2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/ratelimit.h>
27 #include <linux/kthread.h>
28 #include <linux/raid/pq.h>
29 #include <asm/div64.h>
32 #include "extent_map.h"
34 #include "transaction.h"
35 #include "print-tree.h"
38 #include "async-thread.h"
39 #include "check-integrity.h"
40 #include "rcu-string.h"
42 #include "dev-replace.h"
44 static int init_first_rw_device(struct btrfs_trans_handle
*trans
,
45 struct btrfs_root
*root
,
46 struct btrfs_device
*device
);
47 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
);
48 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
);
49 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*device
);
51 static DEFINE_MUTEX(uuid_mutex
);
52 static LIST_HEAD(fs_uuids
);
54 static void lock_chunks(struct btrfs_root
*root
)
56 mutex_lock(&root
->fs_info
->chunk_mutex
);
59 static void unlock_chunks(struct btrfs_root
*root
)
61 mutex_unlock(&root
->fs_info
->chunk_mutex
);
64 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
66 struct btrfs_device
*device
;
67 WARN_ON(fs_devices
->opened
);
68 while (!list_empty(&fs_devices
->devices
)) {
69 device
= list_entry(fs_devices
->devices
.next
,
70 struct btrfs_device
, dev_list
);
71 list_del(&device
->dev_list
);
72 rcu_string_free(device
->name
);
78 static void btrfs_kobject_uevent(struct block_device
*bdev
,
79 enum kobject_action action
)
83 ret
= kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, action
);
85 pr_warn("Sending event '%d' to kobject: '%s' (%p): failed\n",
87 kobject_name(&disk_to_dev(bdev
->bd_disk
)->kobj
),
88 &disk_to_dev(bdev
->bd_disk
)->kobj
);
91 void btrfs_cleanup_fs_uuids(void)
93 struct btrfs_fs_devices
*fs_devices
;
95 while (!list_empty(&fs_uuids
)) {
96 fs_devices
= list_entry(fs_uuids
.next
,
97 struct btrfs_fs_devices
, list
);
98 list_del(&fs_devices
->list
);
99 free_fs_devices(fs_devices
);
103 static noinline
struct btrfs_device
*__find_device(struct list_head
*head
,
106 struct btrfs_device
*dev
;
108 list_for_each_entry(dev
, head
, dev_list
) {
109 if (dev
->devid
== devid
&&
110 (!uuid
|| !memcmp(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
))) {
117 static noinline
struct btrfs_fs_devices
*find_fsid(u8
*fsid
)
119 struct btrfs_fs_devices
*fs_devices
;
121 list_for_each_entry(fs_devices
, &fs_uuids
, list
) {
122 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
129 btrfs_get_bdev_and_sb(const char *device_path
, fmode_t flags
, void *holder
,
130 int flush
, struct block_device
**bdev
,
131 struct buffer_head
**bh
)
135 *bdev
= blkdev_get_by_path(device_path
, flags
, holder
);
138 ret
= PTR_ERR(*bdev
);
139 printk(KERN_INFO
"btrfs: open %s failed\n", device_path
);
144 filemap_write_and_wait((*bdev
)->bd_inode
->i_mapping
);
145 ret
= set_blocksize(*bdev
, 4096);
147 blkdev_put(*bdev
, flags
);
150 invalidate_bdev(*bdev
);
151 *bh
= btrfs_read_dev_super(*bdev
);
154 blkdev_put(*bdev
, flags
);
166 static void requeue_list(struct btrfs_pending_bios
*pending_bios
,
167 struct bio
*head
, struct bio
*tail
)
170 struct bio
*old_head
;
172 old_head
= pending_bios
->head
;
173 pending_bios
->head
= head
;
174 if (pending_bios
->tail
)
175 tail
->bi_next
= old_head
;
177 pending_bios
->tail
= tail
;
181 * we try to collect pending bios for a device so we don't get a large
182 * number of procs sending bios down to the same device. This greatly
183 * improves the schedulers ability to collect and merge the bios.
185 * But, it also turns into a long list of bios to process and that is sure
186 * to eventually make the worker thread block. The solution here is to
187 * make some progress and then put this work struct back at the end of
188 * the list if the block device is congested. This way, multiple devices
189 * can make progress from a single worker thread.
191 static noinline
void run_scheduled_bios(struct btrfs_device
*device
)
194 struct backing_dev_info
*bdi
;
195 struct btrfs_fs_info
*fs_info
;
196 struct btrfs_pending_bios
*pending_bios
;
200 unsigned long num_run
;
201 unsigned long batch_run
= 0;
203 unsigned long last_waited
= 0;
205 int sync_pending
= 0;
206 struct blk_plug plug
;
209 * this function runs all the bios we've collected for
210 * a particular device. We don't want to wander off to
211 * another device without first sending all of these down.
212 * So, setup a plug here and finish it off before we return
214 blk_start_plug(&plug
);
216 bdi
= blk_get_backing_dev_info(device
->bdev
);
217 fs_info
= device
->dev_root
->fs_info
;
218 limit
= btrfs_async_submit_limit(fs_info
);
219 limit
= limit
* 2 / 3;
222 spin_lock(&device
->io_lock
);
227 /* take all the bios off the list at once and process them
228 * later on (without the lock held). But, remember the
229 * tail and other pointers so the bios can be properly reinserted
230 * into the list if we hit congestion
232 if (!force_reg
&& device
->pending_sync_bios
.head
) {
233 pending_bios
= &device
->pending_sync_bios
;
236 pending_bios
= &device
->pending_bios
;
240 pending
= pending_bios
->head
;
241 tail
= pending_bios
->tail
;
242 WARN_ON(pending
&& !tail
);
245 * if pending was null this time around, no bios need processing
246 * at all and we can stop. Otherwise it'll loop back up again
247 * and do an additional check so no bios are missed.
249 * device->running_pending is used to synchronize with the
252 if (device
->pending_sync_bios
.head
== NULL
&&
253 device
->pending_bios
.head
== NULL
) {
255 device
->running_pending
= 0;
258 device
->running_pending
= 1;
261 pending_bios
->head
= NULL
;
262 pending_bios
->tail
= NULL
;
264 spin_unlock(&device
->io_lock
);
269 /* we want to work on both lists, but do more bios on the
270 * sync list than the regular list
273 pending_bios
!= &device
->pending_sync_bios
&&
274 device
->pending_sync_bios
.head
) ||
275 (num_run
> 64 && pending_bios
== &device
->pending_sync_bios
&&
276 device
->pending_bios
.head
)) {
277 spin_lock(&device
->io_lock
);
278 requeue_list(pending_bios
, pending
, tail
);
283 pending
= pending
->bi_next
;
286 if (atomic_dec_return(&fs_info
->nr_async_bios
) < limit
&&
287 waitqueue_active(&fs_info
->async_submit_wait
))
288 wake_up(&fs_info
->async_submit_wait
);
290 BUG_ON(atomic_read(&cur
->bi_cnt
) == 0);
293 * if we're doing the sync list, record that our
294 * plug has some sync requests on it
296 * If we're doing the regular list and there are
297 * sync requests sitting around, unplug before
300 if (pending_bios
== &device
->pending_sync_bios
) {
302 } else if (sync_pending
) {
303 blk_finish_plug(&plug
);
304 blk_start_plug(&plug
);
308 btrfsic_submit_bio(cur
->bi_rw
, cur
);
315 * we made progress, there is more work to do and the bdi
316 * is now congested. Back off and let other work structs
319 if (pending
&& bdi_write_congested(bdi
) && batch_run
> 8 &&
320 fs_info
->fs_devices
->open_devices
> 1) {
321 struct io_context
*ioc
;
323 ioc
= current
->io_context
;
326 * the main goal here is that we don't want to
327 * block if we're going to be able to submit
328 * more requests without blocking.
330 * This code does two great things, it pokes into
331 * the elevator code from a filesystem _and_
332 * it makes assumptions about how batching works.
334 if (ioc
&& ioc
->nr_batch_requests
> 0 &&
335 time_before(jiffies
, ioc
->last_waited
+ HZ
/50UL) &&
337 ioc
->last_waited
== last_waited
)) {
339 * we want to go through our batch of
340 * requests and stop. So, we copy out
341 * the ioc->last_waited time and test
342 * against it before looping
344 last_waited
= ioc
->last_waited
;
349 spin_lock(&device
->io_lock
);
350 requeue_list(pending_bios
, pending
, tail
);
351 device
->running_pending
= 1;
353 spin_unlock(&device
->io_lock
);
354 btrfs_requeue_work(&device
->work
);
357 /* unplug every 64 requests just for good measure */
358 if (batch_run
% 64 == 0) {
359 blk_finish_plug(&plug
);
360 blk_start_plug(&plug
);
369 spin_lock(&device
->io_lock
);
370 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
372 spin_unlock(&device
->io_lock
);
375 blk_finish_plug(&plug
);
378 static void pending_bios_fn(struct btrfs_work
*work
)
380 struct btrfs_device
*device
;
382 device
= container_of(work
, struct btrfs_device
, work
);
383 run_scheduled_bios(device
);
386 static noinline
int device_list_add(const char *path
,
387 struct btrfs_super_block
*disk_super
,
388 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
390 struct btrfs_device
*device
;
391 struct btrfs_fs_devices
*fs_devices
;
392 struct rcu_string
*name
;
393 u64 found_transid
= btrfs_super_generation(disk_super
);
395 fs_devices
= find_fsid(disk_super
->fsid
);
397 fs_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
400 INIT_LIST_HEAD(&fs_devices
->devices
);
401 INIT_LIST_HEAD(&fs_devices
->alloc_list
);
402 list_add(&fs_devices
->list
, &fs_uuids
);
403 memcpy(fs_devices
->fsid
, disk_super
->fsid
, BTRFS_FSID_SIZE
);
404 fs_devices
->latest_devid
= devid
;
405 fs_devices
->latest_trans
= found_transid
;
406 mutex_init(&fs_devices
->device_list_mutex
);
409 device
= __find_device(&fs_devices
->devices
, devid
,
410 disk_super
->dev_item
.uuid
);
413 if (fs_devices
->opened
)
416 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
418 /* we can safely leave the fs_devices entry around */
421 device
->devid
= devid
;
422 device
->dev_stats_valid
= 0;
423 device
->work
.func
= pending_bios_fn
;
424 memcpy(device
->uuid
, disk_super
->dev_item
.uuid
,
426 spin_lock_init(&device
->io_lock
);
428 name
= rcu_string_strdup(path
, GFP_NOFS
);
433 rcu_assign_pointer(device
->name
, name
);
434 INIT_LIST_HEAD(&device
->dev_alloc_list
);
436 /* init readahead state */
437 spin_lock_init(&device
->reada_lock
);
438 device
->reada_curr_zone
= NULL
;
439 atomic_set(&device
->reada_in_flight
, 0);
440 device
->reada_next
= 0;
441 INIT_RADIX_TREE(&device
->reada_zones
, GFP_NOFS
& ~__GFP_WAIT
);
442 INIT_RADIX_TREE(&device
->reada_extents
, GFP_NOFS
& ~__GFP_WAIT
);
444 mutex_lock(&fs_devices
->device_list_mutex
);
445 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
446 mutex_unlock(&fs_devices
->device_list_mutex
);
448 device
->fs_devices
= fs_devices
;
449 fs_devices
->num_devices
++;
450 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
451 name
= rcu_string_strdup(path
, GFP_NOFS
);
454 rcu_string_free(device
->name
);
455 rcu_assign_pointer(device
->name
, name
);
456 if (device
->missing
) {
457 fs_devices
->missing_devices
--;
462 if (found_transid
> fs_devices
->latest_trans
) {
463 fs_devices
->latest_devid
= devid
;
464 fs_devices
->latest_trans
= found_transid
;
466 *fs_devices_ret
= fs_devices
;
470 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
472 struct btrfs_fs_devices
*fs_devices
;
473 struct btrfs_device
*device
;
474 struct btrfs_device
*orig_dev
;
476 fs_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
478 return ERR_PTR(-ENOMEM
);
480 INIT_LIST_HEAD(&fs_devices
->devices
);
481 INIT_LIST_HEAD(&fs_devices
->alloc_list
);
482 INIT_LIST_HEAD(&fs_devices
->list
);
483 mutex_init(&fs_devices
->device_list_mutex
);
484 fs_devices
->latest_devid
= orig
->latest_devid
;
485 fs_devices
->latest_trans
= orig
->latest_trans
;
486 fs_devices
->total_devices
= orig
->total_devices
;
487 memcpy(fs_devices
->fsid
, orig
->fsid
, sizeof(fs_devices
->fsid
));
489 /* We have held the volume lock, it is safe to get the devices. */
490 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
491 struct rcu_string
*name
;
493 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
498 * This is ok to do without rcu read locked because we hold the
499 * uuid mutex so nothing we touch in here is going to disappear.
501 name
= rcu_string_strdup(orig_dev
->name
->str
, GFP_NOFS
);
506 rcu_assign_pointer(device
->name
, name
);
508 device
->devid
= orig_dev
->devid
;
509 device
->work
.func
= pending_bios_fn
;
510 memcpy(device
->uuid
, orig_dev
->uuid
, sizeof(device
->uuid
));
511 spin_lock_init(&device
->io_lock
);
512 INIT_LIST_HEAD(&device
->dev_list
);
513 INIT_LIST_HEAD(&device
->dev_alloc_list
);
515 list_add(&device
->dev_list
, &fs_devices
->devices
);
516 device
->fs_devices
= fs_devices
;
517 fs_devices
->num_devices
++;
521 free_fs_devices(fs_devices
);
522 return ERR_PTR(-ENOMEM
);
525 void btrfs_close_extra_devices(struct btrfs_fs_info
*fs_info
,
526 struct btrfs_fs_devices
*fs_devices
, int step
)
528 struct btrfs_device
*device
, *next
;
530 struct block_device
*latest_bdev
= NULL
;
531 u64 latest_devid
= 0;
532 u64 latest_transid
= 0;
534 mutex_lock(&uuid_mutex
);
536 /* This is the initialized path, it is safe to release the devices. */
537 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
538 if (device
->in_fs_metadata
) {
539 if (!device
->is_tgtdev_for_dev_replace
&&
541 device
->generation
> latest_transid
)) {
542 latest_devid
= device
->devid
;
543 latest_transid
= device
->generation
;
544 latest_bdev
= device
->bdev
;
549 if (device
->devid
== BTRFS_DEV_REPLACE_DEVID
) {
551 * In the first step, keep the device which has
552 * the correct fsid and the devid that is used
553 * for the dev_replace procedure.
554 * In the second step, the dev_replace state is
555 * read from the device tree and it is known
556 * whether the procedure is really active or
557 * not, which means whether this device is
558 * used or whether it should be removed.
560 if (step
== 0 || device
->is_tgtdev_for_dev_replace
) {
565 blkdev_put(device
->bdev
, device
->mode
);
567 fs_devices
->open_devices
--;
569 if (device
->writeable
) {
570 list_del_init(&device
->dev_alloc_list
);
571 device
->writeable
= 0;
572 if (!device
->is_tgtdev_for_dev_replace
)
573 fs_devices
->rw_devices
--;
575 list_del_init(&device
->dev_list
);
576 fs_devices
->num_devices
--;
577 rcu_string_free(device
->name
);
581 if (fs_devices
->seed
) {
582 fs_devices
= fs_devices
->seed
;
586 fs_devices
->latest_bdev
= latest_bdev
;
587 fs_devices
->latest_devid
= latest_devid
;
588 fs_devices
->latest_trans
= latest_transid
;
590 mutex_unlock(&uuid_mutex
);
593 static void __free_device(struct work_struct
*work
)
595 struct btrfs_device
*device
;
597 device
= container_of(work
, struct btrfs_device
, rcu_work
);
600 blkdev_put(device
->bdev
, device
->mode
);
602 rcu_string_free(device
->name
);
606 static void free_device(struct rcu_head
*head
)
608 struct btrfs_device
*device
;
610 device
= container_of(head
, struct btrfs_device
, rcu
);
612 INIT_WORK(&device
->rcu_work
, __free_device
);
613 schedule_work(&device
->rcu_work
);
616 static int __btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
618 struct btrfs_device
*device
;
620 if (--fs_devices
->opened
> 0)
623 mutex_lock(&fs_devices
->device_list_mutex
);
624 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
625 struct btrfs_device
*new_device
;
626 struct rcu_string
*name
;
629 fs_devices
->open_devices
--;
631 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
632 list_del_init(&device
->dev_alloc_list
);
633 fs_devices
->rw_devices
--;
636 if (device
->can_discard
)
637 fs_devices
->num_can_discard
--;
639 new_device
= kmalloc(sizeof(*new_device
), GFP_NOFS
);
640 BUG_ON(!new_device
); /* -ENOMEM */
641 memcpy(new_device
, device
, sizeof(*new_device
));
643 /* Safe because we are under uuid_mutex */
645 name
= rcu_string_strdup(device
->name
->str
, GFP_NOFS
);
646 BUG_ON(device
->name
&& !name
); /* -ENOMEM */
647 rcu_assign_pointer(new_device
->name
, name
);
649 new_device
->bdev
= NULL
;
650 new_device
->writeable
= 0;
651 new_device
->in_fs_metadata
= 0;
652 new_device
->can_discard
= 0;
653 spin_lock_init(&new_device
->io_lock
);
654 list_replace_rcu(&device
->dev_list
, &new_device
->dev_list
);
656 call_rcu(&device
->rcu
, free_device
);
658 mutex_unlock(&fs_devices
->device_list_mutex
);
660 WARN_ON(fs_devices
->open_devices
);
661 WARN_ON(fs_devices
->rw_devices
);
662 fs_devices
->opened
= 0;
663 fs_devices
->seeding
= 0;
668 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
670 struct btrfs_fs_devices
*seed_devices
= NULL
;
673 mutex_lock(&uuid_mutex
);
674 ret
= __btrfs_close_devices(fs_devices
);
675 if (!fs_devices
->opened
) {
676 seed_devices
= fs_devices
->seed
;
677 fs_devices
->seed
= NULL
;
679 mutex_unlock(&uuid_mutex
);
681 while (seed_devices
) {
682 fs_devices
= seed_devices
;
683 seed_devices
= fs_devices
->seed
;
684 __btrfs_close_devices(fs_devices
);
685 free_fs_devices(fs_devices
);
690 static int __btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
691 fmode_t flags
, void *holder
)
693 struct request_queue
*q
;
694 struct block_device
*bdev
;
695 struct list_head
*head
= &fs_devices
->devices
;
696 struct btrfs_device
*device
;
697 struct block_device
*latest_bdev
= NULL
;
698 struct buffer_head
*bh
;
699 struct btrfs_super_block
*disk_super
;
700 u64 latest_devid
= 0;
701 u64 latest_transid
= 0;
708 list_for_each_entry(device
, head
, dev_list
) {
714 ret
= btrfs_get_bdev_and_sb(device
->name
->str
, flags
, holder
, 1,
719 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
720 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
721 if (devid
!= device
->devid
)
724 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
,
728 device
->generation
= btrfs_super_generation(disk_super
);
729 if (!latest_transid
|| device
->generation
> latest_transid
) {
730 latest_devid
= devid
;
731 latest_transid
= device
->generation
;
735 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
736 device
->writeable
= 0;
738 device
->writeable
= !bdev_read_only(bdev
);
742 q
= bdev_get_queue(bdev
);
743 if (blk_queue_discard(q
)) {
744 device
->can_discard
= 1;
745 fs_devices
->num_can_discard
++;
749 device
->in_fs_metadata
= 0;
750 device
->mode
= flags
;
752 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
753 fs_devices
->rotating
= 1;
755 fs_devices
->open_devices
++;
756 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
757 fs_devices
->rw_devices
++;
758 list_add(&device
->dev_alloc_list
,
759 &fs_devices
->alloc_list
);
766 blkdev_put(bdev
, flags
);
769 if (fs_devices
->open_devices
== 0) {
773 fs_devices
->seeding
= seeding
;
774 fs_devices
->opened
= 1;
775 fs_devices
->latest_bdev
= latest_bdev
;
776 fs_devices
->latest_devid
= latest_devid
;
777 fs_devices
->latest_trans
= latest_transid
;
778 fs_devices
->total_rw_bytes
= 0;
783 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
784 fmode_t flags
, void *holder
)
788 mutex_lock(&uuid_mutex
);
789 if (fs_devices
->opened
) {
790 fs_devices
->opened
++;
793 ret
= __btrfs_open_devices(fs_devices
, flags
, holder
);
795 mutex_unlock(&uuid_mutex
);
800 * Look for a btrfs signature on a device. This may be called out of the mount path
801 * and we are not allowed to call set_blocksize during the scan. The superblock
802 * is read via pagecache
804 int btrfs_scan_one_device(const char *path
, fmode_t flags
, void *holder
,
805 struct btrfs_fs_devices
**fs_devices_ret
)
807 struct btrfs_super_block
*disk_super
;
808 struct block_device
*bdev
;
819 * we would like to check all the supers, but that would make
820 * a btrfs mount succeed after a mkfs from a different FS.
