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/kthread.h>
27 #include <asm/div64.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"
38 static int init_first_rw_device(struct btrfs_trans_handle
*trans
,
39 struct btrfs_root
*root
,
40 struct btrfs_device
*device
);
41 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
);
43 static DEFINE_MUTEX(uuid_mutex
);
44 static LIST_HEAD(fs_uuids
);
46 static void lock_chunks(struct btrfs_root
*root
)
48 mutex_lock(&root
->fs_info
->chunk_mutex
);
51 static void unlock_chunks(struct btrfs_root
*root
)
53 mutex_unlock(&root
->fs_info
->chunk_mutex
);
56 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
58 struct btrfs_device
*device
;
59 WARN_ON(fs_devices
->opened
);
60 while (!list_empty(&fs_devices
->devices
)) {
61 device
= list_entry(fs_devices
->devices
.next
,
62 struct btrfs_device
, dev_list
);
63 list_del(&device
->dev_list
);
70 int btrfs_cleanup_fs_uuids(void)
72 struct btrfs_fs_devices
*fs_devices
;
74 while (!list_empty(&fs_uuids
)) {
75 fs_devices
= list_entry(fs_uuids
.next
,
76 struct btrfs_fs_devices
, list
);
77 list_del(&fs_devices
->list
);
78 free_fs_devices(fs_devices
);
83 static noinline
struct btrfs_device
*__find_device(struct list_head
*head
,
86 struct btrfs_device
*dev
;
88 list_for_each_entry(dev
, head
, dev_list
) {
89 if (dev
->devid
== devid
&&
90 (!uuid
|| !memcmp(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
))) {
97 static noinline
struct btrfs_fs_devices
*find_fsid(u8
*fsid
)
99 struct btrfs_fs_devices
*fs_devices
;
101 list_for_each_entry(fs_devices
, &fs_uuids
, list
) {
102 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
108 static void requeue_list(struct btrfs_pending_bios
*pending_bios
,
109 struct bio
*head
, struct bio
*tail
)
112 struct bio
*old_head
;
114 old_head
= pending_bios
->head
;
115 pending_bios
->head
= head
;
116 if (pending_bios
->tail
)
117 tail
->bi_next
= old_head
;
119 pending_bios
->tail
= tail
;
123 * we try to collect pending bios for a device so we don't get a large
124 * number of procs sending bios down to the same device. This greatly
125 * improves the schedulers ability to collect and merge the bios.
127 * But, it also turns into a long list of bios to process and that is sure
128 * to eventually make the worker thread block. The solution here is to
129 * make some progress and then put this work struct back at the end of
130 * the list if the block device is congested. This way, multiple devices
131 * can make progress from a single worker thread.
133 static noinline
int run_scheduled_bios(struct btrfs_device
*device
)
136 struct backing_dev_info
*bdi
;
137 struct btrfs_fs_info
*fs_info
;
138 struct btrfs_pending_bios
*pending_bios
;
142 unsigned long num_run
;
143 unsigned long batch_run
= 0;
145 unsigned long last_waited
= 0;
147 int sync_pending
= 0;
148 struct blk_plug plug
;
151 * this function runs all the bios we've collected for
152 * a particular device. We don't want to wander off to
153 * another device without first sending all of these down.
154 * So, setup a plug here and finish it off before we return
156 blk_start_plug(&plug
);
158 bdi
= blk_get_backing_dev_info(device
->bdev
);
159 fs_info
= device
->dev_root
->fs_info
;
160 limit
= btrfs_async_submit_limit(fs_info
);
161 limit
= limit
* 2 / 3;
164 spin_lock(&device
->io_lock
);
169 /* take all the bios off the list at once and process them
170 * later on (without the lock held). But, remember the
171 * tail and other pointers so the bios can be properly reinserted
172 * into the list if we hit congestion
174 if (!force_reg
&& device
->pending_sync_bios
.head
) {
175 pending_bios
= &device
->pending_sync_bios
;
178 pending_bios
= &device
->pending_bios
;
182 pending
= pending_bios
->head
;
183 tail
= pending_bios
->tail
;
184 WARN_ON(pending
&& !tail
);
187 * if pending was null this time around, no bios need processing
188 * at all and we can stop. Otherwise it'll loop back up again
189 * and do an additional check so no bios are missed.
191 * device->running_pending is used to synchronize with the
194 if (device
->pending_sync_bios
.head
== NULL
&&
195 device
->pending_bios
.head
== NULL
) {
197 device
->running_pending
= 0;
200 device
->running_pending
= 1;
203 pending_bios
->head
= NULL
;
204 pending_bios
->tail
= NULL
;
206 spin_unlock(&device
->io_lock
);
211 /* we want to work on both lists, but do more bios on the
212 * sync list than the regular list
215 pending_bios
!= &device
->pending_sync_bios
&&
216 device
->pending_sync_bios
.head
) ||
217 (num_run
> 64 && pending_bios
== &device
->pending_sync_bios
&&
218 device
->pending_bios
.head
)) {
219 spin_lock(&device
->io_lock
);
220 requeue_list(pending_bios
, pending
, tail
);
225 pending
= pending
->bi_next
;
227 atomic_dec(&fs_info
->nr_async_bios
);
229 if (atomic_read(&fs_info
->nr_async_bios
) < limit
&&
230 waitqueue_active(&fs_info
->async_submit_wait
))
231 wake_up(&fs_info
->async_submit_wait
);
233 BUG_ON(atomic_read(&cur
->bi_cnt
) == 0);
236 * if we're doing the sync list, record that our
237 * plug has some sync requests on it
239 * If we're doing the regular list and there are
240 * sync requests sitting around, unplug before
243 if (pending_bios
== &device
->pending_sync_bios
) {
245 } else if (sync_pending
) {
246 blk_finish_plug(&plug
);
247 blk_start_plug(&plug
);
251 btrfsic_submit_bio(cur
->bi_rw
, cur
);
258 * we made progress, there is more work to do and the bdi
259 * is now congested. Back off and let other work structs
262 if (pending
&& bdi_write_congested(bdi
) && batch_run
> 8 &&
263 fs_info
->fs_devices
->open_devices
> 1) {
264 struct io_context
*ioc
;
266 ioc
= current
->io_context
;
269 * the main goal here is that we don't want to
270 * block if we're going to be able to submit
271 * more requests without blocking.
273 * This code does two great things, it pokes into
274 * the elevator code from a filesystem _and_
275 * it makes assumptions about how batching works.
277 if (ioc
&& ioc
->nr_batch_requests
> 0 &&
278 time_before(jiffies
, ioc
->last_waited
+ HZ
/50UL) &&
280 ioc
->last_waited
== last_waited
)) {
282 * we want to go through our batch of
283 * requests and stop. So, we copy out
284 * the ioc->last_waited time and test
285 * against it before looping
287 last_waited
= ioc
->last_waited
;
292 spin_lock(&device
->io_lock
);
293 requeue_list(pending_bios
, pending
, tail
);
294 device
->running_pending
= 1;
296 spin_unlock(&device
->io_lock
);
297 btrfs_requeue_work(&device
->work
);
300 /* unplug every 64 requests just for good measure */
301 if (batch_run
% 64 == 0) {
302 blk_finish_plug(&plug
);
303 blk_start_plug(&plug
);
312 spin_lock(&device
->io_lock
);
313 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
315 spin_unlock(&device
->io_lock
);
318 blk_finish_plug(&plug
);
322 static void pending_bios_fn(struct btrfs_work
*work
)
324 struct btrfs_device
*device
;
326 device
= container_of(work
, struct btrfs_device
, work
);
327 run_scheduled_bios(device
);
330 static noinline
int device_list_add(const char *path
,
331 struct btrfs_super_block
*disk_super
,
332 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
334 struct btrfs_device
*device
;
335 struct btrfs_fs_devices
*fs_devices
;
336 u64 found_transid
= btrfs_super_generation(disk_super
);
339 fs_devices
= find_fsid(disk_super
->fsid
);
341 fs_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
344 INIT_LIST_HEAD(&fs_devices
->devices
);
345 INIT_LIST_HEAD(&fs_devices
->alloc_list
);
346 list_add(&fs_devices
->list
, &fs_uuids
);
347 memcpy(fs_devices
->fsid
, disk_super
->fsid
, BTRFS_FSID_SIZE
);
348 fs_devices
->latest_devid
= devid
;
349 fs_devices
->latest_trans
= found_transid
;
350 mutex_init(&fs_devices
->device_list_mutex
);
353 device
= __find_device(&fs_devices
->devices
, devid
,
354 disk_super
->dev_item
.uuid
);
357 if (fs_devices
->opened
)
360 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
362 /* we can safely leave the fs_devices entry around */
365 device
->devid
= devid
;
366 device
->work
.func
= pending_bios_fn
;
367 memcpy(device
->uuid
, disk_super
->dev_item
.uuid
,
369 spin_lock_init(&device
->io_lock
);
370 device
->name
= kstrdup(path
, GFP_NOFS
);
375 INIT_LIST_HEAD(&device
->dev_alloc_list
);
377 /* init readahead state */
378 spin_lock_init(&device
->reada_lock
);
379 device
->reada_curr_zone
= NULL
;
380 atomic_set(&device
->reada_in_flight
, 0);
381 device
->reada_next
= 0;
382 INIT_RADIX_TREE(&device
->reada_zones
, GFP_NOFS
& ~__GFP_WAIT
);
383 INIT_RADIX_TREE(&device
->reada_extents
, GFP_NOFS
& ~__GFP_WAIT
);
385 mutex_lock(&fs_devices
->device_list_mutex
);
386 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
387 mutex_unlock(&fs_devices
->device_list_mutex
);
389 device
->fs_devices
= fs_devices
;
390 fs_devices
->num_devices
++;
391 } else if (!device
->name
|| strcmp(device
->name
, path
)) {
392 name
= kstrdup(path
, GFP_NOFS
);
397 if (device
->missing
) {
398 fs_devices
->missing_devices
--;
403 if (found_transid
> fs_devices
->latest_trans
) {
404 fs_devices
->latest_devid
= devid
;
405 fs_devices
->latest_trans
= found_transid
;
407 *fs_devices_ret
= fs_devices
;
411 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
413 struct btrfs_fs_devices
*fs_devices
;
414 struct btrfs_device
*device
;
415 struct btrfs_device
*orig_dev
;
417 fs_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
419 return ERR_PTR(-ENOMEM
);
421 INIT_LIST_HEAD(&fs_devices
->devices
);
422 INIT_LIST_HEAD(&fs_devices
->alloc_list
);
423 INIT_LIST_HEAD(&fs_devices
->list
);
424 mutex_init(&fs_devices
->device_list_mutex
);
425 fs_devices
->latest_devid
= orig
->latest_devid
;
426 fs_devices
->latest_trans
= orig
->latest_trans
;
427 memcpy(fs_devices
->fsid
, orig
->fsid
, sizeof(fs_devices
->fsid
));
429 /* We have held the volume lock, it is safe to get the devices. */
430 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
431 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
435 device
->name
= kstrdup(orig_dev
->name
, GFP_NOFS
);
441 device
->devid
= orig_dev
->devid
;
442 device
->work
.func
= pending_bios_fn
;
443 memcpy(device
->uuid
, orig_dev
->uuid
, sizeof(device
->uuid
));
444 spin_lock_init(&device
->io_lock
);
445 INIT_LIST_HEAD(&device
->dev_list
);
446 INIT_LIST_HEAD(&device
->dev_alloc_list
);
448 list_add(&device
->dev_list
, &fs_devices
->devices
);
449 device
->fs_devices
= fs_devices
;
450 fs_devices
->num_devices
++;
454 free_fs_devices(fs_devices
);
455 return ERR_PTR(-ENOMEM
);
458 int btrfs_close_extra_devices(struct btrfs_fs_devices
*fs_devices
)
460 struct btrfs_device
*device
, *next
;
462 struct block_device
*latest_bdev
= NULL
;
463 u64 latest_devid
= 0;
464 u64 latest_transid
= 0;
466 mutex_lock(&uuid_mutex
);
468 /* This is the initialized path, it is safe to release the devices. */
469 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
470 if (device
->in_fs_metadata
) {
471 if (!latest_transid
||
472 device
->generation
> latest_transid
) {
473 latest_devid
= device
->devid
;
474 latest_transid
= device
->generation
;
475 latest_bdev
= device
->bdev
;
481 blkdev_put(device
->bdev
, device
->mode
);
483 fs_devices
->open_devices
--;
485 if (device
->writeable
) {
486 list_del_init(&device
->dev_alloc_list
);
487 device
->writeable
= 0;
488 fs_devices
->rw_devices
--;
490 list_del_init(&device
->dev_list
);
491 fs_devices
->num_devices
--;
496 if (fs_devices
->seed
) {
497 fs_devices
= fs_devices
->seed
;
501 fs_devices
->latest_bdev
= latest_bdev
;
502 fs_devices
->latest_devid
= latest_devid
;
503 fs_devices
->latest_trans
= latest_transid
;
505 mutex_unlock(&uuid_mutex
);
509 static void __free_device(struct work_struct
*work
)
511 struct btrfs_device
*device
;
513 device
= container_of(work
, struct btrfs_device
, rcu_work
);
516 blkdev_put(device
->bdev
, device
->mode
);
522 static void free_device(struct rcu_head
*head
)
524 struct btrfs_device
*device
;
526 device
= container_of(head
, struct btrfs_device
, rcu
);
528 INIT_WORK(&device
->rcu_work
, __free_device
);
529 schedule_work(&device
->rcu_work
);
532 static int __btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
534 struct btrfs_device
*device
;
536 if (--fs_devices
->opened
> 0)
539 mutex_lock(&fs_devices
->device_list_mutex
);
540 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
