2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/ratelimit.h>
27 #include <linux/kthread.h>
28 #include <linux/raid/pq.h>
29 #include <linux/semaphore.h>
30 #include <asm/div64.h>
32 #include "extent_map.h"
34 #include "transaction.h"
35 #include "print-tree.h"
38 #include "async-thread.h"
39 #include "check-integrity.h"
40 #include "rcu-string.h"
42 #include "dev-replace.h"
45 static int init_first_rw_device(struct btrfs_trans_handle
*trans
,
46 struct btrfs_root
*root
,
47 struct btrfs_device
*device
);
48 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
);
49 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
);
50 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
);
51 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*device
);
53 DEFINE_MUTEX(uuid_mutex
);
54 static LIST_HEAD(fs_uuids
);
56 static struct btrfs_fs_devices
*__alloc_fs_devices(void)
58 struct btrfs_fs_devices
*fs_devs
;
60 fs_devs
= kzalloc(sizeof(*fs_devs
), GFP_NOFS
);
62 return ERR_PTR(-ENOMEM
);
64 mutex_init(&fs_devs
->device_list_mutex
);
66 INIT_LIST_HEAD(&fs_devs
->devices
);
67 INIT_LIST_HEAD(&fs_devs
->resized_devices
);
68 INIT_LIST_HEAD(&fs_devs
->alloc_list
);
69 INIT_LIST_HEAD(&fs_devs
->list
);
75 * alloc_fs_devices - allocate struct btrfs_fs_devices
76 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
79 * Return: a pointer to a new &struct btrfs_fs_devices on success;
80 * ERR_PTR() on error. Returned struct is not linked onto any lists and
81 * can be destroyed with kfree() right away.
83 static struct btrfs_fs_devices
*alloc_fs_devices(const u8
*fsid
)
85 struct btrfs_fs_devices
*fs_devs
;
87 fs_devs
= __alloc_fs_devices();
92 memcpy(fs_devs
->fsid
, fsid
, BTRFS_FSID_SIZE
);
94 generate_random_uuid(fs_devs
->fsid
);
99 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
101 struct btrfs_device
*device
;
102 WARN_ON(fs_devices
->opened
);
103 while (!list_empty(&fs_devices
->devices
)) {
104 device
= list_entry(fs_devices
->devices
.next
,
105 struct btrfs_device
, dev_list
);
106 list_del(&device
->dev_list
);
107 rcu_string_free(device
->name
);
113 static void btrfs_kobject_uevent(struct block_device
*bdev
,
114 enum kobject_action action
)
118 ret
= kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, action
);
120 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
122 kobject_name(&disk_to_dev(bdev
->bd_disk
)->kobj
),
123 &disk_to_dev(bdev
->bd_disk
)->kobj
);
126 void btrfs_cleanup_fs_uuids(void)
128 struct btrfs_fs_devices
*fs_devices
;
130 while (!list_empty(&fs_uuids
)) {
131 fs_devices
= list_entry(fs_uuids
.next
,
132 struct btrfs_fs_devices
, list
);
133 list_del(&fs_devices
->list
);
134 free_fs_devices(fs_devices
);
138 static struct btrfs_device
*__alloc_device(void)
140 struct btrfs_device
*dev
;
142 dev
= kzalloc(sizeof(*dev
), GFP_NOFS
);
144 return ERR_PTR(-ENOMEM
);
146 INIT_LIST_HEAD(&dev
->dev_list
);
147 INIT_LIST_HEAD(&dev
->dev_alloc_list
);
148 INIT_LIST_HEAD(&dev
->resized_list
);
150 spin_lock_init(&dev
->io_lock
);
152 spin_lock_init(&dev
->reada_lock
);
153 atomic_set(&dev
->reada_in_flight
, 0);
154 atomic_set(&dev
->dev_stats_ccnt
, 0);
155 INIT_RADIX_TREE(&dev
->reada_zones
, GFP_NOFS
& ~__GFP_WAIT
);
156 INIT_RADIX_TREE(&dev
->reada_extents
, GFP_NOFS
& ~__GFP_WAIT
);
161 static noinline
struct btrfs_device
*__find_device(struct list_head
*head
,
164 struct btrfs_device
*dev
;
166 list_for_each_entry(dev
, head
, dev_list
) {
167 if (dev
->devid
== devid
&&
168 (!uuid
|| !memcmp(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
))) {
175 static noinline
struct btrfs_fs_devices
*find_fsid(u8
*fsid
)
177 struct btrfs_fs_devices
*fs_devices
;
179 list_for_each_entry(fs_devices
, &fs_uuids
, list
) {
180 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
187 btrfs_get_bdev_and_sb(const char *device_path
, fmode_t flags
, void *holder
,
188 int flush
, struct block_device
**bdev
,
189 struct buffer_head
**bh
)
193 *bdev
= blkdev_get_by_path(device_path
, flags
, holder
);
196 ret
= PTR_ERR(*bdev
);
197 printk(KERN_INFO
"BTRFS: open %s failed\n", device_path
);
202 filemap_write_and_wait((*bdev
)->bd_inode
->i_mapping
);
203 ret
= set_blocksize(*bdev
, 4096);
205 blkdev_put(*bdev
, flags
);
208 invalidate_bdev(*bdev
);
209 *bh
= btrfs_read_dev_super(*bdev
);
212 blkdev_put(*bdev
, flags
);
224 static void requeue_list(struct btrfs_pending_bios
*pending_bios
,
225 struct bio
*head
, struct bio
*tail
)
228 struct bio
*old_head
;
230 old_head
= pending_bios
->head
;
231 pending_bios
->head
= head
;
232 if (pending_bios
->tail
)
233 tail
->bi_next
= old_head
;
235 pending_bios
->tail
= tail
;
239 * we try to collect pending bios for a device so we don't get a large
240 * number of procs sending bios down to the same device. This greatly
241 * improves the schedulers ability to collect and merge the bios.
243 * But, it also turns into a long list of bios to process and that is sure
244 * to eventually make the worker thread block. The solution here is to
245 * make some progress and then put this work struct back at the end of
246 * the list if the block device is congested. This way, multiple devices
247 * can make progress from a single worker thread.
249 static noinline
void run_scheduled_bios(struct btrfs_device
*device
)
252 struct backing_dev_info
*bdi
;
253 struct btrfs_fs_info
*fs_info
;
254 struct btrfs_pending_bios
*pending_bios
;
258 unsigned long num_run
;
259 unsigned long batch_run
= 0;
261 unsigned long last_waited
= 0;
263 int sync_pending
= 0;
264 struct blk_plug plug
;
267 * this function runs all the bios we've collected for
268 * a particular device. We don't want to wander off to
269 * another device without first sending all of these down.
270 * So, setup a plug here and finish it off before we return
272 blk_start_plug(&plug
);
274 bdi
= blk_get_backing_dev_info(device
->bdev
);
275 fs_info
= device
->dev_root
->fs_info
;
276 limit
= btrfs_async_submit_limit(fs_info
);
277 limit
= limit
* 2 / 3;
280 spin_lock(&device
->io_lock
);
285 /* take all the bios off the list at once and process them
286 * later on (without the lock held). But, remember the
287 * tail and other pointers so the bios can be properly reinserted
288 * into the list if we hit congestion
290 if (!force_reg
&& device
->pending_sync_bios
.head
) {
291 pending_bios
= &device
->pending_sync_bios
;
294 pending_bios
= &device
->pending_bios
;
298 pending
= pending_bios
->head
;
299 tail
= pending_bios
->tail
;
300 WARN_ON(pending
&& !tail
);
303 * if pending was null this time around, no bios need processing
304 * at all and we can stop. Otherwise it'll loop back up again
305 * and do an additional check so no bios are missed.
307 * device->running_pending is used to synchronize with the
310 if (device
->pending_sync_bios
.head
== NULL
&&
311 device
->pending_bios
.head
== NULL
) {
313 device
->running_pending
= 0;
316 device
->running_pending
= 1;
319 pending_bios
->head
= NULL
;
320 pending_bios
->tail
= NULL
;
322 spin_unlock(&device
->io_lock
);
327 /* we want to work on both lists, but do more bios on the
328 * sync list than the regular list
331 pending_bios
!= &device
->pending_sync_bios
&&
332 device
->pending_sync_bios
.head
) ||
333 (num_run
> 64 && pending_bios
== &device
->pending_sync_bios
&&
334 device
->pending_bios
.head
)) {
335 spin_lock(&device
->io_lock
);
336 requeue_list(pending_bios
, pending
, tail
);
341 pending
= pending
->bi_next
;
344 if (atomic_dec_return(&fs_info
->nr_async_bios
) < limit
&&
345 waitqueue_active(&fs_info
->async_submit_wait
))
346 wake_up(&fs_info
->async_submit_wait
);
348 BUG_ON(atomic_read(&cur
->bi_cnt
) == 0);
351 * if we're doing the sync list, record that our
352 * plug has some sync requests on it
354 * If we're doing the regular list and there are
355 * sync requests sitting around, unplug before
358 if (pending_bios
== &device
->pending_sync_bios
) {
360 } else if (sync_pending
) {
361 blk_finish_plug(&plug
);
362 blk_start_plug(&plug
);
366 btrfsic_submit_bio(cur
->bi_rw
, cur
);
373 * we made progress, there is more work to do and the bdi
374 * is now congested. Back off and let other work structs
377 if (pending
&& bdi_write_congested(bdi
) && batch_run
> 8 &&
378 fs_info
->fs_devices
->open_devices
> 1) {
379 struct io_context
*ioc
;
381 ioc
= current
->io_context
;
384 * the main goal here is that we don't want to
385 * block if we're going to be able to submit
386 * more requests without blocking.
388 * This code does two great things, it pokes into
389 * the elevator code from a filesystem _and_
390 * it makes assumptions about how batching works.
392 if (ioc
&& ioc
->nr_batch_requests
> 0 &&
393 time_before(jiffies
, ioc
->last_waited
+ HZ
/50UL) &&
395 ioc
->last_waited
== last_waited
)) {
397 * we want to go through our batch of
398 * requests and stop. So, we copy out
399 * the ioc->last_waited time and test
400 * against it before looping
402 last_waited
= ioc
->last_waited
;
407 spin_lock(&device
->io_lock
);
408 requeue_list(pending_bios
, pending
, tail
);
409 device
->running_pending
= 1;
411 spin_unlock(&device
->io_lock
);
412 btrfs_queue_work(fs_info
->submit_workers
,
416 /* unplug every 64 requests just for good measure */
417 if (batch_run
% 64 == 0) {
418 blk_finish_plug(&plug
);
419 blk_start_plug(&plug
);
428 spin_lock(&device
->io_lock
);
429 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
431 spin_unlock(&device
->io_lock
);
434 blk_finish_plug(&plug
);
437 static void pending_bios_fn(struct btrfs_work
*work
)
439 struct btrfs_device
*device
;
441 device
= container_of(work
, struct btrfs_device
, work
);
442 run_scheduled_bios(device
);
446 * Add new device to list of registered devices
449 * 1 - first time device is seen
450 * 0 - device already known
453 static noinline
int device_list_add(const char *path
,
454 struct btrfs_super_block
*disk_super
,
455 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
457 struct btrfs_device
*device
;
458 struct btrfs_fs_devices
*fs_devices
;
459 struct rcu_string
*name
;
461 u64 found_transid
= btrfs_super_generation(disk_super
);
463 fs_devices
= find_fsid(disk_super
->fsid
);
465 fs_devices
= alloc_fs_devices(disk_super
->fsid
);
466 if (IS_ERR(fs_devices
))
467 return PTR_ERR(fs_devices
);
469 list_add(&fs_devices
->list
, &fs_uuids
);
473 device
= __find_device(&fs_devices
->devices
, devid
,
474 disk_super
->dev_item
.uuid
);
478 if (fs_devices
->opened
)
481 device
= btrfs_alloc_device(NULL
, &devid
,
482 disk_super
->dev_item
.uuid
);
483 if (IS_ERR(device
)) {
484 /* we can safely leave the fs_devices entry around */
485 return PTR_ERR(device
);
488 name
= rcu_string_strdup(path
, GFP_NOFS
);
493 rcu_assign_pointer(device
->name
, name
);
495 mutex_lock(&fs_devices
->device_list_mutex
);
496 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
497 fs_devices
->num_devices
++;
498 mutex_unlock(&fs_devices
->device_list_mutex
);
501 device
->fs_devices
= fs_devices
;
502 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
504 * When FS is already mounted.
505 * 1. If you are here and if the device->name is NULL that
506 * means this device was missing at time of FS mount.
507 * 2. If you are here and if the device->name is different
508 * from 'path' that means either
509 * a. The same device disappeared and reappeared with
511 * b. The missing-disk-which-was-replaced, has
514 * We must allow 1 and 2a above. But 2b would be a spurious
517 * Further in case of 1 and 2a above, the disk at 'path'
518 * would have missed some transaction when it was away and
519 * in case of 2a the stale bdev has to be updated as well.
520 * 2b must not be allowed at all time.
524 * For now, we do allow update to btrfs_fs_device through the
525 * btrfs dev scan cli after FS has been mounted. We're still
526 * tracking a problem where systems fail mount by subvolume id
527 * when we reject replacement on a mounted FS.
529 if (!fs_devices
->opened
&& found_transid
< device
->generation
) {
531 * That is if the FS is _not_ mounted and if you
532 * are here, that means there is more than one
533 * disk with same uuid and devid.We keep the one
534 * with larger generation number or the last-in if
535 * generation are equal.
540 name
= rcu_string_strdup(path
, GFP_NOFS
);
543 rcu_string_free(device
->name
);
544 rcu_assign_pointer(device
->name
, name
);
545 if (device
->missing
) {
546 fs_devices
->missing_devices
--;
552 * Unmount does not free the btrfs_device struct but would zero
553 * generation along with most of the other members. So just update
554 * it back. We need it to pick the disk with largest generation
557 if (!fs_devices
->opened
)
558 device
->generation
= found_transid
;
560 *fs_devices_ret
= fs_devices
;
565 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
567 struct btrfs_fs_devices
*fs_devices
;
568 struct btrfs_device
*device
;
569 struct btrfs_device
*orig_dev
;
571 fs_devices
= alloc_fs_devices(orig
->fsid
);
572 if (IS_ERR(fs_devices
))
575 mutex_lock(&orig
->device_list_mutex
);
576 fs_devices
->total_devices
= orig
->total_devices
;
578 /* We have held the volume lock, it is safe to get the devices. */
579 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
580 struct rcu_string
*name
;
582 device
= btrfs_alloc_device(NULL
, &orig_dev
->devid
,
588 * This is ok to do without rcu read locked because we hold the
589 * uuid mutex so nothing we touch in here is going to disappear.
591 if (orig_dev
->name
) {
592 name
= rcu_string_strdup(orig_dev
->name
->str
, GFP_NOFS
);
597 rcu_assign_pointer(device
->name
, name
);
600 list_add(&device
->dev_list
, &fs_devices
->devices
);
601 device
->fs_devices
= fs_devices
;
602 fs_devices
->num_devices
++;
604 mutex_unlock(&orig
->device_list_mutex
);
607 mutex_unlock(&orig
->device_list_mutex
);
608 free_fs_devices(fs_devices
);
609 return ERR_PTR(-ENOMEM
);
612 void btrfs_close_extra_devices(struct btrfs_fs_info
*fs_info
,
613 struct btrfs_fs_devices
*fs_devices
, int step
)
615 struct btrfs_device
*device
, *next
;
616 struct btrfs_device
*latest_dev
= NULL
;
618 mutex_lock(&uuid_mutex
);
620 /* This is the initialized path, it is safe to release the devices. */
621 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
622 if (device
->in_fs_metadata
) {
623 if (!device
->is_tgtdev_for_dev_replace
&&
625 device
->generation
> latest_dev
->generation
)) {
631 if (device
->devid
== BTRFS_DEV_REPLACE_DEVID
) {
633 * In the first step, keep the device which has
634 * the correct fsid and the devid that is used
635 * for the dev_replace procedure.
636 * In the second step, the dev_replace state is
637 * read from the device tree and it is known
638 * whether the procedure is really active or
639 * not, which means whether this device is
640 * used or whether it should be removed.
642 if (step
== 0 || device
->is_tgtdev_for_dev_replace
) {
647 blkdev_put(device
->bdev
, device
->mode
);
649 fs_devices
->open_devices
--;
651 if (device
->writeable
) {
652 list_del_init(&device
->dev_alloc_list
);
653 device
->writeable
= 0;
654 if (!device
->is_tgtdev_for_dev_replace
)
655 fs_devices
->rw_devices
--;
657 list_del_init(&device
->dev_list
);
658 fs_devices
->num_devices
--;
659 rcu_string_free(device
->name
);
663 if (fs_devices
->seed
) {
664 fs_devices
= fs_devices
->seed
;
668 fs_devices
->latest_bdev
= latest_dev
->bdev
;
670 mutex_unlock(&uuid_mutex
);
673 static void __free_device(struct work_struct
*work
)
675 struct btrfs_device
*device
;
677 device
= container_of(work
, struct btrfs_device
, rcu_work
);
680 blkdev_put(device
->bdev
, device
->mode
);
682 rcu_string_free(device
->name
);
686 static void free_device(struct rcu_head
*head
)
688 struct btrfs_device
*device
;
690 device
= container_of(head
, struct btrfs_device
, rcu
);
692 INIT_WORK(&device
->rcu_work
, __free_device
);
693 schedule_work(&device
->rcu_work
);
696 static int __btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
698 struct btrfs_device
*device
;
700 if (--fs_devices
->opened
> 0)
703 mutex_lock(&fs_devices
->device_list_mutex
);
704 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
705 struct btrfs_device
*new_device
;
706 struct rcu_string
*name
;
709 fs_devices
->open_devices
--;
711 if (device
->writeable
&&
712 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
713 list_del_init(&device
->dev_alloc_list
);
714 fs_devices
->rw_devices
--;
718 fs_devices
->missing_devices
--;
720 new_device
= btrfs_alloc_device(NULL
, &device
->devid
,
722 BUG_ON(IS_ERR(new_device
)); /* -ENOMEM */
724 /* Safe because we are under uuid_mutex */
726 name
= rcu_string_strdup(device
->name
->str
, GFP_NOFS
);
727 BUG_ON(!name
); /* -ENOMEM */
728 rcu_assign_pointer(new_device
->name
, name
);
731 list_replace_rcu(&device
->dev_list
, &new_device
->dev_list
);
732 new_device
->fs_devices
= device
->fs_devices
;
734 call_rcu(&device
->rcu
, free_device
);
736 mutex_unlock(&fs_devices
->device_list_mutex
);
738 WARN_ON(fs_devices
->open_devices
);
739 WARN_ON(fs_devices
->rw_devices
);
740 fs_devices
->opened
= 0;
741 fs_devices
->seeding
= 0;
746 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
748 struct btrfs_fs_devices
*seed_devices
= NULL
;
751 mutex_lock(&uuid_mutex
);
752 ret
= __btrfs_close_devices(fs_devices
);
753 if (!fs_devices
->opened
) {
754 seed_devices
= fs_devices
->seed
;
755 fs_devices
->seed
= NULL
;
757 mutex_unlock(&uuid_mutex
);
759 while (seed_devices
) {
760 fs_devices
= seed_devices
;
761 seed_devices
= fs_devices
->seed
;
762 __btrfs_close_devices(fs_devices
);
763 free_fs_devices(fs_devices
);
766 * Wait for rcu kworkers under __btrfs_close_devices
767 * to finish all blkdev_puts so device is really
768 * free when umount is done.
774 static int __btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
775 fmode_t flags
, void *holder
)
777 struct request_queue
*q
;
778 struct block_device
*bdev
;
779 struct list_head
*head
= &fs_devices
->devices
;
780 struct btrfs_device
*device
;
781 struct btrfs_device
*latest_dev
= NULL
;
782 struct buffer_head
*bh
;
783 struct btrfs_super_block
*disk_super
;
790 list_for_each_entry(device
, head
, dev_list
) {
796 /* Just open everything we can; ignore failures here */
797 if (btrfs_get_bdev_and_sb(device
->name
->str
, flags
, holder
, 1,
801 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
802 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
803 if (devid
!= device
->devid
)
806 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
,
810 device
->generation
= btrfs_super_generation(disk_super
);
812 device
->generation
> latest_dev
->generation
)
815 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
816 device
->writeable
= 0;
818 device
->writeable
= !bdev_read_only(bdev
);
822 q
= bdev_get_queue(bdev
);
823 if (blk_queue_discard(q
))
824 device
->can_discard
= 1;
827 device
->in_fs_metadata
= 0;
828 device
->mode
= flags
;
830 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
831 fs_devices
->rotating
= 1;
833 fs_devices
->open_devices
++;
834 if (device
->writeable
&&
835 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
836 fs_devices
->rw_devices
++;
837 list_add(&device
->dev_alloc_list
,
838 &fs_devices
->alloc_list
);
845 blkdev_put(bdev
, flags
);
848 if (fs_devices
->open_devices
== 0) {
852 fs_devices
->seeding
= seeding
;
853 fs_devices
->opened
= 1;
854 fs_devices
->latest_bdev
= latest_dev
->bdev
;
855 fs_devices
->total_rw_bytes
= 0;
860 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
861 fmode_t flags
, void *holder
)
865 mutex_lock(&uuid_mutex
);
866 if (fs_devices
->opened
) {
867 fs_devices
->opened
++;
870 ret
= __btrfs_open_devices(fs_devices
, flags
, holder
);
872 mutex_unlock(&uuid_mutex
);
877 * Look for a btrfs signature on a device. This may be called out of the mount path
878 * and we are not allowed to call set_blocksize during the scan. The superblock
879 * is read via pagecache
881 int btrfs_scan_one_device(const char *path
, fmode_t flags
, void *holder
,
882 struct btrfs_fs_devices
**fs_devices_ret
)
884 struct btrfs_super_block
*disk_super
;
885 struct block_device
*bdev
;
896 * we would like to check all the supers, but that would make
897 * a btrfs mount succeed after a mkfs from a different FS.
