blob: 66dbc8dbddf7584a7c4034203c896cf37fcecb39 [file] [log] [blame]
Stefan Behrense93c89c2012-11-05 17:33:06 +01001/*
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 "compat.h"
30#include "ctree.h"
31#include "extent_map.h"
32#include "disk-io.h"
33#include "transaction.h"
34#include "print-tree.h"
35#include "volumes.h"
36#include "async-thread.h"
37#include "check-integrity.h"
38#include "rcu-string.h"
39#include "dev-replace.h"
40
41static u64 btrfs_get_seconds_since_1970(void);
42static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
43 int scrub_ret);
44static void btrfs_dev_replace_update_device_in_mapping_tree(
45 struct btrfs_fs_info *fs_info,
46 struct btrfs_device *srcdev,
47 struct btrfs_device *tgtdev);
48static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid,
49 char *srcdev_name,
50 struct btrfs_device **device);
51static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info);
52static int btrfs_dev_replace_kthread(void *data);
53static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info);
54
55
56int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
57{
58 struct btrfs_key key;
59 struct btrfs_root *dev_root = fs_info->dev_root;
60 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
61 struct extent_buffer *eb;
62 int slot;
63 int ret = 0;
64 struct btrfs_path *path = NULL;
65 int item_size;
66 struct btrfs_dev_replace_item *ptr;
67 u64 src_devid;
68
69 path = btrfs_alloc_path();
70 if (!path) {
71 ret = -ENOMEM;
72 goto out;
73 }
74
75 key.objectid = 0;
76 key.type = BTRFS_DEV_REPLACE_KEY;
77 key.offset = 0;
78 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
79 if (ret) {
80no_valid_dev_replace_entry_found:
81 ret = 0;
82 dev_replace->replace_state =
83 BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED;
84 dev_replace->cont_reading_from_srcdev_mode =
85 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
86 dev_replace->replace_state = 0;
87 dev_replace->time_started = 0;
88 dev_replace->time_stopped = 0;
89 atomic64_set(&dev_replace->num_write_errors, 0);
90 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
91 dev_replace->cursor_left = 0;
92 dev_replace->committed_cursor_left = 0;
93 dev_replace->cursor_left_last_write_of_item = 0;
94 dev_replace->cursor_right = 0;
95 dev_replace->srcdev = NULL;
96 dev_replace->tgtdev = NULL;
97 dev_replace->is_valid = 0;
98 dev_replace->item_needs_writeback = 0;
99 goto out;
100 }
101 slot = path->slots[0];
102 eb = path->nodes[0];
103 item_size = btrfs_item_size_nr(eb, slot);
104 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
105
106 if (item_size != sizeof(struct btrfs_dev_replace_item)) {
107 pr_warn("btrfs: dev_replace entry found has unexpected size, ignore entry\n");
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;
150 pr_warn("btrfs: cannot mount because device replace operation is ongoing and\n" "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?\n",
151 (unsigned long long)src_devid);
152 }
153 if (!dev_replace->tgtdev &&
154 !btrfs_test_opt(dev_root, DEGRADED)) {
155 ret = -EIO;
156 pr_warn("btrfs: cannot mount because device replace operation is ongoing and\n" "tgtdev (devid %llu) is missing, need to run btrfs dev scan?\n",
157 (unsigned long long)BTRFS_DEV_REPLACE_DEVID);
158 }
159 if (dev_replace->tgtdev) {
160 if (dev_replace->srcdev) {
161 dev_replace->tgtdev->total_bytes =
162 dev_replace->srcdev->total_bytes;
163 dev_replace->tgtdev->disk_total_bytes =
164 dev_replace->srcdev->disk_total_bytes;
165 dev_replace->tgtdev->bytes_used =
166 dev_replace->srcdev->bytes_used;
167 }
168 dev_replace->tgtdev->is_tgtdev_for_dev_replace = 1;
169 btrfs_init_dev_replace_tgtdev_for_resume(fs_info,
170 dev_replace->tgtdev);
171 }
172 break;
173 }
174
175out:
176 if (path)
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_lock(dev_replace);
197 if (!dev_replace->is_valid ||
198 !dev_replace->item_needs_writeback) {
199 btrfs_dev_replace_unlock(dev_replace);
200 return 0;
201 }
202 btrfs_dev_replace_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 pr_warn("btrfs: error %d while searching for dev_replace item!\n",
216 ret);
217 goto out;
218 }
219
220 if (ret == 0 &&
221 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
222 /*
223 * need to delete old one and insert a new one.
