2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/iocontext.h>
24 #include <linux/capability.h>
25 #include <linux/ratelimit.h>
26 #include <linux/kthread.h>
27 #include <linux/raid/pq.h>
28 #include <linux/semaphore.h>
29 #include <linux/uuid.h>
30 #include <asm/div64.h>
32 #include "extent_map.h"
34 #include "transaction.h"
35 #include "print-tree.h"
38 #include "async-thread.h"
39 #include "check-integrity.h"
40 #include "rcu-string.h"
42 #include "dev-replace.h"
45 const struct btrfs_raid_attr btrfs_raid_array
[BTRFS_NR_RAID_TYPES
] = {
46 [BTRFS_RAID_RAID10
] = {
49 .devs_max
= 0, /* 0 == as many as possible */
51 .tolerated_failures
= 1,
55 [BTRFS_RAID_RAID1
] = {
60 .tolerated_failures
= 1,
69 .tolerated_failures
= 0,
73 [BTRFS_RAID_RAID0
] = {
78 .tolerated_failures
= 0,
82 [BTRFS_RAID_SINGLE
] = {
87 .tolerated_failures
= 0,
91 [BTRFS_RAID_RAID5
] = {
96 .tolerated_failures
= 1,
100 [BTRFS_RAID_RAID6
] = {
105 .tolerated_failures
= 2,
111 const u64 btrfs_raid_group
[BTRFS_NR_RAID_TYPES
] = {
112 [BTRFS_RAID_RAID10
] = BTRFS_BLOCK_GROUP_RAID10
,
113 [BTRFS_RAID_RAID1
] = BTRFS_BLOCK_GROUP_RAID1
,
114 [BTRFS_RAID_DUP
] = BTRFS_BLOCK_GROUP_DUP
,
115 [BTRFS_RAID_RAID0
] = BTRFS_BLOCK_GROUP_RAID0
,
116 [BTRFS_RAID_SINGLE
] = 0,
117 [BTRFS_RAID_RAID5
] = BTRFS_BLOCK_GROUP_RAID5
,
118 [BTRFS_RAID_RAID6
] = BTRFS_BLOCK_GROUP_RAID6
,
122 * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
123 * condition is not met. Zero means there's no corresponding
124 * BTRFS_ERROR_DEV_*_NOT_MET value.
126 const int btrfs_raid_mindev_error
[BTRFS_NR_RAID_TYPES
] = {
127 [BTRFS_RAID_RAID10
] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET
,
128 [BTRFS_RAID_RAID1
] = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET
,
129 [BTRFS_RAID_DUP
] = 0,
130 [BTRFS_RAID_RAID0
] = 0,
131 [BTRFS_RAID_SINGLE
] = 0,
132 [BTRFS_RAID_RAID5
] = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET
,
133 [BTRFS_RAID_RAID6
] = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET
,
136 static int init_first_rw_device(struct btrfs_trans_handle
*trans
,
137 struct btrfs_root
*root
,
138 struct btrfs_device
*device
);
139 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
);
140 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
);
141 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
);
142 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*device
);
144 DEFINE_MUTEX(uuid_mutex
);
145 static LIST_HEAD(fs_uuids
);
146 struct list_head
*btrfs_get_fs_uuids(void)
151 static struct btrfs_fs_devices
*__alloc_fs_devices(void)
153 struct btrfs_fs_devices
*fs_devs
;
155 fs_devs
= kzalloc(sizeof(*fs_devs
), GFP_KERNEL
);
157 return ERR_PTR(-ENOMEM
);
159 mutex_init(&fs_devs
->device_list_mutex
);
161 INIT_LIST_HEAD(&fs_devs
->devices
);
162 INIT_LIST_HEAD(&fs_devs
->resized_devices
);
163 INIT_LIST_HEAD(&fs_devs
->alloc_list
);
164 INIT_LIST_HEAD(&fs_devs
->list
);
170 * alloc_fs_devices - allocate struct btrfs_fs_devices
171 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
174 * Return: a pointer to a new &struct btrfs_fs_devices on success;
175 * ERR_PTR() on error. Returned struct is not linked onto any lists and
176 * can be destroyed with kfree() right away.
178 static struct btrfs_fs_devices
*alloc_fs_devices(const u8
*fsid
)
180 struct btrfs_fs_devices
*fs_devs
;
182 fs_devs
= __alloc_fs_devices();
187 memcpy(fs_devs
->fsid
, fsid
, BTRFS_FSID_SIZE
);
189 generate_random_uuid(fs_devs
->fsid
);
194 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
196 struct btrfs_device
*device
;
197 WARN_ON(fs_devices
->opened
);
198 while (!list_empty(&fs_devices
->devices
)) {
199 device
= list_entry(fs_devices
->devices
.next
,
200 struct btrfs_device
, dev_list
);
201 list_del(&device
->dev_list
);
202 rcu_string_free(device
->name
);
208 static void btrfs_kobject_uevent(struct block_device
*bdev
,
209 enum kobject_action action
)
213 ret
= kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, action
);
215 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
217 kobject_name(&disk_to_dev(bdev
->bd_disk
)->kobj
),
218 &disk_to_dev(bdev
->bd_disk
)->kobj
);
221 void btrfs_cleanup_fs_uuids(void)
223 struct btrfs_fs_devices
*fs_devices
;
225 while (!list_empty(&fs_uuids
)) {
226 fs_devices
= list_entry(fs_uuids
.next
,
227 struct btrfs_fs_devices
, list
);
228 list_del(&fs_devices
->list
);
229 free_fs_devices(fs_devices
);
233 static struct btrfs_device
*__alloc_device(void)
235 struct btrfs_device
*dev
;
237 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
239 return ERR_PTR(-ENOMEM
);
241 INIT_LIST_HEAD(&dev
->dev_list
);
242 INIT_LIST_HEAD(&dev
->dev_alloc_list
);
243 INIT_LIST_HEAD(&dev
->resized_list
);
245 spin_lock_init(&dev
->io_lock
);
247 spin_lock_init(&dev
->reada_lock
);
248 atomic_set(&dev
->reada_in_flight
, 0);
249 atomic_set(&dev
->dev_stats_ccnt
, 0);
250 btrfs_device_data_ordered_init(dev
);
251 INIT_RADIX_TREE(&dev
->reada_zones
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
252 INIT_RADIX_TREE(&dev
->reada_extents
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
257 static noinline
struct btrfs_device
*__find_device(struct list_head
*head
,
260 struct btrfs_device
*dev
;
262 list_for_each_entry(dev
, head
, dev_list
) {
263 if (dev
->devid
== devid
&&
264 (!uuid
|| !memcmp(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
))) {
271 static noinline
struct btrfs_fs_devices
*find_fsid(u8
*fsid
)
273 struct btrfs_fs_devices
*fs_devices
;
275 list_for_each_entry(fs_devices
, &fs_uuids
, list
) {
276 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
283 btrfs_get_bdev_and_sb(const char *device_path
, fmode_t flags
, void *holder
,
284 int flush
, struct block_device
**bdev
,
285 struct buffer_head
**bh
)
289 *bdev
= blkdev_get_by_path(device_path
, flags
, holder
);
292 ret
= PTR_ERR(*bdev
);
297 filemap_write_and_wait((*bdev
)->bd_inode
->i_mapping
);
298 ret
= set_blocksize(*bdev
, 4096);
300 blkdev_put(*bdev
, flags
);
303 invalidate_bdev(*bdev
);
304 *bh
= btrfs_read_dev_super(*bdev
);
307 blkdev_put(*bdev
, flags
);
319 static void requeue_list(struct btrfs_pending_bios
*pending_bios
,
320 struct bio
*head
, struct bio
*tail
)
323 struct bio
*old_head
;
325 old_head
= pending_bios
->head
;
326 pending_bios
->head
= head
;
327 if (pending_bios
->tail
)
328 tail
->bi_next
= old_head
;
330 pending_bios
->tail
= tail
;
334 * we try to collect pending bios for a device so we don't get a large
335 * number of procs sending bios down to the same device. This greatly
336 * improves the schedulers ability to collect and merge the bios.
338 * But, it also turns into a long list of bios to process and that is sure
339 * to eventually make the worker thread block. The solution here is to
340 * make some progress and then put this work struct back at the end of
341 * the list if the block device is congested. This way, multiple devices
342 * can make progress from a single worker thread.
344 static noinline
void run_scheduled_bios(struct btrfs_device
*device
)
347 struct backing_dev_info
*bdi
;
348 struct btrfs_fs_info
*fs_info
;
349 struct btrfs_pending_bios
*pending_bios
;
353 unsigned long num_run
;
354 unsigned long batch_run
= 0;
356 unsigned long last_waited
= 0;
358 int sync_pending
= 0;
359 struct blk_plug plug
;
362 * this function runs all the bios we've collected for
363 * a particular device. We don't want to wander off to
364 * another device without first sending all of these down.
365 * So, setup a plug here and finish it off before we return
367 blk_start_plug(&plug
);
369 bdi
= blk_get_backing_dev_info(device
->bdev
);
370 fs_info
= device
->dev_root
->fs_info
;
371 limit
= btrfs_async_submit_limit(fs_info
);
372 limit
= limit
* 2 / 3;
375 spin_lock(&device
->io_lock
);
380 /* take all the bios off the list at once and process them
381 * later on (without the lock held). But, remember the
382 * tail and other pointers so the bios can be properly reinserted
383 * into the list if we hit congestion
385 if (!force_reg
&& device
->pending_sync_bios
.head
) {
386 pending_bios
= &device
->pending_sync_bios
;
389 pending_bios
= &device
->pending_bios
;
393 pending
= pending_bios
->head
;
394 tail
= pending_bios
->tail
;
395 WARN_ON(pending
&& !tail
);
398 * if pending was null this time around, no bios need processing
399 * at all and we can stop. Otherwise it'll loop back up again
400 * and do an additional check so no bios are missed.
402 * device->running_pending is used to synchronize with the
405 if (device
->pending_sync_bios
.head
== NULL
&&
406 device
->pending_bios
.head
== NULL
) {
408 device
->running_pending
= 0;
411 device
->running_pending
= 1;
414 pending_bios
->head
= NULL
;
415 pending_bios
->tail
= NULL
;
417 spin_unlock(&device
->io_lock
);
422 /* we want to work on both lists, but do more bios on the
423 * sync list than the regular list
426 pending_bios
!= &device
->pending_sync_bios
&&
427 device
->pending_sync_bios
.head
) ||
428 (num_run
> 64 && pending_bios
== &device
->pending_sync_bios
&&
429 device
->pending_bios
.head
)) {
430 spin_lock(&device
->io_lock
);
431 requeue_list(pending_bios
, pending
, tail
);
436 pending
= pending
->bi_next
;
440 * atomic_dec_return implies a barrier for waitqueue_active
442 if (atomic_dec_return(&fs_info
->nr_async_bios
) < limit
&&
443 waitqueue_active(&fs_info
->async_submit_wait
))
444 wake_up(&fs_info
->async_submit_wait
);
446 BUG_ON(atomic_read(&cur
->__bi_cnt
) == 0);
449 * if we're doing the sync list, record that our
450 * plug has some sync requests on it
452 * If we're doing the regular list and there are
453 * sync requests sitting around, unplug before
456 if (pending_bios
== &device
->pending_sync_bios
) {
458 } else if (sync_pending
) {
459 blk_finish_plug(&plug
);
460 blk_start_plug(&plug
);
464 btrfsic_submit_bio(cur
);
471 * we made progress, there is more work to do and the bdi
472 * is now congested. Back off and let other work structs
475 if (pending
&& bdi_write_congested(bdi
) && batch_run
> 8 &&
476 fs_info
->fs_devices
->open_devices
> 1) {
477 struct io_context
*ioc
;
479 ioc
= current
->io_context
;
482 * the main goal here is that we don't want to
483 * block if we're going to be able to submit
484 * more requests without blocking.
486 * This code does two great things, it pokes into
487 * the elevator code from a filesystem _and_
488 * it makes assumptions about how batching works.
490 if (ioc
&& ioc
->nr_batch_requests
> 0 &&
491 time_before(jiffies
, ioc
->last_waited
+ HZ
/50UL) &&
493 ioc
->last_waited
== last_waited
)) {
495 * we want to go through our batch of
496 * requests and stop. So, we copy out
497 * the ioc->last_waited time and test
498 * against it before looping
500 last_waited
= ioc
->last_waited
;
504 spin_lock(&device
->io_lock
);
505 requeue_list(pending_bios
, pending
, tail
);
506 device
->running_pending
= 1;
508 spin_unlock(&device
->io_lock
);
509 btrfs_queue_work(fs_info
->submit_workers
,
513 /* unplug every 64 requests just for good measure */
514 if (batch_run
% 64 == 0) {
515 blk_finish_plug(&plug
);
516 blk_start_plug(&plug
);
525 spin_lock(&device
->io_lock
);
526 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
528 spin_unlock(&device
->io_lock
);
531 blk_finish_plug(&plug
);
534 static void pending_bios_fn(struct btrfs_work
*work
)
536 struct btrfs_device
*device
;
538 device
= container_of(work
, struct btrfs_device
, work
);
539 run_scheduled_bios(device
);
543 void btrfs_free_stale_device(struct btrfs_device
*cur_dev
)
545 struct btrfs_fs_devices
*fs_devs
;
546 struct btrfs_device
*dev
;
551 list_for_each_entry(fs_devs
, &fs_uuids
, list
) {
556 if (fs_devs
->seeding
)
559 list_for_each_entry(dev
, &fs_devs
->devices
, dev_list
) {
567 * Todo: This won't be enough. What if the same device
568 * comes back (with new uuid and) with its mapper path?
569 * But for now, this does help as mostly an admin will
570 * either use mapper or non mapper path throughout.
573 del
= strcmp(rcu_str_deref(dev
->name
),
574 rcu_str_deref(cur_dev
->name
));
581 /* delete the stale device */
582 if (fs_devs
->num_devices
== 1) {
583 btrfs_sysfs_remove_fsid(fs_devs
);
584 list_del(&fs_devs
->list
);
585 free_fs_devices(fs_devs
);
587 fs_devs
->num_devices
--;
588 list_del(&dev
->dev_list
);
589 rcu_string_free(dev
->name
);
598 * Add new device to list of registered devices
601 * 1 - first time device is seen
602 * 0 - device already known
605 static noinline
int device_list_add(const char *path
,
606 struct btrfs_super_block
*disk_super
,
607 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
609 struct btrfs_device
*device
;
610 struct btrfs_fs_devices
*fs_devices
;
611 struct rcu_string
*name
;
613 u64 found_transid
= btrfs_super_generation(disk_super
);
615 fs_devices
= find_fsid(disk_super
->fsid
);
617 fs_devices
= alloc_fs_devices(disk_super
->fsid
);
618 if (IS_ERR(fs_devices
))
619 return PTR_ERR(fs_devices
);
621 list_add(&fs_devices
->list
, &fs_uuids
);
625 device
= __find_device(&fs_devices
->devices
, devid
,
626 disk_super
->dev_item
.uuid
);
630 if (fs_devices
->opened
)
633 device
= btrfs_alloc_device(NULL
, &devid
,
634 disk_super
->dev_item
.uuid
);
635 if (IS_ERR(device
)) {
636 /* we can safely leave the fs_devices entry around */
637 return PTR_ERR(device
);
640 name
= rcu_string_strdup(path
, GFP_NOFS
);
645 rcu_assign_pointer(device
->name
, name
);
647 mutex_lock(&fs_devices
->device_list_mutex
);
648 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
649 fs_devices
->num_devices
++;
650 mutex_unlock(&fs_devices
->device_list_mutex
);
653 device
->fs_devices
= fs_devices
;
654 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
656 * When FS is already mounted.
657 * 1. If you are here and if the device->name is NULL that
658 * means this device was missing at time of FS mount.
659 * 2. If you are here and if the device->name is different
660 * from 'path' that means either
661 * a. The same device disappeared and reappeared with
663 * b. The missing-disk-which-was-replaced, has
666 * We must allow 1 and 2a above. But 2b would be a spurious
669 * Further in case of 1 and 2a above, the disk at 'path'
670 * would have missed some transaction when it was away and
671 * in case of 2a the stale bdev has to be updated as well.
672 * 2b must not be allowed at all time.
676 * For now, we do allow update to btrfs_fs_device through the
677 * btrfs dev scan cli after FS has been mounted. We're still
678 * tracking a problem where systems fail mount by subvolume id
679 * when we reject replacement on a mounted FS.
681 if (!fs_devices
->opened
&& found_transid
< device
->generation
) {
683 * That is if the FS is _not_ mounted and if you
684 * are here, that means there is more than one
685 * disk with same uuid and devid.We keep the one
686 * with larger generation number or the last-in if
687 * generation are equal.
692 name
= rcu_string_strdup(path
, GFP_NOFS
);
695 rcu_string_free(device
->name
);
696 rcu_assign_pointer(device
->name
, name
);
697 if (device
->missing
) {
698 fs_devices
->missing_devices
--;
704 * Unmount does not free the btrfs_device struct but would zero
705 * generation along with most of the other members. So just update
706 * it back. We need it to pick the disk with largest generation
709 if (!fs_devices
->opened
)
710 device
->generation
= found_transid
;
713 * if there is new btrfs on an already registered device,
714 * then remove the stale device entry.
717 btrfs_free_stale_device(device
);
719 *fs_devices_ret
= fs_devices
;
724 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
726 struct btrfs_fs_devices
*fs_devices
;
727 struct btrfs_device
*device
;
728 struct btrfs_device
*orig_dev
;
730 fs_devices
= alloc_fs_devices(orig
->fsid
);
731 if (IS_ERR(fs_devices
))
734 mutex_lock(&orig
->device_list_mutex
);
735 fs_devices
->total_devices
= orig
->total_devices
;
737 /* We have held the volume lock, it is safe to get the devices. */
738 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
739 struct rcu_string
*name
;
741 device
= btrfs_alloc_device(NULL
, &orig_dev
->devid
,
747 * This is ok to do without rcu read locked because we hold the
748 * uuid mutex so nothing we touch in here is going to disappear.
750 if (orig_dev
->name
) {
751 name
= rcu_string_strdup(orig_dev
->name
->str
,
757 rcu_assign_pointer(device
->name
, name
);
760 list_add(&device
->dev_list
, &fs_devices
->devices
);
761 device
->fs_devices
= fs_devices
;
762 fs_devices
->num_devices
++;
764 mutex_unlock(&orig
->device_list_mutex
);
767 mutex_unlock(&orig
->device_list_mutex
);
768 free_fs_devices(fs_devices
);
769 return ERR_PTR(-ENOMEM
);
772 void btrfs_close_extra_devices(struct btrfs_fs_devices
*fs_devices
, int step
)
774 struct btrfs_device
*device
, *next
;
775 struct btrfs_device
*latest_dev
= NULL
;
777 mutex_lock(&uuid_mutex
);
779 /* This is the initialized path, it is safe to release the devices. */
780 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
781 if (device
->in_fs_metadata
) {
782 if (!device
->is_tgtdev_for_dev_replace
&&
784 device
->generation
> latest_dev
->generation
)) {
790 if (device
->devid
== BTRFS_DEV_REPLACE_DEVID
) {
792 * In the first step, keep the device which has
793 * the correct fsid and the devid that is used
794 * for the dev_replace procedure.
795 * In the second step, the dev_replace state is
796 * read from the device tree and it is known
797 * whether the procedure is really active or
798 * not, which means whether this device is
799 * used or whether it should be removed.
801 if (step
== 0 || device
->is_tgtdev_for_dev_replace
) {
806 blkdev_put(device
->bdev
, device
->mode
);
808 fs_devices
->open_devices
--;
810 if (device
->writeable
) {
811 list_del_init(&device
->dev_alloc_list
);
812 device
->writeable
= 0;
813 if (!device
->is_tgtdev_for_dev_replace
)
814 fs_devices
->rw_devices
--;
816 list_del_init(&device
->dev_list
);
817 fs_devices
->num_devices
--;
818 rcu_string_free(device
->name
);
822 if (fs_devices
->seed
) {
823 fs_devices
= fs_devices
->seed
;
827 fs_devices
->latest_bdev
= latest_dev
->bdev
;
829 mutex_unlock(&uuid_mutex
);
832 static void __free_device(struct work_struct
*work
)
834 struct btrfs_device
*device
;
836 device
= container_of(work
, struct btrfs_device
, rcu_work
);
837 rcu_string_free(device
->name
);
841 static void free_device(struct rcu_head
*head
)
843 struct btrfs_device
*device
;
845 device
= container_of(head
, struct btrfs_device
, rcu
);
847 INIT_WORK(&device
->rcu_work
, __free_device
);
848 schedule_work(&device
->rcu_work
);
851 static void btrfs_close_bdev(struct btrfs_device
*device
)
853 if (device
->bdev
&& device
->writeable
) {
854 sync_blockdev(device
->bdev
);
855 invalidate_bdev(device
->bdev
);
859 blkdev_put(device
->bdev
, device
->mode
);
862 static void btrfs_close_one_device(struct btrfs_device
*device
)
864 struct btrfs_fs_devices
*fs_devices
= device
->fs_devices
;
865 struct btrfs_device
*new_device
;
866 struct rcu_string
*name
;
869 fs_devices
->open_devices
--;
871 if (device
->writeable
&&
872 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
873 list_del_init(&device
->dev_alloc_list
);
874 fs_devices
->rw_devices
--;
878 fs_devices
->missing_devices
--;
880 btrfs_close_bdev(device
);
882 new_device
= btrfs_alloc_device(NULL
, &device
->devid
,
884 BUG_ON(IS_ERR(new_device
)); /* -ENOMEM */
886 /* Safe because we are under uuid_mutex */
888 name
= rcu_string_strdup(device
->name
->str
, GFP_NOFS
);
889 BUG_ON(!name
); /* -ENOMEM */
890 rcu_assign_pointer(new_device
->name
, name
);
893 list_replace_rcu(&device
->dev_list
, &new_device
->dev_list
);
894 new_device
->fs_devices
= device
->fs_devices
;
896 call_rcu(&device
->rcu
, free_device
);
899 static int __btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
901 struct btrfs_device
*device
, *tmp
;
903 if (--fs_devices
->opened
> 0)
906 mutex_lock(&fs_devices
->device_list_mutex
);
907 list_for_each_entry_safe(device
, tmp
, &fs_devices
->devices
, dev_list
) {
908 btrfs_close_one_device(device
);
910 mutex_unlock(&fs_devices
->device_list_mutex
);
912 WARN_ON(fs_devices
->open_devices
);
913 WARN_ON(fs_devices
->rw_devices
);
914 fs_devices
->opened
= 0;
915 fs_devices
->seeding
= 0;
920 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
922 struct btrfs_fs_devices
*seed_devices
= NULL
;
925 mutex_lock(&uuid_mutex
);
926 ret
= __btrfs_close_devices(fs_devices
);
927 if (!fs_devices
->opened
) {
928 seed_devices
= fs_devices
->seed
;
929 fs_devices
->seed
= NULL
;
931 mutex_unlock(&uuid_mutex
);
933 while (seed_devices
) {
934 fs_devices
= seed_devices
;
935 seed_devices
= fs_devices
->seed
;
936 __btrfs_close_devices(fs_devices
);
937 free_fs_devices(fs_devices
);
940 * Wait for rcu kworkers under __btrfs_close_devices
941 * to finish all blkdev_puts so device is really
942 * free when umount is done.
