btrfs: Fix a deadlock in btrfs_dev_replace_finishing()
[cascardo/linux.git] / fs / btrfs / dev-replace.c
1 /*
2  * Copyright (C) STRATO AG 2012.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/kthread.h>
27 #include <linux/math64.h>
28 #include <asm/div64.h>
29 #include "ctree.h"
30 #include "extent_map.h"
31 #include "disk-io.h"
32 #include "transaction.h"
33 #include "print-tree.h"
34 #include "volumes.h"
35 #include "async-thread.h"
36 #include "check-integrity.h"
37 #include "rcu-string.h"
38 #include "dev-replace.h"
39 #include "sysfs.h"
40
41 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
42                                        int scrub_ret);
43 static 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);
47 static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid,
48                                          char *srcdev_name,
49                                          struct btrfs_device **device);
50 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info);
51 static int btrfs_dev_replace_kthread(void *data);
52 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info);
53
54
55 int 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) {
79 no_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)) {
106                 btrfs_warn(fs_info,
107                         "dev_replace entry found has unexpected size, ignore entry");
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                         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);
155                 }
156                 if (!dev_replace->tgtdev &&
157                     !btrfs_test_opt(dev_root, DEGRADED)) {
158                         ret = -EIO;
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'?",
163                                 BTRFS_DEV_REPLACE_DEVID);
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;
171                                 dev_replace->tgtdev->bytes_used =
172                                         dev_replace->srcdev->bytes_used;
173                         }
174                         dev_replace->tgtdev->is_tgtdev_for_dev_replace = 1;
175                         btrfs_init_dev_replace_tgtdev_for_resume(fs_info,
176                                 dev_replace->tgtdev);
177                 }
178                 break;
179         }
180
181 out:
182         if (path)
183                 btrfs_free_path(path);
184         return ret;
185 }
186
187 /*
188  * called from commit_transaction. Writes changed device replace state to
189  * disk.
190  */
191 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
192                           struct btrfs_fs_info *fs_info)
193 {
194         int ret;
195         struct btrfs_root *dev_root = fs_info->dev_root;
196         struct btrfs_path *path;
197         struct btrfs_key key;
198         struct extent_buffer *eb;
199         struct btrfs_dev_replace_item *ptr;
200         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
201
202         btrfs_dev_replace_lock(dev_replace);
203         if (!dev_replace->is_valid ||
204             !dev_replace->item_needs_writeback) {
205                 btrfs_dev_replace_unlock(dev_replace);
206                 return 0;
207         }
208         btrfs_dev_replace_unlock(dev_replace);
209
210         key.objectid = 0;
211         key.type = BTRFS_DEV_REPLACE_KEY;
212         key.offset = 0;
213
214         path = btrfs_alloc_path();
215         if (!path) {
216                 ret = -ENOMEM;
217                 goto out;
218         }
219         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
220         if (ret < 0) {
221                 btrfs_warn(fs_info, "error %d while searching for dev_replace item!",
222                         ret);
223                 goto out;
224         }
225
226         if (ret == 0 &&
227             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
228                 /*
229                  * need to delete old one and insert a new one.
230                  * Since no attempt is made to recover any old state, if the
231                  * dev_replace state is 'running', the data on the target
232                  * drive is lost.
233                  * It would be possible to recover the state: just make sure
234                  * that the beginning of the item is never changed and always
235                  * contains all the essential information. Then read this
236                  * minimal set of information and use it as a base for the
237                  * new state.
