2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/ratelimit.h>
27 #include <linux/kthread.h>
28 #include <asm/div64.h>
31 #include "extent_map.h"
33 #include "transaction.h"
34 #include "print-tree.h"
36 #include "async-thread.h"
37 #include "check-integrity.h"
38 #include "rcu-string.h"
40 static int init_first_rw_device(struct btrfs_trans_handle
*trans
,
41 struct btrfs_root
*root
,
42 struct btrfs_device
*device
);
43 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
);
44 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
);
45 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*device
);
47 static DEFINE_MUTEX(uuid_mutex
);
48 static LIST_HEAD(fs_uuids
);
50 static void lock_chunks(struct btrfs_root
*root
)
52 mutex_lock(&root
->fs_info
->chunk_mutex
);
55 static void unlock_chunks(struct btrfs_root
*root
)
57 mutex_unlock(&root
->fs_info
->chunk_mutex
);
60 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
62 struct btrfs_device
*device
;
63 WARN_ON(fs_devices
->opened
);
64 while (!list_empty(&fs_devices
->devices
)) {
65 device
= list_entry(fs_devices
->devices
.next
,
66 struct btrfs_device
, dev_list
);
67 list_del(&device
->dev_list
);
68 rcu_string_free(device
->name
);
74 void btrfs_cleanup_fs_uuids(void)
76 struct btrfs_fs_devices
*fs_devices
;
78 while (!list_empty(&fs_uuids
)) {
79 fs_devices
= list_entry(fs_uuids
.next
,
80 struct btrfs_fs_devices
, list
);
81 list_del(&fs_devices
->list
);
82 free_fs_devices(fs_devices
);
86 static noinline
struct btrfs_device
*__find_device(struct list_head
*head
,
89 struct btrfs_device
*dev
;
91 list_for_each_entry(dev
, head
, dev_list
) {
92 if (dev
->devid
== devid
&&
93 (!uuid
|| !memcmp(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
))) {
100 static noinline
struct btrfs_fs_devices
*find_fsid(u8
*fsid
)
102 struct btrfs_fs_devices
*fs_devices
;
104 list_for_each_entry(fs_devices
, &fs_uuids
, list
) {
105 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
111 static void requeue_list(struct btrfs_pending_bios
*pending_bios
,
112 struct bio
*head
, struct bio
*tail
)
115 struct bio
*old_head
;
117 old_head
= pending_bios
->head
;
118 pending_bios
->head
= head
;
119 if (pending_bios
->tail
)
120 tail
->bi_next
= old_head
;
122 pending_bios
->tail
= tail
;
126 * we try to collect pending bios for a device so we don't get a large
127 * number of procs sending bios down to the same device. This greatly
128 * improves the schedulers ability to collect and merge the bios.
130 * But, it also turns into a long list of bios to process and that is sure
131 * to eventually make the worker thread block. The solution here is to
132 * make some progress and then put this work struct back at the end of
133 * the list if the block device is congested. This way, multiple devices
134 * can make progress from a single worker thread.
136 static noinline
void run_scheduled_bios(struct btrfs_device
*device
)
139 struct backing_dev_info
*bdi
;
140 struct btrfs_fs_info
*fs_info
;
141 struct btrfs_pending_bios
*pending_bios
;
145 unsigned long num_run
;
146 unsigned long batch_run
= 0;
148 unsigned long last_waited
= 0;
150 int sync_pending
= 0;
151 struct blk_plug plug
;
154 * this function runs all the bios we've collected for
155 * a particular device. We don't want to wander off to
156 * another device without first sending all of these down.
157 * So, setup a plug here and finish it off before we return
159 blk_start_plug(&plug
);
161 bdi
= blk_get_backing_dev_info(device
->bdev
);
162 fs_info
= device
->dev_root
->fs_info
;
163 limit
= btrfs_async_submit_limit(fs_info
);
164 limit
= limit
* 2 / 3;
167 spin_lock(&device
->io_lock
);
172 /* take all the bios off the list at once and process them
173 * later on (without the lock held). But, remember the
174 * tail and other pointers so the bios can be properly reinserted
175 * into the list if we hit congestion
177 if (!force_reg
&& device
->pending_sync_bios
.head
) {
178 pending_bios
= &device
->pending_sync_bios
;
181 pending_bios
= &device
->pending_bios
;
185 pending
= pending_bios
->head
;
186 tail
= pending_bios
->tail
;
187 WARN_ON(pending
&& !tail
);
190 * if pending was null this time around, no bios need processing
191 * at all and we can stop. Otherwise it'll loop back up again
192 * and do an additional check so no bios are missed.
194 * device->running_pending is used to synchronize with the
197 if (device
->pending_sync_bios
.head
== NULL
&&
198 device
->pending_bios
.head
== NULL
) {
200 device
->running_pending
= 0;
203 device
->running_pending
= 1;
206 pending_bios
->head
= NULL
;
207 pending_bios
->tail
= NULL
;
209 spin_unlock(&device
->io_lock
);
214 /* we want to work on both lists, but do more bios on the
215 * sync list than the regular list
218 pending_bios
!= &device
->pending_sync_bios
&&
219 device
->pending_sync_bios
.head
) ||
220 (num_run
> 64 && pending_bios
== &device
->pending_sync_bios
&&
221 device
->pending_bios
.head
)) {
222 spin_lock(&device
->io_lock
);
223 requeue_list(pending_bios
, pending
, tail
);
228 pending
= pending
->bi_next
;
230 atomic_dec(&fs_info
->nr_async_bios
);
232 if (atomic_read(&fs_info
->nr_async_bios
) < limit
&&
233 waitqueue_active(&fs_info
->async_submit_wait
))
234 wake_up(&fs_info
->async_submit_wait
);
236 BUG_ON(atomic_read(&cur
->bi_cnt
) == 0);
239 * if we're doing the sync list, record that our
240 * plug has some sync requests on it
242 * If we're doing the regular list and there are
243 * sync requests sitting around, unplug before
246 if (pending_bios
== &device
->pending_sync_bios
) {
248 } else if (sync_pending
) {
249 blk_finish_plug(&plug
);
250 blk_start_plug(&plug
);
254 btrfsic_submit_bio(cur
->bi_rw
, cur
);
261 * we made progress, there is more work to do and the bdi
262 * is now congested. Back off and let other work structs
265 if (pending
&& bdi_write_congested(bdi
) && batch_run
> 8 &&
266 fs_info
->fs_devices
->open_devices
> 1) {
267 struct io_context
*ioc
;
269 ioc
= current
->io_context
;
272 * the main goal here is that we don't want to
273 * block if we're going to be able to submit
274 * more requests without blocking.
276 * This code does two great things, it pokes into
277 * the elevator code from a filesystem _and_
278 * it makes assumptions about how batching works.
280 if (ioc
&& ioc
->nr_batch_requests
> 0 &&
281 time_before(jiffies
, ioc
->last_waited
+ HZ
/50UL) &&
283 ioc
->last_waited
== last_waited
)) {
285 * we want to go through our batch of
286 * requests and stop. So, we copy out
287 * the ioc->last_waited time and test
288 * against it before looping
290 last_waited
= ioc
->last_waited
;
295 spin_lock(&device
->io_lock
);
296 requeue_list(pending_bios
, pending
, tail
);
297 device
->running_pending
= 1;
299 spin_unlock(&device
->io_lock
);
300 btrfs_requeue_work(&device
->work
);
303 /* unplug every 64 requests just for good measure */
304 if (batch_run
% 64 == 0) {
305 blk_finish_plug(&plug
);
306 blk_start_plug(&plug
);
315 spin_lock(&device
->io_lock
);
316 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
318 spin_unlock(&device
->io_lock
);
321 blk_finish_plug(&plug
);
324 static void pending_bios_fn(struct btrfs_work
*work
)
326 struct btrfs_device
*device
;
328 device
= container_of(work
, struct btrfs_device
, work
);
329 run_scheduled_bios(device
);
332 static noinline
int device_list_add(const char *path
,
333 struct btrfs_super_block
*disk_super
,
334 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
336 struct btrfs_device
*device
;
337 struct btrfs_fs_devices
*fs_devices
;
338 struct rcu_string
*name
;
339 u64 found_transid
= btrfs_super_generation(disk_super
);
341 fs_devices
= find_fsid(disk_super
->fsid
);
343 fs_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
346 INIT_LIST_HEAD(&fs_devices
->devices
);
347 INIT_LIST_HEAD(&fs_devices
->alloc_list
);
348 list_add(&fs_devices
->list
, &fs_uuids
);
349 memcpy(fs_devices
->fsid
, disk_super
->fsid
, BTRFS_FSID_SIZE
);
350 fs_devices
->latest_devid
= devid
;
351 fs_devices
->latest_trans
= found_transid
;
352 mutex_init(&fs_devices
->device_list_mutex
);
355 device
= __find_device(&fs_devices
->devices
, devid
,
356 disk_super
->dev_item
.uuid
);
359 if (fs_devices
->opened
)
362 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
364 /* we can safely leave the fs_devices entry around */
367 device
->devid
= devid
;
368 device
->dev_stats_valid
= 0;
369 device
->work
.func
= pending_bios_fn
;
370 memcpy(device
->uuid
, disk_super
->dev_item
.uuid
,
372 spin_lock_init(&device
->io_lock
);
374 name
= rcu_string_strdup(path
, GFP_NOFS
);
379 rcu_assign_pointer(device
->name
, name
);
380 INIT_LIST_HEAD(&device
->dev_alloc_list
);
382 /* init readahead state */
383 spin_lock_init(&device
->reada_lock
);
384 device
->reada_curr_zone
= NULL
;
385 atomic_set(&device
->reada_in_flight
, 0);
386 device
->reada_next
= 0;
387 INIT_RADIX_TREE(&device
->reada_zones
, GFP_NOFS
& ~__GFP_WAIT
);
388 INIT_RADIX_TREE(&device
->reada_extents
, GFP_NOFS
& ~__GFP_WAIT
);
390 mutex_lock(&fs_devices
->device_list_mutex
);
391 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
392 mutex_unlock(&fs_devices
->device_list_mutex
);
394 device
->fs_devices
= fs_devices
;
395 fs_devices
->num_devices
++;
396 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
397 name
= rcu_string_strdup(path
, GFP_NOFS
);
400 rcu_string_free(device
->name
);
401 rcu_assign_pointer(device
->name
, name
);
402 if (device
->missing
) {
403 fs_devices
->missing_devices
--;
408 if (found_transid
> fs_devices
->latest_trans
) {
409 fs_devices
->latest_devid
= devid
;
410 fs_devices
->latest_trans
= found_transid
;
412 *fs_devices_ret
= fs_devices
;
416 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
418 struct btrfs_fs_devices
*fs_devices
;
419 struct btrfs_device
*device
;
420 struct btrfs_device
*orig_dev
;
422 fs_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
424 return ERR_PTR(-ENOMEM
);
426 INIT_LIST_HEAD(&fs_devices
->devices
);
427 INIT_LIST_HEAD(&fs_devices
->alloc_list
);
428 INIT_LIST_HEAD(&fs_devices
->list
);
429 mutex_init(&fs_devices
->device_list_mutex
);
430 fs_devices
->latest_devid
= orig
->latest_devid
;
431 fs_devices
->latest_trans
= orig
->latest_trans
;
432 fs_devices
->total_devices
= orig
->total_devices
;
433 memcpy(fs_devices
->fsid
, orig
->fsid
, sizeof(fs_devices
->fsid
));
435 /* We have held the volume lock, it is safe to get the devices. */
436 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
437 struct rcu_string
*name
;
439 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
444 * This is ok to do without rcu read locked because we hold the
445 * uuid mutex so nothing we touch in here is going to disappear.