948 static int __btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
949 fmode_t flags
, void *holder
)
951 struct request_queue
*q
;
952 struct block_device
*bdev
;
953 struct list_head
*head
= &fs_devices
->devices
;
954 struct btrfs_device
*device
;
955 struct btrfs_device
*latest_dev
= NULL
;
956 struct buffer_head
*bh
;
957 struct btrfs_super_block
*disk_super
;
964 list_for_each_entry(device
, head
, dev_list
) {
970 /* Just open everything we can; ignore failures here */
971 if (btrfs_get_bdev_and_sb(device
->name
->str
, flags
, holder
, 1,
975 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
976 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
977 if (devid
!= device
->devid
)
980 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
,
984 device
->generation
= btrfs_super_generation(disk_super
);
986 device
->generation
> latest_dev
->generation
)
989 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
990 device
->writeable
= 0;
992 device
->writeable
= !bdev_read_only(bdev
);
996 q
= bdev_get_queue(bdev
);
997 if (blk_queue_discard(q
))
998 device
->can_discard
= 1;
1000 device
->bdev
= bdev
;
1001 device
->in_fs_metadata
= 0;
1002 device
->mode
= flags
;
1004 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
1005 fs_devices
->rotating
= 1;
1007 fs_devices
->open_devices
++;
1008 if (device
->writeable
&&
1009 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
1010 fs_devices
->rw_devices
++;
1011 list_add(&device
->dev_alloc_list
,
1012 &fs_devices
->alloc_list
);
1019 blkdev_put(bdev
, flags
);
1022 if (fs_devices
->open_devices
== 0) {
1026 fs_devices
->seeding
= seeding
;
1027 fs_devices
->opened
= 1;
1028 fs_devices
->latest_bdev
= latest_dev
->bdev
;
1029 fs_devices
->total_rw_bytes
= 0;
1034 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
1035 fmode_t flags
, void *holder
)
1039 mutex_lock(&uuid_mutex
);
1040 if (fs_devices
->opened
) {
1041 fs_devices
->opened
++;
1044 ret
= __btrfs_open_devices(fs_devices
, flags
, holder
);
1046 mutex_unlock(&uuid_mutex
);
1050 void btrfs_release_disk_super(struct page
*page
)
1056 int btrfs_read_disk_super(struct block_device
*bdev
, u64 bytenr
,
1057 struct page
**page
, struct btrfs_super_block
**disk_super
)
1062 /* make sure our super fits in the device */
1063 if (bytenr
+ PAGE_SIZE
>= i_size_read(bdev
->bd_inode
))
1066 /* make sure our super fits in the page */
1067 if (sizeof(**disk_super
) > PAGE_SIZE
)
1070 /* make sure our super doesn't straddle pages on disk */
1071 index
= bytenr
>> PAGE_SHIFT
;
1072 if ((bytenr
+ sizeof(**disk_super
) - 1) >> PAGE_SHIFT
!= index
)
1075 /* pull in the page with our super */
1076 *page
= read_cache_page_gfp(bdev
->bd_inode
->i_mapping
,
1079 if (IS_ERR_OR_NULL(*page
))
1084 /* align our pointer to the offset of the super block */
1085 *disk_super
= p
+ (bytenr
& ~PAGE_MASK
);
1087 if (btrfs_super_bytenr(*disk_super
) != bytenr
||
1088 btrfs_super_magic(*disk_super
) != BTRFS_MAGIC
) {
1089 btrfs_release_disk_super(*page
);
1093 if ((*disk_super
)->label
[0] &&
1094 (*disk_super
)->label
[BTRFS_LABEL_SIZE
- 1])
1095 (*disk_super
)->label
[BTRFS_LABEL_SIZE
- 1] = '\0';
1101 * Look for a btrfs signature on a device. This may be called out of the mount path
1102 * and we are not allowed to call set_blocksize during the scan. The superblock
1103 * is read via pagecache
1105 int btrfs_scan_one_device(const char *path
, fmode_t flags
, void *holder
,
1106 struct btrfs_fs_devices
**fs_devices_ret
)
1108 struct btrfs_super_block
*disk_super
;
1109 struct block_device
*bdev
;
1118 * we would like to check all the supers, but that would make
1119 * a btrfs mount succeed after a mkfs from a different FS.
1120 * So, we need to add a special mount option to scan for
1121 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1123 bytenr
= btrfs_sb_offset(0);
1124 flags
|= FMODE_EXCL
;
1125 mutex_lock(&uuid_mutex
);
1127 bdev
= blkdev_get_by_path(path
, flags
, holder
);
1129 ret
= PTR_ERR(bdev
);
1133 if (btrfs_read_disk_super(bdev
, bytenr
, &page
, &disk_super
))
1134 goto error_bdev_put
;
1136 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1137 transid
= btrfs_super_generation(disk_super
);
1138 total_devices
= btrfs_super_num_devices(disk_super
);
1140 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
1142 if (disk_super
->label
[0]) {
1143 printk(KERN_INFO
"BTRFS: device label %s ", disk_super
->label
);
1145 printk(KERN_INFO
"BTRFS: device fsid %pU ", disk_super
->fsid
);
1148 printk(KERN_CONT
"devid %llu transid %llu %s\n", devid
, transid
, path
);
1151 if (!ret
&& fs_devices_ret
)
1152 (*fs_devices_ret
)->total_devices
= total_devices
;
1154 btrfs_release_disk_super(page
);
1157 blkdev_put(bdev
, flags
);
1159 mutex_unlock(&uuid_mutex
);
1163 /* helper to account the used device space in the range */
1164 int btrfs_account_dev_extents_size(struct btrfs_device
*device
, u64 start
,
1165 u64 end
, u64
*length
)
1167 struct btrfs_key key
;
1168 struct btrfs_root
*root
= device
->dev_root
;
1169 struct btrfs_dev_extent
*dev_extent
;
1170 struct btrfs_path
*path
;
1174 struct extent_buffer
*l
;
1178 if (start
>= device
->total_bytes
|| device
->is_tgtdev_for_dev_replace
)
1181 path
= btrfs_alloc_path();
1184 path
->reada
= READA_FORWARD
;
1186 key
.objectid
= device
->devid
;
1188 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1190 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1194 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1201 slot
= path
->slots
[0];
1202 if (slot
>= btrfs_header_nritems(l
)) {
1203 ret
= btrfs_next_leaf(root
, path
);
1211 btrfs_item_key_to_cpu(l
, &key
, slot
);
1213 if (key
.objectid
< device
->devid
)
1216 if (key
.objectid
> device
->devid
)
1219 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1222 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1223 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1225 if (key
.offset
<= start
&& extent_end
> end
) {
1226 *length
= end
- start
+ 1;
1228 } else if (key
.offset
<= start
&& extent_end
> start
)
1229 *length
+= extent_end
- start
;
1230 else if (key
.offset
> start
&& extent_end
<= end
)
1231 *length
+= extent_end
- key
.offset
;
1232 else if (key
.offset
> start
&& key
.offset
<= end
) {
1233 *length
+= end
- key
.offset
+ 1;
1235 } else if (key
.offset
> end
)
1243 btrfs_free_path(path
);
1247 static int contains_pending_extent(struct btrfs_transaction
*transaction
,
1248 struct btrfs_device
*device
,
1249 u64
*start
, u64 len
)
1251 struct btrfs_fs_info
*fs_info
= device
->dev_root
->fs_info
;
1252 struct extent_map
*em
;
1253 struct list_head
*search_list
= &fs_info
->pinned_chunks
;
1255 u64 physical_start
= *start
;
1258 search_list
= &transaction
->pending_chunks
;
1260 list_for_each_entry(em
, search_list
, list
) {
1261 struct map_lookup
*map
;
1264 map
= em
->map_lookup
;
1265 for (i
= 0; i
< map
->num_stripes
; i
++) {
1268 if (map
->stripes
[i
].dev
!= device
)
1270 if (map
->stripes
[i
].physical
>= physical_start
+ len
||
1271 map
->stripes
[i
].physical
+ em
->orig_block_len
<=
1275 * Make sure that while processing the pinned list we do
1276 * not override our *start with a lower value, because
1277 * we can have pinned chunks that fall within this
1278 * device hole and that have lower physical addresses
1279 * than the pending chunks we processed before. If we
1280 * do not take this special care we can end up getting
1281 * 2 pending chunks that start at the same physical
1282 * device offsets because the end offset of a pinned
1283 * chunk can be equal to the start offset of some
1286 end
= map
->stripes
[i
].physical
+ em
->orig_block_len
;
1293 if (search_list
!= &fs_info
->pinned_chunks
) {
1294 search_list
= &fs_info
->pinned_chunks
;
1303 * find_free_dev_extent_start - find free space in the specified device
1304 * @device: the device which we search the free space in
1305 * @num_bytes: the size of the free space that we need
1306 * @search_start: the position from which to begin the search
1307 * @start: store the start of the free space.
1308 * @len: the size of the free space. that we find, or the size
1309 * of the max free space if we don't find suitable free space
1311 * this uses a pretty simple search, the expectation is that it is
1312 * called very infrequently and that a given device has a small number
1315 * @start is used to store the start of the free space if we find. But if we
1316 * don't find suitable free space, it will be used to store the start position
1317 * of the max free space.
1319 * @len is used to store the size of the free space that we find.
1320 * But if we don't find suitable free space, it is used to store the size of
1321 * the max free space.
1323 int find_free_dev_extent_start(struct btrfs_transaction
*transaction
,
1324 struct btrfs_device
*device
, u64 num_bytes
,
1325 u64 search_start
, u64
*start
, u64
*len
)
1327 struct btrfs_key key
;
1328 struct btrfs_root
*root
= device
->dev_root
;
1329 struct btrfs_dev_extent
*dev_extent
;
1330 struct btrfs_path
*path
;
1335 u64 search_end
= device
->total_bytes
;
1338 struct extent_buffer
*l
;
1339 u64 min_search_start
;
1342 * We don't want to overwrite the superblock on the drive nor any area
1343 * used by the boot loader (grub for example), so we make sure to start
1344 * at an offset of at least 1MB.
1346 min_search_start
= max(root
->fs_info
->alloc_start
, 1024ull * 1024);
1347 search_start
= max(search_start
, min_search_start
);
1349 path
= btrfs_alloc_path();
1353 max_hole_start
= search_start
;
1357 if (search_start
>= search_end
|| device
->is_tgtdev_for_dev_replace
) {
1362 path
->reada
= READA_FORWARD
;
1363 path
->search_commit_root
= 1;
1364 path
->skip_locking
= 1;
1366 key
.objectid
= device
->devid
;
1367 key
.offset
= search_start
;
1368 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1370 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1374 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1381 slot
= path
->slots
[0];
1382 if (slot
>= btrfs_header_nritems(l
)) {
1383 ret
= btrfs_next_leaf(root
, path
);
1391 btrfs_item_key_to_cpu(l
, &key
, slot
);
1393 if (key
.objectid
< device
->devid
)
1396 if (key
.objectid
> device
->devid
)
1399 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1402 if (key
.offset
> search_start
) {
1403 hole_size
= key
.offset
- search_start
;
1406 * Have to check before we set max_hole_start, otherwise
1407 * we could end up sending back this offset anyway.
1409 if (contains_pending_extent(transaction
, device
,
1412 if (key
.offset
>= search_start
) {
1413 hole_size
= key
.offset
- search_start
;
1420 if (hole_size
> max_hole_size
) {
1421 max_hole_start
= search_start
;
1422 max_hole_size
= hole_size
;
1426 * If this free space is greater than which we need,
1427 * it must be the max free space that we have found
1428 * until now, so max_hole_start must point to the start
1429 * of this free space and the length of this free space
1430 * is stored in max_hole_size. Thus, we return
1431 * max_hole_start and max_hole_size and go back to the
1434 if (hole_size
>= num_bytes
) {
1440 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1441 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1443 if (extent_end
> search_start
)
1444 search_start
= extent_end
;
1451 * At this point, search_start should be the end of
1452 * allocated dev extents, and when shrinking the device,
1453 * search_end may be smaller than search_start.
1455 if (search_end
> search_start
) {
1456 hole_size
= search_end
- search_start
;
1458 if (contains_pending_extent(transaction
, device
, &search_start
,
1460 btrfs_release_path(path
);
1464 if (hole_size
> max_hole_size
) {
1465 max_hole_start
= search_start
;
1466 max_hole_size
= hole_size
;
1471 if (max_hole_size
< num_bytes
)
1477 btrfs_free_path(path
);
1478 *start
= max_hole_start
;
1480 *len
= max_hole_size
;
1484 int find_free_dev_extent(struct btrfs_trans_handle
*trans
,
1485 struct btrfs_device
*device
, u64 num_bytes
,
1486 u64
*start
, u64
*len
)
1488 /* FIXME use last free of some kind */
1489 return find_free_dev_extent_start(trans
->transaction
, device
,
1490 num_bytes
, 0, start
, len
);
1493 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1494 struct btrfs_device
*device
,
1495 u64 start
, u64
*dev_extent_len
)
1498 struct btrfs_path
*path
;
1499 struct btrfs_root
*root
= device
->dev_root
;
1500 struct btrfs_key key
;
1501 struct btrfs_key found_key
;
1502 struct extent_buffer
*leaf
= NULL
;
1503 struct btrfs_dev_extent
*extent
= NULL
;
1505 path
= btrfs_alloc_path();
1509 key
.objectid
= device
->devid
;
1511 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1513 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1515 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1516 BTRFS_DEV_EXTENT_KEY
);
1519 leaf
= path
->nodes
[0];
1520 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1521 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1522 struct btrfs_dev_extent
);
1523 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1524 btrfs_dev_extent_length(leaf
, extent
) < start
);
1526 btrfs_release_path(path
);
1528 } else if (ret
== 0) {
1529 leaf
= path
->nodes
[0];
1530 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1531 struct btrfs_dev_extent
);
1533 btrfs_handle_fs_error(root
->fs_info
, ret
, "Slot search failed");
1537 *dev_extent_len
= btrfs_dev_extent_length(leaf
, extent
);
1539 ret
= btrfs_del_item(trans
, root
, path
);
1541 btrfs_handle_fs_error(root
->fs_info
, ret
,
1542 "Failed to remove dev extent item");
1544 set_bit(BTRFS_TRANS_HAVE_FREE_BGS
, &trans
->transaction
->flags
);
1547 btrfs_free_path(path
);
1551 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1552 struct btrfs_device
*device
,
1553 u64 chunk_tree
, u64 chunk_objectid
,
1554 u64 chunk_offset
, u64 start
, u64 num_bytes
)
1557 struct btrfs_path
*path
;
1558 struct btrfs_root
*root
= device
->dev_root
;
1559 struct btrfs_dev_extent
*extent
;
1560 struct extent_buffer
*leaf
;
1561 struct btrfs_key key
;
1563 WARN_ON(!device
->in_fs_metadata
);
1564 WARN_ON(device
->is_tgtdev_for_dev_replace
);
1565 path
= btrfs_alloc_path();
1569 key
.objectid
= device
->devid
;
1571 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1572 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1577 leaf
= path
->nodes
[0];
1578 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1579 struct btrfs_dev_extent
);
1580 btrfs_set_dev_extent_chunk_tree(leaf
, extent
, chunk_tree
);
1581 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
, chunk_objectid
);
1582 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1584 write_extent_buffer(leaf
, root
->fs_info
->chunk_tree_uuid
,
1585 btrfs_dev_extent_chunk_tree_uuid(extent
), BTRFS_UUID_SIZE
);
1587 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1588 btrfs_mark_buffer_dirty(leaf
);
1590 btrfs_free_path(path
);
1594 static u64
find_next_chunk(struct btrfs_fs_info
*fs_info
)
1596 struct extent_map_tree
*em_tree
;
1597 struct extent_map
*em
;
1601 em_tree
= &fs_info
->mapping_tree
.map_tree
;
1602 read_lock(&em_tree
->lock
);
1603 n
= rb_last(&em_tree
->map
);
1605 em
= rb_entry(n
, struct extent_map
, rb_node
);
1606 ret
= em
->start
+ em
->len
;
1608 read_unlock(&em_tree
->lock
);
1613 static noinline
int find_next_devid(struct btrfs_fs_info
*fs_info
,
1617 struct btrfs_key key
;
1618 struct btrfs_key found_key
;
1619 struct btrfs_path
*path
;
1621 path
= btrfs_alloc_path();
1625 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1626 key
.type
= BTRFS_DEV_ITEM_KEY
;
1627 key
.offset
= (u64
)-1;
1629 ret
= btrfs_search_slot(NULL
, fs_info
->chunk_root
, &key
, path
, 0, 0);
1633 BUG_ON(ret
== 0); /* Corruption */
1635 ret
= btrfs_previous_item(fs_info
->chunk_root
, path
,
1636 BTRFS_DEV_ITEMS_OBJECTID
,
1637 BTRFS_DEV_ITEM_KEY
);
1641 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1643 *devid_ret
= found_key
.offset
+ 1;
1647 btrfs_free_path(path
);
1652 * the device information is stored in the chunk root
1653 * the btrfs_device struct should be fully filled in
1655 static int btrfs_add_device(struct btrfs_trans_handle
*trans
,
1656 struct btrfs_root
*root
,
1657 struct btrfs_device
*device
)
1660 struct btrfs_path
*path
;
1661 struct btrfs_dev_item
*dev_item
;
1662 struct extent_buffer
*leaf
;
1663 struct btrfs_key key
;
1666 root
= root
->fs_info
->chunk_root
;
1668 path
= btrfs_alloc_path();
1672 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1673 key
.type
= BTRFS_DEV_ITEM_KEY
;
1674 key
.offset
= device
->devid
;
1676 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1681 leaf
= path
->nodes
[0];
1682 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1684 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1685 btrfs_set_device_generation(leaf
, dev_item
, 0);
1686 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1687 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1688 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1689 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1690 btrfs_set_device_total_bytes(leaf
, dev_item
,
1691 btrfs_device_get_disk_total_bytes(device
));
1692 btrfs_set_device_bytes_used(leaf
, dev_item
,
1693 btrfs_device_get_bytes_used(device
));
1694 btrfs_set_device_group(leaf
, dev_item
, 0);
1695 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1696 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1697 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1699 ptr
= btrfs_device_uuid(dev_item
);
1700 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1701 ptr
= btrfs_device_fsid(dev_item
);
1702 write_extent_buffer(leaf
, root
->fs_info
->fsid
, ptr
, BTRFS_UUID_SIZE
);
1703 btrfs_mark_buffer_dirty(leaf
);
1707 btrfs_free_path(path
);
1712 * Function to update ctime/mtime for a given device path.
1713 * Mainly used for ctime/mtime based probe like libblkid.
1715 static void update_dev_time(char *path_name
)
1719 filp
= filp_open(path_name
, O_RDWR
, 0);
1722 file_update_time(filp
);
1723 filp_close(filp
, NULL
);
1726 static int btrfs_rm_dev_item(struct btrfs_root
*root
,
1727 struct btrfs_device
*device
)
1730 struct btrfs_path
*path
;
1731 struct btrfs_key key
;
1732 struct btrfs_trans_handle
*trans
;
1734 root
= root
->fs_info
->chunk_root
;
1736 path
= btrfs_alloc_path();
1740 trans
= btrfs_start_transaction(root
, 0);
1741 if (IS_ERR(trans
)) {
1742 btrfs_free_path(path
);
1743 return PTR_ERR(trans
);
1745 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1746 key
.type
= BTRFS_DEV_ITEM_KEY
;
1747 key
.offset
= device
->devid
;
1749 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1758 ret
= btrfs_del_item(trans
, root
, path
);
1762 btrfs_free_path(path
);
1763 btrfs_commit_transaction(trans
, root
);
1768 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1769 * filesystem. It's up to the caller to adjust that number regarding eg. device
1772 static int btrfs_check_raid_min_devices(struct btrfs_fs_info
*fs_info
,
1780 seq
= read_seqbegin(&fs_info
->profiles_lock
);
1782 all_avail
= fs_info
->avail_data_alloc_bits
|
1783 fs_info
->avail_system_alloc_bits
|
1784 fs_info
->avail_metadata_alloc_bits
;
1785 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
1787 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
1788 if (!(all_avail
& btrfs_raid_group
[i
]))
1791 if (num_devices
< btrfs_raid_array
[i
].devs_min
) {
1792 int ret
= btrfs_raid_mindev_error
[i
];
1802 struct btrfs_device
*btrfs_find_next_active_device(struct btrfs_fs_devices
*fs_devs
,
1803 struct btrfs_device
*device
)
1805 struct btrfs_device
*next_device
;
1807 list_for_each_entry(next_device
, &fs_devs
->devices
, dev_list
) {
1808 if (next_device
!= device
&&
1809 !next_device
->missing
&& next_device
->bdev
)
1817 * Helper function to check if the given device is part of s_bdev / latest_bdev
1818 * and replace it with the provided or the next active device, in the context
1819 * where this function called, there should be always be another device (or
1820 * this_dev) which is active.
1822 void btrfs_assign_next_active_device(struct btrfs_fs_info
*fs_info
,
1823 struct btrfs_device
*device
, struct btrfs_device
*this_dev
)
1825 struct btrfs_device
*next_device
;
1828 next_device
= this_dev
;
1830 next_device
= btrfs_find_next_active_device(fs_info
->fs_devices
,
1832 ASSERT(next_device
);
1834 if (fs_info
->sb
->s_bdev
&&
1835 (fs_info
->sb
->s_bdev
== device
->bdev
))
1836 fs_info
->sb
->s_bdev
= next_device
->bdev
;
1838 if (fs_info
->fs_devices
->latest_bdev
== device
->bdev
)
1839 fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1842 int btrfs_rm_device(struct btrfs_root
*root
, char *device_path
, u64 devid
)
1844 struct btrfs_device
*device
;
1845 struct btrfs_fs_devices
*cur_devices
;
1848 bool clear_super
= false;
1849 char *dev_name
= NULL
;
1851 mutex_lock(&uuid_mutex
);
1853 num_devices
= root
->fs_info
->fs_devices
->num_devices
;
1854 btrfs_dev_replace_lock(&root
->fs_info
->dev_replace
, 0);
1855 if (btrfs_dev_replace_is_ongoing(&root
->fs_info
->dev_replace
)) {
1856 WARN_ON(num_devices
< 1);
1859 btrfs_dev_replace_unlock(&root
->fs_info
->dev_replace
, 0);
1861 ret
= btrfs_check_raid_min_devices(root
->fs_info
, num_devices
- 1);
1865 ret
= btrfs_find_device_by_devspec(root
, devid
, device_path
,
1870 if (device
->is_tgtdev_for_dev_replace
) {
1871 ret
= BTRFS_ERROR_DEV_TGT_REPLACE
;
1875 if (device
->writeable
&& root
->fs_info
->fs_devices
->rw_devices
== 1) {
1876 ret
= BTRFS_ERROR_DEV_ONLY_WRITABLE
;
1880 if (device
->writeable
) {
1882 list_del_init(&device
->dev_alloc_list
);
1883 device
->fs_devices
->rw_devices
--;
1884 unlock_chunks(root
);
1885 dev_name
= kstrdup(device
->name
->str
, GFP_KERNEL
);
1893 mutex_unlock(&uuid_mutex
);
1894 ret
= btrfs_shrink_device(device
, 0);
1895 mutex_lock(&uuid_mutex
);
1900 * TODO: the superblock still includes this device in its num_devices
1901 * counter although write_all_supers() is not locked out. This
1902 * could give a filesystem state which requires a degraded mount.
1904 ret
= btrfs_rm_dev_item(root
->fs_info
->chunk_root
, device
);
1908 device
->in_fs_metadata
= 0;
1909 btrfs_scrub_cancel_dev(root
->fs_info
, device
);
1912 * the device list mutex makes sure that we don't change
1913 * the device list while someone else is writing out all
1914 * the device supers. Whoever is writing all supers, should
1915 * lock the device list mutex before getting the number of
1916 * devices in the super block (super_copy). Conversely,
1917 * whoever updates the number of devices in the super block
1918 * (super_copy) should hold the device list mutex.
