Loading...
1/*
2 * Copyright (C) STRATO AG 2012. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18#include <linux/sched.h>
19#include <linux/bio.h>
20#include <linux/slab.h>
21#include <linux/buffer_head.h>
22#include <linux/blkdev.h>
23#include <linux/random.h>
24#include <linux/iocontext.h>
25#include <linux/capability.h>
26#include <linux/kthread.h>
27#include <linux/math64.h>
28#include <asm/div64.h>
29#include "ctree.h"
30#include "extent_map.h"
31#include "disk-io.h"
32#include "transaction.h"
33#include "print-tree.h"
34#include "volumes.h"
35#include "async-thread.h"
36#include "check-integrity.h"
37#include "rcu-string.h"
38#include "dev-replace.h"
39#include "sysfs.h"
40
41static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
42 int scrub_ret);
43static void btrfs_dev_replace_update_device_in_mapping_tree(
44 struct btrfs_fs_info *fs_info,
45 struct btrfs_device *srcdev,
46 struct btrfs_device *tgtdev);
47static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info);
48static int btrfs_dev_replace_kthread(void *data);
49static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info);
50
51
52int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
53{
54 struct btrfs_key key;
55 struct btrfs_root *dev_root = fs_info->dev_root;
56 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
57 struct extent_buffer *eb;
58 int slot;
59 int ret = 0;
60 struct btrfs_path *path = NULL;
61 int item_size;
62 struct btrfs_dev_replace_item *ptr;
63 u64 src_devid;
64
65 path = btrfs_alloc_path();
66 if (!path) {
67 ret = -ENOMEM;
68 goto out;
69 }
70
71 key.objectid = 0;
72 key.type = BTRFS_DEV_REPLACE_KEY;
73 key.offset = 0;
74 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
75 if (ret) {
76no_valid_dev_replace_entry_found:
77 ret = 0;
78 dev_replace->replace_state =
79 BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED;
80 dev_replace->cont_reading_from_srcdev_mode =
81 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
82 dev_replace->replace_state = 0;
83 dev_replace->time_started = 0;
84 dev_replace->time_stopped = 0;
85 atomic64_set(&dev_replace->num_write_errors, 0);
86 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
87 dev_replace->cursor_left = 0;
88 dev_replace->committed_cursor_left = 0;
89 dev_replace->cursor_left_last_write_of_item = 0;
90 dev_replace->cursor_right = 0;
91 dev_replace->srcdev = NULL;
92 dev_replace->tgtdev = NULL;
93 dev_replace->is_valid = 0;
94 dev_replace->item_needs_writeback = 0;
95 goto out;
96 }
97 slot = path->slots[0];
98 eb = path->nodes[0];
99 item_size = btrfs_item_size_nr(eb, slot);
100 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
101
102 if (item_size != sizeof(struct btrfs_dev_replace_item)) {
103 btrfs_warn(fs_info,
104 "dev_replace entry found has unexpected size, ignore entry");
105 goto no_valid_dev_replace_entry_found;
106 }
107
108 src_devid = btrfs_dev_replace_src_devid(eb, ptr);
109 dev_replace->cont_reading_from_srcdev_mode =
110 btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
111 dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
112 dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
113 dev_replace->time_stopped =
114 btrfs_dev_replace_time_stopped(eb, ptr);
115 atomic64_set(&dev_replace->num_write_errors,
116 btrfs_dev_replace_num_write_errors(eb, ptr));
117 atomic64_set(&dev_replace->num_uncorrectable_read_errors,
118 btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
119 dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
120 dev_replace->committed_cursor_left = dev_replace->cursor_left;
121 dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
122 dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
123 dev_replace->is_valid = 1;
124
125 dev_replace->item_needs_writeback = 0;
126 switch (dev_replace->replace_state) {
127 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
128 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
129 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
130 dev_replace->srcdev = NULL;
131 dev_replace->tgtdev = NULL;
132 break;
133 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
134 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
135 dev_replace->srcdev = btrfs_find_device(fs_info, src_devid,
136 NULL, NULL);
137 dev_replace->tgtdev = btrfs_find_device(fs_info,
138 BTRFS_DEV_REPLACE_DEVID,
139 NULL, NULL);
140 /*
141 * allow 'btrfs dev replace_cancel' if src/tgt device is
142 * missing
143 */
144 if (!dev_replace->srcdev &&
145 !btrfs_test_opt(fs_info, DEGRADED)) {
146 ret = -EIO;
147 btrfs_warn(fs_info,
148 "cannot mount because device replace operation is ongoing and");
149 btrfs_warn(fs_info,
150 "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
151 src_devid);
152 }
153 if (!dev_replace->tgtdev &&
154 !btrfs_test_opt(fs_info, DEGRADED)) {
155 ret = -EIO;
156 btrfs_warn(fs_info,
157 "cannot mount because device replace operation is ongoing and");
158 btrfs_warn(fs_info,
159 "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
160 BTRFS_DEV_REPLACE_DEVID);
161 }
162 if (dev_replace->tgtdev) {
163 if (dev_replace->srcdev) {
164 dev_replace->tgtdev->total_bytes =
165 dev_replace->srcdev->total_bytes;
166 dev_replace->tgtdev->disk_total_bytes =
167 dev_replace->srcdev->disk_total_bytes;
168 dev_replace->tgtdev->commit_total_bytes =
169 dev_replace->srcdev->commit_total_bytes;
170 dev_replace->tgtdev->bytes_used =
171 dev_replace->srcdev->bytes_used;
172 dev_replace->tgtdev->commit_bytes_used =
173 dev_replace->srcdev->commit_bytes_used;
174 }
175 dev_replace->tgtdev->is_tgtdev_for_dev_replace = 1;
176 btrfs_init_dev_replace_tgtdev_for_resume(fs_info,
177 dev_replace->tgtdev);
178 }
179 break;
180 }
181
182out:
183 btrfs_free_path(path);
184 return ret;
185}
186
187/*
188 * called from commit_transaction. Writes changed device replace state to
189 * disk.
190 */
191int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
192 struct btrfs_fs_info *fs_info)
193{
194 int ret;
195 struct btrfs_root *dev_root = fs_info->dev_root;
196 struct btrfs_path *path;
197 struct btrfs_key key;
198 struct extent_buffer *eb;
199 struct btrfs_dev_replace_item *ptr;
200 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
201
202 btrfs_dev_replace_lock(dev_replace, 0);
203 if (!dev_replace->is_valid ||
204 !dev_replace->item_needs_writeback) {
205 btrfs_dev_replace_unlock(dev_replace, 0);
206 return 0;
207 }
208 btrfs_dev_replace_unlock(dev_replace, 0);
209
210 key.objectid = 0;
211 key.type = BTRFS_DEV_REPLACE_KEY;
212 key.offset = 0;
213
214 path = btrfs_alloc_path();
215 if (!path) {
216 ret = -ENOMEM;
217 goto out;
218 }
219 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
220 if (ret < 0) {
221 btrfs_warn(fs_info,
222 "error %d while searching for dev_replace item!",
223 ret);
224 goto out;
225 }
226
227 if (ret == 0 &&
228 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
229 /*
230 * need to delete old one and insert a new one.
231 * Since no attempt is made to recover any old state, if the
232 * dev_replace state is 'running', the data on the target
233 * drive is lost.
234 * It would be possible to recover the state: just make sure
235 * that the beginning of the item is never changed and always
236 * contains all the essential information. Then read this
237 * minimal set of information and use it as a base for the
238 * new state.
