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/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/ratelimit.h>
27 #include <linux/kthread.h>
28 #include <linux/raid/pq.h>
29 #include <linux/semaphore.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 static int init_first_rw_device(struct btrfs_trans_handle
*trans
,
46 struct btrfs_root
*root
,
47 struct btrfs_device
*device
);
48 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
);
49 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
);
50 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
);
51 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*device
);
53 DEFINE_MUTEX(uuid_mutex
);
54 static LIST_HEAD(fs_uuids
);
56 static struct btrfs_fs_devices
*__alloc_fs_devices(void)
58 struct btrfs_fs_devices
*fs_devs
;
60 fs_devs
= kzalloc(sizeof(*fs_devs
), GFP_NOFS
);
62 return ERR_PTR(-ENOMEM
);
64 mutex_init(&fs_devs
->device_list_mutex
);
66 INIT_LIST_HEAD(&fs_devs
->devices
);
67 INIT_LIST_HEAD(&fs_devs
->resized_devices
);
68 INIT_LIST_HEAD(&fs_devs
->alloc_list
);
69 INIT_LIST_HEAD(&fs_devs
->list
);
75 * alloc_fs_devices - allocate struct btrfs_fs_devices
76 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
79 * Return: a pointer to a new &struct btrfs_fs_devices on success;
80 * ERR_PTR() on error. Returned struct is not linked onto any lists and
81 * can be destroyed with kfree() right away.
83 static struct btrfs_fs_devices
*alloc_fs_devices(const u8
*fsid
)
85 struct btrfs_fs_devices
*fs_devs
;
87 fs_devs
= __alloc_fs_devices();
92 memcpy(fs_devs
->fsid
, fsid
, BTRFS_FSID_SIZE
);
94 generate_random_uuid(fs_devs
->fsid
);
99 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
101 struct btrfs_device
*device
;
102 WARN_ON(fs_devices
->opened
);
103 while (!list_empty(&fs_devices
->devices
)) {
104 device
= list_entry(fs_devices
->devices
.next
,
105 struct btrfs_device
, dev_list
);
106 list_del(&device
->dev_list
);
107 rcu_string_free(device
->name
);
113 static void btrfs_kobject_uevent(struct block_device
*bdev
,
114 enum kobject_action action
)
118 ret
= kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, action
);
120 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
122 kobject_name(&disk_to_dev(bdev
->bd_disk
)->kobj
),
123 &disk_to_dev(bdev
->bd_disk
)->kobj
);
126 void btrfs_cleanup_fs_uuids(void)
128 struct btrfs_fs_devices
*fs_devices
;
130 while (!list_empty(&fs_uuids
)) {
131 fs_devices
= list_entry(fs_uuids
.next
,
132 struct btrfs_fs_devices
, list
);
133 list_del(&fs_devices
->list
);
134 free_fs_devices(fs_devices
);
138 static struct btrfs_device
*__alloc_device(void)
140 struct btrfs_device
*dev
;
142 dev
= kzalloc(sizeof(*dev
), GFP_NOFS
);
144 return ERR_PTR(-ENOMEM
);
146 INIT_LIST_HEAD(&dev
->dev_list
);
147 INIT_LIST_HEAD(&dev
->dev_alloc_list
);
148 INIT_LIST_HEAD(&dev
->resized_list
);
150 spin_lock_init(&dev
->io_lock
);
152 spin_lock_init(&dev
->reada_lock
);
153 atomic_set(&dev
->reada_in_flight
, 0);
154 atomic_set(&dev
->dev_stats_ccnt
, 0);
155 INIT_RADIX_TREE(&dev
->reada_zones
, GFP_NOFS
& ~__GFP_WAIT
);
156 INIT_RADIX_TREE(&dev
->reada_extents
, GFP_NOFS
& ~__GFP_WAIT
);
161 static noinline
struct btrfs_device
*__find_device(struct list_head
*head
,
164 struct btrfs_device
*dev
;
166 list_for_each_entry(dev
, head
, dev_list
) {
167 if (dev
->devid
== devid
&&
168 (!uuid
|| !memcmp(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
))) {
175 static noinline
struct btrfs_fs_devices
*find_fsid(u8
*fsid
)
177 struct btrfs_fs_devices
*fs_devices
;
179 list_for_each_entry(fs_devices
, &fs_uuids
, list
) {
180 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
187 btrfs_get_bdev_and_sb(const char *device_path
, fmode_t flags
, void *holder
,
188 int flush
, struct block_device
**bdev
,
189 struct buffer_head
**bh
)
193 *bdev
= blkdev_get_by_path(device_path
, flags
, holder
);
196 ret
= PTR_ERR(*bdev
);
197 printk(KERN_INFO
"BTRFS: open %s failed\n", device_path
);
202 filemap_write_and_wait((*bdev
)->bd_inode
->i_mapping
);
203 ret
= set_blocksize(*bdev
, 4096);
205 blkdev_put(*bdev
, flags
);
208 invalidate_bdev(*bdev
);
209 *bh
= btrfs_read_dev_super(*bdev
);
212 blkdev_put(*bdev
, flags
);
224 static void requeue_list(struct btrfs_pending_bios
*pending_bios
,
225 struct bio
*head
, struct bio
*tail
)
228 struct bio
*old_head
;
230 old_head
= pending_bios
->head
;
231 pending_bios
->head
= head
;
232 if (pending_bios
->tail
)
233 tail
->bi_next
= old_head
;
235 pending_bios
->tail
= tail
;
239 * we try to collect pending bios for a device so we don't get a large
240 * number of procs sending bios down to the same device. This greatly
241 * improves the schedulers ability to collect and merge the bios.
243 * But, it also turns into a long list of bios to process and that is sure
244 * to eventually make the worker thread block. The solution here is to
245 * make some progress and then put this work struct back at the end of
246 * the list if the block device is congested. This way, multiple devices
247 * can make progress from a single worker thread.
249 static noinline
void run_scheduled_bios(struct btrfs_device
*device
)
252 struct backing_dev_info
*bdi
;
253 struct btrfs_fs_info
*fs_info
;
254 struct btrfs_pending_bios
*pending_bios
;
258 unsigned long num_run
;
259 unsigned long batch_run
= 0;
261 unsigned long last_waited
= 0;
263 int sync_pending
= 0;
264 struct blk_plug plug
;
267 * this function runs all the bios we've collected for
268 * a particular device. We don't want to wander off to
269 * another device without first sending all of these down.
270 * So, setup a plug here and finish it off before we return
272 blk_start_plug(&plug
);
274 bdi
= blk_get_backing_dev_info(device
->bdev
);
275 fs_info
= device
->dev_root
->fs_info
;
276 limit
= btrfs_async_submit_limit(fs_info
);
277 limit
= limit
* 2 / 3;
280 spin_lock(&device
->io_lock
);
285 /* take all the bios off the list at once and process them
286 * later on (without the lock held). But, remember the
287 * tail and other pointers so the bios can be properly reinserted
288 * into the list if we hit congestion
290 if (!force_reg
&& device
->pending_sync_bios
.head
) {
291 pending_bios
= &device
->pending_sync_bios
;
294 pending_bios
= &device
->pending_bios
;
298 pending
= pending_bios
->head
;
299 tail
= pending_bios
->tail
;
300 WARN_ON(pending
&& !tail
);
303 * if pending was null this time around, no bios need processing
304 * at all and we can stop. Otherwise it'll loop back up again
305 * and do an additional check so no bios are missed.
307 * device->running_pending is used to synchronize with the
310 if (device
->pending_sync_bios
.head
== NULL
&&
311 device
->pending_bios
.head
== NULL
) {
313 device
->running_pending
= 0;
316 device
->running_pending
= 1;
319 pending_bios
->head
= NULL
;
320 pending_bios
->tail
= NULL
;
322 spin_unlock(&device
->io_lock
);
327 /* we want to work on both lists, but do more bios on the
328 * sync list than the regular list
331 pending_bios
!= &device
->pending_sync_bios
&&
332 device
->pending_sync_bios
.head
) ||
333 (num_run
> 64 && pending_bios
== &device
->pending_sync_bios
&&
334 device
->pending_bios
.head
)) {
335 spin_lock(&device
->io_lock
);
336 requeue_list(pending_bios
, pending
, tail
);
341 pending
= pending
->bi_next
;
344 if (atomic_dec_return(&fs_info
->nr_async_bios
) < limit
&&
345 waitqueue_active(&fs_info
->async_submit_wait
))
346 wake_up(&fs_info
->async_submit_wait
);
348 BUG_ON(atomic_read(&cur
->bi_cnt
) == 0);
351 * if we're doing the sync list, record that our
352 * plug has some sync requests on it
354 * If we're doing the regular list and there are
355 * sync requests sitting around, unplug before
358 if (pending_bios
== &device
->pending_sync_bios
) {
360 } else if (sync_pending
) {
361 blk_finish_plug(&plug
);
362 blk_start_plug(&plug
);
366 btrfsic_submit_bio(cur
->bi_rw
, cur
);
373 * we made progress, there is more work to do and the bdi
374 * is now congested. Back off and let other work structs
377 if (pending
&& bdi_write_congested(bdi
) && batch_run
> 8 &&
378 fs_info
->fs_devices
->open_devices
> 1) {
379 struct io_context
*ioc
;
381 ioc
= current
->io_context
;
384 * the main goal here is that we don't want to
385 * block if we're going to be able to submit
386 * more requests without blocking.
388 * This code does two great things, it pokes into
389 * the elevator code from a filesystem _and_
390 * it makes assumptions about how batching works.
392 if (ioc
&& ioc
->nr_batch_requests
> 0 &&
393 time_before(jiffies
, ioc
->last_waited
+ HZ
/50UL) &&
395 ioc
->last_waited
== last_waited
)) {
397 * we want to go through our batch of
398 * requests and stop. So, we copy out
399 * the ioc->last_waited time and test
400 * against it before looping
402 last_waited
= ioc
->last_waited
;
407 spin_lock(&device
->io_lock
);
408 requeue_list(pending_bios
, pending
, tail
);
409 device
->running_pending
= 1;
411 spin_unlock(&device
->io_lock
);
412 btrfs_queue_work(fs_info
->submit_workers
,
416 /* unplug every 64 requests just for good measure */
417 if (batch_run
% 64 == 0) {
418 blk_finish_plug(&plug
);
419 blk_start_plug(&plug
);
428 spin_lock(&device
->io_lock
);
429 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
431 spin_unlock(&device
->io_lock
);
434 blk_finish_plug(&plug
);
437 static void pending_bios_fn(struct btrfs_work
*work
)
439 struct btrfs_device
*device
;
441 device
= container_of(work
, struct btrfs_device
, work
);
442 run_scheduled_bios(device
);
446 * Add new device to list of registered devices
449 * 1 - first time device is seen
450 * 0 - device already known
453 static noinline
int device_list_add(const char *path
,
454 struct btrfs_super_block
*disk_super
,
455 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
457 struct btrfs_device
*device
;
458 struct btrfs_fs_devices
*fs_devices
;
459 struct rcu_string
*name
;
461 u64 found_transid
= btrfs_super_generation(disk_super
);
463 fs_devices
= find_fsid(disk_super
->fsid
);
465 fs_devices
= alloc_fs_devices(disk_super
->fsid
);
466 if (IS_ERR(fs_devices
))
467 return PTR_ERR(fs_devices
);
469 list_add(&fs_devices
->list
, &fs_uuids
);
473 device
= __find_device(&fs_devices
->devices
, devid
,
474 disk_super
->dev_item
.uuid
);
478 if (fs_devices
->opened
)
481 device
= btrfs_alloc_device(NULL
, &devid
,
482 disk_super
->dev_item
.uuid
);
483 if (IS_ERR(device
)) {
484 /* we can safely leave the fs_devices entry around */
485 return PTR_ERR(device
);
488 name
= rcu_string_strdup(path
, GFP_NOFS
);
493 rcu_assign_pointer(device
->name
, name
);
495 mutex_lock(&fs_devices
->device_list_mutex
);
496 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
497 fs_devices
->num_devices
++;
498 mutex_unlock(&fs_devices
->device_list_mutex
);
501 device
->fs_devices
= fs_devices
;
502 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
504 * When FS is already mounted.
505 * 1. If you are here and if the device->name is NULL that
506 * means this device was missing at time of FS mount.
507 * 2. If you are here and if the device->name is different
508 * from 'path' that means either
509 * a. The same device disappeared and reappeared with
511 * b. The missing-disk-which-was-replaced, has
514 * We must allow 1 and 2a above. But 2b would be a spurious
517 * Further in case of 1 and 2a above, the disk at 'path'
518 * would have missed some transaction when it was away and
519 * in case of 2a the stale bdev has to be updated as well.
520 * 2b must not be allowed at all time.
524 * For now, we do allow update to btrfs_fs_device through the
525 * btrfs dev scan cli after FS has been mounted. We're still
526 * tracking a problem where systems fail mount by subvolume id
527 * when we reject replacement on a mounted FS.
529 if (!fs_devices
->opened
&& found_transid
< device
->generation
) {
531 * That is if the FS is _not_ mounted and if you
532 * are here, that means there is more than one
533 * disk with same uuid and devid.We keep the one
534 * with larger generation number or the last-in if
535 * generation are equal.
540 name
= rcu_string_strdup(path
, GFP_NOFS
);
543 rcu_string_free(device
->name
);
544 rcu_assign_pointer(device
->name
, name
);
545 if (device
->missing
) {
546 fs_devices
->missing_devices
--;
552 * Unmount does not free the btrfs_device struct but would zero
553 * generation along with most of the other members. So just update
554 * it back. We need it to pick the disk with largest generation
557 if (!fs_devices
->opened
)
558 device
->generation
= found_transid
;
560 *fs_devices_ret
= fs_devices
;
565 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
567 struct btrfs_fs_devices
*fs_devices
;
568 struct btrfs_device
*device
;
569 struct btrfs_device
*orig_dev
;
571 fs_devices
= alloc_fs_devices(orig
->fsid
);
572 if (IS_ERR(fs_devices
))
575 mutex_lock(&orig
->device_list_mutex
);
576 fs_devices
->total_devices
= orig
->total_devices
;
578 /* We have held the volume lock, it is safe to get the devices. */
579 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
580 struct rcu_string
*name
;
582 device
= btrfs_alloc_device(NULL
, &orig_dev
->devid
,
588 * This is ok to do without rcu read locked because we hold the
589 * uuid mutex so nothing we touch in here is going to disappear.
591 if (orig_dev
->name
) {
592 name
= rcu_string_strdup(orig_dev
->name
->str
, GFP_NOFS
);
597 rcu_assign_pointer(device
->name
, name
);
600 list_add(&device
->dev_list
, &fs_devices
->devices
);
601 device
->fs_devices
= fs_devices
;
602 fs_devices
->num_devices
++;
604 mutex_unlock(&orig
->device_list_mutex
);
607 mutex_unlock(&orig
->device_list_mutex
);
608 free_fs_devices(fs_devices
);
609 return ERR_PTR(-ENOMEM
);
612 void btrfs_close_extra_devices(struct btrfs_fs_devices
*fs_devices
, int step
)
614 struct btrfs_device
*device
, *next
;
615 struct btrfs_device
*latest_dev
= NULL
;
617 mutex_lock(&uuid_mutex
);
619 /* This is the initialized path, it is safe to release the devices. */
620 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
621 if (device
->in_fs_metadata
) {
622 if (!device
->is_tgtdev_for_dev_replace
&&
624 device
->generation
> latest_dev
->generation
)) {
630 if (device
->devid
== BTRFS_DEV_REPLACE_DEVID
) {
632 * In the first step, keep the device which has
633 * the correct fsid and the devid that is used
634 * for the dev_replace procedure.
635 * In the second step, the dev_replace state is
636 * read from the device tree and it is known
637 * whether the procedure is really active or
638 * not, which means whether this device is
639 * used or whether it should be removed.
641 if (step
== 0 || device
->is_tgtdev_for_dev_replace
) {
646 blkdev_put(device
->bdev
, device
->mode
);
648 fs_devices
->open_devices
--;
650 if (device
->writeable
) {
651 list_del_init(&device
->dev_alloc_list
);
652 device
->writeable
= 0;
653 if (!device
->is_tgtdev_for_dev_replace
)
654 fs_devices
->rw_devices
--;
656 list_del_init(&device
->dev_list
);
657 fs_devices
->num_devices
--;
658 rcu_string_free(device
->name
);
662 if (fs_devices
->seed
) {
663 fs_devices
= fs_devices
->seed
;
667 fs_devices
->latest_bdev
= latest_dev
->bdev
;
669 mutex_unlock(&uuid_mutex
);
672 static void __free_device(struct work_struct
*work
)
674 struct btrfs_device
*device
;
676 device
= container_of(work
, struct btrfs_device
, rcu_work
);
679 blkdev_put(device
->bdev
, device
->mode
);
681 rcu_string_free(device
->name
);
685 static void free_device(struct rcu_head
*head
)
687 struct btrfs_device
*device
;
689 device
= container_of(head
, struct btrfs_device
, rcu
);
691 INIT_WORK(&device
->rcu_work
, __free_device
);
692 schedule_work(&device
->rcu_work
);
695 static int __btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
697 struct btrfs_device
*device
;
699 if (--fs_devices
->opened
> 0)
702 mutex_lock(&fs_devices
->device_list_mutex
);
703 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
704 struct btrfs_device
*new_device
;
705 struct rcu_string
*name
;
708 fs_devices
->open_devices
--;
710 if (device
->writeable
&&
711 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
712 list_del_init(&device
->dev_alloc_list
);
713 fs_devices
->rw_devices
--;
717 fs_devices
->missing_devices
--;
719 new_device
= btrfs_alloc_device(NULL
, &device
->devid
,
721 BUG_ON(IS_ERR(new_device
)); /* -ENOMEM */
723 /* Safe because we are under uuid_mutex */
725 name
= rcu_string_strdup(device
->name
->str
, GFP_NOFS
);
726 BUG_ON(!name
); /* -ENOMEM */
727 rcu_assign_pointer(new_device
->name
, name
);
730 list_replace_rcu(&device
->dev_list
, &new_device
->dev_list
);
731 new_device
->fs_devices
= device
->fs_devices
;
733 call_rcu(&device
->rcu
, free_device
);
735 mutex_unlock(&fs_devices
->device_list_mutex
);
737 WARN_ON(fs_devices
->open_devices
);
738 WARN_ON(fs_devices
->rw_devices
);
739 fs_devices
->opened
= 0;
740 fs_devices
->seeding
= 0;
745 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
747 struct btrfs_fs_devices
*seed_devices
= NULL
;
750 mutex_lock(&uuid_mutex
);
751 ret
= __btrfs_close_devices(fs_devices
);
752 if (!fs_devices
->opened
) {
753 seed_devices
= fs_devices
->seed
;
754 fs_devices
->seed
= NULL
;
756 mutex_unlock(&uuid_mutex
);
758 while (seed_devices
) {
759 fs_devices
= seed_devices
;
760 seed_devices
= fs_devices
->seed
;
761 __btrfs_close_devices(fs_devices
);
762 free_fs_devices(fs_devices
);
765 * Wait for rcu kworkers under __btrfs_close_devices
766 * to finish all blkdev_puts so device is really
767 * free when umount is done.
773 static int __btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
774 fmode_t flags
, void *holder
)
776 struct request_queue
*q
;
777 struct block_device
*bdev
;
778 struct list_head
*head
= &fs_devices
->devices
;
779 struct btrfs_device
*device
;
780 struct btrfs_device
*latest_dev
= NULL
;
781 struct buffer_head
*bh
;
782 struct btrfs_super_block
*disk_super
;
789 list_for_each_entry(device
, head
, dev_list
) {
795 /* Just open everything we can; ignore failures here */
796 if (btrfs_get_bdev_and_sb(device
->name
->str
, flags
, holder
, 1,
800 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
801 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
802 if (devid
!= device
->devid
)
805 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
,
809 device
->generation
= btrfs_super_generation(disk_super
);
811 device
->generation
> latest_dev
->generation
)
814 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
815 device
->writeable
= 0;
817 device
->writeable
= !bdev_read_only(bdev
);
821 q
= bdev_get_queue(bdev
);
822 if (blk_queue_discard(q
))
823 device
->can_discard
= 1;
826 device
->in_fs_metadata
= 0;
827 device
->mode
= flags
;
829 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
830 fs_devices
->rotating
= 1;
832 fs_devices
->open_devices
++;
833 if (device
->writeable
&&
834 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
835 fs_devices
->rw_devices
++;
836 list_add(&device
->dev_alloc_list
,
837 &fs_devices
->alloc_list
);
844 blkdev_put(bdev
, flags
);
847 if (fs_devices
->open_devices
== 0) {
851 fs_devices
->seeding
= seeding
;
852 fs_devices
->opened
= 1;
853 fs_devices
->latest_bdev
= latest_dev
->bdev
;
854 fs_devices
->total_rw_bytes
= 0;
859 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
860 fmode_t flags
, void *holder
)
864 mutex_lock(&uuid_mutex
);
865 if (fs_devices
->opened
) {
866 fs_devices
->opened
++;
869 ret
= __btrfs_open_devices(fs_devices
, flags
, holder
);
871 mutex_unlock(&uuid_mutex
);
876 * Look for a btrfs signature on a device. This may be called out of the mount path
877 * and we are not allowed to call set_blocksize during the scan. The superblock
878 * is read via pagecache
880 int btrfs_scan_one_device(const char *path
, fmode_t flags
, void *holder
,
881 struct btrfs_fs_devices
**fs_devices_ret
)
883 struct btrfs_super_block
*disk_super
;
884 struct block_device
*bdev
;
895 * we would like to check all the supers, but that would make
896 * a btrfs mount succeed after a mkfs from a different FS.
