2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/iocontext.h>
24 #include <linux/capability.h>
25 #include <linux/ratelimit.h>
26 #include <linux/kthread.h>
27 #include <linux/raid/pq.h>
28 #include <linux/semaphore.h>
29 #include <linux/uuid.h>
30 #include <asm/div64.h>
32 #include "extent_map.h"
34 #include "transaction.h"
35 #include "print-tree.h"
38 #include "async-thread.h"
39 #include "check-integrity.h"
40 #include "rcu-string.h"
42 #include "dev-replace.h"
45 const struct btrfs_raid_attr btrfs_raid_array
[BTRFS_NR_RAID_TYPES
] = {
46 [BTRFS_RAID_RAID10
] = {
49 .devs_max
= 0, /* 0 == as many as possible */
51 .tolerated_failures
= 1,
55 [BTRFS_RAID_RAID1
] = {
60 .tolerated_failures
= 1,
69 .tolerated_failures
= 0,
73 [BTRFS_RAID_RAID0
] = {
78 .tolerated_failures
= 0,
82 [BTRFS_RAID_SINGLE
] = {
87 .tolerated_failures
= 0,
91 [BTRFS_RAID_RAID5
] = {
96 .tolerated_failures
= 1,
100 [BTRFS_RAID_RAID6
] = {
105 .tolerated_failures
= 2,
111 const u64 btrfs_raid_group
[BTRFS_NR_RAID_TYPES
] = {
112 [BTRFS_RAID_RAID10
] = BTRFS_BLOCK_GROUP_RAID10
,
113 [BTRFS_RAID_RAID1
] = BTRFS_BLOCK_GROUP_RAID1
,
114 [BTRFS_RAID_DUP
] = BTRFS_BLOCK_GROUP_DUP
,
115 [BTRFS_RAID_RAID0
] = BTRFS_BLOCK_GROUP_RAID0
,
116 [BTRFS_RAID_SINGLE
] = 0,
117 [BTRFS_RAID_RAID5
] = BTRFS_BLOCK_GROUP_RAID5
,
118 [BTRFS_RAID_RAID6
] = BTRFS_BLOCK_GROUP_RAID6
,
122 * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
123 * condition is not met. Zero means there's no corresponding
124 * BTRFS_ERROR_DEV_*_NOT_MET value.
126 const int btrfs_raid_mindev_error
[BTRFS_NR_RAID_TYPES
] = {
127 [BTRFS_RAID_RAID10
] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET
,
128 [BTRFS_RAID_RAID1
] = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET
,
129 [BTRFS_RAID_DUP
] = 0,
130 [BTRFS_RAID_RAID0
] = 0,
131 [BTRFS_RAID_SINGLE
] = 0,
132 [BTRFS_RAID_RAID5
] = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET
,
133 [BTRFS_RAID_RAID6
] = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET
,
136 static int init_first_rw_device(struct btrfs_trans_handle
*trans
,
137 struct btrfs_fs_info
*fs_info
);
138 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info
*fs_info
);
139 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
);
140 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
);
141 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*device
);
142 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
,
143 enum btrfs_map_op op
,
144 u64 logical
, u64
*length
,
145 struct btrfs_bio
**bbio_ret
,
146 int mirror_num
, int need_raid_map
);
148 DEFINE_MUTEX(uuid_mutex
);
149 static LIST_HEAD(fs_uuids
);
150 struct list_head
*btrfs_get_fs_uuids(void)
156 * alloc_fs_devices - allocate struct btrfs_fs_devices
157 * @fsid: if not NULL, copy the uuid to fs_devices::fsid
159 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
160 * The returned struct is not linked onto any lists and can be destroyed with
161 * kfree() right away.
163 static struct btrfs_fs_devices
*alloc_fs_devices(const u8
*fsid
)
165 struct btrfs_fs_devices
*fs_devs
;
167 fs_devs
= kzalloc(sizeof(*fs_devs
), GFP_KERNEL
);
169 return ERR_PTR(-ENOMEM
);
171 mutex_init(&fs_devs
->device_list_mutex
);
173 INIT_LIST_HEAD(&fs_devs
->devices
);
174 INIT_LIST_HEAD(&fs_devs
->resized_devices
);
175 INIT_LIST_HEAD(&fs_devs
->alloc_list
);
176 INIT_LIST_HEAD(&fs_devs
->list
);
178 memcpy(fs_devs
->fsid
, fsid
, BTRFS_FSID_SIZE
);
183 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
185 struct btrfs_device
*device
;
186 WARN_ON(fs_devices
->opened
);
187 while (!list_empty(&fs_devices
->devices
)) {
188 device
= list_entry(fs_devices
->devices
.next
,
189 struct btrfs_device
, dev_list
);
190 list_del(&device
->dev_list
);
191 rcu_string_free(device
->name
);
197 static void btrfs_kobject_uevent(struct block_device
*bdev
,
198 enum kobject_action action
)
202 ret
= kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, action
);
204 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
206 kobject_name(&disk_to_dev(bdev
->bd_disk
)->kobj
),
207 &disk_to_dev(bdev
->bd_disk
)->kobj
);
210 void btrfs_cleanup_fs_uuids(void)
212 struct btrfs_fs_devices
*fs_devices
;
214 while (!list_empty(&fs_uuids
)) {
215 fs_devices
= list_entry(fs_uuids
.next
,
216 struct btrfs_fs_devices
, list
);
217 list_del(&fs_devices
->list
);
218 free_fs_devices(fs_devices
);
222 static struct btrfs_device
*__alloc_device(void)
224 struct btrfs_device
*dev
;
226 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
228 return ERR_PTR(-ENOMEM
);
231 * Preallocate a bio that's always going to be used for flushing device
232 * barriers and matches the device lifespan
234 dev
->flush_bio
= bio_alloc_bioset(GFP_KERNEL
, 0, NULL
);
235 if (!dev
->flush_bio
) {
237 return ERR_PTR(-ENOMEM
);
239 bio_get(dev
->flush_bio
);
241 INIT_LIST_HEAD(&dev
->dev_list
);
242 INIT_LIST_HEAD(&dev
->dev_alloc_list
);
243 INIT_LIST_HEAD(&dev
->resized_list
);
245 spin_lock_init(&dev
->io_lock
);
247 spin_lock_init(&dev
->reada_lock
);
248 atomic_set(&dev
->reada_in_flight
, 0);
249 atomic_set(&dev
->dev_stats_ccnt
, 0);
250 btrfs_device_data_ordered_init(dev
);
251 INIT_RADIX_TREE(&dev
->reada_zones
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
252 INIT_RADIX_TREE(&dev
->reada_extents
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
258 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
261 * If devid and uuid are both specified, the match must be exact, otherwise
262 * only devid is used.
264 static struct btrfs_device
*find_device(struct btrfs_fs_devices
*fs_devices
,
265 u64 devid
, const u8
*uuid
)
267 struct list_head
*head
= &fs_devices
->devices
;
268 struct btrfs_device
*dev
;
270 list_for_each_entry(dev
, head
, dev_list
) {
271 if (dev
->devid
== devid
&&
272 (!uuid
|| !memcmp(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
))) {
279 static noinline
struct btrfs_fs_devices
*find_fsid(u8
*fsid
)
281 struct btrfs_fs_devices
*fs_devices
;
283 list_for_each_entry(fs_devices
, &fs_uuids
, list
) {
284 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
291 btrfs_get_bdev_and_sb(const char *device_path
, fmode_t flags
, void *holder
,
292 int flush
, struct block_device
**bdev
,
293 struct buffer_head
**bh
)
297 *bdev
= blkdev_get_by_path(device_path
, flags
, holder
);
300 ret
= PTR_ERR(*bdev
);
305 filemap_write_and_wait((*bdev
)->bd_inode
->i_mapping
);
306 ret
= set_blocksize(*bdev
, BTRFS_BDEV_BLOCKSIZE
);
308 blkdev_put(*bdev
, flags
);
311 invalidate_bdev(*bdev
);
312 *bh
= btrfs_read_dev_super(*bdev
);
315 blkdev_put(*bdev
, flags
);
327 static void requeue_list(struct btrfs_pending_bios
*pending_bios
,
328 struct bio
*head
, struct bio
*tail
)
331 struct bio
*old_head
;
333 old_head
= pending_bios
->head
;
334 pending_bios
->head
= head
;
335 if (pending_bios
->tail
)
336 tail
->bi_next
= old_head
;
338 pending_bios
->tail
= tail
;
342 * we try to collect pending bios for a device so we don't get a large
343 * number of procs sending bios down to the same device. This greatly
344 * improves the schedulers ability to collect and merge the bios.
346 * But, it also turns into a long list of bios to process and that is sure
347 * to eventually make the worker thread block. The solution here is to
348 * make some progress and then put this work struct back at the end of
349 * the list if the block device is congested. This way, multiple devices
350 * can make progress from a single worker thread.
352 static noinline
void run_scheduled_bios(struct btrfs_device
*device
)
354 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
356 struct backing_dev_info
*bdi
;
357 struct btrfs_pending_bios
*pending_bios
;
361 unsigned long num_run
;
362 unsigned long batch_run
= 0;
364 unsigned long last_waited
= 0;
366 int sync_pending
= 0;
367 struct blk_plug plug
;
370 * this function runs all the bios we've collected for
371 * a particular device. We don't want to wander off to
372 * another device without first sending all of these down.
373 * So, setup a plug here and finish it off before we return
375 blk_start_plug(&plug
);
377 bdi
= device
->bdev
->bd_bdi
;
378 limit
= btrfs_async_submit_limit(fs_info
);
379 limit
= limit
* 2 / 3;
382 spin_lock(&device
->io_lock
);
387 /* take all the bios off the list at once and process them
388 * later on (without the lock held). But, remember the
389 * tail and other pointers so the bios can be properly reinserted
390 * into the list if we hit congestion
392 if (!force_reg
&& device
->pending_sync_bios
.head
) {
393 pending_bios
= &device
->pending_sync_bios
;
396 pending_bios
= &device
->pending_bios
;
400 pending
= pending_bios
->head
;
401 tail
= pending_bios
->tail
;
402 WARN_ON(pending
&& !tail
);
405 * if pending was null this time around, no bios need processing
406 * at all and we can stop. Otherwise it'll loop back up again
407 * and do an additional check so no bios are missed.
409 * device->running_pending is used to synchronize with the
412 if (device
->pending_sync_bios
.head
== NULL
&&
413 device
->pending_bios
.head
== NULL
) {
415 device
->running_pending
= 0;
418 device
->running_pending
= 1;
421 pending_bios
->head
= NULL
;
422 pending_bios
->tail
= NULL
;
424 spin_unlock(&device
->io_lock
);
429 /* we want to work on both lists, but do more bios on the
430 * sync list than the regular list
433 pending_bios
!= &device
->pending_sync_bios
&&
434 device
->pending_sync_bios
.head
) ||
435 (num_run
> 64 && pending_bios
== &device
->pending_sync_bios
&&
436 device
->pending_bios
.head
)) {
437 spin_lock(&device
->io_lock
);
438 requeue_list(pending_bios
, pending
, tail
);
443 pending
= pending
->bi_next
;
447 * atomic_dec_return implies a barrier for waitqueue_active
449 if (atomic_dec_return(&fs_info
->nr_async_bios
) < limit
&&
450 waitqueue_active(&fs_info
->async_submit_wait
))
451 wake_up(&fs_info
->async_submit_wait
);
453 BUG_ON(atomic_read(&cur
->__bi_cnt
) == 0);
456 * if we're doing the sync list, record that our
457 * plug has some sync requests on it
459 * If we're doing the regular list and there are
460 * sync requests sitting around, unplug before
463 if (pending_bios
== &device
->pending_sync_bios
) {
465 } else if (sync_pending
) {
466 blk_finish_plug(&plug
);
467 blk_start_plug(&plug
);
471 btrfsic_submit_bio(cur
);
478 * we made progress, there is more work to do and the bdi
479 * is now congested. Back off and let other work structs
482 if (pending
&& bdi_write_congested(bdi
) && batch_run
> 8 &&
483 fs_info
->fs_devices
->open_devices
> 1) {
484 struct io_context
*ioc
;
486 ioc
= current
->io_context
;
489 * the main goal here is that we don't want to
490 * block if we're going to be able to submit
491 * more requests without blocking.
493 * This code does two great things, it pokes into
494 * the elevator code from a filesystem _and_
495 * it makes assumptions about how batching works.
497 if (ioc
&& ioc
->nr_batch_requests
> 0 &&
498 time_before(jiffies
, ioc
->last_waited
+ HZ
/50UL) &&
500 ioc
->last_waited
== last_waited
)) {
502 * we want to go through our batch of
503 * requests and stop. So, we copy out
504 * the ioc->last_waited time and test
505 * against it before looping
507 last_waited
= ioc
->last_waited
;
511 spin_lock(&device
->io_lock
);
512 requeue_list(pending_bios
, pending
, tail
);
513 device
->running_pending
= 1;
515 spin_unlock(&device
->io_lock
);
516 btrfs_queue_work(fs_info
->submit_workers
,
520 /* unplug every 64 requests just for good measure */
521 if (batch_run
% 64 == 0) {
522 blk_finish_plug(&plug
);
523 blk_start_plug(&plug
);
532 spin_lock(&device
->io_lock
);
533 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
535 spin_unlock(&device
->io_lock
);
538 blk_finish_plug(&plug
);
541 static void pending_bios_fn(struct btrfs_work
*work
)
543 struct btrfs_device
*device
;
545 device
= container_of(work
, struct btrfs_device
, work
);
546 run_scheduled_bios(device
);
550 void btrfs_free_stale_device(struct btrfs_device
*cur_dev
)
552 struct btrfs_fs_devices
*fs_devs
;
553 struct btrfs_device
*dev
;
558 list_for_each_entry(fs_devs
, &fs_uuids
, list
) {
563 if (fs_devs
->seeding
)
566 list_for_each_entry(dev
, &fs_devs
->devices
, dev_list
) {
574 * Todo: This won't be enough. What if the same device
575 * comes back (with new uuid and) with its mapper path?
576 * But for now, this does help as mostly an admin will
577 * either use mapper or non mapper path throughout.
580 del
= strcmp(rcu_str_deref(dev
->name
),
581 rcu_str_deref(cur_dev
->name
));
588 /* delete the stale device */
589 if (fs_devs
->num_devices
== 1) {
590 btrfs_sysfs_remove_fsid(fs_devs
);
591 list_del(&fs_devs
->list
);
592 free_fs_devices(fs_devs
);
594 fs_devs
->num_devices
--;
595 list_del(&dev
->dev_list
);
596 rcu_string_free(dev
->name
);
605 * Add new device to list of registered devices
608 * 1 - first time device is seen
609 * 0 - device already known
612 static noinline
int device_list_add(const char *path
,
613 struct btrfs_super_block
*disk_super
,
614 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
616 struct btrfs_device
*device
;
617 struct btrfs_fs_devices
*fs_devices
;
618 struct rcu_string
*name
;
620 u64 found_transid
= btrfs_super_generation(disk_super
);
622 fs_devices
= find_fsid(disk_super
->fsid
);
624 fs_devices
= alloc_fs_devices(disk_super
->fsid
);
625 if (IS_ERR(fs_devices
))
626 return PTR_ERR(fs_devices
);
628 list_add(&fs_devices
->list
, &fs_uuids
);
632 device
= find_device(fs_devices
, devid
,
633 disk_super
->dev_item
.uuid
);
637 if (fs_devices
->opened
)
640 device
= btrfs_alloc_device(NULL
, &devid
,
641 disk_super
->dev_item
.uuid
);
642 if (IS_ERR(device
)) {
643 /* we can safely leave the fs_devices entry around */
644 return PTR_ERR(device
);
647 name
= rcu_string_strdup(path
, GFP_NOFS
);
652 rcu_assign_pointer(device
->name
, name
);
654 mutex_lock(&fs_devices
->device_list_mutex
);
655 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
656 fs_devices
->num_devices
++;
657 mutex_unlock(&fs_devices
->device_list_mutex
);
660 device
->fs_devices
= fs_devices
;
661 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
663 * When FS is already mounted.
664 * 1. If you are here and if the device->name is NULL that
665 * means this device was missing at time of FS mount.
666 * 2. If you are here and if the device->name is different
667 * from 'path' that means either
668 * a. The same device disappeared and reappeared with
670 * b. The missing-disk-which-was-replaced, has
673 * We must allow 1 and 2a above. But 2b would be a spurious
676 * Further in case of 1 and 2a above, the disk at 'path'
677 * would have missed some transaction when it was away and
678 * in case of 2a the stale bdev has to be updated as well.
679 * 2b must not be allowed at all time.
683 * For now, we do allow update to btrfs_fs_device through the
684 * btrfs dev scan cli after FS has been mounted. We're still
685 * tracking a problem where systems fail mount by subvolume id
686 * when we reject replacement on a mounted FS.
688 if (!fs_devices
->opened
&& found_transid
< device
->generation
) {
690 * That is if the FS is _not_ mounted and if you
691 * are here, that means there is more than one
692 * disk with same uuid and devid.We keep the one
693 * with larger generation number or the last-in if
694 * generation are equal.
699 name
= rcu_string_strdup(path
, GFP_NOFS
);
702 rcu_string_free(device
->name
);
703 rcu_assign_pointer(device
->name
, name
);
704 if (device
->missing
) {
705 fs_devices
->missing_devices
--;
711 * Unmount does not free the btrfs_device struct but would zero
712 * generation along with most of the other members. So just update
713 * it back. We need it to pick the disk with largest generation
716 if (!fs_devices
->opened
)
717 device
->generation
= found_transid
;
720 * if there is new btrfs on an already registered device,
721 * then remove the stale device entry.
724 btrfs_free_stale_device(device
);
726 *fs_devices_ret
= fs_devices
;
731 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
733 struct btrfs_fs_devices
*fs_devices
;
734 struct btrfs_device
*device
;
735 struct btrfs_device
*orig_dev
;
737 fs_devices
= alloc_fs_devices(orig
->fsid
);
738 if (IS_ERR(fs_devices
))
741 mutex_lock(&orig
->device_list_mutex
);
742 fs_devices
->total_devices
= orig
->total_devices
;
744 /* We have held the volume lock, it is safe to get the devices. */
745 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
746 struct rcu_string
*name
;
748 device
= btrfs_alloc_device(NULL
, &orig_dev
->devid
,
754 * This is ok to do without rcu read locked because we hold the
755 * uuid mutex so nothing we touch in here is going to disappear.
757 if (orig_dev
->name
) {
758 name
= rcu_string_strdup(orig_dev
->name
->str
,
764 rcu_assign_pointer(device
->name
, name
);
767 list_add(&device
->dev_list
, &fs_devices
->devices
);
768 device
->fs_devices
= fs_devices
;
769 fs_devices
->num_devices
++;
771 mutex_unlock(&orig
->device_list_mutex
);
774 mutex_unlock(&orig
->device_list_mutex
);
775 free_fs_devices(fs_devices
);
776 return ERR_PTR(-ENOMEM
);
779 void btrfs_close_extra_devices(struct btrfs_fs_devices
*fs_devices
, int step
)
781 struct btrfs_device
*device
, *next
;
782 struct btrfs_device
*latest_dev
= NULL
;
784 mutex_lock(&uuid_mutex
);
786 /* This is the initialized path, it is safe to release the devices. */
787 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
788 if (device
->in_fs_metadata
) {
789 if (!device
->is_tgtdev_for_dev_replace
&&
791 device
->generation
> latest_dev
->generation
)) {
797 if (device
->devid
== BTRFS_DEV_REPLACE_DEVID
) {
799 * In the first step, keep the device which has
800 * the correct fsid and the devid that is used
801 * for the dev_replace procedure.
802 * In the second step, the dev_replace state is
803 * read from the device tree and it is known
804 * whether the procedure is really active or
805 * not, which means whether this device is
806 * used or whether it should be removed.
808 if (step
== 0 || device
->is_tgtdev_for_dev_replace
) {
813 blkdev_put(device
->bdev
, device
->mode
);
815 fs_devices
->open_devices
--;
817 if (device
->writeable
) {
818 list_del_init(&device
->dev_alloc_list
);
819 device
->writeable
= 0;
820 if (!device
->is_tgtdev_for_dev_replace
)
821 fs_devices
->rw_devices
--;
823 list_del_init(&device
->dev_list
);
824 fs_devices
->num_devices
--;
825 rcu_string_free(device
->name
);
829 if (fs_devices
->seed
) {
830 fs_devices
= fs_devices
->seed
;
834 fs_devices
->latest_bdev
= latest_dev
->bdev
;
836 mutex_unlock(&uuid_mutex
);
839 static void __free_device(struct work_struct
*work
)
841 struct btrfs_device
*device
;
843 device
= container_of(work
, struct btrfs_device
, rcu_work
);
844 rcu_string_free(device
->name
);
845 bio_put(device
->flush_bio
);
849 static void free_device(struct rcu_head
*head
)
851 struct btrfs_device
*device
;
853 device
= container_of(head
, struct btrfs_device
, rcu
);
855 INIT_WORK(&device
->rcu_work
, __free_device
);
856 schedule_work(&device
->rcu_work
);
859 static void btrfs_close_bdev(struct btrfs_device
*device
)
861 if (device
->bdev
&& device
->writeable
) {
862 sync_blockdev(device
->bdev
);
863 invalidate_bdev(device
->bdev
);
867 blkdev_put(device
->bdev
, device
->mode
);
870 static void btrfs_prepare_close_one_device(struct btrfs_device
*device
)
872 struct btrfs_fs_devices
*fs_devices
= device
->fs_devices
;
873 struct btrfs_device
*new_device
;
874 struct rcu_string
*name
;
877 fs_devices
->open_devices
--;
879 if (device
->writeable
&&
880 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
881 list_del_init(&device
->dev_alloc_list
);
882 fs_devices
->rw_devices
--;
886 fs_devices
->missing_devices
--;
888 new_device
= btrfs_alloc_device(NULL
, &device
->devid
,
890 BUG_ON(IS_ERR(new_device
)); /* -ENOMEM */
892 /* Safe because we are under uuid_mutex */
894 name
= rcu_string_strdup(device
->name
->str
, GFP_NOFS
);
895 BUG_ON(!name
); /* -ENOMEM */
896 rcu_assign_pointer(new_device
->name
, name
);
899 list_replace_rcu(&device
->dev_list
, &new_device
->dev_list
);
900 new_device
->fs_devices
= device
->fs_devices
;
903 static int __btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
905 struct btrfs_device
*device
, *tmp
;
906 struct list_head pending_put
;
908 INIT_LIST_HEAD(&pending_put
);
910 if (--fs_devices
->opened
> 0)
913 mutex_lock(&fs_devices
->device_list_mutex
);
914 list_for_each_entry_safe(device
, tmp
, &fs_devices
->devices
, dev_list
) {
915 btrfs_prepare_close_one_device(device
);
916 list_add(&device
->dev_list
, &pending_put
);
918 mutex_unlock(&fs_devices
->device_list_mutex
);
921 * btrfs_show_devname() is using the device_list_mutex,
922 * sometimes call to blkdev_put() leads vfs calling
923 * into this func. So do put outside of device_list_mutex,
926 while (!list_empty(&pending_put
)) {
927 device
= list_first_entry(&pending_put
,
928 struct btrfs_device
, dev_list
);
929 list_del(&device
->dev_list
);
930 btrfs_close_bdev(device
);
931 call_rcu(&device
->rcu
, free_device
);
934 WARN_ON(fs_devices
->open_devices
);
935 WARN_ON(fs_devices
->rw_devices
);
936 fs_devices
->opened
= 0;
937 fs_devices
->seeding
= 0;
942 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
944 struct btrfs_fs_devices
*seed_devices
= NULL
;
947 mutex_lock(&uuid_mutex
);
948 ret
= __btrfs_close_devices(fs_devices
);
949 if (!fs_devices
->opened
) {
950 seed_devices
= fs_devices
->seed
;
951 fs_devices
->seed
= NULL
;
953 mutex_unlock(&uuid_mutex
);
955 while (seed_devices
) {
956 fs_devices
= seed_devices
;
957 seed_devices
= fs_devices
->seed
;
958 __btrfs_close_devices(fs_devices
);
959 free_fs_devices(fs_devices
);
962 * Wait for rcu kworkers under __btrfs_close_devices
963 * to finish all blkdev_puts so device is really
964 * free when umount is done.