898 * So, we need to add a special mount option to scan for
899 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
901 bytenr
= btrfs_sb_offset(0);
903 mutex_lock(&uuid_mutex
);
905 bdev
= blkdev_get_by_path(path
, flags
, holder
);
912 /* make sure our super fits in the device */
913 if (bytenr
+ PAGE_CACHE_SIZE
>= i_size_read(bdev
->bd_inode
))
916 /* make sure our super fits in the page */
917 if (sizeof(*disk_super
) > PAGE_CACHE_SIZE
)
920 /* make sure our super doesn't straddle pages on disk */
921 index
= bytenr
>> PAGE_CACHE_SHIFT
;
922 if ((bytenr
+ sizeof(*disk_super
) - 1) >> PAGE_CACHE_SHIFT
!= index
)
925 /* pull in the page with our super */
926 page
= read_cache_page_gfp(bdev
->bd_inode
->i_mapping
,
929 if (IS_ERR_OR_NULL(page
))
934 /* align our pointer to the offset of the super block */
935 disk_super
= p
+ (bytenr
& ~PAGE_CACHE_MASK
);
937 if (btrfs_super_bytenr(disk_super
) != bytenr
||
938 btrfs_super_magic(disk_super
) != BTRFS_MAGIC
)
941 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
942 transid
= btrfs_super_generation(disk_super
);
943 total_devices
= btrfs_super_num_devices(disk_super
);
945 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
947 if (disk_super
->label
[0]) {
948 if (disk_super
->label
[BTRFS_LABEL_SIZE
- 1])
949 disk_super
->label
[BTRFS_LABEL_SIZE
- 1] = '\0';
950 printk(KERN_INFO
"BTRFS: device label %s ", disk_super
->label
);
952 printk(KERN_INFO
"BTRFS: device fsid %pU ", disk_super
->fsid
);
955 printk(KERN_CONT
"devid %llu transid %llu %s\n", devid
, transid
, path
);
958 if (!ret
&& fs_devices_ret
)
959 (*fs_devices_ret
)->total_devices
= total_devices
;
963 page_cache_release(page
);
966 blkdev_put(bdev
, flags
);
968 mutex_unlock(&uuid_mutex
);
972 /* helper to account the used device space in the range */
973 int btrfs_account_dev_extents_size(struct btrfs_device
*device
, u64 start
,
974 u64 end
, u64
*length
)
976 struct btrfs_key key
;
977 struct btrfs_root
*root
= device
->dev_root
;
978 struct btrfs_dev_extent
*dev_extent
;
979 struct btrfs_path
*path
;
983 struct extent_buffer
*l
;
987 if (start
>= device
->total_bytes
|| device
->is_tgtdev_for_dev_replace
)
990 path
= btrfs_alloc_path();
995 key
.objectid
= device
->devid
;
997 key
.type
= BTRFS_DEV_EXTENT_KEY
;
999 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1003 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1010 slot
= path
->slots
[0];
1011 if (slot
>= btrfs_header_nritems(l
)) {
1012 ret
= btrfs_next_leaf(root
, path
);
1020 btrfs_item_key_to_cpu(l
, &key
, slot
);
1022 if (key
.objectid
< device
->devid
)
1025 if (key
.objectid
> device
->devid
)
1028 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1031 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1032 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1034 if (key
.offset
<= start
&& extent_end
> end
) {
1035 *length
= end
- start
+ 1;
1037 } else if (key
.offset
<= start
&& extent_end
> start
)
1038 *length
+= extent_end
- start
;
1039 else if (key
.offset
> start
&& extent_end
<= end
)
1040 *length
+= extent_end
- key
.offset
;
1041 else if (key
.offset
> start
&& key
.offset
<= end
) {
1042 *length
+= end
- key
.offset
+ 1;
1044 } else if (key
.offset
> end
)
1052 btrfs_free_path(path
);
1056 static int contains_pending_extent(struct btrfs_trans_handle
*trans
,
1057 struct btrfs_device
*device
,
1058 u64
*start
, u64 len
)
1060 struct extent_map
*em
;
1061 struct list_head
*search_list
= &trans
->transaction
->pending_chunks
;
1065 list_for_each_entry(em
, search_list
, list
) {
1066 struct map_lookup
*map
;
1069 map
= (struct map_lookup
*)em
->bdev
;
1070 for (i
= 0; i
< map
->num_stripes
; i
++) {
1071 if (map
->stripes
[i
].dev
!= device
)
1073 if (map
->stripes
[i
].physical
>= *start
+ len
||
1074 map
->stripes
[i
].physical
+ em
->orig_block_len
<=
1077 *start
= map
->stripes
[i
].physical
+
1082 if (search_list
== &trans
->transaction
->pending_chunks
) {
1083 search_list
= &trans
->root
->fs_info
->pinned_chunks
;
1092 * find_free_dev_extent - find free space in the specified device
1093 * @device: the device which we search the free space in
1094 * @num_bytes: the size of the free space that we need
1095 * @start: store the start of the free space.
1096 * @len: the size of the free space. that we find, or the size of the max
1097 * free space if we don't find suitable free space
1099 * this uses a pretty simple search, the expectation is that it is
1100 * called very infrequently and that a given device has a small number
1103 * @start is used to store the start of the free space if we find. But if we
1104 * don't find suitable free space, it will be used to store the start position
1105 * of the max free space.
1107 * @len is used to store the size of the free space that we find.
1108 * But if we don't find suitable free space, it is used to store the size of
1109 * the max free space.
1111 int find_free_dev_extent(struct btrfs_trans_handle
*trans
,
1112 struct btrfs_device
*device
, u64 num_bytes
,
1113 u64
*start
, u64
*len
)
1115 struct btrfs_key key
;
1116 struct btrfs_root
*root
= device
->dev_root
;
1117 struct btrfs_dev_extent
*dev_extent
;
1118 struct btrfs_path
*path
;
1124 u64 search_end
= device
->total_bytes
;
1127 struct extent_buffer
*l
;
1129 /* FIXME use last free of some kind */
1131 /* we don't want to overwrite the superblock on the drive,
1132 * so we make sure to start at an offset of at least 1MB
1134 search_start
= max(root
->fs_info
->alloc_start
, 1024ull * 1024);
1136 path
= btrfs_alloc_path();
1140 max_hole_start
= search_start
;
1144 if (search_start
>= search_end
|| device
->is_tgtdev_for_dev_replace
) {
1150 path
->search_commit_root
= 1;
1151 path
->skip_locking
= 1;
1153 key
.objectid
= device
->devid
;
1154 key
.offset
= search_start
;
1155 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1157 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1161 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1168 slot
= path
->slots
[0];
1169 if (slot
>= btrfs_header_nritems(l
)) {
1170 ret
= btrfs_next_leaf(root
, path
);
1178 btrfs_item_key_to_cpu(l
, &key
, slot
);
1180 if (key
.objectid
< device
->devid
)
1183 if (key
.objectid
> device
->devid
)
1186 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1189 if (key
.offset
> search_start
) {
1190 hole_size
= key
.offset
- search_start
;
1193 * Have to check before we set max_hole_start, otherwise
1194 * we could end up sending back this offset anyway.
1196 if (contains_pending_extent(trans
, device
,
1201 if (hole_size
> max_hole_size
) {
1202 max_hole_start
= search_start
;
1203 max_hole_size
= hole_size
;
1207 * If this free space is greater than which we need,
1208 * it must be the max free space that we have found
1209 * until now, so max_hole_start must point to the start
1210 * of this free space and the length of this free space
1211 * is stored in max_hole_size. Thus, we return
1212 * max_hole_start and max_hole_size and go back to the
1215 if (hole_size
>= num_bytes
) {
1221 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1222 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1224 if (extent_end
> search_start
)
1225 search_start
= extent_end
;
1232 * At this point, search_start should be the end of
1233 * allocated dev extents, and when shrinking the device,
1234 * search_end may be smaller than search_start.
1236 if (search_end
> search_start
)
1237 hole_size
= search_end
- search_start
;
1239 if (hole_size
> max_hole_size
) {
1240 max_hole_start
= search_start
;
1241 max_hole_size
= hole_size
;
1244 if (contains_pending_extent(trans
, device
, &search_start
, hole_size
)) {
1245 btrfs_release_path(path
);
1250 if (hole_size
< num_bytes
)
1256 btrfs_free_path(path
);
1257 *start
= max_hole_start
;
1259 *len
= max_hole_size
;
1263 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1264 struct btrfs_device
*device
,
1265 u64 start
, u64
*dev_extent_len
)
1268 struct btrfs_path
*path
;
1269 struct btrfs_root
*root
= device
->dev_root
;
1270 struct btrfs_key key
;
1271 struct btrfs_key found_key
;
1272 struct extent_buffer
*leaf
= NULL
;
1273 struct btrfs_dev_extent
*extent
= NULL
;
1275 path
= btrfs_alloc_path();
1279 key
.objectid
= device
->devid
;
1281 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1283 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1285 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1286 BTRFS_DEV_EXTENT_KEY
);
1289 leaf
= path
->nodes
[0];
1290 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1291 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1292 struct btrfs_dev_extent
);
1293 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1294 btrfs_dev_extent_length(leaf
, extent
) < start
);
1296 btrfs_release_path(path
);
1298 } else if (ret
== 0) {
1299 leaf
= path
->nodes
[0];
1300 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1301 struct btrfs_dev_extent
);
1303 btrfs_error(root
->fs_info
, ret
, "Slot search failed");
1307 *dev_extent_len
= btrfs_dev_extent_length(leaf
, extent
);
1309 ret
= btrfs_del_item(trans
, root
, path
);
1311 btrfs_error(root
->fs_info
, ret
,
1312 "Failed to remove dev extent item");
1314 trans
->transaction
->have_free_bgs
= 1;
1317 btrfs_free_path(path
);
1321 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1322 struct btrfs_device
*device
,
1323 u64 chunk_tree
, u64 chunk_objectid
,
1324 u64 chunk_offset
, u64 start
, u64 num_bytes
)
1327 struct btrfs_path
*path
;
1328 struct btrfs_root
*root
= device
->dev_root
;
1329 struct btrfs_dev_extent
*extent
;
1330 struct extent_buffer
*leaf
;
1331 struct btrfs_key key
;
1333 WARN_ON(!device
->in_fs_metadata
);
1334 WARN_ON(device
->is_tgtdev_for_dev_replace
);
1335 path
= btrfs_alloc_path();
1339 key
.objectid
= device
->devid
;
1341 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1342 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1347 leaf
= path
->nodes
[0];
1348 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1349 struct btrfs_dev_extent
);
1350 btrfs_set_dev_extent_chunk_tree(leaf
, extent
, chunk_tree
);
1351 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
, chunk_objectid
);
1352 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1354 write_extent_buffer(leaf
, root
->fs_info
->chunk_tree_uuid
,
1355 btrfs_dev_extent_chunk_tree_uuid(extent
), BTRFS_UUID_SIZE
);
1357 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1358 btrfs_mark_buffer_dirty(leaf
);
1360 btrfs_free_path(path
);
1364 static u64
find_next_chunk(struct btrfs_fs_info
*fs_info
)
1366 struct extent_map_tree
*em_tree
;
1367 struct extent_map
*em
;
1371 em_tree
= &fs_info
->mapping_tree
.map_tree
;
1372 read_lock(&em_tree
->lock
);
1373 n
= rb_last(&em_tree
->map
);
1375 em
= rb_entry(n
, struct extent_map
, rb_node
);
1376 ret
= em
->start
+ em
->len
;
1378 read_unlock(&em_tree
->lock
);
1383 static noinline
int find_next_devid(struct btrfs_fs_info
*fs_info
,
1387 struct btrfs_key key
;
1388 struct btrfs_key found_key
;
1389 struct btrfs_path
*path
;
1391 path
= btrfs_alloc_path();
1395 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1396 key
.type
= BTRFS_DEV_ITEM_KEY
;
1397 key
.offset
= (u64
)-1;
1399 ret
= btrfs_search_slot(NULL
, fs_info
->chunk_root
, &key
, path
, 0, 0);
1403 BUG_ON(ret
== 0); /* Corruption */
1405 ret
= btrfs_previous_item(fs_info
->chunk_root
, path
,
1406 BTRFS_DEV_ITEMS_OBJECTID
,
1407 BTRFS_DEV_ITEM_KEY
);
1411 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1413 *devid_ret
= found_key
.offset
+ 1;
1417 btrfs_free_path(path
);
1422 * the device information is stored in the chunk root
1423 * the btrfs_device struct should be fully filled in
1425 static int btrfs_add_device(struct btrfs_trans_handle
*trans
,
1426 struct btrfs_root
*root
,
1427 struct btrfs_device
*device
)
1430 struct btrfs_path
*path
;
1431 struct btrfs_dev_item
*dev_item
;
1432 struct extent_buffer
*leaf
;
1433 struct btrfs_key key
;
1436 root
= root
->fs_info
->chunk_root
;
1438 path
= btrfs_alloc_path();
1442 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1443 key
.type
= BTRFS_DEV_ITEM_KEY
;
1444 key
.offset
= device
->devid
;
1446 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1451 leaf
= path
->nodes
[0];
1452 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1454 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1455 btrfs_set_device_generation(leaf
, dev_item
, 0);
1456 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1457 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1458 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1459 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1460 btrfs_set_device_total_bytes(leaf
, dev_item
,
1461 btrfs_device_get_disk_total_bytes(device
));
1462 btrfs_set_device_bytes_used(leaf
, dev_item
,
1463 btrfs_device_get_bytes_used(device
));
1464 btrfs_set_device_group(leaf
, dev_item
, 0);
1465 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1466 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1467 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1469 ptr
= btrfs_device_uuid(dev_item
);
1470 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1471 ptr
= btrfs_device_fsid(dev_item
);
1472 write_extent_buffer(leaf
, root
->fs_info
->fsid
, ptr
, BTRFS_UUID_SIZE
);
1473 btrfs_mark_buffer_dirty(leaf
);
1477 btrfs_free_path(path
);
1482 * Function to update ctime/mtime for a given device path.
1483 * Mainly used for ctime/mtime based probe like libblkid.
1485 static void update_dev_time(char *path_name
)
1489 filp
= filp_open(path_name
, O_RDWR
, 0);
1492 file_update_time(filp
);
1493 filp_close(filp
, NULL
);
1497 static int btrfs_rm_dev_item(struct btrfs_root
*root
,
1498 struct btrfs_device
*device
)
1501 struct btrfs_path
*path
;
1502 struct btrfs_key key
;
1503 struct btrfs_trans_handle
*trans
;
1505 root
= root
->fs_info
->chunk_root
;
1507 path
= btrfs_alloc_path();
1511 trans
= btrfs_start_transaction(root
, 0);
1512 if (IS_ERR(trans
)) {
1513 btrfs_free_path(path
);
1514 return PTR_ERR(trans
);
1516 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1517 key
.type
= BTRFS_DEV_ITEM_KEY
;
1518 key
.offset
= device
->devid
;
1520 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1529 ret
= btrfs_del_item(trans
, root
, path
);
1533 btrfs_free_path(path
);
1534 btrfs_commit_transaction(trans
, root
);
1538 int btrfs_rm_device(struct btrfs_root
*root
, char *device_path
)
1540 struct btrfs_device
*device
;
1541 struct btrfs_device
*next_device
;
1542 struct block_device
*bdev
;
1543 struct buffer_head
*bh
= NULL
;
1544 struct btrfs_super_block
*disk_super
;
1545 struct btrfs_fs_devices
*cur_devices
;
1552 bool clear_super
= false;
1554 mutex_lock(&uuid_mutex
);
1557 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
1559 all_avail
= root
->fs_info
->avail_data_alloc_bits
|
1560 root
->fs_info
->avail_system_alloc_bits
|
1561 root
->fs_info
->avail_metadata_alloc_bits
;
1562 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
1564 num_devices
= root
->fs_info
->fs_devices
->num_devices
;
1565 btrfs_dev_replace_lock(&root
->fs_info
->dev_replace
);
1566 if (btrfs_dev_replace_is_ongoing(&root
->fs_info
->dev_replace
)) {
1567 WARN_ON(num_devices
< 1);
1570 btrfs_dev_replace_unlock(&root
->fs_info
->dev_replace
);
1572 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID10
) && num_devices
<= 4) {
1573 ret
= BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET
;
1577 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID1
) && num_devices
<= 2) {
1578 ret
= BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET
;
1582 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID5
) &&
1583 root
->fs_info
->fs_devices
->rw_devices
<= 2) {
1584 ret
= BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET
;
1587 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID6
) &&
1588 root
->fs_info
->fs_devices
->rw_devices
<= 3) {
1589 ret
= BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET
;
1593 if (strcmp(device_path
, "missing") == 0) {
1594 struct list_head
*devices
;
1595 struct btrfs_device
*tmp
;
1598 devices
= &root
->fs_info
->fs_devices
->devices
;
1600 * It is safe to read the devices since the volume_mutex
1603 list_for_each_entry(tmp
, devices
, dev_list
) {
1604 if (tmp
->in_fs_metadata
&&
1605 !tmp
->is_tgtdev_for_dev_replace
&&
1615 ret
= BTRFS_ERROR_DEV_MISSING_NOT_FOUND
;
1619 ret
= btrfs_get_bdev_and_sb(device_path
,
1620 FMODE_WRITE
| FMODE_EXCL
,
1621 root
->fs_info
->bdev_holder
, 0,
1625 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1626 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1627 dev_uuid
= disk_super
->dev_item
.uuid
;
1628 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1636 if (device
->is_tgtdev_for_dev_replace
) {
1637 ret
= BTRFS_ERROR_DEV_TGT_REPLACE
;
1641 if (device
->writeable
&& root
->fs_info
->fs_devices
->rw_devices
== 1) {
1642 ret
= BTRFS_ERROR_DEV_ONLY_WRITABLE
;
1646 if (device
->writeable
) {
1648 list_del_init(&device
->dev_alloc_list
);
1649 device
->fs_devices
->rw_devices
--;
1650 unlock_chunks(root
);
1654 mutex_unlock(&uuid_mutex
);
1655 ret
= btrfs_shrink_device(device
, 0);
1656 mutex_lock(&uuid_mutex
);
1661 * TODO: the superblock still includes this device in its num_devices
1662 * counter although write_all_supers() is not locked out. This
1663 * could give a filesystem state which requires a degraded mount.
1665 ret
= btrfs_rm_dev_item(root
->fs_info
->chunk_root
, device
);
1669 device
->in_fs_metadata
= 0;
1670 btrfs_scrub_cancel_dev(root
->fs_info
, device
);
1673 * the device list mutex makes sure that we don't change
1674 * the device list while someone else is writing out all
1675 * the device supers. Whoever is writing all supers, should
1676 * lock the device list mutex before getting the number of
1677 * devices in the super block (super_copy). Conversely,
1678 * whoever updates the number of devices in the super block
1679 * (super_copy) should hold the device list mutex.
1682 cur_devices
= device
->fs_devices
;
1683 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1684 list_del_rcu(&device
->dev_list
);
1686 device
->fs_devices
->num_devices
--;
1687 device
->fs_devices
->total_devices
--;
1689 if (device
->missing
)
1690 device
->fs_devices
->missing_devices
--;
1692 next_device
= list_entry(root
->fs_info
->fs_devices
->devices
.next
,
1693 struct btrfs_device
, dev_list
);
1694 if (device
->bdev
== root
->fs_info
->sb
->s_bdev
)
1695 root
->fs_info
->sb
->s_bdev
= next_device
->bdev
;
1696 if (device
->bdev
== root
->fs_info
->fs_devices
->latest_bdev
)
1697 root
->fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1700 device
->fs_devices
->open_devices
--;
1701 /* remove sysfs entry */
1702 btrfs_kobj_rm_device(root
->fs_info
, device
);
1705 call_rcu(&device
->rcu
, free_device
);
1707 num_devices
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
1708 btrfs_set_super_num_devices(root
->fs_info
->super_copy
, num_devices
);
1709 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1711 if (cur_devices
->open_devices
== 0) {
1712 struct btrfs_fs_devices
*fs_devices
;
1713 fs_devices
= root
->fs_info
->fs_devices
;
1714 while (fs_devices
) {
1715 if (fs_devices
->seed
== cur_devices
) {
1716 fs_devices
->seed
= cur_devices
->seed
;
1719 fs_devices
= fs_devices
->seed
;
1721 cur_devices
->seed
= NULL
;
1722 __btrfs_close_devices(cur_devices
);
1723 free_fs_devices(cur_devices
);
1726 root
->fs_info
->num_tolerated_disk_barrier_failures
=
1727 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
1730 * at this point, the device is zero sized. We want to
1731 * remove it from the devices list and zero out the old super
1733 if (clear_super
&& disk_super
) {
1737 /* make sure this device isn't detected as part of
1740 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
1741 set_buffer_dirty(bh
);
1742 sync_dirty_buffer(bh
);
1744 /* clear the mirror copies of super block on the disk
1745 * being removed, 0th copy is been taken care above and
1746 * the below would take of the rest
1748 for (i
= 1; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
1749 bytenr
= btrfs_sb_offset(i
);
1750 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
1751 i_size_read(bdev
->bd_inode
))
1755 bh
= __bread(bdev
, bytenr
/ 4096,
1756 BTRFS_SUPER_INFO_SIZE
);
1760 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1762 if (btrfs_super_bytenr(disk_super
) != bytenr
||
1763 btrfs_super_magic(disk_super
) != BTRFS_MAGIC
) {
1766 memset(&disk_super
->magic
, 0,
1767 sizeof(disk_super
->magic
));
1768 set_buffer_dirty(bh
);
1769 sync_dirty_buffer(bh
);
1776 /* Notify udev that device has changed */
1777 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
1779 /* Update ctime/mtime for device path for libblkid */
1780 update_dev_time(device_path
);
1786 blkdev_put(bdev
, FMODE_READ
| FMODE_EXCL
);
1788 mutex_unlock(&uuid_mutex
);
1791 if (device
->writeable
) {
1793 list_add(&device
->dev_alloc_list
,
1794 &root
->fs_info
->fs_devices
->alloc_list
);
1795 device
->fs_devices
->rw_devices
++;
1796 unlock_chunks(root
);
1801 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info
*fs_info
,
1802 struct btrfs_device
*srcdev
)
1804 struct btrfs_fs_devices
*fs_devices
;
1806 WARN_ON(!mutex_is_locked(&fs_info
->fs_devices
->device_list_mutex
));
1809 * in case of fs with no seed, srcdev->fs_devices will point
1810 * to fs_devices of fs_info. However when the dev being replaced is
1811 * a seed dev it will point to the seed's local fs_devices. In short
1812 * srcdev will have its correct fs_devices in both the cases.
