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>
30 #include "extent_map.h"
32 #include "transaction.h"
33 #include "print-tree.h"
35 #include "async-thread.h"
36 #include "check-integrity.h"
37 #include "rcu-string.h"
40 static int init_first_rw_device(struct btrfs_trans_handle
*trans
,
41 struct btrfs_root
*root
,
42 struct btrfs_device
*device
);
43 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
);
44 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
);
45 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*device
);
47 static DEFINE_MUTEX(uuid_mutex
);
48 static LIST_HEAD(fs_uuids
);
50 static void lock_chunks(struct btrfs_root
*root
)
52 mutex_lock(&root
->fs_info
->chunk_mutex
);
55 static void unlock_chunks(struct btrfs_root
*root
)
57 mutex_unlock(&root
->fs_info
->chunk_mutex
);
60 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
62 struct btrfs_device
*device
;
63 WARN_ON(fs_devices
->opened
);
64 while (!list_empty(&fs_devices
->devices
)) {
65 device
= list_entry(fs_devices
->devices
.next
,
66 struct btrfs_device
, dev_list
);
67 list_del(&device
->dev_list
);
68 rcu_string_free(device
->name
);
74 void btrfs_cleanup_fs_uuids(void)
76 struct btrfs_fs_devices
*fs_devices
;
78 while (!list_empty(&fs_uuids
)) {
79 fs_devices
= list_entry(fs_uuids
.next
,
80 struct btrfs_fs_devices
, list
);
81 list_del(&fs_devices
->list
);
82 free_fs_devices(fs_devices
);
86 static noinline
struct btrfs_device
*__find_device(struct list_head
*head
,
89 struct btrfs_device
*dev
;
91 list_for_each_entry(dev
, head
, dev_list
) {
92 if (dev
->devid
== devid
&&
93 (!uuid
|| !memcmp(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
))) {
100 static noinline
struct btrfs_fs_devices
*find_fsid(u8
*fsid
)
102 struct btrfs_fs_devices
*fs_devices
;
104 list_for_each_entry(fs_devices
, &fs_uuids
, list
) {
105 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
112 btrfs_get_bdev_and_sb(const char *device_path
, fmode_t flags
, void *holder
,
113 int flush
, struct block_device
**bdev
,
114 struct buffer_head
**bh
)
118 *bdev
= blkdev_get_by_path(device_path
, flags
, holder
);
121 ret
= PTR_ERR(*bdev
);
122 printk(KERN_INFO
"btrfs: open %s failed\n", device_path
);
127 filemap_write_and_wait((*bdev
)->bd_inode
->i_mapping
);
128 ret
= set_blocksize(*bdev
, 4096);
130 blkdev_put(*bdev
, flags
);
133 invalidate_bdev(*bdev
);
134 *bh
= btrfs_read_dev_super(*bdev
);
137 blkdev_put(*bdev
, flags
);
149 static void requeue_list(struct btrfs_pending_bios
*pending_bios
,
150 struct bio
*head
, struct bio
*tail
)
153 struct bio
*old_head
;
155 old_head
= pending_bios
->head
;
156 pending_bios
->head
= head
;
157 if (pending_bios
->tail
)
158 tail
->bi_next
= old_head
;
160 pending_bios
->tail
= tail
;
164 * we try to collect pending bios for a device so we don't get a large
165 * number of procs sending bios down to the same device. This greatly
166 * improves the schedulers ability to collect and merge the bios.
168 * But, it also turns into a long list of bios to process and that is sure
169 * to eventually make the worker thread block. The solution here is to
170 * make some progress and then put this work struct back at the end of
171 * the list if the block device is congested. This way, multiple devices
172 * can make progress from a single worker thread.
174 static noinline
void run_scheduled_bios(struct btrfs_device
*device
)
177 struct backing_dev_info
*bdi
;
178 struct btrfs_fs_info
*fs_info
;
179 struct btrfs_pending_bios
*pending_bios
;
183 unsigned long num_run
;
184 unsigned long batch_run
= 0;
186 unsigned long last_waited
= 0;
188 int sync_pending
= 0;
189 struct blk_plug plug
;
192 * this function runs all the bios we've collected for
193 * a particular device. We don't want to wander off to
194 * another device without first sending all of these down.
195 * So, setup a plug here and finish it off before we return
197 blk_start_plug(&plug
);
199 bdi
= blk_get_backing_dev_info(device
->bdev
);
200 fs_info
= device
->dev_root
->fs_info
;
201 limit
= btrfs_async_submit_limit(fs_info
);
202 limit
= limit
* 2 / 3;
205 spin_lock(&device
->io_lock
);
210 /* take all the bios off the list at once and process them
211 * later on (without the lock held). But, remember the
212 * tail and other pointers so the bios can be properly reinserted
213 * into the list if we hit congestion
215 if (!force_reg
&& device
->pending_sync_bios
.head
) {
216 pending_bios
= &device
->pending_sync_bios
;
219 pending_bios
= &device
->pending_bios
;
223 pending
= pending_bios
->head
;
224 tail
= pending_bios
->tail
;
225 WARN_ON(pending
&& !tail
);
228 * if pending was null this time around, no bios need processing
229 * at all and we can stop. Otherwise it'll loop back up again
230 * and do an additional check so no bios are missed.
232 * device->running_pending is used to synchronize with the
235 if (device
->pending_sync_bios
.head
== NULL
&&
236 device
->pending_bios
.head
== NULL
) {
238 device
->running_pending
= 0;
241 device
->running_pending
= 1;
244 pending_bios
->head
= NULL
;
245 pending_bios
->tail
= NULL
;
247 spin_unlock(&device
->io_lock
);
252 /* we want to work on both lists, but do more bios on the
253 * sync list than the regular list
256 pending_bios
!= &device
->pending_sync_bios
&&
257 device
->pending_sync_bios
.head
) ||
258 (num_run
> 64 && pending_bios
== &device
->pending_sync_bios
&&
259 device
->pending_bios
.head
)) {
260 spin_lock(&device
->io_lock
);
261 requeue_list(pending_bios
, pending
, tail
);
266 pending
= pending
->bi_next
;
269 if (atomic_dec_return(&fs_info
->nr_async_bios
) < limit
&&
270 waitqueue_active(&fs_info
->async_submit_wait
))
271 wake_up(&fs_info
->async_submit_wait
);
273 BUG_ON(atomic_read(&cur
->bi_cnt
) == 0);
276 * if we're doing the sync list, record that our
277 * plug has some sync requests on it
279 * If we're doing the regular list and there are
280 * sync requests sitting around, unplug before
283 if (pending_bios
== &device
->pending_sync_bios
) {
285 } else if (sync_pending
) {
286 blk_finish_plug(&plug
);
287 blk_start_plug(&plug
);
291 btrfsic_submit_bio(cur
->bi_rw
, cur
);
298 * we made progress, there is more work to do and the bdi
299 * is now congested. Back off and let other work structs
302 if (pending
&& bdi_write_congested(bdi
) && batch_run
> 8 &&
303 fs_info
->fs_devices
->open_devices
> 1) {
304 struct io_context
*ioc
;
306 ioc
= current
->io_context
;
309 * the main goal here is that we don't want to
310 * block if we're going to be able to submit
311 * more requests without blocking.
313 * This code does two great things, it pokes into
314 * the elevator code from a filesystem _and_
315 * it makes assumptions about how batching works.
317 if (ioc
&& ioc
->nr_batch_requests
> 0 &&
318 time_before(jiffies
, ioc
->last_waited
+ HZ
/50UL) &&
320 ioc
->last_waited
== last_waited
)) {
322 * we want to go through our batch of
323 * requests and stop. So, we copy out
324 * the ioc->last_waited time and test
325 * against it before looping
327 last_waited
= ioc
->last_waited
;
332 spin_lock(&device
->io_lock
);
333 requeue_list(pending_bios
, pending
, tail
);
334 device
->running_pending
= 1;
336 spin_unlock(&device
->io_lock
);
337 btrfs_requeue_work(&device
->work
);
340 /* unplug every 64 requests just for good measure */
341 if (batch_run
% 64 == 0) {
342 blk_finish_plug(&plug
);
343 blk_start_plug(&plug
);
352 spin_lock(&device
->io_lock
);
353 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
355 spin_unlock(&device
->io_lock
);
358 blk_finish_plug(&plug
);
361 static void pending_bios_fn(struct btrfs_work
*work
)
363 struct btrfs_device
*device
;
365 device
= container_of(work
, struct btrfs_device
, work
);
366 run_scheduled_bios(device
);
369 static noinline
int device_list_add(const char *path
,
370 struct btrfs_super_block
*disk_super
,
371 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
373 struct btrfs_device
*device
;
374 struct btrfs_fs_devices
*fs_devices
;
375 struct rcu_string
*name
;
376 u64 found_transid
= btrfs_super_generation(disk_super
);
378 fs_devices
= find_fsid(disk_super
->fsid
);
380 fs_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
383 INIT_LIST_HEAD(&fs_devices
->devices
);
384 INIT_LIST_HEAD(&fs_devices
->alloc_list
);
385 list_add(&fs_devices
->list
, &fs_uuids
);
386 memcpy(fs_devices
->fsid
, disk_super
->fsid
, BTRFS_FSID_SIZE
);
387 fs_devices
->latest_devid
= devid
;
388 fs_devices
->latest_trans
= found_transid
;
389 mutex_init(&fs_devices
->device_list_mutex
);
392 device
= __find_device(&fs_devices
->devices
, devid
,
393 disk_super
->dev_item
.uuid
);
396 if (fs_devices
->opened
)
399 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
401 /* we can safely leave the fs_devices entry around */
404 device
->devid
= devid
;
405 device
->dev_stats_valid
= 0;
406 device
->work
.func
= pending_bios_fn
;
407 memcpy(device
->uuid
, disk_super
->dev_item
.uuid
,
409 spin_lock_init(&device
->io_lock
);
411 name
= rcu_string_strdup(path
, GFP_NOFS
);
416 rcu_assign_pointer(device
->name
, name
);
417 INIT_LIST_HEAD(&device
->dev_alloc_list
);
419 /* init readahead state */
420 spin_lock_init(&device
->reada_lock
);
421 device
->reada_curr_zone
= NULL
;
422 atomic_set(&device
->reada_in_flight
, 0);
423 device
->reada_next
= 0;
424 INIT_RADIX_TREE(&device
->reada_zones
, GFP_NOFS
& ~__GFP_WAIT
);
425 INIT_RADIX_TREE(&device
->reada_extents
, GFP_NOFS
& ~__GFP_WAIT
);
427 mutex_lock(&fs_devices
->device_list_mutex
);
428 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
429 mutex_unlock(&fs_devices
->device_list_mutex
);
431 device
->fs_devices
= fs_devices
;
432 fs_devices
->num_devices
++;
433 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
434 name
= rcu_string_strdup(path
, GFP_NOFS
);
437 rcu_string_free(device
->name
);
438 rcu_assign_pointer(device
->name
, name
);
439 if (device
->missing
) {
440 fs_devices
->missing_devices
--;
445 if (found_transid
> fs_devices
->latest_trans
) {
446 fs_devices
->latest_devid
= devid
;
447 fs_devices
->latest_trans
= found_transid
;
449 *fs_devices_ret
= fs_devices
;
453 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
455 struct btrfs_fs_devices
*fs_devices
;
456 struct btrfs_device
*device
;
457 struct btrfs_device
*orig_dev
;
459 fs_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
461 return ERR_PTR(-ENOMEM
);
463 INIT_LIST_HEAD(&fs_devices
->devices
);
464 INIT_LIST_HEAD(&fs_devices
->alloc_list
);
465 INIT_LIST_HEAD(&fs_devices
->list
);
466 mutex_init(&fs_devices
->device_list_mutex
);
467 fs_devices
->latest_devid
= orig
->latest_devid
;
468 fs_devices
->latest_trans
= orig
->latest_trans
;
469 fs_devices
->total_devices
= orig
->total_devices
;
470 memcpy(fs_devices
->fsid
, orig
->fsid
, sizeof(fs_devices
->fsid
));
472 /* We have held the volume lock, it is safe to get the devices. */
473 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
474 struct rcu_string
*name
;
476 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
481 * This is ok to do without rcu read locked because we hold the
482 * uuid mutex so nothing we touch in here is going to disappear.
