2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/ratelimit.h>
27 #include <linux/kthread.h>
30 #include "extent_map.h"
32 #include "transaction.h"
33 #include "print-tree.h"
35 #include "async-thread.h"
36 #include "check-integrity.h"
37 #include "rcu-string.h"
39 #include "dev-replace.h"
41 static int init_first_rw_device(struct btrfs_trans_handle
*trans
,
42 struct btrfs_root
*root
,
43 struct btrfs_device
*device
);
44 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
);
45 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
);
46 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*device
);
48 static DEFINE_MUTEX(uuid_mutex
);
49 static LIST_HEAD(fs_uuids
);
51 static void lock_chunks(struct btrfs_root
*root
)
53 mutex_lock(&root
->fs_info
->chunk_mutex
);
56 static void unlock_chunks(struct btrfs_root
*root
)
58 mutex_unlock(&root
->fs_info
->chunk_mutex
);
61 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
63 struct btrfs_device
*device
;
64 WARN_ON(fs_devices
->opened
);
65 while (!list_empty(&fs_devices
->devices
)) {
66 device
= list_entry(fs_devices
->devices
.next
,
67 struct btrfs_device
, dev_list
);
68 list_del(&device
->dev_list
);
69 rcu_string_free(device
->name
);
75 static void btrfs_kobject_uevent(struct block_device
*bdev
,
76 enum kobject_action action
)
80 ret
= kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, action
);
82 pr_warn("Sending event '%d' to kobject: '%s' (%p): failed\n",
84 kobject_name(&disk_to_dev(bdev
->bd_disk
)->kobj
),
85 &disk_to_dev(bdev
->bd_disk
)->kobj
);
88 void btrfs_cleanup_fs_uuids(void)
90 struct btrfs_fs_devices
*fs_devices
;
92 while (!list_empty(&fs_uuids
)) {
93 fs_devices
= list_entry(fs_uuids
.next
,
94 struct btrfs_fs_devices
, list
);
95 list_del(&fs_devices
->list
);
96 free_fs_devices(fs_devices
);
100 static noinline
struct btrfs_device
*__find_device(struct list_head
*head
,
103 struct btrfs_device
*dev
;
105 list_for_each_entry(dev
, head
, dev_list
) {
106 if (dev
->devid
== devid
&&
107 (!uuid
|| !memcmp(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
))) {
114 static noinline
struct btrfs_fs_devices
*find_fsid(u8
*fsid
)
116 struct btrfs_fs_devices
*fs_devices
;
118 list_for_each_entry(fs_devices
, &fs_uuids
, list
) {
119 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
126 btrfs_get_bdev_and_sb(const char *device_path
, fmode_t flags
, void *holder
,
127 int flush
, struct block_device
**bdev
,
128 struct buffer_head
**bh
)
132 *bdev
= blkdev_get_by_path(device_path
, flags
, holder
);
135 ret
= PTR_ERR(*bdev
);
136 printk(KERN_INFO
"btrfs: open %s failed\n", device_path
);
141 filemap_write_and_wait((*bdev
)->bd_inode
->i_mapping
);
142 ret
= set_blocksize(*bdev
, 4096);
144 blkdev_put(*bdev
, flags
);
147 invalidate_bdev(*bdev
);
148 *bh
= btrfs_read_dev_super(*bdev
);
151 blkdev_put(*bdev
, flags
);
163 static void requeue_list(struct btrfs_pending_bios
*pending_bios
,
164 struct bio
*head
, struct bio
*tail
)
167 struct bio
*old_head
;
169 old_head
= pending_bios
->head
;
170 pending_bios
->head
= head
;
171 if (pending_bios
->tail
)
172 tail
->bi_next
= old_head
;
174 pending_bios
->tail
= tail
;
178 * we try to collect pending bios for a device so we don't get a large
179 * number of procs sending bios down to the same device. This greatly
180 * improves the schedulers ability to collect and merge the bios.
182 * But, it also turns into a long list of bios to process and that is sure
183 * to eventually make the worker thread block. The solution here is to
184 * make some progress and then put this work struct back at the end of
185 * the list if the block device is congested. This way, multiple devices
186 * can make progress from a single worker thread.
188 static noinline
void run_scheduled_bios(struct btrfs_device
*device
)
191 struct backing_dev_info
*bdi
;
192 struct btrfs_fs_info
*fs_info
;
193 struct btrfs_pending_bios
*pending_bios
;
197 unsigned long num_run
;
198 unsigned long batch_run
= 0;
200 unsigned long last_waited
= 0;
202 int sync_pending
= 0;
203 struct blk_plug plug
;
206 * this function runs all the bios we've collected for
207 * a particular device. We don't want to wander off to
208 * another device without first sending all of these down.
209 * So, setup a plug here and finish it off before we return
211 blk_start_plug(&plug
);
213 bdi
= blk_get_backing_dev_info(device
->bdev
);
214 fs_info
= device
->dev_root
->fs_info
;
215 limit
= btrfs_async_submit_limit(fs_info
);
216 limit
= limit
* 2 / 3;
219 spin_lock(&device
->io_lock
);
224 /* take all the bios off the list at once and process them
225 * later on (without the lock held). But, remember the
226 * tail and other pointers so the bios can be properly reinserted
227 * into the list if we hit congestion
229 if (!force_reg
&& device
->pending_sync_bios
.head
) {
230 pending_bios
= &device
->pending_sync_bios
;
233 pending_bios
= &device
->pending_bios
;
237 pending
= pending_bios
->head
;
238 tail
= pending_bios
->tail
;
239 WARN_ON(pending
&& !tail
);
242 * if pending was null this time around, no bios need processing
243 * at all and we can stop. Otherwise it'll loop back up again
244 * and do an additional check so no bios are missed.
246 * device->running_pending is used to synchronize with the
249 if (device
->pending_sync_bios
.head
== NULL
&&
250 device
->pending_bios
.head
== NULL
) {
252 device
->running_pending
= 0;
255 device
->running_pending
= 1;
258 pending_bios
->head
= NULL
;
259 pending_bios
->tail
= NULL
;
261 spin_unlock(&device
->io_lock
);
266 /* we want to work on both lists, but do more bios on the
267 * sync list than the regular list
270 pending_bios
!= &device
->pending_sync_bios
&&
271 device
->pending_sync_bios
.head
) ||
272 (num_run
> 64 && pending_bios
== &device
->pending_sync_bios
&&
273 device
->pending_bios
.head
)) {
274 spin_lock(&device
->io_lock
);
275 requeue_list(pending_bios
, pending
, tail
);
280 pending
= pending
->bi_next
;
283 if (atomic_dec_return(&fs_info
->nr_async_bios
) < limit
&&
284 waitqueue_active(&fs_info
->async_submit_wait
))
285 wake_up(&fs_info
->async_submit_wait
);
287 BUG_ON(atomic_read(&cur
->bi_cnt
) == 0);
290 * if we're doing the sync list, record that our
291 * plug has some sync requests on it
293 * If we're doing the regular list and there are
294 * sync requests sitting around, unplug before
297 if (pending_bios
== &device
->pending_sync_bios
) {
299 } else if (sync_pending
) {
300 blk_finish_plug(&plug
);
301 blk_start_plug(&plug
);
305 btrfsic_submit_bio(cur
->bi_rw
, cur
);
312 * we made progress, there is more work to do and the bdi
313 * is now congested. Back off and let other work structs
316 if (pending
&& bdi_write_congested(bdi
) && batch_run
> 8 &&
317 fs_info
->fs_devices
->open_devices
> 1) {
318 struct io_context
*ioc
;
320 ioc
= current
->io_context
;
323 * the main goal here is that we don't want to
324 * block if we're going to be able to submit
325 * more requests without blocking.
327 * This code does two great things, it pokes into
328 * the elevator code from a filesystem _and_
329 * it makes assumptions about how batching works.
331 if (ioc
&& ioc
->nr_batch_requests
> 0 &&
332 time_before(jiffies
, ioc
->last_waited
+ HZ
/50UL) &&
334 ioc
->last_waited
== last_waited
)) {
336 * we want to go through our batch of
337 * requests and stop. So, we copy out
338 * the ioc->last_waited time and test
339 * against it before looping
341 last_waited
= ioc
->last_waited
;
346 spin_lock(&device
->io_lock
);
347 requeue_list(pending_bios
, pending
, tail
);
348 device
->running_pending
= 1;
350 spin_unlock(&device
->io_lock
);
351 btrfs_requeue_work(&device
->work
);
354 /* unplug every 64 requests just for good measure */
355 if (batch_run
% 64 == 0) {
356 blk_finish_plug(&plug
);
357 blk_start_plug(&plug
);
366 spin_lock(&device
->io_lock
);
367 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
369 spin_unlock(&device
->io_lock
);
372 blk_finish_plug(&plug
);
375 static void pending_bios_fn(struct btrfs_work
*work
)
377 struct btrfs_device
*device
;
379 device
= container_of(work
, struct btrfs_device
, work
);
380 run_scheduled_bios(device
);
383 static noinline
int device_list_add(const char *path
,
384 struct btrfs_super_block
*disk_super
,
385 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
387 struct btrfs_device
*device
;
388 struct btrfs_fs_devices
*fs_devices
;
389 struct rcu_string
*name
;
390 u64 found_transid
= btrfs_super_generation(disk_super
);
392 fs_devices
= find_fsid(disk_super
->fsid
);
394 fs_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
397 INIT_LIST_HEAD(&fs_devices
->devices
);
398 INIT_LIST_HEAD(&fs_devices
->alloc_list
);
399 list_add(&fs_devices
->list
, &fs_uuids
);
400 memcpy(fs_devices
->fsid
, disk_super
->fsid
, BTRFS_FSID_SIZE
);
401 fs_devices
->latest_devid
= devid
;
402 fs_devices
->latest_trans
= found_transid
;
403 mutex_init(&fs_devices
->device_list_mutex
);
406 device
= __find_device(&fs_devices
->devices
, devid
,
407 disk_super
->dev_item
.uuid
);
410 if (fs_devices
->opened
)
413 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
415 /* we can safely leave the fs_devices entry around */
418 device
->devid
= devid
;
419 device
->dev_stats_valid
= 0;
420 device
->work
.func
= pending_bios_fn
;
421 memcpy(device
->uuid
, disk_super
->dev_item
.uuid
,
423 spin_lock_init(&device
->io_lock
);
425 name
= rcu_string_strdup(path
, GFP_NOFS
);
430 rcu_assign_pointer(device
->name
, name
);
431 INIT_LIST_HEAD(&device
->dev_alloc_list
);
433 /* init readahead state */
434 spin_lock_init(&device
->reada_lock
);
435 device
->reada_curr_zone
= NULL
;
436 atomic_set(&device
->reada_in_flight
, 0);
437 device
->reada_next
= 0;
438 INIT_RADIX_TREE(&device
->reada_zones
, GFP_NOFS
& ~__GFP_WAIT
);
439 INIT_RADIX_TREE(&device
->reada_extents
, GFP_NOFS
& ~__GFP_WAIT
);
441 mutex_lock(&fs_devices
->device_list_mutex
);
442 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
443 mutex_unlock(&fs_devices
->device_list_mutex
);
445 device
->fs_devices
= fs_devices
;
446 fs_devices
->num_devices
++;
447 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
448 name
= rcu_string_strdup(path
, GFP_NOFS
);
451 rcu_string_free(device
->name
);
452 rcu_assign_pointer(device
->name
, name
);
453 if (device
->missing
) {
454 fs_devices
->missing_devices
--;
459 if (found_transid
> fs_devices
->latest_trans
) {
460 fs_devices
->latest_devid
= devid
;
461 fs_devices
->latest_trans
= found_transid
;
463 *fs_devices_ret
= fs_devices
;
467 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
469 struct btrfs_fs_devices
*fs_devices
;
470 struct btrfs_device
*device
;
471 struct btrfs_device
*orig_dev
;
473 fs_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
475 return ERR_PTR(-ENOMEM
);
477 INIT_LIST_HEAD(&fs_devices
->devices
);
478 INIT_LIST_HEAD(&fs_devices
->alloc_list
);
479 INIT_LIST_HEAD(&fs_devices
->list
);
480 mutex_init(&fs_devices
->device_list_mutex
);
481 fs_devices
->latest_devid
= orig
->latest_devid
;
482 fs_devices
->latest_trans
= orig
->latest_trans
;
483 fs_devices
->total_devices
= orig
->total_devices
;
484 memcpy(fs_devices
->fsid
, orig
->fsid
, sizeof(fs_devices
->fsid
));
486 /* We have held the volume lock, it is safe to get the devices. */
487 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
488 struct rcu_string
*name
;
490 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
495 * This is ok to do without rcu read locked because we hold the
496 * uuid mutex so nothing we touch in here is going to disappear.
498 name
= rcu_string_strdup(orig_dev
->name
->str
, GFP_NOFS
);
503 rcu_assign_pointer(device
->name
, name
);
505 device
->devid
= orig_dev
->devid
;
506 device
->work
.func
= pending_bios_fn
;
507 memcpy(device
->uuid
, orig_dev
->uuid
, sizeof(device
->uuid
));
508 spin_lock_init(&device
->io_lock
);
509 INIT_LIST_HEAD(&device
->dev_list
);
510 INIT_LIST_HEAD(&device
->dev_alloc_list
);
512 list_add(&device
->dev_list
, &fs_devices
->devices
);
513 device
->fs_devices
= fs_devices
;
514 fs_devices
->num_devices
++;
518 free_fs_devices(fs_devices
);
519 return ERR_PTR(-ENOMEM
);
522 void btrfs_close_extra_devices(struct btrfs_fs_info
*fs_info
,
523 struct btrfs_fs_devices
*fs_devices
, int step
)
525 struct btrfs_device
*device
, *next
;
527 struct block_device
*latest_bdev
= NULL
;
528 u64 latest_devid
= 0;
529 u64 latest_transid
= 0;
531 mutex_lock(&uuid_mutex
);
533 /* This is the initialized path, it is safe to release the devices. */
534 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
535 if (device
->in_fs_metadata
) {
536 if (!device
->is_tgtdev_for_dev_replace
&&
538 device
->generation
> latest_transid
)) {
539 latest_devid
= device
->devid
;
540 latest_transid
= device
->generation
;
541 latest_bdev
= device
->bdev
;
546 if (device
->devid
== BTRFS_DEV_REPLACE_DEVID
) {
548 * In the first step, keep the device which has
549 * the correct fsid and the devid that is used
550 * for the dev_replace procedure.
551 * In the second step, the dev_replace state is
552 * read from the device tree and it is known
553 * whether the procedure is really active or
554 * not, which means whether this device is
555 * used or whether it should be removed.
557 if (step
== 0 || device
->is_tgtdev_for_dev_replace
) {
562 blkdev_put(device
->bdev
, device
->mode
);
564 fs_devices
->open_devices
--;
566 if (device
->writeable
) {
567 list_del_init(&device
->dev_alloc_list
);
568 device
->writeable
= 0;
569 if (!device
->is_tgtdev_for_dev_replace
)
570 fs_devices
->rw_devices
--;
572 list_del_init(&device
->dev_list
);
573 fs_devices
->num_devices
--;
574 rcu_string_free(device
->name
);
578 if (fs_devices
->seed
) {
579 fs_devices
= fs_devices
->seed
;
583 fs_devices
->latest_bdev
= latest_bdev
;
584 fs_devices
->latest_devid
= latest_devid
;
585 fs_devices
->latest_trans
= latest_transid
;
587 mutex_unlock(&uuid_mutex
);
590 static void __free_device(struct work_struct
*work
)
592 struct btrfs_device
*device
;
594 device
= container_of(work
, struct btrfs_device
, rcu_work
);
597 blkdev_put(device
->bdev
, device
->mode
);
599 rcu_string_free(device
->name
);
603 static void free_device(struct rcu_head
*head
)
605 struct btrfs_device
*device
;
607 device
= container_of(head
, struct btrfs_device
, rcu
);
609 INIT_WORK(&device
->rcu_work
, __free_device
);
610 schedule_work(&device
->rcu_work
);
613 static int __btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
615 struct btrfs_device
*device
;
617 if (--fs_devices
->opened
> 0)
620 mutex_lock(&fs_devices
->device_list_mutex
);
621 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
622 struct btrfs_device
*new_device
;
623 struct rcu_string
*name
;
626 fs_devices
->open_devices
--;
628 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
629 list_del_init(&device
->dev_alloc_list
);
630 fs_devices
->rw_devices
--;
633 if (device
->can_discard
)
634 fs_devices
->num_can_discard
--;
636 new_device
= kmalloc(sizeof(*new_device
), GFP_NOFS
);
637 BUG_ON(!new_device
); /* -ENOMEM */
638 memcpy(new_device
, device
, sizeof(*new_device
));
640 /* Safe because we are under uuid_mutex */
642 name
= rcu_string_strdup(device
->name
->str
, GFP_NOFS
);
643 BUG_ON(device
->name
&& !name
); /* -ENOMEM */
644 rcu_assign_pointer(new_device
->name
, name
);
646 new_device
->bdev
= NULL
;
647 new_device
->writeable
= 0;
648 new_device
->in_fs_metadata
= 0;
649 new_device
->can_discard
= 0;
650 list_replace_rcu(&device
->dev_list
, &new_device
->dev_list
);
652 call_rcu(&device
->rcu
, free_device
);
654 mutex_unlock(&fs_devices
->device_list_mutex
);
656 WARN_ON(fs_devices
->open_devices
);
657 WARN_ON(fs_devices
->rw_devices
);
658 fs_devices
->opened
= 0;
659 fs_devices
->seeding
= 0;
664 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
666 struct btrfs_fs_devices
*seed_devices
= NULL
;
669 mutex_lock(&uuid_mutex
);
670 ret
= __btrfs_close_devices(fs_devices
);
671 if (!fs_devices
->opened
) {
672 seed_devices
= fs_devices
->seed
;
673 fs_devices
->seed
= NULL
;
675 mutex_unlock(&uuid_mutex
);
677 while (seed_devices
) {
678 fs_devices
= seed_devices
;
679 seed_devices
= fs_devices
->seed
;
680 __btrfs_close_devices(fs_devices
);
681 free_fs_devices(fs_devices
);
686 static int __btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
687 fmode_t flags
, void *holder
)
689 struct request_queue
*q
;
690 struct block_device
*bdev
;
691 struct list_head
*head
= &fs_devices
->devices
;
692 struct btrfs_device
*device
;
693 struct block_device
*latest_bdev
= NULL
;
694 struct buffer_head
*bh
;
695 struct btrfs_super_block
*disk_super
;
696 u64 latest_devid
= 0;
697 u64 latest_transid
= 0;
704 list_for_each_entry(device
, head
, dev_list
) {
710 ret
= btrfs_get_bdev_and_sb(device
->name
->str
, flags
, holder
, 1,
715 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
716 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
717 if (devid
!= device
->devid
)
720 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
,
724 device
->generation
= btrfs_super_generation(disk_super
);
725 if (!latest_transid
|| device
->generation
> latest_transid
) {
726 latest_devid
= devid
;
727 latest_transid
= device
->generation
;
731 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
732 device
->writeable
= 0;
734 device
->writeable
= !bdev_read_only(bdev
);
738 q
= bdev_get_queue(bdev
);
739 if (blk_queue_discard(q
)) {
740 device
->can_discard
= 1;
741 fs_devices
->num_can_discard
++;
745 device
->in_fs_metadata
= 0;
746 device
->mode
= flags
;
748 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
749 fs_devices
->rotating
= 1;
751 fs_devices
->open_devices
++;
752 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
753 fs_devices
->rw_devices
++;
754 list_add(&device
->dev_alloc_list
,
755 &fs_devices
->alloc_list
);
762 blkdev_put(bdev
, flags
);
765 if (fs_devices
->open_devices
== 0) {
769 fs_devices
->seeding
= seeding
;
770 fs_devices
->opened
= 1;
771 fs_devices
->latest_bdev
= latest_bdev
;
772 fs_devices
->latest_devid
= latest_devid
;
773 fs_devices
->latest_trans
= latest_transid
;
774 fs_devices
->total_rw_bytes
= 0;
779 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
780 fmode_t flags
, void *holder
)
784 mutex_lock(&uuid_mutex
);
785 if (fs_devices
->opened
) {
786 fs_devices
->opened
++;
789 ret
= __btrfs_open_devices(fs_devices
, flags
, holder
);
791 mutex_unlock(&uuid_mutex
);
796 * Look for a btrfs signature on a device. This may be called out of the mount path
797 * and we are not allowed to call set_blocksize during the scan. The superblock
798 * is read via pagecache
800 int btrfs_scan_one_device(const char *path
, fmode_t flags
, void *holder
,
801 struct btrfs_fs_devices
**fs_devices_ret
)
803 struct btrfs_super_block
*disk_super
;
804 struct block_device
*bdev
;
815 * we would like to check all the supers, but that would make
816 * a btrfs mount succeed after a mkfs from a different FS.
