2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/ratelimit.h>
27 #include <linux/kthread.h>
28 #include <linux/raid/pq.h>
29 #include <linux/semaphore.h>
30 #include <asm/div64.h>
32 #include "extent_map.h"
34 #include "transaction.h"
35 #include "print-tree.h"
38 #include "async-thread.h"
39 #include "check-integrity.h"
40 #include "rcu-string.h"
42 #include "dev-replace.h"
44 static int init_first_rw_device(struct btrfs_trans_handle
*trans
,
45 struct btrfs_root
*root
,
46 struct btrfs_device
*device
);
47 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
);
48 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
);
49 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
);
50 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*device
);
52 static DEFINE_MUTEX(uuid_mutex
);
53 static LIST_HEAD(fs_uuids
);
55 static void lock_chunks(struct btrfs_root
*root
)
57 mutex_lock(&root
->fs_info
->chunk_mutex
);
60 static void unlock_chunks(struct btrfs_root
*root
)
62 mutex_unlock(&root
->fs_info
->chunk_mutex
);
65 static struct btrfs_fs_devices
*__alloc_fs_devices(void)
67 struct btrfs_fs_devices
*fs_devs
;
69 fs_devs
= kzalloc(sizeof(*fs_devs
), GFP_NOFS
);
71 return ERR_PTR(-ENOMEM
);
73 mutex_init(&fs_devs
->device_list_mutex
);
75 INIT_LIST_HEAD(&fs_devs
->devices
);
76 INIT_LIST_HEAD(&fs_devs
->alloc_list
);
77 INIT_LIST_HEAD(&fs_devs
->list
);
83 * alloc_fs_devices - allocate struct btrfs_fs_devices
84 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
87 * Return: a pointer to a new &struct btrfs_fs_devices on success;
88 * ERR_PTR() on error. Returned struct is not linked onto any lists and
89 * can be destroyed with kfree() right away.
91 static struct btrfs_fs_devices
*alloc_fs_devices(const u8
*fsid
)
93 struct btrfs_fs_devices
*fs_devs
;
95 fs_devs
= __alloc_fs_devices();
100 memcpy(fs_devs
->fsid
, fsid
, BTRFS_FSID_SIZE
);
102 generate_random_uuid(fs_devs
->fsid
);
107 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
109 struct btrfs_device
*device
;
110 WARN_ON(fs_devices
->opened
);
111 while (!list_empty(&fs_devices
->devices
)) {
112 device
= list_entry(fs_devices
->devices
.next
,
113 struct btrfs_device
, dev_list
);
114 list_del(&device
->dev_list
);
115 rcu_string_free(device
->name
);
121 static void btrfs_kobject_uevent(struct block_device
*bdev
,
122 enum kobject_action action
)
126 ret
= kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, action
);
128 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
130 kobject_name(&disk_to_dev(bdev
->bd_disk
)->kobj
),
131 &disk_to_dev(bdev
->bd_disk
)->kobj
);
134 void btrfs_cleanup_fs_uuids(void)
136 struct btrfs_fs_devices
*fs_devices
;
138 while (!list_empty(&fs_uuids
)) {
139 fs_devices
= list_entry(fs_uuids
.next
,
140 struct btrfs_fs_devices
, list
);
141 list_del(&fs_devices
->list
);
142 free_fs_devices(fs_devices
);
146 static struct btrfs_device
*__alloc_device(void)
148 struct btrfs_device
*dev
;
150 dev
= kzalloc(sizeof(*dev
), GFP_NOFS
);
152 return ERR_PTR(-ENOMEM
);
154 INIT_LIST_HEAD(&dev
->dev_list
);
155 INIT_LIST_HEAD(&dev
->dev_alloc_list
);
157 spin_lock_init(&dev
->io_lock
);
159 spin_lock_init(&dev
->reada_lock
);
160 atomic_set(&dev
->reada_in_flight
, 0);
161 INIT_RADIX_TREE(&dev
->reada_zones
, GFP_NOFS
& ~__GFP_WAIT
);
162 INIT_RADIX_TREE(&dev
->reada_extents
, GFP_NOFS
& ~__GFP_WAIT
);
167 static noinline
struct btrfs_device
*__find_device(struct list_head
*head
,
170 struct btrfs_device
*dev
;
172 list_for_each_entry(dev
, head
, dev_list
) {
173 if (dev
->devid
== devid
&&
174 (!uuid
|| !memcmp(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
))) {
181 static noinline
struct btrfs_fs_devices
*find_fsid(u8
*fsid
)
183 struct btrfs_fs_devices
*fs_devices
;
185 list_for_each_entry(fs_devices
, &fs_uuids
, list
) {
186 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
193 btrfs_get_bdev_and_sb(const char *device_path
, fmode_t flags
, void *holder
,
194 int flush
, struct block_device
**bdev
,
195 struct buffer_head
**bh
)
199 *bdev
= blkdev_get_by_path(device_path
, flags
, holder
);
202 ret
= PTR_ERR(*bdev
);
203 printk(KERN_INFO
"BTRFS: open %s failed\n", device_path
);
208 filemap_write_and_wait((*bdev
)->bd_inode
->i_mapping
);
209 ret
= set_blocksize(*bdev
, 4096);
211 blkdev_put(*bdev
, flags
);
214 invalidate_bdev(*bdev
);
215 *bh
= btrfs_read_dev_super(*bdev
);
218 blkdev_put(*bdev
, flags
);
230 static void requeue_list(struct btrfs_pending_bios
*pending_bios
,
231 struct bio
*head
, struct bio
*tail
)
234 struct bio
*old_head
;
236 old_head
= pending_bios
->head
;
237 pending_bios
->head
= head
;
238 if (pending_bios
->tail
)
239 tail
->bi_next
= old_head
;
241 pending_bios
->tail
= tail
;
245 * we try to collect pending bios for a device so we don't get a large
246 * number of procs sending bios down to the same device. This greatly
247 * improves the schedulers ability to collect and merge the bios.
249 * But, it also turns into a long list of bios to process and that is sure
250 * to eventually make the worker thread block. The solution here is to
251 * make some progress and then put this work struct back at the end of
252 * the list if the block device is congested. This way, multiple devices
253 * can make progress from a single worker thread.
255 static noinline
void run_scheduled_bios(struct btrfs_device
*device
)
258 struct backing_dev_info
*bdi
;
259 struct btrfs_fs_info
*fs_info
;
260 struct btrfs_pending_bios
*pending_bios
;
264 unsigned long num_run
;
265 unsigned long batch_run
= 0;
267 unsigned long last_waited
= 0;
269 int sync_pending
= 0;
270 struct blk_plug plug
;
273 * this function runs all the bios we've collected for
274 * a particular device. We don't want to wander off to
275 * another device without first sending all of these down.
276 * So, setup a plug here and finish it off before we return
278 blk_start_plug(&plug
);
280 bdi
= blk_get_backing_dev_info(device
->bdev
);
281 fs_info
= device
->dev_root
->fs_info
;
282 limit
= btrfs_async_submit_limit(fs_info
);
283 limit
= limit
* 2 / 3;
286 spin_lock(&device
->io_lock
);
291 /* take all the bios off the list at once and process them
292 * later on (without the lock held). But, remember the
293 * tail and other pointers so the bios can be properly reinserted
294 * into the list if we hit congestion
296 if (!force_reg
&& device
->pending_sync_bios
.head
) {
297 pending_bios
= &device
->pending_sync_bios
;
300 pending_bios
= &device
->pending_bios
;
304 pending
= pending_bios
->head
;
305 tail
= pending_bios
->tail
;
306 WARN_ON(pending
&& !tail
);
309 * if pending was null this time around, no bios need processing
310 * at all and we can stop. Otherwise it'll loop back up again
311 * and do an additional check so no bios are missed.
313 * device->running_pending is used to synchronize with the
316 if (device
->pending_sync_bios
.head
== NULL
&&
317 device
->pending_bios
.head
== NULL
) {
319 device
->running_pending
= 0;
322 device
->running_pending
= 1;
325 pending_bios
->head
= NULL
;
326 pending_bios
->tail
= NULL
;
328 spin_unlock(&device
->io_lock
);
333 /* we want to work on both lists, but do more bios on the
334 * sync list than the regular list
337 pending_bios
!= &device
->pending_sync_bios
&&
338 device
->pending_sync_bios
.head
) ||
339 (num_run
> 64 && pending_bios
== &device
->pending_sync_bios
&&
340 device
->pending_bios
.head
)) {
341 spin_lock(&device
->io_lock
);
342 requeue_list(pending_bios
, pending
, tail
);
347 pending
= pending
->bi_next
;
350 if (atomic_dec_return(&fs_info
->nr_async_bios
) < limit
&&
351 waitqueue_active(&fs_info
->async_submit_wait
))
352 wake_up(&fs_info
->async_submit_wait
);
354 BUG_ON(atomic_read(&cur
->bi_cnt
) == 0);
357 * if we're doing the sync list, record that our
358 * plug has some sync requests on it
360 * If we're doing the regular list and there are
361 * sync requests sitting around, unplug before
364 if (pending_bios
== &device
->pending_sync_bios
) {
366 } else if (sync_pending
) {
367 blk_finish_plug(&plug
);
368 blk_start_plug(&plug
);
372 btrfsic_submit_bio(cur
->bi_rw
, cur
);
379 * we made progress, there is more work to do and the bdi
380 * is now congested. Back off and let other work structs
383 if (pending
&& bdi_write_congested(bdi
) && batch_run
> 8 &&
384 fs_info
->fs_devices
->open_devices
> 1) {
385 struct io_context
*ioc
;
387 ioc
= current
->io_context
;
390 * the main goal here is that we don't want to
391 * block if we're going to be able to submit
392 * more requests without blocking.
394 * This code does two great things, it pokes into
395 * the elevator code from a filesystem _and_
396 * it makes assumptions about how batching works.
398 if (ioc
&& ioc
->nr_batch_requests
> 0 &&
399 time_before(jiffies
, ioc
->last_waited
+ HZ
/50UL) &&
401 ioc
->last_waited
== last_waited
)) {
403 * we want to go through our batch of
404 * requests and stop. So, we copy out
405 * the ioc->last_waited time and test
406 * against it before looping
408 last_waited
= ioc
->last_waited
;
413 spin_lock(&device
->io_lock
);
414 requeue_list(pending_bios
, pending
, tail
);
415 device
->running_pending
= 1;
417 spin_unlock(&device
->io_lock
);
418 btrfs_queue_work(fs_info
->submit_workers
,
422 /* unplug every 64 requests just for good measure */
423 if (batch_run
% 64 == 0) {
424 blk_finish_plug(&plug
);
425 blk_start_plug(&plug
);
434 spin_lock(&device
->io_lock
);
435 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
437 spin_unlock(&device
->io_lock
);
440 blk_finish_plug(&plug
);
443 static void pending_bios_fn(struct btrfs_work
*work
)
445 struct btrfs_device
*device
;
447 device
= container_of(work
, struct btrfs_device
, work
);
448 run_scheduled_bios(device
);
452 * Add new device to list of registered devices
455 * 1 - first time device is seen
456 * 0 - device already known
459 static noinline
int device_list_add(const char *path
,
460 struct btrfs_super_block
*disk_super
,
461 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
463 struct btrfs_device
*device
;
464 struct btrfs_fs_devices
*fs_devices
;
465 struct rcu_string
*name
;
467 u64 found_transid
= btrfs_super_generation(disk_super
);
469 fs_devices
= find_fsid(disk_super
->fsid
);
471 fs_devices
= alloc_fs_devices(disk_super
->fsid
);
472 if (IS_ERR(fs_devices
))
473 return PTR_ERR(fs_devices
);
475 list_add(&fs_devices
->list
, &fs_uuids
);
476 fs_devices
->latest_devid
= devid
;
477 fs_devices
->latest_trans
= found_transid
;
481 device
= __find_device(&fs_devices
->devices
, devid
,
482 disk_super
->dev_item
.uuid
);
485 if (fs_devices
->opened
)
488 device
= btrfs_alloc_device(NULL
, &devid
,
489 disk_super
->dev_item
.uuid
);
490 if (IS_ERR(device
)) {
491 /* we can safely leave the fs_devices entry around */
492 return PTR_ERR(device
);
495 name
= rcu_string_strdup(path
, GFP_NOFS
);
500 rcu_assign_pointer(device
->name
, name
);
502 mutex_lock(&fs_devices
->device_list_mutex
);
503 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
504 fs_devices
->num_devices
++;
505 mutex_unlock(&fs_devices
->device_list_mutex
);
508 device
->fs_devices
= fs_devices
;
509 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
510 name
= rcu_string_strdup(path
, GFP_NOFS
);
513 rcu_string_free(device
->name
);
514 rcu_assign_pointer(device
->name
, name
);
515 if (device
->missing
) {
516 fs_devices
->missing_devices
--;
521 if (found_transid
> fs_devices
->latest_trans
) {
522 fs_devices
->latest_devid
= devid
;
523 fs_devices
->latest_trans
= found_transid
;
525 *fs_devices_ret
= fs_devices
;
530 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
532 struct btrfs_fs_devices
*fs_devices
;
533 struct btrfs_device
*device
;
534 struct btrfs_device
*orig_dev
;
536 fs_devices
= alloc_fs_devices(orig
->fsid
);
537 if (IS_ERR(fs_devices
))
540 fs_devices
->latest_devid
= orig
->latest_devid
;
541 fs_devices
->latest_trans
= orig
->latest_trans
;
542 fs_devices
->total_devices
= orig
->total_devices
;
544 /* We have held the volume lock, it is safe to get the devices. */
545 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
546 struct rcu_string
*name
;
548 device
= btrfs_alloc_device(NULL
, &orig_dev
->devid
,
554 * This is ok to do without rcu read locked because we hold the
555 * uuid mutex so nothing we touch in here is going to disappear.
557 name
= rcu_string_strdup(orig_dev
->name
->str
, GFP_NOFS
);
562 rcu_assign_pointer(device
->name
, name
);
564 list_add(&device
->dev_list
, &fs_devices
->devices
);
565 device
->fs_devices
= fs_devices
;
566 fs_devices
->num_devices
++;
570 free_fs_devices(fs_devices
);
571 return ERR_PTR(-ENOMEM
);
574 void btrfs_close_extra_devices(struct btrfs_fs_info
*fs_info
,
575 struct btrfs_fs_devices
*fs_devices
, int step
)
577 struct btrfs_device
*device
, *next
;
579 struct block_device
*latest_bdev
= NULL
;
580 u64 latest_devid
= 0;
581 u64 latest_transid
= 0;
583 mutex_lock(&uuid_mutex
);
585 /* This is the initialized path, it is safe to release the devices. */
586 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
587 if (device
->in_fs_metadata
) {
588 if (!device
->is_tgtdev_for_dev_replace
&&
590 device
->generation
> latest_transid
)) {
591 latest_devid
= device
->devid
;
592 latest_transid
= device
->generation
;
593 latest_bdev
= device
->bdev
;
598 if (device
->devid
== BTRFS_DEV_REPLACE_DEVID
) {
600 * In the first step, keep the device which has
601 * the correct fsid and the devid that is used
602 * for the dev_replace procedure.
603 * In the second step, the dev_replace state is
604 * read from the device tree and it is known
605 * whether the procedure is really active or
606 * not, which means whether this device is
607 * used or whether it should be removed.
609 if (step
== 0 || device
->is_tgtdev_for_dev_replace
) {
614 blkdev_put(device
->bdev
, device
->mode
);
616 fs_devices
->open_devices
--;
618 if (device
->writeable
) {
619 list_del_init(&device
->dev_alloc_list
);
620 device
->writeable
= 0;
621 if (!device
->is_tgtdev_for_dev_replace
)
622 fs_devices
->rw_devices
--;
624 list_del_init(&device
->dev_list
);
625 fs_devices
->num_devices
--;
626 rcu_string_free(device
->name
);
630 if (fs_devices
->seed
) {
631 fs_devices
= fs_devices
->seed
;
635 fs_devices
->latest_bdev
= latest_bdev
;
636 fs_devices
->latest_devid
= latest_devid
;
637 fs_devices
->latest_trans
= latest_transid
;
639 mutex_unlock(&uuid_mutex
);
642 static void __free_device(struct work_struct
*work
)
644 struct btrfs_device
*device
;
646 device
= container_of(work
, struct btrfs_device
, rcu_work
);
649 blkdev_put(device
->bdev
, device
->mode
);
651 rcu_string_free(device
->name
);
655 static void free_device(struct rcu_head
*head
)
657 struct btrfs_device
*device
;
659 device
= container_of(head
, struct btrfs_device
, rcu
);
661 INIT_WORK(&device
->rcu_work
, __free_device
);
662 schedule_work(&device
->rcu_work
);
665 static int __btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
667 struct btrfs_device
*device
;
669 if (--fs_devices
->opened
> 0)
672 mutex_lock(&fs_devices
->device_list_mutex
);
673 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
674 struct btrfs_device
*new_device
;
675 struct rcu_string
*name
;
678 fs_devices
->open_devices
--;
680 if (device
->writeable
&&
681 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
682 list_del_init(&device
->dev_alloc_list
);
683 fs_devices
->rw_devices
--;
686 if (device
->can_discard
)
687 fs_devices
->num_can_discard
--;
689 fs_devices
->missing_devices
--;
691 new_device
= btrfs_alloc_device(NULL
, &device
->devid
,
693 BUG_ON(IS_ERR(new_device
)); /* -ENOMEM */
695 /* Safe because we are under uuid_mutex */
697 name
= rcu_string_strdup(device
->name
->str
, GFP_NOFS
);
698 BUG_ON(!name
); /* -ENOMEM */
699 rcu_assign_pointer(new_device
->name
, name
);
702 list_replace_rcu(&device
->dev_list
, &new_device
->dev_list
);
703 new_device
->fs_devices
= device
->fs_devices
;
705 call_rcu(&device
->rcu
, free_device
);
707 mutex_unlock(&fs_devices
->device_list_mutex
);
709 WARN_ON(fs_devices
->open_devices
);
710 WARN_ON(fs_devices
->rw_devices
);
711 fs_devices
->opened
= 0;
712 fs_devices
->seeding
= 0;
717 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
719 struct btrfs_fs_devices
*seed_devices
= NULL
;
722 mutex_lock(&uuid_mutex
);
723 ret
= __btrfs_close_devices(fs_devices
);
724 if (!fs_devices
->opened
) {
725 seed_devices
= fs_devices
->seed
;
726 fs_devices
->seed
= NULL
;
728 mutex_unlock(&uuid_mutex
);
730 while (seed_devices
) {
731 fs_devices
= seed_devices
;
732 seed_devices
= fs_devices
->seed
;
733 __btrfs_close_devices(fs_devices
);
734 free_fs_devices(fs_devices
);
737 * Wait for rcu kworkers under __btrfs_close_devices
738 * to finish all blkdev_puts so device is really
739 * free when umount is done.
