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
);
55 struct list_head
*btrfs_get_fs_uuids(void)
60 static struct btrfs_fs_devices
*__alloc_fs_devices(void)
62 struct btrfs_fs_devices
*fs_devs
;
64 fs_devs
= kzalloc(sizeof(*fs_devs
), GFP_NOFS
);
66 return ERR_PTR(-ENOMEM
);
68 mutex_init(&fs_devs
->device_list_mutex
);
70 INIT_LIST_HEAD(&fs_devs
->devices
);
71 INIT_LIST_HEAD(&fs_devs
->resized_devices
);
72 INIT_LIST_HEAD(&fs_devs
->alloc_list
);
73 INIT_LIST_HEAD(&fs_devs
->list
);
79 * alloc_fs_devices - allocate struct btrfs_fs_devices
80 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
83 * Return: a pointer to a new &struct btrfs_fs_devices on success;
84 * ERR_PTR() on error. Returned struct is not linked onto any lists and
85 * can be destroyed with kfree() right away.
87 static struct btrfs_fs_devices
*alloc_fs_devices(const u8
*fsid
)
89 struct btrfs_fs_devices
*fs_devs
;
91 fs_devs
= __alloc_fs_devices();
96 memcpy(fs_devs
->fsid
, fsid
, BTRFS_FSID_SIZE
);
98 generate_random_uuid(fs_devs
->fsid
);
103 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
105 struct btrfs_device
*device
;
106 WARN_ON(fs_devices
->opened
);
107 while (!list_empty(&fs_devices
->devices
)) {
108 device
= list_entry(fs_devices
->devices
.next
,
109 struct btrfs_device
, dev_list
);
110 list_del(&device
->dev_list
);
111 rcu_string_free(device
->name
);
117 static void btrfs_kobject_uevent(struct block_device
*bdev
,
118 enum kobject_action action
)
122 ret
= kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, action
);
124 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
126 kobject_name(&disk_to_dev(bdev
->bd_disk
)->kobj
),
127 &disk_to_dev(bdev
->bd_disk
)->kobj
);
130 void btrfs_cleanup_fs_uuids(void)
132 struct btrfs_fs_devices
*fs_devices
;
134 while (!list_empty(&fs_uuids
)) {
135 fs_devices
= list_entry(fs_uuids
.next
,
136 struct btrfs_fs_devices
, list
);
137 list_del(&fs_devices
->list
);
138 free_fs_devices(fs_devices
);
142 static struct btrfs_device
*__alloc_device(void)
144 struct btrfs_device
*dev
;
146 dev
= kzalloc(sizeof(*dev
), GFP_NOFS
);
148 return ERR_PTR(-ENOMEM
);
150 INIT_LIST_HEAD(&dev
->dev_list
);
151 INIT_LIST_HEAD(&dev
->dev_alloc_list
);
152 INIT_LIST_HEAD(&dev
->resized_list
);
154 spin_lock_init(&dev
->io_lock
);
156 spin_lock_init(&dev
->reada_lock
);
157 atomic_set(&dev
->reada_in_flight
, 0);
158 atomic_set(&dev
->dev_stats_ccnt
, 0);
159 INIT_RADIX_TREE(&dev
->reada_zones
, GFP_NOFS
& ~__GFP_WAIT
);
160 INIT_RADIX_TREE(&dev
->reada_extents
, GFP_NOFS
& ~__GFP_WAIT
);
165 static noinline
struct btrfs_device
*__find_device(struct list_head
*head
,
168 struct btrfs_device
*dev
;
170 list_for_each_entry(dev
, head
, dev_list
) {
171 if (dev
->devid
== devid
&&
172 (!uuid
|| !memcmp(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
))) {
179 static noinline
struct btrfs_fs_devices
*find_fsid(u8
*fsid
)
181 struct btrfs_fs_devices
*fs_devices
;
183 list_for_each_entry(fs_devices
, &fs_uuids
, list
) {
184 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
191 btrfs_get_bdev_and_sb(const char *device_path
, fmode_t flags
, void *holder
,
192 int flush
, struct block_device
**bdev
,
193 struct buffer_head
**bh
)
197 *bdev
= blkdev_get_by_path(device_path
, flags
, holder
);
200 ret
= PTR_ERR(*bdev
);
201 printk(KERN_INFO
"BTRFS: open %s failed\n", device_path
);
206 filemap_write_and_wait((*bdev
)->bd_inode
->i_mapping
);
207 ret
= set_blocksize(*bdev
, 4096);
209 blkdev_put(*bdev
, flags
);
212 invalidate_bdev(*bdev
);
213 *bh
= btrfs_read_dev_super(*bdev
);
216 blkdev_put(*bdev
, flags
);
228 static void requeue_list(struct btrfs_pending_bios
*pending_bios
,
229 struct bio
*head
, struct bio
*tail
)
232 struct bio
*old_head
;
234 old_head
= pending_bios
->head
;
235 pending_bios
->head
= head
;
236 if (pending_bios
->tail
)
237 tail
->bi_next
= old_head
;
239 pending_bios
->tail
= tail
;
243 * we try to collect pending bios for a device so we don't get a large
244 * number of procs sending bios down to the same device. This greatly
245 * improves the schedulers ability to collect and merge the bios.
247 * But, it also turns into a long list of bios to process and that is sure
248 * to eventually make the worker thread block. The solution here is to
249 * make some progress and then put this work struct back at the end of
250 * the list if the block device is congested. This way, multiple devices
251 * can make progress from a single worker thread.
253 static noinline
void run_scheduled_bios(struct btrfs_device
*device
)
256 struct backing_dev_info
*bdi
;
257 struct btrfs_fs_info
*fs_info
;
258 struct btrfs_pending_bios
*pending_bios
;
262 unsigned long num_run
;
263 unsigned long batch_run
= 0;
265 unsigned long last_waited
= 0;
267 int sync_pending
= 0;
268 struct blk_plug plug
;
271 * this function runs all the bios we've collected for
272 * a particular device. We don't want to wander off to
273 * another device without first sending all of these down.
274 * So, setup a plug here and finish it off before we return
276 blk_start_plug(&plug
);
278 bdi
= blk_get_backing_dev_info(device
->bdev
);
279 fs_info
= device
->dev_root
->fs_info
;
280 limit
= btrfs_async_submit_limit(fs_info
);
281 limit
= limit
* 2 / 3;
284 spin_lock(&device
->io_lock
);
289 /* take all the bios off the list at once and process them
290 * later on (without the lock held). But, remember the
291 * tail and other pointers so the bios can be properly reinserted
292 * into the list if we hit congestion
294 if (!force_reg
&& device
->pending_sync_bios
.head
) {
295 pending_bios
= &device
->pending_sync_bios
;
298 pending_bios
= &device
->pending_bios
;
302 pending
= pending_bios
->head
;
303 tail
= pending_bios
->tail
;
304 WARN_ON(pending
&& !tail
);
307 * if pending was null this time around, no bios need processing
308 * at all and we can stop. Otherwise it'll loop back up again
309 * and do an additional check so no bios are missed.
311 * device->running_pending is used to synchronize with the
314 if (device
->pending_sync_bios
.head
== NULL
&&
315 device
->pending_bios
.head
== NULL
) {
317 device
->running_pending
= 0;
320 device
->running_pending
= 1;
323 pending_bios
->head
= NULL
;
324 pending_bios
->tail
= NULL
;
326 spin_unlock(&device
->io_lock
);
331 /* we want to work on both lists, but do more bios on the
332 * sync list than the regular list
335 pending_bios
!= &device
->pending_sync_bios
&&
336 device
->pending_sync_bios
.head
) ||
337 (num_run
> 64 && pending_bios
== &device
->pending_sync_bios
&&
338 device
->pending_bios
.head
)) {
339 spin_lock(&device
->io_lock
);
340 requeue_list(pending_bios
, pending
, tail
);
345 pending
= pending
->bi_next
;
348 if (atomic_dec_return(&fs_info
->nr_async_bios
) < limit
&&
349 waitqueue_active(&fs_info
->async_submit_wait
))
350 wake_up(&fs_info
->async_submit_wait
);
352 BUG_ON(atomic_read(&cur
->bi_cnt
) == 0);
355 * if we're doing the sync list, record that our
356 * plug has some sync requests on it
358 * If we're doing the regular list and there are
359 * sync requests sitting around, unplug before
362 if (pending_bios
== &device
->pending_sync_bios
) {
364 } else if (sync_pending
) {
365 blk_finish_plug(&plug
);
366 blk_start_plug(&plug
);
370 btrfsic_submit_bio(cur
->bi_rw
, cur
);
377 * we made progress, there is more work to do and the bdi
378 * is now congested. Back off and let other work structs
381 if (pending
&& bdi_write_congested(bdi
) && batch_run
> 8 &&
382 fs_info
->fs_devices
->open_devices
> 1) {
383 struct io_context
*ioc
;
385 ioc
= current
->io_context
;
388 * the main goal here is that we don't want to
389 * block if we're going to be able to submit
390 * more requests without blocking.
392 * This code does two great things, it pokes into
393 * the elevator code from a filesystem _and_
394 * it makes assumptions about how batching works.
396 if (ioc
&& ioc
->nr_batch_requests
> 0 &&
397 time_before(jiffies
, ioc
->last_waited
+ HZ
/50UL) &&
399 ioc
->last_waited
== last_waited
)) {
401 * we want to go through our batch of
402 * requests and stop. So, we copy out
403 * the ioc->last_waited time and test
404 * against it before looping
406 last_waited
= ioc
->last_waited
;
410 spin_lock(&device
->io_lock
);
411 requeue_list(pending_bios
, pending
, tail
);
412 device
->running_pending
= 1;
414 spin_unlock(&device
->io_lock
);
415 btrfs_queue_work(fs_info
->submit_workers
,
419 /* unplug every 64 requests just for good measure */
420 if (batch_run
% 64 == 0) {
421 blk_finish_plug(&plug
);
422 blk_start_plug(&plug
);
431 spin_lock(&device
->io_lock
);
432 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
434 spin_unlock(&device
->io_lock
);
437 blk_finish_plug(&plug
);
440 static void pending_bios_fn(struct btrfs_work
*work
)
442 struct btrfs_device
*device
;
444 device
= container_of(work
, struct btrfs_device
, work
);
445 run_scheduled_bios(device
);
449 * Add new device to list of registered devices
452 * 1 - first time device is seen
453 * 0 - device already known
456 static noinline
int device_list_add(const char *path
,
457 struct btrfs_super_block
*disk_super
,
458 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
460 struct btrfs_device
*device
;
461 struct btrfs_fs_devices
*fs_devices
;
462 struct rcu_string
*name
;
464 u64 found_transid
= btrfs_super_generation(disk_super
);
466 fs_devices
= find_fsid(disk_super
->fsid
);
468 fs_devices
= alloc_fs_devices(disk_super
->fsid
);
469 if (IS_ERR(fs_devices
))
470 return PTR_ERR(fs_devices
);
472 list_add(&fs_devices
->list
, &fs_uuids
);
476 device
= __find_device(&fs_devices
->devices
, devid
,
477 disk_super
->dev_item
.uuid
);
481 if (fs_devices
->opened
)
484 device
= btrfs_alloc_device(NULL
, &devid
,
485 disk_super
->dev_item
.uuid
);
486 if (IS_ERR(device
)) {
487 /* we can safely leave the fs_devices entry around */
488 return PTR_ERR(device
);
491 name
= rcu_string_strdup(path
, GFP_NOFS
);
496 rcu_assign_pointer(device
->name
, name
);
498 mutex_lock(&fs_devices
->device_list_mutex
);
499 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
500 fs_devices
->num_devices
++;
501 mutex_unlock(&fs_devices
->device_list_mutex
);
504 device
->fs_devices
= fs_devices
;
505 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
507 * When FS is already mounted.
508 * 1. If you are here and if the device->name is NULL that
509 * means this device was missing at time of FS mount.
510 * 2. If you are here and if the device->name is different
511 * from 'path' that means either
512 * a. The same device disappeared and reappeared with
514 * b. The missing-disk-which-was-replaced, has
517 * We must allow 1 and 2a above. But 2b would be a spurious
520 * Further in case of 1 and 2a above, the disk at 'path'
521 * would have missed some transaction when it was away and
522 * in case of 2a the stale bdev has to be updated as well.
523 * 2b must not be allowed at all time.
527 * For now, we do allow update to btrfs_fs_device through the
528 * btrfs dev scan cli after FS has been mounted. We're still
529 * tracking a problem where systems fail mount by subvolume id
530 * when we reject replacement on a mounted FS.
532 if (!fs_devices
->opened
&& found_transid
< device
->generation
) {
534 * That is if the FS is _not_ mounted and if you
535 * are here, that means there is more than one
536 * disk with same uuid and devid.We keep the one
537 * with larger generation number or the last-in if
538 * generation are equal.
543 name
= rcu_string_strdup(path
, GFP_NOFS
);
546 rcu_string_free(device
->name
);
547 rcu_assign_pointer(device
->name
, name
);
548 if (device
->missing
) {
549 fs_devices
->missing_devices
--;
555 * Unmount does not free the btrfs_device struct but would zero
556 * generation along with most of the other members. So just update
557 * it back. We need it to pick the disk with largest generation
560 if (!fs_devices
->opened
)
561 device
->generation
= found_transid
;
563 *fs_devices_ret
= fs_devices
;
568 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
570 struct btrfs_fs_devices
*fs_devices
;
571 struct btrfs_device
*device
;
572 struct btrfs_device
*orig_dev
;
574 fs_devices
= alloc_fs_devices(orig
->fsid
);
575 if (IS_ERR(fs_devices
))
578 mutex_lock(&orig
->device_list_mutex
);
579 fs_devices
->total_devices
= orig
->total_devices
;
581 /* We have held the volume lock, it is safe to get the devices. */
582 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
583 struct rcu_string
*name
;
585 device
= btrfs_alloc_device(NULL
, &orig_dev
->devid
,
591 * This is ok to do without rcu read locked because we hold the
592 * uuid mutex so nothing we touch in here is going to disappear.
594 if (orig_dev
->name
) {
595 name
= rcu_string_strdup(orig_dev
->name
->str
, GFP_NOFS
);
600 rcu_assign_pointer(device
->name
, name
);
603 list_add(&device
->dev_list
, &fs_devices
->devices
);
604 device
->fs_devices
= fs_devices
;
605 fs_devices
->num_devices
++;
607 mutex_unlock(&orig
->device_list_mutex
);
610 mutex_unlock(&orig
->device_list_mutex
);
611 free_fs_devices(fs_devices
);
612 return ERR_PTR(-ENOMEM
);
615 void btrfs_close_extra_devices(struct btrfs_fs_devices
*fs_devices
, int step
)
617 struct btrfs_device
*device
, *next
;
618 struct btrfs_device
*latest_dev
= NULL
;
620 mutex_lock(&uuid_mutex
);
622 /* This is the initialized path, it is safe to release the devices. */
623 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
624 if (device
->in_fs_metadata
) {
625 if (!device
->is_tgtdev_for_dev_replace
&&
627 device
->generation
> latest_dev
->generation
)) {
633 if (device
->devid
== BTRFS_DEV_REPLACE_DEVID
) {
635 * In the first step, keep the device which has
636 * the correct fsid and the devid that is used
637 * for the dev_replace procedure.
638 * In the second step, the dev_replace state is
639 * read from the device tree and it is known
640 * whether the procedure is really active or
641 * not, which means whether this device is
642 * used or whether it should be removed.
644 if (step
== 0 || device
->is_tgtdev_for_dev_replace
) {
649 blkdev_put(device
->bdev
, device
->mode
);
651 fs_devices
->open_devices
--;
653 if (device
->writeable
) {
654 list_del_init(&device
->dev_alloc_list
);
655 device
->writeable
= 0;
656 if (!device
->is_tgtdev_for_dev_replace
)
657 fs_devices
->rw_devices
--;
659 list_del_init(&device
->dev_list
);
660 fs_devices
->num_devices
--;
661 rcu_string_free(device
->name
);
665 if (fs_devices
->seed
) {
666 fs_devices
= fs_devices
->seed
;
670 fs_devices
->latest_bdev
= latest_dev
->bdev
;
672 mutex_unlock(&uuid_mutex
);
675 static void __free_device(struct work_struct
*work
)
677 struct btrfs_device
*device
;
679 device
= container_of(work
, struct btrfs_device
, rcu_work
);
682 blkdev_put(device
->bdev
, device
->mode
);
684 rcu_string_free(device
->name
);
688 static void free_device(struct rcu_head
*head
)
690 struct btrfs_device
*device
;
692 device
= container_of(head
, struct btrfs_device
, rcu
);
694 INIT_WORK(&device
->rcu_work
, __free_device
);
695 schedule_work(&device
->rcu_work
);
698 static int __btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
700 struct btrfs_device
*device
;
702 if (--fs_devices
->opened
> 0)
705 mutex_lock(&fs_devices
->device_list_mutex
);
706 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
707 struct btrfs_device
*new_device
;
708 struct rcu_string
*name
;
711 fs_devices
->open_devices
--;
713 if (device
->writeable
&&
714 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
715 list_del_init(&device
->dev_alloc_list
);
716 fs_devices
->rw_devices
--;
720 fs_devices
->missing_devices
--;
722 new_device
= btrfs_alloc_device(NULL
, &device
->devid
,
724 BUG_ON(IS_ERR(new_device
)); /* -ENOMEM */
726 /* Safe because we are under uuid_mutex */
728 name
= rcu_string_strdup(device
->name
->str
, GFP_NOFS
);
729 BUG_ON(!name
); /* -ENOMEM */
730 rcu_assign_pointer(new_device
->name
, name
);
733 list_replace_rcu(&device
->dev_list
, &new_device
->dev_list
);
734 new_device
->fs_devices
= device
->fs_devices
;
736 call_rcu(&device
->rcu
, free_device
);
738 mutex_unlock(&fs_devices
->device_list_mutex
);
740 WARN_ON(fs_devices
->open_devices
);
741 WARN_ON(fs_devices
->rw_devices
);
742 fs_devices
->opened
= 0;
743 fs_devices
->seeding
= 0;
748 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
750 struct btrfs_fs_devices
*seed_devices
= NULL
;
753 mutex_lock(&uuid_mutex
);
754 ret
= __btrfs_close_devices(fs_devices
);
755 if (!fs_devices
->opened
) {
756 seed_devices
= fs_devices
->seed
;
757 fs_devices
->seed
= NULL
;
759 mutex_unlock(&uuid_mutex
);
761 while (seed_devices
) {
762 fs_devices
= seed_devices
;
763 seed_devices
= fs_devices
->seed
;
764 __btrfs_close_devices(fs_devices
);
765 free_fs_devices(fs_devices
);
768 * Wait for rcu kworkers under __btrfs_close_devices
769 * to finish all blkdev_puts so device is really
770 * free when umount is done.
776 static int __btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
777 fmode_t flags
, void *holder
)
779 struct request_queue
*q
;
780 struct block_device
*bdev
;
781 struct list_head
*head
= &fs_devices
->devices
;
782 struct btrfs_device
*device
;
783 struct btrfs_device
*latest_dev
= NULL
;
784 struct buffer_head
*bh
;
785 struct btrfs_super_block
*disk_super
;
792 list_for_each_entry(device
, head
, dev_list
) {
798 /* Just open everything we can; ignore failures here */
799 if (btrfs_get_bdev_and_sb(device
->name
->str
, flags
, holder
, 1,
803 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
804 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
805 if (devid
!= device
->devid
)
808 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
,
812 device
->generation
= btrfs_super_generation(disk_super
);
814 device
->generation
> latest_dev
->generation
)
817 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
818 device
->writeable
= 0;
820 device
->writeable
= !bdev_read_only(bdev
);
824 q
= bdev_get_queue(bdev
);
825 if (blk_queue_discard(q
))
826 device
->can_discard
= 1;
829 device
->in_fs_metadata
= 0;
830 device
->mode
= flags
;
832 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
833 fs_devices
->rotating
= 1;
835 fs_devices
->open_devices
++;
836 if (device
->writeable
&&
837 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
838 fs_devices
->rw_devices
++;
839 list_add(&device
->dev_alloc_list
,
840 &fs_devices
->alloc_list
);
847 blkdev_put(bdev
, flags
);
850 if (fs_devices
->open_devices
== 0) {
854 fs_devices
->seeding
= seeding
;
855 fs_devices
->opened
= 1;
856 fs_devices
->latest_bdev
= latest_dev
->bdev
;
857 fs_devices
->total_rw_bytes
= 0;
862 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
863 fmode_t flags
, void *holder
)
867 mutex_lock(&uuid_mutex
);
868 if (fs_devices
->opened
) {
869 fs_devices
->opened
++;
872 ret
= __btrfs_open_devices(fs_devices
, flags
, holder
);
874 mutex_unlock(&uuid_mutex
);
879 * Look for a btrfs signature on a device. This may be called out of the mount path
880 * and we are not allowed to call set_blocksize during the scan. The superblock
881 * is read via pagecache
883 int btrfs_scan_one_device(const char *path
, fmode_t flags
, void *holder
,
884 struct btrfs_fs_devices
**fs_devices_ret
)
886 struct btrfs_super_block
*disk_super
;
887 struct block_device
*bdev
;
898 * we would like to check all the supers, but that would make
899 * a btrfs mount succeed after a mkfs from a different FS.
