1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
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/iocontext.h>
24#include <linux/capability.h>
25#include <linux/ratelimit.h>
26#include <linux/kthread.h>
27#include <linux/raid/pq.h>
28#include <linux/semaphore.h>
29#include <linux/uuid.h>
30#include <asm/div64.h>
31#include "ctree.h"
32#include "extent_map.h"
33#include "disk-io.h"
34#include "transaction.h"
35#include "print-tree.h"
36#include "volumes.h"
37#include "raid56.h"
38#include "async-thread.h"
39#include "check-integrity.h"
40#include "rcu-string.h"
41#include "math.h"
42#include "dev-replace.h"
43#include "sysfs.h"
44
45const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
46 [BTRFS_RAID_RAID10] = {
47 .sub_stripes = 2,
48 .dev_stripes = 1,
49 .devs_max = 0,
50 .devs_min = 4,
51 .tolerated_failures = 1,
52 .devs_increment = 2,
53 .ncopies = 2,
54 },
55 [BTRFS_RAID_RAID1] = {
56 .sub_stripes = 1,
57 .dev_stripes = 1,
58 .devs_max = 2,
59 .devs_min = 2,
60 .tolerated_failures = 1,
61 .devs_increment = 2,
62 .ncopies = 2,
63 },
64 [BTRFS_RAID_DUP] = {
65 .sub_stripes = 1,
66 .dev_stripes = 2,
67 .devs_max = 1,
68 .devs_min = 1,
69 .tolerated_failures = 0,
70 .devs_increment = 1,
71 .ncopies = 2,
72 },
73 [BTRFS_RAID_RAID0] = {
74 .sub_stripes = 1,
75 .dev_stripes = 1,
76 .devs_max = 0,
77 .devs_min = 2,
78 .tolerated_failures = 0,
79 .devs_increment = 1,
80 .ncopies = 1,
81 },
82 [BTRFS_RAID_SINGLE] = {
83 .sub_stripes = 1,
84 .dev_stripes = 1,
85 .devs_max = 1,
86 .devs_min = 1,
87 .tolerated_failures = 0,
88 .devs_increment = 1,
89 .ncopies = 1,
90 },
91 [BTRFS_RAID_RAID5] = {
92 .sub_stripes = 1,
93 .dev_stripes = 1,
94 .devs_max = 0,
95 .devs_min = 2,
96 .tolerated_failures = 1,
97 .devs_increment = 1,
98 .ncopies = 2,
99 },
100 [BTRFS_RAID_RAID6] = {
101 .sub_stripes = 1,
102 .dev_stripes = 1,
103 .devs_max = 0,
104 .devs_min = 3,
105 .tolerated_failures = 2,
106 .devs_increment = 1,
107 .ncopies = 3,
108 },
109};
110
111const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = {
112 [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10,
113 [BTRFS_RAID_RAID1] = BTRFS_BLOCK_GROUP_RAID1,
114 [BTRFS_RAID_DUP] = BTRFS_BLOCK_GROUP_DUP,
115 [BTRFS_RAID_RAID0] = BTRFS_BLOCK_GROUP_RAID0,
116 [BTRFS_RAID_SINGLE] = 0,
117 [BTRFS_RAID_RAID5] = BTRFS_BLOCK_GROUP_RAID5,
118 [BTRFS_RAID_RAID6] = BTRFS_BLOCK_GROUP_RAID6,
119};
120
121
122
123
124
125
126const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = {
127 [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
128 [BTRFS_RAID_RAID1] = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
129 [BTRFS_RAID_DUP] = 0,
130 [BTRFS_RAID_RAID0] = 0,
131 [BTRFS_RAID_SINGLE] = 0,
132 [BTRFS_RAID_RAID5] = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
133 [BTRFS_RAID_RAID6] = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
134};
135
136static int init_first_rw_device(struct btrfs_trans_handle *trans,
137 struct btrfs_root *root,
138 struct btrfs_device *device);
139static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
140static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
141static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
142static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
143
144DEFINE_MUTEX(uuid_mutex);
145static LIST_HEAD(fs_uuids);
146struct list_head *btrfs_get_fs_uuids(void)
147{
148 return &fs_uuids;
149}
150
151static struct btrfs_fs_devices *__alloc_fs_devices(void)
152{
153 struct btrfs_fs_devices *fs_devs;
154
155 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
156 if (!fs_devs)
157 return ERR_PTR(-ENOMEM);
158
159 mutex_init(&fs_devs->device_list_mutex);
160
161 INIT_LIST_HEAD(&fs_devs->devices);
162 INIT_LIST_HEAD(&fs_devs->resized_devices);
163 INIT_LIST_HEAD(&fs_devs->alloc_list);
164 INIT_LIST_HEAD(&fs_devs->list);
165
166 return fs_devs;
167}
168
169
170
171
172
173
174
175
176
177
178static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
179{
180 struct btrfs_fs_devices *fs_devs;
181
182 fs_devs = __alloc_fs_devices();
183 if (IS_ERR(fs_devs))
184 return fs_devs;
185
186 if (fsid)
187 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
188 else
189 generate_random_uuid(fs_devs->fsid);
190
191 return fs_devs;
192}
193
194static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
195{
196 struct btrfs_device *device;
197 WARN_ON(fs_devices->opened);
198 while (!list_empty(&fs_devices->devices)) {
199 device = list_entry(fs_devices->devices.next,
200 struct btrfs_device, dev_list);
201 list_del(&device->dev_list);
202 rcu_string_free(device->name);
203 kfree(device);
204 }
205 kfree(fs_devices);
206}
207
208static void btrfs_kobject_uevent(struct block_device *bdev,
209 enum kobject_action action)
210{
211 int ret;
212
213 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
214 if (ret)
215 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
216 action,
217 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
218 &disk_to_dev(bdev->bd_disk)->kobj);
219}
220
221void btrfs_cleanup_fs_uuids(void)
222{
223 struct btrfs_fs_devices *fs_devices;
224
225 while (!list_empty(&fs_uuids)) {
226 fs_devices = list_entry(fs_uuids.next,
227 struct btrfs_fs_devices, list);
228 list_del(&fs_devices->list);
229 free_fs_devices(fs_devices);
230 }
231}
232
233static struct btrfs_device *__alloc_device(void)
234{
235 struct btrfs_device *dev;
236
237 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
238 if (!dev)
239 return ERR_PTR(-ENOMEM);
240
241 INIT_LIST_HEAD(&dev->dev_list);
242 INIT_LIST_HEAD(&dev->dev_alloc_list);
243 INIT_LIST_HEAD(&dev->resized_list);
244
245 spin_lock_init(&dev->io_lock);
246
247 spin_lock_init(&dev->reada_lock);
248 atomic_set(&dev->reada_in_flight, 0);
249 atomic_set(&dev->dev_stats_ccnt, 0);
250 btrfs_device_data_ordered_init(dev);
251 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
252 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
253
254 return dev;
255}
256
257static noinline struct btrfs_device *__find_device(struct list_head *head,
258 u64 devid, u8 *uuid)
259{
260 struct btrfs_device *dev;
261
262 list_for_each_entry(dev, head, dev_list) {
263 if (dev->devid == devid &&
264 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
265 return dev;
266 }
267 }
268 return NULL;
269}
270
271static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
272{
273 struct btrfs_fs_devices *fs_devices;
274
275 list_for_each_entry(fs_devices, &fs_uuids, list) {
276 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
277 return fs_devices;
278 }
279 return NULL;
280}
281
282static int
283btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
284 int flush, struct block_device **bdev,
285 struct buffer_head **bh)
286{
287 int ret;
288
289 *bdev = blkdev_get_by_path(device_path, flags, holder);
290
291 if (IS_ERR(*bdev)) {
292 ret = PTR_ERR(*bdev);
293 goto error;
294 }
295
296 if (flush)
297 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
298 ret = set_blocksize(*bdev, 4096);
299 if (ret) {
300 blkdev_put(*bdev, flags);
301 goto error;
302 }
303 invalidate_bdev(*bdev);
304 *bh = btrfs_read_dev_super(*bdev);
305 if (IS_ERR(*bh)) {
306 ret = PTR_ERR(*bh);
307 blkdev_put(*bdev, flags);
308 goto error;
309 }
310
311 return 0;
312
313error:
314 *bdev = NULL;
315 *bh = NULL;
316 return ret;
317}
318
319static void requeue_list(struct btrfs_pending_bios *pending_bios,
320 struct bio *head, struct bio *tail)
321{
322
323 struct bio *old_head;
324
325 old_head = pending_bios->head;
326 pending_bios->head = head;
327 if (pending_bios->tail)
328 tail->bi_next = old_head;
329 else
330 pending_bios->tail = tail;
331}
332
333
334
335
336
337
338
339
340
341
342
343
344static noinline void run_scheduled_bios(struct btrfs_device *device)
345{
346 struct bio *pending;
347 struct backing_dev_info *bdi;
348 struct btrfs_fs_info *fs_info;
349 struct btrfs_pending_bios *pending_bios;
350 struct bio *tail;
351 struct bio *cur;
352 int again = 0;
353 unsigned long num_run;
354 unsigned long batch_run = 0;
355 unsigned long limit;
356 unsigned long last_waited = 0;
357 int force_reg = 0;
358 int sync_pending = 0;
359 struct blk_plug plug;
360
361
362
363
364
365
366
367 blk_start_plug(&plug);
368
369 bdi = blk_get_backing_dev_info(device->bdev);
370 fs_info = device->dev_root->fs_info;
371 limit = btrfs_async_submit_limit(fs_info);
372 limit = limit * 2 / 3;
373
374loop:
375 spin_lock(&device->io_lock);
376
377loop_lock:
378 num_run = 0;
379
380
381
382
383
384
385 if (!force_reg && device->pending_sync_bios.head) {
386 pending_bios = &device->pending_sync_bios;
387 force_reg = 1;
388 } else {
389 pending_bios = &device->pending_bios;
390 force_reg = 0;
391 }
392
393 pending = pending_bios->head;
394 tail = pending_bios->tail;
395 WARN_ON(pending && !tail);
396
397
398
399
400
401
402
403
404
405 if (device->pending_sync_bios.head == NULL &&
406 device->pending_bios.head == NULL) {
407 again = 0;
408 device->running_pending = 0;
409 } else {
410 again = 1;
411 device->running_pending = 1;
412 }
413
414 pending_bios->head = NULL;
415 pending_bios->tail = NULL;
416
417 spin_unlock(&device->io_lock);
418
419 while (pending) {
420
421 rmb();
422
423
424
425 if ((num_run > 32 &&
426 pending_bios != &device->pending_sync_bios &&
427 device->pending_sync_bios.head) ||
428 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
429 device->pending_bios.head)) {
430 spin_lock(&device->io_lock);
431 requeue_list(pending_bios, pending, tail);
432 goto loop_lock;
433 }
434
435 cur = pending;
436 pending = pending->bi_next;
437 cur->bi_next = NULL;
438
439
440
441
442 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
443 waitqueue_active(&fs_info->async_submit_wait))
444 wake_up(&fs_info->async_submit_wait);
445
446 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
447
448
449
450
451
452
453
454
455
456 if (pending_bios == &device->pending_sync_bios) {
457 sync_pending = 1;
458 } else if (sync_pending) {
459 blk_finish_plug(&plug);
460 blk_start_plug(&plug);
461 sync_pending = 0;
462 }
463
464 btrfsic_submit_bio(cur);
465 num_run++;
466 batch_run++;
467
468 cond_resched();
469
470
471
472
473
474
475 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
476 fs_info->fs_devices->open_devices > 1) {
477 struct io_context *ioc;
478
479 ioc = current->io_context;
480
481
482
483
484
485
486
487
488
489
490 if (ioc && ioc->nr_batch_requests > 0 &&
491 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
492 (last_waited == 0 ||
493 ioc->last_waited == last_waited)) {
494
495
496
497
498
499
500 last_waited = ioc->last_waited;
501 cond_resched();
502 continue;
503 }
504 spin_lock(&device->io_lock);
505 requeue_list(pending_bios, pending, tail);
506 device->running_pending = 1;
507
508 spin_unlock(&device->io_lock);
509 btrfs_queue_work(fs_info->submit_workers,
510 &device->work);
511 goto done;
512 }
513
514 if (batch_run % 64 == 0) {
515 blk_finish_plug(&plug);
516 blk_start_plug(&plug);
517 sync_pending = 0;
518 }
519 }
520
521 cond_resched();
522 if (again)
523 goto loop;
524
525 spin_lock(&device->io_lock);
526 if (device->pending_bios.head || device->pending_sync_bios.head)
527 goto loop_lock;
528 spin_unlock(&device->io_lock);
529
530done:
531 blk_finish_plug(&plug);
532}
533
534static void pending_bios_fn(struct btrfs_work *work)
535{
536 struct btrfs_device *device;
537
538 device = container_of(work, struct btrfs_device, work);
539 run_scheduled_bios(device);
540}
541
542
543void btrfs_free_stale_device(struct btrfs_device *cur_dev)
544{
545 struct btrfs_fs_devices *fs_devs;
546 struct btrfs_device *dev;
547
548 if (!cur_dev->name)
549 return;
550
551 list_for_each_entry(fs_devs, &fs_uuids, list) {
552 int del = 1;
553
554 if (fs_devs->opened)
555 continue;
556 if (fs_devs->seeding)
557 continue;
558
559 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
560
561 if (dev == cur_dev)
562 continue;
563 if (!dev->name)
564 continue;
565
566
567
568
569
570
571
572 rcu_read_lock();
573 del = strcmp(rcu_str_deref(dev->name),
574 rcu_str_deref(cur_dev->name));
575 rcu_read_unlock();
576 if (!del)
577 break;
578 }
579
580 if (!del) {
581
582 if (fs_devs->num_devices == 1) {
583 btrfs_sysfs_remove_fsid(fs_devs);
584 list_del(&fs_devs->list);
585 free_fs_devices(fs_devs);
586 } else {
587 fs_devs->num_devices--;
588 list_del(&dev->dev_list);
589 rcu_string_free(dev->name);
590 kfree(dev);
591 }
592 break;
593 }
594 }
595}
596
597
598
599
600
601
602
603
604
605static noinline int device_list_add(const char *path,
606 struct btrfs_super_block *disk_super,
607 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
608{
609 struct btrfs_device *device;
610 struct btrfs_fs_devices *fs_devices;
611 struct rcu_string *name;
612 int ret = 0;
613 u64 found_transid = btrfs_super_generation(disk_super);
614
615 fs_devices = find_fsid(disk_super->fsid);
616 if (!fs_devices) {
617 fs_devices = alloc_fs_devices(disk_super->fsid);
618 if (IS_ERR(fs_devices))
619 return PTR_ERR(fs_devices);
620
621 list_add(&fs_devices->list, &fs_uuids);
622
623 device = NULL;
624 } else {
625 device = __find_device(&fs_devices->devices, devid,
626 disk_super->dev_item.uuid);
627 }
628
629 if (!device) {
630 if (fs_devices->opened)
631 return -EBUSY;
632
633 device = btrfs_alloc_device(NULL, &devid,
634 disk_super->dev_item.uuid);
635 if (IS_ERR(device)) {
636
637 return PTR_ERR(device);
638 }
639
640 name = rcu_string_strdup(path, GFP_NOFS);
641 if (!name) {
642 kfree(device);
643 return -ENOMEM;
644 }
645 rcu_assign_pointer(device->name, name);
646
647 mutex_lock(&fs_devices->device_list_mutex);
648 list_add_rcu(&device->dev_list, &fs_devices->devices);
649 fs_devices->num_devices++;
650 mutex_unlock(&fs_devices->device_list_mutex);
651
652 ret = 1;
653 device->fs_devices = fs_devices;
654 } else if (!device->name || strcmp(device->name->str, path)) {
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681 if (!fs_devices->opened && found_transid < device->generation) {
682
683
684
685
686
687
688
689 return -EEXIST;
690 }
691
692 name = rcu_string_strdup(path, GFP_NOFS);
693 if (!name)
694 return -ENOMEM;
695 rcu_string_free(device->name);
696 rcu_assign_pointer(device->name, name);
697 if (device->missing) {
698 fs_devices->missing_devices--;
699 device->missing = 0;
700 }
701 }
702
703
704
705
706
707
708
709 if (!fs_devices->opened)
710 device->generation = found_transid;
711
712
713
714
715
716 if (ret > 0)
717 btrfs_free_stale_device(device);
718
719 *fs_devices_ret = fs_devices;
720
721 return ret;
722}
723
724static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
725{
726 struct btrfs_fs_devices *fs_devices;
727 struct btrfs_device *device;
728 struct btrfs_device *orig_dev;
729
730 fs_devices = alloc_fs_devices(orig->fsid);
731 if (IS_ERR(fs_devices))
732 return fs_devices;
733
734 mutex_lock(&orig->device_list_mutex);
735 fs_devices->total_devices = orig->total_devices;
736
737
738 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
739 struct rcu_string *name;
740
741 device = btrfs_alloc_device(NULL, &orig_dev->devid,
742 orig_dev->uuid);
743 if (IS_ERR(device))
744 goto error;
745
746
747
748
749
750 if (orig_dev->name) {
751 name = rcu_string_strdup(orig_dev->name->str,
752 GFP_KERNEL);
753 if (!name) {
754 kfree(device);
755 goto error;
756 }
757 rcu_assign_pointer(device->name, name);
758 }
759
760 list_add(&device->dev_list, &fs_devices->devices);
761 device->fs_devices = fs_devices;
762 fs_devices->num_devices++;
763 }
764 mutex_unlock(&orig->device_list_mutex);
765 return fs_devices;
766error:
767 mutex_unlock(&orig->device_list_mutex);
768 free_fs_devices(fs_devices);
769 return ERR_PTR(-ENOMEM);
770}
771
772void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
773{
774 struct btrfs_device *device, *next;
775 struct btrfs_device *latest_dev = NULL;
776
777 mutex_lock(&uuid_mutex);
778again:
779
780 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
781 if (device->in_fs_metadata) {
782 if (!device->is_tgtdev_for_dev_replace &&
783 (!latest_dev ||
784 device->generation > latest_dev->generation)) {
785 latest_dev = device;
786 }
787 continue;
788 }
789
790 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
791
792
793
794
795
796
797
798
799
800
801 if (step == 0 || device->is_tgtdev_for_dev_replace) {
802 continue;
803 }
804 }
805 if (device->bdev) {
806 blkdev_put(device->bdev, device->mode);
807 device->bdev = NULL;
808 fs_devices->open_devices--;
809 }
810 if (device->writeable) {
811 list_del_init(&device->dev_alloc_list);
812 device->writeable = 0;
813 if (!device->is_tgtdev_for_dev_replace)
814 fs_devices->rw_devices--;
815 }
816 list_del_init(&device->dev_list);
817 fs_devices->num_devices--;
818 rcu_string_free(device->name);
819 kfree(device);
820 }
821
822 if (fs_devices->seed) {
823 fs_devices = fs_devices->seed;
824 goto again;
825 }
826
827 fs_devices->latest_bdev = latest_dev->bdev;
828
829 mutex_unlock(&uuid_mutex);
830}
831
832static void __free_device(struct work_struct *work)
833{
834 struct btrfs_device *device;
835
836 device = container_of(work, struct btrfs_device, rcu_work);
837 rcu_string_free(device->name);
838 kfree(device);
839}
840
841static void free_device(struct rcu_head *head)
842{
843 struct btrfs_device *device;
844
845 device = container_of(head, struct btrfs_device, rcu);
846
847 INIT_WORK(&device->rcu_work, __free_device);
848 schedule_work(&device->rcu_work);
849}
850
851static void btrfs_close_bdev(struct btrfs_device *device)
852{
853 if (device->bdev && device->writeable) {
854 sync_blockdev(device->bdev);
855 invalidate_bdev(device->bdev);
856 }
857
858 if (device->bdev)
859 blkdev_put(device->bdev, device->mode);
860}
861
862static void btrfs_prepare_close_one_device(struct btrfs_device *device)
863{
864 struct btrfs_fs_devices *fs_devices = device->fs_devices;
865 struct btrfs_device *new_device;
866 struct rcu_string *name;
867
868 if (device->bdev)
869 fs_devices->open_devices--;
870
871 if (device->writeable &&
872 device->devid != BTRFS_DEV_REPLACE_DEVID) {
873 list_del_init(&device->dev_alloc_list);
874 fs_devices->rw_devices--;
875 }
876
877 if (device->missing)
878 fs_devices->missing_devices--;
879
880 new_device = btrfs_alloc_device(NULL, &device->devid,
881 device->uuid);
882 BUG_ON(IS_ERR(new_device));
883
884
885 if (device->name) {
886 name = rcu_string_strdup(device->name->str, GFP_NOFS);
887 BUG_ON(!name);
888 rcu_assign_pointer(new_device->name, name);
889 }
890
891 list_replace_rcu(&device->dev_list, &new_device->dev_list);
892 new_device->fs_devices = device->fs_devices;
893}
894
895static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
896{
897 struct btrfs_device *device, *tmp;
898 struct list_head pending_put;
899
900 INIT_LIST_HEAD(&pending_put);
901
902 if (--fs_devices->opened > 0)
903 return 0;
904
905 mutex_lock(&fs_devices->device_list_mutex);
906 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
907 btrfs_prepare_close_one_device(device);
908 list_add(&device->dev_list, &pending_put);
909 }
910 mutex_unlock(&fs_devices->device_list_mutex);
911
912
913
914
915
916
917
918 while (!list_empty(&pending_put)) {
919 device = list_first_entry(&pending_put,
920 struct btrfs_device, dev_list);
921 list_del(&device->dev_list);
922 btrfs_close_bdev(device);
923 call_rcu(&device->rcu, free_device);
924 }
925
926 WARN_ON(fs_devices->open_devices);
927 WARN_ON(fs_devices->rw_devices);
928 fs_devices->opened = 0;
929 fs_devices->seeding = 0;
930
931 return 0;
932}
933
934int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
935{
936 struct btrfs_fs_devices *seed_devices = NULL;
937 int ret;
938
939 mutex_lock(&uuid_mutex);
940 ret = __btrfs_close_devices(fs_devices);
941 if (!fs_devices->opened) {
942 seed_devices = fs_devices->seed;
943 fs_devices->seed = NULL;
944 }
945 mutex_unlock(&uuid_mutex);
946
947 while (seed_devices) {
948 fs_devices = seed_devices;
949 seed_devices = fs_devices->seed;
950 __btrfs_close_devices(fs_devices);
951 free_fs_devices(fs_devices);
952 }
953
954
955
956
957
958 rcu_barrier();
959 return ret;
960}
961
962static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
963 fmode_t flags, void *holder)
964{
965 struct request_queue *q;
966 struct block_device *bdev;
967 struct list_head *head = &fs_devices->devices;
968 struct btrfs_device *device;
969 struct btrfs_device *latest_dev = NULL;
970 struct buffer_head *bh;
971 struct btrfs_super_block *disk_super;
972 u64 devid;
973 int seeding = 1;
974 int ret = 0;
975
976 flags |= FMODE_EXCL;
977
978 list_for_each_entry(device, head, dev_list) {
979 if (device->bdev)
980 continue;
981 if (!device->name)
982 continue;
983
984
985 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
986 &bdev, &bh))
987 continue;
988
989 disk_super = (struct btrfs_super_block *)bh->b_data;
990 devid = btrfs_stack_device_id(&disk_super->dev_item);
991 if (devid != device->devid)
992 goto error_brelse;
993
994 if (memcmp(device->uuid, disk_super->dev_item.uuid,
995 BTRFS_UUID_SIZE))
996 goto error_brelse;
997
998 device->generation = btrfs_super_generation(disk_super);
999 if (!latest_dev ||
1000 device->generation > latest_dev->generation)
1001 latest_dev = device;
1002
1003 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
1004 device->writeable = 0;
1005 } else {
1006 device->writeable = !bdev_read_only(bdev);
1007 seeding = 0;
1008 }
1009
1010 q = bdev_get_queue(bdev);
1011 if (blk_queue_discard(q))
1012 device->can_discard = 1;
1013
1014 device->bdev = bdev;
1015 device->in_fs_metadata = 0;
1016 device->mode = flags;
1017
1018 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1019 fs_devices->rotating = 1;
1020
1021 fs_devices->open_devices++;
1022 if (device->writeable &&
1023 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1024 fs_devices->rw_devices++;
1025 list_add(&device->dev_alloc_list,
1026 &fs_devices->alloc_list);
1027 }
1028 brelse(bh);
1029 continue;
1030
1031error_brelse:
1032 brelse(bh);
1033 blkdev_put(bdev, flags);
1034 continue;
1035 }
1036 if (fs_devices->open_devices == 0) {
1037 ret = -EINVAL;
1038 goto out;
1039 }
1040 fs_devices->seeding = seeding;
1041 fs_devices->opened = 1;
1042 fs_devices->latest_bdev = latest_dev->bdev;
1043 fs_devices->total_rw_bytes = 0;
1044out:
1045 return ret;
1046}
1047
1048int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1049 fmode_t flags, void *holder)
1050{
1051 int ret;
1052
1053 mutex_lock(&uuid_mutex);
1054 if (fs_devices->opened) {
1055 fs_devices->opened++;
1056 ret = 0;
1057 } else {
1058 ret = __btrfs_open_devices(fs_devices, flags, holder);
1059 }
1060 mutex_unlock(&uuid_mutex);
1061 return ret;
1062}
1063
1064void btrfs_release_disk_super(struct page *page)
1065{
1066 kunmap(page);
1067 put_page(page);
1068}
1069
1070int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr,
1071 struct page **page, struct btrfs_super_block **disk_super)
1072{
1073 void *p;
1074 pgoff_t index;
1075