224 * Since no attempt is made to recover any old state, if the
225 * dev_replace state is 'running', the data on the target
226 * drive is lost.
227 * It would be possible to recover the state: just make sure
228 * that the beginning of the item is never changed and always
229 * contains all the essential information. Then read this
230 * minimal set of information and use it as a base for the
231 * new state.
232 */
233 ret = btrfs_del_item(trans, dev_root, path);
234 if (ret != 0) {
235 pr_warn("btrfs: delete too small dev_replace item failed %d!\n",
236 ret);
237 goto out;
238 }
239 ret = 1;
240 }
241
242 if (ret == 1) {
243 /* need to insert a new item */
244 btrfs_release_path(path);
245 ret = btrfs_insert_empty_item(trans, dev_root, path,
246 &key, sizeof(*ptr));
247 if (ret < 0) {
248 pr_warn("btrfs: insert dev_replace item failed %d!\n",
249 ret);
250 goto out;
251 }
252 }
253
254 eb = path->nodes[0];
255 ptr = btrfs_item_ptr(eb, path->slots[0],
256 struct btrfs_dev_replace_item);
257
258 btrfs_dev_replace_lock(dev_replace);
259 if (dev_replace->srcdev)
260 btrfs_set_dev_replace_src_devid(eb, ptr,
261 dev_replace->srcdev->devid);
262 else
263 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
264 btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
265 dev_replace->cont_reading_from_srcdev_mode);
266 btrfs_set_dev_replace_replace_state(eb, ptr,
267 dev_replace->replace_state);
268 btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
269 btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
270 btrfs_set_dev_replace_num_write_errors(eb, ptr,
271 atomic64_read(&dev_replace->num_write_errors));
272 btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
273 atomic64_read(&dev_replace->num_uncorrectable_read_errors));
274 dev_replace->cursor_left_last_write_of_item =
275 dev_replace->cursor_left;
276 btrfs_set_dev_replace_cursor_left(eb, ptr,
277 dev_replace->cursor_left_last_write_of_item);
278 btrfs_set_dev_replace_cursor_right(eb, ptr,
279 dev_replace->cursor_right);
280 dev_replace->item_needs_writeback = 0;
281 btrfs_dev_replace_unlock(dev_replace);
282
283 btrfs_mark_buffer_dirty(eb);
284
285out:
286 btrfs_free_path(path);
287
288 return ret;
289}
290
291void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
292{
293 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
294
295 dev_replace->committed_cursor_left =
296 dev_replace->cursor_left_last_write_of_item;
297}
298
299static u64 btrfs_get_seconds_since_1970(void)
300{
301 struct timespec t = CURRENT_TIME_SEC;
302
303 return t.tv_sec;
304}
305
306int btrfs_dev_replace_start(struct btrfs_root *root,
307 struct btrfs_ioctl_dev_replace_args *args)
308{
309 struct btrfs_trans_handle *trans;
310 struct btrfs_fs_info *fs_info = root->fs_info;
311 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
312 int ret;
313 struct btrfs_device *tgt_device = NULL;
314 struct btrfs_device *src_device = NULL;
315
316 switch (args->start.cont_reading_from_srcdev_mode) {
317 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
318 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
319 break;
320 default:
321 return -EINVAL;
322 }
323
324 if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
325 args->start.tgtdev_name[0] == '\0')
326 return -EINVAL;
327
328 mutex_lock(&fs_info->volume_mutex);
329 ret = btrfs_init_dev_replace_tgtdev(root, args->start.tgtdev_name,
330 &tgt_device);
331 if (ret) {
332 pr_err("btrfs: target device %s is invalid!\n",
333 args->start.tgtdev_name);
334 mutex_unlock(&fs_info->volume_mutex);
335 return -EINVAL;