900 * So, we need to add a special mount option to scan for
901 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
903 bytenr
= btrfs_sb_offset(0);
905 mutex_lock(&uuid_mutex
);
907 bdev
= blkdev_get_by_path(path
, flags
, holder
);
914 /* make sure our super fits in the device */
915 if (bytenr
+ PAGE_CACHE_SIZE
>= i_size_read(bdev
->bd_inode
))
918 /* make sure our super fits in the page */
919 if (sizeof(*disk_super
) > PAGE_CACHE_SIZE
)
922 /* make sure our super doesn't straddle pages on disk */
923 index
= bytenr
>> PAGE_CACHE_SHIFT
;
924 if ((bytenr
+ sizeof(*disk_super
) - 1) >> PAGE_CACHE_SHIFT
!= index
)
927 /* pull in the page with our super */
928 page
= read_cache_page_gfp(bdev
->bd_inode
->i_mapping
,
931 if (IS_ERR_OR_NULL(page
))
936 /* align our pointer to the offset of the super block */
937 disk_super
= p
+ (bytenr
& ~PAGE_CACHE_MASK
);
939 if (btrfs_super_bytenr(disk_super
) != bytenr
||
940 btrfs_super_magic(disk_super
) != BTRFS_MAGIC
)
943 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
944 transid
= btrfs_super_generation(disk_super
);
945 total_devices
= btrfs_super_num_devices(disk_super
);
947 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
949 if (disk_super
->label
[0]) {
950 if (disk_super
->label
[BTRFS_LABEL_SIZE
- 1])
951 disk_super
->label
[BTRFS_LABEL_SIZE
- 1] = '\0';
952 printk(KERN_INFO
"BTRFS: device label %s ", disk_super
->label
);
954 printk(KERN_INFO
"BTRFS: device fsid %pU ", disk_super
->fsid
);
957 printk(KERN_CONT
"devid %llu transid %llu %s\n", devid
, transid
, path
);
960 if (!ret
&& fs_devices_ret
)
961 (*fs_devices_ret
)->total_devices
= total_devices
;
965 page_cache_release(page
);
968 blkdev_put(bdev
, flags
);
970 mutex_unlock(&uuid_mutex
);
974 /* helper to account the used device space in the range */
975 int btrfs_account_dev_extents_size(struct btrfs_device
*device
, u64 start
,
976 u64 end
, u64
*length
)
978 struct btrfs_key key
;
979 struct btrfs_root
*root
= device
->dev_root
;
980 struct btrfs_dev_extent
*dev_extent
;
981 struct btrfs_path
*path
;
985 struct extent_buffer
*l
;
989 if (start
>= device
->total_bytes
|| device
->is_tgtdev_for_dev_replace
)
992 path
= btrfs_alloc_path();
997 key
.objectid
= device
->devid
;
999 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1001 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1005 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1012 slot
= path
->slots
[0];
1013 if (slot
>= btrfs_header_nritems(l
)) {
1014 ret
= btrfs_next_leaf(root
, path
);
1022 btrfs_item_key_to_cpu(l
, &key
, slot
);
1024 if (key
.objectid
< device
->devid
)
1027 if (key
.objectid
> device
->devid
)
1030 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1033 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1034 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1036 if (key
.offset
<= start
&& extent_end
> end
) {
1037 *length
= end
- start
+ 1;
1039 } else if (key
.offset
<= start
&& extent_end
> start
)
1040 *length
+= extent_end
- start
;
1041 else if (key
.offset
> start
&& extent_end
<= end
)
1042 *length
+= extent_end
- key
.offset
;
1043 else if (key
.offset
> start
&& key
.offset
<= end
) {
1044 *length
+= end
- key
.offset
+ 1;
1046 } else if (key
.offset
> end
)
1054 btrfs_free_path(path
);
1058 static int contains_pending_extent(struct btrfs_trans_handle
*trans
,
1059 struct btrfs_device
*device
,
1060 u64
*start
, u64 len
)
1062 struct extent_map
*em
;
1063 struct list_head
*search_list
= &trans
->transaction
->pending_chunks
;
1065 u64 physical_start
= *start
;
1068 list_for_each_entry(em
, search_list
, list
) {
1069 struct map_lookup
*map
;
1072 map
= (struct map_lookup
*)em
->bdev
;
1073 for (i
= 0; i
< map
->num_stripes
; i
++) {
1074 if (map
->stripes
[i
].dev
!= device
)
1076 if (map
->stripes
[i
].physical
>= physical_start
+ len
||
1077 map
->stripes
[i
].physical
+ em
->orig_block_len
<=
1080 *start
= map
->stripes
[i
].physical
+
1085 if (search_list
== &trans
->transaction
->pending_chunks
) {
1086 search_list
= &trans
->root
->fs_info
->pinned_chunks
;
1095 * find_free_dev_extent - find free space in the specified device
1096 * @device: the device which we search the free space in
1097 * @num_bytes: the size of the free space that we need
1098 * @start: store the start of the free space.
1099 * @len: the size of the free space. that we find, or the size of the max
1100 * free space if we don't find suitable free space
1102 * this uses a pretty simple search, the expectation is that it is
1103 * called very infrequently and that a given device has a small number
1106 * @start is used to store the start of the free space if we find. But if we
1107 * don't find suitable free space, it will be used to store the start position
1108 * of the max free space.
1110 * @len is used to store the size of the free space that we find.
1111 * But if we don't find suitable free space, it is used to store the size of
1112 * the max free space.
1114 int find_free_dev_extent(struct btrfs_trans_handle
*trans
,
1115 struct btrfs_device
*device
, u64 num_bytes
,
1116 u64
*start
, u64
*len
)
1118 struct btrfs_key key
;
1119 struct btrfs_root
*root
= device
->dev_root
;
1120 struct btrfs_dev_extent
*dev_extent
;
1121 struct btrfs_path
*path
;
1127 u64 search_end
= device
->total_bytes
;
1130 struct extent_buffer
*l
;
1132 /* FIXME use last free of some kind */
1134 /* we don't want to overwrite the superblock on the drive,
1135 * so we make sure to start at an offset of at least 1MB
1137 search_start
= max(root
->fs_info
->alloc_start
, 1024ull * 1024);
1139 path
= btrfs_alloc_path();
1143 max_hole_start
= search_start
;
1147 if (search_start
>= search_end
|| device
->is_tgtdev_for_dev_replace
) {
1153 path
->search_commit_root
= 1;
1154 path
->skip_locking
= 1;
1156 key
.objectid
= device
->devid
;
1157 key
.offset
= search_start
;
1158 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1160 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1164 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1171 slot
= path
->slots
[0];
1172 if (slot
>= btrfs_header_nritems(l
)) {
1173 ret
= btrfs_next_leaf(root
, path
);
1181 btrfs_item_key_to_cpu(l
, &key
, slot
);
1183 if (key
.objectid
< device
->devid
)
1186 if (key
.objectid
> device
->devid
)
1189 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1192 if (key
.offset
> search_start
) {
1193 hole_size
= key
.offset
- search_start
;
1196 * Have to check before we set max_hole_start, otherwise
1197 * we could end up sending back this offset anyway.
1199 if (contains_pending_extent(trans
, device
,
1202 if (key
.offset
>= search_start
) {
1203 hole_size
= key
.offset
- search_start
;
1210 if (hole_size
> max_hole_size
) {
1211 max_hole_start
= search_start
;
1212 max_hole_size
= hole_size
;
1216 * If this free space is greater than which we need,
1217 * it must be the max free space that we have found
1218 * until now, so max_hole_start must point to the start
1219 * of this free space and the length of this free space
1220 * is stored in max_hole_size. Thus, we return
1221 * max_hole_start and max_hole_size and go back to the
1224 if (hole_size
>= num_bytes
) {
1230 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1231 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1233 if (extent_end
> search_start
)
1234 search_start
= extent_end
;
1241 * At this point, search_start should be the end of
1242 * allocated dev extents, and when shrinking the device,
1243 * search_end may be smaller than search_start.
1245 if (search_end
> search_start
) {
1246 hole_size
= search_end
- search_start
;
1248 if (contains_pending_extent(trans
, device
, &search_start
,
1250 btrfs_release_path(path
);
1254 if (hole_size
> max_hole_size
) {
1255 max_hole_start
= search_start
;
1256 max_hole_size
= hole_size
;
1261 if (max_hole_size
< num_bytes
)
1267 btrfs_free_path(path
);
1268 *start
= max_hole_start
;
1270 *len
= max_hole_size
;
1274 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1275 struct btrfs_device
*device
,
1276 u64 start
, u64
*dev_extent_len
)
1279 struct btrfs_path
*path
;
1280 struct btrfs_root
*root
= device
->dev_root
;
1281 struct btrfs_key key
;
1282 struct btrfs_key found_key
;
1283 struct extent_buffer
*leaf
= NULL
;
1284 struct btrfs_dev_extent
*extent
= NULL
;
1286 path
= btrfs_alloc_path();
1290 key
.objectid
= device
->devid
;
1292 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1294 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1296 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1297 BTRFS_DEV_EXTENT_KEY
);
1300 leaf
= path
->nodes
[0];
1301 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1302 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1303 struct btrfs_dev_extent
);
1304 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1305 btrfs_dev_extent_length(leaf
, extent
) < start
);
1307 btrfs_release_path(path
);
1309 } else if (ret
== 0) {
1310 leaf
= path
->nodes
[0];
1311 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1312 struct btrfs_dev_extent
);
1314 btrfs_error(root
->fs_info
, ret
, "Slot search failed");
1318 *dev_extent_len
= btrfs_dev_extent_length(leaf
, extent
);
1320 ret
= btrfs_del_item(trans
, root
, path
);
1322 btrfs_error(root
->fs_info
, ret
,
1323 "Failed to remove dev extent item");
1325 trans
->transaction
->have_free_bgs
= 1;
1328 btrfs_free_path(path
);
1332 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1333 struct btrfs_device
*device
,
1334 u64 chunk_tree
, u64 chunk_objectid
,
1335 u64 chunk_offset
, u64 start
, u64 num_bytes
)
1338 struct btrfs_path
*path
;
1339 struct btrfs_root
*root
= device
->dev_root
;
1340 struct btrfs_dev_extent
*extent
;
1341 struct extent_buffer
*leaf
;
1342 struct btrfs_key key
;
1344 WARN_ON(!device
->in_fs_metadata
);
1345 WARN_ON(device
->is_tgtdev_for_dev_replace
);
1346 path
= btrfs_alloc_path();
1350 key
.objectid
= device
->devid
;
1352 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1353 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1358 leaf
= path
->nodes
[0];
1359 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1360 struct btrfs_dev_extent
);
1361 btrfs_set_dev_extent_chunk_tree(leaf
, extent
, chunk_tree
);
1362 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
, chunk_objectid
);
1363 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1365 write_extent_buffer(leaf
, root
->fs_info
->chunk_tree_uuid
,
1366 btrfs_dev_extent_chunk_tree_uuid(extent
), BTRFS_UUID_SIZE
);
1368 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1369 btrfs_mark_buffer_dirty(leaf
);
1371 btrfs_free_path(path
);
1375 static u64
find_next_chunk(struct btrfs_fs_info
*fs_info
)
1377 struct extent_map_tree
*em_tree
;
1378 struct extent_map
*em
;
1382 em_tree
= &fs_info
->mapping_tree
.map_tree
;
1383 read_lock(&em_tree
->lock
);
1384 n
= rb_last(&em_tree
->map
);
1386 em
= rb_entry(n
, struct extent_map
, rb_node
);
1387 ret
= em
->start
+ em
->len
;
1389 read_unlock(&em_tree
->lock
);
1394 static noinline
int find_next_devid(struct btrfs_fs_info
*fs_info
,
1398 struct btrfs_key key
;
1399 struct btrfs_key found_key
;
1400 struct btrfs_path
*path
;
1402 path
= btrfs_alloc_path();
1406 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1407 key
.type
= BTRFS_DEV_ITEM_KEY
;
1408 key
.offset
= (u64
)-1;
1410 ret
= btrfs_search_slot(NULL
, fs_info
->chunk_root
, &key
, path
, 0, 0);
1414 BUG_ON(ret
== 0); /* Corruption */
1416 ret
= btrfs_previous_item(fs_info
->chunk_root
, path
,
1417 BTRFS_DEV_ITEMS_OBJECTID
,
1418 BTRFS_DEV_ITEM_KEY
);
1422 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1424 *devid_ret
= found_key
.offset
+ 1;
1428 btrfs_free_path(path
);
1433 * the device information is stored in the chunk root
1434 * the btrfs_device struct should be fully filled in
1436 static int btrfs_add_device(struct btrfs_trans_handle
*trans
,
1437 struct btrfs_root
*root
,
1438 struct btrfs_device
*device
)
1441 struct btrfs_path
*path
;
1442 struct btrfs_dev_item
*dev_item
;
1443 struct extent_buffer
*leaf
;
1444 struct btrfs_key key
;
1447 root
= root
->fs_info
->chunk_root
;
1449 path
= btrfs_alloc_path();
1453 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1454 key
.type
= BTRFS_DEV_ITEM_KEY
;
1455 key
.offset
= device
->devid
;
1457 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1462 leaf
= path
->nodes
[0];
1463 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1465 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1466 btrfs_set_device_generation(leaf
, dev_item
, 0);
1467 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1468 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1469 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1470 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1471 btrfs_set_device_total_bytes(leaf
, dev_item
,
1472 btrfs_device_get_disk_total_bytes(device
));
1473 btrfs_set_device_bytes_used(leaf
, dev_item
,
1474 btrfs_device_get_bytes_used(device
));
1475 btrfs_set_device_group(leaf
, dev_item
, 0);
1476 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1477 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1478 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1480 ptr
= btrfs_device_uuid(dev_item
);
1481 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1482 ptr
= btrfs_device_fsid(dev_item
);
1483 write_extent_buffer(leaf
, root
->fs_info
->fsid
, ptr
, BTRFS_UUID_SIZE
);
1484 btrfs_mark_buffer_dirty(leaf
);
1488 btrfs_free_path(path
);
1493 * Function to update ctime/mtime for a given device path.
1494 * Mainly used for ctime/mtime based probe like libblkid.
1496 static void update_dev_time(char *path_name
)
1500 filp
= filp_open(path_name
, O_RDWR
, 0);
1503 file_update_time(filp
);
1504 filp_close(filp
, NULL
);
1508 static int btrfs_rm_dev_item(struct btrfs_root
*root
,
1509 struct btrfs_device
*device
)
1512 struct btrfs_path
*path
;
1513 struct btrfs_key key
;
1514 struct btrfs_trans_handle
*trans
;
1516 root
= root
->fs_info
->chunk_root
;
1518 path
= btrfs_alloc_path();
1522 trans
= btrfs_start_transaction(root
, 0);
1523 if (IS_ERR(trans
)) {
1524 btrfs_free_path(path
);
1525 return PTR_ERR(trans
);
1527 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1528 key
.type
= BTRFS_DEV_ITEM_KEY
;
1529 key
.offset
= device
->devid
;
1531 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1540 ret
= btrfs_del_item(trans
, root
, path
);
1544 btrfs_free_path(path
);
1545 btrfs_commit_transaction(trans
, root
);
1549 int btrfs_rm_device(struct btrfs_root
*root
, char *device_path
)
1551 struct btrfs_device
*device
;
1552 struct btrfs_device
*next_device
;
1553 struct block_device
*bdev
;
1554 struct buffer_head
*bh
= NULL
;
1555 struct btrfs_super_block
*disk_super
;
1556 struct btrfs_fs_devices
*cur_devices
;
1563 bool clear_super
= false;
1565 mutex_lock(&uuid_mutex
);
1568 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
1570 all_avail
= root
->fs_info
->avail_data_alloc_bits
|
1571 root
->fs_info
->avail_system_alloc_bits
|
1572 root
->fs_info
->avail_metadata_alloc_bits
;
1573 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
1575 num_devices
= root
->fs_info
->fs_devices
->num_devices
;
1576 btrfs_dev_replace_lock(&root
->fs_info
->dev_replace
);
1577 if (btrfs_dev_replace_is_ongoing(&root
->fs_info
->dev_replace
)) {
1578 WARN_ON(num_devices
< 1);
1581 btrfs_dev_replace_unlock(&root
->fs_info
->dev_replace
);
1583 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID10
) && num_devices
<= 4) {
1584 ret
= BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET
;
1588 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID1
) && num_devices
<= 2) {
1589 ret
= BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET
;
1593 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID5
) &&
1594 root
->fs_info
->fs_devices
->rw_devices
<= 2) {
1595 ret
= BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET
;
1598 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID6
) &&
1599 root
->fs_info
->fs_devices
->rw_devices
<= 3) {
1600 ret
= BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET
;
1604 if (strcmp(device_path
, "missing") == 0) {
1605 struct list_head
*devices
;
1606 struct btrfs_device
*tmp
;
1609 devices
= &root
->fs_info
->fs_devices
->devices
;
1611 * It is safe to read the devices since the volume_mutex
1614 list_for_each_entry(tmp
, devices
, dev_list
) {
1615 if (tmp
->in_fs_metadata
&&
1616 !tmp
->is_tgtdev_for_dev_replace
&&
1626 ret
= BTRFS_ERROR_DEV_MISSING_NOT_FOUND
;
1630 ret
= btrfs_get_bdev_and_sb(device_path
,
1631 FMODE_WRITE
| FMODE_EXCL
,
1632 root
->fs_info
->bdev_holder
, 0,
1636 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1637 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1638 dev_uuid
= disk_super
->dev_item
.uuid
;
1639 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1647 if (device
->is_tgtdev_for_dev_replace
) {
1648 ret
= BTRFS_ERROR_DEV_TGT_REPLACE
;
1652 if (device
->writeable
&& root
->fs_info
->fs_devices
->rw_devices
== 1) {
1653 ret
= BTRFS_ERROR_DEV_ONLY_WRITABLE
;
1657 if (device
->writeable
) {
1659 list_del_init(&device
->dev_alloc_list
);
1660 device
->fs_devices
->rw_devices
--;
1661 unlock_chunks(root
);
1665 mutex_unlock(&uuid_mutex
);
1666 ret
= btrfs_shrink_device(device
, 0);
1667 mutex_lock(&uuid_mutex
);
1672 * TODO: the superblock still includes this device in its num_devices
1673 * counter although write_all_supers() is not locked out. This
1674 * could give a filesystem state which requires a degraded mount.
1676 ret
= btrfs_rm_dev_item(root
->fs_info
->chunk_root
, device
);
1680 device
->in_fs_metadata
= 0;
1681 btrfs_scrub_cancel_dev(root
->fs_info
, device
);
1684 * the device list mutex makes sure that we don't change
1685 * the device list while someone else is writing out all
1686 * the device supers. Whoever is writing all supers, should
1687 * lock the device list mutex before getting the number of
1688 * devices in the super block (super_copy). Conversely,
1689 * whoever updates the number of devices in the super block
1690 * (super_copy) should hold the device list mutex.
