xref: /openbmc/linux/fs/btrfs/dev-replace.c (revision be709d48)
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->time_started = 0;
63 		dev_replace->time_stopped = 0;
64 		atomic64_set(&dev_replace->num_write_errors, 0);
65 		atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
66 		dev_replace->cursor_left = 0;
67 		dev_replace->committed_cursor_left = 0;
68 		dev_replace->cursor_left_last_write_of_item = 0;
69 		dev_replace->cursor_right = 0;
70 		dev_replace->srcdev = NULL;
71 		dev_replace->tgtdev = NULL;
72 		dev_replace->is_valid = 0;
73 		dev_replace->item_needs_writeback = 0;
74 		goto out;
75 	}
76 	slot = path->slots[0];
77 	eb = path->nodes[0];
78 	item_size = btrfs_item_size_nr(eb, slot);
79 	ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
80 
81 	if (item_size != sizeof(struct btrfs_dev_replace_item)) {
82 		btrfs_warn(fs_info,
83 			"dev_replace entry found has unexpected size, ignore entry");
84 		goto no_valid_dev_replace_entry_found;
85 	}
86 
87 	src_devid = btrfs_dev_replace_src_devid(eb, ptr);
88 	dev_replace->cont_reading_from_srcdev_mode =
89 		btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
90 	dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
91 	dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
92 	dev_replace->time_stopped =
93 		btrfs_dev_replace_time_stopped(eb, ptr);
94 	atomic64_set(&dev_replace->num_write_errors,
95 		     btrfs_dev_replace_num_write_errors(eb, ptr));
96 	atomic64_set(&dev_replace->num_uncorrectable_read_errors,
97 		     btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
98 	dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
99 	dev_replace->committed_cursor_left = dev_replace->cursor_left;
100 	dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
101 	dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
102 	dev_replace->is_valid = 1;
103 
104 	dev_replace->item_needs_writeback = 0;
105 	switch (dev_replace->replace_state) {
106 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
107 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
108 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
109 		dev_replace->srcdev = NULL;
110 		dev_replace->tgtdev = NULL;
111 		break;
112 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
113 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
114 		dev_replace->srcdev = btrfs_find_device(fs_info->fs_devices,
115 						src_devid, NULL, NULL, true);
116 		dev_replace->tgtdev = btrfs_find_device(fs_info->fs_devices,
117 							BTRFS_DEV_REPLACE_DEVID,
118 							NULL, NULL, true);
119 		/*
120 		 * allow 'btrfs dev replace_cancel' if src/tgt device is
121 		 * missing
122 		 */
123 		if (!dev_replace->srcdev &&
124 		    !btrfs_test_opt(fs_info, DEGRADED)) {
125 			ret = -EIO;
126 			btrfs_warn(fs_info,
127 			   "cannot mount because device replace operation is ongoing and");
128 			btrfs_warn(fs_info,
129 			   "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
130 			   src_devid);
131 		}
132 		if (!dev_replace->tgtdev &&
133 		    !btrfs_test_opt(fs_info, DEGRADED)) {
134 			ret = -EIO;
135 			btrfs_warn(fs_info,
136 			   "cannot mount because device replace operation is ongoing and");
137 			btrfs_warn(fs_info,
138 			   "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
139 				BTRFS_DEV_REPLACE_DEVID);
140 		}
141 		if (dev_replace->tgtdev) {
142 			if (dev_replace->srcdev) {
143 				dev_replace->tgtdev->total_bytes =
144 					dev_replace->srcdev->total_bytes;
145 				dev_replace->tgtdev->disk_total_bytes =
146 					dev_replace->srcdev->disk_total_bytes;
147 				dev_replace->tgtdev->commit_total_bytes =
148 					dev_replace->srcdev->commit_total_bytes;
149 				dev_replace->tgtdev->bytes_used =
150 					dev_replace->srcdev->bytes_used;
151 				dev_replace->tgtdev->commit_bytes_used =
152 					dev_replace->srcdev->commit_bytes_used;
153 			}
154 			set_bit(BTRFS_DEV_STATE_REPLACE_TGT,
155 				&dev_replace->tgtdev->dev_state);
156 
157 			WARN_ON(fs_info->fs_devices->rw_devices == 0);
158 			dev_replace->tgtdev->io_width = fs_info->sectorsize;
159 			dev_replace->tgtdev->io_align = fs_info->sectorsize;
160 			dev_replace->tgtdev->sector_size = fs_info->sectorsize;
161 			dev_replace->tgtdev->fs_info = fs_info;
162 			set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
163 				&dev_replace->tgtdev->dev_state);
164 		}
165 		break;
166 	}
167 
168 out:
169 	btrfs_free_path(path);
170 	return ret;
171 }
172 
173 /*
174  * Initialize a new device for device replace target from a given source dev
175  * and path.