745 static int __btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
746 fmode_t flags
, void *holder
)
748 struct request_queue
*q
;
749 struct block_device
*bdev
;
750 struct list_head
*head
= &fs_devices
->devices
;
751 struct btrfs_device
*device
;
752 struct block_device
*latest_bdev
= NULL
;
753 struct buffer_head
*bh
;
754 struct btrfs_super_block
*disk_super
;
755 u64 latest_devid
= 0;
756 u64 latest_transid
= 0;
763 list_for_each_entry(device
, head
, dev_list
) {
769 /* Just open everything we can; ignore failures here */
770 if (btrfs_get_bdev_and_sb(device
->name
->str
, flags
, holder
, 1,
774 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
775 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
776 if (devid
!= device
->devid
)
779 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
,
783 device
->generation
= btrfs_super_generation(disk_super
);
784 if (!latest_transid
|| device
->generation
> latest_transid
) {
785 latest_devid
= devid
;
786 latest_transid
= device
->generation
;
790 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
791 device
->writeable
= 0;
793 device
->writeable
= !bdev_read_only(bdev
);
797 q
= bdev_get_queue(bdev
);
798 if (blk_queue_discard(q
)) {
799 device
->can_discard
= 1;
800 fs_devices
->num_can_discard
++;
804 device
->in_fs_metadata
= 0;
805 device
->mode
= flags
;
807 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
808 fs_devices
->rotating
= 1;
810 fs_devices
->open_devices
++;
811 if (device
->writeable
&&
812 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
813 fs_devices
->rw_devices
++;
814 list_add(&device
->dev_alloc_list
,
815 &fs_devices
->alloc_list
);
822 blkdev_put(bdev
, flags
);
825 if (fs_devices
->open_devices
== 0) {
829 fs_devices
->seeding
= seeding
;
830 fs_devices
->opened
= 1;
831 fs_devices
->latest_bdev
= latest_bdev
;
832 fs_devices
->latest_devid
= latest_devid
;
833 fs_devices
->latest_trans
= latest_transid
;
834 fs_devices
->total_rw_bytes
= 0;
839 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
840 fmode_t flags
, void *holder
)
844 mutex_lock(&uuid_mutex
);
845 if (fs_devices
->opened
) {
846 fs_devices
->opened
++;
849 ret
= __btrfs_open_devices(fs_devices
, flags
, holder
);
851 mutex_unlock(&uuid_mutex
);
856 * Look for a btrfs signature on a device. This may be called out of the mount path
857 * and we are not allowed to call set_blocksize during the scan. The superblock
858 * is read via pagecache
860 int btrfs_scan_one_device(const char *path
, fmode_t flags
, void *holder
,
861 struct btrfs_fs_devices
**fs_devices_ret
)
863 struct btrfs_super_block
*disk_super
;
864 struct block_device
*bdev
;
875 * we would like to check all the supers, but that would make
876 * a btrfs mount succeed after a mkfs from a different FS.
877 * So, we need to add a special mount option to scan for
878 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
880 bytenr
= btrfs_sb_offset(0);
882 mutex_lock(&uuid_mutex
);
884 bdev
= blkdev_get_by_path(path
, flags
, holder
);
891 /* make sure our super fits in the device */
892 if (bytenr
+ PAGE_CACHE_SIZE
>= i_size_read(bdev
->bd_inode
))
895 /* make sure our super fits in the page */
896 if (sizeof(*disk_super
) > PAGE_CACHE_SIZE
)
899 /* make sure our super doesn't straddle pages on disk */
900 index
= bytenr
>> PAGE_CACHE_SHIFT
;
901 if ((bytenr
+ sizeof(*disk_super
) - 1) >> PAGE_CACHE_SHIFT
!= index
)
904 /* pull in the page with our super */
905 page
= read_cache_page_gfp(bdev
->bd_inode
->i_mapping
,
908 if (IS_ERR_OR_NULL(page
))
913 /* align our pointer to the offset of the super block */
914 disk_super
= p
+ (bytenr
& ~PAGE_CACHE_MASK
);
916 if (btrfs_super_bytenr(disk_super
) != bytenr
||
917 btrfs_super_magic(disk_super
) != BTRFS_MAGIC
)
920 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
921 transid
= btrfs_super_generation(disk_super
);
922 total_devices
= btrfs_super_num_devices(disk_super
);
924 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
926 if (disk_super
->label
[0]) {
927 if (disk_super
->label
[BTRFS_LABEL_SIZE
- 1])
928 disk_super
->label
[BTRFS_LABEL_SIZE
- 1] = '\0';
929 printk(KERN_INFO
"BTRFS: device label %s ", disk_super
->label
);
931 printk(KERN_INFO
"BTRFS: device fsid %pU ", disk_super
->fsid
);
934 printk(KERN_CONT
"devid %llu transid %llu %s\n", devid
, transid
, path
);
937 if (!ret
&& fs_devices_ret
)
938 (*fs_devices_ret
)->total_devices
= total_devices
;
942 page_cache_release(page
);
945 blkdev_put(bdev
, flags
);
947 mutex_unlock(&uuid_mutex
);
951 /* helper to account the used device space in the range */
952 int btrfs_account_dev_extents_size(struct btrfs_device
*device
, u64 start
,
953 u64 end
, u64
*length
)
955 struct btrfs_key key
;
956 struct btrfs_root
*root
= device
->dev_root
;
957 struct btrfs_dev_extent
*dev_extent
;
958 struct btrfs_path
*path
;
962 struct extent_buffer
*l
;
966 if (start
>= device
->total_bytes
|| device
->is_tgtdev_for_dev_replace
)
969 path
= btrfs_alloc_path();
974 key
.objectid
= device
->devid
;
976 key
.type
= BTRFS_DEV_EXTENT_KEY
;
978 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
982 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
989 slot
= path
->slots
[0];
990 if (slot
>= btrfs_header_nritems(l
)) {
991 ret
= btrfs_next_leaf(root
, path
);
999 btrfs_item_key_to_cpu(l
, &key
, slot
);
1001 if (key
.objectid
< device
->devid
)
1004 if (key
.objectid
> device
->devid
)
1007 if (btrfs_key_type(&key
) != BTRFS_DEV_EXTENT_KEY
)
1010 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1011 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1013 if (key
.offset
<= start
&& extent_end
> end
) {
1014 *length
= end
- start
+ 1;
1016 } else if (key
.offset
<= start
&& extent_end
> start
)
1017 *length
+= extent_end
- start
;
1018 else if (key
.offset
> start
&& extent_end
<= end
)
1019 *length
+= extent_end
- key
.offset
;
1020 else if (key
.offset
> start
&& key
.offset
<= end
) {
1021 *length
+= end
- key
.offset
+ 1;
1023 } else if (key
.offset
> end
)
1031 btrfs_free_path(path
);
1035 static int contains_pending_extent(struct btrfs_trans_handle
*trans
,
1036 struct btrfs_device
*device
,
1037 u64
*start
, u64 len
)
1039 struct extent_map
*em
;
1042 list_for_each_entry(em
, &trans
->transaction
->pending_chunks
, list
) {
1043 struct map_lookup
*map
;
1046 map
= (struct map_lookup
*)em
->bdev
;
1047 for (i
= 0; i
< map
->num_stripes
; i
++) {
1048 if (map
->stripes
[i
].dev
!= device
)
1050 if (map
->stripes
[i
].physical
>= *start
+ len
||
1051 map
->stripes
[i
].physical
+ em
->orig_block_len
<=
1054 *start
= map
->stripes
[i
].physical
+
1065 * find_free_dev_extent - find free space in the specified device
1066 * @device: the device which we search the free space in
1067 * @num_bytes: the size of the free space that we need
1068 * @start: store the start of the free space.
1069 * @len: the size of the free space. that we find, or the size of the max
1070 * free space if we don't find suitable free space
1072 * this uses a pretty simple search, the expectation is that it is
1073 * called very infrequently and that a given device has a small number
1076 * @start is used to store the start of the free space if we find. But if we
1077 * don't find suitable free space, it will be used to store the start position
1078 * of the max free space.
1080 * @len is used to store the size of the free space that we find.
1081 * But if we don't find suitable free space, it is used to store the size of
1082 * the max free space.
1084 int find_free_dev_extent(struct btrfs_trans_handle
*trans
,
1085 struct btrfs_device
*device
, u64 num_bytes
,
1086 u64
*start
, u64
*len
)
1088 struct btrfs_key key
;
1089 struct btrfs_root
*root
= device
->dev_root
;
1090 struct btrfs_dev_extent
*dev_extent
;
1091 struct btrfs_path
*path
;
1097 u64 search_end
= device
->total_bytes
;
1100 struct extent_buffer
*l
;
1102 /* FIXME use last free of some kind */
1104 /* we don't want to overwrite the superblock on the drive,
1105 * so we make sure to start at an offset of at least 1MB
1107 search_start
= max(root
->fs_info
->alloc_start
, 1024ull * 1024);
1109 path
= btrfs_alloc_path();
1113 max_hole_start
= search_start
;
1117 if (search_start
>= search_end
|| device
->is_tgtdev_for_dev_replace
) {
1123 path
->search_commit_root
= 1;
1124 path
->skip_locking
= 1;
1126 key
.objectid
= device
->devid
;
1127 key
.offset
= search_start
;
1128 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1130 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1134 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1141 slot
= path
->slots
[0];
1142 if (slot
>= btrfs_header_nritems(l
)) {
1143 ret
= btrfs_next_leaf(root
, path
);
1151 btrfs_item_key_to_cpu(l
, &key
, slot
);
1153 if (key
.objectid
< device
->devid
)
1156 if (key
.objectid
> device
->devid
)
1159 if (btrfs_key_type(&key
) != BTRFS_DEV_EXTENT_KEY
)
1162 if (key
.offset
> search_start
) {
1163 hole_size
= key
.offset
- search_start
;
1166 * Have to check before we set max_hole_start, otherwise
1167 * we could end up sending back this offset anyway.
1169 if (contains_pending_extent(trans
, device
,
1174 if (hole_size
> max_hole_size
) {
1175 max_hole_start
= search_start
;
1176 max_hole_size
= hole_size
;
1180 * If this free space is greater than which we need,
1181 * it must be the max free space that we have found
1182 * until now, so max_hole_start must point to the start
1183 * of this free space and the length of this free space
1184 * is stored in max_hole_size. Thus, we return
1185 * max_hole_start and max_hole_size and go back to the
1188 if (hole_size
>= num_bytes
) {
1194 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1195 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1197 if (extent_end
> search_start
)
1198 search_start
= extent_end
;
1205 * At this point, search_start should be the end of
1206 * allocated dev extents, and when shrinking the device,
1207 * search_end may be smaller than search_start.
1209 if (search_end
> search_start
)
1210 hole_size
= search_end
- search_start
;
1212 if (hole_size
> max_hole_size
) {
1213 max_hole_start
= search_start
;
1214 max_hole_size
= hole_size
;
1217 if (contains_pending_extent(trans
, device
, &search_start
, hole_size
)) {
1218 btrfs_release_path(path
);
1223 if (hole_size
< num_bytes
)
1229 btrfs_free_path(path
);
1230 *start
= max_hole_start
;
1232 *len
= max_hole_size
;
1236 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1237 struct btrfs_device
*device
,
1241 struct btrfs_path
*path
;
1242 struct btrfs_root
*root
= device
->dev_root
;
1243 struct btrfs_key key
;
1244 struct btrfs_key found_key
;
1245 struct extent_buffer
*leaf
= NULL
;
1246 struct btrfs_dev_extent
*extent
= NULL
;
1248 path
= btrfs_alloc_path();
1252 key
.objectid
= device
->devid
;
1254 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1256 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1258 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1259 BTRFS_DEV_EXTENT_KEY
);
1262 leaf
= path
->nodes
[0];
1263 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1264 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1265 struct btrfs_dev_extent
);
1266 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1267 btrfs_dev_extent_length(leaf
, extent
) < start
);
1269 btrfs_release_path(path
);
1271 } else if (ret
== 0) {
1272 leaf
= path
->nodes
[0];
1273 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1274 struct btrfs_dev_extent
);
1276 btrfs_error(root
->fs_info
, ret
, "Slot search failed");
1280 if (device
->bytes_used
> 0) {
1281 u64 len
= btrfs_dev_extent_length(leaf
, extent
);
1282 device
->bytes_used
-= len
;
1283 spin_lock(&root
->fs_info
->free_chunk_lock
);
1284 root
->fs_info
->free_chunk_space
+= len
;
1285 spin_unlock(&root
->fs_info
->free_chunk_lock
);
1287 ret
= btrfs_del_item(trans
, root
, path
);
1289 btrfs_error(root
->fs_info
, ret
,
1290 "Failed to remove dev extent item");
1293 btrfs_free_path(path
);
1297 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1298 struct btrfs_device
*device
,
1299 u64 chunk_tree
, u64 chunk_objectid
,
1300 u64 chunk_offset
, u64 start
, u64 num_bytes
)
1303 struct btrfs_path
*path
;
1304 struct btrfs_root
*root
= device
->dev_root
;
1305 struct btrfs_dev_extent
*extent
;
1306 struct extent_buffer
*leaf
;
1307 struct btrfs_key key
;
1309 WARN_ON(!device
->in_fs_metadata
);
1310 WARN_ON(device
->is_tgtdev_for_dev_replace
);
1311 path
= btrfs_alloc_path();
1315 key
.objectid
= device
->devid
;
1317 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1318 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1323 leaf
= path
->nodes
[0];
1324 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1325 struct btrfs_dev_extent
);
1326 btrfs_set_dev_extent_chunk_tree(leaf
, extent
, chunk_tree
);
1327 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
, chunk_objectid
);
1328 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1330 write_extent_buffer(leaf
, root
->fs_info
->chunk_tree_uuid
,
1331 btrfs_dev_extent_chunk_tree_uuid(extent
), BTRFS_UUID_SIZE
);
1333 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1334 btrfs_mark_buffer_dirty(leaf
);
1336 btrfs_free_path(path
);
1340 static u64
find_next_chunk(struct btrfs_fs_info
*fs_info
)
1342 struct extent_map_tree
*em_tree
;
1343 struct extent_map
*em
;
1347 em_tree
= &fs_info
->mapping_tree
.map_tree
;
1348 read_lock(&em_tree
->lock
);
1349 n
= rb_last(&em_tree
->map
);
1351 em
= rb_entry(n
, struct extent_map
, rb_node
);
1352 ret
= em
->start
+ em
->len
;
1354 read_unlock(&em_tree
->lock
);
1359 static noinline
int find_next_devid(struct btrfs_fs_info
*fs_info
,
1363 struct btrfs_key key
;
1364 struct btrfs_key found_key
;
1365 struct btrfs_path
*path
;
1367 path
= btrfs_alloc_path();
1371 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1372 key
.type
= BTRFS_DEV_ITEM_KEY
;
1373 key
.offset
= (u64
)-1;
1375 ret
= btrfs_search_slot(NULL
, fs_info
->chunk_root
, &key
, path
, 0, 0);
1379 BUG_ON(ret
== 0); /* Corruption */
1381 ret
= btrfs_previous_item(fs_info
->chunk_root
, path
,
1382 BTRFS_DEV_ITEMS_OBJECTID
,
1383 BTRFS_DEV_ITEM_KEY
);
1387 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1389 *devid_ret
= found_key
.offset
+ 1;
1393 btrfs_free_path(path
);
1398 * the device information is stored in the chunk root
1399 * the btrfs_device struct should be fully filled in
1401 static int btrfs_add_device(struct btrfs_trans_handle
*trans
,
1402 struct btrfs_root
*root
,
1403 struct btrfs_device
*device
)
1406 struct btrfs_path
*path
;
1407 struct btrfs_dev_item
*dev_item
;
1408 struct extent_buffer
*leaf
;
1409 struct btrfs_key key
;
1412 root
= root
->fs_info
->chunk_root
;
1414 path
= btrfs_alloc_path();
1418 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1419 key
.type
= BTRFS_DEV_ITEM_KEY
;
1420 key
.offset
= device
->devid
;
1422 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1427 leaf
= path
->nodes
[0];
1428 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1430 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1431 btrfs_set_device_generation(leaf
, dev_item
, 0);
1432 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1433 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1434 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1435 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1436 btrfs_set_device_total_bytes(leaf
, dev_item
, device
->total_bytes
);
1437 btrfs_set_device_bytes_used(leaf
, dev_item
, device
->bytes_used
);
1438 btrfs_set_device_group(leaf
, dev_item
, 0);
1439 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1440 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1441 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1443 ptr
= btrfs_device_uuid(dev_item
);
1444 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1445 ptr
= btrfs_device_fsid(dev_item
);
1446 write_extent_buffer(leaf
, root
->fs_info
->fsid
, ptr
, BTRFS_UUID_SIZE
);
1447 btrfs_mark_buffer_dirty(leaf
);
1451 btrfs_free_path(path
);
1455 static int btrfs_rm_dev_item(struct btrfs_root
*root
,
1456 struct btrfs_device
*device
)
1459 struct btrfs_path
*path
;
1460 struct btrfs_key key
;
1461 struct btrfs_trans_handle
*trans
;
1463 root
= root
->fs_info
->chunk_root
;
1465 path
= btrfs_alloc_path();
1469 trans
= btrfs_start_transaction(root
, 0);
1470 if (IS_ERR(trans
)) {
1471 btrfs_free_path(path
);
1472 return PTR_ERR(trans
);
1474 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1475 key
.type
= BTRFS_DEV_ITEM_KEY
;
1476 key
.offset
= device
->devid
;
1479 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1488 ret
= btrfs_del_item(trans
, root
, path
);
1492 btrfs_free_path(path
);
1493 unlock_chunks(root
);
1494 btrfs_commit_transaction(trans
, root
);
1498 int btrfs_rm_device(struct btrfs_root
*root
, char *device_path
)
1500 struct btrfs_device
*device
;
1501 struct btrfs_device
*next_device
;
1502 struct block_device
*bdev
;
1503 struct buffer_head
*bh
= NULL
;
1504 struct btrfs_super_block
*disk_super
;
1505 struct btrfs_fs_devices
*cur_devices
;
1512 bool clear_super
= false;
1514 mutex_lock(&uuid_mutex
);
1517 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
1519 all_avail
= root
->fs_info
->avail_data_alloc_bits
|
1520 root
->fs_info
->avail_system_alloc_bits
|
1521 root
->fs_info
->avail_metadata_alloc_bits
;
1522 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
1524 num_devices
= root
->fs_info
->fs_devices
->num_devices
;
1525 btrfs_dev_replace_lock(&root
->fs_info
->dev_replace
);
1526 if (btrfs_dev_replace_is_ongoing(&root
->fs_info
->dev_replace
)) {
1527 WARN_ON(num_devices
< 1);
1530 btrfs_dev_replace_unlock(&root
->fs_info
->dev_replace
);
1532 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID10
) && num_devices
<= 4) {
1533 ret
= BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET
;
1537 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID1
) && num_devices
<= 2) {
1538 ret
= BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET
;
1542 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID5
) &&
1543 root
->fs_info
->fs_devices
->rw_devices
<= 2) {
1544 ret
= BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET
;
1547 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID6
) &&
1548 root
->fs_info
->fs_devices
->rw_devices
<= 3) {
1549 ret
= BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET
;
1553 if (strcmp(device_path
, "missing") == 0) {
1554 struct list_head
*devices
;
1555 struct btrfs_device
*tmp
;
1558 devices
= &root
->fs_info
->fs_devices
->devices
;
1560 * It is safe to read the devices since the volume_mutex
1563 list_for_each_entry(tmp
, devices
, dev_list
) {
1564 if (tmp
->in_fs_metadata
&&
1565 !tmp
->is_tgtdev_for_dev_replace
&&
1575 ret
= BTRFS_ERROR_DEV_MISSING_NOT_FOUND
;
1579 ret
= btrfs_get_bdev_and_sb(device_path
,
1580 FMODE_WRITE
| FMODE_EXCL
,
1581 root
->fs_info
->bdev_holder
, 0,
1585 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1586 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1587 dev_uuid
= disk_super
->dev_item
.uuid
;
1588 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1596 if (device
->is_tgtdev_for_dev_replace
) {
1597 ret
= BTRFS_ERROR_DEV_TGT_REPLACE
;
1601 if (device
->writeable
&& root
->fs_info
->fs_devices
->rw_devices
== 1) {
1602 ret
= BTRFS_ERROR_DEV_ONLY_WRITABLE
;
1606 if (device
->writeable
) {
1608 list_del_init(&device
->dev_alloc_list
);
1609 unlock_chunks(root
);
1610 root
->fs_info
->fs_devices
->rw_devices
--;
1614 mutex_unlock(&uuid_mutex
);
1615 ret
= btrfs_shrink_device(device
, 0);
1616 mutex_lock(&uuid_mutex
);
1621 * TODO: the superblock still includes this device in its num_devices
1622 * counter although write_all_supers() is not locked out. This
1623 * could give a filesystem state which requires a degraded mount.
