projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'dev/pending-changes' of git://git.kernel.org/pub/scm/linux/kernel/git...
[deliverable/linux.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
49688107 319 if (btrfs_fs_incompat(fs_info, RAID56)) {
efe120a0 320 btrfs_warn(fs_info, "dev_replace cannot yet handle RAID5/RAID6");
c81d5767 321 return -EOPNOTSUPP;
49688107
SB		 322 }
323
e93c89c1
SB		 324 switch (args->start.cont_reading_from_srcdev_mode) {
325 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
326 case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
327 break;
328 default:
329 return -EINVAL;
330 }
331
332 if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
333 args->start.tgtdev_name[0] == '\0')
334 return -EINVAL;
335
935e5cc9
MX		 336 /*
337 * Here we commit the transaction to make sure commit_total_bytes
338 * of all the devices are updated.
339 */
340 trans = btrfs_attach_transaction(root);
341 if (!IS_ERR(trans)) {
342 ret = btrfs_commit_transaction(trans, root);
343 if (ret)
344 return ret;
345 } else if (PTR_ERR(trans) != -ENOENT) {
346 return PTR_ERR(trans);
347 }
348
349 /* the disk copy procedure reuses the scrub code */
e93c89c1 350 mutex_lock(&fs_info->volume_mutex);
e93c89c1
SB		 351 ret = btrfs_dev_replace_find_srcdev(root, args->start.srcdevid,
352 args->start.srcdev_name,
353 &src_device);
e93c89c1 354 if (ret) {
1c43366d
MX		 355 mutex_unlock(&fs_info->volume_mutex);
356 return ret;
e93c89c1
SB		 357 }
358
1c43366d
MX		 359 ret = btrfs_init_dev_replace_tgtdev(root, args->start.tgtdev_name,
360 src_device, &tgt_device);
361 mutex_unlock(&fs_info->volume_mutex);
362 if (ret)
363 return ret;
e93c89c1
SB		 364
365 btrfs_dev_replace_lock(dev_replace);
366 switch (dev_replace->replace_state) {
367 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
368 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
369 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
370 break;
371 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
372 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
373 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
374 goto leave;
375 }
376
377 dev_replace->cont_reading_from_srcdev_mode =
378 args->start.cont_reading_from_srcdev_mode;
379 WARN_ON(!src_device);
380 dev_replace->srcdev = src_device;
381 WARN_ON(!tgt_device);
382 dev_replace->tgtdev = tgt_device;
383
384 printk_in_rcu(KERN_INFO
efe120a0 385 "BTRFS: dev_replace from %s (devid %llu) to %s started\n",
e93c89c1
SB		 386 src_device->missing ? "<missing disk>" :
387 rcu_str_deref(src_device->name),
388 src_device->devid,
389 rcu_str_deref(tgt_device->name));
390
e93c89c1
SB		 391 /*
392 * from now on, the writes to the srcdev are all duplicated to
393 * go to the tgtdev as well (refer to btrfs_map_block()).
394 */
395 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
4546bcae 396 dev_replace->time_started = get_seconds();
e93c89c1
SB		 397 dev_replace->cursor_left = 0;
398 dev_replace->committed_cursor_left = 0;
399 dev_replace->cursor_left_last_write_of_item = 0;
400 dev_replace->cursor_right = 0;
401 dev_replace->is_valid = 1;
402 dev_replace->item_needs_writeback = 1;
403 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
404 btrfs_dev_replace_unlock(dev_replace);
405
b0244199 406 btrfs_wait_ordered_roots(root->fs_info, -1);
e93c89c1
SB		 407
408 /* force writing the updated state information to disk */
409 trans = btrfs_start_transaction(root, 0);
410 if (IS_ERR(trans)) {
411 ret = PTR_ERR(trans);
412 btrfs_dev_replace_lock(dev_replace);
413 goto leave;
414 }
415
416 ret = btrfs_commit_transaction(trans, root);
417 WARN_ON(ret);
418
419 /* the disk copy procedure reuses the scrub code */
420 ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
7cc8e58d 421 btrfs_device_get_total_bytes(src_device),
e93c89c1
SB		 422 &dev_replace->scrub_progress, 0, 1);
423
424 ret = btrfs_dev_replace_finishing(root->fs_info, ret);
2fc9f6ba
EG		 425 /* don't warn if EINPROGRESS, someone else might be running scrub */
426 if (ret == -EINPROGRESS) {
427 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
428 ret = 0;
429 } else {
430 WARN_ON(ret);
431 }
e93c89c1 432
2fc9f6ba 433 return ret;
e93c89c1
SB		 434
435leave:
436 dev_replace->srcdev = NULL;
437 dev_replace->tgtdev = NULL;
438 btrfs_dev_replace_unlock(dev_replace);
1c43366d 439 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
e93c89c1
SB		 440 return ret;
441}
442
c404e0dc
MX		 443/*
444 * blocked until all flighting bios are finished.