821 * So, we need to add a special mount option to scan for
822 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
824 bytenr
= btrfs_sb_offset(0);
826 mutex_lock(&uuid_mutex
);
828 bdev
= blkdev_get_by_path(path
, flags
, holder
);
832 printk(KERN_INFO
"btrfs: open %s failed\n", path
);
836 /* make sure our super fits in the device */
837 if (bytenr
+ PAGE_CACHE_SIZE
>= i_size_read(bdev
->bd_inode
))
840 /* make sure our super fits in the page */
841 if (sizeof(*disk_super
) > PAGE_CACHE_SIZE
)
844 /* make sure our super doesn't straddle pages on disk */
845 index
= bytenr
>> PAGE_CACHE_SHIFT
;
846 if ((bytenr
+ sizeof(*disk_super
) - 1) >> PAGE_CACHE_SHIFT
!= index
)
849 /* pull in the page with our super */
850 page
= read_cache_page_gfp(bdev
->bd_inode
->i_mapping
,
853 if (IS_ERR_OR_NULL(page
))
858 /* align our pointer to the offset of the super block */
859 disk_super
= p
+ (bytenr
& ~PAGE_CACHE_MASK
);
861 if (btrfs_super_bytenr(disk_super
) != bytenr
||
862 disk_super
->magic
!= cpu_to_le64(BTRFS_MAGIC
))
865 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
866 transid
= btrfs_super_generation(disk_super
);
867 total_devices
= btrfs_super_num_devices(disk_super
);
869 if (disk_super
->label
[0]) {
870 if (disk_super
->label
[BTRFS_LABEL_SIZE
- 1])
871 disk_super
->label
[BTRFS_LABEL_SIZE
- 1] = '\0';
872 printk(KERN_INFO
"device label %s ", disk_super
->label
);
874 printk(KERN_INFO
"device fsid %pU ", disk_super
->fsid
);
877 printk(KERN_CONT
"devid %llu transid %llu %s\n",
878 (unsigned long long)devid
, (unsigned long long)transid
, path
);
880 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
881 if (!ret
&& fs_devices_ret
)
882 (*fs_devices_ret
)->total_devices
= total_devices
;
886 page_cache_release(page
);
889 blkdev_put(bdev
, flags
);
891 mutex_unlock(&uuid_mutex
);
895 /* helper to account the used device space in the range */
896 int btrfs_account_dev_extents_size(struct btrfs_device
*device
, u64 start
,
897 u64 end
, u64
*length
)
899 struct btrfs_key key
;
900 struct btrfs_root
*root
= device
->dev_root
;
901 struct btrfs_dev_extent
*dev_extent
;
902 struct btrfs_path
*path
;
906 struct extent_buffer
*l
;
910 if (start
>= device
->total_bytes
|| device
->is_tgtdev_for_dev_replace
)
913 path
= btrfs_alloc_path();
918 key
.objectid
= device
->devid
;
920 key
.type
= BTRFS_DEV_EXTENT_KEY
;
922 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
926 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
933 slot
= path
->slots
[0];
934 if (slot
>= btrfs_header_nritems(l
)) {
935 ret
= btrfs_next_leaf(root
, path
);
943 btrfs_item_key_to_cpu(l
, &key
, slot
);
945 if (key
.objectid
< device
->devid
)
948 if (key
.objectid
> device
->devid
)
951 if (btrfs_key_type(&key
) != BTRFS_DEV_EXTENT_KEY
)
954 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
955 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
957 if (key
.offset
<= start
&& extent_end
> end
) {
958 *length
= end
- start
+ 1;
960 } else if (key
.offset
<= start
&& extent_end
> start
)
961 *length
+= extent_end
- start
;
962 else if (key
.offset
> start
&& extent_end
<= end
)
963 *length
+= extent_end
- key
.offset
;
964 else if (key
.offset
> start
&& key
.offset
<= end
) {
965 *length
+= end
- key
.offset
+ 1;
967 } else if (key
.offset
> end
)
975 btrfs_free_path(path
);
980 * find_free_dev_extent - find free space in the specified device
981 * @device: the device which we search the free space in
982 * @num_bytes: the size of the free space that we need
983 * @start: store the start of the free space.
984 * @len: the size of the free space. that we find, or the size of the max
985 * free space if we don't find suitable free space
987 * this uses a pretty simple search, the expectation is that it is
988 * called very infrequently and that a given device has a small number
991 * @start is used to store the start of the free space if we find. But if we
992 * don't find suitable free space, it will be used to store the start position
993 * of the max free space.
995 * @len is used to store the size of the free space that we find.
996 * But if we don't find suitable free space, it is used to store the size of
997 * the max free space.
999 int find_free_dev_extent(struct btrfs_device
*device
, u64 num_bytes
,
1000 u64
*start
, u64
*len
)
1002 struct btrfs_key key
;
1003 struct btrfs_root
*root
= device
->dev_root
;
1004 struct btrfs_dev_extent
*dev_extent
;
1005 struct btrfs_path
*path
;
1011 u64 search_end
= device
->total_bytes
;
1014 struct extent_buffer
*l
;
1016 /* FIXME use last free of some kind */
1018 /* we don't want to overwrite the superblock on the drive,
1019 * so we make sure to start at an offset of at least 1MB
1021 search_start
= max(root
->fs_info
->alloc_start
, 1024ull * 1024);
1023 max_hole_start
= search_start
;
1027 if (search_start
>= search_end
|| device
->is_tgtdev_for_dev_replace
) {
1032 path
= btrfs_alloc_path();
1039 key
.objectid
= device
->devid
;
1040 key
.offset
= search_start
;
1041 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1043 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1047 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1054 slot
= path
->slots
[0];
1055 if (slot
>= btrfs_header_nritems(l
)) {
1056 ret
= btrfs_next_leaf(root
, path
);
1064 btrfs_item_key_to_cpu(l
, &key
, slot
);
1066 if (key
.objectid
< device
->devid
)
1069 if (key
.objectid
> device
->devid
)
1072 if (btrfs_key_type(&key
) != BTRFS_DEV_EXTENT_KEY
)
1075 if (key
.offset
> search_start
) {
1076 hole_size
= key
.offset
- search_start
;
1078 if (hole_size
> max_hole_size
) {
1079 max_hole_start
= search_start
;
1080 max_hole_size
= hole_size
;
1084 * If this free space is greater than which we need,
1085 * it must be the max free space that we have found
1086 * until now, so max_hole_start must point to the start
1087 * of this free space and the length of this free space
1088 * is stored in max_hole_size. Thus, we return
1089 * max_hole_start and max_hole_size and go back to the
1092 if (hole_size
>= num_bytes
) {
1098 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1099 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1101 if (extent_end
> search_start
)
1102 search_start
= extent_end
;
1109 * At this point, search_start should be the end of
1110 * allocated dev extents, and when shrinking the device,
1111 * search_end may be smaller than search_start.
1113 if (search_end
> search_start
)
1114 hole_size
= search_end
- search_start
;
1116 if (hole_size
> max_hole_size
) {
1117 max_hole_start
= search_start
;
1118 max_hole_size
= hole_size
;
1122 if (hole_size
< num_bytes
)
1128 btrfs_free_path(path
);
1130 *start
= max_hole_start
;
1132 *len
= max_hole_size
;
1136 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1137 struct btrfs_device
*device
,
1141 struct btrfs_path
*path
;
1142 struct btrfs_root
*root
= device
->dev_root
;
1143 struct btrfs_key key
;
1144 struct btrfs_key found_key
;
1145 struct extent_buffer
*leaf
= NULL
;
1146 struct btrfs_dev_extent
*extent
= NULL
;
1148 path
= btrfs_alloc_path();
1152 key
.objectid
= device
->devid
;
1154 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1156 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1158 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1159 BTRFS_DEV_EXTENT_KEY
);
1162 leaf
= path
->nodes
[0];
1163 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1164 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1165 struct btrfs_dev_extent
);
1166 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1167 btrfs_dev_extent_length(leaf
, extent
) < start
);
1169 btrfs_release_path(path
);
1171 } else if (ret
== 0) {
1172 leaf
= path
->nodes
[0];
1173 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1174 struct btrfs_dev_extent
);
1176 btrfs_error(root
->fs_info
, ret
, "Slot search failed");
1180 if (device
->bytes_used
> 0) {
1181 u64 len
= btrfs_dev_extent_length(leaf
, extent
);
1182 device
->bytes_used
-= len
;
1183 spin_lock(&root
->fs_info
->free_chunk_lock
);
1184 root
->fs_info
->free_chunk_space
+= len
;
1185 spin_unlock(&root
->fs_info
->free_chunk_lock
);
1187 ret
= btrfs_del_item(trans
, root
, path
);
1189 btrfs_error(root
->fs_info
, ret
,
1190 "Failed to remove dev extent item");
1193 btrfs_free_path(path
);
1197 int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1198 struct btrfs_device
*device
,
1199 u64 chunk_tree
, u64 chunk_objectid
,
1200 u64 chunk_offset
, u64 start
, u64 num_bytes
)
1203 struct btrfs_path
*path
;
1204 struct btrfs_root
*root
= device
->dev_root
;
1205 struct btrfs_dev_extent
*extent
;
1206 struct extent_buffer
*leaf
;
1207 struct btrfs_key key
;
1209 WARN_ON(!device
->in_fs_metadata
);
1210 WARN_ON(device
->is_tgtdev_for_dev_replace
);
1211 path
= btrfs_alloc_path();
1215 key
.objectid
= device
->devid
;
1217 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1218 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1223 leaf
= path
->nodes
[0];
1224 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1225 struct btrfs_dev_extent
);
1226 btrfs_set_dev_extent_chunk_tree(leaf
, extent
, chunk_tree
);
1227 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
, chunk_objectid
);
1228 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1230 write_extent_buffer(leaf
, root
->fs_info
->chunk_tree_uuid
,
1231 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent
),
1234 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1235 btrfs_mark_buffer_dirty(leaf
);
1237 btrfs_free_path(path
);
1241 static noinline
int find_next_chunk(struct btrfs_root
*root
,
1242 u64 objectid
, u64
*offset
)
1244 struct btrfs_path
*path
;
1246 struct btrfs_key key
;
1247 struct btrfs_chunk
*chunk
;
1248 struct btrfs_key found_key
;
1250 path
= btrfs_alloc_path();
1254 key
.objectid
= objectid
;
1255 key
.offset
= (u64
)-1;
1256 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
1258 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1262 BUG_ON(ret
== 0); /* Corruption */
1264 ret
= btrfs_previous_item(root
, path
, 0, BTRFS_CHUNK_ITEM_KEY
);
1268 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1270 if (found_key
.objectid
!= objectid
)
1273 chunk
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1274 struct btrfs_chunk
);
1275 *offset
= found_key
.offset
+
1276 btrfs_chunk_length(path
->nodes
[0], chunk
);
1281 btrfs_free_path(path
);
1285 static noinline
int find_next_devid(struct btrfs_root
*root
, u64
*objectid
)
1288 struct btrfs_key key
;
1289 struct btrfs_key found_key
;
1290 struct btrfs_path
*path
;
1292 root
= root
->fs_info
->chunk_root
;
1294 path
= btrfs_alloc_path();
1298 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1299 key
.type
= BTRFS_DEV_ITEM_KEY
;
1300 key
.offset
= (u64
)-1;
1302 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1306 BUG_ON(ret
== 0); /* Corruption */
1308 ret
= btrfs_previous_item(root
, path
, BTRFS_DEV_ITEMS_OBJECTID
,
1309 BTRFS_DEV_ITEM_KEY
);
1313 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1315 *objectid
= found_key
.offset
+ 1;
1319 btrfs_free_path(path
);
1324 * the device information is stored in the chunk root
1325 * the btrfs_device struct should be fully filled in
1327 int btrfs_add_device(struct btrfs_trans_handle
*trans
,
1328 struct btrfs_root
*root
,
1329 struct btrfs_device
*device
)
1332 struct btrfs_path
*path
;
1333 struct btrfs_dev_item
*dev_item
;
1334 struct extent_buffer
*leaf
;
1335 struct btrfs_key key
;
1338 root
= root
->fs_info
->chunk_root
;
1340 path
= btrfs_alloc_path();
1344 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1345 key
.type
= BTRFS_DEV_ITEM_KEY
;
1346 key
.offset
= device
->devid
;
1348 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1353 leaf
= path
->nodes
[0];
1354 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1356 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1357 btrfs_set_device_generation(leaf
, dev_item
, 0);
1358 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1359 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1360 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1361 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1362 btrfs_set_device_total_bytes(leaf
, dev_item
, device
->total_bytes
);
1363 btrfs_set_device_bytes_used(leaf
, dev_item
, device
->bytes_used
);
1364 btrfs_set_device_group(leaf
, dev_item
, 0);
1365 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1366 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1367 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1369 ptr
= (unsigned long)btrfs_device_uuid(dev_item
);
1370 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1371 ptr
= (unsigned long)btrfs_device_fsid(dev_item
);
1372 write_extent_buffer(leaf
, root
->fs_info
->fsid
, ptr
, BTRFS_UUID_SIZE
);
1373 btrfs_mark_buffer_dirty(leaf
);
1377 btrfs_free_path(path
);
1381 static int btrfs_rm_dev_item(struct btrfs_root
*root
,
1382 struct btrfs_device
*device
)
1385 struct btrfs_path
*path
;
1386 struct btrfs_key key
;
1387 struct btrfs_trans_handle
*trans
;
1389 root
= root
->fs_info
->chunk_root
;
1391 path
= btrfs_alloc_path();
1395 trans
= btrfs_start_transaction(root
, 0);
1396 if (IS_ERR(trans
)) {
1397 btrfs_free_path(path
);
1398 return PTR_ERR(trans
);
1400 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1401 key
.type
= BTRFS_DEV_ITEM_KEY
;
1402 key
.offset
= device
->devid
;
1405 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1414 ret
= btrfs_del_item(trans
, root
, path
);
1418 btrfs_free_path(path
);
1419 unlock_chunks(root
);
1420 btrfs_commit_transaction(trans
, root
);
1424 int btrfs_rm_device(struct btrfs_root
*root
, char *device_path
)
1426 struct btrfs_device
*device
;
1427 struct btrfs_device
*next_device
;
1428 struct block_device
*bdev
;
1429 struct buffer_head
*bh
= NULL
;
1430 struct btrfs_super_block
*disk_super
;
1431 struct btrfs_fs_devices
*cur_devices
;
1438 bool clear_super
= false;
1440 mutex_lock(&uuid_mutex
);
1443 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
1445 all_avail
= root
->fs_info
->avail_data_alloc_bits
|
1446 root
->fs_info
->avail_system_alloc_bits
|
1447 root
->fs_info
->avail_metadata_alloc_bits
;
1448 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
1450 num_devices
= root
->fs_info
->fs_devices
->num_devices
;
1451 btrfs_dev_replace_lock(&root
->fs_info
->dev_replace
);
1452 if (btrfs_dev_replace_is_ongoing(&root
->fs_info
->dev_replace
)) {
1453 WARN_ON(num_devices
< 1);
1456 btrfs_dev_replace_unlock(&root
->fs_info
->dev_replace
);
1458 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID10
) && num_devices
<= 4) {
1459 printk(KERN_ERR
"btrfs: unable to go below four devices "
1465 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID1
) && num_devices
<= 2) {
1466 printk(KERN_ERR
"btrfs: unable to go below two "
1467 "devices on raid1\n");
1472 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID5
) &&
1473 root
->fs_info
->fs_devices
->rw_devices
<= 2) {
1474 printk(KERN_ERR
"btrfs: unable to go below two "
1475 "devices on raid5\n");
1479 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID6
) &&
1480 root
->fs_info
->fs_devices
->rw_devices
<= 3) {
1481 printk(KERN_ERR
"btrfs: unable to go below three "
1482 "devices on raid6\n");
1487 if (strcmp(device_path
, "missing") == 0) {
1488 struct list_head
*devices
;
1489 struct btrfs_device
*tmp
;
1492 devices
= &root
->fs_info
->fs_devices
->devices
;
1494 * It is safe to read the devices since the volume_mutex
1497 list_for_each_entry(tmp
, devices
, dev_list
) {
1498 if (tmp
->in_fs_metadata
&&
1499 !tmp
->is_tgtdev_for_dev_replace
&&
1509 printk(KERN_ERR
"btrfs: no missing devices found to "
1514 ret
= btrfs_get_bdev_and_sb(device_path
,
1515 FMODE_WRITE
| FMODE_EXCL
,
1516 root
->fs_info
->bdev_holder
, 0,
1520 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1521 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1522 dev_uuid
= disk_super
->dev_item
.uuid
;
1523 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1531 if (device
->is_tgtdev_for_dev_replace
) {
1532 pr_err("btrfs: unable to remove the dev_replace target dev\n");
1537 if (device
->writeable
&& root
->fs_info
->fs_devices
->rw_devices
== 1) {
1538 printk(KERN_ERR
"btrfs: unable to remove the only writeable "
1544 if (device
->writeable
) {
1546 list_del_init(&device
->dev_alloc_list
);
1547 unlock_chunks(root
);
1548 root
->fs_info
->fs_devices
->rw_devices
--;
1552 ret
= btrfs_shrink_device(device
, 0);
1557 * TODO: the superblock still includes this device in its num_devices
1558 * counter although write_all_supers() is not locked out. This
1559 * could give a filesystem state which requires a degraded mount.
