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 <asm/div64.h>
29 #include "extent_map.h"
31 #include "transaction.h"
32 #include "print-tree.h"
34 #include "async-thread.h"
36 static int init_first_rw_device(struct btrfs_trans_handle
*trans
,
37 struct btrfs_root
*root
,
38 struct btrfs_device
*device
);
39 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
);
41 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
42 (sizeof(struct btrfs_bio_stripe) * (n)))
44 static DEFINE_MUTEX(uuid_mutex
);
45 static LIST_HEAD(fs_uuids
);
47 void btrfs_lock_volumes(void)
49 mutex_lock(&uuid_mutex
);
52 void btrfs_unlock_volumes(void)
54 mutex_unlock(&uuid_mutex
);
57 static void lock_chunks(struct btrfs_root
*root
)
59 mutex_lock(&root
->fs_info
->chunk_mutex
);
62 static void unlock_chunks(struct btrfs_root
*root
)
64 mutex_unlock(&root
->fs_info
->chunk_mutex
);
67 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
69 struct btrfs_device
*device
;
70 WARN_ON(fs_devices
->opened
);
71 while (!list_empty(&fs_devices
->devices
)) {
72 device
= list_entry(fs_devices
->devices
.next
,
73 struct btrfs_device
, dev_list
);
74 list_del(&device
->dev_list
);
81 int btrfs_cleanup_fs_uuids(void)
83 struct btrfs_fs_devices
*fs_devices
;
85 while (!list_empty(&fs_uuids
)) {
86 fs_devices
= list_entry(fs_uuids
.next
,
87 struct btrfs_fs_devices
, list
);
88 list_del(&fs_devices
->list
);
89 free_fs_devices(fs_devices
);
94 static noinline
struct btrfs_device
*__find_device(struct list_head
*head
,
97 struct btrfs_device
*dev
;
99 list_for_each_entry(dev
, head
, dev_list
) {
100 if (dev
->devid
== devid
&&
101 (!uuid
|| !memcmp(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
))) {
108 static noinline
struct btrfs_fs_devices
*find_fsid(u8
*fsid
)
110 struct btrfs_fs_devices
*fs_devices
;
112 list_for_each_entry(fs_devices
, &fs_uuids
, list
) {
113 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
119 static void requeue_list(struct btrfs_pending_bios
*pending_bios
,
120 struct bio
*head
, struct bio
*tail
)
123 struct bio
*old_head
;
125 old_head
= pending_bios
->head
;
126 pending_bios
->head
= head
;
127 if (pending_bios
->tail
)
128 tail
->bi_next
= old_head
;
130 pending_bios
->tail
= tail
;
134 * we try to collect pending bios for a device so we don't get a large
135 * number of procs sending bios down to the same device. This greatly
136 * improves the schedulers ability to collect and merge the bios.
138 * But, it also turns into a long list of bios to process and that is sure
139 * to eventually make the worker thread block. The solution here is to
140 * make some progress and then put this work struct back at the end of
141 * the list if the block device is congested. This way, multiple devices
142 * can make progress from a single worker thread.
144 static noinline
int run_scheduled_bios(struct btrfs_device
*device
)
147 struct backing_dev_info
*bdi
;
148 struct btrfs_fs_info
*fs_info
;
149 struct btrfs_pending_bios
*pending_bios
;
153 unsigned long num_run
;
154 unsigned long batch_run
= 0;
156 unsigned long last_waited
= 0;
158 struct blk_plug plug
;
161 * this function runs all the bios we've collected for
162 * a particular device. We don't want to wander off to
163 * another device without first sending all of these down.
164 * So, setup a plug here and finish it off before we return
166 blk_start_plug(&plug
);
168 bdi
= blk_get_backing_dev_info(device
->bdev
);
169 fs_info
= device
->dev_root
->fs_info
;
170 limit
= btrfs_async_submit_limit(fs_info
);
171 limit
= limit
* 2 / 3;
174 spin_lock(&device
->io_lock
);
179 /* take all the bios off the list at once and process them
180 * later on (without the lock held). But, remember the
181 * tail and other pointers so the bios can be properly reinserted
182 * into the list if we hit congestion
184 if (!force_reg
&& device
->pending_sync_bios
.head
) {
185 pending_bios
= &device
->pending_sync_bios
;
188 pending_bios
= &device
->pending_bios
;
192 pending
= pending_bios
->head
;
193 tail
= pending_bios
->tail
;
194 WARN_ON(pending
&& !tail
);
197 * if pending was null this time around, no bios need processing
198 * at all and we can stop. Otherwise it'll loop back up again
199 * and do an additional check so no bios are missed.
201 * device->running_pending is used to synchronize with the
204 if (device
->pending_sync_bios
.head
== NULL
&&
205 device
->pending_bios
.head
== NULL
) {
207 device
->running_pending
= 0;
210 device
->running_pending
= 1;
213 pending_bios
->head
= NULL
;
214 pending_bios
->tail
= NULL
;
216 spin_unlock(&device
->io_lock
);
221 /* we want to work on both lists, but do more bios on the
222 * sync list than the regular list
225 pending_bios
!= &device
->pending_sync_bios
&&
226 device
->pending_sync_bios
.head
) ||
227 (num_run
> 64 && pending_bios
== &device
->pending_sync_bios
&&
228 device
->pending_bios
.head
)) {
229 spin_lock(&device
->io_lock
);
230 requeue_list(pending_bios
, pending
, tail
);
235 pending
= pending
->bi_next
;
237 atomic_dec(&fs_info
->nr_async_bios
);
239 if (atomic_read(&fs_info
->nr_async_bios
) < limit
&&
240 waitqueue_active(&fs_info
->async_submit_wait
))
241 wake_up(&fs_info
->async_submit_wait
);
243 BUG_ON(atomic_read(&cur
->bi_cnt
) == 0);
245 submit_bio(cur
->bi_rw
, cur
);
252 * we made progress, there is more work to do and the bdi
253 * is now congested. Back off and let other work structs
256 if (pending
&& bdi_write_congested(bdi
) && batch_run
> 8 &&
257 fs_info
->fs_devices
->open_devices
> 1) {
258 struct io_context
*ioc
;
260 ioc
= current
->io_context
;
263 * the main goal here is that we don't want to
264 * block if we're going to be able to submit
265 * more requests without blocking.
267 * This code does two great things, it pokes into
268 * the elevator code from a filesystem _and_
269 * it makes assumptions about how batching works.
271 if (ioc
&& ioc
->nr_batch_requests
> 0 &&
272 time_before(jiffies
, ioc
->last_waited
+ HZ
/50UL) &&
274 ioc
->last_waited
== last_waited
)) {
276 * we want to go through our batch of
277 * requests and stop. So, we copy out
278 * the ioc->last_waited time and test
279 * against it before looping
281 last_waited
= ioc
->last_waited
;
286 spin_lock(&device
->io_lock
);
287 requeue_list(pending_bios
, pending
, tail
);
288 device
->running_pending
= 1;
290 spin_unlock(&device
->io_lock
);
291 btrfs_requeue_work(&device
->work
);
300 spin_lock(&device
->io_lock
);
301 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
303 spin_unlock(&device
->io_lock
);
306 blk_finish_plug(&plug
);
310 static void pending_bios_fn(struct btrfs_work
*work
)
312 struct btrfs_device
*device
;
314 device
= container_of(work
, struct btrfs_device
, work
);
315 run_scheduled_bios(device
);
318 static noinline
int device_list_add(const char *path
,
319 struct btrfs_super_block
*disk_super
,
320 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
322 struct btrfs_device
*device
;
323 struct btrfs_fs_devices
*fs_devices
;
324 u64 found_transid
= btrfs_super_generation(disk_super
);
327 fs_devices
= find_fsid(disk_super
->fsid
);
329 fs_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
332 INIT_LIST_HEAD(&fs_devices
->devices
);
333 INIT_LIST_HEAD(&fs_devices
->alloc_list
);
334 list_add(&fs_devices
->list
, &fs_uuids
);
335 memcpy(fs_devices
->fsid
, disk_super
->fsid
, BTRFS_FSID_SIZE
);
336 fs_devices
->latest_devid
= devid
;
337 fs_devices
->latest_trans
= found_transid
;
338 mutex_init(&fs_devices
->device_list_mutex
);
341 device
= __find_device(&fs_devices
->devices
, devid
,
342 disk_super
->dev_item
.uuid
);
345 if (fs_devices
->opened
)
348 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
350 /* we can safely leave the fs_devices entry around */
353 device
->devid
= devid
;
354 device
->work
.func
= pending_bios_fn
;
355 memcpy(device
->uuid
, disk_super
->dev_item
.uuid
,
357 spin_lock_init(&device
->io_lock
);
358 device
->name
= kstrdup(path
, GFP_NOFS
);
363 INIT_LIST_HEAD(&device
->dev_alloc_list
);
365 mutex_lock(&fs_devices
->device_list_mutex
);
366 list_add(&device
->dev_list
, &fs_devices
->devices
);
367 mutex_unlock(&fs_devices
->device_list_mutex
);
369 device
->fs_devices
= fs_devices
;
370 fs_devices
->num_devices
++;
371 } else if (!device
->name
|| strcmp(device
->name
, path
)) {
372 name
= kstrdup(path
, GFP_NOFS
);
377 if (device
->missing
) {
378 fs_devices
->missing_devices
--;
383 if (found_transid
> fs_devices
->latest_trans
) {
384 fs_devices
->latest_devid
= devid
;
385 fs_devices
->latest_trans
= found_transid
;
387 *fs_devices_ret
= fs_devices
;
391 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
393 struct btrfs_fs_devices
*fs_devices
;
394 struct btrfs_device
*device
;
395 struct btrfs_device
*orig_dev
;
397 fs_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
399 return ERR_PTR(-ENOMEM
);
401 INIT_LIST_HEAD(&fs_devices
->devices
);
402 INIT_LIST_HEAD(&fs_devices
->alloc_list
);
403 INIT_LIST_HEAD(&fs_devices
->list
);
404 mutex_init(&fs_devices
->device_list_mutex
);
405 fs_devices
->latest_devid
= orig
->latest_devid
;
406 fs_devices
->latest_trans
= orig
->latest_trans
;
407 memcpy(fs_devices
->fsid
, orig
->fsid
, sizeof(fs_devices
->fsid
));
409 mutex_lock(&orig
->device_list_mutex
);
410 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
411 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
415 device
->name
= kstrdup(orig_dev
->name
, GFP_NOFS
);
421 device
->devid
= orig_dev
->devid
;
422 device
->work
.func
= pending_bios_fn
;
423 memcpy(device
->uuid
, orig_dev
->uuid
, sizeof(device
->uuid
));
424 spin_lock_init(&device
->io_lock
);
425 INIT_LIST_HEAD(&device
->dev_list
);
426 INIT_LIST_HEAD(&device
->dev_alloc_list
);
428 list_add(&device
->dev_list
, &fs_devices
->devices
);
429 device
->fs_devices
= fs_devices
;
430 fs_devices
->num_devices
++;
432 mutex_unlock(&orig
->device_list_mutex
);
435 mutex_unlock(&orig
->device_list_mutex
);
436 free_fs_devices(fs_devices
);
437 return ERR_PTR(-ENOMEM
);
440 int btrfs_close_extra_devices(struct btrfs_fs_devices
*fs_devices
)
442 struct btrfs_device
*device
, *next
;
444 mutex_lock(&uuid_mutex
);
446 mutex_lock(&fs_devices
->device_list_mutex
);
447 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
448 if (device
->in_fs_metadata
)
452 blkdev_put(device
->bdev
, device
->mode
);
454 fs_devices
->open_devices
--;
456 if (device
->writeable
) {
457 list_del_init(&device
->dev_alloc_list
);
458 device
->writeable
= 0;
459 fs_devices
->rw_devices
--;
461 list_del_init(&device
->dev_list
);
462 fs_devices
->num_devices
--;
466 mutex_unlock(&fs_devices
->device_list_mutex
);
468 if (fs_devices
->seed
) {
469 fs_devices
= fs_devices
->seed
;
473 mutex_unlock(&uuid_mutex
);
477 static int __btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
479 struct btrfs_device
*device
;
481 if (--fs_devices
->opened
> 0)
484 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
486 blkdev_put(device
->bdev
, device
->mode
);
487 fs_devices
->open_devices
--;
489 if (device
->writeable
) {
490 list_del_init(&device
->dev_alloc_list
);
491 fs_devices
->rw_devices
--;
495 device
->writeable
= 0;
496 device
->in_fs_metadata
= 0;
498 WARN_ON(fs_devices
->open_devices
);
499 WARN_ON(fs_devices
->rw_devices
);
500 fs_devices
->opened
= 0;
501 fs_devices
->seeding
= 0;
506 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
508 struct btrfs_fs_devices
*seed_devices
= NULL
;
511 mutex_lock(&uuid_mutex
);
512 ret
= __btrfs_close_devices(fs_devices
);
513 if (!fs_devices
->opened
) {
