149e6139182b804ed6b784fdba9322fc983a41dc
[sfrench/cifs-2.6.git] / fs / btrfs / dev-replace.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) STRATO AG 2012.  All rights reserved.
4  */
5
6 #include <linux/sched.h>
7 #include <linux/bio.h>
8 #include <linux/slab.h>
9 #include <linux/blkdev.h>
10 #include <linux/kthread.h>
11 #include <linux/math64.h>
12 #include "ctree.h"
13 #include "extent_map.h"
14 #include "disk-io.h"
15 #include "transaction.h"
16 #include "print-tree.h"
17 #include "volumes.h"
18 #include "async-thread.h"
19 #include "check-integrity.h"
20 #include "rcu-string.h"
21 #include "dev-replace.h"
22 #include "sysfs.h"
23
24 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
25                                        int scrub_ret);
26 static void btrfs_dev_replace_update_device_in_mapping_tree(
27                                                 struct btrfs_fs_info *fs_info,
28                                                 struct btrfs_device *srcdev,
29                                                 struct btrfs_device *tgtdev);
30 static int btrfs_dev_replace_kthread(void *data);
31
32 int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
33 {
34         struct btrfs_key key;
35         struct btrfs_root *dev_root = fs_info->dev_root;
36         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
37         struct extent_buffer *eb;
38         int slot;
39         int ret = 0;
40         struct btrfs_path *path = NULL;
41         int item_size;
42         struct btrfs_dev_replace_item *ptr;
43         u64 src_devid;
44
45         path = btrfs_alloc_path();
46         if (!path) {
47                 ret = -ENOMEM;
48                 goto out;
49         }
50
51         key.objectid = 0;
52         key.type = BTRFS_DEV_REPLACE_KEY;
53         key.offset = 0;
54         ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
55         if (ret) {
56 no_valid_dev_replace_entry_found:
57                 ret = 0;
58                 dev_replace->replace_state =
59                         BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED;
60                 dev_replace->cont_reading_from_srcdev_mode =
61                     BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
62                 dev_replace->time_started = 0;
63                 dev_replace->time_stopped = 0;
64                 atomic64_set(&dev_replace->num_write_errors, 0);
65                 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
66                 dev_replace->cursor_left = 0;
67                 dev_replace->committed_cursor_left = 0;
68                 dev_replace->cursor_left_last_write_of_item = 0;
69                 dev_replace->cursor_right = 0;
70                 dev_replace->srcdev = NULL;
71                 dev_replace->tgtdev = NULL;
72                 dev_replace->is_valid = 0;
73                 dev_replace->item_needs_writeback = 0;
74                 goto out;
75         }
76         slot = path->slots[0];
77         eb = path->nodes[0];
78         item_size = btrfs_item_size_nr(eb, slot);
79         ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
80
81         if (item_size != sizeof(struct btrfs_dev_replace_item)) {
82                 btrfs_warn(fs_info,
83                         "dev_replace entry found has unexpected size, ignore entry");
84                 goto no_valid_dev_replace_entry_found;
85         }
86
87         src_devid = btrfs_dev_replace_src_devid(eb, ptr);
88         dev_replace->cont_reading_from_srcdev_mode =
89                 btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
90         dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
91         dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
92         dev_replace->time_stopped =
93                 btrfs_dev_replace_time_stopped(eb, ptr);
94         atomic64_set(&dev_replace->num_write_errors,
95                      btrfs_dev_replace_num_write_errors(eb, ptr));
96         atomic64_set(&dev_replace->num_uncorrectable_read_errors,
97                      btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
98         dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
99         dev_replace->committed_cursor_left = dev_replace->cursor_left;
100         dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
101         dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
102         dev_replace->is_valid = 1;
103
104         dev_replace->item_needs_writeback = 0;
105         switch (dev_replace->replace_state) {
106         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
107         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
108         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
109                 dev_replace->srcdev = NULL;
110                 dev_replace->tgtdev = NULL;
111                 break;
112         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
113         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
114                 dev_replace->srcdev = btrfs_find_device(fs_info->fs_devices,
115                                                 src_devid, NULL, NULL, true);
116                 dev_replace->tgtdev = btrfs_find_device(fs_info->fs_devices,
117                                                         BTRFS_DEV_REPLACE_DEVID,
118                                                         NULL, NULL, true);
119                 /*
120                  * allow 'btrfs dev replace_cancel' if src/tgt device is
121                  * missing
122                  */
123                 if (!dev_replace->srcdev &&
124                     !btrfs_test_opt(fs_info, DEGRADED)) {
125                         ret = -EIO;
126                         btrfs_warn(fs_info,
127                            "cannot mount because device replace operation is ongoing and");
128                         btrfs_warn(fs_info,
129                            "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
130                            src_devid);
131                 }
132                 if (!dev_replace->tgtdev &&
133                     !btrfs_test_opt(fs_info, DEGRADED)) {