897 * So, we need to add a special mount option to scan for
898 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
900 bytenr
= btrfs_sb_offset(0);
902 mutex_lock(&uuid_mutex
);
904 bdev
= blkdev_get_by_path(path
, flags
, holder
);
911 /* make sure our super fits in the device */
912 if (bytenr
+ PAGE_CACHE_SIZE
>= i_size_read(bdev
->bd_inode
))
915 /* make sure our super fits in the page */
916 if (sizeof(*disk_super
) > PAGE_CACHE_SIZE
)
919 /* make sure our super doesn't straddle pages on disk */
920 index
= bytenr
>> PAGE_CACHE_SHIFT
;
921 if ((bytenr
+ sizeof(*disk_super
) - 1) >> PAGE_CACHE_SHIFT
!= index
)
924 /* pull in the page with our super */
925 page
= read_cache_page_gfp(bdev
->bd_inode
->i_mapping
,
928 if (IS_ERR_OR_NULL(page
))
933 /* align our pointer to the offset of the super block */
934 disk_super
= p
+ (bytenr
& ~PAGE_CACHE_MASK
);
936 if (btrfs_super_bytenr(disk_super
) != bytenr
||
937 btrfs_super_magic(disk_super
) != BTRFS_MAGIC
)
940 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
941 transid
= btrfs_super_generation(disk_super
);
942 total_devices
= btrfs_super_num_devices(disk_super
);
944 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
946 if (disk_super
->label
[0]) {
947 if (disk_super
->label
[BTRFS_LABEL_SIZE
- 1])
948 disk_super
->label
[BTRFS_LABEL_SIZE
- 1] = '\0';
949 printk(KERN_INFO
"BTRFS: device label %s ", disk_super
->label
);
951 printk(KERN_INFO
"BTRFS: device fsid %pU ", disk_super
->fsid
);
954 printk(KERN_CONT
"devid %llu transid %llu %s\n", devid
, transid
, path
);
957 if (!ret
&& fs_devices_ret
)
958 (*fs_devices_ret
)->total_devices
= total_devices
;
962 page_cache_release(page
);
965 blkdev_put(bdev
, flags
);
967 mutex_unlock(&uuid_mutex
);
971 /* helper to account the used device space in the range */
972 int btrfs_account_dev_extents_size(struct btrfs_device
*device
, u64 start
,
973 u64 end
, u64
*length
)
975 struct btrfs_key key
;
976 struct btrfs_root
*root
= device
->dev_root
;
977 struct btrfs_dev_extent
*dev_extent
;
978 struct btrfs_path
*path
;
982 struct extent_buffer
*l
;
986 if (start
>= device
->total_bytes
|| device
->is_tgtdev_for_dev_replace
)
989 path
= btrfs_alloc_path();
994 key
.objectid
= device
->devid
;
996 key
.type
= BTRFS_DEV_EXTENT_KEY
;
998 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1002 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1009 slot
= path
->slots
[0];
1010 if (slot
>= btrfs_header_nritems(l
)) {
1011 ret
= btrfs_next_leaf(root
, path
);
1019 btrfs_item_key_to_cpu(l
, &key
, slot
);
1021 if (key
.objectid
< device
->devid
)
1024 if (key
.objectid
> device
->devid
)
1027 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1030 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1031 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1033 if (key
.offset
<= start
&& extent_end
> end
) {
1034 *length
= end
- start
+ 1;
1036 } else if (key
.offset
<= start
&& extent_end
> start
)
1037 *length
+= extent_end
- start
;
1038 else if (key
.offset
> start
&& extent_end
<= end
)
1039 *length
+= extent_end
- key
.offset
;
1040 else if (key
.offset
> start
&& key
.offset
<= end
) {
1041 *length
+= end
- key
.offset
+ 1;
1043 } else if (key
.offset
> end
)
1051 btrfs_free_path(path
);
1055 static int contains_pending_extent(struct btrfs_trans_handle
*trans
,
1056 struct btrfs_device
*device
,
1057 u64
*start
, u64 len
)
1059 struct extent_map
*em
;
1060 struct list_head
*search_list
= &trans
->transaction
->pending_chunks
;
1064 list_for_each_entry(em
, search_list
, list
) {
1065 struct map_lookup
*map
;
1068 map
= (struct map_lookup
*)em
->bdev
;
1069 for (i
= 0; i
< map
->num_stripes
; i
++) {
1070 if (map
->stripes
[i
].dev
!= device
)
1072 if (map
->stripes
[i
].physical
>= *start
+ len
||
1073 map
->stripes
[i
].physical
+ em
->orig_block_len
<=
1076 *start
= map
->stripes
[i
].physical
+
1081 if (search_list
== &trans
->transaction
->pending_chunks
) {
1082 search_list
= &trans
->root
->fs_info
->pinned_chunks
;
1091 * find_free_dev_extent - find free space in the specified device
1092 * @device: the device which we search the free space in
1093 * @num_bytes: the size of the free space that we need
1094 * @start: store the start of the free space.
1095 * @len: the size of the free space. that we find, or the size of the max
1096 * free space if we don't find suitable free space
1098 * this uses a pretty simple search, the expectation is that it is
1099 * called very infrequently and that a given device has a small number
1102 * @start is used to store the start of the free space if we find. But if we
1103 * don't find suitable free space, it will be used to store the start position
1104 * of the max free space.
1106 * @len is used to store the size of the free space that we find.
1107 * But if we don't find suitable free space, it is used to store the size of
1108 * the max free space.
1110 int find_free_dev_extent(struct btrfs_trans_handle
*trans
,
1111 struct btrfs_device
*device
, u64 num_bytes
,
1112 u64
*start
, u64
*len
)
1114 struct btrfs_key key
;
1115 struct btrfs_root
*root
= device
->dev_root
;
1116 struct btrfs_dev_extent
*dev_extent
;
1117 struct btrfs_path
*path
;
1123 u64 search_end
= device
->total_bytes
;
1126 struct extent_buffer
*l
;
1128 /* FIXME use last free of some kind */
1130 /* we don't want to overwrite the superblock on the drive,
1131 * so we make sure to start at an offset of at least 1MB
1133 search_start
= max(root
->fs_info
->alloc_start
, 1024ull * 1024);
1135 path
= btrfs_alloc_path();
1139 max_hole_start
= search_start
;
1143 if (search_start
>= search_end
|| device
->is_tgtdev_for_dev_replace
) {
1149 path
->search_commit_root
= 1;
1150 path
->skip_locking
= 1;
1152 key
.objectid
= device
->devid
;
1153 key
.offset
= search_start
;
1154 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1156 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1160 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1167 slot
= path
->slots
[0];
1168 if (slot
>= btrfs_header_nritems(l
)) {
1169 ret
= btrfs_next_leaf(root
, path
);
1177 btrfs_item_key_to_cpu(l
, &key
, slot
);
1179 if (key
.objectid
< device
->devid
)
1182 if (key
.objectid
> device
->devid
)
1185 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1188 if (key
.offset
> search_start
) {
1189 hole_size
= key
.offset
- search_start
;
1192 * Have to check before we set max_hole_start, otherwise
1193 * we could end up sending back this offset anyway.
1195 if (contains_pending_extent(trans
, device
,
1200 if (hole_size
> max_hole_size
) {
1201 max_hole_start
= search_start
;
1202 max_hole_size
= hole_size
;
1206 * If this free space is greater than which we need,
1207 * it must be the max free space that we have found
1208 * until now, so max_hole_start must point to the start
1209 * of this free space and the length of this free space
1210 * is stored in max_hole_size. Thus, we return
1211 * max_hole_start and max_hole_size and go back to the
1214 if (hole_size
>= num_bytes
) {
1220 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1221 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1223 if (extent_end
> search_start
)
1224 search_start
= extent_end
;
1231 * At this point, search_start should be the end of
1232 * allocated dev extents, and when shrinking the device,
1233 * search_end may be smaller than search_start.
1235 if (search_end
> search_start
)
1236 hole_size
= search_end
- search_start
;
1238 if (hole_size
> max_hole_size
) {
1239 max_hole_start
= search_start
;
1240 max_hole_size
= hole_size
;
1243 if (contains_pending_extent(trans
, device
, &search_start
, hole_size
)) {
1244 btrfs_release_path(path
);
1249 if (hole_size
< num_bytes
)
1255 btrfs_free_path(path
);
1256 *start
= max_hole_start
;
1258 *len
= max_hole_size
;
1262 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1263 struct btrfs_device
*device
,
1264 u64 start
, u64
*dev_extent_len
)
1267 struct btrfs_path
*path
;
1268 struct btrfs_root
*root
= device
->dev_root
;
1269 struct btrfs_key key
;
1270 struct btrfs_key found_key
;
1271 struct extent_buffer
*leaf
= NULL
;
1272 struct btrfs_dev_extent
*extent
= NULL
;
1274 path
= btrfs_alloc_path();
1278 key
.objectid
= device
->devid
;
1280 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1282 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1284 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1285 BTRFS_DEV_EXTENT_KEY
);
1288 leaf
= path
->nodes
[0];
1289 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1290 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1291 struct btrfs_dev_extent
);
1292 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1293 btrfs_dev_extent_length(leaf
, extent
) < start
);
1295 btrfs_release_path(path
);
1297 } else if (ret
== 0) {
1298 leaf
= path
->nodes
[0];
1299 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1300 struct btrfs_dev_extent
);
1302 btrfs_error(root
->fs_info
, ret
, "Slot search failed");
1306 *dev_extent_len
= btrfs_dev_extent_length(leaf
, extent
);
1308 ret
= btrfs_del_item(trans
, root
, path
);
1310 btrfs_error(root
->fs_info
, ret
,
1311 "Failed to remove dev extent item");
1313 trans
->transaction
->have_free_bgs
= 1;
1316 btrfs_free_path(path
);
1320 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1321 struct btrfs_device
*device
,
1322 u64 chunk_tree
, u64 chunk_objectid
,
1323 u64 chunk_offset
, u64 start
, u64 num_bytes
)
1326 struct btrfs_path
*path
;
1327 struct btrfs_root
*root
= device
->dev_root
;
1328 struct btrfs_dev_extent
*extent
;
1329 struct extent_buffer
*leaf
;
1330 struct btrfs_key key
;
1332 WARN_ON(!device
->in_fs_metadata
);
1333 WARN_ON(device
->is_tgtdev_for_dev_replace
);
1334 path
= btrfs_alloc_path();
1338 key
.objectid
= device
->devid
;
1340 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1341 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1346 leaf
= path
->nodes
[0];
1347 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1348 struct btrfs_dev_extent
);
1349 btrfs_set_dev_extent_chunk_tree(leaf
, extent
, chunk_tree
);
1350 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
, chunk_objectid
);
1351 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1353 write_extent_buffer(leaf
, root
->fs_info
->chunk_tree_uuid
,
1354 btrfs_dev_extent_chunk_tree_uuid(extent
), BTRFS_UUID_SIZE
);
1356 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1357 btrfs_mark_buffer_dirty(leaf
);
1359 btrfs_free_path(path
);
1363 static u64
find_next_chunk(struct btrfs_fs_info
*fs_info
)
1365 struct extent_map_tree
*em_tree
;
1366 struct extent_map
*em
;
1370 em_tree
= &fs_info
->mapping_tree
.map_tree
;
1371 read_lock(&em_tree
->lock
);
1372 n
= rb_last(&em_tree
->map
);
1374 em
= rb_entry(n
, struct extent_map
, rb_node
);
1375 ret
= em
->start
+ em
->len
;
1377 read_unlock(&em_tree
->lock
);
1382 static noinline
int find_next_devid(struct btrfs_fs_info
*fs_info
,
1386 struct btrfs_key key
;
1387 struct btrfs_key found_key
;
1388 struct btrfs_path
*path
;
1390 path
= btrfs_alloc_path();
1394 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1395 key
.type
= BTRFS_DEV_ITEM_KEY
;
1396 key
.offset
= (u64
)-1;
1398 ret
= btrfs_search_slot(NULL
, fs_info
->chunk_root
, &key
, path
, 0, 0);
1402 BUG_ON(ret
== 0); /* Corruption */
1404 ret
= btrfs_previous_item(fs_info
->chunk_root
, path
,
1405 BTRFS_DEV_ITEMS_OBJECTID
,
1406 BTRFS_DEV_ITEM_KEY
);
1410 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1412 *devid_ret
= found_key
.offset
+ 1;
1416 btrfs_free_path(path
);
1421 * the device information is stored in the chunk root
1422 * the btrfs_device struct should be fully filled in
1424 static int btrfs_add_device(struct btrfs_trans_handle
*trans
,
1425 struct btrfs_root
*root
,
1426 struct btrfs_device
*device
)
1429 struct btrfs_path
*path
;
1430 struct btrfs_dev_item
*dev_item
;
1431 struct extent_buffer
*leaf
;
1432 struct btrfs_key key
;
1435 root
= root
->fs_info
->chunk_root
;
1437 path
= btrfs_alloc_path();
1441 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1442 key
.type
= BTRFS_DEV_ITEM_KEY
;
1443 key
.offset
= device
->devid
;
1445 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1450 leaf
= path
->nodes
[0];
1451 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1453 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1454 btrfs_set_device_generation(leaf
, dev_item
, 0);
1455 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1456 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1457 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1458 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1459 btrfs_set_device_total_bytes(leaf
, dev_item
,
1460 btrfs_device_get_disk_total_bytes(device
));
1461 btrfs_set_device_bytes_used(leaf
, dev_item
,
1462 btrfs_device_get_bytes_used(device
));
1463 btrfs_set_device_group(leaf
, dev_item
, 0);
1464 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1465 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1466 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1468 ptr
= btrfs_device_uuid(dev_item
);
1469 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1470 ptr
= btrfs_device_fsid(dev_item
);
1471 write_extent_buffer(leaf
, root
->fs_info
->fsid
, ptr
, BTRFS_UUID_SIZE
);
1472 btrfs_mark_buffer_dirty(leaf
);
1476 btrfs_free_path(path
);
1481 * Function to update ctime/mtime for a given device path.
1482 * Mainly used for ctime/mtime based probe like libblkid.
1484 static void update_dev_time(char *path_name
)
1488 filp
= filp_open(path_name
, O_RDWR
, 0);
1491 file_update_time(filp
);
1492 filp_close(filp
, NULL
);
1496 static int btrfs_rm_dev_item(struct btrfs_root
*root
,
1497 struct btrfs_device
*device
)
1500 struct btrfs_path
*path
;
1501 struct btrfs_key key
;
1502 struct btrfs_trans_handle
*trans
;
1504 root
= root
->fs_info
->chunk_root
;
1506 path
= btrfs_alloc_path();
1510 trans
= btrfs_start_transaction(root
, 0);
1511 if (IS_ERR(trans
)) {
1512 btrfs_free_path(path
);
1513 return PTR_ERR(trans
);
1515 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1516 key
.type
= BTRFS_DEV_ITEM_KEY
;
1517 key
.offset
= device
->devid
;
1519 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1528 ret
= btrfs_del_item(trans
, root
, path
);
1532 btrfs_free_path(path
);
1533 btrfs_commit_transaction(trans
, root
);
1537 int btrfs_rm_device(struct btrfs_root
*root
, char *device_path
)
1539 struct btrfs_device
*device
;
1540 struct btrfs_device
*next_device
;
1541 struct block_device
*bdev
;
1542 struct buffer_head
*bh
= NULL
;
1543 struct btrfs_super_block
*disk_super
;
1544 struct btrfs_fs_devices
*cur_devices
;
1551 bool clear_super
= false;
1553 mutex_lock(&uuid_mutex
);
1556 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
1558 all_avail
= root
->fs_info
->avail_data_alloc_bits
|
1559 root
->fs_info
->avail_system_alloc_bits
|
1560 root
->fs_info
->avail_metadata_alloc_bits
;
1561 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
1563 num_devices
= root
->fs_info
->fs_devices
->num_devices
;
1564 btrfs_dev_replace_lock(&root
->fs_info
->dev_replace
);
1565 if (btrfs_dev_replace_is_ongoing(&root
->fs_info
->dev_replace
)) {
1566 WARN_ON(num_devices
< 1);
1569 btrfs_dev_replace_unlock(&root
->fs_info
->dev_replace
);
1571 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID10
) && num_devices
<= 4) {
1572 ret
= BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET
;
1576 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID1
) && num_devices
<= 2) {
1577 ret
= BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET
;
1581 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID5
) &&
1582 root
->fs_info
->fs_devices
->rw_devices
<= 2) {
1583 ret
= BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET
;
1586 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID6
) &&
1587 root
->fs_info
->fs_devices
->rw_devices
<= 3) {
1588 ret
= BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET
;
1592 if (strcmp(device_path
, "missing") == 0) {
1593 struct list_head
*devices
;
1594 struct btrfs_device
*tmp
;
1597 devices
= &root
->fs_info
->fs_devices
->devices
;
1599 * It is safe to read the devices since the volume_mutex
1602 list_for_each_entry(tmp
, devices
, dev_list
) {
1603 if (tmp
->in_fs_metadata
&&
1604 !tmp
->is_tgtdev_for_dev_replace
&&
1614 ret
= BTRFS_ERROR_DEV_MISSING_NOT_FOUND
;
1618 ret
= btrfs_get_bdev_and_sb(device_path
,
1619 FMODE_WRITE
| FMODE_EXCL
,
1620 root
->fs_info
->bdev_holder
, 0,
1624 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1625 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1626 dev_uuid
= disk_super
->dev_item
.uuid
;
1627 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1635 if (device
->is_tgtdev_for_dev_replace
) {
1636 ret
= BTRFS_ERROR_DEV_TGT_REPLACE
;
1640 if (device
->writeable
&& root
->fs_info
->fs_devices
->rw_devices
== 1) {
1641 ret
= BTRFS_ERROR_DEV_ONLY_WRITABLE
;
1645 if (device
->writeable
) {
1647 list_del_init(&device
->dev_alloc_list
);
1648 device
->fs_devices
->rw_devices
--;
1649 unlock_chunks(root
);
1653 mutex_unlock(&uuid_mutex
);
1654 ret
= btrfs_shrink_device(device
, 0);
1655 mutex_lock(&uuid_mutex
);
1660 * TODO: the superblock still includes this device in its num_devices
1661 * counter although write_all_supers() is not locked out. This
1662 * could give a filesystem state which requires a degraded mount.
1664 ret
= btrfs_rm_dev_item(root
->fs_info
->chunk_root
, device
);
1668 device
->in_fs_metadata
= 0;
1669 btrfs_scrub_cancel_dev(root
->fs_info
, device
);
1672 * the device list mutex makes sure that we don't change
1673 * the device list while someone else is writing out all
1674 * the device supers. Whoever is writing all supers, should
1675 * lock the device list mutex before getting the number of
1676 * devices in the super block (super_copy). Conversely,
1677 * whoever updates the number of devices in the super block
1678 * (super_copy) should hold the device list mutex.
1681 cur_devices
= device
->fs_devices
;
1682 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1683 list_del_rcu(&device
->dev_list
);
1685 device
->fs_devices
->num_devices
--;
1686 device
->fs_devices
->total_devices
--;
1688 if (device
->missing
)
1689 device
->fs_devices
->missing_devices
--;
1691 next_device
= list_entry(root
->fs_info
->fs_devices
->devices
.next
,
1692 struct btrfs_device
, dev_list
);
1693 if (device
->bdev
== root
->fs_info
->sb
->s_bdev
)
1694 root
->fs_info
->sb
->s_bdev
= next_device
->bdev
;
1695 if (device
->bdev
== root
->fs_info
->fs_devices
->latest_bdev
)
1696 root
->fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1699 device
->fs_devices
->open_devices
--;
1700 /* remove sysfs entry */
1701 btrfs_kobj_rm_device(root
->fs_info
, device
);
1704 call_rcu(&device
->rcu
, free_device
);
1706 num_devices
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
1707 btrfs_set_super_num_devices(root
->fs_info
->super_copy
, num_devices
);
1708 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1710 if (cur_devices
->open_devices
== 0) {
1711 struct btrfs_fs_devices
*fs_devices
;
1712 fs_devices
= root
->fs_info
->fs_devices
;
1713 while (fs_devices
) {
1714 if (fs_devices
->seed
== cur_devices
) {
1715 fs_devices
->seed
= cur_devices
->seed
;
1718 fs_devices
= fs_devices
->seed
;
1720 cur_devices
->seed
= NULL
;
1721 __btrfs_close_devices(cur_devices
);
1722 free_fs_devices(cur_devices
);
1725 root
->fs_info
->num_tolerated_disk_barrier_failures
=
1726 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
1729 * at this point, the device is zero sized. We want to
1730 * remove it from the devices list and zero out the old super
1732 if (clear_super
&& disk_super
) {
1736 /* make sure this device isn't detected as part of
1739 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
1740 set_buffer_dirty(bh
);
1741 sync_dirty_buffer(bh
);
1743 /* clear the mirror copies of super block on the disk
1744 * being removed, 0th copy is been taken care above and
1745 * the below would take of the rest
1747 for (i
= 1; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
1748 bytenr
= btrfs_sb_offset(i
);
1749 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
1750 i_size_read(bdev
->bd_inode
))
1754 bh
= __bread(bdev
, bytenr
/ 4096,
1755 BTRFS_SUPER_INFO_SIZE
);
1759 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1761 if (btrfs_super_bytenr(disk_super
) != bytenr
||
1762 btrfs_super_magic(disk_super
) != BTRFS_MAGIC
) {
1765 memset(&disk_super
->magic
, 0,
1766 sizeof(disk_super
->magic
));
1767 set_buffer_dirty(bh
);
1768 sync_dirty_buffer(bh
);
1775 /* Notify udev that device has changed */
1776 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
1778 /* Update ctime/mtime for device path for libblkid */
1779 update_dev_time(device_path
);
1785 blkdev_put(bdev
, FMODE_READ
| FMODE_EXCL
);
1787 mutex_unlock(&uuid_mutex
);
1790 if (device
->writeable
) {
1792 list_add(&device
->dev_alloc_list
,
1793 &root
->fs_info
->fs_devices
->alloc_list
);
1794 device
->fs_devices
->rw_devices
++;
1795 unlock_chunks(root
);
1800 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info
*fs_info
,
1801 struct btrfs_device
*srcdev
)
1803 struct btrfs_fs_devices
*fs_devices
;
1805 WARN_ON(!mutex_is_locked(&fs_info
->fs_devices
->device_list_mutex
));
1808 * in case of fs with no seed, srcdev->fs_devices will point
1809 * to fs_devices of fs_info. However when the dev being replaced is
1810 * a seed dev it will point to the seed's local fs_devices. In short
1811 * srcdev will have its correct fs_devices in both the cases.