238                  */
239                 ret = btrfs_del_item(trans, dev_root, path);
240                 if (ret != 0) {
241                         btrfs_warn(fs_info, "delete too small dev_replace item failed %d!",
242                                 ret);
243                         goto out;
244                 }
245                 ret = 1;
246         }
247
248         if (ret == 1) {
249                 /* need to insert a new item */
250                 btrfs_release_path(path);
251                 ret = btrfs_insert_empty_item(trans, dev_root, path,
252                                               &key, sizeof(*ptr));
253                 if (ret < 0) {
254                         btrfs_warn(fs_info, "insert dev_replace item failed %d!",
255                                 ret);
256                         goto out;
257                 }
258         }
259
260         eb = path->nodes[0];
261         ptr = btrfs_item_ptr(eb, path->slots[0],
262                              struct btrfs_dev_replace_item);
263
264         btrfs_dev_replace_lock(dev_replace);
265         if (dev_replace->srcdev)
266                 btrfs_set_dev_replace_src_devid(eb, ptr,
267                         dev_replace->srcdev->devid);
268         else
269                 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
270         btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
271                 dev_replace->cont_reading_from_srcdev_mode);
272         btrfs_set_dev_replace_replace_state(eb, ptr,
273                 dev_replace->replace_state);
274         btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
275         btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
276         btrfs_set_dev_replace_num_write_errors(eb, ptr,
277                 atomic64_read(&dev_replace->num_write_errors));
278         btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
279                 atomic64_read(&dev_replace->num_uncorrectable_read_errors));
280         dev_replace->cursor_left_last_write_of_item =
281                 dev_replace->cursor_left;
282         btrfs_set_dev_replace_cursor_left(eb, ptr,
283                 dev_replace->cursor_left_last_write_of_item);
284         btrfs_set_dev_replace_cursor_right(eb, ptr,
285                 dev_replace->cursor_right);
286         dev_replace->item_needs_writeback = 0;
287         btrfs_dev_replace_unlock(dev_replace);
288
289         btrfs_mark_buffer_dirty(eb);
290
291 out:
292         btrfs_free_path(path);
293
294         return ret;
295 }
296
297 void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
298 {
299         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
300
301         dev_replace->committed_cursor_left =
302                 dev_replace->cursor_left_last_write_of_item;
303 }
304
305 int btrfs_dev_replace_start(struct btrfs_root *root,
306                             struct btrfs_ioctl_dev_replace_args *args)
307 {
308         struct btrfs_trans_handle *trans;
309         struct btrfs_fs_info *fs_info = root->fs_info;
310         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
311         int ret;
312         struct btrfs_device *tgt_device = NULL;
313         struct btrfs_device *src_device = NULL;
314
315         if (btrfs_fs_incompat(fs_info, RAID56)) {
316                 btrfs_warn(fs_info, "dev_replace cannot yet handle RAID5/RAID6");
317                 return -EOPNOTSUPP;
318         }
319
320         switch (args->start.cont_reading_from_srcdev_mode) {
321         case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
322         case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
323                 break;
324         default:
325                 return -EINVAL;
326         }
327
328         if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
329             args->start.tgtdev_name[0] == '\0')
330                 return -EINVAL;
331
332         mutex_lock(&fs_info->volume_mutex);
333         ret = btrfs_init_dev_replace_tgtdev(root, args->start.tgtdev_name,
334                                             &tgt_device);
335         if (ret) {
336                 btrfs_err(fs_info, "target device %s is invalid!",
337                        args->start.tgtdev_name);
338                 mutex_unlock(&fs_info->volume_mutex);
339                 return -EINVAL;
340         }
341
342         ret = btrfs_dev_replace_find_srcdev(root, args->start.srcdevid,
343                                             args->start.srcdev_name,
344                                             &src_device);
345         mutex_unlock(&fs_info->volume_mutex);
346         if (ret) {
347                 ret = -EINVAL;
348                 goto leave_no_lock;
349         }
350
351         if (tgt_device->total_bytes < src_device->total_bytes) {
352                 btrfs_err(fs_info, "target device is smaller than source device!");
353                 ret = -EINVAL;
354                 goto leave_no_lock;
355         }
356
357         btrfs_dev_replace_lock(dev_replace);
358         switch (dev_replace->replace_state) {
359         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
360         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
361         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
362                 break;
363         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
364         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
365                 args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
366                 goto leave;
367         }
368
369         dev_replace->cont_reading_from_srcdev_mode =
370                 args->start.cont_reading_from_srcdev_mode;
371         WARN_ON(!src_device);
372         dev_replace->srcdev = src_device;
373         WARN_ON(!tgt_device);
374         dev_replace->tgtdev = tgt_device;
375
376         printk_in_rcu(KERN_INFO
377                       "BTRFS: dev_replace from %s (devid %llu) to %s started\n",
378                       src_device->missing ? "<missing disk>" :
379                         rcu_str_deref(src_device->name),
380                       src_device->devid,
381                       rcu_str_deref(tgt_device->name));
382
383         tgt_device->total_bytes = src_device->total_bytes;
384         tgt_device->disk_total_bytes = src_device->disk_total_bytes;
385         tgt_device->bytes_used = src_device->bytes_used;
386
387         /*
388          * from now on, the writes to the srcdev are all duplicated to
389          * go to the tgtdev as well (refer to btrfs_map_block()).