1814 fs_devices
= srcdev
->fs_devices
;
1816 list_del_rcu(&srcdev
->dev_list
);
1817 list_del_rcu(&srcdev
->dev_alloc_list
);
1818 fs_devices
->num_devices
--;
1819 if (srcdev
->missing
)
1820 fs_devices
->missing_devices
--;
1822 if (srcdev
->writeable
) {
1823 fs_devices
->rw_devices
--;
1824 /* zero out the old super if it is writable */
1825 btrfs_scratch_superblock(srcdev
);
1829 fs_devices
->open_devices
--;
1832 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info
*fs_info
,
1833 struct btrfs_device
*srcdev
)
1835 struct btrfs_fs_devices
*fs_devices
= srcdev
->fs_devices
;
1837 call_rcu(&srcdev
->rcu
, free_device
);
1840 * unless fs_devices is seed fs, num_devices shouldn't go
1843 BUG_ON(!fs_devices
->num_devices
&& !fs_devices
->seeding
);
1845 /* if this is no devs we rather delete the fs_devices */
1846 if (!fs_devices
->num_devices
) {
1847 struct btrfs_fs_devices
*tmp_fs_devices
;
1849 tmp_fs_devices
= fs_info
->fs_devices
;
1850 while (tmp_fs_devices
) {
1851 if (tmp_fs_devices
->seed
== fs_devices
) {
1852 tmp_fs_devices
->seed
= fs_devices
->seed
;
1855 tmp_fs_devices
= tmp_fs_devices
->seed
;
1857 fs_devices
->seed
= NULL
;
1858 __btrfs_close_devices(fs_devices
);
1859 free_fs_devices(fs_devices
);
1863 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
1864 struct btrfs_device
*tgtdev
)
1866 struct btrfs_device
*next_device
;
1868 mutex_lock(&uuid_mutex
);
1870 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
1872 btrfs_scratch_superblock(tgtdev
);
1873 fs_info
->fs_devices
->open_devices
--;
1875 fs_info
->fs_devices
->num_devices
--;
1877 next_device
= list_entry(fs_info
->fs_devices
->devices
.next
,
1878 struct btrfs_device
, dev_list
);
1879 if (tgtdev
->bdev
== fs_info
->sb
->s_bdev
)
1880 fs_info
->sb
->s_bdev
= next_device
->bdev
;
1881 if (tgtdev
->bdev
== fs_info
->fs_devices
->latest_bdev
)
1882 fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1883 list_del_rcu(&tgtdev
->dev_list
);
1885 call_rcu(&tgtdev
->rcu
, free_device
);
1887 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
1888 mutex_unlock(&uuid_mutex
);
1891 static int btrfs_find_device_by_path(struct btrfs_root
*root
, char *device_path
,
1892 struct btrfs_device
**device
)
1895 struct btrfs_super_block
*disk_super
;
1898 struct block_device
*bdev
;
1899 struct buffer_head
*bh
;
1902 ret
= btrfs_get_bdev_and_sb(device_path
, FMODE_READ
,
1903 root
->fs_info
->bdev_holder
, 0, &bdev
, &bh
);
1906 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1907 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1908 dev_uuid
= disk_super
->dev_item
.uuid
;
1909 *device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1914 blkdev_put(bdev
, FMODE_READ
);
1918 int btrfs_find_device_missing_or_by_path(struct btrfs_root
*root
,
1920 struct btrfs_device
**device
)
1923 if (strcmp(device_path
, "missing") == 0) {
1924 struct list_head
*devices
;
1925 struct btrfs_device
*tmp
;
1927 devices
= &root
->fs_info
->fs_devices
->devices
;
1929 * It is safe to read the devices since the volume_mutex
1930 * is held by the caller.
1932 list_for_each_entry(tmp
, devices
, dev_list
) {
1933 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
1940 btrfs_err(root
->fs_info
, "no missing device found");
1946 return btrfs_find_device_by_path(root
, device_path
, device
);
1951 * does all the dirty work required for changing file system's UUID.
1953 static int btrfs_prepare_sprout(struct btrfs_root
*root
)
1955 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
1956 struct btrfs_fs_devices
*old_devices
;
1957 struct btrfs_fs_devices
*seed_devices
;
1958 struct btrfs_super_block
*disk_super
= root
->fs_info
->super_copy
;
1959 struct btrfs_device
*device
;
1962 BUG_ON(!mutex_is_locked(&uuid_mutex
));
1963 if (!fs_devices
->seeding
)
1966 seed_devices
= __alloc_fs_devices();
1967 if (IS_ERR(seed_devices
))
1968 return PTR_ERR(seed_devices
);
1970 old_devices
= clone_fs_devices(fs_devices
);
1971 if (IS_ERR(old_devices
)) {
1972 kfree(seed_devices
);
1973 return PTR_ERR(old_devices
);
1976 list_add(&old_devices
->list
, &fs_uuids
);
1978 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
1979 seed_devices
->opened
= 1;
1980 INIT_LIST_HEAD(&seed_devices
->devices
);
1981 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
1982 mutex_init(&seed_devices
->device_list_mutex
);
1984 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1985 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
1987 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
)
1988 device
->fs_devices
= seed_devices
;
1991 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
1992 unlock_chunks(root
);
1994 fs_devices
->seeding
= 0;
1995 fs_devices
->num_devices
= 0;
1996 fs_devices
->open_devices
= 0;
1997 fs_devices
->missing_devices
= 0;
1998 fs_devices
->rotating
= 0;
1999 fs_devices
->seed
= seed_devices
;
2001 generate_random_uuid(fs_devices
->fsid
);
2002 memcpy(root
->fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2003 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2004 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2006 super_flags
= btrfs_super_flags(disk_super
) &
2007 ~BTRFS_SUPER_FLAG_SEEDING
;
2008 btrfs_set_super_flags(disk_super
, super_flags
);
2014 * strore the expected generation for seed devices in device items.
2016 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
2017 struct btrfs_root
*root
)
2019 struct btrfs_path
*path
;
2020 struct extent_buffer
*leaf
;
2021 struct btrfs_dev_item
*dev_item
;
2022 struct btrfs_device
*device
;
2023 struct btrfs_key key
;
2024 u8 fs_uuid
[BTRFS_UUID_SIZE
];
2025 u8 dev_uuid
[BTRFS_UUID_SIZE
];
2029 path
= btrfs_alloc_path();
2033 root
= root
->fs_info
->chunk_root
;
2034 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2036 key
.type
= BTRFS_DEV_ITEM_KEY
;
2039 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2043 leaf
= path
->nodes
[0];
2045 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
2046 ret
= btrfs_next_leaf(root
, path
);
2051 leaf
= path
->nodes
[0];
2052 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2053 btrfs_release_path(path
);
2057 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2058 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
2059 key
.type
!= BTRFS_DEV_ITEM_KEY
)
2062 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2063 struct btrfs_dev_item
);
2064 devid
= btrfs_device_id(leaf
, dev_item
);
2065 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
2067 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
2069 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
2071 BUG_ON(!device
); /* Logic error */
2073 if (device
->fs_devices
->seeding
) {
2074 btrfs_set_device_generation(leaf
, dev_item
,
2075 device
->generation
);
2076 btrfs_mark_buffer_dirty(leaf
);
2084 btrfs_free_path(path
);
2088 int btrfs_init_new_device(struct btrfs_root
*root
, char *device_path
)
2090 struct request_queue
*q
;
2091 struct btrfs_trans_handle
*trans
;
2092 struct btrfs_device
*device
;
2093 struct block_device
*bdev
;
2094 struct list_head
*devices
;
2095 struct super_block
*sb
= root
->fs_info
->sb
;
2096 struct rcu_string
*name
;
2098 int seeding_dev
= 0;
2101 if ((sb
->s_flags
& MS_RDONLY
) && !root
->fs_info
->fs_devices
->seeding
)
2104 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2105 root
->fs_info
->bdev_holder
);
2107 return PTR_ERR(bdev
);
2109 if (root
->fs_info
->fs_devices
->seeding
) {
2111 down_write(&sb
->s_umount
);
2112 mutex_lock(&uuid_mutex
);
2115 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2117 devices
= &root
->fs_info
->fs_devices
->devices
;
2119 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2120 list_for_each_entry(device
, devices
, dev_list
) {
2121 if (device
->bdev
== bdev
) {
2124 &root
->fs_info
->fs_devices
->device_list_mutex
);
2128 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2130 device
= btrfs_alloc_device(root
->fs_info
, NULL
, NULL
);
2131 if (IS_ERR(device
)) {
2132 /* we can safely leave the fs_devices entry around */
2133 ret
= PTR_ERR(device
);
2137 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2143 rcu_assign_pointer(device
->name
, name
);
2145 trans
= btrfs_start_transaction(root
, 0);
2146 if (IS_ERR(trans
)) {
2147 rcu_string_free(device
->name
);
2149 ret
= PTR_ERR(trans
);
2153 q
= bdev_get_queue(bdev
);
2154 if (blk_queue_discard(q
))
2155 device
->can_discard
= 1;
2156 device
->writeable
= 1;
2157 device
->generation
= trans
->transid
;
2158 device
->io_width
= root
->sectorsize
;
2159 device
->io_align
= root
->sectorsize
;
2160 device
->sector_size
= root
->sectorsize
;
2161 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
2162 device
->disk_total_bytes
= device
->total_bytes
;
2163 device
->commit_total_bytes
= device
->total_bytes
;
2164 device
->dev_root
= root
->fs_info
->dev_root
;
2165 device
->bdev
= bdev
;
2166 device
->in_fs_metadata
= 1;
2167 device
->is_tgtdev_for_dev_replace
= 0;
2168 device
->mode
= FMODE_EXCL
;
2169 device
->dev_stats_valid
= 1;
2170 set_blocksize(device
->bdev
, 4096);
2173 sb
->s_flags
&= ~MS_RDONLY
;
2174 ret
= btrfs_prepare_sprout(root
);
2175 BUG_ON(ret
); /* -ENOMEM */
2178 device
->fs_devices
= root
->fs_info
->fs_devices
;
2180 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2182 list_add_rcu(&device
->dev_list
, &root
->fs_info
->fs_devices
->devices
);
2183 list_add(&device
->dev_alloc_list
,
2184 &root
->fs_info
->fs_devices
->alloc_list
);
2185 root
->fs_info
->fs_devices
->num_devices
++;
2186 root
->fs_info
->fs_devices
->open_devices
++;
2187 root
->fs_info
->fs_devices
->rw_devices
++;
2188 root
->fs_info
->fs_devices
->total_devices
++;
2189 root
->fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
2191 spin_lock(&root
->fs_info
->free_chunk_lock
);
2192 root
->fs_info
->free_chunk_space
+= device
->total_bytes
;
2193 spin_unlock(&root
->fs_info
->free_chunk_lock
);
2195 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
2196 root
->fs_info
->fs_devices
->rotating
= 1;
2198 tmp
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
2199 btrfs_set_super_total_bytes(root
->fs_info
->super_copy
,
2200 tmp
+ device
->total_bytes
);
2202 tmp
= btrfs_super_num_devices(root
->fs_info
->super_copy
);
2203 btrfs_set_super_num_devices(root
->fs_info
->super_copy
,
2206 /* add sysfs device entry */
2207 btrfs_kobj_add_device(root
->fs_info
, device
);
2210 * we've got more storage, clear any full flags on the space
2213 btrfs_clear_space_info_full(root
->fs_info
);
2215 unlock_chunks(root
);
2216 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2220 ret
= init_first_rw_device(trans
, root
, device
);
2221 unlock_chunks(root
);
2223 btrfs_abort_transaction(trans
, root
, ret
);
2228 ret
= btrfs_add_device(trans
, root
, device
);
2230 btrfs_abort_transaction(trans
, root
, ret
);
2235 char fsid_buf
[BTRFS_UUID_UNPARSED_SIZE
];
2237 ret
= btrfs_finish_sprout(trans
, root
);
2239 btrfs_abort_transaction(trans
, root
, ret
);
2243 /* Sprouting would change fsid of the mounted root,
2244 * so rename the fsid on the sysfs
2246 snprintf(fsid_buf
, BTRFS_UUID_UNPARSED_SIZE
, "%pU",
2247 root
->fs_info
->fsid
);
2248 if (kobject_rename(&root
->fs_info
->super_kobj
, fsid_buf
))
2252 root
->fs_info
->num_tolerated_disk_barrier_failures
=
2253 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
2254 ret
= btrfs_commit_transaction(trans
, root
);
2257 mutex_unlock(&uuid_mutex
);
2258 up_write(&sb
->s_umount
);
2260 if (ret
) /* transaction commit */
2263 ret
= btrfs_relocate_sys_chunks(root
);
2265 btrfs_error(root
->fs_info
, ret
,
2266 "Failed to relocate sys chunks after "
2267 "device initialization. This can be fixed "
2268 "using the \"btrfs balance\" command.");
2269 trans
= btrfs_attach_transaction(root
);
2270 if (IS_ERR(trans
)) {
2271 if (PTR_ERR(trans
) == -ENOENT
)
2273 return PTR_ERR(trans
);
2275 ret
= btrfs_commit_transaction(trans
, root
);
2278 /* Update ctime/mtime for libblkid */
2279 update_dev_time(device_path
);
2283 btrfs_end_transaction(trans
, root
);
2284 rcu_string_free(device
->name
);
2285 btrfs_kobj_rm_device(root
->fs_info
, device
);
2288 blkdev_put(bdev
, FMODE_EXCL
);
2290 mutex_unlock(&uuid_mutex
);
2291 up_write(&sb
->s_umount
);
2296 int btrfs_init_dev_replace_tgtdev(struct btrfs_root
*root
, char *device_path
,
2297 struct btrfs_device
*srcdev
,
2298 struct btrfs_device
**device_out
)
2300 struct request_queue
*q
;
2301 struct btrfs_device
*device
;
2302 struct block_device
*bdev
;
2303 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2304 struct list_head
*devices
;
2305 struct rcu_string
*name
;
2306 u64 devid
= BTRFS_DEV_REPLACE_DEVID
;
2310 if (fs_info
->fs_devices
->seeding
) {
2311 btrfs_err(fs_info
, "the filesystem is a seed filesystem!");
2315 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2316 fs_info
->bdev_holder
);
2318 btrfs_err(fs_info
, "target device %s is invalid!", device_path
);
2319 return PTR_ERR(bdev
);
2322 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2324 devices
= &fs_info
->fs_devices
->devices
;
2325 list_for_each_entry(device
, devices
, dev_list
) {
2326 if (device
->bdev
== bdev
) {
2327 btrfs_err(fs_info
, "target device is in the filesystem!");
2334 if (i_size_read(bdev
->bd_inode
) <
2335 btrfs_device_get_total_bytes(srcdev
)) {
2336 btrfs_err(fs_info
, "target device is smaller than source device!");
2342 device
= btrfs_alloc_device(NULL
, &devid
, NULL
);
2343 if (IS_ERR(device
)) {
2344 ret
= PTR_ERR(device
);
2348 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2354 rcu_assign_pointer(device
->name
, name
);
2356 q
= bdev_get_queue(bdev
);
2357 if (blk_queue_discard(q
))
2358 device
->can_discard
= 1;
2359 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2360 device
->writeable
= 1;
2361 device
->generation
= 0;
2362 device
->io_width
= root
->sectorsize
;
2363 device
->io_align
= root
->sectorsize
;
2364 device
->sector_size
= root
->sectorsize
;
2365 device
->total_bytes
= btrfs_device_get_total_bytes(srcdev
);
2366 device
->disk_total_bytes
= btrfs_device_get_disk_total_bytes(srcdev
);
2367 device
->bytes_used
= btrfs_device_get_bytes_used(srcdev
);
2368 ASSERT(list_empty(&srcdev
->resized_list
));
2369 device
->commit_total_bytes
= srcdev
->commit_total_bytes
;
2370 device
->commit_bytes_used
= device
->bytes_used
;
2371 device
->dev_root
= fs_info
->dev_root
;
2372 device
->bdev
= bdev
;
2373 device
->in_fs_metadata
= 1;
2374 device
->is_tgtdev_for_dev_replace
= 1;
2375 device
->mode
= FMODE_EXCL
;
2376 device
->dev_stats_valid
= 1;
2377 set_blocksize(device
->bdev
, 4096);
2378 device
->fs_devices
= fs_info
->fs_devices
;
2379 list_add(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2380 fs_info
->fs_devices
->num_devices
++;
2381 fs_info
->fs_devices
->open_devices
++;
2382 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2384 *device_out
= device
;
2388 blkdev_put(bdev
, FMODE_EXCL
);
2392 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info
*fs_info
,
2393 struct btrfs_device
*tgtdev
)
2395 WARN_ON(fs_info
->fs_devices
->rw_devices
== 0);
2396 tgtdev
->io_width
= fs_info
->dev_root
->sectorsize
;
2397 tgtdev
->io_align
= fs_info
->dev_root
->sectorsize
;
2398 tgtdev
->sector_size
= fs_info
->dev_root
->sectorsize
;
2399 tgtdev
->dev_root
= fs_info
->dev_root
;
2400 tgtdev
->in_fs_metadata
= 1;
2403 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2404 struct btrfs_device
*device
)
2407 struct btrfs_path
*path
;
2408 struct btrfs_root
*root
;
2409 struct btrfs_dev_item
*dev_item
;
2410 struct extent_buffer
*leaf
;
2411 struct btrfs_key key
;
2413 root
= device
->dev_root
->fs_info
->chunk_root
;
2415 path
= btrfs_alloc_path();
2419 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2420 key
.type
= BTRFS_DEV_ITEM_KEY
;
2421 key
.offset
= device
->devid
;
2423 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2432 leaf
= path
->nodes
[0];
2433 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2435 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2436 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2437 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2438 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2439 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2440 btrfs_set_device_total_bytes(leaf
, dev_item
,
2441 btrfs_device_get_disk_total_bytes(device
));
2442 btrfs_set_device_bytes_used(leaf
, dev_item
,
2443 btrfs_device_get_bytes_used(device
));
2444 btrfs_mark_buffer_dirty(leaf
);
2447 btrfs_free_path(path
);
2451 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2452 struct btrfs_device
*device
, u64 new_size
)
2454 struct btrfs_super_block
*super_copy
=
2455 device
->dev_root
->fs_info
->super_copy
;
2456 struct btrfs_fs_devices
*fs_devices
;
2460 if (!device
->writeable
)
2463 lock_chunks(device
->dev_root
);
2464 old_total
= btrfs_super_total_bytes(super_copy
);
2465 diff
= new_size
- device
->total_bytes
;
2467 if (new_size
<= device
->total_bytes
||
2468 device
->is_tgtdev_for_dev_replace
) {
2469 unlock_chunks(device
->dev_root
);
2473 fs_devices
= device
->dev_root
->fs_info
->fs_devices
;
2475 btrfs_set_super_total_bytes(super_copy
, old_total
+ diff
);
2476 device
->fs_devices
->total_rw_bytes
+= diff
;
2478 btrfs_device_set_total_bytes(device
, new_size
);
2479 btrfs_device_set_disk_total_bytes(device
, new_size
);
2480 btrfs_clear_space_info_full(device
->dev_root
->fs_info
);
2481 if (list_empty(&device
->resized_list
))
2482 list_add_tail(&device
->resized_list
,
2483 &fs_devices
->resized_devices
);
2484 unlock_chunks(device
->dev_root
);
2486 return btrfs_update_device(trans
, device
);
2489 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
2490 struct btrfs_root
*root
, u64 chunk_objectid
,
2494 struct btrfs_path
*path
;
2495 struct btrfs_key key
;
2497 root
= root
->fs_info
->chunk_root
;
2498 path
= btrfs_alloc_path();
2502 key
.objectid
= chunk_objectid
;
2503 key
.offset
= chunk_offset
;
2504 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2506 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2509 else if (ret
> 0) { /* Logic error or corruption */
2510 btrfs_error(root
->fs_info
, -ENOENT
,
2511 "Failed lookup while freeing chunk.");
2516 ret
= btrfs_del_item(trans
, root
, path
);
2518 btrfs_error(root
->fs_info
, ret
,
2519 "Failed to delete chunk item.");
2521 btrfs_free_path(path
);
2525 static int btrfs_del_sys_chunk(struct btrfs_root
*root
, u64 chunk_objectid
, u64
2528 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
2529 struct btrfs_disk_key
*disk_key
;
2530 struct btrfs_chunk
*chunk
;
2537 struct btrfs_key key
;
2540 array_size
= btrfs_super_sys_array_size(super_copy
);
2542 ptr
= super_copy
->sys_chunk_array
;
2545 while (cur
< array_size
) {
2546 disk_key
= (struct btrfs_disk_key
*)ptr
;
2547 btrfs_disk_key_to_cpu(&key
, disk_key
);
2549 len
= sizeof(*disk_key
);
2551 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2552 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2553 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2554 len
+= btrfs_chunk_item_size(num_stripes
);
2559 if (key
.objectid
== chunk_objectid
&&
2560 key
.offset
== chunk_offset
) {
2561 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2563 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2569 unlock_chunks(root
);
2573 int btrfs_remove_chunk(struct btrfs_trans_handle
*trans
,
2574 struct btrfs_root
*root
, u64 chunk_offset
)
2576 struct extent_map_tree
*em_tree
;
2577 struct extent_map
*em
;
2578 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2579 struct map_lookup
*map
;
2580 u64 dev_extent_len
= 0;
2581 u64 chunk_objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2585 root
= root
->fs_info
->chunk_root
;
2586 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
2588 read_lock(&em_tree
->lock
);
2589 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
2590 read_unlock(&em_tree
->lock
);
2592 if (!em
|| em
->start
> chunk_offset
||
2593 em
->start
+ em
->len
< chunk_offset
) {
2595 * This is a logic error, but we don't want to just rely on the
2596 * user having built with ASSERT enabled, so if ASSERT doens't
2597 * do anything we still error out.
