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
;
406 spin_lock(&device
->io_lock
);
407 requeue_list(pending_bios
, pending
, tail
);
408 device
->running_pending
= 1;
410 spin_unlock(&device
->io_lock
);
411 btrfs_queue_work(fs_info
->submit_workers
,
415 /* unplug every 64 requests just for good measure */
416 if (batch_run
% 64 == 0) {
417 blk_finish_plug(&plug
);
418 blk_start_plug(&plug
);
427 spin_lock(&device
->io_lock
);
428 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
430 spin_unlock(&device
->io_lock
);
433 blk_finish_plug(&plug
);
436 static void pending_bios_fn(struct btrfs_work
*work
)
438 struct btrfs_device
*device
;
440 device
= container_of(work
, struct btrfs_device
, work
);
441 run_scheduled_bios(device
);
445 * Add new device to list of registered devices
448 * 1 - first time device is seen
449 * 0 - device already known
452 static noinline
int device_list_add(const char *path
,
453 struct btrfs_super_block
*disk_super
,
454 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
456 struct btrfs_device
*device
;
457 struct btrfs_fs_devices
*fs_devices
;
458 struct rcu_string
*name
;
460 u64 found_transid
= btrfs_super_generation(disk_super
);
462 fs_devices
= find_fsid(disk_super
->fsid
);
464 fs_devices
= alloc_fs_devices(disk_super
->fsid
);
465 if (IS_ERR(fs_devices
))
466 return PTR_ERR(fs_devices
);
468 list_add(&fs_devices
->list
, &fs_uuids
);
472 device
= __find_device(&fs_devices
->devices
, devid
,
473 disk_super
->dev_item
.uuid
);
477 if (fs_devices
->opened
)
480 device
= btrfs_alloc_device(NULL
, &devid
,
481 disk_super
->dev_item
.uuid
);
482 if (IS_ERR(device
)) {
483 /* we can safely leave the fs_devices entry around */
484 return PTR_ERR(device
);
487 name
= rcu_string_strdup(path
, GFP_NOFS
);
492 rcu_assign_pointer(device
->name
, name
);
494 mutex_lock(&fs_devices
->device_list_mutex
);
495 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
496 fs_devices
->num_devices
++;
497 mutex_unlock(&fs_devices
->device_list_mutex
);
500 device
->fs_devices
= fs_devices
;
501 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
503 * When FS is already mounted.
504 * 1. If you are here and if the device->name is NULL that
505 * means this device was missing at time of FS mount.
506 * 2. If you are here and if the device->name is different
507 * from 'path' that means either
508 * a. The same device disappeared and reappeared with
510 * b. The missing-disk-which-was-replaced, has
513 * We must allow 1 and 2a above. But 2b would be a spurious
516 * Further in case of 1 and 2a above, the disk at 'path'
517 * would have missed some transaction when it was away and
518 * in case of 2a the stale bdev has to be updated as well.
519 * 2b must not be allowed at all time.
523 * For now, we do allow update to btrfs_fs_device through the
524 * btrfs dev scan cli after FS has been mounted. We're still
525 * tracking a problem where systems fail mount by subvolume id
526 * when we reject replacement on a mounted FS.
528 if (!fs_devices
->opened
&& found_transid
< device
->generation
) {
530 * That is if the FS is _not_ mounted and if you
531 * are here, that means there is more than one
532 * disk with same uuid and devid.We keep the one
533 * with larger generation number or the last-in if
534 * generation are equal.
539 name
= rcu_string_strdup(path
, GFP_NOFS
);
542 rcu_string_free(device
->name
);
543 rcu_assign_pointer(device
->name
, name
);
544 if (device
->missing
) {
545 fs_devices
->missing_devices
--;
551 * Unmount does not free the btrfs_device struct but would zero
552 * generation along with most of the other members. So just update
553 * it back. We need it to pick the disk with largest generation
556 if (!fs_devices
->opened
)
557 device
->generation
= found_transid
;
559 *fs_devices_ret
= fs_devices
;
564 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
566 struct btrfs_fs_devices
*fs_devices
;
567 struct btrfs_device
*device
;
568 struct btrfs_device
*orig_dev
;
570 fs_devices
= alloc_fs_devices(orig
->fsid
);
571 if (IS_ERR(fs_devices
))
574 mutex_lock(&orig
->device_list_mutex
);
575 fs_devices
->total_devices
= orig
->total_devices
;
577 /* We have held the volume lock, it is safe to get the devices. */
578 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
579 struct rcu_string
*name
;
581 device
= btrfs_alloc_device(NULL
, &orig_dev
->devid
,
587 * This is ok to do without rcu read locked because we hold the
588 * uuid mutex so nothing we touch in here is going to disappear.
590 if (orig_dev
->name
) {
591 name
= rcu_string_strdup(orig_dev
->name
->str
, GFP_NOFS
);
596 rcu_assign_pointer(device
->name
, name
);
599 list_add(&device
->dev_list
, &fs_devices
->devices
);
600 device
->fs_devices
= fs_devices
;
601 fs_devices
->num_devices
++;
603 mutex_unlock(&orig
->device_list_mutex
);
606 mutex_unlock(&orig
->device_list_mutex
);
607 free_fs_devices(fs_devices
);
608 return ERR_PTR(-ENOMEM
);
611 void btrfs_close_extra_devices(struct btrfs_fs_devices
*fs_devices
, int step
)
613 struct btrfs_device
*device
, *next
;
614 struct btrfs_device
*latest_dev
= NULL
;
616 mutex_lock(&uuid_mutex
);
618 /* This is the initialized path, it is safe to release the devices. */
619 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
620 if (device
->in_fs_metadata
) {
621 if (!device
->is_tgtdev_for_dev_replace
&&
623 device
->generation
> latest_dev
->generation
)) {
629 if (device
->devid
== BTRFS_DEV_REPLACE_DEVID
) {
631 * In the first step, keep the device which has
632 * the correct fsid and the devid that is used
633 * for the dev_replace procedure.
634 * In the second step, the dev_replace state is
635 * read from the device tree and it is known
636 * whether the procedure is really active or
637 * not, which means whether this device is
638 * used or whether it should be removed.
640 if (step
== 0 || device
->is_tgtdev_for_dev_replace
) {
645 blkdev_put(device
->bdev
, device
->mode
);
647 fs_devices
->open_devices
--;
649 if (device
->writeable
) {
650 list_del_init(&device
->dev_alloc_list
);
651 device
->writeable
= 0;
652 if (!device
->is_tgtdev_for_dev_replace
)
653 fs_devices
->rw_devices
--;
655 list_del_init(&device
->dev_list
);
656 fs_devices
->num_devices
--;
657 rcu_string_free(device
->name
);
661 if (fs_devices
->seed
) {
662 fs_devices
= fs_devices
->seed
;
666 fs_devices
->latest_bdev
= latest_dev
->bdev
;
668 mutex_unlock(&uuid_mutex
);
671 static void __free_device(struct work_struct
*work
)
673 struct btrfs_device
*device
;
675 device
= container_of(work
, struct btrfs_device
, rcu_work
);
678 blkdev_put(device
->bdev
, device
->mode
);
680 rcu_string_free(device
->name
);
684 static void free_device(struct rcu_head
*head
)
686 struct btrfs_device
*device
;
688 device
= container_of(head
, struct btrfs_device
, rcu
);
690 INIT_WORK(&device
->rcu_work
, __free_device
);
691 schedule_work(&device
->rcu_work
);
694 static int __btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
696 struct btrfs_device
*device
;
698 if (--fs_devices
->opened
> 0)
701 mutex_lock(&fs_devices
->device_list_mutex
);
702 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
703 struct btrfs_device
*new_device
;
704 struct rcu_string
*name
;
707 fs_devices
->open_devices
--;
709 if (device
->writeable
&&
710 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
711 list_del_init(&device
->dev_alloc_list
);
712 fs_devices
->rw_devices
--;
716 fs_devices
->missing_devices
--;
718 new_device
= btrfs_alloc_device(NULL
, &device
->devid
,
720 BUG_ON(IS_ERR(new_device
)); /* -ENOMEM */
722 /* Safe because we are under uuid_mutex */
724 name
= rcu_string_strdup(device
->name
->str
, GFP_NOFS
);
725 BUG_ON(!name
); /* -ENOMEM */
726 rcu_assign_pointer(new_device
->name
, name
);
729 list_replace_rcu(&device
->dev_list
, &new_device
->dev_list
);
730 new_device
->fs_devices
= device
->fs_devices
;
732 call_rcu(&device
->rcu
, free_device
);
734 mutex_unlock(&fs_devices
->device_list_mutex
);
736 WARN_ON(fs_devices
->open_devices
);
737 WARN_ON(fs_devices
->rw_devices
);
738 fs_devices
->opened
= 0;
739 fs_devices
->seeding
= 0;
744 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
746 struct btrfs_fs_devices
*seed_devices
= NULL
;
749 mutex_lock(&uuid_mutex
);
750 ret
= __btrfs_close_devices(fs_devices
);
751 if (!fs_devices
->opened
) {
752 seed_devices
= fs_devices
->seed
;
753 fs_devices
->seed
= NULL
;
755 mutex_unlock(&uuid_mutex
);
757 while (seed_devices
) {
758 fs_devices
= seed_devices
;
759 seed_devices
= fs_devices
->seed
;
760 __btrfs_close_devices(fs_devices
);
761 free_fs_devices(fs_devices
);
764 * Wait for rcu kworkers under __btrfs_close_devices
765 * to finish all blkdev_puts so device is really
766 * free when umount is done.
772 static int __btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
773 fmode_t flags
, void *holder
)
775 struct request_queue
*q
;
776 struct block_device
*bdev
;
777 struct list_head
*head
= &fs_devices
->devices
;
778 struct btrfs_device
*device
;
779 struct btrfs_device
*latest_dev
= NULL
;
780 struct buffer_head
*bh
;
781 struct btrfs_super_block
*disk_super
;
788 list_for_each_entry(device
, head
, dev_list
) {
794 /* Just open everything we can; ignore failures here */
795 if (btrfs_get_bdev_and_sb(device
->name
->str
, flags
, holder
, 1,
799 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
800 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
801 if (devid
!= device
->devid
)
804 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
,
808 device
->generation
= btrfs_super_generation(disk_super
);
810 device
->generation
> latest_dev
->generation
)
813 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
814 device
->writeable
= 0;
816 device
->writeable
= !bdev_read_only(bdev
);
820 q
= bdev_get_queue(bdev
);
821 if (blk_queue_discard(q
))
822 device
->can_discard
= 1;
825 device
->in_fs_metadata
= 0;
826 device
->mode
= flags
;
828 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
829 fs_devices
->rotating
= 1;
831 fs_devices
->open_devices
++;
832 if (device
->writeable
&&
833 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
834 fs_devices
->rw_devices
++;
835 list_add(&device
->dev_alloc_list
,
836 &fs_devices
->alloc_list
);
843 blkdev_put(bdev
, flags
);
846 if (fs_devices
->open_devices
== 0) {
850 fs_devices
->seeding
= seeding
;
851 fs_devices
->opened
= 1;
852 fs_devices
->latest_bdev
= latest_dev
->bdev
;
853 fs_devices
->total_rw_bytes
= 0;
858 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
859 fmode_t flags
, void *holder
)
863 mutex_lock(&uuid_mutex
);
864 if (fs_devices
->opened
) {
865 fs_devices
->opened
++;
868 ret
= __btrfs_open_devices(fs_devices
, flags
, holder
);
870 mutex_unlock(&uuid_mutex
);
875 * Look for a btrfs signature on a device. This may be called out of the mount path
876 * and we are not allowed to call set_blocksize during the scan. The superblock
877 * is read via pagecache
879 int btrfs_scan_one_device(const char *path
, fmode_t flags
, void *holder
,
880 struct btrfs_fs_devices
**fs_devices_ret
)
882 struct btrfs_super_block
*disk_super
;
883 struct block_device
*bdev
;
894 * we would like to check all the supers, but that would make
895 * a btrfs mount succeed after a mkfs from a different FS.
896 * So, we need to add a special mount option to scan for
897 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
899 bytenr
= btrfs_sb_offset(0);
901 mutex_lock(&uuid_mutex
);
903 bdev
= blkdev_get_by_path(path
, flags
, holder
);
910 /* make sure our super fits in the device */
911 if (bytenr
+ PAGE_CACHE_SIZE
>= i_size_read(bdev
->bd_inode
))
914 /* make sure our super fits in the page */
915 if (sizeof(*disk_super
) > PAGE_CACHE_SIZE
)
918 /* make sure our super doesn't straddle pages on disk */
919 index
= bytenr
>> PAGE_CACHE_SHIFT
;
920 if ((bytenr
+ sizeof(*disk_super
) - 1) >> PAGE_CACHE_SHIFT
!= index
)
923 /* pull in the page with our super */
924 page
= read_cache_page_gfp(bdev
->bd_inode
->i_mapping
,
927 if (IS_ERR_OR_NULL(page
))
932 /* align our pointer to the offset of the super block */
933 disk_super
= p
+ (bytenr
& ~PAGE_CACHE_MASK
);
935 if (btrfs_super_bytenr(disk_super
) != bytenr
||
936 btrfs_super_magic(disk_super
) != BTRFS_MAGIC
)
939 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
940 transid
= btrfs_super_generation(disk_super
);
941 total_devices
= btrfs_super_num_devices(disk_super
);
943 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
945 if (disk_super
->label
[0]) {
946 if (disk_super
->label
[BTRFS_LABEL_SIZE
- 1])
947 disk_super
->label
[BTRFS_LABEL_SIZE
- 1] = '\0';
948 printk(KERN_INFO
"BTRFS: device label %s ", disk_super
->label
);
950 printk(KERN_INFO
"BTRFS: device fsid %pU ", disk_super
->fsid
);
953 printk(KERN_CONT
"devid %llu transid %llu %s\n", devid
, transid
, path
);
956 if (!ret
&& fs_devices_ret
)
957 (*fs_devices_ret
)->total_devices
= total_devices
;
961 page_cache_release(page
);
964 blkdev_put(bdev
, flags
);
966 mutex_unlock(&uuid_mutex
);
970 /* helper to account the used device space in the range */
971 int btrfs_account_dev_extents_size(struct btrfs_device
*device
, u64 start
,
972 u64 end
, u64
*length
)
974 struct btrfs_key key
;
975 struct btrfs_root
*root
= device
->dev_root
;
976 struct btrfs_dev_extent
*dev_extent
;
977 struct btrfs_path
*path
;
981 struct extent_buffer
*l
;
985 if (start
>= device
->total_bytes
|| device
->is_tgtdev_for_dev_replace
)
988 path
= btrfs_alloc_path();
993 key
.objectid
= device
->devid
;
995 key
.type
= BTRFS_DEV_EXTENT_KEY
;
997 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1001 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1008 slot
= path
->slots
[0];
1009 if (slot
>= btrfs_header_nritems(l
)) {
1010 ret
= btrfs_next_leaf(root
, path
);
1018 btrfs_item_key_to_cpu(l
, &key
, slot
);
1020 if (key
.objectid
< device
->devid
)
1023 if (key
.objectid
> device
->devid
)
1026 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1029 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1030 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1032 if (key
.offset
<= start
&& extent_end
> end
) {
1033 *length
= end
- start
+ 1;
1035 } else if (key
.offset
<= start
&& extent_end
> start
)
1036 *length
+= extent_end
- start
;
1037 else if (key
.offset
> start
&& extent_end
<= end
)
1038 *length
+= extent_end
- key
.offset
;
1039 else if (key
.offset
> start
&& key
.offset
<= end
) {
1040 *length
+= end
- key
.offset
+ 1;
1042 } else if (key
.offset
> end
)
1050 btrfs_free_path(path
);
1054 static int contains_pending_extent(struct btrfs_trans_handle
*trans
,
1055 struct btrfs_device
*device
,
1056 u64
*start
, u64 len
)
1058 struct extent_map
*em
;
1059 struct list_head
*search_list
= &trans
->transaction
->pending_chunks
;
1063 list_for_each_entry(em
, search_list
, list
) {
1064 struct map_lookup
*map
;
1067 map
= (struct map_lookup
*)em
->bdev
;
1068 for (i
= 0; i
< map
->num_stripes
; i
++) {
1069 if (map
->stripes
[i
].dev
!= device
)
1071 if (map
->stripes
[i
].physical
>= *start
+ len
||
1072 map
->stripes
[i
].physical
+ em
->orig_block_len
<=
1075 *start
= map
->stripes
[i
].physical
+
1080 if (search_list
== &trans
->transaction
->pending_chunks
) {
1081 search_list
= &trans
->root
->fs_info
->pinned_chunks
;
1090 * find_free_dev_extent - find free space in the specified device
1091 * @device: the device which we search the free space in
1092 * @num_bytes: the size of the free space that we need
1093 * @start: store the start of the free space.
1094 * @len: the size of the free space. that we find, or the size of the max
1095 * free space if we don't find suitable free space
1097 * this uses a pretty simple search, the expectation is that it is
1098 * called very infrequently and that a given device has a small number
1101 * @start is used to store the start of the free space if we find. But if we
1102 * don't find suitable free space, it will be used to store the start position
1103 * of the max free space.
1105 * @len is used to store the size of the free space that we find.
1106 * But if we don't find suitable free space, it is used to store the size of
1107 * the max free space.
1109 int find_free_dev_extent(struct btrfs_trans_handle
*trans
,
1110 struct btrfs_device
*device
, u64 num_bytes
,
1111 u64
*start
, u64
*len
)
1113 struct btrfs_key key
;
1114 struct btrfs_root
*root
= device
->dev_root
;
1115 struct btrfs_dev_extent
*dev_extent
;
1116 struct btrfs_path
*path
;
1122 u64 search_end
= device
->total_bytes
;
1125 struct extent_buffer
*l
;
1127 /* FIXME use last free of some kind */
1129 /* we don't want to overwrite the superblock on the drive,
1130 * so we make sure to start at an offset of at least 1MB
1132 search_start
= max(root
->fs_info
->alloc_start
, 1024ull * 1024);
1134 path
= btrfs_alloc_path();
1138 max_hole_start
= search_start
;
1142 if (search_start
>= search_end
|| device
->is_tgtdev_for_dev_replace
) {
1148 path
->search_commit_root
= 1;
1149 path
->skip_locking
= 1;
1151 key
.objectid
= device
->devid
;
1152 key
.offset
= search_start
;
1153 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1155 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1159 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1166 slot
= path
->slots
[0];
1167 if (slot
>= btrfs_header_nritems(l
)) {
1168 ret
= btrfs_next_leaf(root
, path
);
1176 btrfs_item_key_to_cpu(l
, &key
, slot
);
1178 if (key
.objectid
< device
->devid
)
1181 if (key
.objectid
> device
->devid
)
1184 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1187 if (key
.offset
> search_start
) {
1188 hole_size
= key
.offset
- search_start
;
1191 * Have to check before we set max_hole_start, otherwise
1192 * we could end up sending back this offset anyway.
1194 if (contains_pending_extent(trans
, device
,
1199 if (hole_size
> max_hole_size
) {
1200 max_hole_start
= search_start
;
1201 max_hole_size
= hole_size
;
1205 * If this free space is greater than which we need,
1206 * it must be the max free space that we have found
1207 * until now, so max_hole_start must point to the start
1208 * of this free space and the length of this free space
1209 * is stored in max_hole_size. Thus, we return
1210 * max_hole_start and max_hole_size and go back to the
1213 if (hole_size
>= num_bytes
) {
1219 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1220 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1222 if (extent_end
> search_start
)
1223 search_start
= extent_end
;
1230 * At this point, search_start should be the end of
1231 * allocated dev extents, and when shrinking the device,
1232 * search_end may be smaller than search_start.