1693 cur_devices
= device
->fs_devices
;
1694 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1695 list_del_rcu(&device
->dev_list
);
1697 device
->fs_devices
->num_devices
--;
1698 device
->fs_devices
->total_devices
--;
1700 if (device
->missing
)
1701 device
->fs_devices
->missing_devices
--;
1703 next_device
= list_entry(root
->fs_info
->fs_devices
->devices
.next
,
1704 struct btrfs_device
, dev_list
);
1705 if (device
->bdev
== root
->fs_info
->sb
->s_bdev
)
1706 root
->fs_info
->sb
->s_bdev
= next_device
->bdev
;
1707 if (device
->bdev
== root
->fs_info
->fs_devices
->latest_bdev
)
1708 root
->fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1711 device
->fs_devices
->open_devices
--;
1712 /* remove sysfs entry */
1713 btrfs_kobj_rm_device(root
->fs_info
, device
);
1716 call_rcu(&device
->rcu
, free_device
);
1718 num_devices
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
1719 btrfs_set_super_num_devices(root
->fs_info
->super_copy
, num_devices
);
1720 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1722 if (cur_devices
->open_devices
== 0) {
1723 struct btrfs_fs_devices
*fs_devices
;
1724 fs_devices
= root
->fs_info
->fs_devices
;
1725 while (fs_devices
) {
1726 if (fs_devices
->seed
== cur_devices
) {
1727 fs_devices
->seed
= cur_devices
->seed
;
1730 fs_devices
= fs_devices
->seed
;
1732 cur_devices
->seed
= NULL
;
1733 __btrfs_close_devices(cur_devices
);
1734 free_fs_devices(cur_devices
);
1737 root
->fs_info
->num_tolerated_disk_barrier_failures
=
1738 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
1741 * at this point, the device is zero sized. We want to
1742 * remove it from the devices list and zero out the old super
1744 if (clear_super
&& disk_super
) {
1748 /* make sure this device isn't detected as part of
1751 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
1752 set_buffer_dirty(bh
);
1753 sync_dirty_buffer(bh
);
1755 /* clear the mirror copies of super block on the disk
1756 * being removed, 0th copy is been taken care above and
1757 * the below would take of the rest
1759 for (i
= 1; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
1760 bytenr
= btrfs_sb_offset(i
);
1761 if (bytenr
+ BTRFS_SUPER_INFO_SIZE
>=
1762 i_size_read(bdev
->bd_inode
))
1766 bh
= __bread(bdev
, bytenr
/ 4096,
1767 BTRFS_SUPER_INFO_SIZE
);
1771 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1773 if (btrfs_super_bytenr(disk_super
) != bytenr
||
1774 btrfs_super_magic(disk_super
) != BTRFS_MAGIC
) {
1777 memset(&disk_super
->magic
, 0,
1778 sizeof(disk_super
->magic
));
1779 set_buffer_dirty(bh
);
1780 sync_dirty_buffer(bh
);
1787 /* Notify udev that device has changed */
1788 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
1790 /* Update ctime/mtime for device path for libblkid */
1791 update_dev_time(device_path
);
1797 blkdev_put(bdev
, FMODE_READ
| FMODE_EXCL
);
1799 mutex_unlock(&uuid_mutex
);
1802 if (device
->writeable
) {
1804 list_add(&device
->dev_alloc_list
,
1805 &root
->fs_info
->fs_devices
->alloc_list
);
1806 device
->fs_devices
->rw_devices
++;
1807 unlock_chunks(root
);
1812 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info
*fs_info
,
1813 struct btrfs_device
*srcdev
)
1815 struct btrfs_fs_devices
*fs_devices
;
1817 WARN_ON(!mutex_is_locked(&fs_info
->fs_devices
->device_list_mutex
));
1820 * in case of fs with no seed, srcdev->fs_devices will point
1821 * to fs_devices of fs_info. However when the dev being replaced is
1822 * a seed dev it will point to the seed's local fs_devices. In short
1823 * srcdev will have its correct fs_devices in both the cases.
1825 fs_devices
= srcdev
->fs_devices
;
1827 list_del_rcu(&srcdev
->dev_list
);
1828 list_del_rcu(&srcdev
->dev_alloc_list
);
1829 fs_devices
->num_devices
--;
1830 if (srcdev
->missing
)
1831 fs_devices
->missing_devices
--;
1833 if (srcdev
->writeable
) {
1834 fs_devices
->rw_devices
--;
1835 /* zero out the old super if it is writable */
1836 btrfs_scratch_superblock(srcdev
);
1840 fs_devices
->open_devices
--;
1843 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info
*fs_info
,
1844 struct btrfs_device
*srcdev
)
1846 struct btrfs_fs_devices
*fs_devices
= srcdev
->fs_devices
;
1848 call_rcu(&srcdev
->rcu
, free_device
);
1851 * unless fs_devices is seed fs, num_devices shouldn't go
1854 BUG_ON(!fs_devices
->num_devices
&& !fs_devices
->seeding
);
1856 /* if this is no devs we rather delete the fs_devices */
1857 if (!fs_devices
->num_devices
) {
1858 struct btrfs_fs_devices
*tmp_fs_devices
;
1860 tmp_fs_devices
= fs_info
->fs_devices
;
1861 while (tmp_fs_devices
) {
1862 if (tmp_fs_devices
->seed
== fs_devices
) {
1863 tmp_fs_devices
->seed
= fs_devices
->seed
;
1866 tmp_fs_devices
= tmp_fs_devices
->seed
;
1868 fs_devices
->seed
= NULL
;
1869 __btrfs_close_devices(fs_devices
);
1870 free_fs_devices(fs_devices
);
1874 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
1875 struct btrfs_device
*tgtdev
)
1877 struct btrfs_device
*next_device
;
1879 mutex_lock(&uuid_mutex
);
1881 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
1883 btrfs_scratch_superblock(tgtdev
);
1884 fs_info
->fs_devices
->open_devices
--;
1886 fs_info
->fs_devices
->num_devices
--;
1888 next_device
= list_entry(fs_info
->fs_devices
->devices
.next
,
1889 struct btrfs_device
, dev_list
);
1890 if (tgtdev
->bdev
== fs_info
->sb
->s_bdev
)
1891 fs_info
->sb
->s_bdev
= next_device
->bdev
;
1892 if (tgtdev
->bdev
== fs_info
->fs_devices
->latest_bdev
)
1893 fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1894 list_del_rcu(&tgtdev
->dev_list
);
1896 call_rcu(&tgtdev
->rcu
, free_device
);
1898 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
1899 mutex_unlock(&uuid_mutex
);
1902 static int btrfs_find_device_by_path(struct btrfs_root
*root
, char *device_path
,
1903 struct btrfs_device
**device
)
1906 struct btrfs_super_block
*disk_super
;
1909 struct block_device
*bdev
;
1910 struct buffer_head
*bh
;
1913 ret
= btrfs_get_bdev_and_sb(device_path
, FMODE_READ
,
1914 root
->fs_info
->bdev_holder
, 0, &bdev
, &bh
);
1917 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1918 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1919 dev_uuid
= disk_super
->dev_item
.uuid
;
1920 *device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1925 blkdev_put(bdev
, FMODE_READ
);
1929 int btrfs_find_device_missing_or_by_path(struct btrfs_root
*root
,
1931 struct btrfs_device
**device
)
1934 if (strcmp(device_path
, "missing") == 0) {
1935 struct list_head
*devices
;
1936 struct btrfs_device
*tmp
;
1938 devices
= &root
->fs_info
->fs_devices
->devices
;
1940 * It is safe to read the devices since the volume_mutex
1941 * is held by the caller.
1943 list_for_each_entry(tmp
, devices
, dev_list
) {
1944 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
1951 btrfs_err(root
->fs_info
, "no missing device found");
1957 return btrfs_find_device_by_path(root
, device_path
, device
);
1962 * does all the dirty work required for changing file system's UUID.
1964 static int btrfs_prepare_sprout(struct btrfs_root
*root
)
1966 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
1967 struct btrfs_fs_devices
*old_devices
;
1968 struct btrfs_fs_devices
*seed_devices
;
1969 struct btrfs_super_block
*disk_super
= root
->fs_info
->super_copy
;
1970 struct btrfs_device
*device
;
1973 BUG_ON(!mutex_is_locked(&uuid_mutex
));
1974 if (!fs_devices
->seeding
)
1977 seed_devices
= __alloc_fs_devices();
1978 if (IS_ERR(seed_devices
))
1979 return PTR_ERR(seed_devices
);
1981 old_devices
= clone_fs_devices(fs_devices
);
1982 if (IS_ERR(old_devices
)) {
1983 kfree(seed_devices
);
1984 return PTR_ERR(old_devices
);
1987 list_add(&old_devices
->list
, &fs_uuids
);
1989 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
1990 seed_devices
->opened
= 1;
1991 INIT_LIST_HEAD(&seed_devices
->devices
);
1992 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
1993 mutex_init(&seed_devices
->device_list_mutex
);
1995 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1996 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
1998 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
)
1999 device
->fs_devices
= seed_devices
;
2002 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
2003 unlock_chunks(root
);
2005 fs_devices
->seeding
= 0;
2006 fs_devices
->num_devices
= 0;
2007 fs_devices
->open_devices
= 0;
2008 fs_devices
->missing_devices
= 0;
2009 fs_devices
->rotating
= 0;
2010 fs_devices
->seed
= seed_devices
;
2012 generate_random_uuid(fs_devices
->fsid
);
2013 memcpy(root
->fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2014 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2015 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2017 super_flags
= btrfs_super_flags(disk_super
) &
2018 ~BTRFS_SUPER_FLAG_SEEDING
;
2019 btrfs_set_super_flags(disk_super
, super_flags
);
2025 * strore the expected generation for seed devices in device items.
2027 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
2028 struct btrfs_root
*root
)
2030 struct btrfs_path
*path
;
2031 struct extent_buffer
*leaf
;
2032 struct btrfs_dev_item
*dev_item
;
2033 struct btrfs_device
*device
;
2034 struct btrfs_key key
;
2035 u8 fs_uuid
[BTRFS_UUID_SIZE
];
2036 u8 dev_uuid
[BTRFS_UUID_SIZE
];
2040 path
= btrfs_alloc_path();
2044 root
= root
->fs_info
->chunk_root
;
2045 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2047 key
.type
= BTRFS_DEV_ITEM_KEY
;
2050 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2054 leaf
= path
->nodes
[0];
2056 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
2057 ret
= btrfs_next_leaf(root
, path
);
2062 leaf
= path
->nodes
[0];
2063 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2064 btrfs_release_path(path
);
2068 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2069 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
2070 key
.type
!= BTRFS_DEV_ITEM_KEY
)
2073 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2074 struct btrfs_dev_item
);
2075 devid
= btrfs_device_id(leaf
, dev_item
);
2076 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
2078 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
2080 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
2082 BUG_ON(!device
); /* Logic error */
2084 if (device
->fs_devices
->seeding
) {
2085 btrfs_set_device_generation(leaf
, dev_item
,
2086 device
->generation
);
2087 btrfs_mark_buffer_dirty(leaf
);
2095 btrfs_free_path(path
);
2099 int btrfs_init_new_device(struct btrfs_root
*root
, char *device_path
)
2101 struct request_queue
*q
;
2102 struct btrfs_trans_handle
*trans
;
2103 struct btrfs_device
*device
;
2104 struct block_device
*bdev
;
2105 struct list_head
*devices
;
2106 struct super_block
*sb
= root
->fs_info
->sb
;
2107 struct rcu_string
*name
;
2109 int seeding_dev
= 0;
2112 if ((sb
->s_flags
& MS_RDONLY
) && !root
->fs_info
->fs_devices
->seeding
)
2115 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2116 root
->fs_info
->bdev_holder
);
2118 return PTR_ERR(bdev
);
2120 if (root
->fs_info
->fs_devices
->seeding
) {
2122 down_write(&sb
->s_umount
);
2123 mutex_lock(&uuid_mutex
);
2126 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2128 devices
= &root
->fs_info
->fs_devices
->devices
;
2130 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2131 list_for_each_entry(device
, devices
, dev_list
) {
2132 if (device
->bdev
== bdev
) {
2135 &root
->fs_info
->fs_devices
->device_list_mutex
);
2139 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2141 device
= btrfs_alloc_device(root
->fs_info
, NULL
, NULL
);
2142 if (IS_ERR(device
)) {
2143 /* we can safely leave the fs_devices entry around */
2144 ret
= PTR_ERR(device
);
2148 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2154 rcu_assign_pointer(device
->name
, name
);
2156 trans
= btrfs_start_transaction(root
, 0);
2157 if (IS_ERR(trans
)) {
2158 rcu_string_free(device
->name
);
2160 ret
= PTR_ERR(trans
);
2164 q
= bdev_get_queue(bdev
);
2165 if (blk_queue_discard(q
))
2166 device
->can_discard
= 1;
2167 device
->writeable
= 1;
2168 device
->generation
= trans
->transid
;
2169 device
->io_width
= root
->sectorsize
;
2170 device
->io_align
= root
->sectorsize
;
2171 device
->sector_size
= root
->sectorsize
;
2172 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
2173 device
->disk_total_bytes
= device
->total_bytes
;
2174 device
->commit_total_bytes
= device
->total_bytes
;
2175 device
->dev_root
= root
->fs_info
->dev_root
;
2176 device
->bdev
= bdev
;
2177 device
->in_fs_metadata
= 1;
2178 device
->is_tgtdev_for_dev_replace
= 0;
2179 device
->mode
= FMODE_EXCL
;
2180 device
->dev_stats_valid
= 1;
2181 set_blocksize(device
->bdev
, 4096);
2184 sb
->s_flags
&= ~MS_RDONLY
;
2185 ret
= btrfs_prepare_sprout(root
);
2186 BUG_ON(ret
); /* -ENOMEM */
2189 device
->fs_devices
= root
->fs_info
->fs_devices
;
2191 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2193 list_add_rcu(&device
->dev_list
, &root
->fs_info
->fs_devices
->devices
);
2194 list_add(&device
->dev_alloc_list
,
2195 &root
->fs_info
->fs_devices
->alloc_list
);
2196 root
->fs_info
->fs_devices
->num_devices
++;
2197 root
->fs_info
->fs_devices
->open_devices
++;
2198 root
->fs_info
->fs_devices
->rw_devices
++;
2199 root
->fs_info
->fs_devices
->total_devices
++;
2200 root
->fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
2202 spin_lock(&root
->fs_info
->free_chunk_lock
);
2203 root
->fs_info
->free_chunk_space
+= device
->total_bytes
;
2204 spin_unlock(&root
->fs_info
->free_chunk_lock
);
2206 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
2207 root
->fs_info
->fs_devices
->rotating
= 1;
2209 tmp
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
2210 btrfs_set_super_total_bytes(root
->fs_info
->super_copy
,
2211 tmp
+ device
->total_bytes
);
2213 tmp
= btrfs_super_num_devices(root
->fs_info
->super_copy
);
2214 btrfs_set_super_num_devices(root
->fs_info
->super_copy
,
2217 /* add sysfs device entry */
2218 btrfs_kobj_add_device(root
->fs_info
, device
);
2221 * we've got more storage, clear any full flags on the space
2224 btrfs_clear_space_info_full(root
->fs_info
);
2226 unlock_chunks(root
);
2227 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2231 ret
= init_first_rw_device(trans
, root
, device
);
2232 unlock_chunks(root
);
2234 btrfs_abort_transaction(trans
, root
, ret
);
2239 ret
= btrfs_add_device(trans
, root
, device
);
2241 btrfs_abort_transaction(trans
, root
, ret
);
2246 char fsid_buf
[BTRFS_UUID_UNPARSED_SIZE
];
2248 ret
= btrfs_finish_sprout(trans
, root
);
2250 btrfs_abort_transaction(trans
, root
, ret
);
2254 /* Sprouting would change fsid of the mounted root,
2255 * so rename the fsid on the sysfs
2257 snprintf(fsid_buf
, BTRFS_UUID_UNPARSED_SIZE
, "%pU",
2258 root
->fs_info
->fsid
);
2259 if (kobject_rename(&root
->fs_info
->fs_devices
->super_kobj
,
2264 root
->fs_info
->num_tolerated_disk_barrier_failures
=
2265 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
2266 ret
= btrfs_commit_transaction(trans
, root
);
2269 mutex_unlock(&uuid_mutex
);
2270 up_write(&sb
->s_umount
);
2272 if (ret
) /* transaction commit */
2275 ret
= btrfs_relocate_sys_chunks(root
);
2277 btrfs_error(root
->fs_info
, ret
,
2278 "Failed to relocate sys chunks after "
2279 "device initialization. This can be fixed "
2280 "using the \"btrfs balance\" command.");
2281 trans
= btrfs_attach_transaction(root
);
2282 if (IS_ERR(trans
)) {
2283 if (PTR_ERR(trans
) == -ENOENT
)
2285 return PTR_ERR(trans
);
2287 ret
= btrfs_commit_transaction(trans
, root
);
2290 /* Update ctime/mtime for libblkid */
2291 update_dev_time(device_path
);
2295 btrfs_end_transaction(trans
, root
);
2296 rcu_string_free(device
->name
);
2297 btrfs_kobj_rm_device(root
->fs_info
, device
);
2300 blkdev_put(bdev
, FMODE_EXCL
);
2302 mutex_unlock(&uuid_mutex
);
2303 up_write(&sb
->s_umount
);
2308 int btrfs_init_dev_replace_tgtdev(struct btrfs_root
*root
, char *device_path
,
2309 struct btrfs_device
*srcdev
,
2310 struct btrfs_device
**device_out
)
2312 struct request_queue
*q
;
2313 struct btrfs_device
*device
;
2314 struct block_device
*bdev
;
2315 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2316 struct list_head
*devices
;
2317 struct rcu_string
*name
;
2318 u64 devid
= BTRFS_DEV_REPLACE_DEVID
;
2322 if (fs_info
->fs_devices
->seeding
) {
2323 btrfs_err(fs_info
, "the filesystem is a seed filesystem!");
2327 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2328 fs_info
->bdev_holder
);
2330 btrfs_err(fs_info
, "target device %s is invalid!", device_path
);
2331 return PTR_ERR(bdev
);
2334 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2336 devices
= &fs_info
->fs_devices
->devices
;
2337 list_for_each_entry(device
, devices
, dev_list
) {
2338 if (device
->bdev
== bdev
) {
2339 btrfs_err(fs_info
, "target device is in the filesystem!");
2346 if (i_size_read(bdev
->bd_inode
) <
2347 btrfs_device_get_total_bytes(srcdev
)) {
2348 btrfs_err(fs_info
, "target device is smaller than source device!");
2354 device
= btrfs_alloc_device(NULL
, &devid
, NULL
);
2355 if (IS_ERR(device
)) {
2356 ret
= PTR_ERR(device
);
2360 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2366 rcu_assign_pointer(device
->name
, name
);
2368 q
= bdev_get_queue(bdev
);
2369 if (blk_queue_discard(q
))
2370 device
->can_discard
= 1;
2371 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2372 device
->writeable
= 1;
2373 device
->generation
= 0;
2374 device
->io_width
= root
->sectorsize
;
2375 device
->io_align
= root
->sectorsize
;
2376 device
->sector_size
= root
->sectorsize
;
2377 device
->total_bytes
= btrfs_device_get_total_bytes(srcdev
);
2378 device
->disk_total_bytes
= btrfs_device_get_disk_total_bytes(srcdev
);
2379 device
->bytes_used
= btrfs_device_get_bytes_used(srcdev
);
2380 ASSERT(list_empty(&srcdev
->resized_list
));
2381 device
->commit_total_bytes
= srcdev
->commit_total_bytes
;
2382 device
->commit_bytes_used
= device
->bytes_used
;
2383 device
->dev_root
= fs_info
->dev_root
;
2384 device
->bdev
= bdev
;
2385 device
->in_fs_metadata
= 1;
2386 device
->is_tgtdev_for_dev_replace
= 1;
2387 device
->mode
= FMODE_EXCL
;
2388 device
->dev_stats_valid
= 1;
2389 set_blocksize(device
->bdev
, 4096);
2390 device
->fs_devices
= fs_info
->fs_devices
;
2391 list_add(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2392 fs_info
->fs_devices
->num_devices
++;
2393 fs_info
->fs_devices
->open_devices
++;
2394 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2396 *device_out
= device
;
2400 blkdev_put(bdev
, FMODE_EXCL
);
2404 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info
*fs_info
,
2405 struct btrfs_device
*tgtdev
)
2407 WARN_ON(fs_info
->fs_devices
->rw_devices
== 0);
2408 tgtdev
->io_width
= fs_info
->dev_root
->sectorsize
;
2409 tgtdev
->io_align
= fs_info
->dev_root
->sectorsize
;
2410 tgtdev
->sector_size
= fs_info
->dev_root
->sectorsize
;
2411 tgtdev
->dev_root
= fs_info
->dev_root
;
2412 tgtdev
->in_fs_metadata
= 1;
2415 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2416 struct btrfs_device
*device
)
2419 struct btrfs_path
*path
;
2420 struct btrfs_root
*root
;
2421 struct btrfs_dev_item
*dev_item
;
2422 struct extent_buffer
*leaf
;
2423 struct btrfs_key key
;
2425 root
= device
->dev_root
->fs_info
->chunk_root
;
2427 path
= btrfs_alloc_path();
2431 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2432 key
.type
= BTRFS_DEV_ITEM_KEY
;
2433 key
.offset
= device
->devid
;
2435 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2444 leaf
= path
->nodes
[0];
2445 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2447 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2448 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2449 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2450 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2451 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2452 btrfs_set_device_total_bytes(leaf
, dev_item
,
2453 btrfs_device_get_disk_total_bytes(device
));
2454 btrfs_set_device_bytes_used(leaf
, dev_item
,
2455 btrfs_device_get_bytes_used(device
));
2456 btrfs_mark_buffer_dirty(leaf
);
2459 btrfs_free_path(path
);
2463 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2464 struct btrfs_device
*device
, u64 new_size
)
2466 struct btrfs_super_block
*super_copy
=
2467 device
->dev_root
->fs_info
->super_copy
;
2468 struct btrfs_fs_devices
*fs_devices
;
2472 if (!device
->writeable
)
2475 lock_chunks(device
->dev_root
);
2476 old_total
= btrfs_super_total_bytes(super_copy
);
2477 diff
= new_size
- device
->total_bytes
;
2479 if (new_size
<= device
->total_bytes
||
2480 device
->is_tgtdev_for_dev_replace
) {
2481 unlock_chunks(device
->dev_root
);
2485 fs_devices
= device
->dev_root
->fs_info
->fs_devices
;
2487 btrfs_set_super_total_bytes(super_copy
, old_total
+ diff
);
2488 device
->fs_devices
->total_rw_bytes
+= diff
;
2490 btrfs_device_set_total_bytes(device
, new_size
);
2491 btrfs_device_set_disk_total_bytes(device
, new_size
);
2492 btrfs_clear_space_info_full(device
->dev_root
->fs_info
);
2493 if (list_empty(&device
->resized_list
))
2494 list_add_tail(&device
->resized_list
,
2495 &fs_devices
->resized_devices
);
2496 unlock_chunks(device
->dev_root
);
2498 return btrfs_update_device(trans
, device
);
2501 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
2502 struct btrfs_root
*root
, u64 chunk_objectid
,
2506 struct btrfs_path
*path
;
2507 struct btrfs_key key
;
2509 root
= root
->fs_info
->chunk_root
;
2510 path
= btrfs_alloc_path();
2514 key
.objectid
= chunk_objectid
;
2515 key
.offset
= chunk_offset
;
2516 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2518 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2521 else if (ret
> 0) { /* Logic error or corruption */
2522 btrfs_error(root
->fs_info
, -ENOENT
,
2523 "Failed lookup while freeing chunk.");
2528 ret
= btrfs_del_item(trans
, root
, path
);
2530 btrfs_error(root
->fs_info
, ret
,
2531 "Failed to delete chunk item.");
2533 btrfs_free_path(path
);
2537 static int btrfs_del_sys_chunk(struct btrfs_root
*root
, u64 chunk_objectid
, u64
2540 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
2541 struct btrfs_disk_key
*disk_key
;
2542 struct btrfs_chunk
*chunk
;
2549 struct btrfs_key key
;
2552 array_size
= btrfs_super_sys_array_size(super_copy
);
2554 ptr
= super_copy
->sys_chunk_array
;
2557 while (cur
< array_size
) {
2558 disk_key
= (struct btrfs_disk_key
*)ptr
;
2559 btrfs_disk_key_to_cpu(&key
, disk_key
);
2561 len
= sizeof(*disk_key
);
2563 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2564 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2565 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2566 len
+= btrfs_chunk_item_size(num_stripes
);
2571 if (key
.objectid
== chunk_objectid
&&
2572 key
.offset
== chunk_offset
) {
2573 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2575 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2581 unlock_chunks(root
);
2585 int btrfs_remove_chunk(struct btrfs_trans_handle
*trans
,
2586 struct btrfs_root
*root
, u64 chunk_offset
)
2588 struct extent_map_tree
*em_tree
;
2589 struct extent_map
*em
;
2590 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2591 struct map_lookup
*map
;
2592 u64 dev_extent_len
= 0;
2593 u64 chunk_objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2597 root
= root
->fs_info
->chunk_root
;
2598 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
2600 read_lock(&em_tree
->lock
);
2601 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
2602 read_unlock(&em_tree
->lock
);
2604 if (!em
|| em
->start
> chunk_offset
||
2605 em
->start
+ em
->len
< chunk_offset
) {
2607 * This is a logic error, but we don't want to just rely on the
2608 * user having built with ASSERT enabled, so if ASSERT doens't
2609 * do anything we still error out.