541 struct btrfs_device
*new_device
;
544 fs_devices
->open_devices
--;
546 if (device
->writeable
) {
547 list_del_init(&device
->dev_alloc_list
);
548 fs_devices
->rw_devices
--;
551 if (device
->can_discard
)
552 fs_devices
->num_can_discard
--;
554 new_device
= kmalloc(sizeof(*new_device
), GFP_NOFS
);
556 memcpy(new_device
, device
, sizeof(*new_device
));
557 new_device
->name
= kstrdup(device
->name
, GFP_NOFS
);
558 BUG_ON(device
->name
&& !new_device
->name
);
559 new_device
->bdev
= NULL
;
560 new_device
->writeable
= 0;
561 new_device
->in_fs_metadata
= 0;
562 new_device
->can_discard
= 0;
563 list_replace_rcu(&device
->dev_list
, &new_device
->dev_list
);
565 call_rcu(&device
->rcu
, free_device
);
567 mutex_unlock(&fs_devices
->device_list_mutex
);
569 WARN_ON(fs_devices
->open_devices
);
570 WARN_ON(fs_devices
->rw_devices
);
571 fs_devices
->opened
= 0;
572 fs_devices
->seeding
= 0;
577 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
579 struct btrfs_fs_devices
*seed_devices
= NULL
;
582 mutex_lock(&uuid_mutex
);
583 ret
= __btrfs_close_devices(fs_devices
);
584 if (!fs_devices
->opened
) {
585 seed_devices
= fs_devices
->seed
;
586 fs_devices
->seed
= NULL
;
588 mutex_unlock(&uuid_mutex
);
590 while (seed_devices
) {
591 fs_devices
= seed_devices
;
592 seed_devices
= fs_devices
->seed
;
593 __btrfs_close_devices(fs_devices
);
594 free_fs_devices(fs_devices
);
599 static int __btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
600 fmode_t flags
, void *holder
)
602 struct request_queue
*q
;
603 struct block_device
*bdev
;
604 struct list_head
*head
= &fs_devices
->devices
;
605 struct btrfs_device
*device
;
606 struct block_device
*latest_bdev
= NULL
;
607 struct buffer_head
*bh
;
608 struct btrfs_super_block
*disk_super
;
609 u64 latest_devid
= 0;
610 u64 latest_transid
= 0;
617 list_for_each_entry(device
, head
, dev_list
) {
623 bdev
= blkdev_get_by_path(device
->name
, flags
, holder
);
625 printk(KERN_INFO
"open %s failed\n", device
->name
);
628 set_blocksize(bdev
, 4096);
630 bh
= btrfs_read_dev_super(bdev
);
634 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
635 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
636 if (devid
!= device
->devid
)
639 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
,
643 device
->generation
= btrfs_super_generation(disk_super
);
644 if (!latest_transid
|| device
->generation
> latest_transid
) {
645 latest_devid
= devid
;
646 latest_transid
= device
->generation
;
650 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
651 device
->writeable
= 0;
653 device
->writeable
= !bdev_read_only(bdev
);
657 q
= bdev_get_queue(bdev
);
658 if (blk_queue_discard(q
)) {
659 device
->can_discard
= 1;
660 fs_devices
->num_can_discard
++;
664 device
->in_fs_metadata
= 0;
665 device
->mode
= flags
;
667 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
668 fs_devices
->rotating
= 1;
670 fs_devices
->open_devices
++;
671 if (device
->writeable
) {
672 fs_devices
->rw_devices
++;
673 list_add(&device
->dev_alloc_list
,
674 &fs_devices
->alloc_list
);
682 blkdev_put(bdev
, flags
);
686 if (fs_devices
->open_devices
== 0) {
690 fs_devices
->seeding
= seeding
;
691 fs_devices
->opened
= 1;
692 fs_devices
->latest_bdev
= latest_bdev
;
693 fs_devices
->latest_devid
= latest_devid
;
694 fs_devices
->latest_trans
= latest_transid
;
695 fs_devices
->total_rw_bytes
= 0;
700 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
701 fmode_t flags
, void *holder
)
705 mutex_lock(&uuid_mutex
);
706 if (fs_devices
->opened
) {
707 fs_devices
->opened
++;
710 ret
= __btrfs_open_devices(fs_devices
, flags
, holder
);
712 mutex_unlock(&uuid_mutex
);
716 int btrfs_scan_one_device(const char *path
, fmode_t flags
, void *holder
,
717 struct btrfs_fs_devices
**fs_devices_ret
)
719 struct btrfs_super_block
*disk_super
;
720 struct block_device
*bdev
;
721 struct buffer_head
*bh
;
727 bdev
= blkdev_get_by_path(path
, flags
, holder
);
734 mutex_lock(&uuid_mutex
);
735 ret
= set_blocksize(bdev
, 4096);
738 bh
= btrfs_read_dev_super(bdev
);
743 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
744 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
745 transid
= btrfs_super_generation(disk_super
);
746 if (disk_super
->label
[0])
747 printk(KERN_INFO
"device label %s ", disk_super
->label
);
749 printk(KERN_INFO
"device fsid %pU ", disk_super
->fsid
);
750 printk(KERN_CONT
"devid %llu transid %llu %s\n",
751 (unsigned long long)devid
, (unsigned long long)transid
, path
);
752 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
756 mutex_unlock(&uuid_mutex
);
757 blkdev_put(bdev
, flags
);
762 /* helper to account the used device space in the range */
763 int btrfs_account_dev_extents_size(struct btrfs_device
*device
, u64 start
,
764 u64 end
, u64
*length
)
766 struct btrfs_key key
;
767 struct btrfs_root
*root
= device
->dev_root
;
768 struct btrfs_dev_extent
*dev_extent
;
769 struct btrfs_path
*path
;
773 struct extent_buffer
*l
;
777 if (start
>= device
->total_bytes
)
780 path
= btrfs_alloc_path();
785 key
.objectid
= device
->devid
;
787 key
.type
= BTRFS_DEV_EXTENT_KEY
;
789 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
793 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
800 slot
= path
->slots
[0];
801 if (slot
>= btrfs_header_nritems(l
)) {
802 ret
= btrfs_next_leaf(root
, path
);
810 btrfs_item_key_to_cpu(l
, &key
, slot
);
812 if (key
.objectid
< device
->devid
)
815 if (key
.objectid
> device
->devid
)
818 if (btrfs_key_type(&key
) != BTRFS_DEV_EXTENT_KEY
)
821 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
822 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
824 if (key
.offset
<= start
&& extent_end
> end
) {
825 *length
= end
- start
+ 1;
827 } else if (key
.offset
<= start
&& extent_end
> start
)
828 *length
+= extent_end
- start
;
829 else if (key
.offset
> start
&& extent_end
<= end
)
830 *length
+= extent_end
- key
.offset
;
831 else if (key
.offset
> start
&& key
.offset
<= end
) {
832 *length
+= end
- key
.offset
+ 1;
834 } else if (key
.offset
> end
)
842 btrfs_free_path(path
);
847 * find_free_dev_extent - find free space in the specified device
848 * @device: the device which we search the free space in
849 * @num_bytes: the size of the free space that we need
850 * @start: store the start of the free space.
851 * @len: the size of the free space. that we find, or the size of the max
852 * free space if we don't find suitable free space
854 * this uses a pretty simple search, the expectation is that it is
855 * called very infrequently and that a given device has a small number
858 * @start is used to store the start of the free space if we find. But if we
859 * don't find suitable free space, it will be used to store the start position
860 * of the max free space.
862 * @len is used to store the size of the free space that we find.
863 * But if we don't find suitable free space, it is used to store the size of
864 * the max free space.
866 int find_free_dev_extent(struct btrfs_device
*device
, u64 num_bytes
,
867 u64
*start
, u64
*len
)
869 struct btrfs_key key
;
870 struct btrfs_root
*root
= device
->dev_root
;
871 struct btrfs_dev_extent
*dev_extent
;
872 struct btrfs_path
*path
;
878 u64 search_end
= device
->total_bytes
;
881 struct extent_buffer
*l
;
883 /* FIXME use last free of some kind */
885 /* we don't want to overwrite the superblock on the drive,
886 * so we make sure to start at an offset of at least 1MB
888 search_start
= max(root
->fs_info
->alloc_start
, 1024ull * 1024);
890 max_hole_start
= search_start
;
894 if (search_start
>= search_end
) {
899 path
= btrfs_alloc_path();
906 key
.objectid
= device
->devid
;
907 key
.offset
= search_start
;
908 key
.type
= BTRFS_DEV_EXTENT_KEY
;
910 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
914 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
921 slot
= path
->slots
[0];
922 if (slot
>= btrfs_header_nritems(l
)) {
923 ret
= btrfs_next_leaf(root
, path
);
931 btrfs_item_key_to_cpu(l
, &key
, slot
);
933 if (key
.objectid
< device
->devid
)
936 if (key
.objectid
> device
->devid
)
939 if (btrfs_key_type(&key
) != BTRFS_DEV_EXTENT_KEY
)
942 if (key
.offset
> search_start
) {
943 hole_size
= key
.offset
- search_start
;
945 if (hole_size
> max_hole_size
) {
946 max_hole_start
= search_start
;
947 max_hole_size
= hole_size
;
951 * If this free space is greater than which we need,
952 * it must be the max free space that we have found
953 * until now, so max_hole_start must point to the start
954 * of this free space and the length of this free space
955 * is stored in max_hole_size. Thus, we return
956 * max_hole_start and max_hole_size and go back to the
959 if (hole_size
>= num_bytes
) {
965 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
966 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
968 if (extent_end
> search_start
)
969 search_start
= extent_end
;
976 * At this point, search_start should be the end of
977 * allocated dev extents, and when shrinking the device,
978 * search_end may be smaller than search_start.
980 if (search_end
> search_start
)
981 hole_size
= search_end
- search_start
;
983 if (hole_size
> max_hole_size
) {
984 max_hole_start
= search_start
;
985 max_hole_size
= hole_size
;
989 if (hole_size
< num_bytes
)
995 btrfs_free_path(path
);
997 *start
= max_hole_start
;
999 *len
= max_hole_size
;
1003 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1004 struct btrfs_device
*device
,
1008 struct btrfs_path
*path
;
1009 struct btrfs_root
*root
= device
->dev_root
;
1010 struct btrfs_key key
;
1011 struct btrfs_key found_key
;
1012 struct extent_buffer
*leaf
= NULL
;
1013 struct btrfs_dev_extent
*extent
= NULL
;
1015 path
= btrfs_alloc_path();
1019 key
.objectid
= device
->devid
;
1021 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1023 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1025 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1026 BTRFS_DEV_EXTENT_KEY
);
1029 leaf
= path
->nodes
[0];
1030 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1031 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1032 struct btrfs_dev_extent
);
1033 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1034 btrfs_dev_extent_length(leaf
, extent
) < start
);
1036 btrfs_release_path(path
);
1038 } else if (ret
== 0) {
1039 leaf
= path
->nodes
[0];
1040 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1041 struct btrfs_dev_extent
);
1045 if (device
->bytes_used
> 0) {
1046 u64 len
= btrfs_dev_extent_length(leaf
, extent
);
1047 device
->bytes_used
-= len
;
1048 spin_lock(&root
->fs_info
->free_chunk_lock
);
1049 root
->fs_info
->free_chunk_space
+= len
;
1050 spin_unlock(&root
->fs_info
->free_chunk_lock
);
1052 ret
= btrfs_del_item(trans
, root
, path
);
1055 btrfs_free_path(path
);
1059 int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1060 struct btrfs_device
*device
,
1061 u64 chunk_tree
, u64 chunk_objectid
,
1062 u64 chunk_offset
, u64 start
, u64 num_bytes
)
1065 struct btrfs_path
*path
;
1066 struct btrfs_root
*root
= device
->dev_root
;
1067 struct btrfs_dev_extent
*extent
;
1068 struct extent_buffer
*leaf
;
1069 struct btrfs_key key
;
1071 WARN_ON(!device
->in_fs_metadata
);
1072 path
= btrfs_alloc_path();
1076 key
.objectid
= device
->devid
;
1078 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1079 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1083 leaf
= path
->nodes
[0];
1084 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1085 struct btrfs_dev_extent
);
1086 btrfs_set_dev_extent_chunk_tree(leaf
, extent
, chunk_tree
);
1087 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
, chunk_objectid
);
1088 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1090 write_extent_buffer(leaf
, root
->fs_info
->chunk_tree_uuid
,
1091 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent
),
1094 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1095 btrfs_mark_buffer_dirty(leaf
);
1096 btrfs_free_path(path
);
1100 static noinline
int find_next_chunk(struct btrfs_root
*root
,
1101 u64 objectid
, u64
*offset
)
1103 struct btrfs_path
*path
;
1105 struct btrfs_key key
;
1106 struct btrfs_chunk
*chunk
;
1107 struct btrfs_key found_key
;
1109 path
= btrfs_alloc_path();
1113 key
.objectid
= objectid
;
1114 key
.offset
= (u64
)-1;
1115 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