239 */
240 ret = btrfs_del_item(trans, dev_root, path);
241 if (ret != 0) {
242 btrfs_warn(fs_info,
243 "delete too small dev_replace item failed %d!",
244 ret);
245 goto out;
246 }
247 ret = 1;
248 }
249
250 if (ret == 1) {
251 /* need to insert a new item */
252 btrfs_release_path(path);
253 ret = btrfs_insert_empty_item(trans, dev_root, path,
254 &key, sizeof(*ptr));
255 if (ret < 0) {
256 btrfs_warn(fs_info,
257 "insert dev_replace item failed %d!", ret);
258 goto out;
259 }
260 }
261
262 eb = path->nodes[0];
263 ptr = btrfs_item_ptr(eb, path->slots[0],
264 struct btrfs_dev_replace_item);
265
266 btrfs_dev_replace_lock(dev_replace, 1);
267 if (dev_replace->srcdev)
268 btrfs_set_dev_replace_src_devid(eb, ptr,
269 dev_replace->srcdev->devid);
270 else
271 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
272 btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
273 dev_replace->cont_reading_from_srcdev_mode);
274 btrfs_set_dev_replace_replace_state(eb, ptr,
275 dev_replace->replace_state);
276 btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
277 btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
278 btrfs_set_dev_replace_num_write_errors(eb, ptr,
279 atomic64_read(&dev_replace->num_write_errors));
280 btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
281 atomic64_read(&dev_replace->num_uncorrectable_read_errors));
282 dev_replace->cursor_left_last_write_of_item =
283 dev_replace->cursor_left;
284 btrfs_set_dev_replace_cursor_left(eb, ptr,
285 dev_replace->cursor_left_last_write_of_item);
286 btrfs_set_dev_replace_cursor_right(eb, ptr,
287 dev_replace->cursor_right);
288 dev_replace->item_needs_writeback = 0;
289 btrfs_dev_replace_unlock(dev_replace, 1);
290
291 btrfs_mark_buffer_dirty(eb);
292
293out:
294 btrfs_free_path(path);
295
296 return ret;
297}
298
299void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
300{
301 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
302
303 dev_replace->committed_cursor_left =
304 dev_replace->cursor_left_last_write_of_item;
305}
306
307int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info, char *tgtdev_name,
308 u64 srcdevid, char *srcdev_name, int read_src)
309{
310 struct btrfs_root *root = fs_info->dev_root;
311 struct btrfs_trans_handle *trans;
312 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
313 int ret;
314 struct btrfs_device *tgt_device = NULL;
315 struct btrfs_device *src_device = NULL;
316
317 /* the disk copy procedure reuses the scrub code */
318 mutex_lock(&fs_info->volume_mutex);
319 ret = btrfs_find_device_by_devspec(fs_info, srcdevid,
320 srcdev_name, &src_device);
321 if (ret) {
322 mutex_unlock(&fs_info->volume_mutex);
323 return ret;
324 }
325
326 ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
327 src_device, &tgt_device);
328 mutex_unlock(&fs_info->volume_mutex);
329 if (ret)
330 return ret;
331
332 /*
333 * Here we commit the transaction to make sure commit_total_bytes
334 * of all the devices are updated.
335 */
336 trans = btrfs_attach_transaction(root);
337 if (!IS_ERR(trans)) {
338 ret = btrfs_commit_transaction(trans);
339 if (ret)
340 return ret;
341 } else if (PTR_ERR(trans) != -ENOENT) {
342 return PTR_ERR(trans);
343 }
344
345 btrfs_dev_replace_lock(dev_replace, 1);
346 switch (dev_replace->replace_state) {
347 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
348 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
349 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
350 break;
351 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
352 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
353 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
354 goto leave;
355 }
356
357 dev_replace->cont_reading_from_srcdev_mode = read_src;
358 WARN_ON(!src_device);
359 dev_replace->srcdev = src_device;
360 WARN_ON(!tgt_device);
361 dev_replace->tgtdev = tgt_device;
362
363 btrfs_info_in_rcu(fs_info,
364 "dev_replace from %s (devid %llu) to %s started",
365 src_device->missing ? "<missing disk>" :
366 rcu_str_deref(src_device->name),
367 src_device->devid,
368 rcu_str_deref(tgt_device->name));
369
370 /*
371 * from now on, the writes to the srcdev are all duplicated to
372 * go to the tgtdev as well (refer to btrfs_map_block()).
373 */
374 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
375 dev_replace->time_started = get_seconds();
376 dev_replace->cursor_left = 0;
377 dev_replace->committed_cursor_left = 0;
378 dev_replace->cursor_left_last_write_of_item = 0;
379 dev_replace->cursor_right = 0;
380 dev_replace->is_valid = 1;
381 dev_replace->item_needs_writeback = 1;
382 atomic64_set(&dev_replace->num_write_errors, 0);
383 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
384 btrfs_dev_replace_unlock(dev_replace, 1);
385
386 ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device);
387 if (ret)
388 btrfs_err(fs_info, "kobj add dev failed %d", ret);
389
390 btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
391
392 /* force writing the updated state information to disk */
393 trans = btrfs_start_transaction(root, 0);
394 if (IS_ERR(trans)) {
395 ret = PTR_ERR(trans);
396 btrfs_dev_replace_lock(dev_replace, 1);
397 goto leave;
398 }
399
400 ret = btrfs_commit_transaction(trans);
401 WARN_ON(ret);
402
403 /* the disk copy procedure reuses the scrub code */
404 ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
405 btrfs_device_get_total_bytes(src_device),
406 &dev_replace->scrub_progress, 0, 1);
407
408 ret = btrfs_dev_replace_finishing(fs_info, ret);
409 if (ret == -EINPROGRESS) {
410 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
411 } else {
412 WARN_ON(ret);
413 }
414
415 return ret;
416
417leave:
418 dev_replace->srcdev = NULL;
419 dev_replace->tgtdev = NULL;
420 btrfs_dev_replace_unlock(dev_replace, 1);
421 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
422 return ret;
423}
424
425int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
426 struct btrfs_ioctl_dev_replace_args *args)
427{
428 int ret;
429
430 switch (args->start.cont_reading_from_srcdev_mode) {
431 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
432 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
433 break;
434 default:
435 return -EINVAL;
436 }
437
438 if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
439 args->start.tgtdev_name[0] == '\0')
440 return -EINVAL;
441
442 ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name,
443 args->start.srcdevid,
444 args->start.srcdev_name,
445 args->start.cont_reading_from_srcdev_mode);
446 args->result = ret;
447 /* don't warn if EINPROGRESS, someone else might be running scrub */
448 if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS)
449 ret = 0;
450
451 return ret;
452}
453
454/*
455 * blocked until all in-flight bios operations are finished.
456 */
457static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
458{
459 set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
460 wait_event(fs_info->replace_wait, !percpu_counter_sum(
461 &fs_info->bio_counter));
462}
463
464/*
465 * we have removed target device, it is safe to allow new bios request.