336 }
337
338 ret = btrfs_dev_replace_find_srcdev(root, args->start.srcdevid,
339 args->start.srcdev_name,
340 &src_device);
341 mutex_unlock(&fs_info->volume_mutex);
342 if (ret) {
343 ret = -EINVAL;
344 goto leave_no_lock;
345 }
346
347 if (tgt_device->total_bytes < src_device->total_bytes) {
348 pr_err("btrfs: target device is smaller than source device!\n");
349 ret = -EINVAL;
350 goto leave_no_lock;
351 }
352
353 btrfs_dev_replace_lock(dev_replace);
354 switch (dev_replace->replace_state) {
355 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
356 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
357 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
358 break;
359 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
360 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
361 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
362 goto leave;
363 }
364
365 dev_replace->cont_reading_from_srcdev_mode =
366 args->start.cont_reading_from_srcdev_mode;
367 WARN_ON(!src_device);
368 dev_replace->srcdev = src_device;
369 WARN_ON(!tgt_device);
370 dev_replace->tgtdev = tgt_device;
371
372 printk_in_rcu(KERN_INFO
373 "btrfs: dev_replace from %s (devid %llu) to %s) started\n",
374 src_device->missing ? "<missing disk>" :
375 rcu_str_deref(src_device->name),
376 src_device->devid,
377 rcu_str_deref(tgt_device->name));
378
379 tgt_device->total_bytes = src_device->total_bytes;
380 tgt_device->disk_total_bytes = src_device->disk_total_bytes;
381 tgt_device->bytes_used = src_device->bytes_used;
382
383 /*
384 * from now on, the writes to the srcdev are all duplicated to
385 * go to the tgtdev as well (refer to btrfs_map_block()).
386 */
387 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
388 dev_replace->time_started = btrfs_get_seconds_since_1970();
389 dev_replace->cursor_left = 0;
390 dev_replace->committed_cursor_left = 0;
391 dev_replace->cursor_left_last_write_of_item = 0;
392 dev_replace->cursor_right = 0;
393 dev_replace->is_valid = 1;
394 dev_replace->item_needs_writeback = 1;
395 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
396 btrfs_dev_replace_unlock(dev_replace);
397
398 btrfs_wait_ordered_extents(root, 0);
399
400 /* force writing the updated state information to disk */
401 trans = btrfs_start_transaction(root, 0);
402 if (IS_ERR(trans)) {
403 ret = PTR_ERR(trans);
404 btrfs_dev_replace_lock(dev_replace);
405 goto leave;
406 }
407
408 ret = btrfs_commit_transaction(trans, root);
409 WARN_ON(ret);
410
411 /* the disk copy procedure reuses the scrub code */
412 ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
413 src_device->total_bytes,
414 &dev_replace->scrub_progress, 0, 1);
415
416 ret = btrfs_dev_replace_finishing(root->fs_info, ret);
417 WARN_ON(ret);
418
419 return 0;
420
421leave:
422 dev_replace->srcdev = NULL;
423 dev_replace->tgtdev = NULL;
424 btrfs_dev_replace_unlock(dev_replace);
425leave_no_lock:
426 if (tgt_device)
427 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
428 return ret;
429}
430
431static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
432 int scrub_ret)
433{
434 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
435 struct btrfs_device *tgt_device;
436 struct btrfs_device *src_device;
437 struct btrfs_root *root = fs_info->tree_root;
438 u8 uuid_tmp[BTRFS_UUID_SIZE];
439 struct btrfs_trans_handle *trans;
440 int ret = 0;
441
442 /* don't allow cancel or unmount to disturb the finishing procedure */
443 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
444
445 btrfs_dev_replace_lock(dev_replace);