2601 free_extent_map(em
);
2604 map
= (struct map_lookup
*)em
->bdev
;
2606 for (i
= 0; i
< map
->num_stripes
; i
++) {
2607 struct btrfs_device
*device
= map
->stripes
[i
].dev
;
2608 ret
= btrfs_free_dev_extent(trans
, device
,
2609 map
->stripes
[i
].physical
,
2612 btrfs_abort_transaction(trans
, root
, ret
);
2616 if (device
->bytes_used
> 0) {
2618 btrfs_device_set_bytes_used(device
,
2619 device
->bytes_used
- dev_extent_len
);
2620 spin_lock(&root
->fs_info
->free_chunk_lock
);
2621 root
->fs_info
->free_chunk_space
+= dev_extent_len
;
2622 spin_unlock(&root
->fs_info
->free_chunk_lock
);
2623 btrfs_clear_space_info_full(root
->fs_info
);
2624 unlock_chunks(root
);
2627 if (map
->stripes
[i
].dev
) {
2628 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2630 btrfs_abort_transaction(trans
, root
, ret
);
2635 ret
= btrfs_free_chunk(trans
, root
, chunk_objectid
, chunk_offset
);
2637 btrfs_abort_transaction(trans
, root
, ret
);
2641 trace_btrfs_chunk_free(root
, map
, chunk_offset
, em
->len
);
2643 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2644 ret
= btrfs_del_sys_chunk(root
, chunk_objectid
, chunk_offset
);
2646 btrfs_abort_transaction(trans
, root
, ret
);
2651 ret
= btrfs_remove_block_group(trans
, extent_root
, chunk_offset
, em
);
2653 btrfs_abort_transaction(trans
, extent_root
, ret
);
2659 free_extent_map(em
);
2663 static int btrfs_relocate_chunk(struct btrfs_root
*root
,
2667 struct btrfs_root
*extent_root
;
2668 struct btrfs_trans_handle
*trans
;
2671 root
= root
->fs_info
->chunk_root
;
2672 extent_root
= root
->fs_info
->extent_root
;
2674 ret
= btrfs_can_relocate(extent_root
, chunk_offset
);
2678 /* step one, relocate all the extents inside this chunk */
2679 ret
= btrfs_relocate_block_group(extent_root
, chunk_offset
);
2683 trans
= btrfs_start_transaction(root
, 0);
2684 if (IS_ERR(trans
)) {
2685 ret
= PTR_ERR(trans
);
2686 btrfs_std_error(root
->fs_info
, ret
);
2691 * step two, delete the device extents and the
2692 * chunk tree entries
2694 ret
= btrfs_remove_chunk(trans
, root
, chunk_offset
);
2695 btrfs_end_transaction(trans
, root
);
2699 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
)
2701 struct btrfs_root
*chunk_root
= root
->fs_info
->chunk_root
;
2702 struct btrfs_path
*path
;
2703 struct extent_buffer
*leaf
;
2704 struct btrfs_chunk
*chunk
;
2705 struct btrfs_key key
;
2706 struct btrfs_key found_key
;
2708 bool retried
= false;
2712 path
= btrfs_alloc_path();
2717 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2718 key
.offset
= (u64
)-1;
2719 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2722 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2725 BUG_ON(ret
== 0); /* Corruption */
2727 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
2734 leaf
= path
->nodes
[0];
2735 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2737 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
2738 struct btrfs_chunk
);
2739 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2740 btrfs_release_path(path
);
2742 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2743 ret
= btrfs_relocate_chunk(chunk_root
,
2752 if (found_key
.offset
== 0)
2754 key
.offset
= found_key
.offset
- 1;
2757 if (failed
&& !retried
) {
2761 } else if (WARN_ON(failed
&& retried
)) {
2765 btrfs_free_path(path
);
2769 static int insert_balance_item(struct btrfs_root
*root
,
2770 struct btrfs_balance_control
*bctl
)
2772 struct btrfs_trans_handle
*trans
;
2773 struct btrfs_balance_item
*item
;
2774 struct btrfs_disk_balance_args disk_bargs
;
2775 struct btrfs_path
*path
;
2776 struct extent_buffer
*leaf
;
2777 struct btrfs_key key
;
2780 path
= btrfs_alloc_path();
2784 trans
= btrfs_start_transaction(root
, 0);
2785 if (IS_ERR(trans
)) {
2786 btrfs_free_path(path
);
2787 return PTR_ERR(trans
);
2790 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2791 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2794 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2799 leaf
= path
->nodes
[0];
2800 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
2802 memset_extent_buffer(leaf
, 0, (unsigned long)item
, sizeof(*item
));
2804 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
2805 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
2806 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
2807 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
2808 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
2809 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
2811 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
2813 btrfs_mark_buffer_dirty(leaf
);
2815 btrfs_free_path(path
);
2816 err
= btrfs_commit_transaction(trans
, root
);
2822 static int del_balance_item(struct btrfs_root
*root
)
2824 struct btrfs_trans_handle
*trans
;
2825 struct btrfs_path
*path
;
2826 struct btrfs_key key
;
2829 path
= btrfs_alloc_path();
2833 trans
= btrfs_start_transaction(root
, 0);
2834 if (IS_ERR(trans
)) {
2835 btrfs_free_path(path
);
2836 return PTR_ERR(trans
);
2839 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2840 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2843 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2851 ret
= btrfs_del_item(trans
, root
, path
);
2853 btrfs_free_path(path
);
2854 err
= btrfs_commit_transaction(trans
, root
);
2861 * This is a heuristic used to reduce the number of chunks balanced on
2862 * resume after balance was interrupted.
2864 static void update_balance_args(struct btrfs_balance_control
*bctl
)
2867 * Turn on soft mode for chunk types that were being converted.
2869 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2870 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2871 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2872 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2873 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2874 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2877 * Turn on usage filter if is not already used. The idea is
2878 * that chunks that we have already balanced should be
2879 * reasonably full. Don't do it for chunks that are being
2880 * converted - that will keep us from relocating unconverted
2881 * (albeit full) chunks.
2883 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2884 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2885 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2886 bctl
->data
.usage
= 90;
2888 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2889 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2890 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2891 bctl
->sys
.usage
= 90;
2893 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2894 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2895 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2896 bctl
->meta
.usage
= 90;
2901 * Should be called with both balance and volume mutexes held to
2902 * serialize other volume operations (add_dev/rm_dev/resize) with
2903 * restriper. Same goes for unset_balance_control.
2905 static void set_balance_control(struct btrfs_balance_control
*bctl
)
2907 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
2909 BUG_ON(fs_info
->balance_ctl
);
2911 spin_lock(&fs_info
->balance_lock
);
2912 fs_info
->balance_ctl
= bctl
;
2913 spin_unlock(&fs_info
->balance_lock
);
2916 static void unset_balance_control(struct btrfs_fs_info
*fs_info
)
2918 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
2920 BUG_ON(!fs_info
->balance_ctl
);
2922 spin_lock(&fs_info
->balance_lock
);
2923 fs_info
->balance_ctl
= NULL
;
2924 spin_unlock(&fs_info
->balance_lock
);
2930 * Balance filters. Return 1 if chunk should be filtered out
2931 * (should not be balanced).
2933 static int chunk_profiles_filter(u64 chunk_type
,
2934 struct btrfs_balance_args
*bargs
)
2936 chunk_type
= chunk_to_extended(chunk_type
) &
2937 BTRFS_EXTENDED_PROFILE_MASK
;
2939 if (bargs
->profiles
& chunk_type
)
2945 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
2946 struct btrfs_balance_args
*bargs
)
2948 struct btrfs_block_group_cache
*cache
;
2949 u64 chunk_used
, user_thresh
;
2952 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
2953 chunk_used
= btrfs_block_group_used(&cache
->item
);
2955 if (bargs
->usage
== 0)
2957 else if (bargs
->usage
> 100)
2958 user_thresh
= cache
->key
.offset
;
2960 user_thresh
= div_factor_fine(cache
->key
.offset
,
2963 if (chunk_used
< user_thresh
)
2966 btrfs_put_block_group(cache
);
2970 static int chunk_devid_filter(struct extent_buffer
*leaf
,
2971 struct btrfs_chunk
*chunk
,
2972 struct btrfs_balance_args
*bargs
)
2974 struct btrfs_stripe
*stripe
;
2975 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2978 for (i
= 0; i
< num_stripes
; i
++) {
2979 stripe
= btrfs_stripe_nr(chunk
, i
);
2980 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
2987 /* [pstart, pend) */
2988 static int chunk_drange_filter(struct extent_buffer
*leaf
,
2989 struct btrfs_chunk
*chunk
,
2991 struct btrfs_balance_args
*bargs
)
2993 struct btrfs_stripe
*stripe
;
2994 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3000 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
3003 if (btrfs_chunk_type(leaf
, chunk
) & (BTRFS_BLOCK_GROUP_DUP
|
3004 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)) {
3005 factor
= num_stripes
/ 2;
3006 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID5
) {
3007 factor
= num_stripes
- 1;
3008 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID6
) {
3009 factor
= num_stripes
- 2;
3011 factor
= num_stripes
;
3014 for (i
= 0; i
< num_stripes
; i
++) {
3015 stripe
= btrfs_stripe_nr(chunk
, i
);
3016 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
3019 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
3020 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
3021 do_div(stripe_length
, factor
);
3023 if (stripe_offset
< bargs
->pend
&&
3024 stripe_offset
+ stripe_length
> bargs
->pstart
)
3031 /* [vstart, vend) */
3032 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
3033 struct btrfs_chunk
*chunk
,
3035 struct btrfs_balance_args
*bargs
)
3037 if (chunk_offset
< bargs
->vend
&&
3038 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
3039 /* at least part of the chunk is inside this vrange */
3045 static int chunk_soft_convert_filter(u64 chunk_type
,
3046 struct btrfs_balance_args
*bargs
)
3048 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
3051 chunk_type
= chunk_to_extended(chunk_type
) &
3052 BTRFS_EXTENDED_PROFILE_MASK
;
3054 if (bargs
->target
== chunk_type
)
3060 static int should_balance_chunk(struct btrfs_root
*root
,
3061 struct extent_buffer
*leaf
,
3062 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
3064 struct btrfs_balance_control
*bctl
= root
->fs_info
->balance_ctl
;
3065 struct btrfs_balance_args
*bargs
= NULL
;
3066 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3069 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
3070 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
3074 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3075 bargs
= &bctl
->data
;
3076 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3078 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3079 bargs
= &bctl
->meta
;
3081 /* profiles filter */
3082 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
3083 chunk_profiles_filter(chunk_type
, bargs
)) {
3088 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3089 chunk_usage_filter(bctl
->fs_info
, chunk_offset
, bargs
)) {
3094 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
3095 chunk_devid_filter(leaf
, chunk
, bargs
)) {
3099 /* drange filter, makes sense only with devid filter */
3100 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
3101 chunk_drange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3106 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
3107 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3111 /* soft profile changing mode */
3112 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
3113 chunk_soft_convert_filter(chunk_type
, bargs
)) {
3118 * limited by count, must be the last filter
3120 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT
)) {
3121 if (bargs
->limit
== 0)
3130 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
3132 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3133 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
3134 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
3135 struct list_head
*devices
;
3136 struct btrfs_device
*device
;
3139 struct btrfs_chunk
*chunk
;
3140 struct btrfs_path
*path
;
3141 struct btrfs_key key
;
3142 struct btrfs_key found_key
;
3143 struct btrfs_trans_handle
*trans
;
3144 struct extent_buffer
*leaf
;
3147 int enospc_errors
= 0;
3148 bool counting
= true;
3149 u64 limit_data
= bctl
->data
.limit
;
3150 u64 limit_meta
= bctl
->meta
.limit
;
3151 u64 limit_sys
= bctl
->sys
.limit
;
3153 /* step one make some room on all the devices */
3154 devices
= &fs_info
->fs_devices
->devices
;
3155 list_for_each_entry(device
, devices
, dev_list
) {
3156 old_size
= btrfs_device_get_total_bytes(device
);
3157 size_to_free
= div_factor(old_size
, 1);
3158 size_to_free
= min(size_to_free
, (u64
)1 * 1024 * 1024);
3159 if (!device
->writeable
||
3160 btrfs_device_get_total_bytes(device
) -
3161 btrfs_device_get_bytes_used(device
) > size_to_free
||
3162 device
->is_tgtdev_for_dev_replace
)
3165 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
3170 trans
= btrfs_start_transaction(dev_root
, 0);
3171 BUG_ON(IS_ERR(trans
));
3173 ret
= btrfs_grow_device(trans
, device
, old_size
);
3176 btrfs_end_transaction(trans
, dev_root
);
3179 /* step two, relocate all the chunks */
3180 path
= btrfs_alloc_path();
3186 /* zero out stat counters */
3187 spin_lock(&fs_info
->balance_lock
);
3188 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
3189 spin_unlock(&fs_info
->balance_lock
);
3192 bctl
->data
.limit
= limit_data
;
3193 bctl
->meta
.limit
= limit_meta
;
3194 bctl
->sys
.limit
= limit_sys
;
3196 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3197 key
.offset
= (u64
)-1;
3198 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3201 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
3202 atomic_read(&fs_info
->balance_cancel_req
)) {
3207 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
3212 * this shouldn't happen, it means the last relocate
3216 BUG(); /* FIXME break ? */
3218 ret
= btrfs_previous_item(chunk_root
, path
, 0,
3219 BTRFS_CHUNK_ITEM_KEY
);
3225 leaf
= path
->nodes
[0];
3226 slot
= path
->slots
[0];
3227 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3229 if (found_key
.objectid
!= key
.objectid
)
3232 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
3235 spin_lock(&fs_info
->balance_lock
);
3236 bctl
->stat
.considered
++;
3237 spin_unlock(&fs_info
->balance_lock
);
3240 ret
= should_balance_chunk(chunk_root
, leaf
, chunk
,
3242 btrfs_release_path(path
);
3247 spin_lock(&fs_info
->balance_lock
);
3248 bctl
->stat
.expected
++;
3249 spin_unlock(&fs_info
->balance_lock
);
3253 ret
= btrfs_relocate_chunk(chunk_root
,
3256 if (ret
&& ret
!= -ENOSPC
)
3258 if (ret
== -ENOSPC
) {
3261 spin_lock(&fs_info
->balance_lock
);
3262 bctl
->stat
.completed
++;
3263 spin_unlock(&fs_info
->balance_lock
);
3266 if (found_key
.offset
== 0)
3268 key
.offset
= found_key
.offset
- 1;
3272 btrfs_release_path(path
);
3277 btrfs_free_path(path
);
3278 if (enospc_errors
) {
3279 btrfs_info(fs_info
, "%d enospc errors during balance",
3289 * alloc_profile_is_valid - see if a given profile is valid and reduced
3290 * @flags: profile to validate
3291 * @extended: if true @flags is treated as an extended profile
3293 static int alloc_profile_is_valid(u64 flags
, int extended
)
3295 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
3296 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
3298 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
3300 /* 1) check that all other bits are zeroed */
3304 /* 2) see if profile is reduced */
3306 return !extended
; /* "0" is valid for usual profiles */
3308 /* true if exactly one bit set */
3309 return (flags
& (flags
- 1)) == 0;
3312 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
3314 /* cancel requested || normal exit path */
3315 return atomic_read(&fs_info
->balance_cancel_req
) ||
3316 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
3317 atomic_read(&fs_info
->balance_cancel_req
) == 0);
3320 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
3324 unset_balance_control(fs_info
);
3325 ret
= del_balance_item(fs_info
->tree_root
);
3327 btrfs_std_error(fs_info
, ret
);
3329 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3333 * Should be called with both balance and volume mutexes held
3335 int btrfs_balance(struct btrfs_balance_control
*bctl
,
3336 struct btrfs_ioctl_balance_args
*bargs
)
3338 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3345 if (btrfs_fs_closing(fs_info
) ||
3346 atomic_read(&fs_info
->balance_pause_req
) ||
3347 atomic_read(&fs_info
->balance_cancel_req
)) {
3352 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
3353 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
3357 * In case of mixed groups both data and meta should be picked,
3358 * and identical options should be given for both of them.
