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)
155 static struct btrfs_fs_devices
*__alloc_fs_devices(void)
157 struct btrfs_fs_devices
*fs_devs
;
159 fs_devs
= kzalloc(sizeof(*fs_devs
), GFP_KERNEL
);
161 return ERR_PTR(-ENOMEM
);
163 mutex_init(&fs_devs
->device_list_mutex
);
165 INIT_LIST_HEAD(&fs_devs
->devices
);
166 INIT_LIST_HEAD(&fs_devs
->resized_devices
);
167 INIT_LIST_HEAD(&fs_devs
->alloc_list
);
168 INIT_LIST_HEAD(&fs_devs
->list
);
174 * alloc_fs_devices - allocate struct btrfs_fs_devices
175 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
178 * Return: a pointer to a new &struct btrfs_fs_devices on success;
179 * ERR_PTR() on error. Returned struct is not linked onto any lists and
180 * can be destroyed with kfree() right away.
182 static struct btrfs_fs_devices
*alloc_fs_devices(const u8
*fsid
)
184 struct btrfs_fs_devices
*fs_devs
;
186 fs_devs
= __alloc_fs_devices();
191 memcpy(fs_devs
->fsid
, fsid
, BTRFS_FSID_SIZE
);
193 generate_random_uuid(fs_devs
->fsid
);
198 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
200 struct btrfs_device
*device
;
201 WARN_ON(fs_devices
->opened
);
202 while (!list_empty(&fs_devices
->devices
)) {
203 device
= list_entry(fs_devices
->devices
.next
,
204 struct btrfs_device
, dev_list
);
205 list_del(&device
->dev_list
);
206 rcu_string_free(device
->name
);
212 static void btrfs_kobject_uevent(struct block_device
*bdev
,
213 enum kobject_action action
)
217 ret
= kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, action
);
219 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
221 kobject_name(&disk_to_dev(bdev
->bd_disk
)->kobj
),
222 &disk_to_dev(bdev
->bd_disk
)->kobj
);
225 void btrfs_cleanup_fs_uuids(void)
227 struct btrfs_fs_devices
*fs_devices
;
229 while (!list_empty(&fs_uuids
)) {
230 fs_devices
= list_entry(fs_uuids
.next
,
231 struct btrfs_fs_devices
, list
);
232 list_del(&fs_devices
->list
);
233 free_fs_devices(fs_devices
);
237 static struct btrfs_device
*__alloc_device(void)
239 struct btrfs_device
*dev
;
241 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
243 return ERR_PTR(-ENOMEM
);
246 * Preallocate a bio that's always going to be used for flushing device
247 * barriers and matches the device lifespan
249 dev
->flush_bio
= bio_alloc_bioset(GFP_KERNEL
, 0, NULL
);
250 if (!dev
->flush_bio
) {
252 return ERR_PTR(-ENOMEM
);
254 bio_get(dev
->flush_bio
);
256 INIT_LIST_HEAD(&dev
->dev_list
);
257 INIT_LIST_HEAD(&dev
->dev_alloc_list
);
258 INIT_LIST_HEAD(&dev
->resized_list
);
260 spin_lock_init(&dev
->io_lock
);
262 spin_lock_init(&dev
->reada_lock
);
263 atomic_set(&dev
->reada_in_flight
, 0);
264 atomic_set(&dev
->dev_stats_ccnt
, 0);
265 btrfs_device_data_ordered_init(dev
);
266 INIT_RADIX_TREE(&dev
->reada_zones
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
267 INIT_RADIX_TREE(&dev
->reada_extents
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
272 static noinline
struct btrfs_device
*__find_device(struct list_head
*head
,
275 struct btrfs_device
*dev
;
277 list_for_each_entry(dev
, head
, dev_list
) {
278 if (dev
->devid
== devid
&&
279 (!uuid
|| !memcmp(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
))) {
286 static noinline
struct btrfs_fs_devices
*find_fsid(u8
*fsid
)
288 struct btrfs_fs_devices
*fs_devices
;
290 list_for_each_entry(fs_devices
, &fs_uuids
, list
) {
291 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
298 btrfs_get_bdev_and_sb(const char *device_path
, fmode_t flags
, void *holder
,
299 int flush
, struct block_device
**bdev
,
300 struct buffer_head
**bh
)
304 *bdev
= blkdev_get_by_path(device_path
, flags
, holder
);
307 ret
= PTR_ERR(*bdev
);
312 filemap_write_and_wait((*bdev
)->bd_inode
->i_mapping
);
313 ret
= set_blocksize(*bdev
, 4096);
315 blkdev_put(*bdev
, flags
);
318 invalidate_bdev(*bdev
);
319 *bh
= btrfs_read_dev_super(*bdev
);
322 blkdev_put(*bdev
, flags
);
334 static void requeue_list(struct btrfs_pending_bios
*pending_bios
,
335 struct bio
*head
, struct bio
*tail
)
338 struct bio
*old_head
;
340 old_head
= pending_bios
->head
;
341 pending_bios
->head
= head
;
342 if (pending_bios
->tail
)
343 tail
->bi_next
= old_head
;
345 pending_bios
->tail
= tail
;
349 * we try to collect pending bios for a device so we don't get a large
350 * number of procs sending bios down to the same device. This greatly
351 * improves the schedulers ability to collect and merge the bios.
353 * But, it also turns into a long list of bios to process and that is sure
354 * to eventually make the worker thread block. The solution here is to
355 * make some progress and then put this work struct back at the end of
356 * the list if the block device is congested. This way, multiple devices
357 * can make progress from a single worker thread.
359 static noinline
void run_scheduled_bios(struct btrfs_device
*device
)
361 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
363 struct backing_dev_info
*bdi
;
364 struct btrfs_pending_bios
*pending_bios
;
368 unsigned long num_run
;
369 unsigned long batch_run
= 0;
371 unsigned long last_waited
= 0;
373 int sync_pending
= 0;
374 struct blk_plug plug
;
377 * this function runs all the bios we've collected for
378 * a particular device. We don't want to wander off to
379 * another device without first sending all of these down.
380 * So, setup a plug here and finish it off before we return
382 blk_start_plug(&plug
);
384 bdi
= device
->bdev
->bd_bdi
;
385 limit
= btrfs_async_submit_limit(fs_info
);
386 limit
= limit
* 2 / 3;
389 spin_lock(&device
->io_lock
);
394 /* take all the bios off the list at once and process them
395 * later on (without the lock held). But, remember the
396 * tail and other pointers so the bios can be properly reinserted
397 * into the list if we hit congestion
399 if (!force_reg
&& device
->pending_sync_bios
.head
) {
400 pending_bios
= &device
->pending_sync_bios
;
403 pending_bios
= &device
->pending_bios
;
407 pending
= pending_bios
->head
;
408 tail
= pending_bios
->tail
;
409 WARN_ON(pending
&& !tail
);
412 * if pending was null this time around, no bios need processing
413 * at all and we can stop. Otherwise it'll loop back up again
414 * and do an additional check so no bios are missed.
416 * device->running_pending is used to synchronize with the
419 if (device
->pending_sync_bios
.head
== NULL
&&
420 device
->pending_bios
.head
== NULL
) {
422 device
->running_pending
= 0;
425 device
->running_pending
= 1;
428 pending_bios
->head
= NULL
;
429 pending_bios
->tail
= NULL
;
431 spin_unlock(&device
->io_lock
);
436 /* we want to work on both lists, but do more bios on the
437 * sync list than the regular list
440 pending_bios
!= &device
->pending_sync_bios
&&
441 device
->pending_sync_bios
.head
) ||
442 (num_run
> 64 && pending_bios
== &device
->pending_sync_bios
&&
443 device
->pending_bios
.head
)) {
444 spin_lock(&device
->io_lock
);
445 requeue_list(pending_bios
, pending
, tail
);
450 pending
= pending
->bi_next
;
454 * atomic_dec_return implies a barrier for waitqueue_active
456 if (atomic_dec_return(&fs_info
->nr_async_bios
) < limit
&&
457 waitqueue_active(&fs_info
->async_submit_wait
))
458 wake_up(&fs_info
->async_submit_wait
);
460 BUG_ON(atomic_read(&cur
->__bi_cnt
) == 0);
463 * if we're doing the sync list, record that our
464 * plug has some sync requests on it
466 * If we're doing the regular list and there are
467 * sync requests sitting around, unplug before
470 if (pending_bios
== &device
->pending_sync_bios
) {
472 } else if (sync_pending
) {
473 blk_finish_plug(&plug
);
474 blk_start_plug(&plug
);
478 btrfsic_submit_bio(cur
);
485 * we made progress, there is more work to do and the bdi
486 * is now congested. Back off and let other work structs
489 if (pending
&& bdi_write_congested(bdi
) && batch_run
> 8 &&
490 fs_info
->fs_devices
->open_devices
> 1) {
491 struct io_context
*ioc
;
493 ioc
= current
->io_context
;
496 * the main goal here is that we don't want to
497 * block if we're going to be able to submit
498 * more requests without blocking.
500 * This code does two great things, it pokes into
501 * the elevator code from a filesystem _and_
502 * it makes assumptions about how batching works.
504 if (ioc
&& ioc
->nr_batch_requests
> 0 &&
505 time_before(jiffies
, ioc
->last_waited
+ HZ
/50UL) &&
507 ioc
->last_waited
== last_waited
)) {
509 * we want to go through our batch of
510 * requests and stop. So, we copy out
511 * the ioc->last_waited time and test
512 * against it before looping
514 last_waited
= ioc
->last_waited
;
518 spin_lock(&device
->io_lock
);
519 requeue_list(pending_bios
, pending
, tail
);
520 device
->running_pending
= 1;
522 spin_unlock(&device
->io_lock
);
523 btrfs_queue_work(fs_info
->submit_workers
,
527 /* unplug every 64 requests just for good measure */
528 if (batch_run
% 64 == 0) {
529 blk_finish_plug(&plug
);
530 blk_start_plug(&plug
);
539 spin_lock(&device
->io_lock
);
540 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
542 spin_unlock(&device
->io_lock
);
545 blk_finish_plug(&plug
);
548 static void pending_bios_fn(struct btrfs_work
*work
)
550 struct btrfs_device
*device
;
552 device
= container_of(work
, struct btrfs_device
, work
);
553 run_scheduled_bios(device
);
557 void btrfs_free_stale_device(struct btrfs_device
*cur_dev
)
559 struct btrfs_fs_devices
*fs_devs
;
560 struct btrfs_device
*dev
;
565 list_for_each_entry(fs_devs
, &fs_uuids
, list
) {
570 if (fs_devs
->seeding
)
573 list_for_each_entry(dev
, &fs_devs
->devices
, dev_list
) {
581 * Todo: This won't be enough. What if the same device
582 * comes back (with new uuid and) with its mapper path?
583 * But for now, this does help as mostly an admin will
584 * either use mapper or non mapper path throughout.
587 del
= strcmp(rcu_str_deref(dev
->name
),
588 rcu_str_deref(cur_dev
->name
));
595 /* delete the stale device */
596 if (fs_devs
->num_devices
== 1) {
597 btrfs_sysfs_remove_fsid(fs_devs
);
598 list_del(&fs_devs
->list
);
599 free_fs_devices(fs_devs
);
601 fs_devs
->num_devices
--;
602 list_del(&dev
->dev_list
);
603 rcu_string_free(dev
->name
);
612 * Add new device to list of registered devices
615 * 1 - first time device is seen
616 * 0 - device already known
619 static noinline
int device_list_add(const char *path
,
620 struct btrfs_super_block
*disk_super
,
621 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
623 struct btrfs_device
*device
;
624 struct btrfs_fs_devices
*fs_devices
;
625 struct rcu_string
*name
;
627 u64 found_transid
= btrfs_super_generation(disk_super
);
629 fs_devices
= find_fsid(disk_super
->fsid
);
631 fs_devices
= alloc_fs_devices(disk_super
->fsid
);
632 if (IS_ERR(fs_devices
))
633 return PTR_ERR(fs_devices
);
635 list_add(&fs_devices
->list
, &fs_uuids
);
639 device
= __find_device(&fs_devices
->devices
, devid
,
640 disk_super
->dev_item
.uuid
);
644 if (fs_devices
->opened
)
647 device
= btrfs_alloc_device(NULL
, &devid
,
648 disk_super
->dev_item
.uuid
);
649 if (IS_ERR(device
)) {
650 /* we can safely leave the fs_devices entry around */
651 return PTR_ERR(device
);
654 name
= rcu_string_strdup(path
, GFP_NOFS
);
659 rcu_assign_pointer(device
->name
, name
);
661 mutex_lock(&fs_devices
->device_list_mutex
);
662 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
663 fs_devices
->num_devices
++;
664 mutex_unlock(&fs_devices
->device_list_mutex
);
667 device
->fs_devices
= fs_devices
;
668 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
670 * When FS is already mounted.
671 * 1. If you are here and if the device->name is NULL that
672 * means this device was missing at time of FS mount.
673 * 2. If you are here and if the device->name is different
674 * from 'path' that means either
675 * a. The same device disappeared and reappeared with
677 * b. The missing-disk-which-was-replaced, has
680 * We must allow 1 and 2a above. But 2b would be a spurious
683 * Further in case of 1 and 2a above, the disk at 'path'
684 * would have missed some transaction when it was away and
685 * in case of 2a the stale bdev has to be updated as well.
686 * 2b must not be allowed at all time.
690 * For now, we do allow update to btrfs_fs_device through the
691 * btrfs dev scan cli after FS has been mounted. We're still
692 * tracking a problem where systems fail mount by subvolume id
693 * when we reject replacement on a mounted FS.
695 if (!fs_devices
->opened
&& found_transid
< device
->generation
) {
697 * That is if the FS is _not_ mounted and if you
698 * are here, that means there is more than one
699 * disk with same uuid and devid.We keep the one
700 * with larger generation number or the last-in if
701 * generation are equal.
706 name
= rcu_string_strdup(path
, GFP_NOFS
);
709 rcu_string_free(device
->name
);
710 rcu_assign_pointer(device
->name
, name
);
711 if (device
->missing
) {
712 fs_devices
->missing_devices
--;
718 * Unmount does not free the btrfs_device struct but would zero
719 * generation along with most of the other members. So just update
720 * it back. We need it to pick the disk with largest generation
723 if (!fs_devices
->opened
)
724 device
->generation
= found_transid
;
727 * if there is new btrfs on an already registered device,
728 * then remove the stale device entry.
731 btrfs_free_stale_device(device
);
733 *fs_devices_ret
= fs_devices
;
738 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
740 struct btrfs_fs_devices
*fs_devices
;
741 struct btrfs_device
*device
;
742 struct btrfs_device
*orig_dev
;
744 fs_devices
= alloc_fs_devices(orig
->fsid
);
745 if (IS_ERR(fs_devices
))
748 mutex_lock(&orig
->device_list_mutex
);
749 fs_devices
->total_devices
= orig
->total_devices
;
751 /* We have held the volume lock, it is safe to get the devices. */
752 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
753 struct rcu_string
*name
;
755 device
= btrfs_alloc_device(NULL
, &orig_dev
->devid
,
761 * This is ok to do without rcu read locked because we hold the
762 * uuid mutex so nothing we touch in here is going to disappear.
764 if (orig_dev
->name
) {
765 name
= rcu_string_strdup(orig_dev
->name
->str
,
771 rcu_assign_pointer(device
->name
, name
);
774 list_add(&device
->dev_list
, &fs_devices
->devices
);
775 device
->fs_devices
= fs_devices
;
776 fs_devices
->num_devices
++;
778 mutex_unlock(&orig
->device_list_mutex
);
781 mutex_unlock(&orig
->device_list_mutex
);
782 free_fs_devices(fs_devices
);
783 return ERR_PTR(-ENOMEM
);
786 void btrfs_close_extra_devices(struct btrfs_fs_devices
*fs_devices
, int step
)
788 struct btrfs_device
*device
, *next
;
789 struct btrfs_device
*latest_dev
= NULL
;
791 mutex_lock(&uuid_mutex
);
793 /* This is the initialized path, it is safe to release the devices. */
794 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
795 if (device
->in_fs_metadata
) {
796 if (!device
->is_tgtdev_for_dev_replace
&&
798 device
->generation
> latest_dev
->generation
)) {
804 if (device
->devid
== BTRFS_DEV_REPLACE_DEVID
) {
806 * In the first step, keep the device which has
807 * the correct fsid and the devid that is used
808 * for the dev_replace procedure.
809 * In the second step, the dev_replace state is
810 * read from the device tree and it is known
811 * whether the procedure is really active or
812 * not, which means whether this device is
813 * used or whether it should be removed.
815 if (step
== 0 || device
->is_tgtdev_for_dev_replace
) {
820 blkdev_put(device
->bdev
, device
->mode
);
822 fs_devices
->open_devices
--;
824 if (device
->writeable
) {
825 list_del_init(&device
->dev_alloc_list
);
826 device
->writeable
= 0;
827 if (!device
->is_tgtdev_for_dev_replace
)
828 fs_devices
->rw_devices
--;
830 list_del_init(&device
->dev_list
);
831 fs_devices
->num_devices
--;
832 rcu_string_free(device
->name
);
836 if (fs_devices
->seed
) {
837 fs_devices
= fs_devices
->seed
;
841 fs_devices
->latest_bdev
= latest_dev
->bdev
;
843 mutex_unlock(&uuid_mutex
);
846 static void __free_device(struct work_struct
*work
)
848 struct btrfs_device
*device
;
850 device
= container_of(work
, struct btrfs_device
, rcu_work
);
851 rcu_string_free(device
->name
);
852 bio_put(device
->flush_bio
);
856 static void free_device(struct rcu_head
*head
)
858 struct btrfs_device
*device
;
860 device
= container_of(head
, struct btrfs_device
, rcu
);
862 INIT_WORK(&device
->rcu_work
, __free_device
);
863 schedule_work(&device
->rcu_work
);
866 static void btrfs_close_bdev(struct btrfs_device
*device
)
868 if (device
->bdev
&& device
->writeable
) {
869 sync_blockdev(device
->bdev
);
870 invalidate_bdev(device
->bdev
);
874 blkdev_put(device
->bdev
, device
->mode
);
877 static void btrfs_prepare_close_one_device(struct btrfs_device
*device
)
879 struct btrfs_fs_devices
*fs_devices
= device
->fs_devices
;
880 struct btrfs_device
*new_device
;
881 struct rcu_string
*name
;
884 fs_devices
->open_devices
--;
886 if (device
->writeable
&&
887 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
888 list_del_init(&device
->dev_alloc_list
);
889 fs_devices
->rw_devices
--;
893 fs_devices
->missing_devices
--;
895 new_device
= btrfs_alloc_device(NULL
, &device
->devid
,
897 BUG_ON(IS_ERR(new_device
)); /* -ENOMEM */
899 /* Safe because we are under uuid_mutex */
901 name
= rcu_string_strdup(device
->name
->str
, GFP_NOFS
);
902 BUG_ON(!name
); /* -ENOMEM */
903 rcu_assign_pointer(new_device
->name
, name
);
906 list_replace_rcu(&device
->dev_list
, &new_device
->dev_list
);
907 new_device
->fs_devices
= device
->fs_devices
;
910 static int __btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
912 struct btrfs_device
*device
, *tmp
;
913 struct list_head pending_put
;
915 INIT_LIST_HEAD(&pending_put
);
917 if (--fs_devices
->opened
> 0)
920 mutex_lock(&fs_devices
->device_list_mutex
);
921 list_for_each_entry_safe(device
, tmp
, &fs_devices
->devices
, dev_list
) {
922 btrfs_prepare_close_one_device(device
);
923 list_add(&device
->dev_list
, &pending_put
);
925 mutex_unlock(&fs_devices
->device_list_mutex
);
928 * btrfs_show_devname() is using the device_list_mutex,
929 * sometimes call to blkdev_put() leads vfs calling
930 * into this func. So do put outside of device_list_mutex,
933 while (!list_empty(&pending_put
)) {
934 device
= list_first_entry(&pending_put
,
935 struct btrfs_device
, dev_list
);
936 list_del(&device
->dev_list
);
937 btrfs_close_bdev(device
);
938 call_rcu(&device
->rcu
, free_device
);
941 WARN_ON(fs_devices
->open_devices
);
942 WARN_ON(fs_devices
->rw_devices
);
943 fs_devices
->opened
= 0;
944 fs_devices
->seeding
= 0;
949 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
951 struct btrfs_fs_devices
*seed_devices
= NULL
;
954 mutex_lock(&uuid_mutex
);
955 ret
= __btrfs_close_devices(fs_devices
);
956 if (!fs_devices
->opened
) {
957 seed_devices
= fs_devices
->seed
;
958 fs_devices
->seed
= NULL
;
960 mutex_unlock(&uuid_mutex
);
962 while (seed_devices
) {
963 fs_devices
= seed_devices
;
964 seed_devices
= fs_devices
->seed
;
965 __btrfs_close_devices(fs_devices
);
966 free_fs_devices(fs_devices
);
969 * Wait for rcu kworkers under __btrfs_close_devices
970 * to finish all blkdev_puts so device is really
971 * free when umount is done.
977 static int __btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
978 fmode_t flags
, void *holder
)
980 struct request_queue
*q
;
981 struct block_device
*bdev
;
982 struct list_head
*head
= &fs_devices
->devices
;
983 struct btrfs_device
*device
;
984 struct btrfs_device
*latest_dev
= NULL
;
985 struct buffer_head
*bh
;
986 struct btrfs_super_block
*disk_super
;
993 list_for_each_entry(device
, head
, dev_list
) {
999 /* Just open everything we can; ignore failures here */
1000 if (btrfs_get_bdev_and_sb(device
->name
->str
, flags
, holder
, 1,
1004 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1005 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1006 if (devid
!= device
->devid
)
1009 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
,
1013 device
->generation
= btrfs_super_generation(disk_super
);
1015 device
->generation
> latest_dev
->generation
)
1016 latest_dev
= device
;
1018 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
1019 device
->writeable
= 0;
1021 device
->writeable
= !bdev_read_only(bdev
);
1025 q
= bdev_get_queue(bdev
);
1026 if (blk_queue_discard(q
))
1027 device
->can_discard
= 1;
1028 if (!blk_queue_nonrot(q
))
1029 fs_devices
->rotating
= 1;
1031 device
->bdev
= bdev
;
1032 device
->in_fs_metadata
= 0;
1033 device
->mode
= flags
;
1035 fs_devices
->open_devices
++;
1036 if (device
->writeable
&&
1037 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
1038 fs_devices
->rw_devices
++;
1039 list_add(&device
->dev_alloc_list
,
1040 &fs_devices
->alloc_list
);
1047 blkdev_put(bdev
, flags
);
1050 if (fs_devices
->open_devices
== 0) {
1054 fs_devices
->seeding
= seeding
;
1055 fs_devices
->opened
= 1;
1056 fs_devices
->latest_bdev
= latest_dev
->bdev
;
1057 fs_devices
->total_rw_bytes
= 0;
1062 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
1063 fmode_t flags
, void *holder
)
1067 mutex_lock(&uuid_mutex
);
1068 if (fs_devices
->opened
) {
1069 fs_devices
->opened
++;
1072 ret
= __btrfs_open_devices(fs_devices
, flags
, holder
);
1074 mutex_unlock(&uuid_mutex
);
1078 void btrfs_release_disk_super(struct page
*page
)
1084 int btrfs_read_disk_super(struct block_device
*bdev
, u64 bytenr
,
1085 struct page
**page
, struct btrfs_super_block
**disk_super
)
1090 /* make sure our super fits in the device */
1091 if (bytenr
+ PAGE_SIZE
>= i_size_read(bdev
->bd_inode
))
1094 /* make sure our super fits in the page */
1095 if (sizeof(**disk_super
) > PAGE_SIZE
)
1098 /* make sure our super doesn't straddle pages on disk */
1099 index
= bytenr
>> PAGE_SHIFT
;
1100 if ((bytenr
+ sizeof(**disk_super
) - 1) >> PAGE_SHIFT
!= index
)
1103 /* pull in the page with our super */
1104 *page
= read_cache_page_gfp(bdev
->bd_inode
->i_mapping
,
1107 if (IS_ERR_OR_NULL(*page
))
1112 /* align our pointer to the offset of the super block */
1113 *disk_super
= p
+ (bytenr
& ~PAGE_MASK
);
1115 if (btrfs_super_bytenr(*disk_super
) != bytenr
||
1116 btrfs_super_magic(*disk_super
) != BTRFS_MAGIC
) {
1117 btrfs_release_disk_super(*page
);
1121 if ((*disk_super
)->label
[0] &&
1122 (*disk_super
)->label
[BTRFS_LABEL_SIZE
- 1])
1123 (*disk_super
)->label
[BTRFS_LABEL_SIZE
- 1] = '\0';
1129 * Look for a btrfs signature on a device. This may be called out of the mount path
1130 * and we are not allowed to call set_blocksize during the scan. The superblock
1131 * is read via pagecache
1133 int btrfs_scan_one_device(const char *path
, fmode_t flags
, void *holder
,
1134 struct btrfs_fs_devices
**fs_devices_ret
)
1136 struct btrfs_super_block
*disk_super
;
1137 struct block_device
*bdev
;
1146 * we would like to check all the supers, but that would make
1147 * a btrfs mount succeed after a mkfs from a different FS.