970 static int __btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
971 fmode_t flags
, void *holder
)
973 struct request_queue
*q
;
974 struct block_device
*bdev
;
975 struct list_head
*head
= &fs_devices
->devices
;
976 struct btrfs_device
*device
;
977 struct btrfs_device
*latest_dev
= NULL
;
978 struct buffer_head
*bh
;
979 struct btrfs_super_block
*disk_super
;
986 list_for_each_entry(device
, head
, dev_list
) {
992 /* Just open everything we can; ignore failures here */
993 if (btrfs_get_bdev_and_sb(device
->name
->str
, flags
, holder
, 1,
997 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
998 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
999 if (devid
!= device
->devid
)
1002 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
,
1006 device
->generation
= btrfs_super_generation(disk_super
);
1008 device
->generation
> latest_dev
->generation
)
1009 latest_dev
= device
;
1011 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
1012 device
->writeable
= 0;
1014 device
->writeable
= !bdev_read_only(bdev
);
1018 q
= bdev_get_queue(bdev
);
1019 if (blk_queue_discard(q
))
1020 device
->can_discard
= 1;
1021 if (!blk_queue_nonrot(q
))
1022 fs_devices
->rotating
= 1;
1024 device
->bdev
= bdev
;
1025 device
->in_fs_metadata
= 0;
1026 device
->mode
= flags
;
1028 fs_devices
->open_devices
++;
1029 if (device
->writeable
&&
1030 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
1031 fs_devices
->rw_devices
++;
1032 list_add(&device
->dev_alloc_list
,
1033 &fs_devices
->alloc_list
);
1040 blkdev_put(bdev
, flags
);
1043 if (fs_devices
->open_devices
== 0) {
1047 fs_devices
->seeding
= seeding
;
1048 fs_devices
->opened
= 1;
1049 fs_devices
->latest_bdev
= latest_dev
->bdev
;
1050 fs_devices
->total_rw_bytes
= 0;
1055 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
1056 fmode_t flags
, void *holder
)
1060 mutex_lock(&uuid_mutex
);
1061 if (fs_devices
->opened
) {
1062 fs_devices
->opened
++;
1065 ret
= __btrfs_open_devices(fs_devices
, flags
, holder
);
1067 mutex_unlock(&uuid_mutex
);
1071 void btrfs_release_disk_super(struct page
*page
)
1077 int btrfs_read_disk_super(struct block_device
*bdev
, u64 bytenr
,
1078 struct page
**page
, struct btrfs_super_block
**disk_super
)
1083 /* make sure our super fits in the device */
1084 if (bytenr
+ PAGE_SIZE
>= i_size_read(bdev
->bd_inode
))
1087 /* make sure our super fits in the page */
1088 if (sizeof(**disk_super
) > PAGE_SIZE
)
1091 /* make sure our super doesn't straddle pages on disk */
1092 index
= bytenr
>> PAGE_SHIFT
;
1093 if ((bytenr
+ sizeof(**disk_super
) - 1) >> PAGE_SHIFT
!= index
)
1096 /* pull in the page with our super */
1097 *page
= read_cache_page_gfp(bdev
->bd_inode
->i_mapping
,
1100 if (IS_ERR_OR_NULL(*page
))
1105 /* align our pointer to the offset of the super block */
1106 *disk_super
= p
+ (bytenr
& ~PAGE_MASK
);
1108 if (btrfs_super_bytenr(*disk_super
) != bytenr
||
1109 btrfs_super_magic(*disk_super
) != BTRFS_MAGIC
) {
1110 btrfs_release_disk_super(*page
);
1114 if ((*disk_super
)->label
[0] &&
1115 (*disk_super
)->label
[BTRFS_LABEL_SIZE
- 1])
1116 (*disk_super
)->label
[BTRFS_LABEL_SIZE
- 1] = '\0';
1122 * Look for a btrfs signature on a device. This may be called out of the mount path
1123 * and we are not allowed to call set_blocksize during the scan. The superblock
1124 * is read via pagecache
1126 int btrfs_scan_one_device(const char *path
, fmode_t flags
, void *holder
,
1127 struct btrfs_fs_devices
**fs_devices_ret
)
1129 struct btrfs_super_block
*disk_super
;
1130 struct block_device
*bdev
;
1139 * we would like to check all the supers, but that would make
1140 * a btrfs mount succeed after a mkfs from a different FS.
1141 * So, we need to add a special mount option to scan for
1142 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1144 bytenr
= btrfs_sb_offset(0);
1145 flags
|= FMODE_EXCL
;
1146 mutex_lock(&uuid_mutex
);
1148 bdev
= blkdev_get_by_path(path
, flags
, holder
);
1150 ret
= PTR_ERR(bdev
);
1154 if (btrfs_read_disk_super(bdev
, bytenr
, &page
, &disk_super
))
1155 goto error_bdev_put
;
1157 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1158 transid
= btrfs_super_generation(disk_super
);
1159 total_devices
= btrfs_super_num_devices(disk_super
);
1161 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
1163 if (disk_super
->label
[0]) {
1164 pr_info("BTRFS: device label %s ", disk_super
->label
);
1166 pr_info("BTRFS: device fsid %pU ", disk_super
->fsid
);
1169 pr_cont("devid %llu transid %llu %s\n", devid
, transid
, path
);
1172 if (!ret
&& fs_devices_ret
)
1173 (*fs_devices_ret
)->total_devices
= total_devices
;
1175 btrfs_release_disk_super(page
);
1178 blkdev_put(bdev
, flags
);
1180 mutex_unlock(&uuid_mutex
);
1184 /* helper to account the used device space in the range */
1185 int btrfs_account_dev_extents_size(struct btrfs_device
*device
, u64 start
,
1186 u64 end
, u64
*length
)
1188 struct btrfs_key key
;
1189 struct btrfs_root
*root
= device
->fs_info
->dev_root
;
1190 struct btrfs_dev_extent
*dev_extent
;
1191 struct btrfs_path
*path
;
1195 struct extent_buffer
*l
;
1199 if (start
>= device
->total_bytes
|| device
->is_tgtdev_for_dev_replace
)
1202 path
= btrfs_alloc_path();
1205 path
->reada
= READA_FORWARD
;
1207 key
.objectid
= device
->devid
;
1209 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1211 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1215 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1222 slot
= path
->slots
[0];
1223 if (slot
>= btrfs_header_nritems(l
)) {
1224 ret
= btrfs_next_leaf(root
, path
);
1232 btrfs_item_key_to_cpu(l
, &key
, slot
);
1234 if (key
.objectid
< device
->devid
)
1237 if (key
.objectid
> device
->devid
)
1240 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1243 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1244 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1246 if (key
.offset
<= start
&& extent_end
> end
) {
1247 *length
= end
- start
+ 1;
1249 } else if (key
.offset
<= start
&& extent_end
> start
)
1250 *length
+= extent_end
- start
;
1251 else if (key
.offset
> start
&& extent_end
<= end
)
1252 *length
+= extent_end
- key
.offset
;
1253 else if (key
.offset
> start
&& key
.offset
<= end
) {
1254 *length
+= end
- key
.offset
+ 1;
1256 } else if (key
.offset
> end
)
1264 btrfs_free_path(path
);
1268 static int contains_pending_extent(struct btrfs_transaction
*transaction
,
1269 struct btrfs_device
*device
,
1270 u64
*start
, u64 len
)
1272 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1273 struct extent_map
*em
;
1274 struct list_head
*search_list
= &fs_info
->pinned_chunks
;
1276 u64 physical_start
= *start
;
1279 search_list
= &transaction
->pending_chunks
;
1281 list_for_each_entry(em
, search_list
, list
) {
1282 struct map_lookup
*map
;
1285 map
= em
->map_lookup
;
1286 for (i
= 0; i
< map
->num_stripes
; i
++) {
1289 if (map
->stripes
[i
].dev
!= device
)
1291 if (map
->stripes
[i
].physical
>= physical_start
+ len
||
1292 map
->stripes
[i
].physical
+ em
->orig_block_len
<=
1296 * Make sure that while processing the pinned list we do
1297 * not override our *start with a lower value, because
1298 * we can have pinned chunks that fall within this
1299 * device hole and that have lower physical addresses
1300 * than the pending chunks we processed before. If we
1301 * do not take this special care we can end up getting
1302 * 2 pending chunks that start at the same physical
1303 * device offsets because the end offset of a pinned
1304 * chunk can be equal to the start offset of some
1307 end
= map
->stripes
[i
].physical
+ em
->orig_block_len
;
1314 if (search_list
!= &fs_info
->pinned_chunks
) {
1315 search_list
= &fs_info
->pinned_chunks
;
1324 * find_free_dev_extent_start - find free space in the specified device
1325 * @device: the device which we search the free space in
1326 * @num_bytes: the size of the free space that we need
1327 * @search_start: the position from which to begin the search
1328 * @start: store the start of the free space.
1329 * @len: the size of the free space. that we find, or the size
1330 * of the max free space if we don't find suitable free space
1332 * this uses a pretty simple search, the expectation is that it is
1333 * called very infrequently and that a given device has a small number
1336 * @start is used to store the start of the free space if we find. But if we
1337 * don't find suitable free space, it will be used to store the start position
1338 * of the max free space.
1340 * @len is used to store the size of the free space that we find.
1341 * But if we don't find suitable free space, it is used to store the size of
1342 * the max free space.
1344 int find_free_dev_extent_start(struct btrfs_transaction
*transaction
,
1345 struct btrfs_device
*device
, u64 num_bytes
,
1346 u64 search_start
, u64
*start
, u64
*len
)
1348 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1349 struct btrfs_root
*root
= fs_info
->dev_root
;
1350 struct btrfs_key key
;
1351 struct btrfs_dev_extent
*dev_extent
;
1352 struct btrfs_path
*path
;
1357 u64 search_end
= device
->total_bytes
;
1360 struct extent_buffer
*l
;
1363 * We don't want to overwrite the superblock on the drive nor any area
1364 * used by the boot loader (grub for example), so we make sure to start
1365 * at an offset of at least 1MB.
1367 search_start
= max_t(u64
, search_start
, SZ_1M
);
1369 path
= btrfs_alloc_path();
1373 max_hole_start
= search_start
;
1377 if (search_start
>= search_end
|| device
->is_tgtdev_for_dev_replace
) {
1382 path
->reada
= READA_FORWARD
;
1383 path
->search_commit_root
= 1;
1384 path
->skip_locking
= 1;
1386 key
.objectid
= device
->devid
;
1387 key
.offset
= search_start
;
1388 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1390 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1394 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1401 slot
= path
->slots
[0];
1402 if (slot
>= btrfs_header_nritems(l
)) {
1403 ret
= btrfs_next_leaf(root
, path
);
1411 btrfs_item_key_to_cpu(l
, &key
, slot
);
1413 if (key
.objectid
< device
->devid
)
1416 if (key
.objectid
> device
->devid
)
1419 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1422 if (key
.offset
> search_start
) {
1423 hole_size
= key
.offset
- search_start
;
1426 * Have to check before we set max_hole_start, otherwise
1427 * we could end up sending back this offset anyway.
1429 if (contains_pending_extent(transaction
, device
,
1432 if (key
.offset
>= search_start
) {
1433 hole_size
= key
.offset
- search_start
;
1440 if (hole_size
> max_hole_size
) {
1441 max_hole_start
= search_start
;
1442 max_hole_size
= hole_size
;
1446 * If this free space is greater than which we need,
1447 * it must be the max free space that we have found
1448 * until now, so max_hole_start must point to the start
1449 * of this free space and the length of this free space
1450 * is stored in max_hole_size. Thus, we return
1451 * max_hole_start and max_hole_size and go back to the
1454 if (hole_size
>= num_bytes
) {
1460 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1461 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1463 if (extent_end
> search_start
)
1464 search_start
= extent_end
;
1471 * At this point, search_start should be the end of
1472 * allocated dev extents, and when shrinking the device,
1473 * search_end may be smaller than search_start.
1475 if (search_end
> search_start
) {
1476 hole_size
= search_end
- search_start
;
1478 if (contains_pending_extent(transaction
, device
, &search_start
,
1480 btrfs_release_path(path
);
1484 if (hole_size
> max_hole_size
) {
1485 max_hole_start
= search_start
;
1486 max_hole_size
= hole_size
;
1491 if (max_hole_size
< num_bytes
)
1497 btrfs_free_path(path
);
1498 *start
= max_hole_start
;
1500 *len
= max_hole_size
;
1504 int find_free_dev_extent(struct btrfs_trans_handle
*trans
,
1505 struct btrfs_device
*device
, u64 num_bytes
,
1506 u64
*start
, u64
*len
)
1508 /* FIXME use last free of some kind */
1509 return find_free_dev_extent_start(trans
->transaction
, device
,
1510 num_bytes
, 0, start
, len
);
1513 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1514 struct btrfs_device
*device
,
1515 u64 start
, u64
*dev_extent_len
)
1517 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1518 struct btrfs_root
*root
= fs_info
->dev_root
;
1520 struct btrfs_path
*path
;
1521 struct btrfs_key key
;
1522 struct btrfs_key found_key
;
1523 struct extent_buffer
*leaf
= NULL
;
1524 struct btrfs_dev_extent
*extent
= NULL
;
1526 path
= btrfs_alloc_path();
1530 key
.objectid
= device
->devid
;
1532 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1534 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1536 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1537 BTRFS_DEV_EXTENT_KEY
);
1540 leaf
= path
->nodes
[0];
1541 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1542 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1543 struct btrfs_dev_extent
);
1544 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1545 btrfs_dev_extent_length(leaf
, extent
) < start
);
1547 btrfs_release_path(path
);
1549 } else if (ret
== 0) {
1550 leaf
= path
->nodes
[0];
1551 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1552 struct btrfs_dev_extent
);
1554 btrfs_handle_fs_error(fs_info
, ret
, "Slot search failed");
1558 *dev_extent_len
= btrfs_dev_extent_length(leaf
, extent
);
1560 ret
= btrfs_del_item(trans
, root
, path
);
1562 btrfs_handle_fs_error(fs_info
, ret
,
1563 "Failed to remove dev extent item");
1565 set_bit(BTRFS_TRANS_HAVE_FREE_BGS
, &trans
->transaction
->flags
);
1568 btrfs_free_path(path
);
1572 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1573 struct btrfs_device
*device
,
1574 u64 chunk_offset
, u64 start
, u64 num_bytes
)
1577 struct btrfs_path
*path
;
1578 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1579 struct btrfs_root
*root
= fs_info
->dev_root
;
1580 struct btrfs_dev_extent
*extent
;
1581 struct extent_buffer
*leaf
;
1582 struct btrfs_key key
;
1584 WARN_ON(!device
->in_fs_metadata
);
1585 WARN_ON(device
->is_tgtdev_for_dev_replace
);
1586 path
= btrfs_alloc_path();
1590 key
.objectid
= device
->devid
;
1592 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1593 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1598 leaf
= path
->nodes
[0];
1599 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1600 struct btrfs_dev_extent
);
1601 btrfs_set_dev_extent_chunk_tree(leaf
, extent
,
1602 BTRFS_CHUNK_TREE_OBJECTID
);
1603 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
,
1604 BTRFS_FIRST_CHUNK_TREE_OBJECTID
);
1605 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1607 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1608 btrfs_mark_buffer_dirty(leaf
);
1610 btrfs_free_path(path
);
1614 static u64
find_next_chunk(struct btrfs_fs_info
*fs_info
)
1616 struct extent_map_tree
*em_tree
;
1617 struct extent_map
*em
;
1621 em_tree
= &fs_info
->mapping_tree
.map_tree
;
1622 read_lock(&em_tree
->lock
);
1623 n
= rb_last(&em_tree
->map
);
1625 em
= rb_entry(n
, struct extent_map
, rb_node
);
1626 ret
= em
->start
+ em
->len
;
1628 read_unlock(&em_tree
->lock
);
1633 static noinline
int find_next_devid(struct btrfs_fs_info
*fs_info
,
1637 struct btrfs_key key
;
1638 struct btrfs_key found_key
;
1639 struct btrfs_path
*path
;
1641 path
= btrfs_alloc_path();
1645 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1646 key
.type
= BTRFS_DEV_ITEM_KEY
;
1647 key
.offset
= (u64
)-1;
1649 ret
= btrfs_search_slot(NULL
, fs_info
->chunk_root
, &key
, path
, 0, 0);
1653 BUG_ON(ret
== 0); /* Corruption */
1655 ret
= btrfs_previous_item(fs_info
->chunk_root
, path
,
1656 BTRFS_DEV_ITEMS_OBJECTID
,
1657 BTRFS_DEV_ITEM_KEY
);
1661 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1663 *devid_ret
= found_key
.offset
+ 1;
1667 btrfs_free_path(path
);
1672 * the device information is stored in the chunk root
1673 * the btrfs_device struct should be fully filled in
1675 static int btrfs_add_device(struct btrfs_trans_handle
*trans
,
1676 struct btrfs_fs_info
*fs_info
,
1677 struct btrfs_device
*device
)
1679 struct btrfs_root
*root
= fs_info
->chunk_root
;
1681 struct btrfs_path
*path
;
1682 struct btrfs_dev_item
*dev_item
;
1683 struct extent_buffer
*leaf
;
1684 struct btrfs_key key
;
1687 path
= btrfs_alloc_path();
1691 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1692 key
.type
= BTRFS_DEV_ITEM_KEY
;
1693 key
.offset
= device
->devid
;
1695 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1700 leaf
= path
->nodes
[0];
1701 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1703 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1704 btrfs_set_device_generation(leaf
, dev_item
, 0);
1705 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1706 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1707 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1708 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1709 btrfs_set_device_total_bytes(leaf
, dev_item
,
1710 btrfs_device_get_disk_total_bytes(device
));
1711 btrfs_set_device_bytes_used(leaf
, dev_item
,
1712 btrfs_device_get_bytes_used(device
));
1713 btrfs_set_device_group(leaf
, dev_item
, 0);
1714 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1715 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1716 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1718 ptr
= btrfs_device_uuid(dev_item
);
1719 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1720 ptr
= btrfs_device_fsid(dev_item
);
1721 write_extent_buffer(leaf
, fs_info
->fsid
, ptr
, BTRFS_FSID_SIZE
);
1722 btrfs_mark_buffer_dirty(leaf
);
1726 btrfs_free_path(path
);
1731 * Function to update ctime/mtime for a given device path.
1732 * Mainly used for ctime/mtime based probe like libblkid.
1734 static void update_dev_time(const char *path_name
)
1738 filp
= filp_open(path_name
, O_RDWR
, 0);
1741 file_update_time(filp
);
1742 filp_close(filp
, NULL
);
1745 static int btrfs_rm_dev_item(struct btrfs_fs_info
*fs_info
,
1746 struct btrfs_device
*device
)
1748 struct btrfs_root
*root
= fs_info
->chunk_root
;
1750 struct btrfs_path
*path
;
1751 struct btrfs_key key
;
1752 struct btrfs_trans_handle
*trans
;
1754 path
= btrfs_alloc_path();
1758 trans
= btrfs_start_transaction(root
, 0);
1759 if (IS_ERR(trans
)) {
1760 btrfs_free_path(path
);
1761 return PTR_ERR(trans
);
1763 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1764 key
.type
= BTRFS_DEV_ITEM_KEY
;
1765 key
.offset
= device
->devid
;
1767 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1776 ret
= btrfs_del_item(trans
, root
, path
);
1780 btrfs_free_path(path
);
1781 btrfs_commit_transaction(trans
);
1786 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1787 * filesystem. It's up to the caller to adjust that number regarding eg. device
1790 static int btrfs_check_raid_min_devices(struct btrfs_fs_info
*fs_info
,
1798 seq
= read_seqbegin(&fs_info
->profiles_lock
);
1800 all_avail
= fs_info
->avail_data_alloc_bits
|
1801 fs_info
->avail_system_alloc_bits
|
1802 fs_info
->avail_metadata_alloc_bits
;
1803 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
1805 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
1806 if (!(all_avail
& btrfs_raid_group
[i
]))
1809 if (num_devices
< btrfs_raid_array
[i
].devs_min
) {
1810 int ret
= btrfs_raid_mindev_error
[i
];
1820 struct btrfs_device
*btrfs_find_next_active_device(struct btrfs_fs_devices
*fs_devs
,
1821 struct btrfs_device
*device
)
1823 struct btrfs_device
*next_device
;
1825 list_for_each_entry(next_device
, &fs_devs
->devices
, dev_list
) {
1826 if (next_device
!= device
&&
1827 !next_device
->missing
&& next_device
->bdev
)
1835 * Helper function to check if the given device is part of s_bdev / latest_bdev
1836 * and replace it with the provided or the next active device, in the context
1837 * where this function called, there should be always be another device (or
1838 * this_dev) which is active.
1840 void btrfs_assign_next_active_device(struct btrfs_fs_info
*fs_info
,
1841 struct btrfs_device
*device
, struct btrfs_device
*this_dev
)
1843 struct btrfs_device
*next_device
;
1846 next_device
= this_dev
;
1848 next_device
= btrfs_find_next_active_device(fs_info
->fs_devices
,
1850 ASSERT(next_device
);
1852 if (fs_info
->sb
->s_bdev
&&
1853 (fs_info
->sb
->s_bdev
== device
->bdev
))
1854 fs_info
->sb
->s_bdev
= next_device
->bdev
;
1856 if (fs_info
->fs_devices
->latest_bdev
== device
->bdev
)
1857 fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1860 int btrfs_rm_device(struct btrfs_fs_info
*fs_info
, const char *device_path
,
1863 struct btrfs_device
*device
;
1864 struct btrfs_fs_devices
*cur_devices
;
1868 mutex_lock(&uuid_mutex
);
1870 num_devices
= fs_info
->fs_devices
->num_devices
;
1871 btrfs_dev_replace_lock(&fs_info
->dev_replace
, 0);
1872 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
1873 WARN_ON(num_devices
< 1);
1876 btrfs_dev_replace_unlock(&fs_info
->dev_replace
, 0);
1878 ret
= btrfs_check_raid_min_devices(fs_info
, num_devices
- 1);
1882 ret
= btrfs_find_device_by_devspec(fs_info
, devid
, device_path
,
1887 if (device
->is_tgtdev_for_dev_replace
) {
1888 ret
= BTRFS_ERROR_DEV_TGT_REPLACE
;
1892 if (device
->writeable
&& fs_info
->fs_devices
->rw_devices
== 1) {
1893 ret
= BTRFS_ERROR_DEV_ONLY_WRITABLE
;
1897 if (device
->writeable
) {
1898 mutex_lock(&fs_info
->chunk_mutex
);
1899 list_del_init(&device
->dev_alloc_list
);
1900 device
->fs_devices
->rw_devices
--;
1901 mutex_unlock(&fs_info
->chunk_mutex
);
1904 mutex_unlock(&uuid_mutex
);
1905 ret
= btrfs_shrink_device(device
, 0);
1906 mutex_lock(&uuid_mutex
);
1911 * TODO: the superblock still includes this device in its num_devices
1912 * counter although write_all_supers() is not locked out. This
1913 * could give a filesystem state which requires a degraded mount.