390          */
391         dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
392         dev_replace->time_started = get_seconds();
393         dev_replace->cursor_left = 0;
394         dev_replace->committed_cursor_left = 0;
395         dev_replace->cursor_left_last_write_of_item = 0;
396         dev_replace->cursor_right = 0;
397         dev_replace->is_valid = 1;
398         dev_replace->item_needs_writeback = 1;
399         args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
400         btrfs_dev_replace_unlock(dev_replace);
401
402         btrfs_wait_ordered_roots(root->fs_info, -1);
403
404         /* force writing the updated state information to disk */
405         trans = btrfs_start_transaction(root, 0);
406         if (IS_ERR(trans)) {
407                 ret = PTR_ERR(trans);
408                 btrfs_dev_replace_lock(dev_replace);
409                 goto leave;
410         }
411
412         ret = btrfs_commit_transaction(trans, root);
413         WARN_ON(ret);
414
415         /* the disk copy procedure reuses the scrub code */
416         ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
417                               src_device->total_bytes,
418                               &dev_replace->scrub_progress, 0, 1);
419
420         ret = btrfs_dev_replace_finishing(root->fs_info, ret);
421         WARN_ON(ret);
422
423         return 0;
424
425 leave:
426         dev_replace->srcdev = NULL;
427         dev_replace->tgtdev = NULL;
428         btrfs_dev_replace_unlock(dev_replace);
429 leave_no_lock:
430         if (tgt_device)
431                 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
432         return ret;
433 }
434
435 /*
436  * blocked until all flighting bios are finished.
437  */
438 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
439 {
440         s64 writers;
441         DEFINE_WAIT(wait);
442
443         set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
444         do {
445                 prepare_to_wait(&fs_info->replace_wait, &wait,
446                                 TASK_UNINTERRUPTIBLE);
447                 writers = percpu_counter_sum(&fs_info->bio_counter);
448                 if (writers)
449                         schedule();
450                 finish_wait(&fs_info->replace_wait, &wait);
451         } while (writers);
452 }
453
454 /*
455  * we have removed target device, it is safe to allow new bios request.
456  */
457 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
458 {
459         clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
460         if (waitqueue_active(&fs_info->replace_wait))
461                 wake_up(&fs_info->replace_wait);
462 }
463
464 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
465                                        int scrub_ret)
466 {
467         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
468         struct btrfs_device *tgt_device;
469         struct btrfs_device *src_device;
470         struct btrfs_root *root = fs_info->tree_root;
471         u8 uuid_tmp[BTRFS_UUID_SIZE];
472         struct btrfs_trans_handle *trans;
473         int ret = 0;
474
475         /* don't allow cancel or unmount to disturb the finishing procedure */
476         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
477
478         btrfs_dev_replace_lock(dev_replace);
479         /* was the operation canceled, or is it finished? */
480         if (dev_replace->replace_state !=
481             BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
482                 btrfs_dev_replace_unlock(dev_replace);
483                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
484                 return 0;
485         }
486
487         tgt_device = dev_replace->tgtdev;
488         src_device = dev_replace->srcdev;
489         btrfs_dev_replace_unlock(dev_replace);
490
491         /*
492          * flush all outstanding I/O and inode extent mappings before the
493          * copy operation is declared as being finished
494          */
495         ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
496         if (ret) {
497                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
498                 return ret;
499         }
500         btrfs_wait_ordered_roots(root->fs_info, -1);
501
502         trans = btrfs_start_transaction(root, 0);
503         if (IS_ERR(trans)) {
504                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
505                 return PTR_ERR(trans);
506         }
507         ret = btrfs_commit_transaction(trans, root);
508         WARN_ON(ret);
509
510         /* keep away write_all_supers() during the finishing procedure */
511         mutex_lock(&root->fs_info->chunk_mutex);
512         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
513         btrfs_dev_replace_lock(dev_replace);
514         dev_replace->replace_state =
515                 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
516                           : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
517         dev_replace->tgtdev = NULL;
518         dev_replace->srcdev = NULL;
519         dev_replace->time_stopped = get_seconds();
520         dev_replace->item_needs_writeback = 1;
521
522         /* replace old device with new one in mapping tree */
523         if (!scrub_ret) {
524                 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
525                                                                 src_device,
526                                                                 tgt_device);
527         } else {
528                 printk_in_rcu(KERN_ERR
529                               "BTRFS: btrfs_scrub_dev(%s, %llu, %s) failed %d\n",
530                               src_device->missing ? "<missing disk>" :
531                                 rcu_str_deref(src_device->name),
532                               src_device->devid,
533                               rcu_str_deref(tgt_device->name), scrub_ret);
534                 btrfs_dev_replace_unlock(dev_replace);
535                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
