;
787 
788 	if (sb_rdonly(fs_info->sb))
789 		return -EROFS;
790 
791 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
792 	btrfs_dev_replace_write_lock(dev_replace);
793 	switch (dev_replace->replace_state) {
794 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
795 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
796 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
797 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
798 		btrfs_dev_replace_write_unlock(dev_replace);
799 		goto leave;
800 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
801 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
802 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
803 		tgt_device = dev_replace->tgtdev;
804 		src_device = dev_replace->srcdev;
805 		dev_replace->tgtdev = NULL;
806 		dev_replace->srcdev = NULL;
807 		break;
808 	}
809 	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
810 	dev_replace->time_stopped = get_seconds();
811 	dev_replace->item_needs_writeback = 1;
812 	btrfs_dev_replace_write_unlock(dev_replace);
813 	btrfs_scrub_cancel(fs_info);
814 
815 	trans = btrfs_start_transaction(root, 0);
816 	if (IS_ERR(trans)) {
817 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
818 		return PTR_ERR(trans);
819 	}
820 	ret = btrfs_commit_transaction(trans);
821 	WARN_ON(ret);
822 
823 	btrfs_info_in_rcu(fs_info,
824 		"dev_replace from %s (devid %llu) to %s canceled",
825 		btrfs_dev_name(src_device), src_device->devid,
826 		btrfs_dev_name(tgt_device));
827 
828 	if (tgt_device)
829 		btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
830 
831 leave:
832 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
833 	return result;
834 }
835 
836 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
837 {
838 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
839 
840 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
841 	btrfs_dev_replace_write_lock(dev_replace);
842 	switch (dev_replace->replace_state) {
843 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
844 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
845 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
846 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
847 		break;
848 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
849 		dev_replace->replace_state =
850 			BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
851 		dev_replace->time_stopped = get_seconds();
852 		dev_replace->item_needs_writeback = 1;
853 		btrfs_info(fs_info, "suspending dev_replace for unmount");
854 		break;
855 	}
856 
857 	btrfs_dev_replace_write_unlock(dev_replace);
858 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
859 }
860 
861 /* resume dev_replace procedure that was interrupted by unmount */
862 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
863 {
864 	struct task_struct *task;
865 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
866 
867 	btrfs_dev_replace_write_lock(dev_replace);
868 	switch (dev_replace->replace_state) {
869 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
870 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
871 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
872 		btrfs_dev_replace_write_unlock(dev_replace);
873 		return 0;
874 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
875 		break;
876 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
877 		dev_replace->replace_state =
878 			BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
879 		break;
880 	}
881 	if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
882 		btrfs_info(fs_info,
883 			   "cannot continue dev_replace, tgtdev is missing");
884 		btrfs_info(fs_info,
885 			   "you may cancel the operation after 'mount -o degraded'");
886 		btrfs_dev_replace_write_unlock(dev_replace);
887 		return 0;
888 	}
889 	btrfs_dev_replace_write_unlock(dev_replace);
890 
891 	/*
892 	 * This could collide with a paused balance, but the exclusive op logic
893 	 * should never allow both to start and pause. We don't want to allow
894 	 * dev-replace to start anyway.
895 	 */
896 	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
897 		btrfs_info(fs_info,
898 		"cannot resume dev-replace, other exclusive operation running");
899 		return 0;
900 	}
901 
902 	task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
903 	return PTR_ERR_OR_ZERO(task);
904 }
905 
906 static int btrfs_dev_replace_kthread(void *data)
907 {
908 	struct btrfs_fs_info *fs_info = data;
909 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
910 	u64 progress;
911 	int ret;
912 
913 	progress = btrfs_dev_replace_progress(fs_info);
914 	progress = div_u64(progress, 10);
915 	btrfs_info_in_rcu(fs_info,
916 		"continuing dev_replace from %s (devid %llu) to target %s @%u%%",
917 		btrfs_dev_name(dev_replace->srcdev),
918 		dev_replace->srcdev->devid,
919 		btrfs_dev_name(dev_replace->tgtdev),
920 		(unsigned int)progress);
921 
922 	ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
923 			      dev_replace->committed_cursor_left,
924 			      btrfs_device_get_total_bytes(dev_replace->srcdev),
925 			      &dev_replace->scrub_progress, 0, 1);
926 	ret = btrfs_dev_replace_finishing(fs_info, ret);
927 	WARN_ON(ret);
928 
929 	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
930 	return 0;
931 }
932 
933 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
934 {
935 	if (!dev_replace->is_valid)
936 		return 0;
937 
938 	switch (dev_replace->replace_state) {
939 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
940 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
941 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
942 		return 0;
943 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
944 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
945 		/*
946 		 * return true even if tgtdev is missing (this is
947 		 * something that can happen if the dev_replace
948 		 * procedure is suspended by an umount and then
949 		 * the tgtdev is missing (or "btrfs dev scan") was
950 		 * not called and the the filesystem is remounted
951 		 * in degraded state. This does not stop the
952 		 * dev_replace procedure. It needs to be canceled
953 		 * manually if the cancellation is wanted.