3360 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
3361 if (mixed
&& (bctl
->flags
& allowed
)) {
3362 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
3363 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
3364 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
3365 btrfs_err(fs_info
, "with mixed groups data and "
3366 "metadata balance options must be the same");
3372 num_devices
= fs_info
->fs_devices
->num_devices
;
3373 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
3374 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
3375 BUG_ON(num_devices
< 1);
3378 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
3379 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
3380 if (num_devices
== 1)
3381 allowed
|= BTRFS_BLOCK_GROUP_DUP
;
3382 else if (num_devices
> 1)
3383 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
3384 if (num_devices
> 2)
3385 allowed
|= BTRFS_BLOCK_GROUP_RAID5
;
3386 if (num_devices
> 3)
3387 allowed
|= (BTRFS_BLOCK_GROUP_RAID10
|
3388 BTRFS_BLOCK_GROUP_RAID6
);
3389 if ((bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3390 (!alloc_profile_is_valid(bctl
->data
.target
, 1) ||
3391 (bctl
->data
.target
& ~allowed
))) {
3392 btrfs_err(fs_info
, "unable to start balance with target "
3393 "data profile %llu",
3398 if ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3399 (!alloc_profile_is_valid(bctl
->meta
.target
, 1) ||
3400 (bctl
->meta
.target
& ~allowed
))) {
3402 "unable to start balance with target metadata profile %llu",
3407 if ((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3408 (!alloc_profile_is_valid(bctl
->sys
.target
, 1) ||
3409 (bctl
->sys
.target
& ~allowed
))) {
3411 "unable to start balance with target system profile %llu",
3417 /* allow dup'ed data chunks only in mixed mode */
3418 if (!mixed
&& (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3419 (bctl
->data
.target
& BTRFS_BLOCK_GROUP_DUP
)) {
3420 btrfs_err(fs_info
, "dup for data is not allowed");
3425 /* allow to reduce meta or sys integrity only if force set */
3426 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3427 BTRFS_BLOCK_GROUP_RAID10
|
3428 BTRFS_BLOCK_GROUP_RAID5
|
3429 BTRFS_BLOCK_GROUP_RAID6
;
3431 seq
= read_seqbegin(&fs_info
->profiles_lock
);
3433 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3434 (fs_info
->avail_system_alloc_bits
& allowed
) &&
3435 !(bctl
->sys
.target
& allowed
)) ||
3436 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3437 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
3438 !(bctl
->meta
.target
& allowed
))) {
3439 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
3440 btrfs_info(fs_info
, "force reducing metadata integrity");
3442 btrfs_err(fs_info
, "balance will reduce metadata "
3443 "integrity, use force if you want this");
3448 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
3450 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3451 int num_tolerated_disk_barrier_failures
;
3452 u64 target
= bctl
->sys
.target
;
3454 num_tolerated_disk_barrier_failures
=
3455 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3456 if (num_tolerated_disk_barrier_failures
> 0 &&
3458 (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID0
|
3459 BTRFS_AVAIL_ALLOC_BIT_SINGLE
)))
3460 num_tolerated_disk_barrier_failures
= 0;
3461 else if (num_tolerated_disk_barrier_failures
> 1 &&
3463 (BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)))
3464 num_tolerated_disk_barrier_failures
= 1;
3466 fs_info
->num_tolerated_disk_barrier_failures
=
3467 num_tolerated_disk_barrier_failures
;
3470 ret
= insert_balance_item(fs_info
->tree_root
, bctl
);
3471 if (ret
&& ret
!= -EEXIST
)
3474 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
3475 BUG_ON(ret
== -EEXIST
);
3476 set_balance_control(bctl
);
3478 BUG_ON(ret
!= -EEXIST
);
3479 spin_lock(&fs_info
->balance_lock
);
3480 update_balance_args(bctl
);
3481 spin_unlock(&fs_info
->balance_lock
);
3484 atomic_inc(&fs_info
->balance_running
);
3485 mutex_unlock(&fs_info
->balance_mutex
);
3487 ret
= __btrfs_balance(fs_info
);
3489 mutex_lock(&fs_info
->balance_mutex
);
3490 atomic_dec(&fs_info
->balance_running
);
3492 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3493 fs_info
->num_tolerated_disk_barrier_failures
=
3494 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3498 memset(bargs
, 0, sizeof(*bargs
));
3499 update_ioctl_balance_args(fs_info
, 0, bargs
);
3502 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
3503 balance_need_close(fs_info
)) {
3504 __cancel_balance(fs_info
);
3507 wake_up(&fs_info
->balance_wait_q
);
3511 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
3512 __cancel_balance(fs_info
);
3515 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3520 static int balance_kthread(void *data
)
3522 struct btrfs_fs_info
*fs_info
= data
;
3525 mutex_lock(&fs_info
->volume_mutex
);
3526 mutex_lock(&fs_info
->balance_mutex
);
3528 if (fs_info
->balance_ctl
) {
3529 btrfs_info(fs_info
, "continuing balance");
3530 ret
= btrfs_balance(fs_info
->balance_ctl
, NULL
);
3533 mutex_unlock(&fs_info
->balance_mutex
);
3534 mutex_unlock(&fs_info
->volume_mutex
);
3539 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
3541 struct task_struct
*tsk
;
3543 spin_lock(&fs_info
->balance_lock
);
3544 if (!fs_info
->balance_ctl
) {
3545 spin_unlock(&fs_info
->balance_lock
);
3548 spin_unlock(&fs_info
->balance_lock
);
3550 if (btrfs_test_opt(fs_info
->tree_root
, SKIP_BALANCE
)) {
3551 btrfs_info(fs_info
, "force skipping balance");
3555 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
3556 return PTR_ERR_OR_ZERO(tsk
);
3559 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
3561 struct btrfs_balance_control
*bctl
;
3562 struct btrfs_balance_item
*item
;
3563 struct btrfs_disk_balance_args disk_bargs
;
3564 struct btrfs_path
*path
;
3565 struct extent_buffer
*leaf
;
3566 struct btrfs_key key
;
3569 path
= btrfs_alloc_path();
3573 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3574 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
3577 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
3580 if (ret
> 0) { /* ret = -ENOENT; */
3585 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
3591 leaf
= path
->nodes
[0];
3592 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3594 bctl
->fs_info
= fs_info
;
3595 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
3596 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
3598 btrfs_balance_data(leaf
, item
, &disk_bargs
);
3599 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
3600 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
3601 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
3602 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
3603 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
3605 WARN_ON(atomic_xchg(&fs_info
->mutually_exclusive_operation_running
, 1));
3607 mutex_lock(&fs_info
->volume_mutex
);
3608 mutex_lock(&fs_info
->balance_mutex
);
3610 set_balance_control(bctl
);
3612 mutex_unlock(&fs_info
->balance_mutex
);
3613 mutex_unlock(&fs_info
->volume_mutex
);
3615 btrfs_free_path(path
);
3619 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
3623 mutex_lock(&fs_info
->balance_mutex
);
3624 if (!fs_info
->balance_ctl
) {
3625 mutex_unlock(&fs_info
->balance_mutex
);
3629 if (atomic_read(&fs_info
->balance_running
)) {
3630 atomic_inc(&fs_info
->balance_pause_req
);
3631 mutex_unlock(&fs_info
->balance_mutex
);
3633 wait_event(fs_info
->balance_wait_q
,
3634 atomic_read(&fs_info
->balance_running
) == 0);
3636 mutex_lock(&fs_info
->balance_mutex
);
3637 /* we are good with balance_ctl ripped off from under us */
3638 BUG_ON(atomic_read(&fs_info
->balance_running
));
3639 atomic_dec(&fs_info
->balance_pause_req
);
3644 mutex_unlock(&fs_info
->balance_mutex
);
3648 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
3650 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
3653 mutex_lock(&fs_info
->balance_mutex
);
3654 if (!fs_info
->balance_ctl
) {
3655 mutex_unlock(&fs_info
->balance_mutex
);
3659 atomic_inc(&fs_info
->balance_cancel_req
);
3661 * if we are running just wait and return, balance item is
3662 * deleted in btrfs_balance in this case
3664 if (atomic_read(&fs_info
->balance_running
)) {
3665 mutex_unlock(&fs_info
->balance_mutex
);
3666 wait_event(fs_info
->balance_wait_q
,
3667 atomic_read(&fs_info
->balance_running
) == 0);
3668 mutex_lock(&fs_info
->balance_mutex
);
3670 /* __cancel_balance needs volume_mutex */
3671 mutex_unlock(&fs_info
->balance_mutex
);
3672 mutex_lock(&fs_info
->volume_mutex
);
3673 mutex_lock(&fs_info
->balance_mutex
);
3675 if (fs_info
->balance_ctl
)
3676 __cancel_balance(fs_info
);
3678 mutex_unlock(&fs_info
->volume_mutex
);
3681 BUG_ON(fs_info
->balance_ctl
|| atomic_read(&fs_info
->balance_running
));
3682 atomic_dec(&fs_info
->balance_cancel_req
);
3683 mutex_unlock(&fs_info
->balance_mutex
);
3687 static int btrfs_uuid_scan_kthread(void *data
)
3689 struct btrfs_fs_info
*fs_info
= data
;
3690 struct btrfs_root
*root
= fs_info
->tree_root
;
3691 struct btrfs_key key
;
3692 struct btrfs_key max_key
;
3693 struct btrfs_path
*path
= NULL
;
3695 struct extent_buffer
*eb
;
3697 struct btrfs_root_item root_item
;
3699 struct btrfs_trans_handle
*trans
= NULL
;
3701 path
= btrfs_alloc_path();
3708 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3711 max_key
.objectid
= (u64
)-1;
3712 max_key
.type
= BTRFS_ROOT_ITEM_KEY
;
3713 max_key
.offset
= (u64
)-1;
3716 ret
= btrfs_search_forward(root
, &key
, path
, 0);
3723 if (key
.type
!= BTRFS_ROOT_ITEM_KEY
||
3724 (key
.objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
3725 key
.objectid
!= BTRFS_FS_TREE_OBJECTID
) ||
3726 key
.objectid
> BTRFS_LAST_FREE_OBJECTID
)
3729 eb
= path
->nodes
[0];
3730 slot
= path
->slots
[0];
3731 item_size
= btrfs_item_size_nr(eb
, slot
);
3732 if (item_size
< sizeof(root_item
))
3735 read_extent_buffer(eb
, &root_item
,
3736 btrfs_item_ptr_offset(eb
, slot
),
3737 (int)sizeof(root_item
));
3738 if (btrfs_root_refs(&root_item
) == 0)
3741 if (!btrfs_is_empty_uuid(root_item
.uuid
) ||
3742 !btrfs_is_empty_uuid(root_item
.received_uuid
)) {
3746 btrfs_release_path(path
);
3748 * 1 - subvol uuid item
3749 * 1 - received_subvol uuid item
3751 trans
= btrfs_start_transaction(fs_info
->uuid_root
, 2);
3752 if (IS_ERR(trans
)) {
3753 ret
= PTR_ERR(trans
);
3761 if (!btrfs_is_empty_uuid(root_item
.uuid
)) {
3762 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
3764 BTRFS_UUID_KEY_SUBVOL
,
3767 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
3773 if (!btrfs_is_empty_uuid(root_item
.received_uuid
)) {
3774 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
3775 root_item
.received_uuid
,
3776 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
3779 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
3787 ret
= btrfs_end_transaction(trans
, fs_info
->uuid_root
);
3793 btrfs_release_path(path
);
3794 if (key
.offset
< (u64
)-1) {
3796 } else if (key
.type
< BTRFS_ROOT_ITEM_KEY
) {
3798 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3799 } else if (key
.objectid
< (u64
)-1) {
3801 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3810 btrfs_free_path(path
);
3811 if (trans
&& !IS_ERR(trans
))
3812 btrfs_end_transaction(trans
, fs_info
->uuid_root
);
3814 btrfs_warn(fs_info
, "btrfs_uuid_scan_kthread failed %d", ret
);
3816 fs_info
->update_uuid_tree_gen
= 1;
3817 up(&fs_info
->uuid_tree_rescan_sem
);
3822 * Callback for btrfs_uuid_tree_iterate().
3824 * 0 check succeeded, the entry is not outdated.
3825 * < 0 if an error occured.
3826 * > 0 if the check failed, which means the caller shall remove the entry.
3828 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info
*fs_info
,
3829 u8
*uuid
, u8 type
, u64 subid
)
3831 struct btrfs_key key
;
3833 struct btrfs_root
*subvol_root
;
3835 if (type
!= BTRFS_UUID_KEY_SUBVOL
&&
3836 type
!= BTRFS_UUID_KEY_RECEIVED_SUBVOL
)
3839 key
.objectid
= subid
;
3840 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3841 key
.offset
= (u64
)-1;
3842 subvol_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
3843 if (IS_ERR(subvol_root
)) {
3844 ret
= PTR_ERR(subvol_root
);
3851 case BTRFS_UUID_KEY_SUBVOL
:
3852 if (memcmp(uuid
, subvol_root
->root_item
.uuid
, BTRFS_UUID_SIZE
))
3855 case BTRFS_UUID_KEY_RECEIVED_SUBVOL
:
3856 if (memcmp(uuid
, subvol_root
->root_item
.received_uuid
,
3866 static int btrfs_uuid_rescan_kthread(void *data
)
3868 struct btrfs_fs_info
*fs_info
= (struct btrfs_fs_info
*)data
;
3872 * 1st step is to iterate through the existing UUID tree and
3873 * to delete all entries that contain outdated data.
3874 * 2nd step is to add all missing entries to the UUID tree.
3876 ret
= btrfs_uuid_tree_iterate(fs_info
, btrfs_check_uuid_tree_entry
);
3878 btrfs_warn(fs_info
, "iterating uuid_tree failed %d", ret
);
3879 up(&fs_info
->uuid_tree_rescan_sem
);
3882 return btrfs_uuid_scan_kthread(data
);
3885 int btrfs_create_uuid_tree(struct btrfs_fs_info
*fs_info
)
3887 struct btrfs_trans_handle
*trans
;
3888 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
3889 struct btrfs_root
*uuid_root
;
3890 struct task_struct
*task
;
3897 trans
= btrfs_start_transaction(tree_root
, 2);
3899 return PTR_ERR(trans
);
3901 uuid_root
= btrfs_create_tree(trans
, fs_info
,
3902 BTRFS_UUID_TREE_OBJECTID
);
3903 if (IS_ERR(uuid_root
)) {
3904 btrfs_abort_transaction(trans
, tree_root
,
3905 PTR_ERR(uuid_root
));
3906 return PTR_ERR(uuid_root
);
3909 fs_info
->uuid_root
= uuid_root
;
3911 ret
= btrfs_commit_transaction(trans
, tree_root
);
3915 down(&fs_info
->uuid_tree_rescan_sem
);
3916 task
= kthread_run(btrfs_uuid_scan_kthread
, fs_info
, "btrfs-uuid");
3918 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3919 btrfs_warn(fs_info
, "failed to start uuid_scan task");
3920 up(&fs_info
->uuid_tree_rescan_sem
);
3921 return PTR_ERR(task
);
3927 int btrfs_check_uuid_tree(struct btrfs_fs_info
*fs_info
)
3929 struct task_struct
*task
;
3931 down(&fs_info
->uuid_tree_rescan_sem
);
3932 task
= kthread_run(btrfs_uuid_rescan_kthread
, fs_info
, "btrfs-uuid");
3934 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3935 btrfs_warn(fs_info
, "failed to start uuid_rescan task");
3936 up(&fs_info
->uuid_tree_rescan_sem
);
3937 return PTR_ERR(task
);
3944 * shrinking a device means finding all of the device extents past
3945 * the new size, and then following the back refs to the chunks.
3946 * The chunk relocation code actually frees the device extent
3948 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
3950 struct btrfs_trans_handle
*trans
;
3951 struct btrfs_root
*root
= device
->dev_root
;
3952 struct btrfs_dev_extent
*dev_extent
= NULL
;
3953 struct btrfs_path
*path
;
3960 bool retried
= false;
3961 struct extent_buffer
*l
;
3962 struct btrfs_key key
;
3963 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
3964 u64 old_total
= btrfs_super_total_bytes(super_copy
);
3965 u64 old_size
= btrfs_device_get_total_bytes(device
);
3966 u64 diff
= old_size
- new_size
;
3968 if (device
->is_tgtdev_for_dev_replace
)
3971 path
= btrfs_alloc_path();
3979 btrfs_device_set_total_bytes(device
, new_size
);
3980 if (device
->writeable
) {
3981 device
->fs_devices
->total_rw_bytes
-= diff
;
3982 spin_lock(&root
->fs_info
->free_chunk_lock
);
3983 root
->fs_info
->free_chunk_space
-= diff
;
3984 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3986 unlock_chunks(root
);
3989 key
.objectid
= device
->devid
;
3990 key
.offset
= (u64
)-1;
3991 key
.type
= BTRFS_DEV_EXTENT_KEY
;
3994 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3998 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
4003 btrfs_release_path(path
);
4008 slot
= path
->slots
[0];
4009 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
4011 if (key
.objectid
!= device
->devid
) {
4012 btrfs_release_path(path
);
4016 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
4017 length
= btrfs_dev_extent_length(l
, dev_extent
);
4019 if (key
.offset
+ length
<= new_size
) {
4020 btrfs_release_path(path
);
4024 chunk_objectid
= btrfs_dev_extent_chunk_objectid(l
, dev_extent
);
4025 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
4026 btrfs_release_path(path
);
4028 ret
= btrfs_relocate_chunk(root
, chunk_objectid
, chunk_offset
);
4029 if (ret
&& ret
!= -ENOSPC
)
4033 } while (key
.offset
-- > 0);
4035 if (failed
&& !retried
) {
4039 } else if (failed
&& retried
) {
4043 btrfs_device_set_total_bytes(device
, old_size
);
4044 if (device
->writeable
)
4045 device
->fs_devices
->total_rw_bytes
+= diff
;
4046 spin_lock(&root
->fs_info
->free_chunk_lock
);
4047 root
->fs_info
->free_chunk_space
+= diff
;
4048 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4049 unlock_chunks(root
);
4053 /* Shrinking succeeded, else we would be at "done". */
4054 trans
= btrfs_start_transaction(root
, 0);
4055 if (IS_ERR(trans
)) {
4056 ret
= PTR_ERR(trans
);
4061 btrfs_device_set_disk_total_bytes(device
, new_size
);
4062 if (list_empty(&device
->resized_list
))
4063 list_add_tail(&device
->resized_list
,
4064 &root
->fs_info
->fs_devices
->resized_devices
);
4066 WARN_ON(diff
> old_total
);
4067 btrfs_set_super_total_bytes(super_copy
, old_total
- diff
);
4068 unlock_chunks(root
);
4070 /* Now btrfs_update_device() will change the on-disk size. */
4071 ret
= btrfs_update_device(trans
, device
);
4072 btrfs_end_transaction(trans
, root
);
4074 btrfs_free_path(path
);
4078 static int btrfs_add_system_chunk(struct btrfs_root
*root
,
4079 struct btrfs_key
*key
,
4080 struct btrfs_chunk
*chunk
, int item_size
)
4082 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
4083 struct btrfs_disk_key disk_key
;
4088 array_size
= btrfs_super_sys_array_size(super_copy
);
4089 if (array_size
+ item_size
+ sizeof(disk_key
)
4090 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4091 unlock_chunks(root
);
4095 ptr
= super_copy
->sys_chunk_array
+ array_size
;
4096 btrfs_cpu_key_to_disk(&disk_key
, key
);
4097 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
4098 ptr
+= sizeof(disk_key
);
4099 memcpy(ptr
, chunk
, item_size
);
4100 item_size
+= sizeof(disk_key
);
4101 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
4102 unlock_chunks(root
);
4108 * sort the devices in descending order by max_avail, total_avail
4110 static int btrfs_cmp_device_info(const void *a
, const void *b
)
4112 const struct btrfs_device_info
*di_a
= a
;
4113 const struct btrfs_device_info
*di_b
= b
;
4115 if (di_a
->max_avail
> di_b
->max_avail
)
4117 if (di_a
->max_avail
< di_b
->max_avail
)
4119 if (di_a
->total_avail
> di_b
->total_avail
)
4121 if (di_a
->total_avail
< di_b
->total_avail
)
4126 static const struct btrfs_raid_attr btrfs_raid_array
[BTRFS_NR_RAID_TYPES
] = {
4127 [BTRFS_RAID_RAID10
] = {
4130 .devs_max
= 0, /* 0 == as many as possible */
4132 .devs_increment
= 2,
4135 [BTRFS_RAID_RAID1
] = {
4140 .devs_increment
= 2,
4143 [BTRFS_RAID_DUP
] = {
4148 .devs_increment
= 1,
4151 [BTRFS_RAID_RAID0
] = {
4156 .devs_increment
= 1,
4159 [BTRFS_RAID_SINGLE
] = {
4164 .devs_increment
= 1,
4167 [BTRFS_RAID_RAID5
] = {
4172 .devs_increment
= 1,
4175 [BTRFS_RAID_RAID6
] = {
4180 .devs_increment
= 1,
4185 static u32
find_raid56_stripe_len(u32 data_devices
, u32 dev_stripe_target
)
4187 /* TODO allow them to set a preferred stripe size */
4191 static void check_raid56_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
4193 if (!(type
& BTRFS_BLOCK_GROUP_RAID56_MASK
))
4196 btrfs_set_fs_incompat(info
, RAID56
);
4199 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4200 - sizeof(struct btrfs_item) \
4201 - sizeof(struct btrfs_chunk)) \
4202 / sizeof(struct btrfs_stripe) + 1)
4204 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4205 - 2 * sizeof(struct btrfs_disk_key) \
4206 - 2 * sizeof(struct btrfs_chunk)) \
4207 / sizeof(struct btrfs_stripe) + 1)
4209 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4210 struct btrfs_root
*extent_root
, u64 start
,
4213 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
4214 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
4215 struct list_head
*cur
;
4216 struct map_lookup
*map
= NULL
;
4217 struct extent_map_tree
*em_tree
;
4218 struct extent_map
*em
;
4219 struct btrfs_device_info
*devices_info
= NULL
;
4221 int num_stripes
; /* total number of stripes to allocate */
4222 int data_stripes
; /* number of stripes that count for
4224 int sub_stripes
; /* sub_stripes info for map */
4225 int dev_stripes
; /* stripes per dev */
4226 int devs_max
; /* max devs to use */
4227 int devs_min
; /* min devs needed */
4228 int devs_increment
; /* ndevs has to be a multiple of this */
4229 int ncopies
; /* how many copies to data has */
4231 u64 max_stripe_size
;
4235 u64 raid_stripe_len
= BTRFS_STRIPE_LEN
;
4241 BUG_ON(!alloc_profile_is_valid(type
, 0));
4243 if (list_empty(&fs_devices
->alloc_list
))
4246 index
= __get_raid_index(type
);
4248 sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
4249 dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
4250 devs_max
= btrfs_raid_array
[index
].devs_max
;
4251 devs_min
= btrfs_raid_array
[index
].devs_min
;
4252 devs_increment
= btrfs_raid_array
[index
].devs_increment
;
4253 ncopies
= btrfs_raid_array
[index
].ncopies
;
4255 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
4256 max_stripe_size
= 1024 * 1024 * 1024;
4257 max_chunk_size
= 10 * max_stripe_size
;
4259 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4260 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
4261 /* for larger filesystems, use larger metadata chunks */
4262 if (fs_devices
->total_rw_bytes
> 50ULL * 1024 * 1024 * 1024)
4263 max_stripe_size
= 1024 * 1024 * 1024;
4265 max_stripe_size
= 256 * 1024 * 1024;
4266 max_chunk_size
= max_stripe_size
;
4268 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4269 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4270 max_stripe_size
= 32 * 1024 * 1024;
4271 max_chunk_size
= 2 * max_stripe_size
;
4273 devs_max
= BTRFS_MAX_DEVS_SYS_CHUNK
;
4275 btrfs_err(info
, "invalid chunk type 0x%llx requested",
4280 /* we don't want a chunk larger than 10% of writeable space */
4281 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
4284 devices_info
= kzalloc(sizeof(*devices_info
) * fs_devices
->rw_devices
,
4289 cur
= fs_devices
->alloc_list
.next
;
4292 * in the first pass through the devices list, we gather information
4293 * about the available holes on each device.
4296 while (cur
!= &fs_devices
->alloc_list
) {
4297 struct btrfs_device
*device
;
4301 device
= list_entry(cur
, struct btrfs_device
, dev_alloc_list
);
4305 if (!device
->writeable
) {
4307 "BTRFS: read-only device in alloc_list\n");
4311 if (!device
->in_fs_metadata
||
4312 device
->is_tgtdev_for_dev_replace
)
4315 if (device
->total_bytes
> device
->bytes_used
)
4316 total_avail
= device
->total_bytes
- device
->bytes_used
;
4320 /* If there is no space on this device, skip it. */
4321 if (total_avail
== 0)
4324 ret
= find_free_dev_extent(trans
, device
,
4325 max_stripe_size
* dev_stripes
,
4326 &dev_offset
, &max_avail
);
4327 if (ret
&& ret
!= -ENOSPC
)
4331 max_avail
= max_stripe_size
* dev_stripes
;
4333 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
4336 if (ndevs
== fs_devices
->rw_devices
) {
4337 WARN(1, "%s: found more than %llu devices\n",
4338 __func__
, fs_devices
->rw_devices
);
4341 devices_info
[ndevs
].dev_offset
= dev_offset
;
4342 devices_info
[ndevs
].max_avail
= max_avail
;
4343 devices_info
[ndevs
].total_avail
= total_avail
;
4344 devices_info
[ndevs
].dev
= device
;
4349 * now sort the devices by hole size / available space
4351 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
4352 btrfs_cmp_device_info
, NULL
);
4354 /* round down to number of usable stripes */
4355 ndevs
-= ndevs
% devs_increment
;
4357 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
4362 if (devs_max
&& ndevs
> devs_max
)
4365 * the primary goal is to maximize the number of stripes, so use as many
4366 * devices as possible, even if the stripes are not maximum sized.
