2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
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>
32 #include "extent_map.h"
34 #include "transaction.h"
35 #include "print-tree.h"
38 #include "async-thread.h"
39 #include "check-integrity.h"
40 #include "rcu-string.h"
42 #include "dev-replace.h"
45 const struct btrfs_raid_attr btrfs_raid_array
[BTRFS_NR_RAID_TYPES
] = {
46 [BTRFS_RAID_RAID10
] = {
49 .devs_max
= 0, /* 0 == as many as possible */
51 .tolerated_failures
= 1,
55 [BTRFS_RAID_RAID1
] = {
60 .tolerated_failures
= 1,
69 .tolerated_failures
= 0,
73 [BTRFS_RAID_RAID0
] = {
78 .tolerated_failures
= 0,
82 [BTRFS_RAID_SINGLE
] = {
87 .tolerated_failures
= 0,
91 [BTRFS_RAID_RAID5
] = {
96 .tolerated_failures
= 1,
100 [BTRFS_RAID_RAID6
] = {
105 .tolerated_failures
= 2,
111 const 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
,
122 * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
123 * condition is not met. Zero means there's no corresponding
124 * BTRFS_ERROR_DEV_*_NOT_MET value.
126 const 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
,
136 static int init_first_rw_device(struct btrfs_trans_handle
*trans
,
137 struct btrfs_root
*root
,
138 struct btrfs_device
*device
);
139 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
);
140 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
);
141 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
);
142 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*device
);
144 DEFINE_MUTEX(uuid_mutex
);
145 static LIST_HEAD(fs_uuids
);
146 struct list_head
*btrfs_get_fs_uuids(void)
151 static struct btrfs_fs_devices
*__alloc_fs_devices(void)
153 struct btrfs_fs_devices
*fs_devs
;
155 fs_devs
= kzalloc(sizeof(*fs_devs
), GFP_KERNEL
);
157 return ERR_PTR(-ENOMEM
);
159 mutex_init(&fs_devs
->device_list_mutex
);
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
);
170 * alloc_fs_devices - allocate struct btrfs_fs_devices
171 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
174 * Return: a pointer to a new &struct btrfs_fs_devices on success;
175 * ERR_PTR() on error. Returned struct is not linked onto any lists and
176 * can be destroyed with kfree() right away.
178 static struct btrfs_fs_devices
*alloc_fs_devices(const u8
*fsid
)
180 struct btrfs_fs_devices
*fs_devs
;
182 fs_devs
= __alloc_fs_devices();
187 memcpy(fs_devs
->fsid
, fsid
, BTRFS_FSID_SIZE
);
189 generate_random_uuid(fs_devs
->fsid
);
194 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
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
);
208 static void btrfs_kobject_uevent(struct block_device
*bdev
,
209 enum kobject_action action
)
213 ret
= kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, action
);
215 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
217 kobject_name(&disk_to_dev(bdev
->bd_disk
)->kobj
),
218 &disk_to_dev(bdev
->bd_disk
)->kobj
);
221 void btrfs_cleanup_fs_uuids(void)
223 struct btrfs_fs_devices
*fs_devices
;
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
);
233 static struct btrfs_device
*__alloc_device(void)
235 struct btrfs_device
*dev
;
237 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
239 return ERR_PTR(-ENOMEM
);
241 INIT_LIST_HEAD(&dev
->dev_list
);
242 INIT_LIST_HEAD(&dev
->dev_alloc_list
);
243 INIT_LIST_HEAD(&dev
->resized_list
);
245 spin_lock_init(&dev
->io_lock
);
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
);
257 static noinline
struct btrfs_device
*__find_device(struct list_head
*head
,
260 struct btrfs_device
*dev
;
262 list_for_each_entry(dev
, head
, dev_list
) {
263 if (dev
->devid
== devid
&&
264 (!uuid
|| !memcmp(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
))) {
271 static noinline
struct btrfs_fs_devices
*find_fsid(u8
*fsid
)
273 struct btrfs_fs_devices
*fs_devices
;
275 list_for_each_entry(fs_devices
, &fs_uuids
, list
) {
276 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
283 btrfs_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
)
289 *bdev
= blkdev_get_by_path(device_path
, flags
, holder
);
292 ret
= PTR_ERR(*bdev
);
297 filemap_write_and_wait((*bdev
)->bd_inode
->i_mapping
);
298 ret
= set_blocksize(*bdev
, 4096);
300 blkdev_put(*bdev
, flags
);
303 invalidate_bdev(*bdev
);
304 *bh
= btrfs_read_dev_super(*bdev
);
307 blkdev_put(*bdev
, flags
);
319 static void requeue_list(struct btrfs_pending_bios
*pending_bios
,
320 struct bio
*head
, struct bio
*tail
)
323 struct bio
*old_head
;
325 old_head
= pending_bios
->head
;
326 pending_bios
->head
= head
;
327 if (pending_bios
->tail
)
328 tail
->bi_next
= old_head
;
330 pending_bios
->tail
= tail
;
334 * we try to collect pending bios for a device so we don't get a large
335 * number of procs sending bios down to the same device. This greatly
336 * improves the schedulers ability to collect and merge the bios.
338 * But, it also turns into a long list of bios to process and that is sure
339 * to eventually make the worker thread block. The solution here is to
340 * make some progress and then put this work struct back at the end of
341 * the list if the block device is congested. This way, multiple devices
342 * can make progress from a single worker thread.
344 static noinline
void run_scheduled_bios(struct btrfs_device
*device
)
346 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
348 struct backing_dev_info
*bdi
;
349 struct btrfs_pending_bios
*pending_bios
;
353 unsigned long num_run
;
354 unsigned long batch_run
= 0;
356 unsigned long last_waited
= 0;
358 int sync_pending
= 0;
359 struct blk_plug plug
;
362 * this function runs all the bios we've collected for
363 * a particular device. We don't want to wander off to
364 * another device without first sending all of these down.
365 * So, setup a plug here and finish it off before we return
367 blk_start_plug(&plug
);
369 bdi
= blk_get_backing_dev_info(device
->bdev
);
370 limit
= btrfs_async_submit_limit(fs_info
);
371 limit
= limit
* 2 / 3;
374 spin_lock(&device
->io_lock
);
379 /* take all the bios off the list at once and process them
380 * later on (without the lock held). But, remember the
381 * tail and other pointers so the bios can be properly reinserted
382 * into the list if we hit congestion
384 if (!force_reg
&& device
->pending_sync_bios
.head
) {
385 pending_bios
= &device
->pending_sync_bios
;
388 pending_bios
= &device
->pending_bios
;
392 pending
= pending_bios
->head
;
393 tail
= pending_bios
->tail
;
394 WARN_ON(pending
&& !tail
);
397 * if pending was null this time around, no bios need processing
398 * at all and we can stop. Otherwise it'll loop back up again
399 * and do an additional check so no bios are missed.
401 * device->running_pending is used to synchronize with the
404 if (device
->pending_sync_bios
.head
== NULL
&&
405 device
->pending_bios
.head
== NULL
) {
407 device
->running_pending
= 0;
410 device
->running_pending
= 1;
413 pending_bios
->head
= NULL
;
414 pending_bios
->tail
= NULL
;
416 spin_unlock(&device
->io_lock
);
421 /* we want to work on both lists, but do more bios on the
422 * sync list than the regular list
425 pending_bios
!= &device
->pending_sync_bios
&&
426 device
->pending_sync_bios
.head
) ||
427 (num_run
> 64 && pending_bios
== &device
->pending_sync_bios
&&
428 device
->pending_bios
.head
)) {
429 spin_lock(&device
->io_lock
);
430 requeue_list(pending_bios
, pending
, tail
);
435 pending
= pending
->bi_next
;
439 * atomic_dec_return implies a barrier for waitqueue_active
441 if (atomic_dec_return(&fs_info
->nr_async_bios
) < limit
&&
442 waitqueue_active(&fs_info
->async_submit_wait
))
443 wake_up(&fs_info
->async_submit_wait
);
445 BUG_ON(atomic_read(&cur
->__bi_cnt
) == 0);
448 * if we're doing the sync list, record that our
449 * plug has some sync requests on it
451 * If we're doing the regular list and there are
452 * sync requests sitting around, unplug before
455 if (pending_bios
== &device
->pending_sync_bios
) {
457 } else if (sync_pending
) {
458 blk_finish_plug(&plug
);
459 blk_start_plug(&plug
);
463 btrfsic_submit_bio(cur
);
470 * we made progress, there is more work to do and the bdi
471 * is now congested. Back off and let other work structs
474 if (pending
&& bdi_write_congested(bdi
) && batch_run
> 8 &&
475 fs_info
->fs_devices
->open_devices
> 1) {
476 struct io_context
*ioc
;
478 ioc
= current
->io_context
;
481 * the main goal here is that we don't want to
482 * block if we're going to be able to submit
483 * more requests without blocking.
485 * This code does two great things, it pokes into
486 * the elevator code from a filesystem _and_
487 * it makes assumptions about how batching works.
489 if (ioc
&& ioc
->nr_batch_requests
> 0 &&
490 time_before(jiffies
, ioc
->last_waited
+ HZ
/50UL) &&
492 ioc
->last_waited
== last_waited
)) {
494 * we want to go through our batch of
495 * requests and stop. So, we copy out
496 * the ioc->last_waited time and test
497 * against it before looping
499 last_waited
= ioc
->last_waited
;
503 spin_lock(&device
->io_lock
);
504 requeue_list(pending_bios
, pending
, tail
);
505 device
->running_pending
= 1;
507 spin_unlock(&device
->io_lock
);
508 btrfs_queue_work(fs_info
->submit_workers
,
512 /* unplug every 64 requests just for good measure */
513 if (batch_run
% 64 == 0) {
514 blk_finish_plug(&plug
);
515 blk_start_plug(&plug
);
524 spin_lock(&device
->io_lock
);
525 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
527 spin_unlock(&device
->io_lock
);
530 blk_finish_plug(&plug
);
533 static void pending_bios_fn(struct btrfs_work
*work
)
535 struct btrfs_device
*device
;
537 device
= container_of(work
, struct btrfs_device
, work
);
538 run_scheduled_bios(device
);
542 void btrfs_free_stale_device(struct btrfs_device
*cur_dev
)
544 struct btrfs_fs_devices
*fs_devs
;
545 struct btrfs_device
*dev
;
550 list_for_each_entry(fs_devs
, &fs_uuids
, list
) {
555 if (fs_devs
->seeding
)
558 list_for_each_entry(dev
, &fs_devs
->devices
, dev_list
) {
566 * Todo: This won't be enough. What if the same device
567 * comes back (with new uuid and) with its mapper path?
568 * But for now, this does help as mostly an admin will
569 * either use mapper or non mapper path throughout.
572 del
= strcmp(rcu_str_deref(dev
->name
),
573 rcu_str_deref(cur_dev
->name
));
580 /* delete the stale device */
581 if (fs_devs
->num_devices
== 1) {
582 btrfs_sysfs_remove_fsid(fs_devs
);
583 list_del(&fs_devs
->list
);
584 free_fs_devices(fs_devs
);
586 fs_devs
->num_devices
--;
587 list_del(&dev
->dev_list
);
588 rcu_string_free(dev
->name
);
597 * Add new device to list of registered devices
600 * 1 - first time device is seen
601 * 0 - device already known
604 static noinline
int device_list_add(const char *path
,
605 struct btrfs_super_block
*disk_super
,
606 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
608 struct btrfs_device
*device
;
609 struct btrfs_fs_devices
*fs_devices
;
610 struct rcu_string
*name
;
612 u64 found_transid
= btrfs_super_generation(disk_super
);
614 fs_devices
= find_fsid(disk_super
->fsid
);
616 fs_devices
= alloc_fs_devices(disk_super
->fsid
);
617 if (IS_ERR(fs_devices
))
618 return PTR_ERR(fs_devices
);
620 list_add(&fs_devices
->list
, &fs_uuids
);
624 device
= __find_device(&fs_devices
->devices
, devid
,
625 disk_super
->dev_item
.uuid
);
629 if (fs_devices
->opened
)
632 device
= btrfs_alloc_device(NULL
, &devid
,
633 disk_super
->dev_item
.uuid
);
634 if (IS_ERR(device
)) {
635 /* we can safely leave the fs_devices entry around */
636 return PTR_ERR(device
);
639 name
= rcu_string_strdup(path
, GFP_NOFS
);
644 rcu_assign_pointer(device
->name
, name
);
646 mutex_lock(&fs_devices
->device_list_mutex
);
647 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
648 fs_devices
->num_devices
++;
649 mutex_unlock(&fs_devices
->device_list_mutex
);
652 device
->fs_devices
= fs_devices
;
653 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
655 * When FS is already mounted.
656 * 1. If you are here and if the device->name is NULL that
657 * means this device was missing at time of FS mount.
658 * 2. If you are here and if the device->name is different
659 * from 'path' that means either
660 * a. The same device disappeared and reappeared with
662 * b. The missing-disk-which-was-replaced, has
665 * We must allow 1 and 2a above. But 2b would be a spurious
668 * Further in case of 1 and 2a above, the disk at 'path'
669 * would have missed some transaction when it was away and
670 * in case of 2a the stale bdev has to be updated as well.
671 * 2b must not be allowed at all time.
675 * For now, we do allow update to btrfs_fs_device through the
676 * btrfs dev scan cli after FS has been mounted. We're still
677 * tracking a problem where systems fail mount by subvolume id
678 * when we reject replacement on a mounted FS.
680 if (!fs_devices
->opened
&& found_transid
< device
->generation
) {
682 * That is if the FS is _not_ mounted and if you
683 * are here, that means there is more than one
684 * disk with same uuid and devid.We keep the one
685 * with larger generation number or the last-in if
686 * generation are equal.
691 name
= rcu_string_strdup(path
, GFP_NOFS
);
694 rcu_string_free(device
->name
);
695 rcu_assign_pointer(device
->name
, name
);
696 if (device
->missing
) {
697 fs_devices
->missing_devices
--;
703 * Unmount does not free the btrfs_device struct but would zero
704 * generation along with most of the other members. So just update
705 * it back. We need it to pick the disk with largest generation
708 if (!fs_devices
->opened
)
709 device
->generation
= found_transid
;
712 * if there is new btrfs on an already registered device,
713 * then remove the stale device entry.
716 btrfs_free_stale_device(device
);
718 *fs_devices_ret
= fs_devices
;
723 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
725 struct btrfs_fs_devices
*fs_devices
;
726 struct btrfs_device
*device
;
727 struct btrfs_device
*orig_dev
;
729 fs_devices
= alloc_fs_devices(orig
->fsid
);
730 if (IS_ERR(fs_devices
))
733 mutex_lock(&orig
->device_list_mutex
);
734 fs_devices
->total_devices
= orig
->total_devices
;
736 /* We have held the volume lock, it is safe to get the devices. */
737 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
738 struct rcu_string
*name
;
740 device
= btrfs_alloc_device(NULL
, &orig_dev
->devid
,
746 * This is ok to do without rcu read locked because we hold the
747 * uuid mutex so nothing we touch in here is going to disappear.
749 if (orig_dev
->name
) {
750 name
= rcu_string_strdup(orig_dev
->name
->str
,
756 rcu_assign_pointer(device
->name
, name
);
759 list_add(&device
->dev_list
, &fs_devices
->devices
);
760 device
->fs_devices
= fs_devices
;
761 fs_devices
->num_devices
++;
763 mutex_unlock(&orig
->device_list_mutex
);
766 mutex_unlock(&orig
->device_list_mutex
);
767 free_fs_devices(fs_devices
);
768 return ERR_PTR(-ENOMEM
);
771 void btrfs_close_extra_devices(struct btrfs_fs_devices
*fs_devices
, int step
)
773 struct btrfs_device
*device
, *next
;
774 struct btrfs_device
*latest_dev
= NULL
;
776 mutex_lock(&uuid_mutex
);
778 /* This is the initialized path, it is safe to release the devices. */
779 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
780 if (device
->in_fs_metadata
) {
781 if (!device
->is_tgtdev_for_dev_replace
&&
783 device
->generation
> latest_dev
->generation
)) {
789 if (device
->devid
== BTRFS_DEV_REPLACE_DEVID
) {
791 * In the first step, keep the device which has
792 * the correct fsid and the devid that is used
793 * for the dev_replace procedure.
794 * In the second step, the dev_replace state is
795 * read from the device tree and it is known
796 * whether the procedure is really active or
797 * not, which means whether this device is
798 * used or whether it should be removed.
800 if (step
== 0 || device
->is_tgtdev_for_dev_replace
) {
805 blkdev_put(device
->bdev
, device
->mode
);
807 fs_devices
->open_devices
--;
809 if (device
->writeable
) {
810 list_del_init(&device
->dev_alloc_list
);
811 device
->writeable
= 0;
812 if (!device
->is_tgtdev_for_dev_replace
)
813 fs_devices
->rw_devices
--;
815 list_del_init(&device
->dev_list
);
816 fs_devices
->num_devices
--;
817 rcu_string_free(device
->name
);
821 if (fs_devices
->seed
) {
822 fs_devices
= fs_devices
->seed
;
826 fs_devices
->latest_bdev
= latest_dev
->bdev
;
828 mutex_unlock(&uuid_mutex
);
831 static void __free_device(struct work_struct
*work
)
833 struct btrfs_device
*device
;
835 device
= container_of(work
, struct btrfs_device
, rcu_work
);
836 rcu_string_free(device
->name
);
840 static void free_device(struct rcu_head
*head
)
842 struct btrfs_device
*device
;
844 device
= container_of(head
, struct btrfs_device
, rcu
);
846 INIT_WORK(&device
->rcu_work
, __free_device
);
847 schedule_work(&device
->rcu_work
);
850 static void btrfs_close_bdev(struct btrfs_device
*device
)
852 if (device
->bdev
&& device
->writeable
) {
853 sync_blockdev(device
->bdev
);
854 invalidate_bdev(device
->bdev
);
858 blkdev_put(device
->bdev
, device
->mode
);
861 static void btrfs_prepare_close_one_device(struct btrfs_device
*device
)
863 struct btrfs_fs_devices
*fs_devices
= device
->fs_devices
;
864 struct btrfs_device
*new_device
;
865 struct rcu_string
*name
;
868 fs_devices
->open_devices
--;
870 if (device
->writeable
&&
871 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
872 list_del_init(&device
->dev_alloc_list
);
873 fs_devices
->rw_devices
--;
877 fs_devices
->missing_devices
--;
879 new_device
= btrfs_alloc_device(NULL
, &device
->devid
,
881 BUG_ON(IS_ERR(new_device
)); /* -ENOMEM */
883 /* Safe because we are under uuid_mutex */
885 name
= rcu_string_strdup(device
->name
->str
, GFP_NOFS
);
886 BUG_ON(!name
); /* -ENOMEM */
887 rcu_assign_pointer(new_device
->name
, name
);
890 list_replace_rcu(&device
->dev_list
, &new_device
->dev_list
);
891 new_device
->fs_devices
= device
->fs_devices
;
894 static int __btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
896 struct btrfs_device
*device
, *tmp
;
897 struct list_head pending_put
;
899 INIT_LIST_HEAD(&pending_put
);
901 if (--fs_devices
->opened
> 0)
904 mutex_lock(&fs_devices
->device_list_mutex
);
905 list_for_each_entry_safe(device
, tmp
, &fs_devices
->devices
, dev_list
) {
906 btrfs_prepare_close_one_device(device
);
907 list_add(&device
->dev_list
, &pending_put
);
909 mutex_unlock(&fs_devices
->device_list_mutex
);
912 * btrfs_show_devname() is using the device_list_mutex,
913 * sometimes call to blkdev_put() leads vfs calling
914 * into this func. So do put outside of device_list_mutex,
917 while (!list_empty(&pending_put
)) {
918 device
= list_first_entry(&pending_put
,
919 struct btrfs_device
, dev_list
);
920 list_del(&device
->dev_list
);
921 btrfs_close_bdev(device
);
922 call_rcu(&device
->rcu
, free_device
);
925 WARN_ON(fs_devices
->open_devices
);
926 WARN_ON(fs_devices
->rw_devices
);
927 fs_devices
->opened
= 0;
928 fs_devices
->seeding
= 0;
933 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
935 struct btrfs_fs_devices
*seed_devices
= NULL
;
938 mutex_lock(&uuid_mutex
);
939 ret
= __btrfs_close_devices(fs_devices
);
940 if (!fs_devices
->opened
) {
941 seed_devices
= fs_devices
->seed
;
942 fs_devices
->seed
= NULL
;
944 mutex_unlock(&uuid_mutex
);
946 while (seed_devices
) {
947 fs_devices
= seed_devices
;
948 seed_devices
= fs_devices
->seed
;
949 __btrfs_close_devices(fs_devices
);
950 free_fs_devices(fs_devices
);
953 * Wait for rcu kworkers under __btrfs_close_devices
954 * to finish all blkdev_puts so device is really
955 * free when umount is done.
