projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: fix race between balance and unused block group deletion
[linux-2.6-block.git] / fs / btrfs / dev-replace.c
CommitLineData
e93c89c1
SB		 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>
e93c89c1
SB		 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"
49c6f736 39#include "sysfs.h"
e93c89c1 40
e93c89c1
SB		 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 int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid,
48 char *srcdev_name,
49 struct btrfs_device **device);
50static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info);
51static int btrfs_dev_replace_kthread(void *data);
52static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info);
53
54
55int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
56{
57 struct btrfs_key key;
58 struct btrfs_root *dev_root = fs_info->dev_root;
59 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
60 struct extent_buffer *eb;
61 int slot;
62 int ret = 0;
63 struct btrfs_path *path = NULL;
64 int item_size;
65 struct btrfs_dev_replace_item *ptr;
66 u64 src_devid;
67
68 path = btrfs_alloc_path();
69 if (!path) {
70 ret = -ENOMEM;
71 goto out;
72 }
73
74 key.objectid = 0;
75 key.type = BTRFS_DEV_REPLACE_KEY;
76 key.offset = 0;
77 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
78 if (ret) {
79no_valid_dev_replace_entry_found:
80 ret = 0;
81 dev_replace->replace_state =
82 BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED;
83 dev_replace->cont_reading_from_srcdev_mode =
84 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
85 dev_replace->replace_state = 0;
86 dev_replace->time_started = 0;
87 dev_replace->time_stopped = 0;
88 atomic64_set(&dev_replace->num_write_errors, 0);
89 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
90 dev_replace->cursor_left = 0;
91 dev_replace->committed_cursor_left = 0;
92 dev_replace->cursor_left_last_write_of_item = 0;
93 dev_replace->cursor_right = 0;
94 dev_replace->srcdev = NULL;
95 dev_replace->tgtdev = NULL;
96 dev_replace->is_valid = 0;
97 dev_replace->item_needs_writeback = 0;
98 goto out;
99 }
100 slot = path->slots[0];
101 eb = path->nodes[0];
102 item_size = btrfs_item_size_nr(eb, slot);
103 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
104
105 if (item_size != sizeof(struct btrfs_dev_replace_item)) {
efe120a0
FH		 106 btrfs_warn(fs_info,
107 "dev_replace entry found has unexpected size, ignore entry");
e93c89c1
SB		 108 goto no_valid_dev_replace_entry_found;
109 }
110
111 src_devid = btrfs_dev_replace_src_devid(eb, ptr);
112 dev_replace->cont_reading_from_srcdev_mode =
113 btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
114 dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
115 dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
116 dev_replace->time_stopped =
117 btrfs_dev_replace_time_stopped(eb, ptr);
118 atomic64_set(&dev_replace->num_write_errors,
119 btrfs_dev_replace_num_write_errors(eb, ptr));
120 atomic64_set(&dev_replace->num_uncorrectable_read_errors,
121 btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
122 dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
123 dev_replace->committed_cursor_left = dev_replace->cursor_left;
124 dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
125 dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
126 dev_replace->is_valid = 1;
127
128 dev_replace->item_needs_writeback = 0;
129 switch (dev_replace->replace_state) {
130 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
131 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
132 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
133 dev_replace->srcdev = NULL;
134 dev_replace->tgtdev = NULL;
135 break;
136 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
137 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
138 dev_replace->srcdev = btrfs_find_device(fs_info, src_devid,
139 NULL, NULL);
140 dev_replace->tgtdev = btrfs_find_device(fs_info,