445 */
446static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
447{
448 s64 writers;
449 DEFINE_WAIT(wait);
450
451 set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
452 do {
453 prepare_to_wait(&fs_info->replace_wait, &wait,
454 TASK_UNINTERRUPTIBLE);
455 writers = percpu_counter_sum(&fs_info->bio_counter);
456 if (writers)
457 schedule();
458 finish_wait(&fs_info->replace_wait, &wait);
459 } while (writers);
460}
461
462/*
463 * we have removed target device, it is safe to allow new bios request.
464 */
465static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
466{
467 clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
468 if (waitqueue_active(&fs_info->replace_wait))
469 wake_up(&fs_info->replace_wait);
470}
471
e93c89c1
SB		 472static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
473 int scrub_ret)
474{
475 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
476 struct btrfs_device *tgt_device;
477 struct btrfs_device *src_device;
478 struct btrfs_root *root = fs_info->tree_root;
479 u8 uuid_tmp[BTRFS_UUID_SIZE];
480 struct btrfs_trans_handle *trans;
481 int ret = 0;
482
483 /* don't allow cancel or unmount to disturb the finishing procedure */
484 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
485
486 btrfs_dev_replace_lock(dev_replace);
487 /* was the operation canceled, or is it finished? */
488 if (dev_replace->replace_state !=
489 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
490 btrfs_dev_replace_unlock(dev_replace);
491 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
492 return 0;
493 }
494
495 tgt_device = dev_replace->tgtdev;
496 src_device = dev_replace->srcdev;
497 btrfs_dev_replace_unlock(dev_replace);
498
e93c89c1
SB		 499 /*
500 * flush all outstanding I/O and inode extent mappings before the
501 * copy operation is declared as being finished
502 */
6c255e67 503 ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
3edb2a68
MX		 504 if (ret) {
505 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
506 return ret;
507 }
b0244199 508 btrfs_wait_ordered_roots(root->fs_info, -1);
e93c89c1
SB		 509
510 trans = btrfs_start_transaction(root, 0);
511 if (IS_ERR(trans)) {
512 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
513 return PTR_ERR(trans);
514 }
515 ret = btrfs_commit_transaction(trans, root);
516 WARN_ON(ret);
517
67a2c45e 518 mutex_lock(&uuid_mutex);
e93c89c1
SB		 519 /* keep away write_all_supers() during the finishing procedure */
520 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2196d6e8 521 mutex_lock(&root->fs_info->chunk_mutex);
e93c89c1
SB		 522 btrfs_dev_replace_lock(dev_replace);
523 dev_replace->replace_state =
524 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
525 : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
526 dev_replace->tgtdev = NULL;
527 dev_replace->srcdev = NULL;
4546bcae 528 dev_replace->time_stopped = get_seconds();
e93c89c1
SB		 529 dev_replace->item_needs_writeback = 1;
530
c404e0dc
MX		 531 /* replace old device with new one in mapping tree */
532 if (!scrub_ret) {
533 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
534 src_device,
535 tgt_device);
536 } else {
e93c89c1 537 printk_in_rcu(KERN_ERR
efe120a0 538 "BTRFS: btrfs_scrub_dev(%s, %llu, %s) failed %d\n",
e93c89c1
SB		 539 src_device->missing ? "<missing disk>" :
540 rcu_str_deref(src_device->name),
541 src_device->devid,
542 rcu_str_deref(tgt_device->name), scrub_ret);
543 btrfs_dev_replace_unlock(dev_replace);
391cd9df 544 mutex_unlock(&root->fs_info->chunk_mutex);
2196d6e8 545 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
67a2c45e 546 mutex_unlock(&uuid_mutex);
e93c89c1
SB		 547 if (tgt_device)
548 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
549 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
550
2fc9f6ba 551 return scrub_ret;
e93c89c1
SB		 552 }
553
554 printk_in_rcu(KERN_INFO
de4c296f 555 "BTRFS: dev_replace from %s (devid %llu) to %s finished\n",
e93c89c1
SB		 556 src_device->missing ? "<missing disk>" :
557 rcu_str_deref(src_device->name),
558 src_device->devid,
559 rcu_str_deref(tgt_device->name));
560 tgt_device->is_tgtdev_for_dev_replace = 0;
561 tgt_device->devid = src_device->devid;
562 src_device->devid = BTRFS_DEV_REPLACE_DEVID;
e93c89c1
SB		 563 memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
564 memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
565 memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
7cc8e58d
MX		 566 btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
567 btrfs_device_set_disk_total_bytes(tgt_device,
568 src_device->disk_total_bytes);
569 btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
935e5cc9