176  *
177  * Return 0 and new device in @device_out, otherwise return < 0
178  */
179 static int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
180 				  const char *device_path,
181 				  struct btrfs_device *srcdev,
182 				  struct btrfs_device **device_out)
183 {
184 	struct btrfs_device *device;
185 	struct block_device *bdev;
186 	struct list_head *devices;
187 	struct rcu_string *name;
188 	u64 devid = BTRFS_DEV_REPLACE_DEVID;
189 	int ret = 0;
190 
191 	*device_out = NULL;
192 	if (fs_info->fs_devices->seeding) {
193 		btrfs_err(fs_info, "the filesystem is a seed filesystem!");
194 		return -EINVAL;
195 	}
196 
197 	bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
198 				  fs_info->bdev_holder);
199 	if (IS_ERR(bdev)) {
200 		btrfs_err(fs_info, "target device %s is invalid!", device_path);
201 		return PTR_ERR(bdev);
202 	}
203 
204 	filemap_write_and_wait(bdev->bd_inode->i_mapping);
205 
206 	devices = &fs_info->fs_devices->devices;
207 	list_for_each_entry(device, devices, dev_list) {
208 		if (device->bdev == bdev) {
209 			btrfs_err(fs_info,
210 				  "target device is in the filesystem!");
211 			ret = -EEXIST;
212 			goto error;
213 		}
214 	}
215 
216 
217 	if (i_size_read(bdev->bd_inode) <
218 	    btrfs_device_get_total_bytes(srcdev)) {
219 		btrfs_err(fs_info,
220 			  "target device is smaller than source device!");
221 		ret = -EINVAL;
222 		goto error;
223 	}
224 
225 
226 	device = btrfs_alloc_device(NULL, &devid, NULL);
227 	if (IS_ERR(device)) {
228 		ret = PTR_ERR(device);
229 		goto error;
230 	}
231 
232 	name = rcu_string_strdup(device_path, GFP_KERNEL);
233 	if (!name) {
234 		btrfs_free_device(device);
235 		ret = -ENOMEM;
236 		goto error;
237 	}
238 	rcu_assign_pointer(device->name, name);
239 
240 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
241 	set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
242 	device->generation = 0;
243 	device->io_width = fs_info->sectorsize;
244 	device->io_align = fs_info->sectorsize;
245 	device->sector_size = fs_info->sectorsize;
246 	device->total_bytes = btrfs_device_get_total_bytes(srcdev);
247 	device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
248 	device->bytes_used = btrfs_device_get_bytes_used(srcdev);
249 	device->commit_total_bytes = srcdev->commit_total_bytes;
250 	device->commit_bytes_used = device->bytes_used;
251 	device->fs_info = fs_info;
252 	device->bdev = bdev;
253 	set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
254 	set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
255 	device->mode = FMODE_EXCL;
256 	device->dev_stats_valid = 1;
257 	set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
258 	device->fs_devices = fs_info->fs_devices;
259 	list_add(&device->dev_list, &fs_info->fs_devices->devices);
260 	fs_info->fs_devices->num_devices++;
261 	fs_info->fs_devices->open_devices++;
262 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
263 
264 	*device_out = device;
265 	return 0;
266 
267 error:
268 	blkdev_put(bdev, FMODE_EXCL);
269 	return ret;
270 }
271 
272 /*
273  * called from commit_transaction. Writes changed device replace state to
274  * disk.