141 BTRFS_DEV_REPLACE_DEVID,
142 NULL, NULL);
143 /*
144 * allow 'btrfs dev replace_cancel' if src/tgt device is
145 * missing
146 */
147 if (!dev_replace->srcdev &&
148 !btrfs_test_opt(dev_root, DEGRADED)) {
149 ret = -EIO;
efe120a0
FH		 150 btrfs_warn(fs_info,
151 "cannot mount because device replace operation is ongoing and");
152 btrfs_warn(fs_info,
153 "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
154 src_devid);
e93c89c1
SB		 155 }
156 if (!dev_replace->tgtdev &&
157 !btrfs_test_opt(dev_root, DEGRADED)) {
158 ret = -EIO;
efe120a0
FH		 159 btrfs_warn(fs_info,
160 "cannot mount because device replace operation is ongoing and");
161 btrfs_warn(fs_info,
162 "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
6e71c47a 163 BTRFS_DEV_REPLACE_DEVID);
e93c89c1
SB		 164 }
165 if (dev_replace->tgtdev) {
166 if (dev_replace->srcdev) {
167 dev_replace->tgtdev->total_bytes =
168 dev_replace->srcdev->total_bytes;
169 dev_replace->tgtdev->disk_total_bytes =
170 dev_replace->srcdev->disk_total_bytes;
935e5cc9
MX		 171 dev_replace->tgtdev->commit_total_bytes =
172 dev_replace->srcdev->commit_total_bytes;
e93c89c1
SB		 173 dev_replace->tgtdev->bytes_used =
174 dev_replace->srcdev->bytes_used;
ce7213c7
MX		 175 dev_replace->tgtdev->commit_bytes_used =
176 dev_replace->srcdev->commit_bytes_used;
e93c89c1
SB		 177 }
178 dev_replace->tgtdev->is_tgtdev_for_dev_replace = 1;
179 btrfs_init_dev_replace_tgtdev_for_resume(fs_info,
180 dev_replace->tgtdev);
181 }
182 break;
183 }
184
185out:
186 if (path)
187 btrfs_free_path(path);
188 return ret;
189}
190
191/*
192 * called from commit_transaction. Writes changed device replace state to
193 * disk.
194 */
195int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
196 struct btrfs_fs_info *fs_info)
197{
198 int ret;
199 struct btrfs_root *dev_root = fs_info->dev_root;
200 struct btrfs_path *path;
201 struct btrfs_key key;
202 struct extent_buffer *eb;
203 struct btrfs_dev_replace_item *ptr;
204 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
205
206 btrfs_dev_replace_lock(dev_replace);
207 if (!dev_replace->is_valid ||
208 !dev_replace->item_needs_writeback) {
209 btrfs_dev_replace_unlock(dev_replace);
210 return 0;
211 }
212 btrfs_dev_replace_unlock(dev_replace);
213
214 key.objectid = 0;
215 key.type = BTRFS_DEV_REPLACE_KEY;
216 key.offset = 0;
217
218 path = btrfs_alloc_path();
219 if (!path) {
220 ret = -ENOMEM;
221 goto out;
222 }
223 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
224 if (ret < 0) {
efe120a0 225 btrfs_warn(fs_info, "error %d while searching for dev_replace item!",
e93c89c1
SB		 226 ret);
227 goto out;
228 }
229
230 if (ret == 0 &&
231 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
232 /*
233 * need to delete old one and insert a new one.
234 * Since no attempt is made to recover any old state, if the
235 * dev_replace state is 'running', the data on the target
236 * drive is lost.
237 * It would be possible to recover the state: just make sure
238 * that the beginning of the item is never changed and always
239 * contains all the essential information. Then read this
240 * minimal set of information and use it as a base for the
241 * new state.
242 */
243 ret = btrfs_del_item(trans, dev_root, path);
244 if (ret != 0) {
efe120a0 245 btrfs_warn(fs_info, "delete too small dev_replace item failed %d!",
e93c89c1
SB		 246 ret);
247 goto out;
248 }
249 ret = 1;
250 }
251
252 if (ret == 1) {
253 /* need to insert a new item */
254 btrfs_release_path(path);
255 ret = btrfs_insert_empty_item(trans, dev_root, path,
256 &key, sizeof(*ptr));
257 if (ret < 0) {
efe120a0 258 btrfs_warn(fs_info, "insert dev_replace item failed %d!",
e93c89c1
SB		 259 ret);
260 goto out;
261 }
262 }
263
264 eb = path->nodes[0];
265 ptr = btrfs_item_ptr(eb, path->slots[0],
266 struct btrfs_dev_replace_item);
267
268 btrfs_dev_replace_lock(dev_replace);
269 if (dev_replace->srcdev)
270 btrfs_set_dev_replace_src_devid(eb, ptr,
