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
);
142 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
,
143 enum btrfs_map_op op
,
144 u64 logical
, u64
*length
,
145 struct btrfs_bio
**bbio_ret
,
146 int mirror_num
, int need_raid_map
);
148 DEFINE_MUTEX(uuid_mutex
);
149 static LIST_HEAD(fs_uuids
);
150 struct list_head
*btrfs_get_fs_uuids(void)
156 * alloc_fs_devices - allocate struct btrfs_fs_devices
157 * @fsid: if not NULL, copy the uuid to fs_devices::fsid
159 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
160 * The returned struct is not linked onto any lists and can be destroyed with
161 * kfree() right away.
163 static struct btrfs_fs_devices
*alloc_fs_devices(const u8
*fsid
)
165 struct btrfs_fs_devices
*fs_devs
;
167 fs_devs
= kzalloc(sizeof(*fs_devs
), GFP_KERNEL
);
169 return ERR_PTR(-ENOMEM
);
171 mutex_init(&fs_devs
->device_list_mutex
);
173 INIT_LIST_HEAD(&fs_devs
->devices
);
174 INIT_LIST_HEAD(&fs_devs
->resized_devices
);
175 INIT_LIST_HEAD(&fs_devs
->alloc_list
);
176 INIT_LIST_HEAD(&fs_devs
->list
);
178 memcpy(fs_devs
->fsid
, fsid
, BTRFS_FSID_SIZE
);
183 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
185 struct btrfs_device
*device
;
186 WARN_ON(fs_devices
->opened
);
187 while (!list_empty(&fs_devices
->devices
)) {
188 device
= list_entry(fs_devices
->devices
.next
,
189 struct btrfs_device
, dev_list
);
190 list_del(&device
->dev_list
);
191 rcu_string_free(device
->name
);
197 static void btrfs_kobject_uevent(struct block_device
*bdev
,
198 enum kobject_action action
)
202 ret
= kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, action
);
204 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
206 kobject_name(&disk_to_dev(bdev
->bd_disk
)->kobj
),
207 &disk_to_dev(bdev
->bd_disk
)->kobj
);
210 void btrfs_cleanup_fs_uuids(void)
212 struct btrfs_fs_devices
*fs_devices
;
214 while (!list_empty(&fs_uuids
)) {
215 fs_devices
= list_entry(fs_uuids
.next
,
216 struct btrfs_fs_devices
, list
);
217 list_del(&fs_devices
->list
);
218 free_fs_devices(fs_devices
);
222 static struct btrfs_device
*__alloc_device(void)
224 struct btrfs_device
*dev
;
226 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
228 return ERR_PTR(-ENOMEM
);
231 * Preallocate a bio that's always going to be used for flushing device
232 * barriers and matches the device lifespan
234 dev
->flush_bio
= bio_alloc_bioset(GFP_KERNEL
, 0, NULL
);
235 if (!dev
->flush_bio
) {
237 return ERR_PTR(-ENOMEM
);
239 bio_get(dev
->flush_bio
);
241 INIT_LIST_HEAD(&dev
->dev_list
);
242 INIT_LIST_HEAD(&dev
->dev_alloc_list
);
243 INIT_LIST_HEAD(&dev
->resized_list
);
245 spin_lock_init(&dev
->io_lock
);
247 spin_lock_init(&dev
->reada_lock
);
248 atomic_set(&dev
->reada_in_flight
, 0);
249 atomic_set(&dev
->dev_stats_ccnt
, 0);
250 btrfs_device_data_ordered_init(dev
);
251 INIT_RADIX_TREE(&dev
->reada_zones
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
252 INIT_RADIX_TREE(&dev
->reada_extents
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
258 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
261 * If devid and uuid are both specified, the match must be exact, otherwise
262 * only devid is used.
264 static struct btrfs_device
*find_device(struct btrfs_fs_devices
*fs_devices
,
265 u64 devid
, const u8
*uuid
)
267 struct list_head
*head
= &fs_devices
->devices
;
268 struct btrfs_device
*dev
;
270 list_for_each_entry(dev
, head
, dev_list
) {
271 if (dev
->devid
== devid
&&
272 (!uuid
|| !memcmp(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
))) {
279 static noinline
struct btrfs_fs_devices
*find_fsid(u8
*fsid
)
281 struct btrfs_fs_devices
*fs_devices
;
283 list_for_each_entry(fs_devices
, &fs_uuids
, list
) {
284 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
291 btrfs_get_bdev_and_sb(const char *device_path
, fmode_t flags
, void *holder
,
292 int flush
, struct block_device
**bdev
,
293 struct buffer_head
**bh
)
297 *bdev
= blkdev_get_by_path(device_path
, flags
, holder
);
300 ret
= PTR_ERR(*bdev
);
305 filemap_write_and_wait((*bdev
)->bd_inode
->i_mapping
);
306 ret
= set_blocksize(*bdev
, BTRFS_BDEV_BLOCKSIZE
);
308 blkdev_put(*bdev
, flags
);
311 invalidate_bdev(*bdev
);
312 *bh
= btrfs_read_dev_super(*bdev
);
315 blkdev_put(*bdev
, flags
);
327 static void requeue_list(struct btrfs_pending_bios
*pending_bios
,
328 struct bio
*head
, struct bio
*tail
)
331 struct bio
*old_head
;
333 old_head
= pending_bios
->head
;
334 pending_bios
->head
= head
;
335 if (pending_bios
->tail
)
336 tail
->bi_next
= old_head
;
338 pending_bios
->tail
= tail
;
342 * we try to collect pending bios for a device so we don't get a large
343 * number of procs sending bios down to the same device. This greatly
344 * improves the schedulers ability to collect and merge the bios.
346 * But, it also turns into a long list of bios to process and that is sure
347 * to eventually make the worker thread block. The solution here is to
348 * make some progress and then put this work struct back at the end of
349 * the list if the block device is congested. This way, multiple devices
350 * can make progress from a single worker thread.
352 static noinline
void run_scheduled_bios(struct btrfs_device
*device
)
354 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
356 struct backing_dev_info
*bdi
;
357 struct btrfs_pending_bios
*pending_bios
;
361 unsigned long num_run
;
362 unsigned long batch_run
= 0;
364 unsigned long last_waited
= 0;
366 int sync_pending
= 0;
367 struct blk_plug plug
;
370 * this function runs all the bios we've collected for
371 * a particular device. We don't want to wander off to
372 * another device without first sending all of these down.
373 * So, setup a plug here and finish it off before we return
375 blk_start_plug(&plug
);
377 bdi
= device
->bdev
->bd_bdi
;
378 limit
= btrfs_async_submit_limit(fs_info
);
379 limit
= limit
* 2 / 3;
382 spin_lock(&device
->io_lock
);
387 /* take all the bios off the list at once and process them
388 * later on (without the lock held). But, remember the
389 * tail and other pointers so the bios can be properly reinserted
390 * into the list if we hit congestion
392 if (!force_reg
&& device
->pending_sync_bios
.head
) {
393 pending_bios
= &device
->pending_sync_bios
;
396 pending_bios
= &device
->pending_bios
;
400 pending
= pending_bios
->head
;
401 tail
= pending_bios
->tail
;
402 WARN_ON(pending
&& !tail
);
405 * if pending was null this time around, no bios need processing
406 * at all and we can stop. Otherwise it'll loop back up again
407 * and do an additional check so no bios are missed.
409 * device->running_pending is used to synchronize with the
412 if (device
->pending_sync_bios
.head
== NULL
&&
413 device
->pending_bios
.head
== NULL
) {
415 device
->running_pending
= 0;
418 device
->running_pending
= 1;
421 pending_bios
->head
= NULL
;
422 pending_bios
->tail
= NULL
;
424 spin_unlock(&device
->io_lock
);
429 /* we want to work on both lists, but do more bios on the
430 * sync list than the regular list
433 pending_bios
!= &device
->pending_sync_bios
&&
434 device
->pending_sync_bios
.head
) ||
435 (num_run
> 64 && pending_bios
== &device
->pending_sync_bios
&&
436 device
->pending_bios
.head
)) {
437 spin_lock(&device
->io_lock
);
438 requeue_list(pending_bios
, pending
, tail
);
443 pending
= pending
->bi_next
;
447 * atomic_dec_return implies a barrier for waitqueue_active
449 if (atomic_dec_return(&fs_info
->nr_async_bios
) < limit
&&
450 waitqueue_active(&fs_info
->async_submit_wait
))
451 wake_up(&fs_info
->async_submit_wait
);
453 BUG_ON(atomic_read(&cur
->__bi_cnt
) == 0);
456 * if we're doing the sync list, record that our
457 * plug has some sync requests on it
459 * If we're doing the regular list and there are
460 * sync requests sitting around, unplug before
463 if (pending_bios
== &device
->pending_sync_bios
) {
465 } else if (sync_pending
) {
466 blk_finish_plug(&plug
);
467 blk_start_plug(&plug
);
471 btrfsic_submit_bio(cur
);
478 * we made progress, there is more work to do and the bdi
479 * is now congested. Back off and let other work structs
482 if (pending
&& bdi_write_congested(bdi
) && batch_run
> 8 &&
483 fs_info
->fs_devices
->open_devices
> 1) {
484 struct io_context
*ioc
;
486 ioc
= current
->io_context
;
489 * the main goal here is that we don't want to
490 * block if we're going to be able to submit
491 * more requests without blocking.
493 * This code does two great things, it pokes into
494 * the elevator code from a filesystem _and_
495 * it makes assumptions about how batching works.
497 if (ioc
&& ioc
->nr_batch_requests
> 0 &&
498 time_before(jiffies
, ioc
->last_waited
+ HZ
/50UL) &&
500 ioc
->last_waited
== last_waited
)) {
502 * we want to go through our batch of
503 * requests and stop. So, we copy out
504 * the ioc->last_waited time and test
505 * against it before looping
507 last_waited
= ioc
->last_waited
;
511 spin_lock(&device
->io_lock
);
512 requeue_list(pending_bios
, pending
, tail
);
513 device
->running_pending
= 1;
515 spin_unlock(&device
->io_lock
);
516 btrfs_queue_work(fs_info
->submit_workers
,
526 spin_lock(&device
->io_lock
);
527 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
529 spin_unlock(&device
->io_lock
);
532 blk_finish_plug(&plug
);
535 static void pending_bios_fn(struct btrfs_work
*work
)
537 struct btrfs_device
*device
;
539 device
= container_of(work
, struct btrfs_device
, work
);
540 run_scheduled_bios(device
);
544 void btrfs_free_stale_device(struct btrfs_device
*cur_dev
)
546 struct btrfs_fs_devices
*fs_devs
;
547 struct btrfs_device
*dev
;
552 list_for_each_entry(fs_devs
, &fs_uuids
, list
) {
557 if (fs_devs
->seeding
)
560 list_for_each_entry(dev
, &fs_devs
->devices
, dev_list
) {
568 * Todo: This won't be enough. What if the same device
569 * comes back (with new uuid and) with its mapper path?
570 * But for now, this does help as mostly an admin will
571 * either use mapper or non mapper path throughout.
574 del
= strcmp(rcu_str_deref(dev
->name
),
575 rcu_str_deref(cur_dev
->name
));
582 /* delete the stale device */
583 if (fs_devs
->num_devices
== 1) {
584 btrfs_sysfs_remove_fsid(fs_devs
);
585 list_del(&fs_devs
->list
);
586 free_fs_devices(fs_devs
);
588 fs_devs
->num_devices
--;
589 list_del(&dev
->dev_list
);
590 rcu_string_free(dev
->name
);
599 * Add new device to list of registered devices
602 * 1 - first time device is seen
603 * 0 - device already known
606 static noinline
int device_list_add(const char *path
,
607 struct btrfs_super_block
*disk_super
,
608 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
610 struct btrfs_device
*device
;
611 struct btrfs_fs_devices
*fs_devices
;
612 struct rcu_string
*name
;
614 u64 found_transid
= btrfs_super_generation(disk_super
);
616 fs_devices
= find_fsid(disk_super
->fsid
);
618 fs_devices
= alloc_fs_devices(disk_super
->fsid
);
619 if (IS_ERR(fs_devices
))
620 return PTR_ERR(fs_devices
);
622 list_add(&fs_devices
->list
, &fs_uuids
);
626 device
= find_device(fs_devices
, devid
,
627 disk_super
->dev_item
.uuid
);
631 if (fs_devices
->opened
)
634 device
= btrfs_alloc_device(NULL
, &devid
,
635 disk_super
->dev_item
.uuid
);
636 if (IS_ERR(device
)) {
637 /* we can safely leave the fs_devices entry around */
638 return PTR_ERR(device
);
641 name
= rcu_string_strdup(path
, GFP_NOFS
);
646 rcu_assign_pointer(device
->name
, name
);
648 mutex_lock(&fs_devices
->device_list_mutex
);
649 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
650 fs_devices
->num_devices
++;
651 mutex_unlock(&fs_devices
->device_list_mutex
);
654 device
->fs_devices
= fs_devices
;
655 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
657 * When FS is already mounted.
658 * 1. If you are here and if the device->name is NULL that
659 * means this device was missing at time of FS mount.
660 * 2. If you are here and if the device->name is different
661 * from 'path' that means either
662 * a. The same device disappeared and reappeared with
664 * b. The missing-disk-which-was-replaced, has
667 * We must allow 1 and 2a above. But 2b would be a spurious
670 * Further in case of 1 and 2a above, the disk at 'path'
671 * would have missed some transaction when it was away and
672 * in case of 2a the stale bdev has to be updated as well.
673 * 2b must not be allowed at all time.
677 * For now, we do allow update to btrfs_fs_device through the
678 * btrfs dev scan cli after FS has been mounted. We're still
679 * tracking a problem where systems fail mount by subvolume id
680 * when we reject replacement on a mounted FS.
682 if (!fs_devices
->opened
&& found_transid
< device
->generation
) {
684 * That is if the FS is _not_ mounted and if you
685 * are here, that means there is more than one
686 * disk with same uuid and devid.We keep the one
687 * with larger generation number or the last-in if
688 * generation are equal.
693 name
= rcu_string_strdup(path
, GFP_NOFS
);
696 rcu_string_free(device
->name
);
697 rcu_assign_pointer(device
->name
, name
);
698 if (device
->missing
) {
699 fs_devices
->missing_devices
--;
705 * Unmount does not free the btrfs_device struct but would zero
706 * generation along with most of the other members. So just update
707 * it back. We need it to pick the disk with largest generation
710 if (!fs_devices
->opened
)
711 device
->generation
= found_transid
;
714 * if there is new btrfs on an already registered device,
715 * then remove the stale device entry.
718 btrfs_free_stale_device(device
);
720 *fs_devices_ret
= fs_devices
;
725 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
727 struct btrfs_fs_devices
*fs_devices
;
728 struct btrfs_device
*device
;
729 struct btrfs_device
*orig_dev
;
731 fs_devices
= alloc_fs_devices(orig
->fsid
);
732 if (IS_ERR(fs_devices
))
735 mutex_lock(&orig
->device_list_mutex
);
736 fs_devices
->total_devices
= orig
->total_devices
;
738 /* We have held the volume lock, it is safe to get the devices. */
739 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
740 struct rcu_string
*name
;
742 device
= btrfs_alloc_device(NULL
, &orig_dev
->devid
,
748 * This is ok to do without rcu read locked because we hold the
749 * uuid mutex so nothing we touch in here is going to disappear.
751 if (orig_dev
->name
) {
752 name
= rcu_string_strdup(orig_dev
->name
->str
,
758 rcu_assign_pointer(device
->name
, name
);
761 list_add(&device
->dev_list
, &fs_devices
->devices
);
762 device
->fs_devices
= fs_devices
;
763 fs_devices
->num_devices
++;
765 mutex_unlock(&orig
->device_list_mutex
);
768 mutex_unlock(&orig
->device_list_mutex
);
769 free_fs_devices(fs_devices
);
770 return ERR_PTR(-ENOMEM
);
773 void btrfs_close_extra_devices(struct btrfs_fs_devices
*fs_devices
, int step
)
775 struct btrfs_device
*device
, *next
;
776 struct btrfs_device
*latest_dev
= NULL
;
778 mutex_lock(&uuid_mutex
);
780 /* This is the initialized path, it is safe to release the devices. */
781 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
782 if (device
->in_fs_metadata
) {
783 if (!device
->is_tgtdev_for_dev_replace
&&
785 device
->generation
> latest_dev
->generation
)) {
791 if (device
->devid
== BTRFS_DEV_REPLACE_DEVID
) {
793 * In the first step, keep the device which has
794 * the correct fsid and the devid that is used
795 * for the dev_replace procedure.
796 * In the second step, the dev_replace state is
797 * read from the device tree and it is known
798 * whether the procedure is really active or
799 * not, which means whether this device is
800 * used or whether it should be removed.
802 if (step
== 0 || device
->is_tgtdev_for_dev_replace
) {
807 blkdev_put(device
->bdev
, device
->mode
);
809 fs_devices
->open_devices
--;
811 if (device
->writeable
) {
812 list_del_init(&device
->dev_alloc_list
);
813 device
->writeable
= 0;
814 if (!device
->is_tgtdev_for_dev_replace
)
815 fs_devices
->rw_devices
--;
817 list_del_init(&device
->dev_list
);
818 fs_devices
->num_devices
--;
819 rcu_string_free(device
->name
);
823 if (fs_devices
->seed
) {
824 fs_devices
= fs_devices
->seed
;
828 fs_devices
->latest_bdev
= latest_dev
->bdev
;
830 mutex_unlock(&uuid_mutex
);
833 static void __free_device(struct work_struct
*work
)
835 struct btrfs_device
*device
;
837 device
= container_of(work
, struct btrfs_device
, rcu_work
);
838 rcu_string_free(device
->name
);
839 bio_put(device
->flush_bio
);
843 static void free_device(struct rcu_head
*head
)
845 struct btrfs_device
*device
;
847 device
= container_of(head
, struct btrfs_device
, rcu
);
849 INIT_WORK(&device
->rcu_work
, __free_device
);
850 schedule_work(&device
->rcu_work
);
853 static void btrfs_close_bdev(struct btrfs_device
*device
)
855 if (device
->bdev
&& device
->writeable
) {
856 sync_blockdev(device
->bdev
);
857 invalidate_bdev(device
->bdev
);
861 blkdev_put(device
->bdev
, device
->mode
);
864 static void btrfs_prepare_close_one_device(struct btrfs_device
*device
)
866 struct btrfs_fs_devices
*fs_devices
= device
->fs_devices
;
867 struct btrfs_device
*new_device
;
868 struct rcu_string
*name
;
871 fs_devices
->open_devices
--;
873 if (device
->writeable
&&
874 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
875 list_del_init(&device
->dev_alloc_list
);
876 fs_devices
->rw_devices
--;
880 fs_devices
->missing_devices
--;
882 new_device
= btrfs_alloc_device(NULL
, &device
->devid
,
884 BUG_ON(IS_ERR(new_device
)); /* -ENOMEM */
886 /* Safe because we are under uuid_mutex */
888 name
= rcu_string_strdup(device
->name
->str
, GFP_NOFS
);
889 BUG_ON(!name
); /* -ENOMEM */
890 rcu_assign_pointer(new_device
->name
, name
);
893 list_replace_rcu(&device
->dev_list
, &new_device
->dev_list
);
894 new_device
->fs_devices
= device
->fs_devices
;
897 static int __btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
899 struct btrfs_device
*device
, *tmp
;
900 struct list_head pending_put
;
902 INIT_LIST_HEAD(&pending_put
);
904 if (--fs_devices
->opened
> 0)
907 mutex_lock(&fs_devices
->device_list_mutex
);
908 list_for_each_entry_safe(device
, tmp
, &fs_devices
->devices
, dev_list
) {
909 btrfs_prepare_close_one_device(device
);
910 list_add(&device
->dev_list
, &pending_put
);
912 mutex_unlock(&fs_devices
->device_list_mutex
);
915 * btrfs_show_devname() is using the device_list_mutex,
916 * sometimes call to blkdev_put() leads vfs calling
917 * into this func. So do put outside of device_list_mutex,
920 while (!list_empty(&pending_put
)) {
921 device
= list_first_entry(&pending_put
,
922 struct btrfs_device
, dev_list
);
923 list_del(&device
->dev_list
);
924 btrfs_close_bdev(device
);
925 call_rcu(&device
->rcu
, free_device
);
928 WARN_ON(fs_devices
->open_devices
);
929 WARN_ON(fs_devices
->rw_devices
);
930 fs_devices
->opened
= 0;
931 fs_devices
->seeding
= 0;
936 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
938 struct btrfs_fs_devices
*seed_devices
= NULL
;
941 mutex_lock(&uuid_mutex
);
942 ret
= __btrfs_close_devices(fs_devices
);
943 if (!fs_devices
->opened
) {
944 seed_devices
= fs_devices
->seed
;
945 fs_devices
->seed
= NULL
;
947 mutex_unlock(&uuid_mutex
);
949 while (seed_devices
) {
950 fs_devices
= seed_devices
;
951 seed_devices
= fs_devices
->seed
;
952 __btrfs_close_devices(fs_devices
);
953 free_fs_devices(fs_devices
);
956 * Wait for rcu kworkers under __btrfs_close_devices
957 * to finish all blkdev_puts so device is really
958 * free when umount is done.