817 * So, we need to add a special mount option to scan for
818 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
820 bytenr
= btrfs_sb_offset(0);
822 mutex_lock(&uuid_mutex
);
824 bdev
= blkdev_get_by_path(path
, flags
, holder
);
828 printk(KERN_INFO
"btrfs: open %s failed\n", path
);
832 /* make sure our super fits in the device */
833 if (bytenr
+ PAGE_CACHE_SIZE
>= i_size_read(bdev
->bd_inode
))
836 /* make sure our super fits in the page */
837 if (sizeof(*disk_super
) > PAGE_CACHE_SIZE
)
840 /* make sure our super doesn't straddle pages on disk */
841 index
= bytenr
>> PAGE_CACHE_SHIFT
;
842 if ((bytenr
+ sizeof(*disk_super
) - 1) >> PAGE_CACHE_SHIFT
!= index
)
845 /* pull in the page with our super */
846 page
= read_cache_page_gfp(bdev
->bd_inode
->i_mapping
,
849 if (IS_ERR_OR_NULL(page
))
854 /* align our pointer to the offset of the super block */
855 disk_super
= p
+ (bytenr
& ~PAGE_CACHE_MASK
);
857 if (btrfs_super_bytenr(disk_super
) != bytenr
||
858 disk_super
->magic
!= cpu_to_le64(BTRFS_MAGIC
))
861 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
862 transid
= btrfs_super_generation(disk_super
);
863 total_devices
= btrfs_super_num_devices(disk_super
);
865 if (disk_super
->label
[0]) {
866 if (disk_super
->label
[BTRFS_LABEL_SIZE
- 1])
867 disk_super
->label
[BTRFS_LABEL_SIZE
- 1] = '\0';
868 printk(KERN_INFO
"device label %s ", disk_super
->label
);
870 printk(KERN_INFO
"device fsid %pU ", disk_super
->fsid
);
873 printk(KERN_CONT
"devid %llu transid %llu %s\n",
874 (unsigned long long)devid
, (unsigned long long)transid
, path
);
876 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
877 if (!ret
&& fs_devices_ret
)
878 (*fs_devices_ret
)->total_devices
= total_devices
;
882 page_cache_release(page
);
885 blkdev_put(bdev
, flags
);
887 mutex_unlock(&uuid_mutex
);
891 /* helper to account the used device space in the range */
892 int btrfs_account_dev_extents_size(struct btrfs_device
*device
, u64 start
,
893 u64 end
, u64
*length
)
895 struct btrfs_key key
;
896 struct btrfs_root
*root
= device
->dev_root
;
897 struct btrfs_dev_extent
*dev_extent
;
898 struct btrfs_path
*path
;
902 struct extent_buffer
*l
;
906 if (start
>= device
->total_bytes
|| device
->is_tgtdev_for_dev_replace
)
909 path
= btrfs_alloc_path();
914 key
.objectid
= device
->devid
;
916 key
.type
= BTRFS_DEV_EXTENT_KEY
;
918 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
922 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
929 slot
= path
->slots
[0];
930 if (slot
>= btrfs_header_nritems(l
)) {
931 ret
= btrfs_next_leaf(root
, path
);
939 btrfs_item_key_to_cpu(l
, &key
, slot
);
941 if (key
.objectid
< device
->devid
)
944 if (key
.objectid
> device
->devid
)
947 if (btrfs_key_type(&key
) != BTRFS_DEV_EXTENT_KEY
)
950 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
951 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
953 if (key
.offset
<= start
&& extent_end
> end
) {
954 *length
= end
- start
+ 1;
956 } else if (key
.offset
<= start
&& extent_end
> start
)
957 *length
+= extent_end
- start
;
958 else if (key
.offset
> start
&& extent_end
<= end
)
959 *length
+= extent_end
- key
.offset
;
960 else if (key
.offset
> start
&& key
.offset
<= end
) {
961 *length
+= end
- key
.offset
+ 1;
963 } else if (key
.offset
> end
)
971 btrfs_free_path(path
);
976 * find_free_dev_extent - find free space in the specified device
977 * @device: the device which we search the free space in
978 * @num_bytes: the size of the free space that we need
979 * @start: store the start of the free space.
980 * @len: the size of the free space. that we find, or the size of the max
981 * free space if we don't find suitable free space
983 * this uses a pretty simple search, the expectation is that it is
984 * called very infrequently and that a given device has a small number
987 * @start is used to store the start of the free space if we find. But if we
988 * don't find suitable free space, it will be used to store the start position
989 * of the max free space.
991 * @len is used to store the size of the free space that we find.
992 * But if we don't find suitable free space, it is used to store the size of
993 * the max free space.
995 int find_free_dev_extent(struct btrfs_device
*device
, u64 num_bytes
,
996 u64
*start
, u64
*len
)
998 struct btrfs_key key
;
999 struct btrfs_root
*root
= device
->dev_root
;
1000 struct btrfs_dev_extent
*dev_extent
;
1001 struct btrfs_path
*path
;
1007 u64 search_end
= device
->total_bytes
;
1010 struct extent_buffer
*l
;
1012 /* FIXME use last free of some kind */
1014 /* we don't want to overwrite the superblock on the drive,
1015 * so we make sure to start at an offset of at least 1MB
1017 search_start
= max(root
->fs_info
->alloc_start
, 1024ull * 1024);
1019 max_hole_start
= search_start
;
1023 if (search_start
>= search_end
|| device
->is_tgtdev_for_dev_replace
) {
1028 path
= btrfs_alloc_path();
1035 key
.objectid
= device
->devid
;
1036 key
.offset
= search_start
;
1037 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1039 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1043 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1050 slot
= path
->slots
[0];
1051 if (slot
>= btrfs_header_nritems(l
)) {
1052 ret
= btrfs_next_leaf(root
, path
);
1060 btrfs_item_key_to_cpu(l
, &key
, slot
);
1062 if (key
.objectid
< device
->devid
)
1065 if (key
.objectid
> device
->devid
)
1068 if (btrfs_key_type(&key
) != BTRFS_DEV_EXTENT_KEY
)
1071 if (key
.offset
> search_start
) {
1072 hole_size
= key
.offset
- search_start
;
1074 if (hole_size
> max_hole_size
) {
1075 max_hole_start
= search_start
;
1076 max_hole_size
= hole_size
;
1080 * If this free space is greater than which we need,
1081 * it must be the max free space that we have found
1082 * until now, so max_hole_start must point to the start
1083 * of this free space and the length of this free space
1084 * is stored in max_hole_size. Thus, we return
1085 * max_hole_start and max_hole_size and go back to the
1088 if (hole_size
>= num_bytes
) {
1094 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1095 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1097 if (extent_end
> search_start
)
1098 search_start
= extent_end
;
1105 * At this point, search_start should be the end of
1106 * allocated dev extents, and when shrinking the device,
1107 * search_end may be smaller than search_start.
1109 if (search_end
> search_start
)
1110 hole_size
= search_end
- search_start
;
1112 if (hole_size
> max_hole_size
) {
1113 max_hole_start
= search_start
;
1114 max_hole_size
= hole_size
;
1118 if (hole_size
< num_bytes
)
1124 btrfs_free_path(path
);
1126 *start
= max_hole_start
;
1128 *len
= max_hole_size
;
1132 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1133 struct btrfs_device
*device
,
1137 struct btrfs_path
*path
;
1138 struct btrfs_root
*root
= device
->dev_root
;
1139 struct btrfs_key key
;
1140 struct btrfs_key found_key
;
1141 struct extent_buffer
*leaf
= NULL
;
1142 struct btrfs_dev_extent
*extent
= NULL
;
1144 path
= btrfs_alloc_path();
1148 key
.objectid
= device
->devid
;
1150 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1152 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1154 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1155 BTRFS_DEV_EXTENT_KEY
);
1158 leaf
= path
->nodes
[0];
1159 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1160 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1161 struct btrfs_dev_extent
);
1162 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1163 btrfs_dev_extent_length(leaf
, extent
) < start
);
1165 btrfs_release_path(path
);
1167 } else if (ret
== 0) {
1168 leaf
= path
->nodes
[0];
1169 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1170 struct btrfs_dev_extent
);
1172 btrfs_error(root
->fs_info
, ret
, "Slot search failed");
1176 if (device
->bytes_used
> 0) {
1177 u64 len
= btrfs_dev_extent_length(leaf
, extent
);
1178 device
->bytes_used
-= len
;
1179 spin_lock(&root
->fs_info
->free_chunk_lock
);
1180 root
->fs_info
->free_chunk_space
+= len
;
1181 spin_unlock(&root
->fs_info
->free_chunk_lock
);
1183 ret
= btrfs_del_item(trans
, root
, path
);
1185 btrfs_error(root
->fs_info
, ret
,
1186 "Failed to remove dev extent item");
1189 btrfs_free_path(path
);
1193 int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1194 struct btrfs_device
*device
,
1195 u64 chunk_tree
, u64 chunk_objectid
,
1196 u64 chunk_offset
, u64 start
, u64 num_bytes
)
1199 struct btrfs_path
*path
;
1200 struct btrfs_root
*root
= device
->dev_root
;
1201 struct btrfs_dev_extent
*extent
;
1202 struct extent_buffer
*leaf
;
1203 struct btrfs_key key
;
1205 WARN_ON(!device
->in_fs_metadata
);
1206 WARN_ON(device
->is_tgtdev_for_dev_replace
);
1207 path
= btrfs_alloc_path();
1211 key
.objectid
= device
->devid
;
1213 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1214 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1219 leaf
= path
->nodes
[0];
1220 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1221 struct btrfs_dev_extent
);
1222 btrfs_set_dev_extent_chunk_tree(leaf
, extent
, chunk_tree
);
1223 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
, chunk_objectid
);
1224 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1226 write_extent_buffer(leaf
, root
->fs_info
->chunk_tree_uuid
,
1227 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent
),
1230 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1231 btrfs_mark_buffer_dirty(leaf
);
1233 btrfs_free_path(path
);
1237 static noinline
int find_next_chunk(struct btrfs_root
*root
,
1238 u64 objectid
, u64
*offset
)
1240 struct btrfs_path
*path
;
1242 struct btrfs_key key
;
1243 struct btrfs_chunk
*chunk
;
1244 struct btrfs_key found_key
;
1246 path
= btrfs_alloc_path();
1250 key
.objectid
= objectid
;
1251 key
.offset
= (u64
)-1;
1252 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
1254 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1258 BUG_ON(ret
== 0); /* Corruption */
1260 ret
= btrfs_previous_item(root
, path
, 0, BTRFS_CHUNK_ITEM_KEY
);
1264 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1266 if (found_key
.objectid
!= objectid
)
1269 chunk
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
1270 struct btrfs_chunk
);
1271 *offset
= found_key
.offset
+
1272 btrfs_chunk_length(path
->nodes
[0], chunk
);
1277 btrfs_free_path(path
);
1281 static noinline
int find_next_devid(struct btrfs_root
*root
, u64
*objectid
)
1284 struct btrfs_key key
;
1285 struct btrfs_key found_key
;
1286 struct btrfs_path
*path
;
1288 root
= root
->fs_info
->chunk_root
;
1290 path
= btrfs_alloc_path();
1294 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1295 key
.type
= BTRFS_DEV_ITEM_KEY
;
1296 key
.offset
= (u64
)-1;
1298 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1302 BUG_ON(ret
== 0); /* Corruption */
1304 ret
= btrfs_previous_item(root
, path
, BTRFS_DEV_ITEMS_OBJECTID
,
1305 BTRFS_DEV_ITEM_KEY
);
1309 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1311 *objectid
= found_key
.offset
+ 1;
1315 btrfs_free_path(path
);
1320 * the device information is stored in the chunk root
1321 * the btrfs_device struct should be fully filled in
1323 int btrfs_add_device(struct btrfs_trans_handle
*trans
,
1324 struct btrfs_root
*root
,
1325 struct btrfs_device
*device
)
1328 struct btrfs_path
*path
;
1329 struct btrfs_dev_item
*dev_item
;
1330 struct extent_buffer
*leaf
;
1331 struct btrfs_key key
;
1334 root
= root
->fs_info
->chunk_root
;
1336 path
= btrfs_alloc_path();
1340 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1341 key
.type
= BTRFS_DEV_ITEM_KEY
;
1342 key
.offset
= device
->devid
;
1344 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1349 leaf
= path
->nodes
[0];
1350 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1352 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1353 btrfs_set_device_generation(leaf
, dev_item
, 0);
1354 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1355 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1356 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1357 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1358 btrfs_set_device_total_bytes(leaf
, dev_item
, device
->total_bytes
);
1359 btrfs_set_device_bytes_used(leaf
, dev_item
, device
->bytes_used
);
1360 btrfs_set_device_group(leaf
, dev_item
, 0);
1361 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1362 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1363 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1365 ptr
= (unsigned long)btrfs_device_uuid(dev_item
);
1366 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1367 ptr
= (unsigned long)btrfs_device_fsid(dev_item
);
1368 write_extent_buffer(leaf
, root
->fs_info
->fsid
, ptr
, BTRFS_UUID_SIZE
);
1369 btrfs_mark_buffer_dirty(leaf
);
1373 btrfs_free_path(path
);
1377 static int btrfs_rm_dev_item(struct btrfs_root
*root
,
1378 struct btrfs_device
*device
)
1381 struct btrfs_path
*path
;
1382 struct btrfs_key key
;
1383 struct btrfs_trans_handle
*trans
;
1385 root
= root
->fs_info
->chunk_root
;
1387 path
= btrfs_alloc_path();
1391 trans
= btrfs_start_transaction(root
, 0);
1392 if (IS_ERR(trans
)) {
1393 btrfs_free_path(path
);
1394 return PTR_ERR(trans
);
1396 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1397 key
.type
= BTRFS_DEV_ITEM_KEY
;
1398 key
.offset
= device
->devid
;
1401 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1410 ret
= btrfs_del_item(trans
, root
, path
);
1414 btrfs_free_path(path
);
1415 unlock_chunks(root
);
1416 btrfs_commit_transaction(trans
, root
);
1420 int btrfs_rm_device(struct btrfs_root
*root
, char *device_path
)
1422 struct btrfs_device
*device
;
1423 struct btrfs_device
*next_device
;
1424 struct block_device
*bdev
;
1425 struct buffer_head
*bh
= NULL
;
1426 struct btrfs_super_block
*disk_super
;
1427 struct btrfs_fs_devices
*cur_devices
;
1434 bool clear_super
= false;
1436 mutex_lock(&uuid_mutex
);
1439 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
1441 all_avail
= root
->fs_info
->avail_data_alloc_bits
|
1442 root
->fs_info
->avail_system_alloc_bits
|
1443 root
->fs_info
->avail_metadata_alloc_bits
;
1444 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
1446 num_devices
= root
->fs_info
->fs_devices
->num_devices
;
1447 btrfs_dev_replace_lock(&root
->fs_info
->dev_replace
);
1448 if (btrfs_dev_replace_is_ongoing(&root
->fs_info
->dev_replace
)) {
1449 WARN_ON(num_devices
< 1);
1452 btrfs_dev_replace_unlock(&root
->fs_info
->dev_replace
);
1454 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID10
) && num_devices
<= 4) {
1455 printk(KERN_ERR
"btrfs: unable to go below four devices "
1461 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID1
) && num_devices
<= 2) {
1462 printk(KERN_ERR
"btrfs: unable to go below two "
1463 "devices on raid1\n");
1468 if (strcmp(device_path
, "missing") == 0) {
1469 struct list_head
*devices
;
1470 struct btrfs_device
*tmp
;
1473 devices
= &root
->fs_info
->fs_devices
->devices
;
1475 * It is safe to read the devices since the volume_mutex
1478 list_for_each_entry(tmp
, devices
, dev_list
) {
1479 if (tmp
->in_fs_metadata
&&
1480 !tmp
->is_tgtdev_for_dev_replace
&&
1490 printk(KERN_ERR
"btrfs: no missing devices found to "
1495 ret
= btrfs_get_bdev_and_sb(device_path
,
1496 FMODE_WRITE
| FMODE_EXCL
,
1497 root
->fs_info
->bdev_holder
, 0,
1501 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1502 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1503 dev_uuid
= disk_super
->dev_item
.uuid
;
1504 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1512 if (device
->is_tgtdev_for_dev_replace
) {
1513 pr_err("btrfs: unable to remove the dev_replace target dev\n");
1518 if (device
->writeable
&& root
->fs_info
->fs_devices
->rw_devices
== 1) {
1519 printk(KERN_ERR
"btrfs: unable to remove the only writeable "
1525 if (device
->writeable
) {
1527 list_del_init(&device
->dev_alloc_list
);
1528 unlock_chunks(root
);
1529 root
->fs_info
->fs_devices
->rw_devices
--;
1533 ret
= btrfs_shrink_device(device
, 0);
1538 * TODO: the superblock still includes this device in its num_devices
1539 * counter although write_all_supers() is not locked out. This
1540 * could give a filesystem state which requires a degraded mount.