1915 ret
= btrfs_rm_dev_item(fs_info
, device
);
1919 device
->in_fs_metadata
= 0;
1920 btrfs_scrub_cancel_dev(fs_info
, device
);
1923 * the device list mutex makes sure that we don't change
1924 * the device list while someone else is writing out all
1925 * the device supers. Whoever is writing all supers, should
1926 * lock the device list mutex before getting the number of
1927 * devices in the super block (super_copy). Conversely,
1928 * whoever updates the number of devices in the super block
1929 * (super_copy) should hold the device list mutex.
1932 cur_devices
= device
->fs_devices
;
1933 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
1934 list_del_rcu(&device
->dev_list
);
1936 device
->fs_devices
->num_devices
--;
1937 device
->fs_devices
->total_devices
--;
1939 if (device
->missing
)
1940 device
->fs_devices
->missing_devices
--;
1942 btrfs_assign_next_active_device(fs_info
, device
, NULL
);
1945 device
->fs_devices
->open_devices
--;
1946 /* remove sysfs entry */
1947 btrfs_sysfs_rm_device_link(fs_info
->fs_devices
, device
);
1950 num_devices
= btrfs_super_num_devices(fs_info
->super_copy
) - 1;
1951 btrfs_set_super_num_devices(fs_info
->super_copy
, num_devices
);
1952 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
1955 * at this point, the device is zero sized and detached from
1956 * the devices list. All that's left is to zero out the old
1957 * supers and free the device.
1959 if (device
->writeable
)
1960 btrfs_scratch_superblocks(device
->bdev
, device
->name
->str
);
1962 btrfs_close_bdev(device
);
1963 call_rcu(&device
->rcu
, free_device
);
1965 if (cur_devices
->open_devices
== 0) {
1966 struct btrfs_fs_devices
*fs_devices
;
1967 fs_devices
= fs_info
->fs_devices
;
1968 while (fs_devices
) {
1969 if (fs_devices
->seed
== cur_devices
) {
1970 fs_devices
->seed
= cur_devices
->seed
;
1973 fs_devices
= fs_devices
->seed
;
1975 cur_devices
->seed
= NULL
;
1976 __btrfs_close_devices(cur_devices
);
1977 free_fs_devices(cur_devices
);
1981 mutex_unlock(&uuid_mutex
);
1985 if (device
->writeable
) {
1986 mutex_lock(&fs_info
->chunk_mutex
);
1987 list_add(&device
->dev_alloc_list
,
1988 &fs_info
->fs_devices
->alloc_list
);
1989 device
->fs_devices
->rw_devices
++;
1990 mutex_unlock(&fs_info
->chunk_mutex
);
1995 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info
*fs_info
,
1996 struct btrfs_device
*srcdev
)
1998 struct btrfs_fs_devices
*fs_devices
;
2000 WARN_ON(!mutex_is_locked(&fs_info
->fs_devices
->device_list_mutex
));
2003 * in case of fs with no seed, srcdev->fs_devices will point
2004 * to fs_devices of fs_info. However when the dev being replaced is
2005 * a seed dev it will point to the seed's local fs_devices. In short
2006 * srcdev will have its correct fs_devices in both the cases.
2008 fs_devices
= srcdev
->fs_devices
;
2010 list_del_rcu(&srcdev
->dev_list
);
2011 list_del_rcu(&srcdev
->dev_alloc_list
);
2012 fs_devices
->num_devices
--;
2013 if (srcdev
->missing
)
2014 fs_devices
->missing_devices
--;
2016 if (srcdev
->writeable
)
2017 fs_devices
->rw_devices
--;
2020 fs_devices
->open_devices
--;
2023 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info
*fs_info
,
2024 struct btrfs_device
*srcdev
)
2026 struct btrfs_fs_devices
*fs_devices
= srcdev
->fs_devices
;
2028 if (srcdev
->writeable
) {
2029 /* zero out the old super if it is writable */
2030 btrfs_scratch_superblocks(srcdev
->bdev
, srcdev
->name
->str
);
2033 btrfs_close_bdev(srcdev
);
2035 call_rcu(&srcdev
->rcu
, free_device
);
2038 * unless fs_devices is seed fs, num_devices shouldn't go
2041 BUG_ON(!fs_devices
->num_devices
&& !fs_devices
->seeding
);
2043 /* if this is no devs we rather delete the fs_devices */
2044 if (!fs_devices
->num_devices
) {
2045 struct btrfs_fs_devices
*tmp_fs_devices
;
2047 tmp_fs_devices
= fs_info
->fs_devices
;
2048 while (tmp_fs_devices
) {
2049 if (tmp_fs_devices
->seed
== fs_devices
) {
2050 tmp_fs_devices
->seed
= fs_devices
->seed
;
2053 tmp_fs_devices
= tmp_fs_devices
->seed
;
2055 fs_devices
->seed
= NULL
;
2056 __btrfs_close_devices(fs_devices
);
2057 free_fs_devices(fs_devices
);
2061 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
2062 struct btrfs_device
*tgtdev
)
2064 mutex_lock(&uuid_mutex
);
2066 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2068 btrfs_sysfs_rm_device_link(fs_info
->fs_devices
, tgtdev
);
2071 fs_info
->fs_devices
->open_devices
--;
2073 fs_info
->fs_devices
->num_devices
--;
2075 btrfs_assign_next_active_device(fs_info
, tgtdev
, NULL
);
2077 list_del_rcu(&tgtdev
->dev_list
);
2079 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2080 mutex_unlock(&uuid_mutex
);
2083 * The update_dev_time() with in btrfs_scratch_superblocks()
2084 * may lead to a call to btrfs_show_devname() which will try
2085 * to hold device_list_mutex. And here this device
2086 * is already out of device list, so we don't have to hold
2087 * the device_list_mutex lock.
2089 btrfs_scratch_superblocks(tgtdev
->bdev
, tgtdev
->name
->str
);
2091 btrfs_close_bdev(tgtdev
);
2092 call_rcu(&tgtdev
->rcu
, free_device
);
2095 static int btrfs_find_device_by_path(struct btrfs_fs_info
*fs_info
,
2096 const char *device_path
,
2097 struct btrfs_device
**device
)
2100 struct btrfs_super_block
*disk_super
;
2103 struct block_device
*bdev
;
2104 struct buffer_head
*bh
;
2107 ret
= btrfs_get_bdev_and_sb(device_path
, FMODE_READ
,
2108 fs_info
->bdev_holder
, 0, &bdev
, &bh
);
2111 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
2112 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
2113 dev_uuid
= disk_super
->dev_item
.uuid
;
2114 *device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, disk_super
->fsid
);
2118 blkdev_put(bdev
, FMODE_READ
);
2122 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info
*fs_info
,
2123 const char *device_path
,
2124 struct btrfs_device
**device
)
2127 if (strcmp(device_path
, "missing") == 0) {
2128 struct list_head
*devices
;
2129 struct btrfs_device
*tmp
;
2131 devices
= &fs_info
->fs_devices
->devices
;
2133 * It is safe to read the devices since the volume_mutex
2134 * is held by the caller.
2136 list_for_each_entry(tmp
, devices
, dev_list
) {
2137 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
2144 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND
;
2148 return btrfs_find_device_by_path(fs_info
, device_path
, device
);
2153 * Lookup a device given by device id, or the path if the id is 0.
2155 int btrfs_find_device_by_devspec(struct btrfs_fs_info
*fs_info
, u64 devid
,
2156 const char *devpath
,
2157 struct btrfs_device
**device
)
2163 *device
= btrfs_find_device(fs_info
, devid
, NULL
, NULL
);
2167 if (!devpath
|| !devpath
[0])
2170 ret
= btrfs_find_device_missing_or_by_path(fs_info
, devpath
,
2177 * does all the dirty work required for changing file system's UUID.
2179 static int btrfs_prepare_sprout(struct btrfs_fs_info
*fs_info
)
2181 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2182 struct btrfs_fs_devices
*old_devices
;
2183 struct btrfs_fs_devices
*seed_devices
;
2184 struct btrfs_super_block
*disk_super
= fs_info
->super_copy
;
2185 struct btrfs_device
*device
;
2188 BUG_ON(!mutex_is_locked(&uuid_mutex
));
2189 if (!fs_devices
->seeding
)
2192 seed_devices
= alloc_fs_devices(NULL
);
2193 if (IS_ERR(seed_devices
))
2194 return PTR_ERR(seed_devices
);
2196 old_devices
= clone_fs_devices(fs_devices
);
2197 if (IS_ERR(old_devices
)) {
2198 kfree(seed_devices
);
2199 return PTR_ERR(old_devices
);
2202 list_add(&old_devices
->list
, &fs_uuids
);
2204 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
2205 seed_devices
->opened
= 1;
2206 INIT_LIST_HEAD(&seed_devices
->devices
);
2207 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
2208 mutex_init(&seed_devices
->device_list_mutex
);
2210 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2211 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
2213 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
)
2214 device
->fs_devices
= seed_devices
;
2216 mutex_lock(&fs_info
->chunk_mutex
);
2217 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
2218 mutex_unlock(&fs_info
->chunk_mutex
);
2220 fs_devices
->seeding
= 0;
2221 fs_devices
->num_devices
= 0;
2222 fs_devices
->open_devices
= 0;
2223 fs_devices
->missing_devices
= 0;
2224 fs_devices
->rotating
= 0;
2225 fs_devices
->seed
= seed_devices
;
2227 generate_random_uuid(fs_devices
->fsid
);
2228 memcpy(fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2229 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2230 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2232 super_flags
= btrfs_super_flags(disk_super
) &
2233 ~BTRFS_SUPER_FLAG_SEEDING
;
2234 btrfs_set_super_flags(disk_super
, super_flags
);
2240 * Store the expected generation for seed devices in device items.
2242 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
2243 struct btrfs_fs_info
*fs_info
)
2245 struct btrfs_root
*root
= fs_info
->chunk_root
;
2246 struct btrfs_path
*path
;
2247 struct extent_buffer
*leaf
;
2248 struct btrfs_dev_item
*dev_item
;
2249 struct btrfs_device
*device
;
2250 struct btrfs_key key
;
2251 u8 fs_uuid
[BTRFS_FSID_SIZE
];
2252 u8 dev_uuid
[BTRFS_UUID_SIZE
];
2256 path
= btrfs_alloc_path();
2260 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2262 key
.type
= BTRFS_DEV_ITEM_KEY
;
2265 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2269 leaf
= path
->nodes
[0];
2271 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
2272 ret
= btrfs_next_leaf(root
, path
);
2277 leaf
= path
->nodes
[0];
2278 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2279 btrfs_release_path(path
);
2283 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2284 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
2285 key
.type
!= BTRFS_DEV_ITEM_KEY
)
2288 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2289 struct btrfs_dev_item
);
2290 devid
= btrfs_device_id(leaf
, dev_item
);
2291 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
2293 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
2295 device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, fs_uuid
);
2296 BUG_ON(!device
); /* Logic error */
2298 if (device
->fs_devices
->seeding
) {
2299 btrfs_set_device_generation(leaf
, dev_item
,
2300 device
->generation
);
2301 btrfs_mark_buffer_dirty(leaf
);
2309 btrfs_free_path(path
);
2313 int btrfs_init_new_device(struct btrfs_fs_info
*fs_info
, const char *device_path
)
2315 struct btrfs_root
*root
= fs_info
->dev_root
;
2316 struct request_queue
*q
;
2317 struct btrfs_trans_handle
*trans
;
2318 struct btrfs_device
*device
;
2319 struct block_device
*bdev
;
2320 struct list_head
*devices
;
2321 struct super_block
*sb
= fs_info
->sb
;
2322 struct rcu_string
*name
;
2324 int seeding_dev
= 0;
2327 if (sb_rdonly(sb
) && !fs_info
->fs_devices
->seeding
)
2330 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2331 fs_info
->bdev_holder
);
2333 return PTR_ERR(bdev
);
2335 if (fs_info
->fs_devices
->seeding
) {
2337 down_write(&sb
->s_umount
);
2338 mutex_lock(&uuid_mutex
);
2341 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2343 devices
= &fs_info
->fs_devices
->devices
;
2345 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2346 list_for_each_entry(device
, devices
, dev_list
) {
2347 if (device
->bdev
== bdev
) {
2350 &fs_info
->fs_devices
->device_list_mutex
);
2354 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2356 device
= btrfs_alloc_device(fs_info
, NULL
, NULL
);
2357 if (IS_ERR(device
)) {
2358 /* we can safely leave the fs_devices entry around */
2359 ret
= PTR_ERR(device
);
2363 name
= rcu_string_strdup(device_path
, GFP_KERNEL
);
2369 rcu_assign_pointer(device
->name
, name
);
2371 trans
= btrfs_start_transaction(root
, 0);
2372 if (IS_ERR(trans
)) {
2373 rcu_string_free(device
->name
);
2375 ret
= PTR_ERR(trans
);
2379 q
= bdev_get_queue(bdev
);
2380 if (blk_queue_discard(q
))
2381 device
->can_discard
= 1;
2382 device
->writeable
= 1;
2383 device
->generation
= trans
->transid
;
2384 device
->io_width
= fs_info
->sectorsize
;
2385 device
->io_align
= fs_info
->sectorsize
;
2386 device
->sector_size
= fs_info
->sectorsize
;
2387 device
->total_bytes
= round_down(i_size_read(bdev
->bd_inode
),
2388 fs_info
->sectorsize
);
2389 device
->disk_total_bytes
= device
->total_bytes
;
2390 device
->commit_total_bytes
= device
->total_bytes
;
2391 device
->fs_info
= fs_info
;
2392 device
->bdev
= bdev
;
2393 device
->in_fs_metadata
= 1;
2394 device
->is_tgtdev_for_dev_replace
= 0;
2395 device
->mode
= FMODE_EXCL
;
2396 device
->dev_stats_valid
= 1;
2397 set_blocksize(device
->bdev
, BTRFS_BDEV_BLOCKSIZE
);
2400 sb
->s_flags
&= ~MS_RDONLY
;
2401 ret
= btrfs_prepare_sprout(fs_info
);
2402 BUG_ON(ret
); /* -ENOMEM */
2405 device
->fs_devices
= fs_info
->fs_devices
;
2407 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2408 mutex_lock(&fs_info
->chunk_mutex
);
2409 list_add_rcu(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2410 list_add(&device
->dev_alloc_list
,
2411 &fs_info
->fs_devices
->alloc_list
);
2412 fs_info
->fs_devices
->num_devices
++;
2413 fs_info
->fs_devices
->open_devices
++;
2414 fs_info
->fs_devices
->rw_devices
++;
2415 fs_info
->fs_devices
->total_devices
++;
2416 fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
2418 atomic64_add(device
->total_bytes
, &fs_info
->free_chunk_space
);
2420 if (!blk_queue_nonrot(q
))
2421 fs_info
->fs_devices
->rotating
= 1;
2423 tmp
= btrfs_super_total_bytes(fs_info
->super_copy
);
2424 btrfs_set_super_total_bytes(fs_info
->super_copy
,
2425 round_down(tmp
+ device
->total_bytes
, fs_info
->sectorsize
));
2427 tmp
= btrfs_super_num_devices(fs_info
->super_copy
);
2428 btrfs_set_super_num_devices(fs_info
->super_copy
, tmp
+ 1);
2430 /* add sysfs device entry */
2431 btrfs_sysfs_add_device_link(fs_info
->fs_devices
, device
);
2434 * we've got more storage, clear any full flags on the space
2437 btrfs_clear_space_info_full(fs_info
);
2439 mutex_unlock(&fs_info
->chunk_mutex
);
2440 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2443 mutex_lock(&fs_info
->chunk_mutex
);
2444 ret
= init_first_rw_device(trans
, fs_info
);
2445 mutex_unlock(&fs_info
->chunk_mutex
);
2447 btrfs_abort_transaction(trans
, ret
);
2452 ret
= btrfs_add_device(trans
, fs_info
, device
);
2454 btrfs_abort_transaction(trans
, ret
);
2459 char fsid_buf
[BTRFS_UUID_UNPARSED_SIZE
];
2461 ret
= btrfs_finish_sprout(trans
, fs_info
);
2463 btrfs_abort_transaction(trans
, ret
);
2467 /* Sprouting would change fsid of the mounted root,
2468 * so rename the fsid on the sysfs
2470 snprintf(fsid_buf
, BTRFS_UUID_UNPARSED_SIZE
, "%pU",
2472 if (kobject_rename(&fs_info
->fs_devices
->fsid_kobj
, fsid_buf
))
2474 "sysfs: failed to create fsid for sprout");
2477 ret
= btrfs_commit_transaction(trans
);
2480 mutex_unlock(&uuid_mutex
);
2481 up_write(&sb
->s_umount
);
2483 if (ret
) /* transaction commit */
2486 ret
= btrfs_relocate_sys_chunks(fs_info
);
2488 btrfs_handle_fs_error(fs_info
, ret
,
2489 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2490 trans
= btrfs_attach_transaction(root
);
2491 if (IS_ERR(trans
)) {
2492 if (PTR_ERR(trans
) == -ENOENT
)
2494 return PTR_ERR(trans
);
2496 ret
= btrfs_commit_transaction(trans
);
2499 /* Update ctime/mtime for libblkid */
2500 update_dev_time(device_path
);
2504 btrfs_end_transaction(trans
);
2505 rcu_string_free(device
->name
);
2506 btrfs_sysfs_rm_device_link(fs_info
->fs_devices
, device
);
2509 blkdev_put(bdev
, FMODE_EXCL
);
2511 mutex_unlock(&uuid_mutex
);
2512 up_write(&sb
->s_umount
);
2517 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
2518 const char *device_path
,
2519 struct btrfs_device
*srcdev
,
2520 struct btrfs_device
**device_out
)
2522 struct request_queue
*q
;
2523 struct btrfs_device
*device
;
2524 struct block_device
*bdev
;
2525 struct list_head
*devices
;
2526 struct rcu_string
*name
;
2527 u64 devid
= BTRFS_DEV_REPLACE_DEVID
;
2531 if (fs_info
->fs_devices
->seeding
) {
2532 btrfs_err(fs_info
, "the filesystem is a seed filesystem!");
2536 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2537 fs_info
->bdev_holder
);
2539 btrfs_err(fs_info
, "target device %s is invalid!", device_path
);
2540 return PTR_ERR(bdev
);
2543 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2545 devices
= &fs_info
->fs_devices
->devices
;
2546 list_for_each_entry(device
, devices
, dev_list
) {
2547 if (device
->bdev
== bdev
) {
2549 "target device is in the filesystem!");
2556 if (i_size_read(bdev
->bd_inode
) <
2557 btrfs_device_get_total_bytes(srcdev
)) {
2559 "target device is smaller than source device!");
2565 device
= btrfs_alloc_device(NULL
, &devid
, NULL
);
2566 if (IS_ERR(device
)) {
2567 ret
= PTR_ERR(device
);
2571 name
= rcu_string_strdup(device_path
, GFP_KERNEL
);
2577 rcu_assign_pointer(device
->name
, name
);
2579 q
= bdev_get_queue(bdev
);
2580 if (blk_queue_discard(q
))
2581 device
->can_discard
= 1;
2582 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2583 device
->writeable
= 1;
2584 device
->generation
= 0;
2585 device
->io_width
= fs_info
->sectorsize
;
2586 device
->io_align
= fs_info
->sectorsize
;
2587 device
->sector_size
= fs_info
->sectorsize
;
2588 device
->total_bytes
= btrfs_device_get_total_bytes(srcdev
);
2589 device
->disk_total_bytes
= btrfs_device_get_disk_total_bytes(srcdev
);
2590 device
->bytes_used
= btrfs_device_get_bytes_used(srcdev
);
2591 ASSERT(list_empty(&srcdev
->resized_list
));
2592 device
->commit_total_bytes
= srcdev
->commit_total_bytes
;
2593 device
->commit_bytes_used
= device
->bytes_used
;
2594 device
->fs_info
= fs_info
;
2595 device
->bdev
= bdev
;
2596 device
->in_fs_metadata
= 1;
2597 device
->is_tgtdev_for_dev_replace
= 1;
2598 device
->mode
= FMODE_EXCL
;
2599 device
->dev_stats_valid
= 1;
2600 set_blocksize(device
->bdev
, BTRFS_BDEV_BLOCKSIZE
);
2601 device
->fs_devices
= fs_info
->fs_devices
;
2602 list_add(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2603 fs_info
->fs_devices
->num_devices
++;
2604 fs_info
->fs_devices
->open_devices
++;
2605 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2607 *device_out
= device
;
2611 blkdev_put(bdev
, FMODE_EXCL
);
2615 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info
*fs_info
,
2616 struct btrfs_device
*tgtdev
)
2618 u32 sectorsize
= fs_info
->sectorsize
;
2620 WARN_ON(fs_info
->fs_devices
->rw_devices
== 0);
2621 tgtdev
->io_width
= sectorsize
;
2622 tgtdev
->io_align
= sectorsize
;
2623 tgtdev
->sector_size
= sectorsize
;
2624 tgtdev
->fs_info
= fs_info
;
2625 tgtdev
->in_fs_metadata
= 1;
2628 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2629 struct btrfs_device
*device
)
2632 struct btrfs_path
*path
;
2633 struct btrfs_root
*root
= device
->fs_info
->chunk_root
;
2634 struct btrfs_dev_item
*dev_item
;
2635 struct extent_buffer
*leaf
;
2636 struct btrfs_key key
;
2638 path
= btrfs_alloc_path();
2642 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2643 key
.type
= BTRFS_DEV_ITEM_KEY
;
2644 key
.offset
= device
->devid
;
2646 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2655 leaf
= path
->nodes
[0];
2656 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2658 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2659 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2660 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2661 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2662 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2663 btrfs_set_device_total_bytes(leaf
, dev_item
,
2664 btrfs_device_get_disk_total_bytes(device
));
2665 btrfs_set_device_bytes_used(leaf
, dev_item
,
2666 btrfs_device_get_bytes_used(device
));
2667 btrfs_mark_buffer_dirty(leaf
);
2670 btrfs_free_path(path
);
2674 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2675 struct btrfs_device
*device
, u64 new_size
)
2677 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
2678 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
2679 struct btrfs_fs_devices
*fs_devices
;
2683 if (!device
->writeable
)
2686 new_size
= round_down(new_size
, fs_info
->sectorsize
);
2688 mutex_lock(&fs_info
->chunk_mutex
);
2689 old_total
= btrfs_super_total_bytes(super_copy
);
2690 diff
= round_down(new_size
- device
->total_bytes
, fs_info
->sectorsize
);
2692 if (new_size
<= device
->total_bytes
||
2693 device
->is_tgtdev_for_dev_replace
) {
2694 mutex_unlock(&fs_info
->chunk_mutex
);
2698 fs_devices
= fs_info
->fs_devices
;
2700 btrfs_set_super_total_bytes(super_copy
,
2701 round_down(old_total
+ diff
, fs_info
->sectorsize
));
2702 device
->fs_devices
->total_rw_bytes
+= diff
;
2704 btrfs_device_set_total_bytes(device
, new_size
);
2705 btrfs_device_set_disk_total_bytes(device
, new_size
);
2706 btrfs_clear_space_info_full(device
->fs_info
);
2707 if (list_empty(&device
->resized_list
))
2708 list_add_tail(&device
->resized_list
,
2709 &fs_devices
->resized_devices
);
2710 mutex_unlock(&fs_info
->chunk_mutex
);
2712 return btrfs_update_device(trans
, device
);
2715 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
2716 struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2718 struct btrfs_root
*root
= fs_info
->chunk_root
;
2720 struct btrfs_path
*path
;
2721 struct btrfs_key key
;
2723 path
= btrfs_alloc_path();
2727 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2728 key
.offset
= chunk_offset
;
2729 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2731 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2734 else if (ret
> 0) { /* Logic error or corruption */
2735 btrfs_handle_fs_error(fs_info
, -ENOENT
,
2736 "Failed lookup while freeing chunk.");
2741 ret
= btrfs_del_item(trans
, root
, path
);
2743 btrfs_handle_fs_error(fs_info
, ret
,
2744 "Failed to delete chunk item.");
2746 btrfs_free_path(path
);
2750 static int btrfs_del_sys_chunk(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2752 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
2753 struct btrfs_disk_key
*disk_key
;
2754 struct btrfs_chunk
*chunk
;
2761 struct btrfs_key key
;
2763 mutex_lock(&fs_info
->chunk_mutex
);
2764 array_size
= btrfs_super_sys_array_size(super_copy
);
2766 ptr
= super_copy
->sys_chunk_array
;
2769 while (cur
< array_size
) {
2770 disk_key
= (struct btrfs_disk_key
*)ptr
;
2771 btrfs_disk_key_to_cpu(&key
, disk_key
);
2773 len
= sizeof(*disk_key
);
2775 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2776 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2777 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2778 len
+= btrfs_chunk_item_size(num_stripes
);
2783 if (key
.objectid
== BTRFS_FIRST_CHUNK_TREE_OBJECTID
&&
2784 key
.offset
== chunk_offset
) {
2785 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2787 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2793 mutex_unlock(&fs_info
->chunk_mutex
);
2797 static struct extent_map
*get_chunk_map(struct btrfs_fs_info
*fs_info
,
2798 u64 logical
, u64 length
)
2800 struct extent_map_tree
*em_tree
;
2801 struct extent_map
*em
;
2803 em_tree
= &fs_info
->mapping_tree
.map_tree
;
2804 read_lock(&em_tree
->lock
);
2805 em
= lookup_extent_mapping(em_tree
, logical
, length
);
2806 read_unlock(&em_tree
->lock
);
2809 btrfs_crit(fs_info
, "unable to find logical %llu length %llu",
2811 return ERR_PTR(-EINVAL
);
2814 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
2816 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2817 logical
, length
, em
->start
, em
->start
+ em
->len
);
2818 free_extent_map(em
);
2819 return ERR_PTR(-EINVAL
);
2822 /* callers are responsible for dropping em's ref. */
2826 int btrfs_remove_chunk(struct btrfs_trans_handle
*trans
,
2827 struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2829 struct extent_map
*em
;
2830 struct map_lookup
*map
;
2831 u64 dev_extent_len
= 0;
2833 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2835 em
= get_chunk_map(fs_info
, chunk_offset
, 1);
2838 * This is a logic error, but we don't want to just rely on the
2839 * user having built with ASSERT enabled, so if ASSERT doesn't
2840 * do anything we still error out.