447 name
= rcu_string_strdup(orig_dev
->name
->str
, GFP_NOFS
);
452 rcu_assign_pointer(device
->name
, name
);
454 device
->devid
= orig_dev
->devid
;
455 device
->work
.func
= pending_bios_fn
;
456 memcpy(device
->uuid
, orig_dev
->uuid
, sizeof(device
->uuid
));
457 spin_lock_init(&device
->io_lock
);
458 INIT_LIST_HEAD(&device
->dev_list
);
459 INIT_LIST_HEAD(&device
->dev_alloc_list
);
461 list_add(&device
->dev_list
, &fs_devices
->devices
);
462 device
->fs_devices
= fs_devices
;
463 fs_devices
->num_devices
++;
467 free_fs_devices(fs_devices
);
468 return ERR_PTR(-ENOMEM
);
471 void btrfs_close_extra_devices(struct btrfs_fs_devices
*fs_devices
)
473 struct btrfs_device
*device
, *next
;
475 struct block_device
*latest_bdev
= NULL
;
476 u64 latest_devid
= 0;
477 u64 latest_transid
= 0;
479 mutex_lock(&uuid_mutex
);
481 /* This is the initialized path, it is safe to release the devices. */
482 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
483 if (device
->in_fs_metadata
) {
484 if (!latest_transid
||
485 device
->generation
> latest_transid
) {
486 latest_devid
= device
->devid
;
487 latest_transid
= device
->generation
;
488 latest_bdev
= device
->bdev
;
494 blkdev_put(device
->bdev
, device
->mode
);
496 fs_devices
->open_devices
--;
498 if (device
->writeable
) {
499 list_del_init(&device
->dev_alloc_list
);
500 device
->writeable
= 0;
501 fs_devices
->rw_devices
--;
503 list_del_init(&device
->dev_list
);
504 fs_devices
->num_devices
--;
505 rcu_string_free(device
->name
);
509 if (fs_devices
->seed
) {
510 fs_devices
= fs_devices
->seed
;
514 fs_devices
->latest_bdev
= latest_bdev
;
515 fs_devices
->latest_devid
= latest_devid
;
516 fs_devices
->latest_trans
= latest_transid
;
518 mutex_unlock(&uuid_mutex
);
521 static void __free_device(struct work_struct
*work
)
523 struct btrfs_device
*device
;
525 device
= container_of(work
, struct btrfs_device
, rcu_work
);
528 blkdev_put(device
->bdev
, device
->mode
);
530 rcu_string_free(device
->name
);
534 static void free_device(struct rcu_head
*head
)
536 struct btrfs_device
*device
;
538 device
= container_of(head
, struct btrfs_device
, rcu
);
540 INIT_WORK(&device
->rcu_work
, __free_device
);
541 schedule_work(&device
->rcu_work
);
544 static int __btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
546 struct btrfs_device
*device
;
548 if (--fs_devices
->opened
> 0)
551 mutex_lock(&fs_devices
->device_list_mutex
);
552 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
553 struct btrfs_device
*new_device
;
554 struct rcu_string
*name
;
557 fs_devices
->open_devices
--;
559 if (device
->writeable
) {
560 list_del_init(&device
->dev_alloc_list
);
561 fs_devices
->rw_devices
--;
564 if (device
->can_discard
)
565 fs_devices
->num_can_discard
--;
567 new_device
= kmalloc(sizeof(*new_device
), GFP_NOFS
);
568 BUG_ON(!new_device
); /* -ENOMEM */
569 memcpy(new_device
, device
, sizeof(*new_device
));
571 /* Safe because we are under uuid_mutex */
572 name
= rcu_string_strdup(device
->name
->str
, GFP_NOFS
);
573 BUG_ON(device
->name
&& !name
); /* -ENOMEM */
574 rcu_assign_pointer(new_device
->name
, name
);
575 new_device
->bdev
= NULL
;
576 new_device
->writeable
= 0;
577 new_device
->in_fs_metadata
= 0;
578 new_device
->can_discard
= 0;
579 list_replace_rcu(&device
->dev_list
, &new_device
->dev_list
);
581 call_rcu(&device
->rcu
, free_device
);
583 mutex_unlock(&fs_devices
->device_list_mutex
);
585 WARN_ON(fs_devices
->open_devices
);
586 WARN_ON(fs_devices
->rw_devices
);
587 fs_devices
->opened
= 0;
588 fs_devices
->seeding
= 0;
593 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
595 struct btrfs_fs_devices
*seed_devices
= NULL
;
598 mutex_lock(&uuid_mutex
);
599 ret
= __btrfs_close_devices(fs_devices
);
600 if (!fs_devices
->opened
) {
601 seed_devices
= fs_devices
->seed
;
602 fs_devices
->seed
= NULL
;
604 mutex_unlock(&uuid_mutex
);
606 while (seed_devices
) {
607 fs_devices
= seed_devices
;
608 seed_devices
= fs_devices
->seed
;
609 __btrfs_close_devices(fs_devices
);
610 free_fs_devices(fs_devices
);
615 static int __btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
616 fmode_t flags
, void *holder
)
618 struct request_queue
*q
;
619 struct block_device
*bdev
;
620 struct list_head
*head
= &fs_devices
->devices
;
621 struct btrfs_device
*device
;
622 struct block_device
*latest_bdev
= NULL
;
623 struct buffer_head
*bh
;
624 struct btrfs_super_block
*disk_super
;
625 u64 latest_devid
= 0;
626 u64 latest_transid
= 0;
633 list_for_each_entry(device
, head
, dev_list
) {
639 bdev
= blkdev_get_by_path(device
->name
->str
, flags
, holder
);
641 printk(KERN_INFO
"open %s failed\n", device
->name
->str
);
644 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
645 invalidate_bdev(bdev
);
646 set_blocksize(bdev
, 4096);
648 bh
= btrfs_read_dev_super(bdev
);
652 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
653 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
654 if (devid
!= device
->devid
)
657 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
,
661 device
->generation
= btrfs_super_generation(disk_super
);
662 if (!latest_transid
|| device
->generation
> latest_transid
) {
663 latest_devid
= devid
;
664 latest_transid
= device
->generation
;
668 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
669 device
->writeable
= 0;
671 device
->writeable
= !bdev_read_only(bdev
);
675 q
= bdev_get_queue(bdev
);
676 if (blk_queue_discard(q
)) {
677 device
->can_discard
= 1;
678 fs_devices
->num_can_discard
++;
682 device
->in_fs_metadata
= 0;
683 device
->mode
= flags
;
685 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
686 fs_devices
->rotating
= 1;
688 fs_devices
->open_devices
++;
689 if (device
->writeable
) {
690 fs_devices
->rw_devices
++;
691 list_add(&device
->dev_alloc_list
,
692 &fs_devices
->alloc_list
);
700 blkdev_put(bdev
, flags
);
704 if (fs_devices
->open_devices
== 0) {
708 fs_devices
->seeding
= seeding
;
709 fs_devices
->opened
= 1;
710 fs_devices
->latest_bdev
= latest_bdev
;
711 fs_devices
->latest_devid
= latest_devid
;
712 fs_devices
->latest_trans
= latest_transid
;
713 fs_devices
->total_rw_bytes
= 0;
718 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
719 fmode_t flags
, void *holder
)
723 mutex_lock(&uuid_mutex
);
724 if (fs_devices
->opened
) {
725 fs_devices
->opened
++;
728 ret
= __btrfs_open_devices(fs_devices
, flags
, holder
);
730 mutex_unlock(&uuid_mutex
);
734 int btrfs_scan_one_device(const char *path
, fmode_t flags
, void *holder
,
735 struct btrfs_fs_devices
**fs_devices_ret
)
737 struct btrfs_super_block
*disk_super
;
738 struct block_device
*bdev
;
739 struct buffer_head
*bh
;
746 bdev
= blkdev_get_by_path(path
, flags
, holder
);
753 mutex_lock(&uuid_mutex
);
754 ret
= set_blocksize(bdev
, 4096);
757 bh
= btrfs_read_dev_super(bdev
);
762 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
763 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
764 transid
= btrfs_super_generation(disk_super
);
765 total_devices
= btrfs_super_num_devices(disk_super
);
766 if (disk_super
->label
[0])
767 printk(KERN_INFO
"device label %s ", disk_super
->label
);
769 printk(KERN_INFO
"device fsid %pU ", disk_super
->fsid
);
770 printk(KERN_CONT
"devid %llu transid %llu %s\n",
771 (unsigned long long)devid
, (unsigned long long)transid
, path
);
772 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
773 if (!ret
&& fs_devices_ret
)
774 (*fs_devices_ret
)->total_devices
= total_devices
;
777 mutex_unlock(&uuid_mutex
);
778 blkdev_put(bdev
, flags
);
783 /* helper to account the used device space in the range */
784 int btrfs_account_dev_extents_size(struct btrfs_device
*device
, u64 start
,
785 u64 end
, u64
*length
)
787 struct btrfs_key key
;
788 struct btrfs_root
*root
= device
->dev_root
;
789 struct btrfs_dev_extent
*dev_extent
;
790 struct btrfs_path
*path
;
794 struct extent_buffer
*l
;
798 if (start
>= device
->total_bytes
)
801 path
= btrfs_alloc_path();
806 key
.objectid
= device
->devid
;
808 key
.type
= BTRFS_DEV_EXTENT_KEY
;
810 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
814 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
821 slot
= path
->slots
[0];
822 if (slot
>= btrfs_header_nritems(l
)) {
823 ret
= btrfs_next_leaf(root
, path
);
831 btrfs_item_key_to_cpu(l
, &key
, slot
);
833 if (key
.objectid
< device
->devid
)
836 if (key
.objectid
> device
->devid
)
839 if (btrfs_key_type(&key
) != BTRFS_DEV_EXTENT_KEY
)
842 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
843 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
845 if (key
.offset
<= start
&& extent_end
> end
) {
846 *length
= end
- start
+ 1;
848 } else if (key
.offset
<= start
&& extent_end
> start
)
849 *length
+= extent_end
- start
;
850 else if (key
.offset
> start
&& extent_end
<= end
)
851 *length
+= extent_end
- key
.offset
;
852 else if (key
.offset
> start
&& key
.offset
<= end
) {
853 *length
+= end
- key
.offset
+ 1;
855 } else if (key
.offset
> end
)
863 btrfs_free_path(path
);
868 * find_free_dev_extent - find free space in the specified device
869 * @device: the device which we search the free space in
870 * @num_bytes: the size of the free space that we need
871 * @start: store the start of the free space.
872 * @len: the size of the free space. that we find, or the size of the max
873 * free space if we don't find suitable free space
875 * this uses a pretty simple search, the expectation is that it is
876 * called very infrequently and that a given device has a small number
879 * @start is used to store the start of the free space if we find. But if we
880 * don't find suitable free space, it will be used to store the start position
881 * of the max free space.
883 * @len is used to store the size of the free space that we find.
884 * But if we don't find suitable free space, it is used to store the size of
885 * the max free space.
887 int find_free_dev_extent(struct btrfs_device
*device
, u64 num_bytes
,
888 u64
*start
, u64
*len
)
890 struct btrfs_key key
;
891 struct btrfs_root
*root
= device
->dev_root
;
892 struct btrfs_dev_extent
*dev_extent
;
893 struct btrfs_path
*path
;
899 u64 search_end
= device
->total_bytes
;
902 struct extent_buffer
*l
;
904 /* FIXME use last free of some kind */
906 /* we don't want to overwrite the superblock on the drive,
907 * so we make sure to start at an offset of at least 1MB
909 search_start
= max(root
->fs_info
->alloc_start
, 1024ull * 1024);
911 max_hole_start
= search_start
;
915 if (search_start
>= search_end
) {
920 path
= btrfs_alloc_path();
927 key
.objectid
= device
->devid
;
928 key
.offset
= search_start
;
929 key
.type
= BTRFS_DEV_EXTENT_KEY
;
931 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
935 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
942 slot
= path
->slots
[0];
943 if (slot
>= btrfs_header_nritems(l
)) {
944 ret
= btrfs_next_leaf(root
, path
);
952 btrfs_item_key_to_cpu(l
, &key
, slot
);
954 if (key
.objectid
< device
->devid
)
957 if (key
.objectid
> device
->devid
)
960 if (btrfs_key_type(&key
) != BTRFS_DEV_EXTENT_KEY
)
963 if (key
.offset
> search_start
) {
964 hole_size
= key
.offset
- search_start
;
966 if (hole_size
> max_hole_size
) {
967 max_hole_start
= search_start
;
968 max_hole_size
= hole_size
;
972 * If this free space is greater than which we need,
973 * it must be the max free space that we have found
974 * until now, so max_hole_start must point to the start
975 * of this free space and the length of this free space
976 * is stored in max_hole_size. Thus, we return
977 * max_hole_start and max_hole_size and go back to the
980 if (hole_size
>= num_bytes
) {
986 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
987 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
989 if (extent_end
> search_start
)
990 search_start
= extent_end
;
997 * At this point, search_start should be the end of
998 * allocated dev extents, and when shrinking the device,
999 * search_end may be smaller than search_start.
1001 if (search_end
> search_start
)
1002 hole_size
= search_end
- search_start
;
1004 if (hole_size
> max_hole_size
) {
1005 max_hole_start
= search_start
;
1006 max_hole_size
= hole_size
;
1010 if (hole_size
< num_bytes
)
1016 btrfs_free_path(path
);
1018 *start
= max_hole_start
;
1020 *len
= max_hole_size
;
1024 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1025 struct btrfs_device
*device
,
1029 struct btrfs_path
*path
;
1030 struct btrfs_root
*root
= device
->dev_root
;
1031 struct btrfs_key key
;
1032 struct btrfs_key found_key
;
1033 struct extent_buffer
*leaf
= NULL
;
1034 struct btrfs_dev_extent
*extent
= NULL
;
1036 path
= btrfs_alloc_path();
1040 key
.objectid
= device
->devid
;
1042 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1044 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1046 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1047 BTRFS_DEV_EXTENT_KEY
);
1050 leaf
= path
->nodes
[0];
1051 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1052 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1053 struct btrfs_dev_extent
);
1054 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1055 btrfs_dev_extent_length(leaf
, extent
) < start
);
1057 btrfs_release_path(path
);
1059 } else if (ret
== 0) {
1060 leaf
= path
->nodes
[0];
1061 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1062 struct btrfs_dev_extent
);
1064 btrfs_error(root
->fs_info
, ret
, "Slot search failed");
1068 if (device
->bytes_used
> 0) {
1069 u64 len
= btrfs_dev_extent_length(leaf
, extent
);
1070 device
->bytes_used
-= len
;
1071 spin_lock(&root
->fs_info
->free_chunk_lock
);
1072 root
->fs_info
->free_chunk_space
+= len
;
1073 spin_unlock(&root
->fs_info
->free_chunk_lock
);
1075 ret
= btrfs_del_item(trans
, root
, path
);
1077 btrfs_error(root
->fs_info
, ret
,
1078 "Failed to remove dev extent item");
1081 btrfs_free_path(path
);
1085 int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1086 struct btrfs_device
*device
,
1087 u64 chunk_tree
, u64 chunk_objectid
,
1088 u64 chunk_offset
, u64 start
, u64 num_bytes
)
1091 struct btrfs_path
*path
;
1092 struct btrfs_root
*root
= device
->dev_root
;
1093 struct btrfs_dev_extent
*extent
;
1094 struct extent_buffer
*leaf
;
1095 struct btrfs_key key
;
1097 WARN_ON(!device
->in_fs_metadata
);
1098 path
= btrfs_alloc_path();
1102 key
.objectid
= device
->devid
;
1104 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1105 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1110 leaf
= path
->nodes
[0];
1111 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1112 struct btrfs_dev_extent
);
1113 btrfs_set_dev_extent_chunk_tree(leaf
, extent
, chunk_tree
);
1114 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
, chunk_objectid
);
1115 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1117 write_extent_buffer(leaf
, root
->fs_info
->chunk_tree_uuid
,
1118 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent
),
1121 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1122 btrfs_mark_buffer_dirty(leaf
);
1124 btrfs_free_path(path
);
1128 static noinline
int find_next_chunk(struct btrfs_root
*root
,
1129 u64 objectid
, u64
*offset
)
1131 struct btrfs_path
*path
;
1133 struct btrfs_key key
;
1134 struct btrfs_chunk
*chunk
;
1135 struct btrfs_key found_key
;
1137 path
= btrfs_alloc_path();
1141 key
.objectid
= objectid
;
1142 key
.offset
= (u64
)-1;
1143 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
1145 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1149 BUG_ON(ret
== 0); /* Corruption */
1151 ret
= btrfs_previous_item(root
, path
, 0, BTRFS_CHUNK_ITEM_KEY
);
1155 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1157 if (found_key
.objectid
!= objectid
)
1160 chunk
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1161 struct btrfs_chunk
);
1162 *offset
= found_key
.offset
+
1163 btrfs_chunk_length(path
->nodes
[0], chunk
);
1168 btrfs_free_path(path
);
1172 static noinline
int find_next_devid(struct btrfs_root
*root
, u64
*objectid
)
1175 struct btrfs_key key
;
1176 struct btrfs_key found_key
;
1177 struct btrfs_path
*path
;
1179 root
= root
->fs_info
->chunk_root
;
1181 path
= btrfs_alloc_path();
1185 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1186 key
.type
= BTRFS_DEV_ITEM_KEY
;
1187 key
.offset
= (u64
)-1;
1189 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1193 BUG_ON(ret
== 0); /* Corruption */
1195 ret
= btrfs_previous_item(root
, path
, BTRFS_DEV_ITEMS_OBJECTID
,
1196 BTRFS_DEV_ITEM_KEY
);
1200 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1202 *objectid
= found_key
.offset
+ 1;
1206 btrfs_free_path(path
);
1211 * the device information is stored in the chunk root
1212 * the btrfs_device struct should be fully filled in
1214 int btrfs_add_device(struct btrfs_trans_handle
*trans
,
1215 struct btrfs_root
*root
,
1216 struct btrfs_device
*device
)
1219 struct btrfs_path
*path
;
1220 struct btrfs_dev_item
*dev_item
;
1221 struct extent_buffer
*leaf
;
1222 struct btrfs_key key
;
1225 root
= root
->fs_info
->chunk_root
;
1227 path
= btrfs_alloc_path();
1231 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1232 key
.type
= BTRFS_DEV_ITEM_KEY
;
1233 key
.offset
= device
->devid
;
1235 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1240 leaf
= path
->nodes
[0];
1241 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1243 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1244 btrfs_set_device_generation(leaf
, dev_item
, 0);
1245 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1246 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1247 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1248 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1249 btrfs_set_device_total_bytes(leaf
, dev_item
, device
->total_bytes
);
1250 btrfs_set_device_bytes_used(leaf
, dev_item
, device
->bytes_used
);
1251 btrfs_set_device_group(leaf
, dev_item
, 0);
1252 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1253 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1254 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1256 ptr
= (unsigned long)btrfs_device_uuid(dev_item
);
1257 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1258 ptr
= (unsigned long)btrfs_device_fsid(dev_item
);
1259 write_extent_buffer(leaf
, root
->fs_info
->fsid
, ptr
, BTRFS_UUID_SIZE
);
1260 btrfs_mark_buffer_dirty(leaf
);
1264 btrfs_free_path(path
);
1268 static int btrfs_rm_dev_item(struct btrfs_root
*root
,
1269 struct btrfs_device
*device
)
1272 struct btrfs_path
*path
;
1273 struct btrfs_key key
;
1274 struct btrfs_trans_handle
*trans
;
1276 root
= root
->fs_info
->chunk_root
;
1278 path
= btrfs_alloc_path();
1282 trans
= btrfs_start_transaction(root
, 0);
1283 if (IS_ERR(trans
)) {
1284 btrfs_free_path(path
);
1285 return PTR_ERR(trans
);
1287 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1288 key
.type
= BTRFS_DEV_ITEM_KEY
;
1289 key
.offset
= device
->devid
;
1292 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1301 ret
= btrfs_del_item(trans
, root
, path
);
1305 btrfs_free_path(path
);
1306 unlock_chunks(root
);
1307 btrfs_commit_transaction(trans
, root
);
1311 int btrfs_rm_device(struct btrfs_root
*root
, char *device_path
)
1313 struct btrfs_device
*device
;
1314 struct btrfs_device
*next_device
;
1315 struct block_device
*bdev
;
1316 struct buffer_head
*bh
= NULL
;
1317 struct btrfs_super_block
*disk_super
;
1318 struct btrfs_fs_devices
*cur_devices
;
1324 bool clear_super
= false;
1326 mutex_lock(&uuid_mutex
);
1328 all_avail
= root
->fs_info
->avail_data_alloc_bits
|
1329 root
->fs_info
->avail_system_alloc_bits
|
1330 root
->fs_info
->avail_metadata_alloc_bits
;
1332 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID10
) &&
1333 root
->fs_info
->fs_devices
->num_devices
<= 4) {
1334 printk(KERN_ERR
"btrfs: unable to go below four devices "
1340 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID1
) &&
1341 root
->fs_info
->fs_devices
->num_devices
<= 2) {
1342 printk(KERN_ERR
"btrfs: unable to go below two "
1343 "devices on raid1\n");
1348 if (strcmp(device_path
, "missing") == 0) {
1349 struct list_head
*devices
;
1350 struct btrfs_device
*tmp
;
1353 devices
= &root
->fs_info
->fs_devices
->devices
;
1355 * It is safe to read the devices since the volume_mutex
1358 list_for_each_entry(tmp
, devices
, dev_list
) {
1359 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
1368 printk(KERN_ERR
"btrfs: no missing devices found to "
1373 bdev
= blkdev_get_by_path(device_path
, FMODE_READ
| FMODE_EXCL
,
1374 root
->fs_info
->bdev_holder
);
1376 ret
= PTR_ERR(bdev
);
1380 set_blocksize(bdev
, 4096);
1381 invalidate_bdev(bdev
);
1382 bh
= btrfs_read_dev_super(bdev
);
1387 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1388 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1389 dev_uuid
= disk_super
->dev_item
.uuid
;
1390 device
= btrfs_find_device(root
, devid
, dev_uuid
,
1398 if (device
->writeable
&& root
->fs_info
->fs_devices
->rw_devices
== 1) {
1399 printk(KERN_ERR
"btrfs: unable to remove the only writeable "
1405 if (device
->writeable
) {
1407 list_del_init(&device
->dev_alloc_list
);
1408 unlock_chunks(root
);
1409 root
->fs_info
->fs_devices
->rw_devices
--;
1413 ret
= btrfs_shrink_device(device
, 0);
1417 ret
= btrfs_rm_dev_item(root
->fs_info
->chunk_root
, device
);
1421 spin_lock(&root
->fs_info
->free_chunk_lock
);
1422 root
->fs_info
->free_chunk_space
= device
->total_bytes
-
1424 spin_unlock(&root
->fs_info
->free_chunk_lock
);
1426 device
->in_fs_metadata
= 0;
1427 btrfs_scrub_cancel_dev(root
, device
);
1430 * the device list mutex makes sure that we don't change
1431 * the device list while someone else is writing out all
1432 * the device supers.