446 /* was the operation canceled, or is it finished? */
447 if (dev_replace->replace_state !=
448 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
449 btrfs_dev_replace_unlock(dev_replace);
450 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
451 return 0;
452 }
453
454 tgt_device = dev_replace->tgtdev;
455 src_device = dev_replace->srcdev;
456 btrfs_dev_replace_unlock(dev_replace);
457
458 /* replace old device with new one in mapping tree */
459 if (!scrub_ret)
460 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
461 src_device,
462 tgt_device);
463
464 /*
465 * flush all outstanding I/O and inode extent mappings before the
466 * copy operation is declared as being finished
467 */
468 btrfs_start_delalloc_inodes(root, 0);
469 btrfs_wait_ordered_extents(root, 0);
470
471 trans = btrfs_start_transaction(root, 0);
472 if (IS_ERR(trans)) {
473 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
474 return PTR_ERR(trans);
475 }
476 ret = btrfs_commit_transaction(trans, root);
477 WARN_ON(ret);
478
479 /* keep away write_all_supers() during the finishing procedure */
480 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
481 btrfs_dev_replace_lock(dev_replace);
482 dev_replace->replace_state =
483 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
484 : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
485 dev_replace->tgtdev = NULL;
486 dev_replace->srcdev = NULL;
487 dev_replace->time_stopped = btrfs_get_seconds_since_1970();
488 dev_replace->item_needs_writeback = 1;
489
490 if (scrub_ret) {
491 printk_in_rcu(KERN_ERR
492 "btrfs: btrfs_scrub_dev(%s, %llu, %s) failed %d\n",
493 src_device->missing ? "<missing disk>" :
494 rcu_str_deref(src_device->name),
495 src_device->devid,
496 rcu_str_deref(tgt_device->name), scrub_ret);
497 btrfs_dev_replace_unlock(dev_replace);
498 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
499 if (tgt_device)
500 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
501 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
502
503 return 0;
504 }
505
506 printk_in_rcu(KERN_INFO
507 "btrfs: dev_replace from %s (devid %llu) to %s) finished\n",
508 src_device->missing ? "<missing disk>" :
509 rcu_str_deref(src_device->name),
510 src_device->devid,
511 rcu_str_deref(tgt_device->name));
512 tgt_device->is_tgtdev_for_dev_replace = 0;
513 tgt_device->devid = src_device->devid;
514 src_device->devid = BTRFS_DEV_REPLACE_DEVID;
515 tgt_device->bytes_used = src_device->bytes_used;
516 memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
517 memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
518 memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
519 tgt_device->total_bytes = src_device->total_bytes;
520 tgt_device->disk_total_bytes = src_device->disk_total_bytes;
521 tgt_device->bytes_used = src_device->bytes_used;
522 if (fs_info->sb->s_bdev == src_device->bdev)
523 fs_info->sb->s_bdev = tgt_device->bdev;
524 if (fs_info->fs_devices->latest_bdev == src_device->bdev)
525 fs_info->fs_devices->latest_bdev = tgt_device->bdev;
526 list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
527
528 btrfs_rm_dev_replace_srcdev(fs_info, src_device);
529 if (src_device->bdev) {
530 /* zero out the old super */
531 btrfs_scratch_superblock(src_device);
532 }
533 /*
534 * this is again a consistent state where no dev_replace procedure
535 * is running, the target device is part of the filesystem, the
536 * source device is not part of the filesystem anymore and its 1st
537 * superblock is scratched out so that it is no longer marked to
538 * belong to this filesystem.