1076
1077 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1078 return 1;
1079
1080
1081 if (sizeof(**disk_super) > PAGE_SIZE)
1082 return 1;
1083
1084
1085 index = bytenr >> PAGE_SHIFT;
1086 if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index)
1087 return 1;
1088
1089
1090 *page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
1091 index, GFP_KERNEL);
1092
1093 if (IS_ERR_OR_NULL(*page))
1094 return 1;
1095
1096 p = kmap(*page);
1097
1098
1099 *disk_super = p + (bytenr & ~PAGE_MASK);
1100
1101 if (btrfs_super_bytenr(*disk_super) != bytenr ||
1102 btrfs_super_magic(*disk_super) != BTRFS_MAGIC) {
1103 btrfs_release_disk_super(*page);
1104 return 1;
1105 }
1106
1107 if ((*disk_super)->label[0] &&
1108 (*disk_super)->label[BTRFS_LABEL_SIZE - 1])
1109 (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0';
1110
1111 return 0;
1112}
1113
1114
1115
1116
1117
1118
1119int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
1120 struct btrfs_fs_devices **fs_devices_ret)
1121{
1122 struct btrfs_super_block *disk_super;
1123 struct block_device *bdev;
1124 struct page *page;
1125 int ret = -EINVAL;
1126 u64 devid;
1127 u64 transid;
1128 u64 total_devices;
1129 u64 bytenr;
1130
1131
1132
1133
1134
1135
1136
1137 bytenr = btrfs_sb_offset(0);
1138 flags |= FMODE_EXCL;
1139 mutex_lock(&uuid_mutex);
1140
1141 bdev = blkdev_get_by_path(path, flags, holder);
1142 if (IS_ERR(bdev)) {
1143 ret = PTR_ERR(bdev);
1144 goto error;
1145 }
1146
1147 if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super))
1148 goto error_bdev_put;
1149
1150 devid = btrfs_stack_device_id(&disk_super->dev_item);
1151 transid = btrfs_super_generation(disk_super);
1152 total_devices = btrfs_super_num_devices(disk_super);
1153
1154 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1155 if (ret > 0) {
1156 if (disk_super->label[0]) {
1157 pr_info("BTRFS: device label %s ", disk_super->label);
1158 } else {
1159 pr_info("BTRFS: device fsid %pU ", disk_super->fsid);
1160 }
1161
1162 pr_cont("devid %llu transid %llu %s\n", devid, transid, path);
1163 ret = 0;
1164 }
1165 if (!ret && fs_devices_ret)
1166 (*fs_devices_ret)->total_devices = total_devices;
1167
1168 btrfs_release_disk_super(page);
1169
1170error_bdev_put:
1171 blkdev_put(bdev, flags);
1172error:
1173 mutex_unlock(&uuid_mutex);
1174 return ret;
1175}
1176
1177
1178int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1179 u64 end, u64 *length)
1180{
1181 struct btrfs_key key;
1182 struct btrfs_root *root = device->dev_root;
1183 struct btrfs_dev_extent *dev_extent;
1184 struct btrfs_path *path;
1185 u64 extent_end;
1186 int ret;
1187 int slot;
1188 struct extent_buffer *l;
1189
1190 *length = 0;
1191
1192 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1193 return 0;
1194
1195 path = btrfs_alloc_path();
1196 if (!path)
1197 return -ENOMEM;
1198 path->reada = READA_FORWARD;
1199
1200 key.objectid = device->devid;
1201 key.offset = start;
1202 key.type = BTRFS_DEV_EXTENT_KEY;
1203
1204 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1205 if (ret < 0)
1206 goto out;
1207 if (ret > 0) {
1208 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1209 if (ret < 0)
1210 goto out;
1211 }
1212
1213 while (1) {
1214 l = path->nodes[0];
1215 slot = path->slots[0];
1216 if (slot >= btrfs_header_nritems(l)) {
1217 ret = btrfs_next_leaf(root, path);
1218 if (ret == 0)
1219 continue;
1220 if (ret < 0)
1221 goto out;
1222
1223 break;
1224 }
1225 btrfs_item_key_to_cpu(l, &key, slot);
1226
1227 if (key.objectid < device->devid)
1228 goto next;
1229
1230 if (key.objectid > device->devid)
1231 break;
1232
1233 if (key.type != BTRFS_DEV_EXTENT_KEY)
1234 goto next;
1235
1236 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1237 extent_end = key.offset + btrfs_dev_extent_length(l,
1238 dev_extent);
1239 if (key.offset <= start && extent_end > end) {
1240 *length = end - start + 1;
1241 break;
1242 } else if (key.offset <= start && extent_end > start)
1243 *length += extent_end - start;
1244 else if (key.offset > start && extent_end <= end)
1245 *length += extent_end - key.offset;
1246 else if (key.offset > start && key.offset <= end) {
1247 *length += end - key.offset + 1;
1248 break;
1249 } else if (key.offset > end)
1250 break;
1251
1252next:
1253 path->slots[0]++;
1254 }
1255 ret = 0;
1256out:
1257 btrfs_free_path(path);
1258 return ret;
1259}
1260
1261static int contains_pending_extent(struct btrfs_transaction *transaction,
1262 struct btrfs_device *device,
1263 u64 *start, u64 len)
1264{
1265 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
1266 struct extent_map *em;
1267 struct list_head *search_list = &fs_info->pinned_chunks;
1268 int ret = 0;
1269 u64 physical_start = *start;
1270
1271 if (transaction)
1272 search_list = &transaction->pending_chunks;
1273again:
1274 list_for_each_entry(em, search_list, list) {
1275 struct map_lookup *map;
1276 int i;
1277
1278 map = em->map_lookup;
1279 for (i = 0; i < map->num_stripes; i++) {
1280 u64 end;
1281
1282 if (map->stripes[i].dev != device)
1283 continue;
1284 if (map->stripes[i].physical >= physical_start + len ||
1285 map->stripes[i].physical + em->orig_block_len <=
1286 physical_start)
1287 continue;
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300 end = map->stripes[i].physical + em->orig_block_len;
1301 if (end > *start) {
1302 *start = end;
1303 ret = 1;
1304 }
1305 }
1306 }
1307 if (search_list != &fs_info->pinned_chunks) {
1308 search_list = &fs_info->pinned_chunks;
1309 goto again;
1310 }
1311
1312 return ret;
1313}
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337int find_free_dev_extent_start(struct btrfs_transaction *transaction,
1338 struct btrfs_device *device, u64 num_bytes,
1339 u64 search_start, u64 *start, u64 *len)
1340{
1341 struct btrfs_key key;
1342 struct btrfs_root *root = device->dev_root;
1343 struct btrfs_dev_extent *dev_extent;
1344 struct btrfs_path *path;
1345 u64 hole_size;
1346 u64 max_hole_start;
1347 u64 max_hole_size;
1348 u64 extent_end;
1349 u64 search_end = device->total_bytes;
1350 int ret;
1351 int slot;
1352 struct extent_buffer *l;
1353 u64 min_search_start;
1354
1355
1356
1357
1358
1359
1360 min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1361 search_start = max(search_start, min_search_start);
1362
1363 path = btrfs_alloc_path();
1364 if (!path)
1365 return -ENOMEM;
1366
1367 max_hole_start = search_start;
1368 max_hole_size = 0;
1369
1370again:
1371 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1372 ret = -ENOSPC;
1373 goto out;
1374 }
1375
1376 path->reada = READA_FORWARD;
1377 path->search_commit_root = 1;
1378 path->skip_locking = 1;
1379
1380 key.objectid = device->devid;
1381 key.offset = search_start;
1382 key.type = BTRFS_DEV_EXTENT_KEY;
1383
1384 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1385 if (ret < 0)
1386 goto out;
1387 if (ret > 0) {
1388 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1389 if (ret < 0)
1390 goto out;
1391 }
1392
1393 while (1) {
1394 l = path->nodes[0];
1395 slot = path->slots[0];
1396 if (slot >= btrfs_header_nritems(l)) {
1397 ret = btrfs_next_leaf(root, path);
1398 if (ret == 0)
1399 continue;
1400 if (ret < 0)
1401 goto out;
1402
1403 break;
1404 }
1405 btrfs_item_key_to_cpu(l, &key, slot);
1406
1407 if (key.objectid < device->devid)
1408 goto next;
1409
1410 if (key.objectid > device->devid)
1411 break;
1412
1413 if (key.type != BTRFS_DEV_EXTENT_KEY)
1414 goto next;
1415
1416 if (key.offset > search_start) {
1417 hole_size = key.offset - search_start;
1418
1419
1420
1421
1422
1423 if (contains_pending_extent(transaction, device,
1424 &search_start,
1425 hole_size)) {
1426 if (key.offset >= search_start) {
1427 hole_size = key.offset - search_start;
1428 } else {
1429 WARN_ON_ONCE(1);
1430 hole_size = 0;
1431 }
1432 }
1433
1434 if (hole_size > max_hole_size) {
1435 max_hole_start = search_start;
1436 max_hole_size = hole_size;
1437 }
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448 if (hole_size >= num_bytes) {
1449 ret = 0;
1450 goto out;
1451 }
1452 }
1453
1454 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1455 extent_end = key.offset + btrfs_dev_extent_length(l,
1456 dev_extent);
1457 if (extent_end > search_start)
1458 search_start = extent_end;
1459next:
1460 path->slots[0]++;
1461 cond_resched();
1462 }
1463
1464
1465
1466
1467
1468
1469 if (search_end > search_start) {
1470 hole_size = search_end - search_start;
1471
1472 if (contains_pending_extent(transaction, device, &search_start,
1473 hole_size)) {
1474 btrfs_release_path(path);
1475 goto again;
1476 }
1477
1478 if (hole_size > max_hole_size) {
1479 max_hole_start = search_start;
1480 max_hole_size = hole_size;
1481 }
1482 }
1483
1484
1485 if (max_hole_size < num_bytes)
1486 ret = -ENOSPC;
1487 else
1488 ret = 0;
1489
1490out:
1491 btrfs_free_path(path);
1492 *start = max_hole_start;
1493 if (len)
1494 *len = max_hole_size;
1495 return ret;
1496}
1497
1498int find_free_dev_extent(struct btrfs_trans_handle *trans,
1499 struct btrfs_device *device, u64 num_bytes,
1500 u64 *start, u64 *len)
1501{
1502
1503 return find_free_dev_extent_start(trans->transaction, device,
1504 num_bytes, 0, start, len);
1505}
1506
1507static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1508 struct btrfs_device *device,
1509 u64 start, u64 *dev_extent_len)
1510{
1511 int ret;
1512 struct btrfs_path *path;
1513 struct btrfs_root *root = device->dev_root;
1514 struct btrfs_key key;
1515 struct btrfs_key found_key;
1516 struct extent_buffer *leaf = NULL;
1517 struct btrfs_dev_extent *extent = NULL;
1518
1519 path = btrfs_alloc_path();
1520 if (!path)
1521 return -ENOMEM;
1522
1523 key.objectid = device->devid;
1524 key.offset = start;
1525 key.type = BTRFS_DEV_EXTENT_KEY;
1526again:
1527 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1528 if (ret > 0) {
1529 ret = btrfs_previous_item(root, path, key.objectid,
1530 BTRFS_DEV_EXTENT_KEY);
1531 if (ret)
1532 goto out;
1533 leaf = path->nodes[0];
1534 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1535 extent = btrfs_item_ptr(leaf, path->slots[0],
1536 struct btrfs_dev_extent);
1537 BUG_ON(found_key.offset > start || found_key.offset +
1538 btrfs_dev_extent_length(leaf, extent) < start);
1539 key = found_key;
1540 btrfs_release_path(path);
1541 goto again;
1542 } else if (ret == 0) {
1543 leaf = path->nodes[0];
1544 extent = btrfs_item_ptr(leaf, path->slots[0],
1545 struct btrfs_dev_extent);
1546 } else {
1547 btrfs_handle_fs_error(root->fs_info, ret, "Slot search failed");
1548 goto out;
1549 }
1550
1551 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1552
1553 ret = btrfs_del_item(trans, root, path);
1554 if (ret) {
1555 btrfs_handle_fs_error(root->fs_info, ret,
1556 "Failed to remove dev extent item");
1557 } else {
1558 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1559 }
1560out:
1561 btrfs_free_path(path);
1562 return ret;
1563}
1564
1565static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1566 struct btrfs_device *device,
1567 u64 chunk_tree, u64 chunk_objectid,
1568 u64 chunk_offset, u64 start, u64 num_bytes)
1569{
1570 int ret;
1571 struct btrfs_path *path;
1572 struct btrfs_root *root = device->dev_root;
1573 struct btrfs_dev_extent *extent;
1574 struct extent_buffer *leaf;
1575 struct btrfs_key key;
1576
1577 WARN_ON(!device->in_fs_metadata);
1578 WARN_ON(device->is_tgtdev_for_dev_replace);
1579 path = btrfs_alloc_path();
1580 if (!path)
1581 return -ENOMEM;
1582
1583 key.objectid = device->devid;
1584 key.offset = start;
1585 key.type = BTRFS_DEV_EXTENT_KEY;
1586 ret = btrfs_insert_empty_item(trans, root, path, &key,
1587 sizeof(*extent));
1588 if (ret)
1589 goto out;
1590
1591 leaf = path->nodes[0];
1592 extent = btrfs_item_ptr(leaf, path->slots[0],
1593 struct btrfs_dev_extent);
1594 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1595 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1596 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1597
1598 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1599 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1600
1601 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1602 btrfs_mark_buffer_dirty(leaf);
1603out:
1604 btrfs_free_path(path);
1605 return ret;
1606}
1607
1608static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1609{
1610 struct extent_map_tree *em_tree;
1611 struct extent_map *em;
1612 struct rb_node *n;
1613 u64 ret = 0;
1614
1615 em_tree = &fs_info->mapping_tree.map_tree;
1616 read_lock(&em_tree->lock);
1617 n = rb_last(&em_tree->map);
1618 if (n) {
1619 em = rb_entry(n, struct extent_map, rb_node);
1620 ret = em->start + em->len;
1621 }
1622 read_unlock(&em_tree->lock);
1623
1624 return ret;
1625}
1626
1627static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1628 u64 *devid_ret)
1629{
1630 int ret;
1631 struct btrfs_key key;
1632 struct btrfs_key found_key;
1633 struct btrfs_path *path;
1634
1635 path = btrfs_alloc_path();
1636 if (!path)
1637 return -ENOMEM;
1638
1639 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1640 key.type = BTRFS_DEV_ITEM_KEY;
1641 key.offset = (u64)-1;
1642
1643 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1644 if (ret < 0)
1645 goto error;
1646
1647 BUG_ON(ret == 0);
1648
1649 ret = btrfs_previous_item(fs_info->chunk_root, path,
1650 BTRFS_DEV_ITEMS_OBJECTID,
1651 BTRFS_DEV_ITEM_KEY);
1652 if (ret) {
1653 *devid_ret = 1;
1654 } else {
1655 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1656 path->slots[0]);
1657 *devid_ret = found_key.offset + 1;
1658 }
1659 ret = 0;
1660error:
1661 btrfs_free_path(path);
1662 return ret;
1663}
1664
1665
1666
1667
1668
1669static int btrfs_add_device(struct btrfs_trans_handle *trans,
1670 struct btrfs_root *root,
1671 struct btrfs_device *device)
1672{
1673 int ret;
1674 struct btrfs_path *path;
1675 struct btrfs_dev_item *dev_item;
1676 struct extent_buffer *leaf;
1677 struct btrfs_key key;
1678 unsigned long ptr;
1679
1680 root = root->fs_info->chunk_root;
1681
1682 path = btrfs_alloc_path();
1683 if (!path)
1684 return -ENOMEM;
1685
1686 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1687 key.type = BTRFS_DEV_ITEM_KEY;
1688 key.offset = device->devid;
1689
1690 ret = btrfs_insert_empty_item(trans, root, path, &key,
1691 sizeof(*dev_item));
1692 if (ret)
1693 goto out;
1694
1695 leaf = path->nodes[0];
1696 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1697
1698 btrfs_set_device_id(leaf, dev_item, device->devid);
1699 btrfs_set_device_generation(leaf, dev_item, 0);
1700 btrfs_set_device_type(leaf, dev_item, device->type);
1701 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1702 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1703 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1704 btrfs_set_device_total_bytes(leaf, dev_item,
1705 btrfs_device_get_disk_total_bytes(device));
1706 btrfs_set_device_bytes_used(leaf, dev_item,
1707 btrfs_device_get_bytes_used(device));
1708 btrfs_set_device_group(leaf, dev_item, 0);
1709 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1710 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1711 btrfs_set_device_start_offset(leaf, dev_item, 0);
1712
1713 ptr = btrfs_device_uuid(dev_item);
1714 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1715 ptr = btrfs_device_fsid(dev_item);
1716 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1717 btrfs_mark_buffer_dirty(leaf);
1718
1719 ret = 0;
1720out:
1721 btrfs_free_path(path);
1722 return ret;
1723}
1724
1725
1726
1727
1728
1729static void update_dev_time(char *path_name)
1730{
1731 struct file *filp;
1732
1733 filp = filp_open(path_name, O_RDWR, 0);
1734 if (IS_ERR(filp))
1735 return;
1736 file_update_time(filp);
1737 filp_close(filp, NULL);
1738}
1739
1740static int btrfs_rm_dev_item(struct btrfs_root *root,
1741 struct btrfs_device *device)
1742{
1743 int ret;
1744 struct btrfs_path *path;
1745 struct btrfs_key key;
1746 struct btrfs_trans_handle *trans;
1747
1748 root = root->fs_info->chunk_root;
1749
1750 path = btrfs_alloc_path();
1751 if (!path)
1752 return -ENOMEM;
1753
1754 trans = btrfs_start_transaction(root, 0);
1755 if (IS_ERR(trans)) {
1756 btrfs_free_path(path);
1757 return PTR_ERR(trans);
1758 }
1759 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1760 key.type = BTRFS_DEV_ITEM_KEY;
1761 key.offset = device->devid;
1762
1763 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1764 if (ret < 0)
1765 goto out;
1766
1767 if (ret > 0) {
1768 ret = -ENOENT;
1769 goto out;
1770 }
1771
1772 ret = btrfs_del_item(trans, root, path);
1773 if (ret)
1774 goto out;
1775out:
1776 btrfs_free_path(path);
1777 btrfs_commit_transaction(trans, root);
1778 return ret;
1779}
1780
1781
1782
1783
1784
1785
1786static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1787 u64 num_devices)
1788{
1789 u64 all_avail;
1790 unsigned seq;
1791 int i;
1792
1793 do {
1794 seq = read_seqbegin(&fs_info->profiles_lock);
1795
1796 all_avail = fs_info->avail_data_alloc_bits |
1797 fs_info->avail_system_alloc_bits |
1798 fs_info->avail_metadata_alloc_bits;
1799 } while (read_seqretry(&fs_info->profiles_lock, seq));
1800
1801 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1802 if (!(all_avail & btrfs_raid_group[i]))
1803 continue;
1804
1805 if (num_devices < btrfs_raid_array[i].devs_min) {
1806 int ret = btrfs_raid_mindev_error[i];
1807
1808 if (ret)
1809 return ret;
1810 }
1811 }
1812
1813 return 0;
1814}
1815
1816struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs,
1817 struct btrfs_device *device)
1818{
1819 struct btrfs_device *next_device;
1820
1821 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1822 if (next_device != device &&
1823 !next_device->missing && next_device->bdev)
1824 return next_device;
1825 }
1826
1827 return NULL;
1828}
1829
1830
1831
1832
1833
1834
1835
1836void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info,
1837 struct btrfs_device *device, struct btrfs_device *this_dev)
1838{
1839 struct btrfs_device *next_device;
1840
1841 if (this_dev)
1842 next_device = this_dev;
1843 else
1844 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1845 device);
1846 ASSERT(next_device);
1847
1848 if (fs_info->sb->s_bdev &&
1849 (fs_info->sb->s_bdev == device->bdev))
1850 fs_info->sb->s_bdev = next_device->bdev;
1851
1852 if (fs_info->fs_devices->latest_bdev == device->bdev)
1853 fs_info->fs_devices->latest_bdev = next_device->bdev;
1854}
1855
1856int btrfs_rm_device(struct btrfs_root *root, char *device_path, u64 devid)
1857{
1858 struct btrfs_device *device;
1859 struct btrfs_fs_devices *cur_devices;
1860 u64 num_devices;
1861 int ret = 0;
1862 bool clear_super = false;
1863
1864 mutex_lock(&uuid_mutex);
1865
1866 num_devices = root->fs_info->fs_devices->num_devices;
1867 btrfs_dev_replace_lock(&root->fs_info->dev_replace, 0);
1868 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1869 WARN_ON(num_devices < 1);
1870 num_devices--;
1871 }
1872 btrfs_dev_replace_unlock(&root->fs_info->dev_replace, 0);
1873
1874 ret = btrfs_check_raid_min_devices(root->fs_info, num_devices - 1);
1875 if (ret)
1876 goto out;
1877
1878 ret = btrfs_find_device_by_devspec(root, devid, device_path,
1879 &device);
1880 if (ret)
1881 goto out;
1882
1883 if (device->is_tgtdev_for_dev_replace) {
1884 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1885 goto out;
1886 }
1887
1888 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1889 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1890 goto out;
1891 }
1892
1893 if (device->writeable) {
1894 lock_chunks(root);
1895 list_del_init(&device->dev_alloc_list);
1896 device->fs_devices->rw_devices--;
1897 unlock_chunks(root);
1898 clear_super = true;
1899 }
1900
1901 mutex_unlock(&uuid_mutex);
1902 ret = btrfs_shrink_device(device, 0);
1903 mutex_lock(&uuid_mutex);
1904 if (ret)
1905 goto error_undo;
1906
1907
1908
1909
1910
1911
1912 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1913 if (ret)
1914 goto error_undo;
1915
1916 device->in_fs_metadata = 0;
1917 btrfs_scrub_cancel_dev(root->fs_info, device);
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929 cur_devices = device->fs_devices;
1930 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1931 list_del_rcu(&device->dev_list);
1932
1933 device->fs_devices->num_devices--;
1934 device->fs_devices->total_devices--;
1935
1936 if (device->missing)
1937 device->fs_devices->missing_devices--;
1938
1939 btrfs_assign_next_active_device(root->fs_info, device, NULL);
1940
1941 if (device->bdev) {
1942 device->fs_devices->open_devices--;
1943
1944 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
1945 }
1946
1947 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1948 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1949 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1950
1951
1952
1953
1954
1955
1956 if (device->writeable)
1957 btrfs_scratch_superblocks(device->bdev, device->name->str);
1958
1959 btrfs_close_bdev(device);
1960 call_rcu(&device->rcu, free_device);
1961
1962 if (cur_devices->open_devices == 0) {
1963 struct btrfs_fs_devices *fs_devices;
1964 fs_devices = root->fs_info->fs_devices;
1965 while (fs_devices) {
1966 if (fs_devices->seed == cur_devices) {
1967 fs_devices->seed = cur_devices->seed;
1968 break;
1969 }
1970 fs_devices = fs_devices->seed;
1971 }
1972 cur_devices->seed = NULL;
1973 __btrfs_close_devices(cur_devices);
1974 free_fs_devices(cur_devices);
1975 }
1976
1977 root->fs_info->num_tolerated_disk_barrier_failures =
1978 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1979
1980out:
1981 mutex_unlock(&uuid_mutex);
1982 return ret;
1983
1984error_undo:
1985 if (device->writeable) {
1986 lock_chunks(root);
1987 list_add(&device->dev_alloc_list,
1988 &root->fs_info->fs_devices->alloc_list);
1989 device->fs_devices->rw_devices++;
1990 unlock_chunks(root);
1991 }
1992 goto out;
1993}
1994
1995void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1996 struct btrfs_device *srcdev)
1997{
1998 struct btrfs_fs_devices *fs_devices;
1999
2000 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
2001
2002
2003
2004
2005
2006
2007
2008 fs_devices = srcdev->fs_devices;
2009
2010 list_del_rcu(&srcdev->dev_list);
2011 list_del_rcu(&srcdev->dev_alloc_list);
2012 fs_devices->num_devices--;
2013 if (srcdev->missing)
2014 fs_devices->missing_devices--;
2015
2016 if (srcdev->writeable)
2017 fs_devices->rw_devices--;
2018
2019 if (srcdev->bdev)
2020 fs_devices->open_devices--;
2021}
2022
2023void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
2024 struct btrfs_device *srcdev)
2025{
2026 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2027
2028 if (srcdev->writeable) {
2029
2030 btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str);
2031 }
2032
2033 btrfs_close_bdev(srcdev);
2034
2035 call_rcu(&srcdev->rcu, free_device);
2036
2037
2038
2039
2040
2041 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
2042
2043
2044 if (!fs_devices->num_devices) {