275  */
276 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
277 			  struct btrfs_fs_info *fs_info)
278 {
279 	int ret;
280 	struct btrfs_root *dev_root = fs_info->dev_root;
281 	struct btrfs_path *path;
282 	struct btrfs_key key;
283 	struct extent_buffer *eb;
284 	struct btrfs_dev_replace_item *ptr;
285 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
286 
287 	down_read(&dev_replace->rwsem);
288 	if (!dev_replace->is_valid ||
289 	    !dev_replace->item_needs_writeback) {
290 		up_read(&dev_replace->rwsem);
291 		return 0;
292 	}
293 	up_read(&dev_replace->rwsem);
294 
295 	key.objectid = 0;
296 	key.type = BTRFS_DEV_REPLACE_KEY;
297 	key.offset = 0;
298 
299 	path = btrfs_alloc_path();
300 	if (!path) {
301 		ret = -ENOMEM;
302 		goto out;
303 	}
304 	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
305 	if (ret < 0) {
306 		btrfs_warn(fs_info,
307 			   "error %d while searching for dev_replace item!",
308 			   ret);
309 		goto out;
310 	}
311 
312 	if (ret == 0 &&
313 	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
314 		/*
315 		 * need to delete old one and insert a new one.
316 		 * Since no attempt is made to recover any old state, if the
317 		 * dev_replace state is 'running', the data on the target
318 		 * drive is lost.
319 		 * It would be possible to recover the state: just make sure
320 		 * that the beginning of the item is never changed and always
321 		 * contains all the essential information. Then read this
322 		 * minimal set of information and use it as a base for the
323 		 * new state.
324 		 */
325 		ret = btrfs_del_item(trans, dev_root, path);
326 		if (ret != 0) {
327 			btrfs_warn(fs_info,
328 				   "delete too small dev_replace item failed %d!",
329 				   ret);
330 			goto out;
331 		}
332 		ret = 1;
333 	}
334 
335 	if (ret == 1) {
336 		/* need to insert a new item */
337 		btrfs_release_path(path);
338 		ret = btrfs_insert_empty_item(trans, dev_root, path,
339 					      &key, sizeof(*ptr));
340 		if (ret < 0) {
341 			btrfs_warn(fs_info,
342 				   "insert dev_replace item failed %d!", ret);
343 			goto out;
344 		}
345 	}
346 
347 	eb = path->nodes[0];
348 	ptr = btrfs_item_ptr(eb, path->slots[0],
349 			     struct btrfs_dev_replace_item);
350 
351 	down_write(&dev_replace->rwsem);
352 	if (dev_replace->srcdev)
353 		btrfs_set_dev_replace_src_devid(eb, ptr,
354 			dev_replace->srcdev->devid);
355 	else
356 		btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
357 	btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
358 		dev_replace->cont_reading_from_srcdev_mode);
359 	btrfs_set_dev_replace_replace_state(eb, ptr,
360 		dev_replace->replace_state);
361 	btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
362 	btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
363 	btrfs_set_dev_replace_num_write_errors(eb, ptr,
364 		atomic64_read(&dev_replace->num_write_errors));
365 	btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
366 		atomic64_read(&dev_replace->num_uncorrectable_read_errors));
367 	dev_replace->cursor_left_last_write_of_item =
368 		dev_replace->cursor_left;
369 	btrfs_set_dev_replace_cursor_left(eb, ptr,
370 		dev_replace->cursor_left_last_write_of_item);
371 	btrfs_set_dev_replace_cursor_right(eb, ptr,
372 		dev_replace->cursor_right);
373 	dev_replace->item_needs_writeback = 0;
374 	up_write(&dev_replace->rwsem);
375 
376 	btrfs_mark_buffer_dirty(eb);
377 
378 out:
379 	btrfs_free_path(path);
380 
381 	return ret;
382 }
383 
384 static char* btrfs_dev_name(struct btrfs_device *device)
385 {
386 	if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
387 		return "<missing disk>";
388 	else
389 		return rcu_str_deref(device->name);
390 }
391 
392 static int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
393 		const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
394 		int read_src)
395 {
396 	struct btrfs_root *root = fs_info->dev_root;
397 	struct btrfs_trans_handle *trans;
398 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
399 	int ret;
400 	struct btrfs_device *tgt_device = NULL;
401 	struct btrfs_device *src_device = NULL;
402 	bool need_unlock;
403 
404 	src_device = btrfs_find_device_by_devspec(fs_info, srcdevid,
405 						  srcdev_name);
406 	if (IS_ERR(src_device))
407 		return PTR_ERR(src_device);
408 
409 	if (btrfs_pinned_by_swapfile(fs_info, src_device)) {
410 		btrfs_warn_in_rcu(fs_info,
411 	  "cannot replace device %s (devid %llu) due to active swapfile",
412 			btrfs_dev_name(src_device), src_device->devid);
413 		return -ETXTBSY;
414 	}
415 
416 	ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
417 					    src_device, &tgt_device);
418 	if (ret)
419 		return ret;
420 
421 	/*
422 	 * Here we commit the transaction to make sure commit_total_bytes
423 	 * of all the devices are updated.