536                 mutex_unlock(&root->fs_info->chunk_mutex);
537                 if (tgt_device)
538                         btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
539                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
540
541                 return 0;
542         }
543
544         printk_in_rcu(KERN_INFO
545                       "BTRFS: dev_replace from %s (devid %llu) to %s finished\n",
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;
553         tgt_device->bytes_used = src_device->bytes_used;
554         memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
555         memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
556         memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
557         tgt_device->total_bytes = src_device->total_bytes;
558         tgt_device->disk_total_bytes = src_device->disk_total_bytes;
559         tgt_device->bytes_used = src_device->bytes_used;
560         if (fs_info->sb->s_bdev == src_device->bdev)
561                 fs_info->sb->s_bdev = tgt_device->bdev;
562         if (fs_info->fs_devices->latest_bdev == src_device->bdev)
563                 fs_info->fs_devices->latest_bdev = tgt_device->bdev;
564         list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
565         if (src_device->fs_devices->seeding)
566                 fs_info->fs_devices->rw_devices++;
567
568         /* replace the sysfs entry */
569         btrfs_kobj_rm_device(fs_info, src_device);
570         btrfs_kobj_add_device(fs_info, tgt_device);
571
572         btrfs_dev_replace_unlock(dev_replace);
573
574         btrfs_rm_dev_replace_blocked(fs_info);
575
576         btrfs_rm_dev_replace_srcdev(fs_info, src_device);
577
578         btrfs_rm_dev_replace_unblocked(fs_info);
579
580         /*
581          * this is again a consistent state where no dev_replace procedure
582          * is running, the target device is part of the filesystem, the
583          * source device is not part of the filesystem anymore and its 1st
584          * superblock is scratched out so that it is no longer marked to
585          * belong to this filesystem.
586          */
587         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
588         mutex_unlock(&root->fs_info->chunk_mutex);
589
590         /* write back the superblocks */
591         trans = btrfs_start_transaction(root, 0);
592         if (!IS_ERR(trans))
593                 btrfs_commit_transaction(trans, root);
594
595         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
596
597         return 0;
598 }
599
600 static void btrfs_dev_replace_update_device_in_mapping_tree(
601                                                 struct btrfs_fs_info *fs_info,
602                                                 struct btrfs_device *srcdev,
603                                                 struct btrfs_device *tgtdev)
604 {
605         struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
606         struct extent_map *em;
607         struct map_lookup *map;
608         u64 start = 0;
609         int i;
610
611         write_lock(&em_tree->lock);
612         do {
613                 em = lookup_extent_mapping(em_tree, start, (u64)-1);
614                 if (!em)
615                         break;
616                 map = (struct map_lookup *)em->bdev;
617                 for (i = 0; i < map->num_stripes; i++)
618                         if (srcdev == map->stripes[i].dev)
619                                 map->stripes[i].dev = tgtdev;
620                 start = em->start + em->len;
621                 free_extent_map(em);
622         } while (start);
623         write_unlock(&em_tree->lock);
624 }
625
626 static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid,
627                                          char *srcdev_name,
628                                          struct btrfs_device **device)
629 {
630         int ret;
631
632         if (srcdevid) {
633                 ret = 0;
634                 *device = btrfs_find_device(root->fs_info, srcdevid, NULL,
635                                             NULL);
636                 if (!*device)
637                         ret = -ENOENT;
638         } else {
639                 ret = btrfs_find_device_missing_or_by_path(root, srcdev_name,
640                                                            device);
641         }
642         return ret;
643 }
644
645 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
646                               struct btrfs_ioctl_dev_replace_args *args)
647 {
648         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
649
650         btrfs_dev_replace_lock(dev_replace);
651         /* even if !dev_replace_is_valid, the values are good enough for
652          * the replace_status ioctl */
653         args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
654         args->status.replace_state = dev_replace->replace_state;
655         args->status.time_started = dev_replace->time_started;
656         args->status.time_stopped = dev_replace->time_stopped;
657         args->status.num_write_errors =
658                 atomic64_read(&dev_replace->num_write_errors);
659         args->status.num_uncorrectable_read_errors =
660                 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
661         switch (dev_replace->replace_state) {
662         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
663         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
664                 args->status.progress_1000 = 0;
665                 break;
666         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
667                 args->status.progress_1000 = 1000;
668                 break;
669         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
670         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
671                 args->status.progress_1000 = div64_u64(dev_replace->cursor_left,
672                         div64_u64(dev_replace->srcdev->total_bytes, 1000));
673                 break;
674         }
675         btrfs_dev_replace_unlock(dev_replace);
676 }
677