964 static int __btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
965 fmode_t flags
, void *holder
)
967 struct request_queue
*q
;
968 struct block_device
*bdev
;
969 struct list_head
*head
= &fs_devices
->devices
;
970 struct btrfs_device
*device
;
971 struct btrfs_device
*latest_dev
= NULL
;
972 struct buffer_head
*bh
;
973 struct btrfs_super_block
*disk_super
;
980 list_for_each_entry(device
, head
, dev_list
) {
986 /* Just open everything we can; ignore failures here */
987 if (btrfs_get_bdev_and_sb(device
->name
->str
, flags
, holder
, 1,
991 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
992 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
993 if (devid
!= device
->devid
)
996 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
,
1000 device
->generation
= btrfs_super_generation(disk_super
);
1002 device
->generation
> latest_dev
->generation
)
1003 latest_dev
= device
;
1005 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
1006 device
->writeable
= 0;
1008 device
->writeable
= !bdev_read_only(bdev
);
1012 q
= bdev_get_queue(bdev
);
1013 if (blk_queue_discard(q
))
1014 device
->can_discard
= 1;
1015 if (!blk_queue_nonrot(q
))
1016 fs_devices
->rotating
= 1;
1018 device
->bdev
= bdev
;
1019 device
->in_fs_metadata
= 0;
1020 device
->mode
= flags
;
1022 fs_devices
->open_devices
++;
1023 if (device
->writeable
&&
1024 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
1025 fs_devices
->rw_devices
++;
1026 list_add(&device
->dev_alloc_list
,
1027 &fs_devices
->alloc_list
);
1034 blkdev_put(bdev
, flags
);
1037 if (fs_devices
->open_devices
== 0) {
1041 fs_devices
->seeding
= seeding
;
1042 fs_devices
->opened
= 1;
1043 fs_devices
->latest_bdev
= latest_dev
->bdev
;
1044 fs_devices
->total_rw_bytes
= 0;
1049 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
1050 fmode_t flags
, void *holder
)
1054 mutex_lock(&uuid_mutex
);
1055 if (fs_devices
->opened
) {
1056 fs_devices
->opened
++;
1059 ret
= __btrfs_open_devices(fs_devices
, flags
, holder
);
1061 mutex_unlock(&uuid_mutex
);
1065 void btrfs_release_disk_super(struct page
*page
)
1071 int btrfs_read_disk_super(struct block_device
*bdev
, u64 bytenr
,
1072 struct page
**page
, struct btrfs_super_block
**disk_super
)
1077 /* make sure our super fits in the device */
1078 if (bytenr
+ PAGE_SIZE
>= i_size_read(bdev
->bd_inode
))
1081 /* make sure our super fits in the page */
1082 if (sizeof(**disk_super
) > PAGE_SIZE
)
1085 /* make sure our super doesn't straddle pages on disk */
1086 index
= bytenr
>> PAGE_SHIFT
;
1087 if ((bytenr
+ sizeof(**disk_super
) - 1) >> PAGE_SHIFT
!= index
)
1090 /* pull in the page with our super */
1091 *page
= read_cache_page_gfp(bdev
->bd_inode
->i_mapping
,
1094 if (IS_ERR_OR_NULL(*page
))
1099 /* align our pointer to the offset of the super block */
1100 *disk_super
= p
+ (bytenr
& ~PAGE_MASK
);
1102 if (btrfs_super_bytenr(*disk_super
) != bytenr
||
1103 btrfs_super_magic(*disk_super
) != BTRFS_MAGIC
) {
1104 btrfs_release_disk_super(*page
);
1108 if ((*disk_super
)->label
[0] &&
1109 (*disk_super
)->label
[BTRFS_LABEL_SIZE
- 1])
1110 (*disk_super
)->label
[BTRFS_LABEL_SIZE
- 1] = '\0';
1116 * Look for a btrfs signature on a device. This may be called out of the mount path
1117 * and we are not allowed to call set_blocksize during the scan. The superblock
1118 * is read via pagecache
1120 int btrfs_scan_one_device(const char *path
, fmode_t flags
, void *holder
,
1121 struct btrfs_fs_devices
**fs_devices_ret
)
1123 struct btrfs_super_block
*disk_super
;
1124 struct block_device
*bdev
;
1133 * we would like to check all the supers, but that would make
1134 * a btrfs mount succeed after a mkfs from a different FS.
1135 * So, we need to add a special mount option to scan for
1136 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1138 bytenr
= btrfs_sb_offset(0);
1139 flags
|= FMODE_EXCL
;
1140 mutex_lock(&uuid_mutex
);
1142 bdev
= blkdev_get_by_path(path
, flags
, holder
);
1144 ret
= PTR_ERR(bdev
);
1148 if (btrfs_read_disk_super(bdev
, bytenr
, &page
, &disk_super
))
1149 goto error_bdev_put
;
1151 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1152 transid
= btrfs_super_generation(disk_super
);
1153 total_devices
= btrfs_super_num_devices(disk_super
);
1155 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
1157 if (disk_super
->label
[0]) {
1158 pr_info("BTRFS: device label %s ", disk_super
->label
);
1160 pr_info("BTRFS: device fsid %pU ", disk_super
->fsid
);
1163 pr_cont("devid %llu transid %llu %s\n", devid
, transid
, path
);
1166 if (!ret
&& fs_devices_ret
)
1167 (*fs_devices_ret
)->total_devices
= total_devices
;
1169 btrfs_release_disk_super(page
);
1172 blkdev_put(bdev
, flags
);
1174 mutex_unlock(&uuid_mutex
);
1178 /* helper to account the used device space in the range */
1179 int btrfs_account_dev_extents_size(struct btrfs_device
*device
, u64 start
,
1180 u64 end
, u64
*length
)
1182 struct btrfs_key key
;
1183 struct btrfs_root
*root
= device
->fs_info
->dev_root
;
1184 struct btrfs_dev_extent
*dev_extent
;
1185 struct btrfs_path
*path
;
1189 struct extent_buffer
*l
;
1193 if (start
>= device
->total_bytes
|| device
->is_tgtdev_for_dev_replace
)
1196 path
= btrfs_alloc_path();
1199 path
->reada
= READA_FORWARD
;
1201 key
.objectid
= device
->devid
;
1203 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1205 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1209 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1216 slot
= path
->slots
[0];
1217 if (slot
>= btrfs_header_nritems(l
)) {
1218 ret
= btrfs_next_leaf(root
, path
);
1226 btrfs_item_key_to_cpu(l
, &key
, slot
);
1228 if (key
.objectid
< device
->devid
)
1231 if (key
.objectid
> device
->devid
)
1234 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1237 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1238 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1240 if (key
.offset
<= start
&& extent_end
> end
) {
1241 *length
= end
- start
+ 1;
1243 } else if (key
.offset
<= start
&& extent_end
> start
)
1244 *length
+= extent_end
- start
;
1245 else if (key
.offset
> start
&& extent_end
<= end
)
1246 *length
+= extent_end
- key
.offset
;
1247 else if (key
.offset
> start
&& key
.offset
<= end
) {
1248 *length
+= end
- key
.offset
+ 1;
1250 } else if (key
.offset
> end
)
1258 btrfs_free_path(path
);
1262 static int contains_pending_extent(struct btrfs_transaction
*transaction
,
1263 struct btrfs_device
*device
,
1264 u64
*start
, u64 len
)
1266 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1267 struct extent_map
*em
;
1268 struct list_head
*search_list
= &fs_info
->pinned_chunks
;
1270 u64 physical_start
= *start
;
1273 search_list
= &transaction
->pending_chunks
;
1275 list_for_each_entry(em
, search_list
, list
) {
1276 struct map_lookup
*map
;
1279 map
= em
->map_lookup
;
1280 for (i
= 0; i
< map
->num_stripes
; i
++) {
1283 if (map
->stripes
[i
].dev
!= device
)
1285 if (map
->stripes
[i
].physical
>= physical_start
+ len
||
1286 map
->stripes
[i
].physical
+ em
->orig_block_len
<=
1290 * Make sure that while processing the pinned list we do
1291 * not override our *start with a lower value, because
1292 * we can have pinned chunks that fall within this
1293 * device hole and that have lower physical addresses
1294 * than the pending chunks we processed before. If we
1295 * do not take this special care we can end up getting
1296 * 2 pending chunks that start at the same physical
1297 * device offsets because the end offset of a pinned
1298 * chunk can be equal to the start offset of some
1301 end
= map
->stripes
[i
].physical
+ em
->orig_block_len
;
1308 if (search_list
!= &fs_info
->pinned_chunks
) {
1309 search_list
= &fs_info
->pinned_chunks
;
1318 * find_free_dev_extent_start - find free space in the specified device
1319 * @device: the device which we search the free space in
1320 * @num_bytes: the size of the free space that we need
1321 * @search_start: the position from which to begin the search
1322 * @start: store the start of the free space.
1323 * @len: the size of the free space. that we find, or the size
1324 * of the max free space if we don't find suitable free space
1326 * this uses a pretty simple search, the expectation is that it is
1327 * called very infrequently and that a given device has a small number
1330 * @start is used to store the start of the free space if we find. But if we
1331 * don't find suitable free space, it will be used to store the start position
1332 * of the max free space.
1334 * @len is used to store the size of the free space that we find.
1335 * But if we don't find suitable free space, it is used to store the size of
1336 * the max free space.
1338 int find_free_dev_extent_start(struct btrfs_transaction
*transaction
,
1339 struct btrfs_device
*device
, u64 num_bytes
,
1340 u64 search_start
, u64
*start
, u64
*len
)
1342 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1343 struct btrfs_root
*root
= fs_info
->dev_root
;
1344 struct btrfs_key key
;
1345 struct btrfs_dev_extent
*dev_extent
;
1346 struct btrfs_path
*path
;
1351 u64 search_end
= device
->total_bytes
;
1354 struct extent_buffer
*l
;
1357 * We don't want to overwrite the superblock on the drive nor any area
1358 * used by the boot loader (grub for example), so we make sure to start
1359 * at an offset of at least 1MB.
1361 search_start
= max_t(u64
, search_start
, SZ_1M
);
1363 path
= btrfs_alloc_path();
1367 max_hole_start
= search_start
;
1371 if (search_start
>= search_end
|| device
->is_tgtdev_for_dev_replace
) {
1376 path
->reada
= READA_FORWARD
;
1377 path
->search_commit_root
= 1;
1378 path
->skip_locking
= 1;
1380 key
.objectid
= device
->devid
;
1381 key
.offset
= search_start
;
1382 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1384 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1388 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1395 slot
= path
->slots
[0];
1396 if (slot
>= btrfs_header_nritems(l
)) {
1397 ret
= btrfs_next_leaf(root
, path
);
1405 btrfs_item_key_to_cpu(l
, &key
, slot
);
1407 if (key
.objectid
< device
->devid
)
1410 if (key
.objectid
> device
->devid
)
1413 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1416 if (key
.offset
> search_start
) {
1417 hole_size
= key
.offset
- search_start
;
1420 * Have to check before we set max_hole_start, otherwise
1421 * we could end up sending back this offset anyway.
1423 if (contains_pending_extent(transaction
, device
,
1426 if (key
.offset
>= search_start
) {
1427 hole_size
= key
.offset
- search_start
;
1434 if (hole_size
> max_hole_size
) {
1435 max_hole_start
= search_start
;
1436 max_hole_size
= hole_size
;
1440 * If this free space is greater than which we need,
1441 * it must be the max free space that we have found
1442 * until now, so max_hole_start must point to the start
1443 * of this free space and the length of this free space
1444 * is stored in max_hole_size. Thus, we return
1445 * max_hole_start and max_hole_size and go back to the
1448 if (hole_size
>= num_bytes
) {
1454 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1455 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1457 if (extent_end
> search_start
)
1458 search_start
= extent_end
;
1465 * At this point, search_start should be the end of
1466 * allocated dev extents, and when shrinking the device,
1467 * search_end may be smaller than search_start.
1469 if (search_end
> search_start
) {
1470 hole_size
= search_end
- search_start
;
1472 if (contains_pending_extent(transaction
, device
, &search_start
,
1474 btrfs_release_path(path
);
1478 if (hole_size
> max_hole_size
) {
1479 max_hole_start
= search_start
;
1480 max_hole_size
= hole_size
;
1485 if (max_hole_size
< num_bytes
)
1491 btrfs_free_path(path
);
1492 *start
= max_hole_start
;
1494 *len
= max_hole_size
;
1498 int find_free_dev_extent(struct btrfs_trans_handle
*trans
,
1499 struct btrfs_device
*device
, u64 num_bytes
,
1500 u64
*start
, u64
*len
)
1502 /* FIXME use last free of some kind */
1503 return find_free_dev_extent_start(trans
->transaction
, device
,
1504 num_bytes
, 0, start
, len
);
1507 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1508 struct btrfs_device
*device
,
1509 u64 start
, u64
*dev_extent_len
)
1511 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1512 struct btrfs_root
*root
= fs_info
->dev_root
;
1514 struct btrfs_path
*path
;
1515 struct btrfs_key key
;
1516 struct btrfs_key found_key
;
1517 struct extent_buffer
*leaf
= NULL
;
1518 struct btrfs_dev_extent
*extent
= NULL
;
1520 path
= btrfs_alloc_path();
1524 key
.objectid
= device
->devid
;
1526 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1528 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1530 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1531 BTRFS_DEV_EXTENT_KEY
);
1534 leaf
= path
->nodes
[0];
1535 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1536 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1537 struct btrfs_dev_extent
);
1538 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1539 btrfs_dev_extent_length(leaf
, extent
) < start
);
1541 btrfs_release_path(path
);
1543 } else if (ret
== 0) {
1544 leaf
= path
->nodes
[0];
1545 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1546 struct btrfs_dev_extent
);
1548 btrfs_handle_fs_error(fs_info
, ret
, "Slot search failed");
1552 *dev_extent_len
= btrfs_dev_extent_length(leaf
, extent
);
1554 ret
= btrfs_del_item(trans
, root
, path
);
1556 btrfs_handle_fs_error(fs_info
, ret
,
1557 "Failed to remove dev extent item");
1559 set_bit(BTRFS_TRANS_HAVE_FREE_BGS
, &trans
->transaction
->flags
);
1562 btrfs_free_path(path
);
1566 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1567 struct btrfs_device
*device
,
1568 u64 chunk_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
,
1596 BTRFS_CHUNK_TREE_OBJECTID
);
1597 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
,
1598 BTRFS_FIRST_CHUNK_TREE_OBJECTID
);
1599 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
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_FSID_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(const 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
, const char *device_path
,
1857 struct btrfs_device
*device
;
1858 struct btrfs_fs_devices
*cur_devices
;
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
);
1898 mutex_unlock(&uuid_mutex
);
1899 ret
= btrfs_shrink_device(device
, 0);
1900 mutex_lock(&uuid_mutex
);
1905 * TODO: the superblock still includes this device in its num_devices
1906 * counter although write_all_supers() is not locked out. This
1907 * could give a filesystem state which requires a degraded mount.
1909 ret
= btrfs_rm_dev_item(fs_info
, device
);
1913 device
->in_fs_metadata
= 0;
1914 btrfs_scrub_cancel_dev(fs_info
, device
);
1917 * the device list mutex makes sure that we don't change
1918 * the device list while someone else is writing out all
1919 * the device supers. Whoever is writing all supers, should
1920 * lock the device list mutex before getting the number of
1921 * devices in the super block (super_copy). Conversely,
1922 * whoever updates the number of devices in the super block
1923 * (super_copy) should hold the device list mutex.
1926 cur_devices
= device
->fs_devices
;
1927 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
1928 list_del_rcu(&device
->dev_list
);
1930 device
->fs_devices
->num_devices
--;
1931 device
->fs_devices
->total_devices
--;
1933 if (device
->missing
)
1934 device
->fs_devices
->missing_devices
--;
1936 btrfs_assign_next_active_device(fs_info
, device
, NULL
);
1939 device
->fs_devices
->open_devices
--;
1940 /* remove sysfs entry */
1941 btrfs_sysfs_rm_device_link(fs_info
->fs_devices
, device
);
1944 num_devices
= btrfs_super_num_devices(fs_info
->super_copy
) - 1;
1945 btrfs_set_super_num_devices(fs_info
->super_copy
, num_devices
);
1946 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
1949 * at this point, the device is zero sized and detached from
1950 * the devices list. All that's left is to zero out the old
1951 * supers and free the device.
1953 if (device
->writeable
)
1954 btrfs_scratch_superblocks(device
->bdev
, device
->name
->str
);
1956 btrfs_close_bdev(device
);
1957 call_rcu(&device
->rcu
, free_device
);
1959 if (cur_devices
->open_devices
== 0) {
1960 struct btrfs_fs_devices
*fs_devices
;
1961 fs_devices
= fs_info
->fs_devices
;
1962 while (fs_devices
) {
1963 if (fs_devices
->seed
== cur_devices
) {
1964 fs_devices
->seed
= cur_devices
->seed
;
1967 fs_devices
= fs_devices
->seed
;
1969 cur_devices
->seed
= NULL
;
1970 __btrfs_close_devices(cur_devices
);
1971 free_fs_devices(cur_devices
);
1975 mutex_unlock(&uuid_mutex
);
1979 if (device
->writeable
) {
1980 mutex_lock(&fs_info
->chunk_mutex
);
1981 list_add(&device
->dev_alloc_list
,
1982 &fs_info
->fs_devices
->alloc_list
);
1983 device
->fs_devices
->rw_devices
++;
1984 mutex_unlock(&fs_info
->chunk_mutex
);
1989 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info
*fs_info
,
1990 struct btrfs_device
*srcdev
)
1992 struct btrfs_fs_devices
*fs_devices
;
1994 WARN_ON(!mutex_is_locked(&fs_info
->fs_devices
->device_list_mutex
));
1997 * in case of fs with no seed, srcdev->fs_devices will point
1998 * to fs_devices of fs_info. However when the dev being replaced is
1999 * a seed dev it will point to the seed's local fs_devices. In short
2000 * srcdev will have its correct fs_devices in both the cases.
2002 fs_devices
= srcdev
->fs_devices
;
2004 list_del_rcu(&srcdev
->dev_list
);
2005 list_del_rcu(&srcdev
->dev_alloc_list
);
2006 fs_devices
->num_devices
--;
2007 if (srcdev
->missing
)
2008 fs_devices
->missing_devices
--;
2010 if (srcdev
->writeable
)
2011 fs_devices
->rw_devices
--;
2014 fs_devices
->open_devices
--;
2017 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info
*fs_info
,
2018 struct btrfs_device
*srcdev
)
2020 struct btrfs_fs_devices
*fs_devices
= srcdev
->fs_devices
;
2022 if (srcdev
->writeable
) {
2023 /* zero out the old super if it is writable */
2024 btrfs_scratch_superblocks(srcdev
->bdev
, srcdev
->name
->str
);
2027 btrfs_close_bdev(srcdev
);
2029 call_rcu(&srcdev
->rcu
, free_device
);
2032 * unless fs_devices is seed fs, num_devices shouldn't go
2035 BUG_ON(!fs_devices
->num_devices
&& !fs_devices
->seeding
);
2037 /* if this is no devs we rather delete the fs_devices */
2038 if (!fs_devices
->num_devices
) {
2039 struct btrfs_fs_devices
*tmp_fs_devices
;
2041 tmp_fs_devices
= fs_info
->fs_devices
;
2042 while (tmp_fs_devices
) {
2043 if (tmp_fs_devices
->seed
== fs_devices
) {
2044 tmp_fs_devices
->seed
= fs_devices
->seed
;
2047 tmp_fs_devices
= tmp_fs_devices
->seed
;
2049 fs_devices
->seed
= NULL
;
2050 __btrfs_close_devices(fs_devices
);
2051 free_fs_devices(fs_devices
);
2055 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
2056 struct btrfs_device
*tgtdev
)
2058 mutex_lock(&uuid_mutex
);
2060 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2062 btrfs_sysfs_rm_device_link(fs_info
->fs_devices
, tgtdev
);
2065 fs_info
->fs_devices
->open_devices
--;
2067 fs_info
->fs_devices
->num_devices
--;
2069 btrfs_assign_next_active_device(fs_info
, tgtdev
, NULL
);
2071 list_del_rcu(&tgtdev
->dev_list
);
2073 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2074 mutex_unlock(&uuid_mutex
);
2077 * The update_dev_time() with in btrfs_scratch_superblocks()
2078 * may lead to a call to btrfs_show_devname() which will try
2079 * to hold device_list_mutex. And here this device
2080 * is already out of device list, so we don't have to hold
2081 * the device_list_mutex lock.
2083 btrfs_scratch_superblocks(tgtdev
->bdev
, tgtdev
->name
->str
);
2085 btrfs_close_bdev(tgtdev
);
2086 call_rcu(&tgtdev
->rcu
, free_device
);
2089 static int btrfs_find_device_by_path(struct btrfs_fs_info
*fs_info
,
2090 const char *device_path
,
2091 struct btrfs_device
**device
)
2094 struct btrfs_super_block
*disk_super
;
2097 struct block_device
*bdev
;
2098 struct buffer_head
*bh
;
2101 ret
= btrfs_get_bdev_and_sb(device_path
, FMODE_READ
,
2102 fs_info
->bdev_holder
, 0, &bdev
, &bh
);
2105 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
2106 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
2107 dev_uuid
= disk_super
->dev_item
.uuid
;
2108 *device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, disk_super
->fsid
);
2112 blkdev_put(bdev
, FMODE_READ
);
2116 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info
*fs_info
,
2117 const char *device_path
,
2118 struct btrfs_device
**device
)
2121 if (strcmp(device_path
, "missing") == 0) {
2122 struct list_head
*devices
;
2123 struct btrfs_device
*tmp
;
2125 devices
= &fs_info
->fs_devices
->devices
;
2127 * It is safe to read the devices since the volume_mutex
2128 * is held by the caller.
2130 list_for_each_entry(tmp
, devices
, dev_list
) {
2131 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
2138 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND
;
2142 return btrfs_find_device_by_path(fs_info
, device_path
, device
);
2147 * Lookup a device given by device id, or the path if the id is 0.
2149 int btrfs_find_device_by_devspec(struct btrfs_fs_info
*fs_info
, u64 devid
,
2150 const char *devpath
,
2151 struct btrfs_device
**device
)
2157 *device
= btrfs_find_device(fs_info
, devid
, NULL
, NULL
);
2161 if (!devpath
|| !devpath
[0])
2164 ret
= btrfs_find_device_missing_or_by_path(fs_info
, devpath
,
2171 * does all the dirty work required for changing file system's UUID.
