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
);
1456 * Function to update ctime/mtime for a given device path.
1457 * Mainly used for ctime/mtime based probe like libblkid.
1459 static void update_dev_time(char *path_name
)
1463 filp
= filp_open(path_name
, O_RDWR
, 0);
1466 file_update_time(filp
);
1467 filp_close(filp
, NULL
);
1471 static int btrfs_rm_dev_item(struct btrfs_root
*root
,
1472 struct btrfs_device
*device
)
1475 struct btrfs_path
*path
;
1476 struct btrfs_key key
;
1477 struct btrfs_trans_handle
*trans
;
1479 root
= root
->fs_info
->chunk_root
;
1481 path
= btrfs_alloc_path();
1485 trans
= btrfs_start_transaction(root
, 0);
1486 if (IS_ERR(trans
)) {
1487 btrfs_free_path(path
);
1488 return PTR_ERR(trans
);
1490 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1491 key
.type
= BTRFS_DEV_ITEM_KEY
;
1492 key
.offset
= device
->devid
;
1495 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1504 ret
= btrfs_del_item(trans
, root
, path
);
1508 btrfs_free_path(path
);
1509 unlock_chunks(root
);
1510 btrfs_commit_transaction(trans
, root
);
1514 int btrfs_rm_device(struct btrfs_root
*root
, char *device_path
)
1516 struct btrfs_device
*device
;
1517 struct btrfs_device
*next_device
;
1518 struct block_device
*bdev
;
1519 struct buffer_head
*bh
= NULL
;
1520 struct btrfs_super_block
*disk_super
;
1521 struct btrfs_fs_devices
*cur_devices
;
1528 bool clear_super
= false;
1530 mutex_lock(&uuid_mutex
);
1533 seq
= read_seqbegin(&root
->fs_info
->profiles_lock
);
1535 all_avail
= root
->fs_info
->avail_data_alloc_bits
|
1536 root
->fs_info
->avail_system_alloc_bits
|
1537 root
->fs_info
->avail_metadata_alloc_bits
;
1538 } while (read_seqretry(&root
->fs_info
->profiles_lock
, seq
));
1540 num_devices
= root
->fs_info
->fs_devices
->num_devices
;
1541 btrfs_dev_replace_lock(&root
->fs_info
->dev_replace
);
1542 if (btrfs_dev_replace_is_ongoing(&root
->fs_info
->dev_replace
)) {
1543 WARN_ON(num_devices
< 1);
1546 btrfs_dev_replace_unlock(&root
->fs_info
->dev_replace
);
1548 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID10
) && num_devices
<= 4) {
1549 ret
= BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET
;
1553 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID1
) && num_devices
<= 2) {
1554 ret
= BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET
;
1558 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID5
) &&
1559 root
->fs_info
->fs_devices
->rw_devices
<= 2) {
1560 ret
= BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET
;
1563 if ((all_avail
& BTRFS_BLOCK_GROUP_RAID6
) &&
1564 root
->fs_info
->fs_devices
->rw_devices
<= 3) {
1565 ret
= BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET
;
1569 if (strcmp(device_path
, "missing") == 0) {
1570 struct list_head
*devices
;
1571 struct btrfs_device
*tmp
;
1574 devices
= &root
->fs_info
->fs_devices
->devices
;
1576 * It is safe to read the devices since the volume_mutex
1579 list_for_each_entry(tmp
, devices
, dev_list
) {
1580 if (tmp
->in_fs_metadata
&&
1581 !tmp
->is_tgtdev_for_dev_replace
&&
1591 ret
= BTRFS_ERROR_DEV_MISSING_NOT_FOUND
;
1595 ret
= btrfs_get_bdev_and_sb(device_path
,
1596 FMODE_WRITE
| FMODE_EXCL
,
1597 root
->fs_info
->bdev_holder
, 0,
1601 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1602 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1603 dev_uuid
= disk_super
->dev_item
.uuid
;
1604 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1612 if (device
->is_tgtdev_for_dev_replace
) {
1613 ret
= BTRFS_ERROR_DEV_TGT_REPLACE
;
1617 if (device
->writeable
&& root
->fs_info
->fs_devices
->rw_devices
== 1) {
1618 ret
= BTRFS_ERROR_DEV_ONLY_WRITABLE
;
1622 if (device
->writeable
) {
1624 list_del_init(&device
->dev_alloc_list
);
1625 unlock_chunks(root
);
1626 root
->fs_info
->fs_devices
->rw_devices
--;
1630 mutex_unlock(&uuid_mutex
);
1631 ret
= btrfs_shrink_device(device
, 0);
1632 mutex_lock(&uuid_mutex
);
1637 * TODO: the superblock still includes this device in its num_devices
1638 * counter although write_all_supers() is not locked out. This
1639 * could give a filesystem state which requires a degraded mount.
1641 ret
= btrfs_rm_dev_item(root
->fs_info
->chunk_root
, device
);
1645 spin_lock(&root
->fs_info
->free_chunk_lock
);
1646 root
->fs_info
->free_chunk_space
= device
->total_bytes
-
1648 spin_unlock(&root
->fs_info
->free_chunk_lock
);
1650 device
->in_fs_metadata
= 0;
1651 btrfs_scrub_cancel_dev(root
->fs_info
, device
);
1654 * the device list mutex makes sure that we don't change
1655 * the device list while someone else is writing out all
1656 * the device supers. Whoever is writing all supers, should
1657 * lock the device list mutex before getting the number of
1658 * devices in the super block (super_copy). Conversely,
1659 * whoever updates the number of devices in the super block
1660 * (super_copy) should hold the device list mutex.
1663 cur_devices
= device
->fs_devices
;
1664 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1665 list_del_rcu(&device
->dev_list
);
1667 device
->fs_devices
->num_devices
--;
1668 device
->fs_devices
->total_devices
--;
1670 if (device
->missing
)
1671 root
->fs_info
->fs_devices
->missing_devices
--;
1673 next_device
= list_entry(root
->fs_info
->fs_devices
->devices
.next
,
1674 struct btrfs_device
, dev_list
);
1675 if (device
->bdev
== root
->fs_info
->sb
->s_bdev
)
1676 root
->fs_info
->sb
->s_bdev
= next_device
->bdev
;
1677 if (device
->bdev
== root
->fs_info
->fs_devices
->latest_bdev
)
1678 root
->fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1681 device
->fs_devices
->open_devices
--;
1683 call_rcu(&device
->rcu
, free_device
);
1685 num_devices
= btrfs_super_num_devices(root
->fs_info
->super_copy
) - 1;
1686 btrfs_set_super_num_devices(root
->fs_info
->super_copy
, num_devices
);
1687 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1689 if (cur_devices
->open_devices
== 0) {
1690 struct btrfs_fs_devices
*fs_devices
;
1691 fs_devices
= root
->fs_info
->fs_devices
;
1692 while (fs_devices
) {
1693 if (fs_devices
->seed
== cur_devices
)
1695 fs_devices
= fs_devices
->seed
;
1697 fs_devices
->seed
= cur_devices
->seed
;
1698 cur_devices
->seed
= NULL
;
1700 __btrfs_close_devices(cur_devices
);
1701 unlock_chunks(root
);
1702 free_fs_devices(cur_devices
);
1705 root
->fs_info
->num_tolerated_disk_barrier_failures
=
1706 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
1709 * at this point, the device is zero sized. We want to
1710 * remove it from the devices list and zero out the old super
1712 if (clear_super
&& disk_super
) {
1713 /* make sure this device isn't detected as part of
1716 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
1717 set_buffer_dirty(bh
);
1718 sync_dirty_buffer(bh
);
1724 /* Notify udev that device has changed */
1725 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
1727 /* Update ctime/mtime for device path for libblkid */
1728 update_dev_time(device_path
);
1734 blkdev_put(bdev
, FMODE_READ
| FMODE_EXCL
);
1736 mutex_unlock(&uuid_mutex
);
1739 if (device
->writeable
) {
1741 list_add(&device
->dev_alloc_list
,
1742 &root
->fs_info
->fs_devices
->alloc_list
);
1743 unlock_chunks(root
);
1744 root
->fs_info
->fs_devices
->rw_devices
++;
1749 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info
*fs_info
,
1750 struct btrfs_device
*srcdev
)
1752 WARN_ON(!mutex_is_locked(&fs_info
->fs_devices
->device_list_mutex
));
1754 list_del_rcu(&srcdev
->dev_list
);
1755 list_del_rcu(&srcdev
->dev_alloc_list
);
1756 fs_info
->fs_devices
->num_devices
--;
1757 if (srcdev
->missing
) {
1758 fs_info
->fs_devices
->missing_devices
--;
1759 fs_info
->fs_devices
->rw_devices
++;
1761 if (srcdev
->can_discard
)
1762 fs_info
->fs_devices
->num_can_discard
--;
1764 fs_info
->fs_devices
->open_devices
--;
1766 /* zero out the old super */
1767 btrfs_scratch_superblock(srcdev
);
1770 call_rcu(&srcdev
->rcu
, free_device
);
1773 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
1774 struct btrfs_device
*tgtdev
)
1776 struct btrfs_device
*next_device
;
1779 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
1781 btrfs_scratch_superblock(tgtdev
);
1782 fs_info
->fs_devices
->open_devices
--;
1784 fs_info
->fs_devices
->num_devices
--;
1785 if (tgtdev
->can_discard
)
1786 fs_info
->fs_devices
->num_can_discard
++;
1788 next_device
= list_entry(fs_info
->fs_devices
->devices
.next
,
1789 struct btrfs_device
, dev_list
);
1790 if (tgtdev
->bdev
== fs_info
->sb
->s_bdev
)
1791 fs_info
->sb
->s_bdev
= next_device
->bdev
;
1792 if (tgtdev
->bdev
== fs_info
->fs_devices
->latest_bdev
)
1793 fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1794 list_del_rcu(&tgtdev
->dev_list
);
1796 call_rcu(&tgtdev
->rcu
, free_device
);
1798 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
1801 static int btrfs_find_device_by_path(struct btrfs_root
*root
, char *device_path
,
1802 struct btrfs_device
**device
)
1805 struct btrfs_super_block
*disk_super
;
1808 struct block_device
*bdev
;
1809 struct buffer_head
*bh
;
1812 ret
= btrfs_get_bdev_and_sb(device_path
, FMODE_READ
,
1813 root
->fs_info
->bdev_holder
, 0, &bdev
, &bh
);
1816 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
1817 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1818 dev_uuid
= disk_super
->dev_item
.uuid
;
1819 *device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1824 blkdev_put(bdev
, FMODE_READ
);
1828 int btrfs_find_device_missing_or_by_path(struct btrfs_root
*root
,
1830 struct btrfs_device
**device
)
1833 if (strcmp(device_path
, "missing") == 0) {
1834 struct list_head
*devices
;
1835 struct btrfs_device
*tmp
;
1837 devices
= &root
->fs_info
->fs_devices
->devices
;
1839 * It is safe to read the devices since the volume_mutex
1840 * is held by the caller.
1842 list_for_each_entry(tmp
, devices
, dev_list
) {
1843 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
1850 btrfs_err(root
->fs_info
, "no missing device found");
1856 return btrfs_find_device_by_path(root
, device_path
, device
);
1861 * does all the dirty work required for changing file system's UUID.
1863 static int btrfs_prepare_sprout(struct btrfs_root
*root
)
1865 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
1866 struct btrfs_fs_devices
*old_devices
;
1867 struct btrfs_fs_devices
*seed_devices
;
1868 struct btrfs_super_block
*disk_super
= root
->fs_info
->super_copy
;
1869 struct btrfs_device
*device
;
1872 BUG_ON(!mutex_is_locked(&uuid_mutex
));
1873 if (!fs_devices
->seeding
)
1876 seed_devices
= __alloc_fs_devices();
1877 if (IS_ERR(seed_devices
))
1878 return PTR_ERR(seed_devices
);
1880 old_devices
= clone_fs_devices(fs_devices
);
1881 if (IS_ERR(old_devices
)) {
1882 kfree(seed_devices
);
1883 return PTR_ERR(old_devices
);
1886 list_add(&old_devices
->list
, &fs_uuids
);
1888 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
1889 seed_devices
->opened
= 1;
1890 INIT_LIST_HEAD(&seed_devices
->devices
);
1891 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
1892 mutex_init(&seed_devices
->device_list_mutex
);
1894 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1895 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
1898 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
1899 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
) {
1900 device
->fs_devices
= seed_devices
;
1903 fs_devices
->seeding
= 0;
1904 fs_devices
->num_devices
= 0;
1905 fs_devices
->open_devices
= 0;
1906 fs_devices
->total_devices
= 0;
1907 fs_devices
->seed
= seed_devices
;
1909 generate_random_uuid(fs_devices
->fsid
);
1910 memcpy(root
->fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
1911 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
1912 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
1914 super_flags
= btrfs_super_flags(disk_super
) &
1915 ~BTRFS_SUPER_FLAG_SEEDING
;
1916 btrfs_set_super_flags(disk_super
, super_flags
);
1922 * strore the expected generation for seed devices in device items.