1813 fs_devices
= srcdev
->fs_devices
;
1815 list_del_rcu(&srcdev
->dev_list
);
1816 list_del_rcu(&srcdev
->dev_alloc_list
);
1817 fs_devices
->num_devices
--;
1818 if (srcdev
->missing
)
1819 fs_devices
->missing_devices
--;
1821 if (srcdev
->writeable
) {
1822 fs_devices
->rw_devices
--;
1823 /* zero out the old super if it is writable */
1824 btrfs_scratch_superblock(srcdev
);
1828 fs_devices
->open_devices
--;
1831 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info
*fs_info
,
1832 struct btrfs_device
*srcdev
)
1834 struct btrfs_fs_devices
*fs_devices
= srcdev
->fs_devices
;
1836 call_rcu(&srcdev
->rcu
, free_device
);
1839 * unless fs_devices is seed fs, num_devices shouldn't go
1842 BUG_ON(!fs_devices
->num_devices
&& !fs_devices
->seeding
);
1844 /* if this is no devs we rather delete the fs_devices */
1845 if (!fs_devices
->num_devices
) {
1846 struct btrfs_fs_devices
*tmp_fs_devices
;
1848 tmp_fs_devices
= fs_info
->fs_devices
;
1849 while (tmp_fs_devices
) {
1850 if (tmp_fs_devices
->seed
== fs_devices
) {
1851 tmp_fs_devices
->seed
= fs_devices
->seed
;
1854 tmp_fs_devices
= tmp_fs_devices
->seed
;
1856 fs_devices
->seed
= NULL
;
1857 __btrfs_close_devices(fs_devices
);
1858 free_fs_devices(fs_devices
);
1862 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
1863 struct btrfs_device
*tgtdev
)
1865 struct btrfs_device
*next_device
;
1867 mutex_lock(&uuid_mutex
);
1869 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
1871 btrfs_scratch_superblock(tgtdev
);
1872 fs_info
->fs_devices
->open_devices
--;
1874 fs_info
->fs_devices
->num_devices
--;
1876 next_device
= list_entry(fs_info
->fs_devices
->devices
.next
,
1877 struct btrfs_device
, dev_list
);
1878 if (tgtdev
->bdev
== fs_info
->sb
->s_bdev
)
1879 fs_info
->sb
->s_bdev
= next_device
->bdev
;
1880 if (tgtdev
->bdev
== fs_info
->fs_devices
->latest_bdev
)
1881 fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1882 list_del_rcu(&tgtdev
->dev_list
);
1884 call_rcu(&tgtdev
->rcu
, free_device
);
1886 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
1887 mutex_unlock(&uuid_mutex
);
1890 static int btrfs_find_device_by_path(struct btrfs_root
*root
, char *device_path
,
1891 struct btrfs_device
**device
)
1894 struct btrfs_super_block
*disk_super
;
1897 struct block_device
*bdev
;
1898 struct buffer_head
*bh
;
1901 ret
= btrfs_get_bdev_and_sb(device_path
, FMODE_READ
,
1902 root
->fs_info
->bdev_holder
, 0, &bdev
, &bh
);
1905 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1906 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1907 dev_uuid
= disk_super
->dev_item
.uuid
;
1908 *device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1913 blkdev_put(bdev
, FMODE_READ
);
1917 int btrfs_find_device_missing_or_by_path(struct btrfs_root
*root
,
1919 struct btrfs_device
**device
)
1922 if (strcmp(device_path
, "missing") == 0) {
1923 struct list_head
*devices
;
1924 struct btrfs_device
*tmp
;
1926 devices
= &root
->fs_info
->fs_devices
->devices
;
1928 * It is safe to read the devices since the volume_mutex
1929 * is held by the caller.
1931 list_for_each_entry(tmp
, devices
, dev_list
) {
1932 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
1939 btrfs_err(root
->fs_info
, "no missing device found");
1945 return btrfs_find_device_by_path(root
, device_path
, device
);
1950 * does all the dirty work required for changing file system's UUID.
1952 static int btrfs_prepare_sprout(struct btrfs_root
*root
)
1954 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
1955 struct btrfs_fs_devices
*old_devices
;
1956 struct btrfs_fs_devices
*seed_devices
;
1957 struct btrfs_super_block
*disk_super
= root
->fs_info
->super_copy
;
1958 struct btrfs_device
*device
;
1961 BUG_ON(!mutex_is_locked(&uuid_mutex
));
1962 if (!fs_devices
->seeding
)
1965 seed_devices
= __alloc_fs_devices();
1966 if (IS_ERR(seed_devices
))
1967 return PTR_ERR(seed_devices
);
1969 old_devices
= clone_fs_devices(fs_devices
);
1970 if (IS_ERR(old_devices
)) {
1971 kfree(seed_devices
);
1972 return PTR_ERR(old_devices
);
1975 list_add(&old_devices
->list
, &fs_uuids
);
1977 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
1978 seed_devices
->opened
= 1;
1979 INIT_LIST_HEAD(&seed_devices
->devices
);
1980 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
1981 mutex_init(&seed_devices
->device_list_mutex
);
1983 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1984 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
1986 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
)
1987 device
->fs_devices
= seed_devices
;
1990 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
1991 unlock_chunks(root
);
1993 fs_devices
->seeding
= 0;
1994 fs_devices
->num_devices
= 0;
1995 fs_devices
->open_devices
= 0;
1996 fs_devices
->missing_devices
= 0;
1997 fs_devices
->rotating
= 0;
1998 fs_devices
->seed
= seed_devices
;
2000 generate_random_uuid(fs_devices
->fsid
);
2001 memcpy(root
->fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2002 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2003 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2005 super_flags
= btrfs_super_flags(disk_super
) &
2006 ~BTRFS_SUPER_FLAG_SEEDING
;
2007 btrfs_set_super_flags(disk_super
, super_flags
);
2013 * strore the expected generation for seed devices in device items.
2015 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
2016 struct btrfs_root
*root
)
2018 struct btrfs_path
*path
;
2019 struct extent_buffer
*leaf
;
2020 struct btrfs_dev_item
*dev_item
;
2021 struct btrfs_device
*device
;
2022 struct btrfs_key key
;
2023 u8 fs_uuid
[BTRFS_UUID_SIZE
];
2024 u8 dev_uuid
[BTRFS_UUID_SIZE
];
2028 path
= btrfs_alloc_path();
2032 root
= root
->fs_info
->chunk_root
;
2033 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2035 key
.type
= BTRFS_DEV_ITEM_KEY
;
2038 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2042 leaf
= path
->nodes
[0];
2044 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
2045 ret
= btrfs_next_leaf(root
, path
);
2050 leaf
= path
->nodes
[0];
2051 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2052 btrfs_release_path(path
);
2056 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2057 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
2058 key
.type
!= BTRFS_DEV_ITEM_KEY
)
2061 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2062 struct btrfs_dev_item
);
2063 devid
= btrfs_device_id(leaf
, dev_item
);
2064 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
2066 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
2068 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
2070 BUG_ON(!device
); /* Logic error */
2072 if (device
->fs_devices
->seeding
) {
2073 btrfs_set_device_generation(leaf
, dev_item
,
2074 device
->generation
);
2075 btrfs_mark_buffer_dirty(leaf
);
2083 btrfs_free_path(path
);
2087 int btrfs_init_new_device(struct btrfs_root
*root
, char *device_path
)
2089 struct request_queue
*q
;
2090 struct btrfs_trans_handle
*trans
;
2091 struct btrfs_device
*device
;
2092 struct block_device
*bdev
;
2093 struct list_head
*devices
;
2094 struct super_block
*sb
= root
->fs_info
->sb
;
2095 struct rcu_string
*name
;
2097 int seeding_dev
= 0;
2100 if ((sb
->s_flags
& MS_RDONLY
) && !root
->fs_info
->fs_devices
->seeding
)
2103 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2104 root
->fs_info
->bdev_holder
);
2106 return PTR_ERR(bdev
);
2108 if (root
->fs_info
->fs_devices
->seeding
) {
2110 down_write(&sb
->s_umount
);
2111 mutex_lock(&uuid_mutex
);
2114 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2116 devices
= &root
->fs_info
->fs_devices
->devices
;
2118 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2119 list_for_each_entry(device
, devices
, dev_list
) {
2120 if (device
->bdev
== bdev
) {
2123 &root
->fs_info
->fs_devices
->device_list_mutex
);
2127 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2129 device
= btrfs_alloc_device(root
->fs_info
, NULL
, NULL
);
2130 if (IS_ERR(device
)) {
2131 /* we can safely leave the fs_devices entry around */
2132 ret
= PTR_ERR(device
);
2136 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2142 rcu_assign_pointer(device
->name
, name
);
2144 trans
= btrfs_start_transaction(root
, 0);
2145 if (IS_ERR(trans
)) {
2146 rcu_string_free(device
->name
);
2148 ret
= PTR_ERR(trans
);
2152 q
= bdev_get_queue(bdev
);
2153 if (blk_queue_discard(q
))
2154 device
->can_discard
= 1;
2155 device
->writeable
= 1;
2156 device
->generation
= trans
->transid
;
2157 device
->io_width
= root
->sectorsize
;
2158 device
->io_align
= root
->sectorsize
;
2159 device
->sector_size
= root
->sectorsize
;
2160 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
2161 device
->disk_total_bytes
= device
->total_bytes
;
2162 device
->commit_total_bytes
= device
->total_bytes
;
2163 device
->dev_root
= root
->fs_info
->dev_root
;
2164 device
->bdev
= bdev
;
2165 device
->in_fs_metadata
= 1;
2166 device
->is_tgtdev_for_dev_replace
= 0;
2167 device
->mode
= FMODE_EXCL
;
2168 device
->dev_stats_valid
= 1;
2169 set_blocksize(device
->bdev
, 4096);
2172 sb
->s_flags
&= ~MS_RDONLY
;
2173 ret
= btrfs_prepare_sprout(root
);
2174 BUG_ON(ret
); /* -ENOMEM */
2177 device
->fs_devices
= root
->fs_info
->fs_devices
;
2179 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2181 list_add_rcu(&device
->dev_list
, &root
->fs_info
->fs_devices
->devices
);
2182 list_add(&device
->dev_alloc_list
,
2183 &root
->fs_info
->fs_devices
->alloc_list
);
2184 root
->fs_info
->fs_devices
->num_devices
++;
2185 root
->fs_info
->fs_devices
->open_devices
++;
2186 root
->fs_info
->fs_devices
->rw_devices
++;
2187 root
->fs_info
->fs_devices
->total_devices
++;
2188 root
->fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
2190 spin_lock(&root
->fs_info
->free_chunk_lock
);
2191 root
->fs_info
->free_chunk_space
+= device
->total_bytes
;
2192 spin_unlock(&root
->fs_info
->free_chunk_lock
);
2194 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
2195 root
->fs_info
->fs_devices
->rotating
= 1;
2197 tmp
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
2198 btrfs_set_super_total_bytes(root
->fs_info
->super_copy
,
2199 tmp
+ device
->total_bytes
);
2201 tmp
= btrfs_super_num_devices(root
->fs_info
->super_copy
);
2202 btrfs_set_super_num_devices(root
->fs_info
->super_copy
,
2205 /* add sysfs device entry */
2206 btrfs_kobj_add_device(root
->fs_info
, device
);
2209 * we've got more storage, clear any full flags on the space
2212 btrfs_clear_space_info_full(root
->fs_info
);
2214 unlock_chunks(root
);
2215 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2219 ret
= init_first_rw_device(trans
, root
, device
);
2220 unlock_chunks(root
);
2222 btrfs_abort_transaction(trans
, root
, ret
);
2227 ret
= btrfs_add_device(trans
, root
, device
);
2229 btrfs_abort_transaction(trans
, root
, ret
);
2234 char fsid_buf
[BTRFS_UUID_UNPARSED_SIZE
];
2236 ret
= btrfs_finish_sprout(trans
, root
);
2238 btrfs_abort_transaction(trans
, root
, ret
);
2242 /* Sprouting would change fsid of the mounted root,
2243 * so rename the fsid on the sysfs
2245 snprintf(fsid_buf
, BTRFS_UUID_UNPARSED_SIZE
, "%pU",
2246 root
->fs_info
->fsid
);
2247 if (kobject_rename(&root
->fs_info
->super_kobj
, fsid_buf
))
2251 root
->fs_info
->num_tolerated_disk_barrier_failures
=
2252 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
2253 ret
= btrfs_commit_transaction(trans
, root
);
2256 mutex_unlock(&uuid_mutex
);
2257 up_write(&sb
->s_umount
);
2259 if (ret
) /* transaction commit */
2262 ret
= btrfs_relocate_sys_chunks(root
);
2264 btrfs_error(root
->fs_info
, ret
,
2265 "Failed to relocate sys chunks after "
2266 "device initialization. This can be fixed "
2267 "using the \"btrfs balance\" command.");
2268 trans
= btrfs_attach_transaction(root
);
2269 if (IS_ERR(trans
)) {
2270 if (PTR_ERR(trans
) == -ENOENT
)
2272 return PTR_ERR(trans
);
2274 ret
= btrfs_commit_transaction(trans
, root
);
2277 /* Update ctime/mtime for libblkid */
2278 update_dev_time(device_path
);
2282 btrfs_end_transaction(trans
, root
);
2283 rcu_string_free(device
->name
);
2284 btrfs_kobj_rm_device(root
->fs_info
, device
);
2287 blkdev_put(bdev
, FMODE_EXCL
);
2289 mutex_unlock(&uuid_mutex
);
2290 up_write(&sb
->s_umount
);
2295 int btrfs_init_dev_replace_tgtdev(struct btrfs_root
*root
, char *device_path
,
2296 struct btrfs_device
*srcdev
,
2297 struct btrfs_device
**device_out
)
2299 struct request_queue
*q
;
2300 struct btrfs_device
*device
;
2301 struct block_device
*bdev
;
2302 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2303 struct list_head
*devices
;
2304 struct rcu_string
*name
;
2305 u64 devid
= BTRFS_DEV_REPLACE_DEVID
;
2309 if (fs_info
->fs_devices
->seeding
) {
2310 btrfs_err(fs_info
, "the filesystem is a seed filesystem!");
2314 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2315 fs_info
->bdev_holder
);
2317 btrfs_err(fs_info
, "target device %s is invalid!", device_path
);
2318 return PTR_ERR(bdev
);
2321 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2323 devices
= &fs_info
->fs_devices
->devices
;
2324 list_for_each_entry(device
, devices
, dev_list
) {
2325 if (device
->bdev
== bdev
) {
2326 btrfs_err(fs_info
, "target device is in the filesystem!");
2333 if (i_size_read(bdev
->bd_inode
) <
2334 btrfs_device_get_total_bytes(srcdev
)) {
2335 btrfs_err(fs_info
, "target device is smaller than source device!");
2341 device
= btrfs_alloc_device(NULL
, &devid
, NULL
);
2342 if (IS_ERR(device
)) {
2343 ret
= PTR_ERR(device
);
2347 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2353 rcu_assign_pointer(device
->name
, name
);
2355 q
= bdev_get_queue(bdev
);
2356 if (blk_queue_discard(q
))
2357 device
->can_discard
= 1;
2358 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2359 device
->writeable
= 1;
2360 device
->generation
= 0;
2361 device
->io_width
= root
->sectorsize
;
2362 device
->io_align
= root
->sectorsize
;
2363 device
->sector_size
= root
->sectorsize
;
2364 device
->total_bytes
= btrfs_device_get_total_bytes(srcdev
);
2365 device
->disk_total_bytes
= btrfs_device_get_disk_total_bytes(srcdev
);
2366 device
->bytes_used
= btrfs_device_get_bytes_used(srcdev
);
2367 ASSERT(list_empty(&srcdev
->resized_list
));
2368 device
->commit_total_bytes
= srcdev
->commit_total_bytes
;
2369 device
->commit_bytes_used
= device
->bytes_used
;
2370 device
->dev_root
= fs_info
->dev_root
;
2371 device
->bdev
= bdev
;
2372 device
->in_fs_metadata
= 1;
2373 device
->is_tgtdev_for_dev_replace
= 1;
2374 device
->mode
= FMODE_EXCL
;
2375 device
->dev_stats_valid
= 1;
2376 set_blocksize(device
->bdev
, 4096);
2377 device
->fs_devices
= fs_info
->fs_devices
;
2378 list_add(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2379 fs_info
->fs_devices
->num_devices
++;
2380 fs_info
->fs_devices
->open_devices
++;
2381 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2383 *device_out
= device
;
2387 blkdev_put(bdev
, FMODE_EXCL
);
2391 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info
*fs_info
,
2392 struct btrfs_device
*tgtdev
)
2394 WARN_ON(fs_info
->fs_devices
->rw_devices
== 0);
2395 tgtdev
->io_width
= fs_info
->dev_root
->sectorsize
;
2396 tgtdev
->io_align
= fs_info
->dev_root
->sectorsize
;
2397 tgtdev
->sector_size
= fs_info
->dev_root
->sectorsize
;
2398 tgtdev
->dev_root
= fs_info
->dev_root
;
2399 tgtdev
->in_fs_metadata
= 1;
2402 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2403 struct btrfs_device
*device
)
2406 struct btrfs_path
*path
;
2407 struct btrfs_root
*root
;
2408 struct btrfs_dev_item
*dev_item
;
2409 struct extent_buffer
*leaf
;
2410 struct btrfs_key key
;
2412 root
= device
->dev_root
->fs_info
->chunk_root
;
2414 path
= btrfs_alloc_path();
2418 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2419 key
.type
= BTRFS_DEV_ITEM_KEY
;
2420 key
.offset
= device
->devid
;
2422 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2431 leaf
= path
->nodes
[0];
2432 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2434 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2435 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2436 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2437 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2438 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2439 btrfs_set_device_total_bytes(leaf
, dev_item
,
2440 btrfs_device_get_disk_total_bytes(device
));
2441 btrfs_set_device_bytes_used(leaf
, dev_item
,
2442 btrfs_device_get_bytes_used(device
));
2443 btrfs_mark_buffer_dirty(leaf
);
2446 btrfs_free_path(path
);
2450 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2451 struct btrfs_device
*device
, u64 new_size
)
2453 struct btrfs_super_block
*super_copy
=
2454 device
->dev_root
->fs_info
->super_copy
;
2455 struct btrfs_fs_devices
*fs_devices
;
2459 if (!device
->writeable
)
2462 lock_chunks(device
->dev_root
);
2463 old_total
= btrfs_super_total_bytes(super_copy
);
2464 diff
= new_size
- device
->total_bytes
;
2466 if (new_size
<= device
->total_bytes
||
2467 device
->is_tgtdev_for_dev_replace
) {
2468 unlock_chunks(device
->dev_root
);
2472 fs_devices
= device
->dev_root
->fs_info
->fs_devices
;
2474 btrfs_set_super_total_bytes(super_copy
, old_total
+ diff
);
2475 device
->fs_devices
->total_rw_bytes
+= diff
;
2477 btrfs_device_set_total_bytes(device
, new_size
);
2478 btrfs_device_set_disk_total_bytes(device
, new_size
);
2479 btrfs_clear_space_info_full(device
->dev_root
->fs_info
);
2480 if (list_empty(&device
->resized_list
))
2481 list_add_tail(&device
->resized_list
,
2482 &fs_devices
->resized_devices
);
2483 unlock_chunks(device
->dev_root
);
2485 return btrfs_update_device(trans
, device
);
2488 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
2489 struct btrfs_root
*root
, u64 chunk_objectid
,
2493 struct btrfs_path
*path
;
2494 struct btrfs_key key
;
2496 root
= root
->fs_info
->chunk_root
;
2497 path
= btrfs_alloc_path();
2501 key
.objectid
= chunk_objectid
;
2502 key
.offset
= chunk_offset
;
2503 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2505 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2508 else if (ret
> 0) { /* Logic error or corruption */
2509 btrfs_error(root
->fs_info
, -ENOENT
,
2510 "Failed lookup while freeing chunk.");
2515 ret
= btrfs_del_item(trans
, root
, path
);
2517 btrfs_error(root
->fs_info
, ret
,
2518 "Failed to delete chunk item.");
2520 btrfs_free_path(path
);
2524 static int btrfs_del_sys_chunk(struct btrfs_root
*root
, u64 chunk_objectid
, u64
2527 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
2528 struct btrfs_disk_key
*disk_key
;
2529 struct btrfs_chunk
*chunk
;
2536 struct btrfs_key key
;
2539 array_size
= btrfs_super_sys_array_size(super_copy
);
2541 ptr
= super_copy
->sys_chunk_array
;
2544 while (cur
< array_size
) {
2545 disk_key
= (struct btrfs_disk_key
*)ptr
;
2546 btrfs_disk_key_to_cpu(&key
, disk_key
);
2548 len
= sizeof(*disk_key
);
2550 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2551 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2552 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2553 len
+= btrfs_chunk_item_size(num_stripes
);
2558 if (key
.objectid
== chunk_objectid
&&
2559 key
.offset
== chunk_offset
) {
2560 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2562 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2568 unlock_chunks(root
);
2572 int btrfs_remove_chunk(struct btrfs_trans_handle
*trans
,
2573 struct btrfs_root
*root
, u64 chunk_offset
)
2575 struct extent_map_tree
*em_tree
;
2576 struct extent_map
*em
;
2577 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2578 struct map_lookup
*map
;
2579 u64 dev_extent_len
= 0;
2580 u64 chunk_objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2584 root
= root
->fs_info
->chunk_root
;
2585 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
2587 read_lock(&em_tree
->lock
);
2588 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
2589 read_unlock(&em_tree
->lock
);
2591 if (!em
|| em
->start
> chunk_offset
||
2592 em
->start
+ em
->len
< chunk_offset
) {
2594 * This is a logic error, but we don't want to just rely on the
2595 * user having built with ASSERT enabled, so if ASSERT doens't
2596 * do anything we still error out.