2173 static int btrfs_prepare_sprout(struct btrfs_fs_info
*fs_info
)
2175 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2176 struct btrfs_fs_devices
*old_devices
;
2177 struct btrfs_fs_devices
*seed_devices
;
2178 struct btrfs_super_block
*disk_super
= fs_info
->super_copy
;
2179 struct btrfs_device
*device
;
2182 BUG_ON(!mutex_is_locked(&uuid_mutex
));
2183 if (!fs_devices
->seeding
)
2186 seed_devices
= alloc_fs_devices(NULL
);
2187 if (IS_ERR(seed_devices
))
2188 return PTR_ERR(seed_devices
);
2190 old_devices
= clone_fs_devices(fs_devices
);
2191 if (IS_ERR(old_devices
)) {
2192 kfree(seed_devices
);
2193 return PTR_ERR(old_devices
);
2196 list_add(&old_devices
->list
, &fs_uuids
);
2198 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
2199 seed_devices
->opened
= 1;
2200 INIT_LIST_HEAD(&seed_devices
->devices
);
2201 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
2202 mutex_init(&seed_devices
->device_list_mutex
);
2204 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2205 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
2207 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
)
2208 device
->fs_devices
= seed_devices
;
2210 mutex_lock(&fs_info
->chunk_mutex
);
2211 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
2212 mutex_unlock(&fs_info
->chunk_mutex
);
2214 fs_devices
->seeding
= 0;
2215 fs_devices
->num_devices
= 0;
2216 fs_devices
->open_devices
= 0;
2217 fs_devices
->missing_devices
= 0;
2218 fs_devices
->rotating
= 0;
2219 fs_devices
->seed
= seed_devices
;
2221 generate_random_uuid(fs_devices
->fsid
);
2222 memcpy(fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2223 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2224 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2226 super_flags
= btrfs_super_flags(disk_super
) &
2227 ~BTRFS_SUPER_FLAG_SEEDING
;
2228 btrfs_set_super_flags(disk_super
, super_flags
);
2234 * Store the expected generation for seed devices in device items.
2236 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
2237 struct btrfs_fs_info
*fs_info
)
2239 struct btrfs_root
*root
= fs_info
->chunk_root
;
2240 struct btrfs_path
*path
;
2241 struct extent_buffer
*leaf
;
2242 struct btrfs_dev_item
*dev_item
;
2243 struct btrfs_device
*device
;
2244 struct btrfs_key key
;
2245 u8 fs_uuid
[BTRFS_FSID_SIZE
];
2246 u8 dev_uuid
[BTRFS_UUID_SIZE
];
2250 path
= btrfs_alloc_path();
2254 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2256 key
.type
= BTRFS_DEV_ITEM_KEY
;
2259 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2263 leaf
= path
->nodes
[0];
2265 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
2266 ret
= btrfs_next_leaf(root
, path
);
2271 leaf
= path
->nodes
[0];
2272 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2273 btrfs_release_path(path
);
2277 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2278 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
2279 key
.type
!= BTRFS_DEV_ITEM_KEY
)
2282 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2283 struct btrfs_dev_item
);
2284 devid
= btrfs_device_id(leaf
, dev_item
);
2285 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
2287 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
2289 device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, fs_uuid
);
2290 BUG_ON(!device
); /* Logic error */
2292 if (device
->fs_devices
->seeding
) {
2293 btrfs_set_device_generation(leaf
, dev_item
,
2294 device
->generation
);
2295 btrfs_mark_buffer_dirty(leaf
);
2303 btrfs_free_path(path
);
2307 int btrfs_init_new_device(struct btrfs_fs_info
*fs_info
, const char *device_path
)
2309 struct btrfs_root
*root
= fs_info
->dev_root
;
2310 struct request_queue
*q
;
2311 struct btrfs_trans_handle
*trans
;
2312 struct btrfs_device
*device
;
2313 struct block_device
*bdev
;
2314 struct list_head
*devices
;
2315 struct super_block
*sb
= fs_info
->sb
;
2316 struct rcu_string
*name
;
2318 int seeding_dev
= 0;
2321 if (sb_rdonly(sb
) && !fs_info
->fs_devices
->seeding
)
2324 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2325 fs_info
->bdev_holder
);
2327 return PTR_ERR(bdev
);
2329 if (fs_info
->fs_devices
->seeding
) {
2331 down_write(&sb
->s_umount
);
2332 mutex_lock(&uuid_mutex
);
2335 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2337 devices
= &fs_info
->fs_devices
->devices
;
2339 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2340 list_for_each_entry(device
, devices
, dev_list
) {
2341 if (device
->bdev
== bdev
) {
2344 &fs_info
->fs_devices
->device_list_mutex
);
2348 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2350 device
= btrfs_alloc_device(fs_info
, NULL
, NULL
);
2351 if (IS_ERR(device
)) {
2352 /* we can safely leave the fs_devices entry around */
2353 ret
= PTR_ERR(device
);
2357 name
= rcu_string_strdup(device_path
, GFP_KERNEL
);
2363 rcu_assign_pointer(device
->name
, name
);
2365 trans
= btrfs_start_transaction(root
, 0);
2366 if (IS_ERR(trans
)) {
2367 rcu_string_free(device
->name
);
2369 ret
= PTR_ERR(trans
);
2373 q
= bdev_get_queue(bdev
);
2374 if (blk_queue_discard(q
))
2375 device
->can_discard
= 1;
2376 device
->writeable
= 1;
2377 device
->generation
= trans
->transid
;
2378 device
->io_width
= fs_info
->sectorsize
;
2379 device
->io_align
= fs_info
->sectorsize
;
2380 device
->sector_size
= fs_info
->sectorsize
;
2381 device
->total_bytes
= round_down(i_size_read(bdev
->bd_inode
),
2382 fs_info
->sectorsize
);
2383 device
->disk_total_bytes
= device
->total_bytes
;
2384 device
->commit_total_bytes
= device
->total_bytes
;
2385 device
->fs_info
= fs_info
;
2386 device
->bdev
= bdev
;
2387 device
->in_fs_metadata
= 1;
2388 device
->is_tgtdev_for_dev_replace
= 0;
2389 device
->mode
= FMODE_EXCL
;
2390 device
->dev_stats_valid
= 1;
2391 set_blocksize(device
->bdev
, BTRFS_BDEV_BLOCKSIZE
);
2394 sb
->s_flags
&= ~MS_RDONLY
;
2395 ret
= btrfs_prepare_sprout(fs_info
);
2396 BUG_ON(ret
); /* -ENOMEM */
2399 device
->fs_devices
= fs_info
->fs_devices
;
2401 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2402 mutex_lock(&fs_info
->chunk_mutex
);
2403 list_add_rcu(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2404 list_add(&device
->dev_alloc_list
,
2405 &fs_info
->fs_devices
->alloc_list
);
2406 fs_info
->fs_devices
->num_devices
++;
2407 fs_info
->fs_devices
->open_devices
++;
2408 fs_info
->fs_devices
->rw_devices
++;
2409 fs_info
->fs_devices
->total_devices
++;
2410 fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
2412 atomic64_add(device
->total_bytes
, &fs_info
->free_chunk_space
);
2414 if (!blk_queue_nonrot(q
))
2415 fs_info
->fs_devices
->rotating
= 1;
2417 tmp
= btrfs_super_total_bytes(fs_info
->super_copy
);
2418 btrfs_set_super_total_bytes(fs_info
->super_copy
,
2419 round_down(tmp
+ device
->total_bytes
, fs_info
->sectorsize
));
2421 tmp
= btrfs_super_num_devices(fs_info
->super_copy
);
2422 btrfs_set_super_num_devices(fs_info
->super_copy
, tmp
+ 1);
2424 /* add sysfs device entry */
2425 btrfs_sysfs_add_device_link(fs_info
->fs_devices
, device
);
2428 * we've got more storage, clear any full flags on the space
2431 btrfs_clear_space_info_full(fs_info
);
2433 mutex_unlock(&fs_info
->chunk_mutex
);
2434 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2437 mutex_lock(&fs_info
->chunk_mutex
);
2438 ret
= init_first_rw_device(trans
, fs_info
);
2439 mutex_unlock(&fs_info
->chunk_mutex
);
2441 btrfs_abort_transaction(trans
, ret
);
2446 ret
= btrfs_add_device(trans
, fs_info
, device
);
2448 btrfs_abort_transaction(trans
, ret
);
2453 char fsid_buf
[BTRFS_UUID_UNPARSED_SIZE
];
2455 ret
= btrfs_finish_sprout(trans
, fs_info
);
2457 btrfs_abort_transaction(trans
, ret
);
2461 /* Sprouting would change fsid of the mounted root,
2462 * so rename the fsid on the sysfs
2464 snprintf(fsid_buf
, BTRFS_UUID_UNPARSED_SIZE
, "%pU",
2466 if (kobject_rename(&fs_info
->fs_devices
->fsid_kobj
, fsid_buf
))
2468 "sysfs: failed to create fsid for sprout");
2471 ret
= btrfs_commit_transaction(trans
);
2474 mutex_unlock(&uuid_mutex
);
2475 up_write(&sb
->s_umount
);
2477 if (ret
) /* transaction commit */
2480 ret
= btrfs_relocate_sys_chunks(fs_info
);
2482 btrfs_handle_fs_error(fs_info
, ret
,
2483 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2484 trans
= btrfs_attach_transaction(root
);
2485 if (IS_ERR(trans
)) {
2486 if (PTR_ERR(trans
) == -ENOENT
)
2488 return PTR_ERR(trans
);
2490 ret
= btrfs_commit_transaction(trans
);
2493 /* Update ctime/mtime for libblkid */
2494 update_dev_time(device_path
);
2498 btrfs_end_transaction(trans
);
2499 rcu_string_free(device
->name
);
2500 btrfs_sysfs_rm_device_link(fs_info
->fs_devices
, device
);
2503 blkdev_put(bdev
, FMODE_EXCL
);
2505 mutex_unlock(&uuid_mutex
);
2506 up_write(&sb
->s_umount
);
2511 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
2512 const char *device_path
,
2513 struct btrfs_device
*srcdev
,
2514 struct btrfs_device
**device_out
)
2516 struct request_queue
*q
;
2517 struct btrfs_device
*device
;
2518 struct block_device
*bdev
;
2519 struct list_head
*devices
;
2520 struct rcu_string
*name
;
2521 u64 devid
= BTRFS_DEV_REPLACE_DEVID
;
2525 if (fs_info
->fs_devices
->seeding
) {
2526 btrfs_err(fs_info
, "the filesystem is a seed filesystem!");
2530 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2531 fs_info
->bdev_holder
);
2533 btrfs_err(fs_info
, "target device %s is invalid!", device_path
);
2534 return PTR_ERR(bdev
);
2537 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2539 devices
= &fs_info
->fs_devices
->devices
;
2540 list_for_each_entry(device
, devices
, dev_list
) {
2541 if (device
->bdev
== bdev
) {
2543 "target device is in the filesystem!");
2550 if (i_size_read(bdev
->bd_inode
) <
2551 btrfs_device_get_total_bytes(srcdev
)) {
2553 "target device is smaller than source device!");
2559 device
= btrfs_alloc_device(NULL
, &devid
, NULL
);
2560 if (IS_ERR(device
)) {
2561 ret
= PTR_ERR(device
);
2565 name
= rcu_string_strdup(device_path
, GFP_KERNEL
);
2571 rcu_assign_pointer(device
->name
, name
);
2573 q
= bdev_get_queue(bdev
);
2574 if (blk_queue_discard(q
))
2575 device
->can_discard
= 1;
2576 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2577 device
->writeable
= 1;
2578 device
->generation
= 0;
2579 device
->io_width
= fs_info
->sectorsize
;
2580 device
->io_align
= fs_info
->sectorsize
;
2581 device
->sector_size
= fs_info
->sectorsize
;
2582 device
->total_bytes
= btrfs_device_get_total_bytes(srcdev
);
2583 device
->disk_total_bytes
= btrfs_device_get_disk_total_bytes(srcdev
);
2584 device
->bytes_used
= btrfs_device_get_bytes_used(srcdev
);
2585 ASSERT(list_empty(&srcdev
->resized_list
));
2586 device
->commit_total_bytes
= srcdev
->commit_total_bytes
;
2587 device
->commit_bytes_used
= device
->bytes_used
;
2588 device
->fs_info
= fs_info
;
2589 device
->bdev
= bdev
;
2590 device
->in_fs_metadata
= 1;
2591 device
->is_tgtdev_for_dev_replace
= 1;
2592 device
->mode
= FMODE_EXCL
;
2593 device
->dev_stats_valid
= 1;
2594 set_blocksize(device
->bdev
, BTRFS_BDEV_BLOCKSIZE
);
2595 device
->fs_devices
= fs_info
->fs_devices
;
2596 list_add(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2597 fs_info
->fs_devices
->num_devices
++;
2598 fs_info
->fs_devices
->open_devices
++;
2599 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2601 *device_out
= device
;
2605 blkdev_put(bdev
, FMODE_EXCL
);
2609 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info
*fs_info
,
2610 struct btrfs_device
*tgtdev
)
2612 u32 sectorsize
= fs_info
->sectorsize
;
2614 WARN_ON(fs_info
->fs_devices
->rw_devices
== 0);
2615 tgtdev
->io_width
= sectorsize
;
2616 tgtdev
->io_align
= sectorsize
;
2617 tgtdev
->sector_size
= sectorsize
;
2618 tgtdev
->fs_info
= fs_info
;
2619 tgtdev
->in_fs_metadata
= 1;
2622 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2623 struct btrfs_device
*device
)
2626 struct btrfs_path
*path
;
2627 struct btrfs_root
*root
= device
->fs_info
->chunk_root
;
2628 struct btrfs_dev_item
*dev_item
;
2629 struct extent_buffer
*leaf
;
2630 struct btrfs_key key
;
2632 path
= btrfs_alloc_path();
2636 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2637 key
.type
= BTRFS_DEV_ITEM_KEY
;
2638 key
.offset
= device
->devid
;
2640 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2649 leaf
= path
->nodes
[0];
2650 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2652 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2653 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2654 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2655 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2656 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2657 btrfs_set_device_total_bytes(leaf
, dev_item
,
2658 btrfs_device_get_disk_total_bytes(device
));
2659 btrfs_set_device_bytes_used(leaf
, dev_item
,
2660 btrfs_device_get_bytes_used(device
));
2661 btrfs_mark_buffer_dirty(leaf
);
2664 btrfs_free_path(path
);
2668 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2669 struct btrfs_device
*device
, u64 new_size
)
2671 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
2672 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
2673 struct btrfs_fs_devices
*fs_devices
;
2677 if (!device
->writeable
)
2680 new_size
= round_down(new_size
, fs_info
->sectorsize
);
2682 mutex_lock(&fs_info
->chunk_mutex
);
2683 old_total
= btrfs_super_total_bytes(super_copy
);
2684 diff
= round_down(new_size
- device
->total_bytes
, fs_info
->sectorsize
);
2686 if (new_size
<= device
->total_bytes
||
2687 device
->is_tgtdev_for_dev_replace
) {
2688 mutex_unlock(&fs_info
->chunk_mutex
);
2692 fs_devices
= fs_info
->fs_devices
;
2694 btrfs_set_super_total_bytes(super_copy
,
2695 round_down(old_total
+ diff
, fs_info
->sectorsize
));
2696 device
->fs_devices
->total_rw_bytes
+= diff
;
2698 btrfs_device_set_total_bytes(device
, new_size
);
2699 btrfs_device_set_disk_total_bytes(device
, new_size
);
2700 btrfs_clear_space_info_full(device
->fs_info
);
2701 if (list_empty(&device
->resized_list
))
2702 list_add_tail(&device
->resized_list
,
2703 &fs_devices
->resized_devices
);
2704 mutex_unlock(&fs_info
->chunk_mutex
);
2706 return btrfs_update_device(trans
, device
);
2709 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
2710 struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2712 struct btrfs_root
*root
= fs_info
->chunk_root
;
2714 struct btrfs_path
*path
;
2715 struct btrfs_key key
;
2717 path
= btrfs_alloc_path();
2721 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2722 key
.offset
= chunk_offset
;
2723 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2725 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2728 else if (ret
> 0) { /* Logic error or corruption */
2729 btrfs_handle_fs_error(fs_info
, -ENOENT
,
2730 "Failed lookup while freeing chunk.");
2735 ret
= btrfs_del_item(trans
, root
, path
);
2737 btrfs_handle_fs_error(fs_info
, ret
,
2738 "Failed to delete chunk item.");
2740 btrfs_free_path(path
);
2744 static int btrfs_del_sys_chunk(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2746 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
2747 struct btrfs_disk_key
*disk_key
;
2748 struct btrfs_chunk
*chunk
;
2755 struct btrfs_key key
;
2757 mutex_lock(&fs_info
->chunk_mutex
);
2758 array_size
= btrfs_super_sys_array_size(super_copy
);
2760 ptr
= super_copy
->sys_chunk_array
;
2763 while (cur
< array_size
) {
2764 disk_key
= (struct btrfs_disk_key
*)ptr
;
2765 btrfs_disk_key_to_cpu(&key
, disk_key
);
2767 len
= sizeof(*disk_key
);
2769 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2770 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2771 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2772 len
+= btrfs_chunk_item_size(num_stripes
);
2777 if (key
.objectid
== BTRFS_FIRST_CHUNK_TREE_OBJECTID
&&
2778 key
.offset
== chunk_offset
) {
2779 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2781 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2787 mutex_unlock(&fs_info
->chunk_mutex
);
2791 static struct extent_map
*get_chunk_map(struct btrfs_fs_info
*fs_info
,
2792 u64 logical
, u64 length
)
2794 struct extent_map_tree
*em_tree
;
2795 struct extent_map
*em
;
2797 em_tree
= &fs_info
->mapping_tree
.map_tree
;
2798 read_lock(&em_tree
->lock
);
2799 em
= lookup_extent_mapping(em_tree
, logical
, length
);
2800 read_unlock(&em_tree
->lock
);
2803 btrfs_crit(fs_info
, "unable to find logical %llu length %llu",
2805 return ERR_PTR(-EINVAL
);
2808 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
2810 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2811 logical
, length
, em
->start
, em
->start
+ em
->len
);
2812 free_extent_map(em
);
2813 return ERR_PTR(-EINVAL
);
2816 /* callers are responsible for dropping em's ref. */
2820 int btrfs_remove_chunk(struct btrfs_trans_handle
*trans
,
2821 struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2823 struct extent_map
*em
;
2824 struct map_lookup
*map
;
2825 u64 dev_extent_len
= 0;
2827 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2829 em
= get_chunk_map(fs_info
, chunk_offset
, 1);
2832 * This is a logic error, but we don't want to just rely on the
2833 * user having built with ASSERT enabled, so if ASSERT doesn't
2834 * do anything we still error out.
2839 map
= em
->map_lookup
;
2840 mutex_lock(&fs_info
->chunk_mutex
);
2841 check_system_chunk(trans
, fs_info
, map
->type
);
2842 mutex_unlock(&fs_info
->chunk_mutex
);
2845 * Take the device list mutex to prevent races with the final phase of
2846 * a device replace operation that replaces the device object associated
2847 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2849 mutex_lock(&fs_devices
->device_list_mutex
);
2850 for (i
= 0; i
< map
->num_stripes
; i
++) {
2851 struct btrfs_device
*device
= map
->stripes
[i
].dev
;
2852 ret
= btrfs_free_dev_extent(trans
, device
,
2853 map
->stripes
[i
].physical
,
2856 mutex_unlock(&fs_devices
->device_list_mutex
);
2857 btrfs_abort_transaction(trans
, ret
);
2861 if (device
->bytes_used
> 0) {
2862 mutex_lock(&fs_info
->chunk_mutex
);
2863 btrfs_device_set_bytes_used(device
,
2864 device
->bytes_used
- dev_extent_len
);
2865 atomic64_add(dev_extent_len
, &fs_info
->free_chunk_space
);
2866 btrfs_clear_space_info_full(fs_info
);
2867 mutex_unlock(&fs_info
->chunk_mutex
);
2870 if (map
->stripes
[i
].dev
) {
2871 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2873 mutex_unlock(&fs_devices
->device_list_mutex
);
2874 btrfs_abort_transaction(trans
, ret
);
2879 mutex_unlock(&fs_devices
->device_list_mutex
);
2881 ret
= btrfs_free_chunk(trans
, fs_info
, chunk_offset
);
2883 btrfs_abort_transaction(trans
, ret
);
2887 trace_btrfs_chunk_free(fs_info
, map
, chunk_offset
, em
->len
);
2889 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2890 ret
= btrfs_del_sys_chunk(fs_info
, chunk_offset
);
2892 btrfs_abort_transaction(trans
, ret
);
2897 ret
= btrfs_remove_block_group(trans
, fs_info
, chunk_offset
, em
);
2899 btrfs_abort_transaction(trans
, ret
);
2905 free_extent_map(em
);
2909 static int btrfs_relocate_chunk(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2911 struct btrfs_root
*root
= fs_info
->chunk_root
;
2912 struct btrfs_trans_handle
*trans
;
2916 * Prevent races with automatic removal of unused block groups.
2917 * After we relocate and before we remove the chunk with offset
2918 * chunk_offset, automatic removal of the block group can kick in,
2919 * resulting in a failure when calling btrfs_remove_chunk() below.
2921 * Make sure to acquire this mutex before doing a tree search (dev
2922 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2923 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2924 * we release the path used to search the chunk/dev tree and before
2925 * the current task acquires this mutex and calls us.
