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/iocontext.h>
24 #include <linux/capability.h>
25 #include <linux/ratelimit.h>
26 #include <linux/kthread.h>
27 #include <linux/raid/pq.h>
28 #include <linux/semaphore.h>
29 #include <linux/uuid.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"
45 const struct btrfs_raid_attr btrfs_raid_array
[BTRFS_NR_RAID_TYPES
] = {
46 [BTRFS_RAID_RAID10
] = {
49 .devs_max
= 0, /* 0 == as many as possible */
51 .tolerated_failures
= 1,
55 [BTRFS_RAID_RAID1
] = {
60 .tolerated_failures
= 1,
69 .tolerated_failures
= 0,
73 [BTRFS_RAID_RAID0
] = {
78 .tolerated_failures
= 0,
82 [BTRFS_RAID_SINGLE
] = {
87 .tolerated_failures
= 0,
91 [BTRFS_RAID_RAID5
] = {
96 .tolerated_failures
= 1,
100 [BTRFS_RAID_RAID6
] = {
105 .tolerated_failures
= 2,
111 const u64 btrfs_raid_group
[BTRFS_NR_RAID_TYPES
] = {
112 [BTRFS_RAID_RAID10
] = BTRFS_BLOCK_GROUP_RAID10
,
113 [BTRFS_RAID_RAID1
] = BTRFS_BLOCK_GROUP_RAID1
,
114 [BTRFS_RAID_DUP
] = BTRFS_BLOCK_GROUP_DUP
,
115 [BTRFS_RAID_RAID0
] = BTRFS_BLOCK_GROUP_RAID0
,
116 [BTRFS_RAID_SINGLE
] = 0,
117 [BTRFS_RAID_RAID5
] = BTRFS_BLOCK_GROUP_RAID5
,
118 [BTRFS_RAID_RAID6
] = BTRFS_BLOCK_GROUP_RAID6
,
122 * Table to convert BTRFS_RAID_* to the error code if minimum number of devices
123 * condition is not met. Zero means there's no corresponding
124 * BTRFS_ERROR_DEV_*_NOT_MET value.
126 const int btrfs_raid_mindev_error
[BTRFS_NR_RAID_TYPES
] = {
127 [BTRFS_RAID_RAID10
] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET
,
128 [BTRFS_RAID_RAID1
] = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET
,
129 [BTRFS_RAID_DUP
] = 0,
130 [BTRFS_RAID_RAID0
] = 0,
131 [BTRFS_RAID_SINGLE
] = 0,
132 [BTRFS_RAID_RAID5
] = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET
,
133 [BTRFS_RAID_RAID6
] = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET
,
136 static int init_first_rw_device(struct btrfs_trans_handle
*trans
,
137 struct btrfs_fs_info
*fs_info
);
138 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info
*fs_info
);
139 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
);
140 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
);
141 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*device
);
142 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
,
143 enum btrfs_map_op op
,
144 u64 logical
, u64
*length
,
145 struct btrfs_bio
**bbio_ret
,
146 int mirror_num
, int need_raid_map
);
148 DEFINE_MUTEX(uuid_mutex
);
149 static LIST_HEAD(fs_uuids
);
150 struct list_head
*btrfs_get_fs_uuids(void)
156 * alloc_fs_devices - allocate struct btrfs_fs_devices
157 * @fsid: if not NULL, copy the uuid to fs_devices::fsid
159 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
160 * The returned struct is not linked onto any lists and can be destroyed with
161 * kfree() right away.
163 static struct btrfs_fs_devices
*alloc_fs_devices(const u8
*fsid
)
165 struct btrfs_fs_devices
*fs_devs
;
167 fs_devs
= kzalloc(sizeof(*fs_devs
), GFP_KERNEL
);
169 return ERR_PTR(-ENOMEM
);
171 mutex_init(&fs_devs
->device_list_mutex
);
173 INIT_LIST_HEAD(&fs_devs
->devices
);
174 INIT_LIST_HEAD(&fs_devs
->resized_devices
);
175 INIT_LIST_HEAD(&fs_devs
->alloc_list
);
176 INIT_LIST_HEAD(&fs_devs
->list
);
178 memcpy(fs_devs
->fsid
, fsid
, BTRFS_FSID_SIZE
);
183 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
185 struct btrfs_device
*device
;
186 WARN_ON(fs_devices
->opened
);
187 while (!list_empty(&fs_devices
->devices
)) {
188 device
= list_entry(fs_devices
->devices
.next
,
189 struct btrfs_device
, dev_list
);
190 list_del(&device
->dev_list
);
191 rcu_string_free(device
->name
);
192 bio_put(device
->flush_bio
);
198 static void btrfs_kobject_uevent(struct block_device
*bdev
,
199 enum kobject_action action
)
203 ret
= kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, action
);
205 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
207 kobject_name(&disk_to_dev(bdev
->bd_disk
)->kobj
),
208 &disk_to_dev(bdev
->bd_disk
)->kobj
);
211 void btrfs_cleanup_fs_uuids(void)
213 struct btrfs_fs_devices
*fs_devices
;
215 while (!list_empty(&fs_uuids
)) {
216 fs_devices
= list_entry(fs_uuids
.next
,
217 struct btrfs_fs_devices
, list
);
218 list_del(&fs_devices
->list
);
219 free_fs_devices(fs_devices
);
223 static struct btrfs_device
*__alloc_device(void)
225 struct btrfs_device
*dev
;
227 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
229 return ERR_PTR(-ENOMEM
);
232 * Preallocate a bio that's always going to be used for flushing device
233 * barriers and matches the device lifespan
235 dev
->flush_bio
= bio_alloc_bioset(GFP_KERNEL
, 0, NULL
);
236 if (!dev
->flush_bio
) {
238 return ERR_PTR(-ENOMEM
);
241 INIT_LIST_HEAD(&dev
->dev_list
);
242 INIT_LIST_HEAD(&dev
->dev_alloc_list
);
243 INIT_LIST_HEAD(&dev
->resized_list
);
245 spin_lock_init(&dev
->io_lock
);
247 spin_lock_init(&dev
->reada_lock
);
248 atomic_set(&dev
->reada_in_flight
, 0);
249 atomic_set(&dev
->dev_stats_ccnt
, 0);
250 btrfs_device_data_ordered_init(dev
);
251 INIT_RADIX_TREE(&dev
->reada_zones
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
252 INIT_RADIX_TREE(&dev
->reada_extents
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
258 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
261 * If devid and uuid are both specified, the match must be exact, otherwise
262 * only devid is used.
264 static struct btrfs_device
*find_device(struct btrfs_fs_devices
*fs_devices
,
265 u64 devid
, const u8
*uuid
)
267 struct list_head
*head
= &fs_devices
->devices
;
268 struct btrfs_device
*dev
;
270 list_for_each_entry(dev
, head
, dev_list
) {
271 if (dev
->devid
== devid
&&
272 (!uuid
|| !memcmp(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
))) {
279 static noinline
struct btrfs_fs_devices
*find_fsid(u8
*fsid
)
281 struct btrfs_fs_devices
*fs_devices
;
283 list_for_each_entry(fs_devices
, &fs_uuids
, list
) {
284 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
291 btrfs_get_bdev_and_sb(const char *device_path
, fmode_t flags
, void *holder
,
292 int flush
, struct block_device
**bdev
,
293 struct buffer_head
**bh
)
297 *bdev
= blkdev_get_by_path(device_path
, flags
, holder
);
300 ret
= PTR_ERR(*bdev
);
305 filemap_write_and_wait((*bdev
)->bd_inode
->i_mapping
);
306 ret
= set_blocksize(*bdev
, BTRFS_BDEV_BLOCKSIZE
);
308 blkdev_put(*bdev
, flags
);
311 invalidate_bdev(*bdev
);
312 *bh
= btrfs_read_dev_super(*bdev
);
315 blkdev_put(*bdev
, flags
);
327 static void requeue_list(struct btrfs_pending_bios
*pending_bios
,
328 struct bio
*head
, struct bio
*tail
)
331 struct bio
*old_head
;
333 old_head
= pending_bios
->head
;
334 pending_bios
->head
= head
;
335 if (pending_bios
->tail
)
336 tail
->bi_next
= old_head
;
338 pending_bios
->tail
= tail
;
342 * we try to collect pending bios for a device so we don't get a large
343 * number of procs sending bios down to the same device. This greatly
344 * improves the schedulers ability to collect and merge the bios.
346 * But, it also turns into a long list of bios to process and that is sure
347 * to eventually make the worker thread block. The solution here is to
348 * make some progress and then put this work struct back at the end of
349 * the list if the block device is congested. This way, multiple devices
350 * can make progress from a single worker thread.
352 static noinline
void run_scheduled_bios(struct btrfs_device
*device
)
354 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
356 struct backing_dev_info
*bdi
;
357 struct btrfs_pending_bios
*pending_bios
;
361 unsigned long num_run
;
362 unsigned long batch_run
= 0;
363 unsigned long last_waited
= 0;
365 int sync_pending
= 0;
366 struct blk_plug plug
;
369 * this function runs all the bios we've collected for
370 * a particular device. We don't want to wander off to
371 * another device without first sending all of these down.
372 * So, setup a plug here and finish it off before we return
374 blk_start_plug(&plug
);
376 bdi
= device
->bdev
->bd_bdi
;
379 spin_lock(&device
->io_lock
);
384 /* take all the bios off the list at once and process them
385 * later on (without the lock held). But, remember the
386 * tail and other pointers so the bios can be properly reinserted
387 * into the list if we hit congestion
389 if (!force_reg
&& device
->pending_sync_bios
.head
) {
390 pending_bios
= &device
->pending_sync_bios
;
393 pending_bios
= &device
->pending_bios
;
397 pending
= pending_bios
->head
;
398 tail
= pending_bios
->tail
;
399 WARN_ON(pending
&& !tail
);
402 * if pending was null this time around, no bios need processing
403 * at all and we can stop. Otherwise it'll loop back up again
404 * and do an additional check so no bios are missed.
406 * device->running_pending is used to synchronize with the
409 if (device
->pending_sync_bios
.head
== NULL
&&
410 device
->pending_bios
.head
== NULL
) {
412 device
->running_pending
= 0;
415 device
->running_pending
= 1;
418 pending_bios
->head
= NULL
;
419 pending_bios
->tail
= NULL
;
421 spin_unlock(&device
->io_lock
);
426 /* we want to work on both lists, but do more bios on the
427 * sync list than the regular list
430 pending_bios
!= &device
->pending_sync_bios
&&
431 device
->pending_sync_bios
.head
) ||
432 (num_run
> 64 && pending_bios
== &device
->pending_sync_bios
&&
433 device
->pending_bios
.head
)) {
434 spin_lock(&device
->io_lock
);
435 requeue_list(pending_bios
, pending
, tail
);
440 pending
= pending
->bi_next
;
443 BUG_ON(atomic_read(&cur
->__bi_cnt
) == 0);
446 * if we're doing the sync list, record that our
447 * plug has some sync requests on it
449 * If we're doing the regular list and there are
450 * sync requests sitting around, unplug before
453 if (pending_bios
== &device
->pending_sync_bios
) {
455 } else if (sync_pending
) {
456 blk_finish_plug(&plug
);
457 blk_start_plug(&plug
);
461 btrfsic_submit_bio(cur
);
468 * we made progress, there is more work to do and the bdi
469 * is now congested. Back off and let other work structs
472 if (pending
&& bdi_write_congested(bdi
) && batch_run
> 8 &&
473 fs_info
->fs_devices
->open_devices
> 1) {
474 struct io_context
*ioc
;
476 ioc
= current
->io_context
;
479 * the main goal here is that we don't want to
480 * block if we're going to be able to submit
481 * more requests without blocking.
483 * This code does two great things, it pokes into
484 * the elevator code from a filesystem _and_
485 * it makes assumptions about how batching works.
487 if (ioc
&& ioc
->nr_batch_requests
> 0 &&
488 time_before(jiffies
, ioc
->last_waited
+ HZ
/50UL) &&
490 ioc
->last_waited
== last_waited
)) {
492 * we want to go through our batch of
493 * requests and stop. So, we copy out
494 * the ioc->last_waited time and test
495 * against it before looping
497 last_waited
= ioc
->last_waited
;
501 spin_lock(&device
->io_lock
);
502 requeue_list(pending_bios
, pending
, tail
);
503 device
->running_pending
= 1;
505 spin_unlock(&device
->io_lock
);
506 btrfs_queue_work(fs_info
->submit_workers
,
516 spin_lock(&device
->io_lock
);
517 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
519 spin_unlock(&device
->io_lock
);
522 blk_finish_plug(&plug
);
525 static void pending_bios_fn(struct btrfs_work
*work
)
527 struct btrfs_device
*device
;
529 device
= container_of(work
, struct btrfs_device
, work
);
530 run_scheduled_bios(device
);
534 static void btrfs_free_stale_device(struct btrfs_device
*cur_dev
)
536 struct btrfs_fs_devices
*fs_devs
;
537 struct btrfs_device
*dev
;
542 list_for_each_entry(fs_devs
, &fs_uuids
, list
) {
547 if (fs_devs
->seeding
)
550 list_for_each_entry(dev
, &fs_devs
->devices
, dev_list
) {
558 * Todo: This won't be enough. What if the same device
559 * comes back (with new uuid and) with its mapper path?
560 * But for now, this does help as mostly an admin will
561 * either use mapper or non mapper path throughout.