134                         ret = -EIO;
135                         btrfs_warn(fs_info,
136                            "cannot mount because device replace operation is ongoing and");
137                         btrfs_warn(fs_info,
138                            "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
139                                 BTRFS_DEV_REPLACE_DEVID);
140                 }
141                 if (dev_replace->tgtdev) {
142                         if (dev_replace->srcdev) {
143                                 dev_replace->tgtdev->total_bytes =
144                                         dev_replace->srcdev->total_bytes;
145                                 dev_replace->tgtdev->disk_total_bytes =
146                                         dev_replace->srcdev->disk_total_bytes;
147                                 dev_replace->tgtdev->commit_total_bytes =
148                                         dev_replace->srcdev->commit_total_bytes;
149                                 dev_replace->tgtdev->bytes_used =
150                                         dev_replace->srcdev->bytes_used;
151                                 dev_replace->tgtdev->commit_bytes_used =
152                                         dev_replace->srcdev->commit_bytes_used;
153                         }
154                         set_bit(BTRFS_DEV_STATE_REPLACE_TGT,
155                                 &dev_replace->tgtdev->dev_state);
156
157                         WARN_ON(fs_info->fs_devices->rw_devices == 0);
158                         dev_replace->tgtdev->io_width = fs_info->sectorsize;
159                         dev_replace->tgtdev->io_align = fs_info->sectorsize;
160                         dev_replace->tgtdev->sector_size = fs_info->sectorsize;
161                         dev_replace->tgtdev->fs_info = fs_info;
162                         set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
163                                 &dev_replace->tgtdev->dev_state);
164                 }
165                 break;
166         }
167
168 out:
169         btrfs_free_path(path);
170         return ret;
171 }
172
173 /*
174  * Initialize a new device for device replace target from a given source dev
175  * and path.
176  *
177  * Return 0 and new device in @device_out, otherwise return < 0
178  */
179 static int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
180                                   const char *device_path,
181                                   struct btrfs_device *srcdev,
182                                   struct btrfs_device **device_out)
183 {
184         struct btrfs_device *device;
185         struct block_device *bdev;
186         struct list_head *devices;
187         struct rcu_string *name;
188         u64 devid = BTRFS_DEV_REPLACE_DEVID;
189         int ret = 0;
190
191         *device_out = NULL;
192         if (fs_info->fs_devices->seeding) {
193                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
194                 return -EINVAL;
195         }
196
197         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
198                                   fs_info->bdev_holder);
199         if (IS_ERR(bdev)) {
200                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
201                 return PTR_ERR(bdev);
202         }
203
204         sync_blockdev(bdev);
205
206         devices = &fs_info->fs_devices->devices;
207         list_for_each_entry(device, devices, dev_list) {
208                 if (device->bdev == bdev) {
209                         btrfs_err(fs_info,
210                                   "target device is in the filesystem!");
211                         ret = -EEXIST;
212                         goto error;
213                 }
214         }
215
216
217         if (i_size_read(bdev->bd_inode) <
218             btrfs_device_get_total_bytes(srcdev)) {
219                 btrfs_err(fs_info,
220                           "target device is smaller than source device!");
221                 ret = -EINVAL;
222                 goto error;
223         }
224
225
226         device = btrfs_alloc_device(NULL, &devid, NULL);
227         if (IS_ERR(device)) {
228                 ret = PTR_ERR(device);
229                 goto error;
230         }
231
232         name = rcu_string_strdup(device_path, GFP_KERNEL);
233         if (!name) {
234                 btrfs_free_device(device);
235                 ret = -ENOMEM;
236                 goto error;
237         }
238         rcu_assign_pointer(device->name, name);
239
240         set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
241         device->generation = 0;
242         device->io_width = fs_info->sectorsize;
243         device->io_align = fs_info->sectorsize;
244         device->sector_size = fs_info->sectorsize;
245         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
246         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
247         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
248         device->commit_total_bytes = srcdev->commit_total_bytes;
249         device->commit_bytes_used = device->bytes_used;
250         device->fs_info = fs_info;
251         device->bdev = bdev;
252         set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
253         set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
254         device->mode = FMODE_EXCL;
255         device->dev_stats_valid = 1;
256         set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
257         device->fs_devices = fs_info->fs_devices;
258
259         mutex_lock(&fs_info->fs_devices->device_list_mutex);
260         list_add(&device->dev_list, &fs_info->fs_devices->devices);
261         fs_info->fs_devices->num_devices++;
262         fs_info->fs_devices->open_devices++;
263         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
264
265         *device_out = device;
266         return 0;
267
268 error:
269         blkdev_put(bdev, FMODE_EXCL);
270         return ret;
271 }
272
273 /*
274  * called from commit_transaction. Writes changed device replace state to
275  * disk.