1148 * So, we need to add a special mount option to scan for
1149 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1151 bytenr
= btrfs_sb_offset(0);
1152 flags
|= FMODE_EXCL
;
1153 mutex_lock(&uuid_mutex
);
1155 bdev
= blkdev_get_by_path(path
, flags
, holder
);
1157 ret
= PTR_ERR(bdev
);
1161 if (btrfs_read_disk_super(bdev
, bytenr
, &page
, &disk_super
))
1162 goto error_bdev_put
;
1164 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1165 transid
= btrfs_super_generation(disk_super
);
1166 total_devices
= btrfs_super_num_devices(disk_super
);
1168 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
1170 if (disk_super
->label
[0]) {
1171 pr_info("BTRFS: device label %s ", disk_super
->label
);
1173 pr_info("BTRFS: device fsid %pU ", disk_super
->fsid
);
1176 pr_cont("devid %llu transid %llu %s\n", devid
, transid
, path
);
1179 if (!ret
&& fs_devices_ret
)
1180 (*fs_devices_ret
)->total_devices
= total_devices
;
1182 btrfs_release_disk_super(page
);
1185 blkdev_put(bdev
, flags
);
1187 mutex_unlock(&uuid_mutex
);
1191 /* helper to account the used device space in the range */
1192 int btrfs_account_dev_extents_size(struct btrfs_device
*device
, u64 start
,
1193 u64 end
, u64
*length
)
1195 struct btrfs_key key
;
1196 struct btrfs_root
*root
= device
->fs_info
->dev_root
;
1197 struct btrfs_dev_extent
*dev_extent
;
1198 struct btrfs_path
*path
;
1202 struct extent_buffer
*l
;
1206 if (start
>= device
->total_bytes
|| device
->is_tgtdev_for_dev_replace
)
1209 path
= btrfs_alloc_path();
1212 path
->reada
= READA_FORWARD
;
1214 key
.objectid
= device
->devid
;
1216 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1218 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1222 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1229 slot
= path
->slots
[0];
1230 if (slot
>= btrfs_header_nritems(l
)) {
1231 ret
= btrfs_next_leaf(root
, path
);
1239 btrfs_item_key_to_cpu(l
, &key
, slot
);
1241 if (key
.objectid
< device
->devid
)
1244 if (key
.objectid
> device
->devid
)
1247 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1250 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1251 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1253 if (key
.offset
<= start
&& extent_end
> end
) {
1254 *length
= end
- start
+ 1;
1256 } else if (key
.offset
<= start
&& extent_end
> start
)
1257 *length
+= extent_end
- start
;
1258 else if (key
.offset
> start
&& extent_end
<= end
)
1259 *length
+= extent_end
- key
.offset
;
1260 else if (key
.offset
> start
&& key
.offset
<= end
) {
1261 *length
+= end
- key
.offset
+ 1;
1263 } else if (key
.offset
> end
)
1271 btrfs_free_path(path
);
1275 static int contains_pending_extent(struct btrfs_transaction
*transaction
,
1276 struct btrfs_device
*device
,
1277 u64
*start
, u64 len
)
1279 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1280 struct extent_map
*em
;
1281 struct list_head
*search_list
= &fs_info
->pinned_chunks
;
1283 u64 physical_start
= *start
;
1286 search_list
= &transaction
->pending_chunks
;
1288 list_for_each_entry(em
, search_list
, list
) {
1289 struct map_lookup
*map
;
1292 map
= em
->map_lookup
;
1293 for (i
= 0; i
< map
->num_stripes
; i
++) {
1296 if (map
->stripes
[i
].dev
!= device
)
1298 if (map
->stripes
[i
].physical
>= physical_start
+ len
||
1299 map
->stripes
[i
].physical
+ em
->orig_block_len
<=
1303 * Make sure that while processing the pinned list we do
1304 * not override our *start with a lower value, because
1305 * we can have pinned chunks that fall within this
1306 * device hole and that have lower physical addresses
1307 * than the pending chunks we processed before. If we
1308 * do not take this special care we can end up getting
1309 * 2 pending chunks that start at the same physical
1310 * device offsets because the end offset of a pinned
1311 * chunk can be equal to the start offset of some
1314 end
= map
->stripes
[i
].physical
+ em
->orig_block_len
;
1321 if (search_list
!= &fs_info
->pinned_chunks
) {
1322 search_list
= &fs_info
->pinned_chunks
;
1331 * find_free_dev_extent_start - find free space in the specified device
1332 * @device: the device which we search the free space in
1333 * @num_bytes: the size of the free space that we need
1334 * @search_start: the position from which to begin the search
1335 * @start: store the start of the free space.
1336 * @len: the size of the free space. that we find, or the size
1337 * of the max free space if we don't find suitable free space
1339 * this uses a pretty simple search, the expectation is that it is
1340 * called very infrequently and that a given device has a small number
1343 * @start is used to store the start of the free space if we find. But if we
1344 * don't find suitable free space, it will be used to store the start position
1345 * of the max free space.
1347 * @len is used to store the size of the free space that we find.
1348 * But if we don't find suitable free space, it is used to store the size of
1349 * the max free space.
1351 int find_free_dev_extent_start(struct btrfs_transaction
*transaction
,
1352 struct btrfs_device
*device
, u64 num_bytes
,
1353 u64 search_start
, u64
*start
, u64
*len
)
1355 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1356 struct btrfs_root
*root
= fs_info
->dev_root
;
1357 struct btrfs_key key
;
1358 struct btrfs_dev_extent
*dev_extent
;
1359 struct btrfs_path
*path
;
1364 u64 search_end
= device
->total_bytes
;
1367 struct extent_buffer
*l
;
1370 * We don't want to overwrite the superblock on the drive nor any area
1371 * used by the boot loader (grub for example), so we make sure to start
1372 * at an offset of at least 1MB.
1374 search_start
= max_t(u64
, search_start
, SZ_1M
);
1376 path
= btrfs_alloc_path();
1380 max_hole_start
= search_start
;
1384 if (search_start
>= search_end
|| device
->is_tgtdev_for_dev_replace
) {
1389 path
->reada
= READA_FORWARD
;
1390 path
->search_commit_root
= 1;
1391 path
->skip_locking
= 1;
1393 key
.objectid
= device
->devid
;
1394 key
.offset
= search_start
;
1395 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1397 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1401 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1408 slot
= path
->slots
[0];
1409 if (slot
>= btrfs_header_nritems(l
)) {
1410 ret
= btrfs_next_leaf(root
, path
);
1418 btrfs_item_key_to_cpu(l
, &key
, slot
);
1420 if (key
.objectid
< device
->devid
)
1423 if (key
.objectid
> device
->devid
)
1426 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1429 if (key
.offset
> search_start
) {
1430 hole_size
= key
.offset
- search_start
;
1433 * Have to check before we set max_hole_start, otherwise
1434 * we could end up sending back this offset anyway.
1436 if (contains_pending_extent(transaction
, device
,
1439 if (key
.offset
>= search_start
) {
1440 hole_size
= key
.offset
- search_start
;
1447 if (hole_size
> max_hole_size
) {
1448 max_hole_start
= search_start
;
1449 max_hole_size
= hole_size
;
1453 * If this free space is greater than which we need,
1454 * it must be the max free space that we have found
1455 * until now, so max_hole_start must point to the start
1456 * of this free space and the length of this free space
1457 * is stored in max_hole_size. Thus, we return
1458 * max_hole_start and max_hole_size and go back to the
1461 if (hole_size
>= num_bytes
) {
1467 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1468 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1470 if (extent_end
> search_start
)
1471 search_start
= extent_end
;
1478 * At this point, search_start should be the end of
1479 * allocated dev extents, and when shrinking the device,
1480 * search_end may be smaller than search_start.
1482 if (search_end
> search_start
) {
1483 hole_size
= search_end
- search_start
;
1485 if (contains_pending_extent(transaction
, device
, &search_start
,
1487 btrfs_release_path(path
);
1491 if (hole_size
> max_hole_size
) {
1492 max_hole_start
= search_start
;
1493 max_hole_size
= hole_size
;
1498 if (max_hole_size
< num_bytes
)
1504 btrfs_free_path(path
);
1505 *start
= max_hole_start
;
1507 *len
= max_hole_size
;
1511 int find_free_dev_extent(struct btrfs_trans_handle
*trans
,
1512 struct btrfs_device
*device
, u64 num_bytes
,
1513 u64
*start
, u64
*len
)
1515 /* FIXME use last free of some kind */
1516 return find_free_dev_extent_start(trans
->transaction
, device
,
1517 num_bytes
, 0, start
, len
);
1520 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1521 struct btrfs_device
*device
,
1522 u64 start
, u64
*dev_extent_len
)
1524 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1525 struct btrfs_root
*root
= fs_info
->dev_root
;
1527 struct btrfs_path
*path
;
1528 struct btrfs_key key
;
1529 struct btrfs_key found_key
;
1530 struct extent_buffer
*leaf
= NULL
;
1531 struct btrfs_dev_extent
*extent
= NULL
;
1533 path
= btrfs_alloc_path();
1537 key
.objectid
= device
->devid
;
1539 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1541 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1543 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1544 BTRFS_DEV_EXTENT_KEY
);
1547 leaf
= path
->nodes
[0];
1548 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1549 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1550 struct btrfs_dev_extent
);
1551 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1552 btrfs_dev_extent_length(leaf
, extent
) < start
);
1554 btrfs_release_path(path
);
1556 } else if (ret
== 0) {
1557 leaf
= path
->nodes
[0];
1558 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1559 struct btrfs_dev_extent
);
1561 btrfs_handle_fs_error(fs_info
, ret
, "Slot search failed");
1565 *dev_extent_len
= btrfs_dev_extent_length(leaf
, extent
);
1567 ret
= btrfs_del_item(trans
, root
, path
);
1569 btrfs_handle_fs_error(fs_info
, ret
,
1570 "Failed to remove dev extent item");
1572 set_bit(BTRFS_TRANS_HAVE_FREE_BGS
, &trans
->transaction
->flags
);
1575 btrfs_free_path(path
);
1579 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1580 struct btrfs_device
*device
,
1581 u64 chunk_tree
, u64 chunk_objectid
,
1582 u64 chunk_offset
, u64 start
, u64 num_bytes
)
1585 struct btrfs_path
*path
;
1586 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1587 struct btrfs_root
*root
= fs_info
->dev_root
;
1588 struct btrfs_dev_extent
*extent
;
1589 struct extent_buffer
*leaf
;
1590 struct btrfs_key key
;
1592 WARN_ON(!device
->in_fs_metadata
);
1593 WARN_ON(device
->is_tgtdev_for_dev_replace
);
1594 path
= btrfs_alloc_path();
1598 key
.objectid
= device
->devid
;
1600 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1601 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1606 leaf
= path
->nodes
[0];
1607 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1608 struct btrfs_dev_extent
);
1609 btrfs_set_dev_extent_chunk_tree(leaf
, extent
, chunk_tree
);
1610 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
, chunk_objectid
);
1611 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1613 write_extent_buffer_chunk_tree_uuid(leaf
, fs_info
->chunk_tree_uuid
);
1615 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1616 btrfs_mark_buffer_dirty(leaf
);
1618 btrfs_free_path(path
);
1622 static u64
find_next_chunk(struct btrfs_fs_info
*fs_info
)
1624 struct extent_map_tree
*em_tree
;
1625 struct extent_map
*em
;
1629 em_tree
= &fs_info
->mapping_tree
.map_tree
;
1630 read_lock(&em_tree
->lock
);
1631 n
= rb_last(&em_tree
->map
);
1633 em
= rb_entry(n
, struct extent_map
, rb_node
);
1634 ret
= em
->start
+ em
->len
;
1636 read_unlock(&em_tree
->lock
);
1641 static noinline
int find_next_devid(struct btrfs_fs_info
*fs_info
,
1645 struct btrfs_key key
;
1646 struct btrfs_key found_key
;
1647 struct btrfs_path
*path
;
1649 path
= btrfs_alloc_path();
1653 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1654 key
.type
= BTRFS_DEV_ITEM_KEY
;
1655 key
.offset
= (u64
)-1;
1657 ret
= btrfs_search_slot(NULL
, fs_info
->chunk_root
, &key
, path
, 0, 0);
1661 BUG_ON(ret
== 0); /* Corruption */
1663 ret
= btrfs_previous_item(fs_info
->chunk_root
, path
,
1664 BTRFS_DEV_ITEMS_OBJECTID
,
1665 BTRFS_DEV_ITEM_KEY
);
1669 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1671 *devid_ret
= found_key
.offset
+ 1;
1675 btrfs_free_path(path
);
1680 * the device information is stored in the chunk root
1681 * the btrfs_device struct should be fully filled in
1683 static int btrfs_add_device(struct btrfs_trans_handle
*trans
,
1684 struct btrfs_fs_info
*fs_info
,
1685 struct btrfs_device
*device
)
1687 struct btrfs_root
*root
= fs_info
->chunk_root
;
1689 struct btrfs_path
*path
;
1690 struct btrfs_dev_item
*dev_item
;
1691 struct extent_buffer
*leaf
;
1692 struct btrfs_key key
;
1695 path
= btrfs_alloc_path();
1699 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1700 key
.type
= BTRFS_DEV_ITEM_KEY
;
1701 key
.offset
= device
->devid
;
1703 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1708 leaf
= path
->nodes
[0];
1709 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1711 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1712 btrfs_set_device_generation(leaf
, dev_item
, 0);
1713 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1714 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1715 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1716 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1717 btrfs_set_device_total_bytes(leaf
, dev_item
,
1718 btrfs_device_get_disk_total_bytes(device
));
1719 btrfs_set_device_bytes_used(leaf
, dev_item
,
1720 btrfs_device_get_bytes_used(device
));
1721 btrfs_set_device_group(leaf
, dev_item
, 0);
1722 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1723 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1724 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1726 ptr
= btrfs_device_uuid(dev_item
);
1727 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1728 ptr
= btrfs_device_fsid(dev_item
);
1729 write_extent_buffer(leaf
, fs_info
->fsid
, ptr
, BTRFS_UUID_SIZE
);
1730 btrfs_mark_buffer_dirty(leaf
);
1734 btrfs_free_path(path
);
1739 * Function to update ctime/mtime for a given device path.
1740 * Mainly used for ctime/mtime based probe like libblkid.
1742 static void update_dev_time(const char *path_name
)
1746 filp
= filp_open(path_name
, O_RDWR
, 0);
1749 file_update_time(filp
);
1750 filp_close(filp
, NULL
);
1753 static int btrfs_rm_dev_item(struct btrfs_fs_info
*fs_info
,
1754 struct btrfs_device
*device
)
1756 struct btrfs_root
*root
= fs_info
->chunk_root
;
1758 struct btrfs_path
*path
;
1759 struct btrfs_key key
;
1760 struct btrfs_trans_handle
*trans
;
1762 path
= btrfs_alloc_path();
1766 trans
= btrfs_start_transaction(root
, 0);
1767 if (IS_ERR(trans
)) {
1768 btrfs_free_path(path
);
1769 return PTR_ERR(trans
);
1771 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1772 key
.type
= BTRFS_DEV_ITEM_KEY
;
1773 key
.offset
= device
->devid
;
1775 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1784 ret
= btrfs_del_item(trans
, root
, path
);
1788 btrfs_free_path(path
);
1789 btrfs_commit_transaction(trans
);
1794 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1795 * filesystem. It's up to the caller to adjust that number regarding eg. device
1798 static int btrfs_check_raid_min_devices(struct btrfs_fs_info
*fs_info
,
1806 seq
= read_seqbegin(&fs_info
->profiles_lock
);
1808 all_avail
= fs_info
->avail_data_alloc_bits
|
1809 fs_info
->avail_system_alloc_bits
|
1810 fs_info
->avail_metadata_alloc_bits
;
1811 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
1813 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
1814 if (!(all_avail
& btrfs_raid_group
[i
]))
1817 if (num_devices
< btrfs_raid_array
[i
].devs_min
) {
1818 int ret
= btrfs_raid_mindev_error
[i
];
1828 struct btrfs_device
*btrfs_find_next_active_device(struct btrfs_fs_devices
*fs_devs
,
1829 struct btrfs_device
*device
)
1831 struct btrfs_device
*next_device
;
1833 list_for_each_entry(next_device
, &fs_devs
->devices
, dev_list
) {
1834 if (next_device
!= device
&&
1835 !next_device
->missing
&& next_device
->bdev
)
1843 * Helper function to check if the given device is part of s_bdev / latest_bdev
1844 * and replace it with the provided or the next active device, in the context
1845 * where this function called, there should be always be another device (or
1846 * this_dev) which is active.
1848 void btrfs_assign_next_active_device(struct btrfs_fs_info
*fs_info
,
1849 struct btrfs_device
*device
, struct btrfs_device
*this_dev
)
1851 struct btrfs_device
*next_device
;
1854 next_device
= this_dev
;
1856 next_device
= btrfs_find_next_active_device(fs_info
->fs_devices
,
1858 ASSERT(next_device
);
1860 if (fs_info
->sb
->s_bdev
&&
1861 (fs_info
->sb
->s_bdev
== device
->bdev
))
1862 fs_info
->sb
->s_bdev
= next_device
->bdev
;
1864 if (fs_info
->fs_devices
->latest_bdev
== device
->bdev
)
1865 fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1868 int btrfs_rm_device(struct btrfs_fs_info
*fs_info
, const char *device_path
,
1871 struct btrfs_device
*device
;
1872 struct btrfs_fs_devices
*cur_devices
;
1875 bool clear_super
= false;
1877 mutex_lock(&uuid_mutex
);
1879 num_devices
= fs_info
->fs_devices
->num_devices
;
1880 btrfs_dev_replace_lock(&fs_info
->dev_replace
, 0);
1881 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
1882 WARN_ON(num_devices
< 1);
1885 btrfs_dev_replace_unlock(&fs_info
->dev_replace
, 0);
1887 ret
= btrfs_check_raid_min_devices(fs_info
, num_devices
- 1);
1891 ret
= btrfs_find_device_by_devspec(fs_info
, devid
, device_path
,
1896 if (device
->is_tgtdev_for_dev_replace
) {
1897 ret
= BTRFS_ERROR_DEV_TGT_REPLACE
;
1901 if (device
->writeable
&& fs_info
->fs_devices
->rw_devices
== 1) {
1902 ret
= BTRFS_ERROR_DEV_ONLY_WRITABLE
;
1906 if (device
->writeable
) {
1907 mutex_lock(&fs_info
->chunk_mutex
);
1908 list_del_init(&device
->dev_alloc_list
);
1909 device
->fs_devices
->rw_devices
--;
1910 mutex_unlock(&fs_info
->chunk_mutex
);
1914 mutex_unlock(&uuid_mutex
);
1915 ret
= btrfs_shrink_device(device
, 0);
1916 mutex_lock(&uuid_mutex
);
1921 * TODO: the superblock still includes this device in its num_devices
1922 * counter although write_all_supers() is not locked out. This
1923 * could give a filesystem state which requires a degraded mount.
1925 ret
= btrfs_rm_dev_item(fs_info
, device
);
1929 device
->in_fs_metadata
= 0;
1930 btrfs_scrub_cancel_dev(fs_info
, device
);
1933 * the device list mutex makes sure that we don't change
1934 * the device list while someone else is writing out all
1935 * the device supers. Whoever is writing all supers, should
1936 * lock the device list mutex before getting the number of
1937 * devices in the super block (super_copy). Conversely,
1938 * whoever updates the number of devices in the super block
1939 * (super_copy) should hold the device list mutex.
1942 cur_devices
= device
->fs_devices
;
1943 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
1944 list_del_rcu(&device
->dev_list
);
1946 device
->fs_devices
->num_devices
--;
1947 device
->fs_devices
->total_devices
--;
1949 if (device
->missing
)
1950 device
->fs_devices
->missing_devices
--;
1952 btrfs_assign_next_active_device(fs_info
, device
, NULL
);
1955 device
->fs_devices
->open_devices
--;
1956 /* remove sysfs entry */
1957 btrfs_sysfs_rm_device_link(fs_info
->fs_devices
, device
);
1960 num_devices
= btrfs_super_num_devices(fs_info
->super_copy
) - 1;
1961 btrfs_set_super_num_devices(fs_info
->super_copy
, num_devices
);
1962 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
1965 * at this point, the device is zero sized and detached from
1966 * the devices list. All that's left is to zero out the old
1967 * supers and free the device.
1969 if (device
->writeable
)
1970 btrfs_scratch_superblocks(device
->bdev
, device
->name
->str
);
1972 btrfs_close_bdev(device
);
1973 call_rcu(&device
->rcu
, free_device
);
1975 if (cur_devices
->open_devices
== 0) {
1976 struct btrfs_fs_devices
*fs_devices
;
1977 fs_devices
= fs_info
->fs_devices
;
1978 while (fs_devices
) {
1979 if (fs_devices
->seed
== cur_devices
) {
1980 fs_devices
->seed
= cur_devices
->seed
;
1983 fs_devices
= fs_devices
->seed
;
1985 cur_devices
->seed
= NULL
;
1986 __btrfs_close_devices(cur_devices
);
1987 free_fs_devices(cur_devices
);
1990 fs_info
->num_tolerated_disk_barrier_failures
=
1991 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
1994 mutex_unlock(&uuid_mutex
);
1998 if (device
->writeable
) {
1999 mutex_lock(&fs_info
->chunk_mutex
);
2000 list_add(&device
->dev_alloc_list
,
2001 &fs_info
->fs_devices
->alloc_list
);
2002 device
->fs_devices
->rw_devices
++;
2003 mutex_unlock(&fs_info
->chunk_mutex
);
2008 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info
*fs_info
,
2009 struct btrfs_device
*srcdev
)
2011 struct btrfs_fs_devices
*fs_devices
;
2013 WARN_ON(!mutex_is_locked(&fs_info
->fs_devices
->device_list_mutex
));
2016 * in case of fs with no seed, srcdev->fs_devices will point
2017 * to fs_devices of fs_info. However when the dev being replaced is
2018 * a seed dev it will point to the seed's local fs_devices. In short
2019 * srcdev will have its correct fs_devices in both the cases.