1561 ret
= btrfs_rm_dev_item(root
->fs_info
->chunk_root
, device
);
1565 spin_lock(&root
->fs_info
->free_chunk_lock
);
1566 root
->fs_info
->free_chunk_space
= device
->total_bytes
-
1568 spin_unlock(&root
->fs_info
->free_chunk_lock
);
1570 device
->in_fs_metadata
= 0;
1571 btrfs_scrub_cancel_dev(root
->fs_info
, device
);
1574 * the device list mutex makes sure that we don't change
1575 * the device list while someone else is writing out all
1576 * the device supers.
1579 cur_devices
= device
->fs_devices
;
1580 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1581 list_del_rcu(&device
->dev_list
);
1583 device
->fs_devices
->num_devices
--;
1584 device
->fs_devices
->total_devices
--;
1586 if (device
->missing
)
1587 root
->fs_info
->fs_devices
->missing_devices
--;
1589 next_device
= list_entry(root
->fs_info
->fs_devices
->devices
.next
,
1590 struct btrfs_device
, dev_list
);
1591 if (device
->bdev
== root
->fs_info
->sb
->s_bdev
)
1592 root
->fs_info
->sb
->s_bdev
= next_device
->bdev
;
1593 if (device
->bdev
== root
->fs_info
->fs_devices
->latest_bdev
)
1594 root
->fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1597 device
->fs_devices
->open_devices
--;
1599 call_rcu(&device
->rcu
, free_device
);
1600 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1602 num_devices
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
1603 btrfs_set_super_num_devices(root
->fs_info
->super_copy
, num_devices
);
1605 if (cur_devices
->open_devices
== 0) {
1606 struct btrfs_fs_devices
*fs_devices
;
1607 fs_devices
= root
->fs_info
->fs_devices
;
1608 while (fs_devices
) {
1609 if (fs_devices
->seed
== cur_devices
)
1611 fs_devices
= fs_devices
->seed
;
1613 fs_devices
->seed
= cur_devices
->seed
;
1614 cur_devices
->seed
= NULL
;
1616 __btrfs_close_devices(cur_devices
);
1617 unlock_chunks(root
);
1618 free_fs_devices(cur_devices
);
1621 root
->fs_info
->num_tolerated_disk_barrier_failures
=
1622 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
1625 * at this point, the device is zero sized. We want to
1626 * remove it from the devices list and zero out the old super
1628 if (clear_super
&& disk_super
) {
1629 /* make sure this device isn't detected as part of
1632 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
1633 set_buffer_dirty(bh
);
1634 sync_dirty_buffer(bh
);
1639 /* Notify udev that device has changed */
1641 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
1646 blkdev_put(bdev
, FMODE_READ
| FMODE_EXCL
);
1648 mutex_unlock(&uuid_mutex
);
1651 if (device
->writeable
) {
1653 list_add(&device
->dev_alloc_list
,
1654 &root
->fs_info
->fs_devices
->alloc_list
);
1655 unlock_chunks(root
);
1656 root
->fs_info
->fs_devices
->rw_devices
++;
1661 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info
*fs_info
,
1662 struct btrfs_device
*srcdev
)
1664 WARN_ON(!mutex_is_locked(&fs_info
->fs_devices
->device_list_mutex
));
1665 list_del_rcu(&srcdev
->dev_list
);
1666 list_del_rcu(&srcdev
->dev_alloc_list
);
1667 fs_info
->fs_devices
->num_devices
--;
1668 if (srcdev
->missing
) {
1669 fs_info
->fs_devices
->missing_devices
--;
1670 fs_info
->fs_devices
->rw_devices
++;
1672 if (srcdev
->can_discard
)
1673 fs_info
->fs_devices
->num_can_discard
--;
1675 fs_info
->fs_devices
->open_devices
--;
1677 call_rcu(&srcdev
->rcu
, free_device
);
1680 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
1681 struct btrfs_device
*tgtdev
)
1683 struct btrfs_device
*next_device
;
1686 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
1688 btrfs_scratch_superblock(tgtdev
);
1689 fs_info
->fs_devices
->open_devices
--;
1691 fs_info
->fs_devices
->num_devices
--;
1692 if (tgtdev
->can_discard
)
1693 fs_info
->fs_devices
->num_can_discard
++;
1695 next_device
= list_entry(fs_info
->fs_devices
->devices
.next
,
1696 struct btrfs_device
, dev_list
);
1697 if (tgtdev
->bdev
== fs_info
->sb
->s_bdev
)
1698 fs_info
->sb
->s_bdev
= next_device
->bdev
;
1699 if (tgtdev
->bdev
== fs_info
->fs_devices
->latest_bdev
)
1700 fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1701 list_del_rcu(&tgtdev
->dev_list
);
1703 call_rcu(&tgtdev
->rcu
, free_device
);
1705 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
1708 int btrfs_find_device_by_path(struct btrfs_root
*root
, char *device_path
,
1709 struct btrfs_device
**device
)
1712 struct btrfs_super_block
*disk_super
;
1715 struct block_device
*bdev
;
1716 struct buffer_head
*bh
;
1719 ret
= btrfs_get_bdev_and_sb(device_path
, FMODE_READ
,
1720 root
->fs_info
->bdev_holder
, 0, &bdev
, &bh
);
1723 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1724 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1725 dev_uuid
= disk_super
->dev_item
.uuid
;
1726 *device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1731 blkdev_put(bdev
, FMODE_READ
);
1735 int btrfs_find_device_missing_or_by_path(struct btrfs_root
*root
,
1737 struct btrfs_device
**device
)
1740 if (strcmp(device_path
, "missing") == 0) {
1741 struct list_head
*devices
;
1742 struct btrfs_device
*tmp
;
1744 devices
= &root
->fs_info
->fs_devices
->devices
;
1746 * It is safe to read the devices since the volume_mutex
1747 * is held by the caller.
1749 list_for_each_entry(tmp
, devices
, dev_list
) {
1750 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
1757 pr_err("btrfs: no missing device found\n");
1763 return btrfs_find_device_by_path(root
, device_path
, device
);
1768 * does all the dirty work required for changing file system's UUID.
1770 static int btrfs_prepare_sprout(struct btrfs_root
*root
)
1772 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
1773 struct btrfs_fs_devices
*old_devices
;
1774 struct btrfs_fs_devices
*seed_devices
;
1775 struct btrfs_super_block
*disk_super
= root
->fs_info
->super_copy
;
1776 struct btrfs_device
*device
;
1779 BUG_ON(!mutex_is_locked(&uuid_mutex
));
1780 if (!fs_devices
->seeding
)
1783 seed_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
1787 old_devices
= clone_fs_devices(fs_devices
);
1788 if (IS_ERR(old_devices
)) {
1789 kfree(seed_devices
);
1790 return PTR_ERR(old_devices
);
1793 list_add(&old_devices
->list
, &fs_uuids
);
1795 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
1796 seed_devices
->opened
= 1;
1797 INIT_LIST_HEAD(&seed_devices
->devices
);
1798 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
1799 mutex_init(&seed_devices
->device_list_mutex
);
1801 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1802 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
1804 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1806 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
1807 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
) {
1808 device
->fs_devices
= seed_devices
;
1811 fs_devices
->seeding
= 0;
1812 fs_devices
->num_devices
= 0;
1813 fs_devices
->open_devices
= 0;
1814 fs_devices
->total_devices
= 0;
1815 fs_devices
->seed
= seed_devices
;
1817 generate_random_uuid(fs_devices
->fsid
);
1818 memcpy(root
->fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
1819 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
1820 super_flags
= btrfs_super_flags(disk_super
) &
1821 ~BTRFS_SUPER_FLAG_SEEDING
;
1822 btrfs_set_super_flags(disk_super
, super_flags
);
1828 * strore the expected generation for seed devices in device items.
1830 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
1831 struct btrfs_root
*root
)
1833 struct btrfs_path
*path
;
1834 struct extent_buffer
*leaf
;
1835 struct btrfs_dev_item
*dev_item
;
1836 struct btrfs_device
*device
;
1837 struct btrfs_key key
;
1838 u8 fs_uuid
[BTRFS_UUID_SIZE
];
1839 u8 dev_uuid
[BTRFS_UUID_SIZE
];
1843 path
= btrfs_alloc_path();
1847 root
= root
->fs_info
->chunk_root
;
1848 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1850 key
.type
= BTRFS_DEV_ITEM_KEY
;
1853 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
1857 leaf
= path
->nodes
[0];
1859 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1860 ret
= btrfs_next_leaf(root
, path
);
1865 leaf
= path
->nodes
[0];
1866 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1867 btrfs_release_path(path
);
1871 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1872 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
1873 key
.type
!= BTRFS_DEV_ITEM_KEY
)
1876 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1877 struct btrfs_dev_item
);
1878 devid
= btrfs_device_id(leaf
, dev_item
);
1879 read_extent_buffer(leaf
, dev_uuid
,
1880 (unsigned long)btrfs_device_uuid(dev_item
),
1882 read_extent_buffer(leaf
, fs_uuid
,
1883 (unsigned long)btrfs_device_fsid(dev_item
),
1885 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1887 BUG_ON(!device
); /* Logic error */
1889 if (device
->fs_devices
->seeding
) {
1890 btrfs_set_device_generation(leaf
, dev_item
,
1891 device
->generation
);
1892 btrfs_mark_buffer_dirty(leaf
);
1900 btrfs_free_path(path
);
1904 int btrfs_init_new_device(struct btrfs_root
*root
, char *device_path
)
1906 struct request_queue
*q
;
1907 struct btrfs_trans_handle
*trans
;
1908 struct btrfs_device
*device
;
1909 struct block_device
*bdev
;
1910 struct list_head
*devices
;
1911 struct super_block
*sb
= root
->fs_info
->sb
;
1912 struct rcu_string
*name
;
1914 int seeding_dev
= 0;
1917 if ((sb
->s_flags
& MS_RDONLY
) && !root
->fs_info
->fs_devices
->seeding
)
1920 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
1921 root
->fs_info
->bdev_holder
);
1923 return PTR_ERR(bdev
);
1925 if (root
->fs_info
->fs_devices
->seeding
) {
1927 down_write(&sb
->s_umount
);
1928 mutex_lock(&uuid_mutex
);
1931 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
1933 devices
= &root
->fs_info
->fs_devices
->devices
;
1935 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1936 list_for_each_entry(device
, devices
, dev_list
) {
1937 if (device
->bdev
== bdev
) {
1940 &root
->fs_info
->fs_devices
->device_list_mutex
);
1944 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1946 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
1948 /* we can safely leave the fs_devices entry around */
1953 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
1959 rcu_assign_pointer(device
->name
, name
);
1961 ret
= find_next_devid(root
, &device
->devid
);
1963 rcu_string_free(device
->name
);
1968 trans
= btrfs_start_transaction(root
, 0);
1969 if (IS_ERR(trans
)) {
1970 rcu_string_free(device
->name
);
1972 ret
= PTR_ERR(trans
);
1978 q
= bdev_get_queue(bdev
);
1979 if (blk_queue_discard(q
))
1980 device
->can_discard
= 1;
1981 device
->writeable
= 1;
1982 device
->work
.func
= pending_bios_fn
;
1983 generate_random_uuid(device
->uuid
);
1984 spin_lock_init(&device
->io_lock
);
1985 device
->generation
= trans
->transid
;
1986 device
->io_width
= root
->sectorsize
;
1987 device
->io_align
= root
->sectorsize
;
1988 device
->sector_size
= root
->sectorsize
;
1989 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
1990 device
->disk_total_bytes
= device
->total_bytes
;
1991 device
->dev_root
= root
->fs_info
->dev_root
;
1992 device
->bdev
= bdev
;
1993 device
->in_fs_metadata
= 1;
1994 device
->is_tgtdev_for_dev_replace
= 0;
1995 device
->mode
= FMODE_EXCL
;
1996 set_blocksize(device
->bdev
, 4096);
1999 sb
->s_flags
&= ~MS_RDONLY
;
2000 ret
= btrfs_prepare_sprout(root
);
2001 BUG_ON(ret
); /* -ENOMEM */
2004 device
->fs_devices
= root
->fs_info
->fs_devices
;
2006 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2007 list_add_rcu(&device
->dev_list
, &root
->fs_info
->fs_devices
->devices
);
2008 list_add(&device
->dev_alloc_list
,
2009 &root
->fs_info
->fs_devices
->alloc_list
);
2010 root
->fs_info
->fs_devices
->num_devices
++;
2011 root
->fs_info
->fs_devices
->open_devices
++;
2012 root
->fs_info
->fs_devices
->rw_devices
++;
2013 root
->fs_info
->fs_devices
->total_devices
++;
2014 if (device
->can_discard
)
2015 root
->fs_info
->fs_devices
->num_can_discard
++;
2016 root
->fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
2018 spin_lock(&root
->fs_info
->free_chunk_lock
);
2019 root
->fs_info
->free_chunk_space
+= device
->total_bytes
;
2020 spin_unlock(&root
->fs_info
->free_chunk_lock
);
2022 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
2023 root
->fs_info
->fs_devices
->rotating
= 1;
2025 total_bytes
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
2026 btrfs_set_super_total_bytes(root
->fs_info
->super_copy
,
2027 total_bytes
+ device
->total_bytes
);
2029 total_bytes
= btrfs_super_num_devices(root
->fs_info
->super_copy
);
2030 btrfs_set_super_num_devices(root
->fs_info
->super_copy
,
2032 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2035 ret
= init_first_rw_device(trans
, root
, device
);
2037 btrfs_abort_transaction(trans
, root
, ret
);
2040 ret
= btrfs_finish_sprout(trans
, root
);
2042 btrfs_abort_transaction(trans
, root
, ret
);
2046 ret
= btrfs_add_device(trans
, root
, device
);
2048 btrfs_abort_transaction(trans
, root
, ret
);
2054 * we've got more storage, clear any full flags on the space
2057 btrfs_clear_space_info_full(root
->fs_info
);
2059 unlock_chunks(root
);
2060 root
->fs_info
->num_tolerated_disk_barrier_failures
=
2061 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
2062 ret
= btrfs_commit_transaction(trans
, root
);
2065 mutex_unlock(&uuid_mutex
);
2066 up_write(&sb
->s_umount
);
2068 if (ret
) /* transaction commit */
2071 ret
= btrfs_relocate_sys_chunks(root
);
2073 btrfs_error(root
->fs_info
, ret
,
2074 "Failed to relocate sys chunks after "
2075 "device initialization. This can be fixed "
2076 "using the \"btrfs balance\" command.");
2077 trans
= btrfs_attach_transaction(root
);
2078 if (IS_ERR(trans
)) {
2079 if (PTR_ERR(trans
) == -ENOENT
)
2081 return PTR_ERR(trans
);
2083 ret
= btrfs_commit_transaction(trans
, root
);
2089 unlock_chunks(root
);
2090 btrfs_end_transaction(trans
, root
);
2091 rcu_string_free(device
->name
);
2094 blkdev_put(bdev
, FMODE_EXCL
);
2096 mutex_unlock(&uuid_mutex
);
2097 up_write(&sb
->s_umount
);
2102 int btrfs_init_dev_replace_tgtdev(struct btrfs_root
*root
, char *device_path
,
2103 struct btrfs_device
**device_out
)
2105 struct request_queue
*q
;
2106 struct btrfs_device
*device
;
2107 struct block_device
*bdev
;
2108 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2109 struct list_head
*devices
;
2110 struct rcu_string
*name
;
2114 if (fs_info
->fs_devices
->seeding
)
2117 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2118 fs_info
->bdev_holder
);
2120 return PTR_ERR(bdev
);
2122 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2124 devices
= &fs_info
->fs_devices
->devices
;
2125 list_for_each_entry(device
, devices
, dev_list
) {
2126 if (device
->bdev
== bdev
) {
2132 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
2138 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2144 rcu_assign_pointer(device
->name
, name
);
2146 q
= bdev_get_queue(bdev
);
2147 if (blk_queue_discard(q
))
2148 device
->can_discard
= 1;
2149 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2150 device
->writeable
= 1;
2151 device
->work
.func
= pending_bios_fn
;
2152 generate_random_uuid(device
->uuid
);
2153 device
->devid
= BTRFS_DEV_REPLACE_DEVID
;
2154 spin_lock_init(&device
->io_lock
);
2155 device
->generation
= 0;
2156 device
->io_width
= root
->sectorsize
;
2157 device
->io_align
= root
->sectorsize
;
2158 device
->sector_size
= root
->sectorsize
;
2159 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
2160 device
->disk_total_bytes
= device
->total_bytes
;
2161 device
->dev_root
= fs_info
->dev_root
;
2162 device
->bdev
= bdev
;
2163 device
->in_fs_metadata
= 1;
2164 device
->is_tgtdev_for_dev_replace
= 1;
2165 device
->mode
= FMODE_EXCL
;
2166 set_blocksize(device
->bdev
, 4096);
2167 device
->fs_devices
= fs_info
->fs_devices
;
2168 list_add(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2169 fs_info
->fs_devices
->num_devices
++;
2170 fs_info
->fs_devices
->open_devices
++;
2171 if (device
->can_discard
)
2172 fs_info
->fs_devices
->num_can_discard
++;
2173 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2175 *device_out
= device
;
2179 blkdev_put(bdev
, FMODE_EXCL
);
2183 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info
*fs_info
,
2184 struct btrfs_device
*tgtdev
)
2186 WARN_ON(fs_info
->fs_devices
->rw_devices
== 0);
2187 tgtdev
->io_width
= fs_info
->dev_root
->sectorsize
;
2188 tgtdev
->io_align
= fs_info
->dev_root
->sectorsize
;
2189 tgtdev
->sector_size
= fs_info
->dev_root
->sectorsize
;
2190 tgtdev
->dev_root
= fs_info
->dev_root
;
2191 tgtdev
->in_fs_metadata
= 1;
2194 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2195 struct btrfs_device
*device
)
2198 struct btrfs_path
*path
;
2199 struct btrfs_root
*root
;
2200 struct btrfs_dev_item
*dev_item
;
2201 struct extent_buffer
*leaf
;
2202 struct btrfs_key key
;
2204 root
= device
->dev_root
->fs_info
->chunk_root
;
2206 path
= btrfs_alloc_path();
2210 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2211 key
.type
= BTRFS_DEV_ITEM_KEY
;
2212 key
.offset
= device
->devid
;
2214 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2223 leaf
= path
->nodes
[0];
2224 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2226 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2227 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2228 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2229 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2230 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2231 btrfs_set_device_total_bytes(leaf
, dev_item
, device
->disk_total_bytes
);
2232 btrfs_set_device_bytes_used(leaf
, dev_item
, device
->bytes_used
);
2233 btrfs_mark_buffer_dirty(leaf
);
2236 btrfs_free_path(path
);
2240 static int __btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2241 struct btrfs_device
*device
, u64 new_size
)
2243 struct btrfs_super_block
*super_copy
=
2244 device
->dev_root
->fs_info
->super_copy
;
2245 u64 old_total
= btrfs_super_total_bytes(super_copy
);
2246 u64 diff
= new_size
- device
->total_bytes
;
2248 if (!device
->writeable
)
2250 if (new_size
<= device
->total_bytes
||
2251 device
->is_tgtdev_for_dev_replace
)
2254 btrfs_set_super_total_bytes(super_copy
, old_total
+ diff
);
2255 device
->fs_devices
->total_rw_bytes
+= diff
;
2257 device
->total_bytes
= new_size
;
2258 device
->disk_total_bytes
= new_size
;
2259 btrfs_clear_space_info_full(device
->dev_root
->fs_info
);
2261 return btrfs_update_device(trans
, device
);
2264 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2265 struct btrfs_device
*device
, u64 new_size
)
2268 lock_chunks(device
->dev_root
);
2269 ret
= __btrfs_grow_device(trans