276  */
277 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans)
278 {
279         struct btrfs_fs_info *fs_info = trans->fs_info;
280         int ret;
281         struct btrfs_root *dev_root = fs_info->dev_root;
282         struct btrfs_path *path;
283         struct btrfs_key key;
284         struct extent_buffer *eb;
285         struct btrfs_dev_replace_item *ptr;
286         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
287
288         down_read(&dev_replace->rwsem);
289         if (!dev_replace->is_valid ||
290             !dev_replace->item_needs_writeback) {
291                 up_read(&dev_replace->rwsem);
292                 return 0;
293         }
294         up_read(&dev_replace->rwsem);
295
296         key.objectid = 0;
297         key.type = BTRFS_DEV_REPLACE_KEY;
298         key.offset = 0;
299
300         path = btrfs_alloc_path();
301         if (!path) {
302                 ret = -ENOMEM;
303                 goto out;
304         }
305         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
306         if (ret < 0) {
307                 btrfs_warn(fs_info,
308                            "error %d while searching for dev_replace item!",
309                            ret);
310                 goto out;
311         }
312
313         if (ret == 0 &&
314             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
315                 /*
316                  * need to delete old one and insert a new one.
317                  * Since no attempt is made to recover any old state, if the
318                  * dev_replace state is 'running', the data on the target
319                  * drive is lost.
320                  * It would be possible to recover the state: just make sure
321                  * that the beginning of the item is never changed and always
322                  * contains all the essential information. Then read this
323                  * minimal set of information and use it as a base for the
324                  * new state.
325                  */
326                 ret = btrfs_del_item(trans, dev_root, path);
327                 if (ret != 0) {
328                         btrfs_warn(fs_info,
329                                    "delete too small dev_replace item failed %d!",
330                                    ret);
331                         goto out;
332                 }
333                 ret = 1;
334         }
335
336         if (ret == 1) {
337                 /* need to insert a new item */
338                 btrfs_release_path(path);
339                 ret = btrfs_insert_empty_item(trans, dev_root, path,
340                                               &key, sizeof(*ptr));
341                 if (ret < 0) {
342                         btrfs_warn(fs_info,
343                                    "insert dev_replace item failed %d!", ret);
344                         goto out;
345                 }
346         }
347
348         eb = path->nodes[0];
349         ptr = btrfs_item_ptr(eb, path->slots[0],
350                              struct btrfs_dev_replace_item);
351
352         down_write(&dev_replace->rwsem);
353         if (dev_replace->srcdev)
354                 btrfs_set_dev_replace_src_devid(eb, ptr,
355                         dev_replace->srcdev->devid);
356         else
357                 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
358         btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
359                 dev_replace->cont_reading_from_srcdev_mode);
360         btrfs_set_dev_replace_replace_state(eb, ptr,
361                 dev_replace->replace_state);
362         btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
363         btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
364         btrfs_set_dev_replace_num_write_errors(eb, ptr,
365                 atomic64_read(&dev_replace->num_write_errors));