484 name
= rcu_string_strdup(orig_dev
->name
->str
, GFP_NOFS
);
489 rcu_assign_pointer(device
->name
, name
);
491 device
->devid
= orig_dev
->devid
;
492 device
->work
.func
= pending_bios_fn
;
493 memcpy(device
->uuid
, orig_dev
->uuid
, sizeof(device
->uuid
));
494 spin_lock_init(&device
->io_lock
);
495 INIT_LIST_HEAD(&device
->dev_list
);
496 INIT_LIST_HEAD(&device
->dev_alloc_list
);
498 list_add(&device
->dev_list
, &fs_devices
->devices
);
499 device
->fs_devices
= fs_devices
;
500 fs_devices
->num_devices
++;
504 free_fs_devices(fs_devices
);
505 return ERR_PTR(-ENOMEM
);
508 void btrfs_close_extra_devices(struct btrfs_fs_devices
*fs_devices
)
510 struct btrfs_device
*device
, *next
;
512 struct block_device
*latest_bdev
= NULL
;
513 u64 latest_devid
= 0;
514 u64 latest_transid
= 0;
516 mutex_lock(&uuid_mutex
);
518 /* This is the initialized path, it is safe to release the devices. */
519 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
520 if (device
->in_fs_metadata
) {
521 if (!latest_transid
||
522 device
->generation
> latest_transid
) {
523 latest_devid
= device
->devid
;
524 latest_transid
= device
->generation
;
525 latest_bdev
= device
->bdev
;
531 blkdev_put(device
->bdev
, device
->mode
);
533 fs_devices
->open_devices
--;
535 if (device
->writeable
) {
536 list_del_init(&device
->dev_alloc_list
);
537 device
->writeable
= 0;
538 fs_devices
->rw_devices
--;
540 list_del_init(&device
->dev_list
);
541 fs_devices
->num_devices
--;
542 rcu_string_free(device
->name
);
546 if (fs_devices
->seed
) {
547 fs_devices
= fs_devices
->seed
;
551 fs_devices
->latest_bdev
= latest_bdev
;
552 fs_devices
->latest_devid
= latest_devid
;
553 fs_devices
->latest_trans
= latest_transid
;
555 mutex_unlock(&uuid_mutex
);
558 static void __free_device(struct work_struct
*work
)
560 struct btrfs_device
*device
;
562 device
= container_of(work
, struct btrfs_device
, rcu_work
);
565 blkdev_put(device
->bdev
, device
->mode
);
567 rcu_string_free(device
->name
);
571 static void free_device(struct rcu_head
*head
)
573 struct btrfs_device
*device
;
575 device
= container_of(head
, struct btrfs_device
, rcu
);
577 INIT_WORK(&device
->rcu_work
, __free_device
);
578 schedule_work(&device
->rcu_work
);
581 static int __btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
583 struct btrfs_device
*device
;
585 if (--fs_devices
->opened
> 0)
588 mutex_lock(&fs_devices
->device_list_mutex
);
589 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
590 struct btrfs_device
*new_device
;
591 struct rcu_string
*name
;
594 fs_devices
->open_devices
--;
596 if (device
->writeable
) {
597 list_del_init(&device
->dev_alloc_list
);
598 fs_devices
->rw_devices
--;
601 if (device
->can_discard
)
602 fs_devices
->num_can_discard
--;
604 new_device
= kmalloc(sizeof(*new_device
), GFP_NOFS
);
605 BUG_ON(!new_device
); /* -ENOMEM */
606 memcpy(new_device
, device
, sizeof(*new_device
));
608 /* Safe because we are under uuid_mutex */
610 name
= rcu_string_strdup(device
->name
->str
, GFP_NOFS
);
611 BUG_ON(device
->name
&& !name
); /* -ENOMEM */
612 rcu_assign_pointer(new_device
->name
, name
);
614 new_device
->bdev
= NULL
;
615 new_device
->writeable
= 0;
616 new_device
->in_fs_metadata
= 0;
617 new_device
->can_discard
= 0;
618 list_replace_rcu(&device
->dev_list
, &new_device
->dev_list
);
620 call_rcu(&device
->rcu
, free_device
);
622 mutex_unlock(&fs_devices
->device_list_mutex
);
624 WARN_ON(fs_devices
->open_devices
);
625 WARN_ON(fs_devices
->rw_devices
);
626 fs_devices
->opened
= 0;
627 fs_devices
->seeding
= 0;
632 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
634 struct btrfs_fs_devices
*seed_devices
= NULL
;
637 mutex_lock(&uuid_mutex
);
638 ret
= __btrfs_close_devices(fs_devices
);
639 if (!fs_devices
->opened
) {
640 seed_devices
= fs_devices
->seed
;
641 fs_devices
->seed
= NULL
;
643 mutex_unlock(&uuid_mutex
);
645 while (seed_devices
) {
646 fs_devices
= seed_devices
;
647 seed_devices
= fs_devices
->seed
;
648 __btrfs_close_devices(fs_devices
);
649 free_fs_devices(fs_devices
);
654 static int __btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
655 fmode_t flags
, void *holder
)
657 struct request_queue
*q
;
658 struct block_device
*bdev
;
659 struct list_head
*head
= &fs_devices
->devices
;
660 struct btrfs_device
*device
;
661 struct block_device
*latest_bdev
= NULL
;
662 struct buffer_head
*bh
;
663 struct btrfs_super_block
*disk_super
;
664 u64 latest_devid
= 0;
665 u64 latest_transid
= 0;
672 list_for_each_entry(device
, head
, dev_list
) {
678 ret
= btrfs_get_bdev_and_sb(device
->name
->str
, flags
, holder
, 1,
683 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
684 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
685 if (devid
!= device
->devid
)
688 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
,
692 device
->generation
= btrfs_super_generation(disk_super
);
693 if (!latest_transid
|| device
->generation
> latest_transid
) {
694 latest_devid
= devid
;
695 latest_transid
= device
->generation
;
699 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
700 device
->writeable
= 0;
702 device
->writeable
= !bdev_read_only(bdev
);
706 q
= bdev_get_queue(bdev
);
707 if (blk_queue_discard(q
)) {
708 device
->can_discard
= 1;
709 fs_devices
->num_can_discard
++;
713 device
->in_fs_metadata
= 0;
714 device
->mode
= flags
;
716 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
717 fs_devices
->rotating
= 1;
719 fs_devices
->open_devices
++;
720 if (device
->writeable
) {
721 fs_devices
->rw_devices
++;
722 list_add(&device
->dev_alloc_list
,
723 &fs_devices
->alloc_list
);
730 blkdev_put(bdev
, flags
);
733 if (fs_devices
->open_devices
== 0) {
737 fs_devices
->seeding
= seeding
;
738 fs_devices
->opened
= 1;
739 fs_devices
->latest_bdev
= latest_bdev
;
740 fs_devices
->latest_devid
= latest_devid
;
741 fs_devices
->latest_trans
= latest_transid
;
742 fs_devices
->total_rw_bytes
= 0;
747 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
748 fmode_t flags
, void *holder
)
752 mutex_lock(&uuid_mutex
);
753 if (fs_devices
->opened
) {
754 fs_devices
->opened
++;
757 ret
= __btrfs_open_devices(fs_devices
, flags
, holder
);
759 mutex_unlock(&uuid_mutex
);
763 int btrfs_scan_one_device(const char *path
, fmode_t flags
, void *holder
,
764 struct btrfs_fs_devices
**fs_devices_ret
)
766 struct btrfs_super_block
*disk_super
;
767 struct block_device
*bdev
;
768 struct buffer_head
*bh
;
775 mutex_lock(&uuid_mutex
);
776 ret
= btrfs_get_bdev_and_sb(path
, flags
, holder
, 0, &bdev
, &bh
);
779 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
780 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
781 transid
= btrfs_super_generation(disk_super
);
782 total_devices
= btrfs_super_num_devices(disk_super
);
783 if (disk_super
->label
[0]) {
784 if (disk_super
->label
[BTRFS_LABEL_SIZE
- 1])
785 disk_super
->label
[BTRFS_LABEL_SIZE
- 1] = '\0';
786 printk(KERN_INFO
"device label %s ", disk_super
->label
);
788 printk(KERN_INFO
"device fsid %pU ", disk_super
->fsid
);
790 printk(KERN_CONT
"devid %llu transid %llu %s\n",
791 (unsigned long long)devid
, (unsigned long long)transid
, path
);
792 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
793 if (!ret
&& fs_devices_ret
)
794 (*fs_devices_ret
)->total_devices
= total_devices
;
796 blkdev_put(bdev
, flags
);
798 mutex_unlock(&uuid_mutex
);
802 /* helper to account the used device space in the range */
803 int btrfs_account_dev_extents_size(struct btrfs_device
*device
, u64 start
,
804 u64 end
, u64
*length
)
806 struct btrfs_key key
;
807 struct btrfs_root
*root
= device
->dev_root
;
808 struct btrfs_dev_extent
*dev_extent
;
809 struct btrfs_path
*path
;
813 struct extent_buffer
*l
;
817 if (start
>= device
->total_bytes
)
820 path
= btrfs_alloc_path();
825 key
.objectid
= device
->devid
;
827 key
.type
= BTRFS_DEV_EXTENT_KEY
;
829 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
833 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
840 slot
= path
->slots
[0];
841 if (slot
>= btrfs_header_nritems(l
)) {
842 ret
= btrfs_next_leaf(root
, path
);
850 btrfs_item_key_to_cpu(l
, &key
, slot
);
852 if (key
.objectid
< device
->devid
)
855 if (key
.objectid
> device
->devid
)
858 if (btrfs_key_type(&key
) != BTRFS_DEV_EXTENT_KEY
)
861 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
862 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
864 if (key
.offset
<= start
&& extent_end
> end
) {
865 *length
= end
- start
+ 1;
867 } else if (key
.offset
<= start
&& extent_end
> start
)
868 *length
+= extent_end
- start
;
869 else if (key
.offset
> start
&& extent_end
<= end
)
870 *length
+= extent_end
- key
.offset
;
871 else if (key
.offset
> start
&& key
.offset
<= end
) {
872 *length
+= end
- key
.offset
+ 1;
874 } else if (key
.offset
> end
)
882 btrfs_free_path(path
);
887 * find_free_dev_extent - find free space in the specified device
888 * @device: the device which we search the free space in
889 * @num_bytes: the size of the free space that we need
890 * @start: store the start of the free space.
891 * @len: the size of the free space. that we find, or the size of the max
892 * free space if we don't find suitable free space
894 * this uses a pretty simple search, the expectation is that it is
895 * called very infrequently and that a given device has a small number
898 * @start is used to store the start of the free space if we find. But if we
899 * don't find suitable free space, it will be used to store the start position
900 * of the max free space.
902 * @len is used to store the size of the free space that we find.
903 * But if we don't find suitable free space, it is used to store the size of
904 * the max free space.
906 int find_free_dev_extent(struct btrfs_device
*device
, u64 num_bytes
,
907 u64
*start
, u64
*len
)
909 struct btrfs_key key
;
910 struct btrfs_root
*root
= device
->dev_root
;
911 struct btrfs_dev_extent
*dev_extent
;
912 struct btrfs_path
*path
;
918 u64 search_end
= device
->total_bytes
;
921 struct extent_buffer
*l
;
923 /* FIXME use last free of some kind */
925 /* we don't want to overwrite the superblock on the drive,
926 * so we make sure to start at an offset of at least 1MB
928 search_start
= max(root
->fs_info
->alloc_start
, 1024ull * 1024);
930 max_hole_start
= search_start
;
934 if (search_start
>= search_end
) {
939 path
= btrfs_alloc_path();
946 key
.objectid
= device
->devid
;
947 key
.offset
= search_start
;
948 key
.type
= BTRFS_DEV_EXTENT_KEY
;
950 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
954 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
961 slot
= path
->slots
[0];
962 if (slot
>= btrfs_header_nritems(l
)) {
963 ret
= btrfs_next_leaf(root
, path
);
971 btrfs_item_key_to_cpu(l
, &key
, slot
);
973 if (key
.objectid
< device
->devid
)
976 if (key
.objectid
> device
->devid
)
979 if (btrfs_key_type(&key
) != BTRFS_DEV_EXTENT_KEY
)
982 if (key
.offset
> search_start
) {
983 hole_size
= key
.offset
- search_start
;
985 if (hole_size
> max_hole_size
) {
986 max_hole_start
= search_start
;
987 max_hole_size
= hole_size
;
991 * If this free space is greater than which we need,
992 * it must be the max free space that we have found
993 * until now, so max_hole_start must point to the start
994 * of this free space and the length of this free space
995 * is stored in max_hole_size. Thus, we return
996 * max_hole_start and max_hole_size and go back to the
999 if (hole_size
>= num_bytes
) {
1005 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1006 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1008 if (extent_end
> search_start
)
1009 search_start
= extent_end
;
1016 * At this point, search_start should be the end of
1017 * allocated dev extents, and when shrinking the device,
1018 * search_end may be smaller than search_start.
1020 if (search_end
> search_start
)
1021 hole_size
= search_end
- search_start
;
1023 if (hole_size
> max_hole_size
) {
1024 max_hole_start
= search_start
;
1025 max_hole_size
= hole_size
;
1029 if (hole_size
< num_bytes
)
1035 btrfs_free_path(path
);
1037 *start
= max_hole_start
;
1039 *len
= max_hole_size
;
1043 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1044 struct btrfs_device
*device
,
1048 struct btrfs_path
*path
;
1049 struct btrfs_root
*root
= device
->dev_root
;
1050 struct btrfs_key key
;
1051 struct btrfs_key found_key
;
1052 struct extent_buffer
*leaf
= NULL
;
1053 struct btrfs_dev_extent
*extent
= NULL
;
1055 path
= btrfs_alloc_path();
1059 key
.objectid
= device
->devid
;
1061 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1063 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1065 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1066 BTRFS_DEV_EXTENT_KEY
);
1069 leaf
= path
->nodes
[0];
1070 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1071 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1072 struct btrfs_dev_extent
);
1073 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1074 btrfs_dev_extent_length(leaf
, extent
) < start
);
1076 btrfs_release_path(path
);
1078 } else if (ret
== 0) {
1079 leaf
= path
->nodes
[0];
1080 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1081 struct btrfs_dev_extent
);
1083 btrfs_error(root
->fs_info
, ret
, "Slot search failed");
1087 if (device
->bytes_used
> 0) {
1088 u64 len
= btrfs_dev_extent_length(leaf
, extent
);
1089 device
->bytes_used
-= len
;
1090 spin_lock(&root
->fs_info
->free_chunk_lock
);
1091 root
->fs_info
->free_chunk_space
+= len
;
1092 spin_unlock(&root
->fs_info
->free_chunk_lock
);
1094 ret
= btrfs_del_item(trans
, root
, path
);
1096 btrfs_error(root
->fs_info
, ret
,
1097 "Failed to remove dev extent item");
1100 btrfs_free_path(path
);
1104 int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1105 struct btrfs_device
*device
,
1106 u64 chunk_tree
, u64 chunk_objectid
,
1107 u64 chunk_offset
, u64 start
, u64 num_bytes
)
1110 struct btrfs_path
*path
;
1111 struct btrfs_root
*root
= device
->dev_root
;
1112 struct btrfs_dev_extent
*extent
;
1113 struct extent_buffer
*leaf
;
1114 struct btrfs_key key
;
1116 WARN_ON(!device
->in_fs_metadata
);
1117 path
= btrfs_alloc_path();
1121 key
.objectid
= device
->devid
;
1123 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1124 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1129 leaf
= path
->nodes
[0];
1130 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1131 struct btrfs_dev_extent
);
1132 btrfs_set_dev_extent_chunk_tree(leaf
, extent
, chunk_tree
);
1133 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
, chunk_objectid
);
1134 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1136 write_extent_buffer(leaf
, root
->fs_info
->chunk_tree_uuid
,
1137 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent
),
1140 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1141 btrfs_mark_buffer_dirty(leaf
);
1143 btrfs_free_path(path
);
1147 static noinline
int find_next_chunk(struct btrfs_root
*root
,
1148 u64 objectid
, u64
*offset
)
1150 struct btrfs_path
*path
;
1152 struct btrfs_key key
;
1153 struct btrfs_chunk
*chunk
;
1154 struct btrfs_key found_key
;
1156 path
= btrfs_alloc_path();
1160 key
.objectid
= objectid
;
1161 key
.offset
= (u64
)-1;
1162 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
1164 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1168 BUG_ON(ret
== 0); /* Corruption */
1170 ret
= btrfs_previous_item(root
, path
, 0, BTRFS_CHUNK_ITEM_KEY
);
1174 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1176 if (found_key
.objectid
!= objectid
)
1179 chunk
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1180 struct btrfs_chunk
);
1181 *offset
= found_key
.offset
+
1182 btrfs_chunk_length(path
->nodes
[0], chunk
);
1187 btrfs_free_path(path
);
1191 static noinline
int find_next_devid(struct btrfs_root
*root
, u64
*objectid
)
1194 struct btrfs_key key
;
1195 struct btrfs_key found_key
;
1196 struct btrfs_path
*path
;
1198 root
= root
->fs_info
->chunk_root
;
1200 path
= btrfs_alloc_path();
1204 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1205 key
.type
= BTRFS_DEV_ITEM_KEY
;
1206 key
.offset
= (u64
)-1;
1208 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1212 BUG_ON(ret
== 0); /* Corruption */
1214 ret
= btrfs_previous_item(root
, path
, BTRFS_DEV_ITEMS_OBJECTID
,
1215 BTRFS_DEV_ITEM_KEY
);
1219 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1221 *objectid
= found_key
.offset
+ 1;
1225 btrfs_free_path(path
);
1230 * the device information is stored in the chunk root
1231 * the btrfs_device struct should be fully filled in
1233 int btrfs_add_device(struct btrfs_trans_handle
*trans
,
1234 struct btrfs_root
*root
,
1235 struct btrfs_device
*device
)
1238 struct btrfs_path
*path
;
1239 struct btrfs_dev_item
*dev_item
;
1240 struct extent_buffer
*leaf
;
1241 struct btrfs_key key
;
1244 root
= root
->fs_info
->chunk_root
;
1246 path
= btrfs_alloc_path();
1250 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1251 key
.type
= BTRFS_DEV_ITEM_KEY
;
1252 key
.offset
= device
->devid
;
1254 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1259 leaf
= path
->nodes
[0];
1260 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1262 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1263 btrfs_set_device_generation(leaf
, dev_item
, 0);
1264 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1265 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1266 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1267 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1268 btrfs_set_device_total_bytes(leaf
, dev_item
, device
->total_bytes
);
1269 btrfs_set_device_bytes_used(leaf
, dev_item
, device
->bytes_used
);
1270 btrfs_set_device_group(leaf
, dev_item
, 0);
1271 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1272 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1273 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1275 ptr
= (unsigned long)btrfs_device_uuid(dev_item
);
1276 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1277 ptr
= (unsigned long)btrfs_device_fsid(dev_item
);
1278 write_extent_buffer(leaf
, root
->fs_info
->fsid
, ptr
, BTRFS_UUID_SIZE
);
1279 btrfs_mark_buffer_dirty(leaf
);
1283 btrfs_free_path(path
);
1287 static int btrfs_rm_dev_item(struct btrfs_root
*root
,
1288 struct btrfs_device
*device
)
1291 struct btrfs_path
*path
;
1292 struct btrfs_key key
;
1293 struct btrfs_trans_handle
*trans
;
1295 root
= root
->fs_info
->chunk_root
;
1297 path
= btrfs_alloc_path();
1301 trans
= btrfs_start_transaction(root
, 0);
1302 if (IS_ERR(trans
)) {
1303 btrfs_free_path(path
);
1304 return PTR_ERR(trans
);
1306 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1307 key
.type
= BTRFS_DEV_ITEM_KEY
;
1308 key
.offset
= device
->devid
;
1311 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1320 ret
= btrfs_del_item(trans
, root
, path
);
1324 btrfs_free_path(path
);
1325 unlock_chunks(root
);
1326 btrfs_commit_transaction(trans
, root
);
1330 int btrfs_rm_device(struct btrfs_root
*root
, char *device_path
)
1332 struct btrfs_device
*device
;
1333 struct btrfs_device
*next_device
;
1334 struct block_device
*bdev
;
1335 struct buffer_head
*bh
= NULL
;
1336 struct btrfs_super_block
*disk_super
;
1337 struct btrfs_fs_devices
*cur_devices
;
1343 bool clear_super
= false;
1345 mutex_lock(&uuid_mutex
);
1347 all_avail
= root
->fs_info
->avail_data_alloc_bits
|
1348 root
->fs_info
->avail_system_alloc_bits
|
1349 root
->fs_info
->avail_metadata_alloc_bits
;
1351 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID10
) &&
1352 root
->fs_info
->fs_devices
->num_devices
<= 4) {
1353 printk(KERN_ERR
"btrfs: unable to go below four devices "
1359 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID1
) &&
1360 root
->fs_info
->fs_devices
->num_devices
<= 2) {
1361 printk(KERN_ERR
"btrfs: unable to go below two "
1362 "devices on raid1\n");
1367 if (strcmp(device_path
, "missing") == 0) {
1368 struct list_head
*devices
;
1369 struct btrfs_device
*tmp
;
1372 devices
= &root
->fs_info
->fs_devices
->devices
;
1374 * It is safe to read the devices since the volume_mutex
1377 list_for_each_entry(tmp
, devices
, dev_list
) {
1378 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
1387 printk(KERN_ERR
"btrfs: no missing devices found to "
1392 ret
= btrfs_get_bdev_and_sb(device_path
,
1393 FMODE_READ
| FMODE_EXCL
,
1394 root
->fs_info
->bdev_holder
, 0,
1398 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1399 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1400 dev_uuid
= disk_super
->dev_item
.uuid
;
1401 device
= btrfs_find_device(root
, devid
, dev_uuid
,
1409 if (device
->writeable
&& root
->fs_info
->fs_devices
->rw_devices
== 1) {
1410 printk(KERN_ERR
"btrfs: unable to remove the only writeable "
1416 if (device
->writeable
) {
1418 list_del_init(&device
->dev_alloc_list
);
1419 unlock_chunks(root
);
1420 root
->fs_info
->fs_devices
->rw_devices
--;
1424 ret
= btrfs_shrink_device(device
, 0);
1428 ret
= btrfs_rm_dev_item(root
->fs_info
->chunk_root
, device
);
1432 spin_lock(&root
->fs_info
->free_chunk_lock
);
1433 root
->fs_info
->free_chunk_space
= device
->total_bytes
-
1435 spin_unlock(&root
->fs_info
->free_chunk_lock
);
1437 device
->in_fs_metadata
= 0;
1438 btrfs_scrub_cancel_dev(root
, device
);
1441 * the device list mutex makes sure that we don't change
1442 * the device list while someone else is writing out all
1443 * the device supers.