271 dev_replace->srcdev->devid);
272 else
273 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
274 btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
275 dev_replace->cont_reading_from_srcdev_mode);
276 btrfs_set_dev_replace_replace_state(eb, ptr,
277 dev_replace->replace_state);
278 btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
279 btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
280 btrfs_set_dev_replace_num_write_errors(eb, ptr,
281 atomic64_read(&dev_replace->num_write_errors));
282 btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
283 atomic64_read(&dev_replace->num_uncorrectable_read_errors));
284 dev_replace->cursor_left_last_write_of_item =
285 dev_replace->cursor_left;
286 btrfs_set_dev_replace_cursor_left(eb, ptr,
287 dev_replace->cursor_left_last_write_of_item);
288 btrfs_set_dev_replace_cursor_right(eb, ptr,
289 dev_replace->cursor_right);
290 dev_replace->item_needs_writeback = 0;
291 btrfs_dev_replace_unlock(dev_replace);
292
293 btrfs_mark_buffer_dirty(eb);
294
295out:
296 btrfs_free_path(path);
297
298 return ret;
299}
300
301void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
302{
303 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
304
305 dev_replace->committed_cursor_left =
306 dev_replace->cursor_left_last_write_of_item;
307}
308
e93c89c1
SB		 309int btrfs_dev_replace_start(struct btrfs_root *root,
310 struct btrfs_ioctl_dev_replace_args *args)
311{
312 struct btrfs_trans_handle *trans;
313 struct btrfs_fs_info *fs_info = root->fs_info;
314 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
315 int ret;
316 struct btrfs_device *tgt_device = NULL;
317 struct btrfs_device *src_device = NULL;
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
935e5cc9
MX		 331 /*
332 * Here we commit the transaction to make sure commit_total_bytes
333 * of all the devices are updated.
334 */
335 trans = btrfs_attach_transaction(root);
336 if (!IS_ERR(trans)) {
337 ret = btrfs_commit_transaction(trans, root);
338 if (ret)
339 return ret;
340 } else if (PTR_ERR(trans) != -ENOENT) {
341 return PTR_ERR(trans);
342 }
343
344 /* the disk copy procedure reuses the scrub code */
e93c89c1 345 mutex_lock(&fs_info->volume_mutex);
e93c89c1
SB		 346 ret = btrfs_dev_replace_find_srcdev(root, args->start.srcdevid,
347 args->start.srcdev_name,
348 &src_device);
e93c89c1 349 if (ret) {
1c43366d
MX		 350 mutex_unlock(&fs_info->volume_mutex);
351 return ret;
e93c89c1
SB		 352 }
353
1c43366d
MX		 354 ret = btrfs_init_dev_replace_tgtdev(root, args->start.tgtdev_name,
355 src_device, &tgt_device);
356 mutex_unlock(&fs_info->volume_mutex);
357 if (ret)
358 return ret;
e93c89c1
SB		 359
360 btrfs_dev_replace_lock(dev_replace);
361 switch (dev_replace->replace_state) {
362 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
363 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
364 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
365 break;
366 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
367 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
368 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
369 goto leave;
370 }
371
372 dev_replace->cont_reading_from_srcdev_mode =
373 args->start.cont_reading_from_srcdev_mode;
374 WARN_ON(!src_device);
375 dev_replace->srcdev = src_device;
376 WARN_ON(!tgt_device);
377 dev_replace->tgtdev = tgt_device;
378
d2ff1b20
AJ		 379 ret = btrfs_kobj_add_device(tgt_device->fs_devices, tgt_device);
380 if (ret)
381 btrfs_error(root->fs_info, ret, "kobj add dev failed");
382
e93c89c1 383 printk_in_rcu(KERN_INFO
efe120a0 384 "BTRFS: dev_replace from %s (devid %llu) to %s started\n",
e93c89c1
SB		 385 src_device->missing ? "<missing disk>" :
386 rcu_str_deref(src_device->name),
387 src_device->devid,
388 rcu_str_deref(tgt_device->name));
389
e93c89c1
SB		 390 /*
391 * from now on, the writes to the srcdev are all duplicated to
392 * go to the tgtdev as well (refer to btrfs_map_block()).