466 */
467static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
468{
469 clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
470 wake_up(&fs_info->replace_wait);
471}
472
473static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
474 int scrub_ret)
475{
476 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
477 struct btrfs_device *tgt_device;
478 struct btrfs_device *src_device;
479 struct btrfs_root *root = fs_info->tree_root;
480 u8 uuid_tmp[BTRFS_UUID_SIZE];
481 struct btrfs_trans_handle *trans;
482 int ret = 0;
483
484 /* don't allow cancel or unmount to disturb the finishing procedure */
485 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
486
487 btrfs_dev_replace_lock(dev_replace, 0);
488 /* was the operation canceled, or is it finished? */
489 if (dev_replace->replace_state !=
490 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
491 btrfs_dev_replace_unlock(dev_replace, 0);
492 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
493 return 0;
494 }
495
496 tgt_device = dev_replace->tgtdev;
497 src_device = dev_replace->srcdev;
498 btrfs_dev_replace_unlock(dev_replace, 0);
499
500 /*
501 * flush all outstanding I/O and inode extent mappings before the
502 * copy operation is declared as being finished
503 */
504 ret = btrfs_start_delalloc_roots(fs_info, 0, -1);
505 if (ret) {
506 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
507 return ret;
508 }
509 btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
510
511 trans = btrfs_start_transaction(root, 0);
512 if (IS_ERR(trans)) {
513 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
514 return PTR_ERR(trans);
515 }
516 ret = btrfs_commit_transaction(trans);
517 WARN_ON(ret);
518
519 mutex_lock(&uuid_mutex);
520 /* keep away write_all_supers() during the finishing procedure */
521 mutex_lock(&fs_info->fs_devices->device_list_mutex);
522 mutex_lock(&fs_info->chunk_mutex);
523 btrfs_dev_replace_lock(dev_replace, 1);
524 dev_replace->replace_state =
525 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
526 : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
527 dev_replace->tgtdev = NULL;
528 dev_replace->srcdev = NULL;
529 dev_replace->time_stopped = get_seconds();
530 dev_replace->item_needs_writeback = 1;
531
532 /* replace old device with new one in mapping tree */
533 if (!scrub_ret) {
534 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
535 src_device,
536 tgt_device);
537 } else {
538 btrfs_err_in_rcu(fs_info,
539 "btrfs_scrub_dev(%s, %llu, %s) failed %d",
540 src_device->missing ? "<missing disk>" :
541 rcu_str_deref(src_device->name),
542 src_device->devid,
543 rcu_str_deref(tgt_device->name), scrub_ret);
544 btrfs_dev_replace_unlock(dev_replace, 1);
545 mutex_unlock(&fs_info->chunk_mutex);
546 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
547 mutex_unlock(&uuid_mutex);
548 if (tgt_device)
549 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
550 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
551
552 return scrub_ret;
553 }
554
555 btrfs_info_in_rcu(fs_info,
556 "dev_replace from %s (devid %llu) to %s finished",
557 src_device->missing ? "<missing disk>" :
558 rcu_str_deref(src_device->name),
559 src_device->devid,
560 rcu_str_deref(tgt_device->name));
561 tgt_device->is_tgtdev_for_dev_replace = 0;
562 tgt_device->devid = src_device->devid;
563 src_device->devid = BTRFS_DEV_REPLACE_DEVID;
564 memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
565 memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
566 memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
567 btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
568 btrfs_device_set_disk_total_bytes(tgt_device,
569 src_device->disk_total_bytes);
570 btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
571 ASSERT(list_empty(&src_device->resized_list));
572 tgt_device->commit_total_bytes = src_device->commit_total_bytes;
573 tgt_device->commit_bytes_used = src_device->bytes_used;
574
575 btrfs_assign_next_active_device(fs_info, src_device, tgt_device);
576
577 list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
578 fs_info->fs_devices->rw_devices++;
579
580 btrfs_dev_replace_unlock(dev_replace, 1);
581
582 btrfs_rm_dev_replace_blocked(fs_info);
583
584 btrfs_rm_dev_replace_remove_srcdev(fs_info, src_device);
585
586 btrfs_rm_dev_replace_unblocked(fs_info);
587
588 /*
589 * this is again a consistent state where no dev_replace procedure
590 * is running, the target device is part of the filesystem, the
591 * source device is not part of the filesystem anymore and its 1st
592 * superblock is scratched out so that it is no longer marked to
593 * belong to this filesystem.
594 */
595 mutex_unlock(&fs_info->chunk_mutex);
596 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
597 mutex_unlock(&uuid_mutex);
598
599 /* replace the sysfs entry */
600 btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device);
601 btrfs_rm_dev_replace_free_srcdev(fs_info, src_device);
602
603 /* write back the superblocks */
604 trans = btrfs_start_transaction(root, 0);
605 if (!IS_ERR(trans))
606 btrfs_commit_transaction(trans);
607
608 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
609
610 return 0;
611}
612
613static void btrfs_dev_replace_update_device_in_mapping_tree(
614 struct btrfs_fs_info *fs_info,
615 struct btrfs_device *srcdev,
616 struct btrfs_device *tgtdev)
617{
618 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
619 struct extent_map *em;
620 struct map_lookup *map;
621 u64 start = 0;
622 int i;
623
624 write_lock(&em_tree->lock);
625 do {
626 em = lookup_extent_mapping(em_tree, start, (u64)-1);
627 if (!em)
628 break;
629 map = em->map_lookup;
630 for (i = 0; i < map->num_stripes; i++)
631 if (srcdev == map->stripes[i].dev)
632 map->stripes[i].dev = tgtdev;
633 start = em->start + em->len;
634 free_extent_map(em);
635 } while (start);
636 write_unlock(&em_tree->lock);
637}
638
639void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
640 struct btrfs_ioctl_dev_replace_args *args)
641{
642 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
643 struct btrfs_device *srcdev;
644
645 btrfs_dev_replace_lock(dev_replace, 0);
646 /* even if !dev_replace_is_valid, the values are good enough for
647 * the replace_status ioctl */
648 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
649 args->status.replace_state = dev_replace->replace_state;
650 args->status.time_started = dev_replace->time_started;
651 args->status.time_stopped = dev_replace->time_stopped;
652 args->status.num_write_errors =
653 atomic64_read(&dev_replace->num_write_errors);
654 args->status.num_uncorrectable_read_errors =
655 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
656 switch (dev_replace->replace_state) {
657 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
658 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
659 args->status.progress_1000 = 0;
660 break;
661 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
662 args->status.progress_1000 = 1000;
663 break;
664 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
665 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
666 srcdev = dev_replace->srcdev;
667 args->status.progress_1000 = div_u64(dev_replace->cursor_left,
668 div_u64(btrfs_device_get_total_bytes(srcdev), 1000));
669 break;
670 }
671 btrfs_dev_replace_unlock(dev_replace, 0);
672}
673
674int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info,
675 struct btrfs_ioctl_dev_replace_args *args)
676{
677 args->result = __btrfs_dev_replace_cancel(fs_info);
678 return 0;
679}
680
681static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
682{
683 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
684 struct btrfs_device *tgt_device = NULL;
685 struct btrfs_trans_handle *trans;
686 struct btrfs_root *root = fs_info->tree_root;
687 u64 result;
688 int ret;
689
690 if (fs_info->sb->s_flags & MS_RDONLY)
691 return -EROFS;
692
693 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
694 btrfs_dev_replace_lock(dev_replace, 1);
695 switch (dev_replace->replace_state) {
696 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
697 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
698 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
699 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
700 btrfs_dev_replace_unlock(dev_replace, 1);
701 goto leave;
702 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
703 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
704 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
705 tgt_device = dev_replace->tgtdev;
706 dev_replace->tgtdev = NULL;
707 dev_replace->srcdev = NULL;
708 break;
709 }
710 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
711 dev_replace->time_stopped = get_seconds();
712 dev_replace->item_needs_writeback = 1;
713 btrfs_dev_replace_unlock(dev_replace, 1);
714 btrfs_scrub_cancel(fs_info);
715
716 trans = btrfs_start_transaction(root, 0);
717 if (IS_ERR(trans)) {
718 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
719 return PTR_ERR(trans);
720 }
721 ret = btrfs_commit_transaction(trans);
722 WARN_ON(ret);
723 if (tgt_device)
724 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
725
726leave:
727 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
728 return result;
729}
730
731void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
732{
733 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
734
735 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
736 btrfs_dev_replace_lock(dev_replace, 1);
737 switch (dev_replace->replace_state) {
738 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
739 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
740 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
741 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
742 break;
743 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
744 dev_replace->replace_state =
745 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
746 dev_replace->time_stopped = get_seconds();
747 dev_replace->item_needs_writeback = 1;
748 btrfs_info(fs_info, "suspending dev_replace for unmount");
749 break;
750 }
751
752 btrfs_dev_replace_unlock(dev_replace, 1);
753 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
754}
755
756/* resume dev_replace procedure that was interrupted by unmount */
757int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
758{
759 struct task_struct *task;
760 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
761
762 btrfs_dev_replace_lock(dev_replace, 1);
763 switch (dev_replace->replace_state) {
764 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
765 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
766 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
767 btrfs_dev_replace_unlock(dev_replace, 1);
768 return 0;
769 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
770 break;
771 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
772 dev_replace->replace_state =
773 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
774 break;
775 }
776 if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
777 btrfs_info(fs_info,
778 "cannot continue dev_replace, tgtdev is missing");
779 btrfs_info(fs_info,
780 "you may cancel the operation after 'mount -o degraded'");
781 btrfs_dev_replace_unlock(dev_replace, 1);
782 return 0;
783 }
784 btrfs_dev_replace_unlock(dev_replace, 1);
785
786 WARN_ON(atomic_xchg(
787 &fs_info->mutually_exclusive_operation_running, 1));
788 task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
789 return PTR_ERR_OR_ZERO(task);
790}
791
792static int btrfs_dev_replace_kthread(void *data)
793{
794 struct btrfs_fs_info *fs_info = data;
795 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
796 struct btrfs_ioctl_dev_replace_args *status_args;
797 u64 progress;
798
799 status_args = kzalloc(sizeof(*status_args), GFP_KERNEL);
800 if (status_args) {
801 btrfs_dev_replace_status(fs_info, status_args);
802 progress = status_args->status.progress_1000;
803 kfree(status_args);
804 progress = div_u64(progress, 10);
805 btrfs_info_in_rcu(fs_info,
806 "continuing dev_replace from %s (devid %llu) to %s @%u%%",
807 dev_replace->srcdev->missing ? "<missing disk>" :
808 rcu_str_deref(dev_replace->srcdev->name),
809 dev_replace->srcdev->devid,
810 dev_replace->tgtdev ?
