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;
363 unsigned long last_waited
= 0;
365 int sync_pending
= 0;
366 struct blk_plug plug
;
369 * this function runs all the bios we've collected for
370 * a particular device. We don't want to wander off to
371 * another device without first sending all of these down.
372 * So, setup a plug here and finish it off before we return
374 blk_start_plug(&plug
);
376 bdi
= device
->bdev
->bd_bdi
;
379 spin_lock(&device
->io_lock
);
384 /* take all the bios off the list at once and process them
385 * later on (without the lock held). But, remember the
386 * tail and other pointers so the bios can be properly reinserted
387 * into the list if we hit congestion
389 if (!force_reg
&& device
->pending_sync_bios
.head
) {
390 pending_bios
= &device
->pending_sync_bios
;
393 pending_bios
= &device
->pending_bios
;
397 pending
= pending_bios
->head
;
398 tail
= pending_bios
->tail
;
399 WARN_ON(pending
&& !tail
);
402 * if pending was null this time around, no bios need processing
403 * at all and we can stop. Otherwise it'll loop back up again
404 * and do an additional check so no bios are missed.
406 * device->running_pending is used to synchronize with the
409 if (device
->pending_sync_bios
.head
== NULL
&&
410 device
->pending_bios
.head
== NULL
) {
412 device
->running_pending
= 0;
415 device
->running_pending
= 1;
418 pending_bios
->head
= NULL
;
419 pending_bios
->tail
= NULL
;
421 spin_unlock(&device
->io_lock
);
426 /* we want to work on both lists, but do more bios on the
427 * sync list than the regular list
430 pending_bios
!= &device
->pending_sync_bios
&&
431 device
->pending_sync_bios
.head
) ||
432 (num_run
> 64 && pending_bios
== &device
->pending_sync_bios
&&
433 device
->pending_bios
.head
)) {
434 spin_lock(&device
->io_lock
);
435 requeue_list(pending_bios
, pending
, tail
);
440 pending
= pending
->bi_next
;
443 BUG_ON(atomic_read(&cur
->__bi_cnt
) == 0);
446 * if we're doing the sync list, record that our
447 * plug has some sync requests on it
449 * If we're doing the regular list and there are
450 * sync requests sitting around, unplug before
453 if (pending_bios
== &device
->pending_sync_bios
) {
455 } else if (sync_pending
) {
456 blk_finish_plug(&plug
);
457 blk_start_plug(&plug
);
461 btrfsic_submit_bio(cur
);
468 * we made progress, there is more work to do and the bdi
469 * is now congested. Back off and let other work structs
472 if (pending
&& bdi_write_congested(bdi
) && batch_run
> 8 &&
473 fs_info
->fs_devices
->open_devices
> 1) {
474 struct io_context
*ioc
;
476 ioc
= current
->io_context
;
479 * the main goal here is that we don't want to
480 * block if we're going to be able to submit
481 * more requests without blocking.
483 * This code does two great things, it pokes into
484 * the elevator code from a filesystem _and_
485 * it makes assumptions about how batching works.
487 if (ioc
&& ioc
->nr_batch_requests
> 0 &&
488 time_before(jiffies
, ioc
->last_waited
+ HZ
/50UL) &&
490 ioc
->last_waited
== last_waited
)) {
492 * we want to go through our batch of
493 * requests and stop. So, we copy out
494 * the ioc->last_waited time and test
495 * against it before looping
497 last_waited
= ioc
->last_waited
;
501 spin_lock(&device
->io_lock
);
502 requeue_list(pending_bios
, pending
, tail
);
503 device
->running_pending
= 1;
505 spin_unlock(&device
->io_lock
);
506 btrfs_queue_work(fs_info
->submit_workers
,
516 spin_lock(&device
->io_lock
);
517 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
519 spin_unlock(&device
->io_lock
);
522 blk_finish_plug(&plug
);
525 static void pending_bios_fn(struct btrfs_work
*work
)
527 struct btrfs_device
*device
;
529 device
= container_of(work
, struct btrfs_device
, work
);
530 run_scheduled_bios(device
);
534 static void btrfs_free_stale_device(struct btrfs_device
*cur_dev
)
536 struct btrfs_fs_devices
*fs_devs
;
537 struct btrfs_device
*dev
;
542 list_for_each_entry(fs_devs
, &fs_uuids
, list
) {
547 if (fs_devs
->seeding
)
550 list_for_each_entry(dev
, &fs_devs
->devices
, dev_list
) {
558 * Todo: This won't be enough. What if the same device
559 * comes back (with new uuid and) with its mapper path?
560 * But for now, this does help as mostly an admin will
561 * either use mapper or non mapper path throughout.
564 del
= strcmp(rcu_str_deref(dev
->name
),
565 rcu_str_deref(cur_dev
->name
));
572 /* delete the stale device */
573 if (fs_devs
->num_devices
== 1) {
574 btrfs_sysfs_remove_fsid(fs_devs
);
575 list_del(&fs_devs
->list
);
576 free_fs_devices(fs_devs
);
578 fs_devs
->num_devices
--;
579 list_del(&dev
->dev_list
);
580 rcu_string_free(dev
->name
);
589 * Add new device to list of registered devices
592 * 1 - first time device is seen
593 * 0 - device already known
596 static noinline
int device_list_add(const char *path
,
597 struct btrfs_super_block
*disk_super
,
598 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
600 struct btrfs_device
*device
;
601 struct btrfs_fs_devices
*fs_devices
;
602 struct rcu_string
*name
;
604 u64 found_transid
= btrfs_super_generation(disk_super
);
606 fs_devices
= find_fsid(disk_super
->fsid
);
608 fs_devices
= alloc_fs_devices(disk_super
->fsid
);
609 if (IS_ERR(fs_devices
))
610 return PTR_ERR(fs_devices
);
612 list_add(&fs_devices
->list
, &fs_uuids
);
616 device
= find_device(fs_devices
, devid
,
617 disk_super
->dev_item
.uuid
);
621 if (fs_devices
->opened
)
624 device
= btrfs_alloc_device(NULL
, &devid
,
625 disk_super
->dev_item
.uuid
);
626 if (IS_ERR(device
)) {
627 /* we can safely leave the fs_devices entry around */
628 return PTR_ERR(device
);
631 name
= rcu_string_strdup(path
, GFP_NOFS
);
636 rcu_assign_pointer(device
->name
, name
);
638 mutex_lock(&fs_devices
->device_list_mutex
);
639 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
640 fs_devices
->num_devices
++;
641 mutex_unlock(&fs_devices
->device_list_mutex
);
644 device
->fs_devices
= fs_devices
;
645 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
647 * When FS is already mounted.
648 * 1. If you are here and if the device->name is NULL that
649 * means this device was missing at time of FS mount.
650 * 2. If you are here and if the device->name is different
651 * from 'path' that means either
652 * a. The same device disappeared and reappeared with
654 * b. The missing-disk-which-was-replaced, has
657 * We must allow 1 and 2a above. But 2b would be a spurious
660 * Further in case of 1 and 2a above, the disk at 'path'
661 * would have missed some transaction when it was away and
662 * in case of 2a the stale bdev has to be updated as well.
663 * 2b must not be allowed at all time.
667 * For now, we do allow update to btrfs_fs_device through the
668 * btrfs dev scan cli after FS has been mounted. We're still
669 * tracking a problem where systems fail mount by subvolume id
670 * when we reject replacement on a mounted FS.
672 if (!fs_devices
->opened
&& found_transid
< device
->generation
) {
674 * That is if the FS is _not_ mounted and if you
675 * are here, that means there is more than one
676 * disk with same uuid and devid.We keep the one
677 * with larger generation number or the last-in if
678 * generation are equal.
683 name
= rcu_string_strdup(path
, GFP_NOFS
);
686 rcu_string_free(device
->name
);
687 rcu_assign_pointer(device
->name
, name
);
688 if (device
->missing
) {
689 fs_devices
->missing_devices
--;
695 * Unmount does not free the btrfs_device struct but would zero
696 * generation along with most of the other members. So just update
697 * it back. We need it to pick the disk with largest generation
700 if (!fs_devices
->opened
)
701 device
->generation
= found_transid
;
704 * if there is new btrfs on an already registered device,
705 * then remove the stale device entry.
708 btrfs_free_stale_device(device
);
710 *fs_devices_ret
= fs_devices
;
715 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
717 struct btrfs_fs_devices
*fs_devices
;
718 struct btrfs_device
*device
;
719 struct btrfs_device
*orig_dev
;
721 fs_devices
= alloc_fs_devices(orig
->fsid
);
722 if (IS_ERR(fs_devices
))
725 mutex_lock(&orig
->device_list_mutex
);
726 fs_devices
->total_devices
= orig
->total_devices
;
728 /* We have held the volume lock, it is safe to get the devices. */
729 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
730 struct rcu_string
*name
;
732 device
= btrfs_alloc_device(NULL
, &orig_dev
->devid
,
738 * This is ok to do without rcu read locked because we hold the
739 * uuid mutex so nothing we touch in here is going to disappear.
741 if (orig_dev
->name
) {
742 name
= rcu_string_strdup(orig_dev
->name
->str
,
748 rcu_assign_pointer(device
->name
, name
);
751 list_add(&device
->dev_list
, &fs_devices
->devices
);
752 device
->fs_devices
= fs_devices
;
753 fs_devices
->num_devices
++;
755 mutex_unlock(&orig
->device_list_mutex
);
758 mutex_unlock(&orig
->device_list_mutex
);
759 free_fs_devices(fs_devices
);
760 return ERR_PTR(-ENOMEM
);
763 void btrfs_close_extra_devices(struct btrfs_fs_devices
*fs_devices
, int step
)
765 struct btrfs_device
*device
, *next
;
766 struct btrfs_device
*latest_dev
= NULL
;
768 mutex_lock(&uuid_mutex
);
770 /* This is the initialized path, it is safe to release the devices. */
771 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
772 if (device
->in_fs_metadata
) {
773 if (!device
->is_tgtdev_for_dev_replace
&&
775 device
->generation
> latest_dev
->generation
)) {
781 if (device
->devid
== BTRFS_DEV_REPLACE_DEVID
) {
783 * In the first step, keep the device which has
784 * the correct fsid and the devid that is used
785 * for the dev_replace procedure.
786 * In the second step, the dev_replace state is
787 * read from the device tree and it is known
788 * whether the procedure is really active or
789 * not, which means whether this device is
790 * used or whether it should be removed.
792 if (step
== 0 || device
->is_tgtdev_for_dev_replace
) {
797 blkdev_put(device
->bdev
, device
->mode
);
799 fs_devices
->open_devices
--;
801 if (device
->writeable
) {
802 list_del_init(&device
->dev_alloc_list
);
803 device
->writeable
= 0;
804 if (!device
->is_tgtdev_for_dev_replace
)
805 fs_devices
->rw_devices
--;
807 list_del_init(&device
->dev_list
);
808 fs_devices
->num_devices
--;
809 rcu_string_free(device
->name
);
813 if (fs_devices
->seed
) {
814 fs_devices
= fs_devices
->seed
;
818 fs_devices
->latest_bdev
= latest_dev
->bdev
;
820 mutex_unlock(&uuid_mutex
);
823 static void __free_device(struct work_struct
*work
)
825 struct btrfs_device
*device
;
827 device
= container_of(work
, struct btrfs_device
, rcu_work
);
828 rcu_string_free(device
->name
);
829 bio_put(device
->flush_bio
);
833 static void free_device(struct rcu_head
*head
)
835 struct btrfs_device
*device
;
837 device
= container_of(head
, struct btrfs_device
, rcu
);
839 INIT_WORK(&device
->rcu_work
, __free_device
);
840 schedule_work(&device
->rcu_work
);
843 static void btrfs_close_bdev(struct btrfs_device
*device
)
845 if (device
->bdev
&& device
->writeable
) {
846 sync_blockdev(device
->bdev
);
847 invalidate_bdev(device
->bdev
);
851 blkdev_put(device
->bdev
, device
->mode
);
854 static void btrfs_prepare_close_one_device(struct btrfs_device
*device
)
856 struct btrfs_fs_devices
*fs_devices
= device
->fs_devices
;
857 struct btrfs_device
*new_device
;
858 struct rcu_string
*name
;
861 fs_devices
->open_devices
--;
863 if (device
->writeable
&&
864 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
865 list_del_init(&device
->dev_alloc_list
);
866 fs_devices
->rw_devices
--;
870 fs_devices
->missing_devices
--;
872 new_device
= btrfs_alloc_device(NULL
, &device
->devid
,
874 BUG_ON(IS_ERR(new_device
)); /* -ENOMEM */
876 /* Safe because we are under uuid_mutex */
878 name
= rcu_string_strdup(device
->name
->str
, GFP_NOFS
);
879 BUG_ON(!name
); /* -ENOMEM */
880 rcu_assign_pointer(new_device
->name
, name
);
883 list_replace_rcu(&device
->dev_list
, &new_device
->dev_list
);
884 new_device
->fs_devices
= device
->fs_devices
;
887 static int __btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
889 struct btrfs_device
*device
, *tmp
;
890 struct list_head pending_put
;
892 INIT_LIST_HEAD(&pending_put
);
894 if (--fs_devices
->opened
> 0)
897 mutex_lock(&fs_devices
->device_list_mutex
);
898 list_for_each_entry_safe(device
, tmp
, &fs_devices
->devices
, dev_list
) {
899 btrfs_prepare_close_one_device(device
);
900 list_add(&device
->dev_list
, &pending_put
);
902 mutex_unlock(&fs_devices
->device_list_mutex
);
905 * btrfs_show_devname() is using the device_list_mutex,
906 * sometimes call to blkdev_put() leads vfs calling
907 * into this func. So do put outside of device_list_mutex,
910 while (!list_empty(&pending_put
)) {
911 device
= list_first_entry(&pending_put
,
912 struct btrfs_device
, dev_list
);
913 list_del(&device
->dev_list
);
914 btrfs_close_bdev(device
);
915 call_rcu(&device
->rcu
, free_device
);
918 WARN_ON(fs_devices
->open_devices
);
919 WARN_ON(fs_devices
->rw_devices
);
920 fs_devices
->opened
= 0;
921 fs_devices
->seeding
= 0;
926 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
928 struct btrfs_fs_devices
*seed_devices
= NULL
;
931 mutex_lock(&uuid_mutex
);
932 ret
= __btrfs_close_devices(fs_devices
);
933 if (!fs_devices
->opened
) {
934 seed_devices
= fs_devices
->seed
;
935 fs_devices
->seed
= NULL
;
937 mutex_unlock(&uuid_mutex
);
939 while (seed_devices
) {
940 fs_devices
= seed_devices
;
941 seed_devices
= fs_devices
->seed
;
942 __btrfs_close_devices(fs_devices
);
943 free_fs_devices(fs_devices
);
946 * Wait for rcu kworkers under __btrfs_close_devices
947 * to finish all blkdev_puts so device is really
948 * free when umount is done.
954 static int __btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
955 fmode_t flags
, void *holder
)
957 struct request_queue
*q
;
958 struct block_device
*bdev
;
959 struct list_head
*head
= &fs_devices
->devices
;
960 struct btrfs_device
*device
;
961 struct btrfs_device
*latest_dev
= NULL
;
962 struct buffer_head
*bh
;
963 struct btrfs_super_block
*disk_super
;
970 list_for_each_entry(device
, head
, dev_list
) {
976 /* Just open everything we can; ignore failures here */
977 if (btrfs_get_bdev_and_sb(device
->name
->str
, flags
, holder
, 1,
981 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
982 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
983 if (devid
!= device
->devid
)
986 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
,
990 device
->generation
= btrfs_super_generation(disk_super
);
992 device
->generation
> latest_dev
->generation
)
995 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
996 device
->writeable
= 0;
998 device
->writeable
= !bdev_read_only(bdev
);
1002 q
= bdev_get_queue(bdev
);
1003 if (blk_queue_discard(q
))
1004 device
->can_discard
= 1;
1005 if (!blk_queue_nonrot(q
))
1006 fs_devices
->rotating
= 1;
1008 device
->bdev
= bdev
;
1009 device
->in_fs_metadata
= 0;
1010 device
->mode
= flags
;
1012 fs_devices
->open_devices
++;
1013 if (device
->writeable
&&
1014 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
1015 fs_devices
->rw_devices
++;
1016 list_add(&device
->dev_alloc_list
,
1017 &fs_devices
->alloc_list
);
1024 blkdev_put(bdev
, flags
);
1027 if (fs_devices
->open_devices
== 0) {
1031 fs_devices
->seeding
= seeding
;
1032 fs_devices
->opened
= 1;
1033 fs_devices
->latest_bdev
= latest_dev
->bdev
;
1034 fs_devices
->total_rw_bytes
= 0;
1039 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
1040 fmode_t flags
, void *holder
)
1044 mutex_lock(&uuid_mutex
);
1045 if (fs_devices
->opened
) {
1046 fs_devices
->opened
++;
1049 ret
= __btrfs_open_devices(fs_devices
, flags
, holder
);
1051 mutex_unlock(&uuid_mutex
);
1055 static void btrfs_release_disk_super(struct page
*page
)
1061 static int btrfs_read_disk_super(struct block_device
*bdev
, u64 bytenr
,
1063 struct btrfs_super_block
**disk_super
)
1068 /* make sure our super fits in the device */
1069 if (bytenr
+ PAGE_SIZE
>= i_size_read(bdev
->bd_inode
))
1072 /* make sure our super fits in the page */
1073 if (sizeof(**disk_super
) > PAGE_SIZE
)
1076 /* make sure our super doesn't straddle pages on disk */
1077 index
= bytenr
>> PAGE_SHIFT
;
1078 if ((bytenr
+ sizeof(**disk_super
) - 1) >> PAGE_SHIFT
!= index
)
1081 /* pull in the page with our super */
1082 *page
= read_cache_page_gfp(bdev
->bd_inode
->i_mapping
,
1085 if (IS_ERR_OR_NULL(*page
))
1090 /* align our pointer to the offset of the super block */
1091 *disk_super
= p
+ (bytenr
& ~PAGE_MASK
);
1093 if (btrfs_super_bytenr(*disk_super
) != bytenr
||
1094 btrfs_super_magic(*disk_super
) != BTRFS_MAGIC
) {
1095 btrfs_release_disk_super(*page
);
1099 if ((*disk_super
)->label
[0] &&
1100 (*disk_super
)->label
[BTRFS_LABEL_SIZE
- 1])
1101 (*disk_super
)->label
[BTRFS_LABEL_SIZE
- 1] = '\0';
1107 * Look for a btrfs signature on a device. This may be called out of the mount path
1108 * and we are not allowed to call set_blocksize during the scan. The superblock
1109 * is read via pagecache
1111 int btrfs_scan_one_device(const char *path
, fmode_t flags
, void *holder
,
1112 struct btrfs_fs_devices
**fs_devices_ret
)
1114 struct btrfs_super_block
*disk_super
;
1115 struct block_device
*bdev
;
1124 * we would like to check all the supers, but that would make
1125 * a btrfs mount succeed after a mkfs from a different FS.
1126 * So, we need to add a special mount option to scan for
1127 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1129 bytenr
= btrfs_sb_offset(0);
1130 flags
|= FMODE_EXCL
;
1131 mutex_lock(&uuid_mutex
);
1133 bdev
= blkdev_get_by_path(path
, flags
, holder
);
1135 ret
= PTR_ERR(bdev
);
1139 if (btrfs_read_disk_super(bdev
, bytenr
, &page
, &disk_super
))
1140 goto error_bdev_put
;
1142 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1143 transid
= btrfs_super_generation(disk_super
);
1144 total_devices
= btrfs_super_num_devices(disk_super
);
1146 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
1148 if (disk_super
->label
[0]) {
1149 pr_info("BTRFS: device label %s ", disk_super
->label
);
1151 pr_info("BTRFS: device fsid %pU ", disk_super
->fsid
);
1154 pr_cont("devid %llu transid %llu %s\n", devid
, transid
, path
);
1157 if (!ret
&& fs_devices_ret
)
1158 (*fs_devices_ret
)->total_devices
= total_devices
;
1160 btrfs_release_disk_super(page
);
1163 blkdev_put(bdev
, flags
);
1165 mutex_unlock(&uuid_mutex
);
1169 /* helper to account the used device space in the range */
1170 int btrfs_account_dev_extents_size(struct btrfs_device
*device
, u64 start
,
1171 u64 end
, u64
*length
)
1173 struct btrfs_key key
;
1174 struct btrfs_root
*root
= device
->fs_info
->dev_root
;
1175 struct btrfs_dev_extent
*dev_extent
;
1176 struct btrfs_path
*path
;
1180 struct extent_buffer
*l
;
1184 if (start
>= device
->total_bytes
|| device
->is_tgtdev_for_dev_replace
)
1187 path
= btrfs_alloc_path();
1190 path
->reada
= READA_FORWARD
;
1192 key
.objectid
= device
->devid
;
1194 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1196 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1200 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1207 slot
= path
->slots
[0];
1208 if (slot
>= btrfs_header_nritems(l
)) {
1209 ret
= btrfs_next_leaf(root
, path
);
1217 btrfs_item_key_to_cpu(l
, &key
, slot
);
1219 if (key
.objectid
< device
->devid
)
1222 if (key
.objectid
> device
->devid
)
1225 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1228 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1229 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1231 if (key
.offset
<= start
&& extent_end
> end
) {
1232 *length
= end
- start
+ 1;
1234 } else if (key
.offset
<= start
&& extent_end
> start
)
1235 *length
+= extent_end
- start
;
1236 else if (key
.offset
> start
&& extent_end
<= end
)
1237 *length
+= extent_end
- key
.offset
;
1238 else if (key
.offset
> start
&& key
.offset
<= end
) {
1239 *length
+= end
- key
.offset
+ 1;
1241 } else if (key
.offset
> end
)
1249 btrfs_free_path(path
);
1253 static int contains_pending_extent(struct btrfs_transaction
*transaction
,
1254 struct btrfs_device
*device
,
1255 u64
*start
, u64 len
)
1257 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1258 struct extent_map
*em
;
1259 struct list_head
*search_list
= &fs_info
->pinned_chunks
;
1261 u64 physical_start
= *start
;
1264 search_list
= &transaction
->pending_chunks
;
1266 list_for_each_entry(em
, search_list
, list
) {
1267 struct map_lookup
*map
;
1270 map
= em
->map_lookup
;
1271 for (i
= 0; i
< map
->num_stripes
; i
++) {
1274 if (map
->stripes
[i
].dev
!= device
)
1276 if (map
->stripes
[i
].physical
>= physical_start
+ len
||
1277 map
->stripes
[i
].physical
+ em
->orig_block_len
<=
1281 * Make sure that while processing the pinned list we do
1282 * not override our *start with a lower value, because
1283 * we can have pinned chunks that fall within this
1284 * device hole and that have lower physical addresses
1285 * than the pending chunks we processed before. If we
1286 * do not take this special care we can end up getting
1287 * 2 pending chunks that start at the same physical
1288 * device offsets because the end offset of a pinned
1289 * chunk can be equal to the start offset of some
1292 end
= map
->stripes
[i
].physical
+ em
->orig_block_len
;
1299 if (search_list
!= &fs_info
->pinned_chunks
) {
1300 search_list
= &fs_info
->pinned_chunks
;
1309 * find_free_dev_extent_start - find free space in the specified device
1310 * @device: the device which we search the free space in
1311 * @num_bytes: the size of the free space that we need
1312 * @search_start: the position from which to begin the search
1313 * @start: store the start of the free space.