1234 if (search_end
> search_start
)
1235 hole_size
= search_end
- search_start
;
1237 if (hole_size
> max_hole_size
) {
1238 max_hole_start
= search_start
;
1239 max_hole_size
= hole_size
;
1242 if (contains_pending_extent(trans
, device
, &search_start
, hole_size
)) {
1243 btrfs_release_path(path
);
1248 if (hole_size
< num_bytes
)
1254 btrfs_free_path(path
);
1255 *start
= max_hole_start
;
1257 *len
= max_hole_size
;
1261 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1262 struct btrfs_device
*device
,
1263 u64 start
, u64
*dev_extent_len
)
1266 struct btrfs_path
*path
;
1267 struct btrfs_root
*root
= device
->dev_root
;
1268 struct btrfs_key key
;
1269 struct btrfs_key found_key
;
1270 struct extent_buffer
*leaf
= NULL
;
1271 struct btrfs_dev_extent
*extent
= NULL
;
1273 path
= btrfs_alloc_path();
1277 key
.objectid
= device
->devid
;
1279 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1281 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1283 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1284 BTRFS_DEV_EXTENT_KEY
);
1287 leaf
= path
->nodes
[0];
1288 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1289 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1290 struct btrfs_dev_extent
);
1291 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1292 btrfs_dev_extent_length(leaf
, extent
) < start
);
1294 btrfs_release_path(path
);
1296 } else if (ret
== 0) {
1297 leaf
= path
->nodes
[0];
1298 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1299 struct btrfs_dev_extent
);
1301 btrfs_error(root
->fs_info
, ret
, "Slot search failed");
1305 *dev_extent_len
= btrfs_dev_extent_length(leaf
, extent
);
1307 ret
= btrfs_del_item(trans
, root
, path
);
1309 btrfs_error(root
->fs_info
, ret
,
1310 "Failed to remove dev extent item");
1312 trans
->transaction
->have_free_bgs
= 1;
1315 btrfs_free_path(path
);
1319 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1320 struct btrfs_device
*device
,
1321 u64 chunk_tree
, u64 chunk_objectid
,
1322 u64 chunk_offset
, u64 start
, u64 num_bytes
)
1325 struct btrfs_path
*path
;
1326 struct btrfs_root
*root
= device
->dev_root
;
1327 struct btrfs_dev_extent
*extent
;
1328 struct extent_buffer
*leaf
;
1329 struct btrfs_key key
;
1331 WARN_ON(!device
->in_fs_metadata
);
1332 WARN_ON(device
->is_tgtdev_for_dev_replace
);
1333 path
= btrfs_alloc_path();
1337 key
.objectid
= device
->devid
;
1339 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1340 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1345 leaf
= path
->nodes
[0];
1346 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1347 struct btrfs_dev_extent
);
1348 btrfs_set_dev_extent_chunk_tree(leaf
, extent
, chunk_tree
);
1349 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
, chunk_objectid
);
1350 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1352 write_extent_buffer(leaf
, root
->fs_info
->chunk_tree_uuid
,
1353 btrfs_dev_extent_chunk_tree_uuid(extent
), BTRFS_UUID_SIZE
);
1355 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1356 btrfs_mark_buffer_dirty(leaf
);
1358 btrfs_free_path(path
);
1362 static u64
find_next_chunk(struct btrfs_fs_info
*fs_info
)
1364 struct extent_map_tree
*em_tree
;
1365 struct extent_map
*em
;
1369 em_tree
= &fs_info
->mapping_tree
.map_tree
;
1370 read_lock(&em_tree
->lock
);
1371 n
= rb_last(&em_tree
->map
);
1373 em
= rb_entry(n
, struct extent_map
, rb_node
);
1374 ret
= em
->start
+ em
->len
;
1376 read_unlock(&em_tree
->lock
);
1381 static noinline
int find_next_devid(struct btrfs_fs_info
*fs_info
,
1385 struct btrfs_key key
;
1386 struct btrfs_key found_key
;
1387 struct btrfs_path
*path
;
1389 path
= btrfs_alloc_path();
1393 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1394 key
.type
= BTRFS_DEV_ITEM_KEY
;
1395 key
.offset
= (u64
)-1;
1397 ret
= btrfs_search_slot(NULL
, fs_info
->chunk_root
, &key
, path
, 0, 0);
1401 BUG_ON(ret
== 0); /* Corruption */
1403 ret
= btrfs_previous_item(fs_info
->chunk_root
, path
,
1404 BTRFS_DEV_ITEMS_OBJECTID
,
1405 BTRFS_DEV_ITEM_KEY
);
1409 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1411 *devid_ret
= found_key
.offset
+ 1;
1415 btrfs_free_path(path
);
1420 * the device information is stored in the chunk root
1421 * the btrfs_device struct should be fully filled in
1423 static int btrfs_add_device(struct btrfs_trans_handle
*trans
,
1424 struct btrfs_root
*root
,
1425 struct btrfs_device
*device
)
1428 struct btrfs_path
*path
;
1429 struct btrfs_dev_item
*dev_item
;
1430 struct extent_buffer
*leaf
;
1431 struct btrfs_key key
;
1434 root
= root
->fs_info
->chunk_root
;
1436 path
= btrfs_alloc_path();
1440 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1441 key
.type
= BTRFS_DEV_ITEM_KEY
;
1442 key
.offset
= device
->devid
;
1444 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1449 leaf
= path
->nodes
[0];
1450 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1452 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1453 btrfs_set_device_generation(leaf
, dev_item
, 0);
1454 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1455 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1456 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1457 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1458 btrfs_set_device_total_bytes(leaf
, dev_item
,
1459 btrfs_device_get_disk_total_bytes(device
));
1460 btrfs_set_device_bytes_used(leaf
, dev_item
,
1461 btrfs_device_get_bytes_used(device
));
1462 btrfs_set_device_group(leaf
, dev_item
, 0);
1463 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1464 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1465 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1467 ptr
= btrfs_device_uuid(dev_item
);
1468 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1469 ptr
= btrfs_device_fsid(dev_item
);
1470 write_extent_buffer(leaf
, root
->fs_info
->fsid
, ptr
, BTRFS_UUID_SIZE
);
1471 btrfs_mark_buffer_dirty(leaf
);
1475 btrfs_free_path(path
);
1480 * Function to update ctime/mtime for a given device path.
1481 * Mainly used for ctime/mtime based probe like libblkid.
1483 static void update_dev_time(char *path_name
)
1487 filp
= filp_open(path_name
, O_RDWR
, 0);
1490 file_update_time(filp
);
1491 filp_close(filp
, NULL
);
1495 static int btrfs_rm_dev_item(struct btrfs_root
*root
,
1496 struct btrfs_device
*device
)
1499 struct btrfs_path
*path
;
1500 struct btrfs_key key
;
1501 struct btrfs_trans_handle
*trans
;
1503 root
= root
->fs_info
->chunk_root
;
1505 path
= btrfs_alloc_path();
1509 trans
= btrfs_start_transaction(root
, 0);
1510 if (IS_ERR(trans
)) {
1511 btrfs_free_path(path
);
1512 return PTR_ERR(trans
);
1514 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1515 key
.type
= BTRFS_DEV_ITEM_KEY
;
1516 key
.offset
= device
->devid
;
1518 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1527 ret
= btrfs_del_item(trans
, root
, path
);
1531 btrfs_free_path(path
);
1532 btrfs_commit_transaction(trans
, root
);
1536 int btrfs_rm_device(struct btrfs_root
*root
, char *device_path
)
1538 struct btrfs_device
*device
;
1539 struct btrfs_device
*next_device
;
1540 struct block_device
*bdev
;
1541 struct buffer_head
*bh
= NULL
;
1542 struct btrfs_super_block
*disk_super
;
1543 struct btrfs_fs_devices
*cur_devices
;
1550 bool clear_super
= false;
1552 mutex_lock(&uuid_mutex
);
1555 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
1557 all_avail
= root
->fs_info
->avail_data_alloc_bits
|
1558 root
->fs_info
->avail_system_alloc_bits
|
1559 root
->fs_info
->avail_metadata_alloc_bits
;
1560 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
1562 num_devices
= root
->fs_info
->fs_devices
->num_devices
;
1563 btrfs_dev_replace_lock(&root
->fs_info
->dev_replace
);
1564 if (btrfs_dev_replace_is_ongoing(&root
->fs_info
->dev_replace
)) {
1565 WARN_ON(num_devices
< 1);
1568 btrfs_dev_replace_unlock(&root
->fs_info
->dev_replace
);
1570 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID10
) && num_devices
<= 4) {
1571 ret
= BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET
;
1575 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID1
) && num_devices
<= 2) {
1576 ret
= BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET
;
1580 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID5
) &&
1581 root
->fs_info
->fs_devices
->rw_devices
<= 2) {
1582 ret
= BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET
;
1585 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID6
) &&
1586 root
->fs_info
->fs_devices
->rw_devices
<= 3) {
1587 ret
= BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET
;
1591 if (strcmp(device_path
, "missing") == 0) {
1592 struct list_head
*devices
;
1593 struct btrfs_device
*tmp
;
1596 devices
= &root
->fs_info
->fs_devices
->devices
;
1598 * It is safe to read the devices since the volume_mutex
1601 list_for_each_entry(tmp
, devices
, dev_list
) {
1602 if (tmp
->in_fs_metadata
&&
1603 !tmp
->is_tgtdev_for_dev_replace
&&
1613 ret
= BTRFS_ERROR_DEV_MISSING_NOT_FOUND
;
1617 ret
= btrfs_get_bdev_and_sb(device_path
,
1618 FMODE_WRITE
| FMODE_EXCL
,
1619 root
->fs_info
->bdev_holder
, 0,
1623 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1624 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1625 dev_uuid
= disk_super
->dev_item
.uuid
;
1626 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1634 if (device
->is_tgtdev_for_dev_replace
) {
1635 ret
= BTRFS_ERROR_DEV_TGT_REPLACE
;
1639 if (device
->writeable
&& root
->fs_info
->fs_devices
->rw_devices
== 1) {
1640 ret
= BTRFS_ERROR_DEV_ONLY_WRITABLE
;
1644 if (device
->writeable
) {
1646 list_del_init(&device
->dev_alloc_list
);
1647 device
->fs_devices
->rw_devices
--;
1648 unlock_chunks(root
);
1652 mutex_unlock(&uuid_mutex
);
1653 ret
= btrfs_shrink_device(device
, 0);
1654 mutex_lock(&uuid_mutex
);
1659 * TODO: the superblock still includes this device in its num_devices
1660 * counter although write_all_supers() is not locked out. This
1661 * could give a filesystem state which requires a degraded mount.
1663 ret
= btrfs_rm_dev_item(root
->fs_info
->chunk_root
, device
);
1667 device
->in_fs_metadata
= 0;
1668 btrfs_scrub_cancel_dev(root
->fs_info
, device
);
1671 * the device list mutex makes sure that we don't change
1672 * the device list while someone else is writing out all
1673 * the device supers. Whoever is writing all supers, should
1674 * lock the device list mutex before getting the number of
1675 * devices in the super block (super_copy). Conversely,
1676 * whoever updates the number of devices in the super block
1677 * (super_copy) should hold the device list mutex.
1680 cur_devices
= device
->fs_devices
;
1681 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1682 list_del_rcu(&device
->dev_list
);
1684 device
->fs_devices
->num_devices
--;
1685 device
->fs_devices
->total_devices
--;
1687 if (device
->missing
)
1688 device
->fs_devices
->missing_devices
--;
1690 next_device
= list_entry(root
->fs_info
->fs_devices
->devices
.next
,
1691 struct btrfs_device
, dev_list
);
1692 if (device
->bdev
== root
->fs_info
->sb
->s_bdev
)
1693 root
->fs_info
->sb
->s_bdev
= next_device
->bdev
;
1694 if (device
->bdev
== root
->fs_info
->fs_devices
->latest_bdev
)
1695 root
->fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1698 device
->fs_devices
->open_devices
--;
1699 /* remove sysfs entry */
1700 btrfs_kobj_rm_device(root
->fs_info
, device
);
1703 call_rcu(&device
->rcu
, free_device
);
1705 num_devices
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
1706 btrfs_set_super_num_devices(root
->fs_info
->super_copy
, num_devices
);
1707 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1709 if (cur_devices
->open_devices
== 0) {
1710 struct btrfs_fs_devices
*fs_devices
;
1711 fs_devices
= root
->fs_info
->fs_devices
;
1712 while (fs_devices
) {
1713 if (fs_devices
->seed
== cur_devices
) {
1714 fs_devices
->seed
= cur_devices
->seed
;
1717 fs_devices
= fs_devices
->seed
;
1719 cur_devices
->seed
= NULL
;
1720 __btrfs_close_devices(cur_devices
);
1721 free_fs_devices(cur_devices
);
1724 root
->fs_info
->num_tolerated_disk_barrier_failures
=
1725 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
1728 * at this point, the device is zero sized. We want to
1729 * remove it from the devices list and zero out the old super
1731 if (clear_super
&& disk_super
) {
1735 /* make sure this device isn't detected as part of
1738 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
1739 set_buffer_dirty(bh
);
1740 sync_dirty_buffer(bh
);
1742 /* clear the mirror copies of super block on the disk
1743 * being removed, 0th copy is been taken care above and
1744 * the below would take of the rest
1746 for (i
= 1; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
1747 bytenr
= btrfs_sb_offset(i
);
1748 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
1749 i_size_read(bdev
->bd_inode
))
1753 bh
= __bread(bdev
, bytenr
/ 4096,
1754 BTRFS_SUPER_INFO_SIZE
);
1758 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1760 if (btrfs_super_bytenr(disk_super
) != bytenr
||
1761 btrfs_super_magic(disk_super
) != BTRFS_MAGIC
) {
1764 memset(&disk_super
->magic
, 0,
1765 sizeof(disk_super
->magic
));
1766 set_buffer_dirty(bh
);
1767 sync_dirty_buffer(bh
);
1774 /* Notify udev that device has changed */
1775 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
1777 /* Update ctime/mtime for device path for libblkid */
1778 update_dev_time(device_path
);
1784 blkdev_put(bdev
, FMODE_READ
| FMODE_EXCL
);
1786 mutex_unlock(&uuid_mutex
);
1789 if (device
->writeable
) {
1791 list_add(&device
->dev_alloc_list
,
1792 &root
->fs_info
->fs_devices
->alloc_list
);
1793 device
->fs_devices
->rw_devices
++;
1794 unlock_chunks(root
);
1799 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info
*fs_info
,
1800 struct btrfs_device
*srcdev
)
1802 struct btrfs_fs_devices
*fs_devices
;
1804 WARN_ON(!mutex_is_locked(&fs_info
->fs_devices
->device_list_mutex
));
1807 * in case of fs with no seed, srcdev->fs_devices will point
1808 * to fs_devices of fs_info. However when the dev being replaced is
1809 * a seed dev it will point to the seed's local fs_devices. In short
1810 * srcdev will have its correct fs_devices in both the cases.
1812 fs_devices
= srcdev
->fs_devices
;
1814 list_del_rcu(&srcdev
->dev_list
);
1815 list_del_rcu(&srcdev
->dev_alloc_list
);
1816 fs_devices
->num_devices
--;
1817 if (srcdev
->missing
)
1818 fs_devices
->missing_devices
--;
1820 if (srcdev
->writeable
) {
1821 fs_devices
->rw_devices
--;
1822 /* zero out the old super if it is writable */
1823 btrfs_scratch_superblock(srcdev
);
1827 fs_devices
->open_devices
--;
1830 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info
*fs_info
,
1831 struct btrfs_device
*srcdev
)
1833 struct btrfs_fs_devices
*fs_devices
= srcdev
->fs_devices
;
1835 call_rcu(&srcdev
->rcu
, free_device
);
1838 * unless fs_devices is seed fs, num_devices shouldn't go
1841 BUG_ON(!fs_devices
->num_devices
&& !fs_devices
->seeding
);
1843 /* if this is no devs we rather delete the fs_devices */
1844 if (!fs_devices
->num_devices
) {
1845 struct btrfs_fs_devices
*tmp_fs_devices
;
1847 tmp_fs_devices
= fs_info
->fs_devices
;
1848 while (tmp_fs_devices
) {
1849 if (tmp_fs_devices
->seed
== fs_devices
) {
1850 tmp_fs_devices
->seed
= fs_devices
->seed
;
1853 tmp_fs_devices
= tmp_fs_devices
->seed
;
1855 fs_devices
->seed
= NULL
;
1856 __btrfs_close_devices(fs_devices
);
1857 free_fs_devices(fs_devices
);
1861 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
1862 struct btrfs_device
*tgtdev
)
1864 struct btrfs_device
*next_device
;
1866 mutex_lock(&uuid_mutex
);
1868 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
1870 btrfs_scratch_superblock(tgtdev
);
1871 fs_info
->fs_devices
->open_devices
--;
1873 fs_info
->fs_devices
->num_devices
--;
1875 next_device
= list_entry(fs_info
->fs_devices
->devices
.next
,
1876 struct btrfs_device
, dev_list
);
1877 if (tgtdev
->bdev
== fs_info
->sb
->s_bdev
)
1878 fs_info
->sb
->s_bdev
= next_device
->bdev
;
1879 if (tgtdev
->bdev
== fs_info
->fs_devices
->latest_bdev
)
1880 fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1881 list_del_rcu(&tgtdev
->dev_list
);
1883 call_rcu(&tgtdev
->rcu
, free_device
);
1885 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
1886 mutex_unlock(&uuid_mutex
);
1889 static int btrfs_find_device_by_path(struct btrfs_root
*root
, char *device_path
,
1890 struct btrfs_device
**device
)
1893 struct btrfs_super_block
*disk_super
;
1896 struct block_device
*bdev
;
1897 struct buffer_head
*bh
;
1900 ret
= btrfs_get_bdev_and_sb(device_path
, FMODE_READ
,
1901 root
->fs_info
->bdev_holder
, 0, &bdev
, &bh
);
1904 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1905 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1906 dev_uuid
= disk_super
->dev_item
.uuid
;
1907 *device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1912 blkdev_put(bdev
, FMODE_READ
);
1916 int btrfs_find_device_missing_or_by_path(struct btrfs_root
*root
,
1918 struct btrfs_device
**device
)
1921 if (strcmp(device_path
, "missing") == 0) {
1922 struct list_head
*devices
;
1923 struct btrfs_device
*tmp
;
1925 devices
= &root
->fs_info
->fs_devices
->devices
;
1927 * It is safe to read the devices since the volume_mutex
1928 * is held by the caller.
1930 list_for_each_entry(tmp
, devices
, dev_list
) {
1931 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
1938 btrfs_err(root
->fs_info
, "no missing device found");
1944 return btrfs_find_device_by_path(root
, device_path
, device
);
1949 * does all the dirty work required for changing file system's UUID.
1951 static int btrfs_prepare_sprout(struct btrfs_root
*root
)
1953 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
1954 struct btrfs_fs_devices
*old_devices
;
1955 struct btrfs_fs_devices
*seed_devices
;
1956 struct btrfs_super_block
*disk_super
= root
->fs_info
->super_copy
;
1957 struct btrfs_device
*device
;
1960 BUG_ON(!mutex_is_locked(&uuid_mutex
));
1961 if (!fs_devices
->seeding
)
1964 seed_devices
= __alloc_fs_devices();
1965 if (IS_ERR(seed_devices
))
1966 return PTR_ERR(seed_devices
);
1968 old_devices
= clone_fs_devices(fs_devices
);
1969 if (IS_ERR(old_devices
)) {
1970 kfree(seed_devices
);
1971 return PTR_ERR(old_devices
);
1974 list_add(&old_devices
->list
, &fs_uuids
);
1976 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
1977 seed_devices
->opened
= 1;
1978 INIT_LIST_HEAD(&seed_devices
->devices
);
1979 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
1980 mutex_init(&seed_devices
->device_list_mutex
);
1982 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1983 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
1985 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
)
1986 device
->fs_devices
= seed_devices
;
1989 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
1990 unlock_chunks(root
);
1992 fs_devices
->seeding
= 0;
1993 fs_devices
->num_devices
= 0;
1994 fs_devices
->open_devices
= 0;
1995 fs_devices
->missing_devices
= 0;
1996 fs_devices
->rotating
= 0;
1997 fs_devices
->seed
= seed_devices
;
1999 generate_random_uuid(fs_devices
->fsid
);
2000 memcpy(root
->fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2001 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2002 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2004 super_flags
= btrfs_super_flags(disk_super
) &
2005 ~BTRFS_SUPER_FLAG_SEEDING
;
2006 btrfs_set_super_flags(disk_super
, super_flags
);
2012 * strore the expected generation for seed devices in device items.