393 */
394 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
4546bcae 395 dev_replace->time_started = get_seconds();
e93c89c1
SB		 396 dev_replace->cursor_left = 0;
397 dev_replace->committed_cursor_left = 0;
398 dev_replace->cursor_left_last_write_of_item = 0;
399 dev_replace->cursor_right = 0;
400 dev_replace->is_valid = 1;
401 dev_replace->item_needs_writeback = 1;
402 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
403 btrfs_dev_replace_unlock(dev_replace);
404
b0244199 405 btrfs_wait_ordered_roots(root->fs_info, -1);
e93c89c1
SB		 406
407 /* force writing the updated state information to disk */
408 trans = btrfs_start_transaction(root, 0);
409 if (IS_ERR(trans)) {
410 ret = PTR_ERR(trans);
411 btrfs_dev_replace_lock(dev_replace);
412 goto leave;
413 }
414
415 ret = btrfs_commit_transaction(trans, root);
416 WARN_ON(ret);
417
418 /* the disk copy procedure reuses the scrub code */
419 ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
7cc8e58d 420 btrfs_device_get_total_bytes(src_device),
e93c89c1
SB		 421 &dev_replace->scrub_progress, 0, 1);
422
423 ret = btrfs_dev_replace_finishing(root->fs_info, ret);
2fc9f6ba
EG		 424 /* don't warn if EINPROGRESS, someone else might be running scrub */
425 if (ret == -EINPROGRESS) {
426 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
427 ret = 0;
428 } else {
429 WARN_ON(ret);
430 }
e93c89c1 431
2fc9f6ba 432 return ret;
e93c89c1
SB		 433
434leave:
435 dev_replace->srcdev = NULL;
436 dev_replace->tgtdev = NULL;
437 btrfs_dev_replace_unlock(dev_replace);
1c43366d 438 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
e93c89c1
SB		 439 return ret;
440}
441
c404e0dc
MX		 442/*
443 * blocked until all flighting bios are finished.
444 */
445static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
446{
c404e0dc 447 set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
7653947f
ZL		 448 wait_event(fs_info->replace_wait, !percpu_counter_sum(
449 &fs_info->bio_counter));
c404e0dc
MX		 450}
451
452/*
453 * we have removed target device, it is safe to allow new bios request.
454 */
455static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
456{
457 clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
458 if (waitqueue_active(&fs_info->replace_wait))
459 wake_up(&fs_info->replace_wait);
460}
461
e93c89c1
SB		 462static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
463 int scrub_ret)
464{
465 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
466 struct btrfs_device *tgt_device;
467 struct btrfs_device *src_device;
468 struct btrfs_root *root = fs_info->tree_root;
469 u8 uuid_tmp[BTRFS_UUID_SIZE];
470 struct btrfs_trans_handle *trans;
471 int ret = 0;
472
473 /* don't allow cancel or unmount to disturb the finishing procedure */
474 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
475
476 btrfs_dev_replace_lock(dev_replace);
477 /* was the operation canceled, or is it finished? */
478 if (dev_replace->replace_state !=
479 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
480 btrfs_dev_replace_unlock(dev_replace);
481 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
482 return 0;
483 }
484
485 tgt_device = dev_replace->tgtdev;
486 src_device = dev_replace->srcdev;
487 btrfs_dev_replace_unlock(dev_replace);
488
e93c89c1
SB		 489 /*
490 * flush all outstanding I/O and inode extent mappings before the
491 * copy operation is declared as being finished
492 */
6c255e67 493 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
3edb2a68
MX		 494 if (ret) {
495 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
496 return ret;
497 }
b0244199 498 btrfs_wait_ordered_roots(root->fs_info, -1);
e93c89c1
SB		 499
500 trans = btrfs_start_transaction(root, 0);
501 if (IS_ERR(trans)) {
502 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
503 return PTR_ERR(trans);
504 }
505 ret = btrfs_commit_transaction(trans, root);
506 WARN_ON(ret);
507
67a2c45e 508 mutex_lock(&uuid_mutex);
e93c89c1
SB		 509 /* keep away write_all_supers() during the finishing procedure */
510 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2196d6e8 511 mutex_lock(&root->fs_info->chunk_mutex);
e93c89c1
SB		 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;
4546bcae 518 dev_replace->time_stopped = get_seconds();
e93c89c1
SB		 519 dev_replace->item_needs_writeback = 1;
520
c404e0dc
MX		 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 {
e93c89c1 527 printk_in_rcu(KERN_ERR
efe120a0 528 "BTRFS: btrfs_scrub_dev(%s, %llu, %s) failed %d\n",
e93c89c1
SB		 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);
391cd9df 534 mutex_unlock(&root->fs_info->chunk_mutex);
2196d6e8 535 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
67a2c45e 536 mutex_unlock(&uuid_mutex);
e93c89c1
SB		 537 if (tgt_device)
538 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
539 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
540
2fc9f6ba 541 return scrub_ret;
e93c89c1
SB		 542 }
543
544 printk_in_rcu(KERN_INFO
de4c296f 545 "BTRFS: dev_replace from %s (devid %llu) to %s finished\n",