1435 cur_devices
= device
->fs_devices
;
1436 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1437 list_del_rcu(&device
->dev_list
);
1439 device
->fs_devices
->num_devices
--;
1440 device
->fs_devices
->total_devices
--;
1442 if (device
->missing
)
1443 root
->fs_info
->fs_devices
->missing_devices
--;
1445 next_device
= list_entry(root
->fs_info
->fs_devices
->devices
.next
,
1446 struct btrfs_device
, dev_list
);
1447 if (device
->bdev
== root
->fs_info
->sb
->s_bdev
)
1448 root
->fs_info
->sb
->s_bdev
= next_device
->bdev
;
1449 if (device
->bdev
== root
->fs_info
->fs_devices
->latest_bdev
)
1450 root
->fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1453 device
->fs_devices
->open_devices
--;
1455 call_rcu(&device
->rcu
, free_device
);
1456 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1458 num_devices
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
1459 btrfs_set_super_num_devices(root
->fs_info
->super_copy
, num_devices
);
1461 if (cur_devices
->open_devices
== 0) {
1462 struct btrfs_fs_devices
*fs_devices
;
1463 fs_devices
= root
->fs_info
->fs_devices
;
1464 while (fs_devices
) {
1465 if (fs_devices
->seed
== cur_devices
)
1467 fs_devices
= fs_devices
->seed
;
1469 fs_devices
->seed
= cur_devices
->seed
;
1470 cur_devices
->seed
= NULL
;
1472 __btrfs_close_devices(cur_devices
);
1473 unlock_chunks(root
);
1474 free_fs_devices(cur_devices
);
1478 * at this point, the device is zero sized. We want to
1479 * remove it from the devices list and zero out the old super
1482 /* make sure this device isn't detected as part of
1485 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
1486 set_buffer_dirty(bh
);
1487 sync_dirty_buffer(bh
);
1496 blkdev_put(bdev
, FMODE_READ
| FMODE_EXCL
);
1498 mutex_unlock(&uuid_mutex
);
1501 if (device
->writeable
) {
1503 list_add(&device
->dev_alloc_list
,
1504 &root
->fs_info
->fs_devices
->alloc_list
);
1505 unlock_chunks(root
);
1506 root
->fs_info
->fs_devices
->rw_devices
++;
1512 * does all the dirty work required for changing file system's UUID.
1514 static int btrfs_prepare_sprout(struct btrfs_root
*root
)
1516 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
1517 struct btrfs_fs_devices
*old_devices
;
1518 struct btrfs_fs_devices
*seed_devices
;
1519 struct btrfs_super_block
*disk_super
= root
->fs_info
->super_copy
;
1520 struct btrfs_device
*device
;
1523 BUG_ON(!mutex_is_locked(&uuid_mutex
));
1524 if (!fs_devices
->seeding
)
1527 seed_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
1531 old_devices
= clone_fs_devices(fs_devices
);
1532 if (IS_ERR(old_devices
)) {
1533 kfree(seed_devices
);
1534 return PTR_ERR(old_devices
);
1537 list_add(&old_devices
->list
, &fs_uuids
);
1539 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
1540 seed_devices
->opened
= 1;
1541 INIT_LIST_HEAD(&seed_devices
->devices
);
1542 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
1543 mutex_init(&seed_devices
->device_list_mutex
);
1545 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1546 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
1548 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1550 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
1551 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
) {
1552 device
->fs_devices
= seed_devices
;
1555 fs_devices
->seeding
= 0;
1556 fs_devices
->num_devices
= 0;
1557 fs_devices
->open_devices
= 0;
1558 fs_devices
->total_devices
= 0;
1559 fs_devices
->seed
= seed_devices
;
1561 generate_random_uuid(fs_devices
->fsid
);
1562 memcpy(root
->fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
1563 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
1564 super_flags
= btrfs_super_flags(disk_super
) &
1565 ~BTRFS_SUPER_FLAG_SEEDING
;
1566 btrfs_set_super_flags(disk_super
, super_flags
);
1572 * strore the expected generation for seed devices in device items.
1574 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
1575 struct btrfs_root
*root
)
1577 struct btrfs_path
*path
;
1578 struct extent_buffer
*leaf
;
1579 struct btrfs_dev_item
*dev_item
;
1580 struct btrfs_device
*device
;
1581 struct btrfs_key key
;
1582 u8 fs_uuid
[BTRFS_UUID_SIZE
];
1583 u8 dev_uuid
[BTRFS_UUID_SIZE
];
1587 path
= btrfs_alloc_path();
1591 root
= root
->fs_info
->chunk_root
;
1592 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1594 key
.type
= BTRFS_DEV_ITEM_KEY
;
1597 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
1601 leaf
= path
->nodes
[0];
1603 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1604 ret
= btrfs_next_leaf(root
, path
);
1609 leaf
= path
->nodes
[0];
1610 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1611 btrfs_release_path(path
);
1615 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1616 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
1617 key
.type
!= BTRFS_DEV_ITEM_KEY
)
1620 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1621 struct btrfs_dev_item
);
1622 devid
= btrfs_device_id(leaf
, dev_item
);
1623 read_extent_buffer(leaf
, dev_uuid
,
1624 (unsigned long)btrfs_device_uuid(dev_item
),
1626 read_extent_buffer(leaf
, fs_uuid
,
1627 (unsigned long)btrfs_device_fsid(dev_item
),
1629 device
= btrfs_find_device(root
, devid
, dev_uuid
, fs_uuid
);
1630 BUG_ON(!device
); /* Logic error */
1632 if (device
->fs_devices
->seeding
) {
1633 btrfs_set_device_generation(leaf
, dev_item
,
1634 device
->generation
);
1635 btrfs_mark_buffer_dirty(leaf
);
1643 btrfs_free_path(path
);
1647 int btrfs_init_new_device(struct btrfs_root
*root
, char *device_path
)
1649 struct request_queue
*q
;
1650 struct btrfs_trans_handle
*trans
;
1651 struct btrfs_device
*device
;
1652 struct block_device
*bdev
;
1653 struct list_head
*devices
;
1654 struct super_block
*sb
= root
->fs_info
->sb
;
1655 struct rcu_string
*name
;
1657 int seeding_dev
= 0;
1660 if ((sb
->s_flags
& MS_RDONLY
) && !root
->fs_info
->fs_devices
->seeding
)
1663 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
1664 root
->fs_info
->bdev_holder
);
1666 return PTR_ERR(bdev
);
1668 if (root
->fs_info
->fs_devices
->seeding
) {
1670 down_write(&sb
->s_umount
);
1671 mutex_lock(&uuid_mutex
);
1674 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
1676 devices
= &root
->fs_info
->fs_devices
->devices
;
1678 * we have the volume lock, so we don't need the extra
1679 * device list mutex while reading the list here.
1681 list_for_each_entry(device
, devices
, dev_list
) {
1682 if (device
->bdev
== bdev
) {
1688 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
1690 /* we can safely leave the fs_devices entry around */
1695 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
1701 rcu_assign_pointer(device
->name
, name
);
1703 ret
= find_next_devid(root
, &device
->devid
);
1705 rcu_string_free(device
->name
);
1710 trans
= btrfs_start_transaction(root
, 0);
1711 if (IS_ERR(trans
)) {
1712 rcu_string_free(device
->name
);
1714 ret
= PTR_ERR(trans
);
1720 q
= bdev_get_queue(bdev
);
1721 if (blk_queue_discard(q
))
1722 device
->can_discard
= 1;
1723 device
->writeable
= 1;
1724 device
->work
.func
= pending_bios_fn
;
1725 generate_random_uuid(device
->uuid
);
1726 spin_lock_init(&device
->io_lock
);
1727 device
->generation
= trans
->transid
;
1728 device
->io_width
= root
->sectorsize
;
1729 device
->io_align
= root
->sectorsize
;
1730 device
->sector_size
= root
->sectorsize
;
1731 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
1732 device
->disk_total_bytes
= device
->total_bytes
;
1733 device
->dev_root
= root
->fs_info
->dev_root
;
1734 device
->bdev
= bdev
;
1735 device
->in_fs_metadata
= 1;
1736 device
->mode
= FMODE_EXCL
;
1737 set_blocksize(device
->bdev
, 4096);
1740 sb
->s_flags
&= ~MS_RDONLY
;
1741 ret
= btrfs_prepare_sprout(root
);
1742 BUG_ON(ret
); /* -ENOMEM */
1745 device
->fs_devices
= root
->fs_info
->fs_devices
;
1748 * we don't want write_supers to jump in here with our device
1751 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1752 list_add_rcu(&device
->dev_list
, &root
->fs_info
->fs_devices
->devices
);
1753 list_add(&device
->dev_alloc_list
,
1754 &root
->fs_info
->fs_devices
->alloc_list
);
1755 root
->fs_info
->fs_devices
->num_devices
++;
1756 root
->fs_info
->fs_devices
->open_devices
++;
1757 root
->fs_info
->fs_devices
->rw_devices
++;
1758 root
->fs_info
->fs_devices
->total_devices
++;
1759 if (device
->can_discard
)
1760 root
->fs_info
->fs_devices
->num_can_discard
++;
1761 root
->fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
1763 spin_lock(&root
->fs_info
->free_chunk_lock
);
1764 root
->fs_info
->free_chunk_space
+= device
->total_bytes
;
1765 spin_unlock(&root
->fs_info
->free_chunk_lock
);
1767 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
1768 root
->fs_info
->fs_devices
->rotating
= 1;
1770 total_bytes
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
1771 btrfs_set_super_total_bytes(root
->fs_info
->super_copy
,
1772 total_bytes
+ device
->total_bytes
);
1774 total_bytes
= btrfs_super_num_devices(root
->fs_info
->super_copy
);
1775 btrfs_set_super_num_devices(root
->fs_info
->super_copy
,
1777 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1780 ret
= init_first_rw_device(trans
, root
, device
);
1783 ret
= btrfs_finish_sprout(trans
, root
);
1787 ret
= btrfs_add_device(trans
, root
, device
);
1793 * we've got more storage, clear any full flags on the space
1796 btrfs_clear_space_info_full(root
->fs_info
);
1798 unlock_chunks(root
);
1799 ret
= btrfs_commit_transaction(trans
, root
);
1802 mutex_unlock(&uuid_mutex
);
1803 up_write(&sb
->s_umount
);
1805 if (ret
) /* transaction commit */
1808 ret
= btrfs_relocate_sys_chunks(root
);
1810 btrfs_error(root
->fs_info
, ret
,
1811 "Failed to relocate sys chunks after "
1812 "device initialization. This can be fixed "
1813 "using the \"btrfs balance\" command.");
1819 unlock_chunks(root
);
1820 btrfs_abort_transaction(trans
, root
, ret
);
1821 btrfs_end_transaction(trans
, root
);
1822 rcu_string_free(device
->name
);
1825 blkdev_put(bdev
, FMODE_EXCL
);
1827 mutex_unlock(&uuid_mutex
);
1828 up_write(&sb
->s_umount
);
1833 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
1834 struct btrfs_device
*device
)
1837 struct btrfs_path
*path
;
1838 struct btrfs_root
*root
;
1839 struct btrfs_dev_item
*dev_item
;
1840 struct extent_buffer
*leaf
;
1841 struct btrfs_key key
;
1843 root
= device
->dev_root
->fs_info
->chunk_root
;
1845 path
= btrfs_alloc_path();
1849 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1850 key
.type
= BTRFS_DEV_ITEM_KEY
;
1851 key
.offset
= device
->devid
;
1853 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
1862 leaf
= path
->nodes
[0];
1863 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1865 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1866 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1867 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1868 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1869 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1870 btrfs_set_device_total_bytes(leaf
, dev_item
, device
->disk_total_bytes
);
1871 btrfs_set_device_bytes_used(leaf
, dev_item
, device
->bytes_used
);
1872 btrfs_mark_buffer_dirty(leaf
);
1875 btrfs_free_path(path
);
1879 static int __btrfs_grow_device(struct btrfs_trans_handle
*trans
,
1880 struct btrfs_device
*device
, u64 new_size
)
1882 struct btrfs_super_block
*super_copy
=
1883 device
->dev_root
->fs_info
->super_copy
;
1884 u64 old_total
= btrfs_super_total_bytes(super_copy
);
1885 u64 diff
= new_size
- device
->total_bytes
;
1887 if (!device
->writeable
)
1889 if (new_size
<= device
->total_bytes
)
1892 btrfs_set_super_total_bytes(super_copy
, old_total
+ diff
);
1893 device
->fs_devices
->total_rw_bytes
+= diff
;
1895 device
->total_bytes
= new_size
;
1896 device
->disk_total_bytes
= new_size
;
1897 btrfs_clear_space_info_full(device
->dev_root
->fs_info
);
1899 return btrfs_update_device(trans
, device
);
1902 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
1903 struct btrfs_device
*device
, u64 new_size
)
1906 lock_chunks(device
->dev_root
);
1907 ret
= __btrfs_grow_device(trans
, device
, new_size
);
1908 unlock_chunks(device
->dev_root
);
1912 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
1913 struct btrfs_root
*root
,
1914 u64 chunk_tree
, u64 chunk_objectid
,
1918 struct btrfs_path
*path
;
1919 struct btrfs_key key
;
1921 root
= root
->fs_info
->chunk_root
;
1922 path
= btrfs_alloc_path();
1926 key
.objectid
= chunk_objectid
;
1927 key
.offset
= chunk_offset
;
1928 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
1930 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1933 else if (ret
> 0) { /* Logic error or corruption */
1934 btrfs_error(root
->fs_info
, -ENOENT
,
1935 "Failed lookup while freeing chunk.");
1940 ret
= btrfs_del_item(trans
, root
, path
);
1942 btrfs_error(root
->fs_info
, ret
,
1943 "Failed to delete chunk item.");
1945 btrfs_free_path(path
);
1949 static int btrfs_del_sys_chunk(struct btrfs_root
*root
, u64 chunk_objectid
, u64
1952 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
1953 struct btrfs_disk_key
*disk_key
;
1954 struct btrfs_chunk
*chunk
;
1961 struct btrfs_key key
;
1963 array_size
= btrfs_super_sys_array_size(super_copy
);
1965 ptr
= super_copy
->sys_chunk_array
;
1968 while (cur
< array_size
) {
1969 disk_key
= (struct btrfs_disk_key
*)ptr
;
1970 btrfs_disk_key_to_cpu(&key
, disk_key
);
1972 len
= sizeof(*disk_key
);
1974 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
1975 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
1976 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
1977 len
+= btrfs_chunk_item_size(num_stripes
);
1982 if (key
.objectid
== chunk_objectid
&&
1983 key
.offset
== chunk_offset
) {
1984 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
1986 btrfs_set_super_sys_array_size(super_copy
, array_size
);
1995 static int btrfs_relocate_chunk(struct btrfs_root
*root
,
1996 u64 chunk_tree
, u64 chunk_objectid
,
1999 struct extent_map_tree
*em_tree
;
2000 struct btrfs_root
*extent_root
;
2001 struct btrfs_trans_handle
*trans
;
2002 struct extent_map
*em
;
2003 struct map_lookup
*map
;
2007 root
= root
->fs_info
->chunk_root
;
2008 extent_root
= root
->fs_info
->extent_root
;
2009 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
2011 ret
= btrfs_can_relocate(extent_root
, chunk_offset
);
2015 /* step one, relocate all the extents inside this chunk */
2016 ret
= btrfs_relocate_block_group(extent_root
, chunk_offset
);
2020 trans
= btrfs_start_transaction(root
, 0);
2021 BUG_ON(IS_ERR(trans
));
2026 * step two, delete the device extents and the
2027 * chunk tree entries
2029 read_lock(&em_tree
->lock
);
2030 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
2031 read_unlock(&em_tree
->lock
);
2033 BUG_ON(!em
|| em
->start
> chunk_offset
||
2034 em
->start
+ em
->len
< chunk_offset
);
2035 map
= (struct map_lookup
*)em
->bdev
;
2037 for (i
= 0; i
< map
->num_stripes
; i
++) {
2038 ret
= btrfs_free_dev_extent(trans
, map
->stripes
[i
].dev
,
2039 map
->stripes
[i
].physical
);
2042 if (map
->stripes
[i
].dev
) {
2043 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2047 ret
= btrfs_free_chunk(trans
, root
, chunk_tree