2600 free_extent_map(em
);
2603 map
= (struct map_lookup
*)em
->bdev
;
2605 for (i
= 0; i
< map
->num_stripes
; i
++) {
2606 struct btrfs_device
*device
= map
->stripes
[i
].dev
;
2607 ret
= btrfs_free_dev_extent(trans
, device
,
2608 map
->stripes
[i
].physical
,
2611 btrfs_abort_transaction(trans
, root
, ret
);
2615 if (device
->bytes_used
> 0) {
2617 btrfs_device_set_bytes_used(device
,
2618 device
->bytes_used
- dev_extent_len
);
2619 spin_lock(&root
->fs_info
->free_chunk_lock
);
2620 root
->fs_info
->free_chunk_space
+= dev_extent_len
;
2621 spin_unlock(&root
->fs_info
->free_chunk_lock
);
2622 btrfs_clear_space_info_full(root
->fs_info
);
2623 unlock_chunks(root
);
2626 if (map
->stripes
[i
].dev
) {
2627 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2629 btrfs_abort_transaction(trans
, root
, ret
);
2634 ret
= btrfs_free_chunk(trans
, root
, chunk_objectid
, chunk_offset
);
2636 btrfs_abort_transaction(trans
, root
, ret
);
2640 trace_btrfs_chunk_free(root
, map
, chunk_offset
, em
->len
);
2642 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2643 ret
= btrfs_del_sys_chunk(root
, chunk_objectid
, chunk_offset
);
2645 btrfs_abort_transaction(trans
, root
, ret
);
2650 ret
= btrfs_remove_block_group(trans
, extent_root
, chunk_offset
, em
);
2652 btrfs_abort_transaction(trans
, extent_root
, ret
);
2658 free_extent_map(em
);
2662 static int btrfs_relocate_chunk(struct btrfs_root
*root
,
2666 struct btrfs_root
*extent_root
;
2667 struct btrfs_trans_handle
*trans
;
2670 root
= root
->fs_info
->chunk_root
;
2671 extent_root
= root
->fs_info
->extent_root
;
2673 ret
= btrfs_can_relocate(extent_root
, chunk_offset
);
2677 /* step one, relocate all the extents inside this chunk */
2678 ret
= btrfs_relocate_block_group(extent_root
, chunk_offset
);
2682 trans
= btrfs_start_transaction(root
, 0);
2683 if (IS_ERR(trans
)) {
2684 ret
= PTR_ERR(trans
);
2685 btrfs_std_error(root
->fs_info
, ret
);
2690 * step two, delete the device extents and the
2691 * chunk tree entries
2693 ret
= btrfs_remove_chunk(trans
, root
, chunk_offset
);
2694 btrfs_end_transaction(trans
, root
);
2698 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
)
2700 struct btrfs_root
*chunk_root
= root
->fs_info
->chunk_root
;
2701 struct btrfs_path
*path
;
2702 struct extent_buffer
*leaf
;
2703 struct btrfs_chunk
*chunk
;
2704 struct btrfs_key key
;
2705 struct btrfs_key found_key
;
2707 bool retried
= false;
2711 path
= btrfs_alloc_path();
2716 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2717 key
.offset
= (u64
)-1;
2718 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2721 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2724 BUG_ON(ret
== 0); /* Corruption */
2726 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
2733 leaf
= path
->nodes
[0];
2734 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2736 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
2737 struct btrfs_chunk
);
2738 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2739 btrfs_release_path(path
);
2741 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2742 ret
= btrfs_relocate_chunk(chunk_root
,
2751 if (found_key
.offset
== 0)
2753 key
.offset
= found_key
.offset
- 1;
2756 if (failed
&& !retried
) {
2760 } else if (WARN_ON(failed
&& retried
)) {
2764 btrfs_free_path(path
);
2768 static int insert_balance_item(struct btrfs_root
*root
,
2769 struct btrfs_balance_control
*bctl
)
2771 struct btrfs_trans_handle
*trans
;
2772 struct btrfs_balance_item
*item
;
2773 struct btrfs_disk_balance_args disk_bargs
;
2774 struct btrfs_path
*path
;
2775 struct extent_buffer
*leaf
;
2776 struct btrfs_key key
;
2779 path
= btrfs_alloc_path();
2783 trans
= btrfs_start_transaction(root
, 0);
2784 if (IS_ERR(trans
)) {
2785 btrfs_free_path(path
);
2786 return PTR_ERR(trans
);
2789 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2790 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2793 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2798 leaf
= path
->nodes
[0];
2799 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
2801 memset_extent_buffer(leaf
, 0, (unsigned long)item
, sizeof(*item
));
2803 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
2804 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
2805 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
2806 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
2807 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
2808 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
2810 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
2812 btrfs_mark_buffer_dirty(leaf
);
2814 btrfs_free_path(path
);
2815 err
= btrfs_commit_transaction(trans
, root
);
2821 static int del_balance_item(struct btrfs_root
*root
)
2823 struct btrfs_trans_handle
*trans
;
2824 struct btrfs_path
*path
;
2825 struct btrfs_key key
;
2828 path
= btrfs_alloc_path();
2832 trans
= btrfs_start_transaction(root
, 0);
2833 if (IS_ERR(trans
)) {
2834 btrfs_free_path(path
);
2835 return PTR_ERR(trans
);
2838 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2839 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2842 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2850 ret
= btrfs_del_item(trans
, root
, path
);
2852 btrfs_free_path(path
);
2853 err
= btrfs_commit_transaction(trans
, root
);
2860 * This is a heuristic used to reduce the number of chunks balanced on
2861 * resume after balance was interrupted.
2863 static void update_balance_args(struct btrfs_balance_control
*bctl
)
2866 * Turn on soft mode for chunk types that were being converted.
2868 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2869 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2870 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2871 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2872 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2873 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2876 * Turn on usage filter if is not already used. The idea is
2877 * that chunks that we have already balanced should be
2878 * reasonably full. Don't do it for chunks that are being
2879 * converted - that will keep us from relocating unconverted
2880 * (albeit full) chunks.
2882 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2883 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2884 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2885 bctl
->data
.usage
= 90;
2887 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2888 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2889 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2890 bctl
->sys
.usage
= 90;
2892 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2893 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2894 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2895 bctl
->meta
.usage
= 90;
2900 * Should be called with both balance and volume mutexes held to
2901 * serialize other volume operations (add_dev/rm_dev/resize) with
2902 * restriper. Same goes for unset_balance_control.
2904 static void set_balance_control(struct btrfs_balance_control
*bctl
)
2906 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
2908 BUG_ON(fs_info
->balance_ctl
);
2910 spin_lock(&fs_info
->balance_lock
);
2911 fs_info
->balance_ctl
= bctl
;
2912 spin_unlock(&fs_info
->balance_lock
);
2915 static void unset_balance_control(struct btrfs_fs_info
*fs_info
)
2917 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
2919 BUG_ON(!fs_info
->balance_ctl
);
2921 spin_lock(&fs_info
->balance_lock
);
2922 fs_info
->balance_ctl
= NULL
;
2923 spin_unlock(&fs_info
->balance_lock
);
2929 * Balance filters. Return 1 if chunk should be filtered out
2930 * (should not be balanced).
2932 static int chunk_profiles_filter(u64 chunk_type
,
2933 struct btrfs_balance_args
*bargs
)
2935 chunk_type
= chunk_to_extended(chunk_type
) &
2936 BTRFS_EXTENDED_PROFILE_MASK
;
2938 if (bargs
->profiles
& chunk_type
)
2944 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
2945 struct btrfs_balance_args
*bargs
)
2947 struct btrfs_block_group_cache
*cache
;
2948 u64 chunk_used
, user_thresh
;
2951 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
2952 chunk_used
= btrfs_block_group_used(&cache
->item
);
2954 if (bargs
->usage
== 0)
2956 else if (bargs
->usage
> 100)
2957 user_thresh
= cache
->key
.offset
;
2959 user_thresh
= div_factor_fine(cache
->key
.offset
,
2962 if (chunk_used
< user_thresh
)
2965 btrfs_put_block_group(cache
);
2969 static int chunk_devid_filter(struct extent_buffer
*leaf
,
2970 struct btrfs_chunk
*chunk
,
2971 struct btrfs_balance_args
*bargs
)
2973 struct btrfs_stripe
*stripe
;
2974 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2977 for (i
= 0; i
< num_stripes
; i
++) {
2978 stripe
= btrfs_stripe_nr(chunk
, i
);
2979 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
2986 /* [pstart, pend) */
2987 static int chunk_drange_filter(struct extent_buffer
*leaf
,
2988 struct btrfs_chunk
*chunk
,
2990 struct btrfs_balance_args
*bargs
)
2992 struct btrfs_stripe
*stripe
;
2993 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2999 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
3002 if (btrfs_chunk_type(leaf
, chunk
) & (BTRFS_BLOCK_GROUP_DUP
|
3003 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)) {
3004 factor
= num_stripes
/ 2;
3005 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID5
) {
3006 factor
= num_stripes
- 1;
3007 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID6
) {
3008 factor
= num_stripes
- 2;
3010 factor
= num_stripes
;
3013 for (i
= 0; i
< num_stripes
; i
++) {
3014 stripe
= btrfs_stripe_nr(chunk
, i
);
3015 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
3018 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
3019 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
3020 do_div(stripe_length
, factor
);
3022 if (stripe_offset
< bargs
->pend
&&
3023 stripe_offset
+ stripe_length
> bargs
->pstart
)
3030 /* [vstart, vend) */
3031 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
3032 struct btrfs_chunk
*chunk
,
3034 struct btrfs_balance_args
*bargs
)
3036 if (chunk_offset
< bargs
->vend
&&
3037 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
3038 /* at least part of the chunk is inside this vrange */
3044 static int chunk_soft_convert_filter(u64 chunk_type
,
3045 struct btrfs_balance_args
*bargs
)
3047 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
3050 chunk_type
= chunk_to_extended(chunk_type
) &
3051 BTRFS_EXTENDED_PROFILE_MASK
;
3053 if (bargs
->target
== chunk_type
)
3059 static int should_balance_chunk(struct btrfs_root
*root
,
3060 struct extent_buffer
*leaf
,
3061 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
3063 struct btrfs_balance_control
*bctl
= root
->fs_info
->balance_ctl
;
3064 struct btrfs_balance_args
*bargs
= NULL
;
3065 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3068 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
3069 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
3073 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3074 bargs
= &bctl
->data
;
3075 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3077 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3078 bargs
= &bctl
->meta
;
3080 /* profiles filter */
3081 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
3082 chunk_profiles_filter(chunk_type
, bargs
)) {
3087 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3088 chunk_usage_filter(bctl
->fs_info
, chunk_offset
, bargs
)) {
3093 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
3094 chunk_devid_filter(leaf
, chunk
, bargs
)) {
3098 /* drange filter, makes sense only with devid filter */
3099 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
3100 chunk_drange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3105 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
3106 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3110 /* soft profile changing mode */
3111 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
3112 chunk_soft_convert_filter(chunk_type
, bargs
)) {
3117 * limited by count, must be the last filter
3119 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT
)) {
3120 if (bargs
->limit
== 0)
3129 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
3131 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3132 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
3133 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
3134 struct list_head
*devices
;
3135 struct btrfs_device
*device
;
3138 struct btrfs_chunk
*chunk
;
3139 struct btrfs_path
*path
;
3140 struct btrfs_key key
;
3141 struct btrfs_key found_key
;
3142 struct btrfs_trans_handle
*trans
;
3143 struct extent_buffer
*leaf
;
3146 int enospc_errors
= 0;
3147 bool counting
= true;
3148 u64 limit_data
= bctl
->data
.limit
;
3149 u64 limit_meta
= bctl
->meta
.limit
;
3150 u64 limit_sys
= bctl
->sys
.limit
;
3152 /* step one make some room on all the devices */
3153 devices
= &fs_info
->fs_devices
->devices
;
3154 list_for_each_entry(device
, devices
, dev_list
) {
3155 old_size
= btrfs_device_get_total_bytes(device
);
3156 size_to_free
= div_factor(old_size
, 1);
3157 size_to_free
= min(size_to_free
, (u64
)1 * 1024 * 1024);
3158 if (!device
->writeable
||
3159 btrfs_device_get_total_bytes(device
) -
3160 btrfs_device_get_bytes_used(device
) > size_to_free
||
3161 device
->is_tgtdev_for_dev_replace
)
3164 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
3169 trans
= btrfs_start_transaction(dev_root
, 0);
3170 BUG_ON(IS_ERR(trans
));
3172 ret
= btrfs_grow_device(trans
, device
, old_size
);
3175 btrfs_end_transaction(trans
, dev_root
);
3178 /* step two, relocate all the chunks */
3179 path
= btrfs_alloc_path();
3185 /* zero out stat counters */
3186 spin_lock(&fs_info
->balance_lock
);
3187 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
3188 spin_unlock(&fs_info
->balance_lock
);
3191 bctl
->data
.limit
= limit_data
;
3192 bctl
->meta
.limit
= limit_meta
;
3193 bctl
->sys
.limit
= limit_sys
;
3195 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3196 key
.offset
= (u64
)-1;
3197 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3200 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
3201 atomic_read(&fs_info
->balance_cancel_req
)) {
3206 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
3211 * this shouldn't happen, it means the last relocate
3215 BUG(); /* FIXME break ? */
3217 ret
= btrfs_previous_item(chunk_root
, path
, 0,
3218 BTRFS_CHUNK_ITEM_KEY
);
3224 leaf
= path
->nodes
[0];
3225 slot
= path
->slots
[0];
3226 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3228 if (found_key
.objectid
!= key
.objectid
)
3231 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
3234 spin_lock(&fs_info
->balance_lock
);
3235 bctl
->stat
.considered
++;
3236 spin_unlock(&fs_info
->balance_lock
);
3239 ret
= should_balance_chunk(chunk_root
, leaf
, chunk
,
3241 btrfs_release_path(path
);
3246 spin_lock(&fs_info
->balance_lock
);
3247 bctl
->stat
.expected
++;
3248 spin_unlock(&fs_info
->balance_lock
);
3252 ret
= btrfs_relocate_chunk(chunk_root
,
3255 if (ret
&& ret
!= -ENOSPC
)
3257 if (ret
== -ENOSPC
) {
3260 spin_lock(&fs_info
->balance_lock
);
3261 bctl
->stat
.completed
++;
3262 spin_unlock(&fs_info
->balance_lock
);
3265 if (found_key
.offset
== 0)
3267 key
.offset
= found_key
.offset
- 1;
3271 btrfs_release_path(path
);
3276 btrfs_free_path(path
);
3277 if (enospc_errors
) {
3278 btrfs_info(fs_info
, "%d enospc errors during balance",
3288 * alloc_profile_is_valid - see if a given profile is valid and reduced
3289 * @flags: profile to validate
3290 * @extended: if true @flags is treated as an extended profile
3292 static int alloc_profile_is_valid(u64 flags
, int extended
)
3294 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
3295 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
3297 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
3299 /* 1) check that all other bits are zeroed */
3303 /* 2) see if profile is reduced */
3305 return !extended
; /* "0" is valid for usual profiles */
3307 /* true if exactly one bit set */
3308 return (flags
& (flags
- 1)) == 0;
3311 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
3313 /* cancel requested || normal exit path */
3314 return atomic_read(&fs_info
->balance_cancel_req
) ||
3315 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
3316 atomic_read(&fs_info
->balance_cancel_req
) == 0);
3319 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
3323 unset_balance_control(fs_info
);
3324 ret
= del_balance_item(fs_info
->tree_root
);
3326 btrfs_std_error(fs_info
, ret
);
3328 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3332 * Should be called with both balance and volume mutexes held
3334 int btrfs_balance(struct btrfs_balance_control
*bctl
,
3335 struct btrfs_ioctl_balance_args
*bargs
)
3337 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3344 if (btrfs_fs_closing(fs_info
) ||
3345 atomic_read(&fs_info
->balance_pause_req
) ||
3346 atomic_read(&fs_info
->balance_cancel_req
)) {
3351 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
3352 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
3356 * In case of mixed groups both data and meta should be picked,
3357 * and identical options should be given for both of them.