4368 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4369 num_stripes
= ndevs
* dev_stripes
;
4372 * this will have to be fixed for RAID1 and RAID10 over
4375 data_stripes
= num_stripes
/ ncopies
;
4377 if (type
& BTRFS_BLOCK_GROUP_RAID5
) {
4378 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 1,
4379 btrfs_super_stripesize(info
->super_copy
));
4380 data_stripes
= num_stripes
- 1;
4382 if (type
& BTRFS_BLOCK_GROUP_RAID6
) {
4383 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 2,
4384 btrfs_super_stripesize(info
->super_copy
));
4385 data_stripes
= num_stripes
- 2;
4389 * Use the number of data stripes to figure out how big this chunk
4390 * is really going to be in terms of logical address space,
4391 * and compare that answer with the max chunk size
4393 if (stripe_size
* data_stripes
> max_chunk_size
) {
4394 u64 mask
= (1ULL << 24) - 1;
4395 stripe_size
= max_chunk_size
;
4396 do_div(stripe_size
, data_stripes
);
4398 /* bump the answer up to a 16MB boundary */
4399 stripe_size
= (stripe_size
+ mask
) & ~mask
;
4401 /* but don't go higher than the limits we found
4402 * while searching for free extents
4404 if (stripe_size
> devices_info
[ndevs
-1].max_avail
)
4405 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4408 do_div(stripe_size
, dev_stripes
);
4410 /* align to BTRFS_STRIPE_LEN */
4411 do_div(stripe_size
, raid_stripe_len
);
4412 stripe_size
*= raid_stripe_len
;
4414 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
4419 map
->num_stripes
= num_stripes
;
4421 for (i
= 0; i
< ndevs
; ++i
) {
4422 for (j
= 0; j
< dev_stripes
; ++j
) {
4423 int s
= i
* dev_stripes
+ j
;
4424 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
4425 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
4429 map
->sector_size
= extent_root
->sectorsize
;
4430 map
->stripe_len
= raid_stripe_len
;
4431 map
->io_align
= raid_stripe_len
;
4432 map
->io_width
= raid_stripe_len
;
4434 map
->sub_stripes
= sub_stripes
;
4436 num_bytes
= stripe_size
* data_stripes
;
4438 trace_btrfs_chunk_alloc(info
->chunk_root
, map
, start
, num_bytes
);
4440 em
= alloc_extent_map();
4446 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
4447 em
->bdev
= (struct block_device
*)map
;
4449 em
->len
= num_bytes
;
4450 em
->block_start
= 0;
4451 em
->block_len
= em
->len
;
4452 em
->orig_block_len
= stripe_size
;
4454 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
4455 write_lock(&em_tree
->lock
);
4456 ret
= add_extent_mapping(em_tree
, em
, 0);
4458 list_add_tail(&em
->list
, &trans
->transaction
->pending_chunks
);
4459 atomic_inc(&em
->refs
);
4461 write_unlock(&em_tree
->lock
);
4463 free_extent_map(em
);
4467 ret
= btrfs_make_block_group(trans
, extent_root
, 0, type
,
4468 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4471 goto error_del_extent
;
4473 for (i
= 0; i
< map
->num_stripes
; i
++) {
4474 num_bytes
= map
->stripes
[i
].dev
->bytes_used
+ stripe_size
;
4475 btrfs_device_set_bytes_used(map
->stripes
[i
].dev
, num_bytes
);
4478 spin_lock(&extent_root
->fs_info
->free_chunk_lock
);
4479 extent_root
->fs_info
->free_chunk_space
-= (stripe_size
*
4481 spin_unlock(&extent_root
->fs_info
->free_chunk_lock
);
4483 free_extent_map(em
);
4484 check_raid56_incompat_flag(extent_root
->fs_info
, type
);
4486 kfree(devices_info
);
4490 write_lock(&em_tree
->lock
);
4491 remove_extent_mapping(em_tree
, em
);
4492 write_unlock(&em_tree
->lock
);
4494 /* One for our allocation */
4495 free_extent_map(em
);
4496 /* One for the tree reference */
4497 free_extent_map(em
);
4498 /* One for the pending_chunks list reference */
4499 free_extent_map(em
);
4501 kfree(devices_info
);
4505 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
4506 struct btrfs_root
*extent_root
,
4507 u64 chunk_offset
, u64 chunk_size
)
4509 struct btrfs_key key
;
4510 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
4511 struct btrfs_device
*device
;
4512 struct btrfs_chunk
*chunk
;
4513 struct btrfs_stripe
*stripe
;
4514 struct extent_map_tree
*em_tree
;
4515 struct extent_map
*em
;
4516 struct map_lookup
*map
;
4523 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
4524 read_lock(&em_tree
->lock
);
4525 em
= lookup_extent_mapping(em_tree
, chunk_offset
, chunk_size
);
4526 read_unlock(&em_tree
->lock
);
4529 btrfs_crit(extent_root
->fs_info
, "unable to find logical "
4530 "%Lu len %Lu", chunk_offset
, chunk_size
);
4534 if (em
->start
!= chunk_offset
|| em
->len
!= chunk_size
) {
4535 btrfs_crit(extent_root
->fs_info
, "found a bad mapping, wanted"
4536 " %Lu-%Lu, found %Lu-%Lu", chunk_offset
,
4537 chunk_size
, em
->start
, em
->len
);
4538 free_extent_map(em
);
4542 map
= (struct map_lookup
*)em
->bdev
;
4543 item_size
= btrfs_chunk_item_size(map
->num_stripes
);
4544 stripe_size
= em
->orig_block_len
;
4546 chunk
= kzalloc(item_size
, GFP_NOFS
);
4552 for (i
= 0; i
< map
->num_stripes
; i
++) {
4553 device
= map
->stripes
[i
].dev
;
4554 dev_offset
= map
->stripes
[i
].physical
;
4556 ret
= btrfs_update_device(trans
, device
);
4559 ret
= btrfs_alloc_dev_extent(trans
, device
,
4560 chunk_root
->root_key
.objectid
,
4561 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4562 chunk_offset
, dev_offset
,
4568 stripe
= &chunk
->stripe
;
4569 for (i
= 0; i
< map
->num_stripes
; i
++) {
4570 device
= map
->stripes
[i
].dev
;
4571 dev_offset
= map
->stripes
[i
].physical
;
4573 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
4574 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
4575 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
4579 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
4580 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
4581 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
4582 btrfs_set_stack_chunk_type(chunk
, map
->type
);
4583 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
4584 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
4585 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
4586 btrfs_set_stack_chunk_sector_size(chunk
, extent_root
->sectorsize
);
4587 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
4589 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
4590 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
4591 key
.offset
= chunk_offset
;
4593 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
4594 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4596 * TODO: Cleanup of inserted chunk root in case of
4599 ret
= btrfs_add_system_chunk(chunk_root
, &key
, chunk
,
4605 free_extent_map(em
);
4610 * Chunk allocation falls into two parts. The first part does works
4611 * that make the new allocated chunk useable, but not do any operation
4612 * that modifies the chunk tree. The second part does the works that
4613 * require modifying the chunk tree. This division is important for the
4614 * bootstrap process of adding storage to a seed btrfs.
4616 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4617 struct btrfs_root
*extent_root
, u64 type
)
4621 chunk_offset
= find_next_chunk(extent_root
->fs_info
);
4622 return __btrfs_alloc_chunk(trans
, extent_root
, chunk_offset
, type
);
4625 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
4626 struct btrfs_root
*root
,
4627 struct btrfs_device
*device
)
4630 u64 sys_chunk_offset
;
4632 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4633 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
4636 chunk_offset
= find_next_chunk(fs_info
);
4637 alloc_profile
= btrfs_get_alloc_profile(extent_root
, 0);
4638 ret
= __btrfs_alloc_chunk(trans
, extent_root
, chunk_offset
,
4643 sys_chunk_offset
= find_next_chunk(root
->fs_info
);
4644 alloc_profile
= btrfs_get_alloc_profile(fs_info
->chunk_root
, 0);
4645 ret
= __btrfs_alloc_chunk(trans
, extent_root
, sys_chunk_offset
,
4650 static inline int btrfs_chunk_max_errors(struct map_lookup
*map
)
4654 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
4655 BTRFS_BLOCK_GROUP_RAID10
|
4656 BTRFS_BLOCK_GROUP_RAID5
|
4657 BTRFS_BLOCK_GROUP_DUP
)) {
4659 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
) {
4668 int btrfs_chunk_readonly(struct btrfs_root
*root
, u64 chunk_offset
)
4670 struct extent_map
*em
;
4671 struct map_lookup
*map
;
4672 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
4677 read_lock(&map_tree
->map_tree
.lock
);
4678 em
= lookup_extent_mapping(&map_tree
->map_tree
, chunk_offset
, 1);
4679 read_unlock(&map_tree
->map_tree
.lock
);
4683 map
= (struct map_lookup
*)em
->bdev
;
4684 for (i
= 0; i
< map
->num_stripes
; i
++) {
4685 if (map
->stripes
[i
].dev
->missing
) {
4690 if (!map
->stripes
[i
].dev
->writeable
) {
4697 * If the number of missing devices is larger than max errors,
4698 * we can not write the data into that chunk successfully, so
4701 if (miss_ndevs
> btrfs_chunk_max_errors(map
))
4704 free_extent_map(em
);
4708 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
4710 extent_map_tree_init(&tree
->map_tree
);
4713 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
4715 struct extent_map
*em
;
4718 write_lock(&tree
->map_tree
.lock
);
4719 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
4721 remove_extent_mapping(&tree
->map_tree
, em
);
4722 write_unlock(&tree
->map_tree
.lock
);
4726 free_extent_map(em
);
4727 /* once for the tree */
4728 free_extent_map(em
);
4732 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
4734 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
4735 struct extent_map
*em
;
4736 struct map_lookup
*map
;
4737 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4740 read_lock(&em_tree
->lock
);
4741 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4742 read_unlock(&em_tree
->lock
);
4745 * We could return errors for these cases, but that could get ugly and
4746 * we'd probably do the same thing which is just not do anything else
4747 * and exit, so return 1 so the callers don't try to use other copies.
4750 btrfs_crit(fs_info
, "No mapping for %Lu-%Lu", logical
,
4755 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
4756 btrfs_crit(fs_info
, "Invalid mapping for %Lu-%Lu, got "
4757 "%Lu-%Lu", logical
, logical
+len
, em
->start
,
4758 em
->start
+ em
->len
);
4759 free_extent_map(em
);
4763 map
= (struct map_lookup
*)em
->bdev
;
4764 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
4765 ret
= map
->num_stripes
;
4766 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
4767 ret
= map
->sub_stripes
;
4768 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID5
)
4770 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
4774 free_extent_map(em
);
4776 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
4777 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
))
4779 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
4784 unsigned long btrfs_full_stripe_len(struct btrfs_root
*root
,
4785 struct btrfs_mapping_tree
*map_tree
,
4788 struct extent_map
*em
;
4789 struct map_lookup
*map
;
4790 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4791 unsigned long len
= root
->sectorsize
;
4793 read_lock(&em_tree
->lock
);
4794 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4795 read_unlock(&em_tree
->lock
);
4798 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4799 map
= (struct map_lookup
*)em
->bdev
;
4800 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
4801 len
= map
->stripe_len
* nr_data_stripes(map
);
4802 free_extent_map(em
);
4806 int btrfs_is_parity_mirror(struct btrfs_mapping_tree
*map_tree
,
4807 u64 logical
, u64 len
, int mirror_num
)
4809 struct extent_map
*em
;
4810 struct map_lookup
*map
;
4811 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4814 read_lock(&em_tree
->lock
);
4815 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4816 read_unlock(&em_tree
->lock
);
4819 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4820 map
= (struct map_lookup
*)em
->bdev
;
4821 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
4823 free_extent_map(em
);
4827 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
4828 struct map_lookup
*map
, int first
, int num
,
4829 int optimal
, int dev_replace_is_ongoing
)
4833 struct btrfs_device
*srcdev
;
4835 if (dev_replace_is_ongoing
&&
4836 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
4837 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
4838 srcdev
= fs_info
->dev_replace
.srcdev
;
4843 * try to avoid the drive that is the source drive for a
4844 * dev-replace procedure, only choose it if no other non-missing
4845 * mirror is available
4847 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
4848 if (map
->stripes
[optimal
].dev
->bdev
&&
4849 (tolerance
|| map
->stripes
[optimal
].dev
!= srcdev
))
4851 for (i
= first
; i
< first
+ num
; i
++) {
4852 if (map
->stripes
[i
].dev
->bdev
&&
4853 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
4858 /* we couldn't find one that doesn't fail. Just return something
4859 * and the io error handling code will clean up eventually
4864 static inline int parity_smaller(u64 a
, u64 b
)
4869 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4870 static void sort_parity_stripes(struct btrfs_bio
*bbio
, int num_stripes
)
4872 struct btrfs_bio_stripe s
;
4879 for (i
= 0; i
< num_stripes
- 1; i
++) {
4880 if (parity_smaller(bbio
->raid_map
[i
],
4881 bbio
->raid_map
[i
+1])) {
4882 s
= bbio
->stripes
[i
];
4883 l
= bbio
->raid_map
[i
];
4884 bbio
->stripes
[i
] = bbio
->stripes
[i
+1];
4885 bbio
->raid_map
[i
] = bbio
->raid_map
[i
+1];
4886 bbio
->stripes
[i
+1] = s
;
4887 bbio
->raid_map
[i
+1] = l
;
4895 static struct btrfs_bio
*alloc_btrfs_bio(int total_stripes
, int real_stripes
)
4897 struct btrfs_bio
*bbio
= kzalloc(
4898 sizeof(struct btrfs_bio
) +
4899 sizeof(struct btrfs_bio_stripe
) * (total_stripes
) +
4900 sizeof(int) * (real_stripes
) +
4901 sizeof(u64
) * (real_stripes
),
4906 atomic_set(&bbio
->error
, 0);
4907 atomic_set(&bbio
->refs
, 1);
4912 void btrfs_get_bbio(struct btrfs_bio
*bbio
)
4914 WARN_ON(!atomic_read(&bbio
->refs
));
4915 atomic_inc(&bbio
->refs
);
4918 void btrfs_put_bbio(struct btrfs_bio
*bbio
)
4922 if (atomic_dec_and_test(&bbio
->refs
))
4926 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
4927 u64 logical
, u64
*length
,
4928 struct btrfs_bio
**bbio_ret
,
4929 int mirror_num
, int need_raid_map
)
4931 struct extent_map
*em
;
4932 struct map_lookup
*map
;
4933 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
4934 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4937 u64 stripe_end_offset
;
4947 int tgtdev_indexes
= 0;
4948 struct btrfs_bio
*bbio
= NULL
;
4949 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
4950 int dev_replace_is_ongoing
= 0;
4951 int num_alloc_stripes
;
4952 int patch_the_first_stripe_for_dev_replace
= 0;
4953 u64 physical_to_patch_in_first_stripe
= 0;
4954 u64 raid56_full_stripe_start
= (u64
)-1;
4956 read_lock(&em_tree
->lock
);
4957 em
= lookup_extent_mapping(em_tree
, logical
, *length
);
4958 read_unlock(&em_tree
->lock
);
4961 btrfs_crit(fs_info
, "unable to find logical %llu len %llu",
4966 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
4967 btrfs_crit(fs_info
, "found a bad mapping, wanted %Lu, "
4968 "found %Lu-%Lu", logical
, em
->start
,
4969 em
->start
+ em
->len
);
4970 free_extent_map(em
);
4974 map
= (struct map_lookup
*)em
->bdev
;
4975 offset
= logical
- em
->start
;
4977 stripe_len
= map
->stripe_len
;
4980 * stripe_nr counts the total number of stripes we have to stride
4981 * to get to this block
4983 do_div(stripe_nr
, stripe_len
);
4985 stripe_offset
= stripe_nr
* stripe_len
;
4986 BUG_ON(offset
< stripe_offset
);
4988 /* stripe_offset is the offset of this block in its stripe*/
4989 stripe_offset
= offset
- stripe_offset
;
4991 /* if we're here for raid56, we need to know the stripe aligned start */
4992 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
4993 unsigned long full_stripe_len
= stripe_len
* nr_data_stripes(map
);
4994 raid56_full_stripe_start
= offset
;
4996 /* allow a write of a full stripe, but make sure we don't
4997 * allow straddling of stripes
4999 do_div(raid56_full_stripe_start
, full_stripe_len
);
5000 raid56_full_stripe_start
*= full_stripe_len
;
5003 if (rw
& REQ_DISCARD
) {
5004 /* we don't discard raid56 yet */
5005 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5009 *length
= min_t(u64
, em
->len
- offset
, *length
);
5010 } else if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
5012 /* For writes to RAID[56], allow a full stripeset across all disks.
5013 For other RAID types and for RAID[56] reads, just allow a single
5014 stripe (on a single disk). */
5015 if ((map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
5017 max_len
= stripe_len
* nr_data_stripes(map
) -
5018 (offset
- raid56_full_stripe_start
);
5020 /* we limit the length of each bio to what fits in a stripe */
5021 max_len
= stripe_len
- stripe_offset
;
5023 *length
= min_t(u64
, em
->len
- offset
, max_len
);
5025 *length
= em
->len
- offset
;
5028 /* This is for when we're called from btrfs_merge_bio_hook() and all
5029 it cares about is the length */
5033 btrfs_dev_replace_lock(dev_replace
);
5034 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
5035 if (!dev_replace_is_ongoing
)
5036 btrfs_dev_replace_unlock(dev_replace
);
5038 if (dev_replace_is_ongoing
&& mirror_num
== map
->num_stripes
+ 1 &&
5039 !(rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) &&
5040 dev_replace
->tgtdev
!= NULL
) {
5042 * in dev-replace case, for repair case (that's the only
5043 * case where the mirror is selected explicitly when
5044 * calling btrfs_map_block), blocks left of the left cursor
5045 * can also be read from the target drive.
5046 * For REQ_GET_READ_MIRRORS, the target drive is added as
5047 * the last one to the array of stripes. For READ, it also
5048 * needs to be supported using the same mirror number.
5049 * If the requested block is not left of the left cursor,
5050 * EIO is returned. This can happen because btrfs_num_copies()
5051 * returns one more in the dev-replace case.
5053 u64 tmp_length
= *length
;
5054 struct btrfs_bio
*tmp_bbio
= NULL
;
5055 int tmp_num_stripes
;
5056 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5057 int index_srcdev
= 0;
5059 u64 physical_of_found
= 0;
5061 ret
= __btrfs_map_block(fs_info
, REQ_GET_READ_MIRRORS
,
5062 logical
, &tmp_length
, &tmp_bbio
, 0, 0);
5064 WARN_ON(tmp_bbio
!= NULL
);
5068 tmp_num_stripes
= tmp_bbio
->num_stripes
;
5069 if (mirror_num
> tmp_num_stripes
) {
5071 * REQ_GET_READ_MIRRORS does not contain this
5072 * mirror, that means that the requested area
5073 * is not left of the left cursor
5076 btrfs_put_bbio(tmp_bbio
);
5081 * process the rest of the function using the mirror_num
5082 * of the source drive. Therefore look it up first.
5083 * At the end, patch the device pointer to the one of the
5086 for (i
= 0; i
< tmp_num_stripes
; i
++) {
5087 if (tmp_bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5089 * In case of DUP, in order to keep it
5090 * simple, only add the mirror with the
5091 * lowest physical address
5094 physical_of_found
<=
5095 tmp_bbio
->stripes
[i
].physical
)
5100 tmp_bbio
->stripes
[i
].physical
;
5105 mirror_num
= index_srcdev
+ 1;
5106 patch_the_first_stripe_for_dev_replace
= 1;
5107 physical_to_patch_in_first_stripe
= physical_of_found
;
5111 btrfs_put_bbio(tmp_bbio
);
5115 btrfs_put_bbio(tmp_bbio
);
5116 } else if (mirror_num
> map
->num_stripes
) {
5122 stripe_nr_orig
= stripe_nr
;
5123 stripe_nr_end
= ALIGN(offset
+ *length
, map
->stripe_len
);
5124 do_div(stripe_nr_end
, map
->stripe_len
);
5125 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
5128 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5129 if (rw
& REQ_DISCARD
)
5130 num_stripes
= min_t(u64
, map
->num_stripes
,
5131 stripe_nr_end
- stripe_nr_orig
);
5132 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
5133 if (!(rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)))
5135 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
5136 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
))
5137 num_stripes
= map
->num_stripes
;
5138 else if (mirror_num
)
5139 stripe_index
= mirror_num
- 1;
5141 stripe_index
= find_live_mirror(fs_info
, map
, 0,
5143 current
->pid
% map
->num_stripes
,
5144 dev_replace_is_ongoing
);
5145 mirror_num
= stripe_index
+ 1;
5148 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
5149 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) {
5150 num_stripes
= map
->num_stripes
;
5151 } else if (mirror_num
) {
5152 stripe_index
= mirror_num
- 1;
5157 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5158 int factor
= map
->num_stripes
/ map
->sub_stripes
;
5160 stripe_index
= do_div(stripe_nr
, factor
);
5161 stripe_index
*= map
->sub_stripes
;
5163 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
))
5164 num_stripes
= map
->sub_stripes
;
5165 else if (rw
& REQ_DISCARD
)
5166 num_stripes
= min_t(u64
, map
->sub_stripes
*
5167 (stripe_nr_end
- stripe_nr_orig
),
5169 else if (mirror_num
)
5170 stripe_index
+= mirror_num
- 1;
5172 int old_stripe_index
= stripe_index
;
5173 stripe_index
= find_live_mirror(fs_info
, map
,
5175 map
->sub_stripes
, stripe_index
+
5176 current
->pid
% map
->sub_stripes
,
5177 dev_replace_is_ongoing
);
5178 mirror_num
= stripe_index
- old_stripe_index
+ 1;
5181 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5182 if (need_raid_map
&&
5183 ((rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
)) ||
5185 /* push stripe_nr back to the start of the full stripe */
5186 stripe_nr
= raid56_full_stripe_start
;
5187 do_div(stripe_nr
, stripe_len
* nr_data_stripes(map
));
5189 /* RAID[56] write or recovery. Return all stripes */
5190 num_stripes
= map
->num_stripes
;
5191 max_errors
= nr_parity_stripes(map
);
5193 *length
= map
->stripe_len
;
5200 * Mirror #0 or #1 means the original data block.