2613 free_extent_map(em
);
2616 map
= (struct map_lookup
*)em
->bdev
;
2618 for (i
= 0; i
< map
->num_stripes
; i
++) {
2619 struct btrfs_device
*device
= map
->stripes
[i
].dev
;
2620 ret
= btrfs_free_dev_extent(trans
, device
,
2621 map
->stripes
[i
].physical
,
2624 btrfs_abort_transaction(trans
, root
, ret
);
2628 if (device
->bytes_used
> 0) {
2630 btrfs_device_set_bytes_used(device
,
2631 device
->bytes_used
- dev_extent_len
);
2632 spin_lock(&root
->fs_info
->free_chunk_lock
);
2633 root
->fs_info
->free_chunk_space
+= dev_extent_len
;
2634 spin_unlock(&root
->fs_info
->free_chunk_lock
);
2635 btrfs_clear_space_info_full(root
->fs_info
);
2636 unlock_chunks(root
);
2639 if (map
->stripes
[i
].dev
) {
2640 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2642 btrfs_abort_transaction(trans
, root
, ret
);
2647 ret
= btrfs_free_chunk(trans
, root
, chunk_objectid
, chunk_offset
);
2649 btrfs_abort_transaction(trans
, root
, ret
);
2653 trace_btrfs_chunk_free(root
, map
, chunk_offset
, em
->len
);
2655 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2656 ret
= btrfs_del_sys_chunk(root
, chunk_objectid
, chunk_offset
);
2658 btrfs_abort_transaction(trans
, root
, ret
);
2663 ret
= btrfs_remove_block_group(trans
, extent_root
, chunk_offset
, em
);
2665 btrfs_abort_transaction(trans
, extent_root
, ret
);
2671 free_extent_map(em
);
2675 static int btrfs_relocate_chunk(struct btrfs_root
*root
,
2679 struct btrfs_root
*extent_root
;
2680 struct btrfs_trans_handle
*trans
;
2683 root
= root
->fs_info
->chunk_root
;
2684 extent_root
= root
->fs_info
->extent_root
;
2686 ret
= btrfs_can_relocate(extent_root
, chunk_offset
);
2690 /* step one, relocate all the extents inside this chunk */
2691 ret
= btrfs_relocate_block_group(extent_root
, chunk_offset
);
2695 trans
= btrfs_start_transaction(root
, 0);
2696 if (IS_ERR(trans
)) {
2697 ret
= PTR_ERR(trans
);
2698 btrfs_std_error(root
->fs_info
, ret
);
2703 * step two, delete the device extents and the
2704 * chunk tree entries
2706 ret
= btrfs_remove_chunk(trans
, root
, chunk_offset
);
2707 btrfs_end_transaction(trans
, root
);
2711 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
)
2713 struct btrfs_root
*chunk_root
= root
->fs_info
->chunk_root
;
2714 struct btrfs_path
*path
;
2715 struct extent_buffer
*leaf
;
2716 struct btrfs_chunk
*chunk
;
2717 struct btrfs_key key
;
2718 struct btrfs_key found_key
;
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
,
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 stripe_length
= div_u64(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
,
3268 if (ret
&& ret
!= -ENOSPC
)
3270 if (ret
== -ENOSPC
) {
3273 spin_lock(&fs_info
->balance_lock
);
3274 bctl
->stat
.completed
++;
3275 spin_unlock(&fs_info
->balance_lock
);
3278 if (found_key
.offset
== 0)
3280 key
.offset
= found_key
.offset
- 1;
3284 btrfs_release_path(path
);
3289 btrfs_free_path(path
);
3290 if (enospc_errors
) {
3291 btrfs_info(fs_info
, "%d enospc errors during balance",
3301 * alloc_profile_is_valid - see if a given profile is valid and reduced
3302 * @flags: profile to validate
3303 * @extended: if true @flags is treated as an extended profile
3305 static int alloc_profile_is_valid(u64 flags
, int extended
)
3307 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
3308 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
3310 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
3312 /* 1) check that all other bits are zeroed */
3316 /* 2) see if profile is reduced */
3318 return !extended
; /* "0" is valid for usual profiles */
3320 /* true if exactly one bit set */
3321 return (flags
& (flags
- 1)) == 0;
3324 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
3326 /* cancel requested || normal exit path */
3327 return atomic_read(&fs_info
->balance_cancel_req
) ||
3328 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
3329 atomic_read(&fs_info
->balance_cancel_req
) == 0);
3332 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
3336 unset_balance_control(fs_info
);
3337 ret
= del_balance_item(fs_info
->tree_root
);
3339 btrfs_std_error(fs_info
, ret
);
3341 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3345 * Should be called with both balance and volume mutexes held
3347 int btrfs_balance(struct btrfs_balance_control
*bctl
,
3348 struct btrfs_ioctl_balance_args
*bargs
)
3350 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3357 if (btrfs_fs_closing(fs_info
) ||
3358 atomic_read(&fs_info
->balance_pause_req
) ||
3359 atomic_read(&fs_info
->balance_cancel_req
)) {
3364 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
3365 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
3369 * In case of mixed groups both data and meta should be picked,
3370 * and identical options should be given for both of them.
3372 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
3373 if (mixed
&& (bctl
->flags
& allowed
)) {
3374 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
3375 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
3376 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
3377 btrfs_err(fs_info
, "with mixed groups data and "
3378 "metadata balance options must be the same");
3384 num_devices
= fs_info
->fs_devices
->num_devices
;
3385 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
3386 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
3387 BUG_ON(num_devices
< 1);
3390 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
3391 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
3392 if (num_devices
== 1)
3393 allowed
|= BTRFS_BLOCK_GROUP_DUP
;
3394 else if (num_devices
> 1)
3395 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
3396 if (num_devices
> 2)
3397 allowed
|= BTRFS_BLOCK_GROUP_RAID5
;
3398 if (num_devices
> 3)
3399 allowed
|= (BTRFS_BLOCK_GROUP_RAID10
|
3400 BTRFS_BLOCK_GROUP_RAID6
);
3401 if ((bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3402 (!alloc_profile_is_valid(bctl
->data
.target
, 1) ||
3403 (bctl
->data
.target
& ~allowed
))) {
3404 btrfs_err(fs_info
, "unable to start balance with target "
3405 "data profile %llu",
3410 if ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3411 (!alloc_profile_is_valid(bctl
->meta
.target
, 1) ||
3412 (bctl
->meta
.target
& ~allowed
))) {
3414 "unable to start balance with target metadata profile %llu",
3419 if ((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3420 (!alloc_profile_is_valid(bctl
->sys
.target
, 1) ||
3421 (bctl
->sys
.target
& ~allowed
))) {
3423 "unable to start balance with target system profile %llu",
3429 /* allow dup'ed data chunks only in mixed mode */
3430 if (!mixed
&& (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3431 (bctl
->data
.target
& BTRFS_BLOCK_GROUP_DUP
)) {
3432 btrfs_err(fs_info
, "dup for data is not allowed");
3437 /* allow to reduce meta or sys integrity only if force set */
3438 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3439 BTRFS_BLOCK_GROUP_RAID10
|
3440 BTRFS_BLOCK_GROUP_RAID5
|
3441 BTRFS_BLOCK_GROUP_RAID6
;
3443 seq
= read_seqbegin(&fs_info
->profiles_lock
);
3445 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3446 (fs_info
->avail_system_alloc_bits
& allowed
) &&
3447 !(bctl
->sys
.target
& allowed
)) ||
3448 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3449 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
3450 !(bctl
->meta
.target
& allowed
))) {
3451 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
3452 btrfs_info(fs_info
, "force reducing metadata integrity");
3454 btrfs_err(fs_info
, "balance will reduce metadata "
3455 "integrity, use force if you want this");
3460 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
3462 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3463 int num_tolerated_disk_barrier_failures
;
3464 u64 target
= bctl
->sys
.target
;
3466 num_tolerated_disk_barrier_failures
=
3467 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3468 if (num_tolerated_disk_barrier_failures
> 0 &&
3470 (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID0
|
3471 BTRFS_AVAIL_ALLOC_BIT_SINGLE
)))
3472 num_tolerated_disk_barrier_failures
= 0;
3473 else if (num_tolerated_disk_barrier_failures
> 1 &&
3475 (BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)))
3476 num_tolerated_disk_barrier_failures
= 1;
3478 fs_info
->num_tolerated_disk_barrier_failures
=
3479 num_tolerated_disk_barrier_failures
;
3482 ret
= insert_balance_item(fs_info
->tree_root
, bctl
);
3483 if (ret
&& ret
!= -EEXIST
)
3486 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
3487 BUG_ON(ret
== -EEXIST
);
3488 set_balance_control(bctl
);
3490 BUG_ON(ret
!= -EEXIST
);
3491 spin_lock(&fs_info
->balance_lock
);
3492 update_balance_args(bctl
);
3493 spin_unlock(&fs_info
->balance_lock
);
3496 atomic_inc(&fs_info
->balance_running
);
3497 mutex_unlock(&fs_info
->balance_mutex
);
3499 ret
= __btrfs_balance(fs_info
);
3501 mutex_lock(&fs_info
->balance_mutex
);
3502 atomic_dec(&fs_info
->balance_running
);
3504 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3505 fs_info
->num_tolerated_disk_barrier_failures
=
3506 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3510 memset(bargs
, 0, sizeof(*bargs
));
3511 update_ioctl_balance_args(fs_info
, 0, bargs
);
3514 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
3515 balance_need_close(fs_info
)) {
3516 __cancel_balance(fs_info
);
3519 wake_up(&fs_info
->balance_wait_q
);
3523 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
3524 __cancel_balance(fs_info
);
3527 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3532 static int balance_kthread(void *data
)
3534 struct btrfs_fs_info
*fs_info
= data
;
3537 mutex_lock(&fs_info
->volume_mutex
);
3538 mutex_lock(&fs_info
->balance_mutex
);
3540 if (fs_info
->balance_ctl
) {
3541 btrfs_info(fs_info
, "continuing balance");
3542 ret
= btrfs_balance(fs_info
->balance_ctl
, NULL
);
3545 mutex_unlock(&fs_info
->balance_mutex
);
3546 mutex_unlock(&fs_info
->volume_mutex
);
3551 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
3553 struct task_struct
*tsk
;
3555 spin_lock(&fs_info
->balance_lock
);
3556 if (!fs_info
->balance_ctl
) {
3557 spin_unlock(&fs_info
->balance_lock
);
3560 spin_unlock(&fs_info
->balance_lock
);
3562 if (btrfs_test_opt(fs_info
->tree_root
, SKIP_BALANCE
)) {
3563 btrfs_info(fs_info
, "force skipping balance");
3567 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
3568 return PTR_ERR_OR_ZERO(tsk
);
3571 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
3573 struct btrfs_balance_control
*bctl
;
3574 struct btrfs_balance_item
*item
;
3575 struct btrfs_disk_balance_args disk_bargs
;
3576 struct btrfs_path
*path
;
3577 struct extent_buffer
*leaf
;
3578 struct btrfs_key key
;
3581 path
= btrfs_alloc_path();
3585 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3586 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
3589 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
3592 if (ret
> 0) { /* ret = -ENOENT; */
3597 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
3603 leaf
= path
->nodes
[0];
3604 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3606 bctl
->fs_info
= fs_info
;
3607 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
3608 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
3610 btrfs_balance_data(leaf
, item
, &disk_bargs
);
3611 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
3612 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
3613 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
3614 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
3615 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
3617 WARN_ON(atomic_xchg(&fs_info
->mutually_exclusive_operation_running
, 1));
3619 mutex_lock(&fs_info
->volume_mutex
);
3620 mutex_lock(&fs_info
->balance_mutex
);
3622 set_balance_control(bctl
);
3624 mutex_unlock(&fs_info
->balance_mutex
);
3625 mutex_unlock(&fs_info
->volume_mutex
);
3627 btrfs_free_path(path
);
3631 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
3635 mutex_lock(&fs_info
->balance_mutex
);
3636 if (!fs_info
->balance_ctl
) {
3637 mutex_unlock(&fs_info
->balance_mutex
);
3641 if (atomic_read(&fs_info
->balance_running
)) {
3642 atomic_inc(&fs_info
->balance_pause_req
);
3643 mutex_unlock(&fs_info
->balance_mutex
);
3645 wait_event(fs_info
->balance_wait_q
,
3646 atomic_read(&fs_info
->balance_running
) == 0);
3648 mutex_lock(&fs_info
->balance_mutex
);
3649 /* we are good with balance_ctl ripped off from under us */
3650 BUG_ON(atomic_read(&fs_info
->balance_running
));
3651 atomic_dec(&fs_info
->balance_pause_req
);
3656 mutex_unlock(&fs_info
->balance_mutex
);
3660 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
3662 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
3665 mutex_lock(&fs_info
->balance_mutex
);
3666 if (!fs_info
->balance_ctl
) {
3667 mutex_unlock(&fs_info
->balance_mutex
);
3671 atomic_inc(&fs_info
->balance_cancel_req
);
3673 * if we are running just wait and return, balance item is
3674 * deleted in btrfs_balance in this case
3676 if (atomic_read(&fs_info
->balance_running
)) {
3677 mutex_unlock(&fs_info
->balance_mutex
);
3678 wait_event(fs_info
->balance_wait_q
,
3679 atomic_read(&fs_info
->balance_running
) == 0);
3680 mutex_lock(&fs_info
->balance_mutex
);
3682 /* __cancel_balance needs volume_mutex */
3683 mutex_unlock(&fs_info
->balance_mutex
);
3684 mutex_lock(&fs_info
->volume_mutex
);
3685 mutex_lock(&fs_info
->balance_mutex
);
3687 if (fs_info
->balance_ctl
)
3688 __cancel_balance(fs_info
);
3690 mutex_unlock(&fs_info
->volume_mutex
);
3693 BUG_ON(fs_info
->balance_ctl
|| atomic_read(&fs_info
->balance_running
));
3694 atomic_dec(&fs_info
->balance_cancel_req
);
3695 mutex_unlock(&fs_info
->balance_mutex
);
3699 static int btrfs_uuid_scan_kthread(void *data
)
3701 struct btrfs_fs_info
*fs_info
= data
;
3702 struct btrfs_root
*root
= fs_info
->tree_root
;
3703 struct btrfs_key key
;
3704 struct btrfs_key max_key
;
3705 struct btrfs_path
*path
= NULL
;
3707 struct extent_buffer
*eb
;
3709 struct btrfs_root_item root_item
;
3711 struct btrfs_trans_handle
*trans
= NULL
;
3713 path
= btrfs_alloc_path();
3720 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3723 max_key
.objectid
= (u64
)-1;
3724 max_key
.type
= BTRFS_ROOT_ITEM_KEY
;
3725 max_key
.offset
= (u64
)-1;
3728 ret
= btrfs_search_forward(root
, &key
, path
, 0);
3735 if (key
.type
!= BTRFS_ROOT_ITEM_KEY
||
3736 (key
.objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
3737 key
.objectid
!= BTRFS_FS_TREE_OBJECTID
) ||
3738 key
.objectid
> BTRFS_LAST_FREE_OBJECTID
)
3741 eb
= path
->nodes
[0];
3742 slot
= path
->slots
[0];
3743 item_size
= btrfs_item_size_nr(eb
, slot
);
3744 if (item_size
< sizeof(root_item
))
3747 read_extent_buffer(eb
, &root_item
,
3748 btrfs_item_ptr_offset(eb
, slot
),
3749 (int)sizeof(root_item
));
3750 if (btrfs_root_refs(&root_item
) == 0)
3753 if (!btrfs_is_empty_uuid(root_item
.uuid
) ||
3754 !btrfs_is_empty_uuid(root_item
.received_uuid
)) {
3758 btrfs_release_path(path
);
3760 * 1 - subvol uuid item
3761 * 1 - received_subvol uuid item
3763 trans
= btrfs_start_transaction(fs_info
->uuid_root
, 2);
3764 if (IS_ERR(trans
)) {
3765 ret
= PTR_ERR(trans
);
3773 if (!btrfs_is_empty_uuid(root_item
.uuid
)) {
3774 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
3776 BTRFS_UUID_KEY_SUBVOL
,
3779 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
3785 if (!btrfs_is_empty_uuid(root_item
.received_uuid
)) {
3786 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
3787 root_item
.received_uuid
,
3788 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
3791 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
3799 ret
= btrfs_end_transaction(trans
, fs_info
->uuid_root
);
3805 btrfs_release_path(path
);
3806 if (key
.offset
< (u64
)-1) {
3808 } else if (key
.type
< BTRFS_ROOT_ITEM_KEY
) {
3810 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3811 } else if (key
.objectid
< (u64
)-1) {
3813 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3822 btrfs_free_path(path
);
3823 if (trans
&& !IS_ERR(trans
))
3824 btrfs_end_transaction(trans
, fs_info
->uuid_root
);
3826 btrfs_warn(fs_info
, "btrfs_uuid_scan_kthread failed %d", ret
);
3828 fs_info
->update_uuid_tree_gen
= 1;
3829 up(&fs_info
->uuid_tree_rescan_sem
);
3834 * Callback for btrfs_uuid_tree_iterate().