e93c89c1
SB		 546 src_device->missing ? "<missing disk>" :
547 rcu_str_deref(src_device->name),
548 src_device->devid,
549 rcu_str_deref(tgt_device->name));
550 tgt_device->is_tgtdev_for_dev_replace = 0;
551 tgt_device->devid = src_device->devid;
552 src_device->devid = BTRFS_DEV_REPLACE_DEVID;
e93c89c1
SB		 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));
7cc8e58d
MX		 556 btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
557 btrfs_device_set_disk_total_bytes(tgt_device,
558 src_device->disk_total_bytes);
559 btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
935e5cc9
MX		 560 ASSERT(list_empty(&src_device->resized_list));
561 tgt_device->commit_total_bytes = src_device->commit_total_bytes;
ce7213c7 562 tgt_device->commit_bytes_used = src_device->bytes_used;
e93c89c1
SB		 563 if (fs_info->sb->s_bdev == src_device->bdev)
564 fs_info->sb->s_bdev = tgt_device->bdev;
565 if (fs_info->fs_devices->latest_bdev == src_device->bdev)
566 fs_info->fs_devices->latest_bdev = tgt_device->bdev;
567 list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
82372bc8 568 fs_info->fs_devices->rw_devices++;
e93c89c1 569
12b894cb
QW		 570 btrfs_dev_replace_unlock(dev_replace);
571
c404e0dc
MX		 572 btrfs_rm_dev_replace_blocked(fs_info);
573
084b6e7c 574 btrfs_rm_dev_replace_remove_srcdev(fs_info, src_device);
1357272f 575
c404e0dc
MX		 576 btrfs_rm_dev_replace_unblocked(fs_info);
577
e93c89c1
SB		 578 /*
579 * this is again a consistent state where no dev_replace procedure
580 * is running, the target device is part of the filesystem, the
581 * source device is not part of the filesystem anymore and its 1st
582 * superblock is scratched out so that it is no longer marked to
583 * belong to this filesystem.
584 */
391cd9df 585 mutex_unlock(&root->fs_info->chunk_mutex);
2196d6e8 586 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
67a2c45e 587 mutex_unlock(&uuid_mutex);
e93c89c1 588
084b6e7c 589 /* replace the sysfs entry */
6c14a164 590 btrfs_kobj_rm_device(fs_info->fs_devices, src_device);
084b6e7c
QW		 591 btrfs_rm_dev_replace_free_srcdev(fs_info, src_device);
592
e93c89c1
SB		 593 /* write back the superblocks */
594 trans = btrfs_start_transaction(root, 0);
595 if (!IS_ERR(trans))
596 btrfs_commit_transaction(trans, root);
597
598 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
599
600 return 0;
601}
602
603static void btrfs_dev_replace_update_device_in_mapping_tree(
604 struct btrfs_fs_info *fs_info,
605 struct btrfs_device *srcdev,
606 struct btrfs_device *tgtdev)
607{
608 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
609 struct extent_map *em;
610 struct map_lookup *map;
611 u64 start = 0;
612 int i;
613
614 write_lock(&em_tree->lock);
615 do {
616 em = lookup_extent_mapping(em_tree, start, (u64)-1);
617 if (!em)
618 break;
619 map = (struct map_lookup *)em->bdev;
620 for (i = 0; i < map->num_stripes; i++)
621 if (srcdev == map->stripes[i].dev)
622 map->stripes[i].dev = tgtdev;
623 start = em->start + em->len;
624 free_extent_map(em);
625 } while (start);
626 write_unlock(&em_tree->lock);
627}
628
629static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid,
630 char *srcdev_name,
631 struct btrfs_device **device)
632{
633 int ret;
634
635 if (srcdevid) {
636 ret = 0;
637 *device = btrfs_find_device(root->fs_info, srcdevid, NULL,
638 NULL);
639 if (!*device)
640 ret = -ENOENT;
641 } else {
642 ret = btrfs_find_device_missing_or_by_path(root, srcdev_name,
643 device);
644 }
645 return ret;
646}
647
648void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
649 struct btrfs_ioctl_dev_replace_args *args)
650{
651 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
7cc8e58d 652 struct btrfs_device *srcdev;
e93c89c1
SB		 653
654 btrfs_dev_replace_lock(dev_replace);
655 /* even if !dev_replace_is_valid, the values are good enough for
656 * the replace_status ioctl */
657 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
658 args->status.replace_state = dev_replace->replace_state;
659 args->status.time_started = dev_replace->time_started;
660 args->status.time_stopped = dev_replace->time_stopped;
661 args->status.num_write_errors =
662 atomic64_read(&dev_replace->num_write_errors);
663 args->status.num_uncorrectable_read_errors =
664 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
665 switch (dev_replace->replace_state) {
666 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
667 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
668 args->status.progress_1000 = 0;
669 break;
670 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
671 args->status.progress_1000 = 1000;
672 break;
673 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
674 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
7cc8e58d 675 srcdev = dev_replace->srcdev;
f8c269d7
DS		 676 args->status.progress_1000 = div_u64(dev_replace->cursor_left,
677 div_u64(btrfs_device_get_total_bytes(srcdev), 1000));
e93c89c1
SB		 678 break;
679 }