811 rcu_str_deref(dev_replace->tgtdev->name) :
812 "<missing target disk>",
813 (unsigned int)progress);
814 }
815 btrfs_dev_replace_continue_on_mount(fs_info);
816 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
817
818 return 0;
819}
820
821static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info)
822{
823 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
824 int ret;
825
826 ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
827 dev_replace->committed_cursor_left,
828 btrfs_device_get_total_bytes(dev_replace->srcdev),
829 &dev_replace->scrub_progress, 0, 1);
830 ret = btrfs_dev_replace_finishing(fs_info, ret);
831 WARN_ON(ret);
832 return 0;
833}
834
835int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
836{
837 if (!dev_replace->is_valid)
838 return 0;
839
840 switch (dev_replace->replace_state) {
841 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
842 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
843 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
844 return 0;
845 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
846 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
847 /*
848 * return true even if tgtdev is missing (this is
849 * something that can happen if the dev_replace
850 * procedure is suspended by an umount and then
851 * the tgtdev is missing (or "btrfs dev scan") was
852 * not called and the the filesystem is remounted
853 * in degraded state. This does not stop the
854 * dev_replace procedure. It needs to be canceled
855 * manually if the cancellation is wanted.
856 */
857 break;
858 }
859 return 1;
860}
861
862void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace, int rw)
863{
864 if (rw == 1) {
865 /* write */
866again:
867 wait_event(dev_replace->read_lock_wq,
868 atomic_read(&dev_replace->blocking_readers) == 0);
869 write_lock(&dev_replace->lock);
870 if (atomic_read(&dev_replace->blocking_readers)) {
871 write_unlock(&dev_replace->lock);
872 goto again;
873 }
874 } else {
875 read_lock(&dev_replace->lock);
876 atomic_inc(&dev_replace->read_locks);
877 }
878}
879
880void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace, int rw)
881{
882 if (rw == 1) {
883 /* write */
884 ASSERT(atomic_read(&dev_replace->blocking_readers) == 0);
885 write_unlock(&dev_replace->lock);
886 } else {
887 ASSERT(atomic_read(&dev_replace->read_locks) > 0);
888 atomic_dec(&dev_replace->read_locks);
889 read_unlock(&dev_replace->lock);
890 }
891}
892
893/* inc blocking cnt and release read lock */
894void btrfs_dev_replace_set_lock_blocking(
895 struct btrfs_dev_replace *dev_replace)
896{
897 /* only set blocking for read lock */
898 ASSERT(atomic_read(&dev_replace->read_locks) > 0);
899 atomic_inc(&dev_replace->blocking_readers);
900 read_unlock(&dev_replace->lock);
901}
902
903/* acquire read lock and dec blocking cnt */
904void btrfs_dev_replace_clear_lock_blocking(
905 struct btrfs_dev_replace *dev_replace)
906{
907 /* only set blocking for read lock */
908 ASSERT(atomic_read(&dev_replace->read_locks) > 0);
909 ASSERT(atomic_read(&dev_replace->blocking_readers) > 0);
910 read_lock(&dev_replace->lock);
911 if (atomic_dec_and_test(&dev_replace->blocking_readers) &&
912 waitqueue_active(&dev_replace->read_lock_wq))
913 wake_up(&dev_replace->read_lock_wq);
914}
915
916void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
917{
918 percpu_counter_inc(&fs_info->bio_counter);
919}
920
921void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
922{
923 percpu_counter_sub(&fs_info->bio_counter, amount);
924
925 if (waitqueue_active(&fs_info->replace_wait))
926 wake_up(&fs_info->replace_wait);
927}
928
929void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
930{
931 while (1) {
932 percpu_counter_inc(&fs_info->bio_counter);
933 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
934 &fs_info->fs_state)))
935 break;
936
937 btrfs_bio_counter_dec(fs_info);
938 wait_event(fs_info->replace_wait,
939 !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
940 &fs_info->fs_state));
941 }
942}
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 "misc.h"
13#include "ctree.h"
14#include "extent_map.h"
15#include "disk-io.h"
16#include "transaction.h"
17#include "print-tree.h"
18#include "volumes.h"
19#include "async-thread.h"
20#include "check-integrity.h"
21#include "rcu-string.h"
22#include "dev-replace.h"
23#include "sysfs.h"
24
25/*
26 * Device replace overview
27 *
28 * [Objective]
29 * To copy all extents (both new and on-disk) from source device to target
30 * device, while still keeping the filesystem read-write.
31 *
32 * [Method]
33 * There are two main methods involved:
34 *
35 * - Write duplication
36 *
37 * All new writes will be written to both target and source devices, so even
38 * if replace gets canceled, sources device still contans up-to-date data.
39 *
40 * Location: handle_ops_on_dev_replace() from __btrfs_map_block()
41 * Start: btrfs_dev_replace_start()
42 * End: btrfs_dev_replace_finishing()
43 * Content: Latest data/metadata
44 *
45 * - Copy existing extents
46 *
47 * This happens by re-using scrub facility, as scrub also iterates through
48 * existing extents from commit root.
49 *
50 * Location: scrub_write_block_to_dev_replace() from
51 * scrub_block_complete()
52 * Content: Data/meta from commit root.
53 *
54 * Due to the content difference, we need to avoid nocow write when dev-replace
55 * is happening. This is done by marking the block group read-only and waiting
56 * for NOCOW writes.
57 *
58 * After replace is done, the finishing part is done by swapping the target and
59 * source devices.