1921 cur_devices
= device
->fs_devices
;
1922 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1923 list_del_rcu(&device
->dev_list
);
1925 device
->fs_devices
->num_devices
--;
1926 device
->fs_devices
->total_devices
--;
1928 if (device
->missing
)
1929 device
->fs_devices
->missing_devices
--;
1931 btrfs_assign_next_active_device(root
->fs_info
, device
, NULL
);
1934 device
->fs_devices
->open_devices
--;
1935 /* remove sysfs entry */
1936 btrfs_sysfs_rm_device_link(root
->fs_info
->fs_devices
, device
);
1939 btrfs_close_bdev(device
);
1941 call_rcu(&device
->rcu
, free_device
);
1943 num_devices
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
1944 btrfs_set_super_num_devices(root
->fs_info
->super_copy
, num_devices
);
1945 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1947 if (cur_devices
->open_devices
== 0) {
1948 struct btrfs_fs_devices
*fs_devices
;
1949 fs_devices
= root
->fs_info
->fs_devices
;
1950 while (fs_devices
) {
1951 if (fs_devices
->seed
== cur_devices
) {
1952 fs_devices
->seed
= cur_devices
->seed
;
1955 fs_devices
= fs_devices
->seed
;
1957 cur_devices
->seed
= NULL
;
1958 __btrfs_close_devices(cur_devices
);
1959 free_fs_devices(cur_devices
);
1962 root
->fs_info
->num_tolerated_disk_barrier_failures
=
1963 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
1966 * at this point, the device is zero sized. We want to
1967 * remove it from the devices list and zero out the old super
1970 struct block_device
*bdev
;
1972 bdev
= blkdev_get_by_path(dev_name
, FMODE_READ
| FMODE_EXCL
,
1973 root
->fs_info
->bdev_holder
);
1974 if (!IS_ERR(bdev
)) {
1975 btrfs_scratch_superblocks(bdev
, dev_name
);
1976 blkdev_put(bdev
, FMODE_READ
| FMODE_EXCL
);
1983 mutex_unlock(&uuid_mutex
);
1987 if (device
->writeable
) {
1989 list_add(&device
->dev_alloc_list
,
1990 &root
->fs_info
->fs_devices
->alloc_list
);
1991 device
->fs_devices
->rw_devices
++;
1992 unlock_chunks(root
);
1997 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info
*fs_info
,
1998 struct btrfs_device
*srcdev
)
2000 struct btrfs_fs_devices
*fs_devices
;
2002 WARN_ON(!mutex_is_locked(&fs_info
->fs_devices
->device_list_mutex
));
2005 * in case of fs with no seed, srcdev->fs_devices will point
2006 * to fs_devices of fs_info. However when the dev being replaced is
2007 * a seed dev it will point to the seed's local fs_devices. In short
2008 * srcdev will have its correct fs_devices in both the cases.
2010 fs_devices
= srcdev
->fs_devices
;
2012 list_del_rcu(&srcdev
->dev_list
);
2013 list_del_rcu(&srcdev
->dev_alloc_list
);
2014 fs_devices
->num_devices
--;
2015 if (srcdev
->missing
)
2016 fs_devices
->missing_devices
--;
2018 if (srcdev
->writeable
)
2019 fs_devices
->rw_devices
--;
2022 fs_devices
->open_devices
--;
2025 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info
*fs_info
,
2026 struct btrfs_device
*srcdev
)
2028 struct btrfs_fs_devices
*fs_devices
= srcdev
->fs_devices
;
2030 if (srcdev
->writeable
) {
2031 /* zero out the old super if it is writable */
2032 btrfs_scratch_superblocks(srcdev
->bdev
, srcdev
->name
->str
);
2035 btrfs_close_bdev(srcdev
);
2037 call_rcu(&srcdev
->rcu
, free_device
);
2040 * unless fs_devices is seed fs, num_devices shouldn't go
2043 BUG_ON(!fs_devices
->num_devices
&& !fs_devices
->seeding
);
2045 /* if this is no devs we rather delete the fs_devices */
2046 if (!fs_devices
->num_devices
) {
2047 struct btrfs_fs_devices
*tmp_fs_devices
;
2049 tmp_fs_devices
= fs_info
->fs_devices
;
2050 while (tmp_fs_devices
) {
2051 if (tmp_fs_devices
->seed
== fs_devices
) {
2052 tmp_fs_devices
->seed
= fs_devices
->seed
;
2055 tmp_fs_devices
= tmp_fs_devices
->seed
;
2057 fs_devices
->seed
= NULL
;
2058 __btrfs_close_devices(fs_devices
);
2059 free_fs_devices(fs_devices
);
2063 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
2064 struct btrfs_device
*tgtdev
)
2066 mutex_lock(&uuid_mutex
);
2068 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2070 btrfs_sysfs_rm_device_link(fs_info
->fs_devices
, tgtdev
);
2073 fs_info
->fs_devices
->open_devices
--;
2075 fs_info
->fs_devices
->num_devices
--;
2077 btrfs_assign_next_active_device(fs_info
, tgtdev
, NULL
);
2079 list_del_rcu(&tgtdev
->dev_list
);
2081 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2082 mutex_unlock(&uuid_mutex
);
2085 * The update_dev_time() with in btrfs_scratch_superblocks()
2086 * may lead to a call to btrfs_show_devname() which will try
2087 * to hold device_list_mutex. And here this device
2088 * is already out of device list, so we don't have to hold
2089 * the device_list_mutex lock.
2091 btrfs_scratch_superblocks(tgtdev
->bdev
, tgtdev
->name
->str
);
2093 btrfs_close_bdev(tgtdev
);
2094 call_rcu(&tgtdev
->rcu
, free_device
);
2097 static int btrfs_find_device_by_path(struct btrfs_root
*root
, char *device_path
,
2098 struct btrfs_device
**device
)
2101 struct btrfs_super_block
*disk_super
;
2104 struct block_device
*bdev
;
2105 struct buffer_head
*bh
;
2108 ret
= btrfs_get_bdev_and_sb(device_path
, FMODE_READ
,
2109 root
->fs_info
->bdev_holder
, 0, &bdev
, &bh
);
2112 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
2113 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
2114 dev_uuid
= disk_super
->dev_item
.uuid
;
2115 *device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
2120 blkdev_put(bdev
, FMODE_READ
);
2124 int btrfs_find_device_missing_or_by_path(struct btrfs_root
*root
,
2126 struct btrfs_device
**device
)
2129 if (strcmp(device_path
, "missing") == 0) {
2130 struct list_head
*devices
;
2131 struct btrfs_device
*tmp
;
2133 devices
= &root
->fs_info
->fs_devices
->devices
;
2135 * It is safe to read the devices since the volume_mutex
2136 * is held by the caller.
2138 list_for_each_entry(tmp
, devices
, dev_list
) {
2139 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
2146 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND
;
2150 return btrfs_find_device_by_path(root
, device_path
, device
);
2155 * Lookup a device given by device id, or the path if the id is 0.
2157 int btrfs_find_device_by_devspec(struct btrfs_root
*root
, u64 devid
,
2159 struct btrfs_device
**device
)
2165 *device
= btrfs_find_device(root
->fs_info
, devid
, NULL
,
2170 if (!devpath
|| !devpath
[0])
2173 ret
= btrfs_find_device_missing_or_by_path(root
, devpath
,
2180 * does all the dirty work required for changing file system's UUID.
2182 static int btrfs_prepare_sprout(struct btrfs_root
*root
)
2184 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
2185 struct btrfs_fs_devices
*old_devices
;
2186 struct btrfs_fs_devices
*seed_devices
;
2187 struct btrfs_super_block
*disk_super
= root
->fs_info
->super_copy
;
2188 struct btrfs_device
*device
;
2191 BUG_ON(!mutex_is_locked(&uuid_mutex
));
2192 if (!fs_devices
->seeding
)
2195 seed_devices
= __alloc_fs_devices();
2196 if (IS_ERR(seed_devices
))
2197 return PTR_ERR(seed_devices
);
2199 old_devices
= clone_fs_devices(fs_devices
);
2200 if (IS_ERR(old_devices
)) {
2201 kfree(seed_devices
);
2202 return PTR_ERR(old_devices
);
2205 list_add(&old_devices
->list
, &fs_uuids
);
2207 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
2208 seed_devices
->opened
= 1;
2209 INIT_LIST_HEAD(&seed_devices
->devices
);
2210 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
2211 mutex_init(&seed_devices
->device_list_mutex
);
2213 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2214 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
2216 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
)
2217 device
->fs_devices
= seed_devices
;
2220 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
2221 unlock_chunks(root
);
2223 fs_devices
->seeding
= 0;
2224 fs_devices
->num_devices
= 0;
2225 fs_devices
->open_devices
= 0;
2226 fs_devices
->missing_devices
= 0;
2227 fs_devices
->rotating
= 0;
2228 fs_devices
->seed
= seed_devices
;
2230 generate_random_uuid(fs_devices
->fsid
);
2231 memcpy(root
->fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2232 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2233 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2235 super_flags
= btrfs_super_flags(disk_super
) &
2236 ~BTRFS_SUPER_FLAG_SEEDING
;
2237 btrfs_set_super_flags(disk_super
, super_flags
);
2243 * Store the expected generation for seed devices in device items.
2245 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
2246 struct btrfs_root
*root
)
2248 struct btrfs_path
*path
;
2249 struct extent_buffer
*leaf
;
2250 struct btrfs_dev_item
*dev_item
;
2251 struct btrfs_device
*device
;
2252 struct btrfs_key key
;
2253 u8 fs_uuid
[BTRFS_UUID_SIZE
];
2254 u8 dev_uuid
[BTRFS_UUID_SIZE
];
2258 path
= btrfs_alloc_path();
2262 root
= root
->fs_info
->chunk_root
;
2263 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2265 key
.type
= BTRFS_DEV_ITEM_KEY
;
2268 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2272 leaf
= path
->nodes
[0];
2274 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
2275 ret
= btrfs_next_leaf(root
, path
);
2280 leaf
= path
->nodes
[0];
2281 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2282 btrfs_release_path(path
);
2286 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2287 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
2288 key
.type
!= BTRFS_DEV_ITEM_KEY
)
2291 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2292 struct btrfs_dev_item
);
2293 devid
= btrfs_device_id(leaf
, dev_item
);
2294 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
2296 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
2298 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
2300 BUG_ON(!device
); /* Logic error */
2302 if (device
->fs_devices
->seeding
) {
2303 btrfs_set_device_generation(leaf
, dev_item
,
2304 device
->generation
);
2305 btrfs_mark_buffer_dirty(leaf
);
2313 btrfs_free_path(path
);
2317 int btrfs_init_new_device(struct btrfs_root
*root
, char *device_path
)
2319 struct request_queue
*q
;
2320 struct btrfs_trans_handle
*trans
;
2321 struct btrfs_device
*device
;
2322 struct block_device
*bdev
;
2323 struct list_head
*devices
;
2324 struct super_block
*sb
= root
->fs_info
->sb
;
2325 struct rcu_string
*name
;
2327 int seeding_dev
= 0;
2330 if ((sb
->s_flags
& MS_RDONLY
) && !root
->fs_info
->fs_devices
->seeding
)
2333 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2334 root
->fs_info
->bdev_holder
);
2336 return PTR_ERR(bdev
);
2338 if (root
->fs_info
->fs_devices
->seeding
) {
2340 down_write(&sb
->s_umount
);
2341 mutex_lock(&uuid_mutex
);
2344 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2346 devices
= &root
->fs_info
->fs_devices
->devices
;
2348 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2349 list_for_each_entry(device
, devices
, dev_list
) {
2350 if (device
->bdev
== bdev
) {
2353 &root
->fs_info
->fs_devices
->device_list_mutex
);
2357 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2359 device
= btrfs_alloc_device(root
->fs_info
, NULL
, NULL
);
2360 if (IS_ERR(device
)) {
2361 /* we can safely leave the fs_devices entry around */
2362 ret
= PTR_ERR(device
);
2366 name
= rcu_string_strdup(device_path
, GFP_KERNEL
);
2372 rcu_assign_pointer(device
->name
, name
);
2374 trans
= btrfs_start_transaction(root
, 0);
2375 if (IS_ERR(trans
)) {
2376 rcu_string_free(device
->name
);
2378 ret
= PTR_ERR(trans
);
2382 q
= bdev_get_queue(bdev
);
2383 if (blk_queue_discard(q
))
2384 device
->can_discard
= 1;
2385 device
->writeable
= 1;
2386 device
->generation
= trans
->transid
;
2387 device
->io_width
= root
->sectorsize
;
2388 device
->io_align
= root
->sectorsize
;
2389 device
->sector_size
= root
->sectorsize
;
2390 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
2391 device
->disk_total_bytes
= device
->total_bytes
;
2392 device
->commit_total_bytes
= device
->total_bytes
;
2393 device
->dev_root
= root
->fs_info
->dev_root
;
2394 device
->bdev
= bdev
;
2395 device
->in_fs_metadata
= 1;
2396 device
->is_tgtdev_for_dev_replace
= 0;
2397 device
->mode
= FMODE_EXCL
;
2398 device
->dev_stats_valid
= 1;
2399 set_blocksize(device
->bdev
, 4096);
2402 sb
->s_flags
&= ~MS_RDONLY
;
2403 ret
= btrfs_prepare_sprout(root
);
2404 BUG_ON(ret
); /* -ENOMEM */
2407 device
->fs_devices
= root
->fs_info
->fs_devices
;
2409 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2411 list_add_rcu(&device
->dev_list
, &root
->fs_info
->fs_devices
->devices
);
2412 list_add(&device
->dev_alloc_list
,
2413 &root
->fs_info
->fs_devices
->alloc_list
);
2414 root
->fs_info
->fs_devices
->num_devices
++;
2415 root
->fs_info
->fs_devices
->open_devices
++;
2416 root
->fs_info
->fs_devices
->rw_devices
++;
2417 root
->fs_info
->fs_devices
->total_devices
++;
2418 root
->fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
2420 spin_lock(&root
->fs_info
->free_chunk_lock
);
2421 root
->fs_info
->free_chunk_space
+= device
->total_bytes
;
2422 spin_unlock(&root
->fs_info
->free_chunk_lock
);
2424 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
2425 root
->fs_info
->fs_devices
->rotating
= 1;
2427 tmp
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
2428 btrfs_set_super_total_bytes(root
->fs_info
->super_copy
,
2429 tmp
+ device
->total_bytes
);
2431 tmp
= btrfs_super_num_devices(root
->fs_info
->super_copy
);
2432 btrfs_set_super_num_devices(root
->fs_info
->super_copy
,
2435 /* add sysfs device entry */
2436 btrfs_sysfs_add_device_link(root
->fs_info
->fs_devices
, device
);
2439 * we've got more storage, clear any full flags on the space
2442 btrfs_clear_space_info_full(root
->fs_info
);
2444 unlock_chunks(root
);
2445 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2449 ret
= init_first_rw_device(trans
, root
, device
);
2450 unlock_chunks(root
);
2452 btrfs_abort_transaction(trans
, ret
);
2457 ret
= btrfs_add_device(trans
, root
, device
);
2459 btrfs_abort_transaction(trans
, ret
);
2464 char fsid_buf
[BTRFS_UUID_UNPARSED_SIZE
];
2466 ret
= btrfs_finish_sprout(trans
, root
);
2468 btrfs_abort_transaction(trans
, ret
);
2472 /* Sprouting would change fsid of the mounted root,
2473 * so rename the fsid on the sysfs
2475 snprintf(fsid_buf
, BTRFS_UUID_UNPARSED_SIZE
, "%pU",
2476 root
->fs_info
->fsid
);
2477 if (kobject_rename(&root
->fs_info
->fs_devices
->fsid_kobj
,
2479 btrfs_warn(root
->fs_info
,
2480 "sysfs: failed to create fsid for sprout");
2483 root
->fs_info
->num_tolerated_disk_barrier_failures
=
2484 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
2485 ret
= btrfs_commit_transaction(trans
, root
);
2488 mutex_unlock(&uuid_mutex
);
2489 up_write(&sb
->s_umount
);
2491 if (ret
) /* transaction commit */
2494 ret
= btrfs_relocate_sys_chunks(root
);
2496 btrfs_handle_fs_error(root
->fs_info
, ret
,
2497 "Failed to relocate sys chunks after "
2498 "device initialization. This can be fixed "
2499 "using the \"btrfs balance\" command.");
2500 trans
= btrfs_attach_transaction(root
);
2501 if (IS_ERR(trans
)) {
2502 if (PTR_ERR(trans
) == -ENOENT
)
2504 return PTR_ERR(trans
);
2506 ret
= btrfs_commit_transaction(trans
, root
);
2509 /* Update ctime/mtime for libblkid */
2510 update_dev_time(device_path
);
2514 btrfs_end_transaction(trans
, root
);
2515 rcu_string_free(device
->name
);
2516 btrfs_sysfs_rm_device_link(root
->fs_info
->fs_devices
, device
);
2519 blkdev_put(bdev
, FMODE_EXCL
);
2521 mutex_unlock(&uuid_mutex
);
2522 up_write(&sb
->s_umount
);
2527 int btrfs_init_dev_replace_tgtdev(struct btrfs_root
*root
, char *device_path
,
2528 struct btrfs_device
*srcdev
,
2529 struct btrfs_device
**device_out
)
2531 struct request_queue
*q
;
2532 struct btrfs_device
*device
;
2533 struct block_device
*bdev
;
2534 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2535 struct list_head
*devices
;
2536 struct rcu_string
*name
;
2537 u64 devid
= BTRFS_DEV_REPLACE_DEVID
;
2541 if (fs_info
->fs_devices
->seeding
) {
2542 btrfs_err(fs_info
, "the filesystem is a seed filesystem!");
2546 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2547 fs_info
->bdev_holder
);
2549 btrfs_err(fs_info
, "target device %s is invalid!", device_path
);
2550 return PTR_ERR(bdev
);
2553 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2555 devices
= &fs_info
->fs_devices
->devices
;
2556 list_for_each_entry(device
, devices
, dev_list
) {
2557 if (device
->bdev
== bdev
) {
2558 btrfs_err(fs_info
, "target device is in the filesystem!");
2565 if (i_size_read(bdev
->bd_inode
) <
2566 btrfs_device_get_total_bytes(srcdev
)) {
2567 btrfs_err(fs_info
, "target device is smaller than source device!");
2573 device
= btrfs_alloc_device(NULL
, &devid
, NULL
);
2574 if (IS_ERR(device
)) {
2575 ret
= PTR_ERR(device
);
2579 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2585 rcu_assign_pointer(device
->name
, name
);
2587 q
= bdev_get_queue(bdev
);
2588 if (blk_queue_discard(q
))
2589 device
->can_discard
= 1;
2590 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2591 device
->writeable
= 1;
2592 device
->generation
= 0;
2593 device
->io_width
= root
->sectorsize
;
2594 device
->io_align
= root
->sectorsize
;
2595 device
->sector_size
= root
->sectorsize
;
2596 device
->total_bytes
= btrfs_device_get_total_bytes(srcdev
);
2597 device
->disk_total_bytes
= btrfs_device_get_disk_total_bytes(srcdev
);
2598 device
->bytes_used
= btrfs_device_get_bytes_used(srcdev
);
2599 ASSERT(list_empty(&srcdev
->resized_list
));
2600 device
->commit_total_bytes
= srcdev
->commit_total_bytes
;
2601 device
->commit_bytes_used
= device
->bytes_used
;
2602 device
->dev_root
= fs_info
->dev_root
;
2603 device
->bdev
= bdev
;
2604 device
->in_fs_metadata
= 1;
2605 device
->is_tgtdev_for_dev_replace
= 1;
2606 device
->mode
= FMODE_EXCL
;
2607 device
->dev_stats_valid
= 1;
2608 set_blocksize(device
->bdev
, 4096);
2609 device
->fs_devices
= fs_info
->fs_devices
;
2610 list_add(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2611 fs_info
->fs_devices
->num_devices
++;
2612 fs_info
->fs_devices
->open_devices
++;
2613 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2615 *device_out
= device
;
2619 blkdev_put(bdev
, FMODE_EXCL
);
2623 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info
*fs_info
,
2624 struct btrfs_device
*tgtdev
)
2626 WARN_ON(fs_info
->fs_devices
->rw_devices
== 0);
2627 tgtdev
->io_width
= fs_info
->dev_root
->sectorsize
;
2628 tgtdev
->io_align
= fs_info
->dev_root
->sectorsize
;
2629 tgtdev
->sector_size
= fs_info
->dev_root
->sectorsize
;
2630 tgtdev
->dev_root
= fs_info
->dev_root
;
2631 tgtdev
->in_fs_metadata
= 1;
2634 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2635 struct btrfs_device
*device
)
2638 struct btrfs_path
*path
;
2639 struct btrfs_root
*root
;
2640 struct btrfs_dev_item
*dev_item
;
2641 struct extent_buffer
*leaf
;
2642 struct btrfs_key key
;
2644 root
= device
->dev_root
->fs_info
->chunk_root
;
2646 path
= btrfs_alloc_path();
2650 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2651 key
.type
= BTRFS_DEV_ITEM_KEY
;
2652 key
.offset
= device
->devid
;
2654 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2663 leaf
= path
->nodes
[0];
2664 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2666 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2667 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2668 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2669 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2670 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2671 btrfs_set_device_total_bytes(leaf
, dev_item
,
2672 btrfs_device_get_disk_total_bytes(device
));
2673 btrfs_set_device_bytes_used(leaf
, dev_item
,
2674 btrfs_device_get_bytes_used(device
));
2675 btrfs_mark_buffer_dirty(leaf
);
2678 btrfs_free_path(path
);
2682 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2683 struct btrfs_device
*device
, u64 new_size
)
2685 struct btrfs_super_block
*super_copy
=
2686 device
->dev_root
->fs_info
->super_copy
;
2687 struct btrfs_fs_devices
*fs_devices
;
2691 if (!device
->writeable
)
2694 lock_chunks(device
->dev_root
);
2695 old_total
= btrfs_super_total_bytes(super_copy
);
2696 diff
= new_size
- device
->total_bytes
;
2698 if (new_size
<= device
->total_bytes
||
2699 device
->is_tgtdev_for_dev_replace
) {
2700 unlock_chunks(device
->dev_root
);
2704 fs_devices
= device
->dev_root
->fs_info
->fs_devices
;
2706 btrfs_set_super_total_bytes(super_copy
, old_total
+ diff
);
2707 device
->fs_devices
->total_rw_bytes
+= diff
;
2709 btrfs_device_set_total_bytes(device
, new_size
);
2710 btrfs_device_set_disk_total_bytes(device
, new_size
);
2711 btrfs_clear_space_info_full(device
->dev_root
->fs_info
);
2712 if (list_empty(&device
->resized_list
))
2713 list_add_tail(&device
->resized_list
,
2714 &fs_devices
->resized_devices
);
2715 unlock_chunks(device
->dev_root
);
2717 return btrfs_update_device(trans
, device
);
2720 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
2721 struct btrfs_root
*root
, u64 chunk_objectid
,
2725 struct btrfs_path
*path
;
2726 struct btrfs_key key
;
2728 root
= root
->fs_info
->chunk_root
;
2729 path
= btrfs_alloc_path();
2733 key
.objectid
= chunk_objectid
;
2734 key
.offset
= chunk_offset
;
2735 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2737 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2740 else if (ret
> 0) { /* Logic error or corruption */
2741 btrfs_handle_fs_error(root
->fs_info
, -ENOENT
,
2742 "Failed lookup while freeing chunk.");
2747 ret
= btrfs_del_item(trans
, root
, path
);
2749 btrfs_handle_fs_error(root
->fs_info
, ret
,
2750 "Failed to delete chunk item.");
2752 btrfs_free_path(path
);
2756 static int btrfs_del_sys_chunk(struct btrfs_root
*root
, u64 chunk_objectid
, u64
2759 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
2760 struct btrfs_disk_key
*disk_key
;
2761 struct btrfs_chunk
*chunk
;
2768 struct btrfs_key key
;
2771 array_size
= btrfs_super_sys_array_size(super_copy
);
2773 ptr
= super_copy
->sys_chunk_array
;
2776 while (cur
< array_size
) {
2777 disk_key
= (struct btrfs_disk_key
*)ptr
;
2778 btrfs_disk_key_to_cpu(&key
, disk_key
);
2780 len
= sizeof(*disk_key
);
2782 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2783 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2784 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2785 len
+= btrfs_chunk_item_size(num_stripes
);
2790 if (key
.objectid
== chunk_objectid
&&
2791 key
.offset
== chunk_offset
) {
2792 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2794 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2800 unlock_chunks(root
);
2804 int btrfs_remove_chunk(struct btrfs_trans_handle
*trans
,
2805 struct btrfs_root
*root
, u64 chunk_offset
)
2807 struct extent_map_tree
*em_tree
;
2808 struct extent_map
*em
;
2809 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2810 struct map_lookup
*map
;
2811 u64 dev_extent_len
= 0;
2812 u64 chunk_objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2814 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
2817 root
= root
->fs_info
->chunk_root
;
2818 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
2820 read_lock(&em_tree
->lock
);
2821 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
2822 read_unlock(&em_tree
->lock
);
2824 if (!em
|| em
->start
> chunk_offset
||
2825 em
->start
+ em
->len
< chunk_offset
) {
2827 * This is a logic error, but we don't want to just rely on the
2828 * user having built with ASSERT enabled, so if ASSERT doesn't
2829 * do anything we still error out.