2045 struct btrfs_fs_devices *tmp_fs_devices;
2046
2047 tmp_fs_devices = fs_info->fs_devices;
2048 while (tmp_fs_devices) {
2049 if (tmp_fs_devices->seed == fs_devices) {
2050 tmp_fs_devices->seed = fs_devices->seed;
2051 break;
2052 }
2053 tmp_fs_devices = tmp_fs_devices->seed;
2054 }
2055 fs_devices->seed = NULL;
2056 __btrfs_close_devices(fs_devices);
2057 free_fs_devices(fs_devices);
2058 }
2059}
2060
2061void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
2062 struct btrfs_device *tgtdev)
2063{
2064 mutex_lock(&uuid_mutex);
2065 WARN_ON(!tgtdev);
2066 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2067
2068 btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev);
2069
2070 if (tgtdev->bdev)
2071 fs_info->fs_devices->open_devices--;
2072
2073 fs_info->fs_devices->num_devices--;
2074
2075 btrfs_assign_next_active_device(fs_info, tgtdev, NULL);
2076
2077 list_del_rcu(&tgtdev->dev_list);
2078
2079 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2080 mutex_unlock(&uuid_mutex);
2081
2082
2083
2084
2085
2086
2087
2088
2089 btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str);
2090
2091 btrfs_close_bdev(tgtdev);
2092 call_rcu(&tgtdev->rcu, free_device);
2093}
2094
2095static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
2096 struct btrfs_device **device)
2097{
2098 int ret = 0;
2099 struct btrfs_super_block *disk_super;
2100 u64 devid;
2101 u8 *dev_uuid;
2102 struct block_device *bdev;
2103 struct buffer_head *bh;
2104
2105 *device = NULL;
2106 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2107 root->fs_info->bdev_holder, 0, &bdev, &bh);
2108 if (ret)
2109 return ret;
2110 disk_super = (struct btrfs_super_block *)bh->b_data;
2111 devid = btrfs_stack_device_id(&disk_super->dev_item);
2112 dev_uuid = disk_super->dev_item.uuid;
2113 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2114 disk_super->fsid);
2115 brelse(bh);
2116 if (!*device)
2117 ret = -ENOENT;
2118 blkdev_put(bdev, FMODE_READ);
2119 return ret;
2120}
2121
2122int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2123 char *device_path,
2124 struct btrfs_device **device)
2125{
2126 *device = NULL;
2127 if (strcmp(device_path, "missing") == 0) {
2128 struct list_head *devices;
2129 struct btrfs_device *tmp;
2130
2131 devices = &root->fs_info->fs_devices->devices;
2132
2133
2134
2135
2136 list_for_each_entry(tmp, devices, dev_list) {
2137 if (tmp->in_fs_metadata && !tmp->bdev) {
2138 *device = tmp;
2139 break;
2140 }
2141 }
2142
2143 if (!*device)
2144 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2145
2146 return 0;
2147 } else {
2148 return btrfs_find_device_by_path(root, device_path, device);
2149 }
2150}
2151
2152
2153
2154
2155int btrfs_find_device_by_devspec(struct btrfs_root *root, u64 devid,
2156 char *devpath,
2157 struct btrfs_device **device)
2158{
2159 int ret;
2160
2161 if (devid) {
2162 ret = 0;
2163 *device = btrfs_find_device(root->fs_info, devid, NULL,
2164 NULL);
2165 if (!*device)
2166 ret = -ENOENT;
2167 } else {
2168 if (!devpath || !devpath[0])
2169 return -EINVAL;
2170
2171 ret = btrfs_find_device_missing_or_by_path(root, devpath,
2172 device);
2173 }
2174 return ret;
2175}
2176
2177
2178
2179
2180static int btrfs_prepare_sprout(struct btrfs_root *root)
2181{
2182 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2183 struct btrfs_fs_devices *old_devices;
2184 struct btrfs_fs_devices *seed_devices;
2185 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2186 struct btrfs_device *device;
2187 u64 super_flags;
2188
2189 BUG_ON(!mutex_is_locked(&uuid_mutex));
2190 if (!fs_devices->seeding)
2191 return -EINVAL;
2192
2193 seed_devices = __alloc_fs_devices();
2194 if (IS_ERR(seed_devices))
2195 return PTR_ERR(seed_devices);
2196
2197 old_devices = clone_fs_devices(fs_devices);
2198 if (IS_ERR(old_devices)) {
2199 kfree(seed_devices);
2200 return PTR_ERR(old_devices);
2201 }
2202
2203 list_add(&old_devices->list, &fs_uuids);
2204
2205 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2206 seed_devices->opened = 1;
2207 INIT_LIST_HEAD(&seed_devices->devices);
2208 INIT_LIST_HEAD(&seed_devices->alloc_list);
2209 mutex_init(&seed_devices->device_list_mutex);
2210
2211 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2212 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2213 synchronize_rcu);
2214 list_for_each_entry(device, &seed_devices->devices, dev_list)
2215 device->fs_devices = seed_devices;
2216
2217 lock_chunks(root);
2218 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2219 unlock_chunks(root);
2220
2221 fs_devices->seeding = 0;
2222 fs_devices->num_devices = 0;
2223 fs_devices->open_devices = 0;
2224 fs_devices->missing_devices = 0;
2225 fs_devices->rotating = 0;
2226 fs_devices->seed = seed_devices;
2227
2228 generate_random_uuid(fs_devices->fsid);
2229 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2230 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2231 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2232
2233 super_flags = btrfs_super_flags(disk_super) &
2234 ~BTRFS_SUPER_FLAG_SEEDING;
2235 btrfs_set_super_flags(disk_super, super_flags);
2236
2237 return 0;
2238}
2239
2240
2241
2242
2243static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2244 struct btrfs_root *root)
2245{
2246 struct btrfs_path *path;
2247 struct extent_buffer *leaf;
2248 struct btrfs_dev_item *dev_item;
2249 struct btrfs_device *device;
2250 struct btrfs_key key;
2251 u8 fs_uuid[BTRFS_UUID_SIZE];
2252 u8 dev_uuid[BTRFS_UUID_SIZE];
2253 u64 devid;
2254 int ret;
2255
2256 path = btrfs_alloc_path();
2257 if (!path)
2258 return -ENOMEM;
2259
2260 root = root->fs_info->chunk_root;
2261 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2262 key.offset = 0;
2263 key.type = BTRFS_DEV_ITEM_KEY;
2264
2265 while (1) {
2266 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2267 if (ret < 0)
2268 goto error;
2269
2270 leaf = path->nodes[0];
2271next_slot:
2272 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2273 ret = btrfs_next_leaf(root, path);
2274 if (ret > 0)
2275 break;
2276 if (ret < 0)
2277 goto error;
2278 leaf = path->nodes[0];
2279 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2280 btrfs_release_path(path);
2281 continue;
2282 }
2283
2284 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2285 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2286 key.type != BTRFS_DEV_ITEM_KEY)
2287 break;
2288
2289 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2290 struct btrfs_dev_item);
2291 devid = btrfs_device_id(leaf, dev_item);
2292 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2293 BTRFS_UUID_SIZE);
2294 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2295 BTRFS_UUID_SIZE);
2296 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2297 fs_uuid);
2298 BUG_ON(!device);
2299
2300 if (device->fs_devices->seeding) {
2301 btrfs_set_device_generation(leaf, dev_item,
2302 device->generation);
2303 btrfs_mark_buffer_dirty(leaf);
2304 }
2305
2306 path->slots[0]++;
2307 goto next_slot;
2308 }
2309 ret = 0;
2310error:
2311 btrfs_free_path(path);
2312 return ret;
2313}
2314
2315int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2316{
2317 struct request_queue *q;
2318 struct btrfs_trans_handle *trans;
2319 struct btrfs_device *device;
2320 struct block_device *bdev;
2321 struct list_head *devices;
2322 struct super_block *sb = root->fs_info->sb;
2323 struct rcu_string *name;
2324 u64 tmp;
2325 int seeding_dev = 0;
2326 int ret = 0;
2327
2328 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2329 return -EROFS;
2330
2331 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2332 root->fs_info->bdev_holder);
2333 if (IS_ERR(bdev))
2334 return PTR_ERR(bdev);
2335
2336 if (root->fs_info->fs_devices->seeding) {
2337 seeding_dev = 1;
2338 down_write(&sb->s_umount);
2339 mutex_lock(&uuid_mutex);
2340 }
2341
2342 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2343
2344 devices = &root->fs_info->fs_devices->devices;
2345
2346 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2347 list_for_each_entry(device, devices, dev_list) {
2348 if (device->bdev == bdev) {
2349 ret = -EEXIST;
2350 mutex_unlock(
2351 &root->fs_info->fs_devices->device_list_mutex);
2352 goto error;
2353 }
2354 }
2355 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2356
2357 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2358 if (IS_ERR(device)) {
2359
2360 ret = PTR_ERR(device);
2361 goto error;
2362 }
2363
2364 name = rcu_string_strdup(device_path, GFP_KERNEL);
2365 if (!name) {
2366 kfree(device);
2367 ret = -ENOMEM;
2368 goto error;
2369 }
2370 rcu_assign_pointer(device->name, name);
2371
2372 trans = btrfs_start_transaction(root, 0);
2373 if (IS_ERR(trans)) {
2374 rcu_string_free(device->name);
2375 kfree(device);
2376 ret = PTR_ERR(trans);
2377 goto error;
2378 }
2379
2380 q = bdev_get_queue(bdev);
2381 if (blk_queue_discard(q))
2382 device->can_discard = 1;
2383 device->writeable = 1;
2384 device->generation = trans->transid;
2385 device->io_width = root->sectorsize;
2386 device->io_align = root->sectorsize;
2387 device->sector_size = root->sectorsize;
2388 device->total_bytes = i_size_read(bdev->bd_inode);
2389 device->disk_total_bytes = device->total_bytes;
2390 device->commit_total_bytes = device->total_bytes;
2391 device->dev_root = root->fs_info->dev_root;
2392 device->bdev = bdev;
2393 device->in_fs_metadata = 1;
2394 device->is_tgtdev_for_dev_replace = 0;
2395 device->mode = FMODE_EXCL;
2396 device->dev_stats_valid = 1;
2397 set_blocksize(device->bdev, 4096);
2398
2399 if (seeding_dev) {
2400 sb->s_flags &= ~MS_RDONLY;
2401 ret = btrfs_prepare_sprout(root);
2402 BUG_ON(ret);
2403 }
2404
2405 device->fs_devices = root->fs_info->fs_devices;
2406
2407 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2408 lock_chunks(root);
2409 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2410 list_add(&device->dev_alloc_list,
2411 &root->fs_info->fs_devices->alloc_list);
2412 root->fs_info->fs_devices->num_devices++;
2413 root->fs_info->fs_devices->open_devices++;
2414 root->fs_info->fs_devices->rw_devices++;
2415 root->fs_info->fs_devices->total_devices++;
2416 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2417
2418 spin_lock(&root->fs_info->free_chunk_lock);
2419 root->fs_info->free_chunk_space += device->total_bytes;
2420 spin_unlock(&root->fs_info->free_chunk_lock);
2421
2422 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2423 root->fs_info->fs_devices->rotating = 1;
2424
2425 tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2426 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2427 tmp + device->total_bytes);
2428
2429 tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2430 btrfs_set_super_num_devices(root->fs_info->super_copy,
2431 tmp + 1);
2432
2433
2434 btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device);
2435
2436
2437
2438
2439
2440 btrfs_clear_space_info_full(root->fs_info);
2441
2442 unlock_chunks(root);
2443 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2444
2445 if (seeding_dev) {
2446 lock_chunks(root);
2447 ret = init_first_rw_device(trans, root, device);
2448 unlock_chunks(root);
2449 if (ret) {
2450 btrfs_abort_transaction(trans, ret);
2451 goto error_trans;
2452 }
2453 }
2454
2455 ret = btrfs_add_device(trans, root, device);
2456 if (ret) {
2457 btrfs_abort_transaction(trans, ret);
2458 goto error_trans;
2459 }
2460
2461 if (seeding_dev) {
2462 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2463
2464 ret = btrfs_finish_sprout(trans, root);
2465 if (ret) {
2466 btrfs_abort_transaction(trans, ret);
2467 goto error_trans;
2468 }
2469
2470
2471
2472
2473 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2474 root->fs_info->fsid);
2475 if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj,
2476 fsid_buf))
2477 btrfs_warn(root->fs_info,
2478 "sysfs: failed to create fsid for sprout");
2479 }
2480
2481 root->fs_info->num_tolerated_disk_barrier_failures =
2482 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2483 ret = btrfs_commit_transaction(trans, root);
2484
2485 if (seeding_dev) {
2486 mutex_unlock(&uuid_mutex);
2487 up_write(&sb->s_umount);
2488
2489 if (ret)
2490 return ret;
2491
2492 ret = btrfs_relocate_sys_chunks(root);
2493 if (ret < 0)
2494 btrfs_handle_fs_error(root->fs_info, ret,
2495 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2496 trans = btrfs_attach_transaction(root);
2497 if (IS_ERR(trans)) {
2498 if (PTR_ERR(trans) == -ENOENT)
2499 return 0;
2500 return PTR_ERR(trans);
2501 }
2502 ret = btrfs_commit_transaction(trans, root);
2503 }
2504
2505
2506 update_dev_time(device_path);
2507 return ret;
2508
2509error_trans:
2510 btrfs_end_transaction(trans, root);
2511 rcu_string_free(device->name);
2512 btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device);
2513 kfree(device);
2514error:
2515 blkdev_put(bdev, FMODE_EXCL);
2516 if (seeding_dev) {
2517 mutex_unlock(&uuid_mutex);
2518 up_write(&sb->s_umount);
2519 }
2520 return ret;
2521}
2522
2523int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2524 struct btrfs_device *srcdev,
2525 struct btrfs_device **device_out)
2526{
2527 struct request_queue *q;
2528 struct btrfs_device *device;
2529 struct block_device *bdev;
2530 struct btrfs_fs_info *fs_info = root->fs_info;
2531 struct list_head *devices;
2532 struct rcu_string *name;
2533 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2534 int ret = 0;
2535
2536 *device_out = NULL;
2537 if (fs_info->fs_devices->seeding) {
2538 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2539 return -EINVAL;
2540 }
2541
2542 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2543 fs_info->bdev_holder);
2544 if (IS_ERR(bdev)) {
2545 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2546 return PTR_ERR(bdev);
2547 }
2548
2549 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2550
2551 devices = &fs_info->fs_devices->devices;
2552 list_for_each_entry(device, devices, dev_list) {
2553 if (device->bdev == bdev) {
2554 btrfs_err(fs_info,
2555 "target device is in the filesystem!");
2556 ret = -EEXIST;
2557 goto error;
2558 }
2559 }
2560
2561
2562 if (i_size_read(bdev->bd_inode) <
2563 btrfs_device_get_total_bytes(srcdev)) {
2564 btrfs_err(fs_info,
2565 "target device is smaller than source device!");
2566 ret = -EINVAL;
2567 goto error;
2568 }
2569
2570
2571 device = btrfs_alloc_device(NULL, &devid, NULL);
2572 if (IS_ERR(device)) {
2573 ret = PTR_ERR(device);
2574 goto error;
2575 }
2576
2577 name = rcu_string_strdup(device_path, GFP_NOFS);
2578 if (!name) {
2579 kfree(device);
2580 ret = -ENOMEM;
2581 goto error;
2582 }
2583 rcu_assign_pointer(device->name, name);
2584
2585 q = bdev_get_queue(bdev);
2586 if (blk_queue_discard(q))
2587 device->can_discard = 1;
2588 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2589 device->writeable = 1;
2590 device->generation = 0;
2591 device->io_width = root->sectorsize;
2592 device->io_align = root->sectorsize;
2593 device->sector_size = root->sectorsize;
2594 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2595 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2596 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2597 ASSERT(list_empty(&srcdev->resized_list));
2598 device->commit_total_bytes = srcdev->commit_total_bytes;
2599 device->commit_bytes_used = device->bytes_used;
2600 device->dev_root = fs_info->dev_root;
2601 device->bdev = bdev;
2602 device->in_fs_metadata = 1;
2603 device->is_tgtdev_for_dev_replace = 1;
2604 device->mode = FMODE_EXCL;
2605 device->dev_stats_valid = 1;
2606 set_blocksize(device->bdev, 4096);
2607 device->fs_devices = fs_info->fs_devices;
2608 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2609 fs_info->fs_devices->num_devices++;
2610 fs_info->fs_devices->open_devices++;
2611 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2612
2613 *device_out = device;
2614 return ret;
2615
2616error:
2617 blkdev_put(bdev, FMODE_EXCL);
2618 return ret;
2619}
2620
2621void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2622 struct btrfs_device *tgtdev)
2623{
2624 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2625 tgtdev->io_width = fs_info->dev_root->sectorsize;
2626 tgtdev->io_align = fs_info->dev_root->sectorsize;
2627 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2628 tgtdev->dev_root = fs_info->dev_root;
2629 tgtdev->in_fs_metadata = 1;
2630}
2631
2632static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2633 struct btrfs_device *device)
2634{
2635 int ret;
2636 struct btrfs_path *path;
2637 struct btrfs_root *root;
2638 struct btrfs_dev_item *dev_item;
2639 struct extent_buffer *leaf;
2640 struct btrfs_key key;
2641
2642 root = device->dev_root->fs_info->chunk_root;
2643
2644 path = btrfs_alloc_path();
2645 if (!path)
2646 return -ENOMEM;
2647
2648 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2649 key.type = BTRFS_DEV_ITEM_KEY;
2650 key.offset = device->devid;
2651
2652 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2653 if (ret < 0)
2654 goto out;
2655
2656 if (ret > 0) {
2657 ret = -ENOENT;
2658 goto out;
2659 }
2660
2661 leaf = path->nodes[0];
2662 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2663
2664 btrfs_set_device_id(leaf, dev_item, device->devid);
2665 btrfs_set_device_type(leaf, dev_item, device->type);
2666 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2667 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2668 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2669 btrfs_set_device_total_bytes(leaf, dev_item,
2670 btrfs_device_get_disk_total_bytes(device));
2671 btrfs_set_device_bytes_used(leaf, dev_item,
2672 btrfs_device_get_bytes_used(device));
2673 btrfs_mark_buffer_dirty(leaf);
2674
2675out:
2676 btrfs_free_path(path);
2677 return ret;
2678}
2679
2680int btrfs_grow_device(struct btrfs_trans_handle *trans,
2681 struct btrfs_device *device, u64 new_size)
2682{
2683 struct btrfs_super_block *super_copy =
2684 device->dev_root->fs_info->super_copy;
2685 struct btrfs_fs_devices *fs_devices;
2686 u64 old_total;
2687 u64 diff;
2688
2689 if (!device->writeable)
2690 return -EACCES;
2691
2692 lock_chunks(device->dev_root);
2693 old_total = btrfs_super_total_bytes(super_copy);
2694 diff = new_size - device->total_bytes;
2695
2696 if (new_size <= device->total_bytes ||
2697 device->is_tgtdev_for_dev_replace) {
2698 unlock_chunks(device->dev_root);
2699 return -EINVAL;
2700 }
2701
2702 fs_devices = device->dev_root->fs_info->fs_devices;
2703
2704 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2705 device->fs_devices->total_rw_bytes += diff;
2706
2707 btrfs_device_set_total_bytes(device, new_size);
2708 btrfs_device_set_disk_total_bytes(device, new_size);
2709 btrfs_clear_space_info_full(device->dev_root->fs_info);
2710 if (list_empty(&device->resized_list))
2711 list_add_tail(&device->resized_list,
2712 &fs_devices->resized_devices);
2713 unlock_chunks(device->dev_root);
2714
2715 return btrfs_update_device(trans, device);
2716}
2717
2718static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2719 struct btrfs_root *root, u64 chunk_objectid,
2720 u64 chunk_offset)
2721{
2722 int ret;
2723 struct btrfs_path *path;
2724 struct btrfs_key key;
2725
2726 root = root->fs_info->chunk_root;
2727 path = btrfs_alloc_path();
2728 if (!path)
2729 return -ENOMEM;
2730
2731 key.objectid = chunk_objectid;
2732 key.offset = chunk_offset;
2733 key.type = BTRFS_CHUNK_ITEM_KEY;
2734
2735 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2736 if (ret < 0)
2737 goto out;
2738 else if (ret > 0) {
2739 btrfs_handle_fs_error(root->fs_info, -ENOENT,
2740 "Failed lookup while freeing chunk.");
2741 ret = -ENOENT;
2742 goto out;
2743 }
2744
2745 ret = btrfs_del_item(trans, root, path);
2746 if (ret < 0)
2747 btrfs_handle_fs_error(root->fs_info, ret,
2748 "Failed to delete chunk item.");
2749out:
2750 btrfs_free_path(path);
2751 return ret;
2752}
2753
2754static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2755 chunk_offset)
2756{
2757 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2758 struct btrfs_disk_key *disk_key;
2759 struct btrfs_chunk *chunk;
2760 u8 *ptr;
2761 int ret = 0;
2762 u32 num_stripes;
2763 u32 array_size;
2764 u32 len = 0;
2765 u32 cur;
2766 struct btrfs_key key;
2767
2768 lock_chunks(root);
2769 array_size = btrfs_super_sys_array_size(super_copy);
2770
2771 ptr = super_copy->sys_chunk_array;
2772 cur = 0;
2773
2774 while (cur < array_size) {
2775 disk_key = (struct btrfs_disk_key *)ptr;
2776 btrfs_disk_key_to_cpu(&key, disk_key);
2777
2778 len = sizeof(*disk_key);
2779
2780 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2781 chunk = (struct btrfs_chunk *)(ptr + len);
2782 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2783 len += btrfs_chunk_item_size(num_stripes);
2784 } else {
2785 ret = -EIO;
2786 break;
2787 }
2788 if (key.objectid == chunk_objectid &&
2789 key.offset == chunk_offset) {
2790 memmove(ptr, ptr + len, array_size - (cur + len));
2791 array_size -= len;
2792 btrfs_set_super_sys_array_size(super_copy, array_size);
2793 } else {
2794 ptr += len;
2795 cur += len;
2796 }
2797 }
2798 unlock_chunks(root);
2799 return ret;
2800}
2801
2802int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2803 struct btrfs_root *root, u64 chunk_offset)
2804{
2805 struct extent_map_tree *em_tree;
2806 struct extent_map *em;
2807 struct btrfs_root *extent_root = root->fs_info->extent_root;
2808 struct map_lookup *map;
2809 u64 dev_extent_len = 0;
2810 u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2811 int i, ret = 0;
2812 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2813
2814
2815 root = root->fs_info->chunk_root;
2816 em_tree = &root->fs_info->mapping_tree.map_tree;
2817
2818 read_lock(&em_tree->lock);
2819 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2820 read_unlock(&em_tree->lock);
2821
2822 if (!em || em->start > chunk_offset ||
2823 em->start + em->len < chunk_offset) {
2824
2825
2826
2827
2828
2829 ASSERT(0);
2830 if (em)
2831 free_extent_map(em);
2832 return -EINVAL;
2833 }
2834 map = em->map_lookup;
2835 lock_chunks(root->fs_info->chunk_root);
2836 check_system_chunk(trans, extent_root, map->type);
2837 unlock_chunks(root->fs_info->chunk_root);
2838
2839
2840
2841
2842
2843
2844 mutex_lock(&fs_devices->device_list_mutex);
2845 for (i = 0; i < map->num_stripes; i++) {
2846 struct btrfs_device *device = map->stripes[i].dev;
2847 ret = btrfs_free_dev_extent(trans, device,
2848 map->stripes[i].physical,
2849 &dev_extent_len);
2850 if (ret) {
2851 mutex_unlock(&fs_devices->device_list_mutex);
2852 btrfs_abort_transaction(trans, ret);
2853 goto out;
2854 }
2855
2856 if (device->bytes_used > 0) {
2857 lock_chunks(root);
2858 btrfs_device_set_bytes_used(device,
2859 device->bytes_used - dev_extent_len);
2860 spin_lock(&root->fs_info->free_chunk_lock);
2861 root->fs_info->free_chunk_space += dev_extent_len;
2862 spin_unlock(&root->fs_info->free_chunk_lock);
2863 btrfs_clear_space_info_full(root->fs_info);