539 */
540 btrfs_dev_replace_unlock(dev_replace);
541 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
542
543 /* write back the superblocks */
544 trans = btrfs_start_transaction(root, 0);
545 if (!IS_ERR(trans))
546 btrfs_commit_transaction(trans, root);
547
548 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
549
550 return 0;
551}
552
553static void btrfs_dev_replace_update_device_in_mapping_tree(
554 struct btrfs_fs_info *fs_info,
555 struct btrfs_device *srcdev,
556 struct btrfs_device *tgtdev)
557{
558 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
559 struct extent_map *em;
560 struct map_lookup *map;
561 u64 start = 0;
562 int i;
563
564 write_lock(&em_tree->lock);
565 do {
566 em = lookup_extent_mapping(em_tree, start, (u64)-1);
567 if (!em)
568 break;
569 map = (struct map_lookup *)em->bdev;
570 for (i = 0; i < map->num_stripes; i++)
571 if (srcdev == map->stripes[i].dev)
572 map->stripes[i].dev = tgtdev;
573 start = em->start + em->len;
574 free_extent_map(em);
575 } while (start);
576 write_unlock(&em_tree->lock);
577}
578
579static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid,
580 char *srcdev_name,
581 struct btrfs_device **device)
582{
583 int ret;
584
585 if (srcdevid) {
586 ret = 0;
587 *device = btrfs_find_device(root->fs_info, srcdevid, NULL,
588 NULL);
589 if (!*device)
590 ret = -ENOENT;
591 } else {
592 ret = btrfs_find_device_missing_or_by_path(root, srcdev_name,
593 device);
594 }
595 return ret;
596}
597
598void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
599 struct btrfs_ioctl_dev_replace_args *args)
600{
601 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
602
603 btrfs_dev_replace_lock(dev_replace);
604 /* even if !dev_replace_is_valid, the values are good enough for
605 * the replace_status ioctl */
606 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
607 args->status.replace_state = dev_replace->replace_state;
608 args->status.time_started = dev_replace->time_started;
609 args->status.time_stopped = dev_replace->time_stopped;
610 args->status.num_write_errors =
611 atomic64_read(&dev_replace->num_write_errors);
612 args->status.num_uncorrectable_read_errors =
613 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
614 switch (dev_replace->replace_state) {
615 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
616 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
617 args->status.progress_1000 = 0;
618 break;
619 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
620 args->status.progress_1000 = 1000;
621 break;
622 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
623 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
624 args->status.progress_1000 = div64_u64(dev_replace->cursor_left,
625 div64_u64(dev_replace->srcdev->total_bytes, 1000));
626 break;
627 }
628 btrfs_dev_replace_unlock(dev_replace);
629}
630
631int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info,
632 struct btrfs_ioctl_dev_replace_args *args)
633{
634 args->result = __btrfs_dev_replace_cancel(fs_info);
635 return 0;
636}
637
638static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
639{
640 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
641 struct btrfs_device *tgt_device = NULL;
642 struct btrfs_trans_handle *trans;
643 struct btrfs_root *root = fs_info->tree_root;
644 u64 result;
645 int ret;
646
647 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
648 btrfs_dev_replace_lock(dev_replace);
649 switch (dev_replace->replace_state) {
650 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
651 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
652 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
653 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
654 btrfs_dev_replace_unlock(dev_replace);
655 goto leave;
656 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
657 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
658 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
659 tgt_device = dev_replace->tgtdev;
660 dev_replace->tgtdev = NULL;
661 dev_replace->srcdev = NULL;
662 break;
663 }
664 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
665 dev_replace->time_stopped = btrfs_get_seconds_since_1970();
666 dev_replace->item_needs_writeback = 1;
667 btrfs_dev_replace_unlock(dev_replace);
668 btrfs_scrub_cancel(fs_info);
669
670 trans = btrfs_start_transaction(root, 0);
671 if (IS_ERR(trans)) {
672 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
673 return PTR_ERR(trans);
674 }
675 ret = btrfs_commit_transaction(trans, root);
676 WARN_ON(ret);
677 if (tgt_device)
678 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
679
680leave:
681 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
682 return result;
683}
684
685void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
686{
687 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
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 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
696 break;
697 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
698 dev_replace->replace_state =
699 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
700 dev_replace->time_stopped = btrfs_get_seconds_since_1970();
701 dev_replace->item_needs_writeback = 1;
702 pr_info("btrfs: suspending dev_replace for unmount\n");
703 break;
704 }
705
706 btrfs_dev_replace_unlock(dev_replace);
707 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
708}
709
710/* resume dev_replace procedure that was interrupted by unmount */
711int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
712{
713 struct task_struct *task;
714 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
715
716 btrfs_dev_replace_lock(dev_replace);
717 switch (dev_replace->replace_state) {
718 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
719 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
720 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
721 btrfs_dev_replace_unlock(dev_replace);
722 return 0;
723 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
724 break;
725 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
726 dev_replace->replace_state =
727 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
728 break;
729 }
730 if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
731 pr_info("btrfs: cannot continue dev_replace, tgtdev is missing\n"
732 "btrfs: you may cancel the operation after 'mount -o degraded'\n");
733 btrfs_dev_replace_unlock(dev_replace);
734 return 0;
735 }
736 btrfs_dev_replace_unlock(dev_replace);
737
738 WARN_ON(atomic_xchg(
739 &fs_info->mutually_exclusive_operation_running, 1));
740 task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
741 return PTR_RET(task);
742}
743
744static int btrfs_dev_replace_kthread(void *data)
745{
746 struct btrfs_fs_info *fs_info = data;
747 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
748 struct btrfs_ioctl_dev_replace_args *status_args;
749 u64 progress;
750
751 status_args = kzalloc(sizeof(*status_args), GFP_NOFS);
752 if (status_args) {
753 btrfs_dev_replace_status(fs_info, status_args);
754 progress = status_args->status.progress_1000;
755 kfree(status_args);
756 do_div(progress, 10);
757 printk_in_rcu(KERN_INFO
758 "btrfs: continuing dev_replace from %s (devid %llu) to %s @%u%%\n",
759 dev_replace->srcdev->missing ? "<missing disk>" :
760 rcu_str_deref(dev_replace->srcdev->name),
761 dev_replace->srcdev->devid,
762 dev_replace->tgtdev ?