1117 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1123 ret
= btrfs_previous_item(root
, path
, 0, BTRFS_CHUNK_ITEM_KEY
);
1127 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1129 if (found_key
.objectid
!= objectid
)
1132 chunk
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1133 struct btrfs_chunk
);
1134 *offset
= found_key
.offset
+
1135 btrfs_chunk_length(path
->nodes
[0], chunk
);
1140 btrfs_free_path(path
);
1144 static noinline
int find_next_devid(struct btrfs_root
*root
, u64
*objectid
)
1147 struct btrfs_key key
;
1148 struct btrfs_key found_key
;
1149 struct btrfs_path
*path
;
1151 root
= root
->fs_info
->chunk_root
;
1153 path
= btrfs_alloc_path();
1157 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1158 key
.type
= BTRFS_DEV_ITEM_KEY
;
1159 key
.offset
= (u64
)-1;
1161 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1167 ret
= btrfs_previous_item(root
, path
, BTRFS_DEV_ITEMS_OBJECTID
,
1168 BTRFS_DEV_ITEM_KEY
);
1172 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1174 *objectid
= found_key
.offset
+ 1;
1178 btrfs_free_path(path
);
1183 * the device information is stored in the chunk root
1184 * the btrfs_device struct should be fully filled in
1186 int btrfs_add_device(struct btrfs_trans_handle
*trans
,
1187 struct btrfs_root
*root
,
1188 struct btrfs_device
*device
)
1191 struct btrfs_path
*path
;
1192 struct btrfs_dev_item
*dev_item
;
1193 struct extent_buffer
*leaf
;
1194 struct btrfs_key key
;
1197 root
= root
->fs_info
->chunk_root
;
1199 path
= btrfs_alloc_path();
1203 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1204 key
.type
= BTRFS_DEV_ITEM_KEY
;
1205 key
.offset
= device
->devid
;
1207 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1212 leaf
= path
->nodes
[0];
1213 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1215 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1216 btrfs_set_device_generation(leaf
, dev_item
, 0);
1217 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1218 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1219 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1220 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1221 btrfs_set_device_total_bytes(leaf
, dev_item
, device
->total_bytes
);
1222 btrfs_set_device_bytes_used(leaf
, dev_item
, device
->bytes_used
);
1223 btrfs_set_device_group(leaf
, dev_item
, 0);
1224 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1225 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1226 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1228 ptr
= (unsigned long)btrfs_device_uuid(dev_item
);
1229 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1230 ptr
= (unsigned long)btrfs_device_fsid(dev_item
);
1231 write_extent_buffer(leaf
, root
->fs_info
->fsid
, ptr
, BTRFS_UUID_SIZE
);
1232 btrfs_mark_buffer_dirty(leaf
);
1236 btrfs_free_path(path
);
1240 static int btrfs_rm_dev_item(struct btrfs_root
*root
,
1241 struct btrfs_device
*device
)
1244 struct btrfs_path
*path
;
1245 struct btrfs_key key
;
1246 struct btrfs_trans_handle
*trans
;
1248 root
= root
->fs_info
->chunk_root
;
1250 path
= btrfs_alloc_path();
1254 trans
= btrfs_start_transaction(root
, 0);
1255 if (IS_ERR(trans
)) {
1256 btrfs_free_path(path
);
1257 return PTR_ERR(trans
);
1259 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1260 key
.type
= BTRFS_DEV_ITEM_KEY
;
1261 key
.offset
= device
->devid
;
1264 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1273 ret
= btrfs_del_item(trans
, root
, path
);
1277 btrfs_free_path(path
);
1278 unlock_chunks(root
);
1279 btrfs_commit_transaction(trans
, root
);
1283 int btrfs_rm_device(struct btrfs_root
*root
, char *device_path
)
1285 struct btrfs_device
*device
;
1286 struct btrfs_device
*next_device
;
1287 struct block_device
*bdev
;
1288 struct buffer_head
*bh
= NULL
;
1289 struct btrfs_super_block
*disk_super
;
1290 struct btrfs_fs_devices
*cur_devices
;
1296 bool clear_super
= false;
1298 mutex_lock(&uuid_mutex
);
1300 all_avail
= root
->fs_info
->avail_data_alloc_bits
|
1301 root
->fs_info
->avail_system_alloc_bits
|
1302 root
->fs_info
->avail_metadata_alloc_bits
;
1304 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID10
) &&
1305 root
->fs_info
->fs_devices
->num_devices
<= 4) {
1306 printk(KERN_ERR
"btrfs: unable to go below four devices "
1312 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID1
) &&
1313 root
->fs_info
->fs_devices
->num_devices
<= 2) {
1314 printk(KERN_ERR
"btrfs: unable to go below two "
1315 "devices on raid1\n");
1320 if (strcmp(device_path
, "missing") == 0) {
1321 struct list_head
*devices
;
1322 struct btrfs_device
*tmp
;
1325 devices
= &root
->fs_info
->fs_devices
->devices
;
1327 * It is safe to read the devices since the volume_mutex
1330 list_for_each_entry(tmp
, devices
, dev_list
) {
1331 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
1340 printk(KERN_ERR
"btrfs: no missing devices found to "
1345 bdev
= blkdev_get_by_path(device_path
, FMODE_READ
| FMODE_EXCL
,
1346 root
->fs_info
->bdev_holder
);
1348 ret
= PTR_ERR(bdev
);
1352 set_blocksize(bdev
, 4096);
1353 bh
= btrfs_read_dev_super(bdev
);
1358 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1359 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1360 dev_uuid
= disk_super
->dev_item
.uuid
;
1361 device
= btrfs_find_device(root
, devid
, dev_uuid
,
1369 if (device
->writeable
&& root
->fs_info
->fs_devices
->rw_devices
== 1) {
1370 printk(KERN_ERR
"btrfs: unable to remove the only writeable "
1376 if (device
->writeable
) {
1378 list_del_init(&device
->dev_alloc_list
);
1379 unlock_chunks(root
);
1380 root
->fs_info
->fs_devices
->rw_devices
--;
1384 ret
= btrfs_shrink_device(device
, 0);
1388 ret
= btrfs_rm_dev_item(root
->fs_info
->chunk_root
, device
);
1392 spin_lock(&root
->fs_info
->free_chunk_lock
);
1393 root
->fs_info
->free_chunk_space
= device
->total_bytes
-
1395 spin_unlock(&root
->fs_info
->free_chunk_lock
);
1397 device
->in_fs_metadata
= 0;
1398 btrfs_scrub_cancel_dev(root
, device
);
1401 * the device list mutex makes sure that we don't change
1402 * the device list while someone else is writing out all
1403 * the device supers.
1406 cur_devices
= device
->fs_devices
;
1407 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1408 list_del_rcu(&device
->dev_list
);
1410 device
->fs_devices
->num_devices
--;
1412 if (device
->missing
)
1413 root
->fs_info
->fs_devices
->missing_devices
--;
1415 next_device
= list_entry(root
->fs_info
->fs_devices
->devices
.next
,
1416 struct btrfs_device
, dev_list
);
1417 if (device
->bdev
== root
->fs_info
->sb
->s_bdev
)
1418 root
->fs_info
->sb
->s_bdev
= next_device
->bdev
;
1419 if (device
->bdev
== root
->fs_info
->fs_devices
->latest_bdev
)
1420 root
->fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1423 device
->fs_devices
->open_devices
--;
1425 call_rcu(&device
->rcu
, free_device
);
1426 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1428 num_devices
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
1429 btrfs_set_super_num_devices(root
->fs_info
->super_copy
, num_devices
);
1431 if (cur_devices
->open_devices
== 0) {
1432 struct btrfs_fs_devices
*fs_devices
;
1433 fs_devices
= root
->fs_info
->fs_devices
;
1434 while (fs_devices
) {
1435 if (fs_devices
->seed
== cur_devices
)
1437 fs_devices
= fs_devices
->seed
;
1439 fs_devices
->seed
= cur_devices
->seed
;
1440 cur_devices
->seed
= NULL
;
1442 __btrfs_close_devices(cur_devices
);
1443 unlock_chunks(root
);
1444 free_fs_devices(cur_devices
);
1448 * at this point, the device is zero sized. We want to
1449 * remove it from the devices list and zero out the old super
1452 /* make sure this device isn't detected as part of
1455 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
1456 set_buffer_dirty(bh
);
1457 sync_dirty_buffer(bh
);
1466 blkdev_put(bdev
, FMODE_READ
| FMODE_EXCL
);
1468 mutex_unlock(&uuid_mutex
);
1471 if (device
->writeable
) {
1473 list_add(&device
->dev_alloc_list
,
1474 &root
->fs_info
->fs_devices
->alloc_list
);
1475 unlock_chunks(root
);
1476 root
->fs_info
->fs_devices
->rw_devices
++;
1482 * does all the dirty work required for changing file system's UUID.
1484 static int btrfs_prepare_sprout(struct btrfs_root
*root
)
1486 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
1487 struct btrfs_fs_devices
*old_devices
;
1488 struct btrfs_fs_devices
*seed_devices
;
1489 struct btrfs_super_block
*disk_super
= root
->fs_info
->super_copy
;
1490 struct btrfs_device
*device
;
1493 BUG_ON(!mutex_is_locked(&uuid_mutex
));
1494 if (!fs_devices
->seeding
)
1497 seed_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
1501 old_devices
= clone_fs_devices(fs_devices
);
1502 if (IS_ERR(old_devices
)) {
1503 kfree(seed_devices
);
1504 return PTR_ERR(old_devices
);
1507 list_add(&old_devices
->list
, &fs_uuids
);
1509 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
1510 seed_devices
->opened
= 1;
1511 INIT_LIST_HEAD(&seed_devices
->devices
);
1512 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
1513 mutex_init(&seed_devices
->device_list_mutex
);
1515 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1516 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
1518 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1520 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
1521 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
) {
1522 device
->fs_devices
= seed_devices
;
1525 fs_devices
->seeding
= 0;
1526 fs_devices
->num_devices
= 0;
1527 fs_devices
->open_devices
= 0;
1528 fs_devices
->seed
= seed_devices
;
1530 generate_random_uuid(fs_devices
->fsid
);
1531 memcpy(root
->fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
1532 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
1533 super_flags
= btrfs_super_flags(disk_super
) &
1534 ~BTRFS_SUPER_FLAG_SEEDING
;
1535 btrfs_set_super_flags(disk_super
, super_flags
);
1541 * strore the expected generation for seed devices in device items.
1543 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
1544 struct btrfs_root
*root
)
1546 struct btrfs_path
*path
;
1547 struct extent_buffer
*leaf
;
1548 struct btrfs_dev_item
*dev_item
;
1549 struct btrfs_device
*device
;
1550 struct btrfs_key key
;
1551 u8 fs_uuid
[BTRFS_UUID_SIZE
];
1552 u8 dev_uuid
[BTRFS_UUID_SIZE
];
1556 path
= btrfs_alloc_path();
1560 root
= root
->fs_info
->chunk_root
;
1561 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1563 key
.type
= BTRFS_DEV_ITEM_KEY
;
1566 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
1570 leaf
= path
->nodes
[0];
1572 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1573 ret
= btrfs_next_leaf(root
, path
);
1578 leaf
= path
->nodes
[0];
1579 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1580 btrfs_release_path(path
);
1584 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1585 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
1586 key
.type
!= BTRFS_DEV_ITEM_KEY
)
1589 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1590 struct btrfs_dev_item
);
1591 devid
= btrfs_device_id(leaf
, dev_item
);
1592 read_extent_buffer(leaf
, dev_uuid
,
1593 (unsigned long)btrfs_device_uuid(dev_item
),
1595 read_extent_buffer(leaf
, fs_uuid
,
1596 (unsigned long)btrfs_device_fsid(dev_item
),
1598 device
= btrfs_find_device(root
, devid
, dev_uuid
, fs_uuid
);
1601 if (device
->fs_devices
->seeding
) {
1602 btrfs_set_device_generation(leaf
, dev_item
,
1603 device
->generation
);
1604 btrfs_mark_buffer_dirty(leaf
);
1612 btrfs_free_path(path
);
1616 int btrfs_init_new_device(struct btrfs_root
*root
, char *device_path
)
1618 struct request_queue
*q
;
1619 struct btrfs_trans_handle
*trans
;
1620 struct btrfs_device
*device
;
1621 struct block_device
*bdev
;
1622 struct list_head
*devices
;
1623 struct super_block
*sb
= root
->fs_info
->sb
;
1625 int seeding_dev
= 0;
1628 if ((sb
->s_flags
& MS_RDONLY
) && !root
->fs_info
->fs_devices
->seeding
)
1631 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
1632 root
->fs_info
->bdev_holder
);
1634 return PTR_ERR(bdev
);
1636 if (root
->fs_info
->fs_devices
->seeding
) {
1638 down_write(&sb
->s_umount
);
1639 mutex_lock(&uuid_mutex
);
1642 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
1644 devices
= &root
->fs_info
->fs_devices
->devices
;
1646 * we have the volume lock, so we don't need the extra
1647 * device list mutex while reading the list here.
