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_remove_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
--;
1834 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info
*fs_info
,
1835 struct btrfs_device
*srcdev
)
1837 struct btrfs_fs_devices
*fs_devices
= srcdev
->fs_devices
;
1839 call_rcu(&srcdev
->rcu
, free_device
);
1842 * unless fs_devices is seed fs, num_devices shouldn't go
1845 BUG_ON(!fs_devices
->num_devices
&& !fs_devices
->seeding
);
1847 /* if this is no devs we rather delete the fs_devices */
1848 if (!fs_devices
->num_devices
) {
1849 struct btrfs_fs_devices
*tmp_fs_devices
;
1851 tmp_fs_devices
= fs_info
->fs_devices
;
1852 while (tmp_fs_devices
) {
1853 if (tmp_fs_devices
->seed
== fs_devices
) {
1854 tmp_fs_devices
->seed
= fs_devices
->seed
;
1857 tmp_fs_devices
= tmp_fs_devices
->seed
;
1859 fs_devices
->seed
= NULL
;
1860 __btrfs_close_devices(fs_devices
);
1861 free_fs_devices(fs_devices
);
1865 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
1866 struct btrfs_device
*tgtdev
)
1868 struct btrfs_device
*next_device
;
1870 mutex_lock(&uuid_mutex
);
1872 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
1874 btrfs_scratch_superblock(tgtdev
);
1875 fs_info
->fs_devices
->open_devices
--;
1877 fs_info
->fs_devices
->num_devices
--;
1879 next_device
= list_entry(fs_info
->fs_devices
->devices
.next
,
1880 struct btrfs_device
, dev_list
);
1881 if (tgtdev
->bdev
== fs_info
->sb
->s_bdev
)
1882 fs_info
->sb
->s_bdev
= next_device
->bdev
;
1883 if (tgtdev
->bdev
== fs_info
->fs_devices
->latest_bdev
)
1884 fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1885 list_del_rcu(&tgtdev
->dev_list
);
1887 call_rcu(&tgtdev
->rcu
, free_device
);
1889 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
1890 mutex_unlock(&uuid_mutex
);
1893 static int btrfs_find_device_by_path(struct btrfs_root
*root
, char *device_path
,
1894 struct btrfs_device
**device
)
1897 struct btrfs_super_block
*disk_super
;
1900 struct block_device
*bdev
;
1901 struct buffer_head
*bh
;
1904 ret
= btrfs_get_bdev_and_sb(device_path
, FMODE_READ
,
1905 root
->fs_info
->bdev_holder
, 0, &bdev
, &bh
);
1908 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1909 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1910 dev_uuid
= disk_super
->dev_item
.uuid
;
1911 *device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1916 blkdev_put(bdev
, FMODE_READ
);
1920 int btrfs_find_device_missing_or_by_path(struct btrfs_root
*root
,
1922 struct btrfs_device
**device
)
1925 if (strcmp(device_path
, "missing") == 0) {
1926 struct list_head
*devices
;
1927 struct btrfs_device
*tmp
;
1929 devices
= &root
->fs_info
->fs_devices
->devices
;
1931 * It is safe to read the devices since the volume_mutex
1932 * is held by the caller.
1934 list_for_each_entry(tmp
, devices
, dev_list
) {
1935 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
1942 btrfs_err(root
->fs_info
, "no missing device found");
1948 return btrfs_find_device_by_path(root
, device_path
, device
);
1953 * does all the dirty work required for changing file system's UUID.
1955 static int btrfs_prepare_sprout(struct btrfs_root
*root
)
1957 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
1958 struct btrfs_fs_devices
*old_devices
;
1959 struct btrfs_fs_devices
*seed_devices
;
1960 struct btrfs_super_block
*disk_super
= root
->fs_info
->super_copy
;
1961 struct btrfs_device
*device
;
1964 BUG_ON(!mutex_is_locked(&uuid_mutex
));
1965 if (!fs_devices
->seeding
)
1968 seed_devices
= __alloc_fs_devices();
1969 if (IS_ERR(seed_devices
))
1970 return PTR_ERR(seed_devices
);
1972 old_devices
= clone_fs_devices(fs_devices
);
1973 if (IS_ERR(old_devices
)) {
1974 kfree(seed_devices
);
1975 return PTR_ERR(old_devices
);
1978 list_add(&old_devices
->list
, &fs_uuids
);
1980 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
1981 seed_devices
->opened
= 1;
1982 INIT_LIST_HEAD(&seed_devices
->devices
);
1983 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
1984 mutex_init(&seed_devices
->device_list_mutex
);
1986 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1987 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
1989 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
)
1990 device
->fs_devices
= seed_devices
;
1993 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
1994 unlock_chunks(root
);
1996 fs_devices
->seeding
= 0;
1997 fs_devices
->num_devices
= 0;
1998 fs_devices
->open_devices
= 0;
1999 fs_devices
->missing_devices
= 0;
2000 fs_devices
->rotating
= 0;
2001 fs_devices
->seed
= seed_devices
;
2003 generate_random_uuid(fs_devices
->fsid
);
2004 memcpy(root
->fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2005 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2006 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2008 super_flags
= btrfs_super_flags(disk_super
) &
2009 ~BTRFS_SUPER_FLAG_SEEDING
;
2010 btrfs_set_super_flags(disk_super
, super_flags
);
2016 * strore the expected generation for seed devices in device items.
2018 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
2019 struct btrfs_root
*root
)
2021 struct btrfs_path
*path
;
2022 struct extent_buffer
*leaf
;
2023 struct btrfs_dev_item
*dev_item
;
2024 struct btrfs_device
*device
;
2025 struct btrfs_key key
;
2026 u8 fs_uuid
[BTRFS_UUID_SIZE
];
2027 u8 dev_uuid
[BTRFS_UUID_SIZE
];
2031 path
= btrfs_alloc_path();
2035 root
= root
->fs_info
->chunk_root
;
2036 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2038 key
.type
= BTRFS_DEV_ITEM_KEY
;
2041 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2045 leaf
= path
->nodes
[0];
2047 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
2048 ret
= btrfs_next_leaf(root
, path
);
2053 leaf
= path
->nodes
[0];
2054 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2055 btrfs_release_path(path
);
2059 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2060 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
2061 key
.type
!= BTRFS_DEV_ITEM_KEY
)
2064 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2065 struct btrfs_dev_item
);
2066 devid
= btrfs_device_id(leaf
, dev_item
);
2067 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
2069 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
2071 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
2073 BUG_ON(!device
); /* Logic error */
2075 if (device
->fs_devices
->seeding
) {
2076 btrfs_set_device_generation(leaf
, dev_item
,
2077 device
->generation
);
2078 btrfs_mark_buffer_dirty(leaf
);
2086 btrfs_free_path(path
);
2090 int btrfs_init_new_device(struct btrfs_root
*root
, char *device_path
)
2092 struct request_queue
*q
;
2093 struct btrfs_trans_handle
*trans
;
2094 struct btrfs_device
*device
;
2095 struct block_device
*bdev
;
2096 struct list_head
*devices
;
2097 struct super_block
*sb
= root
->fs_info
->sb
;
2098 struct rcu_string
*name
;
2100 int seeding_dev
= 0;
2103 if ((sb
->s_flags
& MS_RDONLY
) && !root
->fs_info
->fs_devices
->seeding
)
2106 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2107 root
->fs_info
->bdev_holder
);
2109 return PTR_ERR(bdev
);
2111 if (root
->fs_info
->fs_devices
->seeding
) {
2113 down_write(&sb
->s_umount
);
2114 mutex_lock(&uuid_mutex
);
2117 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2119 devices
= &root
->fs_info
->fs_devices
->devices
;
2121 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2122 list_for_each_entry(device
, devices
, dev_list
) {
2123 if (device
->bdev
== bdev
) {
2126 &root
->fs_info
->fs_devices
->device_list_mutex
);
2130 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2132 device
= btrfs_alloc_device(root
->fs_info
, NULL
, NULL
);
2133 if (IS_ERR(device
)) {
2134 /* we can safely leave the fs_devices entry around */
2135 ret
= PTR_ERR(device
);
2139 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2145 rcu_assign_pointer(device
->name
, name
);
2147 trans
= btrfs_start_transaction(root
, 0);
2148 if (IS_ERR(trans
)) {
2149 rcu_string_free(device
->name
);
2151 ret
= PTR_ERR(trans
);
2155 q
= bdev_get_queue(bdev
);
2156 if (blk_queue_discard(q
))
2157 device
->can_discard
= 1;
2158 device
->writeable
= 1;
2159 device
->generation
= trans
->transid
;
2160 device
->io_width
= root
->sectorsize
;
2161 device
->io_align
= root
->sectorsize
;
2162 device
->sector_size
= root
->sectorsize
;
2163 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
2164 device
->disk_total_bytes
= device
->total_bytes
;
2165 device
->commit_total_bytes
= device
->total_bytes
;
2166 device
->dev_root
= root
->fs_info
->dev_root
;
2167 device
->bdev
= bdev
;
2168 device
->in_fs_metadata
= 1;
2169 device
->is_tgtdev_for_dev_replace
= 0;
2170 device
->mode
= FMODE_EXCL
;
2171 device
->dev_stats_valid
= 1;
2172 set_blocksize(device
->bdev
, 4096);
2175 sb
->s_flags
&= ~MS_RDONLY
;
2176 ret
= btrfs_prepare_sprout(root
);
2177 BUG_ON(ret
); /* -ENOMEM */
2180 device
->fs_devices
= root
->fs_info
->fs_devices
;
2182 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2184 list_add_rcu(&device
->dev_list
, &root
->fs_info
->fs_devices
->devices
);
2185 list_add(&device
->dev_alloc_list
,
2186 &root
->fs_info
->fs_devices
->alloc_list
);
2187 root
->fs_info
->fs_devices
->num_devices
++;
2188 root
->fs_info
->fs_devices
->open_devices
++;
2189 root
->fs_info
->fs_devices
->rw_devices
++;
2190 root
->fs_info
->fs_devices
->total_devices
++;
2191 root
->fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
2193 spin_lock(&root
->fs_info
->free_chunk_lock
);
2194 root
->fs_info
->free_chunk_space
+= device
->total_bytes
;
2195 spin_unlock(&root
->fs_info
->free_chunk_lock
);
2197 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
2198 root
->fs_info
->fs_devices
->rotating
= 1;
2200 tmp
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
2201 btrfs_set_super_total_bytes(root
->fs_info
->super_copy
,
2202 tmp
+ device
->total_bytes
);
2204 tmp
= btrfs_super_num_devices(root
->fs_info
->super_copy
);
2205 btrfs_set_super_num_devices(root
->fs_info
->super_copy
,
2208 /* add sysfs device entry */
2209 btrfs_kobj_add_device(root
->fs_info
, device
);
2212 * we've got more storage, clear any full flags on the space
2215 btrfs_clear_space_info_full(root
->fs_info
);
2217 unlock_chunks(root
);
2218 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2222 ret
= init_first_rw_device(trans
, root
, device
);
2223 unlock_chunks(root
);
2225 btrfs_abort_transaction(trans
, root
, ret
);
2230 ret
= btrfs_add_device(trans
, root
, device
);
2232 btrfs_abort_transaction(trans
, root
, ret
);
2237 char fsid_buf
[BTRFS_UUID_UNPARSED_SIZE
];
2239 ret
= btrfs_finish_sprout(trans
, root
);
2241 btrfs_abort_transaction(trans
, root
, ret
);
2245 /* Sprouting would change fsid of the mounted root,
2246 * so rename the fsid on the sysfs
2248 snprintf(fsid_buf
, BTRFS_UUID_UNPARSED_SIZE
, "%pU",
2249 root
->fs_info
->fsid
);
2250 if (kobject_rename(&root
->fs_info
->super_kobj
, fsid_buf
))
2254 root
->fs_info
->num_tolerated_disk_barrier_failures
=
2255 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
2256 ret
= btrfs_commit_transaction(trans
, root
);
2259 mutex_unlock(&uuid_mutex
);
2260 up_write(&sb
->s_umount
);
2262 if (ret
) /* transaction commit */
2265 ret
= btrfs_relocate_sys_chunks(root
);
2267 btrfs_error(root
->fs_info
, ret
,
2268 "Failed to relocate sys chunks after "
2269 "device initialization. This can be fixed "
2270 "using the \"btrfs balance\" command.");
2271 trans
= btrfs_attach_transaction(root
);
2272 if (IS_ERR(trans
)) {
2273 if (PTR_ERR(trans
) == -ENOENT
)
2275 return PTR_ERR(trans
);
2277 ret
= btrfs_commit_transaction(trans
, root
);
2280 /* Update ctime/mtime for libblkid */
2281 update_dev_time(device_path
);
2285 btrfs_end_transaction(trans
, root
);
2286 rcu_string_free(device
->name
);
2287 btrfs_kobj_rm_device(root
->fs_info
, device
);
2290 blkdev_put(bdev
, FMODE_EXCL
);
2292 mutex_unlock(&uuid_mutex
);
2293 up_write(&sb
->s_umount
);
2298 int btrfs_init_dev_replace_tgtdev(struct btrfs_root
*root
, char *device_path
,
2299 struct btrfs_device
*srcdev
,
2300 struct btrfs_device
**device_out
)
2302 struct request_queue
*q
;
2303 struct btrfs_device
*device
;
2304 struct block_device
*bdev
;
2305 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2306 struct list_head
*devices
;
2307 struct rcu_string
*name
;
2308 u64 devid
= BTRFS_DEV_REPLACE_DEVID
;
2312 if (fs_info
->fs_devices
->seeding
) {
2313 btrfs_err(fs_info
, "the filesystem is a seed filesystem!");
2317 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2318 fs_info
->bdev_holder
);
2320 btrfs_err(fs_info
, "target device %s is invalid!", device_path
);
2321 return PTR_ERR(bdev
);
2324 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2326 devices
= &fs_info
->fs_devices
->devices
;
2327 list_for_each_entry(device
, devices
, dev_list
) {
2328 if (device
->bdev
== bdev
) {
2329 btrfs_err(fs_info
, "target device is in the filesystem!");
2336 if (i_size_read(bdev
->bd_inode
) <
2337 btrfs_device_get_total_bytes(srcdev
)) {
2338 btrfs_err(fs_info
, "target device is smaller than source device!");
2344 device
= btrfs_alloc_device(NULL
, &devid
, NULL
);
2345 if (IS_ERR(device
)) {
2346 ret
= PTR_ERR(device
);
2350 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2356 rcu_assign_pointer(device
->name
, name
);
2358 q
= bdev_get_queue(bdev
);
2359 if (blk_queue_discard(q
))
2360 device
->can_discard
= 1;
2361 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2362 device
->writeable
= 1;
2363 device
->generation
= 0;
2364 device
->io_width
= root
->sectorsize
;
2365 device
->io_align
= root
->sectorsize
;
2366 device
->sector_size
= root
->sectorsize
;
2367 device
->total_bytes
= btrfs_device_get_total_bytes(srcdev
);
2368 device
->disk_total_bytes
= btrfs_device_get_disk_total_bytes(srcdev
);
2369 device
->bytes_used
= btrfs_device_get_bytes_used(srcdev
);
2370 ASSERT(list_empty(&srcdev
->resized_list
));
2371 device
->commit_total_bytes
= srcdev
->commit_total_bytes