514 seed_devices
= fs_devices
->seed
;
515 fs_devices
->seed
= NULL
;
517 mutex_unlock(&uuid_mutex
);
519 while (seed_devices
) {
520 fs_devices
= seed_devices
;
521 seed_devices
= fs_devices
->seed
;
522 __btrfs_close_devices(fs_devices
);
523 free_fs_devices(fs_devices
);
528 static int __btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
529 fmode_t flags
, void *holder
)
531 struct block_device
*bdev
;
532 struct list_head
*head
= &fs_devices
->devices
;
533 struct btrfs_device
*device
;
534 struct block_device
*latest_bdev
= NULL
;
535 struct buffer_head
*bh
;
536 struct btrfs_super_block
*disk_super
;
537 u64 latest_devid
= 0;
538 u64 latest_transid
= 0;
545 list_for_each_entry(device
, head
, dev_list
) {
551 bdev
= blkdev_get_by_path(device
->name
, flags
, holder
);
553 printk(KERN_INFO
"open %s failed\n", device
->name
);
556 set_blocksize(bdev
, 4096);
558 bh
= btrfs_read_dev_super(bdev
);
564 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
565 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
566 if (devid
!= device
->devid
)
569 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
,
573 device
->generation
= btrfs_super_generation(disk_super
);
574 if (!latest_transid
|| device
->generation
> latest_transid
) {
575 latest_devid
= devid
;
576 latest_transid
= device
->generation
;
580 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
581 device
->writeable
= 0;
583 device
->writeable
= !bdev_read_only(bdev
);
588 device
->in_fs_metadata
= 0;
589 device
->mode
= flags
;
591 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
592 fs_devices
->rotating
= 1;
594 fs_devices
->open_devices
++;
595 if (device
->writeable
) {
596 fs_devices
->rw_devices
++;
597 list_add(&device
->dev_alloc_list
,
598 &fs_devices
->alloc_list
);
605 blkdev_put(bdev
, flags
);
609 if (fs_devices
->open_devices
== 0) {
613 fs_devices
->seeding
= seeding
;
614 fs_devices
->opened
= 1;
615 fs_devices
->latest_bdev
= latest_bdev
;
616 fs_devices
->latest_devid
= latest_devid
;
617 fs_devices
->latest_trans
= latest_transid
;
618 fs_devices
->total_rw_bytes
= 0;
623 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
624 fmode_t flags
, void *holder
)
628 mutex_lock(&uuid_mutex
);
629 if (fs_devices
->opened
) {
630 fs_devices
->opened
++;
633 ret
= __btrfs_open_devices(fs_devices
, flags
, holder
);
635 mutex_unlock(&uuid_mutex
);
639 int btrfs_scan_one_device(const char *path
, fmode_t flags
, void *holder
,
640 struct btrfs_fs_devices
**fs_devices_ret
)
642 struct btrfs_super_block
*disk_super
;
643 struct block_device
*bdev
;
644 struct buffer_head
*bh
;
649 mutex_lock(&uuid_mutex
);
652 bdev
= blkdev_get_by_path(path
, flags
, holder
);
659 ret
= set_blocksize(bdev
, 4096);
662 bh
= btrfs_read_dev_super(bdev
);
667 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
668 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
669 transid
= btrfs_super_generation(disk_super
);
670 if (disk_super
->label
[0])
671 printk(KERN_INFO
"device label %s ", disk_super
->label
);
673 /* FIXME, make a readl uuid parser */
674 printk(KERN_INFO
"device fsid %llx-%llx ",
675 *(unsigned long long *)disk_super
->fsid
,
676 *(unsigned long long *)(disk_super
->fsid
+ 8));
678 printk(KERN_CONT
"devid %llu transid %llu %s\n",
679 (unsigned long long)devid
, (unsigned long long)transid
, path
);
680 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
684 blkdev_put(bdev
, flags
);
686 mutex_unlock(&uuid_mutex
);
690 /* helper to account the used device space in the range */
691 int btrfs_account_dev_extents_size(struct btrfs_device
*device
, u64 start
,
692 u64 end
, u64
*length
)
694 struct btrfs_key key
;
695 struct btrfs_root
*root
= device
->dev_root
;
696 struct btrfs_dev_extent
*dev_extent
;
697 struct btrfs_path
*path
;
701 struct extent_buffer
*l
;
705 if (start
>= device
->total_bytes
)
708 path
= btrfs_alloc_path();
713 key
.objectid
= device
->devid
;
715 key
.type
= BTRFS_DEV_EXTENT_KEY
;
717 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
721 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
728 slot
= path
->slots
[0];
729 if (slot
>= btrfs_header_nritems(l
)) {
730 ret
= btrfs_next_leaf(root
, path
);
738 btrfs_item_key_to_cpu(l
, &key
, slot
);
740 if (key
.objectid
< device
->devid
)
743 if (key
.objectid
> device
->devid
)
746 if (btrfs_key_type(&key
) != BTRFS_DEV_EXTENT_KEY
)
749 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
750 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
752 if (key
.offset
<= start
&& extent_end
> end
) {
753 *length
= end
- start
+ 1;
755 } else if (key
.offset
<= start
&& extent_end
> start
)
756 *length
+= extent_end
- start
;
757 else if (key
.offset
> start
&& extent_end
<= end
)
758 *length
+= extent_end
- key
.offset
;
759 else if (key
.offset
> start
&& key
.offset
<= end
) {
760 *length
+= end
- key
.offset
+ 1;
762 } else if (key
.offset
> end
)
770 btrfs_free_path(path
);
775 * find_free_dev_extent - find free space in the specified device
776 * @trans: transaction handler
777 * @device: the device which we search the free space in
778 * @num_bytes: the size of the free space that we need
779 * @start: store the start of the free space.
780 * @len: the size of the free space. that we find, or the size of the max
781 * free space if we don't find suitable free space
783 * this uses a pretty simple search, the expectation is that it is
784 * called very infrequently and that a given device has a small number
787 * @start is used to store the start of the free space if we find. But if we
788 * don't find suitable free space, it will be used to store the start position
789 * of the max free space.
791 * @len is used to store the size of the free space that we find.
792 * But if we don't find suitable free space, it is used to store the size of
793 * the max free space.
795 int find_free_dev_extent(struct btrfs_trans_handle
*trans
,
796 struct btrfs_device
*device
, u64 num_bytes
,
797 u64
*start
, u64
*len
)
799 struct btrfs_key key
;
800 struct btrfs_root
*root
= device
->dev_root
;
801 struct btrfs_dev_extent
*dev_extent
;
802 struct btrfs_path
*path
;
808 u64 search_end
= device
->total_bytes
;
811 struct extent_buffer
*l
;
813 /* FIXME use last free of some kind */
815 /* we don't want to overwrite the superblock on the drive,
816 * so we make sure to start at an offset of at least 1MB
818 search_start
= 1024 * 1024;
820 if (root
->fs_info
->alloc_start
+ num_bytes
<= search_end
)
821 search_start
= max(root
->fs_info
->alloc_start
, search_start
);
823 max_hole_start
= search_start
;
826 if (search_start
>= search_end
) {
831 path
= btrfs_alloc_path();
838 key
.objectid
= device
->devid
;
839 key
.offset
= search_start
;
840 key
.type
= BTRFS_DEV_EXTENT_KEY
;
842 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 0);
846 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
853 slot
= path
->slots
[0];
854 if (slot
>= btrfs_header_nritems(l
)) {
855 ret
= btrfs_next_leaf(root
, path
);
863 btrfs_item_key_to_cpu(l
, &key
, slot
);
865 if (key
.objectid
< device
->devid
)
868 if (key
.objectid
> device
->devid
)
871 if (btrfs_key_type(&key
) != BTRFS_DEV_EXTENT_KEY
)
874 if (key
.offset
> search_start
) {
875 hole_size
= key
.offset
- search_start
;
877 if (hole_size
> max_hole_size
) {
878 max_hole_start
= search_start
;
879 max_hole_size
= hole_size
;
883 * If this free space is greater than which we need,
884 * it must be the max free space that we have found
885 * until now, so max_hole_start must point to the start
886 * of this free space and the length of this free space
887 * is stored in max_hole_size. Thus, we return
888 * max_hole_start and max_hole_size and go back to the
891 if (hole_size
>= num_bytes
) {
897 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
898 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
900 if (extent_end
> search_start
)
901 search_start
= extent_end
;
907 hole_size
= search_end
- search_start
;
908 if (hole_size
> max_hole_size
) {
909 max_hole_start
= search_start
;
910 max_hole_size
= hole_size
;
914 if (hole_size
< num_bytes
)
920 btrfs_free_path(path
);
922 *start
= max_hole_start
;
924 *len
= max_hole_size
;
928 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
929 struct btrfs_device
*device
,
933 struct btrfs_path
*path
;
934 struct btrfs_root
*root
= device
->dev_root
;
935 struct btrfs_key key
;
936 struct btrfs_key found_key
;
937 struct extent_buffer
*leaf
= NULL
;
938 struct btrfs_dev_extent
*extent
= NULL
;
940 path
= btrfs_alloc_path();
944 key
.objectid
= device
->devid
;
946 key
.type
= BTRFS_DEV_EXTENT_KEY
;
948 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
950 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
951 BTRFS_DEV_EXTENT_KEY
);
953 leaf
= path
->nodes
[0];
954 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
955 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
956 struct btrfs_dev_extent
);
957 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
958 btrfs_dev_extent_length(leaf
, extent
) < start
);
960 } else if (ret
== 0) {
961 leaf
= path
->nodes
[0];
962 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
963 struct btrfs_dev_extent
);
967 if (device
->bytes_used
> 0)
968 device
->bytes_used
-= btrfs_dev_extent_length(leaf
, extent
);
969 ret
= btrfs_del_item(trans
, root
, path
);
972 btrfs_free_path(path
);
976 int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
977 struct btrfs_device
*device
,
978 u64 chunk_tree
, u64 chunk_objectid
,
979 u64 chunk_offset
, u64 start
, u64 num_bytes
)
982 struct btrfs_path
*path
;
983 struct btrfs_root
*root
= device
->dev_root
;
984 struct btrfs_dev_extent
*extent
;
985 struct extent_buffer
*leaf
;
986 struct btrfs_key key
;
988 WARN_ON(!device
->in_fs_metadata
);
989 path
= btrfs_alloc_path();
993 key
.objectid
= device
->devid
;
995 key
.type
= BTRFS_DEV_EXTENT_KEY
;
996 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1000 leaf
= path
->nodes
[0];
1001 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1002 struct btrfs_dev_extent
);
1003 btrfs_set_dev_extent_chunk_tree(leaf
, extent
, chunk_tree
);
1004 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
, chunk_objectid
);
1005 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1007 write_extent_buffer(leaf
, root
->fs_info
->chunk_tree_uuid
,
1008 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent
),
1011 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1012 btrfs_mark_buffer_dirty(leaf
);
1013 btrfs_free_path(path
);
1017 static noinline
int find_next_chunk(struct btrfs_root
*root
,
1018 u64 objectid
, u64
*offset
)
1020 struct btrfs_path
*path
;
1022 struct btrfs_key key
;
1023 struct btrfs_chunk
*chunk
;
1024 struct btrfs_key found_key
;
1026 path
= btrfs_alloc_path();
1029 key
.objectid
= objectid
;
1030 key
.offset
= (u64
)-1;
1031 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
1033 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1039 ret
= btrfs_previous_item(root
, path
, 0, BTRFS_CHUNK_ITEM_KEY
);
1043 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1045 if (found_key
.objectid
!= objectid
)
1048 chunk
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1049 struct btrfs_chunk
);
1050 *offset
= found_key
.offset
+
1051 btrfs_chunk_length(path
->nodes
[0], chunk
);
1056 btrfs_free_path(path
);
1060 static noinline
int find_next_devid(struct btrfs_root
*root
, u64
*objectid
)
1063 struct btrfs_key key
;
1064 struct btrfs_key found_key
;
1065 struct btrfs_path
*path
;
1067 root
= root
->fs_info
->chunk_root
;
1069 path