424 	 */
425 	trans = btrfs_attach_transaction(root);
426 	if (!IS_ERR(trans)) {
427 		ret = btrfs_commit_transaction(trans);
428 		if (ret)
429 			return ret;
430 	} else if (PTR_ERR(trans) != -ENOENT) {
431 		return PTR_ERR(trans);
432 	}
433 
434 	need_unlock = true;
435 	down_write(&dev_replace->rwsem);
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 		ASSERT(0);
444 		ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
445 		goto leave;
446 	}
447 
448 	dev_replace->cont_reading_from_srcdev_mode = read_src;
449 	WARN_ON(!src_device);
450 	dev_replace->srcdev = src_device;
451 	dev_replace->tgtdev = tgt_device;
452 
453 	btrfs_info_in_rcu(fs_info,
454 		      "dev_replace from %s (devid %llu) to %s started",
455 		      btrfs_dev_name(src_device),
456 		      src_device->devid,
457 		      rcu_str_deref(tgt_device->name));
458 
459 	/*
460 	 * from now on, the writes to the srcdev are all duplicated to
461 	 * go to the tgtdev as well (refer to btrfs_map_block()).
462 	 */
463 	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
464 	dev_replace->time_started = ktime_get_real_seconds();
465 	dev_replace->cursor_left = 0;
466 	dev_replace->committed_cursor_left = 0;
467 	dev_replace->cursor_left_last_write_of_item = 0;
468 	dev_replace->cursor_right = 0;
469 	dev_replace->is_valid = 1;
470 	dev_replace->item_needs_writeback = 1;
471 	atomic64_set(&dev_replace->num_write_errors, 0);
472 	atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
473 	up_write(&dev_replace->rwsem);
474 	need_unlock = false;
475 
476 	ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device);
477 	if (ret)
478 		btrfs_err(fs_info, "kobj add dev failed %d", ret);
479 
480 	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
481 
482 	/* force writing the updated state information to disk */
483 	trans = btrfs_start_transaction(root, 0);
484 	if (IS_ERR(trans)) {
485 		ret = PTR_ERR(trans);
486 		need_unlock = true;
487 		down_write(&dev_replace->rwsem);
488 		dev_replace->replace_state =
489 			BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
490 		dev_replace->srcdev = NULL;
491 		dev_replace->tgtdev = NULL;
492 		goto leave;
493 	}
494 
495 	ret = btrfs_commit_transaction(trans);
496 	WARN_ON(ret);
497 
498 	/* the disk copy procedure reuses the scrub code */
499 	ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
500 			      btrfs_device_get_total_bytes(src_device),
501 			      &dev_replace->scrub_progress, 0, 1);
502 
503 	ret = btrfs_dev_replace_finishing(fs_info, ret);
504 	if (ret == -EINPROGRESS) {
505 		ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
506 	} else if (ret != -ECANCELED) {
507 		WARN_ON(ret);
508 	}
509 
510 	return ret;
511 
512 leave:
513 	if (need_unlock)
514 		up_write(&dev_replace->rwsem);
515 	btrfs_destroy_dev_replace_tgtdev(tgt_device);
516 	return ret;
517 }
518 
519 int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
520 			    struct btrfs_ioctl_dev_replace_args *args)
521 {
522 	int ret;
523 
524 	switch (args->start.cont_reading_from_srcdev_mode) {
525 	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
526 	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
527 		break;
528 	default:
529 		return -EINVAL;
530 	}
531 
532 	if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
533 	    args->start.tgtdev_name[0] == '\0')
534 		return -EINVAL;
535 
536 	ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name,
537 					args->start.srcdevid,
538 					args->start.srcdev_name,
539 					args->start.cont_reading_from_srcdev_mode);
540 	args->result = ret;
541 	/* don't warn if EINPROGRESS, someone else might be running scrub */
542 	if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS ||
543 	    ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR)
544 		return 0;
545 
546 	return ret;
547 }
548 
549 /*
550  * blocked until all in-flight bios operations are finished.