2927 ASSERT(mutex_is_locked(&fs_info
->delete_unused_bgs_mutex
));
2929 ret
= btrfs_can_relocate(fs_info
, chunk_offset
);
2933 /* step one, relocate all the extents inside this chunk */
2934 btrfs_scrub_pause(fs_info
);
2935 ret
= btrfs_relocate_block_group(fs_info
, chunk_offset
);
2936 btrfs_scrub_continue(fs_info
);
2940 trans
= btrfs_start_trans_remove_block_group(root
->fs_info
,
2942 if (IS_ERR(trans
)) {
2943 ret
= PTR_ERR(trans
);
2944 btrfs_handle_fs_error(root
->fs_info
, ret
, NULL
);
2949 * step two, delete the device extents and the
2950 * chunk tree entries
2952 ret
= btrfs_remove_chunk(trans
, fs_info
, chunk_offset
);
2953 btrfs_end_transaction(trans
);
2957 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info
*fs_info
)
2959 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
2960 struct btrfs_path
*path
;
2961 struct extent_buffer
*leaf
;
2962 struct btrfs_chunk
*chunk
;
2963 struct btrfs_key key
;
2964 struct btrfs_key found_key
;
2966 bool retried
= false;
2970 path
= btrfs_alloc_path();
2975 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2976 key
.offset
= (u64
)-1;
2977 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2980 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
2981 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2983 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
2986 BUG_ON(ret
== 0); /* Corruption */
2988 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
2991 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
2997 leaf
= path
->nodes
[0];
2998 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3000 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
3001 struct btrfs_chunk
);
3002 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3003 btrfs_release_path(path
);
3005 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3006 ret
= btrfs_relocate_chunk(fs_info
, found_key
.offset
);
3012 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3014 if (found_key
.offset
== 0)
3016 key
.offset
= found_key
.offset
- 1;
3019 if (failed
&& !retried
) {
3023 } else if (WARN_ON(failed
&& retried
)) {
3027 btrfs_free_path(path
);
3031 static int insert_balance_item(struct btrfs_fs_info
*fs_info
,
3032 struct btrfs_balance_control
*bctl
)
3034 struct btrfs_root
*root
= fs_info
->tree_root
;
3035 struct btrfs_trans_handle
*trans
;
3036 struct btrfs_balance_item
*item
;
3037 struct btrfs_disk_balance_args disk_bargs
;
3038 struct btrfs_path
*path
;
3039 struct extent_buffer
*leaf
;
3040 struct btrfs_key key
;
3043 path
= btrfs_alloc_path();
3047 trans
= btrfs_start_transaction(root
, 0);
3048 if (IS_ERR(trans
)) {
3049 btrfs_free_path(path
);
3050 return PTR_ERR(trans
);
3053 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3054 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3057 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
3062 leaf
= path
->nodes
[0];
3063 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3065 memzero_extent_buffer(leaf
, (unsigned long)item
, sizeof(*item
));
3067 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
3068 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
3069 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
3070 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
3071 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
3072 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
3074 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
3076 btrfs_mark_buffer_dirty(leaf
);
3078 btrfs_free_path(path
);
3079 err
= btrfs_commit_transaction(trans
);
3085 static int del_balance_item(struct btrfs_fs_info
*fs_info
)
3087 struct btrfs_root
*root
= fs_info
->tree_root
;
3088 struct btrfs_trans_handle
*trans
;
3089 struct btrfs_path
*path
;
3090 struct btrfs_key key
;
3093 path
= btrfs_alloc_path();
3097 trans
= btrfs_start_transaction(root
, 0);
3098 if (IS_ERR(trans
)) {
3099 btrfs_free_path(path
);
3100 return PTR_ERR(trans
);
3103 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3104 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3107 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3115 ret
= btrfs_del_item(trans
, root
, path
);
3117 btrfs_free_path(path
);
3118 err
= btrfs_commit_transaction(trans
);
3125 * This is a heuristic used to reduce the number of chunks balanced on
3126 * resume after balance was interrupted.
3128 static void update_balance_args(struct btrfs_balance_control
*bctl
)
3131 * Turn on soft mode for chunk types that were being converted.
3133 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3134 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3135 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3136 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3137 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3138 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3141 * Turn on usage filter if is not already used. The idea is
3142 * that chunks that we have already balanced should be
3143 * reasonably full. Don't do it for chunks that are being
3144 * converted - that will keep us from relocating unconverted
3145 * (albeit full) chunks.
3147 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3148 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3149 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3150 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3151 bctl
->data
.usage
= 90;
3153 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3154 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3155 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3156 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3157 bctl
->sys
.usage
= 90;
3159 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3160 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3161 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3162 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3163 bctl
->meta
.usage
= 90;
3168 * Should be called with both balance and volume mutexes held to
3169 * serialize other volume operations (add_dev/rm_dev/resize) with
3170 * restriper. Same goes for unset_balance_control.
3172 static void set_balance_control(struct btrfs_balance_control
*bctl
)
3174 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3176 BUG_ON(fs_info
->balance_ctl
);
3178 spin_lock(&fs_info
->balance_lock
);
3179 fs_info
->balance_ctl
= bctl
;
3180 spin_unlock(&fs_info
->balance_lock
);
3183 static void unset_balance_control(struct btrfs_fs_info
*fs_info
)
3185 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3187 BUG_ON(!fs_info
->balance_ctl
);
3189 spin_lock(&fs_info
->balance_lock
);
3190 fs_info
->balance_ctl
= NULL
;
3191 spin_unlock(&fs_info
->balance_lock
);
3197 * Balance filters. Return 1 if chunk should be filtered out
3198 * (should not be balanced).
3200 static int chunk_profiles_filter(u64 chunk_type
,
3201 struct btrfs_balance_args
*bargs
)
3203 chunk_type
= chunk_to_extended(chunk_type
) &
3204 BTRFS_EXTENDED_PROFILE_MASK
;
3206 if (bargs
->profiles
& chunk_type
)
3212 static int chunk_usage_range_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
3213 struct btrfs_balance_args
*bargs
)
3215 struct btrfs_block_group_cache
*cache
;
3217 u64 user_thresh_min
;
3218 u64 user_thresh_max
;
3221 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3222 chunk_used
= btrfs_block_group_used(&cache
->item
);
3224 if (bargs
->usage_min
== 0)
3225 user_thresh_min
= 0;
3227 user_thresh_min
= div_factor_fine(cache
->key
.offset
,
3230 if (bargs
->usage_max
== 0)
3231 user_thresh_max
= 1;
3232 else if (bargs
->usage_max
> 100)
3233 user_thresh_max
= cache
->key
.offset
;
3235 user_thresh_max
= div_factor_fine(cache
->key
.offset
,
3238 if (user_thresh_min
<= chunk_used
&& chunk_used
< user_thresh_max
)
3241 btrfs_put_block_group(cache
);
3245 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
,
3246 u64 chunk_offset
, struct btrfs_balance_args
*bargs
)
3248 struct btrfs_block_group_cache
*cache
;
3249 u64 chunk_used
, user_thresh
;
3252 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3253 chunk_used
= btrfs_block_group_used(&cache
->item
);
3255 if (bargs
->usage_min
== 0)
3257 else if (bargs
->usage
> 100)
3258 user_thresh
= cache
->key
.offset
;
3260 user_thresh
= div_factor_fine(cache
->key
.offset
,
3263 if (chunk_used
< user_thresh
)
3266 btrfs_put_block_group(cache
);
3270 static int chunk_devid_filter(struct extent_buffer
*leaf
,
3271 struct btrfs_chunk
*chunk
,
3272 struct btrfs_balance_args
*bargs
)
3274 struct btrfs_stripe
*stripe
;
3275 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3278 for (i
= 0; i
< num_stripes
; i
++) {
3279 stripe
= btrfs_stripe_nr(chunk
, i
);
3280 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
3287 /* [pstart, pend) */
3288 static int chunk_drange_filter(struct extent_buffer
*leaf
,
3289 struct btrfs_chunk
*chunk
,
3290 struct btrfs_balance_args
*bargs
)
3292 struct btrfs_stripe
*stripe
;
3293 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3299 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
3302 if (btrfs_chunk_type(leaf
, chunk
) & (BTRFS_BLOCK_GROUP_DUP
|
3303 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)) {
3304 factor
= num_stripes
/ 2;
3305 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID5
) {
3306 factor
= num_stripes
- 1;
3307 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID6
) {
3308 factor
= num_stripes
- 2;
3310 factor
= num_stripes
;
3313 for (i
= 0; i
< num_stripes
; i
++) {
3314 stripe
= btrfs_stripe_nr(chunk
, i
);
3315 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
3318 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
3319 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
3320 stripe_length
= div_u64(stripe_length
, factor
);
3322 if (stripe_offset
< bargs
->pend
&&
3323 stripe_offset
+ stripe_length
> bargs
->pstart
)
3330 /* [vstart, vend) */
3331 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
3332 struct btrfs_chunk
*chunk
,
3334 struct btrfs_balance_args
*bargs
)
3336 if (chunk_offset
< bargs
->vend
&&
3337 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
3338 /* at least part of the chunk is inside this vrange */
3344 static int chunk_stripes_range_filter(struct extent_buffer
*leaf
,
3345 struct btrfs_chunk
*chunk
,
3346 struct btrfs_balance_args
*bargs
)
3348 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3350 if (bargs
->stripes_min
<= num_stripes
3351 && num_stripes
<= bargs
->stripes_max
)
3357 static int chunk_soft_convert_filter(u64 chunk_type
,
3358 struct btrfs_balance_args
*bargs
)
3360 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
3363 chunk_type
= chunk_to_extended(chunk_type
) &
3364 BTRFS_EXTENDED_PROFILE_MASK
;
3366 if (bargs
->target
== chunk_type
)
3372 static int should_balance_chunk(struct btrfs_fs_info
*fs_info
,
3373 struct extent_buffer
*leaf
,
3374 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
3376 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3377 struct btrfs_balance_args
*bargs
= NULL
;
3378 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3381 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
3382 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
3386 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3387 bargs
= &bctl
->data
;
3388 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3390 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3391 bargs
= &bctl
->meta
;
3393 /* profiles filter */
3394 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
3395 chunk_profiles_filter(chunk_type
, bargs
)) {
3400 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3401 chunk_usage_filter(fs_info
, chunk_offset
, bargs
)) {
3403 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3404 chunk_usage_range_filter(fs_info
, chunk_offset
, bargs
)) {
3409 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
3410 chunk_devid_filter(leaf
, chunk
, bargs
)) {
3414 /* drange filter, makes sense only with devid filter */
3415 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
3416 chunk_drange_filter(leaf
, chunk
, bargs
)) {
3421 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
3422 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3426 /* stripes filter */
3427 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_STRIPES_RANGE
) &&
3428 chunk_stripes_range_filter(leaf
, chunk
, bargs
)) {
3432 /* soft profile changing mode */
3433 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
3434 chunk_soft_convert_filter(chunk_type
, bargs
)) {
3439 * limited by count, must be the last filter
3441 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT
)) {
3442 if (bargs
->limit
== 0)
3446 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT_RANGE
)) {
3448 * Same logic as the 'limit' filter; the minimum cannot be
3449 * determined here because we do not have the global information
3450 * about the count of all chunks that satisfy the filters.
3452 if (bargs
->limit_max
== 0)
3461 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
3463 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3464 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
3465 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
3466 struct list_head
*devices
;
3467 struct btrfs_device
*device
;
3471 struct btrfs_chunk
*chunk
;
3472 struct btrfs_path
*path
= NULL
;
3473 struct btrfs_key key
;
3474 struct btrfs_key found_key
;
3475 struct btrfs_trans_handle
*trans
;
3476 struct extent_buffer
*leaf
;
3479 int enospc_errors
= 0;
3480 bool counting
= true;
3481 /* The single value limit and min/max limits use the same bytes in the */
3482 u64 limit_data
= bctl
->data
.limit
;
3483 u64 limit_meta
= bctl
->meta
.limit
;
3484 u64 limit_sys
= bctl
->sys
.limit
;
3488 int chunk_reserved
= 0;
3491 /* step one make some room on all the devices */
3492 devices
= &fs_info
->fs_devices
->devices
;
3493 list_for_each_entry(device
, devices
, dev_list
) {
3494 old_size
= btrfs_device_get_total_bytes(device
);
3495 size_to_free
= div_factor(old_size
, 1);
3496 size_to_free
= min_t(u64
, size_to_free
, SZ_1M
);
3497 if (!device
->writeable
||
3498 btrfs_device_get_total_bytes(device
) -
3499 btrfs_device_get_bytes_used(device
) > size_to_free
||
3500 device
->is_tgtdev_for_dev_replace
)
3503 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
3507 /* btrfs_shrink_device never returns ret > 0 */
3512 trans
= btrfs_start_transaction(dev_root
, 0);
3513 if (IS_ERR(trans
)) {
3514 ret
= PTR_ERR(trans
);
3515 btrfs_info_in_rcu(fs_info
,
3516 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3517 rcu_str_deref(device
->name
), ret
,
3518 old_size
, old_size
- size_to_free
);
3522 ret
= btrfs_grow_device(trans
, device
, old_size
);
3524 btrfs_end_transaction(trans
);
3525 /* btrfs_grow_device never returns ret > 0 */
3527 btrfs_info_in_rcu(fs_info
,
3528 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3529 rcu_str_deref(device
->name
), ret
,
3530 old_size
, old_size
- size_to_free
);
3534 btrfs_end_transaction(trans
);
3537 /* step two, relocate all the chunks */
3538 path
= btrfs_alloc_path();
3544 /* zero out stat counters */
3545 spin_lock(&fs_info
->balance_lock
);
3546 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
3547 spin_unlock(&fs_info
->balance_lock
);
3551 * The single value limit and min/max limits use the same bytes
3554 bctl
->data
.limit
= limit_data
;
3555 bctl
->meta
.limit
= limit_meta
;
3556 bctl
->sys
.limit
= limit_sys
;
3558 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3559 key
.offset
= (u64
)-1;
3560 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3563 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
3564 atomic_read(&fs_info
->balance_cancel_req
)) {
3569 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
3570 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
3572 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3577 * this shouldn't happen, it means the last relocate
3581 BUG(); /* FIXME break ? */
3583 ret
= btrfs_previous_item(chunk_root
, path
, 0,
3584 BTRFS_CHUNK_ITEM_KEY
);
3586 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3591 leaf
= path
->nodes
[0];
3592 slot
= path
->slots
[0];
3593 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3595 if (found_key
.objectid
!= key
.objectid
) {
3596 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3600 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
3601 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3604 spin_lock(&fs_info
->balance_lock
);
3605 bctl
->stat
.considered
++;
3606 spin_unlock(&fs_info
->balance_lock
);
3609 ret
= should_balance_chunk(fs_info
, leaf
, chunk
,
3612 btrfs_release_path(path
);
3614 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3619 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3620 spin_lock(&fs_info
->balance_lock
);
3621 bctl
->stat
.expected
++;
3622 spin_unlock(&fs_info
->balance_lock
);
3624 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3626 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3628 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3635 * Apply limit_min filter, no need to check if the LIMITS
3636 * filter is used, limit_min is 0 by default
3638 if (((chunk_type
& BTRFS_BLOCK_GROUP_DATA
) &&
3639 count_data
< bctl
->data
.limit_min
)
3640 || ((chunk_type
& BTRFS_BLOCK_GROUP_METADATA
) &&
3641 count_meta
< bctl
->meta
.limit_min
)
3642 || ((chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) &&
3643 count_sys
< bctl
->sys
.limit_min
)) {
3644 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3648 ASSERT(fs_info
->data_sinfo
);
3649 spin_lock(&fs_info
->data_sinfo
->lock
);
3650 bytes_used
= fs_info
->data_sinfo
->bytes_used
;
3651 spin_unlock(&fs_info
->data_sinfo
->lock
);
3653 if ((chunk_type
& BTRFS_BLOCK_GROUP_DATA
) &&
3654 !chunk_reserved
&& !bytes_used
) {
3655 trans
= btrfs_start_transaction(chunk_root
, 0);
3656 if (IS_ERR(trans
)) {
3657 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3658 ret
= PTR_ERR(trans
);
3662 ret
= btrfs_force_chunk_alloc(trans
, fs_info
,
3663 BTRFS_BLOCK_GROUP_DATA
);
3664 btrfs_end_transaction(trans
);
3666 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3672 ret
= btrfs_relocate_chunk(fs_info
, found_key
.offset
);
3673 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3674 if (ret
&& ret
!= -ENOSPC
)
3676 if (ret
== -ENOSPC
) {
3679 spin_lock(&fs_info
->balance_lock
);
3680 bctl
->stat
.completed
++;
3681 spin_unlock(&fs_info
->balance_lock
);
3684 if (found_key
.offset
== 0)
3686 key
.offset
= found_key
.offset
- 1;
3690 btrfs_release_path(path
);
3695 btrfs_free_path(path
);
3696 if (enospc_errors
) {
3697 btrfs_info(fs_info
, "%d enospc errors during balance",
3707 * alloc_profile_is_valid - see if a given profile is valid and reduced
3708 * @flags: profile to validate
3709 * @extended: if true @flags is treated as an extended profile
3711 static int alloc_profile_is_valid(u64 flags
, int extended
)
3713 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
3714 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
3716 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
3718 /* 1) check that all other bits are zeroed */
3722 /* 2) see if profile is reduced */
3724 return !extended
; /* "0" is valid for usual profiles */
3726 /* true if exactly one bit set */
3727 return (flags
& (flags
- 1)) == 0;
3730 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
3732 /* cancel requested || normal exit path */
3733 return atomic_read(&fs_info
->balance_cancel_req
) ||
3734 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
3735 atomic_read(&fs_info
->balance_cancel_req
) == 0);
3738 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
3742 unset_balance_control(fs_info
);
3743 ret
= del_balance_item(fs_info
);
3745 btrfs_handle_fs_error(fs_info
, ret
, NULL
);
3747 clear_bit(BTRFS_FS_EXCL_OP
, &fs_info
->flags
);
3750 /* Non-zero return value signifies invalidity */
3751 static inline int validate_convert_profile(struct btrfs_balance_args
*bctl_arg
,
3754 return ((bctl_arg
->flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3755 (!alloc_profile_is_valid(bctl_arg
->target
, 1) ||
3756 (bctl_arg
->target
& ~allowed
)));
3760 * Should be called with both balance and volume mutexes held
3762 int btrfs_balance(struct btrfs_balance_control
*bctl
,
3763 struct btrfs_ioctl_balance_args
*bargs
)
3765 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3766 u64 meta_target
, data_target
;
3773 if (btrfs_fs_closing(fs_info
) ||
3774 atomic_read(&fs_info
->balance_pause_req
) ||
3775 atomic_read(&fs_info
->balance_cancel_req
)) {
3780 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
3781 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
3785 * In case of mixed groups both data and meta should be picked,
3786 * and identical options should be given for both of them.
3788 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
3789 if (mixed
&& (bctl
->flags
& allowed
)) {
3790 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
3791 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
3792 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
3794 "with mixed groups data and metadata balance options must be the same");
3800 num_devices
= fs_info
->fs_devices
->num_devices
;
3801 btrfs_dev_replace_lock(&fs_info
->dev_replace
, 0);
3802 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
3803 BUG_ON(num_devices
< 1);
3806 btrfs_dev_replace_unlock(&fs_info
->dev_replace
, 0);
3807 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
| BTRFS_BLOCK_GROUP_DUP
;
3808 if (num_devices
> 1)
3809 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
3810 if (num_devices
> 2)
3811 allowed
|= BTRFS_BLOCK_GROUP_RAID5
;
3812 if (num_devices
> 3)
3813 allowed
|= (BTRFS_BLOCK_GROUP_RAID10
|
3814 BTRFS_BLOCK_GROUP_RAID6
);
3815 if (validate_convert_profile(&bctl
->data
, allowed
)) {
3817 "unable to start balance with target data profile %llu",
3822 if (validate_convert_profile(&bctl
->meta
, allowed
)) {
3824 "unable to start balance with target metadata profile %llu",
3829 if (validate_convert_profile(&bctl
->sys
, allowed
)) {
3831 "unable to start balance with target system profile %llu",
3837 /* allow to reduce meta or sys integrity only if force set */
3838 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3839 BTRFS_BLOCK_GROUP_RAID10
|
3840 BTRFS_BLOCK_GROUP_RAID5
|
3841 BTRFS_BLOCK_GROUP_RAID6
;
3843 seq
= read_seqbegin(&fs_info
->profiles_lock
);
3845 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3846 (fs_info
->avail_system_alloc_bits
& allowed
) &&
3847 !(bctl
->sys
.target
& allowed
)) ||
3848 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3849 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
3850 !(bctl
->meta
.target
& allowed
))) {
3851 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
3853 "force reducing metadata integrity");
3856 "balance will reduce metadata integrity, use force if you want this");
3861 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
3863 /* if we're not converting, the target field is uninitialized */
3864 meta_target
= (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) ?
3865 bctl
->meta
.target
: fs_info
->avail_metadata_alloc_bits
;
3866 data_target
= (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) ?