678 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info,
679                              struct btrfs_ioctl_dev_replace_args *args)
680 {
681         args->result = __btrfs_dev_replace_cancel(fs_info);
682         return 0;
683 }
684
685 static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
686 {
687         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
688         struct btrfs_device *tgt_device = NULL;
689         struct btrfs_trans_handle *trans;
690         struct btrfs_root *root = fs_info->tree_root;
691         u64 result;
692         int ret;
693
694         if (fs_info->sb->s_flags & MS_RDONLY)
695                 return -EROFS;
696
697         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
698         btrfs_dev_replace_lock(dev_replace);
699         switch (dev_replace->replace_state) {
700         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
701         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
702         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
703                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
704                 btrfs_dev_replace_unlock(dev_replace);
705                 goto leave;
706         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
707         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
708                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
709                 tgt_device = dev_replace->tgtdev;
710                 dev_replace->tgtdev = NULL;
711                 dev_replace->srcdev = NULL;
712                 break;
713         }
714         dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
715         dev_replace->time_stopped = get_seconds();
716         dev_replace->item_needs_writeback = 1;
717         btrfs_dev_replace_unlock(dev_replace);
718         btrfs_scrub_cancel(fs_info);
719
720         trans = btrfs_start_transaction(root, 0);
721         if (IS_ERR(trans)) {
722                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
723                 return PTR_ERR(trans);
724         }
725         ret = btrfs_commit_transaction(trans, root);
726         WARN_ON(ret);
727         if (tgt_device)
728                 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
729
730 leave:
731         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
732         return result;
733 }
734
735 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
736 {
737         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
738
739         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
740         btrfs_dev_replace_lock(dev_replace);
741         switch (dev_replace->replace_state) {
742         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
743         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
744         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
745         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
746                 break;
747         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
748                 dev_replace->replace_state =
749                         BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
750                 dev_replace->time_stopped = get_seconds();
751                 dev_replace->item_needs_writeback = 1;
752                 btrfs_info(fs_info, "suspending dev_replace for unmount");
753                 break;
754         }
755
756         btrfs_dev_replace_unlock(dev_replace);
757         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
758 }
759
760 /* resume dev_replace procedure that was interrupted by unmount */
761 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
762 {
763         struct task_struct *task;
764         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
765
766         btrfs_dev_replace_lock(dev_replace);
767         switch (dev_replace->replace_state) {
768         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
769         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
770         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
771                 btrfs_dev_replace_unlock(dev_replace);
772                 return 0;
773         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
774                 break;
775         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
776                 dev_replace->replace_state =
777                         BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
778                 break;
779         }
780         if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
781                 btrfs_info(fs_info, "cannot continue dev_replace, tgtdev is missing");
782                 btrfs_info(fs_info,
783                         "you may cancel the operation after 'mount -o degraded'");
784                 btrfs_dev_replace_unlock(dev_replace);
785                 return 0;
786         }
787         btrfs_dev_replace_unlock(dev_replace);
788
789         WARN_ON(atomic_xchg(
790                 &fs_info->mutually_exclusive_operation_running, 1));
791         task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
792         return PTR_ERR_OR_ZERO(task);
793 }
794
795 static int btrfs_dev_replace_kthread(void *data)
796 {
797         struct btrfs_fs_info *fs_info = data;
798         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
799         struct btrfs_ioctl_dev_replace_args *status_args;
800         u64 progress;
801
802         status_args = kzalloc(sizeof(*status_args), GFP_NOFS);
803         if (status_args) {
804                 btrfs_dev_replace_status(fs_info, status_args);
805                 progress = status_args->status.progress_1000;
806                 kfree(status_args);
807                 do_div(progress, 10);
808                 printk_in_rcu(KERN_INFO
809                         "BTRFS: continuing dev_replace from %s (devid %llu) to %s @%u%%\n",
810                         dev_replace->srcdev->missing ? "<missing disk>" :
811                         rcu_str_deref(dev_replace->srcdev->name),
812                         dev_replace->srcdev->devid,
813                         dev_replace->tgtdev ?