2864 unlock_chunks(root);
2865 }
2866
2867 if (map->stripes[i].dev) {
2868 ret = btrfs_update_device(trans, map->stripes[i].dev);
2869 if (ret) {
2870 mutex_unlock(&fs_devices->device_list_mutex);
2871 btrfs_abort_transaction(trans, ret);
2872 goto out;
2873 }
2874 }
2875 }
2876 mutex_unlock(&fs_devices->device_list_mutex);
2877
2878 ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2879 if (ret) {
2880 btrfs_abort_transaction(trans, ret);
2881 goto out;
2882 }
2883
2884 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2885
2886 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2887 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2888 if (ret) {
2889 btrfs_abort_transaction(trans, ret);
2890 goto out;
2891 }
2892 }
2893
2894 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2895 if (ret) {
2896 btrfs_abort_transaction(trans, ret);
2897 goto out;
2898 }
2899
2900out:
2901
2902 free_extent_map(em);
2903 return ret;
2904}
2905
2906static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2907{
2908 struct btrfs_root *extent_root;
2909 struct btrfs_trans_handle *trans;
2910 int ret;
2911
2912 root = root->fs_info->chunk_root;
2913 extent_root = root->fs_info->extent_root;
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927 ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2928
2929 ret = btrfs_can_relocate(extent_root, chunk_offset);
2930 if (ret)
2931 return -ENOSPC;
2932
2933
2934 btrfs_scrub_pause(root);
2935 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2936 btrfs_scrub_continue(root);
2937 if (ret)
2938 return ret;
2939
2940 trans = btrfs_start_trans_remove_block_group(root->fs_info,
2941 chunk_offset);
2942 if (IS_ERR(trans)) {
2943 ret = PTR_ERR(trans);
2944 btrfs_handle_fs_error(root->fs_info, ret, NULL);
2945 return ret;
2946 }
2947
2948
2949
2950
2951
2952 ret = btrfs_remove_chunk(trans, root, chunk_offset);
2953 btrfs_end_transaction(trans, extent_root);
2954 return ret;
2955}
2956
2957static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2958{
2959 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2960 struct btrfs_path *path;
2961 struct extent_buffer *leaf;
2962 struct btrfs_chunk *chunk;
2963 struct btrfs_key key;
2964 struct btrfs_key found_key;
2965 u64 chunk_type;
2966 bool retried = false;
2967 int failed = 0;
2968 int ret;
2969
2970 path = btrfs_alloc_path();
2971 if (!path)
2972 return -ENOMEM;
2973
2974again:
2975 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2976 key.offset = (u64)-1;
2977 key.type = BTRFS_CHUNK_ITEM_KEY;
2978
2979 while (1) {
2980 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2981 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2982 if (ret < 0) {
2983 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2984 goto error;
2985 }
2986 BUG_ON(ret == 0);
2987
2988 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2989 key.type);
2990 if (ret)
2991 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2992 if (ret < 0)
2993 goto error;
2994 if (ret > 0)
2995 break;
2996
2997 leaf = path->nodes[0];
2998 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2999
3000 chunk = btrfs_item_ptr(leaf, path->slots[0],
3001 struct btrfs_chunk);
3002 chunk_type = btrfs_chunk_type(leaf, chunk);
3003 btrfs_release_path(path);
3004
3005 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3006 ret = btrfs_relocate_chunk(chunk_root,
3007 found_key.offset);
3008 if (ret == -ENOSPC)
3009 failed++;
3010 else
3011 BUG_ON(ret);
3012 }
3013 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
3014
3015 if (found_key.offset == 0)
3016 break;
3017 key.offset = found_key.offset - 1;
3018 }
3019 ret = 0;
3020 if (failed && !retried) {
3021 failed = 0;
3022 retried = true;
3023 goto again;
3024 } else if (WARN_ON(failed && retried)) {
3025 ret = -ENOSPC;
3026 }
3027error:
3028 btrfs_free_path(path);
3029 return ret;
3030}
3031
3032static int insert_balance_item(struct btrfs_root *root,
3033 struct btrfs_balance_control *bctl)
3034{
3035 struct btrfs_trans_handle *trans;
3036 struct btrfs_balance_item *item;
3037 struct btrfs_disk_balance_args disk_bargs;
3038 struct btrfs_path *path;
3039 struct extent_buffer *leaf;
3040 struct btrfs_key key;
3041 int ret, err;
3042
3043 path = btrfs_alloc_path();
3044 if (!path)
3045 return -ENOMEM;
3046
3047 trans = btrfs_start_transaction(root, 0);
3048 if (IS_ERR(trans)) {
3049 btrfs_free_path(path);
3050 return PTR_ERR(trans);
3051 }
3052
3053 key.objectid = BTRFS_BALANCE_OBJECTID;
3054 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3055 key.offset = 0;
3056
3057 ret = btrfs_insert_empty_item(trans, root, path, &key,
3058 sizeof(*item));
3059 if (ret)
3060 goto out;
3061
3062 leaf = path->nodes[0];
3063 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3064
3065 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
3066
3067 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3068 btrfs_set_balance_data(leaf, item, &disk_bargs);
3069 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3070 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3071 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3072 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3073
3074 btrfs_set_balance_flags(leaf, item, bctl->flags);
3075
3076 btrfs_mark_buffer_dirty(leaf);
3077out:
3078 btrfs_free_path(path);
3079 err = btrfs_commit_transaction(trans, root);
3080 if (err && !ret)
3081 ret = err;
3082 return ret;
3083}
3084
3085static int del_balance_item(struct btrfs_root *root)
3086{
3087 struct btrfs_trans_handle *trans;
3088 struct btrfs_path *path;
3089 struct btrfs_key key;
3090 int ret, err;
3091
3092 path = btrfs_alloc_path();
3093 if (!path)
3094 return -ENOMEM;
3095
3096 trans = btrfs_start_transaction(root, 0);
3097 if (IS_ERR(trans)) {
3098 btrfs_free_path(path);
3099 return PTR_ERR(trans);
3100 }
3101
3102 key.objectid = BTRFS_BALANCE_OBJECTID;
3103 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3104 key.offset = 0;
3105
3106 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3107 if (ret < 0)
3108 goto out;
3109 if (ret > 0) {
3110 ret = -ENOENT;
3111 goto out;
3112 }
3113
3114 ret = btrfs_del_item(trans, root, path);
3115out:
3116 btrfs_free_path(path);
3117 err = btrfs_commit_transaction(trans, root);
3118 if (err && !ret)
3119 ret = err;
3120 return ret;
3121}
3122
3123
3124
3125
3126
3127static void update_balance_args(struct btrfs_balance_control *bctl)
3128{
3129
3130
3131
3132 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3133 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3134 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3135 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3136 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3137 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3138
3139
3140
3141
3142
3143
3144
3145
3146 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3147 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3148 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3149 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3150 bctl->data.usage = 90;
3151 }
3152 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3153 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3154 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3155 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3156 bctl->sys.usage = 90;
3157 }
3158 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3159 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3160 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3161 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3162 bctl->meta.usage = 90;
3163 }
3164}
3165
3166
3167
3168
3169
3170
3171static void set_balance_control(struct btrfs_balance_control *bctl)
3172{
3173 struct btrfs_fs_info *fs_info = bctl->fs_info;
3174
3175 BUG_ON(fs_info->balance_ctl);
3176
3177 spin_lock(&fs_info->balance_lock);
3178 fs_info->balance_ctl = bctl;
3179 spin_unlock(&fs_info->balance_lock);
3180}
3181
3182static void unset_balance_control(struct btrfs_fs_info *fs_info)
3183{
3184 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3185
3186 BUG_ON(!fs_info->balance_ctl);
3187
3188 spin_lock(&fs_info->balance_lock);
3189 fs_info->balance_ctl = NULL;
3190 spin_unlock(&fs_info->balance_lock);
3191
3192 kfree(bctl);
3193}
3194
3195
3196
3197
3198
3199static int chunk_profiles_filter(u64 chunk_type,
3200 struct btrfs_balance_args *bargs)
3201{
3202 chunk_type = chunk_to_extended(chunk_type) &
3203 BTRFS_EXTENDED_PROFILE_MASK;
3204
3205 if (bargs->profiles & chunk_type)
3206 return 0;
3207
3208 return 1;
3209}
3210
3211static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3212 struct btrfs_balance_args *bargs)
3213{
3214 struct btrfs_block_group_cache *cache;
3215 u64 chunk_used;
3216 u64 user_thresh_min;
3217 u64 user_thresh_max;
3218 int ret = 1;
3219
3220 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3221 chunk_used = btrfs_block_group_used(&cache->item);
3222
3223 if (bargs->usage_min == 0)
3224 user_thresh_min = 0;
3225 else
3226 user_thresh_min = div_factor_fine(cache->key.offset,
3227 bargs->usage_min);
3228
3229 if (bargs->usage_max == 0)
3230 user_thresh_max = 1;
3231 else if (bargs->usage_max > 100)
3232 user_thresh_max = cache->key.offset;
3233 else
3234 user_thresh_max = div_factor_fine(cache->key.offset,
3235 bargs->usage_max);
3236
3237 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3238 ret = 0;
3239
3240 btrfs_put_block_group(cache);
3241 return ret;
3242}
3243
3244static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3245 u64 chunk_offset, struct btrfs_balance_args *bargs)
3246{
3247 struct btrfs_block_group_cache *cache;
3248 u64 chunk_used, user_thresh;
3249 int ret = 1;
3250
3251 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3252 chunk_used = btrfs_block_group_used(&cache->item);
3253
3254 if (bargs->usage_min == 0)
3255 user_thresh = 1;
3256 else if (bargs->usage > 100)
3257 user_thresh = cache->key.offset;
3258 else
3259 user_thresh = div_factor_fine(cache->key.offset,
3260 bargs->usage);
3261
3262 if (chunk_used < user_thresh)
3263 ret = 0;
3264
3265 btrfs_put_block_group(cache);
3266 return ret;
3267}
3268
3269static int chunk_devid_filter(struct extent_buffer *leaf,
3270 struct btrfs_chunk *chunk,
3271 struct btrfs_balance_args *bargs)
3272{
3273 struct btrfs_stripe *stripe;
3274 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3275 int i;
3276
3277 for (i = 0; i < num_stripes; i++) {
3278 stripe = btrfs_stripe_nr(chunk, i);
3279 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3280 return 0;
3281 }
3282
3283 return 1;
3284}
3285
3286
3287static int chunk_drange_filter(struct extent_buffer *leaf,
3288 struct btrfs_chunk *chunk,
3289 u64 chunk_offset,
3290 struct btrfs_balance_args *bargs)
3291{
3292 struct btrfs_stripe *stripe;
3293 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3294 u64 stripe_offset;
3295 u64 stripe_length;
3296 int factor;
3297 int i;
3298
3299 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3300 return 0;
3301
3302 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3303 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3304 factor = num_stripes / 2;
3305 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3306 factor = num_stripes - 1;
3307 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3308 factor = num_stripes - 2;
3309 } else {
3310 factor = num_stripes;
3311 }
3312
3313 for (i = 0; i < num_stripes; i++) {
3314 stripe = btrfs_stripe_nr(chunk, i);
3315 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3316 continue;
3317
3318 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3319 stripe_length = btrfs_chunk_length(leaf, chunk);
3320 stripe_length = div_u64(stripe_length, factor);
3321
3322 if (stripe_offset < bargs->pend &&
3323 stripe_offset + stripe_length > bargs->pstart)
3324 return 0;
3325 }
3326
3327 return 1;
3328}
3329
3330
3331static int chunk_vrange_filter(struct extent_buffer *leaf,
3332 struct btrfs_chunk *chunk,
3333 u64 chunk_offset,
3334 struct btrfs_balance_args *bargs)
3335{
3336 if (chunk_offset < bargs->vend &&
3337 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3338
3339 return 0;
3340
3341 return 1;
3342}
3343
3344static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3345 struct btrfs_chunk *chunk,
3346 struct btrfs_balance_args *bargs)
3347{
3348 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3349
3350 if (bargs->stripes_min <= num_stripes
3351 && num_stripes <= bargs->stripes_max)
3352 return 0;
3353
3354 return 1;
3355}
3356
3357static int chunk_soft_convert_filter(u64 chunk_type,
3358 struct btrfs_balance_args *bargs)
3359{
3360 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3361 return 0;
3362
3363 chunk_type = chunk_to_extended(chunk_type) &
3364 BTRFS_EXTENDED_PROFILE_MASK;
3365
3366 if (bargs->target == chunk_type)
3367 return 1;
3368
3369 return 0;
3370}
3371
3372static int should_balance_chunk(struct btrfs_root *root,
3373 struct extent_buffer *leaf,
3374 struct btrfs_chunk *chunk, u64 chunk_offset)
3375{
3376 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3377 struct btrfs_balance_args *bargs = NULL;
3378 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3379
3380
3381 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3382 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3383 return 0;
3384 }
3385
3386 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3387 bargs = &bctl->data;
3388 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3389 bargs = &bctl->sys;
3390 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3391 bargs = &bctl->meta;
3392
3393
3394 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3395 chunk_profiles_filter(chunk_type, bargs)) {
3396 return 0;
3397 }
3398
3399
3400 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3401 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3402 return 0;
3403 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3404 chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) {
3405 return 0;
3406 }
3407
3408
3409 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3410 chunk_devid_filter(leaf, chunk, bargs)) {
3411 return 0;
3412 }
3413
3414
3415 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3416 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3417 return 0;
3418 }
3419
3420
3421 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3422 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3423 return 0;
3424 }
3425
3426
3427 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3428 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3429 return 0;
3430 }
3431
3432
3433 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3434 chunk_soft_convert_filter(chunk_type, bargs)) {
3435 return 0;
3436 }
3437
3438
3439
3440
3441 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3442 if (bargs->limit == 0)
3443 return 0;
3444 else
3445 bargs->limit--;
3446 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3447
3448
3449
3450
3451
3452 if (bargs->limit_max == 0)
3453 return 0;
3454 else
3455 bargs->limit_max--;
3456 }
3457
3458 return 1;
3459}
3460
3461static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3462{
3463 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3464 struct btrfs_root *chunk_root = fs_info->chunk_root;
3465 struct btrfs_root *dev_root = fs_info->dev_root;
3466 struct list_head *devices;
3467 struct btrfs_device *device;
3468 u64 old_size;
3469 u64 size_to_free;
3470 u64 chunk_type;
3471 struct btrfs_chunk *chunk;
3472 struct btrfs_path *path = NULL;
3473 struct btrfs_key key;
3474 struct btrfs_key found_key;
3475 struct btrfs_trans_handle *trans;
3476 struct extent_buffer *leaf;
3477 int slot;
3478 int ret;
3479 int enospc_errors = 0;
3480 bool counting = true;
3481
3482 u64 limit_data = bctl->data.limit;
3483 u64 limit_meta = bctl->meta.limit;
3484 u64 limit_sys = bctl->sys.limit;
3485 u32 count_data = 0;
3486 u32 count_meta = 0;
3487 u32 count_sys = 0;
3488 int chunk_reserved = 0;
3489 u64 bytes_used = 0;
3490
3491
3492 devices = &fs_info->fs_devices->devices;
3493 list_for_each_entry(device, devices, dev_list) {
3494 old_size = btrfs_device_get_total_bytes(device);
3495 size_to_free = div_factor(old_size, 1);
3496 size_to_free = min_t(u64, size_to_free, SZ_1M);
3497 if (!device->writeable ||
3498 btrfs_device_get_total_bytes(device) -
3499 btrfs_device_get_bytes_used(device) > size_to_free ||
3500 device->is_tgtdev_for_dev_replace)
3501 continue;
3502
3503 ret = btrfs_shrink_device(device, old_size - size_to_free);
3504 if (ret == -ENOSPC)
3505 break;
3506 if (ret) {
3507
3508 WARN_ON(ret > 0);
3509 goto error;
3510 }
3511
3512 trans = btrfs_start_transaction(dev_root, 0);
3513 if (IS_ERR(trans)) {
3514 ret = PTR_ERR(trans);
3515 btrfs_info_in_rcu(fs_info,
3516 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3517 rcu_str_deref(device->name), ret,
3518 old_size, old_size - size_to_free);
3519 goto error;
3520 }
3521
3522 ret = btrfs_grow_device(trans, device, old_size);
3523 if (ret) {
3524 btrfs_end_transaction(trans, dev_root);
3525
3526 WARN_ON(ret > 0);
3527 btrfs_info_in_rcu(fs_info,
3528 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3529 rcu_str_deref(device->name), ret,
3530 old_size, old_size - size_to_free);
3531 goto error;
3532 }
3533
3534 btrfs_end_transaction(trans, dev_root);
3535 }
3536
3537
3538 path = btrfs_alloc_path();
3539 if (!path) {
3540 ret = -ENOMEM;
3541 goto error;
3542 }
3543
3544
3545 spin_lock(&fs_info->balance_lock);
3546 memset(&bctl->stat, 0, sizeof(bctl->stat));
3547 spin_unlock(&fs_info->balance_lock);
3548again:
3549 if (!counting) {
3550
3551
3552
3553
3554 bctl->data.limit = limit_data;
3555 bctl->meta.limit = limit_meta;
3556 bctl->sys.limit = limit_sys;
3557 }
3558 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3559 key.offset = (u64)-1;
3560 key.type = BTRFS_CHUNK_ITEM_KEY;
3561
3562 while (1) {
3563 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3564 atomic_read(&fs_info->balance_cancel_req)) {
3565 ret = -ECANCELED;
3566 goto error;
3567 }
3568
3569 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3570 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3571 if (ret < 0) {
3572 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3573 goto error;
3574 }
3575
3576
3577
3578
3579
3580 if (ret == 0)
3581 BUG();
3582
3583 ret = btrfs_previous_item(chunk_root, path, 0,
3584 BTRFS_CHUNK_ITEM_KEY);
3585 if (ret) {
3586 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3587 ret = 0;
3588 break;
3589 }
3590
3591 leaf = path->nodes[0];
3592 slot = path->slots[0];
3593 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3594
3595 if (found_key.objectid != key.objectid) {
3596 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3597 break;
3598 }
3599
3600 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3601 chunk_type = btrfs_chunk_type(leaf, chunk);
3602
3603 if (!counting) {
3604 spin_lock(&fs_info->balance_lock);
3605 bctl->stat.considered++;
3606 spin_unlock(&fs_info->balance_lock);
3607 }
3608
3609 ret = should_balance_chunk(chunk_root, leaf, chunk,
3610 found_key.offset);
3611
3612 btrfs_release_path(path);
3613 if (!ret) {
3614 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3615 goto loop;
3616 }
3617
3618 if (counting) {
3619 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3620 spin_lock(&fs_info->balance_lock);
3621 bctl->stat.expected++;
3622 spin_unlock(&fs_info->balance_lock);
3623
3624 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3625 count_data++;
3626 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3627 count_sys++;
3628 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3629 count_meta++;
3630
3631 goto loop;
3632 }
3633
3634
3635
3636
3637
3638 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3639 count_data < bctl->data.limit_min)
3640 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3641 count_meta < bctl->meta.limit_min)
3642 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3643 count_sys < bctl->sys.limit_min)) {
3644 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3645 goto loop;
3646 }
3647
3648 ASSERT(fs_info->data_sinfo);
3649 spin_lock(&fs_info->data_sinfo->lock);
3650 bytes_used = fs_info->data_sinfo->bytes_used;
3651 spin_unlock(&fs_info->data_sinfo->lock);
3652
3653 if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3654 !chunk_reserved && !bytes_used) {
3655 trans = btrfs_start_transaction(chunk_root, 0);
3656 if (IS_ERR(trans)) {
3657 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3658 ret = PTR_ERR(trans);
3659 goto error;
3660 }
3661
3662 ret = btrfs_force_chunk_alloc(trans, chunk_root,
3663 BTRFS_BLOCK_GROUP_DATA);
3664 btrfs_end_transaction(trans, chunk_root);
3665 if (ret < 0) {
3666 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3667 goto error;
3668 }
3669 chunk_reserved = 1;
3670 }
3671
3672 ret = btrfs_relocate_chunk(chunk_root,
3673 found_key.offset);
3674 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3675 if (ret && ret != -ENOSPC)
3676 goto error;
3677 if (ret == -ENOSPC) {
3678 enospc_errors++;
3679 } else {
3680 spin_lock(&fs_info->balance_lock);
3681 bctl->stat.completed++;
3682 spin_unlock(&fs_info->balance_lock);
3683 }
3684loop:
3685 if (found_key.offset == 0)
3686 break;
3687 key.offset = found_key.offset - 1;
3688 }
3689
3690 if (counting) {
3691 btrfs_release_path(path);
3692 counting = false;
3693 goto again;
3694 }
3695error:
3696 btrfs_free_path(path);
3697 if (enospc_errors) {
3698 btrfs_info(fs_info, "%d enospc errors during balance",
3699 enospc_errors);
3700 if (!ret)
3701 ret = -ENOSPC;
3702 }
3703
3704 return ret;
3705}
3706
3707
3708
3709
3710
3711
3712static int alloc_profile_is_valid(u64 flags, int extended)
3713{
3714 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3715 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3716
3717 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3718
3719
3720 if (flags & ~mask)
3721 return 0;
3722
3723
3724 if (flags == 0)
3725 return !extended;
3726
3727
3728 return (flags & (flags - 1)) == 0;
3729}
3730
3731static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3732{
3733
3734 return atomic_read(&fs_info->balance_cancel_req) ||
3735 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3736 atomic_read(&fs_info->balance_cancel_req) == 0);
3737}
3738
3739static void __cancel_balance(struct btrfs_fs_info *fs_info)
3740{
3741 int ret;
3742
3743 unset_balance_control(fs_info);
3744 ret = del_balance_item(fs_info->tree_root);
3745 if (ret)
3746 btrfs_handle_fs_error(fs_info, ret, NULL);
3747
3748 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3749}
3750
3751
3752static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg,
3753 u64 allowed)
3754{
3755 return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3756 (!alloc_profile_is_valid(bctl_arg->target, 1) ||
3757 (bctl_arg->target & ~allowed)));
