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/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/ratelimit.h>
27 #include <linux/kthread.h>
28 #include <linux/raid/pq.h>
29 #include <linux/semaphore.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
,
121 static int init_first_rw_device(struct btrfs_trans_handle
*trans
,
122 struct btrfs_root
*root
,
123 struct btrfs_device
*device
);
124 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
);
125 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
);
126 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
);
127 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*device
);
129 DEFINE_MUTEX(uuid_mutex
);
130 static LIST_HEAD(fs_uuids
);
131 struct list_head
*btrfs_get_fs_uuids(void)
136 static struct btrfs_fs_devices
*__alloc_fs_devices(void)
138 struct btrfs_fs_devices
*fs_devs
;
140 fs_devs
= kzalloc(sizeof(*fs_devs
), GFP_NOFS
);
142 return ERR_PTR(-ENOMEM
);
144 mutex_init(&fs_devs
->device_list_mutex
);
146 INIT_LIST_HEAD(&fs_devs
->devices
);
147 INIT_LIST_HEAD(&fs_devs
->resized_devices
);
148 INIT_LIST_HEAD(&fs_devs
->alloc_list
);
149 INIT_LIST_HEAD(&fs_devs
->list
);
155 * alloc_fs_devices - allocate struct btrfs_fs_devices
156 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
159 * Return: a pointer to a new &struct btrfs_fs_devices on success;
160 * ERR_PTR() on error. Returned struct is not linked onto any lists and
161 * can be destroyed with kfree() right away.
163 static struct btrfs_fs_devices
*alloc_fs_devices(const u8
*fsid
)
165 struct btrfs_fs_devices
*fs_devs
;
167 fs_devs
= __alloc_fs_devices();
172 memcpy(fs_devs
->fsid
, fsid
, BTRFS_FSID_SIZE
);
174 generate_random_uuid(fs_devs
->fsid
);
179 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
181 struct btrfs_device
*device
;
182 WARN_ON(fs_devices
->opened
);
183 while (!list_empty(&fs_devices
->devices
)) {
184 device
= list_entry(fs_devices
->devices
.next
,
185 struct btrfs_device
, dev_list
);
186 list_del(&device
->dev_list
);
187 rcu_string_free(device
->name
);
193 static void btrfs_kobject_uevent(struct block_device
*bdev
,
194 enum kobject_action action
)
198 ret
= kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, action
);
200 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
202 kobject_name(&disk_to_dev(bdev
->bd_disk
)->kobj
),
203 &disk_to_dev(bdev
->bd_disk
)->kobj
);
206 void btrfs_cleanup_fs_uuids(void)
208 struct btrfs_fs_devices
*fs_devices
;
210 while (!list_empty(&fs_uuids
)) {
211 fs_devices
= list_entry(fs_uuids
.next
,
212 struct btrfs_fs_devices
, list
);
213 list_del(&fs_devices
->list
);
214 free_fs_devices(fs_devices
);
218 static struct btrfs_device
*__alloc_device(void)
220 struct btrfs_device
*dev
;
222 dev
= kzalloc(sizeof(*dev
), GFP_NOFS
);
224 return ERR_PTR(-ENOMEM
);
226 INIT_LIST_HEAD(&dev
->dev_list
);
227 INIT_LIST_HEAD(&dev
->dev_alloc_list
);
228 INIT_LIST_HEAD(&dev
->resized_list
);
230 spin_lock_init(&dev
->io_lock
);
232 spin_lock_init(&dev
->reada_lock
);
233 atomic_set(&dev
->reada_in_flight
, 0);
234 atomic_set(&dev
->dev_stats_ccnt
, 0);
235 btrfs_device_data_ordered_init(dev
);
236 INIT_RADIX_TREE(&dev
->reada_zones
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
237 INIT_RADIX_TREE(&dev
->reada_extents
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
242 static noinline
struct btrfs_device
*__find_device(struct list_head
*head
,
245 struct btrfs_device
*dev
;
247 list_for_each_entry(dev
, head
, dev_list
) {
248 if (dev
->devid
== devid
&&
249 (!uuid
|| !memcmp(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
))) {
256 static noinline
struct btrfs_fs_devices
*find_fsid(u8
*fsid
)
258 struct btrfs_fs_devices
*fs_devices
;
260 list_for_each_entry(fs_devices
, &fs_uuids
, list
) {
261 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
268 btrfs_get_bdev_and_sb(const char *device_path
, fmode_t flags
, void *holder
,
269 int flush
, struct block_device
**bdev
,
270 struct buffer_head
**bh
)
274 *bdev
= blkdev_get_by_path(device_path
, flags
, holder
);
277 ret
= PTR_ERR(*bdev
);
282 filemap_write_and_wait((*bdev
)->bd_inode
->i_mapping
);
283 ret
= set_blocksize(*bdev
, 4096);
285 blkdev_put(*bdev
, flags
);
288 invalidate_bdev(*bdev
);
289 *bh
= btrfs_read_dev_super(*bdev
);
292 blkdev_put(*bdev
, flags
);
304 static void requeue_list(struct btrfs_pending_bios
*pending_bios
,
305 struct bio
*head
, struct bio
*tail
)
308 struct bio
*old_head
;
310 old_head
= pending_bios
->head
;
311 pending_bios
->head
= head
;
312 if (pending_bios
->tail
)
313 tail
->bi_next
= old_head
;
315 pending_bios
->tail
= tail
;
319 * we try to collect pending bios for a device so we don't get a large
320 * number of procs sending bios down to the same device. This greatly
321 * improves the schedulers ability to collect and merge the bios.
323 * But, it also turns into a long list of bios to process and that is sure
324 * to eventually make the worker thread block. The solution here is to
325 * make some progress and then put this work struct back at the end of
326 * the list if the block device is congested. This way, multiple devices
327 * can make progress from a single worker thread.
329 static noinline
void run_scheduled_bios(struct btrfs_device
*device
)
332 struct backing_dev_info
*bdi
;
333 struct btrfs_fs_info
*fs_info
;
334 struct btrfs_pending_bios
*pending_bios
;
338 unsigned long num_run
;
339 unsigned long batch_run
= 0;
341 unsigned long last_waited
= 0;
343 int sync_pending
= 0;
344 struct blk_plug plug
;
347 * this function runs all the bios we've collected for
348 * a particular device. We don't want to wander off to
349 * another device without first sending all of these down.
350 * So, setup a plug here and finish it off before we return
352 blk_start_plug(&plug
);
354 bdi
= blk_get_backing_dev_info(device
->bdev
);
355 fs_info
= device
->dev_root
->fs_info
;
356 limit
= btrfs_async_submit_limit(fs_info
);
357 limit
= limit
* 2 / 3;
360 spin_lock(&device
->io_lock
);
365 /* take all the bios off the list at once and process them
366 * later on (without the lock held). But, remember the
367 * tail and other pointers so the bios can be properly reinserted
368 * into the list if we hit congestion
370 if (!force_reg
&& device
->pending_sync_bios
.head
) {
371 pending_bios
= &device
->pending_sync_bios
;
374 pending_bios
= &device
->pending_bios
;
378 pending
= pending_bios
->head
;
379 tail
= pending_bios
->tail
;
380 WARN_ON(pending
&& !tail
);
383 * if pending was null this time around, no bios need processing
384 * at all and we can stop. Otherwise it'll loop back up again
385 * and do an additional check so no bios are missed.
387 * device->running_pending is used to synchronize with the
390 if (device
->pending_sync_bios
.head
== NULL
&&
391 device
->pending_bios
.head
== NULL
) {
393 device
->running_pending
= 0;
396 device
->running_pending
= 1;
399 pending_bios
->head
= NULL
;
400 pending_bios
->tail
= NULL
;
402 spin_unlock(&device
->io_lock
);
407 /* we want to work on both lists, but do more bios on the
408 * sync list than the regular list
411 pending_bios
!= &device
->pending_sync_bios
&&
412 device
->pending_sync_bios
.head
) ||
413 (num_run
> 64 && pending_bios
== &device
->pending_sync_bios
&&
414 device
->pending_bios
.head
)) {
415 spin_lock(&device
->io_lock
);
416 requeue_list(pending_bios
, pending
, tail
);
421 pending
= pending
->bi_next
;
425 * atomic_dec_return implies a barrier for waitqueue_active
427 if (atomic_dec_return(&fs_info
->nr_async_bios
) < limit
&&
428 waitqueue_active(&fs_info
->async_submit_wait
))
429 wake_up(&fs_info
->async_submit_wait
);
431 BUG_ON(atomic_read(&cur
->__bi_cnt
) == 0);
434 * if we're doing the sync list, record that our
435 * plug has some sync requests on it
437 * If we're doing the regular list and there are
438 * sync requests sitting around, unplug before
441 if (pending_bios
== &device
->pending_sync_bios
) {
443 } else if (sync_pending
) {
444 blk_finish_plug(&plug
);
445 blk_start_plug(&plug
);
449 btrfsic_submit_bio(cur
->bi_rw
, cur
);
456 * we made progress, there is more work to do and the bdi
457 * is now congested. Back off and let other work structs
460 if (pending
&& bdi_write_congested(bdi
) && batch_run
> 8 &&
461 fs_info
->fs_devices
->open_devices
> 1) {
462 struct io_context
*ioc
;
464 ioc
= current
->io_context
;
467 * the main goal here is that we don't want to
468 * block if we're going to be able to submit
469 * more requests without blocking.
471 * This code does two great things, it pokes into
472 * the elevator code from a filesystem _and_
473 * it makes assumptions about how batching works.
475 if (ioc
&& ioc
->nr_batch_requests
> 0 &&
476 time_before(jiffies
, ioc
->last_waited
+ HZ
/50UL) &&
478 ioc
->last_waited
== last_waited
)) {
480 * we want to go through our batch of
481 * requests and stop. So, we copy out
482 * the ioc->last_waited time and test
483 * against it before looping
485 last_waited
= ioc
->last_waited
;
489 spin_lock(&device
->io_lock
);
490 requeue_list(pending_bios
, pending
, tail
);
491 device
->running_pending
= 1;
493 spin_unlock(&device
->io_lock
);
494 btrfs_queue_work(fs_info
->submit_workers
,
498 /* unplug every 64 requests just for good measure */
499 if (batch_run
% 64 == 0) {
500 blk_finish_plug(&plug
);
501 blk_start_plug(&plug
);
510 spin_lock(&device
->io_lock
);
511 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
513 spin_unlock(&device
->io_lock
);
516 blk_finish_plug(&plug
);
519 static void pending_bios_fn(struct btrfs_work
*work
)
521 struct btrfs_device
*device
;
523 device
= container_of(work
, struct btrfs_device
, work
);
524 run_scheduled_bios(device
);
528 void btrfs_free_stale_device(struct btrfs_device
*cur_dev
)
530 struct btrfs_fs_devices
*fs_devs
;
531 struct btrfs_device
*dev
;
536 list_for_each_entry(fs_devs
, &fs_uuids
, list
) {
541 if (fs_devs
->seeding
)
544 list_for_each_entry(dev
, &fs_devs
->devices
, dev_list
) {
552 * Todo: This won't be enough. What if the same device
553 * comes back (with new uuid and) with its mapper path?
554 * But for now, this does help as mostly an admin will
555 * either use mapper or non mapper path throughout.
558 del
= strcmp(rcu_str_deref(dev
->name
),
559 rcu_str_deref(cur_dev
->name
));
566 /* delete the stale device */
567 if (fs_devs
->num_devices
== 1) {
568 btrfs_sysfs_remove_fsid(fs_devs
);
569 list_del(&fs_devs
->list
);
570 free_fs_devices(fs_devs
);
573 fs_devs
->num_devices
--;
574 list_del(&dev
->dev_list
);
575 rcu_string_free(dev
->name
);
584 * Add new device to list of registered devices
587 * 1 - first time device is seen
588 * 0 - device already known
591 static noinline
int device_list_add(const char *path
,
592 struct btrfs_super_block
*disk_super
,
593 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
595 struct btrfs_device
*device
;
596 struct btrfs_fs_devices
*fs_devices
;
597 struct rcu_string
*name
;
599 u64 found_transid
= btrfs_super_generation(disk_super
);
601 fs_devices
= find_fsid(disk_super
->fsid
);
603 fs_devices
= alloc_fs_devices(disk_super
->fsid
);
604 if (IS_ERR(fs_devices
))
605 return PTR_ERR(fs_devices
);
607 list_add(&fs_devices
->list
, &fs_uuids
);
611 device
= __find_device(&fs_devices
->devices
, devid
,
612 disk_super
->dev_item
.uuid
);
616 if (fs_devices
->opened
)
619 device
= btrfs_alloc_device(NULL
, &devid
,
620 disk_super
->dev_item
.uuid
);
621 if (IS_ERR(device
)) {
622 /* we can safely leave the fs_devices entry around */
623 return PTR_ERR(device
);
626 name
= rcu_string_strdup(path
, GFP_NOFS
);
631 rcu_assign_pointer(device
->name
, name
);
633 mutex_lock(&fs_devices
->device_list_mutex
);
634 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
635 fs_devices
->num_devices
++;
636 mutex_unlock(&fs_devices
->device_list_mutex
);
639 device
->fs_devices
= fs_devices
;
640 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
642 * When FS is already mounted.
643 * 1. If you are here and if the device->name is NULL that
644 * means this device was missing at time of FS mount.
645 * 2. If you are here and if the device->name is different
646 * from 'path' that means either
647 * a. The same device disappeared and reappeared with
649 * b. The missing-disk-which-was-replaced, has
652 * We must allow 1 and 2a above. But 2b would be a spurious
655 * Further in case of 1 and 2a above, the disk at 'path'
656 * would have missed some transaction when it was away and
657 * in case of 2a the stale bdev has to be updated as well.
658 * 2b must not be allowed at all time.
662 * For now, we do allow update to btrfs_fs_device through the
663 * btrfs dev scan cli after FS has been mounted. We're still
664 * tracking a problem where systems fail mount by subvolume id
665 * when we reject replacement on a mounted FS.
667 if (!fs_devices
->opened
&& found_transid
< device
->generation
) {
669 * That is if the FS is _not_ mounted and if you
670 * are here, that means there is more than one
671 * disk with same uuid and devid.We keep the one
672 * with larger generation number or the last-in if
673 * generation are equal.
678 name
= rcu_string_strdup(path
, GFP_NOFS
);
681 rcu_string_free(device
->name
);
682 rcu_assign_pointer(device
->name
, name
);
683 if (device
->missing
) {
684 fs_devices
->missing_devices
--;
690 * Unmount does not free the btrfs_device struct but would zero
691 * generation along with most of the other members. So just update
692 * it back. We need it to pick the disk with largest generation
695 if (!fs_devices
->opened
)
696 device
->generation
= found_transid
;
699 * if there is new btrfs on an already registered device,
700 * then remove the stale device entry.
702 btrfs_free_stale_device(device
);
704 *fs_devices_ret
= fs_devices
;
709 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
711 struct btrfs_fs_devices
*fs_devices
;
712 struct btrfs_device
*device
;
713 struct btrfs_device
*orig_dev
;
715 fs_devices
= alloc_fs_devices(orig
->fsid
);
716 if (IS_ERR(fs_devices
))
719 mutex_lock(&orig
->device_list_mutex
);
720 fs_devices
->total_devices
= orig
->total_devices
;
722 /* We have held the volume lock, it is safe to get the devices. */
723 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
724 struct rcu_string
*name
;
726 device
= btrfs_alloc_device(NULL
, &orig_dev
->devid
,
732 * This is ok to do without rcu read locked because we hold the
733 * uuid mutex so nothing we touch in here is going to disappear.
735 if (orig_dev
->name
) {
736 name
= rcu_string_strdup(orig_dev
->name
->str
, GFP_NOFS
);
741 rcu_assign_pointer(device
->name
, name
);
744 list_add(&device
->dev_list
, &fs_devices
->devices
);
745 device
->fs_devices
= fs_devices
;
746 fs_devices
->num_devices
++;
748 mutex_unlock(&orig
->device_list_mutex
);
751 mutex_unlock(&orig
->device_list_mutex
);
752 free_fs_devices(fs_devices
);
753 return ERR_PTR(-ENOMEM
);
756 void btrfs_close_extra_devices(struct btrfs_fs_devices
*fs_devices
, int step
)
758 struct btrfs_device
*device
, *next
;
759 struct btrfs_device
*latest_dev
= NULL
;
761 mutex_lock(&uuid_mutex
);
763 /* This is the initialized path, it is safe to release the devices. */
764 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
765 if (device
->in_fs_metadata
) {
766 if (!device
->is_tgtdev_for_dev_replace
&&
768 device
->generation
> latest_dev
->generation
)) {
774 if (device
->devid
== BTRFS_DEV_REPLACE_DEVID
) {
776 * In the first step, keep the device which has
777 * the correct fsid and the devid that is used
778 * for the dev_replace procedure.
779 * In the second step, the dev_replace state is
780 * read from the device tree and it is known
781 * whether the procedure is really active or
782 * not, which means whether this device is
783 * used or whether it should be removed.
785 if (step
== 0 || device
->is_tgtdev_for_dev_replace
) {
790 blkdev_put(device
->bdev
, device
->mode
);
792 fs_devices
->open_devices
--;
794 if (device
->writeable
) {
795 list_del_init(&device
->dev_alloc_list
);
796 device
->writeable
= 0;
797 if (!device
->is_tgtdev_for_dev_replace
)
798 fs_devices
->rw_devices
--;
800 list_del_init(&device
->dev_list
);
801 fs_devices
->num_devices
--;
802 rcu_string_free(device
->name
);
806 if (fs_devices
->seed
) {
807 fs_devices
= fs_devices
->seed
;
811 fs_devices
->latest_bdev
= latest_dev
->bdev
;
813 mutex_unlock(&uuid_mutex
);
816 static void __free_device(struct work_struct
*work
)
818 struct btrfs_device
*device
;
820 device
= container_of(work
, struct btrfs_device
, rcu_work
);
823 blkdev_put(device
->bdev
, device
->mode
);
825 rcu_string_free(device
->name
);
829 static void free_device(struct rcu_head
*head
)
831 struct btrfs_device
*device
;
833 device
= container_of(head
, struct btrfs_device
, rcu
);
835 INIT_WORK(&device
->rcu_work
, __free_device
);
836 schedule_work(&device
->rcu_work
);
839 static int __btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
841 struct btrfs_device
*device
, *tmp
;
843 if (--fs_devices
->opened
> 0)
846 mutex_lock(&fs_devices
->device_list_mutex
);
847 list_for_each_entry_safe(device
, tmp
, &fs_devices
->devices
, dev_list
) {
848 btrfs_close_one_device(device
);
850 mutex_unlock(&fs_devices
->device_list_mutex
);
852 WARN_ON(fs_devices
->open_devices
);
853 WARN_ON(fs_devices
->rw_devices
);
854 fs_devices
->opened
= 0;
855 fs_devices
->seeding
= 0;
860 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
862 struct btrfs_fs_devices
*seed_devices
= NULL
;
865 mutex_lock(&uuid_mutex
);
866 ret
= __btrfs_close_devices(fs_devices
);
867 if (!fs_devices
->opened
) {
868 seed_devices
= fs_devices
->seed
;
869 fs_devices
->seed
= NULL
;
871 mutex_unlock(&uuid_mutex
);
873 while (seed_devices
) {
874 fs_devices
= seed_devices
;
875 seed_devices
= fs_devices
->seed
;
876 __btrfs_close_devices(fs_devices
);
877 free_fs_devices(fs_devices
);
880 * Wait for rcu kworkers under __btrfs_close_devices
881 * to finish all blkdev_puts so device is really
882 * free when umount is done.
888 static int __btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
889 fmode_t flags
, void *holder
)
891 struct request_queue
*q
;
892 struct block_device
*bdev
;
893 struct list_head
*head
= &fs_devices
->devices
;
894 struct btrfs_device
*device
;
895 struct btrfs_device
*latest_dev
= NULL
;
896 struct buffer_head
*bh
;
897 struct btrfs_super_block
*disk_super
;
904 list_for_each_entry(device
, head
, dev_list
) {
910 /* Just open everything we can; ignore failures here */
911 if (btrfs_get_bdev_and_sb(device
->name
->str
, flags
, holder
, 1,
915 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
916 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
917 if (devid
!= device
->devid
)
920 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
,
924 device
->generation
= btrfs_super_generation(disk_super
);
926 device
->generation
> latest_dev
->generation
)
929 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
930 device
->writeable
= 0;
932 device
->writeable
= !bdev_read_only(bdev
);
936 q
= bdev_get_queue(bdev
);
937 if (blk_queue_discard(q
))
938 device
->can_discard
= 1;
941 device
->in_fs_metadata
= 0;
942 device
->mode
= flags
;
944 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
945 fs_devices
->rotating
= 1;
947 fs_devices
->open_devices
++;
948 if (device
->writeable
&&
949 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
950 fs_devices
->rw_devices
++;
951 list_add(&device
->dev_alloc_list
,
952 &fs_devices
->alloc_list
);
959 blkdev_put(bdev
, flags
);
962 if (fs_devices
->open_devices
== 0) {
966 fs_devices
->seeding
= seeding
;
967 fs_devices
->opened
= 1;
968 fs_devices
->latest_bdev
= latest_dev
->bdev
;
969 fs_devices
->total_rw_bytes
= 0;
974 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
975 fmode_t flags
, void *holder
)
979 mutex_lock(&uuid_mutex
);
980 if (fs_devices
->opened
) {
981 fs_devices
->opened
++;
984 ret
= __btrfs_open_devices(fs_devices
, flags
, holder
);
986 mutex_unlock(&uuid_mutex
);
991 * Look for a btrfs signature on a device. This may be called out of the mount path
992 * and we are not allowed to call set_blocksize during the scan. The superblock
993 * is read via pagecache
995 int btrfs_scan_one_device(const char *path
, fmode_t flags
, void *holder
,
996 struct btrfs_fs_devices
**fs_devices_ret
)
998 struct btrfs_super_block
*disk_super
;
999 struct block_device
*bdev
;
1010 * we would like to check all the supers, but that would make
1011 * a btrfs mount succeed after a mkfs from a different FS.