2833 free_extent_map(em
);
2836 map
= em
->map_lookup
;
2837 lock_chunks(root
->fs_info
->chunk_root
);
2838 check_system_chunk(trans
, extent_root
, map
->type
);
2839 unlock_chunks(root
->fs_info
->chunk_root
);
2842 * Take the device list mutex to prevent races with the final phase of
2843 * a device replace operation that replaces the device object associated
2844 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2846 mutex_lock(&fs_devices
->device_list_mutex
);
2847 for (i
= 0; i
< map
->num_stripes
; i
++) {
2848 struct btrfs_device
*device
= map
->stripes
[i
].dev
;
2849 ret
= btrfs_free_dev_extent(trans
, device
,
2850 map
->stripes
[i
].physical
,
2853 mutex_unlock(&fs_devices
->device_list_mutex
);
2854 btrfs_abort_transaction(trans
, ret
);
2858 if (device
->bytes_used
> 0) {
2860 btrfs_device_set_bytes_used(device
,
2861 device
->bytes_used
- dev_extent_len
);
2862 spin_lock(&root
->fs_info
->free_chunk_lock
);
2863 root
->fs_info
->free_chunk_space
+= dev_extent_len
;
2864 spin_unlock(&root
->fs_info
->free_chunk_lock
);
2865 btrfs_clear_space_info_full(root
->fs_info
);
2866 unlock_chunks(root
);
2869 if (map
->stripes
[i
].dev
) {
2870 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2872 mutex_unlock(&fs_devices
->device_list_mutex
);
2873 btrfs_abort_transaction(trans
, ret
);
2878 mutex_unlock(&fs_devices
->device_list_mutex
);
2880 ret
= btrfs_free_chunk(trans
, root
, chunk_objectid
, chunk_offset
);
2882 btrfs_abort_transaction(trans
, ret
);
2886 trace_btrfs_chunk_free(root
, map
, chunk_offset
, em
->len
);
2888 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2889 ret
= btrfs_del_sys_chunk(root
, chunk_objectid
, chunk_offset
);
2891 btrfs_abort_transaction(trans
, ret
);
2896 ret
= btrfs_remove_block_group(trans
, extent_root
, chunk_offset
, em
);
2898 btrfs_abort_transaction(trans
, ret
);
2904 free_extent_map(em
);
2908 static int btrfs_relocate_chunk(struct btrfs_root
*root
, u64 chunk_offset
)
2910 struct btrfs_root
*extent_root
;
2912 struct btrfs_block_group_cache
*block_group
;
2914 root
= root
->fs_info
->chunk_root
;
2915 extent_root
= root
->fs_info
->extent_root
;
2918 * Prevent races with automatic removal of unused block groups.
2919 * After we relocate and before we remove the chunk with offset
2920 * chunk_offset, automatic removal of the block group can kick in,
2921 * resulting in a failure when calling btrfs_remove_chunk() below.
2923 * Make sure to acquire this mutex before doing a tree search (dev
2924 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2925 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2926 * we release the path used to search the chunk/dev tree and before
2927 * the current task acquires this mutex and calls us.
2929 ASSERT(rwsem_is_locked(&root
->fs_info
->bg_delete_sem
));
2931 ret
= btrfs_can_relocate(extent_root
, chunk_offset
);
2935 /* step one, relocate all the extents inside this chunk */
2936 btrfs_scrub_pause(root
);
2937 ret
= btrfs_relocate_block_group(extent_root
, chunk_offset
);
2938 btrfs_scrub_continue(root
);
2943 * step two, flag the chunk as removed and let
2944 * btrfs_delete_unused_bgs() remove it.
2946 block_group
= btrfs_lookup_block_group(root
->fs_info
, chunk_offset
);
2947 spin_lock(&block_group
->lock
);
2948 block_group
->removed
= 1;
2949 spin_unlock(&block_group
->lock
);
2950 btrfs_put_block_group(block_group
);
2955 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
)
2957 struct btrfs_root
*chunk_root
= root
->fs_info
->chunk_root
;
2958 struct btrfs_path
*path
;
2959 struct extent_buffer
*leaf
;
2960 struct btrfs_chunk
*chunk
;
2961 struct btrfs_key key
;
2962 struct btrfs_key found_key
;
2964 bool retried
= false;
2968 path
= btrfs_alloc_path();
2973 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2974 key
.offset
= (u64
)-1;
2975 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2978 down_read(&root
->fs_info
->bg_delete_sem
);
2979 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2981 up_read(&root
->fs_info
->bg_delete_sem
);
2984 BUG_ON(ret
== 0); /* Corruption */
2986 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
2989 up_read(&root
->fs_info
->bg_delete_sem
);
2995 leaf
= path
->nodes
[0];
2996 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2998 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
2999 struct btrfs_chunk
);
3000 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3001 btrfs_release_path(path
);
3003 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3004 ret
= btrfs_relocate_chunk(chunk_root
,
3011 up_read(&root
->fs_info
->bg_delete_sem
);
3013 if (found_key
.offset
== 0)
3015 key
.offset
= found_key
.offset
- 1;
3018 if (failed
&& !retried
) {
3022 } else if (WARN_ON(failed
&& retried
)) {
3026 btrfs_free_path(path
);
3030 static int insert_balance_item(struct btrfs_root
*root
,
3031 struct btrfs_balance_control
*bctl
)
3033 struct btrfs_trans_handle
*trans
;
3034 struct btrfs_balance_item
*item
;
3035 struct btrfs_disk_balance_args disk_bargs
;
3036 struct btrfs_path
*path
;
3037 struct extent_buffer
*leaf
;
3038 struct btrfs_key key
;
3041 path
= btrfs_alloc_path();
3045 trans
= btrfs_start_transaction(root
, 0);
3046 if (IS_ERR(trans
)) {
3047 btrfs_free_path(path
);
3048 return PTR_ERR(trans
);
3051 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3052 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3055 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
3060 leaf
= path
->nodes
[0];
3061 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3063 memset_extent_buffer(leaf
, 0, (unsigned long)item
, sizeof(*item
));
3065 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
3066 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
3067 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
3068 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
3069 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
3070 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
3072 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
3074 btrfs_mark_buffer_dirty(leaf
);
3076 btrfs_free_path(path
);
3077 err
= btrfs_commit_transaction(trans
, root
);
3083 static int del_balance_item(struct btrfs_root
*root
)
3085 struct btrfs_trans_handle
*trans
;
3086 struct btrfs_path
*path
;
3087 struct btrfs_key key
;
3090 path
= btrfs_alloc_path();
3094 trans
= btrfs_start_transaction(root
, 0);
3095 if (IS_ERR(trans
)) {
3096 btrfs_free_path(path
);
3097 return PTR_ERR(trans
);
3100 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3101 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3104 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3112 ret
= btrfs_del_item(trans
, root
, path
);
3114 btrfs_free_path(path
);
3115 err
= btrfs_commit_transaction(trans
, root
);
3122 * This is a heuristic used to reduce the number of chunks balanced on
3123 * resume after balance was interrupted.
3125 static void update_balance_args(struct btrfs_balance_control
*bctl
)
3128 * Turn on soft mode for chunk types that were being converted.
3130 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3131 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3132 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3133 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3134 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3135 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3138 * Turn on usage filter if is not already used. The idea is
3139 * that chunks that we have already balanced should be
3140 * reasonably full. Don't do it for chunks that are being
3141 * converted - that will keep us from relocating unconverted
3142 * (albeit full) chunks.
3144 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3145 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3146 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3147 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3148 bctl
->data
.usage
= 90;
3150 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3151 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3152 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3153 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3154 bctl
->sys
.usage
= 90;
3156 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3157 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3158 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3159 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3160 bctl
->meta
.usage
= 90;
3165 * Should be called with both balance and volume mutexes held to
3166 * serialize other volume operations (add_dev/rm_dev/resize) with
3167 * restriper. Same goes for unset_balance_control.
3169 static void set_balance_control(struct btrfs_balance_control
*bctl
)
3171 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3173 BUG_ON(fs_info
->balance_ctl
);
3175 spin_lock(&fs_info
->balance_lock
);
3176 fs_info
->balance_ctl
= bctl
;
3177 spin_unlock(&fs_info
->balance_lock
);
3180 static void unset_balance_control(struct btrfs_fs_info
*fs_info
)
3182 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3184 BUG_ON(!fs_info
->balance_ctl
);
3186 spin_lock(&fs_info
->balance_lock
);
3187 fs_info
->balance_ctl
= NULL
;
3188 spin_unlock(&fs_info
->balance_lock
);
3194 * Balance filters. Return 1 if chunk should be filtered out
3195 * (should not be balanced).
3197 static int chunk_profiles_filter(u64 chunk_type
,
3198 struct btrfs_balance_args
*bargs
)
3200 chunk_type
= chunk_to_extended(chunk_type
) &
3201 BTRFS_EXTENDED_PROFILE_MASK
;
3203 if (bargs
->profiles
& chunk_type
)
3209 static int chunk_usage_range_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
3210 struct btrfs_balance_args
*bargs
)
3212 struct btrfs_block_group_cache
*cache
;
3214 u64 user_thresh_min
;
3215 u64 user_thresh_max
;
3218 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3219 chunk_used
= btrfs_block_group_used(&cache
->item
);
3221 if (bargs
->usage_min
== 0)
3222 user_thresh_min
= 0;
3224 user_thresh_min
= div_factor_fine(cache
->key
.offset
,
3227 if (bargs
->usage_max
== 0)
3228 user_thresh_max
= 1;
3229 else if (bargs
->usage_max
> 100)
3230 user_thresh_max
= cache
->key
.offset
;
3232 user_thresh_max
= div_factor_fine(cache
->key
.offset
,
3235 if (user_thresh_min
<= chunk_used
&& chunk_used
< user_thresh_max
)
3238 btrfs_put_block_group(cache
);
3242 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
,
3243 u64 chunk_offset
, struct btrfs_balance_args
*bargs
)
3245 struct btrfs_block_group_cache
*cache
;
3246 u64 chunk_used
, user_thresh
;
3249 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3250 chunk_used
= btrfs_block_group_used(&cache
->item
);
3252 if (bargs
->usage_min
== 0)
3254 else if (bargs
->usage
> 100)
3255 user_thresh
= cache
->key
.offset
;
3257 user_thresh
= div_factor_fine(cache
->key
.offset
,
3260 if (chunk_used
< user_thresh
)
3263 btrfs_put_block_group(cache
);
3267 static int chunk_devid_filter(struct extent_buffer
*leaf
,
3268 struct btrfs_chunk
*chunk
,
3269 struct btrfs_balance_args
*bargs
)
3271 struct btrfs_stripe
*stripe
;
3272 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3275 for (i
= 0; i
< num_stripes
; i
++) {
3276 stripe
= btrfs_stripe_nr(chunk
, i
);
3277 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
3284 /* [pstart, pend) */
3285 static int chunk_drange_filter(struct extent_buffer
*leaf
,
3286 struct btrfs_chunk
*chunk
,
3288 struct btrfs_balance_args
*bargs
)
3290 struct btrfs_stripe
*stripe
;
3291 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3297 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
3300 if (btrfs_chunk_type(leaf
, chunk
) & (BTRFS_BLOCK_GROUP_DUP
|
3301 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)) {
3302 factor
= num_stripes
/ 2;
3303 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID5
) {
3304 factor
= num_stripes
- 1;
3305 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID6
) {
3306 factor
= num_stripes
- 2;
3308 factor
= num_stripes
;
3311 for (i
= 0; i
< num_stripes
; i
++) {
3312 stripe
= btrfs_stripe_nr(chunk
, i
);
3313 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
3316 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
3317 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
3318 stripe_length
= div_u64(stripe_length
, factor
);
3320 if (stripe_offset
< bargs
->pend
&&
3321 stripe_offset
+ stripe_length
> bargs
->pstart
)
3328 /* [vstart, vend) */
3329 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
3330 struct btrfs_chunk
*chunk
,
3332 struct btrfs_balance_args
*bargs
)
3334 if (chunk_offset
< bargs
->vend
&&
3335 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
3336 /* at least part of the chunk is inside this vrange */
3342 static int chunk_stripes_range_filter(struct extent_buffer
*leaf
,
3343 struct btrfs_chunk
*chunk
,
3344 struct btrfs_balance_args
*bargs
)
3346 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3348 if (bargs
->stripes_min
<= num_stripes
3349 && num_stripes
<= bargs
->stripes_max
)
3355 static int chunk_soft_convert_filter(u64 chunk_type
,
3356 struct btrfs_balance_args
*bargs
)
3358 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
3361 chunk_type
= chunk_to_extended(chunk_type
) &
3362 BTRFS_EXTENDED_PROFILE_MASK
;
3364 if (bargs
->target
== chunk_type
)
3370 static int should_balance_chunk(struct btrfs_root
*root
,
3371 struct extent_buffer
*leaf
,
3372 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
3374 struct btrfs_balance_control
*bctl
= root
->fs_info
->balance_ctl
;
3375 struct btrfs_balance_args
*bargs
= NULL
;
3376 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3379 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
3380 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
3384 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3385 bargs
= &bctl
->data
;
3386 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3388 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3389 bargs
= &bctl
->meta
;
3391 /* profiles filter */
3392 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
3393 chunk_profiles_filter(chunk_type
, bargs
)) {
3398 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3399 chunk_usage_filter(bctl
->fs_info
, chunk_offset
, bargs
)) {
3401 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3402 chunk_usage_range_filter(bctl
->fs_info
, chunk_offset
, bargs
)) {
3407 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
3408 chunk_devid_filter(leaf
, chunk
, bargs
)) {
3412 /* drange filter, makes sense only with devid filter */
3413 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
3414 chunk_drange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3419 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
3420 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3424 /* stripes filter */
3425 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_STRIPES_RANGE
) &&
3426 chunk_stripes_range_filter(leaf
, chunk
, bargs
)) {
3430 /* soft profile changing mode */
3431 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
3432 chunk_soft_convert_filter(chunk_type
, bargs
)) {
3437 * limited by count, must be the last filter
3439 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT
)) {
3440 if (bargs
->limit
== 0)
3444 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT_RANGE
)) {
3446 * Same logic as the 'limit' filter; the minimum cannot be
3447 * determined here because we do not have the global information
3448 * about the count of all chunks that satisfy the filters.
3450 if (bargs
->limit_max
== 0)
3459 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
3461 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3462 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
3463 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
3464 struct list_head
*devices
;
3465 struct btrfs_device
*device
;
3469 struct btrfs_chunk
*chunk
;
3470 struct btrfs_path
*path
= NULL
;
3471 struct btrfs_key key
;
3472 struct btrfs_key found_key
;
3473 struct btrfs_trans_handle
*trans
;
3474 struct extent_buffer
*leaf
;
3477 int enospc_errors
= 0;
3478 bool counting
= true;
3479 /* The single value limit and min/max limits use the same bytes in the */
3480 u64 limit_data
= bctl
->data
.limit
;
3481 u64 limit_meta
= bctl
->meta
.limit
;
3482 u64 limit_sys
= bctl
->sys
.limit
;
3486 int chunk_reserved
= 0;
3489 /* step one make some room on all the devices */
3490 devices
= &fs_info
->fs_devices
->devices
;
3491 list_for_each_entry(device
, devices
, dev_list
) {
3492 old_size
= btrfs_device_get_total_bytes(device
);
3493 size_to_free
= div_factor(old_size
, 1);
3494 size_to_free
= min_t(u64
, size_to_free
, SZ_1M
);
3495 if (!device
->writeable
||
3496 btrfs_device_get_total_bytes(device
) -
3497 btrfs_device_get_bytes_used(device
) > size_to_free
||
3498 device
->is_tgtdev_for_dev_replace
)
3501 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
3505 /* btrfs_shrink_device never returns ret > 0 */
3510 trans
= btrfs_start_transaction(dev_root
, 0);
3511 if (IS_ERR(trans
)) {
3512 ret
= PTR_ERR(trans
);
3513 btrfs_info_in_rcu(fs_info
,
3514 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3515 rcu_str_deref(device
->name
), ret
,
3516 old_size
, old_size
- size_to_free
);
3520 ret
= btrfs_grow_device(trans
, device
, old_size
);
3522 btrfs_end_transaction(trans
, dev_root
);
3523 /* btrfs_grow_device never returns ret > 0 */
3525 btrfs_info_in_rcu(fs_info
,
3526 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3527 rcu_str_deref(device
->name
), ret
,
3528 old_size
, old_size
- size_to_free
);
3532 btrfs_end_transaction(trans
, dev_root
);
3535 /* step two, relocate all the chunks */
3536 path
= btrfs_alloc_path();
3542 /* zero out stat counters */
3543 spin_lock(&fs_info
->balance_lock
);
3544 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
3545 spin_unlock(&fs_info
->balance_lock
);
3549 * The single value limit and min/max limits use the same bytes
3552 bctl
->data
.limit
= limit_data
;
3553 bctl
->meta
.limit
= limit_meta
;
3554 bctl
->sys
.limit
= limit_sys
;
3556 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3557 key
.offset
= (u64
)-1;
3558 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3561 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
3562 atomic_read(&fs_info
->balance_cancel_req
)) {
3567 down_read(&fs_info
->bg_delete_sem
);
3568 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
3570 up_read(&fs_info
->bg_delete_sem
);
3575 * this shouldn't happen, it means the last relocate
3579 BUG(); /* FIXME break ? */
3581 ret
= btrfs_previous_item(chunk_root
, path
, 0,
3582 BTRFS_CHUNK_ITEM_KEY
);
3584 up_read(&fs_info
->bg_delete_sem
);
3589 leaf
= path
->nodes
[0];
3590 slot
= path
->slots
[0];
3591 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3593 if (found_key
.objectid
!= key
.objectid
) {
3594 up_read(&fs_info
->bg_delete_sem
);
3598 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
3599 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3602 spin_lock(&fs_info
->balance_lock
);
3603 bctl
->stat
.considered
++;
3604 spin_unlock(&fs_info
->balance_lock
);
3607 ret
= should_balance_chunk(chunk_root
, leaf
, chunk
,
3610 btrfs_release_path(path
);
3612 up_read(&fs_info
->bg_delete_sem
);
3617 up_read(&fs_info
->bg_delete_sem
);
3618 spin_lock(&fs_info
->balance_lock
);
3619 bctl
->stat
.expected
++;
3620 spin_unlock(&fs_info
->balance_lock
);
3622 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3624 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3626 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3633 * Apply limit_min filter, no need to check if the LIMITS
3634 * filter is used, limit_min is 0 by default
3636 if (((chunk_type
& BTRFS_BLOCK_GROUP_DATA
) &&
3637 count_data
< bctl
->data
.limit_min
)
3638 || ((chunk_type
& BTRFS_BLOCK_GROUP_METADATA
) &&
3639 count_meta
< bctl
->meta
.limit_min
)
3640 || ((chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) &&
3641 count_sys
< bctl
->sys
.limit_min
)) {
3642 up_read(&fs_info
->bg_delete_sem
);
3646 ASSERT(fs_info
->data_sinfo
);
3647 spin_lock(&fs_info
->data_sinfo
->lock
);
3648 bytes_used
= fs_info
->data_sinfo
->bytes_used
;
3649 spin_unlock(&fs_info
->data_sinfo
->lock
);
3651 if ((chunk_type
& BTRFS_BLOCK_GROUP_DATA
) &&
3652 !chunk_reserved
&& !bytes_used
) {
3653 trans
= btrfs_start_transaction(chunk_root
, 0);
3654 if (IS_ERR(trans
)) {
3655 up_read(&fs_info
->bg_delete_sem
);
3656 ret
= PTR_ERR(trans
);
3660 ret
= btrfs_force_chunk_alloc(trans
, chunk_root
,
3661 BTRFS_BLOCK_GROUP_DATA
);
3662 btrfs_end_transaction(trans
, chunk_root
);
3664 up_read(&fs_info
->bg_delete_sem
);
3670 ret
= btrfs_relocate_chunk(chunk_root
,
3672 up_read(&fs_info
->bg_delete_sem
);
3673 if (ret
&& ret
!= -ENOSPC
)
3675 if (ret
== -ENOSPC
) {
3678 spin_lock(&fs_info
->balance_lock
);
3679 bctl
->stat
.completed
++;
3680 spin_unlock(&fs_info
->balance_lock
);
3683 if (found_key
.offset
== 0)
3685 key
.offset
= found_key
.offset
- 1;
3689 btrfs_release_path(path
);
3694 btrfs_free_path(path
);
3695 if (enospc_errors
) {
3696 btrfs_info(fs_info
, "%d enospc errors during balance",
3706 * alloc_profile_is_valid - see if a given profile is valid and reduced
3707 * @flags: profile to validate
3708 * @extended: if true @flags is treated as an extended profile
3710 static int alloc_profile_is_valid(u64 flags
, int extended
)
3712 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
3713 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
3715 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
3717 /* 1) check that all other bits are zeroed */
3721 /* 2) see if profile is reduced */
3723 return !extended
; /* "0" is valid for usual profiles */
3725 /* true if exactly one bit set */
3726 return (flags
& (flags
- 1)) == 0;
3729 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
3731 /* cancel requested || normal exit path */
3732 return atomic_read(&fs_info
->balance_cancel_req
) ||
3733 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
3734 atomic_read(&fs_info
->balance_cancel_req
) == 0);
3737 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
3741 unset_balance_control(fs_info
);
3742 ret
= del_balance_item(fs_info
->tree_root
);
3744 btrfs_handle_fs_error(fs_info
, ret
, NULL
);
3746 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3749 /* Non-zero return value signifies invalidity */
3750 static inline int validate_convert_profile(struct btrfs_balance_args
*bctl_arg
,
3753 return ((bctl_arg
->flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3754 (!alloc_profile_is_valid(bctl_arg
->target
, 1) ||
3755 (bctl_arg
->target
& ~allowed
)));
3759 * Should be called with both balance and volume mutexes held
3761 int btrfs_balance(struct btrfs_balance_control
*bctl
,
3762 struct btrfs_ioctl_balance_args
*bargs
)
3764 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3771 if (btrfs_fs_closing(fs_info
) ||
3772 atomic_read(&fs_info
->balance_pause_req
) ||
3773 atomic_read(&fs_info
->balance_cancel_req
)) {
3778 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
3779 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
3783 * In case of mixed groups both data and meta should be picked,
3784 * and identical options should be given for both of them.