3758}
3759
3760
3761
3762
3763int btrfs_balance(struct btrfs_balance_control *bctl,
3764 struct btrfs_ioctl_balance_args *bargs)
3765{
3766 struct btrfs_fs_info *fs_info = bctl->fs_info;
3767 u64 allowed;
3768 int mixed = 0;
3769 int ret;
3770 u64 num_devices;
3771 unsigned seq;
3772
3773 if (btrfs_fs_closing(fs_info) ||
3774 atomic_read(&fs_info->balance_pause_req) ||
3775 atomic_read(&fs_info->balance_cancel_req)) {
3776 ret = -EINVAL;
3777 goto out;
3778 }
3779
3780 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3781 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3782 mixed = 1;
3783
3784
3785
3786
3787
3788 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3789 if (mixed && (bctl->flags & allowed)) {
3790 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3791 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3792 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3793 btrfs_err(fs_info,
3794 "with mixed groups data and metadata balance options must be the same");
3795 ret = -EINVAL;
3796 goto out;
3797 }
3798 }
3799
3800 num_devices = fs_info->fs_devices->num_devices;
3801 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
3802 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3803 BUG_ON(num_devices < 1);
3804 num_devices--;
3805 }
3806 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
3807 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP;
3808 if (num_devices > 1)
3809 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3810 if (num_devices > 2)
3811 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3812 if (num_devices > 3)
3813 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3814 BTRFS_BLOCK_GROUP_RAID6);
3815 if (validate_convert_profile(&bctl->data, allowed)) {
3816 btrfs_err(fs_info,
3817 "unable to start balance with target data profile %llu",
3818 bctl->data.target);
3819 ret = -EINVAL;
3820 goto out;
3821 }
3822 if (validate_convert_profile(&bctl->meta, allowed)) {
3823 btrfs_err(fs_info,
3824 "unable to start balance with target metadata profile %llu",
3825 bctl->meta.target);
3826 ret = -EINVAL;
3827 goto out;
3828 }
3829 if (validate_convert_profile(&bctl->sys, allowed)) {
3830 btrfs_err(fs_info,
3831 "unable to start balance with target system profile %llu",
3832 bctl->sys.target);
3833 ret = -EINVAL;
3834 goto out;
3835 }
3836
3837
3838 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3839 BTRFS_BLOCK_GROUP_RAID10 |
3840 BTRFS_BLOCK_GROUP_RAID5 |
3841 BTRFS_BLOCK_GROUP_RAID6;
3842 do {
3843 seq = read_seqbegin(&fs_info->profiles_lock);
3844
3845 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3846 (fs_info->avail_system_alloc_bits & allowed) &&
3847 !(bctl->sys.target & allowed)) ||
3848 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3849 (fs_info->avail_metadata_alloc_bits & allowed) &&
3850 !(bctl->meta.target & allowed))) {
3851 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3852 btrfs_info(fs_info,
3853 "force reducing metadata integrity");
3854 } else {
3855 btrfs_err(fs_info,
3856 "balance will reduce metadata integrity, use force if you want this");
3857 ret = -EINVAL;
3858 goto out;
3859 }
3860 }
3861 } while (read_seqretry(&fs_info->profiles_lock, seq));
3862
3863 if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) <
3864 btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) {
3865 btrfs_warn(fs_info,
3866 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3867 bctl->meta.target, bctl->data.target);
3868 }
3869
3870 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3871 fs_info->num_tolerated_disk_barrier_failures = min(
3872 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info),
3873 btrfs_get_num_tolerated_disk_barrier_failures(
3874 bctl->sys.target));
3875 }
3876
3877 ret = insert_balance_item(fs_info->tree_root, bctl);
3878 if (ret && ret != -EEXIST)
3879 goto out;
3880
3881 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3882 BUG_ON(ret == -EEXIST);
3883 set_balance_control(bctl);
3884 } else {
3885 BUG_ON(ret != -EEXIST);
3886 spin_lock(&fs_info->balance_lock);
3887 update_balance_args(bctl);
3888 spin_unlock(&fs_info->balance_lock);
3889 }
3890
3891 atomic_inc(&fs_info->balance_running);
3892 mutex_unlock(&fs_info->balance_mutex);
3893
3894 ret = __btrfs_balance(fs_info);
3895
3896 mutex_lock(&fs_info->balance_mutex);
3897 atomic_dec(&fs_info->balance_running);
3898
3899 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3900 fs_info->num_tolerated_disk_barrier_failures =
3901 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3902 }
3903
3904 if (bargs) {
3905 memset(bargs, 0, sizeof(*bargs));
3906 update_ioctl_balance_args(fs_info, 0, bargs);
3907 }
3908
3909 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3910 balance_need_close(fs_info)) {
3911 __cancel_balance(fs_info);
3912 }
3913
3914 wake_up(&fs_info->balance_wait_q);
3915
3916 return ret;
3917out:
3918 if (bctl->flags & BTRFS_BALANCE_RESUME)
3919 __cancel_balance(fs_info);
3920 else {
3921 kfree(bctl);
3922 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3923 }
3924 return ret;
3925}
3926
3927static int balance_kthread(void *data)
3928{
3929 struct btrfs_fs_info *fs_info = data;
3930 int ret = 0;
3931
3932 mutex_lock(&fs_info->volume_mutex);
3933 mutex_lock(&fs_info->balance_mutex);
3934
3935 if (fs_info->balance_ctl) {
3936 btrfs_info(fs_info, "continuing balance");
3937 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3938 }
3939
3940 mutex_unlock(&fs_info->balance_mutex);
3941 mutex_unlock(&fs_info->volume_mutex);
3942
3943 return ret;
3944}
3945
3946int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3947{
3948 struct task_struct *tsk;
3949
3950 spin_lock(&fs_info->balance_lock);
3951 if (!fs_info->balance_ctl) {
3952 spin_unlock(&fs_info->balance_lock);
3953 return 0;
3954 }
3955 spin_unlock(&fs_info->balance_lock);
3956
3957 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
3958 btrfs_info(fs_info, "force skipping balance");
3959 return 0;
3960 }
3961
3962 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3963 return PTR_ERR_OR_ZERO(tsk);
3964}
3965
3966int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3967{
3968 struct btrfs_balance_control *bctl;
3969 struct btrfs_balance_item *item;
3970 struct btrfs_disk_balance_args disk_bargs;
3971 struct btrfs_path *path;
3972 struct extent_buffer *leaf;
3973 struct btrfs_key key;
3974 int ret;
3975
3976 path = btrfs_alloc_path();
3977 if (!path)
3978 return -ENOMEM;
3979
3980 key.objectid = BTRFS_BALANCE_OBJECTID;
3981 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3982 key.offset = 0;
3983
3984 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3985 if (ret < 0)
3986 goto out;
3987 if (ret > 0) {
3988 ret = 0;
3989 goto out;
3990 }
3991
3992 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3993 if (!bctl) {
3994 ret = -ENOMEM;
3995 goto out;
3996 }
3997
3998 leaf = path->nodes[0];
3999 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4000
4001 bctl->fs_info = fs_info;
4002 bctl->flags = btrfs_balance_flags(leaf, item);
4003 bctl->flags |= BTRFS_BALANCE_RESUME;
4004
4005 btrfs_balance_data(leaf, item, &disk_bargs);
4006 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4007 btrfs_balance_meta(leaf, item, &disk_bargs);
4008 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4009 btrfs_balance_sys(leaf, item, &disk_bargs);
4010 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4011
4012 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
4013
4014 mutex_lock(&fs_info->volume_mutex);
4015 mutex_lock(&fs_info->balance_mutex);
4016
4017 set_balance_control(bctl);
4018
4019 mutex_unlock(&fs_info->balance_mutex);
4020 mutex_unlock(&fs_info->volume_mutex);
4021out:
4022 btrfs_free_path(path);
4023 return ret;
4024}
4025
4026int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4027{
4028 int ret = 0;
4029
4030 mutex_lock(&fs_info->balance_mutex);
4031 if (!fs_info->balance_ctl) {
4032 mutex_unlock(&fs_info->balance_mutex);
4033 return -ENOTCONN;
4034 }
4035
4036 if (atomic_read(&fs_info->balance_running)) {
4037 atomic_inc(&fs_info->balance_pause_req);
4038 mutex_unlock(&fs_info->balance_mutex);
4039
4040 wait_event(fs_info->balance_wait_q,
4041 atomic_read(&fs_info->balance_running) == 0);
4042
4043 mutex_lock(&fs_info->balance_mutex);
4044
4045 BUG_ON(atomic_read(&fs_info->balance_running));
4046 atomic_dec(&fs_info->balance_pause_req);
4047 } else {
4048 ret = -ENOTCONN;
4049 }
4050
4051 mutex_unlock(&fs_info->balance_mutex);
4052 return ret;
4053}
4054
4055int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4056{
4057 if (fs_info->sb->s_flags & MS_RDONLY)
4058 return -EROFS;
4059
4060 mutex_lock(&fs_info->balance_mutex);
4061 if (!fs_info->balance_ctl) {
4062 mutex_unlock(&fs_info->balance_mutex);
4063 return -ENOTCONN;
4064 }
4065
4066 atomic_inc(&fs_info->balance_cancel_req);
4067
4068
4069
4070
4071 if (atomic_read(&fs_info->balance_running)) {
4072 mutex_unlock(&fs_info->balance_mutex);
4073 wait_event(fs_info->balance_wait_q,
4074 atomic_read(&fs_info->balance_running) == 0);
4075 mutex_lock(&fs_info->balance_mutex);
4076 } else {
4077
4078 mutex_unlock(&fs_info->balance_mutex);
4079 mutex_lock(&fs_info->volume_mutex);
4080 mutex_lock(&fs_info->balance_mutex);
4081
4082 if (fs_info->balance_ctl)
4083 __cancel_balance(fs_info);
4084
4085 mutex_unlock(&fs_info->volume_mutex);
4086 }
4087
4088 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
4089 atomic_dec(&fs_info->balance_cancel_req);
4090 mutex_unlock(&fs_info->balance_mutex);
4091 return 0;
4092}
4093
4094static int btrfs_uuid_scan_kthread(void *data)
4095{
4096 struct btrfs_fs_info *fs_info = data;
4097 struct btrfs_root *root = fs_info->tree_root;
4098 struct btrfs_key key;
4099 struct btrfs_key max_key;
4100 struct btrfs_path *path = NULL;
4101 int ret = 0;
4102 struct extent_buffer *eb;
4103 int slot;
4104 struct btrfs_root_item root_item;
4105 u32 item_size;
4106 struct btrfs_trans_handle *trans = NULL;
4107
4108 path = btrfs_alloc_path();
4109 if (!path) {
4110 ret = -ENOMEM;
4111 goto out;
4112 }
4113
4114 key.objectid = 0;
4115 key.type = BTRFS_ROOT_ITEM_KEY;
4116 key.offset = 0;
4117
4118 max_key.objectid = (u64)-1;
4119 max_key.type = BTRFS_ROOT_ITEM_KEY;
4120 max_key.offset = (u64)-1;
4121
4122 while (1) {
4123 ret = btrfs_search_forward(root, &key, path, 0);
4124 if (ret) {
4125 if (ret > 0)
4126 ret = 0;
4127 break;
4128 }
4129
4130 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4131 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4132 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4133 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4134 goto skip;
4135
4136 eb = path->nodes[0];
4137 slot = path->slots[0];
4138 item_size = btrfs_item_size_nr(eb, slot);
4139 if (item_size < sizeof(root_item))
4140 goto skip;
4141
4142 read_extent_buffer(eb, &root_item,
4143 btrfs_item_ptr_offset(eb, slot),
4144 (int)sizeof(root_item));
4145 if (btrfs_root_refs(&root_item) == 0)
4146 goto skip;
4147
4148 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4149 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4150 if (trans)
4151 goto update_tree;
4152
4153 btrfs_release_path(path);
4154
4155
4156
4157
4158 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4159 if (IS_ERR(trans)) {
4160 ret = PTR_ERR(trans);
4161 break;
4162 }
4163 continue;
4164 } else {
4165 goto skip;
4166 }
4167update_tree:
4168 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4169 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4170 root_item.uuid,
4171 BTRFS_UUID_KEY_SUBVOL,
4172 key.objectid);
4173 if (ret < 0) {
4174 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4175 ret);
4176 break;
4177 }
4178 }
4179
4180 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4181 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
4182 root_item.received_uuid,
4183 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4184 key.objectid);
4185 if (ret < 0) {
4186 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4187 ret);
4188 break;
4189 }
4190 }
4191
4192skip:
4193 if (trans) {
4194 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
4195 trans = NULL;
4196 if (ret)
4197 break;
4198 }
4199
4200 btrfs_release_path(path);
4201 if (key.offset < (u64)-1) {
4202 key.offset++;
4203 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4204 key.offset = 0;
4205 key.type = BTRFS_ROOT_ITEM_KEY;
4206 } else if (key.objectid < (u64)-1) {
4207 key.offset = 0;
4208 key.type = BTRFS_ROOT_ITEM_KEY;
4209 key.objectid++;
4210 } else {
4211 break;
4212 }
4213 cond_resched();
4214 }
4215
4216out:
4217 btrfs_free_path(path);
4218 if (trans && !IS_ERR(trans))
4219 btrfs_end_transaction(trans, fs_info->uuid_root);
4220 if (ret)
4221 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4222 else
4223 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4224 up(&fs_info->uuid_tree_rescan_sem);
4225 return 0;
4226}
4227
4228
4229
4230
4231
4232
4233
4234
4235static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
4236 u8 *uuid, u8 type, u64 subid)
4237{
4238 struct btrfs_key key;
4239 int ret = 0;
4240 struct btrfs_root *subvol_root;
4241
4242 if (type != BTRFS_UUID_KEY_SUBVOL &&
4243 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
4244 goto out;
4245
4246 key.objectid = subid;
4247 key.type = BTRFS_ROOT_ITEM_KEY;
4248 key.offset = (u64)-1;
4249 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
4250 if (IS_ERR(subvol_root)) {
4251 ret = PTR_ERR(subvol_root);
4252 if (ret == -ENOENT)
4253 ret = 1;
4254 goto out;
4255 }
4256
4257 switch (type) {
4258 case BTRFS_UUID_KEY_SUBVOL:
4259 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
4260 ret = 1;
4261 break;
4262 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
4263 if (memcmp(uuid, subvol_root->root_item.received_uuid,
4264 BTRFS_UUID_SIZE))
4265 ret = 1;
4266 break;
4267 }
4268
4269out:
4270 return ret;
4271}
4272
4273static int btrfs_uuid_rescan_kthread(void *data)
4274{
4275 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
4276 int ret;
4277
4278
4279
4280
4281
4282
4283 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4284 if (ret < 0) {
4285 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4286 up(&fs_info->uuid_tree_rescan_sem);
4287 return ret;
4288 }
4289 return btrfs_uuid_scan_kthread(data);
4290}
4291
4292int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4293{
4294 struct btrfs_trans_handle *trans;
4295 struct btrfs_root *tree_root = fs_info->tree_root;
4296 struct btrfs_root *uuid_root;
4297 struct task_struct *task;
4298 int ret;
4299
4300
4301
4302
4303
4304 trans = btrfs_start_transaction(tree_root, 2);
4305 if (IS_ERR(trans))
4306 return PTR_ERR(trans);
4307
4308 uuid_root = btrfs_create_tree(trans, fs_info,
4309 BTRFS_UUID_TREE_OBJECTID);
4310 if (IS_ERR(uuid_root)) {
4311 ret = PTR_ERR(uuid_root);
4312 btrfs_abort_transaction(trans, ret);
4313 btrfs_end_transaction(trans, tree_root);
4314 return ret;
4315 }
4316
4317 fs_info->uuid_root = uuid_root;
4318
4319 ret = btrfs_commit_transaction(trans, tree_root);
4320 if (ret)
4321 return ret;
4322
4323 down(&fs_info->uuid_tree_rescan_sem);
4324 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4325 if (IS_ERR(task)) {
4326
4327 btrfs_warn(fs_info, "failed to start uuid_scan task");
4328 up(&fs_info->uuid_tree_rescan_sem);
4329 return PTR_ERR(task);
4330 }
4331
4332 return 0;
4333}
4334
4335int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4336{
4337 struct task_struct *task;
4338
4339 down(&fs_info->uuid_tree_rescan_sem);
4340 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4341 if (IS_ERR(task)) {
4342
4343 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4344 up(&fs_info->uuid_tree_rescan_sem);
4345 return PTR_ERR(task);
4346 }
4347
4348 return 0;
4349}
4350
4351
4352
4353
4354
4355
4356int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4357{
4358 struct btrfs_trans_handle *trans;
4359 struct btrfs_root *root = device->dev_root;
4360 struct btrfs_dev_extent *dev_extent = NULL;
4361 struct btrfs_path *path;
4362 u64 length;
4363 u64 chunk_offset;
4364 int ret;
4365 int slot;
4366 int failed = 0;
4367 bool retried = false;
4368 bool checked_pending_chunks = false;
4369 struct extent_buffer *l;
4370 struct btrfs_key key;
4371 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4372 u64 old_total = btrfs_super_total_bytes(super_copy);
4373 u64 old_size = btrfs_device_get_total_bytes(device);
4374 u64 diff = old_size - new_size;
4375
4376 if (device->is_tgtdev_for_dev_replace)
4377 return -EINVAL;
4378
4379 path = btrfs_alloc_path();
4380 if (!path)
4381 return -ENOMEM;
4382
4383 path->reada = READA_FORWARD;
4384
4385 lock_chunks(root);
4386
4387 btrfs_device_set_total_bytes(device, new_size);
4388 if (device->writeable) {
4389 device->fs_devices->total_rw_bytes -= diff;
4390 spin_lock(&root->fs_info->free_chunk_lock);
4391 root->fs_info->free_chunk_space -= diff;
4392 spin_unlock(&root->fs_info->free_chunk_lock);
4393 }
4394 unlock_chunks(root);
4395
4396again:
4397 key.objectid = device->devid;
4398 key.offset = (u64)-1;
4399 key.type = BTRFS_DEV_EXTENT_KEY;
4400
4401 do {
4402 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4403 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4404 if (ret < 0) {
4405 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4406 goto done;
4407 }
4408
4409 ret = btrfs_previous_item(root, path, 0, key.type);
4410 if (ret)
4411 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4412 if (ret < 0)
4413 goto done;
4414 if (ret) {
4415 ret = 0;
4416 btrfs_release_path(path);
4417 break;
4418 }
4419
4420 l = path->nodes[0];
4421 slot = path->slots[0];
4422 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4423
4424 if (key.objectid != device->devid) {
4425 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4426 btrfs_release_path(path);
4427 break;
4428 }
4429
4430 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4431 length = btrfs_dev_extent_length(l, dev_extent);
4432
4433 if (key.offset + length <= new_size) {
4434 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4435 btrfs_release_path(path);
4436 break;
4437 }
4438
4439 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4440 btrfs_release_path(path);
4441
4442 ret = btrfs_relocate_chunk(root, chunk_offset);
4443 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4444 if (ret && ret != -ENOSPC)
4445 goto done;
4446 if (ret == -ENOSPC)
4447 failed++;
4448 } while (key.offset-- > 0);
4449
4450 if (failed && !retried) {
4451 failed = 0;
4452 retried = true;
4453 goto again;
4454 } else if (failed && retried) {
4455 ret = -ENOSPC;
4456 goto done;
4457 }
4458
4459
4460 trans = btrfs_start_transaction(root, 0);
4461 if (IS_ERR(trans)) {
4462 ret = PTR_ERR(trans);
4463 goto done;
4464 }
4465
4466 lock_chunks(root);
4467
4468
4469
4470
4471
4472
4473
4474
4475
4476
4477
4478
4479
4480 if (!checked_pending_chunks) {
4481 u64 start = new_size;
4482 u64 len = old_size - new_size;
4483
4484 if (contains_pending_extent(trans->transaction, device,
4485 &start, len)) {
4486 unlock_chunks(root);
4487 checked_pending_chunks = true;
4488 failed = 0;
4489 retried = false;
4490 ret = btrfs_commit_transaction(trans, root);
4491 if (ret)
4492 goto done;
4493 goto again;
4494 }
4495 }
4496
4497 btrfs_device_set_disk_total_bytes(device, new_size);
4498 if (list_empty(&device->resized_list))
4499 list_add_tail(&device->resized_list,
4500 &root->fs_info->fs_devices->resized_devices);
4501
4502 WARN_ON(diff > old_total);
4503 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4504 unlock_chunks(root);
4505
4506
4507 ret = btrfs_update_device(trans, device);
4508 btrfs_end_transaction(trans, root);
4509done:
4510 btrfs_free_path(path);
4511 if (ret) {
4512 lock_chunks(root);
4513 btrfs_device_set_total_bytes(device, old_size);
4514 if (device->writeable)
4515 device->fs_devices->total_rw_bytes += diff;
4516 spin_lock(&root->fs_info->free_chunk_lock);
4517 root->fs_info->free_chunk_space += diff;
4518 spin_unlock(&root->fs_info->free_chunk_lock);
4519 unlock_chunks(root);
4520 }
4521 return ret;
4522}
4523
4524static int btrfs_add_system_chunk(struct btrfs_root *root,
4525 struct btrfs_key *key,
4526 struct btrfs_chunk *chunk, int item_size)
4527{
4528 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4529 struct btrfs_disk_key disk_key;
4530 u32 array_size;
4531 u8 *ptr;
4532
4533 lock_chunks(root);
4534 array_size = btrfs_super_sys_array_size(super_copy);
4535 if (array_size + item_size + sizeof(disk_key)
4536 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4537 unlock_chunks(root);
4538 return -EFBIG;
4539 }
4540
4541 ptr = super_copy->sys_chunk_array + array_size;
4542 btrfs_cpu_key_to_disk(&disk_key, key);
4543 memcpy(ptr, &disk_key, sizeof(disk_key));
4544 ptr += sizeof(disk_key);
4545 memcpy(ptr, chunk, item_size);
4546 item_size += sizeof(disk_key);
4547 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4548 unlock_chunks(root);
4549
4550 return 0;
4551}
4552
4553
4554
4555
4556static int btrfs_cmp_device_info(const void *a, const void *b)
4557{
4558 const struct btrfs_device_info *di_a = a;
4559 const struct btrfs_device_info *di_b = b;
4560
4561 if (di_a->max_avail > di_b->max_avail)
4562 return -1;
4563 if (di_a->max_avail < di_b->max_avail)
4564 return 1;
4565 if (di_a->total_avail > di_b->total_avail)
4566 return -1;
4567 if (di_a->total_avail < di_b->total_avail)
4568 return 1;
4569 return 0;
4570}
4571
4572static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4573{
4574
4575 return SZ_64K;
4576}
4577
4578static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4579{
4580 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4581 return;
4582
4583 btrfs_set_fs_incompat(info, RAID56);
4584}
4585
4586#define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r) \
4587 - sizeof(struct btrfs_chunk)) \
4588 / sizeof(struct btrfs_stripe) + 1)
4589
4590#define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4591 - 2 * sizeof(struct btrfs_disk_key) \
4592 - 2 * sizeof(struct btrfs_chunk)) \
4593 / sizeof(struct btrfs_stripe) + 1)
4594
4595static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4596 struct btrfs_root *extent_root, u64 start,
4597 u64 type)
4598{
4599 struct btrfs_fs_info *info = extent_root->fs_info;
4600 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4601 struct list_head *cur;
4602 struct map_lookup *map = NULL;
4603 struct extent_map_tree *em_tree;
4604 struct extent_map *em;
4605 struct btrfs_device_info *devices_info = NULL;
4606 u64 total_avail;
4607 int num_stripes;
4608 int data_stripes;
4609
4610 int sub_stripes;
4611 int dev_stripes;
4612 int devs_max;
4613 int devs_min;
4614 int devs_increment;
4615 int ncopies;
4616 int ret;
4617 u64 max_stripe_size;
4618 u64 max_chunk_size;
4619 u64 stripe_size;
4620 u64 num_bytes;
4621 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4622 int ndevs;