1012 * So, we need to add a special mount option to scan for
1013 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1015 bytenr
= btrfs_sb_offset(0);
1016 flags
|= FMODE_EXCL
;
1017 mutex_lock(&uuid_mutex
);
1019 bdev
= blkdev_get_by_path(path
, flags
, holder
);
1022 ret
= PTR_ERR(bdev
);
1026 /* make sure our super fits in the device */
1027 if (bytenr
+ PAGE_CACHE_SIZE
>= i_size_read(bdev
->bd_inode
))
1028 goto error_bdev_put
;
1030 /* make sure our super fits in the page */
1031 if (sizeof(*disk_super
) > PAGE_CACHE_SIZE
)
1032 goto error_bdev_put
;
1034 /* make sure our super doesn't straddle pages on disk */
1035 index
= bytenr
>> PAGE_CACHE_SHIFT
;
1036 if ((bytenr
+ sizeof(*disk_super
) - 1) >> PAGE_CACHE_SHIFT
!= index
)
1037 goto error_bdev_put
;
1039 /* pull in the page with our super */
1040 page
= read_cache_page_gfp(bdev
->bd_inode
->i_mapping
,
1043 if (IS_ERR_OR_NULL(page
))
1044 goto error_bdev_put
;
1048 /* align our pointer to the offset of the super block */
1049 disk_super
= p
+ (bytenr
& ~PAGE_CACHE_MASK
);
1051 if (btrfs_super_bytenr(disk_super
) != bytenr
||
1052 btrfs_super_magic(disk_super
) != BTRFS_MAGIC
)
1055 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1056 transid
= btrfs_super_generation(disk_super
);
1057 total_devices
= btrfs_super_num_devices(disk_super
);
1059 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
1061 if (disk_super
->label
[0]) {
1062 if (disk_super
->label
[BTRFS_LABEL_SIZE
- 1])
1063 disk_super
->label
[BTRFS_LABEL_SIZE
- 1] = '\0';
1064 printk(KERN_INFO
"BTRFS: device label %s ", disk_super
->label
);
1066 printk(KERN_INFO
"BTRFS: device fsid %pU ", disk_super
->fsid
);
1069 printk(KERN_CONT
"devid %llu transid %llu %s\n", devid
, transid
, path
);
1072 if (!ret
&& fs_devices_ret
)
1073 (*fs_devices_ret
)->total_devices
= total_devices
;
1077 page_cache_release(page
);
1080 blkdev_put(bdev
, flags
);
1082 mutex_unlock(&uuid_mutex
);
1086 /* helper to account the used device space in the range */
1087 int btrfs_account_dev_extents_size(struct btrfs_device
*device
, u64 start
,
1088 u64 end
, u64
*length
)
1090 struct btrfs_key key
;
1091 struct btrfs_root
*root
= device
->dev_root
;
1092 struct btrfs_dev_extent
*dev_extent
;
1093 struct btrfs_path
*path
;
1097 struct extent_buffer
*l
;
1101 if (start
>= device
->total_bytes
|| device
->is_tgtdev_for_dev_replace
)
1104 path
= btrfs_alloc_path();
1109 key
.objectid
= device
->devid
;
1111 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1113 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1117 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1124 slot
= path
->slots
[0];
1125 if (slot
>= btrfs_header_nritems(l
)) {
1126 ret
= btrfs_next_leaf(root
, path
);
1134 btrfs_item_key_to_cpu(l
, &key
, slot
);
1136 if (key
.objectid
< device
->devid
)
1139 if (key
.objectid
> device
->devid
)
1142 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1145 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1146 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1148 if (key
.offset
<= start
&& extent_end
> end
) {
1149 *length
= end
- start
+ 1;
1151 } else if (key
.offset
<= start
&& extent_end
> start
)
1152 *length
+= extent_end
- start
;
1153 else if (key
.offset
> start
&& extent_end
<= end
)
1154 *length
+= extent_end
- key
.offset
;
1155 else if (key
.offset
> start
&& key
.offset
<= end
) {
1156 *length
+= end
- key
.offset
+ 1;
1158 } else if (key
.offset
> end
)
1166 btrfs_free_path(path
);
1170 static int contains_pending_extent(struct btrfs_transaction
*transaction
,
1171 struct btrfs_device
*device
,
1172 u64
*start
, u64 len
)
1174 struct btrfs_fs_info
*fs_info
= device
->dev_root
->fs_info
;
1175 struct extent_map
*em
;
1176 struct list_head
*search_list
= &fs_info
->pinned_chunks
;
1178 u64 physical_start
= *start
;
1181 search_list
= &transaction
->pending_chunks
;
1183 list_for_each_entry(em
, search_list
, list
) {
1184 struct map_lookup
*map
;
1187 map
= (struct map_lookup
*)em
->bdev
;
1188 for (i
= 0; i
< map
->num_stripes
; i
++) {
1191 if (map
->stripes
[i
].dev
!= device
)
1193 if (map
->stripes
[i
].physical
>= physical_start
+ len
||
1194 map
->stripes
[i
].physical
+ em
->orig_block_len
<=
1198 * Make sure that while processing the pinned list we do
1199 * not override our *start with a lower value, because
1200 * we can have pinned chunks that fall within this
1201 * device hole and that have lower physical addresses
1202 * than the pending chunks we processed before. If we
1203 * do not take this special care we can end up getting
1204 * 2 pending chunks that start at the same physical
1205 * device offsets because the end offset of a pinned
1206 * chunk can be equal to the start offset of some
1209 end
= map
->stripes
[i
].physical
+ em
->orig_block_len
;
1216 if (search_list
!= &fs_info
->pinned_chunks
) {
1217 search_list
= &fs_info
->pinned_chunks
;
1226 * find_free_dev_extent_start - find free space in the specified device
1227 * @device: the device which we search the free space in
1228 * @num_bytes: the size of the free space that we need
1229 * @search_start: the position from which to begin the search
1230 * @start: store the start of the free space.
1231 * @len: the size of the free space. that we find, or the size
1232 * of the max free space if we don't find suitable free space
1234 * this uses a pretty simple search, the expectation is that it is
1235 * called very infrequently and that a given device has a small number
1238 * @start is used to store the start of the free space if we find. But if we
1239 * don't find suitable free space, it will be used to store the start position
1240 * of the max free space.
1242 * @len is used to store the size of the free space that we find.
1243 * But if we don't find suitable free space, it is used to store the size of
1244 * the max free space.
1246 int find_free_dev_extent_start(struct btrfs_transaction
*transaction
,
1247 struct btrfs_device
*device
, u64 num_bytes
,
1248 u64 search_start
, u64
*start
, u64
*len
)
1250 struct btrfs_key key
;
1251 struct btrfs_root
*root
= device
->dev_root
;
1252 struct btrfs_dev_extent
*dev_extent
;
1253 struct btrfs_path
*path
;
1258 u64 search_end
= device
->total_bytes
;
1261 struct extent_buffer
*l
;
1262 u64 min_search_start
;
1265 * We don't want to overwrite the superblock on the drive nor any area
1266 * used by the boot loader (grub for example), so we make sure to start
1267 * at an offset of at least 1MB.
1269 min_search_start
= max(root
->fs_info
->alloc_start
, 1024ull * 1024);
1270 search_start
= max(search_start
, min_search_start
);
1272 path
= btrfs_alloc_path();
1276 max_hole_start
= search_start
;
1280 if (search_start
>= search_end
|| device
->is_tgtdev_for_dev_replace
) {
1286 path
->search_commit_root
= 1;
1287 path
->skip_locking
= 1;
1289 key
.objectid
= device
->devid
;
1290 key
.offset
= search_start
;
1291 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1293 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1297 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1304 slot
= path
->slots
[0];
1305 if (slot
>= btrfs_header_nritems(l
)) {
1306 ret
= btrfs_next_leaf(root
, path
);
1314 btrfs_item_key_to_cpu(l
, &key
, slot
);
1316 if (key
.objectid
< device
->devid
)
1319 if (key
.objectid
> device
->devid
)
1322 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1325 if (key
.offset
> search_start
) {
1326 hole_size
= key
.offset
- search_start
;
1329 * Have to check before we set max_hole_start, otherwise
1330 * we could end up sending back this offset anyway.
1332 if (contains_pending_extent(transaction
, device
,
1335 if (key
.offset
>= search_start
) {
1336 hole_size
= key
.offset
- search_start
;
1343 if (hole_size
> max_hole_size
) {
1344 max_hole_start
= search_start
;
1345 max_hole_size
= hole_size
;
1349 * If this free space is greater than which we need,
1350 * it must be the max free space that we have found
1351 * until now, so max_hole_start must point to the start
1352 * of this free space and the length of this free space
1353 * is stored in max_hole_size. Thus, we return
1354 * max_hole_start and max_hole_size and go back to the
1357 if (hole_size
>= num_bytes
) {
1363 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1364 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1366 if (extent_end
> search_start
)
1367 search_start
= extent_end
;
1374 * At this point, search_start should be the end of
1375 * allocated dev extents, and when shrinking the device,
1376 * search_end may be smaller than search_start.
1378 if (search_end
> search_start
) {
1379 hole_size
= search_end
- search_start
;
1381 if (contains_pending_extent(transaction
, device
, &search_start
,
1383 btrfs_release_path(path
);
1387 if (hole_size
> max_hole_size
) {
1388 max_hole_start
= search_start
;
1389 max_hole_size
= hole_size
;
1394 if (max_hole_size
< num_bytes
)
1400 btrfs_free_path(path
);
1401 *start
= max_hole_start
;
1403 *len
= max_hole_size
;
1407 int find_free_dev_extent(struct btrfs_trans_handle
*trans
,
1408 struct btrfs_device
*device
, u64 num_bytes
,
1409 u64
*start
, u64
*len
)
1411 /* FIXME use last free of some kind */
1412 return find_free_dev_extent_start(trans
->transaction
, device
,
1413 num_bytes
, 0, start
, len
);
1416 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1417 struct btrfs_device
*device
,
1418 u64 start
, u64
*dev_extent_len
)
1421 struct btrfs_path
*path
;
1422 struct btrfs_root
*root
= device
->dev_root
;
1423 struct btrfs_key key
;
1424 struct btrfs_key found_key
;
1425 struct extent_buffer
*leaf
= NULL
;
1426 struct btrfs_dev_extent
*extent
= NULL
;
1428 path
= btrfs_alloc_path();
1432 key
.objectid
= device
->devid
;
1434 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1436 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1438 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1439 BTRFS_DEV_EXTENT_KEY
);
1442 leaf
= path
->nodes
[0];
1443 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1444 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1445 struct btrfs_dev_extent
);
1446 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1447 btrfs_dev_extent_length(leaf
, extent
) < start
);
1449 btrfs_release_path(path
);
1451 } else if (ret
== 0) {
1452 leaf
= path
->nodes
[0];
1453 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1454 struct btrfs_dev_extent
);
1456 btrfs_std_error(root
->fs_info
, ret
, "Slot search failed");
1460 *dev_extent_len
= btrfs_dev_extent_length(leaf
, extent
);
1462 ret
= btrfs_del_item(trans
, root
, path
);
1464 btrfs_std_error(root
->fs_info
, ret
,
1465 "Failed to remove dev extent item");
1467 set_bit(BTRFS_TRANS_HAVE_FREE_BGS
, &trans
->transaction
->flags
);
1470 btrfs_free_path(path
);
1474 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1475 struct btrfs_device
*device
,
1476 u64 chunk_tree
, u64 chunk_objectid
,
1477 u64 chunk_offset
, u64 start
, u64 num_bytes
)
1480 struct btrfs_path
*path
;
1481 struct btrfs_root
*root
= device
->dev_root
;
1482 struct btrfs_dev_extent
*extent
;
1483 struct extent_buffer
*leaf
;
1484 struct btrfs_key key
;
1486 WARN_ON(!device
->in_fs_metadata
);
1487 WARN_ON(device
->is_tgtdev_for_dev_replace
);
1488 path
= btrfs_alloc_path();
1492 key
.objectid
= device
->devid
;
1494 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1495 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1500 leaf
= path
->nodes
[0];
1501 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1502 struct btrfs_dev_extent
);
1503 btrfs_set_dev_extent_chunk_tree(leaf
, extent
, chunk_tree
);
1504 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
, chunk_objectid
);
1505 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1507 write_extent_buffer(leaf
, root
->fs_info
->chunk_tree_uuid
,
1508 btrfs_dev_extent_chunk_tree_uuid(extent
), BTRFS_UUID_SIZE
);
1510 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1511 btrfs_mark_buffer_dirty(leaf
);
1513 btrfs_free_path(path
);
1517 static u64
find_next_chunk(struct btrfs_fs_info
*fs_info
)
1519 struct extent_map_tree
*em_tree
;
1520 struct extent_map
*em
;
1524 em_tree
= &fs_info
->mapping_tree
.map_tree
;
1525 read_lock(&em_tree
->lock
);
1526 n
= rb_last(&em_tree
->map
);
1528 em
= rb_entry(n
, struct extent_map
, rb_node
);
1529 ret
= em
->start
+ em
->len
;
1531 read_unlock(&em_tree
->lock
);
1536 static noinline
int find_next_devid(struct btrfs_fs_info
*fs_info
,
1540 struct btrfs_key key
;
1541 struct btrfs_key found_key
;
1542 struct btrfs_path
*path
;
1544 path
= btrfs_alloc_path();
1548 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1549 key
.type
= BTRFS_DEV_ITEM_KEY
;
1550 key
.offset
= (u64
)-1;
1552 ret
= btrfs_search_slot(NULL
, fs_info
->chunk_root
, &key
, path
, 0, 0);
1556 BUG_ON(ret
== 0); /* Corruption */
1558 ret
= btrfs_previous_item(fs_info
->chunk_root
, path
,
1559 BTRFS_DEV_ITEMS_OBJECTID
,
1560 BTRFS_DEV_ITEM_KEY
);
1564 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1566 *devid_ret
= found_key
.offset
+ 1;
1570 btrfs_free_path(path
);
1575 * the device information is stored in the chunk root
1576 * the btrfs_device struct should be fully filled in
1578 static int btrfs_add_device(struct btrfs_trans_handle
*trans
,
1579 struct btrfs_root
*root
,
1580 struct btrfs_device
*device
)
1583 struct btrfs_path
*path
;
1584 struct btrfs_dev_item
*dev_item
;
1585 struct extent_buffer
*leaf
;
1586 struct btrfs_key key
;
1589 root
= root
->fs_info
->chunk_root
;
1591 path
= btrfs_alloc_path();
1595 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1596 key
.type
= BTRFS_DEV_ITEM_KEY
;
1597 key
.offset
= device
->devid
;
1599 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1604 leaf
= path
->nodes
[0];
1605 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1607 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1608 btrfs_set_device_generation(leaf
, dev_item
, 0);
1609 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1610 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1611 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1612 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1613 btrfs_set_device_total_bytes(leaf
, dev_item
,
1614 btrfs_device_get_disk_total_bytes(device
));
1615 btrfs_set_device_bytes_used(leaf
, dev_item
,
1616 btrfs_device_get_bytes_used(device
));
1617 btrfs_set_device_group(leaf
, dev_item
, 0);
1618 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1619 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1620 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1622 ptr
= btrfs_device_uuid(dev_item
);
1623 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1624 ptr
= btrfs_device_fsid(dev_item
);
1625 write_extent_buffer(leaf
, root
->fs_info
->fsid
, ptr
, BTRFS_UUID_SIZE
);
1626 btrfs_mark_buffer_dirty(leaf
);
1630 btrfs_free_path(path
);
1635 * Function to update ctime/mtime for a given device path.
1636 * Mainly used for ctime/mtime based probe like libblkid.
1638 static void update_dev_time(char *path_name
)
1642 filp
= filp_open(path_name
, O_RDWR
, 0);
1645 file_update_time(filp
);
1646 filp_close(filp
, NULL
);
1650 static int btrfs_rm_dev_item(struct btrfs_root
*root
,
1651 struct btrfs_device
*device
)
1654 struct btrfs_path
*path
;
1655 struct btrfs_key key
;
1656 struct btrfs_trans_handle
*trans
;
1658 root
= root
->fs_info
->chunk_root
;
1660 path
= btrfs_alloc_path();
1664 trans
= btrfs_start_transaction(root
, 0);
1665 if (IS_ERR(trans
)) {
1666 btrfs_free_path(path
);
1667 return PTR_ERR(trans
);
1669 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1670 key
.type
= BTRFS_DEV_ITEM_KEY
;
1671 key
.offset
= device
->devid
;
1673 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1682 ret
= btrfs_del_item(trans
, root
, path
);
1686 btrfs_free_path(path
);
1687 btrfs_commit_transaction(trans
, root
);
1691 int btrfs_rm_device(struct btrfs_root
*root
, char *device_path
)
1693 struct btrfs_device
*device
;
1694 struct btrfs_device
*next_device
;
1695 struct block_device
*bdev
;
1696 struct buffer_head
*bh
= NULL
;
1697 struct btrfs_super_block
*disk_super
;
1698 struct btrfs_fs_devices
*cur_devices
;
1705 bool clear_super
= false;
1707 mutex_lock(&uuid_mutex
);
1710 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
1712 all_avail
= root
->fs_info
->avail_data_alloc_bits
|
1713 root
->fs_info
->avail_system_alloc_bits
|
1714 root
->fs_info
->avail_metadata_alloc_bits
;
1715 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
1717 num_devices
= root
->fs_info
->fs_devices
->num_devices
;
1718 btrfs_dev_replace_lock(&root
->fs_info
->dev_replace
);
1719 if (btrfs_dev_replace_is_ongoing(&root
->fs_info
->dev_replace
)) {
1720 WARN_ON(num_devices
< 1);
1723 btrfs_dev_replace_unlock(&root
->fs_info
->dev_replace
);
1725 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID10
) && num_devices
<= 4) {
1726 ret
= BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET
;
1730 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID1
) && num_devices
<= 2) {
1731 ret
= BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET
;
1735 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID5
) &&
1736 root
->fs_info
->fs_devices
->rw_devices
<= 2) {
1737 ret
= BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET
;
1740 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID6
) &&
1741 root
->fs_info
->fs_devices
->rw_devices
<= 3) {
1742 ret
= BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET
;
1746 if (strcmp(device_path
, "missing") == 0) {
1747 struct list_head
*devices
;
1748 struct btrfs_device
*tmp
;
1751 devices
= &root
->fs_info
->fs_devices
->devices
;
1753 * It is safe to read the devices since the volume_mutex
1756 list_for_each_entry(tmp
, devices
, dev_list
) {
1757 if (tmp
->in_fs_metadata
&&
1758 !tmp
->is_tgtdev_for_dev_replace
&&
1768 ret
= BTRFS_ERROR_DEV_MISSING_NOT_FOUND
;
1772 ret
= btrfs_get_bdev_and_sb(device_path
,
1773 FMODE_WRITE
| FMODE_EXCL
,
1774 root
->fs_info
->bdev_holder
, 0,
1778 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1779 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1780 dev_uuid
= disk_super
->dev_item
.uuid
;
1781 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1789 if (device
->is_tgtdev_for_dev_replace
) {
1790 ret
= BTRFS_ERROR_DEV_TGT_REPLACE
;
1794 if (device
->writeable
&& root
->fs_info
->fs_devices
->rw_devices
== 1) {
1795 ret
= BTRFS_ERROR_DEV_ONLY_WRITABLE
;
1799 if (device
->writeable
) {
1801 list_del_init(&device
->dev_alloc_list
);
1802 device
->fs_devices
->rw_devices
--;
1803 unlock_chunks(root
);
1807 mutex_unlock(&uuid_mutex
);
1808 ret
= btrfs_shrink_device(device
, 0);
1809 mutex_lock(&uuid_mutex
);
1814 * TODO: the superblock still includes this device in its num_devices
1815 * counter although write_all_supers() is not locked out. This
1816 * could give a filesystem state which requires a degraded mount.
1818 ret
= btrfs_rm_dev_item(root
->fs_info
->chunk_root
, device
);
1822 device
->in_fs_metadata
= 0;
1823 btrfs_scrub_cancel_dev(root
->fs_info
, device
);
1826 * the device list mutex makes sure that we don't change
1827 * the device list while someone else is writing out all
1828 * the device supers. Whoever is writing all supers, should
1829 * lock the device list mutex before getting the number of
1830 * devices in the super block (super_copy). Conversely,
1831 * whoever updates the number of devices in the super block
1832 * (super_copy) should hold the device list mutex.
1835 cur_devices
= device
->fs_devices
;
1836 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1837 list_del_rcu(&device
->dev_list
);
1839 device
->fs_devices
->num_devices
--;
1840 device
->fs_devices
->total_devices
--;
1842 if (device
->missing
)
1843 device
->fs_devices
->missing_devices
--;
1845 next_device
= list_entry(root
->fs_info
->fs_devices
->devices
.next
,
1846 struct btrfs_device
, dev_list
);
1847 if (device
->bdev
== root
->fs_info
->sb
->s_bdev
)
1848 root
->fs_info
->sb
->s_bdev
= next_device
->bdev
;
1849 if (device
->bdev
== root
->fs_info
->fs_devices
->latest_bdev
)
1850 root
->fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1853 device
->fs_devices
->open_devices
--;
1854 /* remove sysfs entry */
1855 btrfs_sysfs_rm_device_link(root
->fs_info
->fs_devices
, device
);
1858 call_rcu(&device
->rcu
, free_device
);
1860 num_devices
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
1861 btrfs_set_super_num_devices(root
->fs_info
->super_copy
, num_devices
);
1862 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1864 if (cur_devices
->open_devices
== 0) {
1865 struct btrfs_fs_devices
*fs_devices
;
1866 fs_devices
= root
->fs_info
->fs_devices
;
1867 while (fs_devices
) {
1868 if (fs_devices
->seed
== cur_devices
) {
1869 fs_devices
->seed
= cur_devices
->seed
;
1872 fs_devices
= fs_devices
->seed
;
1874 cur_devices
->seed
= NULL
;
1875 __btrfs_close_devices(cur_devices
);
1876 free_fs_devices(cur_devices
);
1879 root
->fs_info
->num_tolerated_disk_barrier_failures
=
1880 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
1883 * at this point, the device is zero sized. We want to
1884 * remove it from the devices list and zero out the old super
1886 if (clear_super
&& disk_super
) {
1890 /* make sure this device isn't detected as part of
1893 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
1894 set_buffer_dirty(bh
);
1895 sync_dirty_buffer(bh
);
1897 /* clear the mirror copies of super block on the disk
1898 * being removed, 0th copy is been taken care above and
1899 * the below would take of the rest
1901 for (i
= 1; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
1902 bytenr
= btrfs_sb_offset(i
);
1903 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
1904 i_size_read(bdev
->bd_inode
))
1908 bh
= __bread(bdev
, bytenr
/ 4096,
1909 BTRFS_SUPER_INFO_SIZE
);
1913 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1915 if (btrfs_super_bytenr(disk_super
) != bytenr
||
1916 btrfs_super_magic(disk_super
) != BTRFS_MAGIC
) {
1919 memset(&disk_super
->magic
, 0,
1920 sizeof(disk_super
->magic
));
1921 set_buffer_dirty(bh
);
1922 sync_dirty_buffer(bh
);
1929 /* Notify udev that device has changed */
1930 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
1932 /* Update ctime/mtime for device path for libblkid */
1933 update_dev_time(device_path
);
1939 blkdev_put(bdev
, FMODE_READ
| FMODE_EXCL
);
1941 mutex_unlock(&uuid_mutex
);
1944 if (device
->writeable
) {
1946 list_add(&device
->dev_alloc_list
,
1947 &root
->fs_info
->fs_devices
->alloc_list
);
1948 device
->fs_devices
->rw_devices
++;
1949 unlock_chunks(root
);
1954 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info
*fs_info
,
1955 struct btrfs_device
*srcdev
)
1957 struct btrfs_fs_devices
*fs_devices
;
1959 WARN_ON(!mutex_is_locked(&fs_info
->fs_devices
->device_list_mutex
));
1962 * in case of fs with no seed, srcdev->fs_devices will point
1963 * to fs_devices of fs_info. However when the dev being replaced is
1964 * a seed dev it will point to the seed's local fs_devices. In short
1965 * srcdev will have its correct fs_devices in both the cases.