, device
, new_size
);
2270 unlock_chunks(device
->dev_root
);
2274 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
2275 struct btrfs_root
*root
,
2276 u64 chunk_tree
, u64 chunk_objectid
,
2280 struct btrfs_path
*path
;
2281 struct btrfs_key key
;
2283 root
= root
->fs_info
->chunk_root
;
2284 path
= btrfs_alloc_path();
2288 key
.objectid
= chunk_objectid
;
2289 key
.offset
= chunk_offset
;
2290 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2292 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2295 else if (ret
> 0) { /* Logic error or corruption */
2296 btrfs_error(root
->fs_info
, -ENOENT
,
2297 "Failed lookup while freeing chunk.");
2302 ret
= btrfs_del_item(trans
, root
, path
);
2304 btrfs_error(root
->fs_info
, ret
,
2305 "Failed to delete chunk item.");
2307 btrfs_free_path(path
);
2311 static int btrfs_del_sys_chunk(struct btrfs_root
*root
, u64 chunk_objectid
, u64
2314 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
2315 struct btrfs_disk_key
*disk_key
;
2316 struct btrfs_chunk
*chunk
;
2323 struct btrfs_key key
;
2325 array_size
= btrfs_super_sys_array_size(super_copy
);
2327 ptr
= super_copy
->sys_chunk_array
;
2330 while (cur
< array_size
) {
2331 disk_key
= (struct btrfs_disk_key
*)ptr
;
2332 btrfs_disk_key_to_cpu(&key
, disk_key
);
2334 len
= sizeof(*disk_key
);
2336 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2337 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2338 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2339 len
+= btrfs_chunk_item_size(num_stripes
);
2344 if (key
.objectid
== chunk_objectid
&&
2345 key
.offset
== chunk_offset
) {
2346 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2348 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2357 static int btrfs_relocate_chunk(struct btrfs_root
*root
,
2358 u64 chunk_tree
, u64 chunk_objectid
,
2361 struct extent_map_tree
*em_tree
;
2362 struct btrfs_root
*extent_root
;
2363 struct btrfs_trans_handle
*trans
;
2364 struct extent_map
*em
;
2365 struct map_lookup
*map
;
2369 root
= root
->fs_info
->chunk_root
;
2370 extent_root
= root
->fs_info
->extent_root
;
2371 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
2373 ret
= btrfs_can_relocate(extent_root
, chunk_offset
);
2377 /* step one, relocate all the extents inside this chunk */
2378 ret
= btrfs_relocate_block_group(extent_root
, chunk_offset
);
2382 trans
= btrfs_start_transaction(root
, 0);
2383 BUG_ON(IS_ERR(trans
));
2388 * step two, delete the device extents and the
2389 * chunk tree entries
2391 read_lock(&em_tree
->lock
);
2392 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
2393 read_unlock(&em_tree
->lock
);
2395 BUG_ON(!em
|| em
->start
> chunk_offset
||
2396 em
->start
+ em
->len
< chunk_offset
);
2397 map
= (struct map_lookup
*)em
->bdev
;
2399 for (i
= 0; i
< map
->num_stripes
; i
++) {
2400 ret
= btrfs_free_dev_extent(trans
, map
->stripes
[i
].dev
,
2401 map
->stripes
[i
].physical
);
2404 if (map
->stripes
[i
].dev
) {
2405 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2409 ret
= btrfs_free_chunk(trans
, root
, chunk_tree
, chunk_objectid
,
2414 trace_btrfs_chunk_free(root
, map
, chunk_offset
, em
->len
);
2416 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2417 ret
= btrfs_del_sys_chunk(root
, chunk_objectid
, chunk_offset
);
2421 ret
= btrfs_remove_block_group(trans
, extent_root
, chunk_offset
);
2424 write_lock(&em_tree
->lock
);
2425 remove_extent_mapping(em_tree
, em
);
2426 write_unlock(&em_tree
->lock
);
2431 /* once for the tree */
2432 free_extent_map(em
);
2434 free_extent_map(em
);
2436 unlock_chunks(root
);
2437 btrfs_end_transaction(trans
, root
);
2441 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
)
2443 struct btrfs_root
*chunk_root
= root
->fs_info
->chunk_root
;
2444 struct btrfs_path
*path
;
2445 struct extent_buffer
*leaf
;
2446 struct btrfs_chunk
*chunk
;
2447 struct btrfs_key key
;
2448 struct btrfs_key found_key
;
2449 u64 chunk_tree
= chunk_root
->root_key
.objectid
;
2451 bool retried
= false;
2455 path
= btrfs_alloc_path();
2460 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2461 key
.offset
= (u64
)-1;
2462 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2465 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2468 BUG_ON(ret
== 0); /* Corruption */
2470 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
2477 leaf
= path
->nodes
[0];
2478 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2480 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
2481 struct btrfs_chunk
);
2482 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2483 btrfs_release_path(path
);
2485 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2486 ret
= btrfs_relocate_chunk(chunk_root
, chunk_tree
,
2495 if (found_key
.offset
== 0)
2497 key
.offset
= found_key
.offset
- 1;
2500 if (failed
&& !retried
) {
2504 } else if (failed
&& retried
) {
2509 btrfs_free_path(path
);
2513 static int insert_balance_item(struct btrfs_root
*root
,
2514 struct btrfs_balance_control
*bctl
)
2516 struct btrfs_trans_handle
*trans
;
2517 struct btrfs_balance_item
*item
;
2518 struct btrfs_disk_balance_args disk_bargs
;
2519 struct btrfs_path
*path
;
2520 struct extent_buffer
*leaf
;
2521 struct btrfs_key key
;
2524 path
= btrfs_alloc_path();
2528 trans
= btrfs_start_transaction(root
, 0);
2529 if (IS_ERR(trans
)) {
2530 btrfs_free_path(path
);
2531 return PTR_ERR(trans
);
2534 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2535 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2538 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2543 leaf
= path
->nodes
[0];
2544 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
2546 memset_extent_buffer(leaf
, 0, (unsigned long)item
, sizeof(*item
));
2548 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
2549 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
2550 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
2551 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
2552 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
2553 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
2555 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
2557 btrfs_mark_buffer_dirty(leaf
);
2559 btrfs_free_path(path
);
2560 err
= btrfs_commit_transaction(trans
, root
);
2566 static int del_balance_item(struct btrfs_root
*root
)
2568 struct btrfs_trans_handle
*trans
;
2569 struct btrfs_path
*path
;
2570 struct btrfs_key key
;
2573 path
= btrfs_alloc_path();
2577 trans
= btrfs_start_transaction(root
, 0);
2578 if (IS_ERR(trans
)) {
2579 btrfs_free_path(path
);
2580 return PTR_ERR(trans
);
2583 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2584 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2587 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2595 ret
= btrfs_del_item(trans
, root
, path
);
2597 btrfs_free_path(path
);
2598 err
= btrfs_commit_transaction(trans
, root
);
2605 * This is a heuristic used to reduce the number of chunks balanced on
2606 * resume after balance was interrupted.
2608 static void update_balance_args(struct btrfs_balance_control
*bctl
)
2611 * Turn on soft mode for chunk types that were being converted.
2613 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2614 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2615 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2616 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2617 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2618 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2621 * Turn on usage filter if is not already used. The idea is
2622 * that chunks that we have already balanced should be
2623 * reasonably full. Don't do it for chunks that are being
2624 * converted - that will keep us from relocating unconverted
2625 * (albeit full) chunks.
2627 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2628 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2629 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2630 bctl
->data
.usage
= 90;
2632 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2633 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2634 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2635 bctl
->sys
.usage
= 90;
2637 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2638 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2639 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2640 bctl
->meta
.usage
= 90;
2645 * Should be called with both balance and volume mutexes held to
2646 * serialize other volume operations (add_dev/rm_dev/resize) with
2647 * restriper. Same goes for unset_balance_control.
2649 static void set_balance_control(struct btrfs_balance_control
*bctl
)
2651 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
2653 BUG_ON(fs_info
->balance_ctl
);
2655 spin_lock(&fs_info
->balance_lock
);
2656 fs_info
->balance_ctl
= bctl
;
2657 spin_unlock(&fs_info
->balance_lock
);
2660 static void unset_balance_control(struct btrfs_fs_info
*fs_info
)
2662 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
2664 BUG_ON(!fs_info
->balance_ctl
);
2666 spin_lock(&fs_info
->balance_lock
);
2667 fs_info
->balance_ctl
= NULL
;
2668 spin_unlock(&fs_info
->balance_lock
);
2674 * Balance filters. Return 1 if chunk should be filtered out
2675 * (should not be balanced).
2677 static int chunk_profiles_filter(u64 chunk_type
,
2678 struct btrfs_balance_args
*bargs
)
2680 chunk_type
= chunk_to_extended(chunk_type
) &
2681 BTRFS_EXTENDED_PROFILE_MASK
;
2683 if (bargs
->profiles
& chunk_type
)
2689 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
2690 struct btrfs_balance_args
*bargs
)
2692 struct btrfs_block_group_cache
*cache
;
2693 u64 chunk_used
, user_thresh
;
2696 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
2697 chunk_used
= btrfs_block_group_used(&cache
->item
);
2699 if (bargs
->usage
== 0)
2701 else if (bargs
->usage
> 100)
2702 user_thresh
= cache
->key
.offset
;
2704 user_thresh
= div_factor_fine(cache
->key
.offset
,
2707 if (chunk_used
< user_thresh
)
2710 btrfs_put_block_group(cache
);
2714 static int chunk_devid_filter(struct extent_buffer
*leaf
,
2715 struct btrfs_chunk
*chunk
,
2716 struct btrfs_balance_args
*bargs
)
2718 struct btrfs_stripe
*stripe
;
2719 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2722 for (i
= 0; i
< num_stripes
; i
++) {
2723 stripe
= btrfs_stripe_nr(chunk
, i
);
2724 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
2731 /* [pstart, pend) */
2732 static int chunk_drange_filter(struct extent_buffer
*leaf
,
2733 struct btrfs_chunk
*chunk
,
2735 struct btrfs_balance_args
*bargs
)
2737 struct btrfs_stripe
*stripe
;
2738 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2744 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
2747 if (btrfs_chunk_type(leaf
, chunk
) & (BTRFS_BLOCK_GROUP_DUP
|
2748 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)) {
2749 factor
= num_stripes
/ 2;
2750 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID5
) {
2751 factor
= num_stripes
- 1;
2752 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID6
) {
2753 factor
= num_stripes
- 2;
2755 factor
= num_stripes
;
2758 for (i
= 0; i
< num_stripes
; i
++) {
2759 stripe
= btrfs_stripe_nr(chunk
, i
);
2760 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
2763 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
2764 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
2765 do_div(stripe_length
, factor
);
2767 if (stripe_offset
< bargs
->pend
&&
2768 stripe_offset
+ stripe_length
> bargs
->pstart
)
2775 /* [vstart, vend) */
2776 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
2777 struct btrfs_chunk
*chunk
,
2779 struct btrfs_balance_args
*bargs
)
2781 if (chunk_offset
< bargs
->vend
&&
2782 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
2783 /* at least part of the chunk is inside this vrange */
2789 static int chunk_soft_convert_filter(u64 chunk_type
,
2790 struct btrfs_balance_args
*bargs
)
2792 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
2795 chunk_type
= chunk_to_extended(chunk_type
) &
2796 BTRFS_EXTENDED_PROFILE_MASK
;
2798 if (bargs
->target
== chunk_type
)
2804 static int should_balance_chunk(struct btrfs_root
*root
,
2805 struct extent_buffer
*leaf
,
2806 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
2808 struct btrfs_balance_control
*bctl
= root
->fs_info
->balance_ctl
;
2809 struct btrfs_balance_args
*bargs
= NULL
;
2810 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2813 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
2814 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
2818 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
2819 bargs
= &bctl
->data
;
2820 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
2822 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
2823 bargs
= &bctl
->meta
;
2825 /* profiles filter */
2826 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
2827 chunk_profiles_filter(chunk_type
, bargs
)) {
2832 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2833 chunk_usage_filter(bctl
->fs_info
, chunk_offset
, bargs
)) {
2838 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
2839 chunk_devid_filter(leaf
, chunk
, bargs
)) {
2843 /* drange filter, makes sense only with devid filter */
2844 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
2845 chunk_drange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
2850 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
2851 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
2855 /* soft profile changing mode */
2856 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
2857 chunk_soft_convert_filter(chunk_type
, bargs
)) {
2864 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
2866 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
2867 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
2868 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
2869 struct list_head
*devices
;
2870 struct btrfs_device
*device
;
2873 struct btrfs_chunk
*chunk
;
2874 struct btrfs_path
*path
;
2875 struct btrfs_key key
;
2876 struct btrfs_key found_key
;
2877 struct btrfs_trans_handle
*trans
;
2878 struct extent_buffer
*leaf
;
2881 int enospc_errors
= 0;
2882 bool counting
= true;
2884 /* step one make some room on all the devices */
2885 devices
= &fs_info
->fs_devices
->devices
;
2886 list_for_each_entry(device
, devices
, dev_list
) {
2887 old_size
= device
->total_bytes
;
2888 size_to_free
= div_factor(old_size
, 1);
2889 size_to_free
= min(size_to_free
, (u64
)1 * 1024 * 1024);
2890 if (!device
->writeable
||
2891 device
->total_bytes
- device
->bytes_used
> size_to_free
||
2892 device
->is_tgtdev_for_dev_replace
)
2895 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
2900 trans
= btrfs_start_transaction(dev_root
, 0);
2901 BUG_ON(IS_ERR(trans
));
2903 ret
= btrfs_grow_device(trans
, device
, old_size
);
2906 btrfs_end_transaction(trans
, dev_root
);
2909 /* step two, relocate all the chunks */
2910 path
= btrfs_alloc_path();
2916 /* zero out stat counters */
2917 spin_lock(&fs_info
->balance_lock
);
2918 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
2919 spin_unlock(&fs_info
->balance_lock
);
2921 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2922 key
.offset
= (u64
)-1;
2923 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2926 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
2927 atomic_read(&fs_info
->balance_cancel_req
)) {
2932 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2937 * this shouldn't happen, it means the last relocate
2941 BUG(); /* FIXME break ? */
2943 ret
= btrfs_previous_item(chunk_root
, path
, 0,
2944 BTRFS_CHUNK_ITEM_KEY
);
2950 leaf
= path
->nodes
[0];
2951 slot
= path
->slots
[0];
2952 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2954 if (found_key
.objectid
!= key
.objectid
)
2957 /* chunk zero is special */
2958 if (found_key
.offset
== 0)
2961 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
2964 spin_lock(&fs_info
->balance_lock
);
2965 bctl
->stat
.considered
++;
2966 spin_unlock(&fs_info
->balance_lock
);
2969 ret
= should_balance_chunk(chunk_root
, leaf
, chunk
,
2971 btrfs_release_path(path
);
2976 spin_lock(&fs_info
->balance_lock
);
2977 bctl
->stat
.expected
++;
2978 spin_unlock(&fs_info
->balance_lock
);
2982 ret
= btrfs_relocate_chunk(chunk_root
,
2983 chunk_root
->root_key
.objectid
,
2986 if (ret
&& ret
!= -ENOSPC
)
2988 if (ret
== -ENOSPC
) {
2991 spin_lock(&fs_info
->balance_lock
);
2992 bctl
->stat
.completed
++;
2993 spin_unlock(&fs_info
->balance_lock
);
2996 key
.offset
= found_key
.offset
- 1;
3000 btrfs_release_path(path
);
3005 btrfs_free_path(path
);
3006 if (enospc_errors
) {
3007 printk(KERN_INFO
"btrfs: %d enospc errors during balance\n",
3017 * alloc_profile_is_valid - see if a given profile is valid and reduced
3018 * @flags: profile to validate
3019 * @extended: if true @flags is treated as an extended profile
3021 static int alloc_profile_is_valid(u64 flags
, int extended
)
3023 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
3024 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
3026 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
3028 /* 1) check that all other bits are zeroed */
3032 /* 2) see if profile is reduced */
3034 return !extended
; /* "0" is valid for usual profiles */
3036 /* true if exactly one bit set */
3037 return (flags
& (flags
- 1)) == 0;
3040 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
3042 /* cancel requested || normal exit path */
3043 return atomic_read(&fs_info
->balance_cancel_req
) ||
3044 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
3045 atomic_read(&fs_info
->balance_cancel_req
) == 0);
3048 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
3052 unset_balance_control(fs_info
);
3053 ret
= del_balance_item(fs_info
->tree_root
);
3056 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3059 void update_ioctl_balance_args(struct btrfs_fs_info
*fs_info
, int lock
,
3060 struct btrfs_ioctl_balance_args
*bargs
);
3063 * Should be called with both balance and volume mutexes held
3065 int btrfs_balance(struct btrfs_balance_control
*bctl
,
3066 struct btrfs_ioctl_balance_args
*bargs
)
3068 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3075 if (btrfs_fs_closing(fs_info
) ||
3076 atomic_read(&fs_info
->balance_pause_req
) ||
3077 atomic_read(&fs_info
->balance_cancel_req
)) {
3082 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
3083 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
3087 * In case of mixed groups both data and meta should be picked,
3088 * and identical options should be given for both of them.