5201 * Mirror #2 is RAID5 parity block.
5202 * Mirror #3 is RAID6 Q block.
5204 stripe_index
= do_div(stripe_nr
, nr_data_stripes(map
));
5206 stripe_index
= nr_data_stripes(map
) +
5209 /* We distribute the parity blocks across stripes */
5210 tmp
= stripe_nr
+ stripe_index
;
5211 stripe_index
= do_div(tmp
, map
->num_stripes
);
5212 if (!(rw
& (REQ_WRITE
| REQ_DISCARD
|
5213 REQ_GET_READ_MIRRORS
)) && mirror_num
<= 1)
5218 * after this do_div call, stripe_nr is the number of stripes
5219 * on this device we have to walk to find the data, and
5220 * stripe_index is the number of our device in the stripe array
5222 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
5223 mirror_num
= stripe_index
+ 1;
5225 BUG_ON(stripe_index
>= map
->num_stripes
);
5227 num_alloc_stripes
= num_stripes
;
5228 if (dev_replace_is_ongoing
) {
5229 if (rw
& (REQ_WRITE
| REQ_DISCARD
))
5230 num_alloc_stripes
<<= 1;
5231 if (rw
& REQ_GET_READ_MIRRORS
)
5232 num_alloc_stripes
++;
5233 tgtdev_indexes
= num_stripes
;
5236 bbio
= alloc_btrfs_bio(num_alloc_stripes
, tgtdev_indexes
);
5241 if (dev_replace_is_ongoing
)
5242 bbio
->tgtdev_map
= (int *)(bbio
->stripes
+ num_alloc_stripes
);
5244 /* build raid_map */
5245 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
&&
5246 need_raid_map
&& ((rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
)) ||
5251 bbio
->raid_map
= (u64
*)((void *)bbio
->stripes
+
5252 sizeof(struct btrfs_bio_stripe
) *
5254 sizeof(int) * tgtdev_indexes
);
5256 /* Work out the disk rotation on this stripe-set */
5258 rot
= do_div(tmp
, num_stripes
);
5260 /* Fill in the logical address of each stripe */
5261 tmp
= stripe_nr
* nr_data_stripes(map
);
5262 for (i
= 0; i
< nr_data_stripes(map
); i
++)
5263 bbio
->raid_map
[(i
+rot
) % num_stripes
] =
5264 em
->start
+ (tmp
+ i
) * map
->stripe_len
;
5266 bbio
->raid_map
[(i
+rot
) % map
->num_stripes
] = RAID5_P_STRIPE
;
5267 if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5268 bbio
->raid_map
[(i
+rot
+1) % num_stripes
] =
5272 if (rw
& REQ_DISCARD
) {
5274 int sub_stripes
= 0;
5275 u64 stripes_per_dev
= 0;
5276 u32 remaining_stripes
= 0;
5277 u32 last_stripe
= 0;
5280 (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID10
)) {
5281 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5284 sub_stripes
= map
->sub_stripes
;
5286 factor
= map
->num_stripes
/ sub_stripes
;
5287 stripes_per_dev
= div_u64_rem(stripe_nr_end
-
5290 &remaining_stripes
);
5291 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
5292 last_stripe
*= sub_stripes
;
5295 for (i
= 0; i
< num_stripes
; i
++) {
5296 bbio
->stripes
[i
].physical
=
5297 map
->stripes
[stripe_index
].physical
+
5298 stripe_offset
+ stripe_nr
* map
->stripe_len
;
5299 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
5301 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5302 BTRFS_BLOCK_GROUP_RAID10
)) {
5303 bbio
->stripes
[i
].length
= stripes_per_dev
*
5306 if (i
/ sub_stripes
< remaining_stripes
)
5307 bbio
->stripes
[i
].length
+=
5311 * Special for the first stripe and
5314 * |-------|...|-------|
5318 if (i
< sub_stripes
)
5319 bbio
->stripes
[i
].length
-=
5322 if (stripe_index
>= last_stripe
&&
5323 stripe_index
<= (last_stripe
+
5325 bbio
->stripes
[i
].length
-=
5328 if (i
== sub_stripes
- 1)
5331 bbio
->stripes
[i
].length
= *length
;
5334 if (stripe_index
== map
->num_stripes
) {
5335 /* This could only happen for RAID0/10 */
5341 for (i
= 0; i
< num_stripes
; i
++) {
5342 bbio
->stripes
[i
].physical
=
5343 map
->stripes
[stripe_index
].physical
+
5345 stripe_nr
* map
->stripe_len
;
5346 bbio
->stripes
[i
].dev
=
5347 map
->stripes
[stripe_index
].dev
;
5352 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
))
5353 max_errors
= btrfs_chunk_max_errors(map
);
5356 sort_parity_stripes(bbio
, num_stripes
);
5359 if (dev_replace_is_ongoing
&& (rw
& (REQ_WRITE
| REQ_DISCARD
)) &&
5360 dev_replace
->tgtdev
!= NULL
) {
5361 int index_where_to_add
;
5362 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5365 * duplicate the write operations while the dev replace
5366 * procedure is running. Since the copying of the old disk
5367 * to the new disk takes place at run time while the
5368 * filesystem is mounted writable, the regular write
5369 * operations to the old disk have to be duplicated to go
5370 * to the new disk as well.
5371 * Note that device->missing is handled by the caller, and
5372 * that the write to the old disk is already set up in the
5375 index_where_to_add
= num_stripes
;
5376 for (i
= 0; i
< num_stripes
; i
++) {
5377 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5378 /* write to new disk, too */
5379 struct btrfs_bio_stripe
*new =
5380 bbio
->stripes
+ index_where_to_add
;
5381 struct btrfs_bio_stripe
*old
=
5384 new->physical
= old
->physical
;
5385 new->length
= old
->length
;
5386 new->dev
= dev_replace
->tgtdev
;
5387 bbio
->tgtdev_map
[i
] = index_where_to_add
;
5388 index_where_to_add
++;
5393 num_stripes
= index_where_to_add
;
5394 } else if (dev_replace_is_ongoing
&& (rw
& REQ_GET_READ_MIRRORS
) &&
5395 dev_replace
->tgtdev
!= NULL
) {
5396 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5397 int index_srcdev
= 0;
5399 u64 physical_of_found
= 0;
5402 * During the dev-replace procedure, the target drive can
5403 * also be used to read data in case it is needed to repair
5404 * a corrupt block elsewhere. This is possible if the
5405 * requested area is left of the left cursor. In this area,
5406 * the target drive is a full copy of the source drive.
5408 for (i
= 0; i
< num_stripes
; i
++) {
5409 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5411 * In case of DUP, in order to keep it
5412 * simple, only add the mirror with the
5413 * lowest physical address
5416 physical_of_found
<=
5417 bbio
->stripes
[i
].physical
)
5421 physical_of_found
= bbio
->stripes
[i
].physical
;
5425 u64 length
= map
->stripe_len
;
5427 if (physical_of_found
+ length
<=
5428 dev_replace
->cursor_left
) {
5429 struct btrfs_bio_stripe
*tgtdev_stripe
=
5430 bbio
->stripes
+ num_stripes
;
5432 tgtdev_stripe
->physical
= physical_of_found
;
5433 tgtdev_stripe
->length
=
5434 bbio
->stripes
[index_srcdev
].length
;
5435 tgtdev_stripe
->dev
= dev_replace
->tgtdev
;
5436 bbio
->tgtdev_map
[index_srcdev
] = num_stripes
;
5445 bbio
->map_type
= map
->type
;
5446 bbio
->num_stripes
= num_stripes
;
5447 bbio
->max_errors
= max_errors
;
5448 bbio
->mirror_num
= mirror_num
;
5449 bbio
->num_tgtdevs
= tgtdev_indexes
;
5452 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5453 * mirror_num == num_stripes + 1 && dev_replace target drive is
5454 * available as a mirror
5456 if (patch_the_first_stripe_for_dev_replace
&& num_stripes
> 0) {
5457 WARN_ON(num_stripes
> 1);
5458 bbio
->stripes
[0].dev
= dev_replace
->tgtdev
;
5459 bbio
->stripes
[0].physical
= physical_to_patch_in_first_stripe
;
5460 bbio
->mirror_num
= map
->num_stripes
+ 1;
5463 if (dev_replace_is_ongoing
)
5464 btrfs_dev_replace_unlock(dev_replace
);
5465 free_extent_map(em
);
5469 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
5470 u64 logical
, u64
*length
,
5471 struct btrfs_bio
**bbio_ret
, int mirror_num
)
5473 return __btrfs_map_block(fs_info
, rw
, logical
, length
, bbio_ret
,
5477 /* For Scrub/replace */
5478 int btrfs_map_sblock(struct btrfs_fs_info
*fs_info
, int rw
,
5479 u64 logical
, u64
*length
,
5480 struct btrfs_bio
**bbio_ret
, int mirror_num
,
5483 return __btrfs_map_block(fs_info
, rw
, logical
, length
, bbio_ret
,
5484 mirror_num
, need_raid_map
);
5487 int btrfs_rmap_block(struct btrfs_mapping_tree
*map_tree
,
5488 u64 chunk_start
, u64 physical
, u64 devid
,
5489 u64
**logical
, int *naddrs
, int *stripe_len
)
5491 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5492 struct extent_map
*em
;
5493 struct map_lookup
*map
;
5501 read_lock(&em_tree
->lock
);
5502 em
= lookup_extent_mapping(em_tree
, chunk_start
, 1);
5503 read_unlock(&em_tree
->lock
);
5506 printk(KERN_ERR
"BTRFS: couldn't find em for chunk %Lu\n",
5511 if (em
->start
!= chunk_start
) {
5512 printk(KERN_ERR
"BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5513 em
->start
, chunk_start
);
5514 free_extent_map(em
);
5517 map
= (struct map_lookup
*)em
->bdev
;
5520 rmap_len
= map
->stripe_len
;
5522 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5523 do_div(length
, map
->num_stripes
/ map
->sub_stripes
);
5524 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5525 do_div(length
, map
->num_stripes
);
5526 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5527 do_div(length
, nr_data_stripes(map
));
5528 rmap_len
= map
->stripe_len
* nr_data_stripes(map
);
5531 buf
= kzalloc(sizeof(u64
) * map
->num_stripes
, GFP_NOFS
);
5532 BUG_ON(!buf
); /* -ENOMEM */
5534 for (i
= 0; i
< map
->num_stripes
; i
++) {
5535 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
5537 if (map
->stripes
[i
].physical
> physical
||
5538 map
->stripes
[i
].physical
+ length
<= physical
)
5541 stripe_nr
= physical
- map
->stripes
[i
].physical
;
5542 do_div(stripe_nr
, map
->stripe_len
);
5544 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5545 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5546 do_div(stripe_nr
, map
->sub_stripes
);
5547 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5548 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5549 } /* else if RAID[56], multiply by nr_data_stripes().
5550 * Alternatively, just use rmap_len below instead of
5551 * map->stripe_len */
5553 bytenr
= chunk_start
+ stripe_nr
* rmap_len
;
5554 WARN_ON(nr
>= map
->num_stripes
);
5555 for (j
= 0; j
< nr
; j
++) {
5556 if (buf
[j
] == bytenr
)
5560 WARN_ON(nr
>= map
->num_stripes
);
5567 *stripe_len
= rmap_len
;
5569 free_extent_map(em
);
5573 static inline void btrfs_end_bbio(struct btrfs_bio
*bbio
, struct bio
*bio
, int err
)
5575 if (likely(bbio
->flags
& BTRFS_BIO_ORIG_BIO_SUBMITTED
))
5576 bio_endio_nodec(bio
, err
);
5578 bio_endio(bio
, err
);
5579 btrfs_put_bbio(bbio
);
5582 static void btrfs_end_bio(struct bio
*bio
, int err
)
5584 struct btrfs_bio
*bbio
= bio
->bi_private
;
5585 struct btrfs_device
*dev
= bbio
->stripes
[0].dev
;
5586 int is_orig_bio
= 0;
5589 atomic_inc(&bbio
->error
);
5590 if (err
== -EIO
|| err
== -EREMOTEIO
) {
5591 unsigned int stripe_index
=
5592 btrfs_io_bio(bio
)->stripe_index
;
5594 BUG_ON(stripe_index
>= bbio
->num_stripes
);
5595 dev
= bbio
->stripes
[stripe_index
].dev
;
5597 if (bio
->bi_rw
& WRITE
)
5598 btrfs_dev_stat_inc(dev
,
5599 BTRFS_DEV_STAT_WRITE_ERRS
);
5601 btrfs_dev_stat_inc(dev
,
5602 BTRFS_DEV_STAT_READ_ERRS
);
5603 if ((bio
->bi_rw
& WRITE_FLUSH
) == WRITE_FLUSH
)
5604 btrfs_dev_stat_inc(dev
,
5605 BTRFS_DEV_STAT_FLUSH_ERRS
);
5606 btrfs_dev_stat_print_on_error(dev
);
5611 if (bio
== bbio
->orig_bio
)
5614 btrfs_bio_counter_dec(bbio
->fs_info
);
5616 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
5619 bio
= bbio
->orig_bio
;
5622 bio
->bi_private
= bbio
->private;
5623 bio
->bi_end_io
= bbio
->end_io
;
5624 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
5625 /* only send an error to the higher layers if it is
5626 * beyond the tolerance of the btrfs bio
5628 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
5632 * this bio is actually up to date, we didn't
5633 * go over the max number of errors
5635 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5639 btrfs_end_bbio(bbio
, bio
, err
);
5640 } else if (!is_orig_bio
) {
5646 * see run_scheduled_bios for a description of why bios are collected for
5649 * This will add one bio to the pending list for a device and make sure
5650 * the work struct is scheduled.
5652 static noinline
void btrfs_schedule_bio(struct btrfs_root
*root
,
5653 struct btrfs_device
*device
,
5654 int rw
, struct bio
*bio
)
5656 int should_queue
= 1;
5657 struct btrfs_pending_bios
*pending_bios
;
5659 if (device
->missing
|| !device
->bdev
) {
5660 bio_endio(bio
, -EIO
);
5664 /* don't bother with additional async steps for reads, right now */
5665 if (!(rw
& REQ_WRITE
)) {
5667 btrfsic_submit_bio(rw
, bio
);
5673 * nr_async_bios allows us to reliably return congestion to the
5674 * higher layers. Otherwise, the async bio makes it appear we have
5675 * made progress against dirty pages when we've really just put it
5676 * on a queue for later
5678 atomic_inc(&root
->fs_info
->nr_async_bios
);
5679 WARN_ON(bio
->bi_next
);
5680 bio
->bi_next
= NULL
;
5683 spin_lock(&device
->io_lock
);
5684 if (bio
->bi_rw
& REQ_SYNC
)
5685 pending_bios
= &device
->pending_sync_bios
;
5687 pending_bios
= &device
->pending_bios
;
5689 if (pending_bios
->tail
)
5690 pending_bios
->tail
->bi_next
= bio
;
5692 pending_bios
->tail
= bio
;
5693 if (!pending_bios
->head
)
5694 pending_bios
->head
= bio
;
5695 if (device
->running_pending
)
5698 spin_unlock(&device
->io_lock
);
5701 btrfs_queue_work(root
->fs_info
->submit_workers
,
5705 static int bio_size_ok(struct block_device
*bdev
, struct bio
*bio
,
5708 struct bio_vec
*prev
;
5709 struct request_queue
*q
= bdev_get_queue(bdev
);
5710 unsigned int max_sectors
= queue_max_sectors(q
);
5711 struct bvec_merge_data bvm
= {
5713 .bi_sector
= sector
,
5714 .bi_rw
= bio
->bi_rw
,
5717 if (WARN_ON(bio
->bi_vcnt
== 0))
5720 prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
5721 if (bio_sectors(bio
) > max_sectors
)
5724 if (!q
->merge_bvec_fn
)
5727 bvm
.bi_size
= bio
->bi_iter
.bi_size
- prev
->bv_len
;
5728 if (q
->merge_bvec_fn(q
, &bvm
, prev
) < prev
->bv_len
)
5733 static void submit_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
5734 struct bio
*bio
, u64 physical
, int dev_nr
,
5737 struct btrfs_device
*dev
= bbio
->stripes
[dev_nr
].dev
;
5739 bio
->bi_private
= bbio
;
5740 btrfs_io_bio(bio
)->stripe_index
= dev_nr
;
5741 bio
->bi_end_io
= btrfs_end_bio
;
5742 bio
->bi_iter
.bi_sector
= physical
>> 9;
5745 struct rcu_string
*name
;
5748 name
= rcu_dereference(dev
->name
);
5749 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5750 "(%s id %llu), size=%u\n", rw
,
5751 (u64
)bio
->bi_iter
.bi_sector
, (u_long
)dev
->bdev
->bd_dev
,
5752 name
->str
, dev
->devid
, bio
->bi_iter
.bi_size
);
5756 bio
->bi_bdev
= dev
->bdev
;
5758 btrfs_bio_counter_inc_noblocked(root
->fs_info
);
5761 btrfs_schedule_bio(root
, dev
, rw
, bio
);
5763 btrfsic_submit_bio(rw
, bio
);
5766 static int breakup_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
5767 struct bio
*first_bio
, struct btrfs_device
*dev
,
5768 int dev_nr
, int rw
, int async
)
5770 struct bio_vec
*bvec
= first_bio
->bi_io_vec
;
5772 int nr_vecs
= bio_get_nr_vecs(dev
->bdev
);
5773 u64 physical
= bbio
->stripes
[dev_nr
].physical
;
5776 bio
= btrfs_bio_alloc(dev
->bdev
, physical
>> 9, nr_vecs
, GFP_NOFS
);
5780 while (bvec
<= (first_bio
->bi_io_vec
+ first_bio
->bi_vcnt
- 1)) {
5781 if (bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
5782 bvec
->bv_offset
) < bvec
->bv_len
) {
5783 u64 len
= bio
->bi_iter
.bi_size
;
5785 atomic_inc(&bbio
->stripes_pending
);
5786 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
,
5794 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
, rw
, async
);
5798 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
5800 atomic_inc(&bbio
->error
);
5801 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
5802 /* Shoud be the original bio. */
5803 WARN_ON(bio
!= bbio
->orig_bio
);
5805 bio
->bi_private
= bbio
->private;
5806 bio
->bi_end_io
= bbio
->end_io
;
5807 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
5808 bio
->bi_iter
.bi_sector
= logical
>> 9;
5810 btrfs_end_bbio(bbio
, bio
, -EIO
);
5814 int btrfs_map_bio(struct btrfs_root
*root
, int rw
, struct bio
*bio
,
5815 int mirror_num
, int async_submit
)
5817 struct btrfs_device
*dev
;
5818 struct bio
*first_bio
= bio
;
5819 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
5825 struct btrfs_bio
*bbio
= NULL
;
5827 length
= bio
->bi_iter
.bi_size
;
5828 map_length
= length
;
5830 btrfs_bio_counter_inc_blocked(root
->fs_info
);
5831 ret
= __btrfs_map_block(root
->fs_info
, rw
, logical
, &map_length
, &bbio
,
5834 btrfs_bio_counter_dec(root
->fs_info
);
5838 total_devs
= bbio
->num_stripes
;
5839 bbio
->orig_bio
= first_bio
;
5840 bbio
->private = first_bio
->bi_private
;
5841 bbio
->end_io
= first_bio
->bi_end_io
;
5842 bbio
->fs_info
= root
->fs_info
;
5843 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
5845 if (bbio
->raid_map
) {
5846 /* In this case, map_length has been set to the length of
5847 a single stripe; not the whole write */
5849 ret
= raid56_parity_write(root
, bio
, bbio
, map_length
);
5851 ret
= raid56_parity_recover(root
, bio
, bbio
, map_length
,
5855 btrfs_bio_counter_dec(root
->fs_info
);
5859 if (map_length
< length
) {
5860 btrfs_crit(root
->fs_info
, "mapping failed logical %llu bio len %llu len %llu",
5861 logical
, length
, map_length
);
5865 for (dev_nr
= 0; dev_nr
< total_devs
; dev_nr
++) {
5866 dev
= bbio
->stripes
[dev_nr
].dev
;
5867 if (!dev
|| !dev
->bdev
|| (rw
& WRITE
&& !dev
->writeable
)) {
5868 bbio_error(bbio
, first_bio
, logical
);
5873 * Check and see if we're ok with this bio based on it's size
5874 * and offset with the given device.