2845 map
= em
->map_lookup
;
2846 mutex_lock(&fs_info
->chunk_mutex
);
2847 check_system_chunk(trans
, fs_info
, map
->type
);
2848 mutex_unlock(&fs_info
->chunk_mutex
);
2851 * Take the device list mutex to prevent races with the final phase of
2852 * a device replace operation that replaces the device object associated
2853 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2855 mutex_lock(&fs_devices
->device_list_mutex
);
2856 for (i
= 0; i
< map
->num_stripes
; i
++) {
2857 struct btrfs_device
*device
= map
->stripes
[i
].dev
;
2858 ret
= btrfs_free_dev_extent(trans
, device
,
2859 map
->stripes
[i
].physical
,
2862 mutex_unlock(&fs_devices
->device_list_mutex
);
2863 btrfs_abort_transaction(trans
, ret
);
2867 if (device
->bytes_used
> 0) {
2868 mutex_lock(&fs_info
->chunk_mutex
);
2869 btrfs_device_set_bytes_used(device
,
2870 device
->bytes_used
- dev_extent_len
);
2871 atomic64_add(dev_extent_len
, &fs_info
->free_chunk_space
);
2872 btrfs_clear_space_info_full(fs_info
);
2873 mutex_unlock(&fs_info
->chunk_mutex
);
2876 if (map
->stripes
[i
].dev
) {
2877 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2879 mutex_unlock(&fs_devices
->device_list_mutex
);
2880 btrfs_abort_transaction(trans
, ret
);
2885 mutex_unlock(&fs_devices
->device_list_mutex
);
2887 ret
= btrfs_free_chunk(trans
, fs_info
, chunk_offset
);
2889 btrfs_abort_transaction(trans
, ret
);
2893 trace_btrfs_chunk_free(fs_info
, map
, chunk_offset
, em
->len
);
2895 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2896 ret
= btrfs_del_sys_chunk(fs_info
, chunk_offset
);
2898 btrfs_abort_transaction(trans
, ret
);
2903 ret
= btrfs_remove_block_group(trans
, fs_info
, chunk_offset
, em
);
2905 btrfs_abort_transaction(trans
, ret
);
2911 free_extent_map(em
);
2915 static int btrfs_relocate_chunk(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2917 struct btrfs_root
*root
= fs_info
->chunk_root
;
2918 struct btrfs_trans_handle
*trans
;
2922 * Prevent races with automatic removal of unused block groups.
2923 * After we relocate and before we remove the chunk with offset
2924 * chunk_offset, automatic removal of the block group can kick in,
2925 * resulting in a failure when calling btrfs_remove_chunk() below.
2927 * Make sure to acquire this mutex before doing a tree search (dev
2928 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2929 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2930 * we release the path used to search the chunk/dev tree and before
2931 * the current task acquires this mutex and calls us.
2933 ASSERT(mutex_is_locked(&fs_info
->delete_unused_bgs_mutex
));
2935 ret
= btrfs_can_relocate(fs_info
, chunk_offset
);
2939 /* step one, relocate all the extents inside this chunk */
2940 btrfs_scrub_pause(fs_info
);
2941 ret
= btrfs_relocate_block_group(fs_info
, chunk_offset
);
2942 btrfs_scrub_continue(fs_info
);
2946 trans
= btrfs_start_trans_remove_block_group(root
->fs_info
,
2948 if (IS_ERR(trans
)) {
2949 ret
= PTR_ERR(trans
);
2950 btrfs_handle_fs_error(root
->fs_info
, ret
, NULL
);
2955 * step two, delete the device extents and the
2956 * chunk tree entries
2958 ret
= btrfs_remove_chunk(trans
, fs_info
, chunk_offset
);
2959 btrfs_end_transaction(trans
);
2963 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info
*fs_info
)
2965 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
2966 struct btrfs_path
*path
;
2967 struct extent_buffer
*leaf
;
2968 struct btrfs_chunk
*chunk
;
2969 struct btrfs_key key
;
2970 struct btrfs_key found_key
;
2972 bool retried
= false;
2976 path
= btrfs_alloc_path();
2981 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2982 key
.offset
= (u64
)-1;
2983 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2986 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
2987 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2989 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
2992 BUG_ON(ret
== 0); /* Corruption */
2994 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
2997 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3003 leaf
= path
->nodes
[0];
3004 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3006 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
3007 struct btrfs_chunk
);
3008 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3009 btrfs_release_path(path
);
3011 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3012 ret
= btrfs_relocate_chunk(fs_info
, found_key
.offset
);
3018 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3020 if (found_key
.offset
== 0)
3022 key
.offset
= found_key
.offset
- 1;
3025 if (failed
&& !retried
) {
3029 } else if (WARN_ON(failed
&& retried
)) {
3033 btrfs_free_path(path
);
3037 static int insert_balance_item(struct btrfs_fs_info
*fs_info
,
3038 struct btrfs_balance_control
*bctl
)
3040 struct btrfs_root
*root
= fs_info
->tree_root
;
3041 struct btrfs_trans_handle
*trans
;
3042 struct btrfs_balance_item
*item
;
3043 struct btrfs_disk_balance_args disk_bargs
;
3044 struct btrfs_path
*path
;
3045 struct extent_buffer
*leaf
;
3046 struct btrfs_key key
;
3049 path
= btrfs_alloc_path();
3053 trans
= btrfs_start_transaction(root
, 0);
3054 if (IS_ERR(trans
)) {
3055 btrfs_free_path(path
);
3056 return PTR_ERR(trans
);
3059 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3060 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3063 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
3068 leaf
= path
->nodes
[0];
3069 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3071 memzero_extent_buffer(leaf
, (unsigned long)item
, sizeof(*item
));
3073 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
3074 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
3075 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
3076 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
3077 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
3078 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
3080 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
3082 btrfs_mark_buffer_dirty(leaf
);
3084 btrfs_free_path(path
);
3085 err
= btrfs_commit_transaction(trans
);
3091 static int del_balance_item(struct btrfs_fs_info
*fs_info
)
3093 struct btrfs_root
*root
= fs_info
->tree_root
;
3094 struct btrfs_trans_handle
*trans
;
3095 struct btrfs_path
*path
;
3096 struct btrfs_key key
;
3099 path
= btrfs_alloc_path();
3103 trans
= btrfs_start_transaction(root
, 0);
3104 if (IS_ERR(trans
)) {
3105 btrfs_free_path(path
);
3106 return PTR_ERR(trans
);
3109 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3110 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3113 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3121 ret
= btrfs_del_item(trans
, root
, path
);
3123 btrfs_free_path(path
);
3124 err
= btrfs_commit_transaction(trans
);
3131 * This is a heuristic used to reduce the number of chunks balanced on
3132 * resume after balance was interrupted.
3134 static void update_balance_args(struct btrfs_balance_control
*bctl
)
3137 * Turn on soft mode for chunk types that were being converted.
3139 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3140 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3141 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3142 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3143 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3144 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3147 * Turn on usage filter if is not already used. The idea is
3148 * that chunks that we have already balanced should be
3149 * reasonably full. Don't do it for chunks that are being
3150 * converted - that will keep us from relocating unconverted
3151 * (albeit full) chunks.
3153 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3154 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3155 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3156 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3157 bctl
->data
.usage
= 90;
3159 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3160 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3161 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3162 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3163 bctl
->sys
.usage
= 90;
3165 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3166 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3167 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3168 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3169 bctl
->meta
.usage
= 90;
3174 * Should be called with both balance and volume mutexes held to
3175 * serialize other volume operations (add_dev/rm_dev/resize) with
3176 * restriper. Same goes for unset_balance_control.
3178 static void set_balance_control(struct btrfs_balance_control
*bctl
)
3180 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3182 BUG_ON(fs_info
->balance_ctl
);
3184 spin_lock(&fs_info
->balance_lock
);
3185 fs_info
->balance_ctl
= bctl
;
3186 spin_unlock(&fs_info
->balance_lock
);
3189 static void unset_balance_control(struct btrfs_fs_info
*fs_info
)
3191 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3193 BUG_ON(!fs_info
->balance_ctl
);
3195 spin_lock(&fs_info
->balance_lock
);
3196 fs_info
->balance_ctl
= NULL
;
3197 spin_unlock(&fs_info
->balance_lock
);
3203 * Balance filters. Return 1 if chunk should be filtered out
3204 * (should not be balanced).
3206 static int chunk_profiles_filter(u64 chunk_type
,
3207 struct btrfs_balance_args
*bargs
)
3209 chunk_type
= chunk_to_extended(chunk_type
) &
3210 BTRFS_EXTENDED_PROFILE_MASK
;
3212 if (bargs
->profiles
& chunk_type
)
3218 static int chunk_usage_range_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
3219 struct btrfs_balance_args
*bargs
)
3221 struct btrfs_block_group_cache
*cache
;
3223 u64 user_thresh_min
;
3224 u64 user_thresh_max
;
3227 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3228 chunk_used
= btrfs_block_group_used(&cache
->item
);
3230 if (bargs
->usage_min
== 0)
3231 user_thresh_min
= 0;
3233 user_thresh_min
= div_factor_fine(cache
->key
.offset
,
3236 if (bargs
->usage_max
== 0)
3237 user_thresh_max
= 1;
3238 else if (bargs
->usage_max
> 100)
3239 user_thresh_max
= cache
->key
.offset
;
3241 user_thresh_max
= div_factor_fine(cache
->key
.offset
,
3244 if (user_thresh_min
<= chunk_used
&& chunk_used
< user_thresh_max
)
3247 btrfs_put_block_group(cache
);
3251 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
,
3252 u64 chunk_offset
, struct btrfs_balance_args
*bargs
)
3254 struct btrfs_block_group_cache
*cache
;
3255 u64 chunk_used
, user_thresh
;
3258 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3259 chunk_used
= btrfs_block_group_used(&cache
->item
);
3261 if (bargs
->usage_min
== 0)
3263 else if (bargs
->usage
> 100)
3264 user_thresh
= cache
->key
.offset
;
3266 user_thresh
= div_factor_fine(cache
->key
.offset
,
3269 if (chunk_used
< user_thresh
)
3272 btrfs_put_block_group(cache
);
3276 static int chunk_devid_filter(struct extent_buffer
*leaf
,
3277 struct btrfs_chunk
*chunk
,
3278 struct btrfs_balance_args
*bargs
)
3280 struct btrfs_stripe
*stripe
;
3281 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3284 for (i
= 0; i
< num_stripes
; i
++) {
3285 stripe
= btrfs_stripe_nr(chunk
, i
);
3286 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
3293 /* [pstart, pend) */
3294 static int chunk_drange_filter(struct extent_buffer
*leaf
,
3295 struct btrfs_chunk
*chunk
,
3296 struct btrfs_balance_args
*bargs
)
3298 struct btrfs_stripe
*stripe
;
3299 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3305 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
3308 if (btrfs_chunk_type(leaf
, chunk
) & (BTRFS_BLOCK_GROUP_DUP
|
3309 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)) {
3310 factor
= num_stripes
/ 2;
3311 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID5
) {
3312 factor
= num_stripes
- 1;
3313 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID6
) {
3314 factor
= num_stripes
- 2;
3316 factor
= num_stripes
;
3319 for (i
= 0; i
< num_stripes
; i
++) {
3320 stripe
= btrfs_stripe_nr(chunk
, i
);
3321 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
3324 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
3325 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
3326 stripe_length
= div_u64(stripe_length
, factor
);
3328 if (stripe_offset
< bargs
->pend
&&
3329 stripe_offset
+ stripe_length
> bargs
->pstart
)
3336 /* [vstart, vend) */
3337 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
3338 struct btrfs_chunk
*chunk
,
3340 struct btrfs_balance_args
*bargs
)
3342 if (chunk_offset
< bargs
->vend
&&
3343 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
3344 /* at least part of the chunk is inside this vrange */
3350 static int chunk_stripes_range_filter(struct extent_buffer
*leaf
,
3351 struct btrfs_chunk
*chunk
,
3352 struct btrfs_balance_args
*bargs
)
3354 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3356 if (bargs
->stripes_min
<= num_stripes
3357 && num_stripes
<= bargs
->stripes_max
)
3363 static int chunk_soft_convert_filter(u64 chunk_type
,
3364 struct btrfs_balance_args
*bargs
)
3366 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
3369 chunk_type
= chunk_to_extended(chunk_type
) &
3370 BTRFS_EXTENDED_PROFILE_MASK
;
3372 if (bargs
->target
== chunk_type
)
3378 static int should_balance_chunk(struct btrfs_fs_info
*fs_info
,
3379 struct extent_buffer
*leaf
,
3380 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
3382 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3383 struct btrfs_balance_args
*bargs
= NULL
;
3384 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3387 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
3388 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
3392 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3393 bargs
= &bctl
->data
;
3394 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3396 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3397 bargs
= &bctl
->meta
;
3399 /* profiles filter */
3400 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
3401 chunk_profiles_filter(chunk_type
, bargs
)) {
3406 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3407 chunk_usage_filter(fs_info
, chunk_offset
, bargs
)) {
3409 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3410 chunk_usage_range_filter(fs_info
, chunk_offset
, bargs
)) {
3415 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
3416 chunk_devid_filter(leaf
, chunk
, bargs
)) {
3420 /* drange filter, makes sense only with devid filter */
3421 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
3422 chunk_drange_filter(leaf
, chunk
, bargs
)) {
3427 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
3428 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3432 /* stripes filter */
3433 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_STRIPES_RANGE
) &&
3434 chunk_stripes_range_filter(leaf
, chunk
, bargs
)) {
3438 /* soft profile changing mode */
3439 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
3440 chunk_soft_convert_filter(chunk_type
, bargs
)) {
3445 * limited by count, must be the last filter
3447 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT
)) {
3448 if (bargs
->limit
== 0)
3452 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT_RANGE
)) {
3454 * Same logic as the 'limit' filter; the minimum cannot be
3455 * determined here because we do not have the global information
3456 * about the count of all chunks that satisfy the filters.
3458 if (bargs
->limit_max
== 0)
3467 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
3469 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3470 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
3471 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
3472 struct list_head
*devices
;
3473 struct btrfs_device
*device
;
3477 struct btrfs_chunk
*chunk
;
3478 struct btrfs_path
*path
= NULL
;
3479 struct btrfs_key key
;
3480 struct btrfs_key found_key
;
3481 struct btrfs_trans_handle
*trans
;
3482 struct extent_buffer
*leaf
;
3485 int enospc_errors
= 0;
3486 bool counting
= true;
3487 /* The single value limit and min/max limits use the same bytes in the */
3488 u64 limit_data
= bctl
->data
.limit
;
3489 u64 limit_meta
= bctl
->meta
.limit
;
3490 u64 limit_sys
= bctl
->sys
.limit
;
3494 int chunk_reserved
= 0;
3497 /* step one make some room on all the devices */
3498 devices
= &fs_info
->fs_devices
->devices
;
3499 list_for_each_entry(device
, devices
, dev_list
) {
3500 old_size
= btrfs_device_get_total_bytes(device
);
3501 size_to_free
= div_factor(old_size
, 1);
3502 size_to_free
= min_t(u64
, size_to_free
, SZ_1M
);
3503 if (!device
->writeable
||
3504 btrfs_device_get_total_bytes(device
) -
3505 btrfs_device_get_bytes_used(device
) > size_to_free
||
3506 device
->is_tgtdev_for_dev_replace
)
3509 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
3513 /* btrfs_shrink_device never returns ret > 0 */
3518 trans
= btrfs_start_transaction(dev_root
, 0);
3519 if (IS_ERR(trans
)) {
3520 ret
= PTR_ERR(trans
);
3521 btrfs_info_in_rcu(fs_info
,
3522 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3523 rcu_str_deref(device
->name
), ret
,
3524 old_size
, old_size
- size_to_free
);
3528 ret
= btrfs_grow_device(trans
, device
, old_size
);
3530 btrfs_end_transaction(trans
);
3531 /* btrfs_grow_device never returns ret > 0 */
3533 btrfs_info_in_rcu(fs_info
,
3534 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3535 rcu_str_deref(device
->name
), ret
,
3536 old_size
, old_size
- size_to_free
);
3540 btrfs_end_transaction(trans
);
3543 /* step two, relocate all the chunks */
3544 path
= btrfs_alloc_path();
3550 /* zero out stat counters */
3551 spin_lock(&fs_info
->balance_lock
);
3552 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
3553 spin_unlock(&fs_info
->balance_lock
);
3557 * The single value limit and min/max limits use the same bytes
3560 bctl
->data
.limit
= limit_data
;
3561 bctl
->meta
.limit
= limit_meta
;
3562 bctl
->sys
.limit
= limit_sys
;
3564 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3565 key
.offset
= (u64
)-1;
3566 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3569 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
3570 atomic_read(&fs_info
->balance_cancel_req
)) {
3575 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
3576 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
3578 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3583 * this shouldn't happen, it means the last relocate
3587 BUG(); /* FIXME break ? */
3589 ret
= btrfs_previous_item(chunk_root
, path
, 0,
3590 BTRFS_CHUNK_ITEM_KEY
);
3592 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3597 leaf
= path
->nodes
[0];
3598 slot
= path
->slots
[0];
3599 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3601 if (found_key
.objectid
!= key
.objectid
) {
3602 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3606 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
3607 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3610 spin_lock(&fs_info
->balance_lock
);
3611 bctl
->stat
.considered
++;
3612 spin_unlock(&fs_info
->balance_lock
);
3615 ret
= should_balance_chunk(fs_info
, leaf
, chunk
,
3618 btrfs_release_path(path
);
3620 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3625 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3626 spin_lock(&fs_info
->balance_lock
);
3627 bctl
->stat
.expected
++;
3628 spin_unlock(&fs_info
->balance_lock
);
3630 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3632 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3634 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3641 * Apply limit_min filter, no need to check if the LIMITS
3642 * filter is used, limit_min is 0 by default
3644 if (((chunk_type
& BTRFS_BLOCK_GROUP_DATA
) &&
3645 count_data
< bctl
->data
.limit_min
)
3646 || ((chunk_type
& BTRFS_BLOCK_GROUP_METADATA
) &&
3647 count_meta
< bctl
->meta
.limit_min
)
3648 || ((chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) &&
3649 count_sys
< bctl
->sys
.limit_min
)) {
3650 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3654 ASSERT(fs_info
->data_sinfo
);
3655 spin_lock(&fs_info
->data_sinfo
->lock
);
3656 bytes_used
= fs_info
->data_sinfo
->bytes_used
;
3657 spin_unlock(&fs_info
->data_sinfo
->lock
);
3659 if ((chunk_type
& BTRFS_BLOCK_GROUP_DATA
) &&
3660 !chunk_reserved
&& !bytes_used
) {
3661 trans
= btrfs_start_transaction(chunk_root
, 0);
3662 if (IS_ERR(trans
)) {
3663 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3664 ret
= PTR_ERR(trans
);
3668 ret
= btrfs_force_chunk_alloc(trans
, fs_info
,
3669 BTRFS_BLOCK_GROUP_DATA
);
3670 btrfs_end_transaction(trans
);
3672 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3678 ret
= btrfs_relocate_chunk(fs_info
, found_key
.offset
);
3679 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3680 if (ret
&& ret
!= -ENOSPC
)
3682 if (ret
== -ENOSPC
) {
3685 spin_lock(&fs_info
->balance_lock
);
3686 bctl
->stat
.completed
++;
3687 spin_unlock(&fs_info
->balance_lock
);
3690 if (found_key
.offset
== 0)
3692 key
.offset
= found_key
.offset
- 1;
3696 btrfs_release_path(path
);
3701 btrfs_free_path(path
);
3702 if (enospc_errors
) {
3703 btrfs_info(fs_info
, "%d enospc errors during balance",
3713 * alloc_profile_is_valid - see if a given profile is valid and reduced
3714 * @flags: profile to validate
3715 * @extended: if true @flags is treated as an extended profile
3717 static int alloc_profile_is_valid(u64 flags
, int extended
)
3719 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
3720 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
3722 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
3724 /* 1) check that all other bits are zeroed */
3728 /* 2) see if profile is reduced */
3730 return !extended
; /* "0" is valid for usual profiles */
3732 /* true if exactly one bit set */
3733 return (flags
& (flags
- 1)) == 0;
3736 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
3738 /* cancel requested || normal exit path */
3739 return atomic_read(&fs_info
->balance_cancel_req
) ||
3740 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
3741 atomic_read(&fs_info
->balance_cancel_req
) == 0);
3744 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
3748 unset_balance_control(fs_info
);
3749 ret
= del_balance_item(fs_info
);
3751 btrfs_handle_fs_error(fs_info
, ret
, NULL
);
3753 clear_bit(BTRFS_FS_EXCL_OP
, &fs_info
->flags
);
3756 /* Non-zero return value signifies invalidity */
3757 static inline int validate_convert_profile(struct btrfs_balance_args
*bctl_arg
,
3760 return ((bctl_arg
->flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3761 (!alloc_profile_is_valid(bctl_arg
->target
, 1) ||
3762 (bctl_arg
->target
& ~allowed
)));
3766 * Should be called with both balance and volume mutexes held
3768 int btrfs_balance(struct btrfs_balance_control
*bctl
,
3769 struct btrfs_ioctl_balance_args
*bargs
)
3771 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3772 u64 meta_target
, data_target
;
3779 if (btrfs_fs_closing(fs_info
) ||
3780 atomic_read(&fs_info
->balance_pause_req
) ||
3781 atomic_read(&fs_info
->balance_cancel_req
)) {
3786 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
3787 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
3791 * In case of mixed groups both data and meta should be picked,
3792 * and identical options should be given for both of them.