961 static int __btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
962 fmode_t flags
, void *holder
)
964 struct request_queue
*q
;
965 struct block_device
*bdev
;
966 struct list_head
*head
= &fs_devices
->devices
;
967 struct btrfs_device
*device
;
968 struct btrfs_device
*latest_dev
= NULL
;
969 struct buffer_head
*bh
;
970 struct btrfs_super_block
*disk_super
;
977 list_for_each_entry(device
, head
, dev_list
) {
983 /* Just open everything we can; ignore failures here */
984 if (btrfs_get_bdev_and_sb(device
->name
->str
, flags
, holder
, 1,
988 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
989 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
990 if (devid
!= device
->devid
)
993 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
,
997 device
->generation
= btrfs_super_generation(disk_super
);
999 device
->generation
> latest_dev
->generation
)
1000 latest_dev
= device
;
1002 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
1003 device
->writeable
= 0;
1005 device
->writeable
= !bdev_read_only(bdev
);
1009 q
= bdev_get_queue(bdev
);
1010 if (blk_queue_discard(q
))
1011 device
->can_discard
= 1;
1013 device
->bdev
= bdev
;
1014 device
->in_fs_metadata
= 0;
1015 device
->mode
= flags
;
1017 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
1018 fs_devices
->rotating
= 1;
1020 fs_devices
->open_devices
++;
1021 if (device
->writeable
&&
1022 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
1023 fs_devices
->rw_devices
++;
1024 list_add(&device
->dev_alloc_list
,
1025 &fs_devices
->alloc_list
);
1032 blkdev_put(bdev
, flags
);
1035 if (fs_devices
->open_devices
== 0) {
1039 fs_devices
->seeding
= seeding
;
1040 fs_devices
->opened
= 1;
1041 fs_devices
->latest_bdev
= latest_dev
->bdev
;
1042 fs_devices
->total_rw_bytes
= 0;
1047 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
1048 fmode_t flags
, void *holder
)
1052 mutex_lock(&uuid_mutex
);
1053 if (fs_devices
->opened
) {
1054 fs_devices
->opened
++;
1057 ret
= __btrfs_open_devices(fs_devices
, flags
, holder
);
1059 mutex_unlock(&uuid_mutex
);
1063 void btrfs_release_disk_super(struct page
*page
)
1069 int btrfs_read_disk_super(struct block_device
*bdev
, u64 bytenr
,
1070 struct page
**page
, struct btrfs_super_block
**disk_super
)
1075 /* make sure our super fits in the device */
1076 if (bytenr
+ PAGE_SIZE
>= i_size_read(bdev
->bd_inode
))
1079 /* make sure our super fits in the page */
1080 if (sizeof(**disk_super
) > PAGE_SIZE
)
1083 /* make sure our super doesn't straddle pages on disk */
1084 index
= bytenr
>> PAGE_SHIFT
;
1085 if ((bytenr
+ sizeof(**disk_super
) - 1) >> PAGE_SHIFT
!= index
)
1088 /* pull in the page with our super */
1089 *page
= read_cache_page_gfp(bdev
->bd_inode
->i_mapping
,
1092 if (IS_ERR_OR_NULL(*page
))
1097 /* align our pointer to the offset of the super block */
1098 *disk_super
= p
+ (bytenr
& ~PAGE_MASK
);
1100 if (btrfs_super_bytenr(*disk_super
) != bytenr
||
1101 btrfs_super_magic(*disk_super
) != BTRFS_MAGIC
) {
1102 btrfs_release_disk_super(*page
);
1106 if ((*disk_super
)->label
[0] &&
1107 (*disk_super
)->label
[BTRFS_LABEL_SIZE
- 1])
1108 (*disk_super
)->label
[BTRFS_LABEL_SIZE
- 1] = '\0';
1114 * Look for a btrfs signature on a device. This may be called out of the mount path
1115 * and we are not allowed to call set_blocksize during the scan. The superblock
1116 * is read via pagecache
1118 int btrfs_scan_one_device(const char *path
, fmode_t flags
, void *holder
,
1119 struct btrfs_fs_devices
**fs_devices_ret
)
1121 struct btrfs_super_block
*disk_super
;
1122 struct block_device
*bdev
;
1131 * we would like to check all the supers, but that would make
1132 * a btrfs mount succeed after a mkfs from a different FS.
1133 * So, we need to add a special mount option to scan for
1134 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1136 bytenr
= btrfs_sb_offset(0);
1137 flags
|= FMODE_EXCL
;
1138 mutex_lock(&uuid_mutex
);
1140 bdev
= blkdev_get_by_path(path
, flags
, holder
);
1142 ret
= PTR_ERR(bdev
);
1146 if (btrfs_read_disk_super(bdev
, bytenr
, &page
, &disk_super
))
1147 goto error_bdev_put
;
1149 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1150 transid
= btrfs_super_generation(disk_super
);
1151 total_devices
= btrfs_super_num_devices(disk_super
);
1153 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
1155 if (disk_super
->label
[0]) {
1156 pr_info("BTRFS: device label %s ", disk_super
->label
);
1158 pr_info("BTRFS: device fsid %pU ", disk_super
->fsid
);
1161 pr_cont("devid %llu transid %llu %s\n", devid
, transid
, path
);
1164 if (!ret
&& fs_devices_ret
)
1165 (*fs_devices_ret
)->total_devices
= total_devices
;
1167 btrfs_release_disk_super(page
);
1170 blkdev_put(bdev
, flags
);
1172 mutex_unlock(&uuid_mutex
);
1176 /* helper to account the used device space in the range */
1177 int btrfs_account_dev_extents_size(struct btrfs_device
*device
, u64 start
,
1178 u64 end
, u64
*length
)
1180 struct btrfs_key key
;
1181 struct btrfs_root
*root
= device
->fs_info
->dev_root
;
1182 struct btrfs_dev_extent
*dev_extent
;
1183 struct btrfs_path
*path
;
1187 struct extent_buffer
*l
;
1191 if (start
>= device
->total_bytes
|| device
->is_tgtdev_for_dev_replace
)
1194 path
= btrfs_alloc_path();
1197 path
->reada
= READA_FORWARD
;
1199 key
.objectid
= device
->devid
;
1201 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1203 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1207 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1214 slot
= path
->slots
[0];
1215 if (slot
>= btrfs_header_nritems(l
)) {
1216 ret
= btrfs_next_leaf(root
, path
);
1224 btrfs_item_key_to_cpu(l
, &key
, slot
);
1226 if (key
.objectid
< device
->devid
)
1229 if (key
.objectid
> device
->devid
)
1232 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1235 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1236 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1238 if (key
.offset
<= start
&& extent_end
> end
) {
1239 *length
= end
- start
+ 1;
1241 } else if (key
.offset
<= start
&& extent_end
> start
)
1242 *length
+= extent_end
- start
;
1243 else if (key
.offset
> start
&& extent_end
<= end
)
1244 *length
+= extent_end
- key
.offset
;
1245 else if (key
.offset
> start
&& key
.offset
<= end
) {
1246 *length
+= end
- key
.offset
+ 1;
1248 } else if (key
.offset
> end
)
1256 btrfs_free_path(path
);
1260 static int contains_pending_extent(struct btrfs_transaction
*transaction
,
1261 struct btrfs_device
*device
,
1262 u64
*start
, u64 len
)
1264 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1265 struct extent_map
*em
;
1266 struct list_head
*search_list
= &fs_info
->pinned_chunks
;
1268 u64 physical_start
= *start
;
1271 search_list
= &transaction
->pending_chunks
;
1273 list_for_each_entry(em
, search_list
, list
) {
1274 struct map_lookup
*map
;
1277 map
= em
->map_lookup
;
1278 for (i
= 0; i
< map
->num_stripes
; i
++) {
1281 if (map
->stripes
[i
].dev
!= device
)
1283 if (map
->stripes
[i
].physical
>= physical_start
+ len
||
1284 map
->stripes
[i
].physical
+ em
->orig_block_len
<=
1288 * Make sure that while processing the pinned list we do
1289 * not override our *start with a lower value, because
1290 * we can have pinned chunks that fall within this
1291 * device hole and that have lower physical addresses
1292 * than the pending chunks we processed before. If we
1293 * do not take this special care we can end up getting
1294 * 2 pending chunks that start at the same physical
1295 * device offsets because the end offset of a pinned
1296 * chunk can be equal to the start offset of some
1299 end
= map
->stripes
[i
].physical
+ em
->orig_block_len
;
1306 if (search_list
!= &fs_info
->pinned_chunks
) {
1307 search_list
= &fs_info
->pinned_chunks
;
1316 * find_free_dev_extent_start - find free space in the specified device
1317 * @device: the device which we search the free space in
1318 * @num_bytes: the size of the free space that we need
1319 * @search_start: the position from which to begin the search
1320 * @start: store the start of the free space.
1321 * @len: the size of the free space. that we find, or the size
1322 * of the max free space if we don't find suitable free space
1324 * this uses a pretty simple search, the expectation is that it is
1325 * called very infrequently and that a given device has a small number
1328 * @start is used to store the start of the free space if we find. But if we
1329 * don't find suitable free space, it will be used to store the start position
1330 * of the max free space.
1332 * @len is used to store the size of the free space that we find.
1333 * But if we don't find suitable free space, it is used to store the size of
1334 * the max free space.
1336 int find_free_dev_extent_start(struct btrfs_transaction
*transaction
,
1337 struct btrfs_device
*device
, u64 num_bytes
,
1338 u64 search_start
, u64
*start
, u64
*len
)
1340 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1341 struct btrfs_root
*root
= fs_info
->dev_root
;
1342 struct btrfs_key key
;
1343 struct btrfs_dev_extent
*dev_extent
;
1344 struct btrfs_path
*path
;
1349 u64 search_end
= device
->total_bytes
;
1352 struct extent_buffer
*l
;
1353 u64 min_search_start
;
1356 * We don't want to overwrite the superblock on the drive nor any area
1357 * used by the boot loader (grub for example), so we make sure to start
1358 * at an offset of at least 1MB.
1360 min_search_start
= max(fs_info
->alloc_start
, 1024ull * 1024);
1361 search_start
= max(search_start
, min_search_start
);
1363 path
= btrfs_alloc_path();
1367 max_hole_start
= search_start
;
1371 if (search_start
>= search_end
|| device
->is_tgtdev_for_dev_replace
) {
1376 path
->reada
= READA_FORWARD
;
1377 path
->search_commit_root
= 1;
1378 path
->skip_locking
= 1;
1380 key
.objectid
= device
->devid
;
1381 key
.offset
= search_start
;
1382 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1384 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1388 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1395 slot
= path
->slots
[0];
1396 if (slot
>= btrfs_header_nritems(l
)) {
1397 ret
= btrfs_next_leaf(root
, path
);
1405 btrfs_item_key_to_cpu(l
, &key
, slot
);
1407 if (key
.objectid
< device
->devid
)
1410 if (key
.objectid
> device
->devid
)
1413 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1416 if (key
.offset
> search_start
) {
1417 hole_size
= key
.offset
- search_start
;
1420 * Have to check before we set max_hole_start, otherwise
1421 * we could end up sending back this offset anyway.
1423 if (contains_pending_extent(transaction
, device
,
1426 if (key
.offset
>= search_start
) {
1427 hole_size
= key
.offset
- search_start
;
1434 if (hole_size
> max_hole_size
) {
1435 max_hole_start
= search_start
;
1436 max_hole_size
= hole_size
;
1440 * If this free space is greater than which we need,
1441 * it must be the max free space that we have found
1442 * until now, so max_hole_start must point to the start
1443 * of this free space and the length of this free space
1444 * is stored in max_hole_size. Thus, we return
1445 * max_hole_start and max_hole_size and go back to the
1448 if (hole_size
>= num_bytes
) {
1454 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1455 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1457 if (extent_end
> search_start
)
1458 search_start
= extent_end
;
1465 * At this point, search_start should be the end of
1466 * allocated dev extents, and when shrinking the device,
1467 * search_end may be smaller than search_start.
1469 if (search_end
> search_start
) {
1470 hole_size
= search_end
- search_start
;
1472 if (contains_pending_extent(transaction
, device
, &search_start
,
1474 btrfs_release_path(path
);
1478 if (hole_size
> max_hole_size
) {
1479 max_hole_start
= search_start
;
1480 max_hole_size
= hole_size
;
1485 if (max_hole_size
< num_bytes
)
1491 btrfs_free_path(path
);
1492 *start
= max_hole_start
;
1494 *len
= max_hole_size
;
1498 int find_free_dev_extent(struct btrfs_trans_handle
*trans
,
1499 struct btrfs_device
*device
, u64 num_bytes
,
1500 u64
*start
, u64
*len
)
1502 /* FIXME use last free of some kind */
1503 return find_free_dev_extent_start(trans
->transaction
, device
,
1504 num_bytes
, 0, start
, len
);
1507 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1508 struct btrfs_device
*device
,
1509 u64 start
, u64
*dev_extent_len
)
1511 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1512 struct btrfs_root
*root
= fs_info
->dev_root
;
1514 struct btrfs_path
*path
;
1515 struct btrfs_key key
;
1516 struct btrfs_key found_key
;
1517 struct extent_buffer
*leaf
= NULL
;
1518 struct btrfs_dev_extent
*extent
= NULL
;
1520 path
= btrfs_alloc_path();
1524 key
.objectid
= device
->devid
;
1526 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1528 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1530 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1531 BTRFS_DEV_EXTENT_KEY
);
1534 leaf
= path
->nodes
[0];
1535 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1536 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1537 struct btrfs_dev_extent
);
1538 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1539 btrfs_dev_extent_length(leaf
, extent
) < start
);
1541 btrfs_release_path(path
);
1543 } else if (ret
== 0) {
1544 leaf
= path
->nodes
[0];
1545 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1546 struct btrfs_dev_extent
);
1548 btrfs_handle_fs_error(fs_info
, ret
, "Slot search failed");
1552 *dev_extent_len
= btrfs_dev_extent_length(leaf
, extent
);
1554 ret
= btrfs_del_item(trans
, root
, path
);
1556 btrfs_handle_fs_error(fs_info
, ret
,
1557 "Failed to remove dev extent item");
1559 set_bit(BTRFS_TRANS_HAVE_FREE_BGS
, &trans
->transaction
->flags
);
1562 btrfs_free_path(path
);
1566 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1567 struct btrfs_device
*device
,
1568 u64 chunk_tree
, u64 chunk_objectid
,
1569 u64 chunk_offset
, u64 start
, u64 num_bytes
)
1572 struct btrfs_path
*path
;
1573 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1574 struct btrfs_root
*root
= fs_info
->dev_root
;
1575 struct btrfs_dev_extent
*extent
;
1576 struct extent_buffer
*leaf
;
1577 struct btrfs_key key
;
1579 WARN_ON(!device
->in_fs_metadata
);
1580 WARN_ON(device
->is_tgtdev_for_dev_replace
);
1581 path
= btrfs_alloc_path();
1585 key
.objectid
= device
->devid
;
1587 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1588 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1593 leaf
= path
->nodes
[0];
1594 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1595 struct btrfs_dev_extent
);
1596 btrfs_set_dev_extent_chunk_tree(leaf
, extent
, chunk_tree
);
1597 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
, chunk_objectid
);
1598 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1600 write_extent_buffer_chunk_tree_uuid(leaf
, fs_info
->chunk_tree_uuid
);
1602 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1603 btrfs_mark_buffer_dirty(leaf
);
1605 btrfs_free_path(path
);
1609 static u64
find_next_chunk(struct btrfs_fs_info
*fs_info
)
1611 struct extent_map_tree
*em_tree
;
1612 struct extent_map
*em
;
1616 em_tree
= &fs_info
->mapping_tree
.map_tree
;
1617 read_lock(&em_tree
->lock
);
1618 n
= rb_last(&em_tree
->map
);
1620 em
= rb_entry(n
, struct extent_map
, rb_node
);
1621 ret
= em
->start
+ em
->len
;
1623 read_unlock(&em_tree
->lock
);
1628 static noinline
int find_next_devid(struct btrfs_fs_info
*fs_info
,
1632 struct btrfs_key key
;
1633 struct btrfs_key found_key
;
1634 struct btrfs_path
*path
;
1636 path
= btrfs_alloc_path();
1640 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1641 key
.type
= BTRFS_DEV_ITEM_KEY
;
1642 key
.offset
= (u64
)-1;
1644 ret
= btrfs_search_slot(NULL
, fs_info
->chunk_root
, &key
, path
, 0, 0);
1648 BUG_ON(ret
== 0); /* Corruption */
1650 ret
= btrfs_previous_item(fs_info
->chunk_root
, path
,
1651 BTRFS_DEV_ITEMS_OBJECTID
,
1652 BTRFS_DEV_ITEM_KEY
);
1656 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1658 *devid_ret
= found_key
.offset
+ 1;
1662 btrfs_free_path(path
);
1667 * the device information is stored in the chunk root
1668 * the btrfs_device struct should be fully filled in
1670 static int btrfs_add_device(struct btrfs_trans_handle
*trans
,
1671 struct btrfs_fs_info
*fs_info
,
1672 struct btrfs_device
*device
)
1674 struct btrfs_root
*root
= fs_info
->chunk_root
;
1676 struct btrfs_path
*path
;
1677 struct btrfs_dev_item
*dev_item
;
1678 struct extent_buffer
*leaf
;
1679 struct btrfs_key key
;
1682 path
= btrfs_alloc_path();
1686 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1687 key
.type
= BTRFS_DEV_ITEM_KEY
;
1688 key
.offset
= device
->devid
;
1690 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1695 leaf
= path
->nodes
[0];
1696 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
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);
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
, fs_info
->fsid
, ptr
, BTRFS_UUID_SIZE
);
1717 btrfs_mark_buffer_dirty(leaf
);
1721 btrfs_free_path(path
);
1726 * Function to update ctime/mtime for a given device path.
1727 * Mainly used for ctime/mtime based probe like libblkid.
1729 static void update_dev_time(char *path_name
)
1733 filp
= filp_open(path_name
, O_RDWR
, 0);
1736 file_update_time(filp
);
1737 filp_close(filp
, NULL
);
1740 static int btrfs_rm_dev_item(struct btrfs_fs_info
*fs_info
,
1741 struct btrfs_device
*device
)
1743 struct btrfs_root
*root
= fs_info
->chunk_root
;
1745 struct btrfs_path
*path
;
1746 struct btrfs_key key
;
1747 struct btrfs_trans_handle
*trans
;
1749 path
= btrfs_alloc_path();
1753 trans
= btrfs_start_transaction(root
, 0);
1754 if (IS_ERR(trans
)) {
1755 btrfs_free_path(path
);
1756 return PTR_ERR(trans
);
1758 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1759 key
.type
= BTRFS_DEV_ITEM_KEY
;
1760 key
.offset
= device
->devid
;
1762 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1771 ret
= btrfs_del_item(trans
, root
, path
);
1775 btrfs_free_path(path
);
1776 btrfs_commit_transaction(trans
, root
);
1781 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1782 * filesystem. It's up to the caller to adjust that number regarding eg. device
1785 static int btrfs_check_raid_min_devices(struct btrfs_fs_info
*fs_info
,
1793 seq
= read_seqbegin(&fs_info
->profiles_lock
);
1795 all_avail
= fs_info
->avail_data_alloc_bits
|
1796 fs_info
->avail_system_alloc_bits
|
1797 fs_info
->avail_metadata_alloc_bits
;
1798 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
1800 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
1801 if (!(all_avail
& btrfs_raid_group
[i
]))
1804 if (num_devices
< btrfs_raid_array
[i
].devs_min
) {
1805 int ret
= btrfs_raid_mindev_error
[i
];
1815 struct btrfs_device
*btrfs_find_next_active_device(struct btrfs_fs_devices
*fs_devs
,
1816 struct btrfs_device
*device
)
1818 struct btrfs_device
*next_device
;
1820 list_for_each_entry(next_device
, &fs_devs
->devices
, dev_list
) {
1821 if (next_device
!= device
&&
1822 !next_device
->missing
&& next_device
->bdev
)
1830 * Helper function to check if the given device is part of s_bdev / latest_bdev
1831 * and replace it with the provided or the next active device, in the context
1832 * where this function called, there should be always be another device (or
1833 * this_dev) which is active.
1835 void btrfs_assign_next_active_device(struct btrfs_fs_info
*fs_info
,
1836 struct btrfs_device
*device
, struct btrfs_device
*this_dev
)
1838 struct btrfs_device
*next_device
;
1841 next_device
= this_dev
;
1843 next_device
= btrfs_find_next_active_device(fs_info
->fs_devices
,
1845 ASSERT(next_device
);
1847 if (fs_info
->sb
->s_bdev
&&
1848 (fs_info
->sb
->s_bdev
== device
->bdev
))
1849 fs_info
->sb
->s_bdev
= next_device
->bdev
;
1851 if (fs_info
->fs_devices
->latest_bdev
== device
->bdev
)
1852 fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1855 int btrfs_rm_device(struct btrfs_root
*root
, char *device_path
, u64 devid
)
1857 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1858 struct btrfs_device
*device
;
1859 struct btrfs_fs_devices
*cur_devices
;
1862 bool clear_super
= false;
1864 mutex_lock(&uuid_mutex
);
1866 num_devices
= fs_info
->fs_devices
->num_devices
;
1867 btrfs_dev_replace_lock(&fs_info
->dev_replace
, 0);
1868 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
1869 WARN_ON(num_devices
< 1);
1872 btrfs_dev_replace_unlock(&fs_info
->dev_replace
, 0);
1874 ret
= btrfs_check_raid_min_devices(fs_info
, num_devices
- 1);
1878 ret
= btrfs_find_device_by_devspec(root
, devid
, device_path
,
1883 if (device
->is_tgtdev_for_dev_replace
) {
1884 ret
= BTRFS_ERROR_DEV_TGT_REPLACE
;
1888 if (device
->writeable
&& fs_info
->fs_devices
->rw_devices
== 1) {
1889 ret
= BTRFS_ERROR_DEV_ONLY_WRITABLE
;
1893 if (device
->writeable
) {
1894 lock_chunks(fs_info
);
1895 list_del_init(&device
->dev_alloc_list
);
1896 device
->fs_devices
->rw_devices
--;
1897 unlock_chunks(fs_info
);
1901 mutex_unlock(&uuid_mutex
);
1902 ret
= btrfs_shrink_device(device
, 0);
1903 mutex_lock(&uuid_mutex
);
1908 * TODO: the superblock still includes this device in its num_devices
1909 * counter although write_all_supers() is not locked out. This
1910 * could give a filesystem state which requires a degraded mount.
1912 ret
= btrfs_rm_dev_item(fs_info
, device
);
1916 device
->in_fs_metadata
= 0;
1917 btrfs_scrub_cancel_dev(fs_info
, device
);
1920 * the device list mutex makes sure that we don't change
1921 * the device list while someone else is writing out all
1922 * the device supers. Whoever is writing all supers, should
1923 * lock the device list mutex before getting the number of
1924 * devices in the super block (super_copy). Conversely,
1925 * whoever updates the number of devices in the super block
1926 * (super_copy) should hold the device list mutex.
1929 cur_devices
= device
->fs_devices
;
1930 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
1931 list_del_rcu(&device
->dev_list
);
1933 device
->fs_devices
->num_devices
--;
1934 device
->fs_devices
->total_devices
--;
1936 if (device
->missing
)
1937 device
->fs_devices
->missing_devices
--;
1939 btrfs_assign_next_active_device(fs_info
, device
, NULL
);
1942 device
->fs_devices
->open_devices
--;
1943 /* remove sysfs entry */
1944 btrfs_sysfs_rm_device_link(fs_info
->fs_devices
, device
);
1947 num_devices
= btrfs_super_num_devices(fs_info
->super_copy
) - 1;
1948 btrfs_set_super_num_devices(fs_info
->super_copy
, num_devices
);
1949 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
1952 * at this point, the device is zero sized and detached from
1953 * the devices list. All that's left is to zero out the old
1954 * supers and free the device.