1924 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
1925 struct btrfs_root
*root
)
1927 struct btrfs_path
*path
;
1928 struct extent_buffer
*leaf
;
1929 struct btrfs_dev_item
*dev_item
;
1930 struct btrfs_device
*device
;
1931 struct btrfs_key key
;
1932 u8 fs_uuid
[BTRFS_UUID_SIZE
];
1933 u8 dev_uuid
[BTRFS_UUID_SIZE
];
1937 path
= btrfs_alloc_path();
1941 root
= root
->fs_info
->chunk_root
;
1942 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1944 key
.type
= BTRFS_DEV_ITEM_KEY
;
1947 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
1951 leaf
= path
->nodes
[0];
1953 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1954 ret
= btrfs_next_leaf(root
, path
);
1959 leaf
= path
->nodes
[0];
1960 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1961 btrfs_release_path(path
);
1965 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1966 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
1967 key
.type
!= BTRFS_DEV_ITEM_KEY
)
1970 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1971 struct btrfs_dev_item
);
1972 devid
= btrfs_device_id(leaf
, dev_item
);
1973 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
1975 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
1977 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
,
1979 BUG_ON(!device
); /* Logic error */
1981 if (device
->fs_devices
->seeding
) {
1982 btrfs_set_device_generation(leaf
, dev_item
,
1983 device
->generation
);
1984 btrfs_mark_buffer_dirty(leaf
);
1992 btrfs_free_path(path
);
1996 int btrfs_init_new_device(struct btrfs_root
*root
, char *device_path
)
1998 struct request_queue
*q
;
1999 struct btrfs_trans_handle
*trans
;
2000 struct btrfs_device
*device
;
2001 struct block_device
*bdev
;
2002 struct list_head
*devices
;
2003 struct super_block
*sb
= root
->fs_info
->sb
;
2004 struct rcu_string
*name
;
2006 int seeding_dev
= 0;
2009 if ((sb
->s_flags
& MS_RDONLY
) && !root
->fs_info
->fs_devices
->seeding
)
2012 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2013 root
->fs_info
->bdev_holder
);
2015 return PTR_ERR(bdev
);
2017 if (root
->fs_info
->fs_devices
->seeding
) {
2019 down_write(&sb
->s_umount
);
2020 mutex_lock(&uuid_mutex
);
2023 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2025 devices
= &root
->fs_info
->fs_devices
->devices
;
2027 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2028 list_for_each_entry(device
, devices
, dev_list
) {
2029 if (device
->bdev
== bdev
) {
2032 &root
->fs_info
->fs_devices
->device_list_mutex
);
2036 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2038 device
= btrfs_alloc_device(root
->fs_info
, NULL
, NULL
);
2039 if (IS_ERR(device
)) {
2040 /* we can safely leave the fs_devices entry around */
2041 ret
= PTR_ERR(device
);
2045 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2051 rcu_assign_pointer(device
->name
, name
);
2053 trans
= btrfs_start_transaction(root
, 0);
2054 if (IS_ERR(trans
)) {
2055 rcu_string_free(device
->name
);
2057 ret
= PTR_ERR(trans
);
2063 q
= bdev_get_queue(bdev
);
2064 if (blk_queue_discard(q
))
2065 device
->can_discard
= 1;
2066 device
->writeable
= 1;
2067 device
->generation
= trans
->transid
;
2068 device
->io_width
= root
->sectorsize
;
2069 device
->io_align
= root
->sectorsize
;
2070 device
->sector_size
= root
->sectorsize
;
2071 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
2072 device
->disk_total_bytes
= device
->total_bytes
;
2073 device
->dev_root
= root
->fs_info
->dev_root
;
2074 device
->bdev
= bdev
;
2075 device
->in_fs_metadata
= 1;
2076 device
->is_tgtdev_for_dev_replace
= 0;
2077 device
->mode
= FMODE_EXCL
;
2078 device
->dev_stats_valid
= 1;
2079 set_blocksize(device
->bdev
, 4096);
2082 sb
->s_flags
&= ~MS_RDONLY
;
2083 ret
= btrfs_prepare_sprout(root
);
2084 BUG_ON(ret
); /* -ENOMEM */
2087 device
->fs_devices
= root
->fs_info
->fs_devices
;
2089 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2090 list_add_rcu(&device
->dev_list
, &root
->fs_info
->fs_devices
->devices
);
2091 list_add(&device
->dev_alloc_list
,
2092 &root
->fs_info
->fs_devices
->alloc_list
);
2093 root
->fs_info
->fs_devices
->num_devices
++;
2094 root
->fs_info
->fs_devices
->open_devices
++;
2095 root
->fs_info
->fs_devices
->rw_devices
++;
2096 root
->fs_info
->fs_devices
->total_devices
++;
2097 if (device
->can_discard
)
2098 root
->fs_info
->fs_devices
->num_can_discard
++;
2099 root
->fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
2101 spin_lock(&root
->fs_info
->free_chunk_lock
);
2102 root
->fs_info
->free_chunk_space
+= device
->total_bytes
;
2103 spin_unlock(&root
->fs_info
->free_chunk_lock
);
2105 if (!blk_queue_nonrot(bdev_get_queue(bdev
)))
2106 root
->fs_info
->fs_devices
->rotating
= 1;
2108 total_bytes
= btrfs_super_total_bytes(root
->fs_info
->super_copy
);
2109 btrfs_set_super_total_bytes(root
->fs_info
->super_copy
,
2110 total_bytes
+ device
->total_bytes
);
2112 total_bytes
= btrfs_super_num_devices(root
->fs_info
->super_copy
);
2113 btrfs_set_super_num_devices(root
->fs_info
->super_copy
,
2115 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2118 ret
= init_first_rw_device(trans
, root
, device
);
2120 btrfs_abort_transaction(trans
, root
, ret
);
2123 ret
= btrfs_finish_sprout(trans
, root
);
2125 btrfs_abort_transaction(trans
, root
, ret
);
2129 ret
= btrfs_add_device(trans
, root
, device
);
2131 btrfs_abort_transaction(trans
, root
, ret
);
2137 * we've got more storage, clear any full flags on the space
2140 btrfs_clear_space_info_full(root
->fs_info
);
2142 unlock_chunks(root
);
2143 root
->fs_info
->num_tolerated_disk_barrier_failures
=
2144 btrfs_calc_num_tolerated_disk_barrier_failures(root
->fs_info
);
2145 ret
= btrfs_commit_transaction(trans
, root
);
2148 mutex_unlock(&uuid_mutex
);
2149 up_write(&sb
->s_umount
);
2151 if (ret
) /* transaction commit */
2154 ret
= btrfs_relocate_sys_chunks(root
);
2156 btrfs_error(root
->fs_info
, ret
,
2157 "Failed to relocate sys chunks after "
2158 "device initialization. This can be fixed "
2159 "using the \"btrfs balance\" command.");
2160 trans
= btrfs_attach_transaction(root
);
2161 if (IS_ERR(trans
)) {
2162 if (PTR_ERR(trans
) == -ENOENT
)
2164 return PTR_ERR(trans
);
2166 ret
= btrfs_commit_transaction(trans
, root
);
2169 /* Update ctime/mtime for libblkid */
2170 update_dev_time(device_path
);
2174 unlock_chunks(root
);
2175 btrfs_end_transaction(trans
, root
);
2176 rcu_string_free(device
->name
);
2179 blkdev_put(bdev
, FMODE_EXCL
);
2181 mutex_unlock(&uuid_mutex
);
2182 up_write(&sb
->s_umount
);
2187 int btrfs_init_dev_replace_tgtdev(struct btrfs_root
*root
, char *device_path
,
2188 struct btrfs_device
**device_out
)
2190 struct request_queue
*q
;
2191 struct btrfs_device
*device
;
2192 struct block_device
*bdev
;
2193 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2194 struct list_head
*devices
;
2195 struct rcu_string
*name
;
2196 u64 devid
= BTRFS_DEV_REPLACE_DEVID
;
2200 if (fs_info
->fs_devices
->seeding
)
2203 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2204 fs_info
->bdev_holder
);
2206 return PTR_ERR(bdev
);
2208 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2210 devices
= &fs_info
->fs_devices
->devices
;
2211 list_for_each_entry(device
, devices
, dev_list
) {
2212 if (device
->bdev
== bdev
) {
2218 device
= btrfs_alloc_device(NULL
, &devid
, NULL
);
2219 if (IS_ERR(device
)) {
2220 ret
= PTR_ERR(device
);
2224 name
= rcu_string_strdup(device_path
, GFP_NOFS
);
2230 rcu_assign_pointer(device
->name
, name
);
2232 q
= bdev_get_queue(bdev
);
2233 if (blk_queue_discard(q
))
2234 device
->can_discard
= 1;
2235 mutex_lock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2236 device
->writeable
= 1;
2237 device
->generation
= 0;
2238 device
->io_width
= root
->sectorsize
;
2239 device
->io_align
= root
->sectorsize
;
2240 device
->sector_size
= root
->sectorsize
;
2241 device
->total_bytes
= i_size_read(bdev
->bd_inode
);
2242 device
->disk_total_bytes
= device
->total_bytes
;
2243 device
->dev_root
= fs_info
->dev_root
;
2244 device
->bdev
= bdev
;
2245 device
->in_fs_metadata
= 1;
2246 device
->is_tgtdev_for_dev_replace
= 1;
2247 device
->mode
= FMODE_EXCL
;
2248 device
->dev_stats_valid
= 1;
2249 set_blocksize(device
->bdev
, 4096);
2250 device
->fs_devices
= fs_info
->fs_devices
;
2251 list_add(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2252 fs_info
->fs_devices
->num_devices
++;
2253 fs_info
->fs_devices
->open_devices
++;
2254 if (device
->can_discard
)
2255 fs_info
->fs_devices
->num_can_discard
++;
2256 mutex_unlock(&root
->fs_info
->fs_devices
->device_list_mutex
);
2258 *device_out
= device
;
2262 blkdev_put(bdev
, FMODE_EXCL
);
2266 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info
*fs_info
,
2267 struct btrfs_device
*tgtdev
)
2269 WARN_ON(fs_info
->fs_devices
->rw_devices
== 0);
2270 tgtdev
->io_width
= fs_info
->dev_root
->sectorsize
;
2271 tgtdev
->io_align
= fs_info
->dev_root
->sectorsize
;
2272 tgtdev
->sector_size
= fs_info
->dev_root
->sectorsize
;
2273 tgtdev
->dev_root
= fs_info
->dev_root
;
2274 tgtdev
->in_fs_metadata
= 1;
2277 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2278 struct btrfs_device
*device
)
2281 struct btrfs_path
*path
;
2282 struct btrfs_root
*root
;
2283 struct btrfs_dev_item
*dev_item
;
2284 struct extent_buffer
*leaf
;
2285 struct btrfs_key key
;
2287 root
= device
->dev_root
->fs_info
->chunk_root
;
2289 path
= btrfs_alloc_path();
2293 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2294 key
.type
= BTRFS_DEV_ITEM_KEY
;
2295 key
.offset
= device
->devid
;
2297 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2306 leaf
= path
->nodes
[0];
2307 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2309 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2310 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2311 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2312 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2313 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2314 btrfs_set_device_total_bytes(leaf
, dev_item
, device
->disk_total_bytes
);
2315 btrfs_set_device_bytes_used(leaf
, dev_item
, device
->bytes_used
);
2316 btrfs_mark_buffer_dirty(leaf
);
2319 btrfs_free_path(path
);
2323 static int __btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2324 struct btrfs_device
*device
, u64 new_size
)
2326 struct btrfs_super_block
*super_copy
=
2327 device
->dev_root
->fs_info
->super_copy
;
2328 u64 old_total
= btrfs_super_total_bytes(super_copy
);
2329 u64 diff
= new_size
- device
->total_bytes
;
2331 if (!device
->writeable
)
2333 if (new_size
<= device
->total_bytes
||
2334 device
->is_tgtdev_for_dev_replace
)
2337 btrfs_set_super_total_bytes(super_copy
, old_total
+ diff
);
2338 device
->fs_devices
->total_rw_bytes
+= diff
;
2340 device
->total_bytes
= new_size
;
2341 device
->disk_total_bytes
= new_size
;
2342 btrfs_clear_space_info_full(device
->dev_root
->fs_info
);
2344 return btrfs_update_device(trans
, device
);
2347 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2348 struct btrfs_device
*device
, u64 new_size
)
2351 lock_chunks(device
->dev_root
);
2352 ret
= __btrfs_grow_device(trans
, device
, new_size
);
2353 unlock_chunks(device
->dev_root
);
2357 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
2358 struct btrfs_root
*root
,
2359 u64 chunk_tree
, u64 chunk_objectid
,
2363 struct btrfs_path
*path
;
2364 struct btrfs_key key
;
2366 root
= root
->fs_info
->chunk_root
;
2367 path
= btrfs_alloc_path();
2371 key
.objectid
= chunk_objectid
;
2372 key
.offset
= chunk_offset
;
2373 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2375 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2378 else if (ret
> 0) { /* Logic error or corruption */
2379 btrfs_error(root
->fs_info
, -ENOENT
,
2380 "Failed lookup while freeing chunk.");
2385 ret
= btrfs_del_item(trans
, root
, path
);
2387 btrfs_error(root
->fs_info
, ret
,
2388 "Failed to delete chunk item.");
2390 btrfs_free_path(path
);
2394 static int btrfs_del_sys_chunk(struct btrfs_root
*root
, u64 chunk_objectid
, u64
2397 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
2398 struct btrfs_disk_key
*disk_key
;
2399 struct btrfs_chunk
*chunk
;
2406 struct btrfs_key key
;
2408 array_size
= btrfs_super_sys_array_size(super_copy
);
2410 ptr
= super_copy
->sys_chunk_array
;
2413 while (cur
< array_size
) {
2414 disk_key
= (struct btrfs_disk_key
*)ptr
;
2415 btrfs_disk_key_to_cpu(&key
, disk_key
);
2417 len
= sizeof(*disk_key
);
2419 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2420 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2421 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2422 len
+= btrfs_chunk_item_size(num_stripes
);
2427 if (key
.objectid
== chunk_objectid
&&
2428 key
.offset
== chunk_offset
) {
2429 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2431 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2440 static int btrfs_relocate_chunk(struct btrfs_root
*root
,
2441 u64 chunk_tree
, u64 chunk_objectid
,
2444 struct extent_map_tree
*em_tree
;
2445 struct btrfs_root
*extent_root
;
2446 struct btrfs_trans_handle
*trans
;
2447 struct extent_map
*em
;
2448 struct map_lookup
*map
;
2452 root
= root
->fs_info
->chunk_root
;
2453 extent_root
= root
->fs_info
->extent_root
;
2454 em_tree
= &root
->fs_info
->mapping_tree
.map_tree
;
2456 ret
= btrfs_can_relocate(extent_root
, chunk_offset
);
2460 /* step one, relocate all the extents inside this chunk */
2461 ret
= btrfs_relocate_block_group(extent_root
, chunk_offset
);
2465 trans
= btrfs_start_transaction(root
, 0);
2466 if (IS_ERR(trans
)) {
2467 ret
= PTR_ERR(trans
);
2468 btrfs_std_error(root
->fs_info
, ret
);
2475 * step two, delete the device extents and the
2476 * chunk tree entries
2478 read_lock(&em_tree
->lock
);
2479 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
2480 read_unlock(&em_tree
->lock
);
2482 BUG_ON(!em
|| em
->start
> chunk_offset
||
2483 em
->start
+ em
->len
< chunk_offset
);
2484 map
= (struct map_lookup
*)em
->bdev
;
2486 for (i
= 0; i
< map
->num_stripes
; i
++) {
2487 ret
= btrfs_free_dev_extent(trans
, map
->stripes
[i
].dev
,
2488 map
->stripes
[i
].physical
);
2491 if (map
->stripes
[i
].dev
) {
2492 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2496 ret
= btrfs_free_chunk(trans
, root
, chunk_tree
, chunk_objectid
,
2501 trace_btrfs_chunk_free(root
, map
, chunk_offset
, em
->len
);
2503 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2504 ret
= btrfs_del_sys_chunk(root
, chunk_objectid
, chunk_offset
);
2508 ret
= btrfs_remove_block_group(trans
, extent_root
, chunk_offset
);
2511 write_lock(&em_tree
->lock
);
2512 remove_extent_mapping(em_tree
, em
);
2513 write_unlock(&em_tree
->lock
);
2518 /* once for the tree */
2519 free_extent_map(em
);
2521 free_extent_map(em
);
2523 unlock_chunks(root
);
2524 btrfs_end_transaction(trans
, root
);
2528 static int btrfs_relocate_sys_chunks(struct btrfs_root
*root
)
2530 struct btrfs_root
*chunk_root
= root
->fs_info
->chunk_root
;
2531 struct btrfs_path
*path
;
2532 struct extent_buffer
*leaf
;
2533 struct btrfs_chunk
*chunk
;
2534 struct btrfs_key key
;
2535 struct btrfs_key found_key
;
2536 u64 chunk_tree
= chunk_root
->root_key
.objectid
;
2538 bool retried
= false;
2542 path
= btrfs_alloc_path();
2547 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2548 key
.offset
= (u64
)-1;
2549 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2552 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2555 BUG_ON(ret
== 0); /* Corruption */
2557 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
2564 leaf
= path
->nodes
[0];
2565 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2567 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
2568 struct btrfs_chunk
);
2569 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2570 btrfs_release_path(path
);
2572 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2573 ret
= btrfs_relocate_chunk(chunk_root
, chunk_tree
,
2582 if (found_key
.offset
== 0)
2584 key
.offset
= found_key
.offset
- 1;
2587 if (failed
&& !retried
) {
2591 } else if (WARN_ON(failed
&& retried
)) {
2595 btrfs_free_path(path
);
2599 static int insert_balance_item(struct btrfs_root
*root
,
2600 struct btrfs_balance_control
*bctl
)
2602 struct btrfs_trans_handle
*trans
;
2603 struct btrfs_balance_item
*item
;
2604 struct btrfs_disk_balance_args disk_bargs
;
2605 struct btrfs_path
*path
;
2606 struct extent_buffer
*leaf
;
2607 struct btrfs_key key
;
2610 path
= btrfs_alloc_path();
2614 trans
= btrfs_start_transaction(root
, 0);
2615 if (IS_ERR(trans
)) {
2616 btrfs_free_path(path
);
2617 return PTR_ERR(trans
);
2620 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2621 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2624 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2629 leaf
= path
->nodes
[0];
2630 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
2632 memset_extent_buffer(leaf
, 0, (unsigned long)item
, sizeof(*item
));
2634 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
2635 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
2636 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
2637 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
2638 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
2639 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
2641 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
2643 btrfs_mark_buffer_dirty(leaf
);
2645 btrfs_free_path(path
);
2646 err
= btrfs_commit_transaction(trans
, root
);
2652 static int del_balance_item(struct btrfs_root
*root
)
2654 struct btrfs_trans_handle
*trans
;
2655 struct btrfs_path
*path
;
2656 struct btrfs_key key
;
2659 path
= btrfs_alloc_path();
2663 trans
= btrfs_start_transaction(root
, 0);
2664 if (IS_ERR(trans
)) {
2665 btrfs_free_path(path
);
2666 return PTR_ERR(trans
);
2669 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
2670 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
2673 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2681 ret
= btrfs_del_item(trans
, root
, path
);
2683 btrfs_free_path(path
);
2684 err
= btrfs_commit_transaction(trans
, root
);
2691 * This is a heuristic used to reduce the number of chunks balanced on
2692 * resume after balance was interrupted.