3786 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
3787 if (mixed
&& (bctl
->flags
& allowed
)) {
3788 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
3789 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
3790 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
3791 btrfs_err(fs_info
, "with mixed groups data and "
3792 "metadata balance options must be the same");
3798 num_devices
= fs_info
->fs_devices
->num_devices
;
3799 btrfs_dev_replace_lock(&fs_info
->dev_replace
, 0);
3800 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
3801 BUG_ON(num_devices
< 1);
3804 btrfs_dev_replace_unlock(&fs_info
->dev_replace
, 0);
3805 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
| BTRFS_BLOCK_GROUP_DUP
;
3806 if (num_devices
> 1)
3807 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
3808 if (num_devices
> 2)
3809 allowed
|= BTRFS_BLOCK_GROUP_RAID5
;
3810 if (num_devices
> 3)
3811 allowed
|= (BTRFS_BLOCK_GROUP_RAID10
|
3812 BTRFS_BLOCK_GROUP_RAID6
);
3813 if (validate_convert_profile(&bctl
->data
, allowed
)) {
3814 btrfs_err(fs_info
, "unable to start balance with target "
3815 "data profile %llu",
3820 if (validate_convert_profile(&bctl
->meta
, allowed
)) {
3822 "unable to start balance with target metadata profile %llu",
3827 if (validate_convert_profile(&bctl
->sys
, allowed
)) {
3829 "unable to start balance with target system profile %llu",
3835 /* allow to reduce meta or sys integrity only if force set */
3836 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3837 BTRFS_BLOCK_GROUP_RAID10
|
3838 BTRFS_BLOCK_GROUP_RAID5
|
3839 BTRFS_BLOCK_GROUP_RAID6
;
3841 seq
= read_seqbegin(&fs_info
->profiles_lock
);
3843 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3844 (fs_info
->avail_system_alloc_bits
& allowed
) &&
3845 !(bctl
->sys
.target
& allowed
)) ||
3846 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3847 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
3848 !(bctl
->meta
.target
& allowed
))) {
3849 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
3850 btrfs_info(fs_info
, "force reducing metadata integrity");
3852 btrfs_err(fs_info
, "balance will reduce metadata "
3853 "integrity, use force if you want this");
3858 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
3860 if (btrfs_get_num_tolerated_disk_barrier_failures(bctl
->meta
.target
) <
3861 btrfs_get_num_tolerated_disk_barrier_failures(bctl
->data
.target
)) {
3863 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3864 bctl
->meta
.target
, bctl
->data
.target
);
3867 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3868 fs_info
->num_tolerated_disk_barrier_failures
= min(
3869 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
),
3870 btrfs_get_num_tolerated_disk_barrier_failures(
3874 ret
= insert_balance_item(fs_info
->tree_root
, bctl
);
3875 if (ret
&& ret
!= -EEXIST
)
3878 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
3879 BUG_ON(ret
== -EEXIST
);
3880 set_balance_control(bctl
);
3882 BUG_ON(ret
!= -EEXIST
);
3883 spin_lock(&fs_info
->balance_lock
);
3884 update_balance_args(bctl
);
3885 spin_unlock(&fs_info
->balance_lock
);
3888 atomic_inc(&fs_info
->balance_running
);
3889 mutex_unlock(&fs_info
->balance_mutex
);
3891 ret
= __btrfs_balance(fs_info
);
3893 mutex_lock(&fs_info
->balance_mutex
);
3894 atomic_dec(&fs_info
->balance_running
);
3896 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3897 fs_info
->num_tolerated_disk_barrier_failures
=
3898 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3902 memset(bargs
, 0, sizeof(*bargs
));
3903 update_ioctl_balance_args(fs_info
, 0, bargs
);
3906 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
3907 balance_need_close(fs_info
)) {
3908 __cancel_balance(fs_info
);
3911 wake_up(&fs_info
->balance_wait_q
);
3915 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
3916 __cancel_balance(fs_info
);
3919 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3924 static int balance_kthread(void *data
)
3926 struct btrfs_fs_info
*fs_info
= data
;
3929 mutex_lock(&fs_info
->volume_mutex
);
3930 mutex_lock(&fs_info
->balance_mutex
);
3932 if (fs_info
->balance_ctl
) {
3933 btrfs_info(fs_info
, "continuing balance");
3934 ret
= btrfs_balance(fs_info
->balance_ctl
, NULL
);
3937 mutex_unlock(&fs_info
->balance_mutex
);
3938 mutex_unlock(&fs_info
->volume_mutex
);
3943 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
3945 struct task_struct
*tsk
;
3947 spin_lock(&fs_info
->balance_lock
);
3948 if (!fs_info
->balance_ctl
) {
3949 spin_unlock(&fs_info
->balance_lock
);
3952 spin_unlock(&fs_info
->balance_lock
);
3954 if (btrfs_test_opt(fs_info
, SKIP_BALANCE
)) {
3955 btrfs_info(fs_info
, "force skipping balance");
3959 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
3960 return PTR_ERR_OR_ZERO(tsk
);
3963 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
3965 struct btrfs_balance_control
*bctl
;
3966 struct btrfs_balance_item
*item
;
3967 struct btrfs_disk_balance_args disk_bargs
;
3968 struct btrfs_path
*path
;
3969 struct extent_buffer
*leaf
;
3970 struct btrfs_key key
;
3973 path
= btrfs_alloc_path();
3977 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3978 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3981 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
3984 if (ret
> 0) { /* ret = -ENOENT; */
3989 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
3995 leaf
= path
->nodes
[0];
3996 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3998 bctl
->fs_info
= fs_info
;
3999 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
4000 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
4002 btrfs_balance_data(leaf
, item
, &disk_bargs
);
4003 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
4004 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
4005 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
4006 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
4007 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
4009 WARN_ON(atomic_xchg(&fs_info
->mutually_exclusive_operation_running
, 1));
4011 mutex_lock(&fs_info
->volume_mutex
);
4012 mutex_lock(&fs_info
->balance_mutex
);
4014 set_balance_control(bctl
);
4016 mutex_unlock(&fs_info
->balance_mutex
);
4017 mutex_unlock(&fs_info
->volume_mutex
);
4019 btrfs_free_path(path
);
4023 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
4027 mutex_lock(&fs_info
->balance_mutex
);
4028 if (!fs_info
->balance_ctl
) {
4029 mutex_unlock(&fs_info
->balance_mutex
);
4033 if (atomic_read(&fs_info
->balance_running
)) {
4034 atomic_inc(&fs_info
->balance_pause_req
);
4035 mutex_unlock(&fs_info
->balance_mutex
);
4037 wait_event(fs_info
->balance_wait_q
,
4038 atomic_read(&fs_info
->balance_running
) == 0);
4040 mutex_lock(&fs_info
->balance_mutex
);
4041 /* we are good with balance_ctl ripped off from under us */
4042 BUG_ON(atomic_read(&fs_info
->balance_running
));
4043 atomic_dec(&fs_info
->balance_pause_req
);
4048 mutex_unlock(&fs_info
->balance_mutex
);
4052 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
4054 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
4057 mutex_lock(&fs_info
->balance_mutex
);
4058 if (!fs_info
->balance_ctl
) {
4059 mutex_unlock(&fs_info
->balance_mutex
);
4063 atomic_inc(&fs_info
->balance_cancel_req
);
4065 * if we are running just wait and return, balance item is
4066 * deleted in btrfs_balance in this case
4068 if (atomic_read(&fs_info
->balance_running
)) {
4069 mutex_unlock(&fs_info
->balance_mutex
);
4070 wait_event(fs_info
->balance_wait_q
,
4071 atomic_read(&fs_info
->balance_running
) == 0);
4072 mutex_lock(&fs_info
->balance_mutex
);
4074 /* __cancel_balance needs volume_mutex */
4075 mutex_unlock(&fs_info
->balance_mutex
);
4076 mutex_lock(&fs_info
->volume_mutex
);
4077 mutex_lock(&fs_info
->balance_mutex
);
4079 if (fs_info
->balance_ctl
)
4080 __cancel_balance(fs_info
);
4082 mutex_unlock(&fs_info
->volume_mutex
);
4085 BUG_ON(fs_info
->balance_ctl
|| atomic_read(&fs_info
->balance_running
));
4086 atomic_dec(&fs_info
->balance_cancel_req
);
4087 mutex_unlock(&fs_info
->balance_mutex
);
4091 static int btrfs_uuid_scan_kthread(void *data
)
4093 struct btrfs_fs_info
*fs_info
= data
;
4094 struct btrfs_root
*root
= fs_info
->tree_root
;
4095 struct btrfs_key key
;
4096 struct btrfs_key max_key
;
4097 struct btrfs_path
*path
= NULL
;
4099 struct extent_buffer
*eb
;
4101 struct btrfs_root_item root_item
;
4103 struct btrfs_trans_handle
*trans
= NULL
;
4105 path
= btrfs_alloc_path();
4112 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4115 max_key
.objectid
= (u64
)-1;
4116 max_key
.type
= BTRFS_ROOT_ITEM_KEY
;
4117 max_key
.offset
= (u64
)-1;
4120 ret
= btrfs_search_forward(root
, &key
, path
, 0);
4127 if (key
.type
!= BTRFS_ROOT_ITEM_KEY
||
4128 (key
.objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
4129 key
.objectid
!= BTRFS_FS_TREE_OBJECTID
) ||
4130 key
.objectid
> BTRFS_LAST_FREE_OBJECTID
)
4133 eb
= path
->nodes
[0];
4134 slot
= path
->slots
[0];
4135 item_size
= btrfs_item_size_nr(eb
, slot
);
4136 if (item_size
< sizeof(root_item
))
4139 read_extent_buffer(eb
, &root_item
,
4140 btrfs_item_ptr_offset(eb
, slot
),
4141 (int)sizeof(root_item
));
4142 if (btrfs_root_refs(&root_item
) == 0)
4145 if (!btrfs_is_empty_uuid(root_item
.uuid
) ||
4146 !btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4150 btrfs_release_path(path
);
4152 * 1 - subvol uuid item
4153 * 1 - received_subvol uuid item
4155 trans
= btrfs_start_transaction(fs_info
->uuid_root
, 2);
4156 if (IS_ERR(trans
)) {
4157 ret
= PTR_ERR(trans
);
4165 if (!btrfs_is_empty_uuid(root_item
.uuid
)) {
4166 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
4168 BTRFS_UUID_KEY_SUBVOL
,
4171 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4177 if (!btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4178 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
4179 root_item
.received_uuid
,
4180 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
4183 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4191 ret
= btrfs_end_transaction(trans
, fs_info
->uuid_root
);
4197 btrfs_release_path(path
);
4198 if (key
.offset
< (u64
)-1) {
4200 } else if (key
.type
< BTRFS_ROOT_ITEM_KEY
) {
4202 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4203 } else if (key
.objectid
< (u64
)-1) {
4205 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4214 btrfs_free_path(path
);
4215 if (trans
&& !IS_ERR(trans
))
4216 btrfs_end_transaction(trans
, fs_info
->uuid_root
);
4218 btrfs_warn(fs_info
, "btrfs_uuid_scan_kthread failed %d", ret
);
4220 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN
, &fs_info
->flags
);
4221 up(&fs_info
->uuid_tree_rescan_sem
);
4226 * Callback for btrfs_uuid_tree_iterate().
4228 * 0 check succeeded, the entry is not outdated.
4229 * < 0 if an error occurred.
4230 * > 0 if the check failed, which means the caller shall remove the entry.
4232 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info
*fs_info
,
4233 u8
*uuid
, u8 type
, u64 subid
)
4235 struct btrfs_key key
;
4237 struct btrfs_root
*subvol_root
;
4239 if (type
!= BTRFS_UUID_KEY_SUBVOL
&&
4240 type
!= BTRFS_UUID_KEY_RECEIVED_SUBVOL
)
4243 key
.objectid
= subid
;
4244 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4245 key
.offset
= (u64
)-1;
4246 subvol_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
4247 if (IS_ERR(subvol_root
)) {
4248 ret
= PTR_ERR(subvol_root
);
4255 case BTRFS_UUID_KEY_SUBVOL
:
4256 if (memcmp(uuid
, subvol_root
->root_item
.uuid
, BTRFS_UUID_SIZE
))
4259 case BTRFS_UUID_KEY_RECEIVED_SUBVOL
:
4260 if (memcmp(uuid
, subvol_root
->root_item
.received_uuid
,
4270 static int btrfs_uuid_rescan_kthread(void *data
)
4272 struct btrfs_fs_info
*fs_info
= (struct btrfs_fs_info
*)data
;
4276 * 1st step is to iterate through the existing UUID tree and
4277 * to delete all entries that contain outdated data.
4278 * 2nd step is to add all missing entries to the UUID tree.
4280 ret
= btrfs_uuid_tree_iterate(fs_info
, btrfs_check_uuid_tree_entry
);
4282 btrfs_warn(fs_info
, "iterating uuid_tree failed %d", ret
);
4283 up(&fs_info
->uuid_tree_rescan_sem
);
4286 return btrfs_uuid_scan_kthread(data
);
4289 int btrfs_create_uuid_tree(struct btrfs_fs_info
*fs_info
)
4291 struct btrfs_trans_handle
*trans
;
4292 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
4293 struct btrfs_root
*uuid_root
;
4294 struct task_struct
*task
;
4301 trans
= btrfs_start_transaction(tree_root
, 2);
4303 return PTR_ERR(trans
);
4305 uuid_root
= btrfs_create_tree(trans
, fs_info
,
4306 BTRFS_UUID_TREE_OBJECTID
);
4307 if (IS_ERR(uuid_root
)) {
4308 ret
= PTR_ERR(uuid_root
);
4309 btrfs_abort_transaction(trans
, ret
);
4310 btrfs_end_transaction(trans
, tree_root
);
4314 fs_info
->uuid_root
= uuid_root
;
4316 ret
= btrfs_commit_transaction(trans
, tree_root
);
4320 down(&fs_info
->uuid_tree_rescan_sem
);
4321 task
= kthread_run(btrfs_uuid_scan_kthread
, fs_info
, "btrfs-uuid");
4323 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4324 btrfs_warn(fs_info
, "failed to start uuid_scan task");
4325 up(&fs_info
->uuid_tree_rescan_sem
);
4326 return PTR_ERR(task
);
4332 int btrfs_check_uuid_tree(struct btrfs_fs_info
*fs_info
)
4334 struct task_struct
*task
;
4336 down(&fs_info
->uuid_tree_rescan_sem
);
4337 task
= kthread_run(btrfs_uuid_rescan_kthread
, fs_info
, "btrfs-uuid");
4339 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4340 btrfs_warn(fs_info
, "failed to start uuid_rescan task");
4341 up(&fs_info
->uuid_tree_rescan_sem
);
4342 return PTR_ERR(task
);
4349 * shrinking a device means finding all of the device extents past
4350 * the new size, and then following the back refs to the chunks.
4351 * The chunk relocation code actually frees the device extent
4353 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
4355 struct btrfs_trans_handle
*trans
;
4356 struct btrfs_root
*root
= device
->dev_root
;
4357 struct btrfs_dev_extent
*dev_extent
= NULL
;
4358 struct btrfs_path
*path
;
4364 bool retried
= false;
4365 bool checked_pending_chunks
= false;
4366 struct extent_buffer
*l
;
4367 struct btrfs_key key
;
4368 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
4369 u64 old_total
= btrfs_super_total_bytes(super_copy
);
4370 u64 old_size
= btrfs_device_get_total_bytes(device
);
4371 u64 diff
= old_size
- new_size
;
4373 if (device
->is_tgtdev_for_dev_replace
)
4376 path
= btrfs_alloc_path();
4380 path
->reada
= READA_FORWARD
;
4384 btrfs_device_set_total_bytes(device
, new_size
);
4385 if (device
->writeable
) {
4386 device
->fs_devices
->total_rw_bytes
-= diff
;
4387 spin_lock(&root
->fs_info
->free_chunk_lock
);
4388 root
->fs_info
->free_chunk_space
-= diff
;
4389 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4391 unlock_chunks(root
);
4394 key
.objectid
= device
->devid
;
4395 key
.offset
= (u64
)-1;
4396 key
.type
= BTRFS_DEV_EXTENT_KEY
;
4399 down_read(&root
->fs_info
->bg_delete_sem
);
4400 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4402 up_read(&root
->fs_info
->bg_delete_sem
);
4406 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
4408 up_read(&root
->fs_info
->bg_delete_sem
);
4413 btrfs_release_path(path
);
4418 slot
= path
->slots
[0];
4419 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
4421 if (key
.objectid
!= device
->devid
) {
4422 up_read(&root
->fs_info
->bg_delete_sem
);
4423 btrfs_release_path(path
);
4427 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
4428 length
= btrfs_dev_extent_length(l
, dev_extent
);
4430 if (key
.offset
+ length
<= new_size
) {
4431 up_read(&root
->fs_info
->bg_delete_sem
);
4432 btrfs_release_path(path
);
4436 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
4437 btrfs_release_path(path
);
4439 ret
= btrfs_relocate_chunk(root
, chunk_offset
);
4440 up_read(&root
->fs_info
->bg_delete_sem
);
4441 if (ret
&& ret
!= -ENOSPC
)
4445 } while (key
.offset
-- > 0);
4447 if (failed
&& !retried
) {
4451 } else if (failed
&& retried
) {
4456 /* Shrinking succeeded, else we would be at "done". */
4457 trans
= btrfs_start_transaction(root
, 0);
4458 if (IS_ERR(trans
)) {
4459 ret
= PTR_ERR(trans
);
4466 * We checked in the above loop all device extents that were already in
4467 * the device tree. However before we have updated the device's
4468 * total_bytes to the new size, we might have had chunk allocations that
4469 * have not complete yet (new block groups attached to transaction
4470 * handles), and therefore their device extents were not yet in the
4471 * device tree and we missed them in the loop above. So if we have any
4472 * pending chunk using a device extent that overlaps the device range
4473 * that we can not use anymore, commit the current transaction and
4474 * repeat the search on the device tree - this way we guarantee we will
4475 * not have chunks using device extents that end beyond 'new_size'.
4477 if (!checked_pending_chunks
) {
4478 u64 start
= new_size
;
4479 u64 len
= old_size
- new_size
;
4481 if (contains_pending_extent(trans
->transaction
, device
,
4483 unlock_chunks(root
);
4484 checked_pending_chunks
= true;
4487 ret
= btrfs_commit_transaction(trans
, root
);
4494 btrfs_device_set_disk_total_bytes(device
, new_size
);
4495 if (list_empty(&device
->resized_list
))
4496 list_add_tail(&device
->resized_list
,
4497 &root
->fs_info
->fs_devices
->resized_devices
);
4499 WARN_ON(diff
> old_total
);
4500 btrfs_set_super_total_bytes(super_copy
, old_total
- diff
);
4501 unlock_chunks(root
);
4503 /* Now btrfs_update_device() will change the on-disk size. */
4504 ret
= btrfs_update_device(trans
, device
);
4505 btrfs_end_transaction(trans
, root
);
4507 btrfs_free_path(path
);
4510 btrfs_device_set_total_bytes(device
, old_size
);
4511 if (device
->writeable
)
4512 device
->fs_devices
->total_rw_bytes
+= diff
;
4513 spin_lock(&root
->fs_info
->free_chunk_lock
);
4514 root
->fs_info
->free_chunk_space
+= diff
;
4515 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4516 unlock_chunks(root
);
4521 static int btrfs_add_system_chunk(struct btrfs_root
*root
,
4522 struct btrfs_key
*key
,
4523 struct btrfs_chunk
*chunk
, int item_size
)
4525 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
4526 struct btrfs_disk_key disk_key
;
4531 array_size
= btrfs_super_sys_array_size(super_copy
);
4532 if (array_size
+ item_size
+ sizeof(disk_key
)
4533 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4534 unlock_chunks(root
);
4538 ptr
= super_copy
->sys_chunk_array
+ array_size
;
4539 btrfs_cpu_key_to_disk(&disk_key
, key
);
4540 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
4541 ptr
+= sizeof(disk_key
);
4542 memcpy(ptr
, chunk
, item_size
);
4543 item_size
+= sizeof(disk_key
);
4544 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
4545 unlock_chunks(root
);
4551 * sort the devices in descending order by max_avail, total_avail
4553 static int btrfs_cmp_device_info(const void *a
, const void *b
)
4555 const struct btrfs_device_info
*di_a
= a
;
4556 const struct btrfs_device_info
*di_b
= b
;
4558 if (di_a
->max_avail
> di_b
->max_avail
)
4560 if (di_a
->max_avail
< di_b
->max_avail
)
4562 if (di_a
->total_avail
> di_b
->total_avail
)
4564 if (di_a
->total_avail
< di_b
->total_avail
)
4569 static u32
find_raid56_stripe_len(u32 data_devices
, u32 dev_stripe_target
)
4571 /* TODO allow them to set a preferred stripe size */
4575 static void check_raid56_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
4577 if (!(type
& BTRFS_BLOCK_GROUP_RAID56_MASK
))
4580 btrfs_set_fs_incompat(info
, RAID56
);
4583 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r) \
4584 - sizeof(struct btrfs_chunk)) \
4585 / sizeof(struct btrfs_stripe) + 1)
4587 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4588 - 2 * sizeof(struct btrfs_disk_key) \
4589 - 2 * sizeof(struct btrfs_chunk)) \
4590 / sizeof(struct btrfs_stripe) + 1)
4592 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4593 struct btrfs_root
*extent_root
, u64 start
,
4596 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
4597 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
4598 struct list_head
*cur
;
4599 struct map_lookup
*map
= NULL
;
4600 struct extent_map_tree
*em_tree
;
4601 struct extent_map
*em
;
4602 struct btrfs_device_info
*devices_info
= NULL
;
4604 int num_stripes
; /* total number of stripes to allocate */
4605 int data_stripes
; /* number of stripes that count for
4607 int sub_stripes
; /* sub_stripes info for map */
4608 int dev_stripes
; /* stripes per dev */
4609 int devs_max
; /* max devs to use */
4610 int devs_min
; /* min devs needed */
4611 int devs_increment
; /* ndevs has to be a multiple of this */
4612 int ncopies
; /* how many copies to data has */
4614 u64 max_stripe_size
;
4618 u64 raid_stripe_len
= BTRFS_STRIPE_LEN
;
4624 BUG_ON(!alloc_profile_is_valid(type
, 0));
4626 if (list_empty(&fs_devices
->alloc_list
))
4629 index
= __get_raid_index(type
);
4631 sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
4632 dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
4633 devs_max
= btrfs_raid_array
[index
].devs_max
;
4634 devs_min
= btrfs_raid_array
[index
].devs_min
;
4635 devs_increment
= btrfs_raid_array
[index
].devs_increment
;
4636 ncopies
= btrfs_raid_array
[index
].ncopies
;
4638 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
4639 max_stripe_size
= SZ_1G
;
4640 max_chunk_size
= 10 * max_stripe_size
;
4642 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4643 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
4644 /* for larger filesystems, use larger metadata chunks */
4645 if (fs_devices
->total_rw_bytes
> 50ULL * SZ_1G
)
4646 max_stripe_size
= SZ_1G
;
4648 max_stripe_size
= SZ_256M
;
4649 max_chunk_size
= max_stripe_size
;
4651 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4652 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4653 max_stripe_size
= SZ_32M
;
4654 max_chunk_size
= 2 * max_stripe_size
;
4656 devs_max
= BTRFS_MAX_DEVS_SYS_CHUNK
;
4658 btrfs_err(info
, "invalid chunk type 0x%llx requested",
4663 /* we don't want a chunk larger than 10% of writeable space */
4664 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
4667 devices_info
= kcalloc(fs_devices
->rw_devices
, sizeof(*devices_info
),
4672 cur
= fs_devices
->alloc_list
.next
;
4675 * in the first pass through the devices list, we gather information
4676 * about the available holes on each device.
