Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) STRATO AG 2012. All rights reserved.
4 */
5
6#include <linux/sched.h>
7#include <linux/bio.h>
8#include <linux/slab.h>
9#include <linux/buffer_head.h>
10#include <linux/blkdev.h>
11#include <linux/random.h>
12#include <linux/iocontext.h>
13#include <linux/capability.h>
14#include <linux/kthread.h>
15#include <linux/math64.h>
16#include <asm/div64.h>
17#include "ctree.h"
18#include "extent_map.h"
19#include "disk-io.h"
20#include "transaction.h"
21#include "print-tree.h"
22#include "volumes.h"
23#include "async-thread.h"
24#include "check-integrity.h"
25#include "rcu-string.h"
26#include "dev-replace.h"
27#include "sysfs.h"
28
29static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
30 int scrub_ret);
31static void btrfs_dev_replace_update_device_in_mapping_tree(
32 struct btrfs_fs_info *fs_info,
33 struct btrfs_device *srcdev,
34 struct btrfs_device *tgtdev);
35static int btrfs_dev_replace_kthread(void *data);
36static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info);
37
38
39int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
40{
41 struct btrfs_key key;
42 struct btrfs_root *dev_root = fs_info->dev_root;
43 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
44 struct extent_buffer *eb;
45 int slot;
46 int ret = 0;
47 struct btrfs_path *path = NULL;
48 int item_size;
49 struct btrfs_dev_replace_item *ptr;
50 u64 src_devid;
51
52 path = btrfs_alloc_path();
53 if (!path) {
54 ret = -ENOMEM;
55 goto out;
56 }
57
58 key.objectid = 0;
59 key.type = BTRFS_DEV_REPLACE_KEY;
60 key.offset = 0;
61 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
62 if (ret) {
63no_valid_dev_replace_entry_found:
64 ret = 0;
65 dev_replace->replace_state =
66 BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED;
67 dev_replace->cont_reading_from_srcdev_mode =
68 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
69 dev_replace->replace_state = 0;
70 dev_replace->time_started = 0;
71 dev_replace->time_stopped = 0;
72 atomic64_set(&dev_replace->num_write_errors, 0);
73 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
74 dev_replace->cursor_left = 0;
75 dev_replace->committed_cursor_left = 0;
76 dev_replace->cursor_left_last_write_of_item = 0;
77 dev_replace->cursor_right = 0;
78 dev_replace->srcdev = NULL;
79 dev_replace->tgtdev = NULL;
80 dev_replace->is_valid = 0;
81 dev_replace->item_needs_writeback = 0;
82 goto out;
83 }
84 slot = path->slots[0];
85 eb = path->nodes[0];
86 item_size = btrfs_item_size_nr(eb, slot);
87 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
88
89 if (item_size != sizeof(struct btrfs_dev_replace_item)) {
90 btrfs_warn(fs_info,
91 "dev_replace entry found has unexpected size, ignore entry");
92 goto no_valid_dev_replace_entry_found;
93 }
94
95 src_devid = btrfs_dev_replace_src_devid(eb, ptr);
96 dev_replace->cont_reading_from_srcdev_mode =
97 btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
98 dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
99 dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
100 dev_replace->time_stopped =
101 btrfs_dev_replace_time_stopped(eb, ptr);
102 atomic64_set(&dev_replace->num_write_errors,
103 btrfs_dev_replace_num_write_errors(eb, ptr));
104 atomic64_set(&dev_replace->num_uncorrectable_read_errors,
105 btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
106 dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
107 dev_replace->committed_cursor_left = dev_replace->cursor_left;
108 dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
109 dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
110 dev_replace->is_valid = 1;
111
112 dev_replace->item_needs_writeback = 0;
113 switch (dev_replace->replace_state) {
114 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
115 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
116 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
117 dev_replace->srcdev = NULL;
118 dev_replace->tgtdev = NULL;
119 break;
120 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
121 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
122 dev_replace->srcdev = btrfs_find_device(fs_info, src_devid,
123 NULL, NULL);
124 dev_replace->tgtdev = btrfs_find_device(fs_info,
125 BTRFS_DEV_REPLACE_DEVID,
126 NULL, NULL);
127 /*
128 * allow 'btrfs dev replace_cancel' if src/tgt device is
129 * missing
130 */
131 if (!dev_replace->srcdev &&
132 !btrfs_test_opt(fs_info, DEGRADED)) {
133 ret = -EIO;
134 btrfs_warn(fs_info,
135 "cannot mount because device replace operation is ongoing and");
136 btrfs_warn(fs_info,
137 "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
138 src_devid);
139 }
140 if (!dev_replace->tgtdev &&
141 !btrfs_test_opt(fs_info, DEGRADED)) {
142 ret = -EIO;
143 btrfs_warn(fs_info,
144 "cannot mount because device replace operation is ongoing and");
145 btrfs_warn(fs_info,
146 "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
147 BTRFS_DEV_REPLACE_DEVID);
148 }
149 if (dev_replace->tgtdev) {
150 if (dev_replace->srcdev) {
151 dev_replace->tgtdev->total_bytes =
152 dev_replace->srcdev->total_bytes;
153 dev_replace->tgtdev->disk_total_bytes =
154 dev_replace->srcdev->disk_total_bytes;
155 dev_replace->tgtdev->commit_total_bytes =
156 dev_replace->srcdev->commit_total_bytes;
157 dev_replace->tgtdev->bytes_used =
158 dev_replace->srcdev->bytes_used;
159 dev_replace->tgtdev->commit_bytes_used =
160 dev_replace->srcdev->commit_bytes_used;
161 }
162 set_bit(BTRFS_DEV_STATE_REPLACE_TGT,
163 &dev_replace->tgtdev->dev_state);
164
165 WARN_ON(fs_info->fs_devices->rw_devices == 0);
166 dev_replace->tgtdev->io_width = fs_info->sectorsize;
167 dev_replace->tgtdev->io_align = fs_info->sectorsize;
168 dev_replace->tgtdev->sector_size = fs_info->sectorsize;
169 dev_replace->tgtdev->fs_info = fs_info;
170 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
171 &dev_replace->tgtdev->dev_state);
172 }
173 break;
174 }
175
176out:
177 btrfs_free_path(path);
178 return ret;
179}
180
181/*
182 * called from commit_transaction. Writes changed device replace state to
183 * disk.
184 */
185int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
186 struct btrfs_fs_info *fs_info)
187{
188 int ret;
189 struct btrfs_root *dev_root = fs_info->dev_root;
190 struct btrfs_path *path;
191 struct btrfs_key key;
192 struct extent_buffer *eb;
193 struct btrfs_dev_replace_item *ptr;
194 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
195
196 btrfs_dev_replace_read_lock(dev_replace);
197 if (!dev_replace->is_valid ||
198 !dev_replace->item_needs_writeback) {
199 btrfs_dev_replace_read_unlock(dev_replace);
200 return 0;
201 }
202 btrfs_dev_replace_read_unlock(dev_replace);
203
204 key.objectid = 0;
205 key.type = BTRFS_DEV_REPLACE_KEY;
206 key.offset = 0;
207
208 path = btrfs_alloc_path();
209 if (!path) {
210 ret = -ENOMEM;
211 goto out;
212 }
213 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
214 if (ret < 0) {
215 btrfs_warn(fs_info,
216 "error %d while searching for dev_replace item!",
217 ret);
218 goto out;
219 }
220
221 if (ret == 0 &&
222 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
223 /*
224 * need to delete old one and insert a new one.