= btrfs_alloc_path();
1073 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1074 key
.type
= BTRFS_DEV_ITEM_KEY
;
1075 key
.offset
= (u64
)-1;
1077 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1083 ret
= btrfs_previous_item(root
, path
, BTRFS_DEV_ITEMS_OBJECTID
,
1084 BTRFS_DEV_ITEM_KEY
);
1088 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1090 *objectid
= found_key
.offset
+ 1;
1094 btrfs_free_path(path
);
1099 * the device information is stored in the chunk root
1100 * the btrfs_device struct should be fully filled in
1102 int btrfs_add_device(struct btrfs_trans_handle
*trans
,
1103 struct btrfs_root
*root
,
1104 struct btrfs_device
*device
)
1107 struct btrfs_path
*path
;
1108 struct btrfs_dev_item
*dev_item
;
1109 struct extent_buffer
*leaf
;
1110 struct btrfs_key key
;
1113 root
= root
->fs_info
->chunk_root
;
1115 path
= btrfs_alloc_path();
1119 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1120 key
.type
= BTRFS_DEV_ITEM_KEY
;
1121 key
.offset
= device
->devid
;
1123 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1128 leaf
= path
->nodes
[0];
1129 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1131 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1132 btrfs_set_device_generation(leaf
, dev_item
, 0);
1133 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1134 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1135 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1136 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1137 btrfs_set_device_total_bytes(leaf
, dev_item
, device
->total_bytes
);
1138 btrfs_set_device_bytes_used(leaf
, dev_item
, device
->bytes_used
);
1139 btrfs_set_device_group(leaf
, dev_item
, 0);
1140 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1141 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1142 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1144 ptr
= (unsigned long)btrfs_device_uuid(dev_item
);
1145 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1146 ptr
= (unsigned long)btrfs_device_fsid(dev_item
);
1147 write_extent_buffer(leaf
, root
->fs_info
->fsid
, ptr
, BTRFS_UUID_SIZE
);
1148 btrfs_mark_buffer_dirty(leaf
);
1152 btrfs_free_path(path
);
1156 static int btrfs_rm_dev_item(struct btrfs_root
*root
,
1157 struct btrfs_device
*device
)
1160 struct btrfs_path
*path
;
1161 struct btrfs_key key
;
1162 struct btrfs_trans_handle
*trans
;
1164 root
= root
->fs_info
->chunk_root
;
1166 path
= btrfs_alloc_path();
1170 trans
= btrfs_start_transaction(root
, 0);
1171 if (IS_ERR(trans
)) {
1172 btrfs_free_path(path
);
1173 return PTR_ERR(trans
);
1175 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1176 key
.type
= BTRFS_DEV_ITEM_KEY
;
1177 key
.offset
= device
->devid
;
1180 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1189 ret
= btrfs_del_item(trans
, root
, path
);
1193 btrfs_free_path(path
);
1194 unlock_chunks(root
);
1195 btrfs_commit_transaction(trans
, root
);
1199 int btrfs_rm_device(struct btrfs_root
*root
, char *device_path
)
1201 struct btrfs_device
*device
;
1202 struct btrfs_device
*next_device
;
1203 struct block_device
*bdev
;
1204 struct buffer_head
*bh
= NULL
;
1205 struct btrfs_super_block
*disk_super
;
1212 mutex_lock(&uuid_mutex
);
1213 mutex_lock(&root
->fs_info
->volume_mutex
);
1215 all_avail
= root
->fs_info
->avail_data_alloc_bits
|
1216 root
->fs_info
->avail_system_alloc_bits
|
1217 root
->fs_info
->avail_metadata_alloc_bits
;
1219 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID10
) &&
1220 root
->fs_info
->fs_devices
->num_devices
<= 4) {
1221 printk(KERN_ERR
"btrfs: unable to go below four devices "
1227 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID1
) &&
1228 root
->fs_info
->fs_devices
->num_devices
<= 2) {
1229 printk(KERN_ERR
"btrfs: unable to go below two "
1230 "devices on raid1\n");
1235 if (strcmp(device_path
, "missing") == 0) {
1236 struct list_head
*devices
;
1237 struct btrfs_device
*tmp
;
1240 devices
= &root
->fs_info
->fs_devices
->devices
;
1241 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1242 list_for_each_entry(tmp
, devices
, dev_list
) {
1243 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
1248 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1253 printk(KERN_ERR
"btrfs: no missing devices found to "
1258 bdev
= blkdev_get_by_path(device_path
, FMODE_READ
| FMODE_EXCL
,
1259 root
->fs_info
->bdev_holder
);
1261 ret
= PTR_ERR(bdev
);
1265 set_blocksize(bdev
, 4096);
1266 bh
= btrfs_read_dev_super(bdev
);
1271 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1272 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1273 dev_uuid
= disk_super
->dev_item
.uuid
;
1274 device
= btrfs_find_device(root
, devid
, dev_uuid
,
1282 if (device
->writeable
&& root
->fs_info
->fs_devices
->rw_devices
== 1) {
1283 printk(KERN_ERR
"btrfs: unable to remove the only writeable "
1289 if (device
->writeable
) {
1290 list_del_init(&device
->dev_alloc_list
);
1291 root
->fs_info
->fs_devices
->rw_devices
--;
1294 ret
= btrfs_shrink_device(device
, 0);
1298 ret
= btrfs_rm_dev_item(root
->fs_info
->chunk_root
, device
);
1302 device
->in_fs_metadata
= 0;
1305 * the device list mutex makes sure that we don't change
1306 * the device list while someone else is writing out all
1307 * the device supers.
1309 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1310 list_del_init(&device
->dev_list
);
1311 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1313 device
->fs_devices
->num_devices
--;
1315 if (device
->missing
)
1316 root
->fs_info
->fs_devices
->missing_devices
--;
1318 next_device
= list_entry(root
->fs_info
->fs_devices
->devices
.next
,
1319 struct btrfs_device
, dev_list
);
1320 if (device
->bdev
== root
->fs_info
->sb
->s_bdev
)
1321 root
->fs_info
->sb
->s_bdev
= next_device
->bdev
;
1322 if (device
->bdev
== root
->fs_info
->fs_devices
->latest_bdev
)
1323 root
->fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1326 blkdev_put(device
->bdev
, device
->mode
);
1327 device
->bdev
= NULL
;
1328 device
->fs_devices
->open_devices
--;
1331 num_devices
= btrfs_super_num_devices(&root
->fs_info
->super_copy
) - 1;
1332 btrfs_set_super_num_devices(&root
->fs_info
->super_copy
, num_devices
);
1334 if (device
->fs_devices
->open_devices
== 0) {
1335 struct btrfs_fs_devices
*fs_devices
;
1336 fs_devices
= root
->fs_info
->fs_devices
;
1337 while (fs_devices
) {
1338 if (fs_devices
->seed
== device
->fs_devices
)
1340 fs_devices
= fs_devices
->seed
;
1342 fs_devices
->seed
= device
->fs_devices
->seed
;
1343 device
->fs_devices
->seed
= NULL
;
1344 __btrfs_close_devices(device
->fs_devices
);
1345 free_fs_devices(device
->fs_devices
);
1349 * at this point, the device is zero sized. We want to
1350 * remove it from the devices list and zero out the old super
1352 if (device
->writeable
) {
1353 /* make sure this device isn't detected as part of
1356 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
1357 set_buffer_dirty(bh
);
1358 sync_dirty_buffer(bh
);
1361 kfree(device
->name
);
1369 blkdev_put(bdev
, FMODE_READ
| FMODE_EXCL
);
1371 mutex_unlock(&root
->fs_info
->volume_mutex
);
1372 mutex_unlock(&uuid_mutex
);
1375 if (device
->writeable
) {
1376 list_add(&device
->dev_alloc_list
,
1377 &root
->fs_info
->fs_devices
->alloc_list
);
1378 root
->fs_info
->fs_devices
->rw_devices
++;
1384 * does all the dirty work required for changing file system's UUID.
1386 static int btrfs_prepare_sprout(struct btrfs_trans_handle
*trans
,
1387 struct btrfs_root
*root
)
1389 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
1390 struct btrfs_fs_devices
*old_devices
;
1391 struct btrfs_fs_devices
*seed_devices
;
1392 struct btrfs_super_block
*disk_super
= &root
->fs_info
->super_copy
;
1393 struct btrfs_device
*device
;
1396 BUG_ON(!mutex_is_locked(&uuid_mutex
));
1397 if (!fs_devices
->seeding
)
1400 seed_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
1404 old_devices
= clone_fs_devices(fs_devices
);
1405 if (IS_ERR(old_devices
)) {
1406 kfree(seed_devices
);
1407 return PTR_ERR(old_devices
);
1410 list_add(&old_devices
->list
, &fs_uuids
);
1412 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
1413 seed_devices
->opened
= 1;
1414 INIT_LIST_HEAD(&seed_devices
->devices
);
1415 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
1416 mutex_init(&seed_devices
->device_list_mutex
);
1417 list_splice_init(&fs_devices
->devices
, &seed_devices
->devices
);
1418 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
1419 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
) {
1420 device
->fs_devices
= seed_devices
;
1423 fs_devices
->seeding
= 0;
1424 fs_devices
->num_devices
= 0;
1425 fs_devices
->open_devices
= 0;
1426 fs_devices
->seed
= seed_devices
;
1428 generate_random_uuid(fs_devices
->fsid
);
1429 memcpy(root
->fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
1430 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
1431 super_flags
= btrfs_super_flags(disk_super
) &
1432 ~BTRFS_SUPER_FLAG_SEEDING
;
1433 btrfs_set_super_flags(disk_super
, super_flags
);
1439 * strore the expected generation for seed devices in device items.
1441 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
1442 struct btrfs_root
*root
)
1444 struct btrfs_path
*path
;
1445 struct extent_buffer
*leaf
;
1446 struct btrfs_dev_item
*dev_item
;
1447 struct btrfs_device
*device
;
1448 struct btrfs_key key
;
1449 u8 fs_uuid
[BTRFS_UUID_SIZE
];
1450 u8 dev_uuid
[BTRFS_UUID_SIZE
];
1454 path
= btrfs_alloc_path();
1458 root
= root
->fs_info
->chunk_root
;
1459 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1461 key
.type
= BTRFS_DEV_ITEM_KEY
;
1464 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
1468 leaf
= path
->nodes
[0];
1470 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1471 ret
= btrfs_next_leaf(root
, path
);
1476 leaf
= path
->nodes
[0];
1477 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1478 btrfs_release_path(root
, path
);
1482 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1483 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
1484 key
.type
!= BTRFS_DEV_ITEM_KEY
)
1487 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1488 struct btrfs_dev_item
);
1489 devid
= btrfs_device_id(leaf
, dev_item
);
1490 read_extent_buffer(leaf
, dev_uuid
,
1491 (unsigned long)btrfs_device_uuid(dev_item
),
1493 read_extent_buffer(leaf
, fs_uuid
,
1494 (unsigned long)btrfs_device_fsid(dev_item
),
1496 device
= btrfs_find_device(root
, devid
, dev_uuid
, fs_uuid
);
1499 if (device
->fs_devices
->seeding
) {
1500 btrfs_set_device_generation(leaf
, dev_item
,
1501 device
->generation
);
1502 btrfs_mark_buffer_dirty(leaf
);
1510 btrfs_free_path(path
);
1514 int btrfs_init_new_device(struct btrfs_root
*root
, char *device_path
)
1516 struct btrfs_trans_handle
*trans
;
1517 struct btrfs_device
*device
;
1518 struct block_device
*bdev
;
1519 struct list_head
*devices
;
1520 struct super_block
*sb
= root
->fs_info
->sb
;
1522 int seeding_dev
= 0;
1525 if ((sb
->s_flags
& MS_RDONLY
) && !root
->fs_info
->fs_devices
->seeding
)
1528 bdev
= blkdev_get_by_path(device_path
, FMODE_EXCL
,
1529 root
->fs_info
->bdev_holder
);
1531 return PTR_ERR(bdev
);
1533 if (root
->fs_info
->fs_devices
->seeding
) {
1535 down_write(&sb
->s_umount
);
1536 mutex_lock(&uuid_mutex
);
1539 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
1540 mutex_lock(&root
->fs_info
->volume_mutex
);
1542 devices
= &root
->fs_info
->fs_devices
->devices
;
1544 * we have the volume lock, so we don't need the extra
1545 * device list mutex while reading the list here.