366         btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
367                 atomic64_read(&dev_replace->num_uncorrectable_read_errors));
368         dev_replace->cursor_left_last_write_of_item =
369                 dev_replace->cursor_left;
370         btrfs_set_dev_replace_cursor_left(eb, ptr,
371                 dev_replace->cursor_left_last_write_of_item);
372         btrfs_set_dev_replace_cursor_right(eb, ptr,
373                 dev_replace->cursor_right);
374         dev_replace->item_needs_writeback = 0;
375         up_write(&dev_replace->rwsem);
376
377         btrfs_mark_buffer_dirty(eb);
378
379 out:
380         btrfs_free_path(path);
381
382         return ret;
383 }
384
385 static char* btrfs_dev_name(struct btrfs_device *device)
386 {
387         if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
388                 return "<missing disk>";
389         else
390                 return rcu_str_deref(device->name);
391 }
392
393 static int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
394                 const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
395                 int read_src)
396 {
397         struct btrfs_root *root = fs_info->dev_root;
398         struct btrfs_trans_handle *trans;
399         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
400         int ret;
401         struct btrfs_device *tgt_device = NULL;
402         struct btrfs_device *src_device = NULL;
403         bool need_unlock;
404
405         src_device = btrfs_find_device_by_devspec(fs_info, srcdevid,
406                                                   srcdev_name);
407         if (IS_ERR(src_device))
408                 return PTR_ERR(src_device);
409
410         if (btrfs_pinned_by_swapfile(fs_info, src_device)) {
411                 btrfs_warn_in_rcu(fs_info,
412           "cannot replace device %s (devid %llu) due to active swapfile",
413                         btrfs_dev_name(src_device), src_device->devid);
414                 return -ETXTBSY;
415         }
416
417         /*
418          * Here we commit the transaction to make sure commit_total_bytes
419          * of all the devices are updated.
420          */
421         trans = btrfs_attach_transaction(root);
422         if (!IS_ERR(trans)) {
423                 ret = btrfs_commit_transaction(trans);
424                 if (ret)
425                         return ret;
426         } else if (PTR_ERR(trans) != -ENOENT) {
427                 return PTR_ERR(trans);
428         }
429
430         ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
431                                             src_device, &tgt_device);
432         if (ret)
433                 return ret;
434
435         need_unlock = true;
436         down_write(&dev_replace->rwsem);
437         switch (dev_replace->replace_state) {
438         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
439         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
440         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
441                 break;
442         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
443         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
444                 ASSERT(0);
445                 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
446                 goto leave;
447         }
448
449         dev_replace->cont_reading_from_srcdev_mode = read_src;
450         dev_replace->srcdev = src_device;
451         dev_replace->tgtdev = tgt_device;
452
453         btrfs_info_in_rcu(fs_info,
454                       "dev_replace from %s (devid %llu) to %s started",
455                       btrfs_dev_name(src_device),
456                       src_device->devid,
457                       rcu_str_deref(tgt_device->name));
458
459         /*
460          * from now on, the writes to the srcdev are all duplicated to
461          * go to the tgtdev as well (refer to btrfs_map_block()).