2694 static void update_balance_args(struct btrfs_balance_control
*bctl
)
2697 * Turn on soft mode for chunk types that were being converted.
2699 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2700 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2701 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2702 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2703 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
2704 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
2707 * Turn on usage filter if is not already used. The idea is
2708 * that chunks that we have already balanced should be
2709 * reasonably full. Don't do it for chunks that are being
2710 * converted - that will keep us from relocating unconverted
2711 * (albeit full) chunks.
2713 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2714 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2715 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2716 bctl
->data
.usage
= 90;
2718 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2719 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2720 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2721 bctl
->sys
.usage
= 90;
2723 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2724 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
2725 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
2726 bctl
->meta
.usage
= 90;
2731 * Should be called with both balance and volume mutexes held to
2732 * serialize other volume operations (add_dev/rm_dev/resize) with
2733 * restriper. Same goes for unset_balance_control.
2735 static void set_balance_control(struct btrfs_balance_control
*bctl
)
2737 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
2739 BUG_ON(fs_info
->balance_ctl
);
2741 spin_lock(&fs_info
->balance_lock
);
2742 fs_info
->balance_ctl
= bctl
;
2743 spin_unlock(&fs_info
->balance_lock
);
2746 static void unset_balance_control(struct btrfs_fs_info
*fs_info
)
2748 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
2750 BUG_ON(!fs_info
->balance_ctl
);
2752 spin_lock(&fs_info
->balance_lock
);
2753 fs_info
->balance_ctl
= NULL
;
2754 spin_unlock(&fs_info
->balance_lock
);
2760 * Balance filters. Return 1 if chunk should be filtered out
2761 * (should not be balanced).
2763 static int chunk_profiles_filter(u64 chunk_type
,
2764 struct btrfs_balance_args
*bargs
)
2766 chunk_type
= chunk_to_extended(chunk_type
) &
2767 BTRFS_EXTENDED_PROFILE_MASK
;
2769 if (bargs
->profiles
& chunk_type
)
2775 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
2776 struct btrfs_balance_args
*bargs
)
2778 struct btrfs_block_group_cache
*cache
;
2779 u64 chunk_used
, user_thresh
;
2782 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
2783 chunk_used
= btrfs_block_group_used(&cache
->item
);
2785 if (bargs
->usage
== 0)
2787 else if (bargs
->usage
> 100)
2788 user_thresh
= cache
->key
.offset
;
2790 user_thresh
= div_factor_fine(cache
->key
.offset
,
2793 if (chunk_used
< user_thresh
)
2796 btrfs_put_block_group(cache
);
2800 static int chunk_devid_filter(struct extent_buffer
*leaf
,
2801 struct btrfs_chunk
*chunk
,
2802 struct btrfs_balance_args
*bargs
)
2804 struct btrfs_stripe
*stripe
;
2805 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2808 for (i
= 0; i
< num_stripes
; i
++) {
2809 stripe
= btrfs_stripe_nr(chunk
, i
);
2810 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
2817 /* [pstart, pend) */
2818 static int chunk_drange_filter(struct extent_buffer
*leaf
,
2819 struct btrfs_chunk
*chunk
,
2821 struct btrfs_balance_args
*bargs
)
2823 struct btrfs_stripe
*stripe
;
2824 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
2830 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
2833 if (btrfs_chunk_type(leaf
, chunk
) & (BTRFS_BLOCK_GROUP_DUP
|
2834 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)) {
2835 factor
= num_stripes
/ 2;
2836 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID5
) {
2837 factor
= num_stripes
- 1;
2838 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID6
) {
2839 factor
= num_stripes
- 2;
2841 factor
= num_stripes
;
2844 for (i
= 0; i
< num_stripes
; i
++) {
2845 stripe
= btrfs_stripe_nr(chunk
, i
);
2846 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
2849 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
2850 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
2851 do_div(stripe_length
, factor
);
2853 if (stripe_offset
< bargs
->pend
&&
2854 stripe_offset
+ stripe_length
> bargs
->pstart
)
2861 /* [vstart, vend) */
2862 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
2863 struct btrfs_chunk
*chunk
,
2865 struct btrfs_balance_args
*bargs
)
2867 if (chunk_offset
< bargs
->vend
&&
2868 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
2869 /* at least part of the chunk is inside this vrange */
2875 static int chunk_soft_convert_filter(u64 chunk_type
,
2876 struct btrfs_balance_args
*bargs
)
2878 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
2881 chunk_type
= chunk_to_extended(chunk_type
) &
2882 BTRFS_EXTENDED_PROFILE_MASK
;
2884 if (bargs
->target
== chunk_type
)
2890 static int should_balance_chunk(struct btrfs_root
*root
,
2891 struct extent_buffer
*leaf
,
2892 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
2894 struct btrfs_balance_control
*bctl
= root
->fs_info
->balance_ctl
;
2895 struct btrfs_balance_args
*bargs
= NULL
;
2896 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2899 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
2900 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
2904 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
2905 bargs
= &bctl
->data
;
2906 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
2908 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
2909 bargs
= &bctl
->meta
;
2911 /* profiles filter */
2912 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
2913 chunk_profiles_filter(chunk_type
, bargs
)) {
2918 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
2919 chunk_usage_filter(bctl
->fs_info
, chunk_offset
, bargs
)) {
2924 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
2925 chunk_devid_filter(leaf
, chunk
, bargs
)) {
2929 /* drange filter, makes sense only with devid filter */
2930 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
2931 chunk_drange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
2936 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
2937 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
2941 /* soft profile changing mode */
2942 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
2943 chunk_soft_convert_filter(chunk_type
, bargs
)) {
2948 * limited by count, must be the last filter
2950 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT
)) {
2951 if (bargs
->limit
== 0)
2960 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
2962 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
2963 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
2964 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
2965 struct list_head
*devices
;
2966 struct btrfs_device
*device
;
2969 struct btrfs_chunk
*chunk
;
2970 struct btrfs_path
*path
;
2971 struct btrfs_key key
;
2972 struct btrfs_key found_key
;
2973 struct btrfs_trans_handle
*trans
;
2974 struct extent_buffer
*leaf
;
2977 int enospc_errors
= 0;
2978 bool counting
= true;
2979 u64 limit_data
= bctl
->data
.limit
;
2980 u64 limit_meta
= bctl
->meta
.limit
;
2981 u64 limit_sys
= bctl
->sys
.limit
;
2983 /* step one make some room on all the devices */
2984 devices
= &fs_info
->fs_devices
->devices
;
2985 list_for_each_entry(device
, devices
, dev_list
) {
2986 old_size
= device
->total_bytes
;
2987 size_to_free
= div_factor(old_size
, 1);
2988 size_to_free
= min(size_to_free
, (u64
)1 * 1024 * 1024);
2989 if (!device
->writeable
||
2990 device
->total_bytes
- device
->bytes_used
> size_to_free
||
2991 device
->is_tgtdev_for_dev_replace
)
2994 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
2999 trans
= btrfs_start_transaction(dev_root
, 0);
3000 BUG_ON(IS_ERR(trans
));
3002 ret
= btrfs_grow_device(trans
, device
, old_size
);
3005 btrfs_end_transaction(trans
, dev_root
);
3008 /* step two, relocate all the chunks */
3009 path
= btrfs_alloc_path();
3015 /* zero out stat counters */
3016 spin_lock(&fs_info
->balance_lock
);
3017 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
3018 spin_unlock(&fs_info
->balance_lock
);
3021 bctl
->data
.limit
= limit_data
;
3022 bctl
->meta
.limit
= limit_meta
;
3023 bctl
->sys
.limit
= limit_sys
;
3025 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3026 key
.offset
= (u64
)-1;
3027 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3030 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
3031 atomic_read(&fs_info
->balance_cancel_req
)) {
3036 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
3041 * this shouldn't happen, it means the last relocate
3045 BUG(); /* FIXME break ? */
3047 ret
= btrfs_previous_item(chunk_root
, path
, 0,
3048 BTRFS_CHUNK_ITEM_KEY
);
3054 leaf
= path
->nodes
[0];
3055 slot
= path
->slots
[0];
3056 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3058 if (found_key
.objectid
!= key
.objectid
)
3061 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
3064 spin_lock(&fs_info
->balance_lock
);
3065 bctl
->stat
.considered
++;
3066 spin_unlock(&fs_info
->balance_lock
);
3069 ret
= should_balance_chunk(chunk_root
, leaf
, chunk
,
3071 btrfs_release_path(path
);
3076 spin_lock(&fs_info
->balance_lock
);
3077 bctl
->stat
.expected
++;
3078 spin_unlock(&fs_info
->balance_lock
);
3082 ret
= btrfs_relocate_chunk(chunk_root
,
3083 chunk_root
->root_key
.objectid
,
3086 if (ret
&& ret
!= -ENOSPC
)
3088 if (ret
== -ENOSPC
) {
3091 spin_lock(&fs_info
->balance_lock
);
3092 bctl
->stat
.completed
++;
3093 spin_unlock(&fs_info
->balance_lock
);
3096 if (found_key
.offset
== 0)
3098 key
.offset
= found_key
.offset
- 1;
3102 btrfs_release_path(path
);
3107 btrfs_free_path(path
);
3108 if (enospc_errors
) {
3109 btrfs_info(fs_info
, "%d enospc errors during balance",
3119 * alloc_profile_is_valid - see if a given profile is valid and reduced
3120 * @flags: profile to validate
3121 * @extended: if true @flags is treated as an extended profile
3123 static int alloc_profile_is_valid(u64 flags
, int extended
)
3125 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
3126 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
3128 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
3130 /* 1) check that all other bits are zeroed */
3134 /* 2) see if profile is reduced */
3136 return !extended
; /* "0" is valid for usual profiles */
3138 /* true if exactly one bit set */
3139 return (flags
& (flags
- 1)) == 0;
3142 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
3144 /* cancel requested || normal exit path */
3145 return atomic_read(&fs_info
->balance_cancel_req
) ||
3146 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
3147 atomic_read(&fs_info
->balance_cancel_req
) == 0);
3150 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
3154 unset_balance_control(fs_info
);
3155 ret
= del_balance_item(fs_info
->tree_root
);
3157 btrfs_std_error(fs_info
, ret
);
3159 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3163 * Should be called with both balance and volume mutexes held
3165 int btrfs_balance(struct btrfs_balance_control
*bctl
,
3166 struct btrfs_ioctl_balance_args
*bargs
)
3168 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3175 if (btrfs_fs_closing(fs_info
) ||
3176 atomic_read(&fs_info
->balance_pause_req
) ||
3177 atomic_read(&fs_info
->balance_cancel_req
)) {
3182 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
3183 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
3187 * In case of mixed groups both data and meta should be picked,
3188 * and identical options should be given for both of them.
3190 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
3191 if (mixed
&& (bctl
->flags
& allowed
)) {
3192 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
3193 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
3194 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
3195 btrfs_err(fs_info
, "with mixed groups data and "
3196 "metadata balance options must be the same");
3202 num_devices
= fs_info
->fs_devices
->num_devices
;
3203 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
3204 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
3205 BUG_ON(num_devices
< 1);
3208 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
3209 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
3210 if (num_devices
== 1)
3211 allowed
|= BTRFS_BLOCK_GROUP_DUP
;
3212 else if (num_devices
> 1)
3213 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
3214 if (num_devices
> 2)
3215 allowed
|= BTRFS_BLOCK_GROUP_RAID5
;
3216 if (num_devices
> 3)
3217 allowed
|= (BTRFS_BLOCK_GROUP_RAID10
|
3218 BTRFS_BLOCK_GROUP_RAID6
);
3219 if ((bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3220 (!alloc_profile_is_valid(bctl
->data
.target
, 1) ||
3221 (bctl
->data
.target
& ~allowed
))) {
3222 btrfs_err(fs_info
, "unable to start balance with target "
3223 "data profile %llu",
3228 if ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3229 (!alloc_profile_is_valid(bctl
->meta
.target
, 1) ||
3230 (bctl
->meta
.target
& ~allowed
))) {
3232 "unable to start balance with target metadata profile %llu",
3237 if ((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3238 (!alloc_profile_is_valid(bctl
->sys
.target
, 1) ||
3239 (bctl
->sys
.target
& ~allowed
))) {
3241 "unable to start balance with target system profile %llu",
3247 /* allow dup'ed data chunks only in mixed mode */
3248 if (!mixed
&& (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3249 (bctl
->data
.target
& BTRFS_BLOCK_GROUP_DUP
)) {
3250 btrfs_err(fs_info
, "dup for data is not allowed");
3255 /* allow to reduce meta or sys integrity only if force set */
3256 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3257 BTRFS_BLOCK_GROUP_RAID10
|
3258 BTRFS_BLOCK_GROUP_RAID5
|
3259 BTRFS_BLOCK_GROUP_RAID6
;
3261 seq
= read_seqbegin(&fs_info
->profiles_lock
);
3263 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3264 (fs_info
->avail_system_alloc_bits
& allowed
) &&
3265 !(bctl
->sys
.target
& allowed
)) ||
3266 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3267 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
3268 !(bctl
->meta
.target
& allowed
))) {
3269 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
3270 btrfs_info(fs_info
, "force reducing metadata integrity");
3272 btrfs_err(fs_info
, "balance will reduce metadata "
3273 "integrity, use force if you want this");
3278 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
3280 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3281 int num_tolerated_disk_barrier_failures
;
3282 u64 target
= bctl
->sys
.target
;
3284 num_tolerated_disk_barrier_failures
=
3285 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3286 if (num_tolerated_disk_barrier_failures
> 0 &&
3288 (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID0
|
3289 BTRFS_AVAIL_ALLOC_BIT_SINGLE
)))
3290 num_tolerated_disk_barrier_failures
= 0;
3291 else if (num_tolerated_disk_barrier_failures
> 1 &&
3293 (BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)))
3294 num_tolerated_disk_barrier_failures
= 1;
3296 fs_info
->num_tolerated_disk_barrier_failures
=
3297 num_tolerated_disk_barrier_failures
;
3300 ret
= insert_balance_item(fs_info
->tree_root
, bctl
);
3301 if (ret
&& ret
!= -EEXIST
)
3304 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
3305 BUG_ON(ret
== -EEXIST
);
3306 set_balance_control(bctl
);
3308 BUG_ON(ret
!= -EEXIST
);
3309 spin_lock(&fs_info
->balance_lock
);
3310 update_balance_args(bctl
);
3311 spin_unlock(&fs_info
->balance_lock
);
3314 atomic_inc(&fs_info
->balance_running
);
3315 mutex_unlock(&fs_info
->balance_mutex
);
3317 ret
= __btrfs_balance(fs_info
);
3319 mutex_lock(&fs_info
->balance_mutex
);
3320 atomic_dec(&fs_info
->balance_running
);
3322 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) {
3323 fs_info
->num_tolerated_disk_barrier_failures
=
3324 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info
);
3328 memset(bargs
, 0, sizeof(*bargs
));
3329 update_ioctl_balance_args(fs_info
, 0, bargs
);
3332 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
3333 balance_need_close(fs_info
)) {
3334 __cancel_balance(fs_info
);
3337 wake_up(&fs_info
->balance_wait_q
);
3341 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
3342 __cancel_balance(fs_info
);
3345 atomic_set(&fs_info
->mutually_exclusive_operation_running
, 0);
3350 static int balance_kthread(void *data
)
3352 struct btrfs_fs_info
*fs_info
= data
;
3355 mutex_lock(&fs_info
->volume_mutex
);
3356 mutex_lock(&fs_info
->balance_mutex
);
3358 if (fs_info
->balance_ctl
) {