4679 while (cur
!= &fs_devices
->alloc_list
) {
4680 struct btrfs_device
*device
;
4684 device
= list_entry(cur
, struct btrfs_device
, dev_alloc_list
);
4688 if (!device
->writeable
) {
4690 "BTRFS: read-only device in alloc_list\n");
4694 if (!device
->in_fs_metadata
||
4695 device
->is_tgtdev_for_dev_replace
)
4698 if (device
->total_bytes
> device
->bytes_used
)
4699 total_avail
= device
->total_bytes
- device
->bytes_used
;
4703 /* If there is no space on this device, skip it. */
4704 if (total_avail
== 0)
4707 ret
= find_free_dev_extent(trans
, device
,
4708 max_stripe_size
* dev_stripes
,
4709 &dev_offset
, &max_avail
);
4710 if (ret
&& ret
!= -ENOSPC
)
4714 max_avail
= max_stripe_size
* dev_stripes
;
4716 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
4719 if (ndevs
== fs_devices
->rw_devices
) {
4720 WARN(1, "%s: found more than %llu devices\n",
4721 __func__
, fs_devices
->rw_devices
);
4724 devices_info
[ndevs
].dev_offset
= dev_offset
;
4725 devices_info
[ndevs
].max_avail
= max_avail
;
4726 devices_info
[ndevs
].total_avail
= total_avail
;
4727 devices_info
[ndevs
].dev
= device
;
4732 * now sort the devices by hole size / available space
4734 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
4735 btrfs_cmp_device_info
, NULL
);
4737 /* round down to number of usable stripes */
4738 ndevs
-= ndevs
% devs_increment
;
4740 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
4745 if (devs_max
&& ndevs
> devs_max
)
4748 * the primary goal is to maximize the number of stripes, so use as many
4749 * devices as possible, even if the stripes are not maximum sized.
4751 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4752 num_stripes
= ndevs
* dev_stripes
;
4755 * this will have to be fixed for RAID1 and RAID10 over
4758 data_stripes
= num_stripes
/ ncopies
;
4760 if (type
& BTRFS_BLOCK_GROUP_RAID5
) {
4761 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 1,
4762 extent_root
->stripesize
);
4763 data_stripes
= num_stripes
- 1;
4765 if (type
& BTRFS_BLOCK_GROUP_RAID6
) {
4766 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 2,
4767 extent_root
->stripesize
);
4768 data_stripes
= num_stripes
- 2;
4772 * Use the number of data stripes to figure out how big this chunk
4773 * is really going to be in terms of logical address space,
4774 * and compare that answer with the max chunk size
4776 if (stripe_size
* data_stripes
> max_chunk_size
) {
4777 u64 mask
= (1ULL << 24) - 1;
4779 stripe_size
= div_u64(max_chunk_size
, data_stripes
);
4781 /* bump the answer up to a 16MB boundary */
4782 stripe_size
= (stripe_size
+ mask
) & ~mask
;
4784 /* but don't go higher than the limits we found
4785 * while searching for free extents
4787 if (stripe_size
> devices_info
[ndevs
-1].max_avail
)
4788 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4791 stripe_size
= div_u64(stripe_size
, dev_stripes
);
4793 /* align to BTRFS_STRIPE_LEN */
4794 stripe_size
= div_u64(stripe_size
, raid_stripe_len
);
4795 stripe_size
*= raid_stripe_len
;
4797 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
4802 map
->num_stripes
= num_stripes
;
4804 for (i
= 0; i
< ndevs
; ++i
) {
4805 for (j
= 0; j
< dev_stripes
; ++j
) {
4806 int s
= i
* dev_stripes
+ j
;
4807 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
4808 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
4812 map
->sector_size
= extent_root
->sectorsize
;
4813 map
->stripe_len
= raid_stripe_len
;
4814 map
->io_align
= raid_stripe_len
;
4815 map
->io_width
= raid_stripe_len
;
4817 map
->sub_stripes
= sub_stripes
;
4819 num_bytes
= stripe_size
* data_stripes
;
4821 trace_btrfs_chunk_alloc(info
->chunk_root
, map
, start
, num_bytes
);
4823 em
= alloc_extent_map();
4829 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
4830 em
->map_lookup
= map
;
4832 em
->len
= num_bytes
;
4833 em
->block_start
= 0;
4834 em
->block_len
= em
->len
;
4835 em
->orig_block_len
= stripe_size
;
4837 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
4838 write_lock(&em_tree
->lock
);
4839 ret
= add_extent_mapping(em_tree
, em
, 0);
4841 list_add_tail(&em
->list
, &trans
->transaction
->pending_chunks
);
4842 atomic_inc(&em
->refs
);
4844 write_unlock(&em_tree
->lock
);
4846 free_extent_map(em
);
4850 ret
= btrfs_make_block_group(trans
, extent_root
, 0, type
,
4851 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4854 goto error_del_extent
;
4856 for (i
= 0; i
< map
->num_stripes
; i
++) {
4857 num_bytes
= map
->stripes
[i
].dev
->bytes_used
+ stripe_size
;
4858 btrfs_device_set_bytes_used(map
->stripes
[i
].dev
, num_bytes
);
4861 spin_lock(&extent_root
->fs_info
->free_chunk_lock
);
4862 extent_root
->fs_info
->free_chunk_space
-= (stripe_size
*
4864 spin_unlock(&extent_root
->fs_info
->free_chunk_lock
);
4866 free_extent_map(em
);
4867 check_raid56_incompat_flag(extent_root
->fs_info
, type
);
4869 kfree(devices_info
);
4873 write_lock(&em_tree
->lock
);
4874 remove_extent_mapping(em_tree
, em
);
4875 write_unlock(&em_tree
->lock
);
4877 /* One for our allocation */
4878 free_extent_map(em
);
4879 /* One for the tree reference */
4880 free_extent_map(em
);
4881 /* One for the pending_chunks list reference */
4882 free_extent_map(em
);
4884 kfree(devices_info
);
4888 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
4889 struct btrfs_root
*extent_root
,
4890 u64 chunk_offset
, u64 chunk_size
)
4892 struct btrfs_key key
;
4893 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
4894 struct btrfs_device
*device
;
4895 struct btrfs_chunk
*chunk
;
4896 struct btrfs_stripe
*stripe
;
4897 struct extent_map_tree
*em_tree
;
4898 struct extent_map
*em
;
4899 struct map_lookup
*map
;
4906 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
4907 read_lock(&em_tree
->lock
);
4908 em
= lookup_extent_mapping(em_tree
, chunk_offset
, chunk_size
);
4909 read_unlock(&em_tree
->lock
);
4912 btrfs_crit(extent_root
->fs_info
, "unable to find logical "
4913 "%Lu len %Lu", chunk_offset
, chunk_size
);
4917 if (em
->start
!= chunk_offset
|| em
->len
!= chunk_size
) {
4918 btrfs_crit(extent_root
->fs_info
, "found a bad mapping, wanted"
4919 " %Lu-%Lu, found %Lu-%Lu", chunk_offset
,
4920 chunk_size
, em
->start
, em
->len
);
4921 free_extent_map(em
);
4925 map
= em
->map_lookup
;
4926 item_size
= btrfs_chunk_item_size(map
->num_stripes
);
4927 stripe_size
= em
->orig_block_len
;
4929 chunk
= kzalloc(item_size
, GFP_NOFS
);
4936 * Take the device list mutex to prevent races with the final phase of
4937 * a device replace operation that replaces the device object associated
4938 * with the map's stripes, because the device object's id can change
4939 * at any time during that final phase of the device replace operation
4940 * (dev-replace.c:btrfs_dev_replace_finishing()).
4942 mutex_lock(&chunk_root
->fs_info
->fs_devices
->device_list_mutex
);
4943 for (i
= 0; i
< map
->num_stripes
; i
++) {
4944 device
= map
->stripes
[i
].dev
;
4945 dev_offset
= map
->stripes
[i
].physical
;
4947 ret
= btrfs_update_device(trans
, device
);
4950 ret
= btrfs_alloc_dev_extent(trans
, device
,
4951 chunk_root
->root_key
.objectid
,
4952 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4953 chunk_offset
, dev_offset
,
4959 mutex_unlock(&chunk_root
->fs_info
->fs_devices
->device_list_mutex
);
4963 stripe
= &chunk
->stripe
;
4964 for (i
= 0; i
< map
->num_stripes
; i
++) {
4965 device
= map
->stripes
[i
].dev
;
4966 dev_offset
= map
->stripes
[i
].physical
;
4968 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
4969 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
4970 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
4973 mutex_unlock(&chunk_root
->fs_info
->fs_devices
->device_list_mutex
);
4975 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
4976 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
4977 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
4978 btrfs_set_stack_chunk_type(chunk
, map
->type
);
4979 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
4980 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
4981 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
4982 btrfs_set_stack_chunk_sector_size(chunk
, extent_root
->sectorsize
);
4983 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
4985 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
4986 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
4987 key
.offset
= chunk_offset
;
4989 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
4990 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4992 * TODO: Cleanup of inserted chunk root in case of
4995 ret
= btrfs_add_system_chunk(chunk_root
, &key
, chunk
,
5001 free_extent_map(em
);
5006 * Chunk allocation falls into two parts. The first part does works
5007 * that make the new allocated chunk useable, but not do any operation
5008 * that modifies the chunk tree. The second part does the works that
5009 * require modifying the chunk tree. This division is important for the
5010 * bootstrap process of adding storage to a seed btrfs.
5012 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
5013 struct btrfs_root
*extent_root
, u64 type
)
5017 ASSERT(mutex_is_locked(&extent_root
->fs_info
->chunk_mutex
));
5018 chunk_offset
= find_next_chunk(extent_root
->fs_info
);
5019 return __btrfs_alloc_chunk(trans
, extent_root
, chunk_offset
, type
);
5022 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
5023 struct btrfs_root
*root
,
5024 struct btrfs_device
*device
)
5027 u64 sys_chunk_offset
;
5029 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5030 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
5033 chunk_offset
= find_next_chunk(fs_info
);
5034 alloc_profile
= btrfs_get_alloc_profile(extent_root
, 0);
5035 ret
= __btrfs_alloc_chunk(trans
, extent_root
, chunk_offset
,
5040 sys_chunk_offset
= find_next_chunk(root
->fs_info
);
5041 alloc_profile
= btrfs_get_alloc_profile(fs_info
->chunk_root
, 0);
5042 ret
= __btrfs_alloc_chunk(trans
, extent_root
, sys_chunk_offset
,
5047 static inline int btrfs_chunk_max_errors(struct map_lookup
*map
)
5051 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
5052 BTRFS_BLOCK_GROUP_RAID10
|
5053 BTRFS_BLOCK_GROUP_RAID5
|
5054 BTRFS_BLOCK_GROUP_DUP
)) {
5056 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
) {
5065 int btrfs_chunk_readonly(struct btrfs_root
*root
, u64 chunk_offset
)
5067 struct extent_map
*em
;
5068 struct map_lookup
*map
;
5069 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
5074 read_lock(&map_tree
->map_tree
.lock
);
5075 em
= lookup_extent_mapping(&map_tree
->map_tree
, chunk_offset
, 1);
5076 read_unlock(&map_tree
->map_tree
.lock
);
5080 map
= em
->map_lookup
;
5081 for (i
= 0; i
< map
->num_stripes
; i
++) {
5082 if (map
->stripes
[i
].dev
->missing
) {
5087 if (!map
->stripes
[i
].dev
->writeable
) {
5094 * If the number of missing devices is larger than max errors,
5095 * we can not write the data into that chunk successfully, so
5098 if (miss_ndevs
> btrfs_chunk_max_errors(map
))
5101 free_extent_map(em
);
5105 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
5107 extent_map_tree_init(&tree
->map_tree
);
5110 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
5112 struct extent_map
*em
;
5115 write_lock(&tree
->map_tree
.lock
);
5116 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
5118 remove_extent_mapping(&tree
->map_tree
, em
);
5119 write_unlock(&tree
->map_tree
.lock
);
5123 free_extent_map(em
);
5124 /* once for the tree */
5125 free_extent_map(em
);
5129 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
5131 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
5132 struct extent_map
*em
;
5133 struct map_lookup
*map
;
5134 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5137 read_lock(&em_tree
->lock
);
5138 em
= lookup_extent_mapping(em_tree
, logical
, len
);
5139 read_unlock(&em_tree
->lock
);
5142 * We could return errors for these cases, but that could get ugly and
5143 * we'd probably do the same thing which is just not do anything else
5144 * and exit, so return 1 so the callers don't try to use other copies.
5147 btrfs_crit(fs_info
, "No mapping for %Lu-%Lu", logical
,
5152 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
5153 btrfs_crit(fs_info
, "Invalid mapping for %Lu-%Lu, got "
5154 "%Lu-%Lu", logical
, logical
+len
, em
->start
,
5155 em
->start
+ em
->len
);
5156 free_extent_map(em
);
5160 map
= em
->map_lookup
;
5161 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
5162 ret
= map
->num_stripes
;
5163 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5164 ret
= map
->sub_stripes
;
5165 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID5
)
5167 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5171 free_extent_map(em
);
5173 btrfs_dev_replace_lock(&fs_info
->dev_replace
, 0);
5174 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
))
5176 btrfs_dev_replace_unlock(&fs_info
->dev_replace
, 0);
5181 unsigned long btrfs_full_stripe_len(struct btrfs_root
*root
,
5182 struct btrfs_mapping_tree
*map_tree
,
5185 struct extent_map
*em
;
5186 struct map_lookup
*map
;
5187 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5188 unsigned long len
= root
->sectorsize
;
5190 read_lock(&em_tree
->lock
);
5191 em
= lookup_extent_mapping(em_tree
, logical
, len
);
5192 read_unlock(&em_tree
->lock
);
5195 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
5196 map
= em
->map_lookup
;
5197 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5198 len
= map
->stripe_len
* nr_data_stripes(map
);
5199 free_extent_map(em
);
5203 int btrfs_is_parity_mirror(struct btrfs_mapping_tree
*map_tree
,
5204 u64 logical
, u64 len
, int mirror_num
)
5206 struct extent_map
*em
;
5207 struct map_lookup
*map
;
5208 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5211 read_lock(&em_tree
->lock
);
5212 em
= lookup_extent_mapping(em_tree
, logical
, len
);
5213 read_unlock(&em_tree
->lock
);
5216 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
5217 map
= em
->map_lookup
;
5218 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5220 free_extent_map(em
);
5224 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
5225 struct map_lookup
*map
, int first
, int num
,
5226 int optimal
, int dev_replace_is_ongoing
)
5230 struct btrfs_device
*srcdev
;
5232 if (dev_replace_is_ongoing
&&
5233 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
5234 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
5235 srcdev
= fs_info
->dev_replace
.srcdev
;
5240 * try to avoid the drive that is the source drive for a
5241 * dev-replace procedure, only choose it if no other non-missing
5242 * mirror is available
5244 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
5245 if (map
->stripes
[optimal
].dev
->bdev
&&
5246 (tolerance
|| map
->stripes
[optimal
].dev
!= srcdev
))
5248 for (i
= first
; i
< first
+ num
; i
++) {
5249 if (map
->stripes
[i
].dev
->bdev
&&
5250 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
5255 /* we couldn't find one that doesn't fail. Just return something
5256 * and the io error handling code will clean up eventually
5261 static inline int parity_smaller(u64 a
, u64 b
)
5266 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5267 static void sort_parity_stripes(struct btrfs_bio
*bbio
, int num_stripes
)
5269 struct btrfs_bio_stripe s
;
5276 for (i
= 0; i
< num_stripes
- 1; i
++) {
5277 if (parity_smaller(bbio
->raid_map
[i
],
5278 bbio
->raid_map
[i
+1])) {
5279 s
= bbio
->stripes
[i
];
5280 l
= bbio
->raid_map
[i
];
5281 bbio
->stripes
[i
] = bbio
->stripes
[i
+1];
5282 bbio
->raid_map
[i
] = bbio
->raid_map
[i
+1];
5283 bbio
->stripes
[i
+1] = s
;
5284 bbio
->raid_map
[i
+1] = l
;
5292 static struct btrfs_bio
*alloc_btrfs_bio(int total_stripes
, int real_stripes
)
5294 struct btrfs_bio
*bbio
= kzalloc(
5295 /* the size of the btrfs_bio */
5296 sizeof(struct btrfs_bio
) +
5297 /* plus the variable array for the stripes */
5298 sizeof(struct btrfs_bio_stripe
) * (total_stripes
) +
5299 /* plus the variable array for the tgt dev */
5300 sizeof(int) * (real_stripes
) +
5302 * plus the raid_map, which includes both the tgt dev
5305 sizeof(u64
) * (total_stripes
),
5306 GFP_NOFS
|__GFP_NOFAIL
);
5308 atomic_set(&bbio
->error
, 0);
5309 atomic_set(&bbio
->refs
, 1);
5314 void btrfs_get_bbio(struct btrfs_bio
*bbio
)
5316 WARN_ON(!atomic_read(&bbio
->refs
));
5317 atomic_inc(&bbio
->refs
);
5320 void btrfs_put_bbio(struct btrfs_bio
*bbio
)
5324 if (atomic_dec_and_test(&bbio
->refs
))
5328 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
, int op
,
5329 u64 logical
, u64
*length
,
5330 struct btrfs_bio
**bbio_ret
,
5331 int mirror_num
, int need_raid_map
)
5333 struct extent_map
*em
;
5334 struct map_lookup
*map
;
5335 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
5336 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5339 u64 stripe_end_offset
;
5349 int tgtdev_indexes
= 0;
5350 struct btrfs_bio
*bbio
= NULL
;
5351 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
5352 int dev_replace_is_ongoing
= 0;
5353 int num_alloc_stripes
;
5354 int patch_the_first_stripe_for_dev_replace
= 0;
5355 u64 physical_to_patch_in_first_stripe
= 0;
5356 u64 raid56_full_stripe_start
= (u64
)-1;
5358 read_lock(&em_tree
->lock
);
5359 em
= lookup_extent_mapping(em_tree
, logical
, *length
);
5360 read_unlock(&em_tree
->lock
);
5363 btrfs_crit(fs_info
, "unable to find logical %llu len %llu",
5368 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
5369 btrfs_crit(fs_info
, "found a bad mapping, wanted %Lu, "
5370 "found %Lu-%Lu", logical
, em
->start
,
5371 em
->start
+ em
->len
);
5372 free_extent_map(em
);
5376 map
= em
->map_lookup
;
5377 offset
= logical
- em
->start
;
5379 stripe_len
= map
->stripe_len
;
5382 * stripe_nr counts the total number of stripes we have to stride
5383 * to get to this block
5385 stripe_nr
= div64_u64(stripe_nr
, stripe_len
);
5387 stripe_offset
= stripe_nr
* stripe_len
;
5388 if (offset
< stripe_offset
) {
5389 btrfs_crit(fs_info
, "stripe math has gone wrong, "
5390 "stripe_offset=%llu, offset=%llu, start=%llu, "
5391 "logical=%llu, stripe_len=%llu",
5392 stripe_offset
, offset
, em
->start
, logical
,
5394 free_extent_map(em
);
5398 /* stripe_offset is the offset of this block in its stripe*/
5399 stripe_offset
= offset
- stripe_offset
;
5401 /* if we're here for raid56, we need to know the stripe aligned start */
5402 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5403 unsigned long full_stripe_len
= stripe_len
* nr_data_stripes(map
);
5404 raid56_full_stripe_start
= offset
;
5406 /* allow a write of a full stripe, but make sure we don't
5407 * allow straddling of stripes
5409 raid56_full_stripe_start
= div64_u64(raid56_full_stripe_start
,
5411 raid56_full_stripe_start
*= full_stripe_len
;
5414 if (op
== REQ_OP_DISCARD
) {
5415 /* we don't discard raid56 yet */
5416 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5420 *length
= min_t(u64
, em
->len
- offset
, *length
);
5421 } else if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
5423 /* For writes to RAID[56], allow a full stripeset across all disks.
5424 For other RAID types and for RAID[56] reads, just allow a single
5425 stripe (on a single disk). */
5426 if ((map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
5427 (op
== REQ_OP_WRITE
)) {
5428 max_len
= stripe_len
* nr_data_stripes(map
) -
5429 (offset
- raid56_full_stripe_start
);
5431 /* we limit the length of each bio to what fits in a stripe */
5432 max_len
= stripe_len
- stripe_offset
;
5434 *length
= min_t(u64
, em
->len
- offset
, max_len
);
5436 *length
= em
->len
- offset
;
5439 /* This is for when we're called from btrfs_merge_bio_hook() and all
5440 it cares about is the length */
5444 btrfs_dev_replace_lock(dev_replace
, 0);
5445 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
5446 if (!dev_replace_is_ongoing
)
5447 btrfs_dev_replace_unlock(dev_replace
, 0);
5449 btrfs_dev_replace_set_lock_blocking(dev_replace
);
5451 if (dev_replace_is_ongoing
&& mirror_num
== map
->num_stripes
+ 1 &&
5452 op
!= REQ_OP_WRITE
&& op
!= REQ_OP_DISCARD
&&
5453 op
!= REQ_GET_READ_MIRRORS
&& dev_replace
->tgtdev
!= NULL
) {
5455 * in dev-replace case, for repair case (that's the only
5456 * case where the mirror is selected explicitly when
5457 * calling btrfs_map_block), blocks left of the left cursor
5458 * can also be read from the target drive.
5459 * For REQ_GET_READ_MIRRORS, the target drive is added as
5460 * the last one to the array of stripes. For READ, it also
5461 * needs to be supported using the same mirror number.
5462 * If the requested block is not left of the left cursor,
5463 * EIO is returned. This can happen because btrfs_num_copies()
5464 * returns one more in the dev-replace case.
5466 u64 tmp_length
= *length
;
5467 struct btrfs_bio
*tmp_bbio
= NULL
;
5468 int tmp_num_stripes
;
5469 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5470 int index_srcdev
= 0;
5472 u64 physical_of_found
= 0;
5474 ret
= __btrfs_map_block(fs_info
, REQ_GET_READ_MIRRORS
,
5475 logical
, &tmp_length
, &tmp_bbio
, 0, 0);
5477 WARN_ON(tmp_bbio
!= NULL
);
5481 tmp_num_stripes
= tmp_bbio
->num_stripes
;
5482 if (mirror_num
> tmp_num_stripes
) {
5484 * REQ_GET_READ_MIRRORS does not contain this
5485 * mirror, that means that the requested area
5486 * is not left of the left cursor
5489 btrfs_put_bbio(tmp_bbio
);
5494 * process the rest of the function using the mirror_num
5495 * of the source drive. Therefore look it up first.