, chunk_objectid
,
2052 trace_btrfs_chunk_free(root
, map
, chunk_offset
, em
->len
);
2054 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2055 ret
= btrfs_del_sys_chunk(root
, chunk_objectid
, chunk_offset
);
2059 ret
= btrfs_remove_block_group(trans
, extent_root
, chunk_offset
);
2062 write_lock(&em_tree
->lock
);
2063 remove_extent_mapping(em_tree
, em
);
2064 write_unlock(&em_tree
->lock
);
2069 /* once for the tree */
2070 free_extent_map(em
);
2072 free_extent_map(em
);
2074 unlock_chunks(root
);
2075 btrfs_end_transaction(trans
, root
);
2079 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
)
2081 struct btrfs_root
*chunk_root
= root
->fs_info
->chunk_root
;
2082 struct btrfs_path
*path
;
2083 struct extent_buffer
*leaf
;
2084 struct btrfs_chunk
*chunk
;
2085 struct btrfs_key key
;
2086 struct btrfs_key found_key
;
2087 u64 chunk_tree
= chunk_root
->root_key
.objectid
;
2089 bool retried
= false;
2093 path
= btrfs_alloc_path();
2098 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2099 key
.offset
= (u64
)-1;
2100 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2103 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2106 BUG_ON(ret
== 0); /* Corruption */
2108 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
2115 leaf
= path
->nodes
[0];
2116 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2118 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
2119 struct btrfs_chunk
);
2120 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2121 btrfs_release_path(path
);
2123 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2124 ret
= btrfs_relocate_chunk(chunk_root
, chunk_tree
,
2133 if (found_key
.offset
== 0)
2135 key
.offset
= found_key
.offset
- 1;
2138 if (failed
&& !retried
) {
2142 } else if (failed
&& retried
) {
2147 btrfs_free_path(path
);
2151 static int insert_balance_item(struct btrfs_root
*root
,
2152 struct btrfs_balance_control
*bctl
)
2154 struct btrfs_trans_handle
*trans
;
2155 struct btrfs_balance_item
*item
;
2156 struct btrfs_disk_balance_args disk_bargs
;
2157 struct btrfs_path
*path
;
2158 struct extent_buffer
*leaf
;
2159 struct btrfs_key key
;
2162 path
= btrfs_alloc_path();
2166 trans
= btrfs_start_transaction(root
, 0);
2167 if (IS_ERR(trans
)) {
2168 btrfs_free_path(path
);
2169 return PTR_ERR(trans
);
2172 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2173 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2176 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2181 leaf
= path
->nodes
[0];
2182 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
2184 memset_extent_buffer(leaf
, 0, (unsigned long)item
, sizeof(*item
));
2186 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
2187 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
2188 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
2189 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
2190 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
2191 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
2193 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
2195 btrfs_mark_buffer_dirty(leaf
);
2197 btrfs_free_path(path
);
2198 err
= btrfs_commit_transaction(trans
, root
);
2204 static int del_balance_item(struct btrfs_root
*root
)
2206 struct btrfs_trans_handle
*trans
;
2207 struct btrfs_path
*path
;
2208 struct btrfs_key key
;
2211 path
= btrfs_alloc_path();
2215 trans
= btrfs_start_transaction(root
, 0);
2216 if (IS_ERR(trans
)) {
2217 btrfs_free_path(path
);
2218 return PTR_ERR(trans
);
2221 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2222 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2225 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2233 ret
= btrfs_del_item(trans
, root
, path
);
2235 btrfs_free_path(path
);
2236 err
= btrfs_commit_transaction(trans
, root
);
2243 * This is a heuristic used to reduce the number of chunks balanced on
2244 * resume after balance was interrupted.
2246 static void update_balance_args(struct btrfs_balance_control
*bctl
)
2249 * Turn on soft mode for chunk types that were being converted.
2251 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2252 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2253 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2254 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2255 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2256 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2259 * Turn on usage filter if is not already used. The idea is
2260 * that chunks that we have already balanced should be
2261 * reasonably full. Don't do it for chunks that are being
2262 * converted - that will keep us from relocating unconverted
2263 * (albeit full) chunks.
2265 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2266 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2267 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2268 bctl
->data
.usage
= 90;
2270 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2271 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2272 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2273 bctl
->sys
.usage
= 90;
2275 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2276 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2277 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2278 bctl
->meta
.usage
= 90;
2283 * Should be called with both balance and volume mutexes held to
2284 * serialize other volume operations (add_dev/rm_dev/resize) with
2285 * restriper. Same goes for unset_balance_control.
2287 static void set_balance_control(struct btrfs_balance_control
*bctl
)
2289 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
2291 BUG_ON(fs_info
->balance_ctl
);
2293 spin_lock(&fs_info
->balance_lock
);
2294 fs_info
->balance_ctl
= bctl
;
2295 spin_unlock(&fs_info
->balance_lock
);
2298 static void unset_balance_control(struct btrfs_fs_info
*fs_info
)
2300 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
2302 BUG_ON(!fs_info
->balance_ctl
);
2304 spin_lock(&fs_info
->balance_lock
);
2305 fs_info
->balance_ctl
= NULL
;
2306 spin_unlock(&fs_info
->balance_lock
);
2312 * Balance filters. Return 1 if chunk should be filtered out
2313 * (should not be balanced).
2315 static int chunk_profiles_filter(u64 chunk_type
,
2316 struct btrfs_balance_args
*bargs
)
2318 chunk_type
= chunk_to_extended(chunk_type
) &
2319 BTRFS_EXTENDED_PROFILE_MASK
;
2321 if (bargs
->profiles
& chunk_type
)
2327 static u64
div_factor_fine(u64 num
, int factor
)
2339 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
2340 struct btrfs_balance_args
*bargs
)
2342 struct btrfs_block_group_cache
*cache
;
2343 u64 chunk_used
, user_thresh
;
2346 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
2347 chunk_used
= btrfs_block_group_used(&cache
->item
);
2349 user_thresh
= div_factor_fine(cache
->key
.offset
, bargs
->usage
);
2350 if (chunk_used
< user_thresh
)
2353 btrfs_put_block_group(cache
);
2357 static int chunk_devid_filter(struct extent_buffer
*leaf
,
2358 struct btrfs_chunk
*chunk
,
2359 struct btrfs_balance_args
*bargs
)
2361 struct btrfs_stripe
*stripe
;
2362 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2365 for (i
= 0; i
< num_stripes
; i
++) {
2366 stripe
= btrfs_stripe_nr(chunk
, i
);
2367 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
2374 /* [pstart, pend) */
2375 static int chunk_drange_filter(struct extent_buffer
*leaf
,
2376 struct btrfs_chunk
*chunk
,
2378 struct btrfs_balance_args
*bargs
)
2380 struct btrfs_stripe
*stripe
;
2381 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2387 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
2390 if (btrfs_chunk_type(leaf
, chunk
) & (BTRFS_BLOCK_GROUP_DUP
|
2391 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
))
2395 factor
= num_stripes
/ factor
;
2397 for (i
= 0; i
< num_stripes
; i
++) {
2398 stripe
= btrfs_stripe_nr(chunk
, i
);
2399 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
2402 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
2403 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
2404 do_div(stripe_length
, factor
);
2406 if (stripe_offset
< bargs
->pend
&&
2407 stripe_offset
+ stripe_length
> bargs
->pstart
)
2414 /* [vstart, vend) */
2415 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
2416 struct btrfs_chunk
*chunk
,
2418 struct btrfs_balance_args
*bargs
)
2420 if (chunk_offset
< bargs
->vend
&&
2421 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
2422 /* at least part of the chunk is inside this vrange */
2428 static int chunk_soft_convert_filter(u64 chunk_type
,
2429 struct btrfs_balance_args
*bargs
)
2431 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
2434 chunk_type
= chunk_to_extended(chunk_type
) &
2435 BTRFS_EXTENDED_PROFILE_MASK
;
2437 if (bargs
->target
== chunk_type
)
2443 static int should_balance_chunk(struct btrfs_root
*root
,
2444 struct extent_buffer
*leaf
,
2445 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
2447 struct btrfs_balance_control
*bctl
= root
->fs_info
->balance_ctl
;
2448 struct btrfs_balance_args
*bargs
= NULL
;
2449 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2452 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
2453 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
2457 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
2458 bargs
= &bctl
->data
;
2459 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
2461 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
2462 bargs
= &bctl
->meta
;
2464 /* profiles filter */
2465 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
2466 chunk_profiles_filter(chunk_type
, bargs
)) {
2471 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2472 chunk_usage_filter(bctl
->fs_info
, chunk_offset
, bargs
)) {
2477 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
2478 chunk_devid_filter(leaf
, chunk
, bargs
)) {
2482 /* drange filter, makes sense only with devid filter */
2483 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
2484 chunk_drange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
2489 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
2490 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
2494 /* soft profile changing mode */
2495 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
2496 chunk_soft_convert_filter(chunk_type
, bargs
)) {
2503 static u64
div_factor(u64 num
, int factor
)
2512 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
2514 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
2515 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
2516 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
2517 struct list_head
*devices
;
2518 struct btrfs_device
*device
;
2521 struct btrfs_chunk
*chunk
;
2522 struct btrfs_path
*path
;
2523 struct btrfs_key key
;
2524 struct btrfs_key found_key
;
2525 struct btrfs_trans_handle
*trans
;
2526 struct extent_buffer
*leaf
;
2529 int enospc_errors
= 0;
2530 bool counting
= true;
2532 /* step one make some room on all the devices */
2533 devices
= &fs_info
->fs_devices
->devices
;
2534 list_for_each_entry(device
, devices
, dev_list
) {
2535 old_size
= device
->total_bytes
;
2536 size_to_free
= div_factor(old_size
, 1);
2537 size_to_free
= min(size_to_free
, (u64
)1 * 1024 * 1024);
2538 if (!device
->writeable
||
2539 device
->total_bytes
- device
->bytes_used
> size_to_free
)
2542 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
2547 trans
= btrfs_start_transaction(dev_root
, 0);
2548 BUG_ON(IS_ERR(trans
));
2550 ret
= btrfs_grow_device(trans
, device
, old_size
);
2553 btrfs_end_transaction(trans
, dev_root
);
2556 /* step two, relocate all the chunks */
2557 path
= btrfs_alloc_path();
2563 /* zero out stat counters */
2564 spin_lock(&fs_info
->balance_lock
);
2565 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
2566 spin_unlock(&fs_info
->balance_lock
);
2568 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2569 key
.offset
= (u64
)-1;
2570 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2573 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
2574 atomic_read(&fs_info
->balance_cancel_req
)) {
2579 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2584 * this shouldn't happen, it means the last relocate
2588 BUG(); /* FIXME break ? */
2590 ret
= btrfs_previous_item(chunk_root
, path
, 0,
2591 BTRFS_CHUNK_ITEM_KEY
);
2597 leaf
= path
->nodes
[0];
2598 slot
= path
->slots
[0];
2599 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2601 if (found_key
.objectid
!= key
.objectid
)
2604 /* chunk zero is special */
2605 if (found_key
.offset
== 0)
2608 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
2611 spin_lock(&fs_info
->balance_lock
);
2612 bctl
->stat
.considered
++;
2613 spin_unlock(&fs_info
->balance_lock
);
2616 ret
= should_balance_chunk(chunk_root
, leaf
, chunk
,
2618 btrfs_release_path(path
);
2623 spin_lock(&fs_info
->balance_lock
);
2624 bctl
->stat
.expected
++;
2625 spin_unlock(&fs_info
->balance_lock
);
2629 ret
= btrfs_relocate_chunk(chunk_root
,
2630 chunk_root
->root_key
.objectid
,
2633 if (ret
&& ret
!= -ENOSPC
)
2635 if (ret
== -ENOSPC
) {
2638 spin_lock(&fs_info
->balance_lock
);
2639 bctl
->stat
.completed
++;
2640 spin_unlock(&fs_info
->balance_lock
);
2643 key
.offset
= found_key
.offset
- 1;
2647 btrfs_release_path(path
);
2652 btrfs_free_path(path
);
2653 if (enospc_errors
) {
2654 printk(KERN_INFO
"btrfs: %d enospc errors during balance\n",
2664 * alloc_profile_is_valid - see if a given profile is valid and reduced
2665 * @flags: profile to validate
2666 * @extended: if true @flags is treated as an extended profile
2668 static int alloc_profile_is_valid(u64 flags
, int extended
)
2670 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
2671 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
2673 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
2675 /* 1) check that all other bits are zeroed */
2679 /* 2) see if profile is reduced */
2681 return !extended
; /* "0" is valid for usual profiles */
2683 /* true if exactly one bit set */
2684 return (flags
& (flags
- 1)) == 0;
2687 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
2689 /* cancel requested || normal exit path */
2690 return atomic_read(&fs_info
->balance_cancel_req
) ||
2691 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
2692 atomic_read(&fs_info
->balance_cancel_req
) == 0);
2695 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
2699 unset_balance_control(fs_info
);
2700 ret
= del_balance_item(fs_info
->tree_root
);
2704 void update_ioctl_balance_args(struct btrfs_fs_info
*fs_info
, int lock
,
2705 struct btrfs_ioctl_balance_args
*bargs
);
2708 * Should be called with both balance and volume mutexes held
2710 int btrfs_balance(struct btrfs_balance_control
*bctl
,
2711 struct btrfs_ioctl_balance_args
*bargs
)
2713 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
2718 if (btrfs_fs_closing(fs_info
) ||
2719 atomic_read(&fs_info
->balance_pause_req
) ||
2720 atomic_read(&fs_info
->balance_cancel_req
)) {
2725 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
2726 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
2730 * In case of mixed groups both data and meta should be picked,
2731 * and identical options should be given for both of them.