1967 fs_devices
= srcdev
->fs_devices
;
1969 list_del_rcu(&srcdev
->dev_list
);
1970 list_del_rcu(&srcdev
->dev_alloc_list
);
1971 fs_devices
->num_devices
--;
1972 if (srcdev
->missing
)
1973 fs_devices
->missing_devices
--;
1975 if (srcdev
->writeable
) {
1976 fs_devices
->rw_devices
--;
1977 /* zero out the old super if it is writable */
1978 btrfs_scratch_superblocks(srcdev
->bdev
, srcdev
->name
->str
);
1982 fs_devices
->open_devices
--;
1985 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info
*fs_info
,
1986 struct btrfs_device
*srcdev
)
1988 struct btrfs_fs_devices
*fs_devices
= srcdev
->fs_devices
;
1990 call_rcu(&srcdev
->rcu
, free_device
);
1993 * unless fs_devices is seed fs, num_devices shouldn't go
1996 BUG_ON(!fs_devices
->num_devices
&& !fs_devices
->seeding
);
1998 /* if this is no devs we rather delete the fs_devices */
1999 if (!fs_devices
->num_devices
) {
2000 struct btrfs_fs_devices
*tmp_fs_devices
;
2002 tmp_fs_devices
= fs_info
->fs_devices
;
2003 while (tmp_fs_devices
) {
2004 if (tmp_fs_devices
->seed
== fs_devices
) {
2005 tmp_fs_devices
->seed
= fs_devices
->seed
;
2008 tmp_fs_devices
= tmp_fs_devices
->seed
;
2010 fs_devices
->seed
= NULL
;
2011 __btrfs_close_devices(fs_devices
);
2012 free_fs_devices(fs_devices
);
2016 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
2017 struct btrfs_device
*tgtdev
)
2019 struct btrfs_device
*next_device
;
2021 mutex_lock(&uuid_mutex
);
2023 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2025 btrfs_sysfs_rm_device_link(fs_info
->fs_devices
, tgtdev
);
2028 btrfs_scratch_superblocks(tgtdev
->bdev
, tgtdev
->name
->str
);
2029 fs_info
->fs_devices
->open_devices
--;
2031 fs_info
->fs_devices
->num_devices
--;
2033 next_device
= list_entry(fs_info
->fs_devices
->devices
.next
,
2034 struct btrfs_device
, dev_list
);
2035 if (tgtdev
->bdev
== fs_info
->sb
->s_bdev
)
2036 fs_info
->sb
->s_bdev
= next_device
->bdev
;
2037 if (tgtdev
->bdev
== fs_info
->fs_devices
->latest_bdev
)
2038 fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
2039 list_del_rcu(&tgtdev
->dev_list
);
2041 call_rcu(&tgtdev
->rcu
, free_device
);
2043 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2044 mutex_unlock(&uuid_mutex
);
2047 static int btrfs_find_device_by_path(struct btrfs_root
*root
, char *device_path
,
2048 struct btrfs_device
**device
)
2051 struct btrfs_super_block
*disk_super
;
2054 struct block_device
*bdev
;
2055 struct buffer_head
*bh
;
2058 ret
= btrfs_get_bdev_and_sb(device_path
, FMODE_READ
,
2059 root
->fs_info
->bdev_holder
, 0, &bdev
, &bh
);
2062 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
2063 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
2064 dev_uuid
= disk_super
->dev_item
.uuid
;
2065 *device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
2070 blkdev_put(bdev
, FMODE_READ
);
2074 int btrfs_find_device_missing_or_by_path(struct btrfs_root
*root
,
2076 struct btrfs_device
**device
)
2079 if (strcmp(device_path
, "missing") == 0) {
2080 struct list_head
*devices
;
2081 struct btrfs_device
*tmp
;
2083 devices
= &root
->fs_info
->fs_devices
->devices
;
2085 * It is safe to read the devices since the volume_mutex
2086 * is held by the caller.
2088 list_for_each_entry(tmp
, devices
, dev_list
) {
2089 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
2096 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND
;
2100 return btrfs_find_device_by_path(root
, device_path
, device
);
2105 * does all the dirty work required for changing file system's UUID.
2107 static int btrfs_prepare_sprout(struct btrfs_root
*root
)
2109 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
2110 struct btrfs_fs_devices
*old_devices
;
2111 struct btrfs_fs_devices
*seed_devices
;
2112 struct btrfs_super_block
*disk_super
= root
->fs_info
->super_copy
;
2113 struct btrfs_device
*device
;
2116 BUG_ON(!mutex_is_locked(&uuid_mutex
));
2117 if (!fs_devices
->seeding
)
2120 seed_devices
= __alloc_fs_devices();
2121 if (IS_ERR(seed_devices
))
2122 return PTR_ERR(seed_devices
);
2124 old_devices
= clone_fs_devices(fs_devices
);
2125 if (IS_ERR(old_devices
)) {
2126 kfree(seed_devices
);
2127 return PTR_ERR(old_devices
);
2130 list_add(&old_devices
->list
, &fs_uuids
);
2132 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
2133 seed_devices
->opened
= 1;
2134 INIT_LIST_HEAD(&seed_devices
->devices
);
2135 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
2136 mutex_init(&seed_devices
->device_list_mutex
);
2138 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2139 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
2141 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
)
2142 device
->fs_devices
= seed_devices
;
2145 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
2146 unlock_chunks(root
);
2148 fs_devices
->seeding
= 0;
2149 fs_devices
->num_devices
= 0;
2150 fs_devices
->open_devices
= 0;
2151 fs_devices
->missing_devices
= 0;
2152 fs_devices
->rotating
= 0;
2153 fs_devices
->seed
= seed_devices
;
2155 generate_random_uuid(fs_devices
->fsid
);
2156 memcpy(root
->fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2157 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2158 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2160 super_flags
= btrfs_super_flags(disk_super
) &
2161 ~BTRFS_SUPER_FLAG_SEEDING
;
2162 btrfs_set_super_flags(disk_super
, super_flags
);
2168 * strore the expected generation for seed devices in device items.
2170 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
2171 struct btrfs_root
*root
)
2173 struct btrfs_path
*path
;
2174 struct extent_buffer
*leaf
;
2175 struct btrfs_dev_item
*dev_item
;
2176 struct btrfs_device
*device
;
2177 struct btrfs_key key
;
2178 u8 fs_uuid
[BTRFS_UUID_SIZE
];
2179 u8 dev_uuid
[BTRFS_UUID_SIZE
];
2183 path
= btrfs_alloc_path();
2187 root
= root
->fs_info
->chunk_root
;
2188 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2190 key
.type
= BTRFS_DEV_ITEM_KEY
;
2193 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2197 leaf
= path
->nodes
[0];
2199 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
2200 ret
= btrfs_next_leaf(root
, path
);
2205 leaf
= path
->nodes
[0];
2206 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2207 btrfs_release_path(path
);
2211 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2212 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
2213 key
.type
!= BTRFS_DEV_ITEM_KEY
)
2216 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2217 struct btrfs_dev_item
);
2218 devid
= btrfs_device_id(leaf
, dev_item
);
2219 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
2221 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
2223 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
2225 BUG_ON(!device
); /* Logic error */
2227 if (device
->fs_devices
->seeding
) {
2228 btrfs_set_device_generation(leaf
, dev_item
,
2229 device
->generation
);
2230 btrfs_mark_buffer_dirty(leaf
);
2238 btrfs_free_path(path
);
2242 int btrfs_init_new_device(struct btrfs_root
*root
, char *device_path
)
2244 struct request_queue
*q
;
2245 struct btrfs_trans_handle
*trans
;
2246 struct btrfs_device
*device
;
2247 struct block_device
*bdev
;
2248 struct list_head
*devices
;
2249 struct super_block
*sb
= root
->fs_info
->sb
;
2250 struct rcu_string
*name
;
2252 int seeding_dev
= 0;
2255 if ((sb
->s_flags
& MS_RDONLY
) && !root
->fs_info
->fs_devices
->seeding
)
2258 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2259 root
->fs_info
->bdev_holder
);
2261 return PTR_ERR(bdev
);
2263 if (root
->fs_info
->fs_devices
->seeding
) {
2265 down_write(&sb
->s_umount
);
2266 mutex_lock(&uuid_mutex
);
2269 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2271 devices
= &root
->fs_info
->fs_devices
->devices
;
2273 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2274 list_for_each_entry(device
, devices
, dev_list
) {
2275 if (device
->bdev
== bdev
) {
2278 &root
->fs_info
->fs_devices
->device_list_mutex
);
2282 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2284 device
= btrfs_alloc_device(root
->fs_info
, NULL
, NULL
);
2285 if (IS_ERR(device
)) {
2286 /* we can safely leave the fs_devices entry around */
2287 ret
= PTR_ERR(device
);
2291 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2297 rcu_assign_pointer(device
->name
, name
);
2299 trans
= btrfs_start_transaction(root
, 0);
2300 if (IS_ERR(trans
)) {
2301 rcu_string_free(device
->name
);
2303 ret
= PTR_ERR(trans
);
2307 q
= bdev_get_queue(bdev
);
2308 if (blk_queue_discard(q
))
2309 device
->can_discard
= 1;
2310 device
->writeable
= 1;
2311 device
->generation
= trans
->transid
;
2312 device
->io_width
= root
->sectorsize
;
2313 device
->io_align
= root
->sectorsize
;
2314 device
->sector_size
= root
->sectorsize
;
2315 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
2316 device
->disk_total_bytes
= device
->total_bytes
;
2317 device
->commit_total_bytes
= device
->total_bytes
;
2318 device
->dev_root
= root
->fs_info
->dev_root
;
2319 device
->bdev
= bdev
;
2320 device
->in_fs_metadata
= 1;
2321 device
->is_tgtdev_for_dev_replace
= 0;
2322 device
->mode
= FMODE_EXCL
;
2323 device
->dev_stats_valid
= 1;
2324 set_blocksize(device
->bdev
, 4096);
2327 sb
->s_flags
&= ~MS_RDONLY
;
2328 ret
= btrfs_prepare_sprout(root
);
2329 BUG_ON(ret
); /* -ENOMEM */
2332 device
->fs_devices
= root
->fs_info
->fs_devices
;
2334 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2336 list_add_rcu(&device
->dev_list
, &root
->fs_info
->fs_devices
->devices
);
2337 list_add(&device
->dev_alloc_list
,
2338 &root
->fs_info
->fs_devices
->alloc_list
);
2339 root
->fs_info
->fs_devices
->num_devices
++;
2340 root
->fs_info
->fs_devices
->open_devices
++;
2341 root
->fs_info
->fs_devices
->rw_devices
++;
2342 root
->fs_info
->fs_devices
->total_devices
++;
2343 root
->fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
2345 spin_lock(&root
->fs_info
->free_chunk_lock
);
2346 root
->fs_info
->free_chunk_space
+= device
->total_bytes
;
2347 spin_unlock(&root
->fs_info
->free_chunk_lock
);
2349 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
2350 root
->fs_info
->fs_devices
->rotating
= 1;
2352 tmp
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
2353 btrfs_set_super_total_bytes(root
->fs_info
->super_copy
,
2354 tmp
+ device
->total_bytes
);
2356 tmp
= btrfs_super_num_devices(root
->fs_info
->super_copy
);
2357 btrfs_set_super_num_devices(root
->fs_info
->super_copy
,
2360 /* add sysfs device entry */
2361 btrfs_sysfs_add_device_link(root
->fs_info
->fs_devices
, device
);
2364 * we've got more storage, clear any full flags on the space
2367 btrfs_clear_space_info_full(root
->fs_info
);
2369 unlock_chunks(root
);
2370 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2374 ret
= init_first_rw_device(trans
, root
, device
);
2375 unlock_chunks(root
);
2377 btrfs_abort_transaction(trans
, root
, ret
);
2382 ret
= btrfs_add_device(trans
, root
, device
);
2384 btrfs_abort_transaction(trans
, root
, ret
);
2389 char fsid_buf
[BTRFS_UUID_UNPARSED_SIZE
];
2391 ret
= btrfs_finish_sprout(trans
, root
);
2393 btrfs_abort_transaction(trans
, root
, ret
);
2397 /* Sprouting would change fsid of the mounted root,
2398 * so rename the fsid on the sysfs
2400 snprintf(fsid_buf
, BTRFS_UUID_UNPARSED_SIZE
, "%pU",
2401 root
->fs_info
->fsid
);
2402 if (kobject_rename(&root
->fs_info
->fs_devices
->fsid_kobj
,
2404 btrfs_warn(root
->fs_info
,
2405 "sysfs: failed to create fsid for sprout");
2408 root
->fs_info
->num_tolerated_disk_barrier_failures
=
2409 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
2410 ret
= btrfs_commit_transaction(trans
, root
);
2413 mutex_unlock(&uuid_mutex
);
2414 up_write(&sb
->s_umount
);
2416 if (ret
) /* transaction commit */
2419 ret
= btrfs_relocate_sys_chunks(root
);
2421 btrfs_std_error(root
->fs_info
, ret
,
2422 "Failed to relocate sys chunks after "
2423 "device initialization. This can be fixed "
2424 "using the \"btrfs balance\" command.");
2425 trans
= btrfs_attach_transaction(root
);
2426 if (IS_ERR(trans
)) {
2427 if (PTR_ERR(trans
) == -ENOENT
)
2429 return PTR_ERR(trans
);
2431 ret
= btrfs_commit_transaction(trans
, root
);
2434 /* Update ctime/mtime for libblkid */
2435 update_dev_time(device_path
);
2439 btrfs_end_transaction(trans
, root
);
2440 rcu_string_free(device
->name
);
2441 btrfs_sysfs_rm_device_link(root
->fs_info
->fs_devices
, device
);
2444 blkdev_put(bdev
, FMODE_EXCL
);
2446 mutex_unlock(&uuid_mutex
);
2447 up_write(&sb
->s_umount
);
2452 int btrfs_init_dev_replace_tgtdev(struct btrfs_root
*root
, char *device_path
,
2453 struct btrfs_device
*srcdev
,
2454 struct btrfs_device
**device_out
)
2456 struct request_queue
*q
;
2457 struct btrfs_device
*device
;
2458 struct block_device
*bdev
;
2459 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2460 struct list_head
*devices
;
2461 struct rcu_string
*name
;
2462 u64 devid
= BTRFS_DEV_REPLACE_DEVID
;
2466 if (fs_info
->fs_devices
->seeding
) {
2467 btrfs_err(fs_info
, "the filesystem is a seed filesystem!");
2471 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2472 fs_info
->bdev_holder
);
2474 btrfs_err(fs_info
, "target device %s is invalid!", device_path
);
2475 return PTR_ERR(bdev
);
2478 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2480 devices
= &fs_info
->fs_devices
->devices
;
2481 list_for_each_entry(device
, devices
, dev_list
) {
2482 if (device
->bdev
== bdev
) {
2483 btrfs_err(fs_info
, "target device is in the filesystem!");
2490 if (i_size_read(bdev
->bd_inode
) <
2491 btrfs_device_get_total_bytes(srcdev
)) {
2492 btrfs_err(fs_info
, "target device is smaller than source device!");
2498 device
= btrfs_alloc_device(NULL
, &devid
, NULL
);
2499 if (IS_ERR(device
)) {
2500 ret
= PTR_ERR(device
);
2504 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2510 rcu_assign_pointer(device
->name
, name
);
2512 q
= bdev_get_queue(bdev
);
2513 if (blk_queue_discard(q
))
2514 device
->can_discard
= 1;
2515 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2516 device
->writeable
= 1;
2517 device
->generation
= 0;
2518 device
->io_width
= root
->sectorsize
;
2519 device
->io_align
= root
->sectorsize
;
2520 device
->sector_size
= root
->sectorsize
;
2521 device
->total_bytes
= btrfs_device_get_total_bytes(srcdev
);
2522 device
->disk_total_bytes
= btrfs_device_get_disk_total_bytes(srcdev
);
2523 device
->bytes_used
= btrfs_device_get_bytes_used(srcdev
);
2524 ASSERT(list_empty(&srcdev
->resized_list
));
2525 device
->commit_total_bytes
= srcdev
->commit_total_bytes
;
2526 device
->commit_bytes_used
= device
->bytes_used
;
2527 device
->dev_root
= fs_info
->dev_root
;
2528 device
->bdev
= bdev
;
2529 device
->in_fs_metadata
= 1;
2530 device
->is_tgtdev_for_dev_replace
= 1;
2531 device
->mode
= FMODE_EXCL
;
2532 device
->dev_stats_valid
= 1;
2533 set_blocksize(device
->bdev
, 4096);
2534 device
->fs_devices
= fs_info
->fs_devices
;
2535 list_add(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2536 fs_info
->fs_devices
->num_devices
++;
2537 fs_info
->fs_devices
->open_devices
++;
2538 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2540 *device_out
= device
;
2544 blkdev_put(bdev
, FMODE_EXCL
);
2548 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info
*fs_info
,
2549 struct btrfs_device
*tgtdev
)
2551 WARN_ON(fs_info
->fs_devices
->rw_devices
== 0);
2552 tgtdev
->io_width
= fs_info
->dev_root
->sectorsize
;
2553 tgtdev
->io_align
= fs_info
->dev_root
->sectorsize
;
2554 tgtdev
->sector_size
= fs_info
->dev_root
->sectorsize
;
2555 tgtdev
->dev_root
= fs_info
->dev_root
;
2556 tgtdev
->in_fs_metadata
= 1;
2559 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2560 struct btrfs_device
*device
)
2563 struct btrfs_path
*path
;
2564 struct btrfs_root
*root
;
2565 struct btrfs_dev_item
*dev_item
;
2566 struct extent_buffer
*leaf
;
2567 struct btrfs_key key
;
2569 root
= device
->dev_root
->fs_info
->chunk_root
;
2571 path
= btrfs_alloc_path();
2575 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2576 key
.type
= BTRFS_DEV_ITEM_KEY
;
2577 key
.offset
= device
->devid
;
2579 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2588 leaf
= path
->nodes
[0];
2589 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2591 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2592 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2593 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2594 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2595 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2596 btrfs_set_device_total_bytes(leaf
, dev_item
,
2597 btrfs_device_get_disk_total_bytes(device
));
2598 btrfs_set_device_bytes_used(leaf
, dev_item
,
2599 btrfs_device_get_bytes_used(device
));
2600 btrfs_mark_buffer_dirty(leaf
);
2603 btrfs_free_path(path
);
2607 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2608 struct btrfs_device
*device
, u64 new_size
)
2610 struct btrfs_super_block
*super_copy
=
2611 device
->dev_root
->fs_info
->super_copy
;
2612 struct btrfs_fs_devices
*fs_devices
;
2616 if (!device
->writeable
)
2619 lock_chunks(device
->dev_root
);
2620 old_total
= btrfs_super_total_bytes(super_copy
);
2621 diff
= new_size
- device
->total_bytes
;
2623 if (new_size
<= device
->total_bytes
||
2624 device
->is_tgtdev_for_dev_replace
) {
2625 unlock_chunks(device
->dev_root
);
2629 fs_devices
= device
->dev_root
->fs_info
->fs_devices
;
2631 btrfs_set_super_total_bytes(super_copy
, old_total
+ diff
);
2632 device
->fs_devices
->total_rw_bytes
+= diff
;
2634 btrfs_device_set_total_bytes(device
, new_size
);
2635 btrfs_device_set_disk_total_bytes(device
, new_size
);
2636 btrfs_clear_space_info_full(device
->dev_root
->fs_info
);
2637 if (list_empty(&device
->resized_list
))
2638 list_add_tail(&device
->resized_list
,
2639 &fs_devices
->resized_devices
);
2640 unlock_chunks(device
->dev_root
);
2642 return btrfs_update_device(trans
, device
);
2645 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
2646 struct btrfs_root
*root
, u64 chunk_objectid
,
2650 struct btrfs_path
*path
;
2651 struct btrfs_key key
;
2653 root
= root
->fs_info
->chunk_root
;
2654 path
= btrfs_alloc_path();
2658 key
.objectid
= chunk_objectid
;
2659 key
.offset
= chunk_offset
;
2660 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2662 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2665 else if (ret
> 0) { /* Logic error or corruption */
2666 btrfs_std_error(root
->fs_info
, -ENOENT
,
2667 "Failed lookup while freeing chunk.");
2672 ret
= btrfs_del_item(trans
, root
, path
);
2674 btrfs_std_error(root
->fs_info
, ret
,
2675 "Failed to delete chunk item.");
2677 btrfs_free_path(path
);
2681 static int btrfs_del_sys_chunk(struct btrfs_root
*root
, u64 chunk_objectid
, u64
2684 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
2685 struct btrfs_disk_key
*disk_key
;
2686 struct btrfs_chunk
*chunk
;
2693 struct btrfs_key key
;
2696 array_size
= btrfs_super_sys_array_size(super_copy
);
2698 ptr
= super_copy
->sys_chunk_array
;
2701 while (cur
< array_size
) {
2702 disk_key
= (struct btrfs_disk_key
*)ptr
;
2703 btrfs_disk_key_to_cpu(&key
, disk_key
);
2705 len
= sizeof(*disk_key
);
2707 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2708 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2709 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2710 len
+= btrfs_chunk_item_size(num_stripes
);
2715 if (key
.objectid
== chunk_objectid
&&
2716 key
.offset
== chunk_offset
) {
2717 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2719 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2725 unlock_chunks(root
);
2729 int btrfs_remove_chunk(struct btrfs_trans_handle
*trans
,
2730 struct btrfs_root
*root
, u64 chunk_offset
)
2732 struct extent_map_tree
*em_tree
;
2733 struct extent_map
*em
;
2734 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2735 struct map_lookup
*map
;
2736 u64 dev_extent_len
= 0;
2737 u64 chunk_objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2741 root
= root
->fs_info
->chunk_root
;
2742 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
2744 read_lock(&em_tree
->lock
);
2745 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
2746 read_unlock(&em_tree
->lock
);
2748 if (!em
|| em
->start
> chunk_offset
||
2749 em
->start
+ em
->len
< chunk_offset
) {
2751 * This is a logic error, but we don't want to just rely on the
2752 * user having built with ASSERT enabled, so if ASSERT doens't
2753 * do anything we still error out.