2021 fs_devices
= srcdev
->fs_devices
;
2023 list_del_rcu(&srcdev
->dev_list
);
2024 list_del_rcu(&srcdev
->dev_alloc_list
);
2025 fs_devices
->num_devices
--;
2026 if (srcdev
->missing
)
2027 fs_devices
->missing_devices
--;
2029 if (srcdev
->writeable
)
2030 fs_devices
->rw_devices
--;
2033 fs_devices
->open_devices
--;
2036 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info
*fs_info
,
2037 struct btrfs_device
*srcdev
)
2039 struct btrfs_fs_devices
*fs_devices
= srcdev
->fs_devices
;
2041 if (srcdev
->writeable
) {
2042 /* zero out the old super if it is writable */
2043 btrfs_scratch_superblocks(srcdev
->bdev
, srcdev
->name
->str
);
2046 btrfs_close_bdev(srcdev
);
2048 call_rcu(&srcdev
->rcu
, free_device
);
2051 * unless fs_devices is seed fs, num_devices shouldn't go
2054 BUG_ON(!fs_devices
->num_devices
&& !fs_devices
->seeding
);
2056 /* if this is no devs we rather delete the fs_devices */
2057 if (!fs_devices
->num_devices
) {
2058 struct btrfs_fs_devices
*tmp_fs_devices
;
2060 tmp_fs_devices
= fs_info
->fs_devices
;
2061 while (tmp_fs_devices
) {
2062 if (tmp_fs_devices
->seed
== fs_devices
) {
2063 tmp_fs_devices
->seed
= fs_devices
->seed
;
2066 tmp_fs_devices
= tmp_fs_devices
->seed
;
2068 fs_devices
->seed
= NULL
;
2069 __btrfs_close_devices(fs_devices
);
2070 free_fs_devices(fs_devices
);
2074 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
2075 struct btrfs_device
*tgtdev
)
2077 mutex_lock(&uuid_mutex
);
2079 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2081 btrfs_sysfs_rm_device_link(fs_info
->fs_devices
, tgtdev
);
2084 fs_info
->fs_devices
->open_devices
--;
2086 fs_info
->fs_devices
->num_devices
--;
2088 btrfs_assign_next_active_device(fs_info
, tgtdev
, NULL
);
2090 list_del_rcu(&tgtdev
->dev_list
);
2092 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2093 mutex_unlock(&uuid_mutex
);
2096 * The update_dev_time() with in btrfs_scratch_superblocks()
2097 * may lead to a call to btrfs_show_devname() which will try
2098 * to hold device_list_mutex. And here this device
2099 * is already out of device list, so we don't have to hold
2100 * the device_list_mutex lock.
2102 btrfs_scratch_superblocks(tgtdev
->bdev
, tgtdev
->name
->str
);
2104 btrfs_close_bdev(tgtdev
);
2105 call_rcu(&tgtdev
->rcu
, free_device
);
2108 static int btrfs_find_device_by_path(struct btrfs_fs_info
*fs_info
,
2109 const char *device_path
,
2110 struct btrfs_device
**device
)
2113 struct btrfs_super_block
*disk_super
;
2116 struct block_device
*bdev
;
2117 struct buffer_head
*bh
;
2120 ret
= btrfs_get_bdev_and_sb(device_path
, FMODE_READ
,
2121 fs_info
->bdev_holder
, 0, &bdev
, &bh
);
2124 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
2125 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
2126 dev_uuid
= disk_super
->dev_item
.uuid
;
2127 *device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, disk_super
->fsid
);
2131 blkdev_put(bdev
, FMODE_READ
);
2135 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info
*fs_info
,
2136 const char *device_path
,
2137 struct btrfs_device
**device
)
2140 if (strcmp(device_path
, "missing") == 0) {
2141 struct list_head
*devices
;
2142 struct btrfs_device
*tmp
;
2144 devices
= &fs_info
->fs_devices
->devices
;
2146 * It is safe to read the devices since the volume_mutex
2147 * is held by the caller.
2149 list_for_each_entry(tmp
, devices
, dev_list
) {
2150 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
2157 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND
;
2161 return btrfs_find_device_by_path(fs_info
, device_path
, device
);
2166 * Lookup a device given by device id, or the path if the id is 0.
2168 int btrfs_find_device_by_devspec(struct btrfs_fs_info
*fs_info
, u64 devid
,
2169 const char *devpath
,
2170 struct btrfs_device
**device
)
2176 *device
= btrfs_find_device(fs_info
, devid
, NULL
, NULL
);
2180 if (!devpath
|| !devpath
[0])
2183 ret
= btrfs_find_device_missing_or_by_path(fs_info
, devpath
,
2190 * does all the dirty work required for changing file system's UUID.
2192 static int btrfs_prepare_sprout(struct btrfs_fs_info
*fs_info
)
2194 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2195 struct btrfs_fs_devices
*old_devices
;
2196 struct btrfs_fs_devices
*seed_devices
;
2197 struct btrfs_super_block
*disk_super
= fs_info
->super_copy
;
2198 struct btrfs_device
*device
;
2201 BUG_ON(!mutex_is_locked(&uuid_mutex
));
2202 if (!fs_devices
->seeding
)
2205 seed_devices
= __alloc_fs_devices();
2206 if (IS_ERR(seed_devices
))
2207 return PTR_ERR(seed_devices
);
2209 old_devices
= clone_fs_devices(fs_devices
);
2210 if (IS_ERR(old_devices
)) {
2211 kfree(seed_devices
);
2212 return PTR_ERR(old_devices
);
2215 list_add(&old_devices
->list
, &fs_uuids
);
2217 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
2218 seed_devices
->opened
= 1;
2219 INIT_LIST_HEAD(&seed_devices
->devices
);
2220 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
2221 mutex_init(&seed_devices
->device_list_mutex
);
2223 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2224 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
2226 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
)
2227 device
->fs_devices
= seed_devices
;
2229 mutex_lock(&fs_info
->chunk_mutex
);
2230 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
2231 mutex_unlock(&fs_info
->chunk_mutex
);
2233 fs_devices
->seeding
= 0;
2234 fs_devices
->num_devices
= 0;
2235 fs_devices
->open_devices
= 0;
2236 fs_devices
->missing_devices
= 0;
2237 fs_devices
->rotating
= 0;
2238 fs_devices
->seed
= seed_devices
;
2240 generate_random_uuid(fs_devices
->fsid
);
2241 memcpy(fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2242 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2243 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2245 super_flags
= btrfs_super_flags(disk_super
) &
2246 ~BTRFS_SUPER_FLAG_SEEDING
;
2247 btrfs_set_super_flags(disk_super
, super_flags
);
2253 * Store the expected generation for seed devices in device items.
2255 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
2256 struct btrfs_fs_info
*fs_info
)
2258 struct btrfs_root
*root
= fs_info
->chunk_root
;
2259 struct btrfs_path
*path
;
2260 struct extent_buffer
*leaf
;
2261 struct btrfs_dev_item
*dev_item
;
2262 struct btrfs_device
*device
;
2263 struct btrfs_key key
;
2264 u8 fs_uuid
[BTRFS_UUID_SIZE
];
2265 u8 dev_uuid
[BTRFS_UUID_SIZE
];
2269 path
= btrfs_alloc_path();
2273 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2275 key
.type
= BTRFS_DEV_ITEM_KEY
;
2278 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2282 leaf
= path
->nodes
[0];
2284 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
2285 ret
= btrfs_next_leaf(root
, path
);
2290 leaf
= path
->nodes
[0];
2291 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2292 btrfs_release_path(path
);
2296 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2297 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
2298 key
.type
!= BTRFS_DEV_ITEM_KEY
)
2301 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2302 struct btrfs_dev_item
);
2303 devid
= btrfs_device_id(leaf
, dev_item
);
2304 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
2306 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
2308 device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, fs_uuid
);
2309 BUG_ON(!device
); /* Logic error */
2311 if (device
->fs_devices
->seeding
) {
2312 btrfs_set_device_generation(leaf
, dev_item
,
2313 device
->generation
);
2314 btrfs_mark_buffer_dirty(leaf
);
2322 btrfs_free_path(path
);
2326 int btrfs_init_new_device(struct btrfs_fs_info
*fs_info
, const char *device_path
)
2328 struct btrfs_root
*root
= fs_info
->dev_root
;
2329 struct request_queue
*q
;
2330 struct btrfs_trans_handle
*trans
;
2331 struct btrfs_device
*device
;
2332 struct block_device
*bdev
;
2333 struct list_head
*devices
;
2334 struct super_block
*sb
= fs_info
->sb
;
2335 struct rcu_string
*name
;
2337 int seeding_dev
= 0;
2340 if ((sb
->s_flags
& MS_RDONLY
) && !fs_info
->fs_devices
->seeding
)
2343 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2344 fs_info
->bdev_holder
);
2346 return PTR_ERR(bdev
);
2348 if (fs_info
->fs_devices
->seeding
) {
2350 down_write(&sb
->s_umount
);
2351 mutex_lock(&uuid_mutex
);
2354 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2356 devices
= &fs_info
->fs_devices
->devices
;
2358 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2359 list_for_each_entry(device
, devices
, dev_list
) {
2360 if (device
->bdev
== bdev
) {
2363 &fs_info
->fs_devices
->device_list_mutex
);
2367 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2369 device
= btrfs_alloc_device(fs_info
, NULL
, NULL
);
2370 if (IS_ERR(device
)) {
2371 /* we can safely leave the fs_devices entry around */
2372 ret
= PTR_ERR(device
);
2376 name
= rcu_string_strdup(device_path
, GFP_KERNEL
);
2382 rcu_assign_pointer(device
->name
, name
);
2384 trans
= btrfs_start_transaction(root
, 0);
2385 if (IS_ERR(trans
)) {
2386 rcu_string_free(device
->name
);
2388 ret
= PTR_ERR(trans
);
2392 q
= bdev_get_queue(bdev
);
2393 if (blk_queue_discard(q
))
2394 device
->can_discard
= 1;
2395 device
->writeable
= 1;
2396 device
->generation
= trans
->transid
;
2397 device
->io_width
= fs_info
->sectorsize
;
2398 device
->io_align
= fs_info
->sectorsize
;
2399 device
->sector_size
= fs_info
->sectorsize
;
2400 device
->total_bytes
= round_down(i_size_read(bdev
->bd_inode
),
2401 fs_info
->sectorsize
);
2402 device
->disk_total_bytes
= device
->total_bytes
;
2403 device
->commit_total_bytes
= device
->total_bytes
;
2404 device
->fs_info
= fs_info
;
2405 device
->bdev
= bdev
;
2406 device
->in_fs_metadata
= 1;
2407 device
->is_tgtdev_for_dev_replace
= 0;
2408 device
->mode
= FMODE_EXCL
;
2409 device
->dev_stats_valid
= 1;
2410 set_blocksize(device
->bdev
, 4096);
2413 sb
->s_flags
&= ~MS_RDONLY
;
2414 ret
= btrfs_prepare_sprout(fs_info
);
2415 BUG_ON(ret
); /* -ENOMEM */
2418 device
->fs_devices
= fs_info
->fs_devices
;
2420 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2421 mutex_lock(&fs_info
->chunk_mutex
);
2422 list_add_rcu(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2423 list_add(&device
->dev_alloc_list
,
2424 &fs_info
->fs_devices
->alloc_list
);
2425 fs_info
->fs_devices
->num_devices
++;
2426 fs_info
->fs_devices
->open_devices
++;
2427 fs_info
->fs_devices
->rw_devices
++;
2428 fs_info
->fs_devices
->total_devices
++;
2429 fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
2431 atomic64_add(device
->total_bytes
, &fs_info
->free_chunk_space
);
2433 if (!blk_queue_nonrot(q
))
2434 fs_info
->fs_devices
->rotating
= 1;
2436 tmp
= btrfs_super_total_bytes(fs_info
->super_copy
);
2437 btrfs_set_super_total_bytes(fs_info
->super_copy
,
2438 round_down(tmp
+ device
->total_bytes
, fs_info
->sectorsize
));
2440 tmp
= btrfs_super_num_devices(fs_info
->super_copy
);
2441 btrfs_set_super_num_devices(fs_info
->super_copy
, tmp
+ 1);
2443 /* add sysfs device entry */
2444 btrfs_sysfs_add_device_link(fs_info
->fs_devices
, device
);
2447 * we've got more storage, clear any full flags on the space
2450 btrfs_clear_space_info_full(fs_info
);
2452 mutex_unlock(&fs_info
->chunk_mutex
);
2453 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2456 mutex_lock(&fs_info
->chunk_mutex
);
2457 ret
= init_first_rw_device(trans
, fs_info
);
2458 mutex_unlock(&fs_info
->chunk_mutex
);
2460 btrfs_abort_transaction(trans
, ret
);
2465 ret
= btrfs_add_device(trans
, fs_info
, device
);
2467 btrfs_abort_transaction(trans
, ret
);
2472 char fsid_buf
[BTRFS_UUID_UNPARSED_SIZE
];
2474 ret
= btrfs_finish_sprout(trans
, fs_info
);
2476 btrfs_abort_transaction(trans
, ret
);
2480 /* Sprouting would change fsid of the mounted root,
2481 * so rename the fsid on the sysfs
2483 snprintf(fsid_buf
, BTRFS_UUID_UNPARSED_SIZE
, "%pU",
2485 if (kobject_rename(&fs_info
->fs_devices
->fsid_kobj
, fsid_buf
))
2487 "sysfs: failed to create fsid for sprout");
2490 fs_info
->num_tolerated_disk_barrier_failures
=
2491 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
2492 ret
= btrfs_commit_transaction(trans
);
2495 mutex_unlock(&uuid_mutex
);
2496 up_write(&sb
->s_umount
);
2498 if (ret
) /* transaction commit */
2501 ret
= btrfs_relocate_sys_chunks(fs_info
);
2503 btrfs_handle_fs_error(fs_info
, ret
,
2504 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2505 trans
= btrfs_attach_transaction(root
);
2506 if (IS_ERR(trans
)) {
2507 if (PTR_ERR(trans
) == -ENOENT
)
2509 return PTR_ERR(trans
);
2511 ret
= btrfs_commit_transaction(trans
);
2514 /* Update ctime/mtime for libblkid */
2515 update_dev_time(device_path
);
2519 btrfs_end_transaction(trans
);
2520 rcu_string_free(device
->name
);
2521 btrfs_sysfs_rm_device_link(fs_info
->fs_devices
, device
);
2524 blkdev_put(bdev
, FMODE_EXCL
);
2526 mutex_unlock(&uuid_mutex
);
2527 up_write(&sb
->s_umount
);
2532 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
2533 const char *device_path
,
2534 struct btrfs_device
*srcdev
,
2535 struct btrfs_device
**device_out
)
2537 struct request_queue
*q
;
2538 struct btrfs_device
*device
;
2539 struct block_device
*bdev
;
2540 struct list_head
*devices
;
2541 struct rcu_string
*name
;
2542 u64 devid
= BTRFS_DEV_REPLACE_DEVID
;
2546 if (fs_info
->fs_devices
->seeding
) {
2547 btrfs_err(fs_info
, "the filesystem is a seed filesystem!");
2551 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2552 fs_info
->bdev_holder
);
2554 btrfs_err(fs_info
, "target device %s is invalid!", device_path
);
2555 return PTR_ERR(bdev
);
2558 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2560 devices
= &fs_info
->fs_devices
->devices
;
2561 list_for_each_entry(device
, devices
, dev_list
) {
2562 if (device
->bdev
== bdev
) {
2564 "target device is in the filesystem!");
2571 if (i_size_read(bdev
->bd_inode
) <
2572 btrfs_device_get_total_bytes(srcdev
)) {
2574 "target device is smaller than source device!");
2580 device
= btrfs_alloc_device(NULL
, &devid
, NULL
);
2581 if (IS_ERR(device
)) {
2582 ret
= PTR_ERR(device
);
2586 name
= rcu_string_strdup(device_path
, GFP_KERNEL
);
2592 rcu_assign_pointer(device
->name
, name
);
2594 q
= bdev_get_queue(bdev
);
2595 if (blk_queue_discard(q
))
2596 device
->can_discard
= 1;
2597 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2598 device
->writeable
= 1;
2599 device
->generation
= 0;
2600 device
->io_width
= fs_info
->sectorsize
;
2601 device
->io_align
= fs_info
->sectorsize
;
2602 device
->sector_size
= fs_info
->sectorsize
;
2603 device
->total_bytes
= btrfs_device_get_total_bytes(srcdev
);
2604 device
->disk_total_bytes
= btrfs_device_get_disk_total_bytes(srcdev
);
2605 device
->bytes_used
= btrfs_device_get_bytes_used(srcdev
);
2606 ASSERT(list_empty(&srcdev
->resized_list
));
2607 device
->commit_total_bytes
= srcdev
->commit_total_bytes
;
2608 device
->commit_bytes_used
= device
->bytes_used
;
2609 device
->fs_info
= fs_info
;
2610 device
->bdev
= bdev
;
2611 device
->in_fs_metadata
= 1;
2612 device
->is_tgtdev_for_dev_replace
= 1;
2613 device
->mode
= FMODE_EXCL
;
2614 device
->dev_stats_valid
= 1;
2615 set_blocksize(device
->bdev
, 4096);
2616 device
->fs_devices
= fs_info
->fs_devices
;
2617 list_add(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2618 fs_info
->fs_devices
->num_devices
++;
2619 fs_info
->fs_devices
->open_devices
++;
2620 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2622 *device_out
= device
;
2626 blkdev_put(bdev
, FMODE_EXCL
);
2630 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info
*fs_info
,
2631 struct btrfs_device
*tgtdev
)
2633 u32 sectorsize
= fs_info
->sectorsize
;
2635 WARN_ON(fs_info
->fs_devices
->rw_devices
== 0);
2636 tgtdev
->io_width
= sectorsize
;
2637 tgtdev
->io_align
= sectorsize
;
2638 tgtdev
->sector_size
= sectorsize
;
2639 tgtdev
->fs_info
= fs_info
;
2640 tgtdev
->in_fs_metadata
= 1;
2643 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2644 struct btrfs_device
*device
)
2647 struct btrfs_path
*path
;
2648 struct btrfs_root
*root
= device
->fs_info
->chunk_root
;
2649 struct btrfs_dev_item
*dev_item
;
2650 struct extent_buffer
*leaf
;
2651 struct btrfs_key key
;
2653 path
= btrfs_alloc_path();
2657 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2658 key
.type
= BTRFS_DEV_ITEM_KEY
;
2659 key
.offset
= device
->devid
;
2661 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2670 leaf
= path
->nodes
[0];
2671 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2673 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2674 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2675 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2676 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2677 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2678 btrfs_set_device_total_bytes(leaf
, dev_item
,
2679 btrfs_device_get_disk_total_bytes(device
));
2680 btrfs_set_device_bytes_used(leaf
, dev_item
,
2681 btrfs_device_get_bytes_used(device
));
2682 btrfs_mark_buffer_dirty(leaf
);
2685 btrfs_free_path(path
);
2689 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2690 struct btrfs_device
*device
, u64 new_size
)
2692 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
2693 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
2694 struct btrfs_fs_devices
*fs_devices
;
2698 if (!device
->writeable
)
2701 new_size
= round_down(new_size
, fs_info
->sectorsize
);
2703 mutex_lock(&fs_info
->chunk_mutex
);
2704 old_total
= btrfs_super_total_bytes(super_copy
);
2705 diff
= new_size
- device
->total_bytes
;
2707 if (new_size
<= device
->total_bytes
||
2708 device
->is_tgtdev_for_dev_replace
) {
2709 mutex_unlock(&fs_info
->chunk_mutex
);
2713 fs_devices
= fs_info
->fs_devices
;
2715 btrfs_set_super_total_bytes(super_copy
,
2716 round_down(old_total
+ diff
, fs_info
->sectorsize
));
2717 device
->fs_devices
->total_rw_bytes
+= diff
;
2719 btrfs_device_set_total_bytes(device
, new_size
);
2720 btrfs_device_set_disk_total_bytes(device
, new_size
);
2721 btrfs_clear_space_info_full(device
->fs_info
);
2722 if (list_empty(&device
->resized_list
))
2723 list_add_tail(&device
->resized_list
,
2724 &fs_devices
->resized_devices
);
2725 mutex_unlock(&fs_info
->chunk_mutex
);
2727 return btrfs_update_device(trans
, device
);
2730 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
2731 struct btrfs_fs_info
*fs_info
, u64 chunk_objectid
,
2734 struct btrfs_root
*root
= fs_info
->chunk_root
;
2736 struct btrfs_path
*path
;
2737 struct btrfs_key key
;
2739 path
= btrfs_alloc_path();
2743 key
.objectid
= chunk_objectid
;
2744 key
.offset
= chunk_offset
;
2745 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2747 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2750 else if (ret
> 0) { /* Logic error or corruption */
2751 btrfs_handle_fs_error(fs_info
, -ENOENT
,
2752 "Failed lookup while freeing chunk.");
2757 ret
= btrfs_del_item(trans
, root
, path
);
2759 btrfs_handle_fs_error(fs_info
, ret
,
2760 "Failed to delete chunk item.");
2762 btrfs_free_path(path
);
2766 static int btrfs_del_sys_chunk(struct btrfs_fs_info
*fs_info
,
2767 u64 chunk_objectid
, u64 chunk_offset
)
2769 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
2770 struct btrfs_disk_key
*disk_key
;
2771 struct btrfs_chunk
*chunk
;
2778 struct btrfs_key key
;
2780 mutex_lock(&fs_info
->chunk_mutex
);
2781 array_size
= btrfs_super_sys_array_size(super_copy
);
2783 ptr
= super_copy
->sys_chunk_array
;
2786 while (cur
< array_size
) {
2787 disk_key
= (struct btrfs_disk_key
*)ptr
;
2788 btrfs_disk_key_to_cpu(&key
, disk_key
);
2790 len
= sizeof(*disk_key
);
2792 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2793 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2794 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2795 len
+= btrfs_chunk_item_size(num_stripes
);
2800 if (key
.objectid
== chunk_objectid
&&
2801 key
.offset
== chunk_offset
) {
2802 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2804 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2810 mutex_unlock(&fs_info
->chunk_mutex
);
2814 static struct extent_map
*get_chunk_map(struct btrfs_fs_info
*fs_info
,
2815 u64 logical
, u64 length
)
2817 struct extent_map_tree
*em_tree
;
2818 struct extent_map
*em
;
2820 em_tree
= &fs_info
->mapping_tree
.map_tree
;
2821 read_lock(&em_tree
->lock
);
2822 em
= lookup_extent_mapping(em_tree
, logical
, length
);
2823 read_unlock(&em_tree
->lock
);
2826 btrfs_crit(fs_info
, "unable to find logical %llu length %llu",
2828 return ERR_PTR(-EINVAL
);
2831 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
2833 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2834 logical
, length
, em
->start
, em
->start
+ em
->len
);
2835 free_extent_map(em
);
2836 return ERR_PTR(-EINVAL
);
2839 /* callers are responsible for dropping em's ref. */
2843 int btrfs_remove_chunk(struct btrfs_trans_handle
*trans
,
2844 struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2846 struct extent_map
*em
;
2847 struct map_lookup
*map
;
2848 u64 dev_extent_len
= 0;
2849 u64 chunk_objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2851 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2853 em
= get_chunk_map(fs_info
, chunk_offset
, 1);
2856 * This is a logic error, but we don't want to just rely on the
2857 * user having built with ASSERT enabled, so if ASSERT doesn't
2858 * do anything we still error out.