5496 * At the end, patch the device pointer to the one of the
5499 for (i
= 0; i
< tmp_num_stripes
; i
++) {
5500 if (tmp_bbio
->stripes
[i
].dev
->devid
!= srcdev_devid
)
5504 * In case of DUP, in order to keep it simple, only add
5505 * the mirror with the lowest physical address
5508 physical_of_found
<= tmp_bbio
->stripes
[i
].physical
)
5513 physical_of_found
= tmp_bbio
->stripes
[i
].physical
;
5516 btrfs_put_bbio(tmp_bbio
);
5524 mirror_num
= index_srcdev
+ 1;
5525 patch_the_first_stripe_for_dev_replace
= 1;
5526 physical_to_patch_in_first_stripe
= physical_of_found
;
5527 } else if (mirror_num
> map
->num_stripes
) {
5533 stripe_nr_orig
= stripe_nr
;
5534 stripe_nr_end
= ALIGN(offset
+ *length
, map
->stripe_len
);
5535 stripe_nr_end
= div_u64(stripe_nr_end
, map
->stripe_len
);
5536 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
5539 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5540 if (op
== REQ_OP_DISCARD
)
5541 num_stripes
= min_t(u64
, map
->num_stripes
,
5542 stripe_nr_end
- stripe_nr_orig
);
5543 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5545 if (op
!= REQ_OP_WRITE
&& op
!= REQ_OP_DISCARD
&&
5546 op
!= REQ_GET_READ_MIRRORS
)
5548 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
5549 if (op
== REQ_OP_WRITE
|| op
== REQ_OP_DISCARD
||
5550 op
== REQ_GET_READ_MIRRORS
)
5551 num_stripes
= map
->num_stripes
;
5552 else if (mirror_num
)
5553 stripe_index
= mirror_num
- 1;
5555 stripe_index
= find_live_mirror(fs_info
, map
, 0,
5557 current
->pid
% map
->num_stripes
,
5558 dev_replace_is_ongoing
);
5559 mirror_num
= stripe_index
+ 1;
5562 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
5563 if (op
== REQ_OP_WRITE
|| op
== REQ_OP_DISCARD
||
5564 op
== REQ_GET_READ_MIRRORS
) {
5565 num_stripes
= map
->num_stripes
;
5566 } else if (mirror_num
) {
5567 stripe_index
= mirror_num
- 1;
5572 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5573 u32 factor
= map
->num_stripes
/ map
->sub_stripes
;
5575 stripe_nr
= div_u64_rem(stripe_nr
, factor
, &stripe_index
);
5576 stripe_index
*= map
->sub_stripes
;
5578 if (op
== REQ_OP_WRITE
|| op
== REQ_GET_READ_MIRRORS
)
5579 num_stripes
= map
->sub_stripes
;
5580 else if (op
== REQ_OP_DISCARD
)
5581 num_stripes
= min_t(u64
, map
->sub_stripes
*
5582 (stripe_nr_end
- stripe_nr_orig
),
5584 else if (mirror_num
)
5585 stripe_index
+= mirror_num
- 1;
5587 int old_stripe_index
= stripe_index
;
5588 stripe_index
= find_live_mirror(fs_info
, map
,
5590 map
->sub_stripes
, stripe_index
+
5591 current
->pid
% map
->sub_stripes
,
5592 dev_replace_is_ongoing
);
5593 mirror_num
= stripe_index
- old_stripe_index
+ 1;
5596 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5597 if (need_raid_map
&&
5598 (op
== REQ_OP_WRITE
|| op
== REQ_GET_READ_MIRRORS
||
5600 /* push stripe_nr back to the start of the full stripe */
5601 stripe_nr
= div_u64(raid56_full_stripe_start
,
5602 stripe_len
* nr_data_stripes(map
));
5604 /* RAID[56] write or recovery. Return all stripes */
5605 num_stripes
= map
->num_stripes
;
5606 max_errors
= nr_parity_stripes(map
);
5608 *length
= map
->stripe_len
;
5613 * Mirror #0 or #1 means the original data block.
5614 * Mirror #2 is RAID5 parity block.
5615 * Mirror #3 is RAID6 Q block.
5617 stripe_nr
= div_u64_rem(stripe_nr
,
5618 nr_data_stripes(map
), &stripe_index
);
5620 stripe_index
= nr_data_stripes(map
) +
5623 /* We distribute the parity blocks across stripes */
5624 div_u64_rem(stripe_nr
+ stripe_index
, map
->num_stripes
,
5626 if ((op
!= REQ_OP_WRITE
&& op
!= REQ_OP_DISCARD
&&
5627 op
!= REQ_GET_READ_MIRRORS
) && mirror_num
<= 1)
5632 * after this, stripe_nr is the number of stripes on this
5633 * device we have to walk to find the data, and stripe_index is
5634 * the number of our device in the stripe array
5636 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5638 mirror_num
= stripe_index
+ 1;
5640 if (stripe_index
>= map
->num_stripes
) {
5641 btrfs_crit(fs_info
, "stripe index math went horribly wrong, "
5642 "got stripe_index=%u, num_stripes=%u",
5643 stripe_index
, map
->num_stripes
);
5648 num_alloc_stripes
= num_stripes
;
5649 if (dev_replace_is_ongoing
) {
5650 if (op
== REQ_OP_WRITE
|| op
== REQ_OP_DISCARD
)
5651 num_alloc_stripes
<<= 1;
5652 if (op
== REQ_GET_READ_MIRRORS
)
5653 num_alloc_stripes
++;
5654 tgtdev_indexes
= num_stripes
;
5657 bbio
= alloc_btrfs_bio(num_alloc_stripes
, tgtdev_indexes
);
5662 if (dev_replace_is_ongoing
)
5663 bbio
->tgtdev_map
= (int *)(bbio
->stripes
+ num_alloc_stripes
);
5665 /* build raid_map */
5666 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
&&
5668 ((op
== REQ_OP_WRITE
|| op
== REQ_GET_READ_MIRRORS
) ||
5673 bbio
->raid_map
= (u64
*)((void *)bbio
->stripes
+
5674 sizeof(struct btrfs_bio_stripe
) *
5676 sizeof(int) * tgtdev_indexes
);
5678 /* Work out the disk rotation on this stripe-set */
5679 div_u64_rem(stripe_nr
, num_stripes
, &rot
);
5681 /* Fill in the logical address of each stripe */
5682 tmp
= stripe_nr
* nr_data_stripes(map
);
5683 for (i
= 0; i
< nr_data_stripes(map
); i
++)
5684 bbio
->raid_map
[(i
+rot
) % num_stripes
] =
5685 em
->start
+ (tmp
+ i
) * map
->stripe_len
;
5687 bbio
->raid_map
[(i
+rot
) % map
->num_stripes
] = RAID5_P_STRIPE
;
5688 if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5689 bbio
->raid_map
[(i
+rot
+1) % num_stripes
] =
5693 if (op
== REQ_OP_DISCARD
) {
5695 u32 sub_stripes
= 0;
5696 u64 stripes_per_dev
= 0;
5697 u32 remaining_stripes
= 0;
5698 u32 last_stripe
= 0;
5701 (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID10
)) {
5702 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5705 sub_stripes
= map
->sub_stripes
;
5707 factor
= map
->num_stripes
/ sub_stripes
;
5708 stripes_per_dev
= div_u64_rem(stripe_nr_end
-
5711 &remaining_stripes
);
5712 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
5713 last_stripe
*= sub_stripes
;
5716 for (i
= 0; i
< num_stripes
; i
++) {
5717 bbio
->stripes
[i
].physical
=
5718 map
->stripes
[stripe_index
].physical
+
5719 stripe_offset
+ stripe_nr
* map
->stripe_len
;
5720 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
5722 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5723 BTRFS_BLOCK_GROUP_RAID10
)) {
5724 bbio
->stripes
[i
].length
= stripes_per_dev
*
5727 if (i
/ sub_stripes
< remaining_stripes
)
5728 bbio
->stripes
[i
].length
+=
5732 * Special for the first stripe and
5735 * |-------|...|-------|
5739 if (i
< sub_stripes
)
5740 bbio
->stripes
[i
].length
-=
5743 if (stripe_index
>= last_stripe
&&
5744 stripe_index
<= (last_stripe
+
5746 bbio
->stripes
[i
].length
-=
5749 if (i
== sub_stripes
- 1)
5752 bbio
->stripes
[i
].length
= *length
;
5755 if (stripe_index
== map
->num_stripes
) {
5756 /* This could only happen for RAID0/10 */
5762 for (i
= 0; i
< num_stripes
; i
++) {
5763 bbio
->stripes
[i
].physical
=
5764 map
->stripes
[stripe_index
].physical
+
5766 stripe_nr
* map
->stripe_len
;
5767 bbio
->stripes
[i
].dev
=
5768 map
->stripes
[stripe_index
].dev
;
5773 if (op
== REQ_OP_WRITE
|| op
== REQ_GET_READ_MIRRORS
)
5774 max_errors
= btrfs_chunk_max_errors(map
);
5777 sort_parity_stripes(bbio
, num_stripes
);
5780 if (dev_replace_is_ongoing
&&
5781 (op
== REQ_OP_WRITE
|| op
== REQ_OP_DISCARD
) &&
5782 dev_replace
->tgtdev
!= NULL
) {
5783 int index_where_to_add
;
5784 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5787 * duplicate the write operations while the dev replace
5788 * procedure is running. Since the copying of the old disk
5789 * to the new disk takes place at run time while the
5790 * filesystem is mounted writable, the regular write
5791 * operations to the old disk have to be duplicated to go
5792 * to the new disk as well.
5793 * Note that device->missing is handled by the caller, and
5794 * that the write to the old disk is already set up in the
5797 index_where_to_add
= num_stripes
;
5798 for (i
= 0; i
< num_stripes
; i
++) {
5799 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5800 /* write to new disk, too */
5801 struct btrfs_bio_stripe
*new =
5802 bbio
->stripes
+ index_where_to_add
;
5803 struct btrfs_bio_stripe
*old
=
5806 new->physical
= old
->physical
;
5807 new->length
= old
->length
;
5808 new->dev
= dev_replace
->tgtdev
;
5809 bbio
->tgtdev_map
[i
] = index_where_to_add
;
5810 index_where_to_add
++;
5815 num_stripes
= index_where_to_add
;
5816 } else if (dev_replace_is_ongoing
&& (op
== REQ_GET_READ_MIRRORS
) &&
5817 dev_replace
->tgtdev
!= NULL
) {
5818 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5819 int index_srcdev
= 0;
5821 u64 physical_of_found
= 0;
5824 * During the dev-replace procedure, the target drive can
5825 * also be used to read data in case it is needed to repair
5826 * a corrupt block elsewhere. This is possible if the
5827 * requested area is left of the left cursor. In this area,
5828 * the target drive is a full copy of the source drive.
5830 for (i
= 0; i
< num_stripes
; i
++) {
5831 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5833 * In case of DUP, in order to keep it
5834 * simple, only add the mirror with the
5835 * lowest physical address
5838 physical_of_found
<=
5839 bbio
->stripes
[i
].physical
)
5843 physical_of_found
= bbio
->stripes
[i
].physical
;
5847 struct btrfs_bio_stripe
*tgtdev_stripe
=
5848 bbio
->stripes
+ num_stripes
;
5850 tgtdev_stripe
->physical
= physical_of_found
;
5851 tgtdev_stripe
->length
=
5852 bbio
->stripes
[index_srcdev
].length
;
5853 tgtdev_stripe
->dev
= dev_replace
->tgtdev
;
5854 bbio
->tgtdev_map
[index_srcdev
] = num_stripes
;
5862 bbio
->map_type
= map
->type
;
5863 bbio
->num_stripes
= num_stripes
;
5864 bbio
->max_errors
= max_errors
;
5865 bbio
->mirror_num
= mirror_num
;
5866 bbio
->num_tgtdevs
= tgtdev_indexes
;
5869 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5870 * mirror_num == num_stripes + 1 && dev_replace target drive is
5871 * available as a mirror
5873 if (patch_the_first_stripe_for_dev_replace
&& num_stripes
> 0) {
5874 WARN_ON(num_stripes
> 1);
5875 bbio
->stripes
[0].dev
= dev_replace
->tgtdev
;
5876 bbio
->stripes
[0].physical
= physical_to_patch_in_first_stripe
;
5877 bbio
->mirror_num
= map
->num_stripes
+ 1;
5880 if (dev_replace_is_ongoing
) {
5881 btrfs_dev_replace_clear_lock_blocking(dev_replace
);
5882 btrfs_dev_replace_unlock(dev_replace
, 0);
5884 free_extent_map(em
);
5888 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, int op
,
5889 u64 logical
, u64
*length
,
5890 struct btrfs_bio
**bbio_ret
, int mirror_num
)
5892 return __btrfs_map_block(fs_info
, op
, logical
, length
, bbio_ret
,
5896 /* For Scrub/replace */
5897 int btrfs_map_sblock(struct btrfs_fs_info
*fs_info
, int op
,
5898 u64 logical
, u64
*length
,
5899 struct btrfs_bio
**bbio_ret
, int mirror_num
,
5902 return __btrfs_map_block(fs_info
, op
, logical
, length
, bbio_ret
,
5903 mirror_num
, need_raid_map
);
5906 int btrfs_rmap_block(struct btrfs_mapping_tree
*map_tree
,
5907 u64 chunk_start
, u64 physical
, u64 devid
,
5908 u64
**logical
, int *naddrs
, int *stripe_len
)
5910 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5911 struct extent_map
*em
;
5912 struct map_lookup
*map
;
5920 read_lock(&em_tree
->lock
);
5921 em
= lookup_extent_mapping(em_tree
, chunk_start
, 1);
5922 read_unlock(&em_tree
->lock
);
5925 printk(KERN_ERR
"BTRFS: couldn't find em for chunk %Lu\n",
5930 if (em
->start
!= chunk_start
) {
5931 printk(KERN_ERR
"BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5932 em
->start
, chunk_start
);
5933 free_extent_map(em
);
5936 map
= em
->map_lookup
;
5939 rmap_len
= map
->stripe_len
;
5941 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5942 length
= div_u64(length
, map
->num_stripes
/ map
->sub_stripes
);
5943 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5944 length
= div_u64(length
, map
->num_stripes
);
5945 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5946 length
= div_u64(length
, nr_data_stripes(map
));
5947 rmap_len
= map
->stripe_len
* nr_data_stripes(map
);
5950 buf
= kcalloc(map
->num_stripes
, sizeof(u64
), GFP_NOFS
);
5951 BUG_ON(!buf
); /* -ENOMEM */
5953 for (i
= 0; i
< map
->num_stripes
; i
++) {
5954 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
5956 if (map
->stripes
[i
].physical
> physical
||
5957 map
->stripes
[i
].physical
+ length
<= physical
)
5960 stripe_nr
= physical
- map
->stripes
[i
].physical
;
5961 stripe_nr
= div_u64(stripe_nr
, map
->stripe_len
);
5963 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5964 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5965 stripe_nr
= div_u64(stripe_nr
, map
->sub_stripes
);
5966 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5967 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5968 } /* else if RAID[56], multiply by nr_data_stripes().
5969 * Alternatively, just use rmap_len below instead of
5970 * map->stripe_len */
5972 bytenr
= chunk_start
+ stripe_nr
* rmap_len
;
5973 WARN_ON(nr
>= map
->num_stripes
);
5974 for (j
= 0; j
< nr
; j
++) {
5975 if (buf
[j
] == bytenr
)
5979 WARN_ON(nr
>= map
->num_stripes
);
5986 *stripe_len
= rmap_len
;
5988 free_extent_map(em
);
5992 static inline void btrfs_end_bbio(struct btrfs_bio
*bbio
, struct bio
*bio
)
5994 bio
->bi_private
= bbio
->private;
5995 bio
->bi_end_io
= bbio
->end_io
;
5998 btrfs_put_bbio(bbio
);
6001 static void btrfs_end_bio(struct bio
*bio
)
6003 struct btrfs_bio
*bbio
= bio
->bi_private
;
6004 int is_orig_bio
= 0;
6006 if (bio
->bi_error
) {
6007 atomic_inc(&bbio
->error
);
6008 if (bio
->bi_error
== -EIO
|| bio
->bi_error
== -EREMOTEIO
) {
6009 unsigned int stripe_index
=
6010 btrfs_io_bio(bio
)->stripe_index
;
6011 struct btrfs_device
*dev
;
6013 BUG_ON(stripe_index
>= bbio
->num_stripes
);
6014 dev
= bbio
->stripes
[stripe_index
].dev
;
6016 if (bio_op(bio
) == REQ_OP_WRITE
)
6017 btrfs_dev_stat_inc(dev
,
6018 BTRFS_DEV_STAT_WRITE_ERRS
);
6020 btrfs_dev_stat_inc(dev
,
6021 BTRFS_DEV_STAT_READ_ERRS
);
6022 if ((bio
->bi_opf
& WRITE_FLUSH
) == WRITE_FLUSH
)
6023 btrfs_dev_stat_inc(dev
,
6024 BTRFS_DEV_STAT_FLUSH_ERRS
);
6025 btrfs_dev_stat_print_on_error(dev
);
6030 if (bio
== bbio
->orig_bio
)
6033 btrfs_bio_counter_dec(bbio
->fs_info
);
6035 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
6038 bio
= bbio
->orig_bio
;
6041 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
6042 /* only send an error to the higher layers if it is
6043 * beyond the tolerance of the btrfs bio
6045 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
6046 bio
->bi_error
= -EIO
;
6049 * this bio is actually up to date, we didn't
6050 * go over the max number of errors
6055 btrfs_end_bbio(bbio
, bio
);
6056 } else if (!is_orig_bio
) {
6062 * see run_scheduled_bios for a description of why bios are collected for
6065 * This will add one bio to the pending list for a device and make sure
6066 * the work struct is scheduled.
6068 static noinline
void btrfs_schedule_bio(struct btrfs_root
*root
,
6069 struct btrfs_device
*device
,
6072 int should_queue
= 1;
6073 struct btrfs_pending_bios
*pending_bios
;
6075 if (device
->missing
|| !device
->bdev
) {
6080 /* don't bother with additional async steps for reads, right now */
6081 if (bio_op(bio
) == REQ_OP_READ
) {
6083 btrfsic_submit_bio(bio
);
6089 * nr_async_bios allows us to reliably return congestion to the
6090 * higher layers. Otherwise, the async bio makes it appear we have
6091 * made progress against dirty pages when we've really just put it
6092 * on a queue for later
6094 atomic_inc(&root
->fs_info
->nr_async_bios
);
6095 WARN_ON(bio
->bi_next
);
6096 bio
->bi_next
= NULL
;
6098 spin_lock(&device
->io_lock
);
6099 if (bio
->bi_opf
& REQ_SYNC
)
6100 pending_bios
= &device
->pending_sync_bios
;
6102 pending_bios
= &device
->pending_bios
;
6104 if (pending_bios
->tail
)
6105 pending_bios
->tail
->bi_next
= bio
;
6107 pending_bios
->tail
= bio
;
6108 if (!pending_bios
->head
)
6109 pending_bios
->head
= bio
;
6110 if (device
->running_pending
)
6113 spin_unlock(&device
->io_lock
);
6116 btrfs_queue_work(root
->fs_info
->submit_workers
,
6120 static void submit_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
6121 struct bio
*bio
, u64 physical
, int dev_nr
,
6124 struct btrfs_device
*dev
= bbio
->stripes
[dev_nr
].dev
;
6126 bio
->bi_private
= bbio
;
6127 btrfs_io_bio(bio
)->stripe_index
= dev_nr
;
6128 bio
->bi_end_io
= btrfs_end_bio
;
6129 bio
->bi_iter
.bi_sector
= physical
>> 9;
6132 struct rcu_string
*name
;
6135 name
= rcu_dereference(dev
->name
);
6136 pr_debug("btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu "
6137 "(%s id %llu), size=%u\n", bio_op(bio
), bio
->bi_opf
,
6138 (u64
)bio
->bi_iter
.bi_sector
, (u_long
)dev
->bdev
->bd_dev
,
6139 name
->str
, dev
->devid
, bio
->bi_iter
.bi_size
);
6143 bio
->bi_bdev
= dev
->bdev
;
6145 btrfs_bio_counter_inc_noblocked(root
->fs_info
);
6148 btrfs_schedule_bio(root
, dev
, bio
);
6150 btrfsic_submit_bio(bio
);
6153 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
6155 atomic_inc(&bbio
->error
);
6156 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
6157 /* Should be the original bio. */
6158 WARN_ON(bio
!= bbio
->orig_bio
);
6160 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
6161 bio
->bi_iter
.bi_sector
= logical
>> 9;
6162 bio
->bi_error
= -EIO
;
6163 btrfs_end_bbio(bbio
, bio
);
6167 int btrfs_map_bio(struct btrfs_root
*root
, struct bio
*bio
,
6168 int mirror_num
, int async_submit
)
6170 struct btrfs_device
*dev
;
6171 struct bio
*first_bio
= bio
;
6172 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
6178 struct btrfs_bio
*bbio
= NULL
;
6180 length
= bio
->bi_iter
.bi_size
;
6181 map_length
= length
;
6183 btrfs_bio_counter_inc_blocked(root
->fs_info
);
6184 ret
= __btrfs_map_block(root
->fs_info
, bio_op(bio
), logical
,
6185 &map_length
, &bbio
, mirror_num
, 1);
6187 btrfs_bio_counter_dec(root
->fs_info
);
6191 total_devs
= bbio
->num_stripes
;
6192 bbio
->orig_bio
= first_bio
;
6193 bbio
->private = first_bio
->bi_private
;
6194 bbio
->end_io
= first_bio
->bi_end_io
;
6195 bbio
->fs_info
= root
->fs_info
;
6196 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
6198 if ((bbio
->map_type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
6199 ((bio_op(bio
) == REQ_OP_WRITE
) || (mirror_num
> 1))) {
6200 /* In this case, map_length has been set to the length of
6201 a single stripe; not the whole write */
6202 if (bio_op(bio
) == REQ_OP_WRITE
) {
6203 ret
= raid56_parity_write(root
, bio
, bbio
, map_length
);
6205 ret
= raid56_parity_recover(root
, bio
, bbio
, map_length
,
6209 btrfs_bio_counter_dec(root
->fs_info
);
6213 if (map_length
< length
) {
6214 btrfs_crit(root
->fs_info
, "mapping failed logical %llu bio len %llu len %llu",
6215 logical
, length
, map_length
);
6219 for (dev_nr
= 0; dev_nr
< total_devs
; dev_nr
++) {
6220 dev
= bbio
->stripes
[dev_nr
].dev
;
6221 if (!dev
|| !dev
->bdev
||
6222 (bio_op(bio
) == REQ_OP_WRITE
&& !dev
->writeable
)) {
6223 bbio_error(bbio
, first_bio
, logical
);
6227 if (dev_nr
< total_devs
- 1) {
6228 bio
= btrfs_bio_clone(first_bio
, GFP_NOFS
);
6229 BUG_ON(!bio
); /* -ENOMEM */
6233 submit_stripe_bio(root
, bbio
, bio
,
6234 bbio
->stripes
[dev_nr
].physical
, dev_nr
,
6237 btrfs_bio_counter_dec(root
->fs_info
);
6241 struct btrfs_device
*btrfs_find_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
6244 struct btrfs_device
*device
;
6245 struct btrfs_fs_devices
*cur_devices
;
6247 cur_devices
= fs_info
->fs_devices
;
6248 while (cur_devices
) {
6250 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
6251 device
= __find_device(&cur_devices
->devices
,
6256 cur_devices
= cur_devices
->seed
;
6261 static struct btrfs_device
*add_missing_dev(struct btrfs_root
*root
,
6262 struct btrfs_fs_devices
*fs_devices
,
6263 u64 devid
, u8
*dev_uuid
)
6265 struct btrfs_device
*device
;
6267 device
= btrfs_alloc_device(NULL
, &devid
, dev_uuid
);
6271 list_add(&device
->dev_list
, &fs_devices
->devices
);
6272 device
->fs_devices
= fs_devices
;
6273 fs_devices
->num_devices
++;
6275 device
->missing
= 1;
6276 fs_devices
->missing_devices
++;
6282 * btrfs_alloc_device - allocate struct btrfs_device
6283 * @fs_info: used only for generating a new devid, can be NULL if
6284 * devid is provided (i.e. @devid != NULL).
6285 * @devid: a pointer to devid for this device. If NULL a new devid
6287 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6290 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6291 * on error. Returned struct is not linked onto any lists and can be
6292 * destroyed with kfree() right away.