1542 ret
= btrfs_rm_dev_item(root
->fs_info
->chunk_root
, device
);
1546 spin_lock(&root
->fs_info
->free_chunk_lock
);
1547 root
->fs_info
->free_chunk_space
= device
->total_bytes
-
1549 spin_unlock(&root
->fs_info
->free_chunk_lock
);
1551 device
->in_fs_metadata
= 0;
1552 btrfs_scrub_cancel_dev(root
->fs_info
, device
);
1555 * the device list mutex makes sure that we don't change
1556 * the device list while someone else is writing out all
1557 * the device supers.
1560 cur_devices
= device
->fs_devices
;
1561 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1562 list_del_rcu(&device
->dev_list
);
1564 device
->fs_devices
->num_devices
--;
1565 device
->fs_devices
->total_devices
--;
1567 if (device
->missing
)
1568 root
->fs_info
->fs_devices
->missing_devices
--;
1570 next_device
= list_entry(root
->fs_info
->fs_devices
->devices
.next
,
1571 struct btrfs_device
, dev_list
);
1572 if (device
->bdev
== root
->fs_info
->sb
->s_bdev
)
1573 root
->fs_info
->sb
->s_bdev
= next_device
->bdev
;
1574 if (device
->bdev
== root
->fs_info
->fs_devices
->latest_bdev
)
1575 root
->fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1578 device
->fs_devices
->open_devices
--;
1580 call_rcu(&device
->rcu
, free_device
);
1581 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1583 num_devices
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
1584 btrfs_set_super_num_devices(root
->fs_info
->super_copy
, num_devices
);
1586 if (cur_devices
->open_devices
== 0) {
1587 struct btrfs_fs_devices
*fs_devices
;
1588 fs_devices
= root
->fs_info
->fs_devices
;
1589 while (fs_devices
) {
1590 if (fs_devices
->seed
== cur_devices
)
1592 fs_devices
= fs_devices
->seed
;
1594 fs_devices
->seed
= cur_devices
->seed
;
1595 cur_devices
->seed
= NULL
;
1597 __btrfs_close_devices(cur_devices
);
1598 unlock_chunks(root
);
1599 free_fs_devices(cur_devices
);
1602 root
->fs_info
->num_tolerated_disk_barrier_failures
=
1603 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
1606 * at this point, the device is zero sized. We want to
1607 * remove it from the devices list and zero out the old super
1609 if (clear_super
&& disk_super
) {
1610 /* make sure this device isn't detected as part of
1613 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
1614 set_buffer_dirty(bh
);
1615 sync_dirty_buffer(bh
);
1620 /* Notify udev that device has changed */
1622 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
1627 blkdev_put(bdev
, FMODE_READ
| FMODE_EXCL
);
1629 mutex_unlock(&uuid_mutex
);
1632 if (device
->writeable
) {
1634 list_add(&device
->dev_alloc_list
,
1635 &root
->fs_info
->fs_devices
->alloc_list
);
1636 unlock_chunks(root
);
1637 root
->fs_info
->fs_devices
->rw_devices
++;
1642 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info
*fs_info
,
1643 struct btrfs_device
*srcdev
)
1645 WARN_ON(!mutex_is_locked(&fs_info
->fs_devices
->device_list_mutex
));
1646 list_del_rcu(&srcdev
->dev_list
);
1647 list_del_rcu(&srcdev
->dev_alloc_list
);
1648 fs_info
->fs_devices
->num_devices
--;
1649 if (srcdev
->missing
) {
1650 fs_info
->fs_devices
->missing_devices
--;
1651 fs_info
->fs_devices
->rw_devices
++;
1653 if (srcdev
->can_discard
)
1654 fs_info
->fs_devices
->num_can_discard
--;
1656 fs_info
->fs_devices
->open_devices
--;
1658 call_rcu(&srcdev
->rcu
, free_device
);
1661 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
1662 struct btrfs_device
*tgtdev
)
1664 struct btrfs_device
*next_device
;
1667 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
1669 btrfs_scratch_superblock(tgtdev
);
1670 fs_info
->fs_devices
->open_devices
--;
1672 fs_info
->fs_devices
->num_devices
--;
1673 if (tgtdev
->can_discard
)
1674 fs_info
->fs_devices
->num_can_discard
++;
1676 next_device
= list_entry(fs_info
->fs_devices
->devices
.next
,
1677 struct btrfs_device
, dev_list
);
1678 if (tgtdev
->bdev
== fs_info
->sb
->s_bdev
)
1679 fs_info
->sb
->s_bdev
= next_device
->bdev
;
1680 if (tgtdev
->bdev
== fs_info
->fs_devices
->latest_bdev
)
1681 fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1682 list_del_rcu(&tgtdev
->dev_list
);
1684 call_rcu(&tgtdev
->rcu
, free_device
);
1686 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
1689 int btrfs_find_device_by_path(struct btrfs_root
*root
, char *device_path
,
1690 struct btrfs_device
**device
)
1693 struct btrfs_super_block
*disk_super
;
1696 struct block_device
*bdev
;
1697 struct buffer_head
*bh
;
1700 ret
= btrfs_get_bdev_and_sb(device_path
, FMODE_READ
,
1701 root
->fs_info
->bdev_holder
, 0, &bdev
, &bh
);
1704 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1705 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1706 dev_uuid
= disk_super
->dev_item
.uuid
;
1707 *device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1712 blkdev_put(bdev
, FMODE_READ
);
1716 int btrfs_find_device_missing_or_by_path(struct btrfs_root
*root
,
1718 struct btrfs_device
**device
)
1721 if (strcmp(device_path
, "missing") == 0) {
1722 struct list_head
*devices
;
1723 struct btrfs_device
*tmp
;
1725 devices
= &root
->fs_info
->fs_devices
->devices
;
1727 * It is safe to read the devices since the volume_mutex
1728 * is held by the caller.
1730 list_for_each_entry(tmp
, devices
, dev_list
) {
1731 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
1738 pr_err("btrfs: no missing device found\n");
1744 return btrfs_find_device_by_path(root
, device_path
, device
);
1749 * does all the dirty work required for changing file system's UUID.
1751 static int btrfs_prepare_sprout(struct btrfs_root
*root
)
1753 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
1754 struct btrfs_fs_devices
*old_devices
;
1755 struct btrfs_fs_devices
*seed_devices
;
1756 struct btrfs_super_block
*disk_super
= root
->fs_info
->super_copy
;
1757 struct btrfs_device
*device
;
1760 BUG_ON(!mutex_is_locked(&uuid_mutex
));
1761 if (!fs_devices
->seeding
)
1764 seed_devices
= kzalloc(sizeof(*fs_devices
), GFP_NOFS
);
1768 old_devices
= clone_fs_devices(fs_devices
);
1769 if (IS_ERR(old_devices
)) {
1770 kfree(seed_devices
);
1771 return PTR_ERR(old_devices
);
1774 list_add(&old_devices
->list
, &fs_uuids
);
1776 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
1777 seed_devices
->opened
= 1;
1778 INIT_LIST_HEAD(&seed_devices
->devices
);
1779 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
1780 mutex_init(&seed_devices
->device_list_mutex
);
1782 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1783 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
1785 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1787 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
1788 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
) {
1789 device
->fs_devices
= seed_devices
;
1792 fs_devices
->seeding
= 0;
1793 fs_devices
->num_devices
= 0;
1794 fs_devices
->open_devices
= 0;
1795 fs_devices
->total_devices
= 0;
1796 fs_devices
->seed
= seed_devices
;
1798 generate_random_uuid(fs_devices
->fsid
);
1799 memcpy(root
->fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
1800 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
1801 super_flags
= btrfs_super_flags(disk_super
) &
1802 ~BTRFS_SUPER_FLAG_SEEDING
;
1803 btrfs_set_super_flags(disk_super
, super_flags
);
1809 * strore the expected generation for seed devices in device items.
1811 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
1812 struct btrfs_root
*root
)
1814 struct btrfs_path
*path
;
1815 struct extent_buffer
*leaf
;
1816 struct btrfs_dev_item
*dev_item
;
1817 struct btrfs_device
*device
;
1818 struct btrfs_key key
;
1819 u8 fs_uuid
[BTRFS_UUID_SIZE
];
1820 u8 dev_uuid
[BTRFS_UUID_SIZE
];
1824 path
= btrfs_alloc_path();
1828 root
= root
->fs_info
->chunk_root
;
1829 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1831 key
.type
= BTRFS_DEV_ITEM_KEY
;
1834 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
1838 leaf
= path
->nodes
[0];
1840 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1841 ret
= btrfs_next_leaf(root
, path
);
1846 leaf
= path
->nodes
[0];
1847 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1848 btrfs_release_path(path
);
1852 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1853 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
1854 key
.type
!= BTRFS_DEV_ITEM_KEY
)
1857 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1858 struct btrfs_dev_item
);
1859 devid
= btrfs_device_id(leaf
, dev_item
);
1860 read_extent_buffer(leaf
, dev_uuid
,
1861 (unsigned long)btrfs_device_uuid(dev_item
),
1863 read_extent_buffer(leaf
, fs_uuid
,
1864 (unsigned long)btrfs_device_fsid(dev_item
),
1866 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1868 BUG_ON(!device
); /* Logic error */
1870 if (device
->fs_devices
->seeding
) {
1871 btrfs_set_device_generation(leaf
, dev_item
,
1872 device
->generation
);
1873 btrfs_mark_buffer_dirty(leaf
);
1881 btrfs_free_path(path
);
1885 int btrfs_init_new_device(struct btrfs_root
*root
, char *device_path
)
1887 struct request_queue
*q
;
1888 struct btrfs_trans_handle
*trans
;
1889 struct btrfs_device
*device
;
1890 struct block_device
*bdev
;
1891 struct list_head
*devices
;
1892 struct super_block
*sb
= root
->fs_info
->sb
;
1893 struct rcu_string
*name
;
1895 int seeding_dev
= 0;
1898 if ((sb
->s_flags
& MS_RDONLY
) && !root
->fs_info
->fs_devices
->seeding
)
1901 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
1902 root
->fs_info
->bdev_holder
);
1904 return PTR_ERR(bdev
);
1906 if (root
->fs_info
->fs_devices
->seeding
) {
1908 down_write(&sb
->s_umount
);
1909 mutex_lock(&uuid_mutex
);
1912 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
1914 devices
= &root
->fs_info
->fs_devices
->devices
;
1916 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1917 list_for_each_entry(device
, devices
, dev_list
) {
1918 if (device
->bdev
== bdev
) {
1921 &root
->fs_info
->fs_devices
->device_list_mutex
);
1925 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1927 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
1929 /* we can safely leave the fs_devices entry around */
1934 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
1940 rcu_assign_pointer(device
->name
, name
);
1942 ret
= find_next_devid(root
, &device
->devid
);
1944 rcu_string_free(device
->name
);
1949 trans
= btrfs_start_transaction(root
, 0);
1950 if (IS_ERR(trans
)) {
1951 rcu_string_free(device
->name
);
1953 ret
= PTR_ERR(trans
);
1959 q
= bdev_get_queue(bdev
);
1960 if (blk_queue_discard(q
))
1961 device
->can_discard
= 1;
1962 device
->writeable
= 1;
1963 device
->work
.func
= pending_bios_fn
;
1964 generate_random_uuid(device
->uuid
);
1965 spin_lock_init(&device
->io_lock
);
1966 device
->generation
= trans
->transid
;
1967 device
->io_width
= root
->sectorsize
;
1968 device
->io_align
= root
->sectorsize
;
1969 device
->sector_size
= root
->sectorsize
;
1970 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
1971 device
->disk_total_bytes
= device
->total_bytes
;
1972 device
->dev_root
= root
->fs_info
->dev_root
;
1973 device
->bdev
= bdev
;
1974 device
->in_fs_metadata
= 1;
1975 device
->is_tgtdev_for_dev_replace
= 0;
1976 device
->mode
= FMODE_EXCL
;
1977 set_blocksize(device
->bdev
, 4096);
1980 sb
->s_flags
&= ~MS_RDONLY
;
1981 ret
= btrfs_prepare_sprout(root
);
1982 BUG_ON(ret
); /* -ENOMEM */
1985 device
->fs_devices
= root
->fs_info
->fs_devices
;
1987 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1988 list_add_rcu(&device
->dev_list
, &root
->fs_info
->fs_devices
->devices
);
1989 list_add(&device
->dev_alloc_list
,
1990 &root
->fs_info
->fs_devices
->alloc_list
);
1991 root
->fs_info
->fs_devices
->num_devices
++;
1992 root
->fs_info
->fs_devices
->open_devices
++;
1993 root
->fs_info
->fs_devices
->rw_devices
++;
1994 root
->fs_info
->fs_devices
->total_devices
++;
1995 if (device
->can_discard
)
1996 root
->fs_info
->fs_devices
->num_can_discard
++;
1997 root
->fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
1999 spin_lock(&root
->fs_info
->free_chunk_lock
);
2000 root
->fs_info
->free_chunk_space
+= device
->total_bytes
;
2001 spin_unlock(&root
->fs_info
->free_chunk_lock
);
2003 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
2004 root
->fs_info
->fs_devices
->rotating
= 1;
2006 total_bytes
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
2007 btrfs_set_super_total_bytes(root
->fs_info
->super_copy
,
2008 total_bytes
+ device
->total_bytes
);
2010 total_bytes
= btrfs_super_num_devices(root
->fs_info
->super_copy
);
2011 btrfs_set_super_num_devices(root
->fs_info
->super_copy
,
2013 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2016 ret
= init_first_rw_device(trans
, root
, device
);
2018 btrfs_abort_transaction(trans
, root
, ret
);
2021 ret
= btrfs_finish_sprout(trans
, root
);
2023 btrfs_abort_transaction(trans
, root
, ret
);
2027 ret
= btrfs_add_device(trans
, root
, device
);
2029 btrfs_abort_transaction(trans
, root
, ret
);
2035 * we've got more storage, clear any full flags on the space
2038 btrfs_clear_space_info_full(root
->fs_info
);
2040 unlock_chunks(root
);
2041 root
->fs_info
->num_tolerated_disk_barrier_failures
=
2042 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
2043 ret
= btrfs_commit_transaction(trans
, root
);
2046 mutex_unlock(&uuid_mutex
);
2047 up_write(&sb
->s_umount
);
2049 if (ret
) /* transaction commit */
2052 ret
= btrfs_relocate_sys_chunks(root
);
2054 btrfs_error(root
->fs_info
, ret
,
2055 "Failed to relocate sys chunks after "
2056 "device initialization. This can be fixed "
2057 "using the \"btrfs balance\" command.");
2058 trans
= btrfs_attach_transaction(root
);
2059 if (IS_ERR(trans
)) {
2060 if (PTR_ERR(trans
) == -ENOENT
)
2062 return PTR_ERR(trans
);
2064 ret
= btrfs_commit_transaction(trans
, root
);
2070 unlock_chunks(root
);
2071 btrfs_end_transaction(trans
, root
);
2072 rcu_string_free(device
->name
);
2075 blkdev_put(bdev
, FMODE_EXCL
);
2077 mutex_unlock(&uuid_mutex
);
2078 up_write(&sb
->s_umount
);
2083 int btrfs_init_dev_replace_tgtdev(struct btrfs_root
*root
, char *device_path
,
2084 struct btrfs_device
**device_out
)
2086 struct request_queue
*q
;
2087 struct btrfs_device
*device
;
2088 struct block_device
*bdev
;
2089 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2090 struct list_head
*devices
;
2091 struct rcu_string
*name
;
2095 if (fs_info
->fs_devices
->seeding
)
2098 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2099 fs_info
->bdev_holder
);
2101 return PTR_ERR(bdev
);
2103 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2105 devices
= &fs_info
->fs_devices
->devices
;
2106 list_for_each_entry(device
, devices
, dev_list
) {
2107 if (device
->bdev
== bdev
) {
2113 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
2119 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2125 rcu_assign_pointer(device
->name
, name
);
2127 q
= bdev_get_queue(bdev
);
2128 if (blk_queue_discard(q
))
2129 device
->can_discard
= 1;
2130 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2131 device
->writeable
= 1;
2132 device
->work
.func
= pending_bios_fn
;
2133 generate_random_uuid(device
->uuid
);
2134 device
->devid
= BTRFS_DEV_REPLACE_DEVID
;
2135 spin_lock_init(&device
->io_lock
);
2136 device
->generation
= 0;
2137 device
->io_width
= root
->sectorsize
;
2138 device
->io_align
= root
->sectorsize
;
2139 device
->sector_size
= root
->sectorsize
;
2140 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
2141 device
->disk_total_bytes
= device
->total_bytes
;
2142 device
->dev_root
= fs_info
->dev_root
;
2143 device
->bdev
= bdev
;
2144 device
->in_fs_metadata
= 1;
2145 device
->is_tgtdev_for_dev_replace
= 1;
2146 device
->mode
= FMODE_EXCL
;
2147 set_blocksize(device
->bdev
, 4096);
2148 device
->fs_devices
= fs_info
->fs_devices
;
2149 list_add(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2150 fs_info
->fs_devices
->num_devices
++;
2151 fs_info
->fs_devices
->open_devices
++;
2152 if (device
->can_discard
)
2153 fs_info
->fs_devices
->num_can_discard
++;
2154 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2156 *device_out
= device
;
2160 blkdev_put(bdev
, FMODE_EXCL
);
2164 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info
*fs_info
,
2165 struct btrfs_device
*tgtdev
)
2167 WARN_ON(fs_info
->fs_devices
->rw_devices
== 0);
2168 tgtdev
->io_width
= fs_info
->dev_root
->sectorsize
;
2169 tgtdev
->io_align
= fs_info
->dev_root
->sectorsize
;
2170 tgtdev
->sector_size
= fs_info
->dev_root
->sectorsize
;
2171 tgtdev
->dev_root
= fs_info
->dev_root
;
2172 tgtdev
->in_fs_metadata
= 1;
2175 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2176 struct btrfs_device
*device
)
2179 struct btrfs_path
*path
;
2180 struct btrfs_root
*root
;
2181 struct btrfs_dev_item
*dev_item
;
2182 struct extent_buffer
*leaf
;
2183 struct btrfs_key key
;
2185 root
= device
->dev_root
->fs_info
->chunk_root
;
2187 path
= btrfs_alloc_path();
2191 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2192 key
.type
= BTRFS_DEV_ITEM_KEY
;
2193 key
.offset
= device
->devid
;
2195 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2204 leaf
= path
->nodes
[0];
2205 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2207 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2208 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2209 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2210 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2211 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2212 btrfs_set_device_total_bytes(leaf
, dev_item
, device
->disk_total_bytes
);
2213 btrfs_set_device_bytes_used(leaf
, dev_item
, device
->bytes_used
);
2214 btrfs_mark_buffer_dirty(leaf
);
2217 btrfs_free_path(path
);
2221 static int __btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2222 struct btrfs_device
*device
, u64 new_size
)
2224 struct btrfs_super_block
*super_copy
=
2225 device
->dev_root
->fs_info
->super_copy
;
2226 u64 old_total
= btrfs_super_total_bytes(super_copy
);
2227 u64 diff
= new_size
- device
->total_bytes
;
2229 if (!device
->writeable
)
2231 if (new_size
<= device
->total_bytes
||
2232 device
->is_tgtdev_for_dev_replace
)
2235 btrfs_set_super_total_bytes(super_copy
, old_total
+ diff
);
2236 device
->fs_devices
->total_rw_bytes
+= diff
;
2238 device
->total_bytes
= new_size
;
2239 device
->disk_total_bytes
= new_size
;
2240 btrfs_clear_space_info_full(device
->dev_root
->fs_info
);
2242 return btrfs_update_device(trans
, device
);