1956 if (device
->writeable
)
1957 btrfs_scratch_superblocks(device
->bdev
, device
->name
->str
);
1959 btrfs_close_bdev(device
);
1960 call_rcu(&device
->rcu
, free_device
);
1962 if (cur_devices
->open_devices
== 0) {
1963 struct btrfs_fs_devices
*fs_devices
;
1964 fs_devices
= fs_info
->fs_devices
;
1965 while (fs_devices
) {
1966 if (fs_devices
->seed
== cur_devices
) {
1967 fs_devices
->seed
= cur_devices
->seed
;
1970 fs_devices
= fs_devices
->seed
;
1972 cur_devices
->seed
= NULL
;
1973 __btrfs_close_devices(cur_devices
);
1974 free_fs_devices(cur_devices
);
1977 fs_info
->num_tolerated_disk_barrier_failures
=
1978 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
1981 mutex_unlock(&uuid_mutex
);
1985 if (device
->writeable
) {
1986 lock_chunks(fs_info
);
1987 list_add(&device
->dev_alloc_list
,
1988 &fs_info
->fs_devices
->alloc_list
);
1989 device
->fs_devices
->rw_devices
++;
1990 unlock_chunks(fs_info
);
1995 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info
*fs_info
,
1996 struct btrfs_device
*srcdev
)
1998 struct btrfs_fs_devices
*fs_devices
;
2000 WARN_ON(!mutex_is_locked(&fs_info
->fs_devices
->device_list_mutex
));
2003 * in case of fs with no seed, srcdev->fs_devices will point
2004 * to fs_devices of fs_info. However when the dev being replaced is
2005 * a seed dev it will point to the seed's local fs_devices. In short
2006 * srcdev will have its correct fs_devices in both the cases.
2008 fs_devices
= srcdev
->fs_devices
;
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
--;
2016 if (srcdev
->writeable
)
2017 fs_devices
->rw_devices
--;
2020 fs_devices
->open_devices
--;
2023 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info
*fs_info
,
2024 struct btrfs_device
*srcdev
)
2026 struct btrfs_fs_devices
*fs_devices
= srcdev
->fs_devices
;
2028 if (srcdev
->writeable
) {
2029 /* zero out the old super if it is writable */
2030 btrfs_scratch_superblocks(srcdev
->bdev
, srcdev
->name
->str
);
2033 btrfs_close_bdev(srcdev
);
2035 call_rcu(&srcdev
->rcu
, free_device
);
2038 * unless fs_devices is seed fs, num_devices shouldn't go
2041 BUG_ON(!fs_devices
->num_devices
&& !fs_devices
->seeding
);
2043 /* if this is no devs we rather delete the fs_devices */
2044 if (!fs_devices
->num_devices
) {
2045 struct btrfs_fs_devices
*tmp_fs_devices
;
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
;
2053 tmp_fs_devices
= tmp_fs_devices
->seed
;
2055 fs_devices
->seed
= NULL
;
2056 __btrfs_close_devices(fs_devices
);
2057 free_fs_devices(fs_devices
);
2061 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
2062 struct btrfs_device
*tgtdev
)
2064 mutex_lock(&uuid_mutex
);
2066 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2068 btrfs_sysfs_rm_device_link(fs_info
->fs_devices
, tgtdev
);
2071 fs_info
->fs_devices
->open_devices
--;
2073 fs_info
->fs_devices
->num_devices
--;
2075 btrfs_assign_next_active_device(fs_info
, tgtdev
, NULL
);
2077 list_del_rcu(&tgtdev
->dev_list
);
2079 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2080 mutex_unlock(&uuid_mutex
);
2083 * The update_dev_time() with in btrfs_scratch_superblocks()
2084 * may lead to a call to btrfs_show_devname() which will try
2085 * to hold device_list_mutex. And here this device
2086 * is already out of device list, so we don't have to hold
2087 * the device_list_mutex lock.
2089 btrfs_scratch_superblocks(tgtdev
->bdev
, tgtdev
->name
->str
);
2091 btrfs_close_bdev(tgtdev
);
2092 call_rcu(&tgtdev
->rcu
, free_device
);
2095 static int btrfs_find_device_by_path(struct btrfs_root
*root
, char *device_path
,
2096 struct btrfs_device
**device
)
2098 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2100 struct btrfs_super_block
*disk_super
;
2103 struct block_device
*bdev
;
2104 struct buffer_head
*bh
;
2107 ret
= btrfs_get_bdev_and_sb(device_path
, FMODE_READ
,
2108 fs_info
->bdev_holder
, 0, &bdev
, &bh
);
2111 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
2112 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
2113 dev_uuid
= disk_super
->dev_item
.uuid
;
2114 *device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, disk_super
->fsid
);
2118 blkdev_put(bdev
, FMODE_READ
);
2122 int btrfs_find_device_missing_or_by_path(struct btrfs_root
*root
,
2124 struct btrfs_device
**device
)
2126 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2129 if (strcmp(device_path
, "missing") == 0) {
2130 struct list_head
*devices
;
2131 struct btrfs_device
*tmp
;
2133 devices
= &fs_info
->fs_devices
->devices
;
2135 * It is safe to read the devices since the volume_mutex
2136 * is held by the caller.
2138 list_for_each_entry(tmp
, devices
, dev_list
) {
2139 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
2146 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND
;
2150 return btrfs_find_device_by_path(root
, device_path
, device
);
2155 * Lookup a device given by device id, or the path if the id is 0.
2157 int btrfs_find_device_by_devspec(struct btrfs_root
*root
, u64 devid
,
2159 struct btrfs_device
**device
)
2161 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2166 *device
= btrfs_find_device(fs_info
, devid
, NULL
, NULL
);
2170 if (!devpath
|| !devpath
[0])
2173 ret
= btrfs_find_device_missing_or_by_path(root
, devpath
,
2180 * does all the dirty work required for changing file system's UUID.
2182 static int btrfs_prepare_sprout(struct btrfs_root
*root
)
2184 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2185 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2186 struct btrfs_fs_devices
*old_devices
;
2187 struct btrfs_fs_devices
*seed_devices
;
2188 struct btrfs_super_block
*disk_super
= fs_info
->super_copy
;
2189 struct btrfs_device
*device
;
2192 BUG_ON(!mutex_is_locked(&uuid_mutex
));
2193 if (!fs_devices
->seeding
)
2196 seed_devices
= __alloc_fs_devices();
2197 if (IS_ERR(seed_devices
))
2198 return PTR_ERR(seed_devices
);
2200 old_devices
= clone_fs_devices(fs_devices
);
2201 if (IS_ERR(old_devices
)) {
2202 kfree(seed_devices
);
2203 return PTR_ERR(old_devices
);
2206 list_add(&old_devices
->list
, &fs_uuids
);
2208 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
2209 seed_devices
->opened
= 1;
2210 INIT_LIST_HEAD(&seed_devices
->devices
);
2211 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
2212 mutex_init(&seed_devices
->device_list_mutex
);
2214 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2215 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
2217 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
)
2218 device
->fs_devices
= seed_devices
;
2220 lock_chunks(fs_info
);
2221 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
2222 unlock_chunks(fs_info
);
2224 fs_devices
->seeding
= 0;
2225 fs_devices
->num_devices
= 0;
2226 fs_devices
->open_devices
= 0;
2227 fs_devices
->missing_devices
= 0;
2228 fs_devices
->rotating
= 0;
2229 fs_devices
->seed
= seed_devices
;
2231 generate_random_uuid(fs_devices
->fsid
);
2232 memcpy(fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2233 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2234 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2236 super_flags
= btrfs_super_flags(disk_super
) &
2237 ~BTRFS_SUPER_FLAG_SEEDING
;
2238 btrfs_set_super_flags(disk_super
, super_flags
);
2244 * Store the expected generation for seed devices in device items.
2246 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
2247 struct btrfs_fs_info
*fs_info
)
2249 struct btrfs_root
*root
= fs_info
->chunk_root
;
2250 struct btrfs_path
*path
;
2251 struct extent_buffer
*leaf
;
2252 struct btrfs_dev_item
*dev_item
;
2253 struct btrfs_device
*device
;
2254 struct btrfs_key key
;
2255 u8 fs_uuid
[BTRFS_UUID_SIZE
];
2256 u8 dev_uuid
[BTRFS_UUID_SIZE
];
2260 path
= btrfs_alloc_path();
2264 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2266 key
.type
= BTRFS_DEV_ITEM_KEY
;
2269 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2273 leaf
= path
->nodes
[0];
2275 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
2276 ret
= btrfs_next_leaf(root
, path
);
2281 leaf
= path
->nodes
[0];
2282 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2283 btrfs_release_path(path
);
2287 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2288 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
2289 key
.type
!= BTRFS_DEV_ITEM_KEY
)
2292 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2293 struct btrfs_dev_item
);
2294 devid
= btrfs_device_id(leaf
, dev_item
);
2295 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
2297 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
2299 device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, fs_uuid
);
2300 BUG_ON(!device
); /* Logic error */
2302 if (device
->fs_devices
->seeding
) {
2303 btrfs_set_device_generation(leaf
, dev_item
,
2304 device
->generation
);
2305 btrfs_mark_buffer_dirty(leaf
);
2313 btrfs_free_path(path
);
2317 int btrfs_init_new_device(struct btrfs_fs_info
*fs_info
, char *device_path
)
2319 struct btrfs_root
*root
= fs_info
->dev_root
;
2320 struct request_queue
*q
;
2321 struct btrfs_trans_handle
*trans
;
2322 struct btrfs_device
*device
;
2323 struct block_device
*bdev
;
2324 struct list_head
*devices
;
2325 struct super_block
*sb
= fs_info
->sb
;
2326 struct rcu_string
*name
;
2328 int seeding_dev
= 0;
2331 if ((sb
->s_flags
& MS_RDONLY
) && !fs_info
->fs_devices
->seeding
)
2334 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2335 fs_info
->bdev_holder
);
2337 return PTR_ERR(bdev
);
2339 if (fs_info
->fs_devices
->seeding
) {
2341 down_write(&sb
->s_umount
);
2342 mutex_lock(&uuid_mutex
);
2345 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2347 devices
= &fs_info
->fs_devices
->devices
;
2349 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2350 list_for_each_entry(device
, devices
, dev_list
) {
2351 if (device
->bdev
== bdev
) {
2354 &fs_info
->fs_devices
->device_list_mutex
);
2358 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2360 device
= btrfs_alloc_device(fs_info
, NULL
, NULL
);
2361 if (IS_ERR(device
)) {
2362 /* we can safely leave the fs_devices entry around */
2363 ret
= PTR_ERR(device
);
2367 name
= rcu_string_strdup(device_path
, GFP_KERNEL
);
2373 rcu_assign_pointer(device
->name
, name
);
2375 trans
= btrfs_start_transaction(root
, 0);
2376 if (IS_ERR(trans
)) {
2377 rcu_string_free(device
->name
);
2379 ret
= PTR_ERR(trans
);
2383 q
= bdev_get_queue(bdev
);
2384 if (blk_queue_discard(q
))
2385 device
->can_discard
= 1;
2386 device
->writeable
= 1;
2387 device
->generation
= trans
->transid
;
2388 device
->io_width
= fs_info
->sectorsize
;
2389 device
->io_align
= fs_info
->sectorsize
;
2390 device
->sector_size
= fs_info
->sectorsize
;
2391 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
2392 device
->disk_total_bytes
= device
->total_bytes
;
2393 device
->commit_total_bytes
= device
->total_bytes
;
2394 device
->fs_info
= fs_info
;
2395 device
->bdev
= bdev
;
2396 device
->in_fs_metadata
= 1;
2397 device
->is_tgtdev_for_dev_replace
= 0;
2398 device
->mode
= FMODE_EXCL
;
2399 device
->dev_stats_valid
= 1;
2400 set_blocksize(device
->bdev
, 4096);
2403 sb
->s_flags
&= ~MS_RDONLY
;
2404 ret
= btrfs_prepare_sprout(root
);
2405 BUG_ON(ret
); /* -ENOMEM */
2408 device
->fs_devices
= fs_info
->fs_devices
;
2410 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2411 lock_chunks(fs_info
);
2412 list_add_rcu(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2413 list_add(&device
->dev_alloc_list
,
2414 &fs_info
->fs_devices
->alloc_list
);
2415 fs_info
->fs_devices
->num_devices
++;
2416 fs_info
->fs_devices
->open_devices
++;
2417 fs_info
->fs_devices
->rw_devices
++;
2418 fs_info
->fs_devices
->total_devices
++;
2419 fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
2421 spin_lock(&fs_info
->free_chunk_lock
);
2422 fs_info
->free_chunk_space
+= device
->total_bytes
;
2423 spin_unlock(&fs_info
->free_chunk_lock
);
2425 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
2426 fs_info
->fs_devices
->rotating
= 1;
2428 tmp
= btrfs_super_total_bytes(fs_info
->super_copy
);
2429 btrfs_set_super_total_bytes(fs_info
->super_copy
,
2430 tmp
+ device
->total_bytes
);
2432 tmp
= btrfs_super_num_devices(fs_info
->super_copy
);
2433 btrfs_set_super_num_devices(fs_info
->super_copy
, tmp
+ 1);
2435 /* add sysfs device entry */
2436 btrfs_sysfs_add_device_link(fs_info
->fs_devices
, device
);
2439 * we've got more storage, clear any full flags on the space
2442 btrfs_clear_space_info_full(fs_info
);
2444 unlock_chunks(fs_info
);
2445 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2448 lock_chunks(fs_info
);
2449 ret
= init_first_rw_device(trans
, root
, device
);
2450 unlock_chunks(fs_info
);
2452 btrfs_abort_transaction(trans
, ret
);
2457 ret
= btrfs_add_device(trans
, fs_info
, device
);
2459 btrfs_abort_transaction(trans
, ret
);
2464 char fsid_buf
[BTRFS_UUID_UNPARSED_SIZE
];
2466 ret
= btrfs_finish_sprout(trans
, fs_info
);
2468 btrfs_abort_transaction(trans
, ret
);
2472 /* Sprouting would change fsid of the mounted root,
2473 * so rename the fsid on the sysfs
2475 snprintf(fsid_buf
, BTRFS_UUID_UNPARSED_SIZE
, "%pU",
2477 if (kobject_rename(&fs_info
->fs_devices
->fsid_kobj
, fsid_buf
))
2479 "sysfs: failed to create fsid for sprout");
2482 fs_info
->num_tolerated_disk_barrier_failures
=
2483 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
2484 ret
= btrfs_commit_transaction(trans
, root
);
2487 mutex_unlock(&uuid_mutex
);
2488 up_write(&sb
->s_umount
);
2490 if (ret
) /* transaction commit */
2493 ret
= btrfs_relocate_sys_chunks(root
);
2495 btrfs_handle_fs_error(fs_info
, ret
,
2496 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2497 trans
= btrfs_attach_transaction(root
);
2498 if (IS_ERR(trans
)) {
2499 if (PTR_ERR(trans
) == -ENOENT
)
2501 return PTR_ERR(trans
);
2503 ret
= btrfs_commit_transaction(trans
, root
);
2506 /* Update ctime/mtime for libblkid */
2507 update_dev_time(device_path
);
2511 btrfs_end_transaction(trans
, root
);
2512 rcu_string_free(device
->name
);
2513 btrfs_sysfs_rm_device_link(fs_info
->fs_devices
, device
);
2516 blkdev_put(bdev
, FMODE_EXCL
);
2518 mutex_unlock(&uuid_mutex
);
2519 up_write(&sb
->s_umount
);
2524 int btrfs_init_dev_replace_tgtdev(struct btrfs_root
*root
, char *device_path
,
2525 struct btrfs_device
*srcdev
,
2526 struct btrfs_device
**device_out
)
2528 struct request_queue
*q
;
2529 struct btrfs_device
*device
;
2530 struct block_device
*bdev
;
2531 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2532 struct list_head
*devices
;
2533 struct rcu_string
*name
;
2534 u64 devid
= BTRFS_DEV_REPLACE_DEVID
;
2538 if (fs_info
->fs_devices
->seeding
) {
2539 btrfs_err(fs_info
, "the filesystem is a seed filesystem!");
2543 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2544 fs_info
->bdev_holder
);
2546 btrfs_err(fs_info
, "target device %s is invalid!", device_path
);
2547 return PTR_ERR(bdev
);
2550 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2552 devices
= &fs_info
->fs_devices
->devices
;
2553 list_for_each_entry(device
, devices
, dev_list
) {
2554 if (device
->bdev
== bdev
) {
2556 "target device is in the filesystem!");
2563 if (i_size_read(bdev
->bd_inode
) <
2564 btrfs_device_get_total_bytes(srcdev
)) {
2566 "target device is smaller than source device!");
2572 device
= btrfs_alloc_device(NULL
, &devid
, NULL
);
2573 if (IS_ERR(device
)) {
2574 ret
= PTR_ERR(device
);
2578 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2584 rcu_assign_pointer(device
->name
, name
);
2586 q
= bdev_get_queue(bdev
);
2587 if (blk_queue_discard(q
))
2588 device
->can_discard
= 1;
2589 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2590 device
->writeable
= 1;
2591 device
->generation
= 0;
2592 device
->io_width
= fs_info
->sectorsize
;
2593 device
->io_align
= fs_info
->sectorsize
;
2594 device
->sector_size
= fs_info
->sectorsize
;
2595 device
->total_bytes
= btrfs_device_get_total_bytes(srcdev
);
2596 device
->disk_total_bytes
= btrfs_device_get_disk_total_bytes(srcdev
);
2597 device
->bytes_used
= btrfs_device_get_bytes_used(srcdev
);
2598 ASSERT(list_empty(&srcdev
->resized_list
));
2599 device
->commit_total_bytes
= srcdev
->commit_total_bytes
;
2600 device
->commit_bytes_used
= device
->bytes_used
;
2601 device
->fs_info
= fs_info
;
2602 device
->bdev
= bdev
;
2603 device
->in_fs_metadata
= 1;
2604 device
->is_tgtdev_for_dev_replace
= 1;
2605 device
->mode
= FMODE_EXCL
;
2606 device
->dev_stats_valid
= 1;
2607 set_blocksize(device
->bdev
, 4096);
2608 device
->fs_devices
= fs_info
->fs_devices
;
2609 list_add(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2610 fs_info
->fs_devices
->num_devices
++;
2611 fs_info
->fs_devices
->open_devices
++;
2612 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2614 *device_out
= device
;
2618 blkdev_put(bdev
, FMODE_EXCL
);
2622 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info
*fs_info
,
2623 struct btrfs_device
*tgtdev
)
2625 u32 sectorsize
= fs_info
->sectorsize
;
2627 WARN_ON(fs_info
->fs_devices
->rw_devices
== 0);
2628 tgtdev
->io_width
= sectorsize
;
2629 tgtdev
->io_align
= sectorsize
;
2630 tgtdev
->sector_size
= sectorsize
;
2631 tgtdev
->fs_info
= fs_info
;
2632 tgtdev
->in_fs_metadata
= 1;
2635 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2636 struct btrfs_device
*device
)
2639 struct btrfs_path
*path
;
2640 struct btrfs_root
*root
= device
->fs_info
->chunk_root
;
2641 struct btrfs_dev_item
*dev_item
;
2642 struct extent_buffer
*leaf
;
2643 struct btrfs_key key
;
2645 path
= btrfs_alloc_path();
2649 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2650 key
.type
= BTRFS_DEV_ITEM_KEY
;
2651 key
.offset
= device
->devid
;
2653 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2662 leaf
= path
->nodes
[0];
2663 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2665 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2666 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2667 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2668 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2669 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2670 btrfs_set_device_total_bytes(leaf
, dev_item
,
2671 btrfs_device_get_disk_total_bytes(device
));
2672 btrfs_set_device_bytes_used(leaf
, dev_item
,
2673 btrfs_device_get_bytes_used(device
));
2674 btrfs_mark_buffer_dirty(leaf
);
2677 btrfs_free_path(path
);
2681 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2682 struct btrfs_device
*device
, u64 new_size
)
2684 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
2685 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
2686 struct btrfs_fs_devices
*fs_devices
;
2690 if (!device
->writeable
)
2693 lock_chunks(fs_info
);
2694 old_total
= btrfs_super_total_bytes(super_copy
);
2695 diff
= new_size
- device
->total_bytes
;
2697 if (new_size
<= device
->total_bytes
||
2698 device
->is_tgtdev_for_dev_replace
) {
2699 unlock_chunks(fs_info
);
2703 fs_devices
= fs_info
->fs_devices
;
2705 btrfs_set_super_total_bytes(super_copy
, old_total
+ diff
);
2706 device
->fs_devices
->total_rw_bytes
+= diff
;
2708 btrfs_device_set_total_bytes(device
, new_size
);
2709 btrfs_device_set_disk_total_bytes(device
, new_size
);
2710 btrfs_clear_space_info_full(device
->fs_info
);
2711 if (list_empty(&device
->resized_list
))
2712 list_add_tail(&device
->resized_list
,
2713 &fs_devices
->resized_devices
);
2714 unlock_chunks(fs_info
);
2716 return btrfs_update_device(trans
, device
);
2719 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
2720 struct btrfs_fs_info
*fs_info
, u64 chunk_objectid
,
2723 struct btrfs_root
*root
= fs_info
->chunk_root
;
2725 struct btrfs_path
*path
;
2726 struct btrfs_key key
;
2728 path
= btrfs_alloc_path();
2732 key
.objectid
= chunk_objectid
;
2733 key
.offset
= chunk_offset
;
2734 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2736 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2739 else if (ret
> 0) { /* Logic error or corruption */
2740 btrfs_handle_fs_error(fs_info
, -ENOENT
,
2741 "Failed lookup while freeing chunk.");
2746 ret
= btrfs_del_item(trans
, root
, path
);
2748 btrfs_handle_fs_error(fs_info
, ret
,
2749 "Failed to delete chunk item.");
2751 btrfs_free_path(path
);
2755 static int btrfs_del_sys_chunk(struct btrfs_fs_info
*fs_info
,
2756 u64 chunk_objectid
, u64 chunk_offset
)
2758 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
2759 struct btrfs_disk_key
*disk_key
;
2760 struct btrfs_chunk
*chunk
;
2767 struct btrfs_key key
;
2769 lock_chunks(fs_info
);
2770 array_size
= btrfs_super_sys_array_size(super_copy
);
2772 ptr
= super_copy
->sys_chunk_array
;
2775 while (cur
< array_size
) {
2776 disk_key
= (struct btrfs_disk_key
*)ptr
;
2777 btrfs_disk_key_to_cpu(&key
, disk_key
);
2779 len
= sizeof(*disk_key
);
2781 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2782 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2783 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2784 len
+= btrfs_chunk_item_size(num_stripes
);
2789 if (key
.objectid
== chunk_objectid
&&
2790 key
.offset
== chunk_offset
) {
2791 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2793 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2799 unlock_chunks(fs_info
);
2803 int btrfs_remove_chunk(struct btrfs_trans_handle
*trans
,
2804 struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2806 struct extent_map_tree
*em_tree
;
2807 struct extent_map
*em
;
2808 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
2809 struct map_lookup
*map
;
2810 u64 dev_extent_len
= 0;
2811 u64 chunk_objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2813 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2815 em_tree
= &fs_info
->mapping_tree
.map_tree
;
2817 read_lock(&em_tree
->lock
);
2818 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
2819 read_unlock(&em_tree
->lock
);
2821 if (!em
|| em
->start
> chunk_offset
||
2822 em
->start
+ em
->len
< chunk_offset
) {
2824 * This is a logic error, but we don't want to just rely on the
2825 * user having built with ASSERT enabled, so if ASSERT doesn't
2826 * do anything we still error out.