225 * Since no attempt is made to recover any old state, if the
226 * dev_replace state is 'running', the data on the target
227 * drive is lost.
228 * It would be possible to recover the state: just make sure
229 * that the beginning of the item is never changed and always
230 * contains all the essential information. Then read this
231 * minimal set of information and use it as a base for the
232 * new state.
233 */
234 ret = btrfs_del_item(trans, dev_root, path);
235 if (ret != 0) {
236 btrfs_warn(fs_info,
237 "delete too small dev_replace item failed %d!",
238 ret);
239 goto out;
240 }
241 ret = 1;
242 }
243
244 if (ret == 1) {
245 /* need to insert a new item */
246 btrfs_release_path(path);
247 ret = btrfs_insert_empty_item(trans, dev_root, path,
248 &key, sizeof(*ptr));
249 if (ret < 0) {
250 btrfs_warn(fs_info,
251 "insert dev_replace item failed %d!", ret);
252 goto out;
253 }
254 }
255
256 eb = path->nodes[0];
257 ptr = btrfs_item_ptr(eb, path->slots[0],
258 struct btrfs_dev_replace_item);
259
260 btrfs_dev_replace_write_lock(dev_replace);
261 if (dev_replace->srcdev)
262 btrfs_set_dev_replace_src_devid(eb, ptr,
263 dev_replace->srcdev->devid);
264 else
265 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
266 btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
267 dev_replace->cont_reading_from_srcdev_mode);
268 btrfs_set_dev_replace_replace_state(eb, ptr,
269 dev_replace->replace_state);
270 btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
271 btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
272 btrfs_set_dev_replace_num_write_errors(eb, ptr,
273 atomic64_read(&dev_replace->num_write_errors));
274 btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
275 atomic64_read(&dev_replace->num_uncorrectable_read_errors));
276 dev_replace->cursor_left_last_write_of_item =
277 dev_replace->cursor_left;
278 btrfs_set_dev_replace_cursor_left(eb, ptr,
279 dev_replace->cursor_left_last_write_of_item);
280 btrfs_set_dev_replace_cursor_right(eb, ptr,
281 dev_replace->cursor_right);
282 dev_replace->item_needs_writeback = 0;
283 btrfs_dev_replace_write_unlock(dev_replace);
284
285 btrfs_mark_buffer_dirty(eb);
286
287out:
288 btrfs_free_path(path);
289
290 return ret;
291}
292
293void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
294{
295 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
296
297 dev_replace->committed_cursor_left =
298 dev_replace->cursor_left_last_write_of_item;
299}
300
301static char* btrfs_dev_name(struct btrfs_device *device)
302{
303 if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
304 return "<missing disk>";
305 else
306 return rcu_str_deref(device->name);
307}
308
309int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
310 const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
311 int read_src)
312{
313 struct btrfs_root *root = fs_info->dev_root;
314 struct btrfs_trans_handle *trans;
315 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
316 int ret;
317 struct btrfs_device *tgt_device = NULL;
318 struct btrfs_device *src_device = NULL;
319
320 /* the disk copy procedure reuses the scrub code */
321 mutex_lock(&fs_info->volume_mutex);
322 ret = btrfs_find_device_by_devspec(fs_info, srcdevid,
323 srcdev_name, &src_device);
324 if (ret) {
325 mutex_unlock(&fs_info->volume_mutex);
326 return ret;
327 }
328
329 ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
330 src_device, &tgt_device);
331 mutex_unlock(&fs_info->volume_mutex);
332 if (ret)
333 return ret;
334
335 /*
336 * Here we commit the transaction to make sure commit_total_bytes
337 * of all the devices are updated.
338 */
339 trans = btrfs_attach_transaction(root);
340 if (!IS_ERR(trans)) {
341 ret = btrfs_commit_transaction(trans);
342 if (ret)
343 return ret;
344 } else if (PTR_ERR(trans) != -ENOENT) {
345 return PTR_ERR(trans);
346 }
347
348 btrfs_dev_replace_write_lock(dev_replace);
349 switch (dev_replace->replace_state) {
350 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
351 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
352 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
353 break;
354 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
355 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
356 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
357 goto leave;
358 }
359
360 dev_replace->cont_reading_from_srcdev_mode = read_src;
361 WARN_ON(!src_device);
362 dev_replace->srcdev = src_device;
363 WARN_ON(!tgt_device);
364 dev_replace->tgtdev = tgt_device;
365
366 btrfs_info_in_rcu(fs_info,
367 "dev_replace from %s (devid %llu) to %s started",
368 btrfs_dev_name(src_device),
369 src_device->devid,
370 rcu_str_deref(tgt_device->name));
371
372 /*
373 * from now on, the writes to the srcdev are all duplicated to
374 * go to the tgtdev as well (refer to btrfs_map_block()).
375 */
376 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
377 dev_replace->time_started = get_seconds();
378 dev_replace->cursor_left = 0;
379 dev_replace->committed_cursor_left = 0;
380 dev_replace->cursor_left_last_write_of_item = 0;
381 dev_replace->cursor_right = 0;
382 dev_replace->is_valid = 1;
383 dev_replace->item_needs_writeback = 1;
384 atomic64_set(&dev_replace->num_write_errors, 0);
385 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
386 btrfs_dev_replace_write_unlock(dev_replace);
387
388 ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device);
389 if (ret)
390 btrfs_err(fs_info, "kobj add dev failed %d", ret);
391
392 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
393
394 /* force writing the updated state information to disk */
395 trans = btrfs_start_transaction(root, 0);
396 if (IS_ERR(trans)) {
397 ret = PTR_ERR(trans);
398 btrfs_dev_replace_write_lock(dev_replace);
399 goto leave;
400 }
401
402 ret = btrfs_commit_transaction(trans);
403 WARN_ON(ret);
404
405 /* the disk copy procedure reuses the scrub code */
406 ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
407 btrfs_device_get_total_bytes(src_device),
408 &dev_replace->scrub_progress, 0, 1);
409
410 ret = btrfs_dev_replace_finishing(fs_info, ret);
411 if (ret == -EINPROGRESS) {
412 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
413 } else {
414 WARN_ON(ret);
415 }
416
417 return ret;
418
419leave:
420 dev_replace->srcdev = NULL;
421 dev_replace->tgtdev = NULL;
422 btrfs_dev_replace_write_unlock(dev_replace);
423 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
424 return ret;
425}
426
427int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
428 struct btrfs_ioctl_dev_replace_args *args)
429{
430 int ret;
431
432 switch (args->start.cont_reading_from_srcdev_mode) {
433 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
434 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
435 break;
436 default:
437 return -EINVAL;
438 }
439
440 if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
441 args->start.tgtdev_name[0] == '\0')
442 return -EINVAL;
443
444 ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name,
445 args->start.srcdevid,
446 args->start.srcdev_name,
447 args->start.cont_reading_from_srcdev_mode);
448 args->result = ret;
449 /* don't warn if EINPROGRESS, someone else might be running scrub */
450 if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS)
451 ret = 0;
452
453 return ret;
454}
455
456/*
457 * blocked until all in-flight bios operations are finished.