564 del
= strcmp(rcu_str_deref(dev
->name
),
565 rcu_str_deref(cur_dev
->name
));
572 /* delete the stale device */
573 if (fs_devs
->num_devices
== 1) {
574 btrfs_sysfs_remove_fsid(fs_devs
);
575 list_del(&fs_devs
->list
);
576 free_fs_devices(fs_devs
);
579 fs_devs
->num_devices
--;
580 list_del(&dev
->dev_list
);
581 rcu_string_free(dev
->name
);
582 bio_put(dev
->flush_bio
);
591 * Add new device to list of registered devices
594 * 1 - first time device is seen
595 * 0 - device already known
598 static noinline
int device_list_add(const char *path
,
599 struct btrfs_super_block
*disk_super
,
600 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
602 struct btrfs_device
*device
;
603 struct btrfs_fs_devices
*fs_devices
;
604 struct rcu_string
*name
;
606 u64 found_transid
= btrfs_super_generation(disk_super
);
608 fs_devices
= find_fsid(disk_super
->fsid
);
610 fs_devices
= alloc_fs_devices(disk_super
->fsid
);
611 if (IS_ERR(fs_devices
))
612 return PTR_ERR(fs_devices
);
614 list_add(&fs_devices
->list
, &fs_uuids
);
618 device
= find_device(fs_devices
, devid
,
619 disk_super
->dev_item
.uuid
);
623 if (fs_devices
->opened
)
626 device
= btrfs_alloc_device(NULL
, &devid
,
627 disk_super
->dev_item
.uuid
);
628 if (IS_ERR(device
)) {
629 /* we can safely leave the fs_devices entry around */
630 return PTR_ERR(device
);
633 name
= rcu_string_strdup(path
, GFP_NOFS
);
635 bio_put(device
->flush_bio
);
639 rcu_assign_pointer(device
->name
, name
);
641 mutex_lock(&fs_devices
->device_list_mutex
);
642 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
643 fs_devices
->num_devices
++;
644 mutex_unlock(&fs_devices
->device_list_mutex
);
647 device
->fs_devices
= fs_devices
;
648 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
650 * When FS is already mounted.
651 * 1. If you are here and if the device->name is NULL that
652 * means this device was missing at time of FS mount.
653 * 2. If you are here and if the device->name is different
654 * from 'path' that means either
655 * a. The same device disappeared and reappeared with
657 * b. The missing-disk-which-was-replaced, has
660 * We must allow 1 and 2a above. But 2b would be a spurious
663 * Further in case of 1 and 2a above, the disk at 'path'
664 * would have missed some transaction when it was away and
665 * in case of 2a the stale bdev has to be updated as well.
666 * 2b must not be allowed at all time.
670 * For now, we do allow update to btrfs_fs_device through the
671 * btrfs dev scan cli after FS has been mounted. We're still
672 * tracking a problem where systems fail mount by subvolume id
673 * when we reject replacement on a mounted FS.
675 if (!fs_devices
->opened
&& found_transid
< device
->generation
) {
677 * That is if the FS is _not_ mounted and if you
678 * are here, that means there is more than one
679 * disk with same uuid and devid.We keep the one
680 * with larger generation number or the last-in if
681 * generation are equal.
686 name
= rcu_string_strdup(path
, GFP_NOFS
);
689 rcu_string_free(device
->name
);
690 rcu_assign_pointer(device
->name
, name
);
691 if (device
->missing
) {
692 fs_devices
->missing_devices
--;
698 * Unmount does not free the btrfs_device struct but would zero
699 * generation along with most of the other members. So just update
700 * it back. We need it to pick the disk with largest generation
703 if (!fs_devices
->opened
)
704 device
->generation
= found_transid
;
707 * if there is new btrfs on an already registered device,
708 * then remove the stale device entry.
711 btrfs_free_stale_device(device
);
713 *fs_devices_ret
= fs_devices
;
718 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
720 struct btrfs_fs_devices
*fs_devices
;
721 struct btrfs_device
*device
;
722 struct btrfs_device
*orig_dev
;
724 fs_devices
= alloc_fs_devices(orig
->fsid
);
725 if (IS_ERR(fs_devices
))
728 mutex_lock(&orig
->device_list_mutex
);
729 fs_devices
->total_devices
= orig
->total_devices
;
731 /* We have held the volume lock, it is safe to get the devices. */
732 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
733 struct rcu_string
*name
;
735 device
= btrfs_alloc_device(NULL
, &orig_dev
->devid
,
741 * This is ok to do without rcu read locked because we hold the
742 * uuid mutex so nothing we touch in here is going to disappear.
744 if (orig_dev
->name
) {
745 name
= rcu_string_strdup(orig_dev
->name
->str
,
748 bio_put(device
->flush_bio
);
752 rcu_assign_pointer(device
->name
, name
);
755 list_add(&device
->dev_list
, &fs_devices
->devices
);
756 device
->fs_devices
= fs_devices
;
757 fs_devices
->num_devices
++;
759 mutex_unlock(&orig
->device_list_mutex
);
762 mutex_unlock(&orig
->device_list_mutex
);
763 free_fs_devices(fs_devices
);
764 return ERR_PTR(-ENOMEM
);
767 void btrfs_close_extra_devices(struct btrfs_fs_devices
*fs_devices
, int step
)
769 struct btrfs_device
*device
, *next
;
770 struct btrfs_device
*latest_dev
= NULL
;
772 mutex_lock(&uuid_mutex
);
774 /* This is the initialized path, it is safe to release the devices. */
775 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
776 if (device
->in_fs_metadata
) {
777 if (!device
->is_tgtdev_for_dev_replace
&&
779 device
->generation
> latest_dev
->generation
)) {
785 if (device
->devid
== BTRFS_DEV_REPLACE_DEVID
) {
787 * In the first step, keep the device which has
788 * the correct fsid and the devid that is used
789 * for the dev_replace procedure.
790 * In the second step, the dev_replace state is
791 * read from the device tree and it is known
792 * whether the procedure is really active or
793 * not, which means whether this device is
794 * used or whether it should be removed.
796 if (step
== 0 || device
->is_tgtdev_for_dev_replace
) {
801 blkdev_put(device
->bdev
, device
->mode
);
803 fs_devices
->open_devices
--;
805 if (device
->writeable
) {
806 list_del_init(&device
->dev_alloc_list
);
807 device
->writeable
= 0;
808 if (!device
->is_tgtdev_for_dev_replace
)
809 fs_devices
->rw_devices
--;
811 list_del_init(&device
->dev_list
);
812 fs_devices
->num_devices
--;
813 rcu_string_free(device
->name
);
814 bio_put(device
->flush_bio
);
818 if (fs_devices
->seed
) {
819 fs_devices
= fs_devices
->seed
;
823 fs_devices
->latest_bdev
= latest_dev
->bdev
;
825 mutex_unlock(&uuid_mutex
);
828 static void __free_device(struct work_struct
*work
)
830 struct btrfs_device
*device
;
832 device
= container_of(work
, struct btrfs_device
, rcu_work
);
833 rcu_string_free(device
->name
);
834 bio_put(device
->flush_bio
);
838 static void free_device(struct rcu_head
*head
)
840 struct btrfs_device
*device
;
842 device
= container_of(head
, struct btrfs_device
, rcu
);
844 INIT_WORK(&device
->rcu_work
, __free_device
);
845 schedule_work(&device
->rcu_work
);
848 static void btrfs_close_bdev(struct btrfs_device
*device
)
850 if (device
->bdev
&& device
->writeable
) {
851 sync_blockdev(device
->bdev
);
852 invalidate_bdev(device
->bdev
);
856 blkdev_put(device
->bdev
, device
->mode
);
859 static void btrfs_prepare_close_one_device(struct btrfs_device
*device
)
861 struct btrfs_fs_devices
*fs_devices
= device
->fs_devices
;
862 struct btrfs_device
*new_device
;
863 struct rcu_string
*name
;
866 fs_devices
->open_devices
--;
868 if (device
->writeable
&&
869 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
870 list_del_init(&device
->dev_alloc_list
);
871 fs_devices
->rw_devices
--;
875 fs_devices
->missing_devices
--;
877 new_device
= btrfs_alloc_device(NULL
, &device
->devid
,
879 BUG_ON(IS_ERR(new_device
)); /* -ENOMEM */
881 /* Safe because we are under uuid_mutex */
883 name
= rcu_string_strdup(device
->name
->str
, GFP_NOFS
);
884 BUG_ON(!name
); /* -ENOMEM */
885 rcu_assign_pointer(new_device
->name
, name
);
888 list_replace_rcu(&device
->dev_list
, &new_device
->dev_list
);
889 new_device
->fs_devices
= device
->fs_devices
;
892 static int __btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
894 struct btrfs_device
*device
, *tmp
;
895 struct list_head pending_put
;
897 INIT_LIST_HEAD(&pending_put
);
899 if (--fs_devices
->opened
> 0)
902 mutex_lock(&fs_devices
->device_list_mutex
);
903 list_for_each_entry_safe(device
, tmp
, &fs_devices
->devices
, dev_list
) {
904 btrfs_prepare_close_one_device(device
);
905 list_add(&device
->dev_list
, &pending_put
);
907 mutex_unlock(&fs_devices
->device_list_mutex
);
910 * btrfs_show_devname() is using the device_list_mutex,
911 * sometimes call to blkdev_put() leads vfs calling
912 * into this func. So do put outside of device_list_mutex,
915 while (!list_empty(&pending_put
)) {
916 device
= list_first_entry(&pending_put
,
917 struct btrfs_device
, dev_list
);
918 list_del(&device
->dev_list
);
919 btrfs_close_bdev(device
);
920 call_rcu(&device
->rcu
, free_device
);
923 WARN_ON(fs_devices
->open_devices
);
924 WARN_ON(fs_devices
->rw_devices
);
925 fs_devices
->opened
= 0;
926 fs_devices
->seeding
= 0;
931 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
933 struct btrfs_fs_devices
*seed_devices
= NULL
;
936 mutex_lock(&uuid_mutex
);
937 ret
= __btrfs_close_devices(fs_devices
);
938 if (!fs_devices
->opened
) {
939 seed_devices
= fs_devices
->seed
;
940 fs_devices
->seed
= NULL
;
942 mutex_unlock(&uuid_mutex
);
944 while (seed_devices
) {
945 fs_devices
= seed_devices
;
946 seed_devices
= fs_devices
->seed
;
947 __btrfs_close_devices(fs_devices
);
948 free_fs_devices(fs_devices
);
951 * Wait for rcu kworkers under __btrfs_close_devices
952 * to finish all blkdev_puts so device is really
953 * free when umount is done.
959 static int __btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
960 fmode_t flags
, void *holder
)
962 struct request_queue
*q
;
963 struct block_device
*bdev
;
964 struct list_head
*head
= &fs_devices
->devices
;
965 struct btrfs_device
*device
;
966 struct btrfs_device
*latest_dev
= NULL
;
967 struct buffer_head
*bh
;
968 struct btrfs_super_block
*disk_super
;
975 list_for_each_entry(device
, head
, dev_list
) {
981 /* Just open everything we can; ignore failures here */
982 if (btrfs_get_bdev_and_sb(device
->name
->str
, flags
, holder
, 1,
986 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
987 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
988 if (devid
!= device
->devid
)
991 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
,
995 device
->generation
= btrfs_super_generation(disk_super
);
997 device
->generation
> latest_dev
->generation
)
1000 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
1001 device
->writeable
= 0;
1003 device
->writeable
= !bdev_read_only(bdev
);
1007 q
= bdev_get_queue(bdev
);
1008 if (blk_queue_discard(q
))
1009 device
->can_discard
= 1;
1010 if (!blk_queue_nonrot(q
))
1011 fs_devices
->rotating
= 1;
1013 device
->bdev
= bdev
;
1014 device
->in_fs_metadata
= 0;
1015 device
->mode
= flags
;
1017 fs_devices
->open_devices
++;
1018 if (device
->writeable
&&
1019 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
1020 fs_devices
->rw_devices
++;
1021 list_add(&device
->dev_alloc_list
,
1022 &fs_devices
->alloc_list
);
1029 blkdev_put(bdev
, flags
);
1032 if (fs_devices
->open_devices
== 0) {
1036 fs_devices
->seeding
= seeding
;
1037 fs_devices
->opened
= 1;
1038 fs_devices
->latest_bdev
= latest_dev
->bdev
;
1039 fs_devices
->total_rw_bytes
= 0;
1044 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
1045 fmode_t flags
, void *holder
)
1049 mutex_lock(&uuid_mutex
);
1050 if (fs_devices
->opened
) {
1051 fs_devices
->opened
++;
1054 ret
= __btrfs_open_devices(fs_devices
, flags
, holder
);
1056 mutex_unlock(&uuid_mutex
);
1060 static void btrfs_release_disk_super(struct page
*page
)
1066 static int btrfs_read_disk_super(struct block_device
*bdev
, u64 bytenr
,
1068 struct btrfs_super_block
**disk_super
)
1073 /* make sure our super fits in the device */
1074 if (bytenr
+ PAGE_SIZE
>= i_size_read(bdev
->bd_inode
))
1077 /* make sure our super fits in the page */
1078 if (sizeof(**disk_super
) > PAGE_SIZE
)
1081 /* make sure our super doesn't straddle pages on disk */
1082 index
= bytenr
>> PAGE_SHIFT
;
1083 if ((bytenr
+ sizeof(**disk_super
) - 1) >> PAGE_SHIFT
!= index
)
1086 /* pull in the page with our super */
1087 *page
= read_cache_page_gfp(bdev
->bd_inode
->i_mapping
,
1090 if (IS_ERR_OR_NULL(*page
))
1095 /* align our pointer to the offset of the super block */
1096 *disk_super
= p
+ (bytenr
& ~PAGE_MASK
);
1098 if (btrfs_super_bytenr(*disk_super
) != bytenr
||
1099 btrfs_super_magic(*disk_super
) != BTRFS_MAGIC
) {
1100 btrfs_release_disk_super(*page
);
1104 if ((*disk_super
)->label
[0] &&
1105 (*disk_super
)->label
[BTRFS_LABEL_SIZE
- 1])
1106 (*disk_super
)->label
[BTRFS_LABEL_SIZE
- 1] = '\0';
1112 * Look for a btrfs signature on a device. This may be called out of the mount path
1113 * and we are not allowed to call set_blocksize during the scan. The superblock
1114 * is read via pagecache
1116 int btrfs_scan_one_device(const char *path
, fmode_t flags
, void *holder
,
1117 struct btrfs_fs_devices
**fs_devices_ret
)
1119 struct btrfs_super_block
*disk_super
;
1120 struct block_device
*bdev
;
1129 * we would like to check all the supers, but that would make
1130 * a btrfs mount succeed after a mkfs from a different FS.