5876 if (!bio_size_ok(dev
->bdev
, first_bio
,
5877 bbio
->stripes
[dev_nr
].physical
>> 9)) {
5878 ret
= breakup_stripe_bio(root
, bbio
, first_bio
, dev
,
5879 dev_nr
, rw
, async_submit
);
5884 if (dev_nr
< total_devs
- 1) {
5885 bio
= btrfs_bio_clone(first_bio
, GFP_NOFS
);
5886 BUG_ON(!bio
); /* -ENOMEM */
5889 bbio
->flags
|= BTRFS_BIO_ORIG_BIO_SUBMITTED
;
5892 submit_stripe_bio(root
, bbio
, bio
,
5893 bbio
->stripes
[dev_nr
].physical
, dev_nr
, rw
,
5896 btrfs_bio_counter_dec(root
->fs_info
);
5900 struct btrfs_device
*btrfs_find_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
5903 struct btrfs_device
*device
;
5904 struct btrfs_fs_devices
*cur_devices
;
5906 cur_devices
= fs_info
->fs_devices
;
5907 while (cur_devices
) {
5909 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
5910 device
= __find_device(&cur_devices
->devices
,
5915 cur_devices
= cur_devices
->seed
;
5920 static struct btrfs_device
*add_missing_dev(struct btrfs_root
*root
,
5921 struct btrfs_fs_devices
*fs_devices
,
5922 u64 devid
, u8
*dev_uuid
)
5924 struct btrfs_device
*device
;
5926 device
= btrfs_alloc_device(NULL
, &devid
, dev_uuid
);
5930 list_add(&device
->dev_list
, &fs_devices
->devices
);
5931 device
->fs_devices
= fs_devices
;
5932 fs_devices
->num_devices
++;
5934 device
->missing
= 1;
5935 fs_devices
->missing_devices
++;
5941 * btrfs_alloc_device - allocate struct btrfs_device
5942 * @fs_info: used only for generating a new devid, can be NULL if
5943 * devid is provided (i.e. @devid != NULL).
5944 * @devid: a pointer to devid for this device. If NULL a new devid
5946 * @uuid: a pointer to UUID for this device. If NULL a new UUID
5949 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5950 * on error. Returned struct is not linked onto any lists and can be
5951 * destroyed with kfree() right away.
5953 struct btrfs_device
*btrfs_alloc_device(struct btrfs_fs_info
*fs_info
,
5957 struct btrfs_device
*dev
;
5960 if (WARN_ON(!devid
&& !fs_info
))
5961 return ERR_PTR(-EINVAL
);
5963 dev
= __alloc_device();
5972 ret
= find_next_devid(fs_info
, &tmp
);
5975 return ERR_PTR(ret
);
5981 memcpy(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
);
5983 generate_random_uuid(dev
->uuid
);
5985 btrfs_init_work(&dev
->work
, btrfs_submit_helper
,
5986 pending_bios_fn
, NULL
, NULL
);
5991 static int read_one_chunk(struct btrfs_root
*root
, struct btrfs_key
*key
,
5992 struct extent_buffer
*leaf
,
5993 struct btrfs_chunk
*chunk
)
5995 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
5996 struct map_lookup
*map
;
5997 struct extent_map
*em
;
6001 u8 uuid
[BTRFS_UUID_SIZE
];
6006 logical
= key
->offset
;
6007 length
= btrfs_chunk_length(leaf
, chunk
);
6009 read_lock(&map_tree
->map_tree
.lock
);
6010 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
6011 read_unlock(&map_tree
->map_tree
.lock
);
6013 /* already mapped? */
6014 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
6015 free_extent_map(em
);
6018 free_extent_map(em
);
6021 em
= alloc_extent_map();
6024 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6025 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
6027 free_extent_map(em
);
6031 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
6032 em
->bdev
= (struct block_device
*)map
;
6033 em
->start
= logical
;
6036 em
->block_start
= 0;
6037 em
->block_len
= em
->len
;
6039 map
->num_stripes
= num_stripes
;
6040 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
6041 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
6042 map
->sector_size
= btrfs_chunk_sector_size(leaf
, chunk
);
6043 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6044 map
->type
= btrfs_chunk_type(leaf
, chunk
);
6045 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6046 for (i
= 0; i
< num_stripes
; i
++) {
6047 map
->stripes
[i
].physical
=
6048 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
6049 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
6050 read_extent_buffer(leaf
, uuid
, (unsigned long)
6051 btrfs_stripe_dev_uuid_nr(chunk
, i
),
6053 map
->stripes
[i
].dev
= btrfs_find_device(root
->fs_info
, devid
,
6055 if (!map
->stripes
[i
].dev
&& !btrfs_test_opt(root
, DEGRADED
)) {
6056 free_extent_map(em
);
6059 if (!map
->stripes
[i
].dev
) {
6060 map
->stripes
[i
].dev
=
6061 add_missing_dev(root
, root
->fs_info
->fs_devices
,
6063 if (!map
->stripes
[i
].dev
) {
6064 free_extent_map(em
);
6068 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
6071 write_lock(&map_tree
->map_tree
.lock
);
6072 ret
= add_extent_mapping(&map_tree
->map_tree
, em
, 0);
6073 write_unlock(&map_tree
->map_tree
.lock
);
6074 BUG_ON(ret
); /* Tree corruption */
6075 free_extent_map(em
);
6080 static void fill_device_from_item(struct extent_buffer
*leaf
,
6081 struct btrfs_dev_item
*dev_item
,
6082 struct btrfs_device
*device
)
6086 device
->devid
= btrfs_device_id(leaf
, dev_item
);
6087 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
6088 device
->total_bytes
= device
->disk_total_bytes
;
6089 device
->commit_total_bytes
= device
->disk_total_bytes
;
6090 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
6091 device
->commit_bytes_used
= device
->bytes_used
;
6092 device
->type
= btrfs_device_type(leaf
, dev_item
);
6093 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
6094 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
6095 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
6096 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
6097 device
->is_tgtdev_for_dev_replace
= 0;
6099 ptr
= btrfs_device_uuid(dev_item
);
6100 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
6103 static struct btrfs_fs_devices
*open_seed_devices(struct btrfs_root
*root
,
6106 struct btrfs_fs_devices
*fs_devices
;
6109 BUG_ON(!mutex_is_locked(&uuid_mutex
));
6111 fs_devices
= root
->fs_info
->fs_devices
->seed
;
6112 while (fs_devices
) {
6113 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
))
6116 fs_devices
= fs_devices
->seed
;
6119 fs_devices
= find_fsid(fsid
);
6121 if (!btrfs_test_opt(root
, DEGRADED
))
6122 return ERR_PTR(-ENOENT
);
6124 fs_devices
= alloc_fs_devices(fsid
);
6125 if (IS_ERR(fs_devices
))
6128 fs_devices
->seeding
= 1;
6129 fs_devices
->opened
= 1;
6133 fs_devices
= clone_fs_devices(fs_devices
);
6134 if (IS_ERR(fs_devices
))
6137 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
6138 root
->fs_info
->bdev_holder
);
6140 free_fs_devices(fs_devices
);
6141 fs_devices
= ERR_PTR(ret
);
6145 if (!fs_devices
->seeding
) {
6146 __btrfs_close_devices(fs_devices
);
6147 free_fs_devices(fs_devices
);
6148 fs_devices
= ERR_PTR(-EINVAL
);
6152 fs_devices
->seed
= root
->fs_info
->fs_devices
->seed
;
6153 root
->fs_info
->fs_devices
->seed
= fs_devices
;
6158 static int read_one_dev(struct btrfs_root
*root
,
6159 struct extent_buffer
*leaf
,
6160 struct btrfs_dev_item
*dev_item
)
6162 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
6163 struct btrfs_device
*device
;
6166 u8 fs_uuid
[BTRFS_UUID_SIZE
];
6167 u8 dev_uuid
[BTRFS_UUID_SIZE
];
6169 devid
= btrfs_device_id(leaf
, dev_item
);
6170 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
6172 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
6175 if (memcmp(fs_uuid
, root
->fs_info
->fsid
, BTRFS_UUID_SIZE
)) {
6176 fs_devices
= open_seed_devices(root
, fs_uuid
);
6177 if (IS_ERR(fs_devices
))
6178 return PTR_ERR(fs_devices
);
6181 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
, fs_uuid
);
6183 if (!btrfs_test_opt(root
, DEGRADED
))
6186 btrfs_warn(root
->fs_info
, "devid %llu missing", devid
);
6187 device
= add_missing_dev(root
, fs_devices
, devid
, dev_uuid
);
6191 if (!device
->bdev
&& !btrfs_test_opt(root
, DEGRADED
))
6194 if(!device
->bdev
&& !device
->missing
) {
6196 * this happens when a device that was properly setup
6197 * in the device info lists suddenly goes bad.
6198 * device->bdev is NULL, and so we have to set
6199 * device->missing to one here
6201 device
->fs_devices
->missing_devices
++;
6202 device
->missing
= 1;
6205 /* Move the device to its own fs_devices */
6206 if (device
->fs_devices
!= fs_devices
) {
6207 ASSERT(device
->missing
);
6209 list_move(&device
->dev_list
, &fs_devices
->devices
);
6210 device
->fs_devices
->num_devices
--;
6211 fs_devices
->num_devices
++;
6213 device
->fs_devices
->missing_devices
--;
6214 fs_devices
->missing_devices
++;
6216 device
->fs_devices
= fs_devices
;
6220 if (device
->fs_devices
!= root
->fs_info
->fs_devices
) {
6221 BUG_ON(device
->writeable
);
6222 if (device
->generation
!=
6223 btrfs_device_generation(leaf
, dev_item
))
6227 fill_device_from_item(leaf
, dev_item
, device
);
6228 device
->in_fs_metadata
= 1;
6229 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
6230 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
6231 spin_lock(&root
->fs_info
->free_chunk_lock
);
6232 root
->fs_info
->free_chunk_space
+= device
->total_bytes
-
6234 spin_unlock(&root
->fs_info
->free_chunk_lock
);
6240 int btrfs_read_sys_array(struct btrfs_root
*root
)
6242 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
6243 struct extent_buffer
*sb
;
6244 struct btrfs_disk_key
*disk_key
;
6245 struct btrfs_chunk
*chunk
;
6247 unsigned long sb_array_offset
;
6253 struct btrfs_key key
;
6255 ASSERT(BTRFS_SUPER_INFO_SIZE
<= root
->nodesize
);
6257 * This will create extent buffer of nodesize, superblock size is
6258 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6259 * overallocate but we can keep it as-is, only the first page is used.
6261 sb
= btrfs_find_create_tree_block(root
, BTRFS_SUPER_INFO_OFFSET
);
6264 btrfs_set_buffer_uptodate(sb
);
6265 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
6267 * The sb extent buffer is artifical and just used to read the system array.
6268 * btrfs_set_buffer_uptodate() call does not properly mark all it's
6269 * pages up-to-date when the page is larger: extent does not cover the
6270 * whole page and consequently check_page_uptodate does not find all
6271 * the page's extents up-to-date (the hole beyond sb),
6272 * write_extent_buffer then triggers a WARN_ON.
6274 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6275 * but sb spans only this function. Add an explicit SetPageUptodate call
6276 * to silence the warning eg. on PowerPC 64.
6278 if (PAGE_CACHE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
6279 SetPageUptodate(sb
->pages
[0]);
6281 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
6282 array_size
= btrfs_super_sys_array_size(super_copy
);
6284 array_ptr
= super_copy
->sys_chunk_array
;
6285 sb_array_offset
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
6288 while (cur_offset
< array_size
) {
6289 disk_key
= (struct btrfs_disk_key
*)array_ptr
;
6290 len
= sizeof(*disk_key
);
6291 if (cur_offset
+ len
> array_size
)
6292 goto out_short_read
;
6294 btrfs_disk_key_to_cpu(&key
, disk_key
);
6297 sb_array_offset
+= len
;
6300 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6301 chunk
= (struct btrfs_chunk
*)sb_array_offset
;
6303 * At least one btrfs_chunk with one stripe must be
6304 * present, exact stripe count check comes afterwards
6306 len
= btrfs_chunk_item_size(1);
6307 if (cur_offset
+ len
> array_size
)
6308 goto out_short_read
;
6310 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
6311 len
= btrfs_chunk_item_size(num_stripes
);
6312 if (cur_offset
+ len
> array_size
)
6313 goto out_short_read
;
6315 ret
= read_one_chunk(root
, &key
, sb
, chunk
);
6323 sb_array_offset
+= len
;
6326 free_extent_buffer(sb
);
6330 printk(KERN_ERR
"BTRFS: sys_array too short to read %u bytes at offset %u\n",
6332 free_extent_buffer(sb
);
6336 int btrfs_read_chunk_tree(struct btrfs_root
*root
)
6338 struct btrfs_path
*path
;
6339 struct extent_buffer
*leaf
;
6340 struct btrfs_key key
;
6341 struct btrfs_key found_key
;
6345 root
= root
->fs_info
->chunk_root
;
6347 path
= btrfs_alloc_path();
6351 mutex_lock(&uuid_mutex
);
6355 * Read all device items, and then all the chunk items. All
6356 * device items are found before any chunk item (their object id
6357 * is smaller than the lowest possible object id for a chunk
6358 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6360 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
6363 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6367 leaf
= path
->nodes
[0];
6368 slot
= path
->slots
[0];
6369 if (slot
>= btrfs_header_nritems(leaf
)) {
6370 ret
= btrfs_next_leaf(root
, path
);
6377 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6378 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
6379 struct btrfs_dev_item
*dev_item
;
6380 dev_item
= btrfs_item_ptr(leaf
, slot
,
6381 struct btrfs_dev_item
);
6382 ret
= read_one_dev(root
, leaf
, dev_item
);
6385 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6386 struct btrfs_chunk
*chunk
;
6387 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
6388 ret
= read_one_chunk(root
, &found_key
, leaf
, chunk
);
6396 unlock_chunks(root
);
6397 mutex_unlock(&uuid_mutex
);
6399 btrfs_free_path(path
);
6403 void btrfs_init_devices_late(struct btrfs_fs_info
*fs_info
)
6405 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6406 struct btrfs_device
*device
;
6408 while (fs_devices
) {
6409 mutex_lock(&fs_devices
->device_list_mutex
);
6410 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
)
6411 device
->dev_root
= fs_info
->dev_root
;
6412 mutex_unlock(&fs_devices
->device_list_mutex
);
6414 fs_devices
= fs_devices
->seed
;
6418 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
6422 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6423 btrfs_dev_stat_reset(dev
, i
);
6426 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
6428 struct btrfs_key key
;
6429 struct btrfs_key found_key
;
6430 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6431 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6432 struct extent_buffer
*eb
;
6435 struct btrfs_device
*device
;
6436 struct btrfs_path
*path
= NULL
;
6439 path
= btrfs_alloc_path();
6445 mutex_lock(&fs_devices
->device_list_mutex
);
6446 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6448 struct btrfs_dev_stats_item
*ptr
;
6451 key
.type
= BTRFS_DEV_STATS_KEY
;
6452 key
.offset
= device
->devid
;
6453 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
6455 __btrfs_reset_dev_stats(device
);
6456 device
->dev_stats_valid
= 1;
6457 btrfs_release_path(path
);
6460 slot
= path
->slots
[0];
6461 eb
= path
->nodes
[0];
6462 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
6463 item_size
= btrfs_item_size_nr(eb
, slot
);
6465 ptr
= btrfs_item_ptr(eb
, slot
,
6466 struct btrfs_dev_stats_item
);
6468 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6469 if (item_size
>= (1 + i
) * sizeof(__le64
))
6470 btrfs_dev_stat_set(device
, i
,
6471 btrfs_dev_stats_value(eb
, ptr
, i
));
6473 btrfs_dev_stat_reset(device
, i
);
6476 device
->dev_stats_valid
= 1;
6477 btrfs_dev_stat_print_on_load(device
);
6478 btrfs_release_path(path
);
6480 mutex_unlock(&fs_devices
->device_list_mutex
);
6483 btrfs_free_path(path
);
6484 return ret
< 0 ? ret
: 0;
6487 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
6488 struct btrfs_root
*dev_root
,
6489 struct btrfs_device
*device
)
6491 struct btrfs_path
*path
;
6492 struct btrfs_key key
;
6493 struct extent_buffer
*eb
;
6494 struct btrfs_dev_stats_item
*ptr
;
6499 key
.type
= BTRFS_DEV_STATS_KEY
;
6500 key
.offset
= device
->devid
;
6502 path
= btrfs_alloc_path();
6504 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
6506 printk_in_rcu(KERN_WARNING
"BTRFS: "
6507 "error %d while searching for dev_stats item for device %s!\n",
6508 ret
, rcu_str_deref(device
->name
));
6513 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
6514 /* need to delete old one and insert a new one */
6515 ret
= btrfs_del_item(trans
, dev_root
, path
);
6517 printk_in_rcu(KERN_WARNING
"BTRFS: "
6518 "delete too small dev_stats item for device %s failed %d!\n",
6519 rcu_str_deref(device
->name
), ret
);
6526 /* need to insert a new item */
6527 btrfs_release_path(path
);
6528 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
6529 &key
, sizeof(*ptr
));
6531 printk_in_rcu(KERN_WARNING
"BTRFS: "
6532 "insert dev_stats item for device %s failed %d!\n",
6533 rcu_str_deref(device
->name
), ret
);
6538 eb
= path
->nodes
[0];
6539 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
6540 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6541 btrfs_set_dev_stats_value(eb
, ptr
, i
,
6542 btrfs_dev_stat_read(device
, i
));
6543 btrfs_mark_buffer_dirty(eb
);
6546 btrfs_free_path(path
);
6551 * called from commit_transaction. Writes all changed device stats to disk.
6553 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
6554 struct btrfs_fs_info
*fs_info
)
6556 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6557 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6558 struct btrfs_device
*device
;
6562 mutex_lock(&fs_devices
->device_list_mutex
);
6563 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6564 if (!device
->dev_stats_valid
|| !btrfs_dev_stats_dirty(device
))
6567 stats_cnt
= atomic_read(&device
->dev_stats_ccnt
);
6568 ret
= update_dev_stat_item(trans
, dev_root
, device
);
6570 atomic_sub(stats_cnt
, &device
->dev_stats_ccnt
);
6572 mutex_unlock(&fs_devices
->device_list_mutex
);
6577 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
6579 btrfs_dev_stat_inc(dev
, index
);
6580 btrfs_dev_stat_print_on_error(dev
);
6583 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
6585 if (!dev
->dev_stats_valid
)
6587 printk_ratelimited_in_rcu(KERN_ERR
"BTRFS: "
6588 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6589 rcu_str_deref(dev
->name
),
6590 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
6591 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
6592 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
6593 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
6594 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
6597 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
6601 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6602 if (btrfs_dev_stat_read(dev
, i
) != 0)
6604 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
6605 return; /* all values == 0, suppress message */
6607 printk_in_rcu(KERN_INFO
"BTRFS: "
6608 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6609 rcu_str_deref(dev
->name
),
6610 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
6611 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
6612 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
6613 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
6614 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
6617 int btrfs_get_dev_stats(struct btrfs_root
*root
,
6618 struct btrfs_ioctl_get_dev_stats
*stats
)
6620 struct btrfs_device
*dev
;
6621 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
6624 mutex_lock(&fs_devices
->device_list_mutex
);
6625 dev
= btrfs_find_device(root
->fs_info
, stats
->devid
, NULL
, NULL
);
6626 mutex_unlock(&fs_devices
->device_list_mutex
);
6629 btrfs_warn(root
->fs_info
, "get dev_stats failed, device not found");
6631 } else if (!dev
->dev_stats_valid
) {
6632 btrfs_warn(root
->fs_info
, "get dev_stats failed, not yet valid");
6634 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
6635 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6636 if (stats
->nr_items
> i
)
6638 btrfs_dev_stat_read_and_reset(dev
, i
);
6640 btrfs_dev_stat_reset(dev
, i
);
6643 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6644 if (stats
->nr_items
> i
)
6645 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
6647 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
6648 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
6652 int btrfs_scratch_superblock(struct btrfs_device
*device
)
6654 struct buffer_head
*bh
;
6655 struct btrfs_super_block
*disk_super
;
6657 bh
= btrfs_read_dev_super(device
->bdev
);
6660 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
6662 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
6663 set_buffer_dirty(bh
);
6664 sync_dirty_buffer(bh
);
6671 * Update the size of all devices, which is used for writing out the
6674 void btrfs_update_commit_device_size(struct btrfs_fs_info
*fs_info
)
6676 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6677 struct btrfs_device
*curr
, *next
;
6679 if (list_empty(&fs_devices
->resized_devices
))
6682 mutex_lock(&fs_devices
->device_list_mutex
);
6683 lock_chunks(fs_info
->dev_root
);
6684 list_for_each_entry_safe(curr
, next
, &fs_devices
->resized_devices
,
6686 list_del_init(&curr
->resized_list
);
6687 curr
->commit_total_bytes
= curr
->disk_total_bytes
;
6689 unlock_chunks(fs_info
->dev_root
);
6690 mutex_unlock(&fs_devices
->device_list_mutex
);
6693 /* Must be invoked during the transaction commit */
6694 void btrfs_update_commit_device_bytes_used(struct btrfs_root
*root
,
6695 struct btrfs_transaction
*transaction
)
6697 struct extent_map
*em
;
6698 struct map_lookup
*map
;
6699 struct btrfs_device
*dev
;
6702 if (list_empty(&transaction
->pending_chunks
))
6705 /* In order to kick the device replace finish process */
6707 list_for_each_entry(em
, &transaction
->pending_chunks
, list
) {
6708 map
= (struct map_lookup
*)em
->bdev
;
6710 for (i
= 0; i
< map
->num_stripes
; i
++) {
6711 dev
= map
->stripes
[i
].dev
;
6712 dev
->commit_bytes_used
= dev
->bytes_used
;
6715 unlock_chunks(root
);