1314 * @len: the size of the free space. that we find, or the size
1315 * of the max free space if we don't find suitable free space
1317 * this uses a pretty simple search, the expectation is that it is
1318 * called very infrequently and that a given device has a small number
1321 * @start is used to store the start of the free space if we find. But if we
1322 * don't find suitable free space, it will be used to store the start position
1323 * of the max free space.
1325 * @len is used to store the size of the free space that we find.
1326 * But if we don't find suitable free space, it is used to store the size of
1327 * the max free space.
1329 int find_free_dev_extent_start(struct btrfs_transaction
*transaction
,
1330 struct btrfs_device
*device
, u64 num_bytes
,
1331 u64 search_start
, u64
*start
, u64
*len
)
1333 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1334 struct btrfs_root
*root
= fs_info
->dev_root
;
1335 struct btrfs_key key
;
1336 struct btrfs_dev_extent
*dev_extent
;
1337 struct btrfs_path
*path
;
1342 u64 search_end
= device
->total_bytes
;
1345 struct extent_buffer
*l
;
1348 * We don't want to overwrite the superblock on the drive nor any area
1349 * used by the boot loader (grub for example), so we make sure to start
1350 * at an offset of at least 1MB.
1352 search_start
= max_t(u64
, search_start
, SZ_1M
);
1354 path
= btrfs_alloc_path();
1358 max_hole_start
= search_start
;
1362 if (search_start
>= search_end
|| device
->is_tgtdev_for_dev_replace
) {
1367 path
->reada
= READA_FORWARD
;
1368 path
->search_commit_root
= 1;
1369 path
->skip_locking
= 1;
1371 key
.objectid
= device
->devid
;
1372 key
.offset
= search_start
;
1373 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1375 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1379 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1386 slot
= path
->slots
[0];
1387 if (slot
>= btrfs_header_nritems(l
)) {
1388 ret
= btrfs_next_leaf(root
, path
);
1396 btrfs_item_key_to_cpu(l
, &key
, slot
);
1398 if (key
.objectid
< device
->devid
)
1401 if (key
.objectid
> device
->devid
)
1404 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1407 if (key
.offset
> search_start
) {
1408 hole_size
= key
.offset
- search_start
;
1411 * Have to check before we set max_hole_start, otherwise
1412 * we could end up sending back this offset anyway.
1414 if (contains_pending_extent(transaction
, device
,
1417 if (key
.offset
>= search_start
) {
1418 hole_size
= key
.offset
- search_start
;
1425 if (hole_size
> max_hole_size
) {
1426 max_hole_start
= search_start
;
1427 max_hole_size
= hole_size
;
1431 * If this free space is greater than which we need,
1432 * it must be the max free space that we have found
1433 * until now, so max_hole_start must point to the start
1434 * of this free space and the length of this free space
1435 * is stored in max_hole_size. Thus, we return
1436 * max_hole_start and max_hole_size and go back to the
1439 if (hole_size
>= num_bytes
) {
1445 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1446 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1448 if (extent_end
> search_start
)
1449 search_start
= extent_end
;
1456 * At this point, search_start should be the end of
1457 * allocated dev extents, and when shrinking the device,
1458 * search_end may be smaller than search_start.
1460 if (search_end
> search_start
) {
1461 hole_size
= search_end
- search_start
;
1463 if (contains_pending_extent(transaction
, device
, &search_start
,
1465 btrfs_release_path(path
);
1469 if (hole_size
> max_hole_size
) {
1470 max_hole_start
= search_start
;
1471 max_hole_size
= hole_size
;
1476 if (max_hole_size
< num_bytes
)
1482 btrfs_free_path(path
);
1483 *start
= max_hole_start
;
1485 *len
= max_hole_size
;
1489 int find_free_dev_extent(struct btrfs_trans_handle
*trans
,
1490 struct btrfs_device
*device
, u64 num_bytes
,
1491 u64
*start
, u64
*len
)
1493 /* FIXME use last free of some kind */
1494 return find_free_dev_extent_start(trans
->transaction
, device
,
1495 num_bytes
, 0, start
, len
);
1498 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1499 struct btrfs_device
*device
,
1500 u64 start
, u64
*dev_extent_len
)
1502 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1503 struct btrfs_root
*root
= fs_info
->dev_root
;
1505 struct btrfs_path
*path
;
1506 struct btrfs_key key
;
1507 struct btrfs_key found_key
;
1508 struct extent_buffer
*leaf
= NULL
;
1509 struct btrfs_dev_extent
*extent
= NULL
;
1511 path
= btrfs_alloc_path();
1515 key
.objectid
= device
->devid
;
1517 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1519 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1521 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1522 BTRFS_DEV_EXTENT_KEY
);
1525 leaf
= path
->nodes
[0];
1526 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1527 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1528 struct btrfs_dev_extent
);
1529 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1530 btrfs_dev_extent_length(leaf
, extent
) < start
);
1532 btrfs_release_path(path
);
1534 } else if (ret
== 0) {
1535 leaf
= path
->nodes
[0];
1536 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1537 struct btrfs_dev_extent
);
1539 btrfs_handle_fs_error(fs_info
, ret
, "Slot search failed");
1543 *dev_extent_len
= btrfs_dev_extent_length(leaf
, extent
);
1545 ret
= btrfs_del_item(trans
, root
, path
);
1547 btrfs_handle_fs_error(fs_info
, ret
,
1548 "Failed to remove dev extent item");
1550 set_bit(BTRFS_TRANS_HAVE_FREE_BGS
, &trans
->transaction
->flags
);
1553 btrfs_free_path(path
);
1557 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1558 struct btrfs_device
*device
,
1559 u64 chunk_offset
, u64 start
, u64 num_bytes
)
1562 struct btrfs_path
*path
;
1563 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1564 struct btrfs_root
*root
= fs_info
->dev_root
;
1565 struct btrfs_dev_extent
*extent
;
1566 struct extent_buffer
*leaf
;
1567 struct btrfs_key key
;
1569 WARN_ON(!device
->in_fs_metadata
);
1570 WARN_ON(device
->is_tgtdev_for_dev_replace
);
1571 path
= btrfs_alloc_path();
1575 key
.objectid
= device
->devid
;
1577 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1578 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1583 leaf
= path
->nodes
[0];
1584 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1585 struct btrfs_dev_extent
);
1586 btrfs_set_dev_extent_chunk_tree(leaf
, extent
,
1587 BTRFS_CHUNK_TREE_OBJECTID
);
1588 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
,
1589 BTRFS_FIRST_CHUNK_TREE_OBJECTID
);
1590 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1592 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1593 btrfs_mark_buffer_dirty(leaf
);
1595 btrfs_free_path(path
);
1599 static u64
find_next_chunk(struct btrfs_fs_info
*fs_info
)
1601 struct extent_map_tree
*em_tree
;
1602 struct extent_map
*em
;
1606 em_tree
= &fs_info
->mapping_tree
.map_tree
;
1607 read_lock(&em_tree
->lock
);
1608 n
= rb_last(&em_tree
->map
);
1610 em
= rb_entry(n
, struct extent_map
, rb_node
);
1611 ret
= em
->start
+ em
->len
;
1613 read_unlock(&em_tree
->lock
);
1618 static noinline
int find_next_devid(struct btrfs_fs_info
*fs_info
,
1622 struct btrfs_key key
;
1623 struct btrfs_key found_key
;
1624 struct btrfs_path
*path
;
1626 path
= btrfs_alloc_path();
1630 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1631 key
.type
= BTRFS_DEV_ITEM_KEY
;
1632 key
.offset
= (u64
)-1;
1634 ret
= btrfs_search_slot(NULL
, fs_info
->chunk_root
, &key
, path
, 0, 0);
1638 BUG_ON(ret
== 0); /* Corruption */
1640 ret
= btrfs_previous_item(fs_info
->chunk_root
, path
,
1641 BTRFS_DEV_ITEMS_OBJECTID
,
1642 BTRFS_DEV_ITEM_KEY
);
1646 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1648 *devid_ret
= found_key
.offset
+ 1;
1652 btrfs_free_path(path
);
1657 * the device information is stored in the chunk root
1658 * the btrfs_device struct should be fully filled in
1660 static int btrfs_add_device(struct btrfs_trans_handle
*trans
,
1661 struct btrfs_fs_info
*fs_info
,
1662 struct btrfs_device
*device
)
1664 struct btrfs_root
*root
= fs_info
->chunk_root
;
1666 struct btrfs_path
*path
;
1667 struct btrfs_dev_item
*dev_item
;
1668 struct extent_buffer
*leaf
;
1669 struct btrfs_key key
;
1672 path
= btrfs_alloc_path();
1676 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1677 key
.type
= BTRFS_DEV_ITEM_KEY
;
1678 key
.offset
= device
->devid
;
1680 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1685 leaf
= path
->nodes
[0];
1686 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1688 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1689 btrfs_set_device_generation(leaf
, dev_item
, 0);
1690 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1691 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1692 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1693 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1694 btrfs_set_device_total_bytes(leaf
, dev_item
,
1695 btrfs_device_get_disk_total_bytes(device
));
1696 btrfs_set_device_bytes_used(leaf
, dev_item
,
1697 btrfs_device_get_bytes_used(device
));
1698 btrfs_set_device_group(leaf
, dev_item
, 0);
1699 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1700 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1701 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1703 ptr
= btrfs_device_uuid(dev_item
);
1704 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1705 ptr
= btrfs_device_fsid(dev_item
);
1706 write_extent_buffer(leaf
, fs_info
->fsid
, ptr
, BTRFS_FSID_SIZE
);
1707 btrfs_mark_buffer_dirty(leaf
);
1711 btrfs_free_path(path
);
1716 * Function to update ctime/mtime for a given device path.
1717 * Mainly used for ctime/mtime based probe like libblkid.
1719 static void update_dev_time(const char *path_name
)
1723 filp
= filp_open(path_name
, O_RDWR
, 0);
1726 file_update_time(filp
);
1727 filp_close(filp
, NULL
);
1730 static int btrfs_rm_dev_item(struct btrfs_fs_info
*fs_info
,
1731 struct btrfs_device
*device
)
1733 struct btrfs_root
*root
= fs_info
->chunk_root
;
1735 struct btrfs_path
*path
;
1736 struct btrfs_key key
;
1737 struct btrfs_trans_handle
*trans
;
1739 path
= btrfs_alloc_path();
1743 trans
= btrfs_start_transaction(root
, 0);
1744 if (IS_ERR(trans
)) {
1745 btrfs_free_path(path
);
1746 return PTR_ERR(trans
);
1748 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1749 key
.type
= BTRFS_DEV_ITEM_KEY
;
1750 key
.offset
= device
->devid
;
1752 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1761 ret
= btrfs_del_item(trans
, root
, path
);
1765 btrfs_free_path(path
);
1766 btrfs_commit_transaction(trans
);
1771 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1772 * filesystem. It's up to the caller to adjust that number regarding eg. device
1775 static int btrfs_check_raid_min_devices(struct btrfs_fs_info
*fs_info
,
1783 seq
= read_seqbegin(&fs_info
->profiles_lock
);
1785 all_avail
= fs_info
->avail_data_alloc_bits
|
1786 fs_info
->avail_system_alloc_bits
|
1787 fs_info
->avail_metadata_alloc_bits
;
1788 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
1790 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
1791 if (!(all_avail
& btrfs_raid_group
[i
]))
1794 if (num_devices
< btrfs_raid_array
[i
].devs_min
) {
1795 int ret
= btrfs_raid_mindev_error
[i
];
1805 static struct btrfs_device
* btrfs_find_next_active_device(
1806 struct btrfs_fs_devices
*fs_devs
, struct btrfs_device
*device
)
1808 struct btrfs_device
*next_device
;
1810 list_for_each_entry(next_device
, &fs_devs
->devices
, dev_list
) {
1811 if (next_device
!= device
&&
1812 !next_device
->missing
&& next_device
->bdev
)
1820 * Helper function to check if the given device is part of s_bdev / latest_bdev
1821 * and replace it with the provided or the next active device, in the context
1822 * where this function called, there should be always be another device (or
1823 * this_dev) which is active.
1825 void btrfs_assign_next_active_device(struct btrfs_fs_info
*fs_info
,
1826 struct btrfs_device
*device
, struct btrfs_device
*this_dev
)
1828 struct btrfs_device
*next_device
;
1831 next_device
= this_dev
;
1833 next_device
= btrfs_find_next_active_device(fs_info
->fs_devices
,
1835 ASSERT(next_device
);
1837 if (fs_info
->sb
->s_bdev
&&
1838 (fs_info
->sb
->s_bdev
== device
->bdev
))
1839 fs_info
->sb
->s_bdev
= next_device
->bdev
;
1841 if (fs_info
->fs_devices
->latest_bdev
== device
->bdev
)
1842 fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1845 int btrfs_rm_device(struct btrfs_fs_info
*fs_info
, const char *device_path
,
1848 struct btrfs_device
*device
;
1849 struct btrfs_fs_devices
*cur_devices
;
1853 mutex_lock(&uuid_mutex
);
1855 num_devices
= fs_info
->fs_devices
->num_devices
;
1856 btrfs_dev_replace_lock(&fs_info
->dev_replace
, 0);
1857 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
1858 WARN_ON(num_devices
< 1);
1861 btrfs_dev_replace_unlock(&fs_info
->dev_replace
, 0);
1863 ret
= btrfs_check_raid_min_devices(fs_info
, num_devices
- 1);
1867 ret
= btrfs_find_device_by_devspec(fs_info
, devid
, device_path
,
1872 if (device
->is_tgtdev_for_dev_replace
) {
1873 ret
= BTRFS_ERROR_DEV_TGT_REPLACE
;
1877 if (device
->writeable
&& fs_info
->fs_devices
->rw_devices
== 1) {
1878 ret
= BTRFS_ERROR_DEV_ONLY_WRITABLE
;
1882 if (device
->writeable
) {
1883 mutex_lock(&fs_info
->chunk_mutex
);
1884 list_del_init(&device
->dev_alloc_list
);
1885 device
->fs_devices
->rw_devices
--;
1886 mutex_unlock(&fs_info
->chunk_mutex
);
1889 mutex_unlock(&uuid_mutex
);
1890 ret
= btrfs_shrink_device(device
, 0);
1891 mutex_lock(&uuid_mutex
);
1896 * TODO: the superblock still includes this device in its num_devices
1897 * counter although write_all_supers() is not locked out. This
1898 * could give a filesystem state which requires a degraded mount.
1900 ret
= btrfs_rm_dev_item(fs_info
, device
);
1904 device
->in_fs_metadata
= 0;
1905 btrfs_scrub_cancel_dev(fs_info
, device
);
1908 * the device list mutex makes sure that we don't change
1909 * the device list while someone else is writing out all
1910 * the device supers. Whoever is writing all supers, should
1911 * lock the device list mutex before getting the number of
1912 * devices in the super block (super_copy). Conversely,
1913 * whoever updates the number of devices in the super block
1914 * (super_copy) should hold the device list mutex.
1917 cur_devices
= device
->fs_devices
;
1918 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
1919 list_del_rcu(&device
->dev_list
);
1921 device
->fs_devices
->num_devices
--;
1922 device
->fs_devices
->total_devices
--;
1924 if (device
->missing
)
1925 device
->fs_devices
->missing_devices
--;
1927 btrfs_assign_next_active_device(fs_info
, device
, NULL
);
1930 device
->fs_devices
->open_devices
--;
1931 /* remove sysfs entry */
1932 btrfs_sysfs_rm_device_link(fs_info
->fs_devices
, device
);
1935 num_devices
= btrfs_super_num_devices(fs_info
->super_copy
) - 1;
1936 btrfs_set_super_num_devices(fs_info
->super_copy
, num_devices
);
1937 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
1940 * at this point, the device is zero sized and detached from
1941 * the devices list. All that's left is to zero out the old
1942 * supers and free the device.
1944 if (device
->writeable
)
1945 btrfs_scratch_superblocks(device
->bdev
, device
->name
->str
);
1947 btrfs_close_bdev(device
);
1948 call_rcu(&device
->rcu
, free_device
);
1950 if (cur_devices
->open_devices
== 0) {
1951 struct btrfs_fs_devices
*fs_devices
;
1952 fs_devices
= fs_info
->fs_devices
;
1953 while (fs_devices
) {
1954 if (fs_devices
->seed
== cur_devices
) {
1955 fs_devices
->seed
= cur_devices
->seed
;
1958 fs_devices
= fs_devices
->seed
;
1960 cur_devices
->seed
= NULL
;
1961 __btrfs_close_devices(cur_devices
);
1962 free_fs_devices(cur_devices
);
1966 mutex_unlock(&uuid_mutex
);
1970 if (device
->writeable
) {
1971 mutex_lock(&fs_info
->chunk_mutex
);
1972 list_add(&device
->dev_alloc_list
,
1973 &fs_info
->fs_devices
->alloc_list
);
1974 device
->fs_devices
->rw_devices
++;
1975 mutex_unlock(&fs_info
->chunk_mutex
);
1980 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info
*fs_info
,
1981 struct btrfs_device
*srcdev
)
1983 struct btrfs_fs_devices
*fs_devices
;
1985 WARN_ON(!mutex_is_locked(&fs_info
->fs_devices
->device_list_mutex
));
1988 * in case of fs with no seed, srcdev->fs_devices will point
1989 * to fs_devices of fs_info. However when the dev being replaced is
1990 * a seed dev it will point to the seed's local fs_devices. In short
1991 * srcdev will have its correct fs_devices in both the cases.
1993 fs_devices
= srcdev
->fs_devices
;
1995 list_del_rcu(&srcdev
->dev_list
);
1996 list_del_rcu(&srcdev
->dev_alloc_list
);
1997 fs_devices
->num_devices
--;
1998 if (srcdev
->missing
)
1999 fs_devices
->missing_devices
--;
2001 if (srcdev
->writeable
)
2002 fs_devices
->rw_devices
--;
2005 fs_devices
->open_devices
--;
2008 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info
*fs_info
,
2009 struct btrfs_device
*srcdev
)
2011 struct btrfs_fs_devices
*fs_devices
= srcdev
->fs_devices
;
2013 if (srcdev
->writeable
) {
2014 /* zero out the old super if it is writable */
2015 btrfs_scratch_superblocks(srcdev
->bdev
, srcdev
->name
->str
);
2018 btrfs_close_bdev(srcdev
);
2019 call_rcu(&srcdev
->rcu
, free_device
);
2021 /* if this is no devs we rather delete the fs_devices */
2022 if (!fs_devices
->num_devices
) {
2023 struct btrfs_fs_devices
*tmp_fs_devices
;
2026 * On a mounted FS, num_devices can't be zero unless it's a
2027 * seed. In case of a seed device being replaced, the replace
2028 * target added to the sprout FS, so there will be no more
2029 * device left under the seed FS.
2031 ASSERT(fs_devices
->seeding
);
2033 tmp_fs_devices
= fs_info
->fs_devices
;
2034 while (tmp_fs_devices
) {
2035 if (tmp_fs_devices
->seed
== fs_devices
) {
2036 tmp_fs_devices
->seed
= fs_devices
->seed
;
2039 tmp_fs_devices
= tmp_fs_devices
->seed
;
2041 fs_devices
->seed
= NULL
;
2042 __btrfs_close_devices(fs_devices
);
2043 free_fs_devices(fs_devices
);
2047 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
2048 struct btrfs_device
*tgtdev
)
2050 mutex_lock(&uuid_mutex
);
2052 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2054 btrfs_sysfs_rm_device_link(fs_info
->fs_devices
, tgtdev
);
2057 fs_info
->fs_devices
->open_devices
--;
2059 fs_info
->fs_devices
->num_devices
--;
2061 btrfs_assign_next_active_device(fs_info
, tgtdev
, NULL
);
2063 list_del_rcu(&tgtdev
->dev_list
);
2065 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2066 mutex_unlock(&uuid_mutex
);
2069 * The update_dev_time() with in btrfs_scratch_superblocks()
2070 * may lead to a call to btrfs_show_devname() which will try
2071 * to hold device_list_mutex. And here this device
2072 * is already out of device list, so we don't have to hold
2073 * the device_list_mutex lock.
2075 btrfs_scratch_superblocks(tgtdev
->bdev
, tgtdev
->name
->str
);
2077 btrfs_close_bdev(tgtdev
);
2078 call_rcu(&tgtdev
->rcu
, free_device
);
2081 static int btrfs_find_device_by_path(struct btrfs_fs_info
*fs_info
,
2082 const char *device_path
,
2083 struct btrfs_device
**device
)
2086 struct btrfs_super_block
*disk_super
;
2089 struct block_device
*bdev
;
2090 struct buffer_head
*bh
;
2093 ret
= btrfs_get_bdev_and_sb(device_path
, FMODE_READ
,
2094 fs_info
->bdev_holder
, 0, &bdev
, &bh
);
2097 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
2098 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
2099 dev_uuid
= disk_super
->dev_item
.uuid
;
2100 *device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, disk_super
->fsid
);
2104 blkdev_put(bdev
, FMODE_READ
);
2108 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info
*fs_info
,
2109 const char *device_path
,
2110 struct btrfs_device
**device
)
2113 if (strcmp(device_path
, "missing") == 0) {
2114 struct list_head
*devices
;
2115 struct btrfs_device
*tmp
;
2117 devices
= &fs_info
->fs_devices
->devices
;
2119 * It is safe to read the devices since the volume_mutex
2120 * is held by the caller.