3359 btrfs_info(fs_info
, "continuing balance");
3360 ret
= btrfs_balance(fs_info
->balance_ctl
, NULL
);
3363 mutex_unlock(&fs_info
->balance_mutex
);
3364 mutex_unlock(&fs_info
->volume_mutex
);
3369 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
3371 struct task_struct
*tsk
;
3373 spin_lock(&fs_info
->balance_lock
);
3374 if (!fs_info
->balance_ctl
) {
3375 spin_unlock(&fs_info
->balance_lock
);
3378 spin_unlock(&fs_info
->balance_lock
);
3380 if (btrfs_test_opt(fs_info
->tree_root
, SKIP_BALANCE
)) {
3381 btrfs_info(fs_info
, "force skipping balance");
3385 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
3386 return PTR_ERR_OR_ZERO(tsk
);
3389 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
3391 struct btrfs_balance_control
*bctl
;
3392 struct btrfs_balance_item
*item
;
3393 struct btrfs_disk_balance_args disk_bargs
;
3394 struct btrfs_path
*path
;
3395 struct extent_buffer
*leaf
;
3396 struct btrfs_key key
;
3399 path
= btrfs_alloc_path();
3403 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3404 key
.type
= BTRFS_BALANCE_ITEM_KEY
;
3407 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
3410 if (ret
> 0) { /* ret = -ENOENT; */
3415 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
3421 leaf
= path
->nodes
[0];
3422 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3424 bctl
->fs_info
= fs_info
;
3425 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
3426 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
3428 btrfs_balance_data(leaf
, item
, &disk_bargs
);
3429 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
3430 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
3431 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
3432 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
3433 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
3435 WARN_ON(atomic_xchg(&fs_info
->mutually_exclusive_operation_running
, 1));
3437 mutex_lock(&fs_info
->volume_mutex
);
3438 mutex_lock(&fs_info
->balance_mutex
);
3440 set_balance_control(bctl
);
3442 mutex_unlock(&fs_info
->balance_mutex
);
3443 mutex_unlock(&fs_info
->volume_mutex
);
3445 btrfs_free_path(path
);
3449 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
3453 mutex_lock(&fs_info
->balance_mutex
);
3454 if (!fs_info
->balance_ctl
) {
3455 mutex_unlock(&fs_info
->balance_mutex
);
3459 if (atomic_read(&fs_info
->balance_running
)) {
3460 atomic_inc(&fs_info
->balance_pause_req
);
3461 mutex_unlock(&fs_info
->balance_mutex
);
3463 wait_event(fs_info
->balance_wait_q
,
3464 atomic_read(&fs_info
->balance_running
) == 0);
3466 mutex_lock(&fs_info
->balance_mutex
);
3467 /* we are good with balance_ctl ripped off from under us */
3468 BUG_ON(atomic_read(&fs_info
->balance_running
));
3469 atomic_dec(&fs_info
->balance_pause_req
);
3474 mutex_unlock(&fs_info
->balance_mutex
);
3478 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
3480 if (fs_info
->sb
->s_flags
& MS_RDONLY
)
3483 mutex_lock(&fs_info
->balance_mutex
);
3484 if (!fs_info
->balance_ctl
) {
3485 mutex_unlock(&fs_info
->balance_mutex
);
3489 atomic_inc(&fs_info
->balance_cancel_req
);
3491 * if we are running just wait and return, balance item is
3492 * deleted in btrfs_balance in this case
3494 if (atomic_read(&fs_info
->balance_running
)) {
3495 mutex_unlock(&fs_info
->balance_mutex
);
3496 wait_event(fs_info
->balance_wait_q
,
3497 atomic_read(&fs_info
->balance_running
) == 0);
3498 mutex_lock(&fs_info
->balance_mutex
);
3500 /* __cancel_balance needs volume_mutex */
3501 mutex_unlock(&fs_info
->balance_mutex
);
3502 mutex_lock(&fs_info
->volume_mutex
);
3503 mutex_lock(&fs_info
->balance_mutex
);
3505 if (fs_info
->balance_ctl
)
3506 __cancel_balance(fs_info
);
3508 mutex_unlock(&fs_info
->volume_mutex
);
3511 BUG_ON(fs_info
->balance_ctl
|| atomic_read(&fs_info
->balance_running
));
3512 atomic_dec(&fs_info
->balance_cancel_req
);
3513 mutex_unlock(&fs_info
->balance_mutex
);
3517 static int btrfs_uuid_scan_kthread(void *data
)
3519 struct btrfs_fs_info
*fs_info
= data
;
3520 struct btrfs_root
*root
= fs_info
->tree_root
;
3521 struct btrfs_key key
;
3522 struct btrfs_key max_key
;
3523 struct btrfs_path
*path
= NULL
;
3525 struct extent_buffer
*eb
;
3527 struct btrfs_root_item root_item
;
3529 struct btrfs_trans_handle
*trans
= NULL
;
3531 path
= btrfs_alloc_path();
3538 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3541 max_key
.objectid
= (u64
)-1;
3542 max_key
.type
= BTRFS_ROOT_ITEM_KEY
;
3543 max_key
.offset
= (u64
)-1;
3545 path
->keep_locks
= 1;
3548 ret
= btrfs_search_forward(root
, &key
, path
, 0);
3555 if (key
.type
!= BTRFS_ROOT_ITEM_KEY
||
3556 (key
.objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
3557 key
.objectid
!= BTRFS_FS_TREE_OBJECTID
) ||
3558 key
.objectid
> BTRFS_LAST_FREE_OBJECTID
)
3561 eb
= path
->nodes
[0];
3562 slot
= path
->slots
[0];
3563 item_size
= btrfs_item_size_nr(eb
, slot
);
3564 if (item_size
< sizeof(root_item
))
3567 read_extent_buffer(eb
, &root_item
,
3568 btrfs_item_ptr_offset(eb
, slot
),
3569 (int)sizeof(root_item
));
3570 if (btrfs_root_refs(&root_item
) == 0)
3573 if (!btrfs_is_empty_uuid(root_item
.uuid
) ||
3574 !btrfs_is_empty_uuid(root_item
.received_uuid
)) {
3578 btrfs_release_path(path
);
3580 * 1 - subvol uuid item
3581 * 1 - received_subvol uuid item
3583 trans
= btrfs_start_transaction(fs_info
->uuid_root
, 2);
3584 if (IS_ERR(trans
)) {
3585 ret
= PTR_ERR(trans
);
3593 if (!btrfs_is_empty_uuid(root_item
.uuid
)) {
3594 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
3596 BTRFS_UUID_KEY_SUBVOL
,
3599 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
3605 if (!btrfs_is_empty_uuid(root_item
.received_uuid
)) {
3606 ret
= btrfs_uuid_tree_add(trans
, fs_info
->uuid_root
,
3607 root_item
.received_uuid
,
3608 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
3611 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
3619 ret
= btrfs_end_transaction(trans
, fs_info
->uuid_root
);
3625 btrfs_release_path(path
);
3626 if (key
.offset
< (u64
)-1) {
3628 } else if (key
.type
< BTRFS_ROOT_ITEM_KEY
) {
3630 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3631 } else if (key
.objectid
< (u64
)-1) {
3633 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3642 btrfs_free_path(path
);
3643 if (trans
&& !IS_ERR(trans
))
3644 btrfs_end_transaction(trans
, fs_info
->uuid_root
);
3646 btrfs_warn(fs_info
, "btrfs_uuid_scan_kthread failed %d", ret
);
3648 fs_info
->update_uuid_tree_gen
= 1;
3649 up(&fs_info
->uuid_tree_rescan_sem
);
3654 * Callback for btrfs_uuid_tree_iterate().
3656 * 0 check succeeded, the entry is not outdated.
3657 * < 0 if an error occured.
3658 * > 0 if the check failed, which means the caller shall remove the entry.
3660 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info
*fs_info
,
3661 u8
*uuid
, u8 type
, u64 subid
)
3663 struct btrfs_key key
;
3665 struct btrfs_root
*subvol_root
;
3667 if (type
!= BTRFS_UUID_KEY_SUBVOL
&&
3668 type
!= BTRFS_UUID_KEY_RECEIVED_SUBVOL
)
3671 key
.objectid
= subid
;
3672 key
.type
= BTRFS_ROOT_ITEM_KEY
;
3673 key
.offset
= (u64
)-1;
3674 subvol_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
3675 if (IS_ERR(subvol_root
)) {
3676 ret
= PTR_ERR(subvol_root
);
3683 case BTRFS_UUID_KEY_SUBVOL
:
3684 if (memcmp(uuid
, subvol_root
->root_item
.uuid
, BTRFS_UUID_SIZE
))
3687 case BTRFS_UUID_KEY_RECEIVED_SUBVOL
:
3688 if (memcmp(uuid
, subvol_root
->root_item
.received_uuid
,
3698 static int btrfs_uuid_rescan_kthread(void *data
)
3700 struct btrfs_fs_info
*fs_info
= (struct btrfs_fs_info
*)data
;
3704 * 1st step is to iterate through the existing UUID tree and
3705 * to delete all entries that contain outdated data.
3706 * 2nd step is to add all missing entries to the UUID tree.
3708 ret
= btrfs_uuid_tree_iterate(fs_info
, btrfs_check_uuid_tree_entry
);
3710 btrfs_warn(fs_info
, "iterating uuid_tree failed %d", ret
);
3711 up(&fs_info
->uuid_tree_rescan_sem
);
3714 return btrfs_uuid_scan_kthread(data
);
3717 int btrfs_create_uuid_tree(struct btrfs_fs_info
*fs_info
)
3719 struct btrfs_trans_handle
*trans
;
3720 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
3721 struct btrfs_root
*uuid_root
;
3722 struct task_struct
*task
;
3729 trans
= btrfs_start_transaction(tree_root
, 2);
3731 return PTR_ERR(trans
);
3733 uuid_root
= btrfs_create_tree(trans
, fs_info
,
3734 BTRFS_UUID_TREE_OBJECTID
);
3735 if (IS_ERR(uuid_root
)) {
3736 btrfs_abort_transaction(trans
, tree_root
,
3737 PTR_ERR(uuid_root
));
3738 return PTR_ERR(uuid_root
);
3741 fs_info
->uuid_root
= uuid_root
;
3743 ret
= btrfs_commit_transaction(trans
, tree_root
);
3747 down(&fs_info
->uuid_tree_rescan_sem
);
3748 task
= kthread_run(btrfs_uuid_scan_kthread
, fs_info
, "btrfs-uuid");
3750 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3751 btrfs_warn(fs_info
, "failed to start uuid_scan task");
3752 up(&fs_info
->uuid_tree_rescan_sem
);
3753 return PTR_ERR(task
);
3759 int btrfs_check_uuid_tree(struct btrfs_fs_info
*fs_info
)
3761 struct task_struct
*task
;
3763 down(&fs_info
->uuid_tree_rescan_sem
);
3764 task
= kthread_run(btrfs_uuid_rescan_kthread
, fs_info
, "btrfs-uuid");
3766 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3767 btrfs_warn(fs_info
, "failed to start uuid_rescan task");
3768 up(&fs_info
->uuid_tree_rescan_sem
);
3769 return PTR_ERR(task
);
3776 * shrinking a device means finding all of the device extents past
3777 * the new size, and then following the back refs to the chunks.
3778 * The chunk relocation code actually frees the device extent
3780 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
3782 struct btrfs_trans_handle
*trans
;
3783 struct btrfs_root
*root
= device
->dev_root
;
3784 struct btrfs_dev_extent
*dev_extent
= NULL
;
3785 struct btrfs_path
*path
;
3793 bool retried
= false;
3794 struct extent_buffer
*l
;
3795 struct btrfs_key key
;
3796 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
3797 u64 old_total
= btrfs_super_total_bytes(super_copy
);
3798 u64 old_size
= device
->total_bytes
;
3799 u64 diff
= device
->total_bytes
- new_size
;
3801 if (device
->is_tgtdev_for_dev_replace
)
3804 path
= btrfs_alloc_path();
3812 device
->total_bytes
= new_size
;
3813 if (device
->writeable
) {
3814 device
->fs_devices
->total_rw_bytes
-= diff
;
3815 spin_lock(&root
->fs_info
->free_chunk_lock
);
3816 root
->fs_info
->free_chunk_space
-= diff
;
3817 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3819 unlock_chunks(root
);
3822 key
.objectid
= device
->devid
;
3823 key
.offset
= (u64
)-1;
3824 key
.type
= BTRFS_DEV_EXTENT_KEY
;
3827 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3831 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
3836 btrfs_release_path(path
);
3841 slot
= path
->slots
[0];
3842 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
3844 if (key
.objectid
!= device
->devid
) {
3845 btrfs_release_path(path
);
3849 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
3850 length
= btrfs_dev_extent_length(l
, dev_extent
);
3852 if (key
.offset
+ length
<= new_size
) {
3853 btrfs_release_path(path
);
3857 chunk_tree
= btrfs_dev_extent_chunk_tree(l
, dev_extent
);
3858 chunk_objectid
= btrfs_dev_extent_chunk_objectid(l
, dev_extent
);
3859 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
3860 btrfs_release_path(path
);
3862 ret
= btrfs_relocate_chunk(root
, chunk_tree
, chunk_objectid
,
3864 if (ret
&& ret
!= -ENOSPC
)
3868 } while (key
.offset
-- > 0);
3870 if (failed
&& !retried
) {
3874 } else if (failed
&& retried
) {
3878 device
->total_bytes
= old_size
;
3879 if (device
->writeable
)
3880 device
->fs_devices
->total_rw_bytes
+= diff
;
3881 spin_lock(&root
->fs_info
->free_chunk_lock
);
3882 root
->fs_info
->free_chunk_space
+= diff
;
3883 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3884 unlock_chunks(root
);
3888 /* Shrinking succeeded, else we would be at "done". */
3889 trans
= btrfs_start_transaction(root
, 0);
3890 if (IS_ERR(trans
)) {
3891 ret
= PTR_ERR(trans
);
3897 device
->disk_total_bytes
= new_size
;
3898 /* Now btrfs_update_device() will change the on-disk size. */
3899 ret
= btrfs_update_device(trans
, device
);
3901 unlock_chunks(root
);
3902 btrfs_end_transaction(trans
, root
);
3905 WARN_ON(diff
> old_total
);
3906 btrfs_set_super_total_bytes(super_copy
, old_total
- diff
);
3907 unlock_chunks(root
);
3908 btrfs_end_transaction(trans
, root
);
3910 btrfs_free_path(path
);
3914 static int btrfs_add_system_chunk(struct btrfs_root
*root
,
3915 struct btrfs_key
*key
,
3916 struct btrfs_chunk
*chunk
, int item_size
)
3918 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
3919 struct btrfs_disk_key disk_key
;
3923 array_size
= btrfs_super_sys_array_size(super_copy
);
3924 if (array_size
+ item_size
> BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
)
3927 ptr
= super_copy
->sys_chunk_array
+ array_size
;
3928 btrfs_cpu_key_to_disk(&disk_key
, key
);
3929 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
3930 ptr
+= sizeof(disk_key
);
3931 memcpy(ptr
, chunk
, item_size
);
3932 item_size
+= sizeof(disk_key
);
3933 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
3938 * sort the devices in descending order by max_avail, total_avail
3940 static int btrfs_cmp_device_info(const void *a
, const void *b
)
3942 const struct btrfs_device_info
*di_a
= a
;
3943 const struct btrfs_device_info
*di_b
= b
;
3945 if (di_a
->max_avail
> di_b
->max_avail
)
3947 if (di_a
->max_avail
< di_b
->max_avail
)
3949 if (di_a
->total_avail
> di_b
->total_avail
)
3951 if (di_a
->total_avail
< di_b
->total_avail
)
3956 static struct btrfs_raid_attr btrfs_raid_array
[BTRFS_NR_RAID_TYPES
] = {
3957 [BTRFS_RAID_RAID10
] = {
3960 .devs_max
= 0, /* 0 == as many as possible */
3962 .devs_increment
= 2,
3965 [BTRFS_RAID_RAID1
] = {
3970 .devs_increment
= 2,
3973 [BTRFS_RAID_DUP
] = {
3978 .devs_increment
= 1,
3981 [BTRFS_RAID_RAID0
] = {
3986 .devs_increment
= 1,
3989 [BTRFS_RAID_SINGLE
] = {
3994 .devs_increment
= 1,
3997 [BTRFS_RAID_RAID5
] = {
4002 .devs_increment
= 1,
4005 [BTRFS_RAID_RAID6
] = {
4010 .devs_increment
= 1,
4015 static u32
find_raid56_stripe_len(u32 data_devices
, u32 dev_stripe_target
)
4017 /* TODO allow them to set a preferred stripe size */
4021 static void check_raid56_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
4023 if (!(type
& (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
)))
4026 btrfs_set_fs_incompat(info
, RAID56
);
4029 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4030 struct btrfs_root
*extent_root
, u64 start
,
4033 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
4034 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
4035 struct list_head
*cur
;
4036 struct map_lookup
*map
= NULL
;
4037 struct extent_map_tree
*em_tree
;
4038 struct extent_map
*em
;
4039 struct btrfs_device_info
*devices_info
= NULL
;
4041 int num_stripes
; /* total number of stripes to allocate */
4042 int data_stripes
; /* number of stripes that count for
4044 int sub_stripes
; /* sub_stripes info for map */
4045 int dev_stripes
; /* stripes per dev */
4046 int devs_max
; /* max devs to use */
4047 int devs_min
; /* min devs needed */
4048 int devs_increment
; /* ndevs has to be a multiple of this */
4049 int ncopies
; /* how many copies to data has */
4051 u64 max_stripe_size
;
4055 u64 raid_stripe_len
= BTRFS_STRIPE_LEN
;
4061 BUG_ON(!alloc_profile_is_valid(type
, 0));
4063 if (list_empty(&fs_devices
->alloc_list
))
4066 index
= __get_raid_index(type
);
4068 sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
4069 dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
4070 devs_max
= btrfs_raid_array
[index
].devs_max
;
4071 devs_min
= btrfs_raid_array
[index
].devs_min
;
4072 devs_increment
= btrfs_raid_array
[index
].devs_increment
;
4073 ncopies
= btrfs_raid_array
[index
].ncopies
;
4075 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
4076 max_stripe_size
= 1024 * 1024 * 1024;
4077 max_chunk_size
= 10 * max_stripe_size
;
4078 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
4079 /* for larger filesystems, use larger metadata chunks */
4080 if (fs_devices
->total_rw_bytes
> 50ULL * 1024 * 1024 * 1024)
4081 max_stripe_size
= 1024 * 1024 * 1024;
4083 max_stripe_size
= 256 * 1024 * 1024;
4084 max_chunk_size
= max_stripe_size
;
4085 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4086 max_stripe_size
= 32 * 1024 * 1024;
4087 max_chunk_size
= 2 * max_stripe_size
;
4089 btrfs_err(info
, "invalid chunk type 0x%llx requested\n",
4094 /* we don't want a chunk larger than 10% of writeable space */
4095 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
4098 devices_info
= kzalloc(sizeof(*devices_info
) * fs_devices
->rw_devices
,
4103 cur
= fs_devices
->alloc_list
.next
;
4106 * in the first pass through the devices list, we gather information
4107 * about the available holes on each device.