3090 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
3091 if (mixed
&& (bctl
->flags
& allowed
)) {
3092 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
3093 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
3094 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
3095 printk(KERN_ERR
"btrfs: with mixed groups data and "
3096 "metadata balance options must be the same\n");
3102 num_devices
= fs_info
->fs_devices
->num_devices
;
3103 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
3104 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
3105 BUG_ON(num_devices
< 1);
3108 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
3109 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
3110 if (num_devices
== 1)
3111 allowed
|= BTRFS_BLOCK_GROUP_DUP
;
3112 else if (num_devices
< 4)
3113 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
3115 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
|
3116 BTRFS_BLOCK_GROUP_RAID10
|
3117 BTRFS_BLOCK_GROUP_RAID5
|
3118 BTRFS_BLOCK_GROUP_RAID6
);
3120 if ((bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3121 (!alloc_profile_is_valid(bctl
->data
.target
, 1) ||
3122 (bctl
->data
.target
& ~allowed
))) {
3123 printk(KERN_ERR
"btrfs: unable to start balance with target "
3124 "data profile %llu\n",
3125 (unsigned long long)bctl
->data
.target
);
3129 if ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3130 (!alloc_profile_is_valid(bctl
->meta
.target
, 1) ||
3131 (bctl
->meta
.target
& ~allowed
))) {
3132 printk(KERN_ERR
"btrfs: unable to start balance with target "
3133 "metadata profile %llu\n",
3134 (unsigned long long)bctl
->meta
.target
);
3138 if ((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3139 (!alloc_profile_is_valid(bctl
->sys
.target
, 1) ||
3140 (bctl
->sys
.target
& ~allowed
))) {
3141 printk(KERN_ERR
"btrfs: unable to start balance with target "
3142 "system profile %llu\n",
3143 (unsigned long long)bctl
->sys
.target
);
3148 /* allow dup'ed data chunks only in mixed mode */
3149 if (!mixed
&& (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3150 (bctl
->data
.target
& BTRFS_BLOCK_GROUP_DUP
)) {
3151 printk(KERN_ERR
"btrfs: dup for data is not allowed\n");
3156 /* allow to reduce meta or sys integrity only if force set */
3157 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3158 BTRFS_BLOCK_GROUP_RAID10
|
3159 BTRFS_BLOCK_GROUP_RAID5
|
3160 BTRFS_BLOCK_GROUP_RAID6
;
3162 seq
= read_seqbegin(&fs_info
->profiles_lock
);
3164 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3165 (fs_info
->avail_system_alloc_bits
& allowed
) &&
3166 !(bctl
->sys
.target
& allowed
)) ||
3167 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3168 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
3169 !(bctl
->meta
.target
& allowed
))) {
3170 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
3171 printk(KERN_INFO
"btrfs: force reducing metadata "
3174 printk(KERN_ERR
"btrfs: balance will reduce metadata "
3175 "integrity, use force if you want this\n");
3180 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
3182 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3183 int num_tolerated_disk_barrier_failures
;
3184 u64 target
= bctl
->sys
.target
;
3186 num_tolerated_disk_barrier_failures
=
3187 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3188 if (num_tolerated_disk_barrier_failures
> 0 &&
3190 (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID0
|
3191 BTRFS_AVAIL_ALLOC_BIT_SINGLE
)))
3192 num_tolerated_disk_barrier_failures
= 0;
3193 else if (num_tolerated_disk_barrier_failures
> 1 &&
3195 (BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)))
3196 num_tolerated_disk_barrier_failures
= 1;
3198 fs_info
->num_tolerated_disk_barrier_failures
=
3199 num_tolerated_disk_barrier_failures
;
3202 ret
= insert_balance_item(fs_info
->tree_root
, bctl
);
3203 if (ret
&& ret
!= -EEXIST
)
3206 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
3207 BUG_ON(ret
== -EEXIST
);
3208 set_balance_control(bctl
);
3210 BUG_ON(ret
!= -EEXIST
);
3211 spin_lock(&fs_info
->balance_lock
);
3212 update_balance_args(bctl
);
3213 spin_unlock(&fs_info
->balance_lock
);
3216 atomic_inc(&fs_info
->balance_running
);
3217 mutex_unlock(&fs_info
->balance_mutex
);
3219 ret
= __btrfs_balance(fs_info
);
3221 mutex_lock(&fs_info
->balance_mutex
);
3222 atomic_dec(&fs_info
->balance_running
);
3224 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3225 fs_info
->num_tolerated_disk_barrier_failures
=
3226 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3230 memset(bargs
, 0, sizeof(*bargs
));
3231 update_ioctl_balance_args(fs_info
, 0, bargs
);
3234 wake_up(&fs_info
->balance_wait_q
);
3238 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
3239 __cancel_balance(fs_info
);
3242 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3247 static int balance_kthread(void *data
)
3249 struct btrfs_fs_info
*fs_info
= data
;
3252 mutex_lock(&fs_info
->volume_mutex
);
3253 mutex_lock(&fs_info
->balance_mutex
);
3255 if (fs_info
->balance_ctl
) {
3256 printk(KERN_INFO
"btrfs: continuing balance\n");
3257 ret
= btrfs_balance(fs_info
->balance_ctl
, NULL
);
3260 mutex_unlock(&fs_info
->balance_mutex
);
3261 mutex_unlock(&fs_info
->volume_mutex
);
3266 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
3268 struct task_struct
*tsk
;
3270 spin_lock(&fs_info
->balance_lock
);
3271 if (!fs_info
->balance_ctl
) {
3272 spin_unlock(&fs_info
->balance_lock
);
3275 spin_unlock(&fs_info
->balance_lock
);
3277 if (btrfs_test_opt(fs_info
->tree_root
, SKIP_BALANCE
)) {
3278 printk(KERN_INFO
"btrfs: force skipping balance\n");
3282 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
3284 return PTR_ERR(tsk
);
3289 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
3291 struct btrfs_balance_control
*bctl
;
3292 struct btrfs_balance_item
*item
;
3293 struct btrfs_disk_balance_args disk_bargs
;
3294 struct btrfs_path
*path
;
3295 struct extent_buffer
*leaf
;
3296 struct btrfs_key key
;
3299 path
= btrfs_alloc_path();
3303 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3304 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
3307 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
3310 if (ret
> 0) { /* ret = -ENOENT; */
3315 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
3321 leaf
= path
->nodes
[0];
3322 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3324 bctl
->fs_info
= fs_info
;
3325 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
3326 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
3328 btrfs_balance_data(leaf
, item
, &disk_bargs
);
3329 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
3330 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
3331 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
3332 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
3333 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
3335 WARN_ON(atomic_xchg(&fs_info
->mutually_exclusive_operation_running
, 1));
3337 mutex_lock(&fs_info
->volume_mutex
);
3338 mutex_lock(&fs_info
->balance_mutex
);
3340 set_balance_control(bctl
);
3342 mutex_unlock(&fs_info
->balance_mutex
);
3343 mutex_unlock(&fs_info
->volume_mutex
);
3345 btrfs_free_path(path
);
3349 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
3353 mutex_lock(&fs_info
->balance_mutex
);
3354 if (!fs_info
->balance_ctl
) {
3355 mutex_unlock(&fs_info
->balance_mutex
);
3359 if (atomic_read(&fs_info
->balance_running
)) {
3360 atomic_inc(&fs_info
->balance_pause_req
);
3361 mutex_unlock(&fs_info
->balance_mutex
);
3363 wait_event(fs_info
->balance_wait_q
,
3364 atomic_read(&fs_info
->balance_running
) == 0);
3366 mutex_lock(&fs_info
->balance_mutex
);
3367 /* we are good with balance_ctl ripped off from under us */
3368 BUG_ON(atomic_read(&fs_info
->balance_running
));
3369 atomic_dec(&fs_info
->balance_pause_req
);
3374 mutex_unlock(&fs_info
->balance_mutex
);
3378 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
3380 mutex_lock(&fs_info
->balance_mutex
);
3381 if (!fs_info
->balance_ctl
) {
3382 mutex_unlock(&fs_info
->balance_mutex
);
3386 atomic_inc(&fs_info
->balance_cancel_req
);
3388 * if we are running just wait and return, balance item is
3389 * deleted in btrfs_balance in this case
3391 if (atomic_read(&fs_info
->balance_running
)) {
3392 mutex_unlock(&fs_info
->balance_mutex
);
3393 wait_event(fs_info
->balance_wait_q
,
3394 atomic_read(&fs_info
->balance_running
) == 0);
3395 mutex_lock(&fs_info
->balance_mutex
);
3397 /* __cancel_balance needs volume_mutex */
3398 mutex_unlock(&fs_info
->balance_mutex
);
3399 mutex_lock(&fs_info
->volume_mutex
);
3400 mutex_lock(&fs_info
->balance_mutex
);
3402 if (fs_info
->balance_ctl
)
3403 __cancel_balance(fs_info
);
3405 mutex_unlock(&fs_info
->volume_mutex
);
3408 BUG_ON(fs_info
->balance_ctl
|| atomic_read(&fs_info
->balance_running
));
3409 atomic_dec(&fs_info
->balance_cancel_req
);
3410 mutex_unlock(&fs_info
->balance_mutex
);
3415 * shrinking a device means finding all of the device extents past
3416 * the new size, and then following the back refs to the chunks.
3417 * The chunk relocation code actually frees the device extent
3419 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
3421 struct btrfs_trans_handle
*trans
;
3422 struct btrfs_root
*root
= device
->dev_root
;
3423 struct btrfs_dev_extent
*dev_extent
= NULL
;
3424 struct btrfs_path
*path
;
3432 bool retried
= false;
3433 struct extent_buffer
*l
;
3434 struct btrfs_key key
;
3435 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
3436 u64 old_total
= btrfs_super_total_bytes(super_copy
);
3437 u64 old_size
= device
->total_bytes
;
3438 u64 diff
= device
->total_bytes
- new_size
;
3440 if (device
->is_tgtdev_for_dev_replace
)
3443 path
= btrfs_alloc_path();
3451 device
->total_bytes
= new_size
;
3452 if (device
->writeable
) {
3453 device
->fs_devices
->total_rw_bytes
-= diff
;
3454 spin_lock(&root
->fs_info
->free_chunk_lock
);
3455 root
->fs_info
->free_chunk_space
-= diff
;
3456 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3458 unlock_chunks(root
);
3461 key
.objectid
= device
->devid
;
3462 key
.offset
= (u64
)-1;
3463 key
.type
= BTRFS_DEV_EXTENT_KEY
;
3466 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3470 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
3475 btrfs_release_path(path
);
3480 slot
= path
->slots
[0];
3481 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
3483 if (key
.objectid
!= device
->devid
) {
3484 btrfs_release_path(path
);
3488 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
3489 length
= btrfs_dev_extent_length(l
, dev_extent
);
3491 if (key
.offset
+ length
<= new_size
) {
3492 btrfs_release_path(path
);
3496 chunk_tree
= btrfs_dev_extent_chunk_tree(l
, dev_extent
);
3497 chunk_objectid
= btrfs_dev_extent_chunk_objectid(l
, dev_extent
);
3498 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
3499 btrfs_release_path(path
);
3501 ret
= btrfs_relocate_chunk(root
, chunk_tree
, chunk_objectid
,
3503 if (ret
&& ret
!= -ENOSPC
)
3507 } while (key
.offset
-- > 0);
3509 if (failed
&& !retried
) {
3513 } else if (failed
&& retried
) {
3517 device
->total_bytes
= old_size
;
3518 if (device
->writeable
)
3519 device
->fs_devices
->total_rw_bytes
+= diff
;
3520 spin_lock(&root
->fs_info
->free_chunk_lock
);
3521 root
->fs_info
->free_chunk_space
+= diff
;
3522 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3523 unlock_chunks(root
);
3527 /* Shrinking succeeded, else we would be at "done". */
3528 trans
= btrfs_start_transaction(root
, 0);
3529 if (IS_ERR(trans
)) {
3530 ret
= PTR_ERR(trans
);
3536 device
->disk_total_bytes
= new_size
;
3537 /* Now btrfs_update_device() will change the on-disk size. */
3538 ret
= btrfs_update_device(trans
, device
);
3540 unlock_chunks(root
);
3541 btrfs_end_transaction(trans
, root
);
3544 WARN_ON(diff
> old_total
);
3545 btrfs_set_super_total_bytes(super_copy
, old_total
- diff
);
3546 unlock_chunks(root
);
3547 btrfs_end_transaction(trans
, root
);
3549 btrfs_free_path(path
);
3553 static int btrfs_add_system_chunk(struct btrfs_root
*root
,
3554 struct btrfs_key
*key
,
3555 struct btrfs_chunk
*chunk
, int item_size
)
3557 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
3558 struct btrfs_disk_key disk_key
;
3562 array_size
= btrfs_super_sys_array_size(super_copy
);
3563 if (array_size
+ item_size
> BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
)
3566 ptr
= super_copy
->sys_chunk_array
+ array_size
;
3567 btrfs_cpu_key_to_disk(&disk_key
, key
);
3568 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
3569 ptr
+= sizeof(disk_key
);
3570 memcpy(ptr
, chunk
, item_size
);
3571 item_size
+= sizeof(disk_key
);
3572 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
3577 * sort the devices in descending order by max_avail, total_avail
3579 static int btrfs_cmp_device_info(const void *a
, const void *b
)
3581 const struct btrfs_device_info
*di_a
= a
;
3582 const struct btrfs_device_info
*di_b
= b
;
3584 if (di_a
->max_avail
> di_b
->max_avail
)
3586 if (di_a
->max_avail
< di_b
->max_avail
)
3588 if (di_a
->total_avail
> di_b
->total_avail
)
3590 if (di_a
->total_avail
< di_b
->total_avail
)
3595 struct btrfs_raid_attr btrfs_raid_array
[BTRFS_NR_RAID_TYPES
] = {
3596 [BTRFS_RAID_RAID10
] = {
3599 .devs_max
= 0, /* 0 == as many as possible */
3601 .devs_increment
= 2,
3604 [BTRFS_RAID_RAID1
] = {
3609 .devs_increment
= 2,
3612 [BTRFS_RAID_DUP
] = {
3617 .devs_increment
= 1,
3620 [BTRFS_RAID_RAID0
] = {
3625 .devs_increment
= 1,
3628 [BTRFS_RAID_SINGLE
] = {
3633 .devs_increment
= 1,
3636 [BTRFS_RAID_RAID5
] = {
3641 .devs_increment
= 1,
3644 [BTRFS_RAID_RAID6
] = {
3649 .devs_increment
= 1,
3654 static u32
find_raid56_stripe_len(u32 data_devices
, u32 dev_stripe_target
)
3656 /* TODO allow them to set a preferred stripe size */
3660 static void check_raid56_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
3664 if (!(type
& (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
)))
3667 features
= btrfs_super_incompat_flags(info
->super_copy
);
3668 if (features
& BTRFS_FEATURE_INCOMPAT_RAID56
)
3671 features
|= BTRFS_FEATURE_INCOMPAT_RAID56
;
3672 btrfs_set_super_incompat_flags(info
->super_copy
, features
);
3673 printk(KERN_INFO
"btrfs: setting RAID5/6 feature flag\n");
3676 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
3677 struct btrfs_root
*extent_root
,
3678 struct map_lookup
**map_ret
,
3679 u64
*num_bytes_out
, u64
*stripe_size_out
,
3680 u64 start
, u64 type
)
3682 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
3683 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
3684 struct list_head
*cur
;
3685 struct map_lookup
*map
= NULL
;
3686 struct extent_map_tree
*em_tree
;
3687 struct extent_map
*em
;
3688 struct btrfs_device_info
*devices_info
= NULL
;
3690 int num_stripes
; /* total number of stripes to allocate */
3691 int data_stripes
; /* number of stripes that count for
3693 int sub_stripes
; /* sub_stripes info for map */
3694 int dev_stripes
; /* stripes per dev */
3695 int devs_max
; /* max devs to use */
3696 int devs_min
; /* min devs needed */
3697 int devs_increment
; /* ndevs has to be a multiple of this */
3698 int ncopies
; /* how many copies to data has */
3700 u64 max_stripe_size
;
3704 u64 raid_stripe_len
= BTRFS_STRIPE_LEN
;
3710 BUG_ON(!alloc_profile_is_valid(type
, 0));
3712 if (list_empty(&fs_devices
->alloc_list
))
3715 index
= __get_raid_index(type
);
3717 sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
3718 dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
3719 devs_max
= btrfs_raid_array
[index
].devs_max
;
3720 devs_min
= btrfs_raid_array
[index
].devs_min
;
3721 devs_increment
= btrfs_raid_array
[index
].devs_increment
;
3722 ncopies
= btrfs_raid_array
[index
].ncopies
;
3724 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
3725 max_stripe_size
= 1024 * 1024 * 1024;
3726 max_chunk_size
= 10 * max_stripe_size
;
3727 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
3728 /* for larger filesystems, use larger metadata chunks */
3729 if (fs_devices
->total_rw_bytes
> 50ULL * 1024 * 1024 * 1024)
3730 max_stripe_size
= 1024 * 1024 * 1024;
3732 max_stripe_size
= 256 * 1024 * 1024;
3733 max_chunk_size
= max_stripe_size
;
3734 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3735 max_stripe_size
= 32 * 1024 * 1024;
3736 max_chunk_size
= 2 * max_stripe_size
;
3738 printk(KERN_ERR
"btrfs: invalid chunk type 0x%llx requested\n",
3743 /* we don't want a chunk larger than 10% of writeable space */
3744 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
3747 devices_info
= kzalloc(sizeof(*devices_info
) * fs_devices
->rw_devices
,
3752 cur
= fs_devices
->alloc_list
.next
;
3755 * in the first pass through the devices list, we gather information
3756 * about the available holes on each device.
3759 while (cur
!= &fs_devices
->alloc_list
) {
3760 struct btrfs_device
*device
;
3764 device
= list_entry(cur
, struct btrfs_device
, dev_alloc_list
);
3768 if (!device
->writeable
) {
3770 "btrfs: read-only device in alloc_list\n");
3774 if (!device
->in_fs_metadata
||
3775 device
->is_tgtdev_for_dev_replace
)
3778 if (device
->total_bytes
> device
->bytes_used
)
3779 total_avail
= device
->total_bytes
- device
->bytes_used
;
3783 /* If there is no space on this device, skip it. */
3784 if (total_avail
== 0)
3787 ret
= find_free_dev_extent(device
,
3788 max_stripe_size
* dev_stripes
,
3789 &dev_offset
, &max_avail
);
3790 if (ret
&& ret
!= -ENOSPC
)
3794 max_avail
= max_stripe_size
* dev_stripes
;
3796 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
3799 if (ndevs
== fs_devices
->rw_devices
) {
3800 WARN(1, "%s: found more than %llu devices\n",
3801 __func__
, fs_devices
->rw_devices
);
3804 devices_info
[ndevs
].dev_offset
= dev_offset
;
3805 devices_info
[ndevs
].max_avail
= max_avail
;
3806 devices_info
[ndevs
].total_avail
= total_avail
;
3807 devices_info
[ndevs
].dev
= device
;
3812 * now sort the devices by hole size / available space
3814 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
3815 btrfs_cmp_device_info
, NULL
);
3817 /* round down to number of usable stripes */
3818 ndevs
-= ndevs
% devs_increment
;
3820 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
3825 if (devs_max
&& ndevs
> devs_max
)
3828 * the primary goal is to maximize the number of stripes, so use as many
3829 * devices as possible, even if the stripes are not maximum sized.