2122 list_for_each_entry(tmp
, devices
, dev_list
) {
2123 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
2130 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND
;
2134 return btrfs_find_device_by_path(fs_info
, device_path
, device
);
2139 * Lookup a device given by device id, or the path if the id is 0.
2141 int btrfs_find_device_by_devspec(struct btrfs_fs_info
*fs_info
, u64 devid
,
2142 const char *devpath
,
2143 struct btrfs_device
**device
)
2149 *device
= btrfs_find_device(fs_info
, devid
, NULL
, NULL
);
2153 if (!devpath
|| !devpath
[0])
2156 ret
= btrfs_find_device_missing_or_by_path(fs_info
, devpath
,
2163 * does all the dirty work required for changing file system's UUID.
2165 static int btrfs_prepare_sprout(struct btrfs_fs_info
*fs_info
)
2167 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2168 struct btrfs_fs_devices
*old_devices
;
2169 struct btrfs_fs_devices
*seed_devices
;
2170 struct btrfs_super_block
*disk_super
= fs_info
->super_copy
;
2171 struct btrfs_device
*device
;
2174 BUG_ON(!mutex_is_locked(&uuid_mutex
));
2175 if (!fs_devices
->seeding
)
2178 seed_devices
= alloc_fs_devices(NULL
);
2179 if (IS_ERR(seed_devices
))
2180 return PTR_ERR(seed_devices
);
2182 old_devices
= clone_fs_devices(fs_devices
);
2183 if (IS_ERR(old_devices
)) {
2184 kfree(seed_devices
);
2185 return PTR_ERR(old_devices
);
2188 list_add(&old_devices
->list
, &fs_uuids
);
2190 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
2191 seed_devices
->opened
= 1;
2192 INIT_LIST_HEAD(&seed_devices
->devices
);
2193 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
2194 mutex_init(&seed_devices
->device_list_mutex
);
2196 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2197 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
2199 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
)
2200 device
->fs_devices
= seed_devices
;
2202 mutex_lock(&fs_info
->chunk_mutex
);
2203 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
2204 mutex_unlock(&fs_info
->chunk_mutex
);
2206 fs_devices
->seeding
= 0;
2207 fs_devices
->num_devices
= 0;
2208 fs_devices
->open_devices
= 0;
2209 fs_devices
->missing_devices
= 0;
2210 fs_devices
->rotating
= 0;
2211 fs_devices
->seed
= seed_devices
;
2213 generate_random_uuid(fs_devices
->fsid
);
2214 memcpy(fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2215 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2216 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2218 super_flags
= btrfs_super_flags(disk_super
) &
2219 ~BTRFS_SUPER_FLAG_SEEDING
;
2220 btrfs_set_super_flags(disk_super
, super_flags
);
2226 * Store the expected generation for seed devices in device items.
2228 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
2229 struct btrfs_fs_info
*fs_info
)
2231 struct btrfs_root
*root
= fs_info
->chunk_root
;
2232 struct btrfs_path
*path
;
2233 struct extent_buffer
*leaf
;
2234 struct btrfs_dev_item
*dev_item
;
2235 struct btrfs_device
*device
;
2236 struct btrfs_key key
;
2237 u8 fs_uuid
[BTRFS_FSID_SIZE
];
2238 u8 dev_uuid
[BTRFS_UUID_SIZE
];
2242 path
= btrfs_alloc_path();
2246 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2248 key
.type
= BTRFS_DEV_ITEM_KEY
;
2251 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2255 leaf
= path
->nodes
[0];
2257 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
2258 ret
= btrfs_next_leaf(root
, path
);
2263 leaf
= path
->nodes
[0];
2264 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2265 btrfs_release_path(path
);
2269 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2270 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
2271 key
.type
!= BTRFS_DEV_ITEM_KEY
)
2274 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2275 struct btrfs_dev_item
);
2276 devid
= btrfs_device_id(leaf
, dev_item
);
2277 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
2279 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
2281 device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, fs_uuid
);
2282 BUG_ON(!device
); /* Logic error */
2284 if (device
->fs_devices
->seeding
) {
2285 btrfs_set_device_generation(leaf
, dev_item
,
2286 device
->generation
);
2287 btrfs_mark_buffer_dirty(leaf
);
2295 btrfs_free_path(path
);
2299 int btrfs_init_new_device(struct btrfs_fs_info
*fs_info
, const char *device_path
)
2301 struct btrfs_root
*root
= fs_info
->dev_root
;
2302 struct request_queue
*q
;
2303 struct btrfs_trans_handle
*trans
;
2304 struct btrfs_device
*device
;
2305 struct block_device
*bdev
;
2306 struct list_head
*devices
;
2307 struct super_block
*sb
= fs_info
->sb
;
2308 struct rcu_string
*name
;
2310 int seeding_dev
= 0;
2312 bool unlocked
= false;
2314 if (sb_rdonly(sb
) && !fs_info
->fs_devices
->seeding
)
2317 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2318 fs_info
->bdev_holder
);
2320 return PTR_ERR(bdev
);
2322 if (fs_info
->fs_devices
->seeding
) {
2324 down_write(&sb
->s_umount
);
2325 mutex_lock(&uuid_mutex
);
2328 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2330 devices
= &fs_info
->fs_devices
->devices
;
2332 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2333 list_for_each_entry(device
, devices
, dev_list
) {
2334 if (device
->bdev
== bdev
) {
2337 &fs_info
->fs_devices
->device_list_mutex
);
2341 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2343 device
= btrfs_alloc_device(fs_info
, NULL
, NULL
);
2344 if (IS_ERR(device
)) {
2345 /* we can safely leave the fs_devices entry around */
2346 ret
= PTR_ERR(device
);
2350 name
= rcu_string_strdup(device_path
, GFP_KERNEL
);
2356 rcu_assign_pointer(device
->name
, name
);
2358 trans
= btrfs_start_transaction(root
, 0);
2359 if (IS_ERR(trans
)) {
2360 rcu_string_free(device
->name
);
2362 ret
= PTR_ERR(trans
);
2366 q
= bdev_get_queue(bdev
);
2367 if (blk_queue_discard(q
))
2368 device
->can_discard
= 1;
2369 device
->writeable
= 1;
2370 device
->generation
= trans
->transid
;
2371 device
->io_width
= fs_info
->sectorsize
;
2372 device
->io_align
= fs_info
->sectorsize
;
2373 device
->sector_size
= fs_info
->sectorsize
;
2374 device
->total_bytes
= round_down(i_size_read(bdev
->bd_inode
),
2375 fs_info
->sectorsize
);
2376 device
->disk_total_bytes
= device
->total_bytes
;
2377 device
->commit_total_bytes
= device
->total_bytes
;
2378 device
->fs_info
= fs_info
;
2379 device
->bdev
= bdev
;
2380 device
->in_fs_metadata
= 1;
2381 device
->is_tgtdev_for_dev_replace
= 0;
2382 device
->mode
= FMODE_EXCL
;
2383 device
->dev_stats_valid
= 1;
2384 set_blocksize(device
->bdev
, BTRFS_BDEV_BLOCKSIZE
);
2387 sb
->s_flags
&= ~SB_RDONLY
;
2388 ret
= btrfs_prepare_sprout(fs_info
);
2390 btrfs_abort_transaction(trans
, ret
);
2395 device
->fs_devices
= fs_info
->fs_devices
;
2397 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2398 mutex_lock(&fs_info
->chunk_mutex
);
2399 list_add_rcu(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2400 list_add(&device
->dev_alloc_list
,
2401 &fs_info
->fs_devices
->alloc_list
);
2402 fs_info
->fs_devices
->num_devices
++;
2403 fs_info
->fs_devices
->open_devices
++;
2404 fs_info
->fs_devices
->rw_devices
++;
2405 fs_info
->fs_devices
->total_devices
++;
2406 fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
2408 atomic64_add(device
->total_bytes
, &fs_info
->free_chunk_space
);
2410 if (!blk_queue_nonrot(q
))
2411 fs_info
->fs_devices
->rotating
= 1;
2413 tmp
= btrfs_super_total_bytes(fs_info
->super_copy
);
2414 btrfs_set_super_total_bytes(fs_info
->super_copy
,
2415 round_down(tmp
+ device
->total_bytes
, fs_info
->sectorsize
));
2417 tmp
= btrfs_super_num_devices(fs_info
->super_copy
);
2418 btrfs_set_super_num_devices(fs_info
->super_copy
, tmp
+ 1);
2420 /* add sysfs device entry */
2421 btrfs_sysfs_add_device_link(fs_info
->fs_devices
, device
);
2424 * we've got more storage, clear any full flags on the space
2427 btrfs_clear_space_info_full(fs_info
);
2429 mutex_unlock(&fs_info
->chunk_mutex
);
2430 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2433 mutex_lock(&fs_info
->chunk_mutex
);
2434 ret
= init_first_rw_device(trans
, fs_info
);
2435 mutex_unlock(&fs_info
->chunk_mutex
);
2437 btrfs_abort_transaction(trans
, ret
);
2442 ret
= btrfs_add_device(trans
, fs_info
, device
);
2444 btrfs_abort_transaction(trans
, ret
);
2449 char fsid_buf
[BTRFS_UUID_UNPARSED_SIZE
];
2451 ret
= btrfs_finish_sprout(trans
, fs_info
);
2453 btrfs_abort_transaction(trans
, ret
);
2457 /* Sprouting would change fsid of the mounted root,
2458 * so rename the fsid on the sysfs
2460 snprintf(fsid_buf
, BTRFS_UUID_UNPARSED_SIZE
, "%pU",
2462 if (kobject_rename(&fs_info
->fs_devices
->fsid_kobj
, fsid_buf
))
2464 "sysfs: failed to create fsid for sprout");
2467 ret
= btrfs_commit_transaction(trans
);
2470 mutex_unlock(&uuid_mutex
);
2471 up_write(&sb
->s_umount
);
2474 if (ret
) /* transaction commit */
2477 ret
= btrfs_relocate_sys_chunks(fs_info
);
2479 btrfs_handle_fs_error(fs_info
, ret
,
2480 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2481 trans
= btrfs_attach_transaction(root
);
2482 if (IS_ERR(trans
)) {
2483 if (PTR_ERR(trans
) == -ENOENT
)
2485 ret
= PTR_ERR(trans
);
2489 ret
= btrfs_commit_transaction(trans
);
2492 /* Update ctime/mtime for libblkid */
2493 update_dev_time(device_path
);
2497 btrfs_sysfs_rm_device_link(fs_info
->fs_devices
, device
);
2500 sb
->s_flags
|= SB_RDONLY
;
2502 btrfs_end_transaction(trans
);
2503 rcu_string_free(device
->name
);
2506 blkdev_put(bdev
, FMODE_EXCL
);
2507 if (seeding_dev
&& !unlocked
) {
2508 mutex_unlock(&uuid_mutex
);
2509 up_write(&sb
->s_umount
);
2514 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
2515 const char *device_path
,
2516 struct btrfs_device
*srcdev
,
2517 struct btrfs_device
**device_out
)
2519 struct request_queue
*q
;
2520 struct btrfs_device
*device
;
2521 struct block_device
*bdev
;
2522 struct list_head
*devices
;
2523 struct rcu_string
*name
;
2524 u64 devid
= BTRFS_DEV_REPLACE_DEVID
;
2528 if (fs_info
->fs_devices
->seeding
) {
2529 btrfs_err(fs_info
, "the filesystem is a seed filesystem!");
2533 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2534 fs_info
->bdev_holder
);
2536 btrfs_err(fs_info
, "target device %s is invalid!", device_path
);
2537 return PTR_ERR(bdev
);
2540 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2542 devices
= &fs_info
->fs_devices
->devices
;
2543 list_for_each_entry(device
, devices
, dev_list
) {
2544 if (device
->bdev
== bdev
) {
2546 "target device is in the filesystem!");
2553 if (i_size_read(bdev
->bd_inode
) <
2554 btrfs_device_get_total_bytes(srcdev
)) {
2556 "target device is smaller than source device!");
2562 device
= btrfs_alloc_device(NULL
, &devid
, NULL
);
2563 if (IS_ERR(device
)) {
2564 ret
= PTR_ERR(device
);
2568 name
= rcu_string_strdup(device_path
, GFP_KERNEL
);
2574 rcu_assign_pointer(device
->name
, name
);
2576 q
= bdev_get_queue(bdev
);
2577 if (blk_queue_discard(q
))
2578 device
->can_discard
= 1;
2579 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2580 device
->writeable
= 1;
2581 device
->generation
= 0;
2582 device
->io_width
= fs_info
->sectorsize
;
2583 device
->io_align
= fs_info
->sectorsize
;
2584 device
->sector_size
= fs_info
->sectorsize
;
2585 device
->total_bytes
= btrfs_device_get_total_bytes(srcdev
);
2586 device
->disk_total_bytes
= btrfs_device_get_disk_total_bytes(srcdev
);
2587 device
->bytes_used
= btrfs_device_get_bytes_used(srcdev
);
2588 ASSERT(list_empty(&srcdev
->resized_list
));
2589 device
->commit_total_bytes
= srcdev
->commit_total_bytes
;
2590 device
->commit_bytes_used
= device
->bytes_used
;
2591 device
->fs_info
= fs_info
;
2592 device
->bdev
= bdev
;
2593 device
->in_fs_metadata
= 1;
2594 device
->is_tgtdev_for_dev_replace
= 1;
2595 device
->mode
= FMODE_EXCL
;
2596 device
->dev_stats_valid
= 1;
2597 set_blocksize(device
->bdev
, BTRFS_BDEV_BLOCKSIZE
);
2598 device
->fs_devices
= fs_info
->fs_devices
;
2599 list_add(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2600 fs_info
->fs_devices
->num_devices
++;
2601 fs_info
->fs_devices
->open_devices
++;
2602 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2604 *device_out
= device
;
2608 blkdev_put(bdev
, FMODE_EXCL
);
2612 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info
*fs_info
,
2613 struct btrfs_device
*tgtdev
)
2615 u32 sectorsize
= fs_info
->sectorsize
;
2617 WARN_ON(fs_info
->fs_devices
->rw_devices
== 0);
2618 tgtdev
->io_width
= sectorsize
;
2619 tgtdev
->io_align
= sectorsize
;
2620 tgtdev
->sector_size
= sectorsize
;
2621 tgtdev
->fs_info
= fs_info
;
2622 tgtdev
->in_fs_metadata
= 1;
2625 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2626 struct btrfs_device
*device
)
2629 struct btrfs_path
*path
;
2630 struct btrfs_root
*root
= device
->fs_info
->chunk_root
;
2631 struct btrfs_dev_item
*dev_item
;
2632 struct extent_buffer
*leaf
;
2633 struct btrfs_key key
;
2635 path
= btrfs_alloc_path();
2639 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2640 key
.type
= BTRFS_DEV_ITEM_KEY
;
2641 key
.offset
= device
->devid
;
2643 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2652 leaf
= path
->nodes
[0];
2653 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2655 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2656 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2657 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2658 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2659 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2660 btrfs_set_device_total_bytes(leaf
, dev_item
,
2661 btrfs_device_get_disk_total_bytes(device
));
2662 btrfs_set_device_bytes_used(leaf
, dev_item
,
2663 btrfs_device_get_bytes_used(device
));
2664 btrfs_mark_buffer_dirty(leaf
);
2667 btrfs_free_path(path
);
2671 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2672 struct btrfs_device
*device
, u64 new_size
)
2674 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
2675 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
2676 struct btrfs_fs_devices
*fs_devices
;
2680 if (!device
->writeable
)
2683 new_size
= round_down(new_size
, fs_info
->sectorsize
);
2685 mutex_lock(&fs_info
->chunk_mutex
);
2686 old_total
= btrfs_super_total_bytes(super_copy
);
2687 diff
= round_down(new_size
- device
->total_bytes
, fs_info
->sectorsize
);
2689 if (new_size
<= device
->total_bytes
||
2690 device
->is_tgtdev_for_dev_replace
) {
2691 mutex_unlock(&fs_info
->chunk_mutex
);
2695 fs_devices
= fs_info
->fs_devices
;
2697 btrfs_set_super_total_bytes(super_copy
,
2698 round_down(old_total
+ diff
, fs_info
->sectorsize
));
2699 device
->fs_devices
->total_rw_bytes
+= diff
;
2701 btrfs_device_set_total_bytes(device
, new_size
);
2702 btrfs_device_set_disk_total_bytes(device
, new_size
);
2703 btrfs_clear_space_info_full(device
->fs_info
);
2704 if (list_empty(&device
->resized_list
))
2705 list_add_tail(&device
->resized_list
,
2706 &fs_devices
->resized_devices
);
2707 mutex_unlock(&fs_info
->chunk_mutex
);
2709 return btrfs_update_device(trans
, device
);
2712 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
2713 struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2715 struct btrfs_root
*root
= fs_info
->chunk_root
;
2717 struct btrfs_path
*path
;
2718 struct btrfs_key key
;
2720 path
= btrfs_alloc_path();
2724 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2725 key
.offset
= chunk_offset
;
2726 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2728 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2731 else if (ret
> 0) { /* Logic error or corruption */
2732 btrfs_handle_fs_error(fs_info
, -ENOENT
,
2733 "Failed lookup while freeing chunk.");
2738 ret
= btrfs_del_item(trans
, root
, path
);
2740 btrfs_handle_fs_error(fs_info
, ret
,
2741 "Failed to delete chunk item.");
2743 btrfs_free_path(path
);
2747 static int btrfs_del_sys_chunk(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2749 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
2750 struct btrfs_disk_key
*disk_key
;
2751 struct btrfs_chunk
*chunk
;
2758 struct btrfs_key key
;
2760 mutex_lock(&fs_info
->chunk_mutex
);
2761 array_size
= btrfs_super_sys_array_size(super_copy
);
2763 ptr
= super_copy
->sys_chunk_array
;
2766 while (cur
< array_size
) {
2767 disk_key
= (struct btrfs_disk_key
*)ptr
;
2768 btrfs_disk_key_to_cpu(&key
, disk_key
);
2770 len
= sizeof(*disk_key
);
2772 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2773 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2774 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2775 len
+= btrfs_chunk_item_size(num_stripes
);
2780 if (key
.objectid
== BTRFS_FIRST_CHUNK_TREE_OBJECTID
&&
2781 key
.offset
== chunk_offset
) {
2782 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2784 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2790 mutex_unlock(&fs_info
->chunk_mutex
);
2794 static struct extent_map
*get_chunk_map(struct btrfs_fs_info
*fs_info
,
2795 u64 logical
, u64 length
)
2797 struct extent_map_tree
*em_tree
;
2798 struct extent_map
*em
;
2800 em_tree
= &fs_info
->mapping_tree
.map_tree
;
2801 read_lock(&em_tree
->lock
);
2802 em
= lookup_extent_mapping(em_tree
, logical
, length
);
2803 read_unlock(&em_tree
->lock
);
2806 btrfs_crit(fs_info
, "unable to find logical %llu length %llu",
2808 return ERR_PTR(-EINVAL
);
2811 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
2813 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2814 logical
, length
, em
->start
, em
->start
+ em
->len
);
2815 free_extent_map(em
);
2816 return ERR_PTR(-EINVAL
);
2819 /* callers are responsible for dropping em's ref. */
2823 int btrfs_remove_chunk(struct btrfs_trans_handle
*trans
,
2824 struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2826 struct extent_map
*em
;
2827 struct map_lookup
*map
;
2828 u64 dev_extent_len
= 0;
2830 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2832 em
= get_chunk_map(fs_info
, chunk_offset
, 1);
2835 * This is a logic error, but we don't want to just rely on the
2836 * user having built with ASSERT enabled, so if ASSERT doesn't
2837 * do anything we still error out.
2842 map
= em
->map_lookup
;
2843 mutex_lock(&fs_info
->chunk_mutex
);
2844 check_system_chunk(trans
, fs_info
, map
->type
);
2845 mutex_unlock(&fs_info
->chunk_mutex
);
2848 * Take the device list mutex to prevent races with the final phase of
2849 * a device replace operation that replaces the device object associated
2850 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2852 mutex_lock(&fs_devices
->device_list_mutex
);
2853 for (i
= 0; i
< map
->num_stripes
; i
++) {
2854 struct btrfs_device
*device
= map
->stripes
[i
].dev
;
2855 ret
= btrfs_free_dev_extent(trans
, device
,
2856 map
->stripes
[i
].physical
,
2859 mutex_unlock(&fs_devices
->device_list_mutex
);
2860 btrfs_abort_transaction(trans
, ret
);
2864 if (device
->bytes_used
> 0) {
2865 mutex_lock(&fs_info
->chunk_mutex
);
2866 btrfs_device_set_bytes_used(device
,
2867 device
->bytes_used
- dev_extent_len
);
2868 atomic64_add(dev_extent_len
, &fs_info
->free_chunk_space
);
2869 btrfs_clear_space_info_full(fs_info
);
2870 mutex_unlock(&fs_info
->chunk_mutex
);
2873 if (map
->stripes
[i
].dev
) {
2874 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2876 mutex_unlock(&fs_devices
->device_list_mutex
);
2877 btrfs_abort_transaction(trans
, ret
);
2882 mutex_unlock(&fs_devices
->device_list_mutex
);
2884 ret
= btrfs_free_chunk(trans
, fs_info
, chunk_offset
);
2886 btrfs_abort_transaction(trans
, ret
);
2890 trace_btrfs_chunk_free(fs_info
, map
, chunk_offset
, em
->len
);
2892 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2893 ret
= btrfs_del_sys_chunk(fs_info
, chunk_offset
);
2895 btrfs_abort_transaction(trans
, ret
);
2900 ret
= btrfs_remove_block_group(trans
, fs_info
, chunk_offset
, em
);
2902 btrfs_abort_transaction(trans
, ret
);
2908 free_extent_map(em
);
2912 static int btrfs_relocate_chunk(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2914 struct btrfs_root
*root
= fs_info
->chunk_root
;
2915 struct btrfs_trans_handle
*trans
;
2919 * Prevent races with automatic removal of unused block groups.
2920 * After we relocate and before we remove the chunk with offset
2921 * chunk_offset, automatic removal of the block group can kick in,
2922 * resulting in a failure when calling btrfs_remove_chunk() below.
2924 * Make sure to acquire this mutex before doing a tree search (dev
2925 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2926 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2927 * we release the path used to search the chunk/dev tree and before
2928 * the current task acquires this mutex and calls us.