;
2372 device
->commit_bytes_used
= device
->bytes_used
;
2373 device
->dev_root
= fs_info
->dev_root
;
2374 device
->bdev
= bdev
;
2375 device
->in_fs_metadata
= 1;
2376 device
->is_tgtdev_for_dev_replace
= 1;
2377 device
->mode
= FMODE_EXCL
;
2378 device
->dev_stats_valid
= 1;
2379 set_blocksize(device
->bdev
, 4096);
2380 device
->fs_devices
= fs_info
->fs_devices
;
2381 list_add(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2382 fs_info
->fs_devices
->num_devices
++;
2383 fs_info
->fs_devices
->open_devices
++;
2384 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2386 *device_out
= device
;
2390 blkdev_put(bdev
, FMODE_EXCL
);
2394 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info
*fs_info
,
2395 struct btrfs_device
*tgtdev
)
2397 WARN_ON(fs_info
->fs_devices
->rw_devices
== 0);
2398 tgtdev
->io_width
= fs_info
->dev_root
->sectorsize
;
2399 tgtdev
->io_align
= fs_info
->dev_root
->sectorsize
;
2400 tgtdev
->sector_size
= fs_info
->dev_root
->sectorsize
;
2401 tgtdev
->dev_root
= fs_info
->dev_root
;
2402 tgtdev
->in_fs_metadata
= 1;
2405 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2406 struct btrfs_device
*device
)
2409 struct btrfs_path
*path
;
2410 struct btrfs_root
*root
;
2411 struct btrfs_dev_item
*dev_item
;
2412 struct extent_buffer
*leaf
;
2413 struct btrfs_key key
;
2415 root
= device
->dev_root
->fs_info
->chunk_root
;
2417 path
= btrfs_alloc_path();
2421 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2422 key
.type
= BTRFS_DEV_ITEM_KEY
;
2423 key
.offset
= device
->devid
;
2425 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2434 leaf
= path
->nodes
[0];
2435 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2437 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2438 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2439 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2440 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2441 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2442 btrfs_set_device_total_bytes(leaf
, dev_item
,
2443 btrfs_device_get_disk_total_bytes(device
));
2444 btrfs_set_device_bytes_used(leaf
, dev_item
,
2445 btrfs_device_get_bytes_used(device
));
2446 btrfs_mark_buffer_dirty(leaf
);
2449 btrfs_free_path(path
);
2453 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2454 struct btrfs_device
*device
, u64 new_size
)
2456 struct btrfs_super_block
*super_copy
=
2457 device
->dev_root
->fs_info
->super_copy
;
2458 struct btrfs_fs_devices
*fs_devices
;
2462 if (!device
->writeable
)
2465 lock_chunks(device
->dev_root
);
2466 old_total
= btrfs_super_total_bytes(super_copy
);
2467 diff
= new_size
- device
->total_bytes
;
2469 if (new_size
<= device
->total_bytes
||
2470 device
->is_tgtdev_for_dev_replace
) {
2471 unlock_chunks(device
->dev_root
);
2475 fs_devices
= device
->dev_root
->fs_info
->fs_devices
;
2477 btrfs_set_super_total_bytes(super_copy
, old_total
+ diff
);
2478 device
->fs_devices
->total_rw_bytes
+= diff
;
2480 btrfs_device_set_total_bytes(device
, new_size
);
2481 btrfs_device_set_disk_total_bytes(device
, new_size
);
2482 btrfs_clear_space_info_full(device
->dev_root
->fs_info
);
2483 if (list_empty(&device
->resized_list
))
2484 list_add_tail(&device
->resized_list
,
2485 &fs_devices
->resized_devices
);
2486 unlock_chunks(device
->dev_root
);
2488 return btrfs_update_device(trans
, device
);
2491 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
2492 struct btrfs_root
*root
,
2493 u64 chunk_tree
, u64 chunk_objectid
,
2497 struct btrfs_path
*path
;
2498 struct btrfs_key key
;
2500 root
= root
->fs_info
->chunk_root
;
2501 path
= btrfs_alloc_path();
2505 key
.objectid
= chunk_objectid
;
2506 key
.offset
= chunk_offset
;
2507 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2509 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2512 else if (ret
> 0) { /* Logic error or corruption */
2513 btrfs_error(root
->fs_info
, -ENOENT
,
2514 "Failed lookup while freeing chunk.");
2519 ret
= btrfs_del_item(trans
, root
, path
);
2521 btrfs_error(root
->fs_info
, ret
,
2522 "Failed to delete chunk item.");
2524 btrfs_free_path(path
);
2528 static int btrfs_del_sys_chunk(struct btrfs_root
*root
, u64 chunk_objectid
, u64
2531 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
2532 struct btrfs_disk_key
*disk_key
;
2533 struct btrfs_chunk
*chunk
;
2540 struct btrfs_key key
;
2543 array_size
= btrfs_super_sys_array_size(super_copy
);
2545 ptr
= super_copy
->sys_chunk_array
;
2548 while (cur
< array_size
) {
2549 disk_key
= (struct btrfs_disk_key
*)ptr
;
2550 btrfs_disk_key_to_cpu(&key
, disk_key
);
2552 len
= sizeof(*disk_key
);
2554 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2555 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2556 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2557 len
+= btrfs_chunk_item_size(num_stripes
);
2562 if (key
.objectid
== chunk_objectid
&&
2563 key
.offset
== chunk_offset
) {
2564 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2566 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2572 unlock_chunks(root
);
2576 int btrfs_remove_chunk(struct btrfs_trans_handle
*trans
,
2577 struct btrfs_root
*root
, u64 chunk_offset
)
2579 struct extent_map_tree
*em_tree
;
2580 struct extent_map
*em
;
2581 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2582 struct map_lookup
*map
;
2583 u64 dev_extent_len
= 0;
2584 u64 chunk_objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2585 u64 chunk_tree
= root
->fs_info
->chunk_root
->objectid
;
2589 root
= root
->fs_info
->chunk_root
;
2590 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
2592 read_lock(&em_tree
->lock
);
2593 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
2594 read_unlock(&em_tree
->lock
);
2596 if (!em
|| em
->start
> chunk_offset
||
2597 em
->start
+ em
->len
< chunk_offset
) {
2599 * This is a logic error, but we don't want to just rely on the
2600 * user having built with ASSERT enabled, so if ASSERT doens't
2601 * do anything we still error out.
2605 free_extent_map(em
);
2608 map
= (struct map_lookup
*)em
->bdev
;
2610 for (i
= 0; i
< map
->num_stripes
; i
++) {
2611 struct btrfs_device
*device
= map
->stripes
[i
].dev
;
2612 ret
= btrfs_free_dev_extent(trans
, device
,
2613 map
->stripes
[i
].physical
,
2616 btrfs_abort_transaction(trans
, root
, ret
);
2620 if (device
->bytes_used
> 0) {
2622 btrfs_device_set_bytes_used(device
,
2623 device
->bytes_used
- dev_extent_len
);
2624 spin_lock(&root
->fs_info
->free_chunk_lock
);
2625 root
->fs_info
->free_chunk_space
+= dev_extent_len
;
2626 spin_unlock(&root
->fs_info
->free_chunk_lock
);
2627 btrfs_clear_space_info_full(root
->fs_info
);
2628 unlock_chunks(root
);
2631 if (map
->stripes
[i
].dev
) {
2632 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2634 btrfs_abort_transaction(trans
, root
, ret
);
2639 ret
= btrfs_free_chunk(trans
, root
, chunk_tree
, chunk_objectid
,
2642 btrfs_abort_transaction(trans
, root
, ret
);
2646 trace_btrfs_chunk_free(root
, map
, chunk_offset
, em
->len
);
2648 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2649 ret
= btrfs_del_sys_chunk(root
, chunk_objectid
, chunk_offset
);
2651 btrfs_abort_transaction(trans
, root
, ret
);
2656 ret
= btrfs_remove_block_group(trans
, extent_root
, chunk_offset
);
2658 btrfs_abort_transaction(trans
, extent_root
, ret
);
2662 write_lock(&em_tree
->lock
);
2663 remove_extent_mapping(em_tree
, em
);
2664 write_unlock(&em_tree
->lock
);
2666 /* once for the tree */
2667 free_extent_map(em
);
2670 free_extent_map(em
);
2674 static int btrfs_relocate_chunk(struct btrfs_root
*root
,
2675 u64 chunk_tree
, u64 chunk_objectid
,
2678 struct btrfs_root
*extent_root
;
2679 struct btrfs_trans_handle
*trans
;
2682 root
= root
->fs_info
->chunk_root
;
2683 extent_root
= root
->fs_info
->extent_root
;
2685 ret
= btrfs_can_relocate(extent_root
, chunk_offset
);
2689 /* step one, relocate all the extents inside this chunk */
2690 ret
= btrfs_relocate_block_group(extent_root
, chunk_offset
);
2694 trans
= btrfs_start_transaction(root
, 0);
2695 if (IS_ERR(trans
)) {
2696 ret
= PTR_ERR(trans
);
2697 btrfs_std_error(root
->fs_info
, ret
);
2702 * step two, delete the device extents and the
2703 * chunk tree entries
2705 ret
= btrfs_remove_chunk(trans
, root
, chunk_offset
);
2706 btrfs_end_transaction(trans
, root
);
2710 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
)
2712 struct btrfs_root
*chunk_root
= root
->fs_info
->chunk_root
;
2713 struct btrfs_path
*path
;
2714 struct extent_buffer
*leaf
;
2715 struct btrfs_chunk
*chunk
;
2716 struct btrfs_key key
;
2717 struct btrfs_key found_key
;
2718 u64 chunk_tree
= chunk_root
->root_key
.objectid
;
2720 bool retried
= false;
2724 path
= btrfs_alloc_path();
2729 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2730 key
.offset
= (u64
)-1;
2731 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2734 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2737 BUG_ON(ret
== 0); /* Corruption */
2739 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
2746 leaf
= path
->nodes
[0];
2747 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2749 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
2750 struct btrfs_chunk
);
2751 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2752 btrfs_release_path(path
);
2754 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2755 ret
= btrfs_relocate_chunk(chunk_root
, chunk_tree
,
2764 if (found_key
.offset
== 0)
2766 key
.offset
= found_key
.offset
- 1;
2769 if (failed
&& !retried
) {
2773 } else if (WARN_ON(failed
&& retried
)) {
2777 btrfs_free_path(path
);
2781 static int insert_balance_item(struct btrfs_root
*root
,
2782 struct btrfs_balance_control
*bctl
)
2784 struct btrfs_trans_handle
*trans
;
2785 struct btrfs_balance_item
*item
;
2786 struct btrfs_disk_balance_args disk_bargs
;
2787 struct btrfs_path
*path
;
2788 struct extent_buffer
*leaf
;
2789 struct btrfs_key key
;
2792 path
= btrfs_alloc_path();
2796 trans
= btrfs_start_transaction(root
, 0);
2797 if (IS_ERR(trans
)) {
2798 btrfs_free_path(path
);
2799 return PTR_ERR(trans
);
2802 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2803 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2806 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2811 leaf
= path
->nodes
[0];
2812 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
2814 memset_extent_buffer(leaf
, 0, (unsigned long)item
, sizeof(*item
));
2816 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
2817 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
2818 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
2819 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
2820 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
2821 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
2823 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
2825 btrfs_mark_buffer_dirty(leaf
);
2827 btrfs_free_path(path
);
2828 err
= btrfs_commit_transaction(trans
, root
);
2834 static int del_balance_item(struct btrfs_root
*root
)
2836 struct btrfs_trans_handle
*trans
;
2837 struct btrfs_path
*path
;
2838 struct btrfs_key key
;
2841 path
= btrfs_alloc_path();
2845 trans
= btrfs_start_transaction(root
, 0);
2846 if (IS_ERR(trans
)) {
2847 btrfs_free_path(path
);
2848 return PTR_ERR(trans
);
2851 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2852 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2855 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2863 ret
= btrfs_del_item(trans
, root
, path
);
2865 btrfs_free_path(path
);
2866 err
= btrfs_commit_transaction(trans
, root
);
2873 * This is a heuristic used to reduce the number of chunks balanced on
2874 * resume after balance was interrupted.
2876 static void update_balance_args(struct btrfs_balance_control
*bctl
)
2879 * Turn on soft mode for chunk types that were being converted.
2881 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2882 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2883 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2884 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2885 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2886 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2889 * Turn on usage filter if is not already used. The idea is
2890 * that chunks that we have already balanced should be
2891 * reasonably full. Don't do it for chunks that are being
2892 * converted - that will keep us from relocating unconverted
2893 * (albeit full) chunks.
2895 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2896 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2897 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2898 bctl
->data
.usage
= 90;
2900 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2901 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2902 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2903 bctl
->sys
.usage
= 90;
2905 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2906 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2907 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2908 bctl
->meta
.usage
= 90;
2913 * Should be called with both balance and volume mutexes held to
2914 * serialize other volume operations (add_dev/rm_dev/resize) with
2915 * restriper. Same goes for unset_balance_control.
2917 static void set_balance_control(struct btrfs_balance_control
*bctl
)
2919 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
2921 BUG_ON(fs_info
->balance_ctl
);
2923 spin_lock(&fs_info
->balance_lock
);
2924 fs_info
->balance_ctl
= bctl
;
2925 spin_unlock(&fs_info
->balance_lock
);
2928 static void unset_balance_control(struct btrfs_fs_info
*fs_info
)
2930 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
2932 BUG_ON(!fs_info
->balance_ctl
);
2934 spin_lock(&fs_info
->balance_lock
);
2935 fs_info
->balance_ctl
= NULL
;
2936 spin_unlock(&fs_info
->balance_lock
);
2942 * Balance filters. Return 1 if chunk should be filtered out
2943 * (should not be balanced).