1649 list_for_each_entry(device
, devices
, dev_list
) {
1650 if (device
->bdev
== bdev
) {
1656 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
1658 /* we can safely leave the fs_devices entry around */
1663 device
->name
= kstrdup(device_path
, GFP_NOFS
);
1664 if (!device
->name
) {
1670 ret
= find_next_devid(root
, &device
->devid
);
1672 kfree(device
->name
);
1677 trans
= btrfs_start_transaction(root
, 0);
1678 if (IS_ERR(trans
)) {
1679 kfree(device
->name
);
1681 ret
= PTR_ERR(trans
);
1687 q
= bdev_get_queue(bdev
);
1688 if (blk_queue_discard(q
))
1689 device
->can_discard
= 1;
1690 device
->writeable
= 1;
1691 device
->work
.func
= pending_bios_fn
;
1692 generate_random_uuid(device
->uuid
);
1693 spin_lock_init(&device
->io_lock
);
1694 device
->generation
= trans
->transid
;
1695 device
->io_width
= root
->sectorsize
;
1696 device
->io_align
= root
->sectorsize
;
1697 device
->sector_size
= root
->sectorsize
;
1698 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
1699 device
->disk_total_bytes
= device
->total_bytes
;
1700 device
->dev_root
= root
->fs_info
->dev_root
;
1701 device
->bdev
= bdev
;
1702 device
->in_fs_metadata
= 1;
1703 device
->mode
= FMODE_EXCL
;
1704 set_blocksize(device
->bdev
, 4096);
1707 sb
->s_flags
&= ~MS_RDONLY
;
1708 ret
= btrfs_prepare_sprout(root
);
1712 device
->fs_devices
= root
->fs_info
->fs_devices
;
1715 * we don't want write_supers to jump in here with our device
1718 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1719 list_add_rcu(&device
->dev_list
, &root
->fs_info
->fs_devices
->devices
);
1720 list_add(&device
->dev_alloc_list
,
1721 &root
->fs_info
->fs_devices
->alloc_list
);
1722 root
->fs_info
->fs_devices
->num_devices
++;
1723 root
->fs_info
->fs_devices
->open_devices
++;
1724 root
->fs_info
->fs_devices
->rw_devices
++;
1725 if (device
->can_discard
)
1726 root
->fs_info
->fs_devices
->num_can_discard
++;
1727 root
->fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
1729 spin_lock(&root
->fs_info
->free_chunk_lock
);
1730 root
->fs_info
->free_chunk_space
+= device
->total_bytes
;
1731 spin_unlock(&root
->fs_info
->free_chunk_lock
);
1733 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
1734 root
->fs_info
->fs_devices
->rotating
= 1;
1736 total_bytes
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
1737 btrfs_set_super_total_bytes(root
->fs_info
->super_copy
,
1738 total_bytes
+ device
->total_bytes
);
1740 total_bytes
= btrfs_super_num_devices(root
->fs_info
->super_copy
);
1741 btrfs_set_super_num_devices(root
->fs_info
->super_copy
,
1743 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1746 ret
= init_first_rw_device(trans
, root
, device
);
1748 ret
= btrfs_finish_sprout(trans
, root
);
1751 ret
= btrfs_add_device(trans
, root
, device
);
1755 * we've got more storage, clear any full flags on the space
1758 btrfs_clear_space_info_full(root
->fs_info
);
1760 unlock_chunks(root
);
1761 btrfs_commit_transaction(trans
, root
);
1764 mutex_unlock(&uuid_mutex
);
1765 up_write(&sb
->s_umount
);
1767 ret
= btrfs_relocate_sys_chunks(root
);
1773 blkdev_put(bdev
, FMODE_EXCL
);
1775 mutex_unlock(&uuid_mutex
);
1776 up_write(&sb
->s_umount
);
1781 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
1782 struct btrfs_device
*device
)
1785 struct btrfs_path
*path
;
1786 struct btrfs_root
*root
;
1787 struct btrfs_dev_item
*dev_item
;
1788 struct extent_buffer
*leaf
;
1789 struct btrfs_key key
;
1791 root
= device
->dev_root
->fs_info
->chunk_root
;
1793 path
= btrfs_alloc_path();
1797 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1798 key
.type
= BTRFS_DEV_ITEM_KEY
;
1799 key
.offset
= device
->devid
;
1801 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
1810 leaf
= path
->nodes
[0];
1811 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1813 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1814 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1815 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1816 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1817 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1818 btrfs_set_device_total_bytes(leaf
, dev_item
, device
->disk_total_bytes
);
1819 btrfs_set_device_bytes_used(leaf
, dev_item
, device
->bytes_used
);
1820 btrfs_mark_buffer_dirty(leaf
);
1823 btrfs_free_path(path
);
1827 static int __btrfs_grow_device(struct btrfs_trans_handle
*trans
,
1828 struct btrfs_device
*device
, u64 new_size
)
1830 struct btrfs_super_block
*super_copy
=
1831 device
->dev_root
->fs_info
->super_copy
;
1832 u64 old_total
= btrfs_super_total_bytes(super_copy
);
1833 u64 diff
= new_size
- device
->total_bytes
;
1835 if (!device
->writeable
)
1837 if (new_size
<= device
->total_bytes
)
1840 btrfs_set_super_total_bytes(super_copy
, old_total
+ diff
);
1841 device
->fs_devices
->total_rw_bytes
+= diff
;
1843 device
->total_bytes
= new_size
;
1844 device
->disk_total_bytes
= new_size
;
1845 btrfs_clear_space_info_full(device
->dev_root
->fs_info
);
1847 return btrfs_update_device(trans
, device
);
1850 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
1851 struct btrfs_device
*device
, u64 new_size
)
1854 lock_chunks(device
->dev_root
);
1855 ret
= __btrfs_grow_device(trans
, device
, new_size
);
1856 unlock_chunks(device
->dev_root
);
1860 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
1861 struct btrfs_root
*root
,
1862 u64 chunk_tree
, u64 chunk_objectid
,
1866 struct btrfs_path
*path
;
1867 struct btrfs_key key
;
1869 root
= root
->fs_info
->chunk_root
;
1870 path
= btrfs_alloc_path();
1874 key
.objectid
= chunk_objectid
;
1875 key
.offset
= chunk_offset
;
1876 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
1878 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1881 ret
= btrfs_del_item(trans
, root
, path
);
1883 btrfs_free_path(path
);
1887 static int btrfs_del_sys_chunk(struct btrfs_root
*root
, u64 chunk_objectid
, u64
1890 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
1891 struct btrfs_disk_key
*disk_key
;
1892 struct btrfs_chunk
*chunk
;
1899 struct btrfs_key key
;
1901 array_size
= btrfs_super_sys_array_size(super_copy
);
1903 ptr
= super_copy
->sys_chunk_array
;
1906 while (cur
< array_size
) {
1907 disk_key
= (struct btrfs_disk_key
*)ptr
;
1908 btrfs_disk_key_to_cpu(&key
, disk_key
);
1910 len
= sizeof(*disk_key
);
1912 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
1913 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
1914 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
1915 len
+= btrfs_chunk_item_size(num_stripes
);
1920 if (key
.objectid
== chunk_objectid
&&
1921 key
.offset
== chunk_offset
) {
1922 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
1924 btrfs_set_super_sys_array_size(super_copy
, array_size
);
1933 static int btrfs_relocate_chunk(struct btrfs_root
*root
,
1934 u64 chunk_tree
, u64 chunk_objectid
,
1937 struct extent_map_tree
*em_tree
;
1938 struct btrfs_root
*extent_root
;
1939 struct btrfs_trans_handle
*trans
;
1940 struct extent_map
*em
;
1941 struct map_lookup
*map
;
1945 root
= root
->fs_info
->chunk_root
;
1946 extent_root
= root
->fs_info
->extent_root
;
1947 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
1949 ret
= btrfs_can_relocate(extent_root
, chunk_offset
);
1953 /* step one, relocate all the extents inside this chunk */
1954 ret
= btrfs_relocate_block_group(extent_root
, chunk_offset
);
1958 trans
= btrfs_start_transaction(root
, 0);
1959 BUG_ON(IS_ERR(trans
));
1964 * step two, delete the device extents and the
1965 * chunk tree entries
1967 read_lock(&em_tree
->lock
);
1968 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
1969 read_unlock(&em_tree
->lock
);
1971 BUG_ON(!em
|| em
->start
> chunk_offset
||
1972 em
->start
+ em
->len
< chunk_offset
);
1973 map
= (struct map_lookup
*)em
->bdev
;
1975 for (i
= 0; i
< map
->num_stripes
; i
++) {
1976 ret
= btrfs_free_dev_extent(trans
, map
->stripes
[i
].dev
,
1977 map
->stripes
[i
].physical
);
1980 if (map
->stripes
[i
].dev
) {
1981 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
1985 ret
= btrfs_free_chunk(trans
, root
, chunk_tree
, chunk_objectid
,
1990 trace_btrfs_chunk_free(root
, map
, chunk_offset
, em
->len
);
1992 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
1993 ret
= btrfs_del_sys_chunk(root
, chunk_objectid
, chunk_offset
);
1997 ret
= btrfs_remove_block_group(trans
, extent_root
, chunk_offset
);
2000 write_lock(&em_tree
->lock
);
2001 remove_extent_mapping(em_tree
, em
);
2002 write_unlock(&em_tree
->lock
);
2007 /* once for the tree */
2008 free_extent_map(em
);
2010 free_extent_map(em
);
2012 unlock_chunks(root
);
2013 btrfs_end_transaction(trans
, root
);
2017 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
)
2019 struct btrfs_root
*chunk_root
= root
->fs_info
->chunk_root
;
2020 struct btrfs_path
*path
;
2021 struct extent_buffer
*leaf
;
2022 struct btrfs_chunk
*chunk
;
2023 struct btrfs_key key
;
2024 struct btrfs_key found_key
;
2025 u64 chunk_tree
= chunk_root
->root_key
.objectid
;
2027 bool retried
= false;
2031 path
= btrfs_alloc_path();
2036 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2037 key
.offset
= (u64
)-1;
2038 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2041 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2046 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
2053 leaf
= path
->nodes
[0];
2054 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2056 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
2057 struct btrfs_chunk
);
2058 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2059 btrfs_release_path(path
);
2061 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2062 ret
= btrfs_relocate_chunk(chunk_root
, chunk_tree
,
2071 if (found_key
.offset
== 0)
2073 key
.offset
= found_key
.offset
- 1;
2076 if (failed
&& !retried
) {
2080 } else if (failed
&& retried
) {
2085 btrfs_free_path(path
);
2089 static int insert_balance_item(struct btrfs_root
*root
,
2090 struct btrfs_balance_control
*bctl
)
2092 struct btrfs_trans_handle
*trans
;
2093 struct btrfs_balance_item
*item
;
2094 struct btrfs_disk_balance_args disk_bargs
;
2095 struct btrfs_path
*path
;
2096 struct extent_buffer
*leaf
;
2097 struct btrfs_key key
;
2100 path
= btrfs_alloc_path();
2104 trans
= btrfs_start_transaction(root
, 0);
2105 if (IS_ERR(trans
)) {
2106 btrfs_free_path(path
);
2107 return PTR_ERR(trans
);
2110 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2111 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2114 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2119 leaf
= path
->nodes
[0];
2120 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
2122 memset_extent_buffer(leaf
, 0, (unsigned long)item
, sizeof(*item
));
2124 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
2125 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
2126 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
2127 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
2128 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
2129 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
2131 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
2133 btrfs_mark_buffer_dirty(leaf
);
2135 btrfs_free_path(path
);
2136 err
= btrfs_commit_transaction(trans
, root
);
2142 static int del_balance_item(struct btrfs_root
*root
)
2144 struct btrfs_trans_handle
*trans
;
2145 struct btrfs_path
*path
;
2146 struct btrfs_key key
;
2149 path
= btrfs_alloc_path();
2153 trans
= btrfs_start_transaction(root
, 0);
2154 if (IS_ERR(trans
)) {
2155 btrfs_free_path(path
);
2156 return PTR_ERR(trans
);
2159 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2160 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2163 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2171 ret
= btrfs_del_item(trans
, root
, path
);
2173 btrfs_free_path(path
);
2174 err
= btrfs_commit_transaction(trans
, root
);
2181 * This is a heuristic used to reduce the number of chunks balanced on
2182 * resume after balance was interrupted.
2184 static void update_balance_args(struct btrfs_balance_control
*bctl
)
2187 * Turn on soft mode for chunk types that were being converted.
2189 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2190 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2191 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2192 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2193 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2194 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2197 * Turn on usage filter if is not already used. The idea is
2198 * that chunks that we have already balanced should be
2199 * reasonably full. Don't do it for chunks that are being
2200 * converted - that will keep us from relocating unconverted
2201 * (albeit full) chunks.
2203 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2204 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2205 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2206 bctl
->data
.usage
= 90;
2208 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2209 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2210 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2211 bctl
->sys
.usage
= 90;
2213 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2214 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2215 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2216 bctl
->meta
.usage
= 90;
2221 * Should be called with both balance and volume mutexes held to
2222 * serialize other volume operations (add_dev/rm_dev/resize) with
2223 * restriper. Same goes for unset_balance_control.
2225 static void set_balance_control(struct btrfs_balance_control
*bctl
)
2227 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
2229 BUG_ON(fs_info
->balance_ctl
);
2231 spin_lock(&fs_info
->balance_lock
);
2232 fs_info
->balance_ctl
= bctl
;
2233 spin_unlock(&fs_info
->balance_lock
);
2236 static void unset_balance_control(struct btrfs_fs_info
*fs_info
)
2238 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
2240 BUG_ON(!fs_info
->balance_ctl
);
2242 spin_lock(&fs_info
->balance_lock
);
2243 fs_info
->balance_ctl
= NULL
;
2244 spin_unlock(&fs_info
->balance_lock
);
2250 * Balance filters. Return 1 if chunk should be filtered out
2251 * (should not be balanced).