2830 free_extent_map(em
);
2833 map
= em
->map_lookup
;
2834 lock_chunks(fs_info
);
2835 check_system_chunk(trans
, extent_root
, map
->type
);
2836 unlock_chunks(fs_info
);
2839 * Take the device list mutex to prevent races with the final phase of
2840 * a device replace operation that replaces the device object associated
2841 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2843 mutex_lock(&fs_devices
->device_list_mutex
);
2844 for (i
= 0; i
< map
->num_stripes
; i
++) {
2845 struct btrfs_device
*device
= map
->stripes
[i
].dev
;
2846 ret
= btrfs_free_dev_extent(trans
, device
,
2847 map
->stripes
[i
].physical
,
2850 mutex_unlock(&fs_devices
->device_list_mutex
);
2851 btrfs_abort_transaction(trans
, ret
);
2855 if (device
->bytes_used
> 0) {
2856 lock_chunks(fs_info
);
2857 btrfs_device_set_bytes_used(device
,
2858 device
->bytes_used
- dev_extent_len
);
2859 spin_lock(&fs_info
->free_chunk_lock
);
2860 fs_info
->free_chunk_space
+= dev_extent_len
;
2861 spin_unlock(&fs_info
->free_chunk_lock
);
2862 btrfs_clear_space_info_full(fs_info
);
2863 unlock_chunks(fs_info
);
2866 if (map
->stripes
[i
].dev
) {
2867 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2869 mutex_unlock(&fs_devices
->device_list_mutex
);
2870 btrfs_abort_transaction(trans
, ret
);
2875 mutex_unlock(&fs_devices
->device_list_mutex
);
2877 ret
= btrfs_free_chunk(trans
, fs_info
, chunk_objectid
, chunk_offset
);
2879 btrfs_abort_transaction(trans
, ret
);
2883 trace_btrfs_chunk_free(fs_info
, map
, chunk_offset
, em
->len
);
2885 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2886 ret
= btrfs_del_sys_chunk(fs_info
, chunk_objectid
,
2889 btrfs_abort_transaction(trans
, ret
);
2894 ret
= btrfs_remove_block_group(trans
, fs_info
, chunk_offset
, em
);
2896 btrfs_abort_transaction(trans
, ret
);
2902 free_extent_map(em
);
2906 static int btrfs_relocate_chunk(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2908 struct btrfs_root
*root
= fs_info
->chunk_root
;
2909 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
2910 struct btrfs_trans_handle
*trans
;
2914 * Prevent races with automatic removal of unused block groups.
2915 * After we relocate and before we remove the chunk with offset
2916 * chunk_offset, automatic removal of the block group can kick in,
2917 * resulting in a failure when calling btrfs_remove_chunk() below.
2919 * Make sure to acquire this mutex before doing a tree search (dev
2920 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2921 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2922 * we release the path used to search the chunk/dev tree and before
2923 * the current task acquires this mutex and calls us.
2925 ASSERT(mutex_is_locked(&fs_info
->delete_unused_bgs_mutex
));
2927 ret
= btrfs_can_relocate(fs_info
, chunk_offset
);
2931 /* step one, relocate all the extents inside this chunk */
2932 btrfs_scrub_pause(root
);
2933 ret
= btrfs_relocate_block_group(fs_info
, chunk_offset
);
2934 btrfs_scrub_continue(root
);
2938 trans
= btrfs_start_trans_remove_block_group(root
->fs_info
,
2940 if (IS_ERR(trans
)) {
2941 ret
= PTR_ERR(trans
);
2942 btrfs_handle_fs_error(root
->fs_info
, ret
, NULL
);
2947 * step two, delete the device extents and the
2948 * chunk tree entries
2950 ret
= btrfs_remove_chunk(trans
, fs_info
, chunk_offset
);
2951 btrfs_end_transaction(trans
, extent_root
);
2955 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
)
2957 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2958 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
2959 struct btrfs_path
*path
;
2960 struct extent_buffer
*leaf
;
2961 struct btrfs_chunk
*chunk
;
2962 struct btrfs_key key
;
2963 struct btrfs_key found_key
;
2965 bool retried
= false;
2969 path
= btrfs_alloc_path();
2974 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2975 key
.offset
= (u64
)-1;
2976 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2979 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
2980 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2982 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
2985 BUG_ON(ret
== 0); /* Corruption */
2987 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
2990 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
2996 leaf
= path
->nodes
[0];
2997 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2999 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
3000 struct btrfs_chunk
);
3001 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3002 btrfs_release_path(path
);
3004 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3005 ret
= btrfs_relocate_chunk(fs_info
, found_key
.offset
);
3011 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3013 if (found_key
.offset
== 0)
3015 key
.offset
= found_key
.offset
- 1;
3018 if (failed
&& !retried
) {
3022 } else if (WARN_ON(failed
&& retried
)) {
3026 btrfs_free_path(path
);
3030 static int insert_balance_item(struct btrfs_fs_info
*fs_info
,
3031 struct btrfs_balance_control
*bctl
)
3033 struct btrfs_root
*root
= fs_info
->tree_root
;
3034 struct btrfs_trans_handle
*trans
;
3035 struct btrfs_balance_item
*item
;
3036 struct btrfs_disk_balance_args disk_bargs
;
3037 struct btrfs_path
*path
;
3038 struct extent_buffer
*leaf
;
3039 struct btrfs_key key
;
3042 path
= btrfs_alloc_path();
3046 trans
= btrfs_start_transaction(root
, 0);
3047 if (IS_ERR(trans
)) {
3048 btrfs_free_path(path
);
3049 return PTR_ERR(trans
);
3052 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3053 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3056 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
3061 leaf
= path
->nodes
[0];
3062 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3064 memzero_extent_buffer(leaf
, (unsigned long)item
, sizeof(*item
));
3066 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
3067 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
3068 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
3069 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
3070 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
3071 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
3073 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
3075 btrfs_mark_buffer_dirty(leaf
);
3077 btrfs_free_path(path
);
3078 err
= btrfs_commit_transaction(trans
, root
);
3084 static int del_balance_item(struct btrfs_fs_info
*fs_info
)
3086 struct btrfs_root
*root
= fs_info
->tree_root
;
3087 struct btrfs_trans_handle
*trans
;
3088 struct btrfs_path
*path
;
3089 struct btrfs_key key
;
3092 path
= btrfs_alloc_path();
3096 trans
= btrfs_start_transaction(root
, 0);
3097 if (IS_ERR(trans
)) {
3098 btrfs_free_path(path
);
3099 return PTR_ERR(trans
);
3102 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3103 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3106 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3114 ret
= btrfs_del_item(trans
, root
, path
);
3116 btrfs_free_path(path
);
3117 err
= btrfs_commit_transaction(trans
, root
);
3124 * This is a heuristic used to reduce the number of chunks balanced on
3125 * resume after balance was interrupted.
3127 static void update_balance_args(struct btrfs_balance_control
*bctl
)
3130 * Turn on soft mode for chunk types that were being converted.
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
;
3140 * Turn on usage filter if is not already used. The idea is
3141 * that chunks that we have already balanced should be
3142 * reasonably full. Don't do it for chunks that are being
3143 * converted - that will keep us from relocating unconverted
3144 * (albeit full) chunks.
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;
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;
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;
3167 * Should be called with both balance and volume mutexes held to
3168 * serialize other volume operations (add_dev/rm_dev/resize) with
3169 * restriper. Same goes for unset_balance_control.
3171 static void set_balance_control(struct btrfs_balance_control
*bctl
)
3173 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3175 BUG_ON(fs_info
->balance_ctl
);
3177 spin_lock(&fs_info
->balance_lock
);
3178 fs_info
->balance_ctl
= bctl
;
3179 spin_unlock(&fs_info
->balance_lock
);
3182 static void unset_balance_control(struct btrfs_fs_info
*fs_info
)
3184 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3186 BUG_ON(!fs_info
->balance_ctl
);
3188 spin_lock(&fs_info
->balance_lock
);
3189 fs_info
->balance_ctl
= NULL
;
3190 spin_unlock(&fs_info
->balance_lock
);
3196 * Balance filters. Return 1 if chunk should be filtered out
3197 * (should not be balanced).
3199 static int chunk_profiles_filter(u64 chunk_type
,
3200 struct btrfs_balance_args
*bargs
)
3202 chunk_type
= chunk_to_extended(chunk_type
) &
3203 BTRFS_EXTENDED_PROFILE_MASK
;
3205 if (bargs
->profiles
& chunk_type
)
3211 static int chunk_usage_range_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
3212 struct btrfs_balance_args
*bargs
)
3214 struct btrfs_block_group_cache
*cache
;
3216 u64 user_thresh_min
;
3217 u64 user_thresh_max
;
3220 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3221 chunk_used
= btrfs_block_group_used(&cache
->item
);
3223 if (bargs
->usage_min
== 0)
3224 user_thresh_min
= 0;
3226 user_thresh_min
= div_factor_fine(cache
->key
.offset
,
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
;
3234 user_thresh_max
= div_factor_fine(cache
->key
.offset
,
3237 if (user_thresh_min
<= chunk_used
&& chunk_used
< user_thresh_max
)
3240 btrfs_put_block_group(cache
);
3244 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
,
3245 u64 chunk_offset
, struct btrfs_balance_args
*bargs
)
3247 struct btrfs_block_group_cache
*cache
;
3248 u64 chunk_used
, user_thresh
;
3251 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3252 chunk_used
= btrfs_block_group_used(&cache
->item
);
3254 if (bargs
->usage_min
== 0)
3256 else if (bargs
->usage
> 100)
3257 user_thresh
= cache
->key
.offset
;
3259 user_thresh
= div_factor_fine(cache
->key
.offset
,
3262 if (chunk_used
< user_thresh
)
3265 btrfs_put_block_group(cache
);
3269 static int chunk_devid_filter(struct extent_buffer
*leaf
,
3270 struct btrfs_chunk
*chunk
,
3271 struct btrfs_balance_args
*bargs
)
3273 struct btrfs_stripe
*stripe
;
3274 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3277 for (i
= 0; i
< num_stripes
; i
++) {
3278 stripe
= btrfs_stripe_nr(chunk
, i
);
3279 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
3286 /* [pstart, pend) */
3287 static int chunk_drange_filter(struct extent_buffer
*leaf
,
3288 struct btrfs_chunk
*chunk
,
3290 struct btrfs_balance_args
*bargs
)
3292 struct btrfs_stripe
*stripe
;
3293 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3299 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
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;
3310 factor
= num_stripes
;
3313 for (i
= 0; i
< num_stripes
; i
++) {
3314 stripe
= btrfs_stripe_nr(chunk
, i
);
3315 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
3318 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
3319 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
3320 stripe_length
= div_u64(stripe_length
, factor
);
3322 if (stripe_offset
< bargs
->pend
&&
3323 stripe_offset
+ stripe_length
> bargs
->pstart
)
3330 /* [vstart, vend) */
3331 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
3332 struct btrfs_chunk
*chunk
,
3334 struct btrfs_balance_args
*bargs
)
3336 if (chunk_offset
< bargs
->vend
&&
3337 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
3338 /* at least part of the chunk is inside this vrange */
3344 static int chunk_stripes_range_filter(struct extent_buffer
*leaf
,
3345 struct btrfs_chunk
*chunk
,
3346 struct btrfs_balance_args
*bargs
)
3348 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3350 if (bargs
->stripes_min
<= num_stripes
3351 && num_stripes
<= bargs
->stripes_max
)
3357 static int chunk_soft_convert_filter(u64 chunk_type
,
3358 struct btrfs_balance_args
*bargs
)
3360 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
3363 chunk_type
= chunk_to_extended(chunk_type
) &
3364 BTRFS_EXTENDED_PROFILE_MASK
;
3366 if (bargs
->target
== chunk_type
)
3372 static int should_balance_chunk(struct btrfs_root
*root
,
3373 struct extent_buffer
*leaf
,
3374 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
3376 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3377 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3378 struct btrfs_balance_args
*bargs
= NULL
;
3379 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3382 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
3383 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
3387 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3388 bargs
= &bctl
->data
;
3389 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3391 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3392 bargs
= &bctl
->meta
;
3394 /* profiles filter */
3395 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
3396 chunk_profiles_filter(chunk_type
, bargs
)) {
3401 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3402 chunk_usage_filter(fs_info
, chunk_offset
, bargs
)) {
3404 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3405 chunk_usage_range_filter(fs_info
, chunk_offset
, bargs
)) {
3410 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
3411 chunk_devid_filter(leaf
, chunk
, bargs
)) {
3415 /* drange filter, makes sense only with devid filter */
3416 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
3417 chunk_drange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3422 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
3423 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3427 /* stripes filter */
3428 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_STRIPES_RANGE
) &&
3429 chunk_stripes_range_filter(leaf
, chunk
, bargs
)) {
3433 /* soft profile changing mode */
3434 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
3435 chunk_soft_convert_filter(chunk_type
, bargs
)) {
3440 * limited by count, must be the last filter
3442 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT
)) {
3443 if (bargs
->limit
== 0)
3447 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT_RANGE
)) {
3449 * Same logic as the 'limit' filter; the minimum cannot be
3450 * determined here because we do not have the global information
3451 * about the count of all chunks that satisfy the filters.
3453 if (bargs
->limit_max
== 0)
3462 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
3464 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3465 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
3466 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
3467 struct list_head
*devices
;
3468 struct btrfs_device
*device
;
3472 struct btrfs_chunk
*chunk
;
3473 struct btrfs_path
*path
= NULL
;
3474 struct btrfs_key key
;
3475 struct btrfs_key found_key
;
3476 struct btrfs_trans_handle
*trans
;
3477 struct extent_buffer
*leaf
;
3480 int enospc_errors
= 0;
3481 bool counting
= true;
3482 /* The single value limit and min/max limits use the same bytes in the */
3483 u64 limit_data
= bctl
->data
.limit
;
3484 u64 limit_meta
= bctl
->meta
.limit
;
3485 u64 limit_sys
= bctl
->sys
.limit
;
3489 int chunk_reserved
= 0;
3492 /* step one make some room on all the devices */
3493 devices
= &fs_info
->fs_devices
->devices
;
3494 list_for_each_entry(device
, devices
, dev_list
) {
3495 old_size
= btrfs_device_get_total_bytes(device
);
3496 size_to_free
= div_factor(old_size
, 1);
3497 size_to_free
= min_t(u64
, size_to_free
, SZ_1M
);
3498 if (!device
->writeable
||
3499 btrfs_device_get_total_bytes(device
) -
3500 btrfs_device_get_bytes_used(device
) > size_to_free
||
3501 device
->is_tgtdev_for_dev_replace
)
3504 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
3508 /* btrfs_shrink_device never returns ret > 0 */
3513 trans
= btrfs_start_transaction(dev_root
, 0);
3514 if (IS_ERR(trans
)) {
3515 ret
= PTR_ERR(trans
);
3516 btrfs_info_in_rcu(fs_info
,
3517 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3518 rcu_str_deref(device
->name
), ret
,
3519 old_size
, old_size
- size_to_free
);
3523 ret
= btrfs_grow_device(trans
, device
, old_size
);
3525 btrfs_end_transaction(trans
, dev_root
);
3526 /* btrfs_grow_device never returns ret > 0 */
3528 btrfs_info_in_rcu(fs_info
,
3529 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3530 rcu_str_deref(device
->name
), ret
,
3531 old_size
, old_size
- size_to_free
);
3535 btrfs_end_transaction(trans
, dev_root
);
3538 /* step two, relocate all the chunks */
3539 path
= btrfs_alloc_path();
3545 /* zero out stat counters */
3546 spin_lock(&fs_info
->balance_lock
);
3547 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
3548 spin_unlock(&fs_info
->balance_lock
);
3552 * The single value limit and min/max limits use the same bytes
3555 bctl
->data
.limit
= limit_data
;
3556 bctl
->meta
.limit
= limit_meta
;
3557 bctl
->sys
.limit
= limit_sys
;
3559 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3560 key
.offset
= (u64
)-1;
3561 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3564 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
3565 atomic_read(&fs_info
->balance_cancel_req
)) {
3570 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
3571 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
3573 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3578 * this shouldn't happen, it means the last relocate
3582 BUG(); /* FIXME break ? */
3584 ret
= btrfs_previous_item(chunk_root
, path
, 0,
3585 BTRFS_CHUNK_ITEM_KEY
);
3587 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3592 leaf
= path
->nodes
[0];
3593 slot
= path
->slots
[0];
3594 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3596 if (found_key
.objectid
!= key
.objectid
) {
3597 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3601 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
3602 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3605 spin_lock(&fs_info
->balance_lock
);
3606 bctl
->stat
.considered
++;
3607 spin_unlock(&fs_info
->balance_lock
);
3610 ret
= should_balance_chunk(chunk_root
, leaf
, chunk
,
3613 btrfs_release_path(path
);
3615 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3620 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3621 spin_lock(&fs_info
->balance_lock
);
3622 bctl
->stat
.expected
++;
3623 spin_unlock(&fs_info
->balance_lock
);
3625 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3627 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3629 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3636 * Apply limit_min filter, no need to check if the LIMITS
3637 * filter is used, limit_min is 0 by default
3639 if (((chunk_type
& BTRFS_BLOCK_GROUP_DATA
) &&
3640 count_data
< bctl
->data
.limit_min
)
3641 || ((chunk_type
& BTRFS_BLOCK_GROUP_METADATA
) &&
3642 count_meta
< bctl
->meta
.limit_min
)
3643 || ((chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) &&
3644 count_sys
< bctl
->sys
.limit_min
)) {
3645 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3649 ASSERT(fs_info
->data_sinfo
);
3650 spin_lock(&fs_info
->data_sinfo
->lock
);
3651 bytes_used
= fs_info
->data_sinfo
->bytes_used
;
3652 spin_unlock(&fs_info
->data_sinfo
->lock
);
3654 if ((chunk_type
& BTRFS_BLOCK_GROUP_DATA
) &&
3655 !chunk_reserved
&& !bytes_used
) {
3656 trans
= btrfs_start_transaction(chunk_root
, 0);
3657 if (IS_ERR(trans
)) {
3658 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3659 ret
= PTR_ERR(trans
);
3663 ret
= btrfs_force_chunk_alloc(trans
, chunk_root
,
3664 BTRFS_BLOCK_GROUP_DATA
);
3665 btrfs_end_transaction(trans
, chunk_root
);
3667 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3673 ret
= btrfs_relocate_chunk(fs_info
, found_key
.offset
);
3674 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3675 if (ret
&& ret
!= -ENOSPC
)
3677 if (ret
== -ENOSPC
) {
3680 spin_lock(&fs_info
->balance_lock
);
3681 bctl
->stat
.completed
++;
3682 spin_unlock(&fs_info
->balance_lock
);
3685 if (found_key
.offset
== 0)
3687 key
.offset
= found_key
.offset
- 1;
3691 btrfs_release_path(path
);
3696 btrfs_free_path(path
);
3697 if (enospc_errors
) {
3698 btrfs_info(fs_info
, "%d enospc errors during balance",
3708 * alloc_profile_is_valid - see if a given profile is valid and reduced
3709 * @flags: profile to validate
3710 * @extended: if true @flags is treated as an extended profile
3712 static int alloc_profile_is_valid(u64 flags
, int extended
)
3714 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
3715 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
3717 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
3719 /* 1) check that all other bits are zeroed */
3723 /* 2) see if profile is reduced */
3725 return !extended
; /* "0" is valid for usual profiles */
3727 /* true if exactly one bit set */
3728 return (flags
& (flags
- 1)) == 0;
3731 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
3733 /* cancel requested || normal exit path */
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);
3739 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
3743 unset_balance_control(fs_info
);
3744 ret
= del_balance_item(fs_info
);
3746 btrfs_handle_fs_error(fs_info
, ret
, NULL
);
3748 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3751 /* Non-zero return value signifies invalidity */
3752 static inline int validate_convert_profile(struct btrfs_balance_args
*bctl_arg
,
3755 return ((bctl_arg
->flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3756 (!alloc_profile_is_valid(bctl_arg
->target
, 1) ||
3757 (bctl_arg
->target
& ~allowed
)));
3761 * Should be called with both balance and volume mutexes held
3763 int btrfs_balance(struct btrfs_balance_control
*bctl
,
3764 struct btrfs_ioctl_balance_args
*bargs
)
3766 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3773 if (btrfs_fs_closing(fs_info
) ||
3774 atomic_read(&fs_info
->balance_pause_req
) ||
3775 atomic_read(&fs_info
->balance_cancel_req
)) {
3780 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
3781 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
3785 * In case of mixed groups both data and meta should be picked,
3786 * and identical options should be given for both of them.