2930 ASSERT(mutex_is_locked(&fs_info
->delete_unused_bgs_mutex
));
2932 ret
= btrfs_can_relocate(fs_info
, chunk_offset
);
2936 /* step one, relocate all the extents inside this chunk */
2937 btrfs_scrub_pause(fs_info
);
2938 ret
= btrfs_relocate_block_group(fs_info
, chunk_offset
);
2939 btrfs_scrub_continue(fs_info
);
2943 trans
= btrfs_start_trans_remove_block_group(root
->fs_info
,
2945 if (IS_ERR(trans
)) {
2946 ret
= PTR_ERR(trans
);
2947 btrfs_handle_fs_error(root
->fs_info
, ret
, NULL
);
2952 * step two, delete the device extents and the
2953 * chunk tree entries
2955 ret
= btrfs_remove_chunk(trans
, fs_info
, chunk_offset
);
2956 btrfs_end_transaction(trans
);
2960 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info
*fs_info
)
2962 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
2963 struct btrfs_path
*path
;
2964 struct extent_buffer
*leaf
;
2965 struct btrfs_chunk
*chunk
;
2966 struct btrfs_key key
;
2967 struct btrfs_key found_key
;
2969 bool retried
= false;
2973 path
= btrfs_alloc_path();
2978 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2979 key
.offset
= (u64
)-1;
2980 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2983 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
2984 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2986 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
2989 BUG_ON(ret
== 0); /* Corruption */
2991 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
2994 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3000 leaf
= path
->nodes
[0];
3001 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3003 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
3004 struct btrfs_chunk
);
3005 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3006 btrfs_release_path(path
);
3008 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3009 ret
= btrfs_relocate_chunk(fs_info
, found_key
.offset
);
3015 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3017 if (found_key
.offset
== 0)
3019 key
.offset
= found_key
.offset
- 1;
3022 if (failed
&& !retried
) {
3026 } else if (WARN_ON(failed
&& retried
)) {
3030 btrfs_free_path(path
);
3034 static int insert_balance_item(struct btrfs_fs_info
*fs_info
,
3035 struct btrfs_balance_control
*bctl
)
3037 struct btrfs_root
*root
= fs_info
->tree_root
;
3038 struct btrfs_trans_handle
*trans
;
3039 struct btrfs_balance_item
*item
;
3040 struct btrfs_disk_balance_args disk_bargs
;
3041 struct btrfs_path
*path
;
3042 struct extent_buffer
*leaf
;
3043 struct btrfs_key key
;
3046 path
= btrfs_alloc_path();
3050 trans
= btrfs_start_transaction(root
, 0);
3051 if (IS_ERR(trans
)) {
3052 btrfs_free_path(path
);
3053 return PTR_ERR(trans
);
3056 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3057 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3060 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
3065 leaf
= path
->nodes
[0];
3066 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3068 memzero_extent_buffer(leaf
, (unsigned long)item
, sizeof(*item
));
3070 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
3071 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
3072 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
3073 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
3074 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
3075 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
3077 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
3079 btrfs_mark_buffer_dirty(leaf
);
3081 btrfs_free_path(path
);
3082 err
= btrfs_commit_transaction(trans
);
3088 static int del_balance_item(struct btrfs_fs_info
*fs_info
)
3090 struct btrfs_root
*root
= fs_info
->tree_root
;
3091 struct btrfs_trans_handle
*trans
;
3092 struct btrfs_path
*path
;
3093 struct btrfs_key key
;
3096 path
= btrfs_alloc_path();
3100 trans
= btrfs_start_transaction(root
, 0);
3101 if (IS_ERR(trans
)) {
3102 btrfs_free_path(path
);
3103 return PTR_ERR(trans
);
3106 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3107 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3110 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3118 ret
= btrfs_del_item(trans
, root
, path
);
3120 btrfs_free_path(path
);
3121 err
= btrfs_commit_transaction(trans
);
3128 * This is a heuristic used to reduce the number of chunks balanced on
3129 * resume after balance was interrupted.
3131 static void update_balance_args(struct btrfs_balance_control
*bctl
)
3134 * Turn on soft mode for chunk types that were being converted.
3136 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3137 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3138 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3139 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3140 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3141 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3144 * Turn on usage filter if is not already used. The idea is
3145 * that chunks that we have already balanced should be
3146 * reasonably full. Don't do it for chunks that are being
3147 * converted - that will keep us from relocating unconverted
3148 * (albeit full) chunks.
3150 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3151 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3152 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3153 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3154 bctl
->data
.usage
= 90;
3156 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3157 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3158 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3159 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3160 bctl
->sys
.usage
= 90;
3162 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3163 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3164 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3165 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3166 bctl
->meta
.usage
= 90;
3171 * Should be called with both balance and volume mutexes held to
3172 * serialize other volume operations (add_dev/rm_dev/resize) with
3173 * restriper. Same goes for unset_balance_control.
3175 static void set_balance_control(struct btrfs_balance_control
*bctl
)
3177 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3179 BUG_ON(fs_info
->balance_ctl
);
3181 spin_lock(&fs_info
->balance_lock
);
3182 fs_info
->balance_ctl
= bctl
;
3183 spin_unlock(&fs_info
->balance_lock
);
3186 static void unset_balance_control(struct btrfs_fs_info
*fs_info
)
3188 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3190 BUG_ON(!fs_info
->balance_ctl
);
3192 spin_lock(&fs_info
->balance_lock
);
3193 fs_info
->balance_ctl
= NULL
;
3194 spin_unlock(&fs_info
->balance_lock
);
3200 * Balance filters. Return 1 if chunk should be filtered out
3201 * (should not be balanced).
3203 static int chunk_profiles_filter(u64 chunk_type
,
3204 struct btrfs_balance_args
*bargs
)
3206 chunk_type
= chunk_to_extended(chunk_type
) &
3207 BTRFS_EXTENDED_PROFILE_MASK
;
3209 if (bargs
->profiles
& chunk_type
)
3215 static int chunk_usage_range_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
3216 struct btrfs_balance_args
*bargs
)
3218 struct btrfs_block_group_cache
*cache
;
3220 u64 user_thresh_min
;
3221 u64 user_thresh_max
;
3224 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3225 chunk_used
= btrfs_block_group_used(&cache
->item
);
3227 if (bargs
->usage_min
== 0)
3228 user_thresh_min
= 0;
3230 user_thresh_min
= div_factor_fine(cache
->key
.offset
,
3233 if (bargs
->usage_max
== 0)
3234 user_thresh_max
= 1;
3235 else if (bargs
->usage_max
> 100)
3236 user_thresh_max
= cache
->key
.offset
;
3238 user_thresh_max
= div_factor_fine(cache
->key
.offset
,
3241 if (user_thresh_min
<= chunk_used
&& chunk_used
< user_thresh_max
)
3244 btrfs_put_block_group(cache
);
3248 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
,
3249 u64 chunk_offset
, struct btrfs_balance_args
*bargs
)
3251 struct btrfs_block_group_cache
*cache
;
3252 u64 chunk_used
, user_thresh
;
3255 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3256 chunk_used
= btrfs_block_group_used(&cache
->item
);
3258 if (bargs
->usage_min
== 0)
3260 else if (bargs
->usage
> 100)
3261 user_thresh
= cache
->key
.offset
;
3263 user_thresh
= div_factor_fine(cache
->key
.offset
,
3266 if (chunk_used
< user_thresh
)
3269 btrfs_put_block_group(cache
);
3273 static int chunk_devid_filter(struct extent_buffer
*leaf
,
3274 struct btrfs_chunk
*chunk
,
3275 struct btrfs_balance_args
*bargs
)
3277 struct btrfs_stripe
*stripe
;
3278 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3281 for (i
= 0; i
< num_stripes
; i
++) {
3282 stripe
= btrfs_stripe_nr(chunk
, i
);
3283 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
3290 /* [pstart, pend) */
3291 static int chunk_drange_filter(struct extent_buffer
*leaf
,
3292 struct btrfs_chunk
*chunk
,
3293 struct btrfs_balance_args
*bargs
)
3295 struct btrfs_stripe
*stripe
;
3296 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3302 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
3305 if (btrfs_chunk_type(leaf
, chunk
) & (BTRFS_BLOCK_GROUP_DUP
|
3306 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)) {
3307 factor
= num_stripes
/ 2;
3308 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID5
) {
3309 factor
= num_stripes
- 1;
3310 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID6
) {
3311 factor
= num_stripes
- 2;
3313 factor
= num_stripes
;
3316 for (i
= 0; i
< num_stripes
; i
++) {
3317 stripe
= btrfs_stripe_nr(chunk
, i
);
3318 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
3321 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
3322 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
3323 stripe_length
= div_u64(stripe_length
, factor
);
3325 if (stripe_offset
< bargs
->pend
&&
3326 stripe_offset
+ stripe_length
> bargs
->pstart
)
3333 /* [vstart, vend) */
3334 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
3335 struct btrfs_chunk
*chunk
,
3337 struct btrfs_balance_args
*bargs
)
3339 if (chunk_offset
< bargs
->vend
&&
3340 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
3341 /* at least part of the chunk is inside this vrange */
3347 static int chunk_stripes_range_filter(struct extent_buffer
*leaf
,
3348 struct btrfs_chunk
*chunk
,
3349 struct btrfs_balance_args
*bargs
)
3351 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3353 if (bargs
->stripes_min
<= num_stripes
3354 && num_stripes
<= bargs
->stripes_max
)
3360 static int chunk_soft_convert_filter(u64 chunk_type
,
3361 struct btrfs_balance_args
*bargs
)
3363 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
3366 chunk_type
= chunk_to_extended(chunk_type
) &
3367 BTRFS_EXTENDED_PROFILE_MASK
;
3369 if (bargs
->target
== chunk_type
)
3375 static int should_balance_chunk(struct btrfs_fs_info
*fs_info
,
3376 struct extent_buffer
*leaf
,
3377 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
3379 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3380 struct btrfs_balance_args
*bargs
= NULL
;
3381 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3384 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
3385 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
3389 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3390 bargs
= &bctl
->data
;
3391 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3393 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3394 bargs
= &bctl
->meta
;
3396 /* profiles filter */
3397 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
3398 chunk_profiles_filter(chunk_type
, bargs
)) {
3403 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3404 chunk_usage_filter(fs_info
, chunk_offset
, bargs
)) {
3406 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3407 chunk_usage_range_filter(fs_info
, chunk_offset
, bargs
)) {
3412 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
3413 chunk_devid_filter(leaf
, chunk
, bargs
)) {
3417 /* drange filter, makes sense only with devid filter */
3418 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
3419 chunk_drange_filter(leaf
, chunk
, bargs
)) {
3424 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
3425 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3429 /* stripes filter */
3430 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_STRIPES_RANGE
) &&
3431 chunk_stripes_range_filter(leaf
, chunk
, bargs
)) {
3435 /* soft profile changing mode */
3436 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
3437 chunk_soft_convert_filter(chunk_type
, bargs
)) {
3442 * limited by count, must be the last filter
3444 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT
)) {
3445 if (bargs
->limit
== 0)
3449 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT_RANGE
)) {
3451 * Same logic as the 'limit' filter; the minimum cannot be
3452 * determined here because we do not have the global information
3453 * about the count of all chunks that satisfy the filters.
3455 if (bargs
->limit_max
== 0)
3464 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
3466 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3467 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
3468 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
3469 struct list_head
*devices
;
3470 struct btrfs_device
*device
;
3474 struct btrfs_chunk
*chunk
;
3475 struct btrfs_path
*path
= NULL
;
3476 struct btrfs_key key
;
3477 struct btrfs_key found_key
;
3478 struct btrfs_trans_handle
*trans
;
3479 struct extent_buffer
*leaf
;
3482 int enospc_errors
= 0;
3483 bool counting
= true;
3484 /* The single value limit and min/max limits use the same bytes in the */
3485 u64 limit_data
= bctl
->data
.limit
;
3486 u64 limit_meta
= bctl
->meta
.limit
;
3487 u64 limit_sys
= bctl
->sys
.limit
;
3491 int chunk_reserved
= 0;
3494 /* step one make some room on all the devices */
3495 devices
= &fs_info
->fs_devices
->devices
;
3496 list_for_each_entry(device
, devices
, dev_list
) {
3497 old_size
= btrfs_device_get_total_bytes(device
);
3498 size_to_free
= div_factor(old_size
, 1);
3499 size_to_free
= min_t(u64
, size_to_free
, SZ_1M
);
3500 if (!device
->writeable
||
3501 btrfs_device_get_total_bytes(device
) -
3502 btrfs_device_get_bytes_used(device
) > size_to_free
||
3503 device
->is_tgtdev_for_dev_replace
)
3506 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
3510 /* btrfs_shrink_device never returns ret > 0 */
3515 trans
= btrfs_start_transaction(dev_root
, 0);
3516 if (IS_ERR(trans
)) {
3517 ret
= PTR_ERR(trans
);
3518 btrfs_info_in_rcu(fs_info
,
3519 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3520 rcu_str_deref(device
->name
), ret
,
3521 old_size
, old_size
- size_to_free
);
3525 ret
= btrfs_grow_device(trans
, device
, old_size
);
3527 btrfs_end_transaction(trans
);
3528 /* btrfs_grow_device never returns ret > 0 */
3530 btrfs_info_in_rcu(fs_info
,
3531 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3532 rcu_str_deref(device
->name
), ret
,
3533 old_size
, old_size
- size_to_free
);
3537 btrfs_end_transaction(trans
);
3540 /* step two, relocate all the chunks */
3541 path
= btrfs_alloc_path();
3547 /* zero out stat counters */
3548 spin_lock(&fs_info
->balance_lock
);
3549 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
3550 spin_unlock(&fs_info
->balance_lock
);
3554 * The single value limit and min/max limits use the same bytes
3557 bctl
->data
.limit
= limit_data
;
3558 bctl
->meta
.limit
= limit_meta
;
3559 bctl
->sys
.limit
= limit_sys
;
3561 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3562 key
.offset
= (u64
)-1;
3563 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3566 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
3567 atomic_read(&fs_info
->balance_cancel_req
)) {
3572 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
3573 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
3575 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3580 * this shouldn't happen, it means the last relocate
3584 BUG(); /* FIXME break ? */
3586 ret
= btrfs_previous_item(chunk_root
, path
, 0,
3587 BTRFS_CHUNK_ITEM_KEY
);
3589 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3594 leaf
= path
->nodes
[0];
3595 slot
= path
->slots
[0];
3596 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3598 if (found_key
.objectid
!= key
.objectid
) {
3599 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3603 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
3604 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3607 spin_lock(&fs_info
->balance_lock
);
3608 bctl
->stat
.considered
++;
3609 spin_unlock(&fs_info
->balance_lock
);
3612 ret
= should_balance_chunk(fs_info
, leaf
, chunk
,
3615 btrfs_release_path(path
);
3617 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3622 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3623 spin_lock(&fs_info
->balance_lock
);
3624 bctl
->stat
.expected
++;
3625 spin_unlock(&fs_info
->balance_lock
);
3627 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3629 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3631 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3638 * Apply limit_min filter, no need to check if the LIMITS
3639 * filter is used, limit_min is 0 by default
3641 if (((chunk_type
& BTRFS_BLOCK_GROUP_DATA
) &&
3642 count_data
< bctl
->data
.limit_min
)
3643 || ((chunk_type
& BTRFS_BLOCK_GROUP_METADATA
) &&
3644 count_meta
< bctl
->meta
.limit_min
)
3645 || ((chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) &&
3646 count_sys
< bctl
->sys
.limit_min
)) {
3647 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3651 ASSERT(fs_info
->data_sinfo
);
3652 spin_lock(&fs_info
->data_sinfo
->lock
);
3653 bytes_used
= fs_info
->data_sinfo
->bytes_used
;
3654 spin_unlock(&fs_info
->data_sinfo
->lock
);
3656 if ((chunk_type
& BTRFS_BLOCK_GROUP_DATA
) &&
3657 !chunk_reserved
&& !bytes_used
) {
3658 trans
= btrfs_start_transaction(chunk_root
, 0);
3659 if (IS_ERR(trans
)) {
3660 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3661 ret
= PTR_ERR(trans
);
3665 ret
= btrfs_force_chunk_alloc(trans
, fs_info
,
3666 BTRFS_BLOCK_GROUP_DATA
);
3667 btrfs_end_transaction(trans
);
3669 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3675 ret
= btrfs_relocate_chunk(fs_info
, found_key
.offset
);
3676 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3677 if (ret
&& ret
!= -ENOSPC
)
3679 if (ret
== -ENOSPC
) {
3682 spin_lock(&fs_info
->balance_lock
);
3683 bctl
->stat
.completed
++;
3684 spin_unlock(&fs_info
->balance_lock
);
3687 if (found_key
.offset
== 0)
3689 key
.offset
= found_key
.offset
- 1;
3693 btrfs_release_path(path
);
3698 btrfs_free_path(path
);
3699 if (enospc_errors
) {
3700 btrfs_info(fs_info
, "%d enospc errors during balance",
3710 * alloc_profile_is_valid - see if a given profile is valid and reduced
3711 * @flags: profile to validate
3712 * @extended: if true @flags is treated as an extended profile
3714 static int alloc_profile_is_valid(u64 flags
, int extended
)
3716 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
3717 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
3719 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
3721 /* 1) check that all other bits are zeroed */
3725 /* 2) see if profile is reduced */
3727 return !extended
; /* "0" is valid for usual profiles */
3729 /* true if exactly one bit set */
3730 return (flags
& (flags
- 1)) == 0;
3733 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
3735 /* cancel requested || normal exit path */
3736 return atomic_read(&fs_info
->balance_cancel_req
) ||
3737 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
3738 atomic_read(&fs_info
->balance_cancel_req
) == 0);
3741 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
3745 unset_balance_control(fs_info
);
3746 ret
= del_balance_item(fs_info
);
3748 btrfs_handle_fs_error(fs_info
, ret
, NULL
);
3750 clear_bit(BTRFS_FS_EXCL_OP
, &fs_info
->flags
);
3753 /* Non-zero return value signifies invalidity */
3754 static inline int validate_convert_profile(struct btrfs_balance_args
*bctl_arg
,
3757 return ((bctl_arg
->flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3758 (!alloc_profile_is_valid(bctl_arg
->target
, 1) ||
3759 (bctl_arg
->target
& ~allowed
)));
3763 * Should be called with both balance and volume mutexes held
3765 int btrfs_balance(struct btrfs_balance_control
*bctl
,
3766 struct btrfs_ioctl_balance_args
*bargs
)
3768 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3769 u64 meta_target
, data_target
;
3776 if (btrfs_fs_closing(fs_info
) ||
3777 atomic_read(&fs_info
->balance_pause_req
) ||
3778 atomic_read(&fs_info
->balance_cancel_req
)) {
3783 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
3784 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
3788 * In case of mixed groups both data and meta should be picked,
3789 * and identical options should be given for both of them.
3791 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
3792 if (mixed
&& (bctl
->flags
& allowed
)) {
3793 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
3794 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
3795 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
3797 "with mixed groups data and metadata balance options must be the same");
3803 num_devices
= fs_info
->fs_devices
->num_devices
;
3804 btrfs_dev_replace_lock(&fs_info
->dev_replace
, 0);
3805 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
3806 BUG_ON(num_devices
< 1);
3809 btrfs_dev_replace_unlock(&fs_info
->dev_replace
, 0);
3810 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
| BTRFS_BLOCK_GROUP_DUP
;
3811 if (num_devices
> 1)
3812 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
3813 if (num_devices
> 2)
3814 allowed
|= BTRFS_BLOCK_GROUP_RAID5
;
3815 if (num_devices
> 3)
3816 allowed
|= (BTRFS_BLOCK_GROUP_RAID10
|
3817 BTRFS_BLOCK_GROUP_RAID6
);
3818 if (validate_convert_profile(&bctl
->data
, allowed
)) {
3820 "unable to start balance with target data profile %llu",
3825 if (validate_convert_profile(&bctl
->meta
, allowed
)) {
3827 "unable to start balance with target metadata profile %llu",
3832 if (validate_convert_profile(&bctl
->sys
, allowed
)) {
3834 "unable to start balance with target system profile %llu",
3840 /* allow to reduce meta or sys integrity only if force set */
3841 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3842 BTRFS_BLOCK_GROUP_RAID10
|
3843 BTRFS_BLOCK_GROUP_RAID5
|
3844 BTRFS_BLOCK_GROUP_RAID6
;
3846 seq
= read_seqbegin(&fs_info
->profiles_lock
);
3848 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3849 (fs_info
->avail_system_alloc_bits
& allowed
) &&
3850 !(bctl
->sys
.target
& allowed
)) ||
3851 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3852 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
3853 !(bctl
->meta
.target
& allowed
))) {
3854 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
3856 "force reducing metadata integrity");
3859 "balance will reduce metadata integrity, use force if you want this");
3864 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
3866 /* if we're not converting, the target field is uninitialized */
3867 meta_target
= (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) ?
3868 bctl
->meta
.target
: fs_info
->avail_metadata_alloc_bits
;
3869 data_target
= (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) ?
