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 void lock_chunks(struct btrfs_root
*root
)
58 mutex_lock(&root
->fs_info
->chunk_mutex
);
61 static void unlock_chunks(struct btrfs_root
*root
)
63 mutex_unlock(&root
->fs_info
->chunk_mutex
);
66 static struct btrfs_fs_devices
*__alloc_fs_devices(void)
68 struct btrfs_fs_devices
*fs_devs
;
70 fs_devs
= kzalloc(sizeof(*fs_devs
), GFP_NOFS
);
72 return ERR_PTR(-ENOMEM
);
74 mutex_init(&fs_devs
->device_list_mutex
);
76 INIT_LIST_HEAD(&fs_devs
->devices
);
77 INIT_LIST_HEAD(&fs_devs
->resized_devices
);
78 INIT_LIST_HEAD(&fs_devs
->alloc_list
);
79 INIT_LIST_HEAD(&fs_devs
->list
);
85 * alloc_fs_devices - allocate struct btrfs_fs_devices
86 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
89 * Return: a pointer to a new &struct btrfs_fs_devices on success;
90 * ERR_PTR() on error. Returned struct is not linked onto any lists and
91 * can be destroyed with kfree() right away.
93 static struct btrfs_fs_devices
*alloc_fs_devices(const u8
*fsid
)
95 struct btrfs_fs_devices
*fs_devs
;
97 fs_devs
= __alloc_fs_devices();
102 memcpy(fs_devs
->fsid
, fsid
, BTRFS_FSID_SIZE
);
104 generate_random_uuid(fs_devs
->fsid
);
109 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
111 struct btrfs_device
*device
;
112 WARN_ON(fs_devices
->opened
);
113 while (!list_empty(&fs_devices
->devices
)) {
114 device
= list_entry(fs_devices
->devices
.next
,
115 struct btrfs_device
, dev_list
);
116 list_del(&device
->dev_list
);
117 rcu_string_free(device
->name
);
123 static void btrfs_kobject_uevent(struct block_device
*bdev
,
124 enum kobject_action action
)
128 ret
= kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, action
);
130 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
132 kobject_name(&disk_to_dev(bdev
->bd_disk
)->kobj
),
133 &disk_to_dev(bdev
->bd_disk
)->kobj
);
136 void btrfs_cleanup_fs_uuids(void)
138 struct btrfs_fs_devices
*fs_devices
;
140 while (!list_empty(&fs_uuids
)) {
141 fs_devices
= list_entry(fs_uuids
.next
,
142 struct btrfs_fs_devices
, list
);
143 list_del(&fs_devices
->list
);
144 free_fs_devices(fs_devices
);
148 static struct btrfs_device
*__alloc_device(void)
150 struct btrfs_device
*dev
;
152 dev
= kzalloc(sizeof(*dev
), GFP_NOFS
);
154 return ERR_PTR(-ENOMEM
);
156 INIT_LIST_HEAD(&dev
->dev_list
);
157 INIT_LIST_HEAD(&dev
->dev_alloc_list
);
158 INIT_LIST_HEAD(&dev
->resized_list
);
160 spin_lock_init(&dev
->io_lock
);
162 spin_lock_init(&dev
->reada_lock
);
163 atomic_set(&dev
->reada_in_flight
, 0);
164 atomic_set(&dev
->dev_stats_ccnt
, 0);
165 INIT_RADIX_TREE(&dev
->reada_zones
, GFP_NOFS
& ~__GFP_WAIT
);
166 INIT_RADIX_TREE(&dev
->reada_extents
, GFP_NOFS
& ~__GFP_WAIT
);
171 static noinline
struct btrfs_device
*__find_device(struct list_head
*head
,
174 struct btrfs_device
*dev
;
176 list_for_each_entry(dev
, head
, dev_list
) {
177 if (dev
->devid
== devid
&&
178 (!uuid
|| !memcmp(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
))) {
185 static noinline
struct btrfs_fs_devices
*find_fsid(u8
*fsid
)
187 struct btrfs_fs_devices
*fs_devices
;
189 list_for_each_entry(fs_devices
, &fs_uuids
, list
) {
190 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
197 btrfs_get_bdev_and_sb(const char *device_path
, fmode_t flags
, void *holder
,
198 int flush
, struct block_device
**bdev
,
199 struct buffer_head
**bh
)
203 *bdev
= blkdev_get_by_path(device_path
, flags
, holder
);
206 ret
= PTR_ERR(*bdev
);
207 printk(KERN_INFO
"BTRFS: open %s failed\n", device_path
);
212 filemap_write_and_wait((*bdev
)->bd_inode
->i_mapping
);
213 ret
= set_blocksize(*bdev
, 4096);
215 blkdev_put(*bdev
, flags
);
218 invalidate_bdev(*bdev
);
219 *bh
= btrfs_read_dev_super(*bdev
);
222 blkdev_put(*bdev
, flags
);
234 static void requeue_list(struct btrfs_pending_bios
*pending_bios
,
235 struct bio
*head
, struct bio
*tail
)
238 struct bio
*old_head
;
240 old_head
= pending_bios
->head
;
241 pending_bios
->head
= head
;
242 if (pending_bios
->tail
)
243 tail
->bi_next
= old_head
;
245 pending_bios
->tail
= tail
;
249 * we try to collect pending bios for a device so we don't get a large
250 * number of procs sending bios down to the same device. This greatly
251 * improves the schedulers ability to collect and merge the bios.
253 * But, it also turns into a long list of bios to process and that is sure
254 * to eventually make the worker thread block. The solution here is to
255 * make some progress and then put this work struct back at the end of
256 * the list if the block device is congested. This way, multiple devices
257 * can make progress from a single worker thread.
259 static noinline
void run_scheduled_bios(struct btrfs_device
*device
)
262 struct backing_dev_info
*bdi
;
263 struct btrfs_fs_info
*fs_info
;
264 struct btrfs_pending_bios
*pending_bios
;
268 unsigned long num_run
;
269 unsigned long batch_run
= 0;
271 unsigned long last_waited
= 0;
273 int sync_pending
= 0;
274 struct blk_plug plug
;
277 * this function runs all the bios we've collected for
278 * a particular device. We don't want to wander off to
279 * another device without first sending all of these down.
280 * So, setup a plug here and finish it off before we return
282 blk_start_plug(&plug
);
284 bdi
= blk_get_backing_dev_info(device
->bdev
);
285 fs_info
= device
->dev_root
->fs_info
;
286 limit
= btrfs_async_submit_limit(fs_info
);
287 limit
= limit
* 2 / 3;
290 spin_lock(&device
->io_lock
);
295 /* take all the bios off the list at once and process them
296 * later on (without the lock held). But, remember the
297 * tail and other pointers so the bios can be properly reinserted
298 * into the list if we hit congestion
300 if (!force_reg
&& device
->pending_sync_bios
.head
) {
301 pending_bios
= &device
->pending_sync_bios
;
304 pending_bios
= &device
->pending_bios
;
308 pending
= pending_bios
->head
;
309 tail
= pending_bios
->tail
;
310 WARN_ON(pending
&& !tail
);
313 * if pending was null this time around, no bios need processing
314 * at all and we can stop. Otherwise it'll loop back up again
315 * and do an additional check so no bios are missed.
317 * device->running_pending is used to synchronize with the
320 if (device
->pending_sync_bios
.head
== NULL
&&
321 device
->pending_bios
.head
== NULL
) {
323 device
->running_pending
= 0;
326 device
->running_pending
= 1;
329 pending_bios
->head
= NULL
;
330 pending_bios
->tail
= NULL
;
332 spin_unlock(&device
->io_lock
);
337 /* we want to work on both lists, but do more bios on the
338 * sync list than the regular list
341 pending_bios
!= &device
->pending_sync_bios
&&
342 device
->pending_sync_bios
.head
) ||
343 (num_run
> 64 && pending_bios
== &device
->pending_sync_bios
&&
344 device
->pending_bios
.head
)) {
345 spin_lock(&device
->io_lock
);
346 requeue_list(pending_bios
, pending
, tail
);
351 pending
= pending
->bi_next
;
354 if (atomic_dec_return(&fs_info
->nr_async_bios
) < limit
&&
355 waitqueue_active(&fs_info
->async_submit_wait
))
356 wake_up(&fs_info
->async_submit_wait
);
358 BUG_ON(atomic_read(&cur
->bi_cnt
) == 0);
361 * if we're doing the sync list, record that our
362 * plug has some sync requests on it
364 * If we're doing the regular list and there are
365 * sync requests sitting around, unplug before
368 if (pending_bios
== &device
->pending_sync_bios
) {
370 } else if (sync_pending
) {
371 blk_finish_plug(&plug
);
372 blk_start_plug(&plug
);
376 btrfsic_submit_bio(cur
->bi_rw
, cur
);
383 * we made progress, there is more work to do and the bdi
384 * is now congested. Back off and let other work structs
387 if (pending
&& bdi_write_congested(bdi
) && batch_run
> 8 &&
388 fs_info
->fs_devices
->open_devices
> 1) {
389 struct io_context
*ioc
;
391 ioc
= current
->io_context
;
394 * the main goal here is that we don't want to
395 * block if we're going to be able to submit
396 * more requests without blocking.
398 * This code does two great things, it pokes into
399 * the elevator code from a filesystem _and_
400 * it makes assumptions about how batching works.
402 if (ioc
&& ioc
->nr_batch_requests
> 0 &&
403 time_before(jiffies
, ioc
->last_waited
+ HZ
/50UL) &&
405 ioc
->last_waited
== last_waited
)) {
407 * we want to go through our batch of
408 * requests and stop. So, we copy out
409 * the ioc->last_waited time and test
410 * against it before looping
412 last_waited
= ioc
->last_waited
;
417 spin_lock(&device
->io_lock
);
418 requeue_list(pending_bios
, pending
, tail
);
419 device
->running_pending
= 1;
421 spin_unlock(&device
->io_lock
);
422 btrfs_queue_work(fs_info
->submit_workers
,
426 /* unplug every 64 requests just for good measure */
427 if (batch_run
% 64 == 0) {
428 blk_finish_plug(&plug
);
429 blk_start_plug(&plug
);
438 spin_lock(&device
->io_lock
);
439 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
441 spin_unlock(&device
->io_lock
);
444 blk_finish_plug(&plug
);
447 static void pending_bios_fn(struct btrfs_work
*work
)
449 struct btrfs_device
*device
;
451 device
= container_of(work
, struct btrfs_device
, work
);
452 run_scheduled_bios(device
);
456 * Add new device to list of registered devices
459 * 1 - first time device is seen
460 * 0 - device already known
463 static noinline
int device_list_add(const char *path
,
464 struct btrfs_super_block
*disk_super
,
465 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
467 struct btrfs_device
*device
;
468 struct btrfs_fs_devices
*fs_devices
;
469 struct rcu_string
*name
;
471 u64 found_transid
= btrfs_super_generation(disk_super
);
473 fs_devices
= find_fsid(disk_super
->fsid
);
475 fs_devices
= alloc_fs_devices(disk_super
->fsid
);
476 if (IS_ERR(fs_devices
))
477 return PTR_ERR(fs_devices
);
479 list_add(&fs_devices
->list
, &fs_uuids
);
483 device
= __find_device(&fs_devices
->devices
, devid
,
484 disk_super
->dev_item
.uuid
);
488 if (fs_devices
->opened
)
491 device
= btrfs_alloc_device(NULL
, &devid
,
492 disk_super
->dev_item
.uuid
);
493 if (IS_ERR(device
)) {
494 /* we can safely leave the fs_devices entry around */
495 return PTR_ERR(device
);
498 name
= rcu_string_strdup(path
, GFP_NOFS
);
503 rcu_assign_pointer(device
->name
, name
);
505 mutex_lock(&fs_devices
->device_list_mutex
);
506 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
507 fs_devices
->num_devices
++;
508 mutex_unlock(&fs_devices
->device_list_mutex
);
511 device
->fs_devices
= fs_devices
;
512 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
514 * When FS is already mounted.
515 * 1. If you are here and if the device->name is NULL that
516 * means this device was missing at time of FS mount.
517 * 2. If you are here and if the device->name is different
518 * from 'path' that means either
519 * a. The same device disappeared and reappeared with
521 * b. The missing-disk-which-was-replaced, has
524 * We must allow 1 and 2a above. But 2b would be a spurious
527 * Further in case of 1 and 2a above, the disk at 'path'
528 * would have missed some transaction when it was away and
529 * in case of 2a the stale bdev has to be updated as well.
530 * 2b must not be allowed at all time.
534 * For now, we do allow update to btrfs_fs_device through the
535 * btrfs dev scan cli after FS has been mounted. We're still
536 * tracking a problem where systems fail mount by subvolume id
537 * when we reject replacement on a mounted FS.
539 if (!fs_devices
->opened
&& found_transid
< device
->generation
) {
541 * That is if the FS is _not_ mounted and if you
542 * are here, that means there is more than one
543 * disk with same uuid and devid.We keep the one
544 * with larger generation number or the last-in if
545 * generation are equal.
550 name
= rcu_string_strdup(path
, GFP_NOFS
);
553 rcu_string_free(device
->name
);
554 rcu_assign_pointer(device
->name
, name
);
555 if (device
->missing
) {
556 fs_devices
->missing_devices
--;
562 * Unmount does not free the btrfs_device struct but would zero
563 * generation along with most of the other members. So just update
564 * it back. We need it to pick the disk with largest generation
567 if (!fs_devices
->opened
)
568 device
->generation
= found_transid
;
570 *fs_devices_ret
= fs_devices
;
575 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
577 struct btrfs_fs_devices
*fs_devices
;
578 struct btrfs_device
*device
;
579 struct btrfs_device
*orig_dev
;
581 fs_devices
= alloc_fs_devices(orig
->fsid
);
582 if (IS_ERR(fs_devices
))
585 mutex_lock(&orig
->device_list_mutex
);
586 fs_devices
->total_devices
= orig
->total_devices
;
588 /* We have held the volume lock, it is safe to get the devices. */
589 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
590 struct rcu_string
*name
;
592 device
= btrfs_alloc_device(NULL
, &orig_dev
->devid
,
598 * This is ok to do without rcu read locked because we hold the
599 * uuid mutex so nothing we touch in here is going to disappear.
601 if (orig_dev
->name
) {
602 name
= rcu_string_strdup(orig_dev
->name
->str
, GFP_NOFS
);
607 rcu_assign_pointer(device
->name
, name
);
610 list_add(&device
->dev_list
, &fs_devices
->devices
);
611 device
->fs_devices
= fs_devices
;
612 fs_devices
->num_devices
++;
614 mutex_unlock(&orig
->device_list_mutex
);
617 mutex_unlock(&orig
->device_list_mutex
);
618 free_fs_devices(fs_devices
);
619 return ERR_PTR(-ENOMEM
);
622 void btrfs_close_extra_devices(struct btrfs_fs_info
*fs_info
,
623 struct btrfs_fs_devices
*fs_devices
, int step
)
625 struct btrfs_device
*device
, *next
;
626 struct btrfs_device
*latest_dev
= NULL
;
628 mutex_lock(&uuid_mutex
);
630 /* This is the initialized path, it is safe to release the devices. */
631 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
632 if (device
->in_fs_metadata
) {
633 if (!device
->is_tgtdev_for_dev_replace
&&
635 device
->generation
> latest_dev
->generation
)) {
641 if (device
->devid
== BTRFS_DEV_REPLACE_DEVID
) {
643 * In the first step, keep the device which has
644 * the correct fsid and the devid that is used
645 * for the dev_replace procedure.
646 * In the second step, the dev_replace state is
647 * read from the device tree and it is known
648 * whether the procedure is really active or
649 * not, which means whether this device is
650 * used or whether it should be removed.
652 if (step
== 0 || device
->is_tgtdev_for_dev_replace
) {
657 blkdev_put(device
->bdev
, device
->mode
);
659 fs_devices
->open_devices
--;
661 if (device
->writeable
) {
662 list_del_init(&device
->dev_alloc_list
);
663 device
->writeable
= 0;
664 if (!device
->is_tgtdev_for_dev_replace
)
665 fs_devices
->rw_devices
--;
667 list_del_init(&device
->dev_list
);
668 fs_devices
->num_devices
--;
669 rcu_string_free(device
->name
);
673 if (fs_devices
->seed
) {
674 fs_devices
= fs_devices
->seed
;
678 fs_devices
->latest_bdev
= latest_dev
->bdev
;
680 mutex_unlock(&uuid_mutex
);
683 static void __free_device(struct work_struct
*work
)
685 struct btrfs_device
*device
;
687 device
= container_of(work
, struct btrfs_device
, rcu_work
);
690 blkdev_put(device
->bdev
, device
->mode
);
692 rcu_string_free(device
->name
);
696 static void free_device(struct rcu_head
*head
)
698 struct btrfs_device
*device
;
700 device
= container_of(head
, struct btrfs_device
, rcu
);
702 INIT_WORK(&device
->rcu_work
, __free_device
);
703 schedule_work(&device
->rcu_work
);
706 static int __btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
708 struct btrfs_device
*device
;
710 if (--fs_devices
->opened
> 0)
713 mutex_lock(&fs_devices
->device_list_mutex
);
714 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
715 struct btrfs_device
*new_device
;
716 struct rcu_string
*name
;
719 fs_devices
->open_devices
--;
721 if (device
->writeable
&&
722 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
723 list_del_init(&device
->dev_alloc_list
);
724 fs_devices
->rw_devices
--;
728 fs_devices
->missing_devices
--;
730 new_device
= btrfs_alloc_device(NULL
, &device
->devid
,
732 BUG_ON(IS_ERR(new_device
)); /* -ENOMEM */
734 /* Safe because we are under uuid_mutex */
736 name
= rcu_string_strdup(device
->name
->str
, GFP_NOFS
);
737 BUG_ON(!name
); /* -ENOMEM */
738 rcu_assign_pointer(new_device
->name
, name
);
741 list_replace_rcu(&device
->dev_list
, &new_device
->dev_list
);
742 new_device
->fs_devices
= device
->fs_devices
;
744 call_rcu(&device
->rcu
, free_device
);
746 mutex_unlock(&fs_devices
->device_list_mutex
);
748 WARN_ON(fs_devices
->open_devices
);
749 WARN_ON(fs_devices
->rw_devices
);
750 fs_devices
->opened
= 0;
751 fs_devices
->seeding
= 0;
756 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
758 struct btrfs_fs_devices
*seed_devices
= NULL
;
761 mutex_lock(&uuid_mutex
);
762 ret
= __btrfs_close_devices(fs_devices
);
763 if (!fs_devices
->opened
) {
764 seed_devices
= fs_devices
->seed
;
765 fs_devices
->seed
= NULL
;
767 mutex_unlock(&uuid_mutex
);
769 while (seed_devices
) {
770 fs_devices
= seed_devices
;
771 seed_devices
= fs_devices
->seed
;
772 __btrfs_close_devices(fs_devices
);
773 free_fs_devices(fs_devices
);
776 * Wait for rcu kworkers under __btrfs_close_devices
777 * to finish all blkdev_puts so device is really
778 * free when umount is done.
784 static int __btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
785 fmode_t flags
, void *holder
)
787 struct request_queue
*q
;
788 struct block_device
*bdev
;
789 struct list_head
*head
= &fs_devices
->devices
;
790 struct btrfs_device
*device
;
791 struct btrfs_device
*latest_dev
= NULL
;
792 struct buffer_head
*bh
;
793 struct btrfs_super_block
*disk_super
;
800 list_for_each_entry(device
, head
, dev_list
) {
806 /* Just open everything we can; ignore failures here */
807 if (btrfs_get_bdev_and_sb(device
->name
->str
, flags
, holder
, 1,
811 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
812 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
813 if (devid
!= device
->devid
)
816 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
,
820 device
->generation
= btrfs_super_generation(disk_super
);
822 device
->generation
> latest_dev
->generation
)
825 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
826 device
->writeable
= 0;
828 device
->writeable
= !bdev_read_only(bdev
);
832 q
= bdev_get_queue(bdev
);
833 if (blk_queue_discard(q
))
834 device
->can_discard
= 1;
837 device
->in_fs_metadata
= 0;
838 device
->mode
= flags
;
840 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
841 fs_devices
->rotating
= 1;
843 fs_devices
->open_devices
++;
844 if (device
->writeable
&&
845 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
846 fs_devices
->rw_devices
++;
847 list_add(&device
->dev_alloc_list
,
848 &fs_devices
->alloc_list
);
855 blkdev_put(bdev
, flags
);
858 if (fs_devices
->open_devices
== 0) {
862 fs_devices
->seeding
= seeding
;
863 fs_devices
->opened
= 1;
864 fs_devices
->latest_bdev
= latest_dev
->bdev
;
865 fs_devices
->total_rw_bytes
= 0;
870 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
871 fmode_t flags
, void *holder
)
875 mutex_lock(&uuid_mutex
);
876 if (fs_devices
->opened
) {
877 fs_devices
->opened
++;
880 ret
= __btrfs_open_devices(fs_devices
, flags
, holder
);
882 mutex_unlock(&uuid_mutex
);
887 * Look for a btrfs signature on a device. This may be called out of the mount path
888 * and we are not allowed to call set_blocksize during the scan. The superblock
889 * is read via pagecache
891 int btrfs_scan_one_device(const char *path
, fmode_t flags
, void *holder
,
892 struct btrfs_fs_devices
**fs_devices_ret
)
894 struct btrfs_super_block
*disk_super
;
895 struct block_device
*bdev
;
906 * we would like to check all the supers, but that would make
907 * a btrfs mount succeed after a mkfs from a different FS.