1625 ret
= btrfs_rm_dev_item(root
->fs_info
->chunk_root
, device
);
1629 spin_lock(&root
->fs_info
->free_chunk_lock
);
1630 root
->fs_info
->free_chunk_space
= device
->total_bytes
-
1632 spin_unlock(&root
->fs_info
->free_chunk_lock
);
1634 device
->in_fs_metadata
= 0;
1635 btrfs_scrub_cancel_dev(root
->fs_info
, device
);
1638 * the device list mutex makes sure that we don't change
1639 * the device list while someone else is writing out all
1640 * the device supers. Whoever is writing all supers, should
1641 * lock the device list mutex before getting the number of
1642 * devices in the super block (super_copy). Conversely,
1643 * whoever updates the number of devices in the super block
1644 * (super_copy) should hold the device list mutex.
1647 cur_devices
= device
->fs_devices
;
1648 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1649 list_del_rcu(&device
->dev_list
);
1651 device
->fs_devices
->num_devices
--;
1652 device
->fs_devices
->total_devices
--;
1654 if (device
->missing
)
1655 root
->fs_info
->fs_devices
->missing_devices
--;
1657 next_device
= list_entry(root
->fs_info
->fs_devices
->devices
.next
,
1658 struct btrfs_device
, dev_list
);
1659 if (device
->bdev
== root
->fs_info
->sb
->s_bdev
)
1660 root
->fs_info
->sb
->s_bdev
= next_device
->bdev
;
1661 if (device
->bdev
== root
->fs_info
->fs_devices
->latest_bdev
)
1662 root
->fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1665 device
->fs_devices
->open_devices
--;
1667 call_rcu(&device
->rcu
, free_device
);
1669 num_devices
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
1670 btrfs_set_super_num_devices(root
->fs_info
->super_copy
, num_devices
);
1671 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1673 if (cur_devices
->open_devices
== 0) {
1674 struct btrfs_fs_devices
*fs_devices
;
1675 fs_devices
= root
->fs_info
->fs_devices
;
1676 while (fs_devices
) {
1677 if (fs_devices
->seed
== cur_devices
)
1679 fs_devices
= fs_devices
->seed
;
1681 fs_devices
->seed
= cur_devices
->seed
;
1682 cur_devices
->seed
= NULL
;
1684 __btrfs_close_devices(cur_devices
);
1685 unlock_chunks(root
);
1686 free_fs_devices(cur_devices
);
1689 root
->fs_info
->num_tolerated_disk_barrier_failures
=
1690 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
1693 * at this point, the device is zero sized. We want to
1694 * remove it from the devices list and zero out the old super
1696 if (clear_super
&& disk_super
) {
1697 /* make sure this device isn't detected as part of
1700 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
1701 set_buffer_dirty(bh
);
1702 sync_dirty_buffer(bh
);
1707 /* Notify udev that device has changed */
1709 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
1714 blkdev_put(bdev
, FMODE_READ
| FMODE_EXCL
);
1716 mutex_unlock(&uuid_mutex
);
1719 if (device
->writeable
) {
1721 list_add(&device
->dev_alloc_list
,
1722 &root
->fs_info
->fs_devices
->alloc_list
);
1723 unlock_chunks(root
);
1724 root
->fs_info
->fs_devices
->rw_devices
++;
1729 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info
*fs_info
,
1730 struct btrfs_device
*srcdev
)
1732 WARN_ON(!mutex_is_locked(&fs_info
->fs_devices
->device_list_mutex
));
1734 list_del_rcu(&srcdev
->dev_list
);
1735 list_del_rcu(&srcdev
->dev_alloc_list
);
1736 fs_info
->fs_devices
->num_devices
--;
1737 if (srcdev
->missing
) {
1738 fs_info
->fs_devices
->missing_devices
--;
1739 fs_info
->fs_devices
->rw_devices
++;
1741 if (srcdev
->can_discard
)
1742 fs_info
->fs_devices
->num_can_discard
--;
1744 fs_info
->fs_devices
->open_devices
--;
1746 /* zero out the old super */
1747 btrfs_scratch_superblock(srcdev
);
1750 call_rcu(&srcdev
->rcu
, free_device
);
1753 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
1754 struct btrfs_device
*tgtdev
)
1756 struct btrfs_device
*next_device
;
1759 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
1761 btrfs_scratch_superblock(tgtdev
);
1762 fs_info
->fs_devices
->open_devices
--;
1764 fs_info
->fs_devices
->num_devices
--;
1765 if (tgtdev
->can_discard
)
1766 fs_info
->fs_devices
->num_can_discard
++;
1768 next_device
= list_entry(fs_info
->fs_devices
->devices
.next
,
1769 struct btrfs_device
, dev_list
);
1770 if (tgtdev
->bdev
== fs_info
->sb
->s_bdev
)
1771 fs_info
->sb
->s_bdev
= next_device
->bdev
;
1772 if (tgtdev
->bdev
== fs_info
->fs_devices
->latest_bdev
)
1773 fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1774 list_del_rcu(&tgtdev
->dev_list
);
1776 call_rcu(&tgtdev
->rcu
, free_device
);
1778 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
1781 static int btrfs_find_device_by_path(struct btrfs_root
*root
, char *device_path
,
1782 struct btrfs_device
**device
)
1785 struct btrfs_super_block
*disk_super
;
1788 struct block_device
*bdev
;
1789 struct buffer_head
*bh
;
1792 ret
= btrfs_get_bdev_and_sb(device_path
, FMODE_READ
,
1793 root
->fs_info
->bdev_holder
, 0, &bdev
, &bh
);
1796 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1797 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1798 dev_uuid
= disk_super
->dev_item
.uuid
;
1799 *device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1804 blkdev_put(bdev
, FMODE_READ
);
1808 int btrfs_find_device_missing_or_by_path(struct btrfs_root
*root
,
1810 struct btrfs_device
**device
)
1813 if (strcmp(device_path
, "missing") == 0) {
1814 struct list_head
*devices
;
1815 struct btrfs_device
*tmp
;
1817 devices
= &root
->fs_info
->fs_devices
->devices
;
1819 * It is safe to read the devices since the volume_mutex
1820 * is held by the caller.
1822 list_for_each_entry(tmp
, devices
, dev_list
) {
1823 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
1830 btrfs_err(root
->fs_info
, "no missing device found");
1836 return btrfs_find_device_by_path(root
, device_path
, device
);
1841 * does all the dirty work required for changing file system's UUID.
1843 static int btrfs_prepare_sprout(struct btrfs_root
*root
)
1845 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
1846 struct btrfs_fs_devices
*old_devices
;
1847 struct btrfs_fs_devices
*seed_devices
;
1848 struct btrfs_super_block
*disk_super
= root
->fs_info
->super_copy
;
1849 struct btrfs_device
*device
;
1852 BUG_ON(!mutex_is_locked(&uuid_mutex
));
1853 if (!fs_devices
->seeding
)
1856 seed_devices
= __alloc_fs_devices();
1857 if (IS_ERR(seed_devices
))
1858 return PTR_ERR(seed_devices
);
1860 old_devices
= clone_fs_devices(fs_devices
);
1861 if (IS_ERR(old_devices
)) {
1862 kfree(seed_devices
);
1863 return PTR_ERR(old_devices
);
1866 list_add(&old_devices
->list
, &fs_uuids
);
1868 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
1869 seed_devices
->opened
= 1;
1870 INIT_LIST_HEAD(&seed_devices
->devices
);
1871 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
1872 mutex_init(&seed_devices
->device_list_mutex
);
1874 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1875 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
1878 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
1879 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
) {
1880 device
->fs_devices
= seed_devices
;
1883 fs_devices
->seeding
= 0;
1884 fs_devices
->num_devices
= 0;
1885 fs_devices
->open_devices
= 0;
1886 fs_devices
->total_devices
= 0;
1887 fs_devices
->seed
= seed_devices
;
1889 generate_random_uuid(fs_devices
->fsid
);
1890 memcpy(root
->fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
1891 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
1892 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1894 super_flags
= btrfs_super_flags(disk_super
) &
1895 ~BTRFS_SUPER_FLAG_SEEDING
;
1896 btrfs_set_super_flags(disk_super
, super_flags
);
1902 * strore the expected generation for seed devices in device items.
1904 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
1905 struct btrfs_root
*root
)
1907 struct btrfs_path
*path
;
1908 struct extent_buffer
*leaf
;
1909 struct btrfs_dev_item
*dev_item
;
1910 struct btrfs_device
*device
;
1911 struct btrfs_key key
;
1912 u8 fs_uuid
[BTRFS_UUID_SIZE
];
1913 u8 dev_uuid
[BTRFS_UUID_SIZE
];
1917 path
= btrfs_alloc_path();
1921 root
= root
->fs_info
->chunk_root
;
1922 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1924 key
.type
= BTRFS_DEV_ITEM_KEY
;
1927 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
1931 leaf
= path
->nodes
[0];
1933 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1934 ret
= btrfs_next_leaf(root
, path
);
1939 leaf
= path
->nodes
[0];
1940 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1941 btrfs_release_path(path
);
1945 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1946 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
1947 key
.type
!= BTRFS_DEV_ITEM_KEY
)
1950 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1951 struct btrfs_dev_item
);
1952 devid
= btrfs_device_id(leaf
, dev_item
);
1953 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
1955 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
1957 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1959 BUG_ON(!device
); /* Logic error */
1961 if (device
->fs_devices
->seeding
) {
1962 btrfs_set_device_generation(leaf
, dev_item
,
1963 device
->generation
);
1964 btrfs_mark_buffer_dirty(leaf
);
1972 btrfs_free_path(path
);
1976 int btrfs_init_new_device(struct btrfs_root
*root
, char *device_path
)
1978 struct request_queue
*q
;
1979 struct btrfs_trans_handle
*trans
;
1980 struct btrfs_device
*device
;
1981 struct block_device
*bdev
;
1982 struct list_head
*devices
;
1983 struct super_block
*sb
= root
->fs_info
->sb
;
1984 struct rcu_string
*name
;
1986 int seeding_dev
= 0;
1989 if ((sb
->s_flags
& MS_RDONLY
) && !root
->fs_info
->fs_devices
->seeding
)
1992 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
1993 root
->fs_info
->bdev_holder
);
1995 return PTR_ERR(bdev
);
1997 if (root
->fs_info
->fs_devices
->seeding
) {
1999 down_write(&sb
->s_umount
);
2000 mutex_lock(&uuid_mutex
);
2003 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2005 devices
= &root
->fs_info
->fs_devices
->devices
;
2007 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2008 list_for_each_entry(device
, devices
, dev_list
) {
2009 if (device
->bdev
== bdev
) {
2012 &root
->fs_info
->fs_devices
->device_list_mutex
);
2016 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2018 device
= btrfs_alloc_device(root
->fs_info
, NULL
, NULL
);
2019 if (IS_ERR(device
)) {
2020 /* we can safely leave the fs_devices entry around */
2021 ret
= PTR_ERR(device
);
2025 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2031 rcu_assign_pointer(device
->name
, name
);
2033 trans
= btrfs_start_transaction(root
, 0);
2034 if (IS_ERR(trans
)) {
2035 rcu_string_free(device
->name
);
2037 ret
= PTR_ERR(trans
);
2043 q
= bdev_get_queue(bdev
);
2044 if (blk_queue_discard(q
))
2045 device
->can_discard
= 1;
2046 device
->writeable
= 1;
2047 device
->generation
= trans
->transid
;
2048 device
->io_width
= root
->sectorsize
;
2049 device
->io_align
= root
->sectorsize
;
2050 device
->sector_size
= root
->sectorsize
;
2051 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
2052 device
->disk_total_bytes
= device
->total_bytes
;
2053 device
->dev_root
= root
->fs_info
->dev_root
;
2054 device
->bdev
= bdev
;
2055 device
->in_fs_metadata
= 1;
2056 device
->is_tgtdev_for_dev_replace
= 0;
2057 device
->mode
= FMODE_EXCL
;
2058 device
->dev_stats_valid
= 1;
2059 set_blocksize(device
->bdev
, 4096);
2062 sb
->s_flags
&= ~MS_RDONLY
;
2063 ret
= btrfs_prepare_sprout(root
);
2064 BUG_ON(ret
); /* -ENOMEM */
2067 device
->fs_devices
= root
->fs_info
->fs_devices
;
2069 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2070 list_add_rcu(&device
->dev_list
, &root
->fs_info
->fs_devices
->devices
);
2071 list_add(&device
->dev_alloc_list
,
2072 &root
->fs_info
->fs_devices
->alloc_list
);
2073 root
->fs_info
->fs_devices
->num_devices
++;
2074 root
->fs_info
->fs_devices
->open_devices
++;
2075 root
->fs_info
->fs_devices
->rw_devices
++;
2076 root
->fs_info
->fs_devices
->total_devices
++;
2077 if (device
->can_discard
)
2078 root
->fs_info
->fs_devices
->num_can_discard
++;
2079 root
->fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
2081 spin_lock(&root
->fs_info
->free_chunk_lock
);
2082 root
->fs_info
->free_chunk_space
+= device
->total_bytes
;
2083 spin_unlock(&root
->fs_info
->free_chunk_lock
);
2085 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
2086 root
->fs_info
->fs_devices
->rotating
= 1;
2088 total_bytes
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
2089 btrfs_set_super_total_bytes(root
->fs_info
->super_copy
,
2090 total_bytes
+ device
->total_bytes
);
2092 total_bytes
= btrfs_super_num_devices(root
->fs_info
->super_copy
);
2093 btrfs_set_super_num_devices(root
->fs_info
->super_copy
,
2095 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2098 ret
= init_first_rw_device(trans
, root
, device
);
2100 btrfs_abort_transaction(trans
, root
, ret
);
2103 ret
= btrfs_finish_sprout(trans
, root
);
2105 btrfs_abort_transaction(trans
, root
, ret
);
2109 ret
= btrfs_add_device(trans
, root
, device
);
2111 btrfs_abort_transaction(trans
, root
, ret
);
2117 * we've got more storage, clear any full flags on the space
2120 btrfs_clear_space_info_full(root
->fs_info
);
2122 unlock_chunks(root
);
2123 root
->fs_info
->num_tolerated_disk_barrier_failures
=
2124 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
2125 ret
= btrfs_commit_transaction(trans
, root
);
2128 mutex_unlock(&uuid_mutex
);
2129 up_write(&sb
->s_umount
);
2131 if (ret
) /* transaction commit */
2134 ret
= btrfs_relocate_sys_chunks(root
);
2136 btrfs_error(root
->fs_info
, ret
,
2137 "Failed to relocate sys chunks after "
2138 "device initialization. This can be fixed "
2139 "using the \"btrfs balance\" command.");
2140 trans
= btrfs_attach_transaction(root
);
2141 if (IS_ERR(trans
)) {
2142 if (PTR_ERR(trans
) == -ENOENT
)
2144 return PTR_ERR(trans
);
2146 ret
= btrfs_commit_transaction(trans
, root
);
2152 unlock_chunks(root
);
2153 btrfs_end_transaction(trans
, root
);
2154 rcu_string_free(device
->name
);
2157 blkdev_put(bdev
, FMODE_EXCL
);
2159 mutex_unlock(&uuid_mutex
);
2160 up_write(&sb
->s_umount
);
2165 int btrfs_init_dev_replace_tgtdev(struct btrfs_root
*root
, char *device_path
,
2166 struct btrfs_device
**device_out
)
2168 struct request_queue
*q
;
2169 struct btrfs_device
*device
;
2170 struct block_device
*bdev
;
2171 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2172 struct list_head
*devices
;
2173 struct rcu_string
*name
;
2174 u64 devid
= BTRFS_DEV_REPLACE_DEVID
;
2178 if (fs_info
->fs_devices
->seeding
)
2181 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2182 fs_info
->bdev_holder
);
2184 return PTR_ERR(bdev
);
2186 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2188 devices
= &fs_info
->fs_devices
->devices
;
2189 list_for_each_entry(device
, devices
, dev_list
) {
2190 if (device
->bdev
== bdev
) {
2196 device
= btrfs_alloc_device(NULL
, &devid
, NULL
);
2197 if (IS_ERR(device
)) {
2198 ret
= PTR_ERR(device
);
2202 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2208 rcu_assign_pointer(device
->name
, name
);
2210 q
= bdev_get_queue(bdev
);
2211 if (blk_queue_discard(q
))
2212 device
->can_discard
= 1;
2213 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2214 device
->writeable
= 1;
2215 device
->generation
= 0;
2216 device
->io_width
= root
->sectorsize
;
2217 device
->io_align
= root
->sectorsize
;
2218 device
->sector_size
= root
->sectorsize
;
2219 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
2220 device
->disk_total_bytes
= device
->total_bytes
;
2221 device
->dev_root
= fs_info
->dev_root
;
2222 device
->bdev
= bdev
;
2223 device
->in_fs_metadata
= 1;
2224 device
->is_tgtdev_for_dev_replace
= 1;
2225 device
->mode
= FMODE_EXCL
;
2226 device
->dev_stats_valid
= 1;
2227 set_blocksize(device
->bdev
, 4096);
2228 device
->fs_devices
= fs_info
->fs_devices
;
2229 list_add(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2230 fs_info
->fs_devices
->num_devices
++;
2231 fs_info
->fs_devices
->open_devices
++;
2232 if (device
->can_discard
)
2233 fs_info
->fs_devices
->num_can_discard
++;
2234 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2236 *device_out
= device
;
2240 blkdev_put(bdev
, FMODE_EXCL
);
2244 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info
*fs_info
,
2245 struct btrfs_device
*tgtdev
)
2247 WARN_ON(fs_info
->fs_devices
->rw_devices
== 0);
2248 tgtdev
->io_width
= fs_info
->dev_root
->sectorsize
;
2249 tgtdev
->io_align
= fs_info
->dev_root
->sectorsize
;
2250 tgtdev
->sector_size
= fs_info
->dev_root
->sectorsize
;
2251 tgtdev
->dev_root
= fs_info
->dev_root
;
2252 tgtdev
->in_fs_metadata
= 1;
2255 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2256 struct btrfs_device
*device
)
2259 struct btrfs_path
*path
;
2260 struct btrfs_root
*root
;
2261 struct btrfs_dev_item
*dev_item
;
2262 struct extent_buffer
*leaf
;
2263 struct btrfs_key key
;
2265 root
= device
->dev_root
->fs_info
->chunk_root
;
2267 path
= btrfs_alloc_path();
2271 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2272 key
.type
= BTRFS_DEV_ITEM_KEY
;
2273 key
.offset
= device
->devid
;
2275 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2284 leaf
= path
->nodes
[0];
2285 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2287 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2288 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2289 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2290 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2291 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2292 btrfs_set_device_total_bytes(leaf
, dev_item
, device
->disk_total_bytes
);
2293 btrfs_set_device_bytes_used(leaf
, dev_item
, device
->bytes_used
);
2294 btrfs_mark_buffer_dirty(leaf
);
2297 btrfs_free_path(path
);
2301 static int __btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2302 struct btrfs_device
*device
, u64 new_size
)
2304 struct btrfs_super_block
*super_copy
=
2305 device
->dev_root
->fs_info
->super_copy
;
2306 u64 old_total
= btrfs_super_total_bytes(super_copy
);
2307 u64 diff
= new_size
- device
->total_bytes
;
2309 if (!device
->writeable
)
2311 if (new_size
<= device
->total_bytes
||
2312 device
->is_tgtdev_for_dev_replace
)
2315 btrfs_set_super_total_bytes(super_copy
, old_total
+ diff
);
2316 device
->fs_devices
->total_rw_bytes
+= diff
;
2318 device
->total_bytes
= new_size
;
2319 device
->disk_total_bytes
= new_size
;
2320 btrfs_clear_space_info_full(device
->dev_root
->fs_info
);
2322 return btrfs_update_device(trans
, device
);
2325 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2326 struct btrfs_device