2945 static int chunk_profiles_filter(u64 chunk_type
,
2946 struct btrfs_balance_args
*bargs
)
2948 chunk_type
= chunk_to_extended(chunk_type
) &
2949 BTRFS_EXTENDED_PROFILE_MASK
;
2951 if (bargs
->profiles
& chunk_type
)
2957 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
2958 struct btrfs_balance_args
*bargs
)
2960 struct btrfs_block_group_cache
*cache
;
2961 u64 chunk_used
, user_thresh
;
2964 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
2965 chunk_used
= btrfs_block_group_used(&cache
->item
);
2967 if (bargs
->usage
== 0)
2969 else if (bargs
->usage
> 100)
2970 user_thresh
= cache
->key
.offset
;
2972 user_thresh
= div_factor_fine(cache
->key
.offset
,
2975 if (chunk_used
< user_thresh
)
2978 btrfs_put_block_group(cache
);
2982 static int chunk_devid_filter(struct extent_buffer
*leaf
,
2983 struct btrfs_chunk
*chunk
,
2984 struct btrfs_balance_args
*bargs
)
2986 struct btrfs_stripe
*stripe
;
2987 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2990 for (i
= 0; i
< num_stripes
; i
++) {
2991 stripe
= btrfs_stripe_nr(chunk
, i
);
2992 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
2999 /* [pstart, pend) */
3000 static int chunk_drange_filter(struct extent_buffer
*leaf
,
3001 struct btrfs_chunk
*chunk
,
3003 struct btrfs_balance_args
*bargs
)
3005 struct btrfs_stripe
*stripe
;
3006 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3012 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
3015 if (btrfs_chunk_type(leaf
, chunk
) & (BTRFS_BLOCK_GROUP_DUP
|
3016 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)) {
3017 factor
= num_stripes
/ 2;
3018 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID5
) {
3019 factor
= num_stripes
- 1;
3020 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID6
) {
3021 factor
= num_stripes
- 2;
3023 factor
= num_stripes
;
3026 for (i
= 0; i
< num_stripes
; i
++) {
3027 stripe
= btrfs_stripe_nr(chunk
, i
);
3028 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
3031 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
3032 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
3033 do_div(stripe_length
, factor
);
3035 if (stripe_offset
< bargs
->pend
&&
3036 stripe_offset
+ stripe_length
> bargs
->pstart
)
3043 /* [vstart, vend) */
3044 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
3045 struct btrfs_chunk
*chunk
,
3047 struct btrfs_balance_args
*bargs
)
3049 if (chunk_offset
< bargs
->vend
&&
3050 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
3051 /* at least part of the chunk is inside this vrange */
3057 static int chunk_soft_convert_filter(u64 chunk_type
,
3058 struct btrfs_balance_args
*bargs
)
3060 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
3063 chunk_type
= chunk_to_extended(chunk_type
) &
3064 BTRFS_EXTENDED_PROFILE_MASK
;
3066 if (bargs
->target
== chunk_type
)
3072 static int should_balance_chunk(struct btrfs_root
*root
,
3073 struct extent_buffer
*leaf
,
3074 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
3076 struct btrfs_balance_control
*bctl
= root
->fs_info
->balance_ctl
;
3077 struct btrfs_balance_args
*bargs
= NULL
;
3078 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3081 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
3082 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
3086 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3087 bargs
= &bctl
->data
;
3088 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3090 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3091 bargs
= &bctl
->meta
;
3093 /* profiles filter */
3094 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
3095 chunk_profiles_filter(chunk_type
, bargs
)) {
3100 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3101 chunk_usage_filter(bctl
->fs_info
, chunk_offset
, bargs
)) {
3106 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
3107 chunk_devid_filter(leaf
, chunk
, bargs
)) {
3111 /* drange filter, makes sense only with devid filter */
3112 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
3113 chunk_drange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3118 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
3119 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3123 /* soft profile changing mode */
3124 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
3125 chunk_soft_convert_filter(chunk_type
, bargs
)) {
3130 * limited by count, must be the last filter
3132 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT
)) {
3133 if (bargs
->limit
== 0)
3142 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
3144 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3145 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
3146 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
3147 struct list_head
*devices
;
3148 struct btrfs_device
*device
;
3151 struct btrfs_chunk
*chunk
;
3152 struct btrfs_path
*path
;
3153 struct btrfs_key key
;
3154 struct btrfs_key found_key
;
3155 struct btrfs_trans_handle
*trans
;
3156 struct extent_buffer
*leaf
;
3159 int enospc_errors
= 0;
3160 bool counting
= true;
3161 u64 limit_data
= bctl
->data
.limit
;
3162 u64 limit_meta
= bctl
->meta
.limit
;
3163 u64 limit_sys
= bctl
->sys
.limit
;
3165 /* step one make some room on all the devices */
3166 devices
= &fs_info
->fs_devices
->devices
;
3167 list_for_each_entry(device
, devices
, dev_list
) {
3168 old_size
= btrfs_device_get_total_bytes(device
);
3169 size_to_free
= div_factor(old_size
, 1);
3170 size_to_free
= min(size_to_free
, (u64
)1 * 1024 * 1024);
3171 if (!device
->writeable
||
3172 btrfs_device_get_total_bytes(device
) -
3173 btrfs_device_get_bytes_used(device
) > size_to_free
||
3174 device
->is_tgtdev_for_dev_replace
)
3177 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
3182 trans
= btrfs_start_transaction(dev_root
, 0);
3183 BUG_ON(IS_ERR(trans
));
3185 ret
= btrfs_grow_device(trans
, device
, old_size
);
3188 btrfs_end_transaction(trans
, dev_root
);
3191 /* step two, relocate all the chunks */
3192 path
= btrfs_alloc_path();
3198 /* zero out stat counters */
3199 spin_lock(&fs_info
->balance_lock
);
3200 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
3201 spin_unlock(&fs_info
->balance_lock
);
3204 bctl
->data
.limit
= limit_data
;
3205 bctl
->meta
.limit
= limit_meta
;
3206 bctl
->sys
.limit
= limit_sys
;
3208 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3209 key
.offset
= (u64
)-1;
3210 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3213 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
3214 atomic_read(&fs_info
->balance_cancel_req
)) {
3219 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
3224 * this shouldn't happen, it means the last relocate
3228 BUG(); /* FIXME break ? */
3230 ret
= btrfs_previous_item(chunk_root
, path
, 0,
3231 BTRFS_CHUNK_ITEM_KEY
);
3237 leaf
= path
->nodes
[0];
3238 slot
= path
->slots
[0];
3239 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3241 if (found_key
.objectid
!= key
.objectid
)
3244 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
3247 spin_lock(&fs_info
->balance_lock
);
3248 bctl
->stat
.considered
++;
3249 spin_unlock(&fs_info
->balance_lock
);
3252 ret
= should_balance_chunk(chunk_root
, leaf
, chunk
,
3254 btrfs_release_path(path
);
3259 spin_lock(&fs_info
->balance_lock
);
3260 bctl
->stat
.expected
++;
3261 spin_unlock(&fs_info
->balance_lock
);
3265 ret
= btrfs_relocate_chunk(chunk_root
,
3266 chunk_root
->root_key
.objectid
,
3269 if (ret
&& ret
!= -ENOSPC
)
3271 if (ret
== -ENOSPC
) {
3274 spin_lock(&fs_info
->balance_lock
);
3275 bctl
->stat
.completed
++;
3276 spin_unlock(&fs_info
->balance_lock
);
3279 if (found_key
.offset
== 0)
3281 key
.offset
= found_key
.offset
- 1;
3285 btrfs_release_path(path
);
3290 btrfs_free_path(path
);
3291 if (enospc_errors
) {
3292 btrfs_info(fs_info
, "%d enospc errors during balance",
3302 * alloc_profile_is_valid - see if a given profile is valid and reduced
3303 * @flags: profile to validate
3304 * @extended: if true @flags is treated as an extended profile
3306 static int alloc_profile_is_valid(u64 flags
, int extended
)
3308 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
3309 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
3311 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
3313 /* 1) check that all other bits are zeroed */
3317 /* 2) see if profile is reduced */
3319 return !extended
; /* "0" is valid for usual profiles */
3321 /* true if exactly one bit set */
3322 return (flags
& (flags
- 1)) == 0;
3325 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
3327 /* cancel requested || normal exit path */
3328 return atomic_read(&fs_info
->balance_cancel_req
) ||
3329 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
3330 atomic_read(&fs_info
->balance_cancel_req
) == 0);
3333 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
3337 unset_balance_control(fs_info
);
3338 ret
= del_balance_item(fs_info
->tree_root
);
3340 btrfs_std_error(fs_info
, ret
);
3342 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3346 * Should be called with both balance and volume mutexes held
3348 int btrfs_balance(struct btrfs_balance_control
*bctl
,
3349 struct btrfs_ioctl_balance_args
*bargs
)
3351 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3358 if (btrfs_fs_closing(fs_info
) ||
3359 atomic_read(&fs_info
->balance_pause_req
) ||
3360 atomic_read(&fs_info
->balance_cancel_req
)) {
3365 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
3366 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
3370 * In case of mixed groups both data and meta should be picked,
3371 * and identical options should be given for both of them.
3373 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
3374 if (mixed
&& (bctl
->flags
& allowed
)) {
3375 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
3376 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
3377 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
3378 btrfs_err(fs_info
, "with mixed groups data and "
3379 "metadata balance options must be the same");
3385 num_devices
= fs_info
->fs_devices
->num_devices
;
3386 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
3387 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
3388 BUG_ON(num_devices
< 1);
3391 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
3392 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
3393 if (num_devices
== 1)
3394 allowed
|= BTRFS_BLOCK_GROUP_DUP
;
3395 else if (num_devices
> 1)
3396 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
3397 if (num_devices
> 2)
3398 allowed
|= BTRFS_BLOCK_GROUP_RAID5
;
3399 if (num_devices
> 3)
3400 allowed
|= (BTRFS_BLOCK_GROUP_RAID10
|
3401 BTRFS_BLOCK_GROUP_RAID6
);
3402 if ((bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3403 (!alloc_profile_is_valid(bctl
->data
.target
, 1) ||
3404 (bctl
->data
.target
& ~allowed
))) {
3405 btrfs_err(fs_info
, "unable to start balance with target "
3406 "data profile %llu",
3411 if ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3412 (!alloc_profile_is_valid(bctl
->meta
.target
, 1) ||
3413 (bctl
->meta
.target
& ~allowed
))) {
3415 "unable to start balance with target metadata profile %llu",
3420 if ((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3421 (!alloc_profile_is_valid(bctl
->sys
.target
, 1) ||
3422 (bctl
->sys
.target
& ~allowed
))) {
3424 "unable to start balance with target system profile %llu",
3430 /* allow dup'ed data chunks only in mixed mode */
3431 if (!mixed
&& (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3432 (bctl
->data
.target
& BTRFS_BLOCK_GROUP_DUP
)) {
3433 btrfs_err(fs_info
, "dup for data is not allowed");
3438 /* allow to reduce meta or sys integrity only if force set */
3439 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3440 BTRFS_BLOCK_GROUP_RAID10
|
3441 BTRFS_BLOCK_GROUP_RAID5
|
3442 BTRFS_BLOCK_GROUP_RAID6
;
3444 seq
= read_seqbegin(&fs_info
->profiles_lock
);
3446 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3447 (fs_info
->avail_system_alloc_bits
& allowed
) &&
3448 !(bctl
->sys
.target
& allowed
)) ||
3449 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3450 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
3451 !(bctl
->meta
.target
& allowed
))) {
3452 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
3453 btrfs_info(fs_info
, "force reducing metadata integrity");
3455 btrfs_err(fs_info
, "balance will reduce metadata "
3456 "integrity, use force if you want this");
3461 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
3463 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3464 int num_tolerated_disk_barrier_failures
;
3465 u64 target
= bctl
->sys
.target
;
3467 num_tolerated_disk_barrier_failures
=
3468 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3469 if (num_tolerated_disk_barrier_failures
> 0 &&
3471 (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID0
|
3472 BTRFS_AVAIL_ALLOC_BIT_SINGLE
)))
3473 num_tolerated_disk_barrier_failures
= 0;
3474 else if (num_tolerated_disk_barrier_failures
> 1 &&
3476 (BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)))
3477 num_tolerated_disk_barrier_failures
= 1;
3479 fs_info
->num_tolerated_disk_barrier_failures
=
3480 num_tolerated_disk_barrier_failures
;
3483 ret
= insert_balance_item(fs_info
->tree_root
, bctl
);
3484 if (ret
&& ret
!= -EEXIST
)
3487 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
3488 BUG_ON(ret
== -EEXIST
);
3489 set_balance_control(bctl
);
3491 BUG_ON(ret
!= -EEXIST
);
3492 spin_lock(&fs_info
->balance_lock
);
3493 update_balance_args(bctl
);
3494 spin_unlock(&fs_info
->balance_lock
);
3497 atomic_inc(&fs_info
->balance_running
);
3498 mutex_unlock(&fs_info
->balance_mutex
);
3500 ret
= __btrfs_balance(fs_info
);
3502 mutex_lock(&fs_info
->balance_mutex
);
3503 atomic_dec(&fs_info
->balance_running
);
3505 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3506 fs_info
->num_tolerated_disk_barrier_failures
=
3507 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3511 memset(bargs
, 0, sizeof(*bargs
));
3512 update_ioctl_balance_args(fs_info
, 0, bargs
);
3515 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
3516 balance_need_close(fs_info
)) {
3517 __cancel_balance(fs_info
);
3520 wake_up(&fs_info
->balance_wait_q
);
3524 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
3525 __cancel_balance(fs_info
);
3528 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3533 static int balance_kthread(void *data
)
3535 struct btrfs_fs_info
*fs_info
= data
;
3538 mutex_lock(&fs_info
->volume_mutex
);
3539 mutex_lock(&fs_info
->balance_mutex
);
3541 if (fs_info
->balance_ctl
) {
3542 btrfs_info(fs_info
, "continuing balance");
3543 ret
= btrfs_balance(fs_info
->balance_ctl
, NULL
);
3546 mutex_unlock(&fs_info
->balance_mutex
);
3547 mutex_unlock(&fs_info
->volume_mutex
);
3552 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
3554 struct task_struct
*tsk
;
3556 spin_lock(&fs_info
->balance_lock
);
3557 if (!fs_info
->balance_ctl
) {
3558 spin_unlock(&fs_info
->balance_lock
);
3561 spin_unlock(&fs_info
->balance_lock
);
3563 if (btrfs_test_opt(fs_info
->tree_root
, SKIP_BALANCE
)) {
3564 btrfs_info(fs_info
, "force skipping balance");
3568 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
3569 return PTR_ERR_OR_ZERO(tsk
);
3572 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
3574 struct btrfs_balance_control
*bctl
;
3575 struct btrfs_balance_item
*item
;
3576 struct btrfs_disk_balance_args disk_bargs
;
3577 struct btrfs_path
*path
;
3578 struct extent_buffer
*leaf
;
3579 struct btrfs_key key
;
3582 path
= btrfs_alloc_path();
3586 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3587 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
3590 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
3593 if (ret
> 0) { /* ret = -ENOENT; */
3598 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
3604 leaf
= path
->nodes
[0];
3605 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3607 bctl
->fs_info
= fs_info
;
3608 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
3609 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
3611 btrfs_balance_data(leaf
, item
, &disk_bargs
);
3612 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
3613 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
3614 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
3615 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
3616 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
3618 WARN_ON(atomic_xchg(&fs_info
->mutually_exclusive_operation_running
, 1));
3620 mutex_lock(&fs_info
->volume_mutex
);
3621 mutex_lock(&fs_info
->balance_mutex
);
3623 set_balance_control(bctl
);
3625 mutex_unlock(&fs_info
->balance_mutex
);
3626 mutex_unlock(&fs_info
->volume_mutex
);
3628 btrfs_free_path(path
);
3632 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
3636 mutex_lock(&fs_info
->balance_mutex
);
3637 if (!fs_info
->balance_ctl
) {
3638 mutex_unlock(&fs_info
->balance_mutex
);
3642 if (atomic_read(&fs_info
->balance_running
)) {
3643 atomic_inc(&fs_info
->balance_pause_req
);
3644 mutex_unlock(&fs_info
->balance_mutex
);
3646 wait_event(fs_info
->balance_wait_q
,
3647 atomic_read(&fs_info
->balance_running
) == 0);
3649 mutex_lock(&fs_info
->balance_mutex
);
3650 /* we are good with balance_ctl ripped off from under us */
3651 BUG_ON(atomic_read(&fs_info
->balance_running
));
3652 atomic_dec(&fs_info
->balance_pause_req
);
3657 mutex_unlock(&fs_info
->balance_mutex
);
3661 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
3663 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
3666 mutex_lock(&fs_info
->balance_mutex
);
3667 if (!fs_info
->balance_ctl
) {
3668 mutex_unlock(&fs_info
->balance_mutex
);
3672 atomic_inc(&fs_info
->balance_cancel_req
);
3674 * if we are running just wait and return, balance item is
3675 * deleted in btrfs_balance in this case
3677 if (atomic_read(&fs_info
->balance_running
)) {
3678 mutex_unlock(&fs_info
->balance_mutex
);
3679 wait_event(fs_info
->balance_wait_q
,
3680 atomic_read(&fs_info
->balance_running
) == 0);
3681 mutex_lock(&fs_info
->balance_mutex
);
3683 /* __cancel_balance needs volume_mutex */
3684 mutex_unlock(&fs_info
->balance_mutex
);
3685 mutex_lock(&fs_info
->volume_mutex
);
3686 mutex_lock(&fs_info
->balance_mutex
);
3688 if (fs_info
->balance_ctl
)
3689 __cancel_balance(fs_info
);
3691 mutex_unlock(&fs_info
->volume_mutex
);
3694 BUG_ON(fs_info
->balance_ctl
|| atomic_read(&fs_info
->balance_running
));
3695 atomic_dec(&fs_info
->balance_cancel_req
);
3696 mutex_unlock(&fs_info
->balance_mutex
);
3700 static int btrfs_uuid_scan_kthread(void *data
)
3702 struct btrfs_fs_info
*fs_info
= data
;
3703 struct btrfs_root
*root
= fs_info
->tree_root
;
3704 struct btrfs_key key
;
3705 struct btrfs_key max_key
;
3706 struct btrfs_path
*path
= NULL
;
3708 struct extent_buffer
*eb
;
3710 struct btrfs_root_item root_item
;
3712 struct btrfs_trans_handle
*trans
= NULL
;
3714 path
= btrfs_alloc_path();
3721 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3724 max_key
.objectid
= (u64
)-1;
3725 max_key
.type
= BTRFS_ROOT_ITEM_KEY
;
3726 max_key
.offset
= (u64
)-1;
3729 ret
= btrfs_search_forward(root
, &key
, path
, 0);
3736 if (key
.type
!= BTRFS_ROOT_ITEM_KEY
||
3737 (key
.objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
3738 key
.objectid
!= BTRFS_FS_TREE_OBJECTID
) ||
3739 key
.objectid
> BTRFS_LAST_FREE_OBJECTID
)
3742 eb
= path
->nodes
[0];
3743 slot
= path
->slots
[0];
3744 item_size
= btrfs_item_size_nr(eb
, slot
);
3745 if (item_size
< sizeof(root_item
))
3748 read_extent_buffer(eb
, &root_item
,
3749 btrfs_item_ptr_offset(eb
, slot
),
3750 (int)sizeof(root_item
));
3751 if (btrfs_root_refs(&root_item
) == 0)
3754 if (!btrfs_is_empty_uuid(root_item
.uuid
) ||
3755 !btrfs_is_empty_uuid(root_item
.received_uuid
)) {
3759 btrfs_release_path(path
);
3761 * 1 - subvol uuid item
3762 * 1 - received_subvol uuid item
3764 trans
= btrfs_start_transaction(fs_info
->uuid_root
, 2);
3765 if (IS_ERR(trans
)) {
3766 ret
= PTR_ERR(trans
);
3774 if (!btrfs_is_empty_uuid(root_item
.uuid
)) {
3775 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
3777 BTRFS_UUID_KEY_SUBVOL
,
3780 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
3786 if (!btrfs_is_empty_uuid(root_item
.received_uuid
)) {
3787 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
3788 root_item
.received_uuid
,
3789 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
3792 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
3800 ret
= btrfs_end_transaction(trans
, fs_info
->uuid_root
);
3806 btrfs_release_path(path
);
3807 if (key
.offset
< (u64
)-1) {
3809 } else if (key
.type
< BTRFS_ROOT_ITEM_KEY
) {
3811 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3812 } else if (key
.objectid
< (u64
)-1) {
3814 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3823 btrfs_free_path(path
);
3824 if (trans
&& !IS_ERR(trans
))
3825 btrfs_end_transaction(trans
, fs_info
->uuid_root
);
3827 btrfs_warn(fs_info
, "btrfs_uuid_scan_kthread failed %d", ret
);
3829 fs_info
->update_uuid_tree_gen
= 1;
3830 up(&fs_info
->uuid_tree_rescan_sem
);
3835 * Callback for btrfs_uuid_tree_iterate().