908 * So, we need to add a special mount option to scan for
909 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
911 bytenr
= btrfs_sb_offset(0);
913 mutex_lock(&uuid_mutex
);
915 bdev
= blkdev_get_by_path(path
, flags
, holder
);
922 /* make sure our super fits in the device */
923 if (bytenr
+ PAGE_CACHE_SIZE
>= i_size_read(bdev
->bd_inode
))
926 /* make sure our super fits in the page */
927 if (sizeof(*disk_super
) > PAGE_CACHE_SIZE
)
930 /* make sure our super doesn't straddle pages on disk */
931 index
= bytenr
>> PAGE_CACHE_SHIFT
;
932 if ((bytenr
+ sizeof(*disk_super
) - 1) >> PAGE_CACHE_SHIFT
!= index
)
935 /* pull in the page with our super */
936 page
= read_cache_page_gfp(bdev
->bd_inode
->i_mapping
,
939 if (IS_ERR_OR_NULL(page
))
944 /* align our pointer to the offset of the super block */
945 disk_super
= p
+ (bytenr
& ~PAGE_CACHE_MASK
);
947 if (btrfs_super_bytenr(disk_super
) != bytenr
||
948 btrfs_super_magic(disk_super
) != BTRFS_MAGIC
)
951 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
952 transid
= btrfs_super_generation(disk_super
);
953 total_devices
= btrfs_super_num_devices(disk_super
);
955 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
957 if (disk_super
->label
[0]) {
958 if (disk_super
->label
[BTRFS_LABEL_SIZE
- 1])
959 disk_super
->label
[BTRFS_LABEL_SIZE
- 1] = '\0';
960 printk(KERN_INFO
"BTRFS: device label %s ", disk_super
->label
);
962 printk(KERN_INFO
"BTRFS: device fsid %pU ", disk_super
->fsid
);
965 printk(KERN_CONT
"devid %llu transid %llu %s\n", devid
, transid
, path
);
968 if (!ret
&& fs_devices_ret
)
969 (*fs_devices_ret
)->total_devices
= total_devices
;
973 page_cache_release(page
);
976 blkdev_put(bdev
, flags
);
978 mutex_unlock(&uuid_mutex
);
982 /* helper to account the used device space in the range */
983 int btrfs_account_dev_extents_size(struct btrfs_device
*device
, u64 start
,
984 u64 end
, u64
*length
)
986 struct btrfs_key key
;
987 struct btrfs_root
*root
= device
->dev_root
;
988 struct btrfs_dev_extent
*dev_extent
;
989 struct btrfs_path
*path
;
993 struct extent_buffer
*l
;
997 if (start
>= device
->total_bytes
|| device
->is_tgtdev_for_dev_replace
)
1000 path
= btrfs_alloc_path();
1005 key
.objectid
= device
->devid
;
1007 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1009 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1013 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1020 slot
= path
->slots
[0];
1021 if (slot
>= btrfs_header_nritems(l
)) {
1022 ret
= btrfs_next_leaf(root
, path
);
1030 btrfs_item_key_to_cpu(l
, &key
, slot
);
1032 if (key
.objectid
< device
->devid
)
1035 if (key
.objectid
> device
->devid
)
1038 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1041 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1042 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1044 if (key
.offset
<= start
&& extent_end
> end
) {
1045 *length
= end
- start
+ 1;
1047 } else if (key
.offset
<= start
&& extent_end
> start
)
1048 *length
+= extent_end
- start
;
1049 else if (key
.offset
> start
&& extent_end
<= end
)
1050 *length
+= extent_end
- key
.offset
;
1051 else if (key
.offset
> start
&& key
.offset
<= end
) {
1052 *length
+= end
- key
.offset
+ 1;
1054 } else if (key
.offset
> end
)
1062 btrfs_free_path(path
);
1066 static int contains_pending_extent(struct btrfs_trans_handle
*trans
,
1067 struct btrfs_device
*device
,
1068 u64
*start
, u64 len
)
1070 struct extent_map
*em
;
1073 list_for_each_entry(em
, &trans
->transaction
->pending_chunks
, list
) {
1074 struct map_lookup
*map
;
1077 map
= (struct map_lookup
*)em
->bdev
;
1078 for (i
= 0; i
< map
->num_stripes
; i
++) {
1079 if (map
->stripes
[i
].dev
!= device
)
1081 if (map
->stripes
[i
].physical
>= *start
+ len
||
1082 map
->stripes
[i
].physical
+ em
->orig_block_len
<=
1085 *start
= map
->stripes
[i
].physical
+
1096 * find_free_dev_extent - find free space in the specified device
1097 * @device: the device which we search the free space in
1098 * @num_bytes: the size of the free space that we need
1099 * @start: store the start of the free space.
1100 * @len: the size of the free space. that we find, or the size of the max
1101 * free space if we don't find suitable free space
1103 * this uses a pretty simple search, the expectation is that it is
1104 * called very infrequently and that a given device has a small number
1107 * @start is used to store the start of the free space if we find. But if we
1108 * don't find suitable free space, it will be used to store the start position
1109 * of the max free space.
1111 * @len is used to store the size of the free space that we find.
1112 * But if we don't find suitable free space, it is used to store the size of
1113 * the max free space.
1115 int find_free_dev_extent(struct btrfs_trans_handle
*trans
,
1116 struct btrfs_device
*device
, u64 num_bytes
,
1117 u64
*start
, u64
*len
)
1119 struct btrfs_key key
;
1120 struct btrfs_root
*root
= device
->dev_root
;
1121 struct btrfs_dev_extent
*dev_extent
;
1122 struct btrfs_path
*path
;
1128 u64 search_end
= device
->total_bytes
;
1131 struct extent_buffer
*l
;
1133 /* FIXME use last free of some kind */
1135 /* we don't want to overwrite the superblock on the drive,
1136 * so we make sure to start at an offset of at least 1MB
1138 search_start
= max(root
->fs_info
->alloc_start
, 1024ull * 1024);
1140 path
= btrfs_alloc_path();
1144 max_hole_start
= search_start
;
1148 if (search_start
>= search_end
|| device
->is_tgtdev_for_dev_replace
) {
1154 path
->search_commit_root
= 1;
1155 path
->skip_locking
= 1;
1157 key
.objectid
= device
->devid
;
1158 key
.offset
= search_start
;
1159 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1161 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1165 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1172 slot
= path
->slots
[0];
1173 if (slot
>= btrfs_header_nritems(l
)) {
1174 ret
= btrfs_next_leaf(root
, path
);
1182 btrfs_item_key_to_cpu(l
, &key
, slot
);
1184 if (key
.objectid
< device
->devid
)
1187 if (key
.objectid
> device
->devid
)
1190 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1193 if (key
.offset
> search_start
) {
1194 hole_size
= key
.offset
- search_start
;
1197 * Have to check before we set max_hole_start, otherwise
1198 * we could end up sending back this offset anyway.
1200 if (contains_pending_extent(trans
, device
,
1205 if (hole_size
> max_hole_size
) {
1206 max_hole_start
= search_start
;
1207 max_hole_size
= hole_size
;
1211 * If this free space is greater than which we need,
1212 * it must be the max free space that we have found
1213 * until now, so max_hole_start must point to the start
1214 * of this free space and the length of this free space
1215 * is stored in max_hole_size. Thus, we return
1216 * max_hole_start and max_hole_size and go back to the
1219 if (hole_size
>= num_bytes
) {
1225 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1226 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1228 if (extent_end
> search_start
)
1229 search_start
= extent_end
;
1236 * At this point, search_start should be the end of
1237 * allocated dev extents, and when shrinking the device,
1238 * search_end may be smaller than search_start.
1240 if (search_end
> search_start
)
1241 hole_size
= search_end
- search_start
;
1243 if (hole_size
> max_hole_size
) {
1244 max_hole_start
= search_start
;
1245 max_hole_size
= hole_size
;
1248 if (contains_pending_extent(trans
, device
, &search_start
, hole_size
)) {
1249 btrfs_release_path(path
);
1254 if (hole_size
< num_bytes
)
1260 btrfs_free_path(path
);
1261 *start
= max_hole_start
;
1263 *len
= max_hole_size
;
1267 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1268 struct btrfs_device
*device
,
1269 u64 start
, u64
*dev_extent_len
)
1272 struct btrfs_path
*path
;
1273 struct btrfs_root
*root
= device
->dev_root
;
1274 struct btrfs_key key
;
1275 struct btrfs_key found_key
;
1276 struct extent_buffer
*leaf
= NULL
;
1277 struct btrfs_dev_extent
*extent
= NULL
;
1279 path
= btrfs_alloc_path();
1283 key
.objectid
= device
->devid
;
1285 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1287 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1289 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1290 BTRFS_DEV_EXTENT_KEY
);
1293 leaf
= path
->nodes
[0];
1294 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1295 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1296 struct btrfs_dev_extent
);
1297 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1298 btrfs_dev_extent_length(leaf
, extent
) < start
);
1300 btrfs_release_path(path
);
1302 } else if (ret
== 0) {
1303 leaf
= path
->nodes
[0];
1304 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1305 struct btrfs_dev_extent
);
1307 btrfs_error(root
->fs_info
, ret
, "Slot search failed");
1311 *dev_extent_len
= btrfs_dev_extent_length(leaf
, extent
);
1313 ret
= btrfs_del_item(trans
, root
, path
);
1315 btrfs_error(root
->fs_info
, ret
,
1316 "Failed to remove dev extent item");
1319 btrfs_free_path(path
);
1323 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1324 struct btrfs_device
*device
,
1325 u64 chunk_tree
, u64 chunk_objectid
,
1326 u64 chunk_offset
, u64 start
, u64 num_bytes
)
1329 struct btrfs_path
*path
;
1330 struct btrfs_root
*root
= device
->dev_root
;
1331 struct btrfs_dev_extent
*extent
;
1332 struct extent_buffer
*leaf
;
1333 struct btrfs_key key
;
1335 WARN_ON(!device
->in_fs_metadata
);
1336 WARN_ON(device
->is_tgtdev_for_dev_replace
);
1337 path
= btrfs_alloc_path();
1341 key
.objectid
= device
->devid
;
1343 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1344 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1349 leaf
= path
->nodes
[0];
1350 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1351 struct btrfs_dev_extent
);
1352 btrfs_set_dev_extent_chunk_tree(leaf
, extent
, chunk_tree
);
1353 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
, chunk_objectid
);
1354 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1356 write_extent_buffer(leaf
, root
->fs_info
->chunk_tree_uuid
,
1357 btrfs_dev_extent_chunk_tree_uuid(extent
), BTRFS_UUID_SIZE
);
1359 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1360 btrfs_mark_buffer_dirty(leaf
);
1362 btrfs_free_path(path
);
1366 static u64
find_next_chunk(struct btrfs_fs_info
*fs_info
)
1368 struct extent_map_tree
*em_tree
;
1369 struct extent_map
*em
;
1373 em_tree
= &fs_info
->mapping_tree
.map_tree
;
1374 read_lock(&em_tree
->lock
);
1375 n
= rb_last(&em_tree
->map
);
1377 em
= rb_entry(n
, struct extent_map
, rb_node
);
1378 ret
= em
->start
+ em
->len
;
1380 read_unlock(&em_tree
->lock
);
1385 static noinline
int find_next_devid(struct btrfs_fs_info
*fs_info
,
1389 struct btrfs_key key
;
1390 struct btrfs_key found_key
;
1391 struct btrfs_path
*path
;
1393 path
= btrfs_alloc_path();
1397 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1398 key
.type
= BTRFS_DEV_ITEM_KEY
;
1399 key
.offset
= (u64
)-1;
1401 ret
= btrfs_search_slot(NULL
, fs_info
->chunk_root
, &key
, path
, 0, 0);
1405 BUG_ON(ret
== 0); /* Corruption */
1407 ret
= btrfs_previous_item(fs_info
->chunk_root
, path
,
1408 BTRFS_DEV_ITEMS_OBJECTID
,
1409 BTRFS_DEV_ITEM_KEY
);
1413 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1415 *devid_ret
= found_key
.offset
+ 1;
1419 btrfs_free_path(path
);
1424 * the device information is stored in the chunk root
1425 * the btrfs_device struct should be fully filled in
1427 static int btrfs_add_device(struct btrfs_trans_handle
*trans
,
1428 struct btrfs_root
*root
,
1429 struct btrfs_device
*device
)
1432 struct btrfs_path
*path
;
1433 struct btrfs_dev_item
*dev_item
;
1434 struct extent_buffer
*leaf
;
1435 struct btrfs_key key
;
1438 root
= root
->fs_info
->chunk_root
;
1440 path
= btrfs_alloc_path();
1444 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1445 key
.type
= BTRFS_DEV_ITEM_KEY
;
1446 key
.offset
= device
->devid
;
1448 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1453 leaf
= path
->nodes
[0];
1454 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1456 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1457 btrfs_set_device_generation(leaf
, dev_item
, 0);
1458 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1459 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1460 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1461 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1462 btrfs_set_device_total_bytes(leaf
, dev_item
,
1463 btrfs_device_get_disk_total_bytes(device
));
1464 btrfs_set_device_bytes_used(leaf
, dev_item
,
1465 btrfs_device_get_bytes_used(device
));
1466 btrfs_set_device_group(leaf
, dev_item
, 0);
1467 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1468 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1469 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1471 ptr
= btrfs_device_uuid(dev_item
);
1472 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1473 ptr
= btrfs_device_fsid(dev_item
);
1474 write_extent_buffer(leaf
, root
->fs_info
->fsid
, ptr
, BTRFS_UUID_SIZE
);
1475 btrfs_mark_buffer_dirty(leaf
);
1479 btrfs_free_path(path
);
1484 * Function to update ctime/mtime for a given device path.
1485 * Mainly used for ctime/mtime based probe like libblkid.
1487 static void update_dev_time(char *path_name
)
1491 filp
= filp_open(path_name
, O_RDWR
, 0);
1494 file_update_time(filp
);
1495 filp_close(filp
, NULL
);
1499 static int btrfs_rm_dev_item(struct btrfs_root
*root
,
1500 struct btrfs_device
*device
)
1503 struct btrfs_path
*path
;
1504 struct btrfs_key key
;
1505 struct btrfs_trans_handle
*trans
;
1507 root
= root
->fs_info
->chunk_root
;
1509 path
= btrfs_alloc_path();
1513 trans
= btrfs_start_transaction(root
, 0);
1514 if (IS_ERR(trans
)) {
1515 btrfs_free_path(path
);
1516 return PTR_ERR(trans
);
1518 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1519 key
.type
= BTRFS_DEV_ITEM_KEY
;
1520 key
.offset
= device
->devid
;
1522 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1531 ret
= btrfs_del_item(trans
, root
, path
);
1535 btrfs_free_path(path
);
1536 btrfs_commit_transaction(trans
, root
);
1540 int btrfs_rm_device(struct btrfs_root
*root
, char *device_path
)
1542 struct btrfs_device
*device
;
1543 struct btrfs_device
*next_device
;
1544 struct block_device
*bdev
;
1545 struct buffer_head
*bh
= NULL
;
1546 struct btrfs_super_block
*disk_super
;
1547 struct btrfs_fs_devices
*cur_devices
;
1554 bool clear_super
= false;
1556 mutex_lock(&uuid_mutex
);
1559 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
1561 all_avail
= root
->fs_info
->avail_data_alloc_bits
|
1562 root
->fs_info
->avail_system_alloc_bits
|
1563 root
->fs_info
->avail_metadata_alloc_bits
;
1564 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
1566 num_devices
= root
->fs_info
->fs_devices
->num_devices
;
1567 btrfs_dev_replace_lock(&root
->fs_info
->dev_replace
);
1568 if (btrfs_dev_replace_is_ongoing(&root
->fs_info
->dev_replace
)) {
1569 WARN_ON(num_devices
< 1);
1572 btrfs_dev_replace_unlock(&root
->fs_info
->dev_replace
);
1574 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID10
) && num_devices
<= 4) {
1575 ret
= BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET
;
1579 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID1
) && num_devices
<= 2) {
1580 ret
= BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET
;
1584 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID5
) &&
1585 root
->fs_info
->fs_devices
->rw_devices
<= 2) {
1586 ret
= BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET
;
1589 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID6
) &&
1590 root
->fs_info
->fs_devices
->rw_devices
<= 3) {
1591 ret
= BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET
;
1595 if (strcmp(device_path
, "missing") == 0) {
1596 struct list_head
*devices
;
1597 struct btrfs_device
*tmp
;
1600 devices
= &root
->fs_info
->fs_devices
->devices
;
1602 * It is safe to read the devices since the volume_mutex
1605 list_for_each_entry(tmp
, devices
, dev_list
) {
1606 if (tmp
->in_fs_metadata
&&
1607 !tmp
->is_tgtdev_for_dev_replace
&&
1617 ret
= BTRFS_ERROR_DEV_MISSING_NOT_FOUND
;
1621 ret
= btrfs_get_bdev_and_sb(device_path
,
1622 FMODE_WRITE
| FMODE_EXCL
,
1623 root
->fs_info
->bdev_holder
, 0,
1627 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1628 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1629 dev_uuid
= disk_super
->dev_item
.uuid
;
1630 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1638 if (device
->is_tgtdev_for_dev_replace
) {
1639 ret
= BTRFS_ERROR_DEV_TGT_REPLACE
;
1643 if (device
->writeable
&& root
->fs_info
->fs_devices
->rw_devices
== 1) {
1644 ret
= BTRFS_ERROR_DEV_ONLY_WRITABLE
;
1648 if (device
->writeable
) {
1650 list_del_init(&device
->dev_alloc_list
);
1651 device
->fs_devices
->rw_devices
--;
1652 unlock_chunks(root
);
1656 mutex_unlock(&uuid_mutex
);
1657 ret
= btrfs_shrink_device(device
, 0);
1658 mutex_lock(&uuid_mutex
);
1663 * TODO: the superblock still includes this device in its num_devices
1664 * counter although write_all_supers() is not locked out. This
1665 * could give a filesystem state which requires a degraded mount.
1667 ret
= btrfs_rm_dev_item(root
->fs_info
->chunk_root
, device
);
1671 device
->in_fs_metadata
= 0;
1672 btrfs_scrub_cancel_dev(root
->fs_info
, device
);
1675 * the device list mutex makes sure that we don't change
1676 * the device list while someone else is writing out all
1677 * the device supers. Whoever is writing all supers, should
1678 * lock the device list mutex before getting the number of
1679 * devices in the super block (super_copy). Conversely,
1680 * whoever updates the number of devices in the super block
1681 * (super_copy) should hold the device list mutex.
1684 cur_devices
= device
->fs_devices
;
1685 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1686 list_del_rcu(&device
->dev_list
);
1688 device
->fs_devices
->num_devices
--;
1689 device
->fs_devices
->total_devices
--;
1691 if (device
->missing
)
1692 device
->fs_devices
->missing_devices
--;
1694 next_device
= list_entry(root
->fs_info
->fs_devices
->devices
.next
,
1695 struct btrfs_device
, dev_list
);
1696 if (device
->bdev
== root
->fs_info
->sb
->s_bdev
)
1697 root
->fs_info
->sb
->s_bdev
= next_device
->bdev
;
1698 if (device
->bdev
== root
->fs_info
->fs_devices
->latest_bdev
)
1699 root
->fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1702 device
->fs_devices
->open_devices
--;
1703 /* remove sysfs entry */
1704 btrfs_kobj_rm_device(root
->fs_info
, device
);
1707 call_rcu(&device
->rcu
, free_device
);
1709 num_devices
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
1710 btrfs_set_super_num_devices(root
->fs_info
->super_copy
, num_devices
);
1711 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1713 if (cur_devices
->open_devices
== 0) {
1714 struct btrfs_fs_devices
*fs_devices
;
1715 fs_devices
= root
->fs_info
->fs_devices
;
1716 while (fs_devices
) {
1717 if (fs_devices
->seed
== cur_devices
) {
1718 fs_devices
->seed
= cur_devices
->seed
;
1721 fs_devices
= fs_devices
->seed
;
1723 cur_devices
->seed
= NULL
;
1724 __btrfs_close_devices(cur_devices
);
1725 free_fs_devices(cur_devices
);
1728 root
->fs_info
->num_tolerated_disk_barrier_failures
=
1729 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
1732 * at this point, the device is zero sized. We want to
1733 * remove it from the devices list and zero out the old super
1735 if (clear_super
&& disk_super
) {
1739 /* make sure this device isn't detected as part of
1742 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
1743 set_buffer_dirty(bh
);
1744 sync_dirty_buffer(bh
);
1746 /* clear the mirror copies of super block on the disk
1747 * being removed, 0th copy is been taken care above and
1748 * the below would take of the rest
1750 for (i
= 1; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
1751 bytenr
= btrfs_sb_offset(i
);
1752 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
1753 i_size_read(bdev
->bd_inode
))
1757 bh
= __bread(bdev
, bytenr
/ 4096,
1758 BTRFS_SUPER_INFO_SIZE
);
1762 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1764 if (btrfs_super_bytenr(disk_super
) != bytenr
||
1765 btrfs_super_magic(disk_super
) != BTRFS_MAGIC
) {
1768 memset(&disk_super
->magic
, 0,
1769 sizeof(disk_super
->magic
));
1770 set_buffer_dirty(bh
);
1771 sync_dirty_buffer(bh
);
1778 /* Notify udev that device has changed */
1779 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
1781 /* Update ctime/mtime for device path for libblkid */
1782 update_dev_time(device_path
);
1788 blkdev_put(bdev
, FMODE_READ
| FMODE_EXCL
);
1790 mutex_unlock(&uuid_mutex
);
1793 if (device
->writeable
) {
1795 list_add(&device
->dev_alloc_list
,
1796 &root
->fs_info
->fs_devices
->alloc_list
);
1797 device
->fs_devices
->rw_devices
++;
1798 unlock_chunks(root
);
1803 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info
*fs_info
,
1804 struct btrfs_device
*srcdev
)
1806 struct btrfs_fs_devices
*fs_devices
;
1808 WARN_ON(!mutex_is_locked(&fs_info
->fs_devices
->device_list_mutex
));
1811 * in case of fs with no seed, srcdev->fs_devices will point
1812 * to fs_devices of fs_info. However when the dev being replaced is
1813 * a seed dev it will point to the seed's local fs_devices. In short
1814 * srcdev will have its correct fs_devices in both the cases.