763 rcu_str_deref(dev_replace->tgtdev->name) :
764 "<missing target disk>",
765 (unsigned int)progress);
766 }
767 btrfs_dev_replace_continue_on_mount(fs_info);
768 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
769
770 return 0;
771}
772
773static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info)
774{
775 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
776 int ret;
777
778 ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
779 dev_replace->committed_cursor_left,
780 dev_replace->srcdev->total_bytes,
781 &dev_replace->scrub_progress, 0, 1);
782 ret = btrfs_dev_replace_finishing(fs_info, ret);
783 WARN_ON(ret);
784 return 0;
785}
786
787int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
788{
789 if (!dev_replace->is_valid)
790 return 0;
791
792 switch (dev_replace->replace_state) {
793 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
794 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
795 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
796 return 0;
797 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
798 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
799 /*
800 * return true even if tgtdev is missing (this is
801 * something that can happen if the dev_replace
802 * procedure is suspended by an umount and then
803 * the tgtdev is missing (or "btrfs dev scan") was
804 * not called and the the filesystem is remounted
805 * in degraded state. This does not stop the
806 * dev_replace procedure. It needs to be canceled
807 * manually if the cancelation is wanted.
808 */
809 break;
810 }
811 return 1;
812}
813
814void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace)
815{
816 /* the beginning is just an optimization for the typical case */
817 if (atomic_read(&dev_replace->nesting_level) == 0) {
818acquire_lock:
819 /* this is not a nested case where the same thread
820 * is trying to acqurire the same lock twice */
821 mutex_lock(&dev_replace->lock);
822 mutex_lock(&dev_replace->lock_management_lock);
823 dev_replace->lock_owner = current->pid;
824 atomic_inc(&dev_replace->nesting_level);
825 mutex_unlock(&dev_replace->lock_management_lock);
826 return;
827 }
828
829 mutex_lock(&dev_replace->lock_management_lock);
830 if (atomic_read(&dev_replace->nesting_level) > 0 &&
831 dev_replace->lock_owner == current->pid) {
832 WARN_ON(!mutex_is_locked(&dev_replace->lock));
833 atomic_inc(&dev_replace->nesting_level);
834 mutex_unlock(&dev_replace->lock_management_lock);
835 return;
836 }
837
838 mutex_unlock(&dev_replace->lock_management_lock);
839 goto acquire_lock;
840}
841
842void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace)
843{
844 WARN_ON(!mutex_is_locked(&dev_replace->lock));
845 mutex_lock(&dev_replace->lock_management_lock);
846 WARN_ON(atomic_read(&dev_replace->nesting_level) < 1);
847 WARN_ON(dev_replace->lock_owner != current->pid);
848 atomic_dec(&dev_replace->nesting_level);
849 if (atomic_read(&dev_replace->nesting_level) == 0) {
850 dev_replace->lock_owner = 0;
851 mutex_unlock(&dev_replace->lock_management_lock);
852 mutex_unlock(&dev_replace->lock);
853 } else {
854 mutex_unlock(&dev_replace->lock_management_lock);
855 }
856}