MX		 570 ASSERT(list_empty(&src_device->resized_list));
571 tgt_device->commit_total_bytes = src_device->commit_total_bytes;
ce7213c7 572 tgt_device->commit_bytes_used = src_device->bytes_used;
e93c89c1
SB		 573 if (fs_info->sb->s_bdev == src_device->bdev)
574 fs_info->sb->s_bdev = tgt_device->bdev;
575 if (fs_info->fs_devices->latest_bdev == src_device->bdev)
576 fs_info->fs_devices->latest_bdev = tgt_device->bdev;
577 list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
82372bc8 578 fs_info->fs_devices->rw_devices++;
e93c89c1 579
49c6f736
AJ		 580 /* replace the sysfs entry */
581 btrfs_kobj_rm_device(fs_info, src_device);
582 btrfs_kobj_add_device(fs_info, tgt_device);
583
12b894cb
QW		 584 btrfs_dev_replace_unlock(dev_replace);
585
c404e0dc
MX		 586 btrfs_rm_dev_replace_blocked(fs_info);
587
e93c89c1 588 btrfs_rm_dev_replace_srcdev(fs_info, src_device);
1357272f 589
c404e0dc
MX		 590 btrfs_rm_dev_replace_unblocked(fs_info);
591
e93c89c1
SB		 592 /*
593 * this is again a consistent state where no dev_replace procedure
594 * is running, the target device is part of the filesystem, the
595 * source device is not part of the filesystem anymore and its 1st
596 * superblock is scratched out so that it is no longer marked to
597 * belong to this filesystem.
598 */
391cd9df 599 mutex_unlock(&root->fs_info->chunk_mutex);
2196d6e8 600 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
67a2c45e 601 mutex_unlock(&uuid_mutex);
e93c89c1
SB		 602
603 /* write back the superblocks */
604 trans = btrfs_start_transaction(root, 0);
605 if (!IS_ERR(trans))
606 btrfs_commit_transaction(trans, root);
607
608 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
609
610 return 0;
611}
612
613static void btrfs_dev_replace_update_device_in_mapping_tree(
614 struct btrfs_fs_info *fs_info,
615 struct btrfs_device *srcdev,
616 struct btrfs_device *tgtdev)
617{
618 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
619 struct extent_map *em;
620 struct map_lookup *map;
621 u64 start = 0;
622 int i;
623
624 write_lock(&em_tree->lock);
625 do {
626 em = lookup_extent_mapping(em_tree, start, (u64)-1);
627 if (!em)
628 break;
629 map = (struct map_lookup *)em->bdev;
630 for (i = 0; i < map->num_stripes; i++)
631 if (srcdev == map->stripes[i].dev)
632 map->stripes[i].dev = tgtdev;
633 start = em->start + em->len;
634 free_extent_map(em);
635 } while (start);
636 write_unlock(&em_tree->lock);
637}
638
639static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid,
640 char *srcdev_name,
641 struct btrfs_device **device)
642{
643 int ret;
644
645 if (srcdevid) {
646 ret = 0;
647 *device = btrfs_find_device(root->fs_info, srcdevid, NULL,
648 NULL);
649 if (!*device)
650 ret = -ENOENT;
651 } else {
652 ret = btrfs_find_device_missing_or_by_path(root, srcdev_name,
653 device);
654 }
655 return ret;
656}
657
658void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
659 struct btrfs_ioctl_dev_replace_args *args)
660{
661 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
7cc8e58d 662 struct btrfs_device *srcdev;
e93c89c1
SB		 663
664 btrfs_dev_replace_lock(dev_replace);
665 /* even if !dev_replace_is_valid, the values are good enough for
666 * the replace_status ioctl */
667 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
668 args->status.replace_state = dev_replace->replace_state;
669 args->status.time_started = dev_replace->time_started;
670 args->status.time_stopped = dev_replace->time_stopped;
671 args->status.num_write_errors =
672 atomic64_read(&dev_replace->num_write_errors);
673 args->status.num_uncorrectable_read_errors =
674 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
675 switch (dev_replace->replace_state) {
676 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
677 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
678 args->status.progress_1000 = 0;
679 break;
680 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
681 args->status.progress_1000 = 1000;
682 break;
683 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
684 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
7cc8e58d 685 srcdev = dev_replace->srcdev;
e93c89c1 686 args->status.progress_1000 = div64_u64(dev_replace->cursor_left,
7cc8e58d 687 div64_u64(btrfs_device_get_total_bytes(srcdev), 1000));
e93c89c1
SB		 688 break;
689 }
690 btrfs_dev_replace_unlock(dev_replace);
691}
692
693int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info,
694 struct btrfs_ioctl_dev_replace_args *args)
695{
696 args->result = __btrfs_dev_replace_cancel(fs_info);
697 return 0;
698}
699
700static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
701{
702 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
703 struct btrfs_device *tgt_device = NULL;
