2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/iocontext.h>
24 #include <linux/capability.h>
25 #include <linux/ratelimit.h>
26 #include <linux/kthread.h>
27 #include <linux/raid/pq.h>
28 #include <linux/semaphore.h>
29 #include <linux/uuid.h>
30 #include <asm/div64.h>
32 #include "extent_map.h"
34 #include "transaction.h"
35 #include "print-tree.h"
38 #include "async-thread.h"
39 #include "check-integrity.h"
40 #include "rcu-string.h"
42 #include "dev-replace.h"
45 const struct btrfs_raid_attr btrfs_raid_array
[BTRFS_NR_RAID_TYPES
] = {
46 [BTRFS_RAID_RAID10
] = {
49 .devs_max
= 0, /* 0 == as many as possible */
51 .tolerated_failures
= 1,
55 [BTRFS_RAID_RAID1
] = {
60 .tolerated_failures
= 1,
69 .tolerated_failures
= 0,
73 [BTRFS_RAID_RAID0
] = {
78 .tolerated_failures
= 0,
82 [BTRFS_RAID_SINGLE
] = {
87 .tolerated_failures
= 0,
91 [BTRFS_RAID_RAID5
] = {
96 .tolerated_failures
= 1,
100 [BTRFS_RAID_RAID6
] = {
105 .tolerated_failures
= 2,
111 const u64 btrfs_raid_group
[BTRFS_NR_RAID_TYPES
] = {
112 [BTRFS_RAID_RAID10
] = BTRFS_BLOCK_GROUP_RAID10
,
113 [BTRFS_RAID_RAID1
] = BTRFS_BLOCK_GROUP_RAID1
,
114 [BTRFS_RAID_DUP
] = BTRFS_BLOCK_GROUP_DUP
,
115 [BTRFS_RAID_RAID0
] = BTRFS_BLOCK_GROUP_RAID0
,
116 [BTRFS_RAID_SINGLE
] = 0,
117 [BTRFS_RAID_RAID5
] = BTRFS_BLOCK_GROUP_RAID5
,
118 [BTRFS_RAID_RAID6
] = BTRFS_BLOCK_GROUP_RAID6
,
122 * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
123 * condition is not met. Zero means there's no corresponding
124 * BTRFS_ERROR_DEV_*_NOT_MET value.
126 const int btrfs_raid_mindev_error
[BTRFS_NR_RAID_TYPES
] = {
127 [BTRFS_RAID_RAID10
] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET
,
128 [BTRFS_RAID_RAID1
] = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET
,
129 [BTRFS_RAID_DUP
] = 0,
130 [BTRFS_RAID_RAID0
] = 0,
131 [BTRFS_RAID_SINGLE
] = 0,
132 [BTRFS_RAID_RAID5
] = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET
,
133 [BTRFS_RAID_RAID6
] = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET
,
136 static int init_first_rw_device(struct btrfs_trans_handle
*trans
,
137 struct btrfs_fs_info
*fs_info
);
138 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info
*fs_info
);
139 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
);
140 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
);
141 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*device
);
143 DEFINE_MUTEX(uuid_mutex
);
144 static LIST_HEAD(fs_uuids
);
145 struct list_head
*btrfs_get_fs_uuids(void)
150 static struct btrfs_fs_devices
*__alloc_fs_devices(void)
152 struct btrfs_fs_devices
*fs_devs
;
154 fs_devs
= kzalloc(sizeof(*fs_devs
), GFP_KERNEL
);
156 return ERR_PTR(-ENOMEM
);
158 mutex_init(&fs_devs
->device_list_mutex
);
160 INIT_LIST_HEAD(&fs_devs
->devices
);
161 INIT_LIST_HEAD(&fs_devs
->resized_devices
);
162 INIT_LIST_HEAD(&fs_devs
->alloc_list
);
163 INIT_LIST_HEAD(&fs_devs
->list
);
169 * alloc_fs_devices - allocate struct btrfs_fs_devices
170 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
173 * Return: a pointer to a new &struct btrfs_fs_devices on success;
174 * ERR_PTR() on error. Returned struct is not linked onto any lists and
175 * can be destroyed with kfree() right away.
177 static struct btrfs_fs_devices
*alloc_fs_devices(const u8
*fsid
)
179 struct btrfs_fs_devices
*fs_devs
;
181 fs_devs
= __alloc_fs_devices();
186 memcpy(fs_devs
->fsid
, fsid
, BTRFS_FSID_SIZE
);
188 generate_random_uuid(fs_devs
->fsid
);
193 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
195 struct btrfs_device
*device
;
196 WARN_ON(fs_devices
->opened
);
197 while (!list_empty(&fs_devices
->devices
)) {
198 device
= list_entry(fs_devices
->devices
.next
,
199 struct btrfs_device
, dev_list
);
200 list_del(&device
->dev_list
);
201 rcu_string_free(device
->name
);
207 static void btrfs_kobject_uevent(struct block_device
*bdev
,
208 enum kobject_action action
)
212 ret
= kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, action
);
214 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
216 kobject_name(&disk_to_dev(bdev
->bd_disk
)->kobj
),
217 &disk_to_dev(bdev
->bd_disk
)->kobj
);
220 void btrfs_cleanup_fs_uuids(void)
222 struct btrfs_fs_devices
*fs_devices
;
224 while (!list_empty(&fs_uuids
)) {
225 fs_devices
= list_entry(fs_uuids
.next
,
226 struct btrfs_fs_devices
, list
);
227 list_del(&fs_devices
->list
);
228 free_fs_devices(fs_devices
);
232 static struct btrfs_device
*__alloc_device(void)
234 struct btrfs_device
*dev
;
236 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
238 return ERR_PTR(-ENOMEM
);
240 INIT_LIST_HEAD(&dev
->dev_list
);
241 INIT_LIST_HEAD(&dev
->dev_alloc_list
);
242 INIT_LIST_HEAD(&dev
->resized_list
);
244 spin_lock_init(&dev
->io_lock
);
246 spin_lock_init(&dev
->reada_lock
);
247 atomic_set(&dev
->reada_in_flight
, 0);
248 atomic_set(&dev
->dev_stats_ccnt
, 0);
249 btrfs_device_data_ordered_init(dev
);
250 INIT_RADIX_TREE(&dev
->reada_zones
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
251 INIT_RADIX_TREE(&dev
->reada_extents
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
256 static noinline
struct btrfs_device
*__find_device(struct list_head
*head
,
259 struct btrfs_device
*dev
;
261 list_for_each_entry(dev
, head
, dev_list
) {
262 if (dev
->devid
== devid
&&
263 (!uuid
|| !memcmp(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
))) {
270 static noinline
struct btrfs_fs_devices
*find_fsid(u8
*fsid
)
272 struct btrfs_fs_devices
*fs_devices
;
274 list_for_each_entry(fs_devices
, &fs_uuids
, list
) {
275 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
282 btrfs_get_bdev_and_sb(const char *device_path
, fmode_t flags
, void *holder
,
283 int flush
, struct block_device
**bdev
,
284 struct buffer_head
**bh
)
288 *bdev
= blkdev_get_by_path(device_path
, flags
, holder
);
291 ret
= PTR_ERR(*bdev
);
296 filemap_write_and_wait((*bdev
)->bd_inode
->i_mapping
);
297 ret
= set_blocksize(*bdev
, 4096);
299 blkdev_put(*bdev
, flags
);
302 invalidate_bdev(*bdev
);
303 *bh
= btrfs_read_dev_super(*bdev
);
306 blkdev_put(*bdev
, flags
);
318 static void requeue_list(struct btrfs_pending_bios
*pending_bios
,
319 struct bio
*head
, struct bio
*tail
)
322 struct bio
*old_head
;
324 old_head
= pending_bios
->head
;
325 pending_bios
->head
= head
;
326 if (pending_bios
->tail
)
327 tail
->bi_next
= old_head
;
329 pending_bios
->tail
= tail
;
333 * we try to collect pending bios for a device so we don't get a large
334 * number of procs sending bios down to the same device. This greatly
335 * improves the schedulers ability to collect and merge the bios.
337 * But, it also turns into a long list of bios to process and that is sure
338 * to eventually make the worker thread block. The solution here is to
339 * make some progress and then put this work struct back at the end of
340 * the list if the block device is congested. This way, multiple devices
341 * can make progress from a single worker thread.
343 static noinline
void run_scheduled_bios(struct btrfs_device
*device
)
345 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
347 struct backing_dev_info
*bdi
;
348 struct btrfs_pending_bios
*pending_bios
;
352 unsigned long num_run
;
353 unsigned long batch_run
= 0;
355 unsigned long last_waited
= 0;
357 int sync_pending
= 0;
358 struct blk_plug plug
;
361 * this function runs all the bios we've collected for
362 * a particular device. We don't want to wander off to
363 * another device without first sending all of these down.
364 * So, setup a plug here and finish it off before we return
366 blk_start_plug(&plug
);
368 bdi
= device
->bdev
->bd_bdi
;
369 limit
= btrfs_async_submit_limit(fs_info
);
370 limit
= limit
* 2 / 3;
373 spin_lock(&device
->io_lock
);
378 /* take all the bios off the list at once and process them
379 * later on (without the lock held). But, remember the
380 * tail and other pointers so the bios can be properly reinserted
381 * into the list if we hit congestion
383 if (!force_reg
&& device
->pending_sync_bios
.head
) {
384 pending_bios
= &device
->pending_sync_bios
;
387 pending_bios
= &device
->pending_bios
;
391 pending
= pending_bios
->head
;
392 tail
= pending_bios
->tail
;
393 WARN_ON(pending
&& !tail
);
396 * if pending was null this time around, no bios need processing
397 * at all and we can stop. Otherwise it'll loop back up again
398 * and do an additional check so no bios are missed.
400 * device->running_pending is used to synchronize with the
403 if (device
->pending_sync_bios
.head
== NULL
&&
404 device
->pending_bios
.head
== NULL
) {
406 device
->running_pending
= 0;
409 device
->running_pending
= 1;
412 pending_bios
->head
= NULL
;
413 pending_bios
->tail
= NULL
;
415 spin_unlock(&device
->io_lock
);
420 /* we want to work on both lists, but do more bios on the
421 * sync list than the regular list
424 pending_bios
!= &device
->pending_sync_bios
&&
425 device
->pending_sync_bios
.head
) ||
426 (num_run
> 64 && pending_bios
== &device
->pending_sync_bios
&&
427 device
->pending_bios
.head
)) {
428 spin_lock(&device
->io_lock
);
429 requeue_list(pending_bios
, pending
, tail
);
434 pending
= pending
->bi_next
;
438 * atomic_dec_return implies a barrier for waitqueue_active
440 if (atomic_dec_return(&fs_info
->nr_async_bios
) < limit
&&
441 waitqueue_active(&fs_info
->async_submit_wait
))
442 wake_up(&fs_info
->async_submit_wait
);
444 BUG_ON(atomic_read(&cur
->__bi_cnt
) == 0);
447 * if we're doing the sync list, record that our
448 * plug has some sync requests on it
450 * If we're doing the regular list and there are
451 * sync requests sitting around, unplug before
454 if (pending_bios
== &device
->pending_sync_bios
) {
456 } else if (sync_pending
) {
457 blk_finish_plug(&plug
);
458 blk_start_plug(&plug
);
462 btrfsic_submit_bio(cur
);
469 * we made progress, there is more work to do and the bdi
470 * is now congested. Back off and let other work structs
473 if (pending
&& bdi_write_congested(bdi
) && batch_run
> 8 &&
474 fs_info
->fs_devices
->open_devices
> 1) {
475 struct io_context
*ioc
;
477 ioc
= current
->io_context
;
480 * the main goal here is that we don't want to
481 * block if we're going to be able to submit
482 * more requests without blocking.
484 * This code does two great things, it pokes into
485 * the elevator code from a filesystem _and_
486 * it makes assumptions about how batching works.
488 if (ioc
&& ioc
->nr_batch_requests
> 0 &&
489 time_before(jiffies
, ioc
->last_waited
+ HZ
/50UL) &&
491 ioc
->last_waited
== last_waited
)) {
493 * we want to go through our batch of
494 * requests and stop. So, we copy out
495 * the ioc->last_waited time and test
496 * against it before looping
498 last_waited
= ioc
->last_waited
;
502 spin_lock(&device
->io_lock
);
503 requeue_list(pending_bios
, pending
, tail
);
504 device
->running_pending
= 1;
506 spin_unlock(&device
->io_lock
);
507 btrfs_queue_work(fs_info
->submit_workers
,
511 /* unplug every 64 requests just for good measure */
512 if (batch_run
% 64 == 0) {
513 blk_finish_plug(&plug
);
514 blk_start_plug(&plug
);
523 spin_lock(&device
->io_lock
);
524 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
526 spin_unlock(&device
->io_lock
);
529 blk_finish_plug(&plug
);
532 static void pending_bios_fn(struct btrfs_work
*work
)
534 struct btrfs_device
*device
;
536 device
= container_of(work
, struct btrfs_device
, work
);
537 run_scheduled_bios(device
);
541 void btrfs_free_stale_device(struct btrfs_device
*cur_dev
)
543 struct btrfs_fs_devices
*fs_devs
;
544 struct btrfs_device
*dev
;
549 list_for_each_entry(fs_devs
, &fs_uuids
, list
) {
554 if (fs_devs
->seeding
)
557 list_for_each_entry(dev
, &fs_devs
->devices
, dev_list
) {
565 * Todo: This won't be enough. What if the same device
566 * comes back (with new uuid and) with its mapper path?
567 * But for now, this does help as mostly an admin will
568 * either use mapper or non mapper path throughout.
571 del
= strcmp(rcu_str_deref(dev
->name
),
572 rcu_str_deref(cur_dev
->name
));
579 /* delete the stale device */
580 if (fs_devs
->num_devices
== 1) {
581 btrfs_sysfs_remove_fsid(fs_devs
);
582 list_del(&fs_devs
->list
);
583 free_fs_devices(fs_devs
);
585 fs_devs
->num_devices
--;
586 list_del(&dev
->dev_list
);
587 rcu_string_free(dev
->name
);
596 * Add new device to list of registered devices
599 * 1 - first time device is seen
600 * 0 - device already known
603 static noinline
int device_list_add(const char *path
,
604 struct btrfs_super_block
*disk_super
,
605 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
607 struct btrfs_device
*device
;
608 struct btrfs_fs_devices
*fs_devices
;
609 struct rcu_string
*name
;
611 u64 found_transid
= btrfs_super_generation(disk_super
);
613 fs_devices
= find_fsid(disk_super
->fsid
);
615 fs_devices
= alloc_fs_devices(disk_super
->fsid
);
616 if (IS_ERR(fs_devices
))
617 return PTR_ERR(fs_devices
);
619 list_add(&fs_devices
->list
, &fs_uuids
);
623 device
= __find_device(&fs_devices
->devices
, devid
,
624 disk_super
->dev_item
.uuid
);
628 if (fs_devices
->opened
)
631 device
= btrfs_alloc_device(NULL
, &devid
,
632 disk_super
->dev_item
.uuid
);
633 if (IS_ERR(device
)) {
634 /* we can safely leave the fs_devices entry around */
635 return PTR_ERR(device
);
638 name
= rcu_string_strdup(path
, GFP_NOFS
);
643 rcu_assign_pointer(device
->name
, name
);
645 mutex_lock(&fs_devices
->device_list_mutex
);
646 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
647 fs_devices
->num_devices
++;
648 mutex_unlock(&fs_devices
->device_list_mutex
);
651 device
->fs_devices
= fs_devices
;
652 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
654 * When FS is already mounted.
655 * 1. If you are here and if the device->name is NULL that
656 * means this device was missing at time of FS mount.
657 * 2. If you are here and if the device->name is different
658 * from 'path' that means either
659 * a. The same device disappeared and reappeared with
661 * b. The missing-disk-which-was-replaced, has
664 * We must allow 1 and 2a above. But 2b would be a spurious
667 * Further in case of 1 and 2a above, the disk at 'path'
668 * would have missed some transaction when it was away and
669 * in case of 2a the stale bdev has to be updated as well.
670 * 2b must not be allowed at all time.
674 * For now, we do allow update to btrfs_fs_device through the
675 * btrfs dev scan cli after FS has been mounted. We're still
676 * tracking a problem where systems fail mount by subvolume id
677 * when we reject replacement on a mounted FS.
679 if (!fs_devices
->opened
&& found_transid
< device
->generation
) {
681 * That is if the FS is _not_ mounted and if you
682 * are here, that means there is more than one
683 * disk with same uuid and devid.We keep the one
684 * with larger generation number or the last-in if
685 * generation are equal.
690 name
= rcu_string_strdup(path
, GFP_NOFS
);
693 rcu_string_free(device
->name
);
694 rcu_assign_pointer(device
->name
, name
);
695 if (device
->missing
) {
696 fs_devices
->missing_devices
--;
702 * Unmount does not free the btrfs_device struct but would zero
703 * generation along with most of the other members. So just update
704 * it back. We need it to pick the disk with largest generation
707 if (!fs_devices
->opened
)
708 device
->generation
= found_transid
;
711 * if there is new btrfs on an already registered device,
712 * then remove the stale device entry.
715 btrfs_free_stale_device(device
);
717 *fs_devices_ret
= fs_devices
;
722 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
724 struct btrfs_fs_devices
*fs_devices
;
725 struct btrfs_device
*device
;
726 struct btrfs_device
*orig_dev
;
728 fs_devices
= alloc_fs_devices(orig
->fsid
);
729 if (IS_ERR(fs_devices
))
732 mutex_lock(&orig
->device_list_mutex
);
733 fs_devices
->total_devices
= orig
->total_devices
;
735 /* We have held the volume lock, it is safe to get the devices. */
736 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
737 struct rcu_string
*name
;
739 device
= btrfs_alloc_device(NULL
, &orig_dev
->devid
,
745 * This is ok to do without rcu read locked because we hold the
746 * uuid mutex so nothing we touch in here is going to disappear.
748 if (orig_dev
->name
) {
749 name
= rcu_string_strdup(orig_dev
->name
->str
,
755 rcu_assign_pointer(device
->name
, name
);
758 list_add(&device
->dev_list
, &fs_devices
->devices
);
759 device
->fs_devices
= fs_devices
;
760 fs_devices
->num_devices
++;
762 mutex_unlock(&orig
->device_list_mutex
);
765 mutex_unlock(&orig
->device_list_mutex
);
766 free_fs_devices(fs_devices
);
767 return ERR_PTR(-ENOMEM
);
770 void btrfs_close_extra_devices(struct btrfs_fs_devices
*fs_devices
, int step
)
772 struct btrfs_device
*device
, *next
;
773 struct btrfs_device
*latest_dev
= NULL
;
775 mutex_lock(&uuid_mutex
);
777 /* This is the initialized path, it is safe to release the devices. */
778 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
779 if (device
->in_fs_metadata
) {
780 if (!device
->is_tgtdev_for_dev_replace
&&
782 device
->generation
> latest_dev
->generation
)) {
788 if (device
->devid
== BTRFS_DEV_REPLACE_DEVID
) {
790 * In the first step, keep the device which has
791 * the correct fsid and the devid that is used
792 * for the dev_replace procedure.
793 * In the second step, the dev_replace state is
794 * read from the device tree and it is known
795 * whether the procedure is really active or
796 * not, which means whether this device is
797 * used or whether it should be removed.
799 if (step
== 0 || device
->is_tgtdev_for_dev_replace
) {
804 blkdev_put(device
->bdev
, device
->mode
);
806 fs_devices
->open_devices
--;
808 if (device
->writeable
) {
809 list_del_init(&device
->dev_alloc_list
);
810 device
->writeable
= 0;
811 if (!device
->is_tgtdev_for_dev_replace
)
812 fs_devices
->rw_devices
--;
814 list_del_init(&device
->dev_list
);
815 fs_devices
->num_devices
--;
816 rcu_string_free(device
->name
);
820 if (fs_devices
->seed
) {
821 fs_devices
= fs_devices
->seed
;
825 fs_devices
->latest_bdev
= latest_dev
->bdev
;
827 mutex_unlock(&uuid_mutex
);
830 static void __free_device(struct work_struct
*work
)
832 struct btrfs_device
*device
;
834 device
= container_of(work
, struct btrfs_device
, rcu_work
);
835 rcu_string_free(device
->name
);
839 static void free_device(struct rcu_head
*head
)
841 struct btrfs_device
*device
;
843 device
= container_of(head
, struct btrfs_device
, rcu
);
845 INIT_WORK(&device
->rcu_work
, __free_device
);
846 schedule_work(&device
->rcu_work
);
849 static void btrfs_close_bdev(struct btrfs_device
*device
)
851 if (device
->bdev
&& device
->writeable
) {
852 sync_blockdev(device
->bdev
);
853 invalidate_bdev(device
->bdev
);
857 blkdev_put(device
->bdev
, device
->mode
);
860 static void btrfs_prepare_close_one_device(struct btrfs_device
*device
)
862 struct btrfs_fs_devices
*fs_devices
= device
->fs_devices
;
863 struct btrfs_device
*new_device
;
864 struct rcu_string
*name
;
867 fs_devices
->open_devices
--;
869 if (device
->writeable
&&
870 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
871 list_del_init(&device
->dev_alloc_list
);
872 fs_devices
->rw_devices
--;
876 fs_devices
->missing_devices
--;
878 new_device
= btrfs_alloc_device(NULL
, &device
->devid
,
880 BUG_ON(IS_ERR(new_device
)); /* -ENOMEM */
882 /* Safe because we are under uuid_mutex */
884 name
= rcu_string_strdup(device
->name
->str
, GFP_NOFS
);
885 BUG_ON(!name
); /* -ENOMEM */
886 rcu_assign_pointer(new_device
->name
, name
);
889 list_replace_rcu(&device
->dev_list
, &new_device
->dev_list
);
890 new_device
->fs_devices
= device
->fs_devices
;
893 static int __btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
895 struct btrfs_device
*device
, *tmp
;
896 struct list_head pending_put
;
898 INIT_LIST_HEAD(&pending_put
);
900 if (--fs_devices
->opened
> 0)
903 mutex_lock(&fs_devices
->device_list_mutex
);
904 list_for_each_entry_safe(device
, tmp
, &fs_devices
->devices
, dev_list
) {
905 btrfs_prepare_close_one_device(device
);
906 list_add(&device
->dev_list
, &pending_put
);
908 mutex_unlock(&fs_devices
->device_list_mutex
);
911 * btrfs_show_devname() is using the device_list_mutex,
912 * sometimes call to blkdev_put() leads vfs calling
913 * into this func. So do put outside of device_list_mutex,
916 while (!list_empty(&pending_put
)) {
917 device
= list_first_entry(&pending_put
,
918 struct btrfs_device
, dev_list
);
919 list_del(&device
->dev_list
);
920 btrfs_close_bdev(device
);
921 call_rcu(&device
->rcu
, free_device
);
924 WARN_ON(fs_devices
->open_devices
);
925 WARN_ON(fs_devices
->rw_devices
);
926 fs_devices
->opened
= 0;
927 fs_devices
->seeding
= 0;
932 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
934 struct btrfs_fs_devices
*seed_devices
= NULL
;
937 mutex_lock(&uuid_mutex
);
938 ret
= __btrfs_close_devices(fs_devices
);
939 if (!fs_devices
->opened
) {
940 seed_devices
= fs_devices
->seed
;
941 fs_devices
->seed
= NULL
;
943 mutex_unlock(&uuid_mutex
);
945 while (seed_devices
) {
946 fs_devices
= seed_devices
;
947 seed_devices
= fs_devices
->seed
;
948 __btrfs_close_devices(fs_devices
);
949 free_fs_devices(fs_devices
);
952 * Wait for rcu kworkers under __btrfs_close_devices
953 * to finish all blkdev_puts so device is really
954 * free when umount is done.