6294 struct btrfs_device
*btrfs_alloc_device(struct btrfs_fs_info
*fs_info
,
6298 struct btrfs_device
*dev
;
6301 if (WARN_ON(!devid
&& !fs_info
))
6302 return ERR_PTR(-EINVAL
);
6304 dev
= __alloc_device();
6313 ret
= find_next_devid(fs_info
, &tmp
);
6316 return ERR_PTR(ret
);
6322 memcpy(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
);
6324 generate_random_uuid(dev
->uuid
);
6326 btrfs_init_work(&dev
->work
, btrfs_submit_helper
,
6327 pending_bios_fn
, NULL
, NULL
);
6332 /* Return -EIO if any error, otherwise return 0. */
6333 static int btrfs_check_chunk_valid(struct btrfs_root
*root
,
6334 struct extent_buffer
*leaf
,
6335 struct btrfs_chunk
*chunk
, u64 logical
)
6343 length
= btrfs_chunk_length(leaf
, chunk
);
6344 stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6345 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6346 sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6347 type
= btrfs_chunk_type(leaf
, chunk
);
6350 btrfs_err(root
->fs_info
, "invalid chunk num_stripes: %u",
6354 if (!IS_ALIGNED(logical
, root
->sectorsize
)) {
6355 btrfs_err(root
->fs_info
,
6356 "invalid chunk logical %llu", logical
);
6359 if (btrfs_chunk_sector_size(leaf
, chunk
) != root
->sectorsize
) {
6360 btrfs_err(root
->fs_info
, "invalid chunk sectorsize %u",
6361 btrfs_chunk_sector_size(leaf
, chunk
));
6364 if (!length
|| !IS_ALIGNED(length
, root
->sectorsize
)) {
6365 btrfs_err(root
->fs_info
,
6366 "invalid chunk length %llu", length
);
6369 if (!is_power_of_2(stripe_len
) || stripe_len
!= BTRFS_STRIPE_LEN
) {
6370 btrfs_err(root
->fs_info
, "invalid chunk stripe length: %llu",
6374 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK
| BTRFS_BLOCK_GROUP_PROFILE_MASK
) &
6376 btrfs_err(root
->fs_info
, "unrecognized chunk type: %llu",
6377 ~(BTRFS_BLOCK_GROUP_TYPE_MASK
|
6378 BTRFS_BLOCK_GROUP_PROFILE_MASK
) &
6379 btrfs_chunk_type(leaf
, chunk
));
6382 if ((type
& BTRFS_BLOCK_GROUP_RAID10
&& sub_stripes
!= 2) ||
6383 (type
& BTRFS_BLOCK_GROUP_RAID1
&& num_stripes
< 1) ||
6384 (type
& BTRFS_BLOCK_GROUP_RAID5
&& num_stripes
< 2) ||
6385 (type
& BTRFS_BLOCK_GROUP_RAID6
&& num_stripes
< 3) ||
6386 (type
& BTRFS_BLOCK_GROUP_DUP
&& num_stripes
> 2) ||
6387 ((type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) == 0 &&
6388 num_stripes
!= 1)) {
6389 btrfs_err(root
->fs_info
,
6390 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6391 num_stripes
, sub_stripes
,
6392 type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
);
6399 static int read_one_chunk(struct btrfs_root
*root
, struct btrfs_key
*key
,
6400 struct extent_buffer
*leaf
,
6401 struct btrfs_chunk
*chunk
)
6403 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
6404 struct map_lookup
*map
;
6405 struct extent_map
*em
;
6410 u8 uuid
[BTRFS_UUID_SIZE
];
6415 logical
= key
->offset
;
6416 length
= btrfs_chunk_length(leaf
, chunk
);
6417 stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6418 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6420 ret
= btrfs_check_chunk_valid(root
, leaf
, chunk
, logical
);
6424 read_lock(&map_tree
->map_tree
.lock
);
6425 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
6426 read_unlock(&map_tree
->map_tree
.lock
);
6428 /* already mapped? */
6429 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
6430 free_extent_map(em
);
6433 free_extent_map(em
);
6436 em
= alloc_extent_map();
6439 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
6441 free_extent_map(em
);
6445 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
6446 em
->map_lookup
= map
;
6447 em
->start
= logical
;
6450 em
->block_start
= 0;
6451 em
->block_len
= em
->len
;
6453 map
->num_stripes
= num_stripes
;
6454 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
6455 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
6456 map
->sector_size
= btrfs_chunk_sector_size(leaf
, chunk
);
6457 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6458 map
->type
= btrfs_chunk_type(leaf
, chunk
);
6459 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6460 for (i
= 0; i
< num_stripes
; i
++) {
6461 map
->stripes
[i
].physical
=
6462 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
6463 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
6464 read_extent_buffer(leaf
, uuid
, (unsigned long)
6465 btrfs_stripe_dev_uuid_nr(chunk
, i
),
6467 map
->stripes
[i
].dev
= btrfs_find_device(root
->fs_info
, devid
,
6469 if (!map
->stripes
[i
].dev
&&
6470 !btrfs_test_opt(root
->fs_info
, DEGRADED
)) {
6471 free_extent_map(em
);
6474 if (!map
->stripes
[i
].dev
) {
6475 map
->stripes
[i
].dev
=
6476 add_missing_dev(root
, root
->fs_info
->fs_devices
,
6478 if (!map
->stripes
[i
].dev
) {
6479 free_extent_map(em
);
6482 btrfs_warn(root
->fs_info
, "devid %llu uuid %pU is missing",
6485 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
6488 write_lock(&map_tree
->map_tree
.lock
);
6489 ret
= add_extent_mapping(&map_tree
->map_tree
, em
, 0);
6490 write_unlock(&map_tree
->map_tree
.lock
);
6491 BUG_ON(ret
); /* Tree corruption */
6492 free_extent_map(em
);
6497 static void fill_device_from_item(struct extent_buffer
*leaf
,
6498 struct btrfs_dev_item
*dev_item
,
6499 struct btrfs_device
*device
)
6503 device
->devid
= btrfs_device_id(leaf
, dev_item
);
6504 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
6505 device
->total_bytes
= device
->disk_total_bytes
;
6506 device
->commit_total_bytes
= device
->disk_total_bytes
;
6507 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
6508 device
->commit_bytes_used
= device
->bytes_used
;
6509 device
->type
= btrfs_device_type(leaf
, dev_item
);
6510 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
6511 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
6512 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
6513 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
6514 device
->is_tgtdev_for_dev_replace
= 0;
6516 ptr
= btrfs_device_uuid(dev_item
);
6517 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
6520 static struct btrfs_fs_devices
*open_seed_devices(struct btrfs_root
*root
,
6523 struct btrfs_fs_devices
*fs_devices
;
6526 BUG_ON(!mutex_is_locked(&uuid_mutex
));
6528 fs_devices
= root
->fs_info
->fs_devices
->seed
;
6529 while (fs_devices
) {
6530 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
))
6533 fs_devices
= fs_devices
->seed
;
6536 fs_devices
= find_fsid(fsid
);
6538 if (!btrfs_test_opt(root
->fs_info
, DEGRADED
))
6539 return ERR_PTR(-ENOENT
);
6541 fs_devices
= alloc_fs_devices(fsid
);
6542 if (IS_ERR(fs_devices
))
6545 fs_devices
->seeding
= 1;
6546 fs_devices
->opened
= 1;
6550 fs_devices
= clone_fs_devices(fs_devices
);
6551 if (IS_ERR(fs_devices
))
6554 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
6555 root
->fs_info
->bdev_holder
);
6557 free_fs_devices(fs_devices
);
6558 fs_devices
= ERR_PTR(ret
);
6562 if (!fs_devices
->seeding
) {
6563 __btrfs_close_devices(fs_devices
);
6564 free_fs_devices(fs_devices
);
6565 fs_devices
= ERR_PTR(-EINVAL
);
6569 fs_devices
->seed
= root
->fs_info
->fs_devices
->seed
;
6570 root
->fs_info
->fs_devices
->seed
= fs_devices
;
6575 static int read_one_dev(struct btrfs_root
*root
,
6576 struct extent_buffer
*leaf
,
6577 struct btrfs_dev_item
*dev_item
)
6579 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
6580 struct btrfs_device
*device
;
6583 u8 fs_uuid
[BTRFS_UUID_SIZE
];
6584 u8 dev_uuid
[BTRFS_UUID_SIZE
];
6586 devid
= btrfs_device_id(leaf
, dev_item
);
6587 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
6589 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
6592 if (memcmp(fs_uuid
, root
->fs_info
->fsid
, BTRFS_UUID_SIZE
)) {
6593 fs_devices
= open_seed_devices(root
, fs_uuid
);
6594 if (IS_ERR(fs_devices
))
6595 return PTR_ERR(fs_devices
);
6598 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
, fs_uuid
);
6600 if (!btrfs_test_opt(root
->fs_info
, DEGRADED
))
6603 device
= add_missing_dev(root
, fs_devices
, devid
, dev_uuid
);
6606 btrfs_warn(root
->fs_info
, "devid %llu uuid %pU missing",
6609 if (!device
->bdev
&& !btrfs_test_opt(root
->fs_info
, DEGRADED
))
6612 if(!device
->bdev
&& !device
->missing
) {
6614 * this happens when a device that was properly setup
6615 * in the device info lists suddenly goes bad.
6616 * device->bdev is NULL, and so we have to set
6617 * device->missing to one here
6619 device
->fs_devices
->missing_devices
++;
6620 device
->missing
= 1;
6623 /* Move the device to its own fs_devices */
6624 if (device
->fs_devices
!= fs_devices
) {
6625 ASSERT(device
->missing
);
6627 list_move(&device
->dev_list
, &fs_devices
->devices
);
6628 device
->fs_devices
->num_devices
--;
6629 fs_devices
->num_devices
++;
6631 device
->fs_devices
->missing_devices
--;
6632 fs_devices
->missing_devices
++;
6634 device
->fs_devices
= fs_devices
;
6638 if (device
->fs_devices
!= root
->fs_info
->fs_devices
) {
6639 BUG_ON(device
->writeable
);
6640 if (device
->generation
!=
6641 btrfs_device_generation(leaf
, dev_item
))
6645 fill_device_from_item(leaf
, dev_item
, device
);
6646 device
->in_fs_metadata
= 1;
6647 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
6648 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
6649 spin_lock(&root
->fs_info
->free_chunk_lock
);
6650 root
->fs_info
->free_chunk_space
+= device
->total_bytes
-
6652 spin_unlock(&root
->fs_info
->free_chunk_lock
);
6658 int btrfs_read_sys_array(struct btrfs_root
*root
)
6660 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
6661 struct extent_buffer
*sb
;
6662 struct btrfs_disk_key
*disk_key
;
6663 struct btrfs_chunk
*chunk
;
6665 unsigned long sb_array_offset
;
6672 struct btrfs_key key
;
6674 ASSERT(BTRFS_SUPER_INFO_SIZE
<= root
->nodesize
);
6676 * This will create extent buffer of nodesize, superblock size is
6677 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6678 * overallocate but we can keep it as-is, only the first page is used.
6680 sb
= btrfs_find_create_tree_block(root
, BTRFS_SUPER_INFO_OFFSET
);
6683 set_extent_buffer_uptodate(sb
);
6684 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
6686 * The sb extent buffer is artificial and just used to read the system array.
6687 * set_extent_buffer_uptodate() call does not properly mark all it's
6688 * pages up-to-date when the page is larger: extent does not cover the
6689 * whole page and consequently check_page_uptodate does not find all
6690 * the page's extents up-to-date (the hole beyond sb),
6691 * write_extent_buffer then triggers a WARN_ON.
6693 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6694 * but sb spans only this function. Add an explicit SetPageUptodate call
6695 * to silence the warning eg. on PowerPC 64.
6697 if (PAGE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
6698 SetPageUptodate(sb
->pages
[0]);
6700 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
6701 array_size
= btrfs_super_sys_array_size(super_copy
);
6703 array_ptr
= super_copy
->sys_chunk_array
;
6704 sb_array_offset
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
6707 while (cur_offset
< array_size
) {
6708 disk_key
= (struct btrfs_disk_key
*)array_ptr
;
6709 len
= sizeof(*disk_key
);
6710 if (cur_offset
+ len
> array_size
)
6711 goto out_short_read
;
6713 btrfs_disk_key_to_cpu(&key
, disk_key
);
6716 sb_array_offset
+= len
;
6719 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6720 chunk
= (struct btrfs_chunk
*)sb_array_offset
;
6722 * At least one btrfs_chunk with one stripe must be
6723 * present, exact stripe count check comes afterwards
6725 len
= btrfs_chunk_item_size(1);
6726 if (cur_offset
+ len
> array_size
)
6727 goto out_short_read
;
6729 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
6732 "BTRFS: invalid number of stripes %u in sys_array at offset %u\n",
6733 num_stripes
, cur_offset
);
6738 type
= btrfs_chunk_type(sb
, chunk
);
6739 if ((type
& BTRFS_BLOCK_GROUP_SYSTEM
) == 0) {
6740 btrfs_err(root
->fs_info
,
6741 "invalid chunk type %llu in sys_array at offset %u",
6747 len
= btrfs_chunk_item_size(num_stripes
);
6748 if (cur_offset
+ len
> array_size
)
6749 goto out_short_read
;
6751 ret
= read_one_chunk(root
, &key
, sb
, chunk
);
6756 "BTRFS: unexpected item type %u in sys_array at offset %u\n",
6757 (u32
)key
.type
, cur_offset
);
6762 sb_array_offset
+= len
;
6765 clear_extent_buffer_uptodate(sb
);
6766 free_extent_buffer_stale(sb
);
6770 printk(KERN_ERR
"BTRFS: sys_array too short to read %u bytes at offset %u\n",
6772 clear_extent_buffer_uptodate(sb
);
6773 free_extent_buffer_stale(sb
);
6777 int btrfs_read_chunk_tree(struct btrfs_root
*root
)
6779 struct btrfs_path
*path
;
6780 struct extent_buffer
*leaf
;
6781 struct btrfs_key key
;
6782 struct btrfs_key found_key
;
6787 root
= root
->fs_info
->chunk_root
;
6789 path
= btrfs_alloc_path();
6793 mutex_lock(&uuid_mutex
);
6797 * Read all device items, and then all the chunk items. All
6798 * device items are found before any chunk item (their object id
6799 * is smaller than the lowest possible object id for a chunk
6800 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6802 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
6805 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6809 leaf
= path
->nodes
[0];
6810 slot
= path
->slots
[0];
6811 if (slot
>= btrfs_header_nritems(leaf
)) {
6812 ret
= btrfs_next_leaf(root
, path
);
6819 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6820 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
6821 struct btrfs_dev_item
*dev_item
;
6822 dev_item
= btrfs_item_ptr(leaf
, slot
,
6823 struct btrfs_dev_item
);
6824 ret
= read_one_dev(root
, leaf
, dev_item
);
6828 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6829 struct btrfs_chunk
*chunk
;
6830 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
6831 ret
= read_one_chunk(root
, &found_key
, leaf
, chunk
);
6839 * After loading chunk tree, we've got all device information,
6840 * do another round of validation checks.
6842 if (total_dev
!= root
->fs_info
->fs_devices
->total_devices
) {
6843 btrfs_err(root
->fs_info
,
6844 "super_num_devices %llu mismatch with num_devices %llu found here",
6845 btrfs_super_num_devices(root
->fs_info
->super_copy
),
6850 if (btrfs_super_total_bytes(root
->fs_info
->super_copy
) <
6851 root
->fs_info
->fs_devices
->total_rw_bytes
) {
6852 btrfs_err(root
->fs_info
,
6853 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6854 btrfs_super_total_bytes(root
->fs_info
->super_copy
),
6855 root
->fs_info
->fs_devices
->total_rw_bytes
);
6861 unlock_chunks(root
);
6862 mutex_unlock(&uuid_mutex
);
6864 btrfs_free_path(path
);
6868 void btrfs_init_devices_late(struct btrfs_fs_info
*fs_info
)
6870 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6871 struct btrfs_device
*device
;
6873 while (fs_devices
) {
6874 mutex_lock(&fs_devices
->device_list_mutex
);
6875 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
)
6876 device
->dev_root
= fs_info
->dev_root
;
6877 mutex_unlock(&fs_devices
->device_list_mutex
);
6879 fs_devices
= fs_devices
->seed
;
6883 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
6887 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6888 btrfs_dev_stat_reset(dev
, i
);
6891 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
6893 struct btrfs_key key
;
6894 struct btrfs_key found_key
;
6895 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6896 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6897 struct extent_buffer
*eb
;
6900 struct btrfs_device
*device
;
6901 struct btrfs_path
*path
= NULL
;
6904 path
= btrfs_alloc_path();
6910 mutex_lock(&fs_devices
->device_list_mutex
);
6911 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6913 struct btrfs_dev_stats_item
*ptr
;
6915 key
.objectid
= BTRFS_DEV_STATS_OBJECTID
;
6916 key
.type
= BTRFS_PERSISTENT_ITEM_KEY
;
6917 key
.offset
= device
->devid
;
6918 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
6920 __btrfs_reset_dev_stats(device
);
6921 device
->dev_stats_valid
= 1;
6922 btrfs_release_path(path
);
6925 slot
= path
->slots
[0];
6926 eb
= path
->nodes
[0];
6927 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
6928 item_size
= btrfs_item_size_nr(eb
, slot
);
6930 ptr
= btrfs_item_ptr(eb
, slot
,
6931 struct btrfs_dev_stats_item
);
6933 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6934 if (item_size
>= (1 + i
) * sizeof(__le64
))
6935 btrfs_dev_stat_set(device
, i
,
6936 btrfs_dev_stats_value(eb
, ptr
, i
));
6938 btrfs_dev_stat_reset(device
, i
);
6941 device
->dev_stats_valid
= 1;
6942 btrfs_dev_stat_print_on_load(device
);
6943 btrfs_release_path(path
);
6945 mutex_unlock(&fs_devices
->device_list_mutex
);
6948 btrfs_free_path(path
);
6949 return ret
< 0 ? ret
: 0;
6952 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
6953 struct btrfs_root
*dev_root
,
6954 struct btrfs_device
*device
)
6956 struct btrfs_path
*path
;
6957 struct btrfs_key key
;
6958 struct extent_buffer
*eb
;
6959 struct btrfs_dev_stats_item
*ptr
;
6963 key
.objectid
= BTRFS_DEV_STATS_OBJECTID
;
6964 key
.type
= BTRFS_PERSISTENT_ITEM_KEY
;
6965 key
.offset
= device
->devid
;
6967 path
= btrfs_alloc_path();
6969 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
6971 btrfs_warn_in_rcu(dev_root
->fs_info
,
6972 "error %d while searching for dev_stats item for device %s",
6973 ret
, rcu_str_deref(device
->name
));
6978 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
6979 /* need to delete old one and insert a new one */
6980 ret
= btrfs_del_item(trans
, dev_root
, path
);
6982 btrfs_warn_in_rcu(dev_root
->fs_info
,
6983 "delete too small dev_stats item for device %s failed %d",
6984 rcu_str_deref(device
->name
), ret
);
6991 /* need to insert a new item */
6992 btrfs_release_path(path
);
6993 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
6994 &key
, sizeof(*ptr
));
6996 btrfs_warn_in_rcu(dev_root
->fs_info
,
6997 "insert dev_stats item for device %s failed %d",
6998 rcu_str_deref(device
->name
), ret
);
7003 eb
= path
->nodes
[0];
7004 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
7005 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7006 btrfs_set_dev_stats_value(eb
, ptr
, i
,
7007 btrfs_dev_stat_read(device
, i
));
7008 btrfs_mark_buffer_dirty(eb
);
7011 btrfs_free_path(path
);
7016 * called from commit_transaction. Writes all changed device stats to disk.
7018 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
7019 struct btrfs_fs_info
*fs_info
)
7021 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
7022 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7023 struct btrfs_device
*device
;
7027 mutex_lock(&fs_devices
->device_list_mutex
);
7028 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
7029 if (!device
->dev_stats_valid
|| !btrfs_dev_stats_dirty(device
))
7032 stats_cnt
= atomic_read(&device
->dev_stats_ccnt
);
7033 ret
= update_dev_stat_item(trans
, dev_root
, device
);
7035 atomic_sub(stats_cnt
, &device
->dev_stats_ccnt
);
7037 mutex_unlock(&fs_devices
->device_list_mutex
);
7042 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
7044 btrfs_dev_stat_inc(dev
, index
);
7045 btrfs_dev_stat_print_on_error(dev
);
7048 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
7050 if (!dev
->dev_stats_valid
)
7052 btrfs_err_rl_in_rcu(dev
->dev_root
->fs_info
,
7053 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7054 rcu_str_deref(dev
->name
),
7055 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
7056 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
7057 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
7058 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
7059 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
7062 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
7066 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7067 if (btrfs_dev_stat_read(dev
, i
) != 0)
7069 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
7070 return; /* all values == 0, suppress message */
7072 btrfs_info_in_rcu(dev
->dev_root
->fs_info
,
7073 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7074 rcu_str_deref(dev
->name
),
7075 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
7076 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
7077 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
7078 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
7079 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
7082 int btrfs_get_dev_stats(struct btrfs_root
*root
,
7083 struct btrfs_ioctl_get_dev_stats
*stats
)
7085 struct btrfs_device
*dev
;
7086 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
7089 mutex_lock(&fs_devices
->device_list_mutex
);
7090 dev
= btrfs_find_device(root
->fs_info
, stats
->devid
, NULL
, NULL
);
7091 mutex_unlock(&fs_devices
->device_list_mutex
);
7094 btrfs_warn(root
->fs_info
, "get dev_stats failed, device not found");
7096 } else if (!dev
->dev_stats_valid
) {
7097 btrfs_warn(root
->fs_info
, "get dev_stats failed, not yet valid");
7099 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
7100 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
7101 if (stats
->nr_items
> i
)
7103 btrfs_dev_stat_read_and_reset(dev
, i
);
7105 btrfs_dev_stat_reset(dev
, i
);
7108 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7109 if (stats
->nr_items
> i
)
7110 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
7112 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
7113 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
7117 void btrfs_scratch_superblocks(struct block_device
*bdev
, char *device_path
)
7119 struct buffer_head
*bh
;
7120 struct btrfs_super_block
*disk_super
;
7126 for (copy_num
= 0; copy_num
< BTRFS_SUPER_MIRROR_MAX
;
7129 if (btrfs_read_dev_one_super(bdev
, copy_num
, &bh
))
7132 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
7134 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
7135 set_buffer_dirty(bh
);
7136 sync_dirty_buffer(bh
);
7140 /* Notify udev that device has changed */
7141 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
7143 /* Update ctime/mtime for device path for libblkid */
7144 update_dev_time(device_path
);
7148 * Update the size of all devices, which is used for writing out the
7151 void btrfs_update_commit_device_size(struct btrfs_fs_info
*fs_info
)
7153 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7154 struct btrfs_device
*curr
, *next
;
7156 if (list_empty(&fs_devices
->resized_devices
))
7159 mutex_lock(&fs_devices
->device_list_mutex
);
7160 lock_chunks(fs_info
->dev_root
);
7161 list_for_each_entry_safe(curr
, next
, &fs_devices
->resized_devices
,
7163 list_del_init(&curr
->resized_list
);
7164 curr
->commit_total_bytes
= curr
->disk_total_bytes
;
7166 unlock_chunks(fs_info
->dev_root
);
7167 mutex_unlock(&fs_devices
->device_list_mutex
);
7170 /* Must be invoked during the transaction commit */
7171 void btrfs_update_commit_device_bytes_used(struct btrfs_root
*root
,
7172 struct btrfs_transaction
*transaction
)
7174 struct extent_map
*em
;
7175 struct map_lookup
*map
;
7176 struct btrfs_device
*dev
;
7179 if (list_empty(&transaction
->pending_chunks
))
7182 /* In order to kick the device replace finish process */
7184 list_for_each_entry(em
, &transaction
->pending_chunks
, list
) {
7185 map
= em
->map_lookup
;
7187 for (i
= 0; i
< map
->num_stripes
; i
++) {
7188 dev
= map
->stripes
[i
].dev
;
7189 dev
->commit_bytes_used
= dev
->bytes_used
;
7192 unlock_chunks(root
);
7195 void btrfs_set_fs_info_ptr(struct btrfs_fs_info
*fs_info
)
7197 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7198 while (fs_devices
) {
7199 fs_devices
->fs_info
= fs_info
;
7200 fs_devices
= fs_devices
->seed
;
7204 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info
*fs_info
)
7206 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7207 while (fs_devices
) {
7208 fs_devices
->fs_info
= NULL
;
7209 fs_devices
= fs_devices
->seed
;