1131 * So, we need to add a special mount option to scan for
1132 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1134 bytenr
= btrfs_sb_offset(0);
1135 flags
|= FMODE_EXCL
;
1136 mutex_lock(&uuid_mutex
);
1138 bdev
= blkdev_get_by_path(path
, flags
, holder
);
1140 ret
= PTR_ERR(bdev
);
1144 if (btrfs_read_disk_super(bdev
, bytenr
, &page
, &disk_super
))
1145 goto error_bdev_put
;
1147 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
1148 transid
= btrfs_super_generation(disk_super
);
1149 total_devices
= btrfs_super_num_devices(disk_super
);
1151 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
1153 if (disk_super
->label
[0]) {
1154 pr_info("BTRFS: device label %s ", disk_super
->label
);
1156 pr_info("BTRFS: device fsid %pU ", disk_super
->fsid
);
1159 pr_cont("devid %llu transid %llu %s\n", devid
, transid
, path
);
1162 if (!ret
&& fs_devices_ret
)
1163 (*fs_devices_ret
)->total_devices
= total_devices
;
1165 btrfs_release_disk_super(page
);
1168 blkdev_put(bdev
, flags
);
1170 mutex_unlock(&uuid_mutex
);
1174 /* helper to account the used device space in the range */
1175 int btrfs_account_dev_extents_size(struct btrfs_device
*device
, u64 start
,
1176 u64 end
, u64
*length
)
1178 struct btrfs_key key
;
1179 struct btrfs_root
*root
= device
->fs_info
->dev_root
;
1180 struct btrfs_dev_extent
*dev_extent
;
1181 struct btrfs_path
*path
;
1185 struct extent_buffer
*l
;
1189 if (start
>= device
->total_bytes
|| device
->is_tgtdev_for_dev_replace
)
1192 path
= btrfs_alloc_path();
1195 path
->reada
= READA_FORWARD
;
1197 key
.objectid
= device
->devid
;
1199 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1201 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1205 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1212 slot
= path
->slots
[0];
1213 if (slot
>= btrfs_header_nritems(l
)) {
1214 ret
= btrfs_next_leaf(root
, path
);
1222 btrfs_item_key_to_cpu(l
, &key
, slot
);
1224 if (key
.objectid
< device
->devid
)
1227 if (key
.objectid
> device
->devid
)
1230 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1233 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1234 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1236 if (key
.offset
<= start
&& extent_end
> end
) {
1237 *length
= end
- start
+ 1;
1239 } else if (key
.offset
<= start
&& extent_end
> start
)
1240 *length
+= extent_end
- start
;
1241 else if (key
.offset
> start
&& extent_end
<= end
)
1242 *length
+= extent_end
- key
.offset
;
1243 else if (key
.offset
> start
&& key
.offset
<= end
) {
1244 *length
+= end
- key
.offset
+ 1;
1246 } else if (key
.offset
> end
)
1254 btrfs_free_path(path
);
1258 static int contains_pending_extent(struct btrfs_transaction
*transaction
,
1259 struct btrfs_device
*device
,
1260 u64
*start
, u64 len
)
1262 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1263 struct extent_map
*em
;
1264 struct list_head
*search_list
= &fs_info
->pinned_chunks
;
1266 u64 physical_start
= *start
;
1269 search_list
= &transaction
->pending_chunks
;
1271 list_for_each_entry(em
, search_list
, list
) {
1272 struct map_lookup
*map
;
1275 map
= em
->map_lookup
;
1276 for (i
= 0; i
< map
->num_stripes
; i
++) {
1279 if (map
->stripes
[i
].dev
!= device
)
1281 if (map
->stripes
[i
].physical
>= physical_start
+ len
||
1282 map
->stripes
[i
].physical
+ em
->orig_block_len
<=
1286 * Make sure that while processing the pinned list we do
1287 * not override our *start with a lower value, because
1288 * we can have pinned chunks that fall within this
1289 * device hole and that have lower physical addresses
1290 * than the pending chunks we processed before. If we
1291 * do not take this special care we can end up getting
1292 * 2 pending chunks that start at the same physical
1293 * device offsets because the end offset of a pinned
1294 * chunk can be equal to the start offset of some
1297 end
= map
->stripes
[i
].physical
+ em
->orig_block_len
;
1304 if (search_list
!= &fs_info
->pinned_chunks
) {
1305 search_list
= &fs_info
->pinned_chunks
;
1314 * find_free_dev_extent_start - find free space in the specified device
1315 * @device: the device which we search the free space in
1316 * @num_bytes: the size of the free space that we need
1317 * @search_start: the position from which to begin the search
1318 * @start: store the start of the free space.
1319 * @len: the size of the free space. that we find, or the size
1320 * of the max free space if we don't find suitable free space
1322 * this uses a pretty simple search, the expectation is that it is
1323 * called very infrequently and that a given device has a small number
1326 * @start is used to store the start of the free space if we find. But if we
1327 * don't find suitable free space, it will be used to store the start position
1328 * of the max free space.
1330 * @len is used to store the size of the free space that we find.
1331 * But if we don't find suitable free space, it is used to store the size of
1332 * the max free space.
1334 int find_free_dev_extent_start(struct btrfs_transaction
*transaction
,
1335 struct btrfs_device
*device
, u64 num_bytes
,
1336 u64 search_start
, u64
*start
, u64
*len
)
1338 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1339 struct btrfs_root
*root
= fs_info
->dev_root
;
1340 struct btrfs_key key
;
1341 struct btrfs_dev_extent
*dev_extent
;
1342 struct btrfs_path
*path
;
1347 u64 search_end
= device
->total_bytes
;
1350 struct extent_buffer
*l
;
1353 * We don't want to overwrite the superblock on the drive nor any area
1354 * used by the boot loader (grub for example), so we make sure to start
1355 * at an offset of at least 1MB.
1357 search_start
= max_t(u64
, search_start
, SZ_1M
);
1359 path
= btrfs_alloc_path();
1363 max_hole_start
= search_start
;
1367 if (search_start
>= search_end
|| device
->is_tgtdev_for_dev_replace
) {
1372 path
->reada
= READA_FORWARD
;
1373 path
->search_commit_root
= 1;
1374 path
->skip_locking
= 1;
1376 key
.objectid
= device
->devid
;
1377 key
.offset
= search_start
;
1378 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1380 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1384 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1391 slot
= path
->slots
[0];
1392 if (slot
>= btrfs_header_nritems(l
)) {
1393 ret
= btrfs_next_leaf(root
, path
);
1401 btrfs_item_key_to_cpu(l
, &key
, slot
);
1403 if (key
.objectid
< device
->devid
)
1406 if (key
.objectid
> device
->devid
)
1409 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1412 if (key
.offset
> search_start
) {
1413 hole_size
= key
.offset
- search_start
;
1416 * Have to check before we set max_hole_start, otherwise
1417 * we could end up sending back this offset anyway.
1419 if (contains_pending_extent(transaction
, device
,
1422 if (key
.offset
>= search_start
) {
1423 hole_size
= key
.offset
- search_start
;
1430 if (hole_size
> max_hole_size
) {
1431 max_hole_start
= search_start
;
1432 max_hole_size
= hole_size
;
1436 * If this free space is greater than which we need,
1437 * it must be the max free space that we have found
1438 * until now, so max_hole_start must point to the start
1439 * of this free space and the length of this free space
1440 * is stored in max_hole_size. Thus, we return
1441 * max_hole_start and max_hole_size and go back to the
1444 if (hole_size
>= num_bytes
) {
1450 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1451 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1453 if (extent_end
> search_start
)
1454 search_start
= extent_end
;
1461 * At this point, search_start should be the end of
1462 * allocated dev extents, and when shrinking the device,
1463 * search_end may be smaller than search_start.
1465 if (search_end
> search_start
) {
1466 hole_size
= search_end
- search_start
;
1468 if (contains_pending_extent(transaction
, device
, &search_start
,
1470 btrfs_release_path(path
);
1474 if (hole_size
> max_hole_size
) {
1475 max_hole_start
= search_start
;
1476 max_hole_size
= hole_size
;
1481 if (max_hole_size
< num_bytes
)
1487 btrfs_free_path(path
);
1488 *start
= max_hole_start
;
1490 *len
= max_hole_size
;
1494 int find_free_dev_extent(struct btrfs_trans_handle
*trans
,
1495 struct btrfs_device
*device
, u64 num_bytes
,
1496 u64
*start
, u64
*len
)
1498 /* FIXME use last free of some kind */
1499 return find_free_dev_extent_start(trans
->transaction
, device
,
1500 num_bytes
, 0, start
, len
);
1503 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1504 struct btrfs_device
*device
,
1505 u64 start
, u64
*dev_extent_len
)
1507 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1508 struct btrfs_root
*root
= fs_info
->dev_root
;
1510 struct btrfs_path
*path
;
1511 struct btrfs_key key
;
1512 struct btrfs_key found_key
;
1513 struct extent_buffer
*leaf
= NULL
;
1514 struct btrfs_dev_extent
*extent
= NULL
;
1516 path
= btrfs_alloc_path();
1520 key
.objectid
= device
->devid
;
1522 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1524 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1526 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1527 BTRFS_DEV_EXTENT_KEY
);
1530 leaf
= path
->nodes
[0];
1531 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1532 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1533 struct btrfs_dev_extent
);
1534 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1535 btrfs_dev_extent_length(leaf
, extent
) < start
);
1537 btrfs_release_path(path
);
1539 } else if (ret
== 0) {
1540 leaf
= path
->nodes
[0];
1541 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1542 struct btrfs_dev_extent
);
1544 btrfs_handle_fs_error(fs_info
, ret
, "Slot search failed");
1548 *dev_extent_len
= btrfs_dev_extent_length(leaf
, extent
);
1550 ret
= btrfs_del_item(trans
, root
, path
);
1552 btrfs_handle_fs_error(fs_info
, ret
,
1553 "Failed to remove dev extent item");
1555 set_bit(BTRFS_TRANS_HAVE_FREE_BGS
, &trans
->transaction
->flags
);
1558 btrfs_free_path(path
);
1562 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1563 struct btrfs_device
*device
,
1564 u64 chunk_offset
, u64 start
, u64 num_bytes
)
1567 struct btrfs_path
*path
;
1568 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1569 struct btrfs_root
*root
= fs_info
->dev_root
;
1570 struct btrfs_dev_extent
*extent
;
1571 struct extent_buffer
*leaf
;
1572 struct btrfs_key key
;
1574 WARN_ON(!device
->in_fs_metadata
);
1575 WARN_ON(device
->is_tgtdev_for_dev_replace
);
1576 path
= btrfs_alloc_path();
1580 key
.objectid
= device
->devid
;
1582 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1583 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1588 leaf
= path
->nodes
[0];
1589 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1590 struct btrfs_dev_extent
);
1591 btrfs_set_dev_extent_chunk_tree(leaf
, extent
,
1592 BTRFS_CHUNK_TREE_OBJECTID
);
1593 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
,
1594 BTRFS_FIRST_CHUNK_TREE_OBJECTID
);
1595 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1597 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1598 btrfs_mark_buffer_dirty(leaf
);
1600 btrfs_free_path(path
);
1604 static u64
find_next_chunk(struct btrfs_fs_info
*fs_info
)
1606 struct extent_map_tree
*em_tree
;
1607 struct extent_map
*em
;
1611 em_tree
= &fs_info
->mapping_tree
.map_tree
;
1612 read_lock(&em_tree
->lock
);
1613 n
= rb_last(&em_tree
->map
);
1615 em
= rb_entry(n
, struct extent_map
, rb_node
);
1616 ret
= em
->start
+ em
->len
;
1618 read_unlock(&em_tree
->lock
);
1623 static noinline
int find_next_devid(struct btrfs_fs_info
*fs_info
,
1627 struct btrfs_key key
;
1628 struct btrfs_key found_key
;
1629 struct btrfs_path
*path
;
1631 path
= btrfs_alloc_path();
1635 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1636 key
.type
= BTRFS_DEV_ITEM_KEY
;
1637 key
.offset
= (u64
)-1;
1639 ret
= btrfs_search_slot(NULL
, fs_info
->chunk_root
, &key
, path
, 0, 0);
1643 BUG_ON(ret
== 0); /* Corruption */
1645 ret
= btrfs_previous_item(fs_info
->chunk_root
, path
,
1646 BTRFS_DEV_ITEMS_OBJECTID
,
1647 BTRFS_DEV_ITEM_KEY
);
1651 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1653 *devid_ret
= found_key
.offset
+ 1;
1657 btrfs_free_path(path
);
1662 * the device information is stored in the chunk root
1663 * the btrfs_device struct should be fully filled in
1665 static int btrfs_add_device(struct btrfs_trans_handle
*trans
,
1666 struct btrfs_fs_info
*fs_info
,
1667 struct btrfs_device
*device
)
1669 struct btrfs_root
*root
= fs_info
->chunk_root
;
1671 struct btrfs_path
*path
;
1672 struct btrfs_dev_item
*dev_item
;
1673 struct extent_buffer
*leaf
;
1674 struct btrfs_key key
;
1677 path
= btrfs_alloc_path();
1681 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1682 key
.type
= BTRFS_DEV_ITEM_KEY
;
1683 key
.offset
= device
->devid
;
1685 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1690 leaf
= path
->nodes
[0];
1691 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1693 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1694 btrfs_set_device_generation(leaf
, dev_item
, 0);
1695 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1696 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1697 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1698 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1699 btrfs_set_device_total_bytes(leaf
, dev_item
,
1700 btrfs_device_get_disk_total_bytes(device
));
1701 btrfs_set_device_bytes_used(leaf
, dev_item
,
1702 btrfs_device_get_bytes_used(device
));
1703 btrfs_set_device_group(leaf
, dev_item
, 0);
1704 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1705 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1706 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1708 ptr
= btrfs_device_uuid(dev_item
);
1709 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1710 ptr
= btrfs_device_fsid(dev_item
);
1711 write_extent_buffer(leaf
, fs_info
->fsid
, ptr
, BTRFS_FSID_SIZE
);
1712 btrfs_mark_buffer_dirty(leaf
);
1716 btrfs_free_path(path
);
1721 * Function to update ctime/mtime for a given device path.
1722 * Mainly used for ctime/mtime based probe like libblkid.