4110 while (cur
!= &fs_devices
->alloc_list
) {
4111 struct btrfs_device
*device
;
4115 device
= list_entry(cur
, struct btrfs_device
, dev_alloc_list
);
4119 if (!device
->writeable
) {
4121 "BTRFS: read-only device in alloc_list\n");
4125 if (!device
->in_fs_metadata
||
4126 device
->is_tgtdev_for_dev_replace
)
4129 if (device
->total_bytes
> device
->bytes_used
)
4130 total_avail
= device
->total_bytes
- device
->bytes_used
;
4134 /* If there is no space on this device, skip it. */
4135 if (total_avail
== 0)
4138 ret
= find_free_dev_extent(trans
, device
,
4139 max_stripe_size
* dev_stripes
,
4140 &dev_offset
, &max_avail
);
4141 if (ret
&& ret
!= -ENOSPC
)
4145 max_avail
= max_stripe_size
* dev_stripes
;
4147 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
4150 if (ndevs
== fs_devices
->rw_devices
) {
4151 WARN(1, "%s: found more than %llu devices\n",
4152 __func__
, fs_devices
->rw_devices
);
4155 devices_info
[ndevs
].dev_offset
= dev_offset
;
4156 devices_info
[ndevs
].max_avail
= max_avail
;
4157 devices_info
[ndevs
].total_avail
= total_avail
;
4158 devices_info
[ndevs
].dev
= device
;
4163 * now sort the devices by hole size / available space
4165 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
4166 btrfs_cmp_device_info
, NULL
);
4168 /* round down to number of usable stripes */
4169 ndevs
-= ndevs
% devs_increment
;
4171 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
4176 if (devs_max
&& ndevs
> devs_max
)
4179 * the primary goal is to maximize the number of stripes, so use as many
4180 * devices as possible, even if the stripes are not maximum sized.
4182 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4183 num_stripes
= ndevs
* dev_stripes
;
4186 * this will have to be fixed for RAID1 and RAID10 over
4189 data_stripes
= num_stripes
/ ncopies
;
4191 if (type
& BTRFS_BLOCK_GROUP_RAID5
) {
4192 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 1,
4193 btrfs_super_stripesize(info
->super_copy
));
4194 data_stripes
= num_stripes
- 1;
4196 if (type
& BTRFS_BLOCK_GROUP_RAID6
) {
4197 raid_stripe_len
= find_raid56_stripe_len(ndevs
- 2,
4198 btrfs_super_stripesize(info
->super_copy
));
4199 data_stripes
= num_stripes
- 2;
4203 * Use the number of data stripes to figure out how big this chunk
4204 * is really going to be in terms of logical address space,
4205 * and compare that answer with the max chunk size
4207 if (stripe_size
* data_stripes
> max_chunk_size
) {
4208 u64 mask
= (1ULL << 24) - 1;
4209 stripe_size
= max_chunk_size
;
4210 do_div(stripe_size
, data_stripes
);
4212 /* bump the answer up to a 16MB boundary */
4213 stripe_size
= (stripe_size
+ mask
) & ~mask
;
4215 /* but don't go higher than the limits we found
4216 * while searching for free extents
4218 if (stripe_size
> devices_info
[ndevs
-1].max_avail
)
4219 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4222 do_div(stripe_size
, dev_stripes
);
4224 /* align to BTRFS_STRIPE_LEN */
4225 do_div(stripe_size
, raid_stripe_len
);
4226 stripe_size
*= raid_stripe_len
;
4228 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
4233 map
->num_stripes
= num_stripes
;
4235 for (i
= 0; i
< ndevs
; ++i
) {
4236 for (j
= 0; j
< dev_stripes
; ++j
) {
4237 int s
= i
* dev_stripes
+ j
;
4238 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
4239 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
4243 map
->sector_size
= extent_root
->sectorsize
;
4244 map
->stripe_len
= raid_stripe_len
;
4245 map
->io_align
= raid_stripe_len
;
4246 map
->io_width
= raid_stripe_len
;
4248 map
->sub_stripes
= sub_stripes
;
4250 num_bytes
= stripe_size
* data_stripes
;
4252 trace_btrfs_chunk_alloc(info
->chunk_root
, map
, start
, num_bytes
);
4254 em
= alloc_extent_map();
4259 em
->bdev
= (struct block_device
*)map
;
4261 em
->len
= num_bytes
;
4262 em
->block_start
= 0;
4263 em
->block_len
= em
->len
;
4264 em
->orig_block_len
= stripe_size
;
4266 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
4267 write_lock(&em_tree
->lock
);
4268 ret
= add_extent_mapping(em_tree
, em
, 0);
4270 list_add_tail(&em
->list
, &trans
->transaction
->pending_chunks
);
4271 atomic_inc(&em
->refs
);
4273 write_unlock(&em_tree
->lock
);
4275 free_extent_map(em
);
4279 ret
= btrfs_make_block_group(trans
, extent_root
, 0, type
,
4280 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4283 goto error_del_extent
;
4285 free_extent_map(em
);
4286 check_raid56_incompat_flag(extent_root
->fs_info
, type
);
4288 kfree(devices_info
);
4292 write_lock(&em_tree
->lock
);
4293 remove_extent_mapping(em_tree
, em
);
4294 write_unlock(&em_tree
->lock
);
4296 /* One for our allocation */
4297 free_extent_map(em
);
4298 /* One for the tree reference */
4299 free_extent_map(em
);
4302 kfree(devices_info
);
4306 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
4307 struct btrfs_root
*extent_root
,
4308 u64 chunk_offset
, u64 chunk_size
)
4310 struct btrfs_key key
;
4311 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
4312 struct btrfs_device
*device
;
4313 struct btrfs_chunk
*chunk
;
4314 struct btrfs_stripe
*stripe
;
4315 struct extent_map_tree
*em_tree
;
4316 struct extent_map
*em
;
4317 struct map_lookup
*map
;
4324 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
4325 read_lock(&em_tree
->lock
);
4326 em
= lookup_extent_mapping(em_tree
, chunk_offset
, chunk_size
);
4327 read_unlock(&em_tree
->lock
);
4330 btrfs_crit(extent_root
->fs_info
, "unable to find logical "
4331 "%Lu len %Lu", chunk_offset
, chunk_size
);
4335 if (em
->start
!= chunk_offset
|| em
->len
!= chunk_size
) {
4336 btrfs_crit(extent_root
->fs_info
, "found a bad mapping, wanted"
4337 " %Lu-%Lu, found %Lu-%Lu\n", chunk_offset
,
4338 chunk_size
, em
->start
, em
->len
);
4339 free_extent_map(em
);
4343 map
= (struct map_lookup
*)em
->bdev
;
4344 item_size
= btrfs_chunk_item_size(map
->num_stripes
);
4345 stripe_size
= em
->orig_block_len
;
4347 chunk
= kzalloc(item_size
, GFP_NOFS
);
4353 for (i
= 0; i
< map
->num_stripes
; i
++) {
4354 device
= map
->stripes
[i
].dev
;
4355 dev_offset
= map
->stripes
[i
].physical
;
4357 device
->bytes_used
+= stripe_size
;
4358 ret
= btrfs_update_device(trans
, device
);
4361 ret
= btrfs_alloc_dev_extent(trans
, device
,
4362 chunk_root
->root_key
.objectid
,
4363 BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
4364 chunk_offset
, dev_offset
,
4370 spin_lock(&extent_root
->fs_info
->free_chunk_lock
);
4371 extent_root
->fs_info
->free_chunk_space
-= (stripe_size
*
4373 spin_unlock(&extent_root
->fs_info
->free_chunk_lock
);
4375 stripe
= &chunk
->stripe
;
4376 for (i
= 0; i
< map
->num_stripes
; i
++) {
4377 device
= map
->stripes
[i
].dev
;
4378 dev_offset
= map
->stripes
[i
].physical
;
4380 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
4381 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
4382 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
4386 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
4387 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
4388 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
4389 btrfs_set_stack_chunk_type(chunk
, map
->type
);
4390 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
4391 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
4392 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
4393 btrfs_set_stack_chunk_sector_size(chunk
, extent_root
->sectorsize
);
4394 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
4396 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
4397 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
4398 key
.offset
= chunk_offset
;
4400 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
4401 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4403 * TODO: Cleanup of inserted chunk root in case of
4406 ret
= btrfs_add_system_chunk(chunk_root
, &key
, chunk
,
4412 free_extent_map(em
);
4417 * Chunk allocation falls into two parts. The first part does works
4418 * that make the new allocated chunk useable, but not do any operation
4419 * that modifies the chunk tree. The second part does the works that
4420 * require modifying the chunk tree. This division is important for the
4421 * bootstrap process of adding storage to a seed btrfs.
4423 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4424 struct btrfs_root
*extent_root
, u64 type
)
4428 chunk_offset
= find_next_chunk(extent_root
->fs_info
);
4429 return __btrfs_alloc_chunk(trans
, extent_root
, chunk_offset
, type
);
4432 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
4433 struct btrfs_root
*root
,
4434 struct btrfs_device
*device
)
4437 u64 sys_chunk_offset
;
4439 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4440 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
4443 chunk_offset
= find_next_chunk(fs_info
);
4444 alloc_profile
= btrfs_get_alloc_profile(extent_root
, 0);
4445 ret
= __btrfs_alloc_chunk(trans
, extent_root
, chunk_offset
,
4450 sys_chunk_offset
= find_next_chunk(root
->fs_info
);
4451 alloc_profile
= btrfs_get_alloc_profile(fs_info
->chunk_root
, 0);
4452 ret
= __btrfs_alloc_chunk(trans
, extent_root
, sys_chunk_offset
,
4455 btrfs_abort_transaction(trans
, root
, ret
);
4459 ret
= btrfs_add_device(trans
, fs_info
->chunk_root
, device
);
4461 btrfs_abort_transaction(trans
, root
, ret
);
4466 int btrfs_chunk_readonly(struct btrfs_root
*root
, u64 chunk_offset
)
4468 struct extent_map
*em
;
4469 struct map_lookup
*map
;
4470 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
4474 read_lock(&map_tree
->map_tree
.lock
);
4475 em
= lookup_extent_mapping(&map_tree
->map_tree
, chunk_offset
, 1);
4476 read_unlock(&map_tree
->map_tree
.lock
);
4480 if (btrfs_test_opt(root
, DEGRADED
)) {
4481 free_extent_map(em
);
4485 map
= (struct map_lookup
*)em
->bdev
;
4486 for (i
= 0; i
< map
->num_stripes
; i
++) {
4487 if (!map
->stripes
[i
].dev
->writeable
) {
4492 free_extent_map(em
);
4496 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
4498 extent_map_tree_init(&tree
->map_tree
);
4501 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
4503 struct extent_map
*em
;
4506 write_lock(&tree
->map_tree
.lock
);
4507 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
4509 remove_extent_mapping(&tree
->map_tree
, em
);
4510 write_unlock(&tree
->map_tree
.lock
);
4515 free_extent_map(em
);
4516 /* once for the tree */
4517 free_extent_map(em
);
4521 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
4523 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
4524 struct extent_map
*em
;
4525 struct map_lookup
*map
;
4526 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4529 read_lock(&em_tree
->lock
);
4530 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4531 read_unlock(&em_tree
->lock
);
4534 * We could return errors for these cases, but that could get ugly and
4535 * we'd probably do the same thing which is just not do anything else
4536 * and exit, so return 1 so the callers don't try to use other copies.