60 *
61 * Location: btrfs_dev_replace_update_device_in_mapping_tree() from
62 * btrfs_dev_replace_finishing()
63 */
64
65static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
66 int scrub_ret);
67static void btrfs_dev_replace_update_device_in_mapping_tree(
68 struct btrfs_fs_info *fs_info,
69 struct btrfs_device *srcdev,
70 struct btrfs_device *tgtdev);
71static int btrfs_dev_replace_kthread(void *data);
72
73int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
74{
75 struct btrfs_key key;
76 struct btrfs_root *dev_root = fs_info->dev_root;
77 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
78 struct extent_buffer *eb;
79 int slot;
80 int ret = 0;
81 struct btrfs_path *path = NULL;
82 int item_size;
83 struct btrfs_dev_replace_item *ptr;
84 u64 src_devid;
85
86 path = btrfs_alloc_path();
87 if (!path) {
88 ret = -ENOMEM;
89 goto out;
90 }
91
92 key.objectid = 0;
93 key.type = BTRFS_DEV_REPLACE_KEY;
94 key.offset = 0;
95 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
96 if (ret) {
97no_valid_dev_replace_entry_found:
98 ret = 0;
99 dev_replace->replace_state =
100 BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
101 dev_replace->cont_reading_from_srcdev_mode =
102 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
103 dev_replace->time_started = 0;
104 dev_replace->time_stopped = 0;
105 atomic64_set(&dev_replace->num_write_errors, 0);
106 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
107 dev_replace->cursor_left = 0;
108 dev_replace->committed_cursor_left = 0;
109 dev_replace->cursor_left_last_write_of_item = 0;
110 dev_replace->cursor_right = 0;
111 dev_replace->srcdev = NULL;
112 dev_replace->tgtdev = NULL;
113 dev_replace->is_valid = 0;
114 dev_replace->item_needs_writeback = 0;
115 goto out;
116 }
117 slot = path->slots[0];
118 eb = path->nodes[0];
119 item_size = btrfs_item_size_nr(eb, slot);
120 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
121
122 if (item_size != sizeof(struct btrfs_dev_replace_item)) {
123 btrfs_warn(fs_info,
124 "dev_replace entry found has unexpected size, ignore entry");
125 goto no_valid_dev_replace_entry_found;
126 }
127
128 src_devid = btrfs_dev_replace_src_devid(eb, ptr);
129 dev_replace->cont_reading_from_srcdev_mode =
130 btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
131 dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
132 dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
133 dev_replace->time_stopped =
134 btrfs_dev_replace_time_stopped(eb, ptr);
135 atomic64_set(&dev_replace->num_write_errors,
136 btrfs_dev_replace_num_write_errors(eb, ptr));
137 atomic64_set(&dev_replace->num_uncorrectable_read_errors,
138 btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
139 dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
140 dev_replace->committed_cursor_left = dev_replace->cursor_left;
141 dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
142 dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
143 dev_replace->is_valid = 1;
144
145 dev_replace->item_needs_writeback = 0;
146 switch (dev_replace->replace_state) {
147 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
148 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
149 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
150 dev_replace->srcdev = NULL;
151 dev_replace->tgtdev = NULL;
152 break;
153 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
154 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
155 dev_replace->srcdev = btrfs_find_device(fs_info->fs_devices,
156 src_devid, NULL, NULL, true);
157 dev_replace->tgtdev = btrfs_find_device(fs_info->fs_devices,
158 BTRFS_DEV_REPLACE_DEVID,
159 NULL, NULL, true);
160 /*
161 * allow 'btrfs dev replace_cancel' if src/tgt device is
162 * missing
163 */
164 if (!dev_replace->srcdev &&
165 !btrfs_test_opt(fs_info, DEGRADED)) {
166 ret = -EIO;
167 btrfs_warn(fs_info,
168 "cannot mount because device replace operation is ongoing and");
169 btrfs_warn(fs_info,
170 "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
171 src_devid);
172 }
173 if (!dev_replace->tgtdev &&
174 !btrfs_test_opt(fs_info, DEGRADED)) {
175 ret = -EIO;
176 btrfs_warn(fs_info,
177 "cannot mount because device replace operation is ongoing and");
178 btrfs_warn(fs_info,
179 "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
180 BTRFS_DEV_REPLACE_DEVID);
181 }
182 if (dev_replace->tgtdev) {
183 if (dev_replace->srcdev) {
184 dev_replace->tgtdev->total_bytes =
185 dev_replace->srcdev->total_bytes;
186 dev_replace->tgtdev->disk_total_bytes =
187 dev_replace->srcdev->disk_total_bytes;
188 dev_replace->tgtdev->commit_total_bytes =
189 dev_replace->srcdev->commit_total_bytes;
190 dev_replace->tgtdev->bytes_used =
191 dev_replace->srcdev->bytes_used;
192 dev_replace->tgtdev->commit_bytes_used =
193 dev_replace->srcdev->commit_bytes_used;
194 }
195 set_bit(BTRFS_DEV_STATE_REPLACE_TGT,
196 &dev_replace->tgtdev->dev_state);
197
198 WARN_ON(fs_info->fs_devices->rw_devices == 0);
199 dev_replace->tgtdev->io_width = fs_info->sectorsize;
200 dev_replace->tgtdev->io_align = fs_info->sectorsize;
201 dev_replace->tgtdev->sector_size = fs_info->sectorsize;
202 dev_replace->tgtdev->fs_info = fs_info;
203 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
204 &dev_replace->tgtdev->dev_state);
205 }
206 break;
207 }
208
209out:
210 btrfs_free_path(path);
211 return ret;
212}
213
214/*
215 * Initialize a new device for device replace target from a given source dev
216 * and path.
217 *
218 * Return 0 and new device in @device_out, otherwise return < 0
219 */
220static int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
221 const char *device_path,
222 struct btrfs_device *srcdev,
223 struct btrfs_device **device_out)
224{
225 struct btrfs_device *device;
226 struct block_device *bdev;
227 struct list_head *devices;
228 struct rcu_string *name;
229 u64 devid = BTRFS_DEV_REPLACE_DEVID;
230 int ret = 0;
231
232 *device_out = NULL;
233 if (fs_info->fs_devices->seeding) {
234 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
235 return -EINVAL;
236 }
237
238 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
239 fs_info->bdev_holder);
240 if (IS_ERR(bdev)) {
241 btrfs_err(fs_info, "target device %s is invalid!", device_path);
242 return PTR_ERR(bdev);
243 }
244
245 sync_blockdev(bdev);
246
247 devices = &fs_info->fs_devices->devices;
248 list_for_each_entry(device, devices, dev_list) {
249 if (device->bdev == bdev) {
250 btrfs_err(fs_info,
251 "target device is in the filesystem!");
252 ret = -EEXIST;
253 goto error;
254 }
255 }
256
257
258 if (i_size_read(bdev->bd_inode) <
259 btrfs_device_get_total_bytes(srcdev)) {
260 btrfs_err(fs_info,
261 "target device is smaller than source device!");
262 ret = -EINVAL;
263 goto error;
264 }
265
266
267 device = btrfs_alloc_device(NULL, &devid, NULL);
268 if (IS_ERR(device)) {
269 ret = PTR_ERR(device);
270 goto error;
271 }
272
273 name = rcu_string_strdup(device_path, GFP_KERNEL);
274 if (!name) {
275 btrfs_free_device(device);
276 ret = -ENOMEM;
277 goto error;
278 }
279 rcu_assign_pointer(device->name, name);
280
281 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
282 device->generation = 0;
283 device->io_width = fs_info->sectorsize;
284 device->io_align = fs_info->sectorsize;
285 device->sector_size = fs_info->sectorsize;
286 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
287 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
288 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
289 device->commit_total_bytes = srcdev->commit_total_bytes;
290 device->commit_bytes_used = device->bytes_used;
291 device->fs_info = fs_info;
292 device->bdev = bdev;
293 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
294 set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
295 device->mode = FMODE_EXCL;
296 device->dev_stats_valid = 1;
297 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
298 device->fs_devices = fs_info->fs_devices;
299
300 mutex_lock(&fs_info->fs_devices->device_list_mutex);
301 list_add(&device->dev_list, &fs_info->fs_devices->devices);
302 fs_info->fs_devices->num_devices++;
303 fs_info->fs_devices->open_devices++;
304 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
305
306 *device_out = device;
307 return 0;
308
309error:
310 blkdev_put(bdev, FMODE_EXCL);
311 return ret;
312}
313
314/*
315 * called from commit_transaction. Writes changed device replace state to
316 * disk.