458 */
459static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
460{
461 set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
462 wait_event(fs_info->replace_wait, !percpu_counter_sum(
463 &fs_info->bio_counter));
464}
465
466/*
467 * we have removed target device, it is safe to allow new bios request.
468 */
469static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
470{
471 clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
472 wake_up(&fs_info->replace_wait);
473}
474
475static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
476 int scrub_ret)
477{
478 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
479 struct btrfs_device *tgt_device;
480 struct btrfs_device *src_device;
481 struct btrfs_root *root = fs_info->tree_root;
482 u8 uuid_tmp[BTRFS_UUID_SIZE];
483 struct btrfs_trans_handle *trans;
484 int ret = 0;
485
486 /* don't allow cancel or unmount to disturb the finishing procedure */
487 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
488
489 btrfs_dev_replace_read_lock(dev_replace);
490 /* was the operation canceled, or is it finished? */
491 if (dev_replace->replace_state !=
492 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
493 btrfs_dev_replace_read_unlock(dev_replace);
494 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
495 return 0;
496 }
497
498 tgt_device = dev_replace->tgtdev;
499 src_device = dev_replace->srcdev;
500 btrfs_dev_replace_read_unlock(dev_replace);
501
502 /*
503 * flush all outstanding I/O and inode extent mappings before the
504 * copy operation is declared as being finished
505 */
506 ret = btrfs_start_delalloc_roots(fs_info, 0, -1);
507 if (ret) {
508 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
509 return ret;
510 }
511 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
512
513 trans = btrfs_start_transaction(root, 0);
514 if (IS_ERR(trans)) {
515 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
516 return PTR_ERR(trans);
517 }
518 ret = btrfs_commit_transaction(trans);
519 WARN_ON(ret);
520
521 mutex_lock(&uuid_mutex);
522 /* keep away write_all_supers() during the finishing procedure */
523 mutex_lock(&fs_info->fs_devices->device_list_mutex);
524 mutex_lock(&fs_info->chunk_mutex);
525 btrfs_dev_replace_write_lock(dev_replace);
526 dev_replace->replace_state =
527 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
528 : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
529 dev_replace->tgtdev = NULL;
530 dev_replace->srcdev = NULL;
531 dev_replace->time_stopped = get_seconds();
532 dev_replace->item_needs_writeback = 1;
533
534 /* replace old device with new one in mapping tree */
535 if (!scrub_ret) {
536 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
537 src_device,
538 tgt_device);
539 } else {
540 btrfs_err_in_rcu(fs_info,
541 "btrfs_scrub_dev(%s, %llu, %s) failed %d",
542 btrfs_dev_name(src_device),
543 src_device->devid,
544 rcu_str_deref(tgt_device->name), scrub_ret);
545 btrfs_dev_replace_write_unlock(dev_replace);
546 mutex_unlock(&fs_info->chunk_mutex);
547 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
548 mutex_unlock(&uuid_mutex);
549 btrfs_rm_dev_replace_blocked(fs_info);
550 if (tgt_device)
551 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
552 btrfs_rm_dev_replace_unblocked(fs_info);
553 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
554
555 return scrub_ret;
556 }
557
558 btrfs_info_in_rcu(fs_info,
559 "dev_replace from %s (devid %llu) to %s finished",
560 btrfs_dev_name(src_device),
561 src_device->devid,
562 rcu_str_deref(tgt_device->name));
563 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state);
564 tgt_device->devid = src_device->devid;
565 src_device->devid = BTRFS_DEV_REPLACE_DEVID;
566 memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
567 memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
568 memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
569 btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
570 btrfs_device_set_disk_total_bytes(tgt_device,
571 src_device->disk_total_bytes);
572 btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
573 ASSERT(list_empty(&src_device->resized_list));
574 tgt_device->commit_total_bytes = src_device->commit_total_bytes;
575 tgt_device->commit_bytes_used = src_device->bytes_used;
576
577 btrfs_assign_next_active_device(fs_info, src_device, tgt_device);
578
579 list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
580 fs_info->fs_devices->rw_devices++;
581
582 btrfs_dev_replace_write_unlock(dev_replace);
583
584 btrfs_rm_dev_replace_blocked(fs_info);
585
586 btrfs_rm_dev_replace_remove_srcdev(fs_info, src_device);
587
588 btrfs_rm_dev_replace_unblocked(fs_info);
589
590 /*
591 * this is again a consistent state where no dev_replace procedure
592 * is running, the target device is part of the filesystem, the
593 * source device is not part of the filesystem anymore and its 1st
594 * superblock is scratched out so that it is no longer marked to
595 * belong to this filesystem.
596 */
597 mutex_unlock(&fs_info->chunk_mutex);
598 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
599 mutex_unlock(&uuid_mutex);
600
601 /* replace the sysfs entry */
602 btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device);
603 btrfs_rm_dev_replace_free_srcdev(fs_info, src_device);
604
605 /* write back the superblocks */
606 trans = btrfs_start_transaction(root, 0);
607 if (!IS_ERR(trans))
608 btrfs_commit_transaction(trans);
609
610 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
611
612 return 0;
613}
614
615static void btrfs_dev_replace_update_device_in_mapping_tree(
616 struct btrfs_fs_info *fs_info,
617 struct btrfs_device *srcdev,
618 struct btrfs_device *tgtdev)
619{
620 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
621 struct extent_map *em;
622 struct map_lookup *map;
623 u64 start = 0;
624 int i;
625
626 write_lock(&em_tree->lock);
627 do {
628 em = lookup_extent_mapping(em_tree, start, (u64)-1);
629 if (!em)
630 break;
631 map = em->map_lookup;
632 for (i = 0; i < map->num_stripes; i++)
633 if (srcdev == map->stripes[i].dev)
634 map->stripes[i].dev = tgtdev;
635 start = em->start + em->len;
636 free_extent_map(em);
637 } while (start);
638 write_unlock(&em_tree->lock);
639}
640
641/*
642 * Read progress of device replace status according to the state and last
643 * stored position. The value format is the same as for
644 * btrfs_dev_replace::progress_1000
645 */
646static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info)
647{
648 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
649 u64 ret = 0;
650
651 switch (dev_replace->replace_state) {
652 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
653 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
654 ret = 0;
655 break;
656 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
657 ret = 1000;
658 break;
659 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
660 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
661 ret = div64_u64(dev_replace->cursor_left,
662 div_u64(btrfs_device_get_total_bytes(
663 dev_replace->srcdev), 1000));
664 break;
665 }
666
667 return ret;
668}
669
670void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
671 struct btrfs_ioctl_dev_replace_args *args)
672{
673 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
674
675 btrfs_dev_replace_read_lock(dev_replace);
676 /* even if !dev_replace_is_valid, the values are good enough for
677 * the replace_status ioctl */
678 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
679 args->status.replace_state = dev_replace->replace_state;
680 args->status.time_started = dev_replace->time_started;
681 args->status.time_stopped = dev_replace->time_stopped;
682 args->status.num_write_errors =
683 atomic64_read(&dev_replace->num_write_errors);
684 args->status.num_uncorrectable_read_errors =