1446 cur_devices
= device
->fs_devices
;
1447 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1448 list_del_rcu(&device
->dev_list
);
1450 device
->fs_devices
->num_devices
--;
1451 device
->fs_devices
->total_devices
--;
1453 if (device
->missing
)
1454 root
->fs_info
->fs_devices
->missing_devices
--;
1456 next_device
= list_entry(root
->fs_info
->fs_devices
->devices
.next
,
1457 struct btrfs_device
, dev_list
);
1458 if (device
->bdev
== root
->fs_info
->sb
->s_bdev
)
1459 root
->fs_info
->sb
->s_bdev
= next_device
->bdev
;
1460 if (device
->bdev
== root
->fs_info
->fs_devices
->latest_bdev
)
1461 root
->fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1464 device
->fs_devices
->open_devices
--;
1466 call_rcu(&device
->rcu
, free_device
);
1467 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1469 num_devices
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
1470 btrfs_set_super_num_devices(root
->fs_info
->super_copy
, num_devices
);
1472 if (cur_devices
->open_devices
== 0) {
1473 struct btrfs_fs_devices
*fs_devices
;
1474 fs_devices
= root
->fs_info
->fs_devices
;
1475 while (fs_devices
) {
1476 if (fs_devices
->seed
== cur_devices
)
1478 fs_devices
= fs_devices
->seed
;
1480 fs_devices
->seed
= cur_devices
->seed
;
1481 cur_devices
->seed
= NULL
;
1483 __btrfs_close_devices(cur_devices
);
1484 unlock_chunks(root
);
1485 free_fs_devices(cur_devices
);
1488 root
->fs_info
->num_tolerated_disk_barrier_failures
=
1489 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
1492 * at this point, the device is zero sized. We want to
1493 * remove it from the devices list and zero out the old super
1496 /* make sure this device isn't detected as part of
1499 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
1500 set_buffer_dirty(bh
);
1501 sync_dirty_buffer(bh
);
1510 blkdev_put(bdev
, FMODE_READ
| FMODE_EXCL
);
1512 mutex_unlock(&uuid_mutex
);
1515 if (device
->writeable
) {
1517 list_add(&device
->dev_alloc_list
,
1518 &root
->fs_info
->fs_devices
->alloc_list
);
1519 unlock_chunks(root
);
1520 root
->fs_info
->fs_devices
->rw_devices
++;
1526 * does all the dirty work required for changing file system's UUID.
1528 static int btrfs_prepare_sprout(struct btrfs_root
*root
)
1530 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
1531 struct btrfs_fs_devices
*old_devices
;
1532 struct btrfs_fs_devices
*seed_devices
;
1533 struct btrfs_super_block
*disk_super
= root
->fs_info
->super_copy
;
1534 struct btrfs_device
*device
;
1537 BUG_ON(!mutex_is_locked(&uuid_mutex
));
1538 if (!fs_devices
->seeding
)
1541 seed_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
1545 old_devices
= clone_fs_devices(fs_devices
);
1546 if (IS_ERR(old_devices
)) {
1547 kfree(seed_devices
);
1548 return PTR_ERR(old_devices
);
1551 list_add(&old_devices
->list
, &fs_uuids
);
1553 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
1554 seed_devices
->opened
= 1;
1555 INIT_LIST_HEAD(&seed_devices
->devices
);
1556 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
1557 mutex_init(&seed_devices
->device_list_mutex
);
1559 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1560 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
1562 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1564 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
1565 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
) {
1566 device
->fs_devices
= seed_devices
;
1569 fs_devices
->seeding
= 0;
1570 fs_devices
->num_devices
= 0;
1571 fs_devices
->open_devices
= 0;
1572 fs_devices
->total_devices
= 0;
1573 fs_devices
->seed
= seed_devices
;
1575 generate_random_uuid(fs_devices
->fsid
);
1576 memcpy(root
->fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
1577 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
1578 super_flags
= btrfs_super_flags(disk_super
) &
1579 ~BTRFS_SUPER_FLAG_SEEDING
;
1580 btrfs_set_super_flags(disk_super
, super_flags
);
1586 * strore the expected generation for seed devices in device items.
1588 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
1589 struct btrfs_root
*root
)
1591 struct btrfs_path
*path
;
1592 struct extent_buffer
*leaf
;
1593 struct btrfs_dev_item
*dev_item
;
1594 struct btrfs_device
*device
;
1595 struct btrfs_key key
;
1596 u8 fs_uuid
[BTRFS_UUID_SIZE
];
1597 u8 dev_uuid
[BTRFS_UUID_SIZE
];
1601 path
= btrfs_alloc_path();
1605 root
= root
->fs_info
->chunk_root
;
1606 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1608 key
.type
= BTRFS_DEV_ITEM_KEY
;
1611 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
1615 leaf
= path
->nodes
[0];
1617 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1618 ret
= btrfs_next_leaf(root
, path
);
1623 leaf
= path
->nodes
[0];
1624 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1625 btrfs_release_path(path
);
1629 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1630 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
1631 key
.type
!= BTRFS_DEV_ITEM_KEY
)
1634 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1635 struct btrfs_dev_item
);
1636 devid
= btrfs_device_id(leaf
, dev_item
);
1637 read_extent_buffer(leaf
, dev_uuid
,
1638 (unsigned long)btrfs_device_uuid(dev_item
),
1640 read_extent_buffer(leaf
, fs_uuid
,
1641 (unsigned long)btrfs_device_fsid(dev_item
),
1643 device
= btrfs_find_device(root
, devid
, dev_uuid
, fs_uuid
);
1644 BUG_ON(!device
); /* Logic error */
1646 if (device
->fs_devices
->seeding
) {
1647 btrfs_set_device_generation(leaf
, dev_item
,
1648 device
->generation
);
1649 btrfs_mark_buffer_dirty(leaf
);
1657 btrfs_free_path(path
);
1661 int btrfs_init_new_device(struct btrfs_root
*root
, char *device_path
)
1663 struct request_queue
*q
;
1664 struct btrfs_trans_handle
*trans
;
1665 struct btrfs_device
*device
;
1666 struct block_device
*bdev
;
1667 struct list_head
*devices
;
1668 struct super_block
*sb
= root
->fs_info
->sb
;
1669 struct rcu_string
*name
;
1671 int seeding_dev
= 0;
1674 if ((sb
->s_flags
& MS_RDONLY
) && !root
->fs_info
->fs_devices
->seeding
)
1677 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
1678 root
->fs_info
->bdev_holder
);
1680 return PTR_ERR(bdev
);
1682 if (root
->fs_info
->fs_devices
->seeding
) {
1684 down_write(&sb
->s_umount
);
1685 mutex_lock(&uuid_mutex
);
1688 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
1690 devices
= &root
->fs_info
->fs_devices
->devices
;
1692 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1693 list_for_each_entry(device
, devices
, dev_list
) {
1694 if (device
->bdev
== bdev
) {
1697 &root
->fs_info
->fs_devices
->device_list_mutex
);
1701 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1703 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
1705 /* we can safely leave the fs_devices entry around */
1710 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
1716 rcu_assign_pointer(device
->name
, name
);
1718 ret
= find_next_devid(root
, &device
->devid
);
1720 rcu_string_free(device
->name
);
1725 trans
= btrfs_start_transaction(root
, 0);
1726 if (IS_ERR(trans
)) {
1727 rcu_string_free(device
->name
);
1729 ret
= PTR_ERR(trans
);
1735 q
= bdev_get_queue(bdev
);
1736 if (blk_queue_discard(q
))
1737 device
->can_discard
= 1;
1738 device
->writeable
= 1;
1739 device
->work
.func
= pending_bios_fn
;
1740 generate_random_uuid(device
->uuid
);
1741 spin_lock_init(&device
->io_lock
);
1742 device
->generation
= trans
->transid
;
1743 device
->io_width
= root
->sectorsize
;
1744 device
->io_align
= root
->sectorsize
;
1745 device
->sector_size
= root
->sectorsize
;
1746 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
1747 device
->disk_total_bytes
= device
->total_bytes
;
1748 device
->dev_root
= root
->fs_info
->dev_root
;
1749 device
->bdev
= bdev
;
1750 device
->in_fs_metadata
= 1;
1751 device
->mode
= FMODE_EXCL
;
1752 set_blocksize(device
->bdev
, 4096);
1755 sb
->s_flags
&= ~MS_RDONLY
;
1756 ret
= btrfs_prepare_sprout(root
);
1757 BUG_ON(ret
); /* -ENOMEM */
1760 device
->fs_devices
= root
->fs_info
->fs_devices
;
1762 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1763 list_add_rcu(&device
->dev_list
, &root
->fs_info
->fs_devices
->devices
);
1764 list_add(&device
->dev_alloc_list
,
1765 &root
->fs_info
->fs_devices
->alloc_list
);
1766 root
->fs_info
->fs_devices
->num_devices
++;
1767 root
->fs_info
->fs_devices
->open_devices
++;
1768 root
->fs_info
->fs_devices
->rw_devices
++;
1769 root
->fs_info
->fs_devices
->total_devices
++;
1770 if (device
->can_discard
)
1771 root
->fs_info
->fs_devices
->num_can_discard
++;
1772 root
->fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
1774 spin_lock(&root
->fs_info
->free_chunk_lock
);
1775 root
->fs_info
->free_chunk_space
+= device
->total_bytes
;
1776 spin_unlock(&root
->fs_info
->free_chunk_lock
);
1778 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
1779 root
->fs_info
->fs_devices
->rotating
= 1;
1781 total_bytes
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
1782 btrfs_set_super_total_bytes(root
->fs_info
->super_copy
,
1783 total_bytes
+ device
->total_bytes
);
1785 total_bytes
= btrfs_super_num_devices(root
->fs_info
->super_copy
);
1786 btrfs_set_super_num_devices(root
->fs_info
->super_copy
,
1788 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1791 ret
= init_first_rw_device(trans
, root
, device
);
1793 btrfs_abort_transaction(trans
, root
, ret
);
1796 ret
= btrfs_finish_sprout(trans
, root
);
1798 btrfs_abort_transaction(trans
, root
, ret
);
1802 ret
= btrfs_add_device(trans
, root
, device
);
1804 btrfs_abort_transaction(trans
, root
, ret
);
1810 * we've got more storage, clear any full flags on the space
1813 btrfs_clear_space_info_full(root
->fs_info
);
1815 unlock_chunks(root
);
1816 root
->fs_info
->num_tolerated_disk_barrier_failures
=
1817 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
1818 ret
= btrfs_commit_transaction(trans
, root
);
1821 mutex_unlock(&uuid_mutex
);
1822 up_write(&sb
->s_umount
);
1824 if (ret
) /* transaction commit */
1827 ret
= btrfs_relocate_sys_chunks(root
);
1829 btrfs_error(root
->fs_info
, ret
,
1830 "Failed to relocate sys chunks after "
1831 "device initialization. This can be fixed "
1832 "using the \"btrfs balance\" command.");
1833 trans
= btrfs_attach_transaction(root
);
1834 if (IS_ERR(trans
)) {
1835 if (PTR_ERR(trans
) == -ENOENT
)
1837 return PTR_ERR(trans
);
1839 ret
= btrfs_commit_transaction(trans
, root
);
1845 unlock_chunks(root
);
1846 btrfs_end_transaction(trans
, root
);
1847 rcu_string_free(device
->name
);
1850 blkdev_put(bdev
, FMODE_EXCL
);
1852 mutex_unlock(&uuid_mutex
);
1853 up_write(&sb
->s_umount
);
1858 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
1859 struct btrfs_device
*device
)
1862 struct btrfs_path
*path
;
1863 struct btrfs_root
*root
;
1864 struct btrfs_dev_item
*dev_item
;
1865 struct extent_buffer
*leaf
;
1866 struct btrfs_key key
;
1868 root
= device
->dev_root
->fs_info
->chunk_root
;
1870 path
= btrfs_alloc_path();
1874 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1875 key
.type
= BTRFS_DEV_ITEM_KEY
;
1876 key
.offset
= device
->devid
;
1878 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
1887 leaf
= path
->nodes
[0];
1888 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1890 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1891 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1892 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1893 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1894 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1895 btrfs_set_device_total_bytes(leaf
, dev_item
, device
->disk_total_bytes
);
1896 btrfs_set_device_bytes_used(leaf
, dev_item
, device
->bytes_used
);
1897 btrfs_mark_buffer_dirty(leaf
);
1900 btrfs_free_path(path
);
1904 static int __btrfs_grow_device(struct btrfs_trans_handle
*trans
,
1905 struct btrfs_device
*device
, u64 new_size
)
1907 struct btrfs_super_block
*super_copy
=
1908 device
->dev_root
->fs_info
->super_copy
;
1909 u64 old_total
= btrfs_super_total_bytes(super_copy
);
1910 u64 diff
= new_size
- device
->total_bytes
;
1912 if (!device
->writeable
)
1914 if (new_size
<= device
->total_bytes
)
1917 btrfs_set_super_total_bytes(super_copy
, old_total
+ diff
);
1918 device
->fs_devices
->total_rw_bytes
+= diff
;
1920 device
->total_bytes
= new_size
;
1921 device
->disk_total_bytes
= new_size
;
1922 btrfs_clear_space_info_full(device
->dev_root
->fs_info
);
1924 return btrfs_update_device(trans
, device
);
1927 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
1928 struct btrfs_device
*device
, u64 new_size
)
1931 lock_chunks(device
->dev_root
);
1932 ret
= __btrfs_grow_device(trans
, device
, new_size
);
1933 unlock_chunks(device
->dev_root
);
1937 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
1938 struct btrfs_root
*root
,
1939 u64 chunk_tree
, u64 chunk_objectid
,
1943 struct btrfs_path
*path
;
1944 struct btrfs_key key
;
1946 root
= root
->fs_info
->chunk_root
;
1947 path
= btrfs_alloc_path();
1951 key
.objectid
= chunk_objectid
;
1952 key
.offset
= chunk_offset
;
1953 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
1955 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1958 else if (ret
> 0) { /* Logic error or corruption */
1959 btrfs_error(root
->fs_info
, -ENOENT
,
1960 "Failed lookup while freeing chunk.");
1965 ret
= btrfs_del_item(trans
, root
, path
);
1967 btrfs_error(root
->fs_info
, ret
,
1968 "Failed to delete chunk item.");
1970 btrfs_free_path(path
);
1974 static int btrfs_del_sys_chunk(struct btrfs_root
*root
, u64 chunk_objectid
, u64
1977 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
1978 struct btrfs_disk_key
*disk_key
;
1979 struct btrfs_chunk
*chunk
;
1986 struct btrfs_key key
;
1988 array_size
= btrfs_super_sys_array_size(super_copy
);
1990 ptr
= super_copy
->sys_chunk_array
;
1993 while (cur
< array_size
) {
1994 disk_key
= (struct btrfs_disk_key
*)ptr
;
1995 btrfs_disk_key_to_cpu(&key
, disk_key
);
1997 len
= sizeof(*disk_key
);
1999 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2000 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2001 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2002 len
+= btrfs_chunk_item_size(num_stripes
);
2007 if (key
.objectid
== chunk_objectid
&&
2008 key
.offset
== chunk_offset
) {
2009 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2011 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2020 static int btrfs_relocate_chunk(struct btrfs_root
*root
,
2021 u64 chunk_tree
, u64 chunk_objectid
,
2024 struct extent_map_tree
*em_tree
;
2025 struct btrfs_root
*extent_root
;
2026 struct btrfs_trans_handle
*trans
;
2027 struct extent_map
*em
;
2028 struct map_lookup
*map
;
2032 root
= root
->fs_info
->chunk_root
;
2033 extent_root
= root
->fs_info
->extent_root
;
2034 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
2036 ret
= btrfs_can_relocate(extent_root
, chunk_offset
);
2040 /* step one, relocate all the extents inside this chunk */
2041 ret
= btrfs_relocate_block_group(extent_root
, chunk_offset
);
2045 trans
= btrfs_start_transaction(root
, 0);
2046 BUG_ON(IS_ERR(trans
));
2051 * step two, delete the device extents and the
2052 * chunk tree entries
2054 read_lock(&em_tree
->lock
);
2055 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
2056 read_unlock(&em_tree
->lock
);
2058 BUG_ON(!em
|| em
->start
> chunk_offset
||
2059 em
->start
+ em
->len
< chunk_offset
);
2060 map
= (struct map_lookup