2733 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
2734 if (mixed
&& (bctl
->flags
& allowed
)) {
2735 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
2736 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
2737 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
2738 printk(KERN_ERR
"btrfs: with mixed groups data and "
2739 "metadata balance options must be the same\n");
2745 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
2746 if (fs_info
->fs_devices
->num_devices
== 1)
2747 allowed
|= BTRFS_BLOCK_GROUP_DUP
;
2748 else if (fs_info
->fs_devices
->num_devices
< 4)
2749 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
2751 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
|
2752 BTRFS_BLOCK_GROUP_RAID10
);
2754 if ((bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
2755 (!alloc_profile_is_valid(bctl
->data
.target
, 1) ||
2756 (bctl
->data
.target
& ~allowed
))) {
2757 printk(KERN_ERR
"btrfs: unable to start balance with target "
2758 "data profile %llu\n",
2759 (unsigned long long)bctl
->data
.target
);
2763 if ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
2764 (!alloc_profile_is_valid(bctl
->meta
.target
, 1) ||
2765 (bctl
->meta
.target
& ~allowed
))) {
2766 printk(KERN_ERR
"btrfs: unable to start balance with target "
2767 "metadata profile %llu\n",
2768 (unsigned long long)bctl
->meta
.target
);
2772 if ((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
2773 (!alloc_profile_is_valid(bctl
->sys
.target
, 1) ||
2774 (bctl
->sys
.target
& ~allowed
))) {
2775 printk(KERN_ERR
"btrfs: unable to start balance with target "
2776 "system profile %llu\n",
2777 (unsigned long long)bctl
->sys
.target
);
2782 /* allow dup'ed data chunks only in mixed mode */
2783 if (!mixed
&& (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
2784 (bctl
->data
.target
& BTRFS_BLOCK_GROUP_DUP
)) {
2785 printk(KERN_ERR
"btrfs: dup for data is not allowed\n");
2790 /* allow to reduce meta or sys integrity only if force set */
2791 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
2792 BTRFS_BLOCK_GROUP_RAID10
;
2793 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
2794 (fs_info
->avail_system_alloc_bits
& allowed
) &&
2795 !(bctl
->sys
.target
& allowed
)) ||
2796 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
2797 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
2798 !(bctl
->meta
.target
& allowed
))) {
2799 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
2800 printk(KERN_INFO
"btrfs: force reducing metadata "
2803 printk(KERN_ERR
"btrfs: balance will reduce metadata "
2804 "integrity, use force if you want this\n");
2810 ret
= insert_balance_item(fs_info
->tree_root
, bctl
);
2811 if (ret
&& ret
!= -EEXIST
)
2814 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
2815 BUG_ON(ret
== -EEXIST
);
2816 set_balance_control(bctl
);
2818 BUG_ON(ret
!= -EEXIST
);
2819 spin_lock(&fs_info
->balance_lock
);
2820 update_balance_args(bctl
);
2821 spin_unlock(&fs_info
->balance_lock
);
2824 atomic_inc(&fs_info
->balance_running
);
2825 mutex_unlock(&fs_info
->balance_mutex
);
2827 ret
= __btrfs_balance(fs_info
);
2829 mutex_lock(&fs_info
->balance_mutex
);
2830 atomic_dec(&fs_info
->balance_running
);
2833 memset(bargs
, 0, sizeof(*bargs
));
2834 update_ioctl_balance_args(fs_info
, 0, bargs
);
2837 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
2838 balance_need_close(fs_info
)) {
2839 __cancel_balance(fs_info
);
2842 wake_up(&fs_info
->balance_wait_q
);
2846 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
2847 __cancel_balance(fs_info
);
2853 static int balance_kthread(void *data
)
2855 struct btrfs_fs_info
*fs_info
= data
;
2858 mutex_lock(&fs_info
->volume_mutex
);
2859 mutex_lock(&fs_info
->balance_mutex
);
2861 if (fs_info
->balance_ctl
) {
2862 printk(KERN_INFO
"btrfs: continuing balance\n");
2863 ret
= btrfs_balance(fs_info
->balance_ctl
, NULL
);
2866 mutex_unlock(&fs_info
->balance_mutex
);
2867 mutex_unlock(&fs_info
->volume_mutex
);
2872 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
2874 struct task_struct
*tsk
;
2876 spin_lock(&fs_info
->balance_lock
);
2877 if (!fs_info
->balance_ctl
) {
2878 spin_unlock(&fs_info
->balance_lock
);
2881 spin_unlock(&fs_info
->balance_lock
);
2883 if (btrfs_test_opt(fs_info
->tree_root
, SKIP_BALANCE
)) {
2884 printk(KERN_INFO
"btrfs: force skipping balance\n");
2888 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
2890 return PTR_ERR(tsk
);
2895 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
2897 struct btrfs_balance_control
*bctl
;
2898 struct btrfs_balance_item
*item
;
2899 struct btrfs_disk_balance_args disk_bargs
;
2900 struct btrfs_path
*path
;
2901 struct extent_buffer
*leaf
;
2902 struct btrfs_key key
;
2905 path
= btrfs_alloc_path();
2909 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2910 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2913 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
2916 if (ret
> 0) { /* ret = -ENOENT; */
2921 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
2927 leaf
= path
->nodes
[0];
2928 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
2930 bctl
->fs_info
= fs_info
;
2931 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
2932 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
2934 btrfs_balance_data(leaf
, item
, &disk_bargs
);
2935 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
2936 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
2937 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
2938 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
2939 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
2941 mutex_lock(&fs_info
->volume_mutex
);
2942 mutex_lock(&fs_info
->balance_mutex
);
2944 set_balance_control(bctl
);
2946 mutex_unlock(&fs_info
->balance_mutex
);
2947 mutex_unlock(&fs_info
->volume_mutex
);
2949 btrfs_free_path(path
);
2953 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
2957 mutex_lock(&fs_info
->balance_mutex
);
2958 if (!fs_info
->balance_ctl
) {
2959 mutex_unlock(&fs_info
->balance_mutex
);
2963 if (atomic_read(&fs_info
->balance_running
)) {
2964 atomic_inc(&fs_info
->balance_pause_req
);
2965 mutex_unlock(&fs_info
->balance_mutex
);
2967 wait_event(fs_info
->balance_wait_q
,
2968 atomic_read(&fs_info
->balance_running
) == 0);
2970 mutex_lock(&fs_info
->balance_mutex
);
2971 /* we are good with balance_ctl ripped off from under us */
2972 BUG_ON(atomic_read(&fs_info
->balance_running
));
2973 atomic_dec(&fs_info
->balance_pause_req
);
2978 mutex_unlock(&fs_info
->balance_mutex
);
2982 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
2984 mutex_lock(&fs_info
->balance_mutex
);
2985 if (!fs_info
->balance_ctl
) {
2986 mutex_unlock(&fs_info
->balance_mutex
);
2990 atomic_inc(&fs_info
->balance_cancel_req
);
2992 * if we are running just wait and return, balance item is
2993 * deleted in btrfs_balance in this case
2995 if (atomic_read(&fs_info
->balance_running
)) {
2996 mutex_unlock(&fs_info
->balance_mutex
);
2997 wait_event(fs_info
->balance_wait_q
,
2998 atomic_read(&fs_info
->balance_running
) == 0);
2999 mutex_lock(&fs_info
->balance_mutex
);
3001 /* __cancel_balance needs volume_mutex */
3002 mutex_unlock(&fs_info
->balance_mutex
);
3003 mutex_lock(&fs_info
->volume_mutex
);
3004 mutex_lock(&fs_info
->balance_mutex
);
3006 if (fs_info
->balance_ctl
)
3007 __cancel_balance(fs_info
);
3009 mutex_unlock(&fs_info
->volume_mutex
);
3012 BUG_ON(fs_info
->balance_ctl
|| atomic_read(&fs_info
->balance_running
));
3013 atomic_dec(&fs_info
->balance_cancel_req
);
3014 mutex_unlock(&fs_info
->balance_mutex
);
3019 * shrinking a device means finding all of the device extents past
3020 * the new size, and then following the back refs to the chunks.
3021 * The chunk relocation code actually frees the device extent
3023 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
3025 struct btrfs_trans_handle
*trans
;
3026 struct btrfs_root
*root
= device
->dev_root
;
3027 struct btrfs_dev_extent
*dev_extent
= NULL
;
3028 struct btrfs_path
*path
;
3036 bool retried
= false;
3037 struct extent_buffer
*l
;
3038 struct btrfs_key key
;
3039 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
3040 u64 old_total
= btrfs_super_total_bytes(super_copy
);
3041 u64 old_size
= device
->total_bytes
;
3042 u64 diff
= device
->total_bytes
- new_size
;
3044 if (new_size
>= device
->total_bytes
)
3047 path
= btrfs_alloc_path();
3055 device
->total_bytes
= new_size
;
3056 if (device
->writeable
) {
3057 device
->fs_devices
->total_rw_bytes
-= diff
;
3058 spin_lock(&root
->fs_info
->free_chunk_lock
);
3059 root
->fs_info
->free_chunk_space
-= diff
;
3060 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3062 unlock_chunks(root
);
3065 key
.objectid
= device
->devid
;
3066 key
.offset
= (u64
)-1;
3067 key
.type
= BTRFS_DEV_EXTENT_KEY
;
3070 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3074 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
3079 btrfs_release_path(path
);
3084 slot
= path
->slots
[0];
3085 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
3087 if (key
.objectid
!= device
->devid
) {
3088 btrfs_release_path(path
);
3092 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
3093 length
= btrfs_dev_extent_length(l
, dev_extent
);
3095 if (key
.offset
+ length
<= new_size
) {
3096 btrfs_release_path(path
);
3100 chunk_tree
= btrfs_dev_extent_chunk_tree(l
, dev_extent
);
3101 chunk_objectid
= btrfs_dev_extent_chunk_objectid(l
, dev_extent
);
3102 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
3103 btrfs_release_path(path
);
3105 ret
= btrfs_relocate_chunk(root
, chunk_tree
, chunk_objectid
,
3107 if (ret
&& ret
!= -ENOSPC
)
3111 } while (key
.offset
-- > 0);
3113 if (failed
&& !retried
) {
3117 } else if (failed
&& retried
) {
3121 device
->total_bytes
= old_size
;
3122 if (device
->writeable
)
3123 device
->fs_devices
->total_rw_bytes
+= diff
;
3124 spin_lock(&root
->fs_info
->free_chunk_lock
);
3125 root
->fs_info
->free_chunk_space
+= diff
;
3126 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3127 unlock_chunks(root
);
3131 /* Shrinking succeeded, else we would be at "done". */
3132 trans
= btrfs_start_transaction(root
, 0);
3133 if (IS_ERR(trans
)) {
3134 ret
= PTR_ERR(trans
);
3140 device
->disk_total_bytes
= new_size
;
3141 /* Now btrfs_update_device() will change the on-disk size. */
3142 ret
= btrfs_update_device(trans
, device
);
3144 unlock_chunks(root
);
3145 btrfs_end_transaction(trans
, root
);
3148 WARN_ON(diff
> old_total
);
3149 btrfs_set_super_total_bytes(super_copy
, old_total
- diff
);
3150 unlock_chunks(root
);
3151 btrfs_end_transaction(trans
, root
);
3153 btrfs_free_path(path
);
3157 static int btrfs_add_system_chunk(struct btrfs_root
*root
,
3158 struct btrfs_key
*key
,
3159 struct btrfs_chunk
*chunk
, int item_size
)
3161 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
3162 struct btrfs_disk_key disk_key
;
3166 array_size
= btrfs_super_sys_array_size(super_copy
);
3167 if (array_size
+ item_size
> BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
)
3170 ptr
= super_copy
->sys_chunk_array
+ array_size
;
3171 btrfs_cpu_key_to_disk(&disk_key
, key
);
3172 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
3173 ptr
+= sizeof(disk_key
);
3174 memcpy(ptr
, chunk
, item_size
);
3175 item_size
+= sizeof(disk_key
);
3176 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
3181 * sort the devices in descending order by max_avail, total_avail
3183 static int btrfs_cmp_device_info(const void *a
, const void *b
)
3185 const struct btrfs_device_info
*di_a
= a
;
3186 const struct btrfs_device_info
*di_b
= b
;
3188 if (di_a
->max_avail
> di_b
->max_avail
)
3190 if (di_a
->max_avail
< di_b
->max_avail
)
3192 if (di_a
->total_avail
> di_b
->total_avail
)
3194 if (di_a
->total_avail
< di_b
->total_avail
)
3199 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
3200 struct btrfs_root
*extent_root
,
3201 struct map_lookup
**map_ret
,
3202 u64
*num_bytes_out
, u64
*stripe_size_out
,
3203 u64 start
, u64 type
)
3205 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
3206 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
3207 struct list_head
*cur
;
3208 struct map_lookup
*map
= NULL
;
3209 struct extent_map_tree
*em_tree
;
3210 struct extent_map
*em
;
3211 struct btrfs_device_info
*devices_info
= NULL
;
3213 int num_stripes
; /* total number of stripes to allocate */
3214 int sub_stripes
; /* sub_stripes info for map */
3215 int dev_stripes
; /* stripes per dev */
3216 int devs_max
; /* max devs to use */
3217 int devs_min
; /* min devs needed */
3218 int devs_increment
; /* ndevs has to be a multiple of this */
3219 int ncopies
; /* how many copies to data has */
3221 u64 max_stripe_size
;
3229 BUG_ON(!alloc_profile_is_valid(type
, 0));
3231 if (list_empty(&fs_devices
->alloc_list
))
3238 devs_max
= 0; /* 0 == as many as possible */
3242 * define the properties of each RAID type.