2757 free_extent_map(em
);
2760 map
= (struct map_lookup
*)em
->bdev
;
2761 lock_chunks(root
->fs_info
->chunk_root
);
2762 check_system_chunk(trans
, extent_root
, map
->type
);
2763 unlock_chunks(root
->fs_info
->chunk_root
);
2765 for (i
= 0; i
< map
->num_stripes
; i
++) {
2766 struct btrfs_device
*device
= map
->stripes
[i
].dev
;
2767 ret
= btrfs_free_dev_extent(trans
, device
,
2768 map
->stripes
[i
].physical
,
2771 btrfs_abort_transaction(trans
, root
, ret
);
2775 if (device
->bytes_used
> 0) {
2777 btrfs_device_set_bytes_used(device
,
2778 device
->bytes_used
- dev_extent_len
);
2779 spin_lock(&root
->fs_info
->free_chunk_lock
);
2780 root
->fs_info
->free_chunk_space
+= dev_extent_len
;
2781 spin_unlock(&root
->fs_info
->free_chunk_lock
);
2782 btrfs_clear_space_info_full(root
->fs_info
);
2783 unlock_chunks(root
);
2786 if (map
->stripes
[i
].dev
) {
2787 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2789 btrfs_abort_transaction(trans
, root
, ret
);
2794 ret
= btrfs_free_chunk(trans
, root
, chunk_objectid
, chunk_offset
);
2796 btrfs_abort_transaction(trans
, root
, ret
);
2800 trace_btrfs_chunk_free(root
, map
, chunk_offset
, em
->len
);
2802 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2803 ret
= btrfs_del_sys_chunk(root
, chunk_objectid
, chunk_offset
);
2805 btrfs_abort_transaction(trans
, root
, ret
);
2810 ret
= btrfs_remove_block_group(trans
, extent_root
, chunk_offset
, em
);
2812 btrfs_abort_transaction(trans
, extent_root
, ret
);
2818 free_extent_map(em
);
2822 static int btrfs_relocate_chunk(struct btrfs_root
*root
, u64 chunk_offset
)
2824 struct btrfs_root
*extent_root
;
2825 struct btrfs_trans_handle
*trans
;
2828 root
= root
->fs_info
->chunk_root
;
2829 extent_root
= root
->fs_info
->extent_root
;
2832 * Prevent races with automatic removal of unused block groups.
2833 * After we relocate and before we remove the chunk with offset
2834 * chunk_offset, automatic removal of the block group can kick in,
2835 * resulting in a failure when calling btrfs_remove_chunk() below.
2837 * Make sure to acquire this mutex before doing a tree search (dev
2838 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2839 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2840 * we release the path used to search the chunk/dev tree and before
2841 * the current task acquires this mutex and calls us.
2843 ASSERT(mutex_is_locked(&root
->fs_info
->delete_unused_bgs_mutex
));
2845 ret
= btrfs_can_relocate(extent_root
, chunk_offset
);
2849 /* step one, relocate all the extents inside this chunk */
2850 btrfs_scrub_pause(root
);
2851 ret
= btrfs_relocate_block_group(extent_root
, chunk_offset
);
2852 btrfs_scrub_continue(root
);
2856 trans
= btrfs_start_trans_remove_block_group(root
->fs_info
,
2858 if (IS_ERR(trans
)) {
2859 ret
= PTR_ERR(trans
);
2860 btrfs_std_error(root
->fs_info
, ret
, NULL
);
2865 * step two, delete the device extents and the
2866 * chunk tree entries
2868 ret
= btrfs_remove_chunk(trans
, root
, chunk_offset
);
2869 btrfs_end_transaction(trans
, root
);
2873 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
)
2875 struct btrfs_root
*chunk_root
= root
->fs_info
->chunk_root
;
2876 struct btrfs_path
*path
;
2877 struct extent_buffer
*leaf
;
2878 struct btrfs_chunk
*chunk
;
2879 struct btrfs_key key
;
2880 struct btrfs_key found_key
;
2882 bool retried
= false;
2886 path
= btrfs_alloc_path();
2891 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2892 key
.offset
= (u64
)-1;
2893 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2896 mutex_lock(&root
->fs_info
->delete_unused_bgs_mutex
);
2897 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2899 mutex_unlock(&root
->fs_info
->delete_unused_bgs_mutex
);
2902 BUG_ON(ret
== 0); /* Corruption */
2904 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
2907 mutex_unlock(&root
->fs_info
->delete_unused_bgs_mutex
);
2913 leaf
= path
->nodes
[0];
2914 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2916 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
2917 struct btrfs_chunk
);
2918 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2919 btrfs_release_path(path
);
2921 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2922 ret
= btrfs_relocate_chunk(chunk_root
,
2929 mutex_unlock(&root
->fs_info
->delete_unused_bgs_mutex
);
2931 if (found_key
.offset
== 0)
2933 key
.offset
= found_key
.offset
- 1;
2936 if (failed
&& !retried
) {
2940 } else if (WARN_ON(failed
&& retried
)) {
2944 btrfs_free_path(path
);
2948 static int insert_balance_item(struct btrfs_root
*root
,
2949 struct btrfs_balance_control
*bctl
)
2951 struct btrfs_trans_handle
*trans
;
2952 struct btrfs_balance_item
*item
;
2953 struct btrfs_disk_balance_args disk_bargs
;
2954 struct btrfs_path
*path
;
2955 struct extent_buffer
*leaf
;
2956 struct btrfs_key key
;
2959 path
= btrfs_alloc_path();
2963 trans
= btrfs_start_transaction(root
, 0);
2964 if (IS_ERR(trans
)) {
2965 btrfs_free_path(path
);
2966 return PTR_ERR(trans
);
2969 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2970 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2973 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2978 leaf
= path
->nodes
[0];
2979 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
2981 memset_extent_buffer(leaf
, 0, (unsigned long)item
, sizeof(*item
));
2983 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
2984 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
2985 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
2986 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
2987 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
2988 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
2990 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
2992 btrfs_mark_buffer_dirty(leaf
);
2994 btrfs_free_path(path
);
2995 err
= btrfs_commit_transaction(trans
, root
);
3001 static int del_balance_item(struct btrfs_root
*root
)
3003 struct btrfs_trans_handle
*trans
;
3004 struct btrfs_path
*path
;
3005 struct btrfs_key key
;
3008 path
= btrfs_alloc_path();
3012 trans
= btrfs_start_transaction(root
, 0);
3013 if (IS_ERR(trans
)) {
3014 btrfs_free_path(path
);
3015 return PTR_ERR(trans
);
3018 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3019 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
3022 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3030 ret
= btrfs_del_item(trans
, root
, path
);
3032 btrfs_free_path(path
);
3033 err
= btrfs_commit_transaction(trans
, root
);
3040 * This is a heuristic used to reduce the number of chunks balanced on
3041 * resume after balance was interrupted.
3043 static void update_balance_args(struct btrfs_balance_control
*bctl
)
3046 * Turn on soft mode for chunk types that were being converted.
3048 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3049 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3050 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3051 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3052 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3053 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3056 * Turn on usage filter if is not already used. The idea is
3057 * that chunks that we have already balanced should be
3058 * reasonably full. Don't do it for chunks that are being
3059 * converted - that will keep us from relocating unconverted
3060 * (albeit full) chunks.
3062 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3063 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3064 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3065 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3066 bctl
->data
.usage
= 90;
3068 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3069 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3070 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3071 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3072 bctl
->sys
.usage
= 90;
3074 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3075 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3076 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3077 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3078 bctl
->meta
.usage
= 90;
3083 * Should be called with both balance and volume mutexes held to
3084 * serialize other volume operations (add_dev/rm_dev/resize) with
3085 * restriper. Same goes for unset_balance_control.
3087 static void set_balance_control(struct btrfs_balance_control
*bctl
)
3089 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3091 BUG_ON(fs_info
->balance_ctl
);
3093 spin_lock(&fs_info
->balance_lock
);
3094 fs_info
->balance_ctl
= bctl
;
3095 spin_unlock(&fs_info
->balance_lock
);
3098 static void unset_balance_control(struct btrfs_fs_info
*fs_info
)
3100 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3102 BUG_ON(!fs_info
->balance_ctl
);
3104 spin_lock(&fs_info
->balance_lock
);
3105 fs_info
->balance_ctl
= NULL
;
3106 spin_unlock(&fs_info
->balance_lock
);
3112 * Balance filters. Return 1 if chunk should be filtered out
3113 * (should not be balanced).
3115 static int chunk_profiles_filter(u64 chunk_type
,
3116 struct btrfs_balance_args
*bargs
)
3118 chunk_type
= chunk_to_extended(chunk_type
) &
3119 BTRFS_EXTENDED_PROFILE_MASK
;
3121 if (bargs
->profiles
& chunk_type
)
3127 static int chunk_usage_range_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
3128 struct btrfs_balance_args
*bargs
)
3130 struct btrfs_block_group_cache
*cache
;
3132 u64 user_thresh_min
;
3133 u64 user_thresh_max
;
3136 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3137 chunk_used
= btrfs_block_group_used(&cache
->item
);
3139 if (bargs
->usage_min
== 0)
3140 user_thresh_min
= 0;
3142 user_thresh_min
= div_factor_fine(cache
->key
.offset
,
3145 if (bargs
->usage_max
== 0)
3146 user_thresh_max
= 1;
3147 else if (bargs
->usage_max
> 100)
3148 user_thresh_max
= cache
->key
.offset
;
3150 user_thresh_max
= div_factor_fine(cache
->key
.offset
,
3153 if (user_thresh_min
<= chunk_used
&& chunk_used
< user_thresh_max
)
3156 btrfs_put_block_group(cache
);
3160 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
,
3161 u64 chunk_offset
, struct btrfs_balance_args
*bargs
)
3163 struct btrfs_block_group_cache
*cache
;
3164 u64 chunk_used
, user_thresh
;
3167 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3168 chunk_used
= btrfs_block_group_used(&cache
->item
);
3170 if (bargs
->usage_min
== 0)
3172 else if (bargs
->usage
> 100)
3173 user_thresh
= cache
->key
.offset
;
3175 user_thresh
= div_factor_fine(cache
->key
.offset
,
3178 if (chunk_used
< user_thresh
)
3181 btrfs_put_block_group(cache
);
3185 static int chunk_devid_filter(struct extent_buffer
*leaf
,
3186 struct btrfs_chunk
*chunk
,
3187 struct btrfs_balance_args
*bargs
)
3189 struct btrfs_stripe
*stripe
;
3190 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3193 for (i
= 0; i
< num_stripes
; i
++) {
3194 stripe
= btrfs_stripe_nr(chunk
, i
);
3195 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
3202 /* [pstart, pend) */
3203 static int chunk_drange_filter(struct extent_buffer
*leaf
,
3204 struct btrfs_chunk
*chunk
,
3206 struct btrfs_balance_args
*bargs
)
3208 struct btrfs_stripe
*stripe
;
3209 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3215 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
3218 if (btrfs_chunk_type(leaf
, chunk
) & (BTRFS_BLOCK_GROUP_DUP
|
3219 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)) {
3220 factor
= num_stripes
/ 2;
3221 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID5
) {
3222 factor
= num_stripes
- 1;
3223 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID6
) {
3224 factor
= num_stripes
- 2;
3226 factor
= num_stripes
;
3229 for (i
= 0; i
< num_stripes
; i
++) {
3230 stripe
= btrfs_stripe_nr(chunk
, i
);
3231 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
3234 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
3235 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
3236 stripe_length
= div_u64(stripe_length
, factor
);
3238 if (stripe_offset
< bargs
->pend
&&
3239 stripe_offset
+ stripe_length
> bargs
->pstart
)
3246 /* [vstart, vend) */
3247 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
3248 struct btrfs_chunk
*chunk
,
3250 struct btrfs_balance_args
*bargs
)
3252 if (chunk_offset
< bargs
->vend
&&
3253 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
3254 /* at least part of the chunk is inside this vrange */
3260 static int chunk_stripes_range_filter(struct extent_buffer
*leaf
,
3261 struct btrfs_chunk
*chunk
,
3262 struct btrfs_balance_args
*bargs
)
3264 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3266 if (bargs
->stripes_min
<= num_stripes
3267 && num_stripes
<= bargs
->stripes_max
)
3273 static int chunk_soft_convert_filter(u64 chunk_type
,
3274 struct btrfs_balance_args
*bargs
)
3276 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
3279 chunk_type
= chunk_to_extended(chunk_type
) &
3280 BTRFS_EXTENDED_PROFILE_MASK
;
3282 if (bargs
->target
== chunk_type
)
3288 static int should_balance_chunk(struct btrfs_root
*root
,
3289 struct extent_buffer
*leaf
,
3290 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
3292 struct btrfs_balance_control
*bctl
= root
->fs_info
->balance_ctl
;
3293 struct btrfs_balance_args
*bargs
= NULL
;
3294 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3297 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
3298 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
3302 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3303 bargs
= &bctl
->data
;
3304 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3306 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3307 bargs
= &bctl
->meta
;
3309 /* profiles filter */
3310 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
3311 chunk_profiles_filter(chunk_type
, bargs
)) {
3316 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3317 chunk_usage_filter(bctl
->fs_info
, chunk_offset
, bargs
)) {
3319 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3320 chunk_usage_range_filter(bctl
->fs_info
, chunk_offset
, bargs
)) {
3325 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
3326 chunk_devid_filter(leaf
, chunk
, bargs
)) {
3330 /* drange filter, makes sense only with devid filter */
3331 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
3332 chunk_drange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3337 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
3338 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3342 /* stripes filter */
3343 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_STRIPES_RANGE
) &&
3344 chunk_stripes_range_filter(leaf
, chunk
, bargs
)) {
3348 /* soft profile changing mode */
3349 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
3350 chunk_soft_convert_filter(chunk_type
, bargs
)) {
3355 * limited by count, must be the last filter
3357 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT
)) {
3358 if (bargs
->limit
== 0)
3362 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT_RANGE
)) {
3364 * Same logic as the 'limit' filter; the minimum cannot be
3365 * determined here because we do not have the global informatoin
3366 * about the count of all chunks that satisfy the filters.
3368 if (bargs
->limit_max
== 0)
3377 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
3379 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3380 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
3381 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
3382 struct list_head
*devices
;
3383 struct btrfs_device
*device
;
3387 struct btrfs_chunk
*chunk
;
3388 struct btrfs_path
*path
;
3389 struct btrfs_key key
;
3390 struct btrfs_key found_key
;
3391 struct btrfs_trans_handle
*trans
;
3392 struct extent_buffer
*leaf
;
3395 int enospc_errors
= 0;
3396 bool counting
= true;
3397 /* The single value limit and min/max limits use the same bytes in the */
3398 u64 limit_data
= bctl
->data
.limit
;
3399 u64 limit_meta
= bctl
->meta
.limit
;
3400 u64 limit_sys
= bctl
->sys
.limit
;
3404 int chunk_reserved
= 0;
3406 /* step one make some room on all the devices */
3407 devices
= &fs_info
->fs_devices
->devices
;
3408 list_for_each_entry(device
, devices
, dev_list
) {
3409 old_size
= btrfs_device_get_total_bytes(device
);
3410 size_to_free
= div_factor(old_size
, 1);
3411 size_to_free
= min(size_to_free
, (u64
)1 * 1024 * 1024);
3412 if (!device
->writeable
||
3413 btrfs_device_get_total_bytes(device
) -
3414 btrfs_device_get_bytes_used(device
) > size_to_free
||
3415 device
->is_tgtdev_for_dev_replace
)
3418 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
3423 trans
= btrfs_start_transaction(dev_root
, 0);
3424 BUG_ON(IS_ERR(trans
));
3426 ret
= btrfs_grow_device(trans
, device
, old_size
);
3429 btrfs_end_transaction(trans
, dev_root
);
3432 /* step two, relocate all the chunks */
3433 path
= btrfs_alloc_path();
3439 /* zero out stat counters */
3440 spin_lock(&fs_info
->balance_lock
);
3441 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
3442 spin_unlock(&fs_info
->balance_lock
);
3446 * The single value limit and min/max limits use the same bytes
3449 bctl
->data
.limit
= limit_data
;
3450 bctl
->meta
.limit
= limit_meta
;
3451 bctl
->sys
.limit
= limit_sys
;
3453 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3454 key
.offset
= (u64
)-1;
3455 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3458 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
3459 atomic_read(&fs_info
->balance_cancel_req
)) {
3464 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
3465 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
3467 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3472 * this shouldn't happen, it means the last relocate
3476 BUG(); /* FIXME break ? */
3478 ret
= btrfs_previous_item(chunk_root
, path
, 0,
3479 BTRFS_CHUNK_ITEM_KEY
);
3481 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3486 leaf
= path
->nodes
[0];
3487 slot
= path
->slots
[0];
3488 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3490 if (found_key
.objectid
!= key
.objectid
) {
3491 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3495 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
3496 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3499 spin_lock(&fs_info
->balance_lock
);
3500 bctl
->stat
.considered
++;
3501 spin_unlock(&fs_info
->balance_lock
);
3504 ret
= should_balance_chunk(chunk_root
, leaf
, chunk
,
3507 btrfs_release_path(path
);
3509 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3514 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3515 spin_lock(&fs_info
->balance_lock
);
3516 bctl
->stat
.expected
++;
3517 spin_unlock(&fs_info
->balance_lock
);
3519 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3521 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3523 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3530 * Apply limit_min filter, no need to check if the LIMITS
3531 * filter is used, limit_min is 0 by default
3533 if (((chunk_type
& BTRFS_BLOCK_GROUP_DATA
) &&
3534 count_data
< bctl
->data
.limit_min
)
3535 || ((chunk_type
& BTRFS_BLOCK_GROUP_METADATA
) &&
3536 count_meta
< bctl
->meta
.limit_min
)
3537 || ((chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) &&
3538 count_sys
< bctl
->sys
.limit_min
)) {
3539 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3543 if ((chunk_type
& BTRFS_BLOCK_GROUP_DATA
) && !chunk_reserved
) {
3544 trans
= btrfs_start_transaction(chunk_root
, 0);
3545 if (IS_ERR(trans
)) {
3546 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3547 ret
= PTR_ERR(trans
);
3551 ret
= btrfs_force_chunk_alloc(trans
, chunk_root
,
3552 BTRFS_BLOCK_GROUP_DATA
);
3553 btrfs_end_transaction(trans
, chunk_root
);
3555 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3561 ret
= btrfs_relocate_chunk(chunk_root
,
3563 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3564 if (ret
&& ret
!= -ENOSPC
)
3566 if (ret
== -ENOSPC
) {
3569 spin_lock(&fs_info
->balance_lock
);
3570 bctl
->stat
.completed
++;
3571 spin_unlock(&fs_info
->balance_lock
);
3574 if (found_key
.offset
== 0)
3576 key
.offset
= found_key
.offset
- 1;
3580 btrfs_release_path(path
);
3585 btrfs_free_path(path
);
3586 if (enospc_errors
) {
3587 btrfs_info(fs_info
, "%d enospc errors during balance",
3597 * alloc_profile_is_valid - see if a given profile is valid and reduced
3598 * @flags: profile to validate
3599 * @extended: if true @flags is treated as an extended profile
3601 static int alloc_profile_is_valid(u64 flags
, int extended
)
3603 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
3604 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
3606 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
3608 /* 1) check that all other bits are zeroed */
3612 /* 2) see if profile is reduced */
3614 return !extended
; /* "0" is valid for usual profiles */
3616 /* true if exactly one bit set */
3617 return (flags
& (flags
- 1)) == 0;
3620 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
3622 /* cancel requested || normal exit path */
3623 return atomic_read(&fs_info
->balance_cancel_req
) ||
3624 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
3625 atomic_read(&fs_info
->balance_cancel_req
) == 0);
3628 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
3632 unset_balance_control(fs_info
);
3633 ret
= del_balance_item(fs_info
->tree_root
);
3635 btrfs_std_error(fs_info
, ret
, NULL
);
3637 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3640 /* Non-zero return value signifies invalidity */
3641 static inline int validate_convert_profile(struct btrfs_balance_args
*bctl_arg
,
3644 return ((bctl_arg
->flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3645 (!alloc_profile_is_valid(bctl_arg
->target
, 1) ||
3646 (bctl_arg
->target
& ~allowed
)));
3650 * Should be called with both balance and volume mutexes held
3652 int btrfs_balance(struct btrfs_balance_control
*bctl
,
3653 struct btrfs_ioctl_balance_args
*bargs
)
3655 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3662 if (btrfs_fs_closing(fs_info
) ||
3663 atomic_read(&fs_info
->balance_pause_req
) ||
3664 atomic_read(&fs_info
->balance_cancel_req
)) {
3669 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
3670 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
3674 * In case of mixed groups both data and meta should be picked,
3675 * and identical options should be given for both of them.