3836 * 0 check succeeded, the entry is not outdated.
3837 * < 0 if an error occured.
3838 * > 0 if the check failed, which means the caller shall remove the entry.
3840 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info
*fs_info
,
3841 u8
*uuid
, u8 type
, u64 subid
)
3843 struct btrfs_key key
;
3845 struct btrfs_root
*subvol_root
;
3847 if (type
!= BTRFS_UUID_KEY_SUBVOL
&&
3848 type
!= BTRFS_UUID_KEY_RECEIVED_SUBVOL
)
3851 key
.objectid
= subid
;
3852 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3853 key
.offset
= (u64
)-1;
3854 subvol_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
3855 if (IS_ERR(subvol_root
)) {
3856 ret
= PTR_ERR(subvol_root
);
3863 case BTRFS_UUID_KEY_SUBVOL
:
3864 if (memcmp(uuid
, subvol_root
->root_item
.uuid
, BTRFS_UUID_SIZE
))
3867 case BTRFS_UUID_KEY_RECEIVED_SUBVOL
:
3868 if (memcmp(uuid
, subvol_root
->root_item
.received_uuid
,
3878 static int btrfs_uuid_rescan_kthread(void *data
)
3880 struct btrfs_fs_info
*fs_info
= (struct btrfs_fs_info
*)data
;
3884 * 1st step is to iterate through the existing UUID tree and
3885 * to delete all entries that contain outdated data.
3886 * 2nd step is to add all missing entries to the UUID tree.
3888 ret
= btrfs_uuid_tree_iterate(fs_info
, btrfs_check_uuid_tree_entry
);
3890 btrfs_warn(fs_info
, "iterating uuid_tree failed %d", ret
);
3891 up(&fs_info
->uuid_tree_rescan_sem
);
3894 return btrfs_uuid_scan_kthread(data
);
3897 int btrfs_create_uuid_tree(struct btrfs_fs_info
*fs_info
)
3899 struct btrfs_trans_handle
*trans
;
3900 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
3901 struct btrfs_root
*uuid_root
;
3902 struct task_struct
*task
;
3909 trans
= btrfs_start_transaction(tree_root
, 2);
3911 return PTR_ERR(trans
);
3913 uuid_root
= btrfs_create_tree(trans
, fs_info
,
3914 BTRFS_UUID_TREE_OBJECTID
);
3915 if (IS_ERR(uuid_root
)) {
3916 btrfs_abort_transaction(trans
, tree_root
,
3917 PTR_ERR(uuid_root
));
3918 return PTR_ERR(uuid_root
);
3921 fs_info
->uuid_root
= uuid_root
;
3923 ret
= btrfs_commit_transaction(trans
, tree_root
);
3927 down(&fs_info
->uuid_tree_rescan_sem
);
3928 task
= kthread_run(btrfs_uuid_scan_kthread
, fs_info
, "btrfs-uuid");
3930 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3931 btrfs_warn(fs_info
, "failed to start uuid_scan task");
3932 up(&fs_info
->uuid_tree_rescan_sem
);
3933 return PTR_ERR(task
);
3939 int btrfs_check_uuid_tree(struct btrfs_fs_info
*fs_info
)
3941 struct task_struct
*task
;
3943 down(&fs_info
->uuid_tree_rescan_sem
);
3944 task
= kthread_run(btrfs_uuid_rescan_kthread
, fs_info
, "btrfs-uuid");
3946 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3947 btrfs_warn(fs_info
, "failed to start uuid_rescan task");
3948 up(&fs_info
->uuid_tree_rescan_sem
);
3949 return PTR_ERR(task
);
3956 * shrinking a device means finding all of the device extents past
3957 * the new size, and then following the back refs to the chunks.
3958 * The chunk relocation code actually frees the device extent
3960 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
3962 struct btrfs_trans_handle
*trans
;
3963 struct btrfs_root
*root
= device
->dev_root
;
3964 struct btrfs_dev_extent
*dev_extent
= NULL
;
3965 struct btrfs_path
*path
;
3972 bool retried
= false;
3973 struct extent_buffer
*l
;
3974 struct btrfs_key key
;
3975 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
3976 u64 old_total
= btrfs_super_total_bytes(super_copy
);
3977 u64 old_size
= btrfs_device_get_total_bytes(device
);
3978 u64 diff
= old_size
- new_size
;
3980 if (device
->is_tgtdev_for_dev_replace
)
3983 path
= btrfs_alloc_path();
3991 btrfs_device_set_total_bytes(device
, new_size
);
3992 if (device
->writeable
) {
3993 device
->fs_devices
->total_rw_bytes
-= diff
;
3994 spin_lock(&root
->fs_info
->free_chunk_lock
);
3995 root
->fs_info
->free_chunk_space
-= diff
;
3996 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3998 unlock_chunks(root
);
4001 key
.objectid
= device
->devid
;
4002 key
.offset
= (u64
)-1;
4003 key
.type
= BTRFS_DEV_EXTENT_KEY
;
4006 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4010 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
4015 btrfs_release_path(path
);
4020 slot
= path
->slots
[0];
4021 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
4023 if (key
.objectid
!= device
->devid
) {
4024 btrfs_release_path(path
);
4028 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
4029 length
= btrfs_dev_extent_length(l
, dev_extent
);
4031 if (key
.offset
+ length
<= new_size
) {
4032 btrfs_release_path(path
);
4036 chunk_objectid
= btrfs_dev_extent_chunk_objectid(l
, dev_extent
);
4037 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
4038 btrfs_release_path(path
);
4040 ret
= btrfs_relocate_chunk(root
, chunk_objectid
, chunk_offset
);
4041 if (ret
&& ret
!= -ENOSPC
)
4045 } while (key
.offset
-- > 0);
4047 if (failed
&& !retried
) {
4051 } else if (failed
&& retried
) {
4055 btrfs_device_set_total_bytes(device
, old_size
);
4056 if (device
->writeable
)
4057 device
->fs_devices
->total_rw_bytes
+= diff
;
4058 spin_lock(&root
->fs_info
->free_chunk_lock
);
4059 root
->fs_info
->free_chunk_space
+= diff
;
4060 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4061 unlock_chunks(root
);
4065 /* Shrinking succeeded, else we would be at "done". */
4066 trans
= btrfs_start_transaction(root
, 0);
4067 if (IS_ERR(trans
)) {
4068 ret
= PTR_ERR(trans
);
4073 btrfs_device_set_disk_total_bytes(device
, new_size
);
4074 if (list_empty(&device
->resized_list
))
4075 list_add_tail(&device
->resized_list
,
4076 &root
->fs_info
->fs_devices
->resized_devices
);
4078 WARN_ON(diff
> old_total
);
4079 btrfs_set_super_total_bytes(super_copy
, old_total
- diff
);
4080 unlock_chunks(root
);
4082 /* Now btrfs_update_device() will change the on-disk size. */
4083 ret
= btrfs_update_device(trans
, device
);
4084 btrfs_end_transaction(trans
, root
);
4086 btrfs_free_path(path
);
4090 static int btrfs_add_system_chunk(struct btrfs_root
*root
,
4091 struct btrfs_key
*key
,
4092 struct btrfs_chunk
*chunk
, int item_size
)
4094 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
4095 struct btrfs_disk_key disk_key
;
4100 array_size
= btrfs_super_sys_array_size(super_copy
);
4101 if (array_size
+ item_size
+ sizeof(disk_key
)
4102 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4103 unlock_chunks(root
);
4107 ptr
= super_copy
->sys_chunk_array
+ array_size
;
4108 btrfs_cpu_key_to_disk(&disk_key
, key
);
4109 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
4110 ptr
+= sizeof(disk_key
);
4111 memcpy(ptr
, chunk
, item_size
);
4112 item_size
+= sizeof(disk_key
);
4113 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
4114 unlock_chunks(root
);
4120 * sort the devices in descending order by max_avail, total_avail
4122 static int btrfs_cmp_device_info(const void *a
, const void *b
)
4124 const struct btrfs_device_info
*di_a
= a
;
4125 const struct btrfs_device_info
*di_b
= b
;
4127 if (di_a
->max_avail
> di_b
->max_avail
)
4129 if (di_a
->max_avail
< di_b
->max_avail
)
4131 if (di_a
->total_avail
> di_b
->total_avail
)
4133 if (di_a
->total_avail
< di_b
->total_avail
)
4138 static const struct btrfs_raid_attr btrfs_raid_array
[BTRFS_NR_RAID_TYPES
] = {
4139 [BTRFS_RAID_RAID10
] = {
4142 .devs_max
= 0, /* 0 == as many as possible */
4144 .devs_increment
= 2,
4147 [BTRFS_RAID_RAID1
] = {
4152 .devs_increment
= 2,
4155 [BTRFS_RAID_DUP
] = {
4160 .devs_increment
= 1,
4163 [BTRFS_RAID_RAID0
] = {
4168 .devs_increment
= 1,
4171 [BTRFS_RAID_SINGLE
] = {
4176 .devs_increment
= 1,
4179 [BTRFS_RAID_RAID5
] = {
4184 .devs_increment
= 1,
4187 [BTRFS_RAID_RAID6
] = {
4192 .devs_increment
= 1,
4197 static u32
find_raid56_stripe_len(u32 data_devices
, u32 dev_stripe_target
)
4199 /* TODO allow them to set a preferred stripe size */
4203 static void check_raid56_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
4205 if (!(type
& BTRFS_BLOCK_GROUP_RAID56_MASK
))
4208 btrfs_set_fs_incompat(info
, RAID56
);
4211 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4212 - sizeof(struct btrfs_item) \
4213 - sizeof(struct btrfs_chunk)) \
4214 / sizeof(struct btrfs_stripe) + 1)
4216 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4217 - 2 * sizeof(struct btrfs_disk_key) \
4218 - 2 * sizeof(struct btrfs_chunk)) \
4219 / sizeof(struct btrfs_stripe) + 1)
4221 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4222 struct btrfs_root
*extent_root
, u64 start
,
4225 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
4226 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
4227 struct list_head
*cur
;
4228 struct map_lookup
*map
= NULL
;
4229 struct extent_map_tree
*em_tree
;
4230 struct extent_map
*em
;
4231 struct btrfs_device_info
*devices_info
= NULL
;
4233 int num_stripes
; /* total number of stripes to allocate */
4234 int data_stripes
; /* number of stripes that count for
4236 int sub_stripes
; /* sub_stripes info for map */
4237 int dev_stripes
; /* stripes per dev */
4238 int devs_max
; /* max devs to use */
4239 int devs_min
; /* min devs needed */
4240 int devs_increment
; /* ndevs has to be a multiple of this */
4241 int ncopies
; /* how many copies to data has */
4243 u64 max_stripe_size
;
4247 u64 raid_stripe_len
= BTRFS_STRIPE_LEN
;
4253 BUG_ON(!alloc_profile_is_valid(type
, 0));
4255 if (list_empty(&fs_devices
->alloc_list
))
4258 index
= __get_raid_index(type
);
4260 sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
4261 dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
4262 devs_max
= btrfs_raid_array
[index
].devs_max
;
4263 devs_min
= btrfs_raid_array
[index
].devs_min
;
4264 devs_increment
= btrfs_raid_array
[index
].devs_increment
;
4265 ncopies
= btrfs_raid_array
[index
].ncopies
;
4267 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
4268 max_stripe_size
= 1024 * 1024 * 1024;
4269 max_chunk_size
= 10 * max_stripe_size
;
4271 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4272 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
4273 /* for larger filesystems, use larger metadata chunks */
4274 if (fs_devices
->total_rw_bytes
> 50ULL * 1024 * 1024 * 1024)
4275 max_stripe_size
= 1024 * 1024 * 1024;
4277 max_stripe_size
= 256 * 1024 * 1024;
4278 max_chunk_size
= max_stripe_size
;
4280 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4281 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4282 max_stripe_size
= 32 * 1024 * 1024;
4283 max_chunk_size
= 2 * max_stripe_size
;
4285 devs_max
= BTRFS_MAX_DEVS_SYS_CHUNK
;
4287 btrfs_err(info
, "invalid chunk type 0x%llx requested",
4292 /* we don't want a chunk larger than 10% of writeable space */
4293 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
4296 devices_info
= kcalloc(fs_devices
->rw_devices
, sizeof(*devices_info
),
4301 cur
= fs_devices
->alloc_list
.next
;
4304 * in the first pass through the devices list, we gather information
4305 * about the available holes on each device.
4308 while (cur
!= &fs_devices
->alloc_list
) {
4309 struct btrfs_device
*device
;
4313 device
= list_entry(cur
, struct btrfs_device
, dev_alloc_list
);
4317 if (!device
->writeable
) {
4319 "BTRFS: read-only device in alloc_list\n");
4323 if (!device
->in_fs_metadata
||
4324 device
->is_tgtdev_for_dev_replace
)
4327 if (device
->total_bytes
> device
->bytes_used
)
4328 total_avail
= device
->total_bytes
- device
->bytes_used
;
4332 /* If there is no space on this device, skip it. */
4333 if (total_avail
== 0)
4336 ret
= find_free_dev_extent(trans
, device
,
4337 max_stripe_size
* dev_stripes
,
4338 &dev_offset
, &max_avail
);
4339 if (ret
&& ret
!= -ENOSPC
)
4343 max_avail
= max_stripe_size
* dev_stripes
;
4345 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
4348 if (ndevs
== fs_devices
->rw_devices
) {
4349 WARN(1, "%s: found more than %llu devices\n",
4350 __func__
, fs_devices
->rw_devices
);
4353 devices_info
[ndevs
].dev_offset
= dev_offset
;
4354 devices_info
[ndevs
].max_avail
= max_avail
;
4355 devices_info
[ndevs
].total_avail
= total_avail
;
4356 devices_info
[ndevs
].dev
= device
;
4361 * now sort the devices by hole size / available space
4363 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
4364 btrfs_cmp_device_info
, NULL
);
4366 /* round down to number of usable stripes */
4367 ndevs
-= ndevs
% devs_increment
;
4369 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
4374 if (devs_max
&& ndevs
> devs_max
)
4377 * the primary goal is to maximize the number of stripes, so use as many
4378 * devices as possible, even if the stripes are not maximum sized.
4380 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4381 num_stripes
= ndevs
* dev_stripes
;
4384 * this will have to be fixed for RAID1 and RAID10 over
4387 data_stripes
= num_stripes
/ ncopies
;
4389 if (type
& BTRFS_BLOCK_GROUP_RAID5
) {
4390 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 1,
4391 btrfs_super_stripesize(info
->super_copy
));
4392 data_stripes
= num_stripes
- 1;
4394 if (type
& BTRFS_BLOCK_GROUP_RAID6
) {
4395 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 2,
4396 btrfs_super_stripesize(info
->super_copy
));
4397 data_stripes
= num_stripes
- 2;
4401 * Use the number of data stripes to figure out how big this chunk
4402 * is really going to be in terms of logical address space,
4403 * and compare that answer with the max chunk size
4405 if (stripe_size
* data_stripes
> max_chunk_size
) {
4406 u64 mask
= (1ULL << 24) - 1;
4408 stripe_size
= div_u64(max_chunk_size
, data_stripes
);
4410 /* bump the answer up to a 16MB boundary */
4411 stripe_size
= (stripe_size
+ mask
) & ~mask
;
4413 /* but don't go higher than the limits we found
4414 * while searching for free extents
4416 if (stripe_size
> devices_info
[ndevs
-1].max_avail
)
4417 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4420 stripe_size
= div_u64(stripe_size
, dev_stripes
);
4422 /* align to BTRFS_STRIPE_LEN */
4423 stripe_size
= div_u64(stripe_size
, raid_stripe_len
);
4424 stripe_size
*= raid_stripe_len
;
4426 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
4431 map
->num_stripes
= num_stripes
;
4433 for (i
= 0; i
< ndevs
; ++i
) {
4434 for (j
= 0; j
< dev_stripes
; ++j
) {
4435 int s
= i
* dev_stripes
+ j
;
4436 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
4437 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
4441 map
->sector_size
= extent_root
->sectorsize
;
4442 map
->stripe_len
= raid_stripe_len
;
4443 map
->io_align
= raid_stripe_len
;
4444 map
->io_width
= raid_stripe_len
;
4446 map
->sub_stripes
= sub_stripes
;
4448 num_bytes
= stripe_size
* data_stripes
;
4450 trace_btrfs_chunk_alloc(info
->chunk_root
, map
, start
, num_bytes
);
4452 em
= alloc_extent_map();
4458 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
4459 em
->bdev
= (struct block_device
*)map
;
4461 em
->len
= num_bytes
;
4462 em
->block_start
= 0;
4463 em
->block_len
= em
->len
;
4464 em
->orig_block_len
= stripe_size
;
4466 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
4467 write_lock(&em_tree
->lock
);
4468 ret
= add_extent_mapping(em_tree
, em
, 0);
4470 list_add_tail(&em
->list
, &trans
->transaction
->pending_chunks
);
4471 atomic_inc(&em
->refs
);
4473 write_unlock(&em_tree
->lock
);
4475 free_extent_map(em
);
4479 ret
= btrfs_make_block_group(trans
, extent_root
, 0, type
,
4480 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4483 goto error_del_extent
;
4485 for (i
= 0; i
< map
->num_stripes
; i
++) {
4486 num_bytes
= map
->stripes
[i
].dev
->bytes_used
+ stripe_size
;
4487 btrfs_device_set_bytes_used(map
->stripes
[i
].dev
, num_bytes
);
4490 spin_lock(&extent_root
->fs_info
->free_chunk_lock
);
4491 extent_root
->fs_info
->free_chunk_space
-= (stripe_size
*
4493 spin_unlock(&extent_root
->fs_info
->free_chunk_lock
);
4495 free_extent_map(em
);
4496 check_raid56_incompat_flag(extent_root
->fs_info
, type
);
4498 kfree(devices_info
);
4502 write_lock(&em_tree
->lock
);
4503 remove_extent_mapping(em_tree
, em
);
4504 write_unlock(&em_tree
->lock
);
4506 /* One for our allocation */
4507 free_extent_map(em
);
4508 /* One for the tree reference */
4509 free_extent_map(em
);
4510 /* One for the pending_chunks list reference */
4511 free_extent_map(em
);
4513 kfree(devices_info
);
4517 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
4518 struct btrfs_root
*extent_root
,
4519 u64 chunk_offset
, u64 chunk_size
)
4521 struct btrfs_key key
;
4522 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
4523 struct btrfs_device
*device
;
4524 struct btrfs_chunk
*chunk
;
4525 struct btrfs_stripe
*stripe
;
4526 struct extent_map_tree
*em_tree
;
4527 struct extent_map
*em
;
4528 struct map_lookup
*map
;
4535 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
4536 read_lock(&em_tree
->lock
);
4537 em
= lookup_extent_mapping(em_tree
, chunk_offset
, chunk_size
);
4538 read_unlock(&em_tree
->lock
);
4541 btrfs_crit(extent_root
->fs_info
, "unable to find logical "
4542 "%Lu len %Lu", chunk_offset
, chunk_size
);
4546 if (em
->start
!= chunk_offset
|| em
->len
!= chunk_size
) {
4547 btrfs_crit(extent_root
->fs_info
, "found a bad mapping, wanted"
4548 " %Lu-%Lu, found %Lu-%Lu", chunk_offset
,
4549 chunk_size
, em
->start
, em
->len
);
4550 free_extent_map(em
);
4554 map
= (struct map_lookup
*)em
->bdev
;
4555 item_size
= btrfs_chunk_item_size(map
->num_stripes
);
4556 stripe_size
= em
->orig_block_len
;
4558 chunk
= kzalloc(item_size
, GFP_NOFS
);
4564 for (i
= 0; i
< map
->num_stripes
; i
++) {
4565 device
= map
->stripes
[i
].dev
;
4566 dev_offset
= map
->stripes
[i
].physical
;
4568 ret
= btrfs_update_device(trans
, device
);
4571 ret
= btrfs_alloc_dev_extent(trans
, device
,
4572 chunk_root
->root_key
.objectid
,
4573 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4574 chunk_offset
, dev_offset
,
4580 stripe
= &chunk
->stripe
;
4581 for (i
= 0; i
< map
->num_stripes
; i
++) {
4582 device
= map
->stripes
[i
].dev
;
4583 dev_offset
= map
->stripes
[i
].physical
;
4585 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
4586 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
4587 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
4591 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
4592 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
4593 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
4594 btrfs_set_stack_chunk_type(chunk
, map
->type
);
4595 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
4596 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
4597 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
4598 btrfs_set_stack_chunk_sector_size(chunk
, extent_root
->sectorsize
);
4599 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
4601 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
4602 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
4603 key
.offset
= chunk_offset
;
4605 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
4606 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4608 * TODO: Cleanup of inserted chunk root in case of
4611 ret
= btrfs_add_system_chunk(chunk_root
, &key
, chunk
,
4617 free_extent_map(em
);
4622 * Chunk allocation falls into two parts. The first part does works
4623 * that make the new allocated chunk useable, but not do any operation
4624 * that modifies the chunk tree. The second part does the works that
4625 * require modifying the chunk tree. This division is important for the
4626 * bootstrap process of adding storage to a seed btrfs.