2245 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2246 struct btrfs_device
*device
, u64 new_size
)
2249 lock_chunks(device
->dev_root
);
2250 ret
= __btrfs_grow_device(trans
, device
, new_size
);
2251 unlock_chunks(device
->dev_root
);
2255 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
2256 struct btrfs_root
*root
,
2257 u64 chunk_tree
, u64 chunk_objectid
,
2261 struct btrfs_path
*path
;
2262 struct btrfs_key key
;
2264 root
= root
->fs_info
->chunk_root
;
2265 path
= btrfs_alloc_path();
2269 key
.objectid
= chunk_objectid
;
2270 key
.offset
= chunk_offset
;
2271 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2273 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2276 else if (ret
> 0) { /* Logic error or corruption */
2277 btrfs_error(root
->fs_info
, -ENOENT
,
2278 "Failed lookup while freeing chunk.");
2283 ret
= btrfs_del_item(trans
, root
, path
);
2285 btrfs_error(root
->fs_info
, ret
,
2286 "Failed to delete chunk item.");
2288 btrfs_free_path(path
);
2292 static int btrfs_del_sys_chunk(struct btrfs_root
*root
, u64 chunk_objectid
, u64
2295 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
2296 struct btrfs_disk_key
*disk_key
;
2297 struct btrfs_chunk
*chunk
;
2304 struct btrfs_key key
;
2306 array_size
= btrfs_super_sys_array_size(super_copy
);
2308 ptr
= super_copy
->sys_chunk_array
;
2311 while (cur
< array_size
) {
2312 disk_key
= (struct btrfs_disk_key
*)ptr
;
2313 btrfs_disk_key_to_cpu(&key
, disk_key
);
2315 len
= sizeof(*disk_key
);
2317 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2318 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2319 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2320 len
+= btrfs_chunk_item_size(num_stripes
);
2325 if (key
.objectid
== chunk_objectid
&&
2326 key
.offset
== chunk_offset
) {
2327 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2329 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2338 static int btrfs_relocate_chunk(struct btrfs_root
*root
,
2339 u64 chunk_tree
, u64 chunk_objectid
,
2342 struct extent_map_tree
*em_tree
;
2343 struct btrfs_root
*extent_root
;
2344 struct btrfs_trans_handle
*trans
;
2345 struct extent_map
*em
;
2346 struct map_lookup
*map
;
2350 root
= root
->fs_info
->chunk_root
;
2351 extent_root
= root
->fs_info
->extent_root
;
2352 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
2354 ret
= btrfs_can_relocate(extent_root
, chunk_offset
);
2358 /* step one, relocate all the extents inside this chunk */
2359 ret
= btrfs_relocate_block_group(extent_root
, chunk_offset
);
2363 trans
= btrfs_start_transaction(root
, 0);
2364 BUG_ON(IS_ERR(trans
));
2369 * step two, delete the device extents and the
2370 * chunk tree entries
2372 read_lock(&em_tree
->lock
);
2373 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
2374 read_unlock(&em_tree
->lock
);
2376 BUG_ON(!em
|| em
->start
> chunk_offset
||
2377 em
->start
+ em
->len
< chunk_offset
);
2378 map
= (struct map_lookup
*)em
->bdev
;
2380 for (i
= 0; i
< map
->num_stripes
; i
++) {
2381 ret
= btrfs_free_dev_extent(trans
, map
->stripes
[i
].dev
,
2382 map
->stripes
[i
].physical
);
2385 if (map
->stripes
[i
].dev
) {
2386 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2390 ret
= btrfs_free_chunk(trans
, root
, chunk_tree
, chunk_objectid
,
2395 trace_btrfs_chunk_free(root
, map
, chunk_offset
, em
->len
);
2397 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2398 ret
= btrfs_del_sys_chunk(root
, chunk_objectid
, chunk_offset
);
2402 ret
= btrfs_remove_block_group(trans
, extent_root
, chunk_offset
);
2405 write_lock(&em_tree
->lock
);
2406 remove_extent_mapping(em_tree
, em
);
2407 write_unlock(&em_tree
->lock
);
2412 /* once for the tree */
2413 free_extent_map(em
);
2415 free_extent_map(em
);
2417 unlock_chunks(root
);
2418 btrfs_end_transaction(trans
, root
);
2422 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
)
2424 struct btrfs_root
*chunk_root
= root
->fs_info
->chunk_root
;
2425 struct btrfs_path
*path
;
2426 struct extent_buffer
*leaf
;
2427 struct btrfs_chunk
*chunk
;
2428 struct btrfs_key key
;
2429 struct btrfs_key found_key
;
2430 u64 chunk_tree
= chunk_root
->root_key
.objectid
;
2432 bool retried
= false;
2436 path
= btrfs_alloc_path();
2441 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2442 key
.offset
= (u64
)-1;
2443 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2446 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2449 BUG_ON(ret
== 0); /* Corruption */
2451 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
2458 leaf
= path
->nodes
[0];
2459 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2461 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
2462 struct btrfs_chunk
);
2463 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2464 btrfs_release_path(path
);
2466 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2467 ret
= btrfs_relocate_chunk(chunk_root
, chunk_tree
,
2476 if (found_key
.offset
== 0)
2478 key
.offset
= found_key
.offset
- 1;
2481 if (failed
&& !retried
) {
2485 } else if (failed
&& retried
) {
2490 btrfs_free_path(path
);
2494 static int insert_balance_item(struct btrfs_root
*root
,
2495 struct btrfs_balance_control
*bctl
)
2497 struct btrfs_trans_handle
*trans
;
2498 struct btrfs_balance_item
*item
;
2499 struct btrfs_disk_balance_args disk_bargs
;
2500 struct btrfs_path
*path
;
2501 struct extent_buffer
*leaf
;
2502 struct btrfs_key key
;
2505 path
= btrfs_alloc_path();
2509 trans
= btrfs_start_transaction(root
, 0);
2510 if (IS_ERR(trans
)) {
2511 btrfs_free_path(path
);
2512 return PTR_ERR(trans
);
2515 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2516 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2519 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2524 leaf
= path
->nodes
[0];
2525 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
2527 memset_extent_buffer(leaf
, 0, (unsigned long)item
, sizeof(*item
));
2529 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
2530 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
2531 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
2532 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
2533 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
2534 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
2536 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
2538 btrfs_mark_buffer_dirty(leaf
);
2540 btrfs_free_path(path
);
2541 err
= btrfs_commit_transaction(trans
, root
);
2547 static int del_balance_item(struct btrfs_root
*root
)
2549 struct btrfs_trans_handle
*trans
;
2550 struct btrfs_path
*path
;
2551 struct btrfs_key key
;
2554 path
= btrfs_alloc_path();
2558 trans
= btrfs_start_transaction(root
, 0);
2559 if (IS_ERR(trans
)) {
2560 btrfs_free_path(path
);
2561 return PTR_ERR(trans
);
2564 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2565 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2568 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2576 ret
= btrfs_del_item(trans
, root
, path
);
2578 btrfs_free_path(path
);
2579 err
= btrfs_commit_transaction(trans
, root
);
2586 * This is a heuristic used to reduce the number of chunks balanced on
2587 * resume after balance was interrupted.
2589 static void update_balance_args(struct btrfs_balance_control
*bctl
)
2592 * Turn on soft mode for chunk types that were being converted.
2594 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2595 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2596 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2597 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2598 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2599 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2602 * Turn on usage filter if is not already used. The idea is
2603 * that chunks that we have already balanced should be
2604 * reasonably full. Don't do it for chunks that are being
2605 * converted - that will keep us from relocating unconverted
2606 * (albeit full) chunks.
2608 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2609 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2610 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2611 bctl
->data
.usage
= 90;
2613 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2614 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2615 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2616 bctl
->sys
.usage
= 90;
2618 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2619 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2620 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2621 bctl
->meta
.usage
= 90;
2626 * Should be called with both balance and volume mutexes held to
2627 * serialize other volume operations (add_dev/rm_dev/resize) with
2628 * restriper. Same goes for unset_balance_control.
2630 static void set_balance_control(struct btrfs_balance_control
*bctl
)
2632 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
2634 BUG_ON(fs_info
->balance_ctl
);
2636 spin_lock(&fs_info
->balance_lock
);
2637 fs_info
->balance_ctl
= bctl
;
2638 spin_unlock(&fs_info
->balance_lock
);
2641 static void unset_balance_control(struct btrfs_fs_info
*fs_info
)
2643 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
2645 BUG_ON(!fs_info
->balance_ctl
);
2647 spin_lock(&fs_info
->balance_lock
);
2648 fs_info
->balance_ctl
= NULL
;
2649 spin_unlock(&fs_info
->balance_lock
);
2655 * Balance filters. Return 1 if chunk should be filtered out
2656 * (should not be balanced).
2658 static int chunk_profiles_filter(u64 chunk_type
,
2659 struct btrfs_balance_args
*bargs
)
2661 chunk_type
= chunk_to_extended(chunk_type
) &
2662 BTRFS_EXTENDED_PROFILE_MASK
;
2664 if (bargs
->profiles
& chunk_type
)
2670 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
2671 struct btrfs_balance_args
*bargs
)
2673 struct btrfs_block_group_cache
*cache
;
2674 u64 chunk_used
, user_thresh
;
2677 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
2678 chunk_used
= btrfs_block_group_used(&cache
->item
);
2680 if (bargs
->usage
== 0)
2682 else if (bargs
->usage
> 100)
2683 user_thresh
= cache
->key
.offset
;
2685 user_thresh
= div_factor_fine(cache
->key
.offset
,
2688 if (chunk_used
< user_thresh
)
2691 btrfs_put_block_group(cache
);
2695 static int chunk_devid_filter(struct extent_buffer
*leaf
,
2696 struct btrfs_chunk
*chunk
,
2697 struct btrfs_balance_args
*bargs
)
2699 struct btrfs_stripe
*stripe
;
2700 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2703 for (i
= 0; i
< num_stripes
; i
++) {
2704 stripe
= btrfs_stripe_nr(chunk
, i
);
2705 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
2712 /* [pstart, pend) */
2713 static int chunk_drange_filter(struct extent_buffer
*leaf
,
2714 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
);
2725 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
2728 if (btrfs_chunk_type(leaf
, chunk
) & (BTRFS_BLOCK_GROUP_DUP
|
2729 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
))
2733 factor
= num_stripes
/ factor
;
2735 for (i
= 0; i
< num_stripes
; i
++) {
2736 stripe
= btrfs_stripe_nr(chunk
, i
);
2737 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
2740 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
2741 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
2742 do_div(stripe_length
, factor
);
2744 if (stripe_offset
< bargs
->pend
&&
2745 stripe_offset
+ stripe_length
> bargs
->pstart
)
2752 /* [vstart, vend) */
2753 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
2754 struct btrfs_chunk
*chunk
,
2756 struct btrfs_balance_args
*bargs
)
2758 if (chunk_offset
< bargs
->vend
&&
2759 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
2760 /* at least part of the chunk is inside this vrange */
2766 static int chunk_soft_convert_filter(u64 chunk_type
,
2767 struct btrfs_balance_args
*bargs
)
2769 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
2772 chunk_type
= chunk_to_extended(chunk_type
) &
2773 BTRFS_EXTENDED_PROFILE_MASK
;
2775 if (bargs
->target
== chunk_type
)
2781 static int should_balance_chunk(struct btrfs_root
*root
,
2782 struct extent_buffer
*leaf
,
2783 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
2785 struct btrfs_balance_control
*bctl
= root
->fs_info
->balance_ctl
;
2786 struct btrfs_balance_args
*bargs
= NULL
;
2787 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2790 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
2791 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
2795 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
2796 bargs
= &bctl
->data
;
2797 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
2799 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
2800 bargs
= &bctl
->meta
;
2802 /* profiles filter */
2803 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
2804 chunk_profiles_filter(chunk_type
, bargs
)) {
2809 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2810 chunk_usage_filter(bctl
->fs_info
, chunk_offset
, bargs
)) {
2815 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
2816 chunk_devid_filter(leaf
, chunk
, bargs
)) {
2820 /* drange filter, makes sense only with devid filter */
2821 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
2822 chunk_drange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
2827 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
2828 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
2832 /* soft profile changing mode */
2833 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
2834 chunk_soft_convert_filter(chunk_type
, bargs
)) {
2841 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
2843 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
2844 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
2845 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
2846 struct list_head
*devices
;
2847 struct btrfs_device
*device
;
2850 struct btrfs_chunk
*chunk
;
2851 struct btrfs_path
*path
;
2852 struct btrfs_key key
;
2853 struct btrfs_key found_key
;
2854 struct btrfs_trans_handle
*trans
;
2855 struct extent_buffer
*leaf
;
2858 int enospc_errors
= 0;
2859 bool counting
= true;
2861 /* step one make some room on all the devices */
2862 devices
= &fs_info
->fs_devices
->devices
;
2863 list_for_each_entry(device
, devices
, dev_list
) {
2864 old_size
= device
->total_bytes
;
2865 size_to_free
= div_factor(old_size
, 1);
2866 size_to_free
= min(size_to_free
, (u64
)1 * 1024 * 1024);
2867 if (!device
->writeable
||
2868 device
->total_bytes
- device
->bytes_used
> size_to_free
||
2869 device
->is_tgtdev_for_dev_replace
)
2872 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
2877 trans
= btrfs_start_transaction(dev_root
, 0);
2878 BUG_ON(IS_ERR(trans
));
2880 ret
= btrfs_grow_device(trans
, device
, old_size
);
2883 btrfs_end_transaction(trans
, dev_root
);
2886 /* step two, relocate all the chunks */
2887 path
= btrfs_alloc_path();
2893 /* zero out stat counters */
2894 spin_lock(&fs_info
->balance_lock
);
2895 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
2896 spin_unlock(&fs_info
->balance_lock
);
2898 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2899 key
.offset
= (u64
)-1;
2900 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2903 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
2904 atomic_read(&fs_info
->balance_cancel_req
)) {
2909 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2914 * this shouldn't happen, it means the last relocate
2918 BUG(); /* FIXME break ? */
2920 ret
= btrfs_previous_item(chunk_root
, path
, 0,
2921 BTRFS_CHUNK_ITEM_KEY
);
2927 leaf
= path
->nodes
[0];
2928 slot
= path
->slots
[0];
2929 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2931 if (found_key
.objectid
!= key
.objectid
)
2934 /* chunk zero is special */
2935 if (found_key
.offset
== 0)
2938 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
2941 spin_lock(&fs_info
->balance_lock
);
2942 bctl
->stat
.considered
++;
2943 spin_unlock(&fs_info
->balance_lock
);
2946 ret
= should_balance_chunk(chunk_root
, leaf
, chunk
,
2948 btrfs_release_path(path
);
2953 spin_lock(&fs_info
->balance_lock
);
2954 bctl
->stat
.expected
++;
2955 spin_unlock(&fs_info
->balance_lock
);
2959 ret
= btrfs_relocate_chunk(chunk_root
,
2960 chunk_root
->root_key
.objectid
,
2963 if (ret
&& ret
!= -ENOSPC
)
2965 if (ret
== -ENOSPC
) {
2968 spin_lock(&fs_info
->balance_lock
);
2969 bctl
->stat
.completed
++;
2970 spin_unlock(&fs_info
->balance_lock
);
2973 key
.offset
= found_key
.offset
- 1;
2977 btrfs_release_path(path
);
2982 btrfs_free_path(path
);
2983 if (enospc_errors
) {
2984 printk(KERN_INFO
"btrfs: %d enospc errors during balance\n",
2994 * alloc_profile_is_valid - see if a given profile is valid and reduced
2995 * @flags: profile to validate
2996 * @extended: if true @flags is treated as an extended profile
2998 static int alloc_profile_is_valid(u64 flags
, int extended
)
3000 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
3001 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
3003 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
3005 /* 1) check that all other bits are zeroed */
3009 /* 2) see if profile is reduced */
3011 return !extended
; /* "0" is valid for usual profiles */
3013 /* true if exactly one bit set */
3014 return (flags
& (flags
- 1)) == 0;
3017 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
3019 /* cancel requested || normal exit path */
3020 return atomic_read(&fs_info
->balance_cancel_req
) ||
3021 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
3022 atomic_read(&fs_info
->balance_cancel_req
) == 0);
3025 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
3029 unset_balance_control(fs_info
);
3030 ret
= del_balance_item(fs_info
->tree_root
);
3033 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3036 void update_ioctl_balance_args(struct btrfs_fs_info
*fs_info
, int lock
,
3037 struct btrfs_ioctl_balance_args
*bargs
);
3040 * Should be called with both balance and volume mutexes held
3042 int btrfs_balance(struct btrfs_balance_control
*bctl
,
3043 struct btrfs_ioctl_balance_args
*bargs
)
3045 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3052 if (btrfs_fs_closing(fs_info
) ||
3053 atomic_read(&fs_info
->balance_pause_req
) ||
3054 atomic_read(&fs_info
->balance_cancel_req
)) {
3059 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
3060 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
3064 * In case of mixed groups both data and meta should be picked,
3065 * and identical options should be given for both of them.