2863 map
= em
->map_lookup
;
2864 mutex_lock(&fs_info
->chunk_mutex
);
2865 check_system_chunk(trans
, fs_info
, map
->type
);
2866 mutex_unlock(&fs_info
->chunk_mutex
);
2869 * Take the device list mutex to prevent races with the final phase of
2870 * a device replace operation that replaces the device object associated
2871 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2873 mutex_lock(&fs_devices
->device_list_mutex
);
2874 for (i
= 0; i
< map
->num_stripes
; i
++) {
2875 struct btrfs_device
*device
= map
->stripes
[i
].dev
;
2876 ret
= btrfs_free_dev_extent(trans
, device
,
2877 map
->stripes
[i
].physical
,
2880 mutex_unlock(&fs_devices
->device_list_mutex
);
2881 btrfs_abort_transaction(trans
, ret
);
2885 if (device
->bytes_used
> 0) {
2886 mutex_lock(&fs_info
->chunk_mutex
);
2887 btrfs_device_set_bytes_used(device
,
2888 device
->bytes_used
- dev_extent_len
);
2889 atomic64_add(dev_extent_len
, &fs_info
->free_chunk_space
);
2890 btrfs_clear_space_info_full(fs_info
);
2891 mutex_unlock(&fs_info
->chunk_mutex
);
2894 if (map
->stripes
[i
].dev
) {
2895 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2897 mutex_unlock(&fs_devices
->device_list_mutex
);
2898 btrfs_abort_transaction(trans
, ret
);
2903 mutex_unlock(&fs_devices
->device_list_mutex
);
2905 ret
= btrfs_free_chunk(trans
, fs_info
, chunk_objectid
, chunk_offset
);
2907 btrfs_abort_transaction(trans
, ret
);
2911 trace_btrfs_chunk_free(fs_info
, map
, chunk_offset
, em
->len
);
2913 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2914 ret
= btrfs_del_sys_chunk(fs_info
, chunk_objectid
,
2917 btrfs_abort_transaction(trans
, ret
);
2922 ret
= btrfs_remove_block_group(trans
, fs_info
, chunk_offset
, em
);
2924 btrfs_abort_transaction(trans
, ret
);
2930 free_extent_map(em
);
2934 static int btrfs_relocate_chunk(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2936 struct btrfs_root
*root
= fs_info
->chunk_root
;
2937 struct btrfs_trans_handle
*trans
;
2941 * Prevent races with automatic removal of unused block groups.
2942 * After we relocate and before we remove the chunk with offset
2943 * chunk_offset, automatic removal of the block group can kick in,
2944 * resulting in a failure when calling btrfs_remove_chunk() below.
2946 * Make sure to acquire this mutex before doing a tree search (dev
2947 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2948 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2949 * we release the path used to search the chunk/dev tree and before
2950 * the current task acquires this mutex and calls us.
2952 ASSERT(mutex_is_locked(&fs_info
->delete_unused_bgs_mutex
));
2954 ret
= btrfs_can_relocate(fs_info
, chunk_offset
);
2958 /* step one, relocate all the extents inside this chunk */
2959 btrfs_scrub_pause(fs_info
);
2960 ret
= btrfs_relocate_block_group(fs_info
, chunk_offset
);
2961 btrfs_scrub_continue(fs_info
);
2965 trans
= btrfs_start_trans_remove_block_group(root
->fs_info
,
2967 if (IS_ERR(trans
)) {
2968 ret
= PTR_ERR(trans
);
2969 btrfs_handle_fs_error(root
->fs_info
, ret
, NULL
);
2974 * step two, delete the device extents and the
2975 * chunk tree entries
2977 ret
= btrfs_remove_chunk(trans
, fs_info
, chunk_offset
);
2978 btrfs_end_transaction(trans
);
2982 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info
*fs_info
)
2984 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
2985 struct btrfs_path
*path
;
2986 struct extent_buffer
*leaf
;
2987 struct btrfs_chunk
*chunk
;
2988 struct btrfs_key key
;
2989 struct btrfs_key found_key
;
2991 bool retried
= false;
2995 path
= btrfs_alloc_path();
3000 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3001 key
.offset
= (u64
)-1;
3002 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3005 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
3006 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
3008 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3011 BUG_ON(ret
== 0); /* Corruption */
3013 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
3016 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3022 leaf
= path
->nodes
[0];
3023 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3025 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
3026 struct btrfs_chunk
);
3027 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3028 btrfs_release_path(path
);
3030 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3031 ret
= btrfs_relocate_chunk(fs_info
, found_key
.offset
);
3037 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3039 if (found_key
.offset
== 0)
3041 key
.offset
= found_key
.offset
- 1;
3044 if (failed
&& !retried
) {
3048 } else if (WARN_ON(failed
&& retried
)) {
3052 btrfs_free_path(path
);
3056 static int insert_balance_item(struct btrfs_fs_info
*fs_info
,
3057 struct btrfs_balance_control
*bctl
)
3059 struct btrfs_root
*root
= fs_info
->tree_root
;
3060 struct btrfs_trans_handle
*trans
;
3061 struct btrfs_balance_item
*item
;
3062 struct btrfs_disk_balance_args disk_bargs
;
3063 struct btrfs_path
*path
;
3064 struct extent_buffer
*leaf
;
3065 struct btrfs_key key
;
3068 path
= btrfs_alloc_path();
3072 trans
= btrfs_start_transaction(root
, 0);
3073 if (IS_ERR(trans
)) {
3074 btrfs_free_path(path
);
3075 return PTR_ERR(trans
);
3078 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3079 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3082 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
3087 leaf
= path
->nodes
[0];
3088 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3090 memzero_extent_buffer(leaf
, (unsigned long)item
, sizeof(*item
));
3092 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
3093 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
3094 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
3095 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
3096 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
3097 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
3099 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
3101 btrfs_mark_buffer_dirty(leaf
);
3103 btrfs_free_path(path
);
3104 err
= btrfs_commit_transaction(trans
);
3110 static int del_balance_item(struct btrfs_fs_info
*fs_info
)
3112 struct btrfs_root
*root
= fs_info
->tree_root
;
3113 struct btrfs_trans_handle
*trans
;
3114 struct btrfs_path
*path
;
3115 struct btrfs_key key
;
3118 path
= btrfs_alloc_path();
3122 trans
= btrfs_start_transaction(root
, 0);
3123 if (IS_ERR(trans
)) {
3124 btrfs_free_path(path
);
3125 return PTR_ERR(trans
);
3128 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3129 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3132 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3140 ret
= btrfs_del_item(trans
, root
, path
);
3142 btrfs_free_path(path
);
3143 err
= btrfs_commit_transaction(trans
);
3150 * This is a heuristic used to reduce the number of chunks balanced on
3151 * resume after balance was interrupted.
3153 static void update_balance_args(struct btrfs_balance_control
*bctl
)
3156 * Turn on soft mode for chunk types that were being converted.
3158 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3159 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3160 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3161 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3162 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3163 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3166 * Turn on usage filter if is not already used. The idea is
3167 * that chunks that we have already balanced should be
3168 * reasonably full. Don't do it for chunks that are being
3169 * converted - that will keep us from relocating unconverted
3170 * (albeit full) chunks.
3172 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3173 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3174 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3175 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3176 bctl
->data
.usage
= 90;
3178 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3179 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3180 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3181 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3182 bctl
->sys
.usage
= 90;
3184 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3185 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3186 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3187 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3188 bctl
->meta
.usage
= 90;
3193 * Should be called with both balance and volume mutexes held to
3194 * serialize other volume operations (add_dev/rm_dev/resize) with
3195 * restriper. Same goes for unset_balance_control.
3197 static void set_balance_control(struct btrfs_balance_control
*bctl
)
3199 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3201 BUG_ON(fs_info
->balance_ctl
);
3203 spin_lock(&fs_info
->balance_lock
);
3204 fs_info
->balance_ctl
= bctl
;
3205 spin_unlock(&fs_info
->balance_lock
);
3208 static void unset_balance_control(struct btrfs_fs_info
*fs_info
)
3210 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3212 BUG_ON(!fs_info
->balance_ctl
);
3214 spin_lock(&fs_info
->balance_lock
);
3215 fs_info
->balance_ctl
= NULL
;
3216 spin_unlock(&fs_info
->balance_lock
);
3222 * Balance filters. Return 1 if chunk should be filtered out
3223 * (should not be balanced).
3225 static int chunk_profiles_filter(u64 chunk_type
,
3226 struct btrfs_balance_args
*bargs
)
3228 chunk_type
= chunk_to_extended(chunk_type
) &
3229 BTRFS_EXTENDED_PROFILE_MASK
;
3231 if (bargs
->profiles
& chunk_type
)
3237 static int chunk_usage_range_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
3238 struct btrfs_balance_args
*bargs
)
3240 struct btrfs_block_group_cache
*cache
;
3242 u64 user_thresh_min
;
3243 u64 user_thresh_max
;
3246 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3247 chunk_used
= btrfs_block_group_used(&cache
->item
);
3249 if (bargs
->usage_min
== 0)
3250 user_thresh_min
= 0;
3252 user_thresh_min
= div_factor_fine(cache
->key
.offset
,
3255 if (bargs
->usage_max
== 0)
3256 user_thresh_max
= 1;
3257 else if (bargs
->usage_max
> 100)
3258 user_thresh_max
= cache
->key
.offset
;
3260 user_thresh_max
= div_factor_fine(cache
->key
.offset
,
3263 if (user_thresh_min
<= chunk_used
&& chunk_used
< user_thresh_max
)
3266 btrfs_put_block_group(cache
);
3270 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
,
3271 u64 chunk_offset
, struct btrfs_balance_args
*bargs
)
3273 struct btrfs_block_group_cache
*cache
;
3274 u64 chunk_used
, user_thresh
;
3277 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3278 chunk_used
= btrfs_block_group_used(&cache
->item
);
3280 if (bargs
->usage_min
== 0)
3282 else if (bargs
->usage
> 100)
3283 user_thresh
= cache
->key
.offset
;
3285 user_thresh
= div_factor_fine(cache
->key
.offset
,
3288 if (chunk_used
< user_thresh
)
3291 btrfs_put_block_group(cache
);
3295 static int chunk_devid_filter(struct extent_buffer
*leaf
,
3296 struct btrfs_chunk
*chunk
,
3297 struct btrfs_balance_args
*bargs
)
3299 struct btrfs_stripe
*stripe
;
3300 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3303 for (i
= 0; i
< num_stripes
; i
++) {
3304 stripe
= btrfs_stripe_nr(chunk
, i
);
3305 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
3312 /* [pstart, pend) */
3313 static int chunk_drange_filter(struct extent_buffer
*leaf
,
3314 struct btrfs_chunk
*chunk
,
3316 struct btrfs_balance_args
*bargs
)
3318 struct btrfs_stripe
*stripe
;
3319 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3325 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
3328 if (btrfs_chunk_type(leaf
, chunk
) & (BTRFS_BLOCK_GROUP_DUP
|
3329 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)) {
3330 factor
= num_stripes
/ 2;
3331 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID5
) {
3332 factor
= num_stripes
- 1;
3333 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID6
) {
3334 factor
= num_stripes
- 2;
3336 factor
= num_stripes
;
3339 for (i
= 0; i
< num_stripes
; i
++) {
3340 stripe
= btrfs_stripe_nr(chunk
, i
);
3341 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
3344 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
3345 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
3346 stripe_length
= div_u64(stripe_length
, factor
);
3348 if (stripe_offset
< bargs
->pend
&&
3349 stripe_offset
+ stripe_length
> bargs
->pstart
)
3356 /* [vstart, vend) */
3357 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
3358 struct btrfs_chunk
*chunk
,
3360 struct btrfs_balance_args
*bargs
)
3362 if (chunk_offset
< bargs
->vend
&&
3363 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
3364 /* at least part of the chunk is inside this vrange */
3370 static int chunk_stripes_range_filter(struct extent_buffer
*leaf
,
3371 struct btrfs_chunk
*chunk
,
3372 struct btrfs_balance_args
*bargs
)
3374 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3376 if (bargs
->stripes_min
<= num_stripes
3377 && num_stripes
<= bargs
->stripes_max
)
3383 static int chunk_soft_convert_filter(u64 chunk_type
,
3384 struct btrfs_balance_args
*bargs
)
3386 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
3389 chunk_type
= chunk_to_extended(chunk_type
) &
3390 BTRFS_EXTENDED_PROFILE_MASK
;
3392 if (bargs
->target
== chunk_type
)
3398 static int should_balance_chunk(struct btrfs_fs_info
*fs_info
,
3399 struct extent_buffer
*leaf
,
3400 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
3402 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3403 struct btrfs_balance_args
*bargs
= NULL
;
3404 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3407 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
3408 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
3412 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3413 bargs
= &bctl
->data
;
3414 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3416 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3417 bargs
= &bctl
->meta
;
3419 /* profiles filter */
3420 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
3421 chunk_profiles_filter(chunk_type
, bargs
)) {
3426 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3427 chunk_usage_filter(fs_info
, chunk_offset
, bargs
)) {
3429 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3430 chunk_usage_range_filter(fs_info
, chunk_offset
, bargs
)) {
3435 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
3436 chunk_devid_filter(leaf
, chunk
, bargs
)) {
3440 /* drange filter, makes sense only with devid filter */
3441 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
3442 chunk_drange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3447 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
3448 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3452 /* stripes filter */
3453 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_STRIPES_RANGE
) &&
3454 chunk_stripes_range_filter(leaf
, chunk
, bargs
)) {
3458 /* soft profile changing mode */
3459 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
3460 chunk_soft_convert_filter(chunk_type
, bargs
)) {
3465 * limited by count, must be the last filter
3467 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT
)) {
3468 if (bargs
->limit
== 0)
3472 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT_RANGE
)) {
3474 * Same logic as the 'limit' filter; the minimum cannot be
3475 * determined here because we do not have the global information
3476 * about the count of all chunks that satisfy the filters.
3478 if (bargs
->limit_max
== 0)
3487 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
3489 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3490 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
3491 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
3492 struct list_head
*devices
;
3493 struct btrfs_device
*device
;
3497 struct btrfs_chunk
*chunk
;
3498 struct btrfs_path
*path
= NULL
;
3499 struct btrfs_key key
;
3500 struct btrfs_key found_key
;
3501 struct btrfs_trans_handle
*trans
;
3502 struct extent_buffer
*leaf
;
3505 int enospc_errors
= 0;
3506 bool counting
= true;
3507 /* The single value limit and min/max limits use the same bytes in the */
3508 u64 limit_data
= bctl
->data
.limit
;
3509 u64 limit_meta
= bctl
->meta
.limit
;
3510 u64 limit_sys
= bctl
->sys
.limit
;
3514 int chunk_reserved
= 0;
3517 /* step one make some room on all the devices */
3518 devices
= &fs_info
->fs_devices
->devices
;
3519 list_for_each_entry(device
, devices
, dev_list
) {
3520 old_size
= btrfs_device_get_total_bytes(device
);
3521 size_to_free
= div_factor(old_size
, 1);
3522 size_to_free
= min_t(u64
, size_to_free
, SZ_1M
);
3523 if (!device
->writeable
||
3524 btrfs_device_get_total_bytes(device
) -
3525 btrfs_device_get_bytes_used(device
) > size_to_free
||
3526 device
->is_tgtdev_for_dev_replace
)
3529 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
3533 /* btrfs_shrink_device never returns ret > 0 */
3538 trans
= btrfs_start_transaction(dev_root
, 0);
3539 if (IS_ERR(trans
)) {
3540 ret
= PTR_ERR(trans
);
3541 btrfs_info_in_rcu(fs_info
,
3542 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3543 rcu_str_deref(device
->name
), ret
,
3544 old_size
, old_size
- size_to_free
);
3548 ret
= btrfs_grow_device(trans
, device
, old_size
);
3550 btrfs_end_transaction(trans
);
3551 /* btrfs_grow_device never returns ret > 0 */
3553 btrfs_info_in_rcu(fs_info
,
3554 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3555 rcu_str_deref(device
->name
), ret
,
3556 old_size
, old_size
- size_to_free
);
3560 btrfs_end_transaction(trans
);
3563 /* step two, relocate all the chunks */
3564 path
= btrfs_alloc_path();
3570 /* zero out stat counters */
3571 spin_lock(&fs_info
->balance_lock
);
3572 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
3573 spin_unlock(&fs_info
->balance_lock
);
3577 * The single value limit and min/max limits use the same bytes
3580 bctl
->data
.limit
= limit_data
;
3581 bctl
->meta
.limit
= limit_meta
;
3582 bctl
->sys
.limit
= limit_sys
;
3584 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3585 key
.offset
= (u64
)-1;
3586 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3589 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
3590 atomic_read(&fs_info
->balance_cancel_req
)) {
3595 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
3596 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
3598 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3603 * this shouldn't happen, it means the last relocate
3607 BUG(); /* FIXME break ? */
3609 ret
= btrfs_previous_item(chunk_root
, path
, 0,
3610 BTRFS_CHUNK_ITEM_KEY
);
3612 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3617 leaf
= path
->nodes
[0];
3618 slot
= path
->slots
[0];
3619 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3621 if (found_key
.objectid
!= key
.objectid
) {
3622 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3626 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
3627 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3630 spin_lock(&fs_info
->balance_lock
);
3631 bctl
->stat
.considered
++;
3632 spin_unlock(&fs_info
->balance_lock
);
3635 ret
= should_balance_chunk(fs_info
, leaf
, chunk
,
3638 btrfs_release_path(path
);
3640 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3645 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3646 spin_lock(&fs_info
->balance_lock
);
3647 bctl
->stat
.expected
++;
3648 spin_unlock(&fs_info
->balance_lock
);
3650 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3652 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3654 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3661 * Apply limit_min filter, no need to check if the LIMITS
3662 * filter is used, limit_min is 0 by default
3664 if (((chunk_type
& BTRFS_BLOCK_GROUP_DATA
) &&
3665 count_data
< bctl
->data
.limit_min
)
3666 || ((chunk_type
& BTRFS_BLOCK_GROUP_METADATA
) &&
3667 count_meta
< bctl
->meta
.limit_min
)
3668 || ((chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) &&
3669 count_sys
< bctl
->sys
.limit_min
)) {
3670 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3674 ASSERT(fs_info
->data_sinfo
);
3675 spin_lock(&fs_info
->data_sinfo
->lock
);
3676 bytes_used
= fs_info
->data_sinfo
->bytes_used
;
3677 spin_unlock(&fs_info
->data_sinfo
->lock
);
3679 if ((chunk_type
& BTRFS_BLOCK_GROUP_DATA
) &&
3680 !chunk_reserved
&& !bytes_used
) {
3681 trans
= btrfs_start_transaction(chunk_root
, 0);
3682 if (IS_ERR(trans
)) {
3683 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3684 ret
= PTR_ERR(trans
);
3688 ret
= btrfs_force_chunk_alloc(trans
, fs_info
,
3689 BTRFS_BLOCK_GROUP_DATA
);
3690 btrfs_end_transaction(trans
);
3692 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3698 ret
= btrfs_relocate_chunk(fs_info
, found_key
.offset
);
3699 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3700 if (ret
&& ret
!= -ENOSPC
)
3702 if (ret
== -ENOSPC
) {
3705 spin_lock(&fs_info
->balance_lock
);
3706 bctl
->stat
.completed
++;
3707 spin_unlock(&fs_info
->balance_lock
);
3710 if (found_key
.offset
== 0)
3712 key
.offset
= found_key
.offset
- 1;
3716 btrfs_release_path(path
);
3721 btrfs_free_path(path
);
3722 if (enospc_errors
) {
3723 btrfs_info(fs_info
, "%d enospc errors during balance",
3733 * alloc_profile_is_valid - see if a given profile is valid and reduced
3734 * @flags: profile to validate
3735 * @extended: if true @flags is treated as an extended profile
3737 static int alloc_profile_is_valid(u64 flags
, int extended
)
3739 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
3740 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
3742 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
3744 /* 1) check that all other bits are zeroed */
3748 /* 2) see if profile is reduced */
3750 return !extended
; /* "0" is valid for usual profiles */
3752 /* true if exactly one bit set */
3753 return (flags
& (flags
- 1)) == 0;
3756 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
3758 /* cancel requested || normal exit path */
3759 return atomic_read(&fs_info
->balance_cancel_req
) ||
3760 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
3761 atomic_read(&fs_info
->balance_cancel_req
) == 0);
3764 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
3768 unset_balance_control(fs_info
);
3769 ret
= del_balance_item(fs_info
);
3771 btrfs_handle_fs_error(fs_info
, ret
, NULL
);
3773 clear_bit(BTRFS_FS_EXCL_OP
, &fs_info
->flags
);
3776 /* Non-zero return value signifies invalidity */
3777 static inline int validate_convert_profile(struct btrfs_balance_args
*bctl_arg
,
3780 return ((bctl_arg
->flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3781 (!alloc_profile_is_valid(bctl_arg
->target
, 1) ||
3782 (bctl_arg
->target
& ~allowed
)));
3786 * Should be called with both balance and volume mutexes held
3788 int btrfs_balance(struct btrfs_balance_control
*bctl
,
3789 struct btrfs_ioctl_balance_args
*bargs
)
3791 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3792 u64 meta_target
, data_target
;
3799 if (btrfs_fs_closing(fs_info
) ||
3800 atomic_read(&fs_info
->balance_pause_req
) ||
3801 atomic_read(&fs_info
->balance_cancel_req
)) {
3806 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
3807 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
3811 * In case of mixed groups both data and meta should be picked,
3812 * and identical options should be given for both of them.