2253 static int chunk_profiles_filter(u64 chunk_profile
,
2254 struct btrfs_balance_args
*bargs
)
2256 chunk_profile
&= BTRFS_BLOCK_GROUP_PROFILE_MASK
;
2258 if (chunk_profile
== 0)
2259 chunk_profile
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
2261 if (bargs
->profiles
& chunk_profile
)
2267 static u64
div_factor_fine(u64 num
, int factor
)
2279 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
2280 struct btrfs_balance_args
*bargs
)
2282 struct btrfs_block_group_cache
*cache
;
2283 u64 chunk_used
, user_thresh
;
2286 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
2287 chunk_used
= btrfs_block_group_used(&cache
->item
);
2289 user_thresh
= div_factor_fine(cache
->key
.offset
, bargs
->usage
);
2290 if (chunk_used
< user_thresh
)
2293 btrfs_put_block_group(cache
);
2297 static int chunk_devid_filter(struct extent_buffer
*leaf
,
2298 struct btrfs_chunk
*chunk
,
2299 struct btrfs_balance_args
*bargs
)
2301 struct btrfs_stripe
*stripe
;
2302 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2305 for (i
= 0; i
< num_stripes
; i
++) {
2306 stripe
= btrfs_stripe_nr(chunk
, i
);
2307 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
2314 /* [pstart, pend) */
2315 static int chunk_drange_filter(struct extent_buffer
*leaf
,
2316 struct btrfs_chunk
*chunk
,
2318 struct btrfs_balance_args
*bargs
)
2320 struct btrfs_stripe
*stripe
;
2321 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2327 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
2330 if (btrfs_chunk_type(leaf
, chunk
) & (BTRFS_BLOCK_GROUP_DUP
|
2331 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
))
2335 factor
= num_stripes
/ factor
;
2337 for (i
= 0; i
< num_stripes
; i
++) {
2338 stripe
= btrfs_stripe_nr(chunk
, i
);
2339 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
2342 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
2343 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
2344 do_div(stripe_length
, factor
);
2346 if (stripe_offset
< bargs
->pend
&&
2347 stripe_offset
+ stripe_length
> bargs
->pstart
)
2354 /* [vstart, vend) */
2355 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
2356 struct btrfs_chunk
*chunk
,
2358 struct btrfs_balance_args
*bargs
)
2360 if (chunk_offset
< bargs
->vend
&&
2361 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
2362 /* at least part of the chunk is inside this vrange */
2368 static int chunk_soft_convert_filter(u64 chunk_profile
,
2369 struct btrfs_balance_args
*bargs
)
2371 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
2374 chunk_profile
&= BTRFS_BLOCK_GROUP_PROFILE_MASK
;
2376 if (chunk_profile
== 0)
2377 chunk_profile
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
2379 if (bargs
->target
& chunk_profile
)
2385 static int should_balance_chunk(struct btrfs_root
*root
,
2386 struct extent_buffer
*leaf
,
2387 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
2389 struct btrfs_balance_control
*bctl
= root
->fs_info
->balance_ctl
;
2390 struct btrfs_balance_args
*bargs
= NULL
;
2391 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2394 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
2395 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
2399 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
2400 bargs
= &bctl
->data
;
2401 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
2403 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
2404 bargs
= &bctl
->meta
;
2406 /* profiles filter */
2407 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
2408 chunk_profiles_filter(chunk_type
, bargs
)) {
2413 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2414 chunk_usage_filter(bctl
->fs_info
, chunk_offset
, bargs
)) {
2419 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
2420 chunk_devid_filter(leaf
, chunk
, bargs
)) {
2424 /* drange filter, makes sense only with devid filter */
2425 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
2426 chunk_drange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
2431 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
2432 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
2436 /* soft profile changing mode */
2437 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
2438 chunk_soft_convert_filter(chunk_type
, bargs
)) {
2445 static u64
div_factor(u64 num
, int factor
)
2454 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
2456 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
2457 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
2458 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
2459 struct list_head
*devices
;
2460 struct btrfs_device
*device
;
2463 struct btrfs_chunk
*chunk
;
2464 struct btrfs_path
*path
;
2465 struct btrfs_key key
;
2466 struct btrfs_key found_key
;
2467 struct btrfs_trans_handle
*trans
;
2468 struct extent_buffer
*leaf
;
2471 int enospc_errors
= 0;
2472 bool counting
= true;
2474 /* step one make some room on all the devices */
2475 devices
= &fs_info
->fs_devices
->devices
;
2476 list_for_each_entry(device
, devices
, dev_list
) {
2477 old_size
= device
->total_bytes
;
2478 size_to_free
= div_factor(old_size
, 1);
2479 size_to_free
= min(size_to_free
, (u64
)1 * 1024 * 1024);
2480 if (!device
->writeable
||
2481 device
->total_bytes
- device
->bytes_used
> size_to_free
)
2484 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
2489 trans
= btrfs_start_transaction(dev_root
, 0);
2490 BUG_ON(IS_ERR(trans
));
2492 ret
= btrfs_grow_device(trans
, device
, old_size
);
2495 btrfs_end_transaction(trans
, dev_root
);
2498 /* step two, relocate all the chunks */
2499 path
= btrfs_alloc_path();
2505 /* zero out stat counters */
2506 spin_lock(&fs_info
->balance_lock
);
2507 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
2508 spin_unlock(&fs_info
->balance_lock
);
2510 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2511 key
.offset
= (u64
)-1;
2512 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2515 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
2516 atomic_read(&fs_info
->balance_cancel_req
)) {
2521 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2526 * this shouldn't happen, it means the last relocate
2530 BUG(); /* FIXME break ? */
2532 ret
= btrfs_previous_item(chunk_root
, path
, 0,
2533 BTRFS_CHUNK_ITEM_KEY
);
2539 leaf
= path
->nodes
[0];
2540 slot
= path
->slots
[0];
2541 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2543 if (found_key
.objectid
!= key
.objectid
)
2546 /* chunk zero is special */
2547 if (found_key
.offset
== 0)
2550 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
2553 spin_lock(&fs_info
->balance_lock
);
2554 bctl
->stat
.considered
++;
2555 spin_unlock(&fs_info
->balance_lock
);
2558 ret
= should_balance_chunk(chunk_root
, leaf
, chunk
,
2560 btrfs_release_path(path
);
2565 spin_lock(&fs_info
->balance_lock
);
2566 bctl
->stat
.expected
++;
2567 spin_unlock(&fs_info
->balance_lock
);
2571 ret
= btrfs_relocate_chunk(chunk_root
,
2572 chunk_root
->root_key
.objectid
,
2575 if (ret
&& ret
!= -ENOSPC
)
2577 if (ret
== -ENOSPC
) {
2580 spin_lock(&fs_info
->balance_lock
);
2581 bctl
->stat
.completed
++;
2582 spin_unlock(&fs_info
->balance_lock
);
2585 key
.offset
= found_key
.offset
- 1;
2589 btrfs_release_path(path
);
2594 btrfs_free_path(path
);
2595 if (enospc_errors
) {
2596 printk(KERN_INFO
"btrfs: %d enospc errors during balance\n",
2605 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
2607 /* cancel requested || normal exit path */
2608 return atomic_read(&fs_info
->balance_cancel_req
) ||
2609 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
2610 atomic_read(&fs_info
->balance_cancel_req
) == 0);
2613 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
2617 unset_balance_control(fs_info
);
2618 ret
= del_balance_item(fs_info
->tree_root
);
2622 void update_ioctl_balance_args(struct btrfs_fs_info
*fs_info
, int lock
,
2623 struct btrfs_ioctl_balance_args
*bargs
);
2626 * Should be called with both balance and volume mutexes held
2628 int btrfs_balance(struct btrfs_balance_control
*bctl
,
2629 struct btrfs_ioctl_balance_args
*bargs
)
2631 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
2635 if (btrfs_fs_closing(fs_info
) ||
2636 atomic_read(&fs_info
->balance_pause_req
) ||
2637 atomic_read(&fs_info
->balance_cancel_req
)) {
2643 * In case of mixed groups both data and meta should be picked,
2644 * and identical options should be given for both of them.
2646 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
2647 if ((allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
) &&
2648 (bctl
->flags
& (BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
))) {
2649 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
2650 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
2651 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
2652 printk(KERN_ERR
"btrfs: with mixed groups data and "
2653 "metadata balance options must be the same\n");
2660 * Profile changing sanity checks. Skip them if a simple
2661 * balance is requested.
2663 if (!((bctl
->data
.flags
| bctl
->sys
.flags
| bctl
->meta
.flags
) &
2664 BTRFS_BALANCE_ARGS_CONVERT
))
2667 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
2668 if (fs_info
->fs_devices
->num_devices
== 1)
2669 allowed
|= BTRFS_BLOCK_GROUP_DUP
;
2670 else if (fs_info
->fs_devices
->num_devices
< 4)
2671 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
2673 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
|
2674 BTRFS_BLOCK_GROUP_RAID10
);
2676 if (!profile_is_valid(bctl
->data
.target
, 1) ||
2677 bctl
->data
.target
& ~allowed
) {
2678 printk(KERN_ERR
"btrfs: unable to start balance with target "
2679 "data profile %llu\n",
2680 (unsigned long long)bctl
->data
.target
);
2684 if (!profile_is_valid(bctl
->meta
.target
, 1) ||
2685 bctl
->meta
.target
& ~allowed
) {
2686 printk(KERN_ERR
"btrfs: unable to start balance with target "
2687 "metadata profile %llu\n",
2688 (unsigned long long)bctl
->meta
.target
);
2692 if (!profile_is_valid(bctl
->sys
.target
, 1) ||
2693 bctl
->sys
.target
& ~allowed
) {
2694 printk(KERN_ERR
"btrfs: unable to start balance with target "
2695 "system profile %llu\n",
2696 (unsigned long long)bctl
->sys
.target
);
2701 if (bctl
->data
.target
& BTRFS_BLOCK_GROUP_DUP
) {
2702 printk(KERN_ERR
"btrfs: dup for data is not allowed\n");
2707 /* allow to reduce meta or sys integrity only if force set */
2708 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
2709 BTRFS_BLOCK_GROUP_RAID10
;
2710 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
2711 (fs_info
->avail_system_alloc_bits
& allowed
) &&
2712 !(bctl
->sys
.target
& allowed
)) ||
2713 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
2714 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
2715 !(bctl
->meta
.target
& allowed
))) {
2716 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
2717 printk(KERN_INFO
"btrfs: force reducing metadata "
2720 printk(KERN_ERR
"btrfs: balance will reduce metadata "
2721 "integrity, use force if you want this\n");
2728 ret
= insert_balance_item(fs_info
->tree_root
, bctl
);
2729 if (ret
&& ret
!= -EEXIST
)
2732 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
2733 BUG_ON(ret
== -EEXIST
);
2734 set_balance_control(bctl
);
2736 BUG_ON(ret
!= -EEXIST
);
2737 spin_lock(&fs_info
->balance_lock
);
2738 update_balance_args(bctl
);
2739 spin_unlock(&fs_info
->balance_lock
);
2742 atomic_inc(&fs_info
->balance_running
);
2743 mutex_unlock(&fs_info
->balance_mutex
);
2745 ret
= __btrfs_balance(fs_info
);
2747 mutex_lock(&fs_info
->balance_mutex
);
2748 atomic_dec(&fs_info
->balance_running
);
2751 memset(bargs
, 0, sizeof(*bargs
));
2752 update_ioctl_balance_args(fs_info
, 0, bargs
);
2755 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
2756 balance_need_close(fs_info
)) {
2757 __cancel_balance(fs_info
);
2760 wake_up(&fs_info
->balance_wait_q
);
2764 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
2765 __cancel_balance(fs_info
);
2771 static int balance_kthread(void *data
)
2773 struct btrfs_balance_control
*bctl
=
2774 (struct btrfs_balance_control
*)data
;
2775 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
2778 mutex_lock(&fs_info
->volume_mutex
);
2779 mutex_lock(&fs_info
->balance_mutex
);
2781 set_balance_control(bctl
);
2783 if (btrfs_test_opt(fs_info
->tree_root
, SKIP_BALANCE
)) {
2784 printk(KERN_INFO
"btrfs: force skipping balance\n");
2786 printk(KERN_INFO
"btrfs: continuing balance\n");
2787 ret
= btrfs_balance(bctl
, NULL
);
2790 mutex_unlock(&fs_info
->balance_mutex
);
2791 mutex_unlock(&fs_info
->volume_mutex
);
2795 int btrfs_recover_balance(struct btrfs_root
*tree_root
)
2797 struct task_struct
*tsk
;
2798 struct btrfs_balance_control
*bctl
;
2799 struct btrfs_balance_item
*item
;
2800 struct btrfs_disk_balance_args disk_bargs
;
2801 struct btrfs_path
*path
;
2802 struct extent_buffer
*leaf
;
2803 struct btrfs_key key
;
2806 path
= btrfs_alloc_path();
2810 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
2816 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2817 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2820 ret
= btrfs_search_slot(NULL
, tree_root
, &key
, path
, 0, 0);
2823 if (ret
> 0) { /* ret = -ENOENT; */
2828 leaf
= path
->nodes
[0];
2829 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
2831 bctl
->fs_info
= tree_root
->fs_info
;
2832 bctl
->flags
= btrfs_balance_flags(leaf
, item
) | BTRFS_BALANCE_RESUME
;
2834 btrfs_balance_data(leaf
, item
, &disk_bargs
);
2835 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
2836 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
2837 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
2838 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
2839 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
2841 tsk
= kthread_run(balance_kthread
, bctl
, "btrfs-balance");
2850 btrfs_free_path(path
);
2854 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
2858 mutex_lock(&fs_info
->balance_mutex
);
2859 if (!fs_info
->balance_ctl
) {
2860 mutex_unlock(&fs_info
->balance_mutex
);
2864 if (atomic_read(&fs_info
->balance_running
)) {
2865 atomic_inc(&fs_info
->balance_pause_req
);
2866 mutex_unlock(&fs_info
->balance_mutex
);
2868 wait_event(fs_info
->balance_wait_q
,
2869 atomic_read(&fs_info
->balance_running
) == 0);
2871 mutex_lock(&fs_info
->balance_mutex
);
2872 /* we are good with balance_ctl ripped off from under us */
2873 BUG_ON(atomic_read(&fs_info
->balance_running
));
2874 atomic_dec(&fs_info
->balance_pause_req
);
2879 mutex_unlock(&fs_info
->balance_mutex
);
2883 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
2885 mutex_lock(&fs_info
->balance_mutex
);
2886 if (!fs_info
->balance_ctl
) {
2887 mutex_unlock(&fs_info
->balance_mutex
);
2891 atomic_inc(&fs_info
->balance_cancel_req
);
2893 * if we are running just wait and return, balance item is
2894 * deleted in btrfs_balance in this case
2896 if (atomic_read(&fs_info
->balance_running
)) {
2897 mutex_unlock(&fs_info
->balance_mutex
);
2898 wait_event(fs_info
->balance_wait_q
,
2899 atomic_read(&fs_info
->balance_running
) == 0);
2900 mutex_lock(&fs_info
->balance_mutex
);
2902 /* __cancel_balance needs volume_mutex */
2903 mutex_unlock(&fs_info
->balance_mutex
);
2904 mutex_lock(&fs_info
->volume_mutex
);
2905 mutex_lock(&fs_info
->balance_mutex
);
2907 if (fs_info
->balance_ctl
)
2908 __cancel_balance(fs_info
);
2910 mutex_unlock(&fs_info
->volume_mutex
);
2913 BUG_ON(fs_info
->balance_ctl
|| atomic_read(&fs_info
->balance_running
));
2914 atomic_dec(&fs_info
->balance_cancel_req
);
2915 mutex_unlock(&fs_info
->balance_mutex
);
2920 * shrinking a device means finding all of the device extents past
2921 * the new size, and then following the back refs to the chunks.