3867 bctl
->data
.target
: fs_info
->avail_data_alloc_bits
;
3868 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target
) <
3869 btrfs_get_num_tolerated_disk_barrier_failures(data_target
)) {
3871 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3872 meta_target
, data_target
);
3875 ret
= insert_balance_item(fs_info
, bctl
);
3876 if (ret
&& ret
!= -EEXIST
)
3879 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
3880 BUG_ON(ret
== -EEXIST
);
3881 set_balance_control(bctl
);
3883 BUG_ON(ret
!= -EEXIST
);
3884 spin_lock(&fs_info
->balance_lock
);
3885 update_balance_args(bctl
);
3886 spin_unlock(&fs_info
->balance_lock
);
3889 atomic_inc(&fs_info
->balance_running
);
3890 mutex_unlock(&fs_info
->balance_mutex
);
3892 ret
= __btrfs_balance(fs_info
);
3894 mutex_lock(&fs_info
->balance_mutex
);
3895 atomic_dec(&fs_info
->balance_running
);
3898 memset(bargs
, 0, sizeof(*bargs
));
3899 update_ioctl_balance_args(fs_info
, 0, bargs
);
3902 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
3903 balance_need_close(fs_info
)) {
3904 __cancel_balance(fs_info
);
3907 wake_up(&fs_info
->balance_wait_q
);
3911 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
3912 __cancel_balance(fs_info
);
3915 clear_bit(BTRFS_FS_EXCL_OP
, &fs_info
->flags
);
3920 static int balance_kthread(void *data
)
3922 struct btrfs_fs_info
*fs_info
= data
;
3925 mutex_lock(&fs_info
->volume_mutex
);
3926 mutex_lock(&fs_info
->balance_mutex
);
3928 if (fs_info
->balance_ctl
) {
3929 btrfs_info(fs_info
, "continuing balance");
3930 ret
= btrfs_balance(fs_info
->balance_ctl
, NULL
);
3933 mutex_unlock(&fs_info
->balance_mutex
);
3934 mutex_unlock(&fs_info
->volume_mutex
);
3939 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
3941 struct task_struct
*tsk
;
3943 spin_lock(&fs_info
->balance_lock
);
3944 if (!fs_info
->balance_ctl
) {
3945 spin_unlock(&fs_info
->balance_lock
);
3948 spin_unlock(&fs_info
->balance_lock
);
3950 if (btrfs_test_opt(fs_info
, SKIP_BALANCE
)) {
3951 btrfs_info(fs_info
, "force skipping balance");
3955 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
3956 return PTR_ERR_OR_ZERO(tsk
);
3959 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
3961 struct btrfs_balance_control
*bctl
;
3962 struct btrfs_balance_item
*item
;
3963 struct btrfs_disk_balance_args disk_bargs
;
3964 struct btrfs_path
*path
;
3965 struct extent_buffer
*leaf
;
3966 struct btrfs_key key
;
3969 path
= btrfs_alloc_path();
3973 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3974 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3977 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
3980 if (ret
> 0) { /* ret = -ENOENT; */
3985 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
3991 leaf
= path
->nodes
[0];
3992 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3994 bctl
->fs_info
= fs_info
;
3995 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
3996 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
3998 btrfs_balance_data(leaf
, item
, &disk_bargs
);
3999 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
4000 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
4001 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
4002 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
4003 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
4005 WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP
, &fs_info
->flags
));
4007 mutex_lock(&fs_info
->volume_mutex
);
4008 mutex_lock(&fs_info
->balance_mutex
);
4010 set_balance_control(bctl
);
4012 mutex_unlock(&fs_info
->balance_mutex
);
4013 mutex_unlock(&fs_info
->volume_mutex
);
4015 btrfs_free_path(path
);
4019 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
4023 mutex_lock(&fs_info
->balance_mutex
);
4024 if (!fs_info
->balance_ctl
) {
4025 mutex_unlock(&fs_info
->balance_mutex
);
4029 if (atomic_read(&fs_info
->balance_running
)) {
4030 atomic_inc(&fs_info
->balance_pause_req
);
4031 mutex_unlock(&fs_info
->balance_mutex
);
4033 wait_event(fs_info
->balance_wait_q
,
4034 atomic_read(&fs_info
->balance_running
) == 0);
4036 mutex_lock(&fs_info
->balance_mutex
);
4037 /* we are good with balance_ctl ripped off from under us */
4038 BUG_ON(atomic_read(&fs_info
->balance_running
));
4039 atomic_dec(&fs_info
->balance_pause_req
);
4044 mutex_unlock(&fs_info
->balance_mutex
);
4048 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
4050 if (sb_rdonly(fs_info
->sb
))
4053 mutex_lock(&fs_info
->balance_mutex
);
4054 if (!fs_info
->balance_ctl
) {
4055 mutex_unlock(&fs_info
->balance_mutex
);
4059 atomic_inc(&fs_info
->balance_cancel_req
);
4061 * if we are running just wait and return, balance item is
4062 * deleted in btrfs_balance in this case
4064 if (atomic_read(&fs_info
->balance_running
)) {
4065 mutex_unlock(&fs_info
->balance_mutex
);
4066 wait_event(fs_info
->balance_wait_q
,
4067 atomic_read(&fs_info
->balance_running
) == 0);
4068 mutex_lock(&fs_info
->balance_mutex
);
4070 /* __cancel_balance needs volume_mutex */
4071 mutex_unlock(&fs_info
->balance_mutex
);
4072 mutex_lock(&fs_info
->volume_mutex
);
4073 mutex_lock(&fs_info
->balance_mutex
);
4075 if (fs_info
->balance_ctl
)
4076 __cancel_balance(fs_info
);
4078 mutex_unlock(&fs_info
->volume_mutex
);
4081 BUG_ON(fs_info
->balance_ctl
|| atomic_read(&fs_info
->balance_running
));
4082 atomic_dec(&fs_info
->balance_cancel_req
);
4083 mutex_unlock(&fs_info
->balance_mutex
);
4087 static int btrfs_uuid_scan_kthread(void *data
)
4089 struct btrfs_fs_info
*fs_info
= data
;
4090 struct btrfs_root
*root
= fs_info
->tree_root
;
4091 struct btrfs_key key
;
4092 struct btrfs_path
*path
= NULL
;
4094 struct extent_buffer
*eb
;
4096 struct btrfs_root_item root_item
;
4098 struct btrfs_trans_handle
*trans
= NULL
;
4100 path
= btrfs_alloc_path();
4107 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4111 ret
= btrfs_search_forward(root
, &key
, path
, 0);
4118 if (key
.type
!= BTRFS_ROOT_ITEM_KEY
||
4119 (key
.objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
4120 key
.objectid
!= BTRFS_FS_TREE_OBJECTID
) ||
4121 key
.objectid
> BTRFS_LAST_FREE_OBJECTID
)
4124 eb
= path
->nodes
[0];
4125 slot
= path
->slots
[0];
4126 item_size
= btrfs_item_size_nr(eb
, slot
);
4127 if (item_size
< sizeof(root_item
))
4130 read_extent_buffer(eb
, &root_item
,
4131 btrfs_item_ptr_offset(eb
, slot
),
4132 (int)sizeof(root_item
));
4133 if (btrfs_root_refs(&root_item
) == 0)
4136 if (!btrfs_is_empty_uuid(root_item
.uuid
) ||
4137 !btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4141 btrfs_release_path(path
);
4143 * 1 - subvol uuid item
4144 * 1 - received_subvol uuid item
4146 trans
= btrfs_start_transaction(fs_info
->uuid_root
, 2);
4147 if (IS_ERR(trans
)) {
4148 ret
= PTR_ERR(trans
);
4156 if (!btrfs_is_empty_uuid(root_item
.uuid
)) {
4157 ret
= btrfs_uuid_tree_add(trans
, fs_info
,
4159 BTRFS_UUID_KEY_SUBVOL
,
4162 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4168 if (!btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4169 ret
= btrfs_uuid_tree_add(trans
, fs_info
,
4170 root_item
.received_uuid
,
4171 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
4174 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4182 ret
= btrfs_end_transaction(trans
);
4188 btrfs_release_path(path
);
4189 if (key
.offset
< (u64
)-1) {
4191 } else if (key
.type
< BTRFS_ROOT_ITEM_KEY
) {
4193 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4194 } else if (key
.objectid
< (u64
)-1) {
4196 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4205 btrfs_free_path(path
);
4206 if (trans
&& !IS_ERR(trans
))
4207 btrfs_end_transaction(trans
);
4209 btrfs_warn(fs_info
, "btrfs_uuid_scan_kthread failed %d", ret
);
4211 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN
, &fs_info
->flags
);
4212 up(&fs_info
->uuid_tree_rescan_sem
);
4217 * Callback for btrfs_uuid_tree_iterate().
4219 * 0 check succeeded, the entry is not outdated.
4220 * < 0 if an error occurred.
4221 * > 0 if the check failed, which means the caller shall remove the entry.
4223 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info
*fs_info
,
4224 u8
*uuid
, u8 type
, u64 subid
)
4226 struct btrfs_key key
;
4228 struct btrfs_root
*subvol_root
;
4230 if (type
!= BTRFS_UUID_KEY_SUBVOL
&&
4231 type
!= BTRFS_UUID_KEY_RECEIVED_SUBVOL
)
4234 key
.objectid
= subid
;
4235 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4236 key
.offset
= (u64
)-1;
4237 subvol_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
4238 if (IS_ERR(subvol_root
)) {
4239 ret
= PTR_ERR(subvol_root
);
4246 case BTRFS_UUID_KEY_SUBVOL
:
4247 if (memcmp(uuid
, subvol_root
->root_item
.uuid
, BTRFS_UUID_SIZE
))
4250 case BTRFS_UUID_KEY_RECEIVED_SUBVOL
:
4251 if (memcmp(uuid
, subvol_root
->root_item
.received_uuid
,
4261 static int btrfs_uuid_rescan_kthread(void *data
)
4263 struct btrfs_fs_info
*fs_info
= (struct btrfs_fs_info
*)data
;
4267 * 1st step is to iterate through the existing UUID tree and
4268 * to delete all entries that contain outdated data.
4269 * 2nd step is to add all missing entries to the UUID tree.
4271 ret
= btrfs_uuid_tree_iterate(fs_info
, btrfs_check_uuid_tree_entry
);
4273 btrfs_warn(fs_info
, "iterating uuid_tree failed %d", ret
);
4274 up(&fs_info
->uuid_tree_rescan_sem
);
4277 return btrfs_uuid_scan_kthread(data
);
4280 int btrfs_create_uuid_tree(struct btrfs_fs_info
*fs_info
)
4282 struct btrfs_trans_handle
*trans
;
4283 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
4284 struct btrfs_root
*uuid_root
;
4285 struct task_struct
*task
;
4292 trans
= btrfs_start_transaction(tree_root
, 2);
4294 return PTR_ERR(trans
);
4296 uuid_root
= btrfs_create_tree(trans
, fs_info
,
4297 BTRFS_UUID_TREE_OBJECTID
);
4298 if (IS_ERR(uuid_root
)) {
4299 ret
= PTR_ERR(uuid_root
);
4300 btrfs_abort_transaction(trans
, ret
);
4301 btrfs_end_transaction(trans
);
4305 fs_info
->uuid_root
= uuid_root
;
4307 ret
= btrfs_commit_transaction(trans
);
4311 down(&fs_info
->uuid_tree_rescan_sem
);
4312 task
= kthread_run(btrfs_uuid_scan_kthread
, fs_info
, "btrfs-uuid");
4314 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4315 btrfs_warn(fs_info
, "failed to start uuid_scan task");
4316 up(&fs_info
->uuid_tree_rescan_sem
);
4317 return PTR_ERR(task
);
4323 int btrfs_check_uuid_tree(struct btrfs_fs_info
*fs_info
)
4325 struct task_struct
*task
;
4327 down(&fs_info
->uuid_tree_rescan_sem
);
4328 task
= kthread_run(btrfs_uuid_rescan_kthread
, fs_info
, "btrfs-uuid");
4330 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4331 btrfs_warn(fs_info
, "failed to start uuid_rescan task");
4332 up(&fs_info
->uuid_tree_rescan_sem
);
4333 return PTR_ERR(task
);
4340 * shrinking a device means finding all of the device extents past
4341 * the new size, and then following the back refs to the chunks.
4342 * The chunk relocation code actually frees the device extent
4344 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
4346 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
4347 struct btrfs_root
*root
= fs_info
->dev_root
;
4348 struct btrfs_trans_handle
*trans
;
4349 struct btrfs_dev_extent
*dev_extent
= NULL
;
4350 struct btrfs_path
*path
;
4356 bool retried
= false;
4357 bool checked_pending_chunks
= false;
4358 struct extent_buffer
*l
;
4359 struct btrfs_key key
;
4360 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
4361 u64 old_total
= btrfs_super_total_bytes(super_copy
);
4362 u64 old_size
= btrfs_device_get_total_bytes(device
);
4365 new_size
= round_down(new_size
, fs_info
->sectorsize
);
4366 diff
= round_down(old_size
- new_size
, fs_info
->sectorsize
);
4368 if (device
->is_tgtdev_for_dev_replace
)
4371 path
= btrfs_alloc_path();
4375 path
->reada
= READA_FORWARD
;
4377 mutex_lock(&fs_info
->chunk_mutex
);
4379 btrfs_device_set_total_bytes(device
, new_size
);
4380 if (device
->writeable
) {
4381 device
->fs_devices
->total_rw_bytes
-= diff
;
4382 atomic64_sub(diff
, &fs_info
->free_chunk_space
);
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
,
4494 round_down(old_total
- diff
, fs_info
->sectorsize
));
4495 mutex_unlock(&fs_info
->chunk_mutex
);
4497 /* Now btrfs_update_device() will change the on-disk size. */
4498 ret
= btrfs_update_device(trans
, device
);
4499 btrfs_end_transaction(trans
);
4501 btrfs_free_path(path
);
4503 mutex_lock(&fs_info
->chunk_mutex
);
4504 btrfs_device_set_total_bytes(device
, old_size
);
4505 if (device
->writeable
)
4506 device
->fs_devices
->total_rw_bytes
+= diff
;
4507 atomic64_add(diff
, &fs_info
->free_chunk_space
);
4508 mutex_unlock(&fs_info
->chunk_mutex
);
4513 static int btrfs_add_system_chunk(struct btrfs_fs_info
*fs_info
,
4514 struct btrfs_key
*key
,
4515 struct btrfs_chunk
*chunk
, int item_size
)
4517 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
4518 struct btrfs_disk_key disk_key
;
4522 mutex_lock(&fs_info
->chunk_mutex
);
4523 array_size
= btrfs_super_sys_array_size(super_copy
);
4524 if (array_size
+ item_size
+ sizeof(disk_key
)
4525 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4526 mutex_unlock(&fs_info
->chunk_mutex
);
4530 ptr
= super_copy
->sys_chunk_array
+ array_size
;
4531 btrfs_cpu_key_to_disk(&disk_key
, key
);
4532 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
4533 ptr
+= sizeof(disk_key
);
4534 memcpy(ptr
, chunk
, item_size
);
4535 item_size
+= sizeof(disk_key
);
4536 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
4537 mutex_unlock(&fs_info
->chunk_mutex
);
4543 * sort the devices in descending order by max_avail, total_avail
4545 static int btrfs_cmp_device_info(const void *a
, const void *b
)
4547 const struct btrfs_device_info
*di_a
= a
;
4548 const struct btrfs_device_info
*di_b
= b
;
4550 if (di_a
->max_avail
> di_b
->max_avail
)
4552 if (di_a
->max_avail
< di_b
->max_avail
)
4554 if (di_a
->total_avail
> di_b
->total_avail
)
4556 if (di_a
->total_avail
< di_b
->total_avail
)
4561 static void check_raid56_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
4563 if (!(type
& BTRFS_BLOCK_GROUP_RAID56_MASK
))
4566 btrfs_set_fs_incompat(info
, RAID56
);
4569 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info) \
4570 - sizeof(struct btrfs_chunk)) \
4571 / sizeof(struct btrfs_stripe) + 1)
4573 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4574 - 2 * sizeof(struct btrfs_disk_key) \
4575 - 2 * sizeof(struct btrfs_chunk)) \
4576 / sizeof(struct btrfs_stripe) + 1)
4578 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4579 u64 start
, u64 type
)
4581 struct btrfs_fs_info
*info
= trans
->fs_info
;
4582 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
4583 struct btrfs_device
*device
;
4584 struct map_lookup
*map
= NULL
;
4585 struct extent_map_tree
*em_tree
;
4586 struct extent_map
*em
;
4587 struct btrfs_device_info
*devices_info
= NULL
;
4589 int num_stripes
; /* total number of stripes to allocate */
4590 int data_stripes
; /* number of stripes that count for
4592 int sub_stripes
; /* sub_stripes info for map */
4593 int dev_stripes
; /* stripes per dev */
4594 int devs_max
; /* max devs to use */
4595 int devs_min
; /* min devs needed */
4596 int devs_increment
; /* ndevs has to be a multiple of this */
4597 int ncopies
; /* how many copies to data has */
4599 u64 max_stripe_size
;
4608 BUG_ON(!alloc_profile_is_valid(type
, 0));
4610 if (list_empty(&fs_devices
->alloc_list
))
4613 index
= __get_raid_index(type
);
4615 sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
4616 dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
4617 devs_max
= btrfs_raid_array
[index
].devs_max
;
4618 devs_min
= btrfs_raid_array
[index
].devs_min
;
4619 devs_increment
= btrfs_raid_array
[index
].devs_increment
;
4620 ncopies
= btrfs_raid_array
[index
].ncopies
;
4622 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
4623 max_stripe_size
= SZ_1G
;
4624 max_chunk_size
= 10 * max_stripe_size
;
4626 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4627 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
4628 /* for larger filesystems, use larger metadata chunks */
4629 if (fs_devices
->total_rw_bytes
> 50ULL * SZ_1G
)
4630 max_stripe_size
= SZ_1G
;
4632 max_stripe_size
= SZ_256M
;
4633 max_chunk_size
= max_stripe_size
;
4635 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4636 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4637 max_stripe_size
= SZ_32M
;
4638 max_chunk_size
= 2 * max_stripe_size
;
4640 devs_max
= BTRFS_MAX_DEVS_SYS_CHUNK
;
4642 btrfs_err(info
, "invalid chunk type 0x%llx requested",
4647 /* we don't want a chunk larger than 10% of writeable space */
4648 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
4651 devices_info
= kcalloc(fs_devices
->rw_devices
, sizeof(*devices_info
),
4657 * in the first pass through the devices list, we gather information
4658 * about the available holes on each device.
4661 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
4665 if (!device
->writeable
) {
4667 "BTRFS: read-only device in alloc_list\n");
4671 if (!device
->in_fs_metadata
||
4672 device
->is_tgtdev_for_dev_replace
)
4675 if (device
->total_bytes
> device
->bytes_used
)
4676 total_avail
= device
->total_bytes
- device
->bytes_used
;
4680 /* If there is no space on this device, skip it. */
4681 if (total_avail
== 0)
4684 ret
= find_free_dev_extent(trans
, device
,
4685 max_stripe_size
* dev_stripes
,
4686 &dev_offset
, &max_avail
);
4687 if (ret
&& ret
!= -ENOSPC
)
4691 max_avail
= max_stripe_size
* dev_stripes
;
4693 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
4696 if (ndevs
== fs_devices
->rw_devices
) {
4697 WARN(1, "%s: found more than %llu devices\n",
4698 __func__
, fs_devices
->rw_devices
);
4701 devices_info
[ndevs
].dev_offset
= dev_offset
;
4702 devices_info
[ndevs
].max_avail
= max_avail
;
4703 devices_info
[ndevs
].total_avail
= total_avail
;
4704 devices_info
[ndevs
].dev
= device
;
4709 * now sort the devices by hole size / available space
4711 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
4712 btrfs_cmp_device_info
, NULL
);
4714 /* round down to number of usable stripes */
4715 ndevs
= round_down(ndevs
, devs_increment
);
4717 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
4722 ndevs
= min(ndevs
, devs_max
);
4725 * the primary goal is to maximize the number of stripes, so use as many
4726 * devices as possible, even if the stripes are not maximum sized.
4728 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4729 num_stripes
= ndevs
* dev_stripes
;
4732 * this will have to be fixed for RAID1 and RAID10 over
4735 data_stripes
= num_stripes
/ ncopies
;
4737 if (type
& BTRFS_BLOCK_GROUP_RAID5
)
4738 data_stripes
= num_stripes
- 1;
4740 if (type
& BTRFS_BLOCK_GROUP_RAID6
)
4741 data_stripes
= num_stripes
- 2;
4744 * Use the number of data stripes to figure out how big this chunk
4745 * is really going to be in terms of logical address space,
4746 * and compare that answer with the max chunk size
4748 if (stripe_size
* data_stripes
> max_chunk_size
) {
4749 u64 mask
= (1ULL << 24) - 1;
4751 stripe_size
= div_u64(max_chunk_size
, data_stripes
);
4753 /* bump the answer up to a 16MB boundary */
4754 stripe_size
= (stripe_size
+ mask
) & ~mask
;
4756 /* but don't go higher than the limits we found
4757 * while searching for free extents
4759 if (stripe_size
> devices_info
[ndevs
-1].max_avail
)
4760 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4763 stripe_size
= div_u64(stripe_size
, dev_stripes
);
4765 /* align to BTRFS_STRIPE_LEN */
4766 stripe_size
= round_down(stripe_size
, BTRFS_STRIPE_LEN
);
4768 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
4773 map
->num_stripes
= num_stripes
;
4775 for (i
= 0; i
< ndevs
; ++i
) {
4776 for (j
= 0; j
< dev_stripes
; ++j
) {
4777 int s
= i
* dev_stripes
+ j
;
4778 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
4779 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
4783 map
->stripe_len
= BTRFS_STRIPE_LEN
;
4784 map
->io_align
= BTRFS_STRIPE_LEN
;
4785 map
->io_width
= BTRFS_STRIPE_LEN
;
4787 map
->sub_stripes
= sub_stripes
;
4789 num_bytes
= stripe_size
* data_stripes
;
4791 trace_btrfs_chunk_alloc(info
, map
, start
, num_bytes
);
4793 em
= alloc_extent_map();
4799 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
4800 em
->map_lookup
= map
;
4802 em
->len
= num_bytes
;
4803 em
->block_start
= 0;
4804 em
->block_len
= em
->len
;
4805 em
->orig_block_len
= stripe_size
;
4807 em_tree
= &info
->mapping_tree
.map_tree
;
4808 write_lock(&em_tree
->lock
);
4809 ret
= add_extent_mapping(em_tree
, em
, 0);
4811 list_add_tail(&em
->list
, &trans
->transaction
->pending_chunks
);
4812 refcount_inc(&em
->refs
);
4814 write_unlock(&em_tree
->lock
);
4816 free_extent_map(em
);
4820 ret
= btrfs_make_block_group(trans
, info
, 0, type
, start
, num_bytes
);
4822 goto error_del_extent
;
4824 for (i
= 0; i
< map
->num_stripes
; i
++) {
4825 num_bytes
= map
->stripes
[i
].dev
->bytes_used
+ stripe_size
;
4826 btrfs_device_set_bytes_used(map
->stripes
[i
].dev
, num_bytes
);
4829 atomic64_sub(stripe_size
* map
->num_stripes
, &info
->free_chunk_space
);
4831 free_extent_map(em
);
4832 check_raid56_incompat_flag(info
, type
);
4834 kfree(devices_info
);
4838 write_lock(&em_tree
->lock
);
4839 remove_extent_mapping(em_tree
, em
);
4840 write_unlock(&em_tree
->lock
);
4842 /* One for our allocation */
4843 free_extent_map(em
);
4844 /* One for the tree reference */
4845 free_extent_map(em
);
4846 /* One for the pending_chunks list reference */
4847 free_extent_map(em
);
4849 kfree(devices_info
);
4853 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
4854 struct btrfs_fs_info
*fs_info
,
4855 u64 chunk_offset
, u64 chunk_size
)
4857 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
4858 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
4859 struct btrfs_key key
;
4860 struct btrfs_device
*device
;
4861 struct btrfs_chunk
*chunk
;
4862 struct btrfs_stripe
*stripe
;
4863 struct extent_map
*em
;
4864 struct map_lookup
*map
;
4871 em
= get_chunk_map(fs_info
, chunk_offset
, chunk_size
);
4875 map
= em
->map_lookup
;
4876 item_size
= btrfs_chunk_item_size(map
->num_stripes
);
4877 stripe_size
= em
->orig_block_len
;
4879 chunk
= kzalloc(item_size
, GFP_NOFS
);
4886 * Take the device list mutex to prevent races with the final phase of
4887 * a device replace operation that replaces the device object associated
4888 * with the map's stripes, because the device object's id can change
4889 * at any time during that final phase of the device replace operation
4890 * (dev-replace.c:btrfs_dev_replace_finishing()).