317 */
318int btrfs_run_dev_replace(struct btrfs_trans_handle *trans)
319{
320 struct btrfs_fs_info *fs_info = trans->fs_info;
321 int ret;
322 struct btrfs_root *dev_root = fs_info->dev_root;
323 struct btrfs_path *path;
324 struct btrfs_key key;
325 struct extent_buffer *eb;
326 struct btrfs_dev_replace_item *ptr;
327 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
328
329 down_read(&dev_replace->rwsem);
330 if (!dev_replace->is_valid ||
331 !dev_replace->item_needs_writeback) {
332 up_read(&dev_replace->rwsem);
333 return 0;
334 }
335 up_read(&dev_replace->rwsem);
336
337 key.objectid = 0;
338 key.type = BTRFS_DEV_REPLACE_KEY;
339 key.offset = 0;
340
341 path = btrfs_alloc_path();
342 if (!path) {
343 ret = -ENOMEM;
344 goto out;
345 }
346 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
347 if (ret < 0) {
348 btrfs_warn(fs_info,
349 "error %d while searching for dev_replace item!",
350 ret);
351 goto out;
352 }
353
354 if (ret == 0 &&
355 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
356 /*
357 * need to delete old one and insert a new one.
358 * Since no attempt is made to recover any old state, if the
359 * dev_replace state is 'running', the data on the target
360 * drive is lost.
361 * It would be possible to recover the state: just make sure
362 * that the beginning of the item is never changed and always
363 * contains all the essential information. Then read this
364 * minimal set of information and use it as a base for the
365 * new state.
366 */
367 ret = btrfs_del_item(trans, dev_root, path);
368 if (ret != 0) {
369 btrfs_warn(fs_info,
370 "delete too small dev_replace item failed %d!",
371 ret);
372 goto out;
373 }
374 ret = 1;
375 }
376
377 if (ret == 1) {
378 /* need to insert a new item */
379 btrfs_release_path(path);
380 ret = btrfs_insert_empty_item(trans, dev_root, path,
381 &key, sizeof(*ptr));
382 if (ret < 0) {
383 btrfs_warn(fs_info,
384 "insert dev_replace item failed %d!", ret);
385 goto out;
386 }
387 }
388
389 eb = path->nodes[0];
390 ptr = btrfs_item_ptr(eb, path->slots[0],
391 struct btrfs_dev_replace_item);
392
393 down_write(&dev_replace->rwsem);
394 if (dev_replace->srcdev)
395 btrfs_set_dev_replace_src_devid(eb, ptr,
396 dev_replace->srcdev->devid);
397 else
398 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
399 btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
400 dev_replace->cont_reading_from_srcdev_mode);
401 btrfs_set_dev_replace_replace_state(eb, ptr,
402 dev_replace->replace_state);
403 btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
404 btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
405 btrfs_set_dev_replace_num_write_errors(eb, ptr,
406 atomic64_read(&dev_replace->num_write_errors));
407 btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
408 atomic64_read(&dev_replace->num_uncorrectable_read_errors));
409 dev_replace->cursor_left_last_write_of_item =
410 dev_replace->cursor_left;
411 btrfs_set_dev_replace_cursor_left(eb, ptr,
412 dev_replace->cursor_left_last_write_of_item);
413 btrfs_set_dev_replace_cursor_right(eb, ptr,
414 dev_replace->cursor_right);
415 dev_replace->item_needs_writeback = 0;
416 up_write(&dev_replace->rwsem);
417
418 btrfs_mark_buffer_dirty(eb);
419
420out:
421 btrfs_free_path(path);
422
423 return ret;
424}
425
426static char* btrfs_dev_name(struct btrfs_device *device)
427{
428 if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
429 return "<missing disk>";
430 else
431 return rcu_str_deref(device->name);
432}
433
434static int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
435 const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
436 int read_src)
437{
438 struct btrfs_root *root = fs_info->dev_root;
439 struct btrfs_trans_handle *trans;
440 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
441 int ret;
442 struct btrfs_device *tgt_device = NULL;
443 struct btrfs_device *src_device = NULL;
444
445 src_device = btrfs_find_device_by_devspec(fs_info, srcdevid,
446 srcdev_name);
447 if (IS_ERR(src_device))
448 return PTR_ERR(src_device);
449
450 if (btrfs_pinned_by_swapfile(fs_info, src_device)) {
451 btrfs_warn_in_rcu(fs_info,
452 "cannot replace device %s (devid %llu) due to active swapfile",
453 btrfs_dev_name(src_device), src_device->devid);
454 return -ETXTBSY;
455 }
456
457 /*
458 * Here we commit the transaction to make sure commit_total_bytes
459 * of all the devices are updated.
460 */
461 trans = btrfs_attach_transaction(root);
462 if (!IS_ERR(trans)) {
463 ret = btrfs_commit_transaction(trans);
464 if (ret)
465 return ret;
466 } else if (PTR_ERR(trans) != -ENOENT) {
467 return PTR_ERR(trans);
468 }
469
470 ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
471 src_device, &tgt_device);
472 if (ret)
473 return ret;
474
475 down_write(&dev_replace->rwsem);
476 switch (dev_replace->replace_state) {
477 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
478 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
479 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
480 break;
481 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
482 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
483 ASSERT(0);
484 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
485 up_write(&dev_replace->rwsem);
486 goto leave;
487 }
488
489 dev_replace->cont_reading_from_srcdev_mode = read_src;
490 dev_replace->srcdev = src_device;
491 dev_replace->tgtdev = tgt_device;
492
493 btrfs_info_in_rcu(fs_info,
494 "dev_replace from %s (devid %llu) to %s started",
495 btrfs_dev_name(src_device),
496 src_device->devid,
497 rcu_str_deref(tgt_device->name));
498
499 /*
500 * from now on, the writes to the srcdev are all duplicated to
501 * go to the tgtdev as well (refer to btrfs_map_block()).
502 */
503 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
504 dev_replace->time_started = ktime_get_real_seconds();
505 dev_replace->cursor_left = 0;
506 dev_replace->committed_cursor_left = 0;
507 dev_replace->cursor_left_last_write_of_item = 0;
508 dev_replace->cursor_right = 0;
509 dev_replace->is_valid = 1;
510 dev_replace->item_needs_writeback = 1;
511 atomic64_set(&dev_replace->num_write_errors, 0);
512 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
513 up_write(&dev_replace->rwsem);
514
515 ret = btrfs_sysfs_add_devices_dir(tgt_device->fs_devices, tgt_device);
516 if (ret)
517 btrfs_err(fs_info, "kobj add dev failed %d", ret);
518
519 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
520
521 /* Commit dev_replace state and reserve 1 item for it. */
522 trans = btrfs_start_transaction(root, 1);
523 if (IS_ERR(trans)) {
524 ret = PTR_ERR(trans);
525 down_write(&dev_replace->rwsem);
526 dev_replace->replace_state =
527 BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
528 dev_replace->srcdev = NULL;
529 dev_replace->tgtdev = NULL;
530 up_write(&dev_replace->rwsem);
531 goto leave;
532 }
533
534 ret = btrfs_commit_transaction(trans);
535 WARN_ON(ret);
536
537 /* the disk copy procedure reuses the scrub code */
538 ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
539 btrfs_device_get_total_bytes(src_device),
540 &dev_replace->scrub_progress, 0, 1);
541
542 ret = btrfs_dev_replace_finishing(fs_info, ret);
543 if (ret == -EINPROGRESS)
544 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
545
546 return ret;
547
548leave:
549 btrfs_destroy_dev_replace_tgtdev(tgt_device);
550 return ret;
551}
552
553int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
554 struct btrfs_ioctl_dev_replace_args *args)
555{
556 int ret;
557
558 switch (args->start.cont_reading_from_srcdev_mode) {
559 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
560 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
561 break;
562 default:
563 return -EINVAL;
564 }
565
566 if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
567 args->start.tgtdev_name[0] == '\0')
568 return -EINVAL;
569
570 ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name,
571 args->start.srcdevid,
572 args->start.srcdev_name,
573 args->start.cont_reading_from_srcdev_mode);
574 args->result = ret;
575 /* don't warn if EINPROGRESS, someone else might be running scrub */
576 if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS ||
577 ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR)
578 return 0;
579
580 return ret;
581}
582
583/*
584 * blocked until all in-flight bios operations are finished.
585 */
586static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
587{
588 set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
589 wait_event(fs_info->dev_replace.replace_wait, !percpu_counter_sum(
590 &fs_info->dev_replace.bio_counter));
591}
592
593/*
594 * we have removed target device, it is safe to allow new bios request.