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
))) {
3794 "with mixed groups data and metadata balance options must be the same");
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);
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
)) {
3817 "unable to start balance with target data profile %llu",
3822 if (validate_convert_profile(&bctl
->meta
, allowed
)) {
3824 "unable to start balance with target metadata profile %llu",
3829 if (validate_convert_profile(&bctl
->sys
, allowed
)) {
3831 "unable to start balance with target system profile %llu",
3837 /* allow to reduce meta or sys integrity only if force set */
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
;
3843 seq
= read_seqbegin(&fs_info
->profiles_lock
);
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
) {
3853 "force reducing metadata integrity");
3856 "balance will reduce metadata integrity, use force if you want this");
3861 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
3863 if (btrfs_get_num_tolerated_disk_barrier_failures(bctl
->meta
.target
) <
3864 btrfs_get_num_tolerated_disk_barrier_failures(bctl
->data
.target
)) {
3866 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3867 bctl
->meta
.target
, bctl
->data
.target
);
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(
3877 ret
= insert_balance_item(fs_info
, bctl
);
3878 if (ret
&& ret
!= -EEXIST
)
3881 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
3882 BUG_ON(ret
== -EEXIST
);
3883 set_balance_control(bctl
);
3885 BUG_ON(ret
!= -EEXIST
);
3886 spin_lock(&fs_info
->balance_lock
);
3887 update_balance_args(bctl
);
3888 spin_unlock(&fs_info
->balance_lock
);
3891 atomic_inc(&fs_info
->balance_running
);
3892 mutex_unlock(&fs_info
->balance_mutex
);
3894 ret
= __btrfs_balance(fs_info
);
3896 mutex_lock(&fs_info
->balance_mutex
);
3897 atomic_dec(&fs_info
->balance_running
);
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
);
3905 memset(bargs
, 0, sizeof(*bargs
));
3906 update_ioctl_balance_args(fs_info
, 0, bargs
);
3909 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
3910 balance_need_close(fs_info
)) {
3911 __cancel_balance(fs_info
);
3914 wake_up(&fs_info
->balance_wait_q
);
3918 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
3919 __cancel_balance(fs_info
);
3922 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3927 static int balance_kthread(void *data
)
3929 struct btrfs_fs_info
*fs_info
= data
;
3932 mutex_lock(&fs_info
->volume_mutex
);
3933 mutex_lock(&fs_info
->balance_mutex
);
3935 if (fs_info
->balance_ctl
) {
3936 btrfs_info(fs_info
, "continuing balance");
3937 ret
= btrfs_balance(fs_info
->balance_ctl
, NULL
);
3940 mutex_unlock(&fs_info
->balance_mutex
);
3941 mutex_unlock(&fs_info
->volume_mutex
);
3946 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
3948 struct task_struct
*tsk
;
3950 spin_lock(&fs_info
->balance_lock
);
3951 if (!fs_info
->balance_ctl
) {
3952 spin_unlock(&fs_info
->balance_lock
);
3955 spin_unlock(&fs_info
->balance_lock
);
3957 if (btrfs_test_opt(fs_info
, SKIP_BALANCE
)) {
3958 btrfs_info(fs_info
, "force skipping balance");
3962 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
3963 return PTR_ERR_OR_ZERO(tsk
);
3966 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
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
;
3976 path
= btrfs_alloc_path();
3980 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3981 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3984 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
3987 if (ret
> 0) { /* ret = -ENOENT; */
3992 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
3998 leaf
= path
->nodes
[0];
3999 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
4001 bctl
->fs_info
= fs_info
;
4002 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
4003 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
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
);
4012 WARN_ON(atomic_xchg(&fs_info
->mutually_exclusive_operation_running
, 1));
4014 mutex_lock(&fs_info
->volume_mutex
);
4015 mutex_lock(&fs_info
->balance_mutex
);
4017 set_balance_control(bctl
);
4019 mutex_unlock(&fs_info
->balance_mutex
);
4020 mutex_unlock(&fs_info
->volume_mutex
);
4022 btrfs_free_path(path
);
4026 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
4030 mutex_lock(&fs_info
->balance_mutex
);
4031 if (!fs_info
->balance_ctl
) {
4032 mutex_unlock(&fs_info
->balance_mutex
);
4036 if (atomic_read(&fs_info
->balance_running
)) {
4037 atomic_inc(&fs_info
->balance_pause_req
);
4038 mutex_unlock(&fs_info
->balance_mutex
);
4040 wait_event(fs_info
->balance_wait_q
,
4041 atomic_read(&fs_info
->balance_running
) == 0);
4043 mutex_lock(&fs_info
->balance_mutex
);
4044 /* we are good with balance_ctl ripped off from under us */
4045 BUG_ON(atomic_read(&fs_info
->balance_running
));
4046 atomic_dec(&fs_info
->balance_pause_req
);
4051 mutex_unlock(&fs_info
->balance_mutex
);
4055 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
4057 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
4060 mutex_lock(&fs_info
->balance_mutex
);
4061 if (!fs_info
->balance_ctl
) {
4062 mutex_unlock(&fs_info
->balance_mutex
);
4066 atomic_inc(&fs_info
->balance_cancel_req
);
4068 * if we are running just wait and return, balance item is
4069 * deleted in btrfs_balance in this case
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
);
4077 /* __cancel_balance needs volume_mutex */
4078 mutex_unlock(&fs_info
->balance_mutex
);
4079 mutex_lock(&fs_info
->volume_mutex
);
4080 mutex_lock(&fs_info
->balance_mutex
);
4082 if (fs_info
->balance_ctl
)
4083 __cancel_balance(fs_info
);
4085 mutex_unlock(&fs_info
->volume_mutex
);
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
);
4094 static int btrfs_uuid_scan_kthread(void *data
)
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
;
4102 struct extent_buffer
*eb
;
4104 struct btrfs_root_item root_item
;
4106 struct btrfs_trans_handle
*trans
= NULL
;
4108 path
= btrfs_alloc_path();
4115 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4118 max_key
.objectid
= (u64
)-1;
4119 max_key
.type
= BTRFS_ROOT_ITEM_KEY
;
4120 max_key
.offset
= (u64
)-1;
4123 ret
= btrfs_search_forward(root
, &key
, path
, 0);
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
)
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
))
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)
4148 if (!btrfs_is_empty_uuid(root_item
.uuid
) ||
4149 !btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4153 btrfs_release_path(path
);
4155 * 1 - subvol uuid item
4156 * 1 - received_subvol uuid item
4158 trans
= btrfs_start_transaction(fs_info
->uuid_root
, 2);
4159 if (IS_ERR(trans
)) {
4160 ret
= PTR_ERR(trans
);
4168 if (!btrfs_is_empty_uuid(root_item
.uuid
)) {
4169 ret
= btrfs_uuid_tree_add(trans
, fs_info
,
4171 BTRFS_UUID_KEY_SUBVOL
,
4174 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4180 if (!btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4181 ret
= btrfs_uuid_tree_add(trans
, fs_info
,
4182 root_item
.received_uuid
,
4183 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
4186 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4194 ret
= btrfs_end_transaction(trans
, fs_info
->uuid_root
);
4200 btrfs_release_path(path
);
4201 if (key
.offset
< (u64
)-1) {
4203 } else if (key
.type
< BTRFS_ROOT_ITEM_KEY
) {
4205 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4206 } else if (key
.objectid
< (u64
)-1) {
4208 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4217 btrfs_free_path(path
);
4218 if (trans
&& !IS_ERR(trans
))
4219 btrfs_end_transaction(trans
, fs_info
->uuid_root
);
4221 btrfs_warn(fs_info
, "btrfs_uuid_scan_kthread failed %d", ret
);
4223 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN
, &fs_info
->flags
);
4224 up(&fs_info
->uuid_tree_rescan_sem
);
4229 * Callback for btrfs_uuid_tree_iterate().
4231 * 0 check succeeded, the entry is not outdated.
4232 * < 0 if an error occurred.
4233 * > 0 if the check failed, which means the caller shall remove the entry.
4235 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info
*fs_info
,
4236 u8
*uuid
, u8 type
, u64 subid
)
4238 struct btrfs_key key
;
4240 struct btrfs_root
*subvol_root
;
4242 if (type
!= BTRFS_UUID_KEY_SUBVOL
&&
4243 type
!= BTRFS_UUID_KEY_RECEIVED_SUBVOL
)
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
);
4258 case BTRFS_UUID_KEY_SUBVOL
:
4259 if (memcmp(uuid
, subvol_root
->root_item
.uuid
, BTRFS_UUID_SIZE
))
4262 case BTRFS_UUID_KEY_RECEIVED_SUBVOL
:
4263 if (memcmp(uuid
, subvol_root
->root_item
.received_uuid
,
4273 static int btrfs_uuid_rescan_kthread(void *data
)
4275 struct btrfs_fs_info
*fs_info
= (struct btrfs_fs_info
*)data
;
4279 * 1st step is to iterate through the existing UUID tree and
4280 * to delete all entries that contain outdated data.
4281 * 2nd step is to add all missing entries to the UUID tree.
4283 ret
= btrfs_uuid_tree_iterate(fs_info
, btrfs_check_uuid_tree_entry
);
4285 btrfs_warn(fs_info
, "iterating uuid_tree failed %d", ret
);
4286 up(&fs_info
->uuid_tree_rescan_sem
);
4289 return btrfs_uuid_scan_kthread(data
);
4292 int btrfs_create_uuid_tree(struct btrfs_fs_info
*fs_info
)
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
;
4304 trans
= btrfs_start_transaction(tree_root
, 2);
4306 return PTR_ERR(trans
);
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
);
4317 fs_info
->uuid_root
= uuid_root
;
4319 ret
= btrfs_commit_transaction(trans
, tree_root
);
4323 down(&fs_info
->uuid_tree_rescan_sem
);
4324 task
= kthread_run(btrfs_uuid_scan_kthread
, fs_info
, "btrfs-uuid");
4326 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4327 btrfs_warn(fs_info
, "failed to start uuid_scan task");
4328 up(&fs_info
->uuid_tree_rescan_sem
);
4329 return PTR_ERR(task
);
4335 int btrfs_check_uuid_tree(struct btrfs_fs_info
*fs_info
)
4337 struct task_struct
*task
;
4339 down(&fs_info
->uuid_tree_rescan_sem
);
4340 task
= kthread_run(btrfs_uuid_rescan_kthread
, fs_info
, "btrfs-uuid");
4342 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4343 btrfs_warn(fs_info
, "failed to start uuid_rescan task");
4344 up(&fs_info
->uuid_tree_rescan_sem
);
4345 return PTR_ERR(task
);
4352 * shrinking a device means finding all of the device extents past
4353 * the new size, and then following the back refs to the chunks.
4354 * The chunk relocation code actually frees the device extent
4356 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
4358 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
4359 struct btrfs_root
*root
= fs_info
->dev_root
;
4360 struct btrfs_trans_handle
*trans
;
4361 struct btrfs_dev_extent
*dev_extent
= NULL
;
4362 struct btrfs_path
*path
;
4368 bool retried
= false;
4369 bool checked_pending_chunks
= false;
4370 struct extent_buffer
*l
;
4371 struct btrfs_key key
;
4372 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
4373 u64 old_total
= btrfs_super_total_bytes(super_copy
);
4374 u64 old_size
= btrfs_device_get_total_bytes(device
);
4375 u64 diff
= old_size
- new_size
;
4377 if (device
->is_tgtdev_for_dev_replace
)
4380 path
= btrfs_alloc_path();
4384 path
->reada
= READA_FORWARD
;
4386 lock_chunks(fs_info
);
4388 btrfs_device_set_total_bytes(device
, new_size
);
4389 if (device
->writeable
) {
4390 device
->fs_devices
->total_rw_bytes
-= diff
;
4391 spin_lock(&fs_info
->free_chunk_lock
);
4392 fs_info
->free_chunk_space
-= diff
;
4393 spin_unlock(&fs_info
->free_chunk_lock
);
4395 unlock_chunks(fs_info
);
4398 key
.objectid
= device
->devid
;
4399 key
.offset
= (u64
)-1;
4400 key
.type
= BTRFS_DEV_EXTENT_KEY
;
4403 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
4404 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4406 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4410 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
4412 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4417 btrfs_release_path(path
);
4422 slot
= path
->slots
[0];
4423 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
4425 if (key
.objectid
!= device
->devid
) {
4426 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4427 btrfs_release_path(path
);
4431 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
4432 length
= btrfs_dev_extent_length(l
, dev_extent
);
4434 if (key
.offset
+ length
<= new_size
) {
4435 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4436 btrfs_release_path(path
);
4440 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
4441 btrfs_release_path(path
);
4443 ret
= btrfs_relocate_chunk(fs_info
, chunk_offset
);
4444 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4445 if (ret
&& ret
!= -ENOSPC
)
4449 } while (key
.offset
-- > 0);
4451 if (failed
&& !retried
) {
4455 } else if (failed
&& retried
) {
4460 /* Shrinking succeeded, else we would be at "done". */
4461 trans
= btrfs_start_transaction(root
, 0);
4462 if (IS_ERR(trans
)) {
4463 ret
= PTR_ERR(trans
);
4467 lock_chunks(fs_info
);
4470 * We checked in the above loop all device extents that were already in
4471 * the device tree. However before we have updated the device's
4472 * total_bytes to the new size, we might have had chunk allocations that
4473 * have not complete yet (new block groups attached to transaction
4474 * handles), and therefore their device extents were not yet in the
4475 * device tree and we missed them in the loop above. So if we have any
4476 * pending chunk using a device extent that overlaps the device range
4477 * that we can not use anymore, commit the current transaction and
4478 * repeat the search on the device tree - this way we guarantee we will
4479 * not have chunks using device extents that end beyond 'new_size'.
4481 if (!checked_pending_chunks
) {
4482 u64 start
= new_size
;
4483 u64 len
= old_size
- new_size
;
4485 if (contains_pending_extent(trans
->transaction
, device
,
4487 unlock_chunks(fs_info
);
4488 checked_pending_chunks
= true;
4491 ret
= btrfs_commit_transaction(trans
, root
);
4498 btrfs_device_set_disk_total_bytes(device
, new_size
);
4499 if (list_empty(&device
->resized_list
))
4500 list_add_tail(&device
->resized_list
,
4501 &fs_info
->fs_devices
->resized_devices
);
4503 WARN_ON(diff
> old_total
);
4504 btrfs_set_super_total_bytes(super_copy
, old_total
- diff
);
4505 unlock_chunks(fs_info
);
4507 /* Now btrfs_update_device() will change the on-disk size. */
4508 ret
= btrfs_update_device(trans
, device
);
4509 btrfs_end_transaction(trans
, root
);
4511 btrfs_free_path(path
);
4513 lock_chunks(fs_info
);
4514 btrfs_device_set_total_bytes(device
, old_size
);
4515 if (device
->writeable
)
4516 device
->fs_devices
->total_rw_bytes
+= diff
;
4517 spin_lock(&fs_info
->free_chunk_lock
);
4518 fs_info
->free_chunk_space
+= diff
;
4519 spin_unlock(&fs_info
->free_chunk_lock
);
4520 unlock_chunks(fs_info
);
4525 static int btrfs_add_system_chunk(struct btrfs_root
*root
,
4526 struct btrfs_key
*key
,
4527 struct btrfs_chunk
*chunk
, int item_size
)
4529 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4530 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
4531 struct btrfs_disk_key disk_key
;
4535 lock_chunks(fs_info
);
4536 array_size
= btrfs_super_sys_array_size(super_copy
);
4537 if (array_size
+ item_size
+ sizeof(disk_key
)
4538 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4539 unlock_chunks(fs_info
);
4543 ptr
= super_copy
->sys_chunk_array
+ array_size
;
4544 btrfs_cpu_key_to_disk(&disk_key
, key
);
4545 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
4546 ptr
+= sizeof(disk_key
);
4547 memcpy(ptr
, chunk
, item_size
);
4548 item_size
+= sizeof(disk_key
);
4549 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
4550 unlock_chunks(fs_info
);
4556 * sort the devices in descending order by max_avail, total_avail
4558 static int btrfs_cmp_device_info(const void *a
, const void *b
)
4560 const struct btrfs_device_info
*di_a
= a
;
4561 const struct btrfs_device_info
*di_b
= b
;
4563 if (di_a
->max_avail
> di_b
->max_avail
)
4565 if (di_a
->max_avail
< di_b
->max_avail
)
4567 if (di_a
->total_avail
> di_b
->total_avail
)
4569 if (di_a
->total_avail
< di_b
->total_avail
)
4574 static u32
find_raid56_stripe_len(u32 data_devices
, u32 dev_stripe_target
)
4576 /* TODO allow them to set a preferred stripe size */
4580 static void check_raid56_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
4582 if (!(type
& BTRFS_BLOCK_GROUP_RAID56_MASK
))
4585 btrfs_set_fs_incompat(info
, RAID56
);
4588 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info) \
4589 - sizeof(struct btrfs_chunk)) \
4590 / sizeof(struct btrfs_stripe) + 1)
4592 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4593 - 2 * sizeof(struct btrfs_disk_key) \
4594 - 2 * sizeof(struct btrfs_chunk)) \
4595 / sizeof(struct btrfs_stripe) + 1)
4597 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4598 struct btrfs_root
*extent_root
, u64 start
,
4601 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
4602 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
4603 struct list_head
*cur
;
4604 struct map_lookup
*map
= NULL
;
4605 struct extent_map_tree
*em_tree
;
4606 struct extent_map
*em
;
4607 struct btrfs_device_info
*devices_info
= NULL
;
4609 int num_stripes
; /* total number of stripes to allocate */
4610 int data_stripes
; /* number of stripes that count for
4612 int sub_stripes
; /* sub_stripes info for map */
4613 int dev_stripes
; /* stripes per dev */
4614 int devs_max
; /* max devs to use */
4615 int devs_min
; /* min devs needed */
4616 int devs_increment
; /* ndevs has to be a multiple of this */
4617 int ncopies
; /* how many copies to data has */
4619 u64 max_stripe_size
;
4623 u64 raid_stripe_len
= BTRFS_STRIPE_LEN
;
4629 BUG_ON(!alloc_profile_is_valid(type
, 0));
4631 if (list_empty(&fs_devices
->alloc_list
))
4634 index
= __get_raid_index(type
);
4636 sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
4637 dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
4638 devs_max
= btrfs_raid_array
[index
].devs_max
;
4639 devs_min
= btrfs_raid_array
[index
].devs_min
;
4640 devs_increment
= btrfs_raid_array
[index
].devs_increment
;
4641 ncopies
= btrfs_raid_array
[index
].ncopies
;
4643 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
4644 max_stripe_size
= SZ_1G
;
4645 max_chunk_size
= 10 * max_stripe_size
;
4647 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4648 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
4649 /* for larger filesystems, use larger metadata chunks */
4650 if (fs_devices
->total_rw_bytes
> 50ULL * SZ_1G
)
4651 max_stripe_size
= SZ_1G
;
4653 max_stripe_size
= SZ_256M
;
4654 max_chunk_size
= max_stripe_size
;
4656 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4657 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4658 max_stripe_size
= SZ_32M
;
4659 max_chunk_size
= 2 * max_stripe_size
;
4661 devs_max
= BTRFS_MAX_DEVS_SYS_CHUNK
;
4663 btrfs_err(info
, "invalid chunk type 0x%llx requested",
4668 /* we don't want a chunk larger than 10% of writeable space */
4669 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
4672 devices_info
= kcalloc(fs_devices
->rw_devices
, sizeof(*devices_info
),
4677 cur
= fs_devices
->alloc_list
.next
;
4680 * in the first pass through the devices list, we gather information
4681 * about the available holes on each device.
4684 while (cur
!= &fs_devices
->alloc_list
) {
4685 struct btrfs_device
*device
;
4689 device
= list_entry(cur
, struct btrfs_device
, dev_alloc_list
);
4693 if (!device
->writeable
) {
4695 "BTRFS: read-only device in alloc_list\n");
4699 if (!device
->in_fs_metadata
||
4700 device
->is_tgtdev_for_dev_replace
)
4703 if (device
->total_bytes
> device
->bytes_used
)
4704 total_avail
= device
->total_bytes
- device
->bytes_used
;
4708 /* If there is no space on this device, skip it. */
4709 if (total_avail
== 0)
4712 ret
= find_free_dev_extent(trans
, device
,
4713 max_stripe_size
* dev_stripes
,
4714 &dev_offset
, &max_avail
);
4715 if (ret
&& ret
!= -ENOSPC
)
4719 max_avail
= max_stripe_size
* dev_stripes
;
4721 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
4724 if (ndevs
== fs_devices
->rw_devices
) {
4725 WARN(1, "%s: found more than %llu devices\n",
4726 __func__
, fs_devices
->rw_devices
);
4729 devices_info
[ndevs
].dev_offset
= dev_offset
;
4730 devices_info
[ndevs
].max_avail
= max_avail
;
4731 devices_info
[ndevs
].total_avail
= total_avail
;
4732 devices_info
[ndevs
].dev
= device
;
4737 * now sort the devices by hole size / available space
4739 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
4740 btrfs_cmp_device_info
, NULL
);
4742 /* round down to number of usable stripes */
4743 ndevs
-= ndevs
% devs_increment
;
4745 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
4750 if (devs_max
&& ndevs
> devs_max
)
4753 * the primary goal is to maximize the number of stripes, so use as many
4754 * devices as possible, even if the stripes are not maximum sized.