*)em
->bdev
;
2062 for (i
= 0; i
< map
->num_stripes
; i
++) {
2063 ret
= btrfs_free_dev_extent(trans
, map
->stripes
[i
].dev
,
2064 map
->stripes
[i
].physical
);
2067 if (map
->stripes
[i
].dev
) {
2068 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2072 ret
= btrfs_free_chunk(trans
, root
, chunk_tree
, chunk_objectid
,
2077 trace_btrfs_chunk_free(root
, map
, chunk_offset
, em
->len
);
2079 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2080 ret
= btrfs_del_sys_chunk(root
, chunk_objectid
, chunk_offset
);
2084 ret
= btrfs_remove_block_group(trans
, extent_root
, chunk_offset
);
2087 write_lock(&em_tree
->lock
);
2088 remove_extent_mapping(em_tree
, em
);
2089 write_unlock(&em_tree
->lock
);
2094 /* once for the tree */
2095 free_extent_map(em
);
2097 free_extent_map(em
);
2099 unlock_chunks(root
);
2100 btrfs_end_transaction(trans
, root
);
2104 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
)
2106 struct btrfs_root
*chunk_root
= root
->fs_info
->chunk_root
;
2107 struct btrfs_path
*path
;
2108 struct extent_buffer
*leaf
;
2109 struct btrfs_chunk
*chunk
;
2110 struct btrfs_key key
;
2111 struct btrfs_key found_key
;
2112 u64 chunk_tree
= chunk_root
->root_key
.objectid
;
2114 bool retried
= false;
2118 path
= btrfs_alloc_path();
2123 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2124 key
.offset
= (u64
)-1;
2125 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2128 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2131 BUG_ON(ret
== 0); /* Corruption */
2133 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
2140 leaf
= path
->nodes
[0];
2141 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2143 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
2144 struct btrfs_chunk
);
2145 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2146 btrfs_release_path(path
);
2148 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2149 ret
= btrfs_relocate_chunk(chunk_root
, chunk_tree
,
2158 if (found_key
.offset
== 0)
2160 key
.offset
= found_key
.offset
- 1;
2163 if (failed
&& !retried
) {
2167 } else if (failed
&& retried
) {
2172 btrfs_free_path(path
);
2176 static int insert_balance_item(struct btrfs_root
*root
,
2177 struct btrfs_balance_control
*bctl
)
2179 struct btrfs_trans_handle
*trans
;
2180 struct btrfs_balance_item
*item
;
2181 struct btrfs_disk_balance_args disk_bargs
;
2182 struct btrfs_path
*path
;
2183 struct extent_buffer
*leaf
;
2184 struct btrfs_key key
;
2187 path
= btrfs_alloc_path();
2191 trans
= btrfs_start_transaction(root
, 0);
2192 if (IS_ERR(trans
)) {
2193 btrfs_free_path(path
);
2194 return PTR_ERR(trans
);
2197 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2198 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2201 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2206 leaf
= path
->nodes
[0];
2207 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
2209 memset_extent_buffer(leaf
, 0, (unsigned long)item
, sizeof(*item
));
2211 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
2212 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
2213 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
2214 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
2215 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
2216 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
2218 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
2220 btrfs_mark_buffer_dirty(leaf
);
2222 btrfs_free_path(path
);
2223 err
= btrfs_commit_transaction(trans
, root
);
2229 static int del_balance_item(struct btrfs_root
*root
)
2231 struct btrfs_trans_handle
*trans
;
2232 struct btrfs_path
*path
;
2233 struct btrfs_key key
;
2236 path
= btrfs_alloc_path();
2240 trans
= btrfs_start_transaction(root
, 0);
2241 if (IS_ERR(trans
)) {
2242 btrfs_free_path(path
);
2243 return PTR_ERR(trans
);
2246 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2247 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2250 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2258 ret
= btrfs_del_item(trans
, root
, path
);
2260 btrfs_free_path(path
);
2261 err
= btrfs_commit_transaction(trans
, root
);
2268 * This is a heuristic used to reduce the number of chunks balanced on
2269 * resume after balance was interrupted.
2271 static void update_balance_args(struct btrfs_balance_control
*bctl
)
2274 * Turn on soft mode for chunk types that were being converted.
2276 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2277 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2278 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2279 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2280 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2281 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2284 * Turn on usage filter if is not already used. The idea is
2285 * that chunks that we have already balanced should be
2286 * reasonably full. Don't do it for chunks that are being
2287 * converted - that will keep us from relocating unconverted
2288 * (albeit full) chunks.
2290 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2291 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2292 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2293 bctl
->data
.usage
= 90;
2295 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2296 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2297 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2298 bctl
->sys
.usage
= 90;
2300 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2301 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2302 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2303 bctl
->meta
.usage
= 90;
2308 * Should be called with both balance and volume mutexes held to
2309 * serialize other volume operations (add_dev/rm_dev/resize) with
2310 * restriper. Same goes for unset_balance_control.
2312 static void set_balance_control(struct btrfs_balance_control
*bctl
)
2314 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
2316 BUG_ON(fs_info
->balance_ctl
);
2318 spin_lock(&fs_info
->balance_lock
);
2319 fs_info
->balance_ctl
= bctl
;
2320 spin_unlock(&fs_info
->balance_lock
);
2323 static void unset_balance_control(struct btrfs_fs_info
*fs_info
)
2325 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
2327 BUG_ON(!fs_info
->balance_ctl
);
2329 spin_lock(&fs_info
->balance_lock
);
2330 fs_info
->balance_ctl
= NULL
;
2331 spin_unlock(&fs_info
->balance_lock
);
2337 * Balance filters. Return 1 if chunk should be filtered out
2338 * (should not be balanced).
2340 static int chunk_profiles_filter(u64 chunk_type
,
2341 struct btrfs_balance_args
*bargs
)
2343 chunk_type
= chunk_to_extended(chunk_type
) &
2344 BTRFS_EXTENDED_PROFILE_MASK
;
2346 if (bargs
->profiles
& chunk_type
)
2352 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
2353 struct btrfs_balance_args
*bargs
)
2355 struct btrfs_block_group_cache
*cache
;
2356 u64 chunk_used
, user_thresh
;
2359 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
2360 chunk_used
= btrfs_block_group_used(&cache
->item
);
2362 user_thresh
= div_factor_fine(cache
->key
.offset
, bargs
->usage
);
2363 if (chunk_used
< user_thresh
)
2366 btrfs_put_block_group(cache
);
2370 static int chunk_devid_filter(struct extent_buffer
*leaf
,
2371 struct btrfs_chunk
*chunk
,
2372 struct btrfs_balance_args
*bargs
)
2374 struct btrfs_stripe
*stripe
;
2375 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2378 for (i
= 0; i
< num_stripes
; i
++) {
2379 stripe
= btrfs_stripe_nr(chunk
, i
);
2380 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
2387 /* [pstart, pend) */
2388 static int chunk_drange_filter(struct extent_buffer
*leaf
,
2389 struct btrfs_chunk
*chunk
,
2391 struct btrfs_balance_args
*bargs
)
2393 struct btrfs_stripe
*stripe
;
2394 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2400 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
2403 if (btrfs_chunk_type(leaf
, chunk
) & (BTRFS_BLOCK_GROUP_DUP
|
2404 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
))
2408 factor
= num_stripes
/ factor
;
2410 for (i
= 0; i
< num_stripes
; i
++) {
2411 stripe
= btrfs_stripe_nr(chunk
, i
);
2412 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
2415 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
2416 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
2417 do_div(stripe_length
, factor
);
2419 if (stripe_offset
< bargs
->pend
&&
2420 stripe_offset
+ stripe_length
> bargs
->pstart
)
2427 /* [vstart, vend) */
2428 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
2429 struct btrfs_chunk
*chunk
,
2431 struct btrfs_balance_args
*bargs
)
2433 if (chunk_offset
< bargs
->vend
&&
2434 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
2435 /* at least part of the chunk is inside this vrange */
2441 static int chunk_soft_convert_filter(u64 chunk_type
,
2442 struct btrfs_balance_args
*bargs
)
2444 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
2447 chunk_type
= chunk_to_extended(chunk_type
) &
2448 BTRFS_EXTENDED_PROFILE_MASK
;
2450 if (bargs
->target
== chunk_type
)
2456 static int should_balance_chunk(struct btrfs_root
*root
,
2457 struct extent_buffer
*leaf
,
2458 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
2460 struct btrfs_balance_control
*bctl
= root
->fs_info
->balance_ctl
;
2461 struct btrfs_balance_args
*bargs
= NULL
;
2462 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2465 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
2466 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
2470 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
2471 bargs
= &bctl
->data
;
2472 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
2474 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
2475 bargs
= &bctl
->meta
;
2477 /* profiles filter */
2478 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
2479 chunk_profiles_filter(chunk_type
, bargs
)) {
2484 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2485 chunk_usage_filter(bctl
->fs_info
, chunk_offset
, bargs
)) {
2490 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
2491 chunk_devid_filter(leaf
, chunk
, bargs
)) {
2495 /* drange filter, makes sense only with devid filter */
2496 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
2497 chunk_drange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
2502 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
2503 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
2507 /* soft profile changing mode */
2508 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
2509 chunk_soft_convert_filter(chunk_type
, bargs
)) {
2516 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
2518 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
2519 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
2520 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
2521 struct list_head
*devices
;
2522 struct btrfs_device
*device
;
2525 struct btrfs_chunk
*chunk
;
2526 struct btrfs_path
*path
;
2527 struct btrfs_key key
;
2528 struct btrfs_key found_key
;
2529 struct btrfs_trans_handle
*trans
;
2530 struct extent_buffer
*leaf
;
2533 int enospc_errors
= 0;
2534 bool counting
= true;
2536 /* step one make some room on all the devices */
2537 devices
= &fs_info
->fs_devices
->devices
;
2538 list_for_each_entry(device
, devices
, dev_list
) {
2539 old_size
= device
->total_bytes
;
2540 size_to_free
= div_factor(old_size
, 1);
2541 size_to_free
= min(size_to_free
, (u64
)1 * 1024 * 1024);
2542 if (!device
->writeable
||
2543 device
->total_bytes
- device
->bytes_used
> size_to_free
)
2546 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
2551 trans
= btrfs_start_transaction(dev_root
, 0);
2552 BUG_ON(IS_ERR(trans
));
2554 ret
= btrfs_grow_device(trans
, device
, old_size
);
2557 btrfs_end_transaction(trans
, dev_root
);
2560 /* step two, relocate all the chunks */
2561 path
= btrfs_alloc_path();
2567 /* zero out stat counters */
2568 spin_lock(&fs_info
->balance_lock
);
2569 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
2570 spin_unlock(&fs_info
->balance_lock
);
2572 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2573 key
.offset
= (u64
)-1;
2574 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2577 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
2578 atomic_read(&fs_info
->balance_cancel_req
)) {
2583 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2588 * this shouldn't happen, it means the last relocate
2592 BUG(); /* FIXME break ? */
2594 ret
= btrfs_previous_item(chunk_root
, path
, 0,
2595 BTRFS_CHUNK_ITEM_KEY
);
2601 leaf
= path
->nodes
[0];
2602 slot
= path
->slots
[0];
2603 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2605 if (found_key
.objectid
!= key
.objectid
)
2608 /* chunk zero is special */
2609 if (found_key
.offset
== 0)
2612 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
2615 spin_lock(&fs_info
->balance_lock
);
2616 bctl
->stat
.considered
++;
2617 spin_unlock(&fs_info
->balance_lock
);
2620 ret
= should_balance_chunk(chunk_root
, leaf
, chunk
,
2622 btrfs_release_path(path
);
2627 spin_lock(&fs_info
->balance_lock
);
2628 bctl
->stat
.expected
++;
2629 spin_unlock(&fs_info
->balance_lock
);
2633 ret
= btrfs_relocate_chunk(chunk_root
,
2634 chunk_root
->root_key
.objectid
,
2637 if (ret
&& ret
!= -ENOSPC
)
2639 if (ret
== -ENOSPC
) {
2642 spin_lock(&fs_info
->balance_lock
);
2643 bctl
->stat
.completed
++;
2644 spin_unlock(&fs_info
->balance_lock
);
2647 key
.offset
= found_key
.offset
- 1;
2651 btrfs_release_path(path
);
2656 btrfs_free_path(path
);
2657 if (enospc_errors
) {
2658 printk(KERN_INFO
"btrfs: %d enospc errors during balance\n",
2668 * alloc_profile_is_valid - see if a given profile is valid and reduced
2669 * @flags: profile to validate
2670 * @extended: if true @flags is treated as an extended profile
2672 static int alloc_profile_is_valid(u64 flags
, int extended
)
2674 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
2675 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
2677 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
2679 /* 1) check that all other bits are zeroed */
2683 /* 2) see if profile is reduced */
2685 return !extended
; /* "0" is valid for usual profiles */
2687 /* true if exactly one bit set */
2688 return (flags
& (flags
- 1)) == 0;
2691 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
2693 /* cancel requested || normal exit path */
2694 return atomic_read(&fs_info
->balance_cancel_req
) ||
2695 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
2696 atomic_read(&fs_info
->balance_cancel_req
) == 0);
2699 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
2703 unset_balance_control(fs_info
);
2704 ret
= del_balance_item(fs_info
->tree_root
);
2708 void update_ioctl_balance_args(struct btrfs_fs_info
*fs_info
, int lock
,
2709 struct btrfs_ioctl_balance_args
*bargs
);
2712 * Should be called with both balance and volume mutexes held
2714 int btrfs_balance(struct btrfs_balance_control
*bctl
,
2715 struct btrfs_ioctl_balance_args
*bargs
)
2717 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
2722 if (btrfs_fs_closing(fs_info
) ||
2723 atomic_read(&fs_info
->balance_pause_req
) ||
2724 atomic_read(&fs_info
->balance_cancel_req
)) {
2729 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
2730 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
2734 * In case of mixed groups both data and meta should be picked,
2735 * and identical options should be given for both of them.