960 static int __btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
961 fmode_t flags
, void *holder
)
963 struct request_queue
*q
;
964 struct block_device
*bdev
;
965 struct list_head
*head
= &fs_devices
->devices
;
966 struct btrfs_device
*device
;
967 struct btrfs_device
*latest_dev
= NULL
;
968 struct buffer_head
*bh
;
969 struct btrfs_super_block
*disk_super
;
976 list_for_each_entry(device
, head
, dev_list
) {
982 /* Just open everything we can; ignore failures here */
983 if (btrfs_get_bdev_and_sb(device
->name
->str
, flags
, holder
, 1,
987 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
988 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
989 if (devid
!= device
->devid
)
992 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
,
996 device
->generation
= btrfs_super_generation(disk_super
);
998 device
->generation
> latest_dev
->generation
)
1001 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
1002 device
->writeable
= 0;
1004 device
->writeable
= !bdev_read_only(bdev
);
1008 q
= bdev_get_queue(bdev
);
1009 if (blk_queue_discard(q
))
1010 device
->can_discard
= 1;
1012 device
->bdev
= bdev
;
1013 device
->in_fs_metadata
= 0;
1014 device
->mode
= flags
;
1016 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
1017 fs_devices
->rotating
= 1;
1019 fs_devices
->open_devices
++;
1020 if (device
->writeable
&&
1021 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
1022 fs_devices
->rw_devices
++;
1023 list_add(&device
->dev_alloc_list
,
1024 &fs_devices
->alloc_list
);
1031 blkdev_put(bdev
, flags
);
1034 if (fs_devices
->open_devices
== 0) {
1038 fs_devices
->seeding
= seeding
;
1039 fs_devices
->opened
= 1;
1040 fs_devices
->latest_bdev
= latest_dev
->bdev
;
1041 fs_devices
->total_rw_bytes
= 0;
1046 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
1047 fmode_t flags
, void *holder
)
1051 mutex_lock(&uuid_mutex
);
1052 if (fs_devices
->opened
) {
1053 fs_devices
->opened
++;
1056 ret
= __btrfs_open_devices(fs_devices
, flags
, holder
);
1058 mutex_unlock(&uuid_mutex
);
1062 void btrfs_release_disk_super(struct page
*page
)
1068 int btrfs_read_disk_super(struct block_device
*bdev
, u64 bytenr
,
1069 struct page
**page
, struct btrfs_super_block
**disk_super
)
1074 /* make sure our super fits in the device */
1075 if (bytenr
+ PAGE_SIZE
>= i_size_read(bdev
->bd_inode
))
1078 /* make sure our super fits in the page */
1079 if (sizeof(**disk_super
) > PAGE_SIZE
)
1082 /* make sure our super doesn't straddle pages on disk */
1083 index
= bytenr
>> PAGE_SHIFT
;
1084 if ((bytenr
+ sizeof(**disk_super
) - 1) >> PAGE_SHIFT
!= index
)
1087 /* pull in the page with our super */
1088 *page
= read_cache_page_gfp(bdev
->bd_inode
->i_mapping
,
1091 if (IS_ERR_OR_NULL(*page
))
1096 /* align our pointer to the offset of the super block */
1097 *disk_super
= p
+ (bytenr
& ~PAGE_MASK
);
1099 if (btrfs_super_bytenr(*disk_super
) != bytenr
||
1100 btrfs_super_magic(*disk_super
) != BTRFS_MAGIC
) {
1101 btrfs_release_disk_super(*page
);
1105 if ((*disk_super
)->label
[0] &&
1106 (*disk_super
)->label
[BTRFS_LABEL_SIZE
- 1])
1107 (*disk_super
)->label
[BTRFS_LABEL_SIZE
- 1] = '\0';
1113 * Look for a btrfs signature on a device. This may be called out of the mount path
1114 * and we are not allowed to call set_blocksize during the scan. The superblock
1115 * is read via pagecache
1117 int btrfs_scan_one_device(const char *path
, fmode_t flags
, void *holder
,
1118 struct btrfs_fs_devices
**fs_devices_ret
)
1120 struct btrfs_super_block
*disk_super
;
1121 struct block_device
*bdev
;
1130 * we would like to check all the supers, but that would make
1131 * a btrfs mount succeed after a mkfs from a different FS.
1132 * So, we need to add a special mount option to scan for
1133 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1135 bytenr
= btrfs_sb_offset(0);
1136 flags
|= FMODE_EXCL
;
1137 mutex_lock(&uuid_mutex
);
1139 bdev
= blkdev_get_by_path(path
, flags
, holder
);
1141 ret
= PTR_ERR(bdev
);
1145 if (btrfs_read_disk_super(bdev
, bytenr
, &page
, &disk_super
))
1146 goto error_bdev_put
;
1148 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1149 transid
= btrfs_super_generation(disk_super
);
1150 total_devices
= btrfs_super_num_devices(disk_super
);
1152 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
1154 if (disk_super
->label
[0]) {
1155 pr_info("BTRFS: device label %s ", disk_super
->label
);
1157 pr_info("BTRFS: device fsid %pU ", disk_super
->fsid
);
1160 pr_cont("devid %llu transid %llu %s\n", devid
, transid
, path
);
1163 if (!ret
&& fs_devices_ret
)
1164 (*fs_devices_ret
)->total_devices
= total_devices
;
1166 btrfs_release_disk_super(page
);
1169 blkdev_put(bdev
, flags
);
1171 mutex_unlock(&uuid_mutex
);
1175 /* helper to account the used device space in the range */
1176 int btrfs_account_dev_extents_size(struct btrfs_device
*device
, u64 start
,
1177 u64 end
, u64
*length
)
1179 struct btrfs_key key
;
1180 struct btrfs_root
*root
= device
->fs_info
->dev_root
;
1181 struct btrfs_dev_extent
*dev_extent
;
1182 struct btrfs_path
*path
;
1186 struct extent_buffer
*l
;
1190 if (start
>= device
->total_bytes
|| device
->is_tgtdev_for_dev_replace
)
1193 path
= btrfs_alloc_path();
1196 path
->reada
= READA_FORWARD
;
1198 key
.objectid
= device
->devid
;
1200 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1202 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1206 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1213 slot
= path
->slots
[0];
1214 if (slot
>= btrfs_header_nritems(l
)) {
1215 ret
= btrfs_next_leaf(root
, path
);
1223 btrfs_item_key_to_cpu(l
, &key
, slot
);
1225 if (key
.objectid
< device
->devid
)
1228 if (key
.objectid
> device
->devid
)
1231 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1234 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1235 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1237 if (key
.offset
<= start
&& extent_end
> end
) {
1238 *length
= end
- start
+ 1;
1240 } else if (key
.offset
<= start
&& extent_end
> start
)
1241 *length
+= extent_end
- start
;
1242 else if (key
.offset
> start
&& extent_end
<= end
)
1243 *length
+= extent_end
- key
.offset
;
1244 else if (key
.offset
> start
&& key
.offset
<= end
) {
1245 *length
+= end
- key
.offset
+ 1;
1247 } else if (key
.offset
> end
)
1255 btrfs_free_path(path
);
1259 static int contains_pending_extent(struct btrfs_transaction
*transaction
,
1260 struct btrfs_device
*device
,
1261 u64
*start
, u64 len
)
1263 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1264 struct extent_map
*em
;
1265 struct list_head
*search_list
= &fs_info
->pinned_chunks
;
1267 u64 physical_start
= *start
;
1270 search_list
= &transaction
->pending_chunks
;
1272 list_for_each_entry(em
, search_list
, list
) {
1273 struct map_lookup
*map
;
1276 map
= em
->map_lookup
;
1277 for (i
= 0; i
< map
->num_stripes
; i
++) {
1280 if (map
->stripes
[i
].dev
!= device
)
1282 if (map
->stripes
[i
].physical
>= physical_start
+ len
||
1283 map
->stripes
[i
].physical
+ em
->orig_block_len
<=
1287 * Make sure that while processing the pinned list we do
1288 * not override our *start with a lower value, because
1289 * we can have pinned chunks that fall within this
1290 * device hole and that have lower physical addresses
1291 * than the pending chunks we processed before. If we
1292 * do not take this special care we can end up getting
1293 * 2 pending chunks that start at the same physical
1294 * device offsets because the end offset of a pinned
1295 * chunk can be equal to the start offset of some
1298 end
= map
->stripes
[i
].physical
+ em
->orig_block_len
;
1305 if (search_list
!= &fs_info
->pinned_chunks
) {
1306 search_list
= &fs_info
->pinned_chunks
;
1315 * find_free_dev_extent_start - find free space in the specified device
1316 * @device: the device which we search the free space in
1317 * @num_bytes: the size of the free space that we need
1318 * @search_start: the position from which to begin the search
1319 * @start: store the start of the free space.
1320 * @len: the size of the free space. that we find, or the size
1321 * of the max free space if we don't find suitable free space
1323 * this uses a pretty simple search, the expectation is that it is
1324 * called very infrequently and that a given device has a small number
1327 * @start is used to store the start of the free space if we find. But if we
1328 * don't find suitable free space, it will be used to store the start position
1329 * of the max free space.
1331 * @len is used to store the size of the free space that we find.
1332 * But if we don't find suitable free space, it is used to store the size of
1333 * the max free space.
1335 int find_free_dev_extent_start(struct btrfs_transaction
*transaction
,
1336 struct btrfs_device
*device
, u64 num_bytes
,
1337 u64 search_start
, u64
*start
, u64
*len
)
1339 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1340 struct btrfs_root
*root
= fs_info
->dev_root
;
1341 struct btrfs_key key
;
1342 struct btrfs_dev_extent
*dev_extent
;
1343 struct btrfs_path
*path
;
1348 u64 search_end
= device
->total_bytes
;
1351 struct extent_buffer
*l
;
1352 u64 min_search_start
;
1355 * We don't want to overwrite the superblock on the drive nor any area
1356 * used by the boot loader (grub for example), so we make sure to start
1357 * at an offset of at least 1MB.
1359 min_search_start
= max(fs_info
->alloc_start
, 1024ull * 1024);
1360 search_start
= max(search_start
, min_search_start
);
1362 path
= btrfs_alloc_path();
1366 max_hole_start
= search_start
;
1370 if (search_start
>= search_end
|| device
->is_tgtdev_for_dev_replace
) {
1375 path
->reada
= READA_FORWARD
;
1376 path
->search_commit_root
= 1;
1377 path
->skip_locking
= 1;
1379 key
.objectid
= device
->devid
;
1380 key
.offset
= search_start
;
1381 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1383 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1387 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1394 slot
= path
->slots
[0];
1395 if (slot
>= btrfs_header_nritems(l
)) {
1396 ret
= btrfs_next_leaf(root
, path
);
1404 btrfs_item_key_to_cpu(l
, &key
, slot
);
1406 if (key
.objectid
< device
->devid
)
1409 if (key
.objectid
> device
->devid
)
1412 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1415 if (key
.offset
> search_start
) {
1416 hole_size
= key
.offset
- search_start
;
1419 * Have to check before we set max_hole_start, otherwise
1420 * we could end up sending back this offset anyway.
1422 if (contains_pending_extent(transaction
, device
,
1425 if (key
.offset
>= search_start
) {
1426 hole_size
= key
.offset
- search_start
;
1433 if (hole_size
> max_hole_size
) {
1434 max_hole_start
= search_start
;
1435 max_hole_size
= hole_size
;
1439 * If this free space is greater than which we need,
1440 * it must be the max free space that we have found
1441 * until now, so max_hole_start must point to the start
1442 * of this free space and the length of this free space
1443 * is stored in max_hole_size. Thus, we return
1444 * max_hole_start and max_hole_size and go back to the
1447 if (hole_size
>= num_bytes
) {
1453 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1454 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1456 if (extent_end
> search_start
)
1457 search_start
= extent_end
;
1464 * At this point, search_start should be the end of
1465 * allocated dev extents, and when shrinking the device,
1466 * search_end may be smaller than search_start.
1468 if (search_end
> search_start
) {
1469 hole_size
= search_end
- search_start
;
1471 if (contains_pending_extent(transaction
, device
, &search_start
,
1473 btrfs_release_path(path
);
1477 if (hole_size
> max_hole_size
) {
1478 max_hole_start
= search_start
;
1479 max_hole_size
= hole_size
;
1484 if (max_hole_size
< num_bytes
)
1490 btrfs_free_path(path
);
1491 *start
= max_hole_start
;
1493 *len
= max_hole_size
;
1497 int find_free_dev_extent(struct btrfs_trans_handle
*trans
,
1498 struct btrfs_device
*device
, u64 num_bytes
,
1499 u64
*start
, u64
*len
)
1501 /* FIXME use last free of some kind */
1502 return find_free_dev_extent_start(trans
->transaction
, device
,
1503 num_bytes
, 0, start
, len
);
1506 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1507 struct btrfs_device
*device
,
1508 u64 start
, u64
*dev_extent_len
)
1510 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1511 struct btrfs_root
*root
= fs_info
->dev_root
;
1513 struct btrfs_path
*path
;
1514 struct btrfs_key key
;
1515 struct btrfs_key found_key
;
1516 struct extent_buffer
*leaf
= NULL
;
1517 struct btrfs_dev_extent
*extent
= NULL
;
1519 path
= btrfs_alloc_path();
1523 key
.objectid
= device
->devid
;
1525 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1527 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1529 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1530 BTRFS_DEV_EXTENT_KEY
);
1533 leaf
= path
->nodes
[0];
1534 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1535 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1536 struct btrfs_dev_extent
);
1537 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1538 btrfs_dev_extent_length(leaf
, extent
) < start
);
1540 btrfs_release_path(path
);
1542 } else if (ret
== 0) {
1543 leaf
= path
->nodes
[0];
1544 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1545 struct btrfs_dev_extent
);
1547 btrfs_handle_fs_error(fs_info
, ret
, "Slot search failed");
1551 *dev_extent_len
= btrfs_dev_extent_length(leaf
, extent
);
1553 ret
= btrfs_del_item(trans
, root
, path
);
1555 btrfs_handle_fs_error(fs_info
, ret
,
1556 "Failed to remove dev extent item");
1558 set_bit(BTRFS_TRANS_HAVE_FREE_BGS
, &trans
->transaction
->flags
);
1561 btrfs_free_path(path
);
1565 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1566 struct btrfs_device
*device
,
1567 u64 chunk_tree
, u64 chunk_objectid
,
1568 u64 chunk_offset
, u64 start
, u64 num_bytes
)
1571 struct btrfs_path
*path
;
1572 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1573 struct btrfs_root
*root
= fs_info
->dev_root
;
1574 struct btrfs_dev_extent
*extent
;
1575 struct extent_buffer
*leaf
;
1576 struct btrfs_key key
;
1578 WARN_ON(!device
->in_fs_metadata
);
1579 WARN_ON(device
->is_tgtdev_for_dev_replace
);
1580 path
= btrfs_alloc_path();
1584 key
.objectid
= device
->devid
;
1586 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1587 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1592 leaf
= path
->nodes
[0];
1593 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1594 struct btrfs_dev_extent
);
1595 btrfs_set_dev_extent_chunk_tree(leaf
, extent
, chunk_tree
);
1596 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
, chunk_objectid
);
1597 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1599 write_extent_buffer_chunk_tree_uuid(leaf
, fs_info
->chunk_tree_uuid
);
1601 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1602 btrfs_mark_buffer_dirty(leaf
);
1604 btrfs_free_path(path
);
1608 static u64
find_next_chunk(struct btrfs_fs_info
*fs_info
)
1610 struct extent_map_tree
*em_tree
;
1611 struct extent_map
*em
;
1615 em_tree
= &fs_info
->mapping_tree
.map_tree
;
1616 read_lock(&em_tree
->lock
);
1617 n
= rb_last(&em_tree
->map
);
1619 em
= rb_entry(n
, struct extent_map
, rb_node
);
1620 ret
= em
->start
+ em
->len
;
1622 read_unlock(&em_tree
->lock
);
1627 static noinline
int find_next_devid(struct btrfs_fs_info
*fs_info
,
1631 struct btrfs_key key
;
1632 struct btrfs_key found_key
;
1633 struct btrfs_path
*path
;
1635 path
= btrfs_alloc_path();
1639 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1640 key
.type
= BTRFS_DEV_ITEM_KEY
;
1641 key
.offset
= (u64
)-1;
1643 ret
= btrfs_search_slot(NULL
, fs_info
->chunk_root
, &key
, path
, 0, 0);
1647 BUG_ON(ret
== 0); /* Corruption */
1649 ret
= btrfs_previous_item(fs_info
->chunk_root
, path
,
1650 BTRFS_DEV_ITEMS_OBJECTID
,
1651 BTRFS_DEV_ITEM_KEY
);
1655 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1657 *devid_ret
= found_key
.offset
+ 1;
1661 btrfs_free_path(path
);
1666 * the device information is stored in the chunk root
1667 * the btrfs_device struct should be fully filled in
1669 static int btrfs_add_device(struct btrfs_trans_handle
*trans
,
1670 struct btrfs_fs_info
*fs_info
,
1671 struct btrfs_device
*device
)
1673 struct btrfs_root
*root
= fs_info
->chunk_root
;
1675 struct btrfs_path
*path
;
1676 struct btrfs_dev_item
*dev_item
;
1677 struct extent_buffer
*leaf
;
1678 struct btrfs_key key
;
1681 path
= btrfs_alloc_path();
1685 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1686 key
.type
= BTRFS_DEV_ITEM_KEY
;
1687 key
.offset
= device
->devid
;
1689 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1694 leaf
= path
->nodes
[0];
1695 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1697 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1698 btrfs_set_device_generation(leaf
, dev_item
, 0);
1699 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1700 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1701 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1702 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1703 btrfs_set_device_total_bytes(leaf
, dev_item
,
1704 btrfs_device_get_disk_total_bytes(device
));
1705 btrfs_set_device_bytes_used(leaf
, dev_item
,
1706 btrfs_device_get_bytes_used(device
));
1707 btrfs_set_device_group(leaf
, dev_item
, 0);
1708 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1709 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1710 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1712 ptr
= btrfs_device_uuid(dev_item
);
1713 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1714 ptr
= btrfs_device_fsid(dev_item
);
1715 write_extent_buffer(leaf
, fs_info
->fsid
, ptr
, BTRFS_UUID_SIZE
);
1716 btrfs_mark_buffer_dirty(leaf
);
1720 btrfs_free_path(path
);
1725 * Function to update ctime/mtime for a given device path.
1726 * Mainly used for ctime/mtime based probe like libblkid.
1728 static void update_dev_time(char *path_name
)
1732 filp
= filp_open(path_name
, O_RDWR
, 0);
1735 file_update_time(filp
);
1736 filp_close(filp
, NULL
);
1739 static int btrfs_rm_dev_item(struct btrfs_fs_info
*fs_info
,
1740 struct btrfs_device
*device
)
1742 struct btrfs_root
*root
= fs_info
->chunk_root
;
1744 struct btrfs_path
*path
;
1745 struct btrfs_key key
;
1746 struct btrfs_trans_handle
*trans
;
1748 path
= btrfs_alloc_path();
1752 trans
= btrfs_start_transaction(root
, 0);
1753 if (IS_ERR(trans
)) {
1754 btrfs_free_path(path
);
1755 return PTR_ERR(trans
);
1757 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1758 key
.type
= BTRFS_DEV_ITEM_KEY
;
1759 key
.offset
= device
->devid
;
1761 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1770 ret
= btrfs_del_item(trans
, root
, path
);
1774 btrfs_free_path(path
);
1775 btrfs_commit_transaction(trans
);
1780 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1781 * filesystem. It's up to the caller to adjust that number regarding eg. device
1784 static int btrfs_check_raid_min_devices(struct btrfs_fs_info
*fs_info
,
1792 seq
= read_seqbegin(&fs_info
->profiles_lock
);
1794 all_avail
= fs_info
->avail_data_alloc_bits
|
1795 fs_info
->avail_system_alloc_bits
|
1796 fs_info
->avail_metadata_alloc_bits
;
1797 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
1799 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
1800 if (!(all_avail
& btrfs_raid_group
[i
]))
1803 if (num_devices
< btrfs_raid_array
[i
].devs_min
) {
1804 int ret
= btrfs_raid_mindev_error
[i
];
1814 struct btrfs_device
*btrfs_find_next_active_device(struct btrfs_fs_devices
*fs_devs
,
1815 struct btrfs_device
*device
)
1817 struct btrfs_device
*next_device
;
1819 list_for_each_entry(next_device
, &fs_devs
->devices
, dev_list
) {
1820 if (next_device
!= device
&&
1821 !next_device
->missing
&& next_device
->bdev
)
1829 * Helper function to check if the given device is part of s_bdev / latest_bdev
1830 * and replace it with the provided or the next active device, in the context
1831 * where this function called, there should be always be another device (or
1832 * this_dev) which is active.
1834 void btrfs_assign_next_active_device(struct btrfs_fs_info
*fs_info
,
1835 struct btrfs_device
*device
, struct btrfs_device
*this_dev
)
1837 struct btrfs_device
*next_device
;
1840 next_device
= this_dev
;
1842 next_device
= btrfs_find_next_active_device(fs_info
->fs_devices
,
1844 ASSERT(next_device
);
1846 if (fs_info
->sb
->s_bdev
&&
1847 (fs_info
->sb
->s_bdev
== device
->bdev
))
1848 fs_info
->sb
->s_bdev
= next_device
->bdev
;
1850 if (fs_info
->fs_devices
->latest_bdev
== device
->bdev
)
1851 fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1854 int btrfs_rm_device(struct btrfs_fs_info
*fs_info
, char *device_path
, u64 devid
)
1856 struct btrfs_device
*device
;
1857 struct btrfs_fs_devices
*cur_devices
;
1860 bool clear_super
= false;
1862 mutex_lock(&uuid_mutex
);
1864 num_devices
= fs_info
->fs_devices
->num_devices
;
1865 btrfs_dev_replace_lock(&fs_info
->dev_replace
, 0);
1866 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
1867 WARN_ON(num_devices
< 1);
1870 btrfs_dev_replace_unlock(&fs_info
->dev_replace
, 0);
1872 ret
= btrfs_check_raid_min_devices(fs_info
, num_devices
- 1);
1876 ret
= btrfs_find_device_by_devspec(fs_info
, devid
, device_path
,
1881 if (device
->is_tgtdev_for_dev_replace
) {
1882 ret
= BTRFS_ERROR_DEV_TGT_REPLACE
;
1886 if (device
->writeable
&& fs_info
->fs_devices
->rw_devices
== 1) {
1887 ret
= BTRFS_ERROR_DEV_ONLY_WRITABLE
;
1891 if (device
->writeable
) {
1892 mutex_lock(&fs_info
->chunk_mutex
);
1893 list_del_init(&device
->dev_alloc_list
);
1894 device
->fs_devices
->rw_devices
--;
1895 mutex_unlock(&fs_info
->chunk_mutex
);
1899 mutex_unlock(&uuid_mutex
);
1900 ret
= btrfs_shrink_device(device
, 0);
1901 mutex_lock(&uuid_mutex
);
1906 * TODO: the superblock still includes this device in its num_devices
1907 * counter although write_all_supers() is not locked out. This
1908 * could give a filesystem state which requires a degraded mount.