462          */
463         dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
464         dev_replace->time_started = ktime_get_real_seconds();
465         dev_replace->cursor_left = 0;
466         dev_replace->committed_cursor_left = 0;
467         dev_replace->cursor_left_last_write_of_item = 0;
468         dev_replace->cursor_right = 0;
469         dev_replace->is_valid = 1;
470         dev_replace->item_needs_writeback = 1;
471         atomic64_set(&dev_replace->num_write_errors, 0);
472         atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
473         up_write(&dev_replace->rwsem);
474         need_unlock = false;
475
476         ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device);
477         if (ret)
478                 btrfs_err(fs_info, "kobj add dev failed %d", ret);
479
480         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
481
482         /* force writing the updated state information to disk */
483         trans = btrfs_start_transaction(root, 0);
484         if (IS_ERR(trans)) {
485                 ret = PTR_ERR(trans);
486                 need_unlock = true;
487                 down_write(&dev_replace->rwsem);
488                 dev_replace->replace_state =
489                         BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
490                 dev_replace->srcdev = NULL;
491                 dev_replace->tgtdev = NULL;
492                 goto leave;
493         }
494
495         ret = btrfs_commit_transaction(trans);
496         WARN_ON(ret);
497
498         /* the disk copy procedure reuses the scrub code */
499         ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
500                               btrfs_device_get_total_bytes(src_device),
501                               &dev_replace->scrub_progress, 0, 1);
502
503         ret = btrfs_dev_replace_finishing(fs_info, ret);
504         if (ret == -EINPROGRESS) {
505                 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
506         } else if (ret != -ECANCELED) {
507                 WARN_ON(ret);
508         }
509
510         return ret;
511
512 leave:
513         if (need_unlock)
514                 up_write(&dev_replace->rwsem);
515         btrfs_destroy_dev_replace_tgtdev(tgt_device);
516         return ret;
517 }
518
519 int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
520                             struct btrfs_ioctl_dev_replace_args *args)
521 {
522         int ret;
523
524         switch (args->start.cont_reading_from_srcdev_mode) {
525         case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
526         case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
527                 break;
528         default:
529                 return -EINVAL;
530         }
531
532         if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
533             args->start.tgtdev_name[0] == '\0')
534                 return -EINVAL;
535
536         ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name,
537                                         args->start.srcdevid,
538                                         args->start.srcdev_name,
539                                         args->start.cont_reading_from_srcdev_mode);
540         args->result = ret;
541         /* don't warn if EINPROGRESS, someone else might be running scrub */
542         if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS ||
543             ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR)
544                 return 0;
545
546         return ret;
547 }
548
549 /*
550  * blocked until all in-flight bios operations are finished.
551  */
552 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
553 {
554         set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
555         wait_event(fs_info->dev_replace.replace_wait, !percpu_counter_sum(
556                    &fs_info->dev_replace.bio_counter));
557 }
558
559 /*
560  * we have removed target device, it is safe to allow new bios request.
561  */
562 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
563 {
564         clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
565         wake_up(&fs_info->dev_replace.replace_wait);
566 }
567
568 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
569                                        int scrub_ret)
570 {
571         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
572         struct btrfs_device *tgt_device;
573         struct btrfs_device *src_device;
574         struct btrfs_root *root = fs_info->tree_root;
575         u8 uuid_tmp[BTRFS_UUID_SIZE];
576         struct btrfs_trans_handle *trans;
577         int ret = 0;
578
579         /* don't allow cancel or unmount to disturb the finishing procedure */
580         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
581
582         down_read(&dev_replace->rwsem);
583         /* was the operation canceled, or is it finished? */
584         if (dev_replace->replace_state !=
585             BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
586                 up_read(&dev_replace->rwsem);
587                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
588                 return 0;
589         }
590
591         tgt_device = dev_replace->tgtdev;
592         src_device = dev_replace->srcdev;
593         up_read(&dev_replace->rwsem);
594
595         /*
596          * flush all outstanding I/O and inode extent mappings before the
597          * copy operation is declared as being finished
598          */
599         ret = btrfs_start_delalloc_roots(fs_info, -1);
600         if (ret) {
601                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
602                 return ret;
603         }
604         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
605
606         /*
607          * We have to use this loop approach because at this point src_device
608          * has to be available for transaction commit to complete, yet new
609          * chunks shouldn't be allocated on the device.