704 struct btrfs_trans_handle *trans;
705 struct btrfs_root *root = fs_info->tree_root;
706 u64 result;
707 int ret;
708
e649e587
ID		 709 if (fs_info->sb->s_flags & MS_RDONLY)
710 return -EROFS;
711
e93c89c1
SB		 712 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
713 btrfs_dev_replace_lock(dev_replace);
714 switch (dev_replace->replace_state) {
715 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
716 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
717 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
718 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
719 btrfs_dev_replace_unlock(dev_replace);
720 goto leave;
721 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
722 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
723 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
724 tgt_device = dev_replace->tgtdev;
725 dev_replace->tgtdev = NULL;
726 dev_replace->srcdev = NULL;
727 break;
728 }
729 dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
4546bcae 730 dev_replace->time_stopped = get_seconds();
e93c89c1
SB		 731 dev_replace->item_needs_writeback = 1;
732 btrfs_dev_replace_unlock(dev_replace);
733 btrfs_scrub_cancel(fs_info);
734
735 trans = btrfs_start_transaction(root, 0);
736 if (IS_ERR(trans)) {
737 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
738 return PTR_ERR(trans);
739 }
740 ret = btrfs_commit_transaction(trans, root);
741 WARN_ON(ret);
742 if (tgt_device)
743 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
744
745leave:
746 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
747 return result;
748}
749
750void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
751{
752 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
753
754 mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
755 btrfs_dev_replace_lock(dev_replace);
756 switch (dev_replace->replace_state) {
757 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
758 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
759 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
760 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
761 break;
762 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
763 dev_replace->replace_state =
764 BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
4546bcae 765 dev_replace->time_stopped = get_seconds();
e93c89c1 766 dev_replace->item_needs_writeback = 1;
efe120a0 767 btrfs_info(fs_info, "suspending dev_replace for unmount");
e93c89c1
SB		 768 break;
769 }
770
771 btrfs_dev_replace_unlock(dev_replace);
772 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
773}
774
775/* resume dev_replace procedure that was interrupted by unmount */
776int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
777{
778 struct task_struct *task;
779 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
780
781 btrfs_dev_replace_lock(dev_replace);
782 switch (dev_replace->replace_state) {
783 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
784 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
785 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
786 btrfs_dev_replace_unlock(dev_replace);
787 return 0;
788 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
789 break;
790 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
791 dev_replace->replace_state =
792 BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
793 break;
794 }
795 if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
efe120a0
FH		 796 btrfs_info(fs_info, "cannot continue dev_replace, tgtdev is missing");
797 btrfs_info(fs_info,
798 "you may cancel the operation after 'mount -o degraded'");
e93c89c1
SB		 799 btrfs_dev_replace_unlock(dev_replace);
800 return 0;
801 }
802 btrfs_dev_replace_unlock(dev_replace);
803
804 WARN_ON(atomic_xchg(
805 &fs_info->mutually_exclusive_operation_running, 1));
806 task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
8c6ffba0 807 return PTR_ERR_OR_ZERO(task);
e93c89c1
SB		 808}
809
810static int btrfs_dev_replace_kthread(void *data)
811{
812 struct btrfs_fs_info *fs_info = data;
813 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
814 struct btrfs_ioctl_dev_replace_args *status_args;
815 u64 progress;
816
817 status_args = kzalloc(sizeof(*status_args), GFP_NOFS);
818 if (status_args) {
819 btrfs_dev_replace_status(fs_info, status_args);
820 progress = status_args->status.progress_1000;
821 kfree(status_args);
822 do_div(progress, 10);
823 printk_in_rcu(KERN_INFO
efe120a0
FH		 824 "BTRFS: continuing dev_replace from %s (devid %llu) to %s @%u%%\n",
825 dev_replace->srcdev->missing ? "<missing disk>" :
826 rcu_str_deref(dev_replace->srcdev->name),
827 dev_replace->srcdev->devid,
828 dev_replace->tgtdev ?