4539 btrfs_crit(fs_info
, "No mapping for %Lu-%Lu\n", logical
,
4544 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
4545 btrfs_crit(fs_info
, "Invalid mapping for %Lu-%Lu, got "
4546 "%Lu-%Lu\n", logical
, logical
+len
, em
->start
,
4547 em
->start
+ em
->len
);
4548 free_extent_map(em
);
4552 map
= (struct map_lookup
*)em
->bdev
;
4553 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
4554 ret
= map
->num_stripes
;
4555 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
4556 ret
= map
->sub_stripes
;
4557 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID5
)
4559 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
4563 free_extent_map(em
);
4565 btrfs_dev_replace_lock(&fs_info
->dev_replace
);
4566 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
))
4568 btrfs_dev_replace_unlock(&fs_info
->dev_replace
);
4573 unsigned long btrfs_full_stripe_len(struct btrfs_root
*root
,
4574 struct btrfs_mapping_tree
*map_tree
,
4577 struct extent_map
*em
;
4578 struct map_lookup
*map
;
4579 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4580 unsigned long len
= root
->sectorsize
;
4582 read_lock(&em_tree
->lock
);
4583 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4584 read_unlock(&em_tree
->lock
);
4587 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4588 map
= (struct map_lookup
*)em
->bdev
;
4589 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
4590 BTRFS_BLOCK_GROUP_RAID6
)) {
4591 len
= map
->stripe_len
* nr_data_stripes(map
);
4593 free_extent_map(em
);
4597 int btrfs_is_parity_mirror(struct btrfs_mapping_tree
*map_tree
,
4598 u64 logical
, u64 len
, int mirror_num
)
4600 struct extent_map
*em
;
4601 struct map_lookup
*map
;
4602 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4605 read_lock(&em_tree
->lock
);
4606 em
= lookup_extent_mapping(em_tree
, logical
, len
);
4607 read_unlock(&em_tree
->lock
);
4610 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
4611 map
= (struct map_lookup
*)em
->bdev
;
4612 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
4613 BTRFS_BLOCK_GROUP_RAID6
))
4615 free_extent_map(em
);
4619 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
4620 struct map_lookup
*map
, int first
, int num
,
4621 int optimal
, int dev_replace_is_ongoing
)
4625 struct btrfs_device
*srcdev
;
4627 if (dev_replace_is_ongoing
&&
4628 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
4629 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
4630 srcdev
= fs_info
->dev_replace
.srcdev
;
4635 * try to avoid the drive that is the source drive for a
4636 * dev-replace procedure, only choose it if no other non-missing
4637 * mirror is available
4639 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
4640 if (map
->stripes
[optimal
].dev
->bdev
&&
4641 (tolerance
|| map
->stripes
[optimal
].dev
!= srcdev
))
4643 for (i
= first
; i
< first
+ num
; i
++) {
4644 if (map
->stripes
[i
].dev
->bdev
&&
4645 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
4650 /* we couldn't find one that doesn't fail. Just return something
4651 * and the io error handling code will clean up eventually
4656 static inline int parity_smaller(u64 a
, u64 b
)
4661 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4662 static void sort_parity_stripes(struct btrfs_bio
*bbio
, u64
*raid_map
)
4664 struct btrfs_bio_stripe s
;
4671 for (i
= 0; i
< bbio
->num_stripes
- 1; i
++) {
4672 if (parity_smaller(raid_map
[i
], raid_map
[i
+1])) {
4673 s
= bbio
->stripes
[i
];
4675 bbio
->stripes
[i
] = bbio
->stripes
[i
+1];
4676 raid_map
[i
] = raid_map
[i
+1];
4677 bbio
->stripes
[i
+1] = s
;
4685 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
4686 u64 logical
, u64
*length
,
4687 struct btrfs_bio
**bbio_ret
,
4688 int mirror_num
, u64
**raid_map_ret
)
4690 struct extent_map
*em
;
4691 struct map_lookup
*map
;
4692 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
4693 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
4696 u64 stripe_end_offset
;
4701 u64
*raid_map
= NULL
;
4707 struct btrfs_bio
*bbio
= NULL
;
4708 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
4709 int dev_replace_is_ongoing
= 0;
4710 int num_alloc_stripes
;
4711 int patch_the_first_stripe_for_dev_replace
= 0;
4712 u64 physical_to_patch_in_first_stripe
= 0;
4713 u64 raid56_full_stripe_start
= (u64
)-1;
4715 read_lock(&em_tree
->lock
);
4716 em
= lookup_extent_mapping(em_tree
, logical
, *length
);
4717 read_unlock(&em_tree
->lock
);
4720 btrfs_crit(fs_info
, "unable to find logical %llu len %llu",
4725 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
4726 btrfs_crit(fs_info
, "found a bad mapping, wanted %Lu, "
4727 "found %Lu-%Lu\n", logical
, em
->start
,
4728 em
->start
+ em
->len
);
4729 free_extent_map(em
);
4733 map
= (struct map_lookup
*)em
->bdev
;
4734 offset
= logical
- em
->start
;
4736 stripe_len
= map
->stripe_len
;
4739 * stripe_nr counts the total number of stripes we have to stride
4740 * to get to this block
4742 do_div(stripe_nr
, stripe_len
);
4744 stripe_offset
= stripe_nr
* stripe_len
;
4745 BUG_ON(offset
< stripe_offset
);
4747 /* stripe_offset is the offset of this block in its stripe*/
4748 stripe_offset
= offset
- stripe_offset
;
4750 /* if we're here for raid56, we need to know the stripe aligned start */
4751 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
)) {
4752 unsigned long full_stripe_len
= stripe_len
* nr_data_stripes(map
);
4753 raid56_full_stripe_start
= offset
;
4755 /* allow a write of a full stripe, but make sure we don't
4756 * allow straddling of stripes
4758 do_div(raid56_full_stripe_start
, full_stripe_len
);
4759 raid56_full_stripe_start
*= full_stripe_len
;
4762 if (rw
& REQ_DISCARD
) {
4763 /* we don't discard raid56 yet */
4765 (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
)) {
4769 *length
= min_t(u64
, em
->len
- offset
, *length
);
4770 } else if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
4772 /* For writes to RAID[56], allow a full stripeset across all disks.
4773 For other RAID types and for RAID[56] reads, just allow a single
4774 stripe (on a single disk). */
4775 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
) &&
4777 max_len
= stripe_len
* nr_data_stripes(map
) -
4778 (offset
- raid56_full_stripe_start
);
4780 /* we limit the length of each bio to what fits in a stripe */
4781 max_len
= stripe_len
- stripe_offset
;
4783 *length
= min_t(u64
, em
->len
- offset
, max_len
);
4785 *length
= em
->len
- offset
;
4788 /* This is for when we're called from btrfs_merge_bio_hook() and all
4789 it cares about is the length */
4793 btrfs_dev_replace_lock(dev_replace
);
4794 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
4795 if (!dev_replace_is_ongoing
)
4796 btrfs_dev_replace_unlock(dev_replace
);
4798 if (dev_replace_is_ongoing
&& mirror_num
== map
->num_stripes
+ 1 &&
4799 !(rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) &&
4800 dev_replace
->tgtdev
!= NULL
) {
4802 * in dev-replace case, for repair case (that's the only
4803 * case where the mirror is selected explicitly when
4804 * calling btrfs_map_block), blocks left of the left cursor
4805 * can also be read from the target drive.
4806 * For REQ_GET_READ_MIRRORS, the target drive is added as
4807 * the last one to the array of stripes. For READ, it also
4808 * needs to be supported using the same mirror number.
4809 * If the requested block is not left of the left cursor,
4810 * EIO is returned. This can happen because btrfs_num_copies()
4811 * returns one more in the dev-replace case.
4813 u64 tmp_length
= *length
;
4814 struct btrfs_bio
*tmp_bbio
= NULL
;
4815 int tmp_num_stripes
;
4816 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
4817 int index_srcdev
= 0;
4819 u64 physical_of_found
= 0;
4821 ret
= __btrfs_map_block(fs_info
, REQ_GET_READ_MIRRORS
,
4822 logical
, &tmp_length
, &tmp_bbio
, 0, NULL
);
4824 WARN_ON(tmp_bbio
!= NULL
);
4828 tmp_num_stripes
= tmp_bbio
->num_stripes
;
4829 if (mirror_num
> tmp_num_stripes
) {
4831 * REQ_GET_READ_MIRRORS does not contain this
4832 * mirror, that means that the requested area
4833 * is not left of the left cursor
4841 * process the rest of the function using the mirror_num
4842 * of the source drive. Therefore look it up first.
4843 * At the end, patch the device pointer to the one of the
4846 for (i
= 0; i
< tmp_num_stripes
; i
++) {
4847 if (tmp_bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
4849 * In case of DUP, in order to keep it
4850 * simple, only add the mirror with the
4851 * lowest physical address
4854 physical_of_found
<=
4855 tmp_bbio
->stripes
[i
].physical
)
4860 tmp_bbio
->stripes
[i
].physical
;
4865 mirror_num
= index_srcdev
+ 1;
4866 patch_the_first_stripe_for_dev_replace
= 1;
4867 physical_to_patch_in_first_stripe
= physical_of_found
;
4876 } else if (mirror_num
> map
->num_stripes
) {
4882 stripe_nr_orig
= stripe_nr
;
4883 stripe_nr_end
= ALIGN(offset
+ *length
, map
->stripe_len
);
4884 do_div(stripe_nr_end
, map
->stripe_len
);
4885 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
4888 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
4889 if (rw
& REQ_DISCARD
)
4890 num_stripes
= min_t(u64
, map
->num_stripes
,
4891 stripe_nr_end
- stripe_nr_orig
);
4892 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
4893 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
4894 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
))
4895 num_stripes
= map
->num_stripes
;
4896 else if (mirror_num
)
4897 stripe_index
= mirror_num
- 1;
4899 stripe_index
= find_live_mirror(fs_info
, map
, 0,
4901 current
->pid
% map
->num_stripes
,
4902 dev_replace_is_ongoing
);
4903 mirror_num
= stripe_index
+ 1;
4906 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
4907 if (rw
& (REQ_WRITE
| REQ_DISCARD
| REQ_GET_READ_MIRRORS
)) {
4908 num_stripes
= map
->num_stripes
;
4909 } else if (mirror_num
) {
4910 stripe_index
= mirror_num
- 1;
4915 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
4916 int factor
= map
->num_stripes
/ map
->sub_stripes
;
4918 stripe_index
= do_div(stripe_nr
, factor
);
4919 stripe_index
*= map
->sub_stripes
;
4921 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
))
4922 num_stripes
= map
->sub_stripes
;
4923 else if (rw
& REQ_DISCARD
)
4924 num_stripes
= min_t(u64
, map
->sub_stripes
*
4925 (stripe_nr_end
- stripe_nr_orig
),
4927 else if (mirror_num
)
4928 stripe_index
+= mirror_num
- 1;
4930 int old_stripe_index
= stripe_index
;
4931 stripe_index
= find_live_mirror(fs_info
, map
,
4933 map
->sub_stripes
, stripe_index
+
4934 current
->pid
% map
->sub_stripes
,
4935 dev_replace_is_ongoing
);
4936 mirror_num
= stripe_index
- old_stripe_index
+ 1;
4939 } else if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
4940 BTRFS_BLOCK_GROUP_RAID6
)) {
4943 if (bbio_ret
&& ((rw
& REQ_WRITE
) || mirror_num
> 1)
4947 /* push stripe_nr back to the start of the full stripe */
4948 stripe_nr
= raid56_full_stripe_start
;
4949 do_div(stripe_nr
, stripe_len
);
4951 stripe_index
= do_div(stripe_nr
, nr_data_stripes(map
));
4953 /* RAID[56] write or recovery. Return all stripes */
4954 num_stripes
= map
->num_stripes
;
4955 max_errors
= nr_parity_stripes(map
);
4957 raid_map
= kmalloc_array(num_stripes
, sizeof(u64
),
4964 /* Work out the disk rotation on this stripe-set */
4966 rot
= do_div(tmp
, num_stripes
);
4968 /* Fill in the logical address of each stripe */
4969 tmp
= stripe_nr
* nr_data_stripes(map
);
4970 for (i
= 0; i
< nr_data_stripes(map
); i
++)
4971 raid_map
[(i
+rot
) % num_stripes
] =
4972 em
->start
+ (tmp
+ i
) * map
->stripe_len
;
4974 raid_map
[(i
+rot
) % map
->num_stripes
] = RAID5_P_STRIPE
;
4975 if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
4976 raid_map
[(i
+rot
+1) % num_stripes
] =
4979 *length
= map
->stripe_len
;
4984 * Mirror #0 or #1 means the original data block.
4985 * Mirror #2 is RAID5 parity block.
4986 * Mirror #3 is RAID6 Q block.
4988 stripe_index
= do_div(stripe_nr
, nr_data_stripes(map
));
4990 stripe_index
= nr_data_stripes(map
) +
4993 /* We distribute the parity blocks across stripes */
4994 tmp
= stripe_nr
+ stripe_index
;
4995 stripe_index
= do_div(tmp
, map
->num_stripes
);
4999 * after this do_div call, stripe_nr is the number of stripes
5000 * on this device we have to walk to find the data, and
5001 * stripe_index is the number of our device in the stripe array
5003 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
5004 mirror_num
= stripe_index
+ 1;
5006 BUG_ON(stripe_index
>= map
->num_stripes
);
5008 num_alloc_stripes
= num_stripes
;
5009 if (dev_replace_is_ongoing
) {
5010 if (rw
& (REQ_WRITE
| REQ_DISCARD
))
5011 num_alloc_stripes
<<= 1;
5012 if (rw
& REQ_GET_READ_MIRRORS
)
5013 num_alloc_stripes
++;
5015 bbio
= kzalloc(btrfs_bio_size(num_alloc_stripes
), GFP_NOFS
);
5021 atomic_set(&bbio
->error
, 0);
5023 if (rw
& REQ_DISCARD
) {
5025 int sub_stripes
= 0;
5026 u64 stripes_per_dev
= 0;
5027 u32 remaining_stripes
= 0;
5028 u32 last_stripe
= 0;
5031 (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID10
)) {
5032 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5035 sub_stripes
= map
->sub_stripes
;
5037 factor
= map
->num_stripes
/ sub_stripes
;
5038 stripes_per_dev
= div_u64_rem(stripe_nr_end
-
5041 &remaining_stripes
);
5042 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
5043 last_stripe
*= sub_stripes
;
5046 for (i
= 0; i
< num_stripes
; i
++) {
5047 bbio
->stripes
[i
].physical
=
5048 map
->stripes
[stripe_index
].physical
+
5049 stripe_offset
+ stripe_nr
* map
->stripe_len
;
5050 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
5052 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5053 BTRFS_BLOCK_GROUP_RAID10
)) {
5054 bbio
->stripes
[i
].length
= stripes_per_dev
*
5057 if (i
/ sub_stripes
< remaining_stripes
)
5058 bbio
->stripes
[i
].length
+=
5062 * Special for the first stripe and
5065 * |-------|...|-------|
5069 if (i
< sub_stripes
)
5070 bbio
->stripes
[i
].length
-=
5073 if (stripe_index
>= last_stripe
&&
5074 stripe_index
<= (last_stripe
+
5076 bbio
->stripes
[i
].length
-=
5079 if (i
== sub_stripes
- 1)
5082 bbio
->stripes
[i
].length
= *length
;
5085 if (stripe_index
== map
->num_stripes
) {
5086 /* This could only happen for RAID0/10 */
5092 for (i
= 0; i
< num_stripes
; i
++) {
5093 bbio
->stripes
[i
].physical
=
5094 map
->stripes
[stripe_index
].physical
+
5096 stripe_nr
* map
->stripe_len
;
5097 bbio
->stripes
[i
].dev
=
5098 map
->stripes
[stripe_index
].dev
;
5103 if (rw
& (REQ_WRITE
| REQ_GET_READ_MIRRORS
)) {
5104 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
5105 BTRFS_BLOCK_GROUP_RAID10
|
5106 BTRFS_BLOCK_GROUP_RAID5
|
5107 BTRFS_BLOCK_GROUP_DUP
)) {
5109 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
) {
5114 if (dev_replace_is_ongoing
&& (rw
& (REQ_WRITE
| REQ_DISCARD
)) &&
5115 dev_replace
->tgtdev
!= NULL
) {
5116 int index_where_to_add
;
5117 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5120 * duplicate the write operations while the dev replace
5121 * procedure is running. Since the copying of the old disk
5122 * to the new disk takes place at run time while the
5123 * filesystem is mounted writable, the regular write
5124 * operations to the old disk have to be duplicated to go
5125 * to the new disk as well.
5126 * Note that device->missing is handled by the caller, and
5127 * that the write to the old disk is already set up in the
5130 index_where_to_add
= num_stripes
;
5131 for (i
= 0; i
< num_stripes
; i
++) {
5132 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5133 /* write to new disk, too */
5134 struct btrfs_bio_stripe
*new =
5135 bbio
->stripes
+ index_where_to_add
;
5136 struct btrfs_bio_stripe
*old
=
5139 new->physical
= old
->physical
;
5140 new->length
= old
->length
;
5141 new->dev
= dev_replace
->tgtdev
;
5142 index_where_to_add
++;
5146 num_stripes
= index_where_to_add
;
5147 } else if (dev_replace_is_ongoing
&& (rw
& REQ_GET_READ_MIRRORS
) &&
5148 dev_replace
->tgtdev
!= NULL
) {
5149 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5150 int index_srcdev
= 0;
5152 u64 physical_of_found
= 0;
5155 * During the dev-replace procedure, the target drive can
5156 * also be used to read data in case it is needed to repair
5157 * a corrupt block elsewhere. This is possible if the
5158 * requested area is left of the left cursor. In this area,
5159 * the target drive is a full copy of the source drive.