1547 list_for_each_entry(device
, devices
, dev_list
) {
1548 if (device
->bdev
== bdev
) {
1554 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
1556 /* we can safely leave the fs_devices entry around */
1561 device
->name
= kstrdup(device_path
, GFP_NOFS
);
1562 if (!device
->name
) {
1568 ret
= find_next_devid(root
, &device
->devid
);
1570 kfree(device
->name
);
1575 trans
= btrfs_start_transaction(root
, 0);
1576 if (IS_ERR(trans
)) {
1577 kfree(device
->name
);
1579 ret
= PTR_ERR(trans
);
1585 device
->writeable
= 1;
1586 device
->work
.func
= pending_bios_fn
;
1587 generate_random_uuid(device
->uuid
);
1588 spin_lock_init(&device
->io_lock
);
1589 device
->generation
= trans
->transid
;
1590 device
->io_width
= root
->sectorsize
;
1591 device
->io_align
= root
->sectorsize
;
1592 device
->sector_size
= root
->sectorsize
;
1593 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
1594 device
->disk_total_bytes
= device
->total_bytes
;
1595 device
->dev_root
= root
->fs_info
->dev_root
;
1596 device
->bdev
= bdev
;
1597 device
->in_fs_metadata
= 1;
1598 device
->mode
= FMODE_EXCL
;
1599 set_blocksize(device
->bdev
, 4096);
1602 sb
->s_flags
&= ~MS_RDONLY
;
1603 ret
= btrfs_prepare_sprout(trans
, root
);
1607 device
->fs_devices
= root
->fs_info
->fs_devices
;
1610 * we don't want write_supers to jump in here with our device
1613 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1614 list_add(&device
->dev_list
, &root
->fs_info
->fs_devices
->devices
);
1615 list_add(&device
->dev_alloc_list
,
1616 &root
->fs_info
->fs_devices
->alloc_list
);
1617 root
->fs_info
->fs_devices
->num_devices
++;
1618 root
->fs_info
->fs_devices
->open_devices
++;
1619 root
->fs_info
->fs_devices
->rw_devices
++;
1620 root
->fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
1622 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
1623 root
->fs_info
->fs_devices
->rotating
= 1;
1625 total_bytes
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
1626 btrfs_set_super_total_bytes(&root
->fs_info
->super_copy
,
1627 total_bytes
+ device
->total_bytes
);
1629 total_bytes
= btrfs_super_num_devices(&root
->fs_info
->super_copy
);
1630 btrfs_set_super_num_devices(&root
->fs_info
->super_copy
,
1632 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1635 ret
= init_first_rw_device(trans
, root
, device
);
1637 ret
= btrfs_finish_sprout(trans
, root
);
1640 ret
= btrfs_add_device(trans
, root
, device
);
1644 * we've got more storage, clear any full flags on the space
1647 btrfs_clear_space_info_full(root
->fs_info
);
1649 unlock_chunks(root
);
1650 btrfs_commit_transaction(trans
, root
);
1653 mutex_unlock(&uuid_mutex
);
1654 up_write(&sb
->s_umount
);
1656 ret
= btrfs_relocate_sys_chunks(root
);
1660 mutex_unlock(&root
->fs_info
->volume_mutex
);
1663 blkdev_put(bdev
, FMODE_EXCL
);
1665 mutex_unlock(&uuid_mutex
);
1666 up_write(&sb
->s_umount
);
1671 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
1672 struct btrfs_device
*device
)
1675 struct btrfs_path
*path
;
1676 struct btrfs_root
*root
;
1677 struct btrfs_dev_item
*dev_item
;
1678 struct extent_buffer
*leaf
;
1679 struct btrfs_key key
;
1681 root
= device
->dev_root
->fs_info
->chunk_root
;
1683 path
= btrfs_alloc_path();
1687 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1688 key
.type
= BTRFS_DEV_ITEM_KEY
;
1689 key
.offset
= device
->devid
;
1691 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
1700 leaf
= path
->nodes
[0];
1701 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1703 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1704 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1705 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1706 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1707 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1708 btrfs_set_device_total_bytes(leaf
, dev_item
, device
->disk_total_bytes
);
1709 btrfs_set_device_bytes_used(leaf
, dev_item
, device
->bytes_used
);
1710 btrfs_mark_buffer_dirty(leaf
);
1713 btrfs_free_path(path
);
1717 static int __btrfs_grow_device(struct btrfs_trans_handle
*trans
,
1718 struct btrfs_device
*device
, u64 new_size
)
1720 struct btrfs_super_block
*super_copy
=
1721 &device
->dev_root
->fs_info
->super_copy
;
1722 u64 old_total
= btrfs_super_total_bytes(super_copy
);
1723 u64 diff
= new_size
- device
->total_bytes
;
1725 if (!device
->writeable
)
1727 if (new_size
<= device
->total_bytes
)
1730 btrfs_set_super_total_bytes(super_copy
, old_total
+ diff
);
1731 device
->fs_devices
->total_rw_bytes
+= diff
;
1733 device
->total_bytes
= new_size
;
1734 device
->disk_total_bytes
= new_size
;
1735 btrfs_clear_space_info_full(device
->dev_root
->fs_info
);
1737 return btrfs_update_device(trans
, device
);
1740 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
1741 struct btrfs_device
*device
, u64 new_size
)
1744 lock_chunks(device
->dev_root
);
1745 ret
= __btrfs_grow_device(trans
, device
, new_size
);
1746 unlock_chunks(device
->dev_root
);
1750 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
1751 struct btrfs_root
*root
,
1752 u64 chunk_tree
, u64 chunk_objectid
,
1756 struct btrfs_path
*path
;
1757 struct btrfs_key key
;
1759 root
= root
->fs_info
->chunk_root
;
1760 path
= btrfs_alloc_path();
1764 key
.objectid
= chunk_objectid
;
1765 key
.offset
= chunk_offset
;
1766 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
1768 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1771 ret
= btrfs_del_item(trans
, root
, path
);
1774 btrfs_free_path(path
);
1778 static int btrfs_del_sys_chunk(struct btrfs_root
*root
, u64 chunk_objectid
, u64
1781 struct btrfs_super_block
*super_copy
= &root
->fs_info
->super_copy
;
1782 struct btrfs_disk_key
*disk_key
;
1783 struct btrfs_chunk
*chunk
;
1790 struct btrfs_key key
;
1792 array_size
= btrfs_super_sys_array_size(super_copy
);
1794 ptr
= super_copy
->sys_chunk_array
;
1797 while (cur
< array_size
) {
1798 disk_key
= (struct btrfs_disk_key
*)ptr
;
1799 btrfs_disk_key_to_cpu(&key
, disk_key
);
1801 len
= sizeof(*disk_key
);
1803 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
1804 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
1805 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
1806 len
+= btrfs_chunk_item_size(num_stripes
);
1811 if (key
.objectid
== chunk_objectid
&&
1812 key
.offset
== chunk_offset
) {
1813 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
1815 btrfs_set_super_sys_array_size(super_copy
, array_size
);
1824 static int btrfs_relocate_chunk(struct btrfs_root
*root
,
1825 u64 chunk_tree
, u64 chunk_objectid
,
1828 struct extent_map_tree
*em_tree
;
1829 struct btrfs_root
*extent_root
;
1830 struct btrfs_trans_handle
*trans
;
1831 struct extent_map
*em
;
1832 struct map_lookup
*map
;
1836 root
= root
->fs_info
->chunk_root
;
1837 extent_root
= root
->fs_info
->extent_root
;
1838 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
1840 ret
= btrfs_can_relocate(extent_root
, chunk_offset
);
1844 /* step one, relocate all the extents inside this chunk */
1845 ret
= btrfs_relocate_block_group(extent_root
, chunk_offset
);
1849 trans
= btrfs_start_transaction(root
, 0);
1850 BUG_ON(IS_ERR(trans
));
1855 * step two, delete the device extents and the
1856 * chunk tree entries
1858 read_lock(&em_tree
->lock
);
1859 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
1860 read_unlock(&em_tree
->lock
);
1862 BUG_ON(em
->start
> chunk_offset
||
1863 em
->start
+ em
->len
< chunk_offset
);
1864 map
= (struct map_lookup
*)em
->bdev
;
1866 for (i
= 0; i
< map
->num_stripes
; i
++) {
1867 ret
= btrfs_free_dev_extent(trans
, map
->stripes
[i
].dev
,
1868 map
->stripes
[i
].physical
);
1871 if (map
->stripes
[i
].dev
) {
1872 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
1876 ret
= btrfs_free_chunk(trans
, root
, chunk_tree
, chunk_objectid
,
1881 trace_btrfs_chunk_free(root
, map
, chunk_offset
, em
->len
);
1883 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
1884 ret
= btrfs_del_sys_chunk(root
, chunk_objectid
, chunk_offset
);
1888 ret
= btrfs_remove_block_group(trans
, extent_root
, chunk_offset
);
1891 write_lock(&em_tree
->lock
);
1892 remove_extent_mapping(em_tree
, em
);
1893 write_unlock(&em_tree
->lock
);
1898 /* once for the tree */
1899 free_extent_map(em
);
1901 free_extent_map(em
);
1903 unlock_chunks(root
);
1904 btrfs_end_transaction(trans
, root
);
1908 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
)
1910 struct btrfs_root
*chunk_root
= root
->fs_info
->chunk_root
;
1911 struct btrfs_path
*path
;
1912 struct extent_buffer
*leaf
;
1913 struct btrfs_chunk
*chunk
;
1914 struct btrfs_key key
;
1915 struct btrfs_key found_key
;
1916 u64 chunk_tree
= chunk_root
->root_key
.objectid
;
1918 bool retried
= false;
1922 path
= btrfs_alloc_path();
1927 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
1928 key
.offset
= (u64
)-1;
1929 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
1932 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
1937 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
1944 leaf
= path
->nodes
[0];
1945 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1947 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
1948 struct btrfs_chunk
);
1949 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
1950 btrfs_release_path(chunk_root
, path
);
1952 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
1953 ret
= btrfs_relocate_chunk(chunk_root
, chunk_tree
,
1962 if (found_key
.offset
== 0)
1964 key
.offset
= found_key
.offset
- 1;
1967 if (failed
&& !retried
) {
1971 } else if (failed
&& retried
) {
1976 btrfs_free_path(path
);
1980 static u64
div_factor(u64 num
, int factor
)
1989 int btrfs_balance(struct btrfs_root
*dev_root
)
1992 struct list_head
*devices
= &dev_root
->fs_info
->fs_devices
->devices
;
1993 struct btrfs_device
*device
;
1996 struct btrfs_path
*path
;
1997 struct btrfs_key key
;
1998 struct btrfs_root
*chunk_root
= dev_root
->fs_info
->chunk_root
;
1999 struct btrfs_trans_handle
*trans
;
2000 struct btrfs_key found_key
;
2002 if (dev_root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
2005 if (!capable(CAP_SYS_ADMIN
))
2008 mutex_lock(&dev_root
->fs_info
->volume_mutex
);
2009 dev_root
= dev_root
->fs_info
->dev_root
;
2011 /* step one make some room on all the devices */
2012 list_for_each_entry(device
, devices
, dev_list
) {
2013 old_size
= device
->total_bytes
;
2014 size_to_free
= div_factor(old_size
, 1);
2015 size_to_free
= min(size_to_free
, (u64
)1 * 1024 * 1024);
2016 if (!device
->writeable
||
2017 device
->total_bytes
- device
->bytes_used
> size_to_free
)
2020 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
2025 trans
= btrfs_start_transaction(dev_root
, 0);
2026 BUG_ON(IS_ERR(trans
));
2028 ret
= btrfs_grow_device(trans
, device
, old_size
);
2031 btrfs_end_transaction(trans
, dev_root
);
2034 /* step two, relocate all the chunks */
2035 path
= btrfs_alloc_path();
2038 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2039 key
.offset
= (u64
)-1;
2040 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2043 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2048 * this shouldn't happen, it means the last relocate
2054 ret
= btrfs_previous_item(chunk_root
, path
, 0,
2055 BTRFS_CHUNK_ITEM_KEY
);
2059 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
2061 if (found_key
.objectid
!= key
.objectid
)
2064 /* chunk zero is special */
2065 if (found_key
.offset
== 0)
2068 btrfs_release_path(chunk_root
, path
);
2069 ret
= btrfs_relocate_chunk(chunk_root
,
2070 chunk_root
->root_key
.objectid
,
2073 BUG_ON(ret
&& ret
!= -ENOSPC
);
2074 key
.offset
= found_key
.offset
- 1;
2078 btrfs_free_path(path
);
2079 mutex_unlock(&dev_root
->fs_info
->volume_mutex
);
2084 * shrinking a device means finding all of the device extents past
2085 * the new size, and then following the back refs to the chunks.