3831 stripe_size
= devices_info
[ndevs
-1].max_avail
;
3832 num_stripes
= ndevs
* dev_stripes
;
3835 * this will have to be fixed for RAID1 and RAID10 over
3838 data_stripes
= num_stripes
/ ncopies
;
3840 if (type
& BTRFS_BLOCK_GROUP_RAID5
) {
3841 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 1,
3842 btrfs_super_stripesize(info
->super_copy
));
3843 data_stripes
= num_stripes
- 1;
3845 if (type
& BTRFS_BLOCK_GROUP_RAID6
) {
3846 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 2,
3847 btrfs_super_stripesize(info
->super_copy
));
3848 data_stripes
= num_stripes
- 2;
3852 * Use the number of data stripes to figure out how big this chunk
3853 * is really going to be in terms of logical address space,
3854 * and compare that answer with the max chunk size
3856 if (stripe_size
* data_stripes
> max_chunk_size
) {
3857 u64 mask
= (1ULL << 24) - 1;
3858 stripe_size
= max_chunk_size
;
3859 do_div(stripe_size
, data_stripes
);
3861 /* bump the answer up to a 16MB boundary */
3862 stripe_size
= (stripe_size
+ mask
) & ~mask
;
3864 /* but don't go higher than the limits we found
3865 * while searching for free extents
3867 if (stripe_size
> devices_info
[ndevs
-1].max_avail
)
3868 stripe_size
= devices_info
[ndevs
-1].max_avail
;
3871 do_div(stripe_size
, dev_stripes
);
3873 /* align to BTRFS_STRIPE_LEN */
3874 do_div(stripe_size
, raid_stripe_len
);
3875 stripe_size
*= raid_stripe_len
;
3877 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
3882 map
->num_stripes
= num_stripes
;
3884 for (i
= 0; i
< ndevs
; ++i
) {
3885 for (j
= 0; j
< dev_stripes
; ++j
) {
3886 int s
= i
* dev_stripes
+ j
;
3887 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
3888 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
3892 map
->sector_size
= extent_root
->sectorsize
;
3893 map
->stripe_len
= raid_stripe_len
;
3894 map
->io_align
= raid_stripe_len
;
3895 map
->io_width
= raid_stripe_len
;
3897 map
->sub_stripes
= sub_stripes
;
3900 num_bytes
= stripe_size
* data_stripes
;
3902 *stripe_size_out
= stripe_size
;
3903 *num_bytes_out
= num_bytes
;
3905 trace_btrfs_chunk_alloc(info
->chunk_root
, map
, start
, num_bytes
);
3907 em
= alloc_extent_map();
3912 em
->bdev
= (struct block_device
*)map
;
3914 em
->len
= num_bytes
;
3915 em
->block_start
= 0;
3916 em
->block_len
= em
->len
;
3918 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
3919 write_lock(&em_tree
->lock
);
3920 ret
= add_extent_mapping(em_tree
, em
);
3921 write_unlock(&em_tree
->lock
);
3923 free_extent_map(em
);
3927 for (i
= 0; i
< map
->num_stripes
; ++i
) {
3928 struct btrfs_device
*device
;
3931 device
= map
->stripes
[i
].dev
;
3932 dev_offset
= map
->stripes
[i
].physical
;
3934 ret
= btrfs_alloc_dev_extent(trans
, device
,
3935 info
->chunk_root
->root_key
.objectid
,
3936 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
3937 start
, dev_offset
, stripe_size
);
3939 goto error_dev_extent
;
3942 ret
= btrfs_make_block_group(trans
, extent_root
, 0, type
,
3943 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
3946 i
= map
->num_stripes
- 1;
3947 goto error_dev_extent
;
3950 free_extent_map(em
);
3951 check_raid56_incompat_flag(extent_root
->fs_info
, type
);
3953 kfree(devices_info
);
3957 for (; i
>= 0; i
--) {
3958 struct btrfs_device
*device
;
3961 device
= map
->stripes
[i
].dev
;
3962 err
= btrfs_free_dev_extent(trans
, device
, start
);
3964 btrfs_abort_transaction(trans
, extent_root
, err
);
3968 write_lock(&em_tree
->lock
);
3969 remove_extent_mapping(em_tree
, em
);
3970 write_unlock(&em_tree
->lock
);
3972 /* One for our allocation */
3973 free_extent_map(em
);
3974 /* One for the tree reference */
3975 free_extent_map(em
);
3978 kfree(devices_info
);
3982 static int __finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
3983 struct btrfs_root
*extent_root
,
3984 struct map_lookup
*map
, u64 chunk_offset
,
3985 u64 chunk_size
, u64 stripe_size
)
3988 struct btrfs_key key
;
3989 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
3990 struct btrfs_device
*device
;
3991 struct btrfs_chunk
*chunk
;
3992 struct btrfs_stripe
*stripe
;
3993 size_t item_size
= btrfs_chunk_item_size(map
->num_stripes
);
3997 chunk
= kzalloc(item_size
, GFP_NOFS
);
4002 while (index
< map
->num_stripes
) {
4003 device
= map
->stripes
[index
].dev
;
4004 device
->bytes_used
+= stripe_size
;
4005 ret
= btrfs_update_device(trans
, device
);
4011 spin_lock(&extent_root
->fs_info
->free_chunk_lock
);
4012 extent_root
->fs_info
->free_chunk_space
-= (stripe_size
*
4014 spin_unlock(&extent_root
->fs_info
->free_chunk_lock
);
4017 stripe
= &chunk
->stripe
;
4018 while (index
< map
->num_stripes
) {
4019 device
= map
->stripes
[index
].dev
;
4020 dev_offset
= map
->stripes
[index
].physical
;
4022 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
4023 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
4024 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
4029 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
4030 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
4031 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
4032 btrfs_set_stack_chunk_type(chunk
, map
->type
);
4033 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
4034 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
4035 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
4036 btrfs_set_stack_chunk_sector_size(chunk
, extent_root
->sectorsize
);
4037 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
4039 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
4040 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
4041 key
.offset
= chunk_offset
;
4043 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
4045 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4047 * TODO: Cleanup of inserted chunk root in case of
4050 ret
= btrfs_add_system_chunk(chunk_root
, &key
, chunk
,
4060 * Chunk allocation falls into two parts. The first part does works
4061 * that make the new allocated chunk useable, but not do any operation
4062 * that modifies the chunk tree. The second part does the works that
4063 * require modifying the chunk tree. This division is important for the
4064 * bootstrap process of adding storage to a seed btrfs.
4066 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4067 struct btrfs_root
*extent_root
, u64 type
)
4072 struct map_lookup
*map
;
4073 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
4076 ret
= find_next_chunk(chunk_root
, BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4081 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &map
, &chunk_size
,
4082 &stripe_size
, chunk_offset
, type
);
4086 ret
= __finish_chunk_alloc(trans
, extent_root
, map
, chunk_offset
,
4087 chunk_size
, stripe_size
);
4093 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
4094 struct btrfs_root
*root
,
4095 struct btrfs_device
*device
)
4098 u64 sys_chunk_offset
;
4102 u64 sys_stripe_size
;
4104 struct map_lookup
*map
;
4105 struct map_lookup
*sys_map
;
4106 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4107 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
4110 ret
= find_next_chunk(fs_info
->chunk_root
,
4111 BTRFS_FIRST_CHUNK_TREE_OBJECTID
, &chunk_offset
);
4115 alloc_profile
= btrfs_get_alloc_profile(extent_root
, 0);
4116 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &map
, &chunk_size
,
4117 &stripe_size
, chunk_offset
, alloc_profile
);
4121 sys_chunk_offset
= chunk_offset
+ chunk_size
;
4123 alloc_profile
= btrfs_get_alloc_profile(fs_info
->chunk_root
, 0);
4124 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &sys_map
,
4125 &sys_chunk_size
, &sys_stripe_size
,
4126 sys_chunk_offset
, alloc_profile
);
4128 btrfs_abort_transaction(trans
, root
, ret
);
4132 ret
= btrfs_add_device(trans
, fs_info
->chunk_root
, device
);
4134 btrfs_abort_transaction(trans
, root
, ret
);
4139 * Modifying chunk tree needs allocating new blocks from both
4140 * system block group and metadata block group. So we only can
4141 * do operations require modifying the chunk tree after both
4142 * block groups were created.
4144 ret
= __finish_chunk_alloc(trans
, extent_root
, map
, chunk_offset
,
4145 chunk_size
, stripe_size
);
4147 btrfs_abort_transaction(trans
, root
, ret
);
4151 ret
= __finish_chunk_alloc(trans
, extent_root
, sys_map
,
4152 sys_chunk_offset
, sys_chunk_size
,
4155 btrfs_abort_transaction(trans
, root
, ret
);
4162 int btrfs_chunk_readonly(struct btrfs_root
*root
, u64 chunk_offset
)
4164 struct extent_map
*em
;
4165 struct map_lookup
*map
;
4166 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
4170 read_lock(&map_tree
->map_tree
.lock
);
4171 em
= lookup_extent_mapping(&map_tree
->map_tree
, chunk_offset
, 1);
4172 read_unlock(&map_tree
->map_tree
.lock
);
4176 if (btrfs_test_opt(root
, DEGRADED
)) {
4177 free_extent_map(em
);
4181 map
= (struct map_lookup
*)em
->bdev
;
4182 for (i
= 0; i
< map
->num_stripes
; i
++) {
4183 if (!map
->stripes
[i
].dev
->writeable
) {
4188 free_extent_map(em
);
4192 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
4194 extent_map_tree_init(&tree
->map_tree
);
4197 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
4199 struct extent_map
*em
;
4202 write_lock(&tree
->map_tree
.lock
);
4203 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
4205 remove_extent_mapping(&tree
->map_tree
, em
);
4206 write_unlock(&tree
->map_tree
.lock
);
4211 free_extent_map(em
);
4212 /* once for the tree */
4213 free_extent_map(em
);
4217 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
4219 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
4220 struct extent_map
*em
;
4221 struct map_lookup
*map
;
4222 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4225 read_lock(&em_tree
->lock
);
4226 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4227 read_unlock(&em_tree
->lock
);
4230 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4231 map
= (struct map_lookup
*)em
->bdev
;
4232 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
4233 ret
= map
->num_stripes
;
4234 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
4235 ret
= map
->sub_stripes
;
4236 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID5
)
4238 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
4242 free_extent_map(em
);
4244 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
4245 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
))
4247 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
4252 unsigned long btrfs_full_stripe_len(struct btrfs_root
*root
,
4253 struct btrfs_mapping_tree
*map_tree
,
4256 struct extent_map
*em
;
4257 struct map_lookup
*map
;
4258 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4259 unsigned long len
= root
->sectorsize
;
4261 read_lock(&em_tree
->lock
);
4262 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4263 read_unlock(&em_tree
->lock
);
4266 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4267 map
= (struct map_lookup
*)em
->bdev
;
4268 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
4269 BTRFS_BLOCK_GROUP_RAID6
)) {
4270 len
= map
->stripe_len
* nr_data_stripes(map
);
4272 free_extent_map(em
);
4276 int btrfs_is_parity_mirror(struct btrfs_mapping_tree
*map_tree
,
4277 u64 logical
, u64 len
, int mirror_num
)
4279 struct extent_map
*em
;
4280 struct map_lookup
*map
;
4281 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4284 read_lock(&em_tree
->lock
);
4285 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4286 read_unlock(&em_tree
->lock
);
4289 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4290 map
= (struct map_lookup
*)em
->bdev
;
4291 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
4292 BTRFS_BLOCK_GROUP_RAID6
))
4294 free_extent_map(em
);
4298 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
4299 struct map_lookup
*map
, int first
, int num
,
4300 int optimal
, int dev_replace_is_ongoing
)
4304 struct btrfs_device
*srcdev
;
4306 if (dev_replace_is_ongoing
&&
4307 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
4308 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
4309 srcdev
= fs_info
->dev_replace
.srcdev
;
4314 * try to avoid the drive that is the source drive for a
4315 * dev-replace procedure, only choose it if no other non-missing
4316 * mirror is available
4318 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
4319 if (map
->stripes
[optimal
].dev
->bdev
&&
4320 (tolerance
|| map
->stripes
[optimal
].dev
!= srcdev
))
4322 for (i
= first
; i
< first
+ num
; i
++) {
4323 if (map
->stripes
[i
].dev
->bdev
&&
4324 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
4329 /* we couldn't find one that doesn't fail. Just return something
4330 * and the io error handling code will clean up eventually
4335 static inline int parity_smaller(u64 a
, u64 b
)
4340 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4341 static void sort_parity_stripes(struct btrfs_bio
*bbio
, u64
*raid_map
)
4343 struct btrfs_bio_stripe s
;
4350 for (i
= 0; i
< bbio
->num_stripes
- 1; i
++) {
4351 if (parity_smaller(raid_map
[i
], raid_map
[i
+1])) {
4352 s
= bbio
->stripes
[i
];
4354 bbio
->stripes
[i
] = bbio
->stripes
[i
+1];
4355 raid_map
[i
] = raid_map
[i
+1];
4356 bbio
->stripes
[i
+1] = s
;
4364 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
4365 u64 logical
, u64
*length
,
4366 struct btrfs_bio
**bbio_ret
,
4367 int mirror_num
, u64
**raid_map_ret
)
4369 struct extent_map
*em
;
4370 struct map_lookup
*map
;
4371 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
4372 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4375 u64 stripe_end_offset
;
4380 u64
*raid_map
= NULL
;
4386 struct btrfs_bio
*bbio
= NULL
;
4387 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
4388 int dev_replace_is_ongoing
= 0;
4389 int num_alloc_stripes
;
4390 int patch_the_first_stripe_for_dev_replace
= 0;
4391 u64 physical_to_patch_in_first_stripe
= 0;
4392 u64 raid56_full_stripe_start
= (u64
)-1;
4394 read_lock(&em_tree
->lock
);
4395 em
= lookup_extent_mapping(em_tree
, logical
, *length
);
4396 read_unlock(&em_tree
->lock
);
4399 printk(KERN_CRIT
"btrfs: unable to find logical %llu len %llu\n",
4400 (unsigned long long)logical
,
4401 (unsigned long long)*length
);
4405 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4406 map
= (struct map_lookup
*)em
->bdev
;
4407 offset
= logical
- em
->start
;
4409 if (mirror_num
> map
->num_stripes
)
4412 stripe_len
= map
->stripe_len
;
4415 * stripe_nr counts the total number of stripes we have to stride
4416 * to get to this block
4418 do_div(stripe_nr
, stripe_len
);
4420 stripe_offset
= stripe_nr
* stripe_len
;
4421 BUG_ON(offset
< stripe_offset
);
4423 /* stripe_offset is the offset of this block in its stripe*/
4424 stripe_offset
= offset
- stripe_offset
;
4426 /* if we're here for raid56, we need to know the stripe aligned start */
4427 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
)) {
4428 unsigned long full_stripe_len
= stripe_len
* nr_data_stripes(map
);
4429 raid56_full_stripe_start
= offset
;
4431 /* allow a write of a full stripe, but make sure we don't
4432 * allow straddling of stripes
4434 do_div(raid56_full_stripe_start
, full_stripe_len
);
4435 raid56_full_stripe_start
*= full_stripe_len
;
4438 if (rw
& REQ_DISCARD
) {
4439 /* we don't discard raid56 yet */
4441 (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
)) {
4445 *length
= min_t(u64
, em
->len
- offset
, *length
);
4446 } else if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
4448 /* For writes to RAID[56], allow a full stripeset across all disks.
4449 For other RAID types and for RAID[56] reads, just allow a single
4450 stripe (on a single disk). */
4451 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
) &&
4453 max_len
= stripe_len
* nr_data_stripes(map
) -
4454 (offset
- raid56_full_stripe_start
);
4456 /* we limit the length of each bio to what fits in a stripe */
4457 max_len
= stripe_len
- stripe_offset
;
4459 *length
= min_t(u64
, em
->len
- offset
, max_len
);
4461 *length
= em
->len
- offset
;
4464 /* This is for when we're called from btrfs_merge_bio_hook() and all
4465 it cares about is the length */
4469 btrfs_dev_replace_lock(dev_replace
);
4470 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
4471 if (!dev_replace_is_ongoing
)
4472 btrfs_dev_replace_unlock(dev_replace
);
4474 if (dev_replace_is_ongoing
&& mirror_num
== map
->num_stripes
+ 1 &&
4475 !(rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) &&
4476 dev_replace
->tgtdev
!= NULL
) {
4478 * in dev-replace case, for repair case (that's the only
4479 * case where the mirror is selected explicitly when
4480 * calling btrfs_map_block), blocks left of the left cursor
4481 * can also be read from the target drive.
4482 * For REQ_GET_READ_MIRRORS, the target drive is added as
4483 * the last one to the array of stripes. For READ, it also
4484 * needs to be supported using the same mirror number.
4485 * If the requested block is not left of the left cursor,
4486 * EIO is returned. This can happen because btrfs_num_copies()
4487 * returns one more in the dev-replace case.
4489 u64 tmp_length
= *length
;
4490 struct btrfs_bio
*tmp_bbio
= NULL
;
4491 int tmp_num_stripes
;
4492 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
4493 int index_srcdev
= 0;
4495 u64 physical_of_found
= 0;
4497 ret
= __btrfs_map_block(fs_info
, REQ_GET_READ_MIRRORS
,
4498 logical
, &tmp_length
, &tmp_bbio
, 0, NULL
);
4500 WARN_ON(tmp_bbio
!= NULL
);
4504 tmp_num_stripes
= tmp_bbio
->num_stripes
;
4505 if (mirror_num
> tmp_num_stripes
) {
4507 * REQ_GET_READ_MIRRORS does not contain this
4508 * mirror, that means that the requested area
4509 * is not left of the left cursor
4517 * process the rest of the function using the mirror_num
4518 * of the source drive. Therefore look it up first.
4519 * At the end, patch the device pointer to the one of the
4522 for (i
= 0; i
< tmp_num_stripes
; i
++) {
4523 if (tmp_bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
4525 * In case of DUP, in order to keep it
4526 * simple, only add the mirror with the
4527 * lowest physical address
4530 physical_of_found
<=
4531 tmp_bbio
->stripes
[i
].physical
)
4536 tmp_bbio
->stripes
[i
].physical
;
4541 mirror_num
= index_srcdev
+ 1;
4542 patch_the_first_stripe_for_dev_replace
= 1;
4543 physical_to_patch_in_first_stripe
= physical_of_found
;
4552 } else if (mirror_num
> map
->num_stripes
) {
4558 stripe_nr_orig
= stripe_nr
;
4559 stripe_nr_end
= (offset
+ *length
+ map
->stripe_len
- 1) &
4560 (~(map
->stripe_len
- 1));
4561 do_div(stripe_nr_end
, map
->stripe_len
);
4562 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
4565 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
4566 if (rw
& REQ_DISCARD
)
4567 num_stripes
= min_t(u64
, map
->num_stripes
,
4568 stripe_nr_end
- stripe_nr_orig
);
4569 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
4570 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
4571 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
))
4572 num_stripes
= map
->num_stripes
;
4573 else if (mirror_num
)
4574 stripe_index
= mirror_num
- 1;
4576 stripe_index
= find_live_mirror(fs_info
, map
, 0,
4578 current
->pid
% map
->num_stripes
,
4579 dev_replace_is_ongoing
);
4580 mirror_num
= stripe_index
+ 1;
4583 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
4584 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) {
4585 num_stripes
= map
->num_stripes
;
4586 } else if (mirror_num
) {
4587 stripe_index
= mirror_num
- 1;
4592 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
4593 int factor
= map
->num_stripes
/ map
->sub_stripes
;
4595 stripe_index
= do_div(stripe_nr
, factor
);
4596 stripe_index
*= map
->sub_stripes
;
4598 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
))
4599 num_stripes
= map
->sub_stripes
;
4600 else if (rw
& REQ_DISCARD
)
4601 num_stripes
= min_t(u64
, map
->sub_stripes
*
4602 (stripe_nr_end
- stripe_nr_orig
),
4604 else if (mirror_num
)
4605 stripe_index
+= mirror_num
- 1;
4607 int old_stripe_index
= stripe_index
;
4608 stripe_index
= find_live_mirror(fs_info
, map
,
4610 map
->sub_stripes
, stripe_index
+
4611 current
->pid
% map
->sub_stripes
,
4612 dev_replace_is_ongoing
);
4613 mirror_num
= stripe_index
- old_stripe_index
+ 1;
4616 } else if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
4617 BTRFS_BLOCK_GROUP_RAID6
)) {
4620 if (bbio_ret
&& ((rw
& REQ_WRITE
) || mirror_num
> 1)
4624 /* push stripe_nr back to the start of the full stripe */
4625 stripe_nr
= raid56_full_stripe_start
;
4626 do_div(stripe_nr
, stripe_len
);
4628 stripe_index
= do_div(stripe_nr
, nr_data_stripes(map
));
4630 /* RAID[56] write or recovery. Return all stripes */
4631 num_stripes
= map
->num_stripes
;
4632 max_errors
= nr_parity_stripes(map
);
4634 raid_map
= kmalloc(sizeof(u64
) * num_stripes
,
4641 /* Work out the disk rotation on this stripe-set */
4643 rot
= do_div(tmp
, num_stripes
);
4645 /* Fill in the logical address of each stripe */
4646 tmp
= stripe_nr
* nr_data_stripes(map
);
4647 for (i
= 0; i
< nr_data_stripes(map
); i
++)
4648 raid_map
[(i
+rot
) % num_stripes
] =
4649 em
->start
+ (tmp
+ i
) * map
->stripe_len
;
4651 raid_map
[(i
+rot
) % map
->num_stripes
] = RAID5_P_STRIPE
;
4652 if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
4653 raid_map
[(i
+rot
+1) % num_stripes
] =
4656 *length
= map
->stripe_len
;
4661 * Mirror #0 or #1 means the original data block.