3794 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
3795 if (mixed
&& (bctl
->flags
& allowed
)) {
3796 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
3797 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
3798 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
3800 "with mixed groups data and metadata balance options must be the same");
3806 num_devices
= fs_info
->fs_devices
->num_devices
;
3807 btrfs_dev_replace_lock(&fs_info
->dev_replace
, 0);
3808 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
3809 BUG_ON(num_devices
< 1);
3812 btrfs_dev_replace_unlock(&fs_info
->dev_replace
, 0);
3813 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
| BTRFS_BLOCK_GROUP_DUP
;
3814 if (num_devices
> 1)
3815 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
3816 if (num_devices
> 2)
3817 allowed
|= BTRFS_BLOCK_GROUP_RAID5
;
3818 if (num_devices
> 3)
3819 allowed
|= (BTRFS_BLOCK_GROUP_RAID10
|
3820 BTRFS_BLOCK_GROUP_RAID6
);
3821 if (validate_convert_profile(&bctl
->data
, allowed
)) {
3823 "unable to start balance with target data profile %llu",
3828 if (validate_convert_profile(&bctl
->meta
, allowed
)) {
3830 "unable to start balance with target metadata profile %llu",
3835 if (validate_convert_profile(&bctl
->sys
, allowed
)) {
3837 "unable to start balance with target system profile %llu",
3843 /* allow to reduce meta or sys integrity only if force set */
3844 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3845 BTRFS_BLOCK_GROUP_RAID10
|
3846 BTRFS_BLOCK_GROUP_RAID5
|
3847 BTRFS_BLOCK_GROUP_RAID6
;
3849 seq
= read_seqbegin(&fs_info
->profiles_lock
);
3851 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3852 (fs_info
->avail_system_alloc_bits
& allowed
) &&
3853 !(bctl
->sys
.target
& allowed
)) ||
3854 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3855 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
3856 !(bctl
->meta
.target
& allowed
))) {
3857 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
3859 "force reducing metadata integrity");
3862 "balance will reduce metadata integrity, use force if you want this");
3867 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
3869 /* if we're not converting, the target field is uninitialized */
3870 meta_target
= (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) ?
3871 bctl
->meta
.target
: fs_info
->avail_metadata_alloc_bits
;
3872 data_target
= (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) ?
3873 bctl
->data
.target
: fs_info
->avail_data_alloc_bits
;
3874 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target
) <
3875 btrfs_get_num_tolerated_disk_barrier_failures(data_target
)) {
3877 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3878 meta_target
, data_target
);
3881 ret
= insert_balance_item(fs_info
, bctl
);
3882 if (ret
&& ret
!= -EEXIST
)
3885 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
3886 BUG_ON(ret
== -EEXIST
);
3887 set_balance_control(bctl
);
3889 BUG_ON(ret
!= -EEXIST
);
3890 spin_lock(&fs_info
->balance_lock
);
3891 update_balance_args(bctl
);
3892 spin_unlock(&fs_info
->balance_lock
);
3895 atomic_inc(&fs_info
->balance_running
);
3896 mutex_unlock(&fs_info
->balance_mutex
);
3898 ret
= __btrfs_balance(fs_info
);
3900 mutex_lock(&fs_info
->balance_mutex
);
3901 atomic_dec(&fs_info
->balance_running
);
3904 memset(bargs
, 0, sizeof(*bargs
));
3905 update_ioctl_balance_args(fs_info
, 0, bargs
);
3908 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
3909 balance_need_close(fs_info
)) {
3910 __cancel_balance(fs_info
);
3913 wake_up(&fs_info
->balance_wait_q
);
3917 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
3918 __cancel_balance(fs_info
);
3921 clear_bit(BTRFS_FS_EXCL_OP
, &fs_info
->flags
);
3926 static int balance_kthread(void *data
)
3928 struct btrfs_fs_info
*fs_info
= data
;
3931 mutex_lock(&fs_info
->volume_mutex
);
3932 mutex_lock(&fs_info
->balance_mutex
);
3934 if (fs_info
->balance_ctl
) {
3935 btrfs_info(fs_info
, "continuing balance");
3936 ret
= btrfs_balance(fs_info
->balance_ctl
, NULL
);
3939 mutex_unlock(&fs_info
->balance_mutex
);
3940 mutex_unlock(&fs_info
->volume_mutex
);
3945 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
3947 struct task_struct
*tsk
;
3949 spin_lock(&fs_info
->balance_lock
);
3950 if (!fs_info
->balance_ctl
) {
3951 spin_unlock(&fs_info
->balance_lock
);
3954 spin_unlock(&fs_info
->balance_lock
);
3956 if (btrfs_test_opt(fs_info
, SKIP_BALANCE
)) {
3957 btrfs_info(fs_info
, "force skipping balance");
3961 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
3962 return PTR_ERR_OR_ZERO(tsk
);
3965 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
3967 struct btrfs_balance_control
*bctl
;
3968 struct btrfs_balance_item
*item
;
3969 struct btrfs_disk_balance_args disk_bargs
;
3970 struct btrfs_path
*path
;
3971 struct extent_buffer
*leaf
;
3972 struct btrfs_key key
;
3975 path
= btrfs_alloc_path();
3979 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3980 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3983 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
3986 if (ret
> 0) { /* ret = -ENOENT; */
3991 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
3997 leaf
= path
->nodes
[0];
3998 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
4000 bctl
->fs_info
= fs_info
;
4001 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
4002 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
4004 btrfs_balance_data(leaf
, item
, &disk_bargs
);
4005 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
4006 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
4007 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
4008 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
4009 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
4011 WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP
, &fs_info
->flags
));
4013 mutex_lock(&fs_info
->volume_mutex
);
4014 mutex_lock(&fs_info
->balance_mutex
);
4016 set_balance_control(bctl
);
4018 mutex_unlock(&fs_info
->balance_mutex
);
4019 mutex_unlock(&fs_info
->volume_mutex
);
4021 btrfs_free_path(path
);
4025 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
4029 mutex_lock(&fs_info
->balance_mutex
);
4030 if (!fs_info
->balance_ctl
) {
4031 mutex_unlock(&fs_info
->balance_mutex
);
4035 if (atomic_read(&fs_info
->balance_running
)) {
4036 atomic_inc(&fs_info
->balance_pause_req
);
4037 mutex_unlock(&fs_info
->balance_mutex
);
4039 wait_event(fs_info
->balance_wait_q
,
4040 atomic_read(&fs_info
->balance_running
) == 0);
4042 mutex_lock(&fs_info
->balance_mutex
);
4043 /* we are good with balance_ctl ripped off from under us */
4044 BUG_ON(atomic_read(&fs_info
->balance_running
));
4045 atomic_dec(&fs_info
->balance_pause_req
);
4050 mutex_unlock(&fs_info
->balance_mutex
);
4054 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
4056 if (sb_rdonly(fs_info
->sb
))
4059 mutex_lock(&fs_info
->balance_mutex
);
4060 if (!fs_info
->balance_ctl
) {
4061 mutex_unlock(&fs_info
->balance_mutex
);
4065 atomic_inc(&fs_info
->balance_cancel_req
);
4067 * if we are running just wait and return, balance item is
4068 * deleted in btrfs_balance in this case
4070 if (atomic_read(&fs_info
->balance_running
)) {
4071 mutex_unlock(&fs_info
->balance_mutex
);
4072 wait_event(fs_info
->balance_wait_q
,
4073 atomic_read(&fs_info
->balance_running
) == 0);
4074 mutex_lock(&fs_info
->balance_mutex
);
4076 /* __cancel_balance needs volume_mutex */
4077 mutex_unlock(&fs_info
->balance_mutex
);
4078 mutex_lock(&fs_info
->volume_mutex
);
4079 mutex_lock(&fs_info
->balance_mutex
);
4081 if (fs_info
->balance_ctl
)
4082 __cancel_balance(fs_info
);
4084 mutex_unlock(&fs_info
->volume_mutex
);
4087 BUG_ON(fs_info
->balance_ctl
|| atomic_read(&fs_info
->balance_running
));
4088 atomic_dec(&fs_info
->balance_cancel_req
);
4089 mutex_unlock(&fs_info
->balance_mutex
);
4093 static int btrfs_uuid_scan_kthread(void *data
)
4095 struct btrfs_fs_info
*fs_info
= data
;
4096 struct btrfs_root
*root
= fs_info
->tree_root
;
4097 struct btrfs_key key
;
4098 struct btrfs_path
*path
= NULL
;
4100 struct extent_buffer
*eb
;
4102 struct btrfs_root_item root_item
;
4104 struct btrfs_trans_handle
*trans
= NULL
;
4106 path
= btrfs_alloc_path();
4113 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4117 ret
= btrfs_search_forward(root
, &key
, path
, 0);
4124 if (key
.type
!= BTRFS_ROOT_ITEM_KEY
||
4125 (key
.objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
4126 key
.objectid
!= BTRFS_FS_TREE_OBJECTID
) ||
4127 key
.objectid
> BTRFS_LAST_FREE_OBJECTID
)
4130 eb
= path
->nodes
[0];
4131 slot
= path
->slots
[0];
4132 item_size
= btrfs_item_size_nr(eb
, slot
);
4133 if (item_size
< sizeof(root_item
))
4136 read_extent_buffer(eb
, &root_item
,
4137 btrfs_item_ptr_offset(eb
, slot
),
4138 (int)sizeof(root_item
));
4139 if (btrfs_root_refs(&root_item
) == 0)
4142 if (!btrfs_is_empty_uuid(root_item
.uuid
) ||
4143 !btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4147 btrfs_release_path(path
);
4149 * 1 - subvol uuid item
4150 * 1 - received_subvol uuid item
4152 trans
= btrfs_start_transaction(fs_info
->uuid_root
, 2);
4153 if (IS_ERR(trans
)) {
4154 ret
= PTR_ERR(trans
);
4162 if (!btrfs_is_empty_uuid(root_item
.uuid
)) {
4163 ret
= btrfs_uuid_tree_add(trans
, fs_info
,
4165 BTRFS_UUID_KEY_SUBVOL
,
4168 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4174 if (!btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4175 ret
= btrfs_uuid_tree_add(trans
, fs_info
,
4176 root_item
.received_uuid
,
4177 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
4180 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4188 ret
= btrfs_end_transaction(trans
);
4194 btrfs_release_path(path
);
4195 if (key
.offset
< (u64
)-1) {
4197 } else if (key
.type
< BTRFS_ROOT_ITEM_KEY
) {
4199 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4200 } else if (key
.objectid
< (u64
)-1) {
4202 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4211 btrfs_free_path(path
);
4212 if (trans
&& !IS_ERR(trans
))
4213 btrfs_end_transaction(trans
);
4215 btrfs_warn(fs_info
, "btrfs_uuid_scan_kthread failed %d", ret
);
4217 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN
, &fs_info
->flags
);
4218 up(&fs_info
->uuid_tree_rescan_sem
);
4223 * Callback for btrfs_uuid_tree_iterate().
4225 * 0 check succeeded, the entry is not outdated.
4226 * < 0 if an error occurred.
4227 * > 0 if the check failed, which means the caller shall remove the entry.
4229 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info
*fs_info
,
4230 u8
*uuid
, u8 type
, u64 subid
)
4232 struct btrfs_key key
;
4234 struct btrfs_root
*subvol_root
;
4236 if (type
!= BTRFS_UUID_KEY_SUBVOL
&&
4237 type
!= BTRFS_UUID_KEY_RECEIVED_SUBVOL
)
4240 key
.objectid
= subid
;
4241 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4242 key
.offset
= (u64
)-1;
4243 subvol_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
4244 if (IS_ERR(subvol_root
)) {
4245 ret
= PTR_ERR(subvol_root
);
4252 case BTRFS_UUID_KEY_SUBVOL
:
4253 if (memcmp(uuid
, subvol_root
->root_item
.uuid
, BTRFS_UUID_SIZE
))
4256 case BTRFS_UUID_KEY_RECEIVED_SUBVOL
:
4257 if (memcmp(uuid
, subvol_root
->root_item
.received_uuid
,
4267 static int btrfs_uuid_rescan_kthread(void *data
)
4269 struct btrfs_fs_info
*fs_info
= (struct btrfs_fs_info
*)data
;
4273 * 1st step is to iterate through the existing UUID tree and
4274 * to delete all entries that contain outdated data.
4275 * 2nd step is to add all missing entries to the UUID tree.
4277 ret
= btrfs_uuid_tree_iterate(fs_info
, btrfs_check_uuid_tree_entry
);
4279 btrfs_warn(fs_info
, "iterating uuid_tree failed %d", ret
);
4280 up(&fs_info
->uuid_tree_rescan_sem
);
4283 return btrfs_uuid_scan_kthread(data
);
4286 int btrfs_create_uuid_tree(struct btrfs_fs_info
*fs_info
)
4288 struct btrfs_trans_handle
*trans
;
4289 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
4290 struct btrfs_root
*uuid_root
;
4291 struct task_struct
*task
;
4298 trans
= btrfs_start_transaction(tree_root
, 2);
4300 return PTR_ERR(trans
);
4302 uuid_root
= btrfs_create_tree(trans
, fs_info
,
4303 BTRFS_UUID_TREE_OBJECTID
);
4304 if (IS_ERR(uuid_root
)) {
4305 ret
= PTR_ERR(uuid_root
);
4306 btrfs_abort_transaction(trans
, ret
);
4307 btrfs_end_transaction(trans
);
4311 fs_info
->uuid_root
= uuid_root
;
4313 ret
= btrfs_commit_transaction(trans
);
4317 down(&fs_info
->uuid_tree_rescan_sem
);
4318 task
= kthread_run(btrfs_uuid_scan_kthread
, fs_info
, "btrfs-uuid");
4320 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4321 btrfs_warn(fs_info
, "failed to start uuid_scan task");
4322 up(&fs_info
->uuid_tree_rescan_sem
);
4323 return PTR_ERR(task
);
4329 int btrfs_check_uuid_tree(struct btrfs_fs_info
*fs_info
)
4331 struct task_struct
*task
;
4333 down(&fs_info
->uuid_tree_rescan_sem
);
4334 task
= kthread_run(btrfs_uuid_rescan_kthread
, fs_info
, "btrfs-uuid");
4336 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4337 btrfs_warn(fs_info
, "failed to start uuid_rescan task");
4338 up(&fs_info
->uuid_tree_rescan_sem
);
4339 return PTR_ERR(task
);
4346 * shrinking a device means finding all of the device extents past
4347 * the new size, and then following the back refs to the chunks.
4348 * The chunk relocation code actually frees the device extent
4350 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
4352 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
4353 struct btrfs_root
*root
= fs_info
->dev_root
;
4354 struct btrfs_trans_handle
*trans
;
4355 struct btrfs_dev_extent
*dev_extent
= NULL
;
4356 struct btrfs_path
*path
;
4362 bool retried
= false;
4363 bool checked_pending_chunks
= false;
4364 struct extent_buffer
*l
;
4365 struct btrfs_key key
;
4366 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
4367 u64 old_total
= btrfs_super_total_bytes(super_copy
);
4368 u64 old_size
= btrfs_device_get_total_bytes(device
);
4371 new_size
= round_down(new_size
, fs_info
->sectorsize
);
4372 diff
= round_down(old_size
- new_size
, fs_info
->sectorsize
);
4374 if (device
->is_tgtdev_for_dev_replace
)
4377 path
= btrfs_alloc_path();
4381 path
->reada
= READA_FORWARD
;
4383 mutex_lock(&fs_info
->chunk_mutex
);
4385 btrfs_device_set_total_bytes(device
, new_size
);
4386 if (device
->writeable
) {
4387 device
->fs_devices
->total_rw_bytes
-= diff
;
4388 atomic64_sub(diff
, &fs_info
->free_chunk_space
);
4390 mutex_unlock(&fs_info
->chunk_mutex
);
4393 key
.objectid
= device
->devid
;
4394 key
.offset
= (u64
)-1;
4395 key
.type
= BTRFS_DEV_EXTENT_KEY
;
4398 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
4399 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4401 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4405 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
4407 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4412 btrfs_release_path(path
);
4417 slot
= path
->slots
[0];
4418 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
4420 if (key
.objectid
!= device
->devid
) {
4421 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4422 btrfs_release_path(path
);
4426 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
4427 length
= btrfs_dev_extent_length(l
, dev_extent
);
4429 if (key
.offset
+ length
<= new_size
) {
4430 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4431 btrfs_release_path(path
);
4435 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
4436 btrfs_release_path(path
);
4438 ret
= btrfs_relocate_chunk(fs_info
, chunk_offset
);
4439 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4440 if (ret
&& ret
!= -ENOSPC
)
4444 } while (key
.offset
-- > 0);
4446 if (failed
&& !retried
) {
4450 } else if (failed
&& retried
) {
4455 /* Shrinking succeeded, else we would be at "done". */
4456 trans
= btrfs_start_transaction(root
, 0);
4457 if (IS_ERR(trans
)) {
4458 ret
= PTR_ERR(trans
);
4462 mutex_lock(&fs_info
->chunk_mutex
);
4465 * We checked in the above loop all device extents that were already in
4466 * the device tree. However before we have updated the device's
4467 * total_bytes to the new size, we might have had chunk allocations that
4468 * have not complete yet (new block groups attached to transaction
4469 * handles), and therefore their device extents were not yet in the
4470 * device tree and we missed them in the loop above. So if we have any
4471 * pending chunk using a device extent that overlaps the device range
4472 * that we can not use anymore, commit the current transaction and
4473 * repeat the search on the device tree - this way we guarantee we will
4474 * not have chunks using device extents that end beyond 'new_size'.
4476 if (!checked_pending_chunks
) {
4477 u64 start
= new_size
;
4478 u64 len
= old_size
- new_size
;
4480 if (contains_pending_extent(trans
->transaction
, device
,
4482 mutex_unlock(&fs_info
->chunk_mutex
);
4483 checked_pending_chunks
= true;
4486 ret
= btrfs_commit_transaction(trans
);
4493 btrfs_device_set_disk_total_bytes(device
, new_size
);
4494 if (list_empty(&device
->resized_list
))
4495 list_add_tail(&device
->resized_list
,
4496 &fs_info
->fs_devices
->resized_devices
);
4498 WARN_ON(diff
> old_total
);
4499 btrfs_set_super_total_bytes(super_copy
,
4500 round_down(old_total
- diff
, fs_info
->sectorsize
));
4501 mutex_unlock(&fs_info
->chunk_mutex
);
4503 /* Now btrfs_update_device() will change the on-disk size. */
4504 ret
= btrfs_update_device(trans
, device
);
4505 btrfs_end_transaction(trans
);
4507 btrfs_free_path(path
);
4509 mutex_lock(&fs_info
->chunk_mutex
);
4510 btrfs_device_set_total_bytes(device
, old_size
);
4511 if (device
->writeable
)
4512 device
->fs_devices
->total_rw_bytes
+= diff
;
4513 atomic64_add(diff
, &fs_info
->free_chunk_space
);
4514 mutex_unlock(&fs_info
->chunk_mutex
);
4519 static int btrfs_add_system_chunk(struct btrfs_fs_info
*fs_info
,
4520 struct btrfs_key
*key
,
4521 struct btrfs_chunk
*chunk
, int item_size
)
4523 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
4524 struct btrfs_disk_key disk_key
;
4528 mutex_lock(&fs_info
->chunk_mutex
);
4529 array_size
= btrfs_super_sys_array_size(super_copy
);
4530 if (array_size
+ item_size
+ sizeof(disk_key
)
4531 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4532 mutex_unlock(&fs_info
->chunk_mutex
);
4536 ptr
= super_copy
->sys_chunk_array
+ array_size
;
4537 btrfs_cpu_key_to_disk(&disk_key
, key
);
4538 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
4539 ptr
+= sizeof(disk_key
);
4540 memcpy(ptr
, chunk
, item_size
);
4541 item_size
+= sizeof(disk_key
);
4542 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
4543 mutex_unlock(&fs_info
->chunk_mutex
);
4549 * sort the devices in descending order by max_avail, total_avail
4551 static int btrfs_cmp_device_info(const void *a
, const void *b
)
4553 const struct btrfs_device_info
*di_a
= a
;
4554 const struct btrfs_device_info
*di_b
= b
;
4556 if (di_a
->max_avail
> di_b
->max_avail
)
4558 if (di_a
->max_avail
< di_b
->max_avail
)
4560 if (di_a
->total_avail
> di_b
->total_avail
)
4562 if (di_a
->total_avail
< di_b
->total_avail
)
4567 static void check_raid56_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
4569 if (!(type
& BTRFS_BLOCK_GROUP_RAID56_MASK
))
4572 btrfs_set_fs_incompat(info
, RAID56
);
4575 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info) \
4576 - sizeof(struct btrfs_chunk)) \
4577 / sizeof(struct btrfs_stripe) + 1)
4579 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4580 - 2 * sizeof(struct btrfs_disk_key) \
4581 - 2 * sizeof(struct btrfs_chunk)) \
4582 / sizeof(struct btrfs_stripe) + 1)
4584 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4585 u64 start
, u64 type
)
4587 struct btrfs_fs_info
*info
= trans
->fs_info
;
4588 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
4589 struct btrfs_device
*device
;
4590 struct map_lookup
*map
= NULL
;
4591 struct extent_map_tree
*em_tree
;
4592 struct extent_map
*em
;
4593 struct btrfs_device_info
*devices_info
= NULL
;
4595 int num_stripes
; /* total number of stripes to allocate */
4596 int data_stripes
; /* number of stripes that count for
4598 int sub_stripes
; /* sub_stripes info for map */
4599 int dev_stripes
; /* stripes per dev */
4600 int devs_max
; /* max devs to use */
4601 int devs_min
; /* min devs needed */
4602 int devs_increment
; /* ndevs has to be a multiple of this */
4603 int ncopies
; /* how many copies to data has */
4605 u64 max_stripe_size
;
4614 BUG_ON(!alloc_profile_is_valid(type
, 0));
4616 if (list_empty(&fs_devices
->alloc_list
))
4619 index
= __get_raid_index(type
);
4621 sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
4622 dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
4623 devs_max
= btrfs_raid_array
[index
].devs_max
;
4624 devs_min
= btrfs_raid_array
[index
].devs_min
;
4625 devs_increment
= btrfs_raid_array
[index
].devs_increment
;
4626 ncopies
= btrfs_raid_array
[index
].ncopies
;
4628 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
4629 max_stripe_size
= SZ_1G
;
4630 max_chunk_size
= 10 * max_stripe_size
;
4632 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4633 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
4634 /* for larger filesystems, use larger metadata chunks */
4635 if (fs_devices
->total_rw_bytes
> 50ULL * SZ_1G
)
4636 max_stripe_size
= SZ_1G
;
4638 max_stripe_size
= SZ_256M
;
4639 max_chunk_size
= max_stripe_size
;
4641 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4642 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4643 max_stripe_size
= SZ_32M
;
4644 max_chunk_size
= 2 * max_stripe_size
;
4646 devs_max
= BTRFS_MAX_DEVS_SYS_CHUNK
;
4648 btrfs_err(info
, "invalid chunk type 0x%llx requested",
4653 /* we don't want a chunk larger than 10% of writeable space */
4654 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
4657 devices_info
= kcalloc(fs_devices
->rw_devices
, sizeof(*devices_info
),
4663 * in the first pass through the devices list, we gather information
4664 * about the available holes on each device.