3814 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
3815 if (mixed
&& (bctl
->flags
& allowed
)) {
3816 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
3817 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
3818 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
3820 "with mixed groups data and metadata balance options must be the same");
3826 num_devices
= fs_info
->fs_devices
->num_devices
;
3827 btrfs_dev_replace_lock(&fs_info
->dev_replace
, 0);
3828 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
3829 BUG_ON(num_devices
< 1);
3832 btrfs_dev_replace_unlock(&fs_info
->dev_replace
, 0);
3833 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
| BTRFS_BLOCK_GROUP_DUP
;
3834 if (num_devices
> 1)
3835 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
3836 if (num_devices
> 2)
3837 allowed
|= BTRFS_BLOCK_GROUP_RAID5
;
3838 if (num_devices
> 3)
3839 allowed
|= (BTRFS_BLOCK_GROUP_RAID10
|
3840 BTRFS_BLOCK_GROUP_RAID6
);
3841 if (validate_convert_profile(&bctl
->data
, allowed
)) {
3843 "unable to start balance with target data profile %llu",
3848 if (validate_convert_profile(&bctl
->meta
, allowed
)) {
3850 "unable to start balance with target metadata profile %llu",
3855 if (validate_convert_profile(&bctl
->sys
, allowed
)) {
3857 "unable to start balance with target system profile %llu",
3863 /* allow to reduce meta or sys integrity only if force set */
3864 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3865 BTRFS_BLOCK_GROUP_RAID10
|
3866 BTRFS_BLOCK_GROUP_RAID5
|
3867 BTRFS_BLOCK_GROUP_RAID6
;
3869 seq
= read_seqbegin(&fs_info
->profiles_lock
);
3871 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3872 (fs_info
->avail_system_alloc_bits
& allowed
) &&
3873 !(bctl
->sys
.target
& allowed
)) ||
3874 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3875 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
3876 !(bctl
->meta
.target
& allowed
))) {
3877 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
3879 "force reducing metadata integrity");
3882 "balance will reduce metadata integrity, use force if you want this");
3887 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
3889 /* if we're not converting, the target field is uninitialized */
3890 meta_target
= (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) ?
3891 bctl
->meta
.target
: fs_info
->avail_metadata_alloc_bits
;
3892 data_target
= (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) ?
3893 bctl
->data
.target
: fs_info
->avail_data_alloc_bits
;
3894 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target
) <
3895 btrfs_get_num_tolerated_disk_barrier_failures(data_target
)) {
3897 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3898 meta_target
, data_target
);
3901 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3902 fs_info
->num_tolerated_disk_barrier_failures
= min(
3903 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
),
3904 btrfs_get_num_tolerated_disk_barrier_failures(
3908 ret
= insert_balance_item(fs_info
, bctl
);
3909 if (ret
&& ret
!= -EEXIST
)
3912 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
3913 BUG_ON(ret
== -EEXIST
);
3914 set_balance_control(bctl
);
3916 BUG_ON(ret
!= -EEXIST
);
3917 spin_lock(&fs_info
->balance_lock
);
3918 update_balance_args(bctl
);
3919 spin_unlock(&fs_info
->balance_lock
);
3922 atomic_inc(&fs_info
->balance_running
);
3923 mutex_unlock(&fs_info
->balance_mutex
);
3925 ret
= __btrfs_balance(fs_info
);
3927 mutex_lock(&fs_info
->balance_mutex
);
3928 atomic_dec(&fs_info
->balance_running
);
3930 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3931 fs_info
->num_tolerated_disk_barrier_failures
=
3932 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3936 memset(bargs
, 0, sizeof(*bargs
));
3937 update_ioctl_balance_args(fs_info
, 0, bargs
);
3940 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
3941 balance_need_close(fs_info
)) {
3942 __cancel_balance(fs_info
);
3945 wake_up(&fs_info
->balance_wait_q
);
3949 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
3950 __cancel_balance(fs_info
);
3953 clear_bit(BTRFS_FS_EXCL_OP
, &fs_info
->flags
);
3958 static int balance_kthread(void *data
)
3960 struct btrfs_fs_info
*fs_info
= data
;
3963 mutex_lock(&fs_info
->volume_mutex
);
3964 mutex_lock(&fs_info
->balance_mutex
);
3966 if (fs_info
->balance_ctl
) {
3967 btrfs_info(fs_info
, "continuing balance");
3968 ret
= btrfs_balance(fs_info
->balance_ctl
, NULL
);
3971 mutex_unlock(&fs_info
->balance_mutex
);
3972 mutex_unlock(&fs_info
->volume_mutex
);
3977 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
3979 struct task_struct
*tsk
;
3981 spin_lock(&fs_info
->balance_lock
);
3982 if (!fs_info
->balance_ctl
) {
3983 spin_unlock(&fs_info
->balance_lock
);
3986 spin_unlock(&fs_info
->balance_lock
);
3988 if (btrfs_test_opt(fs_info
, SKIP_BALANCE
)) {
3989 btrfs_info(fs_info
, "force skipping balance");
3993 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
3994 return PTR_ERR_OR_ZERO(tsk
);
3997 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
3999 struct btrfs_balance_control
*bctl
;
4000 struct btrfs_balance_item
*item
;
4001 struct btrfs_disk_balance_args disk_bargs
;
4002 struct btrfs_path
*path
;
4003 struct extent_buffer
*leaf
;
4004 struct btrfs_key key
;
4007 path
= btrfs_alloc_path();
4011 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
4012 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
4015 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
4018 if (ret
> 0) { /* ret = -ENOENT; */
4023 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
4029 leaf
= path
->nodes
[0];
4030 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
4032 bctl
->fs_info
= fs_info
;
4033 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
4034 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
4036 btrfs_balance_data(leaf
, item
, &disk_bargs
);
4037 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
4038 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
4039 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
4040 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
4041 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
4043 WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP
, &fs_info
->flags
));
4045 mutex_lock(&fs_info
->volume_mutex
);
4046 mutex_lock(&fs_info
->balance_mutex
);
4048 set_balance_control(bctl
);
4050 mutex_unlock(&fs_info
->balance_mutex
);
4051 mutex_unlock(&fs_info
->volume_mutex
);
4053 btrfs_free_path(path
);
4057 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
4061 mutex_lock(&fs_info
->balance_mutex
);
4062 if (!fs_info
->balance_ctl
) {
4063 mutex_unlock(&fs_info
->balance_mutex
);
4067 if (atomic_read(&fs_info
->balance_running
)) {
4068 atomic_inc(&fs_info
->balance_pause_req
);
4069 mutex_unlock(&fs_info
->balance_mutex
);
4071 wait_event(fs_info
->balance_wait_q
,
4072 atomic_read(&fs_info
->balance_running
) == 0);
4074 mutex_lock(&fs_info
->balance_mutex
);
4075 /* we are good with balance_ctl ripped off from under us */
4076 BUG_ON(atomic_read(&fs_info
->balance_running
));
4077 atomic_dec(&fs_info
->balance_pause_req
);
4082 mutex_unlock(&fs_info
->balance_mutex
);
4086 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
4088 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
4091 mutex_lock(&fs_info
->balance_mutex
);
4092 if (!fs_info
->balance_ctl
) {
4093 mutex_unlock(&fs_info
->balance_mutex
);
4097 atomic_inc(&fs_info
->balance_cancel_req
);
4099 * if we are running just wait and return, balance item is
4100 * deleted in btrfs_balance in this case
4102 if (atomic_read(&fs_info
->balance_running
)) {
4103 mutex_unlock(&fs_info
->balance_mutex
);
4104 wait_event(fs_info
->balance_wait_q
,
4105 atomic_read(&fs_info
->balance_running
) == 0);
4106 mutex_lock(&fs_info
->balance_mutex
);
4108 /* __cancel_balance needs volume_mutex */
4109 mutex_unlock(&fs_info
->balance_mutex
);
4110 mutex_lock(&fs_info
->volume_mutex
);
4111 mutex_lock(&fs_info
->balance_mutex
);
4113 if (fs_info
->balance_ctl
)
4114 __cancel_balance(fs_info
);
4116 mutex_unlock(&fs_info
->volume_mutex
);
4119 BUG_ON(fs_info
->balance_ctl
|| atomic_read(&fs_info
->balance_running
));
4120 atomic_dec(&fs_info
->balance_cancel_req
);
4121 mutex_unlock(&fs_info
->balance_mutex
);
4125 static int btrfs_uuid_scan_kthread(void *data
)
4127 struct btrfs_fs_info
*fs_info
= data
;
4128 struct btrfs_root
*root
= fs_info
->tree_root
;
4129 struct btrfs_key key
;
4130 struct btrfs_key max_key
;
4131 struct btrfs_path
*path
= NULL
;
4133 struct extent_buffer
*eb
;
4135 struct btrfs_root_item root_item
;
4137 struct btrfs_trans_handle
*trans
= NULL
;
4139 path
= btrfs_alloc_path();
4146 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4149 max_key
.objectid
= (u64
)-1;
4150 max_key
.type
= BTRFS_ROOT_ITEM_KEY
;
4151 max_key
.offset
= (u64
)-1;
4154 ret
= btrfs_search_forward(root
, &key
, path
, 0);
4161 if (key
.type
!= BTRFS_ROOT_ITEM_KEY
||
4162 (key
.objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
4163 key
.objectid
!= BTRFS_FS_TREE_OBJECTID
) ||
4164 key
.objectid
> BTRFS_LAST_FREE_OBJECTID
)
4167 eb
= path
->nodes
[0];
4168 slot
= path
->slots
[0];
4169 item_size
= btrfs_item_size_nr(eb
, slot
);
4170 if (item_size
< sizeof(root_item
))
4173 read_extent_buffer(eb
, &root_item
,
4174 btrfs_item_ptr_offset(eb
, slot
),
4175 (int)sizeof(root_item
));
4176 if (btrfs_root_refs(&root_item
) == 0)
4179 if (!btrfs_is_empty_uuid(root_item
.uuid
) ||
4180 !btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4184 btrfs_release_path(path
);
4186 * 1 - subvol uuid item
4187 * 1 - received_subvol uuid item
4189 trans
= btrfs_start_transaction(fs_info
->uuid_root
, 2);
4190 if (IS_ERR(trans
)) {
4191 ret
= PTR_ERR(trans
);
4199 if (!btrfs_is_empty_uuid(root_item
.uuid
)) {
4200 ret
= btrfs_uuid_tree_add(trans
, fs_info
,
4202 BTRFS_UUID_KEY_SUBVOL
,
4205 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4211 if (!btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4212 ret
= btrfs_uuid_tree_add(trans
, fs_info
,
4213 root_item
.received_uuid
,
4214 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
4217 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4225 ret
= btrfs_end_transaction(trans
);
4231 btrfs_release_path(path
);
4232 if (key
.offset
< (u64
)-1) {
4234 } else if (key
.type
< BTRFS_ROOT_ITEM_KEY
) {
4236 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4237 } else if (key
.objectid
< (u64
)-1) {
4239 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4248 btrfs_free_path(path
);
4249 if (trans
&& !IS_ERR(trans
))
4250 btrfs_end_transaction(trans
);
4252 btrfs_warn(fs_info
, "btrfs_uuid_scan_kthread failed %d", ret
);
4254 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN
, &fs_info
->flags
);
4255 up(&fs_info
->uuid_tree_rescan_sem
);
4260 * Callback for btrfs_uuid_tree_iterate().
4262 * 0 check succeeded, the entry is not outdated.
4263 * < 0 if an error occurred.
4264 * > 0 if the check failed, which means the caller shall remove the entry.
4266 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info
*fs_info
,
4267 u8
*uuid
, u8 type
, u64 subid
)
4269 struct btrfs_key key
;
4271 struct btrfs_root
*subvol_root
;
4273 if (type
!= BTRFS_UUID_KEY_SUBVOL
&&
4274 type
!= BTRFS_UUID_KEY_RECEIVED_SUBVOL
)
4277 key
.objectid
= subid
;
4278 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4279 key
.offset
= (u64
)-1;
4280 subvol_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
4281 if (IS_ERR(subvol_root
)) {
4282 ret
= PTR_ERR(subvol_root
);
4289 case BTRFS_UUID_KEY_SUBVOL
:
4290 if (memcmp(uuid
, subvol_root
->root_item
.uuid
, BTRFS_UUID_SIZE
))
4293 case BTRFS_UUID_KEY_RECEIVED_SUBVOL
:
4294 if (memcmp(uuid
, subvol_root
->root_item
.received_uuid
,
4304 static int btrfs_uuid_rescan_kthread(void *data
)
4306 struct btrfs_fs_info
*fs_info
= (struct btrfs_fs_info
*)data
;
4310 * 1st step is to iterate through the existing UUID tree and
4311 * to delete all entries that contain outdated data.
4312 * 2nd step is to add all missing entries to the UUID tree.
4314 ret
= btrfs_uuid_tree_iterate(fs_info
, btrfs_check_uuid_tree_entry
);
4316 btrfs_warn(fs_info
, "iterating uuid_tree failed %d", ret
);
4317 up(&fs_info
->uuid_tree_rescan_sem
);
4320 return btrfs_uuid_scan_kthread(data
);
4323 int btrfs_create_uuid_tree(struct btrfs_fs_info
*fs_info
)
4325 struct btrfs_trans_handle
*trans
;
4326 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
4327 struct btrfs_root
*uuid_root
;
4328 struct task_struct
*task
;
4335 trans
= btrfs_start_transaction(tree_root
, 2);
4337 return PTR_ERR(trans
);
4339 uuid_root
= btrfs_create_tree(trans
, fs_info
,
4340 BTRFS_UUID_TREE_OBJECTID
);
4341 if (IS_ERR(uuid_root
)) {
4342 ret
= PTR_ERR(uuid_root
);
4343 btrfs_abort_transaction(trans
, ret
);
4344 btrfs_end_transaction(trans
);
4348 fs_info
->uuid_root
= uuid_root
;
4350 ret
= btrfs_commit_transaction(trans
);
4354 down(&fs_info
->uuid_tree_rescan_sem
);
4355 task
= kthread_run(btrfs_uuid_scan_kthread
, fs_info
, "btrfs-uuid");
4357 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4358 btrfs_warn(fs_info
, "failed to start uuid_scan task");
4359 up(&fs_info
->uuid_tree_rescan_sem
);
4360 return PTR_ERR(task
);
4366 int btrfs_check_uuid_tree(struct btrfs_fs_info
*fs_info
)
4368 struct task_struct
*task
;
4370 down(&fs_info
->uuid_tree_rescan_sem
);
4371 task
= kthread_run(btrfs_uuid_rescan_kthread
, fs_info
, "btrfs-uuid");
4373 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4374 btrfs_warn(fs_info
, "failed to start uuid_rescan task");
4375 up(&fs_info
->uuid_tree_rescan_sem
);
4376 return PTR_ERR(task
);
4383 * shrinking a device means finding all of the device extents past
4384 * the new size, and then following the back refs to the chunks.
4385 * The chunk relocation code actually frees the device extent
4387 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
4389 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
4390 struct btrfs_root
*root
= fs_info
->dev_root
;
4391 struct btrfs_trans_handle
*trans
;
4392 struct btrfs_dev_extent
*dev_extent
= NULL
;
4393 struct btrfs_path
*path
;
4399 bool retried
= false;
4400 bool checked_pending_chunks
= false;
4401 struct extent_buffer
*l
;
4402 struct btrfs_key key
;
4403 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
4404 u64 old_total
= btrfs_super_total_bytes(super_copy
);
4405 u64 old_size
= btrfs_device_get_total_bytes(device
);
4408 new_size
= round_down(new_size
, fs_info
->sectorsize
);
4409 diff
= old_size
- new_size
;
4411 if (device
->is_tgtdev_for_dev_replace
)
4414 path
= btrfs_alloc_path();
4418 path
->reada
= READA_FORWARD
;
4420 mutex_lock(&fs_info
->chunk_mutex
);
4422 btrfs_device_set_total_bytes(device
, new_size
);
4423 if (device
->writeable
) {
4424 device
->fs_devices
->total_rw_bytes
-= diff
;
4425 atomic64_sub(diff
, &fs_info
->free_chunk_space
);
4427 mutex_unlock(&fs_info
->chunk_mutex
);
4430 key
.objectid
= device
->devid
;
4431 key
.offset
= (u64
)-1;
4432 key
.type
= BTRFS_DEV_EXTENT_KEY
;
4435 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
4436 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4438 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4442 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
4444 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4449 btrfs_release_path(path
);
4454 slot
= path
->slots
[0];
4455 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
4457 if (key
.objectid
!= device
->devid
) {
4458 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4459 btrfs_release_path(path
);
4463 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
4464 length
= btrfs_dev_extent_length(l
, dev_extent
);
4466 if (key
.offset
+ length
<= new_size
) {
4467 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4468 btrfs_release_path(path
);
4472 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
4473 btrfs_release_path(path
);
4475 ret
= btrfs_relocate_chunk(fs_info
, chunk_offset
);
4476 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4477 if (ret
&& ret
!= -ENOSPC
)
4481 } while (key
.offset
-- > 0);
4483 if (failed
&& !retried
) {
4487 } else if (failed
&& retried
) {
4492 /* Shrinking succeeded, else we would be at "done". */
4493 trans
= btrfs_start_transaction(root
, 0);
4494 if (IS_ERR(trans
)) {
4495 ret
= PTR_ERR(trans
);
4499 mutex_lock(&fs_info
->chunk_mutex
);
4502 * We checked in the above loop all device extents that were already in
4503 * the device tree. However before we have updated the device's
4504 * total_bytes to the new size, we might have had chunk allocations that
4505 * have not complete yet (new block groups attached to transaction
4506 * handles), and therefore their device extents were not yet in the
4507 * device tree and we missed them in the loop above. So if we have any
4508 * pending chunk using a device extent that overlaps the device range
4509 * that we can not use anymore, commit the current transaction and
4510 * repeat the search on the device tree - this way we guarantee we will
4511 * not have chunks using device extents that end beyond 'new_size'.
4513 if (!checked_pending_chunks
) {
4514 u64 start
= new_size
;
4515 u64 len
= old_size
- new_size
;
4517 if (contains_pending_extent(trans
->transaction
, device
,
4519 mutex_unlock(&fs_info
->chunk_mutex
);
4520 checked_pending_chunks
= true;
4523 ret
= btrfs_commit_transaction(trans
);
4530 btrfs_device_set_disk_total_bytes(device
, new_size
);
4531 if (list_empty(&device
->resized_list
))
4532 list_add_tail(&device
->resized_list
,
4533 &fs_info
->fs_devices
->resized_devices
);
4535 WARN_ON(diff
> old_total
);
4536 btrfs_set_super_total_bytes(super_copy
,
4537 round_down(old_total
- diff
, fs_info
->sectorsize
));
4538 mutex_unlock(&fs_info
->chunk_mutex
);
4540 /* Now btrfs_update_device() will change the on-disk size. */
4541 ret
= btrfs_update_device(trans
, device
);
4542 btrfs_end_transaction(trans
);
4544 btrfs_free_path(path
);
4546 mutex_lock(&fs_info
->chunk_mutex
);
4547 btrfs_device_set_total_bytes(device
, old_size
);
4548 if (device
->writeable
)
4549 device
->fs_devices
->total_rw_bytes
+= diff
;
4550 atomic64_add(diff
, &fs_info
->free_chunk_space
);
4551 mutex_unlock(&fs_info
->chunk_mutex
);
4556 static int btrfs_add_system_chunk(struct btrfs_fs_info
*fs_info
,
4557 struct btrfs_key
*key
,
4558 struct btrfs_chunk
*chunk
, int item_size
)
4560 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
4561 struct btrfs_disk_key disk_key
;
4565 mutex_lock(&fs_info
->chunk_mutex
);
4566 array_size
= btrfs_super_sys_array_size(super_copy
);
4567 if (array_size
+ item_size
+ sizeof(disk_key
)
4568 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4569 mutex_unlock(&fs_info
->chunk_mutex
);
4573 ptr
= super_copy
->sys_chunk_array
+ array_size
;
4574 btrfs_cpu_key_to_disk(&disk_key
, key
);
4575 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
4576 ptr
+= sizeof(disk_key
);
4577 memcpy(ptr
, chunk
, item_size
);
4578 item_size
+= sizeof(disk_key
);
4579 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
4580 mutex_unlock(&fs_info
->chunk_mutex
);
4586 * sort the devices in descending order by max_avail, total_avail
4588 static int btrfs_cmp_device_info(const void *a
, const void *b
)
4590 const struct btrfs_device_info
*di_a
= a
;
4591 const struct btrfs_device_info
*di_b
= b
;
4593 if (di_a
->max_avail
> di_b
->max_avail
)
4595 if (di_a
->max_avail
< di_b
->max_avail
)
4597 if (di_a
->total_avail
> di_b
->total_avail
)
4599 if (di_a
->total_avail
< di_b
->total_avail
)
4604 static u32
find_raid56_stripe_len(u32 data_devices
, u32 dev_stripe_target
)
4606 /* TODO allow them to set a preferred stripe size */
4610 static void check_raid56_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
4612 if (!(type
& BTRFS_BLOCK_GROUP_RAID56_MASK
))
4615 btrfs_set_fs_incompat(info
, RAID56
);
4618 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info) \
4619 - sizeof(struct btrfs_chunk)) \
4620 / sizeof(struct btrfs_stripe) + 1)
4622 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4623 - 2 * sizeof(struct btrfs_disk_key) \
4624 - 2 * sizeof(struct btrfs_chunk)) \
4625 / sizeof(struct btrfs_stripe) + 1)
4627 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4628 u64 start
, u64 type
)
4630 struct btrfs_fs_info
*info
= trans
->fs_info
;
4631 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
4632 struct list_head
*cur
;
4633 struct map_lookup
*map
= NULL
;
4634 struct extent_map_tree
*em_tree
;
4635 struct extent_map
*em
;
4636 struct btrfs_device_info
*devices_info
= NULL
;
4638 int num_stripes
; /* total number of stripes to allocate */
4639 int data_stripes
; /* number of stripes that count for
4641 int sub_stripes
; /* sub_stripes info for map */
4642 int dev_stripes
; /* stripes per dev */
4643 int devs_max
; /* max devs to use */
4644 int devs_min
; /* min devs needed */
4645 int devs_increment
; /* ndevs has to be a multiple of this */
4646 int ncopies
; /* how many copies to data has */
4648 u64 max_stripe_size
;
4652 u64 raid_stripe_len
= BTRFS_STRIPE_LEN
;
4658 BUG_ON(!alloc_profile_is_valid(type
, 0));
4660 if (list_empty(&fs_devices
->alloc_list
))
4663 index
= __get_raid_index(type
);
4665 sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
4666 dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
4667 devs_max
= btrfs_raid_array
[index
].devs_max
;
4668 devs_min
= btrfs_raid_array
[index
].devs_min
;
4669 devs_increment
= btrfs_raid_array
[index
].devs_increment
;
4670 ncopies
= btrfs_raid_array
[index
].ncopies
;
4672 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
4673 max_stripe_size
= SZ_1G
;
4674 max_chunk_size
= 10 * max_stripe_size
;
4676 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4677 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
4678 /* for larger filesystems, use larger metadata chunks */
4679 if (fs_devices
->total_rw_bytes
> 50ULL * SZ_1G
)
4680 max_stripe_size
= SZ_1G
;
4682 max_stripe_size
= SZ_256M
;
4683 max_chunk_size
= max_stripe_size
;
4685 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4686 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4687 max_stripe_size
= SZ_32M
;
4688 max_chunk_size
= 2 * max_stripe_size
;
4690 devs_max
= BTRFS_MAX_DEVS_SYS_CHUNK
;
4692 btrfs_err(info
, "invalid chunk type 0x%llx requested",
4697 /* we don't want a chunk larger than 10% of writeable space */
4698 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
4701 devices_info
= kcalloc(fs_devices
->rw_devices
, sizeof(*devices_info
),
4706 cur
= fs_devices
->alloc_list
.next
;
4709 * in the first pass through the devices list, we gather information
4710 * about the available holes on each device.