1910 ret
= btrfs_rm_dev_item(fs_info
, device
);
1914 device
->in_fs_metadata
= 0;
1915 btrfs_scrub_cancel_dev(fs_info
, device
);
1918 * the device list mutex makes sure that we don't change
1919 * the device list while someone else is writing out all
1920 * the device supers. Whoever is writing all supers, should
1921 * lock the device list mutex before getting the number of
1922 * devices in the super block (super_copy). Conversely,
1923 * whoever updates the number of devices in the super block
1924 * (super_copy) should hold the device list mutex.
1927 cur_devices
= device
->fs_devices
;
1928 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
1929 list_del_rcu(&device
->dev_list
);
1931 device
->fs_devices
->num_devices
--;
1932 device
->fs_devices
->total_devices
--;
1934 if (device
->missing
)
1935 device
->fs_devices
->missing_devices
--;
1937 btrfs_assign_next_active_device(fs_info
, device
, NULL
);
1940 device
->fs_devices
->open_devices
--;
1941 /* remove sysfs entry */
1942 btrfs_sysfs_rm_device_link(fs_info
->fs_devices
, device
);
1945 num_devices
= btrfs_super_num_devices(fs_info
->super_copy
) - 1;
1946 btrfs_set_super_num_devices(fs_info
->super_copy
, num_devices
);
1947 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
1950 * at this point, the device is zero sized and detached from
1951 * the devices list. All that's left is to zero out the old
1952 * supers and free the device.
1954 if (device
->writeable
)
1955 btrfs_scratch_superblocks(device
->bdev
, device
->name
->str
);
1957 btrfs_close_bdev(device
);
1958 call_rcu(&device
->rcu
, free_device
);
1960 if (cur_devices
->open_devices
== 0) {
1961 struct btrfs_fs_devices
*fs_devices
;
1962 fs_devices
= fs_info
->fs_devices
;
1963 while (fs_devices
) {
1964 if (fs_devices
->seed
== cur_devices
) {
1965 fs_devices
->seed
= cur_devices
->seed
;
1968 fs_devices
= fs_devices
->seed
;
1970 cur_devices
->seed
= NULL
;
1971 __btrfs_close_devices(cur_devices
);
1972 free_fs_devices(cur_devices
);
1975 fs_info
->num_tolerated_disk_barrier_failures
=
1976 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
1979 mutex_unlock(&uuid_mutex
);
1983 if (device
->writeable
) {
1984 mutex_lock(&fs_info
->chunk_mutex
);
1985 list_add(&device
->dev_alloc_list
,
1986 &fs_info
->fs_devices
->alloc_list
);
1987 device
->fs_devices
->rw_devices
++;
1988 mutex_unlock(&fs_info
->chunk_mutex
);
1993 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info
*fs_info
,
1994 struct btrfs_device
*srcdev
)
1996 struct btrfs_fs_devices
*fs_devices
;
1998 WARN_ON(!mutex_is_locked(&fs_info
->fs_devices
->device_list_mutex
));
2001 * in case of fs with no seed, srcdev->fs_devices will point
2002 * to fs_devices of fs_info. However when the dev being replaced is
2003 * a seed dev it will point to the seed's local fs_devices. In short
2004 * srcdev will have its correct fs_devices in both the cases.
2006 fs_devices
= srcdev
->fs_devices
;
2008 list_del_rcu(&srcdev
->dev_list
);
2009 list_del_rcu(&srcdev
->dev_alloc_list
);
2010 fs_devices
->num_devices
--;
2011 if (srcdev
->missing
)
2012 fs_devices
->missing_devices
--;
2014 if (srcdev
->writeable
)
2015 fs_devices
->rw_devices
--;
2018 fs_devices
->open_devices
--;
2021 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info
*fs_info
,
2022 struct btrfs_device
*srcdev
)
2024 struct btrfs_fs_devices
*fs_devices
= srcdev
->fs_devices
;
2026 if (srcdev
->writeable
) {
2027 /* zero out the old super if it is writable */
2028 btrfs_scratch_superblocks(srcdev
->bdev
, srcdev
->name
->str
);
2031 btrfs_close_bdev(srcdev
);
2033 call_rcu(&srcdev
->rcu
, free_device
);
2036 * unless fs_devices is seed fs, num_devices shouldn't go
2039 BUG_ON(!fs_devices
->num_devices
&& !fs_devices
->seeding
);
2041 /* if this is no devs we rather delete the fs_devices */
2042 if (!fs_devices
->num_devices
) {
2043 struct btrfs_fs_devices
*tmp_fs_devices
;
2045 tmp_fs_devices
= fs_info
->fs_devices
;
2046 while (tmp_fs_devices
) {
2047 if (tmp_fs_devices
->seed
== fs_devices
) {
2048 tmp_fs_devices
->seed
= fs_devices
->seed
;
2051 tmp_fs_devices
= tmp_fs_devices
->seed
;
2053 fs_devices
->seed
= NULL
;
2054 __btrfs_close_devices(fs_devices
);
2055 free_fs_devices(fs_devices
);
2059 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
2060 struct btrfs_device
*tgtdev
)
2062 mutex_lock(&uuid_mutex
);
2064 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2066 btrfs_sysfs_rm_device_link(fs_info
->fs_devices
, tgtdev
);
2069 fs_info
->fs_devices
->open_devices
--;
2071 fs_info
->fs_devices
->num_devices
--;
2073 btrfs_assign_next_active_device(fs_info
, tgtdev
, NULL
);
2075 list_del_rcu(&tgtdev
->dev_list
);
2077 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2078 mutex_unlock(&uuid_mutex
);
2081 * The update_dev_time() with in btrfs_scratch_superblocks()
2082 * may lead to a call to btrfs_show_devname() which will try
2083 * to hold device_list_mutex. And here this device
2084 * is already out of device list, so we don't have to hold
2085 * the device_list_mutex lock.
2087 btrfs_scratch_superblocks(tgtdev
->bdev
, tgtdev
->name
->str
);
2089 btrfs_close_bdev(tgtdev
);
2090 call_rcu(&tgtdev
->rcu
, free_device
);
2093 static int btrfs_find_device_by_path(struct btrfs_fs_info
*fs_info
,
2095 struct btrfs_device
**device
)
2098 struct btrfs_super_block
*disk_super
;
2101 struct block_device
*bdev
;
2102 struct buffer_head
*bh
;
2105 ret
= btrfs_get_bdev_and_sb(device_path
, FMODE_READ
,
2106 fs_info
->bdev_holder
, 0, &bdev
, &bh
);
2109 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
2110 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
2111 dev_uuid
= disk_super
->dev_item
.uuid
;
2112 *device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, disk_super
->fsid
);
2116 blkdev_put(bdev
, FMODE_READ
);
2120 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info
*fs_info
,
2122 struct btrfs_device
**device
)
2125 if (strcmp(device_path
, "missing") == 0) {
2126 struct list_head
*devices
;
2127 struct btrfs_device
*tmp
;
2129 devices
= &fs_info
->fs_devices
->devices
;
2131 * It is safe to read the devices since the volume_mutex
2132 * is held by the caller.
2134 list_for_each_entry(tmp
, devices
, dev_list
) {
2135 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
2142 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND
;
2146 return btrfs_find_device_by_path(fs_info
, device_path
, device
);
2151 * Lookup a device given by device id, or the path if the id is 0.
2153 int btrfs_find_device_by_devspec(struct btrfs_fs_info
*fs_info
, u64 devid
,
2154 char *devpath
, struct btrfs_device
**device
)
2160 *device
= btrfs_find_device(fs_info
, devid
, NULL
, NULL
);
2164 if (!devpath
|| !devpath
[0])
2167 ret
= btrfs_find_device_missing_or_by_path(fs_info
, devpath
,
2174 * does all the dirty work required for changing file system's UUID.
2176 static int btrfs_prepare_sprout(struct btrfs_fs_info
*fs_info
)
2178 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2179 struct btrfs_fs_devices
*old_devices
;
2180 struct btrfs_fs_devices
*seed_devices
;
2181 struct btrfs_super_block
*disk_super
= fs_info
->super_copy
;
2182 struct btrfs_device
*device
;
2185 BUG_ON(!mutex_is_locked(&uuid_mutex
));
2186 if (!fs_devices
->seeding
)
2189 seed_devices
= __alloc_fs_devices();
2190 if (IS_ERR(seed_devices
))
2191 return PTR_ERR(seed_devices
);
2193 old_devices
= clone_fs_devices(fs_devices
);
2194 if (IS_ERR(old_devices
)) {
2195 kfree(seed_devices
);
2196 return PTR_ERR(old_devices
);
2199 list_add(&old_devices
->list
, &fs_uuids
);
2201 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
2202 seed_devices
->opened
= 1;
2203 INIT_LIST_HEAD(&seed_devices
->devices
);
2204 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
2205 mutex_init(&seed_devices
->device_list_mutex
);
2207 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2208 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
2210 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
)
2211 device
->fs_devices
= seed_devices
;
2213 mutex_lock(&fs_info
->chunk_mutex
);
2214 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
2215 mutex_unlock(&fs_info
->chunk_mutex
);
2217 fs_devices
->seeding
= 0;
2218 fs_devices
->num_devices
= 0;
2219 fs_devices
->open_devices
= 0;
2220 fs_devices
->missing_devices
= 0;
2221 fs_devices
->rotating
= 0;
2222 fs_devices
->seed
= seed_devices
;
2224 generate_random_uuid(fs_devices
->fsid
);
2225 memcpy(fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2226 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2227 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2229 super_flags
= btrfs_super_flags(disk_super
) &
2230 ~BTRFS_SUPER_FLAG_SEEDING
;
2231 btrfs_set_super_flags(disk_super
, super_flags
);
2237 * Store the expected generation for seed devices in device items.
2239 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
2240 struct btrfs_fs_info
*fs_info
)
2242 struct btrfs_root
*root
= fs_info
->chunk_root
;
2243 struct btrfs_path
*path
;
2244 struct extent_buffer
*leaf
;
2245 struct btrfs_dev_item
*dev_item
;
2246 struct btrfs_device
*device
;
2247 struct btrfs_key key
;
2248 u8 fs_uuid
[BTRFS_UUID_SIZE
];
2249 u8 dev_uuid
[BTRFS_UUID_SIZE
];
2253 path
= btrfs_alloc_path();
2257 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2259 key
.type
= BTRFS_DEV_ITEM_KEY
;
2262 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2266 leaf
= path
->nodes
[0];
2268 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
2269 ret
= btrfs_next_leaf(root
, path
);
2274 leaf
= path
->nodes
[0];
2275 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2276 btrfs_release_path(path
);
2280 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2281 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
2282 key
.type
!= BTRFS_DEV_ITEM_KEY
)
2285 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2286 struct btrfs_dev_item
);
2287 devid
= btrfs_device_id(leaf
, dev_item
);
2288 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
2290 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
2292 device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, fs_uuid
);
2293 BUG_ON(!device
); /* Logic error */
2295 if (device
->fs_devices
->seeding
) {
2296 btrfs_set_device_generation(leaf
, dev_item
,
2297 device
->generation
);
2298 btrfs_mark_buffer_dirty(leaf
);
2306 btrfs_free_path(path
);
2310 int btrfs_init_new_device(struct btrfs_fs_info
*fs_info
, char *device_path
)
2312 struct btrfs_root
*root
= fs_info
->dev_root
;
2313 struct request_queue
*q
;
2314 struct btrfs_trans_handle
*trans
;
2315 struct btrfs_device
*device
;
2316 struct block_device
*bdev
;
2317 struct list_head
*devices
;
2318 struct super_block
*sb
= fs_info
->sb
;
2319 struct rcu_string
*name
;
2321 int seeding_dev
= 0;
2324 if ((sb
->s_flags
& MS_RDONLY
) && !fs_info
->fs_devices
->seeding
)
2327 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2328 fs_info
->bdev_holder
);
2330 return PTR_ERR(bdev
);
2332 if (fs_info
->fs_devices
->seeding
) {
2334 down_write(&sb
->s_umount
);
2335 mutex_lock(&uuid_mutex
);
2338 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2340 devices
= &fs_info
->fs_devices
->devices
;
2342 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2343 list_for_each_entry(device
, devices
, dev_list
) {
2344 if (device
->bdev
== bdev
) {
2347 &fs_info
->fs_devices
->device_list_mutex
);
2351 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2353 device
= btrfs_alloc_device(fs_info
, NULL
, NULL
);
2354 if (IS_ERR(device
)) {
2355 /* we can safely leave the fs_devices entry around */
2356 ret
= PTR_ERR(device
);
2360 name
= rcu_string_strdup(device_path
, GFP_KERNEL
);
2366 rcu_assign_pointer(device
->name
, name
);
2368 trans
= btrfs_start_transaction(root
, 0);
2369 if (IS_ERR(trans
)) {
2370 rcu_string_free(device
->name
);
2372 ret
= PTR_ERR(trans
);
2376 q
= bdev_get_queue(bdev
);
2377 if (blk_queue_discard(q
))
2378 device
->can_discard
= 1;
2379 device
->writeable
= 1;
2380 device
->generation
= trans
->transid
;
2381 device
->io_width
= fs_info
->sectorsize
;
2382 device
->io_align
= fs_info
->sectorsize
;
2383 device
->sector_size
= fs_info
->sectorsize
;
2384 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
2385 device
->disk_total_bytes
= device
->total_bytes
;
2386 device
->commit_total_bytes
= device
->total_bytes
;
2387 device
->fs_info
= fs_info
;
2388 device
->bdev
= bdev
;
2389 device
->in_fs_metadata
= 1;
2390 device
->is_tgtdev_for_dev_replace
= 0;
2391 device
->mode
= FMODE_EXCL
;
2392 device
->dev_stats_valid
= 1;
2393 set_blocksize(device
->bdev
, 4096);
2396 sb
->s_flags
&= ~MS_RDONLY
;
2397 ret
= btrfs_prepare_sprout(fs_info
);
2398 BUG_ON(ret
); /* -ENOMEM */
2401 device
->fs_devices
= fs_info
->fs_devices
;
2403 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2404 mutex_lock(&fs_info
->chunk_mutex
);
2405 list_add_rcu(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2406 list_add(&device
->dev_alloc_list
,
2407 &fs_info
->fs_devices
->alloc_list
);
2408 fs_info
->fs_devices
->num_devices
++;
2409 fs_info
->fs_devices
->open_devices
++;
2410 fs_info
->fs_devices
->rw_devices
++;
2411 fs_info
->fs_devices
->total_devices
++;
2412 fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
2414 spin_lock(&fs_info
->free_chunk_lock
);
2415 fs_info
->free_chunk_space
+= device
->total_bytes
;
2416 spin_unlock(&fs_info
->free_chunk_lock
);
2418 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
2419 fs_info
->fs_devices
->rotating
= 1;
2421 tmp
= btrfs_super_total_bytes(fs_info
->super_copy
);
2422 btrfs_set_super_total_bytes(fs_info
->super_copy
,
2423 tmp
+ device
->total_bytes
);
2425 tmp
= btrfs_super_num_devices(fs_info
->super_copy
);
2426 btrfs_set_super_num_devices(fs_info
->super_copy
, tmp
+ 1);
2428 /* add sysfs device entry */
2429 btrfs_sysfs_add_device_link(fs_info
->fs_devices
, device
);
2432 * we've got more storage, clear any full flags on the space
2435 btrfs_clear_space_info_full(fs_info
);
2437 mutex_unlock(&fs_info
->chunk_mutex
);
2438 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2441 mutex_lock(&fs_info
->chunk_mutex
);
2442 ret
= init_first_rw_device(trans
, fs_info
);
2443 mutex_unlock(&fs_info
->chunk_mutex
);
2445 btrfs_abort_transaction(trans
, ret
);
2450 ret
= btrfs_add_device(trans
, fs_info
, device
);
2452 btrfs_abort_transaction(trans
, ret
);
2457 char fsid_buf
[BTRFS_UUID_UNPARSED_SIZE
];
2459 ret
= btrfs_finish_sprout(trans
, fs_info
);
2461 btrfs_abort_transaction(trans
, ret
);
2465 /* Sprouting would change fsid of the mounted root,
2466 * so rename the fsid on the sysfs
2468 snprintf(fsid_buf
, BTRFS_UUID_UNPARSED_SIZE
, "%pU",
2470 if (kobject_rename(&fs_info
->fs_devices
->fsid_kobj
, fsid_buf
))
2472 "sysfs: failed to create fsid for sprout");
2475 fs_info
->num_tolerated_disk_barrier_failures
=
2476 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
2477 ret
= btrfs_commit_transaction(trans
);
2480 mutex_unlock(&uuid_mutex
);
2481 up_write(&sb
->s_umount
);
2483 if (ret
) /* transaction commit */
2486 ret
= btrfs_relocate_sys_chunks(fs_info
);
2488 btrfs_handle_fs_error(fs_info
, ret
,
2489 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2490 trans
= btrfs_attach_transaction(root
);
2491 if (IS_ERR(trans
)) {
2492 if (PTR_ERR(trans
) == -ENOENT
)
2494 return PTR_ERR(trans
);
2496 ret
= btrfs_commit_transaction(trans
);
2499 /* Update ctime/mtime for libblkid */
2500 update_dev_time(device_path
);
2504 btrfs_end_transaction(trans
);
2505 rcu_string_free(device
->name
);
2506 btrfs_sysfs_rm_device_link(fs_info
->fs_devices
, device
);
2509 blkdev_put(bdev
, FMODE_EXCL
);
2511 mutex_unlock(&uuid_mutex
);
2512 up_write(&sb
->s_umount
);
2517 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
2519 struct btrfs_device
*srcdev
,
2520 struct btrfs_device
**device_out
)
2522 struct request_queue
*q
;
2523 struct btrfs_device
*device
;
2524 struct block_device
*bdev
;
2525 struct list_head
*devices
;
2526 struct rcu_string
*name
;
2527 u64 devid
= BTRFS_DEV_REPLACE_DEVID
;
2531 if (fs_info
->fs_devices
->seeding
) {
2532 btrfs_err(fs_info
, "the filesystem is a seed filesystem!");
2536 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2537 fs_info
->bdev_holder
);
2539 btrfs_err(fs_info
, "target device %s is invalid!", device_path
);
2540 return PTR_ERR(bdev
);
2543 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2545 devices
= &fs_info
->fs_devices
->devices
;
2546 list_for_each_entry(device
, devices
, dev_list
) {
2547 if (device
->bdev
== bdev
) {
2549 "target device is in the filesystem!");
2556 if (i_size_read(bdev
->bd_inode
) <
2557 btrfs_device_get_total_bytes(srcdev
)) {
2559 "target device is smaller than source device!");
2565 device
= btrfs_alloc_device(NULL
, &devid
, NULL
);
2566 if (IS_ERR(device
)) {
2567 ret
= PTR_ERR(device
);
2571 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2577 rcu_assign_pointer(device
->name
, name
);
2579 q
= bdev_get_queue(bdev
);
2580 if (blk_queue_discard(q
))
2581 device
->can_discard
= 1;
2582 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2583 device
->writeable
= 1;
2584 device
->generation
= 0;
2585 device
->io_width
= fs_info
->sectorsize
;
2586 device
->io_align
= fs_info
->sectorsize
;
2587 device
->sector_size
= fs_info
->sectorsize
;
2588 device
->total_bytes
= btrfs_device_get_total_bytes(srcdev
);
2589 device
->disk_total_bytes
= btrfs_device_get_disk_total_bytes(srcdev
);
2590 device
->bytes_used
= btrfs_device_get_bytes_used(srcdev
);
2591 ASSERT(list_empty(&srcdev
->resized_list
));
2592 device
->commit_total_bytes
= srcdev
->commit_total_bytes
;
2593 device
->commit_bytes_used
= device
->bytes_used
;
2594 device
->fs_info
= fs_info
;
2595 device
->bdev
= bdev
;
2596 device
->in_fs_metadata
= 1;
2597 device
->is_tgtdev_for_dev_replace
= 1;
2598 device
->mode
= FMODE_EXCL
;
2599 device
->dev_stats_valid
= 1;
2600 set_blocksize(device
->bdev
, 4096);
2601 device
->fs_devices
= fs_info
->fs_devices
;
2602 list_add(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2603 fs_info
->fs_devices
->num_devices
++;
2604 fs_info
->fs_devices
->open_devices
++;
2605 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2607 *device_out
= device
;
2611 blkdev_put(bdev
, FMODE_EXCL
);
2615 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info
*fs_info
,
2616 struct btrfs_device
*tgtdev
)
2618 u32 sectorsize
= fs_info
->sectorsize
;
2620 WARN_ON(fs_info
->fs_devices
->rw_devices
== 0);
2621 tgtdev
->io_width
= sectorsize
;
2622 tgtdev
->io_align
= sectorsize
;
2623 tgtdev
->sector_size
= sectorsize
;
2624 tgtdev
->fs_info
= fs_info
;
2625 tgtdev
->in_fs_metadata
= 1;
2628 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2629 struct btrfs_device
*device
)
2632 struct btrfs_path
*path
;
2633 struct btrfs_root
*root
= device
->fs_info
->chunk_root
;
2634 struct btrfs_dev_item
*dev_item
;
2635 struct extent_buffer
*leaf
;
2636 struct btrfs_key key
;
2638 path
= btrfs_alloc_path();
2642 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2643 key
.type
= BTRFS_DEV_ITEM_KEY
;
2644 key
.offset
= device
->devid
;
2646 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2655 leaf
= path
->nodes
[0];
2656 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2658 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2659 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2660 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2661 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2662 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2663 btrfs_set_device_total_bytes(leaf
, dev_item
,
2664 btrfs_device_get_disk_total_bytes(device
));
2665 btrfs_set_device_bytes_used(leaf
, dev_item
,
2666 btrfs_device_get_bytes_used(device
));
2667 btrfs_mark_buffer_dirty(leaf
);
2670 btrfs_free_path(path
);
2674 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2675 struct btrfs_device
*device
, u64 new_size
)
2677 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
2678 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
2679 struct btrfs_fs_devices
*fs_devices
;
2683 if (!device
->writeable
)
2686 mutex_lock(&fs_info
->chunk_mutex
);
2687 old_total
= btrfs_super_total_bytes(super_copy
);
2688 diff
= new_size
- device
->total_bytes
;
2690 if (new_size
<= device
->total_bytes
||
2691 device
->is_tgtdev_for_dev_replace
) {
2692 mutex_unlock(&fs_info
->chunk_mutex
);
2696 fs_devices
= fs_info
->fs_devices
;
2698 btrfs_set_super_total_bytes(super_copy
, old_total
+ diff
);
2699 device
->fs_devices
->total_rw_bytes
+= diff
;
2701 btrfs_device_set_total_bytes(device
, new_size
);
2702 btrfs_device_set_disk_total_bytes(device
, new_size
);
2703 btrfs_clear_space_info_full(device
->fs_info
);
2704 if (list_empty(&device
->resized_list
))
2705 list_add_tail(&device
->resized_list
,
2706 &fs_devices
->resized_devices
);
2707 mutex_unlock(&fs_info
->chunk_mutex
);
2709 return btrfs_update_device(trans
, device
);
2712 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
2713 struct btrfs_fs_info
*fs_info
, u64 chunk_objectid
,
2716 struct btrfs_root
*root
= fs_info
->chunk_root
;
2718 struct btrfs_path
*path
;
2719 struct btrfs_key key
;
2721 path
= btrfs_alloc_path();
2725 key
.objectid
= chunk_objectid
;
2726 key
.offset
= chunk_offset
;
2727 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2729 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2732 else if (ret
> 0) { /* Logic error or corruption */
2733 btrfs_handle_fs_error(fs_info
, -ENOENT
,
2734 "Failed lookup while freeing chunk.");
2739 ret
= btrfs_del_item(trans
, root
, path
);
2741 btrfs_handle_fs_error(fs_info
, ret
,
2742 "Failed to delete chunk item.");
2744 btrfs_free_path(path
);
2748 static int btrfs_del_sys_chunk(struct btrfs_fs_info
*fs_info
,
2749 u64 chunk_objectid
, u64 chunk_offset
)
2751 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
2752 struct btrfs_disk_key
*disk_key
;
2753 struct btrfs_chunk
*chunk
;
2760 struct btrfs_key key
;
2762 mutex_lock(&fs_info
->chunk_mutex
);
2763 array_size
= btrfs_super_sys_array_size(super_copy
);
2765 ptr
= super_copy
->sys_chunk_array
;
2768 while (cur
< array_size
) {
2769 disk_key
= (struct btrfs_disk_key
*)ptr
;
2770 btrfs_disk_key_to_cpu(&key
, disk_key
);
2772 len
= sizeof(*disk_key
);
2774 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2775 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2776 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2777 len
+= btrfs_chunk_item_size(num_stripes
);
2782 if (key
.objectid
== chunk_objectid
&&
2783 key
.offset
== chunk_offset
) {
2784 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2786 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2792 mutex_unlock(&fs_info
->chunk_mutex
);
2796 int btrfs_remove_chunk(struct btrfs_trans_handle
*trans
,
2797 struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2799 struct extent_map_tree
*em_tree
;
2800 struct extent_map
*em
;
2801 struct map_lookup
*map
;
2802 u64 dev_extent_len
= 0;
2803 u64 chunk_objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2805 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2807 em_tree
= &fs_info
->mapping_tree
.map_tree
;
2809 read_lock(&em_tree
->lock
);
2810 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
2811 read_unlock(&em_tree
->lock
);
2813 if (!em
|| em
->start
> chunk_offset
||
2814 em
->start
+ em
->len
< chunk_offset
) {
2816 * This is a logic error, but we don't want to just rely on the
2817 * user having built with ASSERT enabled, so if ASSERT doesn't
2818 * do anything we still error out.