4623 int i;
4624 int j;
4625 int index;
4626
4627 BUG_ON(!alloc_profile_is_valid(type, 0));
4628
4629 if (list_empty(&fs_devices->alloc_list))
4630 return -ENOSPC;
4631
4632 index = __get_raid_index(type);
4633
4634 sub_stripes = btrfs_raid_array[index].sub_stripes;
4635 dev_stripes = btrfs_raid_array[index].dev_stripes;
4636 devs_max = btrfs_raid_array[index].devs_max;
4637 devs_min = btrfs_raid_array[index].devs_min;
4638 devs_increment = btrfs_raid_array[index].devs_increment;
4639 ncopies = btrfs_raid_array[index].ncopies;
4640
4641 if (type & BTRFS_BLOCK_GROUP_DATA) {
4642 max_stripe_size = SZ_1G;
4643 max_chunk_size = 10 * max_stripe_size;
4644 if (!devs_max)
4645 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4646 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4647
4648 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4649 max_stripe_size = SZ_1G;
4650 else
4651 max_stripe_size = SZ_256M;
4652 max_chunk_size = max_stripe_size;
4653 if (!devs_max)
4654 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4655 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4656 max_stripe_size = SZ_32M;
4657 max_chunk_size = 2 * max_stripe_size;
4658 if (!devs_max)
4659 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4660 } else {
4661 btrfs_err(info, "invalid chunk type 0x%llx requested",
4662 type);
4663 BUG_ON(1);
4664 }
4665
4666
4667 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4668 max_chunk_size);
4669
4670 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4671 GFP_NOFS);
4672 if (!devices_info)
4673 return -ENOMEM;
4674
4675 cur = fs_devices->alloc_list.next;
4676
4677
4678
4679
4680
4681 ndevs = 0;
4682 while (cur != &fs_devices->alloc_list) {
4683 struct btrfs_device *device;
4684 u64 max_avail;
4685 u64 dev_offset;
4686
4687 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4688
4689 cur = cur->next;
4690
4691 if (!device->writeable) {
4692 WARN(1, KERN_ERR
4693 "BTRFS: read-only device in alloc_list\n");
4694 continue;
4695 }
4696
4697 if (!device->in_fs_metadata ||
4698 device->is_tgtdev_for_dev_replace)
4699 continue;
4700
4701 if (device->total_bytes > device->bytes_used)
4702 total_avail = device->total_bytes - device->bytes_used;
4703 else
4704 total_avail = 0;
4705
4706
4707 if (total_avail == 0)
4708 continue;
4709
4710 ret = find_free_dev_extent(trans, device,
4711 max_stripe_size * dev_stripes,
4712 &dev_offset, &max_avail);
4713 if (ret && ret != -ENOSPC)
4714 goto error;
4715
4716 if (ret == 0)
4717 max_avail = max_stripe_size * dev_stripes;
4718
4719 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4720 continue;
4721
4722 if (ndevs == fs_devices->rw_devices) {
4723 WARN(1, "%s: found more than %llu devices\n",
4724 __func__, fs_devices->rw_devices);
4725 break;
4726 }
4727 devices_info[ndevs].dev_offset = dev_offset;
4728 devices_info[ndevs].max_avail = max_avail;
4729 devices_info[ndevs].total_avail = total_avail;
4730 devices_info[ndevs].dev = device;
4731 ++ndevs;
4732 }
4733
4734
4735
4736
4737 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4738 btrfs_cmp_device_info, NULL);
4739
4740
4741 ndevs -= ndevs % devs_increment;
4742
4743 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4744 ret = -ENOSPC;
4745 goto error;
4746 }
4747
4748 if (devs_max && ndevs > devs_max)
4749 ndevs = devs_max;
4750
4751
4752
4753
4754 stripe_size = devices_info[ndevs-1].max_avail;
4755 num_stripes = ndevs * dev_stripes;
4756
4757
4758
4759
4760
4761 data_stripes = num_stripes / ncopies;
4762
4763 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4764 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4765 extent_root->stripesize);
4766 data_stripes = num_stripes - 1;
4767 }
4768 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4769 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4770 extent_root->stripesize);
4771 data_stripes = num_stripes - 2;
4772 }
4773
4774
4775
4776
4777
4778
4779 if (stripe_size * data_stripes > max_chunk_size) {
4780 u64 mask = (1ULL << 24) - 1;
4781
4782 stripe_size = div_u64(max_chunk_size, data_stripes);
4783
4784
4785 stripe_size = (stripe_size + mask) & ~mask;
4786
4787
4788
4789
4790 if (stripe_size > devices_info[ndevs-1].max_avail)
4791 stripe_size = devices_info[ndevs-1].max_avail;
4792 }
4793
4794 stripe_size = div_u64(stripe_size, dev_stripes);
4795
4796
4797 stripe_size = div_u64(stripe_size, raid_stripe_len);
4798 stripe_size *= raid_stripe_len;
4799
4800 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4801 if (!map) {
4802 ret = -ENOMEM;
4803 goto error;
4804 }
4805 map->num_stripes = num_stripes;
4806
4807 for (i = 0; i < ndevs; ++i) {
4808 for (j = 0; j < dev_stripes; ++j) {
4809 int s = i * dev_stripes + j;
4810 map->stripes[s].dev = devices_info[i].dev;
4811 map->stripes[s].physical = devices_info[i].dev_offset +
4812 j * stripe_size;
4813 }
4814 }
4815 map->sector_size = extent_root->sectorsize;
4816 map->stripe_len = raid_stripe_len;
4817 map->io_align = raid_stripe_len;
4818 map->io_width = raid_stripe_len;
4819 map->type = type;
4820 map->sub_stripes = sub_stripes;
4821
4822 num_bytes = stripe_size * data_stripes;
4823
4824 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4825
4826 em = alloc_extent_map();
4827 if (!em) {
4828 kfree(map);
4829 ret = -ENOMEM;
4830 goto error;
4831 }
4832 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4833 em->map_lookup = map;
4834 em->start = start;
4835 em->len = num_bytes;
4836 em->block_start = 0;
4837 em->block_len = em->len;
4838 em->orig_block_len = stripe_size;
4839
4840 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4841 write_lock(&em_tree->lock);
4842 ret = add_extent_mapping(em_tree, em, 0);
4843 if (!ret) {
4844 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4845 atomic_inc(&em->refs);
4846 }
4847 write_unlock(&em_tree->lock);
4848 if (ret) {
4849 free_extent_map(em);
4850 goto error;
4851 }
4852
4853 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4854 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4855 start, num_bytes);
4856 if (ret)
4857 goto error_del_extent;
4858
4859 for (i = 0; i < map->num_stripes; i++) {
4860 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4861 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4862 }
4863
4864 spin_lock(&extent_root->fs_info->free_chunk_lock);
4865 extent_root->fs_info->free_chunk_space -= (stripe_size *
4866 map->num_stripes);
4867 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4868
4869 free_extent_map(em);
4870 check_raid56_incompat_flag(extent_root->fs_info, type);
4871
4872 kfree(devices_info);
4873 return 0;
4874
4875error_del_extent:
4876 write_lock(&em_tree->lock);
4877 remove_extent_mapping(em_tree, em);
4878 write_unlock(&em_tree->lock);
4879
4880
4881 free_extent_map(em);
4882
4883 free_extent_map(em);
4884
4885 free_extent_map(em);
4886error:
4887 kfree(devices_info);
4888 return ret;
4889}
4890
4891int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4892 struct btrfs_root *extent_root,
4893 u64 chunk_offset, u64 chunk_size)
4894{
4895 struct btrfs_key key;
4896 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4897 struct btrfs_device *device;
4898 struct btrfs_chunk *chunk;
4899 struct btrfs_stripe *stripe;
4900 struct extent_map_tree *em_tree;
4901 struct extent_map *em;
4902 struct map_lookup *map;
4903 size_t item_size;
4904 u64 dev_offset;
4905 u64 stripe_size;
4906 int i = 0;
4907 int ret = 0;
4908
4909 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4910 read_lock(&em_tree->lock);
4911 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4912 read_unlock(&em_tree->lock);
4913
4914 if (!em) {
4915 btrfs_crit(extent_root->fs_info,
4916 "unable to find logical %Lu len %Lu",
4917 chunk_offset, chunk_size);
4918 return -EINVAL;
4919 }
4920
4921 if (em->start != chunk_offset || em->len != chunk_size) {
4922 btrfs_crit(extent_root->fs_info,
4923 "found a bad mapping, wanted %Lu-%Lu, found %Lu-%Lu",
4924 chunk_offset, chunk_size, em->start, em->len);
4925 free_extent_map(em);
4926 return -EINVAL;
4927 }
4928
4929 map = em->map_lookup;
4930 item_size = btrfs_chunk_item_size(map->num_stripes);
4931 stripe_size = em->orig_block_len;
4932
4933 chunk = kzalloc(item_size, GFP_NOFS);
4934 if (!chunk) {
4935 ret = -ENOMEM;
4936 goto out;
4937 }
4938
4939
4940
4941
4942
4943
4944
4945
4946 mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4947 for (i = 0; i < map->num_stripes; i++) {
4948 device = map->stripes[i].dev;
4949 dev_offset = map->stripes[i].physical;
4950
4951 ret = btrfs_update_device(trans, device);
4952 if (ret)
4953 break;
4954 ret = btrfs_alloc_dev_extent(trans, device,
4955 chunk_root->root_key.objectid,
4956 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4957 chunk_offset, dev_offset,
4958 stripe_size);
4959 if (ret)
4960 break;
4961 }
4962 if (ret) {
4963 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4964 goto out;
4965 }
4966
4967 stripe = &chunk->stripe;
4968 for (i = 0; i < map->num_stripes; i++) {
4969 device = map->stripes[i].dev;
4970 dev_offset = map->stripes[i].physical;
4971
4972 btrfs_set_stack_stripe_devid(stripe, device->devid);
4973 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4974 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4975 stripe++;
4976 }
4977 mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex);
4978
4979 btrfs_set_stack_chunk_length(chunk, chunk_size);
4980 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4981 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4982 btrfs_set_stack_chunk_type(chunk, map->type);
4983 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4984 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4985 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4986 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4987 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4988
4989 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4990 key.type = BTRFS_CHUNK_ITEM_KEY;
4991 key.offset = chunk_offset;
4992
4993 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4994 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4995
4996
4997
4998
4999 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
5000 item_size);
5001 }
5002
5003out:
5004 kfree(chunk);
5005 free_extent_map(em);
5006 return ret;
5007}
5008
5009
5010
5011
5012
5013
5014
5015
5016int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5017 struct btrfs_root *extent_root, u64 type)
5018{
5019 u64 chunk_offset;
5020
5021 ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
5022 chunk_offset = find_next_chunk(extent_root->fs_info);
5023 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
5024}
5025
5026static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
5027 struct btrfs_root *root,
5028 struct btrfs_device *device)
5029{
5030 u64 chunk_offset;
5031 u64 sys_chunk_offset;
5032 u64 alloc_profile;
5033 struct btrfs_fs_info *fs_info = root->fs_info;
5034 struct btrfs_root *extent_root = fs_info->extent_root;
5035 int ret;
5036
5037 chunk_offset = find_next_chunk(fs_info);
5038 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
5039 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
5040 alloc_profile);
5041 if (ret)
5042 return ret;
5043
5044 sys_chunk_offset = find_next_chunk(root->fs_info);
5045 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
5046 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
5047 alloc_profile);
5048 return ret;
5049}
5050
5051static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5052{
5053 int max_errors;
5054
5055 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
5056 BTRFS_BLOCK_GROUP_RAID10 |
5057 BTRFS_BLOCK_GROUP_RAID5 |
5058 BTRFS_BLOCK_GROUP_DUP)) {
5059 max_errors = 1;
5060 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
5061 max_errors = 2;
5062 } else {
5063 max_errors = 0;
5064 }
5065
5066 return max_errors;
5067}
5068
5069int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
5070{
5071 struct extent_map *em;
5072 struct map_lookup *map;
5073 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5074 int readonly = 0;
5075 int miss_ndevs = 0;
5076 int i;
5077
5078 read_lock(&map_tree->map_tree.lock);
5079 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
5080 read_unlock(&map_tree->map_tree.lock);
5081 if (!em)
5082 return 1;
5083
5084 map = em->map_lookup;
5085 for (i = 0; i < map->num_stripes; i++) {
5086 if (map->stripes[i].dev->missing) {
5087 miss_ndevs++;
5088 continue;
5089 }
5090
5091 if (!map->stripes[i].dev->writeable) {
5092 readonly = 1;
5093 goto end;
5094 }
5095 }
5096
5097
5098
5099
5100
5101
5102 if (miss_ndevs > btrfs_chunk_max_errors(map))
5103 readonly = 1;
5104end:
5105 free_extent_map(em);
5106 return readonly;
5107}
5108
5109void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
5110{
5111 extent_map_tree_init(&tree->map_tree);
5112}
5113
5114void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
5115{
5116 struct extent_map *em;
5117
5118 while (1) {
5119 write_lock(&tree->map_tree.lock);
5120 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
5121 if (em)
5122 remove_extent_mapping(&tree->map_tree, em);
5123 write_unlock(&tree->map_tree.lock);
5124 if (!em)
5125 break;
5126
5127 free_extent_map(em);
5128
5129 free_extent_map(em);
5130 }
5131}
5132
5133int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5134{
5135 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5136 struct extent_map *em;
5137 struct map_lookup *map;
5138 struct extent_map_tree *em_tree = &map_tree->map_tree;
5139 int ret;
5140
5141 read_lock(&em_tree->lock);
5142 em = lookup_extent_mapping(em_tree, logical, len);
5143 read_unlock(&em_tree->lock);
5144
5145
5146
5147
5148
5149
5150 if (!em) {
5151 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
5152 logical+len);
5153 return 1;
5154 }
5155
5156 if (em->start > logical || em->start + em->len < logical) {
5157 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got %Lu-%Lu",
5158 logical, logical+len, em->start,
5159 em->start + em->len);
5160 free_extent_map(em);
5161 return 1;
5162 }
5163
5164 map = em->map_lookup;
5165 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
5166 ret = map->num_stripes;
5167 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5168 ret = map->sub_stripes;
5169 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5170 ret = 2;
5171 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5172 ret = 3;
5173 else
5174 ret = 1;
5175 free_extent_map(em);
5176
5177 btrfs_dev_replace_lock(&fs_info->dev_replace, 0);
5178 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
5179 ret++;
5180 btrfs_dev_replace_unlock(&fs_info->dev_replace, 0);
5181
5182 return ret;
5183}
5184
5185unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
5186 struct btrfs_mapping_tree *map_tree,
5187 u64 logical)
5188{
5189 struct extent_map *em;
5190 struct map_lookup *map;
5191 struct extent_map_tree *em_tree = &map_tree->map_tree;
5192 unsigned long len = root->sectorsize;
5193
5194 read_lock(&em_tree->lock);
5195 em = lookup_extent_mapping(em_tree, logical, len);
5196 read_unlock(&em_tree->lock);
5197 BUG_ON(!em);
5198
5199 BUG_ON(em->start > logical || em->start + em->len < logical);
5200 map = em->map_lookup;
5201 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5202 len = map->stripe_len * nr_data_stripes(map);
5203 free_extent_map(em);
5204 return len;
5205}
5206
5207int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
5208 u64 logical, u64 len, int mirror_num)
5209{
5210 struct extent_map *em;
5211 struct map_lookup *map;
5212 struct extent_map_tree *em_tree = &map_tree->map_tree;
5213 int ret = 0;
5214
5215 read_lock(&em_tree->lock);
5216 em = lookup_extent_mapping(em_tree, logical, len);
5217 read_unlock(&em_tree->lock);
5218 BUG_ON(!em);
5219
5220 BUG_ON(em->start > logical || em->start + em->len < logical);
5221 map = em->map_lookup;
5222 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5223 ret = 1;
5224 free_extent_map(em);
5225 return ret;
5226}
5227
5228static int find_live_mirror(struct btrfs_fs_info *fs_info,
5229 struct map_lookup *map, int first, int num,
5230 int optimal, int dev_replace_is_ongoing)
5231{
5232 int i;
5233 int tolerance;
5234 struct btrfs_device *srcdev;
5235
5236 if (dev_replace_is_ongoing &&
5237 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5238 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5239 srcdev = fs_info->dev_replace.srcdev;
5240 else
5241 srcdev = NULL;
5242
5243
5244
5245
5246
5247
5248 for (tolerance = 0; tolerance < 2; tolerance++) {
5249 if (map->stripes[optimal].dev->bdev &&
5250 (tolerance || map->stripes[optimal].dev != srcdev))
5251 return optimal;
5252 for (i = first; i < first + num; i++) {
5253 if (map->stripes[i].dev->bdev &&
5254 (tolerance || map->stripes[i].dev != srcdev))
5255 return i;
5256 }
5257 }
5258
5259
5260
5261
5262 return optimal;
5263}
5264
5265static inline int parity_smaller(u64 a, u64 b)
5266{
5267 return a > b;
5268}
5269
5270
5271static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5272{
5273 struct btrfs_bio_stripe s;
5274 int i;
5275 u64 l;
5276 int again = 1;
5277
5278 while (again) {
5279 again = 0;
5280 for (i = 0; i < num_stripes - 1; i++) {
5281 if (parity_smaller(bbio->raid_map[i],
5282 bbio->raid_map[i+1])) {
5283 s = bbio->stripes[i];
5284 l = bbio->raid_map[i];
5285 bbio->stripes[i] = bbio->stripes[i+1];
5286 bbio->raid_map[i] = bbio->raid_map[i+1];
5287 bbio->stripes[i+1] = s;
5288 bbio->raid_map[i+1] = l;
5289
5290 again = 1;
5291 }
5292 }
5293 }
5294}
5295
5296static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5297{
5298 struct btrfs_bio *bbio = kzalloc(
5299
5300 sizeof(struct btrfs_bio) +
5301
5302 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5303
5304 sizeof(int) * (real_stripes) +
5305
5306
5307
5308
5309 sizeof(u64) * (total_stripes),
5310 GFP_NOFS|__GFP_NOFAIL);
5311
5312 atomic_set(&bbio->error, 0);
5313 atomic_set(&bbio->refs, 1);
5314
5315 return bbio;
5316}
5317
5318void btrfs_get_bbio(struct btrfs_bio *bbio)
5319{
5320 WARN_ON(!atomic_read(&bbio->refs));
5321 atomic_inc(&bbio->refs);
5322}
5323
5324void btrfs_put_bbio(struct btrfs_bio *bbio)
5325{
5326 if (!bbio)
5327 return;
5328 if (atomic_dec_and_test(&bbio->refs))
5329 kfree(bbio);
5330}
5331
5332static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int op,
5333 u64 logical, u64 *length,
5334 struct btrfs_bio **bbio_ret,
5335 int mirror_num, int need_raid_map)
5336{
5337 struct extent_map *em;
5338 struct map_lookup *map;
5339 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5340 struct extent_map_tree *em_tree = &map_tree->map_tree;
5341 u64 offset;
5342 u64 stripe_offset;
5343 u64 stripe_end_offset;
5344 u64 stripe_nr;
5345 u64 stripe_nr_orig;
5346 u64 stripe_nr_end;
5347 u64 stripe_len;
5348 u32 stripe_index;
5349 int i;
5350 int ret = 0;
5351 int num_stripes;
5352 int max_errors = 0;
5353 int tgtdev_indexes = 0;
5354 struct btrfs_bio *bbio = NULL;
5355 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5356 int dev_replace_is_ongoing = 0;
5357 int num_alloc_stripes;
5358 int patch_the_first_stripe_for_dev_replace = 0;
5359 u64 physical_to_patch_in_first_stripe = 0;
5360 u64 raid56_full_stripe_start = (u64)-1;
5361
5362 read_lock(&em_tree->lock);
5363 em = lookup_extent_mapping(em_tree, logical, *length);
5364 read_unlock(&em_tree->lock);
5365
5366 if (!em) {
5367 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5368 logical, *length);
5369 return -EINVAL;
5370 }
5371
5372 if (em->start > logical || em->start + em->len < logical) {
5373 btrfs_crit(fs_info,
5374 "found a bad mapping, wanted %Lu, found %Lu-%Lu",
5375 logical, em->start, em->start + em->len);
5376 free_extent_map(em);
5377 return -EINVAL;
5378 }
5379
5380 map = em->map_lookup;
5381 offset = logical - em->start;
5382
5383 stripe_len = map->stripe_len;
5384 stripe_nr = offset;
5385
5386
5387
5388
5389 stripe_nr = div64_u64(stripe_nr, stripe_len);
5390
5391 stripe_offset = stripe_nr * stripe_len;
5392 if (offset < stripe_offset) {
5393 btrfs_crit(fs_info,
5394 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5395 stripe_offset, offset, em->start, logical,
5396 stripe_len);
5397 free_extent_map(em);
5398 return -EINVAL;
5399 }
5400
5401
5402 stripe_offset = offset - stripe_offset;
5403
5404
5405 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5406 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5407 raid56_full_stripe_start = offset;
5408
5409
5410
5411
5412 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5413 full_stripe_len);
5414 raid56_full_stripe_start *= full_stripe_len;
5415 }
5416
5417 if (op == REQ_OP_DISCARD) {
5418
5419 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5420 ret = -EOPNOTSUPP;
5421 goto out;
5422 }
5423 *length = min_t(u64, em->len - offset, *length);
5424 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5425 u64 max_len;
5426
5427
5428
5429 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5430 (op == REQ_OP_WRITE)) {
5431 max_len = stripe_len * nr_data_stripes(map) -
5432 (offset - raid56_full_stripe_start);
5433 } else {
5434
5435 max_len = stripe_len - stripe_offset;
5436 }
5437 *length = min_t(u64, em->len - offset, max_len);
5438 } else {
5439 *length = em->len - offset;
5440 }
5441
5442
5443
5444 if (!bbio_ret)
5445 goto out;
5446
5447 btrfs_dev_replace_lock(dev_replace, 0);
5448 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5449 if (!dev_replace_is_ongoing)
5450 btrfs_dev_replace_unlock(dev_replace, 0);
5451 else
5452 btrfs_dev_replace_set_lock_blocking(dev_replace);
5453
5454 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5455 op != REQ_OP_WRITE && op != REQ_OP_DISCARD &&
5456 op != REQ_GET_READ_MIRRORS && dev_replace->tgtdev != NULL) {
5457
5458
5459
5460
5461
5462
5463
5464
5465
5466
5467
5468
5469 u64 tmp_length = *length;
5470 struct btrfs_bio *tmp_bbio = NULL;
5471 int tmp_num_stripes;
5472 u64 srcdev_devid = dev_replace->srcdev->devid;
5473 int index_srcdev = 0;
5474 int found = 0;
5475 u64 physical_of_found = 0;
5476
5477 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5478 logical, &tmp_length, &tmp_bbio, 0, 0);
5479 if (ret) {
5480 WARN_ON(tmp_bbio != NULL);
5481 goto out;
5482 }
5483
5484 tmp_num_stripes = tmp_bbio->num_stripes;
5485 if (mirror_num > tmp_num_stripes) {
5486
5487
5488
5489
5490
5491 ret = -EIO;
5492 btrfs_put_bbio(tmp_bbio);
5493 goto out;
5494 }
5495
5496
5497
5498
5499
5500
5501
5502 for (i = 0; i < tmp_num_stripes; i++) {
5503 if (tmp_bbio->stripes[i].dev->devid != srcdev_devid)
5504 continue;
5505
5506
5507
5508
5509
5510 if (found &&
5511 physical_of_found <= tmp_bbio->stripes[i].physical)
5512 continue;
5513
5514 index_srcdev = i;
5515 found = 1;
5516 physical_of_found = tmp_bbio->stripes[i].physical;
5517 }
5518
5519 btrfs_put_bbio(tmp_bbio);
5520
5521 if (!found) {
5522 WARN_ON(1);
5523 ret = -EIO;
5524 goto out;
5525 }
5526
5527 mirror_num = index_srcdev + 1;