3243 * FIXME: move this to a global table and use it in all RAID
3246 if (type
& (BTRFS_BLOCK_GROUP_DUP
)) {
3250 } else if (type
& (BTRFS_BLOCK_GROUP_RAID0
)) {
3252 } else if (type
& (BTRFS_BLOCK_GROUP_RAID1
)) {
3257 } else if (type
& (BTRFS_BLOCK_GROUP_RAID10
)) {
3266 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
3267 max_stripe_size
= 1024 * 1024 * 1024;
3268 max_chunk_size
= 10 * max_stripe_size
;
3269 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
3270 /* for larger filesystems, use larger metadata chunks */
3271 if (fs_devices
->total_rw_bytes
> 50ULL * 1024 * 1024 * 1024)
3272 max_stripe_size
= 1024 * 1024 * 1024;
3274 max_stripe_size
= 256 * 1024 * 1024;
3275 max_chunk_size
= max_stripe_size
;
3276 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3277 max_stripe_size
= 32 * 1024 * 1024;
3278 max_chunk_size
= 2 * max_stripe_size
;
3280 printk(KERN_ERR
"btrfs: invalid chunk type 0x%llx requested\n",
3285 /* we don't want a chunk larger than 10% of writeable space */
3286 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
3289 devices_info
= kzalloc(sizeof(*devices_info
) * fs_devices
->rw_devices
,
3294 cur
= fs_devices
->alloc_list
.next
;
3297 * in the first pass through the devices list, we gather information
3298 * about the available holes on each device.
3301 while (cur
!= &fs_devices
->alloc_list
) {
3302 struct btrfs_device
*device
;
3306 device
= list_entry(cur
, struct btrfs_device
, dev_alloc_list
);
3310 if (!device
->writeable
) {
3312 "btrfs: read-only device in alloc_list\n");
3317 if (!device
->in_fs_metadata
)
3320 if (device
->total_bytes
> device
->bytes_used
)
3321 total_avail
= device
->total_bytes
- device
->bytes_used
;
3325 /* If there is no space on this device, skip it. */
3326 if (total_avail
== 0)
3329 ret
= find_free_dev_extent(device
,
3330 max_stripe_size
* dev_stripes
,
3331 &dev_offset
, &max_avail
);
3332 if (ret
&& ret
!= -ENOSPC
)
3336 max_avail
= max_stripe_size
* dev_stripes
;
3338 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
3341 devices_info
[ndevs
].dev_offset
= dev_offset
;
3342 devices_info
[ndevs
].max_avail
= max_avail
;
3343 devices_info
[ndevs
].total_avail
= total_avail
;
3344 devices_info
[ndevs
].dev
= device
;
3349 * now sort the devices by hole size / available space
3351 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
3352 btrfs_cmp_device_info
, NULL
);
3354 /* round down to number of usable stripes */
3355 ndevs
-= ndevs
% devs_increment
;
3357 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
3362 if (devs_max
&& ndevs
> devs_max
)
3365 * the primary goal is to maximize the number of stripes, so use as many
3366 * devices as possible, even if the stripes are not maximum sized.
3368 stripe_size
= devices_info
[ndevs
-1].max_avail
;
3369 num_stripes
= ndevs
* dev_stripes
;
3371 if (stripe_size
* ndevs
> max_chunk_size
* ncopies
) {
3372 stripe_size
= max_chunk_size
* ncopies
;
3373 do_div(stripe_size
, ndevs
);
3376 do_div(stripe_size
, dev_stripes
);
3378 /* align to BTRFS_STRIPE_LEN */
3379 do_div(stripe_size
, BTRFS_STRIPE_LEN
);
3380 stripe_size
*= BTRFS_STRIPE_LEN
;
3382 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
3387 map
->num_stripes
= num_stripes
;
3389 for (i
= 0; i
< ndevs
; ++i
) {
3390 for (j
= 0; j
< dev_stripes
; ++j
) {
3391 int s
= i
* dev_stripes
+ j
;
3392 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
3393 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
3397 map
->sector_size
= extent_root
->sectorsize
;
3398 map
->stripe_len
= BTRFS_STRIPE_LEN
;
3399 map
->io_align
= BTRFS_STRIPE_LEN
;
3400 map
->io_width
= BTRFS_STRIPE_LEN
;
3402 map
->sub_stripes
= sub_stripes
;
3405 num_bytes
= stripe_size
* (num_stripes
/ ncopies
);
3407 *stripe_size_out
= stripe_size
;
3408 *num_bytes_out
= num_bytes
;
3410 trace_btrfs_chunk_alloc(info
->chunk_root
, map
, start
, num_bytes
);
3412 em
= alloc_extent_map();
3417 em
->bdev
= (struct block_device
*)map
;
3419 em
->len
= num_bytes
;
3420 em
->block_start
= 0;
3421 em
->block_len
= em
->len
;
3423 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
3424 write_lock(&em_tree
->lock
);
3425 ret
= add_extent_mapping(em_tree
, em
);
3426 write_unlock(&em_tree
->lock
);
3427 free_extent_map(em
);
3431 ret
= btrfs_make_block_group(trans
, extent_root
, 0, type
,
3432 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
3437 for (i
= 0; i
< map
->num_stripes
; ++i
) {
3438 struct btrfs_device
*device
;
3441 device
= map
->stripes
[i
].dev
;
3442 dev_offset
= map
->stripes
[i
].physical
;
3444 ret
= btrfs_alloc_dev_extent(trans
, device
,
3445 info
->chunk_root
->root_key
.objectid
,
3446 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
3447 start
, dev_offset
, stripe_size
);
3449 btrfs_abort_transaction(trans
, extent_root
, ret
);
3454 kfree(devices_info
);
3459 kfree(devices_info
);
3463 static int __finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
3464 struct btrfs_root
*extent_root
,
3465 struct map_lookup
*map
, u64 chunk_offset
,
3466 u64 chunk_size
, u64 stripe_size
)
3469 struct btrfs_key key
;
3470 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
3471 struct btrfs_device
*device
;
3472 struct btrfs_chunk
*chunk
;
3473 struct btrfs_stripe
*stripe
;
3474 size_t item_size
= btrfs_chunk_item_size(map
->num_stripes
);
3478 chunk
= kzalloc(item_size
, GFP_NOFS
);
3483 while (index
< map
->num_stripes
) {
3484 device
= map
->stripes
[index
].dev
;
3485 device
->bytes_used
+= stripe_size
;
3486 ret
= btrfs_update_device(trans
, device
);
3492 spin_lock(&extent_root
->fs_info
->free_chunk_lock
);
3493 extent_root
->fs_info
->free_chunk_space
-= (stripe_size
*
3495 spin_unlock(&extent_root
->fs_info
->free_chunk_lock
);
3498 stripe
= &chunk
->stripe
;
3499 while (index
< map
->num_stripes
) {
3500 device
= map
->stripes
[index
].dev
;
3501 dev_offset
= map
->stripes
[index
].physical
;
3503 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
3504 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
3505 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
3510 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
3511 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
3512 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
3513 btrfs_set_stack_chunk_type(chunk
, map
->type
);
3514 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
3515 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
3516 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
3517 btrfs_set_stack_chunk_sector_size(chunk
, extent_root
->sectorsize
);
3518 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
3520 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3521 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3522 key
.offset
= chunk_offset
;
3524 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
3526 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3528 * TODO: Cleanup of inserted chunk root in case of
3531 ret
= btrfs_add_system_chunk(chunk_root
, &key
, chunk
,
3541 * Chunk allocation falls into two parts. The first part does works
3542 * that make the new allocated chunk useable, but not do any operation
3543 * that modifies the chunk tree. The second part does the works that
3544 * require modifying the chunk tree. This division is important for the
3545 * bootstrap process of adding storage to a seed btrfs.
3547 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
3548 struct btrfs_root
*extent_root
, u64 type
)
3553 struct map_lookup
*map
;
3554 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
3557 ret
= find_next_chunk(chunk_root
, BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
3562 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &map
, &chunk_size
,
3563 &stripe_size
, chunk_offset
, type
);
3567 ret
= __finish_chunk_alloc(trans
, extent_root
, map
, chunk_offset
,
3568 chunk_size
, stripe_size
);
3574 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
3575 struct btrfs_root
*root
,
3576 struct btrfs_device
*device
)
3579 u64 sys_chunk_offset
;
3583 u64 sys_stripe_size
;
3585 struct map_lookup
*map
;
3586 struct map_lookup
*sys_map
;
3587 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3588 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
3591 ret
= find_next_chunk(fs_info
->chunk_root
,
3592 BTRFS_FIRST_CHUNK_TREE_OBJECTID
, &chunk_offset
);
3596 alloc_profile
= BTRFS_BLOCK_GROUP_METADATA
|
3597 fs_info
->avail_metadata_alloc_bits
;
3598 alloc_profile
= btrfs_reduce_alloc_profile(root
, alloc_profile
);
3600 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &map
, &chunk_size
,
3601 &stripe_size
, chunk_offset
, alloc_profile
);
3605 sys_chunk_offset
= chunk_offset
+ chunk_size
;
3607 alloc_profile
= BTRFS_BLOCK_GROUP_SYSTEM
|
3608 fs_info
->avail_system_alloc_bits
;
3609 alloc_profile
= btrfs_reduce_alloc_profile(root
, alloc_profile
);
3611 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &sys_map
,
3612 &sys_chunk_size
, &sys_stripe_size
,
3613 sys_chunk_offset
, alloc_profile
);
3617 ret
= btrfs_add_device(trans
, fs_info
->chunk_root
, device
);
3622 * Modifying chunk tree needs allocating new blocks from both
3623 * system block group and metadata block group. So we only can
3624 * do operations require modifying the chunk tree after both
3625 * block groups were created.
3627 ret
= __finish_chunk_alloc(trans
, extent_root
, map
, chunk_offset
,
3628 chunk_size
, stripe_size
);
3632 ret
= __finish_chunk_alloc(trans
, extent_root
, sys_map
,
3633 sys_chunk_offset
, sys_chunk_size
,
3641 btrfs_abort_transaction(trans
, root
, ret
);
3645 int btrfs_chunk_readonly(struct btrfs_root
*root
, u64 chunk_offset
)
3647 struct extent_map
*em
;
3648 struct map_lookup
*map
;
3649 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
3653 read_lock(&map_tree
->map_tree
.lock
);
3654 em
= lookup_extent_mapping(&map_tree
->map_tree
, chunk_offset
, 1);
3655 read_unlock(&map_tree
->map_tree
.lock
);
3659 if (btrfs_test_opt(root
, DEGRADED
)) {
3660 free_extent_map(em
);
3664 map
= (struct map_lookup
*)em
->bdev
;
3665 for (i
= 0; i
< map
->num_stripes
; i
++) {
3666 if (!map
->stripes
[i
].dev
->writeable
) {
3671 free_extent_map(em
);
3675 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
3677 extent_map_tree_init(&tree
->map_tree
);
3680 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
3682 struct extent_map
*em
;
3685 write_lock(&tree
->map_tree
.lock
);
3686 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
3688 remove_extent_mapping(&tree
->map_tree
, em
);
3689 write_unlock(&tree
->map_tree
.lock
);
3694 free_extent_map(em
);
3695 /* once for the tree */
3696 free_extent_map(em
);
3700 int btrfs_num_copies(struct btrfs_mapping_tree
*map_tree
, u64 logical
, u64 len
)
3702 struct extent_map
*em
;
3703 struct map_lookup
*map
;
3704 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
3707 read_lock(&em_tree
->lock
);
3708 em
= lookup_extent_mapping(em_tree
, logical
, len
);
3709 read_unlock(&em_tree
->lock
);
3712 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
3713 map
= (struct map_lookup
*)em
->bdev
;
3714 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
3715 ret
= map
->num_stripes
;
3716 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
3717 ret
= map
->sub_stripes
;
3720 free_extent_map(em
);
3724 static int find_live_mirror(struct map_lookup
*map
, int first
, int num
,
3728 if (map
->stripes
[optimal
].dev
->bdev
)
3730 for (i
= first
; i
< first
+ num
; i
++) {
3731 if (map
->stripes
[i
].dev
->bdev
)
3734 /* we couldn't find one that doesn't fail. Just return something
3735 * and the io error handling code will clean up eventually
3740 static int __btrfs_map_block(struct btrfs_mapping_tree
*map_tree
, int rw
,
3741 u64 logical
, u64
*length
,
3742 struct btrfs_bio
**bbio_ret
,
3745 struct extent_map
*em
;
3746 struct map_lookup
*map
;
3747 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
3750 u64 stripe_end_offset
;
3759 struct btrfs_bio
*bbio
= NULL
;
3761 read_lock(&em_tree
->lock
);
3762 em
= lookup_extent_mapping(em_tree
, logical
, *length
);
3763 read_unlock(&em_tree
->lock
);
3766 printk(KERN_CRIT
"unable to find logical %llu len %llu\n",
3767 (unsigned long long)logical
,
3768 (unsigned long long)*length
);
3772 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
3773 map
= (struct map_lookup
*)em
->bdev
;
3774 offset
= logical
- em
->start
;
3776 if (mirror_num
> map
->num_stripes
)
3781 * stripe_nr counts the total number of stripes we have to stride
3782 * to get to this block
3784 do_div(stripe_nr
, map
->stripe_len
);
3786 stripe_offset
= stripe_nr
* map
->stripe_len
;
3787 BUG_ON(offset
< stripe_offset
);
3789 /* stripe_offset is the offset of this block in its stripe*/
3790 stripe_offset
= offset
- stripe_offset
;
3792 if (rw
& REQ_DISCARD
)
3793 *length
= min_t(u64
, em
->len
- offset
, *length
);
3794 else if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
3795 /* we limit the length of each bio to what fits in a stripe */
3796 *length
= min_t(u64
, em
->len
- offset
,
3797 map
->stripe_len
- stripe_offset
);
3799 *length
= em
->len
- offset
;
3807 stripe_nr_orig
= stripe_nr
;
3808 stripe_nr_end
= (offset
+ *length
+ map
->stripe_len
- 1) &
3809 (~(map
->stripe_len
- 1));
3810 do_div(stripe_nr_end
, map
->stripe_len
);
3811 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
3813 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
3814 if (rw
& REQ_DISCARD
)
3815 num_stripes
= min_t(u64
, map
->num_stripes
,
3816 stripe_nr_end
- stripe_nr_orig
);
3817 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
3818 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
3819 if (rw
& (REQ_WRITE
| REQ_DISCARD
))
3820 num_stripes
= map
->num_stripes
;
3821 else if (mirror_num
)
3822 stripe_index
= mirror_num
- 1;
3824 stripe_index
= find_live_mirror(map
, 0,
3826 current
->pid
% map
->num_stripes
);