2822 free_extent_map(em
);
2825 map
= em
->map_lookup
;
2826 mutex_lock(&fs_info
->chunk_mutex
);
2827 check_system_chunk(trans
, fs_info
, map
->type
);
2828 mutex_unlock(&fs_info
->chunk_mutex
);
2831 * Take the device list mutex to prevent races with the final phase of
2832 * a device replace operation that replaces the device object associated
2833 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2835 mutex_lock(&fs_devices
->device_list_mutex
);
2836 for (i
= 0; i
< map
->num_stripes
; i
++) {
2837 struct btrfs_device
*device
= map
->stripes
[i
].dev
;
2838 ret
= btrfs_free_dev_extent(trans
, device
,
2839 map
->stripes
[i
].physical
,
2842 mutex_unlock(&fs_devices
->device_list_mutex
);
2843 btrfs_abort_transaction(trans
, ret
);
2847 if (device
->bytes_used
> 0) {
2848 mutex_lock(&fs_info
->chunk_mutex
);
2849 btrfs_device_set_bytes_used(device
,
2850 device
->bytes_used
- dev_extent_len
);
2851 spin_lock(&fs_info
->free_chunk_lock
);
2852 fs_info
->free_chunk_space
+= dev_extent_len
;
2853 spin_unlock(&fs_info
->free_chunk_lock
);
2854 btrfs_clear_space_info_full(fs_info
);
2855 mutex_unlock(&fs_info
->chunk_mutex
);
2858 if (map
->stripes
[i
].dev
) {
2859 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2861 mutex_unlock(&fs_devices
->device_list_mutex
);
2862 btrfs_abort_transaction(trans
, ret
);
2867 mutex_unlock(&fs_devices
->device_list_mutex
);
2869 ret
= btrfs_free_chunk(trans
, fs_info
, chunk_objectid
, chunk_offset
);
2871 btrfs_abort_transaction(trans
, ret
);
2875 trace_btrfs_chunk_free(fs_info
, map
, chunk_offset
, em
->len
);
2877 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2878 ret
= btrfs_del_sys_chunk(fs_info
, chunk_objectid
,
2881 btrfs_abort_transaction(trans
, ret
);
2886 ret
= btrfs_remove_block_group(trans
, fs_info
, chunk_offset
, em
);
2888 btrfs_abort_transaction(trans
, ret
);
2894 free_extent_map(em
);
2898 static int btrfs_relocate_chunk(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2900 struct btrfs_root
*root
= fs_info
->chunk_root
;
2901 struct btrfs_trans_handle
*trans
;
2905 * Prevent races with automatic removal of unused block groups.
2906 * After we relocate and before we remove the chunk with offset
2907 * chunk_offset, automatic removal of the block group can kick in,
2908 * resulting in a failure when calling btrfs_remove_chunk() below.
2910 * Make sure to acquire this mutex before doing a tree search (dev
2911 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2912 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2913 * we release the path used to search the chunk/dev tree and before
2914 * the current task acquires this mutex and calls us.
2916 ASSERT(mutex_is_locked(&fs_info
->delete_unused_bgs_mutex
));
2918 ret
= btrfs_can_relocate(fs_info
, chunk_offset
);
2922 /* step one, relocate all the extents inside this chunk */
2923 btrfs_scrub_pause(fs_info
);
2924 ret
= btrfs_relocate_block_group(fs_info
, chunk_offset
);
2925 btrfs_scrub_continue(fs_info
);
2929 trans
= btrfs_start_trans_remove_block_group(root
->fs_info
,
2931 if (IS_ERR(trans
)) {
2932 ret
= PTR_ERR(trans
);
2933 btrfs_handle_fs_error(root
->fs_info
, ret
, NULL
);
2938 * step two, delete the device extents and the
2939 * chunk tree entries
2941 ret
= btrfs_remove_chunk(trans
, fs_info
, chunk_offset
);
2942 btrfs_end_transaction(trans
);
2946 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info
*fs_info
)
2948 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
2949 struct btrfs_path
*path
;
2950 struct extent_buffer
*leaf
;
2951 struct btrfs_chunk
*chunk
;
2952 struct btrfs_key key
;
2953 struct btrfs_key found_key
;
2955 bool retried
= false;
2959 path
= btrfs_alloc_path();
2964 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2965 key
.offset
= (u64
)-1;
2966 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2969 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
2970 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2972 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
2975 BUG_ON(ret
== 0); /* Corruption */
2977 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
2980 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
2986 leaf
= path
->nodes
[0];
2987 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2989 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
2990 struct btrfs_chunk
);
2991 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2992 btrfs_release_path(path
);
2994 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2995 ret
= btrfs_relocate_chunk(fs_info
, found_key
.offset
);
3001 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3003 if (found_key
.offset
== 0)
3005 key
.offset
= found_key
.offset
- 1;
3008 if (failed
&& !retried
) {
3012 } else if (WARN_ON(failed
&& retried
)) {
3016 btrfs_free_path(path
);
3020 static int insert_balance_item(struct btrfs_fs_info
*fs_info
,
3021 struct btrfs_balance_control
*bctl
)
3023 struct btrfs_root
*root
= fs_info
->tree_root
;
3024 struct btrfs_trans_handle
*trans
;
3025 struct btrfs_balance_item
*item
;
3026 struct btrfs_disk_balance_args disk_bargs
;
3027 struct btrfs_path
*path
;
3028 struct extent_buffer
*leaf
;
3029 struct btrfs_key key
;
3032 path
= btrfs_alloc_path();
3036 trans
= btrfs_start_transaction(root
, 0);
3037 if (IS_ERR(trans
)) {
3038 btrfs_free_path(path
);
3039 return PTR_ERR(trans
);
3042 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3043 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3046 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
3051 leaf
= path
->nodes
[0];
3052 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3054 memzero_extent_buffer(leaf
, (unsigned long)item
, sizeof(*item
));
3056 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
3057 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
3058 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
3059 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
3060 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
3061 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
3063 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
3065 btrfs_mark_buffer_dirty(leaf
);
3067 btrfs_free_path(path
);
3068 err
= btrfs_commit_transaction(trans
);
3074 static int del_balance_item(struct btrfs_fs_info
*fs_info
)
3076 struct btrfs_root
*root
= fs_info
->tree_root
;
3077 struct btrfs_trans_handle
*trans
;
3078 struct btrfs_path
*path
;
3079 struct btrfs_key key
;
3082 path
= btrfs_alloc_path();
3086 trans
= btrfs_start_transaction(root
, 0);
3087 if (IS_ERR(trans
)) {
3088 btrfs_free_path(path
);
3089 return PTR_ERR(trans
);
3092 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3093 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3096 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3104 ret
= btrfs_del_item(trans
, root
, path
);
3106 btrfs_free_path(path
);
3107 err
= btrfs_commit_transaction(trans
);
3114 * This is a heuristic used to reduce the number of chunks balanced on
3115 * resume after balance was interrupted.
3117 static void update_balance_args(struct btrfs_balance_control
*bctl
)
3120 * Turn on soft mode for chunk types that were being converted.
3122 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3123 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3124 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3125 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3126 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3127 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3130 * Turn on usage filter if is not already used. The idea is
3131 * that chunks that we have already balanced should be
3132 * reasonably full. Don't do it for chunks that are being
3133 * converted - that will keep us from relocating unconverted
3134 * (albeit full) chunks.
3136 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3137 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3138 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3139 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3140 bctl
->data
.usage
= 90;
3142 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3143 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3144 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3145 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3146 bctl
->sys
.usage
= 90;
3148 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3149 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3150 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3151 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3152 bctl
->meta
.usage
= 90;
3157 * Should be called with both balance and volume mutexes held to
3158 * serialize other volume operations (add_dev/rm_dev/resize) with
3159 * restriper. Same goes for unset_balance_control.
3161 static void set_balance_control(struct btrfs_balance_control
*bctl
)
3163 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3165 BUG_ON(fs_info
->balance_ctl
);
3167 spin_lock(&fs_info
->balance_lock
);
3168 fs_info
->balance_ctl
= bctl
;
3169 spin_unlock(&fs_info
->balance_lock
);
3172 static void unset_balance_control(struct btrfs_fs_info
*fs_info
)
3174 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3176 BUG_ON(!fs_info
->balance_ctl
);
3178 spin_lock(&fs_info
->balance_lock
);
3179 fs_info
->balance_ctl
= NULL
;
3180 spin_unlock(&fs_info
->balance_lock
);
3186 * Balance filters. Return 1 if chunk should be filtered out
3187 * (should not be balanced).
3189 static int chunk_profiles_filter(u64 chunk_type
,
3190 struct btrfs_balance_args
*bargs
)
3192 chunk_type
= chunk_to_extended(chunk_type
) &
3193 BTRFS_EXTENDED_PROFILE_MASK
;
3195 if (bargs
->profiles
& chunk_type
)
3201 static int chunk_usage_range_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
3202 struct btrfs_balance_args
*bargs
)
3204 struct btrfs_block_group_cache
*cache
;
3206 u64 user_thresh_min
;
3207 u64 user_thresh_max
;
3210 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3211 chunk_used
= btrfs_block_group_used(&cache
->item
);
3213 if (bargs
->usage_min
== 0)
3214 user_thresh_min
= 0;
3216 user_thresh_min
= div_factor_fine(cache
->key
.offset
,
3219 if (bargs
->usage_max
== 0)
3220 user_thresh_max
= 1;
3221 else if (bargs
->usage_max
> 100)
3222 user_thresh_max
= cache
->key
.offset
;
3224 user_thresh_max
= div_factor_fine(cache
->key
.offset
,
3227 if (user_thresh_min
<= chunk_used
&& chunk_used
< user_thresh_max
)
3230 btrfs_put_block_group(cache
);
3234 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
,
3235 u64 chunk_offset
, struct btrfs_balance_args
*bargs
)
3237 struct btrfs_block_group_cache
*cache
;
3238 u64 chunk_used
, user_thresh
;
3241 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3242 chunk_used
= btrfs_block_group_used(&cache
->item
);
3244 if (bargs
->usage_min
== 0)
3246 else if (bargs
->usage
> 100)
3247 user_thresh
= cache
->key
.offset
;
3249 user_thresh
= div_factor_fine(cache
->key
.offset
,
3252 if (chunk_used
< user_thresh
)
3255 btrfs_put_block_group(cache
);
3259 static int chunk_devid_filter(struct extent_buffer
*leaf
,
3260 struct btrfs_chunk
*chunk
,
3261 struct btrfs_balance_args
*bargs
)
3263 struct btrfs_stripe
*stripe
;
3264 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3267 for (i
= 0; i
< num_stripes
; i
++) {
3268 stripe
= btrfs_stripe_nr(chunk
, i
);
3269 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
3276 /* [pstart, pend) */
3277 static int chunk_drange_filter(struct extent_buffer
*leaf
,
3278 struct btrfs_chunk
*chunk
,
3280 struct btrfs_balance_args
*bargs
)
3282 struct btrfs_stripe
*stripe
;
3283 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3289 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
3292 if (btrfs_chunk_type(leaf
, chunk
) & (BTRFS_BLOCK_GROUP_DUP
|
3293 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)) {
3294 factor
= num_stripes
/ 2;
3295 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID5
) {
3296 factor
= num_stripes
- 1;
3297 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID6
) {
3298 factor
= num_stripes
- 2;
3300 factor
= num_stripes
;
3303 for (i
= 0; i
< num_stripes
; i
++) {
3304 stripe
= btrfs_stripe_nr(chunk
, i
);
3305 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
3308 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
3309 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
3310 stripe_length
= div_u64(stripe_length
, factor
);
3312 if (stripe_offset
< bargs
->pend
&&
3313 stripe_offset
+ stripe_length
> bargs
->pstart
)
3320 /* [vstart, vend) */
3321 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
3322 struct btrfs_chunk
*chunk
,
3324 struct btrfs_balance_args
*bargs
)
3326 if (chunk_offset
< bargs
->vend
&&
3327 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
3328 /* at least part of the chunk is inside this vrange */
3334 static int chunk_stripes_range_filter(struct extent_buffer
*leaf
,
3335 struct btrfs_chunk
*chunk
,
3336 struct btrfs_balance_args
*bargs
)
3338 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3340 if (bargs
->stripes_min
<= num_stripes
3341 && num_stripes
<= bargs
->stripes_max
)
3347 static int chunk_soft_convert_filter(u64 chunk_type
,
3348 struct btrfs_balance_args
*bargs
)
3350 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
3353 chunk_type
= chunk_to_extended(chunk_type
) &
3354 BTRFS_EXTENDED_PROFILE_MASK
;
3356 if (bargs
->target
== chunk_type
)
3362 static int should_balance_chunk(struct btrfs_fs_info
*fs_info
,
3363 struct extent_buffer
*leaf
,
3364 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
3366 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3367 struct btrfs_balance_args
*bargs
= NULL
;
3368 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3371 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
3372 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
3376 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3377 bargs
= &bctl
->data
;
3378 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3380 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3381 bargs
= &bctl
->meta
;
3383 /* profiles filter */
3384 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
3385 chunk_profiles_filter(chunk_type
, bargs
)) {
3390 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3391 chunk_usage_filter(fs_info
, chunk_offset
, bargs
)) {
3393 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3394 chunk_usage_range_filter(fs_info
, chunk_offset
, bargs
)) {
3399 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
3400 chunk_devid_filter(leaf
, chunk
, bargs
)) {
3404 /* drange filter, makes sense only with devid filter */
3405 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
3406 chunk_drange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3411 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
3412 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3416 /* stripes filter */
3417 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_STRIPES_RANGE
) &&
3418 chunk_stripes_range_filter(leaf
, chunk
, bargs
)) {
3422 /* soft profile changing mode */
3423 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
3424 chunk_soft_convert_filter(chunk_type
, bargs
)) {
3429 * limited by count, must be the last filter
3431 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT
)) {
3432 if (bargs
->limit
== 0)
3436 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT_RANGE
)) {
3438 * Same logic as the 'limit' filter; the minimum cannot be
3439 * determined here because we do not have the global information
3440 * about the count of all chunks that satisfy the filters.
3442 if (bargs
->limit_max
== 0)
3451 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
3453 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3454 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
3455 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
3456 struct list_head
*devices
;
3457 struct btrfs_device
*device
;
3461 struct btrfs_chunk
*chunk
;
3462 struct btrfs_path
*path
= NULL
;
3463 struct btrfs_key key
;
3464 struct btrfs_key found_key
;
3465 struct btrfs_trans_handle
*trans
;
3466 struct extent_buffer
*leaf
;
3469 int enospc_errors
= 0;
3470 bool counting
= true;
3471 /* The single value limit and min/max limits use the same bytes in the */
3472 u64 limit_data
= bctl
->data
.limit
;
3473 u64 limit_meta
= bctl
->meta
.limit
;
3474 u64 limit_sys
= bctl
->sys
.limit
;
3478 int chunk_reserved
= 0;
3481 /* step one make some room on all the devices */
3482 devices
= &fs_info
->fs_devices
->devices
;
3483 list_for_each_entry(device
, devices
, dev_list
) {
3484 old_size
= btrfs_device_get_total_bytes(device
);
3485 size_to_free
= div_factor(old_size
, 1);
3486 size_to_free
= min_t(u64
, size_to_free
, SZ_1M
);
3487 if (!device
->writeable
||
3488 btrfs_device_get_total_bytes(device
) -
3489 btrfs_device_get_bytes_used(device
) > size_to_free
||
3490 device
->is_tgtdev_for_dev_replace
)
3493 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
3497 /* btrfs_shrink_device never returns ret > 0 */
3502 trans
= btrfs_start_transaction(dev_root
, 0);
3503 if (IS_ERR(trans
)) {
3504 ret
= PTR_ERR(trans
);
3505 btrfs_info_in_rcu(fs_info
,
3506 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3507 rcu_str_deref(device
->name
), ret
,
3508 old_size
, old_size
- size_to_free
);
3512 ret
= btrfs_grow_device(trans
, device
, old_size
);
3514 btrfs_end_transaction(trans
);
3515 /* btrfs_grow_device never returns ret > 0 */
3517 btrfs_info_in_rcu(fs_info
,
3518 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3519 rcu_str_deref(device
->name
), ret
,
3520 old_size
, old_size
- size_to_free
);
3524 btrfs_end_transaction(trans
);
3527 /* step two, relocate all the chunks */
3528 path
= btrfs_alloc_path();
3534 /* zero out stat counters */
3535 spin_lock(&fs_info
->balance_lock
);
3536 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
3537 spin_unlock(&fs_info
->balance_lock
);
3541 * The single value limit and min/max limits use the same bytes
3544 bctl
->data
.limit
= limit_data
;
3545 bctl
->meta
.limit
= limit_meta
;
3546 bctl
->sys
.limit
= limit_sys
;
3548 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3549 key
.offset
= (u64
)-1;
3550 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3553 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
3554 atomic_read(&fs_info
->balance_cancel_req
)) {
3559 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
3560 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
3562 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3567 * this shouldn't happen, it means the last relocate
3571 BUG(); /* FIXME break ? */
3573 ret
= btrfs_previous_item(chunk_root
, path
, 0,
3574 BTRFS_CHUNK_ITEM_KEY
);
3576 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3581 leaf
= path
->nodes
[0];
3582 slot
= path
->slots
[0];
3583 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3585 if (found_key
.objectid
!= key
.objectid
) {
3586 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3590 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
3591 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3594 spin_lock(&fs_info
->balance_lock
);
3595 bctl
->stat
.considered
++;
3596 spin_unlock(&fs_info
->balance_lock
);
3599 ret
= should_balance_chunk(fs_info
, leaf
, chunk
,
3602 btrfs_release_path(path
);
3604 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3609 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3610 spin_lock(&fs_info
->balance_lock
);
3611 bctl
->stat
.expected
++;
3612 spin_unlock(&fs_info
->balance_lock
);
3614 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3616 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3618 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3625 * Apply limit_min filter, no need to check if the LIMITS
3626 * filter is used, limit_min is 0 by default
3628 if (((chunk_type
& BTRFS_BLOCK_GROUP_DATA
) &&
3629 count_data
< bctl
->data
.limit_min
)
3630 || ((chunk_type
& BTRFS_BLOCK_GROUP_METADATA
) &&
3631 count_meta
< bctl
->meta
.limit_min
)
3632 || ((chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) &&
3633 count_sys
< bctl
->sys
.limit_min
)) {
3634 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3638 ASSERT(fs_info
->data_sinfo
);
3639 spin_lock(&fs_info
->data_sinfo
->lock
);
3640 bytes_used
= fs_info
->data_sinfo
->bytes_used
;
3641 spin_unlock(&fs_info
->data_sinfo
->lock
);
3643 if ((chunk_type
& BTRFS_BLOCK_GROUP_DATA
) &&
3644 !chunk_reserved
&& !bytes_used
) {
3645 trans
= btrfs_start_transaction(chunk_root
, 0);
3646 if (IS_ERR(trans
)) {
3647 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3648 ret
= PTR_ERR(trans
);
3652 ret
= btrfs_force_chunk_alloc(trans
, fs_info
,
3653 BTRFS_BLOCK_GROUP_DATA
);
3654 btrfs_end_transaction(trans
);
3656 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3662 ret
= btrfs_relocate_chunk(fs_info
, found_key
.offset
);
3663 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3664 if (ret
&& ret
!= -ENOSPC
)
3666 if (ret
== -ENOSPC
) {
3669 spin_lock(&fs_info
->balance_lock
);
3670 bctl
->stat
.completed
++;
3671 spin_unlock(&fs_info
->balance_lock
);
3674 if (found_key
.offset
== 0)
3676 key
.offset
= found_key
.offset
- 1;
3680 btrfs_release_path(path
);
3685 btrfs_free_path(path
);
3686 if (enospc_errors
) {
3687 btrfs_info(fs_info
, "%d enospc errors during balance",
3697 * alloc_profile_is_valid - see if a given profile is valid and reduced
3698 * @flags: profile to validate
3699 * @extended: if true @flags is treated as an extended profile
3701 static int alloc_profile_is_valid(u64 flags
, int extended
)
3703 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
3704 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
3706 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
3708 /* 1) check that all other bits are zeroed */
3712 /* 2) see if profile is reduced */
3714 return !extended
; /* "0" is valid for usual profiles */
3716 /* true if exactly one bit set */
3717 return (flags
& (flags
- 1)) == 0;
3720 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
3722 /* cancel requested || normal exit path */
3723 return atomic_read(&fs_info
->balance_cancel_req
) ||
3724 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
3725 atomic_read(&fs_info
->balance_cancel_req
) == 0);
3728 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
3732 unset_balance_control(fs_info
);
3733 ret
= del_balance_item(fs_info
);
3735 btrfs_handle_fs_error(fs_info
, ret
, NULL
);
3737 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3740 /* Non-zero return value signifies invalidity */
3741 static inline int validate_convert_profile(struct btrfs_balance_args
*bctl_arg
,
3744 return ((bctl_arg
->flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3745 (!alloc_profile_is_valid(bctl_arg
->target
, 1) ||
3746 (bctl_arg
->target
& ~allowed
)));
3750 * Should be called with both balance and volume mutexes held
3752 int btrfs_balance(struct btrfs_balance_control
*bctl
,
3753 struct btrfs_ioctl_balance_args
*bargs
)
3755 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3762 if (btrfs_fs_closing(fs_info
) ||
3763 atomic_read(&fs_info
->balance_pause_req
) ||
3764 atomic_read(&fs_info
->balance_cancel_req
)) {
3769 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
3770 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
3774 * In case of mixed groups both data and meta should be picked,
3775 * and identical options should be given for both of them.