685 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
686 args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
687 btrfs_dev_replace_read_unlock(dev_replace);
688}
689
690int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
691{
692 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
693 struct btrfs_device *tgt_device = NULL;
694 struct btrfs_device *src_device = NULL;
695 struct btrfs_trans_handle *trans;
696 struct btrfs_root *root = fs_info->tree_root;
697 int result;
698 int ret;
699
700 if (sb_rdonly(fs_info->sb))
701 return -EROFS;
702
703 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
704 btrfs_dev_replace_write_lock(dev_replace);
705 switch (dev_replace->replace_state) {
706 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
707 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
708 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
709 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
710 btrfs_dev_replace_write_unlock(dev_replace);
711 goto leave;
712 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
713 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
714 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
715 tgt_device = dev_replace->tgtdev;
716 src_device = dev_replace->srcdev;
717 dev_replace->tgtdev = NULL;
718 dev_replace->srcdev = NULL;
719 break;
720 }
721 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
722 dev_replace->time_stopped = get_seconds();
723 dev_replace->item_needs_writeback = 1;
724 btrfs_dev_replace_write_unlock(dev_replace);
725 btrfs_scrub_cancel(fs_info);
726
727 trans = btrfs_start_transaction(root, 0);
728 if (IS_ERR(trans)) {
729 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
730 return PTR_ERR(trans);
731 }
732 ret = btrfs_commit_transaction(trans);
733 WARN_ON(ret);
734
735 btrfs_info_in_rcu(fs_info,
736 "dev_replace from %s (devid %llu) to %s canceled",
737 btrfs_dev_name(src_device), src_device->devid,
738 btrfs_dev_name(tgt_device));
739
740 if (tgt_device)
741 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
742
743leave:
744 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
745 return result;
746}
747
748void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
749{
750 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
751
752 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
753 btrfs_dev_replace_write_lock(dev_replace);
754 switch (dev_replace->replace_state) {
755 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
756 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
757 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
758 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
759 break;
760 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
761 dev_replace->replace_state =
762 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
763 dev_replace->time_stopped = get_seconds();
764 dev_replace->item_needs_writeback = 1;
765 btrfs_info(fs_info, "suspending dev_replace for unmount");
766 break;
767 }
768
769 btrfs_dev_replace_write_unlock(dev_replace);
770 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
771}
772
773/* resume dev_replace procedure that was interrupted by unmount */
774int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
775{
776 struct task_struct *task;
777 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
778
779 btrfs_dev_replace_write_lock(dev_replace);
780 switch (dev_replace->replace_state) {
781 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
782 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
783 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
784 btrfs_dev_replace_write_unlock(dev_replace);
785 return 0;
786 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
787 break;
788 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
789 dev_replace->replace_state =
790 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
791 break;
792 }
793 if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
794 btrfs_info(fs_info,
795 "cannot continue dev_replace, tgtdev is missing");
796 btrfs_info(fs_info,
797 "you may cancel the operation after 'mount -o degraded'");
798 btrfs_dev_replace_write_unlock(dev_replace);
799 return 0;
800 }
801 btrfs_dev_replace_write_unlock(dev_replace);
802
803 WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags));
804 task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
805 return PTR_ERR_OR_ZERO(task);
806}
807
808static int btrfs_dev_replace_kthread(void *data)
809{
810 struct btrfs_fs_info *fs_info = data;
811 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
812 u64 progress;
813
814 progress = btrfs_dev_replace_progress(fs_info);
815 progress = div_u64(progress, 10);
816 btrfs_info_in_rcu(fs_info,
817 "continuing dev_replace from %s (devid %llu) to target %s @%u%%",
818 btrfs_dev_name(dev_replace->srcdev),
819 dev_replace->srcdev->devid,
820 btrfs_dev_name(dev_replace->tgtdev),
821 (unsigned int)progress);
822
823 btrfs_dev_replace_continue_on_mount(fs_info);
824 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
825
826 return 0;
827}
828
829static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info)
830{
831 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
832 int ret;
833
834 ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
835 dev_replace->committed_cursor_left,
836 btrfs_device_get_total_bytes(dev_replace->srcdev),
837 &dev_replace->scrub_progress, 0, 1);
838 ret = btrfs_dev_replace_finishing(fs_info, ret);
839 WARN_ON(ret);
840 return 0;
841}
842
843int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
844{
845 if (!dev_replace->is_valid)
846 return 0;
847
848 switch (dev_replace->replace_state) {
849 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
850 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
851 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
852 return 0;
853 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
854 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
855 /*
856 * return true even if tgtdev is missing (this is
857 * something that can happen if the dev_replace
858 * procedure is suspended by an umount and then
859 * the tgtdev is missing (or "btrfs dev scan") was
860 * not called and the the filesystem is remounted
861 * in degraded state. This does not stop the
862 * dev_replace procedure. It needs to be canceled
863 * manually if the cancellation is wanted.
864 */
865 break;
866 }
867 return 1;
868}
869
870void btrfs_dev_replace_read_lock(struct btrfs_dev_replace *dev_replace)
871{
872 read_lock(&dev_replace->lock);
873 atomic_inc(&dev_replace->read_locks);
874}
875
876void btrfs_dev_replace_read_unlock(struct btrfs_dev_replace *dev_replace)
877{
878 ASSERT(atomic_read(&dev_replace->read_locks) > 0);
879 atomic_dec(&dev_replace->read_locks);
880 read_unlock(&dev_replace->lock);
881}
882
883void btrfs_dev_replace_write_lock(struct btrfs_dev_replace *dev_replace)
884{
885again:
886 wait_event(dev_replace->read_lock_wq,
887 atomic_read(&dev_replace->blocking_readers) == 0);
888 write_lock(&dev_replace->lock);
889 if (atomic_read(&dev_replace->blocking_readers)) {
890 write_unlock(&dev_replace->lock);
891 goto again;
892 }
893}
894
895void btrfs_dev_replace_write_unlock(struct btrfs_dev_replace *dev_replace)
896{
897 ASSERT(atomic_read(&dev_replace->blocking_readers) == 0);
898 write_unlock(&dev_replace->lock);
899}
900
901/* inc blocking cnt and release read lock */
902void btrfs_dev_replace_set_lock_blocking(
903 struct btrfs_dev_replace *dev_replace)
904{
905 /* only set blocking for read lock */
906 ASSERT(atomic_read(&dev_replace->read_locks) > 0);
907 atomic_inc(&dev_replace->blocking_readers);
908 read_unlock(&dev_replace->lock);
909}
910
911/* acquire read lock and dec blocking cnt */
912void btrfs_dev_replace_clear_lock_blocking(