4713 while (cur
!= &fs_devices
->alloc_list
) {
4714 struct btrfs_device
*device
;
4718 device
= list_entry(cur
, struct btrfs_device
, dev_alloc_list
);
4722 if (!device
->writeable
) {
4724 "BTRFS: read-only device in alloc_list\n");
4728 if (!device
->in_fs_metadata
||
4729 device
->is_tgtdev_for_dev_replace
)
4732 if (device
->total_bytes
> device
->bytes_used
)
4733 total_avail
= device
->total_bytes
- device
->bytes_used
;
4737 /* If there is no space on this device, skip it. */
4738 if (total_avail
== 0)
4741 ret
= find_free_dev_extent(trans
, device
,
4742 max_stripe_size
* dev_stripes
,
4743 &dev_offset
, &max_avail
);
4744 if (ret
&& ret
!= -ENOSPC
)
4748 max_avail
= max_stripe_size
* dev_stripes
;
4750 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
4753 if (ndevs
== fs_devices
->rw_devices
) {
4754 WARN(1, "%s: found more than %llu devices\n",
4755 __func__
, fs_devices
->rw_devices
);
4758 devices_info
[ndevs
].dev_offset
= dev_offset
;
4759 devices_info
[ndevs
].max_avail
= max_avail
;
4760 devices_info
[ndevs
].total_avail
= total_avail
;
4761 devices_info
[ndevs
].dev
= device
;
4766 * now sort the devices by hole size / available space
4768 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
4769 btrfs_cmp_device_info
, NULL
);
4771 /* round down to number of usable stripes */
4772 ndevs
-= ndevs
% devs_increment
;
4774 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
4779 if (devs_max
&& ndevs
> devs_max
)
4782 * the primary goal is to maximize the number of stripes, so use as many
4783 * devices as possible, even if the stripes are not maximum sized.
4785 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4786 num_stripes
= ndevs
* dev_stripes
;
4789 * this will have to be fixed for RAID1 and RAID10 over
4792 data_stripes
= num_stripes
/ ncopies
;
4794 if (type
& BTRFS_BLOCK_GROUP_RAID5
) {
4795 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 1,
4797 data_stripes
= num_stripes
- 1;
4799 if (type
& BTRFS_BLOCK_GROUP_RAID6
) {
4800 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 2,
4802 data_stripes
= num_stripes
- 2;
4806 * Use the number of data stripes to figure out how big this chunk
4807 * is really going to be in terms of logical address space,
4808 * and compare that answer with the max chunk size
4810 if (stripe_size
* data_stripes
> max_chunk_size
) {
4811 u64 mask
= (1ULL << 24) - 1;
4813 stripe_size
= div_u64(max_chunk_size
, data_stripes
);
4815 /* bump the answer up to a 16MB boundary */
4816 stripe_size
= (stripe_size
+ mask
) & ~mask
;
4818 /* but don't go higher than the limits we found
4819 * while searching for free extents
4821 if (stripe_size
> devices_info
[ndevs
-1].max_avail
)
4822 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4825 stripe_size
= div_u64(stripe_size
, dev_stripes
);
4827 /* align to BTRFS_STRIPE_LEN */
4828 stripe_size
= div64_u64(stripe_size
, raid_stripe_len
);
4829 stripe_size
*= raid_stripe_len
;
4831 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
4836 map
->num_stripes
= num_stripes
;
4838 for (i
= 0; i
< ndevs
; ++i
) {
4839 for (j
= 0; j
< dev_stripes
; ++j
) {
4840 int s
= i
* dev_stripes
+ j
;
4841 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
4842 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
4846 map
->sector_size
= info
->sectorsize
;
4847 map
->stripe_len
= raid_stripe_len
;
4848 map
->io_align
= raid_stripe_len
;
4849 map
->io_width
= raid_stripe_len
;
4851 map
->sub_stripes
= sub_stripes
;
4853 num_bytes
= stripe_size
* data_stripes
;
4855 trace_btrfs_chunk_alloc(info
, map
, start
, num_bytes
);
4857 em
= alloc_extent_map();
4863 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
4864 em
->map_lookup
= map
;
4866 em
->len
= num_bytes
;
4867 em
->block_start
= 0;
4868 em
->block_len
= em
->len
;
4869 em
->orig_block_len
= stripe_size
;
4871 em_tree
= &info
->mapping_tree
.map_tree
;
4872 write_lock(&em_tree
->lock
);
4873 ret
= add_extent_mapping(em_tree
, em
, 0);
4875 list_add_tail(&em
->list
, &trans
->transaction
->pending_chunks
);
4876 refcount_inc(&em
->refs
);
4878 write_unlock(&em_tree
->lock
);
4880 free_extent_map(em
);
4884 ret
= btrfs_make_block_group(trans
, info
, 0, type
,
4885 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4888 goto error_del_extent
;
4890 for (i
= 0; i
< map
->num_stripes
; i
++) {
4891 num_bytes
= map
->stripes
[i
].dev
->bytes_used
+ stripe_size
;
4892 btrfs_device_set_bytes_used(map
->stripes
[i
].dev
, num_bytes
);
4895 atomic64_sub(stripe_size
* map
->num_stripes
, &info
->free_chunk_space
);
4897 free_extent_map(em
);
4898 check_raid56_incompat_flag(info
, type
);
4900 kfree(devices_info
);
4904 write_lock(&em_tree
->lock
);
4905 remove_extent_mapping(em_tree
, em
);
4906 write_unlock(&em_tree
->lock
);
4908 /* One for our allocation */
4909 free_extent_map(em
);
4910 /* One for the tree reference */
4911 free_extent_map(em
);
4912 /* One for the pending_chunks list reference */
4913 free_extent_map(em
);
4915 kfree(devices_info
);
4919 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
4920 struct btrfs_fs_info
*fs_info
,
4921 u64 chunk_offset
, u64 chunk_size
)
4923 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
4924 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
4925 struct btrfs_key key
;
4926 struct btrfs_device
*device
;
4927 struct btrfs_chunk
*chunk
;
4928 struct btrfs_stripe
*stripe
;
4929 struct extent_map
*em
;
4930 struct map_lookup
*map
;
4937 em
= get_chunk_map(fs_info
, chunk_offset
, chunk_size
);
4941 map
= em
->map_lookup
;
4942 item_size
= btrfs_chunk_item_size(map
->num_stripes
);
4943 stripe_size
= em
->orig_block_len
;
4945 chunk
= kzalloc(item_size
, GFP_NOFS
);
4952 * Take the device list mutex to prevent races with the final phase of
4953 * a device replace operation that replaces the device object associated
4954 * with the map's stripes, because the device object's id can change
4955 * at any time during that final phase of the device replace operation
4956 * (dev-replace.c:btrfs_dev_replace_finishing()).
4958 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
4959 for (i
= 0; i
< map
->num_stripes
; i
++) {
4960 device
= map
->stripes
[i
].dev
;
4961 dev_offset
= map
->stripes
[i
].physical
;
4963 ret
= btrfs_update_device(trans
, device
);
4966 ret
= btrfs_alloc_dev_extent(trans
, device
,
4967 chunk_root
->root_key
.objectid
,
4968 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4969 chunk_offset
, dev_offset
,
4975 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
4979 stripe
= &chunk
->stripe
;
4980 for (i
= 0; i
< map
->num_stripes
; i
++) {
4981 device
= map
->stripes
[i
].dev
;
4982 dev_offset
= map
->stripes
[i
].physical
;
4984 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
4985 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
4986 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
4989 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
4991 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
4992 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
4993 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
4994 btrfs_set_stack_chunk_type(chunk
, map
->type
);
4995 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
4996 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
4997 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
4998 btrfs_set_stack_chunk_sector_size(chunk
, fs_info
->sectorsize
);
4999 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
5001 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
5002 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
5003 key
.offset
= chunk_offset
;
5005 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
5006 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
5008 * TODO: Cleanup of inserted chunk root in case of
5011 ret
= btrfs_add_system_chunk(fs_info
, &key
, chunk
, item_size
);
5016 free_extent_map(em
);
5021 * Chunk allocation falls into two parts. The first part does works
5022 * that make the new allocated chunk useable, but not do any operation
5023 * that modifies the chunk tree. The second part does the works that
5024 * require modifying the chunk tree. This division is important for the
5025 * bootstrap process of adding storage to a seed btrfs.
5027 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
5028 struct btrfs_fs_info
*fs_info
, u64 type
)
5032 ASSERT(mutex_is_locked(&fs_info
->chunk_mutex
));
5033 chunk_offset
= find_next_chunk(fs_info
);
5034 return __btrfs_alloc_chunk(trans
, chunk_offset
, type
);
5037 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
5038 struct btrfs_fs_info
*fs_info
)
5041 u64 sys_chunk_offset
;
5045 chunk_offset
= find_next_chunk(fs_info
);
5046 alloc_profile
= btrfs_metadata_alloc_profile(fs_info
);
5047 ret
= __btrfs_alloc_chunk(trans
, chunk_offset
, alloc_profile
);
5051 sys_chunk_offset
= find_next_chunk(fs_info
);
5052 alloc_profile
= btrfs_system_alloc_profile(fs_info
);
5053 ret
= __btrfs_alloc_chunk(trans
, sys_chunk_offset
, alloc_profile
);
5057 static inline int btrfs_chunk_max_errors(struct map_lookup
*map
)
5061 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
5062 BTRFS_BLOCK_GROUP_RAID10
|
5063 BTRFS_BLOCK_GROUP_RAID5
|
5064 BTRFS_BLOCK_GROUP_DUP
)) {
5066 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
) {
5075 int btrfs_chunk_readonly(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
5077 struct extent_map
*em
;
5078 struct map_lookup
*map
;
5083 em
= get_chunk_map(fs_info
, chunk_offset
, 1);
5087 map
= em
->map_lookup
;
5088 for (i
= 0; i
< map
->num_stripes
; i
++) {
5089 if (map
->stripes
[i
].dev
->missing
) {
5094 if (!map
->stripes
[i
].dev
->writeable
) {
5101 * If the number of missing devices is larger than max errors,
5102 * we can not write the data into that chunk successfully, so
5105 if (miss_ndevs
> btrfs_chunk_max_errors(map
))
5108 free_extent_map(em
);
5112 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
5114 extent_map_tree_init(&tree
->map_tree
);
5117 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
5119 struct extent_map
*em
;
5122 write_lock(&tree
->map_tree
.lock
);
5123 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
5125 remove_extent_mapping(&tree
->map_tree
, em
);
5126 write_unlock(&tree
->map_tree
.lock
);
5130 free_extent_map(em
);
5131 /* once for the tree */
5132 free_extent_map(em
);
5136 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
5138 struct extent_map
*em
;
5139 struct map_lookup
*map
;
5142 em
= get_chunk_map(fs_info
, logical
, len
);
5145 * We could return errors for these cases, but that could get
5146 * ugly and we'd probably do the same thing which is just not do
5147 * anything else and exit, so return 1 so the callers don't try
5148 * to use other copies.
5152 map
= em
->map_lookup
;
5153 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
5154 ret
= map
->num_stripes
;
5155 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5156 ret
= map
->sub_stripes
;
5157 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID5
)
5159 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5163 free_extent_map(em
);
5165 btrfs_dev_replace_lock(&fs_info
->dev_replace
, 0);
5166 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
) &&
5167 fs_info
->dev_replace
.tgtdev
)
5169 btrfs_dev_replace_unlock(&fs_info
->dev_replace
, 0);
5174 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info
*fs_info
,
5175 struct btrfs_mapping_tree
*map_tree
,
5178 struct extent_map
*em
;
5179 struct map_lookup
*map
;
5180 unsigned long len
= fs_info
->sectorsize
;
5182 em
= get_chunk_map(fs_info
, logical
, len
);
5183 WARN_ON(IS_ERR(em
));
5185 map
= em
->map_lookup
;
5186 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5187 len
= map
->stripe_len
* nr_data_stripes(map
);
5188 free_extent_map(em
);
5192 int btrfs_is_parity_mirror(struct btrfs_fs_info
*fs_info
,
5193 u64 logical
, u64 len
, int mirror_num
)
5195 struct extent_map
*em
;
5196 struct map_lookup
*map
;
5199 em
= get_chunk_map(fs_info
, logical
, len
);
5200 WARN_ON(IS_ERR(em
));
5202 map
= em
->map_lookup
;
5203 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5205 free_extent_map(em
);
5209 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
5210 struct map_lookup
*map
, int first
, int num
,
5211 int optimal
, int dev_replace_is_ongoing
)
5215 struct btrfs_device
*srcdev
;
5217 if (dev_replace_is_ongoing
&&
5218 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
5219 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
5220 srcdev
= fs_info
->dev_replace
.srcdev
;
5225 * try to avoid the drive that is the source drive for a
5226 * dev-replace procedure, only choose it if no other non-missing
5227 * mirror is available
5229 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
5230 if (map
->stripes
[optimal
].dev
->bdev
&&
5231 (tolerance
|| map
->stripes
[optimal
].dev
!= srcdev
))
5233 for (i
= first
; i
< first
+ num
; i
++) {
5234 if (map
->stripes
[i
].dev
->bdev
&&
5235 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
5240 /* we couldn't find one that doesn't fail. Just return something
5241 * and the io error handling code will clean up eventually
5246 static inline int parity_smaller(u64 a
, u64 b
)
5251 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5252 static void sort_parity_stripes(struct btrfs_bio
*bbio
, int num_stripes
)
5254 struct btrfs_bio_stripe s
;
5261 for (i
= 0; i
< num_stripes
- 1; i
++) {
5262 if (parity_smaller(bbio
->raid_map
[i
],
5263 bbio
->raid_map
[i
+1])) {
5264 s
= bbio
->stripes
[i
];
5265 l
= bbio
->raid_map
[i
];
5266 bbio
->stripes
[i
] = bbio
->stripes
[i
+1];
5267 bbio
->raid_map
[i
] = bbio
->raid_map
[i
+1];
5268 bbio
->stripes
[i
+1] = s
;
5269 bbio
->raid_map
[i
+1] = l
;
5277 static struct btrfs_bio
*alloc_btrfs_bio(int total_stripes
, int real_stripes
)
5279 struct btrfs_bio
*bbio
= kzalloc(
5280 /* the size of the btrfs_bio */
5281 sizeof(struct btrfs_bio
) +
5282 /* plus the variable array for the stripes */
5283 sizeof(struct btrfs_bio_stripe
) * (total_stripes
) +
5284 /* plus the variable array for the tgt dev */
5285 sizeof(int) * (real_stripes
) +
5287 * plus the raid_map, which includes both the tgt dev
5290 sizeof(u64
) * (total_stripes
),
5291 GFP_NOFS
|__GFP_NOFAIL
);
5293 atomic_set(&bbio
->error
, 0);
5294 refcount_set(&bbio
->refs
, 1);
5299 void btrfs_get_bbio(struct btrfs_bio
*bbio
)
5301 WARN_ON(!refcount_read(&bbio
->refs
));
5302 refcount_inc(&bbio
->refs
);
5305 void btrfs_put_bbio(struct btrfs_bio
*bbio
)
5309 if (refcount_dec_and_test(&bbio
->refs
))
5313 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5315 * Please note that, discard won't be sent to target device of device
5318 static int __btrfs_map_block_for_discard(struct btrfs_fs_info
*fs_info
,
5319 u64 logical
, u64 length
,
5320 struct btrfs_bio
**bbio_ret
)
5322 struct extent_map
*em
;
5323 struct map_lookup
*map
;
5324 struct btrfs_bio
*bbio
;
5328 u64 stripe_end_offset
;
5335 u32 sub_stripes
= 0;
5336 u64 stripes_per_dev
= 0;
5337 u32 remaining_stripes
= 0;
5338 u32 last_stripe
= 0;
5342 /* discard always return a bbio */
5345 em
= get_chunk_map(fs_info
, logical
, length
);
5349 map
= em
->map_lookup
;
5350 /* we don't discard raid56 yet */
5351 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5356 offset
= logical
- em
->start
;
5357 length
= min_t(u64
, em
->len
- offset
, length
);
5359 stripe_len
= map
->stripe_len
;
5361 * stripe_nr counts the total number of stripes we have to stride
5362 * to get to this block
5364 stripe_nr
= div64_u64(offset
, stripe_len
);
5366 /* stripe_offset is the offset of this block in its stripe */
5367 stripe_offset
= offset
- stripe_nr
* stripe_len
;
5369 stripe_nr_end
= round_up(offset
+ length
, map
->stripe_len
);
5370 stripe_nr_end
= div64_u64(stripe_nr_end
, map
->stripe_len
);
5371 stripe_cnt
= stripe_nr_end
- stripe_nr
;
5372 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
5375 * after this, stripe_nr is the number of stripes on this
5376 * device we have to walk to find the data, and stripe_index is
5377 * the number of our device in the stripe array
5381 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5382 BTRFS_BLOCK_GROUP_RAID10
)) {
5383 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5386 sub_stripes
= map
->sub_stripes
;
5388 factor
= map
->num_stripes
/ sub_stripes
;
5389 num_stripes
= min_t(u64
, map
->num_stripes
,
5390 sub_stripes
* stripe_cnt
);
5391 stripe_nr
= div_u64_rem(stripe_nr
, factor
, &stripe_index
);
5392 stripe_index
*= sub_stripes
;
5393 stripes_per_dev
= div_u64_rem(stripe_cnt
, factor
,
5394 &remaining_stripes
);
5395 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
5396 last_stripe
*= sub_stripes
;
5397 } else if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
5398 BTRFS_BLOCK_GROUP_DUP
)) {
5399 num_stripes
= map
->num_stripes
;
5401 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5405 bbio
= alloc_btrfs_bio(num_stripes
, 0);
5411 for (i
= 0; i
< num_stripes
; i
++) {
5412 bbio
->stripes
[i
].physical
=
5413 map
->stripes
[stripe_index
].physical
+
5414 stripe_offset
+ stripe_nr
* map
->stripe_len
;
5415 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
5417 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5418 BTRFS_BLOCK_GROUP_RAID10
)) {
5419 bbio
->stripes
[i
].length
= stripes_per_dev
*
5422 if (i
/ sub_stripes
< remaining_stripes
)
5423 bbio
->stripes
[i
].length
+=
5427 * Special for the first stripe and
5430 * |-------|...|-------|
5434 if (i
< sub_stripes
)
5435 bbio
->stripes
[i
].length
-=
5438 if (stripe_index
>= last_stripe
&&
5439 stripe_index
<= (last_stripe
+
5441 bbio
->stripes
[i
].length
-=
5444 if (i
== sub_stripes
- 1)
5447 bbio
->stripes
[i
].length
= length
;
5451 if (stripe_index
== map
->num_stripes
) {
5458 bbio
->map_type
= map
->type
;
5459 bbio
->num_stripes
= num_stripes
;
5461 free_extent_map(em
);
5466 * In dev-replace case, for repair case (that's the only case where the mirror
5467 * is selected explicitly when calling btrfs_map_block), blocks left of the
5468 * left cursor can also be read from the target drive.
5470 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5472 * For READ, it also needs to be supported using the same mirror number.
5474 * If the requested block is not left of the left cursor, EIO is returned. This
5475 * can happen because btrfs_num_copies() returns one more in the dev-replace
5478 static int get_extra_mirror_from_replace(struct btrfs_fs_info
*fs_info
,
5479 u64 logical
, u64 length
,
5480 u64 srcdev_devid
, int *mirror_num
,
5483 struct btrfs_bio
*bbio
= NULL
;
5485 int index_srcdev
= 0;
5487 u64 physical_of_found
= 0;
5491 ret
= __btrfs_map_block(fs_info
, BTRFS_MAP_GET_READ_MIRRORS
,
5492 logical
, &length
, &bbio
, 0, 0);
5494 ASSERT(bbio
== NULL
);
5498 num_stripes
= bbio
->num_stripes
;
5499 if (*mirror_num
> num_stripes
) {
5501 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5502 * that means that the requested area is not left of the left
5505 btrfs_put_bbio(bbio
);
5510 * process the rest of the function using the mirror_num of the source
5511 * drive. Therefore look it up first. At the end, patch the device
5512 * pointer to the one of the target drive.
5514 for (i
= 0; i
< num_stripes
; i
++) {
5515 if (bbio
->stripes
[i
].dev
->devid
!= srcdev_devid
)
5519 * In case of DUP, in order to keep it simple, only add the
5520 * mirror with the lowest physical address
5523 physical_of_found
<= bbio
->stripes
[i
].physical
)
5528 physical_of_found
= bbio
->stripes
[i
].physical
;
5531 btrfs_put_bbio(bbio
);
5537 *mirror_num
= index_srcdev
+ 1;
5538 *physical
= physical_of_found
;
5542 static void handle_ops_on_dev_replace(enum btrfs_map_op op
,
5543 struct btrfs_bio
**bbio_ret
,
5544 struct btrfs_dev_replace
*dev_replace
,
5545 int *num_stripes_ret
, int *max_errors_ret
)
5547 struct btrfs_bio
*bbio
= *bbio_ret
;
5548 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5549 int tgtdev_indexes
= 0;
5550 int num_stripes
= *num_stripes_ret
;
5551 int max_errors
= *max_errors_ret
;
5554 if (op
== BTRFS_MAP_WRITE
) {
5555 int index_where_to_add
;
5558 * duplicate the write operations while the dev replace
5559 * procedure is running. Since the copying of the old disk to
5560 * the new disk takes place at run time while the filesystem is
5561 * mounted writable, the regular write operations to the old
5562 * disk have to be duplicated to go to the new disk as well.
5564 * Note that device->missing is handled by the caller, and that
5565 * the write to the old disk is already set up in the stripes
5568 index_where_to_add
= num_stripes
;
5569 for (i
= 0; i
< num_stripes
; i
++) {
5570 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5571 /* write to new disk, too */
5572 struct btrfs_bio_stripe
*new =
5573 bbio
->stripes
+ index_where_to_add
;
5574 struct btrfs_bio_stripe
*old
=
5577 new->physical
= old
->physical
;
5578 new->length
= old
->length
;
5579 new->dev
= dev_replace
->tgtdev
;
5580 bbio
->tgtdev_map
[i
] = index_where_to_add
;
5581 index_where_to_add
++;
5586 num_stripes
= index_where_to_add
;
5587 } else if (op
== BTRFS_MAP_GET_READ_MIRRORS
) {
5588 int index_srcdev
= 0;
5590 u64 physical_of_found
= 0;
5593 * During the dev-replace procedure, the target drive can also
5594 * be used to read data in case it is needed to repair a corrupt
5595 * block elsewhere. This is possible if the requested area is
5596 * left of the left cursor. In this area, the target drive is a
5597 * full copy of the source drive.