1724 static void update_dev_time(const char *path_name
)
1728 filp
= filp_open(path_name
, O_RDWR
, 0);
1731 file_update_time(filp
);
1732 filp_close(filp
, NULL
);
1735 static int btrfs_rm_dev_item(struct btrfs_fs_info
*fs_info
,
1736 struct btrfs_device
*device
)
1738 struct btrfs_root
*root
= fs_info
->chunk_root
;
1740 struct btrfs_path
*path
;
1741 struct btrfs_key key
;
1742 struct btrfs_trans_handle
*trans
;
1744 path
= btrfs_alloc_path();
1748 trans
= btrfs_start_transaction(root
, 0);
1749 if (IS_ERR(trans
)) {
1750 btrfs_free_path(path
);
1751 return PTR_ERR(trans
);
1753 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1754 key
.type
= BTRFS_DEV_ITEM_KEY
;
1755 key
.offset
= device
->devid
;
1757 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1761 btrfs_abort_transaction(trans
, ret
);
1762 btrfs_end_transaction(trans
);
1766 ret
= btrfs_del_item(trans
, root
, path
);
1768 btrfs_abort_transaction(trans
, ret
);
1769 btrfs_end_transaction(trans
);
1773 btrfs_free_path(path
);
1775 ret
= btrfs_commit_transaction(trans
);
1780 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1781 * filesystem. It's up to the caller to adjust that number regarding eg. device
1784 static int btrfs_check_raid_min_devices(struct btrfs_fs_info
*fs_info
,
1792 seq
= read_seqbegin(&fs_info
->profiles_lock
);
1794 all_avail
= fs_info
->avail_data_alloc_bits
|
1795 fs_info
->avail_system_alloc_bits
|
1796 fs_info
->avail_metadata_alloc_bits
;
1797 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
1799 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
1800 if (!(all_avail
& btrfs_raid_group
[i
]))
1803 if (num_devices
< btrfs_raid_array
[i
].devs_min
) {
1804 int ret
= btrfs_raid_mindev_error
[i
];
1814 static struct btrfs_device
* btrfs_find_next_active_device(
1815 struct btrfs_fs_devices
*fs_devs
, struct btrfs_device
*device
)
1817 struct btrfs_device
*next_device
;
1819 list_for_each_entry(next_device
, &fs_devs
->devices
, dev_list
) {
1820 if (next_device
!= device
&&
1821 !next_device
->missing
&& next_device
->bdev
)
1829 * Helper function to check if the given device is part of s_bdev / latest_bdev
1830 * and replace it with the provided or the next active device, in the context
1831 * where this function called, there should be always be another device (or
1832 * this_dev) which is active.
1834 void btrfs_assign_next_active_device(struct btrfs_fs_info
*fs_info
,
1835 struct btrfs_device
*device
, struct btrfs_device
*this_dev
)
1837 struct btrfs_device
*next_device
;
1840 next_device
= this_dev
;
1842 next_device
= btrfs_find_next_active_device(fs_info
->fs_devices
,
1844 ASSERT(next_device
);
1846 if (fs_info
->sb
->s_bdev
&&
1847 (fs_info
->sb
->s_bdev
== device
->bdev
))
1848 fs_info
->sb
->s_bdev
= next_device
->bdev
;
1850 if (fs_info
->fs_devices
->latest_bdev
== device
->bdev
)
1851 fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1854 int btrfs_rm_device(struct btrfs_fs_info
*fs_info
, const char *device_path
,
1857 struct btrfs_device
*device
;
1858 struct btrfs_fs_devices
*cur_devices
;
1862 mutex_lock(&uuid_mutex
);
1864 num_devices
= fs_info
->fs_devices
->num_devices
;
1865 btrfs_dev_replace_lock(&fs_info
->dev_replace
, 0);
1866 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
1867 WARN_ON(num_devices
< 1);
1870 btrfs_dev_replace_unlock(&fs_info
->dev_replace
, 0);
1872 ret
= btrfs_check_raid_min_devices(fs_info
, num_devices
- 1);
1876 ret
= btrfs_find_device_by_devspec(fs_info
, devid
, device_path
,
1881 if (device
->is_tgtdev_for_dev_replace
) {
1882 ret
= BTRFS_ERROR_DEV_TGT_REPLACE
;
1886 if (device
->writeable
&& fs_info
->fs_devices
->rw_devices
== 1) {
1887 ret
= BTRFS_ERROR_DEV_ONLY_WRITABLE
;
1891 if (device
->writeable
) {
1892 mutex_lock(&fs_info
->chunk_mutex
);
1893 list_del_init(&device
->dev_alloc_list
);
1894 device
->fs_devices
->rw_devices
--;
1895 mutex_unlock(&fs_info
->chunk_mutex
);
1898 mutex_unlock(&uuid_mutex
);
1899 ret
= btrfs_shrink_device(device
, 0);
1900 mutex_lock(&uuid_mutex
);
1905 * TODO: the superblock still includes this device in its num_devices
1906 * counter although write_all_supers() is not locked out. This
1907 * could give a filesystem state which requires a degraded mount.
1909 ret
= btrfs_rm_dev_item(fs_info
, device
);
1913 device
->in_fs_metadata
= 0;
1914 btrfs_scrub_cancel_dev(fs_info
, device
);
1917 * the device list mutex makes sure that we don't change
1918 * the device list while someone else is writing out all
1919 * the device supers. Whoever is writing all supers, should
1920 * lock the device list mutex before getting the number of
1921 * devices in the super block (super_copy). Conversely,
1922 * whoever updates the number of devices in the super block
1923 * (super_copy) should hold the device list mutex.
1926 cur_devices
= device
->fs_devices
;
1927 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
1928 list_del_rcu(&device
->dev_list
);
1930 device
->fs_devices
->num_devices
--;
1931 device
->fs_devices
->total_devices
--;
1933 if (device
->missing
)
1934 device
->fs_devices
->missing_devices
--;
1936 btrfs_assign_next_active_device(fs_info
, device
, NULL
);
1939 device
->fs_devices
->open_devices
--;
1940 /* remove sysfs entry */
1941 btrfs_sysfs_rm_device_link(fs_info
->fs_devices
, device
);
1944 num_devices
= btrfs_super_num_devices(fs_info
->super_copy
) - 1;
1945 btrfs_set_super_num_devices(fs_info
->super_copy
, num_devices
);
1946 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
1949 * at this point, the device is zero sized and detached from
1950 * the devices list. All that's left is to zero out the old
1951 * supers and free the device.
1953 if (device
->writeable
)
1954 btrfs_scratch_superblocks(device
->bdev
, device
->name
->str
);
1956 btrfs_close_bdev(device
);
1957 call_rcu(&device
->rcu
, free_device
);
1959 if (cur_devices
->open_devices
== 0) {
1960 struct btrfs_fs_devices
*fs_devices
;
1961 fs_devices
= fs_info
->fs_devices
;
1962 while (fs_devices
) {
1963 if (fs_devices
->seed
== cur_devices
) {
1964 fs_devices
->seed
= cur_devices
->seed
;
1967 fs_devices
= fs_devices
->seed
;
1969 cur_devices
->seed
= NULL
;
1970 __btrfs_close_devices(cur_devices
);
1971 free_fs_devices(cur_devices
);
1975 mutex_unlock(&uuid_mutex
);
1979 if (device
->writeable
) {
1980 mutex_lock(&fs_info
->chunk_mutex
);
1981 list_add(&device
->dev_alloc_list
,
1982 &fs_info
->fs_devices
->alloc_list
);
1983 device
->fs_devices
->rw_devices
++;
1984 mutex_unlock(&fs_info
->chunk_mutex
);
1989 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info
*fs_info
,
1990 struct btrfs_device
*srcdev
)
1992 struct btrfs_fs_devices
*fs_devices
;
1994 WARN_ON(!mutex_is_locked(&fs_info
->fs_devices
->device_list_mutex
));
1997 * in case of fs with no seed, srcdev->fs_devices will point
1998 * to fs_devices of fs_info. However when the dev being replaced is
1999 * a seed dev it will point to the seed's local fs_devices. In short
2000 * srcdev will have its correct fs_devices in both the cases.
2002 fs_devices
= srcdev
->fs_devices
;
2004 list_del_rcu(&srcdev
->dev_list
);
2005 list_del(&srcdev
->dev_alloc_list
);
2006 fs_devices
->num_devices
--;
2007 if (srcdev
->missing
)
2008 fs_devices
->missing_devices
--;
2010 if (srcdev
->writeable
)
2011 fs_devices
->rw_devices
--;
2014 fs_devices
->open_devices
--;
2017 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info
*fs_info
,
2018 struct btrfs_device
*srcdev
)
2020 struct btrfs_fs_devices
*fs_devices
= srcdev
->fs_devices
;
2022 if (srcdev
->writeable
) {
2023 /* zero out the old super if it is writable */
2024 btrfs_scratch_superblocks(srcdev
->bdev
, srcdev
->name
->str
);
2027 btrfs_close_bdev(srcdev
);
2028 call_rcu(&srcdev
->rcu
, free_device
);
2030 /* if this is no devs we rather delete the fs_devices */
2031 if (!fs_devices
->num_devices
) {
2032 struct btrfs_fs_devices
*tmp_fs_devices
;
2035 * On a mounted FS, num_devices can't be zero unless it's a
2036 * seed. In case of a seed device being replaced, the replace
2037 * target added to the sprout FS, so there will be no more
2038 * device left under the seed FS.
2040 ASSERT(fs_devices
->seeding
);
2042 tmp_fs_devices
= fs_info
->fs_devices
;
2043 while (tmp_fs_devices
) {
2044 if (tmp_fs_devices
->seed
== fs_devices
) {
2045 tmp_fs_devices
->seed
= fs_devices
->seed
;
2048 tmp_fs_devices
= tmp_fs_devices
->seed
;
2050 fs_devices
->seed
= NULL
;
2051 __btrfs_close_devices(fs_devices
);
2052 free_fs_devices(fs_devices
);
2056 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
2057 struct btrfs_device
*tgtdev
)
2059 mutex_lock(&uuid_mutex
);
2061 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2063 btrfs_sysfs_rm_device_link(fs_info
->fs_devices
, tgtdev
);
2066 fs_info
->fs_devices
->open_devices
--;
2068 fs_info
->fs_devices
->num_devices
--;
2070 btrfs_assign_next_active_device(fs_info
, tgtdev
, NULL
);
2072 list_del_rcu(&tgtdev
->dev_list
);
2074 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2075 mutex_unlock(&uuid_mutex
);
2078 * The update_dev_time() with in btrfs_scratch_superblocks()
2079 * may lead to a call to btrfs_show_devname() which will try
2080 * to hold device_list_mutex. And here this device
2081 * is already out of device list, so we don't have to hold
2082 * the device_list_mutex lock.
2084 btrfs_scratch_superblocks(tgtdev
->bdev
, tgtdev
->name
->str
);
2086 btrfs_close_bdev(tgtdev
);
2087 call_rcu(&tgtdev
->rcu
, free_device
);
2090 static int btrfs_find_device_by_path(struct btrfs_fs_info
*fs_info
,
2091 const char *device_path
,
2092 struct btrfs_device
**device
)
2095 struct btrfs_super_block
*disk_super
;
2098 struct block_device
*bdev
;
2099 struct buffer_head
*bh
;
2102 ret
= btrfs_get_bdev_and_sb(device_path
, FMODE_READ
,
2103 fs_info
->bdev_holder
, 0, &bdev
, &bh
);
2106 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
2107 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
2108 dev_uuid
= disk_super
->dev_item
.uuid
;
2109 *device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, disk_super
->fsid
);
2113 blkdev_put(bdev
, FMODE_READ
);
2117 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info
*fs_info
,
2118 const char *device_path
,
2119 struct btrfs_device
**device
)
2122 if (strcmp(device_path
, "missing") == 0) {
2123 struct list_head
*devices
;
2124 struct btrfs_device
*tmp
;
2126 devices
= &fs_info
->fs_devices
->devices
;
2128 * It is safe to read the devices since the volume_mutex
2129 * is held by the caller.
2131 list_for_each_entry(tmp
, devices
, dev_list
) {
2132 if (tmp
->in_fs_metadata
&& !tmp
->bdev
) {
2139 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND
;
2143 return btrfs_find_device_by_path(fs_info
, device_path
, device
);
2148 * Lookup a device given by device id, or the path if the id is 0.
2150 int btrfs_find_device_by_devspec(struct btrfs_fs_info
*fs_info
, u64 devid
,
2151 const char *devpath
,
2152 struct btrfs_device
**device
)
2158 *device
= btrfs_find_device(fs_info
, devid
, NULL
, NULL
);
2162 if (!devpath
|| !devpath
[0])
2165 ret
= btrfs_find_device_missing_or_by_path(fs_info
, devpath
,
2172 * does all the dirty work required for changing file system's UUID.
2174 static int btrfs_prepare_sprout(struct btrfs_fs_info
*fs_info
)
2176 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2177 struct btrfs_fs_devices
*old_devices
;
2178 struct btrfs_fs_devices
*seed_devices
;
2179 struct btrfs_super_block
*disk_super
= fs_info
->super_copy
;
2180 struct btrfs_device
*device
;
2183 BUG_ON(!mutex_is_locked(&uuid_mutex
));
2184 if (!fs_devices
->seeding
)
2187 seed_devices
= alloc_fs_devices(NULL
);
2188 if (IS_ERR(seed_devices
))
2189 return PTR_ERR(seed_devices
);
2191 old_devices
= clone_fs_devices(fs_devices
);
2192 if (IS_ERR(old_devices
)) {
2193 kfree(seed_devices
);
2194 return PTR_ERR(old_devices
);
2197 list_add(&old_devices
->list
, &fs_uuids
);
2199 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
2200 seed_devices
->opened
= 1;
2201 INIT_LIST_HEAD(&seed_devices
->devices
);
2202 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
2203 mutex_init(&seed_devices
->device_list_mutex
);
2205 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2206 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
2208 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
)
2209 device
->fs_devices
= seed_devices
;
2211 mutex_lock(&fs_info
->chunk_mutex
);
2212 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
2213 mutex_unlock(&fs_info
->chunk_mutex
);
2215 fs_devices
->seeding
= 0;
2216 fs_devices
->num_devices
= 0;
2217 fs_devices
->open_devices
= 0;
2218 fs_devices
->missing_devices
= 0;
2219 fs_devices
->rotating
= 0;
2220 fs_devices
->seed
= seed_devices
;
2222 generate_random_uuid(fs_devices
->fsid
);
2223 memcpy(fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2224 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2225 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2227 super_flags
= btrfs_super_flags(disk_super
) &
2228 ~BTRFS_SUPER_FLAG_SEEDING
;
2229 btrfs_set_super_flags(disk_super
, super_flags
);
2235 * Store the expected generation for seed devices in device items.