3677 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
3678 if (mixed
&& (bctl
->flags
& allowed
)) {
3679 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
3680 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
3681 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
3682 btrfs_err(fs_info
, "with mixed groups data and "
3683 "metadata balance options must be the same");
3689 num_devices
= fs_info
->fs_devices
->num_devices
;
3690 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
3691 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
3692 BUG_ON(num_devices
< 1);
3695 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
3696 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
3697 if (num_devices
== 1)
3698 allowed
|= BTRFS_BLOCK_GROUP_DUP
;
3699 else if (num_devices
> 1)
3700 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
3701 if (num_devices
> 2)
3702 allowed
|= BTRFS_BLOCK_GROUP_RAID5
;
3703 if (num_devices
> 3)
3704 allowed
|= (BTRFS_BLOCK_GROUP_RAID10
|
3705 BTRFS_BLOCK_GROUP_RAID6
);
3706 if (validate_convert_profile(&bctl
->data
, allowed
)) {
3707 btrfs_err(fs_info
, "unable to start balance with target "
3708 "data profile %llu",
3713 if (validate_convert_profile(&bctl
->meta
, allowed
)) {
3715 "unable to start balance with target metadata profile %llu",
3720 if (validate_convert_profile(&bctl
->sys
, allowed
)) {
3722 "unable to start balance with target system profile %llu",
3728 /* allow dup'ed data chunks only in mixed mode */
3729 if (!mixed
&& (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3730 (bctl
->data
.target
& BTRFS_BLOCK_GROUP_DUP
)) {
3731 btrfs_err(fs_info
, "dup for data is not allowed");
3736 /* allow to reduce meta or sys integrity only if force set */
3737 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3738 BTRFS_BLOCK_GROUP_RAID10
|
3739 BTRFS_BLOCK_GROUP_RAID5
|
3740 BTRFS_BLOCK_GROUP_RAID6
;
3742 seq
= read_seqbegin(&fs_info
->profiles_lock
);
3744 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3745 (fs_info
->avail_system_alloc_bits
& allowed
) &&
3746 !(bctl
->sys
.target
& allowed
)) ||
3747 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3748 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
3749 !(bctl
->meta
.target
& allowed
))) {
3750 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
3751 btrfs_info(fs_info
, "force reducing metadata integrity");
3753 btrfs_err(fs_info
, "balance will reduce metadata "
3754 "integrity, use force if you want this");
3759 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
3761 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3762 fs_info
->num_tolerated_disk_barrier_failures
= min(
3763 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
),
3764 btrfs_get_num_tolerated_disk_barrier_failures(
3768 ret
= insert_balance_item(fs_info
->tree_root
, bctl
);
3769 if (ret
&& ret
!= -EEXIST
)
3772 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
3773 BUG_ON(ret
== -EEXIST
);
3774 set_balance_control(bctl
);
3776 BUG_ON(ret
!= -EEXIST
);
3777 spin_lock(&fs_info
->balance_lock
);
3778 update_balance_args(bctl
);
3779 spin_unlock(&fs_info
->balance_lock
);
3782 atomic_inc(&fs_info
->balance_running
);
3783 mutex_unlock(&fs_info
->balance_mutex
);
3785 ret
= __btrfs_balance(fs_info
);
3787 mutex_lock(&fs_info
->balance_mutex
);
3788 atomic_dec(&fs_info
->balance_running
);
3790 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3791 fs_info
->num_tolerated_disk_barrier_failures
=
3792 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3796 memset(bargs
, 0, sizeof(*bargs
));
3797 update_ioctl_balance_args(fs_info
, 0, bargs
);
3800 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
3801 balance_need_close(fs_info
)) {
3802 __cancel_balance(fs_info
);
3805 wake_up(&fs_info
->balance_wait_q
);
3809 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
3810 __cancel_balance(fs_info
);
3813 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3818 static int balance_kthread(void *data
)
3820 struct btrfs_fs_info
*fs_info
= data
;
3823 mutex_lock(&fs_info
->volume_mutex
);
3824 mutex_lock(&fs_info
->balance_mutex
);
3826 if (fs_info
->balance_ctl
) {
3827 btrfs_info(fs_info
, "continuing balance");
3828 ret
= btrfs_balance(fs_info
->balance_ctl
, NULL
);
3831 mutex_unlock(&fs_info
->balance_mutex
);
3832 mutex_unlock(&fs_info
->volume_mutex
);
3837 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
3839 struct task_struct
*tsk
;
3841 spin_lock(&fs_info
->balance_lock
);
3842 if (!fs_info
->balance_ctl
) {
3843 spin_unlock(&fs_info
->balance_lock
);
3846 spin_unlock(&fs_info
->balance_lock
);
3848 if (btrfs_test_opt(fs_info
->tree_root
, SKIP_BALANCE
)) {
3849 btrfs_info(fs_info
, "force skipping balance");
3854 * A ro->rw remount sequence should continue with the paused balance
3855 * regardless of who pauses it, system or the user as of now, so set
3858 spin_lock(&fs_info
->balance_lock
);
3859 fs_info
->balance_ctl
->flags
|= BTRFS_BALANCE_RESUME
;
3860 spin_unlock(&fs_info
->balance_lock
);
3862 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
3863 return PTR_ERR_OR_ZERO(tsk
);
3866 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
3868 struct btrfs_balance_control
*bctl
;
3869 struct btrfs_balance_item
*item
;
3870 struct btrfs_disk_balance_args disk_bargs
;
3871 struct btrfs_path
*path
;
3872 struct extent_buffer
*leaf
;
3873 struct btrfs_key key
;
3876 path
= btrfs_alloc_path();
3880 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3881 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
3884 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
3887 if (ret
> 0) { /* ret = -ENOENT; */
3892 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
3898 leaf
= path
->nodes
[0];
3899 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3901 bctl
->fs_info
= fs_info
;
3902 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
3903 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
3905 btrfs_balance_data(leaf
, item
, &disk_bargs
);
3906 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
3907 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
3908 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
3909 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
3910 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
3912 WARN_ON(atomic_xchg(&fs_info
->mutually_exclusive_operation_running
, 1));
3914 mutex_lock(&fs_info
->volume_mutex
);
3915 mutex_lock(&fs_info
->balance_mutex
);
3917 set_balance_control(bctl
);
3919 mutex_unlock(&fs_info
->balance_mutex
);
3920 mutex_unlock(&fs_info
->volume_mutex
);
3922 btrfs_free_path(path
);
3926 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
3930 mutex_lock(&fs_info
->balance_mutex
);
3931 if (!fs_info
->balance_ctl
) {
3932 mutex_unlock(&fs_info
->balance_mutex
);
3936 if (atomic_read(&fs_info
->balance_running
)) {
3937 atomic_inc(&fs_info
->balance_pause_req
);
3938 mutex_unlock(&fs_info
->balance_mutex
);
3940 wait_event(fs_info
->balance_wait_q
,
3941 atomic_read(&fs_info
->balance_running
) == 0);
3943 mutex_lock(&fs_info
->balance_mutex
);
3944 /* we are good with balance_ctl ripped off from under us */
3945 BUG_ON(atomic_read(&fs_info
->balance_running
));
3946 atomic_dec(&fs_info
->balance_pause_req
);
3951 mutex_unlock(&fs_info
->balance_mutex
);
3955 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
3957 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
3960 mutex_lock(&fs_info
->balance_mutex
);
3961 if (!fs_info
->balance_ctl
) {
3962 mutex_unlock(&fs_info
->balance_mutex
);
3966 atomic_inc(&fs_info
->balance_cancel_req
);
3968 * if we are running just wait and return, balance item is
3969 * deleted in btrfs_balance in this case
3971 if (atomic_read(&fs_info
->balance_running
)) {
3972 mutex_unlock(&fs_info
->balance_mutex
);
3973 wait_event(fs_info
->balance_wait_q
,
3974 atomic_read(&fs_info
->balance_running
) == 0);
3975 mutex_lock(&fs_info
->balance_mutex
);
3977 /* __cancel_balance needs volume_mutex */
3978 mutex_unlock(&fs_info
->balance_mutex
);
3979 mutex_lock(&fs_info
->volume_mutex
);
3980 mutex_lock(&fs_info
->balance_mutex
);
3982 if (fs_info
->balance_ctl
)
3983 __cancel_balance(fs_info
);
3985 mutex_unlock(&fs_info
->volume_mutex
);
3988 BUG_ON(fs_info
->balance_ctl
|| atomic_read(&fs_info
->balance_running
));
3989 atomic_dec(&fs_info
->balance_cancel_req
);
3990 mutex_unlock(&fs_info
->balance_mutex
);
3994 static int btrfs_uuid_scan_kthread(void *data
)
3996 struct btrfs_fs_info
*fs_info
= data
;
3997 struct btrfs_root
*root
= fs_info
->tree_root
;
3998 struct btrfs_key key
;
3999 struct btrfs_key max_key
;
4000 struct btrfs_path
*path
= NULL
;
4002 struct extent_buffer
*eb
;
4004 struct btrfs_root_item root_item
;
4006 struct btrfs_trans_handle
*trans
= NULL
;
4008 path
= btrfs_alloc_path();
4015 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4018 max_key
.objectid
= (u64
)-1;
4019 max_key
.type
= BTRFS_ROOT_ITEM_KEY
;
4020 max_key
.offset
= (u64
)-1;
4023 ret
= btrfs_search_forward(root
, &key
, path
, 0);
4030 if (key
.type
!= BTRFS_ROOT_ITEM_KEY
||
4031 (key
.objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
4032 key
.objectid
!= BTRFS_FS_TREE_OBJECTID
) ||
4033 key
.objectid
> BTRFS_LAST_FREE_OBJECTID
)
4036 eb
= path
->nodes
[0];
4037 slot
= path
->slots
[0];
4038 item_size
= btrfs_item_size_nr(eb
, slot
);
4039 if (item_size
< sizeof(root_item
))
4042 read_extent_buffer(eb
, &root_item
,
4043 btrfs_item_ptr_offset(eb
, slot
),
4044 (int)sizeof(root_item
));
4045 if (btrfs_root_refs(&root_item
) == 0)
4048 if (!btrfs_is_empty_uuid(root_item
.uuid
) ||
4049 !btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4053 btrfs_release_path(path
);
4055 * 1 - subvol uuid item
4056 * 1 - received_subvol uuid item
4058 trans
= btrfs_start_transaction(fs_info
->uuid_root
, 2);
4059 if (IS_ERR(trans
)) {
4060 ret
= PTR_ERR(trans
);
4068 if (!btrfs_is_empty_uuid(root_item
.uuid
)) {
4069 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
4071 BTRFS_UUID_KEY_SUBVOL
,
4074 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4080 if (!btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4081 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
4082 root_item
.received_uuid
,
4083 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
4086 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4094 ret
= btrfs_end_transaction(trans
, fs_info
->uuid_root
);
4100 btrfs_release_path(path
);
4101 if (key
.offset
< (u64
)-1) {
4103 } else if (key
.type
< BTRFS_ROOT_ITEM_KEY
) {
4105 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4106 } else if (key
.objectid
< (u64
)-1) {
4108 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4117 btrfs_free_path(path
);
4118 if (trans
&& !IS_ERR(trans
))
4119 btrfs_end_transaction(trans
, fs_info
->uuid_root
);
4121 btrfs_warn(fs_info
, "btrfs_uuid_scan_kthread failed %d", ret
);
4123 fs_info
->update_uuid_tree_gen
= 1;
4124 up(&fs_info
->uuid_tree_rescan_sem
);
4129 * Callback for btrfs_uuid_tree_iterate().
4131 * 0 check succeeded, the entry is not outdated.
4132 * < 0 if an error occured.
4133 * > 0 if the check failed, which means the caller shall remove the entry.
4135 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info
*fs_info
,
4136 u8
*uuid
, u8 type
, u64 subid
)
4138 struct btrfs_key key
;
4140 struct btrfs_root
*subvol_root
;
4142 if (type
!= BTRFS_UUID_KEY_SUBVOL
&&
4143 type
!= BTRFS_UUID_KEY_RECEIVED_SUBVOL
)
4146 key
.objectid
= subid
;
4147 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4148 key
.offset
= (u64
)-1;
4149 subvol_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
4150 if (IS_ERR(subvol_root
)) {
4151 ret
= PTR_ERR(subvol_root
);
4158 case BTRFS_UUID_KEY_SUBVOL
:
4159 if (memcmp(uuid
, subvol_root
->root_item
.uuid
, BTRFS_UUID_SIZE
))
4162 case BTRFS_UUID_KEY_RECEIVED_SUBVOL
:
4163 if (memcmp(uuid
, subvol_root
->root_item
.received_uuid
,
4173 static int btrfs_uuid_rescan_kthread(void *data
)
4175 struct btrfs_fs_info
*fs_info
= (struct btrfs_fs_info
*)data
;
4179 * 1st step is to iterate through the existing UUID tree and
4180 * to delete all entries that contain outdated data.
4181 * 2nd step is to add all missing entries to the UUID tree.
4183 ret
= btrfs_uuid_tree_iterate(fs_info
, btrfs_check_uuid_tree_entry
);
4185 btrfs_warn(fs_info
, "iterating uuid_tree failed %d", ret
);
4186 up(&fs_info
->uuid_tree_rescan_sem
);
4189 return btrfs_uuid_scan_kthread(data
);
4192 int btrfs_create_uuid_tree(struct btrfs_fs_info
*fs_info
)
4194 struct btrfs_trans_handle
*trans
;
4195 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
4196 struct btrfs_root
*uuid_root
;
4197 struct task_struct
*task
;
4204 trans
= btrfs_start_transaction(tree_root
, 2);
4206 return PTR_ERR(trans
);
4208 uuid_root
= btrfs_create_tree(trans
, fs_info
,
4209 BTRFS_UUID_TREE_OBJECTID
);
4210 if (IS_ERR(uuid_root
)) {
4211 ret
= PTR_ERR(uuid_root
);
4212 btrfs_abort_transaction(trans
, tree_root
, ret
);
4216 fs_info
->uuid_root
= uuid_root
;
4218 ret
= btrfs_commit_transaction(trans
, tree_root
);
4222 down(&fs_info
->uuid_tree_rescan_sem
);
4223 task
= kthread_run(btrfs_uuid_scan_kthread
, fs_info
, "btrfs-uuid");
4225 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4226 btrfs_warn(fs_info
, "failed to start uuid_scan task");
4227 up(&fs_info
->uuid_tree_rescan_sem
);
4228 return PTR_ERR(task
);
4234 int btrfs_check_uuid_tree(struct btrfs_fs_info
*fs_info
)
4236 struct task_struct
*task
;
4238 down(&fs_info
->uuid_tree_rescan_sem
);
4239 task
= kthread_run(btrfs_uuid_rescan_kthread
, fs_info
, "btrfs-uuid");
4241 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4242 btrfs_warn(fs_info
, "failed to start uuid_rescan task");
4243 up(&fs_info
->uuid_tree_rescan_sem
);
4244 return PTR_ERR(task
);
4251 * shrinking a device means finding all of the device extents past
4252 * the new size, and then following the back refs to the chunks.
4253 * The chunk relocation code actually frees the device extent
4255 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
4257 struct btrfs_trans_handle
*trans
;
4258 struct btrfs_root
*root
= device
->dev_root
;
4259 struct btrfs_dev_extent
*dev_extent
= NULL
;
4260 struct btrfs_path
*path
;
4266 bool retried
= false;
4267 bool checked_pending_chunks
= false;
4268 struct extent_buffer
*l
;
4269 struct btrfs_key key
;
4270 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
4271 u64 old_total
= btrfs_super_total_bytes(super_copy
);
4272 u64 old_size
= btrfs_device_get_total_bytes(device
);
4273 u64 diff
= old_size
- new_size
;
4275 if (device
->is_tgtdev_for_dev_replace
)
4278 path
= btrfs_alloc_path();
4286 btrfs_device_set_total_bytes(device
, new_size
);
4287 if (device
->writeable
) {
4288 device
->fs_devices
->total_rw_bytes
-= diff
;
4289 spin_lock(&root
->fs_info
->free_chunk_lock
);
4290 root
->fs_info
->free_chunk_space
-= diff
;
4291 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4293 unlock_chunks(root
);
4296 key
.objectid
= device
->devid
;
4297 key
.offset
= (u64
)-1;
4298 key
.type
= BTRFS_DEV_EXTENT_KEY
;
4301 mutex_lock(&root
->fs_info
->delete_unused_bgs_mutex
);
4302 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4304 mutex_unlock(&root
->fs_info
->delete_unused_bgs_mutex
);
4308 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
4310 mutex_unlock(&root
->fs_info
->delete_unused_bgs_mutex
);
4315 btrfs_release_path(path
);
4320 slot
= path
->slots
[0];
4321 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
4323 if (key
.objectid
!= device
->devid
) {
4324 mutex_unlock(&root
->fs_info
->delete_unused_bgs_mutex
);
4325 btrfs_release_path(path
);
4329 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
4330 length
= btrfs_dev_extent_length(l
, dev_extent
);
4332 if (key
.offset
+ length
<= new_size
) {
4333 mutex_unlock(&root
->fs_info
->delete_unused_bgs_mutex
);
4334 btrfs_release_path(path
);
4338 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
4339 btrfs_release_path(path
);
4341 ret
= btrfs_relocate_chunk(root
, chunk_offset
);
4342 mutex_unlock(&root
->fs_info
->delete_unused_bgs_mutex
);
4343 if (ret
&& ret
!= -ENOSPC
)
4347 } while (key
.offset
-- > 0);
4349 if (failed
&& !retried
) {
4353 } else if (failed
&& retried
) {
4358 /* Shrinking succeeded, else we would be at "done". */
4359 trans
= btrfs_start_transaction(root
, 0);
4360 if (IS_ERR(trans
)) {
4361 ret
= PTR_ERR(trans
);
4368 * We checked in the above loop all device extents that were already in
4369 * the device tree. However before we have updated the device's
4370 * total_bytes to the new size, we might have had chunk allocations that
4371 * have not complete yet (new block groups attached to transaction
4372 * handles), and therefore their device extents were not yet in the
4373 * device tree and we missed them in the loop above. So if we have any
4374 * pending chunk using a device extent that overlaps the device range
4375 * that we can not use anymore, commit the current transaction and
4376 * repeat the search on the device tree - this way we guarantee we will
4377 * not have chunks using device extents that end beyond 'new_size'.
4379 if (!checked_pending_chunks
) {
4380 u64 start
= new_size
;
4381 u64 len
= old_size
- new_size
;
4383 if (contains_pending_extent(trans
->transaction
, device
,
4385 unlock_chunks(root
);
4386 checked_pending_chunks
= true;
4389 ret
= btrfs_commit_transaction(trans
, root
);
4396 btrfs_device_set_disk_total_bytes(device
, new_size
);
4397 if (list_empty(&device
->resized_list
))
4398 list_add_tail(&device
->resized_list
,
4399 &root
->fs_info
->fs_devices
->resized_devices
);
4401 WARN_ON(diff
> old_total
);
4402 btrfs_set_super_total_bytes(super_copy
, old_total
- diff
);
4403 unlock_chunks(root
);
4405 /* Now btrfs_update_device() will change the on-disk size. */
4406 ret
= btrfs_update_device(trans
, device
);
4407 btrfs_end_transaction(trans
, root
);
4409 btrfs_free_path(path
);
4412 btrfs_device_set_total_bytes(device
, old_size
);
4413 if (device
->writeable
)
4414 device
->fs_devices
->total_rw_bytes
+= diff
;
4415 spin_lock(&root
->fs_info
->free_chunk_lock
);
4416 root
->fs_info
->free_chunk_space
+= diff
;
4417 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4418 unlock_chunks(root
);
4423 static int btrfs_add_system_chunk(struct btrfs_root
*root
,
4424 struct btrfs_key
*key
,
4425 struct btrfs_chunk
*chunk
, int item_size
)
4427 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
4428 struct btrfs_disk_key disk_key
;
4433 array_size
= btrfs_super_sys_array_size(super_copy
);
4434 if (array_size
+ item_size
+ sizeof(disk_key
)
4435 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4436 unlock_chunks(root
);
4440 ptr
= super_copy
->sys_chunk_array
+ array_size
;
4441 btrfs_cpu_key_to_disk(&disk_key
, key
);
4442 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
4443 ptr
+= sizeof(disk_key
);
4444 memcpy(ptr
, chunk
, item_size
);
4445 item_size
+= sizeof(disk_key
);
4446 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
4447 unlock_chunks(root
);
4453 * sort the devices in descending order by max_avail, total_avail
4455 static int btrfs_cmp_device_info(const void *a
, const void *b
)
4457 const struct btrfs_device_info
*di_a
= a
;
4458 const struct btrfs_device_info
*di_b
= b
;
4460 if (di_a
->max_avail
> di_b
->max_avail
)
4462 if (di_a
->max_avail
< di_b
->max_avail
)
4464 if (di_a
->total_avail
> di_b
->total_avail
)
4466 if (di_a
->total_avail
< di_b
->total_avail
)
4471 static u32
find_raid56_stripe_len(u32 data_devices
, u32 dev_stripe_target
)
4473 /* TODO allow them to set a preferred stripe size */
4477 static void check_raid56_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
4479 if (!(type
& BTRFS_BLOCK_GROUP_RAID56_MASK
))
4482 btrfs_set_fs_incompat(info
, RAID56
);
4485 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4486 - sizeof(struct btrfs_item) \
4487 - sizeof(struct btrfs_chunk)) \
4488 / sizeof(struct btrfs_stripe) + 1)
4490 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4491 - 2 * sizeof(struct btrfs_disk_key) \
4492 - 2 * sizeof(struct btrfs_chunk)) \
4493 / sizeof(struct btrfs_stripe) + 1)
4495 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4496 struct btrfs_root
*extent_root
, u64 start
,
4499 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
4500 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
4501 struct list_head
*cur
;
4502 struct map_lookup
*map
= NULL
;
4503 struct extent_map_tree
*em_tree
;
4504 struct extent_map
*em
;
4505 struct btrfs_device_info
*devices_info
= NULL
;
4507 int num_stripes
; /* total number of stripes to allocate */
4508 int data_stripes
; /* number of stripes that count for
4510 int sub_stripes
; /* sub_stripes info for map */
4511 int dev_stripes
; /* stripes per dev */
4512 int devs_max
; /* max devs to use */
4513 int devs_min
; /* min devs needed */
4514 int devs_increment
; /* ndevs has to be a multiple of this */
4515 int ncopies
; /* how many copies to data has */
4517 u64 max_stripe_size
;
4521 u64 raid_stripe_len
= BTRFS_STRIPE_LEN
;
4527 BUG_ON(!alloc_profile_is_valid(type
, 0));
4529 if (list_empty(&fs_devices
->alloc_list
))
4532 index
= __get_raid_index(type
);
4534 sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
4535 dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
4536 devs_max
= btrfs_raid_array
[index
].devs_max
;
4537 devs_min
= btrfs_raid_array
[index
].devs_min
;
4538 devs_increment
= btrfs_raid_array
[index
].devs_increment
;
4539 ncopies
= btrfs_raid_array
[index
].ncopies
;
4541 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
4542 max_stripe_size
= 1024 * 1024 * 1024;
4543 max_chunk_size
= 10 * max_stripe_size
;
4545 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4546 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
4547 /* for larger filesystems, use larger metadata chunks */
4548 if (fs_devices
->total_rw_bytes
> 50ULL * 1024 * 1024 * 1024)
4549 max_stripe_size
= 1024 * 1024 * 1024;
4551 max_stripe_size
= 256 * 1024 * 1024;
4552 max_chunk_size
= max_stripe_size
;
4554 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4555 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4556 max_stripe_size
= 32 * 1024 * 1024;
4557 max_chunk_size
= 2 * max_stripe_size
;
4559 devs_max
= BTRFS_MAX_DEVS_SYS_CHUNK
;
4561 btrfs_err(info
, "invalid chunk type 0x%llx requested",
4566 /* we don't want a chunk larger than 10% of writeable space */
4567 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
4570 devices_info
= kcalloc(fs_devices
->rw_devices
, sizeof(*devices_info
),
4575 cur
= fs_devices
->alloc_list
.next
;
4578 * in the first pass through the devices list, we gather information
4579 * about the available holes on each device.