1816 fs_devices
= srcdev
->fs_devices
;
1818 list_del_rcu(&srcdev
->dev_list
);
1819 list_del_rcu(&srcdev
->dev_alloc_list
);
1820 fs_devices
->num_devices
--;
1821 if (srcdev
->missing
)
1822 fs_devices
->missing_devices
--;
1824 if (srcdev
->writeable
) {
1825 fs_devices
->rw_devices
--;
1826 /* zero out the old super if it is writable */
1827 btrfs_scratch_superblock(srcdev
);
1831 fs_devices
->open_devices
--;
1833 call_rcu(&srcdev
->rcu
, free_device
);
1836 * unless fs_devices is seed fs, num_devices shouldn't go
1839 BUG_ON(!fs_devices
->num_devices
&& !fs_devices
->seeding
);
1841 /* if this is no devs we rather delete the fs_devices */
1842 if (!fs_devices
->num_devices
) {
1843 struct btrfs_fs_devices
*tmp_fs_devices
;
1845 tmp_fs_devices
= fs_info
->fs_devices
;
1846 while (tmp_fs_devices
) {
1847 if (tmp_fs_devices
->seed
== fs_devices
) {
1848 tmp_fs_devices
->seed
= fs_devices
->seed
;
1851 tmp_fs_devices
= tmp_fs_devices
->seed
;
1853 fs_devices
->seed
= NULL
;
1854 __btrfs_close_devices(fs_devices
);
1855 free_fs_devices(fs_devices
);
1859 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
1860 struct btrfs_device
*tgtdev
)
1862 struct btrfs_device
*next_device
;
1864 mutex_lock(&uuid_mutex
);
1866 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
1868 btrfs_scratch_superblock(tgtdev
);
1869 fs_info
->fs_devices
->open_devices
--;
1871 fs_info
->fs_devices
->num_devices
--;
1873 next_device
= list_entry(fs_info
->fs_devices
->devices
.next
,
1874 struct btrfs_device
, dev_list
);
1875 if (tgtdev
->bdev
== fs_info
->sb
->s_bdev
)
1876 fs_info
->sb
->s_bdev
= next_device
->bdev
;
1877 if (tgtdev
->bdev
== fs_info
->fs_devices
->latest_bdev
)
1878 fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1879 list_del_rcu(&tgtdev
->dev_list
);
1881 call_rcu(&tgtdev
->rcu
, free_device
);
1883 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
1884 mutex_unlock(&uuid_mutex
);
1887 static int btrfs_find_device_by_path(struct btrfs_root
*root
, char *device_path
,
1888 struct btrfs_device
**device
)
1891 struct btrfs_super_block
*disk_super
;
1894 struct block_device
*bdev
;
1895 struct buffer_head
*bh
;
1898 ret
= btrfs_get_bdev_and_sb(device_path
, FMODE_READ
,
1899 root
->fs_info
->bdev_holder
, 0, &bdev
, &bh
);
1902 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1903 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1904 dev_uuid
= disk_super
->dev_item
.uuid
;
1905 *device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1910 blkdev_put(bdev
, FMODE_READ
);
1914 int btrfs_find_device_missing_or_by_path(struct btrfs_root
*root
,
1916 struct btrfs_device
**device
)
1919 if (strcmp(device_path
, "missing") == 0) {
1920 struct list_head
*devices
;
1921 struct btrfs_device
*tmp
;
1923 devices
= &root
->fs_info
->fs_devices
->devices
;
1925 * It is safe to read the devices since the volume_mutex
1926 * is held by the caller.
1928 list_for_each_entry(tmp
, devices
, dev_list
) {
1929 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
1936 btrfs_err(root
->fs_info
, "no missing device found");
1942 return btrfs_find_device_by_path(root
, device_path
, device
);
1947 * does all the dirty work required for changing file system's UUID.
1949 static int btrfs_prepare_sprout(struct btrfs_root
*root
)
1951 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
1952 struct btrfs_fs_devices
*old_devices
;
1953 struct btrfs_fs_devices
*seed_devices
;
1954 struct btrfs_super_block
*disk_super
= root
->fs_info
->super_copy
;
1955 struct btrfs_device
*device
;
1958 BUG_ON(!mutex_is_locked(&uuid_mutex
));
1959 if (!fs_devices
->seeding
)
1962 seed_devices
= __alloc_fs_devices();
1963 if (IS_ERR(seed_devices
))
1964 return PTR_ERR(seed_devices
);
1966 old_devices
= clone_fs_devices(fs_devices
);
1967 if (IS_ERR(old_devices
)) {
1968 kfree(seed_devices
);
1969 return PTR_ERR(old_devices
);
1972 list_add(&old_devices
->list
, &fs_uuids
);
1974 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
1975 seed_devices
->opened
= 1;
1976 INIT_LIST_HEAD(&seed_devices
->devices
);
1977 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
1978 mutex_init(&seed_devices
->device_list_mutex
);
1980 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1981 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
1983 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
)
1984 device
->fs_devices
= seed_devices
;
1987 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
1988 unlock_chunks(root
);
1990 fs_devices
->seeding
= 0;
1991 fs_devices
->num_devices
= 0;
1992 fs_devices
->open_devices
= 0;
1993 fs_devices
->missing_devices
= 0;
1994 fs_devices
->rotating
= 0;
1995 fs_devices
->seed
= seed_devices
;
1997 generate_random_uuid(fs_devices
->fsid
);
1998 memcpy(root
->fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
1999 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2000 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2002 super_flags
= btrfs_super_flags(disk_super
) &
2003 ~BTRFS_SUPER_FLAG_SEEDING
;
2004 btrfs_set_super_flags(disk_super
, super_flags
);
2010 * strore the expected generation for seed devices in device items.
2012 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
2013 struct btrfs_root
*root
)
2015 struct btrfs_path
*path
;
2016 struct extent_buffer
*leaf
;
2017 struct btrfs_dev_item
*dev_item
;
2018 struct btrfs_device
*device
;
2019 struct btrfs_key key
;
2020 u8 fs_uuid
[BTRFS_UUID_SIZE
];
2021 u8 dev_uuid
[BTRFS_UUID_SIZE
];
2025 path
= btrfs_alloc_path();
2029 root
= root
->fs_info
->chunk_root
;
2030 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2032 key
.type
= BTRFS_DEV_ITEM_KEY
;
2035 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2039 leaf
= path
->nodes
[0];
2041 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
2042 ret
= btrfs_next_leaf(root
, path
);
2047 leaf
= path
->nodes
[0];
2048 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2049 btrfs_release_path(path
);
2053 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2054 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
2055 key
.type
!= BTRFS_DEV_ITEM_KEY
)
2058 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2059 struct btrfs_dev_item
);
2060 devid
= btrfs_device_id(leaf
, dev_item
);
2061 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
2063 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
2065 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
2067 BUG_ON(!device
); /* Logic error */
2069 if (device
->fs_devices
->seeding
) {
2070 btrfs_set_device_generation(leaf
, dev_item
,
2071 device
->generation
);
2072 btrfs_mark_buffer_dirty(leaf
);
2080 btrfs_free_path(path
);
2084 int btrfs_init_new_device(struct btrfs_root
*root
, char *device_path
)
2086 struct request_queue
*q
;
2087 struct btrfs_trans_handle
*trans
;
2088 struct btrfs_device
*device
;
2089 struct block_device
*bdev
;
2090 struct list_head
*devices
;
2091 struct super_block
*sb
= root
->fs_info
->sb
;
2092 struct rcu_string
*name
;
2094 int seeding_dev
= 0;
2097 if ((sb
->s_flags
& MS_RDONLY
) && !root
->fs_info
->fs_devices
->seeding
)
2100 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2101 root
->fs_info
->bdev_holder
);
2103 return PTR_ERR(bdev
);
2105 if (root
->fs_info
->fs_devices
->seeding
) {
2107 down_write(&sb
->s_umount
);
2108 mutex_lock(&uuid_mutex
);
2111 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2113 devices
= &root
->fs_info
->fs_devices
->devices
;
2115 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2116 list_for_each_entry(device
, devices
, dev_list
) {
2117 if (device
->bdev
== bdev
) {
2120 &root
->fs_info
->fs_devices
->device_list_mutex
);
2124 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2126 device
= btrfs_alloc_device(root
->fs_info
, NULL
, NULL
);
2127 if (IS_ERR(device
)) {
2128 /* we can safely leave the fs_devices entry around */
2129 ret
= PTR_ERR(device
);
2133 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2139 rcu_assign_pointer(device
->name
, name
);
2141 trans
= btrfs_start_transaction(root
, 0);
2142 if (IS_ERR(trans
)) {
2143 rcu_string_free(device
->name
);
2145 ret
= PTR_ERR(trans
);
2149 q
= bdev_get_queue(bdev
);
2150 if (blk_queue_discard(q
))
2151 device
->can_discard
= 1;
2152 device
->writeable
= 1;
2153 device
->generation
= trans
->transid
;
2154 device
->io_width
= root
->sectorsize
;
2155 device
->io_align
= root
->sectorsize
;
2156 device
->sector_size
= root
->sectorsize
;
2157 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
2158 device
->disk_total_bytes
= device
->total_bytes
;
2159 device
->commit_total_bytes
= device
->total_bytes
;
2160 device
->dev_root
= root
->fs_info
->dev_root
;
2161 device
->bdev
= bdev
;
2162 device
->in_fs_metadata
= 1;
2163 device
->is_tgtdev_for_dev_replace
= 0;
2164 device
->mode
= FMODE_EXCL
;
2165 device
->dev_stats_valid
= 1;
2166 set_blocksize(device
->bdev
, 4096);
2169 sb
->s_flags
&= ~MS_RDONLY
;
2170 ret
= btrfs_prepare_sprout(root
);
2171 BUG_ON(ret
); /* -ENOMEM */
2174 device
->fs_devices
= root
->fs_info
->fs_devices
;
2176 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2178 list_add_rcu(&device
->dev_list
, &root
->fs_info
->fs_devices
->devices
);
2179 list_add(&device
->dev_alloc_list
,
2180 &root
->fs_info
->fs_devices
->alloc_list
);
2181 root
->fs_info
->fs_devices
->num_devices
++;
2182 root
->fs_info
->fs_devices
->open_devices
++;
2183 root
->fs_info
->fs_devices
->rw_devices
++;
2184 root
->fs_info
->fs_devices
->total_devices
++;
2185 root
->fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
2187 spin_lock(&root
->fs_info
->free_chunk_lock
);
2188 root
->fs_info
->free_chunk_space
+= device
->total_bytes
;
2189 spin_unlock(&root
->fs_info
->free_chunk_lock
);
2191 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
2192 root
->fs_info
->fs_devices
->rotating
= 1;
2194 tmp
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
2195 btrfs_set_super_total_bytes(root
->fs_info
->super_copy
,
2196 tmp
+ device
->total_bytes
);
2198 tmp
= btrfs_super_num_devices(root
->fs_info
->super_copy
);
2199 btrfs_set_super_num_devices(root
->fs_info
->super_copy
,
2202 /* add sysfs device entry */
2203 btrfs_kobj_add_device(root
->fs_info
, device
);
2206 * we've got more storage, clear any full flags on the space
2209 btrfs_clear_space_info_full(root
->fs_info
);
2211 unlock_chunks(root
);
2212 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2216 ret
= init_first_rw_device(trans
, root
, device
);
2217 unlock_chunks(root
);
2219 btrfs_abort_transaction(trans
, root
, ret
);
2224 ret
= btrfs_add_device(trans
, root
, device
);
2226 btrfs_abort_transaction(trans
, root
, ret
);
2231 char fsid_buf
[BTRFS_UUID_UNPARSED_SIZE
];
2233 ret
= btrfs_finish_sprout(trans
, root
);
2235 btrfs_abort_transaction(trans
, root
, ret
);
2239 /* Sprouting would change fsid of the mounted root,
2240 * so rename the fsid on the sysfs
2242 snprintf(fsid_buf
, BTRFS_UUID_UNPARSED_SIZE
, "%pU",
2243 root
->fs_info
->fsid
);
2244 if (kobject_rename(&root
->fs_info
->super_kobj
, fsid_buf
))
2248 root
->fs_info
->num_tolerated_disk_barrier_failures
=
2249 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
2250 ret
= btrfs_commit_transaction(trans
, root
);
2253 mutex_unlock(&uuid_mutex
);
2254 up_write(&sb
->s_umount
);
2256 if (ret
) /* transaction commit */
2259 ret
= btrfs_relocate_sys_chunks(root
);
2261 btrfs_error(root
->fs_info
, ret
,
2262 "Failed to relocate sys chunks after "
2263 "device initialization. This can be fixed "
2264 "using the \"btrfs balance\" command.");
2265 trans
= btrfs_attach_transaction(root
);
2266 if (IS_ERR(trans
)) {
2267 if (PTR_ERR(trans
) == -ENOENT
)
2269 return PTR_ERR(trans
);
2271 ret
= btrfs_commit_transaction(trans
, root
);
2274 /* Update ctime/mtime for libblkid */
2275 update_dev_time(device_path
);
2279 btrfs_end_transaction(trans
, root
);
2280 rcu_string_free(device
->name
);
2281 btrfs_kobj_rm_device(root
->fs_info
, device
);
2284 blkdev_put(bdev
, FMODE_EXCL
);
2286 mutex_unlock(&uuid_mutex
);
2287 up_write(&sb
->s_umount
);
2292 int btrfs_init_dev_replace_tgtdev(struct btrfs_root
*root
, char *device_path
,
2293 struct btrfs_device
*srcdev
,
2294 struct btrfs_device
**device_out
)
2296 struct request_queue
*q
;
2297 struct btrfs_device
*device
;
2298 struct block_device
*bdev
;
2299 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2300 struct list_head
*devices
;
2301 struct rcu_string
*name
;
2302 u64 devid
= BTRFS_DEV_REPLACE_DEVID
;
2306 if (fs_info
->fs_devices
->seeding
) {
2307 btrfs_err(fs_info
, "the filesystem is a seed filesystem!");
2311 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2312 fs_info
->bdev_holder
);
2314 btrfs_err(fs_info
, "target device %s is invalid!", device_path
);
2315 return PTR_ERR(bdev
);
2318 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2320 devices
= &fs_info
->fs_devices
->devices
;
2321 list_for_each_entry(device
, devices
, dev_list
) {
2322 if (device
->bdev
== bdev
) {
2323 btrfs_err(fs_info
, "target device is in the filesystem!");
2330 if (i_size_read(bdev
->bd_inode
) <
2331 btrfs_device_get_total_bytes(srcdev
)) {
2332 btrfs_err(fs_info
, "target device is smaller than source device!");
2338 device
= btrfs_alloc_device(NULL
, &devid
, NULL
);
2339 if (IS_ERR(device
)) {
2340 ret
= PTR_ERR(device
);
2344 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2350 rcu_assign_pointer(device
->name
, name
);
2352 q
= bdev_get_queue(bdev
);
2353 if (blk_queue_discard(q
))
2354 device
->can_discard
= 1;
2355 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2356 device
->writeable
= 1;
2357 device
->generation
= 0;
2358 device
->io_width
= root
->sectorsize
;
2359 device
->io_align
= root
->sectorsize
;
2360 device
->sector_size
= root
->sectorsize
;
2361 device
->total_bytes
= btrfs_device_get_total_bytes(srcdev
);
2362 device
->disk_total_bytes
= btrfs_device_get_disk_total_bytes(srcdev
);
2363 device
->bytes_used
= btrfs_device_get_bytes_used(srcdev
);
2364 ASSERT(list_empty(&srcdev
->resized_list
));
2365 device
->commit_total_bytes
= srcdev
->commit_total_bytes
;
2366 device
->commit_bytes_used
= device
->bytes_used
;
2367 device
->dev_root
= fs_info
->dev_root
;
2368 device
->bdev
= bdev
;
2369 device
->in_fs_metadata
= 1;
2370 device
->is_tgtdev_for_dev_replace
= 1;
2371 device
->mode
= FMODE_EXCL
;
2372 device
->dev_stats_valid
= 1;
2373 set_blocksize(device
->bdev
, 4096);
2374 device
->fs_devices
= fs_info
->fs_devices
;
2375 list_add(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2376 fs_info
->fs_devices
->num_devices
++;
2377 fs_info
->fs_devices
->open_devices
++;
2378 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2380 *device_out
= device
;
2384 blkdev_put(bdev
, FMODE_EXCL
);
2388 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info
*fs_info
,
2389 struct btrfs_device
*tgtdev
)
2391 WARN_ON(fs_info
->fs_devices
->rw_devices
== 0);
2392 tgtdev
->io_width
= fs_info
->dev_root
->sectorsize
;
2393 tgtdev
->io_align
= fs_info
->dev_root
->sectorsize
;
2394 tgtdev
->sector_size
= fs_info
->dev_root
->sectorsize
;
2395 tgtdev
->dev_root
= fs_info
->dev_root
;
2396 tgtdev
->in_fs_metadata
= 1;
2399 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2400 struct btrfs_device
*device
)
2403 struct btrfs_path
*path
;
2404 struct btrfs_root
*root
;
2405 struct btrfs_dev_item
*dev_item
;
2406 struct extent_buffer
*leaf
;
2407 struct btrfs_key key
;
2409 root
= device
->dev_root
->fs_info
->chunk_root
;
2411 path
= btrfs_alloc_path();
2415 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2416 key
.type
= BTRFS_DEV_ITEM_KEY
;
2417 key
.offset
= device
->devid
;
2419 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2428 leaf
= path
->nodes
[0];
2429 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2431 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2432 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2433 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2434 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2435 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2436 btrfs_set_device_total_bytes(leaf
, dev_item
,
2437 btrfs_device_get_disk_total_bytes(device
));
2438 btrfs_set_device_bytes_used(leaf
, dev_item
,
2439 btrfs_device_get_bytes_used(device
));
2440 btrfs_mark_buffer_dirty(leaf
);
2443 btrfs_free_path(path
);
2447 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2448 struct btrfs_device
*device
, u64 new_size
)
2450 struct btrfs_super_block
*super_copy
=
2451 device
->dev_root
->fs_info
->super_copy
;
2452 struct btrfs_fs_devices
*fs_devices
;
2456 if (!device
->writeable
)
2459 lock_chunks(device
->dev_root
);
2460 old_total
= btrfs_super_total_bytes(super_copy
);
2461 diff
= new_size
- device
->total_bytes
;
2463 if (new_size
<= device
->total_bytes
||
2464 device
->is_tgtdev_for_dev_replace
) {
2465 unlock_chunks(device
->dev_root
);
2469 fs_devices
= device
->dev_root
->fs_info
->fs_devices
;
2471 btrfs_set_super_total_bytes(super_copy
, old_total
+ diff
);
2472 device
->fs_devices
->total_rw_bytes
+= diff
;
2474 btrfs_device_set_total_bytes(device
, new_size
);
2475 btrfs_device_set_disk_total_bytes(device
, new_size
);
2476 btrfs_clear_space_info_full(device
->dev_root
->fs_info
);
2477 if (list_empty(&device
->resized_list
))
2478 list_add_tail(&device
->resized_list
,
2479 &fs_devices
->resized_devices
);
2480 unlock_chunks(device
->dev_root
);
2482 return btrfs_update_device(trans
, device
);
2485 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
2486 struct btrfs_root
*root
,
2487 u64 chunk_tree
, u64 chunk_objectid
,
2491 struct btrfs_path
*path
;
2492 struct btrfs_key key
;
2494 root
= root
->fs_info
->chunk_root
;
2495 path
= btrfs_alloc_path();
2499 key
.objectid
= chunk_objectid
;
2500 key
.offset
= chunk_offset
;
2501 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2503 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2506 else if (ret
> 0) { /* Logic error or corruption */
2507 btrfs_error(root
->fs_info
, -ENOENT
,
2508 "Failed lookup while freeing chunk.");
2513 ret
= btrfs_del_item(trans
, root
, path
);
2515 btrfs_error(root
->fs_info
, ret
,
2516 "Failed to delete chunk item.");
2518 btrfs_free_path(path
);
2522 static int btrfs_del_sys_chunk(struct btrfs_root
*root
, u64 chunk_objectid
, u64
2525 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
2526 struct btrfs_disk_key
*disk_key
;
2527 struct btrfs_chunk
*chunk
;
2534 struct btrfs_key key
;
2537 array_size
= btrfs_super_sys_array_size(super_copy
);
2539 ptr
= super_copy
->sys_chunk_array
;
2542 while (cur
< array_size
) {
2543 disk_key
= (struct btrfs_disk_key
*)ptr
;
2544 btrfs_disk_key_to_cpu(&key
, disk_key
);
2546 len
= sizeof(*disk_key
);
2548 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2549 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2550 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2551 len
+= btrfs_chunk_item_size(num_stripes
);
2556 if (key
.objectid
== chunk_objectid
&&
2557 key
.offset
== chunk_offset
) {
2558 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2560 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2566 unlock_chunks(root
);
2570 int btrfs_remove_chunk(struct btrfs_trans_handle
*trans
,
2571 struct btrfs_root
*root
, u64 chunk_offset
)
2573 struct extent_map_tree
*em_tree
;
2574 struct extent_map
*em
;
2575 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2576 struct map_lookup
*map
;
2577 u64 dev_extent_len
= 0;
2578 u64 chunk_objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2579 u64 chunk_tree
= root
->fs_info
->chunk_root
->objectid
;
2583 root
= root
->fs_info
->chunk_root
;
2584 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
2586 read_lock(&em_tree
->lock
);
2587 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
2588 read_unlock(&em_tree
->lock
);
2590 if (!em
|| em
->start
> chunk_offset
||
2591 em
->start
+ em
->len
< chunk_offset
) {
2593 * This is a logic error, but we don't want to just rely on the
2594 * user having built with ASSERT enabled, so if ASSERT doens't
2595 * do anything we still error out.