2014 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
2015 struct btrfs_root
*root
)
2017 struct btrfs_path
*path
;
2018 struct extent_buffer
*leaf
;
2019 struct btrfs_dev_item
*dev_item
;
2020 struct btrfs_device
*device
;
2021 struct btrfs_key key
;
2022 u8 fs_uuid
[BTRFS_UUID_SIZE
];
2023 u8 dev_uuid
[BTRFS_UUID_SIZE
];
2027 path
= btrfs_alloc_path();
2031 root
= root
->fs_info
->chunk_root
;
2032 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2034 key
.type
= BTRFS_DEV_ITEM_KEY
;
2037 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2041 leaf
= path
->nodes
[0];
2043 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
2044 ret
= btrfs_next_leaf(root
, path
);
2049 leaf
= path
->nodes
[0];
2050 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2051 btrfs_release_path(path
);
2055 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2056 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
2057 key
.type
!= BTRFS_DEV_ITEM_KEY
)
2060 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2061 struct btrfs_dev_item
);
2062 devid
= btrfs_device_id(leaf
, dev_item
);
2063 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
2065 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
2067 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
2069 BUG_ON(!device
); /* Logic error */
2071 if (device
->fs_devices
->seeding
) {
2072 btrfs_set_device_generation(leaf
, dev_item
,
2073 device
->generation
);
2074 btrfs_mark_buffer_dirty(leaf
);
2082 btrfs_free_path(path
);
2086 int btrfs_init_new_device(struct btrfs_root
*root
, char *device_path
)
2088 struct request_queue
*q
;
2089 struct btrfs_trans_handle
*trans
;
2090 struct btrfs_device
*device
;
2091 struct block_device
*bdev
;
2092 struct list_head
*devices
;
2093 struct super_block
*sb
= root
->fs_info
->sb
;
2094 struct rcu_string
*name
;
2096 int seeding_dev
= 0;
2099 if ((sb
->s_flags
& MS_RDONLY
) && !root
->fs_info
->fs_devices
->seeding
)
2102 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2103 root
->fs_info
->bdev_holder
);
2105 return PTR_ERR(bdev
);
2107 if (root
->fs_info
->fs_devices
->seeding
) {
2109 down_write(&sb
->s_umount
);
2110 mutex_lock(&uuid_mutex
);
2113 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2115 devices
= &root
->fs_info
->fs_devices
->devices
;
2117 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2118 list_for_each_entry(device
, devices
, dev_list
) {
2119 if (device
->bdev
== bdev
) {
2122 &root
->fs_info
->fs_devices
->device_list_mutex
);
2126 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2128 device
= btrfs_alloc_device(root
->fs_info
, NULL
, NULL
);
2129 if (IS_ERR(device
)) {
2130 /* we can safely leave the fs_devices entry around */
2131 ret
= PTR_ERR(device
);
2135 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2141 rcu_assign_pointer(device
->name
, name
);
2143 trans
= btrfs_start_transaction(root
, 0);
2144 if (IS_ERR(trans
)) {
2145 rcu_string_free(device
->name
);
2147 ret
= PTR_ERR(trans
);
2151 q
= bdev_get_queue(bdev
);
2152 if (blk_queue_discard(q
))
2153 device
->can_discard
= 1;
2154 device
->writeable
= 1;
2155 device
->generation
= trans
->transid
;
2156 device
->io_width
= root
->sectorsize
;
2157 device
->io_align
= root
->sectorsize
;
2158 device
->sector_size
= root
->sectorsize
;
2159 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
2160 device
->disk_total_bytes
= device
->total_bytes
;
2161 device
->commit_total_bytes
= device
->total_bytes
;
2162 device
->dev_root
= root
->fs_info
->dev_root
;
2163 device
->bdev
= bdev
;
2164 device
->in_fs_metadata
= 1;
2165 device
->is_tgtdev_for_dev_replace
= 0;
2166 device
->mode
= FMODE_EXCL
;
2167 device
->dev_stats_valid
= 1;
2168 set_blocksize(device
->bdev
, 4096);
2171 sb
->s_flags
&= ~MS_RDONLY
;
2172 ret
= btrfs_prepare_sprout(root
);
2173 BUG_ON(ret
); /* -ENOMEM */
2176 device
->fs_devices
= root
->fs_info
->fs_devices
;
2178 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2180 list_add_rcu(&device
->dev_list
, &root
->fs_info
->fs_devices
->devices
);
2181 list_add(&device
->dev_alloc_list
,
2182 &root
->fs_info
->fs_devices
->alloc_list
);
2183 root
->fs_info
->fs_devices
->num_devices
++;
2184 root
->fs_info
->fs_devices
->open_devices
++;
2185 root
->fs_info
->fs_devices
->rw_devices
++;
2186 root
->fs_info
->fs_devices
->total_devices
++;
2187 root
->fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
2189 spin_lock(&root
->fs_info
->free_chunk_lock
);
2190 root
->fs_info
->free_chunk_space
+= device
->total_bytes
;
2191 spin_unlock(&root
->fs_info
->free_chunk_lock
);
2193 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
2194 root
->fs_info
->fs_devices
->rotating
= 1;
2196 tmp
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
2197 btrfs_set_super_total_bytes(root
->fs_info
->super_copy
,
2198 tmp
+ device
->total_bytes
);
2200 tmp
= btrfs_super_num_devices(root
->fs_info
->super_copy
);
2201 btrfs_set_super_num_devices(root
->fs_info
->super_copy
,
2204 /* add sysfs device entry */
2205 btrfs_kobj_add_device(root
->fs_info
, device
);
2208 * we've got more storage, clear any full flags on the space
2211 btrfs_clear_space_info_full(root
->fs_info
);
2213 unlock_chunks(root
);
2214 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2218 ret
= init_first_rw_device(trans
, root
, device
);
2219 unlock_chunks(root
);
2221 btrfs_abort_transaction(trans
, root
, ret
);
2226 ret
= btrfs_add_device(trans
, root
, device
);
2228 btrfs_abort_transaction(trans
, root
, ret
);
2233 char fsid_buf
[BTRFS_UUID_UNPARSED_SIZE
];
2235 ret
= btrfs_finish_sprout(trans
, root
);
2237 btrfs_abort_transaction(trans
, root
, ret
);
2241 /* Sprouting would change fsid of the mounted root,
2242 * so rename the fsid on the sysfs
2244 snprintf(fsid_buf
, BTRFS_UUID_UNPARSED_SIZE
, "%pU",
2245 root
->fs_info
->fsid
);
2246 if (kobject_rename(&root
->fs_info
->super_kobj
, fsid_buf
))
2250 root
->fs_info
->num_tolerated_disk_barrier_failures
=
2251 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
2252 ret
= btrfs_commit_transaction(trans
, root
);
2255 mutex_unlock(&uuid_mutex
);
2256 up_write(&sb
->s_umount
);
2258 if (ret
) /* transaction commit */
2261 ret
= btrfs_relocate_sys_chunks(root
);
2263 btrfs_error(root
->fs_info
, ret
,
2264 "Failed to relocate sys chunks after "
2265 "device initialization. This can be fixed "
2266 "using the \"btrfs balance\" command.");
2267 trans
= btrfs_attach_transaction(root
);
2268 if (IS_ERR(trans
)) {
2269 if (PTR_ERR(trans
) == -ENOENT
)
2271 return PTR_ERR(trans
);
2273 ret
= btrfs_commit_transaction(trans
, root
);
2276 /* Update ctime/mtime for libblkid */
2277 update_dev_time(device_path
);
2281 btrfs_end_transaction(trans
, root
);
2282 rcu_string_free(device
->name
);
2283 btrfs_kobj_rm_device(root
->fs_info
, device
);
2286 blkdev_put(bdev
, FMODE_EXCL
);
2288 mutex_unlock(&uuid_mutex
);
2289 up_write(&sb
->s_umount
);
2294 int btrfs_init_dev_replace_tgtdev(struct btrfs_root
*root
, char *device_path
,
2295 struct btrfs_device
*srcdev
,
2296 struct btrfs_device
**device_out
)
2298 struct request_queue
*q
;
2299 struct btrfs_device
*device
;
2300 struct block_device
*bdev
;
2301 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2302 struct list_head
*devices
;
2303 struct rcu_string
*name
;
2304 u64 devid
= BTRFS_DEV_REPLACE_DEVID
;
2308 if (fs_info
->fs_devices
->seeding
) {
2309 btrfs_err(fs_info
, "the filesystem is a seed filesystem!");
2313 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2314 fs_info
->bdev_holder
);
2316 btrfs_err(fs_info
, "target device %s is invalid!", device_path
);
2317 return PTR_ERR(bdev
);
2320 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2322 devices
= &fs_info
->fs_devices
->devices
;
2323 list_for_each_entry(device
, devices
, dev_list
) {
2324 if (device
->bdev
== bdev
) {
2325 btrfs_err(fs_info
, "target device is in the filesystem!");
2332 if (i_size_read(bdev
->bd_inode
) <
2333 btrfs_device_get_total_bytes(srcdev
)) {
2334 btrfs_err(fs_info
, "target device is smaller than source device!");
2340 device
= btrfs_alloc_device(NULL
, &devid
, NULL
);
2341 if (IS_ERR(device
)) {
2342 ret
= PTR_ERR(device
);
2346 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2352 rcu_assign_pointer(device
->name
, name
);
2354 q
= bdev_get_queue(bdev
);
2355 if (blk_queue_discard(q
))
2356 device
->can_discard
= 1;
2357 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2358 device
->writeable
= 1;
2359 device
->generation
= 0;
2360 device
->io_width
= root
->sectorsize
;
2361 device
->io_align
= root
->sectorsize
;
2362 device
->sector_size
= root
->sectorsize
;
2363 device
->total_bytes
= btrfs_device_get_total_bytes(srcdev
);
2364 device
->disk_total_bytes
= btrfs_device_get_disk_total_bytes(srcdev
);
2365 device
->bytes_used
= btrfs_device_get_bytes_used(srcdev
);
2366 ASSERT(list_empty(&srcdev
->resized_list
));
2367 device
->commit_total_bytes
= srcdev
->commit_total_bytes
;
2368 device
->commit_bytes_used
= device
->bytes_used
;
2369 device
->dev_root
= fs_info
->dev_root
;
2370 device
->bdev
= bdev
;
2371 device
->in_fs_metadata
= 1;
2372 device
->is_tgtdev_for_dev_replace
= 1;
2373 device
->mode
= FMODE_EXCL
;
2374 device
->dev_stats_valid
= 1;
2375 set_blocksize(device
->bdev
, 4096);
2376 device
->fs_devices
= fs_info
->fs_devices
;
2377 list_add(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2378 fs_info
->fs_devices
->num_devices
++;
2379 fs_info
->fs_devices
->open_devices
++;
2380 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2382 *device_out
= device
;
2386 blkdev_put(bdev
, FMODE_EXCL
);
2390 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info
*fs_info
,
2391 struct btrfs_device
*tgtdev
)
2393 WARN_ON(fs_info
->fs_devices
->rw_devices
== 0);
2394 tgtdev
->io_width
= fs_info
->dev_root
->sectorsize
;
2395 tgtdev
->io_align
= fs_info
->dev_root
->sectorsize
;
2396 tgtdev
->sector_size
= fs_info
->dev_root
->sectorsize
;
2397 tgtdev
->dev_root
= fs_info
->dev_root
;
2398 tgtdev
->in_fs_metadata
= 1;
2401 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2402 struct btrfs_device
*device
)
2405 struct btrfs_path
*path
;
2406 struct btrfs_root
*root
;
2407 struct btrfs_dev_item
*dev_item
;
2408 struct extent_buffer
*leaf
;
2409 struct btrfs_key key
;
2411 root
= device
->dev_root
->fs_info
->chunk_root
;
2413 path
= btrfs_alloc_path();
2417 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2418 key
.type
= BTRFS_DEV_ITEM_KEY
;
2419 key
.offset
= device
->devid
;
2421 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2430 leaf
= path
->nodes
[0];
2431 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2433 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2434 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2435 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2436 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2437 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2438 btrfs_set_device_total_bytes(leaf
, dev_item
,
2439 btrfs_device_get_disk_total_bytes(device
));
2440 btrfs_set_device_bytes_used(leaf
, dev_item
,
2441 btrfs_device_get_bytes_used(device
));
2442 btrfs_mark_buffer_dirty(leaf
);
2445 btrfs_free_path(path
);
2449 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2450 struct btrfs_device
*device
, u64 new_size
)
2452 struct btrfs_super_block
*super_copy
=
2453 device
->dev_root
->fs_info
->super_copy
;
2454 struct btrfs_fs_devices
*fs_devices
;
2458 if (!device
->writeable
)
2461 lock_chunks(device
->dev_root
);
2462 old_total
= btrfs_super_total_bytes(super_copy
);
2463 diff
= new_size
- device
->total_bytes
;
2465 if (new_size
<= device
->total_bytes
||
2466 device
->is_tgtdev_for_dev_replace
) {
2467 unlock_chunks(device
->dev_root
);
2471 fs_devices
= device
->dev_root
->fs_info
->fs_devices
;
2473 btrfs_set_super_total_bytes(super_copy
, old_total
+ diff
);
2474 device
->fs_devices
->total_rw_bytes
+= diff
;
2476 btrfs_device_set_total_bytes(device
, new_size
);
2477 btrfs_device_set_disk_total_bytes(device
, new_size
);
2478 btrfs_clear_space_info_full(device
->dev_root
->fs_info
);
2479 if (list_empty(&device
->resized_list
))
2480 list_add_tail(&device
->resized_list
,
2481 &fs_devices
->resized_devices
);
2482 unlock_chunks(device
->dev_root
);
2484 return btrfs_update_device(trans
, device
);
2487 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
2488 struct btrfs_root
*root
, u64 chunk_objectid
,
2492 struct btrfs_path
*path
;
2493 struct btrfs_key key
;
2495 root
= root
->fs_info
->chunk_root
;
2496 path
= btrfs_alloc_path();
2500 key
.objectid
= chunk_objectid
;
2501 key
.offset
= chunk_offset
;
2502 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2504 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2507 else if (ret
> 0) { /* Logic error or corruption */
2508 btrfs_error(root
->fs_info
, -ENOENT
,
2509 "Failed lookup while freeing chunk.");
2514 ret
= btrfs_del_item(trans
, root
, path
);
2516 btrfs_error(root
->fs_info
, ret
,
2517 "Failed to delete chunk item.");
2519 btrfs_free_path(path
);
2523 static int btrfs_del_sys_chunk(struct btrfs_root
*root
, u64 chunk_objectid
, u64
2526 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
2527 struct btrfs_disk_key
*disk_key
;
2528 struct btrfs_chunk
*chunk
;
2535 struct btrfs_key key
;
2538 array_size
= btrfs_super_sys_array_size(super_copy
);
2540 ptr
= super_copy
->sys_chunk_array
;
2543 while (cur
< array_size
) {
2544 disk_key
= (struct btrfs_disk_key
*)ptr
;
2545 btrfs_disk_key_to_cpu(&key
, disk_key
);
2547 len
= sizeof(*disk_key
);
2549 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2550 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2551 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2552 len
+= btrfs_chunk_item_size(num_stripes
);
2557 if (key
.objectid
== chunk_objectid
&&
2558 key
.offset
== chunk_offset
) {
2559 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2561 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2567 unlock_chunks(root
);
2571 int btrfs_remove_chunk(struct btrfs_trans_handle
*trans
,
2572 struct btrfs_root
*root
, u64 chunk_offset
)
2574 struct extent_map_tree
*em_tree
;
2575 struct extent_map
*em
;
2576 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2577 struct map_lookup
*map
;
2578 u64 dev_extent_len
= 0;
2579 u64 chunk_objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2583 root
= root
->fs_info
->chunk_root
;
2584 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
2586 read_lock(&em_tree
->lock
);
2587 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
2588 read_unlock(&em_tree
->lock
);
2590 if (!em
|| em
->start
> chunk_offset
||
2591 em
->start
+ em
->len
< chunk_offset
) {
2593 * This is a logic error, but we don't want to just rely on the
2594 * user having built with ASSERT enabled, so if ASSERT doens't
2595 * do anything we still error out.
2599 free_extent_map(em
);
2602 map
= (struct map_lookup
*)em
->bdev
;
2604 for (i
= 0; i
< map
->num_stripes
; i
++) {
2605 struct btrfs_device
*device
= map
->stripes
[i
].dev
;
2606 ret
= btrfs_free_dev_extent(trans
, device
,
2607 map
->stripes
[i
].physical
,
2610 btrfs_abort_transaction(trans
, root
, ret
);
2614 if (device
->bytes_used
> 0) {
2616 btrfs_device_set_bytes_used(device
,
2617 device
->bytes_used
- dev_extent_len
);
2618 spin_lock(&root
->fs_info
->free_chunk_lock
);
2619 root
->fs_info
->free_chunk_space
+= dev_extent_len
;
2620 spin_unlock(&root
->fs_info
->free_chunk_lock
);
2621 btrfs_clear_space_info_full(root
->fs_info
);
2622 unlock_chunks(root
);
2625 if (map
->stripes
[i
].dev
) {
2626 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2628 btrfs_abort_transaction(trans
, root
, ret
);
2633 ret
= btrfs_free_chunk(trans
, root
, chunk_objectid
, chunk_offset
);
2635 btrfs_abort_transaction(trans
, root
, ret
);
2639 trace_btrfs_chunk_free(root
, map
, chunk_offset
, em
->len
);
2641 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2642 ret
= btrfs_del_sys_chunk(root
, chunk_objectid
, chunk_offset
);
2644 btrfs_abort_transaction(trans
, root
, ret
);
2649 ret
= btrfs_remove_block_group(trans
, extent_root
, chunk_offset
, em
);
2651 btrfs_abort_transaction(trans
, extent_root
, ret
);
2657 free_extent_map(em
);
2661 static int btrfs_relocate_chunk(struct btrfs_root
*root
,
2665 struct btrfs_root
*extent_root
;
2666 struct btrfs_trans_handle
*trans
;
2669 root
= root
->fs_info
->chunk_root
;
2670 extent_root
= root
->fs_info
->extent_root
;
2672 ret
= btrfs_can_relocate(extent_root
, chunk_offset
);
2676 /* step one, relocate all the extents inside this chunk */
2677 ret
= btrfs_relocate_block_group(extent_root
, chunk_offset
);
2681 trans
= btrfs_start_transaction(root
, 0);
2682 if (IS_ERR(trans
)) {
2683 ret
= PTR_ERR(trans
);
2684 btrfs_std_error(root
->fs_info
, ret
);
2689 * step two, delete the device extents and the
2690 * chunk tree entries
2692 ret
= btrfs_remove_chunk(trans
, root
, chunk_offset
);
2693 btrfs_end_transaction(trans
, root
);
2697 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
)
2699 struct btrfs_root
*chunk_root
= root
->fs_info
->chunk_root
;
2700 struct btrfs_path
*path
;
2701 struct extent_buffer
*leaf
;
2702 struct btrfs_chunk
*chunk
;
2703 struct btrfs_key key
;
2704 struct btrfs_key found_key
;
2706 bool retried
= false;
2710 path
= btrfs_alloc_path();
2715 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2716 key
.offset
= (u64
)-1;
2717 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2720 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2723 BUG_ON(ret
== 0); /* Corruption */
2725 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
2732 leaf
= path
->nodes
[0];
2733 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2735 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
2736 struct btrfs_chunk
);
2737 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2738 btrfs_release_path(path
);
2740 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2741 ret
= btrfs_relocate_chunk(chunk_root
,
2750 if (found_key
.offset
== 0)
2752 key
.offset
= found_key
.offset
- 1;
2755 if (failed
&& !retried
) {
2759 } else if (WARN_ON(failed
&& retried
)) {
2763 btrfs_free_path(path
);
2767 static int insert_balance_item(struct btrfs_root
*root
,
2768 struct btrfs_balance_control
*bctl
)
2770 struct btrfs_trans_handle
*trans
;
2771 struct btrfs_balance_item
*item
;
2772 struct btrfs_disk_balance_args disk_bargs
;
2773 struct btrfs_path
*path
;
2774 struct extent_buffer
*leaf
;
2775 struct btrfs_key key
;
2778 path
= btrfs_alloc_path();
2782 trans
= btrfs_start_transaction(root
, 0);
2783 if (IS_ERR(trans
)) {
2784 btrfs_free_path(path
);
2785 return PTR_ERR(trans
);
2788 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2789 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2792 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2797 leaf
= path
->nodes
[0];
2798 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
2800 memset_extent_buffer(leaf
, 0, (unsigned long)item
, sizeof(*item
));
2802 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
2803 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
2804 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
2805 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
2806 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
2807 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
2809 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
2811 btrfs_mark_buffer_dirty(leaf
);
2813 btrfs_free_path(path
);
2814 err
= btrfs_commit_transaction(trans
, root
);
2820 static int del_balance_item(struct btrfs_root
*root
)
2822 struct btrfs_trans_handle
*trans
;
2823 struct btrfs_path
*path
;
2824 struct btrfs_key key
;
2827 path
= btrfs_alloc_path();
2831 trans
= btrfs_start_transaction(root
, 0);
2832 if (IS_ERR(trans
)) {
2833 btrfs_free_path(path
);
2834 return PTR_ERR(trans
);
2837 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2838 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2841 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2849 ret
= btrfs_del_item(trans
, root
, path
);
2851 btrfs_free_path(path
);
2852 err
= btrfs_commit_transaction(trans
, root
);
2859 * This is a heuristic used to reduce the number of chunks balanced on
2860 * resume after balance was interrupted.