4582 while (cur
!= &fs_devices
->alloc_list
) {
4583 struct btrfs_device
*device
;
4587 device
= list_entry(cur
, struct btrfs_device
, dev_alloc_list
);
4591 if (!device
->writeable
) {
4593 "BTRFS: read-only device in alloc_list\n");
4597 if (!device
->in_fs_metadata
||
4598 device
->is_tgtdev_for_dev_replace
)
4601 if (device
->total_bytes
> device
->bytes_used
)
4602 total_avail
= device
->total_bytes
- device
->bytes_used
;
4606 /* If there is no space on this device, skip it. */
4607 if (total_avail
== 0)
4610 ret
= find_free_dev_extent(trans
, device
,
4611 max_stripe_size
* dev_stripes
,
4612 &dev_offset
, &max_avail
);
4613 if (ret
&& ret
!= -ENOSPC
)
4617 max_avail
= max_stripe_size
* dev_stripes
;
4619 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
4622 if (ndevs
== fs_devices
->rw_devices
) {
4623 WARN(1, "%s: found more than %llu devices\n",
4624 __func__
, fs_devices
->rw_devices
);
4627 devices_info
[ndevs
].dev_offset
= dev_offset
;
4628 devices_info
[ndevs
].max_avail
= max_avail
;
4629 devices_info
[ndevs
].total_avail
= total_avail
;
4630 devices_info
[ndevs
].dev
= device
;
4635 * now sort the devices by hole size / available space
4637 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
4638 btrfs_cmp_device_info
, NULL
);
4640 /* round down to number of usable stripes */
4641 ndevs
-= ndevs
% devs_increment
;
4643 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
4648 if (devs_max
&& ndevs
> devs_max
)
4651 * The primary goal is to maximize the number of stripes, so use as
4652 * many devices as possible, even if the stripes are not maximum sized.
4654 * The DUP profile stores more than one stripe per device, the
4655 * max_avail is the total size so we have to adjust.
4657 stripe_size
= div_u64(devices_info
[ndevs
- 1].max_avail
, dev_stripes
);
4658 num_stripes
= ndevs
* dev_stripes
;
4661 * this will have to be fixed for RAID1 and RAID10 over
4664 data_stripes
= num_stripes
/ ncopies
;
4666 if (type
& BTRFS_BLOCK_GROUP_RAID5
) {
4667 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 1,
4668 btrfs_super_stripesize(info
->super_copy
));
4669 data_stripes
= num_stripes
- 1;
4671 if (type
& BTRFS_BLOCK_GROUP_RAID6
) {
4672 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 2,
4673 btrfs_super_stripesize(info
->super_copy
));
4674 data_stripes
= num_stripes
- 2;
4678 * Use the number of data stripes to figure out how big this chunk
4679 * is really going to be in terms of logical address space,
4680 * and compare that answer with the max chunk size
4682 if (stripe_size
* data_stripes
> max_chunk_size
) {
4683 u64 mask
= (1ULL << 24) - 1;
4685 stripe_size
= div_u64(max_chunk_size
, data_stripes
);
4687 /* bump the answer up to a 16MB boundary */
4688 stripe_size
= (stripe_size
+ mask
) & ~mask
;
4690 /* but don't go higher than the limits we found
4691 * while searching for free extents
4693 if (stripe_size
> devices_info
[ndevs
-1].max_avail
)
4694 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4697 /* align to BTRFS_STRIPE_LEN */
4698 stripe_size
= div_u64(stripe_size
, raid_stripe_len
);
4699 stripe_size
*= raid_stripe_len
;
4701 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
4706 map
->num_stripes
= num_stripes
;
4708 for (i
= 0; i
< ndevs
; ++i
) {
4709 for (j
= 0; j
< dev_stripes
; ++j
) {
4710 int s
= i
* dev_stripes
+ j
;
4711 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
4712 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
4716 map
->sector_size
= extent_root
->sectorsize
;
4717 map
->stripe_len
= raid_stripe_len
;
4718 map
->io_align
= raid_stripe_len
;
4719 map
->io_width
= raid_stripe_len
;
4721 map
->sub_stripes
= sub_stripes
;
4723 num_bytes
= stripe_size
* data_stripes
;
4725 trace_btrfs_chunk_alloc(info
->chunk_root
, map
, start
, num_bytes
);
4727 em
= alloc_extent_map();
4733 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
4734 em
->bdev
= (struct block_device
*)map
;
4736 em
->len
= num_bytes
;
4737 em
->block_start
= 0;
4738 em
->block_len
= em
->len
;
4739 em
->orig_block_len
= stripe_size
;
4741 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
4742 write_lock(&em_tree
->lock
);
4743 ret
= add_extent_mapping(em_tree
, em
, 0);
4745 list_add_tail(&em
->list
, &trans
->transaction
->pending_chunks
);
4746 atomic_inc(&em
->refs
);
4748 write_unlock(&em_tree
->lock
);
4750 free_extent_map(em
);
4754 ret
= btrfs_make_block_group(trans
, extent_root
, 0, type
,
4755 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4758 goto error_del_extent
;
4760 for (i
= 0; i
< map
->num_stripes
; i
++) {
4761 num_bytes
= map
->stripes
[i
].dev
->bytes_used
+ stripe_size
;
4762 btrfs_device_set_bytes_used(map
->stripes
[i
].dev
, num_bytes
);
4765 spin_lock(&extent_root
->fs_info
->free_chunk_lock
);
4766 extent_root
->fs_info
->free_chunk_space
-= (stripe_size
*
4768 spin_unlock(&extent_root
->fs_info
->free_chunk_lock
);
4770 free_extent_map(em
);
4771 check_raid56_incompat_flag(extent_root
->fs_info
, type
);
4773 kfree(devices_info
);
4777 write_lock(&em_tree
->lock
);
4778 remove_extent_mapping(em_tree
, em
);
4779 write_unlock(&em_tree
->lock
);
4781 /* One for our allocation */
4782 free_extent_map(em
);
4783 /* One for the tree reference */
4784 free_extent_map(em
);
4785 /* One for the pending_chunks list reference */
4786 free_extent_map(em
);
4788 kfree(devices_info
);
4792 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
4793 struct btrfs_root
*extent_root
,
4794 u64 chunk_offset
, u64 chunk_size
)
4796 struct btrfs_key key
;
4797 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
4798 struct btrfs_device
*device
;
4799 struct btrfs_chunk
*chunk
;
4800 struct btrfs_stripe
*stripe
;
4801 struct extent_map_tree
*em_tree
;
4802 struct extent_map
*em
;
4803 struct map_lookup
*map
;
4810 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
4811 read_lock(&em_tree
->lock
);
4812 em
= lookup_extent_mapping(em_tree
, chunk_offset
, chunk_size
);
4813 read_unlock(&em_tree
->lock
);
4816 btrfs_crit(extent_root
->fs_info
, "unable to find logical "
4817 "%Lu len %Lu", chunk_offset
, chunk_size
);
4821 if (em
->start
!= chunk_offset
|| em
->len
!= chunk_size
) {
4822 btrfs_crit(extent_root
->fs_info
, "found a bad mapping, wanted"
4823 " %Lu-%Lu, found %Lu-%Lu", chunk_offset
,
4824 chunk_size
, em
->start
, em
->len
);
4825 free_extent_map(em
);
4829 map
= (struct map_lookup
*)em
->bdev
;
4830 item_size
= btrfs_chunk_item_size(map
->num_stripes
);
4831 stripe_size
= em
->orig_block_len
;
4833 chunk
= kzalloc(item_size
, GFP_NOFS
);
4839 for (i
= 0; i
< map
->num_stripes
; i
++) {
4840 device
= map
->stripes
[i
].dev
;
4841 dev_offset
= map
->stripes
[i
].physical
;
4843 ret
= btrfs_update_device(trans
, device
);
4846 ret
= btrfs_alloc_dev_extent(trans
, device
,
4847 chunk_root
->root_key
.objectid
,
4848 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4849 chunk_offset
, dev_offset
,
4855 stripe
= &chunk
->stripe
;
4856 for (i
= 0; i
< map
->num_stripes
; i
++) {
4857 device
= map
->stripes
[i
].dev
;
4858 dev_offset
= map
->stripes
[i
].physical
;
4860 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
4861 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
4862 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
4866 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
4867 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
4868 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
4869 btrfs_set_stack_chunk_type(chunk
, map
->type
);
4870 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
4871 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
4872 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
4873 btrfs_set_stack_chunk_sector_size(chunk
, extent_root
->sectorsize
);
4874 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
4876 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
4877 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
4878 key
.offset
= chunk_offset
;
4880 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
4881 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4883 * TODO: Cleanup of inserted chunk root in case of
4886 ret
= btrfs_add_system_chunk(chunk_root
, &key
, chunk
,
4892 free_extent_map(em
);
4897 * Chunk allocation falls into two parts. The first part does works
4898 * that make the new allocated chunk useable, but not do any operation
4899 * that modifies the chunk tree. The second part does the works that
4900 * require modifying the chunk tree. This division is important for the
4901 * bootstrap process of adding storage to a seed btrfs.
4903 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4904 struct btrfs_root
*extent_root
, u64 type
)
4908 ASSERT(mutex_is_locked(&extent_root
->fs_info
->chunk_mutex
));
4909 chunk_offset
= find_next_chunk(extent_root
->fs_info
);
4910 return __btrfs_alloc_chunk(trans
, extent_root
, chunk_offset
, type
);
4913 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
4914 struct btrfs_root
*root
,
4915 struct btrfs_device
*device
)
4918 u64 sys_chunk_offset
;
4920 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4921 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
4924 chunk_offset
= find_next_chunk(fs_info
);
4925 alloc_profile
= btrfs_get_alloc_profile(extent_root
, 0);
4926 ret
= __btrfs_alloc_chunk(trans
, extent_root
, chunk_offset
,
4931 sys_chunk_offset
= find_next_chunk(root
->fs_info
);
4932 alloc_profile
= btrfs_get_alloc_profile(fs_info
->chunk_root
, 0);
4933 ret
= __btrfs_alloc_chunk(trans
, extent_root
, sys_chunk_offset
,
4938 static inline int btrfs_chunk_max_errors(struct map_lookup
*map
)
4942 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
4943 BTRFS_BLOCK_GROUP_RAID10
|
4944 BTRFS_BLOCK_GROUP_RAID5
|
4945 BTRFS_BLOCK_GROUP_DUP
)) {
4947 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
) {
4956 int btrfs_chunk_readonly(struct btrfs_root
*root
, u64 chunk_offset
)
4958 struct extent_map
*em
;
4959 struct map_lookup
*map
;
4960 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
4965 read_lock(&map_tree
->map_tree
.lock
);
4966 em
= lookup_extent_mapping(&map_tree
->map_tree
, chunk_offset
, 1);
4967 read_unlock(&map_tree
->map_tree
.lock
);
4971 map
= (struct map_lookup
*)em
->bdev
;
4972 for (i
= 0; i
< map
->num_stripes
; i
++) {
4973 if (map
->stripes
[i
].dev
->missing
) {
4978 if (!map
->stripes
[i
].dev
->writeable
) {
4985 * If the number of missing devices is larger than max errors,
4986 * we can not write the data into that chunk successfully, so
4989 if (miss_ndevs
> btrfs_chunk_max_errors(map
))
4992 free_extent_map(em
);
4996 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
4998 extent_map_tree_init(&tree
->map_tree
);
5001 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
5003 struct extent_map
*em
;
5006 write_lock(&tree
->map_tree
.lock
);
5007 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
5009 remove_extent_mapping(&tree
->map_tree
, em
);
5010 write_unlock(&tree
->map_tree
.lock
);
5014 free_extent_map(em
);
5015 /* once for the tree */
5016 free_extent_map(em
);
5020 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
5022 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
5023 struct extent_map
*em
;
5024 struct map_lookup
*map
;
5025 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5028 read_lock(&em_tree
->lock
);
5029 em
= lookup_extent_mapping(em_tree
, logical
, len
);
5030 read_unlock(&em_tree
->lock
);
5033 * We could return errors for these cases, but that could get ugly and
5034 * we'd probably do the same thing which is just not do anything else
5035 * and exit, so return 1 so the callers don't try to use other copies.
5038 btrfs_crit(fs_info
, "No mapping for %Lu-%Lu", logical
,
5043 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
5044 btrfs_crit(fs_info
, "Invalid mapping for %Lu-%Lu, got "
5045 "%Lu-%Lu", logical
, logical
+len
, em
->start
,
5046 em
->start
+ em
->len
);
5047 free_extent_map(em
);
5051 map
= (struct map_lookup
*)em
->bdev
;
5052 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
5053 ret
= map
->num_stripes
;
5054 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5055 ret
= map
->sub_stripes
;
5056 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID5
)
5058 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5060 * There could be two corrupted data stripes, we need
5061 * to loop retry in order to rebuild the correct data.
5063 * Fail a stripe at a time on every retry except the
5064 * stripe under reconstruction.
5066 ret
= map
->num_stripes
;
5069 free_extent_map(em
);
5071 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
5072 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
))
5074 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
5079 unsigned long btrfs_full_stripe_len(struct btrfs_root
*root
,
5080 struct btrfs_mapping_tree
*map_tree
,
5083 struct extent_map
*em
;
5084 struct map_lookup
*map
;
5085 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5086 unsigned long len
= root
->sectorsize
;
5088 read_lock(&em_tree
->lock
);
5089 em
= lookup_extent_mapping(em_tree
, logical
, len
);
5090 read_unlock(&em_tree
->lock
);
5093 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
5094 map
= (struct map_lookup
*)em
->bdev
;
5095 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5096 len
= map
->stripe_len
* nr_data_stripes(map
);
5097 free_extent_map(em
);
5101 int btrfs_is_parity_mirror(struct btrfs_mapping_tree
*map_tree
,
5102 u64 logical
, u64 len
, int mirror_num
)
5104 struct extent_map
*em
;
5105 struct map_lookup
*map
;
5106 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5109 read_lock(&em_tree
->lock
);
5110 em
= lookup_extent_mapping(em_tree
, logical
, len
);
5111 read_unlock(&em_tree
->lock
);
5114 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
5115 map
= (struct map_lookup
*)em
->bdev
;
5116 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5118 free_extent_map(em
);
5122 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
5123 struct map_lookup
*map
, int first
, int num
,
5124 int optimal
, int dev_replace_is_ongoing
)
5128 struct btrfs_device
*srcdev
;
5130 if (dev_replace_is_ongoing
&&
5131 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
5132 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
5133 srcdev
= fs_info
->dev_replace
.srcdev
;
5138 * try to avoid the drive that is the source drive for a
5139 * dev-replace procedure, only choose it if no other non-missing
5140 * mirror is available
5142 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
5143 if (map
->stripes
[optimal
].dev
->bdev
&&
5144 (tolerance
|| map
->stripes
[optimal
].dev
!= srcdev
))
5146 for (i
= first
; i
< first
+ num
; i
++) {
5147 if (map
->stripes
[i
].dev
->bdev
&&
5148 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
5153 /* we couldn't find one that doesn't fail. Just return something
5154 * and the io error handling code will clean up eventually
5159 static inline int parity_smaller(u64 a
, u64 b
)
5164 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5165 static void sort_parity_stripes(struct btrfs_bio
*bbio
, int num_stripes
)
5167 struct btrfs_bio_stripe s
;
5174 for (i
= 0; i
< num_stripes
- 1; i
++) {
5175 if (parity_smaller(bbio
->raid_map
[i
],
5176 bbio
->raid_map
[i
+1])) {
5177 s
= bbio
->stripes
[i
];
5178 l
= bbio
->raid_map
[i
];
5179 bbio
->stripes
[i
] = bbio
->stripes
[i
+1];
5180 bbio
->raid_map
[i
] = bbio
->raid_map
[i
+1];
5181 bbio
->stripes
[i
+1] = s
;
5182 bbio
->raid_map
[i
+1] = l
;
5190 static struct btrfs_bio
*alloc_btrfs_bio(int total_stripes
, int real_stripes
)
5192 struct btrfs_bio
*bbio
= kzalloc(
5193 /* the size of the btrfs_bio */
5194 sizeof(struct btrfs_bio
) +
5195 /* plus the variable array for the stripes */
5196 sizeof(struct btrfs_bio_stripe
) * (total_stripes
) +
5197 /* plus the variable array for the tgt dev */
5198 sizeof(int) * (real_stripes
) +
5200 * plus the raid_map, which includes both the tgt dev
5203 sizeof(u64
) * (total_stripes
),
5204 GFP_NOFS
|__GFP_NOFAIL
);
5206 atomic_set(&bbio
->error
, 0);
5207 atomic_set(&bbio
->refs
, 1);
5212 void btrfs_get_bbio(struct btrfs_bio
*bbio
)
5214 WARN_ON(!atomic_read(&bbio
->refs
));
5215 atomic_inc(&bbio
->refs
);
5218 void btrfs_put_bbio(struct btrfs_bio
*bbio
)
5222 if (atomic_dec_and_test(&bbio
->refs
))
5226 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
5227 u64 logical
, u64
*length
,
5228 struct btrfs_bio
**bbio_ret
,
5229 int mirror_num
, int need_raid_map
)
5231 struct extent_map
*em
;
5232 struct map_lookup
*map
;
5233 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
5234 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5237 u64 stripe_end_offset
;
5247 int tgtdev_indexes
= 0;
5248 struct btrfs_bio
*bbio
= NULL
;
5249 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
5250 int dev_replace_is_ongoing
= 0;
5251 int num_alloc_stripes
;
5252 int patch_the_first_stripe_for_dev_replace
= 0;
5253 u64 physical_to_patch_in_first_stripe
= 0;
5254 u64 raid56_full_stripe_start
= (u64
)-1;
5256 read_lock(&em_tree
->lock
);
5257 em
= lookup_extent_mapping(em_tree
, logical
, *length
);
5258 read_unlock(&em_tree
->lock
);
5261 btrfs_crit(fs_info
, "unable to find logical %llu len %llu",
5266 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
5267 btrfs_crit(fs_info
, "found a bad mapping, wanted %Lu, "
5268 "found %Lu-%Lu", logical
, em
->start
,
5269 em
->start
+ em
->len
);
5270 free_extent_map(em
);
5274 map
= (struct map_lookup
*)em
->bdev
;
5275 offset
= logical
- em
->start
;
5277 stripe_len
= map
->stripe_len
;
5280 * stripe_nr counts the total number of stripes we have to stride
5281 * to get to this block
5283 stripe_nr
= div64_u64(stripe_nr
, stripe_len
);
5285 stripe_offset
= stripe_nr
* stripe_len
;
5286 BUG_ON(offset
< stripe_offset
);
5288 /* stripe_offset is the offset of this block in its stripe*/
5289 stripe_offset
= offset
- stripe_offset
;
5291 /* if we're here for raid56, we need to know the stripe aligned start */
5292 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5293 unsigned long full_stripe_len
= stripe_len
* nr_data_stripes(map
);
5294 raid56_full_stripe_start
= offset
;
5296 /* allow a write of a full stripe, but make sure we don't
5297 * allow straddling of stripes
5299 raid56_full_stripe_start
= div64_u64(raid56_full_stripe_start
,
5301 raid56_full_stripe_start
*= full_stripe_len
;
5304 if (rw
& REQ_DISCARD
) {
5305 /* we don't discard raid56 yet */
5306 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5310 *length
= min_t(u64
, em
->len
- offset
, *length
);
5311 } else if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
5313 /* For writes to RAID[56], allow a full stripeset across all disks.
5314 For other RAID types and for RAID[56] reads, just allow a single
5315 stripe (on a single disk). */
5316 if ((map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
5318 max_len
= stripe_len
* nr_data_stripes(map
) -
5319 (offset
- raid56_full_stripe_start
);
5321 /* we limit the length of each bio to what fits in a stripe */
5322 max_len
= stripe_len
- stripe_offset
;
5324 *length
= min_t(u64
, em
->len
- offset
, max_len
);
5326 *length
= em
->len
- offset
;
5329 /* This is for when we're called from btrfs_merge_bio_hook() and all
5330 it cares about is the length */
5334 btrfs_dev_replace_lock(dev_replace
);
5335 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
5336 if (!dev_replace_is_ongoing
)
5337 btrfs_dev_replace_unlock(dev_replace
);
5339 if (dev_replace_is_ongoing
&& mirror_num
== map
->num_stripes
+ 1 &&
5340 !(rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) &&
5341 dev_replace
->tgtdev
!= NULL
) {
5343 * in dev-replace case, for repair case (that's the only
5344 * case where the mirror is selected explicitly when
5345 * calling btrfs_map_block), blocks left of the left cursor
5346 * can also be read from the target drive.
5347 * For REQ_GET_READ_MIRRORS, the target drive is added as
5348 * the last one to the array of stripes. For READ, it also
5349 * needs to be supported using the same mirror number.
5350 * If the requested block is not left of the left cursor,
5351 * EIO is returned. This can happen because btrfs_num_copies()
5352 * returns one more in the dev-replace case.
5354 u64 tmp_length
= *length
;
5355 struct btrfs_bio
*tmp_bbio
= NULL
;
5356 int tmp_num_stripes
;
5357 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5358 int index_srcdev
= 0;
5360 u64 physical_of_found
= 0;
5362 ret
= __btrfs_map_block(fs_info
, REQ_GET_READ_MIRRORS
,
5363 logical
, &tmp_length
, &tmp_bbio
, 0, 0);
5365 WARN_ON(tmp_bbio
!= NULL
);
5369 tmp_num_stripes
= tmp_bbio
->num_stripes
;
5370 if (mirror_num
> tmp_num_stripes
) {
5372 * REQ_GET_READ_MIRRORS does not contain this
5373 * mirror, that means that the requested area
5374 * is not left of the left cursor
5377 btrfs_put_bbio(tmp_bbio
);
5382 * process the rest of the function using the mirror_num
5383 * of the source drive. Therefore look it up first.