2086 * The chunk relocation code actually frees the device extent
2088 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
2090 struct btrfs_trans_handle
*trans
;
2091 struct btrfs_root
*root
= device
->dev_root
;
2092 struct btrfs_dev_extent
*dev_extent
= NULL
;
2093 struct btrfs_path
*path
;
2101 bool retried
= false;
2102 struct extent_buffer
*l
;
2103 struct btrfs_key key
;
2104 struct btrfs_super_block
*super_copy
= &root
->fs_info
->super_copy
;
2105 u64 old_total
= btrfs_super_total_bytes(super_copy
);
2106 u64 old_size
= device
->total_bytes
;
2107 u64 diff
= device
->total_bytes
- new_size
;
2109 if (new_size
>= device
->total_bytes
)
2112 path
= btrfs_alloc_path();
2120 device
->total_bytes
= new_size
;
2121 if (device
->writeable
)
2122 device
->fs_devices
->total_rw_bytes
-= diff
;
2123 unlock_chunks(root
);
2126 key
.objectid
= device
->devid
;
2127 key
.offset
= (u64
)-1;
2128 key
.type
= BTRFS_DEV_EXTENT_KEY
;
2131 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2135 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
2140 btrfs_release_path(root
, path
);
2145 slot
= path
->slots
[0];
2146 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
2148 if (key
.objectid
!= device
->devid
) {
2149 btrfs_release_path(root
, path
);
2153 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
2154 length
= btrfs_dev_extent_length(l
, dev_extent
);
2156 if (key
.offset
+ length
<= new_size
) {
2157 btrfs_release_path(root
, path
);
2161 chunk_tree
= btrfs_dev_extent_chunk_tree(l
, dev_extent
);
2162 chunk_objectid
= btrfs_dev_extent_chunk_objectid(l
, dev_extent
);
2163 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
2164 btrfs_release_path(root
, path
);
2166 ret
= btrfs_relocate_chunk(root
, chunk_tree
, chunk_objectid
,
2168 if (ret
&& ret
!= -ENOSPC
)
2175 if (failed
&& !retried
) {
2179 } else if (failed
&& retried
) {
2183 device
->total_bytes
= old_size
;
2184 if (device
->writeable
)
2185 device
->fs_devices
->total_rw_bytes
+= diff
;
2186 unlock_chunks(root
);
2190 /* Shrinking succeeded, else we would be at "done". */
2191 trans
= btrfs_start_transaction(root
, 0);
2192 if (IS_ERR(trans
)) {
2193 ret
= PTR_ERR(trans
);
2199 device
->disk_total_bytes
= new_size
;
2200 /* Now btrfs_update_device() will change the on-disk size. */
2201 ret
= btrfs_update_device(trans
, device
);
2203 unlock_chunks(root
);
2204 btrfs_end_transaction(trans
, root
);
2207 WARN_ON(diff
> old_total
);
2208 btrfs_set_super_total_bytes(super_copy
, old_total
- diff
);
2209 unlock_chunks(root
);
2210 btrfs_end_transaction(trans
, root
);
2212 btrfs_free_path(path
);
2216 static int btrfs_add_system_chunk(struct btrfs_trans_handle
*trans
,
2217 struct btrfs_root
*root
,
2218 struct btrfs_key
*key
,
2219 struct btrfs_chunk
*chunk
, int item_size
)
2221 struct btrfs_super_block
*super_copy
= &root
->fs_info
->super_copy
;
2222 struct btrfs_disk_key disk_key
;
2226 array_size
= btrfs_super_sys_array_size(super_copy
);
2227 if (array_size
+ item_size
> BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
)
2230 ptr
= super_copy
->sys_chunk_array
+ array_size
;
2231 btrfs_cpu_key_to_disk(&disk_key
, key
);
2232 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
2233 ptr
+= sizeof(disk_key
);
2234 memcpy(ptr
, chunk
, item_size
);
2235 item_size
+= sizeof(disk_key
);
2236 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
2240 static noinline u64
chunk_bytes_by_type(u64 type
, u64 calc_size
,
2241 int num_stripes
, int sub_stripes
)
2243 if (type
& (BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_DUP
))
2245 else if (type
& BTRFS_BLOCK_GROUP_RAID10
)
2246 return calc_size
* (num_stripes
/ sub_stripes
);
2248 return calc_size
* num_stripes
;
2251 /* Used to sort the devices by max_avail(descending sort) */
2252 int btrfs_cmp_device_free_bytes(const void *dev_info1
, const void *dev_info2
)
2254 if (((struct btrfs_device_info
*)dev_info1
)->max_avail
>
2255 ((struct btrfs_device_info
*)dev_info2
)->max_avail
)
2257 else if (((struct btrfs_device_info
*)dev_info1
)->max_avail
<
2258 ((struct btrfs_device_info
*)dev_info2
)->max_avail
)
2264 static int __btrfs_calc_nstripes(struct btrfs_fs_devices
*fs_devices
, u64 type
,
2265 int *num_stripes
, int *min_stripes
,
2272 if (type
& (BTRFS_BLOCK_GROUP_RAID0
)) {
2273 *num_stripes
= fs_devices
->rw_devices
;
2276 if (type
& (BTRFS_BLOCK_GROUP_DUP
)) {
2280 if (type
& (BTRFS_BLOCK_GROUP_RAID1
)) {
2281 if (fs_devices
->rw_devices
< 2)
2286 if (type
& (BTRFS_BLOCK_GROUP_RAID10
)) {
2287 *num_stripes
= fs_devices
->rw_devices
;
2288 if (*num_stripes
< 4)
2290 *num_stripes
&= ~(u32
)1;
2298 static u64
__btrfs_calc_stripe_size(struct btrfs_fs_devices
*fs_devices
,
2299 u64 proposed_size
, u64 type
,
2300 int num_stripes
, int small_stripe
)
2302 int min_stripe_size
= 1 * 1024 * 1024;
2303 u64 calc_size
= proposed_size
;
2304 u64 max_chunk_size
= calc_size
;
2307 if (type
& (BTRFS_BLOCK_GROUP_RAID1
|
2308 BTRFS_BLOCK_GROUP_DUP
|
2309 BTRFS_BLOCK_GROUP_RAID10
))
2312 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
2313 max_chunk_size
= 10 * calc_size
;
2314 min_stripe_size
= 64 * 1024 * 1024;
2315 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
2316 max_chunk_size
= 256 * 1024 * 1024;
2317 min_stripe_size
= 32 * 1024 * 1024;
2318 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2319 calc_size
= 8 * 1024 * 1024;
2320 max_chunk_size
= calc_size
* 2;
2321 min_stripe_size
= 1 * 1024 * 1024;
2324 /* we don't want a chunk larger than 10% of writeable space */
2325 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
2328 if (calc_size
* num_stripes
> max_chunk_size
* ncopies
) {
2329 calc_size
= max_chunk_size
* ncopies
;
2330 do_div(calc_size
, num_stripes
);
2331 do_div(calc_size
, BTRFS_STRIPE_LEN
);
2332 calc_size
*= BTRFS_STRIPE_LEN
;
2335 /* we don't want tiny stripes */
2337 calc_size
= max_t(u64
, min_stripe_size
, calc_size
);
2340 * we're about to do_div by the BTRFS_STRIPE_LEN so lets make sure
2341 * we end up with something bigger than a stripe
2343 calc_size
= max_t(u64
, calc_size
, BTRFS_STRIPE_LEN
);
2345 do_div(calc_size
, BTRFS_STRIPE_LEN
);
2346 calc_size
*= BTRFS_STRIPE_LEN
;
2351 static struct map_lookup
*__shrink_map_lookup_stripes(struct map_lookup
*map
,
2354 struct map_lookup
*new;
2355 size_t len
= map_lookup_size(num_stripes
);
2357 BUG_ON(map
->num_stripes
< num_stripes
);
2359 if (map
->num_stripes
== num_stripes
)
2362 new = kmalloc(len
, GFP_NOFS
);
2364 /* just change map->num_stripes */
2365 map
->num_stripes
= num_stripes
;
2369 memcpy(new, map
, len
);
2370 new->num_stripes
= num_stripes
;
2376 * helper to allocate device space from btrfs_device_info, in which we stored
2377 * max free space information of every device. It is used when we can not
2378 * allocate chunks by default size.
2380 * By this helper, we can allocate a new chunk as larger as possible.
2382 static int __btrfs_alloc_tiny_space(struct btrfs_trans_handle
*trans
,
2383 struct btrfs_fs_devices
*fs_devices
,
2384 struct btrfs_device_info
*devices
,
2385 int nr_device
, u64 type
,
2386 struct map_lookup
**map_lookup
,
2387 int min_stripes
, u64
*stripe_size
)
2389 int i
, index
, sort_again
= 0;
2390 int min_devices
= min_stripes
;
2391 u64 max_avail
, min_free
;
2392 struct map_lookup
*map
= *map_lookup
;
2395 if (nr_device
< min_stripes
)
2398 btrfs_descending_sort_devices(devices
, nr_device
);
2400 max_avail
= devices
[0].max_avail
;
2404 for (i
= 0; i
< nr_device
; i
++) {
2406 * if dev_offset = 0, it means the free space of this device
2407 * is less than what we need, and we didn't search max avail
2408 * extent on this device, so do it now.