595 */
596static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
597{
598 clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
599 wake_up(&fs_info->dev_replace.replace_wait);
600}
601
602/*
603 * When finishing the device replace, before swapping the source device with the
604 * target device we must update the chunk allocation state in the target device,
605 * as it is empty because replace works by directly copying the chunks and not
606 * through the normal chunk allocation path.
607 */
608static int btrfs_set_target_alloc_state(struct btrfs_device *srcdev,
609 struct btrfs_device *tgtdev)
610{
611 struct extent_state *cached_state = NULL;
612 u64 start = 0;
613 u64 found_start;
614 u64 found_end;
615 int ret = 0;
616
617 lockdep_assert_held(&srcdev->fs_info->chunk_mutex);
618
619 while (!find_first_extent_bit(&srcdev->alloc_state, start,
620 &found_start, &found_end,
621 CHUNK_ALLOCATED, &cached_state)) {
622 ret = set_extent_bits(&tgtdev->alloc_state, found_start,
623 found_end, CHUNK_ALLOCATED);
624 if (ret)
625 break;
626 start = found_end + 1;
627 }
628
629 free_extent_state(cached_state);
630 return ret;
631}
632
633static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
634 int scrub_ret)
635{
636 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
637 struct btrfs_device *tgt_device;
638 struct btrfs_device *src_device;
639 struct btrfs_root *root = fs_info->tree_root;
640 u8 uuid_tmp[BTRFS_UUID_SIZE];
641 struct btrfs_trans_handle *trans;
642 int ret = 0;
643
644 /* don't allow cancel or unmount to disturb the finishing procedure */
645 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
646
647 down_read(&dev_replace->rwsem);
648 /* was the operation canceled, or is it finished? */
649 if (dev_replace->replace_state !=
650 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
651 up_read(&dev_replace->rwsem);
652 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
653 return 0;
654 }
655
656 tgt_device = dev_replace->tgtdev;
657 src_device = dev_replace->srcdev;
658 up_read(&dev_replace->rwsem);
659
660 /*
661 * flush all outstanding I/O and inode extent mappings before the
662 * copy operation is declared as being finished
663 */
664 ret = btrfs_start_delalloc_roots(fs_info, -1);
665 if (ret) {
666 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
667 return ret;
668 }
669 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
670
671 /*
672 * We have to use this loop approach because at this point src_device
673 * has to be available for transaction commit to complete, yet new
674 * chunks shouldn't be allocated on the device.
675 */
676 while (1) {
677 trans = btrfs_start_transaction(root, 0);
678 if (IS_ERR(trans)) {
679 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
680 return PTR_ERR(trans);
681 }
682 ret = btrfs_commit_transaction(trans);
683 WARN_ON(ret);
684
685 /* Prevent write_all_supers() during the finishing procedure */
686 mutex_lock(&fs_info->fs_devices->device_list_mutex);
687 /* Prevent new chunks being allocated on the source device */
688 mutex_lock(&fs_info->chunk_mutex);
689
690 if (!list_empty(&src_device->post_commit_list)) {
691 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
692 mutex_unlock(&fs_info->chunk_mutex);
693 } else {
694 break;
695 }
696 }
697
698 down_write(&dev_replace->rwsem);
699 dev_replace->replace_state =
700 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
701 : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
702 dev_replace->tgtdev = NULL;
703 dev_replace->srcdev = NULL;
704 dev_replace->time_stopped = ktime_get_real_seconds();
705 dev_replace->item_needs_writeback = 1;
706
707 /*
708 * Update allocation state in the new device and replace the old device
709 * with the new one in the mapping tree.
710 */
711 if (!scrub_ret) {
712 scrub_ret = btrfs_set_target_alloc_state(src_device, tgt_device);
713 if (scrub_ret)
714 goto error;
715 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
716 src_device,
717 tgt_device);
718 } else {
719 if (scrub_ret != -ECANCELED)
720 btrfs_err_in_rcu(fs_info,
721 "btrfs_scrub_dev(%s, %llu, %s) failed %d",
722 btrfs_dev_name(src_device),
723 src_device->devid,
724 rcu_str_deref(tgt_device->name), scrub_ret);
725error:
726 up_write(&dev_replace->rwsem);
727 mutex_unlock(&fs_info->chunk_mutex);
728 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
729 btrfs_rm_dev_replace_blocked(fs_info);
730 if (tgt_device)
731 btrfs_destroy_dev_replace_tgtdev(tgt_device);
732 btrfs_rm_dev_replace_unblocked(fs_info);
733 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
734
735 return scrub_ret;
736 }
737
738 btrfs_info_in_rcu(fs_info,
739 "dev_replace from %s (devid %llu) to %s finished",
740 btrfs_dev_name(src_device),
741 src_device->devid,
742 rcu_str_deref(tgt_device->name));
743 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state);
744 tgt_device->devid = src_device->devid;
745 src_device->devid = BTRFS_DEV_REPLACE_DEVID;
746 memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
747 memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
748 memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
749 btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
750 btrfs_device_set_disk_total_bytes(tgt_device,
751 src_device->disk_total_bytes);
752 btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
753 tgt_device->commit_bytes_used = src_device->bytes_used;
754
755 btrfs_assign_next_active_device(src_device, tgt_device);
756
757 list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
758 fs_info->fs_devices->rw_devices++;
759
760 up_write(&dev_replace->rwsem);
761 btrfs_rm_dev_replace_blocked(fs_info);
762
763 btrfs_rm_dev_replace_remove_srcdev(src_device);
764
765 btrfs_rm_dev_replace_unblocked(fs_info);
766
767 /*
768 * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will
769 * update on-disk dev stats value during commit transaction
770 */
771 atomic_inc(&tgt_device->dev_stats_ccnt);
772
773 /*
774 * this is again a consistent state where no dev_replace procedure
775 * is running, the target device is part of the filesystem, the
776 * source device is not part of the filesystem anymore and its 1st
777 * superblock is scratched out so that it is no longer marked to
778 * belong to this filesystem.