2922 * The chunk relocation code actually frees the device extent
2924 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
2926 struct btrfs_trans_handle
*trans
;
2927 struct btrfs_root
*root
= device
->dev_root
;
2928 struct btrfs_dev_extent
*dev_extent
= NULL
;
2929 struct btrfs_path
*path
;
2937 bool retried
= false;
2938 struct extent_buffer
*l
;
2939 struct btrfs_key key
;
2940 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
2941 u64 old_total
= btrfs_super_total_bytes(super_copy
);
2942 u64 old_size
= device
->total_bytes
;
2943 u64 diff
= device
->total_bytes
- new_size
;
2945 if (new_size
>= device
->total_bytes
)
2948 path
= btrfs_alloc_path();
2956 device
->total_bytes
= new_size
;
2957 if (device
->writeable
) {
2958 device
->fs_devices
->total_rw_bytes
-= diff
;
2959 spin_lock(&root
->fs_info
->free_chunk_lock
);
2960 root
->fs_info
->free_chunk_space
-= diff
;
2961 spin_unlock(&root
->fs_info
->free_chunk_lock
);
2963 unlock_chunks(root
);
2966 key
.objectid
= device
->devid
;
2967 key
.offset
= (u64
)-1;
2968 key
.type
= BTRFS_DEV_EXTENT_KEY
;
2971 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2975 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
2980 btrfs_release_path(path
);
2985 slot
= path
->slots
[0];
2986 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
2988 if (key
.objectid
!= device
->devid
) {
2989 btrfs_release_path(path
);
2993 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
2994 length
= btrfs_dev_extent_length(l
, dev_extent
);
2996 if (key
.offset
+ length
<= new_size
) {
2997 btrfs_release_path(path
);
3001 chunk_tree
= btrfs_dev_extent_chunk_tree(l
, dev_extent
);
3002 chunk_objectid
= btrfs_dev_extent_chunk_objectid(l
, dev_extent
);
3003 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
3004 btrfs_release_path(path
);
3006 ret
= btrfs_relocate_chunk(root
, chunk_tree
, chunk_objectid
,
3008 if (ret
&& ret
!= -ENOSPC
)
3015 if (failed
&& !retried
) {
3019 } else if (failed
&& retried
) {
3023 device
->total_bytes
= old_size
;
3024 if (device
->writeable
)
3025 device
->fs_devices
->total_rw_bytes
+= diff
;
3026 spin_lock(&root
->fs_info
->free_chunk_lock
);
3027 root
->fs_info
->free_chunk_space
+= diff
;
3028 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3029 unlock_chunks(root
);
3033 /* Shrinking succeeded, else we would be at "done". */
3034 trans
= btrfs_start_transaction(root
, 0);
3035 if (IS_ERR(trans
)) {
3036 ret
= PTR_ERR(trans
);
3042 device
->disk_total_bytes
= new_size
;
3043 /* Now btrfs_update_device() will change the on-disk size. */
3044 ret
= btrfs_update_device(trans
, device
);
3046 unlock_chunks(root
);
3047 btrfs_end_transaction(trans
, root
);
3050 WARN_ON(diff
> old_total
);
3051 btrfs_set_super_total_bytes(super_copy
, old_total
- diff
);
3052 unlock_chunks(root
);
3053 btrfs_end_transaction(trans
, root
);
3055 btrfs_free_path(path
);
3059 static int btrfs_add_system_chunk(struct btrfs_root
*root
,
3060 struct btrfs_key
*key
,
3061 struct btrfs_chunk
*chunk
, int item_size
)
3063 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
3064 struct btrfs_disk_key disk_key
;
3068 array_size
= btrfs_super_sys_array_size(super_copy
);
3069 if (array_size
+ item_size
> BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
)
3072 ptr
= super_copy
->sys_chunk_array
+ array_size
;
3073 btrfs_cpu_key_to_disk(&disk_key
, key
);
3074 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
3075 ptr
+= sizeof(disk_key
);
3076 memcpy(ptr
, chunk
, item_size
);
3077 item_size
+= sizeof(disk_key
);
3078 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
3083 * sort the devices in descending order by max_avail, total_avail
3085 static int btrfs_cmp_device_info(const void *a
, const void *b
)
3087 const struct btrfs_device_info
*di_a
= a
;
3088 const struct btrfs_device_info
*di_b
= b
;
3090 if (di_a
->max_avail
> di_b
->max_avail
)
3092 if (di_a
->max_avail
< di_b
->max_avail
)
3094 if (di_a
->total_avail
> di_b
->total_avail
)
3096 if (di_a
->total_avail
< di_b
->total_avail
)
3101 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
3102 struct btrfs_root
*extent_root
,
3103 struct map_lookup
**map_ret
,
3104 u64
*num_bytes_out
, u64
*stripe_size_out
,
3105 u64 start
, u64 type
)
3107 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
3108 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
3109 struct list_head
*cur
;
3110 struct map_lookup
*map
= NULL
;
3111 struct extent_map_tree
*em_tree
;
3112 struct extent_map
*em
;
3113 struct btrfs_device_info
*devices_info
= NULL
;
3115 int num_stripes
; /* total number of stripes to allocate */
3116 int sub_stripes
; /* sub_stripes info for map */
3117 int dev_stripes
; /* stripes per dev */
3118 int devs_max
; /* max devs to use */
3119 int devs_min
; /* min devs needed */
3120 int devs_increment
; /* ndevs has to be a multiple of this */
3121 int ncopies
; /* how many copies to data has */
3123 u64 max_stripe_size
;
3131 if ((type
& BTRFS_BLOCK_GROUP_RAID1
) &&
3132 (type
& BTRFS_BLOCK_GROUP_DUP
)) {
3134 type
&= ~BTRFS_BLOCK_GROUP_DUP
;
3137 if (list_empty(&fs_devices
->alloc_list
))
3144 devs_max
= 0; /* 0 == as many as possible */
3148 * define the properties of each RAID type.
3149 * FIXME: move this to a global table and use it in all RAID
3152 if (type
& (BTRFS_BLOCK_GROUP_DUP
)) {
3156 } else if (type
& (BTRFS_BLOCK_GROUP_RAID0
)) {
3158 } else if (type
& (BTRFS_BLOCK_GROUP_RAID1
)) {
3163 } else if (type
& (BTRFS_BLOCK_GROUP_RAID10
)) {
3172 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
3173 max_stripe_size
= 1024 * 1024 * 1024;
3174 max_chunk_size
= 10 * max_stripe_size
;
3175 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
3176 /* for larger filesystems, use larger metadata chunks */
3177 if (fs_devices
->total_rw_bytes
> 50ULL * 1024 * 1024 * 1024)
3178 max_stripe_size
= 1024 * 1024 * 1024;
3180 max_stripe_size
= 256 * 1024 * 1024;
3181 max_chunk_size
= max_stripe_size
;
3182 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3183 max_stripe_size
= 32 * 1024 * 1024;
3184 max_chunk_size
= 2 * max_stripe_size
;
3186 printk(KERN_ERR
"btrfs: invalid chunk type 0x%llx requested\n",
3191 /* we don't want a chunk larger than 10% of writeable space */
3192 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
3195 devices_info
= kzalloc(sizeof(*devices_info
) * fs_devices
->rw_devices
,
3200 cur
= fs_devices
->alloc_list
.next
;
3203 * in the first pass through the devices list, we gather information
3204 * about the available holes on each device.
3207 while (cur
!= &fs_devices
->alloc_list
) {
3208 struct btrfs_device
*device
;
3212 device
= list_entry(cur
, struct btrfs_device
, dev_alloc_list
);
3216 if (!device
->writeable
) {
3218 "btrfs: read-only device in alloc_list\n");
3223 if (!device
->in_fs_metadata
)
3226 if (device
->total_bytes
> device
->bytes_used
)
3227 total_avail
= device
->total_bytes
- device
->bytes_used
;
3231 /* If there is no space on this device, skip it. */
3232 if (total_avail
== 0)
3235 ret
= find_free_dev_extent(device
,
3236 max_stripe_size
* dev_stripes
,
3237 &dev_offset
, &max_avail
);
3238 if (ret
&& ret
!= -ENOSPC
)
3242 max_avail
= max_stripe_size
* dev_stripes
;
3244 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
3247 devices_info
[ndevs
].dev_offset
= dev_offset
;
3248 devices_info
[ndevs
].max_avail
= max_avail
;
3249 devices_info
[ndevs
].total_avail
= total_avail
;
3250 devices_info
[ndevs
].dev
= device
;
3255 * now sort the devices by hole size / available space
3257 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
3258 btrfs_cmp_device_info
, NULL
);
3260 /* round down to number of usable stripes */
3261 ndevs
-= ndevs
% devs_increment
;
3263 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
3268 if (devs_max
&& ndevs
> devs_max
)
3271 * the primary goal is to maximize the number of stripes, so use as many
3272 * devices as possible, even if the stripes are not maximum sized.
3274 stripe_size
= devices_info
[ndevs
-1].max_avail
;
3275 num_stripes
= ndevs
* dev_stripes
;
3277 if (stripe_size
* num_stripes
> max_chunk_size
* ncopies
) {
3278 stripe_size
= max_chunk_size
* ncopies
;
3279 do_div(stripe_size
, num_stripes
);
3282 do_div(stripe_size
, dev_stripes
);
3283 do_div(stripe_size
, BTRFS_STRIPE_LEN
);
3284 stripe_size
*= BTRFS_STRIPE_LEN
;
3286 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
3291 map
->num_stripes
= num_stripes
;
3293 for (i
= 0; i
< ndevs
; ++i
) {
3294 for (j
= 0; j
< dev_stripes
; ++j
) {
3295 int s
= i
* dev_stripes
+ j
;
3296 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
3297 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
3301 map
->sector_size
= extent_root
->sectorsize
;
3302 map
->stripe_len
= BTRFS_STRIPE_LEN
;
3303 map
->io_align
= BTRFS_STRIPE_LEN
;
3304 map
->io_width
= BTRFS_STRIPE_LEN
;
3306 map
->sub_stripes
= sub_stripes
;
3309 num_bytes
= stripe_size
* (num_stripes
/ ncopies
);
3311 *stripe_size_out
= stripe_size
;
3312 *num_bytes_out
= num_bytes
;
3314 trace_btrfs_chunk_alloc(info
->chunk_root
, map
, start
, num_bytes
);
3316 em
= alloc_extent_map();
3321 em
->bdev
= (struct block_device
*)map
;
3323 em
->len
= num_bytes
;
3324 em
->block_start
= 0;
3325 em
->block_len
= em
->len
;
3327 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
3328 write_lock(&em_tree
->lock
);
3329 ret
= add_extent_mapping(em_tree
, em
);
3330 write_unlock(&em_tree
->lock
);
3332 free_extent_map(em
);
3334 ret
= btrfs_make_block_group(trans
, extent_root
, 0, type
,
3335 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
3339 for (i
= 0; i
< map
->num_stripes
; ++i
) {
3340 struct btrfs_device
*device
;
3343 device
= map
->stripes
[i
].dev
;
3344 dev_offset
= map
->stripes
[i
].physical
;
3346 ret
= btrfs_alloc_dev_extent(trans
, device
,
3347 info
->chunk_root
->root_key
.objectid
,
3348 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
3349 start
, dev_offset
, stripe_size
);
3353 kfree(devices_info
);
3358 kfree(devices_info
);
3362 static int __finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
3363 struct btrfs_root
*extent_root
,
3364 struct map_lookup
*map
, u64 chunk_offset
,
3365 u64 chunk_size
, u64 stripe_size
)
3368 struct btrfs_key key
;
3369 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
3370 struct btrfs_device
*device
;
3371 struct btrfs_chunk
*chunk
;
3372 struct btrfs_stripe
*stripe
;
3373 size_t item_size
= btrfs_chunk_item_size(map
->num_stripes
);
3377 chunk
= kzalloc(item_size
, GFP_NOFS
);
3382 while (index
< map
->num_stripes
) {
3383 device
= map
->stripes
[index
].dev
;
3384 device
->bytes_used
+= stripe_size
;
3385 ret
= btrfs_update_device(trans
, device
);
3390 spin_lock(&extent_root
->fs_info
->free_chunk_lock
);
3391 extent_root
->fs_info
->free_chunk_space
-= (stripe_size
*
3393 spin_unlock(&extent_root
->fs_info
->free_chunk_lock
);
3396 stripe
= &chunk
->stripe
;
3397 while (index
< map
->num_stripes
) {
3398 device
= map
->stripes
[index
].dev
;
3399 dev_offset
= map
->stripes
[index
].physical
;
3401 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
3402 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
3403 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
3408 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
3409 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
3410 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
3411 btrfs_set_stack_chunk_type(chunk
, map
->type
);
3412 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
3413 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
3414 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
3415 btrfs_set_stack_chunk_sector_size(chunk
, extent_root
->sectorsize
);
3416 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
3418 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3419 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3420 key
.offset
= chunk_offset
;
3422 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
3425 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3426 ret
= btrfs_add_system_chunk(chunk_root
, &key
, chunk
,
3436 * Chunk allocation falls into two parts. The first part does works
3437 * that make the new allocated chunk useable, but not do any operation
3438 * that modifies the chunk tree. The second part does the works that
3439 * require modifying the chunk tree. This division is important for the
3440 * bootstrap process of adding storage to a seed btrfs.