551  */
552 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
553 {
554 	set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
555 	wait_event(fs_info->dev_replace.replace_wait, !percpu_counter_sum(
556 		   &fs_info->dev_replace.bio_counter));
557 }
558 
559 /*
560  * we have removed target device, it is safe to allow new bios request.
561  */
562 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
563 {
564 	clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
565 	wake_up(&fs_info->dev_replace.replace_wait);
566 }
567 
568 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
569 				       int scrub_ret)
570 {
571 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
572 	struct btrfs_device *tgt_device;
573 	struct btrfs_device *src_device;
574 	struct btrfs_root *root = fs_info->tree_root;
575 	u8 uuid_tmp[BTRFS_UUID_SIZE];
576 	struct btrfs_trans_handle *trans;
577 	int ret = 0;
578 
579 	/* don't allow cancel or unmount to disturb the finishing procedure */
580 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
581 
582 	down_read(&dev_replace->rwsem);
583 	/* was the operation canceled, or is it finished? */
584 	if (dev_replace->replace_state !=
585 	    BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
586 		up_read(&dev_replace->rwsem);
587 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
588 		return 0;
589 	}
590 
591 	tgt_device = dev_replace->tgtdev;
592 	src_device = dev_replace->srcdev;
593 	up_read(&dev_replace->rwsem);
594 
595 	/*
596 	 * flush all outstanding I/O and inode extent mappings before the
597 	 * copy operation is declared as being finished
598 	 */
599 	ret = btrfs_start_delalloc_roots(fs_info, -1);
600 	if (ret) {
601 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
602 		return ret;
603 	}
604 	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
605 
606 	trans = btrfs_start_transaction(root, 0);
607 	if (IS_ERR(trans)) {
608 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
609 		return PTR_ERR(trans);
610 	}
611 	ret = btrfs_commit_transaction(trans);
612 	WARN_ON(ret);
613 
614 	/* keep away write_all_supers() during the finishing procedure */
615 	mutex_lock(&fs_info->fs_devices->device_list_mutex);
616 	mutex_lock(&fs_info->chunk_mutex);
617 	down_write(&dev_replace->rwsem);
618 	dev_replace->replace_state =
619 		scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
620 			  : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
621 	dev_replace->tgtdev = NULL;
622 	dev_replace->srcdev = NULL;
623 	dev_replace->time_stopped = ktime_get_real_seconds();
624 	dev_replace->item_needs_writeback = 1;
625 
626 	/* replace old device with new one in mapping tree */
627 	if (!scrub_ret) {
628 		btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
629 								src_device,
630 								tgt_device);
631 	} else {
632 		if (scrub_ret != -ECANCELED)
633 			btrfs_err_in_rcu(fs_info,
634 				 "btrfs_scrub_dev(%s, %llu, %s) failed %d",
635 				 btrfs_dev_name(src_device),
636 				 src_device->devid,
637 				 rcu_str_deref(tgt_device->name), scrub_ret);
638 		up_write(&dev_replace->rwsem);
639 		mutex_unlock(&fs_info->chunk_mutex);
640 		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
641 		btrfs_rm_dev_replace_blocked(fs_info);
642 		if (tgt_device)
643 			btrfs_destroy_dev_replace_tgtdev(tgt_device);
644 		btrfs_rm_dev_replace_unblocked(fs_info);
645 		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
646 
647 		return scrub_ret;
648 	}
649 
650 	btrfs_info_in_rcu(fs_info,
651 			  "dev_replace from %s (devid %llu) to %s finished",
652 			  btrfs_dev_name(src_device),
653 			  src_device->devid,
654 			  rcu_str_deref(tgt_device->name));
655 	clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state);
656 	tgt_device->devid = src_device->devid;
657 	src_device->devid = BTRFS_DEV_REPLACE_DEVID;
658 	memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
659 	memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
660 	memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
661 	btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
662 	btrfs_device_set_disk_total_bytes(tgt_device,
663 					  src_device->disk_total_bytes);
664 	btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
665 	ASSERT(list_empty(&src_device->resized_list));
666 	tgt_device->commit_total_bytes = src_device->commit_total_bytes;
667 	tgt_device->commit_bytes_used = src_device->bytes_used;
668 
669 	btrfs_assign_next_active_device(src_device, tgt_device);
670 
671 	list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
672 	fs_info->fs_devices->rw_devices++;
673 