4892 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
4893 for (i
= 0; i
< map
->num_stripes
; i
++) {
4894 device
= map
->stripes
[i
].dev
;
4895 dev_offset
= map
->stripes
[i
].physical
;
4897 ret
= btrfs_update_device(trans
, device
);
4900 ret
= btrfs_alloc_dev_extent(trans
, device
, chunk_offset
,
4901 dev_offset
, stripe_size
);
4906 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
4910 stripe
= &chunk
->stripe
;
4911 for (i
= 0; i
< map
->num_stripes
; i
++) {
4912 device
= map
->stripes
[i
].dev
;
4913 dev_offset
= map
->stripes
[i
].physical
;
4915 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
4916 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
4917 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
4920 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
4922 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
4923 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
4924 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
4925 btrfs_set_stack_chunk_type(chunk
, map
->type
);
4926 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
4927 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
4928 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
4929 btrfs_set_stack_chunk_sector_size(chunk
, fs_info
->sectorsize
);
4930 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
4932 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
4933 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
4934 key
.offset
= chunk_offset
;
4936 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
4937 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4939 * TODO: Cleanup of inserted chunk root in case of
4942 ret
= btrfs_add_system_chunk(fs_info
, &key
, chunk
, item_size
);
4947 free_extent_map(em
);
4952 * Chunk allocation falls into two parts. The first part does works
4953 * that make the new allocated chunk useable, but not do any operation
4954 * that modifies the chunk tree. The second part does the works that
4955 * require modifying the chunk tree. This division is important for the
4956 * bootstrap process of adding storage to a seed btrfs.
4958 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4959 struct btrfs_fs_info
*fs_info
, u64 type
)
4963 ASSERT(mutex_is_locked(&fs_info
->chunk_mutex
));
4964 chunk_offset
= find_next_chunk(fs_info
);
4965 return __btrfs_alloc_chunk(trans
, chunk_offset
, type
);
4968 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
4969 struct btrfs_fs_info
*fs_info
)
4972 u64 sys_chunk_offset
;
4976 chunk_offset
= find_next_chunk(fs_info
);
4977 alloc_profile
= btrfs_metadata_alloc_profile(fs_info
);
4978 ret
= __btrfs_alloc_chunk(trans
, chunk_offset
, alloc_profile
);
4982 sys_chunk_offset
= find_next_chunk(fs_info
);
4983 alloc_profile
= btrfs_system_alloc_profile(fs_info
);
4984 ret
= __btrfs_alloc_chunk(trans
, sys_chunk_offset
, alloc_profile
);
4988 static inline int btrfs_chunk_max_errors(struct map_lookup
*map
)
4992 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
4993 BTRFS_BLOCK_GROUP_RAID10
|
4994 BTRFS_BLOCK_GROUP_RAID5
|
4995 BTRFS_BLOCK_GROUP_DUP
)) {
4997 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
) {
5006 int btrfs_chunk_readonly(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
5008 struct extent_map
*em
;
5009 struct map_lookup
*map
;
5014 em
= get_chunk_map(fs_info
, chunk_offset
, 1);
5018 map
= em
->map_lookup
;
5019 for (i
= 0; i
< map
->num_stripes
; i
++) {
5020 if (map
->stripes
[i
].dev
->missing
) {
5025 if (!map
->stripes
[i
].dev
->writeable
) {
5032 * If the number of missing devices is larger than max errors,
5033 * we can not write the data into that chunk successfully, so
5036 if (miss_ndevs
> btrfs_chunk_max_errors(map
))
5039 free_extent_map(em
);
5043 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
5045 extent_map_tree_init(&tree
->map_tree
);
5048 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
5050 struct extent_map
*em
;
5053 write_lock(&tree
->map_tree
.lock
);
5054 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
5056 remove_extent_mapping(&tree
->map_tree
, em
);
5057 write_unlock(&tree
->map_tree
.lock
);
5061 free_extent_map(em
);
5062 /* once for the tree */
5063 free_extent_map(em
);
5067 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
5069 struct extent_map
*em
;
5070 struct map_lookup
*map
;
5073 em
= get_chunk_map(fs_info
, logical
, len
);
5076 * We could return errors for these cases, but that could get
5077 * ugly and we'd probably do the same thing which is just not do
5078 * anything else and exit, so return 1 so the callers don't try
5079 * to use other copies.
5083 map
= em
->map_lookup
;
5084 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
5085 ret
= map
->num_stripes
;
5086 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5087 ret
= map
->sub_stripes
;
5088 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID5
)
5090 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5094 free_extent_map(em
);
5096 btrfs_dev_replace_lock(&fs_info
->dev_replace
, 0);
5097 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
) &&
5098 fs_info
->dev_replace
.tgtdev
)
5100 btrfs_dev_replace_unlock(&fs_info
->dev_replace
, 0);
5105 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info
*fs_info
,
5108 struct extent_map
*em
;
5109 struct map_lookup
*map
;
5110 unsigned long len
= fs_info
->sectorsize
;
5112 em
= get_chunk_map(fs_info
, logical
, len
);
5114 if (!WARN_ON(IS_ERR(em
))) {
5115 map
= em
->map_lookup
;
5116 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5117 len
= map
->stripe_len
* nr_data_stripes(map
);
5118 free_extent_map(em
);
5123 int btrfs_is_parity_mirror(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
5125 struct extent_map
*em
;
5126 struct map_lookup
*map
;
5129 em
= get_chunk_map(fs_info
, logical
, len
);
5131 if(!WARN_ON(IS_ERR(em
))) {
5132 map
= em
->map_lookup
;
5133 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5135 free_extent_map(em
);
5140 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
5141 struct map_lookup
*map
, int first
, int num
,
5142 int optimal
, int dev_replace_is_ongoing
)
5146 struct btrfs_device
*srcdev
;
5148 if (dev_replace_is_ongoing
&&
5149 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
5150 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
5151 srcdev
= fs_info
->dev_replace
.srcdev
;
5156 * try to avoid the drive that is the source drive for a
5157 * dev-replace procedure, only choose it if no other non-missing
5158 * mirror is available
5160 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
5161 if (map
->stripes
[optimal
].dev
->bdev
&&
5162 (tolerance
|| map
->stripes
[optimal
].dev
!= srcdev
))
5164 for (i
= first
; i
< first
+ num
; i
++) {
5165 if (map
->stripes
[i
].dev
->bdev
&&
5166 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
5171 /* we couldn't find one that doesn't fail. Just return something
5172 * and the io error handling code will clean up eventually
5177 static inline int parity_smaller(u64 a
, u64 b
)
5182 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5183 static void sort_parity_stripes(struct btrfs_bio
*bbio
, int num_stripes
)
5185 struct btrfs_bio_stripe s
;
5192 for (i
= 0; i
< num_stripes
- 1; i
++) {
5193 if (parity_smaller(bbio
->raid_map
[i
],
5194 bbio
->raid_map
[i
+1])) {
5195 s
= bbio
->stripes
[i
];
5196 l
= bbio
->raid_map
[i
];
5197 bbio
->stripes
[i
] = bbio
->stripes
[i
+1];
5198 bbio
->raid_map
[i
] = bbio
->raid_map
[i
+1];
5199 bbio
->stripes
[i
+1] = s
;
5200 bbio
->raid_map
[i
+1] = l
;
5208 static struct btrfs_bio
*alloc_btrfs_bio(int total_stripes
, int real_stripes
)
5210 struct btrfs_bio
*bbio
= kzalloc(
5211 /* the size of the btrfs_bio */
5212 sizeof(struct btrfs_bio
) +
5213 /* plus the variable array for the stripes */
5214 sizeof(struct btrfs_bio_stripe
) * (total_stripes
) +
5215 /* plus the variable array for the tgt dev */
5216 sizeof(int) * (real_stripes
) +
5218 * plus the raid_map, which includes both the tgt dev
5221 sizeof(u64
) * (total_stripes
),
5222 GFP_NOFS
|__GFP_NOFAIL
);
5224 atomic_set(&bbio
->error
, 0);
5225 refcount_set(&bbio
->refs
, 1);
5230 void btrfs_get_bbio(struct btrfs_bio
*bbio
)
5232 WARN_ON(!refcount_read(&bbio
->refs
));
5233 refcount_inc(&bbio
->refs
);
5236 void btrfs_put_bbio(struct btrfs_bio
*bbio
)
5240 if (refcount_dec_and_test(&bbio
->refs
))
5244 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5246 * Please note that, discard won't be sent to target device of device
5249 static int __btrfs_map_block_for_discard(struct btrfs_fs_info
*fs_info
,
5250 u64 logical
, u64 length
,
5251 struct btrfs_bio
**bbio_ret
)
5253 struct extent_map
*em
;
5254 struct map_lookup
*map
;
5255 struct btrfs_bio
*bbio
;
5259 u64 stripe_end_offset
;
5266 u32 sub_stripes
= 0;
5267 u64 stripes_per_dev
= 0;
5268 u32 remaining_stripes
= 0;
5269 u32 last_stripe
= 0;
5273 /* discard always return a bbio */
5276 em
= get_chunk_map(fs_info
, logical
, length
);
5280 map
= em
->map_lookup
;
5281 /* we don't discard raid56 yet */
5282 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5287 offset
= logical
- em
->start
;
5288 length
= min_t(u64
, em
->len
- offset
, length
);
5290 stripe_len
= map
->stripe_len
;
5292 * stripe_nr counts the total number of stripes we have to stride
5293 * to get to this block
5295 stripe_nr
= div64_u64(offset
, stripe_len
);
5297 /* stripe_offset is the offset of this block in its stripe */
5298 stripe_offset
= offset
- stripe_nr
* stripe_len
;
5300 stripe_nr_end
= round_up(offset
+ length
, map
->stripe_len
);
5301 stripe_nr_end
= div64_u64(stripe_nr_end
, map
->stripe_len
);
5302 stripe_cnt
= stripe_nr_end
- stripe_nr
;
5303 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
5306 * after this, stripe_nr is the number of stripes on this
5307 * device we have to walk to find the data, and stripe_index is
5308 * the number of our device in the stripe array
5312 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5313 BTRFS_BLOCK_GROUP_RAID10
)) {
5314 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5317 sub_stripes
= map
->sub_stripes
;
5319 factor
= map
->num_stripes
/ sub_stripes
;
5320 num_stripes
= min_t(u64
, map
->num_stripes
,
5321 sub_stripes
* stripe_cnt
);
5322 stripe_nr
= div_u64_rem(stripe_nr
, factor
, &stripe_index
);
5323 stripe_index
*= sub_stripes
;
5324 stripes_per_dev
= div_u64_rem(stripe_cnt
, factor
,
5325 &remaining_stripes
);
5326 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
5327 last_stripe
*= sub_stripes
;
5328 } else if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
5329 BTRFS_BLOCK_GROUP_DUP
)) {
5330 num_stripes
= map
->num_stripes
;
5332 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5336 bbio
= alloc_btrfs_bio(num_stripes
, 0);
5342 for (i
= 0; i
< num_stripes
; i
++) {
5343 bbio
->stripes
[i
].physical
=
5344 map
->stripes
[stripe_index
].physical
+
5345 stripe_offset
+ stripe_nr
* map
->stripe_len
;
5346 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
5348 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5349 BTRFS_BLOCK_GROUP_RAID10
)) {
5350 bbio
->stripes
[i
].length
= stripes_per_dev
*
5353 if (i
/ sub_stripes
< remaining_stripes
)
5354 bbio
->stripes
[i
].length
+=
5358 * Special for the first stripe and
5361 * |-------|...|-------|
5365 if (i
< sub_stripes
)
5366 bbio
->stripes
[i
].length
-=
5369 if (stripe_index
>= last_stripe
&&
5370 stripe_index
<= (last_stripe
+
5372 bbio
->stripes
[i
].length
-=
5375 if (i
== sub_stripes
- 1)
5378 bbio
->stripes
[i
].length
= length
;
5382 if (stripe_index
== map
->num_stripes
) {
5389 bbio
->map_type
= map
->type
;
5390 bbio
->num_stripes
= num_stripes
;
5392 free_extent_map(em
);
5397 * In dev-replace case, for repair case (that's the only case where the mirror
5398 * is selected explicitly when calling btrfs_map_block), blocks left of the
5399 * left cursor can also be read from the target drive.
5401 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5403 * For READ, it also needs to be supported using the same mirror number.
5405 * If the requested block is not left of the left cursor, EIO is returned. This
5406 * can happen because btrfs_num_copies() returns one more in the dev-replace
5409 static int get_extra_mirror_from_replace(struct btrfs_fs_info
*fs_info
,
5410 u64 logical
, u64 length
,
5411 u64 srcdev_devid
, int *mirror_num
,
5414 struct btrfs_bio
*bbio
= NULL
;
5416 int index_srcdev
= 0;
5418 u64 physical_of_found
= 0;
5422 ret
= __btrfs_map_block(fs_info
, BTRFS_MAP_GET_READ_MIRRORS
,
5423 logical
, &length
, &bbio
, 0, 0);
5425 ASSERT(bbio
== NULL
);
5429 num_stripes
= bbio
->num_stripes
;
5430 if (*mirror_num
> num_stripes
) {
5432 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5433 * that means that the requested area is not left of the left
5436 btrfs_put_bbio(bbio
);
5441 * process the rest of the function using the mirror_num of the source
5442 * drive. Therefore look it up first. At the end, patch the device
5443 * pointer to the one of the target drive.
5445 for (i
= 0; i
< num_stripes
; i
++) {
5446 if (bbio
->stripes
[i
].dev
->devid
!= srcdev_devid
)
5450 * In case of DUP, in order to keep it simple, only add the
5451 * mirror with the lowest physical address
5454 physical_of_found
<= bbio
->stripes
[i
].physical
)
5459 physical_of_found
= bbio
->stripes
[i
].physical
;
5462 btrfs_put_bbio(bbio
);
5468 *mirror_num
= index_srcdev
+ 1;
5469 *physical
= physical_of_found
;
5473 static void handle_ops_on_dev_replace(enum btrfs_map_op op
,
5474 struct btrfs_bio
**bbio_ret
,
5475 struct btrfs_dev_replace
*dev_replace
,
5476 int *num_stripes_ret
, int *max_errors_ret
)
5478 struct btrfs_bio
*bbio
= *bbio_ret
;
5479 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5480 int tgtdev_indexes
= 0;
5481 int num_stripes
= *num_stripes_ret
;
5482 int max_errors
= *max_errors_ret
;
5485 if (op
== BTRFS_MAP_WRITE
) {
5486 int index_where_to_add
;
5489 * duplicate the write operations while the dev replace
5490 * procedure is running. Since the copying of the old disk to
5491 * the new disk takes place at run time while the filesystem is
5492 * mounted writable, the regular write operations to the old
5493 * disk have to be duplicated to go to the new disk as well.
5495 * Note that device->missing is handled by the caller, and that
5496 * the write to the old disk is already set up in the stripes
5499 index_where_to_add
= num_stripes
;
5500 for (i
= 0; i
< num_stripes
; i
++) {
5501 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5502 /* write to new disk, too */
5503 struct btrfs_bio_stripe
*new =
5504 bbio
->stripes
+ index_where_to_add
;
5505 struct btrfs_bio_stripe
*old
=
5508 new->physical
= old
->physical
;
5509 new->length
= old
->length
;
5510 new->dev
= dev_replace
->tgtdev
;
5511 bbio
->tgtdev_map
[i
] = index_where_to_add
;
5512 index_where_to_add
++;
5517 num_stripes
= index_where_to_add
;
5518 } else if (op
== BTRFS_MAP_GET_READ_MIRRORS
) {
5519 int index_srcdev
= 0;
5521 u64 physical_of_found
= 0;
5524 * During the dev-replace procedure, the target drive can also
5525 * be used to read data in case it is needed to repair a corrupt
5526 * block elsewhere. This is possible if the requested area is
5527 * left of the left cursor. In this area, the target drive is a
5528 * full copy of the source drive.
5530 for (i
= 0; i
< num_stripes
; i
++) {
5531 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5533 * In case of DUP, in order to keep it simple,
5534 * only add the mirror with the lowest physical
5538 physical_of_found
<=
5539 bbio
->stripes
[i
].physical
)
5543 physical_of_found
= bbio
->stripes
[i
].physical
;
5547 struct btrfs_bio_stripe
*tgtdev_stripe
=
5548 bbio
->stripes
+ num_stripes
;
5550 tgtdev_stripe
->physical
= physical_of_found
;
5551 tgtdev_stripe
->length
=
5552 bbio
->stripes
[index_srcdev
].length
;
5553 tgtdev_stripe
->dev
= dev_replace
->tgtdev
;
5554 bbio
->tgtdev_map
[index_srcdev
] = num_stripes
;
5561 *num_stripes_ret
= num_stripes
;
5562 *max_errors_ret
= max_errors
;
5563 bbio
->num_tgtdevs
= tgtdev_indexes
;
5567 static bool need_full_stripe(enum btrfs_map_op op
)
5569 return (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
);
5572 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
,
5573 enum btrfs_map_op op
,
5574 u64 logical
, u64
*length
,
5575 struct btrfs_bio
**bbio_ret
,
5576 int mirror_num
, int need_raid_map
)
5578 struct extent_map
*em
;
5579 struct map_lookup
*map
;
5589 int tgtdev_indexes
= 0;
5590 struct btrfs_bio
*bbio
= NULL
;
5591 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
5592 int dev_replace_is_ongoing
= 0;
5593 int num_alloc_stripes
;
5594 int patch_the_first_stripe_for_dev_replace
= 0;
5595 u64 physical_to_patch_in_first_stripe
= 0;
5596 u64 raid56_full_stripe_start
= (u64
)-1;
5598 if (op
== BTRFS_MAP_DISCARD
)
5599 return __btrfs_map_block_for_discard(fs_info
, logical
,
5602 em
= get_chunk_map(fs_info
, logical
, *length
);
5606 map
= em
->map_lookup
;
5607 offset
= logical
- em
->start
;
5609 stripe_len
= map
->stripe_len
;
5612 * stripe_nr counts the total number of stripes we have to stride
5613 * to get to this block
5615 stripe_nr
= div64_u64(stripe_nr
, stripe_len
);
5617 stripe_offset
= stripe_nr
* stripe_len
;
5618 if (offset
< stripe_offset
) {
5620 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5621 stripe_offset
, offset
, em
->start
, logical
,
5623 free_extent_map(em
);
5627 /* stripe_offset is the offset of this block in its stripe*/
5628 stripe_offset
= offset
- stripe_offset
;
5630 /* if we're here for raid56, we need to know the stripe aligned start */
5631 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5632 unsigned long full_stripe_len
= stripe_len
* nr_data_stripes(map
);
5633 raid56_full_stripe_start
= offset
;
5635 /* allow a write of a full stripe, but make sure we don't
5636 * allow straddling of stripes
5638 raid56_full_stripe_start
= div64_u64(raid56_full_stripe_start
,
5640 raid56_full_stripe_start
*= full_stripe_len
;
5643 if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
5645 /* For writes to RAID[56], allow a full stripeset across all disks.
5646 For other RAID types and for RAID[56] reads, just allow a single
5647 stripe (on a single disk). */
5648 if ((map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
5649 (op
== BTRFS_MAP_WRITE
)) {
5650 max_len
= stripe_len
* nr_data_stripes(map
) -
5651 (offset
- raid56_full_stripe_start
);
5653 /* we limit the length of each bio to what fits in a stripe */
5654 max_len
= stripe_len
- stripe_offset
;
5656 *length
= min_t(u64
, em
->len
- offset
, max_len
);
5658 *length
= em
->len
- offset
;
5661 /* This is for when we're called from btrfs_merge_bio_hook() and all
5662 it cares about is the length */
5666 btrfs_dev_replace_lock(dev_replace
, 0);
5667 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
5668 if (!dev_replace_is_ongoing
)
5669 btrfs_dev_replace_unlock(dev_replace
, 0);
5671 btrfs_dev_replace_set_lock_blocking(dev_replace
);
5673 if (dev_replace_is_ongoing
&& mirror_num
== map
->num_stripes
+ 1 &&
5674 !need_full_stripe(op
) && dev_replace
->tgtdev
!= NULL
) {
5675 ret
= get_extra_mirror_from_replace(fs_info
, logical
, *length
,
5676 dev_replace
->srcdev
->devid
,
5678 &physical_to_patch_in_first_stripe
);
5682 patch_the_first_stripe_for_dev_replace
= 1;
5683 } else if (mirror_num
> map
->num_stripes
) {
5689 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5690 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5692 if (op
!= BTRFS_MAP_WRITE
&& op
!= BTRFS_MAP_GET_READ_MIRRORS
)
5694 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
5695 if (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
)
5696 num_stripes
= map
->num_stripes
;
5697 else if (mirror_num
)
5698 stripe_index
= mirror_num
- 1;
5700 stripe_index
= find_live_mirror(fs_info
, map
, 0,
5702 current
->pid
% map
->num_stripes
,
5703 dev_replace_is_ongoing
);
5704 mirror_num
= stripe_index
+ 1;
5707 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
5708 if (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
) {
5709 num_stripes
= map
->num_stripes
;
5710 } else if (mirror_num
) {
5711 stripe_index
= mirror_num
- 1;
5716 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5717 u32 factor
= map
->num_stripes
/ map
->sub_stripes
;
5719 stripe_nr
= div_u64_rem(stripe_nr
, factor
, &stripe_index
);
5720 stripe_index
*= map
->sub_stripes
;
5722 if (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
)
5723 num_stripes
= map
->sub_stripes
;
5724 else if (mirror_num
)
5725 stripe_index
+= mirror_num
- 1;
5727 int old_stripe_index
= stripe_index
;
5728 stripe_index
= find_live_mirror(fs_info
, map
,
5730 map
->sub_stripes
, stripe_index
+
5731 current
->pid
% map
->sub_stripes
,
5732 dev_replace_is_ongoing
);
5733 mirror_num
= stripe_index
- old_stripe_index
+ 1;
5736 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5737 if (need_raid_map
&&
5738 (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
||
5740 /* push stripe_nr back to the start of the full stripe */
5741 stripe_nr
= div64_u64(raid56_full_stripe_start
,
5742 stripe_len
* nr_data_stripes(map
));
5744 /* RAID[56] write or recovery. Return all stripes */
5745 num_stripes
= map
->num_stripes
;
5746 max_errors
= nr_parity_stripes(map
);
5748 *length
= map
->stripe_len
;
5753 * Mirror #0 or #1 means the original data block.