2737 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
2738 if (mixed
&& (bctl
->flags
& allowed
)) {
2739 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
2740 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
2741 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
2742 printk(KERN_ERR
"btrfs: with mixed groups data and "
2743 "metadata balance options must be the same\n");
2749 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
2750 if (fs_info
->fs_devices
->num_devices
== 1)
2751 allowed
|= BTRFS_BLOCK_GROUP_DUP
;
2752 else if (fs_info
->fs_devices
->num_devices
< 4)
2753 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
2755 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
|
2756 BTRFS_BLOCK_GROUP_RAID10
);
2758 if ((bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
2759 (!alloc_profile_is_valid(bctl
->data
.target
, 1) ||
2760 (bctl
->data
.target
& ~allowed
))) {
2761 printk(KERN_ERR
"btrfs: unable to start balance with target "
2762 "data profile %llu\n",
2763 (unsigned long long)bctl
->data
.target
);
2767 if ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
2768 (!alloc_profile_is_valid(bctl
->meta
.target
, 1) ||
2769 (bctl
->meta
.target
& ~allowed
))) {
2770 printk(KERN_ERR
"btrfs: unable to start balance with target "
2771 "metadata profile %llu\n",
2772 (unsigned long long)bctl
->meta
.target
);
2776 if ((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
2777 (!alloc_profile_is_valid(bctl
->sys
.target
, 1) ||
2778 (bctl
->sys
.target
& ~allowed
))) {
2779 printk(KERN_ERR
"btrfs: unable to start balance with target "
2780 "system profile %llu\n",
2781 (unsigned long long)bctl
->sys
.target
);
2786 /* allow dup'ed data chunks only in mixed mode */
2787 if (!mixed
&& (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
2788 (bctl
->data
.target
& BTRFS_BLOCK_GROUP_DUP
)) {
2789 printk(KERN_ERR
"btrfs: dup for data is not allowed\n");
2794 /* allow to reduce meta or sys integrity only if force set */
2795 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
2796 BTRFS_BLOCK_GROUP_RAID10
;
2797 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
2798 (fs_info
->avail_system_alloc_bits
& allowed
) &&
2799 !(bctl
->sys
.target
& allowed
)) ||
2800 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
2801 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
2802 !(bctl
->meta
.target
& allowed
))) {
2803 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
2804 printk(KERN_INFO
"btrfs: force reducing metadata "
2807 printk(KERN_ERR
"btrfs: balance will reduce metadata "
2808 "integrity, use force if you want this\n");
2814 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
2815 int num_tolerated_disk_barrier_failures
;
2816 u64 target
= bctl
->sys
.target
;
2818 num_tolerated_disk_barrier_failures
=
2819 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
2820 if (num_tolerated_disk_barrier_failures
> 0 &&
2822 (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID0
|
2823 BTRFS_AVAIL_ALLOC_BIT_SINGLE
)))
2824 num_tolerated_disk_barrier_failures
= 0;
2825 else if (num_tolerated_disk_barrier_failures
> 1 &&
2827 (BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)))
2828 num_tolerated_disk_barrier_failures
= 1;
2830 fs_info
->num_tolerated_disk_barrier_failures
=
2831 num_tolerated_disk_barrier_failures
;
2834 ret
= insert_balance_item(fs_info
->tree_root
, bctl
);
2835 if (ret
&& ret
!= -EEXIST
)
2838 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
2839 BUG_ON(ret
== -EEXIST
);
2840 set_balance_control(bctl
);
2842 BUG_ON(ret
!= -EEXIST
);
2843 spin_lock(&fs_info
->balance_lock
);
2844 update_balance_args(bctl
);
2845 spin_unlock(&fs_info
->balance_lock
);
2848 atomic_inc(&fs_info
->balance_running
);
2849 mutex_unlock(&fs_info
->balance_mutex
);
2851 ret
= __btrfs_balance(fs_info
);
2853 mutex_lock(&fs_info
->balance_mutex
);
2854 atomic_dec(&fs_info
->balance_running
);
2857 memset(bargs
, 0, sizeof(*bargs
));
2858 update_ioctl_balance_args(fs_info
, 0, bargs
);
2861 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
2862 balance_need_close(fs_info
)) {
2863 __cancel_balance(fs_info
);
2866 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
2867 fs_info
->num_tolerated_disk_barrier_failures
=
2868 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
2871 wake_up(&fs_info
->balance_wait_q
);
2875 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
2876 __cancel_balance(fs_info
);
2882 static int balance_kthread(void *data
)
2884 struct btrfs_fs_info
*fs_info
= data
;
2887 mutex_lock(&fs_info
->volume_mutex
);
2888 mutex_lock(&fs_info
->balance_mutex
);
2890 if (fs_info
->balance_ctl
) {
2891 printk(KERN_INFO
"btrfs: continuing balance\n");
2892 ret
= btrfs_balance(fs_info
->balance_ctl
, NULL
);
2895 mutex_unlock(&fs_info
->balance_mutex
);
2896 mutex_unlock(&fs_info
->volume_mutex
);
2901 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
2903 struct task_struct
*tsk
;
2905 spin_lock(&fs_info
->balance_lock
);
2906 if (!fs_info
->balance_ctl
) {
2907 spin_unlock(&fs_info
->balance_lock
);
2910 spin_unlock(&fs_info
->balance_lock
);
2912 if (btrfs_test_opt(fs_info
->tree_root
, SKIP_BALANCE
)) {
2913 printk(KERN_INFO
"btrfs: force skipping balance\n");
2917 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
2919 return PTR_ERR(tsk
);
2924 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
2926 struct btrfs_balance_control
*bctl
;
2927 struct btrfs_balance_item
*item
;
2928 struct btrfs_disk_balance_args disk_bargs
;
2929 struct btrfs_path
*path
;
2930 struct extent_buffer
*leaf
;
2931 struct btrfs_key key
;
2934 path
= btrfs_alloc_path();
2938 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2939 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2942 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
2945 if (ret
> 0) { /* ret = -ENOENT; */
2950 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
2956 leaf
= path
->nodes
[0];
2957 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
2959 bctl
->fs_info
= fs_info
;
2960 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
2961 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
2963 btrfs_balance_data(leaf
, item
, &disk_bargs
);
2964 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
2965 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
2966 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
2967 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
2968 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
2970 mutex_lock(&fs_info
->volume_mutex
);
2971 mutex_lock(&fs_info
->balance_mutex
);
2973 set_balance_control(bctl
);
2975 mutex_unlock(&fs_info
->balance_mutex
);
2976 mutex_unlock(&fs_info
->volume_mutex
);
2978 btrfs_free_path(path
);
2982 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
2986 mutex_lock(&fs_info
->balance_mutex
);
2987 if (!fs_info
->balance_ctl
) {
2988 mutex_unlock(&fs_info
->balance_mutex
);
2992 if (atomic_read(&fs_info
->balance_running
)) {
2993 atomic_inc(&fs_info
->balance_pause_req
);
2994 mutex_unlock(&fs_info
->balance_mutex
);
2996 wait_event(fs_info
->balance_wait_q
,
2997 atomic_read(&fs_info
->balance_running
) == 0);
2999 mutex_lock(&fs_info
->balance_mutex
);
3000 /* we are good with balance_ctl ripped off from under us */
3001 BUG_ON(atomic_read(&fs_info
->balance_running
));
3002 atomic_dec(&fs_info
->balance_pause_req
);
3007 mutex_unlock(&fs_info
->balance_mutex
);
3011 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
3013 mutex_lock(&fs_info
->balance_mutex
);
3014 if (!fs_info
->balance_ctl
) {
3015 mutex_unlock(&fs_info
->balance_mutex
);
3019 atomic_inc(&fs_info
->balance_cancel_req
);
3021 * if we are running just wait and return, balance item is
3022 * deleted in btrfs_balance in this case
3024 if (atomic_read(&fs_info
->balance_running
)) {
3025 mutex_unlock(&fs_info
->balance_mutex
);
3026 wait_event(fs_info
->balance_wait_q
,
3027 atomic_read(&fs_info
->balance_running
) == 0);
3028 mutex_lock(&fs_info
->balance_mutex
);
3030 /* __cancel_balance needs volume_mutex */
3031 mutex_unlock(&fs_info
->balance_mutex
);
3032 mutex_lock(&fs_info
->volume_mutex
);
3033 mutex_lock(&fs_info
->balance_mutex
);
3035 if (fs_info
->balance_ctl
)
3036 __cancel_balance(fs_info
);
3038 mutex_unlock(&fs_info
->volume_mutex
);
3041 BUG_ON(fs_info
->balance_ctl
|| atomic_read(&fs_info
->balance_running
));
3042 atomic_dec(&fs_info
->balance_cancel_req
);
3043 mutex_unlock(&fs_info
->balance_mutex
);
3048 * shrinking a device means finding all of the device extents past
3049 * the new size, and then following the back refs to the chunks.
3050 * The chunk relocation code actually frees the device extent
3052 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
3054 struct btrfs_trans_handle
*trans
;
3055 struct btrfs_root
*root
= device
->dev_root
;
3056 struct btrfs_dev_extent
*dev_extent
= NULL
;
3057 struct btrfs_path
*path
;
3065 bool retried
= false;
3066 struct extent_buffer
*l
;
3067 struct btrfs_key key
;
3068 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
3069 u64 old_total
= btrfs_super_total_bytes(super_copy
);
3070 u64 old_size
= device
->total_bytes
;
3071 u64 diff
= device
->total_bytes
- new_size
;
3073 path
= btrfs_alloc_path();
3081 device
->total_bytes
= new_size
;
3082 if (device
->writeable
) {
3083 device
->fs_devices
->total_rw_bytes
-= diff
;
3084 spin_lock(&root
->fs_info
->free_chunk_lock
);
3085 root
->fs_info
->free_chunk_space
-= diff
;
3086 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3088 unlock_chunks(root
);
3091 key
.objectid
= device
->devid
;
3092 key
.offset
= (u64
)-1;
3093 key
.type
= BTRFS_DEV_EXTENT_KEY
;
3096 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3100 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
3105 btrfs_release_path(path
);
3110 slot
= path
->slots
[0];
3111 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
3113 if (key
.objectid
!= device
->devid
) {
3114 btrfs_release_path(path
);
3118 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
3119 length
= btrfs_dev_extent_length(l
, dev_extent
);
3121 if (key
.offset
+ length
<= new_size
) {
3122 btrfs_release_path(path
);
3126 chunk_tree
= btrfs_dev_extent_chunk_tree(l
, dev_extent
);
3127 chunk_objectid
= btrfs_dev_extent_chunk_objectid(l
, dev_extent
);
3128 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
3129 btrfs_release_path(path
);
3131 ret
= btrfs_relocate_chunk(root
, chunk_tree
, chunk_objectid
,
3133 if (ret
&& ret
!= -ENOSPC
)
3137 } while (key
.offset
-- > 0);
3139 if (failed
&& !retried
) {
3143 } else if (failed
&& retried
) {
3147 device
->total_bytes
= old_size
;
3148 if (device
->writeable
)
3149 device
->fs_devices
->total_rw_bytes
+= diff
;
3150 spin_lock(&root
->fs_info
->free_chunk_lock
);
3151 root
->fs_info
->free_chunk_space
+= diff
;
3152 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3153 unlock_chunks(root
);
3157 /* Shrinking succeeded, else we would be at "done". */
3158 trans
= btrfs_start_transaction(root
, 0);
3159 if (IS_ERR(trans
)) {
3160 ret
= PTR_ERR(trans
);
3166 device
->disk_total_bytes
= new_size
;
3167 /* Now btrfs_update_device() will change the on-disk size. */
3168 ret
= btrfs_update_device(trans
, device
);
3170 unlock_chunks(root
);
3171 btrfs_end_transaction(trans
, root
);
3174 WARN_ON(diff
> old_total
);
3175 btrfs_set_super_total_bytes(super_copy
, old_total
- diff
);
3176 unlock_chunks(root
);
3177 btrfs_end_transaction(trans
, root
);
3179 btrfs_free_path(path
);
3183 static int btrfs_add_system_chunk(struct btrfs_root
*root
,
3184 struct btrfs_key
*key
,
3185 struct btrfs_chunk
*chunk
, int item_size
)
3187 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
3188 struct btrfs_disk_key disk_key
;
3192 array_size
= btrfs_super_sys_array_size(super_copy
);
3193 if (array_size
+ item_size
> BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
)
3196 ptr
= super_copy
->sys_chunk_array
+ array_size
;
3197 btrfs_cpu_key_to_disk(&disk_key
, key
);
3198 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
3199 ptr
+= sizeof(disk_key
);
3200 memcpy(ptr
, chunk
, item_size
);
3201 item_size
+= sizeof(disk_key
);
3202 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
3207 * sort the devices in descending order by max_avail, total_avail
3209 static int btrfs_cmp_device_info(const void *a
, const void *b
)
3211 const struct btrfs_device_info
*di_a
= a
;
3212 const struct btrfs_device_info
*di_b
= b
;
3214 if (di_a
->max_avail
> di_b
->max_avail
)
3216 if (di_a
->max_avail
< di_b
->max_avail
)
3218 if (di_a
->total_avail
> di_b
->total_avail
)
3220 if (di_a
->total_avail
< di_b
->total_avail
)
3225 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
3226 struct btrfs_root
*extent_root
,
3227 struct map_lookup
**map_ret
,
3228 u64
*num_bytes_out
, u64
*stripe_size_out
,
3229 u64 start
, u64 type
)
3231 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
3232 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
3233 struct list_head
*cur
;
3234 struct map_lookup
*map
= NULL
;
3235 struct extent_map_tree
*em_tree
;
3236 struct extent_map
*em
;
3237 struct btrfs_device_info
*devices_info
= NULL
;
3239 int num_stripes
; /* total number of stripes to allocate */
3240 int sub_stripes
; /* sub_stripes info for map */
3241 int dev_stripes
; /* stripes per dev */
3242 int devs_max
; /* max devs to use */
3243 int devs_min
; /* min devs needed */
3244 int devs_increment
; /* ndevs has to be a multiple of this */
3245 int ncopies
; /* how many copies to data has */
3247 u64 max_stripe_size
;
3255 BUG_ON(!alloc_profile_is_valid(type
, 0));
3257 if (list_empty(&fs_devices
->alloc_list
))
3264 devs_max
= 0; /* 0 == as many as possible */
3268 * define the properties of each RAID type.