3870 bctl
->data
.target
: fs_info
->avail_data_alloc_bits
;
3871 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target
) <
3872 btrfs_get_num_tolerated_disk_barrier_failures(data_target
)) {
3874 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3875 meta_target
, data_target
);
3878 ret
= insert_balance_item(fs_info
, bctl
);
3879 if (ret
&& ret
!= -EEXIST
)
3882 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
3883 BUG_ON(ret
== -EEXIST
);
3884 set_balance_control(bctl
);
3886 BUG_ON(ret
!= -EEXIST
);
3887 spin_lock(&fs_info
->balance_lock
);
3888 update_balance_args(bctl
);
3889 spin_unlock(&fs_info
->balance_lock
);
3892 atomic_inc(&fs_info
->balance_running
);
3893 mutex_unlock(&fs_info
->balance_mutex
);
3895 ret
= __btrfs_balance(fs_info
);
3897 mutex_lock(&fs_info
->balance_mutex
);
3898 atomic_dec(&fs_info
->balance_running
);
3901 memset(bargs
, 0, sizeof(*bargs
));
3902 update_ioctl_balance_args(fs_info
, 0, bargs
);
3905 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
3906 balance_need_close(fs_info
)) {
3907 __cancel_balance(fs_info
);
3910 wake_up(&fs_info
->balance_wait_q
);
3914 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
3915 __cancel_balance(fs_info
);
3918 clear_bit(BTRFS_FS_EXCL_OP
, &fs_info
->flags
);
3923 static int balance_kthread(void *data
)
3925 struct btrfs_fs_info
*fs_info
= data
;
3928 mutex_lock(&fs_info
->volume_mutex
);
3929 mutex_lock(&fs_info
->balance_mutex
);
3931 if (fs_info
->balance_ctl
) {
3932 btrfs_info(fs_info
, "continuing balance");
3933 ret
= btrfs_balance(fs_info
->balance_ctl
, NULL
);
3936 mutex_unlock(&fs_info
->balance_mutex
);
3937 mutex_unlock(&fs_info
->volume_mutex
);
3942 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
3944 struct task_struct
*tsk
;
3946 spin_lock(&fs_info
->balance_lock
);
3947 if (!fs_info
->balance_ctl
) {
3948 spin_unlock(&fs_info
->balance_lock
);
3951 spin_unlock(&fs_info
->balance_lock
);
3953 if (btrfs_test_opt(fs_info
, SKIP_BALANCE
)) {
3954 btrfs_info(fs_info
, "force skipping balance");
3958 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
3959 return PTR_ERR_OR_ZERO(tsk
);
3962 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
3964 struct btrfs_balance_control
*bctl
;
3965 struct btrfs_balance_item
*item
;
3966 struct btrfs_disk_balance_args disk_bargs
;
3967 struct btrfs_path
*path
;
3968 struct extent_buffer
*leaf
;
3969 struct btrfs_key key
;
3972 path
= btrfs_alloc_path();
3976 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3977 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3980 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
3983 if (ret
> 0) { /* ret = -ENOENT; */
3988 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
3994 leaf
= path
->nodes
[0];
3995 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3997 bctl
->fs_info
= fs_info
;
3998 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
3999 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
4001 btrfs_balance_data(leaf
, item
, &disk_bargs
);
4002 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
4003 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
4004 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
4005 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
4006 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
4008 WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP
, &fs_info
->flags
));
4010 mutex_lock(&fs_info
->volume_mutex
);
4011 mutex_lock(&fs_info
->balance_mutex
);
4013 set_balance_control(bctl
);
4015 mutex_unlock(&fs_info
->balance_mutex
);
4016 mutex_unlock(&fs_info
->volume_mutex
);
4018 btrfs_free_path(path
);
4022 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
4026 mutex_lock(&fs_info
->balance_mutex
);
4027 if (!fs_info
->balance_ctl
) {
4028 mutex_unlock(&fs_info
->balance_mutex
);
4032 if (atomic_read(&fs_info
->balance_running
)) {
4033 atomic_inc(&fs_info
->balance_pause_req
);
4034 mutex_unlock(&fs_info
->balance_mutex
);
4036 wait_event(fs_info
->balance_wait_q
,
4037 atomic_read(&fs_info
->balance_running
) == 0);
4039 mutex_lock(&fs_info
->balance_mutex
);
4040 /* we are good with balance_ctl ripped off from under us */
4041 BUG_ON(atomic_read(&fs_info
->balance_running
));
4042 atomic_dec(&fs_info
->balance_pause_req
);
4047 mutex_unlock(&fs_info
->balance_mutex
);
4051 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
4053 if (sb_rdonly(fs_info
->sb
))
4056 mutex_lock(&fs_info
->balance_mutex
);
4057 if (!fs_info
->balance_ctl
) {
4058 mutex_unlock(&fs_info
->balance_mutex
);
4062 atomic_inc(&fs_info
->balance_cancel_req
);
4064 * if we are running just wait and return, balance item is
4065 * deleted in btrfs_balance in this case
4067 if (atomic_read(&fs_info
->balance_running
)) {
4068 mutex_unlock(&fs_info
->balance_mutex
);
4069 wait_event(fs_info
->balance_wait_q
,
4070 atomic_read(&fs_info
->balance_running
) == 0);
4071 mutex_lock(&fs_info
->balance_mutex
);
4073 /* __cancel_balance needs volume_mutex */
4074 mutex_unlock(&fs_info
->balance_mutex
);
4075 mutex_lock(&fs_info
->volume_mutex
);
4076 mutex_lock(&fs_info
->balance_mutex
);
4078 if (fs_info
->balance_ctl
)
4079 __cancel_balance(fs_info
);
4081 mutex_unlock(&fs_info
->volume_mutex
);
4084 BUG_ON(fs_info
->balance_ctl
|| atomic_read(&fs_info
->balance_running
));
4085 atomic_dec(&fs_info
->balance_cancel_req
);
4086 mutex_unlock(&fs_info
->balance_mutex
);
4090 static int btrfs_uuid_scan_kthread(void *data
)
4092 struct btrfs_fs_info
*fs_info
= data
;
4093 struct btrfs_root
*root
= fs_info
->tree_root
;
4094 struct btrfs_key key
;
4095 struct btrfs_path
*path
= NULL
;
4097 struct extent_buffer
*eb
;
4099 struct btrfs_root_item root_item
;
4101 struct btrfs_trans_handle
*trans
= NULL
;
4103 path
= btrfs_alloc_path();
4110 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4114 ret
= btrfs_search_forward(root
, &key
, path
, 0);
4121 if (key
.type
!= BTRFS_ROOT_ITEM_KEY
||
4122 (key
.objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
4123 key
.objectid
!= BTRFS_FS_TREE_OBJECTID
) ||
4124 key
.objectid
> BTRFS_LAST_FREE_OBJECTID
)
4127 eb
= path
->nodes
[0];
4128 slot
= path
->slots
[0];
4129 item_size
= btrfs_item_size_nr(eb
, slot
);
4130 if (item_size
< sizeof(root_item
))
4133 read_extent_buffer(eb
, &root_item
,
4134 btrfs_item_ptr_offset(eb
, slot
),
4135 (int)sizeof(root_item
));
4136 if (btrfs_root_refs(&root_item
) == 0)
4139 if (!btrfs_is_empty_uuid(root_item
.uuid
) ||
4140 !btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4144 btrfs_release_path(path
);
4146 * 1 - subvol uuid item
4147 * 1 - received_subvol uuid item
4149 trans
= btrfs_start_transaction(fs_info
->uuid_root
, 2);
4150 if (IS_ERR(trans
)) {
4151 ret
= PTR_ERR(trans
);
4159 if (!btrfs_is_empty_uuid(root_item
.uuid
)) {
4160 ret
= btrfs_uuid_tree_add(trans
, fs_info
,
4162 BTRFS_UUID_KEY_SUBVOL
,
4165 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4171 if (!btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4172 ret
= btrfs_uuid_tree_add(trans
, fs_info
,
4173 root_item
.received_uuid
,
4174 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
4177 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4185 ret
= btrfs_end_transaction(trans
);
4191 btrfs_release_path(path
);
4192 if (key
.offset
< (u64
)-1) {
4194 } else if (key
.type
< BTRFS_ROOT_ITEM_KEY
) {
4196 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4197 } else if (key
.objectid
< (u64
)-1) {
4199 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4208 btrfs_free_path(path
);
4209 if (trans
&& !IS_ERR(trans
))
4210 btrfs_end_transaction(trans
);
4212 btrfs_warn(fs_info
, "btrfs_uuid_scan_kthread failed %d", ret
);
4214 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN
, &fs_info
->flags
);
4215 up(&fs_info
->uuid_tree_rescan_sem
);
4220 * Callback for btrfs_uuid_tree_iterate().
4222 * 0 check succeeded, the entry is not outdated.
4223 * < 0 if an error occurred.
4224 * > 0 if the check failed, which means the caller shall remove the entry.
4226 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info
*fs_info
,
4227 u8
*uuid
, u8 type
, u64 subid
)
4229 struct btrfs_key key
;
4231 struct btrfs_root
*subvol_root
;
4233 if (type
!= BTRFS_UUID_KEY_SUBVOL
&&
4234 type
!= BTRFS_UUID_KEY_RECEIVED_SUBVOL
)
4237 key
.objectid
= subid
;
4238 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4239 key
.offset
= (u64
)-1;
4240 subvol_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
4241 if (IS_ERR(subvol_root
)) {
4242 ret
= PTR_ERR(subvol_root
);
4249 case BTRFS_UUID_KEY_SUBVOL
:
4250 if (memcmp(uuid
, subvol_root
->root_item
.uuid
, BTRFS_UUID_SIZE
))
4253 case BTRFS_UUID_KEY_RECEIVED_SUBVOL
:
4254 if (memcmp(uuid
, subvol_root
->root_item
.received_uuid
,
4264 static int btrfs_uuid_rescan_kthread(void *data
)
4266 struct btrfs_fs_info
*fs_info
= (struct btrfs_fs_info
*)data
;
4270 * 1st step is to iterate through the existing UUID tree and
4271 * to delete all entries that contain outdated data.
4272 * 2nd step is to add all missing entries to the UUID tree.
4274 ret
= btrfs_uuid_tree_iterate(fs_info
, btrfs_check_uuid_tree_entry
);
4276 btrfs_warn(fs_info
, "iterating uuid_tree failed %d", ret
);
4277 up(&fs_info
->uuid_tree_rescan_sem
);
4280 return btrfs_uuid_scan_kthread(data
);
4283 int btrfs_create_uuid_tree(struct btrfs_fs_info
*fs_info
)
4285 struct btrfs_trans_handle
*trans
;
4286 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
4287 struct btrfs_root
*uuid_root
;
4288 struct task_struct
*task
;
4295 trans
= btrfs_start_transaction(tree_root
, 2);
4297 return PTR_ERR(trans
);
4299 uuid_root
= btrfs_create_tree(trans
, fs_info
,
4300 BTRFS_UUID_TREE_OBJECTID
);
4301 if (IS_ERR(uuid_root
)) {
4302 ret
= PTR_ERR(uuid_root
);
4303 btrfs_abort_transaction(trans
, ret
);
4304 btrfs_end_transaction(trans
);
4308 fs_info
->uuid_root
= uuid_root
;
4310 ret
= btrfs_commit_transaction(trans
);
4314 down(&fs_info
->uuid_tree_rescan_sem
);
4315 task
= kthread_run(btrfs_uuid_scan_kthread
, fs_info
, "btrfs-uuid");
4317 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4318 btrfs_warn(fs_info
, "failed to start uuid_scan task");
4319 up(&fs_info
->uuid_tree_rescan_sem
);
4320 return PTR_ERR(task
);
4326 int btrfs_check_uuid_tree(struct btrfs_fs_info
*fs_info
)
4328 struct task_struct
*task
;
4330 down(&fs_info
->uuid_tree_rescan_sem
);
4331 task
= kthread_run(btrfs_uuid_rescan_kthread
, fs_info
, "btrfs-uuid");
4333 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4334 btrfs_warn(fs_info
, "failed to start uuid_rescan task");
4335 up(&fs_info
->uuid_tree_rescan_sem
);
4336 return PTR_ERR(task
);
4343 * shrinking a device means finding all of the device extents past
4344 * the new size, and then following the back refs to the chunks.
4345 * The chunk relocation code actually frees the device extent
4347 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
4349 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
4350 struct btrfs_root
*root
= fs_info
->dev_root
;
4351 struct btrfs_trans_handle
*trans
;
4352 struct btrfs_dev_extent
*dev_extent
= NULL
;
4353 struct btrfs_path
*path
;
4359 bool retried
= false;
4360 bool checked_pending_chunks
= false;
4361 struct extent_buffer
*l
;
4362 struct btrfs_key key
;
4363 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
4364 u64 old_total
= btrfs_super_total_bytes(super_copy
);
4365 u64 old_size
= btrfs_device_get_total_bytes(device
);
4368 new_size
= round_down(new_size
, fs_info
->sectorsize
);
4369 diff
= round_down(old_size
- new_size
, fs_info
->sectorsize
);
4371 if (device
->is_tgtdev_for_dev_replace
)
4374 path
= btrfs_alloc_path();
4378 path
->reada
= READA_FORWARD
;
4380 mutex_lock(&fs_info
->chunk_mutex
);
4382 btrfs_device_set_total_bytes(device
, new_size
);
4383 if (device
->writeable
) {
4384 device
->fs_devices
->total_rw_bytes
-= diff
;
4385 atomic64_sub(diff
, &fs_info
->free_chunk_space
);
4387 mutex_unlock(&fs_info
->chunk_mutex
);
4390 key
.objectid
= device
->devid
;
4391 key
.offset
= (u64
)-1;
4392 key
.type
= BTRFS_DEV_EXTENT_KEY
;
4395 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
4396 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4398 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4402 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
4404 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4409 btrfs_release_path(path
);
4414 slot
= path
->slots
[0];
4415 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
4417 if (key
.objectid
!= device
->devid
) {
4418 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4419 btrfs_release_path(path
);
4423 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
4424 length
= btrfs_dev_extent_length(l
, dev_extent
);
4426 if (key
.offset
+ length
<= new_size
) {
4427 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4428 btrfs_release_path(path
);
4432 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
4433 btrfs_release_path(path
);
4435 ret
= btrfs_relocate_chunk(fs_info
, chunk_offset
);
4436 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4437 if (ret
&& ret
!= -ENOSPC
)
4441 } while (key
.offset
-- > 0);
4443 if (failed
&& !retried
) {
4447 } else if (failed
&& retried
) {
4452 /* Shrinking succeeded, else we would be at "done". */
4453 trans
= btrfs_start_transaction(root
, 0);
4454 if (IS_ERR(trans
)) {
4455 ret
= PTR_ERR(trans
);
4459 mutex_lock(&fs_info
->chunk_mutex
);
4462 * We checked in the above loop all device extents that were already in
4463 * the device tree. However before we have updated the device's
4464 * total_bytes to the new size, we might have had chunk allocations that
4465 * have not complete yet (new block groups attached to transaction
4466 * handles), and therefore their device extents were not yet in the
4467 * device tree and we missed them in the loop above. So if we have any
4468 * pending chunk using a device extent that overlaps the device range
4469 * that we can not use anymore, commit the current transaction and
4470 * repeat the search on the device tree - this way we guarantee we will
4471 * not have chunks using device extents that end beyond 'new_size'.
4473 if (!checked_pending_chunks
) {
4474 u64 start
= new_size
;
4475 u64 len
= old_size
- new_size
;
4477 if (contains_pending_extent(trans
->transaction
, device
,
4479 mutex_unlock(&fs_info
->chunk_mutex
);
4480 checked_pending_chunks
= true;
4483 ret
= btrfs_commit_transaction(trans
);
4490 btrfs_device_set_disk_total_bytes(device
, new_size
);
4491 if (list_empty(&device
->resized_list
))
4492 list_add_tail(&device
->resized_list
,
4493 &fs_info
->fs_devices
->resized_devices
);
4495 WARN_ON(diff
> old_total
);
4496 btrfs_set_super_total_bytes(super_copy
,
4497 round_down(old_total
- diff
, fs_info
->sectorsize
));
4498 mutex_unlock(&fs_info
->chunk_mutex
);
4500 /* Now btrfs_update_device() will change the on-disk size. */
4501 ret
= btrfs_update_device(trans
, device
);
4502 btrfs_end_transaction(trans
);
4504 btrfs_free_path(path
);
4506 mutex_lock(&fs_info
->chunk_mutex
);
4507 btrfs_device_set_total_bytes(device
, old_size
);
4508 if (device
->writeable
)
4509 device
->fs_devices
->total_rw_bytes
+= diff
;
4510 atomic64_add(diff
, &fs_info
->free_chunk_space
);
4511 mutex_unlock(&fs_info
->chunk_mutex
);
4516 static int btrfs_add_system_chunk(struct btrfs_fs_info
*fs_info
,
4517 struct btrfs_key
*key
,
4518 struct btrfs_chunk
*chunk
, int item_size
)
4520 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
4521 struct btrfs_disk_key disk_key
;
4525 mutex_lock(&fs_info
->chunk_mutex
);
4526 array_size
= btrfs_super_sys_array_size(super_copy
);
4527 if (array_size
+ item_size
+ sizeof(disk_key
)
4528 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4529 mutex_unlock(&fs_info
->chunk_mutex
);
4533 ptr
= super_copy
->sys_chunk_array
+ array_size
;
4534 btrfs_cpu_key_to_disk(&disk_key
, key
);
4535 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
4536 ptr
+= sizeof(disk_key
);
4537 memcpy(ptr
, chunk
, item_size
);
4538 item_size
+= sizeof(disk_key
);
4539 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
4540 mutex_unlock(&fs_info
->chunk_mutex
);
4546 * sort the devices in descending order by max_avail, total_avail
4548 static int btrfs_cmp_device_info(const void *a
, const void *b
)
4550 const struct btrfs_device_info
*di_a
= a
;
4551 const struct btrfs_device_info
*di_b
= b
;
4553 if (di_a
->max_avail
> di_b
->max_avail
)
4555 if (di_a
->max_avail
< di_b
->max_avail
)
4557 if (di_a
->total_avail
> di_b
->total_avail
)
4559 if (di_a
->total_avail
< di_b
->total_avail
)
4564 static void check_raid56_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
4566 if (!(type
& BTRFS_BLOCK_GROUP_RAID56_MASK
))
4569 btrfs_set_fs_incompat(info
, RAID56
);
4572 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info) \
4573 - sizeof(struct btrfs_chunk)) \
4574 / sizeof(struct btrfs_stripe) + 1)
4576 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4577 - 2 * sizeof(struct btrfs_disk_key) \
4578 - 2 * sizeof(struct btrfs_chunk)) \
4579 / sizeof(struct btrfs_stripe) + 1)
4581 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4582 u64 start
, u64 type
)
4584 struct btrfs_fs_info
*info
= trans
->fs_info
;
4585 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
4586 struct btrfs_device
*device
;
4587 struct map_lookup
*map
= NULL
;
4588 struct extent_map_tree
*em_tree
;
4589 struct extent_map
*em
;
4590 struct btrfs_device_info
*devices_info
= NULL
;
4592 int num_stripes
; /* total number of stripes to allocate */
4593 int data_stripes
; /* number of stripes that count for
4595 int sub_stripes
; /* sub_stripes info for map */
4596 int dev_stripes
; /* stripes per dev */
4597 int devs_max
; /* max devs to use */
4598 int devs_min
; /* min devs needed */
4599 int devs_increment
; /* ndevs has to be a multiple of this */
4600 int ncopies
; /* how many copies to data has */
4602 u64 max_stripe_size
;
4611 BUG_ON(!alloc_profile_is_valid(type
, 0));
4613 if (list_empty(&fs_devices
->alloc_list
))
4616 index
= __get_raid_index(type
);
4618 sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
4619 dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
4620 devs_max
= btrfs_raid_array
[index
].devs_max
;
4621 devs_min
= btrfs_raid_array
[index
].devs_min
;
4622 devs_increment
= btrfs_raid_array
[index
].devs_increment
;
4623 ncopies
= btrfs_raid_array
[index
].ncopies
;
4625 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
4626 max_stripe_size
= SZ_1G
;
4627 max_chunk_size
= 10 * max_stripe_size
;
4629 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4630 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
4631 /* for larger filesystems, use larger metadata chunks */
4632 if (fs_devices
->total_rw_bytes
> 50ULL * SZ_1G
)
4633 max_stripe_size
= SZ_1G
;
4635 max_stripe_size
= SZ_256M
;
4636 max_chunk_size
= max_stripe_size
;
4638 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4639 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4640 max_stripe_size
= SZ_32M
;
4641 max_chunk_size
= 2 * max_stripe_size
;
4643 devs_max
= BTRFS_MAX_DEVS_SYS_CHUNK
;
4645 btrfs_err(info
, "invalid chunk type 0x%llx requested",
4650 /* we don't want a chunk larger than 10% of writeable space */
4651 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
4654 devices_info
= kcalloc(fs_devices
->rw_devices
, sizeof(*devices_info
),
4660 * in the first pass through the devices list, we gather information
4661 * about the available holes on each device.
4664 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
4668 if (!device
->writeable
) {
4670 "BTRFS: read-only device in alloc_list\n");
4674 if (!device
->in_fs_metadata
||
4675 device
->is_tgtdev_for_dev_replace
)
4678 if (device
->total_bytes
> device
->bytes_used
)
4679 total_avail
= device
->total_bytes
- device
->bytes_used
;
4683 /* If there is no space on this device, skip it. */
4684 if (total_avail
== 0)
4687 ret
= find_free_dev_extent(trans
, device
,
4688 max_stripe_size
* dev_stripes
,
4689 &dev_offset
, &max_avail
);
4690 if (ret
&& ret
!= -ENOSPC
)
4694 max_avail
= max_stripe_size
* dev_stripes
;
4696 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
4699 if (ndevs
== fs_devices
->rw_devices
) {
4700 WARN(1, "%s: found more than %llu devices\n",
4701 __func__
, fs_devices
->rw_devices
);
4704 devices_info
[ndevs
].dev_offset
= dev_offset
;
4705 devices_info
[ndevs
].max_avail
= max_avail
;
4706 devices_info
[ndevs
].total_avail
= total_avail
;
4707 devices_info
[ndevs
].dev
= device
;
4712 * now sort the devices by hole size / available space
4714 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
4715 btrfs_cmp_device_info
, NULL
);
4717 /* round down to number of usable stripes */
4718 ndevs
= round_down(ndevs
, devs_increment
);
4720 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
4725 ndevs
= min(ndevs
, devs_max
);
4728 * the primary goal is to maximize the number of stripes, so use as many
4729 * devices as possible, even if the stripes are not maximum sized.