913 struct btrfs_dev_replace *dev_replace)
914{
915 /* only set blocking for read lock */
916 ASSERT(atomic_read(&dev_replace->read_locks) > 0);
917 ASSERT(atomic_read(&dev_replace->blocking_readers) > 0);
918 read_lock(&dev_replace->lock);
919 if (atomic_dec_and_test(&dev_replace->blocking_readers) &&
920 waitqueue_active(&dev_replace->read_lock_wq))
921 wake_up(&dev_replace->read_lock_wq);
922}
923
924void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
925{
926 percpu_counter_inc(&fs_info->bio_counter);
927}
928
929void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
930{
931 percpu_counter_sub(&fs_info->bio_counter, amount);
932
933 if (waitqueue_active(&fs_info->replace_wait))
934 wake_up(&fs_info->replace_wait);
935}
936
937void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
938{
939 while (1) {
940 percpu_counter_inc(&fs_info->bio_counter);
941 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
942 &fs_info->fs_state)))
943 break;
944
945 btrfs_bio_counter_dec(fs_info);
946 wait_event(fs_info->replace_wait,
947 !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
948 &fs_info->fs_state));
949 }
950}
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
40static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
41 int scrub_ret);
42static void btrfs_dev_replace_update_device_in_mapping_tree(
43 struct btrfs_fs_info *fs_info,
44 struct btrfs_device *srcdev,
45 struct btrfs_device *tgtdev);
46static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid,
47 char *srcdev_name,
48 struct btrfs_device **device);
49static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info);
50static int btrfs_dev_replace_kthread(void *data);
51static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info);
52
53
54int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
55{
56 struct btrfs_key key;
57 struct btrfs_root *dev_root = fs_info->dev_root;
58 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
59 struct extent_buffer *eb;
60 int slot;
61 int ret = 0;
62 struct btrfs_path *path = NULL;
63 int item_size;
64 struct btrfs_dev_replace_item *ptr;
65 u64 src_devid;
66
67 path = btrfs_alloc_path();
68 if (!path) {
69 ret = -ENOMEM;
70 goto out;
71 }
72
73 key.objectid = 0;
74 key.type = BTRFS_DEV_REPLACE_KEY;
75 key.offset = 0;
76 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
77 if (ret) {
78no_valid_dev_replace_entry_found:
79 ret = 0;
80 dev_replace->replace_state =
81 BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED;
82 dev_replace->cont_reading_from_srcdev_mode =
83 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
84 dev_replace->replace_state = 0;
85 dev_replace->time_started = 0;
86 dev_replace->time_stopped = 0;
87 atomic64_set(&dev_replace->num_write_errors, 0);
88 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
89 dev_replace->cursor_left = 0;
90 dev_replace->committed_cursor_left = 0;
91 dev_replace->cursor_left_last_write_of_item = 0;
92 dev_replace->cursor_right = 0;
93 dev_replace->srcdev = NULL;
94 dev_replace->tgtdev = NULL;
95 dev_replace->is_valid = 0;
96 dev_replace->item_needs_writeback = 0;
97 goto out;
98 }
99 slot = path->slots[0];
100 eb = path->nodes[0];
101 item_size = btrfs_item_size_nr(eb, slot);
102 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
103
104 if (item_size != sizeof(struct btrfs_dev_replace_item)) {
105 btrfs_warn(fs_info,
106 "dev_replace entry found has unexpected size, ignore entry");
107 goto no_valid_dev_replace_entry_found;
108 }
109
110 src_devid = btrfs_dev_replace_src_devid(eb, ptr);
111 dev_replace->cont_reading_from_srcdev_mode =
112 btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
113 dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
114 dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
115 dev_replace->time_stopped =
116 btrfs_dev_replace_time_stopped(eb, ptr);
117 atomic64_set(&dev_replace->num_write_errors,
118 btrfs_dev_replace_num_write_errors(eb, ptr));
119 atomic64_set(&dev_replace->num_uncorrectable_read_errors,
120 btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
121 dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
122 dev_replace->committed_cursor_left = dev_replace->cursor_left;
123 dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
124 dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
125 dev_replace->is_valid = 1;
126
127 dev_replace->item_needs_writeback = 0;
128 switch (dev_replace->replace_state) {
129 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
130 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
131 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
132 dev_replace->srcdev = NULL;
133 dev_replace->tgtdev = NULL;
134 break;
135 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
136 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
137 dev_replace->srcdev = btrfs_find_device(fs_info, src_devid,
138 NULL, NULL);
139 dev_replace->tgtdev = btrfs_find_device(fs_info,
140 BTRFS_DEV_REPLACE_DEVID,
141 NULL, NULL);
142 /*
143 * allow 'btrfs dev replace_cancel' if src/tgt device is
144 * missing
145 */
146 if (!dev_replace->srcdev &&
147 !btrfs_test_opt(dev_root, DEGRADED)) {
148 ret = -EIO;
149 btrfs_warn(fs_info,
150 "cannot mount because device replace operation is ongoing and");
151 btrfs_warn(fs_info,
152 "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
153 src_devid);
154 }
155 if (!dev_replace->tgtdev &&
156 !btrfs_test_opt(dev_root, DEGRADED)) {
157 ret = -EIO;
158 btrfs_warn(fs_info,
159 "cannot mount because device replace operation is ongoing and");
160 btrfs_warn(fs_info,
161 "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
162 BTRFS_DEV_REPLACE_DEVID);
163 }
164 if (dev_replace->tgtdev) {
165 if (dev_replace->srcdev) {
166 dev_replace->tgtdev->total_bytes =
167 dev_replace->srcdev->total_bytes;
168 dev_replace->tgtdev->disk_total_bytes =
169 dev_replace->srcdev->disk_total_bytes;
170 dev_replace->tgtdev->bytes_used =
171 dev_replace->srcdev->bytes_used;
172 }
173 dev_replace->tgtdev->is_tgtdev_for_dev_replace = 1;
174 btrfs_init_dev_replace_tgtdev_for_resume(fs_info,
175 dev_replace->tgtdev);
176 }
177 break;
178 }
179
180out:
181 if (path)
182 btrfs_free_path(path);
183 return ret;
184}
185
186/*
187 * called from commit_transaction. Writes changed device replace state to
188 * disk.
189 */
190int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
191 struct btrfs_fs_info *fs_info)
192{
193 int ret;
194 struct btrfs_root *dev_root = fs_info->dev_root;
195 struct btrfs_path *path;
196 struct btrfs_key key;
197 struct extent_buffer *eb;
198 struct btrfs_dev_replace_item *ptr;
199 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
200
201 btrfs_dev_replace_lock(dev_replace);
202 if (!dev_replace->is_valid ||
203 !dev_replace->item_needs_writeback) {
204 btrfs_dev_replace_unlock(dev_replace);
205 return 0;
206 }
207 btrfs_dev_replace_unlock(dev_replace);
208
209 key.objectid = 0;
210 key.type = BTRFS_DEV_REPLACE_KEY;
211 key.offset = 0;
212
213 path = btrfs_alloc_path();
214 if (!path) {
215 ret = -ENOMEM;
216 goto out;
217 }
218 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
219 if (ret < 0) {
220 btrfs_warn(fs_info, "error %d while searching for dev_replace item!",
221 ret);
222 goto out;
223 }
224
225 if (ret == 0 &&
226 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
227 /*
228 * need to delete old one and insert a new one.
229 * Since no attempt is made to recover any old state, if the
230 * dev_replace state is 'running', the data on the target
231 * drive is lost.
232 * It would be possible to recover the state: just make sure
233 * that the beginning of the item is never changed and always
234 * contains all the essential information. Then read this
235 * minimal set of information and use it as a base for the
236 * new state.