2862 static void update_balance_args(struct btrfs_balance_control
*bctl
)
2865 * Turn on soft mode for chunk types that were being converted.
2867 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2868 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2869 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2870 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2871 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2872 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2875 * Turn on usage filter if is not already used. The idea is
2876 * that chunks that we have already balanced should be
2877 * reasonably full. Don't do it for chunks that are being
2878 * converted - that will keep us from relocating unconverted
2879 * (albeit full) chunks.
2881 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2882 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2883 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2884 bctl
->data
.usage
= 90;
2886 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2887 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2888 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2889 bctl
->sys
.usage
= 90;
2891 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2892 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2893 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2894 bctl
->meta
.usage
= 90;
2899 * Should be called with both balance and volume mutexes held to
2900 * serialize other volume operations (add_dev/rm_dev/resize) with
2901 * restriper. Same goes for unset_balance_control.
2903 static void set_balance_control(struct btrfs_balance_control
*bctl
)
2905 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
2907 BUG_ON(fs_info
->balance_ctl
);
2909 spin_lock(&fs_info
->balance_lock
);
2910 fs_info
->balance_ctl
= bctl
;
2911 spin_unlock(&fs_info
->balance_lock
);
2914 static void unset_balance_control(struct btrfs_fs_info
*fs_info
)
2916 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
2918 BUG_ON(!fs_info
->balance_ctl
);
2920 spin_lock(&fs_info
->balance_lock
);
2921 fs_info
->balance_ctl
= NULL
;
2922 spin_unlock(&fs_info
->balance_lock
);
2928 * Balance filters. Return 1 if chunk should be filtered out
2929 * (should not be balanced).
2931 static int chunk_profiles_filter(u64 chunk_type
,
2932 struct btrfs_balance_args
*bargs
)
2934 chunk_type
= chunk_to_extended(chunk_type
) &
2935 BTRFS_EXTENDED_PROFILE_MASK
;
2937 if (bargs
->profiles
& chunk_type
)
2943 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
2944 struct btrfs_balance_args
*bargs
)
2946 struct btrfs_block_group_cache
*cache
;
2947 u64 chunk_used
, user_thresh
;
2950 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
2951 chunk_used
= btrfs_block_group_used(&cache
->item
);
2953 if (bargs
->usage
== 0)
2955 else if (bargs
->usage
> 100)
2956 user_thresh
= cache
->key
.offset
;
2958 user_thresh
= div_factor_fine(cache
->key
.offset
,
2961 if (chunk_used
< user_thresh
)
2964 btrfs_put_block_group(cache
);
2968 static int chunk_devid_filter(struct extent_buffer
*leaf
,
2969 struct btrfs_chunk
*chunk
,
2970 struct btrfs_balance_args
*bargs
)
2972 struct btrfs_stripe
*stripe
;
2973 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2976 for (i
= 0; i
< num_stripes
; i
++) {
2977 stripe
= btrfs_stripe_nr(chunk
, i
);
2978 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
2985 /* [pstart, pend) */
2986 static int chunk_drange_filter(struct extent_buffer
*leaf
,
2987 struct btrfs_chunk
*chunk
,
2989 struct btrfs_balance_args
*bargs
)
2991 struct btrfs_stripe
*stripe
;
2992 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2998 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
3001 if (btrfs_chunk_type(leaf
, chunk
) & (BTRFS_BLOCK_GROUP_DUP
|
3002 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)) {
3003 factor
= num_stripes
/ 2;
3004 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID5
) {
3005 factor
= num_stripes
- 1;
3006 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID6
) {
3007 factor
= num_stripes
- 2;
3009 factor
= num_stripes
;
3012 for (i
= 0; i
< num_stripes
; i
++) {
3013 stripe
= btrfs_stripe_nr(chunk
, i
);
3014 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
3017 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
3018 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
3019 stripe_length
= div_u64(stripe_length
, factor
);
3021 if (stripe_offset
< bargs
->pend
&&
3022 stripe_offset
+ stripe_length
> bargs
->pstart
)
3029 /* [vstart, vend) */
3030 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
3031 struct btrfs_chunk
*chunk
,
3033 struct btrfs_balance_args
*bargs
)
3035 if (chunk_offset
< bargs
->vend
&&
3036 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
3037 /* at least part of the chunk is inside this vrange */
3043 static int chunk_soft_convert_filter(u64 chunk_type
,
3044 struct btrfs_balance_args
*bargs
)
3046 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
3049 chunk_type
= chunk_to_extended(chunk_type
) &
3050 BTRFS_EXTENDED_PROFILE_MASK
;
3052 if (bargs
->target
== chunk_type
)
3058 static int should_balance_chunk(struct btrfs_root
*root
,
3059 struct extent_buffer
*leaf
,
3060 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
3062 struct btrfs_balance_control
*bctl
= root
->fs_info
->balance_ctl
;
3063 struct btrfs_balance_args
*bargs
= NULL
;
3064 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3067 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
3068 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
3072 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3073 bargs
= &bctl
->data
;
3074 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3076 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3077 bargs
= &bctl
->meta
;
3079 /* profiles filter */
3080 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
3081 chunk_profiles_filter(chunk_type
, bargs
)) {
3086 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3087 chunk_usage_filter(bctl
->fs_info
, chunk_offset
, bargs
)) {
3092 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
3093 chunk_devid_filter(leaf
, chunk
, bargs
)) {
3097 /* drange filter, makes sense only with devid filter */
3098 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
3099 chunk_drange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3104 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
3105 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3109 /* soft profile changing mode */
3110 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
3111 chunk_soft_convert_filter(chunk_type
, bargs
)) {
3116 * limited by count, must be the last filter
3118 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT
)) {
3119 if (bargs
->limit
== 0)
3128 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
3130 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3131 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
3132 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
3133 struct list_head
*devices
;
3134 struct btrfs_device
*device
;
3137 struct btrfs_chunk
*chunk
;
3138 struct btrfs_path
*path
;
3139 struct btrfs_key key
;
3140 struct btrfs_key found_key
;
3141 struct btrfs_trans_handle
*trans
;
3142 struct extent_buffer
*leaf
;
3145 int enospc_errors
= 0;
3146 bool counting
= true;
3147 u64 limit_data
= bctl
->data
.limit
;
3148 u64 limit_meta
= bctl
->meta
.limit
;
3149 u64 limit_sys
= bctl
->sys
.limit
;
3151 /* step one make some room on all the devices */
3152 devices
= &fs_info
->fs_devices
->devices
;
3153 list_for_each_entry(device
, devices
, dev_list
) {
3154 old_size
= btrfs_device_get_total_bytes(device
);
3155 size_to_free
= div_factor(old_size
, 1);
3156 size_to_free
= min(size_to_free
, (u64
)1 * 1024 * 1024);
3157 if (!device
->writeable
||
3158 btrfs_device_get_total_bytes(device
) -
3159 btrfs_device_get_bytes_used(device
) > size_to_free
||
3160 device
->is_tgtdev_for_dev_replace
)
3163 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
3168 trans
= btrfs_start_transaction(dev_root
, 0);
3169 BUG_ON(IS_ERR(trans
));
3171 ret
= btrfs_grow_device(trans
, device
, old_size
);
3174 btrfs_end_transaction(trans
, dev_root
);
3177 /* step two, relocate all the chunks */
3178 path
= btrfs_alloc_path();
3184 /* zero out stat counters */
3185 spin_lock(&fs_info
->balance_lock
);
3186 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
3187 spin_unlock(&fs_info
->balance_lock
);
3190 bctl
->data
.limit
= limit_data
;
3191 bctl
->meta
.limit
= limit_meta
;
3192 bctl
->sys
.limit
= limit_sys
;
3194 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3195 key
.offset
= (u64
)-1;
3196 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3199 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
3200 atomic_read(&fs_info
->balance_cancel_req
)) {
3205 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
3210 * this shouldn't happen, it means the last relocate
3214 BUG(); /* FIXME break ? */
3216 ret
= btrfs_previous_item(chunk_root
, path
, 0,
3217 BTRFS_CHUNK_ITEM_KEY
);
3223 leaf
= path
->nodes
[0];
3224 slot
= path
->slots
[0];
3225 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3227 if (found_key
.objectid
!= key
.objectid
)
3230 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
3233 spin_lock(&fs_info
->balance_lock
);
3234 bctl
->stat
.considered
++;
3235 spin_unlock(&fs_info
->balance_lock
);
3238 ret
= should_balance_chunk(chunk_root
, leaf
, chunk
,
3240 btrfs_release_path(path
);
3245 spin_lock(&fs_info
->balance_lock
);
3246 bctl
->stat
.expected
++;
3247 spin_unlock(&fs_info
->balance_lock
);
3251 ret
= btrfs_relocate_chunk(chunk_root
,
3254 if (ret
&& ret
!= -ENOSPC
)
3256 if (ret
== -ENOSPC
) {
3259 spin_lock(&fs_info
->balance_lock
);
3260 bctl
->stat
.completed
++;
3261 spin_unlock(&fs_info
->balance_lock
);
3264 if (found_key
.offset
== 0)
3266 key
.offset
= found_key
.offset
- 1;
3270 btrfs_release_path(path
);
3275 btrfs_free_path(path
);
3276 if (enospc_errors
) {
3277 btrfs_info(fs_info
, "%d enospc errors during balance",
3287 * alloc_profile_is_valid - see if a given profile is valid and reduced
3288 * @flags: profile to validate
3289 * @extended: if true @flags is treated as an extended profile
3291 static int alloc_profile_is_valid(u64 flags
, int extended
)
3293 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
3294 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
3296 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
3298 /* 1) check that all other bits are zeroed */
3302 /* 2) see if profile is reduced */
3304 return !extended
; /* "0" is valid for usual profiles */
3306 /* true if exactly one bit set */
3307 return (flags
& (flags
- 1)) == 0;
3310 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
3312 /* cancel requested || normal exit path */
3313 return atomic_read(&fs_info
->balance_cancel_req
) ||
3314 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
3315 atomic_read(&fs_info
->balance_cancel_req
) == 0);
3318 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
3322 unset_balance_control(fs_info
);
3323 ret
= del_balance_item(fs_info
->tree_root
);
3325 btrfs_std_error(fs_info
, ret
);
3327 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3331 * Should be called with both balance and volume mutexes held
3333 int btrfs_balance(struct btrfs_balance_control
*bctl
,
3334 struct btrfs_ioctl_balance_args
*bargs
)
3336 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3343 if (btrfs_fs_closing(fs_info
) ||
3344 atomic_read(&fs_info
->balance_pause_req
) ||
3345 atomic_read(&fs_info
->balance_cancel_req
)) {
3350 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
3351 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
3355 * In case of mixed groups both data and meta should be picked,
3356 * and identical options should be given for both of them.
3358 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
3359 if (mixed
&& (bctl
->flags
& allowed
)) {
3360 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
3361 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
3362 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
3363 btrfs_err(fs_info
, "with mixed groups data and "
3364 "metadata balance options must be the same");
3370 num_devices
= fs_info
->fs_devices
->num_devices
;
3371 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
3372 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
3373 BUG_ON(num_devices
< 1);
3376 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
3377 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
3378 if (num_devices
== 1)
3379 allowed
|= BTRFS_BLOCK_GROUP_DUP
;
3380 else if (num_devices
> 1)
3381 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
3382 if (num_devices
> 2)
3383 allowed
|= BTRFS_BLOCK_GROUP_RAID5
;
3384 if (num_devices
> 3)
3385 allowed
|= (BTRFS_BLOCK_GROUP_RAID10
|
3386 BTRFS_BLOCK_GROUP_RAID6
);
3387 if ((bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3388 (!alloc_profile_is_valid(bctl
->data
.target
, 1) ||
3389 (bctl
->data
.target
& ~allowed
))) {
3390 btrfs_err(fs_info
, "unable to start balance with target "
3391 "data profile %llu",
3396 if ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3397 (!alloc_profile_is_valid(bctl
->meta
.target
, 1) ||
3398 (bctl
->meta
.target
& ~allowed
))) {
3400 "unable to start balance with target metadata profile %llu",
3405 if ((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3406 (!alloc_profile_is_valid(bctl
->sys
.target
, 1) ||
3407 (bctl
->sys
.target
& ~allowed
))) {
3409 "unable to start balance with target system profile %llu",
3415 /* allow dup'ed data chunks only in mixed mode */
3416 if (!mixed
&& (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3417 (bctl
->data
.target
& BTRFS_BLOCK_GROUP_DUP
)) {
3418 btrfs_err(fs_info
, "dup for data is not allowed");
3423 /* allow to reduce meta or sys integrity only if force set */
3424 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3425 BTRFS_BLOCK_GROUP_RAID10
|
3426 BTRFS_BLOCK_GROUP_RAID5
|
3427 BTRFS_BLOCK_GROUP_RAID6
;
3429 seq
= read_seqbegin(&fs_info
->profiles_lock
);
3431 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3432 (fs_info
->avail_system_alloc_bits
& allowed
) &&
3433 !(bctl
->sys
.target
& allowed
)) ||
3434 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3435 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
3436 !(bctl
->meta
.target
& allowed
))) {
3437 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
3438 btrfs_info(fs_info
, "force reducing metadata integrity");
3440 btrfs_err(fs_info
, "balance will reduce metadata "
3441 "integrity, use force if you want this");
3446 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
3448 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3449 int num_tolerated_disk_barrier_failures
;
3450 u64 target
= bctl
->sys
.target
;
3452 num_tolerated_disk_barrier_failures
=
3453 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3454 if (num_tolerated_disk_barrier_failures
> 0 &&
3456 (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID0
|
3457 BTRFS_AVAIL_ALLOC_BIT_SINGLE
)))
3458 num_tolerated_disk_barrier_failures
= 0;
3459 else if (num_tolerated_disk_barrier_failures
> 1 &&
3461 (BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)))
3462 num_tolerated_disk_barrier_failures
= 1;
3464 fs_info
->num_tolerated_disk_barrier_failures
=
3465 num_tolerated_disk_barrier_failures
;
3468 ret
= insert_balance_item(fs_info
->tree_root
, bctl
);
3469 if (ret
&& ret
!= -EEXIST
)
3472 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
3473 BUG_ON(ret
== -EEXIST
);
3474 set_balance_control(bctl
);
3476 BUG_ON(ret
!= -EEXIST
);
3477 spin_lock(&fs_info
->balance_lock
);
3478 update_balance_args(bctl
);
3479 spin_unlock(&fs_info
->balance_lock
);
3482 atomic_inc(&fs_info
->balance_running
);
3483 mutex_unlock(&fs_info
->balance_mutex
);
3485 ret
= __btrfs_balance(fs_info
);
3487 mutex_lock(&fs_info
->balance_mutex
);
3488 atomic_dec(&fs_info
->balance_running
);
3490 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3491 fs_info
->num_tolerated_disk_barrier_failures
=
3492 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3496 memset(bargs
, 0, sizeof(*bargs
));
3497 update_ioctl_balance_args(fs_info
, 0, bargs
);
3500 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
3501 balance_need_close(fs_info
)) {
3502 __cancel_balance(fs_info
);
3505 wake_up(&fs_info
->balance_wait_q
);
3509 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
3510 __cancel_balance(fs_info
);
3513 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3518 static int balance_kthread(void *data
)
3520 struct btrfs_fs_info
*fs_info
= data
;
3523 mutex_lock(&fs_info
->volume_mutex
);
3524 mutex_lock(&fs_info
->balance_mutex
);
3526 if (fs_info
->balance_ctl
) {
3527 btrfs_info(fs_info
, "continuing balance");
3528 ret
= btrfs_balance(fs_info
->balance_ctl
, NULL
);
3531 mutex_unlock(&fs_info
->balance_mutex
);
3532 mutex_unlock(&fs_info
->volume_mutex
);
3537 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
3539 struct task_struct
*tsk
;
3541 spin_lock(&fs_info
->balance_lock
);
3542 if (!fs_info
->balance_ctl
) {
3543 spin_unlock(&fs_info
->balance_lock
);
3546 spin_unlock(&fs_info
->balance_lock
);
3548 if (btrfs_test_opt(fs_info
->tree_root
, SKIP_BALANCE
)) {
3549 btrfs_info(fs_info
, "force skipping balance");
3553 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
3554 return PTR_ERR_OR_ZERO(tsk
);
3557 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
3559 struct btrfs_balance_control
*bctl
;
3560 struct btrfs_balance_item
*item
;
3561 struct btrfs_disk_balance_args disk_bargs
;
3562 struct btrfs_path
*path
;
3563 struct extent_buffer
*leaf
;
3564 struct btrfs_key key
;
3567 path
= btrfs_alloc_path();
3571 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3572 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
3575 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
3578 if (ret
> 0) { /* ret = -ENOENT; */
3583 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
3589 leaf
= path
->nodes
[0];
3590 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3592 bctl
->fs_info
= fs_info
;
3593 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
3594 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
3596 btrfs_balance_data(leaf
, item
, &disk_bargs
);
3597 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
3598 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
3599 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
3600 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
3601 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
3603 WARN_ON(atomic_xchg(&fs_info
->mutually_exclusive_operation_running
, 1));
3605 mutex_lock(&fs_info
->volume_mutex
);
3606 mutex_lock(&fs_info
->balance_mutex
);
3608 set_balance_control(bctl
);
3610 mutex_unlock(&fs_info
->balance_mutex
);
3611 mutex_unlock(&fs_info
->volume_mutex
);
3613 btrfs_free_path(path
);
3617 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
3621 mutex_lock(&fs_info
->balance_mutex
);
3622 if (!fs_info
->balance_ctl
) {
3623 mutex_unlock(&fs_info
->balance_mutex
);
3627 if (atomic_read(&fs_info
->balance_running
)) {
3628 atomic_inc(&fs_info
->balance_pause_req
);
3629 mutex_unlock(&fs_info
->balance_mutex
);
3631 wait_event(fs_info
->balance_wait_q
,
3632 atomic_read(&fs_info
->balance_running
) == 0);
3634 mutex_lock(&fs_info
->balance_mutex
);
3635 /* we are good with balance_ctl ripped off from under us */
3636 BUG_ON(atomic_read(&fs_info
->balance_running
));
3637 atomic_dec(&fs_info
->balance_pause_req
);
3642 mutex_unlock(&fs_info
->balance_mutex
);
3646 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
3648 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
3651 mutex_lock(&fs_info
->balance_mutex
);
3652 if (!fs_info
->balance_ctl
) {
3653 mutex_unlock(&fs_info
->balance_mutex
);
3657 atomic_inc(&fs_info
->balance_cancel_req
);
3659 * if we are running just wait and return, balance item is
3660 * deleted in btrfs_balance in this case
3662 if (atomic_read(&fs_info
->balance_running
)) {
3663 mutex_unlock(&fs_info
->balance_mutex
);
3664 wait_event(fs_info
->balance_wait_q
,
3665 atomic_read(&fs_info
->balance_running
) == 0);
3666 mutex_lock(&fs_info
->balance_mutex
);
3668 /* __cancel_balance needs volume_mutex */
3669 mutex_unlock(&fs_info
->balance_mutex
);
3670 mutex_lock(&fs_info
->volume_mutex
);
3671 mutex_lock(&fs_info
->balance_mutex
);
3673 if (fs_info
->balance_ctl
)
3674 __cancel_balance(fs_info
);
3676 mutex_unlock(&fs_info
->volume_mutex
);
3679 BUG_ON(fs_info
->balance_ctl
|| atomic_read(&fs_info
->balance_running
));
3680 atomic_dec(&fs_info
->balance_cancel_req
);
3681 mutex_unlock(&fs_info
->balance_mutex
);
3685 static int btrfs_uuid_scan_kthread(void *data
)
3687 struct btrfs_fs_info
*fs_info
= data
;
3688 struct btrfs_root
*root
= fs_info
->tree_root
;
3689 struct btrfs_key key
;
3690 struct btrfs_key max_key
;
3691 struct btrfs_path
*path
= NULL
;
3693 struct extent_buffer
*eb
;
3695 struct btrfs_root_item root_item
;
3697 struct btrfs_trans_handle
*trans
= NULL
;
3699 path
= btrfs_alloc_path();
3706 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3709 max_key
.objectid
= (u64
)-1;
3710 max_key
.type
= BTRFS_ROOT_ITEM_KEY
;
3711 max_key
.offset
= (u64
)-1;
3714 ret
= btrfs_search_forward(root
, &key
, path
, 0);
3721 if (key
.type
!= BTRFS_ROOT_ITEM_KEY
||
3722 (key
.objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
3723 key
.objectid
!= BTRFS_FS_TREE_OBJECTID
) ||
3724 key
.objectid
> BTRFS_LAST_FREE_OBJECTID
)
3727 eb
= path
->nodes
[0];
3728 slot
= path
->slots
[0];
3729 item_size
= btrfs_item_size_nr(eb
, slot
);
3730 if (item_size
< sizeof(root_item
))
3733 read_extent_buffer(eb
, &root_item
,
3734 btrfs_item_ptr_offset(eb
, slot
),
3735 (int)sizeof(root_item
));
3736 if (btrfs_root_refs(&root_item
) == 0)
3739 if (!btrfs_is_empty_uuid(root_item
.uuid
) ||
3740 !btrfs_is_empty_uuid(root_item
.received_uuid
)) {
3744 btrfs_release_path(path
);
3746 * 1 - subvol uuid item
3747 * 1 - received_subvol uuid item
3749 trans
= btrfs_start_transaction(fs_info
->uuid_root
, 2);
3750 if (IS_ERR(trans
)) {
3751 ret
= PTR_ERR(trans
);
3759 if (!btrfs_is_empty_uuid(root_item
.uuid
)) {
3760 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
3762 BTRFS_UUID_KEY_SUBVOL
,
3765 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
3771 if (!btrfs_is_empty_uuid(root_item
.received_uuid
)) {
3772 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
3773 root_item
.received_uuid
,
3774 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
3777 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
3785 ret
= btrfs_end_transaction(trans
, fs_info
->uuid_root
);
3791 btrfs_release_path(path
);
3792 if (key
.offset
< (u64
)-1) {
3794 } else if (key
.type
< BTRFS_ROOT_ITEM_KEY
) {
3796 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3797 } else if (key
.objectid
< (u64
)-1) {
3799 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3808 btrfs_free_path(path
);
3809 if (trans
&& !IS_ERR(trans
))
3810 btrfs_end_transaction(trans
, fs_info
->uuid_root
);
3812 btrfs_warn(fs_info
, "btrfs_uuid_scan_kthread failed %d", ret
);
3814 fs_info
->update_uuid_tree_gen
= 1;
3815 up(&fs_info
->uuid_tree_rescan_sem
);
3820 * Callback for btrfs_uuid_tree_iterate().