*device
, u64 new_size
)
2329 lock_chunks(device
->dev_root
);
2330 ret
= __btrfs_grow_device(trans
, device
, new_size
);
2331 unlock_chunks(device
->dev_root
);
2335 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
2336 struct btrfs_root
*root
,
2337 u64 chunk_tree
, u64 chunk_objectid
,
2341 struct btrfs_path
*path
;
2342 struct btrfs_key key
;
2344 root
= root
->fs_info
->chunk_root
;
2345 path
= btrfs_alloc_path();
2349 key
.objectid
= chunk_objectid
;
2350 key
.offset
= chunk_offset
;
2351 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2353 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2356 else if (ret
> 0) { /* Logic error or corruption */
2357 btrfs_error(root
->fs_info
, -ENOENT
,
2358 "Failed lookup while freeing chunk.");
2363 ret
= btrfs_del_item(trans
, root
, path
);
2365 btrfs_error(root
->fs_info
, ret
,
2366 "Failed to delete chunk item.");
2368 btrfs_free_path(path
);
2372 static int btrfs_del_sys_chunk(struct btrfs_root
*root
, u64 chunk_objectid
, u64
2375 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
2376 struct btrfs_disk_key
*disk_key
;
2377 struct btrfs_chunk
*chunk
;
2384 struct btrfs_key key
;
2386 array_size
= btrfs_super_sys_array_size(super_copy
);
2388 ptr
= super_copy
->sys_chunk_array
;
2391 while (cur
< array_size
) {
2392 disk_key
= (struct btrfs_disk_key
*)ptr
;
2393 btrfs_disk_key_to_cpu(&key
, disk_key
);
2395 len
= sizeof(*disk_key
);
2397 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2398 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2399 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2400 len
+= btrfs_chunk_item_size(num_stripes
);
2405 if (key
.objectid
== chunk_objectid
&&
2406 key
.offset
== chunk_offset
) {
2407 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2409 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2418 static int btrfs_relocate_chunk(struct btrfs_root
*root
,
2419 u64 chunk_tree
, u64 chunk_objectid
,
2422 struct extent_map_tree
*em_tree
;
2423 struct btrfs_root
*extent_root
;
2424 struct btrfs_trans_handle
*trans
;
2425 struct extent_map
*em
;
2426 struct map_lookup
*map
;
2430 root
= root
->fs_info
->chunk_root
;
2431 extent_root
= root
->fs_info
->extent_root
;
2432 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
2434 ret
= btrfs_can_relocate(extent_root
, chunk_offset
);
2438 /* step one, relocate all the extents inside this chunk */
2439 ret
= btrfs_relocate_block_group(extent_root
, chunk_offset
);
2443 trans
= btrfs_start_transaction(root
, 0);
2444 if (IS_ERR(trans
)) {
2445 ret
= PTR_ERR(trans
);
2446 btrfs_std_error(root
->fs_info
, ret
);
2453 * step two, delete the device extents and the
2454 * chunk tree entries
2456 read_lock(&em_tree
->lock
);
2457 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
2458 read_unlock(&em_tree
->lock
);
2460 BUG_ON(!em
|| em
->start
> chunk_offset
||
2461 em
->start
+ em
->len
< chunk_offset
);
2462 map
= (struct map_lookup
*)em
->bdev
;
2464 for (i
= 0; i
< map
->num_stripes
; i
++) {
2465 ret
= btrfs_free_dev_extent(trans
, map
->stripes
[i
].dev
,
2466 map
->stripes
[i
].physical
);
2469 if (map
->stripes
[i
].dev
) {
2470 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2474 ret
= btrfs_free_chunk(trans
, root
, chunk_tree
, chunk_objectid
,
2479 trace_btrfs_chunk_free(root
, map
, chunk_offset
, em
->len
);
2481 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2482 ret
= btrfs_del_sys_chunk(root
, chunk_objectid
, chunk_offset
);
2486 ret
= btrfs_remove_block_group(trans
, extent_root
, chunk_offset
);
2489 write_lock(&em_tree
->lock
);
2490 remove_extent_mapping(em_tree
, em
);
2491 write_unlock(&em_tree
->lock
);
2496 /* once for the tree */
2497 free_extent_map(em
);
2499 free_extent_map(em
);
2501 unlock_chunks(root
);
2502 btrfs_end_transaction(trans
, root
);
2506 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
)
2508 struct btrfs_root
*chunk_root
= root
->fs_info
->chunk_root
;
2509 struct btrfs_path
*path
;
2510 struct extent_buffer
*leaf
;
2511 struct btrfs_chunk
*chunk
;
2512 struct btrfs_key key
;
2513 struct btrfs_key found_key
;
2514 u64 chunk_tree
= chunk_root
->root_key
.objectid
;
2516 bool retried
= false;
2520 path
= btrfs_alloc_path();
2525 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2526 key
.offset
= (u64
)-1;
2527 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2530 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2533 BUG_ON(ret
== 0); /* Corruption */
2535 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
2542 leaf
= path
->nodes
[0];
2543 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2545 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
2546 struct btrfs_chunk
);
2547 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2548 btrfs_release_path(path
);
2550 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2551 ret
= btrfs_relocate_chunk(chunk_root
, chunk_tree
,
2560 if (found_key
.offset
== 0)
2562 key
.offset
= found_key
.offset
- 1;
2565 if (failed
&& !retried
) {
2569 } else if (WARN_ON(failed
&& retried
)) {
2573 btrfs_free_path(path
);
2577 static int insert_balance_item(struct btrfs_root
*root
,
2578 struct btrfs_balance_control
*bctl
)
2580 struct btrfs_trans_handle
*trans
;
2581 struct btrfs_balance_item
*item
;
2582 struct btrfs_disk_balance_args disk_bargs
;
2583 struct btrfs_path
*path
;
2584 struct extent_buffer
*leaf
;
2585 struct btrfs_key key
;
2588 path
= btrfs_alloc_path();
2592 trans
= btrfs_start_transaction(root
, 0);
2593 if (IS_ERR(trans
)) {
2594 btrfs_free_path(path
);
2595 return PTR_ERR(trans
);
2598 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2599 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2602 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2607 leaf
= path
->nodes
[0];
2608 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
2610 memset_extent_buffer(leaf
, 0, (unsigned long)item
, sizeof(*item
));
2612 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
2613 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
2614 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
2615 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
2616 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
2617 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
2619 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
2621 btrfs_mark_buffer_dirty(leaf
);
2623 btrfs_free_path(path
);
2624 err
= btrfs_commit_transaction(trans
, root
);
2630 static int del_balance_item(struct btrfs_root
*root
)
2632 struct btrfs_trans_handle
*trans
;
2633 struct btrfs_path
*path
;
2634 struct btrfs_key key
;
2637 path
= btrfs_alloc_path();
2641 trans
= btrfs_start_transaction(root
, 0);
2642 if (IS_ERR(trans
)) {
2643 btrfs_free_path(path
);
2644 return PTR_ERR(trans
);
2647 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2648 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2651 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2659 ret
= btrfs_del_item(trans
, root
, path
);
2661 btrfs_free_path(path
);
2662 err
= btrfs_commit_transaction(trans
, root
);
2669 * This is a heuristic used to reduce the number of chunks balanced on
2670 * resume after balance was interrupted.
2672 static void update_balance_args(struct btrfs_balance_control
*bctl
)
2675 * Turn on soft mode for chunk types that were being converted.
2677 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2678 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2679 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2680 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2681 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2682 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2685 * Turn on usage filter if is not already used. The idea is
2686 * that chunks that we have already balanced should be
2687 * reasonably full. Don't do it for chunks that are being
2688 * converted - that will keep us from relocating unconverted
2689 * (albeit full) chunks.
2691 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2692 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2693 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2694 bctl
->data
.usage
= 90;
2696 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2697 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2698 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2699 bctl
->sys
.usage
= 90;
2701 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2702 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2703 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2704 bctl
->meta
.usage
= 90;
2709 * Should be called with both balance and volume mutexes held to
2710 * serialize other volume operations (add_dev/rm_dev/resize) with
2711 * restriper. Same goes for unset_balance_control.
2713 static void set_balance_control(struct btrfs_balance_control
*bctl
)
2715 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
2717 BUG_ON(fs_info
->balance_ctl
);
2719 spin_lock(&fs_info
->balance_lock
);
2720 fs_info
->balance_ctl
= bctl
;
2721 spin_unlock(&fs_info
->balance_lock
);
2724 static void unset_balance_control(struct btrfs_fs_info
*fs_info
)
2726 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
2728 BUG_ON(!fs_info
->balance_ctl
);
2730 spin_lock(&fs_info
->balance_lock
);
2731 fs_info
->balance_ctl
= NULL
;
2732 spin_unlock(&fs_info
->balance_lock
);
2738 * Balance filters. Return 1 if chunk should be filtered out
2739 * (should not be balanced).
2741 static int chunk_profiles_filter(u64 chunk_type
,
2742 struct btrfs_balance_args
*bargs
)
2744 chunk_type
= chunk_to_extended(chunk_type
) &
2745 BTRFS_EXTENDED_PROFILE_MASK
;
2747 if (bargs
->profiles
& chunk_type
)
2753 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
2754 struct btrfs_balance_args
*bargs
)
2756 struct btrfs_block_group_cache
*cache
;
2757 u64 chunk_used
, user_thresh
;
2760 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
2761 chunk_used
= btrfs_block_group_used(&cache
->item
);
2763 if (bargs
->usage
== 0)
2765 else if (bargs
->usage
> 100)
2766 user_thresh
= cache
->key
.offset
;
2768 user_thresh
= div_factor_fine(cache
->key
.offset
,
2771 if (chunk_used
< user_thresh
)
2774 btrfs_put_block_group(cache
);
2778 static int chunk_devid_filter(struct extent_buffer
*leaf
,
2779 struct btrfs_chunk
*chunk
,
2780 struct btrfs_balance_args
*bargs
)
2782 struct btrfs_stripe
*stripe
;
2783 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2786 for (i
= 0; i
< num_stripes
; i
++) {
2787 stripe
= btrfs_stripe_nr(chunk
, i
);
2788 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
2795 /* [pstart, pend) */
2796 static int chunk_drange_filter(struct extent_buffer
*leaf
,
2797 struct btrfs_chunk
*chunk
,
2799 struct btrfs_balance_args
*bargs
)
2801 struct btrfs_stripe
*stripe
;
2802 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2808 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
2811 if (btrfs_chunk_type(leaf
, chunk
) & (BTRFS_BLOCK_GROUP_DUP
|
2812 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)) {
2813 factor
= num_stripes
/ 2;
2814 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID5
) {
2815 factor
= num_stripes
- 1;
2816 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID6
) {
2817 factor
= num_stripes
- 2;
2819 factor
= num_stripes
;
2822 for (i
= 0; i
< num_stripes
; i
++) {
2823 stripe
= btrfs_stripe_nr(chunk
, i
);
2824 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
2827 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
2828 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
2829 do_div(stripe_length
, factor
);
2831 if (stripe_offset
< bargs
->pend
&&
2832 stripe_offset
+ stripe_length
> bargs
->pstart
)
2839 /* [vstart, vend) */
2840 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
2841 struct btrfs_chunk
*chunk
,
2843 struct btrfs_balance_args
*bargs
)
2845 if (chunk_offset
< bargs
->vend
&&
2846 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
2847 /* at least part of the chunk is inside this vrange */
2853 static int chunk_soft_convert_filter(u64 chunk_type
,
2854 struct btrfs_balance_args
*bargs
)
2856 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
2859 chunk_type
= chunk_to_extended(chunk_type
) &
2860 BTRFS_EXTENDED_PROFILE_MASK
;
2862 if (bargs
->target
== chunk_type
)
2868 static int should_balance_chunk(struct btrfs_root
*root
,
2869 struct extent_buffer
*leaf
,
2870 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
2872 struct btrfs_balance_control
*bctl
= root
->fs_info
->balance_ctl
;
2873 struct btrfs_balance_args
*bargs
= NULL
;
2874 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2877 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
2878 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
2882 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
2883 bargs
= &bctl
->data
;
2884 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
2886 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
2887 bargs
= &bctl
->meta
;
2889 /* profiles filter */
2890 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
2891 chunk_profiles_filter(chunk_type
, bargs
)) {
2896 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2897 chunk_usage_filter(bctl
->fs_info
, chunk_offset
, bargs
)) {
2902 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
2903 chunk_devid_filter(leaf
, chunk
, bargs
)) {
2907 /* drange filter, makes sense only with devid filter */
2908 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
2909 chunk_drange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
2914 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
2915 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
2919 /* soft profile changing mode */
2920 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
2921 chunk_soft_convert_filter(chunk_type
, bargs
)) {
2928 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
2930 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
2931 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
2932 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
2933 struct list_head
*devices
;
2934 struct btrfs_device
*device
;
2937 struct btrfs_chunk
*chunk
;
2938 struct btrfs_path
*path
;
2939 struct btrfs_key key
;
2940 struct btrfs_key found_key
;
2941 struct btrfs_trans_handle
*trans
;
2942 struct extent_buffer
*leaf
;
2945 int enospc_errors
= 0;
2946 bool counting
= true;
2948 /* step one make some room on all the devices */
2949 devices
= &fs_info
->fs_devices
->devices
;
2950 list_for_each_entry(device
, devices
, dev_list
) {
2951 old_size
= device
->total_bytes
;
2952 size_to_free
= div_factor(old_size
, 1);
2953 size_to_free
= min(size_to_free
, (u64
)1 * 1024 * 1024);
2954 if (!device
->writeable
||
2955 device
->total_bytes
- device
->bytes_used
> size_to_free
||
2956 device
->is_tgtdev_for_dev_replace
)
2959 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
2964 trans
= btrfs_start_transaction(dev_root
, 0);
2965 BUG_ON(IS_ERR(trans
));
2967 ret
= btrfs_grow_device(trans
, device
, old_size
);
2970 btrfs_end_transaction(trans
, dev_root
);
2973 /* step two, relocate all the chunks */
2974 path
= btrfs_alloc_path();
2980 /* zero out stat counters */
2981 spin_lock(&fs_info
->balance_lock
);
2982 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
2983 spin_unlock(&fs_info
->balance_lock
);
2985 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2986 key
.offset
= (u64
)-1;
2987 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2990 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
2991 atomic_read(&fs_info
->balance_cancel_req
)) {
2996 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
3001 * this shouldn't happen, it means the last relocate
3005 BUG(); /* FIXME break ? */
3007 ret
= btrfs_previous_item(chunk_root
, path
, 0,
3008 BTRFS_CHUNK_ITEM_KEY
);
3014 leaf
= path
->nodes
[0];
3015 slot
= path
->slots
[0];
3016 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3018 if (found_key
.objectid
!= key
.objectid
)
3021 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
3024 spin_lock(&fs_info
->balance_lock
);
3025 bctl
->stat
.considered
++;
3026 spin_unlock(&fs_info
->balance_lock
);
3029 ret
= should_balance_chunk(chunk_root
, leaf
, chunk
,
3031 btrfs_release_path(path
);
3036 spin_lock(&fs_info
->balance_lock
);
3037 bctl
->stat
.expected
++;
3038 spin_unlock(&fs_info
->balance_lock
);
3042 ret
= btrfs_relocate_chunk(chunk_root
,
3043 chunk_root
->root_key
.objectid
,
3046 if (ret
&& ret
!= -ENOSPC
)
3048 if (ret
== -ENOSPC
) {
3051 spin_lock(&fs_info
->balance_lock
);
3052 bctl
->stat
.completed
++;
3053 spin_unlock(&fs_info
->balance_lock
);
3056 if (found_key
.offset
== 0)
3058 key
.offset
= found_key
.offset
- 1;
3062 btrfs_release_path(path
);
3067 btrfs_free_path(path
);
3068 if (enospc_errors
) {
3069 btrfs_info(fs_info
, "%d enospc errors during balance",
3079 * alloc_profile_is_valid - see if a given profile is valid and reduced
3080 * @flags: profile to validate
3081 * @extended: if true @flags is treated as an extended profile
3083 static int alloc_profile_is_valid(u64 flags
, int extended
)
3085 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
3086 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
3088 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
3090 /* 1) check that all other bits are zeroed */
3094 /* 2) see if profile is reduced */
3096 return !extended
; /* "0" is valid for usual profiles */
3098 /* true if exactly one bit set */
3099 return (flags
& (flags
- 1)) == 0;
3102 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
3104 /* cancel requested || normal exit path */
3105 return atomic_read(&fs_info
->balance_cancel_req
) ||
3106 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
3107 atomic_read(&fs_info
->balance_cancel_req
) == 0);
3110 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
3114 unset_balance_control(fs_info
);
3115 ret
= del_balance_item(fs_info
->tree_root
);
3117 btrfs_std_error(fs_info
, ret
);
3119 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3123 * Should be called with both balance and volume mutexes held
3125 int btrfs_balance(struct btrfs_balance_control
*bctl
,
3126 struct btrfs_ioctl_balance_args
*bargs
)
3128 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3135 if (btrfs_fs_closing(fs_info
) ||
3136 atomic_read(&fs_info
->balance_pause_req
) ||
3137 atomic_read(&fs_info
->balance_cancel_req
)) {
3142 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
3143 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
3147 * In case of mixed groups both data and meta should be picked,
3148 * and identical options should be given for both of them.