5528 patch_the_first_stripe_for_dev_replace = 1;
5529 physical_to_patch_in_first_stripe = physical_of_found;
5530 } else if (mirror_num > map->num_stripes) {
5531 mirror_num = 0;
5532 }
5533
5534 num_stripes = 1;
5535 stripe_index = 0;
5536 stripe_nr_orig = stripe_nr;
5537 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5538 stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5539 stripe_end_offset = stripe_nr_end * map->stripe_len -
5540 (offset + *length);
5541
5542 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5543 if (op == REQ_OP_DISCARD)
5544 num_stripes = min_t(u64, map->num_stripes,
5545 stripe_nr_end - stripe_nr_orig);
5546 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5547 &stripe_index);
5548 if (op != REQ_OP_WRITE && op != REQ_OP_DISCARD &&
5549 op != REQ_GET_READ_MIRRORS)
5550 mirror_num = 1;
5551 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5552 if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD ||
5553 op == REQ_GET_READ_MIRRORS)
5554 num_stripes = map->num_stripes;
5555 else if (mirror_num)
5556 stripe_index = mirror_num - 1;
5557 else {
5558 stripe_index = find_live_mirror(fs_info, map, 0,
5559 map->num_stripes,
5560 current->pid % map->num_stripes,
5561 dev_replace_is_ongoing);
5562 mirror_num = stripe_index + 1;
5563 }
5564
5565 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5566 if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD ||
5567 op == REQ_GET_READ_MIRRORS) {
5568 num_stripes = map->num_stripes;
5569 } else if (mirror_num) {
5570 stripe_index = mirror_num - 1;
5571 } else {
5572 mirror_num = 1;
5573 }
5574
5575 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5576 u32 factor = map->num_stripes / map->sub_stripes;
5577
5578 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5579 stripe_index *= map->sub_stripes;
5580
5581 if (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS)
5582 num_stripes = map->sub_stripes;
5583 else if (op == REQ_OP_DISCARD)
5584 num_stripes = min_t(u64, map->sub_stripes *
5585 (stripe_nr_end - stripe_nr_orig),
5586 map->num_stripes);
5587 else if (mirror_num)
5588 stripe_index += mirror_num - 1;
5589 else {
5590 int old_stripe_index = stripe_index;
5591 stripe_index = find_live_mirror(fs_info, map,
5592 stripe_index,
5593 map->sub_stripes, stripe_index +
5594 current->pid % map->sub_stripes,
5595 dev_replace_is_ongoing);
5596 mirror_num = stripe_index - old_stripe_index + 1;
5597 }
5598
5599 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5600 if (need_raid_map &&
5601 (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS ||
5602 mirror_num > 1)) {
5603
5604 stripe_nr = div_u64(raid56_full_stripe_start,
5605 stripe_len * nr_data_stripes(map));
5606
5607
5608 num_stripes = map->num_stripes;
5609 max_errors = nr_parity_stripes(map);
5610
5611 *length = map->stripe_len;
5612 stripe_index = 0;
5613 stripe_offset = 0;
5614 } else {
5615
5616
5617
5618
5619
5620 stripe_nr = div_u64_rem(stripe_nr,
5621 nr_data_stripes(map), &stripe_index);
5622 if (mirror_num > 1)
5623 stripe_index = nr_data_stripes(map) +
5624 mirror_num - 2;
5625
5626
5627 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5628 &stripe_index);
5629 if ((op != REQ_OP_WRITE && op != REQ_OP_DISCARD &&
5630 op != REQ_GET_READ_MIRRORS) && mirror_num <= 1)
5631 mirror_num = 1;
5632 }
5633 } else {
5634
5635
5636
5637
5638
5639 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5640 &stripe_index);
5641 mirror_num = stripe_index + 1;
5642 }
5643 if (stripe_index >= map->num_stripes) {
5644 btrfs_crit(fs_info,
5645 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5646 stripe_index, map->num_stripes);
5647 ret = -EINVAL;
5648 goto out;
5649 }
5650
5651 num_alloc_stripes = num_stripes;
5652 if (dev_replace_is_ongoing) {
5653 if (op == REQ_OP_WRITE || op == REQ_OP_DISCARD)
5654 num_alloc_stripes <<= 1;
5655 if (op == REQ_GET_READ_MIRRORS)
5656 num_alloc_stripes++;
5657 tgtdev_indexes = num_stripes;
5658 }
5659
5660 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5661 if (!bbio) {
5662 ret = -ENOMEM;
5663 goto out;
5664 }
5665 if (dev_replace_is_ongoing)
5666 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5667
5668
5669 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5670 need_raid_map &&
5671 ((op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS) ||
5672 mirror_num > 1)) {
5673 u64 tmp;
5674 unsigned rot;
5675
5676 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5677 sizeof(struct btrfs_bio_stripe) *
5678 num_alloc_stripes +
5679 sizeof(int) * tgtdev_indexes);
5680
5681
5682 div_u64_rem(stripe_nr, num_stripes, &rot);
5683
5684
5685 tmp = stripe_nr * nr_data_stripes(map);
5686 for (i = 0; i < nr_data_stripes(map); i++)
5687 bbio->raid_map[(i+rot) % num_stripes] =
5688 em->start + (tmp + i) * map->stripe_len;
5689
5690 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5691 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5692 bbio->raid_map[(i+rot+1) % num_stripes] =
5693 RAID6_Q_STRIPE;
5694 }
5695
5696 if (op == REQ_OP_DISCARD) {
5697 u32 factor = 0;
5698 u32 sub_stripes = 0;
5699 u64 stripes_per_dev = 0;
5700 u32 remaining_stripes = 0;
5701 u32 last_stripe = 0;
5702
5703 if (map->type &
5704 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5705 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5706 sub_stripes = 1;
5707 else
5708 sub_stripes = map->sub_stripes;
5709
5710 factor = map->num_stripes / sub_stripes;
5711 stripes_per_dev = div_u64_rem(stripe_nr_end -
5712 stripe_nr_orig,
5713 factor,
5714 &remaining_stripes);
5715 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5716 last_stripe *= sub_stripes;
5717 }
5718
5719 for (i = 0; i < num_stripes; i++) {
5720 bbio->stripes[i].physical =
5721 map->stripes[stripe_index].physical +
5722 stripe_offset + stripe_nr * map->stripe_len;
5723 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5724
5725 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5726 BTRFS_BLOCK_GROUP_RAID10)) {
5727 bbio->stripes[i].length = stripes_per_dev *
5728 map->stripe_len;
5729
5730 if (i / sub_stripes < remaining_stripes)
5731 bbio->stripes[i].length +=
5732 map->stripe_len;
5733
5734
5735
5736
5737
5738
5739
5740
5741
5742 if (i < sub_stripes)
5743 bbio->stripes[i].length -=
5744 stripe_offset;
5745
5746 if (stripe_index >= last_stripe &&
5747 stripe_index <= (last_stripe +
5748 sub_stripes - 1))
5749 bbio->stripes[i].length -=
5750 stripe_end_offset;
5751
5752 if (i == sub_stripes - 1)
5753 stripe_offset = 0;
5754 } else
5755 bbio->stripes[i].length = *length;
5756
5757 stripe_index++;
5758 if (stripe_index == map->num_stripes) {
5759
5760 stripe_index = 0;
5761 stripe_nr++;
5762 }
5763 }
5764 } else {
5765 for (i = 0; i < num_stripes; i++) {
5766 bbio->stripes[i].physical =
5767 map->stripes[stripe_index].physical +
5768 stripe_offset +
5769 stripe_nr * map->stripe_len;
5770 bbio->stripes[i].dev =
5771 map->stripes[stripe_index].dev;
5772 stripe_index++;
5773 }
5774 }
5775
5776 if (op == REQ_OP_WRITE || op == REQ_GET_READ_MIRRORS)
5777 max_errors = btrfs_chunk_max_errors(map);
5778
5779 if (bbio->raid_map)
5780 sort_parity_stripes(bbio, num_stripes);
5781
5782 tgtdev_indexes = 0;
5783 if (dev_replace_is_ongoing &&
5784 (op == REQ_OP_WRITE || op == REQ_OP_DISCARD) &&
5785 dev_replace->tgtdev != NULL) {
5786 int index_where_to_add;
5787 u64 srcdev_devid = dev_replace->srcdev->devid;
5788
5789
5790
5791
5792
5793
5794
5795
5796
5797
5798
5799
5800 index_where_to_add = num_stripes;
5801 for (i = 0; i < num_stripes; i++) {
5802 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5803
5804 struct btrfs_bio_stripe *new =
5805 bbio->stripes + index_where_to_add;
5806 struct btrfs_bio_stripe *old =
5807 bbio->stripes + i;
5808
5809 new->physical = old->physical;
5810 new->length = old->length;
5811 new->dev = dev_replace->tgtdev;
5812 bbio->tgtdev_map[i] = index_where_to_add;
5813 index_where_to_add++;
5814 max_errors++;
5815 tgtdev_indexes++;
5816 }
5817 }
5818 num_stripes = index_where_to_add;
5819 } else if (dev_replace_is_ongoing && (op == REQ_GET_READ_MIRRORS) &&
5820 dev_replace->tgtdev != NULL) {
5821 u64 srcdev_devid = dev_replace->srcdev->devid;
5822 int index_srcdev = 0;
5823 int found = 0;
5824 u64 physical_of_found = 0;
5825
5826
5827
5828
5829
5830
5831
5832
5833 for (i = 0; i < num_stripes; i++) {
5834 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5835
5836
5837
5838
5839
5840 if (found &&
5841 physical_of_found <=
5842 bbio->stripes[i].physical)
5843 continue;
5844 index_srcdev = i;
5845 found = 1;
5846 physical_of_found = bbio->stripes[i].physical;
5847 }
5848 }
5849 if (found) {
5850 struct btrfs_bio_stripe *tgtdev_stripe =
5851 bbio->stripes + num_stripes;
5852
5853 tgtdev_stripe->physical = physical_of_found;
5854 tgtdev_stripe->length =
5855 bbio->stripes[index_srcdev].length;
5856 tgtdev_stripe->dev = dev_replace->tgtdev;
5857 bbio->tgtdev_map[index_srcdev] = num_stripes;
5858
5859 tgtdev_indexes++;
5860 num_stripes++;
5861 }
5862 }
5863
5864 *bbio_ret = bbio;
5865 bbio->map_type = map->type;
5866 bbio->num_stripes = num_stripes;
5867 bbio->max_errors = max_errors;
5868 bbio->mirror_num = mirror_num;
5869 bbio->num_tgtdevs = tgtdev_indexes;
5870
5871
5872
5873
5874
5875
5876 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5877 WARN_ON(num_stripes > 1);
5878 bbio->stripes[0].dev = dev_replace->tgtdev;
5879 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5880 bbio->mirror_num = map->num_stripes + 1;
5881 }
5882out:
5883 if (dev_replace_is_ongoing) {
5884 btrfs_dev_replace_clear_lock_blocking(dev_replace);
5885 btrfs_dev_replace_unlock(dev_replace, 0);
5886 }
5887 free_extent_map(em);
5888 return ret;
5889}
5890
5891int btrfs_map_block(struct btrfs_fs_info *fs_info, int op,
5892 u64 logical, u64 *length,
5893 struct btrfs_bio **bbio_ret, int mirror_num)
5894{
5895 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5896 mirror_num, 0);
5897}
5898
5899
5900int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int op,
5901 u64 logical, u64 *length,
5902 struct btrfs_bio **bbio_ret, int mirror_num,
5903 int need_raid_map)
5904{
5905 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
5906 mirror_num, need_raid_map);
5907}
5908
5909int btrfs_rmap_block(struct btrfs_fs_info *fs_info,
5910 u64 chunk_start, u64 physical, u64 devid,
5911 u64 **logical, int *naddrs, int *stripe_len)
5912{
5913 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5914 struct extent_map_tree *em_tree = &map_tree->map_tree;
5915 struct extent_map *em;
5916 struct map_lookup *map;
5917 u64 *buf;
5918 u64 bytenr;
5919 u64 length;
5920 u64 stripe_nr;
5921 u64 rmap_len;
5922 int i, j, nr = 0;
5923
5924 read_lock(&em_tree->lock);
5925 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5926 read_unlock(&em_tree->lock);
5927
5928 if (!em) {
5929 btrfs_err(fs_info, "couldn't find em for chunk %Lu",
5930 chunk_start);
5931 return -EIO;
5932 }
5933
5934 if (em->start != chunk_start) {
5935 btrfs_err(fs_info, "bad chunk start, em=%Lu, wanted=%Lu",
5936 em->start, chunk_start);
5937 free_extent_map(em);
5938 return -EIO;
5939 }
5940 map = em->map_lookup;
5941
5942 length = em->len;
5943 rmap_len = map->stripe_len;
5944
5945 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5946 length = div_u64(length, map->num_stripes / map->sub_stripes);
5947 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5948 length = div_u64(length, map->num_stripes);
5949 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5950 length = div_u64(length, nr_data_stripes(map));
5951 rmap_len = map->stripe_len * nr_data_stripes(map);
5952 }
5953
5954 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5955 BUG_ON(!buf);
5956
5957 for (i = 0; i < map->num_stripes; i++) {
5958 if (devid && map->stripes[i].dev->devid != devid)
5959 continue;
5960 if (map->stripes[i].physical > physical ||
5961 map->stripes[i].physical + length <= physical)
5962 continue;
5963
5964 stripe_nr = physical - map->stripes[i].physical;
5965 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5966
5967 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5968 stripe_nr = stripe_nr * map->num_stripes + i;
5969 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5970 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5971 stripe_nr = stripe_nr * map->num_stripes + i;
5972 }
5973
5974
5975
5976 bytenr = chunk_start + stripe_nr * rmap_len;
5977 WARN_ON(nr >= map->num_stripes);
5978 for (j = 0; j < nr; j++) {
5979 if (buf[j] == bytenr)
5980 break;
5981 }
5982 if (j == nr) {
5983 WARN_ON(nr >= map->num_stripes);
5984 buf[nr++] = bytenr;
5985 }
5986 }
5987
5988 *logical = buf;
5989 *naddrs = nr;
5990 *stripe_len = rmap_len;
5991
5992 free_extent_map(em);
5993 return 0;
5994}
5995
5996static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
5997{
5998 bio->bi_private = bbio->private;
5999 bio->bi_end_io = bbio->end_io;
6000 bio_endio(bio);
6001
6002 btrfs_put_bbio(bbio);
6003}
6004
6005static void btrfs_end_bio(struct bio *bio)
6006{
6007 struct btrfs_bio *bbio = bio->bi_private;
6008 int is_orig_bio = 0;
6009
6010 if (bio->bi_error) {
6011 atomic_inc(&bbio->error);
6012 if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) {
6013 unsigned int stripe_index =
6014 btrfs_io_bio(bio)->stripe_index;
6015 struct btrfs_device *dev;
6016
6017 BUG_ON(stripe_index >= bbio->num_stripes);
6018 dev = bbio->stripes[stripe_index].dev;
6019 if (dev->bdev) {
6020 if (bio_op(bio) == REQ_OP_WRITE)
6021 btrfs_dev_stat_inc(dev,
6022 BTRFS_DEV_STAT_WRITE_ERRS);
6023 else
6024 btrfs_dev_stat_inc(dev,
6025 BTRFS_DEV_STAT_READ_ERRS);
6026 if ((bio->bi_opf & WRITE_FLUSH) == WRITE_FLUSH)
6027 btrfs_dev_stat_inc(dev,
6028 BTRFS_DEV_STAT_FLUSH_ERRS);
6029 btrfs_dev_stat_print_on_error(dev);
6030 }
6031 }
6032 }
6033
6034 if (bio == bbio->orig_bio)
6035 is_orig_bio = 1;
6036
6037 btrfs_bio_counter_dec(bbio->fs_info);
6038
6039 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6040 if (!is_orig_bio) {
6041 bio_put(bio);
6042 bio = bbio->orig_bio;
6043 }
6044
6045 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6046
6047
6048
6049 if (atomic_read(&bbio->error) > bbio->max_errors) {
6050 bio->bi_error = -EIO;
6051 } else {
6052
6053
6054
6055
6056 bio->bi_error = 0;
6057 }
6058
6059 btrfs_end_bbio(bbio, bio);
6060 } else if (!is_orig_bio) {
6061 bio_put(bio);
6062 }
6063}
6064
6065
6066
6067
6068
6069
6070
6071
6072static noinline void btrfs_schedule_bio(struct btrfs_root *root,
6073 struct btrfs_device *device,
6074 struct bio *bio)
6075{
6076 int should_queue = 1;
6077 struct btrfs_pending_bios *pending_bios;
6078
6079 if (device->missing || !device->bdev) {
6080 bio_io_error(bio);
6081 return;
6082 }
6083
6084
6085 if (bio_op(bio) == REQ_OP_READ) {
6086 bio_get(bio);
6087 btrfsic_submit_bio(bio);
6088 bio_put(bio);
6089 return;
6090 }
6091
6092
6093
6094
6095
6096
6097
6098 atomic_inc(&root->fs_info->nr_async_bios);
6099 WARN_ON(bio->bi_next);
6100 bio->bi_next = NULL;
6101
6102 spin_lock(&device->io_lock);
6103 if (bio->bi_opf & REQ_SYNC)
6104 pending_bios = &device->pending_sync_bios;
6105 else
6106 pending_bios = &device->pending_bios;
6107
6108 if (pending_bios->tail)
6109 pending_bios->tail->bi_next = bio;
6110
6111 pending_bios->tail = bio;
6112 if (!pending_bios->head)
6113 pending_bios->head = bio;
6114 if (device->running_pending)
6115 should_queue = 0;
6116
6117 spin_unlock(&device->io_lock);
6118
6119 if (should_queue)
6120 btrfs_queue_work(root->fs_info->submit_workers,
6121 &device->work);
6122}
6123
6124static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
6125 struct bio *bio, u64 physical, int dev_nr,
6126 int async)
6127{
6128 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
6129
6130 bio->bi_private = bbio;
6131 btrfs_io_bio(bio)->stripe_index = dev_nr;
6132 bio->bi_end_io = btrfs_end_bio;
6133 bio->bi_iter.bi_sector = physical >> 9;
6134#ifdef DEBUG
6135 {
6136 struct rcu_string *name;
6137
6138 rcu_read_lock();
6139 name = rcu_dereference(dev->name);
6140 btrfs_debug(fs_info,
6141 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6142 bio_op(bio), bio->bi_opf,
6143 (u64)bio->bi_iter.bi_sector,
6144 (u_long)dev->bdev->bd_dev, name->str, dev->devid,
6145 bio->bi_iter.bi_size);
6146 rcu_read_unlock();
6147 }
6148#endif
6149 bio->bi_bdev = dev->bdev;
6150
6151 btrfs_bio_counter_inc_noblocked(root->fs_info);
6152
6153 if (async)
6154 btrfs_schedule_bio(root, dev, bio);
6155 else
6156 btrfsic_submit_bio(bio);
6157}
6158
6159static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6160{
6161 atomic_inc(&bbio->error);
6162 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6163
6164 WARN_ON(bio != bbio->orig_bio);
6165
6166 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6167 bio->bi_iter.bi_sector = logical >> 9;
6168 bio->bi_error = -EIO;
6169 btrfs_end_bbio(bbio, bio);
6170 }
6171}
6172
6173int btrfs_map_bio(struct btrfs_root *root, struct bio *bio,
6174 int mirror_num, int async_submit)
6175{
6176 struct btrfs_device *dev;
6177 struct bio *first_bio = bio;
6178 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6179 u64 length = 0;
6180 u64 map_length;
6181 int ret;
6182 int dev_nr;
6183 int total_devs;
6184 struct btrfs_bio *bbio = NULL;
6185
6186 length = bio->bi_iter.bi_size;
6187 map_length = length;
6188
6189 btrfs_bio_counter_inc_blocked(root->fs_info);
6190 ret = __btrfs_map_block(root->fs_info, bio_op(bio), logical,
6191 &map_length, &bbio, mirror_num, 1);
6192 if (ret) {
6193 btrfs_bio_counter_dec(root->fs_info);
6194 return ret;
6195 }
6196
6197 total_devs = bbio->num_stripes;
6198 bbio->orig_bio = first_bio;
6199 bbio->private = first_bio->bi_private;
6200 bbio->end_io = first_bio->bi_end_io;
6201 bbio->fs_info = root->fs_info;
6202 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6203
6204 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6205 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6206
6207
6208 if (bio_op(bio) == REQ_OP_WRITE) {
6209 ret = raid56_parity_write(root, bio, bbio, map_length);
6210 } else {
6211 ret = raid56_parity_recover(root, bio, bbio, map_length,
6212 mirror_num, 1);
6213 }
6214
6215 btrfs_bio_counter_dec(root->fs_info);
6216 return ret;
6217 }
6218
6219 if (map_length < length) {
6220 btrfs_crit(root->fs_info,
6221 "mapping failed logical %llu bio len %llu len %llu",
6222 logical, length, map_length);
6223 BUG();
6224 }
6225
6226 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6227 dev = bbio->stripes[dev_nr].dev;
6228 if (!dev || !dev->bdev ||
6229 (bio_op(bio) == REQ_OP_WRITE && !dev->writeable)) {
6230 bbio_error(bbio, first_bio, logical);
6231 continue;
6232 }
6233
6234 if (dev_nr < total_devs - 1) {
6235 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6236 BUG_ON(!bio);
6237 } else
6238 bio = first_bio;
6239
6240 submit_stripe_bio(root, bbio, bio,
6241 bbio->stripes[dev_nr].physical, dev_nr,
6242 async_submit);
6243 }
6244 btrfs_bio_counter_dec(root->fs_info);
6245 return 0;
6246}
6247
6248struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6249 u8 *uuid, u8 *fsid)
6250{
6251 struct btrfs_device *device;
6252 struct btrfs_fs_devices *cur_devices;
6253
6254 cur_devices = fs_info->fs_devices;
6255 while (cur_devices) {
6256 if (!fsid ||
6257 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6258 device = __find_device(&cur_devices->devices,
6259 devid, uuid);
6260 if (device)
6261 return device;
6262 }
6263 cur_devices = cur_devices->seed;
6264 }
6265 return NULL;
6266}
6267
6268static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6269 struct btrfs_fs_devices *fs_devices,
6270 u64 devid, u8 *dev_uuid)
6271{
6272 struct btrfs_device *device;
6273
6274 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6275 if (IS_ERR(device))
6276 return NULL;
6277
6278 list_add(&device->dev_list, &fs_devices->devices);
6279 device->fs_devices = fs_devices;
6280 fs_devices->num_devices++;
6281
6282 device->missing = 1;
6283 fs_devices->missing_devices++;
6284
6285 return device;
6286}
6287
6288
6289
6290
6291
6292
6293
6294
6295
6296
6297
6298
6299
6300
6301struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6302 const u64 *devid,
6303 const u8 *uuid)
6304{
6305 struct btrfs_device *dev;
6306 u64 tmp;
6307
6308 if (WARN_ON(!devid && !fs_info))
6309 return ERR_PTR(-EINVAL);
6310
6311 dev = __alloc_device();
6312 if (IS_ERR(dev))
6313 return dev;
6314
6315 if (devid)
6316 tmp = *devid;
6317 else {
6318 int ret;
6319
6320 ret = find_next_devid(fs_info, &tmp);
6321 if (ret) {
6322 kfree(dev);
6323 return ERR_PTR(ret);
6324 }
6325 }
6326 dev->devid = tmp;
6327
6328 if (uuid)
6329 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6330 else
6331 generate_random_uuid(dev->uuid);
6332
6333 btrfs_init_work(&dev->work, btrfs_submit_helper,
6334 pending_bios_fn, NULL, NULL);
6335
6336 return dev;
6337}
6338
6339
6340static int btrfs_check_chunk_valid(struct btrfs_root *root,
6341 struct extent_buffer *leaf,
6342 struct btrfs_chunk *chunk, u64 logical)
6343{
6344 u64 length;
6345 u64 stripe_len;
6346 u16 num_stripes;
6347 u16 sub_stripes;
6348 u64 type;
6349
6350 length = btrfs_chunk_length(leaf, chunk);
6351 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6352 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6353 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6354 type = btrfs_chunk_type(leaf, chunk);
6355
6356 if (!num_stripes) {
6357 btrfs_err(root->fs_info, "invalid chunk num_stripes: %u",
6358 num_stripes);
6359 return -EIO;
6360 }
6361 if (!IS_ALIGNED(logical, root->sectorsize)) {
6362 btrfs_err(root->fs_info,
6363 "invalid chunk logical %llu", logical);
6364 return -EIO;
6365 }
6366 if (btrfs_chunk_sector_size(leaf, chunk) != root->sectorsize) {
6367 btrfs_err(root->fs_info, "invalid chunk sectorsize %u",
6368 btrfs_chunk_sector_size(leaf, chunk));
6369 return -EIO;
6370 }
6371 if (!length || !IS_ALIGNED(length, root->sectorsize)) {
6372 btrfs_err(root->fs_info,
6373 "invalid chunk length %llu", length);
6374 return -EIO;
6375 }
6376 if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) {
6377 btrfs_err(root->fs_info, "invalid chunk stripe length: %llu",
6378 stripe_len);
6379 return -EIO;
6380 }
6381 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6382 type) {
6383 btrfs_err(root->fs_info, "unrecognized chunk type: %llu",
6384 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
6385 BTRFS_BLOCK_GROUP_PROFILE_MASK) &
6386 btrfs_chunk_type(leaf, chunk));
6387 return -EIO;
6388 }
6389 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) ||
6390 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
6391 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
6392 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
6393 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
6394 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
6395 num_stripes != 1)) {
6396 btrfs_err(root->fs_info,
6397 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6398 num_stripes, sub_stripes,