4731 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4732 num_stripes
= ndevs
* dev_stripes
;
4735 * this will have to be fixed for RAID1 and RAID10 over
4738 data_stripes
= num_stripes
/ ncopies
;
4740 if (type
& BTRFS_BLOCK_GROUP_RAID5
)
4741 data_stripes
= num_stripes
- 1;
4743 if (type
& BTRFS_BLOCK_GROUP_RAID6
)
4744 data_stripes
= num_stripes
- 2;
4747 * Use the number of data stripes to figure out how big this chunk
4748 * is really going to be in terms of logical address space,
4749 * and compare that answer with the max chunk size
4751 if (stripe_size
* data_stripes
> max_chunk_size
) {
4752 u64 mask
= (1ULL << 24) - 1;
4754 stripe_size
= div_u64(max_chunk_size
, data_stripes
);
4756 /* bump the answer up to a 16MB boundary */
4757 stripe_size
= (stripe_size
+ mask
) & ~mask
;
4759 /* but don't go higher than the limits we found
4760 * while searching for free extents
4762 if (stripe_size
> devices_info
[ndevs
-1].max_avail
)
4763 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4766 stripe_size
= div_u64(stripe_size
, dev_stripes
);
4768 /* align to BTRFS_STRIPE_LEN */
4769 stripe_size
= round_down(stripe_size
, BTRFS_STRIPE_LEN
);
4771 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
4776 map
->num_stripes
= num_stripes
;
4778 for (i
= 0; i
< ndevs
; ++i
) {
4779 for (j
= 0; j
< dev_stripes
; ++j
) {
4780 int s
= i
* dev_stripes
+ j
;
4781 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
4782 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
4786 map
->stripe_len
= BTRFS_STRIPE_LEN
;
4787 map
->io_align
= BTRFS_STRIPE_LEN
;
4788 map
->io_width
= BTRFS_STRIPE_LEN
;
4790 map
->sub_stripes
= sub_stripes
;
4792 num_bytes
= stripe_size
* data_stripes
;
4794 trace_btrfs_chunk_alloc(info
, map
, start
, num_bytes
);
4796 em
= alloc_extent_map();
4802 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
4803 em
->map_lookup
= map
;
4805 em
->len
= num_bytes
;
4806 em
->block_start
= 0;
4807 em
->block_len
= em
->len
;
4808 em
->orig_block_len
= stripe_size
;
4810 em_tree
= &info
->mapping_tree
.map_tree
;
4811 write_lock(&em_tree
->lock
);
4812 ret
= add_extent_mapping(em_tree
, em
, 0);
4814 write_unlock(&em_tree
->lock
);
4815 free_extent_map(em
);
4819 list_add_tail(&em
->list
, &trans
->transaction
->pending_chunks
);
4820 refcount_inc(&em
->refs
);
4821 write_unlock(&em_tree
->lock
);
4823 ret
= btrfs_make_block_group(trans
, info
, 0, type
, start
, num_bytes
);
4825 goto error_del_extent
;
4827 for (i
= 0; i
< map
->num_stripes
; i
++) {
4828 num_bytes
= map
->stripes
[i
].dev
->bytes_used
+ stripe_size
;
4829 btrfs_device_set_bytes_used(map
->stripes
[i
].dev
, num_bytes
);
4832 atomic64_sub(stripe_size
* map
->num_stripes
, &info
->free_chunk_space
);
4834 free_extent_map(em
);
4835 check_raid56_incompat_flag(info
, type
);
4837 kfree(devices_info
);
4841 write_lock(&em_tree
->lock
);
4842 remove_extent_mapping(em_tree
, em
);
4843 write_unlock(&em_tree
->lock
);
4845 /* One for our allocation */
4846 free_extent_map(em
);
4847 /* One for the tree reference */
4848 free_extent_map(em
);
4849 /* One for the pending_chunks list reference */
4850 free_extent_map(em
);
4852 kfree(devices_info
);
4856 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
4857 struct btrfs_fs_info
*fs_info
,
4858 u64 chunk_offset
, u64 chunk_size
)
4860 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
4861 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
4862 struct btrfs_key key
;
4863 struct btrfs_device
*device
;
4864 struct btrfs_chunk
*chunk
;
4865 struct btrfs_stripe
*stripe
;
4866 struct extent_map
*em
;
4867 struct map_lookup
*map
;
4874 em
= get_chunk_map(fs_info
, chunk_offset
, chunk_size
);
4878 map
= em
->map_lookup
;
4879 item_size
= btrfs_chunk_item_size(map
->num_stripes
);
4880 stripe_size
= em
->orig_block_len
;
4882 chunk
= kzalloc(item_size
, GFP_NOFS
);
4889 * Take the device list mutex to prevent races with the final phase of
4890 * a device replace operation that replaces the device object associated
4891 * with the map's stripes, because the device object's id can change
4892 * at any time during that final phase of the device replace operation
4893 * (dev-replace.c:btrfs_dev_replace_finishing()).
4895 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
4896 for (i
= 0; i
< map
->num_stripes
; i
++) {
4897 device
= map
->stripes
[i
].dev
;
4898 dev_offset
= map
->stripes
[i
].physical
;
4900 ret
= btrfs_update_device(trans
, device
);
4903 ret
= btrfs_alloc_dev_extent(trans
, device
, chunk_offset
,
4904 dev_offset
, stripe_size
);
4909 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
4913 stripe
= &chunk
->stripe
;
4914 for (i
= 0; i
< map
->num_stripes
; i
++) {
4915 device
= map
->stripes
[i
].dev
;
4916 dev_offset
= map
->stripes
[i
].physical
;
4918 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
4919 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
4920 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
4923 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
4925 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
4926 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
4927 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
4928 btrfs_set_stack_chunk_type(chunk
, map
->type
);
4929 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
4930 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
4931 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
4932 btrfs_set_stack_chunk_sector_size(chunk
, fs_info
->sectorsize
);
4933 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
4935 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
4936 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
4937 key
.offset
= chunk_offset
;
4939 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
4940 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4942 * TODO: Cleanup of inserted chunk root in case of
4945 ret
= btrfs_add_system_chunk(fs_info
, &key
, chunk
, item_size
);
4950 free_extent_map(em
);
4955 * Chunk allocation falls into two parts. The first part does works
4956 * that make the new allocated chunk useable, but not do any operation
4957 * that modifies the chunk tree. The second part does the works that
4958 * require modifying the chunk tree. This division is important for the
4959 * bootstrap process of adding storage to a seed btrfs.
4961 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4962 struct btrfs_fs_info
*fs_info
, u64 type
)
4966 ASSERT(mutex_is_locked(&fs_info
->chunk_mutex
));
4967 chunk_offset
= find_next_chunk(fs_info
);
4968 return __btrfs_alloc_chunk(trans
, chunk_offset
, type
);
4971 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
4972 struct btrfs_fs_info
*fs_info
)
4975 u64 sys_chunk_offset
;
4979 chunk_offset
= find_next_chunk(fs_info
);
4980 alloc_profile
= btrfs_metadata_alloc_profile(fs_info
);
4981 ret
= __btrfs_alloc_chunk(trans
, chunk_offset
, alloc_profile
);
4985 sys_chunk_offset
= find_next_chunk(fs_info
);
4986 alloc_profile
= btrfs_system_alloc_profile(fs_info
);
4987 ret
= __btrfs_alloc_chunk(trans
, sys_chunk_offset
, alloc_profile
);
4991 static inline int btrfs_chunk_max_errors(struct map_lookup
*map
)
4995 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
4996 BTRFS_BLOCK_GROUP_RAID10
|
4997 BTRFS_BLOCK_GROUP_RAID5
|
4998 BTRFS_BLOCK_GROUP_DUP
)) {
5000 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
) {
5009 int btrfs_chunk_readonly(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
5011 struct extent_map
*em
;
5012 struct map_lookup
*map
;
5017 em
= get_chunk_map(fs_info
, chunk_offset
, 1);
5021 map
= em
->map_lookup
;
5022 for (i
= 0; i
< map
->num_stripes
; i
++) {
5023 if (map
->stripes
[i
].dev
->missing
) {
5028 if (!map
->stripes
[i
].dev
->writeable
) {
5035 * If the number of missing devices is larger than max errors,
5036 * we can not write the data into that chunk successfully, so
5039 if (miss_ndevs
> btrfs_chunk_max_errors(map
))
5042 free_extent_map(em
);
5046 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
5048 extent_map_tree_init(&tree
->map_tree
);
5051 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
5053 struct extent_map
*em
;
5056 write_lock(&tree
->map_tree
.lock
);
5057 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
5059 remove_extent_mapping(&tree
->map_tree
, em
);
5060 write_unlock(&tree
->map_tree
.lock
);
5064 free_extent_map(em
);
5065 /* once for the tree */
5066 free_extent_map(em
);
5070 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
5072 struct extent_map
*em
;
5073 struct map_lookup
*map
;
5076 em
= get_chunk_map(fs_info
, logical
, len
);
5079 * We could return errors for these cases, but that could get
5080 * ugly and we'd probably do the same thing which is just not do
5081 * anything else and exit, so return 1 so the callers don't try
5082 * to use other copies.
5086 map
= em
->map_lookup
;
5087 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
5088 ret
= map
->num_stripes
;
5089 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5090 ret
= map
->sub_stripes
;
5091 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID5
)
5093 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5097 free_extent_map(em
);
5099 btrfs_dev_replace_lock(&fs_info
->dev_replace
, 0);
5100 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
) &&
5101 fs_info
->dev_replace
.tgtdev
)
5103 btrfs_dev_replace_unlock(&fs_info
->dev_replace
, 0);
5108 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info
*fs_info
,
5111 struct extent_map
*em
;
5112 struct map_lookup
*map
;
5113 unsigned long len
= fs_info
->sectorsize
;
5115 em
= get_chunk_map(fs_info
, logical
, len
);
5117 if (!WARN_ON(IS_ERR(em
))) {
5118 map
= em
->map_lookup
;
5119 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5120 len
= map
->stripe_len
* nr_data_stripes(map
);
5121 free_extent_map(em
);
5126 int btrfs_is_parity_mirror(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
5128 struct extent_map
*em
;
5129 struct map_lookup
*map
;
5132 em
= get_chunk_map(fs_info
, logical
, len
);
5134 if(!WARN_ON(IS_ERR(em
))) {
5135 map
= em
->map_lookup
;
5136 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5138 free_extent_map(em
);
5143 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
5144 struct map_lookup
*map
, int first
, int num
,
5145 int optimal
, int dev_replace_is_ongoing
)
5149 struct btrfs_device
*srcdev
;
5151 if (dev_replace_is_ongoing
&&
5152 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
5153 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
5154 srcdev
= fs_info
->dev_replace
.srcdev
;
5159 * try to avoid the drive that is the source drive for a
5160 * dev-replace procedure, only choose it if no other non-missing
5161 * mirror is available
5163 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
5164 if (map
->stripes
[optimal
].dev
->bdev
&&
5165 (tolerance
|| map
->stripes
[optimal
].dev
!= srcdev
))
5167 for (i
= first
; i
< first
+ num
; i
++) {
5168 if (map
->stripes
[i
].dev
->bdev
&&
5169 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
5174 /* we couldn't find one that doesn't fail. Just return something
5175 * and the io error handling code will clean up eventually
5180 static inline int parity_smaller(u64 a
, u64 b
)
5185 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5186 static void sort_parity_stripes(struct btrfs_bio
*bbio
, int num_stripes
)
5188 struct btrfs_bio_stripe s
;
5195 for (i
= 0; i
< num_stripes
- 1; i
++) {
5196 if (parity_smaller(bbio
->raid_map
[i
],
5197 bbio
->raid_map
[i
+1])) {
5198 s
= bbio
->stripes
[i
];
5199 l
= bbio
->raid_map
[i
];
5200 bbio
->stripes
[i
] = bbio
->stripes
[i
+1];
5201 bbio
->raid_map
[i
] = bbio
->raid_map
[i
+1];
5202 bbio
->stripes
[i
+1] = s
;
5203 bbio
->raid_map
[i
+1] = l
;
5211 static struct btrfs_bio
*alloc_btrfs_bio(int total_stripes
, int real_stripes
)
5213 struct btrfs_bio
*bbio
= kzalloc(
5214 /* the size of the btrfs_bio */
5215 sizeof(struct btrfs_bio
) +
5216 /* plus the variable array for the stripes */
5217 sizeof(struct btrfs_bio_stripe
) * (total_stripes
) +
5218 /* plus the variable array for the tgt dev */
5219 sizeof(int) * (real_stripes
) +
5221 * plus the raid_map, which includes both the tgt dev
5224 sizeof(u64
) * (total_stripes
),
5225 GFP_NOFS
|__GFP_NOFAIL
);
5227 atomic_set(&bbio
->error
, 0);
5228 refcount_set(&bbio
->refs
, 1);
5233 void btrfs_get_bbio(struct btrfs_bio
*bbio
)
5235 WARN_ON(!refcount_read(&bbio
->refs
));
5236 refcount_inc(&bbio
->refs
);
5239 void btrfs_put_bbio(struct btrfs_bio
*bbio
)
5243 if (refcount_dec_and_test(&bbio
->refs
))
5247 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5249 * Please note that, discard won't be sent to target device of device
5252 static int __btrfs_map_block_for_discard(struct btrfs_fs_info
*fs_info
,
5253 u64 logical
, u64 length
,
5254 struct btrfs_bio
**bbio_ret
)
5256 struct extent_map
*em
;
5257 struct map_lookup
*map
;
5258 struct btrfs_bio
*bbio
;
5262 u64 stripe_end_offset
;
5269 u32 sub_stripes
= 0;
5270 u64 stripes_per_dev
= 0;
5271 u32 remaining_stripes
= 0;
5272 u32 last_stripe
= 0;
5276 /* discard always return a bbio */
5279 em
= get_chunk_map(fs_info
, logical
, length
);
5283 map
= em
->map_lookup
;
5284 /* we don't discard raid56 yet */
5285 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5290 offset
= logical
- em
->start
;
5291 length
= min_t(u64
, em
->len
- offset
, length
);
5293 stripe_len
= map
->stripe_len
;
5295 * stripe_nr counts the total number of stripes we have to stride
5296 * to get to this block
5298 stripe_nr
= div64_u64(offset
, stripe_len
);
5300 /* stripe_offset is the offset of this block in its stripe */
5301 stripe_offset
= offset
- stripe_nr
* stripe_len
;
5303 stripe_nr_end
= round_up(offset
+ length
, map
->stripe_len
);
5304 stripe_nr_end
= div64_u64(stripe_nr_end
, map
->stripe_len
);
5305 stripe_cnt
= stripe_nr_end
- stripe_nr
;
5306 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
5309 * after this, stripe_nr is the number of stripes on this
5310 * device we have to walk to find the data, and stripe_index is
5311 * the number of our device in the stripe array
5315 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5316 BTRFS_BLOCK_GROUP_RAID10
)) {
5317 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5320 sub_stripes
= map
->sub_stripes
;
5322 factor
= map
->num_stripes
/ sub_stripes
;
5323 num_stripes
= min_t(u64
, map
->num_stripes
,
5324 sub_stripes
* stripe_cnt
);
5325 stripe_nr
= div_u64_rem(stripe_nr
, factor
, &stripe_index
);
5326 stripe_index
*= sub_stripes
;
5327 stripes_per_dev
= div_u64_rem(stripe_cnt
, factor
,
5328 &remaining_stripes
);
5329 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
5330 last_stripe
*= sub_stripes
;
5331 } else if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
5332 BTRFS_BLOCK_GROUP_DUP
)) {
5333 num_stripes
= map
->num_stripes
;
5335 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5339 bbio
= alloc_btrfs_bio(num_stripes
, 0);
5345 for (i
= 0; i
< num_stripes
; i
++) {
5346 bbio
->stripes
[i
].physical
=
5347 map
->stripes
[stripe_index
].physical
+
5348 stripe_offset
+ stripe_nr
* map
->stripe_len
;
5349 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
5351 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5352 BTRFS_BLOCK_GROUP_RAID10
)) {
5353 bbio
->stripes
[i
].length
= stripes_per_dev
*
5356 if (i
/ sub_stripes
< remaining_stripes
)
5357 bbio
->stripes
[i
].length
+=
5361 * Special for the first stripe and
5364 * |-------|...|-------|
5368 if (i
< sub_stripes
)
5369 bbio
->stripes
[i
].length
-=
5372 if (stripe_index
>= last_stripe
&&
5373 stripe_index
<= (last_stripe
+
5375 bbio
->stripes
[i
].length
-=
5378 if (i
== sub_stripes
- 1)
5381 bbio
->stripes
[i
].length
= length
;
5385 if (stripe_index
== map
->num_stripes
) {
5392 bbio
->map_type
= map
->type
;
5393 bbio
->num_stripes
= num_stripes
;
5395 free_extent_map(em
);
5400 * In dev-replace case, for repair case (that's the only case where the mirror
5401 * is selected explicitly when calling btrfs_map_block), blocks left of the
5402 * left cursor can also be read from the target drive.
5404 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5406 * For READ, it also needs to be supported using the same mirror number.
5408 * If the requested block is not left of the left cursor, EIO is returned. This
5409 * can happen because btrfs_num_copies() returns one more in the dev-replace
5412 static int get_extra_mirror_from_replace(struct btrfs_fs_info
*fs_info
,
5413 u64 logical
, u64 length
,
5414 u64 srcdev_devid
, int *mirror_num
,
5417 struct btrfs_bio
*bbio
= NULL
;
5419 int index_srcdev
= 0;
5421 u64 physical_of_found
= 0;
5425 ret
= __btrfs_map_block(fs_info
, BTRFS_MAP_GET_READ_MIRRORS
,
5426 logical
, &length
, &bbio
, 0, 0);
5428 ASSERT(bbio
== NULL
);
5432 num_stripes
= bbio
->num_stripes
;
5433 if (*mirror_num
> num_stripes
) {
5435 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5436 * that means that the requested area is not left of the left
5439 btrfs_put_bbio(bbio
);
5444 * process the rest of the function using the mirror_num of the source
5445 * drive. Therefore look it up first. At the end, patch the device
5446 * pointer to the one of the target drive.
5448 for (i
= 0; i
< num_stripes
; i
++) {
5449 if (bbio
->stripes
[i
].dev
->devid
!= srcdev_devid
)
5453 * In case of DUP, in order to keep it simple, only add the
5454 * mirror with the lowest physical address
5457 physical_of_found
<= bbio
->stripes
[i
].physical
)
5462 physical_of_found
= bbio
->stripes
[i
].physical
;
5465 btrfs_put_bbio(bbio
);
5471 *mirror_num
= index_srcdev
+ 1;
5472 *physical
= physical_of_found
;
5476 static void handle_ops_on_dev_replace(enum btrfs_map_op op
,
5477 struct btrfs_bio
**bbio_ret
,
5478 struct btrfs_dev_replace
*dev_replace
,
5479 int *num_stripes_ret
, int *max_errors_ret
)
5481 struct btrfs_bio
*bbio
= *bbio_ret
;
5482 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5483 int tgtdev_indexes
= 0;
5484 int num_stripes
= *num_stripes_ret
;
5485 int max_errors
= *max_errors_ret
;
5488 if (op
== BTRFS_MAP_WRITE
) {
5489 int index_where_to_add
;
5492 * duplicate the write operations while the dev replace
5493 * procedure is running. Since the copying of the old disk to
5494 * the new disk takes place at run time while the filesystem is
5495 * mounted writable, the regular write operations to the old
5496 * disk have to be duplicated to go to the new disk as well.
5498 * Note that device->missing is handled by the caller, and that
5499 * the write to the old disk is already set up in the stripes
5502 index_where_to_add
= num_stripes
;
5503 for (i
= 0; i
< num_stripes
; i
++) {
5504 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5505 /* write to new disk, too */
5506 struct btrfs_bio_stripe
*new =
5507 bbio
->stripes
+ index_where_to_add
;
5508 struct btrfs_bio_stripe
*old
=
5511 new->physical
= old
->physical
;
5512 new->length
= old
->length
;
5513 new->dev
= dev_replace
->tgtdev
;
5514 bbio
->tgtdev_map
[i
] = index_where_to_add
;
5515 index_where_to_add
++;
5520 num_stripes
= index_where_to_add
;
5521 } else if (op
== BTRFS_MAP_GET_READ_MIRRORS
) {
5522 int index_srcdev
= 0;
5524 u64 physical_of_found
= 0;
5527 * During the dev-replace procedure, the target drive can also
5528 * be used to read data in case it is needed to repair a corrupt
5529 * block elsewhere. This is possible if the requested area is
5530 * left of the left cursor. In this area, the target drive is a
5531 * full copy of the source drive.
5533 for (i
= 0; i
< num_stripes
; i
++) {
5534 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5536 * In case of DUP, in order to keep it simple,
5537 * only add the mirror with the lowest physical
5541 physical_of_found
<=
5542 bbio
->stripes
[i
].physical
)
5546 physical_of_found
= bbio
->stripes
[i
].physical
;
5550 struct btrfs_bio_stripe
*tgtdev_stripe
=
5551 bbio
->stripes
+ num_stripes
;
5553 tgtdev_stripe
->physical
= physical_of_found
;
5554 tgtdev_stripe
->length
=
5555 bbio
->stripes
[index_srcdev
].length
;
5556 tgtdev_stripe
->dev
= dev_replace
->tgtdev
;
5557 bbio
->tgtdev_map
[index_srcdev
] = num_stripes
;
5564 *num_stripes_ret
= num_stripes
;
5565 *max_errors_ret
= max_errors
;
5566 bbio
->num_tgtdevs
= tgtdev_indexes
;
5570 static bool need_full_stripe(enum btrfs_map_op op
)
5572 return (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
);
5575 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
,
5576 enum btrfs_map_op op
,
5577 u64 logical
, u64
*length
,
5578 struct btrfs_bio
**bbio_ret
,
5579 int mirror_num
, int need_raid_map
)
5581 struct extent_map
*em
;
5582 struct map_lookup
*map
;
5592 int tgtdev_indexes
= 0;
5593 struct btrfs_bio
*bbio
= NULL
;
5594 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
5595 int dev_replace_is_ongoing
= 0;
5596 int num_alloc_stripes
;
5597 int patch_the_first_stripe_for_dev_replace
= 0;
5598 u64 physical_to_patch_in_first_stripe
= 0;
5599 u64 raid56_full_stripe_start
= (u64
)-1;
5601 if (op
== BTRFS_MAP_DISCARD
)
5602 return __btrfs_map_block_for_discard(fs_info
, logical
,
5605 em
= get_chunk_map(fs_info
, logical
, *length
);
5609 map
= em
->map_lookup
;
5610 offset
= logical
- em
->start
;
5612 stripe_len
= map
->stripe_len
;
5615 * stripe_nr counts the total number of stripes we have to stride
5616 * to get to this block
5618 stripe_nr
= div64_u64(stripe_nr
, stripe_len
);
5620 stripe_offset
= stripe_nr
* stripe_len
;
5621 if (offset
< stripe_offset
) {
5623 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5624 stripe_offset
, offset
, em
->start
, logical
,
5626 free_extent_map(em
);
5630 /* stripe_offset is the offset of this block in its stripe*/
5631 stripe_offset
= offset
- stripe_offset
;
5633 /* if we're here for raid56, we need to know the stripe aligned start */
5634 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5635 unsigned long full_stripe_len
= stripe_len
* nr_data_stripes(map
);
5636 raid56_full_stripe_start
= offset
;
5638 /* allow a write of a full stripe, but make sure we don't
5639 * allow straddling of stripes
5641 raid56_full_stripe_start
= div64_u64(raid56_full_stripe_start
,
5643 raid56_full_stripe_start
*= full_stripe_len
;
5646 if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
5648 /* For writes to RAID[56], allow a full stripeset across all disks.