4756 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4757 num_stripes
= ndevs
* dev_stripes
;
4760 * this will have to be fixed for RAID1 and RAID10 over
4763 data_stripes
= num_stripes
/ ncopies
;
4765 if (type
& BTRFS_BLOCK_GROUP_RAID5
) {
4766 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 1,
4768 data_stripes
= num_stripes
- 1;
4770 if (type
& BTRFS_BLOCK_GROUP_RAID6
) {
4771 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 2,
4773 data_stripes
= num_stripes
- 2;
4777 * Use the number of data stripes to figure out how big this chunk
4778 * is really going to be in terms of logical address space,
4779 * and compare that answer with the max chunk size
4781 if (stripe_size
* data_stripes
> max_chunk_size
) {
4782 u64 mask
= (1ULL << 24) - 1;
4784 stripe_size
= div_u64(max_chunk_size
, data_stripes
);
4786 /* bump the answer up to a 16MB boundary */
4787 stripe_size
= (stripe_size
+ mask
) & ~mask
;
4789 /* but don't go higher than the limits we found
4790 * while searching for free extents
4792 if (stripe_size
> devices_info
[ndevs
-1].max_avail
)
4793 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4796 stripe_size
= div_u64(stripe_size
, dev_stripes
);
4798 /* align to BTRFS_STRIPE_LEN */
4799 stripe_size
= div_u64(stripe_size
, raid_stripe_len
);
4800 stripe_size
*= raid_stripe_len
;
4802 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
4807 map
->num_stripes
= num_stripes
;
4809 for (i
= 0; i
< ndevs
; ++i
) {
4810 for (j
= 0; j
< dev_stripes
; ++j
) {
4811 int s
= i
* dev_stripes
+ j
;
4812 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
4813 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
4817 map
->sector_size
= info
->sectorsize
;
4818 map
->stripe_len
= raid_stripe_len
;
4819 map
->io_align
= raid_stripe_len
;
4820 map
->io_width
= raid_stripe_len
;
4822 map
->sub_stripes
= sub_stripes
;
4824 num_bytes
= stripe_size
* data_stripes
;
4826 trace_btrfs_chunk_alloc(info
, map
, start
, num_bytes
);
4828 em
= alloc_extent_map();
4834 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
4835 em
->map_lookup
= map
;
4837 em
->len
= num_bytes
;
4838 em
->block_start
= 0;
4839 em
->block_len
= em
->len
;
4840 em
->orig_block_len
= stripe_size
;
4842 em_tree
= &info
->mapping_tree
.map_tree
;
4843 write_lock(&em_tree
->lock
);
4844 ret
= add_extent_mapping(em_tree
, em
, 0);
4846 list_add_tail(&em
->list
, &trans
->transaction
->pending_chunks
);
4847 atomic_inc(&em
->refs
);
4849 write_unlock(&em_tree
->lock
);
4851 free_extent_map(em
);
4855 ret
= btrfs_make_block_group(trans
, extent_root
, 0, type
,
4856 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4859 goto error_del_extent
;
4861 for (i
= 0; i
< map
->num_stripes
; i
++) {
4862 num_bytes
= map
->stripes
[i
].dev
->bytes_used
+ stripe_size
;
4863 btrfs_device_set_bytes_used(map
->stripes
[i
].dev
, num_bytes
);
4866 spin_lock(&info
->free_chunk_lock
);
4867 info
->free_chunk_space
-= (stripe_size
* map
->num_stripes
);
4868 spin_unlock(&info
->free_chunk_lock
);
4870 free_extent_map(em
);
4871 check_raid56_incompat_flag(info
, type
);
4873 kfree(devices_info
);
4877 write_lock(&em_tree
->lock
);
4878 remove_extent_mapping(em_tree
, em
);
4879 write_unlock(&em_tree
->lock
);
4881 /* One for our allocation */
4882 free_extent_map(em
);
4883 /* One for the tree reference */
4884 free_extent_map(em
);
4885 /* One for the pending_chunks list reference */
4886 free_extent_map(em
);
4888 kfree(devices_info
);
4892 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
4893 struct btrfs_fs_info
*fs_info
,
4894 u64 chunk_offset
, u64 chunk_size
)
4896 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
4897 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
4898 struct btrfs_key key
;
4899 struct btrfs_device
*device
;
4900 struct btrfs_chunk
*chunk
;
4901 struct btrfs_stripe
*stripe
;
4902 struct extent_map_tree
*em_tree
;
4903 struct extent_map
*em
;
4904 struct map_lookup
*map
;
4911 em_tree
= &fs_info
->mapping_tree
.map_tree
;
4912 read_lock(&em_tree
->lock
);
4913 em
= lookup_extent_mapping(em_tree
, chunk_offset
, chunk_size
);
4914 read_unlock(&em_tree
->lock
);
4917 btrfs_crit(fs_info
, "unable to find logical %Lu len %Lu",
4918 chunk_offset
, chunk_size
);
4922 if (em
->start
!= chunk_offset
|| em
->len
!= chunk_size
) {
4924 "found a bad mapping, wanted %Lu-%Lu, found %Lu-%Lu",
4925 chunk_offset
, chunk_size
, em
->start
, em
->len
);
4926 free_extent_map(em
);
4930 map
= em
->map_lookup
;
4931 item_size
= btrfs_chunk_item_size(map
->num_stripes
);
4932 stripe_size
= em
->orig_block_len
;
4934 chunk
= kzalloc(item_size
, GFP_NOFS
);
4941 * Take the device list mutex to prevent races with the final phase of
4942 * a device replace operation that replaces the device object associated
4943 * with the map's stripes, because the device object's id can change
4944 * at any time during that final phase of the device replace operation
4945 * (dev-replace.c:btrfs_dev_replace_finishing()).
4947 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
4948 for (i
= 0; i
< map
->num_stripes
; i
++) {
4949 device
= map
->stripes
[i
].dev
;
4950 dev_offset
= map
->stripes
[i
].physical
;
4952 ret
= btrfs_update_device(trans
, device
);
4955 ret
= btrfs_alloc_dev_extent(trans
, device
,
4956 chunk_root
->root_key
.objectid
,
4957 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4958 chunk_offset
, dev_offset
,
4964 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
4968 stripe
= &chunk
->stripe
;
4969 for (i
= 0; i
< map
->num_stripes
; i
++) {
4970 device
= map
->stripes
[i
].dev
;
4971 dev_offset
= map
->stripes
[i
].physical
;
4973 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
4974 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
4975 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
4978 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
4980 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
4981 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
4982 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
4983 btrfs_set_stack_chunk_type(chunk
, map
->type
);
4984 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
4985 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
4986 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
4987 btrfs_set_stack_chunk_sector_size(chunk
, fs_info
->sectorsize
);
4988 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
4990 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
4991 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
4992 key
.offset
= chunk_offset
;
4994 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
4995 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4997 * TODO: Cleanup of inserted chunk root in case of
5000 ret
= btrfs_add_system_chunk(chunk_root
, &key
, chunk
,
5006 free_extent_map(em
);
5011 * Chunk allocation falls into two parts. The first part does works
5012 * that make the new allocated chunk useable, but not do any operation
5013 * that modifies the chunk tree. The second part does the works that
5014 * require modifying the chunk tree. This division is important for the
5015 * bootstrap process of adding storage to a seed btrfs.
5017 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
5018 struct btrfs_root
*extent_root
, u64 type
)
5020 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
5023 ASSERT(mutex_is_locked(&fs_info
->chunk_mutex
));
5024 chunk_offset
= find_next_chunk(fs_info
);
5025 return __btrfs_alloc_chunk(trans
, extent_root
, chunk_offset
, type
);
5028 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
5029 struct btrfs_root
*root
,
5030 struct btrfs_device
*device
)
5033 u64 sys_chunk_offset
;
5035 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5036 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
5039 chunk_offset
= find_next_chunk(fs_info
);
5040 alloc_profile
= btrfs_get_alloc_profile(extent_root
, 0);
5041 ret
= __btrfs_alloc_chunk(trans
, extent_root
, chunk_offset
,
5046 sys_chunk_offset
= find_next_chunk(fs_info
);
5047 alloc_profile
= btrfs_get_alloc_profile(fs_info
->chunk_root
, 0);
5048 ret
= __btrfs_alloc_chunk(trans
, extent_root
, sys_chunk_offset
,
5053 static inline int btrfs_chunk_max_errors(struct map_lookup
*map
)
5057 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
5058 BTRFS_BLOCK_GROUP_RAID10
|
5059 BTRFS_BLOCK_GROUP_RAID5
|
5060 BTRFS_BLOCK_GROUP_DUP
)) {
5062 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
) {
5071 int btrfs_chunk_readonly(struct btrfs_root
*root
, u64 chunk_offset
)
5073 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5074 struct extent_map
*em
;
5075 struct map_lookup
*map
;
5076 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
5081 read_lock(&map_tree
->map_tree
.lock
);
5082 em
= lookup_extent_mapping(&map_tree
->map_tree
, chunk_offset
, 1);
5083 read_unlock(&map_tree
->map_tree
.lock
);
5087 map
= em
->map_lookup
;
5088 for (i
= 0; i
< map
->num_stripes
; i
++) {
5089 if (map
->stripes
[i
].dev
->missing
) {
5094 if (!map
->stripes
[i
].dev
->writeable
) {
5101 * If the number of missing devices is larger than max errors,
5102 * we can not write the data into that chunk successfully, so
5105 if (miss_ndevs
> btrfs_chunk_max_errors(map
))
5108 free_extent_map(em
);
5112 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
5114 extent_map_tree_init(&tree
->map_tree
);
5117 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
5119 struct extent_map
*em
;
5122 write_lock(&tree
->map_tree
.lock
);
5123 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
5125 remove_extent_mapping(&tree
->map_tree
, em
);
5126 write_unlock(&tree
->map_tree
.lock
);
5130 free_extent_map(em
);
5131 /* once for the tree */
5132 free_extent_map(em
);
5136 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
5138 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
5139 struct extent_map
*em
;
5140 struct map_lookup
*map
;
5141 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5144 read_lock(&em_tree
->lock
);
5145 em
= lookup_extent_mapping(em_tree
, logical
, len
);
5146 read_unlock(&em_tree
->lock
);
5149 * We could return errors for these cases, but that could get ugly and
5150 * we'd probably do the same thing which is just not do anything else
5151 * and exit, so return 1 so the callers don't try to use other copies.
5154 btrfs_crit(fs_info
, "No mapping for %Lu-%Lu", logical
,
5159 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
5160 btrfs_crit(fs_info
, "Invalid mapping for %Lu-%Lu, got %Lu-%Lu",
5161 logical
, logical
+len
, em
->start
,
5162 em
->start
+ em
->len
);
5163 free_extent_map(em
);
5167 map
= em
->map_lookup
;
5168 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
5169 ret
= map
->num_stripes
;
5170 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5171 ret
= map
->sub_stripes
;
5172 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID5
)
5174 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5178 free_extent_map(em
);
5180 btrfs_dev_replace_lock(&fs_info
->dev_replace
, 0);
5181 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
))
5183 btrfs_dev_replace_unlock(&fs_info
->dev_replace
, 0);
5188 unsigned long btrfs_full_stripe_len(struct btrfs_root
*root
,
5189 struct btrfs_mapping_tree
*map_tree
,
5192 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5193 struct extent_map
*em
;
5194 struct map_lookup
*map
;
5195 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5196 unsigned long len
= fs_info
->sectorsize
;
5198 read_lock(&em_tree
->lock
);
5199 em
= lookup_extent_mapping(em_tree
, logical
, len
);
5200 read_unlock(&em_tree
->lock
);
5203 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
5204 map
= em
->map_lookup
;
5205 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5206 len
= map
->stripe_len
* nr_data_stripes(map
);
5207 free_extent_map(em
);
5211 int btrfs_is_parity_mirror(struct btrfs_mapping_tree
*map_tree
,
5212 u64 logical
, u64 len
, int mirror_num
)
5214 struct extent_map
*em
;
5215 struct map_lookup
*map
;
5216 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5219 read_lock(&em_tree
->lock
);
5220 em
= lookup_extent_mapping(em_tree
, logical
, len
);
5221 read_unlock(&em_tree
->lock
);
5224 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
5225 map
= em
->map_lookup
;
5226 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5228 free_extent_map(em
);
5232 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
5233 struct map_lookup
*map
, int first
, int num
,
5234 int optimal
, int dev_replace_is_ongoing
)
5238 struct btrfs_device
*srcdev
;
5240 if (dev_replace_is_ongoing
&&
5241 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
5242 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
5243 srcdev
= fs_info
->dev_replace
.srcdev
;
5248 * try to avoid the drive that is the source drive for a
5249 * dev-replace procedure, only choose it if no other non-missing
5250 * mirror is available
5252 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
5253 if (map
->stripes
[optimal
].dev
->bdev
&&
5254 (tolerance
|| map
->stripes
[optimal
].dev
!= srcdev
))
5256 for (i
= first
; i
< first
+ num
; i
++) {
5257 if (map
->stripes
[i
].dev
->bdev
&&
5258 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
5263 /* we couldn't find one that doesn't fail. Just return something
5264 * and the io error handling code will clean up eventually
5269 static inline int parity_smaller(u64 a
, u64 b
)
5274 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5275 static void sort_parity_stripes(struct btrfs_bio
*bbio
, int num_stripes
)
5277 struct btrfs_bio_stripe s
;
5284 for (i
= 0; i
< num_stripes
- 1; i
++) {
5285 if (parity_smaller(bbio
->raid_map
[i
],
5286 bbio
->raid_map
[i
+1])) {
5287 s
= bbio
->stripes
[i
];
5288 l
= bbio
->raid_map
[i
];
5289 bbio
->stripes
[i
] = bbio
->stripes
[i
+1];
5290 bbio
->raid_map
[i
] = bbio
->raid_map
[i
+1];
5291 bbio
->stripes
[i
+1] = s
;
5292 bbio
->raid_map
[i
+1] = l
;
5300 static struct btrfs_bio
*alloc_btrfs_bio(int total_stripes
, int real_stripes
)
5302 struct btrfs_bio
*bbio
= kzalloc(
5303 /* the size of the btrfs_bio */
5304 sizeof(struct btrfs_bio
) +
5305 /* plus the variable array for the stripes */
5306 sizeof(struct btrfs_bio_stripe
) * (total_stripes
) +
5307 /* plus the variable array for the tgt dev */
5308 sizeof(int) * (real_stripes
) +
5310 * plus the raid_map, which includes both the tgt dev
5313 sizeof(u64
) * (total_stripes
),
5314 GFP_NOFS
|__GFP_NOFAIL
);
5316 atomic_set(&bbio
->error
, 0);
5317 atomic_set(&bbio
->refs
, 1);
5322 void btrfs_get_bbio(struct btrfs_bio
*bbio
)
5324 WARN_ON(!atomic_read(&bbio
->refs
));
5325 atomic_inc(&bbio
->refs
);
5328 void btrfs_put_bbio(struct btrfs_bio
*bbio
)
5332 if (atomic_dec_and_test(&bbio
->refs
))
5336 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
,
5337 enum btrfs_map_op op
,
5338 u64 logical
, u64
*length
,
5339 struct btrfs_bio
**bbio_ret
,
5340 int mirror_num
, int need_raid_map
)
5342 struct extent_map
*em
;
5343 struct map_lookup
*map
;
5344 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
5345 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5348 u64 stripe_end_offset
;
5358 int tgtdev_indexes
= 0;
5359 struct btrfs_bio
*bbio
= NULL
;
5360 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
5361 int dev_replace_is_ongoing
= 0;
5362 int num_alloc_stripes
;
5363 int patch_the_first_stripe_for_dev_replace
= 0;
5364 u64 physical_to_patch_in_first_stripe
= 0;
5365 u64 raid56_full_stripe_start
= (u64
)-1;
5367 read_lock(&em_tree
->lock
);
5368 em
= lookup_extent_mapping(em_tree
, logical
, *length
);
5369 read_unlock(&em_tree
->lock
);
5372 btrfs_crit(fs_info
, "unable to find logical %llu len %llu",
5377 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
5379 "found a bad mapping, wanted %Lu, found %Lu-%Lu",
5380 logical
, em
->start
, em
->start
+ em
->len
);
5381 free_extent_map(em
);
5385 map
= em
->map_lookup
;
5386 offset
= logical
- em
->start
;
5388 stripe_len
= map
->stripe_len
;
5391 * stripe_nr counts the total number of stripes we have to stride
5392 * to get to this block
5394 stripe_nr
= div64_u64(stripe_nr
, stripe_len
);
5396 stripe_offset
= stripe_nr
* stripe_len
;
5397 if (offset
< stripe_offset
) {
5399 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5400 stripe_offset
, offset
, em
->start
, logical
,
5402 free_extent_map(em
);
5406 /* stripe_offset is the offset of this block in its stripe*/
5407 stripe_offset
= offset
- stripe_offset
;
5409 /* if we're here for raid56, we need to know the stripe aligned start */
5410 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5411 unsigned long full_stripe_len
= stripe_len
* nr_data_stripes(map
);
5412 raid56_full_stripe_start
= offset
;
5414 /* allow a write of a full stripe, but make sure we don't
5415 * allow straddling of stripes
5417 raid56_full_stripe_start
= div64_u64(raid56_full_stripe_start
,
5419 raid56_full_stripe_start
*= full_stripe_len
;
5422 if (op
== BTRFS_MAP_DISCARD
) {
5423 /* we don't discard raid56 yet */
5424 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5428 *length
= min_t(u64
, em
->len
- offset
, *length
);
5429 } else if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
5431 /* For writes to RAID[56], allow a full stripeset across all disks.
5432 For other RAID types and for RAID[56] reads, just allow a single
5433 stripe (on a single disk). */
5434 if ((map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
5435 (op
== BTRFS_MAP_WRITE
)) {
5436 max_len
= stripe_len
* nr_data_stripes(map
) -
5437 (offset
- raid56_full_stripe_start
);
5439 /* we limit the length of each bio to what fits in a stripe */
5440 max_len
= stripe_len
- stripe_offset
;
5442 *length
= min_t(u64
, em
->len
- offset
, max_len
);
5444 *length
= em
->len
- offset
;
5447 /* This is for when we're called from btrfs_merge_bio_hook() and all
5448 it cares about is the length */
5452 btrfs_dev_replace_lock(dev_replace
, 0);
5453 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
5454 if (!dev_replace_is_ongoing
)
5455 btrfs_dev_replace_unlock(dev_replace
, 0);
5457 btrfs_dev_replace_set_lock_blocking(dev_replace
);
5459 if (dev_replace_is_ongoing
&& mirror_num
== map
->num_stripes
+ 1 &&
5460 op
!= BTRFS_MAP_WRITE
&& op
!= BTRFS_MAP_DISCARD
&&
5461 op
!= BTRFS_MAP_GET_READ_MIRRORS
&& dev_replace
->tgtdev
!= NULL
) {
5463 * in dev-replace case, for repair case (that's the only
5464 * case where the mirror is selected explicitly when
5465 * calling btrfs_map_block), blocks left of the left cursor
5466 * can also be read from the target drive.
5467 * For REQ_GET_READ_MIRRORS, the target drive is added as
5468 * the last one to the array of stripes. For READ, it also
5469 * needs to be supported using the same mirror number.
5470 * If the requested block is not left of the left cursor,
5471 * EIO is returned. This can happen because btrfs_num_copies()
5472 * returns one more in the dev-replace case.
5474 u64 tmp_length
= *length
;
5475 struct btrfs_bio
*tmp_bbio
= NULL
;
5476 int tmp_num_stripes
;
5477 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5478 int index_srcdev
= 0;
5480 u64 physical_of_found
= 0;
5482 ret
= __btrfs_map_block(fs_info
, BTRFS_MAP_GET_READ_MIRRORS
,
5483 logical
, &tmp_length
, &tmp_bbio
, 0, 0);
5485 WARN_ON(tmp_bbio
!= NULL
);
5489 tmp_num_stripes
= tmp_bbio
->num_stripes
;
5490 if (mirror_num
> tmp_num_stripes
) {
5492 * BTRFS_MAP_GET_READ_MIRRORS does not contain this
5493 * mirror, that means that the requested area
5494 * is not left of the left cursor
5497 btrfs_put_bbio(tmp_bbio
);
5502 * process the rest of the function using the mirror_num
5503 * of the source drive. Therefore look it up first.
5504 * At the end, patch the device pointer to the one of the
5507 for (i
= 0; i
< tmp_num_stripes
; i
++) {
5508 if (tmp_bbio
->stripes
[i
].dev
->devid
!= srcdev_devid
)
5512 * In case of DUP, in order to keep it simple, only add
5513 * the mirror with the lowest physical address
5516 physical_of_found
<= tmp_bbio
->stripes
[i
].physical
)
5521 physical_of_found
= tmp_bbio
->stripes
[i
].physical
;
5524 btrfs_put_bbio(tmp_bbio
);
5532 mirror_num
= index_srcdev
+ 1;
5533 patch_the_first_stripe_for_dev_replace
= 1;
5534 physical_to_patch_in_first_stripe
= physical_of_found
;
5535 } else if (mirror_num
> map
->num_stripes
) {
5541 stripe_nr_orig
= stripe_nr
;
5542 stripe_nr_end
= ALIGN(offset
+ *length
, map
->stripe_len
);
5543 stripe_nr_end
= div_u64(stripe_nr_end
, map
->stripe_len
);
5544 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
5547 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5548 if (op
== BTRFS_MAP_DISCARD
)
5549 num_stripes
= min_t(u64
, map
->num_stripes
,
5550 stripe_nr_end
- stripe_nr_orig
);
5551 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5553 if (op
!= BTRFS_MAP_WRITE
&& op
!= BTRFS_MAP_DISCARD
&&
5554 op
!= BTRFS_MAP_GET_READ_MIRRORS
)
5556 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
5557 if (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_DISCARD
||
5558 op
== BTRFS_MAP_GET_READ_MIRRORS
)
5559 num_stripes
= map
->num_stripes
;
5560 else if (mirror_num
)
5561 stripe_index
= mirror_num
- 1;
5563 stripe_index
= find_live_mirror(fs_info
, map
, 0,
5565 current
->pid
% map
->num_stripes
,
5566 dev_replace_is_ongoing
);
5567 mirror_num
= stripe_index
+ 1;
5570 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
5571 if (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_DISCARD
||
5572 op
== BTRFS_MAP_GET_READ_MIRRORS
) {
5573 num_stripes
= map
->num_stripes
;
5574 } else if (mirror_num
) {
5575 stripe_index
= mirror_num
- 1;
5580 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5581 u32 factor
= map
->num_stripes
/ map
->sub_stripes
;
5583 stripe_nr
= div_u64_rem(stripe_nr
, factor
, &stripe_index
);
5584 stripe_index
*= map
->sub_stripes
;
5586 if (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
)
5587 num_stripes
= map
->sub_stripes
;
5588 else if (op
== BTRFS_MAP_DISCARD
)
5589 num_stripes
= min_t(u64
, map
->sub_stripes
*
5590 (stripe_nr_end
- stripe_nr_orig
),
5592 else if (mirror_num
)
5593 stripe_index
+= mirror_num
- 1;
5595 int old_stripe_index
= stripe_index
;
5596 stripe_index
= find_live_mirror(fs_info
, map
,
5598 map
->sub_stripes
, stripe_index
+
5599 current
->pid
% map
->sub_stripes
,
5600 dev_replace_is_ongoing
);
5601 mirror_num
= stripe_index
- old_stripe_index
+ 1;
5604 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5605 if (need_raid_map
&&
5606 (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
||
5608 /* push stripe_nr back to the start of the full stripe */
5609 stripe_nr
= div_u64(raid56_full_stripe_start
,
5610 stripe_len
* nr_data_stripes(map
));
5612 /* RAID[56] write or recovery. Return all stripes */
5613 num_stripes
= map
->num_stripes
;
5614 max_errors
= nr_parity_stripes(map
);
5616 *length
= map
->stripe_len
;
5621 * Mirror #0 or #1 means the original data block.