3837 * 0 check succeeded, the entry is not outdated.
3838 * < 0 if an error occured.
3839 * > 0 if the check failed, which means the caller shall remove the entry.
3841 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info
*fs_info
,
3842 u8
*uuid
, u8 type
, u64 subid
)
3844 struct btrfs_key key
;
3846 struct btrfs_root
*subvol_root
;
3848 if (type
!= BTRFS_UUID_KEY_SUBVOL
&&
3849 type
!= BTRFS_UUID_KEY_RECEIVED_SUBVOL
)
3852 key
.objectid
= subid
;
3853 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3854 key
.offset
= (u64
)-1;
3855 subvol_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
3856 if (IS_ERR(subvol_root
)) {
3857 ret
= PTR_ERR(subvol_root
);
3864 case BTRFS_UUID_KEY_SUBVOL
:
3865 if (memcmp(uuid
, subvol_root
->root_item
.uuid
, BTRFS_UUID_SIZE
))
3868 case BTRFS_UUID_KEY_RECEIVED_SUBVOL
:
3869 if (memcmp(uuid
, subvol_root
->root_item
.received_uuid
,
3879 static int btrfs_uuid_rescan_kthread(void *data
)
3881 struct btrfs_fs_info
*fs_info
= (struct btrfs_fs_info
*)data
;
3885 * 1st step is to iterate through the existing UUID tree and
3886 * to delete all entries that contain outdated data.
3887 * 2nd step is to add all missing entries to the UUID tree.
3889 ret
= btrfs_uuid_tree_iterate(fs_info
, btrfs_check_uuid_tree_entry
);
3891 btrfs_warn(fs_info
, "iterating uuid_tree failed %d", ret
);
3892 up(&fs_info
->uuid_tree_rescan_sem
);
3895 return btrfs_uuid_scan_kthread(data
);
3898 int btrfs_create_uuid_tree(struct btrfs_fs_info
*fs_info
)
3900 struct btrfs_trans_handle
*trans
;
3901 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
3902 struct btrfs_root
*uuid_root
;
3903 struct task_struct
*task
;
3910 trans
= btrfs_start_transaction(tree_root
, 2);
3912 return PTR_ERR(trans
);
3914 uuid_root
= btrfs_create_tree(trans
, fs_info
,
3915 BTRFS_UUID_TREE_OBJECTID
);
3916 if (IS_ERR(uuid_root
)) {
3917 btrfs_abort_transaction(trans
, tree_root
,
3918 PTR_ERR(uuid_root
));
3919 return PTR_ERR(uuid_root
);
3922 fs_info
->uuid_root
= uuid_root
;
3924 ret
= btrfs_commit_transaction(trans
, tree_root
);
3928 down(&fs_info
->uuid_tree_rescan_sem
);
3929 task
= kthread_run(btrfs_uuid_scan_kthread
, fs_info
, "btrfs-uuid");
3931 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3932 btrfs_warn(fs_info
, "failed to start uuid_scan task");
3933 up(&fs_info
->uuid_tree_rescan_sem
);
3934 return PTR_ERR(task
);
3940 int btrfs_check_uuid_tree(struct btrfs_fs_info
*fs_info
)
3942 struct task_struct
*task
;
3944 down(&fs_info
->uuid_tree_rescan_sem
);
3945 task
= kthread_run(btrfs_uuid_rescan_kthread
, fs_info
, "btrfs-uuid");
3947 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3948 btrfs_warn(fs_info
, "failed to start uuid_rescan task");
3949 up(&fs_info
->uuid_tree_rescan_sem
);
3950 return PTR_ERR(task
);
3957 * shrinking a device means finding all of the device extents past
3958 * the new size, and then following the back refs to the chunks.
3959 * The chunk relocation code actually frees the device extent
3961 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
3963 struct btrfs_trans_handle
*trans
;
3964 struct btrfs_root
*root
= device
->dev_root
;
3965 struct btrfs_dev_extent
*dev_extent
= NULL
;
3966 struct btrfs_path
*path
;
3974 bool retried
= false;
3975 struct extent_buffer
*l
;
3976 struct btrfs_key key
;
3977 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
3978 u64 old_total
= btrfs_super_total_bytes(super_copy
);
3979 u64 old_size
= btrfs_device_get_total_bytes(device
);
3980 u64 diff
= old_size
- new_size
;
3982 if (device
->is_tgtdev_for_dev_replace
)
3985 path
= btrfs_alloc_path();
3993 btrfs_device_set_total_bytes(device
, new_size
);
3994 if (device
->writeable
) {
3995 device
->fs_devices
->total_rw_bytes
-= diff
;
3996 spin_lock(&root
->fs_info
->free_chunk_lock
);
3997 root
->fs_info
->free_chunk_space
-= diff
;
3998 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4000 unlock_chunks(root
);
4003 key
.objectid
= device
->devid
;
4004 key
.offset
= (u64
)-1;
4005 key
.type
= BTRFS_DEV_EXTENT_KEY
;
4008 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4012 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
4017 btrfs_release_path(path
);
4022 slot
= path
->slots
[0];
4023 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
4025 if (key
.objectid
!= device
->devid
) {
4026 btrfs_release_path(path
);
4030 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
4031 length
= btrfs_dev_extent_length(l
, dev_extent
);
4033 if (key
.offset
+ length
<= new_size
) {
4034 btrfs_release_path(path
);
4038 chunk_tree
= btrfs_dev_extent_chunk_tree(l
, dev_extent
);
4039 chunk_objectid
= btrfs_dev_extent_chunk_objectid(l
, dev_extent
);
4040 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
4041 btrfs_release_path(path
);
4043 ret
= btrfs_relocate_chunk(root
, chunk_tree
, chunk_objectid
,
4045 if (ret
&& ret
!= -ENOSPC
)
4049 } while (key
.offset
-- > 0);
4051 if (failed
&& !retried
) {
4055 } else if (failed
&& retried
) {
4059 btrfs_device_set_total_bytes(device
, old_size
);
4060 if (device
->writeable
)
4061 device
->fs_devices
->total_rw_bytes
+= diff
;
4062 spin_lock(&root
->fs_info
->free_chunk_lock
);
4063 root
->fs_info
->free_chunk_space
+= diff
;
4064 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4065 unlock_chunks(root
);
4069 /* Shrinking succeeded, else we would be at "done". */
4070 trans
= btrfs_start_transaction(root
, 0);
4071 if (IS_ERR(trans
)) {
4072 ret
= PTR_ERR(trans
);
4077 btrfs_device_set_disk_total_bytes(device
, new_size
);
4078 if (list_empty(&device
->resized_list
))
4079 list_add_tail(&device
->resized_list
,
4080 &root
->fs_info
->fs_devices
->resized_devices
);
4082 WARN_ON(diff
> old_total
);
4083 btrfs_set_super_total_bytes(super_copy
, old_total
- diff
);
4084 unlock_chunks(root
);
4086 /* Now btrfs_update_device() will change the on-disk size. */
4087 ret
= btrfs_update_device(trans
, device
);
4088 btrfs_end_transaction(trans
, root
);
4090 btrfs_free_path(path
);
4094 static int btrfs_add_system_chunk(struct btrfs_root
*root
,
4095 struct btrfs_key
*key
,
4096 struct btrfs_chunk
*chunk
, int item_size
)
4098 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
4099 struct btrfs_disk_key disk_key
;
4104 array_size
= btrfs_super_sys_array_size(super_copy
);
4105 if (array_size
+ item_size
+ sizeof(disk_key
)
4106 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4107 unlock_chunks(root
);
4111 ptr
= super_copy
->sys_chunk_array
+ array_size
;
4112 btrfs_cpu_key_to_disk(&disk_key
, key
);
4113 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
4114 ptr
+= sizeof(disk_key
);
4115 memcpy(ptr
, chunk
, item_size
);
4116 item_size
+= sizeof(disk_key
);
4117 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
4118 unlock_chunks(root
);
4124 * sort the devices in descending order by max_avail, total_avail
4126 static int btrfs_cmp_device_info(const void *a
, const void *b
)
4128 const struct btrfs_device_info
*di_a
= a
;
4129 const struct btrfs_device_info
*di_b
= b
;
4131 if (di_a
->max_avail
> di_b
->max_avail
)
4133 if (di_a
->max_avail
< di_b
->max_avail
)
4135 if (di_a
->total_avail
> di_b
->total_avail
)
4137 if (di_a
->total_avail
< di_b
->total_avail
)
4142 static struct btrfs_raid_attr btrfs_raid_array
[BTRFS_NR_RAID_TYPES
] = {
4143 [BTRFS_RAID_RAID10
] = {
4146 .devs_max
= 0, /* 0 == as many as possible */
4148 .devs_increment
= 2,
4151 [BTRFS_RAID_RAID1
] = {
4156 .devs_increment
= 2,
4159 [BTRFS_RAID_DUP
] = {
4164 .devs_increment
= 1,
4167 [BTRFS_RAID_RAID0
] = {
4172 .devs_increment
= 1,
4175 [BTRFS_RAID_SINGLE
] = {
4180 .devs_increment
= 1,
4183 [BTRFS_RAID_RAID5
] = {
4188 .devs_increment
= 1,
4191 [BTRFS_RAID_RAID6
] = {
4196 .devs_increment
= 1,
4201 static u32
find_raid56_stripe_len(u32 data_devices
, u32 dev_stripe_target
)
4203 /* TODO allow them to set a preferred stripe size */
4207 static void check_raid56_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
4209 if (!(type
& (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
)))
4212 btrfs_set_fs_incompat(info
, RAID56
);
4215 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4216 - sizeof(struct btrfs_item) \
4217 - sizeof(struct btrfs_chunk)) \
4218 / sizeof(struct btrfs_stripe) + 1)
4220 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4221 - 2 * sizeof(struct btrfs_disk_key) \
4222 - 2 * sizeof(struct btrfs_chunk)) \
4223 / sizeof(struct btrfs_stripe) + 1)
4225 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4226 struct btrfs_root
*extent_root
, u64 start
,
4229 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
4230 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
4231 struct list_head
*cur
;
4232 struct map_lookup
*map
= NULL
;
4233 struct extent_map_tree
*em_tree
;
4234 struct extent_map
*em
;
4235 struct btrfs_device_info
*devices_info
= NULL
;
4237 int num_stripes
; /* total number of stripes to allocate */
4238 int data_stripes
; /* number of stripes that count for
4240 int sub_stripes
; /* sub_stripes info for map */
4241 int dev_stripes
; /* stripes per dev */
4242 int devs_max
; /* max devs to use */
4243 int devs_min
; /* min devs needed */
4244 int devs_increment
; /* ndevs has to be a multiple of this */
4245 int ncopies
; /* how many copies to data has */
4247 u64 max_stripe_size
;
4251 u64 raid_stripe_len
= BTRFS_STRIPE_LEN
;
4257 BUG_ON(!alloc_profile_is_valid(type
, 0));
4259 if (list_empty(&fs_devices
->alloc_list
))
4262 index
= __get_raid_index(type
);
4264 sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
4265 dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
4266 devs_max
= btrfs_raid_array
[index
].devs_max
;
4267 devs_min
= btrfs_raid_array
[index
].devs_min
;
4268 devs_increment
= btrfs_raid_array
[index
].devs_increment
;
4269 ncopies
= btrfs_raid_array
[index
].ncopies
;
4271 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
4272 max_stripe_size
= 1024 * 1024 * 1024;
4273 max_chunk_size
= 10 * max_stripe_size
;
4275 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4276 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
4277 /* for larger filesystems, use larger metadata chunks */
4278 if (fs_devices
->total_rw_bytes
> 50ULL * 1024 * 1024 * 1024)
4279 max_stripe_size
= 1024 * 1024 * 1024;
4281 max_stripe_size
= 256 * 1024 * 1024;
4282 max_chunk_size
= max_stripe_size
;
4284 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4285 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4286 max_stripe_size
= 32 * 1024 * 1024;
4287 max_chunk_size
= 2 * max_stripe_size
;
4289 devs_max
= BTRFS_MAX_DEVS_SYS_CHUNK
;
4291 btrfs_err(info
, "invalid chunk type 0x%llx requested",
4296 /* we don't want a chunk larger than 10% of writeable space */
4297 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
4300 devices_info
= kzalloc(sizeof(*devices_info
) * fs_devices
->rw_devices
,
4305 cur
= fs_devices
->alloc_list
.next
;
4308 * in the first pass through the devices list, we gather information
4309 * about the available holes on each device.