680 btrfs_dev_replace_unlock(dev_replace);
681}
682
683int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info,
684 struct btrfs_ioctl_dev_replace_args *args)
685{
686 args->result = __btrfs_dev_replace_cancel(fs_info);
687 return 0;
688}
689
690static u64 __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_trans_handle *trans;
695 struct btrfs_root *root = fs_info->tree_root;
696 u64 result;
697 int ret;
698
e649e587
ID		 699 if (fs_info->sb->s_flags & MS_RDONLY)
700 return -EROFS;
701
e93c89c1
SB		 702 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
703 btrfs_dev_replace_lock(dev_replace);
704 switch (dev_replace->replace_state) {
705 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
706 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
707 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
708 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
709 btrfs_dev_replace_unlock(dev_replace);
710 goto leave;
711 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
712 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
713 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
714 tgt_device = dev_replace->tgtdev;
715 dev_replace->tgtdev = NULL;
716 dev_replace->srcdev = NULL;
717 break;
718 }
719 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
4546bcae 720 dev_replace->time_stopped = get_seconds();
e93c89c1
SB		 721 dev_replace->item_needs_writeback = 1;
722 btrfs_dev_replace_unlock(dev_replace);
723 btrfs_scrub_cancel(fs_info);
724
725 trans = btrfs_start_transaction(root, 0);
726 if (IS_ERR(trans)) {
727 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
728 return PTR_ERR(trans);
729 }
730 ret = btrfs_commit_transaction(trans, root);
731 WARN_ON(ret);
732 if (tgt_device)
733 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
734
735leave:
736 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
737 return result;
738}
739
740void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
741{
742 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
743
744 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
745 btrfs_dev_replace_lock(dev_replace);
746 switch (dev_replace->replace_state) {
747 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
748 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
749 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
750 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
751 break;
752 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
753 dev_replace->replace_state =
754 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
4546bcae 755 dev_replace->time_stopped = get_seconds();
e93c89c1 756 dev_replace->item_needs_writeback = 1;
efe120a0 757 btrfs_info(fs_info, "suspending dev_replace for unmount");
e93c89c1
SB		 758 break;
759 }
760
761 btrfs_dev_replace_unlock(dev_replace);
762 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
763}
764
765/* resume dev_replace procedure that was interrupted by unmount */
766int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
767{
768 struct task_struct *task;
769 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
770
771 btrfs_dev_replace_lock(dev_replace);
772 switch (dev_replace->replace_state) {
773 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
774 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
775 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
776 btrfs_dev_replace_unlock(dev_replace);
777 return 0;
778 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
779 break;
780 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
781 dev_replace->replace_state =
782 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
783 break;
784 }
785 if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
efe120a0
FH		 786 btrfs_info(fs_info, "cannot continue dev_replace, tgtdev is missing");
787 btrfs_info(fs_info,
788 "you may cancel the operation after 'mount -o degraded'");
e93c89c1
SB		 789 btrfs_dev_replace_unlock(dev_replace);
790 return 0;
791 }
792 btrfs_dev_replace_unlock(dev_replace);
793
794 WARN_ON(atomic_xchg(
795 &fs_info->mutually_exclusive_operation_running, 1));
796 task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
8c6ffba0 797 return PTR_ERR_OR_ZERO(task);
e93c89c1
SB		 798}
799
800static int btrfs_dev_replace_kthread(void *data)
801{
802 struct btrfs_fs_info *fs_info = data;
803 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
804 struct btrfs_ioctl_dev_replace_args *status_args;
805 u64 progress;
806
807 status_args = kzalloc(sizeof(*status_args), GFP_NOFS);
808 if (status_args) {
809 btrfs_dev_replace_status(fs_info, status_args);
810 progress = status_args->status.progress_1000;
811 kfree(status_args);
f8c269d7 812 progress = div_u64(progress, 10);
e93c89c1 813 printk_in_rcu(KERN_INFO
efe120a0
FH		 814 "BTRFS: continuing dev_replace from %s (devid %llu) to %s @%u%%\n",
815 dev_replace->srcdev->missing ? "<missing disk>" :
816 rcu_str_deref(dev_replace->srcdev->name),
817 dev_replace->srcdev->devid,
818 dev_replace->tgtdev ?