4628 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4629 struct btrfs_root
*extent_root
, u64 type
)
4633 ASSERT(mutex_is_locked(&extent_root
->fs_info
->chunk_mutex
));
4634 chunk_offset
= find_next_chunk(extent_root
->fs_info
);
4635 return __btrfs_alloc_chunk(trans
, extent_root
, chunk_offset
, type
);
4638 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
4639 struct btrfs_root
*root
,
4640 struct btrfs_device
*device
)
4643 u64 sys_chunk_offset
;
4645 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4646 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
4649 chunk_offset
= find_next_chunk(fs_info
);
4650 alloc_profile
= btrfs_get_alloc_profile(extent_root
, 0);
4651 ret
= __btrfs_alloc_chunk(trans
, extent_root
, chunk_offset
,
4656 sys_chunk_offset
= find_next_chunk(root
->fs_info
);
4657 alloc_profile
= btrfs_get_alloc_profile(fs_info
->chunk_root
, 0);
4658 ret
= __btrfs_alloc_chunk(trans
, extent_root
, sys_chunk_offset
,
4663 static inline int btrfs_chunk_max_errors(struct map_lookup
*map
)
4667 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
4668 BTRFS_BLOCK_GROUP_RAID10
|
4669 BTRFS_BLOCK_GROUP_RAID5
|
4670 BTRFS_BLOCK_GROUP_DUP
)) {
4672 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
) {
4681 int btrfs_chunk_readonly(struct btrfs_root
*root
, u64 chunk_offset
)
4683 struct extent_map
*em
;
4684 struct map_lookup
*map
;
4685 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
4690 read_lock(&map_tree
->map_tree
.lock
);
4691 em
= lookup_extent_mapping(&map_tree
->map_tree
, chunk_offset
, 1);
4692 read_unlock(&map_tree
->map_tree
.lock
);
4696 map
= (struct map_lookup
*)em
->bdev
;
4697 for (i
= 0; i
< map
->num_stripes
; i
++) {
4698 if (map
->stripes
[i
].dev
->missing
) {
4703 if (!map
->stripes
[i
].dev
->writeable
) {
4710 * If the number of missing devices is larger than max errors,
4711 * we can not write the data into that chunk successfully, so
4714 if (miss_ndevs
> btrfs_chunk_max_errors(map
))
4717 free_extent_map(em
);
4721 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
4723 extent_map_tree_init(&tree
->map_tree
);
4726 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
4728 struct extent_map
*em
;
4731 write_lock(&tree
->map_tree
.lock
);
4732 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
4734 remove_extent_mapping(&tree
->map_tree
, em
);
4735 write_unlock(&tree
->map_tree
.lock
);
4739 free_extent_map(em
);
4740 /* once for the tree */
4741 free_extent_map(em
);
4745 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
4747 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
4748 struct extent_map
*em
;
4749 struct map_lookup
*map
;
4750 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4753 read_lock(&em_tree
->lock
);
4754 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4755 read_unlock(&em_tree
->lock
);
4758 * We could return errors for these cases, but that could get ugly and
4759 * we'd probably do the same thing which is just not do anything else
4760 * and exit, so return 1 so the callers don't try to use other copies.
4763 btrfs_crit(fs_info
, "No mapping for %Lu-%Lu", logical
,
4768 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
4769 btrfs_crit(fs_info
, "Invalid mapping for %Lu-%Lu, got "
4770 "%Lu-%Lu", logical
, logical
+len
, em
->start
,
4771 em
->start
+ em
->len
);
4772 free_extent_map(em
);
4776 map
= (struct map_lookup
*)em
->bdev
;
4777 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
4778 ret
= map
->num_stripes
;
4779 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
4780 ret
= map
->sub_stripes
;
4781 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID5
)
4783 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
4787 free_extent_map(em
);
4789 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
4790 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
))
4792 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
4797 unsigned long btrfs_full_stripe_len(struct btrfs_root
*root
,
4798 struct btrfs_mapping_tree
*map_tree
,
4801 struct extent_map
*em
;
4802 struct map_lookup
*map
;
4803 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4804 unsigned long len
= root
->sectorsize
;
4806 read_lock(&em_tree
->lock
);
4807 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4808 read_unlock(&em_tree
->lock
);
4811 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4812 map
= (struct map_lookup
*)em
->bdev
;
4813 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
4814 len
= map
->stripe_len
* nr_data_stripes(map
);
4815 free_extent_map(em
);
4819 int btrfs_is_parity_mirror(struct btrfs_mapping_tree
*map_tree
,
4820 u64 logical
, u64 len
, int mirror_num
)
4822 struct extent_map
*em
;
4823 struct map_lookup
*map
;
4824 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4827 read_lock(&em_tree
->lock
);
4828 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4829 read_unlock(&em_tree
->lock
);
4832 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4833 map
= (struct map_lookup
*)em
->bdev
;
4834 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
4836 free_extent_map(em
);
4840 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
4841 struct map_lookup
*map
, int first
, int num
,
4842 int optimal
, int dev_replace_is_ongoing
)
4846 struct btrfs_device
*srcdev
;
4848 if (dev_replace_is_ongoing
&&
4849 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
4850 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
4851 srcdev
= fs_info
->dev_replace
.srcdev
;
4856 * try to avoid the drive that is the source drive for a
4857 * dev-replace procedure, only choose it if no other non-missing
4858 * mirror is available
4860 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
4861 if (map
->stripes
[optimal
].dev
->bdev
&&
4862 (tolerance
|| map
->stripes
[optimal
].dev
!= srcdev
))
4864 for (i
= first
; i
< first
+ num
; i
++) {
4865 if (map
->stripes
[i
].dev
->bdev
&&
4866 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
4871 /* we couldn't find one that doesn't fail. Just return something
4872 * and the io error handling code will clean up eventually
4877 static inline int parity_smaller(u64 a
, u64 b
)
4882 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4883 static void sort_parity_stripes(struct btrfs_bio
*bbio
, int num_stripes
)
4885 struct btrfs_bio_stripe s
;
4892 for (i
= 0; i
< num_stripes
- 1; i
++) {
4893 if (parity_smaller(bbio
->raid_map
[i
],
4894 bbio
->raid_map
[i
+1])) {
4895 s
= bbio
->stripes
[i
];
4896 l
= bbio
->raid_map
[i
];
4897 bbio
->stripes
[i
] = bbio
->stripes
[i
+1];
4898 bbio
->raid_map
[i
] = bbio
->raid_map
[i
+1];
4899 bbio
->stripes
[i
+1] = s
;
4900 bbio
->raid_map
[i
+1] = l
;
4908 static struct btrfs_bio
*alloc_btrfs_bio(int total_stripes
, int real_stripes
)
4910 struct btrfs_bio
*bbio
= kzalloc(
4911 /* the size of the btrfs_bio */
4912 sizeof(struct btrfs_bio
) +
4913 /* plus the variable array for the stripes */
4914 sizeof(struct btrfs_bio_stripe
) * (total_stripes
) +
4915 /* plus the variable array for the tgt dev */
4916 sizeof(int) * (real_stripes
) +
4918 * plus the raid_map, which includes both the tgt dev
4921 sizeof(u64
) * (total_stripes
),
4926 atomic_set(&bbio
->error
, 0);
4927 atomic_set(&bbio
->refs
, 1);
4932 void btrfs_get_bbio(struct btrfs_bio
*bbio
)
4934 WARN_ON(!atomic_read(&bbio
->refs
));
4935 atomic_inc(&bbio
->refs
);
4938 void btrfs_put_bbio(struct btrfs_bio
*bbio
)
4942 if (atomic_dec_and_test(&bbio
->refs
))
4946 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
4947 u64 logical
, u64
*length
,
4948 struct btrfs_bio
**bbio_ret
,
4949 int mirror_num
, int need_raid_map
)
4951 struct extent_map
*em
;
4952 struct map_lookup
*map
;
4953 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
4954 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4957 u64 stripe_end_offset
;
4967 int tgtdev_indexes
= 0;
4968 struct btrfs_bio
*bbio
= NULL
;
4969 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
4970 int dev_replace_is_ongoing
= 0;
4971 int num_alloc_stripes
;
4972 int patch_the_first_stripe_for_dev_replace
= 0;
4973 u64 physical_to_patch_in_first_stripe
= 0;
4974 u64 raid56_full_stripe_start
= (u64
)-1;
4976 read_lock(&em_tree
->lock
);
4977 em
= lookup_extent_mapping(em_tree
, logical
, *length
);
4978 read_unlock(&em_tree
->lock
);
4981 btrfs_crit(fs_info
, "unable to find logical %llu len %llu",
4986 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
4987 btrfs_crit(fs_info
, "found a bad mapping, wanted %Lu, "
4988 "found %Lu-%Lu", logical
, em
->start
,
4989 em
->start
+ em
->len
);
4990 free_extent_map(em
);
4994 map
= (struct map_lookup
*)em
->bdev
;
4995 offset
= logical
- em
->start
;
4997 stripe_len
= map
->stripe_len
;
5000 * stripe_nr counts the total number of stripes we have to stride
5001 * to get to this block
5003 stripe_nr
= div64_u64(stripe_nr
, stripe_len
);
5005 stripe_offset
= stripe_nr
* stripe_len
;
5006 BUG_ON(offset
< stripe_offset
);
5008 /* stripe_offset is the offset of this block in its stripe*/
5009 stripe_offset
= offset
- stripe_offset
;
5011 /* if we're here for raid56, we need to know the stripe aligned start */
5012 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5013 unsigned long full_stripe_len
= stripe_len
* nr_data_stripes(map
);
5014 raid56_full_stripe_start
= offset
;
5016 /* allow a write of a full stripe, but make sure we don't
5017 * allow straddling of stripes
5019 raid56_full_stripe_start
= div64_u64(raid56_full_stripe_start
,
5021 raid56_full_stripe_start
*= full_stripe_len
;
5024 if (rw
& REQ_DISCARD
) {
5025 /* we don't discard raid56 yet */
5026 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5030 *length
= min_t(u64
, em
->len
- offset
, *length
);
5031 } else if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
5033 /* For writes to RAID[56], allow a full stripeset across all disks.
5034 For other RAID types and for RAID[56] reads, just allow a single
5035 stripe (on a single disk). */
5036 if ((map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
5038 max_len
= stripe_len
* nr_data_stripes(map
) -
5039 (offset
- raid56_full_stripe_start
);
5041 /* we limit the length of each bio to what fits in a stripe */
5042 max_len
= stripe_len
- stripe_offset
;
5044 *length
= min_t(u64
, em
->len
- offset
, max_len
);
5046 *length
= em
->len
- offset
;
5049 /* This is for when we're called from btrfs_merge_bio_hook() and all
5050 it cares about is the length */
5054 btrfs_dev_replace_lock(dev_replace
);
5055 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
5056 if (!dev_replace_is_ongoing
)
5057 btrfs_dev_replace_unlock(dev_replace
);
5059 if (dev_replace_is_ongoing
&& mirror_num
== map
->num_stripes
+ 1 &&
5060 !(rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) &&
5061 dev_replace
->tgtdev
!= NULL
) {
5063 * in dev-replace case, for repair case (that's the only
5064 * case where the mirror is selected explicitly when
5065 * calling btrfs_map_block), blocks left of the left cursor
5066 * can also be read from the target drive.
5067 * For REQ_GET_READ_MIRRORS, the target drive is added as
5068 * the last one to the array of stripes. For READ, it also
5069 * needs to be supported using the same mirror number.
5070 * If the requested block is not left of the left cursor,
5071 * EIO is returned. This can happen because btrfs_num_copies()
5072 * returns one more in the dev-replace case.
5074 u64 tmp_length
= *length
;
5075 struct btrfs_bio
*tmp_bbio
= NULL
;
5076 int tmp_num_stripes
;
5077 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5078 int index_srcdev
= 0;
5080 u64 physical_of_found
= 0;
5082 ret
= __btrfs_map_block(fs_info
, REQ_GET_READ_MIRRORS
,
5083 logical
, &tmp_length
, &tmp_bbio
, 0, 0);
5085 WARN_ON(tmp_bbio
!= NULL
);
5089 tmp_num_stripes
= tmp_bbio
->num_stripes
;
5090 if (mirror_num
> tmp_num_stripes
) {
5092 * REQ_GET_READ_MIRRORS does not contain this
5093 * mirror, that means that the requested area
5094 * is not left of the left cursor
5097 btrfs_put_bbio(tmp_bbio
);
5102 * process the rest of the function using the mirror_num
5103 * of the source drive. Therefore look it up first.
5104 * At the end, patch the device pointer to the one of the
5107 for (i
= 0; i
< tmp_num_stripes
; i
++) {
5108 if (tmp_bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5110 * In case of DUP, in order to keep it
5111 * simple, only add the mirror with the
5112 * lowest physical address
5115 physical_of_found
<=
5116 tmp_bbio
->stripes
[i
].physical
)
5121 tmp_bbio
->stripes
[i
].physical
;
5126 mirror_num
= index_srcdev
+ 1;
5127 patch_the_first_stripe_for_dev_replace
= 1;
5128 physical_to_patch_in_first_stripe
= physical_of_found
;
5132 btrfs_put_bbio(tmp_bbio
);
5136 btrfs_put_bbio(tmp_bbio
);
5137 } else if (mirror_num
> map
->num_stripes
) {
5143 stripe_nr_orig
= stripe_nr
;
5144 stripe_nr_end
= ALIGN(offset
+ *length
, map
->stripe_len
);
5145 stripe_nr_end
= div_u64(stripe_nr_end
, map
->stripe_len
);
5146 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
5149 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5150 if (rw
& REQ_DISCARD
)
5151 num_stripes
= min_t(u64
, map
->num_stripes
,
5152 stripe_nr_end
- stripe_nr_orig
);
5153 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5155 if (!(rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)))
5157 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
5158 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
))
5159 num_stripes
= map
->num_stripes
;
5160 else if (mirror_num
)
5161 stripe_index
= mirror_num
- 1;
5163 stripe_index
= find_live_mirror(fs_info
, map
, 0,
5165 current
->pid
% map
->num_stripes
,
5166 dev_replace_is_ongoing
);
5167 mirror_num
= stripe_index
+ 1;
5170 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
5171 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) {
5172 num_stripes
= map
->num_stripes
;
5173 } else if (mirror_num
) {
5174 stripe_index
= mirror_num
- 1;
5179 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5180 u32 factor
= map
->num_stripes
/ map
->sub_stripes
;
5182 stripe_nr
= div_u64_rem(stripe_nr
, factor
, &stripe_index
);
5183 stripe_index
*= map
->sub_stripes
;
5185 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
))
5186 num_stripes
= map
->sub_stripes
;
5187 else if (rw
& REQ_DISCARD
)
5188 num_stripes
= min_t(u64
, map
->sub_stripes
*
5189 (stripe_nr_end
- stripe_nr_orig
),
5191 else if (mirror_num
)
5192 stripe_index
+= mirror_num
- 1;
5194 int old_stripe_index
= stripe_index
;
5195 stripe_index
= find_live_mirror(fs_info
, map
,
5197 map
->sub_stripes
, stripe_index
+
5198 current
->pid
% map
->sub_stripes
,
5199 dev_replace_is_ongoing
);
5200 mirror_num
= stripe_index
- old_stripe_index
+ 1;
5203 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5204 if (need_raid_map
&&
5205 ((rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
)) ||
5207 /* push stripe_nr back to the start of the full stripe */
5208 stripe_nr
= div_u64(raid56_full_stripe_start
,
5209 stripe_len
* nr_data_stripes(map
));
5211 /* RAID[56] write or recovery. Return all stripes */
5212 num_stripes
= map
->num_stripes
;
5213 max_errors
= nr_parity_stripes(map
);
5215 *length
= map
->stripe_len
;
5220 * Mirror #0 or #1 means the original data block.
5221 * Mirror #2 is RAID5 parity block.
5222 * Mirror #3 is RAID6 Q block.