4312 while (cur
!= &fs_devices
->alloc_list
) {
4313 struct btrfs_device
*device
;
4317 device
= list_entry(cur
, struct btrfs_device
, dev_alloc_list
);
4321 if (!device
->writeable
) {
4323 "BTRFS: read-only device in alloc_list\n");
4327 if (!device
->in_fs_metadata
||
4328 device
->is_tgtdev_for_dev_replace
)
4331 if (device
->total_bytes
> device
->bytes_used
)
4332 total_avail
= device
->total_bytes
- device
->bytes_used
;
4336 /* If there is no space on this device, skip it. */
4337 if (total_avail
== 0)
4340 ret
= find_free_dev_extent(trans
, device
,
4341 max_stripe_size
* dev_stripes
,
4342 &dev_offset
, &max_avail
);
4343 if (ret
&& ret
!= -ENOSPC
)
4347 max_avail
= max_stripe_size
* dev_stripes
;
4349 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
4352 if (ndevs
== fs_devices
->rw_devices
) {
4353 WARN(1, "%s: found more than %llu devices\n",
4354 __func__
, fs_devices
->rw_devices
);
4357 devices_info
[ndevs
].dev_offset
= dev_offset
;
4358 devices_info
[ndevs
].max_avail
= max_avail
;
4359 devices_info
[ndevs
].total_avail
= total_avail
;
4360 devices_info
[ndevs
].dev
= device
;
4365 * now sort the devices by hole size / available space
4367 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
4368 btrfs_cmp_device_info
, NULL
);
4370 /* round down to number of usable stripes */
4371 ndevs
-= ndevs
% devs_increment
;
4373 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
4378 if (devs_max
&& ndevs
> devs_max
)
4381 * the primary goal is to maximize the number of stripes, so use as many
4382 * devices as possible, even if the stripes are not maximum sized.
4384 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4385 num_stripes
= ndevs
* dev_stripes
;
4388 * this will have to be fixed for RAID1 and RAID10 over
4391 data_stripes
= num_stripes
/ ncopies
;
4393 if (type
& BTRFS_BLOCK_GROUP_RAID5
) {
4394 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 1,
4395 btrfs_super_stripesize(info
->super_copy
));
4396 data_stripes
= num_stripes
- 1;
4398 if (type
& BTRFS_BLOCK_GROUP_RAID6
) {
4399 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 2,
4400 btrfs_super_stripesize(info
->super_copy
));
4401 data_stripes
= num_stripes
- 2;
4405 * Use the number of data stripes to figure out how big this chunk
4406 * is really going to be in terms of logical address space,
4407 * and compare that answer with the max chunk size
4409 if (stripe_size
* data_stripes
> max_chunk_size
) {
4410 u64 mask
= (1ULL << 24) - 1;
4411 stripe_size
= max_chunk_size
;
4412 do_div(stripe_size
, data_stripes
);
4414 /* bump the answer up to a 16MB boundary */
4415 stripe_size
= (stripe_size
+ mask
) & ~mask
;
4417 /* but don't go higher than the limits we found
4418 * while searching for free extents
4420 if (stripe_size
> devices_info
[ndevs
-1].max_avail
)
4421 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4424 do_div(stripe_size
, dev_stripes
);
4426 /* align to BTRFS_STRIPE_LEN */
4427 do_div(stripe_size
, raid_stripe_len
);
4428 stripe_size
*= raid_stripe_len
;
4430 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
4435 map
->num_stripes
= num_stripes
;
4437 for (i
= 0; i
< ndevs
; ++i
) {
4438 for (j
= 0; j
< dev_stripes
; ++j
) {
4439 int s
= i
* dev_stripes
+ j
;
4440 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
4441 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
4445 map
->sector_size
= extent_root
->sectorsize
;
4446 map
->stripe_len
= raid_stripe_len
;
4447 map
->io_align
= raid_stripe_len
;
4448 map
->io_width
= raid_stripe_len
;
4450 map
->sub_stripes
= sub_stripes
;
4452 num_bytes
= stripe_size
* data_stripes
;
4454 trace_btrfs_chunk_alloc(info
->chunk_root
, map
, start
, num_bytes
);
4456 em
= alloc_extent_map();
4462 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
4463 em
->bdev
= (struct block_device
*)map
;
4465 em
->len
= num_bytes
;
4466 em
->block_start
= 0;
4467 em
->block_len
= em
->len
;
4468 em
->orig_block_len
= stripe_size
;
4470 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
4471 write_lock(&em_tree
->lock
);
4472 ret
= add_extent_mapping(em_tree
, em
, 0);
4474 list_add_tail(&em
->list
, &trans
->transaction
->pending_chunks
);
4475 atomic_inc(&em
->refs
);
4477 write_unlock(&em_tree
->lock
);
4479 free_extent_map(em
);
4483 ret
= btrfs_make_block_group(trans
, extent_root
, 0, type
,
4484 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4487 goto error_del_extent
;
4489 for (i
= 0; i
< map
->num_stripes
; i
++) {
4490 num_bytes
= map
->stripes
[i
].dev
->bytes_used
+ stripe_size
;
4491 btrfs_device_set_bytes_used(map
->stripes
[i
].dev
, num_bytes
);
4494 spin_lock(&extent_root
->fs_info
->free_chunk_lock
);
4495 extent_root
->fs_info
->free_chunk_space
-= (stripe_size
*
4497 spin_unlock(&extent_root
->fs_info
->free_chunk_lock
);
4499 free_extent_map(em
);
4500 check_raid56_incompat_flag(extent_root
->fs_info
, type
);
4502 kfree(devices_info
);
4506 write_lock(&em_tree
->lock
);
4507 remove_extent_mapping(em_tree
, em
);
4508 write_unlock(&em_tree
->lock
);
4510 /* One for our allocation */
4511 free_extent_map(em
);
4512 /* One for the tree reference */
4513 free_extent_map(em
);
4515 kfree(devices_info
);
4519 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
4520 struct btrfs_root
*extent_root
,
4521 u64 chunk_offset
, u64 chunk_size
)
4523 struct btrfs_key key
;
4524 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
4525 struct btrfs_device
*device
;
4526 struct btrfs_chunk
*chunk
;
4527 struct btrfs_stripe
*stripe
;
4528 struct extent_map_tree
*em_tree
;
4529 struct extent_map
*em
;
4530 struct map_lookup
*map
;
4537 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
4538 read_lock(&em_tree
->lock
);
4539 em
= lookup_extent_mapping(em_tree
, chunk_offset
, chunk_size
);
4540 read_unlock(&em_tree
->lock
);
4543 btrfs_crit(extent_root
->fs_info
, "unable to find logical "
4544 "%Lu len %Lu", chunk_offset
, chunk_size
);
4548 if (em
->start
!= chunk_offset
|| em
->len
!= chunk_size
) {
4549 btrfs_crit(extent_root
->fs_info
, "found a bad mapping, wanted"
4550 " %Lu-%Lu, found %Lu-%Lu", chunk_offset
,
4551 chunk_size
, em
->start
, em
->len
);
4552 free_extent_map(em
);
4556 map
= (struct map_lookup
*)em
->bdev
;
4557 item_size
= btrfs_chunk_item_size(map
->num_stripes
);
4558 stripe_size
= em
->orig_block_len
;
4560 chunk
= kzalloc(item_size
, GFP_NOFS
);
4566 for (i
= 0; i
< map
->num_stripes
; i
++) {
4567 device
= map
->stripes
[i
].dev
;
4568 dev_offset
= map
->stripes
[i
].physical
;
4570 ret
= btrfs_update_device(trans
, device
);
4573 ret
= btrfs_alloc_dev_extent(trans
, device
,
4574 chunk_root
->root_key
.objectid
,
4575 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4576 chunk_offset
, dev_offset
,
4582 stripe
= &chunk
->stripe
;
4583 for (i
= 0; i
< map
->num_stripes
; i
++) {
4584 device
= map
->stripes
[i
].dev
;
4585 dev_offset
= map
->stripes
[i
].physical
;
4587 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
4588 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
4589 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
4593 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
4594 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
4595 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
4596 btrfs_set_stack_chunk_type(chunk
, map
->type
);
4597 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
4598 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
4599 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
4600 btrfs_set_stack_chunk_sector_size(chunk
, extent_root
->sectorsize
);
4601 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
4603 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
4604 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
4605 key
.offset
= chunk_offset
;
4607 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
4608 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4610 * TODO: Cleanup of inserted chunk root in case of
4613 ret
= btrfs_add_system_chunk(chunk_root
, &key
, chunk
,
4619 free_extent_map(em
);
4624 * Chunk allocation falls into two parts. The first part does works
4625 * that make the new allocated chunk useable, but not do any operation
4626 * that modifies the chunk tree. The second part does the works that
4627 * require modifying the chunk tree. This division is important for the
4628 * bootstrap process of adding storage to a seed btrfs.
4630 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4631 struct btrfs_root
*extent_root
, u64 type
)
4635 chunk_offset
= find_next_chunk(extent_root
->fs_info
);
4636 return __btrfs_alloc_chunk(trans
, extent_root
, chunk_offset
, type
);
4639 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
4640 struct btrfs_root
*root
,
4641 struct btrfs_device
*device
)
4644 u64 sys_chunk_offset
;
4646 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4647 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
4650 chunk_offset
= find_next_chunk(fs_info
);
4651 alloc_profile
= btrfs_get_alloc_profile(extent_root
, 0);
4652 ret
= __btrfs_alloc_chunk(trans
, extent_root
, chunk_offset
,
4657 sys_chunk_offset
= find_next_chunk(root
->fs_info
);
4658 alloc_profile
= btrfs_get_alloc_profile(fs_info
->chunk_root
, 0);
4659 ret
= __btrfs_alloc_chunk(trans
, extent_root
, sys_chunk_offset
,
4664 static inline int btrfs_chunk_max_errors(struct map_lookup
*map
)
4668 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
4669 BTRFS_BLOCK_GROUP_RAID10
|
4670 BTRFS_BLOCK_GROUP_RAID5
|
4671 BTRFS_BLOCK_GROUP_DUP
)) {
4673 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
) {
4682 int btrfs_chunk_readonly(struct btrfs_root
*root
, u64 chunk_offset
)
4684 struct extent_map
*em
;
4685 struct map_lookup
*map
;
4686 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
4691 read_lock(&map_tree
->map_tree
.lock
);
4692 em
= lookup_extent_mapping(&map_tree
->map_tree
, chunk_offset
, 1);
4693 read_unlock(&map_tree
->map_tree
.lock
);
4697 map
= (struct map_lookup
*)em
->bdev
;
4698 for (i
= 0; i
< map
->num_stripes
; i
++) {
4699 if (map
->stripes
[i
].dev
->missing
) {
4704 if (!map
->stripes
[i
].dev
->writeable
) {
4711 * If the number of missing devices is larger than max errors,
4712 * we can not write the data into that chunk successfully, so
4715 if (miss_ndevs
> btrfs_chunk_max_errors(map
))
4718 free_extent_map(em
);
4722 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
4724 extent_map_tree_init(&tree
->map_tree
);
4727 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
4729 struct extent_map
*em
;
4732 write_lock(&tree
->map_tree
.lock
);
4733 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
4735 remove_extent_mapping(&tree
->map_tree
, em
);
4736 write_unlock(&tree
->map_tree
.lock
);
4740 free_extent_map(em
);
4741 /* once for the tree */
4742 free_extent_map(em
);
4746 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
4748 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
4749 struct extent_map
*em
;
4750 struct map_lookup
*map
;
4751 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4754 read_lock(&em_tree
->lock
);
4755 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4756 read_unlock(&em_tree
->lock
);
4759 * We could return errors for these cases, but that could get ugly and
4760 * we'd probably do the same thing which is just not do anything else
4761 * and exit, so return 1 so the callers don't try to use other copies.
4764 btrfs_crit(fs_info
, "No mapping for %Lu-%Lu", logical
,
4769 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
4770 btrfs_crit(fs_info
, "Invalid mapping for %Lu-%Lu, got "
4771 "%Lu-%Lu", logical
, logical
+len
, em
->start
,
4772 em
->start
+ em
->len
);
4773 free_extent_map(em
);
4777 map
= (struct map_lookup
*)em
->bdev
;
4778 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
4779 ret
= map
->num_stripes
;
4780 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
4781 ret
= map
->sub_stripes
;
4782 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID5
)
4784 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
4788 free_extent_map(em
);
4790 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
4791 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
))
4793 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
4798 unsigned long btrfs_full_stripe_len(struct btrfs_root
*root
,
4799 struct btrfs_mapping_tree
*map_tree
,
4802 struct extent_map
*em
;
4803 struct map_lookup
*map
;
4804 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4805 unsigned long len
= root
->sectorsize
;
4807 read_lock(&em_tree
->lock
);
4808 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4809 read_unlock(&em_tree
->lock
);
4812 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4813 map
= (struct map_lookup
*)em
->bdev
;
4814 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
4815 BTRFS_BLOCK_GROUP_RAID6
)) {
4816 len
= map
->stripe_len
* nr_data_stripes(map
);
4818 free_extent_map(em
);
4822 int btrfs_is_parity_mirror(struct btrfs_mapping_tree
*map_tree
,
4823 u64 logical
, u64 len
, int mirror_num
)
4825 struct extent_map
*em
;
4826 struct map_lookup
*map
;
4827 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4830 read_lock(&em_tree
->lock
);
4831 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4832 read_unlock(&em_tree
->lock
);
4835 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4836 map
= (struct map_lookup
*)em
->bdev
;
4837 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
4838 BTRFS_BLOCK_GROUP_RAID6
))
4840 free_extent_map(em
);
4844 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
4845 struct map_lookup
*map
, int first
, int num
,
4846 int optimal
, int dev_replace_is_ongoing
)
4850 struct btrfs_device
*srcdev
;
4852 if (dev_replace_is_ongoing
&&
4853 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
4854 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
4855 srcdev
= fs_info
->dev_replace
.srcdev
;
4860 * try to avoid the drive that is the source drive for a
4861 * dev-replace procedure, only choose it if no other non-missing
4862 * mirror is available
4864 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
4865 if (map
->stripes
[optimal
].dev
->bdev
&&
4866 (tolerance
|| map
->stripes
[optimal
].dev
!= srcdev
))
4868 for (i
= first
; i
< first
+ num
; i
++) {
4869 if (map
->stripes
[i
].dev
->bdev
&&
4870 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
4875 /* we couldn't find one that doesn't fail. Just return something
4876 * and the io error handling code will clean up eventually
4881 static inline int parity_smaller(u64 a
, u64 b
)
4886 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4887 static void sort_parity_stripes(struct btrfs_bio
*bbio
, u64
*raid_map
)
4889 struct btrfs_bio_stripe s
;
4896 for (i
= 0; i
< bbio
->num_stripes
- 1; i
++) {
4897 if (parity_smaller(raid_map
[i
], raid_map
[i
+1])) {
4898 s
= bbio
->stripes
[i
];
4900 bbio
->stripes
[i
] = bbio
->stripes
[i
+1];
4901 raid_map
[i
] = raid_map
[i
+1];
4902 bbio
->stripes
[i
+1] = s
;
4910 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
4911 u64 logical
, u64
*length
,
4912 struct btrfs_bio
**bbio_ret
,
4913 int mirror_num
, u64
**raid_map_ret
)
4915 struct extent_map
*em
;
4916 struct map_lookup
*map
;
4917 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
4918 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4921 u64 stripe_end_offset
;
4926 u64
*raid_map
= NULL
;
4932 struct btrfs_bio
*bbio
= NULL
;
4933 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
4934 int dev_replace_is_ongoing
= 0;
4935 int num_alloc_stripes
;
4936 int patch_the_first_stripe_for_dev_replace
= 0;
4937 u64 physical_to_patch_in_first_stripe
= 0;
4938 u64 raid56_full_stripe_start
= (u64
)-1;
4940 read_lock(&em_tree
->lock
);
4941 em
= lookup_extent_mapping(em_tree
, logical
, *length
);
4942 read_unlock(&em_tree
->lock
);
4945 btrfs_crit(fs_info
, "unable to find logical %llu len %llu",
4950 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
4951 btrfs_crit(fs_info
, "found a bad mapping, wanted %Lu, "
4952 "found %Lu-%Lu", logical
, em
->start
,
4953 em
->start
+ em
->len
);
4954 free_extent_map(em
);
4958 map
= (struct map_lookup
*)em
->bdev
;
4959 offset
= logical
- em
->start
;
4961 stripe_len
= map
->stripe_len
;
4964 * stripe_nr counts the total number of stripes we have to stride
4965 * to get to this block
4967 do_div(stripe_nr
, stripe_len
);
4969 stripe_offset
= stripe_nr
* stripe_len
;
4970 BUG_ON(offset
< stripe_offset
);
4972 /* stripe_offset is the offset of this block in its stripe*/
4973 stripe_offset
= offset
- stripe_offset
;
4975 /* if we're here for raid56, we need to know the stripe aligned start */
4976 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
)) {
4977 unsigned long full_stripe_len
= stripe_len
* nr_data_stripes(map
);
4978 raid56_full_stripe_start
= offset
;
4980 /* allow a write of a full stripe, but make sure we don't
4981 * allow straddling of stripes
4983 do_div(raid56_full_stripe_start
, full_stripe_len
);
4984 raid56_full_stripe_start
*= full_stripe_len
;
4987 if (rw
& REQ_DISCARD
) {
4988 /* we don't discard raid56 yet */
4990 (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
)) {
4994 *length
= min_t(u64
, em
->len
- offset
, *length
);
4995 } else if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
4997 /* For writes to RAID[56], allow a full stripeset across all disks.