5599 for (i
= 0; i
< num_stripes
; i
++) {
5600 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5602 * In case of DUP, in order to keep it simple,
5603 * only add the mirror with the lowest physical
5607 physical_of_found
<=
5608 bbio
->stripes
[i
].physical
)
5612 physical_of_found
= bbio
->stripes
[i
].physical
;
5616 struct btrfs_bio_stripe
*tgtdev_stripe
=
5617 bbio
->stripes
+ num_stripes
;
5619 tgtdev_stripe
->physical
= physical_of_found
;
5620 tgtdev_stripe
->length
=
5621 bbio
->stripes
[index_srcdev
].length
;
5622 tgtdev_stripe
->dev
= dev_replace
->tgtdev
;
5623 bbio
->tgtdev_map
[index_srcdev
] = num_stripes
;
5630 *num_stripes_ret
= num_stripes
;
5631 *max_errors_ret
= max_errors
;
5632 bbio
->num_tgtdevs
= tgtdev_indexes
;
5636 static bool need_full_stripe(enum btrfs_map_op op
)
5638 return (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
);
5641 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
,
5642 enum btrfs_map_op op
,
5643 u64 logical
, u64
*length
,
5644 struct btrfs_bio
**bbio_ret
,
5645 int mirror_num
, int need_raid_map
)
5647 struct extent_map
*em
;
5648 struct map_lookup
*map
;
5658 int tgtdev_indexes
= 0;
5659 struct btrfs_bio
*bbio
= NULL
;
5660 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
5661 int dev_replace_is_ongoing
= 0;
5662 int num_alloc_stripes
;
5663 int patch_the_first_stripe_for_dev_replace
= 0;
5664 u64 physical_to_patch_in_first_stripe
= 0;
5665 u64 raid56_full_stripe_start
= (u64
)-1;
5667 if (op
== BTRFS_MAP_DISCARD
)
5668 return __btrfs_map_block_for_discard(fs_info
, logical
,
5671 em
= get_chunk_map(fs_info
, logical
, *length
);
5675 map
= em
->map_lookup
;
5676 offset
= logical
- em
->start
;
5678 stripe_len
= map
->stripe_len
;
5681 * stripe_nr counts the total number of stripes we have to stride
5682 * to get to this block
5684 stripe_nr
= div64_u64(stripe_nr
, stripe_len
);
5686 stripe_offset
= stripe_nr
* stripe_len
;
5687 if (offset
< stripe_offset
) {
5689 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5690 stripe_offset
, offset
, em
->start
, logical
,
5692 free_extent_map(em
);
5696 /* stripe_offset is the offset of this block in its stripe*/
5697 stripe_offset
= offset
- stripe_offset
;
5699 /* if we're here for raid56, we need to know the stripe aligned start */
5700 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5701 unsigned long full_stripe_len
= stripe_len
* nr_data_stripes(map
);
5702 raid56_full_stripe_start
= offset
;
5704 /* allow a write of a full stripe, but make sure we don't
5705 * allow straddling of stripes
5707 raid56_full_stripe_start
= div64_u64(raid56_full_stripe_start
,
5709 raid56_full_stripe_start
*= full_stripe_len
;
5712 if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
5714 /* For writes to RAID[56], allow a full stripeset across all disks.
5715 For other RAID types and for RAID[56] reads, just allow a single
5716 stripe (on a single disk). */
5717 if ((map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
5718 (op
== BTRFS_MAP_WRITE
)) {
5719 max_len
= stripe_len
* nr_data_stripes(map
) -
5720 (offset
- raid56_full_stripe_start
);
5722 /* we limit the length of each bio to what fits in a stripe */
5723 max_len
= stripe_len
- stripe_offset
;
5725 *length
= min_t(u64
, em
->len
- offset
, max_len
);
5727 *length
= em
->len
- offset
;
5730 /* This is for when we're called from btrfs_merge_bio_hook() and all
5731 it cares about is the length */
5735 btrfs_dev_replace_lock(dev_replace
, 0);
5736 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
5737 if (!dev_replace_is_ongoing
)
5738 btrfs_dev_replace_unlock(dev_replace
, 0);
5740 btrfs_dev_replace_set_lock_blocking(dev_replace
);
5742 if (dev_replace_is_ongoing
&& mirror_num
== map
->num_stripes
+ 1 &&
5743 !need_full_stripe(op
) && dev_replace
->tgtdev
!= NULL
) {
5744 ret
= get_extra_mirror_from_replace(fs_info
, logical
, *length
,
5745 dev_replace
->srcdev
->devid
,
5747 &physical_to_patch_in_first_stripe
);
5751 patch_the_first_stripe_for_dev_replace
= 1;
5752 } else if (mirror_num
> map
->num_stripes
) {
5758 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5759 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5761 if (op
!= BTRFS_MAP_WRITE
&& op
!= BTRFS_MAP_GET_READ_MIRRORS
)
5763 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
5764 if (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
)
5765 num_stripes
= map
->num_stripes
;
5766 else if (mirror_num
)
5767 stripe_index
= mirror_num
- 1;
5769 stripe_index
= find_live_mirror(fs_info
, map
, 0,
5771 current
->pid
% map
->num_stripes
,
5772 dev_replace_is_ongoing
);
5773 mirror_num
= stripe_index
+ 1;
5776 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
5777 if (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
) {
5778 num_stripes
= map
->num_stripes
;
5779 } else if (mirror_num
) {
5780 stripe_index
= mirror_num
- 1;
5785 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5786 u32 factor
= map
->num_stripes
/ map
->sub_stripes
;
5788 stripe_nr
= div_u64_rem(stripe_nr
, factor
, &stripe_index
);
5789 stripe_index
*= map
->sub_stripes
;
5791 if (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
)
5792 num_stripes
= map
->sub_stripes
;
5793 else if (mirror_num
)
5794 stripe_index
+= mirror_num
- 1;
5796 int old_stripe_index
= stripe_index
;
5797 stripe_index
= find_live_mirror(fs_info
, map
,
5799 map
->sub_stripes
, stripe_index
+
5800 current
->pid
% map
->sub_stripes
,
5801 dev_replace_is_ongoing
);
5802 mirror_num
= stripe_index
- old_stripe_index
+ 1;
5805 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5806 if (need_raid_map
&&
5807 (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
||
5809 /* push stripe_nr back to the start of the full stripe */
5810 stripe_nr
= div64_u64(raid56_full_stripe_start
,
5811 stripe_len
* nr_data_stripes(map
));
5813 /* RAID[56] write or recovery. Return all stripes */
5814 num_stripes
= map
->num_stripes
;
5815 max_errors
= nr_parity_stripes(map
);
5817 *length
= map
->stripe_len
;
5822 * Mirror #0 or #1 means the original data block.
5823 * Mirror #2 is RAID5 parity block.
5824 * Mirror #3 is RAID6 Q block.
5826 stripe_nr
= div_u64_rem(stripe_nr
,
5827 nr_data_stripes(map
), &stripe_index
);
5829 stripe_index
= nr_data_stripes(map
) +
5832 /* We distribute the parity blocks across stripes */
5833 div_u64_rem(stripe_nr
+ stripe_index
, map
->num_stripes
,
5835 if ((op
!= BTRFS_MAP_WRITE
&&
5836 op
!= BTRFS_MAP_GET_READ_MIRRORS
) &&
5842 * after this, stripe_nr is the number of stripes on this
5843 * device we have to walk to find the data, and stripe_index is
5844 * the number of our device in the stripe array
5846 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5848 mirror_num
= stripe_index
+ 1;
5850 if (stripe_index
>= map
->num_stripes
) {
5852 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5853 stripe_index
, map
->num_stripes
);
5858 num_alloc_stripes
= num_stripes
;
5859 if (dev_replace_is_ongoing
&& dev_replace
->tgtdev
!= NULL
) {
5860 if (op
== BTRFS_MAP_WRITE
)
5861 num_alloc_stripes
<<= 1;
5862 if (op
== BTRFS_MAP_GET_READ_MIRRORS
)
5863 num_alloc_stripes
++;
5864 tgtdev_indexes
= num_stripes
;
5867 bbio
= alloc_btrfs_bio(num_alloc_stripes
, tgtdev_indexes
);
5872 if (dev_replace_is_ongoing
&& dev_replace
->tgtdev
!= NULL
)
5873 bbio
->tgtdev_map
= (int *)(bbio
->stripes
+ num_alloc_stripes
);
5875 /* build raid_map */
5876 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
&& need_raid_map
&&
5877 (need_full_stripe(op
) || mirror_num
> 1)) {
5881 bbio
->raid_map
= (u64
*)((void *)bbio
->stripes
+
5882 sizeof(struct btrfs_bio_stripe
) *
5884 sizeof(int) * tgtdev_indexes
);
5886 /* Work out the disk rotation on this stripe-set */
5887 div_u64_rem(stripe_nr
, num_stripes
, &rot
);
5889 /* Fill in the logical address of each stripe */
5890 tmp
= stripe_nr
* nr_data_stripes(map
);
5891 for (i
= 0; i
< nr_data_stripes(map
); i
++)
5892 bbio
->raid_map
[(i
+rot
) % num_stripes
] =
5893 em
->start
+ (tmp
+ i
) * map
->stripe_len
;
5895 bbio
->raid_map
[(i
+rot
) % map
->num_stripes
] = RAID5_P_STRIPE
;
5896 if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5897 bbio
->raid_map
[(i
+rot
+1) % num_stripes
] =
5902 for (i
= 0; i
< num_stripes
; i
++) {
5903 bbio
->stripes
[i
].physical
=
5904 map
->stripes
[stripe_index
].physical
+
5906 stripe_nr
* map
->stripe_len
;
5907 bbio
->stripes
[i
].dev
=
5908 map
->stripes
[stripe_index
].dev
;
5912 if (need_full_stripe(op
))
5913 max_errors
= btrfs_chunk_max_errors(map
);
5916 sort_parity_stripes(bbio
, num_stripes
);
5918 if (dev_replace_is_ongoing
&& dev_replace
->tgtdev
!= NULL
&&
5919 need_full_stripe(op
)) {
5920 handle_ops_on_dev_replace(op
, &bbio
, dev_replace
, &num_stripes
,
5925 bbio
->map_type
= map
->type
;
5926 bbio
->num_stripes
= num_stripes
;
5927 bbio
->max_errors
= max_errors
;
5928 bbio
->mirror_num
= mirror_num
;
5931 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5932 * mirror_num == num_stripes + 1 && dev_replace target drive is
5933 * available as a mirror
5935 if (patch_the_first_stripe_for_dev_replace
&& num_stripes
> 0) {
5936 WARN_ON(num_stripes
> 1);
5937 bbio
->stripes
[0].dev
= dev_replace
->tgtdev
;
5938 bbio
->stripes
[0].physical
= physical_to_patch_in_first_stripe
;
5939 bbio
->mirror_num
= map
->num_stripes
+ 1;
5942 if (dev_replace_is_ongoing
) {
5943 btrfs_dev_replace_clear_lock_blocking(dev_replace
);
5944 btrfs_dev_replace_unlock(dev_replace
, 0);
5946 free_extent_map(em
);
5950 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, enum btrfs_map_op op
,
5951 u64 logical
, u64
*length
,
5952 struct btrfs_bio
**bbio_ret
, int mirror_num
)
5954 return __btrfs_map_block(fs_info
, op
, logical
, length
, bbio_ret
,
5958 /* For Scrub/replace */
5959 int btrfs_map_sblock(struct btrfs_fs_info
*fs_info
, enum btrfs_map_op op
,
5960 u64 logical
, u64
*length
,
5961 struct btrfs_bio
**bbio_ret
)
5963 return __btrfs_map_block(fs_info
, op
, logical
, length
, bbio_ret
, 0, 1);
5966 int btrfs_rmap_block(struct btrfs_fs_info
*fs_info
,
5967 u64 chunk_start
, u64 physical
, u64 devid
,
5968 u64
**logical
, int *naddrs
, int *stripe_len
)
5970 struct extent_map
*em
;
5971 struct map_lookup
*map
;
5979 em
= get_chunk_map(fs_info
, chunk_start
, 1);
5983 map
= em
->map_lookup
;
5985 rmap_len
= map
->stripe_len
;
5987 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5988 length
= div_u64(length
, map
->num_stripes
/ map
->sub_stripes
);
5989 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5990 length
= div_u64(length
, map
->num_stripes
);
5991 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5992 length
= div_u64(length
, nr_data_stripes(map
));
5993 rmap_len
= map
->stripe_len
* nr_data_stripes(map
);
5996 buf
= kcalloc(map
->num_stripes
, sizeof(u64
), GFP_NOFS
);
5997 BUG_ON(!buf
); /* -ENOMEM */
5999 for (i
= 0; i
< map
->num_stripes
; i
++) {
6000 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
6002 if (map
->stripes
[i
].physical
> physical
||
6003 map
->stripes
[i
].physical
+ length
<= physical
)
6006 stripe_nr
= physical
- map
->stripes
[i
].physical
;
6007 stripe_nr
= div64_u64(stripe_nr
, map
->stripe_len
);
6009 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
6010 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
6011 stripe_nr
= div_u64(stripe_nr
, map
->sub_stripes
);
6012 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
6013 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
6014 } /* else if RAID[56], multiply by nr_data_stripes().
6015 * Alternatively, just use rmap_len below instead of
6016 * map->stripe_len */
6018 bytenr
= chunk_start
+ stripe_nr
* rmap_len
;
6019 WARN_ON(nr
>= map
->num_stripes
);
6020 for (j
= 0; j
< nr
; j
++) {
6021 if (buf
[j
] == bytenr
)
6025 WARN_ON(nr
>= map
->num_stripes
);
6032 *stripe_len
= rmap_len
;
6034 free_extent_map(em
);
6038 static inline void btrfs_end_bbio(struct btrfs_bio
*bbio
, struct bio
*bio
)
6040 bio
->bi_private
= bbio
->private;
6041 bio
->bi_end_io
= bbio
->end_io
;
6044 btrfs_put_bbio(bbio
);
6047 static void btrfs_end_bio(struct bio
*bio
)
6049 struct btrfs_bio
*bbio
= bio
->bi_private
;
6050 int is_orig_bio
= 0;
6052 if (bio
->bi_status
) {
6053 atomic_inc(&bbio
->error
);
6054 if (bio
->bi_status
== BLK_STS_IOERR
||
6055 bio
->bi_status
== BLK_STS_TARGET
) {
6056 unsigned int stripe_index
=
6057 btrfs_io_bio(bio
)->stripe_index
;
6058 struct btrfs_device
*dev
;
6060 BUG_ON(stripe_index
>= bbio
->num_stripes
);
6061 dev
= bbio
->stripes
[stripe_index
].dev
;
6063 if (bio_op(bio
) == REQ_OP_WRITE
)
6064 btrfs_dev_stat_inc(dev
,
6065 BTRFS_DEV_STAT_WRITE_ERRS
);
6067 btrfs_dev_stat_inc(dev
,
6068 BTRFS_DEV_STAT_READ_ERRS
);
6069 if (bio
->bi_opf
& REQ_PREFLUSH
)
6070 btrfs_dev_stat_inc(dev
,
6071 BTRFS_DEV_STAT_FLUSH_ERRS
);
6072 btrfs_dev_stat_print_on_error(dev
);
6077 if (bio
== bbio
->orig_bio
)
6080 btrfs_bio_counter_dec(bbio
->fs_info
);
6082 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
6085 bio
= bbio
->orig_bio
;
6088 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
6089 /* only send an error to the higher layers if it is
6090 * beyond the tolerance of the btrfs bio
6092 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
6093 bio
->bi_status
= BLK_STS_IOERR
;
6096 * this bio is actually up to date, we didn't
6097 * go over the max number of errors
6102 btrfs_end_bbio(bbio
, bio
);
6103 } else if (!is_orig_bio
) {
6109 * see run_scheduled_bios for a description of why bios are collected for
6112 * This will add one bio to the pending list for a device and make sure
6113 * the work struct is scheduled.
6115 static noinline
void btrfs_schedule_bio(struct btrfs_device
*device
,
6118 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
6119 int should_queue
= 1;
6120 struct btrfs_pending_bios
*pending_bios
;
6122 if (device
->missing
|| !device
->bdev
) {
6127 /* don't bother with additional async steps for reads, right now */
6128 if (bio_op(bio
) == REQ_OP_READ
) {
6130 btrfsic_submit_bio(bio
);
6136 * nr_async_bios allows us to reliably return congestion to the
6137 * higher layers. Otherwise, the async bio makes it appear we have
6138 * made progress against dirty pages when we've really just put it
6139 * on a queue for later
6141 atomic_inc(&fs_info
->nr_async_bios
);
6142 WARN_ON(bio
->bi_next
);
6143 bio
->bi_next
= NULL
;
6145 spin_lock(&device
->io_lock
);
6146 if (op_is_sync(bio
->bi_opf
))
6147 pending_bios
= &device
->pending_sync_bios
;
6149 pending_bios
= &device
->pending_bios
;
6151 if (pending_bios
->tail
)
6152 pending_bios
->tail
->bi_next
= bio
;
6154 pending_bios
->tail
= bio
;
6155 if (!pending_bios
->head
)
6156 pending_bios
->head
= bio
;
6157 if (device
->running_pending
)
6160 spin_unlock(&device
->io_lock
);
6163 btrfs_queue_work(fs_info
->submit_workers
, &device
->work
);
6166 static void submit_stripe_bio(struct btrfs_bio
*bbio
, struct bio
*bio
,
6167 u64 physical
, int dev_nr
, int async
)
6169 struct btrfs_device
*dev
= bbio
->stripes
[dev_nr
].dev
;
6170 struct btrfs_fs_info
*fs_info
= bbio
->fs_info
;
6172 bio
->bi_private
= bbio
;
6173 btrfs_io_bio(bio
)->stripe_index
= dev_nr
;
6174 bio
->bi_end_io
= btrfs_end_bio
;
6175 bio
->bi_iter
.bi_sector
= physical
>> 9;
6178 struct rcu_string
*name
;
6181 name
= rcu_dereference(dev
->name
);
6182 btrfs_debug(fs_info
,
6183 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6184 bio_op(bio
), bio
->bi_opf
,
6185 (u64
)bio
->bi_iter
.bi_sector
,
6186 (u_long
)dev
->bdev
->bd_dev
, name
->str
, dev
->devid
,
6187 bio
->bi_iter
.bi_size
);
6191 bio
->bi_bdev
= dev
->bdev
;
6193 btrfs_bio_counter_inc_noblocked(fs_info
);
6196 btrfs_schedule_bio(dev
, bio
);
6198 btrfsic_submit_bio(bio
);
6201 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
6203 atomic_inc(&bbio
->error
);
6204 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
6205 /* Should be the original bio. */
6206 WARN_ON(bio
!= bbio
->orig_bio
);
6208 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
6209 bio
->bi_iter
.bi_sector
= logical
>> 9;
6210 bio
->bi_status
= BLK_STS_IOERR
;
6211 btrfs_end_bbio(bbio
, bio
);
6215 int btrfs_map_bio(struct btrfs_fs_info
*fs_info
, struct bio
*bio
,
6216 int mirror_num
, int async_submit
)
6218 struct btrfs_device
*dev
;
6219 struct bio
*first_bio
= bio
;
6220 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
6226 struct btrfs_bio
*bbio
= NULL
;
6228 length
= bio
->bi_iter
.bi_size
;
6229 map_length
= length
;
6231 btrfs_bio_counter_inc_blocked(fs_info
);
6232 ret
= __btrfs_map_block(fs_info
, bio_op(bio
), logical
,
6233 &map_length
, &bbio
, mirror_num
, 1);
6235 btrfs_bio_counter_dec(fs_info
);
6239 total_devs
= bbio
->num_stripes
;
6240 bbio
->orig_bio
= first_bio
;
6241 bbio
->private = first_bio
->bi_private
;
6242 bbio
->end_io
= first_bio
->bi_end_io
;
6243 bbio
->fs_info
= fs_info
;
6244 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
6246 if ((bbio
->map_type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
6247 ((bio_op(bio
) == REQ_OP_WRITE
) || (mirror_num
> 1))) {
6248 /* In this case, map_length has been set to the length of
6249 a single stripe; not the whole write */
6250 if (bio_op(bio
) == REQ_OP_WRITE
) {
6251 ret
= raid56_parity_write(fs_info
, bio
, bbio
,
6254 ret
= raid56_parity_recover(fs_info
, bio
, bbio
,
6255 map_length
, mirror_num
, 1);
6258 btrfs_bio_counter_dec(fs_info
);
6262 if (map_length
< length
) {
6264 "mapping failed logical %llu bio len %llu len %llu",
6265 logical
, length
, map_length
);
6269 for (dev_nr
= 0; dev_nr
< total_devs
; dev_nr
++) {
6270 dev
= bbio
->stripes
[dev_nr
].dev
;
6271 if (!dev
|| !dev
->bdev
||
6272 (bio_op(first_bio
) == REQ_OP_WRITE
&& !dev
->writeable
)) {
6273 bbio_error(bbio
, first_bio
, logical
);
6277 if (dev_nr
< total_devs
- 1)
6278 bio
= btrfs_bio_clone(first_bio
);
6282 submit_stripe_bio(bbio
, bio
, bbio
->stripes
[dev_nr
].physical
,
6283 dev_nr
, async_submit
);
6285 btrfs_bio_counter_dec(fs_info
);
6289 struct btrfs_device
*btrfs_find_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
6292 struct btrfs_device
*device
;
6293 struct btrfs_fs_devices
*cur_devices
;
6295 cur_devices
= fs_info
->fs_devices
;
6296 while (cur_devices
) {
6298 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
6299 device
= __find_device(&cur_devices
->devices
,
6304 cur_devices
= cur_devices
->seed
;
6309 static struct btrfs_device
*add_missing_dev(struct btrfs_fs_devices
*fs_devices
,
6310 u64 devid
, u8
*dev_uuid
)
6312 struct btrfs_device
*device
;
6314 device
= btrfs_alloc_device(NULL
, &devid
, dev_uuid
);
6318 list_add(&device
->dev_list
, &fs_devices
->devices
);
6319 device
->fs_devices
= fs_devices
;
6320 fs_devices
->num_devices
++;
6322 device
->missing
= 1;
6323 fs_devices
->missing_devices
++;
6329 * btrfs_alloc_device - allocate struct btrfs_device
6330 * @fs_info: used only for generating a new devid, can be NULL if
6331 * devid is provided (i.e. @devid != NULL).
6332 * @devid: a pointer to devid for this device. If NULL a new devid
6334 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6337 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6338 * on error. Returned struct is not linked onto any lists and can be
6339 * destroyed with kfree() right away.