3067 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
3068 if (mixed
&& (bctl
->flags
& allowed
)) {
3069 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
3070 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
3071 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
3072 printk(KERN_ERR
"btrfs: with mixed groups data and "
3073 "metadata balance options must be the same\n");
3079 num_devices
= fs_info
->fs_devices
->num_devices
;
3080 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
3081 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
3082 BUG_ON(num_devices
< 1);
3085 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
3086 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
3087 if (num_devices
== 1)
3088 allowed
|= BTRFS_BLOCK_GROUP_DUP
;
3089 else if (num_devices
< 4)
3090 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
3092 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
|
3093 BTRFS_BLOCK_GROUP_RAID10
);
3095 if ((bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3096 (!alloc_profile_is_valid(bctl
->data
.target
, 1) ||
3097 (bctl
->data
.target
& ~allowed
))) {
3098 printk(KERN_ERR
"btrfs: unable to start balance with target "
3099 "data profile %llu\n",
3100 (unsigned long long)bctl
->data
.target
);
3104 if ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3105 (!alloc_profile_is_valid(bctl
->meta
.target
, 1) ||
3106 (bctl
->meta
.target
& ~allowed
))) {
3107 printk(KERN_ERR
"btrfs: unable to start balance with target "
3108 "metadata profile %llu\n",
3109 (unsigned long long)bctl
->meta
.target
);
3113 if ((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3114 (!alloc_profile_is_valid(bctl
->sys
.target
, 1) ||
3115 (bctl
->sys
.target
& ~allowed
))) {
3116 printk(KERN_ERR
"btrfs: unable to start balance with target "
3117 "system profile %llu\n",
3118 (unsigned long long)bctl
->sys
.target
);
3123 /* allow dup'ed data chunks only in mixed mode */
3124 if (!mixed
&& (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3125 (bctl
->data
.target
& BTRFS_BLOCK_GROUP_DUP
)) {
3126 printk(KERN_ERR
"btrfs: dup for data is not allowed\n");
3131 /* allow to reduce meta or sys integrity only if force set */
3132 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3133 BTRFS_BLOCK_GROUP_RAID10
;
3135 seq
= read_seqbegin(&fs_info
->profiles_lock
);
3137 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3138 (fs_info
->avail_system_alloc_bits
& allowed
) &&
3139 !(bctl
->sys
.target
& allowed
)) ||
3140 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3141 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
3142 !(bctl
->meta
.target
& allowed
))) {
3143 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
3144 printk(KERN_INFO
"btrfs: force reducing metadata "
3147 printk(KERN_ERR
"btrfs: balance will reduce metadata "
3148 "integrity, use force if you want this\n");
3153 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
3155 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3156 int num_tolerated_disk_barrier_failures
;
3157 u64 target
= bctl
->sys
.target
;
3159 num_tolerated_disk_barrier_failures
=
3160 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3161 if (num_tolerated_disk_barrier_failures
> 0 &&
3163 (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID0
|
3164 BTRFS_AVAIL_ALLOC_BIT_SINGLE
)))
3165 num_tolerated_disk_barrier_failures
= 0;
3166 else if (num_tolerated_disk_barrier_failures
> 1 &&
3168 (BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)))
3169 num_tolerated_disk_barrier_failures
= 1;
3171 fs_info
->num_tolerated_disk_barrier_failures
=
3172 num_tolerated_disk_barrier_failures
;
3175 ret
= insert_balance_item(fs_info
->tree_root
, bctl
);
3176 if (ret
&& ret
!= -EEXIST
)
3179 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
3180 BUG_ON(ret
== -EEXIST
);
3181 set_balance_control(bctl
);
3183 BUG_ON(ret
!= -EEXIST
);
3184 spin_lock(&fs_info
->balance_lock
);
3185 update_balance_args(bctl
);
3186 spin_unlock(&fs_info
->balance_lock
);
3189 atomic_inc(&fs_info
->balance_running
);
3190 mutex_unlock(&fs_info
->balance_mutex
);
3192 ret
= __btrfs_balance(fs_info
);
3194 mutex_lock(&fs_info
->balance_mutex
);
3195 atomic_dec(&fs_info
->balance_running
);
3197 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3198 fs_info
->num_tolerated_disk_barrier_failures
=
3199 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3203 memset(bargs
, 0, sizeof(*bargs
));
3204 update_ioctl_balance_args(fs_info
, 0, bargs
);
3207 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
3208 balance_need_close(fs_info
)) {
3209 __cancel_balance(fs_info
);
3212 wake_up(&fs_info
->balance_wait_q
);
3216 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
3217 __cancel_balance(fs_info
);
3220 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3225 static int balance_kthread(void *data
)
3227 struct btrfs_fs_info
*fs_info
= data
;
3230 mutex_lock(&fs_info
->volume_mutex
);
3231 mutex_lock(&fs_info
->balance_mutex
);
3233 if (fs_info
->balance_ctl
) {
3234 printk(KERN_INFO
"btrfs: continuing balance\n");
3235 ret
= btrfs_balance(fs_info
->balance_ctl
, NULL
);
3238 mutex_unlock(&fs_info
->balance_mutex
);
3239 mutex_unlock(&fs_info
->volume_mutex
);
3244 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
3246 struct task_struct
*tsk
;
3248 spin_lock(&fs_info
->balance_lock
);
3249 if (!fs_info
->balance_ctl
) {
3250 spin_unlock(&fs_info
->balance_lock
);
3253 spin_unlock(&fs_info
->balance_lock
);
3255 if (btrfs_test_opt(fs_info
->tree_root
, SKIP_BALANCE
)) {
3256 printk(KERN_INFO
"btrfs: force skipping balance\n");
3260 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
3262 return PTR_ERR(tsk
);
3267 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
3269 struct btrfs_balance_control
*bctl
;
3270 struct btrfs_balance_item
*item
;
3271 struct btrfs_disk_balance_args disk_bargs
;
3272 struct btrfs_path
*path
;
3273 struct extent_buffer
*leaf
;
3274 struct btrfs_key key
;
3277 path
= btrfs_alloc_path();
3281 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3282 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
3285 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
3288 if (ret
> 0) { /* ret = -ENOENT; */
3293 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
3299 leaf
= path
->nodes
[0];
3300 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3302 bctl
->fs_info
= fs_info
;
3303 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
3304 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
3306 btrfs_balance_data(leaf
, item
, &disk_bargs
);
3307 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
3308 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
3309 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
3310 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
3311 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
3313 WARN_ON(atomic_xchg(&fs_info
->mutually_exclusive_operation_running
, 1));
3315 mutex_lock(&fs_info
->volume_mutex
);
3316 mutex_lock(&fs_info
->balance_mutex
);
3318 set_balance_control(bctl
);
3320 mutex_unlock(&fs_info
->balance_mutex
);
3321 mutex_unlock(&fs_info
->volume_mutex
);
3323 btrfs_free_path(path
);
3327 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
3331 mutex_lock(&fs_info
->balance_mutex
);
3332 if (!fs_info
->balance_ctl
) {
3333 mutex_unlock(&fs_info
->balance_mutex
);
3337 if (atomic_read(&fs_info
->balance_running
)) {
3338 atomic_inc(&fs_info
->balance_pause_req
);
3339 mutex_unlock(&fs_info
->balance_mutex
);
3341 wait_event(fs_info
->balance_wait_q
,
3342 atomic_read(&fs_info
->balance_running
) == 0);
3344 mutex_lock(&fs_info
->balance_mutex
);
3345 /* we are good with balance_ctl ripped off from under us */
3346 BUG_ON(atomic_read(&fs_info
->balance_running
));
3347 atomic_dec(&fs_info
->balance_pause_req
);
3352 mutex_unlock(&fs_info
->balance_mutex
);
3356 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
3358 mutex_lock(&fs_info
->balance_mutex
);
3359 if (!fs_info
->balance_ctl
) {
3360 mutex_unlock(&fs_info
->balance_mutex
);
3364 atomic_inc(&fs_info
->balance_cancel_req
);
3366 * if we are running just wait and return, balance item is
3367 * deleted in btrfs_balance in this case
3369 if (atomic_read(&fs_info
->balance_running
)) {
3370 mutex_unlock(&fs_info
->balance_mutex
);
3371 wait_event(fs_info
->balance_wait_q
,
3372 atomic_read(&fs_info
->balance_running
) == 0);
3373 mutex_lock(&fs_info
->balance_mutex
);
3375 /* __cancel_balance needs volume_mutex */
3376 mutex_unlock(&fs_info
->balance_mutex
);
3377 mutex_lock(&fs_info
->volume_mutex
);
3378 mutex_lock(&fs_info
->balance_mutex
);
3380 if (fs_info
->balance_ctl
)
3381 __cancel_balance(fs_info
);
3383 mutex_unlock(&fs_info
->volume_mutex
);
3386 BUG_ON(fs_info
->balance_ctl
|| atomic_read(&fs_info
->balance_running
));
3387 atomic_dec(&fs_info
->balance_cancel_req
);
3388 mutex_unlock(&fs_info
->balance_mutex
);
3393 * shrinking a device means finding all of the device extents past
3394 * the new size, and then following the back refs to the chunks.
3395 * The chunk relocation code actually frees the device extent
3397 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
3399 struct btrfs_trans_handle
*trans
;
3400 struct btrfs_root
*root
= device
->dev_root
;
3401 struct btrfs_dev_extent
*dev_extent
= NULL
;
3402 struct btrfs_path
*path
;
3410 bool retried
= false;
3411 struct extent_buffer
*l
;
3412 struct btrfs_key key
;
3413 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
3414 u64 old_total
= btrfs_super_total_bytes(super_copy
);
3415 u64 old_size
= device
->total_bytes
;
3416 u64 diff
= device
->total_bytes
- new_size
;
3418 if (device
->is_tgtdev_for_dev_replace
)
3421 path
= btrfs_alloc_path();
3429 device
->total_bytes
= new_size
;
3430 if (device
->writeable
) {
3431 device
->fs_devices
->total_rw_bytes
-= diff
;
3432 spin_lock(&root
->fs_info
->free_chunk_lock
);
3433 root
->fs_info
->free_chunk_space
-= diff
;
3434 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3436 unlock_chunks(root
);
3439 key
.objectid
= device
->devid
;
3440 key
.offset
= (u64
)-1;
3441 key
.type
= BTRFS_DEV_EXTENT_KEY
;
3444 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3448 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
3453 btrfs_release_path(path
);
3458 slot
= path
->slots
[0];
3459 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
3461 if (key
.objectid
!= device
->devid
) {
3462 btrfs_release_path(path
);
3466 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
3467 length
= btrfs_dev_extent_length(l
, dev_extent
);
3469 if (key
.offset
+ length
<= new_size
) {
3470 btrfs_release_path(path
);
3474 chunk_tree
= btrfs_dev_extent_chunk_tree(l
, dev_extent
);
3475 chunk_objectid
= btrfs_dev_extent_chunk_objectid(l
, dev_extent
);
3476 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
3477 btrfs_release_path(path
);
3479 ret
= btrfs_relocate_chunk(root
, chunk_tree
, chunk_objectid
,
3481 if (ret
&& ret
!= -ENOSPC
)
3485 } while (key
.offset
-- > 0);
3487 if (failed
&& !retried
) {
3491 } else if (failed
&& retried
) {
3495 device
->total_bytes
= old_size
;
3496 if (device
->writeable
)
3497 device
->fs_devices
->total_rw_bytes
+= diff
;
3498 spin_lock(&root
->fs_info
->free_chunk_lock
);
3499 root
->fs_info
->free_chunk_space
+= diff
;
3500 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3501 unlock_chunks(root
);
3505 /* Shrinking succeeded, else we would be at "done". */
3506 trans
= btrfs_start_transaction(root
, 0);
3507 if (IS_ERR(trans
)) {
3508 ret
= PTR_ERR(trans
);
3514 device
->disk_total_bytes
= new_size
;
3515 /* Now btrfs_update_device() will change the on-disk size. */
3516 ret
= btrfs_update_device(trans
, device
);
3518 unlock_chunks(root
);
3519 btrfs_end_transaction(trans
, root
);
3522 WARN_ON(diff
> old_total
);
3523 btrfs_set_super_total_bytes(super_copy
, old_total
- diff
);
3524 unlock_chunks(root
);
3525 btrfs_end_transaction(trans
, root
);
3527 btrfs_free_path(path
);
3531 static int btrfs_add_system_chunk(struct btrfs_root
*root
,
3532 struct btrfs_key
*key
,
3533 struct btrfs_chunk
*chunk
, int item_size
)
3535 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
3536 struct btrfs_disk_key disk_key
;
3540 array_size
= btrfs_super_sys_array_size(super_copy
);
3541 if (array_size
+ item_size
> BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
)
3544 ptr
= super_copy
->sys_chunk_array
+ array_size
;
3545 btrfs_cpu_key_to_disk(&disk_key
, key
);
3546 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
3547 ptr
+= sizeof(disk_key
);
3548 memcpy(ptr
, chunk
, item_size
);
3549 item_size
+= sizeof(disk_key
);
3550 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
3555 * sort the devices in descending order by max_avail, total_avail
3557 static int btrfs_cmp_device_info(const void *a
, const void *b
)
3559 const struct btrfs_device_info
*di_a
= a
;
3560 const struct btrfs_device_info
*di_b
= b
;
3562 if (di_a
->max_avail
> di_b
->max_avail
)
3564 if (di_a
->max_avail
< di_b
->max_avail
)
3566 if (di_a
->total_avail
> di_b
->total_avail
)
3568 if (di_a
->total_avail
< di_b
->total_avail
)
3573 struct btrfs_raid_attr btrfs_raid_array
[BTRFS_NR_RAID_TYPES
] = {
3574 [BTRFS_RAID_RAID10
] = {
3577 .devs_max
= 0, /* 0 == as many as possible */
3579 .devs_increment
= 2,
3582 [BTRFS_RAID_RAID1
] = {
3587 .devs_increment
= 2,
3590 [BTRFS_RAID_DUP
] = {
3595 .devs_increment
= 1,
3598 [BTRFS_RAID_RAID0
] = {
3603 .devs_increment
= 1,
3606 [BTRFS_RAID_SINGLE
] = {
3611 .devs_increment
= 1,
3616 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
3617 struct btrfs_root
*extent_root
,
3618 struct map_lookup
**map_ret
,
3619 u64
*num_bytes_out
, u64
*stripe_size_out
,
3620 u64 start
, u64 type
)
3622 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
3623 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
3624 struct list_head
*cur
;
3625 struct map_lookup
*map
= NULL
;
3626 struct extent_map_tree
*em_tree
;
3627 struct extent_map
*em
;
3628 struct btrfs_device_info
*devices_info
= NULL
;
3630 int num_stripes
; /* total number of stripes to allocate */
3631 int sub_stripes
; /* sub_stripes info for map */
3632 int dev_stripes
; /* stripes per dev */
3633 int devs_max
; /* max devs to use */
3634 int devs_min
; /* min devs needed */
3635 int devs_increment
; /* ndevs has to be a multiple of this */
3636 int ncopies
; /* how many copies to data has */
3638 u64 max_stripe_size
;
3647 BUG_ON(!alloc_profile_is_valid(type
, 0));
3649 if (list_empty(&fs_devices
->alloc_list
))
3652 index
= __get_raid_index(type
);
3654 sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
3655 dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
3656 devs_max
= btrfs_raid_array
[index
].devs_max
;
3657 devs_min
= btrfs_raid_array
[index
].devs_min
;
3658 devs_increment
= btrfs_raid_array
[index
].devs_increment
;
3659 ncopies
= btrfs_raid_array
[index
].ncopies
;
3661 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
3662 max_stripe_size
= 1024 * 1024 * 1024;
3663 max_chunk_size
= 10 * max_stripe_size
;
3664 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
3665 /* for larger filesystems, use larger metadata chunks */
3666 if (fs_devices
->total_rw_bytes
> 50ULL * 1024 * 1024 * 1024)
3667 max_stripe_size
= 1024 * 1024 * 1024;
3669 max_stripe_size
= 256 * 1024 * 1024;
3670 max_chunk_size
= max_stripe_size
;
3671 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3672 max_stripe_size
= 32 * 1024 * 1024;
3673 max_chunk_size
= 2 * max_stripe_size
;
3675 printk(KERN_ERR
"btrfs: invalid chunk type 0x%llx requested\n",
3680 /* we don't want a chunk larger than 10% of writeable space */
3681 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
3684 devices_info
= kzalloc(sizeof(*devices_info
) * fs_devices
->rw_devices
,
3689 cur
= fs_devices
->alloc_list
.next
;
3692 * in the first pass through the devices list, we gather information
3693 * about the available holes on each device.