3269 * FIXME: move this to a global table and use it in all RAID
3272 if (type
& (BTRFS_BLOCK_GROUP_DUP
)) {
3276 } else if (type
& (BTRFS_BLOCK_GROUP_RAID0
)) {
3278 } else if (type
& (BTRFS_BLOCK_GROUP_RAID1
)) {
3283 } else if (type
& (BTRFS_BLOCK_GROUP_RAID10
)) {
3292 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
3293 max_stripe_size
= 1024 * 1024 * 1024;
3294 max_chunk_size
= 10 * max_stripe_size
;
3295 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
3296 /* for larger filesystems, use larger metadata chunks */
3297 if (fs_devices
->total_rw_bytes
> 50ULL * 1024 * 1024 * 1024)
3298 max_stripe_size
= 1024 * 1024 * 1024;
3300 max_stripe_size
= 256 * 1024 * 1024;
3301 max_chunk_size
= max_stripe_size
;
3302 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3303 max_stripe_size
= 32 * 1024 * 1024;
3304 max_chunk_size
= 2 * max_stripe_size
;
3306 printk(KERN_ERR
"btrfs: invalid chunk type 0x%llx requested\n",
3311 /* we don't want a chunk larger than 10% of writeable space */
3312 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
3315 devices_info
= kzalloc(sizeof(*devices_info
) * fs_devices
->rw_devices
,
3320 cur
= fs_devices
->alloc_list
.next
;
3323 * in the first pass through the devices list, we gather information
3324 * about the available holes on each device.
3327 while (cur
!= &fs_devices
->alloc_list
) {
3328 struct btrfs_device
*device
;
3332 device
= list_entry(cur
, struct btrfs_device
, dev_alloc_list
);
3336 if (!device
->writeable
) {
3338 "btrfs: read-only device in alloc_list\n");
3342 if (!device
->in_fs_metadata
)
3345 if (device
->total_bytes
> device
->bytes_used
)
3346 total_avail
= device
->total_bytes
- device
->bytes_used
;
3350 /* If there is no space on this device, skip it. */
3351 if (total_avail
== 0)
3354 ret
= find_free_dev_extent(device
,
3355 max_stripe_size
* dev_stripes
,
3356 &dev_offset
, &max_avail
);
3357 if (ret
&& ret
!= -ENOSPC
)
3361 max_avail
= max_stripe_size
* dev_stripes
;
3363 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
3366 devices_info
[ndevs
].dev_offset
= dev_offset
;
3367 devices_info
[ndevs
].max_avail
= max_avail
;
3368 devices_info
[ndevs
].total_avail
= total_avail
;
3369 devices_info
[ndevs
].dev
= device
;
3374 * now sort the devices by hole size / available space
3376 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
3377 btrfs_cmp_device_info
, NULL
);
3379 /* round down to number of usable stripes */
3380 ndevs
-= ndevs
% devs_increment
;
3382 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
3387 if (devs_max
&& ndevs
> devs_max
)
3390 * the primary goal is to maximize the number of stripes, so use as many
3391 * devices as possible, even if the stripes are not maximum sized.
3393 stripe_size
= devices_info
[ndevs
-1].max_avail
;
3394 num_stripes
= ndevs
* dev_stripes
;
3396 if (stripe_size
* ndevs
> max_chunk_size
* ncopies
) {
3397 stripe_size
= max_chunk_size
* ncopies
;
3398 do_div(stripe_size
, ndevs
);
3401 do_div(stripe_size
, dev_stripes
);
3403 /* align to BTRFS_STRIPE_LEN */
3404 do_div(stripe_size
, BTRFS_STRIPE_LEN
);
3405 stripe_size
*= BTRFS_STRIPE_LEN
;
3407 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
3412 map
->num_stripes
= num_stripes
;
3414 for (i
= 0; i
< ndevs
; ++i
) {
3415 for (j
= 0; j
< dev_stripes
; ++j
) {
3416 int s
= i
* dev_stripes
+ j
;
3417 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
3418 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
3422 map
->sector_size
= extent_root
->sectorsize
;
3423 map
->stripe_len
= BTRFS_STRIPE_LEN
;
3424 map
->io_align
= BTRFS_STRIPE_LEN
;
3425 map
->io_width
= BTRFS_STRIPE_LEN
;
3427 map
->sub_stripes
= sub_stripes
;
3430 num_bytes
= stripe_size
* (num_stripes
/ ncopies
);
3432 *stripe_size_out
= stripe_size
;
3433 *num_bytes_out
= num_bytes
;
3435 trace_btrfs_chunk_alloc(info
->chunk_root
, map
, start
, num_bytes
);
3437 em
= alloc_extent_map();
3442 em
->bdev
= (struct block_device
*)map
;
3444 em
->len
= num_bytes
;
3445 em
->block_start
= 0;
3446 em
->block_len
= em
->len
;
3448 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
3449 write_lock(&em_tree
->lock
);
3450 ret
= add_extent_mapping(em_tree
, em
);
3451 write_unlock(&em_tree
->lock
);
3452 free_extent_map(em
);
3456 ret
= btrfs_make_block_group(trans
, extent_root
, 0, type
,
3457 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
3462 for (i
= 0; i
< map
->num_stripes
; ++i
) {
3463 struct btrfs_device
*device
;
3466 device
= map
->stripes
[i
].dev
;
3467 dev_offset
= map
->stripes
[i
].physical
;
3469 ret
= btrfs_alloc_dev_extent(trans
, device
,
3470 info
->chunk_root
->root_key
.objectid
,
3471 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
3472 start
, dev_offset
, stripe_size
);
3474 btrfs_abort_transaction(trans
, extent_root
, ret
);
3479 kfree(devices_info
);
3484 kfree(devices_info
);
3488 static int __finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
3489 struct btrfs_root
*extent_root
,
3490 struct map_lookup
*map
, u64 chunk_offset
,
3491 u64 chunk_size
, u64 stripe_size
)
3494 struct btrfs_key key
;
3495 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
3496 struct btrfs_device
*device
;
3497 struct btrfs_chunk
*chunk
;
3498 struct btrfs_stripe
*stripe
;
3499 size_t item_size
= btrfs_chunk_item_size(map
->num_stripes
);
3503 chunk
= kzalloc(item_size
, GFP_NOFS
);
3508 while (index
< map
->num_stripes
) {
3509 device
= map
->stripes
[index
].dev
;
3510 device
->bytes_used
+= stripe_size
;
3511 ret
= btrfs_update_device(trans
, device
);
3517 spin_lock(&extent_root
->fs_info
->free_chunk_lock
);
3518 extent_root
->fs_info
->free_chunk_space
-= (stripe_size
*
3520 spin_unlock(&extent_root
->fs_info
->free_chunk_lock
);
3523 stripe
= &chunk
->stripe
;
3524 while (index
< map
->num_stripes
) {
3525 device
= map
->stripes
[index
].dev
;
3526 dev_offset
= map
->stripes
[index
].physical
;
3528 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
3529 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
3530 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
3535 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
3536 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
3537 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
3538 btrfs_set_stack_chunk_type(chunk
, map
->type
);
3539 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
3540 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
3541 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
3542 btrfs_set_stack_chunk_sector_size(chunk
, extent_root
->sectorsize
);
3543 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
3545 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3546 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3547 key
.offset
= chunk_offset
;
3549 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
3551 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3553 * TODO: Cleanup of inserted chunk root in case of
3556 ret
= btrfs_add_system_chunk(chunk_root
, &key
, chunk
,
3566 * Chunk allocation falls into two parts. The first part does works
3567 * that make the new allocated chunk useable, but not do any operation
3568 * that modifies the chunk tree. The second part does the works that
3569 * require modifying the chunk tree. This division is important for the
3570 * bootstrap process of adding storage to a seed btrfs.
3572 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
3573 struct btrfs_root
*extent_root
, u64 type
)
3578 struct map_lookup
*map
;
3579 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
3582 ret
= find_next_chunk(chunk_root
, BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
3587 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &map
, &chunk_size
,
3588 &stripe_size
, chunk_offset
, type
);
3592 ret
= __finish_chunk_alloc(trans
, extent_root
, map
, chunk_offset
,
3593 chunk_size
, stripe_size
);
3599 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
3600 struct btrfs_root
*root
,
3601 struct btrfs_device
*device
)
3604 u64 sys_chunk_offset
;
3608 u64 sys_stripe_size
;
3610 struct map_lookup
*map
;
3611 struct map_lookup
*sys_map
;
3612 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3613 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
3616 ret
= find_next_chunk(fs_info
->chunk_root
,
3617 BTRFS_FIRST_CHUNK_TREE_OBJECTID
, &chunk_offset
);
3621 alloc_profile
= BTRFS_BLOCK_GROUP_METADATA
|
3622 fs_info
->avail_metadata_alloc_bits
;
3623 alloc_profile
= btrfs_reduce_alloc_profile(root
, alloc_profile
);
3625 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &map
, &chunk_size
,
3626 &stripe_size
, chunk_offset
, alloc_profile
);
3630 sys_chunk_offset
= chunk_offset
+ chunk_size
;
3632 alloc_profile
= BTRFS_BLOCK_GROUP_SYSTEM
|
3633 fs_info
->avail_system_alloc_bits
;
3634 alloc_profile
= btrfs_reduce_alloc_profile(root
, alloc_profile
);
3636 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &sys_map
,
3637 &sys_chunk_size
, &sys_stripe_size
,
3638 sys_chunk_offset
, alloc_profile
);
3640 btrfs_abort_transaction(trans
, root
, ret
);
3644 ret
= btrfs_add_device(trans
, fs_info
->chunk_root
, device
);
3646 btrfs_abort_transaction(trans
, root
, ret
);
3651 * Modifying chunk tree needs allocating new blocks from both
3652 * system block group and metadata block group. So we only can
3653 * do operations require modifying the chunk tree after both
3654 * block groups were created.
3656 ret
= __finish_chunk_alloc(trans
, extent_root
, map
, chunk_offset
,
3657 chunk_size
, stripe_size
);
3659 btrfs_abort_transaction(trans
, root
, ret
);
3663 ret
= __finish_chunk_alloc(trans
, extent_root
, sys_map
,
3664 sys_chunk_offset
, sys_chunk_size
,
3667 btrfs_abort_transaction(trans
, root
, ret
);
3674 int btrfs_chunk_readonly(struct btrfs_root
*root
, u64 chunk_offset
)
3676 struct extent_map
*em
;
3677 struct map_lookup
*map
;
3678 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
3682 read_lock(&map_tree
->map_tree
.lock
);
3683 em
= lookup_extent_mapping(&map_tree
->map_tree
, chunk_offset
, 1);
3684 read_unlock(&map_tree
->map_tree
.lock
);
3688 if (btrfs_test_opt(root
, DEGRADED
)) {
3689 free_extent_map(em
);
3693 map
= (struct map_lookup
*)em
->bdev
;
3694 for (i
= 0; i
< map
->num_stripes
; i
++) {
3695 if (!map
->stripes
[i
].dev
->writeable
) {
3700 free_extent_map(em
);
3704 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
3706 extent_map_tree_init(&tree
->map_tree
);
3709 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
3711 struct extent_map
*em
;
3714 write_lock(&tree
->map_tree
.lock
);
3715 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
3717 remove_extent_mapping(&tree
->map_tree
, em
);
3718 write_unlock(&tree
->map_tree
.lock
);
3723 free_extent_map(em
);
3724 /* once for the tree */
3725 free_extent_map(em
);
3729 int btrfs_num_copies(struct btrfs_mapping_tree
*map_tree
, u64 logical
, u64 len
)
3731 struct extent_map
*em
;
3732 struct map_lookup
*map
;
3733 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
3736 read_lock(&em_tree
->lock
);
3737 em
= lookup_extent_mapping(em_tree
, logical
, len
);
3738 read_unlock(&em_tree
->lock
);
3741 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
3742 map
= (struct map_lookup
*)em
->bdev
;
3743 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
3744 ret
= map
->num_stripes
;
3745 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
3746 ret
= map
->sub_stripes
;
3749 free_extent_map(em
);
3753 static int find_live_mirror(struct map_lookup
*map
, int first
, int num
,
3757 if (map
->stripes
[optimal
].dev
->bdev
)
3759 for (i
= first
; i
< first
+ num
; i
++) {
3760 if (map
->stripes
[i
].dev
->bdev
)
3763 /* we couldn't find one that doesn't fail. Just return something
3764 * and the io error handling code will clean up eventually
3769 static int __btrfs_map_block(struct btrfs_mapping_tree
*map_tree
, int rw
,
3770 u64 logical
, u64
*length
,
3771 struct btrfs_bio
**bbio_ret
,
3774 struct extent_map
*em
;
3775 struct map_lookup
*map
;
3776 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
3779 u64 stripe_end_offset
;
3788 struct btrfs_bio
*bbio
= NULL
;
3790 read_lock(&em_tree
->lock
);
3791 em
= lookup_extent_mapping(em_tree
, logical
, *length
);
3792 read_unlock(&em_tree
->lock
);
3795 printk(KERN_CRIT
"btrfs: unable to find logical %llu len %llu\n",
3796 (unsigned long long)logical
,
3797 (unsigned long long)*length
);
3801 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
3802 map
= (struct map_lookup
*)em
->bdev
;
3803 offset
= logical
- em
->start
;
3805 if (mirror_num
> map
->num_stripes
)
3810 * stripe_nr counts the total number of stripes we have to stride
3811 * to get to this block
3813 do_div(stripe_nr
, map
->stripe_len
);
3815 stripe_offset
= stripe_nr
* map
->stripe_len
;
3816 BUG_ON(offset
< stripe_offset
);
3818 /* stripe_offset is the offset of this block in its stripe*/
3819 stripe_offset
= offset
- stripe_offset
;
3821 if (rw
& REQ_DISCARD
)
3822 *length
= min_t(u64
, em
->len
- offset
, *length
);
3823 else if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
3824 /* we limit the length of each bio to what fits in a stripe */
3825 *length
= min_t(u64
, em
->len
- offset
,
3826 map
->stripe_len
- stripe_offset
);
3828 *length
= em
->len
- offset
;
3836 stripe_nr_orig
= stripe_nr
;
3837 stripe_nr_end
= (offset
+ *length
+ map
->stripe_len
- 1) &
3838 (~(map