674 	up_write(&dev_replace->rwsem);
675 	btrfs_rm_dev_replace_blocked(fs_info);
676 
677 	btrfs_rm_dev_replace_remove_srcdev(src_device);
678 
679 	btrfs_rm_dev_replace_unblocked(fs_info);
680 
681 	/*
682 	 * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will
683 	 * update on-disk dev stats value during commit transaction
684 	 */
685 	atomic_inc(&tgt_device->dev_stats_ccnt);
686 
687 	/*
688 	 * this is again a consistent state where no dev_replace procedure
689 	 * is running, the target device is part of the filesystem, the
690 	 * source device is not part of the filesystem anymore and its 1st
691 	 * superblock is scratched out so that it is no longer marked to
692 	 * belong to this filesystem.
693 	 */
694 	mutex_unlock(&fs_info->chunk_mutex);
695 	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
696 
697 	/* replace the sysfs entry */
698 	btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device);
699 	btrfs_rm_dev_replace_free_srcdev(fs_info, src_device);
700 
701 	/* write back the superblocks */
702 	trans = btrfs_start_transaction(root, 0);
703 	if (!IS_ERR(trans))
704 		btrfs_commit_transaction(trans);
705 
706 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
707 
708 	return 0;
709 }
710 
711 static void btrfs_dev_replace_update_device_in_mapping_tree(
712 						struct btrfs_fs_info *fs_info,
713 						struct btrfs_device *srcdev,
714 						struct btrfs_device *tgtdev)
715 {
716 	struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
717 	struct extent_map *em;
718 	struct map_lookup *map;
719 	u64 start = 0;
720 	int i;
721 
722 	write_lock(&em_tree->lock);
723 	do {
724 		em = lookup_extent_mapping(em_tree, start, (u64)-1);
725 		if (!em)
726 			break;
727 		map = em->map_lookup;
728 		for (i = 0; i < map->num_stripes; i++)
729 			if (srcdev == map->stripes[i].dev)
730 				map->stripes[i].dev = tgtdev;
731 		start = em->start + em->len;
732 		free_extent_map(em);
733 	} while (start);
734 	write_unlock(&em_tree->lock);
735 }
736 
737 /*
738  * Read progress of device replace status according to the state and last
739  * stored position. The value format is the same as for
740  * btrfs_dev_replace::progress_1000
741  */
742 static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info)
743 {
744 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
745 	u64 ret = 0;
746 
747 	switch (dev_replace->replace_state) {
748 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
749 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
750 		ret = 0;
751 		break;
752 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
753 		ret = 1000;
754 		break;
755 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
756 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
757 		ret = div64_u64(dev_replace->cursor_left,
758 				div_u64(btrfs_device_get_total_bytes(
759 						dev_replace->srcdev), 1000));
760 		break;
761 	}
762 
763 	return ret;
764 }
765 
766 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
767 			      struct btrfs_ioctl_dev_replace_args *args)
768 {
769 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
770 
771 	down_read(&dev_replace->rwsem);
772 	/* even if !dev_replace_is_valid, the values are good enough for
773 	 * the replace_status ioctl */
774 	args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
775 	args->status.replace_state = dev_replace->replace_state;
776 	args->status.time_started = dev_replace->time_started;
777 	args->status.time_stopped = dev_replace->time_stopped;
778 	args->status.num_write_errors =
779 		atomic64_read(&dev_replace->num_write_errors);
780 	args->status.num_uncorrectable_read_errors =
781 		atomic64_read(&dev_replace->num_uncorrectable_read_errors);
782 	args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
783 	up_read(&dev_replace->rwsem);
784 }
785 
786 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
787 {
788 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
789 	struct btrfs_device *tgt_device = NULL;
790 	struct btrfs_device *src_device = NULL;
791 	struct btrfs_trans_handle *trans;
792 	struct btrfs_root *root = fs_info->tree_root;
793 	int result;
794 	int ret;
795 
796 	if (sb_rdonly(fs_info->sb))
797 		return -EROFS;
798 
799 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
800 	down_write(&dev_replace->rwsem);
801 	switch (dev_replace->replace_state) {
802 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
803 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
804 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
805 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
806 		up_write(&dev_replace->rwsem);