2599 free_extent_map(em
);
2602 map
= (struct map_lookup
*)em
->bdev
;
2604 for (i
= 0; i
< map
->num_stripes
; i
++) {
2605 struct btrfs_device
*device
= map
->stripes
[i
].dev
;
2606 ret
= btrfs_free_dev_extent(trans
, device
,
2607 map
->stripes
[i
].physical
,
2610 btrfs_abort_transaction(trans
, root
, ret
);
2614 if (device
->bytes_used
> 0) {
2616 btrfs_device_set_bytes_used(device
,
2617 device
->bytes_used
- dev_extent_len
);
2618 spin_lock(&root
->fs_info
->free_chunk_lock
);
2619 root
->fs_info
->free_chunk_space
+= dev_extent_len
;
2620 spin_unlock(&root
->fs_info
->free_chunk_lock
);
2621 btrfs_clear_space_info_full(root
->fs_info
);
2622 unlock_chunks(root
);
2625 if (map
->stripes
[i
].dev
) {
2626 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2628 btrfs_abort_transaction(trans
, root
, ret
);
2633 ret
= btrfs_free_chunk(trans
, root
, chunk_tree
, chunk_objectid
,
2636 btrfs_abort_transaction(trans
, root
, ret
);
2640 trace_btrfs_chunk_free(root
, map
, chunk_offset
, em
->len
);
2642 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2643 ret
= btrfs_del_sys_chunk(root
, chunk_objectid
, chunk_offset
);
2645 btrfs_abort_transaction(trans
, root
, ret
);
2650 ret
= btrfs_remove_block_group(trans
, extent_root
, chunk_offset
);
2652 btrfs_abort_transaction(trans
, extent_root
, ret
);
2656 write_lock(&em_tree
->lock
);
2657 remove_extent_mapping(em_tree
, em
);
2658 write_unlock(&em_tree
->lock
);
2660 /* once for the tree */
2661 free_extent_map(em
);
2664 free_extent_map(em
);
2668 static int btrfs_relocate_chunk(struct btrfs_root
*root
,
2669 u64 chunk_tree
, u64 chunk_objectid
,
2672 struct btrfs_root
*extent_root
;
2673 struct btrfs_trans_handle
*trans
;
2676 root
= root
->fs_info
->chunk_root
;
2677 extent_root
= root
->fs_info
->extent_root
;
2679 ret
= btrfs_can_relocate(extent_root
, chunk_offset
);
2683 /* step one, relocate all the extents inside this chunk */
2684 ret
= btrfs_relocate_block_group(extent_root
, chunk_offset
);
2688 trans
= btrfs_start_transaction(root
, 0);
2689 if (IS_ERR(trans
)) {
2690 ret
= PTR_ERR(trans
);
2691 btrfs_std_error(root
->fs_info
, ret
);
2696 * step two, delete the device extents and the
2697 * chunk tree entries
2699 ret
= btrfs_remove_chunk(trans
, root
, chunk_offset
);
2700 btrfs_end_transaction(trans
, root
);
2704 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
)
2706 struct btrfs_root
*chunk_root
= root
->fs_info
->chunk_root
;
2707 struct btrfs_path
*path
;
2708 struct extent_buffer
*leaf
;
2709 struct btrfs_chunk
*chunk
;
2710 struct btrfs_key key
;
2711 struct btrfs_key found_key
;
2712 u64 chunk_tree
= chunk_root
->root_key
.objectid
;
2714 bool retried
= false;
2718 path
= btrfs_alloc_path();
2723 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2724 key
.offset
= (u64
)-1;
2725 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2728 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2731 BUG_ON(ret
== 0); /* Corruption */
2733 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
2740 leaf
= path
->nodes
[0];
2741 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2743 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
2744 struct btrfs_chunk
);
2745 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2746 btrfs_release_path(path
);
2748 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2749 ret
= btrfs_relocate_chunk(chunk_root
, chunk_tree
,
2758 if (found_key
.offset
== 0)
2760 key
.offset
= found_key
.offset
- 1;
2763 if (failed
&& !retried
) {
2767 } else if (WARN_ON(failed
&& retried
)) {
2771 btrfs_free_path(path
);
2775 static int insert_balance_item(struct btrfs_root
*root
,
2776 struct btrfs_balance_control
*bctl
)
2778 struct btrfs_trans_handle
*trans
;
2779 struct btrfs_balance_item
*item
;
2780 struct btrfs_disk_balance_args disk_bargs
;
2781 struct btrfs_path
*path
;
2782 struct extent_buffer
*leaf
;
2783 struct btrfs_key key
;
2786 path
= btrfs_alloc_path();
2790 trans
= btrfs_start_transaction(root
, 0);
2791 if (IS_ERR(trans
)) {
2792 btrfs_free_path(path
);
2793 return PTR_ERR(trans
);
2796 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2797 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2800 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2805 leaf
= path
->nodes
[0];
2806 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
2808 memset_extent_buffer(leaf
, 0, (unsigned long)item
, sizeof(*item
));
2810 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
2811 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
2812 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
2813 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
2814 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
2815 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
2817 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
2819 btrfs_mark_buffer_dirty(leaf
);
2821 btrfs_free_path(path
);
2822 err
= btrfs_commit_transaction(trans
, root
);
2828 static int del_balance_item(struct btrfs_root
*root
)
2830 struct btrfs_trans_handle
*trans
;
2831 struct btrfs_path
*path
;
2832 struct btrfs_key key
;
2835 path
= btrfs_alloc_path();
2839 trans
= btrfs_start_transaction(root
, 0);
2840 if (IS_ERR(trans
)) {
2841 btrfs_free_path(path
);
2842 return PTR_ERR(trans
);
2845 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2846 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2849 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2857 ret
= btrfs_del_item(trans
, root
, path
);
2859 btrfs_free_path(path
);
2860 err
= btrfs_commit_transaction(trans
, root
);
2867 * This is a heuristic used to reduce the number of chunks balanced on
2868 * resume after balance was interrupted.
2870 static void update_balance_args(struct btrfs_balance_control
*bctl
)
2873 * Turn on soft mode for chunk types that were being converted.
2875 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2876 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2877 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2878 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2879 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2880 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2883 * Turn on usage filter if is not already used. The idea is
2884 * that chunks that we have already balanced should be
2885 * reasonably full. Don't do it for chunks that are being
2886 * converted - that will keep us from relocating unconverted
2887 * (albeit full) chunks.
2889 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2890 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2891 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2892 bctl
->data
.usage
= 90;
2894 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2895 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2896 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2897 bctl
->sys
.usage
= 90;
2899 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2900 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2901 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2902 bctl
->meta
.usage
= 90;
2907 * Should be called with both balance and volume mutexes held to
2908 * serialize other volume operations (add_dev/rm_dev/resize) with
2909 * restriper. Same goes for unset_balance_control.
2911 static void set_balance_control(struct btrfs_balance_control
*bctl
)
2913 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
2915 BUG_ON(fs_info
->balance_ctl
);
2917 spin_lock(&fs_info
->balance_lock
);
2918 fs_info
->balance_ctl
= bctl
;
2919 spin_unlock(&fs_info
->balance_lock
);
2922 static void unset_balance_control(struct btrfs_fs_info
*fs_info
)
2924 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
2926 BUG_ON(!fs_info
->balance_ctl
);
2928 spin_lock(&fs_info
->balance_lock
);
2929 fs_info
->balance_ctl
= NULL
;
2930 spin_unlock(&fs_info
->balance_lock
);
2936 * Balance filters. Return 1 if chunk should be filtered out
2937 * (should not be balanced).
2939 static int chunk_profiles_filter(u64 chunk_type
,
2940 struct btrfs_balance_args
*bargs
)
2942 chunk_type
= chunk_to_extended(chunk_type
) &
2943 BTRFS_EXTENDED_PROFILE_MASK
;
2945 if (bargs
->profiles
& chunk_type
)
2951 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
2952 struct btrfs_balance_args
*bargs
)
2954 struct btrfs_block_group_cache
*cache
;
2955 u64 chunk_used
, user_thresh
;
2958 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
2959 chunk_used
= btrfs_block_group_used(&cache
->item
);
2961 if (bargs
->usage
== 0)
2963 else if (bargs
->usage
> 100)
2964 user_thresh
= cache
->key
.offset
;
2966 user_thresh
= div_factor_fine(cache
->key
.offset
,
2969 if (chunk_used
< user_thresh
)
2972 btrfs_put_block_group(cache
);
2976 static int chunk_devid_filter(struct extent_buffer
*leaf
,
2977 struct btrfs_chunk
*chunk
,
2978 struct btrfs_balance_args
*bargs
)
2980 struct btrfs_stripe
*stripe
;
2981 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2984 for (i
= 0; i
< num_stripes
; i
++) {
2985 stripe
= btrfs_stripe_nr(chunk
, i
);
2986 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
2993 /* [pstart, pend) */
2994 static int chunk_drange_filter(struct extent_buffer
*leaf
,
2995 struct btrfs_chunk
*chunk
,
2997 struct btrfs_balance_args
*bargs
)
2999 struct btrfs_stripe
*stripe
;
3000 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3006 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
3009 if (btrfs_chunk_type(leaf
, chunk
) & (BTRFS_BLOCK_GROUP_DUP
|
3010 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)) {
3011 factor
= num_stripes
/ 2;
3012 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID5
) {
3013 factor
= num_stripes
- 1;
3014 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID6
) {
3015 factor
= num_stripes
- 2;
3017 factor
= num_stripes
;
3020 for (i
= 0; i
< num_stripes
; i
++) {
3021 stripe
= btrfs_stripe_nr(chunk
, i
);
3022 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
3025 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
3026 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
3027 do_div(stripe_length
, factor
);
3029 if (stripe_offset
< bargs
->pend
&&
3030 stripe_offset
+ stripe_length
> bargs
->pstart
)
3037 /* [vstart, vend) */
3038 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
3039 struct btrfs_chunk
*chunk
,
3041 struct btrfs_balance_args
*bargs
)
3043 if (chunk_offset
< bargs
->vend
&&
3044 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
3045 /* at least part of the chunk is inside this vrange */
3051 static int chunk_soft_convert_filter(u64 chunk_type
,
3052 struct btrfs_balance_args
*bargs
)
3054 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
3057 chunk_type
= chunk_to_extended(chunk_type
) &
3058 BTRFS_EXTENDED_PROFILE_MASK
;
3060 if (bargs
->target
== chunk_type
)
3066 static int should_balance_chunk(struct btrfs_root
*root
,
3067 struct extent_buffer
*leaf
,
3068 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
3070 struct btrfs_balance_control
*bctl
= root
->fs_info
->balance_ctl
;
3071 struct btrfs_balance_args
*bargs
= NULL
;
3072 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3075 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
3076 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
3080 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3081 bargs
= &bctl
->data
;
3082 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3084 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3085 bargs
= &bctl
->meta
;
3087 /* profiles filter */
3088 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
3089 chunk_profiles_filter(chunk_type
, bargs
)) {
3094 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3095 chunk_usage_filter(bctl
->fs_info
, chunk_offset
, bargs
)) {
3100 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
3101 chunk_devid_filter(leaf
, chunk
, bargs
)) {
3105 /* drange filter, makes sense only with devid filter */
3106 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
3107 chunk_drange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3112 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
3113 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3117 /* soft profile changing mode */
3118 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
3119 chunk_soft_convert_filter(chunk_type
, bargs
)) {
3124 * limited by count, must be the last filter
3126 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT
)) {
3127 if (bargs
->limit
== 0)
3136 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
3138 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3139 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
3140 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
3141 struct list_head
*devices
;
3142 struct btrfs_device
*device
;
3145 struct btrfs_chunk
*chunk
;
3146 struct btrfs_path
*path
;
3147 struct btrfs_key key
;
3148 struct btrfs_key found_key
;
3149 struct btrfs_trans_handle
*trans
;
3150 struct extent_buffer
*leaf
;
3153 int enospc_errors
= 0;
3154 bool counting
= true;
3155 u64 limit_data
= bctl
->data
.limit
;
3156 u64 limit_meta
= bctl
->meta
.limit
;
3157 u64 limit_sys
= bctl
->sys
.limit
;
3159 /* step one make some room on all the devices */
3160 devices
= &fs_info
->fs_devices
->devices
;
3161 list_for_each_entry(device
, devices
, dev_list
) {
3162 old_size
= btrfs_device_get_total_bytes(device
);
3163 size_to_free
= div_factor(old_size
, 1);
3164 size_to_free
= min(size_to_free
, (u64
)1 * 1024 * 1024);
3165 if (!device
->writeable
||
3166 btrfs_device_get_total_bytes(device
) -
3167 btrfs_device_get_bytes_used(device
) > size_to_free
||
3168 device
->is_tgtdev_for_dev_replace
)
3171 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
3176 trans
= btrfs_start_transaction(dev_root
, 0);
3177 BUG_ON(IS_ERR(trans
));
3179 ret
= btrfs_grow_device(trans
, device
, old_size
);
3182 btrfs_end_transaction(trans
, dev_root
);
3185 /* step two, relocate all the chunks */
3186 path
= btrfs_alloc_path();
3192 /* zero out stat counters */
3193 spin_lock(&fs_info
->balance_lock
);
3194 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
3195 spin_unlock(&fs_info
->balance_lock
);
3198 bctl
->data
.limit
= limit_data
;
3199 bctl
->meta
.limit
= limit_meta
;
3200 bctl
->sys
.limit
= limit_sys
;
3202 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3203 key
.offset
= (u64
)-1;
3204 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3207 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
3208 atomic_read(&fs_info
->balance_cancel_req
)) {
3213 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
3218 * this shouldn't happen, it means the last relocate
3222 BUG(); /* FIXME break ? */
3224 ret
= btrfs_previous_item(chunk_root
, path
, 0,
3225 BTRFS_CHUNK_ITEM_KEY
);
3231 leaf
= path
->nodes
[0];
3232 slot
= path
->slots
[0];
3233 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3235 if (found_key
.objectid
!= key
.objectid
)
3238 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
3241 spin_lock(&fs_info
->balance_lock
);
3242 bctl
->stat
.considered
++;
3243 spin_unlock(&fs_info
->balance_lock
);
3246 ret
= should_balance_chunk(chunk_root
, leaf
, chunk
,
3248 btrfs_release_path(path
);
3253 spin_lock(&fs_info
->balance_lock
);
3254 bctl
->stat
.expected
++;
3255 spin_unlock(&fs_info
->balance_lock
);
3259 ret
= btrfs_relocate_chunk(chunk_root
,
3260 chunk_root
->root_key
.objectid
,
3263 if (ret
&& ret
!= -ENOSPC
)
3265 if (ret
== -ENOSPC
) {
3268 spin_lock(&fs_info
->balance_lock
);
3269 bctl
->stat
.completed
++;
3270 spin_unlock(&fs_info
->balance_lock
);
3273 if (found_key
.offset
== 0)
3275 key
.offset
= found_key
.offset
- 1;
3279 btrfs_release_path(path
);
3284 btrfs_free_path(path
);
3285 if (enospc_errors
) {
3286 btrfs_info(fs_info
, "%d enospc errors during balance",
3296 * alloc_profile_is_valid - see if a given profile is valid and reduced
3297 * @flags: profile to validate
3298 * @extended: if true @flags is treated as an extended profile
3300 static int alloc_profile_is_valid(u64 flags
, int extended
)
3302 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
3303 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
3305 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
3307 /* 1) check that all other bits are zeroed */
3311 /* 2) see if profile is reduced */
3313 return !extended
; /* "0" is valid for usual profiles */
3315 /* true if exactly one bit set */
3316 return (flags
& (flags
- 1)) == 0;
3319 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
3321 /* cancel requested || normal exit path */
3322 return atomic_read(&fs_info
->balance_cancel_req
) ||
3323 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
3324 atomic_read(&fs_info
->balance_cancel_req
) == 0);
3327 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
3331 unset_balance_control(fs_info
);
3332 ret
= del_balance_item(fs_info
->tree_root
);
3334 btrfs_std_error(fs_info
, ret
);
3336 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3340 * Should be called with both balance and volume mutexes held
3342 int btrfs_balance(struct btrfs_balance_control
*bctl
,
3343 struct btrfs_ioctl_balance_args
*bargs
)
3345 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3352 if (btrfs_fs_closing(fs_info
) ||
3353 atomic_read(&fs_info
->balance_pause_req
) ||
3354 atomic_read(&fs_info
->balance_cancel_req
)) {
3359 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
3360 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
3364 * In case of mixed groups both data and meta should be picked,
3365 * and identical options should be given for both of them.