237 */
238 ret = btrfs_del_item(trans, dev_root, path);
239 if (ret != 0) {
240 btrfs_warn(fs_info, "delete too small dev_replace item failed %d!",
241 ret);
242 goto out;
243 }
244 ret = 1;
245 }
246
247 if (ret == 1) {
248 /* need to insert a new item */
249 btrfs_release_path(path);
250 ret = btrfs_insert_empty_item(trans, dev_root, path,
251 &key, sizeof(*ptr));
252 if (ret < 0) {
253 btrfs_warn(fs_info, "insert dev_replace item failed %d!",
254 ret);
255 goto out;
256 }
257 }
258
259 eb = path->nodes[0];
260 ptr = btrfs_item_ptr(eb, path->slots[0],
261 struct btrfs_dev_replace_item);
262
263 btrfs_dev_replace_lock(dev_replace);
264 if (dev_replace->srcdev)
265 btrfs_set_dev_replace_src_devid(eb, ptr,
266 dev_replace->srcdev->devid);
267 else
268 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
269 btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
270 dev_replace->cont_reading_from_srcdev_mode);
271 btrfs_set_dev_replace_replace_state(eb, ptr,
272 dev_replace->replace_state);
273 btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
274 btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
275 btrfs_set_dev_replace_num_write_errors(eb, ptr,
276 atomic64_read(&dev_replace->num_write_errors));
277 btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
278 atomic64_read(&dev_replace->num_uncorrectable_read_errors));
279 dev_replace->cursor_left_last_write_of_item =
280 dev_replace->cursor_left;
281 btrfs_set_dev_replace_cursor_left(eb, ptr,
282 dev_replace->cursor_left_last_write_of_item);
283 btrfs_set_dev_replace_cursor_right(eb, ptr,
284 dev_replace->cursor_right);
285 dev_replace->item_needs_writeback = 0;
286 btrfs_dev_replace_unlock(dev_replace);
287
288 btrfs_mark_buffer_dirty(eb);
289
290out:
291 btrfs_free_path(path);
292
293 return ret;
294}
295
296void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
297{
298 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
299
300 dev_replace->committed_cursor_left =
301 dev_replace->cursor_left_last_write_of_item;
302}
303
304int btrfs_dev_replace_start(struct btrfs_root *root,
305 struct btrfs_ioctl_dev_replace_args *args)
306{
307 struct btrfs_trans_handle *trans;
308 struct btrfs_fs_info *fs_info = root->fs_info;
309 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
310 int ret;
311 struct btrfs_device *tgt_device = NULL;
312 struct btrfs_device *src_device = NULL;
313
314 if (btrfs_fs_incompat(fs_info, RAID56)) {
315 btrfs_warn(fs_info, "dev_replace cannot yet handle RAID5/RAID6");
316 return -EINVAL;
317 }
318
319 switch (args->start.cont_reading_from_srcdev_mode) {
320 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
321 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
322 break;
323 default:
324 return -EINVAL;
325 }
326
327 if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
328 args->start.tgtdev_name[0] == '\0')
329 return -EINVAL;
330
331 mutex_lock(&fs_info->volume_mutex);
332 ret = btrfs_init_dev_replace_tgtdev(root, args->start.tgtdev_name,
333 &tgt_device);
334 if (ret) {
335 btrfs_err(fs_info, "target device %s is invalid!",
336 args->start.tgtdev_name);
337 mutex_unlock(&fs_info->volume_mutex);
338 return -EINVAL;
339 }
340
341 ret = btrfs_dev_replace_find_srcdev(root, args->start.srcdevid,
342 args->start.srcdev_name,
343 &src_device);
344 mutex_unlock(&fs_info->volume_mutex);
345 if (ret) {
346 ret = -EINVAL;
347 goto leave_no_lock;
348 }
349
350 if (tgt_device->total_bytes < src_device->total_bytes) {
351 btrfs_err(fs_info, "target device is smaller than source device!");
352 ret = -EINVAL;
353 goto leave_no_lock;
354 }
355
356 btrfs_dev_replace_lock(dev_replace);
357 switch (dev_replace->replace_state) {
358 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
359 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
360 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
361 break;
362 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
363 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
364 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
365 goto leave;
366 }
367
368 dev_replace->cont_reading_from_srcdev_mode =
369 args->start.cont_reading_from_srcdev_mode;
370 WARN_ON(!src_device);
371 dev_replace->srcdev = src_device;
372 WARN_ON(!tgt_device);
373 dev_replace->tgtdev = tgt_device;
374
375 printk_in_rcu(KERN_INFO
376 "BTRFS: dev_replace from %s (devid %llu) to %s started\n",
377 src_device->missing ? "<missing disk>" :
378 rcu_str_deref(src_device->name),
379 src_device->devid,
380 rcu_str_deref(tgt_device->name));
381
382 tgt_device->total_bytes = src_device->total_bytes;
383 tgt_device->disk_total_bytes = src_device->disk_total_bytes;
384 tgt_device->bytes_used = src_device->bytes_used;
385
386 /*
387 * from now on, the writes to the srcdev are all duplicated to
388 * go to the tgtdev as well (refer to btrfs_map_block()).
389 */
390 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
391 dev_replace->time_started = get_seconds();
392 dev_replace->cursor_left = 0;
393 dev_replace->committed_cursor_left = 0;
394 dev_replace->cursor_left_last_write_of_item = 0;
395 dev_replace->cursor_right = 0;
396 dev_replace->is_valid = 1;
397 dev_replace->item_needs_writeback = 1;
398 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
399 btrfs_dev_replace_unlock(dev_replace);
400
401 btrfs_wait_ordered_roots(root->fs_info, -1);
402
403 /* force writing the updated state information to disk */
404 trans = btrfs_start_transaction(root, 0);
405 if (IS_ERR(trans)) {
406 ret = PTR_ERR(trans);
407 btrfs_dev_replace_lock(dev_replace);
408 goto leave;
409 }
410
411 ret = btrfs_commit_transaction(trans, root);
412 WARN_ON(ret);
413
414 /* the disk copy procedure reuses the scrub code */
415 ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
416 src_device->total_bytes,
417 &dev_replace->scrub_progress, 0, 1);
418
419 ret = btrfs_dev_replace_finishing(root->fs_info, ret);
420 WARN_ON(ret);
421
422 return 0;
423
424leave:
425 dev_replace->srcdev = NULL;
426 dev_replace->tgtdev = NULL;
427 btrfs_dev_replace_unlock(dev_replace);
428leave_no_lock:
429 if (tgt_device)
430 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
431 return ret;
432}
433
434/*
435 * blocked until all flighting bios are finished.
436 */
437static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
438{
439 s64 writers;
440 DEFINE_WAIT(wait);
441
442 set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
443 do {
444 prepare_to_wait(&fs_info->replace_wait, &wait,
445 TASK_UNINTERRUPTIBLE);
446 writers = percpu_counter_sum(&fs_info->bio_counter);
447 if (writers)
448 schedule();
449 finish_wait(&fs_info->replace_wait, &wait);
450 } while (writers);
451}
452
453/*
454 * we have removed target device, it is safe to allow new bios request.