->stripe_len
- 1));
3839 do_div(stripe_nr_end
, map
->stripe_len
);
3840 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
3842 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
3843 if (rw
& REQ_DISCARD
)
3844 num_stripes
= min_t(u64
, map
->num_stripes
,
3845 stripe_nr_end
- stripe_nr_orig
);
3846 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
3847 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
3848 if (rw
& (REQ_WRITE
| REQ_DISCARD
))
3849 num_stripes
= map
->num_stripes
;
3850 else if (mirror_num
)
3851 stripe_index
= mirror_num
- 1;
3853 stripe_index
= find_live_mirror(map
, 0,
3855 current
->pid
% map
->num_stripes
);
3856 mirror_num
= stripe_index
+ 1;
3859 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
3860 if (rw
& (REQ_WRITE
| REQ_DISCARD
)) {
3861 num_stripes
= map
->num_stripes
;
3862 } else if (mirror_num
) {
3863 stripe_index
= mirror_num
- 1;
3868 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
3869 int factor
= map
->num_stripes
/ map
->sub_stripes
;
3871 stripe_index
= do_div(stripe_nr
, factor
);
3872 stripe_index
*= map
->sub_stripes
;
3875 num_stripes
= map
->sub_stripes
;
3876 else if (rw
& REQ_DISCARD
)
3877 num_stripes
= min_t(u64
, map
->sub_stripes
*
3878 (stripe_nr_end
- stripe_nr_orig
),
3880 else if (mirror_num
)
3881 stripe_index
+= mirror_num
- 1;
3883 int old_stripe_index
= stripe_index
;
3884 stripe_index
= find_live_mirror(map
, stripe_index
,
3885 map
->sub_stripes
, stripe_index
+
3886 current
->pid
% map
->sub_stripes
);
3887 mirror_num
= stripe_index
- old_stripe_index
+ 1;
3891 * after this do_div call, stripe_nr is the number of stripes
3892 * on this device we have to walk to find the data, and
3893 * stripe_index is the number of our device in the stripe array
3895 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
3896 mirror_num
= stripe_index
+ 1;
3898 BUG_ON(stripe_index
>= map
->num_stripes
);
3900 bbio
= kzalloc(btrfs_bio_size(num_stripes
), GFP_NOFS
);
3905 atomic_set(&bbio
->error
, 0);
3907 if (rw
& REQ_DISCARD
) {
3909 int sub_stripes
= 0;
3910 u64 stripes_per_dev
= 0;
3911 u32 remaining_stripes
= 0;
3912 u32 last_stripe
= 0;
3915 (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID10
)) {
3916 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
3919 sub_stripes
= map
->sub_stripes
;
3921 factor
= map
->num_stripes
/ sub_stripes
;
3922 stripes_per_dev
= div_u64_rem(stripe_nr_end
-
3925 &remaining_stripes
);
3926 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
3927 last_stripe
*= sub_stripes
;
3930 for (i
= 0; i
< num_stripes
; i
++) {
3931 bbio
->stripes
[i
].physical
=
3932 map
->stripes
[stripe_index
].physical
+
3933 stripe_offset
+ stripe_nr
* map
->stripe_len
;
3934 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
3936 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
3937 BTRFS_BLOCK_GROUP_RAID10
)) {
3938 bbio
->stripes
[i
].length
= stripes_per_dev
*
3941 if (i
/ sub_stripes
< remaining_stripes
)
3942 bbio
->stripes
[i
].length
+=
3946 * Special for the first stripe and
3949 * |-------|...|-------|
3953 if (i
< sub_stripes
)
3954 bbio
->stripes
[i
].length
-=
3957 if (stripe_index
>= last_stripe
&&
3958 stripe_index
<= (last_stripe
+
3960 bbio
->stripes
[i
].length
-=
3963 if (i
== sub_stripes
- 1)
3966 bbio
->stripes
[i
].length
= *length
;
3969 if (stripe_index
== map
->num_stripes
) {
3970 /* This could only happen for RAID0/10 */
3976 for (i
= 0; i
< num_stripes
; i
++) {
3977 bbio
->stripes
[i
].physical
=
3978 map
->stripes
[stripe_index
].physical
+
3980 stripe_nr
* map
->stripe_len
;
3981 bbio
->stripes
[i
].dev
=
3982 map
->stripes
[stripe_index
].dev
;
3987 if (rw
& REQ_WRITE
) {
3988 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
3989 BTRFS_BLOCK_GROUP_RAID10
|
3990 BTRFS_BLOCK_GROUP_DUP
)) {
3996 bbio
->num_stripes
= num_stripes
;
3997 bbio
->max_errors
= max_errors
;
3998 bbio
->mirror_num
= mirror_num
;
4000 free_extent_map(em
);
4004 int btrfs_map_block(struct btrfs_mapping_tree
*map_tree
, int rw
,
4005 u64 logical
, u64
*length
,
4006 struct btrfs_bio
**bbio_ret
, int mirror_num
)
4008 return __btrfs_map_block(map_tree
, rw
, logical
, length
, bbio_ret
,
4012 int btrfs_rmap_block(struct btrfs_mapping_tree
*map_tree
,
4013 u64 chunk_start
, u64 physical
, u64 devid
,
4014 u64
**logical
, int *naddrs
, int *stripe_len
)
4016 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4017 struct extent_map
*em
;
4018 struct map_lookup
*map
;
4025 read_lock(&em_tree
->lock
);
4026 em
= lookup_extent_mapping(em_tree
, chunk_start
, 1);
4027 read_unlock(&em_tree
->lock
);
4029 BUG_ON(!em
|| em
->start
!= chunk_start
);
4030 map
= (struct map_lookup
*)em
->bdev
;
4033 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
4034 do_div(length
, map
->num_stripes
/ map
->sub_stripes
);
4035 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
4036 do_div(length
, map
->num_stripes
);
4038 buf
= kzalloc(sizeof(u64
) * map
->num_stripes
, GFP_NOFS
);
4039 BUG_ON(!buf
); /* -ENOMEM */
4041 for (i
= 0; i
< map
->num_stripes
; i
++) {
4042 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
4044 if (map
->stripes
[i
].physical
> physical
||
4045 map
->stripes
[i
].physical
+ length
<= physical
)
4048 stripe_nr
= physical
- map
->stripes
[i
].physical
;
4049 do_div(stripe_nr
, map
->stripe_len
);
4051 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
4052 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
4053 do_div(stripe_nr
, map
->sub_stripes
);
4054 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
4055 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
4057 bytenr
= chunk_start
+ stripe_nr
* map
->stripe_len
;
4058 WARN_ON(nr
>= map
->num_stripes
);
4059 for (j
= 0; j
< nr
; j
++) {
4060 if (buf
[j
] == bytenr
)
4064 WARN_ON(nr
>= map
->num_stripes
);
4071 *stripe_len
= map
->stripe_len
;
4073 free_extent_map(em
);
4077 static void *merge_stripe_index_into_bio_private(void *bi_private
,
4078 unsigned int stripe_index
)
4081 * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
4083 * The alternative solution (instead of stealing bits from the
4084 * pointer) would be to allocate an intermediate structure
4085 * that contains the old private pointer plus the stripe_index.
4087 BUG_ON((((uintptr_t)bi_private
) & 3) != 0);
4088 BUG_ON(stripe_index
> 3);
4089 return (void *)(((uintptr_t)bi_private
) | stripe_index
);
4092 static struct btrfs_bio
*extract_bbio_from_bio_private(void *bi_private
)
4094 return (struct btrfs_bio
*)(((uintptr_t)bi_private
) & ~((uintptr_t)3));
4097 static unsigned int extract_stripe_index_from_bio_private(void *bi_private
)
4099 return (unsigned int)((uintptr_t)bi_private
) & 3;
4102 static void btrfs_end_bio(struct bio
*bio
, int err
)
4104 struct btrfs_bio
*bbio
= extract_bbio_from_bio_private(bio
->bi_private
);
4105 int is_orig_bio
= 0;
4108 atomic_inc(&bbio
->error
);
4109 if (err
== -EIO
|| err
== -EREMOTEIO
) {
4110 unsigned int stripe_index
=
4111 extract_stripe_index_from_bio_private(
4113 struct btrfs_device
*dev
;
4115 BUG_ON(stripe_index
>= bbio
->num_stripes
);
4116 dev
= bbio
->stripes
[stripe_index
].dev
;
4118 if (bio
->bi_rw
& WRITE
)
4119 btrfs_dev_stat_inc(dev
,
4120 BTRFS_DEV_STAT_WRITE_ERRS
);
4122 btrfs_dev_stat_inc(dev
,
4123 BTRFS_DEV_STAT_READ_ERRS
);
4124 if ((bio
->bi_rw
& WRITE_FLUSH
) == WRITE_FLUSH
)
4125 btrfs_dev_stat_inc(dev
,
4126 BTRFS_DEV_STAT_FLUSH_ERRS
);
4127 btrfs_dev_stat_print_on_error(dev
);
4132 if (bio
== bbio
->orig_bio
)
4135 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
4138 bio
= bbio
->orig_bio
;
4140 bio
->bi_private
= bbio
->private;
4141 bio
->bi_end_io
= bbio
->end_io
;
4142 bio
->bi_bdev
= (struct block_device
*)
4143 (unsigned long)bbio
->mirror_num
;
4144 /* only send an error to the higher layers if it is
4145 * beyond the tolerance of the multi-bio
4147 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
4151 * this bio is actually up to date, we didn't
4152 * go over the max number of errors
4154 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
4159 bio_endio(bio
, err
);
4160 } else if (!is_orig_bio
) {
4165 struct async_sched
{
4168 struct btrfs_fs_info
*info
;
4169 struct btrfs_work work
;
4173 * see run_scheduled_bios for a description of why bios are collected for
4176 * This will add one bio to the pending list for a device and make sure
4177 * the work struct is scheduled.
4179 static noinline
void schedule_bio(struct btrfs_root
*root
,
4180 struct btrfs_device
*device
,
4181 int rw
, struct bio
*bio
)
4183 int should_queue
= 1;
4184 struct btrfs_pending_bios
*pending_bios
;
4186 /* don't bother with additional async steps for reads, right now */
4187 if (!(rw
& REQ_WRITE
)) {
4189 btrfsic_submit_bio(rw
, bio
);
4195 * nr_async_bios allows us to reliably return congestion to the
4196 * higher layers. Otherwise, the async bio makes it appear we have
4197 * made progress against dirty pages when we've really just put it
4198 * on a queue for later
4200 atomic_inc(&root
->fs_info
->nr_async_bios
);
4201 WARN_ON(bio
->bi_next
);
4202 bio
->bi_next
= NULL
;
4205 spin_lock(&device
->io_lock
);
4206 if (bio
->bi_rw
& REQ_SYNC
)
4207 pending_bios
= &device
->pending_sync_bios
;
4209 pending_bios
= &device
->pending_bios
;
4211 if (pending_bios
->tail
)
4212 pending_bios
->tail
->bi_next
= bio
;
4214 pending_bios
->tail
= bio
;
4215 if (!pending_bios
->head
)
4216 pending_bios
->head
= bio
;
4217 if (device
->running_pending
)
4220 spin_unlock(&device
->io_lock
);
4223 btrfs_queue_worker(&root
->fs_info
->submit_workers
,
4227 static int bio_size_ok(struct block_device
*bdev
, struct bio
*bio
,
4230 struct bio_vec
*prev
;
4231 struct request_queue
*q
= bdev_get_queue(bdev
);
4232 unsigned short max_sectors
= queue_max_sectors(q
);
4233 struct bvec_merge_data bvm
= {
4235 .bi_sector
= sector
,
4236 .bi_rw
= bio
->bi_rw
,
4239 if (bio
->bi_vcnt
== 0) {
4244 prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
4245 if ((bio
->bi_size
>> 9) > max_sectors
)
4248 if (!q
->merge_bvec_fn
)
4251 bvm
.bi_size
= bio
->bi_size
- prev
->bv_len
;
4252 if (q
->merge_bvec_fn(q
, &bvm
, prev
) < prev
->bv_len
)
4257 static void submit_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
4258 struct bio
*bio
, u64 physical
, int dev_nr
,
4261 struct btrfs_device
*dev
= bbio
->stripes
[dev_nr
].dev
;
4263 bio
->bi_private
= bbio
;
4264 bio
->bi_private
= merge_stripe_index_into_bio_private(
4265 bio
->bi_private
, (unsigned int)dev_nr
);
4266 bio
->bi_end_io
= btrfs_end_bio
;
4267 bio
->bi_sector
= physical
>> 9;
4270 struct rcu_string
*name
;
4273 name
= rcu_dereference(dev
->name
);
4274 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
4275 "(%s id %llu), size=%u\n", rw
,
4276 (u64
)bio
->bi_sector
, (u_long
)dev
->bdev
->bd_dev
,
4277 name
->str
, dev
->devid
, bio
->bi_size
);
4281 bio
->bi_bdev
= dev
->bdev
;
4283 schedule_bio(root
, dev
, rw
, bio
);
4285 btrfsic_submit_bio(rw
, bio
);
4288 static int breakup_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
4289 struct bio
*first_bio
, struct btrfs_device
*dev
,
4290 int dev_nr
, int rw
, int async
)
4292 struct bio_vec
*bvec
= first_bio
->bi_io_vec
;
4294 int nr_vecs
= bio_get_nr_vecs(dev
->bdev
);
4295 u64 physical
= bbio
->stripes
[dev_nr
].physical
;
4298 bio
= btrfs_bio_alloc(dev
->bdev
, physical
>> 9, nr_vecs
, GFP_NOFS
);
4302 while (bvec
<= (first_bio
->bi_io_vec
+ first_bio
->bi_vcnt
- 1)) {
4303 if (bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
4304 bvec
->bv_offset
) < bvec
->bv_len
) {
4305 u64 len
= bio
->bi_size
;
4307 atomic_inc(&bbio
->stripes_pending
);
4308 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
,
4316 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
, rw
, async
);
4320 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
4322 atomic_inc(&bbio
->error
);
4323 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
4324 bio
->bi_private
= bbio
->private;
4325 bio
->bi_end_io
= bbio
->end_io
;
4326 bio
->bi_bdev
= (struct block_device
*)
4327 (unsigned long)bbio
->mirror_num
;
4328 bio
->bi_sector
= logical
>> 9;
4330 bio_endio(bio
, -EIO
);
4334 int btrfs_map_bio(struct btrfs_root
*root
, int rw
, struct bio
*bio
,
4335 int mirror_num
, int async_submit
)
4337 struct btrfs_mapping_tree
*map_tree
;
4338 struct btrfs_device
*dev
;
4339 struct bio
*first_bio
= bio
;
4340 u64 logical
= (u64
)bio
->bi_sector
<< 9;
4346 struct btrfs_bio
*bbio
= NULL
;
4348 length
= bio
->bi_size
;
4349 map_tree
= &root
->fs_info
->mapping_tree
;
4350 map_length
= length
;
4352 ret
= btrfs_map_block(map_tree
, rw
, logical
, &map_length
, &bbio
,
4354 if (ret
) /* -ENOMEM */
4357 total_devs
= bbio
->num_stripes
;
4358 if (map_length
< length
) {
4359 printk(KERN_CRIT
"btrfs: mapping failed logical %llu bio len %llu "
4360 "len %llu\n", (unsigned long long)logical
,
4361 (unsigned long long)length
,
4362 (unsigned long long)map_length
);
4366 bbio
->orig_bio
= first_bio
;
4367 bbio
->private = first_bio
->bi_private
;
4368 bbio
->end_io
= first_bio
->bi_end_io
;
4369 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
4371 while (dev_nr
< total_devs
) {
4372 dev
= bbio
->stripes
[dev_nr
].dev
;
4373 if (!dev
|| !dev
->bdev
|| (rw
& WRITE
&& !dev
->writeable
)) {
4374 bbio_error(bbio
, first_bio
, logical
);
4380 * Check and see if we're ok with this bio based on it's size
4381 * and offset with the given device.