3367 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
3368 if (mixed
&& (bctl
->flags
& allowed
)) {
3369 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
3370 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
3371 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
3372 btrfs_err(fs_info
, "with mixed groups data and "
3373 "metadata balance options must be the same");
3379 num_devices
= fs_info
->fs_devices
->num_devices
;
3380 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
3381 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
3382 BUG_ON(num_devices
< 1);
3385 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
3386 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
3387 if (num_devices
== 1)
3388 allowed
|= BTRFS_BLOCK_GROUP_DUP
;
3389 else if (num_devices
> 1)
3390 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
3391 if (num_devices
> 2)
3392 allowed
|= BTRFS_BLOCK_GROUP_RAID5
;
3393 if (num_devices
> 3)
3394 allowed
|= (BTRFS_BLOCK_GROUP_RAID10
|
3395 BTRFS_BLOCK_GROUP_RAID6
);
3396 if ((bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3397 (!alloc_profile_is_valid(bctl
->data
.target
, 1) ||
3398 (bctl
->data
.target
& ~allowed
))) {
3399 btrfs_err(fs_info
, "unable to start balance with target "
3400 "data profile %llu",
3405 if ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3406 (!alloc_profile_is_valid(bctl
->meta
.target
, 1) ||
3407 (bctl
->meta
.target
& ~allowed
))) {
3409 "unable to start balance with target metadata profile %llu",
3414 if ((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3415 (!alloc_profile_is_valid(bctl
->sys
.target
, 1) ||
3416 (bctl
->sys
.target
& ~allowed
))) {
3418 "unable to start balance with target system profile %llu",
3424 /* allow dup'ed data chunks only in mixed mode */
3425 if (!mixed
&& (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3426 (bctl
->data
.target
& BTRFS_BLOCK_GROUP_DUP
)) {
3427 btrfs_err(fs_info
, "dup for data is not allowed");
3432 /* allow to reduce meta or sys integrity only if force set */
3433 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3434 BTRFS_BLOCK_GROUP_RAID10
|
3435 BTRFS_BLOCK_GROUP_RAID5
|
3436 BTRFS_BLOCK_GROUP_RAID6
;
3438 seq
= read_seqbegin(&fs_info
->profiles_lock
);
3440 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3441 (fs_info
->avail_system_alloc_bits
& allowed
) &&
3442 !(bctl
->sys
.target
& allowed
)) ||
3443 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3444 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
3445 !(bctl
->meta
.target
& allowed
))) {
3446 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
3447 btrfs_info(fs_info
, "force reducing metadata integrity");
3449 btrfs_err(fs_info
, "balance will reduce metadata "
3450 "integrity, use force if you want this");
3455 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
3457 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3458 int num_tolerated_disk_barrier_failures
;
3459 u64 target
= bctl
->sys
.target
;
3461 num_tolerated_disk_barrier_failures
=
3462 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3463 if (num_tolerated_disk_barrier_failures
> 0 &&
3465 (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID0
|
3466 BTRFS_AVAIL_ALLOC_BIT_SINGLE
)))
3467 num_tolerated_disk_barrier_failures
= 0;
3468 else if (num_tolerated_disk_barrier_failures
> 1 &&
3470 (BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)))
3471 num_tolerated_disk_barrier_failures
= 1;
3473 fs_info
->num_tolerated_disk_barrier_failures
=
3474 num_tolerated_disk_barrier_failures
;
3477 ret
= insert_balance_item(fs_info
->tree_root
, bctl
);
3478 if (ret
&& ret
!= -EEXIST
)
3481 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
3482 BUG_ON(ret
== -EEXIST
);
3483 set_balance_control(bctl
);
3485 BUG_ON(ret
!= -EEXIST
);
3486 spin_lock(&fs_info
->balance_lock
);
3487 update_balance_args(bctl
);
3488 spin_unlock(&fs_info
->balance_lock
);
3491 atomic_inc(&fs_info
->balance_running
);
3492 mutex_unlock(&fs_info
->balance_mutex
);
3494 ret
= __btrfs_balance(fs_info
);
3496 mutex_lock(&fs_info
->balance_mutex
);
3497 atomic_dec(&fs_info
->balance_running
);
3499 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3500 fs_info
->num_tolerated_disk_barrier_failures
=
3501 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3505 memset(bargs
, 0, sizeof(*bargs
));
3506 update_ioctl_balance_args(fs_info
, 0, bargs
);
3509 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
3510 balance_need_close(fs_info
)) {
3511 __cancel_balance(fs_info
);
3514 wake_up(&fs_info
->balance_wait_q
);
3518 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
3519 __cancel_balance(fs_info
);
3522 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3527 static int balance_kthread(void *data
)
3529 struct btrfs_fs_info
*fs_info
= data
;
3532 mutex_lock(&fs_info
->volume_mutex
);
3533 mutex_lock(&fs_info
->balance_mutex
);
3535 if (fs_info
->balance_ctl
) {
3536 btrfs_info(fs_info
, "continuing balance");
3537 ret
= btrfs_balance(fs_info
->balance_ctl
, NULL
);
3540 mutex_unlock(&fs_info
->balance_mutex
);
3541 mutex_unlock(&fs_info
->volume_mutex
);
3546 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
3548 struct task_struct
*tsk
;
3550 spin_lock(&fs_info
->balance_lock
);
3551 if (!fs_info
->balance_ctl
) {
3552 spin_unlock(&fs_info
->balance_lock
);
3555 spin_unlock(&fs_info
->balance_lock
);
3557 if (btrfs_test_opt(fs_info
->tree_root
, SKIP_BALANCE
)) {
3558 btrfs_info(fs_info
, "force skipping balance");
3562 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
3563 return PTR_ERR_OR_ZERO(tsk
);
3566 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
3568 struct btrfs_balance_control
*bctl
;
3569 struct btrfs_balance_item
*item
;
3570 struct btrfs_disk_balance_args disk_bargs
;
3571 struct btrfs_path
*path
;
3572 struct extent_buffer
*leaf
;
3573 struct btrfs_key key
;
3576 path
= btrfs_alloc_path();
3580 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3581 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
3584 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
3587 if (ret
> 0) { /* ret = -ENOENT; */
3592 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
3598 leaf
= path
->nodes
[0];
3599 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3601 bctl
->fs_info
= fs_info
;
3602 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
3603 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
3605 btrfs_balance_data(leaf
, item
, &disk_bargs
);
3606 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
3607 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
3608 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
3609 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
3610 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
3612 WARN_ON(atomic_xchg(&fs_info
->mutually_exclusive_operation_running
, 1));
3614 mutex_lock(&fs_info
->volume_mutex
);
3615 mutex_lock(&fs_info
->balance_mutex
);
3617 set_balance_control(bctl
);
3619 mutex_unlock(&fs_info
->balance_mutex
);
3620 mutex_unlock(&fs_info
->volume_mutex
);
3622 btrfs_free_path(path
);
3626 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
3630 mutex_lock(&fs_info
->balance_mutex
);
3631 if (!fs_info
->balance_ctl
) {
3632 mutex_unlock(&fs_info
->balance_mutex
);
3636 if (atomic_read(&fs_info
->balance_running
)) {
3637 atomic_inc(&fs_info
->balance_pause_req
);
3638 mutex_unlock(&fs_info
->balance_mutex
);
3640 wait_event(fs_info
->balance_wait_q
,
3641 atomic_read(&fs_info
->balance_running
) == 0);
3643 mutex_lock(&fs_info
->balance_mutex
);
3644 /* we are good with balance_ctl ripped off from under us */
3645 BUG_ON(atomic_read(&fs_info
->balance_running
));
3646 atomic_dec(&fs_info
->balance_pause_req
);
3651 mutex_unlock(&fs_info
->balance_mutex
);
3655 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
3657 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
3660 mutex_lock(&fs_info
->balance_mutex
);
3661 if (!fs_info
->balance_ctl
) {
3662 mutex_unlock(&fs_info
->balance_mutex
);
3666 atomic_inc(&fs_info
->balance_cancel_req
);
3668 * if we are running just wait and return, balance item is
3669 * deleted in btrfs_balance in this case
3671 if (atomic_read(&fs_info
->balance_running
)) {
3672 mutex_unlock(&fs_info
->balance_mutex
);
3673 wait_event(fs_info
->balance_wait_q
,
3674 atomic_read(&fs_info
->balance_running
) == 0);
3675 mutex_lock(&fs_info
->balance_mutex
);
3677 /* __cancel_balance needs volume_mutex */
3678 mutex_unlock(&fs_info
->balance_mutex
);
3679 mutex_lock(&fs_info
->volume_mutex
);
3680 mutex_lock(&fs_info
->balance_mutex
);
3682 if (fs_info
->balance_ctl
)
3683 __cancel_balance(fs_info
);
3685 mutex_unlock(&fs_info
->volume_mutex
);
3688 BUG_ON(fs_info
->balance_ctl
|| atomic_read(&fs_info
->balance_running
));
3689 atomic_dec(&fs_info
->balance_cancel_req
);
3690 mutex_unlock(&fs_info
->balance_mutex
);
3694 static int btrfs_uuid_scan_kthread(void *data
)
3696 struct btrfs_fs_info
*fs_info
= data
;
3697 struct btrfs_root
*root
= fs_info
->tree_root
;
3698 struct btrfs_key key
;
3699 struct btrfs_key max_key
;
3700 struct btrfs_path
*path
= NULL
;
3702 struct extent_buffer
*eb
;
3704 struct btrfs_root_item root_item
;
3706 struct btrfs_trans_handle
*trans
= NULL
;
3708 path
= btrfs_alloc_path();
3715 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3718 max_key
.objectid
= (u64
)-1;
3719 max_key
.type
= BTRFS_ROOT_ITEM_KEY
;
3720 max_key
.offset
= (u64
)-1;
3723 ret
= btrfs_search_forward(root
, &key
, path
, 0);
3730 if (key
.type
!= BTRFS_ROOT_ITEM_KEY
||
3731 (key
.objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
3732 key
.objectid
!= BTRFS_FS_TREE_OBJECTID
) ||
3733 key
.objectid
> BTRFS_LAST_FREE_OBJECTID
)
3736 eb
= path
->nodes
[0];
3737 slot
= path
->slots
[0];
3738 item_size
= btrfs_item_size_nr(eb
, slot
);
3739 if (item_size
< sizeof(root_item
))
3742 read_extent_buffer(eb
, &root_item
,
3743 btrfs_item_ptr_offset(eb
, slot
),
3744 (int)sizeof(root_item
));
3745 if (btrfs_root_refs(&root_item
) == 0)
3748 if (!btrfs_is_empty_uuid(root_item
.uuid
) ||
3749 !btrfs_is_empty_uuid(root_item
.received_uuid
)) {
3753 btrfs_release_path(path
);
3755 * 1 - subvol uuid item
3756 * 1 - received_subvol uuid item
3758 trans
= btrfs_start_transaction(fs_info
->uuid_root
, 2);
3759 if (IS_ERR(trans
)) {
3760 ret
= PTR_ERR(trans
);
3768 if (!btrfs_is_empty_uuid(root_item
.uuid
)) {
3769 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
3771 BTRFS_UUID_KEY_SUBVOL
,
3774 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
3780 if (!btrfs_is_empty_uuid(root_item
.received_uuid
)) {
3781 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
3782 root_item
.received_uuid
,
3783 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
3786 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
3794 ret
= btrfs_end_transaction(trans
, fs_info
->uuid_root
);
3800 btrfs_release_path(path
);
3801 if (key
.offset
< (u64
)-1) {
3803 } else if (key
.type
< BTRFS_ROOT_ITEM_KEY
) {
3805 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3806 } else if (key
.objectid
< (u64
)-1) {
3808 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3817 btrfs_free_path(path
);
3818 if (trans
&& !IS_ERR(trans
))
3819 btrfs_end_transaction(trans
, fs_info
->uuid_root
);
3821 btrfs_warn(fs_info
, "btrfs_uuid_scan_kthread failed %d", ret
);
3823 fs_info
->update_uuid_tree_gen
= 1;
3824 up(&fs_info
->uuid_tree_rescan_sem
);
3829 * Callback for btrfs_uuid_tree_iterate().
3831 * 0 check succeeded, the entry is not outdated.
3832 * < 0 if an error occured.
3833 * > 0 if the check failed, which means the caller shall remove the entry.
3835 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info
*fs_info
,
3836 u8
*uuid
, u8 type
, u64 subid
)
3838 struct btrfs_key key
;
3840 struct btrfs_root
*subvol_root
;
3842 if (type
!= BTRFS_UUID_KEY_SUBVOL
&&
3843 type
!= BTRFS_UUID_KEY_RECEIVED_SUBVOL
)
3846 key
.objectid
= subid
;
3847 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3848 key
.offset
= (u64
)-1;
3849 subvol_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
3850 if (IS_ERR(subvol_root
)) {
3851 ret
= PTR_ERR(subvol_root
);
3858 case BTRFS_UUID_KEY_SUBVOL
:
3859 if (memcmp(uuid
, subvol_root
->root_item
.uuid
, BTRFS_UUID_SIZE
))
3862 case BTRFS_UUID_KEY_RECEIVED_SUBVOL
:
3863 if (memcmp(uuid
, subvol_root
->root_item
.received_uuid
,
3873 static int btrfs_uuid_rescan_kthread(void *data
)
3875 struct btrfs_fs_info
*fs_info
= (struct btrfs_fs_info
*)data
;
3879 * 1st step is to iterate through the existing UUID tree and
3880 * to delete all entries that contain outdated data.
3881 * 2nd step is to add all missing entries to the UUID tree.
3883 ret
= btrfs_uuid_tree_iterate(fs_info
, btrfs_check_uuid_tree_entry
);
3885 btrfs_warn(fs_info
, "iterating uuid_tree failed %d", ret
);
3886 up(&fs_info
->uuid_tree_rescan_sem
);
3889 return btrfs_uuid_scan_kthread(data
);
3892 int btrfs_create_uuid_tree(struct btrfs_fs_info
*fs_info
)
3894 struct btrfs_trans_handle
*trans
;
3895 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
3896 struct btrfs_root
*uuid_root
;
3897 struct task_struct
*task
;
3904 trans
= btrfs_start_transaction(tree_root
, 2);
3906 return PTR_ERR(trans
);
3908 uuid_root
= btrfs_create_tree(trans
, fs_info
,
3909 BTRFS_UUID_TREE_OBJECTID
);
3910 if (IS_ERR(uuid_root
)) {
3911 btrfs_abort_transaction(trans
, tree_root
,
3912 PTR_ERR(uuid_root
));
3913 return PTR_ERR(uuid_root
);
3916 fs_info
->uuid_root
= uuid_root
;
3918 ret
= btrfs_commit_transaction(trans
, tree_root
);
3922 down(&fs_info
->uuid_tree_rescan_sem
);
3923 task
= kthread_run(btrfs_uuid_scan_kthread
, fs_info
, "btrfs-uuid");
3925 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3926 btrfs_warn(fs_info
, "failed to start uuid_scan task");
3927 up(&fs_info
->uuid_tree_rescan_sem
);
3928 return PTR_ERR(task
);
3934 int btrfs_check_uuid_tree(struct btrfs_fs_info
*fs_info
)
3936 struct task_struct
*task
;
3938 down(&fs_info
->uuid_tree_rescan_sem
);
3939 task
= kthread_run(btrfs_uuid_rescan_kthread
, fs_info
, "btrfs-uuid");
3941 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3942 btrfs_warn(fs_info
, "failed to start uuid_rescan task");
3943 up(&fs_info
->uuid_tree_rescan_sem
);
3944 return PTR_ERR(task
);
3951 * shrinking a device means finding all of the device extents past
3952 * the new size, and then following the back refs to the chunks.
3953 * The chunk relocation code actually frees the device extent
3955 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
3957 struct btrfs_trans_handle
*trans
;
3958 struct btrfs_root
*root
= device
->dev_root
;
3959 struct btrfs_dev_extent
*dev_extent
= NULL
;
3960 struct btrfs_path
*path
;
3968 bool retried
= false;
3969 struct extent_buffer
*l
;
3970 struct btrfs_key key
;
3971 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
3972 u64 old_total
= btrfs_super_total_bytes(super_copy
);
3973 u64 old_size
= btrfs_device_get_total_bytes(device
);
3974 u64 diff
= old_size
- new_size
;
3976 if (device
->is_tgtdev_for_dev_replace
)
3979 path
= btrfs_alloc_path();
3987 btrfs_device_set_total_bytes(device
, new_size
);
3988 if (device
->writeable
) {
3989 device
->fs_devices
->total_rw_bytes
-= diff
;
3990 spin_lock(&root
->fs_info
->free_chunk_lock
);
3991 root
->fs_info
->free_chunk_space
-= diff
;
3992 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3994 unlock_chunks(root
);
3997 key
.objectid
= device
->devid
;
3998 key
.offset
= (u64
)-1;
3999 key
.type
= BTRFS_DEV_EXTENT_KEY
;
4002 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4006 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
4011 btrfs_release_path(path
);
4016 slot
= path
->slots
[0];
4017 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
4019 if (key
.objectid
!= device
->devid
) {
4020 btrfs_release_path(path
);
4024 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
4025 length
= btrfs_dev_extent_length(l
, dev_extent
);
4027 if (key
.offset
+ length
<= new_size
) {
4028 btrfs_release_path(path
);
4032 chunk_tree
= btrfs_dev_extent_chunk_tree(l
, dev_extent
);
4033 chunk_objectid
= btrfs_dev_extent_chunk_objectid(l
, dev_extent
);
4034 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
4035 btrfs_release_path(path
);
4037 ret
= btrfs_relocate_chunk(root
, chunk_tree
, chunk_objectid
,
4039 if (ret
&& ret
!= -ENOSPC
)
4043 } while (key
.offset
-- > 0);
4045 if (failed
&& !retried
) {
4049 } else if (failed
&& retried
) {
4053 btrfs_device_set_total_bytes(device
, old_size
);
4054 if (device
->writeable
)
4055 device
->fs_devices
->total_rw_bytes
+= diff
;
4056 spin_lock(&root
->fs_info
->free_chunk_lock
);
4057 root
->fs_info
->free_chunk_space
+= diff
;
4058 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4059 unlock_chunks(root
);
4063 /* Shrinking succeeded, else we would be at "done". */
4064 trans
= btrfs_start_transaction(root
, 0);
4065 if (IS_ERR(trans
)) {
4066 ret
= PTR_ERR(trans
);
4071 btrfs_device_set_disk_total_bytes(device
, new_size
);
4072 if (list_empty(&device
->resized_list
))
4073 list_add_tail(&device
->resized_list
,
4074 &root
->fs_info
->fs_devices
->resized_devices
);
4076 WARN_ON(diff
> old_total
);
4077 btrfs_set_super_total_bytes(super_copy
, old_total
- diff
);
4078 unlock_chunks(root
);
4080 /* Now btrfs_update_device() will change the on-disk size. */
4081 ret
= btrfs_update_device(trans
, device
);
4082 btrfs_end_transaction(trans
, root
);
4084 btrfs_free_path(path
);
4088 static int btrfs_add_system_chunk(struct btrfs_root
*root
,
4089 struct btrfs_key
*key
,
4090 struct btrfs_chunk
*chunk
, int item_size
)
4092 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
4093 struct btrfs_disk_key disk_key
;
4098 array_size
= btrfs_super_sys_array_size(super_copy
);
4099 if (array_size
+ item_size
+ sizeof(disk_key
)
4100 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4101 unlock_chunks(root
);
4105 ptr
= super_copy
->sys_chunk_array
+ array_size
;
4106 btrfs_cpu_key_to_disk(&disk_key
, key
);
4107 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
4108 ptr
+= sizeof(disk_key
);
4109 memcpy(ptr
, chunk
, item_size
);
4110 item_size
+= sizeof(disk_key
);
4111 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
4112 unlock_chunks(root
);
4118 * sort the devices in descending order by max_avail, total_avail
4120 static int btrfs_cmp_device_info(const void *a
, const void *b
)
4122 const struct btrfs_device_info
*di_a
= a
;
4123 const struct btrfs_device_info
*di_b
= b
;
4125 if (di_a
->max_avail
> di_b
->max_avail
)
4127 if (di_a
->max_avail
< di_b
->max_avail
)
4129 if (di_a
->total_avail
> di_b
->total_avail
)
4131 if (di_a
->total_avail
< di_b
->total_avail
)
4136 static struct btrfs_raid_attr btrfs_raid_array
[BTRFS_NR_RAID_TYPES
] = {
4137 [BTRFS_RAID_RAID10
] = {
4140 .devs_max
= 0, /* 0 == as many as possible */
4142 .devs_increment
= 2,
4145 [BTRFS_RAID_RAID1
] = {
4150 .devs_increment
= 2,
4153 [BTRFS_RAID_DUP
] = {
4158 .devs_increment
= 1,
4161 [BTRFS_RAID_RAID0
] = {
4166 .devs_increment
= 1,
4169 [BTRFS_RAID_SINGLE
] = {
4174 .devs_increment
= 1,
4177 [BTRFS_RAID_RAID5
] = {
4182 .devs_increment
= 1,
4185 [BTRFS_RAID_RAID6
] = {
4190 .devs_increment
= 1,
4195 static u32
find_raid56_stripe_len(u32 data_devices
, u32 dev_stripe_target
)
4197 /* TODO allow them to set a preferred stripe size */
4201 static void check_raid56_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
4203 if (!(type
& (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
)))
4206 btrfs_set_fs_incompat(info
, RAID56
);
4209 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4210 - sizeof(struct btrfs_item) \
4211 - sizeof(struct btrfs_chunk)) \
4212 / sizeof(struct btrfs_stripe) + 1)
4214 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4215 - 2 * sizeof(struct btrfs_disk_key) \
4216 - 2 * sizeof(struct btrfs_chunk)) \
4217 / sizeof(struct btrfs_stripe) + 1)
4219 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4220 struct btrfs_root
*extent_root
, u64 start
,
4223 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
4224 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
4225 struct list_head
*cur
;
4226 struct map_lookup
*map
= NULL
;
4227 struct extent_map_tree
*em_tree
;
4228 struct extent_map
*em
;
4229 struct btrfs_device_info
*devices_info
= NULL
;
4231 int num_stripes
; /* total number of stripes to allocate */
4232 int data_stripes
; /* number of stripes that count for
4234 int sub_stripes
; /* sub_stripes info for map */
4235 int dev_stripes
; /* stripes per dev */
4236 int devs_max
; /* max devs to use */
4237 int devs_min
; /* min devs needed */
4238 int devs_increment
; /* ndevs has to be a multiple of this */
4239 int ncopies
; /* how many copies to data has */
4241 u64 max_stripe_size
;
4245 u64 raid_stripe_len
= BTRFS_STRIPE_LEN
;
4251 BUG_ON(!alloc_profile_is_valid(type
, 0));
4253 if (list_empty(&fs_devices
->alloc_list
))
4256 index
= __get_raid_index(type
);
4258 sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
4259 dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
4260 devs_max
= btrfs_raid_array
[index
].devs_max
;
4261 devs_min
= btrfs_raid_array
[index
].devs_min
;
4262 devs_increment
= btrfs_raid_array
[index
].devs_increment
;
4263 ncopies
= btrfs_raid_array
[index
].ncopies
;
4265 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
4266 max_stripe_size
= 1024 * 1024 * 1024;
4267 max_chunk_size
= 10 * max_stripe_size
;
4269 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4270 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
4271 /* for larger filesystems, use larger metadata chunks */
4272 if (fs_devices
->total_rw_bytes
> 50ULL * 1024 * 1024 * 1024)
4273 max_stripe_size
= 1024 * 1024 * 1024;
4275 max_stripe_size
= 256 * 1024 * 1024;
4276 max_chunk_size
= max_stripe_size
;
4278 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4279 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4280 max_stripe_size
= 32 * 1024 * 1024;
4281 max_chunk_size
= 2 * max_stripe_size
;
4283 devs_max
= BTRFS_MAX_DEVS_SYS_CHUNK
;
4285 btrfs_err(info
, "invalid chunk type 0x%llx requested",
4290 /* we don't want a chunk larger than 10% of writeable space */
4291 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
4294 devices_info
= kzalloc(sizeof(*devices_info
) * fs_devices
->rw_devices
,
4299 cur
= fs_devices
->alloc_list
.next
;
4302 * in the first pass through the devices list, we gather information
4303 * about the available holes on each device.