610          */
611         while (1) {
612                 trans = btrfs_start_transaction(root, 0);
613                 if (IS_ERR(trans)) {
614                         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
615                         return PTR_ERR(trans);
616                 }
617                 ret = btrfs_commit_transaction(trans);
618                 WARN_ON(ret);
619
620                 /* Prevent write_all_supers() during the finishing procedure */
621                 mutex_lock(&fs_info->fs_devices->device_list_mutex);
622                 /* Prevent new chunks being allocated on the source device */
623                 mutex_lock(&fs_info->chunk_mutex);
624
625                 if (!list_empty(&src_device->post_commit_list)) {
626                         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
627                         mutex_unlock(&fs_info->chunk_mutex);
628                 } else {
629                         break;
630                 }
631         }
632
633         down_write(&dev_replace->rwsem);
634         dev_replace->replace_state =
635                 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
636                           : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
637         dev_replace->tgtdev = NULL;
638         dev_replace->srcdev = NULL;
639         dev_replace->time_stopped = ktime_get_real_seconds();
640         dev_replace->item_needs_writeback = 1;
641
642         /* replace old device with new one in mapping tree */
643         if (!scrub_ret) {
644                 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
645                                                                 src_device,
646                                                                 tgt_device);
647         } else {
648                 if (scrub_ret != -ECANCELED)
649                         btrfs_err_in_rcu(fs_info,
650                                  "btrfs_scrub_dev(%s, %llu, %s) failed %d",
651                                  btrfs_dev_name(src_device),
652                                  src_device->devid,
653                                  rcu_str_deref(tgt_device->name), scrub_ret);
654                 up_write(&dev_replace->rwsem);
655                 mutex_unlock(&fs_info->chunk_mutex);
656                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
657                 btrfs_rm_dev_replace_blocked(fs_info);
658                 if (tgt_device)
659                         btrfs_destroy_dev_replace_tgtdev(tgt_device);
660                 btrfs_rm_dev_replace_unblocked(fs_info);
661                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
662
663                 return scrub_ret;
664         }
665
666         btrfs_info_in_rcu(fs_info,
667                           "dev_replace from %s (devid %llu) to %s finished",
668                           btrfs_dev_name(src_device),
669                           src_device->devid,
670                           rcu_str_deref(tgt_device->name));
671         clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state);
672         tgt_device->devid = src_device->devid;
673         src_device->devid = BTRFS_DEV_REPLACE_DEVID;
674         memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
675         memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
676         memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
677         btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
678         btrfs_device_set_disk_total_bytes(tgt_device,
679                                           src_device->disk_total_bytes);
680         btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
681         tgt_device->commit_total_bytes = src_device->commit_total_bytes;
682         tgt_device->commit_bytes_used = src_device->bytes_used;
683
684         btrfs_assign_next_active_device(src_device, tgt_device);
685
686         list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
687         fs_info->fs_devices->rw_devices++;
688
689         up_write(&dev_replace->rwsem);
690         btrfs_rm_dev_replace_blocked(fs_info);
691
692         btrfs_rm_dev_replace_remove_srcdev(src_device);
693
694         btrfs_rm_dev_replace_unblocked(fs_info);
695
696         /*
697          * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will
698          * update on-disk dev stats value during commit transaction
699          */
700         atomic_inc(&tgt_device->dev_stats_ccnt);
701
702         /*
703          * this is again a consistent state where no dev_replace procedure
704          * is running, the target device is part of the filesystem, the
705          * source device is not part of the filesystem anymore and its 1st
706          * superblock is scratched out so that it is no longer marked to
707          * belong to this filesystem.
708          */
709         mutex_unlock(&fs_info->chunk_mutex);
710         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
711
712         /* replace the sysfs entry */
713         btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device);
714         btrfs_rm_dev_replace_free_srcdev(src_device);
715
716         /* write back the superblocks */
717         trans = btrfs_start_transaction(root, 0);
718         if (!IS_ERR(trans))
719                 btrfs_commit_transaction(trans);
720
721         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
722
723         return 0;
724 }
725
726 static void btrfs_dev_replace_update_device_in_mapping_tree(
727                                                 struct btrfs_fs_info *fs_info,
728                                                 struct btrfs_device *srcdev,
729                                                 struct btrfs_device *tgtdev)
730 {
731         struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
732         struct extent_map *em;
733         struct map_lookup *map;
734         u64 start = 0;
735         int i;
736
737         write_lock(&em_tree->lock);
738         do {
739                 em = lookup_extent_mapping(em_tree, start, (u64)-1);
740                 if (!em)
741                         break;
742                 map = em->map_lookup;
743                 for (i = 0; i < map->num_stripes; i++)
744                         if (srcdev == map->stripes[i].dev)
745                                 map->stripes[i].dev = tgtdev;
746                 start = em->start + em->len;
747                 free_extent_map(em);
748         } while (start);
749         write_unlock(&em_tree->lock);
750 }
751
752 /*