3822 * 0 check succeeded, the entry is not outdated.
3823 * < 0 if an error occured.
3824 * > 0 if the check failed, which means the caller shall remove the entry.
3826 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info
*fs_info
,
3827 u8
*uuid
, u8 type
, u64 subid
)
3829 struct btrfs_key key
;
3831 struct btrfs_root
*subvol_root
;
3833 if (type
!= BTRFS_UUID_KEY_SUBVOL
&&
3834 type
!= BTRFS_UUID_KEY_RECEIVED_SUBVOL
)
3837 key
.objectid
= subid
;
3838 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3839 key
.offset
= (u64
)-1;
3840 subvol_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
3841 if (IS_ERR(subvol_root
)) {
3842 ret
= PTR_ERR(subvol_root
);
3849 case BTRFS_UUID_KEY_SUBVOL
:
3850 if (memcmp(uuid
, subvol_root
->root_item
.uuid
, BTRFS_UUID_SIZE
))
3853 case BTRFS_UUID_KEY_RECEIVED_SUBVOL
:
3854 if (memcmp(uuid
, subvol_root
->root_item
.received_uuid
,
3864 static int btrfs_uuid_rescan_kthread(void *data
)
3866 struct btrfs_fs_info
*fs_info
= (struct btrfs_fs_info
*)data
;
3870 * 1st step is to iterate through the existing UUID tree and
3871 * to delete all entries that contain outdated data.
3872 * 2nd step is to add all missing entries to the UUID tree.
3874 ret
= btrfs_uuid_tree_iterate(fs_info
, btrfs_check_uuid_tree_entry
);
3876 btrfs_warn(fs_info
, "iterating uuid_tree failed %d", ret
);
3877 up(&fs_info
->uuid_tree_rescan_sem
);
3880 return btrfs_uuid_scan_kthread(data
);
3883 int btrfs_create_uuid_tree(struct btrfs_fs_info
*fs_info
)
3885 struct btrfs_trans_handle
*trans
;
3886 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
3887 struct btrfs_root
*uuid_root
;
3888 struct task_struct
*task
;
3895 trans
= btrfs_start_transaction(tree_root
, 2);
3897 return PTR_ERR(trans
);
3899 uuid_root
= btrfs_create_tree(trans
, fs_info
,
3900 BTRFS_UUID_TREE_OBJECTID
);
3901 if (IS_ERR(uuid_root
)) {
3902 btrfs_abort_transaction(trans
, tree_root
,
3903 PTR_ERR(uuid_root
));
3904 return PTR_ERR(uuid_root
);
3907 fs_info
->uuid_root
= uuid_root
;
3909 ret
= btrfs_commit_transaction(trans
, tree_root
);
3913 down(&fs_info
->uuid_tree_rescan_sem
);
3914 task
= kthread_run(btrfs_uuid_scan_kthread
, fs_info
, "btrfs-uuid");
3916 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3917 btrfs_warn(fs_info
, "failed to start uuid_scan task");
3918 up(&fs_info
->uuid_tree_rescan_sem
);
3919 return PTR_ERR(task
);
3925 int btrfs_check_uuid_tree(struct btrfs_fs_info
*fs_info
)
3927 struct task_struct
*task
;
3929 down(&fs_info
->uuid_tree_rescan_sem
);
3930 task
= kthread_run(btrfs_uuid_rescan_kthread
, fs_info
, "btrfs-uuid");
3932 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3933 btrfs_warn(fs_info
, "failed to start uuid_rescan task");
3934 up(&fs_info
->uuid_tree_rescan_sem
);
3935 return PTR_ERR(task
);
3942 * shrinking a device means finding all of the device extents past
3943 * the new size, and then following the back refs to the chunks.
3944 * The chunk relocation code actually frees the device extent
3946 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
3948 struct btrfs_trans_handle
*trans
;
3949 struct btrfs_root
*root
= device
->dev_root
;
3950 struct btrfs_dev_extent
*dev_extent
= NULL
;
3951 struct btrfs_path
*path
;
3958 bool retried
= false;
3959 struct extent_buffer
*l
;
3960 struct btrfs_key key
;
3961 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
3962 u64 old_total
= btrfs_super_total_bytes(super_copy
);
3963 u64 old_size
= btrfs_device_get_total_bytes(device
);
3964 u64 diff
= old_size
- new_size
;
3966 if (device
->is_tgtdev_for_dev_replace
)
3969 path
= btrfs_alloc_path();
3977 btrfs_device_set_total_bytes(device
, new_size
);
3978 if (device
->writeable
) {
3979 device
->fs_devices
->total_rw_bytes
-= diff
;
3980 spin_lock(&root
->fs_info
->free_chunk_lock
);
3981 root
->fs_info
->free_chunk_space
-= diff
;
3982 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3984 unlock_chunks(root
);
3987 key
.objectid
= device
->devid
;
3988 key
.offset
= (u64
)-1;
3989 key
.type
= BTRFS_DEV_EXTENT_KEY
;
3992 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3996 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
4001 btrfs_release_path(path
);
4006 slot
= path
->slots
[0];
4007 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
4009 if (key
.objectid
!= device
->devid
) {
4010 btrfs_release_path(path
);
4014 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
4015 length
= btrfs_dev_extent_length(l
, dev_extent
);
4017 if (key
.offset
+ length
<= new_size
) {
4018 btrfs_release_path(path
);
4022 chunk_objectid
= btrfs_dev_extent_chunk_objectid(l
, dev_extent
);
4023 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
4024 btrfs_release_path(path
);
4026 ret
= btrfs_relocate_chunk(root
, chunk_objectid
, chunk_offset
);
4027 if (ret
&& ret
!= -ENOSPC
)
4031 } while (key
.offset
-- > 0);
4033 if (failed
&& !retried
) {
4037 } else if (failed
&& retried
) {
4041 btrfs_device_set_total_bytes(device
, old_size
);
4042 if (device
->writeable
)
4043 device
->fs_devices
->total_rw_bytes
+= diff
;
4044 spin_lock(&root
->fs_info
->free_chunk_lock
);
4045 root
->fs_info
->free_chunk_space
+= diff
;
4046 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4047 unlock_chunks(root
);
4051 /* Shrinking succeeded, else we would be at "done". */
4052 trans
= btrfs_start_transaction(root
, 0);
4053 if (IS_ERR(trans
)) {
4054 ret
= PTR_ERR(trans
);
4059 btrfs_device_set_disk_total_bytes(device
, new_size
);
4060 if (list_empty(&device
->resized_list
))
4061 list_add_tail(&device
->resized_list
,
4062 &root
->fs_info
->fs_devices
->resized_devices
);
4064 WARN_ON(diff
> old_total
);
4065 btrfs_set_super_total_bytes(super_copy
, old_total
- diff
);
4066 unlock_chunks(root
);
4068 /* Now btrfs_update_device() will change the on-disk size. */
4069 ret
= btrfs_update_device(trans
, device
);
4070 btrfs_end_transaction(trans
, root
);
4072 btrfs_free_path(path
);
4076 static int btrfs_add_system_chunk(struct btrfs_root
*root
,
4077 struct btrfs_key
*key
,
4078 struct btrfs_chunk
*chunk
, int item_size
)
4080 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
4081 struct btrfs_disk_key disk_key
;
4086 array_size
= btrfs_super_sys_array_size(super_copy
);
4087 if (array_size
+ item_size
+ sizeof(disk_key
)
4088 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4089 unlock_chunks(root
);
4093 ptr
= super_copy
->sys_chunk_array
+ array_size
;
4094 btrfs_cpu_key_to_disk(&disk_key
, key
);
4095 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
4096 ptr
+= sizeof(disk_key
);
4097 memcpy(ptr
, chunk
, item_size
);
4098 item_size
+= sizeof(disk_key
);
4099 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
4100 unlock_chunks(root
);
4106 * sort the devices in descending order by max_avail, total_avail
4108 static int btrfs_cmp_device_info(const void *a
, const void *b
)
4110 const struct btrfs_device_info
*di_a
= a
;
4111 const struct btrfs_device_info
*di_b
= b
;
4113 if (di_a
->max_avail
> di_b
->max_avail
)
4115 if (di_a
->max_avail
< di_b
->max_avail
)
4117 if (di_a
->total_avail
> di_b
->total_avail
)
4119 if (di_a
->total_avail
< di_b
->total_avail
)
4124 static const struct btrfs_raid_attr btrfs_raid_array
[BTRFS_NR_RAID_TYPES
] = {
4125 [BTRFS_RAID_RAID10
] = {
4128 .devs_max
= 0, /* 0 == as many as possible */
4130 .devs_increment
= 2,
4133 [BTRFS_RAID_RAID1
] = {
4138 .devs_increment
= 2,
4141 [BTRFS_RAID_DUP
] = {
4146 .devs_increment
= 1,
4149 [BTRFS_RAID_RAID0
] = {
4154 .devs_increment
= 1,
4157 [BTRFS_RAID_SINGLE
] = {
4162 .devs_increment
= 1,
4165 [BTRFS_RAID_RAID5
] = {
4170 .devs_increment
= 1,
4173 [BTRFS_RAID_RAID6
] = {
4178 .devs_increment
= 1,
4183 static u32
find_raid56_stripe_len(u32 data_devices
, u32 dev_stripe_target
)
4185 /* TODO allow them to set a preferred stripe size */
4189 static void check_raid56_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
4191 if (!(type
& BTRFS_BLOCK_GROUP_RAID56_MASK
))
4194 btrfs_set_fs_incompat(info
, RAID56
);
4197 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4198 - sizeof(struct btrfs_item) \
4199 - sizeof(struct btrfs_chunk)) \
4200 / sizeof(struct btrfs_stripe) + 1)
4202 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4203 - 2 * sizeof(struct btrfs_disk_key) \
4204 - 2 * sizeof(struct btrfs_chunk)) \
4205 / sizeof(struct btrfs_stripe) + 1)
4207 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4208 struct btrfs_root
*extent_root
, u64 start
,
4211 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
4212 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
4213 struct list_head
*cur
;
4214 struct map_lookup
*map
= NULL
;
4215 struct extent_map_tree
*em_tree
;
4216 struct extent_map
*em
;
4217 struct btrfs_device_info
*devices_info
= NULL
;
4219 int num_stripes
; /* total number of stripes to allocate */
4220 int data_stripes
; /* number of stripes that count for
4222 int sub_stripes
; /* sub_stripes info for map */
4223 int dev_stripes
; /* stripes per dev */
4224 int devs_max
; /* max devs to use */
4225 int devs_min
; /* min devs needed */
4226 int devs_increment
; /* ndevs has to be a multiple of this */
4227 int ncopies
; /* how many copies to data has */
4229 u64 max_stripe_size
;
4233 u64 raid_stripe_len
= BTRFS_STRIPE_LEN
;
4239 BUG_ON(!alloc_profile_is_valid(type
, 0));
4241 if (list_empty(&fs_devices
->alloc_list
))
4244 index
= __get_raid_index(type
);
4246 sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
4247 dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
4248 devs_max
= btrfs_raid_array
[index
].devs_max
;
4249 devs_min
= btrfs_raid_array
[index
].devs_min
;
4250 devs_increment
= btrfs_raid_array
[index
].devs_increment
;
4251 ncopies
= btrfs_raid_array
[index
].ncopies
;
4253 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
4254 max_stripe_size
= 1024 * 1024 * 1024;
4255 max_chunk_size
= 10 * max_stripe_size
;
4257 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4258 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
4259 /* for larger filesystems, use larger metadata chunks */
4260 if (fs_devices
->total_rw_bytes
> 50ULL * 1024 * 1024 * 1024)
4261 max_stripe_size
= 1024 * 1024 * 1024;
4263 max_stripe_size
= 256 * 1024 * 1024;
4264 max_chunk_size
= max_stripe_size
;
4266 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4267 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4268 max_stripe_size
= 32 * 1024 * 1024;
4269 max_chunk_size
= 2 * max_stripe_size
;
4271 devs_max
= BTRFS_MAX_DEVS_SYS_CHUNK
;
4273 btrfs_err(info
, "invalid chunk type 0x%llx requested",
4278 /* we don't want a chunk larger than 10% of writeable space */
4279 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
4282 devices_info
= kcalloc(fs_devices
->rw_devices
, sizeof(*devices_info
),
4287 cur
= fs_devices
->alloc_list
.next
;
4290 * in the first pass through the devices list, we gather information
4291 * about the available holes on each device.
4294 while (cur
!= &fs_devices
->alloc_list
) {
4295 struct btrfs_device
*device
;
4299 device
= list_entry(cur
, struct btrfs_device
, dev_alloc_list
);
4303 if (!device
->writeable
) {
4305 "BTRFS: read-only device in alloc_list\n");
4309 if (!device
->in_fs_metadata
||
4310 device
->is_tgtdev_for_dev_replace
)
4313 if (device
->total_bytes
> device
->bytes_used
)
4314 total_avail
= device
->total_bytes
- device
->bytes_used
;
4318 /* If there is no space on this device, skip it. */
4319 if (total_avail
== 0)
4322 ret
= find_free_dev_extent(trans
, device
,
4323 max_stripe_size
* dev_stripes
,
4324 &dev_offset
, &max_avail
);
4325 if (ret
&& ret
!= -ENOSPC
)
4329 max_avail
= max_stripe_size
* dev_stripes
;
4331 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
4334 if (ndevs
== fs_devices
->rw_devices
) {
4335 WARN(1, "%s: found more than %llu devices\n",
4336 __func__
, fs_devices
->rw_devices
);
4339 devices_info
[ndevs
].dev_offset
= dev_offset
;
4340 devices_info
[ndevs
].max_avail
= max_avail
;
4341 devices_info
[ndevs
].total_avail
= total_avail
;
4342 devices_info
[ndevs
].dev
= device
;
4347 * now sort the devices by hole size / available space
4349 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
4350 btrfs_cmp_device_info
, NULL
);
4352 /* round down to number of usable stripes */
4353 ndevs
-= ndevs
% devs_increment
;
4355 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
4360 if (devs_max
&& ndevs
> devs_max
)
4363 * the primary goal is to maximize the number of stripes, so use as many
4364 * devices as possible, even if the stripes are not maximum sized.
4366 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4367 num_stripes
= ndevs
* dev_stripes
;
4370 * this will have to be fixed for RAID1 and RAID10 over
4373 data_stripes
= num_stripes
/ ncopies
;
4375 if (type
& BTRFS_BLOCK_GROUP_RAID5
) {
4376 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 1,
4377 btrfs_super_stripesize(info
->super_copy
));
4378 data_stripes
= num_stripes
- 1;
4380 if (type
& BTRFS_BLOCK_GROUP_RAID6
) {
4381 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 2,
4382 btrfs_super_stripesize(info
->super_copy
));
4383 data_stripes
= num_stripes
- 2;
4387 * Use the number of data stripes to figure out how big this chunk
4388 * is really going to be in terms of logical address space,
4389 * and compare that answer with the max chunk size
4391 if (stripe_size
* data_stripes
> max_chunk_size
) {
4392 u64 mask
= (1ULL << 24) - 1;
4394 stripe_size
= div_u64(max_chunk_size
, data_stripes
);
4396 /* bump the answer up to a 16MB boundary */
4397 stripe_size
= (stripe_size
+ mask
) & ~mask
;
4399 /* but don't go higher than the limits we found
4400 * while searching for free extents
4402 if (stripe_size
> devices_info
[ndevs
-1].max_avail
)
4403 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4406 stripe_size
= div_u64(stripe_size
, dev_stripes
);
4408 /* align to BTRFS_STRIPE_LEN */
4409 stripe_size
= div_u64(stripe_size
, raid_stripe_len
);
4410 stripe_size
*= raid_stripe_len
;
4412 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
4417 map
->num_stripes
= num_stripes
;
4419 for (i
= 0; i
< ndevs
; ++i
) {
4420 for (j
= 0; j
< dev_stripes
; ++j
) {
4421 int s
= i
* dev_stripes
+ j
;
4422 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
4423 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
4427 map
->sector_size
= extent_root
->sectorsize
;
4428 map
->stripe_len
= raid_stripe_len
;
4429 map
->io_align
= raid_stripe_len
;
4430 map
->io_width
= raid_stripe_len
;
4432 map
->sub_stripes
= sub_stripes
;
4434 num_bytes
= stripe_size
* data_stripes
;
4436 trace_btrfs_chunk_alloc(info
->chunk_root
, map
, start
, num_bytes
);
4438 em
= alloc_extent_map();
4444 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
4445 em
->bdev
= (struct block_device
*)map
;
4447 em
->len
= num_bytes
;
4448 em
->block_start
= 0;
4449 em
->block_len
= em
->len
;
4450 em
->orig_block_len
= stripe_size
;
4452 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
4453 write_lock(&em_tree
->lock
);
4454 ret
= add_extent_mapping(em_tree
, em
, 0);
4456 list_add_tail(&em
->list
, &trans
->transaction
->pending_chunks
);
4457 atomic_inc(&em
->refs
);
4459 write_unlock(&em_tree
->lock
);
4461 free_extent_map(em
);
4465 ret
= btrfs_make_block_group(trans
, extent_root
, 0, type
,
4466 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4469 goto error_del_extent
;
4471 for (i
= 0; i
< map
->num_stripes
; i
++) {
4472 num_bytes
= map
->stripes
[i
].dev
->bytes_used
+ stripe_size
;
4473 btrfs_device_set_bytes_used(map
->stripes
[i
].dev
, num_bytes
);
4476 spin_lock(&extent_root
->fs_info
->free_chunk_lock
);
4477 extent_root
->fs_info
->free_chunk_space
-= (stripe_size
*
4479 spin_unlock(&extent_root
->fs_info
->free_chunk_lock
);
4481 free_extent_map(em
);
4482 check_raid56_incompat_flag(extent_root
->fs_info
, type
);
4484 kfree(devices_info
);
4488 write_lock(&em_tree
->lock
);
4489 remove_extent_mapping(em_tree
, em
);
4490 write_unlock(&em_tree
->lock
);
4492 /* One for our allocation */
4493 free_extent_map(em
);
4494 /* One for the tree reference */
4495 free_extent_map(em
);
4496 /* One for the pending_chunks list reference */
4497 free_extent_map(em
);
4499 kfree(devices_info
);
4503 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
4504 struct btrfs_root
*extent_root
,
4505 u64 chunk_offset
, u64 chunk_size
)
4507 struct btrfs_key key
;
4508 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
4509 struct btrfs_device
*device
;
4510 struct btrfs_chunk
*chunk
;
4511 struct btrfs_stripe
*stripe
;
4512 struct extent_map_tree
*em_tree
;
4513 struct extent_map
*em
;
4514 struct map_lookup
*map
;
4521 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
4522 read_lock(&em_tree
->lock
);
4523 em
= lookup_extent_mapping(em_tree
, chunk_offset
, chunk_size
);
4524 read_unlock(&em_tree
->lock
);
4527 btrfs_crit(extent_root
->fs_info
, "unable to find logical "
4528 "%Lu len %Lu", chunk_offset
, chunk_size
);
4532 if (em
->start
!= chunk_offset
|| em
->len
!= chunk_size
) {
4533 btrfs_crit(extent_root
->fs_info
, "found a bad mapping, wanted"
4534 " %Lu-%Lu, found %Lu-%Lu", chunk_offset
,
4535 chunk_size
, em
->start
, em
->len
);
4536 free_extent_map(em
);
4540 map
= (struct map_lookup
*)em
->bdev
;
4541 item_size
= btrfs_chunk_item_size(map
->num_stripes
);
4542 stripe_size
= em
->orig_block_len
;
4544 chunk
= kzalloc(item_size
, GFP_NOFS
);
4550 for (i
= 0; i
< map
->num_stripes
; i
++) {
4551 device
= map
->stripes
[i
].dev
;
4552 dev_offset
= map
->stripes
[i
].physical
;
4554 ret
= btrfs_update_device(trans
, device
);
4557 ret
= btrfs_alloc_dev_extent(trans
, device
,
4558 chunk_root
->root_key
.objectid
,
4559 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4560 chunk_offset
, dev_offset
,
4566 stripe
= &chunk
->stripe
;
4567 for (i
= 0; i
< map
->num_stripes
; i
++) {
4568 device
= map
->stripes
[i
].dev
;
4569 dev_offset
= map
->stripes
[i
].physical
;
4571 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
4572 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
4573 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
4577 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
4578 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
4579 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
4580 btrfs_set_stack_chunk_type(chunk
, map
->type
);
4581 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
4582 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
4583 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
4584 btrfs_set_stack_chunk_sector_size(chunk
, extent_root
->sectorsize
);
4585 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
4587 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
4588 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
4589 key
.offset
= chunk_offset
;
4591 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
4592 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4594 * TODO: Cleanup of inserted chunk root in case of
4597 ret
= btrfs_add_system_chunk(chunk_root
, &key
, chunk
,
4603 free_extent_map(em
);
4608 * Chunk allocation falls into two parts. The first part does works
4609 * that make the new allocated chunk useable, but not do any operation
4610 * that modifies the chunk tree. The second part does the works that
4611 * require modifying the chunk tree. This division is important for the
4612 * bootstrap process of adding storage to a seed btrfs.