4662 * Mirror #2 is RAID5 parity block.
4663 * Mirror #3 is RAID6 Q block.
4665 stripe_index
= do_div(stripe_nr
, nr_data_stripes(map
));
4667 stripe_index
= nr_data_stripes(map
) +
4670 /* We distribute the parity blocks across stripes */
4671 tmp
= stripe_nr
+ stripe_index
;
4672 stripe_index
= do_div(tmp
, map
->num_stripes
);
4676 * after this do_div call, stripe_nr is the number of stripes
4677 * on this device we have to walk to find the data, and
4678 * stripe_index is the number of our device in the stripe array
4680 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
4681 mirror_num
= stripe_index
+ 1;
4683 BUG_ON(stripe_index
>= map
->num_stripes
);
4685 num_alloc_stripes
= num_stripes
;
4686 if (dev_replace_is_ongoing
) {
4687 if (rw
& (REQ_WRITE
| REQ_DISCARD
))
4688 num_alloc_stripes
<<= 1;
4689 if (rw
& REQ_GET_READ_MIRRORS
)
4690 num_alloc_stripes
++;
4692 bbio
= kzalloc(btrfs_bio_size(num_alloc_stripes
), GFP_NOFS
);
4697 atomic_set(&bbio
->error
, 0);
4699 if (rw
& REQ_DISCARD
) {
4701 int sub_stripes
= 0;
4702 u64 stripes_per_dev
= 0;
4703 u32 remaining_stripes
= 0;
4704 u32 last_stripe
= 0;
4707 (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID10
)) {
4708 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
4711 sub_stripes
= map
->sub_stripes
;
4713 factor
= map
->num_stripes
/ sub_stripes
;
4714 stripes_per_dev
= div_u64_rem(stripe_nr_end
-
4717 &remaining_stripes
);
4718 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
4719 last_stripe
*= sub_stripes
;
4722 for (i
= 0; i
< num_stripes
; i
++) {
4723 bbio
->stripes
[i
].physical
=
4724 map
->stripes
[stripe_index
].physical
+
4725 stripe_offset
+ stripe_nr
* map
->stripe_len
;
4726 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
4728 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
4729 BTRFS_BLOCK_GROUP_RAID10
)) {
4730 bbio
->stripes
[i
].length
= stripes_per_dev
*
4733 if (i
/ sub_stripes
< remaining_stripes
)
4734 bbio
->stripes
[i
].length
+=
4738 * Special for the first stripe and
4741 * |-------|...|-------|
4745 if (i
< sub_stripes
)
4746 bbio
->stripes
[i
].length
-=
4749 if (stripe_index
>= last_stripe
&&
4750 stripe_index
<= (last_stripe
+
4752 bbio
->stripes
[i
].length
-=
4755 if (i
== sub_stripes
- 1)
4758 bbio
->stripes
[i
].length
= *length
;
4761 if (stripe_index
== map
->num_stripes
) {
4762 /* This could only happen for RAID0/10 */
4768 for (i
= 0; i
< num_stripes
; i
++) {
4769 bbio
->stripes
[i
].physical
=
4770 map
->stripes
[stripe_index
].physical
+
4772 stripe_nr
* map
->stripe_len
;
4773 bbio
->stripes
[i
].dev
=
4774 map
->stripes
[stripe_index
].dev
;
4779 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
)) {
4780 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
4781 BTRFS_BLOCK_GROUP_RAID10
|
4782 BTRFS_BLOCK_GROUP_RAID5
|
4783 BTRFS_BLOCK_GROUP_DUP
)) {
4785 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
) {
4790 if (dev_replace_is_ongoing
&& (rw
& (REQ_WRITE
| REQ_DISCARD
)) &&
4791 dev_replace
->tgtdev
!= NULL
) {
4792 int index_where_to_add
;
4793 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
4796 * duplicate the write operations while the dev replace
4797 * procedure is running. Since the copying of the old disk
4798 * to the new disk takes place at run time while the
4799 * filesystem is mounted writable, the regular write
4800 * operations to the old disk have to be duplicated to go
4801 * to the new disk as well.
4802 * Note that device->missing is handled by the caller, and
4803 * that the write to the old disk is already set up in the
4806 index_where_to_add
= num_stripes
;
4807 for (i
= 0; i
< num_stripes
; i
++) {
4808 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
4809 /* write to new disk, too */
4810 struct btrfs_bio_stripe
*new =
4811 bbio
->stripes
+ index_where_to_add
;
4812 struct btrfs_bio_stripe
*old
=
4815 new->physical
= old
->physical
;
4816 new->length
= old
->length
;
4817 new->dev
= dev_replace
->tgtdev
;
4818 index_where_to_add
++;
4822 num_stripes
= index_where_to_add
;
4823 } else if (dev_replace_is_ongoing
&& (rw
& REQ_GET_READ_MIRRORS
) &&
4824 dev_replace
->tgtdev
!= NULL
) {
4825 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
4826 int index_srcdev
= 0;
4828 u64 physical_of_found
= 0;
4831 * During the dev-replace procedure, the target drive can
4832 * also be used to read data in case it is needed to repair
4833 * a corrupt block elsewhere. This is possible if the
4834 * requested area is left of the left cursor. In this area,
4835 * the target drive is a full copy of the source drive.
4837 for (i
= 0; i
< num_stripes
; i
++) {
4838 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
4840 * In case of DUP, in order to keep it
4841 * simple, only add the mirror with the
4842 * lowest physical address
4845 physical_of_found
<=
4846 bbio
->stripes
[i
].physical
)
4850 physical_of_found
= bbio
->stripes
[i
].physical
;
4854 u64 length
= map
->stripe_len
;
4856 if (physical_of_found
+ length
<=
4857 dev_replace
->cursor_left
) {
4858 struct btrfs_bio_stripe
*tgtdev_stripe
=
4859 bbio
->stripes
+ num_stripes
;
4861 tgtdev_stripe
->physical
= physical_of_found
;
4862 tgtdev_stripe
->length
=
4863 bbio
->stripes
[index_srcdev
].length
;
4864 tgtdev_stripe
->dev
= dev_replace
->tgtdev
;
4872 bbio
->num_stripes
= num_stripes
;
4873 bbio
->max_errors
= max_errors
;
4874 bbio
->mirror_num
= mirror_num
;
4877 * this is the case that REQ_READ && dev_replace_is_ongoing &&
4878 * mirror_num == num_stripes + 1 && dev_replace target drive is
4879 * available as a mirror
4881 if (patch_the_first_stripe_for_dev_replace
&& num_stripes
> 0) {
4882 WARN_ON(num_stripes
> 1);
4883 bbio
->stripes
[0].dev
= dev_replace
->tgtdev
;
4884 bbio
->stripes
[0].physical
= physical_to_patch_in_first_stripe
;
4885 bbio
->mirror_num
= map
->num_stripes
+ 1;
4888 sort_parity_stripes(bbio
, raid_map
);
4889 *raid_map_ret
= raid_map
;
4892 if (dev_replace_is_ongoing
)
4893 btrfs_dev_replace_unlock(dev_replace
);
4894 free_extent_map(em
);
4898 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
4899 u64 logical
, u64
*length
,
4900 struct btrfs_bio
**bbio_ret
, int mirror_num
)
4902 return __btrfs_map_block(fs_info
, rw
, logical
, length
, bbio_ret
,
4906 int btrfs_rmap_block(struct btrfs_mapping_tree
*map_tree
,
4907 u64 chunk_start
, u64 physical
, u64 devid
,
4908 u64
**logical
, int *naddrs
, int *stripe_len
)
4910 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4911 struct extent_map
*em
;
4912 struct map_lookup
*map
;
4920 read_lock(&em_tree
->lock
);
4921 em
= lookup_extent_mapping(em_tree
, chunk_start
, 1);
4922 read_unlock(&em_tree
->lock
);
4924 BUG_ON(!em
|| em
->start
!= chunk_start
);
4925 map
= (struct map_lookup
*)em
->bdev
;
4928 rmap_len
= map
->stripe_len
;
4930 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
4931 do_div(length
, map
->num_stripes
/ map
->sub_stripes
);
4932 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
4933 do_div(length
, map
->num_stripes
);
4934 else if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
4935 BTRFS_BLOCK_GROUP_RAID6
)) {
4936 do_div(length
, nr_data_stripes(map
));
4937 rmap_len
= map
->stripe_len
* nr_data_stripes(map
);
4940 buf
= kzalloc(sizeof(u64
) * map
->num_stripes
, GFP_NOFS
);
4941 BUG_ON(!buf
); /* -ENOMEM */
4943 for (i
= 0; i
< map
->num_stripes
; i
++) {
4944 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
4946 if (map
->stripes
[i
].physical
> physical
||
4947 map
->stripes
[i
].physical
+ length
<= physical
)
4950 stripe_nr
= physical
- map
->stripes
[i
].physical
;
4951 do_div(stripe_nr
, map
->stripe_len
);
4953 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
4954 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
4955 do_div(stripe_nr
, map
->sub_stripes
);
4956 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
4957 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
4958 } /* else if RAID[56], multiply by nr_data_stripes().
4959 * Alternatively, just use rmap_len below instead of
4960 * map->stripe_len */
4962 bytenr
= chunk_start
+ stripe_nr
* rmap_len
;
4963 WARN_ON(nr
>= map
->num_stripes
);
4964 for (j
= 0; j
< nr
; j
++) {
4965 if (buf
[j
] == bytenr
)
4969 WARN_ON(nr
>= map
->num_stripes
);
4976 *stripe_len
= rmap_len
;
4978 free_extent_map(em
);
4982 static void *merge_stripe_index_into_bio_private(void *bi_private
,
4983 unsigned int stripe_index
)
4986 * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
4988 * The alternative solution (instead of stealing bits from the
4989 * pointer) would be to allocate an intermediate structure
4990 * that contains the old private pointer plus the stripe_index.
4992 BUG_ON((((uintptr_t)bi_private
) & 3) != 0);
4993 BUG_ON(stripe_index
> 3);
4994 return (void *)(((uintptr_t)bi_private
) | stripe_index
);
4997 static struct btrfs_bio
*extract_bbio_from_bio_private(void *bi_private
)
4999 return (struct btrfs_bio
*)(((uintptr_t)bi_private
) & ~((uintptr_t)3));
5002 static unsigned int extract_stripe_index_from_bio_private(void *bi_private
)
5004 return (unsigned int)((uintptr_t)bi_private
) & 3;
5007 static void btrfs_end_bio(struct bio
*bio
, int err
)
5009 struct btrfs_bio
*bbio
= extract_bbio_from_bio_private(bio
->bi_private
);
5010 int is_orig_bio
= 0;
5013 atomic_inc(&bbio
->error
);
5014 if (err
== -EIO
|| err
== -EREMOTEIO
) {
5015 unsigned int stripe_index
=
5016 extract_stripe_index_from_bio_private(
5018 struct btrfs_device
*dev
;
5020 BUG_ON(stripe_index
>= bbio
->num_stripes
);
5021 dev
= bbio
->stripes
[stripe_index
].dev
;
5023 if (bio
->bi_rw
& WRITE
)
5024 btrfs_dev_stat_inc(dev
,
5025 BTRFS_DEV_STAT_WRITE_ERRS
);
5027 btrfs_dev_stat_inc(dev
,
5028 BTRFS_DEV_STAT_READ_ERRS
);
5029 if ((bio
->bi_rw
& WRITE_FLUSH
) == WRITE_FLUSH
)
5030 btrfs_dev_stat_inc(dev
,
5031 BTRFS_DEV_STAT_FLUSH_ERRS
);
5032 btrfs_dev_stat_print_on_error(dev
);
5037 if (bio
== bbio
->orig_bio
)
5040 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
5043 bio
= bbio
->orig_bio
;
5045 bio
->bi_private
= bbio
->private;
5046 bio
->bi_end_io
= bbio
->end_io
;
5047 bio
->bi_bdev
= (struct block_device
*)
5048 (unsigned long)bbio
->mirror_num
;
5049 /* only send an error to the higher layers if it is
5050 * beyond the tolerance of the btrfs bio
5052 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
5056 * this bio is actually up to date, we didn't
5057 * go over the max number of errors
5059 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5064 bio_endio(bio
, err
);
5065 } else if (!is_orig_bio
) {
5070 struct async_sched
{
5073 struct btrfs_fs_info
*info
;
5074 struct btrfs_work work
;
5078 * see run_scheduled_bios for a description of why bios are collected for
5081 * This will add one bio to the pending list for a device and make sure
5082 * the work struct is scheduled.
5084 noinline
void btrfs_schedule_bio(struct btrfs_root
*root
,
5085 struct btrfs_device
*device
,
5086 int rw
, struct bio
*bio
)
5088 int should_queue
= 1;
5089 struct btrfs_pending_bios
*pending_bios
;
5091 if (device
->missing
|| !device
->bdev
) {
5092 bio_endio(bio
, -EIO
);
5096 /* don't bother with additional async steps for reads, right now */
5097 if (!(rw
& REQ_WRITE
)) {
5099 btrfsic_submit_bio(rw
, bio
);
5105 * nr_async_bios allows us to reliably return congestion to the
5106 * higher layers. Otherwise, the async bio makes it appear we have
5107 * made progress against dirty pages when we've really just put it
5108 * on a queue for later
5110 atomic_inc(&root
->fs_info
->nr_async_bios
);
5111 WARN_ON(bio
->bi_next
);
5112 bio
->bi_next
= NULL
;
5115 spin_lock(&device
->io_lock
);
5116 if (bio
->bi_rw
& REQ_SYNC
)
5117 pending_bios
= &device
->pending_sync_bios
;
5119 pending_bios
= &device
->pending_bios
;
5121 if (pending_bios
->tail
)
5122 pending_bios
->tail
->bi_next
= bio
;
5124 pending_bios
->tail
= bio
;
5125 if (!pending_bios
->head
)
5126 pending_bios
->head
= bio
;
5127 if (device
->running_pending
)
5130 spin_unlock(&device
->io_lock
);
5133 btrfs_queue_worker(&root
->fs_info
->submit_workers
,
5137 static int bio_size_ok(struct block_device
*bdev
, struct bio
*bio
,
5140 struct bio_vec
*prev
;
5141 struct request_queue
*q
= bdev_get_queue(bdev
);
5142 unsigned short max_sectors
= queue_max_sectors(q
);
5143 struct bvec_merge_data bvm
= {
5145 .bi_sector
= sector
,
5146 .bi_rw
= bio
->bi_rw
,
5149 if (bio
->bi_vcnt
== 0) {
5154 prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
5155 if ((bio
->bi_size
>> 9) > max_sectors
)
5158 if (!q
->merge_bvec_fn
)
5161 bvm
.bi_size
= bio
->bi_size
- prev
->bv_len
;
5162 if (q
->merge_bvec_fn(q
, &bvm
, prev
) < prev
->bv_len
)
5167 static void submit_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
5168 struct bio
*bio
, u64 physical
, int dev_nr
,
5171 struct btrfs_device
*dev
= bbio
->stripes
[dev_nr
].dev
;
5173 bio
->bi_private
= bbio
;
5174 bio
->bi_private
= merge_stripe_index_into_bio_private(
5175 bio
->bi_private
, (unsigned int)dev_nr
);
5176 bio
->bi_end_io
= btrfs_end_bio
;
5177 bio
->bi_sector
= physical
>> 9;
5180 struct rcu_string
*name
;
5183 name
= rcu_dereference(dev
->name
);
5184 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5185 "(%s id %llu), size=%u\n", rw
,
5186 (u64
)bio
->bi_sector
, (u_long
)dev
->bdev
->bd_dev
,
5187 name
->str
, dev
->devid
, bio
->bi_size
);
5191 bio
->bi_bdev
= dev
->bdev
;
5193 btrfs_schedule_bio(root
, dev
, rw
, bio
);
5195 btrfsic_submit_bio(rw
, bio
);
5198 static int breakup_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
5199 struct bio
*first_bio
, struct btrfs_device
*dev
,
5200 int dev_nr
, int rw
, int async
)
5202 struct bio_vec
*bvec
= first_bio
->bi_io_vec
;
5204 int nr_vecs
= bio_get_nr_vecs(dev
->bdev
);
5205 u64 physical
= bbio
->stripes
[dev_nr
].physical
;
5208 bio
= btrfs_bio_alloc(dev
->bdev
, physical
>> 9, nr_vecs
, GFP_NOFS
);
5212 while (bvec
<= (first_bio
->bi_io_vec
+ first_bio
->bi_vcnt
- 1)) {
5213 if (bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
5214 bvec
->bv_offset
) < bvec
->bv_len
) {
5215 u64 len
= bio
->bi_size
;
5217 atomic_inc(&bbio
->stripes_pending
);
5218 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
,
5226 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
, rw
, async
);
5230 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
5232 atomic_inc(&bbio
->error
);
5233 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
5234 bio
->bi_private
= bbio
->private;
5235 bio
->bi_end_io
= bbio
->end_io
;
5236 bio
->bi_bdev
= (struct block_device
*)
5237 (unsigned long)bbio
->mirror_num
;
5238 bio
->bi_sector
= logical
>> 9;
5240 bio_endio(bio
, -EIO
);
5244 int btrfs_map_bio(struct btrfs_root
*root
, int rw
, struct bio
*bio
,
5245 int mirror_num
, int async_submit
)
5247 struct btrfs_device
*dev
;
5248 struct bio
*first_bio
= bio
;
5249 u64 logical
= (u64
)bio
->bi_sector
<< 9;
5252 u64
*raid_map
= NULL
;
5256 struct btrfs_bio
*bbio
= NULL
;
5258 length
= bio
->bi_size
;
5259 map_length
= length
;
5261 ret
= __btrfs_map_block(root
->fs_info
, rw
, logical
, &map_length
, &bbio
,
5262 mirror_num
, &raid_map
);
5263 if (ret
) /* -ENOMEM */
5266 total_devs
= bbio
->num_stripes
;
5267 bbio
->orig_bio
= first_bio
;
5268 bbio
->private = first_bio
->bi_private
;
5269 bbio
->end_io
= first_bio
->bi_end_io
;
5270 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
5273 /* In this case, map_length has been set to the length of
5274 a single stripe; not the whole write */
5276 return raid56_parity_write(root
, bio
, bbio
,
5277 raid_map
, map_length
);
5279 return raid56_parity_recover(root
, bio
, bbio
,
5280 raid_map
, map_length
,
5285 if (map_length
< length
) {
5286 printk(KERN_CRIT
"btrfs: mapping failed logical %llu bio len %llu "
5287 "len %llu\n", (unsigned long long)logical
,
5288 (unsigned long long)length
,
5289 (unsigned long long)map_length
);
5293 while (dev_nr
< total_devs
) {
5294 dev
= bbio
->stripes
[dev_nr
].dev
;
5295 if (!dev
|| !dev
->bdev
|| (rw
& WRITE
&& !dev
->writeable
)) {
5296 bbio_error(bbio
, first_bio
, logical
);
5302 * Check and see if we're ok with this bio based on it's size
5303 * and offset with the given device.