4306 while (cur
!= &fs_devices
->alloc_list
) {
4307 struct btrfs_device
*device
;
4311 device
= list_entry(cur
, struct btrfs_device
, dev_alloc_list
);
4315 if (!device
->writeable
) {
4317 "BTRFS: read-only device in alloc_list\n");
4321 if (!device
->in_fs_metadata
||
4322 device
->is_tgtdev_for_dev_replace
)
4325 if (device
->total_bytes
> device
->bytes_used
)
4326 total_avail
= device
->total_bytes
- device
->bytes_used
;
4330 /* If there is no space on this device, skip it. */
4331 if (total_avail
== 0)
4334 ret
= find_free_dev_extent(trans
, device
,
4335 max_stripe_size
* dev_stripes
,
4336 &dev_offset
, &max_avail
);
4337 if (ret
&& ret
!= -ENOSPC
)
4341 max_avail
= max_stripe_size
* dev_stripes
;
4343 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
4346 if (ndevs
== fs_devices
->rw_devices
) {
4347 WARN(1, "%s: found more than %llu devices\n",
4348 __func__
, fs_devices
->rw_devices
);
4351 devices_info
[ndevs
].dev_offset
= dev_offset
;
4352 devices_info
[ndevs
].max_avail
= max_avail
;
4353 devices_info
[ndevs
].total_avail
= total_avail
;
4354 devices_info
[ndevs
].dev
= device
;
4359 * now sort the devices by hole size / available space
4361 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
4362 btrfs_cmp_device_info
, NULL
);
4364 /* round down to number of usable stripes */
4365 ndevs
-= ndevs
% devs_increment
;
4367 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
4372 if (devs_max
&& ndevs
> devs_max
)
4375 * the primary goal is to maximize the number of stripes, so use as many
4376 * devices as possible, even if the stripes are not maximum sized.
4378 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4379 num_stripes
= ndevs
* dev_stripes
;
4382 * this will have to be fixed for RAID1 and RAID10 over
4385 data_stripes
= num_stripes
/ ncopies
;
4387 if (type
& BTRFS_BLOCK_GROUP_RAID5
) {
4388 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 1,
4389 btrfs_super_stripesize(info
->super_copy
));
4390 data_stripes
= num_stripes
- 1;
4392 if (type
& BTRFS_BLOCK_GROUP_RAID6
) {
4393 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 2,
4394 btrfs_super_stripesize(info
->super_copy
));
4395 data_stripes
= num_stripes
- 2;
4399 * Use the number of data stripes to figure out how big this chunk
4400 * is really going to be in terms of logical address space,
4401 * and compare that answer with the max chunk size
4403 if (stripe_size
* data_stripes
> max_chunk_size
) {
4404 u64 mask
= (1ULL << 24) - 1;
4405 stripe_size
= max_chunk_size
;
4406 do_div(stripe_size
, data_stripes
);
4408 /* bump the answer up to a 16MB boundary */
4409 stripe_size
= (stripe_size
+ mask
) & ~mask
;
4411 /* but don't go higher than the limits we found
4412 * while searching for free extents
4414 if (stripe_size
> devices_info
[ndevs
-1].max_avail
)
4415 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4418 do_div(stripe_size
, dev_stripes
);
4420 /* align to BTRFS_STRIPE_LEN */
4421 do_div(stripe_size
, raid_stripe_len
);
4422 stripe_size
*= raid_stripe_len
;
4424 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
4429 map
->num_stripes
= num_stripes
;
4431 for (i
= 0; i
< ndevs
; ++i
) {
4432 for (j
= 0; j
< dev_stripes
; ++j
) {
4433 int s
= i
* dev_stripes
+ j
;
4434 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
4435 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
4439 map
->sector_size
= extent_root
->sectorsize
;
4440 map
->stripe_len
= raid_stripe_len
;
4441 map
->io_align
= raid_stripe_len
;
4442 map
->io_width
= raid_stripe_len
;
4444 map
->sub_stripes
= sub_stripes
;
4446 num_bytes
= stripe_size
* data_stripes
;
4448 trace_btrfs_chunk_alloc(info
->chunk_root
, map
, start
, num_bytes
);
4450 em
= alloc_extent_map();
4456 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
4457 em
->bdev
= (struct block_device
*)map
;
4459 em
->len
= num_bytes
;
4460 em
->block_start
= 0;
4461 em
->block_len
= em
->len
;
4462 em
->orig_block_len
= stripe_size
;
4464 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
4465 write_lock(&em_tree
->lock
);
4466 ret
= add_extent_mapping(em_tree
, em
, 0);
4468 list_add_tail(&em
->list
, &trans
->transaction
->pending_chunks
);
4469 atomic_inc(&em
->refs
);
4471 write_unlock(&em_tree
->lock
);
4473 free_extent_map(em
);
4477 ret
= btrfs_make_block_group(trans
, extent_root
, 0, type
,
4478 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4481 goto error_del_extent
;
4483 for (i
= 0; i
< map
->num_stripes
; i
++) {
4484 num_bytes
= map
->stripes
[i
].dev
->bytes_used
+ stripe_size
;
4485 btrfs_device_set_bytes_used(map
->stripes
[i
].dev
, num_bytes
);
4488 spin_lock(&extent_root
->fs_info
->free_chunk_lock
);
4489 extent_root
->fs_info
->free_chunk_space
-= (stripe_size
*
4491 spin_unlock(&extent_root
->fs_info
->free_chunk_lock
);
4493 free_extent_map(em
);
4494 check_raid56_incompat_flag(extent_root
->fs_info
, type
);
4496 kfree(devices_info
);
4500 write_lock(&em_tree
->lock
);
4501 remove_extent_mapping(em_tree
, em
);
4502 write_unlock(&em_tree
->lock
);
4504 /* One for our allocation */
4505 free_extent_map(em
);
4506 /* One for the tree reference */
4507 free_extent_map(em
);
4509 kfree(devices_info
);
4513 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
4514 struct btrfs_root
*extent_root
,
4515 u64 chunk_offset
, u64 chunk_size
)
4517 struct btrfs_key key
;
4518 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
4519 struct btrfs_device
*device
;
4520 struct btrfs_chunk
*chunk
;
4521 struct btrfs_stripe
*stripe
;
4522 struct extent_map_tree
*em_tree
;
4523 struct extent_map
*em
;
4524 struct map_lookup
*map
;
4531 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
4532 read_lock(&em_tree
->lock
);
4533 em
= lookup_extent_mapping(em_tree
, chunk_offset
, chunk_size
);
4534 read_unlock(&em_tree
->lock
);
4537 btrfs_crit(extent_root
->fs_info
, "unable to find logical "
4538 "%Lu len %Lu", chunk_offset
, chunk_size
);
4542 if (em
->start
!= chunk_offset
|| em
->len
!= chunk_size
) {
4543 btrfs_crit(extent_root
->fs_info
, "found a bad mapping, wanted"
4544 " %Lu-%Lu, found %Lu-%Lu", chunk_offset
,
4545 chunk_size
, em
->start
, em
->len
);
4546 free_extent_map(em
);
4550 map
= (struct map_lookup
*)em
->bdev
;
4551 item_size
= btrfs_chunk_item_size(map
->num_stripes
);
4552 stripe_size
= em
->orig_block_len
;
4554 chunk
= kzalloc(item_size
, GFP_NOFS
);
4560 for (i
= 0; i
< map
->num_stripes
; i
++) {
4561 device
= map
->stripes
[i
].dev
;
4562 dev_offset
= map
->stripes
[i
].physical
;
4564 ret
= btrfs_update_device(trans
, device
);
4567 ret
= btrfs_alloc_dev_extent(trans
, device
,
4568 chunk_root
->root_key
.objectid
,
4569 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4570 chunk_offset
, dev_offset
,
4576 stripe
= &chunk
->stripe
;
4577 for (i
= 0; i
< map
->num_stripes
; i
++) {
4578 device
= map
->stripes
[i
].dev
;
4579 dev_offset
= map
->stripes
[i
].physical
;
4581 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
4582 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
4583 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
4587 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
4588 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
4589 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
4590 btrfs_set_stack_chunk_type(chunk
, map
->type
);
4591 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
4592 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
4593 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
4594 btrfs_set_stack_chunk_sector_size(chunk
, extent_root
->sectorsize
);
4595 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
4597 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
4598 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
4599 key
.offset
= chunk_offset
;
4601 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
4602 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4604 * TODO: Cleanup of inserted chunk root in case of
4607 ret
= btrfs_add_system_chunk(chunk_root
, &key
, chunk
,
4613 free_extent_map(em
);
4618 * Chunk allocation falls into two parts. The first part does works
4619 * that make the new allocated chunk useable, but not do any operation
4620 * that modifies the chunk tree. The second part does the works that
4621 * require modifying the chunk tree. This division is important for the
4622 * bootstrap process of adding storage to a seed btrfs.
4624 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4625 struct btrfs_root
*extent_root
, u64 type
)
4629 chunk_offset
= find_next_chunk(extent_root
->fs_info
);
4630 return __btrfs_alloc_chunk(trans
, extent_root
, chunk_offset
, type
);
4633 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
4634 struct btrfs_root
*root
,
4635 struct btrfs_device
*device
)
4638 u64 sys_chunk_offset
;
4640 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4641 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
4644 chunk_offset
= find_next_chunk(fs_info
);
4645 alloc_profile
= btrfs_get_alloc_profile(extent_root
, 0);
4646 ret
= __btrfs_alloc_chunk(trans
, extent_root
, chunk_offset
,
4651 sys_chunk_offset
= find_next_chunk(root
->fs_info
);
4652 alloc_profile
= btrfs_get_alloc_profile(fs_info
->chunk_root
, 0);
4653 ret
= __btrfs_alloc_chunk(trans
, extent_root
, sys_chunk_offset
,
4658 static inline int btrfs_chunk_max_errors(struct map_lookup
*map
)
4662 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
4663 BTRFS_BLOCK_GROUP_RAID10
|
4664 BTRFS_BLOCK_GROUP_RAID5
|
4665 BTRFS_BLOCK_GROUP_DUP
)) {
4667 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
) {
4676 int btrfs_chunk_readonly(struct btrfs_root
*root
, u64 chunk_offset
)
4678 struct extent_map
*em
;
4679 struct map_lookup
*map
;
4680 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
4685 read_lock(&map_tree
->map_tree
.lock
);
4686 em
= lookup_extent_mapping(&map_tree
->map_tree
, chunk_offset
, 1);
4687 read_unlock(&map_tree
->map_tree
.lock
);
4691 map
= (struct map_lookup
*)em
->bdev
;
4692 for (i
= 0; i
< map
->num_stripes
; i
++) {
4693 if (map
->stripes
[i
].dev
->missing
) {
4698 if (!map
->stripes
[i
].dev
->writeable
) {
4705 * If the number of missing devices is larger than max errors,
4706 * we can not write the data into that chunk successfully, so
4709 if (miss_ndevs
> btrfs_chunk_max_errors(map
))
4712 free_extent_map(em
);
4716 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
4718 extent_map_tree_init(&tree
->map_tree
);
4721 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
4723 struct extent_map
*em
;
4726 write_lock(&tree
->map_tree
.lock
);
4727 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
4729 remove_extent_mapping(&tree
->map_tree
, em
);
4730 write_unlock(&tree
->map_tree
.lock
);
4734 free_extent_map(em
);
4735 /* once for the tree */
4736 free_extent_map(em
);
4740 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
4742 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
4743 struct extent_map
*em
;
4744 struct map_lookup
*map
;
4745 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4748 read_lock(&em_tree
->lock
);
4749 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4750 read_unlock(&em_tree
->lock
);
4753 * We could return errors for these cases, but that could get ugly and
4754 * we'd probably do the same thing which is just not do anything else
4755 * and exit, so return 1 so the callers don't try to use other copies.
4758 btrfs_crit(fs_info
, "No mapping for %Lu-%Lu", logical
,
4763 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
4764 btrfs_crit(fs_info
, "Invalid mapping for %Lu-%Lu, got "
4765 "%Lu-%Lu", logical
, logical
+len
, em
->start
,
4766 em
->start
+ em
->len
);
4767 free_extent_map(em
);
4771 map
= (struct map_lookup
*)em
->bdev
;
4772 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
4773 ret
= map
->num_stripes
;
4774 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
4775 ret
= map
->sub_stripes
;
4776 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID5
)
4778 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
4782 free_extent_map(em
);
4784 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
4785 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
))
4787 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
4792 unsigned long btrfs_full_stripe_len(struct btrfs_root
*root
,
4793 struct btrfs_mapping_tree
*map_tree
,
4796 struct extent_map
*em
;
4797 struct map_lookup
*map
;
4798 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4799 unsigned long len
= root
->sectorsize
;
4801 read_lock(&em_tree
->lock
);
4802 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4803 read_unlock(&em_tree
->lock
);
4806 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4807 map
= (struct map_lookup
*)em
->bdev
;
4808 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
4809 BTRFS_BLOCK_GROUP_RAID6
)) {
4810 len
= map
->stripe_len
* nr_data_stripes(map
);
4812 free_extent_map(em
);
4816 int btrfs_is_parity_mirror(struct btrfs_mapping_tree
*map_tree
,
4817 u64 logical
, u64 len
, int mirror_num
)
4819 struct extent_map
*em
;
4820 struct map_lookup
*map
;
4821 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4824 read_lock(&em_tree
->lock
);
4825 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4826 read_unlock(&em_tree
->lock
);
4829 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4830 map
= (struct map_lookup
*)em
->bdev
;
4831 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
4832 BTRFS_BLOCK_GROUP_RAID6
))
4834 free_extent_map(em
);
4838 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
4839 struct map_lookup
*map
, int first
, int num
,
4840 int optimal
, int dev_replace_is_ongoing
)
4844 struct btrfs_device
*srcdev
;
4846 if (dev_replace_is_ongoing
&&
4847 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
4848 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
4849 srcdev
= fs_info
->dev_replace
.srcdev
;
4854 * try to avoid the drive that is the source drive for a
4855 * dev-replace procedure, only choose it if no other non-missing
4856 * mirror is available
4858 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
4859 if (map
->stripes
[optimal
].dev
->bdev
&&
4860 (tolerance
|| map
->stripes
[optimal
].dev
!= srcdev
))
4862 for (i
= first
; i
< first
+ num
; i
++) {
4863 if (map
->stripes
[i
].dev
->bdev
&&
4864 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
4869 /* we couldn't find one that doesn't fail. Just return something
4870 * and the io error handling code will clean up eventually
4875 static inline int parity_smaller(u64 a
, u64 b
)
4880 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4881 static void sort_parity_stripes(struct btrfs_bio
*bbio
, u64
*raid_map
)
4883 struct btrfs_bio_stripe s
;
4890 for (i
= 0; i
< bbio
->num_stripes
- 1; i
++) {
4891 if (parity_smaller(raid_map
[i
], raid_map
[i
+1])) {
4892 s
= bbio
->stripes
[i
];
4894 bbio
->stripes
[i
] = bbio
->stripes
[i
+1];
4895 raid_map
[i
] = raid_map
[i
+1];
4896 bbio
->stripes
[i
+1] = s
;
4904 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
4905 u64 logical
, u64
*length
,
4906 struct btrfs_bio
**bbio_ret
,
4907 int mirror_num
, u64
**raid_map_ret
)
4909 struct extent_map
*em
;
4910 struct map_lookup
*map
;
4911 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
4912 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4915 u64 stripe_end_offset
;
4920 u64
*raid_map
= NULL
;
4926 struct btrfs_bio
*bbio
= NULL
;
4927 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
4928 int dev_replace_is_ongoing
= 0;
4929 int num_alloc_stripes
;
4930 int patch_the_first_stripe_for_dev_replace
= 0;
4931 u64 physical_to_patch_in_first_stripe
= 0;
4932 u64 raid56_full_stripe_start
= (u64
)-1;
4934 read_lock(&em_tree
->lock
);
4935 em
= lookup_extent_mapping(em_tree
, logical
, *length
);
4936 read_unlock(&em_tree
->lock
);
4939 btrfs_crit(fs_info
, "unable to find logical %llu len %llu",
4944 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
4945 btrfs_crit(fs_info
, "found a bad mapping, wanted %Lu, "
4946 "found %Lu-%Lu", logical
, em
->start
,
4947 em
->start
+ em
->len
);
4948 free_extent_map(em
);
4952 map
= (struct map_lookup
*)em
->bdev
;
4953 offset
= logical
- em
->start
;
4955 stripe_len
= map
->stripe_len
;
4958 * stripe_nr counts the total number of stripes we have to stride
4959 * to get to this block
4961 do_div(stripe_nr
, stripe_len
);
4963 stripe_offset
= stripe_nr
* stripe_len
;
4964 BUG_ON(offset
< stripe_offset
);
4966 /* stripe_offset is the offset of this block in its stripe*/
4967 stripe_offset
= offset
- stripe_offset
;
4969 /* if we're here for raid56, we need to know the stripe aligned start */
4970 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
)) {
4971 unsigned long full_stripe_len
= stripe_len
* nr_data_stripes(map
);
4972 raid56_full_stripe_start
= offset
;
4974 /* allow a write of a full stripe, but make sure we don't
4975 * allow straddling of stripes
4977 do_div(raid56_full_stripe_start
, full_stripe_len
);
4978 raid56_full_stripe_start
*= full_stripe_len
;
4981 if (rw
& REQ_DISCARD
) {
4982 /* we don't discard raid56 yet */
4984 (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
)) {
4988 *length
= min_t(u64
, em
->len
- offset
, *length
);
4989 } else if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
4991 /* For writes to RAID[56], allow a full stripeset across all disks.
4992 For other RAID types and for RAID[56] reads, just allow a single
4993 stripe (on a single disk). */
4994 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
) &&
4996 max_len
= stripe_len
* nr_data_stripes(map
) -
4997 (offset
- raid56_full_stripe_start
);
4999 /* we limit the length of each bio to what fits in a stripe */
5000 max_len
= stripe_len
- stripe_offset
;
5002 *length
= min_t(u64
, em
->len
- offset
, max_len
);
5004 *length
= em
->len
- offset
;
5007 /* This is for when we're called from btrfs_merge_bio_hook() and all
5008 it cares about is the length */
5012 btrfs_dev_replace_lock(dev_replace
);
5013 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
5014 if (!dev_replace_is_ongoing
)
5015 btrfs_dev_replace_unlock(dev_replace
);
5017 if (dev_replace_is_ongoing
&& mirror_num
== map
->num_stripes
+ 1 &&
5018 !(rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) &&
5019 dev_replace
->tgtdev
!= NULL
) {
5021 * in dev-replace case, for repair case (that's the only
5022 * case where the mirror is selected explicitly when
5023 * calling btrfs_map_block), blocks left of the left cursor
5024 * can also be read from the target drive.
5025 * For REQ_GET_READ_MIRRORS, the target drive is added as
5026 * the last one to the array of stripes. For READ, it also
5027 * needs to be supported using the same mirror number.
5028 * If the requested block is not left of the left cursor,
5029 * EIO is returned. This can happen because btrfs_num_copies()
5030 * returns one more in the dev-replace case.
5032 u64 tmp_length
= *length
;
5033 struct btrfs_bio
*tmp_bbio
= NULL
;
5034 int tmp_num_stripes
;
5035 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5036 int index_srcdev
= 0;
5038 u64 physical_of_found
= 0;
5040 ret
= __btrfs_map_block(fs_info
, REQ_GET_READ_MIRRORS
,
5041 logical
, &tmp_length
, &tmp_bbio
, 0, NULL
);
5043 WARN_ON(tmp_bbio
!= NULL
);
5047 tmp_num_stripes
= tmp_bbio
->num_stripes
;
5048 if (mirror_num
> tmp_num_stripes
) {
5050 * REQ_GET_READ_MIRRORS does not contain this
5051 * mirror, that means that the requested area
5052 * is not left of the left cursor
5060 * process the rest of the function using the mirror_num
5061 * of the source drive. Therefore look it up first.