5622 * Mirror #2 is RAID5 parity block.
5623 * Mirror #3 is RAID6 Q block.
5625 stripe_nr
= div_u64_rem(stripe_nr
,
5626 nr_data_stripes(map
), &stripe_index
);
5628 stripe_index
= nr_data_stripes(map
) +
5631 /* We distribute the parity blocks across stripes */
5632 div_u64_rem(stripe_nr
+ stripe_index
, map
->num_stripes
,
5634 if ((op
!= BTRFS_MAP_WRITE
&& op
!= BTRFS_MAP_DISCARD
&&
5635 op
!= BTRFS_MAP_GET_READ_MIRRORS
) && mirror_num
<= 1)
5640 * after this, stripe_nr is the number of stripes on this
5641 * device we have to walk to find the data, and stripe_index is
5642 * the number of our device in the stripe array
5644 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5646 mirror_num
= stripe_index
+ 1;
5648 if (stripe_index
>= map
->num_stripes
) {
5650 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5651 stripe_index
, map
->num_stripes
);
5656 num_alloc_stripes
= num_stripes
;
5657 if (dev_replace_is_ongoing
) {
5658 if (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_DISCARD
)
5659 num_alloc_stripes
<<= 1;
5660 if (op
== BTRFS_MAP_GET_READ_MIRRORS
)
5661 num_alloc_stripes
++;
5662 tgtdev_indexes
= num_stripes
;
5665 bbio
= alloc_btrfs_bio(num_alloc_stripes
, tgtdev_indexes
);
5670 if (dev_replace_is_ongoing
)
5671 bbio
->tgtdev_map
= (int *)(bbio
->stripes
+ num_alloc_stripes
);
5673 /* build raid_map */
5674 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
&&
5676 ((op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
) ||
5681 bbio
->raid_map
= (u64
*)((void *)bbio
->stripes
+
5682 sizeof(struct btrfs_bio_stripe
) *
5684 sizeof(int) * tgtdev_indexes
);
5686 /* Work out the disk rotation on this stripe-set */
5687 div_u64_rem(stripe_nr
, num_stripes
, &rot
);
5689 /* Fill in the logical address of each stripe */
5690 tmp
= stripe_nr
* nr_data_stripes(map
);
5691 for (i
= 0; i
< nr_data_stripes(map
); i
++)
5692 bbio
->raid_map
[(i
+rot
) % num_stripes
] =
5693 em
->start
+ (tmp
+ i
) * map
->stripe_len
;
5695 bbio
->raid_map
[(i
+rot
) % map
->num_stripes
] = RAID5_P_STRIPE
;
5696 if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5697 bbio
->raid_map
[(i
+rot
+1) % num_stripes
] =
5701 if (op
== BTRFS_MAP_DISCARD
) {
5703 u32 sub_stripes
= 0;
5704 u64 stripes_per_dev
= 0;
5705 u32 remaining_stripes
= 0;
5706 u32 last_stripe
= 0;
5709 (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID10
)) {
5710 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5713 sub_stripes
= map
->sub_stripes
;
5715 factor
= map
->num_stripes
/ sub_stripes
;
5716 stripes_per_dev
= div_u64_rem(stripe_nr_end
-
5719 &remaining_stripes
);
5720 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
5721 last_stripe
*= sub_stripes
;
5724 for (i
= 0; i
< num_stripes
; i
++) {
5725 bbio
->stripes
[i
].physical
=
5726 map
->stripes
[stripe_index
].physical
+
5727 stripe_offset
+ stripe_nr
* map
->stripe_len
;
5728 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
5730 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5731 BTRFS_BLOCK_GROUP_RAID10
)) {
5732 bbio
->stripes
[i
].length
= stripes_per_dev
*
5735 if (i
/ sub_stripes
< remaining_stripes
)
5736 bbio
->stripes
[i
].length
+=
5740 * Special for the first stripe and
5743 * |-------|...|-------|
5747 if (i
< sub_stripes
)
5748 bbio
->stripes
[i
].length
-=
5751 if (stripe_index
>= last_stripe
&&
5752 stripe_index
<= (last_stripe
+
5754 bbio
->stripes
[i
].length
-=
5757 if (i
== sub_stripes
- 1)
5760 bbio
->stripes
[i
].length
= *length
;
5763 if (stripe_index
== map
->num_stripes
) {
5764 /* This could only happen for RAID0/10 */
5770 for (i
= 0; i
< num_stripes
; i
++) {
5771 bbio
->stripes
[i
].physical
=
5772 map
->stripes
[stripe_index
].physical
+
5774 stripe_nr
* map
->stripe_len
;
5775 bbio
->stripes
[i
].dev
=
5776 map
->stripes
[stripe_index
].dev
;
5781 if (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
)
5782 max_errors
= btrfs_chunk_max_errors(map
);
5785 sort_parity_stripes(bbio
, num_stripes
);
5788 if (dev_replace_is_ongoing
&&
5789 (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_DISCARD
) &&
5790 dev_replace
->tgtdev
!= NULL
) {
5791 int index_where_to_add
;
5792 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5795 * duplicate the write operations while the dev replace
5796 * procedure is running. Since the copying of the old disk
5797 * to the new disk takes place at run time while the
5798 * filesystem is mounted writable, the regular write
5799 * operations to the old disk have to be duplicated to go
5800 * to the new disk as well.
5801 * Note that device->missing is handled by the caller, and
5802 * that the write to the old disk is already set up in the
5805 index_where_to_add
= num_stripes
;
5806 for (i
= 0; i
< num_stripes
; i
++) {
5807 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5808 /* write to new disk, too */
5809 struct btrfs_bio_stripe
*new =
5810 bbio
->stripes
+ index_where_to_add
;
5811 struct btrfs_bio_stripe
*old
=
5814 new->physical
= old
->physical
;
5815 new->length
= old
->length
;
5816 new->dev
= dev_replace
->tgtdev
;
5817 bbio
->tgtdev_map
[i
] = index_where_to_add
;
5818 index_where_to_add
++;
5823 num_stripes
= index_where_to_add
;
5824 } else if (dev_replace_is_ongoing
&&
5825 op
== BTRFS_MAP_GET_READ_MIRRORS
&&
5826 dev_replace
->tgtdev
!= NULL
) {
5827 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5828 int index_srcdev
= 0;
5830 u64 physical_of_found
= 0;
5833 * During the dev-replace procedure, the target drive can
5834 * also be used to read data in case it is needed to repair
5835 * a corrupt block elsewhere. This is possible if the
5836 * requested area is left of the left cursor. In this area,
5837 * the target drive is a full copy of the source drive.
5839 for (i
= 0; i
< num_stripes
; i
++) {
5840 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5842 * In case of DUP, in order to keep it
5843 * simple, only add the mirror with the
5844 * lowest physical address
5847 physical_of_found
<=
5848 bbio
->stripes
[i
].physical
)
5852 physical_of_found
= bbio
->stripes
[i
].physical
;
5856 struct btrfs_bio_stripe
*tgtdev_stripe
=
5857 bbio
->stripes
+ num_stripes
;
5859 tgtdev_stripe
->physical
= physical_of_found
;
5860 tgtdev_stripe
->length
=
5861 bbio
->stripes
[index_srcdev
].length
;
5862 tgtdev_stripe
->dev
= dev_replace
->tgtdev
;
5863 bbio
->tgtdev_map
[index_srcdev
] = num_stripes
;
5871 bbio
->map_type
= map
->type
;
5872 bbio
->num_stripes
= num_stripes
;
5873 bbio
->max_errors
= max_errors
;
5874 bbio
->mirror_num
= mirror_num
;
5875 bbio
->num_tgtdevs
= tgtdev_indexes
;
5878 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5879 * mirror_num == num_stripes + 1 && dev_replace target drive is
5880 * available as a mirror
5882 if (patch_the_first_stripe_for_dev_replace
&& num_stripes
> 0) {
5883 WARN_ON(num_stripes
> 1);
5884 bbio
->stripes
[0].dev
= dev_replace
->tgtdev
;
5885 bbio
->stripes
[0].physical
= physical_to_patch_in_first_stripe
;
5886 bbio
->mirror_num
= map
->num_stripes
+ 1;
5889 if (dev_replace_is_ongoing
) {
5890 btrfs_dev_replace_clear_lock_blocking(dev_replace
);
5891 btrfs_dev_replace_unlock(dev_replace
, 0);
5893 free_extent_map(em
);
5897 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, enum btrfs_map_op op
,
5898 u64 logical
, u64
*length
,
5899 struct btrfs_bio
**bbio_ret
, int mirror_num
)
5901 return __btrfs_map_block(fs_info
, op
, logical
, length
, bbio_ret
,
5905 /* For Scrub/replace */
5906 int btrfs_map_sblock(struct btrfs_fs_info
*fs_info
, enum btrfs_map_op op
,
5907 u64 logical
, u64
*length
,
5908 struct btrfs_bio
**bbio_ret
, int mirror_num
,
5911 return __btrfs_map_block(fs_info
, op
, logical
, length
, bbio_ret
,
5912 mirror_num
, need_raid_map
);
5915 int btrfs_rmap_block(struct btrfs_fs_info
*fs_info
,
5916 u64 chunk_start
, u64 physical
, u64 devid
,
5917 u64
**logical
, int *naddrs
, int *stripe_len
)
5919 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
5920 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5921 struct extent_map
*em
;
5922 struct map_lookup
*map
;
5930 read_lock(&em_tree
->lock
);
5931 em
= lookup_extent_mapping(em_tree
, chunk_start
, 1);
5932 read_unlock(&em_tree
->lock
);
5935 btrfs_err(fs_info
, "couldn't find em for chunk %Lu",
5940 if (em
->start
!= chunk_start
) {
5941 btrfs_err(fs_info
, "bad chunk start, em=%Lu, wanted=%Lu",
5942 em
->start
, chunk_start
);
5943 free_extent_map(em
);
5946 map
= em
->map_lookup
;
5949 rmap_len
= map
->stripe_len
;
5951 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5952 length
= div_u64(length
, map
->num_stripes
/ map
->sub_stripes
);
5953 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5954 length
= div_u64(length
, map
->num_stripes
);
5955 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5956 length
= div_u64(length
, nr_data_stripes(map
));
5957 rmap_len
= map
->stripe_len
* nr_data_stripes(map
);
5960 buf
= kcalloc(map
->num_stripes
, sizeof(u64
), GFP_NOFS
);
5961 BUG_ON(!buf
); /* -ENOMEM */
5963 for (i
= 0; i
< map
->num_stripes
; i
++) {
5964 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
5966 if (map
->stripes
[i
].physical
> physical
||
5967 map
->stripes
[i
].physical
+ length
<= physical
)
5970 stripe_nr
= physical
- map
->stripes
[i
].physical
;
5971 stripe_nr
= div_u64(stripe_nr
, map
->stripe_len
);
5973 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5974 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5975 stripe_nr
= div_u64(stripe_nr
, map
->sub_stripes
);
5976 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5977 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5978 } /* else if RAID[56], multiply by nr_data_stripes().
5979 * Alternatively, just use rmap_len below instead of
5980 * map->stripe_len */
5982 bytenr
= chunk_start
+ stripe_nr
* rmap_len
;
5983 WARN_ON(nr
>= map
->num_stripes
);
5984 for (j
= 0; j
< nr
; j
++) {
5985 if (buf
[j
] == bytenr
)
5989 WARN_ON(nr
>= map
->num_stripes
);
5996 *stripe_len
= rmap_len
;
5998 free_extent_map(em
);
6002 static inline void btrfs_end_bbio(struct btrfs_bio
*bbio
, struct bio
*bio
)
6004 bio
->bi_private
= bbio
->private;
6005 bio
->bi_end_io
= bbio
->end_io
;
6008 btrfs_put_bbio(bbio
);
6011 static void btrfs_end_bio(struct bio
*bio
)
6013 struct btrfs_bio
*bbio
= bio
->bi_private
;
6014 int is_orig_bio
= 0;
6016 if (bio
->bi_error
) {
6017 atomic_inc(&bbio
->error
);
6018 if (bio
->bi_error
== -EIO
|| bio
->bi_error
== -EREMOTEIO
) {
6019 unsigned int stripe_index
=
6020 btrfs_io_bio(bio
)->stripe_index
;
6021 struct btrfs_device
*dev
;
6023 BUG_ON(stripe_index
>= bbio
->num_stripes
);
6024 dev
= bbio
->stripes
[stripe_index
].dev
;
6026 if (bio_op(bio
) == REQ_OP_WRITE
)
6027 btrfs_dev_stat_inc(dev
,
6028 BTRFS_DEV_STAT_WRITE_ERRS
);
6030 btrfs_dev_stat_inc(dev
,
6031 BTRFS_DEV_STAT_READ_ERRS
);
6032 if ((bio
->bi_opf
& WRITE_FLUSH
) == WRITE_FLUSH
)
6033 btrfs_dev_stat_inc(dev
,
6034 BTRFS_DEV_STAT_FLUSH_ERRS
);
6035 btrfs_dev_stat_print_on_error(dev
);
6040 if (bio
== bbio
->orig_bio
)
6043 btrfs_bio_counter_dec(bbio
->fs_info
);
6045 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
6048 bio
= bbio
->orig_bio
;
6051 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
6052 /* only send an error to the higher layers if it is
6053 * beyond the tolerance of the btrfs bio
6055 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
6056 bio
->bi_error
= -EIO
;
6059 * this bio is actually up to date, we didn't
6060 * go over the max number of errors
6065 btrfs_end_bbio(bbio
, bio
);
6066 } else if (!is_orig_bio
) {
6072 * see run_scheduled_bios for a description of why bios are collected for
6075 * This will add one bio to the pending list for a device and make sure
6076 * the work struct is scheduled.
6078 static noinline
void btrfs_schedule_bio(struct btrfs_root
*root
,
6079 struct btrfs_device
*device
,
6082 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
6083 int should_queue
= 1;
6084 struct btrfs_pending_bios
*pending_bios
;
6086 if (device
->missing
|| !device
->bdev
) {
6091 /* don't bother with additional async steps for reads, right now */
6092 if (bio_op(bio
) == REQ_OP_READ
) {
6094 btrfsic_submit_bio(bio
);
6100 * nr_async_bios allows us to reliably return congestion to the
6101 * higher layers. Otherwise, the async bio makes it appear we have
6102 * made progress against dirty pages when we've really just put it
6103 * on a queue for later
6105 atomic_inc(&fs_info
->nr_async_bios
);
6106 WARN_ON(bio
->bi_next
);
6107 bio
->bi_next
= NULL
;
6109 spin_lock(&device
->io_lock
);
6110 if (bio
->bi_opf
& REQ_SYNC
)
6111 pending_bios
= &device
->pending_sync_bios
;
6113 pending_bios
= &device
->pending_bios
;
6115 if (pending_bios
->tail
)
6116 pending_bios
->tail
->bi_next
= bio
;
6118 pending_bios
->tail
= bio
;
6119 if (!pending_bios
->head
)
6120 pending_bios
->head
= bio
;
6121 if (device
->running_pending
)
6124 spin_unlock(&device
->io_lock
);
6127 btrfs_queue_work(fs_info
->submit_workers
, &device
->work
);
6130 static void submit_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
6131 struct bio
*bio
, u64 physical
, int dev_nr
,
6134 struct btrfs_device
*dev
= bbio
->stripes
[dev_nr
].dev
;
6136 bio
->bi_private
= bbio
;
6137 btrfs_io_bio(bio
)->stripe_index
= dev_nr
;
6138 bio
->bi_end_io
= btrfs_end_bio
;
6139 bio
->bi_iter
.bi_sector
= physical
>> 9;
6142 struct rcu_string
*name
;
6145 name
= rcu_dereference(dev
->name
);
6146 btrfs_debug(fs_info
,
6147 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6148 bio_op(bio
), bio
->bi_opf
,
6149 (u64
)bio
->bi_iter
.bi_sector
,
6150 (u_long
)dev
->bdev
->bd_dev
, name
->str
, dev
->devid
,
6151 bio
->bi_iter
.bi_size
);
6155 bio
->bi_bdev
= dev
->bdev
;
6157 btrfs_bio_counter_inc_noblocked(root
->fs_info
);
6160 btrfs_schedule_bio(root
, dev
, bio
);
6162 btrfsic_submit_bio(bio
);
6165 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
6167 atomic_inc(&bbio
->error
);
6168 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
6169 /* Should be the original bio. */
6170 WARN_ON(bio
!= bbio
->orig_bio
);
6172 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
6173 bio
->bi_iter
.bi_sector
= logical
>> 9;
6174 bio
->bi_error
= -EIO
;
6175 btrfs_end_bbio(bbio
, bio
);
6179 int btrfs_map_bio(struct btrfs_root
*root
, struct bio
*bio
,
6180 int mirror_num
, int async_submit
)
6182 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6183 struct btrfs_device
*dev
;
6184 struct bio
*first_bio
= bio
;
6185 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
6191 struct btrfs_bio
*bbio
= NULL
;
6193 length
= bio
->bi_iter
.bi_size
;
6194 map_length
= length
;
6196 btrfs_bio_counter_inc_blocked(fs_info
);
6197 ret
= __btrfs_map_block(fs_info
, bio_op(bio
), logical
,
6198 &map_length
, &bbio
, mirror_num
, 1);
6200 btrfs_bio_counter_dec(fs_info
);
6204 total_devs
= bbio
->num_stripes
;
6205 bbio
->orig_bio
= first_bio
;
6206 bbio
->private = first_bio
->bi_private
;
6207 bbio
->end_io
= first_bio
->bi_end_io
;
6208 bbio
->fs_info
= fs_info
;
6209 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
6211 if ((bbio
->map_type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
6212 ((bio_op(bio
) == REQ_OP_WRITE
) || (mirror_num
> 1))) {
6213 /* In this case, map_length has been set to the length of
6214 a single stripe; not the whole write */
6215 if (bio_op(bio
) == REQ_OP_WRITE
) {
6216 ret
= raid56_parity_write(root
, bio
, bbio
, map_length
);
6218 ret
= raid56_parity_recover(root
, bio
, bbio
, map_length
,
6222 btrfs_bio_counter_dec(fs_info
);
6226 if (map_length
< length
) {
6228 "mapping failed logical %llu bio len %llu len %llu",
6229 logical
, length
, map_length
);
6233 for (dev_nr
= 0; dev_nr
< total_devs
; dev_nr
++) {
6234 dev
= bbio
->stripes
[dev_nr
].dev
;
6235 if (!dev
|| !dev
->bdev
||
6236 (bio_op(bio
) == REQ_OP_WRITE
&& !dev
->writeable
)) {
6237 bbio_error(bbio
, first_bio
, logical
);
6241 if (dev_nr
< total_devs
- 1) {
6242 bio
= btrfs_bio_clone(first_bio
, GFP_NOFS
);
6243 BUG_ON(!bio
); /* -ENOMEM */
6247 submit_stripe_bio(root
, bbio
, bio
,
6248 bbio
->stripes
[dev_nr
].physical
, dev_nr
,
6251 btrfs_bio_counter_dec(fs_info
);
6255 struct btrfs_device
*btrfs_find_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
6258 struct btrfs_device
*device
;
6259 struct btrfs_fs_devices
*cur_devices
;
6261 cur_devices
= fs_info
->fs_devices
;
6262 while (cur_devices
) {
6264 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
6265 device
= __find_device(&cur_devices
->devices
,
6270 cur_devices
= cur_devices
->seed
;
6275 static struct btrfs_device
*add_missing_dev(struct btrfs_root
*root
,
6276 struct btrfs_fs_devices
*fs_devices
,
6277 u64 devid
, u8
*dev_uuid
)
6279 struct btrfs_device
*device
;
6281 device
= btrfs_alloc_device(NULL
, &devid
, dev_uuid
);
6285 list_add(&device
->dev_list
, &fs_devices
->devices
);
6286 device
->fs_devices
= fs_devices
;
6287 fs_devices
->num_devices
++;
6289 device
->missing
= 1;
6290 fs_devices
->missing_devices
++;
6296 * btrfs_alloc_device - allocate struct btrfs_device
6297 * @fs_info: used only for generating a new devid, can be NULL if
6298 * devid is provided (i.e. @devid != NULL).
6299 * @devid: a pointer to devid for this device. If NULL a new devid
6301 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6304 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6305 * on error. Returned struct is not linked onto any lists and can be
6306 * destroyed with kfree() right away.