3442 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
3443 struct btrfs_root
*extent_root
, u64 type
)
3448 struct map_lookup
*map
;
3449 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
3452 ret
= find_next_chunk(chunk_root
, BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
3457 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &map
, &chunk_size
,
3458 &stripe_size
, chunk_offset
, type
);
3462 ret
= __finish_chunk_alloc(trans
, extent_root
, map
, chunk_offset
,
3463 chunk_size
, stripe_size
);
3468 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
3469 struct btrfs_root
*root
,
3470 struct btrfs_device
*device
)
3473 u64 sys_chunk_offset
;
3477 u64 sys_stripe_size
;
3479 struct map_lookup
*map
;
3480 struct map_lookup
*sys_map
;
3481 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3482 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
3485 ret
= find_next_chunk(fs_info
->chunk_root
,
3486 BTRFS_FIRST_CHUNK_TREE_OBJECTID
, &chunk_offset
);
3490 alloc_profile
= BTRFS_BLOCK_GROUP_METADATA
|
3491 fs_info
->avail_metadata_alloc_bits
;
3492 alloc_profile
= btrfs_reduce_alloc_profile(root
, alloc_profile
);
3494 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &map
, &chunk_size
,
3495 &stripe_size
, chunk_offset
, alloc_profile
);
3498 sys_chunk_offset
= chunk_offset
+ chunk_size
;
3500 alloc_profile
= BTRFS_BLOCK_GROUP_SYSTEM
|
3501 fs_info
->avail_system_alloc_bits
;
3502 alloc_profile
= btrfs_reduce_alloc_profile(root
, alloc_profile
);
3504 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &sys_map
,
3505 &sys_chunk_size
, &sys_stripe_size
,
3506 sys_chunk_offset
, alloc_profile
);
3509 ret
= btrfs_add_device(trans
, fs_info
->chunk_root
, device
);
3513 * Modifying chunk tree needs allocating new blocks from both
3514 * system block group and metadata block group. So we only can
3515 * do operations require modifying the chunk tree after both
3516 * block groups were created.
3518 ret
= __finish_chunk_alloc(trans
, extent_root
, map
, chunk_offset
,
3519 chunk_size
, stripe_size
);
3522 ret
= __finish_chunk_alloc(trans
, extent_root
, sys_map
,
3523 sys_chunk_offset
, sys_chunk_size
,
3529 int btrfs_chunk_readonly(struct btrfs_root
*root
, u64 chunk_offset
)
3531 struct extent_map
*em
;
3532 struct map_lookup
*map
;
3533 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
3537 read_lock(&map_tree
->map_tree
.lock
);
3538 em
= lookup_extent_mapping(&map_tree
->map_tree
, chunk_offset
, 1);
3539 read_unlock(&map_tree
->map_tree
.lock
);
3543 if (btrfs_test_opt(root
, DEGRADED
)) {
3544 free_extent_map(em
);
3548 map
= (struct map_lookup
*)em
->bdev
;
3549 for (i
= 0; i
< map
->num_stripes
; i
++) {
3550 if (!map
->stripes
[i
].dev
->writeable
) {
3555 free_extent_map(em
);
3559 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
3561 extent_map_tree_init(&tree
->map_tree
);
3564 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
3566 struct extent_map
*em
;
3569 write_lock(&tree
->map_tree
.lock
);
3570 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
3572 remove_extent_mapping(&tree
->map_tree
, em
);
3573 write_unlock(&tree
->map_tree
.lock
);
3578 free_extent_map(em
);
3579 /* once for the tree */
3580 free_extent_map(em
);
3584 int btrfs_num_copies(struct btrfs_mapping_tree
*map_tree
, u64 logical
, u64 len
)
3586 struct extent_map
*em
;
3587 struct map_lookup
*map
;
3588 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
3591 read_lock(&em_tree
->lock
);
3592 em
= lookup_extent_mapping(em_tree
, logical
, len
);
3593 read_unlock(&em_tree
->lock
);
3596 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
3597 map
= (struct map_lookup
*)em
->bdev
;
3598 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
3599 ret
= map
->num_stripes
;
3600 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
3601 ret
= map
->sub_stripes
;
3604 free_extent_map(em
);
3608 static int find_live_mirror(struct map_lookup
*map
, int first
, int num
,
3612 if (map
->stripes
[optimal
].dev
->bdev
)
3614 for (i
= first
; i
< first
+ num
; i
++) {
3615 if (map
->stripes
[i
].dev
->bdev
)
3618 /* we couldn't find one that doesn't fail. Just return something
3619 * and the io error handling code will clean up eventually
3624 static int __btrfs_map_block(struct btrfs_mapping_tree
*map_tree
, int rw
,
3625 u64 logical
, u64
*length
,
3626 struct btrfs_bio
**bbio_ret
,
3629 struct extent_map
*em
;
3630 struct map_lookup
*map
;
3631 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
3634 u64 stripe_end_offset
;
3643 struct btrfs_bio
*bbio
= NULL
;
3645 read_lock(&em_tree
->lock
);
3646 em
= lookup_extent_mapping(em_tree
, logical
, *length
);
3647 read_unlock(&em_tree
->lock
);
3650 printk(KERN_CRIT
"unable to find logical %llu len %llu\n",
3651 (unsigned long long)logical
,
3652 (unsigned long long)*length
);
3656 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
3657 map
= (struct map_lookup
*)em
->bdev
;
3658 offset
= logical
- em
->start
;
3660 if (mirror_num
> map
->num_stripes
)
3665 * stripe_nr counts the total number of stripes we have to stride
3666 * to get to this block
3668 do_div(stripe_nr
, map
->stripe_len
);
3670 stripe_offset
= stripe_nr
* map
->stripe_len
;
3671 BUG_ON(offset
< stripe_offset
);
3673 /* stripe_offset is the offset of this block in its stripe*/
3674 stripe_offset
= offset
- stripe_offset
;
3676 if (rw
& REQ_DISCARD
)
3677 *length
= min_t(u64
, em
->len
- offset
, *length
);
3678 else if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
3679 /* we limit the length of each bio to what fits in a stripe */
3680 *length
= min_t(u64
, em
->len
- offset
,
3681 map
->stripe_len
- stripe_offset
);
3683 *length
= em
->len
- offset
;
3691 stripe_nr_orig
= stripe_nr
;
3692 stripe_nr_end
= (offset
+ *length
+ map
->stripe_len
- 1) &
3693 (~(map
->stripe_len
- 1));
3694 do_div(stripe_nr_end
, map
->stripe_len
);
3695 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
3697 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
3698 if (rw
& REQ_DISCARD
)
3699 num_stripes
= min_t(u64
, map
->num_stripes
,
3700 stripe_nr_end
- stripe_nr_orig
);
3701 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
3702 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
3703 if (rw
& (REQ_WRITE
| REQ_DISCARD
))
3704 num_stripes
= map
->num_stripes
;
3705 else if (mirror_num
)
3706 stripe_index
= mirror_num
- 1;
3708 stripe_index
= find_live_mirror(map
, 0,
3710 current
->pid
% map
->num_stripes
);
3711 mirror_num
= stripe_index
+ 1;
3714 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
3715 if (rw
& (REQ_WRITE
| REQ_DISCARD
)) {
3716 num_stripes
= map
->num_stripes
;
3717 } else if (mirror_num
) {
3718 stripe_index
= mirror_num
- 1;
3723 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
3724 int factor
= map
->num_stripes
/ map
->sub_stripes
;
3726 stripe_index
= do_div(stripe_nr
, factor
);
3727 stripe_index
*= map
->sub_stripes
;
3730 num_stripes
= map
->sub_stripes
;
3731 else if (rw
& REQ_DISCARD
)
3732 num_stripes
= min_t(u64
, map
->sub_stripes
*
3733 (stripe_nr_end
- stripe_nr_orig
),
3735 else if (mirror_num
)
3736 stripe_index
+= mirror_num
- 1;
3738 stripe_index
= find_live_mirror(map
, stripe_index
,
3739 map
->sub_stripes
, stripe_index
+
3740 current
->pid
% map
->sub_stripes
);
3741 mirror_num
= stripe_index
+ 1;
3745 * after this do_div call, stripe_nr is the number of stripes
3746 * on this device we have to walk to find the data, and
3747 * stripe_index is the number of our device in the stripe array
3749 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
3750 mirror_num
= stripe_index
+ 1;
3752 BUG_ON(stripe_index
>= map
->num_stripes
);
3754 bbio
= kzalloc(btrfs_bio_size(num_stripes
), GFP_NOFS
);
3759 atomic_set(&bbio
->error
, 0);
3761 if (rw
& REQ_DISCARD
) {
3763 int sub_stripes
= 0;
3764 u64 stripes_per_dev
= 0;
3765 u32 remaining_stripes
= 0;
3768 (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID10
)) {
3769 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
3772 sub_stripes
= map
->sub_stripes
;
3774 factor
= map
->num_stripes
/ sub_stripes
;
3775 stripes_per_dev
= div_u64_rem(stripe_nr_end
-
3778 &remaining_stripes
);
3781 for (i
= 0; i
< num_stripes
; i
++) {
3782 bbio
->stripes
[i
].physical
=
3783 map
->stripes
[stripe_index
].physical
+
3784 stripe_offset
+ stripe_nr
* map
->stripe_len
;
3785 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
3787 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
3788 BTRFS_BLOCK_GROUP_RAID10
)) {
3789 bbio
->stripes
[i
].length
= stripes_per_dev
*
3791 if (i
/ sub_stripes
< remaining_stripes
)
3792 bbio
->stripes
[i
].length
+=
3794 if (i
< sub_stripes
)
3795 bbio
->stripes
[i
].length
-=
3797 if ((i
/ sub_stripes
+ 1) %
3798 sub_stripes
== remaining_stripes
)
3799 bbio
->stripes
[i
].length
-=
3801 if (i
== sub_stripes
- 1)
3804 bbio
->stripes
[i
].length
= *length
;
3807 if (stripe_index
== map
->num_stripes
) {
3808 /* This could only happen for RAID0/10 */
3814 for (i
= 0; i
< num_stripes
; i
++) {
3815 bbio
->stripes
[i
].physical
=
3816 map
->stripes
[stripe_index
].physical
+
3818 stripe_nr
* map
->stripe_len
;
3819 bbio
->stripes
[i
].dev
=
3820 map
->stripes
[stripe_index
].dev
;
3825 if (rw
& REQ_WRITE
) {
3826 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
3827 BTRFS_BLOCK_GROUP_RAID10
|
3828 BTRFS_BLOCK_GROUP_DUP
)) {
3834 bbio
->num_stripes
= num_stripes
;
3835 bbio
->max_errors
= max_errors
;
3836 bbio
->mirror_num
= mirror_num
;
3838 free_extent_map(em
);
3842 int btrfs_map_block(struct btrfs_mapping_tree
*map_tree
, int rw
,
3843 u64 logical
, u64
*length
,
3844 struct btrfs_bio
**bbio_ret
, int mirror_num
)
3846 return __btrfs_map_block(map_tree
, rw
, logical
, length
, bbio_ret
,
3850 int btrfs_rmap_block(struct btrfs_mapping_tree
*map_tree
,
3851 u64 chunk_start
, u64 physical
, u64 devid
,
3852 u64
**logical
, int *naddrs
, int *stripe_len
)
3854 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
3855 struct extent_map
*em
;
3856 struct map_lookup
*map
;
3863 read_lock(&em_tree
->lock
);
3864 em
= lookup_extent_mapping(em_tree
, chunk_start
, 1);
3865 read_unlock(&em_tree
->lock
);
3867 BUG_ON(!em
|| em
->start
!= chunk_start
);
3868 map
= (struct map_lookup
*)em
->bdev
;
3871 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
3872 do_div(length
, map
->num_stripes
/ map
->sub_stripes
);
3873 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
3874 do_div(length
, map
->num_stripes
);
3876 buf
= kzalloc(sizeof(u64
) * map
->num_stripes
, GFP_NOFS
);
3879 for (i
= 0; i
< map
->num_stripes
; i
++) {
3880 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
3882 if (map
->stripes
[i
].physical
> physical
||
3883 map
->stripes
[i
].physical
+ length
<= physical
)
3886 stripe_nr
= physical
- map
->stripes
[i
].physical
;
3887 do_div(stripe_nr
, map
->stripe_len
);
3889 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
3890 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
3891 do_div(stripe_nr
, map
->sub_stripes
);
3892 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
3893 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
3895 bytenr
= chunk_start
+ stripe_nr
* map
->stripe_len
;
3896 WARN_ON(nr
>= map
->num_stripes
);
3897 for (j
= 0; j
< nr
; j
++) {
3898 if (buf
[j
] == bytenr
)
3902 WARN_ON(nr
>= map
->num_stripes
);
3909 *stripe_len
= map
->stripe_len
;
3911 free_extent_map(em
);
3915 static void btrfs_end_bio(struct bio
*bio
, int err
)
3917 struct btrfs_bio
*bbio
= bio
->bi_private
;
3918 int is_orig_bio
= 0;
3921 atomic_inc(&bbio
->error
);
3923 if (bio
== bbio
->orig_bio
)
3926 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
3929 bio
= bbio
->orig_bio
;
3931 bio
->bi_private
= bbio
->private;
3932 bio
->bi_end_io
= bbio
->end_io
;
3933 bio
->bi_bdev
= (struct block_device
*)
3934 (unsigned long)bbio
->mirror_num
;
3935 /* only send an error to the higher layers if it is
3936 * beyond the tolerance of the multi-bio
3938 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
3942 * this bio is actually up to date, we didn't
3943 * go over the max number of errors
3945 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
3950 bio_endio(bio
, err
);
3951 } else if (!is_orig_bio
) {
3956 struct async_sched
{
3959 struct btrfs_fs_info
*info
;
3960 struct btrfs_work work
;
3964 * see run_scheduled_bios for a description of why bios are collected for
3967 * This will add one bio to the pending list for a device and make sure
3968 * the work struct is scheduled.