5305 if (!bio_size_ok(dev
->bdev
, first_bio
,
5306 bbio
->stripes
[dev_nr
].physical
>> 9)) {
5307 ret
= breakup_stripe_bio(root
, bbio
, first_bio
, dev
,
5308 dev_nr
, rw
, async_submit
);
5314 if (dev_nr
< total_devs
- 1) {
5315 bio
= bio_clone(first_bio
, GFP_NOFS
);
5316 BUG_ON(!bio
); /* -ENOMEM */
5321 submit_stripe_bio(root
, bbio
, bio
,
5322 bbio
->stripes
[dev_nr
].physical
, dev_nr
, rw
,
5329 struct btrfs_device
*btrfs_find_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
5332 struct btrfs_device
*device
;
5333 struct btrfs_fs_devices
*cur_devices
;
5335 cur_devices
= fs_info
->fs_devices
;
5336 while (cur_devices
) {
5338 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
5339 device
= __find_device(&cur_devices
->devices
,
5344 cur_devices
= cur_devices
->seed
;
5349 static struct btrfs_device
*add_missing_dev(struct btrfs_root
*root
,
5350 u64 devid
, u8
*dev_uuid
)
5352 struct btrfs_device
*device
;
5353 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
5355 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
5358 list_add(&device
->dev_list
,
5359 &fs_devices
->devices
);
5360 device
->dev_root
= root
->fs_info
->dev_root
;
5361 device
->devid
= devid
;
5362 device
->work
.func
= pending_bios_fn
;
5363 device
->fs_devices
= fs_devices
;
5364 device
->missing
= 1;
5365 fs_devices
->num_devices
++;
5366 fs_devices
->missing_devices
++;
5367 spin_lock_init(&device
->io_lock
);
5368 INIT_LIST_HEAD(&device
->dev_alloc_list
);
5369 memcpy(device
->uuid
, dev_uuid
, BTRFS_UUID_SIZE
);
5373 static int read_one_chunk(struct btrfs_root
*root
, struct btrfs_key
*key
,
5374 struct extent_buffer
*leaf
,
5375 struct btrfs_chunk
*chunk
)
5377 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
5378 struct map_lookup
*map
;
5379 struct extent_map
*em
;
5383 u8 uuid
[BTRFS_UUID_SIZE
];
5388 logical
= key
->offset
;
5389 length
= btrfs_chunk_length(leaf
, chunk
);
5391 read_lock(&map_tree
->map_tree
.lock
);
5392 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
5393 read_unlock(&map_tree
->map_tree
.lock
);
5395 /* already mapped? */
5396 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
5397 free_extent_map(em
);
5400 free_extent_map(em
);
5403 em
= alloc_extent_map();
5406 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
5407 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
5409 free_extent_map(em
);
5413 em
->bdev
= (struct block_device
*)map
;
5414 em
->start
= logical
;
5417 em
->block_start
= 0;
5418 em
->block_len
= em
->len
;
5420 map
->num_stripes
= num_stripes
;
5421 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
5422 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
5423 map
->sector_size
= btrfs_chunk_sector_size(leaf
, chunk
);
5424 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
5425 map
->type
= btrfs_chunk_type(leaf
, chunk
);
5426 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
5427 for (i
= 0; i
< num_stripes
; i
++) {
5428 map
->stripes
[i
].physical
=
5429 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
5430 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
5431 read_extent_buffer(leaf
, uuid
, (unsigned long)
5432 btrfs_stripe_dev_uuid_nr(chunk
, i
),
5434 map
->stripes
[i
].dev
= btrfs_find_device(root
->fs_info
, devid
,
5436 if (!map
->stripes
[i
].dev
&& !btrfs_test_opt(root
, DEGRADED
)) {
5438 free_extent_map(em
);
5441 if (!map
->stripes
[i
].dev
) {
5442 map
->stripes
[i
].dev
=
5443 add_missing_dev(root
, devid
, uuid
);
5444 if (!map
->stripes
[i
].dev
) {
5446 free_extent_map(em
);
5450 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
5453 write_lock(&map_tree
->map_tree
.lock
);
5454 ret
= add_extent_mapping(&map_tree
->map_tree
, em
);
5455 write_unlock(&map_tree
->map_tree
.lock
);
5456 BUG_ON(ret
); /* Tree corruption */
5457 free_extent_map(em
);
5462 static void fill_device_from_item(struct extent_buffer
*leaf
,
5463 struct btrfs_dev_item
*dev_item
,
5464 struct btrfs_device
*device
)
5468 device
->devid
= btrfs_device_id(leaf
, dev_item
);
5469 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
5470 device
->total_bytes
= device
->disk_total_bytes
;
5471 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
5472 device
->type
= btrfs_device_type(leaf
, dev_item
);
5473 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
5474 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
5475 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
5476 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
5477 device
->is_tgtdev_for_dev_replace
= 0;
5479 ptr
= (unsigned long)btrfs_device_uuid(dev_item
);
5480 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
5483 static int open_seed_devices(struct btrfs_root
*root
, u8
*fsid
)
5485 struct btrfs_fs_devices
*fs_devices
;
5488 BUG_ON(!mutex_is_locked(&uuid_mutex
));
5490 fs_devices
= root
->fs_info
->fs_devices
->seed
;
5491 while (fs_devices
) {
5492 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
5496 fs_devices
= fs_devices
->seed
;
5499 fs_devices
= find_fsid(fsid
);
5505 fs_devices
= clone_fs_devices(fs_devices
);
5506 if (IS_ERR(fs_devices
)) {
5507 ret
= PTR_ERR(fs_devices
);
5511 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
5512 root
->fs_info
->bdev_holder
);
5514 free_fs_devices(fs_devices
);
5518 if (!fs_devices
->seeding
) {
5519 __btrfs_close_devices(fs_devices
);
5520 free_fs_devices(fs_devices
);
5525 fs_devices
->seed
= root
->fs_info
->fs_devices
->seed
;
5526 root
->fs_info
->fs_devices
->seed
= fs_devices
;
5531 static int read_one_dev(struct btrfs_root
*root
,
5532 struct extent_buffer
*leaf
,
5533 struct btrfs_dev_item
*dev_item
)
5535 struct btrfs_device
*device
;
5538 u8 fs_uuid
[BTRFS_UUID_SIZE
];
5539 u8 dev_uuid
[BTRFS_UUID_SIZE
];
5541 devid
= btrfs_device_id(leaf
, dev_item
);
5542 read_extent_buffer(leaf
, dev_uuid
,
5543 (unsigned long)btrfs_device_uuid(dev_item
),
5545 read_extent_buffer(leaf
, fs_uuid
,
5546 (unsigned long)btrfs_device_fsid(dev_item
),
5549 if (memcmp(fs_uuid
, root
->fs_info
->fsid
, BTRFS_UUID_SIZE
)) {
5550 ret
= open_seed_devices(root
, fs_uuid
);
5551 if (ret
&& !btrfs_test_opt(root
, DEGRADED
))
5555 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
, fs_uuid
);
5556 if (!device
|| !device
->bdev
) {
5557 if (!btrfs_test_opt(root
, DEGRADED
))
5561 printk(KERN_WARNING
"warning devid %llu missing\n",
5562 (unsigned long long)devid
);
5563 device
= add_missing_dev(root
, devid
, dev_uuid
);
5566 } else if (!device
->missing
) {
5568 * this happens when a device that was properly setup
5569 * in the device info lists suddenly goes bad.
5570 * device->bdev is NULL, and so we have to set
5571 * device->missing to one here
5573 root
->fs_info
->fs_devices
->missing_devices
++;
5574 device
->missing
= 1;
5578 if (device
->fs_devices
!= root
->fs_info
->fs_devices
) {
5579 BUG_ON(device
->writeable
);
5580 if (device
->generation
!=
5581 btrfs_device_generation(leaf
, dev_item
))
5585 fill_device_from_item(leaf
, dev_item
, device
);
5586 device
->dev_root
= root
->fs_info
->dev_root
;
5587 device
->in_fs_metadata
= 1;
5588 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
5589 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
5590 spin_lock(&root
->fs_info
->free_chunk_lock
);
5591 root
->fs_info
->free_chunk_space
+= device
->total_bytes
-
5593 spin_unlock(&root
->fs_info
->free_chunk_lock
);
5599 int btrfs_read_sys_array(struct btrfs_root
*root
)
5601 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
5602 struct extent_buffer
*sb
;
5603 struct btrfs_disk_key
*disk_key
;
5604 struct btrfs_chunk
*chunk
;
5606 unsigned long sb_ptr
;
5612 struct btrfs_key key
;
5614 sb
= btrfs_find_create_tree_block(root
, BTRFS_SUPER_INFO_OFFSET
,
5615 BTRFS_SUPER_INFO_SIZE
);
5618 btrfs_set_buffer_uptodate(sb
);
5619 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
5621 * The sb extent buffer is artifical and just used to read the system array.
5622 * btrfs_set_buffer_uptodate() call does not properly mark all it's
5623 * pages up-to-date when the page is larger: extent does not cover the
5624 * whole page and consequently check_page_uptodate does not find all
5625 * the page's extents up-to-date (the hole beyond sb),
5626 * write_extent_buffer then triggers a WARN_ON.
5628 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5629 * but sb spans only this function. Add an explicit SetPageUptodate call
5630 * to silence the warning eg. on PowerPC 64.
5632 if (PAGE_CACHE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
5633 SetPageUptodate(sb
->pages
[0]);
5635 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
5636 array_size
= btrfs_super_sys_array_size(super_copy
);
5638 ptr
= super_copy
->sys_chunk_array
;
5639 sb_ptr
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
5642 while (cur
< array_size
) {
5643 disk_key
= (struct btrfs_disk_key
*)ptr
;
5644 btrfs_disk_key_to_cpu(&key
, disk_key
);
5646 len
= sizeof(*disk_key
); ptr
+= len
;
5650 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
5651 chunk
= (struct btrfs_chunk
*)sb_ptr
;
5652 ret
= read_one_chunk(root
, &key
, sb
, chunk
);
5655 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
5656 len
= btrfs_chunk_item_size(num_stripes
);
5665 free_extent_buffer(sb
);
5669 int btrfs_read_chunk_tree(struct btrfs_root
*root
)
5671 struct btrfs_path
*path
;
5672 struct extent_buffer
*leaf
;
5673 struct btrfs_key key
;
5674 struct btrfs_key found_key
;
5678 root
= root
->fs_info
->chunk_root
;
5680 path
= btrfs_alloc_path();
5684 mutex_lock(&uuid_mutex
);
5687 /* first we search for all of the device items, and then we
5688 * read in all of the chunk items. This way we can create chunk
5689 * mappings that reference all of the devices that are afound
5691 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
5695 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5699 leaf
= path
->nodes
[0];
5700 slot
= path
->slots
[0];
5701 if (slot
>= btrfs_header_nritems(leaf
)) {
5702 ret
= btrfs_next_leaf(root
, path
);
5709 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
5710 if (key
.objectid
== BTRFS_DEV_ITEMS_OBJECTID
) {
5711 if (found_key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
)
5713 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
5714 struct btrfs_dev_item
*dev_item
;
5715 dev_item
= btrfs_item_ptr(leaf
, slot
,
5716 struct btrfs_dev_item
);
5717 ret
= read_one_dev(root
, leaf
, dev_item
);
5721 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
5722 struct btrfs_chunk
*chunk
;
5723 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
5724 ret
= read_one_chunk(root
, &found_key
, leaf
, chunk
);
5730 if (key
.objectid
== BTRFS_DEV_ITEMS_OBJECTID
) {
5732 btrfs_release_path(path
);
5737 unlock_chunks(root
);
5738 mutex_unlock(&uuid_mutex
);
5740 btrfs_free_path(path
);
5744 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
5748 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
5749 btrfs_dev_stat_reset(dev
, i
);
5752 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
5754 struct btrfs_key key
;
5755 struct btrfs_key found_key
;
5756 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
5757 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
5758 struct extent_buffer
*eb
;
5761 struct btrfs_device
*device
;
5762 struct btrfs_path
*path
= NULL
;
5765 path
= btrfs_alloc_path();
5771 mutex_lock(&fs_devices
->device_list_mutex
);
5772 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
5774 struct btrfs_dev_stats_item
*ptr
;
5777 key
.type
= BTRFS_DEV_STATS_KEY
;
5778 key
.offset
= device
->devid
;
5779 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
5781 __btrfs_reset_dev_stats(device
);
5782 device
->dev_stats_valid
= 1;
5783 btrfs_release_path(path
);
5786 slot
= path
->slots
[0];
5787 eb
= path
->nodes
[0];
5788 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
5789 item_size
= btrfs_item_size_nr(eb
, slot
);
5791 ptr
= btrfs_item_ptr(eb
, slot
,
5792 struct btrfs_dev_stats_item
);
5794 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
5795 if (item_size
>= (1 + i
) * sizeof(__le64
))
5796 btrfs_dev_stat_set(device
, i
,
5797 btrfs_dev_stats_value(eb
, ptr
, i
));
5799 btrfs_dev_stat_reset(device
, i
);
5802 device
->dev_stats_valid
= 1;
5803 btrfs_dev_stat_print_on_load(device
);
5804 btrfs_release_path(path
);
5806 mutex_unlock(&fs_devices
->device_list_mutex
);
5809 btrfs_free_path(path
);
5810 return ret
< 0 ? ret
: 0;
5813 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
5814 struct btrfs_root
*dev_root
,
5815 struct btrfs_device
*device
)
5817 struct btrfs_path
*path
;
5818 struct btrfs_key key
;
5819 struct extent_buffer
*eb
;
5820 struct btrfs_dev_stats_item
*ptr
;
5825 key
.type
= BTRFS_DEV_STATS_KEY
;
5826 key
.offset
= device
->devid
;
5828 path
= btrfs_alloc_path();
5830 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
5832 printk_in_rcu(KERN_WARNING
"btrfs: error %d while searching for dev_stats item for device %s!\n",
5833 ret
, rcu_str_deref(device
->name
));
5838 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
5839 /* need to delete old one and insert a new one */
5840 ret
= btrfs_del_item(trans
, dev_root
, path
);
5842 printk_in_rcu(KERN_WARNING
"btrfs: delete too small dev_stats item for device %s failed %d!\n",
5843 rcu_str_deref(device
->name
), ret
);
5850 /* need to insert a new item */
5851 btrfs_release_path(path
);
5852 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
5853 &key
, sizeof(*ptr
));
5855 printk_in_rcu(KERN_WARNING
"btrfs: insert dev_stats item for device %s failed %d!\n",
5856 rcu_str_deref(device
->name
), ret
);
5861 eb
= path
->nodes
[0];
5862 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
5863 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
5864 btrfs_set_dev_stats_value(eb
, ptr
, i
,
5865 btrfs_dev_stat_read(device
, i
));
5866 btrfs_mark_buffer_dirty(eb
);
5869 btrfs_free_path(path
);
5874 * called from commit_transaction. Writes all changed device stats to disk.
5876 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
5877 struct btrfs_fs_info
*fs_info
)
5879 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
5880 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
5881 struct btrfs_device
*device
;
5884 mutex_lock(&fs_devices
->device_list_mutex
);
5885 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
5886 if (!device
->dev_stats_valid
|| !device
->dev_stats_dirty
)
5889 ret
= update_dev_stat_item(trans
, dev_root
, device
);
5891 device
->dev_stats_dirty
= 0;
5893 mutex_unlock(&fs_devices
->device_list_mutex
);
5898 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
5900 btrfs_dev_stat_inc(dev
, index
);
5901 btrfs_dev_stat_print_on_error(dev
);
5904 void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
5906 if (!dev
->dev_stats_valid
)
5908 printk_ratelimited_in_rcu(KERN_ERR
5909 "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5910 rcu_str_deref(dev
->name
),
5911 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
5912 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
5913 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
5914 btrfs_dev_stat_read(dev
,
5915 BTRFS_DEV_STAT_CORRUPTION_ERRS
),
5916 btrfs_dev_stat_read(dev
,
5917 BTRFS_DEV_STAT_GENERATION_ERRS
));
5920 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
5924 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
5925 if (btrfs_dev_stat_read(dev
, i
) != 0)
5927 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
5928 return; /* all values == 0, suppress message */
5930 printk_in_rcu(KERN_INFO
"btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5931 rcu_str_deref(dev
->name
),
5932 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
5933 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
5934 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
5935 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
5936 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
5939 int btrfs_get_dev_stats(struct btrfs_root
*root
,
5940 struct btrfs_ioctl_get_dev_stats
*stats
)
5942 struct btrfs_device
*dev
;
5943 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
5946 mutex_lock(&fs_devices
->device_list_mutex
);
5947 dev
= btrfs_find_device(root
->fs_info
, stats
->devid
, NULL
, NULL
);
5948 mutex_unlock(&fs_devices
->device_list_mutex
);
5952 "btrfs: get dev_stats failed, device not found\n");
5954 } else if (!dev
->dev_stats_valid
) {
5956 "btrfs: get dev_stats failed, not yet valid\n");
5958 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
5959 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
5960 if (stats
->nr_items
> i
)
5962 btrfs_dev_stat_read_and_reset(dev
, i
);
5964 btrfs_dev_stat_reset(dev
, i
);
5967 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
5968 if (stats
->nr_items
> i
)
5969 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
5971 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
5972 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
5976 int btrfs_scratch_superblock(struct btrfs_device
*device
)
5978 struct buffer_head
*bh
;
5979 struct btrfs_super_block
*disk_super
;
5981 bh
= btrfs_read_dev_super(device
->bdev
);
5984 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
5986 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
5987 set_buffer_dirty(bh
);
5988 sync_dirty_buffer(bh
);