3777 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
3778 if (mixed
&& (bctl
->flags
& allowed
)) {
3779 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
3780 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
3781 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
3783 "with mixed groups data and metadata balance options must be the same");
3789 num_devices
= fs_info
->fs_devices
->num_devices
;
3790 btrfs_dev_replace_lock(&fs_info
->dev_replace
, 0);
3791 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
3792 BUG_ON(num_devices
< 1);
3795 btrfs_dev_replace_unlock(&fs_info
->dev_replace
, 0);
3796 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
| BTRFS_BLOCK_GROUP_DUP
;
3797 if (num_devices
> 1)
3798 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
3799 if (num_devices
> 2)
3800 allowed
|= BTRFS_BLOCK_GROUP_RAID5
;
3801 if (num_devices
> 3)
3802 allowed
|= (BTRFS_BLOCK_GROUP_RAID10
|
3803 BTRFS_BLOCK_GROUP_RAID6
);
3804 if (validate_convert_profile(&bctl
->data
, allowed
)) {
3806 "unable to start balance with target data profile %llu",
3811 if (validate_convert_profile(&bctl
->meta
, allowed
)) {
3813 "unable to start balance with target metadata profile %llu",
3818 if (validate_convert_profile(&bctl
->sys
, allowed
)) {
3820 "unable to start balance with target system profile %llu",
3826 /* allow to reduce meta or sys integrity only if force set */
3827 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3828 BTRFS_BLOCK_GROUP_RAID10
|
3829 BTRFS_BLOCK_GROUP_RAID5
|
3830 BTRFS_BLOCK_GROUP_RAID6
;
3832 seq
= read_seqbegin(&fs_info
->profiles_lock
);
3834 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3835 (fs_info
->avail_system_alloc_bits
& allowed
) &&
3836 !(bctl
->sys
.target
& allowed
)) ||
3837 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3838 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
3839 !(bctl
->meta
.target
& allowed
))) {
3840 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
3842 "force reducing metadata integrity");
3845 "balance will reduce metadata integrity, use force if you want this");
3850 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
3852 if (btrfs_get_num_tolerated_disk_barrier_failures(bctl
->meta
.target
) <
3853 btrfs_get_num_tolerated_disk_barrier_failures(bctl
->data
.target
)) {
3855 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3856 bctl
->meta
.target
, bctl
->data
.target
);
3859 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3860 fs_info
->num_tolerated_disk_barrier_failures
= min(
3861 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
),
3862 btrfs_get_num_tolerated_disk_barrier_failures(
3866 ret
= insert_balance_item(fs_info
, bctl
);
3867 if (ret
&& ret
!= -EEXIST
)
3870 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
3871 BUG_ON(ret
== -EEXIST
);
3872 set_balance_control(bctl
);
3874 BUG_ON(ret
!= -EEXIST
);
3875 spin_lock(&fs_info
->balance_lock
);
3876 update_balance_args(bctl
);
3877 spin_unlock(&fs_info
->balance_lock
);
3880 atomic_inc(&fs_info
->balance_running
);
3881 mutex_unlock(&fs_info
->balance_mutex
);
3883 ret
= __btrfs_balance(fs_info
);
3885 mutex_lock(&fs_info
->balance_mutex
);
3886 atomic_dec(&fs_info
->balance_running
);
3888 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3889 fs_info
->num_tolerated_disk_barrier_failures
=
3890 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3894 memset(bargs
, 0, sizeof(*bargs
));
3895 update_ioctl_balance_args(fs_info
, 0, bargs
);
3898 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
3899 balance_need_close(fs_info
)) {
3900 __cancel_balance(fs_info
);
3903 wake_up(&fs_info
->balance_wait_q
);
3907 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
3908 __cancel_balance(fs_info
);
3911 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3916 static int balance_kthread(void *data
)
3918 struct btrfs_fs_info
*fs_info
= data
;
3921 mutex_lock(&fs_info
->volume_mutex
);
3922 mutex_lock(&fs_info
->balance_mutex
);
3924 if (fs_info
->balance_ctl
) {
3925 btrfs_info(fs_info
, "continuing balance");
3926 ret
= btrfs_balance(fs_info
->balance_ctl
, NULL
);
3929 mutex_unlock(&fs_info
->balance_mutex
);
3930 mutex_unlock(&fs_info
->volume_mutex
);
3935 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
3937 struct task_struct
*tsk
;
3939 spin_lock(&fs_info
->balance_lock
);
3940 if (!fs_info
->balance_ctl
) {
3941 spin_unlock(&fs_info
->balance_lock
);
3944 spin_unlock(&fs_info
->balance_lock
);
3946 if (btrfs_test_opt(fs_info
, SKIP_BALANCE
)) {
3947 btrfs_info(fs_info
, "force skipping balance");
3951 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
3952 return PTR_ERR_OR_ZERO(tsk
);
3955 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
3957 struct btrfs_balance_control
*bctl
;
3958 struct btrfs_balance_item
*item
;
3959 struct btrfs_disk_balance_args disk_bargs
;
3960 struct btrfs_path
*path
;
3961 struct extent_buffer
*leaf
;
3962 struct btrfs_key key
;
3965 path
= btrfs_alloc_path();
3969 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3970 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3973 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
3976 if (ret
> 0) { /* ret = -ENOENT; */
3981 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
3987 leaf
= path
->nodes
[0];
3988 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3990 bctl
->fs_info
= fs_info
;
3991 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
3992 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
3994 btrfs_balance_data(leaf
, item
, &disk_bargs
);
3995 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
3996 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
3997 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
3998 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
3999 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
4001 WARN_ON(atomic_xchg(&fs_info
->mutually_exclusive_operation_running
, 1));
4003 mutex_lock(&fs_info
->volume_mutex
);
4004 mutex_lock(&fs_info
->balance_mutex
);
4006 set_balance_control(bctl
);
4008 mutex_unlock(&fs_info
->balance_mutex
);
4009 mutex_unlock(&fs_info
->volume_mutex
);
4011 btrfs_free_path(path
);
4015 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
4019 mutex_lock(&fs_info
->balance_mutex
);
4020 if (!fs_info
->balance_ctl
) {
4021 mutex_unlock(&fs_info
->balance_mutex
);
4025 if (atomic_read(&fs_info
->balance_running
)) {
4026 atomic_inc(&fs_info
->balance_pause_req
);
4027 mutex_unlock(&fs_info
->balance_mutex
);
4029 wait_event(fs_info
->balance_wait_q
,
4030 atomic_read(&fs_info
->balance_running
) == 0);
4032 mutex_lock(&fs_info
->balance_mutex
);
4033 /* we are good with balance_ctl ripped off from under us */
4034 BUG_ON(atomic_read(&fs_info
->balance_running
));
4035 atomic_dec(&fs_info
->balance_pause_req
);
4040 mutex_unlock(&fs_info
->balance_mutex
);
4044 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
4046 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
4049 mutex_lock(&fs_info
->balance_mutex
);
4050 if (!fs_info
->balance_ctl
) {
4051 mutex_unlock(&fs_info
->balance_mutex
);
4055 atomic_inc(&fs_info
->balance_cancel_req
);
4057 * if we are running just wait and return, balance item is
4058 * deleted in btrfs_balance in this case
4060 if (atomic_read(&fs_info
->balance_running
)) {
4061 mutex_unlock(&fs_info
->balance_mutex
);
4062 wait_event(fs_info
->balance_wait_q
,
4063 atomic_read(&fs_info
->balance_running
) == 0);
4064 mutex_lock(&fs_info
->balance_mutex
);
4066 /* __cancel_balance needs volume_mutex */
4067 mutex_unlock(&fs_info
->balance_mutex
);
4068 mutex_lock(&fs_info
->volume_mutex
);
4069 mutex_lock(&fs_info
->balance_mutex
);
4071 if (fs_info
->balance_ctl
)
4072 __cancel_balance(fs_info
);
4074 mutex_unlock(&fs_info
->volume_mutex
);
4077 BUG_ON(fs_info
->balance_ctl
|| atomic_read(&fs_info
->balance_running
));
4078 atomic_dec(&fs_info
->balance_cancel_req
);
4079 mutex_unlock(&fs_info
->balance_mutex
);
4083 static int btrfs_uuid_scan_kthread(void *data
)
4085 struct btrfs_fs_info
*fs_info
= data
;
4086 struct btrfs_root
*root
= fs_info
->tree_root
;
4087 struct btrfs_key key
;
4088 struct btrfs_key max_key
;
4089 struct btrfs_path
*path
= NULL
;
4091 struct extent_buffer
*eb
;
4093 struct btrfs_root_item root_item
;
4095 struct btrfs_trans_handle
*trans
= NULL
;
4097 path
= btrfs_alloc_path();
4104 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4107 max_key
.objectid
= (u64
)-1;
4108 max_key
.type
= BTRFS_ROOT_ITEM_KEY
;
4109 max_key
.offset
= (u64
)-1;
4112 ret
= btrfs_search_forward(root
, &key
, path
, 0);
4119 if (key
.type
!= BTRFS_ROOT_ITEM_KEY
||
4120 (key
.objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
4121 key
.objectid
!= BTRFS_FS_TREE_OBJECTID
) ||
4122 key
.objectid
> BTRFS_LAST_FREE_OBJECTID
)
4125 eb
= path
->nodes
[0];
4126 slot
= path
->slots
[0];
4127 item_size
= btrfs_item_size_nr(eb
, slot
);
4128 if (item_size
< sizeof(root_item
))
4131 read_extent_buffer(eb
, &root_item
,
4132 btrfs_item_ptr_offset(eb
, slot
),
4133 (int)sizeof(root_item
));
4134 if (btrfs_root_refs(&root_item
) == 0)
4137 if (!btrfs_is_empty_uuid(root_item
.uuid
) ||
4138 !btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4142 btrfs_release_path(path
);
4144 * 1 - subvol uuid item
4145 * 1 - received_subvol uuid item
4147 trans
= btrfs_start_transaction(fs_info
->uuid_root
, 2);
4148 if (IS_ERR(trans
)) {
4149 ret
= PTR_ERR(trans
);
4157 if (!btrfs_is_empty_uuid(root_item
.uuid
)) {
4158 ret
= btrfs_uuid_tree_add(trans
, fs_info
,
4160 BTRFS_UUID_KEY_SUBVOL
,
4163 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4169 if (!btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4170 ret
= btrfs_uuid_tree_add(trans
, fs_info
,
4171 root_item
.received_uuid
,
4172 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
4175 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4183 ret
= btrfs_end_transaction(trans
);
4189 btrfs_release_path(path
);
4190 if (key
.offset
< (u64
)-1) {
4192 } else if (key
.type
< BTRFS_ROOT_ITEM_KEY
) {
4194 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4195 } else if (key
.objectid
< (u64
)-1) {
4197 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4206 btrfs_free_path(path
);
4207 if (trans
&& !IS_ERR(trans
))
4208 btrfs_end_transaction(trans
);
4210 btrfs_warn(fs_info
, "btrfs_uuid_scan_kthread failed %d", ret
);
4212 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN
, &fs_info
->flags
);
4213 up(&fs_info
->uuid_tree_rescan_sem
);
4218 * Callback for btrfs_uuid_tree_iterate().
4220 * 0 check succeeded, the entry is not outdated.
4221 * < 0 if an error occurred.
4222 * > 0 if the check failed, which means the caller shall remove the entry.
4224 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info
*fs_info
,
4225 u8
*uuid
, u8 type
, u64 subid
)
4227 struct btrfs_key key
;
4229 struct btrfs_root
*subvol_root
;
4231 if (type
!= BTRFS_UUID_KEY_SUBVOL
&&
4232 type
!= BTRFS_UUID_KEY_RECEIVED_SUBVOL
)
4235 key
.objectid
= subid
;
4236 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4237 key
.offset
= (u64
)-1;
4238 subvol_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
4239 if (IS_ERR(subvol_root
)) {
4240 ret
= PTR_ERR(subvol_root
);
4247 case BTRFS_UUID_KEY_SUBVOL
:
4248 if (memcmp(uuid
, subvol_root
->root_item
.uuid
, BTRFS_UUID_SIZE
))
4251 case BTRFS_UUID_KEY_RECEIVED_SUBVOL
:
4252 if (memcmp(uuid
, subvol_root
->root_item
.received_uuid
,
4262 static int btrfs_uuid_rescan_kthread(void *data
)
4264 struct btrfs_fs_info
*fs_info
= (struct btrfs_fs_info
*)data
;
4268 * 1st step is to iterate through the existing UUID tree and
4269 * to delete all entries that contain outdated data.
4270 * 2nd step is to add all missing entries to the UUID tree.
4272 ret
= btrfs_uuid_tree_iterate(fs_info
, btrfs_check_uuid_tree_entry
);
4274 btrfs_warn(fs_info
, "iterating uuid_tree failed %d", ret
);
4275 up(&fs_info
->uuid_tree_rescan_sem
);
4278 return btrfs_uuid_scan_kthread(data
);
4281 int btrfs_create_uuid_tree(struct btrfs_fs_info
*fs_info
)
4283 struct btrfs_trans_handle
*trans
;
4284 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
4285 struct btrfs_root
*uuid_root
;
4286 struct task_struct
*task
;
4293 trans
= btrfs_start_transaction(tree_root
, 2);
4295 return PTR_ERR(trans
);
4297 uuid_root
= btrfs_create_tree(trans
, fs_info
,
4298 BTRFS_UUID_TREE_OBJECTID
);
4299 if (IS_ERR(uuid_root
)) {
4300 ret
= PTR_ERR(uuid_root
);
4301 btrfs_abort_transaction(trans
, ret
);
4302 btrfs_end_transaction(trans
);
4306 fs_info
->uuid_root
= uuid_root
;
4308 ret
= btrfs_commit_transaction(trans
);
4312 down(&fs_info
->uuid_tree_rescan_sem
);
4313 task
= kthread_run(btrfs_uuid_scan_kthread
, fs_info
, "btrfs-uuid");
4315 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4316 btrfs_warn(fs_info
, "failed to start uuid_scan task");
4317 up(&fs_info
->uuid_tree_rescan_sem
);
4318 return PTR_ERR(task
);
4324 int btrfs_check_uuid_tree(struct btrfs_fs_info
*fs_info
)
4326 struct task_struct
*task
;
4328 down(&fs_info
->uuid_tree_rescan_sem
);
4329 task
= kthread_run(btrfs_uuid_rescan_kthread
, fs_info
, "btrfs-uuid");
4331 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4332 btrfs_warn(fs_info
, "failed to start uuid_rescan task");
4333 up(&fs_info
->uuid_tree_rescan_sem
);
4334 return PTR_ERR(task
);
4341 * shrinking a device means finding all of the device extents past
4342 * the new size, and then following the back refs to the chunks.
4343 * The chunk relocation code actually frees the device extent
4345 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
4347 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
4348 struct btrfs_root
*root
= fs_info
->dev_root
;
4349 struct btrfs_trans_handle
*trans
;
4350 struct btrfs_dev_extent
*dev_extent
= NULL
;
4351 struct btrfs_path
*path
;
4357 bool retried
= false;
4358 bool checked_pending_chunks
= false;
4359 struct extent_buffer
*l
;
4360 struct btrfs_key key
;
4361 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
4362 u64 old_total
= btrfs_super_total_bytes(super_copy
);
4363 u64 old_size
= btrfs_device_get_total_bytes(device
);
4364 u64 diff
= old_size
- new_size
;
4366 if (device
->is_tgtdev_for_dev_replace
)
4369 path
= btrfs_alloc_path();
4373 path
->reada
= READA_FORWARD
;
4375 mutex_lock(&fs_info
->chunk_mutex
);
4377 btrfs_device_set_total_bytes(device
, new_size
);
4378 if (device
->writeable
) {
4379 device
->fs_devices
->total_rw_bytes
-= diff
;
4380 spin_lock(&fs_info
->free_chunk_lock
);
4381 fs_info
->free_chunk_space
-= diff
;
4382 spin_unlock(&fs_info
->free_chunk_lock
);
4384 mutex_unlock(&fs_info
->chunk_mutex
);
4387 key
.objectid
= device
->devid
;
4388 key
.offset
= (u64
)-1;
4389 key
.type
= BTRFS_DEV_EXTENT_KEY
;
4392 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
4393 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4395 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4399 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
4401 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4406 btrfs_release_path(path
);
4411 slot
= path
->slots
[0];
4412 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
4414 if (key
.objectid
!= device
->devid
) {
4415 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4416 btrfs_release_path(path
);
4420 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
4421 length
= btrfs_dev_extent_length(l
, dev_extent
);
4423 if (key
.offset
+ length
<= new_size
) {
4424 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4425 btrfs_release_path(path
);
4429 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
4430 btrfs_release_path(path
);
4432 ret
= btrfs_relocate_chunk(fs_info
, chunk_offset
);
4433 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4434 if (ret
&& ret
!= -ENOSPC
)
4438 } while (key
.offset
-- > 0);
4440 if (failed
&& !retried
) {
4444 } else if (failed
&& retried
) {
4449 /* Shrinking succeeded, else we would be at "done". */
4450 trans
= btrfs_start_transaction(root
, 0);
4451 if (IS_ERR(trans
)) {
4452 ret
= PTR_ERR(trans
);
4456 mutex_lock(&fs_info
->chunk_mutex
);
4459 * We checked in the above loop all device extents that were already in
4460 * the device tree. However before we have updated the device's
4461 * total_bytes to the new size, we might have had chunk allocations that
4462 * have not complete yet (new block groups attached to transaction
4463 * handles), and therefore their device extents were not yet in the
4464 * device tree and we missed them in the loop above. So if we have any
4465 * pending chunk using a device extent that overlaps the device range
4466 * that we can not use anymore, commit the current transaction and
4467 * repeat the search on the device tree - this way we guarantee we will
4468 * not have chunks using device extents that end beyond 'new_size'.
4470 if (!checked_pending_chunks
) {
4471 u64 start
= new_size
;
4472 u64 len
= old_size
- new_size
;
4474 if (contains_pending_extent(trans
->transaction
, device
,
4476 mutex_unlock(&fs_info
->chunk_mutex
);
4477 checked_pending_chunks
= true;
4480 ret
= btrfs_commit_transaction(trans
);
4487 btrfs_device_set_disk_total_bytes(device
, new_size
);
4488 if (list_empty(&device
->resized_list
))
4489 list_add_tail(&device
->resized_list
,
4490 &fs_info
->fs_devices
->resized_devices
);
4492 WARN_ON(diff
> old_total
);
4493 btrfs_set_super_total_bytes(super_copy
, old_total
- diff
);
4494 mutex_unlock(&fs_info
->chunk_mutex
);
4496 /* Now btrfs_update_device() will change the on-disk size. */
4497 ret
= btrfs_update_device(trans
, device
);
4498 btrfs_end_transaction(trans
);
4500 btrfs_free_path(path
);
4502 mutex_lock(&fs_info
->chunk_mutex
);
4503 btrfs_device_set_total_bytes(device
, old_size
);
4504 if (device
->writeable
)
4505 device
->fs_devices
->total_rw_bytes
+= diff
;
4506 spin_lock(&fs_info
->free_chunk_lock
);
4507 fs_info
->free_chunk_space
+= diff
;
4508 spin_unlock(&fs_info
->free_chunk_lock
);
4509 mutex_unlock(&fs_info
->chunk_mutex
);
4514 static int btrfs_add_system_chunk(struct btrfs_fs_info
*fs_info
,
4515 struct btrfs_key
*key
,
4516 struct btrfs_chunk
*chunk
, int item_size
)
4518 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
4519 struct btrfs_disk_key disk_key
;
4523 mutex_lock(&fs_info
->chunk_mutex
);
4524 array_size
= btrfs_super_sys_array_size(super_copy
);
4525 if (array_size
+ item_size
+ sizeof(disk_key
)
4526 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4527 mutex_unlock(&fs_info
->chunk_mutex
);
4531 ptr
= super_copy
->sys_chunk_array
+ array_size
;
4532 btrfs_cpu_key_to_disk(&disk_key
, key
);
4533 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
4534 ptr
+= sizeof(disk_key
);
4535 memcpy(ptr
, chunk
, item_size
);
4536 item_size
+= sizeof(disk_key
);
4537 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
4538 mutex_unlock(&fs_info
->chunk_mutex
);
4544 * sort the devices in descending order by max_avail, total_avail
4546 static int btrfs_cmp_device_info(const void *a
, const void *b
)
4548 const struct btrfs_device_info
*di_a
= a
;
4549 const struct btrfs_device_info
*di_b
= b
;
4551 if (di_a
->max_avail
> di_b
->max_avail
)
4553 if (di_a
->max_avail
< di_b
->max_avail
)
4555 if (di_a
->total_avail
> di_b
->total_avail
)
4557 if (di_a
->total_avail
< di_b
->total_avail
)
4562 static u32
find_raid56_stripe_len(u32 data_devices
, u32 dev_stripe_target
)
4564 /* TODO allow them to set a preferred stripe size */
4568 static void check_raid56_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
4570 if (!(type
& BTRFS_BLOCK_GROUP_RAID56_MASK
))
4573 btrfs_set_fs_incompat(info
, RAID56
);
4576 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info) \
4577 - sizeof(struct btrfs_chunk)) \
4578 / sizeof(struct btrfs_stripe) + 1)
4580 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4581 - 2 * sizeof(struct btrfs_disk_key) \
4582 - 2 * sizeof(struct btrfs_chunk)) \
4583 / sizeof(struct btrfs_stripe) + 1)
4585 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4586 u64 start
, u64 type
)
4588 struct btrfs_fs_info
*info
= trans
->fs_info
;
4589 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
4590 struct list_head
*cur
;
4591 struct map_lookup
*map
= NULL
;
4592 struct extent_map_tree
*em_tree
;
4593 struct extent_map
*em
;
4594 struct btrfs_device_info
*devices_info
= NULL
;
4596 int num_stripes
; /* total number of stripes to allocate */
4597 int data_stripes
; /* number of stripes that count for
4599 int sub_stripes
; /* sub_stripes info for map */
4600 int dev_stripes
; /* stripes per dev */
4601 int devs_max
; /* max devs to use */
4602 int devs_min
; /* min devs needed */
4603 int devs_increment
; /* ndevs has to be a multiple of this */
4604 int ncopies
; /* how many copies to data has */
4606 u64 max_stripe_size
;
4610 u64 raid_stripe_len
= BTRFS_STRIPE_LEN
;
4616 BUG_ON(!alloc_profile_is_valid(type
, 0));
4618 if (list_empty(&fs_devices
->alloc_list
))
4621 index
= __get_raid_index(type
);
4623 sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
4624 dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
4625 devs_max
= btrfs_raid_array
[index
].devs_max
;
4626 devs_min
= btrfs_raid_array
[index
].devs_min
;
4627 devs_increment
= btrfs_raid_array
[index
].devs_increment
;
4628 ncopies
= btrfs_raid_array
[index
].ncopies
;
4630 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
4631 max_stripe_size
= SZ_1G
;
4632 max_chunk_size
= 10 * max_stripe_size
;
4634 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4635 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
4636 /* for larger filesystems, use larger metadata chunks */
4637 if (fs_devices
->total_rw_bytes
> 50ULL * SZ_1G
)
4638 max_stripe_size
= SZ_1G
;
4640 max_stripe_size
= SZ_256M
;
4641 max_chunk_size
= max_stripe_size
;
4643 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4644 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4645 max_stripe_size
= SZ_32M
;
4646 max_chunk_size
= 2 * max_stripe_size
;
4648 devs_max
= BTRFS_MAX_DEVS_SYS_CHUNK
;
4650 btrfs_err(info
, "invalid chunk type 0x%llx requested",
4655 /* we don't want a chunk larger than 10% of writeable space */
4656 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
4659 devices_info
= kcalloc(fs_devices
->rw_devices
, sizeof(*devices_info
),
4664 cur
= fs_devices
->alloc_list
.next
;
4667 * in the first pass through the devices list, we gather information
4668 * about the available holes on each device.