3150 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
3151 if (mixed
&& (bctl
->flags
& allowed
)) {
3152 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
3153 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
3154 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
3155 btrfs_err(fs_info
, "with mixed groups data and "
3156 "metadata balance options must be the same");
3162 num_devices
= fs_info
->fs_devices
->num_devices
;
3163 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
3164 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
3165 BUG_ON(num_devices
< 1);
3168 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
3169 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
3170 if (num_devices
== 1)
3171 allowed
|= BTRFS_BLOCK_GROUP_DUP
;
3172 else if (num_devices
> 1)
3173 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
3174 if (num_devices
> 2)
3175 allowed
|= BTRFS_BLOCK_GROUP_RAID5
;
3176 if (num_devices
> 3)
3177 allowed
|= (BTRFS_BLOCK_GROUP_RAID10
|
3178 BTRFS_BLOCK_GROUP_RAID6
);
3179 if ((bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3180 (!alloc_profile_is_valid(bctl
->data
.target
, 1) ||
3181 (bctl
->data
.target
& ~allowed
))) {
3182 btrfs_err(fs_info
, "unable to start balance with target "
3183 "data profile %llu",
3188 if ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3189 (!alloc_profile_is_valid(bctl
->meta
.target
, 1) ||
3190 (bctl
->meta
.target
& ~allowed
))) {
3192 "unable to start balance with target metadata profile %llu",
3197 if ((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3198 (!alloc_profile_is_valid(bctl
->sys
.target
, 1) ||
3199 (bctl
->sys
.target
& ~allowed
))) {
3201 "unable to start balance with target system profile %llu",
3207 /* allow dup'ed data chunks only in mixed mode */
3208 if (!mixed
&& (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3209 (bctl
->data
.target
& BTRFS_BLOCK_GROUP_DUP
)) {
3210 btrfs_err(fs_info
, "dup for data is not allowed");
3215 /* allow to reduce meta or sys integrity only if force set */
3216 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3217 BTRFS_BLOCK_GROUP_RAID10
|
3218 BTRFS_BLOCK_GROUP_RAID5
|
3219 BTRFS_BLOCK_GROUP_RAID6
;
3221 seq
= read_seqbegin(&fs_info
->profiles_lock
);
3223 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3224 (fs_info
->avail_system_alloc_bits
& allowed
) &&
3225 !(bctl
->sys
.target
& allowed
)) ||
3226 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3227 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
3228 !(bctl
->meta
.target
& allowed
))) {
3229 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
3230 btrfs_info(fs_info
, "force reducing metadata integrity");
3232 btrfs_err(fs_info
, "balance will reduce metadata "
3233 "integrity, use force if you want this");
3238 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
3240 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3241 int num_tolerated_disk_barrier_failures
;
3242 u64 target
= bctl
->sys
.target
;
3244 num_tolerated_disk_barrier_failures
=
3245 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3246 if (num_tolerated_disk_barrier_failures
> 0 &&
3248 (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID0
|
3249 BTRFS_AVAIL_ALLOC_BIT_SINGLE
)))
3250 num_tolerated_disk_barrier_failures
= 0;
3251 else if (num_tolerated_disk_barrier_failures
> 1 &&
3253 (BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)))
3254 num_tolerated_disk_barrier_failures
= 1;
3256 fs_info
->num_tolerated_disk_barrier_failures
=
3257 num_tolerated_disk_barrier_failures
;
3260 ret
= insert_balance_item(fs_info
->tree_root
, bctl
);
3261 if (ret
&& ret
!= -EEXIST
)
3264 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
3265 BUG_ON(ret
== -EEXIST
);
3266 set_balance_control(bctl
);
3268 BUG_ON(ret
!= -EEXIST
);
3269 spin_lock(&fs_info
->balance_lock
);
3270 update_balance_args(bctl
);
3271 spin_unlock(&fs_info
->balance_lock
);
3274 atomic_inc(&fs_info
->balance_running
);
3275 mutex_unlock(&fs_info
->balance_mutex
);
3277 ret
= __btrfs_balance(fs_info
);
3279 mutex_lock(&fs_info
->balance_mutex
);
3280 atomic_dec(&fs_info
->balance_running
);
3282 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3283 fs_info
->num_tolerated_disk_barrier_failures
=
3284 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3288 memset(bargs
, 0, sizeof(*bargs
));
3289 update_ioctl_balance_args(fs_info
, 0, bargs
);
3292 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
3293 balance_need_close(fs_info
)) {
3294 __cancel_balance(fs_info
);
3297 wake_up(&fs_info
->balance_wait_q
);
3301 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
3302 __cancel_balance(fs_info
);
3305 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3310 static int balance_kthread(void *data
)
3312 struct btrfs_fs_info
*fs_info
= data
;
3315 mutex_lock(&fs_info
->volume_mutex
);
3316 mutex_lock(&fs_info
->balance_mutex
);
3318 if (fs_info
->balance_ctl
) {
3319 btrfs_info(fs_info
, "continuing balance");
3320 ret
= btrfs_balance(fs_info
->balance_ctl
, NULL
);
3323 mutex_unlock(&fs_info
->balance_mutex
);
3324 mutex_unlock(&fs_info
->volume_mutex
);
3329 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
3331 struct task_struct
*tsk
;
3333 spin_lock(&fs_info
->balance_lock
);
3334 if (!fs_info
->balance_ctl
) {
3335 spin_unlock(&fs_info
->balance_lock
);
3338 spin_unlock(&fs_info
->balance_lock
);
3340 if (btrfs_test_opt(fs_info
->tree_root
, SKIP_BALANCE
)) {
3341 btrfs_info(fs_info
, "force skipping balance");
3345 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
3346 return PTR_ERR_OR_ZERO(tsk
);
3349 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
3351 struct btrfs_balance_control
*bctl
;
3352 struct btrfs_balance_item
*item
;
3353 struct btrfs_disk_balance_args disk_bargs
;
3354 struct btrfs_path
*path
;
3355 struct extent_buffer
*leaf
;
3356 struct btrfs_key key
;
3359 path
= btrfs_alloc_path();
3363 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3364 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
3367 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
3370 if (ret
> 0) { /* ret = -ENOENT; */
3375 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
3381 leaf
= path
->nodes
[0];
3382 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3384 bctl
->fs_info
= fs_info
;
3385 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
3386 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
3388 btrfs_balance_data(leaf
, item
, &disk_bargs
);
3389 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
3390 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
3391 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
3392 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
3393 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
3395 WARN_ON(atomic_xchg(&fs_info
->mutually_exclusive_operation_running
, 1));
3397 mutex_lock(&fs_info
->volume_mutex
);
3398 mutex_lock(&fs_info
->balance_mutex
);
3400 set_balance_control(bctl
);
3402 mutex_unlock(&fs_info
->balance_mutex
);
3403 mutex_unlock(&fs_info
->volume_mutex
);
3405 btrfs_free_path(path
);
3409 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
3413 mutex_lock(&fs_info
->balance_mutex
);
3414 if (!fs_info
->balance_ctl
) {
3415 mutex_unlock(&fs_info
->balance_mutex
);
3419 if (atomic_read(&fs_info
->balance_running
)) {
3420 atomic_inc(&fs_info
->balance_pause_req
);
3421 mutex_unlock(&fs_info
->balance_mutex
);
3423 wait_event(fs_info
->balance_wait_q
,
3424 atomic_read(&fs_info
->balance_running
) == 0);
3426 mutex_lock(&fs_info
->balance_mutex
);
3427 /* we are good with balance_ctl ripped off from under us */
3428 BUG_ON(atomic_read(&fs_info
->balance_running
));
3429 atomic_dec(&fs_info
->balance_pause_req
);
3434 mutex_unlock(&fs_info
->balance_mutex
);
3438 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
3440 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
3443 mutex_lock(&fs_info
->balance_mutex
);
3444 if (!fs_info
->balance_ctl
) {
3445 mutex_unlock(&fs_info
->balance_mutex
);
3449 atomic_inc(&fs_info
->balance_cancel_req
);
3451 * if we are running just wait and return, balance item is
3452 * deleted in btrfs_balance in this case
3454 if (atomic_read(&fs_info
->balance_running
)) {
3455 mutex_unlock(&fs_info
->balance_mutex
);
3456 wait_event(fs_info
->balance_wait_q
,
3457 atomic_read(&fs_info
->balance_running
) == 0);
3458 mutex_lock(&fs_info
->balance_mutex
);
3460 /* __cancel_balance needs volume_mutex */
3461 mutex_unlock(&fs_info
->balance_mutex
);
3462 mutex_lock(&fs_info
->volume_mutex
);
3463 mutex_lock(&fs_info
->balance_mutex
);
3465 if (fs_info
->balance_ctl
)
3466 __cancel_balance(fs_info
);
3468 mutex_unlock(&fs_info
->volume_mutex
);
3471 BUG_ON(fs_info
->balance_ctl
|| atomic_read(&fs_info
->balance_running
));
3472 atomic_dec(&fs_info
->balance_cancel_req
);
3473 mutex_unlock(&fs_info
->balance_mutex
);
3477 static int btrfs_uuid_scan_kthread(void *data
)
3479 struct btrfs_fs_info
*fs_info
= data
;
3480 struct btrfs_root
*root
= fs_info
->tree_root
;
3481 struct btrfs_key key
;
3482 struct btrfs_key max_key
;
3483 struct btrfs_path
*path
= NULL
;
3485 struct extent_buffer
*eb
;
3487 struct btrfs_root_item root_item
;
3489 struct btrfs_trans_handle
*trans
= NULL
;
3491 path
= btrfs_alloc_path();
3498 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3501 max_key
.objectid
= (u64
)-1;
3502 max_key
.type
= BTRFS_ROOT_ITEM_KEY
;
3503 max_key
.offset
= (u64
)-1;
3505 path
->keep_locks
= 1;
3508 ret
= btrfs_search_forward(root
, &key
, path
, 0);
3515 if (key
.type
!= BTRFS_ROOT_ITEM_KEY
||
3516 (key
.objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
3517 key
.objectid
!= BTRFS_FS_TREE_OBJECTID
) ||
3518 key
.objectid
> BTRFS_LAST_FREE_OBJECTID
)
3521 eb
= path
->nodes
[0];
3522 slot
= path
->slots
[0];
3523 item_size
= btrfs_item_size_nr(eb
, slot
);
3524 if (item_size
< sizeof(root_item
))
3527 read_extent_buffer(eb
, &root_item
,
3528 btrfs_item_ptr_offset(eb
, slot
),
3529 (int)sizeof(root_item
));
3530 if (btrfs_root_refs(&root_item
) == 0)
3533 if (!btrfs_is_empty_uuid(root_item
.uuid
) ||
3534 !btrfs_is_empty_uuid(root_item
.received_uuid
)) {
3538 btrfs_release_path(path
);
3540 * 1 - subvol uuid item
3541 * 1 - received_subvol uuid item
3543 trans
= btrfs_start_transaction(fs_info
->uuid_root
, 2);
3544 if (IS_ERR(trans
)) {
3545 ret
= PTR_ERR(trans
);
3553 if (!btrfs_is_empty_uuid(root_item
.uuid
)) {
3554 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
3556 BTRFS_UUID_KEY_SUBVOL
,
3559 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
3565 if (!btrfs_is_empty_uuid(root_item
.received_uuid
)) {
3566 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
3567 root_item
.received_uuid
,
3568 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
3571 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
3579 ret
= btrfs_end_transaction(trans
, fs_info
->uuid_root
);
3585 btrfs_release_path(path
);
3586 if (key
.offset
< (u64
)-1) {
3588 } else if (key
.type
< BTRFS_ROOT_ITEM_KEY
) {
3590 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3591 } else if (key
.objectid
< (u64
)-1) {
3593 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3602 btrfs_free_path(path
);
3603 if (trans
&& !IS_ERR(trans
))
3604 btrfs_end_transaction(trans
, fs_info
->uuid_root
);
3606 btrfs_warn(fs_info
, "btrfs_uuid_scan_kthread failed %d", ret
);
3608 fs_info
->update_uuid_tree_gen
= 1;
3609 up(&fs_info
->uuid_tree_rescan_sem
);
3614 * Callback for btrfs_uuid_tree_iterate().
3616 * 0 check succeeded, the entry is not outdated.
3617 * < 0 if an error occured.
3618 * > 0 if the check failed, which means the caller shall remove the entry.
3620 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info
*fs_info
,
3621 u8
*uuid
, u8 type
, u64 subid
)
3623 struct btrfs_key key
;
3625 struct btrfs_root
*subvol_root
;
3627 if (type
!= BTRFS_UUID_KEY_SUBVOL
&&
3628 type
!= BTRFS_UUID_KEY_RECEIVED_SUBVOL
)
3631 key
.objectid
= subid
;
3632 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3633 key
.offset
= (u64
)-1;
3634 subvol_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
3635 if (IS_ERR(subvol_root
)) {
3636 ret
= PTR_ERR(subvol_root
);
3643 case BTRFS_UUID_KEY_SUBVOL
:
3644 if (memcmp(uuid
, subvol_root
->root_item
.uuid
, BTRFS_UUID_SIZE
))
3647 case BTRFS_UUID_KEY_RECEIVED_SUBVOL
:
3648 if (memcmp(uuid
, subvol_root
->root_item
.received_uuid
,
3658 static int btrfs_uuid_rescan_kthread(void *data
)
3660 struct btrfs_fs_info
*fs_info
= (struct btrfs_fs_info
*)data
;
3664 * 1st step is to iterate through the existing UUID tree and
3665 * to delete all entries that contain outdated data.
3666 * 2nd step is to add all missing entries to the UUID tree.
3668 ret
= btrfs_uuid_tree_iterate(fs_info
, btrfs_check_uuid_tree_entry
);
3670 btrfs_warn(fs_info
, "iterating uuid_tree failed %d", ret
);
3671 up(&fs_info
->uuid_tree_rescan_sem
);
3674 return btrfs_uuid_scan_kthread(data
);
3677 int btrfs_create_uuid_tree(struct btrfs_fs_info
*fs_info
)
3679 struct btrfs_trans_handle
*trans
;
3680 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
3681 struct btrfs_root
*uuid_root
;
3682 struct task_struct
*task
;
3689 trans
= btrfs_start_transaction(tree_root
, 2);
3691 return PTR_ERR(trans
);
3693 uuid_root
= btrfs_create_tree(trans
, fs_info
,
3694 BTRFS_UUID_TREE_OBJECTID
);
3695 if (IS_ERR(uuid_root
)) {
3696 btrfs_abort_transaction(trans
, tree_root
,
3697 PTR_ERR(uuid_root
));
3698 return PTR_ERR(uuid_root
);
3701 fs_info
->uuid_root
= uuid_root
;
3703 ret
= btrfs_commit_transaction(trans
, tree_root
);
3707 down(&fs_info
->uuid_tree_rescan_sem
);
3708 task
= kthread_run(btrfs_uuid_scan_kthread
, fs_info
, "btrfs-uuid");
3710 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3711 btrfs_warn(fs_info
, "failed to start uuid_scan task");
3712 up(&fs_info
->uuid_tree_rescan_sem
);
3713 return PTR_ERR(task
);
3719 int btrfs_check_uuid_tree(struct btrfs_fs_info
*fs_info
)
3721 struct task_struct
*task
;
3723 down(&fs_info
->uuid_tree_rescan_sem
);
3724 task
= kthread_run(btrfs_uuid_rescan_kthread
, fs_info
, "btrfs-uuid");
3726 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3727 btrfs_warn(fs_info
, "failed to start uuid_rescan task");
3728 up(&fs_info
->uuid_tree_rescan_sem
);
3729 return PTR_ERR(task
);
3736 * shrinking a device means finding all of the device extents past
3737 * the new size, and then following the back refs to the chunks.
3738 * The chunk relocation code actually frees the device extent
3740 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
3742 struct btrfs_trans_handle
*trans
;
3743 struct btrfs_root
*root
= device
->dev_root
;
3744 struct btrfs_dev_extent
*dev_extent
= NULL
;
3745 struct btrfs_path
*path
;
3753 bool retried
= false;
3754 struct extent_buffer
*l
;
3755 struct btrfs_key key
;
3756 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
3757 u64 old_total
= btrfs_super_total_bytes(super_copy
);
3758 u64 old_size
= device
->total_bytes
;
3759 u64 diff
= device
->total_bytes
- new_size
;
3761 if (device
->is_tgtdev_for_dev_replace
)
3764 path
= btrfs_alloc_path();
3772 device
->total_bytes
= new_size
;
3773 if (device
->writeable
) {
3774 device
->fs_devices
->total_rw_bytes
-= diff
;
3775 spin_lock(&root
->fs_info
->free_chunk_lock
);
3776 root
->fs_info
->free_chunk_space
-= diff
;
3777 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3779 unlock_chunks(root
);
3782 key
.objectid
= device
->devid
;
3783 key
.offset
= (u64
)-1;
3784 key
.type
= BTRFS_DEV_EXTENT_KEY
;
3787 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3791 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
3796 btrfs_release_path(path
);
3801 slot
= path
->slots
[0];
3802 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
3804 if (key
.objectid
!= device
->devid
) {
3805 btrfs_release_path(path
);
3809 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
3810 length
= btrfs_dev_extent_length(l
, dev_extent
);
3812 if (key
.offset
+ length
<= new_size
) {
3813 btrfs_release_path(path
);
3817 chunk_tree
= btrfs_dev_extent_chunk_tree(l
, dev_extent
);
3818 chunk_objectid
= btrfs_dev_extent_chunk_objectid(l
, dev_extent
);
3819 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
3820 btrfs_release_path(path
);
3822 ret
= btrfs_relocate_chunk(root
, chunk_tree
, chunk_objectid
,
3824 if (ret
&& ret
!= -ENOSPC
)
3828 } while (key
.offset
-- > 0);
3830 if (failed
&& !retried
) {
3834 } else if (failed
&& retried
) {
3838 device
->total_bytes
= old_size
;
3839 if (device
->writeable
)
3840 device
->fs_devices
->total_rw_bytes
+= diff
;
3841 spin_lock(&root
->fs_info
->free_chunk_lock
);
3842 root
->fs_info
->free_chunk_space
+= diff
;
3843 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3844 unlock_chunks(root
);
3848 /* Shrinking succeeded, else we would be at "done". */
3849 trans
= btrfs_start_transaction(root
, 0);
3850 if (IS_ERR(trans
)) {
3851 ret
= PTR_ERR(trans
);
3857 device
->disk_total_bytes
= new_size
;
3858 /* Now btrfs_update_device() will change the on-disk size. */
3859 ret
= btrfs_update_device(trans
, device
);
3861 unlock_chunks(root
);
3862 btrfs_end_transaction(trans
, root
);
3865 WARN_ON(diff
> old_total
);
3866 btrfs_set_super_total_bytes(super_copy
, old_total
- diff
);
3867 unlock_chunks(root
);
3868 btrfs_end_transaction(trans
, root
);
3870 btrfs_free_path(path
);
3874 static int btrfs_add_system_chunk(struct btrfs_root
*root
,
3875 struct btrfs_key
*key
,
3876 struct btrfs_chunk
*chunk
, int item_size
)
3878 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
3879 struct btrfs_disk_key disk_key
;
3883 array_size
= btrfs_super_sys_array_size(super_copy
);
3884 if (array_size
+ item_size
> BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
)
3887 ptr
= super_copy
->sys_chunk_array
+ array_size
;
3888 btrfs_cpu_key_to_disk(&disk_key
, key
);
3889 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
3890 ptr
+= sizeof(disk_key
);
3891 memcpy(ptr
, chunk
, item_size
);
3892 item_size
+= sizeof(disk_key
);
3893 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
3898 * sort the devices in descending order by max_avail, total_avail
3900 static int btrfs_cmp_device_info(const void *a
, const void *b
)
3902 const struct btrfs_device_info
*di_a
= a
;
3903 const struct btrfs_device_info
*di_b
= b
;
3905 if (di_a
->max_avail
> di_b
->max_avail
)
3907 if (di_a
->max_avail
< di_b
->max_avail
)
3909 if (di_a
->total_avail
> di_b
->total_avail
)
3911 if (di_a
->total_avail
< di_b
->total_avail
)
3916 static struct btrfs_raid_attr btrfs_raid_array
[BTRFS_NR_RAID_TYPES
] = {
3917 [BTRFS_RAID_RAID10
] = {
3920 .devs_max
= 0, /* 0 == as many as possible */
3922 .devs_increment
= 2,
3925 [BTRFS_RAID_RAID1
] = {
3930 .devs_increment
= 2,
3933 [BTRFS_RAID_DUP
] = {
3938 .devs_increment
= 1,
3941 [BTRFS_RAID_RAID0
] = {
3946 .devs_increment
= 1,
3949 [BTRFS_RAID_SINGLE
] = {
3954 .devs_increment
= 1,
3957 [BTRFS_RAID_RAID5
] = {
3962 .devs_increment
= 1,
3965 [BTRFS_RAID_RAID6
] = {
3970 .devs_increment
= 1,
3975 static u32
find_raid56_stripe_len(u32 data_devices
, u32 dev_stripe_target
)
3977 /* TODO allow them to set a preferred stripe size */
3981 static void check_raid56_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
3983 if (!(type
& (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
)))
3986 btrfs_set_fs_incompat(info
, RAID56
);
3989 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
3990 struct btrfs_root
*extent_root
, u64 start
,
3993 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
3994 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
3995 struct list_head
*cur
;
3996 struct map_lookup
*map
= NULL
;
3997 struct extent_map_tree
*em_tree
;
3998 struct extent_map
*em
;
3999 struct btrfs_device_info
*devices_info
= NULL
;
4001 int num_stripes
; /* total number of stripes to allocate */
4002 int data_stripes
; /* number of stripes that count for
4004 int sub_stripes
; /* sub_stripes info for map */
4005 int dev_stripes
; /* stripes per dev */
4006 int devs_max
; /* max devs to use */
4007 int devs_min
; /* min devs needed */
4008 int devs_increment
; /* ndevs has to be a multiple of this */
4009 int ncopies
; /* how many copies to data has */
4011 u64 max_stripe_size
;
4015 u64 raid_stripe_len
= BTRFS_STRIPE_LEN
;
4021 BUG_ON(!alloc_profile_is_valid(type
, 0));
4023 if (list_empty(&fs_devices
->alloc_list
))
4026 index
= __get_raid_index(type
);
4028 sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
4029 dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
4030 devs_max
= btrfs_raid_array
[index
].devs_max
;
4031 devs_min
= btrfs_raid_array
[index
].devs_min
;
4032 devs_increment
= btrfs_raid_array
[index
].devs_increment
;
4033 ncopies
= btrfs_raid_array
[index
].ncopies
;
4035 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
4036 max_stripe_size
= 1024 * 1024 * 1024;
4037 max_chunk_size
= 10 * max_stripe_size
;
4038 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
4039 /* for larger filesystems, use larger metadata chunks */
4040 if (fs_devices
->total_rw_bytes
> 50ULL * 1024 * 1024 * 1024)
4041 max_stripe_size
= 1024 * 1024 * 1024;
4043 max_stripe_size
= 256 * 1024 * 1024;
4044 max_chunk_size
= max_stripe_size
;
4045 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4046 max_stripe_size
= 32 * 1024 * 1024;
4047 max_chunk_size
= 2 * max_stripe_size
;
4049 btrfs_err(info
, "invalid chunk type 0x%llx requested\n",
4054 /* we don't want a chunk larger than 10% of writeable space */
4055 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
4058 devices_info
= kzalloc(sizeof(*devices_info
) * fs_devices
->rw_devices
,
4063 cur
= fs_devices
->alloc_list
.next
;
4066 * in the first pass through the devices list, we gather information
4067 * about the available holes on each device.