4667 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
4671 if (!device
->writeable
) {
4673 "BTRFS: read-only device in alloc_list\n");
4677 if (!device
->in_fs_metadata
||
4678 device
->is_tgtdev_for_dev_replace
)
4681 if (device
->total_bytes
> device
->bytes_used
)
4682 total_avail
= device
->total_bytes
- device
->bytes_used
;
4686 /* If there is no space on this device, skip it. */
4687 if (total_avail
== 0)
4690 ret
= find_free_dev_extent(trans
, device
,
4691 max_stripe_size
* dev_stripes
,
4692 &dev_offset
, &max_avail
);
4693 if (ret
&& ret
!= -ENOSPC
)
4697 max_avail
= max_stripe_size
* dev_stripes
;
4699 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
4702 if (ndevs
== fs_devices
->rw_devices
) {
4703 WARN(1, "%s: found more than %llu devices\n",
4704 __func__
, fs_devices
->rw_devices
);
4707 devices_info
[ndevs
].dev_offset
= dev_offset
;
4708 devices_info
[ndevs
].max_avail
= max_avail
;
4709 devices_info
[ndevs
].total_avail
= total_avail
;
4710 devices_info
[ndevs
].dev
= device
;
4715 * now sort the devices by hole size / available space
4717 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
4718 btrfs_cmp_device_info
, NULL
);
4720 /* round down to number of usable stripes */
4721 ndevs
= round_down(ndevs
, devs_increment
);
4723 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
4728 ndevs
= min(ndevs
, devs_max
);
4731 * the primary goal is to maximize the number of stripes, so use as many
4732 * devices as possible, even if the stripes are not maximum sized.
4734 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4735 num_stripes
= ndevs
* dev_stripes
;
4738 * this will have to be fixed for RAID1 and RAID10 over
4741 data_stripes
= num_stripes
/ ncopies
;
4743 if (type
& BTRFS_BLOCK_GROUP_RAID5
)
4744 data_stripes
= num_stripes
- 1;
4746 if (type
& BTRFS_BLOCK_GROUP_RAID6
)
4747 data_stripes
= num_stripes
- 2;
4750 * Use the number of data stripes to figure out how big this chunk
4751 * is really going to be in terms of logical address space,
4752 * and compare that answer with the max chunk size
4754 if (stripe_size
* data_stripes
> max_chunk_size
) {
4755 u64 mask
= (1ULL << 24) - 1;
4757 stripe_size
= div_u64(max_chunk_size
, data_stripes
);
4759 /* bump the answer up to a 16MB boundary */
4760 stripe_size
= (stripe_size
+ mask
) & ~mask
;
4762 /* but don't go higher than the limits we found
4763 * while searching for free extents
4765 if (stripe_size
> devices_info
[ndevs
-1].max_avail
)
4766 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4769 stripe_size
= div_u64(stripe_size
, dev_stripes
);
4771 /* align to BTRFS_STRIPE_LEN */
4772 stripe_size
= round_down(stripe_size
, BTRFS_STRIPE_LEN
);
4774 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
4779 map
->num_stripes
= num_stripes
;
4781 for (i
= 0; i
< ndevs
; ++i
) {
4782 for (j
= 0; j
< dev_stripes
; ++j
) {
4783 int s
= i
* dev_stripes
+ j
;
4784 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
4785 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
4789 map
->stripe_len
= BTRFS_STRIPE_LEN
;
4790 map
->io_align
= BTRFS_STRIPE_LEN
;
4791 map
->io_width
= BTRFS_STRIPE_LEN
;
4793 map
->sub_stripes
= sub_stripes
;
4795 num_bytes
= stripe_size
* data_stripes
;
4797 trace_btrfs_chunk_alloc(info
, map
, start
, num_bytes
);
4799 em
= alloc_extent_map();
4805 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
4806 em
->map_lookup
= map
;
4808 em
->len
= num_bytes
;
4809 em
->block_start
= 0;
4810 em
->block_len
= em
->len
;
4811 em
->orig_block_len
= stripe_size
;
4813 em_tree
= &info
->mapping_tree
.map_tree
;
4814 write_lock(&em_tree
->lock
);
4815 ret
= add_extent_mapping(em_tree
, em
, 0);
4817 list_add_tail(&em
->list
, &trans
->transaction
->pending_chunks
);
4818 refcount_inc(&em
->refs
);
4820 write_unlock(&em_tree
->lock
);
4822 free_extent_map(em
);
4826 ret
= btrfs_make_block_group(trans
, info
, 0, type
, start
, num_bytes
);
4828 goto error_del_extent
;
4830 for (i
= 0; i
< map
->num_stripes
; i
++) {
4831 num_bytes
= map
->stripes
[i
].dev
->bytes_used
+ stripe_size
;
4832 btrfs_device_set_bytes_used(map
->stripes
[i
].dev
, num_bytes
);
4835 atomic64_sub(stripe_size
* map
->num_stripes
, &info
->free_chunk_space
);
4837 free_extent_map(em
);
4838 check_raid56_incompat_flag(info
, type
);
4840 kfree(devices_info
);
4844 write_lock(&em_tree
->lock
);
4845 remove_extent_mapping(em_tree
, em
);
4846 write_unlock(&em_tree
->lock
);
4848 /* One for our allocation */
4849 free_extent_map(em
);
4850 /* One for the tree reference */
4851 free_extent_map(em
);
4852 /* One for the pending_chunks list reference */
4853 free_extent_map(em
);
4855 kfree(devices_info
);
4859 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
4860 struct btrfs_fs_info
*fs_info
,
4861 u64 chunk_offset
, u64 chunk_size
)
4863 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
4864 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
4865 struct btrfs_key key
;
4866 struct btrfs_device
*device
;
4867 struct btrfs_chunk
*chunk
;
4868 struct btrfs_stripe
*stripe
;
4869 struct extent_map
*em
;
4870 struct map_lookup
*map
;
4877 em
= get_chunk_map(fs_info
, chunk_offset
, chunk_size
);
4881 map
= em
->map_lookup
;
4882 item_size
= btrfs_chunk_item_size(map
->num_stripes
);
4883 stripe_size
= em
->orig_block_len
;
4885 chunk
= kzalloc(item_size
, GFP_NOFS
);
4892 * Take the device list mutex to prevent races with the final phase of
4893 * a device replace operation that replaces the device object associated
4894 * with the map's stripes, because the device object's id can change
4895 * at any time during that final phase of the device replace operation
4896 * (dev-replace.c:btrfs_dev_replace_finishing()).
4898 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
4899 for (i
= 0; i
< map
->num_stripes
; i
++) {
4900 device
= map
->stripes
[i
].dev
;
4901 dev_offset
= map
->stripes
[i
].physical
;
4903 ret
= btrfs_update_device(trans
, device
);
4906 ret
= btrfs_alloc_dev_extent(trans
, device
, chunk_offset
,
4907 dev_offset
, stripe_size
);
4912 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
4916 stripe
= &chunk
->stripe
;
4917 for (i
= 0; i
< map
->num_stripes
; i
++) {
4918 device
= map
->stripes
[i
].dev
;
4919 dev_offset
= map
->stripes
[i
].physical
;
4921 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
4922 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
4923 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
4926 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
4928 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
4929 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
4930 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
4931 btrfs_set_stack_chunk_type(chunk
, map
->type
);
4932 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
4933 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
4934 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
4935 btrfs_set_stack_chunk_sector_size(chunk
, fs_info
->sectorsize
);
4936 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
4938 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
4939 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
4940 key
.offset
= chunk_offset
;
4942 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
4943 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4945 * TODO: Cleanup of inserted chunk root in case of
4948 ret
= btrfs_add_system_chunk(fs_info
, &key
, chunk
, item_size
);
4953 free_extent_map(em
);
4958 * Chunk allocation falls into two parts. The first part does works
4959 * that make the new allocated chunk useable, but not do any operation
4960 * that modifies the chunk tree. The second part does the works that
4961 * require modifying the chunk tree. This division is important for the
4962 * bootstrap process of adding storage to a seed btrfs.
4964 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4965 struct btrfs_fs_info
*fs_info
, u64 type
)
4969 ASSERT(mutex_is_locked(&fs_info
->chunk_mutex
));
4970 chunk_offset
= find_next_chunk(fs_info
);
4971 return __btrfs_alloc_chunk(trans
, chunk_offset
, type
);
4974 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
4975 struct btrfs_fs_info
*fs_info
)
4978 u64 sys_chunk_offset
;
4982 chunk_offset
= find_next_chunk(fs_info
);
4983 alloc_profile
= btrfs_metadata_alloc_profile(fs_info
);
4984 ret
= __btrfs_alloc_chunk(trans
, chunk_offset
, alloc_profile
);
4988 sys_chunk_offset
= find_next_chunk(fs_info
);
4989 alloc_profile
= btrfs_system_alloc_profile(fs_info
);
4990 ret
= __btrfs_alloc_chunk(trans
, sys_chunk_offset
, alloc_profile
);
4994 static inline int btrfs_chunk_max_errors(struct map_lookup
*map
)
4998 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
4999 BTRFS_BLOCK_GROUP_RAID10
|
5000 BTRFS_BLOCK_GROUP_RAID5
|
5001 BTRFS_BLOCK_GROUP_DUP
)) {
5003 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
) {
5012 int btrfs_chunk_readonly(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
5014 struct extent_map
*em
;
5015 struct map_lookup
*map
;
5020 em
= get_chunk_map(fs_info
, chunk_offset
, 1);
5024 map
= em
->map_lookup
;
5025 for (i
= 0; i
< map
->num_stripes
; i
++) {
5026 if (map
->stripes
[i
].dev
->missing
) {
5031 if (!map
->stripes
[i
].dev
->writeable
) {
5038 * If the number of missing devices is larger than max errors,
5039 * we can not write the data into that chunk successfully, so
5042 if (miss_ndevs
> btrfs_chunk_max_errors(map
))
5045 free_extent_map(em
);
5049 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
5051 extent_map_tree_init(&tree
->map_tree
);
5054 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
5056 struct extent_map
*em
;
5059 write_lock(&tree
->map_tree
.lock
);
5060 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
5062 remove_extent_mapping(&tree
->map_tree
, em
);
5063 write_unlock(&tree
->map_tree
.lock
);
5067 free_extent_map(em
);
5068 /* once for the tree */
5069 free_extent_map(em
);
5073 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
5075 struct extent_map
*em
;
5076 struct map_lookup
*map
;
5079 em
= get_chunk_map(fs_info
, logical
, len
);
5082 * We could return errors for these cases, but that could get
5083 * ugly and we'd probably do the same thing which is just not do
5084 * anything else and exit, so return 1 so the callers don't try
5085 * to use other copies.
5089 map
= em
->map_lookup
;
5090 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
5091 ret
= map
->num_stripes
;
5092 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5093 ret
= map
->sub_stripes
;
5094 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID5
)
5096 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5100 free_extent_map(em
);
5102 btrfs_dev_replace_lock(&fs_info
->dev_replace
, 0);
5103 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
) &&
5104 fs_info
->dev_replace
.tgtdev
)
5106 btrfs_dev_replace_unlock(&fs_info
->dev_replace
, 0);
5111 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info
*fs_info
,
5114 struct extent_map
*em
;
5115 struct map_lookup
*map
;
5116 unsigned long len
= fs_info
->sectorsize
;
5118 em
= get_chunk_map(fs_info
, logical
, len
);
5120 if (!WARN_ON(IS_ERR(em
))) {
5121 map
= em
->map_lookup
;
5122 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5123 len
= map
->stripe_len
* nr_data_stripes(map
);
5124 free_extent_map(em
);
5129 int btrfs_is_parity_mirror(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
5131 struct extent_map
*em
;
5132 struct map_lookup
*map
;
5135 em
= get_chunk_map(fs_info
, logical
, len
);
5137 if(!WARN_ON(IS_ERR(em
))) {
5138 map
= em
->map_lookup
;
5139 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5141 free_extent_map(em
);
5146 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
5147 struct map_lookup
*map
, int first
, int num
,
5148 int optimal
, int dev_replace_is_ongoing
)
5152 struct btrfs_device
*srcdev
;
5154 if (dev_replace_is_ongoing
&&
5155 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
5156 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
5157 srcdev
= fs_info
->dev_replace
.srcdev
;
5162 * try to avoid the drive that is the source drive for a
5163 * dev-replace procedure, only choose it if no other non-missing
5164 * mirror is available
5166 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
5167 if (map
->stripes
[optimal
].dev
->bdev
&&
5168 (tolerance
|| map
->stripes
[optimal
].dev
!= srcdev
))
5170 for (i
= first
; i
< first
+ num
; i
++) {
5171 if (map
->stripes
[i
].dev
->bdev
&&
5172 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
5177 /* we couldn't find one that doesn't fail. Just return something
5178 * and the io error handling code will clean up eventually
5183 static inline int parity_smaller(u64 a
, u64 b
)
5188 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5189 static void sort_parity_stripes(struct btrfs_bio
*bbio
, int num_stripes
)
5191 struct btrfs_bio_stripe s
;
5198 for (i
= 0; i
< num_stripes
- 1; i
++) {
5199 if (parity_smaller(bbio
->raid_map
[i
],
5200 bbio
->raid_map
[i
+1])) {
5201 s
= bbio
->stripes
[i
];
5202 l
= bbio
->raid_map
[i
];
5203 bbio
->stripes
[i
] = bbio
->stripes
[i
+1];
5204 bbio
->raid_map
[i
] = bbio
->raid_map
[i
+1];
5205 bbio
->stripes
[i
+1] = s
;
5206 bbio
->raid_map
[i
+1] = l
;
5214 static struct btrfs_bio
*alloc_btrfs_bio(int total_stripes
, int real_stripes
)
5216 struct btrfs_bio
*bbio
= kzalloc(
5217 /* the size of the btrfs_bio */
5218 sizeof(struct btrfs_bio
) +
5219 /* plus the variable array for the stripes */
5220 sizeof(struct btrfs_bio_stripe
) * (total_stripes
) +
5221 /* plus the variable array for the tgt dev */
5222 sizeof(int) * (real_stripes
) +
5224 * plus the raid_map, which includes both the tgt dev
5227 sizeof(u64
) * (total_stripes
),
5228 GFP_NOFS
|__GFP_NOFAIL
);
5230 atomic_set(&bbio
->error
, 0);
5231 refcount_set(&bbio
->refs
, 1);
5236 void btrfs_get_bbio(struct btrfs_bio
*bbio
)
5238 WARN_ON(!refcount_read(&bbio
->refs
));
5239 refcount_inc(&bbio
->refs
);
5242 void btrfs_put_bbio(struct btrfs_bio
*bbio
)
5246 if (refcount_dec_and_test(&bbio
->refs
))
5250 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5252 * Please note that, discard won't be sent to target device of device
5255 static int __btrfs_map_block_for_discard(struct btrfs_fs_info
*fs_info
,
5256 u64 logical
, u64 length
,
5257 struct btrfs_bio
**bbio_ret
)
5259 struct extent_map
*em
;
5260 struct map_lookup
*map
;
5261 struct btrfs_bio
*bbio
;
5265 u64 stripe_end_offset
;
5272 u32 sub_stripes
= 0;
5273 u64 stripes_per_dev
= 0;
5274 u32 remaining_stripes
= 0;
5275 u32 last_stripe
= 0;
5279 /* discard always return a bbio */
5282 em
= get_chunk_map(fs_info
, logical
, length
);
5286 map
= em
->map_lookup
;
5287 /* we don't discard raid56 yet */
5288 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5293 offset
= logical
- em
->start
;
5294 length
= min_t(u64
, em
->len
- offset
, length
);
5296 stripe_len
= map
->stripe_len
;
5298 * stripe_nr counts the total number of stripes we have to stride
5299 * to get to this block
5301 stripe_nr
= div64_u64(offset
, stripe_len
);
5303 /* stripe_offset is the offset of this block in its stripe */
5304 stripe_offset
= offset
- stripe_nr
* stripe_len
;
5306 stripe_nr_end
= round_up(offset
+ length
, map
->stripe_len
);
5307 stripe_nr_end
= div64_u64(stripe_nr_end
, map
->stripe_len
);
5308 stripe_cnt
= stripe_nr_end
- stripe_nr
;
5309 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
5312 * after this, stripe_nr is the number of stripes on this
5313 * device we have to walk to find the data, and stripe_index is
5314 * the number of our device in the stripe array
5318 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5319 BTRFS_BLOCK_GROUP_RAID10
)) {
5320 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5323 sub_stripes
= map
->sub_stripes
;
5325 factor
= map
->num_stripes
/ sub_stripes
;
5326 num_stripes
= min_t(u64
, map
->num_stripes
,
5327 sub_stripes
* stripe_cnt
);
5328 stripe_nr
= div_u64_rem(stripe_nr
, factor
, &stripe_index
);
5329 stripe_index
*= sub_stripes
;
5330 stripes_per_dev
= div_u64_rem(stripe_cnt
, factor
,
5331 &remaining_stripes
);
5332 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
5333 last_stripe
*= sub_stripes
;
5334 } else if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
5335 BTRFS_BLOCK_GROUP_DUP
)) {
5336 num_stripes
= map
->num_stripes
;
5338 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5342 bbio
= alloc_btrfs_bio(num_stripes
, 0);
5348 for (i
= 0; i
< num_stripes
; i
++) {
5349 bbio
->stripes
[i
].physical
=
5350 map
->stripes
[stripe_index
].physical
+
5351 stripe_offset
+ stripe_nr
* map
->stripe_len
;
5352 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
5354 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5355 BTRFS_BLOCK_GROUP_RAID10
)) {
5356 bbio
->stripes
[i
].length
= stripes_per_dev
*
5359 if (i
/ sub_stripes
< remaining_stripes
)
5360 bbio
->stripes
[i
].length
+=
5364 * Special for the first stripe and
5367 * |-------|...|-------|
5371 if (i
< sub_stripes
)
5372 bbio
->stripes
[i
].length
-=
5375 if (stripe_index
>= last_stripe
&&
5376 stripe_index
<= (last_stripe
+
5378 bbio
->stripes
[i
].length
-=
5381 if (i
== sub_stripes
- 1)
5384 bbio
->stripes
[i
].length
= length
;
5388 if (stripe_index
== map
->num_stripes
) {
5395 bbio
->map_type
= map
->type
;
5396 bbio
->num_stripes
= num_stripes
;
5398 free_extent_map(em
);
5403 * In dev-replace case, for repair case (that's the only case where the mirror
5404 * is selected explicitly when calling btrfs_map_block), blocks left of the
5405 * left cursor can also be read from the target drive.
5407 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5409 * For READ, it also needs to be supported using the same mirror number.
5411 * If the requested block is not left of the left cursor, EIO is returned. This
5412 * can happen because btrfs_num_copies() returns one more in the dev-replace
5415 static int get_extra_mirror_from_replace(struct btrfs_fs_info
*fs_info
,
5416 u64 logical
, u64 length
,
5417 u64 srcdev_devid
, int *mirror_num
,
5420 struct btrfs_bio
*bbio
= NULL
;
5422 int index_srcdev
= 0;
5424 u64 physical_of_found
= 0;
5428 ret
= __btrfs_map_block(fs_info
, BTRFS_MAP_GET_READ_MIRRORS
,
5429 logical
, &length
, &bbio
, 0, 0);
5431 ASSERT(bbio
== NULL
);
5435 num_stripes
= bbio
->num_stripes
;
5436 if (*mirror_num
> num_stripes
) {
5438 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5439 * that means that the requested area is not left of the left
5442 btrfs_put_bbio(bbio
);
5447 * process the rest of the function using the mirror_num of the source
5448 * drive. Therefore look it up first. At the end, patch the device
5449 * pointer to the one of the target drive.
5451 for (i
= 0; i
< num_stripes
; i
++) {
5452 if (bbio
->stripes
[i
].dev
->devid
!= srcdev_devid
)
5456 * In case of DUP, in order to keep it simple, only add the
5457 * mirror with the lowest physical address
5460 physical_of_found
<= bbio
->stripes
[i
].physical
)
5465 physical_of_found
= bbio
->stripes
[i
].physical
;
5468 btrfs_put_bbio(bbio
);
5474 *mirror_num
= index_srcdev
+ 1;
5475 *physical
= physical_of_found
;
5479 static void handle_ops_on_dev_replace(enum btrfs_map_op op
,
5480 struct btrfs_bio
**bbio_ret
,
5481 struct btrfs_dev_replace
*dev_replace
,
5482 int *num_stripes_ret
, int *max_errors_ret
)
5484 struct btrfs_bio
*bbio
= *bbio_ret
;
5485 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5486 int tgtdev_indexes
= 0;
5487 int num_stripes
= *num_stripes_ret
;
5488 int max_errors
= *max_errors_ret
;
5491 if (op
== BTRFS_MAP_WRITE
) {
5492 int index_where_to_add
;
5495 * duplicate the write operations while the dev replace
5496 * procedure is running. Since the copying of the old disk to
5497 * the new disk takes place at run time while the filesystem is
5498 * mounted writable, the regular write operations to the old
5499 * disk have to be duplicated to go to the new disk as well.
5501 * Note that device->missing is handled by the caller, and that
5502 * the write to the old disk is already set up in the stripes
5505 index_where_to_add
= num_stripes
;
5506 for (i
= 0; i
< num_stripes
; i
++) {
5507 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5508 /* write to new disk, too */
5509 struct btrfs_bio_stripe
*new =
5510 bbio
->stripes
+ index_where_to_add
;
5511 struct btrfs_bio_stripe
*old
=
5514 new->physical
= old
->physical
;
5515 new->length
= old
->length
;
5516 new->dev
= dev_replace
->tgtdev
;
5517 bbio
->tgtdev_map
[i
] = index_where_to_add
;
5518 index_where_to_add
++;
5523 num_stripes
= index_where_to_add
;
5524 } else if (op
== BTRFS_MAP_GET_READ_MIRRORS
) {
5525 int index_srcdev
= 0;
5527 u64 physical_of_found
= 0;
5530 * During the dev-replace procedure, the target drive can also
5531 * be used to read data in case it is needed to repair a corrupt
5532 * block elsewhere. This is possible if the requested area is
5533 * left of the left cursor. In this area, the target drive is a
5534 * full copy of the source drive.