4671 while (cur
!= &fs_devices
->alloc_list
) {
4672 struct btrfs_device
*device
;
4676 device
= list_entry(cur
, struct btrfs_device
, dev_alloc_list
);
4680 if (!device
->writeable
) {
4682 "BTRFS: read-only device in alloc_list\n");
4686 if (!device
->in_fs_metadata
||
4687 device
->is_tgtdev_for_dev_replace
)
4690 if (device
->total_bytes
> device
->bytes_used
)
4691 total_avail
= device
->total_bytes
- device
->bytes_used
;
4695 /* If there is no space on this device, skip it. */
4696 if (total_avail
== 0)
4699 ret
= find_free_dev_extent(trans
, device
,
4700 max_stripe_size
* dev_stripes
,
4701 &dev_offset
, &max_avail
);
4702 if (ret
&& ret
!= -ENOSPC
)
4706 max_avail
= max_stripe_size
* dev_stripes
;
4708 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
4711 if (ndevs
== fs_devices
->rw_devices
) {
4712 WARN(1, "%s: found more than %llu devices\n",
4713 __func__
, fs_devices
->rw_devices
);
4716 devices_info
[ndevs
].dev_offset
= dev_offset
;
4717 devices_info
[ndevs
].max_avail
= max_avail
;
4718 devices_info
[ndevs
].total_avail
= total_avail
;
4719 devices_info
[ndevs
].dev
= device
;
4724 * now sort the devices by hole size / available space
4726 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
4727 btrfs_cmp_device_info
, NULL
);
4729 /* round down to number of usable stripes */
4730 ndevs
-= ndevs
% devs_increment
;
4732 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
4737 if (devs_max
&& ndevs
> devs_max
)
4740 * the primary goal is to maximize the number of stripes, so use as many
4741 * devices as possible, even if the stripes are not maximum sized.
4743 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4744 num_stripes
= ndevs
* dev_stripes
;
4747 * this will have to be fixed for RAID1 and RAID10 over
4750 data_stripes
= num_stripes
/ ncopies
;
4752 if (type
& BTRFS_BLOCK_GROUP_RAID5
) {
4753 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 1,
4755 data_stripes
= num_stripes
- 1;
4757 if (type
& BTRFS_BLOCK_GROUP_RAID6
) {
4758 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 2,
4760 data_stripes
= num_stripes
- 2;
4764 * Use the number of data stripes to figure out how big this chunk
4765 * is really going to be in terms of logical address space,
4766 * and compare that answer with the max chunk size
4768 if (stripe_size
* data_stripes
> max_chunk_size
) {
4769 u64 mask
= (1ULL << 24) - 1;
4771 stripe_size
= div_u64(max_chunk_size
, data_stripes
);
4773 /* bump the answer up to a 16MB boundary */
4774 stripe_size
= (stripe_size
+ mask
) & ~mask
;
4776 /* but don't go higher than the limits we found
4777 * while searching for free extents
4779 if (stripe_size
> devices_info
[ndevs
-1].max_avail
)
4780 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4783 stripe_size
= div_u64(stripe_size
, dev_stripes
);
4785 /* align to BTRFS_STRIPE_LEN */
4786 stripe_size
= div_u64(stripe_size
, raid_stripe_len
);
4787 stripe_size
*= raid_stripe_len
;
4789 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
4794 map
->num_stripes
= num_stripes
;
4796 for (i
= 0; i
< ndevs
; ++i
) {
4797 for (j
= 0; j
< dev_stripes
; ++j
) {
4798 int s
= i
* dev_stripes
+ j
;
4799 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
4800 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
4804 map
->sector_size
= info
->sectorsize
;
4805 map
->stripe_len
= raid_stripe_len
;
4806 map
->io_align
= raid_stripe_len
;
4807 map
->io_width
= raid_stripe_len
;
4809 map
->sub_stripes
= sub_stripes
;
4811 num_bytes
= stripe_size
* data_stripes
;
4813 trace_btrfs_chunk_alloc(info
, map
, start
, num_bytes
);
4815 em
= alloc_extent_map();
4821 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
4822 em
->map_lookup
= map
;
4824 em
->len
= num_bytes
;
4825 em
->block_start
= 0;
4826 em
->block_len
= em
->len
;
4827 em
->orig_block_len
= stripe_size
;
4829 em_tree
= &info
->mapping_tree
.map_tree
;
4830 write_lock(&em_tree
->lock
);
4831 ret
= add_extent_mapping(em_tree
, em
, 0);
4833 list_add_tail(&em
->list
, &trans
->transaction
->pending_chunks
);
4834 atomic_inc(&em
->refs
);
4836 write_unlock(&em_tree
->lock
);
4838 free_extent_map(em
);
4842 ret
= btrfs_make_block_group(trans
, info
, 0, type
,
4843 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4846 goto error_del_extent
;
4848 for (i
= 0; i
< map
->num_stripes
; i
++) {
4849 num_bytes
= map
->stripes
[i
].dev
->bytes_used
+ stripe_size
;
4850 btrfs_device_set_bytes_used(map
->stripes
[i
].dev
, num_bytes
);
4853 spin_lock(&info
->free_chunk_lock
);
4854 info
->free_chunk_space
-= (stripe_size
* map
->num_stripes
);
4855 spin_unlock(&info
->free_chunk_lock
);
4857 free_extent_map(em
);
4858 check_raid56_incompat_flag(info
, type
);
4860 kfree(devices_info
);
4864 write_lock(&em_tree
->lock
);
4865 remove_extent_mapping(em_tree
, em
);
4866 write_unlock(&em_tree
->lock
);
4868 /* One for our allocation */
4869 free_extent_map(em
);
4870 /* One for the tree reference */
4871 free_extent_map(em
);
4872 /* One for the pending_chunks list reference */
4873 free_extent_map(em
);
4875 kfree(devices_info
);
4879 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
4880 struct btrfs_fs_info
*fs_info
,
4881 u64 chunk_offset
, u64 chunk_size
)
4883 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
4884 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
4885 struct btrfs_key key
;
4886 struct btrfs_device
*device
;
4887 struct btrfs_chunk
*chunk
;
4888 struct btrfs_stripe
*stripe
;
4889 struct extent_map_tree
*em_tree
;
4890 struct extent_map
*em
;
4891 struct map_lookup
*map
;
4898 em_tree
= &fs_info
->mapping_tree
.map_tree
;
4899 read_lock(&em_tree
->lock
);
4900 em
= lookup_extent_mapping(em_tree
, chunk_offset
, chunk_size
);
4901 read_unlock(&em_tree
->lock
);
4904 btrfs_crit(fs_info
, "unable to find logical %Lu len %Lu",
4905 chunk_offset
, chunk_size
);
4909 if (em
->start
!= chunk_offset
|| em
->len
!= chunk_size
) {
4911 "found a bad mapping, wanted %Lu-%Lu, found %Lu-%Lu",
4912 chunk_offset
, chunk_size
, em
->start
, em
->len
);
4913 free_extent_map(em
);
4917 map
= em
->map_lookup
;
4918 item_size
= btrfs_chunk_item_size(map
->num_stripes
);
4919 stripe_size
= em
->orig_block_len
;
4921 chunk
= kzalloc(item_size
, GFP_NOFS
);
4928 * Take the device list mutex to prevent races with the final phase of
4929 * a device replace operation that replaces the device object associated
4930 * with the map's stripes, because the device object's id can change
4931 * at any time during that final phase of the device replace operation
4932 * (dev-replace.c:btrfs_dev_replace_finishing()).
4934 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
4935 for (i
= 0; i
< map
->num_stripes
; i
++) {
4936 device
= map
->stripes
[i
].dev
;
4937 dev_offset
= map
->stripes
[i
].physical
;
4939 ret
= btrfs_update_device(trans
, device
);
4942 ret
= btrfs_alloc_dev_extent(trans
, device
,
4943 chunk_root
->root_key
.objectid
,
4944 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4945 chunk_offset
, dev_offset
,
4951 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
4955 stripe
= &chunk
->stripe
;
4956 for (i
= 0; i
< map
->num_stripes
; i
++) {
4957 device
= map
->stripes
[i
].dev
;
4958 dev_offset
= map
->stripes
[i
].physical
;
4960 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
4961 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
4962 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
4965 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
4967 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
4968 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
4969 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
4970 btrfs_set_stack_chunk_type(chunk
, map
->type
);
4971 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
4972 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
4973 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
4974 btrfs_set_stack_chunk_sector_size(chunk
, fs_info
->sectorsize
);
4975 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
4977 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
4978 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
4979 key
.offset
= chunk_offset
;
4981 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
4982 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4984 * TODO: Cleanup of inserted chunk root in case of
4987 ret
= btrfs_add_system_chunk(fs_info
, &key
, chunk
, item_size
);
4992 free_extent_map(em
);
4997 * Chunk allocation falls into two parts. The first part does works
4998 * that make the new allocated chunk useable, but not do any operation
4999 * that modifies the chunk tree. The second part does the works that
5000 * require modifying the chunk tree. This division is important for the
5001 * bootstrap process of adding storage to a seed btrfs.
5003 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
5004 struct btrfs_fs_info
*fs_info
, u64 type
)
5008 ASSERT(mutex_is_locked(&fs_info
->chunk_mutex
));
5009 chunk_offset
= find_next_chunk(fs_info
);
5010 return __btrfs_alloc_chunk(trans
, chunk_offset
, type
);
5013 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
5014 struct btrfs_fs_info
*fs_info
)
5016 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
5018 u64 sys_chunk_offset
;
5022 chunk_offset
= find_next_chunk(fs_info
);
5023 alloc_profile
= btrfs_get_alloc_profile(extent_root
, 0);
5024 ret
= __btrfs_alloc_chunk(trans
, chunk_offset
, alloc_profile
);
5028 sys_chunk_offset
= find_next_chunk(fs_info
);
5029 alloc_profile
= btrfs_get_alloc_profile(fs_info
->chunk_root
, 0);
5030 ret
= __btrfs_alloc_chunk(trans
, sys_chunk_offset
, alloc_profile
);
5034 static inline int btrfs_chunk_max_errors(struct map_lookup
*map
)
5038 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
5039 BTRFS_BLOCK_GROUP_RAID10
|
5040 BTRFS_BLOCK_GROUP_RAID5
|
5041 BTRFS_BLOCK_GROUP_DUP
)) {
5043 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
) {
5052 int btrfs_chunk_readonly(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
5054 struct extent_map
*em
;
5055 struct map_lookup
*map
;
5056 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
5061 read_lock(&map_tree
->map_tree
.lock
);
5062 em
= lookup_extent_mapping(&map_tree
->map_tree
, chunk_offset
, 1);
5063 read_unlock(&map_tree
->map_tree
.lock
);
5067 map
= em
->map_lookup
;
5068 for (i
= 0; i
< map
->num_stripes
; i
++) {
5069 if (map
->stripes
[i
].dev
->missing
) {
5074 if (!map
->stripes
[i
].dev
->writeable
) {
5081 * If the number of missing devices is larger than max errors,
5082 * we can not write the data into that chunk successfully, so
5085 if (miss_ndevs
> btrfs_chunk_max_errors(map
))
5088 free_extent_map(em
);
5092 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
5094 extent_map_tree_init(&tree
->map_tree
);
5097 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
5099 struct extent_map
*em
;
5102 write_lock(&tree
->map_tree
.lock
);
5103 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
5105 remove_extent_mapping(&tree
->map_tree
, em
);
5106 write_unlock(&tree
->map_tree
.lock
);
5110 free_extent_map(em
);
5111 /* once for the tree */
5112 free_extent_map(em
);
5116 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
5118 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
5119 struct extent_map
*em
;
5120 struct map_lookup
*map
;
5121 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5124 read_lock(&em_tree
->lock
);
5125 em
= lookup_extent_mapping(em_tree
, logical
, len
);
5126 read_unlock(&em_tree
->lock
);
5129 * We could return errors for these cases, but that could get ugly and
5130 * we'd probably do the same thing which is just not do anything else
5131 * and exit, so return 1 so the callers don't try to use other copies.
5134 btrfs_crit(fs_info
, "No mapping for %Lu-%Lu", logical
,
5139 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
5140 btrfs_crit(fs_info
, "Invalid mapping for %Lu-%Lu, got %Lu-%Lu",
5141 logical
, logical
+len
, em
->start
,
5142 em
->start
+ em
->len
);
5143 free_extent_map(em
);
5147 map
= em
->map_lookup
;
5148 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
5149 ret
= map
->num_stripes
;
5150 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5151 ret
= map
->sub_stripes
;
5152 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID5
)
5154 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5158 free_extent_map(em
);
5160 btrfs_dev_replace_lock(&fs_info
->dev_replace
, 0);
5161 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
))
5163 btrfs_dev_replace_unlock(&fs_info
->dev_replace
, 0);
5168 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info
*fs_info
,
5169 struct btrfs_mapping_tree
*map_tree
,
5172 struct extent_map
*em
;
5173 struct map_lookup
*map
;
5174 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5175 unsigned long len
= fs_info
->sectorsize
;
5177 read_lock(&em_tree
->lock
);
5178 em
= lookup_extent_mapping(em_tree
, logical
, len
);
5179 read_unlock(&em_tree
->lock
);
5182 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
5183 map
= em
->map_lookup
;
5184 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5185 len
= map
->stripe_len
* nr_data_stripes(map
);
5186 free_extent_map(em
);
5190 int btrfs_is_parity_mirror(struct btrfs_mapping_tree
*map_tree
,
5191 u64 logical
, u64 len
, int mirror_num
)
5193 struct extent_map
*em
;
5194 struct map_lookup
*map
;
5195 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5198 read_lock(&em_tree
->lock
);
5199 em
= lookup_extent_mapping(em_tree
, logical
, len
);
5200 read_unlock(&em_tree
->lock
);
5203 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
5204 map
= em
->map_lookup
;
5205 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5207 free_extent_map(em
);
5211 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
5212 struct map_lookup
*map
, int first
, int num
,
5213 int optimal
, int dev_replace_is_ongoing
)
5217 struct btrfs_device
*srcdev
;
5219 if (dev_replace_is_ongoing
&&
5220 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
5221 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
5222 srcdev
= fs_info
->dev_replace
.srcdev
;
5227 * try to avoid the drive that is the source drive for a
5228 * dev-replace procedure, only choose it if no other non-missing
5229 * mirror is available
5231 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
5232 if (map
->stripes
[optimal
].dev
->bdev
&&
5233 (tolerance
|| map
->stripes
[optimal
].dev
!= srcdev
))
5235 for (i
= first
; i
< first
+ num
; i
++) {
5236 if (map
->stripes
[i
].dev
->bdev
&&
5237 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
5242 /* we couldn't find one that doesn't fail. Just return something
5243 * and the io error handling code will clean up eventually
5248 static inline int parity_smaller(u64 a
, u64 b
)
5253 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5254 static void sort_parity_stripes(struct btrfs_bio
*bbio
, int num_stripes
)
5256 struct btrfs_bio_stripe s
;
5263 for (i
= 0; i
< num_stripes
- 1; i
++) {
5264 if (parity_smaller(bbio
->raid_map
[i
],
5265 bbio
->raid_map
[i
+1])) {
5266 s
= bbio
->stripes
[i
];
5267 l
= bbio
->raid_map
[i
];
5268 bbio
->stripes
[i
] = bbio
->stripes
[i
+1];
5269 bbio
->raid_map
[i
] = bbio
->raid_map
[i
+1];
5270 bbio
->stripes
[i
+1] = s
;
5271 bbio
->raid_map
[i
+1] = l
;
5279 static struct btrfs_bio
*alloc_btrfs_bio(int total_stripes
, int real_stripes
)
5281 struct btrfs_bio
*bbio
= kzalloc(
5282 /* the size of the btrfs_bio */
5283 sizeof(struct btrfs_bio
) +
5284 /* plus the variable array for the stripes */
5285 sizeof(struct btrfs_bio_stripe
) * (total_stripes
) +
5286 /* plus the variable array for the tgt dev */
5287 sizeof(int) * (real_stripes
) +
5289 * plus the raid_map, which includes both the tgt dev
5292 sizeof(u64
) * (total_stripes
),
5293 GFP_NOFS
|__GFP_NOFAIL
);
5295 atomic_set(&bbio
->error
, 0);
5296 atomic_set(&bbio
->refs
, 1);
5301 void btrfs_get_bbio(struct btrfs_bio
*bbio
)
5303 WARN_ON(!atomic_read(&bbio
->refs
));
5304 atomic_inc(&bbio
->refs
);
5307 void btrfs_put_bbio(struct btrfs_bio
*bbio
)
5311 if (atomic_dec_and_test(&bbio
->refs
))
5315 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
,
5316 enum btrfs_map_op op
,
5317 u64 logical
, u64
*length
,
5318 struct btrfs_bio
**bbio_ret
,
5319 int mirror_num
, int need_raid_map
)
5321 struct extent_map
*em
;
5322 struct map_lookup
*map
;
5323 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
5324 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5327 u64 stripe_end_offset
;
5337 int tgtdev_indexes
= 0;
5338 struct btrfs_bio
*bbio
= NULL
;
5339 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
5340 int dev_replace_is_ongoing
= 0;
5341 int num_alloc_stripes
;
5342 int patch_the_first_stripe_for_dev_replace
= 0;
5343 u64 physical_to_patch_in_first_stripe
= 0;
5344 u64 raid56_full_stripe_start
= (u64
)-1;
5346 read_lock(&em_tree
->lock
);
5347 em
= lookup_extent_mapping(em_tree
, logical
, *length
);
5348 read_unlock(&em_tree
->lock
);
5351 btrfs_crit(fs_info
, "unable to find logical %llu len %llu",
5356 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
5358 "found a bad mapping, wanted %Lu, found %Lu-%Lu",
5359 logical
, em
->start
, em
->start
+ em
->len
);
5360 free_extent_map(em
);
5364 map
= em
->map_lookup
;
5365 offset
= logical
- em
->start
;
5367 stripe_len
= map
->stripe_len
;
5370 * stripe_nr counts the total number of stripes we have to stride
5371 * to get to this block
5373 stripe_nr
= div64_u64(stripe_nr
, stripe_len
);
5375 stripe_offset
= stripe_nr
* stripe_len
;
5376 if (offset
< stripe_offset
) {
5378 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5379 stripe_offset
, offset
, em
->start
, logical
,
5381 free_extent_map(em
);
5385 /* stripe_offset is the offset of this block in its stripe*/
5386 stripe_offset
= offset
- stripe_offset
;
5388 /* if we're here for raid56, we need to know the stripe aligned start */
5389 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5390 unsigned long full_stripe_len
= stripe_len
* nr_data_stripes(map
);
5391 raid56_full_stripe_start
= offset
;
5393 /* allow a write of a full stripe, but make sure we don't
5394 * allow straddling of stripes
5396 raid56_full_stripe_start
= div64_u64(raid56_full_stripe_start
,
5398 raid56_full_stripe_start
*= full_stripe_len
;
5401 if (op
== BTRFS_MAP_DISCARD
) {
5402 /* we don't discard raid56 yet */
5403 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5407 *length
= min_t(u64
, em
->len
- offset
, *length
);
5408 } else if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
5410 /* For writes to RAID[56], allow a full stripeset across all disks.
5411 For other RAID types and for RAID[56] reads, just allow a single
5412 stripe (on a single disk). */
5413 if ((map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
5414 (op
== BTRFS_MAP_WRITE
)) {
5415 max_len
= stripe_len
* nr_data_stripes(map
) -
5416 (offset
- raid56_full_stripe_start
);
5418 /* we limit the length of each bio to what fits in a stripe */
5419 max_len
= stripe_len
- stripe_offset
;
5421 *length
= min_t(u64
, em
->len
- offset
, max_len
);
5423 *length
= em
->len
- offset
;
5426 /* This is for when we're called from btrfs_merge_bio_hook() and all
5427 it cares about is the length */
5431 btrfs_dev_replace_lock(dev_replace
, 0);
5432 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
5433 if (!dev_replace_is_ongoing
)
5434 btrfs_dev_replace_unlock(dev_replace
, 0);
5436 btrfs_dev_replace_set_lock_blocking(dev_replace
);
5438 if (dev_replace_is_ongoing
&& mirror_num
== map
->num_stripes
+ 1 &&
5439 op
!= BTRFS_MAP_WRITE
&& op
!= BTRFS_MAP_DISCARD
&&
5440 op
!= BTRFS_MAP_GET_READ_MIRRORS
&& dev_replace
->tgtdev
!= NULL
) {
5442 * in dev-replace case, for repair case (that's the only
5443 * case where the mirror is selected explicitly when
5444 * calling btrfs_map_block), blocks left of the left cursor
5445 * can also be read from the target drive.
5446 * For REQ_GET_READ_MIRRORS, the target drive is added as
5447 * the last one to the array of stripes. For READ, it also
5448 * needs to be supported using the same mirror number.
5449 * If the requested block is not left of the left cursor,
5450 * EIO is returned. This can happen because btrfs_num_copies()
5451 * returns one more in the dev-replace case.
5453 u64 tmp_length
= *length
;
5454 struct btrfs_bio
*tmp_bbio
= NULL
;
5455 int tmp_num_stripes
;
5456 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5457 int index_srcdev
= 0;
5459 u64 physical_of_found
= 0;
5461 ret
= __btrfs_map_block(fs_info
, BTRFS_MAP_GET_READ_MIRRORS
,
5462 logical
, &tmp_length
, &tmp_bbio
, 0, 0);
5464 WARN_ON(tmp_bbio
!= NULL
);
5468 tmp_num_stripes
= tmp_bbio
->num_stripes
;
5469 if (mirror_num
> tmp_num_stripes
) {
5471 * BTRFS_MAP_GET_READ_MIRRORS does not contain this
5472 * mirror, that means that the requested area
5473 * is not left of the left cursor
5476 btrfs_put_bbio(tmp_bbio
);
5481 * process the rest of the function using the mirror_num
5482 * of the source drive. Therefore look it up first.