5754 * Mirror #2 is RAID5 parity block.
5755 * Mirror #3 is RAID6 Q block.
5757 stripe_nr
= div_u64_rem(stripe_nr
,
5758 nr_data_stripes(map
), &stripe_index
);
5760 stripe_index
= nr_data_stripes(map
) +
5763 /* We distribute the parity blocks across stripes */
5764 div_u64_rem(stripe_nr
+ stripe_index
, map
->num_stripes
,
5766 if ((op
!= BTRFS_MAP_WRITE
&&
5767 op
!= BTRFS_MAP_GET_READ_MIRRORS
) &&
5773 * after this, stripe_nr is the number of stripes on this
5774 * device we have to walk to find the data, and stripe_index is
5775 * the number of our device in the stripe array
5777 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5779 mirror_num
= stripe_index
+ 1;
5781 if (stripe_index
>= map
->num_stripes
) {
5783 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5784 stripe_index
, map
->num_stripes
);
5789 num_alloc_stripes
= num_stripes
;
5790 if (dev_replace_is_ongoing
&& dev_replace
->tgtdev
!= NULL
) {
5791 if (op
== BTRFS_MAP_WRITE
)
5792 num_alloc_stripes
<<= 1;
5793 if (op
== BTRFS_MAP_GET_READ_MIRRORS
)
5794 num_alloc_stripes
++;
5795 tgtdev_indexes
= num_stripes
;
5798 bbio
= alloc_btrfs_bio(num_alloc_stripes
, tgtdev_indexes
);
5803 if (dev_replace_is_ongoing
&& dev_replace
->tgtdev
!= NULL
)
5804 bbio
->tgtdev_map
= (int *)(bbio
->stripes
+ num_alloc_stripes
);
5806 /* build raid_map */
5807 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
&& need_raid_map
&&
5808 (need_full_stripe(op
) || mirror_num
> 1)) {
5812 bbio
->raid_map
= (u64
*)((void *)bbio
->stripes
+
5813 sizeof(struct btrfs_bio_stripe
) *
5815 sizeof(int) * tgtdev_indexes
);
5817 /* Work out the disk rotation on this stripe-set */
5818 div_u64_rem(stripe_nr
, num_stripes
, &rot
);
5820 /* Fill in the logical address of each stripe */
5821 tmp
= stripe_nr
* nr_data_stripes(map
);
5822 for (i
= 0; i
< nr_data_stripes(map
); i
++)
5823 bbio
->raid_map
[(i
+rot
) % num_stripes
] =
5824 em
->start
+ (tmp
+ i
) * map
->stripe_len
;
5826 bbio
->raid_map
[(i
+rot
) % map
->num_stripes
] = RAID5_P_STRIPE
;
5827 if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5828 bbio
->raid_map
[(i
+rot
+1) % num_stripes
] =
5833 for (i
= 0; i
< num_stripes
; i
++) {
5834 bbio
->stripes
[i
].physical
=
5835 map
->stripes
[stripe_index
].physical
+
5837 stripe_nr
* map
->stripe_len
;
5838 bbio
->stripes
[i
].dev
=
5839 map
->stripes
[stripe_index
].dev
;
5843 if (need_full_stripe(op
))
5844 max_errors
= btrfs_chunk_max_errors(map
);
5847 sort_parity_stripes(bbio
, num_stripes
);
5849 if (dev_replace_is_ongoing
&& dev_replace
->tgtdev
!= NULL
&&
5850 need_full_stripe(op
)) {
5851 handle_ops_on_dev_replace(op
, &bbio
, dev_replace
, &num_stripes
,
5856 bbio
->map_type
= map
->type
;
5857 bbio
->num_stripes
= num_stripes
;
5858 bbio
->max_errors
= max_errors
;
5859 bbio
->mirror_num
= mirror_num
;
5862 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5863 * mirror_num == num_stripes + 1 && dev_replace target drive is
5864 * available as a mirror
5866 if (patch_the_first_stripe_for_dev_replace
&& num_stripes
> 0) {
5867 WARN_ON(num_stripes
> 1);
5868 bbio
->stripes
[0].dev
= dev_replace
->tgtdev
;
5869 bbio
->stripes
[0].physical
= physical_to_patch_in_first_stripe
;
5870 bbio
->mirror_num
= map
->num_stripes
+ 1;
5873 if (dev_replace_is_ongoing
) {
5874 btrfs_dev_replace_clear_lock_blocking(dev_replace
);
5875 btrfs_dev_replace_unlock(dev_replace
, 0);
5877 free_extent_map(em
);
5881 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, enum btrfs_map_op op
,
5882 u64 logical
, u64
*length
,
5883 struct btrfs_bio
**bbio_ret
, int mirror_num
)
5885 return __btrfs_map_block(fs_info
, op
, logical
, length
, bbio_ret
,
5889 /* For Scrub/replace */
5890 int btrfs_map_sblock(struct btrfs_fs_info
*fs_info
, enum btrfs_map_op op
,
5891 u64 logical
, u64
*length
,
5892 struct btrfs_bio
**bbio_ret
)
5894 return __btrfs_map_block(fs_info
, op
, logical
, length
, bbio_ret
, 0, 1);
5897 int btrfs_rmap_block(struct btrfs_fs_info
*fs_info
,
5898 u64 chunk_start
, u64 physical
, u64 devid
,
5899 u64
**logical
, int *naddrs
, int *stripe_len
)
5901 struct extent_map
*em
;
5902 struct map_lookup
*map
;
5910 em
= get_chunk_map(fs_info
, chunk_start
, 1);
5914 map
= em
->map_lookup
;
5916 rmap_len
= map
->stripe_len
;
5918 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5919 length
= div_u64(length
, map
->num_stripes
/ map
->sub_stripes
);
5920 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5921 length
= div_u64(length
, map
->num_stripes
);
5922 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5923 length
= div_u64(length
, nr_data_stripes(map
));
5924 rmap_len
= map
->stripe_len
* nr_data_stripes(map
);
5927 buf
= kcalloc(map
->num_stripes
, sizeof(u64
), GFP_NOFS
);
5928 BUG_ON(!buf
); /* -ENOMEM */
5930 for (i
= 0; i
< map
->num_stripes
; i
++) {
5931 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
5933 if (map
->stripes
[i
].physical
> physical
||
5934 map
->stripes
[i
].physical
+ length
<= physical
)
5937 stripe_nr
= physical
- map
->stripes
[i
].physical
;
5938 stripe_nr
= div64_u64(stripe_nr
, map
->stripe_len
);
5940 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5941 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5942 stripe_nr
= div_u64(stripe_nr
, map
->sub_stripes
);
5943 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5944 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5945 } /* else if RAID[56], multiply by nr_data_stripes().
5946 * Alternatively, just use rmap_len below instead of
5947 * map->stripe_len */
5949 bytenr
= chunk_start
+ stripe_nr
* rmap_len
;
5950 WARN_ON(nr
>= map
->num_stripes
);
5951 for (j
= 0; j
< nr
; j
++) {
5952 if (buf
[j
] == bytenr
)
5956 WARN_ON(nr
>= map
->num_stripes
);
5963 *stripe_len
= rmap_len
;
5965 free_extent_map(em
);
5969 static inline void btrfs_end_bbio(struct btrfs_bio
*bbio
, struct bio
*bio
)
5971 bio
->bi_private
= bbio
->private;
5972 bio
->bi_end_io
= bbio
->end_io
;
5975 btrfs_put_bbio(bbio
);
5978 static void btrfs_end_bio(struct bio
*bio
)
5980 struct btrfs_bio
*bbio
= bio
->bi_private
;
5981 int is_orig_bio
= 0;
5983 if (bio
->bi_status
) {
5984 atomic_inc(&bbio
->error
);
5985 if (bio
->bi_status
== BLK_STS_IOERR
||
5986 bio
->bi_status
== BLK_STS_TARGET
) {
5987 unsigned int stripe_index
=
5988 btrfs_io_bio(bio
)->stripe_index
;
5989 struct btrfs_device
*dev
;
5991 BUG_ON(stripe_index
>= bbio
->num_stripes
);
5992 dev
= bbio
->stripes
[stripe_index
].dev
;
5994 if (bio_op(bio
) == REQ_OP_WRITE
)
5995 btrfs_dev_stat_inc(dev
,
5996 BTRFS_DEV_STAT_WRITE_ERRS
);
5998 btrfs_dev_stat_inc(dev
,
5999 BTRFS_DEV_STAT_READ_ERRS
);
6000 if (bio
->bi_opf
& REQ_PREFLUSH
)
6001 btrfs_dev_stat_inc(dev
,
6002 BTRFS_DEV_STAT_FLUSH_ERRS
);
6003 btrfs_dev_stat_print_on_error(dev
);
6008 if (bio
== bbio
->orig_bio
)
6011 btrfs_bio_counter_dec(bbio
->fs_info
);
6013 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
6016 bio
= bbio
->orig_bio
;
6019 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
6020 /* only send an error to the higher layers if it is
6021 * beyond the tolerance of the btrfs bio
6023 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
6024 bio
->bi_status
= BLK_STS_IOERR
;
6027 * this bio is actually up to date, we didn't
6028 * go over the max number of errors
6033 btrfs_end_bbio(bbio
, bio
);
6034 } else if (!is_orig_bio
) {
6040 * see run_scheduled_bios for a description of why bios are collected for
6043 * This will add one bio to the pending list for a device and make sure
6044 * the work struct is scheduled.
6046 static noinline
void btrfs_schedule_bio(struct btrfs_device
*device
,
6049 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
6050 int should_queue
= 1;
6051 struct btrfs_pending_bios
*pending_bios
;
6053 if (device
->missing
|| !device
->bdev
) {
6058 /* don't bother with additional async steps for reads, right now */
6059 if (bio_op(bio
) == REQ_OP_READ
) {
6061 btrfsic_submit_bio(bio
);
6067 * nr_async_bios allows us to reliably return congestion to the
6068 * higher layers. Otherwise, the async bio makes it appear we have
6069 * made progress against dirty pages when we've really just put it
6070 * on a queue for later
6072 atomic_inc(&fs_info
->nr_async_bios
);
6073 WARN_ON(bio
->bi_next
);
6074 bio
->bi_next
= NULL
;
6076 spin_lock(&device
->io_lock
);
6077 if (op_is_sync(bio
->bi_opf
))
6078 pending_bios
= &device
->pending_sync_bios
;
6080 pending_bios
= &device
->pending_bios
;
6082 if (pending_bios
->tail
)
6083 pending_bios
->tail
->bi_next
= bio
;
6085 pending_bios
->tail
= bio
;
6086 if (!pending_bios
->head
)
6087 pending_bios
->head
= bio
;
6088 if (device
->running_pending
)
6091 spin_unlock(&device
->io_lock
);
6094 btrfs_queue_work(fs_info
->submit_workers
, &device
->work
);
6097 static void submit_stripe_bio(struct btrfs_bio
*bbio
, struct bio
*bio
,
6098 u64 physical
, int dev_nr
, int async
)
6100 struct btrfs_device
*dev
= bbio
->stripes
[dev_nr
].dev
;
6101 struct btrfs_fs_info
*fs_info
= bbio
->fs_info
;
6103 bio
->bi_private
= bbio
;
6104 btrfs_io_bio(bio
)->stripe_index
= dev_nr
;
6105 bio
->bi_end_io
= btrfs_end_bio
;
6106 bio
->bi_iter
.bi_sector
= physical
>> 9;
6109 struct rcu_string
*name
;
6112 name
= rcu_dereference(dev
->name
);
6113 btrfs_debug(fs_info
,
6114 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6115 bio_op(bio
), bio
->bi_opf
,
6116 (u64
)bio
->bi_iter
.bi_sector
,
6117 (u_long
)dev
->bdev
->bd_dev
, name
->str
, dev
->devid
,
6118 bio
->bi_iter
.bi_size
);
6122 bio_set_dev(bio
, dev
->bdev
);
6124 btrfs_bio_counter_inc_noblocked(fs_info
);
6127 btrfs_schedule_bio(dev
, bio
);
6129 btrfsic_submit_bio(bio
);
6132 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
6134 atomic_inc(&bbio
->error
);
6135 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
6136 /* Should be the original bio. */
6137 WARN_ON(bio
!= bbio
->orig_bio
);
6139 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
6140 bio
->bi_iter
.bi_sector
= logical
>> 9;
6141 bio
->bi_status
= BLK_STS_IOERR
;
6142 btrfs_end_bbio(bbio
, bio
);
6146 blk_status_t
btrfs_map_bio(struct btrfs_fs_info
*fs_info
, struct bio
*bio
,
6147 int mirror_num
, int async_submit
)
6149 struct btrfs_device
*dev
;
6150 struct bio
*first_bio
= bio
;
6151 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
6157 struct btrfs_bio
*bbio
= NULL
;
6159 length
= bio
->bi_iter
.bi_size
;
6160 map_length
= length
;
6162 btrfs_bio_counter_inc_blocked(fs_info
);
6163 ret
= __btrfs_map_block(fs_info
, btrfs_op(bio
), logical
,
6164 &map_length
, &bbio
, mirror_num
, 1);
6166 btrfs_bio_counter_dec(fs_info
);
6167 return errno_to_blk_status(ret
);
6170 total_devs
= bbio
->num_stripes
;
6171 bbio
->orig_bio
= first_bio
;
6172 bbio
->private = first_bio
->bi_private
;
6173 bbio
->end_io
= first_bio
->bi_end_io
;
6174 bbio
->fs_info
= fs_info
;
6175 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
6177 if ((bbio
->map_type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
6178 ((bio_op(bio
) == REQ_OP_WRITE
) || (mirror_num
> 1))) {
6179 /* In this case, map_length has been set to the length of
6180 a single stripe; not the whole write */
6181 if (bio_op(bio
) == REQ_OP_WRITE
) {
6182 ret
= raid56_parity_write(fs_info
, bio
, bbio
,
6185 ret
= raid56_parity_recover(fs_info
, bio
, bbio
,
6186 map_length
, mirror_num
, 1);
6189 btrfs_bio_counter_dec(fs_info
);
6190 return errno_to_blk_status(ret
);
6193 if (map_length
< length
) {
6195 "mapping failed logical %llu bio len %llu len %llu",
6196 logical
, length
, map_length
);
6200 for (dev_nr
= 0; dev_nr
< total_devs
; dev_nr
++) {
6201 dev
= bbio
->stripes
[dev_nr
].dev
;
6202 if (!dev
|| !dev
->bdev
||
6203 (bio_op(first_bio
) == REQ_OP_WRITE
&& !dev
->writeable
)) {
6204 bbio_error(bbio
, first_bio
, logical
);
6208 if (dev_nr
< total_devs
- 1)
6209 bio
= btrfs_bio_clone(first_bio
);
6213 submit_stripe_bio(bbio
, bio
, bbio
->stripes
[dev_nr
].physical
,
6214 dev_nr
, async_submit
);
6216 btrfs_bio_counter_dec(fs_info
);
6220 struct btrfs_device
*btrfs_find_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
6223 struct btrfs_device
*device
;
6224 struct btrfs_fs_devices
*cur_devices
;
6226 cur_devices
= fs_info
->fs_devices
;
6227 while (cur_devices
) {
6229 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_FSID_SIZE
)) {
6230 device
= find_device(cur_devices
, devid
, uuid
);
6234 cur_devices
= cur_devices
->seed
;
6239 static struct btrfs_device
*add_missing_dev(struct btrfs_fs_devices
*fs_devices
,
6240 u64 devid
, u8
*dev_uuid
)
6242 struct btrfs_device
*device
;
6244 device
= btrfs_alloc_device(NULL
, &devid
, dev_uuid
);
6248 list_add(&device
->dev_list
, &fs_devices
->devices
);
6249 device
->fs_devices
= fs_devices
;
6250 fs_devices
->num_devices
++;
6252 device
->missing
= 1;
6253 fs_devices
->missing_devices
++;
6259 * btrfs_alloc_device - allocate struct btrfs_device
6260 * @fs_info: used only for generating a new devid, can be NULL if
6261 * devid is provided (i.e. @devid != NULL).
6262 * @devid: a pointer to devid for this device. If NULL a new devid
6264 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6267 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6268 * on error. Returned struct is not linked onto any lists and can be
6269 * destroyed with kfree() right away.