3696 while (cur
!= &fs_devices
->alloc_list
) {
3697 struct btrfs_device
*device
;
3701 device
= list_entry(cur
, struct btrfs_device
, dev_alloc_list
);
3705 if (!device
->writeable
) {
3707 "btrfs: read-only device in alloc_list\n");
3711 if (!device
->in_fs_metadata
||
3712 device
->is_tgtdev_for_dev_replace
)
3715 if (device
->total_bytes
> device
->bytes_used
)
3716 total_avail
= device
->total_bytes
- device
->bytes_used
;
3720 /* If there is no space on this device, skip it. */
3721 if (total_avail
== 0)
3724 ret
= find_free_dev_extent(device
,
3725 max_stripe_size
* dev_stripes
,
3726 &dev_offset
, &max_avail
);
3727 if (ret
&& ret
!= -ENOSPC
)
3731 max_avail
= max_stripe_size
* dev_stripes
;
3733 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
3736 if (ndevs
== fs_devices
->rw_devices
) {
3737 WARN(1, "%s: found more than %llu devices\n",
3738 __func__
, fs_devices
->rw_devices
);
3741 devices_info
[ndevs
].dev_offset
= dev_offset
;
3742 devices_info
[ndevs
].max_avail
= max_avail
;
3743 devices_info
[ndevs
].total_avail
= total_avail
;
3744 devices_info
[ndevs
].dev
= device
;
3749 * now sort the devices by hole size / available space
3751 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
3752 btrfs_cmp_device_info
, NULL
);
3754 /* round down to number of usable stripes */
3755 ndevs
-= ndevs
% devs_increment
;
3757 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
3762 if (devs_max
&& ndevs
> devs_max
)
3765 * the primary goal is to maximize the number of stripes, so use as many
3766 * devices as possible, even if the stripes are not maximum sized.
3768 stripe_size
= devices_info
[ndevs
-1].max_avail
;
3769 num_stripes
= ndevs
* dev_stripes
;
3771 if (stripe_size
* ndevs
> max_chunk_size
* ncopies
) {
3772 stripe_size
= max_chunk_size
* ncopies
;
3773 do_div(stripe_size
, ndevs
);
3776 do_div(stripe_size
, dev_stripes
);
3778 /* align to BTRFS_STRIPE_LEN */
3779 do_div(stripe_size
, BTRFS_STRIPE_LEN
);
3780 stripe_size
*= BTRFS_STRIPE_LEN
;
3782 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
3787 map
->num_stripes
= num_stripes
;
3789 for (i
= 0; i
< ndevs
; ++i
) {
3790 for (j
= 0; j
< dev_stripes
; ++j
) {
3791 int s
= i
* dev_stripes
+ j
;
3792 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
3793 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
3797 map
->sector_size
= extent_root
->sectorsize
;
3798 map
->stripe_len
= BTRFS_STRIPE_LEN
;
3799 map
->io_align
= BTRFS_STRIPE_LEN
;
3800 map
->io_width
= BTRFS_STRIPE_LEN
;
3802 map
->sub_stripes
= sub_stripes
;
3805 num_bytes
= stripe_size
* (num_stripes
/ ncopies
);
3807 *stripe_size_out
= stripe_size
;
3808 *num_bytes_out
= num_bytes
;
3810 trace_btrfs_chunk_alloc(info
->chunk_root
, map
, start
, num_bytes
);
3812 em
= alloc_extent_map();
3817 em
->bdev
= (struct block_device
*)map
;
3819 em
->len
= num_bytes
;
3820 em
->block_start
= 0;
3821 em
->block_len
= em
->len
;
3823 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
3824 write_lock(&em_tree
->lock
);
3825 ret
= add_extent_mapping(em_tree
, em
);
3826 write_unlock(&em_tree
->lock
);
3828 free_extent_map(em
);
3832 for (i
= 0; i
< map
->num_stripes
; ++i
) {
3833 struct btrfs_device
*device
;
3836 device
= map
->stripes
[i
].dev
;
3837 dev_offset
= map
->stripes
[i
].physical
;
3839 ret
= btrfs_alloc_dev_extent(trans
, device
,
3840 info
->chunk_root
->root_key
.objectid
,
3841 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
3842 start
, dev_offset
, stripe_size
);
3844 goto error_dev_extent
;
3847 ret
= btrfs_make_block_group(trans
, extent_root
, 0, type
,
3848 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
3851 i
= map
->num_stripes
- 1;
3852 goto error_dev_extent
;
3855 free_extent_map(em
);
3856 kfree(devices_info
);
3860 for (; i
>= 0; i
--) {
3861 struct btrfs_device
*device
;
3864 device
= map
->stripes
[i
].dev
;
3865 err
= btrfs_free_dev_extent(trans
, device
, start
);
3867 btrfs_abort_transaction(trans
, extent_root
, err
);
3871 write_lock(&em_tree
->lock
);
3872 remove_extent_mapping(em_tree
, em
);
3873 write_unlock(&em_tree
->lock
);
3875 /* One for our allocation */
3876 free_extent_map(em
);
3877 /* One for the tree reference */
3878 free_extent_map(em
);
3881 kfree(devices_info
);
3885 static int __finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
3886 struct btrfs_root
*extent_root
,
3887 struct map_lookup
*map
, u64 chunk_offset
,
3888 u64 chunk_size
, u64 stripe_size
)
3891 struct btrfs_key key
;
3892 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
3893 struct btrfs_device
*device
;
3894 struct btrfs_chunk
*chunk
;
3895 struct btrfs_stripe
*stripe
;
3896 size_t item_size
= btrfs_chunk_item_size(map
->num_stripes
);
3900 chunk
= kzalloc(item_size
, GFP_NOFS
);
3905 while (index
< map
->num_stripes
) {
3906 device
= map
->stripes
[index
].dev
;
3907 device
->bytes_used
+= stripe_size
;
3908 ret
= btrfs_update_device(trans
, device
);
3914 spin_lock(&extent_root
->fs_info
->free_chunk_lock
);
3915 extent_root
->fs_info
->free_chunk_space
-= (stripe_size
*
3917 spin_unlock(&extent_root
->fs_info
->free_chunk_lock
);
3920 stripe
= &chunk
->stripe
;
3921 while (index
< map
->num_stripes
) {
3922 device
= map
->stripes
[index
].dev
;
3923 dev_offset
= map
->stripes
[index
].physical
;
3925 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
3926 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
3927 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
3932 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
3933 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
3934 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
3935 btrfs_set_stack_chunk_type(chunk
, map
->type
);
3936 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
3937 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
3938 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
3939 btrfs_set_stack_chunk_sector_size(chunk
, extent_root
->sectorsize
);
3940 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
3942 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3943 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3944 key
.offset
= chunk_offset
;
3946 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
3948 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3950 * TODO: Cleanup of inserted chunk root in case of
3953 ret
= btrfs_add_system_chunk(chunk_root
, &key
, chunk
,
3963 * Chunk allocation falls into two parts. The first part does works
3964 * that make the new allocated chunk useable, but not do any operation
3965 * that modifies the chunk tree. The second part does the works that
3966 * require modifying the chunk tree. This division is important for the
3967 * bootstrap process of adding storage to a seed btrfs.
3969 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
3970 struct btrfs_root
*extent_root
, u64 type
)
3975 struct map_lookup
*map
;
3976 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
3979 ret
= find_next_chunk(chunk_root
, BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
3984 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &map
, &chunk_size
,
3985 &stripe_size
, chunk_offset
, type
);
3989 ret
= __finish_chunk_alloc(trans
, extent_root
, map
, chunk_offset
,
3990 chunk_size
, stripe_size
);
3996 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
3997 struct btrfs_root
*root
,
3998 struct btrfs_device
*device
)
4001 u64 sys_chunk_offset
;
4005 u64 sys_stripe_size
;
4007 struct map_lookup
*map
;
4008 struct map_lookup
*sys_map
;
4009 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4010 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
4013 ret
= find_next_chunk(fs_info
->chunk_root
,
4014 BTRFS_FIRST_CHUNK_TREE_OBJECTID
, &chunk_offset
);
4018 alloc_profile
= btrfs_get_alloc_profile(extent_root
, 0);
4019 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &map
, &chunk_size
,
4020 &stripe_size
, chunk_offset
, alloc_profile
);
4024 sys_chunk_offset
= chunk_offset
+ chunk_size
;
4026 alloc_profile
= btrfs_get_alloc_profile(fs_info
->chunk_root
, 0);
4027 ret
= __btrfs_alloc_chunk(trans
, extent_root
, &sys_map
,
4028 &sys_chunk_size
, &sys_stripe_size
,
4029 sys_chunk_offset
, alloc_profile
);
4031 btrfs_abort_transaction(trans
, root
, ret
);
4035 ret
= btrfs_add_device(trans
, fs_info
->chunk_root
, device
);
4037 btrfs_abort_transaction(trans
, root
, ret
);
4042 * Modifying chunk tree needs allocating new blocks from both
4043 * system block group and metadata block group. So we only can
4044 * do operations require modifying the chunk tree after both
4045 * block groups were created.
4047 ret
= __finish_chunk_alloc(trans
, extent_root
, map
, chunk_offset
,
4048 chunk_size
, stripe_size
);
4050 btrfs_abort_transaction(trans
, root
, ret
);
4054 ret
= __finish_chunk_alloc(trans
, extent_root
, sys_map
,
4055 sys_chunk_offset
, sys_chunk_size
,
4058 btrfs_abort_transaction(trans
, root
, ret
);
4065 int btrfs_chunk_readonly(struct btrfs_root
*root
, u64 chunk_offset
)
4067 struct extent_map
*em
;
4068 struct map_lookup
*map
;
4069 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
4073 read_lock(&map_tree
->map_tree
.lock
);
4074 em
= lookup_extent_mapping(&map_tree
->map_tree
, chunk_offset
, 1);
4075 read_unlock(&map_tree
->map_tree
.lock
);
4079 if (btrfs_test_opt(root
, DEGRADED
)) {
4080 free_extent_map(em
);
4084 map
= (struct map_lookup
*)em
->bdev
;
4085 for (i
= 0; i
< map
->num_stripes
; i
++) {
4086 if (!map
->stripes
[i
].dev
->writeable
) {
4091 free_extent_map(em
);
4095 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
4097 extent_map_tree_init(&tree
->map_tree
);
4100 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
4102 struct extent_map
*em
;
4105 write_lock(&tree
->map_tree
.lock
);
4106 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
4108 remove_extent_mapping(&tree
->map_tree
, em
);
4109 write_unlock(&tree
->map_tree
.lock
);
4114 free_extent_map(em
);
4115 /* once for the tree */
4116 free_extent_map(em
);
4120 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
4122 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
4123 struct extent_map
*em
;
4124 struct map_lookup
*map
;
4125 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4128 read_lock(&em_tree
->lock
);
4129 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4130 read_unlock(&em_tree
->lock
);
4133 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4134 map
= (struct map_lookup
*)em
->bdev
;
4135 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
4136 ret
= map
->num_stripes
;
4137 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
4138 ret
= map
->sub_stripes
;
4141 free_extent_map(em
);
4143 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
4144 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
))
4146 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
4151 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
4152 struct map_lookup
*map
, int first
, int num
,
4153 int optimal
, int dev_replace_is_ongoing
)
4157 struct btrfs_device
*srcdev
;
4159 if (dev_replace_is_ongoing
&&
4160 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
4161 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
4162 srcdev
= fs_info
->dev_replace
.srcdev
;
4167 * try to avoid the drive that is the source drive for a
4168 * dev-replace procedure, only choose it if no other non-missing
4169 * mirror is available
4171 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
4172 if (map
->stripes
[optimal
].dev
->bdev
&&
4173 (tolerance
|| map
->stripes
[optimal
].dev
!= srcdev
))
4175 for (i
= first
; i
< first
+ num
; i
++) {
4176 if (map
->stripes
[i
].dev
->bdev
&&
4177 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
4182 /* we couldn't find one that doesn't fail. Just return something
4183 * and the io error handling code will clean up eventually
4188 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
4189 u64 logical
, u64
*length
,
4190 struct btrfs_bio
**bbio_ret
,
4193 struct extent_map
*em
;
4194 struct map_lookup
*map
;
4195 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
4196 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4199 u64 stripe_end_offset
;
4208 struct btrfs_bio
*bbio
= NULL
;
4209 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
4210 int dev_replace_is_ongoing
= 0;
4211 int num_alloc_stripes
;
4212 int patch_the_first_stripe_for_dev_replace
= 0;
4213 u64 physical_to_patch_in_first_stripe
= 0;
4215 read_lock(&em_tree
->lock
);
4216 em
= lookup_extent_mapping(em_tree
, logical
, *length
);
4217 read_unlock(&em_tree
->lock
);
4220 printk(KERN_CRIT
"btrfs: unable to find logical %llu len %llu\n",
4221 (unsigned long long)logical
,
4222 (unsigned long long)*length
);
4226 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4227 map
= (struct map_lookup
*)em
->bdev
;
4228 offset
= logical
- em
->start
;
4232 * stripe_nr counts the total number of stripes we have to stride
4233 * to get to this block
4235 do_div(stripe_nr
, map
->stripe_len
);
4237 stripe_offset
= stripe_nr
* map
->stripe_len
;
4238 BUG_ON(offset
< stripe_offset
);
4240 /* stripe_offset is the offset of this block in its stripe*/
4241 stripe_offset
= offset
- stripe_offset
;
4243 if (rw
& REQ_DISCARD
)
4244 *length
= min_t(u64
, em
->len
- offset
, *length
);
4245 else if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
4246 /* we limit the length of each bio to what fits in a stripe */
4247 *length
= min_t(u64
, em
->len
- offset
,
4248 map
->stripe_len
- stripe_offset
);
4250 *length
= em
->len
- offset
;
4256 btrfs_dev_replace_lock(dev_replace
);
4257 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
4258 if (!dev_replace_is_ongoing
)
4259 btrfs_dev_replace_unlock(dev_replace
);
4261 if (dev_replace_is_ongoing
&& mirror_num
== map
->num_stripes
+ 1 &&
4262 !(rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) &&
4263 dev_replace
->tgtdev
!= NULL
) {
4265 * in dev-replace case, for repair case (that's the only
4266 * case where the mirror is selected explicitly when
4267 * calling btrfs_map_block), blocks left of the left cursor
4268 * can also be read from the target drive.
4269 * For REQ_GET_READ_MIRRORS, the target drive is added as
4270 * the last one to the array of stripes. For READ, it also
4271 * needs to be supported using the same mirror number.
4272 * If the requested block is not left of the left cursor,
4273 * EIO is returned. This can happen because btrfs_num_copies()
4274 * returns one more in the dev-replace case.
4276 u64 tmp_length
= *length
;
4277 struct btrfs_bio
*tmp_bbio
= NULL
;
4278 int tmp_num_stripes
;
4279 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
4280 int index_srcdev
= 0;
4282 u64 physical_of_found
= 0;
4284 ret
= __btrfs_map_block(fs_info
, REQ_GET_READ_MIRRORS
,
4285 logical
, &tmp_length
, &tmp_bbio
, 0);
4287 WARN_ON(tmp_bbio
!= NULL
);
4291 tmp_num_stripes
= tmp_bbio
->num_stripes
;
4292 if (mirror_num
> tmp_num_stripes
) {
4294 * REQ_GET_READ_MIRRORS does not contain this
4295 * mirror, that means that the requested area
4296 * is not left of the left cursor
4304 * process the rest of the function using the mirror_num
4305 * of the source drive. Therefore look it up first.