5483 * At the end, patch the device pointer to the one of the
5486 for (i
= 0; i
< tmp_num_stripes
; i
++) {
5487 if (tmp_bbio
->stripes
[i
].dev
->devid
!= srcdev_devid
)
5491 * In case of DUP, in order to keep it simple, only add
5492 * the mirror with the lowest physical address
5495 physical_of_found
<= tmp_bbio
->stripes
[i
].physical
)
5500 physical_of_found
= tmp_bbio
->stripes
[i
].physical
;
5503 btrfs_put_bbio(tmp_bbio
);
5511 mirror_num
= index_srcdev
+ 1;
5512 patch_the_first_stripe_for_dev_replace
= 1;
5513 physical_to_patch_in_first_stripe
= physical_of_found
;
5514 } else if (mirror_num
> map
->num_stripes
) {
5520 stripe_nr_orig
= stripe_nr
;
5521 stripe_nr_end
= ALIGN(offset
+ *length
, map
->stripe_len
);
5522 stripe_nr_end
= div_u64(stripe_nr_end
, map
->stripe_len
);
5523 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
5526 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5527 if (op
== BTRFS_MAP_DISCARD
)
5528 num_stripes
= min_t(u64
, map
->num_stripes
,
5529 stripe_nr_end
- stripe_nr_orig
);
5530 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5532 if (op
!= BTRFS_MAP_WRITE
&& op
!= BTRFS_MAP_DISCARD
&&
5533 op
!= BTRFS_MAP_GET_READ_MIRRORS
)
5535 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
5536 if (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_DISCARD
||
5537 op
== BTRFS_MAP_GET_READ_MIRRORS
)
5538 num_stripes
= map
->num_stripes
;
5539 else if (mirror_num
)
5540 stripe_index
= mirror_num
- 1;
5542 stripe_index
= find_live_mirror(fs_info
, map
, 0,
5544 current
->pid
% map
->num_stripes
,
5545 dev_replace_is_ongoing
);
5546 mirror_num
= stripe_index
+ 1;
5549 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
5550 if (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_DISCARD
||
5551 op
== BTRFS_MAP_GET_READ_MIRRORS
) {
5552 num_stripes
= map
->num_stripes
;
5553 } else if (mirror_num
) {
5554 stripe_index
= mirror_num
- 1;
5559 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5560 u32 factor
= map
->num_stripes
/ map
->sub_stripes
;
5562 stripe_nr
= div_u64_rem(stripe_nr
, factor
, &stripe_index
);
5563 stripe_index
*= map
->sub_stripes
;
5565 if (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
)
5566 num_stripes
= map
->sub_stripes
;
5567 else if (op
== BTRFS_MAP_DISCARD
)
5568 num_stripes
= min_t(u64
, map
->sub_stripes
*
5569 (stripe_nr_end
- stripe_nr_orig
),
5571 else if (mirror_num
)
5572 stripe_index
+= mirror_num
- 1;
5574 int old_stripe_index
= stripe_index
;
5575 stripe_index
= find_live_mirror(fs_info
, map
,
5577 map
->sub_stripes
, stripe_index
+
5578 current
->pid
% map
->sub_stripes
,
5579 dev_replace_is_ongoing
);
5580 mirror_num
= stripe_index
- old_stripe_index
+ 1;
5583 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5584 if (need_raid_map
&&
5585 (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
||
5587 /* push stripe_nr back to the start of the full stripe */
5588 stripe_nr
= div_u64(raid56_full_stripe_start
,
5589 stripe_len
* nr_data_stripes(map
));
5591 /* RAID[56] write or recovery. Return all stripes */
5592 num_stripes
= map
->num_stripes
;
5593 max_errors
= nr_parity_stripes(map
);
5595 *length
= map
->stripe_len
;
5600 * Mirror #0 or #1 means the original data block.
5601 * Mirror #2 is RAID5 parity block.
5602 * Mirror #3 is RAID6 Q block.
5604 stripe_nr
= div_u64_rem(stripe_nr
,
5605 nr_data_stripes(map
), &stripe_index
);
5607 stripe_index
= nr_data_stripes(map
) +
5610 /* We distribute the parity blocks across stripes */
5611 div_u64_rem(stripe_nr
+ stripe_index
, map
->num_stripes
,
5613 if ((op
!= BTRFS_MAP_WRITE
&& op
!= BTRFS_MAP_DISCARD
&&
5614 op
!= BTRFS_MAP_GET_READ_MIRRORS
) && mirror_num
<= 1)
5619 * after this, stripe_nr is the number of stripes on this
5620 * device we have to walk to find the data, and stripe_index is
5621 * the number of our device in the stripe array
5623 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5625 mirror_num
= stripe_index
+ 1;
5627 if (stripe_index
>= map
->num_stripes
) {
5629 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5630 stripe_index
, map
->num_stripes
);
5635 num_alloc_stripes
= num_stripes
;
5636 if (dev_replace_is_ongoing
) {
5637 if (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_DISCARD
)
5638 num_alloc_stripes
<<= 1;
5639 if (op
== BTRFS_MAP_GET_READ_MIRRORS
)
5640 num_alloc_stripes
++;
5641 tgtdev_indexes
= num_stripes
;
5644 bbio
= alloc_btrfs_bio(num_alloc_stripes
, tgtdev_indexes
);
5649 if (dev_replace_is_ongoing
)
5650 bbio
->tgtdev_map
= (int *)(bbio
->stripes
+ num_alloc_stripes
);
5652 /* build raid_map */
5653 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
&&
5655 ((op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
) ||
5660 bbio
->raid_map
= (u64
*)((void *)bbio
->stripes
+
5661 sizeof(struct btrfs_bio_stripe
) *
5663 sizeof(int) * tgtdev_indexes
);
5665 /* Work out the disk rotation on this stripe-set */
5666 div_u64_rem(stripe_nr
, num_stripes
, &rot
);
5668 /* Fill in the logical address of each stripe */
5669 tmp
= stripe_nr
* nr_data_stripes(map
);
5670 for (i
= 0; i
< nr_data_stripes(map
); i
++)
5671 bbio
->raid_map
[(i
+rot
) % num_stripes
] =
5672 em
->start
+ (tmp
+ i
) * map
->stripe_len
;
5674 bbio
->raid_map
[(i
+rot
) % map
->num_stripes
] = RAID5_P_STRIPE
;
5675 if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5676 bbio
->raid_map
[(i
+rot
+1) % num_stripes
] =
5680 if (op
== BTRFS_MAP_DISCARD
) {
5682 u32 sub_stripes
= 0;
5683 u64 stripes_per_dev
= 0;
5684 u32 remaining_stripes
= 0;
5685 u32 last_stripe
= 0;
5688 (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID10
)) {
5689 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5692 sub_stripes
= map
->sub_stripes
;
5694 factor
= map
->num_stripes
/ sub_stripes
;
5695 stripes_per_dev
= div_u64_rem(stripe_nr_end
-
5698 &remaining_stripes
);
5699 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
5700 last_stripe
*= sub_stripes
;
5703 for (i
= 0; i
< num_stripes
; i
++) {
5704 bbio
->stripes
[i
].physical
=
5705 map
->stripes
[stripe_index
].physical
+
5706 stripe_offset
+ stripe_nr
* map
->stripe_len
;
5707 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
5709 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5710 BTRFS_BLOCK_GROUP_RAID10
)) {
5711 bbio
->stripes
[i
].length
= stripes_per_dev
*
5714 if (i
/ sub_stripes
< remaining_stripes
)
5715 bbio
->stripes
[i
].length
+=
5719 * Special for the first stripe and
5722 * |-------|...|-------|
5726 if (i
< sub_stripes
)
5727 bbio
->stripes
[i
].length
-=
5730 if (stripe_index
>= last_stripe
&&
5731 stripe_index
<= (last_stripe
+
5733 bbio
->stripes
[i
].length
-=
5736 if (i
== sub_stripes
- 1)
5739 bbio
->stripes
[i
].length
= *length
;
5742 if (stripe_index
== map
->num_stripes
) {
5743 /* This could only happen for RAID0/10 */
5749 for (i
= 0; i
< num_stripes
; i
++) {
5750 bbio
->stripes
[i
].physical
=
5751 map
->stripes
[stripe_index
].physical
+
5753 stripe_nr
* map
->stripe_len
;
5754 bbio
->stripes
[i
].dev
=
5755 map
->stripes
[stripe_index
].dev
;
5760 if (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
)
5761 max_errors
= btrfs_chunk_max_errors(map
);
5764 sort_parity_stripes(bbio
, num_stripes
);
5767 if (dev_replace_is_ongoing
&&
5768 (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_DISCARD
) &&
5769 dev_replace
->tgtdev
!= NULL
) {
5770 int index_where_to_add
;
5771 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5774 * duplicate the write operations while the dev replace
5775 * procedure is running. Since the copying of the old disk
5776 * to the new disk takes place at run time while the
5777 * filesystem is mounted writable, the regular write
5778 * operations to the old disk have to be duplicated to go
5779 * to the new disk as well.
5780 * Note that device->missing is handled by the caller, and
5781 * that the write to the old disk is already set up in the
5784 index_where_to_add
= num_stripes
;
5785 for (i
= 0; i
< num_stripes
; i
++) {
5786 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5787 /* write to new disk, too */
5788 struct btrfs_bio_stripe
*new =
5789 bbio
->stripes
+ index_where_to_add
;
5790 struct btrfs_bio_stripe
*old
=
5793 new->physical
= old
->physical
;
5794 new->length
= old
->length
;
5795 new->dev
= dev_replace
->tgtdev
;
5796 bbio
->tgtdev_map
[i
] = index_where_to_add
;
5797 index_where_to_add
++;
5802 num_stripes
= index_where_to_add
;
5803 } else if (dev_replace_is_ongoing
&&
5804 op
== BTRFS_MAP_GET_READ_MIRRORS
&&
5805 dev_replace
->tgtdev
!= NULL
) {
5806 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5807 int index_srcdev
= 0;
5809 u64 physical_of_found
= 0;
5812 * During the dev-replace procedure, the target drive can
5813 * also be used to read data in case it is needed to repair
5814 * a corrupt block elsewhere. This is possible if the
5815 * requested area is left of the left cursor. In this area,
5816 * the target drive is a full copy of the source drive.
5818 for (i
= 0; i
< num_stripes
; i
++) {
5819 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5821 * In case of DUP, in order to keep it
5822 * simple, only add the mirror with the
5823 * lowest physical address
5826 physical_of_found
<=
5827 bbio
->stripes
[i
].physical
)
5831 physical_of_found
= bbio
->stripes
[i
].physical
;
5835 struct btrfs_bio_stripe
*tgtdev_stripe
=
5836 bbio
->stripes
+ num_stripes
;
5838 tgtdev_stripe
->physical
= physical_of_found
;
5839 tgtdev_stripe
->length
=
5840 bbio
->stripes
[index_srcdev
].length
;
5841 tgtdev_stripe
->dev
= dev_replace
->tgtdev
;
5842 bbio
->tgtdev_map
[index_srcdev
] = num_stripes
;
5850 bbio
->map_type
= map
->type
;
5851 bbio
->num_stripes
= num_stripes
;
5852 bbio
->max_errors
= max_errors
;
5853 bbio
->mirror_num
= mirror_num
;
5854 bbio
->num_tgtdevs
= tgtdev_indexes
;
5857 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5858 * mirror_num == num_stripes + 1 && dev_replace target drive is
5859 * available as a mirror
5861 if (patch_the_first_stripe_for_dev_replace
&& num_stripes
> 0) {
5862 WARN_ON(num_stripes
> 1);
5863 bbio
->stripes
[0].dev
= dev_replace
->tgtdev
;
5864 bbio
->stripes
[0].physical
= physical_to_patch_in_first_stripe
;
5865 bbio
->mirror_num
= map
->num_stripes
+ 1;
5868 if (dev_replace_is_ongoing
) {
5869 btrfs_dev_replace_clear_lock_blocking(dev_replace
);
5870 btrfs_dev_replace_unlock(dev_replace
, 0);
5872 free_extent_map(em
);
5876 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, enum btrfs_map_op op
,
5877 u64 logical
, u64
*length
,
5878 struct btrfs_bio
**bbio_ret
, int mirror_num
)
5880 return __btrfs_map_block(fs_info
, op
, logical
, length
, bbio_ret
,
5884 /* For Scrub/replace */
5885 int btrfs_map_sblock(struct btrfs_fs_info
*fs_info
, enum btrfs_map_op op
,
5886 u64 logical
, u64
*length
,
5887 struct btrfs_bio
**bbio_ret
, int mirror_num
,
5890 return __btrfs_map_block(fs_info
, op
, logical
, length
, bbio_ret
,
5891 mirror_num
, need_raid_map
);
5894 int btrfs_rmap_block(struct btrfs_fs_info
*fs_info
,
5895 u64 chunk_start
, u64 physical
, u64 devid
,
5896 u64
**logical
, int *naddrs
, int *stripe_len
)
5898 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
5899 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5900 struct extent_map
*em
;
5901 struct map_lookup
*map
;
5909 read_lock(&em_tree
->lock
);
5910 em
= lookup_extent_mapping(em_tree
, chunk_start
, 1);
5911 read_unlock(&em_tree
->lock
);
5914 btrfs_err(fs_info
, "couldn't find em for chunk %Lu",
5919 if (em
->start
!= chunk_start
) {
5920 btrfs_err(fs_info
, "bad chunk start, em=%Lu, wanted=%Lu",
5921 em
->start
, chunk_start
);
5922 free_extent_map(em
);
5925 map
= em
->map_lookup
;
5928 rmap_len
= map
->stripe_len
;
5930 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5931 length
= div_u64(length
, map
->num_stripes
/ map
->sub_stripes
);
5932 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5933 length
= div_u64(length
, map
->num_stripes
);
5934 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5935 length
= div_u64(length
, nr_data_stripes(map
));
5936 rmap_len
= map
->stripe_len
* nr_data_stripes(map
);
5939 buf
= kcalloc(map
->num_stripes
, sizeof(u64
), GFP_NOFS
);
5940 BUG_ON(!buf
); /* -ENOMEM */
5942 for (i
= 0; i
< map
->num_stripes
; i
++) {
5943 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
5945 if (map
->stripes
[i
].physical
> physical
||
5946 map
->stripes
[i
].physical
+ length
<= physical
)
5949 stripe_nr
= physical
- map
->stripes
[i
].physical
;
5950 stripe_nr
= div_u64(stripe_nr
, map
->stripe_len
);
5952 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5953 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5954 stripe_nr
= div_u64(stripe_nr
, map
->sub_stripes
);
5955 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5956 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5957 } /* else if RAID[56], multiply by nr_data_stripes().
5958 * Alternatively, just use rmap_len below instead of
5959 * map->stripe_len */
5961 bytenr
= chunk_start
+ stripe_nr
* rmap_len
;
5962 WARN_ON(nr
>= map
->num_stripes
);
5963 for (j
= 0; j
< nr
; j
++) {
5964 if (buf
[j
] == bytenr
)
5968 WARN_ON(nr
>= map
->num_stripes
);
5975 *stripe_len
= rmap_len
;
5977 free_extent_map(em
);
5981 static inline void btrfs_end_bbio(struct btrfs_bio
*bbio
, struct bio
*bio
)
5983 bio
->bi_private
= bbio
->private;
5984 bio
->bi_end_io
= bbio
->end_io
;
5987 btrfs_put_bbio(bbio
);
5990 static void btrfs_end_bio(struct bio
*bio
)
5992 struct btrfs_bio
*bbio
= bio
->bi_private
;
5993 int is_orig_bio
= 0;
5995 if (bio
->bi_error
) {
5996 atomic_inc(&bbio
->error
);
5997 if (bio
->bi_error
== -EIO
|| bio
->bi_error
== -EREMOTEIO
) {
5998 unsigned int stripe_index
=
5999 btrfs_io_bio(bio
)->stripe_index
;
6000 struct btrfs_device
*dev
;
6002 BUG_ON(stripe_index
>= bbio
->num_stripes
);
6003 dev
= bbio
->stripes
[stripe_index
].dev
;
6005 if (bio_op(bio
) == REQ_OP_WRITE
)
6006 btrfs_dev_stat_inc(dev
,
6007 BTRFS_DEV_STAT_WRITE_ERRS
);
6009 btrfs_dev_stat_inc(dev
,
6010 BTRFS_DEV_STAT_READ_ERRS
);
6011 if (bio
->bi_opf
& REQ_PREFLUSH
)
6012 btrfs_dev_stat_inc(dev
,
6013 BTRFS_DEV_STAT_FLUSH_ERRS
);
6014 btrfs_dev_stat_print_on_error(dev
);
6019 if (bio
== bbio
->orig_bio
)
6022 btrfs_bio_counter_dec(bbio
->fs_info
);
6024 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
6027 bio
= bbio
->orig_bio
;
6030 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
6031 /* only send an error to the higher layers if it is
6032 * beyond the tolerance of the btrfs bio
6034 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
6035 bio
->bi_error
= -EIO
;
6038 * this bio is actually up to date, we didn't
6039 * go over the max number of errors
6044 btrfs_end_bbio(bbio
, bio
);
6045 } else if (!is_orig_bio
) {
6051 * see run_scheduled_bios for a description of why bios are collected for
6054 * This will add one bio to the pending list for a device and make sure
6055 * the work struct is scheduled.
6057 static noinline
void btrfs_schedule_bio(struct btrfs_device
*device
,
6060 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
6061 int should_queue
= 1;
6062 struct btrfs_pending_bios
*pending_bios
;
6064 if (device
->missing
|| !device
->bdev
) {
6069 /* don't bother with additional async steps for reads, right now */
6070 if (bio_op(bio
) == REQ_OP_READ
) {
6072 btrfsic_submit_bio(bio
);
6078 * nr_async_bios allows us to reliably return congestion to the
6079 * higher layers. Otherwise, the async bio makes it appear we have
6080 * made progress against dirty pages when we've really just put it
6081 * on a queue for later
6083 atomic_inc(&fs_info
->nr_async_bios
);
6084 WARN_ON(bio
->bi_next
);
6085 bio
->bi_next
= NULL
;
6087 spin_lock(&device
->io_lock
);
6088 if (op_is_sync(bio
->bi_opf
))
6089 pending_bios
= &device
->pending_sync_bios
;
6091 pending_bios
= &device
->pending_bios
;
6093 if (pending_bios
->tail
)
6094 pending_bios
->tail
->bi_next
= bio
;
6096 pending_bios
->tail
= bio
;
6097 if (!pending_bios
->head
)
6098 pending_bios
->head
= bio
;
6099 if (device
->running_pending
)
6102 spin_unlock(&device
->io_lock
);
6105 btrfs_queue_work(fs_info
->submit_workers
, &device
->work
);
6108 static void submit_stripe_bio(struct btrfs_bio
*bbio
, struct bio
*bio
,
6109 u64 physical
, int dev_nr
, int async
)
6111 struct btrfs_device
*dev
= bbio
->stripes
[dev_nr
].dev
;
6112 struct btrfs_fs_info
*fs_info
= bbio
->fs_info
;
6114 bio
->bi_private
= bbio
;
6115 btrfs_io_bio(bio
)->stripe_index
= dev_nr
;
6116 bio
->bi_end_io
= btrfs_end_bio
;
6117 bio
->bi_iter
.bi_sector
= physical
>> 9;
6120 struct rcu_string
*name
;
6123 name
= rcu_dereference(dev
->name
);
6124 btrfs_debug(fs_info
,
6125 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6126 bio_op(bio
), bio
->bi_opf
,
6127 (u64
)bio
->bi_iter
.bi_sector
,
6128 (u_long
)dev
->bdev
->bd_dev
, name
->str
, dev
->devid
,
6129 bio
->bi_iter
.bi_size
);
6133 bio
->bi_bdev
= dev
->bdev
;
6135 btrfs_bio_counter_inc_noblocked(fs_info
);
6138 btrfs_schedule_bio(dev
, bio
);
6140 btrfsic_submit_bio(bio
);
6143 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
6145 atomic_inc(&bbio
->error
);
6146 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
6147 /* Should be the original bio. */
6148 WARN_ON(bio
!= bbio
->orig_bio
);
6150 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
6151 bio
->bi_iter
.bi_sector
= logical
>> 9;
6152 bio
->bi_error
= -EIO
;
6153 btrfs_end_bbio(bbio
, bio
);
6157 int btrfs_map_bio(struct btrfs_fs_info
*fs_info
, struct bio
*bio
,
6158 int mirror_num
, int async_submit
)
6160 struct btrfs_device
*dev
;
6161 struct bio
*first_bio
= bio
;
6162 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
6168 struct btrfs_bio
*bbio
= NULL
;
6170 length
= bio
->bi_iter
.bi_size
;
6171 map_length
= length
;
6173 btrfs_bio_counter_inc_blocked(fs_info
);
6174 ret
= __btrfs_map_block(fs_info
, bio_op(bio
), logical
,
6175 &map_length
, &bbio
, mirror_num
, 1);
6177 btrfs_bio_counter_dec(fs_info
);
6181 total_devs
= bbio
->num_stripes
;
6182 bbio
->orig_bio
= first_bio
;
6183 bbio
->private = first_bio
->bi_private
;
6184 bbio
->end_io
= first_bio
->bi_end_io
;
6185 bbio
->fs_info
= fs_info
;
6186 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
6188 if ((bbio
->map_type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
6189 ((bio_op(bio
) == REQ_OP_WRITE
) || (mirror_num
> 1))) {
6190 /* In this case, map_length has been set to the length of
6191 a single stripe; not the whole write */
6192 if (bio_op(bio
) == REQ_OP_WRITE
) {
6193 ret
= raid56_parity_write(fs_info
, bio
, bbio
,
6196 ret
= raid56_parity_recover(fs_info
, bio
, bbio
,
6197 map_length
, mirror_num
, 1);
6200 btrfs_bio_counter_dec(fs_info
);
6204 if (map_length
< length
) {
6206 "mapping failed logical %llu bio len %llu len %llu",
6207 logical
, length
, map_length
);
6211 for (dev_nr
= 0; dev_nr
< total_devs
; dev_nr
++) {
6212 dev
= bbio
->stripes
[dev_nr
].dev
;
6213 if (!dev
|| !dev
->bdev
||
6214 (bio_op(bio
) == REQ_OP_WRITE
&& !dev
->writeable
)) {
6215 bbio_error(bbio
, first_bio
, logical
);
6219 if (dev_nr
< total_devs
- 1) {
6220 bio
= btrfs_bio_clone(first_bio
, GFP_NOFS
);
6221 BUG_ON(!bio
); /* -ENOMEM */
6225 submit_stripe_bio(bbio
, bio
, bbio
->stripes
[dev_nr
].physical
,
6226 dev_nr
, async_submit
);
6228 btrfs_bio_counter_dec(fs_info
);
6232 struct btrfs_device
*btrfs_find_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
6235 struct btrfs_device
*device
;
6236 struct btrfs_fs_devices
*cur_devices
;
6238 cur_devices
= fs_info
->fs_devices
;
6239 while (cur_devices
) {
6241 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
6242 device
= __find_device(&cur_devices
->devices
,
6247 cur_devices
= cur_devices
->seed
;
6252 static struct btrfs_device
*add_missing_dev(struct btrfs_fs_devices
*fs_devices
,
6253 u64 devid
, u8
*dev_uuid
)
6255 struct btrfs_device
*device
;
6257 device
= btrfs_alloc_device(NULL
, &devid
, dev_uuid
);
6261 list_add(&device
->dev_list
, &fs_devices
->devices
);
6262 device
->fs_devices
= fs_devices
;
6263 fs_devices
->num_devices
++;
6265 device
->missing
= 1;
6266 fs_devices
->missing_devices
++;
6272 * btrfs_alloc_device - allocate struct btrfs_device
6273 * @fs_info: used only for generating a new devid, can be NULL if
6274 * devid is provided (i.e. @devid != NULL).
6275 * @devid: a pointer to devid for this device. If NULL a new devid
6277 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6280 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6281 * on error. Returned struct is not linked onto any lists and can be
6282 * destroyed with kfree() right away.