6271 struct btrfs_device
*btrfs_alloc_device(struct btrfs_fs_info
*fs_info
,
6275 struct btrfs_device
*dev
;
6278 if (WARN_ON(!devid
&& !fs_info
))
6279 return ERR_PTR(-EINVAL
);
6281 dev
= __alloc_device();
6290 ret
= find_next_devid(fs_info
, &tmp
);
6293 return ERR_PTR(ret
);
6299 memcpy(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
);
6301 generate_random_uuid(dev
->uuid
);
6303 btrfs_init_work(&dev
->work
, btrfs_submit_helper
,
6304 pending_bios_fn
, NULL
, NULL
);
6309 /* Return -EIO if any error, otherwise return 0. */
6310 static int btrfs_check_chunk_valid(struct btrfs_fs_info
*fs_info
,
6311 struct extent_buffer
*leaf
,
6312 struct btrfs_chunk
*chunk
, u64 logical
)
6320 length
= btrfs_chunk_length(leaf
, chunk
);
6321 stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6322 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6323 sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6324 type
= btrfs_chunk_type(leaf
, chunk
);
6327 btrfs_err(fs_info
, "invalid chunk num_stripes: %u",
6331 if (!IS_ALIGNED(logical
, fs_info
->sectorsize
)) {
6332 btrfs_err(fs_info
, "invalid chunk logical %llu", logical
);
6335 if (btrfs_chunk_sector_size(leaf
, chunk
) != fs_info
->sectorsize
) {
6336 btrfs_err(fs_info
, "invalid chunk sectorsize %u",
6337 btrfs_chunk_sector_size(leaf
, chunk
));
6340 if (!length
|| !IS_ALIGNED(length
, fs_info
->sectorsize
)) {
6341 btrfs_err(fs_info
, "invalid chunk length %llu", length
);
6344 if (!is_power_of_2(stripe_len
) || stripe_len
!= BTRFS_STRIPE_LEN
) {
6345 btrfs_err(fs_info
, "invalid chunk stripe length: %llu",
6349 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK
| BTRFS_BLOCK_GROUP_PROFILE_MASK
) &
6351 btrfs_err(fs_info
, "unrecognized chunk type: %llu",
6352 ~(BTRFS_BLOCK_GROUP_TYPE_MASK
|
6353 BTRFS_BLOCK_GROUP_PROFILE_MASK
) &
6354 btrfs_chunk_type(leaf
, chunk
));
6357 if ((type
& BTRFS_BLOCK_GROUP_RAID10
&& sub_stripes
!= 2) ||
6358 (type
& BTRFS_BLOCK_GROUP_RAID1
&& num_stripes
< 1) ||
6359 (type
& BTRFS_BLOCK_GROUP_RAID5
&& num_stripes
< 2) ||
6360 (type
& BTRFS_BLOCK_GROUP_RAID6
&& num_stripes
< 3) ||
6361 (type
& BTRFS_BLOCK_GROUP_DUP
&& num_stripes
> 2) ||
6362 ((type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) == 0 &&
6363 num_stripes
!= 1)) {
6365 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6366 num_stripes
, sub_stripes
,
6367 type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
);
6374 static int read_one_chunk(struct btrfs_fs_info
*fs_info
, struct btrfs_key
*key
,
6375 struct extent_buffer
*leaf
,
6376 struct btrfs_chunk
*chunk
)
6378 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
6379 struct map_lookup
*map
;
6380 struct extent_map
*em
;
6384 u8 uuid
[BTRFS_UUID_SIZE
];
6389 logical
= key
->offset
;
6390 length
= btrfs_chunk_length(leaf
, chunk
);
6391 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6393 ret
= btrfs_check_chunk_valid(fs_info
, leaf
, chunk
, logical
);
6397 read_lock(&map_tree
->map_tree
.lock
);
6398 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
6399 read_unlock(&map_tree
->map_tree
.lock
);
6401 /* already mapped? */
6402 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
6403 free_extent_map(em
);
6406 free_extent_map(em
);
6409 em
= alloc_extent_map();
6412 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
6414 free_extent_map(em
);
6418 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
6419 em
->map_lookup
= map
;
6420 em
->start
= logical
;
6423 em
->block_start
= 0;
6424 em
->block_len
= em
->len
;
6426 map
->num_stripes
= num_stripes
;
6427 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
6428 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
6429 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6430 map
->type
= btrfs_chunk_type(leaf
, chunk
);
6431 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6432 for (i
= 0; i
< num_stripes
; i
++) {
6433 map
->stripes
[i
].physical
=
6434 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
6435 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
6436 read_extent_buffer(leaf
, uuid
, (unsigned long)
6437 btrfs_stripe_dev_uuid_nr(chunk
, i
),
6439 map
->stripes
[i
].dev
= btrfs_find_device(fs_info
, devid
,
6441 if (!map
->stripes
[i
].dev
&&
6442 !btrfs_test_opt(fs_info
, DEGRADED
)) {
6443 free_extent_map(em
);
6444 btrfs_report_missing_device(fs_info
, devid
, uuid
);
6447 if (!map
->stripes
[i
].dev
) {
6448 map
->stripes
[i
].dev
=
6449 add_missing_dev(fs_info
->fs_devices
, devid
,
6451 if (!map
->stripes
[i
].dev
) {
6452 free_extent_map(em
);
6455 btrfs_report_missing_device(fs_info
, devid
, uuid
);
6457 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
6460 write_lock(&map_tree
->map_tree
.lock
);
6461 ret
= add_extent_mapping(&map_tree
->map_tree
, em
, 0);
6462 write_unlock(&map_tree
->map_tree
.lock
);
6463 BUG_ON(ret
); /* Tree corruption */
6464 free_extent_map(em
);
6469 static void fill_device_from_item(struct extent_buffer
*leaf
,
6470 struct btrfs_dev_item
*dev_item
,
6471 struct btrfs_device
*device
)
6475 device
->devid
= btrfs_device_id(leaf
, dev_item
);
6476 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
6477 device
->total_bytes
= device
->disk_total_bytes
;
6478 device
->commit_total_bytes
= device
->disk_total_bytes
;
6479 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
6480 device
->commit_bytes_used
= device
->bytes_used
;
6481 device
->type
= btrfs_device_type(leaf
, dev_item
);
6482 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
6483 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
6484 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
6485 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
6486 device
->is_tgtdev_for_dev_replace
= 0;
6488 ptr
= btrfs_device_uuid(dev_item
);
6489 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
6492 static struct btrfs_fs_devices
*open_seed_devices(struct btrfs_fs_info
*fs_info
,
6495 struct btrfs_fs_devices
*fs_devices
;
6498 BUG_ON(!mutex_is_locked(&uuid_mutex
));
6501 fs_devices
= fs_info
->fs_devices
->seed
;
6502 while (fs_devices
) {
6503 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_FSID_SIZE
))
6506 fs_devices
= fs_devices
->seed
;
6509 fs_devices
= find_fsid(fsid
);
6511 if (!btrfs_test_opt(fs_info
, DEGRADED
))
6512 return ERR_PTR(-ENOENT
);
6514 fs_devices
= alloc_fs_devices(fsid
);
6515 if (IS_ERR(fs_devices
))
6518 fs_devices
->seeding
= 1;
6519 fs_devices
->opened
= 1;
6523 fs_devices
= clone_fs_devices(fs_devices
);
6524 if (IS_ERR(fs_devices
))
6527 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
6528 fs_info
->bdev_holder
);
6530 free_fs_devices(fs_devices
);
6531 fs_devices
= ERR_PTR(ret
);
6535 if (!fs_devices
->seeding
) {
6536 __btrfs_close_devices(fs_devices
);
6537 free_fs_devices(fs_devices
);
6538 fs_devices
= ERR_PTR(-EINVAL
);
6542 fs_devices
->seed
= fs_info
->fs_devices
->seed
;
6543 fs_info
->fs_devices
->seed
= fs_devices
;
6548 static int read_one_dev(struct btrfs_fs_info
*fs_info
,
6549 struct extent_buffer
*leaf
,
6550 struct btrfs_dev_item
*dev_item
)
6552 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6553 struct btrfs_device
*device
;
6556 u8 fs_uuid
[BTRFS_FSID_SIZE
];
6557 u8 dev_uuid
[BTRFS_UUID_SIZE
];
6559 devid
= btrfs_device_id(leaf
, dev_item
);
6560 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
6562 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
6565 if (memcmp(fs_uuid
, fs_info
->fsid
, BTRFS_FSID_SIZE
)) {
6566 fs_devices
= open_seed_devices(fs_info
, fs_uuid
);
6567 if (IS_ERR(fs_devices
))
6568 return PTR_ERR(fs_devices
);
6571 device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, fs_uuid
);
6573 if (!btrfs_test_opt(fs_info
, DEGRADED
)) {
6574 btrfs_report_missing_device(fs_info
, devid
, dev_uuid
);
6578 device
= add_missing_dev(fs_devices
, devid
, dev_uuid
);
6581 btrfs_report_missing_device(fs_info
, devid
, dev_uuid
);
6583 if (!device
->bdev
) {
6584 btrfs_report_missing_device(fs_info
, devid
, dev_uuid
);
6585 if (!btrfs_test_opt(fs_info
, DEGRADED
))
6589 if(!device
->bdev
&& !device
->missing
) {
6591 * this happens when a device that was properly setup
6592 * in the device info lists suddenly goes bad.
6593 * device->bdev is NULL, and so we have to set
6594 * device->missing to one here
6596 device
->fs_devices
->missing_devices
++;
6597 device
->missing
= 1;
6600 /* Move the device to its own fs_devices */
6601 if (device
->fs_devices
!= fs_devices
) {
6602 ASSERT(device
->missing
);
6604 list_move(&device
->dev_list
, &fs_devices
->devices
);
6605 device
->fs_devices
->num_devices
--;
6606 fs_devices
->num_devices
++;
6608 device
->fs_devices
->missing_devices
--;
6609 fs_devices
->missing_devices
++;
6611 device
->fs_devices
= fs_devices
;
6615 if (device
->fs_devices
!= fs_info
->fs_devices
) {
6616 BUG_ON(device
->writeable
);
6617 if (device
->generation
!=
6618 btrfs_device_generation(leaf
, dev_item
))
6622 fill_device_from_item(leaf
, dev_item
, device
);
6623 device
->in_fs_metadata
= 1;
6624 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
6625 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
6626 atomic64_add(device
->total_bytes
- device
->bytes_used
,
6627 &fs_info
->free_chunk_space
);
6633 int btrfs_read_sys_array(struct btrfs_fs_info
*fs_info
)
6635 struct btrfs_root
*root
= fs_info
->tree_root
;
6636 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
6637 struct extent_buffer
*sb
;
6638 struct btrfs_disk_key
*disk_key
;
6639 struct btrfs_chunk
*chunk
;
6641 unsigned long sb_array_offset
;
6648 struct btrfs_key key
;
6650 ASSERT(BTRFS_SUPER_INFO_SIZE
<= fs_info
->nodesize
);
6652 * This will create extent buffer of nodesize, superblock size is
6653 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6654 * overallocate but we can keep it as-is, only the first page is used.
6656 sb
= btrfs_find_create_tree_block(fs_info
, BTRFS_SUPER_INFO_OFFSET
);
6659 set_extent_buffer_uptodate(sb
);
6660 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
6662 * The sb extent buffer is artificial and just used to read the system array.
6663 * set_extent_buffer_uptodate() call does not properly mark all it's
6664 * pages up-to-date when the page is larger: extent does not cover the
6665 * whole page and consequently check_page_uptodate does not find all
6666 * the page's extents up-to-date (the hole beyond sb),
6667 * write_extent_buffer then triggers a WARN_ON.
6669 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6670 * but sb spans only this function. Add an explicit SetPageUptodate call
6671 * to silence the warning eg. on PowerPC 64.
6673 if (PAGE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
6674 SetPageUptodate(sb
->pages
[0]);
6676 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
6677 array_size
= btrfs_super_sys_array_size(super_copy
);
6679 array_ptr
= super_copy
->sys_chunk_array
;
6680 sb_array_offset
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
6683 while (cur_offset
< array_size
) {
6684 disk_key
= (struct btrfs_disk_key
*)array_ptr
;
6685 len
= sizeof(*disk_key
);
6686 if (cur_offset
+ len
> array_size
)
6687 goto out_short_read
;
6689 btrfs_disk_key_to_cpu(&key
, disk_key
);
6692 sb_array_offset
+= len
;
6695 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6696 chunk
= (struct btrfs_chunk
*)sb_array_offset
;
6698 * At least one btrfs_chunk with one stripe must be
6699 * present, exact stripe count check comes afterwards
6701 len
= btrfs_chunk_item_size(1);
6702 if (cur_offset
+ len
> array_size
)
6703 goto out_short_read
;
6705 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
6708 "invalid number of stripes %u in sys_array at offset %u",
6709 num_stripes
, cur_offset
);
6714 type
= btrfs_chunk_type(sb
, chunk
);
6715 if ((type
& BTRFS_BLOCK_GROUP_SYSTEM
) == 0) {
6717 "invalid chunk type %llu in sys_array at offset %u",
6723 len
= btrfs_chunk_item_size(num_stripes
);
6724 if (cur_offset
+ len
> array_size
)
6725 goto out_short_read
;
6727 ret
= read_one_chunk(fs_info
, &key
, sb
, chunk
);
6732 "unexpected item type %u in sys_array at offset %u",
6733 (u32
)key
.type
, cur_offset
);
6738 sb_array_offset
+= len
;
6741 clear_extent_buffer_uptodate(sb
);
6742 free_extent_buffer_stale(sb
);
6746 btrfs_err(fs_info
, "sys_array too short to read %u bytes at offset %u",
6748 clear_extent_buffer_uptodate(sb
);
6749 free_extent_buffer_stale(sb
);
6753 void btrfs_report_missing_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
6756 btrfs_warn_rl(fs_info
, "devid %llu uuid %pU is missing", devid
, uuid
);
6760 * Check if all chunks in the fs are OK for read-write degraded mount
6762 * Return true if all chunks meet the minimal RW mount requirements.
6763 * Return false if any chunk doesn't meet the minimal RW mount requirements.
6765 bool btrfs_check_rw_degradable(struct btrfs_fs_info
*fs_info
)
6767 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
6768 struct extent_map
*em
;
6772 read_lock(&map_tree
->map_tree
.lock
);
6773 em
= lookup_extent_mapping(&map_tree
->map_tree
, 0, (u64
)-1);
6774 read_unlock(&map_tree
->map_tree
.lock
);
6775 /* No chunk at all? Return false anyway */
6781 struct map_lookup
*map
;
6786 map
= em
->map_lookup
;
6788 btrfs_get_num_tolerated_disk_barrier_failures(
6790 for (i
= 0; i
< map
->num_stripes
; i
++) {
6791 struct btrfs_device
*dev
= map
->stripes
[i
].dev
;
6793 if (!dev
|| !dev
->bdev
|| dev
->missing
||
6794 dev
->last_flush_error
)
6797 if (missing
> max_tolerated
) {
6799 "chunk %llu missing %d devices, max tolerance is %d for writeable mount",
6800 em
->start
, missing
, max_tolerated
);
6801 free_extent_map(em
);
6805 next_start
= extent_map_end(em
);
6806 free_extent_map(em
);
6808 read_lock(&map_tree
->map_tree
.lock
);
6809 em
= lookup_extent_mapping(&map_tree
->map_tree
, next_start
,
6810 (u64
)(-1) - next_start
);
6811 read_unlock(&map_tree
->map_tree
.lock
);
6817 int btrfs_read_chunk_tree(struct btrfs_fs_info
*fs_info
)
6819 struct btrfs_root
*root
= fs_info
->chunk_root
;
6820 struct btrfs_path
*path
;
6821 struct extent_buffer
*leaf
;
6822 struct btrfs_key key
;
6823 struct btrfs_key found_key
;
6828 path
= btrfs_alloc_path();
6832 mutex_lock(&uuid_mutex
);
6833 mutex_lock(&fs_info
->chunk_mutex
);
6836 * Read all device items, and then all the chunk items. All
6837 * device items are found before any chunk item (their object id
6838 * is smaller than the lowest possible object id for a chunk
6839 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6841 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
6844 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6848 leaf
= path
->nodes
[0];
6849 slot
= path
->slots
[0];
6850 if (slot
>= btrfs_header_nritems(leaf
)) {
6851 ret
= btrfs_next_leaf(root
, path
);
6858 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6859 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
6860 struct btrfs_dev_item
*dev_item
;
6861 dev_item
= btrfs_item_ptr(leaf
, slot
,
6862 struct btrfs_dev_item
);
6863 ret
= read_one_dev(fs_info
, leaf
, dev_item
);
6867 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6868 struct btrfs_chunk
*chunk
;
6869 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
6870 ret
= read_one_chunk(fs_info
, &found_key
, leaf
, chunk
);
6878 * After loading chunk tree, we've got all device information,
6879 * do another round of validation checks.
6881 if (total_dev
!= fs_info
->fs_devices
->total_devices
) {
6883 "super_num_devices %llu mismatch with num_devices %llu found here",
6884 btrfs_super_num_devices(fs_info
->super_copy
),
6889 if (btrfs_super_total_bytes(fs_info
->super_copy
) <
6890 fs_info
->fs_devices
->total_rw_bytes
) {
6892 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6893 btrfs_super_total_bytes(fs_info
->super_copy
),
6894 fs_info
->fs_devices
->total_rw_bytes
);
6900 mutex_unlock(&fs_info
->chunk_mutex
);
6901 mutex_unlock(&uuid_mutex
);
6903 btrfs_free_path(path
);
6907 void btrfs_init_devices_late(struct btrfs_fs_info
*fs_info
)
6909 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6910 struct btrfs_device
*device
;
6912 while (fs_devices
) {
6913 mutex_lock(&fs_devices
->device_list_mutex
);
6914 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
)
6915 device
->fs_info
= fs_info
;
6916 mutex_unlock(&fs_devices
->device_list_mutex
);
6918 fs_devices
= fs_devices
->seed
;
6922 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
6926 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6927 btrfs_dev_stat_reset(dev
, i
);
6930 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
6932 struct btrfs_key key
;
6933 struct btrfs_key found_key
;
6934 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6935 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6936 struct extent_buffer
*eb
;
6939 struct btrfs_device
*device
;
6940 struct btrfs_path
*path
= NULL
;
6943 path
= btrfs_alloc_path();
6949 mutex_lock(&fs_devices
->device_list_mutex
);
6950 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6952 struct btrfs_dev_stats_item
*ptr
;
6954 key
.objectid
= BTRFS_DEV_STATS_OBJECTID
;
6955 key
.type
= BTRFS_PERSISTENT_ITEM_KEY
;
6956 key
.offset
= device
->devid
;
6957 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
6959 __btrfs_reset_dev_stats(device
);
6960 device
->dev_stats_valid
= 1;
6961 btrfs_release_path(path
);
6964 slot
= path
->slots
[0];
6965 eb
= path
->nodes
[0];
6966 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
6967 item_size
= btrfs_item_size_nr(eb
, slot
);
6969 ptr
= btrfs_item_ptr(eb
, slot
,
6970 struct btrfs_dev_stats_item
);
6972 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6973 if (item_size
>= (1 + i
) * sizeof(__le64
))
6974 btrfs_dev_stat_set(device
, i
,
6975 btrfs_dev_stats_value(eb
, ptr
, i
));
6977 btrfs_dev_stat_reset(device
, i
);
6980 device
->dev_stats_valid
= 1;
6981 btrfs_dev_stat_print_on_load(device
);
6982 btrfs_release_path(path
);
6984 mutex_unlock(&fs_devices
->device_list_mutex
);
6987 btrfs_free_path(path
);
6988 return ret
< 0 ? ret
: 0;
6991 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
6992 struct btrfs_fs_info
*fs_info
,
6993 struct btrfs_device
*device
)
6995 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6996 struct btrfs_path
*path
;
6997 struct btrfs_key key
;
6998 struct extent_buffer
*eb
;
6999 struct btrfs_dev_stats_item
*ptr
;
7003 key
.objectid
= BTRFS_DEV_STATS_OBJECTID
;
7004 key
.type
= BTRFS_PERSISTENT_ITEM_KEY
;
7005 key
.offset
= device
->devid
;
7007 path
= btrfs_alloc_path();
7010 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
7012 btrfs_warn_in_rcu(fs_info
,
7013 "error %d while searching for dev_stats item for device %s",
7014 ret
, rcu_str_deref(device
->name
));
7019 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
7020 /* need to delete old one and insert a new one */
7021 ret
= btrfs_del_item(trans
, dev_root
, path
);
7023 btrfs_warn_in_rcu(fs_info
,
7024 "delete too small dev_stats item for device %s failed %d",
7025 rcu_str_deref(device
->name
), ret
);
7032 /* need to insert a new item */
7033 btrfs_release_path(path
);
7034 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
7035 &key
, sizeof(*ptr
));
7037 btrfs_warn_in_rcu(fs_info
,
7038 "insert dev_stats item for device %s failed %d",
7039 rcu_str_deref(device
->name
), ret
);
7044 eb
= path
->nodes
[0];
7045 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
7046 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7047 btrfs_set_dev_stats_value(eb
, ptr
, i
,
7048 btrfs_dev_stat_read(device
, i
));
7049 btrfs_mark_buffer_dirty(eb
);
7052 btrfs_free_path(path
);
7057 * called from commit_transaction. Writes all changed device stats to disk.
7059 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
7060 struct btrfs_fs_info
*fs_info
)
7062 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7063 struct btrfs_device
*device
;
7067 mutex_lock(&fs_devices
->device_list_mutex
);
7068 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
7069 if (!device
->dev_stats_valid
|| !btrfs_dev_stats_dirty(device
))
7072 stats_cnt
= atomic_read(&device
->dev_stats_ccnt
);
7073 ret
= update_dev_stat_item(trans
, fs_info
, device
);
7075 atomic_sub(stats_cnt
, &device
->dev_stats_ccnt
);
7077 mutex_unlock(&fs_devices
->device_list_mutex
);
7082 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
7084 btrfs_dev_stat_inc(dev
, index
);
7085 btrfs_dev_stat_print_on_error(dev
);
7088 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
7090 if (!dev
->dev_stats_valid
)
7092 btrfs_err_rl_in_rcu(dev
->fs_info
,
7093 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7094 rcu_str_deref(dev
->name
),
7095 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
7096 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
7097 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
7098 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
7099 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
7102 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
7106 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7107 if (btrfs_dev_stat_read(dev
, i
) != 0)
7109 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
7110 return; /* all values == 0, suppress message */
7112 btrfs_info_in_rcu(dev
->fs_info
,
7113 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7114 rcu_str_deref(dev
->name
),
7115 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
7116 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
7117 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
7118 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
7119 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
7122 int btrfs_get_dev_stats(struct btrfs_fs_info
*fs_info
,
7123 struct btrfs_ioctl_get_dev_stats
*stats
)
7125 struct btrfs_device
*dev
;
7126 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7129 mutex_lock(&fs_devices
->device_list_mutex
);
7130 dev
= btrfs_find_device(fs_info
, stats
->devid
, NULL
, NULL
);
7131 mutex_unlock(&fs_devices
->device_list_mutex
);
7134 btrfs_warn(fs_info
, "get dev_stats failed, device not found");
7136 } else if (!dev
->dev_stats_valid
) {
7137 btrfs_warn(fs_info
, "get dev_stats failed, not yet valid");
7139 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
7140 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
7141 if (stats
->nr_items
> i
)
7143 btrfs_dev_stat_read_and_reset(dev
, i
);
7145 btrfs_dev_stat_reset(dev
, i
);
7148 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7149 if (stats
->nr_items
> i
)
7150 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
7152 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
7153 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
7157 void btrfs_scratch_superblocks(struct block_device
*bdev
, const char *device_path
)
7159 struct buffer_head
*bh
;
7160 struct btrfs_super_block
*disk_super
;
7166 for (copy_num
= 0; copy_num
< BTRFS_SUPER_MIRROR_MAX
;
7169 if (btrfs_read_dev_one_super(bdev
, copy_num
, &bh
))
7172 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
7174 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
7175 set_buffer_dirty(bh
);
7176 sync_dirty_buffer(bh
);
7180 /* Notify udev that device has changed */
7181 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
7183 /* Update ctime/mtime for device path for libblkid */
7184 update_dev_time(device_path
);
7188 * Update the size of all devices, which is used for writing out the
7191 void btrfs_update_commit_device_size(struct btrfs_fs_info
*fs_info
)
7193 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7194 struct btrfs_device
*curr
, *next
;
7196 if (list_empty(&fs_devices
->resized_devices
))
7199 mutex_lock(&fs_devices
->device_list_mutex
);
7200 mutex_lock(&fs_info
->chunk_mutex
);
7201 list_for_each_entry_safe(curr
, next
, &fs_devices
->resized_devices
,
7203 list_del_init(&curr
->resized_list
);
7204 curr
->commit_total_bytes
= curr
->disk_total_bytes
;
7206 mutex_unlock(&fs_info
->chunk_mutex
);
7207 mutex_unlock(&fs_devices
->device_list_mutex
);
7210 /* Must be invoked during the transaction commit */
7211 void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info
*fs_info
,
7212 struct btrfs_transaction
*transaction
)
7214 struct extent_map
*em
;
7215 struct map_lookup
*map
;
7216 struct btrfs_device
*dev
;
7219 if (list_empty(&transaction
->pending_chunks
))
7222 /* In order to kick the device replace finish process */
7223 mutex_lock(&fs_info
->chunk_mutex
);
7224 list_for_each_entry(em
, &transaction
->pending_chunks
, list
) {
7225 map
= em
->map_lookup
;
7227 for (i
= 0; i
< map
->num_stripes
; i
++) {
7228 dev
= map
->stripes
[i
].dev
;
7229 dev
->commit_bytes_used
= dev
->bytes_used
;
7232 mutex_unlock(&fs_info
->chunk_mutex
);
7235 void btrfs_set_fs_info_ptr(struct btrfs_fs_info
*fs_info
)
7237 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7238 while (fs_devices
) {
7239 fs_devices
->fs_info
= fs_info
;
7240 fs_devices
= fs_devices
->seed
;
7244 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info
*fs_info
)
7246 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7247 while (fs_devices
) {
7248 fs_devices
->fs_info
= NULL
;
7249 fs_devices
= fs_devices
->seed
;