2237 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
2238 struct btrfs_fs_info
*fs_info
)
2240 struct btrfs_root
*root
= fs_info
->chunk_root
;
2241 struct btrfs_path
*path
;
2242 struct extent_buffer
*leaf
;
2243 struct btrfs_dev_item
*dev_item
;
2244 struct btrfs_device
*device
;
2245 struct btrfs_key key
;
2246 u8 fs_uuid
[BTRFS_FSID_SIZE
];
2247 u8 dev_uuid
[BTRFS_UUID_SIZE
];
2251 path
= btrfs_alloc_path();
2255 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2257 key
.type
= BTRFS_DEV_ITEM_KEY
;
2260 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2264 leaf
= path
->nodes
[0];
2266 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
2267 ret
= btrfs_next_leaf(root
, path
);
2272 leaf
= path
->nodes
[0];
2273 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2274 btrfs_release_path(path
);
2278 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2279 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
2280 key
.type
!= BTRFS_DEV_ITEM_KEY
)
2283 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2284 struct btrfs_dev_item
);
2285 devid
= btrfs_device_id(leaf
, dev_item
);
2286 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
2288 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
2290 device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, fs_uuid
);
2291 BUG_ON(!device
); /* Logic error */
2293 if (device
->fs_devices
->seeding
) {
2294 btrfs_set_device_generation(leaf
, dev_item
,
2295 device
->generation
);
2296 btrfs_mark_buffer_dirty(leaf
);
2304 btrfs_free_path(path
);
2308 int btrfs_init_new_device(struct btrfs_fs_info
*fs_info
, const char *device_path
)
2310 struct btrfs_root
*root
= fs_info
->dev_root
;
2311 struct request_queue
*q
;
2312 struct btrfs_trans_handle
*trans
;
2313 struct btrfs_device
*device
;
2314 struct block_device
*bdev
;
2315 struct list_head
*devices
;
2316 struct super_block
*sb
= fs_info
->sb
;
2317 struct rcu_string
*name
;
2319 int seeding_dev
= 0;
2321 bool unlocked
= false;
2323 if (sb_rdonly(sb
) && !fs_info
->fs_devices
->seeding
)
2326 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2327 fs_info
->bdev_holder
);
2329 return PTR_ERR(bdev
);
2331 if (fs_info
->fs_devices
->seeding
) {
2333 down_write(&sb
->s_umount
);
2334 mutex_lock(&uuid_mutex
);
2337 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2339 devices
= &fs_info
->fs_devices
->devices
;
2341 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2342 list_for_each_entry(device
, devices
, dev_list
) {
2343 if (device
->bdev
== bdev
) {
2346 &fs_info
->fs_devices
->device_list_mutex
);
2350 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2352 device
= btrfs_alloc_device(fs_info
, NULL
, NULL
);
2353 if (IS_ERR(device
)) {
2354 /* we can safely leave the fs_devices entry around */
2355 ret
= PTR_ERR(device
);
2359 name
= rcu_string_strdup(device_path
, GFP_KERNEL
);
2361 bio_put(device
->flush_bio
);
2366 rcu_assign_pointer(device
->name
, name
);
2368 trans
= btrfs_start_transaction(root
, 0);
2369 if (IS_ERR(trans
)) {
2370 rcu_string_free(device
->name
);
2371 bio_put(device
->flush_bio
);
2373 ret
= PTR_ERR(trans
);
2377 q
= bdev_get_queue(bdev
);
2378 if (blk_queue_discard(q
))
2379 device
->can_discard
= 1;
2380 device
->writeable
= 1;
2381 device
->generation
= trans
->transid
;
2382 device
->io_width
= fs_info
->sectorsize
;
2383 device
->io_align
= fs_info
->sectorsize
;
2384 device
->sector_size
= fs_info
->sectorsize
;
2385 device
->total_bytes
= round_down(i_size_read(bdev
->bd_inode
),
2386 fs_info
->sectorsize
);
2387 device
->disk_total_bytes
= device
->total_bytes
;
2388 device
->commit_total_bytes
= device
->total_bytes
;
2389 device
->fs_info
= fs_info
;
2390 device
->bdev
= bdev
;
2391 device
->in_fs_metadata
= 1;
2392 device
->is_tgtdev_for_dev_replace
= 0;
2393 device
->mode
= FMODE_EXCL
;
2394 device
->dev_stats_valid
= 1;
2395 set_blocksize(device
->bdev
, BTRFS_BDEV_BLOCKSIZE
);
2398 sb
->s_flags
&= ~SB_RDONLY
;
2399 ret
= btrfs_prepare_sprout(fs_info
);
2401 btrfs_abort_transaction(trans
, ret
);
2406 device
->fs_devices
= fs_info
->fs_devices
;
2408 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2409 mutex_lock(&fs_info
->chunk_mutex
);
2410 list_add_rcu(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2411 list_add(&device
->dev_alloc_list
,
2412 &fs_info
->fs_devices
->alloc_list
);
2413 fs_info
->fs_devices
->num_devices
++;
2414 fs_info
->fs_devices
->open_devices
++;
2415 fs_info
->fs_devices
->rw_devices
++;
2416 fs_info
->fs_devices
->total_devices
++;
2417 fs_info
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
2419 atomic64_add(device
->total_bytes
, &fs_info
->free_chunk_space
);
2421 if (!blk_queue_nonrot(q
))
2422 fs_info
->fs_devices
->rotating
= 1;
2424 tmp
= btrfs_super_total_bytes(fs_info
->super_copy
);
2425 btrfs_set_super_total_bytes(fs_info
->super_copy
,
2426 round_down(tmp
+ device
->total_bytes
, fs_info
->sectorsize
));
2428 tmp
= btrfs_super_num_devices(fs_info
->super_copy
);
2429 btrfs_set_super_num_devices(fs_info
->super_copy
, tmp
+ 1);
2431 /* add sysfs device entry */
2432 btrfs_sysfs_add_device_link(fs_info
->fs_devices
, device
);
2435 * we've got more storage, clear any full flags on the space
2438 btrfs_clear_space_info_full(fs_info
);
2440 mutex_unlock(&fs_info
->chunk_mutex
);
2441 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2444 mutex_lock(&fs_info
->chunk_mutex
);
2445 ret
= init_first_rw_device(trans
, fs_info
);
2446 mutex_unlock(&fs_info
->chunk_mutex
);
2448 btrfs_abort_transaction(trans
, ret
);
2453 ret
= btrfs_add_device(trans
, fs_info
, device
);
2455 btrfs_abort_transaction(trans
, ret
);
2460 char fsid_buf
[BTRFS_UUID_UNPARSED_SIZE
];
2462 ret
= btrfs_finish_sprout(trans
, fs_info
);
2464 btrfs_abort_transaction(trans
, ret
);
2468 /* Sprouting would change fsid of the mounted root,
2469 * so rename the fsid on the sysfs
2471 snprintf(fsid_buf
, BTRFS_UUID_UNPARSED_SIZE
, "%pU",
2473 if (kobject_rename(&fs_info
->fs_devices
->fsid_kobj
, fsid_buf
))
2475 "sysfs: failed to create fsid for sprout");
2478 ret
= btrfs_commit_transaction(trans
);
2481 mutex_unlock(&uuid_mutex
);
2482 up_write(&sb
->s_umount
);
2485 if (ret
) /* transaction commit */
2488 ret
= btrfs_relocate_sys_chunks(fs_info
);
2490 btrfs_handle_fs_error(fs_info
, ret
,
2491 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2492 trans
= btrfs_attach_transaction(root
);
2493 if (IS_ERR(trans
)) {
2494 if (PTR_ERR(trans
) == -ENOENT
)
2496 ret
= PTR_ERR(trans
);
2500 ret
= btrfs_commit_transaction(trans
);
2503 /* Update ctime/mtime for libblkid */
2504 update_dev_time(device_path
);
2508 btrfs_sysfs_rm_device_link(fs_info
->fs_devices
, device
);
2511 sb
->s_flags
|= SB_RDONLY
;
2513 btrfs_end_transaction(trans
);
2514 rcu_string_free(device
->name
);
2515 bio_put(device
->flush_bio
);
2518 blkdev_put(bdev
, FMODE_EXCL
);
2519 if (seeding_dev
&& !unlocked
) {
2520 mutex_unlock(&uuid_mutex
);
2521 up_write(&sb
->s_umount
);
2526 int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info
*fs_info
,
2527 const char *device_path
,
2528 struct btrfs_device
*srcdev
,
2529 struct btrfs_device
**device_out
)
2531 struct request_queue
*q
;
2532 struct btrfs_device
*device
;
2533 struct block_device
*bdev
;
2534 struct list_head
*devices
;
2535 struct rcu_string
*name
;
2536 u64 devid
= BTRFS_DEV_REPLACE_DEVID
;
2540 if (fs_info
->fs_devices
->seeding
) {
2541 btrfs_err(fs_info
, "the filesystem is a seed filesystem!");
2545 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2546 fs_info
->bdev_holder
);
2548 btrfs_err(fs_info
, "target device %s is invalid!", device_path
);
2549 return PTR_ERR(bdev
);
2552 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2554 devices
= &fs_info
->fs_devices
->devices
;
2555 list_for_each_entry(device
, devices
, dev_list
) {
2556 if (device
->bdev
== bdev
) {
2558 "target device is in the filesystem!");
2565 if (i_size_read(bdev
->bd_inode
) <
2566 btrfs_device_get_total_bytes(srcdev
)) {
2568 "target device is smaller than source device!");
2574 device
= btrfs_alloc_device(NULL
, &devid
, NULL
);
2575 if (IS_ERR(device
)) {
2576 ret
= PTR_ERR(device
);
2580 name
= rcu_string_strdup(device_path
, GFP_KERNEL
);
2582 bio_put(device
->flush_bio
);
2587 rcu_assign_pointer(device
->name
, name
);
2589 q
= bdev_get_queue(bdev
);
2590 if (blk_queue_discard(q
))
2591 device
->can_discard
= 1;
2592 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2593 device
->writeable
= 1;
2594 device
->generation
= 0;
2595 device
->io_width
= fs_info
->sectorsize
;
2596 device
->io_align
= fs_info
->sectorsize
;
2597 device
->sector_size
= fs_info
->sectorsize
;
2598 device
->total_bytes
= btrfs_device_get_total_bytes(srcdev
);
2599 device
->disk_total_bytes
= btrfs_device_get_disk_total_bytes(srcdev
);
2600 device
->bytes_used
= btrfs_device_get_bytes_used(srcdev
);
2601 ASSERT(list_empty(&srcdev
->resized_list
));
2602 device
->commit_total_bytes
= srcdev
->commit_total_bytes
;
2603 device
->commit_bytes_used
= device
->bytes_used
;
2604 device
->fs_info
= fs_info
;
2605 device
->bdev
= bdev
;
2606 device
->in_fs_metadata
= 1;
2607 device
->is_tgtdev_for_dev_replace
= 1;
2608 device
->mode
= FMODE_EXCL
;
2609 device
->dev_stats_valid
= 1;
2610 set_blocksize(device
->bdev
, BTRFS_BDEV_BLOCKSIZE
);
2611 device
->fs_devices
= fs_info
->fs_devices
;
2612 list_add(&device
->dev_list
, &fs_info
->fs_devices
->devices
);
2613 fs_info
->fs_devices
->num_devices
++;
2614 fs_info
->fs_devices
->open_devices
++;
2615 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2617 *device_out
= device
;
2621 blkdev_put(bdev
, FMODE_EXCL
);
2625 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info
*fs_info
,
2626 struct btrfs_device
*tgtdev
)
2628 u32 sectorsize
= fs_info
->sectorsize
;
2630 WARN_ON(fs_info
->fs_devices
->rw_devices
== 0);
2631 tgtdev
->io_width
= sectorsize
;
2632 tgtdev
->io_align
= sectorsize
;
2633 tgtdev
->sector_size
= sectorsize
;
2634 tgtdev
->fs_info
= fs_info
;
2635 tgtdev
->in_fs_metadata
= 1;
2638 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2639 struct btrfs_device
*device
)
2642 struct btrfs_path
*path
;
2643 struct btrfs_root
*root
= device
->fs_info
->chunk_root
;
2644 struct btrfs_dev_item
*dev_item
;
2645 struct extent_buffer
*leaf
;
2646 struct btrfs_key key
;
2648 path
= btrfs_alloc_path();
2652 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2653 key
.type
= BTRFS_DEV_ITEM_KEY
;
2654 key
.offset
= device
->devid
;
2656 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2665 leaf
= path
->nodes
[0];
2666 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2668 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2669 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2670 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2671 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2672 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2673 btrfs_set_device_total_bytes(leaf
, dev_item
,
2674 btrfs_device_get_disk_total_bytes(device
));
2675 btrfs_set_device_bytes_used(leaf
, dev_item
,
2676 btrfs_device_get_bytes_used(device
));
2677 btrfs_mark_buffer_dirty(leaf
);
2680 btrfs_free_path(path
);
2684 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2685 struct btrfs_device
*device
, u64 new_size
)
2687 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
2688 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
2689 struct btrfs_fs_devices
*fs_devices
;
2693 if (!device
->writeable
)
2696 new_size
= round_down(new_size
, fs_info
->sectorsize
);
2698 mutex_lock(&fs_info
->chunk_mutex
);
2699 old_total
= btrfs_super_total_bytes(super_copy
);
2700 diff
= round_down(new_size
- device
->total_bytes
, fs_info
->sectorsize
);
2702 if (new_size
<= device
->total_bytes
||
2703 device
->is_tgtdev_for_dev_replace
) {
2704 mutex_unlock(&fs_info
->chunk_mutex
);
2708 fs_devices
= fs_info
->fs_devices
;
2710 btrfs_set_super_total_bytes(super_copy
,
2711 round_down(old_total
+ diff
, fs_info
->sectorsize
));
2712 device
->fs_devices
->total_rw_bytes
+= diff
;
2714 btrfs_device_set_total_bytes(device
, new_size
);
2715 btrfs_device_set_disk_total_bytes(device
, new_size
);
2716 btrfs_clear_space_info_full(device
->fs_info
);
2717 if (list_empty(&device
->resized_list
))
2718 list_add_tail(&device
->resized_list
,
2719 &fs_devices
->resized_devices
);
2720 mutex_unlock(&fs_info
->chunk_mutex
);
2722 return btrfs_update_device(trans
, device
);
2725 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
,
2726 struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2728 struct btrfs_root
*root
= fs_info
->chunk_root
;
2730 struct btrfs_path
*path
;
2731 struct btrfs_key key
;
2733 path
= btrfs_alloc_path();
2737 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2738 key
.offset
= chunk_offset
;
2739 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2741 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2744 else if (ret
> 0) { /* Logic error or corruption */
2745 btrfs_handle_fs_error(fs_info
, -ENOENT
,
2746 "Failed lookup while freeing chunk.");
2751 ret
= btrfs_del_item(trans
, root
, path
);
2753 btrfs_handle_fs_error(fs_info
, ret
,
2754 "Failed to delete chunk item.");
2756 btrfs_free_path(path
);
2760 static int btrfs_del_sys_chunk(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2762 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
2763 struct btrfs_disk_key
*disk_key
;
2764 struct btrfs_chunk
*chunk
;
2771 struct btrfs_key key
;
2773 mutex_lock(&fs_info
->chunk_mutex
);
2774 array_size
= btrfs_super_sys_array_size(super_copy
);
2776 ptr
= super_copy
->sys_chunk_array
;
2779 while (cur
< array_size
) {
2780 disk_key
= (struct btrfs_disk_key
*)ptr
;
2781 btrfs_disk_key_to_cpu(&key
, disk_key
);
2783 len
= sizeof(*disk_key
);
2785 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2786 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2787 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2788 len
+= btrfs_chunk_item_size(num_stripes
);
2793 if (key
.objectid
== BTRFS_FIRST_CHUNK_TREE_OBJECTID
&&
2794 key
.offset
== chunk_offset
) {
2795 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2797 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2803 mutex_unlock(&fs_info
->chunk_mutex
);
2807 static struct extent_map
*get_chunk_map(struct btrfs_fs_info
*fs_info
,
2808 u64 logical
, u64 length
)
2810 struct extent_map_tree
*em_tree
;
2811 struct extent_map
*em
;
2813 em_tree
= &fs_info
->mapping_tree
.map_tree
;
2814 read_lock(&em_tree
->lock
);
2815 em
= lookup_extent_mapping(em_tree
, logical
, length
);
2816 read_unlock(&em_tree
->lock
);
2819 btrfs_crit(fs_info
, "unable to find logical %llu length %llu",
2821 return ERR_PTR(-EINVAL
);
2824 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
2826 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2827 logical
, length
, em
->start
, em
->start
+ em
->len
);
2828 free_extent_map(em
);
2829 return ERR_PTR(-EINVAL
);
2832 /* callers are responsible for dropping em's ref. */
2836 int btrfs_remove_chunk(struct btrfs_trans_handle
*trans
,
2837 struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2839 struct extent_map
*em
;
2840 struct map_lookup
*map
;
2841 u64 dev_extent_len
= 0;
2843 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2845 em
= get_chunk_map(fs_info
, chunk_offset
, 1);
2848 * This is a logic error, but we don't want to just rely on the
2849 * user having built with ASSERT enabled, so if ASSERT doesn't
2850 * do anything we still error out.