4998 For other RAID types and for RAID[56] reads, just allow a single
4999 stripe (on a single disk). */
5000 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
) &&
5002 max_len
= stripe_len
* nr_data_stripes(map
) -
5003 (offset
- raid56_full_stripe_start
);
5005 /* we limit the length of each bio to what fits in a stripe */
5006 max_len
= stripe_len
- stripe_offset
;
5008 *length
= min_t(u64
, em
->len
- offset
, max_len
);
5010 *length
= em
->len
- offset
;
5013 /* This is for when we're called from btrfs_merge_bio_hook() and all
5014 it cares about is the length */
5018 btrfs_dev_replace_lock(dev_replace
);
5019 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
5020 if (!dev_replace_is_ongoing
)
5021 btrfs_dev_replace_unlock(dev_replace
);
5023 if (dev_replace_is_ongoing
&& mirror_num
== map
->num_stripes
+ 1 &&
5024 !(rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) &&
5025 dev_replace
->tgtdev
!= NULL
) {
5027 * in dev-replace case, for repair case (that's the only
5028 * case where the mirror is selected explicitly when
5029 * calling btrfs_map_block), blocks left of the left cursor
5030 * can also be read from the target drive.
5031 * For REQ_GET_READ_MIRRORS, the target drive is added as
5032 * the last one to the array of stripes. For READ, it also
5033 * needs to be supported using the same mirror number.
5034 * If the requested block is not left of the left cursor,
5035 * EIO is returned. This can happen because btrfs_num_copies()
5036 * returns one more in the dev-replace case.
5038 u64 tmp_length
= *length
;
5039 struct btrfs_bio
*tmp_bbio
= NULL
;
5040 int tmp_num_stripes
;
5041 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5042 int index_srcdev
= 0;
5044 u64 physical_of_found
= 0;
5046 ret
= __btrfs_map_block(fs_info
, REQ_GET_READ_MIRRORS
,
5047 logical
, &tmp_length
, &tmp_bbio
, 0, NULL
);
5049 WARN_ON(tmp_bbio
!= NULL
);
5053 tmp_num_stripes
= tmp_bbio
->num_stripes
;
5054 if (mirror_num
> tmp_num_stripes
) {
5056 * REQ_GET_READ_MIRRORS does not contain this
5057 * mirror, that means that the requested area
5058 * is not left of the left cursor
5066 * process the rest of the function using the mirror_num
5067 * of the source drive. Therefore look it up first.
5068 * At the end, patch the device pointer to the one of the
5071 for (i
= 0; i
< tmp_num_stripes
; i
++) {
5072 if (tmp_bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5074 * In case of DUP, in order to keep it
5075 * simple, only add the mirror with the
5076 * lowest physical address
5079 physical_of_found
<=
5080 tmp_bbio
->stripes
[i
].physical
)
5085 tmp_bbio
->stripes
[i
].physical
;
5090 mirror_num
= index_srcdev
+ 1;
5091 patch_the_first_stripe_for_dev_replace
= 1;
5092 physical_to_patch_in_first_stripe
= physical_of_found
;
5101 } else if (mirror_num
> map
->num_stripes
) {
5107 stripe_nr_orig
= stripe_nr
;
5108 stripe_nr_end
= ALIGN(offset
+ *length
, map
->stripe_len
);
5109 do_div(stripe_nr_end
, map
->stripe_len
);
5110 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
5113 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5114 if (rw
& REQ_DISCARD
)
5115 num_stripes
= min_t(u64
, map
->num_stripes
,
5116 stripe_nr_end
- stripe_nr_orig
);
5117 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
5118 if (!(rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)))
5120 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
5121 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
))
5122 num_stripes
= map
->num_stripes
;
5123 else if (mirror_num
)
5124 stripe_index
= mirror_num
- 1;
5126 stripe_index
= find_live_mirror(fs_info
, map
, 0,
5128 current
->pid
% map
->num_stripes
,
5129 dev_replace_is_ongoing
);
5130 mirror_num
= stripe_index
+ 1;
5133 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
5134 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) {
5135 num_stripes
= map
->num_stripes
;
5136 } else if (mirror_num
) {
5137 stripe_index
= mirror_num
- 1;
5142 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5143 int factor
= map
->num_stripes
/ map
->sub_stripes
;
5145 stripe_index
= do_div(stripe_nr
, factor
);
5146 stripe_index
*= map
->sub_stripes
;
5148 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
))
5149 num_stripes
= map
->sub_stripes
;
5150 else if (rw
& REQ_DISCARD
)
5151 num_stripes
= min_t(u64
, map
->sub_stripes
*
5152 (stripe_nr_end
- stripe_nr_orig
),
5154 else if (mirror_num
)
5155 stripe_index
+= mirror_num
- 1;
5157 int old_stripe_index
= stripe_index
;
5158 stripe_index
= find_live_mirror(fs_info
, map
,
5160 map
->sub_stripes
, stripe_index
+
5161 current
->pid
% map
->sub_stripes
,
5162 dev_replace_is_ongoing
);
5163 mirror_num
= stripe_index
- old_stripe_index
+ 1;
5166 } else if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
5167 BTRFS_BLOCK_GROUP_RAID6
)) {
5170 if (bbio_ret
&& ((rw
& REQ_WRITE
) || mirror_num
> 1)
5174 /* push stripe_nr back to the start of the full stripe */
5175 stripe_nr
= raid56_full_stripe_start
;
5176 do_div(stripe_nr
, stripe_len
);
5178 stripe_index
= do_div(stripe_nr
, nr_data_stripes(map
));
5180 /* RAID[56] write or recovery. Return all stripes */
5181 num_stripes
= map
->num_stripes
;
5182 max_errors
= nr_parity_stripes(map
);
5184 raid_map
= kmalloc_array(num_stripes
, sizeof(u64
),
5191 /* Work out the disk rotation on this stripe-set */
5193 rot
= do_div(tmp
, num_stripes
);
5195 /* Fill in the logical address of each stripe */
5196 tmp
= stripe_nr
* nr_data_stripes(map
);
5197 for (i
= 0; i
< nr_data_stripes(map
); i
++)
5198 raid_map
[(i
+rot
) % num_stripes
] =
5199 em
->start
+ (tmp
+ i
) * map
->stripe_len
;
5201 raid_map
[(i
+rot
) % map
->num_stripes
] = RAID5_P_STRIPE
;
5202 if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5203 raid_map
[(i
+rot
+1) % num_stripes
] =
5206 *length
= map
->stripe_len
;
5211 * Mirror #0 or #1 means the original data block.
5212 * Mirror #2 is RAID5 parity block.
5213 * Mirror #3 is RAID6 Q block.
5215 stripe_index
= do_div(stripe_nr
, nr_data_stripes(map
));
5217 stripe_index
= nr_data_stripes(map
) +
5220 /* We distribute the parity blocks across stripes */
5221 tmp
= stripe_nr
+ stripe_index
;
5222 stripe_index
= do_div(tmp
, map
->num_stripes
);
5223 if (!(rw
& (REQ_WRITE
| REQ_DISCARD
|
5224 REQ_GET_READ_MIRRORS
)) && mirror_num
<= 1)
5229 * after this do_div call, stripe_nr is the number of stripes
5230 * on this device we have to walk to find the data, and
5231 * stripe_index is the number of our device in the stripe array
5233 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
5234 mirror_num
= stripe_index
+ 1;
5236 BUG_ON(stripe_index
>= map
->num_stripes
);
5238 num_alloc_stripes
= num_stripes
;
5239 if (dev_replace_is_ongoing
) {
5240 if (rw
& (REQ_WRITE
| REQ_DISCARD
))
5241 num_alloc_stripes
<<= 1;
5242 if (rw
& REQ_GET_READ_MIRRORS
)
5243 num_alloc_stripes
++;
5245 bbio
= kzalloc(btrfs_bio_size(num_alloc_stripes
), GFP_NOFS
);
5251 atomic_set(&bbio
->error
, 0);
5253 if (rw
& REQ_DISCARD
) {
5255 int sub_stripes
= 0;
5256 u64 stripes_per_dev
= 0;
5257 u32 remaining_stripes
= 0;
5258 u32 last_stripe
= 0;
5261 (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID10
)) {
5262 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5265 sub_stripes
= map
->sub_stripes
;
5267 factor
= map
->num_stripes
/ sub_stripes
;
5268 stripes_per_dev
= div_u64_rem(stripe_nr_end
-
5271 &remaining_stripes
);
5272 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
5273 last_stripe
*= sub_stripes
;
5276 for (i
= 0; i
< num_stripes
; i
++) {
5277 bbio
->stripes
[i
].physical
=
5278 map
->stripes
[stripe_index
].physical
+
5279 stripe_offset
+ stripe_nr
* map
->stripe_len
;
5280 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
5282 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5283 BTRFS_BLOCK_GROUP_RAID10
)) {
5284 bbio
->stripes
[i
].length
= stripes_per_dev
*
5287 if (i
/ sub_stripes
< remaining_stripes
)
5288 bbio
->stripes
[i
].length
+=
5292 * Special for the first stripe and
5295 * |-------|...|-------|
5299 if (i
< sub_stripes
)
5300 bbio
->stripes
[i
].length
-=
5303 if (stripe_index
>= last_stripe
&&
5304 stripe_index
<= (last_stripe
+
5306 bbio
->stripes
[i
].length
-=
5309 if (i
== sub_stripes
- 1)
5312 bbio
->stripes
[i
].length
= *length
;
5315 if (stripe_index
== map
->num_stripes
) {
5316 /* This could only happen for RAID0/10 */
5322 for (i
= 0; i
< num_stripes
; i
++) {
5323 bbio
->stripes
[i
].physical
=
5324 map
->stripes
[stripe_index
].physical
+
5326 stripe_nr
* map
->stripe_len
;
5327 bbio
->stripes
[i
].dev
=
5328 map
->stripes
[stripe_index
].dev
;
5333 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
))
5334 max_errors
= btrfs_chunk_max_errors(map
);
5336 if (dev_replace_is_ongoing
&& (rw
& (REQ_WRITE
| REQ_DISCARD
)) &&
5337 dev_replace
->tgtdev
!= NULL
) {
5338 int index_where_to_add
;
5339 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5342 * duplicate the write operations while the dev replace
5343 * procedure is running. Since the copying of the old disk
5344 * to the new disk takes place at run time while the
5345 * filesystem is mounted writable, the regular write
5346 * operations to the old disk have to be duplicated to go
5347 * to the new disk as well.
5348 * Note that device->missing is handled by the caller, and
5349 * that the write to the old disk is already set up in the
5352 index_where_to_add
= num_stripes
;
5353 for (i
= 0; i
< num_stripes
; i
++) {
5354 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5355 /* write to new disk, too */
5356 struct btrfs_bio_stripe
*new =
5357 bbio
->stripes
+ index_where_to_add
;
5358 struct btrfs_bio_stripe
*old
=
5361 new->physical
= old
->physical
;
5362 new->length
= old
->length
;
5363 new->dev
= dev_replace
->tgtdev
;
5364 index_where_to_add
++;
5368 num_stripes
= index_where_to_add
;
5369 } else if (dev_replace_is_ongoing
&& (rw
& REQ_GET_READ_MIRRORS
) &&
5370 dev_replace
->tgtdev
!= NULL
) {
5371 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5372 int index_srcdev
= 0;
5374 u64 physical_of_found
= 0;
5377 * During the dev-replace procedure, the target drive can
5378 * also be used to read data in case it is needed to repair
5379 * a corrupt block elsewhere. This is possible if the
5380 * requested area is left of the left cursor. In this area,
5381 * the target drive is a full copy of the source drive.
5383 for (i
= 0; i
< num_stripes
; i
++) {
5384 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5386 * In case of DUP, in order to keep it
5387 * simple, only add the mirror with the
5388 * lowest physical address
5391 physical_of_found
<=
5392 bbio
->stripes
[i
].physical
)
5396 physical_of_found
= bbio
->stripes
[i
].physical
;
5400 u64 length
= map
->stripe_len
;
5402 if (physical_of_found
+ length
<=
5403 dev_replace
->cursor_left
) {
5404 struct btrfs_bio_stripe
*tgtdev_stripe
=
5405 bbio
->stripes
+ num_stripes
;
5407 tgtdev_stripe
->physical
= physical_of_found
;
5408 tgtdev_stripe
->length
=
5409 bbio
->stripes
[index_srcdev
].length
;
5410 tgtdev_stripe
->dev
= dev_replace
->tgtdev
;
5418 bbio
->num_stripes
= num_stripes
;
5419 bbio
->max_errors
= max_errors
;
5420 bbio
->mirror_num
= mirror_num
;
5423 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5424 * mirror_num == num_stripes + 1 && dev_replace target drive is
5425 * available as a mirror
5427 if (patch_the_first_stripe_for_dev_replace
&& num_stripes
> 0) {
5428 WARN_ON(num_stripes
> 1);
5429 bbio
->stripes
[0].dev
= dev_replace
->tgtdev
;
5430 bbio
->stripes
[0].physical
= physical_to_patch_in_first_stripe
;
5431 bbio
->mirror_num
= map
->num_stripes
+ 1;
5434 sort_parity_stripes(bbio
, raid_map
);
5435 *raid_map_ret
= raid_map
;
5438 if (dev_replace_is_ongoing
)
5439 btrfs_dev_replace_unlock(dev_replace
);
5440 free_extent_map(em
);
5444 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
5445 u64 logical
, u64
*length
,
5446 struct btrfs_bio
**bbio_ret
, int mirror_num
)
5448 return __btrfs_map_block(fs_info
, rw
, logical
, length
, bbio_ret
,
5452 int btrfs_rmap_block(struct btrfs_mapping_tree
*map_tree
,
5453 u64 chunk_start
, u64 physical
, u64 devid
,
5454 u64
**logical
, int *naddrs
, int *stripe_len
)
5456 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5457 struct extent_map
*em
;
5458 struct map_lookup
*map
;
5466 read_lock(&em_tree
->lock
);
5467 em
= lookup_extent_mapping(em_tree
, chunk_start
, 1);
5468 read_unlock(&em_tree
->lock
);
5471 printk(KERN_ERR
"BTRFS: couldn't find em for chunk %Lu\n",
5476 if (em
->start
!= chunk_start
) {
5477 printk(KERN_ERR
"BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5478 em
->start
, chunk_start
);
5479 free_extent_map(em
);
5482 map
= (struct map_lookup
*)em
->bdev
;
5485 rmap_len
= map
->stripe_len
;
5487 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5488 do_div(length
, map
->num_stripes
/ map
->sub_stripes
);
5489 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5490 do_div(length
, map
->num_stripes
);
5491 else if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
5492 BTRFS_BLOCK_GROUP_RAID6
)) {
5493 do_div(length
, nr_data_stripes(map
));
5494 rmap_len
= map
->stripe_len
* nr_data_stripes(map
);
5497 buf
= kzalloc(sizeof(u64
) * map
->num_stripes
, GFP_NOFS
);
5498 BUG_ON(!buf
); /* -ENOMEM */
5500 for (i
= 0; i
< map
->num_stripes
; i
++) {
5501 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
5503 if (map
->stripes
[i
].physical
> physical
||
5504 map
->stripes
[i
].physical
+ length
<= physical
)
5507 stripe_nr
= physical
- map
->stripes
[i
].physical
;
5508 do_div(stripe_nr
, map
->stripe_len
);
5510 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5511 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5512 do_div(stripe_nr
, map
->sub_stripes
);
5513 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5514 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5515 } /* else if RAID[56], multiply by nr_data_stripes().