4383 if (!bio_size_ok(dev
->bdev
, first_bio
,
4384 bbio
->stripes
[dev_nr
].physical
>> 9)) {
4385 ret
= breakup_stripe_bio(root
, bbio
, first_bio
, dev
,
4386 dev_nr
, rw
, async_submit
);
4392 if (dev_nr
< total_devs
- 1) {
4393 bio
= bio_clone(first_bio
, GFP_NOFS
);
4394 BUG_ON(!bio
); /* -ENOMEM */
4399 submit_stripe_bio(root
, bbio
, bio
,
4400 bbio
->stripes
[dev_nr
].physical
, dev_nr
, rw
,
4407 struct btrfs_device
*btrfs_find_device(struct btrfs_root
*root
, u64 devid
,
4410 struct btrfs_device
*device
;
4411 struct btrfs_fs_devices
*cur_devices
;
4413 cur_devices
= root
->fs_info
->fs_devices
;
4414 while (cur_devices
) {
4416 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
4417 device
= __find_device(&cur_devices
->devices
,
4422 cur_devices
= cur_devices
->seed
;
4427 static struct btrfs_device
*add_missing_dev(struct btrfs_root
*root
,
4428 u64 devid
, u8
*dev_uuid
)
4430 struct btrfs_device
*device
;
4431 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
4433 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
4436 list_add(&device
->dev_list
,
4437 &fs_devices
->devices
);
4438 device
->dev_root
= root
->fs_info
->dev_root
;
4439 device
->devid
= devid
;
4440 device
->work
.func
= pending_bios_fn
;
4441 device
->fs_devices
= fs_devices
;
4442 device
->missing
= 1;
4443 fs_devices
->num_devices
++;
4444 fs_devices
->missing_devices
++;
4445 spin_lock_init(&device
->io_lock
);
4446 INIT_LIST_HEAD(&device
->dev_alloc_list
);
4447 memcpy(device
->uuid
, dev_uuid
, BTRFS_UUID_SIZE
);
4451 static int read_one_chunk(struct btrfs_root
*root
, struct btrfs_key
*key
,
4452 struct extent_buffer
*leaf
,
4453 struct btrfs_chunk
*chunk
)
4455 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
4456 struct map_lookup
*map
;
4457 struct extent_map
*em
;
4461 u8 uuid
[BTRFS_UUID_SIZE
];
4466 logical
= key
->offset
;
4467 length
= btrfs_chunk_length(leaf
, chunk
);
4469 read_lock(&map_tree
->map_tree
.lock
);
4470 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
4471 read_unlock(&map_tree
->map_tree
.lock
);
4473 /* already mapped? */
4474 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
4475 free_extent_map(em
);
4478 free_extent_map(em
);
4481 em
= alloc_extent_map();
4484 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
4485 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
4487 free_extent_map(em
);
4491 em
->bdev
= (struct block_device
*)map
;
4492 em
->start
= logical
;
4494 em
->block_start
= 0;
4495 em
->block_len
= em
->len
;
4497 map
->num_stripes
= num_stripes
;
4498 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
4499 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
4500 map
->sector_size
= btrfs_chunk_sector_size(leaf
, chunk
);
4501 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
4502 map
->type
= btrfs_chunk_type(leaf
, chunk
);
4503 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
4504 for (i
= 0; i
< num_stripes
; i
++) {
4505 map
->stripes
[i
].physical
=
4506 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
4507 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
4508 read_extent_buffer(leaf
, uuid
, (unsigned long)
4509 btrfs_stripe_dev_uuid_nr(chunk
, i
),
4511 map
->stripes
[i
].dev
= btrfs_find_device(root
, devid
, uuid
,
4513 if (!map
->stripes
[i
].dev
&& !btrfs_test_opt(root
, DEGRADED
)) {
4515 free_extent_map(em
);
4518 if (!map
->stripes
[i
].dev
) {
4519 map
->stripes
[i
].dev
=
4520 add_missing_dev(root
, devid
, uuid
);
4521 if (!map
->stripes
[i
].dev
) {
4523 free_extent_map(em
);
4527 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
4530 write_lock(&map_tree
->map_tree
.lock
);
4531 ret
= add_extent_mapping(&map_tree
->map_tree
, em
);
4532 write_unlock(&map_tree
->map_tree
.lock
);
4533 BUG_ON(ret
); /* Tree corruption */
4534 free_extent_map(em
);
4539 static void fill_device_from_item(struct extent_buffer
*leaf
,
4540 struct btrfs_dev_item
*dev_item
,
4541 struct btrfs_device
*device
)
4545 device
->devid
= btrfs_device_id(leaf
, dev_item
);
4546 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
4547 device
->total_bytes
= device
->disk_total_bytes
;
4548 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
4549 device
->type
= btrfs_device_type(leaf
, dev_item
);
4550 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
4551 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
4552 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
4554 ptr
= (unsigned long)btrfs_device_uuid(dev_item
);
4555 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
4558 static int open_seed_devices(struct btrfs_root
*root
, u8
*fsid
)
4560 struct btrfs_fs_devices
*fs_devices
;
4563 BUG_ON(!mutex_is_locked(&uuid_mutex
));
4565 fs_devices
= root
->fs_info
->fs_devices
->seed
;
4566 while (fs_devices
) {
4567 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
4571 fs_devices
= fs_devices
->seed
;
4574 fs_devices
= find_fsid(fsid
);
4580 fs_devices
= clone_fs_devices(fs_devices
);
4581 if (IS_ERR(fs_devices
)) {
4582 ret
= PTR_ERR(fs_devices
);
4586 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
4587 root
->fs_info
->bdev_holder
);
4589 free_fs_devices(fs_devices
);
4593 if (!fs_devices
->seeding
) {
4594 __btrfs_close_devices(fs_devices
);
4595 free_fs_devices(fs_devices
);
4600 fs_devices
->seed
= root
->fs_info
->fs_devices
->seed
;
4601 root
->fs_info
->fs_devices
->seed
= fs_devices
;
4606 static int read_one_dev(struct btrfs_root
*root
,
4607 struct extent_buffer
*leaf
,
4608 struct btrfs_dev_item
*dev_item
)
4610 struct btrfs_device
*device
;
4613 u8 fs_uuid
[BTRFS_UUID_SIZE
];
4614 u8 dev_uuid
[BTRFS_UUID_SIZE
];
4616 devid
= btrfs_device_id(leaf
, dev_item
);
4617 read_extent_buffer(leaf
, dev_uuid
,
4618 (unsigned long)btrfs_device_uuid(dev_item
),
4620 read_extent_buffer(leaf
, fs_uuid
,
4621 (unsigned long)btrfs_device_fsid(dev_item
),
4624 if (memcmp(fs_uuid
, root
->fs_info
->fsid
, BTRFS_UUID_SIZE
)) {
4625 ret
= open_seed_devices(root
, fs_uuid
);
4626 if (ret
&& !btrfs_test_opt(root
, DEGRADED
))
4630 device
= btrfs_find_device(root
, devid
, dev_uuid
, fs_uuid
);
4631 if (!device
|| !device
->bdev
) {
4632 if (!btrfs_test_opt(root
, DEGRADED
))
4636 printk(KERN_WARNING
"warning devid %llu missing\n",
4637 (unsigned long long)devid
);
4638 device
= add_missing_dev(root
, devid
, dev_uuid
);
4641 } else if (!device
->missing
) {
4643 * this happens when a device that was properly setup
4644 * in the device info lists suddenly goes bad.
4645 * device->bdev is NULL, and so we have to set
4646 * device->missing to one here
4648 root
->fs_info
->fs_devices
->missing_devices
++;
4649 device
->missing
= 1;
4653 if (device
->fs_devices
!= root
->fs_info
->fs_devices
) {
4654 BUG_ON(device
->writeable
);
4655 if (device
->generation
!=
4656 btrfs_device_generation(leaf
, dev_item
))
4660 fill_device_from_item(leaf
, dev_item
, device
);
4661 device
->dev_root
= root
->fs_info
->dev_root
;
4662 device
->in_fs_metadata
= 1;
4663 if (device
->writeable
) {
4664 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
4665 spin_lock(&root
->fs_info
->free_chunk_lock
);
4666 root
->fs_info
->free_chunk_space
+= device
->total_bytes
-
4668 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4674 int btrfs_read_sys_array(struct btrfs_root
*root
)
4676 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
4677 struct extent_buffer
*sb
;
4678 struct btrfs_disk_key
*disk_key
;
4679 struct btrfs_chunk
*chunk
;
4681 unsigned long sb_ptr
;
4687 struct btrfs_key key
;
4689 sb
= btrfs_find_create_tree_block(root
, BTRFS_SUPER_INFO_OFFSET
,
4690 BTRFS_SUPER_INFO_SIZE
);
4693 btrfs_set_buffer_uptodate(sb
);
4694 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
4696 * The sb extent buffer is artifical and just used to read the system array.
4697 * btrfs_set_buffer_uptodate() call does not properly mark all it's
4698 * pages up-to-date when the page is larger: extent does not cover the
4699 * whole page and consequently check_page_uptodate does not find all
4700 * the page's extents up-to-date (the hole beyond sb),
4701 * write_extent_buffer then triggers a WARN_ON.
4703 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
4704 * but sb spans only this function. Add an explicit SetPageUptodate call
4705 * to silence the warning eg. on PowerPC 64.
4707 if (PAGE_CACHE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
4708 SetPageUptodate(sb
->pages
[0]);
4710 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
4711 array_size
= btrfs_super_sys_array_size(super_copy
);
4713 ptr
= super_copy
->sys_chunk_array
;
4714 sb_ptr
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
4717 while (cur
< array_size
) {
4718 disk_key
= (struct btrfs_disk_key
*)ptr
;
4719 btrfs_disk_key_to_cpu(&key
, disk_key
);
4721 len
= sizeof(*disk_key
); ptr
+= len
;
4725 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
4726 chunk
= (struct btrfs_chunk
*)sb_ptr
;
4727 ret
= read_one_chunk(root
, &key
, sb
, chunk
);
4730 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
4731 len
= btrfs_chunk_item_size(num_stripes
);
4740 free_extent_buffer(sb
);
4744 int btrfs_read_chunk_tree(struct btrfs_root
*root
)
4746 struct btrfs_path
*path
;
4747 struct extent_buffer
*leaf
;
4748 struct btrfs_key key
;
4749 struct btrfs_key found_key
;
4753 root
= root
->fs_info
->chunk_root
;
4755 path
= btrfs_alloc_path();
4759 mutex_lock(&uuid_mutex
);
4762 /* first we search for all of the device items, and then we
4763 * read in all of the chunk items. This way we can create chunk
4764 * mappings that reference all of the devices that are afound
4766 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
4770 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4774 leaf
= path
->nodes
[0];
4775 slot
= path
->slots
[0];
4776 if (slot
>= btrfs_header_nritems(leaf
)) {
4777 ret
= btrfs_next_leaf(root
, path
);
4784 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
4785 if (key
.objectid
== BTRFS_DEV_ITEMS_OBJECTID
) {
4786 if (found_key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
)
4788 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
4789 struct btrfs_dev_item
*dev_item
;
4790 dev_item
= btrfs_item_ptr(leaf
, slot
,
4791 struct btrfs_dev_item
);
4792 ret
= read_one_dev(root
, leaf
, dev_item
);
4796 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
4797 struct btrfs_chunk
*chunk
;
4798 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
4799 ret
= read_one_chunk(root
, &found_key
, leaf
, chunk
);
4805 if (key
.objectid
== BTRFS_DEV_ITEMS_OBJECTID
) {
4807 btrfs_release_path(path
);
4812 unlock_chunks(root
);
4813 mutex_unlock(&uuid_mutex
);
4815 btrfs_free_path(path
);
4819 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
4823 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
4824 btrfs_dev_stat_reset(dev
, i
);
4827 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
4829 struct btrfs_key key
;
4830 struct btrfs_key found_key
;
4831 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
4832 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
4833 struct extent_buffer
*eb
;
4836 struct btrfs_device
*device
;
4837 struct btrfs_path
*path
= NULL
;
4840 path
= btrfs_alloc_path();
4846 mutex_lock(&fs_devices
->device_list_mutex
);
4847 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
4849 struct btrfs_dev_stats_item
*ptr
;
4852 key
.type
= BTRFS_DEV_STATS_KEY
;
4853 key
.offset
= device
->devid
;
4854 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
4856 __btrfs_reset_dev_stats(device
);
4857 device
->dev_stats_valid
= 1;
4858 btrfs_release_path(path
);
4861 slot
= path
->slots
[0];
4862 eb
= path
->nodes
[0];
4863 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
4864 item_size
= btrfs_item_size_nr(eb
, slot
);
4866 ptr
= btrfs_item_ptr(eb
, slot
,
4867 struct btrfs_dev_stats_item
);
4869 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
4870 if (item_size
>= (1 + i
) * sizeof(__le64
))
4871 btrfs_dev_stat_set(device
, i
,
4872 btrfs_dev_stats_value(eb
, ptr
, i
));
4874 btrfs_dev_stat_reset(device
, i
);
4877 device
->dev_stats_valid
= 1;
4878 btrfs_dev_stat_print_on_load(device
);
4879 btrfs_release_path(path
);
4881 mutex_unlock(&fs_devices
->device_list_mutex
);
4884 btrfs_free_path(path
);
4885 return ret
< 0 ? ret
: 0;
4888 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
4889 struct btrfs_root
*dev_root
,
4890 struct btrfs_device
*device
)
4892 struct btrfs_path
*path
;
4893 struct btrfs_key key
;
4894 struct extent_buffer
*eb
;
4895 struct btrfs_dev_stats_item
*ptr
;
4900 key
.type
= BTRFS_DEV_STATS_KEY
;
4901 key
.offset
= device
->devid
;
4903 path
= btrfs_alloc_path();
4905 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
4907 printk_in_rcu(KERN_WARNING
"btrfs: error %d while searching for dev_stats item for device %s!\n",
4908 ret
, rcu_str_deref(device
->name
));
4913 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
4914 /* need to delete old one and insert a new one */
4915 ret
= btrfs_del_item(trans
, dev_root
, path
);
4917 printk_in_rcu(KERN_WARNING
"btrfs: delete too small dev_stats item for device %s failed %d!\n",
4918 rcu_str_deref(device
->name
), ret
);
4925 /* need to insert a new item */
4926 btrfs_release_path(path
);
4927 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
4928 &key
, sizeof(*ptr
));
4930 printk_in_rcu(KERN_WARNING
"btrfs: insert dev_stats item for device %s failed %d!\n",
4931 rcu_str_deref(device
->name
), ret
);
4936 eb
= path
->nodes
[0];
4937 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
4938 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
4939 btrfs_set_dev_stats_value(eb
, ptr
, i
,
4940 btrfs_dev_stat_read(device
, i
));
4941 btrfs_mark_buffer_dirty(eb
);
4944 btrfs_free_path(path
);
4949 * called from commit_transaction. Writes all changed device stats to disk.
4951 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
4952 struct btrfs_fs_info
*fs_info
)
4954 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
4955 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
4956 struct btrfs_device
*device
;
4959 mutex_lock(&fs_devices
->device_list_mutex
);
4960 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
4961 if (!device
->dev_stats_valid
|| !device
->dev_stats_dirty
)
4964 ret
= update_dev_stat_item(trans
, dev_root
, device
);
4966 device
->dev_stats_dirty
= 0;
4968 mutex_unlock(&fs_devices
->device_list_mutex
);
4973 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
4975 btrfs_dev_stat_inc(dev
, index
);
4976 btrfs_dev_stat_print_on_error(dev
);
4979 void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
4981 if (!dev
->dev_stats_valid
)
4983 printk_ratelimited_in_rcu(KERN_ERR
4984 "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
4985 rcu_str_deref(dev
->name
),
4986 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
4987 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
4988 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
4989 btrfs_dev_stat_read(dev
,
4990 BTRFS_DEV_STAT_CORRUPTION_ERRS
),
4991 btrfs_dev_stat_read(dev
,
4992 BTRFS_DEV_STAT_GENERATION_ERRS
));
4995 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
4999 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
5000 if (btrfs_dev_stat_read(dev
, i
) != 0)
5002 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
5003 return; /* all values == 0, suppress message */
5005 printk_in_rcu(KERN_INFO
"btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5006 rcu_str_deref(dev
->name
),
5007 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
5008 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
5009 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
5010 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
5011 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
5014 int btrfs_get_dev_stats(struct btrfs_root
*root
,
5015 struct btrfs_ioctl_get_dev_stats
*stats
)
5017 struct btrfs_device
*dev
;
5018 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
5021 mutex_lock(&fs_devices
->device_list_mutex
);
5022 dev
= btrfs_find_device(root
, stats
->devid
, NULL
, NULL
);
5023 mutex_unlock(&fs_devices
->device_list_mutex
);
5027 "btrfs: get dev_stats failed, device not found\n");
5029 } else if (!dev
->dev_stats_valid
) {
5031 "btrfs: get dev_stats failed, not yet valid\n");
5033 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
5034 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
5035 if (stats
->nr_items
> i
)
5037 btrfs_dev_stat_read_and_reset(dev
, i
);
5039 btrfs_dev_stat_reset(dev
, i
);
5042 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
5043 if (stats
->nr_items
> i
)
5044 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
5046 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
5047 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;