5161 for (i
= 0; i
< num_stripes
; i
++) {
5162 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5164 * In case of DUP, in order to keep it
5165 * simple, only add the mirror with the
5166 * lowest physical address
5169 physical_of_found
<=
5170 bbio
->stripes
[i
].physical
)
5174 physical_of_found
= bbio
->stripes
[i
].physical
;
5178 u64 length
= map
->stripe_len
;
5180 if (physical_of_found
+ length
<=
5181 dev_replace
->cursor_left
) {
5182 struct btrfs_bio_stripe
*tgtdev_stripe
=
5183 bbio
->stripes
+ num_stripes
;
5185 tgtdev_stripe
->physical
= physical_of_found
;
5186 tgtdev_stripe
->length
=
5187 bbio
->stripes
[index_srcdev
].length
;
5188 tgtdev_stripe
->dev
= dev_replace
->tgtdev
;
5196 bbio
->num_stripes
= num_stripes
;
5197 bbio
->max_errors
= max_errors
;
5198 bbio
->mirror_num
= mirror_num
;
5201 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5202 * mirror_num == num_stripes + 1 && dev_replace target drive is
5203 * available as a mirror
5205 if (patch_the_first_stripe_for_dev_replace
&& num_stripes
> 0) {
5206 WARN_ON(num_stripes
> 1);
5207 bbio
->stripes
[0].dev
= dev_replace
->tgtdev
;
5208 bbio
->stripes
[0].physical
= physical_to_patch_in_first_stripe
;
5209 bbio
->mirror_num
= map
->num_stripes
+ 1;
5212 sort_parity_stripes(bbio
, raid_map
);
5213 *raid_map_ret
= raid_map
;
5216 if (dev_replace_is_ongoing
)
5217 btrfs_dev_replace_unlock(dev_replace
);
5218 free_extent_map(em
);
5222 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, int rw
,
5223 u64 logical
, u64
*length
,
5224 struct btrfs_bio
**bbio_ret
, int mirror_num
)
5226 return __btrfs_map_block(fs_info
, rw
, logical
, length
, bbio_ret
,
5230 int btrfs_rmap_block(struct btrfs_mapping_tree
*map_tree
,
5231 u64 chunk_start
, u64 physical
, u64 devid
,
5232 u64
**logical
, int *naddrs
, int *stripe_len
)
5234 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
5235 struct extent_map
*em
;
5236 struct map_lookup
*map
;
5244 read_lock(&em_tree
->lock
);
5245 em
= lookup_extent_mapping(em_tree
, chunk_start
, 1);
5246 read_unlock(&em_tree
->lock
);
5249 printk(KERN_ERR
"BTRFS: couldn't find em for chunk %Lu\n",
5254 if (em
->start
!= chunk_start
) {
5255 printk(KERN_ERR
"BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5256 em
->start
, chunk_start
);
5257 free_extent_map(em
);
5260 map
= (struct map_lookup
*)em
->bdev
;
5263 rmap_len
= map
->stripe_len
;
5265 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5266 do_div(length
, map
->num_stripes
/ map
->sub_stripes
);
5267 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5268 do_div(length
, map
->num_stripes
);
5269 else if (map
->type
& (BTRFS_BLOCK_GROUP_RAID5
|
5270 BTRFS_BLOCK_GROUP_RAID6
)) {
5271 do_div(length
, nr_data_stripes(map
));
5272 rmap_len
= map
->stripe_len
* nr_data_stripes(map
);
5275 buf
= kzalloc(sizeof(u64
) * map
->num_stripes
, GFP_NOFS
);
5276 BUG_ON(!buf
); /* -ENOMEM */
5278 for (i
= 0; i
< map
->num_stripes
; i
++) {
5279 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
5281 if (map
->stripes
[i
].physical
> physical
||
5282 map
->stripes
[i
].physical
+ length
<= physical
)
5285 stripe_nr
= physical
- map
->stripes
[i
].physical
;
5286 do_div(stripe_nr
, map
->stripe_len
);
5288 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5289 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5290 do_div(stripe_nr
, map
->sub_stripes
);
5291 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5292 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5293 } /* else if RAID[56], multiply by nr_data_stripes().
5294 * Alternatively, just use rmap_len below instead of
5295 * map->stripe_len */
5297 bytenr
= chunk_start
+ stripe_nr
* rmap_len
;
5298 WARN_ON(nr
>= map
->num_stripes
);
5299 for (j
= 0; j
< nr
; j
++) {
5300 if (buf
[j
] == bytenr
)
5304 WARN_ON(nr
>= map
->num_stripes
);
5311 *stripe_len
= rmap_len
;
5313 free_extent_map(em
);
5317 static void btrfs_end_bio(struct bio
*bio
, int err
)
5319 struct btrfs_bio
*bbio
= bio
->bi_private
;
5320 struct btrfs_device
*dev
= bbio
->stripes
[0].dev
;
5321 int is_orig_bio
= 0;
5324 atomic_inc(&bbio
->error
);
5325 if (err
== -EIO
|| err
== -EREMOTEIO
) {
5326 unsigned int stripe_index
=
5327 btrfs_io_bio(bio
)->stripe_index
;
5329 BUG_ON(stripe_index
>= bbio
->num_stripes
);
5330 dev
= bbio
->stripes
[stripe_index
].dev
;
5332 if (bio
->bi_rw
& WRITE
)
5333 btrfs_dev_stat_inc(dev
,
5334 BTRFS_DEV_STAT_WRITE_ERRS
);
5336 btrfs_dev_stat_inc(dev
,
5337 BTRFS_DEV_STAT_READ_ERRS
);
5338 if ((bio
->bi_rw
& WRITE_FLUSH
) == WRITE_FLUSH
)
5339 btrfs_dev_stat_inc(dev
,
5340 BTRFS_DEV_STAT_FLUSH_ERRS
);
5341 btrfs_dev_stat_print_on_error(dev
);
5346 if (bio
== bbio
->orig_bio
)
5349 btrfs_bio_counter_dec(bbio
->fs_info
);
5351 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
5354 bio
= bbio
->orig_bio
;
5358 * We have original bio now. So increment bi_remaining to
5359 * account for it in endio
5361 atomic_inc(&bio
->bi_remaining
);
5363 bio
->bi_private
= bbio
->private;
5364 bio
->bi_end_io
= bbio
->end_io
;
5365 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
5366 /* only send an error to the higher layers if it is
5367 * beyond the tolerance of the btrfs bio
5369 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
5373 * this bio is actually up to date, we didn't
5374 * go over the max number of errors
5376 set_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5381 bio_endio(bio
, err
);
5382 } else if (!is_orig_bio
) {
5388 * see run_scheduled_bios for a description of why bios are collected for
5391 * This will add one bio to the pending list for a device and make sure
5392 * the work struct is scheduled.
5394 static noinline
void btrfs_schedule_bio(struct btrfs_root
*root
,
5395 struct btrfs_device
*device
,
5396 int rw
, struct bio
*bio
)
5398 int should_queue
= 1;
5399 struct btrfs_pending_bios
*pending_bios
;
5401 if (device
->missing
|| !device
->bdev
) {
5402 bio_endio(bio
, -EIO
);
5406 /* don't bother with additional async steps for reads, right now */
5407 if (!(rw
& REQ_WRITE
)) {
5409 btrfsic_submit_bio(rw
, bio
);
5415 * nr_async_bios allows us to reliably return congestion to the
5416 * higher layers. Otherwise, the async bio makes it appear we have
5417 * made progress against dirty pages when we've really just put it
5418 * on a queue for later
5420 atomic_inc(&root
->fs_info
->nr_async_bios
);
5421 WARN_ON(bio
->bi_next
);
5422 bio
->bi_next
= NULL
;
5425 spin_lock(&device
->io_lock
);
5426 if (bio
->bi_rw
& REQ_SYNC
)
5427 pending_bios
= &device
->pending_sync_bios
;
5429 pending_bios
= &device
->pending_bios
;
5431 if (pending_bios
->tail
)
5432 pending_bios
->tail
->bi_next
= bio
;
5434 pending_bios
->tail
= bio
;
5435 if (!pending_bios
->head
)
5436 pending_bios
->head
= bio
;
5437 if (device
->running_pending
)
5440 spin_unlock(&device
->io_lock
);
5443 btrfs_queue_work(root
->fs_info
->submit_workers
,
5447 static int bio_size_ok(struct block_device
*bdev
, struct bio
*bio
,
5450 struct bio_vec
*prev
;
5451 struct request_queue
*q
= bdev_get_queue(bdev
);
5452 unsigned int max_sectors
= queue_max_sectors(q
);
5453 struct bvec_merge_data bvm
= {
5455 .bi_sector
= sector
,
5456 .bi_rw
= bio
->bi_rw
,
5459 if (WARN_ON(bio
->bi_vcnt
== 0))
5462 prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
5463 if (bio_sectors(bio
) > max_sectors
)
5466 if (!q
->merge_bvec_fn
)
5469 bvm
.bi_size
= bio
->bi_iter
.bi_size
- prev
->bv_len
;
5470 if (q
->merge_bvec_fn(q
, &bvm
, prev
) < prev
->bv_len
)
5475 static void submit_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
5476 struct bio
*bio
, u64 physical
, int dev_nr
,
5479 struct btrfs_device
*dev
= bbio
->stripes
[dev_nr
].dev
;
5481 bio
->bi_private
= bbio
;
5482 btrfs_io_bio(bio
)->stripe_index
= dev_nr
;
5483 bio
->bi_end_io
= btrfs_end_bio
;
5484 bio
->bi_iter
.bi_sector
= physical
>> 9;
5487 struct rcu_string
*name
;
5490 name
= rcu_dereference(dev
->name
);
5491 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5492 "(%s id %llu), size=%u\n", rw
,
5493 (u64
)bio
->bi_sector
, (u_long
)dev
->bdev
->bd_dev
,
5494 name
->str
, dev
->devid
, bio
->bi_size
);
5498 bio
->bi_bdev
= dev
->bdev
;
5500 btrfs_bio_counter_inc_noblocked(root
->fs_info
);
5503 btrfs_schedule_bio(root
, dev
, rw
, bio
);
5505 btrfsic_submit_bio(rw
, bio
);
5508 static int breakup_stripe_bio(struct btrfs_root
*root
, struct btrfs_bio
*bbio
,
5509 struct bio
*first_bio
, struct btrfs_device
*dev
,
5510 int dev_nr
, int rw
, int async
)
5512 struct bio_vec
*bvec
= first_bio
->bi_io_vec
;
5514 int nr_vecs
= bio_get_nr_vecs(dev
->bdev
);
5515 u64 physical
= bbio
->stripes
[dev_nr
].physical
;
5518 bio
= btrfs_bio_alloc(dev
->bdev
, physical
>> 9, nr_vecs
, GFP_NOFS
);
5522 while (bvec
<= (first_bio
->bi_io_vec
+ first_bio
->bi_vcnt
- 1)) {
5523 if (bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
5524 bvec
->bv_offset
) < bvec
->bv_len
) {
5525 u64 len
= bio
->bi_iter
.bi_size
;
5527 atomic_inc(&bbio
->stripes_pending
);
5528 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
,
5536 submit_stripe_bio(root
, bbio
, bio
, physical
, dev_nr
, rw
, async
);
5540 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
5542 atomic_inc(&bbio
->error
);
5543 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
5544 bio
->bi_private
= bbio
->private;
5545 bio
->bi_end_io
= bbio
->end_io
;
5546 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
5547 bio
->bi_iter
.bi_sector
= logical
>> 9;
5549 bio_endio(bio
, -EIO
);
5553 int btrfs_map_bio(struct btrfs_root
*root
, int rw
, struct bio
*bio
,
5554 int mirror_num
, int async_submit
)
5556 struct btrfs_device
*dev
;
5557 struct bio
*first_bio
= bio
;
5558 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
5561 u64
*raid_map
= NULL
;
5565 struct btrfs_bio
*bbio
= NULL
;
5567 length
= bio
->bi_iter
.bi_size
;
5568 map_length
= length
;
5570 btrfs_bio_counter_inc_blocked(root
->fs_info
);
5571 ret
= __btrfs_map_block(root
->fs_info
, rw
, logical
, &map_length
, &bbio
,
5572 mirror_num
, &raid_map
);
5574 btrfs_bio_counter_dec(root
->fs_info
);
5578 total_devs
= bbio
->num_stripes
;
5579 bbio
->orig_bio
= first_bio
;
5580 bbio
->private = first_bio
->bi_private
;
5581 bbio
->end_io
= first_bio
->bi_end_io
;
5582 bbio
->fs_info
= root
->fs_info
;
5583 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
5586 /* In this case, map_length has been set to the length of
5587 a single stripe; not the whole write */
5589 ret
= raid56_parity_write(root
, bio
, bbio
,
5590 raid_map
, map_length
);
5592 ret
= raid56_parity_recover(root
, bio
, bbio
,
5593 raid_map
, map_length
,
5597 * FIXME, replace dosen't support raid56 yet, please fix
5600 btrfs_bio_counter_dec(root
->fs_info
);
5604 if (map_length
< length
) {
5605 btrfs_crit(root
->fs_info
, "mapping failed logical %llu bio len %llu len %llu",
5606 logical
, length
, map_length
);
5610 while (dev_nr
< total_devs
) {
5611 dev
= bbio
->stripes
[dev_nr
].dev
;
5612 if (!dev
|| !dev
->bdev
|| (rw
& WRITE
&& !dev
->writeable
)) {
5613 bbio_error(bbio
, first_bio
, logical
);
5619 * Check and see if we're ok with this bio based on it's size
5620 * and offset with the given device.
5622 if (!bio_size_ok(dev
->bdev
, first_bio
,
5623 bbio
->stripes
[dev_nr
].physical
>> 9)) {
5624 ret
= breakup_stripe_bio(root
, bbio
, first_bio
, dev
,
5625 dev_nr
, rw
, async_submit
);
5631 if (dev_nr
< total_devs
- 1) {
5632 bio
= btrfs_bio_clone(first_bio
, GFP_NOFS
);
5633 BUG_ON(!bio
); /* -ENOMEM */
5638 submit_stripe_bio(root
, bbio
, bio
,
5639 bbio
->stripes
[dev_nr
].physical
, dev_nr
, rw
,
5643 btrfs_bio_counter_dec(root
->fs_info
);
5647 struct btrfs_device
*btrfs_find_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
5650 struct btrfs_device
*device
;
5651 struct btrfs_fs_devices
*cur_devices
;
5653 cur_devices
= fs_info
->fs_devices
;
5654 while (cur_devices
) {
5656 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
5657 device
= __find_device(&cur_devices
->devices
,
5662 cur_devices
= cur_devices
->seed
;
5667 static struct btrfs_device
*add_missing_dev(struct btrfs_root
*root
,
5668 u64 devid
, u8
*dev_uuid
)
5670 struct btrfs_device
*device
;
5671 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
5673 device
= btrfs_alloc_device(NULL
, &devid
, dev_uuid
);
5677 list_add(&device
->dev_list
, &fs_devices
->devices
);
5678 device
->fs_devices
= fs_devices
;
5679 fs_devices
->num_devices
++;
5681 device
->missing
= 1;
5682 fs_devices
->missing_devices
++;
5688 * btrfs_alloc_device - allocate struct btrfs_device
5689 * @fs_info: used only for generating a new devid, can be NULL if
5690 * devid is provided (i.e. @devid != NULL).
5691 * @devid: a pointer to devid for this device. If NULL a new devid
5693 * @uuid: a pointer to UUID for this device. If NULL a new UUID
5696 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5697 * on error. Returned struct is not linked onto any lists and can be
5698 * destroyed with kfree() right away.