5536 for (i
= 0; i
< num_stripes
; i
++) {
5537 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5539 * In case of DUP, in order to keep it simple,
5540 * only add the mirror with the lowest physical
5544 physical_of_found
<=
5545 bbio
->stripes
[i
].physical
)
5549 physical_of_found
= bbio
->stripes
[i
].physical
;
5553 struct btrfs_bio_stripe
*tgtdev_stripe
=
5554 bbio
->stripes
+ num_stripes
;
5556 tgtdev_stripe
->physical
= physical_of_found
;
5557 tgtdev_stripe
->length
=
5558 bbio
->stripes
[index_srcdev
].length
;
5559 tgtdev_stripe
->dev
= dev_replace
->tgtdev
;
5560 bbio
->tgtdev_map
[index_srcdev
] = num_stripes
;
5567 *num_stripes_ret
= num_stripes
;
5568 *max_errors_ret
= max_errors
;
5569 bbio
->num_tgtdevs
= tgtdev_indexes
;
5573 static bool need_full_stripe(enum btrfs_map_op op
)
5575 return (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
);
5578 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
,
5579 enum btrfs_map_op op
,
5580 u64 logical
, u64
*length
,
5581 struct btrfs_bio
**bbio_ret
,
5582 int mirror_num
, int need_raid_map
)
5584 struct extent_map
*em
;
5585 struct map_lookup
*map
;
5595 int tgtdev_indexes
= 0;
5596 struct btrfs_bio
*bbio
= NULL
;
5597 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
5598 int dev_replace_is_ongoing
= 0;
5599 int num_alloc_stripes
;
5600 int patch_the_first_stripe_for_dev_replace
= 0;
5601 u64 physical_to_patch_in_first_stripe
= 0;
5602 u64 raid56_full_stripe_start
= (u64
)-1;
5604 if (op
== BTRFS_MAP_DISCARD
)
5605 return __btrfs_map_block_for_discard(fs_info
, logical
,
5608 em
= get_chunk_map(fs_info
, logical
, *length
);
5612 map
= em
->map_lookup
;
5613 offset
= logical
- em
->start
;
5615 stripe_len
= map
->stripe_len
;
5618 * stripe_nr counts the total number of stripes we have to stride
5619 * to get to this block
5621 stripe_nr
= div64_u64(stripe_nr
, stripe_len
);
5623 stripe_offset
= stripe_nr
* stripe_len
;
5624 if (offset
< stripe_offset
) {
5626 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5627 stripe_offset
, offset
, em
->start
, logical
,
5629 free_extent_map(em
);
5633 /* stripe_offset is the offset of this block in its stripe*/
5634 stripe_offset
= offset
- stripe_offset
;
5636 /* if we're here for raid56, we need to know the stripe aligned start */
5637 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5638 unsigned long full_stripe_len
= stripe_len
* nr_data_stripes(map
);
5639 raid56_full_stripe_start
= offset
;
5641 /* allow a write of a full stripe, but make sure we don't
5642 * allow straddling of stripes
5644 raid56_full_stripe_start
= div64_u64(raid56_full_stripe_start
,
5646 raid56_full_stripe_start
*= full_stripe_len
;
5649 if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
5651 /* For writes to RAID[56], allow a full stripeset across all disks.
5652 For other RAID types and for RAID[56] reads, just allow a single
5653 stripe (on a single disk). */
5654 if ((map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
5655 (op
== BTRFS_MAP_WRITE
)) {
5656 max_len
= stripe_len
* nr_data_stripes(map
) -
5657 (offset
- raid56_full_stripe_start
);
5659 /* we limit the length of each bio to what fits in a stripe */
5660 max_len
= stripe_len
- stripe_offset
;
5662 *length
= min_t(u64
, em
->len
- offset
, max_len
);
5664 *length
= em
->len
- offset
;
5667 /* This is for when we're called from btrfs_merge_bio_hook() and all
5668 it cares about is the length */
5672 btrfs_dev_replace_lock(dev_replace
, 0);
5673 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
5674 if (!dev_replace_is_ongoing
)
5675 btrfs_dev_replace_unlock(dev_replace
, 0);
5677 btrfs_dev_replace_set_lock_blocking(dev_replace
);
5679 if (dev_replace_is_ongoing
&& mirror_num
== map
->num_stripes
+ 1 &&
5680 !need_full_stripe(op
) && dev_replace
->tgtdev
!= NULL
) {
5681 ret
= get_extra_mirror_from_replace(fs_info
, logical
, *length
,
5682 dev_replace
->srcdev
->devid
,
5684 &physical_to_patch_in_first_stripe
);
5688 patch_the_first_stripe_for_dev_replace
= 1;
5689 } else if (mirror_num
> map
->num_stripes
) {
5695 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5696 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5698 if (op
!= BTRFS_MAP_WRITE
&& op
!= BTRFS_MAP_GET_READ_MIRRORS
)
5700 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
5701 if (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
)
5702 num_stripes
= map
->num_stripes
;
5703 else if (mirror_num
)
5704 stripe_index
= mirror_num
- 1;
5706 stripe_index
= find_live_mirror(fs_info
, map
, 0,
5708 current
->pid
% map
->num_stripes
,
5709 dev_replace_is_ongoing
);
5710 mirror_num
= stripe_index
+ 1;
5713 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
5714 if (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
) {
5715 num_stripes
= map
->num_stripes
;
5716 } else if (mirror_num
) {
5717 stripe_index
= mirror_num
- 1;
5722 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5723 u32 factor
= map
->num_stripes
/ map
->sub_stripes
;
5725 stripe_nr
= div_u64_rem(stripe_nr
, factor
, &stripe_index
);
5726 stripe_index
*= map
->sub_stripes
;
5728 if (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
)
5729 num_stripes
= map
->sub_stripes
;
5730 else if (mirror_num
)
5731 stripe_index
+= mirror_num
- 1;
5733 int old_stripe_index
= stripe_index
;
5734 stripe_index
= find_live_mirror(fs_info
, map
,
5736 map
->sub_stripes
, stripe_index
+
5737 current
->pid
% map
->sub_stripes
,
5738 dev_replace_is_ongoing
);
5739 mirror_num
= stripe_index
- old_stripe_index
+ 1;
5742 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5743 if (need_raid_map
&&
5744 (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
||
5746 /* push stripe_nr back to the start of the full stripe */
5747 stripe_nr
= div64_u64(raid56_full_stripe_start
,
5748 stripe_len
* nr_data_stripes(map
));
5750 /* RAID[56] write or recovery. Return all stripes */
5751 num_stripes
= map
->num_stripes
;
5752 max_errors
= nr_parity_stripes(map
);
5754 *length
= map
->stripe_len
;
5759 * Mirror #0 or #1 means the original data block.
5760 * Mirror #2 is RAID5 parity block.
5761 * Mirror #3 is RAID6 Q block.
5763 stripe_nr
= div_u64_rem(stripe_nr
,
5764 nr_data_stripes(map
), &stripe_index
);
5766 stripe_index
= nr_data_stripes(map
) +
5769 /* We distribute the parity blocks across stripes */
5770 div_u64_rem(stripe_nr
+ stripe_index
, map
->num_stripes
,
5772 if ((op
!= BTRFS_MAP_WRITE
&&
5773 op
!= BTRFS_MAP_GET_READ_MIRRORS
) &&
5779 * after this, stripe_nr is the number of stripes on this
5780 * device we have to walk to find the data, and stripe_index is
5781 * the number of our device in the stripe array
5783 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5785 mirror_num
= stripe_index
+ 1;
5787 if (stripe_index
>= map
->num_stripes
) {
5789 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5790 stripe_index
, map
->num_stripes
);
5795 num_alloc_stripes
= num_stripes
;
5796 if (dev_replace_is_ongoing
&& dev_replace
->tgtdev
!= NULL
) {
5797 if (op
== BTRFS_MAP_WRITE
)
5798 num_alloc_stripes
<<= 1;
5799 if (op
== BTRFS_MAP_GET_READ_MIRRORS
)
5800 num_alloc_stripes
++;
5801 tgtdev_indexes
= num_stripes
;
5804 bbio
= alloc_btrfs_bio(num_alloc_stripes
, tgtdev_indexes
);
5809 if (dev_replace_is_ongoing
&& dev_replace
->tgtdev
!= NULL
)
5810 bbio
->tgtdev_map
= (int *)(bbio
->stripes
+ num_alloc_stripes
);
5812 /* build raid_map */
5813 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
&& need_raid_map
&&
5814 (need_full_stripe(op
) || mirror_num
> 1)) {
5818 bbio
->raid_map
= (u64
*)((void *)bbio
->stripes
+
5819 sizeof(struct btrfs_bio_stripe
) *
5821 sizeof(int) * tgtdev_indexes
);
5823 /* Work out the disk rotation on this stripe-set */
5824 div_u64_rem(stripe_nr
, num_stripes
, &rot
);
5826 /* Fill in the logical address of each stripe */
5827 tmp
= stripe_nr
* nr_data_stripes(map
);
5828 for (i
= 0; i
< nr_data_stripes(map
); i
++)
5829 bbio
->raid_map
[(i
+rot
) % num_stripes
] =
5830 em
->start
+ (tmp
+ i
) * map
->stripe_len
;
5832 bbio
->raid_map
[(i
+rot
) % map
->num_stripes
] = RAID5_P_STRIPE
;
5833 if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5834 bbio
->raid_map
[(i
+rot
+1) % num_stripes
] =
5839 for (i
= 0; i
< num_stripes
; i
++) {
5840 bbio
->stripes
[i
].physical
=
5841 map
->stripes
[stripe_index
].physical
+
5843 stripe_nr
* map
->stripe_len
;
5844 bbio
->stripes
[i
].dev
=
5845 map
->stripes
[stripe_index
].dev
;
5849 if (need_full_stripe(op
))
5850 max_errors
= btrfs_chunk_max_errors(map
);
5853 sort_parity_stripes(bbio
, num_stripes
);
5855 if (dev_replace_is_ongoing
&& dev_replace
->tgtdev
!= NULL
&&
5856 need_full_stripe(op
)) {
5857 handle_ops_on_dev_replace(op
, &bbio
, dev_replace
, &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
;
5868 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5869 * mirror_num == num_stripes + 1 && dev_replace target drive is
5870 * available as a mirror
5872 if (patch_the_first_stripe_for_dev_replace
&& num_stripes
> 0) {
5873 WARN_ON(num_stripes
> 1);
5874 bbio
->stripes
[0].dev
= dev_replace
->tgtdev
;
5875 bbio
->stripes
[0].physical
= physical_to_patch_in_first_stripe
;
5876 bbio
->mirror_num
= map
->num_stripes
+ 1;
5879 if (dev_replace_is_ongoing
) {
5880 btrfs_dev_replace_clear_lock_blocking(dev_replace
);
5881 btrfs_dev_replace_unlock(dev_replace
, 0);
5883 free_extent_map(em
);
5887 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, enum btrfs_map_op op
,
5888 u64 logical
, u64
*length
,
5889 struct btrfs_bio
**bbio_ret
, int mirror_num
)
5891 return __btrfs_map_block(fs_info
, op
, logical
, length
, bbio_ret
,
5895 /* For Scrub/replace */
5896 int btrfs_map_sblock(struct btrfs_fs_info
*fs_info
, enum btrfs_map_op op
,
5897 u64 logical
, u64
*length
,
5898 struct btrfs_bio
**bbio_ret
)
5900 return __btrfs_map_block(fs_info
, op
, logical
, length
, bbio_ret
, 0, 1);
5903 int btrfs_rmap_block(struct btrfs_fs_info
*fs_info
,
5904 u64 chunk_start
, u64 physical
, u64 devid
,
5905 u64
**logical
, int *naddrs
, int *stripe_len
)
5907 struct extent_map
*em
;
5908 struct map_lookup
*map
;
5916 em
= get_chunk_map(fs_info
, chunk_start
, 1);
5920 map
= em
->map_lookup
;
5922 rmap_len
= map
->stripe_len
;
5924 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5925 length
= div_u64(length
, map
->num_stripes
/ map
->sub_stripes
);
5926 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5927 length
= div_u64(length
, map
->num_stripes
);
5928 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5929 length
= div_u64(length
, nr_data_stripes(map
));
5930 rmap_len
= map
->stripe_len
* nr_data_stripes(map
);
5933 buf
= kcalloc(map
->num_stripes
, sizeof(u64
), GFP_NOFS
);
5934 BUG_ON(!buf
); /* -ENOMEM */
5936 for (i
= 0; i
< map
->num_stripes
; i
++) {
5937 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
5939 if (map
->stripes
[i
].physical
> physical
||
5940 map
->stripes
[i
].physical
+ length
<= physical
)
5943 stripe_nr
= physical
- map
->stripes
[i
].physical
;
5944 stripe_nr
= div64_u64(stripe_nr
, map
->stripe_len
);
5946 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5947 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5948 stripe_nr
= div_u64(stripe_nr
, map
->sub_stripes
);
5949 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5950 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5951 } /* else if RAID[56], multiply by nr_data_stripes().
5952 * Alternatively, just use rmap_len below instead of
5953 * map->stripe_len */
5955 bytenr
= chunk_start
+ stripe_nr
* rmap_len
;
5956 WARN_ON(nr
>= map
->num_stripes
);
5957 for (j
= 0; j
< nr
; j
++) {
5958 if (buf
[j
] == bytenr
)
5962 WARN_ON(nr
>= map
->num_stripes
);
5969 *stripe_len
= rmap_len
;
5971 free_extent_map(em
);
5975 static inline void btrfs_end_bbio(struct btrfs_bio
*bbio
, struct bio
*bio
)
5977 bio
->bi_private
= bbio
->private;
5978 bio
->bi_end_io
= bbio
->end_io
;
5981 btrfs_put_bbio(bbio
);
5984 static void btrfs_end_bio(struct bio
*bio
)
5986 struct btrfs_bio
*bbio
= bio
->bi_private
;
5987 int is_orig_bio
= 0;
5989 if (bio
->bi_status
) {
5990 atomic_inc(&bbio
->error
);
5991 if (bio
->bi_status
== BLK_STS_IOERR
||
5992 bio
->bi_status
== BLK_STS_TARGET
) {
5993 unsigned int stripe_index
=
5994 btrfs_io_bio(bio
)->stripe_index
;
5995 struct btrfs_device
*dev
;
5997 BUG_ON(stripe_index
>= bbio
->num_stripes
);
5998 dev
= bbio
->stripes
[stripe_index
].dev
;
6000 if (bio_op(bio
) == REQ_OP_WRITE
)
6001 btrfs_dev_stat_inc(dev
,
6002 BTRFS_DEV_STAT_WRITE_ERRS
);
6004 btrfs_dev_stat_inc(dev
,
6005 BTRFS_DEV_STAT_READ_ERRS
);
6006 if (bio
->bi_opf
& REQ_PREFLUSH
)
6007 btrfs_dev_stat_inc(dev
,
6008 BTRFS_DEV_STAT_FLUSH_ERRS
);
6009 btrfs_dev_stat_print_on_error(dev
);
6014 if (bio
== bbio
->orig_bio
)
6017 btrfs_bio_counter_dec(bbio
->fs_info
);
6019 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
6022 bio
= bbio
->orig_bio
;
6025 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
6026 /* only send an error to the higher layers if it is
6027 * beyond the tolerance of the btrfs bio
6029 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
6030 bio
->bi_status
= BLK_STS_IOERR
;
6033 * this bio is actually up to date, we didn't
6034 * go over the max number of errors
6039 btrfs_end_bbio(bbio
, bio
);
6040 } else if (!is_orig_bio
) {
6046 * see run_scheduled_bios for a description of why bios are collected for
6049 * This will add one bio to the pending list for a device and make sure
6050 * the work struct is scheduled.
6052 static noinline
void btrfs_schedule_bio(struct btrfs_device
*device
,
6055 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
6056 int should_queue
= 1;
6057 struct btrfs_pending_bios
*pending_bios
;
6059 if (device
->missing
|| !device
->bdev
) {
6064 /* don't bother with additional async steps for reads, right now */
6065 if (bio_op(bio
) == REQ_OP_READ
) {
6067 btrfsic_submit_bio(bio
);
6073 * nr_async_bios allows us to reliably return congestion to the
6074 * higher layers. Otherwise, the async bio makes it appear we have
6075 * made progress against dirty pages when we've really just put it
6076 * on a queue for later
6078 atomic_inc(&fs_info
->nr_async_bios
);
6079 WARN_ON(bio
->bi_next
);
6080 bio
->bi_next
= NULL
;
6082 spin_lock(&device
->io_lock
);
6083 if (op_is_sync(bio
->bi_opf
))
6084 pending_bios
= &device
->pending_sync_bios
;
6086 pending_bios
= &device
->pending_bios
;
6088 if (pending_bios
->tail
)
6089 pending_bios
->tail
->bi_next
= bio
;
6091 pending_bios
->tail
= bio
;
6092 if (!pending_bios
->head
)
6093 pending_bios
->head
= bio
;
6094 if (device
->running_pending
)
6097 spin_unlock(&device
->io_lock
);
6100 btrfs_queue_work(fs_info
->submit_workers
, &device
->work
);
6103 static void submit_stripe_bio(struct btrfs_bio
*bbio
, struct bio
*bio
,
6104 u64 physical
, int dev_nr
, int async
)
6106 struct btrfs_device
*dev
= bbio
->stripes
[dev_nr
].dev
;
6107 struct btrfs_fs_info
*fs_info
= bbio
->fs_info
;
6109 bio
->bi_private
= bbio
;
6110 btrfs_io_bio(bio
)->stripe_index
= dev_nr
;
6111 bio
->bi_end_io
= btrfs_end_bio
;
6112 bio
->bi_iter
.bi_sector
= physical
>> 9;
6115 struct rcu_string
*name
;
6118 name
= rcu_dereference(dev
->name
);
6119 btrfs_debug(fs_info
,
6120 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6121 bio_op(bio
), bio
->bi_opf
,
6122 (u64
)bio
->bi_iter
.bi_sector
,
6123 (u_long
)dev
->bdev
->bd_dev
, name
->str
, dev
->devid
,
6124 bio
->bi_iter
.bi_size
);
6128 bio_set_dev(bio
, dev
->bdev
);
6130 btrfs_bio_counter_inc_noblocked(fs_info
);
6133 btrfs_schedule_bio(dev
, bio
);
6135 btrfsic_submit_bio(bio
);
6138 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
6140 atomic_inc(&bbio
->error
);
6141 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
6142 /* Should be the original bio. */
6143 WARN_ON(bio
!= bbio
->orig_bio
);
6145 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
6146 bio
->bi_iter
.bi_sector
= logical
>> 9;
6147 bio
->bi_status
= BLK_STS_IOERR
;
6148 btrfs_end_bbio(bbio
, bio
);
6152 blk_status_t
btrfs_map_bio(struct btrfs_fs_info
*fs_info
, struct bio
*bio
,
6153 int mirror_num
, int async_submit
)
6155 struct btrfs_device
*dev
;
6156 struct bio
*first_bio
= bio
;
6157 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
6163 struct btrfs_bio
*bbio
= NULL
;
6165 length
= bio
->bi_iter
.bi_size
;
6166 map_length
= length
;
6168 btrfs_bio_counter_inc_blocked(fs_info
);
6169 ret
= __btrfs_map_block(fs_info
, btrfs_op(bio
), logical
,
6170 &map_length
, &bbio
, mirror_num
, 1);
6172 btrfs_bio_counter_dec(fs_info
);
6173 return errno_to_blk_status(ret
);
6176 total_devs
= bbio
->num_stripes
;
6177 bbio
->orig_bio
= first_bio
;
6178 bbio
->private = first_bio
->bi_private
;
6179 bbio
->end_io
= first_bio
->bi_end_io
;
6180 bbio
->fs_info
= fs_info
;
6181 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
6183 if ((bbio
->map_type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
6184 ((bio_op(bio
) == REQ_OP_WRITE
) || (mirror_num
> 1))) {
6185 /* In this case, map_length has been set to the length of
6186 a single stripe; not the whole write */
6187 if (bio_op(bio
) == REQ_OP_WRITE
) {
6188 ret
= raid56_parity_write(fs_info
, bio
, bbio
,
6191 ret
= raid56_parity_recover(fs_info
, bio
, bbio
,
6192 map_length
, mirror_num
, 1);
6195 btrfs_bio_counter_dec(fs_info
);
6196 return errno_to_blk_status(ret
);
6199 if (map_length
< length
) {
6201 "mapping failed logical %llu bio len %llu len %llu",
6202 logical
, length
, map_length
);
6206 for (dev_nr
= 0; dev_nr
< total_devs
; dev_nr
++) {
6207 dev
= bbio
->stripes
[dev_nr
].dev
;
6208 if (!dev
|| !dev
->bdev
||
6209 (bio_op(first_bio
) == REQ_OP_WRITE
&& !dev
->writeable
)) {
6210 bbio_error(bbio
, first_bio
, logical
);
6214 if (dev_nr
< total_devs
- 1)
6215 bio
= btrfs_bio_clone(first_bio
);
6219 submit_stripe_bio(bbio
, bio
, bbio
->stripes
[dev_nr
].physical
,
6220 dev_nr
, async_submit
);
6222 btrfs_bio_counter_dec(fs_info
);
6226 struct btrfs_device
*btrfs_find_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
6229 struct btrfs_device
*device
;
6230 struct btrfs_fs_devices
*cur_devices
;
6232 cur_devices
= fs_info
->fs_devices
;
6233 while (cur_devices
) {
6235 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_FSID_SIZE
)) {
6236 device
= find_device(cur_devices
, devid
, uuid
);
6240 cur_devices
= cur_devices
->seed
;
6245 static struct btrfs_device
*add_missing_dev(struct btrfs_fs_devices
*fs_devices
,
6246 u64 devid
, u8
*dev_uuid
)
6248 struct btrfs_device
*device
;
6250 device
= btrfs_alloc_device(NULL
, &devid
, dev_uuid
);
6254 list_add(&device
->dev_list
, &fs_devices
->devices
);
6255 device
->fs_devices
= fs_devices
;
6256 fs_devices
->num_devices
++;
6258 device
->missing
= 1;
6259 fs_devices
->missing_devices
++;
6265 * btrfs_alloc_device - allocate struct btrfs_device
6266 * @fs_info: used only for generating a new devid, can be NULL if
6267 * devid is provided (i.e. @devid != NULL).
6268 * @devid: a pointer to devid for this device. If NULL a new devid
6270 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6273 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6274 * on error. Returned struct is not linked onto any lists and can be
6275 * destroyed with kfree() right away.