819 rcu_str_deref(dev_replace->tgtdev->name) :
820 "<missing target disk>",
821 (unsigned int)progress);
e93c89c1
SB		 822 }
823 btrfs_dev_replace_continue_on_mount(fs_info);
824 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
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,
7cc8e58d 836 btrfs_device_get_total_bytes(dev_replace->srcdev),
e93c89c1
SB		 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 cancelation is wanted.
864 */
865 break;
866 }
867 return 1;
868}
869
870void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace)
871{
872 /* the beginning is just an optimization for the typical case */
873 if (atomic_read(&dev_replace->nesting_level) == 0) {
874acquire_lock:
875 /* this is not a nested case where the same thread
876 * is trying to acqurire the same lock twice */
877 mutex_lock(&dev_replace->lock);
878 mutex_lock(&dev_replace->lock_management_lock);
879 dev_replace->lock_owner = current->pid;
880 atomic_inc(&dev_replace->nesting_level);
881 mutex_unlock(&dev_replace->lock_management_lock);
882 return;
883 }
884
885 mutex_lock(&dev_replace->lock_management_lock);
886 if (atomic_read(&dev_replace->nesting_level) > 0 &&
887 dev_replace->lock_owner == current->pid) {
888 WARN_ON(!mutex_is_locked(&dev_replace->lock));
889 atomic_inc(&dev_replace->nesting_level);
890 mutex_unlock(&dev_replace->lock_management_lock);
891 return;
892 }
893
894 mutex_unlock(&dev_replace->lock_management_lock);
895 goto acquire_lock;
896}
897
898void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace)
899{
900 WARN_ON(!mutex_is_locked(&dev_replace->lock));
901 mutex_lock(&dev_replace->lock_management_lock);
902 WARN_ON(atomic_read(&dev_replace->nesting_level) < 1);
903 WARN_ON(dev_replace->lock_owner != current->pid);
904 atomic_dec(&dev_replace->nesting_level);
905 if (atomic_read(&dev_replace->nesting_level) == 0) {
906 dev_replace->lock_owner = 0;
907 mutex_unlock(&dev_replace->lock_management_lock);
908 mutex_unlock(&dev_replace->lock);
909 } else {
910 mutex_unlock(&dev_replace->lock_management_lock);
911 }
912}
c404e0dc
MX		 913
914void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
915{
916 percpu_counter_inc(&fs_info->bio_counter);
917}
918
4245215d 919void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
c404e0dc 920{
4245215d 921 percpu_counter_sub(&fs_info->bio_counter, amount);
c404e0dc
MX		 922
923 if (waitqueue_active(&fs_info->replace_wait))
924 wake_up(&fs_info->replace_wait);
925}
926
927void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
928{
09dd7a01
ZL		 929 while (1) {
930 percpu_counter_inc(&fs_info->bio_counter);
931 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
932 &fs_info->fs_state)))
933 break;
934
c404e0dc
MX		 935 btrfs_bio_counter_dec(fs_info);
936 wait_event(fs_info->replace_wait,
937 !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
938 &fs_info->fs_state));
c404e0dc 939 }
c404e0dc 940}
Linux 6.x block layer and io_uring tree(s)