4070 while (cur
!= &fs_devices
->alloc_list
) {
4071 struct btrfs_device
*device
;
4075 device
= list_entry(cur
, struct btrfs_device
, dev_alloc_list
);
4079 if (!device
->writeable
) {
4081 "BTRFS: read-only device in alloc_list\n");
4085 if (!device
->in_fs_metadata
||
4086 device
->is_tgtdev_for_dev_replace
)
4089 if (device
->total_bytes
> device
->bytes_used
)
4090 total_avail
= device
->total_bytes
- device
->bytes_used
;
4094 /* If there is no space on this device, skip it. */
4095 if (total_avail
== 0)
4098 ret
= find_free_dev_extent(trans
, device
,
4099 max_stripe_size
* dev_stripes
,
4100 &dev_offset
, &max_avail
);
4101 if (ret
&& ret
!= -ENOSPC
)
4105 max_avail
= max_stripe_size
* dev_stripes
;
4107 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
4110 if (ndevs
== fs_devices
->rw_devices
) {
4111 WARN(1, "%s: found more than %llu devices\n",
4112 __func__
, fs_devices
->rw_devices
);
4115 devices_info
[ndevs
].dev_offset
= dev_offset
;
4116 devices_info
[ndevs
].max_avail
= max_avail
;
4117 devices_info
[ndevs
].total_avail
= total_avail
;
4118 devices_info
[ndevs
].dev
= device
;
4123 * now sort the devices by hole size / available space
4125 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
4126 btrfs_cmp_device_info
, NULL
);
4128 /* round down to number of usable stripes */
4129 ndevs
-= ndevs
% devs_increment
;
4131 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
4136 if (devs_max
&& ndevs
> devs_max
)
4139 * the primary goal is to maximize the number of stripes, so use as many
4140 * devices as possible, even if the stripes are not maximum sized.
4142 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4143 num_stripes
= ndevs
* dev_stripes
;
4146 * this will have to be fixed for RAID1 and RAID10 over
4149 data_stripes
= num_stripes
/ ncopies
;
4151 if (type
& BTRFS_BLOCK_GROUP_RAID5
) {
4152 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 1,
4153 btrfs_super_stripesize(info
->super_copy
));
4154 data_stripes
= num_stripes
- 1;
4156 if (type
& BTRFS_BLOCK_GROUP_RAID6
) {
4157 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 2,
4158 btrfs_super_stripesize(info
->super_copy
));
4159 data_stripes
= num_stripes
- 2;
4163 * Use the number of data stripes to figure out how big this chunk
4164 * is really going to be in terms of logical address space,
4165 * and compare that answer with the max chunk size
4167 if (stripe_size
* data_stripes
> max_chunk_size
) {
4168 u64 mask
= (1ULL << 24) - 1;
4169 stripe_size
= max_chunk_size
;
4170 do_div(stripe_size
, data_stripes
);
4172 /* bump the answer up to a 16MB boundary */
4173 stripe_size
= (stripe_size
+ mask
) & ~mask
;
4175 /* but don't go higher than the limits we found
4176 * while searching for free extents
4178 if (stripe_size
> devices_info
[ndevs
-1].max_avail
)
4179 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4182 do_div(stripe_size
, dev_stripes
);
4184 /* align to BTRFS_STRIPE_LEN */
4185 do_div(stripe_size
, raid_stripe_len
);
4186 stripe_size
*= raid_stripe_len
;
4188 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
4193 map
->num_stripes
= num_stripes
;
4195 for (i
= 0; i
< ndevs
; ++i
) {
4196 for (j
= 0; j
< dev_stripes
; ++j
) {
4197 int s
= i
* dev_stripes
+ j
;
4198 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
4199 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
4203 map
->sector_size
= extent_root
->sectorsize
;
4204 map
->stripe_len
= raid_stripe_len
;
4205 map
->io_align
= raid_stripe_len
;
4206 map
->io_width
= raid_stripe_len
;
4208 map
->sub_stripes
= sub_stripes
;
4210 num_bytes
= stripe_size
* data_stripes
;
4212 trace_btrfs_chunk_alloc(info
->chunk_root
, map
, start
, num_bytes
);
4214 em
= alloc_extent_map();
4219 em
->bdev
= (struct block_device
*)map
;
4221 em
->len
= num_bytes
;
4222 em
->block_start
= 0;
4223 em
->block_len
= em
->len
;
4224 em
->orig_block_len
= stripe_size
;
4226 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
4227 write_lock(&em_tree
->lock
);
4228 ret
= add_extent_mapping(em_tree
, em
, 0);
4230 list_add_tail(&em
->list
, &trans
->transaction
->pending_chunks
);
4231 atomic_inc(&em
->refs
);
4233 write_unlock(&em_tree
->lock
);
4235 free_extent_map(em
);
4239 ret
= btrfs_make_block_group(trans
, extent_root
, 0, type
,
4240 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4243 goto error_del_extent
;
4245 free_extent_map(em
);
4246 check_raid56_incompat_flag(extent_root
->fs_info
, type
);
4248 kfree(devices_info
);
4252 write_lock(&em_tree
->lock
);
4253 remove_extent_mapping(em_tree
, em
);
4254 write_unlock(&em_tree
->lock
);
4256 /* One for our allocation */
4257 free_extent_map(em
);
4258 /* One for the tree reference */
4259 free_extent_map(em
);
4262 kfree(devices_info
);
4266 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
4267 struct btrfs_root
*extent_root
,
4268 u64 chunk_offset
, u64 chunk_size
)
4270 struct btrfs_key key
;
4271 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
4272 struct btrfs_device
*device
;
4273 struct btrfs_chunk
*chunk
;
4274 struct btrfs_stripe
*stripe
;
4275 struct extent_map_tree
*em_tree
;
4276 struct extent_map
*em
;
4277 struct map_lookup
*map
;
4284 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
4285 read_lock(&em_tree
->lock
);
4286 em
= lookup_extent_mapping(em_tree
, chunk_offset
, chunk_size
);
4287 read_unlock(&em_tree
->lock
);
4290 btrfs_crit(extent_root
->fs_info
, "unable to find logical "
4291 "%Lu len %Lu", chunk_offset
, chunk_size
);
4295 if (em
->start
!= chunk_offset
|| em
->len
!= chunk_size
) {
4296 btrfs_crit(extent_root
->fs_info
, "found a bad mapping, wanted"
4297 " %Lu-%Lu, found %Lu-%Lu\n", chunk_offset
,
4298 chunk_size
, em
->start
, em
->len
);
4299 free_extent_map(em
);
4303 map
= (struct map_lookup
*)em
->bdev
;
4304 item_size
= btrfs_chunk_item_size(map
->num_stripes
);
4305 stripe_size
= em
->orig_block_len
;
4307 chunk
= kzalloc(item_size
, GFP_NOFS
);
4313 for (i
= 0; i
< map
->num_stripes
; i
++) {
4314 device
= map
->stripes
[i
].dev
;
4315 dev_offset
= map
->stripes
[i
].physical
;
4317 device
->bytes_used
+= stripe_size
;
4318 ret
= btrfs_update_device(trans
, device
);
4321 ret
= btrfs_alloc_dev_extent(trans
, device
,
4322 chunk_root
->root_key
.objectid
,
4323 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4324 chunk_offset
, dev_offset
,
4330 spin_lock(&extent_root
->fs_info
->free_chunk_lock
);
4331 extent_root
->fs_info
->free_chunk_space
-= (stripe_size
*
4333 spin_unlock(&extent_root
->fs_info
->free_chunk_lock
);
4335 stripe
= &chunk
->stripe
;
4336 for (i
= 0; i
< map
->num_stripes
; i
++) {
4337 device
= map
->stripes
[i
].dev
;
4338 dev_offset
= map
->stripes
[i
].physical
;
4340 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
4341 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
4342 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
4346 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
4347 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
4348 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
4349 btrfs_set_stack_chunk_type(chunk
, map
->type
);
4350 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
4351 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
4352 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
4353 btrfs_set_stack_chunk_sector_size(chunk
, extent_root
->sectorsize
);
4354 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
4356 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
4357 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
4358 key
.offset
= chunk_offset
;
4360 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
4361 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4363 * TODO: Cleanup of inserted chunk root in case of
4366 ret
= btrfs_add_system_chunk(chunk_root
, &key
, chunk
,
4372 free_extent_map(em
);
4377 * Chunk allocation falls into two parts. The first part does works
4378 * that make the new allocated chunk useable, but not do any operation
4379 * that modifies the chunk tree. The second part does the works that
4380 * require modifying the chunk tree. This division is important for the
4381 * bootstrap process of adding storage to a seed btrfs.
4383 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4384 struct btrfs_root
*extent_root
, u64 type
)
4388 chunk_offset
= find_next_chunk(extent_root
->fs_info
);
4389 return __btrfs_alloc_chunk(trans
, extent_root
, chunk_offset
, type
);
4392 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
4393 struct btrfs_root
*root
,
4394 struct btrfs_device
*device
)
4397 u64 sys_chunk_offset
;
4399 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4400 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
4403 chunk_offset
= find_next_chunk(fs_info
);
4404 alloc_profile
= btrfs_get_alloc_profile(extent_root
, 0);
4405 ret
= __btrfs_alloc_chunk(trans
, extent_root
, chunk_offset
,
4410 sys_chunk_offset
= find_next_chunk(root
->fs_info
);
4411 alloc_profile
= btrfs_get_alloc_profile(fs_info
->chunk_root
, 0);
4412 ret
= __btrfs_alloc_chunk(trans
, extent_root
, sys_chunk_offset
,
4415 btrfs_abort_transaction(trans
, root
, ret
);
4419 ret
= btrfs_add_device(trans
, fs_info
->chunk_root
, device
);
4421 btrfs_abort_transaction(trans
, root
, ret
);
4426 int btrfs_chunk_readonly(struct btrfs_root
*root
, u64 chunk_offset
)
4428 struct extent_map
*em
;
4429 struct map_lookup
*map
;
4430 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
4434 read_lock(&map_tree
->map_tree
.lock
);
4435 em
= lookup_extent_mapping(&map_tree
->map_tree
, chunk_offset
, 1);
4436 read_unlock(&map_tree
->map_tree
.lock
);
4440 if (btrfs_test_opt(root
, DEGRADED
)) {
4441 free_extent_map(em
);
4445 map
= (struct map_lookup
*)em
->bdev
;
4446 for (i
= 0; i
< map
->num_stripes
; i
++) {
4447 if (!map
->stripes
[i
].dev
->writeable
) {
4452 free_extent_map(em
);
4456 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
4458 extent_map_tree_init(&tree
->map_tree
);
4461 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
4463 struct extent_map
*em
;
4466 write_lock(&tree
->map_tree
.lock
);
4467 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
4469 remove_extent_mapping(&tree
->map_tree
, em
);
4470 write_unlock(&tree
->map_tree
.lock
);
4475 free_extent_map(em
);
4476 /* once for the tree */
4477 free_extent_map(em
);
4481 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
4483 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
4484 struct extent_map
*em
;
4485 struct map_lookup
*map
;
4486 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4489 read_lock(&em_tree
->lock
);
4490 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4491 read_unlock(&em_tree
->lock
);
4494 * We could return errors for these cases, but that could get ugly and
4495 * we'd probably do the same thing which is just not do anything else
4496 * and exit, so return 1 so the callers don't try to use other copies.
4499 btrfs_crit(fs_info
, "No mapping for %Lu-%Lu\n", logical
,
4504 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
4505 btrfs_crit(fs_info
, "Invalid mapping for %Lu-%Lu, got "
4506 "%Lu-%Lu\n", logical
, logical
+len
, em
->start
,
4507 em
->start
+ em
->len
);
4508 free_extent_map(em
);
4512 map
= (struct map_lookup
*)em
->bdev
;
4513 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
4514 ret
= map
->num_stripes
;
4515 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
4516 ret
= map
->sub_stripes
;
4517 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID5
)
4519 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
4523 free_extent_map(em
);
4525 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
4526 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
))
4528 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
4533 unsigned long btrfs_full_stripe_len(struct btrfs_root
*root
,
4534 struct btrfs_mapping_tree
*map_tree
,
4537 struct extent_map
*em
;
4538 struct map_lookup
*map
;
4539 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4540 unsigned long len
= root
->sectorsize
;
4542 read_lock(&em_tree
->lock
);
4543 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4544 read_unlock(&em_tree
->lock
);
4547 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4548 map
= (struct map_lookup
*)em
->bdev
;
4549 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
4550 BTRFS_BLOCK_GROUP_RAID6
)) {
4551 len
= map
->stripe_len
* nr_data_stripes(map
);
4553 free_extent_map(em
);
4557 int btrfs_is_parity_mirror(struct btrfs_mapping_tree
*map_tree
,
4558 u64 logical
, u64 len
, int mirror_num
)
4560 struct extent_map
*em
;
4561 struct map_lookup
*map
;
4562 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4565 read_lock(&em_tree
->lock
);
4566 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4567 read_unlock(&em_tree
->lock
);
4570 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4571 map
= (struct map_lookup
*)em
->bdev
;
4572 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
4573 BTRFS_BLOCK_GROUP_RAID6
))
4575 free_extent_map(em
);
4579 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
4580 struct map_lookup
*map
, int first
, int num
,
4581 int optimal
, int dev_replace_is_ongoing
)
4585 struct btrfs_device
*srcdev
;
4587 if (dev_replace_is_ongoing
&&
4588 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
4589 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
4590 srcdev
= fs_info
->dev_replace
.srcdev
;
4595 * try to avoid the drive that is the source drive for a
4596 * dev-replace procedure, only choose it if no other non-missing
4597 * mirror is available
4599 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
4600 if (map
->stripes
[optimal
].dev
->bdev
&&
4601 (tolerance
|| map
->stripes
[optimal
].dev
!= srcdev
))
4603 for (i
= first
; i
< first
+ num
; i
++) {
4604 if (map
->stripes
[i
].dev
->bdev
&&
4605 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
4610 /* we couldn't find one that doesn't fail. Just return something
4611 * and the io error handling code will clean up eventually
4616 static inline int parity_smaller(u64 a
, u64 b
)
4621 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4622 static void sort_parity_stripes(struct btrfs_bio
*bbio
, u64
*raid_map
)
4624 struct btrfs_bio_stripe s
;
4631 for (i
= 0; i
< bbio
->num_stripes
- 1; i
++) {
4632 if (parity_smaller(raid_map
[i
], raid_map
[i
+1])) {
4633 s
= bbio
->stripes
[i
];
4635 bbio
->stripes
[i
] = bbio
->stripes
[i
+1];
4636 raid_map
[i
] = raid_map
[i
+1];
4637 bbio
->stripes
[i
+1] = s
;
4645 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
4646 u64 logical
, u64
*length
,
4647 struct btrfs_bio
**bbio_ret
,
4648 int mirror_num
, u64
**raid_map_ret
)
4650 struct extent_map
*em
;
4651 struct map_lookup
*map
;
4652 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
4653 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4656 u64 stripe_end_offset
;
4661 u64
*raid_map
= NULL
;
4667 struct btrfs_bio
*bbio
= NULL
;
4668 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
4669 int dev_replace_is_ongoing
= 0;
4670 int num_alloc_stripes
;
4671 int patch_the_first_stripe_for_dev_replace
= 0;
4672 u64 physical_to_patch_in_first_stripe
= 0;
4673 u64 raid56_full_stripe_start
= (u64
)-1;
4675 read_lock(&em_tree
->lock
);
4676 em
= lookup_extent_mapping(em_tree
, logical
, *length
);
4677 read_unlock(&em_tree
->lock
);
4680 btrfs_crit(fs_info
, "unable to find logical %llu len %llu",
4685 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
4686 btrfs_crit(fs_info
, "found a bad mapping, wanted %Lu, "
4687 "found %Lu-%Lu\n", logical
, em
->start
,
4688 em
->start
+ em
->len
);
4689 free_extent_map(em
);
4693 map
= (struct map_lookup
*)em
->bdev
;
4694 offset
= logical
- em
->start
;
4696 stripe_len
= map
->stripe_len
;
4699 * stripe_nr counts the total number of stripes we have to stride
4700 * to get to this block
4702 do_div(stripe_nr
, stripe_len
);
4704 stripe_offset
= stripe_nr
* stripe_len
;
4705 BUG_ON(offset
< stripe_offset
);
4707 /* stripe_offset is the offset of this block in its stripe*/
4708 stripe_offset
= offset
- stripe_offset
;
4710 /* if we're here for raid56, we need to know the stripe aligned start */
4711 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
)) {
4712 unsigned long full_stripe_len
= stripe_len
* nr_data_stripes(map
);
4713 raid56_full_stripe_start
= offset
;
4715 /* allow a write of a full stripe, but make sure we don't
4716 * allow straddling of stripes
4718 do_div(raid56_full_stripe_start
, full_stripe_len
);
4719 raid56_full_stripe_start
*= full_stripe_len
;
4722 if (rw
& REQ_DISCARD
) {
4723 /* we don't discard raid56 yet */
4725 (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
)) {
4729 *length
= min_t(u64
, em
->len
- offset
, *length
);
4730 } else if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
4732 /* For writes to RAID[56], allow a full stripeset across all disks.
4733 For other RAID types and for RAID[56] reads, just allow a single
4734 stripe (on a single disk). */
4735 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
) &&
4737 max_len
= stripe_len
* nr_data_stripes(map
) -
4738 (offset
- raid56_full_stripe_start
);
4740 /* we limit the length of each bio to what fits in a stripe */
4741 max_len
= stripe_len
- stripe_offset
;
4743 *length
= min_t(u64
, em
->len
- offset
, max_len
);
4745 *length
= em
->len
- offset
;
4748 /* This is for when we're called from btrfs_merge_bio_hook() and all
4749 it cares about is the length */
4753 btrfs_dev_replace_lock(dev_replace
);
4754 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
4755 if (!dev_replace_is_ongoing
)
4756 btrfs_dev_replace_unlock(dev_replace
);
4758 if (dev_replace_is_ongoing
&& mirror_num
== map
->num_stripes
+ 1 &&
4759 !(rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) &&
4760 dev_replace
->tgtdev
!= NULL
) {
4762 * in dev-replace case, for repair case (that's the only
4763 * case where the mirror is selected explicitly when
4764 * calling btrfs_map_block), blocks left of the left cursor
4765 * can also be read from the target drive.
4766 * For REQ_GET_READ_MIRRORS, the target drive is added as
4767 * the last one to the array of stripes. For READ, it also
4768 * needs to be supported using the same mirror number.
4769 * If the requested block is not left of the left cursor,
4770 * EIO is returned. This can happen because btrfs_num_copies()
4771 * returns one more in the dev-replace case.
4773 u64 tmp_length
= *length
;
4774 struct btrfs_bio
*tmp_bbio
= NULL
;
4775 int tmp_num_stripes
;
4776 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
4777 int index_srcdev
= 0;
4779 u64 physical_of_found
= 0;
4781 ret
= __btrfs_map_block(fs_info
, REQ_GET_READ_MIRRORS
,
4782 logical
, &tmp_length
, &tmp_bbio
, 0, NULL
);
4784 WARN_ON(tmp_bbio
!= NULL
);
4788 tmp_num_stripes
= tmp_bbio
->num_stripes
;
4789 if (mirror_num
> tmp_num_stripes
) {
4791 * REQ_GET_READ_MIRRORS does not contain this
4792 * mirror, that means that the requested area
4793 * is not left of the left cursor
4801 * process the rest of the function using the mirror_num
4802 * of the source drive. Therefore look it up first.