6341 struct btrfs_device
*btrfs_alloc_device(struct btrfs_fs_info
*fs_info
,
6345 struct btrfs_device
*dev
;
6348 if (WARN_ON(!devid
&& !fs_info
))
6349 return ERR_PTR(-EINVAL
);
6351 dev
= __alloc_device();
6360 ret
= find_next_devid(fs_info
, &tmp
);
6363 return ERR_PTR(ret
);
6369 memcpy(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
);
6371 generate_random_uuid(dev
->uuid
);
6373 btrfs_init_work(&dev
->work
, btrfs_submit_helper
,
6374 pending_bios_fn
, NULL
, NULL
);
6379 /* Return -EIO if any error, otherwise return 0. */
6380 static int btrfs_check_chunk_valid(struct btrfs_fs_info
*fs_info
,
6381 struct extent_buffer
*leaf
,
6382 struct btrfs_chunk
*chunk
, u64 logical
)
6390 length
= btrfs_chunk_length(leaf
, chunk
);
6391 stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6392 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6393 sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6394 type
= btrfs_chunk_type(leaf
, chunk
);
6397 btrfs_err(fs_info
, "invalid chunk num_stripes: %u",
6401 if (!IS_ALIGNED(logical
, fs_info
->sectorsize
)) {
6402 btrfs_err(fs_info
, "invalid chunk logical %llu", logical
);
6405 if (btrfs_chunk_sector_size(leaf
, chunk
) != fs_info
->sectorsize
) {
6406 btrfs_err(fs_info
, "invalid chunk sectorsize %u",
6407 btrfs_chunk_sector_size(leaf
, chunk
));
6410 if (!length
|| !IS_ALIGNED(length
, fs_info
->sectorsize
)) {
6411 btrfs_err(fs_info
, "invalid chunk length %llu", length
);
6414 if (!is_power_of_2(stripe_len
) || stripe_len
!= BTRFS_STRIPE_LEN
) {
6415 btrfs_err(fs_info
, "invalid chunk stripe length: %llu",
6419 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK
| BTRFS_BLOCK_GROUP_PROFILE_MASK
) &
6421 btrfs_err(fs_info
, "unrecognized chunk type: %llu",
6422 ~(BTRFS_BLOCK_GROUP_TYPE_MASK
|
6423 BTRFS_BLOCK_GROUP_PROFILE_MASK
) &
6424 btrfs_chunk_type(leaf
, chunk
));
6427 if ((type
& BTRFS_BLOCK_GROUP_RAID10
&& sub_stripes
!= 2) ||
6428 (type
& BTRFS_BLOCK_GROUP_RAID1
&& num_stripes
< 1) ||
6429 (type
& BTRFS_BLOCK_GROUP_RAID5
&& num_stripes
< 2) ||
6430 (type
& BTRFS_BLOCK_GROUP_RAID6
&& num_stripes
< 3) ||
6431 (type
& BTRFS_BLOCK_GROUP_DUP
&& num_stripes
> 2) ||
6432 ((type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) == 0 &&
6433 num_stripes
!= 1)) {
6435 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6436 num_stripes
, sub_stripes
,
6437 type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
);
6444 static int read_one_chunk(struct btrfs_fs_info
*fs_info
, struct btrfs_key
*key
,
6445 struct extent_buffer
*leaf
,
6446 struct btrfs_chunk
*chunk
)
6448 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
6449 struct map_lookup
*map
;
6450 struct extent_map
*em
;
6455 u8 uuid
[BTRFS_UUID_SIZE
];
6460 logical
= key
->offset
;
6461 length
= btrfs_chunk_length(leaf
, chunk
);
6462 stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6463 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6465 ret
= btrfs_check_chunk_valid(fs_info
, leaf
, chunk
, logical
);
6469 read_lock(&map_tree
->map_tree
.lock
);
6470 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
6471 read_unlock(&map_tree
->map_tree
.lock
);
6473 /* already mapped? */
6474 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
6475 free_extent_map(em
);
6478 free_extent_map(em
);
6481 em
= alloc_extent_map();
6484 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
6486 free_extent_map(em
);
6490 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
6491 em
->map_lookup
= map
;
6492 em
->start
= logical
;
6495 em
->block_start
= 0;
6496 em
->block_len
= em
->len
;
6498 map
->num_stripes
= num_stripes
;
6499 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
6500 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
6501 map
->sector_size
= btrfs_chunk_sector_size(leaf
, chunk
);
6502 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6503 map
->type
= btrfs_chunk_type(leaf
, chunk
);
6504 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6505 for (i
= 0; i
< num_stripes
; i
++) {
6506 map
->stripes
[i
].physical
=
6507 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
6508 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
6509 read_extent_buffer(leaf
, uuid
, (unsigned long)
6510 btrfs_stripe_dev_uuid_nr(chunk
, i
),
6512 map
->stripes
[i
].dev
= btrfs_find_device(fs_info
, devid
,
6514 if (!map
->stripes
[i
].dev
&&
6515 !btrfs_test_opt(fs_info
, DEGRADED
)) {
6516 free_extent_map(em
);
6519 if (!map
->stripes
[i
].dev
) {
6520 map
->stripes
[i
].dev
=
6521 add_missing_dev(fs_info
->fs_devices
, devid
,
6523 if (!map
->stripes
[i
].dev
) {
6524 free_extent_map(em
);
6527 btrfs_warn(fs_info
, "devid %llu uuid %pU is missing",
6530 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
6533 write_lock(&map_tree
->map_tree
.lock
);
6534 ret
= add_extent_mapping(&map_tree
->map_tree
, em
, 0);
6535 write_unlock(&map_tree
->map_tree
.lock
);
6536 BUG_ON(ret
); /* Tree corruption */
6537 free_extent_map(em
);
6542 static void fill_device_from_item(struct extent_buffer
*leaf
,
6543 struct btrfs_dev_item
*dev_item
,
6544 struct btrfs_device
*device
)
6548 device
->devid
= btrfs_device_id(leaf
, dev_item
);
6549 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
6550 device
->total_bytes
= device
->disk_total_bytes
;
6551 device
->commit_total_bytes
= device
->disk_total_bytes
;
6552 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
6553 device
->commit_bytes_used
= device
->bytes_used
;
6554 device
->type
= btrfs_device_type(leaf
, dev_item
);
6555 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
6556 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
6557 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
6558 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
6559 device
->is_tgtdev_for_dev_replace
= 0;
6561 ptr
= btrfs_device_uuid(dev_item
);
6562 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
6565 static struct btrfs_fs_devices
*open_seed_devices(struct btrfs_fs_info
*fs_info
,
6568 struct btrfs_fs_devices
*fs_devices
;
6571 BUG_ON(!mutex_is_locked(&uuid_mutex
));
6573 fs_devices
= fs_info
->fs_devices
->seed
;
6574 while (fs_devices
) {
6575 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
))
6578 fs_devices
= fs_devices
->seed
;
6581 fs_devices
= find_fsid(fsid
);
6583 if (!btrfs_test_opt(fs_info
, DEGRADED
))
6584 return ERR_PTR(-ENOENT
);
6586 fs_devices
= alloc_fs_devices(fsid
);
6587 if (IS_ERR(fs_devices
))
6590 fs_devices
->seeding
= 1;
6591 fs_devices
->opened
= 1;
6595 fs_devices
= clone_fs_devices(fs_devices
);
6596 if (IS_ERR(fs_devices
))
6599 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
6600 fs_info
->bdev_holder
);
6602 free_fs_devices(fs_devices
);
6603 fs_devices
= ERR_PTR(ret
);
6607 if (!fs_devices
->seeding
) {
6608 __btrfs_close_devices(fs_devices
);
6609 free_fs_devices(fs_devices
);
6610 fs_devices
= ERR_PTR(-EINVAL
);
6614 fs_devices
->seed
= fs_info
->fs_devices
->seed
;
6615 fs_info
->fs_devices
->seed
= fs_devices
;
6620 static int read_one_dev(struct btrfs_fs_info
*fs_info
,
6621 struct extent_buffer
*leaf
,
6622 struct btrfs_dev_item
*dev_item
)
6624 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6625 struct btrfs_device
*device
;
6628 u8 fs_uuid
[BTRFS_UUID_SIZE
];
6629 u8 dev_uuid
[BTRFS_UUID_SIZE
];
6631 devid
= btrfs_device_id(leaf
, dev_item
);
6632 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
6634 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
6637 if (memcmp(fs_uuid
, fs_info
->fsid
, BTRFS_UUID_SIZE
)) {
6638 fs_devices
= open_seed_devices(fs_info
, fs_uuid
);
6639 if (IS_ERR(fs_devices
))
6640 return PTR_ERR(fs_devices
);
6643 device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, fs_uuid
);
6645 if (!btrfs_test_opt(fs_info
, DEGRADED
))
6648 device
= add_missing_dev(fs_devices
, devid
, dev_uuid
);
6651 btrfs_warn(fs_info
, "devid %llu uuid %pU missing",
6654 if (!device
->bdev
&& !btrfs_test_opt(fs_info
, DEGRADED
))
6657 if(!device
->bdev
&& !device
->missing
) {
6659 * this happens when a device that was properly setup
6660 * in the device info lists suddenly goes bad.
6661 * device->bdev is NULL, and so we have to set
6662 * device->missing to one here
6664 device
->fs_devices
->missing_devices
++;
6665 device
->missing
= 1;
6668 /* Move the device to its own fs_devices */
6669 if (device
->fs_devices
!= fs_devices
) {
6670 ASSERT(device
->missing
);
6672 list_move(&device
->dev_list
, &fs_devices
->devices
);
6673 device
->fs_devices
->num_devices
--;
6674 fs_devices
->num_devices
++;
6676 device
->fs_devices
->missing_devices
--;
6677 fs_devices
->missing_devices
++;
6679 device
->fs_devices
= fs_devices
;
6683 if (device
->fs_devices
!= fs_info
->fs_devices
) {
6684 BUG_ON(device
->writeable
);
6685 if (device
->generation
!=
6686 btrfs_device_generation(leaf
, dev_item
))
6690 fill_device_from_item(leaf
, dev_item
, device
);
6691 device
->in_fs_metadata
= 1;
6692 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
6693 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
6694 atomic64_add(device
->total_bytes
- device
->bytes_used
,
6695 &fs_info
->free_chunk_space
);
6701 int btrfs_read_sys_array(struct btrfs_fs_info
*fs_info
)
6703 struct btrfs_root
*root
= fs_info
->tree_root
;
6704 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
6705 struct extent_buffer
*sb
;
6706 struct btrfs_disk_key
*disk_key
;
6707 struct btrfs_chunk
*chunk
;
6709 unsigned long sb_array_offset
;
6716 struct btrfs_key key
;
6718 ASSERT(BTRFS_SUPER_INFO_SIZE
<= fs_info
->nodesize
);
6720 * This will create extent buffer of nodesize, superblock size is
6721 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6722 * overallocate but we can keep it as-is, only the first page is used.
6724 sb
= btrfs_find_create_tree_block(fs_info
, BTRFS_SUPER_INFO_OFFSET
);
6727 set_extent_buffer_uptodate(sb
);
6728 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
6730 * The sb extent buffer is artificial and just used to read the system array.
6731 * set_extent_buffer_uptodate() call does not properly mark all it's
6732 * pages up-to-date when the page is larger: extent does not cover the
6733 * whole page and consequently check_page_uptodate does not find all
6734 * the page's extents up-to-date (the hole beyond sb),
6735 * write_extent_buffer then triggers a WARN_ON.
6737 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6738 * but sb spans only this function. Add an explicit SetPageUptodate call
6739 * to silence the warning eg. on PowerPC 64.
6741 if (PAGE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
6742 SetPageUptodate(sb
->pages
[0]);
6744 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
6745 array_size
= btrfs_super_sys_array_size(super_copy
);
6747 array_ptr
= super_copy
->sys_chunk_array
;
6748 sb_array_offset
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
6751 while (cur_offset
< array_size
) {
6752 disk_key
= (struct btrfs_disk_key
*)array_ptr
;
6753 len
= sizeof(*disk_key
);
6754 if (cur_offset
+ len
> array_size
)
6755 goto out_short_read
;
6757 btrfs_disk_key_to_cpu(&key
, disk_key
);
6760 sb_array_offset
+= len
;
6763 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6764 chunk
= (struct btrfs_chunk
*)sb_array_offset
;
6766 * At least one btrfs_chunk with one stripe must be
6767 * present, exact stripe count check comes afterwards
6769 len
= btrfs_chunk_item_size(1);
6770 if (cur_offset
+ len
> array_size
)
6771 goto out_short_read
;
6773 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
6776 "invalid number of stripes %u in sys_array at offset %u",
6777 num_stripes
, cur_offset
);
6782 type
= btrfs_chunk_type(sb
, chunk
);
6783 if ((type
& BTRFS_BLOCK_GROUP_SYSTEM
) == 0) {
6785 "invalid chunk type %llu in sys_array at offset %u",
6791 len
= btrfs_chunk_item_size(num_stripes
);
6792 if (cur_offset
+ len
> array_size
)
6793 goto out_short_read
;
6795 ret
= read_one_chunk(fs_info
, &key
, sb
, chunk
);
6800 "unexpected item type %u in sys_array at offset %u",
6801 (u32
)key
.type
, cur_offset
);
6806 sb_array_offset
+= len
;
6809 clear_extent_buffer_uptodate(sb
);
6810 free_extent_buffer_stale(sb
);
6814 btrfs_err(fs_info
, "sys_array too short to read %u bytes at offset %u",
6816 clear_extent_buffer_uptodate(sb
);
6817 free_extent_buffer_stale(sb
);
6821 int btrfs_read_chunk_tree(struct btrfs_fs_info
*fs_info
)
6823 struct btrfs_root
*root
= fs_info
->chunk_root
;
6824 struct btrfs_path
*path
;
6825 struct extent_buffer
*leaf
;
6826 struct btrfs_key key
;
6827 struct btrfs_key found_key
;
6832 path
= btrfs_alloc_path();
6836 mutex_lock(&uuid_mutex
);
6837 mutex_lock(&fs_info
->chunk_mutex
);
6840 * Read all device items, and then all the chunk items. All
6841 * device items are found before any chunk item (their object id
6842 * is smaller than the lowest possible object id for a chunk
6843 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6845 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
6848 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6852 leaf
= path
->nodes
[0];
6853 slot
= path
->slots
[0];
6854 if (slot
>= btrfs_header_nritems(leaf
)) {
6855 ret
= btrfs_next_leaf(root
, path
);
6862 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6863 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
6864 struct btrfs_dev_item
*dev_item
;
6865 dev_item
= btrfs_item_ptr(leaf
, slot
,
6866 struct btrfs_dev_item
);
6867 ret
= read_one_dev(fs_info
, leaf
, dev_item
);
6871 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6872 struct btrfs_chunk
*chunk
;
6873 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
6874 ret
= read_one_chunk(fs_info
, &found_key
, leaf
, chunk
);
6882 * After loading chunk tree, we've got all device information,
6883 * do another round of validation checks.
6885 if (total_dev
!= fs_info
->fs_devices
->total_devices
) {
6887 "super_num_devices %llu mismatch with num_devices %llu found here",
6888 btrfs_super_num_devices(fs_info
->super_copy
),
6893 if (btrfs_super_total_bytes(fs_info
->super_copy
) <
6894 fs_info
->fs_devices
->total_rw_bytes
) {
6896 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6897 btrfs_super_total_bytes(fs_info
->super_copy
),
6898 fs_info
->fs_devices
->total_rw_bytes
);
6904 mutex_unlock(&fs_info
->chunk_mutex
);
6905 mutex_unlock(&uuid_mutex
);
6907 btrfs_free_path(path
);
6911 void btrfs_init_devices_late(struct btrfs_fs_info
*fs_info
)
6913 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6914 struct btrfs_device
*device
;
6916 while (fs_devices
) {
6917 mutex_lock(&fs_devices
->device_list_mutex
);
6918 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
)
6919 device
->fs_info
= fs_info
;
6920 mutex_unlock(&fs_devices
->device_list_mutex
);
6922 fs_devices
= fs_devices
->seed
;
6926 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
6930 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6931 btrfs_dev_stat_reset(dev
, i
);
6934 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
6936 struct btrfs_key key
;
6937 struct btrfs_key found_key
;
6938 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6939 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6940 struct extent_buffer
*eb
;
6943 struct btrfs_device
*device
;
6944 struct btrfs_path
*path
= NULL
;
6947 path
= btrfs_alloc_path();
6953 mutex_lock(&fs_devices
->device_list_mutex
);
6954 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6956 struct btrfs_dev_stats_item
*ptr
;
6958 key
.objectid
= BTRFS_DEV_STATS_OBJECTID
;
6959 key
.type
= BTRFS_PERSISTENT_ITEM_KEY
;
6960 key
.offset
= device
->devid
;
6961 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
6963 __btrfs_reset_dev_stats(device
);
6964 device
->dev_stats_valid
= 1;
6965 btrfs_release_path(path
);
6968 slot
= path
->slots
[0];
6969 eb
= path
->nodes
[0];
6970 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
6971 item_size
= btrfs_item_size_nr(eb
, slot
);
6973 ptr
= btrfs_item_ptr(eb
, slot
,
6974 struct btrfs_dev_stats_item
);
6976 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6977 if (item_size
>= (1 + i
) * sizeof(__le64
))
6978 btrfs_dev_stat_set(device
, i
,
6979 btrfs_dev_stats_value(eb
, ptr
, i
));
6981 btrfs_dev_stat_reset(device
, i
);
6984 device
->dev_stats_valid
= 1;
6985 btrfs_dev_stat_print_on_load(device
);
6986 btrfs_release_path(path
);
6988 mutex_unlock(&fs_devices
->device_list_mutex
);
6991 btrfs_free_path(path
);
6992 return ret
< 0 ? ret
: 0;
6995 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
6996 struct btrfs_fs_info
*fs_info
,
6997 struct btrfs_device
*device
)
6999 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
7000 struct btrfs_path
*path
;
7001 struct btrfs_key key
;
7002 struct extent_buffer
*eb
;
7003 struct btrfs_dev_stats_item
*ptr
;
7007 key
.objectid
= BTRFS_DEV_STATS_OBJECTID
;
7008 key
.type
= BTRFS_PERSISTENT_ITEM_KEY
;
7009 key
.offset
= device
->devid
;
7011 path
= btrfs_alloc_path();
7014 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
7016 btrfs_warn_in_rcu(fs_info
,
7017 "error %d while searching for dev_stats item for device %s",
7018 ret
, rcu_str_deref(device
->name
));
7023 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
7024 /* need to delete old one and insert a new one */
7025 ret
= btrfs_del_item(trans
, dev_root
, path
);
7027 btrfs_warn_in_rcu(fs_info
,
7028 "delete too small dev_stats item for device %s failed %d",
7029 rcu_str_deref(device
->name
), ret
);
7036 /* need to insert a new item */
7037 btrfs_release_path(path
);
7038 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
7039 &key
, sizeof(*ptr
));
7041 btrfs_warn_in_rcu(fs_info
,
7042 "insert dev_stats item for device %s failed %d",
7043 rcu_str_deref(device
->name
), ret
);
7048 eb
= path
->nodes
[0];
7049 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
7050 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7051 btrfs_set_dev_stats_value(eb
, ptr
, i
,
7052 btrfs_dev_stat_read(device
, i
));
7053 btrfs_mark_buffer_dirty(eb
);
7056 btrfs_free_path(path
);
7061 * called from commit_transaction. Writes all changed device stats to disk.
7063 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
7064 struct btrfs_fs_info
*fs_info
)
7066 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7067 struct btrfs_device
*device
;
7071 mutex_lock(&fs_devices
->device_list_mutex
);
7072 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
7073 if (!device
->dev_stats_valid
|| !btrfs_dev_stats_dirty(device
))
7076 stats_cnt
= atomic_read(&device
->dev_stats_ccnt
);
7077 ret
= update_dev_stat_item(trans
, fs_info
, device
);
7079 atomic_sub(stats_cnt
, &device
->dev_stats_ccnt
);
7081 mutex_unlock(&fs_devices
->device_list_mutex
);
7086 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
7088 btrfs_dev_stat_inc(dev
, index
);
7089 btrfs_dev_stat_print_on_error(dev
);
7092 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
7094 if (!dev
->dev_stats_valid
)
7096 btrfs_err_rl_in_rcu(dev
->fs_info
,
7097 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7098 rcu_str_deref(dev
->name
),
7099 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
7100 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
7101 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
7102 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
7103 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
7106 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
7110 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7111 if (btrfs_dev_stat_read(dev
, i
) != 0)
7113 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
7114 return; /* all values == 0, suppress message */
7116 btrfs_info_in_rcu(dev
->fs_info
,
7117 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7118 rcu_str_deref(dev
->name
),
7119 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
7120 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
7121 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
7122 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
7123 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
7126 int btrfs_get_dev_stats(struct btrfs_fs_info
*fs_info
,
7127 struct btrfs_ioctl_get_dev_stats
*stats
)
7129 struct btrfs_device
*dev
;
7130 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7133 mutex_lock(&fs_devices
->device_list_mutex
);
7134 dev
= btrfs_find_device(fs_info
, stats
->devid
, NULL
, NULL
);
7135 mutex_unlock(&fs_devices
->device_list_mutex
);
7138 btrfs_warn(fs_info
, "get dev_stats failed, device not found");
7140 } else if (!dev
->dev_stats_valid
) {
7141 btrfs_warn(fs_info
, "get dev_stats failed, not yet valid");
7143 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
7144 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
7145 if (stats
->nr_items
> i
)
7147 btrfs_dev_stat_read_and_reset(dev
, i
);
7149 btrfs_dev_stat_reset(dev
, i
);
7152 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7153 if (stats
->nr_items
> i
)
7154 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
7156 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
7157 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
7161 void btrfs_scratch_superblocks(struct block_device
*bdev
, const char *device_path
)
7163 struct buffer_head
*bh
;
7164 struct btrfs_super_block
*disk_super
;
7170 for (copy_num
= 0; copy_num
< BTRFS_SUPER_MIRROR_MAX
;
7173 if (btrfs_read_dev_one_super(bdev
, copy_num
, &bh
))
7176 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
7178 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
7179 set_buffer_dirty(bh
);
7180 sync_dirty_buffer(bh
);
7184 /* Notify udev that device has changed */
7185 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
7187 /* Update ctime/mtime for device path for libblkid */
7188 update_dev_time(device_path
);
7192 * Update the size of all devices, which is used for writing out the
7195 void btrfs_update_commit_device_size(struct btrfs_fs_info
*fs_info
)
7197 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7198 struct btrfs_device
*curr
, *next
;
7200 if (list_empty(&fs_devices
->resized_devices
))
7203 mutex_lock(&fs_devices
->device_list_mutex
);
7204 mutex_lock(&fs_info
->chunk_mutex
);
7205 list_for_each_entry_safe(curr
, next
, &fs_devices
->resized_devices
,
7207 list_del_init(&curr
->resized_list
);
7208 curr
->commit_total_bytes
= curr
->disk_total_bytes
;
7210 mutex_unlock(&fs_info
->chunk_mutex
);
7211 mutex_unlock(&fs_devices
->device_list_mutex
);
7214 /* Must be invoked during the transaction commit */
7215 void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info
*fs_info
,
7216 struct btrfs_transaction
*transaction
)
7218 struct extent_map
*em
;
7219 struct map_lookup
*map
;
7220 struct btrfs_device
*dev
;
7223 if (list_empty(&transaction
->pending_chunks
))
7226 /* In order to kick the device replace finish process */
7227 mutex_lock(&fs_info
->chunk_mutex
);
7228 list_for_each_entry(em
, &transaction
->pending_chunks
, list
) {
7229 map
= em
->map_lookup
;
7231 for (i
= 0; i
< map
->num_stripes
; i
++) {
7232 dev
= map
->stripes
[i
].dev
;
7233 dev
->commit_bytes_used
= dev
->bytes_used
;
7236 mutex_unlock(&fs_info
->chunk_mutex
);
7239 void btrfs_set_fs_info_ptr(struct btrfs_fs_info
*fs_info
)
7241 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7242 while (fs_devices
) {
7243 fs_devices
->fs_info
= fs_info
;
7244 fs_devices
= fs_devices
->seed
;
7248 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info
*fs_info
)
7250 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7251 while (fs_devices
) {
7252 fs_devices
->fs_info
= NULL
;
7253 fs_devices
= fs_devices
->seed
;