5516 * Alternatively, just use rmap_len below instead of
5517 * map->stripe_len */
5519 bytenr
= chunk_start
+ stripe_nr
* rmap_len
;
5520 WARN_ON(nr
>= map
->num_stripes
);
5521 for (j
= 0; j
< nr
; j
++) {
5522 if (buf
[j
] == bytenr
)
5526 WARN_ON(nr
>= map
->num_stripes
);
5533 *stripe_len
= rmap_len
;
5535 free_extent_map(em
);
5539 static inline void btrfs_end_bbio(struct btrfs_bio
*bbio
, struct bio
*bio
, int err
)
5541 if (likely(bbio
->flags
& BTRFS_BIO_ORIG_BIO_SUBMITTED
))
5542 bio_endio_nodec(bio
, err
);
5544 bio_endio(bio
, err
);
5548 static void btrfs_end_bio(struct bio
*bio
, int err
)
5550 struct btrfs_bio
*bbio
= bio
->bi_private
;
5551 struct btrfs_device
*dev
= bbio
->stripes
[0].dev
;
5552 int is_orig_bio
= 0;
5555 atomic_inc(&bbio
->error
);
5556 if (err
== -EIO
|| err
== -EREMOTEIO
) {
5557 unsigned int stripe_index
=
5558 btrfs_io_bio(bio
)->stripe_index
;
5560 BUG_ON(stripe_index
>= bbio
->num_stripes
);
5561 dev
= bbio
->stripes
[stripe_index
].dev
;
5563 if (bio
->bi_rw
& WRITE
)
5564 btrfs_dev_stat_inc(dev
,
5565 BTRFS_DEV_STAT_WRITE_ERRS
);
5567 btrfs_dev_stat_inc(dev
,
5568 BTRFS_DEV_STAT_READ_ERRS
);
5569 if ((bio
->bi_rw
& WRITE_FLUSH
) == WRITE_FLUSH
)
5570 btrfs_dev_stat_inc(dev
,
5571 BTRFS_DEV_STAT_FLUSH_ERRS
);
5572 btrfs_dev_stat_print_on_error(dev
);
5577 if (bio
== bbio
->orig_bio
)
5580 btrfs_bio_counter_dec(bbio
->fs_info
);
5582 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
5585 bio
= bbio
->orig_bio
;
5588 bio
->bi_private
= bbio
->private;
5589 bio
->bi_end_io
= bbio
->end_io
;
5590 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
5591 /* only send an error to the higher layers if it is
5592 * beyond the tolerance of the btrfs bio
5594 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
5598 * this bio is actually up to date, we didn't
5599 * go over the max number of errors
5601 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5605 btrfs_end_bbio(bbio
, bio
, err
);
5606 } else if (!is_orig_bio
) {
5612 * see run_scheduled_bios for a description of why bios are collected for
5615 * This will add one bio to the pending list for a device and make sure
5616 * the work struct is scheduled.
5618 static noinline
void btrfs_schedule_bio(struct btrfs_root
*root
,
5619 struct btrfs_device
*device
,
5620 int rw
, struct bio
*bio
)
5622 int should_queue
= 1;
5623 struct btrfs_pending_bios
*pending_bios
;
5625 if (device
->missing
|| !device
->bdev
) {
5626 bio_endio(bio
, -EIO
);
5630 /* don't bother with additional async steps for reads, right now */
5631 if (!(rw
& REQ_WRITE
)) {
5633 btrfsic_submit_bio(rw
, bio
);
5639 * nr_async_bios allows us to reliably return congestion to the
5640 * higher layers. Otherwise, the async bio makes it appear we have
5641 * made progress against dirty pages when we've really just put it
5642 * on a queue for later
5644 atomic_inc(&root
->fs_info
->nr_async_bios
);
5645 WARN_ON(bio
->bi_next
);
5646 bio
->bi_next
= NULL
;
5649 spin_lock(&device
->io_lock
);
5650 if (bio
->bi_rw
& REQ_SYNC
)
5651 pending_bios
= &device
->pending_sync_bios
;
5653 pending_bios
= &device
->pending_bios
;
5655 if (pending_bios
->tail
)
5656 pending_bios
->tail
->bi_next
= bio
;
5658 pending_bios
->tail
= bio
;
5659 if (!pending_bios
->head
)
5660 pending_bios
->head
= bio
;
5661 if (device
->running_pending
)
5664 spin_unlock(&device
->io_lock
);
5667 btrfs_queue_work(root
->fs_info
->submit_workers
,
5671 static int bio_size_ok(struct block_device
*bdev
, struct bio
*bio
,
5674 struct bio_vec
*prev
;
5675 struct request_queue
*q
= bdev_get_queue(bdev
);
5676 unsigned int max_sectors
= queue_max_sectors(q
);
5677 struct bvec_merge_data bvm
= {
5679 .bi_sector
= sector
,
5680 .bi_rw
= bio
->bi_rw
,
5683 if (WARN_ON(bio
->bi_vcnt
== 0))
5686 prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
5687 if (bio_sectors(bio
) > max_sectors
)
5690 if (!q
->merge_bvec_fn
)
5693 bvm
.bi_size
= bio
->bi_iter
.bi_size
- prev
->bv_len
;
5694 if (q
->merge_bvec_fn(q
, &bvm
, prev
) < prev
->bv_len
)
5699 static void submit_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
5700 struct bio
*bio
, u64 physical
, int dev_nr
,
5703 struct btrfs_device
*dev
= bbio
->stripes
[dev_nr
].dev
;
5705 bio
->bi_private
= bbio
;
5706 btrfs_io_bio(bio
)->stripe_index
= dev_nr
;
5707 bio
->bi_end_io
= btrfs_end_bio
;
5708 bio
->bi_iter
.bi_sector
= physical
>> 9;
5711 struct rcu_string
*name
;
5714 name
= rcu_dereference(dev
->name
);
5715 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5716 "(%s id %llu), size=%u\n", rw
,
5717 (u64
)bio
->bi_iter
.bi_sector
, (u_long
)dev
->bdev
->bd_dev
,
5718 name
->str
, dev
->devid
, bio
->bi_iter
.bi_size
);
5722 bio
->bi_bdev
= dev
->bdev
;
5724 btrfs_bio_counter_inc_noblocked(root
->fs_info
);
5727 btrfs_schedule_bio(root
, dev
, rw
, bio
);
5729 btrfsic_submit_bio(rw
, bio
);
5732 static int breakup_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
5733 struct bio
*first_bio
, struct btrfs_device
*dev
,
5734 int dev_nr
, int rw
, int async
)
5736 struct bio_vec
*bvec
= first_bio
->bi_io_vec
;
5738 int nr_vecs
= bio_get_nr_vecs(dev
->bdev
);
5739 u64 physical
= bbio
->stripes
[dev_nr
].physical
;
5742 bio
= btrfs_bio_alloc(dev
->bdev
, physical
>> 9, nr_vecs
, GFP_NOFS
);
5746 while (bvec
<= (first_bio
->bi_io_vec
+ first_bio
->bi_vcnt
- 1)) {
5747 if (bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
5748 bvec
->bv_offset
) < bvec
->bv_len
) {
5749 u64 len
= bio
->bi_iter
.bi_size
;
5751 atomic_inc(&bbio
->stripes_pending
);
5752 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
,
5760 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
, rw
, async
);
5764 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
5766 atomic_inc(&bbio
->error
);
5767 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
5768 /* Shoud be the original bio. */
5769 WARN_ON(bio
!= bbio
->orig_bio
);
5771 bio
->bi_private
= bbio
->private;
5772 bio
->bi_end_io
= bbio
->end_io
;
5773 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
5774 bio
->bi_iter
.bi_sector
= logical
>> 9;
5776 btrfs_end_bbio(bbio
, bio
, -EIO
);
5780 int btrfs_map_bio(struct btrfs_root
*root
, int rw
, struct bio
*bio
,
5781 int mirror_num
, int async_submit
)
5783 struct btrfs_device
*dev
;
5784 struct bio
*first_bio
= bio
;
5785 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
5788 u64
*raid_map
= NULL
;
5792 struct btrfs_bio
*bbio
= NULL
;
5794 length
= bio
->bi_iter
.bi_size
;
5795 map_length
= length
;
5797 btrfs_bio_counter_inc_blocked(root
->fs_info
);
5798 ret
= __btrfs_map_block(root
->fs_info
, rw
, logical
, &map_length
, &bbio
,
5799 mirror_num
, &raid_map
);
5801 btrfs_bio_counter_dec(root
->fs_info
);
5805 total_devs
= bbio
->num_stripes
;
5806 bbio
->orig_bio
= first_bio
;
5807 bbio
->private = first_bio
->bi_private
;
5808 bbio
->end_io
= first_bio
->bi_end_io
;
5809 bbio
->fs_info
= root
->fs_info
;
5810 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
5813 /* In this case, map_length has been set to the length of
5814 a single stripe; not the whole write */
5816 ret
= raid56_parity_write(root
, bio
, bbio
,
5817 raid_map
, map_length
);
5819 ret
= raid56_parity_recover(root
, bio
, bbio
,
5820 raid_map
, map_length
,
5824 * FIXME, replace dosen't support raid56 yet, please fix
5827 btrfs_bio_counter_dec(root
->fs_info
);
5831 if (map_length
< length
) {
5832 btrfs_crit(root
->fs_info
, "mapping failed logical %llu bio len %llu len %llu",
5833 logical
, length
, map_length
);
5837 while (dev_nr
< total_devs
) {
5838 dev
= bbio
->stripes
[dev_nr
].dev
;
5839 if (!dev
|| !dev
->bdev
|| (rw
& WRITE
&& !dev
->writeable
)) {
5840 bbio_error(bbio
, first_bio
, logical
);
5846 * Check and see if we're ok with this bio based on it's size
5847 * and offset with the given device.
5849 if (!bio_size_ok(dev
->bdev
, first_bio
,
5850 bbio
->stripes
[dev_nr
].physical
>> 9)) {
5851 ret
= breakup_stripe_bio(root
, bbio
, first_bio
, dev
,
5852 dev_nr
, rw
, async_submit
);
5858 if (dev_nr
< total_devs
- 1) {
5859 bio
= btrfs_bio_clone(first_bio
, GFP_NOFS
);
5860 BUG_ON(!bio
); /* -ENOMEM */
5863 bbio
->flags
|= BTRFS_BIO_ORIG_BIO_SUBMITTED
;
5866 submit_stripe_bio(root
, bbio
, bio
,
5867 bbio
->stripes
[dev_nr
].physical
, dev_nr
, rw
,
5871 btrfs_bio_counter_dec(root
->fs_info
);
5875 struct btrfs_device
*btrfs_find_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
5878 struct btrfs_device
*device
;
5879 struct btrfs_fs_devices
*cur_devices
;
5881 cur_devices
= fs_info
->fs_devices
;
5882 while (cur_devices
) {
5884 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
5885 device
= __find_device(&cur_devices
->devices
,
5890 cur_devices
= cur_devices
->seed
;
5895 static struct btrfs_device
*add_missing_dev(struct btrfs_root
*root
,
5896 struct btrfs_fs_devices
*fs_devices
,
5897 u64 devid
, u8
*dev_uuid
)
5899 struct btrfs_device
*device
;
5901 device
= btrfs_alloc_device(NULL
, &devid
, dev_uuid
);
5905 list_add(&device
->dev_list
, &fs_devices
->devices
);
5906 device
->fs_devices
= fs_devices
;
5907 fs_devices
->num_devices
++;
5909 device
->missing
= 1;
5910 fs_devices
->missing_devices
++;
5916 * btrfs_alloc_device - allocate struct btrfs_device
5917 * @fs_info: used only for generating a new devid, can be NULL if
5918 * devid is provided (i.e. @devid != NULL).
5919 * @devid: a pointer to devid for this device. If NULL a new devid
5921 * @uuid: a pointer to UUID for this device. If NULL a new UUID
5924 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5925 * on error. Returned struct is not linked onto any lists and can be
5926 * destroyed with kfree() right away.