807 		break;
808 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
809 		tgt_device = dev_replace->tgtdev;
810 		src_device = dev_replace->srcdev;
811 		up_write(&dev_replace->rwsem);
812 		ret = btrfs_scrub_cancel(fs_info);
813 		if (ret < 0) {
814 			result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
815 		} else {
816 			result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
817 			/*
818 			 * btrfs_dev_replace_finishing() will handle the
819 			 * cleanup part
820 			 */
821 			btrfs_info_in_rcu(fs_info,
822 				"dev_replace from %s (devid %llu) to %s canceled",
823 				btrfs_dev_name(src_device), src_device->devid,
824 				btrfs_dev_name(tgt_device));
825 		}
826 		break;
827 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
828 		/*
829 		 * Scrub doing the replace isn't running so we need to do the
830 		 * cleanup step of btrfs_dev_replace_finishing() here
831 		 */
832 		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
833 		tgt_device = dev_replace->tgtdev;
834 		src_device = dev_replace->srcdev;
835 		dev_replace->tgtdev = NULL;
836 		dev_replace->srcdev = NULL;
837 		dev_replace->replace_state =
838 				BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
839 		dev_replace->time_stopped = ktime_get_real_seconds();
840 		dev_replace->item_needs_writeback = 1;
841 
842 		up_write(&dev_replace->rwsem);
843 
844 		/* Scrub for replace must not be running in suspended state */
845 		ret = btrfs_scrub_cancel(fs_info);
846 		ASSERT(ret != -ENOTCONN);
847 
848 		trans = btrfs_start_transaction(root, 0);
849 		if (IS_ERR(trans)) {
850 			mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
851 			return PTR_ERR(trans);
852 		}
853 		ret = btrfs_commit_transaction(trans);
854 		WARN_ON(ret);
855 
856 		btrfs_info_in_rcu(fs_info,
857 		"suspended dev_replace from %s (devid %llu) to %s canceled",
858 			btrfs_dev_name(src_device), src_device->devid,
859 			btrfs_dev_name(tgt_device));
860 
861 		if (tgt_device)
862 			btrfs_destroy_dev_replace_tgtdev(tgt_device);
863 		break;
864 	default:
865 		up_write(&dev_replace->rwsem);
866 		result = -EINVAL;
867 	}
868 
869 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
870 	return result;
871 }
872 
873 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
874 {
875 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
876 
877 	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
878 	down_write(&dev_replace->rwsem);
879 
880 	switch (dev_replace->replace_state) {
881 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
882 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
883 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
884 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
885 		break;
886 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
887 		dev_replace->replace_state =
888 			BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
889 		dev_replace->time_stopped = ktime_get_real_seconds();
890 		dev_replace->item_needs_writeback = 1;
891 		btrfs_info(fs_info, "suspending dev_replace for unmount");
892 		break;
893 	}
894 
895 	up_write(&dev_replace->rwsem);
896 	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
897 }
898 
899 /* resume dev_replace procedure that was interrupted by unmount */
900 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
901 {
902 	struct task_struct *task;
903 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
904 
905 	down_write(&dev_replace->rwsem);
906 
907 	switch (dev_replace->replace_state) {
908 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
909 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
910 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
911 		up_write(&dev_replace->rwsem);
912 		return 0;
913 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
914 		break;
915 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
916 		dev_replace->replace_state =
917 			BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
918 		break;
919 	}
920 	if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
921 		btrfs_info(fs_info,
922 			   "cannot continue dev_replace, tgtdev is missing");
923 		btrfs_info(fs_info,
924 			   "you may cancel the operation after 'mount -o degraded'");
925 		dev_replace->replace_state =
926 					BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
927 		up_write(&dev_replace->rwsem);
928 		return 0;
929 	}
930 	up_write(&dev_replace->rwsem);
931 
932 	/*
933 	 * This could collide with a paused balance, but the exclusive op logic
934 	 * should never allow both to start and pause. We don't want to allow
935 	 * dev-replace to start anyway.