2855 map
= em
->map_lookup
;
2856 mutex_lock(&fs_info
->chunk_mutex
);
2857 check_system_chunk(trans
, fs_info
, map
->type
);
2858 mutex_unlock(&fs_info
->chunk_mutex
);
2861 * Take the device list mutex to prevent races with the final phase of
2862 * a device replace operation that replaces the device object associated
2863 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2865 mutex_lock(&fs_devices
->device_list_mutex
);
2866 for (i
= 0; i
< map
->num_stripes
; i
++) {
2867 struct btrfs_device
*device
= map
->stripes
[i
].dev
;
2868 ret
= btrfs_free_dev_extent(trans
, device
,
2869 map
->stripes
[i
].physical
,
2872 mutex_unlock(&fs_devices
->device_list_mutex
);
2873 btrfs_abort_transaction(trans
, ret
);
2877 if (device
->bytes_used
> 0) {
2878 mutex_lock(&fs_info
->chunk_mutex
);
2879 btrfs_device_set_bytes_used(device
,
2880 device
->bytes_used
- dev_extent_len
);
2881 atomic64_add(dev_extent_len
, &fs_info
->free_chunk_space
);
2882 btrfs_clear_space_info_full(fs_info
);
2883 mutex_unlock(&fs_info
->chunk_mutex
);
2886 if (map
->stripes
[i
].dev
) {
2887 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2889 mutex_unlock(&fs_devices
->device_list_mutex
);
2890 btrfs_abort_transaction(trans
, ret
);
2895 mutex_unlock(&fs_devices
->device_list_mutex
);
2897 ret
= btrfs_free_chunk(trans
, fs_info
, chunk_offset
);
2899 btrfs_abort_transaction(trans
, ret
);
2903 trace_btrfs_chunk_free(fs_info
, map
, chunk_offset
, em
->len
);
2905 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2906 ret
= btrfs_del_sys_chunk(fs_info
, chunk_offset
);
2908 btrfs_abort_transaction(trans
, ret
);
2913 ret
= btrfs_remove_block_group(trans
, fs_info
, chunk_offset
, em
);
2915 btrfs_abort_transaction(trans
, ret
);
2921 free_extent_map(em
);
2925 static int btrfs_relocate_chunk(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2927 struct btrfs_root
*root
= fs_info
->chunk_root
;
2928 struct btrfs_trans_handle
*trans
;
2932 * Prevent races with automatic removal of unused block groups.
2933 * After we relocate and before we remove the chunk with offset
2934 * chunk_offset, automatic removal of the block group can kick in,
2935 * resulting in a failure when calling btrfs_remove_chunk() below.
2937 * Make sure to acquire this mutex before doing a tree search (dev
2938 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2939 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2940 * we release the path used to search the chunk/dev tree and before
2941 * the current task acquires this mutex and calls us.
2943 ASSERT(mutex_is_locked(&fs_info
->delete_unused_bgs_mutex
));
2945 ret
= btrfs_can_relocate(fs_info
, chunk_offset
);
2949 /* step one, relocate all the extents inside this chunk */
2950 btrfs_scrub_pause(fs_info
);
2951 ret
= btrfs_relocate_block_group(fs_info
, chunk_offset
);
2952 btrfs_scrub_continue(fs_info
);
2956 trans
= btrfs_start_trans_remove_block_group(root
->fs_info
,
2958 if (IS_ERR(trans
)) {
2959 ret
= PTR_ERR(trans
);
2960 btrfs_handle_fs_error(root
->fs_info
, ret
, NULL
);
2965 * step two, delete the device extents and the
2966 * chunk tree entries
2968 ret
= btrfs_remove_chunk(trans
, fs_info
, chunk_offset
);
2969 btrfs_end_transaction(trans
);
2973 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info
*fs_info
)
2975 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
2976 struct btrfs_path
*path
;
2977 struct extent_buffer
*leaf
;
2978 struct btrfs_chunk
*chunk
;
2979 struct btrfs_key key
;
2980 struct btrfs_key found_key
;
2982 bool retried
= false;
2986 path
= btrfs_alloc_path();
2991 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2992 key
.offset
= (u64
)-1;
2993 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2996 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
2997 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2999 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3002 BUG_ON(ret
== 0); /* Corruption */
3004 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
3007 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3013 leaf
= path
->nodes
[0];
3014 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3016 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
3017 struct btrfs_chunk
);
3018 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3019 btrfs_release_path(path
);
3021 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3022 ret
= btrfs_relocate_chunk(fs_info
, found_key
.offset
);
3028 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3030 if (found_key
.offset
== 0)
3032 key
.offset
= found_key
.offset
- 1;
3035 if (failed
&& !retried
) {
3039 } else if (WARN_ON(failed
&& retried
)) {
3043 btrfs_free_path(path
);
3047 static int insert_balance_item(struct btrfs_fs_info
*fs_info
,
3048 struct btrfs_balance_control
*bctl
)
3050 struct btrfs_root
*root
= fs_info
->tree_root
;
3051 struct btrfs_trans_handle
*trans
;
3052 struct btrfs_balance_item
*item
;
3053 struct btrfs_disk_balance_args disk_bargs
;
3054 struct btrfs_path
*path
;
3055 struct extent_buffer
*leaf
;
3056 struct btrfs_key key
;
3059 path
= btrfs_alloc_path();
3063 trans
= btrfs_start_transaction(root
, 0);
3064 if (IS_ERR(trans
)) {
3065 btrfs_free_path(path
);
3066 return PTR_ERR(trans
);
3069 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3070 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3073 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
3078 leaf
= path
->nodes
[0];
3079 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3081 memzero_extent_buffer(leaf
, (unsigned long)item
, sizeof(*item
));
3083 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
3084 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
3085 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
3086 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
3087 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
3088 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
3090 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
3092 btrfs_mark_buffer_dirty(leaf
);
3094 btrfs_free_path(path
);
3095 err
= btrfs_commit_transaction(trans
);
3101 static int del_balance_item(struct btrfs_fs_info
*fs_info
)
3103 struct btrfs_root
*root
= fs_info
->tree_root
;
3104 struct btrfs_trans_handle
*trans
;
3105 struct btrfs_path
*path
;
3106 struct btrfs_key key
;
3109 path
= btrfs_alloc_path();
3113 trans
= btrfs_start_transaction(root
, 0);
3114 if (IS_ERR(trans
)) {
3115 btrfs_free_path(path
);
3116 return PTR_ERR(trans
);
3119 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3120 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3123 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3131 ret
= btrfs_del_item(trans
, root
, path
);
3133 btrfs_free_path(path
);
3134 err
= btrfs_commit_transaction(trans
);
3141 * This is a heuristic used to reduce the number of chunks balanced on
3142 * resume after balance was interrupted.
3144 static void update_balance_args(struct btrfs_balance_control
*bctl
)
3147 * Turn on soft mode for chunk types that were being converted.
3149 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3150 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3151 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3152 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3153 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3154 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3157 * Turn on usage filter if is not already used. The idea is
3158 * that chunks that we have already balanced should be
3159 * reasonably full. Don't do it for chunks that are being
3160 * converted - that will keep us from relocating unconverted
3161 * (albeit full) chunks.
3163 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3164 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3165 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3166 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3167 bctl
->data
.usage
= 90;
3169 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3170 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3171 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3172 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3173 bctl
->sys
.usage
= 90;
3175 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3176 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3177 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3178 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3179 bctl
->meta
.usage
= 90;
3184 * Should be called with both balance and volume mutexes held to
3185 * serialize other volume operations (add_dev/rm_dev/resize) with
3186 * restriper. Same goes for unset_balance_control.
3188 static void set_balance_control(struct btrfs_balance_control
*bctl
)
3190 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3192 BUG_ON(fs_info
->balance_ctl
);
3194 spin_lock(&fs_info
->balance_lock
);
3195 fs_info
->balance_ctl
= bctl
;
3196 spin_unlock(&fs_info
->balance_lock
);
3199 static void unset_balance_control(struct btrfs_fs_info
*fs_info
)
3201 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3203 BUG_ON(!fs_info
->balance_ctl
);
3205 spin_lock(&fs_info
->balance_lock
);
3206 fs_info
->balance_ctl
= NULL
;
3207 spin_unlock(&fs_info
->balance_lock
);
3213 * Balance filters. Return 1 if chunk should be filtered out
3214 * (should not be balanced).