5649 For other RAID types and for RAID[56] reads, just allow a single
5650 stripe (on a single disk). */
5651 if ((map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
5652 (op
== BTRFS_MAP_WRITE
)) {
5653 max_len
= stripe_len
* nr_data_stripes(map
) -
5654 (offset
- raid56_full_stripe_start
);
5656 /* we limit the length of each bio to what fits in a stripe */
5657 max_len
= stripe_len
- stripe_offset
;
5659 *length
= min_t(u64
, em
->len
- offset
, max_len
);
5661 *length
= em
->len
- offset
;
5664 /* This is for when we're called from btrfs_merge_bio_hook() and all
5665 it cares about is the length */
5669 btrfs_dev_replace_lock(dev_replace
, 0);
5670 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
5671 if (!dev_replace_is_ongoing
)
5672 btrfs_dev_replace_unlock(dev_replace
, 0);
5674 btrfs_dev_replace_set_lock_blocking(dev_replace
);
5676 if (dev_replace_is_ongoing
&& mirror_num
== map
->num_stripes
+ 1 &&
5677 !need_full_stripe(op
) && dev_replace
->tgtdev
!= NULL
) {
5678 ret
= get_extra_mirror_from_replace(fs_info
, logical
, *length
,
5679 dev_replace
->srcdev
->devid
,
5681 &physical_to_patch_in_first_stripe
);
5685 patch_the_first_stripe_for_dev_replace
= 1;
5686 } else if (mirror_num
> map
->num_stripes
) {
5692 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5693 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5695 if (!need_full_stripe(op
))
5697 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
5698 if (need_full_stripe(op
))
5699 num_stripes
= map
->num_stripes
;
5700 else if (mirror_num
)
5701 stripe_index
= mirror_num
- 1;
5703 stripe_index
= find_live_mirror(fs_info
, map
, 0,
5705 current
->pid
% map
->num_stripes
,
5706 dev_replace_is_ongoing
);
5707 mirror_num
= stripe_index
+ 1;
5710 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
5711 if (need_full_stripe(op
)) {
5712 num_stripes
= map
->num_stripes
;
5713 } else if (mirror_num
) {
5714 stripe_index
= mirror_num
- 1;
5719 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5720 u32 factor
= map
->num_stripes
/ map
->sub_stripes
;
5722 stripe_nr
= div_u64_rem(stripe_nr
, factor
, &stripe_index
);
5723 stripe_index
*= map
->sub_stripes
;
5725 if (need_full_stripe(op
))
5726 num_stripes
= map
->sub_stripes
;
5727 else if (mirror_num
)
5728 stripe_index
+= mirror_num
- 1;
5730 int old_stripe_index
= stripe_index
;
5731 stripe_index
= find_live_mirror(fs_info
, map
,
5733 map
->sub_stripes
, stripe_index
+
5734 current
->pid
% map
->sub_stripes
,
5735 dev_replace_is_ongoing
);
5736 mirror_num
= stripe_index
- old_stripe_index
+ 1;
5739 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5740 if (need_raid_map
&& (need_full_stripe(op
) || mirror_num
> 1)) {
5741 /* push stripe_nr back to the start of the full stripe */
5742 stripe_nr
= div64_u64(raid56_full_stripe_start
,
5743 stripe_len
* nr_data_stripes(map
));
5745 /* RAID[56] write or recovery. Return all stripes */
5746 num_stripes
= map
->num_stripes
;
5747 max_errors
= nr_parity_stripes(map
);
5749 *length
= map
->stripe_len
;
5754 * Mirror #0 or #1 means the original data block.
5755 * Mirror #2 is RAID5 parity block.
5756 * Mirror #3 is RAID6 Q block.
5758 stripe_nr
= div_u64_rem(stripe_nr
,
5759 nr_data_stripes(map
), &stripe_index
);
5761 stripe_index
= nr_data_stripes(map
) +
5764 /* We distribute the parity blocks across stripes */
5765 div_u64_rem(stripe_nr
+ stripe_index
, map
->num_stripes
,
5767 if (!need_full_stripe(op
) && mirror_num
<= 1)
5772 * after this, stripe_nr is the number of stripes on this
5773 * device we have to walk to find the data, and stripe_index is
5774 * the number of our device in the stripe array
5776 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5778 mirror_num
= stripe_index
+ 1;
5780 if (stripe_index
>= map
->num_stripes
) {
5782 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5783 stripe_index
, map
->num_stripes
);
5788 num_alloc_stripes
= num_stripes
;
5789 if (dev_replace_is_ongoing
&& dev_replace
->tgtdev
!= NULL
) {
5790 if (op
== BTRFS_MAP_WRITE
)
5791 num_alloc_stripes
<<= 1;
5792 if (op
== BTRFS_MAP_GET_READ_MIRRORS
)
5793 num_alloc_stripes
++;
5794 tgtdev_indexes
= num_stripes
;
5797 bbio
= alloc_btrfs_bio(num_alloc_stripes
, tgtdev_indexes
);
5802 if (dev_replace_is_ongoing
&& dev_replace
->tgtdev
!= NULL
)
5803 bbio
->tgtdev_map
= (int *)(bbio
->stripes
+ num_alloc_stripes
);
5805 /* build raid_map */
5806 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
&& need_raid_map
&&
5807 (need_full_stripe(op
) || mirror_num
> 1)) {
5811 bbio
->raid_map
= (u64
*)((void *)bbio
->stripes
+
5812 sizeof(struct btrfs_bio_stripe
) *
5814 sizeof(int) * tgtdev_indexes
);
5816 /* Work out the disk rotation on this stripe-set */
5817 div_u64_rem(stripe_nr
, num_stripes
, &rot
);
5819 /* Fill in the logical address of each stripe */
5820 tmp
= stripe_nr
* nr_data_stripes(map
);
5821 for (i
= 0; i
< nr_data_stripes(map
); i
++)
5822 bbio
->raid_map
[(i
+rot
) % num_stripes
] =
5823 em
->start
+ (tmp
+ i
) * map
->stripe_len
;
5825 bbio
->raid_map
[(i
+rot
) % map
->num_stripes
] = RAID5_P_STRIPE
;
5826 if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5827 bbio
->raid_map
[(i
+rot
+1) % num_stripes
] =
5832 for (i
= 0; i
< num_stripes
; i
++) {
5833 bbio
->stripes
[i
].physical
=
5834 map
->stripes
[stripe_index
].physical
+
5836 stripe_nr
* map
->stripe_len
;
5837 bbio
->stripes
[i
].dev
=
5838 map
->stripes
[stripe_index
].dev
;
5842 if (need_full_stripe(op
))
5843 max_errors
= btrfs_chunk_max_errors(map
);
5846 sort_parity_stripes(bbio
, num_stripes
);
5848 if (dev_replace_is_ongoing
&& dev_replace
->tgtdev
!= NULL
&&
5849 need_full_stripe(op
)) {
5850 handle_ops_on_dev_replace(op
, &bbio
, dev_replace
, &num_stripes
,
5855 bbio
->map_type
= map
->type
;
5856 bbio
->num_stripes
= num_stripes
;
5857 bbio
->max_errors
= max_errors
;
5858 bbio
->mirror_num
= mirror_num
;
5861 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5862 * mirror_num == num_stripes + 1 && dev_replace target drive is
5863 * available as a mirror
5865 if (patch_the_first_stripe_for_dev_replace
&& num_stripes
> 0) {
5866 WARN_ON(num_stripes
> 1);
5867 bbio
->stripes
[0].dev
= dev_replace
->tgtdev
;
5868 bbio
->stripes
[0].physical
= physical_to_patch_in_first_stripe
;
5869 bbio
->mirror_num
= map
->num_stripes
+ 1;
5872 if (dev_replace_is_ongoing
) {
5873 btrfs_dev_replace_clear_lock_blocking(dev_replace
);
5874 btrfs_dev_replace_unlock(dev_replace
, 0);
5876 free_extent_map(em
);
5880 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, enum btrfs_map_op op
,
5881 u64 logical
, u64
*length
,
5882 struct btrfs_bio
**bbio_ret
, int mirror_num
)
5884 return __btrfs_map_block(fs_info
, op
, logical
, length
, bbio_ret
,
5888 /* For Scrub/replace */
5889 int btrfs_map_sblock(struct btrfs_fs_info
*fs_info
, enum btrfs_map_op op
,
5890 u64 logical
, u64
*length
,
5891 struct btrfs_bio
**bbio_ret
)
5893 return __btrfs_map_block(fs_info
, op
, logical
, length
, bbio_ret
, 0, 1);
5896 int btrfs_rmap_block(struct btrfs_fs_info
*fs_info
,
5897 u64 chunk_start
, u64 physical
, u64 devid
,
5898 u64
**logical
, int *naddrs
, int *stripe_len
)
5900 struct extent_map
*em
;
5901 struct map_lookup
*map
;
5909 em
= get_chunk_map(fs_info
, chunk_start
, 1);
5913 map
= em
->map_lookup
;
5915 rmap_len
= map
->stripe_len
;
5917 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5918 length
= div_u64(length
, map
->num_stripes
/ map
->sub_stripes
);
5919 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5920 length
= div_u64(length
, map
->num_stripes
);
5921 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5922 length
= div_u64(length
, nr_data_stripes(map
));
5923 rmap_len
= map
->stripe_len
* nr_data_stripes(map
);
5926 buf
= kcalloc(map
->num_stripes
, sizeof(u64
), GFP_NOFS
);
5927 BUG_ON(!buf
); /* -ENOMEM */
5929 for (i
= 0; i
< map
->num_stripes
; i
++) {
5930 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
5932 if (map
->stripes
[i
].physical
> physical
||
5933 map
->stripes
[i
].physical
+ length
<= physical
)
5936 stripe_nr
= physical
- map
->stripes
[i
].physical
;
5937 stripe_nr
= div64_u64(stripe_nr
, map
->stripe_len
);
5939 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5940 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5941 stripe_nr
= div_u64(stripe_nr
, map
->sub_stripes
);
5942 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5943 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5944 } /* else if RAID[56], multiply by nr_data_stripes().
5945 * Alternatively, just use rmap_len below instead of
5946 * map->stripe_len */
5948 bytenr
= chunk_start
+ stripe_nr
* rmap_len
;
5949 WARN_ON(nr
>= map
->num_stripes
);
5950 for (j
= 0; j
< nr
; j
++) {
5951 if (buf
[j
] == bytenr
)
5955 WARN_ON(nr
>= map
->num_stripes
);
5962 *stripe_len
= rmap_len
;
5964 free_extent_map(em
);
5968 static inline void btrfs_end_bbio(struct btrfs_bio
*bbio
, struct bio
*bio
)
5970 bio
->bi_private
= bbio
->private;
5971 bio
->bi_end_io
= bbio
->end_io
;
5974 btrfs_put_bbio(bbio
);
5977 static void btrfs_end_bio(struct bio
*bio
)
5979 struct btrfs_bio
*bbio
= bio
->bi_private
;
5980 int is_orig_bio
= 0;
5982 if (bio
->bi_status
) {
5983 atomic_inc(&bbio
->error
);
5984 if (bio
->bi_status
== BLK_STS_IOERR
||
5985 bio
->bi_status
== BLK_STS_TARGET
) {
5986 unsigned int stripe_index
=
5987 btrfs_io_bio(bio
)->stripe_index
;
5988 struct btrfs_device
*dev
;
5990 BUG_ON(stripe_index
>= bbio
->num_stripes
);
5991 dev
= bbio
->stripes
[stripe_index
].dev
;
5993 if (bio_op(bio
) == REQ_OP_WRITE
)
5994 btrfs_dev_stat_inc(dev
,
5995 BTRFS_DEV_STAT_WRITE_ERRS
);
5997 btrfs_dev_stat_inc(dev
,
5998 BTRFS_DEV_STAT_READ_ERRS
);
5999 if (bio
->bi_opf
& REQ_PREFLUSH
)
6000 btrfs_dev_stat_inc(dev
,
6001 BTRFS_DEV_STAT_FLUSH_ERRS
);
6002 btrfs_dev_stat_print_on_error(dev
);
6007 if (bio
== bbio
->orig_bio
)
6010 btrfs_bio_counter_dec(bbio
->fs_info
);
6012 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
6015 bio
= bbio
->orig_bio
;
6018 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
6019 /* only send an error to the higher layers if it is
6020 * beyond the tolerance of the btrfs bio
6022 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
6023 bio
->bi_status
= BLK_STS_IOERR
;
6026 * this bio is actually up to date, we didn't
6027 * go over the max number of errors
6029 bio
->bi_status
= BLK_STS_OK
;
6032 btrfs_end_bbio(bbio
, bio
);
6033 } else if (!is_orig_bio
) {
6039 * see run_scheduled_bios for a description of why bios are collected for
6042 * This will add one bio to the pending list for a device and make sure
6043 * the work struct is scheduled.
6045 static noinline
void btrfs_schedule_bio(struct btrfs_device
*device
,
6048 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
6049 int should_queue
= 1;
6050 struct btrfs_pending_bios
*pending_bios
;
6052 if (device
->missing
|| !device
->bdev
) {
6057 /* don't bother with additional async steps for reads, right now */
6058 if (bio_op(bio
) == REQ_OP_READ
) {
6060 btrfsic_submit_bio(bio
);
6065 WARN_ON(bio
->bi_next
);
6066 bio
->bi_next
= NULL
;
6068 spin_lock(&device
->io_lock
);
6069 if (op_is_sync(bio
->bi_opf
))
6070 pending_bios
= &device
->pending_sync_bios
;
6072 pending_bios
= &device
->pending_bios
;
6074 if (pending_bios
->tail
)
6075 pending_bios
->tail
->bi_next
= bio
;
6077 pending_bios
->tail
= bio
;
6078 if (!pending_bios
->head
)
6079 pending_bios
->head
= bio
;
6080 if (device
->running_pending
)
6083 spin_unlock(&device
->io_lock
);
6086 btrfs_queue_work(fs_info
->submit_workers
, &device
->work
);
6089 static void submit_stripe_bio(struct btrfs_bio
*bbio
, struct bio
*bio
,
6090 u64 physical
, int dev_nr
, int async
)
6092 struct btrfs_device
*dev
= bbio
->stripes
[dev_nr
].dev
;
6093 struct btrfs_fs_info
*fs_info
= bbio
->fs_info
;
6095 bio
->bi_private
= bbio
;
6096 btrfs_io_bio(bio
)->stripe_index
= dev_nr
;
6097 bio
->bi_end_io
= btrfs_end_bio
;
6098 bio
->bi_iter
.bi_sector
= physical
>> 9;
6101 struct rcu_string
*name
;
6104 name
= rcu_dereference(dev
->name
);
6105 btrfs_debug(fs_info
,
6106 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6107 bio_op(bio
), bio
->bi_opf
,
6108 (u64
)bio
->bi_iter
.bi_sector
,
6109 (u_long
)dev
->bdev
->bd_dev
, name
->str
, dev
->devid
,
6110 bio
->bi_iter
.bi_size
);
6114 bio_set_dev(bio
, dev
->bdev
);
6116 btrfs_bio_counter_inc_noblocked(fs_info
);
6119 btrfs_schedule_bio(dev
, bio
);
6121 btrfsic_submit_bio(bio
);
6124 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
6126 atomic_inc(&bbio
->error
);
6127 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
6128 /* Should be the original bio. */
6129 WARN_ON(bio
!= bbio
->orig_bio
);
6131 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
6132 bio
->bi_iter
.bi_sector
= logical
>> 9;
6133 if (atomic_read(&bbio
->error
) > bbio
->max_errors
)
6134 bio
->bi_status
= BLK_STS_IOERR
;
6136 bio
->bi_status
= BLK_STS_OK
;
6137 btrfs_end_bbio(bbio
, bio
);
6141 blk_status_t
btrfs_map_bio(struct btrfs_fs_info
*fs_info
, struct bio
*bio
,
6142 int mirror_num
, int async_submit
)
6144 struct btrfs_device
*dev
;
6145 struct bio
*first_bio
= bio
;
6146 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
6152 struct btrfs_bio
*bbio
= NULL
;
6154 length
= bio
->bi_iter
.bi_size
;
6155 map_length
= length
;
6157 btrfs_bio_counter_inc_blocked(fs_info
);
6158 ret
= __btrfs_map_block(fs_info
, btrfs_op(bio
), logical
,
6159 &map_length
, &bbio
, mirror_num
, 1);
6161 btrfs_bio_counter_dec(fs_info
);
6162 return errno_to_blk_status(ret
);
6165 total_devs
= bbio
->num_stripes
;
6166 bbio
->orig_bio
= first_bio
;
6167 bbio
->private = first_bio
->bi_private
;
6168 bbio
->end_io
= first_bio
->bi_end_io
;
6169 bbio
->fs_info
= fs_info
;
6170 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
6172 if ((bbio
->map_type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
6173 ((bio_op(bio
) == REQ_OP_WRITE
) || (mirror_num
> 1))) {
6174 /* In this case, map_length has been set to the length of
6175 a single stripe; not the whole write */
6176 if (bio_op(bio
) == REQ_OP_WRITE
) {
6177 ret
= raid56_parity_write(fs_info
, bio
, bbio
,
6180 ret
= raid56_parity_recover(fs_info
, bio
, bbio
,
6181 map_length
, mirror_num
, 1);
6184 btrfs_bio_counter_dec(fs_info
);
6185 return errno_to_blk_status(ret
);
6188 if (map_length
< length
) {
6190 "mapping failed logical %llu bio len %llu len %llu",
6191 logical
, length
, map_length
);
6195 for (dev_nr
= 0; dev_nr
< total_devs
; dev_nr
++) {
6196 dev
= bbio
->stripes
[dev_nr
].dev
;
6197 if (!dev
|| !dev
->bdev
||
6198 (bio_op(first_bio
) == REQ_OP_WRITE
&& !dev
->writeable
)) {
6199 bbio_error(bbio
, first_bio
, logical
);
6203 if (dev_nr
< total_devs
- 1)
6204 bio
= btrfs_bio_clone(first_bio
);
6208 submit_stripe_bio(bbio
, bio
, bbio
->stripes
[dev_nr
].physical
,
6209 dev_nr
, async_submit
);
6211 btrfs_bio_counter_dec(fs_info
);
6215 struct btrfs_device
*btrfs_find_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
6218 struct btrfs_device
*device
;
6219 struct btrfs_fs_devices
*cur_devices
;
6221 cur_devices
= fs_info
->fs_devices
;
6222 while (cur_devices
) {
6224 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_FSID_SIZE
)) {
6225 device
= find_device(cur_devices
, devid
, uuid
);
6229 cur_devices
= cur_devices
->seed
;
6234 static struct btrfs_device
*add_missing_dev(struct btrfs_fs_devices
*fs_devices
,
6235 u64 devid
, u8
*dev_uuid
)
6237 struct btrfs_device
*device
;
6239 device
= btrfs_alloc_device(NULL
, &devid
, dev_uuid
);
6243 list_add(&device
->dev_list
, &fs_devices
->devices
);
6244 device
->fs_devices
= fs_devices
;
6245 fs_devices
->num_devices
++;
6247 device
->missing
= 1;
6248 fs_devices
->missing_devices
++;
6254 * btrfs_alloc_device - allocate struct btrfs_device
6255 * @fs_info: used only for generating a new devid, can be NULL if
6256 * devid is provided (i.e. @devid != NULL).
6257 * @devid: a pointer to devid for this device. If NULL a new devid
6259 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6262 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6263 * on error. Returned struct is not linked onto any lists and can be
6264 * destroyed with kfree() right away.