5224 stripe_nr
= div_u64_rem(stripe_nr
,
5225 nr_data_stripes(map
), &stripe_index
);
5227 stripe_index
= nr_data_stripes(map
) +
5230 /* We distribute the parity blocks across stripes */
5231 div_u64_rem(stripe_nr
+ stripe_index
, map
->num_stripes
,
5233 if (!(rw
& (REQ_WRITE
| REQ_DISCARD
|
5234 REQ_GET_READ_MIRRORS
)) && mirror_num
<= 1)
5239 * after this, stripe_nr is the number of stripes on this
5240 * device we have to walk to find the data, and stripe_index is
5241 * the number of our device in the stripe array
5243 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5245 mirror_num
= stripe_index
+ 1;
5247 BUG_ON(stripe_index
>= map
->num_stripes
);
5249 num_alloc_stripes
= num_stripes
;
5250 if (dev_replace_is_ongoing
) {
5251 if (rw
& (REQ_WRITE
| REQ_DISCARD
))
5252 num_alloc_stripes
<<= 1;
5253 if (rw
& REQ_GET_READ_MIRRORS
)
5254 num_alloc_stripes
++;
5255 tgtdev_indexes
= num_stripes
;
5258 bbio
= alloc_btrfs_bio(num_alloc_stripes
, tgtdev_indexes
);
5263 if (dev_replace_is_ongoing
)
5264 bbio
->tgtdev_map
= (int *)(bbio
->stripes
+ num_alloc_stripes
);
5266 /* build raid_map */
5267 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
&&
5268 need_raid_map
&& ((rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
)) ||
5273 bbio
->raid_map
= (u64
*)((void *)bbio
->stripes
+
5274 sizeof(struct btrfs_bio_stripe
) *
5276 sizeof(int) * tgtdev_indexes
);
5278 /* Work out the disk rotation on this stripe-set */
5279 div_u64_rem(stripe_nr
, num_stripes
, &rot
);
5281 /* Fill in the logical address of each stripe */
5282 tmp
= stripe_nr
* nr_data_stripes(map
);
5283 for (i
= 0; i
< nr_data_stripes(map
); i
++)
5284 bbio
->raid_map
[(i
+rot
) % num_stripes
] =
5285 em
->start
+ (tmp
+ i
) * map
->stripe_len
;
5287 bbio
->raid_map
[(i
+rot
) % map
->num_stripes
] = RAID5_P_STRIPE
;
5288 if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5289 bbio
->raid_map
[(i
+rot
+1) % num_stripes
] =
5293 if (rw
& REQ_DISCARD
) {
5295 u32 sub_stripes
= 0;
5296 u64 stripes_per_dev
= 0;
5297 u32 remaining_stripes
= 0;
5298 u32 last_stripe
= 0;
5301 (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID10
)) {
5302 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5305 sub_stripes
= map
->sub_stripes
;
5307 factor
= map
->num_stripes
/ sub_stripes
;
5308 stripes_per_dev
= div_u64_rem(stripe_nr_end
-
5311 &remaining_stripes
);
5312 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
5313 last_stripe
*= sub_stripes
;
5316 for (i
= 0; i
< num_stripes
; i
++) {
5317 bbio
->stripes
[i
].physical
=
5318 map
->stripes
[stripe_index
].physical
+
5319 stripe_offset
+ stripe_nr
* map
->stripe_len
;
5320 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
5322 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5323 BTRFS_BLOCK_GROUP_RAID10
)) {
5324 bbio
->stripes
[i
].length
= stripes_per_dev
*
5327 if (i
/ sub_stripes
< remaining_stripes
)
5328 bbio
->stripes
[i
].length
+=
5332 * Special for the first stripe and
5335 * |-------|...|-------|
5339 if (i
< sub_stripes
)
5340 bbio
->stripes
[i
].length
-=
5343 if (stripe_index
>= last_stripe
&&
5344 stripe_index
<= (last_stripe
+
5346 bbio
->stripes
[i
].length
-=
5349 if (i
== sub_stripes
- 1)
5352 bbio
->stripes
[i
].length
= *length
;
5355 if (stripe_index
== map
->num_stripes
) {
5356 /* This could only happen for RAID0/10 */
5362 for (i
= 0; i
< num_stripes
; i
++) {
5363 bbio
->stripes
[i
].physical
=
5364 map
->stripes
[stripe_index
].physical
+
5366 stripe_nr
* map
->stripe_len
;
5367 bbio
->stripes
[i
].dev
=
5368 map
->stripes
[stripe_index
].dev
;
5373 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
))
5374 max_errors
= btrfs_chunk_max_errors(map
);
5377 sort_parity_stripes(bbio
, num_stripes
);
5380 if (dev_replace_is_ongoing
&& (rw
& (REQ_WRITE
| REQ_DISCARD
)) &&
5381 dev_replace
->tgtdev
!= NULL
) {
5382 int index_where_to_add
;
5383 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5386 * duplicate the write operations while the dev replace
5387 * procedure is running. Since the copying of the old disk
5388 * to the new disk takes place at run time while the
5389 * filesystem is mounted writable, the regular write
5390 * operations to the old disk have to be duplicated to go
5391 * to the new disk as well.
5392 * Note that device->missing is handled by the caller, and
5393 * that the write to the old disk is already set up in the
5396 index_where_to_add
= num_stripes
;
5397 for (i
= 0; i
< num_stripes
; i
++) {
5398 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5399 /* write to new disk, too */
5400 struct btrfs_bio_stripe
*new =
5401 bbio
->stripes
+ index_where_to_add
;
5402 struct btrfs_bio_stripe
*old
=
5405 new->physical
= old
->physical
;
5406 new->length
= old
->length
;
5407 new->dev
= dev_replace
->tgtdev
;
5408 bbio
->tgtdev_map
[i
] = index_where_to_add
;
5409 index_where_to_add
++;
5414 num_stripes
= index_where_to_add
;
5415 } else if (dev_replace_is_ongoing
&& (rw
& REQ_GET_READ_MIRRORS
) &&
5416 dev_replace
->tgtdev
!= NULL
) {
5417 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5418 int index_srcdev
= 0;
5420 u64 physical_of_found
= 0;
5423 * During the dev-replace procedure, the target drive can
5424 * also be used to read data in case it is needed to repair
5425 * a corrupt block elsewhere. This is possible if the
5426 * requested area is left of the left cursor. In this area,
5427 * the target drive is a full copy of the source drive.
5429 for (i
= 0; i
< num_stripes
; i
++) {
5430 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5432 * In case of DUP, in order to keep it
5433 * simple, only add the mirror with the
5434 * lowest physical address
5437 physical_of_found
<=
5438 bbio
->stripes
[i
].physical
)
5442 physical_of_found
= bbio
->stripes
[i
].physical
;
5446 if (physical_of_found
+ map
->stripe_len
<=
5447 dev_replace
->cursor_left
) {
5448 struct btrfs_bio_stripe
*tgtdev_stripe
=
5449 bbio
->stripes
+ num_stripes
;
5451 tgtdev_stripe
->physical
= physical_of_found
;
5452 tgtdev_stripe
->length
=
5453 bbio
->stripes
[index_srcdev
].length
;
5454 tgtdev_stripe
->dev
= dev_replace
->tgtdev
;
5455 bbio
->tgtdev_map
[index_srcdev
] = num_stripes
;
5464 bbio
->map_type
= map
->type
;
5465 bbio
->num_stripes
= num_stripes
;
5466 bbio
->max_errors
= max_errors
;
5467 bbio
->mirror_num
= mirror_num
;
5468 bbio
->num_tgtdevs
= tgtdev_indexes
;
5471 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5472 * mirror_num == num_stripes + 1 && dev_replace target drive is
5473 * available as a mirror
5475 if (patch_the_first_stripe_for_dev_replace
&& num_stripes
> 0) {
5476 WARN_ON(num_stripes
> 1);
5477 bbio
->stripes
[0].dev
= dev_replace
->tgtdev
;
5478 bbio
->stripes
[0].physical
= physical_to_patch_in_first_stripe
;
5479 bbio
->mirror_num
= map
->num_stripes
+ 1;
5482 if (dev_replace_is_ongoing
)
5483 btrfs_dev_replace_unlock(dev_replace
);
5484 free_extent_map(em
);
5488 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
5489 u64 logical
, u64
*length
,
5490 struct btrfs_bio
**bbio_ret
, int mirror_num
)
5492 return __btrfs_map_block(fs_info
, rw
, logical
, length
, bbio_ret
,
5496 /* For Scrub/replace */
5497 int btrfs_map_sblock(struct btrfs_fs_info
*fs_info
, int rw
,
5498 u64 logical
, u64
*length
,
5499 struct btrfs_bio
**bbio_ret
, int mirror_num
,
5502 return __btrfs_map_block(fs_info
, rw
, logical
, length
, bbio_ret
,
5503 mirror_num
, need_raid_map
);
5506 int btrfs_rmap_block(struct btrfs_mapping_tree
*map_tree
,
5507 u64 chunk_start
, u64 physical
, u64 devid
,
5508 u64
**logical
, int *naddrs
, int *stripe_len
)
5510 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5511 struct extent_map
*em
;
5512 struct map_lookup
*map
;
5520 read_lock(&em_tree
->lock
);
5521 em
= lookup_extent_mapping(em_tree
, chunk_start
, 1);
5522 read_unlock(&em_tree
->lock
);
5525 printk(KERN_ERR
"BTRFS: couldn't find em for chunk %Lu\n",
5530 if (em
->start
!= chunk_start
) {
5531 printk(KERN_ERR
"BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5532 em
->start
, chunk_start
);
5533 free_extent_map(em
);
5536 map
= (struct map_lookup
*)em
->bdev
;
5539 rmap_len
= map
->stripe_len
;
5541 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5542 length
= div_u64(length
, map
->num_stripes
/ map
->sub_stripes
);
5543 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5544 length
= div_u64(length
, map
->num_stripes
);
5545 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5546 length
= div_u64(length
, nr_data_stripes(map
));
5547 rmap_len
= map
->stripe_len
* nr_data_stripes(map
);
5550 buf
= kcalloc(map
->num_stripes
, sizeof(u64
), GFP_NOFS
);
5551 BUG_ON(!buf
); /* -ENOMEM */
5553 for (i
= 0; i
< map
->num_stripes
; i
++) {
5554 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
5556 if (map
->stripes
[i
].physical
> physical
||
5557 map
->stripes
[i
].physical
+ length
<= physical
)
5560 stripe_nr
= physical
- map
->stripes
[i
].physical
;
5561 stripe_nr
= div_u64(stripe_nr
, map
->stripe_len
);
5563 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5564 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5565 stripe_nr
= div_u64(stripe_nr
, map
->sub_stripes
);
5566 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5567 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5568 } /* else if RAID[56], multiply by nr_data_stripes().
5569 * Alternatively, just use rmap_len below instead of
5570 * map->stripe_len */
5572 bytenr
= chunk_start
+ stripe_nr
* rmap_len
;
5573 WARN_ON(nr
>= map
->num_stripes
);
5574 for (j
= 0; j
< nr
; j
++) {
5575 if (buf
[j
] == bytenr
)
5579 WARN_ON(nr
>= map
->num_stripes
);
5586 *stripe_len
= rmap_len
;
5588 free_extent_map(em
);
5592 static inline void btrfs_end_bbio(struct btrfs_bio
*bbio
, struct bio
*bio
, int err
)
5594 if (likely(bbio
->flags
& BTRFS_BIO_ORIG_BIO_SUBMITTED
))
5595 bio_endio_nodec(bio
, err
);
5597 bio_endio(bio
, err
);
5598 btrfs_put_bbio(bbio
);
5601 static void btrfs_end_bio(struct bio
*bio
, int err
)
5603 struct btrfs_bio
*bbio
= bio
->bi_private
;
5604 struct btrfs_device
*dev
= bbio
->stripes
[0].dev
;
5605 int is_orig_bio
= 0;
5608 atomic_inc(&bbio
->error
);
5609 if (err
== -EIO
|| err
== -EREMOTEIO
) {
5610 unsigned int stripe_index
=
5611 btrfs_io_bio(bio
)->stripe_index
;
5613 BUG_ON(stripe_index
>= bbio
->num_stripes
);
5614 dev
= bbio
->stripes
[stripe_index
].dev
;
5616 if (bio
->bi_rw
& WRITE
)
5617 btrfs_dev_stat_inc(dev
,
5618 BTRFS_DEV_STAT_WRITE_ERRS
);
5620 btrfs_dev_stat_inc(dev
,
5621 BTRFS_DEV_STAT_READ_ERRS
);
5622 if ((bio
->bi_rw
& WRITE_FLUSH
) == WRITE_FLUSH
)
5623 btrfs_dev_stat_inc(dev
,
5624 BTRFS_DEV_STAT_FLUSH_ERRS
);
5625 btrfs_dev_stat_print_on_error(dev
);
5630 if (bio
== bbio
->orig_bio
)
5633 btrfs_bio_counter_dec(bbio
->fs_info
);
5635 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
5638 bio
= bbio
->orig_bio
;
5641 bio
->bi_private
= bbio
->private;
5642 bio
->bi_end_io
= bbio
->end_io
;
5643 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
5644 /* only send an error to the higher layers if it is
5645 * beyond the tolerance of the btrfs bio
5647 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
5651 * this bio is actually up to date, we didn't
5652 * go over the max number of errors
5654 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5658 btrfs_end_bbio(bbio
, bio
, err
);
5659 } else if (!is_orig_bio
) {
5665 * see run_scheduled_bios for a description of why bios are collected for
5668 * This will add one bio to the pending list for a device and make sure
5669 * the work struct is scheduled.
5671 static noinline
void btrfs_schedule_bio(struct btrfs_root
*root
,
5672 struct btrfs_device
*device
,
5673 int rw
, struct bio
*bio
)
5675 int should_queue
= 1;
5676 struct btrfs_pending_bios
*pending_bios
;
5678 if (device
->missing
|| !device
->bdev
) {
5679 bio_endio(bio
, -EIO
);
5683 /* don't bother with additional async steps for reads, right now */
5684 if (!(rw
& REQ_WRITE
)) {
5686 btrfsic_submit_bio(rw
, bio
);
5692 * nr_async_bios allows us to reliably return congestion to the
5693 * higher layers. Otherwise, the async bio makes it appear we have
5694 * made progress against dirty pages when we've really just put it
5695 * on a queue for later
5697 atomic_inc(&root
->fs_info
->nr_async_bios
);
5698 WARN_ON(bio
->bi_next
);
5699 bio
->bi_next
= NULL
;
5702 spin_lock(&device
->io_lock
);
5703 if (bio
->bi_rw
& REQ_SYNC
)
5704 pending_bios
= &device
->pending_sync_bios
;
5706 pending_bios
= &device
->pending_bios
;
5708 if (pending_bios
->tail
)
5709 pending_bios
->tail
->bi_next
= bio
;
5711 pending_bios
->tail
= bio
;
5712 if (!pending_bios
->head
)
5713 pending_bios
->head
= bio
;
5714 if (device
->running_pending
)
5717 spin_unlock(&device
->io_lock
);
5720 btrfs_queue_work(root
->fs_info
->submit_workers
,
5724 static int bio_size_ok(struct block_device
*bdev
, struct bio
*bio
,
5727 struct bio_vec
*prev
;
5728 struct request_queue
*q
= bdev_get_queue(bdev
);
5729 unsigned int max_sectors
= queue_max_sectors(q
);
5730 struct bvec_merge_data bvm
= {
5732 .bi_sector
= sector
,
5733 .bi_rw
= bio
->bi_rw
,
5736 if (WARN_ON(bio
->bi_vcnt
== 0))
5739 prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
5740 if (bio_sectors(bio
) > max_sectors
)
5743 if (!q
->merge_bvec_fn
)
5746 bvm
.bi_size
= bio
->bi_iter
.bi_size
- prev
->bv_len
;
5747 if (q
->merge_bvec_fn(q
, &bvm
, prev
) < prev
->bv_len
)
5752 static void submit_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
5753 struct bio
*bio
, u64 physical
, int dev_nr
,
5756 struct btrfs_device
*dev
= bbio
->stripes
[dev_nr
].dev
;
5758 bio
->bi_private
= bbio
;
5759 btrfs_io_bio(bio
)->stripe_index
= dev_nr
;
5760 bio
->bi_end_io
= btrfs_end_bio
;
5761 bio
->bi_iter
.bi_sector
= physical
>> 9;
5764 struct rcu_string
*name
;
5767 name
= rcu_dereference(dev
->name
);
5768 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5769 "(%s id %llu), size=%u\n", rw
,
5770 (u64
)bio
->bi_iter
.bi_sector
, (u_long
)dev
->bdev
->bd_dev
,
5771 name
->str
, dev
->devid
, bio
->bi_iter
.bi_size
);
5775 bio
->bi_bdev
= dev
->bdev
;
5777 btrfs_bio_counter_inc_noblocked(root
->fs_info
);
5780 btrfs_schedule_bio(root
, dev
, rw
, bio
);
5782 btrfsic_submit_bio(rw
, bio
);
5785 static int breakup_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
5786 struct bio
*first_bio
, struct btrfs_device
*dev
,
5787 int dev_nr
, int rw
, int async
)
5789 struct bio_vec
*bvec
= first_bio
->bi_io_vec
;
5791 int nr_vecs
= bio_get_nr_vecs(dev
->bdev
);
5792 u64 physical
= bbio
->stripes
[dev_nr
].physical
;
5795 bio
= btrfs_bio_alloc(dev
->bdev
, physical
>> 9, nr_vecs
, GFP_NOFS
);
5799 while (bvec
<= (first_bio
->bi_io_vec
+ first_bio
->bi_vcnt
- 1)) {
5800 if (bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
5801 bvec
->bv_offset
) < bvec
->bv_len
) {
5802 u64 len
= bio
->bi_iter
.bi_size
;
5804 atomic_inc(&bbio
->stripes_pending
);
5805 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
,
5813 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
, rw
, async
);
5817 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
5819 atomic_inc(&bbio
->error
);
5820 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
5821 /* Shoud be the original bio. */
5822 WARN_ON(bio
!= bbio
->orig_bio
);
5824 bio
->bi_private
= bbio
->private;
5825 bio
->bi_end_io
= bbio
->end_io
;
5826 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
5827 bio
->bi_iter
.bi_sector
= logical
>> 9;
5829 btrfs_end_bbio(bbio
, bio
, -EIO
);
5833 int btrfs_map_bio(struct btrfs_root
*root
, int rw
, struct bio
*bio
,
5834 int mirror_num
, int async_submit
)
5836 struct btrfs_device
*dev
;
5837 struct bio
*first_bio
= bio
;
5838 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
5844 struct btrfs_bio
*bbio
= NULL
;
5846 length
= bio
->bi_iter
.bi_size
;
5847 map_length
= length
;
5849 btrfs_bio_counter_inc_blocked(root
->fs_info
);
5850 ret
= __btrfs_map_block(root
->fs_info
, rw
, logical
, &map_length
, &bbio
,
5853 btrfs_bio_counter_dec(root
->fs_info
);
5857 total_devs
= bbio
->num_stripes
;
5858 bbio
->orig_bio
= first_bio
;
5859 bbio
->private = first_bio
->bi_private
;
5860 bbio
->end_io
= first_bio
->bi_end_io
;
5861 bbio
->fs_info
= root
->fs_info
;
5862 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
5864 if (bbio
->raid_map
) {
5865 /* In this case, map_length has been set to the length of
5866 a single stripe; not the whole write */
5868 ret
= raid56_parity_write(root
, bio
, bbio
, map_length
);
5870 ret
= raid56_parity_recover(root
, bio
, bbio
, map_length
,
5874 btrfs_bio_counter_dec(root
->fs_info
);
5878 if (map_length
< length
) {
5879 btrfs_crit(root
->fs_info
, "mapping failed logical %llu bio len %llu len %llu",
5880 logical
, length
, map_length
);
5884 for (dev_nr
= 0; dev_nr
< total_devs
; dev_nr
++) {
5885 dev
= bbio
->stripes
[dev_nr
].dev
;
5886 if (!dev
|| !dev
->bdev
|| (rw
& WRITE
&& !dev
->writeable
)) {
5887 bbio_error(bbio
, first_bio
, logical
);
5892 * Check and see if we're ok with this bio based on it's size
5893 * and offset with the given device.
5895 if (!bio_size_ok(dev
->bdev
, first_bio
,
5896 bbio
->stripes
[dev_nr
].physical
>> 9)) {
5897 ret
= breakup_stripe_bio(root
, bbio
, first_bio
, dev
,
5898 dev_nr
, rw
, async_submit
);
5903 if (dev_nr
< total_devs
- 1) {
5904 bio
= btrfs_bio_clone(first_bio
, GFP_NOFS
);
5905 BUG_ON(!bio
); /* -ENOMEM */
5908 bbio
->flags
|= BTRFS_BIO_ORIG_BIO_SUBMITTED
;
5911 submit_stripe_bio(root
, bbio
, bio
,
5912 bbio
->stripes
[dev_nr
].physical
, dev_nr
, rw
,
5915 btrfs_bio_counter_dec(root
->fs_info
);
5919 struct btrfs_device
*btrfs_find_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
5922 struct btrfs_device
*device
;
5923 struct btrfs_fs_devices
*cur_devices
;
5925 cur_devices
= fs_info
->fs_devices
;
5926 while (cur_devices
) {
5928 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
5929 device
= __find_device(&cur_devices
->devices
,
5934 cur_devices
= cur_devices
->seed
;
5939 static struct btrfs_device
*add_missing_dev(struct btrfs_root
*root
,
5940 struct btrfs_fs_devices
*fs_devices
,
5941 u64 devid
, u8
*dev_uuid
)
5943 struct btrfs_device
*device
;
5945 device
= btrfs_alloc_device(NULL
, &devid
, dev_uuid
);
5949 list_add(&device
->dev_list
, &fs_devices
->devices
);
5950 device
->fs_devices
= fs_devices
;
5951 fs_devices
->num_devices
++;
5953 device
->missing
= 1;
5954 fs_devices
->missing_devices
++;
5960 * btrfs_alloc_device - allocate struct btrfs_device
5961 * @fs_info: used only for generating a new devid, can be NULL if
5962 * devid is provided (i.e. @devid != NULL).