3970 static noinline
int schedule_bio(struct btrfs_root
*root
,
3971 struct btrfs_device
*device
,
3972 int rw
, struct bio
*bio
)
3974 int should_queue
= 1;
3975 struct btrfs_pending_bios
*pending_bios
;
3977 /* don't bother with additional async steps for reads, right now */
3978 if (!(rw
& REQ_WRITE
)) {
3980 btrfsic_submit_bio(rw
, bio
);
3986 * nr_async_bios allows us to reliably return congestion to the
3987 * higher layers. Otherwise, the async bio makes it appear we have
3988 * made progress against dirty pages when we've really just put it
3989 * on a queue for later
3991 atomic_inc(&root
->fs_info
->nr_async_bios
);
3992 WARN_ON(bio
->bi_next
);
3993 bio
->bi_next
= NULL
;
3996 spin_lock(&device
->io_lock
);
3997 if (bio
->bi_rw
& REQ_SYNC
)
3998 pending_bios
= &device
->pending_sync_bios
;
4000 pending_bios
= &device
->pending_bios
;
4002 if (pending_bios
->tail
)
4003 pending_bios
->tail
->bi_next
= bio
;
4005 pending_bios
->tail
= bio
;
4006 if (!pending_bios
->head
)
4007 pending_bios
->head
= bio
;
4008 if (device
->running_pending
)
4011 spin_unlock(&device
->io_lock
);
4014 btrfs_queue_worker(&root
->fs_info
->submit_workers
,
4019 int btrfs_map_bio(struct btrfs_root
*root
, int rw
, struct bio
*bio
,
4020 int mirror_num
, int async_submit
)
4022 struct btrfs_mapping_tree
*map_tree
;
4023 struct btrfs_device
*dev
;
4024 struct bio
*first_bio
= bio
;
4025 u64 logical
= (u64
)bio
->bi_sector
<< 9;
4031 struct btrfs_bio
*bbio
= NULL
;
4033 length
= bio
->bi_size
;
4034 map_tree
= &root
->fs_info
->mapping_tree
;
4035 map_length
= length
;
4037 ret
= btrfs_map_block(map_tree
, rw
, logical
, &map_length
, &bbio
,
4041 total_devs
= bbio
->num_stripes
;
4042 if (map_length
< length
) {
4043 printk(KERN_CRIT
"mapping failed logical %llu bio len %llu "
4044 "len %llu\n", (unsigned long long)logical
,
4045 (unsigned long long)length
,
4046 (unsigned long long)map_length
);
4050 bbio
->orig_bio
= first_bio
;
4051 bbio
->private = first_bio
->bi_private
;
4052 bbio
->end_io
= first_bio
->bi_end_io
;
4053 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
4055 while (dev_nr
< total_devs
) {
4056 if (dev_nr
< total_devs
- 1) {
4057 bio
= bio_clone(first_bio
, GFP_NOFS
);
4062 bio
->bi_private
= bbio
;
4063 bio
->bi_end_io
= btrfs_end_bio
;
4064 bio
->bi_sector
= bbio
->stripes
[dev_nr
].physical
>> 9;
4065 dev
= bbio
->stripes
[dev_nr
].dev
;
4066 if (dev
&& dev
->bdev
&& (rw
!= WRITE
|| dev
->writeable
)) {
4067 pr_debug("btrfs_map_bio: rw %d, secor=%llu, dev=%lu "
4068 "(%s id %llu), size=%u\n", rw
,
4069 (u64
)bio
->bi_sector
, (u_long
)dev
->bdev
->bd_dev
,
4070 dev
->name
, dev
->devid
, bio
->bi_size
);
4071 bio
->bi_bdev
= dev
->bdev
;
4073 schedule_bio(root
, dev
, rw
, bio
);
4075 btrfsic_submit_bio(rw
, bio
);
4077 bio
->bi_bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4078 bio
->bi_sector
= logical
>> 9;
4079 bio_endio(bio
, -EIO
);
4086 struct btrfs_device
*btrfs_find_device(struct btrfs_root
*root
, u64 devid
,
4089 struct btrfs_device
*device
;
4090 struct btrfs_fs_devices
*cur_devices
;
4092 cur_devices
= root
->fs_info
->fs_devices
;
4093 while (cur_devices
) {
4095 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
4096 device
= __find_device(&cur_devices
->devices
,
4101 cur_devices
= cur_devices
->seed
;
4106 static struct btrfs_device
*add_missing_dev(struct btrfs_root
*root
,
4107 u64 devid
, u8
*dev_uuid
)
4109 struct btrfs_device
*device
;
4110 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
4112 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
4115 list_add(&device
->dev_list
,
4116 &fs_devices
->devices
);
4117 device
->dev_root
= root
->fs_info
->dev_root
;
4118 device
->devid
= devid
;
4119 device
->work
.func
= pending_bios_fn
;
4120 device
->fs_devices
= fs_devices
;
4121 device
->missing
= 1;
4122 fs_devices
->num_devices
++;
4123 fs_devices
->missing_devices
++;
4124 spin_lock_init(&device
->io_lock
);
4125 INIT_LIST_HEAD(&device
->dev_alloc_list
);
4126 memcpy(device
->uuid
, dev_uuid
, BTRFS_UUID_SIZE
);
4130 static int read_one_chunk(struct btrfs_root
*root
, struct btrfs_key
*key
,
4131 struct extent_buffer
*leaf
,
4132 struct btrfs_chunk
*chunk
)
4134 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
4135 struct map_lookup
*map
;
4136 struct extent_map
*em
;
4140 u8 uuid
[BTRFS_UUID_SIZE
];
4145 logical
= key
->offset
;
4146 length
= btrfs_chunk_length(leaf
, chunk
);
4148 read_lock(&map_tree
->map_tree
.lock
);
4149 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
4150 read_unlock(&map_tree
->map_tree
.lock
);
4152 /* already mapped? */
4153 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
4154 free_extent_map(em
);
4157 free_extent_map(em
);
4160 em
= alloc_extent_map();
4163 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
4164 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
4166 free_extent_map(em
);
4170 em
->bdev
= (struct block_device
*)map
;
4171 em
->start
= logical
;
4173 em
->block_start
= 0;
4174 em
->block_len
= em
->len
;
4176 map
->num_stripes
= num_stripes
;
4177 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
4178 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
4179 map
->sector_size
= btrfs_chunk_sector_size(leaf
, chunk
);
4180 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
4181 map
->type
= btrfs_chunk_type(leaf
, chunk
);
4182 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
4183 for (i
= 0; i
< num_stripes
; i
++) {
4184 map
->stripes
[i
].physical
=
4185 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
4186 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
4187 read_extent_buffer(leaf
, uuid
, (unsigned long)
4188 btrfs_stripe_dev_uuid_nr(chunk
, i
),
4190 map
->stripes
[i
].dev
= btrfs_find_device(root
, devid
, uuid
,
4192 if (!map
->stripes
[i
].dev
&& !btrfs_test_opt(root
, DEGRADED
)) {
4194 free_extent_map(em
);
4197 if (!map
->stripes
[i
].dev
) {
4198 map
->stripes
[i
].dev
=
4199 add_missing_dev(root
, devid
, uuid
);
4200 if (!map
->stripes
[i
].dev
) {
4202 free_extent_map(em
);
4206 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
4209 write_lock(&map_tree
->map_tree
.lock
);
4210 ret
= add_extent_mapping(&map_tree
->map_tree
, em
);
4211 write_unlock(&map_tree
->map_tree
.lock
);
4213 free_extent_map(em
);
4218 static int fill_device_from_item(struct extent_buffer
*leaf
,
4219 struct btrfs_dev_item
*dev_item
,
4220 struct btrfs_device
*device
)
4224 device
->devid
= btrfs_device_id(leaf
, dev_item
);
4225 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
4226 device
->total_bytes
= device
->disk_total_bytes
;
4227 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
4228 device
->type
= btrfs_device_type(leaf
, dev_item
);
4229 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
4230 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
4231 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
4233 ptr
= (unsigned long)btrfs_device_uuid(dev_item
);
4234 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
4239 static int open_seed_devices(struct btrfs_root
*root
, u8
*fsid
)
4241 struct btrfs_fs_devices
*fs_devices
;
4244 BUG_ON(!mutex_is_locked(&uuid_mutex
));
4246 fs_devices
= root
->fs_info
->fs_devices
->seed
;
4247 while (fs_devices
) {
4248 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
4252 fs_devices
= fs_devices
->seed
;
4255 fs_devices
= find_fsid(fsid
);
4261 fs_devices
= clone_fs_devices(fs_devices
);
4262 if (IS_ERR(fs_devices
)) {
4263 ret
= PTR_ERR(fs_devices
);
4267 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
4268 root
->fs_info
->bdev_holder
);
4272 if (!fs_devices
->seeding
) {
4273 __btrfs_close_devices(fs_devices
);
4274 free_fs_devices(fs_devices
);
4279 fs_devices
->seed
= root
->fs_info
->fs_devices
->seed
;
4280 root
->fs_info
->fs_devices
->seed
= fs_devices
;
4285 static int read_one_dev(struct btrfs_root
*root
,
4286 struct extent_buffer
*leaf
,
4287 struct btrfs_dev_item
*dev_item
)
4289 struct btrfs_device
*device
;
4292 u8 fs_uuid
[BTRFS_UUID_SIZE
];
4293 u8 dev_uuid
[BTRFS_UUID_SIZE
];
4295 devid
= btrfs_device_id(leaf
, dev_item
);
4296 read_extent_buffer(leaf
, dev_uuid
,
4297 (unsigned long)btrfs_device_uuid(dev_item
),
4299 read_extent_buffer(leaf
, fs_uuid
,
4300 (unsigned long)btrfs_device_fsid(dev_item
),
4303 if (memcmp(fs_uuid
, root
->fs_info
->fsid
, BTRFS_UUID_SIZE
)) {
4304 ret
= open_seed_devices(root
, fs_uuid
);
4305 if (ret
&& !btrfs_test_opt(root
, DEGRADED
))
4309 device
= btrfs_find_device(root
, devid
, dev_uuid
, fs_uuid
);
4310 if (!device
|| !device
->bdev
) {
4311 if (!btrfs_test_opt(root
, DEGRADED
))
4315 printk(KERN_WARNING
"warning devid %llu missing\n",
4316 (unsigned long long)devid
);
4317 device
= add_missing_dev(root
, devid
, dev_uuid
);
4320 } else if (!device
->missing
) {
4322 * this happens when a device that was properly setup
4323 * in the device info lists suddenly goes bad.
4324 * device->bdev is NULL, and so we have to set
4325 * device->missing to one here
4327 root
->fs_info
->fs_devices
->missing_devices
++;
4328 device
->missing
= 1;
4332 if (device
->fs_devices
!= root
->fs_info
->fs_devices
) {
4333 BUG_ON(device
->writeable
);
4334 if (device
->generation
!=
4335 btrfs_device_generation(leaf
, dev_item
))
4339 fill_device_from_item(leaf
, dev_item
, device
);
4340 device
->dev_root
= root
->fs_info
->dev_root
;
4341 device
->in_fs_metadata
= 1;
4342 if (device
->writeable
) {
4343 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
4344 spin_lock(&root
->fs_info
->free_chunk_lock
);
4345 root
->fs_info
->free_chunk_space
+= device
->total_bytes
-
4347 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4353 int btrfs_read_sys_array(struct btrfs_root
*root
)
4355 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
4356 struct extent_buffer
*sb
;
4357 struct btrfs_disk_key
*disk_key
;
4358 struct btrfs_chunk
*chunk
;
4360 unsigned long sb_ptr
;
4366 struct btrfs_key key
;
4368 sb
= btrfs_find_create_tree_block(root
, BTRFS_SUPER_INFO_OFFSET
,
4369 BTRFS_SUPER_INFO_SIZE
);
4372 btrfs_set_buffer_uptodate(sb
);
4373 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
4375 * The sb extent buffer is artifical and just used to read the system array.
4376 * btrfs_set_buffer_uptodate() call does not properly mark all it's
4377 * pages up-to-date when the page is larger: extent does not cover the
4378 * whole page and consequently check_page_uptodate does not find all
4379 * the page's extents up-to-date (the hole beyond sb),
4380 * write_extent_buffer then triggers a WARN_ON.
4382 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
4383 * but sb spans only this function. Add an explicit SetPageUptodate call
4384 * to silence the warning eg. on PowerPC 64.
4386 if (PAGE_CACHE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
4387 SetPageUptodate(sb
->first_page
);
4389 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
4390 array_size
= btrfs_super_sys_array_size(super_copy
);
4392 ptr
= super_copy
->sys_chunk_array
;
4393 sb_ptr
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
4396 while (cur
< array_size
) {
4397 disk_key
= (struct btrfs_disk_key
*)ptr
;
4398 btrfs_disk_key_to_cpu(&key
, disk_key
);
4400 len
= sizeof(*disk_key
); ptr
+= len
;
4404 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
4405 chunk
= (struct btrfs_chunk
*)sb_ptr
;
4406 ret
= read_one_chunk(root
, &key
, sb
, chunk
);
4409 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
4410 len
= btrfs_chunk_item_size(num_stripes
);
4419 free_extent_buffer(sb
);
4423 int btrfs_read_chunk_tree(struct btrfs_root
*root
)
4425 struct btrfs_path
*path
;
4426 struct extent_buffer
*leaf
;
4427 struct btrfs_key key
;
4428 struct btrfs_key found_key
;
4432 root
= root
->fs_info
->chunk_root
;
4434 path
= btrfs_alloc_path();
4438 mutex_lock(&uuid_mutex
);
4441 /* first we search for all of the device items, and then we
4442 * read in all of the chunk items. This way we can create chunk
4443 * mappings that reference all of the devices that are afound
4445 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
4449 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4453 leaf
= path
->nodes
[0];
4454 slot
= path
->slots
[0];
4455 if (slot
>= btrfs_header_nritems(leaf
)) {
4456 ret
= btrfs_next_leaf(root
, path
);
4463 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
4464 if (key
.objectid
== BTRFS_DEV_ITEMS_OBJECTID
) {
4465 if (found_key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
)
4467 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
4468 struct btrfs_dev_item
*dev_item
;
4469 dev_item
= btrfs_item_ptr(leaf
, slot
,
4470 struct btrfs_dev_item
);
4471 ret
= read_one_dev(root
, leaf
, dev_item
);
4475 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
4476 struct btrfs_chunk
*chunk
;
4477 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
4478 ret
= read_one_chunk(root
, &found_key
, leaf
, chunk
);
4484 if (key
.objectid
== BTRFS_DEV_ITEMS_OBJECTID
) {
4486 btrfs_release_path(path
);
4491 unlock_chunks(root
);
4492 mutex_unlock(&uuid_mutex
);
4494 btrfs_free_path(path
);