4306 * At the end, patch the device pointer to the one of the
4309 for (i
= 0; i
< tmp_num_stripes
; i
++) {
4310 if (tmp_bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
4312 * In case of DUP, in order to keep it
4313 * simple, only add the mirror with the
4314 * lowest physical address
4317 physical_of_found
<=
4318 tmp_bbio
->stripes
[i
].physical
)
4323 tmp_bbio
->stripes
[i
].physical
;
4328 mirror_num
= index_srcdev
+ 1;
4329 patch_the_first_stripe_for_dev_replace
= 1;
4330 physical_to_patch_in_first_stripe
= physical_of_found
;
4339 } else if (mirror_num
> map
->num_stripes
) {
4345 stripe_nr_orig
= stripe_nr
;
4346 stripe_nr_end
= (offset
+ *length
+ map
->stripe_len
- 1) &
4347 (~(map
->stripe_len
- 1));
4348 do_div(stripe_nr_end
, map
->stripe_len
);
4349 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
4351 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
4352 if (rw
& REQ_DISCARD
)
4353 num_stripes
= min_t(u64
, map
->num_stripes
,
4354 stripe_nr_end
- stripe_nr_orig
);
4355 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
4356 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
4357 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
))
4358 num_stripes
= map
->num_stripes
;
4359 else if (mirror_num
)
4360 stripe_index
= mirror_num
- 1;
4362 stripe_index
= find_live_mirror(fs_info
, map
, 0,
4364 current
->pid
% map
->num_stripes
,
4365 dev_replace_is_ongoing
);
4366 mirror_num
= stripe_index
+ 1;
4369 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
4370 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) {
4371 num_stripes
= map
->num_stripes
;
4372 } else if (mirror_num
) {
4373 stripe_index
= mirror_num
- 1;
4378 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
4379 int factor
= map
->num_stripes
/ map
->sub_stripes
;
4381 stripe_index
= do_div(stripe_nr
, factor
);
4382 stripe_index
*= map
->sub_stripes
;
4384 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
))
4385 num_stripes
= map
->sub_stripes
;
4386 else if (rw
& REQ_DISCARD
)
4387 num_stripes
= min_t(u64
, map
->sub_stripes
*
4388 (stripe_nr_end
- stripe_nr_orig
),
4390 else if (mirror_num
)
4391 stripe_index
+= mirror_num
- 1;
4393 int old_stripe_index
= stripe_index
;
4394 stripe_index
= find_live_mirror(fs_info
, map
,
4396 map
->sub_stripes
, stripe_index
+
4397 current
->pid
% map
->sub_stripes
,
4398 dev_replace_is_ongoing
);
4399 mirror_num
= stripe_index
- old_stripe_index
+ 1;
4403 * after this do_div call, stripe_nr is the number of stripes
4404 * on this device we have to walk to find the data, and
4405 * stripe_index is the number of our device in the stripe array
4407 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
4408 mirror_num
= stripe_index
+ 1;
4410 BUG_ON(stripe_index
>= map
->num_stripes
);
4412 num_alloc_stripes
= num_stripes
;
4413 if (dev_replace_is_ongoing
) {
4414 if (rw
& (REQ_WRITE
| REQ_DISCARD
))
4415 num_alloc_stripes
<<= 1;
4416 if (rw
& REQ_GET_READ_MIRRORS
)
4417 num_alloc_stripes
++;
4419 bbio
= kzalloc(btrfs_bio_size(num_alloc_stripes
), GFP_NOFS
);
4424 atomic_set(&bbio
->error
, 0);
4426 if (rw
& REQ_DISCARD
) {
4428 int sub_stripes
= 0;
4429 u64 stripes_per_dev
= 0;
4430 u32 remaining_stripes
= 0;
4431 u32 last_stripe
= 0;
4434 (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID10
)) {
4435 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
4438 sub_stripes
= map
->sub_stripes
;
4440 factor
= map
->num_stripes
/ sub_stripes
;
4441 stripes_per_dev
= div_u64_rem(stripe_nr_end
-
4444 &remaining_stripes
);
4445 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
4446 last_stripe
*= sub_stripes
;
4449 for (i
= 0; i
< num_stripes
; i
++) {
4450 bbio
->stripes
[i
].physical
=
4451 map
->stripes
[stripe_index
].physical
+
4452 stripe_offset
+ stripe_nr
* map
->stripe_len
;
4453 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
4455 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
4456 BTRFS_BLOCK_GROUP_RAID10
)) {
4457 bbio
->stripes
[i
].length
= stripes_per_dev
*
4460 if (i
/ sub_stripes
< remaining_stripes
)
4461 bbio
->stripes
[i
].length
+=
4465 * Special for the first stripe and
4468 * |-------|...|-------|
4472 if (i
< sub_stripes
)
4473 bbio
->stripes
[i
].length
-=
4476 if (stripe_index
>= last_stripe
&&
4477 stripe_index
<= (last_stripe
+
4479 bbio
->stripes
[i
].length
-=
4482 if (i
== sub_stripes
- 1)
4485 bbio
->stripes
[i
].length
= *length
;
4488 if (stripe_index
== map
->num_stripes
) {
4489 /* This could only happen for RAID0/10 */
4495 for (i
= 0; i
< num_stripes
; i
++) {
4496 bbio
->stripes
[i
].physical
=
4497 map
->stripes
[stripe_index
].physical
+
4499 stripe_nr
* map
->stripe_len
;
4500 bbio
->stripes
[i
].dev
=
4501 map
->stripes
[stripe_index
].dev
;
4506 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
)) {
4507 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
4508 BTRFS_BLOCK_GROUP_RAID10
|
4509 BTRFS_BLOCK_GROUP_DUP
)) {
4514 if (dev_replace_is_ongoing
&& (rw
& (REQ_WRITE
| REQ_DISCARD
)) &&
4515 dev_replace
->tgtdev
!= NULL
) {
4516 int index_where_to_add
;
4517 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
4520 * duplicate the write operations while the dev replace
4521 * procedure is running. Since the copying of the old disk
4522 * to the new disk takes place at run time while the
4523 * filesystem is mounted writable, the regular write
4524 * operations to the old disk have to be duplicated to go
4525 * to the new disk as well.
4526 * Note that device->missing is handled by the caller, and
4527 * that the write to the old disk is already set up in the
4530 index_where_to_add
= num_stripes
;
4531 for (i
= 0; i
< num_stripes
; i
++) {
4532 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
4533 /* write to new disk, too */
4534 struct btrfs_bio_stripe
*new =
4535 bbio
->stripes
+ index_where_to_add
;
4536 struct btrfs_bio_stripe
*old
=
4539 new->physical
= old
->physical
;
4540 new->length
= old
->length
;
4541 new->dev
= dev_replace
->tgtdev
;
4542 index_where_to_add
++;
4546 num_stripes
= index_where_to_add
;
4547 } else if (dev_replace_is_ongoing
&& (rw
& REQ_GET_READ_MIRRORS
) &&
4548 dev_replace
->tgtdev
!= NULL
) {
4549 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
4550 int index_srcdev
= 0;
4552 u64 physical_of_found
= 0;
4555 * During the dev-replace procedure, the target drive can
4556 * also be used to read data in case it is needed to repair
4557 * a corrupt block elsewhere. This is possible if the
4558 * requested area is left of the left cursor. In this area,
4559 * the target drive is a full copy of the source drive.
4561 for (i
= 0; i
< num_stripes
; i
++) {
4562 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
4564 * In case of DUP, in order to keep it
4565 * simple, only add the mirror with the
4566 * lowest physical address
4569 physical_of_found
<=
4570 bbio
->stripes
[i
].physical
)
4574 physical_of_found
= bbio
->stripes
[i
].physical
;
4578 u64 length
= map
->stripe_len
;
4580 if (physical_of_found
+ length
<=
4581 dev_replace
->cursor_left
) {
4582 struct btrfs_bio_stripe
*tgtdev_stripe
=
4583 bbio
->stripes
+ num_stripes
;
4585 tgtdev_stripe
->physical
= physical_of_found
;
4586 tgtdev_stripe
->length
=
4587 bbio
->stripes
[index_srcdev
].length
;
4588 tgtdev_stripe
->dev
= dev_replace
->tgtdev
;
4596 bbio
->num_stripes
= num_stripes
;
4597 bbio
->max_errors
= max_errors
;
4598 bbio
->mirror_num
= mirror_num
;
4601 * this is the case that REQ_READ && dev_replace_is_ongoing &&
4602 * mirror_num == num_stripes + 1 && dev_replace target drive is
4603 * available as a mirror
4605 if (patch_the_first_stripe_for_dev_replace
&& num_stripes
> 0) {
4606 WARN_ON(num_stripes
> 1);
4607 bbio
->stripes
[0].dev
= dev_replace
->tgtdev
;
4608 bbio
->stripes
[0].physical
= physical_to_patch_in_first_stripe
;
4609 bbio
->mirror_num
= map
->num_stripes
+ 1;
4612 if (dev_replace_is_ongoing
)
4613 btrfs_dev_replace_unlock(dev_replace
);
4614 free_extent_map(em
);
4618 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
4619 u64 logical
, u64
*length
,
4620 struct btrfs_bio
**bbio_ret
, int mirror_num
)
4622 return __btrfs_map_block(fs_info
, rw
, logical
, length
, bbio_ret
,
4626 int btrfs_rmap_block(struct btrfs_mapping_tree
*map_tree
,
4627 u64 chunk_start
, u64 physical
, u64 devid
,
4628 u64
**logical
, int *naddrs
, int *stripe_len
)
4630 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4631 struct extent_map
*em
;
4632 struct map_lookup
*map
;
4639 read_lock(&em_tree
->lock
);
4640 em
= lookup_extent_mapping(em_tree
, chunk_start
, 1);
4641 read_unlock(&em_tree
->lock
);
4643 BUG_ON(!em
|| em
->start
!= chunk_start
);
4644 map
= (struct map_lookup
*)em
->bdev
;
4647 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
4648 do_div(length
, map
->num_stripes
/ map
->sub_stripes
);
4649 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
4650 do_div(length
, map
->num_stripes
);
4652 buf
= kzalloc(sizeof(u64
) * map
->num_stripes
, GFP_NOFS
);
4653 BUG_ON(!buf
); /* -ENOMEM */
4655 for (i
= 0; i
< map
->num_stripes
; i
++) {
4656 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
4658 if (map
->stripes
[i
].physical
> physical
||
4659 map
->stripes
[i
].physical
+ length
<= physical
)
4662 stripe_nr
= physical
- map
->stripes
[i
].physical
;
4663 do_div(stripe_nr
, map
->stripe_len
);
4665 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
4666 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
4667 do_div(stripe_nr
, map
->sub_stripes
);
4668 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
4669 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
4671 bytenr
= chunk_start
+ stripe_nr
* map
->stripe_len
;
4672 WARN_ON(nr
>= map
->num_stripes
);
4673 for (j
= 0; j
< nr
; j
++) {
4674 if (buf
[j
] == bytenr
)
4678 WARN_ON(nr
>= map
->num_stripes
);
4685 *stripe_len
= map
->stripe_len
;
4687 free_extent_map(em
);
4691 static void *merge_stripe_index_into_bio_private(void *bi_private
,
4692 unsigned int stripe_index
)
4695 * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
4697 * The alternative solution (instead of stealing bits from the
4698 * pointer) would be to allocate an intermediate structure
4699 * that contains the old private pointer plus the stripe_index.
4701 BUG_ON((((uintptr_t)bi_private
) & 3) != 0);
4702 BUG_ON(stripe_index
> 3);
4703 return (void *)(((uintptr_t)bi_private
) | stripe_index
);
4706 static struct btrfs_bio
*extract_bbio_from_bio_private(void *bi_private
)
4708 return (struct btrfs_bio
*)(((uintptr_t)bi_private
) & ~((uintptr_t)3));
4711 static unsigned int extract_stripe_index_from_bio_private(void *bi_private
)
4713 return (unsigned int)((uintptr_t)bi_private
) & 3;
4716 static void btrfs_end_bio(struct bio
*bio
, int err
)
4718 struct btrfs_bio
*bbio
= extract_bbio_from_bio_private(bio
->bi_private
);
4719 int is_orig_bio
= 0;
4722 atomic_inc(&bbio
->error
);
4723 if (err
== -EIO
|| err
== -EREMOTEIO
) {
4724 unsigned int stripe_index
=
4725 extract_stripe_index_from_bio_private(
4727 struct btrfs_device
*dev
;
4729 BUG_ON(stripe_index
>= bbio
->num_stripes
);
4730 dev
= bbio
->stripes
[stripe_index
].dev
;
4732 if (bio
->bi_rw
& WRITE
)
4733 btrfs_dev_stat_inc(dev
,
4734 BTRFS_DEV_STAT_WRITE_ERRS
);
4736 btrfs_dev_stat_inc(dev
,
4737 BTRFS_DEV_STAT_READ_ERRS
);
4738 if ((bio
->bi_rw
& WRITE_FLUSH
) == WRITE_FLUSH
)
4739 btrfs_dev_stat_inc(dev
,
4740 BTRFS_DEV_STAT_FLUSH_ERRS
);
4741 btrfs_dev_stat_print_on_error(dev
);
4746 if (bio
== bbio
->orig_bio
)
4749 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
4752 bio
= bbio
->orig_bio
;
4754 bio
->bi_private
= bbio
->private;
4755 bio
->bi_end_io
= bbio
->end_io
;
4756 bio
->bi_bdev
= (struct block_device
*)
4757 (unsigned long)bbio
->mirror_num
;
4758 /* only send an error to the higher layers if it is
4759 * beyond the tolerance of the multi-bio
4761 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
4765 * this bio is actually up to date, we didn't
4766 * go over the max number of errors
4768 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
4773 bio_endio(bio
, err
);
4774 } else if (!is_orig_bio
) {
4779 struct async_sched
{
4782 struct btrfs_fs_info
*info
;
4783 struct btrfs_work work
;
4787 * see run_scheduled_bios for a description of why bios are collected for
4790 * This will add one bio to the pending list for a device and make sure
4791 * the work struct is scheduled.
4793 static noinline
void schedule_bio(struct btrfs_root
*root
,
4794 struct btrfs_device
*device
,
4795 int rw
, struct bio
*bio
)
4797 int should_queue
= 1;
4798 struct btrfs_pending_bios
*pending_bios
;
4800 /* don't bother with additional async steps for reads, right now */
4801 if (!(rw
& REQ_WRITE
)) {
4803 btrfsic_submit_bio(rw
, bio
);
4809 * nr_async_bios allows us to reliably return congestion to the
4810 * higher layers. Otherwise, the async bio makes it appear we have
4811 * made progress against dirty pages when we've really just put it
4812 * on a queue for later
4814 atomic_inc(&root
->fs_info
->nr_async_bios
);
4815 WARN_ON(bio
->bi_next
);
4816 bio
->bi_next
= NULL
;
4819 spin_lock(&device
->io_lock
);
4820 if (bio
->bi_rw
& REQ_SYNC
)
4821 pending_bios
= &device
->pending_sync_bios
;
4823 pending_bios
= &device
->pending_bios
;
4825 if (pending_bios
->tail
)
4826 pending_bios
->tail
->bi_next
= bio
;
4828 pending_bios
->tail
= bio
;
4829 if (!pending_bios
->head
)
4830 pending_bios
->head
= bio
;
4831 if (device
->running_pending
)
4834 spin_unlock(&device
->io_lock
);
4837 btrfs_queue_worker(&root
->fs_info
->submit_workers
,
4841 static int bio_size_ok(struct block_device
*bdev
, struct bio
*bio
,
4844 struct bio_vec
*prev
;
4845 struct request_queue
*q
= bdev_get_queue(bdev
);
4846 unsigned short max_sectors
= queue_max_sectors(q
);
4847 struct bvec_merge_data bvm
= {
4849 .bi_sector
= sector
,
4850 .bi_rw
= bio
->bi_rw
,
4853 if (bio
->bi_vcnt
== 0) {
4858 prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
4859 if ((bio
->bi_size
>> 9) > max_sectors
)
4862 if (!q
->merge_bvec_fn
)
4865 bvm
.bi_size
= bio
->bi_size
- prev
->bv_len
;
4866 if (q
->merge_bvec_fn(q
, &bvm
, prev
) < prev
->bv_len
)
4871 static void submit_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
4872 struct bio
*bio
, u64 physical
, int dev_nr
,
4875 struct btrfs_device
*dev
= bbio
->stripes
[dev_nr
].dev
;
4877 bio
->bi_private
= bbio
;
4878 bio
->bi_private
= merge_stripe_index_into_bio_private(
4879 bio
->bi_private
, (unsigned int)dev_nr
);
4880 bio
->bi_end_io
= btrfs_end_bio
;
4881 bio
->bi_sector
= physical
>> 9;
4884 struct rcu_string
*name
;
4887 name
= rcu_dereference(dev
->name
);
4888 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
4889 "(%s id %llu), size=%u\n", rw
,
4890 (u64
)bio
->bi_sector
, (u_long
)dev
->bdev
->bd_dev
,
4891 name
->str
, dev
->devid
, bio
->bi_size
);
4895 bio
->bi_bdev
= dev
->bdev
;
4897 schedule_bio(root
, dev
, rw
, bio
);
4899 btrfsic_submit_bio(rw
, bio
);
4902 static int breakup_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
4903 struct bio
*first_bio
, struct btrfs_device
*dev
,
4904 int dev_nr
, int rw
, int async
)
4906 struct bio_vec
*bvec
= first_bio
->bi_io_vec
;
4908 int nr_vecs
= bio_get_nr_vecs(dev
->bdev
);
4909 u64 physical
= bbio
->stripes
[dev_nr
].physical
;
4912 bio
= btrfs_bio_alloc(dev
->bdev
, physical
>> 9, nr_vecs
, GFP_NOFS
);
4916 while (bvec
<= (first_bio
->bi_io_vec
+ first_bio
->bi_vcnt
- 1)) {
4917 if (bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
4918 bvec
->bv_offset
) < bvec
->bv_len
) {
4919 u64 len
= bio
->bi_size
;
4921 atomic_inc(&bbio
->stripes_pending
);
4922 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
,
4930 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
, rw
, async
);
4934 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
4936 atomic_inc(&bbio
->error
);
4937 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
4938 bio
->bi_private
= bbio
->private;
4939 bio
->bi_end_io
= bbio
->end_io
;
4940 bio
->bi_bdev
= (struct block_device
*)
4941 (unsigned long)bbio
->mirror_num
;
4942 bio
->bi_sector
= logical
>> 9;
4944 bio_endio(bio
, -EIO
);
4948 int btrfs_map_bio(struct btrfs_root
*root
, int rw
, struct bio
*bio
,
4949 int mirror_num
, int async_submit
)
4951 struct btrfs_device
*dev
;
4952 struct bio
*first_bio
= bio
;
4953 u64 logical
= (u64
)bio
->bi_sector
<< 9;
4959 struct btrfs_bio
*bbio
= NULL
;
4961 length
= bio
->bi_size
;
4962 map_length
= length
;
4964 ret
= btrfs_map_block(root
->fs_info
, rw
, logical
, &map_length
, &bbio
,
4969 total_devs
= bbio
->num_stripes
;
4970 if (map_length
< length
) {
4971 printk(KERN_CRIT
"btrfs: mapping failed logical %llu bio len %llu "
4972 "len %llu\n", (unsigned long long)logical
,
4973 (unsigned long long)length
,
4974 (unsigned long long)map_length
);
4978 bbio
->orig_bio
= first_bio
;
4979 bbio
->private = first_bio
->bi_private
;
4980 bbio
->end_io
= first_bio
->bi_end_io
;
4981 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
4983 while (dev_nr
< total_devs
) {
4984 dev
= bbio
->stripes
[dev_nr
].dev
;
4985 if (!dev
|| !dev
->bdev
|| (rw
& WRITE
&& !dev
->writeable
)) {
4986 bbio_error(bbio
, first_bio
, logical
);
4992 * Check and see if we're ok with this bio based on it's size
4993 * and offset with the given device.