2410 if (!devices
[i
].dev_offset
) {
2411 ret
= find_free_dev_extent(trans
, devices
[i
].dev
,
2413 &devices
[i
].dev_offset
,
2414 &devices
[i
].max_avail
);
2415 if (ret
!= 0 && ret
!= -ENOSPC
)
2421 /* we update the max avail free extent of each devices, sort again */
2423 btrfs_descending_sort_devices(devices
, nr_device
);
2425 if (type
& BTRFS_BLOCK_GROUP_DUP
)
2428 if (!devices
[min_devices
- 1].max_avail
)
2431 max_avail
= devices
[min_devices
- 1].max_avail
;
2432 if (type
& BTRFS_BLOCK_GROUP_DUP
)
2433 do_div(max_avail
, 2);
2435 max_avail
= __btrfs_calc_stripe_size(fs_devices
, max_avail
, type
,
2437 if (type
& BTRFS_BLOCK_GROUP_DUP
)
2438 min_free
= max_avail
* 2;
2440 min_free
= max_avail
;
2442 if (min_free
> devices
[min_devices
- 1].max_avail
)
2445 map
= __shrink_map_lookup_stripes(map
, min_stripes
);
2446 *stripe_size
= max_avail
;
2449 for (i
= 0; i
< min_stripes
; i
++) {
2450 map
->stripes
[i
].dev
= devices
[index
].dev
;
2451 map
->stripes
[i
].physical
= devices
[index
].dev_offset
;
2452 if (type
& BTRFS_BLOCK_GROUP_DUP
) {
2454 map
->stripes
[i
].dev
= devices
[index
].dev
;
2455 map
->stripes
[i
].physical
= devices
[index
].dev_offset
+
2465 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
2466 struct btrfs_root
*extent_root
,
2467 struct map_lookup
**map_ret
,
2468 u64
*num_bytes
, u64
*stripe_size
,
2469 u64 start
, u64 type
)
2471 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
2472 struct btrfs_device
*device
= NULL
;
2473 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
2474 struct list_head
*cur
;
2475 struct map_lookup
*map
;
2476 struct extent_map_tree
*em_tree
;
2477 struct extent_map
*em
;
2478 struct btrfs_device_info
*devices_info
;
2479 struct list_head private_devs
;
2480 u64 calc_size
= 1024 * 1024 * 1024;
2487 int min_devices
; /* the min number of devices we need */
2492 if ((type
& BTRFS_BLOCK_GROUP_RAID1
) &&
2493 (type
& BTRFS_BLOCK_GROUP_DUP
)) {
2495 type
&= ~BTRFS_BLOCK_GROUP_DUP
;
2497 if (list_empty(&fs_devices
->alloc_list
))
2500 ret
= __btrfs_calc_nstripes(fs_devices
, type
, &num_stripes
,
2501 &min_stripes
, &sub_stripes
);
2505 devices_info
= kzalloc(sizeof(*devices_info
) * fs_devices
->rw_devices
,
2510 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
2515 map
->num_stripes
= num_stripes
;
2517 cur
= fs_devices
->alloc_list
.next
;
2521 calc_size
= __btrfs_calc_stripe_size(fs_devices
, calc_size
, type
,
2524 if (type
& BTRFS_BLOCK_GROUP_DUP
) {
2525 min_free
= calc_size
* 2;
2528 min_free
= calc_size
;
2529 min_devices
= min_stripes
;
2532 INIT_LIST_HEAD(&private_devs
);
2533 while (index
< num_stripes
) {
2534 device
= list_entry(cur
, struct btrfs_device
, dev_alloc_list
);
2535 BUG_ON(!device
->writeable
);
2536 if (device
->total_bytes
> device
->bytes_used
)
2537 avail
= device
->total_bytes
- device
->bytes_used
;
2542 if (device
->in_fs_metadata
&& avail
>= min_free
) {
2543 ret
= find_free_dev_extent(trans
, device
, min_free
,
2544 &devices_info
[i
].dev_offset
,
2545 &devices_info
[i
].max_avail
);
2547 list_move_tail(&device
->dev_alloc_list
,
2549 map
->stripes
[index
].dev
= device
;
2550 map
->stripes
[index
].physical
=
2551 devices_info
[i
].dev_offset
;
2553 if (type
& BTRFS_BLOCK_GROUP_DUP
) {
2554 map
->stripes
[index
].dev
= device
;
2555 map
->stripes
[index
].physical
=
2556 devices_info
[i
].dev_offset
+
2560 } else if (ret
!= -ENOSPC
)
2563 devices_info
[i
].dev
= device
;
2565 } else if (device
->in_fs_metadata
&&
2566 avail
>= BTRFS_STRIPE_LEN
) {
2567 devices_info
[i
].dev
= device
;
2568 devices_info
[i
].max_avail
= avail
;
2572 if (cur
== &fs_devices
->alloc_list
)
2576 list_splice(&private_devs
, &fs_devices
->alloc_list
);
2577 if (index
< num_stripes
) {
2578 if (index
>= min_stripes
) {
2579 num_stripes
= index
;
2580 if (type
& (BTRFS_BLOCK_GROUP_RAID10
)) {
2581 num_stripes
/= sub_stripes
;
2582 num_stripes
*= sub_stripes
;
2585 map
= __shrink_map_lookup_stripes(map
, num_stripes
);
2586 } else if (i
>= min_devices
) {
2587 ret
= __btrfs_alloc_tiny_space(trans
, fs_devices
,
2588 devices_info
, i
, type
,
2598 map
->sector_size
= extent_root
->sectorsize
;
2599 map
->stripe_len
= BTRFS_STRIPE_LEN
;
2600 map
->io_align
= BTRFS_STRIPE_LEN
;
2601 map
->io_width
= BTRFS_STRIPE_LEN
;
2603 map
->sub_stripes
= sub_stripes
;
2606 *stripe_size
= calc_size
;
2607 *num_bytes
= chunk_bytes_by_type(type
, calc_size
,
2608 map
->num_stripes
, sub_stripes
);
2610 trace_btrfs_chunk_alloc(info
->chunk_root
, map
, start
, *num_bytes
);
2612 em
= alloc_extent_map(GFP_NOFS
);
2617 em
->bdev
= (struct block_device
*)map
;
2619 em
->len
= *num_bytes
;
2620 em
->block_start
= 0;
2621 em
->block_len
= em
->len
;
2623 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
2624 write_lock(&em_tree
->lock
);
2625 ret
= add_extent_mapping(em_tree
, em
);
2626 write_unlock(&em_tree
->lock
);
2628 free_extent_map(em
);
2630 ret
= btrfs_make_block_group(trans
, extent_root
, 0, type
,
2631 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
2636 while (index
< map
->num_stripes
) {
2637 device
= map
->stripes
[index
].dev
;
2638 dev_offset
= map
->stripes
[index
].physical
;
2640 ret
= btrfs_alloc_dev_extent(trans
, device
,
2641 info
->chunk_root
->root_key
.objectid
,
2642 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
2643 start
, dev_offset
, calc_size
);
2648 kfree(devices_info
);
2653 kfree(devices_info
);
2657 static int __finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
2658 struct btrfs_root
*extent_root
,
2659 struct map_lookup
*map
, u64 chunk_offset
,
2660 u64 chunk_size
, u64 stripe_size
)
2663 struct btrfs_key key
;
2664 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
2665 struct btrfs_device
*device
;
2666 struct btrfs_chunk
*chunk
;
2667 struct btrfs_stripe
*stripe
;
2668 size_t item_size
= btrfs_chunk_item_size(map
->num_stripes
);
2672 chunk
= kzalloc(item_size
, GFP_NOFS
);
2677 while (index
< map
->num_stripes
) {
2678 device
= map
->stripes
[index
].dev
;
2679 device
->bytes_used
+= stripe_size
;
2680 ret
= btrfs_update_device(trans
, device
);
2686 stripe
= &chunk
->stripe
;
2687 while (index
< map
->num_stripes
) {
2688 device
= map
->stripes
[index
].dev
;
2689 dev_offset
= map
->stripes
[index
].physical
;
2691 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
2692 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
2693 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
2698 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
2699 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
2700 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
2701 btrfs_set_stack_chunk_type(chunk
, map
->type
);
2702 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
2703 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
2704 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
2705 btrfs_set_stack_chunk_sector_size(chunk
, extent_root
->sectorsize
);
2706 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
2708 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2709 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2710 key
.offset
= chunk_offset
;
2712 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
2715 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2716 ret
= btrfs_add_system_chunk(trans
, chunk_root
, &key
, chunk
,
2726 * Chunk allocation falls into two parts. The first part does works
2727 * that make the new allocated chunk useable, but not do any operation
2728 * that modifies the chunk tree. The second part does the works that
2729 * require modifying the chunk tree. This division is important for the
2730 * bootstrap process of adding storage to a seed btrfs.
2732 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
2733 struct btrfs_root
*extent_root
, u64 type
)
2738 struct map_lookup
*map
;
2739 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
2742 ret
= find_next_chunk(chunk_root
, BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
2747 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &map
, &chunk_size
,
2748 &stripe_size
, chunk_offset
, type
);
2752 ret
= __finish_chunk_alloc(trans
, extent_root
, map
, chunk_offset
,
2753 chunk_size
, stripe_size
);
2758 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
2759 struct btrfs_root
*root
,
2760 struct btrfs_device
*device
)
2763 u64 sys_chunk_offset
;
2767 u64 sys_stripe_size
;
2769 struct map_lookup
*map
;
2770 struct map_lookup
*sys_map
;
2771 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2772 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
2775 ret
= find_next_chunk(fs_info
->chunk_root
,
2776 BTRFS_FIRST_CHUNK_TREE_OBJECTID
, &chunk_offset
);
2779 alloc_profile
= BTRFS_BLOCK_GROUP_METADATA
|
2780 (fs_info
->metadata_alloc_profile
&
2781 fs_info
->avail_metadata_alloc_bits
);
2782 alloc_profile
= btrfs_reduce_alloc_profile(root
, alloc_profile
);
2784 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &map
, &chunk_size
,
2785 &stripe_size
, chunk_offset
, alloc_profile
);
2788 sys_chunk_offset
= chunk_offset
+ chunk_size
;
2790 alloc_profile
= BTRFS_BLOCK_GROUP_SYSTEM
|
2791 (fs_info
->system_alloc_profile
&
2792 fs_info
->avail_system_alloc_bits
);
2793 alloc_profile
= btrfs_reduce_alloc_profile(root
, alloc_profile
);
2795 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &sys_map
,
2796 &sys_chunk_size
, &sys_stripe_size
,
2797 sys_chunk_offset
, alloc_profile
);
2800 ret
= btrfs_add_device(trans
, fs_info
->chunk_root
, device
);
2804 * Modifying chunk tree needs allocating new blocks from both
2805 * system block group and metadata block group. So we only can
2806 * do operations require modifying the chunk tree after both
2807 * block groups were created.