779 */
780 mutex_unlock(&fs_info->chunk_mutex);
781 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
782
783 /* replace the sysfs entry */
784 btrfs_sysfs_remove_devices_dir(fs_info->fs_devices, src_device);
785 btrfs_sysfs_update_devid(tgt_device);
786 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &src_device->dev_state))
787 btrfs_scratch_superblocks(fs_info, src_device->bdev,
788 src_device->name->str);
789
790 /* write back the superblocks */
791 trans = btrfs_start_transaction(root, 0);
792 if (!IS_ERR(trans))
793 btrfs_commit_transaction(trans);
794
795 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
796
797 btrfs_rm_dev_replace_free_srcdev(src_device);
798
799 return 0;
800}
801
802static void btrfs_dev_replace_update_device_in_mapping_tree(
803 struct btrfs_fs_info *fs_info,
804 struct btrfs_device *srcdev,
805 struct btrfs_device *tgtdev)
806{
807 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
808 struct extent_map *em;
809 struct map_lookup *map;
810 u64 start = 0;
811 int i;
812
813 write_lock(&em_tree->lock);
814 do {
815 em = lookup_extent_mapping(em_tree, start, (u64)-1);
816 if (!em)
817 break;
818 map = em->map_lookup;
819 for (i = 0; i < map->num_stripes; i++)
820 if (srcdev == map->stripes[i].dev)
821 map->stripes[i].dev = tgtdev;
822 start = em->start + em->len;
823 free_extent_map(em);
824 } while (start);
825 write_unlock(&em_tree->lock);
826}
827
828/*
829 * Read progress of device replace status according to the state and last
830 * stored position. The value format is the same as for
831 * btrfs_dev_replace::progress_1000
832 */
833static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info)
834{
835 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
836 u64 ret = 0;
837
838 switch (dev_replace->replace_state) {
839 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
840 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
841 ret = 0;
842 break;
843 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
844 ret = 1000;
845 break;
846 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
847 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
848 ret = div64_u64(dev_replace->cursor_left,
849 div_u64(btrfs_device_get_total_bytes(
850 dev_replace->srcdev), 1000));
851 break;
852 }
853
854 return ret;
855}
856
857void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
858 struct btrfs_ioctl_dev_replace_args *args)
859{
860 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
861
862 down_read(&dev_replace->rwsem);
863 /* even if !dev_replace_is_valid, the values are good enough for
864 * the replace_status ioctl */
865 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
866 args->status.replace_state = dev_replace->replace_state;
867 args->status.time_started = dev_replace->time_started;
868 args->status.time_stopped = dev_replace->time_stopped;
869 args->status.num_write_errors =
870 atomic64_read(&dev_replace->num_write_errors);
871 args->status.num_uncorrectable_read_errors =
872 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
873 args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
874 up_read(&dev_replace->rwsem);
875}
876
877int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
878{
879 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
880 struct btrfs_device *tgt_device = NULL;
881 struct btrfs_device *src_device = NULL;
882 struct btrfs_trans_handle *trans;
883 struct btrfs_root *root = fs_info->tree_root;
884 int result;
885 int ret;
886
887 if (sb_rdonly(fs_info->sb))
888 return -EROFS;
889
890 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
891 down_write(&dev_replace->rwsem);
892 switch (dev_replace->replace_state) {
893 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
894 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
895 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
896 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
897 up_write(&dev_replace->rwsem);
898 break;
899 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
900 tgt_device = dev_replace->tgtdev;
901 src_device = dev_replace->srcdev;
902 up_write(&dev_replace->rwsem);
903 ret = btrfs_scrub_cancel(fs_info);
904 if (ret < 0) {
905 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
906 } else {
907 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
908 /*
909 * btrfs_dev_replace_finishing() will handle the
910 * cleanup part
911 */
912 btrfs_info_in_rcu(fs_info,
913 "dev_replace from %s (devid %llu) to %s canceled",
914 btrfs_dev_name(src_device), src_device->devid,
915 btrfs_dev_name(tgt_device));
916 }
917 break;
918 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
919 /*
920 * Scrub doing the replace isn't running so we need to do the
921 * cleanup step of btrfs_dev_replace_finishing() here
922 */
923 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
924 tgt_device = dev_replace->tgtdev;
925 src_device = dev_replace->srcdev;
926 dev_replace->tgtdev = NULL;
927 dev_replace->srcdev = NULL;
928 dev_replace->replace_state =
929 BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
930 dev_replace->time_stopped = ktime_get_real_seconds();
931 dev_replace->item_needs_writeback = 1;
932
933 up_write(&dev_replace->rwsem);
934
935 /* Scrub for replace must not be running in suspended state */
936 ret = btrfs_scrub_cancel(fs_info);
937 ASSERT(ret != -ENOTCONN);
938
939 trans = btrfs_start_transaction(root, 0);
940 if (IS_ERR(trans)) {
941 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
942 return PTR_ERR(trans);
943 }
944 ret = btrfs_commit_transaction(trans);
945 WARN_ON(ret);
946
947 btrfs_info_in_rcu(fs_info,
948 "suspended dev_replace from %s (devid %llu) to %s canceled",
949 btrfs_dev_name(src_device), src_device->devid,
950 btrfs_dev_name(tgt_device));
951
952 if (tgt_device)
953 btrfs_destroy_dev_replace_tgtdev(tgt_device);
954 break;
955 default:
956 up_write(&dev_replace->rwsem);
957 result = -EINVAL;
958 }
959
960 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
961 return result;
962}
963
964void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
965{
966 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
967
968 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
969 down_write(&dev_replace->rwsem);
970
971 switch (dev_replace->replace_state) {
972 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
973 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
974 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
975 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
976 break;
977 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
978 dev_replace->replace_state =
979 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
980 dev_replace->time_stopped = ktime_get_real_seconds();
981 dev_replace->item_needs_writeback = 1;
982 btrfs_info(fs_info, "suspending dev_replace for unmount");
983 break;
984 }
985
986 up_write(&dev_replace->rwsem);
987 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
988}
989
990/* resume dev_replace procedure that was interrupted by unmount */
991int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
992{
993 struct task_struct *task;
994 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
995
996 down_write(&dev_replace->rwsem);
997
998 switch (dev_replace->replace_state) {
999 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1000 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1001 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1002 up_write(&dev_replace->rwsem);
1003 return 0;
1004 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1005 break;
1006 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1007 dev_replace->replace_state =
1008 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
1009 break;
1010 }
1011 if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
1012 btrfs_info(fs_info,
1013 "cannot continue dev_replace, tgtdev is missing");
1014 btrfs_info(fs_info,
1015 "you may cancel the operation after 'mount -o degraded'");
1016 dev_replace->replace_state =
1017 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
1018 up_write(&dev_replace->rwsem);
1019 return 0;
1020 }
1021 up_write(&dev_replace->rwsem);
1022
1023 /*
1024 * This could collide with a paused balance, but the exclusive op logic
1025 * should never allow both to start and pause. We don't want to allow
1026 * dev-replace to start anyway.
1027 */
1028 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
1029 down_write(&dev_replace->rwsem);
1030 dev_replace->replace_state =
1031 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
1032 up_write(&dev_replace->rwsem);
1033 btrfs_info(fs_info,
1034 "cannot resume dev-replace, other exclusive operation running");
1035 return 0;
1036 }
1037
1038 task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
1039 return PTR_ERR_OR_ZERO(task);
1040}
1041
1042static int btrfs_dev_replace_kthread(void *data)
1043{
1044 struct btrfs_fs_info *fs_info = data;
1045 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1046 u64 progress;
1047 int ret;
1048
1049 progress = btrfs_dev_replace_progress(fs_info);
1050 progress = div_u64(progress, 10);
1051 btrfs_info_in_rcu(fs_info,
1052 "continuing dev_replace from %s (devid %llu) to target %s @%u%%",
1053 btrfs_dev_name(dev_replace->srcdev),
1054 dev_replace->srcdev->devid,
1055 btrfs_dev_name(dev_replace->tgtdev),
1056 (unsigned int)progress);
1057
1058 ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
1059 dev_replace->committed_cursor_left,
1060 btrfs_device_get_total_bytes(dev_replace->srcdev),
1061 &dev_replace->scrub_progress, 0, 1);
1062 ret = btrfs_dev_replace_finishing(fs_info, ret);
1063 WARN_ON(ret && ret != -ECANCELED);
1064
1065 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
1066 return 0;
1067}
1068
1069int __pure btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
1070{
1071 if (!dev_replace->is_valid)
1072 return 0;
1073
1074 switch (dev_replace->replace_state) {
1075 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1076 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1077 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1078 return 0;
1079 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1080 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1081 /*
1082 * return true even if tgtdev is missing (this is
1083 * something that can happen if the dev_replace
1084 * procedure is suspended by an umount and then
1085 * the tgtdev is missing (or "btrfs dev scan") was
1086 * not called and the filesystem is remounted
1087 * in degraded state. This does not stop the
1088 * dev_replace procedure. It needs to be canceled
1089 * manually if the cancellation is wanted.
1090 */
1091 break;
1092 }
1093 return 1;
1094}
1095
1096void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
1097{
1098 percpu_counter_inc(&fs_info->dev_replace.bio_counter);
1099}
1100
1101void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
1102{
1103 percpu_counter_sub(&fs_info->dev_replace.bio_counter, amount);
1104 cond_wake_up_nomb(&fs_info->dev_replace.replace_wait);
1105}
1106
1107void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
1108{
1109 while (1) {
1110 percpu_counter_inc(&fs_info->dev_replace.bio_counter);
1111 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1112 &fs_info->fs_state)))
1113 break;
1114
1115 btrfs_bio_counter_dec(fs_info);
1116 wait_event(fs_info->dev_replace.replace_wait,
1117 !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1118 &fs_info->fs_state));
1119 }
1120}