6277 struct btrfs_device
*btrfs_alloc_device(struct btrfs_fs_info
*fs_info
,
6281 struct btrfs_device
*dev
;
6284 if (WARN_ON(!devid
&& !fs_info
))
6285 return ERR_PTR(-EINVAL
);
6287 dev
= __alloc_device();
6296 ret
= find_next_devid(fs_info
, &tmp
);
6299 return ERR_PTR(ret
);
6305 memcpy(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
);
6307 generate_random_uuid(dev
->uuid
);
6309 btrfs_init_work(&dev
->work
, btrfs_submit_helper
,
6310 pending_bios_fn
, NULL
, NULL
);
6315 /* Return -EIO if any error, otherwise return 0. */
6316 static int btrfs_check_chunk_valid(struct btrfs_fs_info
*fs_info
,
6317 struct extent_buffer
*leaf
,
6318 struct btrfs_chunk
*chunk
, u64 logical
)
6326 length
= btrfs_chunk_length(leaf
, chunk
);
6327 stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6328 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6329 sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6330 type
= btrfs_chunk_type(leaf
, chunk
);
6333 btrfs_err(fs_info
, "invalid chunk num_stripes: %u",
6337 if (!IS_ALIGNED(logical
, fs_info
->sectorsize
)) {
6338 btrfs_err(fs_info
, "invalid chunk logical %llu", logical
);
6341 if (btrfs_chunk_sector_size(leaf
, chunk
) != fs_info
->sectorsize
) {
6342 btrfs_err(fs_info
, "invalid chunk sectorsize %u",
6343 btrfs_chunk_sector_size(leaf
, chunk
));
6346 if (!length
|| !IS_ALIGNED(length
, fs_info
->sectorsize
)) {
6347 btrfs_err(fs_info
, "invalid chunk length %llu", length
);
6350 if (!is_power_of_2(stripe_len
) || stripe_len
!= BTRFS_STRIPE_LEN
) {
6351 btrfs_err(fs_info
, "invalid chunk stripe length: %llu",
6355 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK
| BTRFS_BLOCK_GROUP_PROFILE_MASK
) &
6357 btrfs_err(fs_info
, "unrecognized chunk type: %llu",
6358 ~(BTRFS_BLOCK_GROUP_TYPE_MASK
|
6359 BTRFS_BLOCK_GROUP_PROFILE_MASK
) &
6360 btrfs_chunk_type(leaf
, chunk
));
6363 if ((type
& BTRFS_BLOCK_GROUP_RAID10
&& sub_stripes
!= 2) ||
6364 (type
& BTRFS_BLOCK_GROUP_RAID1
&& num_stripes
< 1) ||
6365 (type
& BTRFS_BLOCK_GROUP_RAID5
&& num_stripes
< 2) ||
6366 (type
& BTRFS_BLOCK_GROUP_RAID6
&& num_stripes
< 3) ||
6367 (type
& BTRFS_BLOCK_GROUP_DUP
&& num_stripes
> 2) ||
6368 ((type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) == 0 &&
6369 num_stripes
!= 1)) {
6371 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6372 num_stripes
, sub_stripes
,
6373 type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
);
6380 static int read_one_chunk(struct btrfs_fs_info
*fs_info
, struct btrfs_key
*key
,
6381 struct extent_buffer
*leaf
,
6382 struct btrfs_chunk
*chunk
)
6384 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
6385 struct map_lookup
*map
;
6386 struct extent_map
*em
;
6390 u8 uuid
[BTRFS_UUID_SIZE
];
6395 logical
= key
->offset
;
6396 length
= btrfs_chunk_length(leaf
, chunk
);
6397 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6399 ret
= btrfs_check_chunk_valid(fs_info
, leaf
, chunk
, logical
);
6403 read_lock(&map_tree
->map_tree
.lock
);
6404 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
6405 read_unlock(&map_tree
->map_tree
.lock
);
6407 /* already mapped? */
6408 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
6409 free_extent_map(em
);
6412 free_extent_map(em
);
6415 em
= alloc_extent_map();
6418 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
6420 free_extent_map(em
);
6424 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
6425 em
->map_lookup
= map
;
6426 em
->start
= logical
;
6429 em
->block_start
= 0;
6430 em
->block_len
= em
->len
;
6432 map
->num_stripes
= num_stripes
;
6433 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
6434 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
6435 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6436 map
->type
= btrfs_chunk_type(leaf
, chunk
);
6437 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6438 for (i
= 0; i
< num_stripes
; i
++) {
6439 map
->stripes
[i
].physical
=
6440 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
6441 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
6442 read_extent_buffer(leaf
, uuid
, (unsigned long)
6443 btrfs_stripe_dev_uuid_nr(chunk
, i
),
6445 map
->stripes
[i
].dev
= btrfs_find_device(fs_info
, devid
,
6447 if (!map
->stripes
[i
].dev
&&
6448 !btrfs_test_opt(fs_info
, DEGRADED
)) {
6449 free_extent_map(em
);
6450 btrfs_report_missing_device(fs_info
, devid
, uuid
);
6453 if (!map
->stripes
[i
].dev
) {
6454 map
->stripes
[i
].dev
=
6455 add_missing_dev(fs_info
->fs_devices
, devid
,
6457 if (!map
->stripes
[i
].dev
) {
6458 free_extent_map(em
);
6461 btrfs_report_missing_device(fs_info
, devid
, uuid
);
6463 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
6466 write_lock(&map_tree
->map_tree
.lock
);
6467 ret
= add_extent_mapping(&map_tree
->map_tree
, em
, 0);
6468 write_unlock(&map_tree
->map_tree
.lock
);
6469 BUG_ON(ret
); /* Tree corruption */
6470 free_extent_map(em
);
6475 static void fill_device_from_item(struct extent_buffer
*leaf
,
6476 struct btrfs_dev_item
*dev_item
,
6477 struct btrfs_device
*device
)
6481 device
->devid
= btrfs_device_id(leaf
, dev_item
);
6482 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
6483 device
->total_bytes
= device
->disk_total_bytes
;
6484 device
->commit_total_bytes
= device
->disk_total_bytes
;
6485 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
6486 device
->commit_bytes_used
= device
->bytes_used
;
6487 device
->type
= btrfs_device_type(leaf
, dev_item
);
6488 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
6489 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
6490 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
6491 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
6492 device
->is_tgtdev_for_dev_replace
= 0;
6494 ptr
= btrfs_device_uuid(dev_item
);
6495 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
6498 static struct btrfs_fs_devices
*open_seed_devices(struct btrfs_fs_info
*fs_info
,
6501 struct btrfs_fs_devices
*fs_devices
;
6504 BUG_ON(!mutex_is_locked(&uuid_mutex
));
6507 fs_devices
= fs_info
->fs_devices
->seed
;
6508 while (fs_devices
) {
6509 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_FSID_SIZE
))
6512 fs_devices
= fs_devices
->seed
;
6515 fs_devices
= find_fsid(fsid
);
6517 if (!btrfs_test_opt(fs_info
, DEGRADED
))
6518 return ERR_PTR(-ENOENT
);
6520 fs_devices
= alloc_fs_devices(fsid
);
6521 if (IS_ERR(fs_devices
))
6524 fs_devices
->seeding
= 1;
6525 fs_devices
->opened
= 1;
6529 fs_devices
= clone_fs_devices(fs_devices
);
6530 if (IS_ERR(fs_devices
))
6533 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
6534 fs_info
->bdev_holder
);
6536 free_fs_devices(fs_devices
);
6537 fs_devices
= ERR_PTR(ret
);
6541 if (!fs_devices
->seeding
) {
6542 __btrfs_close_devices(fs_devices
);
6543 free_fs_devices(fs_devices
);
6544 fs_devices
= ERR_PTR(-EINVAL
);
6548 fs_devices
->seed
= fs_info
->fs_devices
->seed
;
6549 fs_info
->fs_devices
->seed
= fs_devices
;
6554 static int read_one_dev(struct btrfs_fs_info
*fs_info
,
6555 struct extent_buffer
*leaf
,
6556 struct btrfs_dev_item
*dev_item
)
6558 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6559 struct btrfs_device
*device
;
6562 u8 fs_uuid
[BTRFS_FSID_SIZE
];
6563 u8 dev_uuid
[BTRFS_UUID_SIZE
];
6565 devid
= btrfs_device_id(leaf
, dev_item
);
6566 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
6568 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
6571 if (memcmp(fs_uuid
, fs_info
->fsid
, BTRFS_FSID_SIZE
)) {
6572 fs_devices
= open_seed_devices(fs_info
, fs_uuid
);
6573 if (IS_ERR(fs_devices
))
6574 return PTR_ERR(fs_devices
);
6577 device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, fs_uuid
);
6579 if (!btrfs_test_opt(fs_info
, DEGRADED
)) {
6580 btrfs_report_missing_device(fs_info
, devid
, dev_uuid
);
6584 device
= add_missing_dev(fs_devices
, devid
, dev_uuid
);
6587 btrfs_report_missing_device(fs_info
, devid
, dev_uuid
);
6589 if (!device
->bdev
) {
6590 btrfs_report_missing_device(fs_info
, devid
, dev_uuid
);
6591 if (!btrfs_test_opt(fs_info
, DEGRADED
))
6595 if(!device
->bdev
&& !device
->missing
) {
6597 * this happens when a device that was properly setup
6598 * in the device info lists suddenly goes bad.
6599 * device->bdev is NULL, and so we have to set
6600 * device->missing to one here
6602 device
->fs_devices
->missing_devices
++;
6603 device
->missing
= 1;
6606 /* Move the device to its own fs_devices */
6607 if (device
->fs_devices
!= fs_devices
) {
6608 ASSERT(device
->missing
);
6610 list_move(&device
->dev_list
, &fs_devices
->devices
);
6611 device
->fs_devices
->num_devices
--;
6612 fs_devices
->num_devices
++;
6614 device
->fs_devices
->missing_devices
--;
6615 fs_devices
->missing_devices
++;
6617 device
->fs_devices
= fs_devices
;
6621 if (device
->fs_devices
!= fs_info
->fs_devices
) {
6622 BUG_ON(device
->writeable
);
6623 if (device
->generation
!=
6624 btrfs_device_generation(leaf
, dev_item
))
6628 fill_device_from_item(leaf
, dev_item
, device
);
6629 device
->in_fs_metadata
= 1;
6630 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
6631 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
6632 atomic64_add(device
->total_bytes
- device
->bytes_used
,
6633 &fs_info
->free_chunk_space
);
6639 int btrfs_read_sys_array(struct btrfs_fs_info
*fs_info
)
6641 struct btrfs_root
*root
= fs_info
->tree_root
;
6642 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
6643 struct extent_buffer
*sb
;
6644 struct btrfs_disk_key
*disk_key
;
6645 struct btrfs_chunk
*chunk
;
6647 unsigned long sb_array_offset
;
6654 struct btrfs_key key
;
6656 ASSERT(BTRFS_SUPER_INFO_SIZE
<= fs_info
->nodesize
);
6658 * This will create extent buffer of nodesize, superblock size is
6659 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6660 * overallocate but we can keep it as-is, only the first page is used.
6662 sb
= btrfs_find_create_tree_block(fs_info
, BTRFS_SUPER_INFO_OFFSET
);
6665 set_extent_buffer_uptodate(sb
);
6666 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
6668 * The sb extent buffer is artificial and just used to read the system array.
6669 * set_extent_buffer_uptodate() call does not properly mark all it's
6670 * pages up-to-date when the page is larger: extent does not cover the
6671 * whole page and consequently check_page_uptodate does not find all
6672 * the page's extents up-to-date (the hole beyond sb),
6673 * write_extent_buffer then triggers a WARN_ON.
6675 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6676 * but sb spans only this function. Add an explicit SetPageUptodate call
6677 * to silence the warning eg. on PowerPC 64.
6679 if (PAGE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
6680 SetPageUptodate(sb
->pages
[0]);
6682 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
6683 array_size
= btrfs_super_sys_array_size(super_copy
);
6685 array_ptr
= super_copy
->sys_chunk_array
;
6686 sb_array_offset
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
6689 while (cur_offset
< array_size
) {
6690 disk_key
= (struct btrfs_disk_key
*)array_ptr
;
6691 len
= sizeof(*disk_key
);
6692 if (cur_offset
+ len
> array_size
)
6693 goto out_short_read
;
6695 btrfs_disk_key_to_cpu(&key
, disk_key
);
6698 sb_array_offset
+= len
;
6701 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6702 chunk
= (struct btrfs_chunk
*)sb_array_offset
;
6704 * At least one btrfs_chunk with one stripe must be
6705 * present, exact stripe count check comes afterwards
6707 len
= btrfs_chunk_item_size(1);
6708 if (cur_offset
+ len
> array_size
)
6709 goto out_short_read
;
6711 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
6714 "invalid number of stripes %u in sys_array at offset %u",
6715 num_stripes
, cur_offset
);
6720 type
= btrfs_chunk_type(sb
, chunk
);
6721 if ((type
& BTRFS_BLOCK_GROUP_SYSTEM
) == 0) {
6723 "invalid chunk type %llu in sys_array at offset %u",
6729 len
= btrfs_chunk_item_size(num_stripes
);
6730 if (cur_offset
+ len
> array_size
)
6731 goto out_short_read
;
6733 ret
= read_one_chunk(fs_info
, &key
, sb
, chunk
);
6738 "unexpected item type %u in sys_array at offset %u",
6739 (u32
)key
.type
, cur_offset
);
6744 sb_array_offset
+= len
;
6747 clear_extent_buffer_uptodate(sb
);
6748 free_extent_buffer_stale(sb
);
6752 btrfs_err(fs_info
, "sys_array too short to read %u bytes at offset %u",
6754 clear_extent_buffer_uptodate(sb
);
6755 free_extent_buffer_stale(sb
);
6759 void btrfs_report_missing_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
6762 btrfs_warn_rl(fs_info
, "devid %llu uuid %pU is missing", devid
, uuid
);
6766 * Check if all chunks in the fs are OK for read-write degraded mount
6768 * Return true if all chunks meet the minimal RW mount requirements.
6769 * Return false if any chunk doesn't meet the minimal RW mount requirements.
6771 bool btrfs_check_rw_degradable(struct btrfs_fs_info
*fs_info
)
6773 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
6774 struct extent_map
*em
;
6778 read_lock(&map_tree
->map_tree
.lock
);
6779 em
= lookup_extent_mapping(&map_tree
->map_tree
, 0, (u64
)-1);
6780 read_unlock(&map_tree
->map_tree
.lock
);
6781 /* No chunk at all? Return false anyway */
6787 struct map_lookup
*map
;
6792 map
= em
->map_lookup
;
6794 btrfs_get_num_tolerated_disk_barrier_failures(
6796 for (i
= 0; i
< map
->num_stripes
; i
++) {
6797 struct btrfs_device
*dev
= map
->stripes
[i
].dev
;
6799 if (!dev
|| !dev
->bdev
|| dev
->missing
||
6800 dev
->last_flush_error
)
6803 if (missing
> max_tolerated
) {
6805 "chunk %llu missing %d devices, max tolerance is %d for writeable mount",
6806 em
->start
, missing
, max_tolerated
);
6807 free_extent_map(em
);
6811 next_start
= extent_map_end(em
);
6812 free_extent_map(em
);
6814 read_lock(&map_tree
->map_tree
.lock
);
6815 em
= lookup_extent_mapping(&map_tree
->map_tree
, next_start
,
6816 (u64
)(-1) - next_start
);
6817 read_unlock(&map_tree
->map_tree
.lock
);
6823 int btrfs_read_chunk_tree(struct btrfs_fs_info
*fs_info
)
6825 struct btrfs_root
*root
= fs_info
->chunk_root
;
6826 struct btrfs_path
*path
;
6827 struct extent_buffer
*leaf
;
6828 struct btrfs_key key
;
6829 struct btrfs_key found_key
;
6834 path
= btrfs_alloc_path();
6838 mutex_lock(&uuid_mutex
);
6839 mutex_lock(&fs_info
->chunk_mutex
);
6842 * Read all device items, and then all the chunk items. All
6843 * device items are found before any chunk item (their object id
6844 * is smaller than the lowest possible object id for a chunk
6845 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6847 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
6850 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6854 leaf
= path
->nodes
[0];
6855 slot
= path
->slots
[0];
6856 if (slot
>= btrfs_header_nritems(leaf
)) {
6857 ret
= btrfs_next_leaf(root
, path
);
6864 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6865 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
6866 struct btrfs_dev_item
*dev_item
;
6867 dev_item
= btrfs_item_ptr(leaf
, slot
,
6868 struct btrfs_dev_item
);
6869 ret
= read_one_dev(fs_info
, leaf
, dev_item
);
6873 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6874 struct btrfs_chunk
*chunk
;
6875 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
6876 ret
= read_one_chunk(fs_info
, &found_key
, leaf
, chunk
);
6884 * After loading chunk tree, we've got all device information,
6885 * do another round of validation checks.
6887 if (total_dev
!= fs_info
->fs_devices
->total_devices
) {
6889 "super_num_devices %llu mismatch with num_devices %llu found here",
6890 btrfs_super_num_devices(fs_info
->super_copy
),
6895 if (btrfs_super_total_bytes(fs_info
->super_copy
) <
6896 fs_info
->fs_devices
->total_rw_bytes
) {
6898 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6899 btrfs_super_total_bytes(fs_info
->super_copy
),
6900 fs_info
->fs_devices
->total_rw_bytes
);
6906 mutex_unlock(&fs_info
->chunk_mutex
);
6907 mutex_unlock(&uuid_mutex
);
6909 btrfs_free_path(path
);
6913 void btrfs_init_devices_late(struct btrfs_fs_info
*fs_info
)
6915 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6916 struct btrfs_device
*device
;
6918 while (fs_devices
) {
6919 mutex_lock(&fs_devices
->device_list_mutex
);
6920 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
)
6921 device
->fs_info
= fs_info
;
6922 mutex_unlock(&fs_devices
->device_list_mutex
);
6924 fs_devices
= fs_devices
->seed
;
6928 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
6932 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6933 btrfs_dev_stat_reset(dev
, i
);
6936 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
6938 struct btrfs_key key
;
6939 struct btrfs_key found_key
;
6940 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6941 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6942 struct extent_buffer
*eb
;
6945 struct btrfs_device
*device
;
6946 struct btrfs_path
*path
= NULL
;
6949 path
= btrfs_alloc_path();
6955 mutex_lock(&fs_devices
->device_list_mutex
);
6956 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6958 struct btrfs_dev_stats_item
*ptr
;
6960 key
.objectid
= BTRFS_DEV_STATS_OBJECTID
;
6961 key
.type
= BTRFS_PERSISTENT_ITEM_KEY
;
6962 key
.offset
= device
->devid
;
6963 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
6965 __btrfs_reset_dev_stats(device
);
6966 device
->dev_stats_valid
= 1;
6967 btrfs_release_path(path
);
6970 slot
= path
->slots
[0];
6971 eb
= path
->nodes
[0];
6972 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
6973 item_size
= btrfs_item_size_nr(eb
, slot
);
6975 ptr
= btrfs_item_ptr(eb
, slot
,
6976 struct btrfs_dev_stats_item
);
6978 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6979 if (item_size
>= (1 + i
) * sizeof(__le64
))
6980 btrfs_dev_stat_set(device
, i
,
6981 btrfs_dev_stats_value(eb
, ptr
, i
));
6983 btrfs_dev_stat_reset(device
, i
);
6986 device
->dev_stats_valid
= 1;
6987 btrfs_dev_stat_print_on_load(device
);
6988 btrfs_release_path(path
);
6990 mutex_unlock(&fs_devices
->device_list_mutex
);
6993 btrfs_free_path(path
);
6994 return ret
< 0 ? ret
: 0;
6997 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
6998 struct btrfs_fs_info
*fs_info
,
6999 struct btrfs_device
*device
)
7001 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
7002 struct btrfs_path
*path
;
7003 struct btrfs_key key
;
7004 struct extent_buffer
*eb
;
7005 struct btrfs_dev_stats_item
*ptr
;
7009 key
.objectid
= BTRFS_DEV_STATS_OBJECTID
;
7010 key
.type
= BTRFS_PERSISTENT_ITEM_KEY
;
7011 key
.offset
= device
->devid
;
7013 path
= btrfs_alloc_path();
7016 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
7018 btrfs_warn_in_rcu(fs_info
,
7019 "error %d while searching for dev_stats item for device %s",
7020 ret
, rcu_str_deref(device
->name
));
7025 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
7026 /* need to delete old one and insert a new one */
7027 ret
= btrfs_del_item(trans
, dev_root
, path
);
7029 btrfs_warn_in_rcu(fs_info
,
7030 "delete too small dev_stats item for device %s failed %d",
7031 rcu_str_deref(device
->name
), ret
);
7038 /* need to insert a new item */
7039 btrfs_release_path(path
);
7040 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
7041 &key
, sizeof(*ptr
));
7043 btrfs_warn_in_rcu(fs_info
,
7044 "insert dev_stats item for device %s failed %d",
7045 rcu_str_deref(device
->name
), ret
);
7050 eb
= path
->nodes
[0];
7051 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
7052 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7053 btrfs_set_dev_stats_value(eb
, ptr
, i
,
7054 btrfs_dev_stat_read(device
, i
));
7055 btrfs_mark_buffer_dirty(eb
);
7058 btrfs_free_path(path
);
7063 * called from commit_transaction. Writes all changed device stats to disk.
7065 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
7066 struct btrfs_fs_info
*fs_info
)
7068 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7069 struct btrfs_device
*device
;
7073 mutex_lock(&fs_devices
->device_list_mutex
);
7074 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
7075 if (!device
->dev_stats_valid
|| !btrfs_dev_stats_dirty(device
))
7078 stats_cnt
= atomic_read(&device
->dev_stats_ccnt
);
7079 ret
= update_dev_stat_item(trans
, fs_info
, device
);
7081 atomic_sub(stats_cnt
, &device
->dev_stats_ccnt
);
7083 mutex_unlock(&fs_devices
->device_list_mutex
);
7088 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
7090 btrfs_dev_stat_inc(dev
, index
);
7091 btrfs_dev_stat_print_on_error(dev
);
7094 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
7096 if (!dev
->dev_stats_valid
)
7098 btrfs_err_rl_in_rcu(dev
->fs_info
,
7099 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7100 rcu_str_deref(dev
->name
),
7101 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
7102 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
7103 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
7104 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
7105 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
7108 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
7112 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7113 if (btrfs_dev_stat_read(dev
, i
) != 0)
7115 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
7116 return; /* all values == 0, suppress message */
7118 btrfs_info_in_rcu(dev
->fs_info
,
7119 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7120 rcu_str_deref(dev
->name
),
7121 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
7122 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
7123 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
7124 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
7125 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
7128 int btrfs_get_dev_stats(struct btrfs_fs_info
*fs_info
,
7129 struct btrfs_ioctl_get_dev_stats
*stats
)
7131 struct btrfs_device
*dev
;
7132 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7135 mutex_lock(&fs_devices
->device_list_mutex
);
7136 dev
= btrfs_find_device(fs_info
, stats
->devid
, NULL
, NULL
);
7137 mutex_unlock(&fs_devices
->device_list_mutex
);
7140 btrfs_warn(fs_info
, "get dev_stats failed, device not found");
7142 } else if (!dev
->dev_stats_valid
) {
7143 btrfs_warn(fs_info
, "get dev_stats failed, not yet valid");
7145 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
7146 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
7147 if (stats
->nr_items
> i
)
7149 btrfs_dev_stat_read_and_reset(dev
, i
);
7151 btrfs_dev_stat_reset(dev
, i
);
7154 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7155 if (stats
->nr_items
> i
)
7156 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
7158 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
7159 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
7163 void btrfs_scratch_superblocks(struct block_device
*bdev
, const char *device_path
)
7165 struct buffer_head
*bh
;
7166 struct btrfs_super_block
*disk_super
;
7172 for (copy_num
= 0; copy_num
< BTRFS_SUPER_MIRROR_MAX
;
7175 if (btrfs_read_dev_one_super(bdev
, copy_num
, &bh
))
7178 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
7180 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
7181 set_buffer_dirty(bh
);
7182 sync_dirty_buffer(bh
);
7186 /* Notify udev that device has changed */
7187 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
7189 /* Update ctime/mtime for device path for libblkid */
7190 update_dev_time(device_path
);
7194 * Update the size of all devices, which is used for writing out the
7197 void btrfs_update_commit_device_size(struct btrfs_fs_info
*fs_info
)
7199 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7200 struct btrfs_device
*curr
, *next
;
7202 if (list_empty(&fs_devices
->resized_devices
))
7205 mutex_lock(&fs_devices
->device_list_mutex
);
7206 mutex_lock(&fs_info
->chunk_mutex
);
7207 list_for_each_entry_safe(curr
, next
, &fs_devices
->resized_devices
,
7209 list_del_init(&curr
->resized_list
);
7210 curr
->commit_total_bytes
= curr
->disk_total_bytes
;
7212 mutex_unlock(&fs_info
->chunk_mutex
);
7213 mutex_unlock(&fs_devices
->device_list_mutex
);
7216 /* Must be invoked during the transaction commit */
7217 void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info
*fs_info
,
7218 struct btrfs_transaction
*transaction
)
7220 struct extent_map
*em
;
7221 struct map_lookup
*map
;
7222 struct btrfs_device
*dev
;
7225 if (list_empty(&transaction
->pending_chunks
))
7228 /* In order to kick the device replace finish process */
7229 mutex_lock(&fs_info
->chunk_mutex
);
7230 list_for_each_entry(em
, &transaction
->pending_chunks
, list
) {
7231 map
= em
->map_lookup
;
7233 for (i
= 0; i
< map
->num_stripes
; i
++) {
7234 dev
= map
->stripes
[i
].dev
;
7235 dev
->commit_bytes_used
= dev
->bytes_used
;
7238 mutex_unlock(&fs_info
->chunk_mutex
);
7241 void btrfs_set_fs_info_ptr(struct btrfs_fs_info
*fs_info
)
7243 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7244 while (fs_devices
) {
7245 fs_devices
->fs_info
= fs_info
;
7246 fs_devices
= fs_devices
->seed
;
7250 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info
*fs_info
)
7252 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7253 while (fs_devices
) {
7254 fs_devices
->fs_info
= NULL
;
7255 fs_devices
= fs_devices
->seed
;