3827 mirror_num
= stripe_index
+ 1;
3830 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
3831 if (rw
& (REQ_WRITE
| REQ_DISCARD
)) {
3832 num_stripes
= map
->num_stripes
;
3833 } else if (mirror_num
) {
3834 stripe_index
= mirror_num
- 1;
3839 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
3840 int factor
= map
->num_stripes
/ map
->sub_stripes
;
3842 stripe_index
= do_div(stripe_nr
, factor
);
3843 stripe_index
*= map
->sub_stripes
;
3846 num_stripes
= map
->sub_stripes
;
3847 else if (rw
& REQ_DISCARD
)
3848 num_stripes
= min_t(u64
, map
->sub_stripes
*
3849 (stripe_nr_end
- stripe_nr_orig
),
3851 else if (mirror_num
)
3852 stripe_index
+= mirror_num
- 1;
3854 int old_stripe_index
= stripe_index
;
3855 stripe_index
= find_live_mirror(map
, stripe_index
,
3856 map
->sub_stripes
, stripe_index
+
3857 current
->pid
% map
->sub_stripes
);
3858 mirror_num
= stripe_index
- old_stripe_index
+ 1;
3862 * after this do_div call, stripe_nr is the number of stripes
3863 * on this device we have to walk to find the data, and
3864 * stripe_index is the number of our device in the stripe array
3866 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
3867 mirror_num
= stripe_index
+ 1;
3869 BUG_ON(stripe_index
>= map
->num_stripes
);
3871 bbio
= kzalloc(btrfs_bio_size(num_stripes
), GFP_NOFS
);
3876 atomic_set(&bbio
->error
, 0);
3878 if (rw
& REQ_DISCARD
) {
3880 int sub_stripes
= 0;
3881 u64 stripes_per_dev
= 0;
3882 u32 remaining_stripes
= 0;
3883 u32 last_stripe
= 0;
3886 (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID10
)) {
3887 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
3890 sub_stripes
= map
->sub_stripes
;
3892 factor
= map
->num_stripes
/ sub_stripes
;
3893 stripes_per_dev
= div_u64_rem(stripe_nr_end
-
3896 &remaining_stripes
);
3897 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
3898 last_stripe
*= sub_stripes
;
3901 for (i
= 0; i
< num_stripes
; i
++) {
3902 bbio
->stripes
[i
].physical
=
3903 map
->stripes
[stripe_index
].physical
+
3904 stripe_offset
+ stripe_nr
* map
->stripe_len
;
3905 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
3907 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
3908 BTRFS_BLOCK_GROUP_RAID10
)) {
3909 bbio
->stripes
[i
].length
= stripes_per_dev
*
3912 if (i
/ sub_stripes
< remaining_stripes
)
3913 bbio
->stripes
[i
].length
+=
3917 * Special for the first stripe and
3920 * |-------|...|-------|
3924 if (i
< sub_stripes
)
3925 bbio
->stripes
[i
].length
-=
3928 if (stripe_index
>= last_stripe
&&
3929 stripe_index
<= (last_stripe
+
3931 bbio
->stripes
[i
].length
-=
3934 if (i
== sub_stripes
- 1)
3937 bbio
->stripes
[i
].length
= *length
;
3940 if (stripe_index
== map
->num_stripes
) {
3941 /* This could only happen for RAID0/10 */
3947 for (i
= 0; i
< num_stripes
; i
++) {
3948 bbio
->stripes
[i
].physical
=
3949 map
->stripes
[stripe_index
].physical
+
3951 stripe_nr
* map
->stripe_len
;
3952 bbio
->stripes
[i
].dev
=
3953 map
->stripes
[stripe_index
].dev
;
3958 if (rw
& REQ_WRITE
) {
3959 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
3960 BTRFS_BLOCK_GROUP_RAID10
|
3961 BTRFS_BLOCK_GROUP_DUP
)) {
3967 bbio
->num_stripes
= num_stripes
;
3968 bbio
->max_errors
= max_errors
;
3969 bbio
->mirror_num
= mirror_num
;
3971 free_extent_map(em
);
3975 int btrfs_map_block(struct btrfs_mapping_tree
*map_tree
, int rw
,
3976 u64 logical
, u64
*length
,
3977 struct btrfs_bio
**bbio_ret
, int mirror_num
)
3979 return __btrfs_map_block(map_tree
, rw
, logical
, length
, bbio_ret
,
3983 int btrfs_rmap_block(struct btrfs_mapping_tree
*map_tree
,
3984 u64 chunk_start
, u64 physical
, u64 devid
,
3985 u64
**logical
, int *naddrs
, int *stripe_len
)
3987 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
3988 struct extent_map
*em
;
3989 struct map_lookup
*map
;
3996 read_lock(&em_tree
->lock
);
3997 em
= lookup_extent_mapping(em_tree
, chunk_start
, 1);
3998 read_unlock(&em_tree
->lock
);
4000 BUG_ON(!em
|| em
->start
!= chunk_start
);
4001 map
= (struct map_lookup
*)em
->bdev
;
4004 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
4005 do_div(length
, map
->num_stripes
/ map
->sub_stripes
);
4006 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
4007 do_div(length
, map
->num_stripes
);
4009 buf
= kzalloc(sizeof(u64
) * map
->num_stripes
, GFP_NOFS
);
4010 BUG_ON(!buf
); /* -ENOMEM */
4012 for (i
= 0; i
< map
->num_stripes
; i
++) {
4013 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
4015 if (map
->stripes
[i
].physical
> physical
||
4016 map
->stripes
[i
].physical
+ length
<= physical
)
4019 stripe_nr
= physical
- map
->stripes
[i
].physical
;
4020 do_div(stripe_nr
, map
->stripe_len
);
4022 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
4023 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
4024 do_div(stripe_nr
, map
->sub_stripes
);
4025 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
4026 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
4028 bytenr
= chunk_start
+ stripe_nr
* map
->stripe_len
;
4029 WARN_ON(nr
>= map
->num_stripes
);
4030 for (j
= 0; j
< nr
; j
++) {
4031 if (buf
[j
] == bytenr
)
4035 WARN_ON(nr
>= map
->num_stripes
);
4042 *stripe_len
= map
->stripe_len
;
4044 free_extent_map(em
);
4048 static void *merge_stripe_index_into_bio_private(void *bi_private
,
4049 unsigned int stripe_index
)
4052 * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
4054 * The alternative solution (instead of stealing bits from the
4055 * pointer) would be to allocate an intermediate structure
4056 * that contains the old private pointer plus the stripe_index.
4058 BUG_ON((((uintptr_t)bi_private
) & 3) != 0);
4059 BUG_ON(stripe_index
> 3);
4060 return (void *)(((uintptr_t)bi_private
) | stripe_index
);
4063 static struct btrfs_bio
*extract_bbio_from_bio_private(void *bi_private
)
4065 return (struct btrfs_bio
*)(((uintptr_t)bi_private
) & ~((uintptr_t)3));
4068 static unsigned int extract_stripe_index_from_bio_private(void *bi_private
)
4070 return (unsigned int)((uintptr_t)bi_private
) & 3;
4073 static void btrfs_end_bio(struct bio
*bio
, int err
)
4075 struct btrfs_bio
*bbio
= extract_bbio_from_bio_private(bio
->bi_private
);
4076 int is_orig_bio
= 0;
4079 atomic_inc(&bbio
->error
);
4080 if (err
== -EIO
|| err
== -EREMOTEIO
) {
4081 unsigned int stripe_index
=
4082 extract_stripe_index_from_bio_private(
4084 struct btrfs_device
*dev
;
4086 BUG_ON(stripe_index
>= bbio
->num_stripes
);
4087 dev
= bbio
->stripes
[stripe_index
].dev
;
4089 if (bio
->bi_rw
& WRITE
)
4090 btrfs_dev_stat_inc(dev
,
4091 BTRFS_DEV_STAT_WRITE_ERRS
);
4093 btrfs_dev_stat_inc(dev
,
4094 BTRFS_DEV_STAT_READ_ERRS
);
4095 if ((bio
->bi_rw
& WRITE_FLUSH
) == WRITE_FLUSH
)
4096 btrfs_dev_stat_inc(dev
,
4097 BTRFS_DEV_STAT_FLUSH_ERRS
);
4098 btrfs_dev_stat_print_on_error(dev
);
4103 if (bio
== bbio
->orig_bio
)
4106 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
4109 bio
= bbio
->orig_bio
;
4111 bio
->bi_private
= bbio
->private;
4112 bio
->bi_end_io
= bbio
->end_io
;
4113 bio
->bi_bdev
= (struct block_device
*)
4114 (unsigned long)bbio
->mirror_num
;
4115 /* only send an error to the higher layers if it is
4116 * beyond the tolerance of the multi-bio
4118 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
4122 * this bio is actually up to date, we didn't
4123 * go over the max number of errors
4125 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
4130 bio_endio(bio
, err
);
4131 } else if (!is_orig_bio
) {
4136 struct async_sched
{
4139 struct btrfs_fs_info
*info
;
4140 struct btrfs_work work
;
4144 * see run_scheduled_bios for a description of why bios are collected for
4147 * This will add one bio to the pending list for a device and make sure
4148 * the work struct is scheduled.
4150 static noinline
void schedule_bio(struct btrfs_root
*root
,
4151 struct btrfs_device
*device
,
4152 int rw
, struct bio
*bio
)
4154 int should_queue
= 1;
4155 struct btrfs_pending_bios
*pending_bios
;
4157 /* don't bother with additional async steps for reads, right now */
4158 if (!(rw
& REQ_WRITE
)) {
4160 btrfsic_submit_bio(rw
, bio
);
4166 * nr_async_bios allows us to reliably return congestion to the
4167 * higher layers. Otherwise, the async bio makes it appear we have
4168 * made progress against dirty pages when we've really just put it
4169 * on a queue for later
4171 atomic_inc(&root
->fs_info
->nr_async_bios
);
4172 WARN_ON(bio
->bi_next
);
4173 bio
->bi_next
= NULL
;
4176 spin_lock(&device
->io_lock
);
4177 if (bio
->bi_rw
& REQ_SYNC
)
4178 pending_bios
= &device
->pending_sync_bios
;
4180 pending_bios
= &device
->pending_bios
;
4182 if (pending_bios
->tail
)
4183 pending_bios
->tail
->bi_next
= bio
;
4185 pending_bios
->tail
= bio
;
4186 if (!pending_bios
->head
)
4187 pending_bios
->head
= bio
;
4188 if (device
->running_pending
)
4191 spin_unlock(&device
->io_lock
);
4194 btrfs_queue_worker(&root
->fs_info
->submit_workers
,
4198 int btrfs_map_bio(struct btrfs_root
*root
, int rw
, struct bio
*bio
,
4199 int mirror_num
, int async_submit
)
4201 struct btrfs_mapping_tree
*map_tree
;
4202 struct btrfs_device
*dev
;
4203 struct bio
*first_bio
= bio
;
4204 u64 logical
= (u64
)bio
->bi_sector
<< 9;
4210 struct btrfs_bio
*bbio
= NULL
;
4212 length
= bio
->bi_size
;
4213 map_tree
= &root
->fs_info
->mapping_tree
;
4214 map_length
= length
;
4216 ret
= btrfs_map_block(map_tree
, rw
, logical
, &map_length
, &bbio
,
4218 if (ret
) /* -ENOMEM */
4221 total_devs
= bbio
->num_stripes
;
4222 if (map_length
< length
) {
4223 printk(KERN_CRIT
"mapping failed logical %llu bio len %llu "
4224 "len %llu\n", (unsigned long long)logical
,
4225 (unsigned long long)length
,
4226 (unsigned long long)map_length
);
4230 bbio
->orig_bio
= first_bio
;
4231 bbio
->private = first_bio
->bi_private
;
4232 bbio
->end_io
= first_bio
->bi_end_io
;
4233 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
4235 while (dev_nr
< total_devs
) {
4236 if (dev_nr
< total_devs
- 1) {
4237 bio
= bio_clone(first_bio
, GFP_NOFS
);
4238 BUG_ON(!bio
); /* -ENOMEM */
4242 bio
->bi_private
= bbio
;
4243 bio
->bi_private
= merge_stripe_index_into_bio_private(
4244 bio
->bi_private
, (unsigned int)dev_nr
);
4245 bio
->bi_end_io
= btrfs_end_bio
;
4246 bio
->bi_sector
= bbio
->stripes
[dev_nr
].physical
>> 9;
4247 dev
= bbio
->stripes
[dev_nr
].dev
;
4248 if (dev
&& dev
->bdev
&& (rw
!= WRITE
|| dev
->writeable
)) {
4250 struct rcu_string
*name
;
4253 name
= rcu_dereference(dev
->name
);
4254 pr_debug("btrfs_map_bio: rw %d, secor=%llu, dev=%lu "
4255 "(%s id %llu), size=%u\n", rw
,
4256 (u64
)bio
->bi_sector
, (u_long
)dev
->bdev
->bd_dev
,
4257 name
->str
, dev
->devid
, bio
->bi_size
);
4260 bio
->bi_bdev
= dev
->bdev
;
4262 schedule_bio(root
, dev
, rw
, bio
);
4264 btrfsic_submit_bio(rw
, bio
);
4266 bio
->bi_bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4267 bio
->bi_sector
= logical
>> 9;
4268 bio_endio(bio
, -EIO
);
4275 struct btrfs_device
*btrfs_find_device(struct btrfs_root
*root
, u64 devid
,
4278 struct btrfs_device
*device
;
4279 struct btrfs_fs_devices
*cur_devices
;
4281 cur_devices
= root
->fs_info
->fs_devices
;
4282 while (cur_devices
) {
4284 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
4285 device
= __find_device(&cur_devices
->devices
,
4290 cur_devices
= cur_devices
->seed
;
4295 static struct btrfs_device
*add_missing_dev(struct btrfs_root
*root
,
4296 u64 devid
, u8
*dev_uuid
)
4298 struct btrfs_device
*device
;
4299 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
4301 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
4304 list_add(&device
->dev_list
,
4305 &fs_devices
->devices
);
4306 device
->dev_root
= root
->fs_info
->dev_root
;
4307 device
->devid
= devid
;
4308 device
->work
.func
= pending_bios_fn
;
4309 device
->fs_devices
= fs_devices
;
4310 device
->missing
= 1;
4311 fs_devices
->num_devices
++;
4312 fs_devices
->missing_devices
++;
4313 spin_lock_init(&device
->io_lock
);
4314 INIT_LIST_HEAD(&device
->dev_alloc_list
);
4315 memcpy(device
->uuid
, dev_uuid
, BTRFS_UUID_SIZE
);
4319 static int read_one_chunk(struct btrfs_root
*root
, struct btrfs_key
*key
,
4320 struct extent_buffer
*leaf
,
4321 struct btrfs_chunk
*chunk
)
4323 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
4324 struct map_lookup
*map
;
4325 struct extent_map
*em
;
4329 u8 uuid
[BTRFS_UUID_SIZE
];
4334 logical
= key
->offset
;
4335 length
= btrfs_chunk_length(leaf
, chunk
);
4337 read_lock(&map_tree
->map_tree
.lock
);
4338 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
4339 read_unlock(&map_tree
->map_tree
.lock
);
4341 /* already mapped? */
4342 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
4343 free_extent_map(em
);
4346 free_extent_map(em
);
4349 em
= alloc_extent_map();
4352 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
4353 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
4355 free_extent_map(em
);
4359 em
->bdev
= (struct block_device
*)map
;
4360 em
->start
= logical
;
4362 em
->block_start
= 0;
4363 em
->block_len
= em
->len
;
4365 map
->num_stripes
= num_stripes
;
4366 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
4367 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
4368 map
->sector_size
= btrfs_chunk_sector_size(leaf
, chunk
);
4369 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
4370 map
->type
= btrfs_chunk_type(leaf
, chunk
);
4371 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
4372 for (i
= 0; i
< num_stripes
; i
++) {
4373 map
->stripes
[i
].physical
=
4374 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
4375 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
4376 read_extent_buffer(leaf
, uuid
, (unsigned long)
4377 btrfs_stripe_dev_uuid_nr(chunk
, i
),
4379 map
->stripes
[i
].dev
= btrfs_find_device(root
, devid
, uuid
,
4381 if (!map
->stripes
[i
].dev
&& !btrfs_test_opt(root
, DEGRADED
)) {
4383 free_extent_map(em
);
4386 if (!map
->stripes
[i
].dev