5928 struct btrfs_device
*btrfs_alloc_device(struct btrfs_fs_info
*fs_info
,
5932 struct btrfs_device
*dev
;
5935 if (WARN_ON(!devid
&& !fs_info
))
5936 return ERR_PTR(-EINVAL
);
5938 dev
= __alloc_device();
5947 ret
= find_next_devid(fs_info
, &tmp
);
5950 return ERR_PTR(ret
);
5956 memcpy(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
);
5958 generate_random_uuid(dev
->uuid
);
5960 btrfs_init_work(&dev
->work
, btrfs_submit_helper
,
5961 pending_bios_fn
, NULL
, NULL
);
5966 static int read_one_chunk(struct btrfs_root
*root
, struct btrfs_key
*key
,
5967 struct extent_buffer
*leaf
,
5968 struct btrfs_chunk
*chunk
)
5970 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
5971 struct map_lookup
*map
;
5972 struct extent_map
*em
;
5976 u8 uuid
[BTRFS_UUID_SIZE
];
5981 logical
= key
->offset
;
5982 length
= btrfs_chunk_length(leaf
, chunk
);
5984 read_lock(&map_tree
->map_tree
.lock
);
5985 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
5986 read_unlock(&map_tree
->map_tree
.lock
);
5988 /* already mapped? */
5989 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
5990 free_extent_map(em
);
5993 free_extent_map(em
);
5996 em
= alloc_extent_map();
5999 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6000 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
6002 free_extent_map(em
);
6006 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
6007 em
->bdev
= (struct block_device
*)map
;
6008 em
->start
= logical
;
6011 em
->block_start
= 0;
6012 em
->block_len
= em
->len
;
6014 map
->num_stripes
= num_stripes
;
6015 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
6016 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
6017 map
->sector_size
= btrfs_chunk_sector_size(leaf
, chunk
);
6018 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6019 map
->type
= btrfs_chunk_type(leaf
, chunk
);
6020 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6021 for (i
= 0; i
< num_stripes
; i
++) {
6022 map
->stripes
[i
].physical
=
6023 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
6024 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
6025 read_extent_buffer(leaf
, uuid
, (unsigned long)
6026 btrfs_stripe_dev_uuid_nr(chunk
, i
),
6028 map
->stripes
[i
].dev
= btrfs_find_device(root
->fs_info
, devid
,
6030 if (!map
->stripes
[i
].dev
&& !btrfs_test_opt(root
, DEGRADED
)) {
6031 free_extent_map(em
);
6034 if (!map
->stripes
[i
].dev
) {
6035 map
->stripes
[i
].dev
=
6036 add_missing_dev(root
, root
->fs_info
->fs_devices
,
6038 if (!map
->stripes
[i
].dev
) {
6039 free_extent_map(em
);
6043 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
6046 write_lock(&map_tree
->map_tree
.lock
);
6047 ret
= add_extent_mapping(&map_tree
->map_tree
, em
, 0);
6048 write_unlock(&map_tree
->map_tree
.lock
);
6049 BUG_ON(ret
); /* Tree corruption */
6050 free_extent_map(em
);
6055 static void fill_device_from_item(struct extent_buffer
*leaf
,
6056 struct btrfs_dev_item
*dev_item
,
6057 struct btrfs_device
*device
)
6061 device
->devid
= btrfs_device_id(leaf
, dev_item
);
6062 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
6063 device
->total_bytes
= device
->disk_total_bytes
;
6064 device
->commit_total_bytes
= device
->disk_total_bytes
;
6065 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
6066 device
->commit_bytes_used
= device
->bytes_used
;
6067 device
->type
= btrfs_device_type(leaf
, dev_item
);
6068 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
6069 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
6070 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
6071 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
6072 device
->is_tgtdev_for_dev_replace
= 0;
6074 ptr
= btrfs_device_uuid(dev_item
);
6075 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
6078 static struct btrfs_fs_devices
*open_seed_devices(struct btrfs_root
*root
,
6081 struct btrfs_fs_devices
*fs_devices
;
6084 BUG_ON(!mutex_is_locked(&uuid_mutex
));
6086 fs_devices
= root
->fs_info
->fs_devices
->seed
;
6087 while (fs_devices
) {
6088 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
))
6091 fs_devices
= fs_devices
->seed
;
6094 fs_devices
= find_fsid(fsid
);
6096 if (!btrfs_test_opt(root
, DEGRADED
))
6097 return ERR_PTR(-ENOENT
);
6099 fs_devices
= alloc_fs_devices(fsid
);
6100 if (IS_ERR(fs_devices
))
6103 fs_devices
->seeding
= 1;
6104 fs_devices
->opened
= 1;
6108 fs_devices
= clone_fs_devices(fs_devices
);
6109 if (IS_ERR(fs_devices
))
6112 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
6113 root
->fs_info
->bdev_holder
);
6115 free_fs_devices(fs_devices
);
6116 fs_devices
= ERR_PTR(ret
);
6120 if (!fs_devices
->seeding
) {
6121 __btrfs_close_devices(fs_devices
);
6122 free_fs_devices(fs_devices
);
6123 fs_devices
= ERR_PTR(-EINVAL
);
6127 fs_devices
->seed
= root
->fs_info
->fs_devices
->seed
;
6128 root
->fs_info
->fs_devices
->seed
= fs_devices
;
6133 static int read_one_dev(struct btrfs_root
*root
,
6134 struct extent_buffer
*leaf
,
6135 struct btrfs_dev_item
*dev_item
)
6137 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
6138 struct btrfs_device
*device
;
6141 u8 fs_uuid
[BTRFS_UUID_SIZE
];
6142 u8 dev_uuid
[BTRFS_UUID_SIZE
];
6144 devid
= btrfs_device_id(leaf
, dev_item
);
6145 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
6147 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
6150 if (memcmp(fs_uuid
, root
->fs_info
->fsid
, BTRFS_UUID_SIZE
)) {
6151 fs_devices
= open_seed_devices(root
, fs_uuid
);
6152 if (IS_ERR(fs_devices
))
6153 return PTR_ERR(fs_devices
);
6156 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
, fs_uuid
);
6158 if (!btrfs_test_opt(root
, DEGRADED
))
6161 btrfs_warn(root
->fs_info
, "devid %llu missing", devid
);
6162 device
= add_missing_dev(root
, fs_devices
, devid
, dev_uuid
);
6166 if (!device
->bdev
&& !btrfs_test_opt(root
, DEGRADED
))
6169 if(!device
->bdev
&& !device
->missing
) {
6171 * this happens when a device that was properly setup
6172 * in the device info lists suddenly goes bad.
6173 * device->bdev is NULL, and so we have to set
6174 * device->missing to one here
6176 device
->fs_devices
->missing_devices
++;
6177 device
->missing
= 1;
6180 /* Move the device to its own fs_devices */
6181 if (device
->fs_devices
!= fs_devices
) {
6182 ASSERT(device
->missing
);
6184 list_move(&device
->dev_list
, &fs_devices
->devices
);
6185 device
->fs_devices
->num_devices
--;
6186 fs_devices
->num_devices
++;
6188 device
->fs_devices
->missing_devices
--;
6189 fs_devices
->missing_devices
++;
6191 device
->fs_devices
= fs_devices
;
6195 if (device
->fs_devices
!= root
->fs_info
->fs_devices
) {
6196 BUG_ON(device
->writeable
);
6197 if (device
->generation
!=
6198 btrfs_device_generation(leaf
, dev_item
))
6202 fill_device_from_item(leaf
, dev_item
, device
);
6203 device
->in_fs_metadata
= 1;
6204 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
6205 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
6206 spin_lock(&root
->fs_info
->free_chunk_lock
);
6207 root
->fs_info
->free_chunk_space
+= device
->total_bytes
-
6209 spin_unlock(&root
->fs_info
->free_chunk_lock
);
6215 int btrfs_read_sys_array(struct btrfs_root
*root
)
6217 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
6218 struct extent_buffer
*sb
;
6219 struct btrfs_disk_key
*disk_key
;
6220 struct btrfs_chunk
*chunk
;
6222 unsigned long sb_ptr
;
6228 struct btrfs_key key
;
6230 sb
= btrfs_find_create_tree_block(root
, BTRFS_SUPER_INFO_OFFSET
,
6231 BTRFS_SUPER_INFO_SIZE
);
6234 btrfs_set_buffer_uptodate(sb
);
6235 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
6237 * The sb extent buffer is artifical and just used to read the system array.
6238 * btrfs_set_buffer_uptodate() call does not properly mark all it's
6239 * pages up-to-date when the page is larger: extent does not cover the
6240 * whole page and consequently check_page_uptodate does not find all
6241 * the page's extents up-to-date (the hole beyond sb),
6242 * write_extent_buffer then triggers a WARN_ON.
6244 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6245 * but sb spans only this function. Add an explicit SetPageUptodate call
6246 * to silence the warning eg. on PowerPC 64.
6248 if (PAGE_CACHE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
6249 SetPageUptodate(sb
->pages
[0]);
6251 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
6252 array_size
= btrfs_super_sys_array_size(super_copy
);
6254 ptr
= super_copy
->sys_chunk_array
;
6255 sb_ptr
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
6258 while (cur
< array_size
) {
6259 disk_key
= (struct btrfs_disk_key
*)ptr
;
6260 btrfs_disk_key_to_cpu(&key
, disk_key
);
6262 len
= sizeof(*disk_key
); ptr
+= len
;
6266 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6267 chunk
= (struct btrfs_chunk
*)sb_ptr
;
6268 ret
= read_one_chunk(root
, &key
, sb
, chunk
);
6271 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
6272 len
= btrfs_chunk_item_size(num_stripes
);
6281 free_extent_buffer(sb
);
6285 int btrfs_read_chunk_tree(struct btrfs_root
*root
)
6287 struct btrfs_path
*path
;
6288 struct extent_buffer
*leaf
;
6289 struct btrfs_key key
;
6290 struct btrfs_key found_key
;
6294 root
= root
->fs_info
->chunk_root
;
6296 path
= btrfs_alloc_path();
6300 mutex_lock(&uuid_mutex
);
6304 * Read all device items, and then all the chunk items. All
6305 * device items are found before any chunk item (their object id
6306 * is smaller than the lowest possible object id for a chunk
6307 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6309 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
6312 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6316 leaf
= path
->nodes
[0];
6317 slot
= path
->slots
[0];
6318 if (slot
>= btrfs_header_nritems(leaf
)) {
6319 ret
= btrfs_next_leaf(root
, path
);
6326 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6327 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
6328 struct btrfs_dev_item
*dev_item
;
6329 dev_item
= btrfs_item_ptr(leaf
, slot
,
6330 struct btrfs_dev_item
);
6331 ret
= read_one_dev(root
, leaf
, dev_item
);
6334 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6335 struct btrfs_chunk
*chunk
;
6336 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
6337 ret
= read_one_chunk(root
, &found_key
, leaf
, chunk
);
6345 unlock_chunks(root
);
6346 mutex_unlock(&uuid_mutex
);
6348 btrfs_free_path(path
);
6352 void btrfs_init_devices_late(struct btrfs_fs_info
*fs_info
)
6354 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6355 struct btrfs_device
*device
;
6357 while (fs_devices
) {
6358 mutex_lock(&fs_devices
->device_list_mutex
);
6359 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
)
6360 device
->dev_root
= fs_info
->dev_root
;
6361 mutex_unlock(&fs_devices
->device_list_mutex
);
6363 fs_devices
= fs_devices
->seed
;
6367 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
6371 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6372 btrfs_dev_stat_reset(dev
, i
);
6375 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
6377 struct btrfs_key key
;
6378 struct btrfs_key found_key
;
6379 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6380 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6381 struct extent_buffer
*eb
;
6384 struct btrfs_device
*device
;
6385 struct btrfs_path
*path
= NULL
;
6388 path
= btrfs_alloc_path();
6394 mutex_lock(&fs_devices
->device_list_mutex
);
6395 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6397 struct btrfs_dev_stats_item
*ptr
;
6400 key
.type
= BTRFS_DEV_STATS_KEY
;
6401 key
.offset
= device
->devid
;
6402 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
6404 __btrfs_reset_dev_stats(device
);
6405 device
->dev_stats_valid
= 1;
6406 btrfs_release_path(path
);
6409 slot
= path
->slots
[0];
6410 eb
= path
->nodes
[0];
6411 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
6412 item_size
= btrfs_item_size_nr(eb
, slot
);
6414 ptr
= btrfs_item_ptr(eb
, slot
,
6415 struct btrfs_dev_stats_item
);
6417 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6418 if (item_size
>= (1 + i
) * sizeof(__le64
))
6419 btrfs_dev_stat_set(device
, i
,
6420 btrfs_dev_stats_value(eb
, ptr
, i
));
6422 btrfs_dev_stat_reset(device
, i
);
6425 device
->dev_stats_valid
= 1;
6426 btrfs_dev_stat_print_on_load(device
);
6427 btrfs_release_path(path
);
6429 mutex_unlock(&fs_devices
->device_list_mutex
);
6432 btrfs_free_path(path
);
6433 return ret
< 0 ? ret
: 0;
6436 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
6437 struct btrfs_root
*dev_root
,
6438 struct btrfs_device
*device
)
6440 struct btrfs_path
*path
;
6441 struct btrfs_key key
;
6442 struct extent_buffer
*eb
;
6443 struct btrfs_dev_stats_item
*ptr
;
6448 key
.type
= BTRFS_DEV_STATS_KEY
;
6449 key
.offset
= device
->devid
;
6451 path
= btrfs_alloc_path();
6453 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
6455 printk_in_rcu(KERN_WARNING
"BTRFS: "
6456 "error %d while searching for dev_stats item for device %s!\n",
6457 ret
, rcu_str_deref(device
->name
));
6462 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
6463 /* need to delete old one and insert a new one */
6464 ret
= btrfs_del_item(trans
, dev_root
, path
);
6466 printk_in_rcu(KERN_WARNING
"BTRFS: "
6467 "delete too small dev_stats item for device %s failed %d!\n",
6468 rcu_str_deref(device
->name
), ret
);
6475 /* need to insert a new item */
6476 btrfs_release_path(path
);
6477 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
6478 &key
, sizeof(*ptr
));
6480 printk_in_rcu(KERN_WARNING
"BTRFS: "
6481 "insert dev_stats item for device %s failed %d!\n",
6482 rcu_str_deref(device
->name
), ret
);
6487 eb
= path
->nodes
[0];
6488 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
6489 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6490 btrfs_set_dev_stats_value(eb
, ptr
, i
,
6491 btrfs_dev_stat_read(device
, i
));
6492 btrfs_mark_buffer_dirty(eb
);
6495 btrfs_free_path(path
);
6500 * called from commit_transaction. Writes all changed device stats to disk.
6502 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
6503 struct btrfs_fs_info
*fs_info
)
6505 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6506 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6507 struct btrfs_device
*device
;
6511 mutex_lock(&fs_devices
->device_list_mutex
);
6512 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6513 if (!device
->dev_stats_valid
|| !btrfs_dev_stats_dirty(device
))
6516 stats_cnt
= atomic_read(&device
->dev_stats_ccnt
);
6517 ret
= update_dev_stat_item(trans
, dev_root
, device
);
6519 atomic_sub(stats_cnt
, &device
->dev_stats_ccnt
);
6521 mutex_unlock(&fs_devices
->device_list_mutex
);
6526 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
6528 btrfs_dev_stat_inc(dev
, index
);
6529 btrfs_dev_stat_print_on_error(dev
);
6532 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
6534 if (!dev
->dev_stats_valid
)
6536 printk_ratelimited_in_rcu(KERN_ERR
"BTRFS: "
6537 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6538 rcu_str_deref(dev
->name
),
6539 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
6540 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
6541 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
6542 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
6543 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
6546 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
6550 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6551 if (btrfs_dev_stat_read(dev
, i
) != 0)
6553 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
6554 return; /* all values == 0, suppress message */
6556 printk_in_rcu(KERN_INFO
"BTRFS: "
6557 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6558 rcu_str_deref(dev
->name
),
6559 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
6560 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
6561 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
6562 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
6563 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
6566 int btrfs_get_dev_stats(struct btrfs_root
*root
,
6567 struct btrfs_ioctl_get_dev_stats
*stats
)
6569 struct btrfs_device
*dev
;
6570 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
6573 mutex_lock(&fs_devices
->device_list_mutex
);
6574 dev
= btrfs_find_device(root
->fs_info
, stats
->devid
, NULL
, NULL
);
6575 mutex_unlock(&fs_devices
->device_list_mutex
);
6578 btrfs_warn(root
->fs_info
, "get dev_stats failed, device not found");
6580 } else if (!dev
->dev_stats_valid
) {
6581 btrfs_warn(root
->fs_info
, "get dev_stats failed, not yet valid");
6583 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
6584 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6585 if (stats
->nr_items
> i
)
6587 btrfs_dev_stat_read_and_reset(dev
, i
);
6589 btrfs_dev_stat_reset(dev
, i
);
6592 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6593 if (stats
->nr_items
> i
)
6594 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
6596 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
6597 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
6601 int btrfs_scratch_superblock(struct btrfs_device
*device
)
6603 struct buffer_head
*bh
;
6604 struct btrfs_super_block
*disk_super
;
6606 bh
= btrfs_read_dev_super(device
->bdev
);
6609 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
6611 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
6612 set_buffer_dirty(bh
);
6613 sync_dirty_buffer(bh
);
6620 * Update the size of all devices, which is used for writing out the
6623 void btrfs_update_commit_device_size(struct btrfs_fs_info
*fs_info
)
6625 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6626 struct btrfs_device
*curr
, *next
;
6628 if (list_empty(&fs_devices
->resized_devices
))
6631 mutex_lock(&fs_devices
->device_list_mutex
);
6632 lock_chunks(fs_info
->dev_root
);
6633 list_for_each_entry_safe(curr
, next
, &fs_devices
->resized_devices
,
6635 list_del_init(&curr
->resized_list
);
6636 curr
->commit_total_bytes
= curr
->disk_total_bytes
;
6638 unlock_chunks(fs_info
->dev_root
);
6639 mutex_unlock(&fs_devices
->device_list_mutex
);
6642 /* Must be invoked during the transaction commit */
6643 void btrfs_update_commit_device_bytes_used(struct btrfs_root
*root
,
6644 struct btrfs_transaction
*transaction
)
6646 struct extent_map
*em
;
6647 struct map_lookup
*map
;
6648 struct btrfs_device
*dev
;
6651 if (list_empty(&transaction
->pending_chunks
))
6654 /* In order to kick the device replace finish process */
6656 list_for_each_entry(em
, &transaction
->pending_chunks
, list
) {
6657 map
= (struct map_lookup
*)em
->bdev
;
6659 for (i
= 0; i
< map
->num_stripes
; i
++) {
6660 dev
= map
->stripes
[i
].dev
;
6661 dev
->commit_bytes_used
= dev
->bytes_used
;
6664 unlock_chunks(root
);