753  * Read progress of device replace status according to the state and last
754  * stored position. The value format is the same as for
755  * btrfs_dev_replace::progress_1000
756  */
757 static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info)
758 {
759         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
760         u64 ret = 0;
761
762         switch (dev_replace->replace_state) {
763         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
764         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
765                 ret = 0;
766                 break;
767         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
768                 ret = 1000;
769                 break;
770         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
771         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
772                 ret = div64_u64(dev_replace->cursor_left,
773                                 div_u64(btrfs_device_get_total_bytes(
774                                                 dev_replace->srcdev), 1000));
775                 break;
776         }
777
778         return ret;
779 }
780
781 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
782                               struct btrfs_ioctl_dev_replace_args *args)
783 {
784         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
785
786         down_read(&dev_replace->rwsem);
787         /* even if !dev_replace_is_valid, the values are good enough for
788          * the replace_status ioctl */
789         args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
790         args->status.replace_state = dev_replace->replace_state;
791         args->status.time_started = dev_replace->time_started;
792         args->status.time_stopped = dev_replace->time_stopped;
793         args->status.num_write_errors =
794                 atomic64_read(&dev_replace->num_write_errors);
795         args->status.num_uncorrectable_read_errors =
796                 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
797         args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
798         up_read(&dev_replace->rwsem);
799 }
800
801 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
802 {
803         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
804         struct btrfs_device *tgt_device = NULL;
805         struct btrfs_device *src_device = NULL;
806         struct btrfs_trans_handle *trans;
807         struct btrfs_root *root = fs_info->tree_root;
808         int result;
809         int ret;
810
811         if (sb_rdonly(fs_info->sb))
812                 return -EROFS;
813
814         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
815         down_write(&dev_replace->rwsem);
816         switch (dev_replace->replace_state) {
817         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
818         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
819         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
820                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
821                 up_write(&dev_replace->rwsem);
822                 break;
823         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
824                 tgt_device = dev_replace->tgtdev;
825                 src_device = dev_replace->srcdev;
826                 up_write(&dev_replace->rwsem);
827                 ret = btrfs_scrub_cancel(fs_info);
828                 if (ret < 0) {
829                         result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
830                 } else {
831                         result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
832                         /*
833                          * btrfs_dev_replace_finishing() will handle the
834                          * cleanup part
835                          */
836                         btrfs_info_in_rcu(fs_info,
837                                 "dev_replace from %s (devid %llu) to %s canceled",
838                                 btrfs_dev_name(src_device), src_device->devid,
839                                 btrfs_dev_name(tgt_device));
840                 }
841                 break;
842         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
843                 /*
844                  * Scrub doing the replace isn't running so we need to do the
845                  * cleanup step of btrfs_dev_replace_finishing() here
846                  */
847                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
848                 tgt_device = dev_replace->tgtdev;
849                 src_device = dev_replace->srcdev;
850                 dev_replace->tgtdev = NULL;
851                 dev_replace->srcdev = NULL;
852                 dev_replace->replace_state =
853                                 BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
854                 dev_replace->time_stopped = ktime_get_real_seconds();
855                 dev_replace->item_needs_writeback = 1;
856
857                 up_write(&dev_replace->rwsem);
858
859                 /* Scrub for replace must not be running in suspended state */
860                 ret = btrfs_scrub_cancel(fs_info);
861                 ASSERT(ret != -ENOTCONN);
862
863                 trans = btrfs_start_transaction(root, 0);
864                 if (IS_ERR(trans)) {
865                         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
866                         return PTR_ERR(trans);
867                 }
868                 ret = btrfs_commit_transaction(trans);
869                 WARN_ON(ret);
870
871                 btrfs_info_in_rcu(fs_info,
872                 "suspended dev_replace from %s (devid %llu) to %s canceled",
873                         btrfs_dev_name(src_device), src_device->devid,
874                         btrfs_dev_name(tgt_device));
875
876                 if (tgt_device)
877                         btrfs_destroy_dev_replace_tgtdev(tgt_device);
878                 break;
879         default:
880                 up_write(&dev_replace->rwsem);
881                 result = -EINVAL;
882         }
883
884         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
885         return result;
886 }
887
888 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
889 {
890         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
891
892         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
893         down_write(&dev_replace->rwsem);
894
895         switch (dev_replace->replace_state) {
896         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