4614 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4615 struct btrfs_root
*extent_root
, u64 type
)
4619 chunk_offset
= find_next_chunk(extent_root
->fs_info
);
4620 return __btrfs_alloc_chunk(trans
, extent_root
, chunk_offset
, type
);
4623 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
4624 struct btrfs_root
*root
,
4625 struct btrfs_device
*device
)
4628 u64 sys_chunk_offset
;
4630 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4631 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
4634 chunk_offset
= find_next_chunk(fs_info
);
4635 alloc_profile
= btrfs_get_alloc_profile(extent_root
, 0);
4636 ret
= __btrfs_alloc_chunk(trans
, extent_root
, chunk_offset
,
4641 sys_chunk_offset
= find_next_chunk(root
->fs_info
);
4642 alloc_profile
= btrfs_get_alloc_profile(fs_info
->chunk_root
, 0);
4643 ret
= __btrfs_alloc_chunk(trans
, extent_root
, sys_chunk_offset
,
4648 static inline int btrfs_chunk_max_errors(struct map_lookup
*map
)
4652 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
4653 BTRFS_BLOCK_GROUP_RAID10
|
4654 BTRFS_BLOCK_GROUP_RAID5
|
4655 BTRFS_BLOCK_GROUP_DUP
)) {
4657 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
) {
4666 int btrfs_chunk_readonly(struct btrfs_root
*root
, u64 chunk_offset
)
4668 struct extent_map
*em
;
4669 struct map_lookup
*map
;
4670 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
4675 read_lock(&map_tree
->map_tree
.lock
);
4676 em
= lookup_extent_mapping(&map_tree
->map_tree
, chunk_offset
, 1);
4677 read_unlock(&map_tree
->map_tree
.lock
);
4681 map
= (struct map_lookup
*)em
->bdev
;
4682 for (i
= 0; i
< map
->num_stripes
; i
++) {
4683 if (map
->stripes
[i
].dev
->missing
) {
4688 if (!map
->stripes
[i
].dev
->writeable
) {
4695 * If the number of missing devices is larger than max errors,
4696 * we can not write the data into that chunk successfully, so
4699 if (miss_ndevs
> btrfs_chunk_max_errors(map
))
4702 free_extent_map(em
);
4706 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
4708 extent_map_tree_init(&tree
->map_tree
);
4711 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
4713 struct extent_map
*em
;
4716 write_lock(&tree
->map_tree
.lock
);
4717 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
4719 remove_extent_mapping(&tree
->map_tree
, em
);
4720 write_unlock(&tree
->map_tree
.lock
);
4724 free_extent_map(em
);
4725 /* once for the tree */
4726 free_extent_map(em
);
4730 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
4732 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
4733 struct extent_map
*em
;
4734 struct map_lookup
*map
;
4735 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4738 read_lock(&em_tree
->lock
);
4739 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4740 read_unlock(&em_tree
->lock
);
4743 * We could return errors for these cases, but that could get ugly and
4744 * we'd probably do the same thing which is just not do anything else
4745 * and exit, so return 1 so the callers don't try to use other copies.
4748 btrfs_crit(fs_info
, "No mapping for %Lu-%Lu", logical
,
4753 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
4754 btrfs_crit(fs_info
, "Invalid mapping for %Lu-%Lu, got "
4755 "%Lu-%Lu", logical
, logical
+len
, em
->start
,
4756 em
->start
+ em
->len
);
4757 free_extent_map(em
);
4761 map
= (struct map_lookup
*)em
->bdev
;
4762 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
4763 ret
= map
->num_stripes
;
4764 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
4765 ret
= map
->sub_stripes
;
4766 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID5
)
4768 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
4772 free_extent_map(em
);
4774 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
4775 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
))
4777 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
4782 unsigned long btrfs_full_stripe_len(struct btrfs_root
*root
,
4783 struct btrfs_mapping_tree
*map_tree
,
4786 struct extent_map
*em
;
4787 struct map_lookup
*map
;
4788 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4789 unsigned long len
= root
->sectorsize
;
4791 read_lock(&em_tree
->lock
);
4792 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4793 read_unlock(&em_tree
->lock
);
4796 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4797 map
= (struct map_lookup
*)em
->bdev
;
4798 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
4799 len
= map
->stripe_len
* nr_data_stripes(map
);
4800 free_extent_map(em
);
4804 int btrfs_is_parity_mirror(struct btrfs_mapping_tree
*map_tree
,
4805 u64 logical
, u64 len
, int mirror_num
)
4807 struct extent_map
*em
;
4808 struct map_lookup
*map
;
4809 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4812 read_lock(&em_tree
->lock
);
4813 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4814 read_unlock(&em_tree
->lock
);
4817 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4818 map
= (struct map_lookup
*)em
->bdev
;
4819 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
4821 free_extent_map(em
);
4825 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
4826 struct map_lookup
*map
, int first
, int num
,
4827 int optimal
, int dev_replace_is_ongoing
)
4831 struct btrfs_device
*srcdev
;
4833 if (dev_replace_is_ongoing
&&
4834 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
4835 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
4836 srcdev
= fs_info
->dev_replace
.srcdev
;
4841 * try to avoid the drive that is the source drive for a
4842 * dev-replace procedure, only choose it if no other non-missing
4843 * mirror is available
4845 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
4846 if (map
->stripes
[optimal
].dev
->bdev
&&
4847 (tolerance
|| map
->stripes
[optimal
].dev
!= srcdev
))
4849 for (i
= first
; i
< first
+ num
; i
++) {
4850 if (map
->stripes
[i
].dev
->bdev
&&
4851 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
4856 /* we couldn't find one that doesn't fail. Just return something
4857 * and the io error handling code will clean up eventually
4862 static inline int parity_smaller(u64 a
, u64 b
)
4867 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4868 static void sort_parity_stripes(struct btrfs_bio
*bbio
, int num_stripes
)
4870 struct btrfs_bio_stripe s
;
4877 for (i
= 0; i
< num_stripes
- 1; i
++) {
4878 if (parity_smaller(bbio
->raid_map
[i
],
4879 bbio
->raid_map
[i
+1])) {
4880 s
= bbio
->stripes
[i
];
4881 l
= bbio
->raid_map
[i
];
4882 bbio
->stripes
[i
] = bbio
->stripes
[i
+1];
4883 bbio
->raid_map
[i
] = bbio
->raid_map
[i
+1];
4884 bbio
->stripes
[i
+1] = s
;
4885 bbio
->raid_map
[i
+1] = l
;
4893 static struct btrfs_bio
*alloc_btrfs_bio(int total_stripes
, int real_stripes
)
4895 struct btrfs_bio
*bbio
= kzalloc(
4896 /* the size of the btrfs_bio */
4897 sizeof(struct btrfs_bio
) +
4898 /* plus the variable array for the stripes */
4899 sizeof(struct btrfs_bio_stripe
) * (total_stripes
) +
4900 /* plus the variable array for the tgt dev */
4901 sizeof(int) * (real_stripes
) +
4903 * plus the raid_map, which includes both the tgt dev
4906 sizeof(u64
) * (total_stripes
),
4911 atomic_set(&bbio
->error
, 0);
4912 atomic_set(&bbio
->refs
, 1);
4917 void btrfs_get_bbio(struct btrfs_bio
*bbio
)
4919 WARN_ON(!atomic_read(&bbio
->refs
));
4920 atomic_inc(&bbio
->refs
);
4923 void btrfs_put_bbio(struct btrfs_bio
*bbio
)
4927 if (atomic_dec_and_test(&bbio
->refs
))
4931 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
4932 u64 logical
, u64
*length
,
4933 struct btrfs_bio
**bbio_ret
,
4934 int mirror_num
, int need_raid_map
)
4936 struct extent_map
*em
;
4937 struct map_lookup
*map
;
4938 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
4939 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4942 u64 stripe_end_offset
;
4952 int tgtdev_indexes
= 0;
4953 struct btrfs_bio
*bbio
= NULL
;
4954 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
4955 int dev_replace_is_ongoing
= 0;
4956 int num_alloc_stripes
;
4957 int patch_the_first_stripe_for_dev_replace
= 0;
4958 u64 physical_to_patch_in_first_stripe
= 0;
4959 u64 raid56_full_stripe_start
= (u64
)-1;
4961 read_lock(&em_tree
->lock
);
4962 em
= lookup_extent_mapping(em_tree
, logical
, *length
);
4963 read_unlock(&em_tree
->lock
);
4966 btrfs_crit(fs_info
, "unable to find logical %llu len %llu",
4971 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
4972 btrfs_crit(fs_info
, "found a bad mapping, wanted %Lu, "
4973 "found %Lu-%Lu", logical
, em
->start
,
4974 em
->start
+ em
->len
);
4975 free_extent_map(em
);
4979 map
= (struct map_lookup
*)em
->bdev
;
4980 offset
= logical
- em
->start
;
4982 stripe_len
= map
->stripe_len
;
4985 * stripe_nr counts the total number of stripes we have to stride
4986 * to get to this block
4988 stripe_nr
= div64_u64(stripe_nr
, stripe_len
);
4990 stripe_offset
= stripe_nr
* stripe_len
;
4991 BUG_ON(offset
< stripe_offset
);
4993 /* stripe_offset is the offset of this block in its stripe*/
4994 stripe_offset
= offset
- stripe_offset
;
4996 /* if we're here for raid56, we need to know the stripe aligned start */
4997 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
4998 unsigned long full_stripe_len
= stripe_len
* nr_data_stripes(map
);
4999 raid56_full_stripe_start
= offset
;
5001 /* allow a write of a full stripe, but make sure we don't
5002 * allow straddling of stripes
5004 raid56_full_stripe_start
= div64_u64(raid56_full_stripe_start
,
5006 raid56_full_stripe_start
*= full_stripe_len
;
5009 if (rw
& REQ_DISCARD
) {
5010 /* we don't discard raid56 yet */
5011 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5015 *length
= min_t(u64
, em
->len
- offset
, *length
);
5016 } else if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
5018 /* For writes to RAID[56], allow a full stripeset across all disks.
5019 For other RAID types and for RAID[56] reads, just allow a single
5020 stripe (on a single disk). */
5021 if ((map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
5023 max_len
= stripe_len
* nr_data_stripes(map
) -
5024 (offset
- raid56_full_stripe_start
);
5026 /* we limit the length of each bio to what fits in a stripe */
5027 max_len
= stripe_len
- stripe_offset
;
5029 *length
= min_t(u64
, em
->len
- offset
, max_len
);
5031 *length
= em
->len
- offset
;
5034 /* This is for when we're called from btrfs_merge_bio_hook() and all
5035 it cares about is the length */
5039 btrfs_dev_replace_lock(dev_replace
);
5040 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
5041 if (!dev_replace_is_ongoing
)
5042 btrfs_dev_replace_unlock(dev_replace
);
5044 if (dev_replace_is_ongoing
&& mirror_num
== map
->num_stripes
+ 1 &&
5045 !(rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) &&
5046 dev_replace
->tgtdev
!= NULL
) {
5048 * in dev-replace case, for repair case (that's the only
5049 * case where the mirror is selected explicitly when
5050 * calling btrfs_map_block), blocks left of the left cursor
5051 * can also be read from the target drive.
5052 * For REQ_GET_READ_MIRRORS, the target drive is added as
5053 * the last one to the array of stripes. For READ, it also
5054 * needs to be supported using the same mirror number.
5055 * If the requested block is not left of the left cursor,
5056 * EIO is returned. This can happen because btrfs_num_copies()
5057 * returns one more in the dev-replace case.
5059 u64 tmp_length
= *length
;
5060 struct btrfs_bio
*tmp_bbio
= NULL
;
5061 int tmp_num_stripes
;
5062 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5063 int index_srcdev
= 0;
5065 u64 physical_of_found
= 0;
5067 ret
= __btrfs_map_block(fs_info
, REQ_GET_READ_MIRRORS
,
5068 logical
, &tmp_length
, &tmp_bbio
, 0, 0);
5070 WARN_ON(tmp_bbio
!= NULL
);
5074 tmp_num_stripes
= tmp_bbio
->num_stripes
;
5075 if (mirror_num
> tmp_num_stripes
) {
5077 * REQ_GET_READ_MIRRORS does not contain this
5078 * mirror, that means that the requested area
5079 * is not left of the left cursor
5082 btrfs_put_bbio(tmp_bbio
);
5087 * process the rest of the function using the mirror_num
5088 * of the source drive. Therefore look it up first.
5089 * At the end, patch the device pointer to the one of the
5092 for (i
= 0; i
< tmp_num_stripes
; i
++) {
5093 if (tmp_bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5095 * In case of DUP, in order to keep it
5096 * simple, only add the mirror with the
5097 * lowest physical address
5100 physical_of_found
<=
5101 tmp_bbio
->stripes
[i
].physical
)
5106 tmp_bbio
->stripes
[i
].physical
;
5111 mirror_num
= index_srcdev
+ 1;
5112 patch_the_first_stripe_for_dev_replace
= 1;
5113 physical_to_patch_in_first_stripe
= physical_of_found
;
5117 btrfs_put_bbio(tmp_bbio
);
5121 btrfs_put_bbio(tmp_bbio
);
5122 } else if (mirror_num
> map
->num_stripes
) {
5128 stripe_nr_orig
= stripe_nr
;
5129 stripe_nr_end
= ALIGN(offset
+ *length
, map
->stripe_len
);
5130 stripe_nr_end
= div_u64(stripe_nr_end
, map
->stripe_len
);
5131 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
5134 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5135 if (rw
& REQ_DISCARD
)
5136 num_stripes
= min_t(u64
, map
->num_stripes
,
5137 stripe_nr_end
- stripe_nr_orig
);
5138 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5140 if (!(rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)))
5142 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
5143 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
))
5144 num_stripes
= map
->num_stripes
;
5145 else if (mirror_num
)
5146 stripe_index
= mirror_num
- 1;
5148 stripe_index
= find_live_mirror(fs_info
, map
, 0,
5150 current
->pid
% map
->num_stripes
,
5151 dev_replace_is_ongoing
);
5152 mirror_num
= stripe_index
+ 1;
5155 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
5156 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) {
5157 num_stripes
= map
->num_stripes
;
5158 } else if (mirror_num
) {
5159 stripe_index
= mirror_num
- 1;
5164 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5165 u32 factor
= map
->num_stripes
/ map
->sub_stripes
;
5167 stripe_nr
= div_u64_rem(stripe_nr
, factor
, &stripe_index
);
5168 stripe_index
*= map
->sub_stripes
;
5170 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
))
5171 num_stripes
= map
->sub_stripes
;
5172 else if (rw
& REQ_DISCARD
)
5173 num_stripes
= min_t(u64
, map
->sub_stripes
*
5174 (stripe_nr_end
- stripe_nr_orig
),
5176 else if (mirror_num
)
5177 stripe_index
+= mirror_num
- 1;
5179 int old_stripe_index
= stripe_index
;
5180 stripe_index
= find_live_mirror(fs_info
, map
,
5182 map
->sub_stripes
, stripe_index
+
5183 current
->pid
% map
->sub_stripes
,
5184 dev_replace_is_ongoing
);
5185 mirror_num
= stripe_index
- old_stripe_index
+ 1;
5188 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5189 if (need_raid_map
&&
5190 ((rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
)) ||
5192 /* push stripe_nr back to the start of the full stripe */
5193 stripe_nr
= div_u64(raid56_full_stripe_start
,
5194 stripe_len
* nr_data_stripes(map
));
5196 /* RAID[56] write or recovery. Return all stripes */
5197 num_stripes
= map
->num_stripes
;
5198 max_errors
= nr_parity_stripes(map
);
5200 *length
= map
->stripe_len
;
5205 * Mirror #0 or #1 means the original data block.
5206 * Mirror #2 is RAID5 parity block.
5207 * Mirror #3 is RAID6 Q block.