2809 ret
= __finish_chunk_alloc(trans
, extent_root
, map
, chunk_offset
,
2810 chunk_size
, stripe_size
);
2813 ret
= __finish_chunk_alloc(trans
, extent_root
, sys_map
,
2814 sys_chunk_offset
, sys_chunk_size
,
2820 int btrfs_chunk_readonly(struct btrfs_root
*root
, u64 chunk_offset
)
2822 struct extent_map
*em
;
2823 struct map_lookup
*map
;
2824 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
2828 read_lock(&map_tree
->map_tree
.lock
);
2829 em
= lookup_extent_mapping(&map_tree
->map_tree
, chunk_offset
, 1);
2830 read_unlock(&map_tree
->map_tree
.lock
);
2834 if (btrfs_test_opt(root
, DEGRADED
)) {
2835 free_extent_map(em
);
2839 map
= (struct map_lookup
*)em
->bdev
;
2840 for (i
= 0; i
< map
->num_stripes
; i
++) {
2841 if (!map
->stripes
[i
].dev
->writeable
) {
2846 free_extent_map(em
);
2850 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
2852 extent_map_tree_init(&tree
->map_tree
, GFP_NOFS
);
2855 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
2857 struct extent_map
*em
;
2860 write_lock(&tree
->map_tree
.lock
);
2861 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
2863 remove_extent_mapping(&tree
->map_tree
, em
);
2864 write_unlock(&tree
->map_tree
.lock
);
2869 free_extent_map(em
);
2870 /* once for the tree */
2871 free_extent_map(em
);
2875 int btrfs_num_copies(struct btrfs_mapping_tree
*map_tree
, u64 logical
, u64 len
)
2877 struct extent_map
*em
;
2878 struct map_lookup
*map
;
2879 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
2882 read_lock(&em_tree
->lock
);
2883 em
= lookup_extent_mapping(em_tree
, logical
, len
);
2884 read_unlock(&em_tree
->lock
);
2887 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
2888 map
= (struct map_lookup
*)em
->bdev
;
2889 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
2890 ret
= map
->num_stripes
;
2891 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
2892 ret
= map
->sub_stripes
;
2895 free_extent_map(em
);
2899 static int find_live_mirror(struct map_lookup
*map
, int first
, int num
,
2903 if (map
->stripes
[optimal
].dev
->bdev
)
2905 for (i
= first
; i
< first
+ num
; i
++) {
2906 if (map
->stripes
[i
].dev
->bdev
)
2909 /* we couldn't find one that doesn't fail. Just return something
2910 * and the io error handling code will clean up eventually
2915 static int __btrfs_map_block(struct btrfs_mapping_tree
*map_tree
, int rw
,
2916 u64 logical
, u64
*length
,
2917 struct btrfs_multi_bio
**multi_ret
,
2920 struct extent_map
*em
;
2921 struct map_lookup
*map
;
2922 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
2925 u64 stripe_end_offset
;
2929 int stripes_allocated
= 8;
2930 int stripes_required
= 1;
2935 struct btrfs_multi_bio
*multi
= NULL
;
2937 if (multi_ret
&& !(rw
& (REQ_WRITE
| REQ_DISCARD
)))
2938 stripes_allocated
= 1;
2941 multi
= kzalloc(btrfs_multi_bio_size(stripes_allocated
),
2946 atomic_set(&multi
->error
, 0);
2949 read_lock(&em_tree
->lock
);
2950 em
= lookup_extent_mapping(em_tree
, logical
, *length
);
2951 read_unlock(&em_tree
->lock
);
2954 printk(KERN_CRIT
"unable to find logical %llu len %llu\n",
2955 (unsigned long long)logical
,
2956 (unsigned long long)*length
);
2960 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
2961 map
= (struct map_lookup
*)em
->bdev
;
2962 offset
= logical
- em
->start
;
2964 if (mirror_num
> map
->num_stripes
)
2967 /* if our multi bio struct is too small, back off and try again */
2968 if (rw
& REQ_WRITE
) {
2969 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
2970 BTRFS_BLOCK_GROUP_DUP
)) {
2971 stripes_required
= map
->num_stripes
;
2973 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
2974 stripes_required
= map
->sub_stripes
;
2978 if (rw
& REQ_DISCARD
) {
2979 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
2980 BTRFS_BLOCK_GROUP_RAID1
|
2981 BTRFS_BLOCK_GROUP_DUP
|
2982 BTRFS_BLOCK_GROUP_RAID10
)) {
2983 stripes_required
= map
->num_stripes
;
2986 if (multi_ret
&& (rw
& (REQ_WRITE
| REQ_DISCARD
)) &&
2987 stripes_allocated
< stripes_required
) {
2988 stripes_allocated
= map
->num_stripes
;
2989 free_extent_map(em
);
2995 * stripe_nr counts the total number of stripes we have to stride
2996 * to get to this block
2998 do_div(stripe_nr
, map
->stripe_len
);
3000 stripe_offset
= stripe_nr
* map
->stripe_len
;
3001 BUG_ON(offset
< stripe_offset
);
3003 /* stripe_offset is the offset of this block in its stripe*/
3004 stripe_offset
= offset
- stripe_offset
;
3006 if (rw
& REQ_DISCARD
)
3007 *length
= min_t(u64
, em
->len
- offset
, *length
);
3008 else if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
3009 BTRFS_BLOCK_GROUP_RAID1
|
3010 BTRFS_BLOCK_GROUP_RAID10
|
3011 BTRFS_BLOCK_GROUP_DUP
)) {
3012 /* we limit the length of each bio to what fits in a stripe */
3013 *length
= min_t(u64
, em
->len
- offset
,
3014 map
->stripe_len
- stripe_offset
);
3016 *length
= em
->len
- offset
;
3024 stripe_nr_orig
= stripe_nr
;
3025 stripe_nr_end
= (offset
+ *length
+ map
->stripe_len
- 1) &
3026 (~(map
->stripe_len
- 1));
3027 do_div(stripe_nr_end
, map
->stripe_len
);
3028 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
3030 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
3031 if (rw
& REQ_DISCARD
)
3032 num_stripes
= min_t(u64
, map
->num_stripes
,
3033 stripe_nr_end
- stripe_nr_orig
);
3034 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
3035 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
3036 if (rw
& (REQ_WRITE
| REQ_DISCARD
))
3037 num_stripes
= map
->num_stripes
;
3038 else if (mirror_num
)
3039 stripe_index
= mirror_num
- 1;
3041 stripe_index
= find_live_mirror(map
, 0,
3043 current
->pid
% map
->num_stripes
);
3046 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
3047 if (rw
& (REQ_WRITE
| REQ_DISCARD
))
3048 num_stripes
= map
->num_stripes
;
3049 else if (mirror_num
)
3050 stripe_index
= mirror_num
- 1;
3052 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
3053 int factor
= map
->num_stripes
/ map
->sub_stripes
;
3055 stripe_index
= do_div(stripe_nr
, factor
);
3056 stripe_index
*= map
->sub_stripes
;
3059 num_stripes
= map
->sub_stripes
;
3060 else if (rw
& REQ_DISCARD
)
3061 num_stripes
= min_t(u64
, map
->sub_stripes
*
3062 (stripe_nr_end
- stripe_nr_orig
),
3064 else if (mirror_num
)
3065 stripe_index
+= mirror_num
- 1;
3067 stripe_index
= find_live_mirror(map
, stripe_index
,
3068 map
->sub_stripes
, stripe_index
+
3069 current
->pid
% map
->sub_stripes
);
3073 * after this do_div call, stripe_nr is the number of stripes
3074 * on this device we have to walk to find the data, and
3075 * stripe_index is the number of our device in the stripe array
3077 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
3079 BUG_ON(stripe_index
>= map
->num_stripes
);
3081 if (rw
& REQ_DISCARD
) {
3082 for (i
= 0; i
< num_stripes
; i
++) {
3083 multi
->stripes
[i
].physical
=
3084 map
->stripes
[stripe_index
].physical
+
3085 stripe_offset
+ stripe_nr
* map
->stripe_len
;
3086 multi
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
3088 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
3090 u32 last_stripe
= 0;
3093 div_u64_rem(stripe_nr_end
- 1,
3097 for (j
= 0; j
< map
->num_stripes
; j
++) {
3100 div_u64_rem(stripe_nr_end
- 1 - j
,
3101 map
->num_stripes
, &test
);
3102 if (test
== stripe_index
)
3105 stripes
= stripe_nr_end
- 1 - j
;
3106 do_div(stripes
, map
->num_stripes
);
3107 multi
->stripes
[i
].length
= map
->stripe_len
*
3108 (stripes
- stripe_nr
+ 1);
3111 multi
->stripes
[i
].length
-=
3115 if (stripe_index
== last_stripe
)
3116 multi
->stripes
[i
].length
-=
3118 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
3121 int factor
= map
->num_stripes
/
3123 u32 last_stripe
= 0;
3125 div_u64_rem(stripe_nr_end
- 1,
3126 factor
, &last_stripe
);
3127 last_stripe
*= map
->sub_stripes
;
3129 for (j
= 0; j
< factor
; j
++) {
3132 div_u64_rem(stripe_nr_end
- 1 - j
,
3136 stripe_index
/ map
->sub_stripes
)
3139 stripes
= stripe_nr_end
- 1 - j
;
3140 do_div(stripes
, factor
);
3141 multi
->stripes
[i
].length
= map
->stripe_len
*
3142 (stripes
- stripe_nr
+ 1);
3144 if (i
< map
->sub_stripes
) {
3145 multi
->stripes
[i
].length
-=
3147 if (i
== map
->sub_stripes
- 1)
3150 if (stripe_index
>= last_stripe
&&
3151 stripe_index
<= (last_stripe
+
3152 map
->sub_stripes
- 1)) {
3153 multi
->stripes
[i
].length
-=
3157 multi
->stripes
[i
].length
= *length
;
3160 if (stripe_index
== map
->num_stripes
) {
3161 /* This could only happen for RAID0/10 */
3167 for (i
= 0; i
< num_stripes
; i
++) {
3168 multi
->stripes
[i
].physical
=
3169 map
->stripes
[stripe_index
].physical
+
3171 stripe_nr
* map
->stripe_len
;
3172 multi
->stripes
[i
].dev
=
3173 map
->stripes
[stripe_index
].dev
;
3179 multi
->num_stripes
= num_stripes
;
3180 multi
->max_errors
= max_errors
;
3183 free_extent_map(em
);
3187 int btrfs_map_block(struct btrfs_mapping_tree
*map_tree
, int rw
,
3188 u64 logical
, u64
*length
,
3189 struct btrfs_multi_bio
**multi_ret
, int mirror_num
)
3191 return __btrfs_map_block(map_tree
, rw
, logical
, length
, multi_ret
,
3195 int btrfs_rmap_block(struct btrfs_mapping_tree
*map_tree
,
3196 u64 chunk_start
, u64 physical
, u64 devid
,
3197 u64
**logical
, int *naddrs
, int *stripe_len
)
3199 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
3200 struct extent_map
*em
;
3201 struct map_lookup
*map
;
3208 read_lock(&em_tree
->lock
);
3209 em
= lookup_extent_mapping(em_tree
, chunk_start
, 1);
3210 read_unlock(&em_tree
->lock
);
3212 BUG_ON(!em
|| em
->start
!= chunk_start
);
3213 map
= (struct map_lookup
*)em
->bdev
;
3216 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
3217 do_div(length
, map
->num_stripes
/ map
->sub_stripes
);
3218 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
3219 do_div(length
, map
->num_stripes
);
3221 buf
= kzalloc(sizeof(u64
) * map
->num_stripes
, GFP_NOFS
);
3224 for (i
= 0; i
< map
->num_stripes
; i
++) {
3225 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
3227 if (map
->stripes
[i
].physical
> physical
||
3228 map
->stripes
[i
].physical
+ length
<= physical
)
3231 stripe_nr
= physical
- map
->stripes
[i
].physical
;
3232 do_div(stripe_nr
, map
->stripe_len
);
3234 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
3235 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
3236 do_div(stripe_nr
, map
->sub_stripes
);
3237 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
3238 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
3240 bytenr
= chunk_start
+ stripe_nr
* map
->stripe_len
;
3241 WARN_ON(nr
>= map
->num_stripes
);
3242 for (j
= 0; j
< nr
; j
++) {
3243 if (buf
[j
] == bytenr
)
3247 WARN_ON(nr
>= map
->num_stripes
);
3254 *stripe_len
= map
->stripe_len
;
3256 free_extent_map(em
);
3260 static void end_bio_multi_stripe(struct bio
*bio
, int err
)
3262 struct btrfs_multi_bio
*multi
= bio
->bi_private
;
3263 int is_orig_bio
= 0;
3266 atomic_inc(&multi
->error
);
3268 if (bio
== multi
->orig_bio
)
3271 if (atomic_dec_and_test(&multi
->stripes_pending
)) {
3274 bio
= multi
->orig_bio
;
3276 bio
->bi_private
= multi
->private;
3277 bio
->bi_end_io
= multi
->end_io
;
3278 /* only send an error to the higher layers if it is
3279 * beyond the tolerance of the multi-bio
3281 if (atomic_read(&multi
->error
) > multi
->max_errors
) {
3285 * this bio is actually up to date, we didn't
3286 * go over the max number of errors
3288 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
3293 bio_endio(bio
, err
);
3294 } else if (!is_orig_bio
) {
3299 struct async_sched
{
3302 struct btrfs_fs_info
*info
;
3303 struct btrfs_work work
;
3307 * see run_scheduled_bios for a description of why bios are collected for
3310 * This will add one bio to the pending list for a device and make sure
3311 * the work struct is scheduled.