) {
4387 map
->stripes
[i
].dev
=
4388 add_missing_dev(root
, devid
, uuid
);
4389 if (!map
->stripes
[i
].dev
) {
4391 free_extent_map(em
);
4395 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
4398 write_lock(&map_tree
->map_tree
.lock
);
4399 ret
= add_extent_mapping(&map_tree
->map_tree
, em
);
4400 write_unlock(&map_tree
->map_tree
.lock
);
4401 BUG_ON(ret
); /* Tree corruption */
4402 free_extent_map(em
);
4407 static void fill_device_from_item(struct extent_buffer
*leaf
,
4408 struct btrfs_dev_item
*dev_item
,
4409 struct btrfs_device
*device
)
4413 device
->devid
= btrfs_device_id(leaf
, dev_item
);
4414 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
4415 device
->total_bytes
= device
->disk_total_bytes
;
4416 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
4417 device
->type
= btrfs_device_type(leaf
, dev_item
);
4418 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
4419 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
4420 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
4422 ptr
= (unsigned long)btrfs_device_uuid(dev_item
);
4423 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
4426 static int open_seed_devices(struct btrfs_root
*root
, u8
*fsid
)
4428 struct btrfs_fs_devices
*fs_devices
;
4431 BUG_ON(!mutex_is_locked(&uuid_mutex
));
4433 fs_devices
= root
->fs_info
->fs_devices
->seed
;
4434 while (fs_devices
) {
4435 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
4439 fs_devices
= fs_devices
->seed
;
4442 fs_devices
= find_fsid(fsid
);
4448 fs_devices
= clone_fs_devices(fs_devices
);
4449 if (IS_ERR(fs_devices
)) {
4450 ret
= PTR_ERR(fs_devices
);
4454 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
4455 root
->fs_info
->bdev_holder
);
4457 free_fs_devices(fs_devices
);
4461 if (!fs_devices
->seeding
) {
4462 __btrfs_close_devices(fs_devices
);
4463 free_fs_devices(fs_devices
);
4468 fs_devices
->seed
= root
->fs_info
->fs_devices
->seed
;
4469 root
->fs_info
->fs_devices
->seed
= fs_devices
;
4474 static int read_one_dev(struct btrfs_root
*root
,
4475 struct extent_buffer
*leaf
,
4476 struct btrfs_dev_item
*dev_item
)
4478 struct btrfs_device
*device
;
4481 u8 fs_uuid
[BTRFS_UUID_SIZE
];
4482 u8 dev_uuid
[BTRFS_UUID_SIZE
];
4484 devid
= btrfs_device_id(leaf
, dev_item
);
4485 read_extent_buffer(leaf
, dev_uuid
,
4486 (unsigned long)btrfs_device_uuid(dev_item
),
4488 read_extent_buffer(leaf
, fs_uuid
,
4489 (unsigned long)btrfs_device_fsid(dev_item
),
4492 if (memcmp(fs_uuid
, root
->fs_info
->fsid
, BTRFS_UUID_SIZE
)) {
4493 ret
= open_seed_devices(root
, fs_uuid
);
4494 if (ret
&& !btrfs_test_opt(root
, DEGRADED
))
4498 device
= btrfs_find_device(root
, devid
, dev_uuid
, fs_uuid
);
4499 if (!device
|| !device
->bdev
) {
4500 if (!btrfs_test_opt(root
, DEGRADED
))
4504 printk(KERN_WARNING
"warning devid %llu missing\n",
4505 (unsigned long long)devid
);
4506 device
= add_missing_dev(root
, devid
, dev_uuid
);
4509 } else if (!device
->missing
) {
4511 * this happens when a device that was properly setup
4512 * in the device info lists suddenly goes bad.
4513 * device->bdev is NULL, and so we have to set
4514 * device->missing to one here
4516 root
->fs_info
->fs_devices
->missing_devices
++;
4517 device
->missing
= 1;
4521 if (device
->fs_devices
!= root
->fs_info
->fs_devices
) {
4522 BUG_ON(device
->writeable
);
4523 if (device
->generation
!=
4524 btrfs_device_generation(leaf
, dev_item
))
4528 fill_device_from_item(leaf
, dev_item
, device
);
4529 device
->dev_root
= root
->fs_info
->dev_root
;
4530 device
->in_fs_metadata
= 1;
4531 if (device
->writeable
) {
4532 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
4533 spin_lock(&root
->fs_info
->free_chunk_lock
);
4534 root
->fs_info
->free_chunk_space
+= device
->total_bytes
-
4536 spin_unlock(&root
->fs_info
->free_chunk_lock
);
4542 int btrfs_read_sys_array(struct btrfs_root
*root
)
4544 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
4545 struct extent_buffer
*sb
;
4546 struct btrfs_disk_key
*disk_key
;
4547 struct btrfs_chunk
*chunk
;
4549 unsigned long sb_ptr
;
4555 struct btrfs_key key
;
4557 sb
= btrfs_find_create_tree_block(root
, BTRFS_SUPER_INFO_OFFSET
,
4558 BTRFS_SUPER_INFO_SIZE
);
4561 btrfs_set_buffer_uptodate(sb
);
4562 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
4564 * The sb extent buffer is artifical and just used to read the system array.
4565 * btrfs_set_buffer_uptodate() call does not properly mark all it's
4566 * pages up-to-date when the page is larger: extent does not cover the
4567 * whole page and consequently check_page_uptodate does not find all
4568 * the page's extents up-to-date (the hole beyond sb),
4569 * write_extent_buffer then triggers a WARN_ON.
4571 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
4572 * but sb spans only this function. Add an explicit SetPageUptodate call
4573 * to silence the warning eg. on PowerPC 64.
4575 if (PAGE_CACHE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
4576 SetPageUptodate(sb
->pages
[0]);
4578 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
4579 array_size
= btrfs_super_sys_array_size(super_copy
);
4581 ptr
= super_copy
->sys_chunk_array
;
4582 sb_ptr
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
4585 while (cur
< array_size
) {
4586 disk_key
= (struct btrfs_disk_key
*)ptr
;
4587 btrfs_disk_key_to_cpu(&key
, disk_key
);
4589 len
= sizeof(*disk_key
); ptr
+= len
;
4593 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
4594 chunk
= (struct btrfs_chunk
*)sb_ptr
;
4595 ret
= read_one_chunk(root
, &key
, sb
, chunk
);
4598 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
4599 len
= btrfs_chunk_item_size(num_stripes
);
4608 free_extent_buffer(sb
);
4612 struct btrfs_device
*btrfs_find_device_for_logical(struct btrfs_root
*root
,
4613 u64 logical
, int mirror_num
)
4615 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
4618 struct btrfs_bio
*bbio
= NULL
;
4619 struct btrfs_device
*device
;
4621 BUG_ON(mirror_num
== 0);
4622 ret
= btrfs_map_block(map_tree
, WRITE
, logical
, &map_length
, &bbio
,
4625 BUG_ON(bbio
!= NULL
);
4628 BUG_ON(mirror_num
!= bbio
->mirror_num
);
4629 device
= bbio
->stripes
[mirror_num
- 1].dev
;
4634 int btrfs_read_chunk_tree(struct btrfs_root
*root
)
4636 struct btrfs_path
*path
;
4637 struct extent_buffer
*leaf
;
4638 struct btrfs_key key
;
4639 struct btrfs_key found_key
;
4643 root
= root
->fs_info
->chunk_root
;
4645 path
= btrfs_alloc_path();
4649 mutex_lock(&uuid_mutex
);
4652 /* first we search for all of the device items, and then we
4653 * read in all of the chunk items. This way we can create chunk
4654 * mappings that reference all of the devices that are afound
4656 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
4660 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4664 leaf
= path
->nodes
[0];
4665 slot
= path
->slots
[0];
4666 if (slot
>= btrfs_header_nritems(leaf
)) {
4667 ret
= btrfs_next_leaf(root
, path
);
4674 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
4675 if (key
.objectid
== BTRFS_DEV_ITEMS_OBJECTID
) {
4676 if (found_key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
)
4678 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
4679 struct btrfs_dev_item
*dev_item
;
4680 dev_item
= btrfs_item_ptr(leaf
, slot
,
4681 struct btrfs_dev_item
);
4682 ret
= read_one_dev(root
, leaf
, dev_item
);
4686 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
4687 struct btrfs_chunk
*chunk
;
4688 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
4689 ret
= read_one_chunk(root
, &found_key
, leaf
, chunk
);
4695 if (key
.objectid
== BTRFS_DEV_ITEMS_OBJECTID
) {
4697 btrfs_release_path(path
);
4702 unlock_chunks(root
);
4703 mutex_unlock(&uuid_mutex
);
4705 btrfs_free_path(path
);
4709 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
4713 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
4714 btrfs_dev_stat_reset(dev
, i
);
4717 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
4719 struct btrfs_key key
;
4720 struct btrfs_key found_key
;
4721 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
4722 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
4723 struct extent_buffer
*eb
;
4726 struct btrfs_device
*device
;
4727 struct btrfs_path
*path
= NULL
;
4730 path
= btrfs_alloc_path();
4736 mutex_lock(&fs_devices
->device_list_mutex
);
4737 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
4739 struct btrfs_dev_stats_item
*ptr
;
4742 key
.type
= BTRFS_DEV_STATS_KEY
;
4743 key
.offset
= device
->devid
;
4744 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
4746 __btrfs_reset_dev_stats(device
);
4747 device
->dev_stats_valid
= 1;
4748 btrfs_release_path(path
);
4751 slot
= path
->slots
[0];
4752 eb
= path
->nodes
[0];
4753 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
4754 item_size
= btrfs_item_size_nr(eb
, slot
);
4756 ptr
= btrfs_item_ptr(eb
, slot
,
4757 struct btrfs_dev_stats_item
);
4759 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
4760 if (item_size
>= (1 + i
) * sizeof(__le64
))
4761 btrfs_dev_stat_set(device
, i
,
4762 btrfs_dev_stats_value(eb
, ptr
, i
));
4764 btrfs_dev_stat_reset(device
, i
);
4767 device
->dev_stats_valid
= 1;
4768 btrfs_dev_stat_print_on_load(device
);
4769 btrfs_release_path(path
);
4771 mutex_unlock(&fs_devices
->device_list_mutex
);
4774 btrfs_free_path(path
);
4775 return ret
< 0 ? ret
: 0;
4778 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
4779 struct btrfs_root
*dev_root
,
4780 struct btrfs_device
*device
)
4782 struct btrfs_path
*path
;
4783 struct btrfs_key key
;
4784 struct extent_buffer
*eb
;
4785 struct btrfs_dev_stats_item
*ptr
;
4790 key
.type
= BTRFS_DEV_STATS_KEY
;
4791 key
.offset
= device
->devid
;
4793 path
= btrfs_alloc_path();
4795 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
4797 printk_in_rcu(KERN_WARNING
"btrfs: error %d while searching for dev_stats item for device %s!\n",
4798 ret
, rcu_str_deref(device
->name
));
4803 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
4804 /* need to delete old one and insert a new one */
4805 ret
= btrfs_del_item(trans
, dev_root
, path
);
4807 printk_in_rcu(KERN_WARNING
"btrfs: delete too small dev_stats item for device %s failed %d!\n",
4808 rcu_str_deref(device
->name
), ret
);
4815 /* need to insert a new item */
4816 btrfs_release_path(path
);
4817 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
4818 &key
, sizeof(*ptr
));
4820 printk_in_rcu(KERN_WARNING
"btrfs: insert dev_stats item for device %s failed %d!\n",
4821 rcu_str_deref(device
->name
), ret
);
4826 eb
= path
->nodes
[0];
4827 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
4828 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
4829 btrfs_set_dev_stats_value(eb
, ptr
, i
,
4830 btrfs_dev_stat_read(device
, i
));
4831 btrfs_mark_buffer_dirty(eb
);
4834 btrfs_free_path(path
);
4839 * called from commit_transaction. Writes all changed device stats to disk.
4841 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
4842 struct btrfs_fs_info
*fs_info
)
4844 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
4845 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
4846 struct btrfs_device
*device
;
4849 mutex_lock(&fs_devices
->device_list_mutex
);
4850 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
4851 if (!device
->dev_stats_valid
|| !device
->dev_stats_dirty
)
4854 ret
= update_dev_stat_item(trans
, dev_root
, device
);
4856 device
->dev_stats_dirty
= 0;
4858 mutex_unlock(&fs_devices
->device_list_mutex
);
4863 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
4865 btrfs_dev_stat_inc(dev
, index
);
4866 btrfs_dev_stat_print_on_error(dev
);
4869 void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
4871 if (!dev
->dev_stats_valid
)
4873 printk_ratelimited_in_rcu(KERN_ERR
4874 "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
4875 rcu_str_deref(dev
->name
),
4876 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
4877 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
4878 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
4879 btrfs_dev_stat_read(dev
,
4880 BTRFS_DEV_STAT_CORRUPTION_ERRS
),
4881 btrfs_dev_stat_read(dev
,
4882 BTRFS_DEV_STAT_GENERATION_ERRS
));
4885 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
4889 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
4890 if (btrfs_dev_stat_read(dev
, i
) != 0)
4892 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
4893 return; /* all values == 0, suppress message */
4895 printk_in_rcu(KERN_INFO
"btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
4896 rcu_str_deref(dev
->name
),
4897 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
4898 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
4899 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
4900 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
4901 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
4904 int btrfs_get_dev_stats(struct btrfs_root
*root
,
4905 struct btrfs_ioctl_get_dev_stats
*stats
)
4907 struct btrfs_device
*dev
;
4908 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
4911 mutex_lock(&fs_devices
->device_list_mutex
);
4912 dev
= btrfs_find_device(root
, stats
->devid
, NULL
, NULL
);
4913 mutex_unlock(&fs_devices
->device_list_mutex
);
4917 "btrfs: get dev_stats failed, device not found\n");
4919 } else if (!dev
->dev_stats_valid
) {
4921 "btrfs: get dev_stats failed, not yet valid\n");
4923 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
4924 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
4925 if (stats
->nr_items
> i
)
4927 btrfs_dev_stat_read_and_reset(dev
, i
);
4929 btrfs_dev_stat_reset(dev
, i
);
4932 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
4933 if (stats
->nr_items
> i
)
4934 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
4936 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
4937 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;