6266 struct btrfs_device
*btrfs_alloc_device(struct btrfs_fs_info
*fs_info
,
6270 struct btrfs_device
*dev
;
6273 if (WARN_ON(!devid
&& !fs_info
))
6274 return ERR_PTR(-EINVAL
);
6276 dev
= __alloc_device();
6285 ret
= find_next_devid(fs_info
, &tmp
);
6288 return ERR_PTR(ret
);
6294 memcpy(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
);
6296 generate_random_uuid(dev
->uuid
);
6298 btrfs_init_work(&dev
->work
, btrfs_submit_helper
,
6299 pending_bios_fn
, NULL
, NULL
);
6304 /* Return -EIO if any error, otherwise return 0. */
6305 static int btrfs_check_chunk_valid(struct btrfs_fs_info
*fs_info
,
6306 struct extent_buffer
*leaf
,
6307 struct btrfs_chunk
*chunk
, u64 logical
)
6315 length
= btrfs_chunk_length(leaf
, chunk
);
6316 stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6317 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6318 sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6319 type
= btrfs_chunk_type(leaf
, chunk
);
6322 btrfs_err(fs_info
, "invalid chunk num_stripes: %u",
6326 if (!IS_ALIGNED(logical
, fs_info
->sectorsize
)) {
6327 btrfs_err(fs_info
, "invalid chunk logical %llu", logical
);
6330 if (btrfs_chunk_sector_size(leaf
, chunk
) != fs_info
->sectorsize
) {
6331 btrfs_err(fs_info
, "invalid chunk sectorsize %u",
6332 btrfs_chunk_sector_size(leaf
, chunk
));
6335 if (!length
|| !IS_ALIGNED(length
, fs_info
->sectorsize
)) {
6336 btrfs_err(fs_info
, "invalid chunk length %llu", length
);
6339 if (!is_power_of_2(stripe_len
) || stripe_len
!= BTRFS_STRIPE_LEN
) {
6340 btrfs_err(fs_info
, "invalid chunk stripe length: %llu",
6344 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK
| BTRFS_BLOCK_GROUP_PROFILE_MASK
) &
6346 btrfs_err(fs_info
, "unrecognized chunk type: %llu",
6347 ~(BTRFS_BLOCK_GROUP_TYPE_MASK
|
6348 BTRFS_BLOCK_GROUP_PROFILE_MASK
) &
6349 btrfs_chunk_type(leaf
, chunk
));
6352 if ((type
& BTRFS_BLOCK_GROUP_RAID10
&& sub_stripes
!= 2) ||
6353 (type
& BTRFS_BLOCK_GROUP_RAID1
&& num_stripes
< 1) ||
6354 (type
& BTRFS_BLOCK_GROUP_RAID5
&& num_stripes
< 2) ||
6355 (type
& BTRFS_BLOCK_GROUP_RAID6
&& num_stripes
< 3) ||
6356 (type
& BTRFS_BLOCK_GROUP_DUP
&& num_stripes
> 2) ||
6357 ((type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) == 0 &&
6358 num_stripes
!= 1)) {
6360 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6361 num_stripes
, sub_stripes
,
6362 type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
);
6369 static void btrfs_report_missing_device(struct btrfs_fs_info
*fs_info
,
6370 u64 devid
, u8
*uuid
, bool error
)
6373 btrfs_err_rl(fs_info
, "devid %llu uuid %pU is missing",
6376 btrfs_warn_rl(fs_info
, "devid %llu uuid %pU is missing",
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
, true);
6453 if (!map
->stripes
[i
].dev
) {
6454 map
->stripes
[i
].dev
=
6455 add_missing_dev(fs_info
->fs_devices
, devid
,
6457 if (IS_ERR(map
->stripes
[i
].dev
)) {
6458 free_extent_map(em
);
6460 "failed to init missing dev %llu: %ld",
6461 devid
, PTR_ERR(map
->stripes
[i
].dev
));
6462 return PTR_ERR(map
->stripes
[i
].dev
);
6464 btrfs_report_missing_device(fs_info
, devid
, uuid
, false);
6466 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
6469 write_lock(&map_tree
->map_tree
.lock
);
6470 ret
= add_extent_mapping(&map_tree
->map_tree
, em
, 0);
6471 write_unlock(&map_tree
->map_tree
.lock
);
6472 BUG_ON(ret
); /* Tree corruption */
6473 free_extent_map(em
);
6478 static void fill_device_from_item(struct extent_buffer
*leaf
,
6479 struct btrfs_dev_item
*dev_item
,
6480 struct btrfs_device
*device
)
6484 device
->devid
= btrfs_device_id(leaf
, dev_item
);
6485 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
6486 device
->total_bytes
= device
->disk_total_bytes
;
6487 device
->commit_total_bytes
= device
->disk_total_bytes
;
6488 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
6489 device
->commit_bytes_used
= device
->bytes_used
;
6490 device
->type
= btrfs_device_type(leaf
, dev_item
);
6491 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
6492 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
6493 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
6494 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
6495 device
->is_tgtdev_for_dev_replace
= 0;
6497 ptr
= btrfs_device_uuid(dev_item
);
6498 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
6501 static struct btrfs_fs_devices
*open_seed_devices(struct btrfs_fs_info
*fs_info
,
6504 struct btrfs_fs_devices
*fs_devices
;
6507 BUG_ON(!mutex_is_locked(&uuid_mutex
));
6510 fs_devices
= fs_info
->fs_devices
->seed
;
6511 while (fs_devices
) {
6512 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_FSID_SIZE
))
6515 fs_devices
= fs_devices
->seed
;
6518 fs_devices
= find_fsid(fsid
);
6520 if (!btrfs_test_opt(fs_info
, DEGRADED
))
6521 return ERR_PTR(-ENOENT
);
6523 fs_devices
= alloc_fs_devices(fsid
);
6524 if (IS_ERR(fs_devices
))
6527 fs_devices
->seeding
= 1;
6528 fs_devices
->opened
= 1;
6532 fs_devices
= clone_fs_devices(fs_devices
);
6533 if (IS_ERR(fs_devices
))
6536 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
6537 fs_info
->bdev_holder
);
6539 free_fs_devices(fs_devices
);
6540 fs_devices
= ERR_PTR(ret
);
6544 if (!fs_devices
->seeding
) {
6545 __btrfs_close_devices(fs_devices
);
6546 free_fs_devices(fs_devices
);
6547 fs_devices
= ERR_PTR(-EINVAL
);
6551 fs_devices
->seed
= fs_info
->fs_devices
->seed
;
6552 fs_info
->fs_devices
->seed
= fs_devices
;
6557 static int read_one_dev(struct btrfs_fs_info
*fs_info
,
6558 struct extent_buffer
*leaf
,
6559 struct btrfs_dev_item
*dev_item
)
6561 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6562 struct btrfs_device
*device
;
6565 u8 fs_uuid
[BTRFS_FSID_SIZE
];
6566 u8 dev_uuid
[BTRFS_UUID_SIZE
];
6568 devid
= btrfs_device_id(leaf
, dev_item
);
6569 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
6571 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
6574 if (memcmp(fs_uuid
, fs_info
->fsid
, BTRFS_FSID_SIZE
)) {
6575 fs_devices
= open_seed_devices(fs_info
, fs_uuid
);
6576 if (IS_ERR(fs_devices
))
6577 return PTR_ERR(fs_devices
);
6580 device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, fs_uuid
);
6582 if (!btrfs_test_opt(fs_info
, DEGRADED
)) {
6583 btrfs_report_missing_device(fs_info
, devid
,
6588 device
= add_missing_dev(fs_devices
, devid
, dev_uuid
);
6589 if (IS_ERR(device
)) {
6591 "failed to add missing dev %llu: %ld",
6592 devid
, PTR_ERR(device
));
6593 return PTR_ERR(device
);
6595 btrfs_report_missing_device(fs_info
, devid
, dev_uuid
, false);
6597 if (!device
->bdev
) {
6598 if (!btrfs_test_opt(fs_info
, DEGRADED
)) {
6599 btrfs_report_missing_device(fs_info
,
6600 devid
, dev_uuid
, true);
6603 btrfs_report_missing_device(fs_info
, devid
,
6607 if(!device
->bdev
&& !device
->missing
) {
6609 * this happens when a device that was properly setup
6610 * in the device info lists suddenly goes bad.
6611 * device->bdev is NULL, and so we have to set
6612 * device->missing to one here
6614 device
->fs_devices
->missing_devices
++;
6615 device
->missing
= 1;
6618 /* Move the device to its own fs_devices */
6619 if (device
->fs_devices
!= fs_devices
) {
6620 ASSERT(device
->missing
);
6622 list_move(&device
->dev_list
, &fs_devices
->devices
);
6623 device
->fs_devices
->num_devices
--;
6624 fs_devices
->num_devices
++;
6626 device
->fs_devices
->missing_devices
--;
6627 fs_devices
->missing_devices
++;
6629 device
->fs_devices
= fs_devices
;
6633 if (device
->fs_devices
!= fs_info
->fs_devices
) {
6634 BUG_ON(device
->writeable
);
6635 if (device
->generation
!=
6636 btrfs_device_generation(leaf
, dev_item
))
6640 fill_device_from_item(leaf
, dev_item
, device
);
6641 device
->in_fs_metadata
= 1;
6642 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
6643 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
6644 atomic64_add(device
->total_bytes
- device
->bytes_used
,
6645 &fs_info
->free_chunk_space
);
6651 int btrfs_read_sys_array(struct btrfs_fs_info
*fs_info
)
6653 struct btrfs_root
*root
= fs_info
->tree_root
;
6654 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
6655 struct extent_buffer
*sb
;
6656 struct btrfs_disk_key
*disk_key
;
6657 struct btrfs_chunk
*chunk
;
6659 unsigned long sb_array_offset
;
6666 struct btrfs_key key
;
6668 ASSERT(BTRFS_SUPER_INFO_SIZE
<= fs_info
->nodesize
);
6670 * This will create extent buffer of nodesize, superblock size is
6671 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6672 * overallocate but we can keep it as-is, only the first page is used.
6674 sb
= btrfs_find_create_tree_block(fs_info
, BTRFS_SUPER_INFO_OFFSET
);
6677 set_extent_buffer_uptodate(sb
);
6678 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
6680 * The sb extent buffer is artificial and just used to read the system array.
6681 * set_extent_buffer_uptodate() call does not properly mark all it's
6682 * pages up-to-date when the page is larger: extent does not cover the
6683 * whole page and consequently check_page_uptodate does not find all
6684 * the page's extents up-to-date (the hole beyond sb),
6685 * write_extent_buffer then triggers a WARN_ON.
6687 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6688 * but sb spans only this function. Add an explicit SetPageUptodate call
6689 * to silence the warning eg. on PowerPC 64.
6691 if (PAGE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
6692 SetPageUptodate(sb
->pages
[0]);
6694 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
6695 array_size
= btrfs_super_sys_array_size(super_copy
);
6697 array_ptr
= super_copy
->sys_chunk_array
;
6698 sb_array_offset
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
6701 while (cur_offset
< array_size
) {
6702 disk_key
= (struct btrfs_disk_key
*)array_ptr
;
6703 len
= sizeof(*disk_key
);
6704 if (cur_offset
+ len
> array_size
)
6705 goto out_short_read
;
6707 btrfs_disk_key_to_cpu(&key
, disk_key
);
6710 sb_array_offset
+= len
;
6713 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6714 chunk
= (struct btrfs_chunk
*)sb_array_offset
;
6716 * At least one btrfs_chunk with one stripe must be
6717 * present, exact stripe count check comes afterwards
6719 len
= btrfs_chunk_item_size(1);
6720 if (cur_offset
+ len
> array_size
)
6721 goto out_short_read
;
6723 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
6726 "invalid number of stripes %u in sys_array at offset %u",
6727 num_stripes
, cur_offset
);
6732 type
= btrfs_chunk_type(sb
, chunk
);
6733 if ((type
& BTRFS_BLOCK_GROUP_SYSTEM
) == 0) {
6735 "invalid chunk type %llu in sys_array at offset %u",
6741 len
= btrfs_chunk_item_size(num_stripes
);
6742 if (cur_offset
+ len
> array_size
)
6743 goto out_short_read
;
6745 ret
= read_one_chunk(fs_info
, &key
, sb
, chunk
);
6750 "unexpected item type %u in sys_array at offset %u",
6751 (u32
)key
.type
, cur_offset
);
6756 sb_array_offset
+= len
;
6759 clear_extent_buffer_uptodate(sb
);
6760 free_extent_buffer_stale(sb
);
6764 btrfs_err(fs_info
, "sys_array too short to read %u bytes at offset %u",
6766 clear_extent_buffer_uptodate(sb
);
6767 free_extent_buffer_stale(sb
);
6772 * Check if all chunks in the fs are OK for read-write degraded mount
6774 * Return true if all chunks meet the minimal RW mount requirements.
6775 * Return false if any chunk doesn't meet the minimal RW mount requirements.
6777 bool btrfs_check_rw_degradable(struct btrfs_fs_info
*fs_info
)
6779 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
6780 struct extent_map
*em
;
6784 read_lock(&map_tree
->map_tree
.lock
);
6785 em
= lookup_extent_mapping(&map_tree
->map_tree
, 0, (u64
)-1);
6786 read_unlock(&map_tree
->map_tree
.lock
);
6787 /* No chunk at all? Return false anyway */
6793 struct map_lookup
*map
;
6798 map
= em
->map_lookup
;
6800 btrfs_get_num_tolerated_disk_barrier_failures(
6802 for (i
= 0; i
< map
->num_stripes
; i
++) {
6803 struct btrfs_device
*dev
= map
->stripes
[i
].dev
;
6805 if (!dev
|| !dev
->bdev
|| dev
->missing
||
6806 dev
->last_flush_error
)
6809 if (missing
> max_tolerated
) {
6811 "chunk %llu missing %d devices, max tolerance is %d for writeable mount",
6812 em
->start
, missing
, max_tolerated
);
6813 free_extent_map(em
);
6817 next_start
= extent_map_end(em
);
6818 free_extent_map(em
);
6820 read_lock(&map_tree
->map_tree
.lock
);
6821 em
= lookup_extent_mapping(&map_tree
->map_tree
, next_start
,
6822 (u64
)(-1) - next_start
);
6823 read_unlock(&map_tree
->map_tree
.lock
);
6829 int btrfs_read_chunk_tree(struct btrfs_fs_info
*fs_info
)
6831 struct btrfs_root
*root
= fs_info
->chunk_root
;
6832 struct btrfs_path
*path
;
6833 struct extent_buffer
*leaf
;
6834 struct btrfs_key key
;
6835 struct btrfs_key found_key
;
6840 path
= btrfs_alloc_path();
6844 mutex_lock(&uuid_mutex
);
6845 mutex_lock(&fs_info
->chunk_mutex
);
6848 * Read all device items, and then all the chunk items. All
6849 * device items are found before any chunk item (their object id
6850 * is smaller than the lowest possible object id for a chunk
6851 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6853 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
6856 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6860 leaf
= path
->nodes
[0];
6861 slot
= path
->slots
[0];
6862 if (slot
>= btrfs_header_nritems(leaf
)) {
6863 ret
= btrfs_next_leaf(root
, path
);
6870 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6871 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
6872 struct btrfs_dev_item
*dev_item
;
6873 dev_item
= btrfs_item_ptr(leaf
, slot
,
6874 struct btrfs_dev_item
);
6875 ret
= read_one_dev(fs_info
, leaf
, dev_item
);
6879 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6880 struct btrfs_chunk
*chunk
;
6881 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
6882 ret
= read_one_chunk(fs_info
, &found_key
, leaf
, chunk
);
6890 * After loading chunk tree, we've got all device information,
6891 * do another round of validation checks.
6893 if (total_dev
!= fs_info
->fs_devices
->total_devices
) {
6895 "super_num_devices %llu mismatch with num_devices %llu found here",
6896 btrfs_super_num_devices(fs_info
->super_copy
),
6901 if (btrfs_super_total_bytes(fs_info
->super_copy
) <
6902 fs_info
->fs_devices
->total_rw_bytes
) {
6904 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6905 btrfs_super_total_bytes(fs_info
->super_copy
),
6906 fs_info
->fs_devices
->total_rw_bytes
);
6912 mutex_unlock(&fs_info
->chunk_mutex
);
6913 mutex_unlock(&uuid_mutex
);
6915 btrfs_free_path(path
);
6919 void btrfs_init_devices_late(struct btrfs_fs_info
*fs_info
)
6921 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6922 struct btrfs_device
*device
;
6924 while (fs_devices
) {
6925 mutex_lock(&fs_devices
->device_list_mutex
);
6926 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
)
6927 device
->fs_info
= fs_info
;
6928 mutex_unlock(&fs_devices
->device_list_mutex
);
6930 fs_devices
= fs_devices
->seed
;
6934 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
6938 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6939 btrfs_dev_stat_reset(dev
, i
);
6942 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
6944 struct btrfs_key key
;
6945 struct btrfs_key found_key
;
6946 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6947 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6948 struct extent_buffer
*eb
;
6951 struct btrfs_device
*device
;
6952 struct btrfs_path
*path
= NULL
;
6955 path
= btrfs_alloc_path();
6961 mutex_lock(&fs_devices
->device_list_mutex
);
6962 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6964 struct btrfs_dev_stats_item
*ptr
;
6966 key
.objectid
= BTRFS_DEV_STATS_OBJECTID
;
6967 key
.type
= BTRFS_PERSISTENT_ITEM_KEY
;
6968 key
.offset
= device
->devid
;
6969 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
6971 __btrfs_reset_dev_stats(device
);
6972 device
->dev_stats_valid
= 1;
6973 btrfs_release_path(path
);
6976 slot
= path
->slots
[0];
6977 eb
= path
->nodes
[0];
6978 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
6979 item_size
= btrfs_item_size_nr(eb
, slot
);
6981 ptr
= btrfs_item_ptr(eb
, slot
,
6982 struct btrfs_dev_stats_item
);
6984 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6985 if (item_size
>= (1 + i
) * sizeof(__le64
))
6986 btrfs_dev_stat_set(device
, i
,
6987 btrfs_dev_stats_value(eb
, ptr
, i
));
6989 btrfs_dev_stat_reset(device
, i
);
6992 device
->dev_stats_valid
= 1;
6993 btrfs_dev_stat_print_on_load(device
);
6994 btrfs_release_path(path
);
6996 mutex_unlock(&fs_devices
->device_list_mutex
);
6999 btrfs_free_path(path
);
7000 return ret
< 0 ? ret
: 0;
7003 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
7004 struct btrfs_fs_info
*fs_info
,
7005 struct btrfs_device
*device
)
7007 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
7008 struct btrfs_path
*path
;
7009 struct btrfs_key key
;
7010 struct extent_buffer
*eb
;
7011 struct btrfs_dev_stats_item
*ptr
;
7015 key
.objectid
= BTRFS_DEV_STATS_OBJECTID
;
7016 key
.type
= BTRFS_PERSISTENT_ITEM_KEY
;
7017 key
.offset
= device
->devid
;
7019 path
= btrfs_alloc_path();
7022 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
7024 btrfs_warn_in_rcu(fs_info
,
7025 "error %d while searching for dev_stats item for device %s",
7026 ret
, rcu_str_deref(device
->name
));
7031 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
7032 /* need to delete old one and insert a new one */
7033 ret
= btrfs_del_item(trans
, dev_root
, path
);
7035 btrfs_warn_in_rcu(fs_info
,
7036 "delete too small dev_stats item for device %s failed %d",
7037 rcu_str_deref(device
->name
), ret
);
7044 /* need to insert a new item */
7045 btrfs_release_path(path
);
7046 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
7047 &key
, sizeof(*ptr
));
7049 btrfs_warn_in_rcu(fs_info
,
7050 "insert dev_stats item for device %s failed %d",
7051 rcu_str_deref(device
->name
), ret
);
7056 eb
= path
->nodes
[0];
7057 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
7058 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7059 btrfs_set_dev_stats_value(eb
, ptr
, i
,
7060 btrfs_dev_stat_read(device
, i
));
7061 btrfs_mark_buffer_dirty(eb
);
7064 btrfs_free_path(path
);
7069 * called from commit_transaction. Writes all changed device stats to disk.
7071 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
7072 struct btrfs_fs_info
*fs_info
)
7074 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7075 struct btrfs_device
*device
;
7079 mutex_lock(&fs_devices
->device_list_mutex
);
7080 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
7081 if (!device
->dev_stats_valid
|| !btrfs_dev_stats_dirty(device
))
7084 stats_cnt
= atomic_read(&device
->dev_stats_ccnt
);
7085 ret
= update_dev_stat_item(trans
, fs_info
, device
);
7087 atomic_sub(stats_cnt
, &device
->dev_stats_ccnt
);
7089 mutex_unlock(&fs_devices
->device_list_mutex
);
7094 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
7096 btrfs_dev_stat_inc(dev
, index
);
7097 btrfs_dev_stat_print_on_error(dev
);
7100 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
7102 if (!dev
->dev_stats_valid
)
7104 btrfs_err_rl_in_rcu(dev
->fs_info
,
7105 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7106 rcu_str_deref(dev
->name
),
7107 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
7108 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
7109 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
7110 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
7111 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
7114 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
7118 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7119 if (btrfs_dev_stat_read(dev
, i
) != 0)
7121 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
7122 return; /* all values == 0, suppress message */
7124 btrfs_info_in_rcu(dev
->fs_info
,
7125 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7126 rcu_str_deref(dev
->name
),
7127 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
7128 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
7129 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
7130 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
7131 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
7134 int btrfs_get_dev_stats(struct btrfs_fs_info
*fs_info
,
7135 struct btrfs_ioctl_get_dev_stats
*stats
)
7137 struct btrfs_device
*dev
;
7138 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7141 mutex_lock(&fs_devices
->device_list_mutex
);
7142 dev
= btrfs_find_device(fs_info
, stats
->devid
, NULL
, NULL
);
7143 mutex_unlock(&fs_devices
->device_list_mutex
);
7146 btrfs_warn(fs_info
, "get dev_stats failed, device not found");
7148 } else if (!dev
->dev_stats_valid
) {
7149 btrfs_warn(fs_info
, "get dev_stats failed, not yet valid");
7151 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
7152 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
7153 if (stats
->nr_items
> i
)
7155 btrfs_dev_stat_read_and_reset(dev
, i
);
7157 btrfs_dev_stat_reset(dev
, i
);
7160 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7161 if (stats
->nr_items
> i
)
7162 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
7164 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
7165 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
7169 void btrfs_scratch_superblocks(struct block_device
*bdev
, const char *device_path
)
7171 struct buffer_head
*bh
;
7172 struct btrfs_super_block
*disk_super
;
7178 for (copy_num
= 0; copy_num
< BTRFS_SUPER_MIRROR_MAX
;
7181 if (btrfs_read_dev_one_super(bdev
, copy_num
, &bh
))
7184 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
7186 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
7187 set_buffer_dirty(bh
);
7188 sync_dirty_buffer(bh
);
7192 /* Notify udev that device has changed */
7193 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
7195 /* Update ctime/mtime for device path for libblkid */
7196 update_dev_time(device_path
);
7200 * Update the size of all devices, which is used for writing out the
7203 void btrfs_update_commit_device_size(struct btrfs_fs_info
*fs_info
)
7205 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7206 struct btrfs_device
*curr
, *next
;
7208 if (list_empty(&fs_devices
->resized_devices
))
7211 mutex_lock(&fs_devices
->device_list_mutex
);
7212 mutex_lock(&fs_info
->chunk_mutex
);
7213 list_for_each_entry_safe(curr
, next
, &fs_devices
->resized_devices
,
7215 list_del_init(&curr
->resized_list
);
7216 curr
->commit_total_bytes
= curr
->disk_total_bytes
;
7218 mutex_unlock(&fs_info
->chunk_mutex
);
7219 mutex_unlock(&fs_devices
->device_list_mutex
);
7222 /* Must be invoked during the transaction commit */
7223 void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info
*fs_info
,
7224 struct btrfs_transaction
*transaction
)
7226 struct extent_map
*em
;
7227 struct map_lookup
*map
;
7228 struct btrfs_device
*dev
;
7231 if (list_empty(&transaction
->pending_chunks
))
7234 /* In order to kick the device replace finish process */
7235 mutex_lock(&fs_info
->chunk_mutex
);
7236 list_for_each_entry(em
, &transaction
->pending_chunks
, list
) {
7237 map
= em
->map_lookup
;
7239 for (i
= 0; i
< map
->num_stripes
; i
++) {
7240 dev
= map
->stripes
[i
].dev
;
7241 dev
->commit_bytes_used
= dev
->bytes_used
;
7244 mutex_unlock(&fs_info
->chunk_mutex
);
7247 void btrfs_set_fs_info_ptr(struct btrfs_fs_info
*fs_info
)
7249 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7250 while (fs_devices
) {
7251 fs_devices
->fs_info
= fs_info
;
7252 fs_devices
= fs_devices
->seed
;
7256 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info
*fs_info
)
7258 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7259 while (fs_devices
) {
7260 fs_devices
->fs_info
= NULL
;
7261 fs_devices
= fs_devices
->seed
;