897         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
898         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
899         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
900                 break;
901         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
902                 dev_replace->replace_state =
903                         BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
904                 dev_replace->time_stopped = ktime_get_real_seconds();
905                 dev_replace->item_needs_writeback = 1;
906                 btrfs_info(fs_info, "suspending dev_replace for unmount");
907                 break;
908         }
909
910         up_write(&dev_replace->rwsem);
911         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
912 }
913
914 /* resume dev_replace procedure that was interrupted by unmount */
915 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
916 {
917         struct task_struct *task;
918         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
919
920         down_write(&dev_replace->rwsem);
921
922         switch (dev_replace->replace_state) {
923         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
924         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
925         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
926                 up_write(&dev_replace->rwsem);
927                 return 0;
928         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
929                 break;
930         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
931                 dev_replace->replace_state =
932                         BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
933                 break;
934         }
935         if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
936                 btrfs_info(fs_info,
937                            "cannot continue dev_replace, tgtdev is missing");
938                 btrfs_info(fs_info,
939                            "you may cancel the operation after 'mount -o degraded'");
940                 dev_replace->replace_state =
941                                         BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
942                 up_write(&dev_replace->rwsem);
943                 return 0;
944         }
945         up_write(&dev_replace->rwsem);
946
947         /*
948          * This could collide with a paused balance, but the exclusive op logic
949          * should never allow both to start and pause. We don't want to allow
950          * dev-replace to start anyway.
951          */
952         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
953                 down_write(&dev_replace->rwsem);
954                 dev_replace->replace_state =
955                                         BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
956                 up_write(&dev_replace->rwsem);
957                 btrfs_info(fs_info,
958                 "cannot resume dev-replace, other exclusive operation running");
959                 return 0;
960         }
961
962         task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
963         return PTR_ERR_OR_ZERO(task);
964 }
965
966 static int btrfs_dev_replace_kthread(void *data)
967 {
968         struct btrfs_fs_info *fs_info = data;
969         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
970         u64 progress;
971         int ret;
972
973         progress = btrfs_dev_replace_progress(fs_info);
974         progress = div_u64(progress, 10);
975         btrfs_info_in_rcu(fs_info,
976                 "continuing dev_replace from %s (devid %llu) to target %s @%u%%",
977                 btrfs_dev_name(dev_replace->srcdev),
978                 dev_replace->srcdev->devid,
979                 btrfs_dev_name(dev_replace->tgtdev),
980                 (unsigned int)progress);
981
982         ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
983                               dev_replace->committed_cursor_left,
984                               btrfs_device_get_total_bytes(dev_replace->srcdev),
985                               &dev_replace->scrub_progress, 0, 1);
986         ret = btrfs_dev_replace_finishing(fs_info, ret);
987         WARN_ON(ret && ret != -ECANCELED);
988
989         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
990         return 0;
991 }
992
993 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
994 {
995         if (!dev_replace->is_valid)
996                 return 0;
997
998         switch (dev_replace->replace_state) {
999         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
1000         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
1001         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
1002                 return 0;
1003         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1004         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1005                 /*
1006                  * return true even if tgtdev is missing (this is
1007                  * something that can happen if the dev_replace
1008                  * procedure is suspended by an umount and then
1009                  * the tgtdev is missing (or "btrfs dev scan") was
1010                  * not called and the filesystem is remounted
1011                  * in degraded state. This does not stop the
1012                  * dev_replace procedure. It needs to be canceled
1013                  * manually if the cancellation is wanted.
1014                  */
1015                 break;
1016         }
1017         return 1;
1018 }
1019
1020 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
1021 {
1022         percpu_counter_inc(&fs_info->dev_replace.bio_counter);
1023 }
1024
1025 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
1026 {
1027         percpu_counter_sub(&fs_info->dev_replace.bio_counter, amount);
1028         cond_wake_up_nomb(&fs_info->dev_replace.replace_wait);
1029 }
1030
1031 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
1032 {
1033         while (1) {
1034                 percpu_counter_inc(&fs_info->dev_replace.bio_counter);
1035                 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1036                                      &fs_info->fs_state)))
1037                         break;
1038
1039                 btrfs_bio_counter_dec(fs_info);
1040                 wait_event(fs_info->dev_replace.replace_wait,
1041                            !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1042                                      &fs_info->fs_state));
1043         }
1044 }