936 	 */
937 	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
938 		down_write(&dev_replace->rwsem);
939 		dev_replace->replace_state =
940 					BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
941 		up_write(&dev_replace->rwsem);
942 		btrfs_info(fs_info,
943 		"cannot resume dev-replace, other exclusive operation running");
944 		return 0;
945 	}
946 
947 	task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
948 	return PTR_ERR_OR_ZERO(task);
949 }
950 
951 static int btrfs_dev_replace_kthread(void *data)
952 {
953 	struct btrfs_fs_info *fs_info = data;
954 	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
955 	u64 progress;
956 	int ret;
957 
958 	progress = btrfs_dev_replace_progress(fs_info);
959 	progress = div_u64(progress, 10);
960 	btrfs_info_in_rcu(fs_info,
961 		"continuing dev_replace from %s (devid %llu) to target %s @%u%%",
962 		btrfs_dev_name(dev_replace->srcdev),
963 		dev_replace->srcdev->devid,
964 		btrfs_dev_name(dev_replace->tgtdev),
965 		(unsigned int)progress);
966 
967 	ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
968 			      dev_replace->committed_cursor_left,
969 			      btrfs_device_get_total_bytes(dev_replace->srcdev),
970 			      &dev_replace->scrub_progress, 0, 1);
971 	ret = btrfs_dev_replace_finishing(fs_info, ret);
972 	WARN_ON(ret && ret != -ECANCELED);
973 
974 	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
975 	return 0;
976 }
977 
978 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
979 {
980 	if (!dev_replace->is_valid)
981 		return 0;
982 
983 	switch (dev_replace->replace_state) {
984 	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
985 	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
986 	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
987 		return 0;
988 	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
989 	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
990 		/*
991 		 * return true even if tgtdev is missing (this is
992 		 * something that can happen if the dev_replace
993 		 * procedure is suspended by an umount and then
994 		 * the tgtdev is missing (or "btrfs dev scan") was
995 		 * not called and the filesystem is remounted
996 		 * in degraded state. This does not stop the
997 		 * dev_replace procedure. It needs to be canceled
998 		 * manually if the cancellation is wanted.
999 		 */
1000 		break;
1001 	}
1002 	return 1;
1003 }
1004 
1005 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
1006 {
1007 	percpu_counter_inc(&fs_info->dev_replace.bio_counter);
1008 }
1009 
1010 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
1011 {
1012 	percpu_counter_sub(&fs_info->dev_replace.bio_counter, amount);
1013 	cond_wake_up_nomb(&fs_info->dev_replace.replace_wait);
1014 }
1015 
1016 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
1017 {
1018 	while (1) {
1019 		percpu_counter_inc(&fs_info->dev_replace.bio_counter);
1020 		if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1021 				     &fs_info->fs_state)))
1022 			break;
1023 
1024 		btrfs_bio_counter_dec(fs_info);
1025 		wait_event(fs_info->dev_replace.replace_wait,
1026 			   !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1027 				     &fs_info->fs_state));
1028 	}
1029 }
1030