3359 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
3360 if (mixed
&& (bctl
->flags
& allowed
)) {
3361 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
3362 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
3363 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
3364 btrfs_err(fs_info
, "with mixed groups data and "
3365 "metadata balance options must be the same");
3371 num_devices
= fs_info
->fs_devices
->num_devices
;
3372 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
3373 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
3374 BUG_ON(num_devices
< 1);
3377 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
3378 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
3379 if (num_devices
== 1)
3380 allowed
|= BTRFS_BLOCK_GROUP_DUP
;
3381 else if (num_devices
> 1)
3382 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
3383 if (num_devices
> 2)
3384 allowed
|= BTRFS_BLOCK_GROUP_RAID5
;
3385 if (num_devices
> 3)
3386 allowed
|= (BTRFS_BLOCK_GROUP_RAID10
|
3387 BTRFS_BLOCK_GROUP_RAID6
);
3388 if ((bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3389 (!alloc_profile_is_valid(bctl
->data
.target
, 1) ||
3390 (bctl
->data
.target
& ~allowed
))) {
3391 btrfs_err(fs_info
, "unable to start balance with target "
3392 "data profile %llu",
3397 if ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3398 (!alloc_profile_is_valid(bctl
->meta
.target
, 1) ||
3399 (bctl
->meta
.target
& ~allowed
))) {
3401 "unable to start balance with target metadata profile %llu",
3406 if ((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3407 (!alloc_profile_is_valid(bctl
->sys
.target
, 1) ||
3408 (bctl
->sys
.target
& ~allowed
))) {
3410 "unable to start balance with target system profile %llu",
3416 /* allow dup'ed data chunks only in mixed mode */
3417 if (!mixed
&& (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3418 (bctl
->data
.target
& BTRFS_BLOCK_GROUP_DUP
)) {
3419 btrfs_err(fs_info
, "dup for data is not allowed");
3424 /* allow to reduce meta or sys integrity only if force set */
3425 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3426 BTRFS_BLOCK_GROUP_RAID10
|
3427 BTRFS_BLOCK_GROUP_RAID5
|
3428 BTRFS_BLOCK_GROUP_RAID6
;
3430 seq
= read_seqbegin(&fs_info
->profiles_lock
);
3432 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3433 (fs_info
->avail_system_alloc_bits
& allowed
) &&
3434 !(bctl
->sys
.target
& allowed
)) ||
3435 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3436 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
3437 !(bctl
->meta
.target
& allowed
))) {
3438 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
3439 btrfs_info(fs_info
, "force reducing metadata integrity");
3441 btrfs_err(fs_info
, "balance will reduce metadata "
3442 "integrity, use force if you want this");
3447 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
3449 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3450 int num_tolerated_disk_barrier_failures
;
3451 u64 target
= bctl
->sys
.target
;
3453 num_tolerated_disk_barrier_failures
=
3454 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3455 if (num_tolerated_disk_barrier_failures
> 0 &&
3457 (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID0
|
3458 BTRFS_AVAIL_ALLOC_BIT_SINGLE
)))
3459 num_tolerated_disk_barrier_failures
= 0;
3460 else if (num_tolerated_disk_barrier_failures
> 1 &&
3462 (BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)))
3463 num_tolerated_disk_barrier_failures
= 1;
3465 fs_info
->num_tolerated_disk_barrier_failures
=
3466 num_tolerated_disk_barrier_failures
;
3469 ret
= insert_balance_item(fs_info
->tree_root
, bctl
);
3470 if (ret
&& ret
!= -EEXIST
)
3473 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
3474 BUG_ON(ret
== -EEXIST
);
3475 set_balance_control(bctl
);
3477 BUG_ON(ret
!= -EEXIST
);
3478 spin_lock(&fs_info
->balance_lock
);
3479 update_balance_args(bctl
);
3480 spin_unlock(&fs_info
->balance_lock
);
3483 atomic_inc(&fs_info
->balance_running
);
3484 mutex_unlock(&fs_info
->balance_mutex
);
3486 ret
= __btrfs_balance(fs_info
);
3488 mutex_lock(&fs_info
->balance_mutex
);
3489 atomic_dec(&fs_info
->balance_running
);
3491 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3492 fs_info
->num_tolerated_disk_barrier_failures
=
3493 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3497 memset(bargs
, 0, sizeof(*bargs
));
3498 update_ioctl_balance_args(fs_info
, 0, bargs
);
3501 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
3502 balance_need_close(fs_info
)) {
3503 __cancel_balance(fs_info
);
3506 wake_up(&fs_info
->balance_wait_q
);
3510 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
3511 __cancel_balance(fs_info
);
3514 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3519 static int balance_kthread(void *data
)
3521 struct btrfs_fs_info
*fs_info
= data
;
3524 mutex_lock(&fs_info
->volume_mutex
);
3525 mutex_lock(&fs_info
->balance_mutex
);
3527 if (fs_info
->balance_ctl
) {
3528 btrfs_info(fs_info
, "continuing balance");
3529 ret
= btrfs_balance(fs_info
->balance_ctl
, NULL
);
3532 mutex_unlock(&fs_info
->balance_mutex
);
3533 mutex_unlock(&fs_info
->volume_mutex
);
3538 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
3540 struct task_struct
*tsk
;
3542 spin_lock(&fs_info
->balance_lock
);
3543 if (!fs_info
->balance_ctl
) {
3544 spin_unlock(&fs_info
->balance_lock
);
3547 spin_unlock(&fs_info
->balance_lock
);
3549 if (btrfs_test_opt(fs_info
->tree_root
, SKIP_BALANCE
)) {
3550 btrfs_info(fs_info
, "force skipping balance");
3554 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
3555 return PTR_ERR_OR_ZERO(tsk
);
3558 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
3560 struct btrfs_balance_control
*bctl
;
3561 struct btrfs_balance_item
*item
;
3562 struct btrfs_disk_balance_args disk_bargs
;
3563 struct btrfs_path
*path
;
3564 struct extent_buffer
*leaf
;
3565 struct btrfs_key key
;
3568 path
= btrfs_alloc_path();
3572 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3573 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
3576 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
3579 if (ret
> 0) { /* ret = -ENOENT; */
3584 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
3590 leaf
= path
->nodes
[0];
3591 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3593 bctl
->fs_info
= fs_info
;
3594 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
3595 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
3597 btrfs_balance_data(leaf
, item
, &disk_bargs
);
3598 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
3599 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
3600 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
3601 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
3602 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
3604 WARN_ON(atomic_xchg(&fs_info
->mutually_exclusive_operation_running
, 1));
3606 mutex_lock(&fs_info
->volume_mutex
);
3607 mutex_lock(&fs_info
->balance_mutex
);
3609 set_balance_control(bctl
);
3611 mutex_unlock(&fs_info
->balance_mutex
);
3612 mutex_unlock(&fs_info
->volume_mutex
);
3614 btrfs_free_path(path
);
3618 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
3622 mutex_lock(&fs_info
->balance_mutex
);
3623 if (!fs_info
->balance_ctl
) {
3624 mutex_unlock(&fs_info
->balance_mutex
);
3628 if (atomic_read(&fs_info
->balance_running
)) {
3629 atomic_inc(&fs_info
->balance_pause_req
);
3630 mutex_unlock(&fs_info
->balance_mutex
);
3632 wait_event(fs_info
->balance_wait_q
,
3633 atomic_read(&fs_info
->balance_running
) == 0);
3635 mutex_lock(&fs_info
->balance_mutex
);
3636 /* we are good with balance_ctl ripped off from under us */
3637 BUG_ON(atomic_read(&fs_info
->balance_running
));
3638 atomic_dec(&fs_info
->balance_pause_req
);
3643 mutex_unlock(&fs_info
->balance_mutex
);
3647 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
3649 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
3652 mutex_lock(&fs_info
->balance_mutex
);
3653 if (!fs_info
->balance_ctl
) {
3654 mutex_unlock(&fs_info
->balance_mutex
);
3658 atomic_inc(&fs_info
->balance_cancel_req
);
3660 * if we are running just wait and return, balance item is
3661 * deleted in btrfs_balance in this case
3663 if (atomic_read(&fs_info
->balance_running
)) {
3664 mutex_unlock(&fs_info
->balance_mutex
);
3665 wait_event(fs_info
->balance_wait_q
,
3666 atomic_read(&fs_info
->balance_running
) == 0);
3667 mutex_lock(&fs_info
->balance_mutex
);
3669 /* __cancel_balance needs volume_mutex */
3670 mutex_unlock(&fs_info
->balance_mutex
);
3671 mutex_lock(&fs_info
->volume_mutex
);
3672 mutex_lock(&fs_info
->balance_mutex
);
3674 if (fs_info
->balance_ctl
)
3675 __cancel_balance(fs_info
);
3677 mutex_unlock(&fs_info
->volume_mutex
);
3680 BUG_ON(fs_info
->balance_ctl
|| atomic_read(&fs_info
->balance_running
));
3681 atomic_dec(&fs_info
->balance_cancel_req
);
3682 mutex_unlock(&fs_info
->balance_mutex
);
3686 static int btrfs_uuid_scan_kthread(void *data
)
3688 struct btrfs_fs_info
*fs_info
= data
;
3689 struct btrfs_root
*root
= fs_info
->tree_root
;
3690 struct btrfs_key key
;
3691 struct btrfs_key max_key
;
3692 struct btrfs_path
*path
= NULL
;
3694 struct extent_buffer
*eb
;
3696 struct btrfs_root_item root_item
;
3698 struct btrfs_trans_handle
*trans
= NULL
;
3700 path
= btrfs_alloc_path();
3707 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3710 max_key
.objectid
= (u64
)-1;
3711 max_key
.type
= BTRFS_ROOT_ITEM_KEY
;
3712 max_key
.offset
= (u64
)-1;
3715 ret
= btrfs_search_forward(root
, &key
, path
, 0);
3722 if (key
.type
!= BTRFS_ROOT_ITEM_KEY
||
3723 (key
.objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
3724 key
.objectid
!= BTRFS_FS_TREE_OBJECTID
) ||
3725 key
.objectid
> BTRFS_LAST_FREE_OBJECTID
)
3728 eb
= path
->nodes
[0];
3729 slot
= path
->slots
[0];
3730 item_size
= btrfs_item_size_nr(eb
, slot
);
3731 if (item_size
< sizeof(root_item
))
3734 read_extent_buffer(eb
, &root_item
,
3735 btrfs_item_ptr_offset(eb
, slot
),
3736 (int)sizeof(root_item
));
3737 if (btrfs_root_refs(&root_item
) == 0)
3740 if (!btrfs_is_empty_uuid(root_item
.uuid
) ||
3741 !btrfs_is_empty_uuid(root_item
.received_uuid
)) {
3745 btrfs_release_path(path
);
3747 * 1 - subvol uuid item
3748 * 1 - received_subvol uuid item
3750 trans
= btrfs_start_transaction(fs_info
->uuid_root
, 2);
3751 if (IS_ERR(trans
)) {
3752 ret
= PTR_ERR(trans
);
3760 if (!btrfs_is_empty_uuid(root_item
.uuid
)) {
3761 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
3763 BTRFS_UUID_KEY_SUBVOL
,
3766 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
3772 if (!btrfs_is_empty_uuid(root_item
.received_uuid
)) {
3773 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
3774 root_item
.received_uuid
,
3775 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
3778 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
3786 ret
= btrfs_end_transaction(trans
, fs_info
->uuid_root
);
3792 btrfs_release_path(path
);
3793 if (key
.offset
< (u64
)-1) {
3795 } else if (key
.type
< BTRFS_ROOT_ITEM_KEY
) {
3797 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3798 } else if (key
.objectid
< (u64
)-1) {
3800 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3809 btrfs_free_path(path
);
3810 if (trans
&& !IS_ERR(trans
))
3811 btrfs_end_transaction(trans
, fs_info
->uuid_root
);
3813 btrfs_warn(fs_info
, "btrfs_uuid_scan_kthread failed %d", ret
);
3815 fs_info
->update_uuid_tree_gen
= 1;
3816 up(&fs_info
->uuid_tree_rescan_sem
);
3821 * Callback for btrfs_uuid_tree_iterate().
3823 * 0 check succeeded, the entry is not outdated.
3824 * < 0 if an error occured.
3825 * > 0 if the check failed, which means the caller shall remove the entry.
3827 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info
*fs_info
,
3828 u8
*uuid
, u8 type
, u64 subid
)
3830 struct btrfs_key key
;
3832 struct btrfs_root
*subvol_root
;
3834 if (type
!= BTRFS_UUID_KEY_SUBVOL
&&
3835 type
!= BTRFS_UUID_KEY_RECEIVED_SUBVOL
)
3838 key
.objectid
= subid
;
3839 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3840 key
.offset
= (u64
)-1;
3841 subvol_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
3842 if (IS_ERR(subvol_root
)) {
3843 ret
= PTR_ERR(subvol_root
);
3850 case BTRFS_UUID_KEY_SUBVOL
:
3851 if (memcmp(uuid
, subvol_root
->root_item
.uuid
, BTRFS_UUID_SIZE
))
3854 case BTRFS_UUID_KEY_RECEIVED_SUBVOL
:
3855 if (memcmp(uuid
, subvol_root
->root_item
.received_uuid
,
3865 static int btrfs_uuid_rescan_kthread(void *data
)
3867 struct btrfs_fs_info
*fs_info
= (struct btrfs_fs_info
*)data
;
3871 * 1st step is to iterate through the existing UUID tree and
3872 * to delete all entries that contain outdated data.
3873 * 2nd step is to add all missing entries to the UUID tree.
3875 ret
= btrfs_uuid_tree_iterate(fs_info
, btrfs_check_uuid_tree_entry
);
3877 btrfs_warn(fs_info
, "iterating uuid_tree failed %d", ret
);
3878 up(&fs_info
->uuid_tree_rescan_sem
);
3881 return btrfs_uuid_scan_kthread(data
);
3884 int btrfs_create_uuid_tree(struct btrfs_fs_info
*fs_info
)
3886 struct btrfs_trans_handle
*trans
;
3887 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
3888 struct btrfs_root
*uuid_root
;
3889 struct task_struct
*task
;
3896 trans
= btrfs_start_transaction(tree_root
, 2);
3898 return PTR_ERR(trans
);
3900 uuid_root
= btrfs_create_tree(trans
, fs_info
,
3901 BTRFS_UUID_TREE_OBJECTID
);
3902 if (IS_ERR(uuid_root
)) {
3903 btrfs_abort_transaction(trans
, tree_root
,
3904 PTR_ERR(uuid_root
));
3905 return PTR_ERR(uuid_root
);
3908 fs_info
->uuid_root
= uuid_root
;
3910 ret
= btrfs_commit_transaction(trans
, tree_root
);
3914 down(&fs_info
->uuid_tree_rescan_sem
);
3915 task
= kthread_run(btrfs_uuid_scan_kthread
, fs_info
, "btrfs-uuid");
3917 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3918 btrfs_warn(fs_info
, "failed to start uuid_scan task");
3919 up(&fs_info
->uuid_tree_rescan_sem
);
3920 return PTR_ERR(task
);
3926 int btrfs_check_uuid_tree(struct btrfs_fs_info
*fs_info
)
3928 struct task_struct
*task
;
3930 down(&fs_info
->uuid_tree_rescan_sem
);
3931 task
= kthread_run(btrfs_uuid_rescan_kthread
, fs_info
, "btrfs-uuid");
3933 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3934 btrfs_warn(fs_info
, "failed to start uuid_rescan task");
3935 up(&fs_info
->uuid_tree_rescan_sem
);
3936 return PTR_ERR(task
);
3943 * shrinking a device means finding all of the device extents past
3944 * the new size, and then following the back refs to the chunks.
3945 * The chunk relocation code actually frees the device extent
3947 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
3949 struct btrfs_trans_handle
*trans
;
3950 struct btrfs_root
*root
= device
->dev_root
;
3951 struct btrfs_dev_extent
*dev_extent
= NULL
;
3952 struct btrfs_path
*path
;
3959 bool retried
= false;
3960 struct extent_buffer
*l
;
3961 struct btrfs_key key
;
3962 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
3963 u64 old_total
= btrfs_super_total_bytes(super_copy
);
3964 u64 old_size
= btrfs_device_get_total_bytes(device
);
3965 u64 diff
= old_size
- new_size
;
3967 if (device
->is_tgtdev_for_dev_replace
)
3970 path
= btrfs_alloc_path();
3978 btrfs_device_set_total_bytes(device
, new_size
);
3979 if (device
->writeable
) {
3980 device
->fs_devices
->total_rw_bytes
-= diff
;
3981 spin_lock(&root
->fs_info
->free_chunk_lock
);
3982 root
->fs_info
->free_chunk_space
-= diff
;
3983 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3985 unlock_chunks(root
);
3988 key
.objectid
= device
->devid
;
3989 key
.offset
= (u64
)-1;
3990 key
.type
= BTRFS_DEV_EXTENT_KEY
;
3993 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3997 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
4002 btrfs_release_path(path
);
4007 slot
= path
->slots
[0];
4008 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
4010 if (key
.objectid
!= device
->devid
) {
4011 btrfs_release_path(path
);
4015 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
4016 length
= btrfs_dev_extent_length(l
, dev_extent
);
4018 if (key
.offset
+ length
<= new_size
) {
4019 btrfs_release_path(path
);
4023 chunk_objectid
= btrfs_dev_extent_chunk_objectid(l
, dev_extent
);
4024 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
4025 btrfs_release_path(path
);
4027 ret
= btrfs_relocate_chunk(root
, chunk_objectid
, chunk_offset
);
4028 if (ret
&& ret
!= -ENOSPC
)
4032 } while (key
.offset
-- > 0);
4034 if (failed
&& !retried
) {
4038 } else if (failed
&& retried
) {
4042 btrfs_device_set_total_bytes(device
, old_size
);
4043 if (device
->writeable
)
4044 device
->fs_devices
->total_rw_bytes
+= diff
;
4045 spin_lock(&root
->fs_info
->free_chunk_lock
);
4046 root
->fs_info
->free_chunk_space
+= diff
;
4047 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4048 unlock_chunks(root
);
4052 /* Shrinking succeeded, else we would be at "done". */
4053 trans
= btrfs_start_transaction(root
, 0);
4054 if (IS_ERR(trans
)) {
4055 ret
= PTR_ERR(trans
);
4060 btrfs_device_set_disk_total_bytes(device
, new_size
);
4061 if (list_empty(&device
->resized_list
))
4062 list_add_tail(&device
->resized_list
,
4063 &root
->fs_info
->fs_devices
->resized_devices
);
4065 WARN_ON(diff
> old_total
);
4066 btrfs_set_super_total_bytes(super_copy
, old_total
- diff
);
4067 unlock_chunks(root
);
4069 /* Now btrfs_update_device() will change the on-disk size. */
4070 ret
= btrfs_update_device(trans
, device
);
4071 btrfs_end_transaction(trans
, root
);
4073 btrfs_free_path(path
);
4077 static int btrfs_add_system_chunk(struct btrfs_root
*root
,
4078 struct btrfs_key
*key
,
4079 struct btrfs_chunk
*chunk
, int item_size
)
4081 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
4082 struct btrfs_disk_key disk_key
;
4087 array_size
= btrfs_super_sys_array_size(super_copy
);
4088 if (array_size
+ item_size
+ sizeof(disk_key
)
4089 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4090 unlock_chunks(root
);
4094 ptr
= super_copy
->sys_chunk_array
+ array_size
;
4095 btrfs_cpu_key_to_disk(&disk_key
, key
);
4096 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
4097 ptr
+= sizeof(disk_key
);
4098 memcpy(ptr
, chunk
, item_size
);
4099 item_size
+= sizeof(disk_key
);
4100 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
4101 unlock_chunks(root
);
4107 * sort the devices in descending order by max_avail, total_avail
4109 static int btrfs_cmp_device_info(const void *a
, const void *b
)
4111 const struct btrfs_device_info
*di_a
= a
;
4112 const struct btrfs_device_info
*di_b
= b
;
4114 if (di_a
->max_avail
> di_b
->max_avail
)
4116 if (di_a
->max_avail
< di_b
->max_avail
)
4118 if (di_a
->total_avail
> di_b
->total_avail
)
4120 if (di_a
->total_avail
< di_b
->total_avail
)
4125 static const struct btrfs_raid_attr btrfs_raid_array
[BTRFS_NR_RAID_TYPES
] = {
4126 [BTRFS_RAID_RAID10
] = {
4129 .devs_max
= 0, /* 0 == as many as possible */
4131 .devs_increment
= 2,
4134 [BTRFS_RAID_RAID1
] = {
4139 .devs_increment
= 2,
4142 [BTRFS_RAID_DUP
] = {
4147 .devs_increment
= 1,
4150 [BTRFS_RAID_RAID0
] = {
4155 .devs_increment
= 1,
4158 [BTRFS_RAID_SINGLE
] = {
4163 .devs_increment
= 1,
4166 [BTRFS_RAID_RAID5
] = {
4171 .devs_increment
= 1,
4174 [BTRFS_RAID_RAID6
] = {
4179 .devs_increment
= 1,
4184 static u32
find_raid56_stripe_len(u32 data_devices
, u32 dev_stripe_target
)
4186 /* TODO allow them to set a preferred stripe size */
4190 static void check_raid56_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
4192 if (!(type
& BTRFS_BLOCK_GROUP_RAID56_MASK
))
4195 btrfs_set_fs_incompat(info
, RAID56
);
4198 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4199 - sizeof(struct btrfs_item) \
4200 - sizeof(struct btrfs_chunk)) \
4201 / sizeof(struct btrfs_stripe) + 1)
4203 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4204 - 2 * sizeof(struct btrfs_disk_key) \
4205 - 2 * sizeof(struct btrfs_chunk)) \
4206 / sizeof(struct btrfs_stripe) + 1)
4208 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4209 struct btrfs_root
*extent_root
, u64 start
,
4212 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
4213 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
4214 struct list_head
*cur
;
4215 struct map_lookup
*map
= NULL
;
4216 struct extent_map_tree
*em_tree
;
4217 struct extent_map
*em
;
4218 struct btrfs_device_info
*devices_info
= NULL
;
4220 int num_stripes
; /* total number of stripes to allocate */
4221 int data_stripes
; /* number of stripes that count for
4223 int sub_stripes
; /* sub_stripes info for map */
4224 int dev_stripes
; /* stripes per dev */
4225 int devs_max
; /* max devs to use */
4226 int devs_min
; /* min devs needed */
4227 int devs_increment
; /* ndevs has to be a multiple of this */
4228 int ncopies
; /* how many copies to data has */
4230 u64 max_stripe_size
;
4234 u64 raid_stripe_len
= BTRFS_STRIPE_LEN
;
4240 BUG_ON(!alloc_profile_is_valid(type
, 0));
4242 if (list_empty(&fs_devices
->alloc_list
))
4245 index
= __get_raid_index(type
);
4247 sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
4248 dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
4249 devs_max
= btrfs_raid_array
[index
].devs_max
;
4250 devs_min
= btrfs_raid_array
[index
].devs_min
;
4251 devs_increment
= btrfs_raid_array
[index
].devs_increment
;
4252 ncopies
= btrfs_raid_array
[index
].ncopies
;
4254 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
4255 max_stripe_size
= 1024 * 1024 * 1024;
4256 max_chunk_size
= 10 * max_stripe_size
;
4258 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4259 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
4260 /* for larger filesystems, use larger metadata chunks */
4261 if (fs_devices
->total_rw_bytes
> 50ULL * 1024 * 1024 * 1024)
4262 max_stripe_size
= 1024 * 1024 * 1024;
4264 max_stripe_size
= 256 * 1024 * 1024;
4265 max_chunk_size
= max_stripe_size
;
4267 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4268 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4269 max_stripe_size
= 32 * 1024 * 1024;
4270 max_chunk_size
= 2 * max_stripe_size
;
4272 devs_max
= BTRFS_MAX_DEVS_SYS_CHUNK
;
4274 btrfs_err(info
, "invalid chunk type 0x%llx requested",
4279 /* we don't want a chunk larger than 10% of writeable space */
4280 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
4283 devices_info
= kzalloc(sizeof(*devices_info
) * fs_devices
->rw_devices
,
4288 cur
= fs_devices
->alloc_list
.next
;
4291 * in the first pass through the devices list, we gather information
4292 * about the available holes on each device.
4295 while (cur
!= &fs_devices
->alloc_list
) {
4296 struct btrfs_device
*device
;
4300 device
= list_entry(cur
, struct btrfs_device
, dev_alloc_list
);
4304 if (!device
->writeable
) {
4306 "BTRFS: read-only device in alloc_list\n");
4310 if (!device
->in_fs_metadata
||
4311 device
->is_tgtdev_for_dev_replace
)
4314 if (device
->total_bytes
> device
->bytes_used
)
4315 total_avail
= device
->total_bytes
- device
->bytes_used
;
4319 /* If there is no space on this device, skip it. */
4320 if (total_avail
== 0)
4323 ret
= find_free_dev_extent(trans
, device
,
4324 max_stripe_size
* dev_stripes
,
4325 &dev_offset
, &max_avail
);
4326 if (ret
&& ret
!= -ENOSPC
)
4330 max_avail
= max_stripe_size
* dev_stripes
;
4332 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
4335 if (ndevs
== fs_devices
->rw_devices
) {
4336 WARN(1, "%s: found more than %llu devices\n",
4337 __func__
, fs_devices
->rw_devices
);
4340 devices_info
[ndevs
].dev_offset
= dev_offset
;
4341 devices_info
[ndevs
].max_avail
= max_avail
;
4342 devices_info
[ndevs
].total_avail
= total_avail
;
4343 devices_info
[ndevs
].dev
= device
;
4348 * now sort the devices by hole size / available space
4350 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
4351 btrfs_cmp_device_info
, NULL
);
4353 /* round down to number of usable stripes */
4354 ndevs
-= ndevs
% devs_increment
;
4356 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
4361 if (devs_max
&& ndevs
> devs_max
)
4364 * the primary goal is to maximize the number of stripes, so use as many
4365 * devices as possible, even if the stripes are not maximum sized.