5062 * At the end, patch the device pointer to the one of the
5065 for (i
= 0; i
< tmp_num_stripes
; i
++) {
5066 if (tmp_bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5068 * In case of DUP, in order to keep it
5069 * simple, only add the mirror with the
5070 * lowest physical address
5073 physical_of_found
<=
5074 tmp_bbio
->stripes
[i
].physical
)
5079 tmp_bbio
->stripes
[i
].physical
;
5084 mirror_num
= index_srcdev
+ 1;
5085 patch_the_first_stripe_for_dev_replace
= 1;
5086 physical_to_patch_in_first_stripe
= physical_of_found
;
5095 } else if (mirror_num
> map
->num_stripes
) {
5101 stripe_nr_orig
= stripe_nr
;
5102 stripe_nr_end
= ALIGN(offset
+ *length
, map
->stripe_len
);
5103 do_div(stripe_nr_end
, map
->stripe_len
);
5104 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
5107 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5108 if (rw
& REQ_DISCARD
)
5109 num_stripes
= min_t(u64
, map
->num_stripes
,
5110 stripe_nr_end
- stripe_nr_orig
);
5111 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
5112 if (!(rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)))
5114 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
5115 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
))
5116 num_stripes
= map
->num_stripes
;
5117 else if (mirror_num
)
5118 stripe_index
= mirror_num
- 1;
5120 stripe_index
= find_live_mirror(fs_info
, map
, 0,
5122 current
->pid
% map
->num_stripes
,
5123 dev_replace_is_ongoing
);
5124 mirror_num
= stripe_index
+ 1;
5127 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
5128 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) {
5129 num_stripes
= map
->num_stripes
;
5130 } else if (mirror_num
) {
5131 stripe_index
= mirror_num
- 1;
5136 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5137 int factor
= map
->num_stripes
/ map
->sub_stripes
;
5139 stripe_index
= do_div(stripe_nr
, factor
);
5140 stripe_index
*= map
->sub_stripes
;
5142 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
))
5143 num_stripes
= map
->sub_stripes
;
5144 else if (rw
& REQ_DISCARD
)
5145 num_stripes
= min_t(u64
, map
->sub_stripes
*
5146 (stripe_nr_end
- stripe_nr_orig
),
5148 else if (mirror_num
)
5149 stripe_index
+= mirror_num
- 1;
5151 int old_stripe_index
= stripe_index
;
5152 stripe_index
= find_live_mirror(fs_info
, map
,
5154 map
->sub_stripes
, stripe_index
+
5155 current
->pid
% map
->sub_stripes
,
5156 dev_replace_is_ongoing
);
5157 mirror_num
= stripe_index
- old_stripe_index
+ 1;
5160 } else if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
5161 BTRFS_BLOCK_GROUP_RAID6
)) {
5164 if (bbio_ret
&& ((rw
& REQ_WRITE
) || mirror_num
> 1)
5168 /* push stripe_nr back to the start of the full stripe */
5169 stripe_nr
= raid56_full_stripe_start
;
5170 do_div(stripe_nr
, stripe_len
);
5172 stripe_index
= do_div(stripe_nr
, nr_data_stripes(map
));
5174 /* RAID[56] write or recovery. Return all stripes */
5175 num_stripes
= map
->num_stripes
;
5176 max_errors
= nr_parity_stripes(map
);
5178 raid_map
= kmalloc_array(num_stripes
, sizeof(u64
),
5185 /* Work out the disk rotation on this stripe-set */
5187 rot
= do_div(tmp
, num_stripes
);
5189 /* Fill in the logical address of each stripe */
5190 tmp
= stripe_nr
* nr_data_stripes(map
);
5191 for (i
= 0; i
< nr_data_stripes(map
); i
++)
5192 raid_map
[(i
+rot
) % num_stripes
] =
5193 em
->start
+ (tmp
+ i
) * map
->stripe_len
;
5195 raid_map
[(i
+rot
) % map
->num_stripes
] = RAID5_P_STRIPE
;
5196 if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5197 raid_map
[(i
+rot
+1) % num_stripes
] =
5200 *length
= map
->stripe_len
;
5205 * Mirror #0 or #1 means the original data block.
5206 * Mirror #2 is RAID5 parity block.
5207 * Mirror #3 is RAID6 Q block.
5209 stripe_index
= do_div(stripe_nr
, nr_data_stripes(map
));
5211 stripe_index
= nr_data_stripes(map
) +
5214 /* We distribute the parity blocks across stripes */
5215 tmp
= stripe_nr
+ stripe_index
;
5216 stripe_index
= do_div(tmp
, map
->num_stripes
);
5217 if (!(rw
& (REQ_WRITE
| REQ_DISCARD
|
5218 REQ_GET_READ_MIRRORS
)) && mirror_num
<= 1)
5223 * after this do_div call, stripe_nr is the number of stripes
5224 * on this device we have to walk to find the data, and
5225 * stripe_index is the number of our device in the stripe array
5227 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
5228 mirror_num
= stripe_index
+ 1;
5230 BUG_ON(stripe_index
>= map
->num_stripes
);
5232 num_alloc_stripes
= num_stripes
;
5233 if (dev_replace_is_ongoing
) {
5234 if (rw
& (REQ_WRITE
| REQ_DISCARD
))
5235 num_alloc_stripes
<<= 1;
5236 if (rw
& REQ_GET_READ_MIRRORS
)
5237 num_alloc_stripes
++;
5239 bbio
= kzalloc(btrfs_bio_size(num_alloc_stripes
), GFP_NOFS
);
5245 atomic_set(&bbio
->error
, 0);
5247 if (rw
& REQ_DISCARD
) {
5249 int sub_stripes
= 0;
5250 u64 stripes_per_dev
= 0;
5251 u32 remaining_stripes
= 0;
5252 u32 last_stripe
= 0;
5255 (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID10
)) {
5256 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5259 sub_stripes
= map
->sub_stripes
;
5261 factor
= map
->num_stripes
/ sub_stripes
;
5262 stripes_per_dev
= div_u64_rem(stripe_nr_end
-
5265 &remaining_stripes
);
5266 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
5267 last_stripe
*= sub_stripes
;
5270 for (i
= 0; i
< num_stripes
; i
++) {
5271 bbio
->stripes
[i
].physical
=
5272 map
->stripes
[stripe_index
].physical
+
5273 stripe_offset
+ stripe_nr
* map
->stripe_len
;
5274 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
5276 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5277 BTRFS_BLOCK_GROUP_RAID10
)) {
5278 bbio
->stripes
[i
].length
= stripes_per_dev
*
5281 if (i
/ sub_stripes
< remaining_stripes
)
5282 bbio
->stripes
[i
].length
+=
5286 * Special for the first stripe and
5289 * |-------|...|-------|
5293 if (i
< sub_stripes
)
5294 bbio
->stripes
[i
].length
-=
5297 if (stripe_index
>= last_stripe
&&
5298 stripe_index
<= (last_stripe
+
5300 bbio
->stripes
[i
].length
-=
5303 if (i
== sub_stripes
- 1)
5306 bbio
->stripes
[i
].length
= *length
;
5309 if (stripe_index
== map
->num_stripes
) {
5310 /* This could only happen for RAID0/10 */
5316 for (i
= 0; i
< num_stripes
; i
++) {
5317 bbio
->stripes
[i
].physical
=
5318 map
->stripes
[stripe_index
].physical
+
5320 stripe_nr
* map
->stripe_len
;
5321 bbio
->stripes
[i
].dev
=
5322 map
->stripes
[stripe_index
].dev
;
5327 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
))
5328 max_errors
= btrfs_chunk_max_errors(map
);
5330 if (dev_replace_is_ongoing
&& (rw
& (REQ_WRITE
| REQ_DISCARD
)) &&
5331 dev_replace
->tgtdev
!= NULL
) {
5332 int index_where_to_add
;
5333 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5336 * duplicate the write operations while the dev replace
5337 * procedure is running. Since the copying of the old disk
5338 * to the new disk takes place at run time while the
5339 * filesystem is mounted writable, the regular write
5340 * operations to the old disk have to be duplicated to go
5341 * to the new disk as well.
5342 * Note that device->missing is handled by the caller, and
5343 * that the write to the old disk is already set up in the
5346 index_where_to_add
= num_stripes
;
5347 for (i
= 0; i
< num_stripes
; i
++) {
5348 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5349 /* write to new disk, too */
5350 struct btrfs_bio_stripe
*new =
5351 bbio
->stripes
+ index_where_to_add
;
5352 struct btrfs_bio_stripe
*old
=
5355 new->physical
= old
->physical
;
5356 new->length
= old
->length
;
5357 new->dev
= dev_replace
->tgtdev
;
5358 index_where_to_add
++;
5362 num_stripes
= index_where_to_add
;
5363 } else if (dev_replace_is_ongoing
&& (rw
& REQ_GET_READ_MIRRORS
) &&
5364 dev_replace
->tgtdev
!= NULL
) {
5365 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5366 int index_srcdev
= 0;
5368 u64 physical_of_found
= 0;
5371 * During the dev-replace procedure, the target drive can
5372 * also be used to read data in case it is needed to repair
5373 * a corrupt block elsewhere. This is possible if the
5374 * requested area is left of the left cursor. In this area,
5375 * the target drive is a full copy of the source drive.
5377 for (i
= 0; i
< num_stripes
; i
++) {
5378 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5380 * In case of DUP, in order to keep it
5381 * simple, only add the mirror with the
5382 * lowest physical address
5385 physical_of_found
<=
5386 bbio
->stripes
[i
].physical
)
5390 physical_of_found
= bbio
->stripes
[i
].physical
;
5394 u64 length
= map
->stripe_len
;
5396 if (physical_of_found
+ length
<=
5397 dev_replace
->cursor_left
) {
5398 struct btrfs_bio_stripe
*tgtdev_stripe
=
5399 bbio
->stripes
+ num_stripes
;
5401 tgtdev_stripe
->physical
= physical_of_found
;
5402 tgtdev_stripe
->length
=
5403 bbio
->stripes
[index_srcdev
].length
;
5404 tgtdev_stripe
->dev
= dev_replace
->tgtdev
;
5412 bbio
->num_stripes
= num_stripes
;
5413 bbio
->max_errors
= max_errors
;
5414 bbio
->mirror_num
= mirror_num
;
5417 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5418 * mirror_num == num_stripes + 1 && dev_replace target drive is
5419 * available as a mirror
5421 if (patch_the_first_stripe_for_dev_replace
&& num_stripes
> 0) {
5422 WARN_ON(num_stripes
> 1);
5423 bbio
->stripes
[0].dev
= dev_replace
->tgtdev
;
5424 bbio
->stripes
[0].physical
= physical_to_patch_in_first_stripe
;
5425 bbio
->mirror_num
= map
->num_stripes
+ 1;
5428 sort_parity_stripes(bbio
, raid_map
);
5429 *raid_map_ret
= raid_map
;
5432 if (dev_replace_is_ongoing
)
5433 btrfs_dev_replace_unlock(dev_replace
);
5434 free_extent_map(em
);
5438 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
5439 u64 logical
, u64
*length
,
5440 struct btrfs_bio
**bbio_ret
, int mirror_num
)
5442 return __btrfs_map_block(fs_info
, rw
, logical
, length
, bbio_ret
,
5446 int btrfs_rmap_block(struct btrfs_mapping_tree
*map_tree
,
5447 u64 chunk_start
, u64 physical
, u64 devid
,
5448 u64
**logical
, int *naddrs
, int *stripe_len
)
5450 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5451 struct extent_map
*em
;
5452 struct map_lookup
*map
;
5460 read_lock(&em_tree
->lock
);
5461 em
= lookup_extent_mapping(em_tree
, chunk_start
, 1);
5462 read_unlock(&em_tree
->lock
);
5465 printk(KERN_ERR
"BTRFS: couldn't find em for chunk %Lu\n",
5470 if (em
->start
!= chunk_start
) {
5471 printk(KERN_ERR
"BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5472 em
->start
, chunk_start
);
5473 free_extent_map(em
);
5476 map
= (struct map_lookup
*)em
->bdev
;
5479 rmap_len
= map
->stripe_len
;
5481 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5482 do_div(length
, map
->num_stripes
/ map
->sub_stripes
);
5483 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5484 do_div(length
, map
->num_stripes
);
5485 else if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
5486 BTRFS_BLOCK_GROUP_RAID6
)) {
5487 do_div(length
, nr_data_stripes(map
));
5488 rmap_len
= map
->stripe_len
* nr_data_stripes(map
);
5491 buf
= kzalloc(sizeof(u64
) * map
->num_stripes
, GFP_NOFS
);
5492 BUG_ON(!buf
); /* -ENOMEM */
5494 for (i
= 0; i
< map
->num_stripes
; i
++) {
5495 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
5497 if (map
->stripes
[i
].physical
> physical
||
5498 map
->stripes
[i
].physical
+ length
<= physical
)
5501 stripe_nr
= physical
- map
->stripes
[i
].physical
;
5502 do_div(stripe_nr
, map
->stripe_len
);
5504 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5505 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5506 do_div(stripe_nr
, map
->sub_stripes
);
5507 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5508 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5509 } /* else if RAID[56], multiply by nr_data_stripes().
5510 * Alternatively, just use rmap_len below instead of
5511 * map->stripe_len */
5513 bytenr
= chunk_start
+ stripe_nr
* rmap_len
;
5514 WARN_ON(nr
>= map
->num_stripes
);
5515 for (j
= 0; j
< nr
; j
++) {
5516 if (buf
[j
] == bytenr
)
5520 WARN_ON(nr
>= map
->num_stripes
);
5527 *stripe_len
= rmap_len
;
5529 free_extent_map(em
);
5533 static inline void btrfs_end_bbio(struct btrfs_bio
*bbio
, struct bio
*bio
, int err
)
5535 if (likely(bbio
->flags
& BTRFS_BIO_ORIG_BIO_SUBMITTED
))
5536 bio_endio_nodec(bio
, err
);
5538 bio_endio(bio
, err
);
5542 static void btrfs_end_bio(struct bio
*bio
, int err
)
5544 struct btrfs_bio
*bbio
= bio
->bi_private
;
5545 struct btrfs_device
*dev
= bbio
->stripes
[0].dev
;
5546 int is_orig_bio
= 0;
5549 atomic_inc(&bbio
->error
);
5550 if (err
== -EIO
|| err
== -EREMOTEIO
) {
5551 unsigned int stripe_index
=
5552 btrfs_io_bio(bio
)->stripe_index
;
5554 BUG_ON(stripe_index
>= bbio
->num_stripes
);
5555 dev
= bbio
->stripes
[stripe_index
].dev
;
5557 if (bio
->bi_rw
& WRITE
)
5558 btrfs_dev_stat_inc(dev
,
5559 BTRFS_DEV_STAT_WRITE_ERRS
);
5561 btrfs_dev_stat_inc(dev
,
5562 BTRFS_DEV_STAT_READ_ERRS
);
5563 if ((bio
->bi_rw
& WRITE_FLUSH
) == WRITE_FLUSH
)
5564 btrfs_dev_stat_inc(dev
,
5565 BTRFS_DEV_STAT_FLUSH_ERRS
);
5566 btrfs_dev_stat_print_on_error(dev
);
5571 if (bio
== bbio
->orig_bio
)
5574 btrfs_bio_counter_dec(bbio
->fs_info
);
5576 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
5579 bio
= bbio
->orig_bio
;
5582 bio
->bi_private
= bbio
->private;
5583 bio
->bi_end_io
= bbio
->end_io
;
5584 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
5585 /* only send an error to the higher layers if it is
5586 * beyond the tolerance of the btrfs bio
5588 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
5592 * this bio is actually up to date, we didn't
5593 * go over the max number of errors
5595 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5599 btrfs_end_bbio(bbio
, bio
, err
);
5600 } else if (!is_orig_bio
) {
5606 * see run_scheduled_bios for a description of why bios are collected for
5609 * This will add one bio to the pending list for a device and make sure
5610 * the work struct is scheduled.
5612 static noinline
void btrfs_schedule_bio(struct btrfs_root
*root
,
5613 struct btrfs_device
*device
,
5614 int rw
, struct bio
*bio
)
5616 int should_queue
= 1;
5617 struct btrfs_pending_bios
*pending_bios
;
5619 if (device
->missing
|| !device
->bdev
) {
5620 bio_endio(bio
, -EIO
);
5624 /* don't bother with additional async steps for reads, right now */
5625 if (!(rw
& REQ_WRITE
)) {
5627 btrfsic_submit_bio(rw
, bio
);
5633 * nr_async_bios allows us to reliably return congestion to the
5634 * higher layers. Otherwise, the async bio makes it appear we have
5635 * made progress against dirty pages when we've really just put it
5636 * on a queue for later
5638 atomic_inc(&root
->fs_info
->nr_async_bios
);
5639 WARN_ON(bio
->bi_next
);
5640 bio
->bi_next
= NULL
;
5643 spin_lock(&device
->io_lock
);
5644 if (bio
->bi_rw
& REQ_SYNC
)
5645 pending_bios
= &device
->pending_sync_bios
;
5647 pending_bios
= &device
->pending_bios
;
5649 if (pending_bios
->tail
)
5650 pending_bios
->tail
->bi_next
= bio
;
5652 pending_bios
->tail
= bio
;
5653 if (!pending_bios
->head
)
5654 pending_bios
->head
= bio
;
5655 if (device
->running_pending
)
5658 spin_unlock(&device
->io_lock
);
5661 btrfs_queue_work(root
->fs_info
->submit_workers
,
5665 static int bio_size_ok(struct block_device
*bdev
, struct bio
*bio
,
5668 struct bio_vec
*prev
;
5669 struct request_queue
*q
= bdev_get_queue(bdev
);
5670 unsigned int max_sectors
= queue_max_sectors(q
);
5671 struct bvec_merge_data bvm
= {
5673 .bi_sector
= sector
,
5674 .bi_rw
= bio
->bi_rw
,
5677 if (WARN_ON(bio
->bi_vcnt
== 0))
5680 prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
5681 if (bio_sectors(bio
) > max_sectors
)
5684 if (!q
->merge_bvec_fn
)
5687 bvm
.bi_size
= bio
->bi_iter
.bi_size
- prev
->bv_len
;
5688 if (q
->merge_bvec_fn(q
, &bvm
, prev
) < prev
->bv_len
)
5693 static void submit_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
5694 struct bio
*bio
, u64 physical
, int dev_nr
,
5697 struct btrfs_device
*dev
= bbio
->stripes
[dev_nr
].dev
;
5699 bio
->bi_private
= bbio
;
5700 btrfs_io_bio(bio
)->stripe_index
= dev_nr
;
5701 bio
->bi_end_io
= btrfs_end_bio
;
5702 bio
->bi_iter
.bi_sector
= physical
>> 9;
5705 struct rcu_string
*name
;
5708 name
= rcu_dereference(dev
->name
);
5709 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5710 "(%s id %llu), size=%u\n", rw
,
5711 (u64
)bio
->bi_iter
.bi_sector
, (u_long
)dev
->bdev
->bd_dev
,
5712 name
->str
, dev
->devid
, bio
->bi_iter
.bi_size
);
5716 bio
->bi_bdev
= dev
->bdev
;
5718 btrfs_bio_counter_inc_noblocked(root
->fs_info
);
5721 btrfs_schedule_bio(root
, dev
, rw
, bio
);
5723 btrfsic_submit_bio(rw
, bio
);
5726 static int breakup_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
5727 struct bio
*first_bio
, struct btrfs_device
*dev
,
5728 int dev_nr
, int rw
, int async
)
5730 struct bio_vec
*bvec
= first_bio
->bi_io_vec
;
5732 int nr_vecs
= bio_get_nr_vecs(dev
->bdev
);
5733 u64 physical
= bbio
->stripes
[dev_nr
].physical
;
5736 bio
= btrfs_bio_alloc(dev
->bdev
, physical
>> 9, nr_vecs
, GFP_NOFS
);
5740 while (bvec
<= (first_bio
->bi_io_vec
+ first_bio
->bi_vcnt
- 1)) {
5741 if (bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
5742 bvec
->bv_offset
) < bvec
->bv_len
) {
5743 u64 len
= bio
->bi_iter
.bi_size
;
5745 atomic_inc(&bbio
->stripes_pending
);
5746 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
,
5754 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
, rw
, async
);
5758 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
5760 atomic_inc(&bbio
->error
);
5761 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
5762 /* Shoud be the original bio. */
5763 WARN_ON(bio
!= bbio
->orig_bio
);
5765 bio
->bi_private
= bbio
->private;
5766 bio
->bi_end_io
= bbio
->end_io
;
5767 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
5768 bio
->bi_iter
.bi_sector
= logical
>> 9;
5770 btrfs_end_bbio(bbio
, bio
, -EIO
);
5774 int btrfs_map_bio(struct btrfs_root
*root
, int rw
, struct bio
*bio
,
5775 int mirror_num
, int async_submit
)
5777 struct btrfs_device
*dev
;
5778 struct bio
*first_bio
= bio
;
5779 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
5782 u64
*raid_map
= NULL
;
5786 struct btrfs_bio
*bbio
= NULL
;
5788 length
= bio
->bi_iter
.bi_size
;
5789 map_length
= length
;
5791 btrfs_bio_counter_inc_blocked(root
->fs_info
);
5792 ret
= __btrfs_map_block(root
->fs_info
, rw
, logical
, &map_length
, &bbio
,
5793 mirror_num
, &raid_map
);
5795 btrfs_bio_counter_dec(root
->fs_info
);
5799 total_devs
= bbio
->num_stripes
;
5800 bbio
->orig_bio
= first_bio
;
5801 bbio
->private = first_bio
->bi_private
;
5802 bbio
->end_io
= first_bio
->bi_end_io
;
5803 bbio
->fs_info
= root
->fs_info
;
5804 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
5807 /* In this case, map_length has been set to the length of
5808 a single stripe; not the whole write */
5810 ret
= raid56_parity_write(root
, bio
, bbio
,
5811 raid_map
, map_length
);
5813 ret
= raid56_parity_recover(root
, bio
, bbio
,
5814 raid_map
, map_length
,
5818 * FIXME, replace dosen't support raid56 yet, please fix
5821 btrfs_bio_counter_dec(root
->fs_info
);
5825 if (map_length
< length
) {
5826 btrfs_crit(root
->fs_info
, "mapping failed logical %llu bio len %llu len %llu",
5827 logical
, length
, map_length
);
5831 while (dev_nr
< total_devs
) {
5832 dev
= bbio
->stripes
[dev_nr
].dev
;
5833 if (!dev
|| !dev
->bdev
|| (rw
& WRITE
&& !dev
->writeable
)) {
5834 bbio_error(bbio
, first_bio
, logical
);
5840 * Check and see if we're ok with this bio based on it's size
5841 * and offset with the given device.