5963 * @devid: a pointer to devid for this device. If NULL a new devid
5965 * @uuid: a pointer to UUID for this device. If NULL a new UUID
5968 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5969 * on error. Returned struct is not linked onto any lists and can be
5970 * destroyed with kfree() right away.
5972 struct btrfs_device
*btrfs_alloc_device(struct btrfs_fs_info
*fs_info
,
5976 struct btrfs_device
*dev
;
5979 if (WARN_ON(!devid
&& !fs_info
))
5980 return ERR_PTR(-EINVAL
);
5982 dev
= __alloc_device();
5991 ret
= find_next_devid(fs_info
, &tmp
);
5994 return ERR_PTR(ret
);
6000 memcpy(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
);
6002 generate_random_uuid(dev
->uuid
);
6004 btrfs_init_work(&dev
->work
, btrfs_submit_helper
,
6005 pending_bios_fn
, NULL
, NULL
);
6010 static int read_one_chunk(struct btrfs_root
*root
, struct btrfs_key
*key
,
6011 struct extent_buffer
*leaf
,
6012 struct btrfs_chunk
*chunk
)
6014 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
6015 struct map_lookup
*map
;
6016 struct extent_map
*em
;
6020 u8 uuid
[BTRFS_UUID_SIZE
];
6025 logical
= key
->offset
;
6026 length
= btrfs_chunk_length(leaf
, chunk
);
6028 read_lock(&map_tree
->map_tree
.lock
);
6029 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
6030 read_unlock(&map_tree
->map_tree
.lock
);
6032 /* already mapped? */
6033 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
6034 free_extent_map(em
);
6037 free_extent_map(em
);
6040 em
= alloc_extent_map();
6043 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6044 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
6046 free_extent_map(em
);
6050 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
6051 em
->bdev
= (struct block_device
*)map
;
6052 em
->start
= logical
;
6055 em
->block_start
= 0;
6056 em
->block_len
= em
->len
;
6058 map
->num_stripes
= num_stripes
;
6059 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
6060 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
6061 map
->sector_size
= btrfs_chunk_sector_size(leaf
, chunk
);
6062 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6063 map
->type
= btrfs_chunk_type(leaf
, chunk
);
6064 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6065 for (i
= 0; i
< num_stripes
; i
++) {
6066 map
->stripes
[i
].physical
=
6067 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
6068 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
6069 read_extent_buffer(leaf
, uuid
, (unsigned long)
6070 btrfs_stripe_dev_uuid_nr(chunk
, i
),
6072 map
->stripes
[i
].dev
= btrfs_find_device(root
->fs_info
, devid
,
6074 if (!map
->stripes
[i
].dev
&& !btrfs_test_opt(root
, DEGRADED
)) {
6075 free_extent_map(em
);
6078 if (!map
->stripes
[i
].dev
) {
6079 map
->stripes
[i
].dev
=
6080 add_missing_dev(root
, root
->fs_info
->fs_devices
,
6082 if (!map
->stripes
[i
].dev
) {
6083 free_extent_map(em
);
6087 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
6090 write_lock(&map_tree
->map_tree
.lock
);
6091 ret
= add_extent_mapping(&map_tree
->map_tree
, em
, 0);
6092 write_unlock(&map_tree
->map_tree
.lock
);
6093 BUG_ON(ret
); /* Tree corruption */
6094 free_extent_map(em
);
6099 static void fill_device_from_item(struct extent_buffer
*leaf
,
6100 struct btrfs_dev_item
*dev_item
,
6101 struct btrfs_device
*device
)
6105 device
->devid
= btrfs_device_id(leaf
, dev_item
);
6106 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
6107 device
->total_bytes
= device
->disk_total_bytes
;
6108 device
->commit_total_bytes
= device
->disk_total_bytes
;
6109 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
6110 device
->commit_bytes_used
= device
->bytes_used
;
6111 device
->type
= btrfs_device_type(leaf
, dev_item
);
6112 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
6113 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
6114 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
6115 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
6116 device
->is_tgtdev_for_dev_replace
= 0;
6118 ptr
= btrfs_device_uuid(dev_item
);
6119 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
6122 static struct btrfs_fs_devices
*open_seed_devices(struct btrfs_root
*root
,
6125 struct btrfs_fs_devices
*fs_devices
;
6128 BUG_ON(!mutex_is_locked(&uuid_mutex
));
6130 fs_devices
= root
->fs_info
->fs_devices
->seed
;
6131 while (fs_devices
) {
6132 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
))
6135 fs_devices
= fs_devices
->seed
;
6138 fs_devices
= find_fsid(fsid
);
6140 if (!btrfs_test_opt(root
, DEGRADED
))
6141 return ERR_PTR(-ENOENT
);
6143 fs_devices
= alloc_fs_devices(fsid
);
6144 if (IS_ERR(fs_devices
))
6147 fs_devices
->seeding
= 1;
6148 fs_devices
->opened
= 1;
6152 fs_devices
= clone_fs_devices(fs_devices
);
6153 if (IS_ERR(fs_devices
))
6156 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
6157 root
->fs_info
->bdev_holder
);
6159 free_fs_devices(fs_devices
);
6160 fs_devices
= ERR_PTR(ret
);
6164 if (!fs_devices
->seeding
) {
6165 __btrfs_close_devices(fs_devices
);
6166 free_fs_devices(fs_devices
);
6167 fs_devices
= ERR_PTR(-EINVAL
);
6171 fs_devices
->seed
= root
->fs_info
->fs_devices
->seed
;
6172 root
->fs_info
->fs_devices
->seed
= fs_devices
;
6177 static int read_one_dev(struct btrfs_root
*root
,
6178 struct extent_buffer
*leaf
,
6179 struct btrfs_dev_item
*dev_item
)
6181 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
6182 struct btrfs_device
*device
;
6185 u8 fs_uuid
[BTRFS_UUID_SIZE
];
6186 u8 dev_uuid
[BTRFS_UUID_SIZE
];
6188 devid
= btrfs_device_id(leaf
, dev_item
);
6189 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
6191 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
6194 if (memcmp(fs_uuid
, root
->fs_info
->fsid
, BTRFS_UUID_SIZE
)) {
6195 fs_devices
= open_seed_devices(root
, fs_uuid
);
6196 if (IS_ERR(fs_devices
))
6197 return PTR_ERR(fs_devices
);
6200 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
, fs_uuid
);
6202 if (!btrfs_test_opt(root
, DEGRADED
))
6205 btrfs_warn(root
->fs_info
, "devid %llu missing", devid
);
6206 device
= add_missing_dev(root
, fs_devices
, devid
, dev_uuid
);
6210 if (!device
->bdev
&& !btrfs_test_opt(root
, DEGRADED
))
6213 if(!device
->bdev
&& !device
->missing
) {
6215 * this happens when a device that was properly setup
6216 * in the device info lists suddenly goes bad.
6217 * device->bdev is NULL, and so we have to set
6218 * device->missing to one here
6220 device
->fs_devices
->missing_devices
++;
6221 device
->missing
= 1;
6224 /* Move the device to its own fs_devices */
6225 if (device
->fs_devices
!= fs_devices
) {
6226 ASSERT(device
->missing
);
6228 list_move(&device
->dev_list
, &fs_devices
->devices
);
6229 device
->fs_devices
->num_devices
--;
6230 fs_devices
->num_devices
++;
6232 device
->fs_devices
->missing_devices
--;
6233 fs_devices
->missing_devices
++;
6235 device
->fs_devices
= fs_devices
;
6239 if (device
->fs_devices
!= root
->fs_info
->fs_devices
) {
6240 BUG_ON(device
->writeable
);
6241 if (device
->generation
!=
6242 btrfs_device_generation(leaf
, dev_item
))
6246 fill_device_from_item(leaf
, dev_item
, device
);
6247 device
->in_fs_metadata
= 1;
6248 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
6249 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
6250 spin_lock(&root
->fs_info
->free_chunk_lock
);
6251 root
->fs_info
->free_chunk_space
+= device
->total_bytes
-
6253 spin_unlock(&root
->fs_info
->free_chunk_lock
);
6259 int btrfs_read_sys_array(struct btrfs_root
*root
)
6261 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
6262 struct extent_buffer
*sb
;
6263 struct btrfs_disk_key
*disk_key
;
6264 struct btrfs_chunk
*chunk
;
6266 unsigned long sb_array_offset
;
6272 struct btrfs_key key
;
6274 ASSERT(BTRFS_SUPER_INFO_SIZE
<= root
->nodesize
);
6276 * This will create extent buffer of nodesize, superblock size is
6277 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6278 * overallocate but we can keep it as-is, only the first page is used.
6280 sb
= btrfs_find_create_tree_block(root
, BTRFS_SUPER_INFO_OFFSET
);
6283 btrfs_set_buffer_uptodate(sb
);
6284 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
6286 * The sb extent buffer is artifical and just used to read the system array.
6287 * btrfs_set_buffer_uptodate() call does not properly mark all it's
6288 * pages up-to-date when the page is larger: extent does not cover the
6289 * whole page and consequently check_page_uptodate does not find all
6290 * the page's extents up-to-date (the hole beyond sb),
6291 * write_extent_buffer then triggers a WARN_ON.
6293 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6294 * but sb spans only this function. Add an explicit SetPageUptodate call
6295 * to silence the warning eg. on PowerPC 64.
6297 if (PAGE_CACHE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
6298 SetPageUptodate(sb
->pages
[0]);
6300 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
6301 array_size
= btrfs_super_sys_array_size(super_copy
);
6303 array_ptr
= super_copy
->sys_chunk_array
;
6304 sb_array_offset
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
6307 while (cur_offset
< array_size
) {
6308 disk_key
= (struct btrfs_disk_key
*)array_ptr
;
6309 len
= sizeof(*disk_key
);
6310 if (cur_offset
+ len
> array_size
)
6311 goto out_short_read
;
6313 btrfs_disk_key_to_cpu(&key
, disk_key
);
6316 sb_array_offset
+= len
;
6319 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6320 chunk
= (struct btrfs_chunk
*)sb_array_offset
;
6322 * At least one btrfs_chunk with one stripe must be
6323 * present, exact stripe count check comes afterwards
6325 len
= btrfs_chunk_item_size(1);
6326 if (cur_offset
+ len
> array_size
)
6327 goto out_short_read
;
6329 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
6330 len
= btrfs_chunk_item_size(num_stripes
);
6331 if (cur_offset
+ len
> array_size
)
6332 goto out_short_read
;
6334 ret
= read_one_chunk(root
, &key
, sb
, chunk
);
6342 sb_array_offset
+= len
;
6345 free_extent_buffer(sb
);
6349 printk(KERN_ERR
"BTRFS: sys_array too short to read %u bytes at offset %u\n",
6351 free_extent_buffer(sb
);
6355 int btrfs_read_chunk_tree(struct btrfs_root
*root
)
6357 struct btrfs_path
*path
;
6358 struct extent_buffer
*leaf
;
6359 struct btrfs_key key
;
6360 struct btrfs_key found_key
;
6364 root
= root
->fs_info
->chunk_root
;
6366 path
= btrfs_alloc_path();
6370 mutex_lock(&uuid_mutex
);
6374 * Read all device items, and then all the chunk items. All
6375 * device items are found before any chunk item (their object id
6376 * is smaller than the lowest possible object id for a chunk
6377 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6379 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
6382 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6386 leaf
= path
->nodes
[0];
6387 slot
= path
->slots
[0];
6388 if (slot
>= btrfs_header_nritems(leaf
)) {
6389 ret
= btrfs_next_leaf(root
, path
);
6396 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6397 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
6398 struct btrfs_dev_item
*dev_item
;
6399 dev_item
= btrfs_item_ptr(leaf
, slot
,
6400 struct btrfs_dev_item
);
6401 ret
= read_one_dev(root
, leaf
, dev_item
);
6404 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6405 struct btrfs_chunk
*chunk
;
6406 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
6407 ret
= read_one_chunk(root
, &found_key
, leaf
, chunk
);
6415 unlock_chunks(root
);
6416 mutex_unlock(&uuid_mutex
);
6418 btrfs_free_path(path
);
6422 void btrfs_init_devices_late(struct btrfs_fs_info
*fs_info
)
6424 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6425 struct btrfs_device
*device
;
6427 while (fs_devices
) {
6428 mutex_lock(&fs_devices
->device_list_mutex
);
6429 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
)
6430 device
->dev_root
= fs_info
->dev_root
;
6431 mutex_unlock(&fs_devices
->device_list_mutex
);
6433 fs_devices
= fs_devices
->seed
;
6437 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
6441 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6442 btrfs_dev_stat_reset(dev
, i
);
6445 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
6447 struct btrfs_key key
;
6448 struct btrfs_key found_key
;
6449 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6450 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6451 struct extent_buffer
*eb
;
6454 struct btrfs_device
*device
;
6455 struct btrfs_path
*path
= NULL
;
6458 path
= btrfs_alloc_path();
6464 mutex_lock(&fs_devices
->device_list_mutex
);
6465 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6467 struct btrfs_dev_stats_item
*ptr
;
6470 key
.type
= BTRFS_DEV_STATS_KEY
;
6471 key
.offset
= device
->devid
;
6472 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
6474 __btrfs_reset_dev_stats(device
);
6475 device
->dev_stats_valid
= 1;
6476 btrfs_release_path(path
);
6479 slot
= path
->slots
[0];
6480 eb
= path
->nodes
[0];
6481 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
6482 item_size
= btrfs_item_size_nr(eb
, slot
);
6484 ptr
= btrfs_item_ptr(eb
, slot
,
6485 struct btrfs_dev_stats_item
);
6487 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6488 if (item_size
>= (1 + i
) * sizeof(__le64
))
6489 btrfs_dev_stat_set(device
, i
,
6490 btrfs_dev_stats_value(eb
, ptr
, i
));
6492 btrfs_dev_stat_reset(device
, i
);
6495 device
->dev_stats_valid
= 1;
6496 btrfs_dev_stat_print_on_load(device
);
6497 btrfs_release_path(path
);
6499 mutex_unlock(&fs_devices
->device_list_mutex
);
6502 btrfs_free_path(path
);
6503 return ret
< 0 ? ret
: 0;
6506 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
6507 struct btrfs_root
*dev_root
,
6508 struct btrfs_device
*device
)
6510 struct btrfs_path
*path
;
6511 struct btrfs_key key
;
6512 struct extent_buffer
*eb
;
6513 struct btrfs_dev_stats_item
*ptr
;
6518 key
.type
= BTRFS_DEV_STATS_KEY
;
6519 key
.offset
= device
->devid
;
6521 path
= btrfs_alloc_path();
6523 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
6525 printk_in_rcu(KERN_WARNING
"BTRFS: "
6526 "error %d while searching for dev_stats item for device %s!\n",
6527 ret
, rcu_str_deref(device
->name
));
6532 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
6533 /* need to delete old one and insert a new one */
6534 ret
= btrfs_del_item(trans
, dev_root
, path
);
6536 printk_in_rcu(KERN_WARNING
"BTRFS: "
6537 "delete too small dev_stats item for device %s failed %d!\n",
6538 rcu_str_deref(device
->name
), ret
);
6545 /* need to insert a new item */
6546 btrfs_release_path(path
);
6547 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
6548 &key
, sizeof(*ptr
));
6550 printk_in_rcu(KERN_WARNING
"BTRFS: "
6551 "insert dev_stats item for device %s failed %d!\n",
6552 rcu_str_deref(device
->name
), ret
);
6557 eb
= path
->nodes
[0];
6558 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
6559 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6560 btrfs_set_dev_stats_value(eb
, ptr
, i
,
6561 btrfs_dev_stat_read(device
, i
));
6562 btrfs_mark_buffer_dirty(eb
);
6565 btrfs_free_path(path
);
6570 * called from commit_transaction. Writes all changed device stats to disk.
6572 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
6573 struct btrfs_fs_info
*fs_info
)
6575 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6576 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6577 struct btrfs_device
*device
;
6581 mutex_lock(&fs_devices
->device_list_mutex
);
6582 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6583 if (!device
->dev_stats_valid
|| !btrfs_dev_stats_dirty(device
))
6586 stats_cnt
= atomic_read(&device
->dev_stats_ccnt
);
6587 ret
= update_dev_stat_item(trans
, dev_root
, device
);
6589 atomic_sub(stats_cnt
, &device
->dev_stats_ccnt
);
6591 mutex_unlock(&fs_devices
->device_list_mutex
);
6596 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
6598 btrfs_dev_stat_inc(dev
, index
);
6599 btrfs_dev_stat_print_on_error(dev
);
6602 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
6604 if (!dev
->dev_stats_valid
)
6606 printk_ratelimited_in_rcu(KERN_ERR
"BTRFS: "
6607 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6608 rcu_str_deref(dev
->name
),
6609 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
6610 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
6611 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
6612 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
6613 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
6616 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
6620 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6621 if (btrfs_dev_stat_read(dev
, i
) != 0)
6623 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
6624 return; /* all values == 0, suppress message */
6626 printk_in_rcu(KERN_INFO
"BTRFS: "
6627 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6628 rcu_str_deref(dev
->name
),
6629 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
6630 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
6631 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
6632 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
6633 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
6636 int btrfs_get_dev_stats(struct btrfs_root
*root
,
6637 struct btrfs_ioctl_get_dev_stats
*stats
)
6639 struct btrfs_device
*dev
;
6640 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
6643 mutex_lock(&fs_devices
->device_list_mutex
);
6644 dev
= btrfs_find_device(root
->fs_info
, stats
->devid
, NULL
, NULL
);
6645 mutex_unlock(&fs_devices
->device_list_mutex
);
6648 btrfs_warn(root
->fs_info
, "get dev_stats failed, device not found");
6650 } else if (!dev
->dev_stats_valid
) {
6651 btrfs_warn(root
->fs_info
, "get dev_stats failed, not yet valid");
6653 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
6654 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6655 if (stats
->nr_items
> i
)
6657 btrfs_dev_stat_read_and_reset(dev
, i
);
6659 btrfs_dev_stat_reset(dev
, i
);
6662 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6663 if (stats
->nr_items
> i
)
6664 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
6666 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
6667 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
6671 int btrfs_scratch_superblock(struct btrfs_device
*device
)
6673 struct buffer_head
*bh
;
6674 struct btrfs_super_block
*disk_super
;
6676 bh
= btrfs_read_dev_super(device
->bdev
);
6679 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
6681 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
6682 set_buffer_dirty(bh
);
6683 sync_dirty_buffer(bh
);
6690 * Update the size of all devices, which is used for writing out the
6693 void btrfs_update_commit_device_size(struct btrfs_fs_info
*fs_info
)
6695 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6696 struct btrfs_device
*curr
, *next
;
6698 if (list_empty(&fs_devices
->resized_devices
))
6701 mutex_lock(&fs_devices
->device_list_mutex
);
6702 lock_chunks(fs_info
->dev_root
);
6703 list_for_each_entry_safe(curr
, next
, &fs_devices
->resized_devices
,
6705 list_del_init(&curr
->resized_list
);
6706 curr
->commit_total_bytes
= curr
->disk_total_bytes
;
6708 unlock_chunks(fs_info
->dev_root
);
6709 mutex_unlock(&fs_devices
->device_list_mutex
);
6712 /* Must be invoked during the transaction commit */
6713 void btrfs_update_commit_device_bytes_used(struct btrfs_root
*root
,
6714 struct btrfs_transaction
*transaction
)
6716 struct extent_map
*em
;
6717 struct map_lookup
*map
;
6718 struct btrfs_device
*dev
;
6721 if (list_empty(&transaction
->pending_chunks
))
6724 /* In order to kick the device replace finish process */
6726 list_for_each_entry(em
, &transaction
->pending_chunks
, list
) {
6727 map
= (struct map_lookup
*)em
->bdev
;
6729 for (i
= 0; i
< map
->num_stripes
; i
++) {
6730 dev
= map
->stripes
[i
].dev
;
6731 dev
->commit_bytes_used
= dev
->bytes_used
;
6734 unlock_chunks(root
);