4367 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4368 num_stripes
= ndevs
* dev_stripes
;
4371 * this will have to be fixed for RAID1 and RAID10 over
4374 data_stripes
= num_stripes
/ ncopies
;
4376 if (type
& BTRFS_BLOCK_GROUP_RAID5
) {
4377 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 1,
4378 btrfs_super_stripesize(info
->super_copy
));
4379 data_stripes
= num_stripes
- 1;
4381 if (type
& BTRFS_BLOCK_GROUP_RAID6
) {
4382 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 2,
4383 btrfs_super_stripesize(info
->super_copy
));
4384 data_stripes
= num_stripes
- 2;
4388 * Use the number of data stripes to figure out how big this chunk
4389 * is really going to be in terms of logical address space,
4390 * and compare that answer with the max chunk size
4392 if (stripe_size
* data_stripes
> max_chunk_size
) {
4393 u64 mask
= (1ULL << 24) - 1;
4394 stripe_size
= max_chunk_size
;
4395 do_div(stripe_size
, data_stripes
);
4397 /* bump the answer up to a 16MB boundary */
4398 stripe_size
= (stripe_size
+ mask
) & ~mask
;
4400 /* but don't go higher than the limits we found
4401 * while searching for free extents
4403 if (stripe_size
> devices_info
[ndevs
-1].max_avail
)
4404 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4407 do_div(stripe_size
, dev_stripes
);
4409 /* align to BTRFS_STRIPE_LEN */
4410 do_div(stripe_size
, raid_stripe_len
);
4411 stripe_size
*= raid_stripe_len
;
4413 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
4418 map
->num_stripes
= num_stripes
;
4420 for (i
= 0; i
< ndevs
; ++i
) {
4421 for (j
= 0; j
< dev_stripes
; ++j
) {
4422 int s
= i
* dev_stripes
+ j
;
4423 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
4424 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
4428 map
->sector_size
= extent_root
->sectorsize
;
4429 map
->stripe_len
= raid_stripe_len
;
4430 map
->io_align
= raid_stripe_len
;
4431 map
->io_width
= raid_stripe_len
;
4433 map
->sub_stripes
= sub_stripes
;
4435 num_bytes
= stripe_size
* data_stripes
;
4437 trace_btrfs_chunk_alloc(info
->chunk_root
, map
, start
, num_bytes
);
4439 em
= alloc_extent_map();
4445 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
4446 em
->bdev
= (struct block_device
*)map
;
4448 em
->len
= num_bytes
;
4449 em
->block_start
= 0;
4450 em
->block_len
= em
->len
;
4451 em
->orig_block_len
= stripe_size
;
4453 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
4454 write_lock(&em_tree
->lock
);
4455 ret
= add_extent_mapping(em_tree
, em
, 0);
4457 list_add_tail(&em
->list
, &trans
->transaction
->pending_chunks
);
4458 atomic_inc(&em
->refs
);
4460 write_unlock(&em_tree
->lock
);
4462 free_extent_map(em
);
4466 ret
= btrfs_make_block_group(trans
, extent_root
, 0, type
,
4467 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4470 goto error_del_extent
;
4472 for (i
= 0; i
< map
->num_stripes
; i
++) {
4473 num_bytes
= map
->stripes
[i
].dev
->bytes_used
+ stripe_size
;
4474 btrfs_device_set_bytes_used(map
->stripes
[i
].dev
, num_bytes
);
4477 spin_lock(&extent_root
->fs_info
->free_chunk_lock
);
4478 extent_root
->fs_info
->free_chunk_space
-= (stripe_size
*
4480 spin_unlock(&extent_root
->fs_info
->free_chunk_lock
);
4482 free_extent_map(em
);
4483 check_raid56_incompat_flag(extent_root
->fs_info
, type
);
4485 kfree(devices_info
);
4489 write_lock(&em_tree
->lock
);
4490 remove_extent_mapping(em_tree
, em
);
4491 write_unlock(&em_tree
->lock
);
4493 /* One for our allocation */
4494 free_extent_map(em
);
4495 /* One for the tree reference */
4496 free_extent_map(em
);
4497 /* One for the pending_chunks list reference */
4498 free_extent_map(em
);
4500 kfree(devices_info
);
4504 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
4505 struct btrfs_root
*extent_root
,
4506 u64 chunk_offset
, u64 chunk_size
)
4508 struct btrfs_key key
;
4509 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
4510 struct btrfs_device
*device
;
4511 struct btrfs_chunk
*chunk
;
4512 struct btrfs_stripe
*stripe
;
4513 struct extent_map_tree
*em_tree
;
4514 struct extent_map
*em
;
4515 struct map_lookup
*map
;
4522 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
4523 read_lock(&em_tree
->lock
);
4524 em
= lookup_extent_mapping(em_tree
, chunk_offset
, chunk_size
);
4525 read_unlock(&em_tree
->lock
);
4528 btrfs_crit(extent_root
->fs_info
, "unable to find logical "
4529 "%Lu len %Lu", chunk_offset
, chunk_size
);
4533 if (em
->start
!= chunk_offset
|| em
->len
!= chunk_size
) {
4534 btrfs_crit(extent_root
->fs_info
, "found a bad mapping, wanted"
4535 " %Lu-%Lu, found %Lu-%Lu", chunk_offset
,
4536 chunk_size
, em
->start
, em
->len
);
4537 free_extent_map(em
);
4541 map
= (struct map_lookup
*)em
->bdev
;
4542 item_size
= btrfs_chunk_item_size(map
->num_stripes
);
4543 stripe_size
= em
->orig_block_len
;
4545 chunk
= kzalloc(item_size
, GFP_NOFS
);
4551 for (i
= 0; i
< map
->num_stripes
; i
++) {
4552 device
= map
->stripes
[i
].dev
;
4553 dev_offset
= map
->stripes
[i
].physical
;
4555 ret
= btrfs_update_device(trans
, device
);
4558 ret
= btrfs_alloc_dev_extent(trans
, device
,
4559 chunk_root
->root_key
.objectid
,
4560 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4561 chunk_offset
, dev_offset
,
4567 stripe
= &chunk
->stripe
;
4568 for (i
= 0; i
< map
->num_stripes
; i
++) {
4569 device
= map
->stripes
[i
].dev
;
4570 dev_offset
= map
->stripes
[i
].physical
;
4572 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
4573 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
4574 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
4578 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
4579 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
4580 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
4581 btrfs_set_stack_chunk_type(chunk
, map
->type
);
4582 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
4583 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
4584 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
4585 btrfs_set_stack_chunk_sector_size(chunk
, extent_root
->sectorsize
);
4586 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
4588 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
4589 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
4590 key
.offset
= chunk_offset
;
4592 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
4593 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4595 * TODO: Cleanup of inserted chunk root in case of
4598 ret
= btrfs_add_system_chunk(chunk_root
, &key
, chunk
,
4604 free_extent_map(em
);
4609 * Chunk allocation falls into two parts. The first part does works
4610 * that make the new allocated chunk useable, but not do any operation
4611 * that modifies the chunk tree. The second part does the works that
4612 * require modifying the chunk tree. This division is important for the
4613 * bootstrap process of adding storage to a seed btrfs.
4615 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4616 struct btrfs_root
*extent_root
, u64 type
)
4620 chunk_offset
= find_next_chunk(extent_root
->fs_info
);
4621 return __btrfs_alloc_chunk(trans
, extent_root
, chunk_offset
, type
);
4624 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
4625 struct btrfs_root
*root
,
4626 struct btrfs_device
*device
)
4629 u64 sys_chunk_offset
;
4631 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4632 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
4635 chunk_offset
= find_next_chunk(fs_info
);
4636 alloc_profile
= btrfs_get_alloc_profile(extent_root
, 0);
4637 ret
= __btrfs_alloc_chunk(trans
, extent_root
, chunk_offset
,
4642 sys_chunk_offset
= find_next_chunk(root
->fs_info
);
4643 alloc_profile
= btrfs_get_alloc_profile(fs_info
->chunk_root
, 0);
4644 ret
= __btrfs_alloc_chunk(trans
, extent_root
, sys_chunk_offset
,
4649 static inline int btrfs_chunk_max_errors(struct map_lookup
*map
)
4653 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
4654 BTRFS_BLOCK_GROUP_RAID10
|
4655 BTRFS_BLOCK_GROUP_RAID5
|
4656 BTRFS_BLOCK_GROUP_DUP
)) {
4658 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
) {
4667 int btrfs_chunk_readonly(struct btrfs_root
*root
, u64 chunk_offset
)
4669 struct extent_map
*em
;
4670 struct map_lookup
*map
;
4671 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
4676 read_lock(&map_tree
->map_tree
.lock
);
4677 em
= lookup_extent_mapping(&map_tree
->map_tree
, chunk_offset
, 1);
4678 read_unlock(&map_tree
->map_tree
.lock
);
4682 map
= (struct map_lookup
*)em
->bdev
;
4683 for (i
= 0; i
< map
->num_stripes
; i
++) {
4684 if (map
->stripes
[i
].dev
->missing
) {
4689 if (!map
->stripes
[i
].dev
->writeable
) {
4696 * If the number of missing devices is larger than max errors,
4697 * we can not write the data into that chunk successfully, so
4700 if (miss_ndevs
> btrfs_chunk_max_errors(map
))
4703 free_extent_map(em
);
4707 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
4709 extent_map_tree_init(&tree
->map_tree
);
4712 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
4714 struct extent_map
*em
;
4717 write_lock(&tree
->map_tree
.lock
);
4718 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
4720 remove_extent_mapping(&tree
->map_tree
, em
);
4721 write_unlock(&tree
->map_tree
.lock
);
4725 free_extent_map(em
);
4726 /* once for the tree */
4727 free_extent_map(em
);
4731 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
4733 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
4734 struct extent_map
*em
;
4735 struct map_lookup
*map
;
4736 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4739 read_lock(&em_tree
->lock
);
4740 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4741 read_unlock(&em_tree
->lock
);
4744 * We could return errors for these cases, but that could get ugly and
4745 * we'd probably do the same thing which is just not do anything else
4746 * and exit, so return 1 so the callers don't try to use other copies.
4749 btrfs_crit(fs_info
, "No mapping for %Lu-%Lu", logical
,
4754 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
4755 btrfs_crit(fs_info
, "Invalid mapping for %Lu-%Lu, got "
4756 "%Lu-%Lu", logical
, logical
+len
, em
->start
,
4757 em
->start
+ em
->len
);
4758 free_extent_map(em
);
4762 map
= (struct map_lookup
*)em
->bdev
;
4763 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
4764 ret
= map
->num_stripes
;
4765 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
4766 ret
= map
->sub_stripes
;
4767 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID5
)
4769 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
4773 free_extent_map(em
);
4775 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
4776 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
))
4778 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
4783 unsigned long btrfs_full_stripe_len(struct btrfs_root
*root
,
4784 struct btrfs_mapping_tree
*map_tree
,
4787 struct extent_map
*em
;
4788 struct map_lookup
*map
;
4789 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4790 unsigned long len
= root
->sectorsize
;
4792 read_lock(&em_tree
->lock
);
4793 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4794 read_unlock(&em_tree
->lock
);
4797 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4798 map
= (struct map_lookup
*)em
->bdev
;
4799 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
4800 len
= map
->stripe_len
* nr_data_stripes(map
);
4801 free_extent_map(em
);
4805 int btrfs_is_parity_mirror(struct btrfs_mapping_tree
*map_tree
,
4806 u64 logical
, u64 len
, int mirror_num
)
4808 struct extent_map
*em
;
4809 struct map_lookup
*map
;
4810 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4813 read_lock(&em_tree
->lock
);
4814 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4815 read_unlock(&em_tree
->lock
);
4818 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4819 map
= (struct map_lookup
*)em
->bdev
;
4820 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
4822 free_extent_map(em
);
4826 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
4827 struct map_lookup
*map
, int first
, int num
,
4828 int optimal
, int dev_replace_is_ongoing
)
4832 struct btrfs_device
*srcdev
;
4834 if (dev_replace_is_ongoing
&&
4835 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
4836 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
4837 srcdev
= fs_info
->dev_replace
.srcdev
;
4842 * try to avoid the drive that is the source drive for a
4843 * dev-replace procedure, only choose it if no other non-missing
4844 * mirror is available
4846 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
4847 if (map
->stripes
[optimal
].dev
->bdev
&&
4848 (tolerance
|| map
->stripes
[optimal
].dev
!= srcdev
))
4850 for (i
= first
; i
< first
+ num
; i
++) {
4851 if (map
->stripes
[i
].dev
->bdev
&&
4852 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
4857 /* we couldn't find one that doesn't fail. Just return something
4858 * and the io error handling code will clean up eventually
4863 static inline int parity_smaller(u64 a
, u64 b
)
4868 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4869 static void sort_parity_stripes(struct btrfs_bio
*bbio
, int num_stripes
)
4871 struct btrfs_bio_stripe s
;
4878 for (i
= 0; i
< num_stripes
- 1; i
++) {
4879 if (parity_smaller(bbio
->raid_map
[i
],
4880 bbio
->raid_map
[i
+1])) {
4881 s
= bbio
->stripes
[i
];
4882 l
= bbio
->raid_map
[i
];
4883 bbio
->stripes
[i
] = bbio
->stripes
[i
+1];
4884 bbio
->raid_map
[i
] = bbio
->raid_map
[i
+1];
4885 bbio
->stripes
[i
+1] = s
;
4886 bbio
->raid_map
[i
+1] = l
;
4894 static struct btrfs_bio
*alloc_btrfs_bio(int total_stripes
, int real_stripes
)
4896 struct btrfs_bio
*bbio
= kzalloc(
4897 sizeof(struct btrfs_bio
) +
4898 sizeof(struct btrfs_bio_stripe
) * (total_stripes
) +
4899 sizeof(int) * (real_stripes
) +
4900 sizeof(u64
) * (real_stripes
),
4905 atomic_set(&bbio
->error
, 0);
4906 atomic_set(&bbio
->refs
, 1);
4911 void btrfs_get_bbio(struct btrfs_bio
*bbio
)
4913 WARN_ON(!atomic_read(&bbio
->refs
));
4914 atomic_inc(&bbio
->refs
);
4917 void btrfs_put_bbio(struct btrfs_bio
*bbio
)
4921 if (atomic_dec_and_test(&bbio
->refs
))
4925 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
4926 u64 logical
, u64
*length
,
4927 struct btrfs_bio
**bbio_ret
,
4928 int mirror_num
, int need_raid_map
)
4930 struct extent_map
*em
;
4931 struct map_lookup
*map
;
4932 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
4933 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4936 u64 stripe_end_offset
;
4946 int tgtdev_indexes
= 0;
4947 struct btrfs_bio
*bbio
= NULL
;
4948 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
4949 int dev_replace_is_ongoing
= 0;
4950 int num_alloc_stripes
;
4951 int patch_the_first_stripe_for_dev_replace
= 0;
4952 u64 physical_to_patch_in_first_stripe
= 0;
4953 u64 raid56_full_stripe_start
= (u64
)-1;
4955 read_lock(&em_tree
->lock
);
4956 em
= lookup_extent_mapping(em_tree
, logical
, *length
);
4957 read_unlock(&em_tree
->lock
);
4960 btrfs_crit(fs_info
, "unable to find logical %llu len %llu",
4965 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
4966 btrfs_crit(fs_info
, "found a bad mapping, wanted %Lu, "
4967 "found %Lu-%Lu", logical
, em
->start
,
4968 em
->start
+ em
->len
);
4969 free_extent_map(em
);
4973 map
= (struct map_lookup
*)em
->bdev
;
4974 offset
= logical
- em
->start
;
4976 stripe_len
= map
->stripe_len
;
4979 * stripe_nr counts the total number of stripes we have to stride
4980 * to get to this block
4982 do_div(stripe_nr
, stripe_len
);
4984 stripe_offset
= stripe_nr
* stripe_len
;
4985 BUG_ON(offset
< stripe_offset
);
4987 /* stripe_offset is the offset of this block in its stripe*/
4988 stripe_offset
= offset
- stripe_offset
;
4990 /* if we're here for raid56, we need to know the stripe aligned start */
4991 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
4992 unsigned long full_stripe_len
= stripe_len
* nr_data_stripes(map
);
4993 raid56_full_stripe_start
= offset
;
4995 /* allow a write of a full stripe, but make sure we don't
4996 * allow straddling of stripes
4998 do_div(raid56_full_stripe_start
, full_stripe_len
);
4999 raid56_full_stripe_start
*= full_stripe_len
;
5002 if (rw
& REQ_DISCARD
) {
5003 /* we don't discard raid56 yet */
5004 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5008 *length
= min_t(u64
, em
->len
- offset
, *length
);
5009 } else if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
5011 /* For writes to RAID[56], allow a full stripeset across all disks.
5012 For other RAID types and for RAID[56] reads, just allow a single
5013 stripe (on a single disk). */
5014 if ((map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
5016 max_len
= stripe_len
* nr_data_stripes(map
) -
5017 (offset
- raid56_full_stripe_start
);
5019 /* we limit the length of each bio to what fits in a stripe */
5020 max_len
= stripe_len
- stripe_offset
;
5022 *length
= min_t(u64
, em
->len
- offset
, max_len
);
5024 *length
= em
->len
- offset
;
5027 /* This is for when we're called from btrfs_merge_bio_hook() and all
5028 it cares about is the length */
5032 btrfs_dev_replace_lock(dev_replace
);
5033 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
5034 if (!dev_replace_is_ongoing
)
5035 btrfs_dev_replace_unlock(dev_replace
);
5037 if (dev_replace_is_ongoing
&& mirror_num
== map
->num_stripes
+ 1 &&
5038 !(rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) &&
5039 dev_replace
->tgtdev
!= NULL
) {
5041 * in dev-replace case, for repair case (that's the only
5042 * case where the mirror is selected explicitly when
5043 * calling btrfs_map_block), blocks left of the left cursor
5044 * can also be read from the target drive.
5045 * For REQ_GET_READ_MIRRORS, the target drive is added as
5046 * the last one to the array of stripes. For READ, it also
5047 * needs to be supported using the same mirror number.
5048 * If the requested block is not left of the left cursor,
5049 * EIO is returned. This can happen because btrfs_num_copies()
5050 * returns one more in the dev-replace case.
5052 u64 tmp_length
= *length
;
5053 struct btrfs_bio
*tmp_bbio
= NULL
;
5054 int tmp_num_stripes
;
5055 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5056 int index_srcdev
= 0;
5058 u64 physical_of_found
= 0;
5060 ret
= __btrfs_map_block(fs_info
, REQ_GET_READ_MIRRORS
,
5061 logical
, &tmp_length
, &tmp_bbio
, 0, 0);
5063 WARN_ON(tmp_bbio
!= NULL
);
5067 tmp_num_stripes
= tmp_bbio
->num_stripes
;
5068 if (mirror_num
> tmp_num_stripes
) {
5070 * REQ_GET_READ_MIRRORS does not contain this
5071 * mirror, that means that the requested area
5072 * is not left of the left cursor
5075 btrfs_put_bbio(tmp_bbio
);
5080 * process the rest of the function using the mirror_num
5081 * of the source drive. Therefore look it up first.
5082 * At the end, patch the device pointer to the one of the
5085 for (i
= 0; i
< tmp_num_stripes
; i
++) {
5086 if (tmp_bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5088 * In case of DUP, in order to keep it
5089 * simple, only add the mirror with the
5090 * lowest physical address
5093 physical_of_found
<=
5094 tmp_bbio
->stripes
[i
].physical
)
5099 tmp_bbio
->stripes
[i
].physical
;
5104 mirror_num
= index_srcdev
+ 1;
5105 patch_the_first_stripe_for_dev_replace
= 1;
5106 physical_to_patch_in_first_stripe
= physical_of_found
;
5110 btrfs_put_bbio(tmp_bbio
);
5114 btrfs_put_bbio(tmp_bbio
);
5115 } else if (mirror_num
> map
->num_stripes
) {
5121 stripe_nr_orig
= stripe_nr
;
5122 stripe_nr_end
= ALIGN(offset
+ *length
, map
->stripe_len
);
5123 do_div(stripe_nr_end
, map
->stripe_len
);
5124 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
5127 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5128 if (rw
& REQ_DISCARD
)
5129 num_stripes
= min_t(u64
, map
->num_stripes
,
5130 stripe_nr_end
- stripe_nr_orig
);
5131 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
5132 if (!(rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)))
5134 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
5135 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
))
5136 num_stripes
= map
->num_stripes
;
5137 else if (mirror_num
)
5138 stripe_index
= mirror_num
- 1;
5140 stripe_index
= find_live_mirror(fs_info
, map
, 0,
5142 current
->pid
% map
->num_stripes
,
5143 dev_replace_is_ongoing
);
5144 mirror_num
= stripe_index
+ 1;
5147 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
5148 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) {
5149 num_stripes
= map
->num_stripes
;
5150 } else if (mirror_num
) {
5151 stripe_index
= mirror_num
- 1;
5156 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5157 int factor
= map
->num_stripes
/ map
->sub_stripes
;
5159 stripe_index
= do_div(stripe_nr
, factor
);
5160 stripe_index
*= map
->sub_stripes
;
5162 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
))
5163 num_stripes
= map
->sub_stripes
;
5164 else if (rw
& REQ_DISCARD
)
5165 num_stripes
= min_t(u64
, map
->sub_stripes
*
5166 (stripe_nr_end
- stripe_nr_orig
),
5168 else if (mirror_num
)
5169 stripe_index
+= mirror_num
- 1;
5171 int old_stripe_index
= stripe_index
;
5172 stripe_index
= find_live_mirror(fs_info
, map
,
5174 map
->sub_stripes
, stripe_index
+
5175 current
->pid
% map
->sub_stripes
,
5176 dev_replace_is_ongoing
);
5177 mirror_num
= stripe_index
- old_stripe_index
+ 1;
5180 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5181 if (need_raid_map
&&
5182 ((rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
)) ||
5184 /* push stripe_nr back to the start of the full stripe */
5185 stripe_nr
= raid56_full_stripe_start
;
5186 do_div(stripe_nr
, stripe_len
* nr_data_stripes(map
));
5188 /* RAID[56] write or recovery. Return all stripes */
5189 num_stripes
= map
->num_stripes
;
5190 max_errors
= nr_parity_stripes(map
);
5192 *length
= map
->stripe_len
;
5199 * Mirror #0 or #1 means the original data block.