954 		 */
955 		break;
956 	}
957 	return 1;
958 }
959 
960 void btrfs_dev_replace_read_lock(struct btrfs_dev_replace *dev_replace)
961 {
962 	read_lock(&dev_replace->lock);
963 	atomic_inc(&dev_replace->read_locks);
964 }
965 
966 void btrfs_dev_replace_read_unlock(struct btrfs_dev_replace *dev_replace)
967 {
968 	ASSERT(atomic_read(&dev_replace->read_locks) > 0);
969 	atomic_dec(&dev_replace->read_locks);
970 	read_unlock(&dev_replace->lock);
971 }
972 
973 void btrfs_dev_replace_write_lock(struct btrfs_dev_replace *dev_replace)
974 {
975 again:
976 	wait_event(dev_replace->read_lock_wq,
977 		   atomic_read(&dev_replace->blocking_readers) == 0);
978 	write_lock(&dev_replace->lock);
979 	if (atomic_read(&dev_replace->blocking_readers)) {
980 		write_unlock(&dev_replace->lock);
981 		goto again;
982 	}
983 }
984 
985 void btrfs_dev_replace_write_unlock(struct btrfs_dev_replace *dev_replace)
986 {
987 	ASSERT(atomic_read(&dev_replace->blocking_readers) == 0);
988 	write_unlock(&dev_replace->lock);
989 }
990 
991 /* inc blocking cnt and release read lock */
992 void btrfs_dev_replace_set_lock_blocking(
993 					struct btrfs_dev_replace *dev_replace)
994 {
995 	/* only set blocking for read lock */
996 	ASSERT(atomic_read(&dev_replace->read_locks) > 0);
997 	atomic_inc(&dev_replace->blocking_readers);
998 	read_unlock(&dev_replace->lock);
999 }
1000 
1001 /* acquire read lock and dec blocking cnt */
1002 void btrfs_dev_replace_clear_lock_blocking(
1003 					struct btrfs_dev_replace *dev_replace)
1004 {
1005 	/* only set blocking for read lock */
1006 	ASSERT(atomic_read(&dev_replace->read_locks) > 0);
1007 	ASSERT(atomic_read(&dev_replace->blocking_readers) > 0);
1008 	read_lock(&dev_replace->lock);
1009 	/* Barrier implied by atomic_dec_and_test */
1010 	if (atomic_dec_and_test(&dev_replace->blocking_readers))
1011 		cond_wake_up_nomb(&dev_replace->read_lock_wq);
1012 }
1013 
1014 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
1015 {
1016 	percpu_counter_inc(&fs_info->bio_counter);
1017 }
1018 
1019 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
1020 {
1021 	percpu_counter_sub(&fs_info->bio_counter, amount);
1022 	cond_wake_up_nomb(&fs_info->replace_wait);
1023 }
1024 
1025 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
1026 {
1027 	while (1) {
1028 		percpu_counter_inc(&fs_info->bio_counter);
1029 		if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1030 				     &fs_info->fs_state)))
1031 			break;
1032 
1033 		btrfs_bio_counter_dec(fs_info);
1034 		wait_event(fs_info->replace_wait,
1035 			   !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1036 				     &fs_info->fs_state));
1037 	}
1038 }
1039