5384 * At the end, patch the device pointer to the one of the
5387 for (i
= 0; i
< tmp_num_stripes
; i
++) {
5388 if (tmp_bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5390 * In case of DUP, in order to keep it
5391 * simple, only add the mirror with the
5392 * lowest physical address
5395 physical_of_found
<=
5396 tmp_bbio
->stripes
[i
].physical
)
5401 tmp_bbio
->stripes
[i
].physical
;
5406 mirror_num
= index_srcdev
+ 1;
5407 patch_the_first_stripe_for_dev_replace
= 1;
5408 physical_to_patch_in_first_stripe
= physical_of_found
;
5412 btrfs_put_bbio(tmp_bbio
);
5416 btrfs_put_bbio(tmp_bbio
);
5417 } else if (mirror_num
> map
->num_stripes
) {
5423 stripe_nr_orig
= stripe_nr
;
5424 stripe_nr_end
= ALIGN(offset
+ *length
, map
->stripe_len
);
5425 stripe_nr_end
= div_u64(stripe_nr_end
, map
->stripe_len
);
5426 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
5429 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5430 if (rw
& REQ_DISCARD
)
5431 num_stripes
= min_t(u64
, map
->num_stripes
,
5432 stripe_nr_end
- stripe_nr_orig
);
5433 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5435 if (!(rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)))
5437 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
5438 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
))
5439 num_stripes
= map
->num_stripes
;
5440 else if (mirror_num
)
5441 stripe_index
= mirror_num
- 1;
5443 stripe_index
= find_live_mirror(fs_info
, map
, 0,
5445 current
->pid
% map
->num_stripes
,
5446 dev_replace_is_ongoing
);
5447 mirror_num
= stripe_index
+ 1;
5450 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
5451 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) {
5452 num_stripes
= map
->num_stripes
;
5453 } else if (mirror_num
) {
5454 stripe_index
= mirror_num
- 1;
5459 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5460 u32 factor
= map
->num_stripes
/ map
->sub_stripes
;
5462 stripe_nr
= div_u64_rem(stripe_nr
, factor
, &stripe_index
);
5463 stripe_index
*= map
->sub_stripes
;
5465 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
))
5466 num_stripes
= map
->sub_stripes
;
5467 else if (rw
& REQ_DISCARD
)
5468 num_stripes
= min_t(u64
, map
->sub_stripes
*
5469 (stripe_nr_end
- stripe_nr_orig
),
5471 else if (mirror_num
)
5472 stripe_index
+= mirror_num
- 1;
5474 int old_stripe_index
= stripe_index
;
5475 stripe_index
= find_live_mirror(fs_info
, map
,
5477 map
->sub_stripes
, stripe_index
+
5478 current
->pid
% map
->sub_stripes
,
5479 dev_replace_is_ongoing
);
5480 mirror_num
= stripe_index
- old_stripe_index
+ 1;
5483 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5484 if (need_raid_map
&&
5485 ((rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
)) ||
5487 /* push stripe_nr back to the start of the full stripe */
5488 stripe_nr
= div_u64(raid56_full_stripe_start
,
5489 stripe_len
* nr_data_stripes(map
));
5491 /* RAID[56] write or recovery. Return all stripes */
5492 num_stripes
= map
->num_stripes
;
5493 max_errors
= nr_parity_stripes(map
);
5495 *length
= map
->stripe_len
;
5500 * Mirror #0 or #1 means the original data block.
5501 * Mirror #2 is RAID5 parity block.
5502 * Mirror #3 is RAID6 Q block.
5504 stripe_nr
= div_u64_rem(stripe_nr
,
5505 nr_data_stripes(map
), &stripe_index
);
5507 stripe_index
= nr_data_stripes(map
) +
5510 /* We distribute the parity blocks across stripes */
5511 div_u64_rem(stripe_nr
+ stripe_index
, map
->num_stripes
,
5513 if (!(rw
& (REQ_WRITE
| REQ_DISCARD
|
5514 REQ_GET_READ_MIRRORS
)) && mirror_num
<= 1)
5519 * after this, stripe_nr is the number of stripes on this
5520 * device we have to walk to find the data, and stripe_index is
5521 * the number of our device in the stripe array
5523 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5525 mirror_num
= stripe_index
+ 1;
5527 BUG_ON(stripe_index
>= map
->num_stripes
);
5529 num_alloc_stripes
= num_stripes
;
5530 if (dev_replace_is_ongoing
) {
5531 if (rw
& (REQ_WRITE
| REQ_DISCARD
))
5532 num_alloc_stripes
<<= 1;
5533 if (rw
& REQ_GET_READ_MIRRORS
)
5534 num_alloc_stripes
++;
5535 tgtdev_indexes
= num_stripes
;
5538 bbio
= alloc_btrfs_bio(num_alloc_stripes
, tgtdev_indexes
);
5543 if (dev_replace_is_ongoing
)
5544 bbio
->tgtdev_map
= (int *)(bbio
->stripes
+ num_alloc_stripes
);
5546 /* build raid_map */
5547 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
&&
5548 need_raid_map
&& ((rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
)) ||
5553 bbio
->raid_map
= (u64
*)((void *)bbio
->stripes
+
5554 sizeof(struct btrfs_bio_stripe
) *
5556 sizeof(int) * tgtdev_indexes
);
5558 /* Work out the disk rotation on this stripe-set */
5559 div_u64_rem(stripe_nr
, num_stripes
, &rot
);
5561 /* Fill in the logical address of each stripe */
5562 tmp
= stripe_nr
* nr_data_stripes(map
);
5563 for (i
= 0; i
< nr_data_stripes(map
); i
++)
5564 bbio
->raid_map
[(i
+rot
) % num_stripes
] =
5565 em
->start
+ (tmp
+ i
) * map
->stripe_len
;
5567 bbio
->raid_map
[(i
+rot
) % map
->num_stripes
] = RAID5_P_STRIPE
;
5568 if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5569 bbio
->raid_map
[(i
+rot
+1) % num_stripes
] =
5573 if (rw
& REQ_DISCARD
) {
5575 u32 sub_stripes
= 0;
5576 u64 stripes_per_dev
= 0;
5577 u32 remaining_stripes
= 0;
5578 u32 last_stripe
= 0;
5581 (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID10
)) {
5582 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5585 sub_stripes
= map
->sub_stripes
;
5587 factor
= map
->num_stripes
/ sub_stripes
;
5588 stripes_per_dev
= div_u64_rem(stripe_nr_end
-
5591 &remaining_stripes
);
5592 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
5593 last_stripe
*= sub_stripes
;
5596 for (i
= 0; i
< num_stripes
; i
++) {
5597 bbio
->stripes
[i
].physical
=
5598 map
->stripes
[stripe_index
].physical
+
5599 stripe_offset
+ stripe_nr
* map
->stripe_len
;
5600 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
5602 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5603 BTRFS_BLOCK_GROUP_RAID10
)) {
5604 bbio
->stripes
[i
].length
= stripes_per_dev
*
5607 if (i
/ sub_stripes
< remaining_stripes
)
5608 bbio
->stripes
[i
].length
+=
5612 * Special for the first stripe and
5615 * |-------|...|-------|
5619 if (i
< sub_stripes
)
5620 bbio
->stripes
[i
].length
-=
5623 if (stripe_index
>= last_stripe
&&
5624 stripe_index
<= (last_stripe
+
5626 bbio
->stripes
[i
].length
-=
5629 if (i
== sub_stripes
- 1)
5632 bbio
->stripes
[i
].length
= *length
;
5635 if (stripe_index
== map
->num_stripes
) {
5636 /* This could only happen for RAID0/10 */
5642 for (i
= 0; i
< num_stripes
; i
++) {
5643 bbio
->stripes
[i
].physical
=
5644 map
->stripes
[stripe_index
].physical
+
5646 stripe_nr
* map
->stripe_len
;
5647 bbio
->stripes
[i
].dev
=
5648 map
->stripes
[stripe_index
].dev
;
5653 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
))
5654 max_errors
= btrfs_chunk_max_errors(map
);
5657 sort_parity_stripes(bbio
, num_stripes
);
5660 if (dev_replace_is_ongoing
&& (rw
& (REQ_WRITE
| REQ_DISCARD
)) &&
5661 dev_replace
->tgtdev
!= NULL
) {
5662 int index_where_to_add
;
5663 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5666 * duplicate the write operations while the dev replace
5667 * procedure is running. Since the copying of the old disk
5668 * to the new disk takes place at run time while the
5669 * filesystem is mounted writable, the regular write
5670 * operations to the old disk have to be duplicated to go
5671 * to the new disk as well.
5672 * Note that device->missing is handled by the caller, and
5673 * that the write to the old disk is already set up in the
5676 index_where_to_add
= num_stripes
;
5677 for (i
= 0; i
< num_stripes
; i
++) {
5678 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5679 /* write to new disk, too */
5680 struct btrfs_bio_stripe
*new =
5681 bbio
->stripes
+ index_where_to_add
;
5682 struct btrfs_bio_stripe
*old
=
5685 new->physical
= old
->physical
;
5686 new->length
= old
->length
;
5687 new->dev
= dev_replace
->tgtdev
;
5688 bbio
->tgtdev_map
[i
] = index_where_to_add
;
5689 index_where_to_add
++;
5694 num_stripes
= index_where_to_add
;
5695 } else if (dev_replace_is_ongoing
&& (rw
& REQ_GET_READ_MIRRORS
) &&
5696 dev_replace
->tgtdev
!= NULL
) {
5697 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5698 int index_srcdev
= 0;
5700 u64 physical_of_found
= 0;
5703 * During the dev-replace procedure, the target drive can
5704 * also be used to read data in case it is needed to repair
5705 * a corrupt block elsewhere. This is possible if the
5706 * requested area is left of the left cursor. In this area,
5707 * the target drive is a full copy of the source drive.
5709 for (i
= 0; i
< num_stripes
; i
++) {
5710 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5712 * In case of DUP, in order to keep it
5713 * simple, only add the mirror with the
5714 * lowest physical address
5717 physical_of_found
<=
5718 bbio
->stripes
[i
].physical
)
5722 physical_of_found
= bbio
->stripes
[i
].physical
;
5726 if (physical_of_found
+ map
->stripe_len
<=
5727 dev_replace
->cursor_left
) {
5728 struct btrfs_bio_stripe
*tgtdev_stripe
=
5729 bbio
->stripes
+ num_stripes
;
5731 tgtdev_stripe
->physical
= physical_of_found
;
5732 tgtdev_stripe
->length
=
5733 bbio
->stripes
[index_srcdev
].length
;
5734 tgtdev_stripe
->dev
= dev_replace
->tgtdev
;
5735 bbio
->tgtdev_map
[index_srcdev
] = num_stripes
;
5744 bbio
->map_type
= map
->type
;
5745 bbio
->num_stripes
= num_stripes
;
5746 bbio
->max_errors
= max_errors
;
5747 bbio
->mirror_num
= mirror_num
;
5748 bbio
->num_tgtdevs
= tgtdev_indexes
;
5751 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5752 * mirror_num == num_stripes + 1 && dev_replace target drive is
5753 * available as a mirror
5755 if (patch_the_first_stripe_for_dev_replace
&& num_stripes
> 0) {
5756 WARN_ON(num_stripes
> 1);
5757 bbio
->stripes
[0].dev
= dev_replace
->tgtdev
;
5758 bbio
->stripes
[0].physical
= physical_to_patch_in_first_stripe
;
5759 bbio
->mirror_num
= map
->num_stripes
+ 1;
5762 if (dev_replace_is_ongoing
)
5763 btrfs_dev_replace_unlock(dev_replace
);
5764 free_extent_map(em
);
5768 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
5769 u64 logical
, u64
*length
,
5770 struct btrfs_bio
**bbio_ret
, int mirror_num
)
5772 return __btrfs_map_block(fs_info
, rw
, logical
, length
, bbio_ret
,
5776 /* For Scrub/replace */
5777 int btrfs_map_sblock(struct btrfs_fs_info
*fs_info
, int rw
,
5778 u64 logical
, u64
*length
,
5779 struct btrfs_bio
**bbio_ret
, int mirror_num
,
5782 return __btrfs_map_block(fs_info
, rw
, logical
, length
, bbio_ret
,
5783 mirror_num
, need_raid_map
);
5786 int btrfs_rmap_block(struct btrfs_mapping_tree
*map_tree
,
5787 u64 chunk_start
, u64 physical
, u64 devid
,
5788 u64
**logical
, int *naddrs
, int *stripe_len
)
5790 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5791 struct extent_map
*em
;
5792 struct map_lookup
*map
;
5800 read_lock(&em_tree
->lock
);
5801 em
= lookup_extent_mapping(em_tree
, chunk_start
, 1);
5802 read_unlock(&em_tree
->lock
);
5805 printk(KERN_ERR
"BTRFS: couldn't find em for chunk %Lu\n",
5810 if (em
->start
!= chunk_start
) {
5811 printk(KERN_ERR
"BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5812 em
->start
, chunk_start
);
5813 free_extent_map(em
);
5816 map
= (struct map_lookup
*)em
->bdev
;
5819 rmap_len
= map
->stripe_len
;
5821 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5822 length
= div_u64(length
, map
->num_stripes
/ map
->sub_stripes
);
5823 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5824 length
= div_u64(length
, map
->num_stripes
);
5825 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5826 length
= div_u64(length
, nr_data_stripes(map
));
5827 rmap_len
= map
->stripe_len
* nr_data_stripes(map
);
5830 buf
= kcalloc(map
->num_stripes
, sizeof(u64
), GFP_NOFS
);
5831 BUG_ON(!buf
); /* -ENOMEM */
5833 for (i
= 0; i
< map
->num_stripes
; i
++) {
5834 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
5836 if (map
->stripes
[i
].physical
> physical
||
5837 map
->stripes
[i
].physical
+ length
<= physical
)
5840 stripe_nr
= physical
- map
->stripes
[i
].physical
;
5841 stripe_nr
= div_u64(stripe_nr
, map
->stripe_len
);
5843 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5844 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5845 stripe_nr
= div_u64(stripe_nr
, map
->sub_stripes
);
5846 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5847 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5848 } /* else if RAID[56], multiply by nr_data_stripes().
5849 * Alternatively, just use rmap_len below instead of
5850 * map->stripe_len */
5852 bytenr
= chunk_start
+ stripe_nr
* rmap_len
;
5853 WARN_ON(nr
>= map
->num_stripes
);
5854 for (j
= 0; j
< nr
; j
++) {
5855 if (buf
[j
] == bytenr
)
5859 WARN_ON(nr
>= map
->num_stripes
);
5866 *stripe_len
= rmap_len
;
5868 free_extent_map(em
);
5872 static inline void btrfs_end_bbio(struct btrfs_bio
*bbio
, struct bio
*bio
)
5874 bio
->bi_private
= bbio
->private;
5875 bio
->bi_end_io
= bbio
->end_io
;
5878 btrfs_put_bbio(bbio
);
5881 static void btrfs_end_bio(struct bio
*bio
)
5883 struct btrfs_bio
*bbio
= bio
->bi_private
;
5884 int is_orig_bio
= 0;
5886 if (bio
->bi_error
) {
5887 atomic_inc(&bbio
->error
);
5888 if (bio
->bi_error
== -EIO
|| bio
->bi_error
== -EREMOTEIO
) {
5889 unsigned int stripe_index
=
5890 btrfs_io_bio(bio
)->stripe_index
;
5891 struct btrfs_device
*dev
;
5893 BUG_ON(stripe_index
>= bbio
->num_stripes
);
5894 dev
= bbio
->stripes
[stripe_index
].dev
;
5896 if (bio
->bi_rw
& WRITE
)
5897 btrfs_dev_stat_inc(dev
,
5898 BTRFS_DEV_STAT_WRITE_ERRS
);
5900 btrfs_dev_stat_inc(dev
,
5901 BTRFS_DEV_STAT_READ_ERRS
);
5902 if ((bio
->bi_rw
& WRITE_FLUSH
) == WRITE_FLUSH
)
5903 btrfs_dev_stat_inc(dev
,
5904 BTRFS_DEV_STAT_FLUSH_ERRS
);
5905 btrfs_dev_stat_print_on_error(dev
);
5910 if (bio
== bbio
->orig_bio
)
5913 btrfs_bio_counter_dec(bbio
->fs_info
);
5915 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
5918 bio
= bbio
->orig_bio
;
5921 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
5922 /* only send an error to the higher layers if it is
5923 * beyond the tolerance of the btrfs bio
5925 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
5926 bio
->bi_error
= -EIO
;
5929 * this bio is actually up to date, we didn't
5930 * go over the max number of errors
5935 btrfs_end_bbio(bbio
, bio
);
5936 } else if (!is_orig_bio
) {
5942 * see run_scheduled_bios for a description of why bios are collected for
5945 * This will add one bio to the pending list for a device and make sure
5946 * the work struct is scheduled.
5948 static noinline
void btrfs_schedule_bio(struct btrfs_root
*root
,
5949 struct btrfs_device
*device
,
5950 int rw
, struct bio
*bio
)
5952 int should_queue
= 1;
5953 struct btrfs_pending_bios
*pending_bios
;
5955 if (device
->missing
|| !device
->bdev
) {
5960 /* don't bother with additional async steps for reads, right now */
5961 if (!(rw
& REQ_WRITE
)) {
5963 btrfsic_submit_bio(rw
, bio
);
5969 * nr_async_bios allows us to reliably return congestion to the
5970 * higher layers. Otherwise, the async bio makes it appear we have
5971 * made progress against dirty pages when we've really just put it
5972 * on a queue for later
5974 atomic_inc(&root
->fs_info
->nr_async_bios
);
5975 WARN_ON(bio
->bi_next
);
5976 bio
->bi_next
= NULL
;
5979 spin_lock(&device
->io_lock
);
5980 if (bio
->bi_rw
& REQ_SYNC
)
5981 pending_bios
= &device
->pending_sync_bios
;
5983 pending_bios
= &device
->pending_bios
;
5985 if (pending_bios
->tail
)
5986 pending_bios
->tail
->bi_next
= bio
;
5988 pending_bios
->tail
= bio
;
5989 if (!pending_bios
->head
)
5990 pending_bios
->head
= bio
;
5991 if (device
->running_pending
)
5994 spin_unlock(&device
->io_lock
);
5997 btrfs_queue_work(root
->fs_info
->submit_workers
,
6001 static void submit_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
6002 struct bio
*bio
, u64 physical
, int dev_nr
,
6005 struct btrfs_device
*dev
= bbio
->stripes
[dev_nr
].dev
;
6007 bio
->bi_private
= bbio
;
6008 btrfs_io_bio(bio
)->stripe_index
= dev_nr
;
6009 bio
->bi_end_io
= btrfs_end_bio
;
6010 bio
->bi_iter
.bi_sector
= physical
>> 9;
6013 struct rcu_string
*name
;
6016 name
= rcu_dereference(dev
->name
);
6017 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
6018 "(%s id %llu), size=%u\n", rw
,
6019 (u64
)bio
->bi_iter
.bi_sector
, (u_long
)dev
->bdev
->bd_dev
,
6020 name
->str
, dev
->devid
, bio
->bi_iter
.bi_size
);
6024 bio
->bi_bdev
= dev
->bdev
;
6026 btrfs_bio_counter_inc_noblocked(root
->fs_info
);
6029 btrfs_schedule_bio(root
, dev
, rw
, bio
);
6031 btrfsic_submit_bio(rw
, bio
);
6034 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
6036 atomic_inc(&bbio
->error
);
6037 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
6038 /* Shoud be the original bio. */
6039 WARN_ON(bio
!= bbio
->orig_bio
);
6041 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
6042 bio
->bi_iter
.bi_sector
= logical
>> 9;
6043 bio
->bi_error
= -EIO
;
6044 btrfs_end_bbio(bbio
, bio
);
6048 int btrfs_map_bio(struct btrfs_root
*root
, int rw
, struct bio
*bio
,
6049 int mirror_num
, int async_submit
)
6051 struct btrfs_device
*dev
;
6052 struct bio
*first_bio
= bio
;
6053 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
6059 struct btrfs_bio
*bbio
= NULL
;
6061 length
= bio
->bi_iter
.bi_size
;
6062 map_length
= length
;
6064 btrfs_bio_counter_inc_blocked(root
->fs_info
);
6065 ret
= __btrfs_map_block(root
->fs_info
, rw
, logical
, &map_length
, &bbio
,
6068 btrfs_bio_counter_dec(root
->fs_info
);
6072 total_devs
= bbio
->num_stripes
;
6073 bbio
->orig_bio
= first_bio
;
6074 bbio
->private = first_bio
->bi_private
;
6075 bbio
->end_io
= first_bio
->bi_end_io
;
6076 bbio
->fs_info
= root
->fs_info
;
6077 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
6079 if (bbio
->raid_map
) {
6080 /* In this case, map_length has been set to the length of
6081 a single stripe; not the whole write */
6083 ret
= raid56_parity_write(root
, bio
, bbio
, map_length
);
6085 ret
= raid56_parity_recover(root
, bio
, bbio
, map_length
,
6089 btrfs_bio_counter_dec(root
->fs_info
);
6093 if (map_length
< length
) {
6094 btrfs_crit(root
->fs_info
, "mapping failed logical %llu bio len %llu len %llu",
6095 logical
, length
, map_length
);
6099 for (dev_nr
= 0; dev_nr
< total_devs
; dev_nr
++) {
6100 dev
= bbio
->stripes
[dev_nr
].dev
;
6101 if (!dev
|| !dev
->bdev
|| (rw
& WRITE
&& !dev
->writeable
)) {
6102 bbio_error(bbio
, first_bio
, logical
);
6106 if (dev_nr
< total_devs
- 1) {
6107 bio
= btrfs_bio_clone(first_bio
, GFP_NOFS
);
6108 BUG_ON(!bio
); /* -ENOMEM */
6112 submit_stripe_bio(root
, bbio
, bio
,
6113 bbio
->stripes
[dev_nr
].physical
, dev_nr
, rw
,
6116 btrfs_bio_counter_dec(root
->fs_info
);
6120 struct btrfs_device
*btrfs_find_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
6123 struct btrfs_device
*device
;
6124 struct btrfs_fs_devices
*cur_devices
;
6126 cur_devices
= fs_info
->fs_devices
;
6127 while (cur_devices
) {
6129 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
6130 device
= __find_device(&cur_devices
->devices
,
6135 cur_devices
= cur_devices
->seed
;
6140 static struct btrfs_device
*add_missing_dev(struct btrfs_root
*root
,
6141 struct btrfs_fs_devices
*fs_devices
,
6142 u64 devid
, u8
*dev_uuid
)
6144 struct btrfs_device
*device
;
6146 device
= btrfs_alloc_device(NULL
, &devid
, dev_uuid
);
6150 list_add(&device
->dev_list
, &fs_devices
->devices
);
6151 device
->fs_devices
= fs_devices
;
6152 fs_devices
->num_devices
++;
6154 device
->missing
= 1;
6155 fs_devices
->missing_devices
++;
6161 * btrfs_alloc_device - allocate struct btrfs_device
6162 * @fs_info: used only for generating a new devid, can be NULL if
6163 * devid is provided (i.e. @devid != NULL).