4995 if (!bio_size_ok(dev
->bdev
, first_bio
,
4996 bbio
->stripes
[dev_nr
].physical
>> 9)) {
4997 ret
= breakup_stripe_bio(root
, bbio
, first_bio
, dev
,
4998 dev_nr
, rw
, async_submit
);
5004 if (dev_nr
< total_devs
- 1) {
5005 bio
= bio_clone(first_bio
, GFP_NOFS
);
5006 BUG_ON(!bio
); /* -ENOMEM */
5011 submit_stripe_bio(root
, bbio
, bio
,
5012 bbio
->stripes
[dev_nr
].physical
, dev_nr
, rw
,
5019 struct btrfs_device
*btrfs_find_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
5022 struct btrfs_device
*device
;
5023 struct btrfs_fs_devices
*cur_devices
;
5025 cur_devices
= fs_info
->fs_devices
;
5026 while (cur_devices
) {
5028 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
5029 device
= __find_device(&cur_devices
->devices
,
5034 cur_devices
= cur_devices
->seed
;
5039 static struct btrfs_device
*add_missing_dev(struct btrfs_root
*root
,
5040 u64 devid
, u8
*dev_uuid
)
5042 struct btrfs_device
*device
;
5043 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
5045 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
5048 list_add(&device
->dev_list
,
5049 &fs_devices
->devices
);
5050 device
->dev_root
= root
->fs_info
->dev_root
;
5051 device
->devid
= devid
;
5052 device
->work
.func
= pending_bios_fn
;
5053 device
->fs_devices
= fs_devices
;
5054 device
->missing
= 1;
5055 fs_devices
->num_devices
++;
5056 fs_devices
->missing_devices
++;
5057 spin_lock_init(&device
->io_lock
);
5058 INIT_LIST_HEAD(&device
->dev_alloc_list
);
5059 memcpy(device
->uuid
, dev_uuid
, BTRFS_UUID_SIZE
);
5063 static int read_one_chunk(struct btrfs_root
*root
, struct btrfs_key
*key
,
5064 struct extent_buffer
*leaf
,
5065 struct btrfs_chunk
*chunk
)
5067 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
5068 struct map_lookup
*map
;
5069 struct extent_map
*em
;
5073 u8 uuid
[BTRFS_UUID_SIZE
];
5078 logical
= key
->offset
;
5079 length
= btrfs_chunk_length(leaf
, chunk
);
5081 read_lock(&map_tree
->map_tree
.lock
);
5082 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
5083 read_unlock(&map_tree
->map_tree
.lock
);
5085 /* already mapped? */
5086 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
5087 free_extent_map(em
);
5090 free_extent_map(em
);
5093 em
= alloc_extent_map();
5096 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
5097 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
5099 free_extent_map(em
);
5103 em
->bdev
= (struct block_device
*)map
;
5104 em
->start
= logical
;
5107 em
->block_start
= 0;
5108 em
->block_len
= em
->len
;
5110 map
->num_stripes
= num_stripes
;
5111 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
5112 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
5113 map
->sector_size
= btrfs_chunk_sector_size(leaf
, chunk
);
5114 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
5115 map
->type
= btrfs_chunk_type(leaf
, chunk
);
5116 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
5117 for (i
= 0; i
< num_stripes
; i
++) {
5118 map
->stripes
[i
].physical
=
5119 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
5120 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
5121 read_extent_buffer(leaf
, uuid
, (unsigned long)
5122 btrfs_stripe_dev_uuid_nr(chunk
, i
),
5124 map
->stripes
[i
].dev
= btrfs_find_device(root
->fs_info
, devid
,
5126 if (!map
->stripes
[i
].dev
&& !btrfs_test_opt(root
, DEGRADED
)) {
5128 free_extent_map(em
);
5131 if (!map
->stripes
[i
].dev
) {
5132 map
->stripes
[i
].dev
=
5133 add_missing_dev(root
, devid
, uuid
);
5134 if (!map
->stripes
[i
].dev
) {
5136 free_extent_map(em
);
5140 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
5143 write_lock(&map_tree
->map_tree
.lock
);
5144 ret
= add_extent_mapping(&map_tree
->map_tree
, em
);
5145 write_unlock(&map_tree
->map_tree
.lock
);
5146 BUG_ON(ret
); /* Tree corruption */
5147 free_extent_map(em
);
5152 static void fill_device_from_item(struct extent_buffer
*leaf
,
5153 struct btrfs_dev_item
*dev_item
,
5154 struct btrfs_device
*device
)
5158 device
->devid
= btrfs_device_id(leaf
, dev_item
);
5159 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
5160 device
->total_bytes
= device
->disk_total_bytes
;
5161 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
5162 device
->type
= btrfs_device_type(leaf
, dev_item
);
5163 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
5164 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
5165 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
5166 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
5167 device
->is_tgtdev_for_dev_replace
= 0;
5169 ptr
= (unsigned long)btrfs_device_uuid(dev_item
);
5170 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
5173 static int open_seed_devices(struct btrfs_root
*root
, u8
*fsid
)
5175 struct btrfs_fs_devices
*fs_devices
;
5178 BUG_ON(!mutex_is_locked(&uuid_mutex
));
5180 fs_devices
= root
->fs_info
->fs_devices
->seed
;
5181 while (fs_devices
) {
5182 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
5186 fs_devices
= fs_devices
->seed
;
5189 fs_devices
= find_fsid(fsid
);
5195 fs_devices
= clone_fs_devices(fs_devices
);
5196 if (IS_ERR(fs_devices
)) {
5197 ret
= PTR_ERR(fs_devices
);
5201 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
5202 root
->fs_info
->bdev_holder
);
5204 free_fs_devices(fs_devices
);
5208 if (!fs_devices
->seeding
) {
5209 __btrfs_close_devices(fs_devices
);
5210 free_fs_devices(fs_devices
);
5215 fs_devices
->seed
= root
->fs_info
->fs_devices
->seed
;
5216 root
->fs_info
->fs_devices
->seed
= fs_devices
;
5221 static int read_one_dev(struct btrfs_root
*root
,
5222 struct extent_buffer
*leaf
,
5223 struct btrfs_dev_item
*dev_item
)
5225 struct btrfs_device
*device
;
5228 u8 fs_uuid
[BTRFS_UUID_SIZE
];
5229 u8 dev_uuid
[BTRFS_UUID_SIZE
];
5231 devid
= btrfs_device_id(leaf
, dev_item
);
5232 read_extent_buffer(leaf
, dev_uuid
,
5233 (unsigned long)btrfs_device_uuid(dev_item
),
5235 read_extent_buffer(leaf
, fs_uuid
,
5236 (unsigned long)btrfs_device_fsid(dev_item
),
5239 if (memcmp(fs_uuid
, root
->fs_info
->fsid
, BTRFS_UUID_SIZE
)) {
5240 ret
= open_seed_devices(root
, fs_uuid
);
5241 if (ret
&& !btrfs_test_opt(root
, DEGRADED
))
5245 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
, fs_uuid
);
5246 if (!device
|| !device
->bdev
) {
5247 if (!btrfs_test_opt(root
, DEGRADED
))
5251 printk(KERN_WARNING
"warning devid %llu missing\n",
5252 (unsigned long long)devid
);
5253 device
= add_missing_dev(root
, devid
, dev_uuid
);
5256 } else if (!device
->missing
) {
5258 * this happens when a device that was properly setup
5259 * in the device info lists suddenly goes bad.
5260 * device->bdev is NULL, and so we have to set
5261 * device->missing to one here
5263 root
->fs_info
->fs_devices
->missing_devices
++;
5264 device
->missing
= 1;
5268 if (device
->fs_devices
!= root
->fs_info
->fs_devices
) {
5269 BUG_ON(device
->writeable
);
5270 if (device
->generation
!=
5271 btrfs_device_generation(leaf
, dev_item
))
5275 fill_device_from_item(leaf
, dev_item
, device
);
5276 device
->dev_root
= root
->fs_info
->dev_root
;
5277 device
->in_fs_metadata
= 1;
5278 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
5279 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
5280 spin_lock(&root
->fs_info
->free_chunk_lock
);
5281 root
->fs_info
->free_chunk_space
+= device
->total_bytes
-
5283 spin_unlock(&root
->fs_info
->free_chunk_lock
);
5289 int btrfs_read_sys_array(struct btrfs_root
*root
)
5291 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
5292 struct extent_buffer
*sb
;
5293 struct btrfs_disk_key
*disk_key
;
5294 struct btrfs_chunk
*chunk
;
5296 unsigned long sb_ptr
;
5302 struct btrfs_key key
;
5304 sb
= btrfs_find_create_tree_block(root
, BTRFS_SUPER_INFO_OFFSET
,
5305 BTRFS_SUPER_INFO_SIZE
);
5308 btrfs_set_buffer_uptodate(sb
);
5309 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
5311 * The sb extent buffer is artifical and just used to read the system array.
5312 * btrfs_set_buffer_uptodate() call does not properly mark all it's
5313 * pages up-to-date when the page is larger: extent does not cover the
5314 * whole page and consequently check_page_uptodate does not find all
5315 * the page's extents up-to-date (the hole beyond sb),
5316 * write_extent_buffer then triggers a WARN_ON.
5318 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5319 * but sb spans only this function. Add an explicit SetPageUptodate call
5320 * to silence the warning eg. on PowerPC 64.
5322 if (PAGE_CACHE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
5323 SetPageUptodate(sb
->pages
[0]);
5325 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
5326 array_size
= btrfs_super_sys_array_size(super_copy
);
5328 ptr
= super_copy
->sys_chunk_array
;
5329 sb_ptr
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
5332 while (cur
< array_size
) {
5333 disk_key
= (struct btrfs_disk_key
*)ptr
;
5334 btrfs_disk_key_to_cpu(&key
, disk_key
);
5336 len
= sizeof(*disk_key
); ptr
+= len
;
5340 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
5341 chunk
= (struct btrfs_chunk
*)sb_ptr
;
5342 ret
= read_one_chunk(root
, &key
, sb
, chunk
);
5345 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
5346 len
= btrfs_chunk_item_size(num_stripes
);
5355 free_extent_buffer(sb
);
5359 int btrfs_read_chunk_tree(struct btrfs_root
*root
)
5361 struct btrfs_path
*path
;
5362 struct extent_buffer
*leaf
;
5363 struct btrfs_key key
;
5364 struct btrfs_key found_key
;
5368 root
= root
->fs_info
->chunk_root
;
5370 path
= btrfs_alloc_path();
5374 mutex_lock(&uuid_mutex
);
5377 /* first we search for all of the device items, and then we
5378 * read in all of the chunk items. This way we can create chunk
5379 * mappings that reference all of the devices that are afound
5381 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
5385 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5389 leaf
= path
->nodes
[0];
5390 slot
= path
->slots
[0];
5391 if (slot
>= btrfs_header_nritems(leaf
)) {
5392 ret
= btrfs_next_leaf(root
, path
);
5399 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
5400 if (key
.objectid
== BTRFS_DEV_ITEMS_OBJECTID
) {
5401 if (found_key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
)
5403 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
5404 struct btrfs_dev_item
*dev_item
;
5405 dev_item
= btrfs_item_ptr(leaf
, slot
,
5406 struct btrfs_dev_item
);
5407 ret
= read_one_dev(root
, leaf
, dev_item
);
5411 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
5412 struct btrfs_chunk
*chunk
;
5413 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
5414 ret
= read_one_chunk(root
, &found_key
, leaf
, chunk
);
5420 if (key
.objectid
== BTRFS_DEV_ITEMS_OBJECTID
) {
5422 btrfs_release_path(path
);
5427 unlock_chunks(root
);
5428 mutex_unlock(&uuid_mutex
);
5430 btrfs_free_path(path
);
5434 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
5438 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
5439 btrfs_dev_stat_reset(dev
, i
);
5442 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
5444 struct btrfs_key key
;
5445 struct btrfs_key found_key
;
5446 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
5447 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
5448 struct extent_buffer
*eb
;
5451 struct btrfs_device
*device
;
5452 struct btrfs_path
*path
= NULL
;
5455 path
= btrfs_alloc_path();
5461 mutex_lock(&fs_devices
->device_list_mutex
);
5462 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
5464 struct btrfs_dev_stats_item
*ptr
;
5467 key
.type
= BTRFS_DEV_STATS_KEY
;
5468 key
.offset
= device
->devid
;
5469 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
5471 __btrfs_reset_dev_stats(device
);
5472 device
->dev_stats_valid
= 1;
5473 btrfs_release_path(path
);
5476 slot
= path
->slots
[0];
5477 eb
= path
->nodes
[0];
5478 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
5479 item_size
= btrfs_item_size_nr(eb
, slot
);
5481 ptr
= btrfs_item_ptr(eb
, slot
,
5482 struct btrfs_dev_stats_item
);
5484 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
5485 if (item_size
>= (1 + i
) * sizeof(__le64
))
5486 btrfs_dev_stat_set(device
, i
,
5487 btrfs_dev_stats_value(eb
, ptr
, i
));
5489 btrfs_dev_stat_reset(device
, i
);
5492 device
->dev_stats_valid
= 1;
5493 btrfs_dev_stat_print_on_load(device
);
5494 btrfs_release_path(path
);
5496 mutex_unlock(&fs_devices
->device_list_mutex
);
5499 btrfs_free_path(path
);
5500 return ret
< 0 ? ret
: 0;
5503 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
5504 struct btrfs_root
*dev_root
,
5505 struct btrfs_device
*device
)
5507 struct btrfs_path
*path
;
5508 struct btrfs_key key
;
5509 struct extent_buffer
*eb
;
5510 struct btrfs_dev_stats_item
*ptr
;
5515 key
.type
= BTRFS_DEV_STATS_KEY
;
5516 key
.offset
= device
->devid
;
5518 path
= btrfs_alloc_path();
5520 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
5522 printk_in_rcu(KERN_WARNING
"btrfs: error %d while searching for dev_stats item for device %s!\n",
5523 ret
, rcu_str_deref(device
->name
));
5528 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
5529 /* need to delete old one and insert a new one */
5530 ret
= btrfs_del_item(trans
, dev_root
, path
);
5532 printk_in_rcu(KERN_WARNING
"btrfs: delete too small dev_stats item for device %s failed %d!\n",
5533 rcu_str_deref(device
->name
), ret
);
5540 /* need to insert a new item */
5541 btrfs_release_path(path
);
5542 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
5543 &key
, sizeof(*ptr
));
5545 printk_in_rcu(KERN_WARNING
"btrfs: insert dev_stats item for device %s failed %d!\n",
5546 rcu_str_deref(device
->name
), ret
);
5551 eb
= path
->nodes
[0];
5552 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
5553 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
5554 btrfs_set_dev_stats_value(eb
, ptr
, i
,
5555 btrfs_dev_stat_read(device
, i
));
5556 btrfs_mark_buffer_dirty(eb
);
5559 btrfs_free_path(path
);
5564 * called from commit_transaction. Writes all changed device stats to disk.
5566 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
5567 struct btrfs_fs_info
*fs_info
)
5569 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
5570 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
5571 struct btrfs_device
*device
;
5574 mutex_lock(&fs_devices
->device_list_mutex
);
5575 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
5576 if (!device
->dev_stats_valid
|| !device
->dev_stats_dirty
)
5579 ret
= update_dev_stat_item(trans
, dev_root
, device
);
5581 device
->dev_stats_dirty
= 0;
5583 mutex_unlock(&fs_devices
->device_list_mutex
);
5588 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
5590 btrfs_dev_stat_inc(dev
, index
);
5591 btrfs_dev_stat_print_on_error(dev
);
5594 void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
5596 if (!dev
->dev_stats_valid
)
5598 printk_ratelimited_in_rcu(KERN_ERR
5599 "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5600 rcu_str_deref(dev
->name
),
5601 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
5602 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
5603 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
5604 btrfs_dev_stat_read(dev
,
5605 BTRFS_DEV_STAT_CORRUPTION_ERRS
),
5606 btrfs_dev_stat_read(dev
,
5607 BTRFS_DEV_STAT_GENERATION_ERRS
));
5610 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
5614 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
5615 if (btrfs_dev_stat_read(dev
, i
) != 0)
5617 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
5618 return; /* all values == 0, suppress message */
5620 printk_in_rcu(KERN_INFO
"btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5621 rcu_str_deref(dev
->name
),
5622 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
5623 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
5624 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
5625 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
5626 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
5629 int btrfs_get_dev_stats(struct btrfs_root
*root
,
5630 struct btrfs_ioctl_get_dev_stats
*stats
)
5632 struct btrfs_device
*dev
;
5633 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
5636 mutex_lock(&fs_devices
->device_list_mutex
);
5637 dev
= btrfs_find_device(root
->fs_info
, stats
->devid
, NULL
, NULL
);
5638 mutex_unlock(&fs_devices
->device_list_mutex
);
5642 "btrfs: get dev_stats failed, device not found\n");
5644 } else if (!dev
->dev_stats_valid
) {
5646 "btrfs: get dev_stats failed, not yet valid\n");
5648 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
5649 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
5650 if (stats
->nr_items
> i
)
5652 btrfs_dev_stat_read_and_reset(dev
, i
);
5654 btrfs_dev_stat_reset(dev
, i
);
5657 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
5658 if (stats
->nr_items
> i
)
5659 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
5661 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
5662 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
5666 int btrfs_scratch_superblock(struct btrfs_device
*device
)
5668 struct buffer_head
*bh
;
5669 struct btrfs_super_block
*disk_super
;
5671 bh
= btrfs_read_dev_super(device
->bdev
);
5674 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
5676 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
5677 set_buffer_dirty(bh
);
5678 sync_dirty_buffer(bh
);