6308 struct btrfs_device
*btrfs_alloc_device(struct btrfs_fs_info
*fs_info
,
6312 struct btrfs_device
*dev
;
6315 if (WARN_ON(!devid
&& !fs_info
))
6316 return ERR_PTR(-EINVAL
);
6318 dev
= __alloc_device();
6327 ret
= find_next_devid(fs_info
, &tmp
);
6330 return ERR_PTR(ret
);
6336 memcpy(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
);
6338 generate_random_uuid(dev
->uuid
);
6340 btrfs_init_work(&dev
->work
, btrfs_submit_helper
,
6341 pending_bios_fn
, NULL
, NULL
);
6346 /* Return -EIO if any error, otherwise return 0. */
6347 static int btrfs_check_chunk_valid(struct btrfs_root
*root
,
6348 struct extent_buffer
*leaf
,
6349 struct btrfs_chunk
*chunk
, u64 logical
)
6351 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6358 length
= btrfs_chunk_length(leaf
, chunk
);
6359 stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6360 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6361 sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6362 type
= btrfs_chunk_type(leaf
, chunk
);
6365 btrfs_err(fs_info
, "invalid chunk num_stripes: %u",
6369 if (!IS_ALIGNED(logical
, fs_info
->sectorsize
)) {
6370 btrfs_err(fs_info
, "invalid chunk logical %llu", logical
);
6373 if (btrfs_chunk_sector_size(leaf
, chunk
) != fs_info
->sectorsize
) {
6374 btrfs_err(fs_info
, "invalid chunk sectorsize %u",
6375 btrfs_chunk_sector_size(leaf
, chunk
));
6378 if (!length
|| !IS_ALIGNED(length
, fs_info
->sectorsize
)) {
6379 btrfs_err(fs_info
, "invalid chunk length %llu", length
);
6382 if (!is_power_of_2(stripe_len
) || stripe_len
!= BTRFS_STRIPE_LEN
) {
6383 btrfs_err(fs_info
, "invalid chunk stripe length: %llu",
6387 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK
| BTRFS_BLOCK_GROUP_PROFILE_MASK
) &
6389 btrfs_err(fs_info
, "unrecognized chunk type: %llu",
6390 ~(BTRFS_BLOCK_GROUP_TYPE_MASK
|
6391 BTRFS_BLOCK_GROUP_PROFILE_MASK
) &
6392 btrfs_chunk_type(leaf
, chunk
));
6395 if ((type
& BTRFS_BLOCK_GROUP_RAID10
&& sub_stripes
!= 2) ||
6396 (type
& BTRFS_BLOCK_GROUP_RAID1
&& num_stripes
< 1) ||
6397 (type
& BTRFS_BLOCK_GROUP_RAID5
&& num_stripes
< 2) ||
6398 (type
& BTRFS_BLOCK_GROUP_RAID6
&& num_stripes
< 3) ||
6399 (type
& BTRFS_BLOCK_GROUP_DUP
&& num_stripes
> 2) ||
6400 ((type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) == 0 &&
6401 num_stripes
!= 1)) {
6403 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6404 num_stripes
, sub_stripes
,
6405 type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
);
6412 static int read_one_chunk(struct btrfs_root
*root
, struct btrfs_key
*key
,
6413 struct extent_buffer
*leaf
,
6414 struct btrfs_chunk
*chunk
)
6416 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6417 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
6418 struct map_lookup
*map
;
6419 struct extent_map
*em
;
6424 u8 uuid
[BTRFS_UUID_SIZE
];
6429 logical
= key
->offset
;
6430 length
= btrfs_chunk_length(leaf
, chunk
);
6431 stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6432 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6434 ret
= btrfs_check_chunk_valid(root
, leaf
, chunk
, logical
);
6438 read_lock(&map_tree
->map_tree
.lock
);
6439 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
6440 read_unlock(&map_tree
->map_tree
.lock
);
6442 /* already mapped? */
6443 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
6444 free_extent_map(em
);
6447 free_extent_map(em
);
6450 em
= alloc_extent_map();
6453 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
6455 free_extent_map(em
);
6459 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
6460 em
->map_lookup
= map
;
6461 em
->start
= logical
;
6464 em
->block_start
= 0;
6465 em
->block_len
= em
->len
;
6467 map
->num_stripes
= num_stripes
;
6468 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
6469 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
6470 map
->sector_size
= btrfs_chunk_sector_size(leaf
, chunk
);
6471 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6472 map
->type
= btrfs_chunk_type(leaf
, chunk
);
6473 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6474 for (i
= 0; i
< num_stripes
; i
++) {
6475 map
->stripes
[i
].physical
=
6476 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
6477 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
6478 read_extent_buffer(leaf
, uuid
, (unsigned long)
6479 btrfs_stripe_dev_uuid_nr(chunk
, i
),
6481 map
->stripes
[i
].dev
= btrfs_find_device(fs_info
, devid
,
6483 if (!map
->stripes
[i
].dev
&&
6484 !btrfs_test_opt(fs_info
, DEGRADED
)) {
6485 free_extent_map(em
);
6488 if (!map
->stripes
[i
].dev
) {
6489 map
->stripes
[i
].dev
=
6490 add_missing_dev(root
, fs_info
->fs_devices
,
6492 if (!map
->stripes
[i
].dev
) {
6493 free_extent_map(em
);
6496 btrfs_warn(fs_info
, "devid %llu uuid %pU is missing",
6499 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
6502 write_lock(&map_tree
->map_tree
.lock
);
6503 ret
= add_extent_mapping(&map_tree
->map_tree
, em
, 0);
6504 write_unlock(&map_tree
->map_tree
.lock
);
6505 BUG_ON(ret
); /* Tree corruption */
6506 free_extent_map(em
);
6511 static void fill_device_from_item(struct extent_buffer
*leaf
,
6512 struct btrfs_dev_item
*dev_item
,
6513 struct btrfs_device
*device
)
6517 device
->devid
= btrfs_device_id(leaf
, dev_item
);
6518 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
6519 device
->total_bytes
= device
->disk_total_bytes
;
6520 device
->commit_total_bytes
= device
->disk_total_bytes
;
6521 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
6522 device
->commit_bytes_used
= device
->bytes_used
;
6523 device
->type
= btrfs_device_type(leaf
, dev_item
);
6524 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
6525 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
6526 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
6527 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
6528 device
->is_tgtdev_for_dev_replace
= 0;
6530 ptr
= btrfs_device_uuid(dev_item
);
6531 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
6534 static struct btrfs_fs_devices
*open_seed_devices(struct btrfs_root
*root
,
6537 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6538 struct btrfs_fs_devices
*fs_devices
;
6541 BUG_ON(!mutex_is_locked(&uuid_mutex
));
6543 fs_devices
= fs_info
->fs_devices
->seed
;
6544 while (fs_devices
) {
6545 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
))
6548 fs_devices
= fs_devices
->seed
;
6551 fs_devices
= find_fsid(fsid
);
6553 if (!btrfs_test_opt(fs_info
, DEGRADED
))
6554 return ERR_PTR(-ENOENT
);
6556 fs_devices
= alloc_fs_devices(fsid
);
6557 if (IS_ERR(fs_devices
))
6560 fs_devices
->seeding
= 1;
6561 fs_devices
->opened
= 1;
6565 fs_devices
= clone_fs_devices(fs_devices
);
6566 if (IS_ERR(fs_devices
))
6569 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
6570 fs_info
->bdev_holder
);
6572 free_fs_devices(fs_devices
);
6573 fs_devices
= ERR_PTR(ret
);
6577 if (!fs_devices
->seeding
) {
6578 __btrfs_close_devices(fs_devices
);
6579 free_fs_devices(fs_devices
);
6580 fs_devices
= ERR_PTR(-EINVAL
);
6584 fs_devices
->seed
= fs_info
->fs_devices
->seed
;
6585 fs_info
->fs_devices
->seed
= fs_devices
;
6590 static int read_one_dev(struct btrfs_root
*root
,
6591 struct extent_buffer
*leaf
,
6592 struct btrfs_dev_item
*dev_item
)
6594 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6595 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6596 struct btrfs_device
*device
;
6599 u8 fs_uuid
[BTRFS_UUID_SIZE
];
6600 u8 dev_uuid
[BTRFS_UUID_SIZE
];
6602 devid
= btrfs_device_id(leaf
, dev_item
);
6603 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
6605 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
6608 if (memcmp(fs_uuid
, fs_info
->fsid
, BTRFS_UUID_SIZE
)) {
6609 fs_devices
= open_seed_devices(root
, fs_uuid
);
6610 if (IS_ERR(fs_devices
))
6611 return PTR_ERR(fs_devices
);
6614 device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, fs_uuid
);
6616 if (!btrfs_test_opt(fs_info
, DEGRADED
))
6619 device
= add_missing_dev(root
, fs_devices
, devid
, dev_uuid
);
6622 btrfs_warn(fs_info
, "devid %llu uuid %pU missing",
6625 if (!device
->bdev
&& !btrfs_test_opt(fs_info
, DEGRADED
))
6628 if(!device
->bdev
&& !device
->missing
) {
6630 * this happens when a device that was properly setup
6631 * in the device info lists suddenly goes bad.
6632 * device->bdev is NULL, and so we have to set
6633 * device->missing to one here
6635 device
->fs_devices
->missing_devices
++;
6636 device
->missing
= 1;
6639 /* Move the device to its own fs_devices */
6640 if (device
->fs_devices
!= fs_devices
) {
6641 ASSERT(device
->missing
);
6643 list_move(&device
->dev_list
, &fs_devices
->devices
);
6644 device
->fs_devices
->num_devices
--;
6645 fs_devices
->num_devices
++;
6647 device
->fs_devices
->missing_devices
--;
6648 fs_devices
->missing_devices
++;
6650 device
->fs_devices
= fs_devices
;
6654 if (device
->fs_devices
!= fs_info
->fs_devices
) {
6655 BUG_ON(device
->writeable
);
6656 if (device
->generation
!=
6657 btrfs_device_generation(leaf
, dev_item
))
6661 fill_device_from_item(leaf
, dev_item
, device
);
6662 device
->in_fs_metadata
= 1;
6663 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
6664 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
6665 spin_lock(&fs_info
->free_chunk_lock
);
6666 fs_info
->free_chunk_space
+= device
->total_bytes
-
6668 spin_unlock(&fs_info
->free_chunk_lock
);
6674 int btrfs_read_sys_array(struct btrfs_fs_info
*fs_info
)
6676 struct btrfs_root
*root
= fs_info
->tree_root
;
6677 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
6678 struct extent_buffer
*sb
;
6679 struct btrfs_disk_key
*disk_key
;
6680 struct btrfs_chunk
*chunk
;
6682 unsigned long sb_array_offset
;
6689 struct btrfs_key key
;
6691 ASSERT(BTRFS_SUPER_INFO_SIZE
<= fs_info
->nodesize
);
6693 * This will create extent buffer of nodesize, superblock size is
6694 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6695 * overallocate but we can keep it as-is, only the first page is used.
6697 sb
= btrfs_find_create_tree_block(root
, BTRFS_SUPER_INFO_OFFSET
);
6700 set_extent_buffer_uptodate(sb
);
6701 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
6703 * The sb extent buffer is artificial and just used to read the system array.
6704 * set_extent_buffer_uptodate() call does not properly mark all it's
6705 * pages up-to-date when the page is larger: extent does not cover the
6706 * whole page and consequently check_page_uptodate does not find all
6707 * the page's extents up-to-date (the hole beyond sb),
6708 * write_extent_buffer then triggers a WARN_ON.
6710 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6711 * but sb spans only this function. Add an explicit SetPageUptodate call
6712 * to silence the warning eg. on PowerPC 64.
6714 if (PAGE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
6715 SetPageUptodate(sb
->pages
[0]);
6717 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
6718 array_size
= btrfs_super_sys_array_size(super_copy
);
6720 array_ptr
= super_copy
->sys_chunk_array
;
6721 sb_array_offset
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
6724 while (cur_offset
< array_size
) {
6725 disk_key
= (struct btrfs_disk_key
*)array_ptr
;
6726 len
= sizeof(*disk_key
);
6727 if (cur_offset
+ len
> array_size
)
6728 goto out_short_read
;
6730 btrfs_disk_key_to_cpu(&key
, disk_key
);
6733 sb_array_offset
+= len
;
6736 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6737 chunk
= (struct btrfs_chunk
*)sb_array_offset
;
6739 * At least one btrfs_chunk with one stripe must be
6740 * present, exact stripe count check comes afterwards
6742 len
= btrfs_chunk_item_size(1);
6743 if (cur_offset
+ len
> array_size
)
6744 goto out_short_read
;
6746 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
6749 "invalid number of stripes %u in sys_array at offset %u",
6750 num_stripes
, cur_offset
);
6755 type
= btrfs_chunk_type(sb
, chunk
);
6756 if ((type
& BTRFS_BLOCK_GROUP_SYSTEM
) == 0) {
6758 "invalid chunk type %llu in sys_array at offset %u",
6764 len
= btrfs_chunk_item_size(num_stripes
);
6765 if (cur_offset
+ len
> array_size
)
6766 goto out_short_read
;
6768 ret
= read_one_chunk(root
, &key
, sb
, chunk
);
6773 "unexpected item type %u in sys_array at offset %u",
6774 (u32
)key
.type
, cur_offset
);
6779 sb_array_offset
+= len
;
6782 clear_extent_buffer_uptodate(sb
);
6783 free_extent_buffer_stale(sb
);
6787 btrfs_err(fs_info
, "sys_array too short to read %u bytes at offset %u",
6789 clear_extent_buffer_uptodate(sb
);
6790 free_extent_buffer_stale(sb
);
6794 int btrfs_read_chunk_tree(struct btrfs_fs_info
*fs_info
)
6796 struct btrfs_root
*root
= fs_info
->chunk_root
;
6797 struct btrfs_path
*path
;
6798 struct extent_buffer
*leaf
;
6799 struct btrfs_key key
;
6800 struct btrfs_key found_key
;
6805 path
= btrfs_alloc_path();
6809 mutex_lock(&uuid_mutex
);
6810 lock_chunks(fs_info
);
6813 * Read all device items, and then all the chunk items. All
6814 * device items are found before any chunk item (their object id
6815 * is smaller than the lowest possible object id for a chunk
6816 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6818 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
6821 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6825 leaf
= path
->nodes
[0];
6826 slot
= path
->slots
[0];
6827 if (slot
>= btrfs_header_nritems(leaf
)) {
6828 ret
= btrfs_next_leaf(root
, path
);
6835 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6836 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
6837 struct btrfs_dev_item
*dev_item
;
6838 dev_item
= btrfs_item_ptr(leaf
, slot
,
6839 struct btrfs_dev_item
);
6840 ret
= read_one_dev(root
, leaf
, dev_item
);
6844 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6845 struct btrfs_chunk
*chunk
;
6846 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
6847 ret
= read_one_chunk(root
, &found_key
, leaf
, chunk
);
6855 * After loading chunk tree, we've got all device information,
6856 * do another round of validation checks.
6858 if (total_dev
!= fs_info
->fs_devices
->total_devices
) {
6860 "super_num_devices %llu mismatch with num_devices %llu found here",
6861 btrfs_super_num_devices(fs_info
->super_copy
),
6866 if (btrfs_super_total_bytes(fs_info
->super_copy
) <
6867 fs_info
->fs_devices
->total_rw_bytes
) {
6869 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6870 btrfs_super_total_bytes(fs_info
->super_copy
),
6871 fs_info
->fs_devices
->total_rw_bytes
);
6877 unlock_chunks(fs_info
);
6878 mutex_unlock(&uuid_mutex
);
6880 btrfs_free_path(path
);
6884 void btrfs_init_devices_late(struct btrfs_fs_info
*fs_info
)
6886 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6887 struct btrfs_device
*device
;
6889 while (fs_devices
) {
6890 mutex_lock(&fs_devices
->device_list_mutex
);
6891 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
)
6892 device
->fs_info
= fs_info
;
6893 mutex_unlock(&fs_devices
->device_list_mutex
);
6895 fs_devices
= fs_devices
->seed
;
6899 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
6903 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6904 btrfs_dev_stat_reset(dev
, i
);
6907 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
6909 struct btrfs_key key
;
6910 struct btrfs_key found_key
;
6911 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6912 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6913 struct extent_buffer
*eb
;
6916 struct btrfs_device
*device
;
6917 struct btrfs_path
*path
= NULL
;
6920 path
= btrfs_alloc_path();
6926 mutex_lock(&fs_devices
->device_list_mutex
);
6927 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6929 struct btrfs_dev_stats_item
*ptr
;
6931 key
.objectid
= BTRFS_DEV_STATS_OBJECTID
;
6932 key
.type
= BTRFS_PERSISTENT_ITEM_KEY
;
6933 key
.offset
= device
->devid
;
6934 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
6936 __btrfs_reset_dev_stats(device
);
6937 device
->dev_stats_valid
= 1;
6938 btrfs_release_path(path
);
6941 slot
= path
->slots
[0];
6942 eb
= path
->nodes
[0];
6943 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
6944 item_size
= btrfs_item_size_nr(eb
, slot
);
6946 ptr
= btrfs_item_ptr(eb
, slot
,
6947 struct btrfs_dev_stats_item
);
6949 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6950 if (item_size
>= (1 + i
) * sizeof(__le64
))
6951 btrfs_dev_stat_set(device
, i
,
6952 btrfs_dev_stats_value(eb
, ptr
, i
));
6954 btrfs_dev_stat_reset(device
, i
);
6957 device
->dev_stats_valid
= 1;
6958 btrfs_dev_stat_print_on_load(device
);
6959 btrfs_release_path(path
);
6961 mutex_unlock(&fs_devices
->device_list_mutex
);
6964 btrfs_free_path(path
);
6965 return ret
< 0 ? ret
: 0;
6968 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
6969 struct btrfs_fs_info
*fs_info
,
6970 struct btrfs_device
*device
)
6972 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6973 struct btrfs_path
*path
;
6974 struct btrfs_key key
;
6975 struct extent_buffer
*eb
;
6976 struct btrfs_dev_stats_item
*ptr
;
6980 key
.objectid
= BTRFS_DEV_STATS_OBJECTID
;
6981 key
.type
= BTRFS_PERSISTENT_ITEM_KEY
;
6982 key
.offset
= device
->devid
;
6984 path
= btrfs_alloc_path();
6986 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
6988 btrfs_warn_in_rcu(fs_info
,
6989 "error %d while searching for dev_stats item for device %s",
6990 ret
, rcu_str_deref(device
->name
));
6995 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
6996 /* need to delete old one and insert a new one */
6997 ret
= btrfs_del_item(trans
, dev_root
, path
);
6999 btrfs_warn_in_rcu(fs_info
,
7000 "delete too small dev_stats item for device %s failed %d",
7001 rcu_str_deref(device
->name
), ret
);
7008 /* need to insert a new item */
7009 btrfs_release_path(path
);
7010 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
7011 &key
, sizeof(*ptr
));
7013 btrfs_warn_in_rcu(fs_info
,
7014 "insert dev_stats item for device %s failed %d",
7015 rcu_str_deref(device
->name
), ret
);
7020 eb
= path
->nodes
[0];
7021 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
7022 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7023 btrfs_set_dev_stats_value(eb
, ptr
, i
,
7024 btrfs_dev_stat_read(device
, i
));
7025 btrfs_mark_buffer_dirty(eb
);
7028 btrfs_free_path(path
);
7033 * called from commit_transaction. Writes all changed device stats to disk.
7035 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
7036 struct btrfs_fs_info
*fs_info
)
7038 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7039 struct btrfs_device
*device
;
7043 mutex_lock(&fs_devices
->device_list_mutex
);
7044 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
7045 if (!device
->dev_stats_valid
|| !btrfs_dev_stats_dirty(device
))
7048 stats_cnt
= atomic_read(&device
->dev_stats_ccnt
);
7049 ret
= update_dev_stat_item(trans
, fs_info
, device
);
7051 atomic_sub(stats_cnt
, &device
->dev_stats_ccnt
);
7053 mutex_unlock(&fs_devices
->device_list_mutex
);
7058 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
7060 btrfs_dev_stat_inc(dev
, index
);
7061 btrfs_dev_stat_print_on_error(dev
);
7064 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
7066 if (!dev
->dev_stats_valid
)
7068 btrfs_err_rl_in_rcu(dev
->fs_info
,
7069 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7070 rcu_str_deref(dev
->name
),
7071 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
7072 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
7073 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
7074 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
7075 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
7078 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
7082 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7083 if (btrfs_dev_stat_read(dev
, i
) != 0)
7085 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
7086 return; /* all values == 0, suppress message */
7088 btrfs_info_in_rcu(dev
->fs_info
,
7089 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7090 rcu_str_deref(dev
->name
),
7091 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
7092 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
7093 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
7094 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
7095 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
7098 int btrfs_get_dev_stats(struct btrfs_root
*root
,
7099 struct btrfs_ioctl_get_dev_stats
*stats
)
7101 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7102 struct btrfs_device
*dev
;
7103 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7106 mutex_lock(&fs_devices
->device_list_mutex
);
7107 dev
= btrfs_find_device(fs_info
, stats
->devid
, NULL
, NULL
);
7108 mutex_unlock(&fs_devices
->device_list_mutex
);
7111 btrfs_warn(fs_info
, "get dev_stats failed, device not found");
7113 } else if (!dev
->dev_stats_valid
) {
7114 btrfs_warn(fs_info
, "get dev_stats failed, not yet valid");
7116 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
7117 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
7118 if (stats
->nr_items
> i
)
7120 btrfs_dev_stat_read_and_reset(dev
, i
);
7122 btrfs_dev_stat_reset(dev
, i
);
7125 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7126 if (stats
->nr_items
> i
)
7127 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
7129 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
7130 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
7134 void btrfs_scratch_superblocks(struct block_device
*bdev
, char *device_path
)
7136 struct buffer_head
*bh
;
7137 struct btrfs_super_block
*disk_super
;
7143 for (copy_num
= 0; copy_num
< BTRFS_SUPER_MIRROR_MAX
;
7146 if (btrfs_read_dev_one_super(bdev
, copy_num
, &bh
))
7149 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
7151 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
7152 set_buffer_dirty(bh
);
7153 sync_dirty_buffer(bh
);
7157 /* Notify udev that device has changed */
7158 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
7160 /* Update ctime/mtime for device path for libblkid */
7161 update_dev_time(device_path
);
7165 * Update the size of all devices, which is used for writing out the
7168 void btrfs_update_commit_device_size(struct btrfs_fs_info
*fs_info
)
7170 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7171 struct btrfs_device
*curr
, *next
;
7173 if (list_empty(&fs_devices
->resized_devices
))
7176 mutex_lock(&fs_devices
->device_list_mutex
);
7177 lock_chunks(fs_info
);
7178 list_for_each_entry_safe(curr
, next
, &fs_devices
->resized_devices
,
7180 list_del_init(&curr
->resized_list
);
7181 curr
->commit_total_bytes
= curr
->disk_total_bytes
;
7183 unlock_chunks(fs_info
);
7184 mutex_unlock(&fs_devices
->device_list_mutex
);
7187 /* Must be invoked during the transaction commit */
7188 void btrfs_update_commit_device_bytes_used(struct btrfs_root
*root
,
7189 struct btrfs_transaction
*transaction
)
7191 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7192 struct extent_map
*em
;
7193 struct map_lookup
*map
;
7194 struct btrfs_device
*dev
;
7197 if (list_empty(&transaction
->pending_chunks
))
7200 /* In order to kick the device replace finish process */
7201 lock_chunks(fs_info
);
7202 list_for_each_entry(em
, &transaction
->pending_chunks
, list
) {
7203 map
= em
->map_lookup
;
7205 for (i
= 0; i
< map
->num_stripes
; i
++) {
7206 dev
= map
->stripes
[i
].dev
;
7207 dev
->commit_bytes_used
= dev
->bytes_used
;
7210 unlock_chunks(fs_info
);
7213 void btrfs_set_fs_info_ptr(struct btrfs_fs_info
*fs_info
)
7215 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7216 while (fs_devices
) {
7217 fs_devices
->fs_info
= fs_info
;
7218 fs_devices
= fs_devices
->seed
;
7222 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info
*fs_info
)
7224 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7225 while (fs_devices
) {
7226 fs_devices
->fs_info
= NULL
;
7227 fs_devices
= fs_devices
->seed
;