6164 * @devid: a pointer to devid for this device. If NULL a new devid
6166 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6169 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6170 * on error. Returned struct is not linked onto any lists and can be
6171 * destroyed with kfree() right away.
6173 struct btrfs_device
*btrfs_alloc_device(struct btrfs_fs_info
*fs_info
,
6177 struct btrfs_device
*dev
;
6180 if (WARN_ON(!devid
&& !fs_info
))
6181 return ERR_PTR(-EINVAL
);
6183 dev
= __alloc_device();
6192 ret
= find_next_devid(fs_info
, &tmp
);
6195 return ERR_PTR(ret
);
6201 memcpy(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
);
6203 generate_random_uuid(dev
->uuid
);
6205 btrfs_init_work(&dev
->work
, btrfs_submit_helper
,
6206 pending_bios_fn
, NULL
, NULL
);
6211 static int read_one_chunk(struct btrfs_root
*root
, struct btrfs_key
*key
,
6212 struct extent_buffer
*leaf
,
6213 struct btrfs_chunk
*chunk
)
6215 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
6216 struct map_lookup
*map
;
6217 struct extent_map
*em
;
6221 u8 uuid
[BTRFS_UUID_SIZE
];
6226 logical
= key
->offset
;
6227 length
= btrfs_chunk_length(leaf
, chunk
);
6229 read_lock(&map_tree
->map_tree
.lock
);
6230 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
6231 read_unlock(&map_tree
->map_tree
.lock
);
6233 /* already mapped? */
6234 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
6235 free_extent_map(em
);
6238 free_extent_map(em
);
6241 em
= alloc_extent_map();
6244 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6245 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
6247 free_extent_map(em
);
6251 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
6252 em
->bdev
= (struct block_device
*)map
;
6253 em
->start
= logical
;
6256 em
->block_start
= 0;
6257 em
->block_len
= em
->len
;
6259 map
->num_stripes
= num_stripes
;
6260 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
6261 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
6262 map
->sector_size
= btrfs_chunk_sector_size(leaf
, chunk
);
6263 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6264 map
->type
= btrfs_chunk_type(leaf
, chunk
);
6265 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6266 for (i
= 0; i
< num_stripes
; i
++) {
6267 map
->stripes
[i
].physical
=
6268 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
6269 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
6270 read_extent_buffer(leaf
, uuid
, (unsigned long)
6271 btrfs_stripe_dev_uuid_nr(chunk
, i
),
6273 map
->stripes
[i
].dev
= btrfs_find_device(root
->fs_info
, devid
,
6275 if (!map
->stripes
[i
].dev
&& !btrfs_test_opt(root
, DEGRADED
)) {
6276 free_extent_map(em
);
6279 if (!map
->stripes
[i
].dev
) {
6280 map
->stripes
[i
].dev
=
6281 add_missing_dev(root
, root
->fs_info
->fs_devices
,
6283 if (!map
->stripes
[i
].dev
) {
6284 free_extent_map(em
);
6287 btrfs_warn(root
->fs_info
, "devid %llu uuid %pU is missing",
6290 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
6293 write_lock(&map_tree
->map_tree
.lock
);
6294 ret
= add_extent_mapping(&map_tree
->map_tree
, em
, 0);
6295 write_unlock(&map_tree
->map_tree
.lock
);
6296 BUG_ON(ret
); /* Tree corruption */
6297 free_extent_map(em
);
6302 static void fill_device_from_item(struct extent_buffer
*leaf
,
6303 struct btrfs_dev_item
*dev_item
,
6304 struct btrfs_device
*device
)
6308 device
->devid
= btrfs_device_id(leaf
, dev_item
);
6309 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
6310 device
->total_bytes
= device
->disk_total_bytes
;
6311 device
->commit_total_bytes
= device
->disk_total_bytes
;
6312 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
6313 device
->commit_bytes_used
= device
->bytes_used
;
6314 device
->type
= btrfs_device_type(leaf
, dev_item
);
6315 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
6316 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
6317 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
6318 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
6319 device
->is_tgtdev_for_dev_replace
= 0;
6321 ptr
= btrfs_device_uuid(dev_item
);
6322 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
6325 static struct btrfs_fs_devices
*open_seed_devices(struct btrfs_root
*root
,
6328 struct btrfs_fs_devices
*fs_devices
;
6331 BUG_ON(!mutex_is_locked(&uuid_mutex
));
6333 fs_devices
= root
->fs_info
->fs_devices
->seed
;
6334 while (fs_devices
) {
6335 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
))
6338 fs_devices
= fs_devices
->seed
;
6341 fs_devices
= find_fsid(fsid
);
6343 if (!btrfs_test_opt(root
, DEGRADED
))
6344 return ERR_PTR(-ENOENT
);
6346 fs_devices
= alloc_fs_devices(fsid
);
6347 if (IS_ERR(fs_devices
))
6350 fs_devices
->seeding
= 1;
6351 fs_devices
->opened
= 1;
6355 fs_devices
= clone_fs_devices(fs_devices
);
6356 if (IS_ERR(fs_devices
))
6359 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
6360 root
->fs_info
->bdev_holder
);
6362 free_fs_devices(fs_devices
);
6363 fs_devices
= ERR_PTR(ret
);
6367 if (!fs_devices
->seeding
) {
6368 __btrfs_close_devices(fs_devices
);
6369 free_fs_devices(fs_devices
);
6370 fs_devices
= ERR_PTR(-EINVAL
);
6374 fs_devices
->seed
= root
->fs_info
->fs_devices
->seed
;
6375 root
->fs_info
->fs_devices
->seed
= fs_devices
;
6380 static int read_one_dev(struct btrfs_root
*root
,
6381 struct extent_buffer
*leaf
,
6382 struct btrfs_dev_item
*dev_item
)
6384 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
6385 struct btrfs_device
*device
;
6388 u8 fs_uuid
[BTRFS_UUID_SIZE
];
6389 u8 dev_uuid
[BTRFS_UUID_SIZE
];
6391 devid
= btrfs_device_id(leaf
, dev_item
);
6392 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
6394 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
6397 if (memcmp(fs_uuid
, root
->fs_info
->fsid
, BTRFS_UUID_SIZE
)) {
6398 fs_devices
= open_seed_devices(root
, fs_uuid
);
6399 if (IS_ERR(fs_devices
))
6400 return PTR_ERR(fs_devices
);
6403 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
, fs_uuid
);
6405 if (!btrfs_test_opt(root
, DEGRADED
))
6408 device
= add_missing_dev(root
, fs_devices
, devid
, dev_uuid
);
6411 btrfs_warn(root
->fs_info
, "devid %llu uuid %pU missing",
6414 if (!device
->bdev
&& !btrfs_test_opt(root
, DEGRADED
))
6417 if(!device
->bdev
&& !device
->missing
) {
6419 * this happens when a device that was properly setup
6420 * in the device info lists suddenly goes bad.
6421 * device->bdev is NULL, and so we have to set
6422 * device->missing to one here
6424 device
->fs_devices
->missing_devices
++;
6425 device
->missing
= 1;
6428 /* Move the device to its own fs_devices */
6429 if (device
->fs_devices
!= fs_devices
) {
6430 ASSERT(device
->missing
);
6432 list_move(&device
->dev_list
, &fs_devices
->devices
);
6433 device
->fs_devices
->num_devices
--;
6434 fs_devices
->num_devices
++;
6436 device
->fs_devices
->missing_devices
--;
6437 fs_devices
->missing_devices
++;
6439 device
->fs_devices
= fs_devices
;
6443 if (device
->fs_devices
!= root
->fs_info
->fs_devices
) {
6444 BUG_ON(device
->writeable
);
6445 if (device
->generation
!=
6446 btrfs_device_generation(leaf
, dev_item
))
6450 fill_device_from_item(leaf
, dev_item
, device
);
6451 device
->in_fs_metadata
= 1;
6452 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
6453 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
6454 spin_lock(&root
->fs_info
->free_chunk_lock
);
6455 root
->fs_info
->free_chunk_space
+= device
->total_bytes
-
6457 spin_unlock(&root
->fs_info
->free_chunk_lock
);
6463 int btrfs_read_sys_array(struct btrfs_root
*root
)
6465 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
6466 struct extent_buffer
*sb
;
6467 struct btrfs_disk_key
*disk_key
;
6468 struct btrfs_chunk
*chunk
;
6470 unsigned long sb_array_offset
;
6476 struct btrfs_key key
;
6478 ASSERT(BTRFS_SUPER_INFO_SIZE
<= root
->nodesize
);
6480 * This will create extent buffer of nodesize, superblock size is
6481 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6482 * overallocate but we can keep it as-is, only the first page is used.
6484 sb
= btrfs_find_create_tree_block(root
, BTRFS_SUPER_INFO_OFFSET
);
6487 btrfs_set_buffer_uptodate(sb
);
6488 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
6490 * The sb extent buffer is artifical and just used to read the system array.
6491 * btrfs_set_buffer_uptodate() call does not properly mark all it's
6492 * pages up-to-date when the page is larger: extent does not cover the
6493 * whole page and consequently check_page_uptodate does not find all
6494 * the page's extents up-to-date (the hole beyond sb),
6495 * write_extent_buffer then triggers a WARN_ON.
6497 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6498 * but sb spans only this function. Add an explicit SetPageUptodate call
6499 * to silence the warning eg. on PowerPC 64.
6501 if (PAGE_CACHE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
6502 SetPageUptodate(sb
->pages
[0]);
6504 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
6505 array_size
= btrfs_super_sys_array_size(super_copy
);
6507 array_ptr
= super_copy
->sys_chunk_array
;
6508 sb_array_offset
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
6511 while (cur_offset
< array_size
) {
6512 disk_key
= (struct btrfs_disk_key
*)array_ptr
;
6513 len
= sizeof(*disk_key
);
6514 if (cur_offset
+ len
> array_size
)
6515 goto out_short_read
;
6517 btrfs_disk_key_to_cpu(&key
, disk_key
);
6520 sb_array_offset
+= len
;
6523 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6524 chunk
= (struct btrfs_chunk
*)sb_array_offset
;
6526 * At least one btrfs_chunk with one stripe must be
6527 * present, exact stripe count check comes afterwards
6529 len
= btrfs_chunk_item_size(1);
6530 if (cur_offset
+ len
> array_size
)
6531 goto out_short_read
;
6533 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
6536 "BTRFS: invalid number of stripes %u in sys_array at offset %u\n",
6537 num_stripes
, cur_offset
);
6542 len
= btrfs_chunk_item_size(num_stripes
);
6543 if (cur_offset
+ len
> array_size
)
6544 goto out_short_read
;
6546 ret
= read_one_chunk(root
, &key
, sb
, chunk
);
6554 sb_array_offset
+= len
;
6557 free_extent_buffer(sb
);
6561 printk(KERN_ERR
"BTRFS: sys_array too short to read %u bytes at offset %u\n",
6563 free_extent_buffer(sb
);
6567 int btrfs_read_chunk_tree(struct btrfs_root
*root
)
6569 struct btrfs_path
*path
;
6570 struct extent_buffer
*leaf
;
6571 struct btrfs_key key
;
6572 struct btrfs_key found_key
;
6576 root
= root
->fs_info
->chunk_root
;
6578 path
= btrfs_alloc_path();
6582 mutex_lock(&uuid_mutex
);
6586 * Read all device items, and then all the chunk items. All
6587 * device items are found before any chunk item (their object id
6588 * is smaller than the lowest possible object id for a chunk
6589 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6591 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
6594 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6598 leaf
= path
->nodes
[0];
6599 slot
= path
->slots
[0];
6600 if (slot
>= btrfs_header_nritems(leaf
)) {
6601 ret
= btrfs_next_leaf(root
, path
);
6608 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6609 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
6610 struct btrfs_dev_item
*dev_item
;
6611 dev_item
= btrfs_item_ptr(leaf
, slot
,
6612 struct btrfs_dev_item
);
6613 ret
= read_one_dev(root
, leaf
, dev_item
);
6616 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6617 struct btrfs_chunk
*chunk
;
6618 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
6619 ret
= read_one_chunk(root
, &found_key
, leaf
, chunk
);
6627 unlock_chunks(root
);
6628 mutex_unlock(&uuid_mutex
);
6630 btrfs_free_path(path
);
6634 void btrfs_init_devices_late(struct btrfs_fs_info
*fs_info
)
6636 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6637 struct btrfs_device
*device
;
6639 while (fs_devices
) {
6640 mutex_lock(&fs_devices
->device_list_mutex
);
6641 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
)
6642 device
->dev_root
= fs_info
->dev_root
;
6643 mutex_unlock(&fs_devices
->device_list_mutex
);
6645 fs_devices
= fs_devices
->seed
;
6649 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
6653 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6654 btrfs_dev_stat_reset(dev
, i
);
6657 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
6659 struct btrfs_key key
;
6660 struct btrfs_key found_key
;
6661 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6662 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6663 struct extent_buffer
*eb
;
6666 struct btrfs_device
*device
;
6667 struct btrfs_path
*path
= NULL
;
6670 path
= btrfs_alloc_path();
6676 mutex_lock(&fs_devices
->device_list_mutex
);
6677 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6679 struct btrfs_dev_stats_item
*ptr
;
6682 key
.type
= BTRFS_DEV_STATS_KEY
;
6683 key
.offset
= device
->devid
;
6684 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
6686 __btrfs_reset_dev_stats(device
);
6687 device
->dev_stats_valid
= 1;
6688 btrfs_release_path(path
);
6691 slot
= path
->slots
[0];
6692 eb
= path
->nodes
[0];
6693 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
6694 item_size
= btrfs_item_size_nr(eb
, slot
);
6696 ptr
= btrfs_item_ptr(eb
, slot
,
6697 struct btrfs_dev_stats_item
);
6699 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6700 if (item_size
>= (1 + i
) * sizeof(__le64
))
6701 btrfs_dev_stat_set(device
, i
,
6702 btrfs_dev_stats_value(eb
, ptr
, i
));
6704 btrfs_dev_stat_reset(device
, i
);
6707 device
->dev_stats_valid
= 1;
6708 btrfs_dev_stat_print_on_load(device
);
6709 btrfs_release_path(path
);
6711 mutex_unlock(&fs_devices
->device_list_mutex
);
6714 btrfs_free_path(path
);
6715 return ret
< 0 ? ret
: 0;
6718 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
6719 struct btrfs_root
*dev_root
,
6720 struct btrfs_device
*device
)
6722 struct btrfs_path
*path
;
6723 struct btrfs_key key
;
6724 struct extent_buffer
*eb
;
6725 struct btrfs_dev_stats_item
*ptr
;
6730 key
.type
= BTRFS_DEV_STATS_KEY
;
6731 key
.offset
= device
->devid
;
6733 path
= btrfs_alloc_path();
6735 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
6737 btrfs_warn_in_rcu(dev_root
->fs_info
,
6738 "error %d while searching for dev_stats item for device %s",
6739 ret
, rcu_str_deref(device
->name
));
6744 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
6745 /* need to delete old one and insert a new one */
6746 ret
= btrfs_del_item(trans
, dev_root
, path
);
6748 btrfs_warn_in_rcu(dev_root
->fs_info
,
6749 "delete too small dev_stats item for device %s failed %d",
6750 rcu_str_deref(device
->name
), ret
);
6757 /* need to insert a new item */
6758 btrfs_release_path(path
);
6759 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
6760 &key
, sizeof(*ptr
));
6762 btrfs_warn_in_rcu(dev_root
->fs_info
,
6763 "insert dev_stats item for device %s failed %d",
6764 rcu_str_deref(device
->name
), ret
);
6769 eb
= path
->nodes
[0];
6770 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
6771 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6772 btrfs_set_dev_stats_value(eb
, ptr
, i
,
6773 btrfs_dev_stat_read(device
, i
));
6774 btrfs_mark_buffer_dirty(eb
);
6777 btrfs_free_path(path
);
6782 * called from commit_transaction. Writes all changed device stats to disk.
6784 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
6785 struct btrfs_fs_info
*fs_info
)
6787 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6788 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6789 struct btrfs_device
*device
;
6793 mutex_lock(&fs_devices
->device_list_mutex
);
6794 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6795 if (!device
->dev_stats_valid
|| !btrfs_dev_stats_dirty(device
))
6798 stats_cnt
= atomic_read(&device
->dev_stats_ccnt
);
6799 ret
= update_dev_stat_item(trans
, dev_root
, device
);
6801 atomic_sub(stats_cnt
, &device
->dev_stats_ccnt
);
6803 mutex_unlock(&fs_devices
->device_list_mutex
);
6808 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
6810 btrfs_dev_stat_inc(dev
, index
);
6811 btrfs_dev_stat_print_on_error(dev
);
6814 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
6816 if (!dev
->dev_stats_valid
)
6818 btrfs_err_rl_in_rcu(dev
->dev_root
->fs_info
,
6819 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6820 rcu_str_deref(dev
->name
),
6821 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
6822 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
6823 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
6824 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
6825 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
6828 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
6832 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6833 if (btrfs_dev_stat_read(dev
, i
) != 0)
6835 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
6836 return; /* all values == 0, suppress message */
6838 btrfs_info_in_rcu(dev
->dev_root
->fs_info
,
6839 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
6840 rcu_str_deref(dev
->name
),
6841 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
6842 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
6843 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
6844 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
6845 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
6848 int btrfs_get_dev_stats(struct btrfs_root
*root
,
6849 struct btrfs_ioctl_get_dev_stats
*stats
)
6851 struct btrfs_device
*dev
;
6852 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
6855 mutex_lock(&fs_devices
->device_list_mutex
);
6856 dev
= btrfs_find_device(root
->fs_info
, stats
->devid
, NULL
, NULL
);
6857 mutex_unlock(&fs_devices
->device_list_mutex
);
6860 btrfs_warn(root
->fs_info
, "get dev_stats failed, device not found");
6862 } else if (!dev
->dev_stats_valid
) {
6863 btrfs_warn(root
->fs_info
, "get dev_stats failed, not yet valid");
6865 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
6866 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6867 if (stats
->nr_items
> i
)
6869 btrfs_dev_stat_read_and_reset(dev
, i
);
6871 btrfs_dev_stat_reset(dev
, i
);
6874 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6875 if (stats
->nr_items
> i
)
6876 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
6878 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
6879 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
6883 void btrfs_scratch_superblocks(struct block_device
*bdev
, char *device_path
)
6885 struct buffer_head
*bh
;
6886 struct btrfs_super_block
*disk_super
;
6892 for (copy_num
= 0; copy_num
< BTRFS_SUPER_MIRROR_MAX
;
6895 if (btrfs_read_dev_one_super(bdev
, copy_num
, &bh
))
6898 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
6900 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
6901 set_buffer_dirty(bh
);
6902 sync_dirty_buffer(bh
);
6906 /* Notify udev that device has changed */
6907 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
6909 /* Update ctime/mtime for device path for libblkid */
6910 update_dev_time(device_path
);
6914 * Update the size of all devices, which is used for writing out the
6917 void btrfs_update_commit_device_size(struct btrfs_fs_info
*fs_info
)
6919 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6920 struct btrfs_device
*curr
, *next
;
6922 if (list_empty(&fs_devices
->resized_devices
))
6925 mutex_lock(&fs_devices
->device_list_mutex
);
6926 lock_chunks(fs_info
->dev_root
);
6927 list_for_each_entry_safe(curr
, next
, &fs_devices
->resized_devices
,
6929 list_del_init(&curr
->resized_list
);
6930 curr
->commit_total_bytes
= curr
->disk_total_bytes
;
6932 unlock_chunks(fs_info
->dev_root
);
6933 mutex_unlock(&fs_devices
->device_list_mutex
);
6936 /* Must be invoked during the transaction commit */
6937 void btrfs_update_commit_device_bytes_used(struct btrfs_root
*root
,
6938 struct btrfs_transaction
*transaction
)
6940 struct extent_map
*em
;
6941 struct map_lookup
*map
;
6942 struct btrfs_device
*dev
;
6945 if (list_empty(&transaction
->pending_chunks
))
6948 /* In order to kick the device replace finish process */
6950 list_for_each_entry(em
, &transaction
->pending_chunks
, list
) {
6951 map
= (struct map_lookup
*)em
->bdev
;
6953 for (i
= 0; i
< map
->num_stripes
; i
++) {
6954 dev
= map
->stripes
[i
].dev
;
6955 dev
->commit_bytes_used
= dev
->bytes_used
;
6958 unlock_chunks(root
);
6961 void btrfs_set_fs_info_ptr(struct btrfs_fs_info
*fs_info
)
6963 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6964 while (fs_devices
) {
6965 fs_devices
->fs_info
= fs_info
;
6966 fs_devices
= fs_devices
->seed
;
6970 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info
*fs_info
)
6972 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6973 while (fs_devices
) {
6974 fs_devices
->fs_info
= NULL
;
6975 fs_devices
= fs_devices
->seed
;
6979 void btrfs_close_one_device(struct btrfs_device
*device
)
6981 struct btrfs_fs_devices
*fs_devices
= device
->fs_devices
;
6982 struct btrfs_device
*new_device
;
6983 struct rcu_string
*name
;
6986 fs_devices
->open_devices
--;
6988 if (device
->writeable
&&
6989 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
6990 list_del_init(&device
->dev_alloc_list
);
6991 fs_devices
->rw_devices
--;
6994 if (device
->missing
)
6995 fs_devices
->missing_devices
--;
6997 new_device
= btrfs_alloc_device(NULL
, &device
->devid
,
6999 BUG_ON(IS_ERR(new_device
)); /* -ENOMEM */
7001 /* Safe because we are under uuid_mutex */
7003 name
= rcu_string_strdup(device
->name
->str
, GFP_NOFS
);
7004 BUG_ON(!name
); /* -ENOMEM */
7005 rcu_assign_pointer(new_device
->name
, name
);
7008 list_replace_rcu(&device
->dev_list
, &new_device
->dev_list
);
7009 new_device
->fs_devices
= device
->fs_devices
;
7011 call_rcu(&device
->rcu
, free_device
);