5209 stripe_nr
= div_u64_rem(stripe_nr
,
5210 nr_data_stripes(map
), &stripe_index
);
5212 stripe_index
= nr_data_stripes(map
) +
5215 /* We distribute the parity blocks across stripes */
5216 div_u64_rem(stripe_nr
+ stripe_index
, map
->num_stripes
,
5218 if (!(rw
& (REQ_WRITE
| REQ_DISCARD
|
5219 REQ_GET_READ_MIRRORS
)) && mirror_num
<= 1)
5224 * after this, stripe_nr is the number of stripes on this
5225 * device we have to walk to find the data, and stripe_index is
5226 * the number of our device in the stripe array
5228 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5230 mirror_num
= stripe_index
+ 1;
5232 BUG_ON(stripe_index
>= map
->num_stripes
);
5234 num_alloc_stripes
= num_stripes
;
5235 if (dev_replace_is_ongoing
) {
5236 if (rw
& (REQ_WRITE
| REQ_DISCARD
))
5237 num_alloc_stripes
<<= 1;
5238 if (rw
& REQ_GET_READ_MIRRORS
)
5239 num_alloc_stripes
++;
5240 tgtdev_indexes
= num_stripes
;
5243 bbio
= alloc_btrfs_bio(num_alloc_stripes
, tgtdev_indexes
);
5248 if (dev_replace_is_ongoing
)
5249 bbio
->tgtdev_map
= (int *)(bbio
->stripes
+ num_alloc_stripes
);
5251 /* build raid_map */
5252 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
&&
5253 need_raid_map
&& ((rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
)) ||
5258 bbio
->raid_map
= (u64
*)((void *)bbio
->stripes
+
5259 sizeof(struct btrfs_bio_stripe
) *
5261 sizeof(int) * tgtdev_indexes
);
5263 /* Work out the disk rotation on this stripe-set */
5264 div_u64_rem(stripe_nr
, num_stripes
, &rot
);
5266 /* Fill in the logical address of each stripe */
5267 tmp
= stripe_nr
* nr_data_stripes(map
);
5268 for (i
= 0; i
< nr_data_stripes(map
); i
++)
5269 bbio
->raid_map
[(i
+rot
) % num_stripes
] =
5270 em
->start
+ (tmp
+ i
) * map
->stripe_len
;
5272 bbio
->raid_map
[(i
+rot
) % map
->num_stripes
] = RAID5_P_STRIPE
;
5273 if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5274 bbio
->raid_map
[(i
+rot
+1) % num_stripes
] =
5278 if (rw
& REQ_DISCARD
) {
5280 u32 sub_stripes
= 0;
5281 u64 stripes_per_dev
= 0;
5282 u32 remaining_stripes
= 0;
5283 u32 last_stripe
= 0;
5286 (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID10
)) {
5287 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5290 sub_stripes
= map
->sub_stripes
;
5292 factor
= map
->num_stripes
/ sub_stripes
;
5293 stripes_per_dev
= div_u64_rem(stripe_nr_end
-
5296 &remaining_stripes
);
5297 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
5298 last_stripe
*= sub_stripes
;
5301 for (i
= 0; i
< num_stripes
; i
++) {
5302 bbio
->stripes
[i
].physical
=
5303 map
->stripes
[stripe_index
].physical
+
5304 stripe_offset
+ stripe_nr
* map
->stripe_len
;
5305 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
5307 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5308 BTRFS_BLOCK_GROUP_RAID10
)) {
5309 bbio
->stripes
[i
].length
= stripes_per_dev
*
5312 if (i
/ sub_stripes
< remaining_stripes
)
5313 bbio
->stripes
[i
].length
+=
5317 * Special for the first stripe and
5320 * |-------|...|-------|
5324 if (i
< sub_stripes
)
5325 bbio
->stripes
[i
].length
-=
5328 if (stripe_index
>= last_stripe
&&
5329 stripe_index
<= (last_stripe
+
5331 bbio
->stripes
[i
].length
-=
5334 if (i
== sub_stripes
- 1)
5337 bbio
->stripes
[i
].length
= *length
;
5340 if (stripe_index
== map
->num_stripes
) {
5341 /* This could only happen for RAID0/10 */
5347 for (i
= 0; i
< num_stripes
; i
++) {
5348 bbio
->stripes
[i
].physical
=
5349 map
->stripes
[stripe_index
].physical
+
5351 stripe_nr
* map
->stripe_len
;
5352 bbio
->stripes
[i
].dev
=
5353 map
->stripes
[stripe_index
].dev
;
5358 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
))
5359 max_errors
= btrfs_chunk_max_errors(map
);
5362 sort_parity_stripes(bbio
, num_stripes
);
5365 if (dev_replace_is_ongoing
&& (rw
& (REQ_WRITE
| REQ_DISCARD
)) &&
5366 dev_replace
->tgtdev
!= NULL
) {
5367 int index_where_to_add
;
5368 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5371 * duplicate the write operations while the dev replace
5372 * procedure is running. Since the copying of the old disk
5373 * to the new disk takes place at run time while the
5374 * filesystem is mounted writable, the regular write
5375 * operations to the old disk have to be duplicated to go
5376 * to the new disk as well.
5377 * Note that device->missing is handled by the caller, and
5378 * that the write to the old disk is already set up in the
5381 index_where_to_add
= num_stripes
;
5382 for (i
= 0; i
< num_stripes
; i
++) {
5383 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5384 /* write to new disk, too */
5385 struct btrfs_bio_stripe
*new =
5386 bbio
->stripes
+ index_where_to_add
;
5387 struct btrfs_bio_stripe
*old
=
5390 new->physical
= old
->physical
;
5391 new->length
= old
->length
;
5392 new->dev
= dev_replace
->tgtdev
;
5393 bbio
->tgtdev_map
[i
] = index_where_to_add
;
5394 index_where_to_add
++;
5399 num_stripes
= index_where_to_add
;
5400 } else if (dev_replace_is_ongoing
&& (rw
& REQ_GET_READ_MIRRORS
) &&
5401 dev_replace
->tgtdev
!= NULL
) {
5402 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5403 int index_srcdev
= 0;
5405 u64 physical_of_found
= 0;
5408 * During the dev-replace procedure, the target drive can
5409 * also be used to read data in case it is needed to repair
5410 * a corrupt block elsewhere. This is possible if the
5411 * requested area is left of the left cursor. In this area,
5412 * the target drive is a full copy of the source drive.
5414 for (i
= 0; i
< num_stripes
; i
++) {
5415 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5417 * In case of DUP, in order to keep it
5418 * simple, only add the mirror with the
5419 * lowest physical address
5422 physical_of_found
<=
5423 bbio
->stripes
[i
].physical
)
5427 physical_of_found
= bbio
->stripes
[i
].physical
;
5431 if (physical_of_found
+ map
->stripe_len
<=
5432 dev_replace
->cursor_left
) {
5433 struct btrfs_bio_stripe
*tgtdev_stripe
=
5434 bbio
->stripes
+ num_stripes
;
5436 tgtdev_stripe
->physical
= physical_of_found
;
5437 tgtdev_stripe
->length
=
5438 bbio
->stripes
[index_srcdev
].length
;
5439 tgtdev_stripe
->dev
= dev_replace
->tgtdev
;
5440 bbio
->tgtdev_map
[index_srcdev
] = num_stripes
;
5449 bbio
->map_type
= map
->type
;
5450 bbio
->num_stripes
= num_stripes
;
5451 bbio
->max_errors
= max_errors
;
5452 bbio
->mirror_num
= mirror_num
;
5453 bbio
->num_tgtdevs
= tgtdev_indexes
;
5456 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5457 * mirror_num == num_stripes + 1 && dev_replace target drive is
5458 * available as a mirror
5460 if (patch_the_first_stripe_for_dev_replace
&& num_stripes
> 0) {
5461 WARN_ON(num_stripes
> 1);
5462 bbio
->stripes
[0].dev
= dev_replace
->tgtdev
;
5463 bbio
->stripes
[0].physical
= physical_to_patch_in_first_stripe
;
5464 bbio
->mirror_num
= map
->num_stripes
+ 1;
5467 if (dev_replace_is_ongoing
)
5468 btrfs_dev_replace_unlock(dev_replace
);
5469 free_extent_map(em
);
5473 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
5474 u64 logical
, u64
*length
,
5475 struct btrfs_bio
**bbio_ret
, int mirror_num
)
5477 return __btrfs_map_block(fs_info
, rw
, logical
, length
, bbio_ret
,
5481 /* For Scrub/replace */
5482 int btrfs_map_sblock(struct btrfs_fs_info
*fs_info
, int rw
,
5483 u64 logical
, u64
*length
,
5484 struct btrfs_bio
**bbio_ret
, int mirror_num
,
5487 return __btrfs_map_block(fs_info
, rw
, logical
, length
, bbio_ret
,
5488 mirror_num
, need_raid_map
);
5491 int btrfs_rmap_block(struct btrfs_mapping_tree
*map_tree
,
5492 u64 chunk_start
, u64 physical
, u64 devid
,
5493 u64
**logical
, int *naddrs
, int *stripe_len
)
5495 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5496 struct extent_map
*em
;
5497 struct map_lookup
*map
;
5505 read_lock(&em_tree
->lock
);
5506 em
= lookup_extent_mapping(em_tree
, chunk_start
, 1);
5507 read_unlock(&em_tree
->lock
);
5510 printk(KERN_ERR
"BTRFS: couldn't find em for chunk %Lu\n",
5515 if (em
->start
!= chunk_start
) {
5516 printk(KERN_ERR
"BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5517 em
->start
, chunk_start
);
5518 free_extent_map(em
);
5521 map
= (struct map_lookup
*)em
->bdev
;
5524 rmap_len
= map
->stripe_len
;
5526 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5527 length
= div_u64(length
, map
->num_stripes
/ map
->sub_stripes
);
5528 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5529 length
= div_u64(length
, map
->num_stripes
);
5530 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5531 length
= div_u64(length
, nr_data_stripes(map
));
5532 rmap_len
= map
->stripe_len
* nr_data_stripes(map
);
5535 buf
= kcalloc(map
->num_stripes
, sizeof(u64
), GFP_NOFS
);
5536 BUG_ON(!buf
); /* -ENOMEM */
5538 for (i
= 0; i
< map
->num_stripes
; i
++) {
5539 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
5541 if (map
->stripes
[i
].physical
> physical
||
5542 map
->stripes
[i
].physical
+ length
<= physical
)
5545 stripe_nr
= physical
- map
->stripes
[i
].physical
;
5546 stripe_nr
= div_u64(stripe_nr
, map
->stripe_len
);
5548 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5549 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5550 stripe_nr
= div_u64(stripe_nr
, map
->sub_stripes
);
5551 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5552 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5553 } /* else if RAID[56], multiply by nr_data_stripes().
5554 * Alternatively, just use rmap_len below instead of
5555 * map->stripe_len */
5557 bytenr
= chunk_start
+ stripe_nr
* rmap_len
;
5558 WARN_ON(nr
>= map
->num_stripes
);
5559 for (j
= 0; j
< nr
; j
++) {
5560 if (buf
[j
] == bytenr
)
5564 WARN_ON(nr
>= map
->num_stripes
);
5571 *stripe_len
= rmap_len
;
5573 free_extent_map(em
);
5577 static inline void btrfs_end_bbio(struct btrfs_bio
*bbio
, struct bio
*bio
, int err
)
5579 if (likely(bbio
->flags
& BTRFS_BIO_ORIG_BIO_SUBMITTED
))
5580 bio_endio_nodec(bio
, err
);
5582 bio_endio(bio
, err
);
5583 btrfs_put_bbio(bbio
);
5586 static void btrfs_end_bio(struct bio
*bio
, int err
)
5588 struct btrfs_bio
*bbio
= bio
->bi_private
;
5589 struct btrfs_device
*dev
= bbio
->stripes
[0].dev
;
5590 int is_orig_bio
= 0;
5593 atomic_inc(&bbio
->error
);
5594 if (err
== -EIO
|| err
== -EREMOTEIO
) {
5595 unsigned int stripe_index
=
5596 btrfs_io_bio(bio
)->stripe_index
;
5598 BUG_ON(stripe_index
>= bbio
->num_stripes
);
5599 dev
= bbio
->stripes
[stripe_index
].dev
;
5601 if (bio
->bi_rw
& WRITE
)
5602 btrfs_dev_stat_inc(dev
,
5603 BTRFS_DEV_STAT_WRITE_ERRS
);
5605 btrfs_dev_stat_inc(dev
,
5606 BTRFS_DEV_STAT_READ_ERRS
);
5607 if ((bio
->bi_rw
& WRITE_FLUSH
) == WRITE_FLUSH
)
5608 btrfs_dev_stat_inc(dev
,
5609 BTRFS_DEV_STAT_FLUSH_ERRS
);
5610 btrfs_dev_stat_print_on_error(dev
);
5615 if (bio
== bbio
->orig_bio
)
5618 btrfs_bio_counter_dec(bbio
->fs_info
);
5620 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
5623 bio
= bbio
->orig_bio
;
5626 bio
->bi_private
= bbio
->private;
5627 bio
->bi_end_io
= bbio
->end_io
;
5628 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
5629 /* only send an error to the higher layers if it is
5630 * beyond the tolerance of the btrfs bio
5632 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
5636 * this bio is actually up to date, we didn't
5637 * go over the max number of errors
5639 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5643 btrfs_end_bbio(bbio
, bio
, err
);
5644 } else if (!is_orig_bio
) {
5650 * see run_scheduled_bios for a description of why bios are collected for
5653 * This will add one bio to the pending list for a device and make sure
5654 * the work struct is scheduled.
5656 static noinline
void btrfs_schedule_bio(struct btrfs_root
*root
,
5657 struct btrfs_device
*device
,
5658 int rw
, struct bio
*bio
)
5660 int should_queue
= 1;
5661 struct btrfs_pending_bios
*pending_bios
;
5663 if (device
->missing
|| !device
->bdev
) {
5664 bio_endio(bio
, -EIO
);
5668 /* don't bother with additional async steps for reads, right now */
5669 if (!(rw
& REQ_WRITE
)) {
5671 btrfsic_submit_bio(rw
, bio
);
5677 * nr_async_bios allows us to reliably return congestion to the
5678 * higher layers. Otherwise, the async bio makes it appear we have
5679 * made progress against dirty pages when we've really just put it
5680 * on a queue for later
5682 atomic_inc(&root
->fs_info
->nr_async_bios
);
5683 WARN_ON(bio
->bi_next
);
5684 bio
->bi_next
= NULL
;
5687 spin_lock(&device
->io_lock
);
5688 if (bio
->bi_rw
& REQ_SYNC
)
5689 pending_bios
= &device
->pending_sync_bios
;
5691 pending_bios
= &device
->pending_bios
;
5693 if (pending_bios
->tail
)
5694 pending_bios
->tail
->bi_next
= bio
;
5696 pending_bios
->tail
= bio
;
5697 if (!pending_bios
->head
)
5698 pending_bios
->head
= bio
;
5699 if (device
->running_pending
)
5702 spin_unlock(&device
->io_lock
);
5705 btrfs_queue_work(root
->fs_info
->submit_workers
,
5709 static int bio_size_ok(struct block_device
*bdev
, struct bio
*bio
,
5712 struct bio_vec
*prev
;
5713 struct request_queue
*q
= bdev_get_queue(bdev
);
5714 unsigned int max_sectors
= queue_max_sectors(q
);
5715 struct bvec_merge_data bvm
= {
5717 .bi_sector
= sector
,
5718 .bi_rw
= bio
->bi_rw
,
5721 if (WARN_ON(bio
->bi_vcnt
== 0))
5724 prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
5725 if (bio_sectors(bio
) > max_sectors
)
5728 if (!q
->merge_bvec_fn
)
5731 bvm
.bi_size
= bio
->bi_iter
.bi_size
- prev
->bv_len
;
5732 if (q
->merge_bvec_fn(q
, &bvm
, prev
) < prev
->bv_len
)
5737 static void submit_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
5738 struct bio
*bio
, u64 physical
, int dev_nr
,
5741 struct btrfs_device
*dev
= bbio
->stripes
[dev_nr
].dev
;
5743 bio
->bi_private
= bbio
;
5744 btrfs_io_bio(bio
)->stripe_index
= dev_nr
;
5745 bio
->bi_end_io
= btrfs_end_bio
;
5746 bio
->bi_iter
.bi_sector
= physical
>> 9;
5749 struct rcu_string
*name
;
5752 name
= rcu_dereference(dev
->name
);
5753 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5754 "(%s id %llu), size=%u\n", rw
,
5755 (u64
)bio
->bi_iter
.bi_sector
, (u_long
)dev
->bdev
->bd_dev
,
5756 name
->str
, dev
->devid
, bio
->bi_iter
.bi_size
);
5760 bio
->bi_bdev
= dev
->bdev
;
5762 btrfs_bio_counter_inc_noblocked(root
->fs_info
);
5765 btrfs_schedule_bio(root
, dev
, rw
, bio
);
5767 btrfsic_submit_bio(rw
, bio
);
5770 static int breakup_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
5771 struct bio
*first_bio
, struct btrfs_device
*dev
,
5772 int dev_nr
, int rw
, int async
)
5774 struct bio_vec
*bvec
= first_bio
->bi_io_vec
;
5776 int nr_vecs
= bio_get_nr_vecs(dev
->bdev
);
5777 u64 physical
= bbio
->stripes
[dev_nr
].physical
;
5780 bio
= btrfs_bio_alloc(dev
->bdev
, physical
>> 9, nr_vecs
, GFP_NOFS
);
5784 while (bvec
<= (first_bio
->bi_io_vec
+ first_bio
->bi_vcnt
- 1)) {
5785 if (bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
5786 bvec
->bv_offset
) < bvec
->bv_len
) {
5787 u64 len
= bio
->bi_iter
.bi_size
;
5789 atomic_inc(&bbio
->stripes_pending
);
5790 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
,
5798 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
, rw
, async
);
5802 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
5804 atomic_inc(&bbio
->error
);
5805 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
5806 /* Shoud be the original bio. */
5807 WARN_ON(bio
!= bbio
->orig_bio
);
5809 bio
->bi_private
= bbio
->private;
5810 bio
->bi_end_io
= bbio
->end_io
;
5811 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
5812 bio
->bi_iter
.bi_sector
= logical
>> 9;
5814 btrfs_end_bbio(bbio
, bio
, -EIO
);
5818 int btrfs_map_bio(struct btrfs_root
*root
, int rw
, struct bio
*bio
,
5819 int mirror_num
, int async_submit
)
5821 struct btrfs_device
*dev
;
5822 struct bio
*first_bio
= bio
;
5823 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
5829 struct btrfs_bio
*bbio
= NULL
;
5831 length
= bio
->bi_iter
.bi_size
;
5832 map_length
= length
;
5834 btrfs_bio_counter_inc_blocked(root
->fs_info
);
5835 ret
= __btrfs_map_block(root
->fs_info
, rw
, logical
, &map_length
, &bbio
,
5838 btrfs_bio_counter_dec(root
->fs_info
);
5842 total_devs
= bbio
->num_stripes
;
5843 bbio
->orig_bio
= first_bio
;
5844 bbio
->private = first_bio
->bi_private
;
5845 bbio
->end_io
= first_bio
->bi_end_io
;
5846 bbio
->fs_info
= root
->fs_info
;
5847 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
5849 if (bbio
->raid_map
) {
5850 /* In this case, map_length has been set to the length of
5851 a single stripe; not the whole write */
5853 ret
= raid56_parity_write(root
, bio
, bbio
, map_length
);
5855 ret
= raid56_parity_recover(root
, bio
, bbio
, map_length
,
5859 btrfs_bio_counter_dec(root
->fs_info
);
5863 if (map_length
< length
) {
5864 btrfs_crit(root
->fs_info
, "mapping failed logical %llu bio len %llu len %llu",
5865 logical
, length
, map_length
);
5869 for (dev_nr
= 0; dev_nr
< total_devs
; dev_nr
++) {
5870 dev
= bbio
->stripes
[dev_nr
].dev
;
5871 if (!dev
|| !dev
->bdev
|| (rw
& WRITE
&& !dev
->writeable
)) {
5872 bbio_error(bbio
, first_bio
, logical
);
5877 * Check and see if we're ok with this bio based on it's size
5878 * and offset with the given device.
5880 if (!bio_size_ok(dev
->bdev
, first_bio
,
5881 bbio
->stripes
[dev_nr
].physical
>> 9)) {
5882 ret
= breakup_stripe_bio(root
, bbio
, first_bio
, dev
,
5883 dev_nr
, rw
, async_submit
);
5888 if (dev_nr
< total_devs
- 1) {
5889 bio
= btrfs_bio_clone(first_bio
, GFP_NOFS
);
5890 BUG_ON(!bio
); /* -ENOMEM */
5893 bbio
->flags
|= BTRFS_BIO_ORIG_BIO_SUBMITTED
;
5896 submit_stripe_bio(root
, bbio
, bio
,
5897 bbio
->stripes
[dev_nr
].physical
, dev_nr
, rw
,
5900 btrfs_bio_counter_dec(root
->fs_info
);
5904 struct btrfs_device
*btrfs_find_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
5907 struct btrfs_device
*device
;
5908 struct btrfs_fs_devices
*cur_devices
;
5910 cur_devices
= fs_info
->fs_devices
;
5911 while (cur_devices
) {
5913 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
5914 device
= __find_device(&cur_devices
->devices
,
5919 cur_devices
= cur_devices
->seed
;
5924 static struct btrfs_device
*add_missing_dev(struct btrfs_root
*root
,
5925 struct btrfs_fs_devices
*fs_devices
,
5926 u64 devid
, u8
*dev_uuid
)
5928 struct btrfs_device
*device
;
5930 device
= btrfs_alloc_device(NULL
, &devid
, dev_uuid
);
5934 list_add(&device
->dev_list
, &fs_devices
->devices
);
5935 device
->fs_devices
= fs_devices
;
5936 fs_devices
->num_devices
++;
5938 device
->missing
= 1;
5939 fs_devices
->missing_devices
++;
5945 * btrfs_alloc_device - allocate struct btrfs_device
5946 * @fs_info: used only for generating a new devid, can be NULL if
5947 * devid is provided (i.e. @devid != NULL).