3313 static noinline
int schedule_bio(struct btrfs_root
*root
,
3314 struct btrfs_device
*device
,
3315 int rw
, struct bio
*bio
)
3317 int should_queue
= 1;
3318 struct btrfs_pending_bios
*pending_bios
;
3320 /* don't bother with additional async steps for reads, right now */
3321 if (!(rw
& REQ_WRITE
)) {
3323 submit_bio(rw
, bio
);
3329 * nr_async_bios allows us to reliably return congestion to the
3330 * higher layers. Otherwise, the async bio makes it appear we have
3331 * made progress against dirty pages when we've really just put it
3332 * on a queue for later
3334 atomic_inc(&root
->fs_info
->nr_async_bios
);
3335 WARN_ON(bio
->bi_next
);
3336 bio
->bi_next
= NULL
;
3339 spin_lock(&device
->io_lock
);
3340 if (bio
->bi_rw
& REQ_SYNC
)
3341 pending_bios
= &device
->pending_sync_bios
;
3343 pending_bios
= &device
->pending_bios
;
3345 if (pending_bios
->tail
)
3346 pending_bios
->tail
->bi_next
= bio
;
3348 pending_bios
->tail
= bio
;
3349 if (!pending_bios
->head
)
3350 pending_bios
->head
= bio
;
3351 if (device
->running_pending
)
3354 spin_unlock(&device
->io_lock
);
3357 btrfs_queue_worker(&root
->fs_info
->submit_workers
,
3362 int btrfs_map_bio(struct btrfs_root
*root
, int rw
, struct bio
*bio
,
3363 int mirror_num
, int async_submit
)
3365 struct btrfs_mapping_tree
*map_tree
;
3366 struct btrfs_device
*dev
;
3367 struct bio
*first_bio
= bio
;
3368 u64 logical
= (u64
)bio
->bi_sector
<< 9;
3371 struct btrfs_multi_bio
*multi
= NULL
;
3376 length
= bio
->bi_size
;
3377 map_tree
= &root
->fs_info
->mapping_tree
;
3378 map_length
= length
;
3380 ret
= btrfs_map_block(map_tree
, rw
, logical
, &map_length
, &multi
,
3384 total_devs
= multi
->num_stripes
;
3385 if (map_length
< length
) {
3386 printk(KERN_CRIT
"mapping failed logical %llu bio len %llu "
3387 "len %llu\n", (unsigned long long)logical
,
3388 (unsigned long long)length
,
3389 (unsigned long long)map_length
);
3392 multi
->end_io
= first_bio
->bi_end_io
;
3393 multi
->private = first_bio
->bi_private
;
3394 multi
->orig_bio
= first_bio
;
3395 atomic_set(&multi
->stripes_pending
, multi
->num_stripes
);
3397 while (dev_nr
< total_devs
) {
3398 if (total_devs
> 1) {
3399 if (dev_nr
< total_devs
- 1) {
3400 bio
= bio_clone(first_bio
, GFP_NOFS
);
3405 bio
->bi_private
= multi
;
3406 bio
->bi_end_io
= end_bio_multi_stripe
;
3408 bio
->bi_sector
= multi
->stripes
[dev_nr
].physical
>> 9;
3409 dev
= multi
->stripes
[dev_nr
].dev
;
3410 if (dev
&& dev
->bdev
&& (rw
!= WRITE
|| dev
->writeable
)) {
3411 bio
->bi_bdev
= dev
->bdev
;
3413 schedule_bio(root
, dev
, rw
, bio
);
3415 submit_bio(rw
, bio
);
3417 bio
->bi_bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
3418 bio
->bi_sector
= logical
>> 9;
3419 bio_endio(bio
, -EIO
);
3423 if (total_devs
== 1)
3428 struct btrfs_device
*btrfs_find_device(struct btrfs_root
*root
, u64 devid
,
3431 struct btrfs_device
*device
;
3432 struct btrfs_fs_devices
*cur_devices
;
3434 cur_devices
= root
->fs_info
->fs_devices
;
3435 while (cur_devices
) {
3437 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
3438 device
= __find_device(&cur_devices
->devices
,
3443 cur_devices
= cur_devices
->seed
;
3448 static struct btrfs_device
*add_missing_dev(struct btrfs_root
*root
,
3449 u64 devid
, u8
*dev_uuid
)
3451 struct btrfs_device
*device
;
3452 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
3454 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
3457 list_add(&device
->dev_list
,
3458 &fs_devices
->devices
);
3459 device
->dev_root
= root
->fs_info
->dev_root
;
3460 device
->devid
= devid
;
3461 device
->work
.func
= pending_bios_fn
;
3462 device
->fs_devices
= fs_devices
;
3463 device
->missing
= 1;
3464 fs_devices
->num_devices
++;
3465 fs_devices
->missing_devices
++;
3466 spin_lock_init(&device
->io_lock
);
3467 INIT_LIST_HEAD(&device
->dev_alloc_list
);
3468 memcpy(device
->uuid
, dev_uuid
, BTRFS_UUID_SIZE
);
3472 static int read_one_chunk(struct btrfs_root
*root
, struct btrfs_key
*key
,
3473 struct extent_buffer
*leaf
,
3474 struct btrfs_chunk
*chunk
)
3476 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
3477 struct map_lookup
*map
;
3478 struct extent_map
*em
;
3482 u8 uuid
[BTRFS_UUID_SIZE
];
3487 logical
= key
->offset
;
3488 length
= btrfs_chunk_length(leaf
, chunk
);
3490 read_lock(&map_tree
->map_tree
.lock
);
3491 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
3492 read_unlock(&map_tree
->map_tree
.lock
);
3494 /* already mapped? */
3495 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
3496 free_extent_map(em
);
3499 free_extent_map(em
);
3502 em
= alloc_extent_map(GFP_NOFS
);
3505 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3506 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
3508 free_extent_map(em
);
3512 em
->bdev
= (struct block_device
*)map
;
3513 em
->start
= logical
;
3515 em
->block_start
= 0;
3516 em
->block_len
= em
->len
;
3518 map
->num_stripes
= num_stripes
;
3519 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
3520 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
3521 map
->sector_size
= btrfs_chunk_sector_size(leaf
, chunk
);
3522 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
3523 map
->type
= btrfs_chunk_type(leaf
, chunk
);
3524 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
3525 for (i
= 0; i
< num_stripes
; i
++) {
3526 map
->stripes
[i
].physical
=
3527 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
3528 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
3529 read_extent_buffer(leaf
, uuid
, (unsigned long)
3530 btrfs_stripe_dev_uuid_nr(chunk
, i
),
3532 map
->stripes
[i
].dev
= btrfs_find_device(root
, devid
, uuid
,
3534 if (!map
->stripes
[i
].dev
&& !btrfs_test_opt(root
, DEGRADED
)) {
3536 free_extent_map(em
);
3539 if (!map
->stripes
[i
].dev
) {
3540 map
->stripes
[i
].dev
=
3541 add_missing_dev(root
, devid
, uuid
);
3542 if (!map
->stripes
[i
].dev
) {
3544 free_extent_map(em
);
3548 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
3551 write_lock(&map_tree
->map_tree
.lock
);
3552 ret
= add_extent_mapping(&map_tree
->map_tree
, em
);
3553 write_unlock(&map_tree
->map_tree
.lock
);
3555 free_extent_map(em
);
3560 static int fill_device_from_item(struct extent_buffer
*leaf
,
3561 struct btrfs_dev_item
*dev_item
,
3562 struct btrfs_device
*device
)
3566 device
->devid
= btrfs_device_id(leaf
, dev_item
);
3567 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
3568 device
->total_bytes
= device
->disk_total_bytes
;
3569 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
3570 device
->type
= btrfs_device_type(leaf
, dev_item
);
3571 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
3572 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
3573 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
3575 ptr
= (unsigned long)btrfs_device_uuid(dev_item
);
3576 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
3581 static int open_seed_devices(struct btrfs_root
*root
, u8
*fsid
)
3583 struct btrfs_fs_devices
*fs_devices
;
3586 mutex_lock(&uuid_mutex
);
3588 fs_devices
= root
->fs_info
->fs_devices
->seed
;
3589 while (fs_devices
) {
3590 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
3594 fs_devices
= fs_devices
->seed
;
3597 fs_devices
= find_fsid(fsid
);
3603 fs_devices
= clone_fs_devices(fs_devices
);
3604 if (IS_ERR(fs_devices
)) {
3605 ret
= PTR_ERR(fs_devices
);
3609 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
3610 root
->fs_info
->bdev_holder
);
3614 if (!fs_devices
->seeding
) {
3615 __btrfs_close_devices(fs_devices
);
3616 free_fs_devices(fs_devices
);
3621 fs_devices
->seed
= root
->fs_info
->fs_devices
->seed
;
3622 root
->fs_info
->fs_devices
->seed
= fs_devices
;
3624 mutex_unlock(&uuid_mutex
);
3628 static int read_one_dev(struct btrfs_root
*root
,
3629 struct extent_buffer
*leaf
,
3630 struct btrfs_dev_item
*dev_item
)
3632 struct btrfs_device
*device
;
3635 u8 fs_uuid
[BTRFS_UUID_SIZE
];
3636 u8 dev_uuid
[BTRFS_UUID_SIZE
];
3638 devid
= btrfs_device_id(leaf
, dev_item
);
3639 read_extent_buffer(leaf
, dev_uuid
,
3640 (unsigned long)btrfs_device_uuid(dev_item
),
3642 read_extent_buffer(leaf
, fs_uuid
,
3643 (unsigned long)btrfs_device_fsid(dev_item
),
3646 if (memcmp(fs_uuid
, root
->fs_info
->fsid
, BTRFS_UUID_SIZE
)) {
3647 ret
= open_seed_devices(root
, fs_uuid
);
3648 if (ret
&& !btrfs_test_opt(root
, DEGRADED
))
3652 device
= btrfs_find_device(root
, devid
, dev_uuid
, fs_uuid
);
3653 if (!device
|| !device
->bdev
) {
3654 if (!btrfs_test_opt(root
, DEGRADED
))
3658 printk(KERN_WARNING
"warning devid %llu missing\n",
3659 (unsigned long long)devid
);
3660 device
= add_missing_dev(root
, devid
, dev_uuid
);
3663 } else if (!device
->missing
) {
3665 * this happens when a device that was properly setup
3666 * in the device info lists suddenly goes bad.
3667 * device->bdev is NULL, and so we have to set
3668 * device->missing to one here
3670 root
->fs_info
->fs_devices
->missing_devices
++;
3671 device
->missing
= 1;
3675 if (device
->fs_devices
!= root
->fs_info
->fs_devices
) {
3676 BUG_ON(device
->writeable
);
3677 if (device
->generation
!=
3678 btrfs_device_generation(leaf
, dev_item
))
3682 fill_device_from_item(leaf
, dev_item
, device
);
3683 device
->dev_root
= root
->fs_info
->dev_root
;
3684 device
->in_fs_metadata
= 1;
3685 if (device
->writeable
)
3686 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
3691 int btrfs_read_super_device(struct btrfs_root
*root
, struct extent_buffer
*buf
)
3693 struct btrfs_dev_item
*dev_item
;
3695 dev_item
= (struct btrfs_dev_item
*)offsetof(struct btrfs_super_block
,
3697 return read_one_dev(root
, buf
, dev_item
);
3700 int btrfs_read_sys_array(struct btrfs_root
*root
)
3702 struct btrfs_super_block
*super_copy
= &root
->fs_info
->super_copy
;
3703 struct extent_buffer
*sb
;
3704 struct btrfs_disk_key
*disk_key
;
3705 struct btrfs_chunk
*chunk
;
3707 unsigned long sb_ptr
;
3713 struct btrfs_key key
;
3715 sb
= btrfs_find_create_tree_block(root
, BTRFS_SUPER_INFO_OFFSET
,
3716 BTRFS_SUPER_INFO_SIZE
);
3719 btrfs_set_buffer_uptodate(sb
);
3720 btrfs_set_buffer_lockdep_class(sb
, 0);
3722 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
3723 array_size
= btrfs_super_sys_array_size(super_copy
);
3725 ptr
= super_copy
->sys_chunk_array
;
3726 sb_ptr
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
3729 while (cur
< array_size
) {
3730 disk_key
= (struct btrfs_disk_key
*)ptr
;
3731 btrfs_disk_key_to_cpu(&key
, disk_key
);
3733 len
= sizeof(*disk_key
); ptr
+= len
;
3737 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
3738 chunk
= (struct btrfs_chunk
*)sb_ptr
;
3739 ret
= read_one_chunk(root
, &key
, sb
, chunk
);
3742 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
3743 len
= btrfs_chunk_item_size(num_stripes
);
3752 free_extent_buffer(sb
);
3756 int btrfs_read_chunk_tree(struct btrfs_root
*root
)
3758 struct btrfs_path
*path
;
3759 struct extent_buffer
*leaf
;
3760 struct btrfs_key key
;
3761 struct btrfs_key found_key
;
3765 root
= root
->fs_info
->chunk_root
;
3767 path
= btrfs_alloc_path();
3771 /* first we search for all of the device items, and then we
3772 * read in all of the chunk items. This way we can create chunk
3773 * mappings that reference all of the devices that are afound
3775 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
3779 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3783 leaf
= path
->nodes
[0];
3784 slot
= path
->slots
[0];
3785 if (slot
>= btrfs_header_nritems(leaf
)) {
3786 ret
= btrfs_next_leaf(root
, path
);
3793 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3794 if (key
.objectid
== BTRFS_DEV_ITEMS_OBJECTID
) {
3795 if (found_key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
)
3797 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
3798 struct btrfs_dev_item
*dev_item
;
3799 dev_item
= btrfs_item_ptr(leaf
, slot
,
3800 struct btrfs_dev_item
);
3801 ret
= read_one_dev(root
, leaf
, dev_item
);
3805 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
3806 struct btrfs_chunk
*chunk
;
3807 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
3808 ret
= read_one_chunk(root
, &found_key
, leaf
, chunk
);
3814 if (key
.objectid
== BTRFS_DEV_ITEMS_OBJECTID
) {
3816 btrfs_release_path(root
, path
);
3821 btrfs_free_path(path
);