6284 struct btrfs_device
*btrfs_alloc_device(struct btrfs_fs_info
*fs_info
,
6288 struct btrfs_device
*dev
;
6291 if (WARN_ON(!devid
&& !fs_info
))
6292 return ERR_PTR(-EINVAL
);
6294 dev
= __alloc_device();
6303 ret
= find_next_devid(fs_info
, &tmp
);
6306 return ERR_PTR(ret
);
6312 memcpy(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
);
6314 generate_random_uuid(dev
->uuid
);
6316 btrfs_init_work(&dev
->work
, btrfs_submit_helper
,
6317 pending_bios_fn
, NULL
, NULL
);
6322 /* Return -EIO if any error, otherwise return 0. */
6323 static int btrfs_check_chunk_valid(struct btrfs_fs_info
*fs_info
,
6324 struct extent_buffer
*leaf
,
6325 struct btrfs_chunk
*chunk
, u64 logical
)
6333 length
= btrfs_chunk_length(leaf
, chunk
);
6334 stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6335 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6336 sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6337 type
= btrfs_chunk_type(leaf
, chunk
);
6340 btrfs_err(fs_info
, "invalid chunk num_stripes: %u",
6344 if (!IS_ALIGNED(logical
, fs_info
->sectorsize
)) {
6345 btrfs_err(fs_info
, "invalid chunk logical %llu", logical
);
6348 if (btrfs_chunk_sector_size(leaf
, chunk
) != fs_info
->sectorsize
) {
6349 btrfs_err(fs_info
, "invalid chunk sectorsize %u",
6350 btrfs_chunk_sector_size(leaf
, chunk
));
6353 if (!length
|| !IS_ALIGNED(length
, fs_info
->sectorsize
)) {
6354 btrfs_err(fs_info
, "invalid chunk length %llu", length
);
6357 if (!is_power_of_2(stripe_len
) || stripe_len
!= BTRFS_STRIPE_LEN
) {
6358 btrfs_err(fs_info
, "invalid chunk stripe length: %llu",
6362 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK
| BTRFS_BLOCK_GROUP_PROFILE_MASK
) &
6364 btrfs_err(fs_info
, "unrecognized chunk type: %llu",
6365 ~(BTRFS_BLOCK_GROUP_TYPE_MASK
|
6366 BTRFS_BLOCK_GROUP_PROFILE_MASK
) &
6367 btrfs_chunk_type(leaf
, chunk
));
6370 if ((type
& BTRFS_BLOCK_GROUP_RAID10
&& sub_stripes
!= 2) ||
6371 (type
& BTRFS_BLOCK_GROUP_RAID1
&& num_stripes
< 1) ||
6372 (type
& BTRFS_BLOCK_GROUP_RAID5
&& num_stripes
< 2) ||
6373 (type
& BTRFS_BLOCK_GROUP_RAID6
&& num_stripes
< 3) ||
6374 (type
& BTRFS_BLOCK_GROUP_DUP
&& num_stripes
> 2) ||
6375 ((type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) == 0 &&
6376 num_stripes
!= 1)) {
6378 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6379 num_stripes
, sub_stripes
,
6380 type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
);
6387 static int read_one_chunk(struct btrfs_fs_info
*fs_info
, struct btrfs_key
*key
,
6388 struct extent_buffer
*leaf
,
6389 struct btrfs_chunk
*chunk
)
6391 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
6392 struct map_lookup
*map
;
6393 struct extent_map
*em
;
6398 u8 uuid
[BTRFS_UUID_SIZE
];
6403 logical
= key
->offset
;
6404 length
= btrfs_chunk_length(leaf
, chunk
);
6405 stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6406 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6408 ret
= btrfs_check_chunk_valid(fs_info
, leaf
, chunk
, logical
);
6412 read_lock(&map_tree
->map_tree
.lock
);
6413 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
6414 read_unlock(&map_tree
->map_tree
.lock
);
6416 /* already mapped? */
6417 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
6418 free_extent_map(em
);
6421 free_extent_map(em
);
6424 em
= alloc_extent_map();
6427 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
6429 free_extent_map(em
);
6433 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
6434 em
->map_lookup
= map
;
6435 em
->start
= logical
;
6438 em
->block_start
= 0;
6439 em
->block_len
= em
->len
;
6441 map
->num_stripes
= num_stripes
;
6442 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
6443 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
6444 map
->sector_size
= btrfs_chunk_sector_size(leaf
, chunk
);
6445 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6446 map
->type
= btrfs_chunk_type(leaf
, chunk
);
6447 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6448 for (i
= 0; i
< num_stripes
; i
++) {
6449 map
->stripes
[i
].physical
=
6450 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
6451 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
6452 read_extent_buffer(leaf
, uuid
, (unsigned long)
6453 btrfs_stripe_dev_uuid_nr(chunk
, i
),
6455 map
->stripes
[i
].dev
= btrfs_find_device(fs_info
, devid
,
6457 if (!map
->stripes
[i
].dev
&&
6458 !btrfs_test_opt(fs_info
, DEGRADED
)) {
6459 free_extent_map(em
);
6462 if (!map
->stripes
[i
].dev
) {
6463 map
->stripes
[i
].dev
=
6464 add_missing_dev(fs_info
->fs_devices
, devid
,
6466 if (!map
->stripes
[i
].dev
) {
6467 free_extent_map(em
);
6470 btrfs_warn(fs_info
, "devid %llu uuid %pU is missing",
6473 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
6476 write_lock(&map_tree
->map_tree
.lock
);
6477 ret
= add_extent_mapping(&map_tree
->map_tree
, em
, 0);
6478 write_unlock(&map_tree
->map_tree
.lock
);
6479 BUG_ON(ret
); /* Tree corruption */
6480 free_extent_map(em
);
6485 static void fill_device_from_item(struct extent_buffer
*leaf
,
6486 struct btrfs_dev_item
*dev_item
,
6487 struct btrfs_device
*device
)
6491 device
->devid
= btrfs_device_id(leaf
, dev_item
);
6492 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
6493 device
->total_bytes
= device
->disk_total_bytes
;
6494 device
->commit_total_bytes
= device
->disk_total_bytes
;
6495 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
6496 device
->commit_bytes_used
= device
->bytes_used
;
6497 device
->type
= btrfs_device_type(leaf
, dev_item
);
6498 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
6499 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
6500 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
6501 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
6502 device
->is_tgtdev_for_dev_replace
= 0;
6504 ptr
= btrfs_device_uuid(dev_item
);
6505 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
6508 static struct btrfs_fs_devices
*open_seed_devices(struct btrfs_fs_info
*fs_info
,
6511 struct btrfs_fs_devices
*fs_devices
;
6514 BUG_ON(!mutex_is_locked(&uuid_mutex
));
6516 fs_devices
= fs_info
->fs_devices
->seed
;
6517 while (fs_devices
) {
6518 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
))
6521 fs_devices
= fs_devices
->seed
;
6524 fs_devices
= find_fsid(fsid
);
6526 if (!btrfs_test_opt(fs_info
, DEGRADED
))
6527 return ERR_PTR(-ENOENT
);
6529 fs_devices
= alloc_fs_devices(fsid
);
6530 if (IS_ERR(fs_devices
))
6533 fs_devices
->seeding
= 1;
6534 fs_devices
->opened
= 1;
6538 fs_devices
= clone_fs_devices(fs_devices
);
6539 if (IS_ERR(fs_devices
))
6542 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
6543 fs_info
->bdev_holder
);
6545 free_fs_devices(fs_devices
);
6546 fs_devices
= ERR_PTR(ret
);
6550 if (!fs_devices
->seeding
) {
6551 __btrfs_close_devices(fs_devices
);
6552 free_fs_devices(fs_devices
);
6553 fs_devices
= ERR_PTR(-EINVAL
);
6557 fs_devices
->seed
= fs_info
->fs_devices
->seed
;
6558 fs_info
->fs_devices
->seed
= fs_devices
;
6563 static int read_one_dev(struct btrfs_fs_info
*fs_info
,
6564 struct extent_buffer
*leaf
,
6565 struct btrfs_dev_item
*dev_item
)
6567 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6568 struct btrfs_device
*device
;
6571 u8 fs_uuid
[BTRFS_UUID_SIZE
];
6572 u8 dev_uuid
[BTRFS_UUID_SIZE
];
6574 devid
= btrfs_device_id(leaf
, dev_item
);
6575 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
6577 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
6580 if (memcmp(fs_uuid
, fs_info
->fsid
, BTRFS_UUID_SIZE
)) {
6581 fs_devices
= open_seed_devices(fs_info
, fs_uuid
);
6582 if (IS_ERR(fs_devices
))
6583 return PTR_ERR(fs_devices
);
6586 device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, fs_uuid
);
6588 if (!btrfs_test_opt(fs_info
, DEGRADED
))
6591 device
= add_missing_dev(fs_devices
, devid
, dev_uuid
);
6594 btrfs_warn(fs_info
, "devid %llu uuid %pU missing",
6597 if (!device
->bdev
&& !btrfs_test_opt(fs_info
, DEGRADED
))
6600 if(!device
->bdev
&& !device
->missing
) {
6602 * this happens when a device that was properly setup
6603 * in the device info lists suddenly goes bad.
6604 * device->bdev is NULL, and so we have to set
6605 * device->missing to one here
6607 device
->fs_devices
->missing_devices
++;
6608 device
->missing
= 1;
6611 /* Move the device to its own fs_devices */
6612 if (device
->fs_devices
!= fs_devices
) {
6613 ASSERT(device
->missing
);
6615 list_move(&device
->dev_list
, &fs_devices
->devices
);
6616 device
->fs_devices
->num_devices
--;
6617 fs_devices
->num_devices
++;
6619 device
->fs_devices
->missing_devices
--;
6620 fs_devices
->missing_devices
++;
6622 device
->fs_devices
= fs_devices
;
6626 if (device
->fs_devices
!= fs_info
->fs_devices
) {
6627 BUG_ON(device
->writeable
);
6628 if (device
->generation
!=
6629 btrfs_device_generation(leaf
, dev_item
))
6633 fill_device_from_item(leaf
, dev_item
, device
);
6634 device
->in_fs_metadata
= 1;
6635 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
6636 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
6637 spin_lock(&fs_info
->free_chunk_lock
);
6638 fs_info
->free_chunk_space
+= device
->total_bytes
-
6640 spin_unlock(&fs_info
->free_chunk_lock
);
6646 int btrfs_read_sys_array(struct btrfs_fs_info
*fs_info
)
6648 struct btrfs_root
*root
= fs_info
->tree_root
;
6649 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
6650 struct extent_buffer
*sb
;
6651 struct btrfs_disk_key
*disk_key
;
6652 struct btrfs_chunk
*chunk
;
6654 unsigned long sb_array_offset
;
6661 struct btrfs_key key
;
6663 ASSERT(BTRFS_SUPER_INFO_SIZE
<= fs_info
->nodesize
);
6665 * This will create extent buffer of nodesize, superblock size is
6666 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6667 * overallocate but we can keep it as-is, only the first page is used.
6669 sb
= btrfs_find_create_tree_block(fs_info
, BTRFS_SUPER_INFO_OFFSET
);
6672 set_extent_buffer_uptodate(sb
);
6673 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
6675 * The sb extent buffer is artificial and just used to read the system array.
6676 * set_extent_buffer_uptodate() call does not properly mark all it's
6677 * pages up-to-date when the page is larger: extent does not cover the
6678 * whole page and consequently check_page_uptodate does not find all
6679 * the page's extents up-to-date (the hole beyond sb),
6680 * write_extent_buffer then triggers a WARN_ON.
6682 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6683 * but sb spans only this function. Add an explicit SetPageUptodate call
6684 * to silence the warning eg. on PowerPC 64.
6686 if (PAGE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
6687 SetPageUptodate(sb
->pages
[0]);
6689 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
6690 array_size
= btrfs_super_sys_array_size(super_copy
);
6692 array_ptr
= super_copy
->sys_chunk_array
;
6693 sb_array_offset
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
6696 while (cur_offset
< array_size
) {
6697 disk_key
= (struct btrfs_disk_key
*)array_ptr
;
6698 len
= sizeof(*disk_key
);
6699 if (cur_offset
+ len
> array_size
)
6700 goto out_short_read
;
6702 btrfs_disk_key_to_cpu(&key
, disk_key
);
6705 sb_array_offset
+= len
;
6708 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6709 chunk
= (struct btrfs_chunk
*)sb_array_offset
;
6711 * At least one btrfs_chunk with one stripe must be
6712 * present, exact stripe count check comes afterwards
6714 len
= btrfs_chunk_item_size(1);
6715 if (cur_offset
+ len
> array_size
)
6716 goto out_short_read
;
6718 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
6721 "invalid number of stripes %u in sys_array at offset %u",
6722 num_stripes
, cur_offset
);
6727 type
= btrfs_chunk_type(sb
, chunk
);
6728 if ((type
& BTRFS_BLOCK_GROUP_SYSTEM
) == 0) {
6730 "invalid chunk type %llu in sys_array at offset %u",
6736 len
= btrfs_chunk_item_size(num_stripes
);
6737 if (cur_offset
+ len
> array_size
)
6738 goto out_short_read
;
6740 ret
= read_one_chunk(fs_info
, &key
, sb
, chunk
);
6745 "unexpected item type %u in sys_array at offset %u",
6746 (u32
)key
.type
, cur_offset
);
6751 sb_array_offset
+= len
;
6754 clear_extent_buffer_uptodate(sb
);
6755 free_extent_buffer_stale(sb
);
6759 btrfs_err(fs_info
, "sys_array too short to read %u bytes at offset %u",
6761 clear_extent_buffer_uptodate(sb
);
6762 free_extent_buffer_stale(sb
);
6766 int btrfs_read_chunk_tree(struct btrfs_fs_info
*fs_info
)
6768 struct btrfs_root
*root
= fs_info
->chunk_root
;
6769 struct btrfs_path
*path
;
6770 struct extent_buffer
*leaf
;
6771 struct btrfs_key key
;
6772 struct btrfs_key found_key
;
6777 path
= btrfs_alloc_path();
6781 mutex_lock(&uuid_mutex
);
6782 mutex_lock(&fs_info
->chunk_mutex
);
6785 * Read all device items, and then all the chunk items. All
6786 * device items are found before any chunk item (their object id
6787 * is smaller than the lowest possible object id for a chunk
6788 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6790 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
6793 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6797 leaf
= path
->nodes
[0];
6798 slot
= path
->slots
[0];
6799 if (slot
>= btrfs_header_nritems(leaf
)) {
6800 ret
= btrfs_next_leaf(root
, path
);
6807 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6808 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
6809 struct btrfs_dev_item
*dev_item
;
6810 dev_item
= btrfs_item_ptr(leaf
, slot
,
6811 struct btrfs_dev_item
);
6812 ret
= read_one_dev(fs_info
, leaf
, dev_item
);
6816 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6817 struct btrfs_chunk
*chunk
;
6818 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
6819 ret
= read_one_chunk(fs_info
, &found_key
, leaf
, chunk
);
6827 * After loading chunk tree, we've got all device information,
6828 * do another round of validation checks.
6830 if (total_dev
!= fs_info
->fs_devices
->total_devices
) {
6832 "super_num_devices %llu mismatch with num_devices %llu found here",
6833 btrfs_super_num_devices(fs_info
->super_copy
),
6838 if (btrfs_super_total_bytes(fs_info
->super_copy
) <
6839 fs_info
->fs_devices
->total_rw_bytes
) {
6841 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6842 btrfs_super_total_bytes(fs_info
->super_copy
),
6843 fs_info
->fs_devices
->total_rw_bytes
);
6849 mutex_unlock(&fs_info
->chunk_mutex
);
6850 mutex_unlock(&uuid_mutex
);
6852 btrfs_free_path(path
);
6856 void btrfs_init_devices_late(struct btrfs_fs_info
*fs_info
)
6858 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6859 struct btrfs_device
*device
;
6861 while (fs_devices
) {
6862 mutex_lock(&fs_devices
->device_list_mutex
);
6863 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
)
6864 device
->fs_info
= fs_info
;
6865 mutex_unlock(&fs_devices
->device_list_mutex
);
6867 fs_devices
= fs_devices
->seed
;
6871 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
6875 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6876 btrfs_dev_stat_reset(dev
, i
);
6879 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
6881 struct btrfs_key key
;
6882 struct btrfs_key found_key
;
6883 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6884 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6885 struct extent_buffer
*eb
;
6888 struct btrfs_device
*device
;
6889 struct btrfs_path
*path
= NULL
;
6892 path
= btrfs_alloc_path();
6898 mutex_lock(&fs_devices
->device_list_mutex
);
6899 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6901 struct btrfs_dev_stats_item
*ptr
;
6903 key
.objectid
= BTRFS_DEV_STATS_OBJECTID
;
6904 key
.type
= BTRFS_PERSISTENT_ITEM_KEY
;
6905 key
.offset
= device
->devid
;
6906 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
6908 __btrfs_reset_dev_stats(device
);
6909 device
->dev_stats_valid
= 1;
6910 btrfs_release_path(path
);
6913 slot
= path
->slots
[0];
6914 eb
= path
->nodes
[0];
6915 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
6916 item_size
= btrfs_item_size_nr(eb
, slot
);
6918 ptr
= btrfs_item_ptr(eb
, slot
,
6919 struct btrfs_dev_stats_item
);
6921 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6922 if (item_size
>= (1 + i
) * sizeof(__le64
))
6923 btrfs_dev_stat_set(device
, i
,
6924 btrfs_dev_stats_value(eb
, ptr
, i
));
6926 btrfs_dev_stat_reset(device
, i
);
6929 device
->dev_stats_valid
= 1;
6930 btrfs_dev_stat_print_on_load(device
);
6931 btrfs_release_path(path
);
6933 mutex_unlock(&fs_devices
->device_list_mutex
);
6936 btrfs_free_path(path
);
6937 return ret
< 0 ? ret
: 0;
6940 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
6941 struct btrfs_fs_info
*fs_info
,
6942 struct btrfs_device
*device
)
6944 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6945 struct btrfs_path
*path
;
6946 struct btrfs_key key
;
6947 struct extent_buffer
*eb
;
6948 struct btrfs_dev_stats_item
*ptr
;
6952 key
.objectid
= BTRFS_DEV_STATS_OBJECTID
;
6953 key
.type
= BTRFS_PERSISTENT_ITEM_KEY
;
6954 key
.offset
= device
->devid
;
6956 path
= btrfs_alloc_path();
6958 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
6960 btrfs_warn_in_rcu(fs_info
,
6961 "error %d while searching for dev_stats item for device %s",
6962 ret
, rcu_str_deref(device
->name
));
6967 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
6968 /* need to delete old one and insert a new one */
6969 ret
= btrfs_del_item(trans
, dev_root
, path
);
6971 btrfs_warn_in_rcu(fs_info
,
6972 "delete too small dev_stats item for device %s failed %d",
6973 rcu_str_deref(device
->name
), ret
);
6980 /* need to insert a new item */
6981 btrfs_release_path(path
);
6982 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
6983 &key
, sizeof(*ptr
));
6985 btrfs_warn_in_rcu(fs_info
,
6986 "insert dev_stats item for device %s failed %d",
6987 rcu_str_deref(device
->name
), ret
);
6992 eb
= path
->nodes
[0];
6993 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
6994 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6995 btrfs_set_dev_stats_value(eb
, ptr
, i
,
6996 btrfs_dev_stat_read(device
, i
));
6997 btrfs_mark_buffer_dirty(eb
);
7000 btrfs_free_path(path
);
7005 * called from commit_transaction. Writes all changed device stats to disk.
7007 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
7008 struct btrfs_fs_info
*fs_info
)
7010 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7011 struct btrfs_device
*device
;
7015 mutex_lock(&fs_devices
->device_list_mutex
);
7016 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
7017 if (!device
->dev_stats_valid
|| !btrfs_dev_stats_dirty(device
))
7020 stats_cnt
= atomic_read(&device
->dev_stats_ccnt
);
7021 ret
= update_dev_stat_item(trans
, fs_info
, device
);
7023 atomic_sub(stats_cnt
, &device
->dev_stats_ccnt
);
7025 mutex_unlock(&fs_devices
->device_list_mutex
);
7030 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
7032 btrfs_dev_stat_inc(dev
, index
);
7033 btrfs_dev_stat_print_on_error(dev
);
7036 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
7038 if (!dev
->dev_stats_valid
)
7040 btrfs_err_rl_in_rcu(dev
->fs_info
,
7041 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7042 rcu_str_deref(dev
->name
),
7043 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
7044 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
7045 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
7046 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
7047 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
7050 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
7054 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7055 if (btrfs_dev_stat_read(dev
, i
) != 0)
7057 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
7058 return; /* all values == 0, suppress message */
7060 btrfs_info_in_rcu(dev
->fs_info
,
7061 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7062 rcu_str_deref(dev
->name
),
7063 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
7064 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
7065 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
7066 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
7067 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
7070 int btrfs_get_dev_stats(struct btrfs_fs_info
*fs_info
,
7071 struct btrfs_ioctl_get_dev_stats
*stats
)
7073 struct btrfs_device
*dev
;
7074 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7077 mutex_lock(&fs_devices
->device_list_mutex
);
7078 dev
= btrfs_find_device(fs_info
, stats
->devid
, NULL
, NULL
);
7079 mutex_unlock(&fs_devices
->device_list_mutex
);
7082 btrfs_warn(fs_info
, "get dev_stats failed, device not found");
7084 } else if (!dev
->dev_stats_valid
) {
7085 btrfs_warn(fs_info
, "get dev_stats failed, not yet valid");
7087 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
7088 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
7089 if (stats
->nr_items
> i
)
7091 btrfs_dev_stat_read_and_reset(dev
, i
);
7093 btrfs_dev_stat_reset(dev
, i
);
7096 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7097 if (stats
->nr_items
> i
)
7098 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
7100 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
7101 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
7105 void btrfs_scratch_superblocks(struct block_device
*bdev
, char *device_path
)
7107 struct buffer_head
*bh
;
7108 struct btrfs_super_block
*disk_super
;
7114 for (copy_num
= 0; copy_num
< BTRFS_SUPER_MIRROR_MAX
;
7117 if (btrfs_read_dev_one_super(bdev
, copy_num
, &bh
))
7120 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
7122 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
7123 set_buffer_dirty(bh
);
7124 sync_dirty_buffer(bh
);
7128 /* Notify udev that device has changed */
7129 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
7131 /* Update ctime/mtime for device path for libblkid */
7132 update_dev_time(device_path
);
7136 * Update the size of all devices, which is used for writing out the
7139 void btrfs_update_commit_device_size(struct btrfs_fs_info
*fs_info
)
7141 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7142 struct btrfs_device
*curr
, *next
;
7144 if (list_empty(&fs_devices
->resized_devices
))
7147 mutex_lock(&fs_devices
->device_list_mutex
);
7148 mutex_lock(&fs_info
->chunk_mutex
);
7149 list_for_each_entry_safe(curr
, next
, &fs_devices
->resized_devices
,
7151 list_del_init(&curr
->resized_list
);
7152 curr
->commit_total_bytes
= curr
->disk_total_bytes
;
7154 mutex_unlock(&fs_info
->chunk_mutex
);
7155 mutex_unlock(&fs_devices
->device_list_mutex
);
7158 /* Must be invoked during the transaction commit */
7159 void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info
*fs_info
,
7160 struct btrfs_transaction
*transaction
)
7162 struct extent_map
*em
;
7163 struct map_lookup
*map
;
7164 struct btrfs_device
*dev
;
7167 if (list_empty(&transaction
->pending_chunks
))
7170 /* In order to kick the device replace finish process */
7171 mutex_lock(&fs_info
->chunk_mutex
);
7172 list_for_each_entry(em
, &transaction
->pending_chunks
, list
) {
7173 map
= em
->map_lookup
;
7175 for (i
= 0; i
< map
->num_stripes
; i
++) {
7176 dev
= map
->stripes
[i
].dev
;
7177 dev
->commit_bytes_used
= dev
->bytes_used
;
7180 mutex_unlock(&fs_info
->chunk_mutex
);
7183 void btrfs_set_fs_info_ptr(struct btrfs_fs_info
*fs_info
)
7185 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7186 while (fs_devices
) {
7187 fs_devices
->fs_info
= fs_info
;
7188 fs_devices
= fs_devices
->seed
;
7192 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info
*fs_info
)
7194 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7195 while (fs_devices
) {
7196 fs_devices
->fs_info
= NULL
;
7197 fs_devices
= fs_devices
->seed
;