5948 * @devid: a pointer to devid for this device. If NULL a new devid
5950 * @uuid: a pointer to UUID for this device. If NULL a new UUID
5953 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5954 * on error. Returned struct is not linked onto any lists and can be
5955 * destroyed with kfree() right away.
5957 struct btrfs_device
*btrfs_alloc_device(struct btrfs_fs_info
*fs_info
,
5961 struct btrfs_device
*dev
;
5964 if (WARN_ON(!devid
&& !fs_info
))
5965 return ERR_PTR(-EINVAL
);
5967 dev
= __alloc_device();
5976 ret
= find_next_devid(fs_info
, &tmp
);
5979 return ERR_PTR(ret
);
5985 memcpy(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
);
5987 generate_random_uuid(dev
->uuid
);
5989 btrfs_init_work(&dev
->work
, btrfs_submit_helper
,
5990 pending_bios_fn
, NULL
, NULL
);
5995 static int read_one_chunk(struct btrfs_root
*root
, struct btrfs_key
*key
,
5996 struct extent_buffer
*leaf
,
5997 struct btrfs_chunk
*chunk
)
5999 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
6000 struct map_lookup
*map
;
6001 struct extent_map
*em
;
6005 u8 uuid
[BTRFS_UUID_SIZE
];
6010 logical
= key
->offset
;
6011 length
= btrfs_chunk_length(leaf
, chunk
);
6013 read_lock(&map_tree
->map_tree
.lock
);
6014 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
6015 read_unlock(&map_tree
->map_tree
.lock
);
6017 /* already mapped? */
6018 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
6019 free_extent_map(em
);
6022 free_extent_map(em
);
6025 em
= alloc_extent_map();
6028 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6029 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
6031 free_extent_map(em
);
6035 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
6036 em
->bdev
= (struct block_device
*)map
;
6037 em
->start
= logical
;
6040 em
->block_start
= 0;
6041 em
->block_len
= em
->len
;
6043 map
->num_stripes
= num_stripes
;
6044 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
6045 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
6046 map
->sector_size
= btrfs_chunk_sector_size(leaf
, chunk
);
6047 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6048 map
->type
= btrfs_chunk_type(leaf
, chunk
);
6049 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6050 for (i
= 0; i
< num_stripes
; i
++) {
6051 map
->stripes
[i
].physical
=
6052 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
6053 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
6054 read_extent_buffer(leaf
, uuid
, (unsigned long)
6055 btrfs_stripe_dev_uuid_nr(chunk
, i
),
6057 map
->stripes
[i
].dev
= btrfs_find_device(root
->fs_info
, devid
,
6059 if (!map
->stripes
[i
].dev
&& !btrfs_test_opt(root
, DEGRADED
)) {
6060 free_extent_map(em
);
6063 if (!map
->stripes
[i
].dev
) {
6064 map
->stripes
[i
].dev
=
6065 add_missing_dev(root
, root
->fs_info
->fs_devices
,
6067 if (!map
->stripes
[i
].dev
) {
6068 free_extent_map(em
);
6072 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
6075 write_lock(&map_tree
->map_tree
.lock
);
6076 ret
= add_extent_mapping(&map_tree
->map_tree
, em
, 0);
6077 write_unlock(&map_tree
->map_tree
.lock
);
6078 BUG_ON(ret
); /* Tree corruption */
6079 free_extent_map(em
);
6084 static void fill_device_from_item(struct extent_buffer
*leaf
,
6085 struct btrfs_dev_item
*dev_item
,
6086 struct btrfs_device
*device
)
6090 device
->devid
= btrfs_device_id(leaf
, dev_item
);
6091 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
6092 device
->total_bytes
= device
->disk_total_bytes
;
6093 device
->commit_total_bytes
= device
->disk_total_bytes
;
6094 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
6095 device
->commit_bytes_used
= device
->bytes_used
;
6096 device
->type
= btrfs_device_type(leaf
, dev_item
);
6097 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
6098 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
6099 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
6100 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
6101 device
->is_tgtdev_for_dev_replace
= 0;
6103 ptr
= btrfs_device_uuid(dev_item
);
6104 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
6107 static struct btrfs_fs_devices
*open_seed_devices(struct btrfs_root
*root
,
6110 struct btrfs_fs_devices
*fs_devices
;
6113 BUG_ON(!mutex_is_locked(&uuid_mutex
));
6115 fs_devices
= root
->fs_info
->fs_devices
->seed
;
6116 while (fs_devices
) {
6117 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
))
6120 fs_devices
= fs_devices
->seed
;
6123 fs_devices
= find_fsid(fsid
);
6125 if (!btrfs_test_opt(root
, DEGRADED
))
6126 return ERR_PTR(-ENOENT
);
6128 fs_devices
= alloc_fs_devices(fsid
);
6129 if (IS_ERR(fs_devices
))
6132 fs_devices
->seeding
= 1;
6133 fs_devices
->opened
= 1;
6137 fs_devices
= clone_fs_devices(fs_devices
);
6138 if (IS_ERR(fs_devices
))
6141 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
6142 root
->fs_info
->bdev_holder
);
6144 free_fs_devices(fs_devices
);
6145 fs_devices
= ERR_PTR(ret
);
6149 if (!fs_devices
->seeding
) {
6150 __btrfs_close_devices(fs_devices
);
6151 free_fs_devices(fs_devices
);
6152 fs_devices
= ERR_PTR(-EINVAL
);
6156 fs_devices
->seed
= root
->fs_info
->fs_devices
->seed
;
6157 root
->fs_info
->fs_devices
->seed
= fs_devices
;
6162 static int read_one_dev(struct btrfs_root
*root
,
6163 struct extent_buffer
*leaf
,
6164 struct btrfs_dev_item
*dev_item
)
6166 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
6167 struct btrfs_device
*device
;
6170 u8 fs_uuid
[BTRFS_UUID_SIZE
];
6171 u8 dev_uuid
[BTRFS_UUID_SIZE
];
6173 devid
= btrfs_device_id(leaf
, dev_item
);
6174 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
6176 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
6179 if (memcmp(fs_uuid
, root
->fs_info
->fsid
, BTRFS_UUID_SIZE
)) {
6180 fs_devices
= open_seed_devices(root
, fs_uuid
);
6181 if (IS_ERR(fs_devices
))
6182 return PTR_ERR(fs_devices
);
6185 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
, fs_uuid
);
6187 if (!btrfs_test_opt(root
, DEGRADED
))
6190 btrfs_warn(root
->fs_info
, "devid %llu missing", devid
);
6191 device
= add_missing_dev(root
, fs_devices
, devid
, dev_uuid
);
6195 if (!device
->bdev
&& !btrfs_test_opt(root
, DEGRADED
))
6198 if(!device
->bdev
&& !device
->missing
) {
6200 * this happens when a device that was properly setup
6201 * in the device info lists suddenly goes bad.
6202 * device->bdev is NULL, and so we have to set
6203 * device->missing to one here
6205 device
->fs_devices
->missing_devices
++;
6206 device
->missing
= 1;
6209 /* Move the device to its own fs_devices */
6210 if (device
->fs_devices
!= fs_devices
) {
6211 ASSERT(device
->missing
);
6213 list_move(&device
->dev_list
, &fs_devices
->devices
);
6214 device
->fs_devices
->num_devices
--;
6215 fs_devices
->num_devices
++;
6217 device
->fs_devices
->missing_devices
--;
6218 fs_devices
->missing_devices
++;
6220 device
->fs_devices
= fs_devices
;
6224 if (device
->fs_devices
!= root
->fs_info
->fs_devices
) {
6225 BUG_ON(device
->writeable
);
6226 if (device
->generation
!=
6227 btrfs_device_generation(leaf
, dev_item
))
6231 fill_device_from_item(leaf
, dev_item
, device
);
6232 device
->in_fs_metadata
= 1;
6233 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
6234 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
6235 spin_lock(&root
->fs_info
->free_chunk_lock
);
6236 root
->fs_info
->free_chunk_space
+= device
->total_bytes
-
6238 spin_unlock(&root
->fs_info
->free_chunk_lock
);
6244 int btrfs_read_sys_array(struct btrfs_root
*root
)
6246 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
6247 struct extent_buffer
*sb
;
6248 struct btrfs_disk_key
*disk_key
;
6249 struct btrfs_chunk
*chunk
;
6251 unsigned long sb_array_offset
;
6257 struct btrfs_key key
;
6259 ASSERT(BTRFS_SUPER_INFO_SIZE
<= root
->nodesize
);
6261 * This will create extent buffer of nodesize, superblock size is
6262 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6263 * overallocate but we can keep it as-is, only the first page is used.
6265 sb
= btrfs_find_create_tree_block(root
, BTRFS_SUPER_INFO_OFFSET
);
6268 btrfs_set_buffer_uptodate(sb
);
6269 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
6271 * The sb extent buffer is artifical and just used to read the system array.
6272 * btrfs_set_buffer_uptodate() call does not properly mark all it's
6273 * pages up-to-date when the page is larger: extent does not cover the
6274 * whole page and consequently check_page_uptodate does not find all
6275 * the page's extents up-to-date (the hole beyond sb),
6276 * write_extent_buffer then triggers a WARN_ON.
6278 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6279 * but sb spans only this function. Add an explicit SetPageUptodate call
6280 * to silence the warning eg. on PowerPC 64.
6282 if (PAGE_CACHE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
6283 SetPageUptodate(sb
->pages
[0]);
6285 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
6286 array_size
= btrfs_super_sys_array_size(super_copy
);
6288 array_ptr
= super_copy
->sys_chunk_array
;
6289 sb_array_offset
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
6292 while (cur_offset
< array_size
) {
6293 disk_key
= (struct btrfs_disk_key
*)array_ptr
;
6294 len
= sizeof(*disk_key
);
6295 if (cur_offset
+ len
> array_size
)
6296 goto out_short_read
;
6298 btrfs_disk_key_to_cpu(&key
, disk_key
);
6301 sb_array_offset
+= len
;
6304 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6305 chunk
= (struct btrfs_chunk
*)sb_array_offset
;
6307 * At least one btrfs_chunk with one stripe must be
6308 * present, exact stripe count check comes afterwards
6310 len
= btrfs_chunk_item_size(1);
6311 if (cur_offset
+ len
> array_size
)
6312 goto out_short_read
;
6314 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
6315 len
= btrfs_chunk_item_size(num_stripes
);
6316 if (cur_offset
+ len
> array_size
)
6317 goto out_short_read
;
6319 ret
= read_one_chunk(root
, &key
, sb
, chunk
);
6327 sb_array_offset
+= len
;
6330 free_extent_buffer(sb
);
6334 printk(KERN_ERR
"BTRFS: sys_array too short to read %u bytes at offset %u\n",
6336 free_extent_buffer(sb
);
6340 int btrfs_read_chunk_tree(struct btrfs_root
*root
)
6342 struct btrfs_path
*path
;
6343 struct extent_buffer
*leaf
;
6344 struct btrfs_key key
;
6345 struct btrfs_key found_key
;
6349 root
= root
->fs_info
->chunk_root
;
6351 path
= btrfs_alloc_path();
6355 mutex_lock(&uuid_mutex
);
6359 * Read all device items, and then all the chunk items. All
6360 * device items are found before any chunk item (their object id
6361 * is smaller than the lowest possible object id for a chunk
6362 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6364 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
6367 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6371 leaf
= path
->nodes
[0];
6372 slot
= path
->slots
[0];
6373 if (slot
>= btrfs_header_nritems(leaf
)) {
6374 ret
= btrfs_next_leaf(root
, path
);
6381 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6382 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
6383 struct btrfs_dev_item
*dev_item
;
6384 dev_item
= btrfs_item_ptr(leaf
, slot
,
6385 struct btrfs_dev_item
);
6386 ret
= read_one_dev(root
, leaf
, dev_item
);
6389 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6390 struct btrfs_chunk
*chunk
;
6391 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
6392 ret
= read_one_chunk(root
, &found_key
, leaf
, chunk
);
6400 unlock_chunks(root
);
6401 mutex_unlock(&uuid_mutex
);
6403 btrfs_free_path(path
);
6407 void btrfs_init_devices_late(struct btrfs_fs_info
*fs_info
)
6409 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6410 struct btrfs_device
*device
;
6412 while (fs_devices
) {
6413 mutex_lock(&fs_devices
->device_list_mutex
);
6414 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
)
6415 device
->dev_root
= fs_info
->dev_root
;
6416 mutex_unlock(&fs_devices
->device_list_mutex
);
6418 fs_devices
= fs_devices
->seed
;
6422 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
6426 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6427 btrfs_dev_stat_reset(dev
, i
);
6430 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
6432 struct btrfs_key key
;
6433 struct btrfs_key found_key
;
6434 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6435 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6436 struct extent_buffer
*eb
;
6439 struct btrfs_device
*device
;
6440 struct btrfs_path
*path
= NULL
;
6443 path
= btrfs_alloc_path();
6449 mutex_lock(&fs_devices
->device_list_mutex
);
6450 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6452 struct btrfs_dev_stats_item
*ptr
;
6455 key
.type
= BTRFS_DEV_STATS_KEY
;
6456 key
.offset
= device
->devid
;
6457 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
6459 __btrfs_reset_dev_stats(device
);
6460 device
->dev_stats_valid
= 1;
6461 btrfs_release_path(path
);
6464 slot
= path
->slots
[0];
6465 eb
= path
->nodes
[0];
6466 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
6467 item_size
= btrfs_item_size_nr(eb
, slot
);
6469 ptr
= btrfs_item_ptr(eb
, slot
,
6470 struct btrfs_dev_stats_item
);
6472 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6473 if (item_size
>= (1 + i
) * sizeof(__le64
))
6474 btrfs_dev_stat_set(device
, i
,
6475 btrfs_dev_stats_value(eb
, ptr
, i
));
6477 btrfs_dev_stat_reset(device
, i
);
6480 device
->dev_stats_valid
= 1;
6481 btrfs_dev_stat_print_on_load(device
);
6482 btrfs_release_path(path
);
6484 mutex_unlock(&fs_devices
->device_list_mutex
);
6487 btrfs_free_path(path
);
6488 return ret
< 0 ? ret
: 0;
6491 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
6492 struct btrfs_root
*dev_root
,
6493 struct btrfs_device
*device
)
6495 struct btrfs_path
*path
;
6496 struct btrfs_key key
;
6497 struct extent_buffer
*eb
;
6498 struct btrfs_dev_stats_item
*ptr
;
6503 key
.type
= BTRFS_DEV_STATS_KEY
;
6504 key
.offset
= device
->devid
;
6506 path
= btrfs_alloc_path();
6508 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
6510 printk_in_rcu(KERN_WARNING
"BTRFS: "
6511 "error %d while searching for dev_stats item for device %s!\n",
6512 ret
, rcu_str_deref(device
->name
));
6517 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
6518 /* need to delete old one and insert a new one */
6519 ret
= btrfs_del_item(trans
, dev_root
, path
);
6521 printk_in_rcu(KERN_WARNING
"BTRFS: "
6522 "delete too small dev_stats item for device %s failed %d!\n",
6523 rcu_str_deref(device
->name
), ret
);
6530 /* need to insert a new item */
6531 btrfs_release_path(path
);
6532 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
6533 &key
, sizeof(*ptr
));
6535 printk_in_rcu(KERN_WARNING
"BTRFS: "
6536 "insert dev_stats item for device %s failed %d!\n",
6537 rcu_str_deref(device
->name
), ret
);
6542 eb
= path
->nodes
[0];
6543 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
6544 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6545 btrfs_set_dev_stats_value(eb
, ptr
, i
,
6546 btrfs_dev_stat_read(device
, i
));
6547 btrfs_mark_buffer_dirty(eb
);
6550 btrfs_free_path(path
);
6555 * called from commit_transaction. Writes all changed device stats to disk.
6557 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
6558 struct btrfs_fs_info
*fs_info
)
6560 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6561 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6562 struct btrfs_device
*device
;
6566 mutex_lock(&fs_devices
->device_list_mutex
);
6567 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6568 if (!device
->dev_stats_valid
|| !btrfs_dev_stats_dirty(device
))
6571 stats_cnt
= atomic_read(&device
->dev_stats_ccnt
);
6572 ret
= update_dev_stat_item(trans
, dev_root
, device
);
6574 atomic_sub(stats_cnt
, &device
->dev_stats_ccnt
);
6576 mutex_unlock(&fs_devices
->device_list_mutex
);
6581 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
6583 btrfs_dev_stat_inc(dev
, index
);
6584 btrfs_dev_stat_print_on_error(dev
);
6587 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
6589 if (!dev
->dev_stats_valid
)
6591 printk_ratelimited_in_rcu(KERN_ERR
"BTRFS: "
6592 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6593 rcu_str_deref(dev
->name
),
6594 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
6595 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
6596 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
6597 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
6598 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
6601 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
6605 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6606 if (btrfs_dev_stat_read(dev
, i
) != 0)
6608 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
6609 return; /* all values == 0, suppress message */
6611 printk_in_rcu(KERN_INFO
"BTRFS: "
6612 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6613 rcu_str_deref(dev
->name
),
6614 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
6615 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
6616 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
6617 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
6618 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
6621 int btrfs_get_dev_stats(struct btrfs_root
*root
,
6622 struct btrfs_ioctl_get_dev_stats
*stats
)
6624 struct btrfs_device
*dev
;
6625 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
6628 mutex_lock(&fs_devices
->device_list_mutex
);
6629 dev
= btrfs_find_device(root
->fs_info
, stats
->devid
, NULL
, NULL
);
6630 mutex_unlock(&fs_devices
->device_list_mutex
);
6633 btrfs_warn(root
->fs_info
, "get dev_stats failed, device not found");
6635 } else if (!dev
->dev_stats_valid
) {
6636 btrfs_warn(root
->fs_info
, "get dev_stats failed, not yet valid");
6638 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
6639 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6640 if (stats
->nr_items
> i
)
6642 btrfs_dev_stat_read_and_reset(dev
, i
);
6644 btrfs_dev_stat_reset(dev
, i
);
6647 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6648 if (stats
->nr_items
> i
)
6649 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
6651 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
6652 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
6656 int btrfs_scratch_superblock(struct btrfs_device
*device
)
6658 struct buffer_head
*bh
;
6659 struct btrfs_super_block
*disk_super
;
6661 bh
= btrfs_read_dev_super(device
->bdev
);
6664 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
6666 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
6667 set_buffer_dirty(bh
);
6668 sync_dirty_buffer(bh
);
6675 * Update the size of all devices, which is used for writing out the
6678 void btrfs_update_commit_device_size(struct btrfs_fs_info
*fs_info
)
6680 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6681 struct btrfs_device
*curr
, *next
;
6683 if (list_empty(&fs_devices
->resized_devices
))
6686 mutex_lock(&fs_devices
->device_list_mutex
);
6687 lock_chunks(fs_info
->dev_root
);
6688 list_for_each_entry_safe(curr
, next
, &fs_devices
->resized_devices
,
6690 list_del_init(&curr
->resized_list
);
6691 curr
->commit_total_bytes
= curr
->disk_total_bytes
;
6693 unlock_chunks(fs_info
->dev_root
);
6694 mutex_unlock(&fs_devices
->device_list_mutex
);
6697 /* Must be invoked during the transaction commit */
6698 void btrfs_update_commit_device_bytes_used(struct btrfs_root
*root
,
6699 struct btrfs_transaction
*transaction
)
6701 struct extent_map
*em
;
6702 struct map_lookup
*map
;
6703 struct btrfs_device
*dev
;
6706 if (list_empty(&transaction
->pending_chunks
))
6709 /* In order to kick the device replace finish process */
6711 list_for_each_entry(em
, &transaction
->pending_chunks
, list
) {
6712 map
= (struct map_lookup
*)em
->bdev
;
6714 for (i
= 0; i
< map
->num_stripes
; i
++) {
6715 dev
= map
->stripes
[i
].dev
;
6716 dev
->commit_bytes_used
= dev
->bytes_used
;
6719 unlock_chunks(root
);