814                         rcu_str_deref(dev_replace->tgtdev->name) :
815                         "<missing target disk>",
816                         (unsigned int)progress);
817         }
818         btrfs_dev_replace_continue_on_mount(fs_info);
819         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
820
821         return 0;
822 }
823
824 static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info)
825 {
826         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
827         int ret;
828
829         ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
830                               dev_replace->committed_cursor_left,
831                               dev_replace->srcdev->total_bytes,
832                               &dev_replace->scrub_progress, 0, 1);
833         ret = btrfs_dev_replace_finishing(fs_info, ret);
834         WARN_ON(ret);
835         return 0;
836 }
837
838 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
839 {
840         if (!dev_replace->is_valid)
841                 return 0;
842
843         switch (dev_replace->replace_state) {
844         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
845         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
846         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
847                 return 0;
848         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
849         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
850                 /*
851                  * return true even if tgtdev is missing (this is
852                  * something that can happen if the dev_replace
853                  * procedure is suspended by an umount and then
854                  * the tgtdev is missing (or "btrfs dev scan") was
855                  * not called and the the filesystem is remounted
856                  * in degraded state. This does not stop the
857                  * dev_replace procedure. It needs to be canceled
858                  * manually if the cancelation is wanted.
859                  */
860                 break;
861         }
862         return 1;
863 }
864
865 void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace)
866 {
867         /* the beginning is just an optimization for the typical case */
868         if (atomic_read(&dev_replace->nesting_level) == 0) {
869 acquire_lock:
870                 /* this is not a nested case where the same thread
871                  * is trying to acqurire the same lock twice */
872                 mutex_lock(&dev_replace->lock);
873                 mutex_lock(&dev_replace->lock_management_lock);
874                 dev_replace->lock_owner = current->pid;
875                 atomic_inc(&dev_replace->nesting_level);
876                 mutex_unlock(&dev_replace->lock_management_lock);
877                 return;
878         }
879
880         mutex_lock(&dev_replace->lock_management_lock);
881         if (atomic_read(&dev_replace->nesting_level) > 0 &&
882             dev_replace->lock_owner == current->pid) {
883                 WARN_ON(!mutex_is_locked(&dev_replace->lock));
884                 atomic_inc(&dev_replace->nesting_level);
885                 mutex_unlock(&dev_replace->lock_management_lock);
886                 return;
887         }
888
889         mutex_unlock(&dev_replace->lock_management_lock);
890         goto acquire_lock;
891 }
892
893 void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace)
894 {
895         WARN_ON(!mutex_is_locked(&dev_replace->lock));
896         mutex_lock(&dev_replace->lock_management_lock);
897         WARN_ON(atomic_read(&dev_replace->nesting_level) < 1);
898         WARN_ON(dev_replace->lock_owner != current->pid);
899         atomic_dec(&dev_replace->nesting_level);
900         if (atomic_read(&dev_replace->nesting_level) == 0) {
901                 dev_replace->lock_owner = 0;
902                 mutex_unlock(&dev_replace->lock_management_lock);
903                 mutex_unlock(&dev_replace->lock);
904         } else {
905                 mutex_unlock(&dev_replace->lock_management_lock);
906         }
907 }
908
909 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
910 {
911         percpu_counter_inc(&fs_info->bio_counter);
912 }
913
914 void btrfs_bio_counter_dec(struct btrfs_fs_info *fs_info)
915 {
916         percpu_counter_dec(&fs_info->bio_counter);
917
918         if (waitqueue_active(&fs_info->replace_wait))
919                 wake_up(&fs_info->replace_wait);
920 }
921
922 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
923 {
924         DEFINE_WAIT(wait);
925 again:
926         percpu_counter_inc(&fs_info->bio_counter);
927         if (test_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state)) {
928                 btrfs_bio_counter_dec(fs_info);
929                 wait_event(fs_info->replace_wait,
930                            !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
931                                      &fs_info->fs_state));
932                 goto again;
933         }
934
935 }