6399 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
6400 return -EIO;
6401 }
6402
6403 return 0;
6404}
6405
6406static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6407 struct extent_buffer *leaf,
6408 struct btrfs_chunk *chunk)
6409{
6410 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6411 struct map_lookup *map;
6412 struct extent_map *em;
6413 u64 logical;
6414 u64 length;
6415 u64 stripe_len;
6416 u64 devid;
6417 u8 uuid[BTRFS_UUID_SIZE];
6418 int num_stripes;
6419 int ret;
6420 int i;
6421
6422 logical = key->offset;
6423 length = btrfs_chunk_length(leaf, chunk);
6424 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6425 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6426
6427 ret = btrfs_check_chunk_valid(root, leaf, chunk, logical);
6428 if (ret)
6429 return ret;
6430
6431 read_lock(&map_tree->map_tree.lock);
6432 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6433 read_unlock(&map_tree->map_tree.lock);
6434
6435
6436 if (em && em->start <= logical && em->start + em->len > logical) {
6437 free_extent_map(em);
6438 return 0;
6439 } else if (em) {
6440 free_extent_map(em);
6441 }
6442
6443 em = alloc_extent_map();
6444 if (!em)
6445 return -ENOMEM;
6446 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6447 if (!map) {
6448 free_extent_map(em);
6449 return -ENOMEM;
6450 }
6451
6452 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6453 em->map_lookup = map;
6454 em->start = logical;
6455 em->len = length;
6456 em->orig_start = 0;
6457 em->block_start = 0;
6458 em->block_len = em->len;
6459
6460 map->num_stripes = num_stripes;
6461 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6462 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6463 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6464 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6465 map->type = btrfs_chunk_type(leaf, chunk);
6466 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6467 for (i = 0; i < num_stripes; i++) {
6468 map->stripes[i].physical =
6469 btrfs_stripe_offset_nr(leaf, chunk, i);
6470 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6471 read_extent_buffer(leaf, uuid, (unsigned long)
6472 btrfs_stripe_dev_uuid_nr(chunk, i),
6473 BTRFS_UUID_SIZE);
6474 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6475 uuid, NULL);
6476 if (!map->stripes[i].dev &&
6477 !btrfs_test_opt(root->fs_info, DEGRADED)) {
6478 free_extent_map(em);
6479 return -EIO;
6480 }
6481 if (!map->stripes[i].dev) {
6482 map->stripes[i].dev =
6483 add_missing_dev(root, root->fs_info->fs_devices,
6484 devid, uuid);
6485 if (!map->stripes[i].dev) {
6486 free_extent_map(em);
6487 return -EIO;
6488 }
6489 btrfs_warn(root->fs_info,
6490 "devid %llu uuid %pU is missing",
6491 devid, uuid);
6492 }
6493 map->stripes[i].dev->in_fs_metadata = 1;
6494 }
6495
6496 write_lock(&map_tree->map_tree.lock);
6497 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6498 write_unlock(&map_tree->map_tree.lock);
6499 BUG_ON(ret);
6500 free_extent_map(em);
6501
6502 return 0;
6503}
6504
6505static void fill_device_from_item(struct extent_buffer *leaf,
6506 struct btrfs_dev_item *dev_item,
6507 struct btrfs_device *device)
6508{
6509 unsigned long ptr;
6510
6511 device->devid = btrfs_device_id(leaf, dev_item);
6512 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6513 device->total_bytes = device->disk_total_bytes;
6514 device->commit_total_bytes = device->disk_total_bytes;
6515 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6516 device->commit_bytes_used = device->bytes_used;
6517 device->type = btrfs_device_type(leaf, dev_item);
6518 device->io_align = btrfs_device_io_align(leaf, dev_item);
6519 device->io_width = btrfs_device_io_width(leaf, dev_item);
6520 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6521 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6522 device->is_tgtdev_for_dev_replace = 0;
6523
6524 ptr = btrfs_device_uuid(dev_item);
6525 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6526}
6527
6528static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6529 u8 *fsid)
6530{
6531 struct btrfs_fs_devices *fs_devices;
6532 int ret;
6533
6534 BUG_ON(!mutex_is_locked(&uuid_mutex));
6535
6536 fs_devices = root->fs_info->fs_devices->seed;
6537 while (fs_devices) {
6538 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6539 return fs_devices;
6540
6541 fs_devices = fs_devices->seed;
6542 }
6543
6544 fs_devices = find_fsid(fsid);
6545 if (!fs_devices) {
6546 if (!btrfs_test_opt(root->fs_info, DEGRADED))
6547 return ERR_PTR(-ENOENT);
6548
6549 fs_devices = alloc_fs_devices(fsid);
6550 if (IS_ERR(fs_devices))
6551 return fs_devices;
6552
6553 fs_devices->seeding = 1;
6554 fs_devices->opened = 1;
6555 return fs_devices;
6556 }
6557
6558 fs_devices = clone_fs_devices(fs_devices);
6559 if (IS_ERR(fs_devices))
6560 return fs_devices;
6561
6562 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6563 root->fs_info->bdev_holder);
6564 if (ret) {
6565 free_fs_devices(fs_devices);
6566 fs_devices = ERR_PTR(ret);
6567 goto out;
6568 }
6569
6570 if (!fs_devices->seeding) {
6571 __btrfs_close_devices(fs_devices);
6572 free_fs_devices(fs_devices);
6573 fs_devices = ERR_PTR(-EINVAL);
6574 goto out;
6575 }
6576
6577 fs_devices->seed = root->fs_info->fs_devices->seed;
6578 root->fs_info->fs_devices->seed = fs_devices;
6579out:
6580 return fs_devices;
6581}
6582
6583static int read_one_dev(struct btrfs_root *root,
6584 struct extent_buffer *leaf,
6585 struct btrfs_dev_item *dev_item)
6586{
6587 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6588 struct btrfs_device *device;
6589 u64 devid;
6590 int ret;
6591 u8 fs_uuid[BTRFS_UUID_SIZE];
6592 u8 dev_uuid[BTRFS_UUID_SIZE];
6593
6594 devid = btrfs_device_id(leaf, dev_item);
6595 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6596 BTRFS_UUID_SIZE);
6597 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6598 BTRFS_UUID_SIZE);
6599
6600 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6601 fs_devices = open_seed_devices(root, fs_uuid);
6602 if (IS_ERR(fs_devices))
6603 return PTR_ERR(fs_devices);
6604 }
6605
6606 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6607 if (!device) {
6608 if (!btrfs_test_opt(root->fs_info, DEGRADED))
6609 return -EIO;
6610
6611 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6612 if (!device)
6613 return -ENOMEM;
6614 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6615 devid, dev_uuid);
6616 } else {
6617 if (!device->bdev && !btrfs_test_opt(root->fs_info, DEGRADED))
6618 return -EIO;
6619
6620 if(!device->bdev && !device->missing) {
6621
6622
6623
6624
6625
6626
6627 device->fs_devices->missing_devices++;
6628 device->missing = 1;
6629 }
6630
6631
6632 if (device->fs_devices != fs_devices) {
6633 ASSERT(device->missing);
6634
6635 list_move(&device->dev_list, &fs_devices->devices);
6636 device->fs_devices->num_devices--;
6637 fs_devices->num_devices++;
6638
6639 device->fs_devices->missing_devices--;
6640 fs_devices->missing_devices++;
6641
6642 device->fs_devices = fs_devices;
6643 }
6644 }
6645
6646 if (device->fs_devices != root->fs_info->fs_devices) {
6647 BUG_ON(device->writeable);
6648 if (device->generation !=
6649 btrfs_device_generation(leaf, dev_item))
6650 return -EINVAL;
6651 }
6652
6653 fill_device_from_item(leaf, dev_item, device);
6654 device->in_fs_metadata = 1;
6655 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6656 device->fs_devices->total_rw_bytes += device->total_bytes;
6657 spin_lock(&root->fs_info->free_chunk_lock);
6658 root->fs_info->free_chunk_space += device->total_bytes -
6659 device->bytes_used;
6660 spin_unlock(&root->fs_info->free_chunk_lock);
6661 }
6662 ret = 0;
6663 return ret;
6664}
6665
6666int btrfs_read_sys_array(struct btrfs_root *root)
6667{
6668 struct btrfs_fs_info *fs_info = root->fs_info;
6669 struct btrfs_super_block *super_copy = fs_info->super_copy;
6670 struct extent_buffer *sb;
6671 struct btrfs_disk_key *disk_key;
6672 struct btrfs_chunk *chunk;
6673 u8 *array_ptr;
6674 unsigned long sb_array_offset;
6675 int ret = 0;
6676 u32 num_stripes;
6677 u32 array_size;
6678 u32 len = 0;
6679 u32 cur_offset;
6680 u64 type;
6681 struct btrfs_key key;
6682
6683 ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6684
6685
6686
6687
6688
6689 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6690 if (IS_ERR(sb))
6691 return PTR_ERR(sb);
6692 set_extent_buffer_uptodate(sb);
6693 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6694
6695
6696
6697
6698
6699
6700
6701
6702
6703
6704
6705
6706 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6707 SetPageUptodate(sb->pages[0]);
6708
6709 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6710 array_size = btrfs_super_sys_array_size(super_copy);
6711
6712 array_ptr = super_copy->sys_chunk_array;
6713 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6714 cur_offset = 0;
6715
6716 while (cur_offset < array_size) {
6717 disk_key = (struct btrfs_disk_key *)array_ptr;
6718 len = sizeof(*disk_key);
6719 if (cur_offset + len > array_size)
6720 goto out_short_read;
6721
6722 btrfs_disk_key_to_cpu(&key, disk_key);
6723
6724 array_ptr += len;
6725 sb_array_offset += len;
6726 cur_offset += len;
6727
6728 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6729 chunk = (struct btrfs_chunk *)sb_array_offset;
6730
6731
6732
6733
6734 len = btrfs_chunk_item_size(1);
6735 if (cur_offset + len > array_size)
6736 goto out_short_read;
6737
6738 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6739 if (!num_stripes) {
6740 btrfs_err(fs_info,
6741 "invalid number of stripes %u in sys_array at offset %u",
6742 num_stripes, cur_offset);
6743 ret = -EIO;
6744 break;
6745 }
6746
6747 type = btrfs_chunk_type(sb, chunk);
6748 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6749 btrfs_err(fs_info,
6750 "invalid chunk type %llu in sys_array at offset %u",
6751 type, cur_offset);
6752 ret = -EIO;
6753 break;
6754 }
6755
6756 len = btrfs_chunk_item_size(num_stripes);
6757 if (cur_offset + len > array_size)
6758 goto out_short_read;
6759
6760 ret = read_one_chunk(root, &key, sb, chunk);
6761 if (ret)
6762 break;
6763 } else {
6764 btrfs_err(fs_info,
6765 "unexpected item type %u in sys_array at offset %u",
6766 (u32)key.type, cur_offset);
6767 ret = -EIO;
6768 break;
6769 }
6770 array_ptr += len;
6771 sb_array_offset += len;
6772 cur_offset += len;
6773 }
6774 clear_extent_buffer_uptodate(sb);
6775 free_extent_buffer_stale(sb);
6776 return ret;
6777
6778out_short_read:
6779 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6780 len, cur_offset);
6781 clear_extent_buffer_uptodate(sb);
6782 free_extent_buffer_stale(sb);
6783 return -EIO;
6784}
6785
6786int btrfs_read_chunk_tree(struct btrfs_root *root)
6787{
6788 struct btrfs_path *path;
6789 struct extent_buffer *leaf;
6790 struct btrfs_key key;
6791 struct btrfs_key found_key;
6792 int ret;
6793 int slot;
6794 u64 total_dev = 0;
6795
6796 root = root->fs_info->chunk_root;
6797
6798 path = btrfs_alloc_path();
6799 if (!path)
6800 return -ENOMEM;
6801
6802 mutex_lock(&uuid_mutex);
6803 lock_chunks(root);
6804
6805
6806
6807
6808
6809
6810
6811 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6812 key.offset = 0;
6813 key.type = 0;
6814 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6815 if (ret < 0)
6816 goto error;
6817 while (1) {
6818 leaf = path->nodes[0];
6819 slot = path->slots[0];
6820 if (slot >= btrfs_header_nritems(leaf)) {
6821 ret = btrfs_next_leaf(root, path);
6822 if (ret == 0)
6823 continue;
6824 if (ret < 0)
6825 goto error;
6826 break;
6827 }
6828 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6829 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6830 struct btrfs_dev_item *dev_item;
6831 dev_item = btrfs_item_ptr(leaf, slot,
6832 struct btrfs_dev_item);
6833 ret = read_one_dev(root, leaf, dev_item);
6834 if (ret)
6835 goto error;
6836 total_dev++;
6837 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6838 struct btrfs_chunk *chunk;
6839 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6840 ret = read_one_chunk(root, &found_key, leaf, chunk);
6841 if (ret)
6842 goto error;
6843 }
6844 path->slots[0]++;
6845 }
6846
6847
6848
6849
6850
6851 if (total_dev != root->fs_info->fs_devices->total_devices) {
6852 btrfs_err(root->fs_info,
6853 "super_num_devices %llu mismatch with num_devices %llu found here",
6854 btrfs_super_num_devices(root->fs_info->super_copy),
6855 total_dev);
6856 ret = -EINVAL;
6857 goto error;
6858 }
6859 if (btrfs_super_total_bytes(root->fs_info->super_copy) <
6860 root->fs_info->fs_devices->total_rw_bytes) {
6861 btrfs_err(root->fs_info,
6862 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6863 btrfs_super_total_bytes(root->fs_info->super_copy),
6864 root->fs_info->fs_devices->total_rw_bytes);
6865 ret = -EINVAL;
6866 goto error;
6867 }
6868 ret = 0;
6869error:
6870 unlock_chunks(root);
6871 mutex_unlock(&uuid_mutex);
6872
6873 btrfs_free_path(path);
6874 return ret;
6875}
6876
6877void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6878{
6879 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6880 struct btrfs_device *device;
6881
6882 while (fs_devices) {
6883 mutex_lock(&fs_devices->device_list_mutex);
6884 list_for_each_entry(device, &fs_devices->devices, dev_list)
6885 device->dev_root = fs_info->dev_root;
6886 mutex_unlock(&fs_devices->device_list_mutex);
6887
6888 fs_devices = fs_devices->seed;
6889 }
6890}
6891
6892static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6893{
6894 int i;
6895
6896 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6897 btrfs_dev_stat_reset(dev, i);
6898}
6899
6900int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6901{
6902 struct btrfs_key key;
6903 struct btrfs_key found_key;
6904 struct btrfs_root *dev_root = fs_info->dev_root;
6905 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6906 struct extent_buffer *eb;
6907 int slot;
6908 int ret = 0;
6909 struct btrfs_device *device;
6910 struct btrfs_path *path = NULL;
6911 int i;
6912
6913 path = btrfs_alloc_path();
6914 if (!path) {
6915 ret = -ENOMEM;
6916 goto out;
6917 }
6918
6919 mutex_lock(&fs_devices->device_list_mutex);
6920 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6921 int item_size;
6922 struct btrfs_dev_stats_item *ptr;
6923
6924 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6925 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6926 key.offset = device->devid;
6927 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6928 if (ret) {
6929 __btrfs_reset_dev_stats(device);
6930 device->dev_stats_valid = 1;
6931 btrfs_release_path(path);
6932 continue;
6933 }
6934 slot = path->slots[0];
6935 eb = path->nodes[0];
6936 btrfs_item_key_to_cpu(eb, &found_key, slot);
6937 item_size = btrfs_item_size_nr(eb, slot);
6938
6939 ptr = btrfs_item_ptr(eb, slot,
6940 struct btrfs_dev_stats_item);
6941
6942 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6943 if (item_size >= (1 + i) * sizeof(__le64))
6944 btrfs_dev_stat_set(device, i,
6945 btrfs_dev_stats_value(eb, ptr, i));
6946 else
6947 btrfs_dev_stat_reset(device, i);
6948 }
6949
6950 device->dev_stats_valid = 1;
6951 btrfs_dev_stat_print_on_load(device);
6952 btrfs_release_path(path);
6953 }
6954 mutex_unlock(&fs_devices->device_list_mutex);
6955
6956out:
6957 btrfs_free_path(path);
6958 return ret < 0 ? ret : 0;
6959}
6960
6961static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6962 struct btrfs_root *dev_root,
6963 struct btrfs_device *device)
6964{
6965 struct btrfs_path *path;
6966 struct btrfs_key key;
6967 struct extent_buffer *eb;
6968 struct btrfs_dev_stats_item *ptr;
6969 int ret;
6970 int i;
6971
6972 key.objectid = BTRFS_DEV_STATS_OBJECTID;
6973 key.type = BTRFS_PERSISTENT_ITEM_KEY;
6974 key.offset = device->devid;
6975
6976 path = btrfs_alloc_path();
6977 BUG_ON(!path);
6978 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6979 if (ret < 0) {
6980 btrfs_warn_in_rcu(dev_root->fs_info,
6981 "error %d while searching for dev_stats item for device %s",
6982 ret, rcu_str_deref(device->name));
6983 goto out;
6984 }
6985
6986 if (ret == 0 &&
6987 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6988
6989 ret = btrfs_del_item(trans, dev_root, path);
6990 if (ret != 0) {
6991 btrfs_warn_in_rcu(dev_root->fs_info,
6992 "delete too small dev_stats item for device %s failed %d",
6993 rcu_str_deref(device->name), ret);
6994 goto out;
6995 }
6996 ret = 1;
6997 }
6998
6999 if (ret == 1) {
7000
7001 btrfs_release_path(path);
7002 ret = btrfs_insert_empty_item(trans, dev_root, path,
7003 &key, sizeof(*ptr));
7004 if (ret < 0) {
7005 btrfs_warn_in_rcu(dev_root->fs_info,
7006 "insert dev_stats item for device %s failed %d",
7007 rcu_str_deref(device->name), ret);
7008 goto out;
7009 }
7010 }
7011
7012 eb = path->nodes[0];
7013 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7014 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7015 btrfs_set_dev_stats_value(eb, ptr, i,
7016 btrfs_dev_stat_read(device, i));
7017 btrfs_mark_buffer_dirty(eb);
7018
7019out:
7020 btrfs_free_path(path);
7021 return ret;
7022}
7023
7024
7025
7026
7027int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
7028 struct btrfs_fs_info *fs_info)
7029{
7030 struct btrfs_root *dev_root = fs_info->dev_root;
7031 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7032 struct btrfs_device *device;
7033 int stats_cnt;
7034 int ret = 0;
7035
7036 mutex_lock(&fs_devices->device_list_mutex);
7037 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7038 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
7039 continue;
7040
7041 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7042 ret = update_dev_stat_item(trans, dev_root, device);
7043 if (!ret)
7044 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7045 }
7046 mutex_unlock(&fs_devices->device_list_mutex);
7047
7048 return ret;
7049}
7050
7051void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7052{
7053 btrfs_dev_stat_inc(dev, index);
7054 btrfs_dev_stat_print_on_error(dev);
7055}
7056
7057static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7058{
7059 if (!dev->dev_stats_valid)
7060 return;
7061 btrfs_err_rl_in_rcu(dev->dev_root->fs_info,
7062 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7063 rcu_str_deref(dev->name),
7064 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7065 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7066 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7067 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7068 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7069}
7070
7071static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7072{
7073 int i;
7074
7075 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7076 if (btrfs_dev_stat_read(dev, i) != 0)
7077 break;
7078 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7079 return;
7080
7081 btrfs_info_in_rcu(dev->dev_root->fs_info,
7082 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7083 rcu_str_deref(dev->name),
7084 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7085 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7086 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7087 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7088 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7089}
7090
7091int btrfs_get_dev_stats(struct btrfs_root *root,
7092 struct btrfs_ioctl_get_dev_stats *stats)
7093{
7094 struct btrfs_device *dev;
7095 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
7096 int i;
7097
7098 mutex_lock(&fs_devices->device_list_mutex);
7099 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
7100 mutex_unlock(&fs_devices->device_list_mutex);
7101
7102 if (!dev) {
7103 btrfs_warn(root->fs_info,
7104 "get dev_stats failed, device not found");
7105 return -ENODEV;
7106 } else if (!dev->dev_stats_valid) {
7107 btrfs_warn(root->fs_info,
7108 "get dev_stats failed, not yet valid");
7109 return -ENODEV;
7110 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7111 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7112 if (stats->nr_items > i)
7113 stats->values[i] =
7114 btrfs_dev_stat_read_and_reset(dev, i);
7115 else
7116 btrfs_dev_stat_reset(dev, i);
7117 }
7118 } else {
7119 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7120 if (stats->nr_items > i)
7121 stats->values[i] = btrfs_dev_stat_read(dev, i);
7122 }
7123 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7124 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7125 return 0;
7126}
7127
7128void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path)
7129{
7130 struct buffer_head *bh;
7131 struct btrfs_super_block *disk_super;
7132 int copy_num;
7133
7134 if (!bdev)
7135 return;
7136
7137 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX;
7138 copy_num++) {
7139
7140 if (btrfs_read_dev_one_super(bdev, copy_num, &bh))
7141 continue;
7142
7143 disk_super = (struct btrfs_super_block *)bh->b_data;
7144
7145 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
7146 set_buffer_dirty(bh);
7147 sync_dirty_buffer(bh);
7148 brelse(bh);
7149 }
7150
7151
7152 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
7153
7154
7155 update_dev_time(device_path);
7156}
7157
7158
7159
7160
7161
7162void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
7163{
7164 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7165 struct btrfs_device *curr, *next;
7166
7167 if (list_empty(&fs_devices->resized_devices))
7168 return;
7169
7170 mutex_lock(&fs_devices->device_list_mutex);
7171 lock_chunks(fs_info->dev_root);
7172 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
7173 resized_list) {
7174 list_del_init(&curr->resized_list);
7175 curr->commit_total_bytes = curr->disk_total_bytes;
7176 }
7177 unlock_chunks(fs_info->dev_root);
7178 mutex_unlock(&fs_devices->device_list_mutex);
7179}
7180
7181
7182void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
7183 struct btrfs_transaction *transaction)
7184{
7185 struct extent_map *em;
7186 struct map_lookup *map;
7187 struct btrfs_device *dev;
7188 int i;
7189
7190 if (list_empty(&transaction->pending_chunks))
7191 return;
7192
7193
7194 lock_chunks(root);
7195 list_for_each_entry(em, &transaction->pending_chunks, list) {
7196 map = em->map_lookup;
7197
7198 for (i = 0; i < map->num_stripes; i++) {
7199 dev = map->stripes[i].dev;
7200 dev->commit_bytes_used = dev->bytes_used;
7201 }
7202 }
7203 unlock_chunks(root);
7204}
7205
7206void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7207{
7208 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7209 while (fs_devices) {
7210 fs_devices->fs_info = fs_info;
7211 fs_devices = fs_devices->seed;
7212 }
7213}
7214
7215void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7216{
7217 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7218 while (fs_devices) {
7219 fs_devices->fs_info = NULL;
7220 fs_devices = fs_devices->seed;
7221 }
7222}
7223