455 */
456static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
457{
458 clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
459 if (waitqueue_active(&fs_info->replace_wait))
460 wake_up(&fs_info->replace_wait);
461}
462
463static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
464 int scrub_ret)
465{
466 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
467 struct btrfs_device *tgt_device;
468 struct btrfs_device *src_device;
469 struct btrfs_root *root = fs_info->tree_root;
470 u8 uuid_tmp[BTRFS_UUID_SIZE];
471 struct btrfs_trans_handle *trans;
472 int ret = 0;
473
474 /* don't allow cancel or unmount to disturb the finishing procedure */
475 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
476
477 btrfs_dev_replace_lock(dev_replace);
478 /* was the operation canceled, or is it finished? */
479 if (dev_replace->replace_state !=
480 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
481 btrfs_dev_replace_unlock(dev_replace);
482 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
483 return 0;
484 }
485
486 tgt_device = dev_replace->tgtdev;
487 src_device = dev_replace->srcdev;
488 btrfs_dev_replace_unlock(dev_replace);
489
490 /*
491 * flush all outstanding I/O and inode extent mappings before the
492 * copy operation is declared as being finished
493 */
494 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
495 if (ret) {
496 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
497 return ret;
498 }
499 btrfs_wait_ordered_roots(root->fs_info, -1);
500
501 trans = btrfs_start_transaction(root, 0);
502 if (IS_ERR(trans)) {
503 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
504 return PTR_ERR(trans);
505 }
506 ret = btrfs_commit_transaction(trans, root);
507 WARN_ON(ret);
508
509 /* keep away write_all_supers() during the finishing procedure */
510 mutex_lock(&root->fs_info->chunk_mutex);
511 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
512 btrfs_dev_replace_lock(dev_replace);
513 dev_replace->replace_state =
514 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
515 : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
516 dev_replace->tgtdev = NULL;
517 dev_replace->srcdev = NULL;
518 dev_replace->time_stopped = get_seconds();
519 dev_replace->item_needs_writeback = 1;
520
521 /* replace old device with new one in mapping tree */
522 if (!scrub_ret) {
523 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
524 src_device,
525 tgt_device);
526 } else {
527 printk_in_rcu(KERN_ERR
528 "BTRFS: btrfs_scrub_dev(%s, %llu, %s) failed %d\n",
529 src_device->missing ? "<missing disk>" :
530 rcu_str_deref(src_device->name),
531 src_device->devid,
532 rcu_str_deref(tgt_device->name), scrub_ret);
533 btrfs_dev_replace_unlock(dev_replace);
534 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
535 mutex_unlock(&root->fs_info->chunk_mutex);
536 if (tgt_device)
537 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
538 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
539
540 return 0;
541 }
542
543 printk_in_rcu(KERN_INFO
544 "BTRFS: dev_replace from %s (devid %llu) to %s) finished\n",
545 src_device->missing ? "<missing disk>" :
546 rcu_str_deref(src_device->name),
547 src_device->devid,
548 rcu_str_deref(tgt_device->name));
549 tgt_device->is_tgtdev_for_dev_replace = 0;
550 tgt_device->devid = src_device->devid;
551 src_device->devid = BTRFS_DEV_REPLACE_DEVID;
552 tgt_device->bytes_used = src_device->bytes_used;
553 memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
554 memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
555 memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
556 tgt_device->total_bytes = src_device->total_bytes;
557 tgt_device->disk_total_bytes = src_device->disk_total_bytes;
558 tgt_device->bytes_used = src_device->bytes_used;
559 if (fs_info->sb->s_bdev == src_device->bdev)
560 fs_info->sb->s_bdev = tgt_device->bdev;
561 if (fs_info->fs_devices->latest_bdev == src_device->bdev)
562 fs_info->fs_devices->latest_bdev = tgt_device->bdev;
563 list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
564
565 btrfs_rm_dev_replace_blocked(fs_info);
566
567 btrfs_rm_dev_replace_srcdev(fs_info, src_device);
568
569 btrfs_rm_dev_replace_unblocked(fs_info);
570
571 /*
572 * this is again a consistent state where no dev_replace procedure
573 * is running, the target device is part of the filesystem, the
574 * source device is not part of the filesystem anymore and its 1st
575 * superblock is scratched out so that it is no longer marked to
576 * belong to this filesystem.
577 */
578 btrfs_dev_replace_unlock(dev_replace);
579 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
580 mutex_unlock(&root->fs_info->chunk_mutex);
581
582 /* write back the superblocks */
583 trans = btrfs_start_transaction(root, 0);
584 if (!IS_ERR(trans))
585 btrfs_commit_transaction(trans, root);
586
587 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
588
589 return 0;
590}
591
592static void btrfs_dev_replace_update_device_in_mapping_tree(
593 struct btrfs_fs_info *fs_info,
594 struct btrfs_device *srcdev,
595 struct btrfs_device *tgtdev)
596{
597 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
598 struct extent_map *em;
599 struct map_lookup *map;
600 u64 start = 0;
601 int i;
602
603 write_lock(&em_tree->lock);
604 do {
605 em = lookup_extent_mapping(em_tree, start, (u64)-1);
606 if (!em)
607 break;
608 map = (struct map_lookup *)em->bdev;
609 for (i = 0; i < map->num_stripes; i++)
610 if (srcdev == map->stripes[i].dev)
611 map->stripes[i].dev = tgtdev;
612 start = em->start + em->len;
613 free_extent_map(em);
614 } while (start);
615 write_unlock(&em_tree->lock);
616}
617
618static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid,
619 char *srcdev_name,
620 struct btrfs_device **device)
621{
622 int ret;
623
624 if (srcdevid) {
625 ret = 0;
626 *device = btrfs_find_device(root->fs_info, srcdevid, NULL,
627 NULL);
628 if (!*device)
629 ret = -ENOENT;
630 } else {
631 ret = btrfs_find_device_missing_or_by_path(root, srcdev_name,
632 device);
633 }
634 return ret;
635}
636
637void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
638 struct btrfs_ioctl_dev_replace_args *args)
639{
640 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
641
642 btrfs_dev_replace_lock(dev_replace);
643 /* even if !dev_replace_is_valid, the values are good enough for
644 * the replace_status ioctl */
645 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
646 args->status.replace_state = dev_replace->replace_state;
647 args->status.time_started = dev_replace->time_started;
648 args->status.time_stopped = dev_replace->time_stopped;
649 args->status.num_write_errors =
650 atomic64_read(&dev_replace->num_write_errors);
651 args->status.num_uncorrectable_read_errors =
652 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
653 switch (dev_replace->replace_state) {
654 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
655 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
656 args->status.progress_1000 = 0;
657 break;
658 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
659 args->status.progress_1000 = 1000;
660 break;
661 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
662 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
663 args->status.progress_1000 = div64_u64(dev_replace->cursor_left,
664 div64_u64(dev_replace->srcdev->total_bytes, 1000));
665 break;
666 }
667 btrfs_dev_replace_unlock(dev_replace);
668}
669
670int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info,
671 struct btrfs_ioctl_dev_replace_args *args)
672{
673 args->result = __btrfs_dev_replace_cancel(fs_info);
674 return 0;
675}
676
677static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
678{
679 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
680 struct btrfs_device *tgt_device = NULL;
681 struct btrfs_trans_handle *trans;
682 struct btrfs_root *root = fs_info->tree_root;
683 u64 result;
684 int ret;
685
686 if (fs_info->sb->s_flags & MS_RDONLY)
687 return -EROFS;
688
689 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
690 btrfs_dev_replace_lock(dev_replace);
691 switch (dev_replace->replace_state) {