3216 static int chunk_profiles_filter(u64 chunk_type
,
3217 struct btrfs_balance_args
*bargs
)
3219 chunk_type
= chunk_to_extended(chunk_type
) &
3220 BTRFS_EXTENDED_PROFILE_MASK
;
3222 if (bargs
->profiles
& chunk_type
)
3228 static int chunk_usage_range_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
3229 struct btrfs_balance_args
*bargs
)
3231 struct btrfs_block_group_cache
*cache
;
3233 u64 user_thresh_min
;
3234 u64 user_thresh_max
;
3237 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3238 chunk_used
= btrfs_block_group_used(&cache
->item
);
3240 if (bargs
->usage_min
== 0)
3241 user_thresh_min
= 0;
3243 user_thresh_min
= div_factor_fine(cache
->key
.offset
,
3246 if (bargs
->usage_max
== 0)
3247 user_thresh_max
= 1;
3248 else if (bargs
->usage_max
> 100)
3249 user_thresh_max
= cache
->key
.offset
;
3251 user_thresh_max
= div_factor_fine(cache
->key
.offset
,
3254 if (user_thresh_min
<= chunk_used
&& chunk_used
< user_thresh_max
)
3257 btrfs_put_block_group(cache
);
3261 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
,
3262 u64 chunk_offset
, struct btrfs_balance_args
*bargs
)
3264 struct btrfs_block_group_cache
*cache
;
3265 u64 chunk_used
, user_thresh
;
3268 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3269 chunk_used
= btrfs_block_group_used(&cache
->item
);
3271 if (bargs
->usage_min
== 0)
3273 else if (bargs
->usage
> 100)
3274 user_thresh
= cache
->key
.offset
;
3276 user_thresh
= div_factor_fine(cache
->key
.offset
,
3279 if (chunk_used
< user_thresh
)
3282 btrfs_put_block_group(cache
);
3286 static int chunk_devid_filter(struct extent_buffer
*leaf
,
3287 struct btrfs_chunk
*chunk
,
3288 struct btrfs_balance_args
*bargs
)
3290 struct btrfs_stripe
*stripe
;
3291 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3294 for (i
= 0; i
< num_stripes
; i
++) {
3295 stripe
= btrfs_stripe_nr(chunk
, i
);
3296 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
3303 /* [pstart, pend) */
3304 static int chunk_drange_filter(struct extent_buffer
*leaf
,
3305 struct btrfs_chunk
*chunk
,
3306 struct btrfs_balance_args
*bargs
)
3308 struct btrfs_stripe
*stripe
;
3309 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3315 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
3318 if (btrfs_chunk_type(leaf
, chunk
) & (BTRFS_BLOCK_GROUP_DUP
|
3319 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)) {
3320 factor
= num_stripes
/ 2;
3321 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID5
) {
3322 factor
= num_stripes
- 1;
3323 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID6
) {
3324 factor
= num_stripes
- 2;
3326 factor
= num_stripes
;
3329 for (i
= 0; i
< num_stripes
; i
++) {
3330 stripe
= btrfs_stripe_nr(chunk
, i
);
3331 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
3334 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
3335 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
3336 stripe_length
= div_u64(stripe_length
, factor
);
3338 if (stripe_offset
< bargs
->pend
&&
3339 stripe_offset
+ stripe_length
> bargs
->pstart
)
3346 /* [vstart, vend) */
3347 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
3348 struct btrfs_chunk
*chunk
,
3350 struct btrfs_balance_args
*bargs
)
3352 if (chunk_offset
< bargs
->vend
&&
3353 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
3354 /* at least part of the chunk is inside this vrange */
3360 static int chunk_stripes_range_filter(struct extent_buffer
*leaf
,
3361 struct btrfs_chunk
*chunk
,
3362 struct btrfs_balance_args
*bargs
)
3364 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3366 if (bargs
->stripes_min
<= num_stripes
3367 && num_stripes
<= bargs
->stripes_max
)
3373 static int chunk_soft_convert_filter(u64 chunk_type
,
3374 struct btrfs_balance_args
*bargs
)
3376 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
3379 chunk_type
= chunk_to_extended(chunk_type
) &
3380 BTRFS_EXTENDED_PROFILE_MASK
;
3382 if (bargs
->target
== chunk_type
)
3388 static int should_balance_chunk(struct btrfs_fs_info
*fs_info
,
3389 struct extent_buffer
*leaf
,
3390 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
3392 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3393 struct btrfs_balance_args
*bargs
= NULL
;
3394 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3397 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
3398 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
3402 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3403 bargs
= &bctl
->data
;
3404 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3406 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3407 bargs
= &bctl
->meta
;
3409 /* profiles filter */
3410 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
3411 chunk_profiles_filter(chunk_type
, bargs
)) {
3416 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3417 chunk_usage_filter(fs_info
, chunk_offset
, bargs
)) {
3419 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3420 chunk_usage_range_filter(fs_info
, chunk_offset
, bargs
)) {
3425 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
3426 chunk_devid_filter(leaf
, chunk
, bargs
)) {
3430 /* drange filter, makes sense only with devid filter */
3431 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
3432 chunk_drange_filter(leaf
, chunk
, bargs
)) {
3437 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
3438 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3442 /* stripes filter */
3443 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_STRIPES_RANGE
) &&
3444 chunk_stripes_range_filter(leaf
, chunk
, bargs
)) {
3448 /* soft profile changing mode */
3449 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
3450 chunk_soft_convert_filter(chunk_type
, bargs
)) {
3455 * limited by count, must be the last filter
3457 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT
)) {
3458 if (bargs
->limit
== 0)
3462 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT_RANGE
)) {
3464 * Same logic as the 'limit' filter; the minimum cannot be
3465 * determined here because we do not have the global information
3466 * about the count of all chunks that satisfy the filters.
3468 if (bargs
->limit_max
== 0)
3477 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
3479 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3480 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
3481 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
3482 struct list_head
*devices
;
3483 struct btrfs_device
*device
;
3487 struct btrfs_chunk
*chunk
;
3488 struct btrfs_path
*path
= NULL
;
3489 struct btrfs_key key
;
3490 struct btrfs_key found_key
;
3491 struct btrfs_trans_handle
*trans
;
3492 struct extent_buffer
*leaf
;
3495 int enospc_errors
= 0;
3496 bool counting
= true;
3497 /* The single value limit and min/max limits use the same bytes in the */
3498 u64 limit_data
= bctl
->data
.limit
;
3499 u64 limit_meta
= bctl
->meta
.limit
;
3500 u64 limit_sys
= bctl
->sys
.limit
;
3504 int chunk_reserved
= 0;
3507 /* step one make some room on all the devices */
3508 devices
= &fs_info
->fs_devices
->devices
;
3509 list_for_each_entry(device
, devices
, dev_list
) {
3510 old_size
= btrfs_device_get_total_bytes(device
);
3511 size_to_free
= div_factor(old_size
, 1);
3512 size_to_free
= min_t(u64
, size_to_free
, SZ_1M
);
3513 if (!device
->writeable
||
3514 btrfs_device_get_total_bytes(device
) -
3515 btrfs_device_get_bytes_used(device
) > size_to_free
||
3516 device
->is_tgtdev_for_dev_replace
)
3519 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
3523 /* btrfs_shrink_device never returns ret > 0 */
3528 trans
= btrfs_start_transaction(dev_root
, 0);
3529 if (IS_ERR(trans
)) {
3530 ret
= PTR_ERR(trans
);
3531 btrfs_info_in_rcu(fs_info
,
3532 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3533 rcu_str_deref(device
->name
), ret
,
3534 old_size
, old_size
- size_to_free
);
3538 ret
= btrfs_grow_device(trans
, device
, old_size
);
3540 btrfs_end_transaction(trans
);
3541 /* btrfs_grow_device never returns ret > 0 */
3543 btrfs_info_in_rcu(fs_info
,
3544 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3545 rcu_str_deref(device
->name
), ret
,
3546 old_size
, old_size
- size_to_free
);
3550 btrfs_end_transaction(trans
);
3553 /* step two, relocate all the chunks */
3554 path
= btrfs_alloc_path();
3560 /* zero out stat counters */
3561 spin_lock(&fs_info
->balance_lock
);
3562 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
3563 spin_unlock(&fs_info
->balance_lock
);
3567 * The single value limit and min/max limits use the same bytes
3570 bctl
->data
.limit
= limit_data
;
3571 bctl
->meta
.limit
= limit_meta
;
3572 bctl
->sys
.limit
= limit_sys
;
3574 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3575 key
.offset
= (u64
)-1;
3576 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3579 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
3580 atomic_read(&fs_info
->balance_cancel_req
)) {
3585 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
3586 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
3588 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3593 * this shouldn't happen, it means the last relocate
3597 BUG(); /* FIXME break ? */
3599 ret
= btrfs_previous_item(chunk_root
, path
, 0,
3600 BTRFS_CHUNK_ITEM_KEY
);
3602 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3607 leaf
= path
->nodes
[0];
3608 slot
= path
->slots
[0];
3609 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3611 if (found_key
.objectid
!= key
.objectid
) {
3612 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3616 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
3617 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3620 spin_lock(&fs_info
->balance_lock
);
3621 bctl
->stat
.considered
++;
3622 spin_unlock(&fs_info
->balance_lock
);
3625 ret
= should_balance_chunk(fs_info
, leaf
, chunk
,
3628 btrfs_release_path(path
);
3630 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3635 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3636 spin_lock(&fs_info
->balance_lock
);
3637 bctl
->stat
.expected
++;
3638 spin_unlock(&fs_info
->balance_lock
);
3640 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3642 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3644 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3651 * Apply limit_min filter, no need to check if the LIMITS
3652 * filter is used, limit_min is 0 by default
3654 if (((chunk_type
& BTRFS_BLOCK_GROUP_DATA
) &&
3655 count_data
< bctl
->data
.limit_min
)
3656 || ((chunk_type
& BTRFS_BLOCK_GROUP_METADATA
) &&
3657 count_meta
< bctl
->meta
.limit_min
)
3658 || ((chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) &&
3659 count_sys
< bctl
->sys
.limit_min
)) {
3660 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3664 ASSERT(fs_info
->data_sinfo
);
3665 spin_lock(&fs_info
->data_sinfo
->lock
);
3666 bytes_used
= fs_info
->data_sinfo
->bytes_used
;
3667 spin_unlock(&fs_info
->data_sinfo
->lock
);
3669 if ((chunk_type
& BTRFS_BLOCK_GROUP_DATA
) &&
3670 !chunk_reserved
&& !bytes_used
) {
3671 trans
= btrfs_start_transaction(chunk_root
, 0);
3672 if (IS_ERR(trans
)) {
3673 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3674 ret
= PTR_ERR(trans
);
3678 ret
= btrfs_force_chunk_alloc(trans
, fs_info
,
3679 BTRFS_BLOCK_GROUP_DATA
);
3680 btrfs_end_transaction(trans
);
3682 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3688 ret
= btrfs_relocate_chunk(fs_info
, found_key
.offset
);
3689 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3690 if (ret
&& ret
!= -ENOSPC
)
3692 if (ret
== -ENOSPC
) {
3695 spin_lock(&fs_info
->balance_lock
);
3696 bctl
->stat
.completed
++;
3697 spin_unlock(&fs_info
->balance_lock
);
3700 if (found_key
.offset
== 0)
3702 key
.offset
= found_key
.offset
- 1;
3706 btrfs_release_path(path
);
3711 btrfs_free_path(path
);
3712 if (enospc_errors
) {
3713 btrfs_info(fs_info
, "%d enospc errors during balance",
3723 * alloc_profile_is_valid - see if a given profile is valid and reduced
3724 * @flags: profile to validate
3725 * @extended: if true @flags is treated as an extended profile
3727 static int alloc_profile_is_valid(u64 flags
, int extended
)
3729 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
3730 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
3732 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
3734 /* 1) check that all other bits are zeroed */
3738 /* 2) see if profile is reduced */
3740 return !extended
; /* "0" is valid for usual profiles */
3742 /* true if exactly one bit set */
3743 return (flags
& (flags
- 1)) == 0;
3746 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
3748 /* cancel requested || normal exit path */
3749 return atomic_read(&fs_info
->balance_cancel_req
) ||
3750 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
3751 atomic_read(&fs_info
->balance_cancel_req
) == 0);
3754 static void __cancel_balance(struct btrfs_fs_info
*fs_info
)
3758 unset_balance_control(fs_info
);
3759 ret
= del_balance_item(fs_info
);
3761 btrfs_handle_fs_error(fs_info
, ret
, NULL
);
3763 clear_bit(BTRFS_FS_EXCL_OP
, &fs_info
->flags
);
3766 /* Non-zero return value signifies invalidity */
3767 static inline int validate_convert_profile(struct btrfs_balance_args
*bctl_arg
,
3770 return ((bctl_arg
->flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3771 (!alloc_profile_is_valid(bctl_arg
->target
, 1) ||
3772 (bctl_arg
->target
& ~allowed
)));
3776 * Should be called with both balance and volume mutexes held
3778 int btrfs_balance(struct btrfs_balance_control
*bctl
,
3779 struct btrfs_ioctl_balance_args
*bargs
)
3781 struct btrfs_fs_info
*fs_info
= bctl
->fs_info
;
3782 u64 meta_target
, data_target
;
3789 if (btrfs_fs_closing(fs_info
) ||
3790 atomic_read(&fs_info
->balance_pause_req
) ||
3791 atomic_read(&fs_info
->balance_cancel_req
)) {
3796 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
3797 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
3801 * In case of mixed groups both data and meta should be picked,
3802 * and identical options should be given for both of them.
3804 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
3805 if (mixed
&& (bctl
->flags
& allowed
)) {
3806 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
3807 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
3808 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
3810 "with mixed groups data and metadata balance options must be the same");
3816 num_devices
= fs_info
->fs_devices
->num_devices
;
3817 btrfs_dev_replace_lock(&fs_info
->dev_replace
, 0);
3818 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
3819 BUG_ON(num_devices
< 1);
3822 btrfs_dev_replace_unlock(&fs_info
->dev_replace
, 0);
3823 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
| BTRFS_BLOCK_GROUP_DUP
;
3824 if (num_devices
> 1)
3825 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
3826 if (num_devices
> 2)
3827 allowed
|= BTRFS_BLOCK_GROUP_RAID5
;
3828 if (num_devices
> 3)
3829 allowed
|= (BTRFS_BLOCK_GROUP_RAID10
|
3830 BTRFS_BLOCK_GROUP_RAID6
);
3831 if (validate_convert_profile(&bctl
->data
, allowed
)) {
3833 "unable to start balance with target data profile %llu",
3838 if (validate_convert_profile(&bctl
->meta
, allowed
)) {
3840 "unable to start balance with target metadata profile %llu",
3845 if (validate_convert_profile(&bctl
->sys
, allowed
)) {
3847 "unable to start balance with target system profile %llu",
3853 /* allow to reduce meta or sys integrity only if force set */
3854 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3855 BTRFS_BLOCK_GROUP_RAID10
|
3856 BTRFS_BLOCK_GROUP_RAID5
|
3857 BTRFS_BLOCK_GROUP_RAID6
;
3859 seq
= read_seqbegin(&fs_info
->profiles_lock
);
3861 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3862 (fs_info
->avail_system_alloc_bits
& allowed
) &&
3863 !(bctl
->sys
.target
& allowed
)) ||
3864 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3865 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
3866 !(bctl
->meta
.target
& allowed
))) {
3867 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
3869 "force reducing metadata integrity");
3872 "balance will reduce metadata integrity, use force if you want this");
3877 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
3879 /* if we're not converting, the target field is uninitialized */
3880 meta_target
= (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) ?
3881 bctl
->meta
.target
: fs_info
->avail_metadata_alloc_bits
;
3882 data_target
= (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) ?