5700 struct btrfs_device
*btrfs_alloc_device(struct btrfs_fs_info
*fs_info
,
5704 struct btrfs_device
*dev
;
5707 if (WARN_ON(!devid
&& !fs_info
))
5708 return ERR_PTR(-EINVAL
);
5710 dev
= __alloc_device();
5719 ret
= find_next_devid(fs_info
, &tmp
);
5722 return ERR_PTR(ret
);
5728 memcpy(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
);
5730 generate_random_uuid(dev
->uuid
);
5732 btrfs_init_work(&dev
->work
, pending_bios_fn
, NULL
, NULL
);
5737 static int read_one_chunk(struct btrfs_root
*root
, struct btrfs_key
*key
,
5738 struct extent_buffer
*leaf
,
5739 struct btrfs_chunk
*chunk
)
5741 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
5742 struct map_lookup
*map
;
5743 struct extent_map
*em
;
5747 u8 uuid
[BTRFS_UUID_SIZE
];
5752 logical
= key
->offset
;
5753 length
= btrfs_chunk_length(leaf
, chunk
);
5755 read_lock(&map_tree
->map_tree
.lock
);
5756 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
5757 read_unlock(&map_tree
->map_tree
.lock
);
5759 /* already mapped? */
5760 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
5761 free_extent_map(em
);
5764 free_extent_map(em
);
5767 em
= alloc_extent_map();
5770 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
5771 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
5773 free_extent_map(em
);
5777 em
->bdev
= (struct block_device
*)map
;
5778 em
->start
= logical
;
5781 em
->block_start
= 0;
5782 em
->block_len
= em
->len
;
5784 map
->num_stripes
= num_stripes
;
5785 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
5786 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
5787 map
->sector_size
= btrfs_chunk_sector_size(leaf
, chunk
);
5788 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
5789 map
->type
= btrfs_chunk_type(leaf
, chunk
);
5790 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
5791 for (i
= 0; i
< num_stripes
; i
++) {
5792 map
->stripes
[i
].physical
=
5793 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
5794 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
5795 read_extent_buffer(leaf
, uuid
, (unsigned long)
5796 btrfs_stripe_dev_uuid_nr(chunk
, i
),
5798 map
->stripes
[i
].dev
= btrfs_find_device(root
->fs_info
, devid
,
5800 if (!map
->stripes
[i
].dev
&& !btrfs_test_opt(root
, DEGRADED
)) {
5802 free_extent_map(em
);
5805 if (!map
->stripes
[i
].dev
) {
5806 map
->stripes
[i
].dev
=
5807 add_missing_dev(root
, devid
, uuid
);
5808 if (!map
->stripes
[i
].dev
) {
5810 free_extent_map(em
);
5814 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
5817 write_lock(&map_tree
->map_tree
.lock
);
5818 ret
= add_extent_mapping(&map_tree
->map_tree
, em
, 0);
5819 write_unlock(&map_tree
->map_tree
.lock
);
5820 BUG_ON(ret
); /* Tree corruption */
5821 free_extent_map(em
);
5826 static void fill_device_from_item(struct extent_buffer
*leaf
,
5827 struct btrfs_dev_item
*dev_item
,
5828 struct btrfs_device
*device
)
5832 device
->devid
= btrfs_device_id(leaf
, dev_item
);
5833 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
5834 device
->total_bytes
= device
->disk_total_bytes
;
5835 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
5836 device
->type
= btrfs_device_type(leaf
, dev_item
);
5837 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
5838 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
5839 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
5840 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
5841 device
->is_tgtdev_for_dev_replace
= 0;
5843 ptr
= btrfs_device_uuid(dev_item
);
5844 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
5847 static int open_seed_devices(struct btrfs_root
*root
, u8
*fsid
)
5849 struct btrfs_fs_devices
*fs_devices
;
5852 BUG_ON(!mutex_is_locked(&uuid_mutex
));
5854 fs_devices
= root
->fs_info
->fs_devices
->seed
;
5855 while (fs_devices
) {
5856 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_UUID_SIZE
)) {
5860 fs_devices
= fs_devices
->seed
;
5863 fs_devices
= find_fsid(fsid
);
5869 fs_devices
= clone_fs_devices(fs_devices
);
5870 if (IS_ERR(fs_devices
)) {
5871 ret
= PTR_ERR(fs_devices
);
5875 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
5876 root
->fs_info
->bdev_holder
);
5878 free_fs_devices(fs_devices
);
5882 if (!fs_devices
->seeding
) {
5883 __btrfs_close_devices(fs_devices
);
5884 free_fs_devices(fs_devices
);
5889 fs_devices
->seed
= root
->fs_info
->fs_devices
->seed
;
5890 root
->fs_info
->fs_devices
->seed
= fs_devices
;
5895 static int read_one_dev(struct btrfs_root
*root
,
5896 struct extent_buffer
*leaf
,
5897 struct btrfs_dev_item
*dev_item
)
5899 struct btrfs_device
*device
;
5902 u8 fs_uuid
[BTRFS_UUID_SIZE
];
5903 u8 dev_uuid
[BTRFS_UUID_SIZE
];
5905 devid
= btrfs_device_id(leaf
, dev_item
);
5906 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
5908 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
5911 if (memcmp(fs_uuid
, root
->fs_info
->fsid
, BTRFS_UUID_SIZE
)) {
5912 ret
= open_seed_devices(root
, fs_uuid
);
5913 if (ret
&& !btrfs_test_opt(root
, DEGRADED
))
5917 device
= btrfs_find_device(root
->fs_info
, devid
, dev_uuid
, fs_uuid
);
5918 if (!device
|| !device
->bdev
) {
5919 if (!btrfs_test_opt(root
, DEGRADED
))
5923 btrfs_warn(root
->fs_info
, "devid %llu missing", devid
);
5924 device
= add_missing_dev(root
, devid
, dev_uuid
);
5927 } else if (!device
->missing
) {
5929 * this happens when a device that was properly setup
5930 * in the device info lists suddenly goes bad.
5931 * device->bdev is NULL, and so we have to set
5932 * device->missing to one here
5934 root
->fs_info
->fs_devices
->missing_devices
++;
5935 device
->missing
= 1;
5939 if (device
->fs_devices
!= root
->fs_info
->fs_devices
) {
5940 BUG_ON(device
->writeable
);
5941 if (device
->generation
!=
5942 btrfs_device_generation(leaf
, dev_item
))
5946 fill_device_from_item(leaf
, dev_item
, device
);
5947 device
->in_fs_metadata
= 1;
5948 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
5949 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
5950 spin_lock(&root
->fs_info
->free_chunk_lock
);
5951 root
->fs_info
->free_chunk_space
+= device
->total_bytes
-
5953 spin_unlock(&root
->fs_info
->free_chunk_lock
);
5959 int btrfs_read_sys_array(struct btrfs_root
*root
)
5961 struct btrfs_super_block
*super_copy
= root
->fs_info
->super_copy
;
5962 struct extent_buffer
*sb
;
5963 struct btrfs_disk_key
*disk_key
;
5964 struct btrfs_chunk
*chunk
;
5966 unsigned long sb_ptr
;
5972 struct btrfs_key key
;
5974 sb
= btrfs_find_create_tree_block(root
, BTRFS_SUPER_INFO_OFFSET
,
5975 BTRFS_SUPER_INFO_SIZE
);
5978 btrfs_set_buffer_uptodate(sb
);
5979 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
5981 * The sb extent buffer is artifical and just used to read the system array.
5982 * btrfs_set_buffer_uptodate() call does not properly mark all it's
5983 * pages up-to-date when the page is larger: extent does not cover the
5984 * whole page and consequently check_page_uptodate does not find all
5985 * the page's extents up-to-date (the hole beyond sb),
5986 * write_extent_buffer then triggers a WARN_ON.
5988 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5989 * but sb spans only this function. Add an explicit SetPageUptodate call
5990 * to silence the warning eg. on PowerPC 64.
5992 if (PAGE_CACHE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
5993 SetPageUptodate(sb
->pages
[0]);
5995 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
5996 array_size
= btrfs_super_sys_array_size(super_copy
);
5998 ptr
= super_copy
->sys_chunk_array
;
5999 sb_ptr
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
6002 while (cur
< array_size
) {
6003 disk_key
= (struct btrfs_disk_key
*)ptr
;
6004 btrfs_disk_key_to_cpu(&key
, disk_key
);
6006 len
= sizeof(*disk_key
); ptr
+= len
;
6010 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6011 chunk
= (struct btrfs_chunk
*)sb_ptr
;
6012 ret
= read_one_chunk(root
, &key
, sb
, chunk
);
6015 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
6016 len
= btrfs_chunk_item_size(num_stripes
);
6025 free_extent_buffer(sb
);
6029 int btrfs_read_chunk_tree(struct btrfs_root
*root
)
6031 struct btrfs_path
*path
;
6032 struct extent_buffer
*leaf
;
6033 struct btrfs_key key
;
6034 struct btrfs_key found_key
;
6038 root
= root
->fs_info
->chunk_root
;
6040 path
= btrfs_alloc_path();
6044 mutex_lock(&uuid_mutex
);
6048 * Read all device items, and then all the chunk items. All
6049 * device items are found before any chunk item (their object id
6050 * is smaller than the lowest possible object id for a chunk
6051 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6053 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
6056 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6060 leaf
= path
->nodes
[0];
6061 slot
= path
->slots
[0];
6062 if (slot
>= btrfs_header_nritems(leaf
)) {
6063 ret
= btrfs_next_leaf(root
, path
);
6070 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6071 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
6072 struct btrfs_dev_item
*dev_item
;
6073 dev_item
= btrfs_item_ptr(leaf
, slot
,
6074 struct btrfs_dev_item
);
6075 ret
= read_one_dev(root
, leaf
, dev_item
);
6078 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6079 struct btrfs_chunk
*chunk
;
6080 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
6081 ret
= read_one_chunk(root
, &found_key
, leaf
, chunk
);
6089 unlock_chunks(root
);
6090 mutex_unlock(&uuid_mutex
);
6092 btrfs_free_path(path
);
6096 void btrfs_init_devices_late(struct btrfs_fs_info
*fs_info
)
6098 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6099 struct btrfs_device
*device
;
6101 mutex_lock(&fs_devices
->device_list_mutex
);
6102 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
)
6103 device
->dev_root
= fs_info
->dev_root
;
6104 mutex_unlock(&fs_devices
->device_list_mutex
);
6107 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
6111 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6112 btrfs_dev_stat_reset(dev
, i
);
6115 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
6117 struct btrfs_key key
;
6118 struct btrfs_key found_key
;
6119 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6120 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6121 struct extent_buffer
*eb
;
6124 struct btrfs_device
*device
;
6125 struct btrfs_path
*path
= NULL
;
6128 path
= btrfs_alloc_path();
6134 mutex_lock(&fs_devices
->device_list_mutex
);
6135 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6137 struct btrfs_dev_stats_item
*ptr
;
6140 key
.type
= BTRFS_DEV_STATS_KEY
;
6141 key
.offset
= device
->devid
;
6142 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
6144 __btrfs_reset_dev_stats(device
);
6145 device
->dev_stats_valid
= 1;
6146 btrfs_release_path(path
);
6149 slot
= path
->slots
[0];
6150 eb
= path
->nodes
[0];
6151 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
6152 item_size
= btrfs_item_size_nr(eb
, slot
);
6154 ptr
= btrfs_item_ptr(eb
, slot
,
6155 struct btrfs_dev_stats_item
);
6157 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6158 if (item_size
>= (1 + i
) * sizeof(__le64
))
6159 btrfs_dev_stat_set(device
, i
,
6160 btrfs_dev_stats_value(eb
, ptr
, i
));
6162 btrfs_dev_stat_reset(device
, i
);
6165 device
->dev_stats_valid
= 1;
6166 btrfs_dev_stat_print_on_load(device
);
6167 btrfs_release_path(path
);
6169 mutex_unlock(&fs_devices
->device_list_mutex
);
6172 btrfs_free_path(path
);
6173 return ret
< 0 ? ret
: 0;
6176 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
6177 struct btrfs_root
*dev_root
,
6178 struct btrfs_device
*device
)
6180 struct btrfs_path
*path
;
6181 struct btrfs_key key
;
6182 struct extent_buffer
*eb
;
6183 struct btrfs_dev_stats_item
*ptr
;
6188 key
.type
= BTRFS_DEV_STATS_KEY
;
6189 key
.offset
= device
->devid
;
6191 path
= btrfs_alloc_path();
6193 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
6195 printk_in_rcu(KERN_WARNING
"BTRFS: "
6196 "error %d while searching for dev_stats item for device %s!\n",
6197 ret
, rcu_str_deref(device
->name
));
6202 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
6203 /* need to delete old one and insert a new one */
6204 ret
= btrfs_del_item(trans
, dev_root
, path
);
6206 printk_in_rcu(KERN_WARNING
"BTRFS: "
6207 "delete too small dev_stats item for device %s failed %d!\n",
6208 rcu_str_deref(device
->name
), ret
);
6215 /* need to insert a new item */
6216 btrfs_release_path(path
);
6217 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
6218 &key
, sizeof(*ptr
));
6220 printk_in_rcu(KERN_WARNING
"BTRFS: "
6221 "insert dev_stats item for device %s failed %d!\n",
6222 rcu_str_deref(device
->name
), ret
);
6227 eb
= path
->nodes
[0];
6228 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
6229 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6230 btrfs_set_dev_stats_value(eb
, ptr
, i
,
6231 btrfs_dev_stat_read(device
, i
));
6232 btrfs_mark_buffer_dirty(eb
);
6235 btrfs_free_path(path
);
6240 * called from commit_transaction. Writes all changed device stats to disk.
6242 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
6243 struct btrfs_fs_info
*fs_info
)
6245 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6246 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6247 struct btrfs_device
*device
;
6250 mutex_lock(&fs_devices
->device_list_mutex
);
6251 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6252 if (!device
->dev_stats_valid
|| !device
->dev_stats_dirty
)
6255 ret
= update_dev_stat_item(trans
, dev_root
, device
);
6257 device
->dev_stats_dirty
= 0;
6259 mutex_unlock(&fs_devices
->device_list_mutex
);
6264 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
6266 btrfs_dev_stat_inc(dev
, index
);
6267 btrfs_dev_stat_print_on_error(dev
);
6270 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
6272 if (!dev
->dev_stats_valid
)
6274 printk_ratelimited_in_rcu(KERN_ERR
"BTRFS: "
6275 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6276 rcu_str_deref(dev
->name
),
6277 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
6278 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
6279 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
6280 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
6281 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
6284 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
6288 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6289 if (btrfs_dev_stat_read(dev
, i
) != 0)
6291 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
6292 return; /* all values == 0, suppress message */
6294 printk_in_rcu(KERN_INFO
"BTRFS: "
6295 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6296 rcu_str_deref(dev
->name
),
6297 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
6298 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
6299 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
6300 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
6301 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
6304 int btrfs_get_dev_stats(struct btrfs_root
*root
,
6305 struct btrfs_ioctl_get_dev_stats
*stats
)
6307 struct btrfs_device
*dev
;
6308 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
6311 mutex_lock(&fs_devices
->device_list_mutex
);
6312 dev
= btrfs_find_device(root
->fs_info
, stats
->devid
, NULL
, NULL
);
6313 mutex_unlock(&fs_devices
->device_list_mutex
);
6316 btrfs_warn(root
->fs_info
, "get dev_stats failed, device not found");
6318 } else if (!dev
->dev_stats_valid
) {
6319 btrfs_warn(root
->fs_info
, "get dev_stats failed, not yet valid");
6321 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
6322 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
6323 if (stats
->nr_items
> i
)
6325 btrfs_dev_stat_read_and_reset(dev
, i
);
6327 btrfs_dev_stat_reset(dev
, i
);
6330 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6331 if (stats
->nr_items
> i
)
6332 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
6334 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
6335 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
6339 int btrfs_scratch_superblock(struct btrfs_device
*device
)
6341 struct buffer_head
*bh
;
6342 struct btrfs_super_block
*disk_super
;
6344 bh
= btrfs_read_dev_super(device
->bdev
);
6347 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
6349 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
6350 set_buffer_dirty(bh
);
6351 sync_dirty_buffer(bh
);