4803 * At the end, patch the device pointer to the one of the
4806 for (i
= 0; i
< tmp_num_stripes
; i
++) {
4807 if (tmp_bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
4809 * In case of DUP, in order to keep it
4810 * simple, only add the mirror with the
4811 * lowest physical address
4814 physical_of_found
<=
4815 tmp_bbio
->stripes
[i
].physical
)
4820 tmp_bbio
->stripes
[i
].physical
;
4825 mirror_num
= index_srcdev
+ 1;
4826 patch_the_first_stripe_for_dev_replace
= 1;
4827 physical_to_patch_in_first_stripe
= physical_of_found
;
4836 } else if (mirror_num
> map
->num_stripes
) {
4842 stripe_nr_orig
= stripe_nr
;
4843 stripe_nr_end
= ALIGN(offset
+ *length
, map
->stripe_len
);
4844 do_div(stripe_nr_end
, map
->stripe_len
);
4845 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
4848 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
4849 if (rw
& REQ_DISCARD
)
4850 num_stripes
= min_t(u64
, map
->num_stripes
,
4851 stripe_nr_end
- stripe_nr_orig
);
4852 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
4853 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
4854 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
))
4855 num_stripes
= map
->num_stripes
;
4856 else if (mirror_num
)
4857 stripe_index
= mirror_num
- 1;
4859 stripe_index
= find_live_mirror(fs_info
, map
, 0,
4861 current
->pid
% map
->num_stripes
,
4862 dev_replace_is_ongoing
);
4863 mirror_num
= stripe_index
+ 1;
4866 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
4867 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) {
4868 num_stripes
= map
->num_stripes
;
4869 } else if (mirror_num
) {
4870 stripe_index
= mirror_num
- 1;
4875 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
4876 int factor
= map
->num_stripes
/ map
->sub_stripes
;
4878 stripe_index
= do_div(stripe_nr
, factor
);
4879 stripe_index
*= map
->sub_stripes
;
4881 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
))
4882 num_stripes
= map
->sub_stripes
;
4883 else if (rw
& REQ_DISCARD
)
4884 num_stripes
= min_t(u64
, map
->sub_stripes
*
4885 (stripe_nr_end
- stripe_nr_orig
),
4887 else if (mirror_num
)
4888 stripe_index
+= mirror_num
- 1;
4890 int old_stripe_index
= stripe_index
;
4891 stripe_index
= find_live_mirror(fs_info
, map
,
4893 map
->sub_stripes
, stripe_index
+
4894 current
->pid
% map
->sub_stripes
,
4895 dev_replace_is_ongoing
);
4896 mirror_num
= stripe_index
- old_stripe_index
+ 1;
4899 } else if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
4900 BTRFS_BLOCK_GROUP_RAID6
)) {
4903 if (bbio_ret
&& ((rw
& REQ_WRITE
) || mirror_num
> 1)
4907 /* push stripe_nr back to the start of the full stripe */
4908 stripe_nr
= raid56_full_stripe_start
;
4909 do_div(stripe_nr
, stripe_len
);
4911 stripe_index
= do_div(stripe_nr
, nr_data_stripes(map
));
4913 /* RAID[56] write or recovery. Return all stripes */
4914 num_stripes
= map
->num_stripes
;
4915 max_errors
= nr_parity_stripes(map
);
4917 raid_map
= kmalloc_array(num_stripes
, sizeof(u64
),
4924 /* Work out the disk rotation on this stripe-set */
4926 rot
= do_div(tmp
, num_stripes
);
4928 /* Fill in the logical address of each stripe */
4929 tmp
= stripe_nr
* nr_data_stripes(map
);
4930 for (i
= 0; i
< nr_data_stripes(map
); i
++)
4931 raid_map
[(i
+rot
) % num_stripes
] =
4932 em
->start
+ (tmp
+ i
) * map
->stripe_len
;
4934 raid_map
[(i
+rot
) % map
->num_stripes
] = RAID5_P_STRIPE
;
4935 if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
4936 raid_map
[(i
+rot
+1) % num_stripes
] =
4939 *length
= map
->stripe_len
;
4944 * Mirror #0 or #1 means the original data block.
4945 * Mirror #2 is RAID5 parity block.
4946 * Mirror #3 is RAID6 Q block.
4948 stripe_index
= do_div(stripe_nr
, nr_data_stripes(map
));
4950 stripe_index
= nr_data_stripes(map
) +
4953 /* We distribute the parity blocks across stripes */
4954 tmp
= stripe_nr
+ stripe_index
;
4955 stripe_index
= do_div(tmp
, map
->num_stripes
);
4959 * after this do_div call, stripe_nr is the number of stripes
4960 * on this device we have to walk to find the data, and
4961 * stripe_index is the number of our device in the stripe array
4963 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
4964 mirror_num
= stripe_index
+ 1;
4966 BUG_ON(stripe_index
>= map
->num_stripes
);
4968 num_alloc_stripes
= num_stripes
;
4969 if (dev_replace_is_ongoing
) {
4970 if (rw
& (REQ_WRITE
| REQ_DISCARD
))
4971 num_alloc_stripes
<<= 1;
4972 if (rw
& REQ_GET_READ_MIRRORS
)
4973 num_alloc_stripes
++;
4975 bbio
= kzalloc(btrfs_bio_size(num_alloc_stripes
), GFP_NOFS
);
4981 atomic_set(&bbio
->error
, 0);
4983 if (rw
& REQ_DISCARD
) {
4985 int sub_stripes
= 0;
4986 u64 stripes_per_dev
= 0;
4987 u32 remaining_stripes
= 0;
4988 u32 last_stripe
= 0;
4991 (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID10
)) {
4992 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
4995 sub_stripes
= map
->sub_stripes
;
4997 factor
= map
->num_stripes
/ sub_stripes
;
4998 stripes_per_dev
= div_u64_rem(stripe_nr_end
-
5001 &remaining_stripes
);
5002 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
5003 last_stripe
*= sub_stripes
;
5006 for (i
= 0; i
< num_stripes
; i
++) {
5007 bbio
->stripes
[i
].physical
=
5008 map
->stripes
[stripe_index
].physical
+
5009 stripe_offset
+ stripe_nr
* map
->stripe_len
;
5010 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
5012 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5013 BTRFS_BLOCK_GROUP_RAID10
)) {
5014 bbio
->stripes
[i
].length
= stripes_per_dev
*
5017 if (i
/ sub_stripes
< remaining_stripes
)
5018 bbio
->stripes
[i
].length
+=
5022 * Special for the first stripe and
5025 * |-------|...|-------|
5029 if (i
< sub_stripes
)
5030 bbio
->stripes
[i
].length
-=
5033 if (stripe_index
>= last_stripe
&&
5034 stripe_index
<= (last_stripe
+
5036 bbio
->stripes
[i
].length
-=
5039 if (i
== sub_stripes
- 1)
5042 bbio
->stripes
[i
].length
= *length
;
5045 if (stripe_index
== map
->num_stripes
) {
5046 /* This could only happen for RAID0/10 */
5052 for (i
= 0; i
< num_stripes
; i
++) {
5053 bbio
->stripes
[i
].physical
=
5054 map
->stripes
[stripe_index
].physical
+
5056 stripe_nr
* map
->stripe_len
;
5057 bbio
->stripes
[i
].dev
=
5058 map
->stripes
[stripe_index
].dev
;
5063 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
)) {
5064 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
5065 BTRFS_BLOCK_GROUP_RAID10
|
5066 BTRFS_BLOCK_GROUP_RAID5
|
5067 BTRFS_BLOCK_GROUP_DUP
)) {
5069 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
) {
5074 if (dev_replace_is_ongoing
&& (rw
& (REQ_WRITE
| REQ_DISCARD
)) &&
5075 dev_replace
->tgtdev
!= NULL
) {
5076 int index_where_to_add
;
5077 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5080 * duplicate the write operations while the dev replace
5081 * procedure is running. Since the copying of the old disk
5082 * to the new disk takes place at run time while the
5083 * filesystem is mounted writable, the regular write
5084 * operations to the old disk have to be duplicated to go
5085 * to the new disk as well.
5086 * Note that device->missing is handled by the caller, and
5087 * that the write to the old disk is already set up in the
5090 index_where_to_add
= num_stripes
;
5091 for (i
= 0; i
< num_stripes
; i
++) {
5092 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5093 /* write to new disk, too */
5094 struct btrfs_bio_stripe
*new =
5095 bbio
->stripes
+ index_where_to_add
;
5096 struct btrfs_bio_stripe
*old
=
5099 new->physical
= old
->physical
;
5100 new->length
= old
->length
;
5101 new->dev
= dev_replace
->tgtdev
;
5102 index_where_to_add
++;
5106 num_stripes
= index_where_to_add
;
5107 } else if (dev_replace_is_ongoing
&& (rw
& REQ_GET_READ_MIRRORS
) &&
5108 dev_replace
->tgtdev
!= NULL
) {
5109 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5110 int index_srcdev
= 0;
5112 u64 physical_of_found
= 0;
5115 * During the dev-replace procedure, the target drive can
5116 * also be used to read data in case it is needed to repair
5117 * a corrupt block elsewhere. This is possible if the
5118 * requested area is left of the left cursor. In this area,
5119 * the target drive is a full copy of the source drive.
5121 for (i
= 0; i
< num_stripes
; i
++) {
5122 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5124 * In case of DUP, in order to keep it
5125 * simple, only add the mirror with the
5126 * lowest physical address
5129 physical_of_found
<=
5130 bbio
->stripes
[i
].physical
)
5134 physical_of_found
= bbio
->stripes
[i
].physical
;
5138 u64 length
= map
->stripe_len
;
5140 if (physical_of_found
+ length
<=
5141 dev_replace
->cursor_left
) {
5142 struct btrfs_bio_stripe
*tgtdev_stripe
=
5143 bbio
->stripes
+ num_stripes
;
5145 tgtdev_stripe
->physical
= physical_of_found
;
5146 tgtdev_stripe
->length
=
5147 bbio
->stripes
[index_srcdev
].length
;
5148 tgtdev_stripe
->dev
= dev_replace
->tgtdev
;
5156 bbio
->num_stripes
= num_stripes
;
5157 bbio
->max_errors
= max_errors
;
5158 bbio
->mirror_num
= mirror_num
;
5161 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5162 * mirror_num == num_stripes + 1 && dev_replace target drive is
5163 * available as a mirror
5165 if (patch_the_first_stripe_for_dev_replace
&& num_stripes
> 0) {
5166 WARN_ON(num_stripes
> 1);
5167 bbio
->stripes
[0].dev
= dev_replace
->tgtdev
;
5168 bbio
->stripes
[0].physical
= physical_to_patch_in_first_stripe
;
5169 bbio
->mirror_num
= map
->num_stripes
+ 1;
5172 sort_parity_stripes(bbio
, raid_map
);
5173 *raid_map_ret
= raid_map
;
5176 if (dev_replace_is_ongoing
)
5177 btrfs_dev_replace_unlock(dev_replace
);
5178 free_extent_map(em
);
5182 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
5183 u64 logical
, u64
*length
,
5184 struct btrfs_bio
**bbio_ret
, int mirror_num
)
5186 return __btrfs_map_block(fs_info
, rw
, logical
, length
, bbio_ret
,
5190 int btrfs_rmap_block(struct btrfs_mapping_tree
*map_tree
,
5191 u64 chunk_start
, u64 physical
, u64 devid
,
5192 u64
**logical
, int *naddrs
, int *stripe_len
)
5194 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5195 struct extent_map
*em
;
5196 struct map_lookup
*map
;
5204 read_lock(&em_tree
->lock
);
5205 em
= lookup_extent_mapping(em_tree
, chunk_start
, 1);
5206 read_unlock(&em_tree
->lock
);
5209 printk(KERN_ERR
"BTRFS: couldn't find em for chunk %Lu\n",
5214 if (em
->start
!= chunk_start
) {
5215 printk(KERN_ERR
"BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5216 em
->start
, chunk_start
);
5217 free_extent_map(em
);
5220 map
= (struct map_lookup
*)em
->bdev
;
5223 rmap_len
= map
->stripe_len
;
5225 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5226 do_div(length
, map
->num_stripes
/ map
->sub_stripes
);
5227 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5228 do_div(length
, map
->num_stripes
);
5229 else if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
5230 BTRFS_BLOCK_GROUP_RAID6
)) {
5231 do_div(length
, nr_data_stripes(map
));
5232 rmap_len
= map
->stripe_len
* nr_data_stripes(map
);
5235 buf
= kzalloc(sizeof(u64
) * map
->num_stripes
, GFP_NOFS
);
5236 BUG_ON(!buf
); /* -ENOMEM */
5238 for (i
= 0; i
< map
->num_stripes
; i
++) {
5239 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
5241 if (map
->stripes
[i
].physical
> physical
||
5242 map
->stripes
[i
].physical
+ length
<= physical
)
5245 stripe_nr
= physical
- map
->stripes
[i
].physical
;
5246 do_div(stripe_nr
, map
->stripe_len
);
5248 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5249 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5250 do_div(stripe_nr
, map
->sub_stripes
);
5251 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5252 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5253 } /* else if RAID[56], multiply by nr_data_stripes().
5254 * Alternatively, just use rmap_len below instead of
5255 * map->stripe_len */
5257 bytenr
= chunk_start
+ stripe_nr
* rmap_len
;
5258 WARN_ON(nr
>= map
->num_stripes
);
5259 for (j
= 0; j
< nr
; j
++) {
5260 if (buf
[j
] == bytenr
)
5264 WARN_ON(nr
>= map
->num_stripes
);
5271 *stripe_len
= rmap_len
;
5273 free_extent_map(em
);
5277 static void btrfs_end_bio(struct bio
*bio
, int err
)
5279 struct btrfs_bio
*bbio
= bio
->bi_private
;
5280 struct btrfs_device
*dev
= bbio
->stripes
[0].dev
;
5281 int is_orig_bio
= 0;
5284 atomic_inc(&bbio
->error
);
5285 if (err
== -EIO
|| err
== -EREMOTEIO
) {
5286 unsigned int stripe_index
=
5287 btrfs_io_bio(bio
)->stripe_index
;
5289 BUG_ON(stripe_index
>= bbio
->num_stripes
);
5290 dev
= bbio
->stripes
[stripe_index
].dev
;
5292 if (bio
->bi_rw
& WRITE
)
5293 btrfs_dev_stat_inc(dev
,
5294 BTRFS_DEV_STAT_WRITE_ERRS
);
5296 btrfs_dev_stat_inc(dev
,
5297 BTRFS_DEV_STAT_READ_ERRS
);
5298 if ((bio
->bi_rw
& WRITE_FLUSH
) == WRITE_FLUSH
)
5299 btrfs_dev_stat_inc(dev
,
5300 BTRFS_DEV_STAT_FLUSH_ERRS
);
5301 btrfs_dev_stat_print_on_error(dev
);
5306 if (bio
== bbio
->orig_bio
)
5309 btrfs_bio_counter_dec(bbio
->fs_info
);
5311 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
5314 bio
= bbio
->orig_bio
;
5318 * We have original bio now. So increment bi_remaining to
5319 * account for it in endio
5321 atomic_inc(&bio
->bi_remaining
);
5323 bio
->bi_private
= bbio
->private;
5324 bio
->bi_end_io
= bbio
->end_io
;
5325 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
5326 /* only send an error to the higher layers if it is
5327 * beyond the tolerance of the btrfs bio
5329 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
5333 * this bio is actually up to date, we didn't
5334 * go over the max number of errors
5336 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5341 bio_endio(bio
, err
);
5342 } else if (!is_orig_bio
) {
5348 * see run_scheduled_bios for a description of why bios are collected for
5351 * This will add one bio to the pending list for a device and make sure
5352 * the work struct is scheduled.
5354 static noinline
void btrfs_schedule_bio(struct btrfs_root
*root
,
5355 struct btrfs_device
*device
,
5356 int rw
, struct bio
*bio
)
5358 int should_queue
= 1;
5359 struct btrfs_pending_bios
*pending_bios
;
5361 if (device
->missing
|| !device
->bdev
) {
5362 bio_endio(bio
, -EIO
);
5366 /* don't bother with additional async steps for reads, right now */
5367 if (!(rw
& REQ_WRITE
)) {
5369 btrfsic_submit_bio(rw
, bio
);
5375 * nr_async_bios allows us to reliably return congestion to the
5376 * higher layers. Otherwise, the async bio makes it appear we have
5377 * made progress against dirty pages when we've really just put it
5378 * on a queue for later
5380 atomic_inc(&root
->fs_info
->nr_async_bios
);
5381 WARN_ON(bio
->bi_next
);
5382 bio
->bi_next
= NULL
;
5385 spin_lock(&device
->io_lock
);
5386 if (bio
->bi_rw
& REQ_SYNC
)
5387 pending_bios
= &device
->pending_sync_bios
;
5389 pending_bios
= &device
->pending_bios
;
5391 if (pending_bios
->tail
)
5392 pending_bios
->tail
->bi_next
= bio
;
5394 pending_bios
->tail
= bio
;
5395 if (!pending_bios
->head
)
5396 pending_bios
->head
= bio
;
5397 if (device
->running_pending
)
5400 spin_unlock(&device
->io_lock
);
5403 btrfs_queue_work(root
->fs_info
->submit_workers
,
5407 static int bio_size_ok(struct block_device
*bdev
, struct bio
*bio
,
5410 struct bio_vec
*prev
;
5411 struct request_queue
*q
= bdev_get_queue(bdev
);
5412 unsigned int max_sectors
= queue_max_sectors(q
);
5413 struct bvec_merge_data bvm
= {
5415 .bi_sector
= sector
,
5416 .bi_rw
= bio
->bi_rw
,
5419 if (WARN_ON(bio
->bi_vcnt
== 0))
5422 prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
5423 if (bio_sectors(bio
) > max_sectors
)
5426 if (!q
->merge_bvec_fn
)
5429 bvm
.bi_size
= bio
->bi_iter
.bi_size
- prev
->bv_len
;
5430 if (q
->merge_bvec_fn(q
, &bvm
, prev
) < prev
->bv_len
)
5435 static void submit_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
5436 struct bio
*bio
, u64 physical
, int dev_nr
,
5439 struct btrfs_device
*dev
= bbio
->stripes
[dev_nr
].dev
;
5441 bio
->bi_private
= bbio
;
5442 btrfs_io_bio(bio
)->stripe_index
= dev_nr
;
5443 bio
->bi_end_io
= btrfs_end_bio
;
5444 bio
->bi_iter
.bi_sector
= physical
>> 9;
5447 struct rcu_string
*name
;
5450 name
= rcu_dereference(dev
->name
);
5451 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5452 "(%s id %llu), size=%u\n", rw
,
5453 (u64
)bio
->bi_sector
, (u_long
)dev
->bdev
->bd_dev
,
5454 name
->str
, dev
->devid
, bio
->bi_size
);
5458 bio
->bi_bdev
= dev
->bdev
;
5460 btrfs_bio_counter_inc_noblocked(root
->fs_info
);
5463 btrfs_schedule_bio(root
, dev
, rw
, bio
);
5465 btrfsic_submit_bio(rw
, bio
);
5468 static int breakup_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
5469 struct bio
*first_bio
, struct btrfs_device
*dev
,
5470 int dev_nr
, int rw
, int async
)
5472 struct bio_vec
*bvec
= first_bio
->bi_io_vec
;
5474 int nr_vecs
= bio_get_nr_vecs(dev
->bdev
);
5475 u64 physical
= bbio
->stripes
[dev_nr
].physical
;
5478 bio
= btrfs_bio_alloc(dev
->bdev
, physical
>> 9, nr_vecs
, GFP_NOFS
);
5482 while (bvec
<= (first_bio
->bi_io_vec
+ first_bio
->bi_vcnt
- 1)) {
5483 if (bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
5484 bvec
->bv_offset
) < bvec
->bv_len
) {
5485 u64 len
= bio
->bi_iter
.bi_size
;
5487 atomic_inc(&bbio
->stripes_pending
);
5488 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
,
5496 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
, rw
, async
);
5500 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
5502 atomic_inc(&bbio
->error
);
5503 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
5504 bio
->bi_private
= bbio
->private;
5505 bio
->bi_end_io
= bbio
->end_io
;
5506 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
5507 bio
->bi_iter
.bi_sector
= logical
>> 9;
5509 bio_endio(bio
, -EIO
);
5513 int btrfs_map_bio(struct btrfs_root
*root
, int rw
, struct bio
*bio
,
5514 int mirror_num
, int async_submit
)
5516 struct btrfs_device
*dev
;
5517 struct bio
*first_bio
= bio
;
5518 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
5521 u64
*raid_map
= NULL
;
5525 struct btrfs_bio
*bbio
= NULL
;
5527 length
= bio
->bi_iter
.bi_size
;
5528 map_length
= length
;
5530 btrfs_bio_counter_inc_blocked(root
->fs_info
);
5531 ret
= __btrfs_map_block(root
->fs_info
, rw
, logical
, &map_length
, &bbio
,
5532 mirror_num
, &raid_map
);
5534 btrfs_bio_counter_dec(root
->fs_info
);
5538 total_devs
= bbio
->num_stripes
;
5539 bbio
->orig_bio
= first_bio
;
5540 bbio
->private = first_bio
->bi_private
;
5541 bbio
->end_io
= first_bio
->bi_end_io
;
5542 bbio
->fs_info
= root
->fs_info
;
5543 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
5546 /* In this case, map_length has been set to the length of
5547 a single stripe; not the whole write */
5549 ret
= raid56_parity_write(root
, bio
, bbio
,
5550 raid_map
, map_length
);
5552 ret
= raid56_parity_recover(root
, bio
, bbio
,
5553 raid_map
, map_length
,
5557 * FIXME, replace dosen't support raid56 yet, please fix
5560 btrfs_bio_counter_dec(root
->fs_info
);
5564 if (map_length
< length
) {
5565 btrfs_crit(root
->fs_info
, "mapping failed logical %llu bio len %llu len %llu",
5566 logical
, length
, map_length
);
5570 while (dev_nr
< total_devs
) {
5571 dev
= bbio
->stripes
[dev_nr
].dev
;
5572 if (!dev
|| !dev
->bdev
|| (rw
& WRITE
&& !dev
->writeable
)) {
5573 bbio_error(bbio
, first_bio
, logical
);
5579 * Check and see if we're ok with this bio based on it's size
5580 * and offset with the given device.