5200 * Mirror #2 is RAID5 parity block.
5201 * Mirror #3 is RAID6 Q block.
5203 stripe_index
= do_div(stripe_nr
, nr_data_stripes(map
));
5205 stripe_index
= nr_data_stripes(map
) +
5208 /* We distribute the parity blocks across stripes */
5209 tmp
= stripe_nr
+ stripe_index
;
5210 stripe_index
= do_div(tmp
, map
->num_stripes
);
5211 if (!(rw
& (REQ_WRITE
| REQ_DISCARD
|
5212 REQ_GET_READ_MIRRORS
)) && mirror_num
<= 1)
5217 * after this do_div call, stripe_nr is the number of stripes
5218 * on this device we have to walk to find the data, and
5219 * stripe_index is the number of our device in the stripe array
5221 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
5222 mirror_num
= stripe_index
+ 1;
5224 BUG_ON(stripe_index
>= map
->num_stripes
);
5226 num_alloc_stripes
= num_stripes
;
5227 if (dev_replace_is_ongoing
) {
5228 if (rw
& (REQ_WRITE
| REQ_DISCARD
))
5229 num_alloc_stripes
<<= 1;
5230 if (rw
& REQ_GET_READ_MIRRORS
)
5231 num_alloc_stripes
++;
5232 tgtdev_indexes
= num_stripes
;
5235 bbio
= alloc_btrfs_bio(num_alloc_stripes
, tgtdev_indexes
);
5240 if (dev_replace_is_ongoing
)
5241 bbio
->tgtdev_map
= (int *)(bbio
->stripes
+ num_alloc_stripes
);
5243 /* build raid_map */
5244 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
&&
5245 need_raid_map
&& ((rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
)) ||
5250 bbio
->raid_map
= (u64
*)((void *)bbio
->stripes
+
5251 sizeof(struct btrfs_bio_stripe
) *
5253 sizeof(int) * tgtdev_indexes
);
5255 /* Work out the disk rotation on this stripe-set */
5257 rot
= do_div(tmp
, num_stripes
);
5259 /* Fill in the logical address of each stripe */
5260 tmp
= stripe_nr
* nr_data_stripes(map
);
5261 for (i
= 0; i
< nr_data_stripes(map
); i
++)
5262 bbio
->raid_map
[(i
+rot
) % num_stripes
] =
5263 em
->start
+ (tmp
+ i
) * map
->stripe_len
;
5265 bbio
->raid_map
[(i
+rot
) % map
->num_stripes
] = RAID5_P_STRIPE
;
5266 if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5267 bbio
->raid_map
[(i
+rot
+1) % num_stripes
] =
5271 if (rw
& REQ_DISCARD
) {
5273 int sub_stripes
= 0;
5274 u64 stripes_per_dev
= 0;
5275 u32 remaining_stripes
= 0;
5276 u32 last_stripe
= 0;
5279 (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID10
)) {
5280 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5283 sub_stripes
= map
->sub_stripes
;
5285 factor
= map
->num_stripes
/ sub_stripes
;
5286 stripes_per_dev
= div_u64_rem(stripe_nr_end
-
5289 &remaining_stripes
);
5290 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
5291 last_stripe
*= sub_stripes
;
5294 for (i
= 0; i
< num_stripes
; i
++) {
5295 bbio
->stripes
[i
].physical
=
5296 map
->stripes
[stripe_index
].physical
+
5297 stripe_offset
+ stripe_nr
* map
->stripe_len
;
5298 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
5300 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5301 BTRFS_BLOCK_GROUP_RAID10
)) {
5302 bbio
->stripes
[i
].length
= stripes_per_dev
*
5305 if (i
/ sub_stripes
< remaining_stripes
)
5306 bbio
->stripes
[i
].length
+=
5310 * Special for the first stripe and
5313 * |-------|...|-------|
5317 if (i
< sub_stripes
)
5318 bbio
->stripes
[i
].length
-=
5321 if (stripe_index
>= last_stripe
&&
5322 stripe_index
<= (last_stripe
+
5324 bbio
->stripes
[i
].length
-=
5327 if (i
== sub_stripes
- 1)
5330 bbio
->stripes
[i
].length
= *length
;
5333 if (stripe_index
== map
->num_stripes
) {
5334 /* This could only happen for RAID0/10 */
5340 for (i
= 0; i
< num_stripes
; i
++) {
5341 bbio
->stripes
[i
].physical
=
5342 map
->stripes
[stripe_index
].physical
+
5344 stripe_nr
* map
->stripe_len
;
5345 bbio
->stripes
[i
].dev
=
5346 map
->stripes
[stripe_index
].dev
;
5351 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
))
5352 max_errors
= btrfs_chunk_max_errors(map
);
5355 sort_parity_stripes(bbio
, num_stripes
);
5358 if (dev_replace_is_ongoing
&& (rw
& (REQ_WRITE
| REQ_DISCARD
)) &&
5359 dev_replace
->tgtdev
!= NULL
) {
5360 int index_where_to_add
;
5361 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5364 * duplicate the write operations while the dev replace
5365 * procedure is running. Since the copying of the old disk
5366 * to the new disk takes place at run time while the
5367 * filesystem is mounted writable, the regular write
5368 * operations to the old disk have to be duplicated to go
5369 * to the new disk as well.
5370 * Note that device->missing is handled by the caller, and
5371 * that the write to the old disk is already set up in the
5374 index_where_to_add
= num_stripes
;
5375 for (i
= 0; i
< num_stripes
; i
++) {
5376 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5377 /* write to new disk, too */
5378 struct btrfs_bio_stripe
*new =
5379 bbio
->stripes
+ index_where_to_add
;
5380 struct btrfs_bio_stripe
*old
=
5383 new->physical
= old
->physical
;
5384 new->length
= old
->length
;
5385 new->dev
= dev_replace
->tgtdev
;
5386 bbio
->tgtdev_map
[i
] = index_where_to_add
;
5387 index_where_to_add
++;
5392 num_stripes
= index_where_to_add
;
5393 } else if (dev_replace_is_ongoing
&& (rw
& REQ_GET_READ_MIRRORS
) &&
5394 dev_replace
->tgtdev
!= NULL
) {
5395 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5396 int index_srcdev
= 0;
5398 u64 physical_of_found
= 0;
5401 * During the dev-replace procedure, the target drive can
5402 * also be used to read data in case it is needed to repair
5403 * a corrupt block elsewhere. This is possible if the
5404 * requested area is left of the left cursor. In this area,
5405 * the target drive is a full copy of the source drive.
5407 for (i
= 0; i
< num_stripes
; i
++) {
5408 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5410 * In case of DUP, in order to keep it
5411 * simple, only add the mirror with the
5412 * lowest physical address
5415 physical_of_found
<=
5416 bbio
->stripes
[i
].physical
)
5420 physical_of_found
= bbio
->stripes
[i
].physical
;
5424 u64 length
= map
->stripe_len
;
5426 if (physical_of_found
+ length
<=
5427 dev_replace
->cursor_left
) {
5428 struct btrfs_bio_stripe
*tgtdev_stripe
=
5429 bbio
->stripes
+ num_stripes
;
5431 tgtdev_stripe
->physical
= physical_of_found
;
5432 tgtdev_stripe
->length
=
5433 bbio
->stripes
[index_srcdev
].length
;
5434 tgtdev_stripe
->dev
= dev_replace
->tgtdev
;
5435 bbio
->tgtdev_map
[index_srcdev
] = num_stripes
;
5444 bbio
->map_type
= map
->type
;
5445 bbio
->num_stripes
= num_stripes
;
5446 bbio
->max_errors
= max_errors
;
5447 bbio
->mirror_num
= mirror_num
;
5448 bbio
->num_tgtdevs
= tgtdev_indexes
;
5451 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5452 * mirror_num == num_stripes + 1 && dev_replace target drive is
5453 * available as a mirror
5455 if (patch_the_first_stripe_for_dev_replace
&& num_stripes
> 0) {
5456 WARN_ON(num_stripes
> 1);
5457 bbio
->stripes
[0].dev
= dev_replace
->tgtdev
;
5458 bbio
->stripes
[0].physical
= physical_to_patch_in_first_stripe
;
5459 bbio
->mirror_num
= map
->num_stripes
+ 1;
5462 if (dev_replace_is_ongoing
)
5463 btrfs_dev_replace_unlock(dev_replace
);
5464 free_extent_map(em
);
5468 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
5469 u64 logical
, u64
*length
,
5470 struct btrfs_bio
**bbio_ret
, int mirror_num
)
5472 return __btrfs_map_block(fs_info
, rw
, logical
, length
, bbio_ret
,
5476 /* For Scrub/replace */
5477 int btrfs_map_sblock(struct btrfs_fs_info
*fs_info
, int rw
,
5478 u64 logical
, u64
*length
,
5479 struct btrfs_bio
**bbio_ret
, int mirror_num
,
5482 return __btrfs_map_block(fs_info
, rw
, logical
, length
, bbio_ret
,
5483 mirror_num
, need_raid_map
);
5486 int btrfs_rmap_block(struct btrfs_mapping_tree
*map_tree
,
5487 u64 chunk_start
, u64 physical
, u64 devid
,
5488 u64
**logical
, int *naddrs
, int *stripe_len
)
5490 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5491 struct extent_map
*em
;
5492 struct map_lookup
*map
;
5500 read_lock(&em_tree
->lock
);
5501 em
= lookup_extent_mapping(em_tree
, chunk_start
, 1);
5502 read_unlock(&em_tree
->lock
);
5505 printk(KERN_ERR
"BTRFS: couldn't find em for chunk %Lu\n",
5510 if (em
->start
!= chunk_start
) {
5511 printk(KERN_ERR
"BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5512 em
->start
, chunk_start
);
5513 free_extent_map(em
);
5516 map
= (struct map_lookup
*)em
->bdev
;
5519 rmap_len
= map
->stripe_len
;
5521 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5522 do_div(length
, map
->num_stripes
/ map
->sub_stripes
);
5523 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5524 do_div(length
, map
->num_stripes
);
5525 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5526 do_div(length
, nr_data_stripes(map
));
5527 rmap_len
= map
->stripe_len
* nr_data_stripes(map
);
5530 buf
= kzalloc(sizeof(u64
) * map
->num_stripes
, GFP_NOFS
);
5531 BUG_ON(!buf
); /* -ENOMEM */
5533 for (i
= 0; i
< map
->num_stripes
; i
++) {
5534 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
5536 if (map
->stripes
[i
].physical
> physical
||
5537 map
->stripes
[i
].physical
+ length
<= physical
)
5540 stripe_nr
= physical
- map
->stripes
[i
].physical
;
5541 do_div(stripe_nr
, map
->stripe_len
);
5543 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5544 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5545 do_div(stripe_nr
, map
->sub_stripes
);
5546 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5547 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5548 } /* else if RAID[56], multiply by nr_data_stripes().
5549 * Alternatively, just use rmap_len below instead of
5550 * map->stripe_len */
5552 bytenr
= chunk_start
+ stripe_nr
* rmap_len
;
5553 WARN_ON(nr
>= map
->num_stripes
);
5554 for (j
= 0; j
< nr
; j
++) {
5555 if (buf
[j
] == bytenr
)
5559 WARN_ON(nr
>= map
->num_stripes
);
5566 *stripe_len
= rmap_len
;
5568 free_extent_map(em
);
5572 static inline void btrfs_end_bbio(struct btrfs_bio
*bbio
, struct bio
*bio
, int err
)
5574 if (likely(bbio
->flags
& BTRFS_BIO_ORIG_BIO_SUBMITTED
))
5575 bio_endio_nodec(bio
, err
);
5577 bio_endio(bio
, err
);
5578 btrfs_put_bbio(bbio
);
5581 static void btrfs_end_bio(struct bio
*bio
, int err
)
5583 struct btrfs_bio
*bbio
= bio
->bi_private
;
5584 struct btrfs_device
*dev
= bbio
->stripes
[0].dev
;
5585 int is_orig_bio
= 0;
5588 atomic_inc(&bbio
->error
);
5589 if (err
== -EIO
|| err
== -EREMOTEIO
) {
5590 unsigned int stripe_index
=
5591 btrfs_io_bio(bio
)->stripe_index
;
5593 BUG_ON(stripe_index
>= bbio
->num_stripes
);
5594 dev
= bbio
->stripes
[stripe_index
].dev
;
5596 if (bio
->bi_rw
& WRITE
)
5597 btrfs_dev_stat_inc(dev
,
5598 BTRFS_DEV_STAT_WRITE_ERRS
);
5600 btrfs_dev_stat_inc(dev
,
5601 BTRFS_DEV_STAT_READ_ERRS
);
5602 if ((bio
->bi_rw
& WRITE_FLUSH
) == WRITE_FLUSH
)
5603 btrfs_dev_stat_inc(dev
,
5604 BTRFS_DEV_STAT_FLUSH_ERRS
);
5605 btrfs_dev_stat_print_on_error(dev
);
5610 if (bio
== bbio
->orig_bio
)
5613 btrfs_bio_counter_dec(bbio
->fs_info
);
5615 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
5618 bio
= bbio
->orig_bio
;
5621 bio
->bi_private
= bbio
->private;
5622 bio
->bi_end_io
= bbio
->end_io
;
5623 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
5624 /* only send an error to the higher layers if it is
5625 * beyond the tolerance of the btrfs bio
5627 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
5631 * this bio is actually up to date, we didn't
5632 * go over the max number of errors
5634 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5638 btrfs_end_bbio(bbio
, bio
, err
);
5639 } else if (!is_orig_bio
) {
5645 * see run_scheduled_bios for a description of why bios are collected for
5648 * This will add one bio to the pending list for a device and make sure
5649 * the work struct is scheduled.
5651 static noinline
void btrfs_schedule_bio(struct btrfs_root
*root
,
5652 struct btrfs_device
*device
,
5653 int rw
, struct bio
*bio
)
5655 int should_queue
= 1;
5656 struct btrfs_pending_bios
*pending_bios
;
5658 if (device
->missing
|| !device
->bdev
) {
5659 bio_endio(bio
, -EIO
);
5663 /* don't bother with additional async steps for reads, right now */
5664 if (!(rw
& REQ_WRITE
)) {
5666 btrfsic_submit_bio(rw
, bio
);
5672 * nr_async_bios allows us to reliably return congestion to the
5673 * higher layers. Otherwise, the async bio makes it appear we have
5674 * made progress against dirty pages when we've really just put it
5675 * on a queue for later
5677 atomic_inc(&root
->fs_info
->nr_async_bios
);
5678 WARN_ON(bio
->bi_next
);
5679 bio
->bi_next
= NULL
;
5682 spin_lock(&device
->io_lock
);
5683 if (bio
->bi_rw
& REQ_SYNC
)
5684 pending_bios
= &device
->pending_sync_bios
;
5686 pending_bios
= &device
->pending_bios
;
5688 if (pending_bios
->tail
)
5689 pending_bios
->tail
->bi_next
= bio
;
5691 pending_bios
->tail
= bio
;
5692 if (!pending_bios
->head
)
5693 pending_bios
->head
= bio
;
5694 if (device
->running_pending
)
5697 spin_unlock(&device
->io_lock
);
5700 btrfs_queue_work(root
->fs_info
->submit_workers
,
5704 static int bio_size_ok(struct block_device
*bdev
, struct bio
*bio
,
5707 struct bio_vec
*prev
;
5708 struct request_queue
*q
= bdev_get_queue(bdev
);
5709 unsigned int max_sectors
= queue_max_sectors(q
);
5710 struct bvec_merge_data bvm
= {
5712 .bi_sector
= sector
,
5713 .bi_rw
= bio
->bi_rw
,
5716 if (WARN_ON(bio
->bi_vcnt
== 0))
5719 prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
5720 if (bio_sectors(bio
) > max_sectors
)
5723 if (!q
->merge_bvec_fn
)
5726 bvm
.bi_size
= bio
->bi_iter
.bi_size
- prev
->bv_len
;
5727 if (q
->merge_bvec_fn(q
, &bvm
, prev
) < prev
->bv_len
)
5732 static void submit_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
5733 struct bio
*bio
, u64 physical
, int dev_nr
,
5736 struct btrfs_device
*dev
= bbio
->stripes
[dev_nr
].dev
;
5738 bio
->bi_private
= bbio
;
5739 btrfs_io_bio(bio
)->stripe_index
= dev_nr
;
5740 bio
->bi_end_io
= btrfs_end_bio
;
5741 bio
->bi_iter
.bi_sector
= physical
>> 9;
5744 struct rcu_string
*name
;
5747 name
= rcu_dereference(dev
->name
);
5748 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5749 "(%s id %llu), size=%u\n", rw
,
5750 (u64
)bio
->bi_iter
.bi_sector
, (u_long
)dev
->bdev
->bd_dev
,
5751 name
->str
, dev
->devid
, bio
->bi_iter
.bi_size
);
5755 bio
->bi_bdev
= dev
->bdev
;
5757 btrfs_bio_counter_inc_noblocked(root
->fs_info
);
5760 btrfs_schedule_bio(root
, dev
, rw
, bio
);
5762 btrfsic_submit_bio(rw
, bio
);
5765 static int breakup_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
5766 struct bio
*first_bio
, struct btrfs_device
*dev
,
5767 int dev_nr
, int rw
, int async
)
5769 struct bio_vec
*bvec
= first_bio
->bi_io_vec
;
5771 int nr_vecs
= bio_get_nr_vecs(dev
->bdev
);
5772 u64 physical
= bbio
->stripes
[dev_nr
].physical
;
5775 bio
= btrfs_bio_alloc(dev
->bdev
, physical
>> 9, nr_vecs
, GFP_NOFS
);
5779 while (bvec
<= (first_bio
->bi_io_vec
+ first_bio
->bi_vcnt
- 1)) {
5780 if (bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
5781 bvec
->bv_offset
) < bvec
->bv_len
) {
5782 u64 len
= bio
->bi_iter
.bi_size
;
5784 atomic_inc(&bbio
->stripes_pending
);
5785 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
,
5793 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
, rw
, async
);
5797 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
5799 atomic_inc(&bbio
->error
);
5800 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
5801 /* Shoud be the original bio. */
5802 WARN_ON(bio
!= bbio
->orig_bio
);
5804 bio
->bi_private
= bbio
->private;
5805 bio
->bi_end_io
= bbio
->end_io
;
5806 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
5807 bio
->bi_iter
.bi_sector
= logical
>> 9;
5809 btrfs_end_bbio(bbio
, bio
, -EIO
);
5813 int btrfs_map_bio(struct btrfs_root
*root
, int rw
, struct bio
*bio
,
5814 int mirror_num
, int async_submit
)
5816 struct btrfs_device
*dev
;
5817 struct bio
*first_bio
= bio
;
5818 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
5824 struct btrfs_bio
*bbio
= NULL
;
5826 length
= bio
->bi_iter
.bi_size
;
5827 map_length
= length
;
5829 btrfs_bio_counter_inc_blocked(root
->fs_info
);
5830 ret
= __btrfs_map_block(root
->fs_info
, rw
, logical
, &map_length
, &bbio
,
5833 btrfs_bio_counter_dec(root
->fs_info
);
5837 total_devs
= bbio
->num_stripes
;
5838 bbio
->orig_bio
= first_bio
;
5839 bbio
->private = first_bio
->bi_private
;
5840 bbio
->end_io
= first_bio
->bi_end_io
;
5841 bbio
->fs_info
= root
->fs_info
;
5842 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
5844 if (bbio
->raid_map
) {
5845 /* In this case, map_length has been set to the length of
5846 a single stripe; not the whole write */
5848 ret
= raid56_parity_write(root
, bio
, bbio
, map_length
);
5850 ret
= raid56_parity_recover(root
, bio
, bbio
, map_length
,
5854 btrfs_bio_counter_dec(root
->fs_info
);
5858 if (map_length
< length
) {
5859 btrfs_crit(root
->fs_info
, "mapping failed logical %llu bio len %llu len %llu",
5860 logical
, length
, map_length
);
5864 for (dev_nr
= 0; dev_nr
< total_devs
; dev_nr
++) {
5865 dev
= bbio
->stripes
[dev_nr
].dev
;
5866 if (!dev
|| !dev
->bdev
|| (rw
& WRITE
&& !dev
->writeable
)) {
5867 bbio_error(bbio
, first_bio
, logical
);
5872 * Check and see if we're ok with this bio based on it's size
5873 * and offset with the given device.