692 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
693 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
694 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
695 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
696 btrfs_dev_replace_unlock(dev_replace);
697 goto leave;
698 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
699 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
700 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
701 tgt_device = dev_replace->tgtdev;
702 dev_replace->tgtdev = NULL;
703 dev_replace->srcdev = NULL;
704 break;
705 }
706 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
707 dev_replace->time_stopped = get_seconds();
708 dev_replace->item_needs_writeback = 1;
709 btrfs_dev_replace_unlock(dev_replace);
710 btrfs_scrub_cancel(fs_info);
711
712 trans = btrfs_start_transaction(root, 0);
713 if (IS_ERR(trans)) {
714 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
715 return PTR_ERR(trans);
716 }
717 ret = btrfs_commit_transaction(trans, root);
718 WARN_ON(ret);
719 if (tgt_device)
720 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
721
722leave:
723 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
724 return result;
725}
726
727void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
728{
729 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
730
731 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
732 btrfs_dev_replace_lock(dev_replace);
733 switch (dev_replace->replace_state) {
734 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
735 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
736 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
737 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
738 break;
739 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
740 dev_replace->replace_state =
741 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
742 dev_replace->time_stopped = get_seconds();
743 dev_replace->item_needs_writeback = 1;
744 btrfs_info(fs_info, "suspending dev_replace for unmount");
745 break;
746 }
747
748 btrfs_dev_replace_unlock(dev_replace);
749 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
750}
751
752/* resume dev_replace procedure that was interrupted by unmount */
753int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
754{
755 struct task_struct *task;
756 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
757
758 btrfs_dev_replace_lock(dev_replace);
759 switch (dev_replace->replace_state) {
760 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
761 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
762 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
763 btrfs_dev_replace_unlock(dev_replace);
764 return 0;
765 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
766 break;
767 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
768 dev_replace->replace_state =
769 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
770 break;
771 }
772 if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
773 btrfs_info(fs_info, "cannot continue dev_replace, tgtdev is missing");
774 btrfs_info(fs_info,
775 "you may cancel the operation after 'mount -o degraded'");
776 btrfs_dev_replace_unlock(dev_replace);
777 return 0;
778 }
779 btrfs_dev_replace_unlock(dev_replace);
780
781 WARN_ON(atomic_xchg(
782 &fs_info->mutually_exclusive_operation_running, 1));
783 task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
784 return PTR_ERR_OR_ZERO(task);
785}
786
787static int btrfs_dev_replace_kthread(void *data)
788{
789 struct btrfs_fs_info *fs_info = data;
790 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
791 struct btrfs_ioctl_dev_replace_args *status_args;
792 u64 progress;
793
794 status_args = kzalloc(sizeof(*status_args), GFP_NOFS);
795 if (status_args) {
796 btrfs_dev_replace_status(fs_info, status_args);
797 progress = status_args->status.progress_1000;
798 kfree(status_args);
799 do_div(progress, 10);
800 printk_in_rcu(KERN_INFO
801 "BTRFS: continuing dev_replace from %s (devid %llu) to %s @%u%%\n",
802 dev_replace->srcdev->missing ? "<missing disk>" :
803 rcu_str_deref(dev_replace->srcdev->name),
804 dev_replace->srcdev->devid,
805 dev_replace->tgtdev ?
806 rcu_str_deref(dev_replace->tgtdev->name) :
807 "<missing target disk>",
808 (unsigned int)progress);
809 }
810 btrfs_dev_replace_continue_on_mount(fs_info);
811 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
812
813 return 0;
814}
815
816static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info)
817{
818 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
819 int ret;
820
821 ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
822 dev_replace->committed_cursor_left,
823 dev_replace->srcdev->total_bytes,
824 &dev_replace->scrub_progress, 0, 1);
825 ret = btrfs_dev_replace_finishing(fs_info, ret);
826 WARN_ON(ret);
827 return 0;
828}
829
830int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
831{
832 if (!dev_replace->is_valid)
833 return 0;
834
835 switch (dev_replace->replace_state) {
836 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
837 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
838 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
839 return 0;
840 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
841 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
842 /*
843 * return true even if tgtdev is missing (this is
844 * something that can happen if the dev_replace
845 * procedure is suspended by an umount and then
846 * the tgtdev is missing (or "btrfs dev scan") was
847 * not called and the the filesystem is remounted
848 * in degraded state. This does not stop the
849 * dev_replace procedure. It needs to be canceled
850 * manually if the cancelation is wanted.
851 */
852 break;
853 }
854 return 1;
855}
856
857void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace)
858{
859 /* the beginning is just an optimization for the typical case */
860 if (atomic_read(&dev_replace->nesting_level) == 0) {
861acquire_lock:
862 /* this is not a nested case where the same thread
863 * is trying to acqurire the same lock twice */
864 mutex_lock(&dev_replace->lock);
865 mutex_lock(&dev_replace->lock_management_lock);
866 dev_replace->lock_owner = current->pid;
867 atomic_inc(&dev_replace->nesting_level);
868 mutex_unlock(&dev_replace->lock_management_lock);
869 return;
870 }
871
872 mutex_lock(&dev_replace->lock_management_lock);
873 if (atomic_read(&dev_replace->nesting_level) > 0 &&
874 dev_replace->lock_owner == current->pid) {
875 WARN_ON(!mutex_is_locked(&dev_replace->lock));
876 atomic_inc(&dev_replace->nesting_level);
877 mutex_unlock(&dev_replace->lock_management_lock);
878 return;
879 }
880
881 mutex_unlock(&dev_replace->lock_management_lock);
882 goto acquire_lock;
883}
884
885void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace)
886{
887 WARN_ON(!mutex_is_locked(&dev_replace->lock));
888 mutex_lock(&dev_replace->lock_management_lock);
889 WARN_ON(atomic_read(&dev_replace->nesting_level) < 1);
890 WARN_ON(dev_replace->lock_owner != current->pid);
891 atomic_dec(&dev_replace->nesting_level);
892 if (atomic_read(&dev_replace->nesting_level) == 0) {
893 dev_replace->lock_owner = 0;
894 mutex_unlock(&dev_replace->lock_management_lock);
895 mutex_unlock(&dev_replace->lock);
896 } else {
897 mutex_unlock(&dev_replace->lock_management_lock);
898 }
899}
900
901void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
902{
903 percpu_counter_inc(&fs_info->bio_counter);
904}
905
906void btrfs_bio_counter_dec(struct btrfs_fs_info *fs_info)
907{
908 percpu_counter_dec(&fs_info->bio_counter);
909
910 if (waitqueue_active(&fs_info->replace_wait))
911 wake_up(&fs_info->replace_wait);
912}
913
914void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
915{
916 DEFINE_WAIT(wait);
917again:
918 percpu_counter_inc(&fs_info->bio_counter);
919 if (test_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state)) {
920 btrfs_bio_counter_dec(fs_info);
921 wait_event(fs_info->replace_wait,
922 !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
923 &fs_info->fs_state));
924 goto again;
925 }
926
927}