5582 if (!bio_size_ok(dev
->bdev
, first_bio
,
5583 bbio
->stripes
[dev_nr
].physical
>> 9)) {
5584 ret
= breakup_stripe_bio(root
, bbio
, first_bio
, dev
,
5585 dev_nr
, rw
, async_submit
);
5591 if (dev_nr
< total_devs
- 1) {
5592 bio
= btrfs_bio_clone(first_bio
, GFP_NOFS
);
5593 BUG_ON(!bio
); /* -ENOMEM */
5598 submit_stripe_bio(root
, bbio
, bio
,
5599 bbio
->stripes
[dev_nr
].physical
, dev_nr
, rw
,
5603 btrfs_bio_counter_dec(root
->fs_info
);
5607 struct btrfs_device
*btrfs_find_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
5610 struct btrfs_device
*device
;
5611 struct btrfs_fs_devices
*cur_devices
;
5613 cur_devices
= fs_info
->fs_devices
;
5614 while (cur_devices
) {
5616 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
5617 device
= __find_device(&cur_devices
->devices
,
5622 cur_devices
= cur_devices
->seed
;
5627 static struct btrfs_device
*add_missing_dev(struct btrfs_root
*root
,
5628 u64 devid
, u8
*dev_uuid
)
5630 struct btrfs_device
*device
;
5631 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
5633 device
= btrfs_alloc_device(NULL
, &devid
, dev_uuid
);
5637 list_add(&device
->dev_list
, &fs_devices
->devices
);
5638 device
->fs_devices
= fs_devices
;
5639 fs_devices
->num_devices
++;
5641 device
->missing
= 1;
5642 fs_devices
->missing_devices
++;
5648 * btrfs_alloc_device - allocate struct btrfs_device
5649 * @fs_info: used only for generating a new devid, can be NULL if
5650 * devid is provided (i.e. @devid != NULL).
5651 * @devid: a pointer to devid for this device. If NULL a new devid
5653 * @uuid: a pointer to UUID for this device. If NULL a new UUID
5656 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5657 * on error. Returned struct is not linked onto any lists and can be
5658 * destroyed with kfree() right away.
5660 struct btrfs_device
*btrfs_alloc_device(struct btrfs_fs_info
*fs_info
,
5664 struct btrfs_device
*dev
;
5667 if (WARN_ON(!devid
&& !fs_info
))
5668 return ERR_PTR(-EINVAL
);
5670 dev
= __alloc_device();
5679 ret
= find_next_devid(fs_info
, &tmp
);
5682 return ERR_PTR(ret
);
5688 memcpy(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
);
5690 generate_random_uuid(dev
->uuid
);
5692 btrfs_init_work(&dev
->work
, pending_bios_fn
, NULL
, NULL
);
5697 static int read_one_chunk(struct btrfs_root
*root
, struct btrfs_key
*key
,
5698 struct extent_buffer
*leaf
,
5699 struct btrfs_chunk
*chunk
)
5701 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
5702 struct map_lookup
*map
;
5703 struct extent_map
*em
;
5707 u8 uuid
[BTRFS_UUID_SIZE
];
5712 logical
= key
->offset
;
5713 length
= btrfs_chunk_length(leaf
, chunk
);
5715 read_lock(&map_tree
->map_tree
.lock
);
5716 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
5717 read_unlock(&map_tree
->map_tree
.lock
);
5719 /* already mapped? */
5720 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
5721 free_extent_map(em
);
5724 free_extent_map(em
);
5727 em
= alloc_extent_map();
5730 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
5731 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
5733 free_extent_map(em
);
5737 em
->bdev
= (struct block_device
*)map
;
5738 em
->start
= logical
;
5741 em
->block_start
= 0;
5742 em
->block_len
= em
->len
;
5744 map
->num_stripes
= num_stripes
;
5745 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
5746 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
5747 map
->sector_size
= btrfs_chunk_sector_size(leaf
, chunk
);
5748 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
5749 map
->type
= btrfs_chunk_type(leaf
, chunk
);
5750 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
5751 for (i
= 0; i
< num_stripes
; i
++) {
5752 map
->stripes
[i
].physical
=
5753 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
5754 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
5755 read_extent_buffer(leaf
, uuid
, (unsigned long)
5756 btrfs_stripe_dev_uuid_nr(chunk
, i
),
5758 map
->stripes
[i
].dev
= btrfs_find_device(root
->fs_info
, devid
,
5760 if (!map
->stripes
[i
].dev
&& !btrfs_test_opt(root
, DEGRADED
)) {
5762 free_extent_map(em
);
5765 if (!map
->stripes
[i
].dev
) {
5766 map
->stripes
[i
].dev
=
5767 add_missing_dev(root
, devid
, uuid
);
5768 if (!map
->stripes
[i
].dev
) {
5770 free_extent_map(em
);
5774 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
5777 write_lock(&map_tree
->map_tree
.lock
);
5778 ret
= add_extent_mapping(&map_tree
->map_tree
, em
, 0);
5779 write_unlock(&map_tree
->map_tree
.lock
);
5780 BUG_ON(ret
); /* Tree corruption */
5781 free_extent_map(em
);
5786 static void fill_device_from_item(struct extent_buffer
*leaf
,
5787 struct btrfs_dev_item
*dev_item
,
5788 struct btrfs_device
*device
)
5792 device
->devid
= btrfs_device_id(leaf
, dev_item
);
5793 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
5794 device
->total_bytes
= device
->disk_total_bytes
;
5795 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
5796 device
->type
= btrfs_device_type(leaf
, dev_item
);
5797 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
5798 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
5799 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
5800 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
5801 device
->is_tgtdev_for_dev_replace
= 0;
5803 ptr
= btrfs_device_uuid(dev_item
);
5804 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
5807 static int open_seed_devices(struct btrfs_root
*root
, u8
*fsid
)
5809 struct btrfs_fs_devices
*fs_devices
;
5812 BUG_ON(!mutex_is_locked(&uuid_mutex
));
5814 fs_devices
= root
->fs_info
->fs_devices
->seed
;
5815 while (fs_devices
) {
5816 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
5820 fs_devices
= fs_devices
->seed
;
5823 fs_devices
= find_fsid(fsid
);
5829 fs_devices
= clone_fs_devices(fs_devices
);
5830 if (IS_ERR(fs_devices
)) {
5831 ret
= PTR_ERR(fs_devices
);
5835 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
5836 root
->fs_info
->bdev_holder
);
5838 free_fs_devices(fs_devices
);
5842 if (!fs_devices
->seeding
) {
5843 __btrfs_close_devices(fs_devices
);
5844 free_fs_devices(fs_devices
);
5849 fs_devices
->seed
= root
->fs_info
->fs_devices
->seed
;
5850 root
->fs_info
->fs_devices
->seed
= fs_devices
;
5855 static int read_one_dev(struct btrfs_root
*root
,
5856 struct extent_buffer
*leaf
,
5857 struct btrfs_dev_item
*dev_item
)
5859 struct btrfs_device
*device
;
5862 u8 fs_uuid
[BTRFS_UUID_SIZE
];
5863 u8 dev_uuid
[BTRFS_UUID_SIZE
];
5865 devid
= btrfs_device_id(leaf
, dev_item
);
5866 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
5868 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
5871 if (memcmp(fs_uuid
, root
->fs_info
->fsid
, BTRFS_UUID_SIZE
)) {
5872 ret
= open_seed_devices(root
, fs_uuid
);
5873 if (ret
&& !btrfs_test_opt(root
, DEGRADED
))
5877 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
, fs_uuid
);
5878 if (!device
|| !device
->bdev
) {
5879 if (!btrfs_test_opt(root
, DEGRADED
))
5883 btrfs_warn(root
->fs_info
, "devid %llu missing", devid
);
5884 device
= add_missing_dev(root
, devid
, dev_uuid
);
5887 } else if (!device
->missing
) {
5889 * this happens when a device that was properly setup
5890 * in the device info lists suddenly goes bad.
5891 * device->bdev is NULL, and so we have to set
5892 * device->missing to one here
5894 root
->fs_info
->fs_devices
->missing_devices
++;
5895 device
->missing
= 1;
5899 if (device
->fs_devices
!= root
->fs_info
->fs_devices
) {
5900 BUG_ON(device
->writeable
);
5901 if (device
->generation
!=
5902 btrfs_device_generation(leaf
, dev_item
))
5906 fill_device_from_item(leaf
, dev_item
, device
);
5907 device
->in_fs_metadata
= 1;
5908 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
5909 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
5910 spin_lock(&root
->fs_info
->free_chunk_lock
);
5911 root
->fs_info
->free_chunk_space
+= device
->total_bytes
-
5913 spin_unlock(&root
->fs_info
->free_chunk_lock
);
5919 int btrfs_read_sys_array(struct btrfs_root
*root
)
5921 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
5922 struct extent_buffer
*sb
;
5923 struct btrfs_disk_key
*disk_key
;
5924 struct btrfs_chunk
*chunk
;
5926 unsigned long sb_ptr
;
5932 struct btrfs_key key
;
5934 sb
= btrfs_find_create_tree_block(root
, BTRFS_SUPER_INFO_OFFSET
,
5935 BTRFS_SUPER_INFO_SIZE
);
5938 btrfs_set_buffer_uptodate(sb
);
5939 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
5941 * The sb extent buffer is artifical and just used to read the system array.
5942 * btrfs_set_buffer_uptodate() call does not properly mark all it's
5943 * pages up-to-date when the page is larger: extent does not cover the
5944 * whole page and consequently check_page_uptodate does not find all
5945 * the page's extents up-to-date (the hole beyond sb),
5946 * write_extent_buffer then triggers a WARN_ON.
5948 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5949 * but sb spans only this function. Add an explicit SetPageUptodate call
5950 * to silence the warning eg. on PowerPC 64.
5952 if (PAGE_CACHE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
5953 SetPageUptodate(sb
->pages
[0]);
5955 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
5956 array_size
= btrfs_super_sys_array_size(super_copy
);
5958 ptr
= super_copy
->sys_chunk_array
;
5959 sb_ptr
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
5962 while (cur
< array_size
) {
5963 disk_key
= (struct btrfs_disk_key
*)ptr
;
5964 btrfs_disk_key_to_cpu(&key
, disk_key
);
5966 len
= sizeof(*disk_key
); ptr
+= len
;
5970 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
5971 chunk
= (struct btrfs_chunk
*)sb_ptr
;
5972 ret
= read_one_chunk(root
, &key
, sb
, chunk
);
5975 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
5976 len
= btrfs_chunk_item_size(num_stripes
);
5985 free_extent_buffer(sb
);
5989 int btrfs_read_chunk_tree(struct btrfs_root
*root
)
5991 struct btrfs_path
*path
;
5992 struct extent_buffer
*leaf
;
5993 struct btrfs_key key
;
5994 struct btrfs_key found_key
;
5998 root
= root
->fs_info
->chunk_root
;
6000 path
= btrfs_alloc_path();
6004 mutex_lock(&uuid_mutex
);
6008 * Read all device items, and then all the chunk items. All
6009 * device items are found before any chunk item (their object id
6010 * is smaller than the lowest possible object id for a chunk
6011 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6013 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
6016 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6020 leaf
= path
->nodes
[0];
6021 slot
= path
->slots
[0];
6022 if (slot
>= btrfs_header_nritems(leaf
)) {
6023 ret
= btrfs_next_leaf(root
, path
);
6030 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6031 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
6032 struct btrfs_dev_item
*dev_item
;
6033 dev_item
= btrfs_item_ptr(leaf
, slot
,
6034 struct btrfs_dev_item
);
6035 ret
= read_one_dev(root
, leaf
, dev_item
);
6038 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6039 struct btrfs_chunk
*chunk
;
6040 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
6041 ret
= read_one_chunk(root
, &found_key
, leaf
, chunk
);
6049 unlock_chunks(root
);
6050 mutex_unlock(&uuid_mutex
);
6052 btrfs_free_path(path
);
6056 void btrfs_init_devices_late(struct btrfs_fs_info
*fs_info
)
6058 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6059 struct btrfs_device
*device
;
6061 mutex_lock(&fs_devices
->device_list_mutex
);
6062 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
)
6063 device
->dev_root
= fs_info
->dev_root
;
6064 mutex_unlock(&fs_devices
->device_list_mutex
);
6067 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
6071 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6072 btrfs_dev_stat_reset(dev
, i
);
6075 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
6077 struct btrfs_key key
;
6078 struct btrfs_key found_key
;
6079 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6080 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6081 struct extent_buffer
*eb
;
6084 struct btrfs_device
*device
;
6085 struct btrfs_path
*path
= NULL
;
6088 path
= btrfs_alloc_path();
6094 mutex_lock(&fs_devices
->device_list_mutex
);
6095 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6097 struct btrfs_dev_stats_item
*ptr
;
6100 key
.type
= BTRFS_DEV_STATS_KEY
;
6101 key
.offset
= device
->devid
;
6102 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
6104 __btrfs_reset_dev_stats(device
);
6105 device
->dev_stats_valid
= 1;
6106 btrfs_release_path(path
);
6109 slot
= path
->slots
[0];
6110 eb
= path
->nodes
[0];
6111 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
6112 item_size
= btrfs_item_size_nr(eb
, slot
);
6114 ptr
= btrfs_item_ptr(eb
, slot
,
6115 struct btrfs_dev_stats_item
);
6117 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6118 if (item_size
>= (1 + i
) * sizeof(__le64
))
6119 btrfs_dev_stat_set(device
, i
,
6120 btrfs_dev_stats_value(eb
, ptr
, i
));
6122 btrfs_dev_stat_reset(device
, i
);
6125 device
->dev_stats_valid
= 1;
6126 btrfs_dev_stat_print_on_load(device
);
6127 btrfs_release_path(path
);
6129 mutex_unlock(&fs_devices
->device_list_mutex
);
6132 btrfs_free_path(path
);
6133 return ret
< 0 ? ret
: 0;
6136 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
6137 struct btrfs_root
*dev_root
,
6138 struct btrfs_device
*device
)
6140 struct btrfs_path
*path
;
6141 struct btrfs_key key
;
6142 struct extent_buffer
*eb
;
6143 struct btrfs_dev_stats_item
*ptr
;
6148 key
.type
= BTRFS_DEV_STATS_KEY
;
6149 key
.offset
= device
->devid
;
6151 path
= btrfs_alloc_path();
6153 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
6155 printk_in_rcu(KERN_WARNING
"BTRFS: "
6156 "error %d while searching for dev_stats item for device %s!\n",
6157 ret
, rcu_str_deref(device
->name
));
6162 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
6163 /* need to delete old one and insert a new one */
6164 ret
= btrfs_del_item(trans
, dev_root
, path
);
6166 printk_in_rcu(KERN_WARNING
"BTRFS: "
6167 "delete too small dev_stats item for device %s failed %d!\n",
6168 rcu_str_deref(device
->name
), ret
);
6175 /* need to insert a new item */
6176 btrfs_release_path(path
);
6177 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
6178 &key
, sizeof(*ptr
));
6180 printk_in_rcu(KERN_WARNING
"BTRFS: "
6181 "insert dev_stats item for device %s failed %d!\n",
6182 rcu_str_deref(device
->name
), ret
);
6187 eb
= path
->nodes
[0];
6188 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
6189 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6190 btrfs_set_dev_stats_value(eb
, ptr
, i
,
6191 btrfs_dev_stat_read(device
, i
));
6192 btrfs_mark_buffer_dirty(eb
);
6195 btrfs_free_path(path
);
6200 * called from commit_transaction. Writes all changed device stats to disk.
6202 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
6203 struct btrfs_fs_info
*fs_info
)
6205 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6206 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6207 struct btrfs_device
*device
;
6210 mutex_lock(&fs_devices
->device_list_mutex
);
6211 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6212 if (!device
->dev_stats_valid
|| !device
->dev_stats_dirty
)
6215 ret
= update_dev_stat_item(trans
, dev_root
, device
);
6217 device
->dev_stats_dirty
= 0;
6219 mutex_unlock(&fs_devices
->device_list_mutex
);
6224 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
6226 btrfs_dev_stat_inc(dev
, index
);
6227 btrfs_dev_stat_print_on_error(dev
);
6230 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
6232 if (!dev
->dev_stats_valid
)
6234 printk_ratelimited_in_rcu(KERN_ERR
"BTRFS: "
6235 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6236 rcu_str_deref(dev
->name
),
6237 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
6238 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
6239 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
6240 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
6241 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
6244 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
6248 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6249 if (btrfs_dev_stat_read(dev
, i
) != 0)
6251 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
6252 return; /* all values == 0, suppress message */
6254 printk_in_rcu(KERN_INFO
"BTRFS: "
6255 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6256 rcu_str_deref(dev
->name
),
6257 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
6258 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
6259 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
6260 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
6261 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
6264 int btrfs_get_dev_stats(struct btrfs_root
*root
,
6265 struct btrfs_ioctl_get_dev_stats
*stats
)
6267 struct btrfs_device
*dev
;
6268 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
6271 mutex_lock(&fs_devices
->device_list_mutex
);
6272 dev
= btrfs_find_device(root
->fs_info
, stats
->devid
, NULL
, NULL
);
6273 mutex_unlock(&fs_devices
->device_list_mutex
);
6276 btrfs_warn(root
->fs_info
, "get dev_stats failed, device not found");
6278 } else if (!dev
->dev_stats_valid
) {
6279 btrfs_warn(root
->fs_info
, "get dev_stats failed, not yet valid");
6281 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
6282 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6283 if (stats
->nr_items
> i
)
6285 btrfs_dev_stat_read_and_reset(dev
, i
);
6287 btrfs_dev_stat_reset(dev
, i
);
6290 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6291 if (stats
->nr_items
> i
)
6292 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
6294 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
6295 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
6299 int btrfs_scratch_superblock(struct btrfs_device
*device
)
6301 struct buffer_head
*bh
;
6302 struct btrfs_super_block
*disk_super
;
6304 bh
= btrfs_read_dev_super(device
->bdev
);
6307 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
6309 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
6310 set_buffer_dirty(bh
);
6311 sync_dirty_buffer(bh
);