5875 if (!bio_size_ok(dev
->bdev
, first_bio
,
5876 bbio
->stripes
[dev_nr
].physical
>> 9)) {
5877 ret
= breakup_stripe_bio(root
, bbio
, first_bio
, dev
,
5878 dev_nr
, rw
, async_submit
);
5883 if (dev_nr
< total_devs
- 1) {
5884 bio
= btrfs_bio_clone(first_bio
, GFP_NOFS
);
5885 BUG_ON(!bio
); /* -ENOMEM */
5888 bbio
->flags
|= BTRFS_BIO_ORIG_BIO_SUBMITTED
;
5891 submit_stripe_bio(root
, bbio
, bio
,
5892 bbio
->stripes
[dev_nr
].physical
, dev_nr
, rw
,
5895 btrfs_bio_counter_dec(root
->fs_info
);
5899 struct btrfs_device
*btrfs_find_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
5902 struct btrfs_device
*device
;
5903 struct btrfs_fs_devices
*cur_devices
;
5905 cur_devices
= fs_info
->fs_devices
;
5906 while (cur_devices
) {
5908 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
5909 device
= __find_device(&cur_devices
->devices
,
5914 cur_devices
= cur_devices
->seed
;
5919 static struct btrfs_device
*add_missing_dev(struct btrfs_root
*root
,
5920 struct btrfs_fs_devices
*fs_devices
,
5921 u64 devid
, u8
*dev_uuid
)
5923 struct btrfs_device
*device
;
5925 device
= btrfs_alloc_device(NULL
, &devid
, dev_uuid
);
5929 list_add(&device
->dev_list
, &fs_devices
->devices
);
5930 device
->fs_devices
= fs_devices
;
5931 fs_devices
->num_devices
++;
5933 device
->missing
= 1;
5934 fs_devices
->missing_devices
++;
5940 * btrfs_alloc_device - allocate struct btrfs_device
5941 * @fs_info: used only for generating a new devid, can be NULL if
5942 * devid is provided (i.e. @devid != NULL).
5943 * @devid: a pointer to devid for this device. If NULL a new devid
5945 * @uuid: a pointer to UUID for this device. If NULL a new UUID
5948 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5949 * on error. Returned struct is not linked onto any lists and can be
5950 * destroyed with kfree() right away.
5952 struct btrfs_device
*btrfs_alloc_device(struct btrfs_fs_info
*fs_info
,
5956 struct btrfs_device
*dev
;
5959 if (WARN_ON(!devid
&& !fs_info
))
5960 return ERR_PTR(-EINVAL
);
5962 dev
= __alloc_device();
5971 ret
= find_next_devid(fs_info
, &tmp
);
5974 return ERR_PTR(ret
);
5980 memcpy(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
);
5982 generate_random_uuid(dev
->uuid
);
5984 btrfs_init_work(&dev
->work
, btrfs_submit_helper
,
5985 pending_bios_fn
, NULL
, NULL
);
5990 static int read_one_chunk(struct btrfs_root
*root
, struct btrfs_key
*key
,
5991 struct extent_buffer
*leaf
,
5992 struct btrfs_chunk
*chunk
)
5994 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
5995 struct map_lookup
*map
;
5996 struct extent_map
*em
;
6000 u8 uuid
[BTRFS_UUID_SIZE
];
6005 logical
= key
->offset
;
6006 length
= btrfs_chunk_length(leaf
, chunk
);
6008 read_lock(&map_tree
->map_tree
.lock
);
6009 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
6010 read_unlock(&map_tree
->map_tree
.lock
);
6012 /* already mapped? */
6013 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
6014 free_extent_map(em
);
6017 free_extent_map(em
);
6020 em
= alloc_extent_map();
6023 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6024 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
6026 free_extent_map(em
);
6030 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
6031 em
->bdev
= (struct block_device
*)map
;
6032 em
->start
= logical
;
6035 em
->block_start
= 0;
6036 em
->block_len
= em
->len
;
6038 map
->num_stripes
= num_stripes
;
6039 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
6040 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
6041 map
->sector_size
= btrfs_chunk_sector_size(leaf
, chunk
);
6042 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6043 map
->type
= btrfs_chunk_type(leaf
, chunk
);
6044 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6045 for (i
= 0; i
< num_stripes
; i
++) {
6046 map
->stripes
[i
].physical
=
6047 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
6048 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
6049 read_extent_buffer(leaf
, uuid
, (unsigned long)
6050 btrfs_stripe_dev_uuid_nr(chunk
, i
),
6052 map
->stripes
[i
].dev
= btrfs_find_device(root
->fs_info
, devid
,
6054 if (!map
->stripes
[i
].dev
&& !btrfs_test_opt(root
, DEGRADED
)) {
6055 free_extent_map(em
);
6058 if (!map
->stripes
[i
].dev
) {
6059 map
->stripes
[i
].dev
=
6060 add_missing_dev(root
, root
->fs_info
->fs_devices
,
6062 if (!map
->stripes
[i
].dev
) {
6063 free_extent_map(em
);
6067 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
6070 write_lock(&map_tree
->map_tree
.lock
);
6071 ret
= add_extent_mapping(&map_tree
->map_tree
, em
, 0);
6072 write_unlock(&map_tree
->map_tree
.lock
);
6073 BUG_ON(ret
); /* Tree corruption */
6074 free_extent_map(em
);
6079 static void fill_device_from_item(struct extent_buffer
*leaf
,
6080 struct btrfs_dev_item
*dev_item
,
6081 struct btrfs_device
*device
)
6085 device
->devid
= btrfs_device_id(leaf
, dev_item
);
6086 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
6087 device
->total_bytes
= device
->disk_total_bytes
;
6088 device
->commit_total_bytes
= device
->disk_total_bytes
;
6089 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
6090 device
->commit_bytes_used
= device
->bytes_used
;
6091 device
->type
= btrfs_device_type(leaf
, dev_item
);
6092 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
6093 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
6094 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
6095 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
6096 device
->is_tgtdev_for_dev_replace
= 0;
6098 ptr
= btrfs_device_uuid(dev_item
);
6099 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
6102 static struct btrfs_fs_devices
*open_seed_devices(struct btrfs_root
*root
,
6105 struct btrfs_fs_devices
*fs_devices
;
6108 BUG_ON(!mutex_is_locked(&uuid_mutex
));
6110 fs_devices
= root
->fs_info
->fs_devices
->seed
;
6111 while (fs_devices
) {
6112 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
))
6115 fs_devices
= fs_devices
->seed
;
6118 fs_devices
= find_fsid(fsid
);
6120 if (!btrfs_test_opt(root
, DEGRADED
))
6121 return ERR_PTR(-ENOENT
);
6123 fs_devices
= alloc_fs_devices(fsid
);
6124 if (IS_ERR(fs_devices
))
6127 fs_devices
->seeding
= 1;
6128 fs_devices
->opened
= 1;
6132 fs_devices
= clone_fs_devices(fs_devices
);
6133 if (IS_ERR(fs_devices
))
6136 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
6137 root
->fs_info
->bdev_holder
);
6139 free_fs_devices(fs_devices
);
6140 fs_devices
= ERR_PTR(ret
);
6144 if (!fs_devices
->seeding
) {
6145 __btrfs_close_devices(fs_devices
);
6146 free_fs_devices(fs_devices
);
6147 fs_devices
= ERR_PTR(-EINVAL
);
6151 fs_devices
->seed
= root
->fs_info
->fs_devices
->seed
;
6152 root
->fs_info
->fs_devices
->seed
= fs_devices
;
6157 static int read_one_dev(struct btrfs_root
*root
,
6158 struct extent_buffer
*leaf
,
6159 struct btrfs_dev_item
*dev_item
)
6161 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
6162 struct btrfs_device
*device
;
6165 u8 fs_uuid
[BTRFS_UUID_SIZE
];
6166 u8 dev_uuid
[BTRFS_UUID_SIZE
];
6168 devid
= btrfs_device_id(leaf
, dev_item
);
6169 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
6171 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
6174 if (memcmp(fs_uuid
, root
->fs_info
->fsid
, BTRFS_UUID_SIZE
)) {
6175 fs_devices
= open_seed_devices(root
, fs_uuid
);
6176 if (IS_ERR(fs_devices
))
6177 return PTR_ERR(fs_devices
);
6180 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
, fs_uuid
);
6182 if (!btrfs_test_opt(root
, DEGRADED
))
6185 btrfs_warn(root
->fs_info
, "devid %llu missing", devid
);
6186 device
= add_missing_dev(root
, fs_devices
, devid
, dev_uuid
);
6190 if (!device
->bdev
&& !btrfs_test_opt(root
, DEGRADED
))
6193 if(!device
->bdev
&& !device
->missing
) {
6195 * this happens when a device that was properly setup
6196 * in the device info lists suddenly goes bad.
6197 * device->bdev is NULL, and so we have to set
6198 * device->missing to one here
6200 device
->fs_devices
->missing_devices
++;
6201 device
->missing
= 1;
6204 /* Move the device to its own fs_devices */
6205 if (device
->fs_devices
!= fs_devices
) {
6206 ASSERT(device
->missing
);
6208 list_move(&device
->dev_list
, &fs_devices
->devices
);
6209 device
->fs_devices
->num_devices
--;
6210 fs_devices
->num_devices
++;
6212 device
->fs_devices
->missing_devices
--;
6213 fs_devices
->missing_devices
++;
6215 device
->fs_devices
= fs_devices
;
6219 if (device
->fs_devices
!= root
->fs_info
->fs_devices
) {
6220 BUG_ON(device
->writeable
);
6221 if (device
->generation
!=
6222 btrfs_device_generation(leaf
, dev_item
))
6226 fill_device_from_item(leaf
, dev_item
, device
);
6227 device
->in_fs_metadata
= 1;
6228 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
6229 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
6230 spin_lock(&root
->fs_info
->free_chunk_lock
);
6231 root
->fs_info
->free_chunk_space
+= device
->total_bytes
-
6233 spin_unlock(&root
->fs_info
->free_chunk_lock
);
6239 int btrfs_read_sys_array(struct btrfs_root
*root
)
6241 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
6242 struct extent_buffer
*sb
;
6243 struct btrfs_disk_key
*disk_key
;
6244 struct btrfs_chunk
*chunk
;
6246 unsigned long sb_array_offset
;
6252 struct btrfs_key key
;
6254 ASSERT(BTRFS_SUPER_INFO_SIZE
<= root
->nodesize
);
6256 * This will create extent buffer of nodesize, superblock size is
6257 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6258 * overallocate but we can keep it as-is, only the first page is used.
6260 sb
= btrfs_find_create_tree_block(root
, BTRFS_SUPER_INFO_OFFSET
);
6263 btrfs_set_buffer_uptodate(sb
);
6264 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
6266 * The sb extent buffer is artifical and just used to read the system array.
6267 * btrfs_set_buffer_uptodate() call does not properly mark all it's
6268 * pages up-to-date when the page is larger: extent does not cover the
6269 * whole page and consequently check_page_uptodate does not find all
6270 * the page's extents up-to-date (the hole beyond sb),
6271 * write_extent_buffer then triggers a WARN_ON.
6273 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6274 * but sb spans only this function. Add an explicit SetPageUptodate call
6275 * to silence the warning eg. on PowerPC 64.
6277 if (PAGE_CACHE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
6278 SetPageUptodate(sb
->pages
[0]);
6280 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
6281 array_size
= btrfs_super_sys_array_size(super_copy
);
6283 array_ptr
= super_copy
->sys_chunk_array
;
6284 sb_array_offset
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
6287 while (cur_offset
< array_size
) {
6288 disk_key
= (struct btrfs_disk_key
*)array_ptr
;
6289 len
= sizeof(*disk_key
);
6290 if (cur_offset
+ len
> array_size
)
6291 goto out_short_read
;
6293 btrfs_disk_key_to_cpu(&key
, disk_key
);
6296 sb_array_offset
+= len
;
6299 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6300 chunk
= (struct btrfs_chunk
*)sb_array_offset
;
6302 * At least one btrfs_chunk with one stripe must be
6303 * present, exact stripe count check comes afterwards
6305 len
= btrfs_chunk_item_size(1);
6306 if (cur_offset
+ len
> array_size
)
6307 goto out_short_read
;
6309 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
6310 len
= btrfs_chunk_item_size(num_stripes
);
6311 if (cur_offset
+ len
> array_size
)
6312 goto out_short_read
;
6314 ret
= read_one_chunk(root
, &key
, sb
, chunk
);
6322 sb_array_offset
+= len
;
6325 free_extent_buffer(sb
);
6329 printk(KERN_ERR
"BTRFS: sys_array too short to read %u bytes at offset %u\n",
6331 free_extent_buffer(sb
);
6335 int btrfs_read_chunk_tree(struct btrfs_root
*root
)
6337 struct btrfs_path
*path
;
6338 struct extent_buffer
*leaf
;
6339 struct btrfs_key key
;
6340 struct btrfs_key found_key
;
6344 root
= root
->fs_info
->chunk_root
;
6346 path
= btrfs_alloc_path();
6350 mutex_lock(&uuid_mutex
);
6354 * Read all device items, and then all the chunk items. All
6355 * device items are found before any chunk item (their object id
6356 * is smaller than the lowest possible object id for a chunk
6357 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6359 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
6362 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6366 leaf
= path
->nodes
[0];
6367 slot
= path
->slots
[0];
6368 if (slot
>= btrfs_header_nritems(leaf
)) {
6369 ret
= btrfs_next_leaf(root
, path
);
6376 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6377 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
6378 struct btrfs_dev_item
*dev_item
;
6379 dev_item
= btrfs_item_ptr(leaf
, slot
,
6380 struct btrfs_dev_item
);
6381 ret
= read_one_dev(root
, leaf
, dev_item
);
6384 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6385 struct btrfs_chunk
*chunk
;
6386 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
6387 ret
= read_one_chunk(root
, &found_key
, leaf
, chunk
);
6395 unlock_chunks(root
);
6396 mutex_unlock(&uuid_mutex
);
6398 btrfs_free_path(path
);
6402 void btrfs_init_devices_late(struct btrfs_fs_info
*fs_info
)
6404 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6405 struct btrfs_device
*device
;
6407 while (fs_devices
) {
6408 mutex_lock(&fs_devices
->device_list_mutex
);
6409 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
)
6410 device
->dev_root
= fs_info
->dev_root
;
6411 mutex_unlock(&fs_devices
->device_list_mutex
);
6413 fs_devices
= fs_devices
->seed
;
6417 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
6421 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6422 btrfs_dev_stat_reset(dev
, i
);
6425 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
6427 struct btrfs_key key
;
6428 struct btrfs_key found_key
;
6429 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6430 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6431 struct extent_buffer
*eb
;
6434 struct btrfs_device
*device
;
6435 struct btrfs_path
*path
= NULL
;
6438 path
= btrfs_alloc_path();
6444 mutex_lock(&fs_devices
->device_list_mutex
);
6445 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6447 struct btrfs_dev_stats_item
*ptr
;
6450 key
.type
= BTRFS_DEV_STATS_KEY
;
6451 key
.offset
= device
->devid
;
6452 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
6454 __btrfs_reset_dev_stats(device
);
6455 device
->dev_stats_valid
= 1;
6456 btrfs_release_path(path
);
6459 slot
= path
->slots
[0];
6460 eb
= path
->nodes
[0];
6461 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
6462 item_size
= btrfs_item_size_nr(eb
, slot
);
6464 ptr
= btrfs_item_ptr(eb
, slot
,
6465 struct btrfs_dev_stats_item
);
6467 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6468 if (item_size
>= (1 + i
) * sizeof(__le64
))
6469 btrfs_dev_stat_set(device
, i
,
6470 btrfs_dev_stats_value(eb
, ptr
, i
));
6472 btrfs_dev_stat_reset(device
, i
);
6475 device
->dev_stats_valid
= 1;
6476 btrfs_dev_stat_print_on_load(device
);
6477 btrfs_release_path(path
);
6479 mutex_unlock(&fs_devices
->device_list_mutex
);
6482 btrfs_free_path(path
);
6483 return ret
< 0 ? ret
: 0;
6486 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
6487 struct btrfs_root
*dev_root
,
6488 struct btrfs_device
*device
)
6490 struct btrfs_path
*path
;
6491 struct btrfs_key key
;
6492 struct extent_buffer
*eb
;
6493 struct btrfs_dev_stats_item
*ptr
;
6498 key
.type
= BTRFS_DEV_STATS_KEY
;
6499 key
.offset
= device
->devid
;
6501 path
= btrfs_alloc_path();
6503 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
6505 printk_in_rcu(KERN_WARNING
"BTRFS: "
6506 "error %d while searching for dev_stats item for device %s!\n",
6507 ret
, rcu_str_deref(device
->name
));
6512 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
6513 /* need to delete old one and insert a new one */
6514 ret
= btrfs_del_item(trans
, dev_root
, path
);
6516 printk_in_rcu(KERN_WARNING
"BTRFS: "
6517 "delete too small dev_stats item for device %s failed %d!\n",
6518 rcu_str_deref(device
->name
), ret
);
6525 /* need to insert a new item */
6526 btrfs_release_path(path
);
6527 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
6528 &key
, sizeof(*ptr
));
6530 printk_in_rcu(KERN_WARNING
"BTRFS: "
6531 "insert dev_stats item for device %s failed %d!\n",
6532 rcu_str_deref(device
->name
), ret
);
6537 eb
= path
->nodes
[0];
6538 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
6539 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6540 btrfs_set_dev_stats_value(eb
, ptr
, i
,
6541 btrfs_dev_stat_read(device
, i
));
6542 btrfs_mark_buffer_dirty(eb
);
6545 btrfs_free_path(path
);
6550 * called from commit_transaction. Writes all changed device stats to disk.
6552 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
6553 struct btrfs_fs_info
*fs_info
)
6555 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6556 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6557 struct btrfs_device
*device
;
6561 mutex_lock(&fs_devices
->device_list_mutex
);
6562 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6563 if (!device
->dev_stats_valid
|| !btrfs_dev_stats_dirty(device
))
6566 stats_cnt
= atomic_read(&device
->dev_stats_ccnt
);
6567 ret
= update_dev_stat_item(trans
, dev_root
, device
);
6569 atomic_sub(stats_cnt
, &device
->dev_stats_ccnt
);
6571 mutex_unlock(&fs_devices
->device_list_mutex
);
6576 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
6578 btrfs_dev_stat_inc(dev
, index
);
6579 btrfs_dev_stat_print_on_error(dev
);
6582 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
6584 if (!dev
->dev_stats_valid
)
6586 printk_ratelimited_in_rcu(KERN_ERR
"BTRFS: "
6587 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6588 rcu_str_deref(dev
->name
),
6589 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
6590 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
6591 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
6592 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
6593 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
6596 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
6600 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6601 if (btrfs_dev_stat_read(dev
, i
) != 0)
6603 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
6604 return; /* all values == 0, suppress message */
6606 printk_in_rcu(KERN_INFO
"BTRFS: "
6607 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6608 rcu_str_deref(dev
->name
),
6609 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
6610 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
6611 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
6612 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
6613 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
6616 int btrfs_get_dev_stats(struct btrfs_root
*root
,
6617 struct btrfs_ioctl_get_dev_stats
*stats
)
6619 struct btrfs_device
*dev
;
6620 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
6623 mutex_lock(&fs_devices
->device_list_mutex
);
6624 dev
= btrfs_find_device(root
->fs_info
, stats
->devid
, NULL
, NULL
);
6625 mutex_unlock(&fs_devices
->device_list_mutex
);
6628 btrfs_warn(root
->fs_info
, "get dev_stats failed, device not found");
6630 } else if (!dev
->dev_stats_valid
) {
6631 btrfs_warn(root
->fs_info
, "get dev_stats failed, not yet valid");
6633 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
6634 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6635 if (stats
->nr_items
> i
)
6637 btrfs_dev_stat_read_and_reset(dev
, i
);
6639 btrfs_dev_stat_reset(dev
, i
);
6642 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6643 if (stats
->nr_items
> i
)
6644 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
6646 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
6647 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
6651 int btrfs_scratch_superblock(struct btrfs_device
*device
)
6653 struct buffer_head
*bh
;
6654 struct btrfs_super_block
*disk_super
;
6656 bh
= btrfs_read_dev_super(device
->bdev
);
6659 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
6661 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
6662 set_buffer_dirty(bh
);
6663 sync_dirty_buffer(bh
);
6670 * Update the size of all devices, which is used for writing out the
6673 void btrfs_update_commit_device_size(struct btrfs_fs_info
*fs_info
)
6675 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6676 struct btrfs_device
*curr
, *next
;
6678 if (list_empty(&fs_devices
->resized_devices
))
6681 mutex_lock(&fs_devices
->device_list_mutex
);
6682 lock_chunks(fs_info
->dev_root
);
6683 list_for_each_entry_safe(curr
, next
, &fs_devices
->resized_devices
,
6685 list_del_init(&curr
->resized_list
);
6686 curr
->commit_total_bytes
= curr
->disk_total_bytes
;
6688 unlock_chunks(fs_info
->dev_root
);
6689 mutex_unlock(&fs_devices
->device_list_mutex
);
6692 /* Must be invoked during the transaction commit */
6693 void btrfs_update_commit_device_bytes_used(struct btrfs_root
*root
,
6694 struct btrfs_transaction
*transaction
)
6696 struct extent_map
*em
;
6697 struct map_lookup
*map
;
6698 struct btrfs_device
*dev
;
6701 if (list_empty(&transaction
->pending_chunks
))
6704 /* In order to kick the device replace finish process */
6706 list_for_each_entry(em
, &transaction
->pending_chunks
, list
) {
6707 map
= (struct map_lookup
*)em
->bdev
;
6709 for (i
= 0; i
< map
->num_stripes
; i
++) {
6710 dev
= map
->stripes
[i
].dev
;
6711 dev
->commit_bytes_used
= dev
->bytes_used
;
6714 unlock_chunks(root
);