5843 if (!bio_size_ok(dev
->bdev
, first_bio
,
5844 bbio
->stripes
[dev_nr
].physical
>> 9)) {
5845 ret
= breakup_stripe_bio(root
, bbio
, first_bio
, dev
,
5846 dev_nr
, rw
, async_submit
);
5852 if (dev_nr
< total_devs
- 1) {
5853 bio
= btrfs_bio_clone(first_bio
, GFP_NOFS
);
5854 BUG_ON(!bio
); /* -ENOMEM */
5857 bbio
->flags
|= BTRFS_BIO_ORIG_BIO_SUBMITTED
;
5860 submit_stripe_bio(root
, bbio
, bio
,
5861 bbio
->stripes
[dev_nr
].physical
, dev_nr
, rw
,
5865 btrfs_bio_counter_dec(root
->fs_info
);
5869 struct btrfs_device
*btrfs_find_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
5872 struct btrfs_device
*device
;
5873 struct btrfs_fs_devices
*cur_devices
;
5875 cur_devices
= fs_info
->fs_devices
;
5876 while (cur_devices
) {
5878 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
5879 device
= __find_device(&cur_devices
->devices
,
5884 cur_devices
= cur_devices
->seed
;
5889 static struct btrfs_device
*add_missing_dev(struct btrfs_root
*root
,
5890 struct btrfs_fs_devices
*fs_devices
,
5891 u64 devid
, u8
*dev_uuid
)
5893 struct btrfs_device
*device
;
5895 device
= btrfs_alloc_device(NULL
, &devid
, dev_uuid
);
5899 list_add(&device
->dev_list
, &fs_devices
->devices
);
5900 device
->fs_devices
= fs_devices
;
5901 fs_devices
->num_devices
++;
5903 device
->missing
= 1;
5904 fs_devices
->missing_devices
++;
5910 * btrfs_alloc_device - allocate struct btrfs_device
5911 * @fs_info: used only for generating a new devid, can be NULL if
5912 * devid is provided (i.e. @devid != NULL).
5913 * @devid: a pointer to devid for this device. If NULL a new devid
5915 * @uuid: a pointer to UUID for this device. If NULL a new UUID
5918 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5919 * on error. Returned struct is not linked onto any lists and can be
5920 * destroyed with kfree() right away.
5922 struct btrfs_device
*btrfs_alloc_device(struct btrfs_fs_info
*fs_info
,
5926 struct btrfs_device
*dev
;
5929 if (WARN_ON(!devid
&& !fs_info
))
5930 return ERR_PTR(-EINVAL
);
5932 dev
= __alloc_device();
5941 ret
= find_next_devid(fs_info
, &tmp
);
5944 return ERR_PTR(ret
);
5950 memcpy(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
);
5952 generate_random_uuid(dev
->uuid
);
5954 btrfs_init_work(&dev
->work
, btrfs_submit_helper
,
5955 pending_bios_fn
, NULL
, NULL
);
5960 static int read_one_chunk(struct btrfs_root
*root
, struct btrfs_key
*key
,
5961 struct extent_buffer
*leaf
,
5962 struct btrfs_chunk
*chunk
)
5964 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
5965 struct map_lookup
*map
;
5966 struct extent_map
*em
;
5970 u8 uuid
[BTRFS_UUID_SIZE
];
5975 logical
= key
->offset
;
5976 length
= btrfs_chunk_length(leaf
, chunk
);
5978 read_lock(&map_tree
->map_tree
.lock
);
5979 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
5980 read_unlock(&map_tree
->map_tree
.lock
);
5982 /* already mapped? */
5983 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
5984 free_extent_map(em
);
5987 free_extent_map(em
);
5990 em
= alloc_extent_map();
5993 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
5994 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
5996 free_extent_map(em
);
6000 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
6001 em
->bdev
= (struct block_device
*)map
;
6002 em
->start
= logical
;
6005 em
->block_start
= 0;
6006 em
->block_len
= em
->len
;
6008 map
->num_stripes
= num_stripes
;
6009 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
6010 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
6011 map
->sector_size
= btrfs_chunk_sector_size(leaf
, chunk
);
6012 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6013 map
->type
= btrfs_chunk_type(leaf
, chunk
);
6014 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6015 for (i
= 0; i
< num_stripes
; i
++) {
6016 map
->stripes
[i
].physical
=
6017 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
6018 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
6019 read_extent_buffer(leaf
, uuid
, (unsigned long)
6020 btrfs_stripe_dev_uuid_nr(chunk
, i
),
6022 map
->stripes
[i
].dev
= btrfs_find_device(root
->fs_info
, devid
,
6024 if (!map
->stripes
[i
].dev
&& !btrfs_test_opt(root
, DEGRADED
)) {
6025 free_extent_map(em
);
6028 if (!map
->stripes
[i
].dev
) {
6029 map
->stripes
[i
].dev
=
6030 add_missing_dev(root
, root
->fs_info
->fs_devices
,
6032 if (!map
->stripes
[i
].dev
) {
6033 free_extent_map(em
);
6037 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
6040 write_lock(&map_tree
->map_tree
.lock
);
6041 ret
= add_extent_mapping(&map_tree
->map_tree
, em
, 0);
6042 write_unlock(&map_tree
->map_tree
.lock
);
6043 BUG_ON(ret
); /* Tree corruption */
6044 free_extent_map(em
);
6049 static void fill_device_from_item(struct extent_buffer
*leaf
,
6050 struct btrfs_dev_item
*dev_item
,
6051 struct btrfs_device
*device
)
6055 device
->devid
= btrfs_device_id(leaf
, dev_item
);
6056 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
6057 device
->total_bytes
= device
->disk_total_bytes
;
6058 device
->commit_total_bytes
= device
->disk_total_bytes
;
6059 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
6060 device
->commit_bytes_used
= device
->bytes_used
;
6061 device
->type
= btrfs_device_type(leaf
, dev_item
);
6062 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
6063 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
6064 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
6065 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
6066 device
->is_tgtdev_for_dev_replace
= 0;
6068 ptr
= btrfs_device_uuid(dev_item
);
6069 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
6072 static struct btrfs_fs_devices
*open_seed_devices(struct btrfs_root
*root
,
6075 struct btrfs_fs_devices
*fs_devices
;
6078 BUG_ON(!mutex_is_locked(&uuid_mutex
));
6080 fs_devices
= root
->fs_info
->fs_devices
->seed
;
6081 while (fs_devices
) {
6082 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
))
6085 fs_devices
= fs_devices
->seed
;
6088 fs_devices
= find_fsid(fsid
);
6090 if (!btrfs_test_opt(root
, DEGRADED
))
6091 return ERR_PTR(-ENOENT
);
6093 fs_devices
= alloc_fs_devices(fsid
);
6094 if (IS_ERR(fs_devices
))
6097 fs_devices
->seeding
= 1;
6098 fs_devices
->opened
= 1;
6102 fs_devices
= clone_fs_devices(fs_devices
);
6103 if (IS_ERR(fs_devices
))
6106 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
6107 root
->fs_info
->bdev_holder
);
6109 free_fs_devices(fs_devices
);
6110 fs_devices
= ERR_PTR(ret
);
6114 if (!fs_devices
->seeding
) {
6115 __btrfs_close_devices(fs_devices
);
6116 free_fs_devices(fs_devices
);
6117 fs_devices
= ERR_PTR(-EINVAL
);
6121 fs_devices
->seed
= root
->fs_info
->fs_devices
->seed
;
6122 root
->fs_info
->fs_devices
->seed
= fs_devices
;
6127 static int read_one_dev(struct btrfs_root
*root
,
6128 struct extent_buffer
*leaf
,
6129 struct btrfs_dev_item
*dev_item
)
6131 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
6132 struct btrfs_device
*device
;
6135 u8 fs_uuid
[BTRFS_UUID_SIZE
];
6136 u8 dev_uuid
[BTRFS_UUID_SIZE
];
6138 devid
= btrfs_device_id(leaf
, dev_item
);
6139 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
6141 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
6144 if (memcmp(fs_uuid
, root
->fs_info
->fsid
, BTRFS_UUID_SIZE
)) {
6145 fs_devices
= open_seed_devices(root
, fs_uuid
);
6146 if (IS_ERR(fs_devices
))
6147 return PTR_ERR(fs_devices
);
6150 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
, fs_uuid
);
6152 if (!btrfs_test_opt(root
, DEGRADED
))
6155 btrfs_warn(root
->fs_info
, "devid %llu missing", devid
);
6156 device
= add_missing_dev(root
, fs_devices
, devid
, dev_uuid
);
6160 if (!device
->bdev
&& !btrfs_test_opt(root
, DEGRADED
))
6163 if(!device
->bdev
&& !device
->missing
) {
6165 * this happens when a device that was properly setup
6166 * in the device info lists suddenly goes bad.
6167 * device->bdev is NULL, and so we have to set
6168 * device->missing to one here
6170 device
->fs_devices
->missing_devices
++;
6171 device
->missing
= 1;
6174 /* Move the device to its own fs_devices */
6175 if (device
->fs_devices
!= fs_devices
) {
6176 ASSERT(device
->missing
);
6178 list_move(&device
->dev_list
, &fs_devices
->devices
);
6179 device
->fs_devices
->num_devices
--;
6180 fs_devices
->num_devices
++;
6182 device
->fs_devices
->missing_devices
--;
6183 fs_devices
->missing_devices
++;
6185 device
->fs_devices
= fs_devices
;
6189 if (device
->fs_devices
!= root
->fs_info
->fs_devices
) {
6190 BUG_ON(device
->writeable
);
6191 if (device
->generation
!=
6192 btrfs_device_generation(leaf
, dev_item
))
6196 fill_device_from_item(leaf
, dev_item
, device
);
6197 device
->in_fs_metadata
= 1;
6198 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
6199 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
6200 spin_lock(&root
->fs_info
->free_chunk_lock
);
6201 root
->fs_info
->free_chunk_space
+= device
->total_bytes
-
6203 spin_unlock(&root
->fs_info
->free_chunk_lock
);
6209 int btrfs_read_sys_array(struct btrfs_root
*root
)
6211 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
6212 struct extent_buffer
*sb
;
6213 struct btrfs_disk_key
*disk_key
;
6214 struct btrfs_chunk
*chunk
;
6216 unsigned long sb_ptr
;
6222 struct btrfs_key key
;
6224 sb
= btrfs_find_create_tree_block(root
, BTRFS_SUPER_INFO_OFFSET
,
6225 BTRFS_SUPER_INFO_SIZE
);
6228 btrfs_set_buffer_uptodate(sb
);
6229 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
6231 * The sb extent buffer is artifical and just used to read the system array.
6232 * btrfs_set_buffer_uptodate() call does not properly mark all it's
6233 * pages up-to-date when the page is larger: extent does not cover the
6234 * whole page and consequently check_page_uptodate does not find all
6235 * the page's extents up-to-date (the hole beyond sb),
6236 * write_extent_buffer then triggers a WARN_ON.
6238 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6239 * but sb spans only this function. Add an explicit SetPageUptodate call
6240 * to silence the warning eg. on PowerPC 64.
6242 if (PAGE_CACHE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
6243 SetPageUptodate(sb
->pages
[0]);
6245 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
6246 array_size
= btrfs_super_sys_array_size(super_copy
);
6248 ptr
= super_copy
->sys_chunk_array
;
6249 sb_ptr
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
6252 while (cur
< array_size
) {
6253 disk_key
= (struct btrfs_disk_key
*)ptr
;
6254 btrfs_disk_key_to_cpu(&key
, disk_key
);
6256 len
= sizeof(*disk_key
); ptr
+= len
;
6260 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6261 chunk
= (struct btrfs_chunk
*)sb_ptr
;
6262 ret
= read_one_chunk(root
, &key
, sb
, chunk
);
6265 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
6266 len
= btrfs_chunk_item_size(num_stripes
);
6275 free_extent_buffer(sb
);
6279 int btrfs_read_chunk_tree(struct btrfs_root
*root
)
6281 struct btrfs_path
*path
;
6282 struct extent_buffer
*leaf
;
6283 struct btrfs_key key
;
6284 struct btrfs_key found_key
;
6288 root
= root
->fs_info
->chunk_root
;
6290 path
= btrfs_alloc_path();
6294 mutex_lock(&uuid_mutex
);
6298 * Read all device items, and then all the chunk items. All
6299 * device items are found before any chunk item (their object id
6300 * is smaller than the lowest possible object id for a chunk
6301 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6303 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
6306 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6310 leaf
= path
->nodes
[0];
6311 slot
= path
->slots
[0];
6312 if (slot
>= btrfs_header_nritems(leaf
)) {
6313 ret
= btrfs_next_leaf(root
, path
);
6320 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6321 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
6322 struct btrfs_dev_item
*dev_item
;
6323 dev_item
= btrfs_item_ptr(leaf
, slot
,
6324 struct btrfs_dev_item
);
6325 ret
= read_one_dev(root
, leaf
, dev_item
);
6328 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6329 struct btrfs_chunk
*chunk
;
6330 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
6331 ret
= read_one_chunk(root
, &found_key
, leaf
, chunk
);
6339 unlock_chunks(root
);
6340 mutex_unlock(&uuid_mutex
);
6342 btrfs_free_path(path
);
6346 void btrfs_init_devices_late(struct btrfs_fs_info
*fs_info
)
6348 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6349 struct btrfs_device
*device
;
6351 while (fs_devices
) {
6352 mutex_lock(&fs_devices
->device_list_mutex
);
6353 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
)
6354 device
->dev_root
= fs_info
->dev_root
;
6355 mutex_unlock(&fs_devices
->device_list_mutex
);
6357 fs_devices
= fs_devices
->seed
;
6361 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
6365 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6366 btrfs_dev_stat_reset(dev
, i
);
6369 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
6371 struct btrfs_key key
;
6372 struct btrfs_key found_key
;
6373 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6374 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6375 struct extent_buffer
*eb
;
6378 struct btrfs_device
*device
;
6379 struct btrfs_path
*path
= NULL
;
6382 path
= btrfs_alloc_path();
6388 mutex_lock(&fs_devices
->device_list_mutex
);
6389 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6391 struct btrfs_dev_stats_item
*ptr
;
6394 key
.type
= BTRFS_DEV_STATS_KEY
;
6395 key
.offset
= device
->devid
;
6396 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
6398 __btrfs_reset_dev_stats(device
);
6399 device
->dev_stats_valid
= 1;
6400 btrfs_release_path(path
);
6403 slot
= path
->slots
[0];
6404 eb
= path
->nodes
[0];
6405 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
6406 item_size
= btrfs_item_size_nr(eb
, slot
);
6408 ptr
= btrfs_item_ptr(eb
, slot
,
6409 struct btrfs_dev_stats_item
);
6411 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6412 if (item_size
>= (1 + i
) * sizeof(__le64
))
6413 btrfs_dev_stat_set(device
, i
,
6414 btrfs_dev_stats_value(eb
, ptr
, i
));
6416 btrfs_dev_stat_reset(device
, i
);
6419 device
->dev_stats_valid
= 1;
6420 btrfs_dev_stat_print_on_load(device
);
6421 btrfs_release_path(path
);
6423 mutex_unlock(&fs_devices
->device_list_mutex
);
6426 btrfs_free_path(path
);
6427 return ret
< 0 ? ret
: 0;
6430 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
6431 struct btrfs_root
*dev_root
,
6432 struct btrfs_device
*device
)
6434 struct btrfs_path
*path
;
6435 struct btrfs_key key
;
6436 struct extent_buffer
*eb
;
6437 struct btrfs_dev_stats_item
*ptr
;
6442 key
.type
= BTRFS_DEV_STATS_KEY
;
6443 key
.offset
= device
->devid
;
6445 path
= btrfs_alloc_path();
6447 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
6449 printk_in_rcu(KERN_WARNING
"BTRFS: "
6450 "error %d while searching for dev_stats item for device %s!\n",
6451 ret
, rcu_str_deref(device
->name
));
6456 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
6457 /* need to delete old one and insert a new one */
6458 ret
= btrfs_del_item(trans
, dev_root
, path
);
6460 printk_in_rcu(KERN_WARNING
"BTRFS: "
6461 "delete too small dev_stats item for device %s failed %d!\n",
6462 rcu_str_deref(device
->name
), ret
);
6469 /* need to insert a new item */
6470 btrfs_release_path(path
);
6471 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
6472 &key
, sizeof(*ptr
));
6474 printk_in_rcu(KERN_WARNING
"BTRFS: "
6475 "insert dev_stats item for device %s failed %d!\n",
6476 rcu_str_deref(device
->name
), ret
);
6481 eb
= path
->nodes
[0];
6482 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
6483 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6484 btrfs_set_dev_stats_value(eb
, ptr
, i
,
6485 btrfs_dev_stat_read(device
, i
));
6486 btrfs_mark_buffer_dirty(eb
);
6489 btrfs_free_path(path
);
6494 * called from commit_transaction. Writes all changed device stats to disk.
6496 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
6497 struct btrfs_fs_info
*fs_info
)
6499 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6500 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6501 struct btrfs_device
*device
;
6505 mutex_lock(&fs_devices
->device_list_mutex
);
6506 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6507 if (!device
->dev_stats_valid
|| !btrfs_dev_stats_dirty(device
))
6510 stats_cnt
= atomic_read(&device
->dev_stats_ccnt
);
6511 ret
= update_dev_stat_item(trans
, dev_root
, device
);
6513 atomic_sub(stats_cnt
, &device
->dev_stats_ccnt
);
6515 mutex_unlock(&fs_devices
->device_list_mutex
);
6520 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
6522 btrfs_dev_stat_inc(dev
, index
);
6523 btrfs_dev_stat_print_on_error(dev
);
6526 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
6528 if (!dev
->dev_stats_valid
)
6530 printk_ratelimited_in_rcu(KERN_ERR
"BTRFS: "
6531 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6532 rcu_str_deref(dev
->name
),
6533 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
6534 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
6535 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
6536 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
6537 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
6540 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
6544 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6545 if (btrfs_dev_stat_read(dev
, i
) != 0)
6547 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
6548 return; /* all values == 0, suppress message */
6550 printk_in_rcu(KERN_INFO
"BTRFS: "
6551 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6552 rcu_str_deref(dev
->name
),
6553 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
6554 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
6555 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
6556 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
6557 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
6560 int btrfs_get_dev_stats(struct btrfs_root
*root
,
6561 struct btrfs_ioctl_get_dev_stats
*stats
)
6563 struct btrfs_device
*dev
;
6564 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
6567 mutex_lock(&fs_devices
->device_list_mutex
);
6568 dev
= btrfs_find_device(root
->fs_info
, stats
->devid
, NULL
, NULL
);
6569 mutex_unlock(&fs_devices
->device_list_mutex
);
6572 btrfs_warn(root
->fs_info
, "get dev_stats failed, device not found");
6574 } else if (!dev
->dev_stats_valid
) {
6575 btrfs_warn(root
->fs_info
, "get dev_stats failed, not yet valid");
6577 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
6578 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6579 if (stats
->nr_items
> i
)
6581 btrfs_dev_stat_read_and_reset(dev
, i
);
6583 btrfs_dev_stat_reset(dev
, i
);
6586 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6587 if (stats
->nr_items
> i
)
6588 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
6590 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
6591 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
6595 int btrfs_scratch_superblock(struct btrfs_device
*device
)
6597 struct buffer_head
*bh
;
6598 struct btrfs_super_block
*disk_super
;
6600 bh
= btrfs_read_dev_super(device
->bdev
);
6603 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
6605 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
6606 set_buffer_dirty(bh
);
6607 sync_dirty_buffer(bh
);
6614 * Update the size of all devices, which is used for writing out the
6617 void btrfs_update_commit_device_size(struct btrfs_fs_info
*fs_info
)
6619 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6620 struct btrfs_device
*curr
, *next
;
6622 if (list_empty(&fs_devices
->resized_devices
))
6625 mutex_lock(&fs_devices
->device_list_mutex
);
6626 lock_chunks(fs_info
->dev_root
);
6627 list_for_each_entry_safe(curr
, next
, &fs_devices
->resized_devices
,
6629 list_del_init(&curr
->resized_list
);
6630 curr
->commit_total_bytes
= curr
->disk_total_bytes
;
6632 unlock_chunks(fs_info
->dev_root
);
6633 mutex_unlock(&fs_devices
->device_list_mutex
);
6636 /* Must be invoked during the transaction commit */
6637 void btrfs_update_commit_device_bytes_used(struct btrfs_root
*root
,
6638 struct btrfs_transaction
*transaction
)
6640 struct extent_map
*em
;
6641 struct map_lookup
*map
;
6642 struct btrfs_device
*dev
;
6645 if (list_empty(&transaction
->pending_chunks
))
6648 /* In order to kick the device replace finish process */
6650 list_for_each_entry(em
, &transaction
->pending_chunks
, list
) {
6651 map
= (struct map_lookup
*)em
->bdev
;
6653 for (i
= 0; i
< map
->num_stripes
; i
++) {
6654 dev
= map
->stripes
[i
].dev
;
6655 dev
->commit_bytes_used
= dev
->bytes_used
;
6658 unlock_chunks(root
);