829 rcu_str_deref(dev_replace->tgtdev->name) :
830 "<missing target disk>",
831 (unsigned int)progress);
e93c89c1
SB		 832 }
833 btrfs_dev_replace_continue_on_mount(fs_info);
834 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
835
836 return 0;
837}
838
839static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info)
840{
841 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
842 int ret;
843
844 ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
845 dev_replace->committed_cursor_left,
7cc8e58d 846 btrfs_device_get_total_bytes(dev_replace->srcdev),
e93c89c1
SB		 847 &dev_replace->scrub_progress, 0, 1);
848 ret = btrfs_dev_replace_finishing(fs_info, ret);
849 WARN_ON(ret);
850 return 0;
851}
852
853int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
854{
855 if (!dev_replace->is_valid)
856 return 0;
857
858 switch (dev_replace->replace_state) {
859 case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
860 case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
861 case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
862 return 0;
863 case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
864 case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
865 /*
866 * return true even if tgtdev is missing (this is
867 * something that can happen if the dev_replace
868 * procedure is suspended by an umount and then
869 * the tgtdev is missing (or "btrfs dev scan") was
870 * not called and the the filesystem is remounted
871 * in degraded state. This does not stop the
872 * dev_replace procedure. It needs to be canceled
873 * manually if the cancelation is wanted.
874 */
875 break;
876 }
877 return 1;
878}
879
880void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace)
881{
882 /* the beginning is just an optimization for the typical case */
883 if (atomic_read(&dev_replace->nesting_level) == 0) {
884acquire_lock:
885 /* this is not a nested case where the same thread
886 * is trying to acqurire the same lock twice */
887 mutex_lock(&dev_replace->lock);
888 mutex_lock(&dev_replace->lock_management_lock);
889 dev_replace->lock_owner = current->pid;
890 atomic_inc(&dev_replace->nesting_level);
891 mutex_unlock(&dev_replace->lock_management_lock);
892 return;
893 }
894
895 mutex_lock(&dev_replace->lock_management_lock);
896 if (atomic_read(&dev_replace->nesting_level) > 0 &&
897 dev_replace->lock_owner == current->pid) {
898 WARN_ON(!mutex_is_locked(&dev_replace->lock));
899 atomic_inc(&dev_replace->nesting_level);
900 mutex_unlock(&dev_replace->lock_management_lock);
901 return;
902 }
903
904 mutex_unlock(&dev_replace->lock_management_lock);
905 goto acquire_lock;
906}
907
908void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace)
909{
910 WARN_ON(!mutex_is_locked(&dev_replace->lock));
911 mutex_lock(&dev_replace->lock_management_lock);
912 WARN_ON(atomic_read(&dev_replace->nesting_level) < 1);
913 WARN_ON(dev_replace->lock_owner != current->pid);
914 atomic_dec(&dev_replace->nesting_level);
915 if (atomic_read(&dev_replace->nesting_level) == 0) {
916 dev_replace->lock_owner = 0;
917 mutex_unlock(&dev_replace->lock_management_lock);
918 mutex_unlock(&dev_replace->lock);
919 } else {
920 mutex_unlock(&dev_replace->lock_management_lock);
921 }
922}
c404e0dc
MX		 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_dec(struct btrfs_fs_info *fs_info)
930{
931 percpu_counter_dec(&fs_info->bio_counter);
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 DEFINE_WAIT(wait);
940again:
941 percpu_counter_inc(&fs_info->bio_counter);
942 if (test_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state)) {
943 btrfs_bio_counter_dec(fs_info);
944 wait_event(fs_info->replace_wait,
945 !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
946 &fs_info->fs_state));
947 goto again;
948 }
949
950}
Linux kernel - Deliverables
This page took 0.168742 seconds and 5 git commands to generate.