3883 bctl
->data
.target
: fs_info
->avail_data_alloc_bits
;
3884 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target
) <
3885 btrfs_get_num_tolerated_disk_barrier_failures(data_target
)) {
3887 "metadata profile 0x%llx has lower redundancy than data profile 0x%llx",
3888 meta_target
, data_target
);
3891 ret
= insert_balance_item(fs_info
, bctl
);
3892 if (ret
&& ret
!= -EEXIST
)
3895 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
3896 BUG_ON(ret
== -EEXIST
);
3897 set_balance_control(bctl
);
3899 BUG_ON(ret
!= -EEXIST
);
3900 spin_lock(&fs_info
->balance_lock
);
3901 update_balance_args(bctl
);
3902 spin_unlock(&fs_info
->balance_lock
);
3905 atomic_inc(&fs_info
->balance_running
);
3906 mutex_unlock(&fs_info
->balance_mutex
);
3908 ret
= __btrfs_balance(fs_info
);
3910 mutex_lock(&fs_info
->balance_mutex
);
3911 atomic_dec(&fs_info
->balance_running
);
3914 memset(bargs
, 0, sizeof(*bargs
));
3915 update_ioctl_balance_args(fs_info
, 0, bargs
);
3918 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
3919 balance_need_close(fs_info
)) {
3920 __cancel_balance(fs_info
);
3923 wake_up(&fs_info
->balance_wait_q
);
3927 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
3928 __cancel_balance(fs_info
);
3931 clear_bit(BTRFS_FS_EXCL_OP
, &fs_info
->flags
);
3936 static int balance_kthread(void *data
)
3938 struct btrfs_fs_info
*fs_info
= data
;
3941 mutex_lock(&fs_info
->volume_mutex
);
3942 mutex_lock(&fs_info
->balance_mutex
);
3944 if (fs_info
->balance_ctl
) {
3945 btrfs_info(fs_info
, "continuing balance");
3946 ret
= btrfs_balance(fs_info
->balance_ctl
, NULL
);
3949 mutex_unlock(&fs_info
->balance_mutex
);
3950 mutex_unlock(&fs_info
->volume_mutex
);
3955 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
3957 struct task_struct
*tsk
;
3959 spin_lock(&fs_info
->balance_lock
);
3960 if (!fs_info
->balance_ctl
) {
3961 spin_unlock(&fs_info
->balance_lock
);
3964 spin_unlock(&fs_info
->balance_lock
);
3966 if (btrfs_test_opt(fs_info
, SKIP_BALANCE
)) {
3967 btrfs_info(fs_info
, "force skipping balance");
3972 * A ro->rw remount sequence should continue with the paused balance
3973 * regardless of who pauses it, system or the user as of now, so set
3976 spin_lock(&fs_info
->balance_lock
);
3977 fs_info
->balance_ctl
->flags
|= BTRFS_BALANCE_RESUME
;
3978 spin_unlock(&fs_info
->balance_lock
);
3980 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
3981 return PTR_ERR_OR_ZERO(tsk
);
3984 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
3986 struct btrfs_balance_control
*bctl
;
3987 struct btrfs_balance_item
*item
;
3988 struct btrfs_disk_balance_args disk_bargs
;
3989 struct btrfs_path
*path
;
3990 struct extent_buffer
*leaf
;
3991 struct btrfs_key key
;
3994 path
= btrfs_alloc_path();
3998 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3999 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
4002 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
4005 if (ret
> 0) { /* ret = -ENOENT; */
4010 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
4016 leaf
= path
->nodes
[0];
4017 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
4019 bctl
->fs_info
= fs_info
;
4020 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
4021 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
4023 btrfs_balance_data(leaf
, item
, &disk_bargs
);
4024 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
4025 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
4026 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
4027 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
4028 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
4030 WARN_ON(test_and_set_bit(BTRFS_FS_EXCL_OP
, &fs_info
->flags
));
4032 mutex_lock(&fs_info
->volume_mutex
);
4033 mutex_lock(&fs_info
->balance_mutex
);
4035 set_balance_control(bctl
);
4037 mutex_unlock(&fs_info
->balance_mutex
);
4038 mutex_unlock(&fs_info
->volume_mutex
);
4040 btrfs_free_path(path
);
4044 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
4048 mutex_lock(&fs_info
->balance_mutex
);
4049 if (!fs_info
->balance_ctl
) {
4050 mutex_unlock(&fs_info
->balance_mutex
);
4054 if (atomic_read(&fs_info
->balance_running
)) {
4055 atomic_inc(&fs_info
->balance_pause_req
);
4056 mutex_unlock(&fs_info
->balance_mutex
);
4058 wait_event(fs_info
->balance_wait_q
,
4059 atomic_read(&fs_info
->balance_running
) == 0);
4061 mutex_lock(&fs_info
->balance_mutex
);
4062 /* we are good with balance_ctl ripped off from under us */
4063 BUG_ON(atomic_read(&fs_info
->balance_running
));
4064 atomic_dec(&fs_info
->balance_pause_req
);
4069 mutex_unlock(&fs_info
->balance_mutex
);
4073 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
4075 if (sb_rdonly(fs_info
->sb
))
4078 mutex_lock(&fs_info
->balance_mutex
);
4079 if (!fs_info
->balance_ctl
) {
4080 mutex_unlock(&fs_info
->balance_mutex
);
4084 atomic_inc(&fs_info
->balance_cancel_req
);
4086 * if we are running just wait and return, balance item is
4087 * deleted in btrfs_balance in this case
4089 if (atomic_read(&fs_info
->balance_running
)) {
4090 mutex_unlock(&fs_info
->balance_mutex
);
4091 wait_event(fs_info
->balance_wait_q
,
4092 atomic_read(&fs_info
->balance_running
) == 0);
4093 mutex_lock(&fs_info
->balance_mutex
);
4095 /* __cancel_balance needs volume_mutex */
4096 mutex_unlock(&fs_info
->balance_mutex
);
4097 mutex_lock(&fs_info
->volume_mutex
);
4098 mutex_lock(&fs_info
->balance_mutex
);
4100 if (fs_info
->balance_ctl
)
4101 __cancel_balance(fs_info
);
4103 mutex_unlock(&fs_info
->volume_mutex
);
4106 BUG_ON(fs_info
->balance_ctl
|| atomic_read(&fs_info
->balance_running
));
4107 atomic_dec(&fs_info
->balance_cancel_req
);
4108 mutex_unlock(&fs_info
->balance_mutex
);
4112 static int btrfs_uuid_scan_kthread(void *data
)
4114 struct btrfs_fs_info
*fs_info
= data
;
4115 struct btrfs_root
*root
= fs_info
->tree_root
;
4116 struct btrfs_key key
;
4117 struct btrfs_path
*path
= NULL
;
4119 struct extent_buffer
*eb
;
4121 struct btrfs_root_item root_item
;
4123 struct btrfs_trans_handle
*trans
= NULL
;
4125 path
= btrfs_alloc_path();
4132 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4136 ret
= btrfs_search_forward(root
, &key
, path
, 0);
4143 if (key
.type
!= BTRFS_ROOT_ITEM_KEY
||
4144 (key
.objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
4145 key
.objectid
!= BTRFS_FS_TREE_OBJECTID
) ||
4146 key
.objectid
> BTRFS_LAST_FREE_OBJECTID
)
4149 eb
= path
->nodes
[0];
4150 slot
= path
->slots
[0];
4151 item_size
= btrfs_item_size_nr(eb
, slot
);
4152 if (item_size
< sizeof(root_item
))
4155 read_extent_buffer(eb
, &root_item
,
4156 btrfs_item_ptr_offset(eb
, slot
),
4157 (int)sizeof(root_item
));
4158 if (btrfs_root_refs(&root_item
) == 0)
4161 if (!btrfs_is_empty_uuid(root_item
.uuid
) ||
4162 !btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4166 btrfs_release_path(path
);
4168 * 1 - subvol uuid item
4169 * 1 - received_subvol uuid item
4171 trans
= btrfs_start_transaction(fs_info
->uuid_root
, 2);
4172 if (IS_ERR(trans
)) {
4173 ret
= PTR_ERR(trans
);
4181 if (!btrfs_is_empty_uuid(root_item
.uuid
)) {
4182 ret
= btrfs_uuid_tree_add(trans
, fs_info
,
4184 BTRFS_UUID_KEY_SUBVOL
,
4187 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4193 if (!btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4194 ret
= btrfs_uuid_tree_add(trans
, fs_info
,
4195 root_item
.received_uuid
,
4196 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
4199 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4207 ret
= btrfs_end_transaction(trans
);
4213 btrfs_release_path(path
);
4214 if (key
.offset
< (u64
)-1) {
4216 } else if (key
.type
< BTRFS_ROOT_ITEM_KEY
) {
4218 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4219 } else if (key
.objectid
< (u64
)-1) {
4221 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4230 btrfs_free_path(path
);
4231 if (trans
&& !IS_ERR(trans
))
4232 btrfs_end_transaction(trans
);
4234 btrfs_warn(fs_info
, "btrfs_uuid_scan_kthread failed %d", ret
);
4236 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN
, &fs_info
->flags
);
4237 up(&fs_info
->uuid_tree_rescan_sem
);
4242 * Callback for btrfs_uuid_tree_iterate().
4244 * 0 check succeeded, the entry is not outdated.
4245 * < 0 if an error occurred.
4246 * > 0 if the check failed, which means the caller shall remove the entry.
4248 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info
*fs_info
,
4249 u8
*uuid
, u8 type
, u64 subid
)
4251 struct btrfs_key key
;
4253 struct btrfs_root
*subvol_root
;
4255 if (type
!= BTRFS_UUID_KEY_SUBVOL
&&
4256 type
!= BTRFS_UUID_KEY_RECEIVED_SUBVOL
)
4259 key
.objectid
= subid
;
4260 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4261 key
.offset
= (u64
)-1;
4262 subvol_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
4263 if (IS_ERR(subvol_root
)) {
4264 ret
= PTR_ERR(subvol_root
);
4271 case BTRFS_UUID_KEY_SUBVOL
:
4272 if (memcmp(uuid
, subvol_root
->root_item
.uuid
, BTRFS_UUID_SIZE
))
4275 case BTRFS_UUID_KEY_RECEIVED_SUBVOL
:
4276 if (memcmp(uuid
, subvol_root
->root_item
.received_uuid
,
4286 static int btrfs_uuid_rescan_kthread(void *data
)
4288 struct btrfs_fs_info
*fs_info
= (struct btrfs_fs_info
*)data
;
4292 * 1st step is to iterate through the existing UUID tree and
4293 * to delete all entries that contain outdated data.
4294 * 2nd step is to add all missing entries to the UUID tree.
4296 ret
= btrfs_uuid_tree_iterate(fs_info
, btrfs_check_uuid_tree_entry
);
4298 btrfs_warn(fs_info
, "iterating uuid_tree failed %d", ret
);
4299 up(&fs_info
->uuid_tree_rescan_sem
);
4302 return btrfs_uuid_scan_kthread(data
);
4305 int btrfs_create_uuid_tree(struct btrfs_fs_info
*fs_info
)
4307 struct btrfs_trans_handle
*trans
;
4308 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
4309 struct btrfs_root
*uuid_root
;
4310 struct task_struct
*task
;
4317 trans
= btrfs_start_transaction(tree_root
, 2);
4319 return PTR_ERR(trans
);
4321 uuid_root
= btrfs_create_tree(trans
, fs_info
,
4322 BTRFS_UUID_TREE_OBJECTID
);
4323 if (IS_ERR(uuid_root
)) {
4324 ret
= PTR_ERR(uuid_root
);
4325 btrfs_abort_transaction(trans
, ret
);
4326 btrfs_end_transaction(trans
);
4330 fs_info
->uuid_root
= uuid_root
;
4332 ret
= btrfs_commit_transaction(trans
);
4336 down(&fs_info
->uuid_tree_rescan_sem
);
4337 task
= kthread_run(btrfs_uuid_scan_kthread
, fs_info
, "btrfs-uuid");
4339 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4340 btrfs_warn(fs_info
, "failed to start uuid_scan task");
4341 up(&fs_info
->uuid_tree_rescan_sem
);
4342 return PTR_ERR(task
);
4348 int btrfs_check_uuid_tree(struct btrfs_fs_info
*fs_info
)
4350 struct task_struct
*task
;
4352 down(&fs_info
->uuid_tree_rescan_sem
);
4353 task
= kthread_run(btrfs_uuid_rescan_kthread
, fs_info
, "btrfs-uuid");
4355 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4356 btrfs_warn(fs_info
, "failed to start uuid_rescan task");
4357 up(&fs_info
->uuid_tree_rescan_sem
);
4358 return PTR_ERR(task
);
4365 * shrinking a device means finding all of the device extents past
4366 * the new size, and then following the back refs to the chunks.
4367 * The chunk relocation code actually frees the device extent
4369 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
4371 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
4372 struct btrfs_root
*root
= fs_info
->dev_root
;
4373 struct btrfs_trans_handle
*trans
;
4374 struct btrfs_dev_extent
*dev_extent
= NULL
;
4375 struct btrfs_path
*path
;
4381 bool retried
= false;
4382 bool checked_pending_chunks
= false;
4383 struct extent_buffer
*l
;
4384 struct btrfs_key key
;
4385 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
4386 u64 old_total
= btrfs_super_total_bytes(super_copy
);
4387 u64 old_size
= btrfs_device_get_total_bytes(device
);
4390 new_size
= round_down(new_size
, fs_info
->sectorsize
);
4391 diff
= round_down(old_size
- new_size
, fs_info
->sectorsize
);
4393 if (device
->is_tgtdev_for_dev_replace
)
4396 path
= btrfs_alloc_path();
4400 path
->reada
= READA_FORWARD
;
4402 mutex_lock(&fs_info
->chunk_mutex
);
4404 btrfs_device_set_total_bytes(device
, new_size
);
4405 if (device
->writeable
) {
4406 device
->fs_devices
->total_rw_bytes
-= diff
;
4407 atomic64_sub(diff
, &fs_info
->free_chunk_space
);
4409 mutex_unlock(&fs_info
->chunk_mutex
);
4412 key
.objectid
= device
->devid
;
4413 key
.offset
= (u64
)-1;
4414 key
.type
= BTRFS_DEV_EXTENT_KEY
;
4417 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
4418 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4420 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4424 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
4426 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4431 btrfs_release_path(path
);
4436 slot
= path
->slots
[0];
4437 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
4439 if (key
.objectid
!= device
->devid
) {
4440 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4441 btrfs_release_path(path
);
4445 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
4446 length
= btrfs_dev_extent_length(l
, dev_extent
);
4448 if (key
.offset
+ length
<= new_size
) {
4449 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4450 btrfs_release_path(path
);
4454 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
4455 btrfs_release_path(path
);
4457 ret
= btrfs_relocate_chunk(fs_info
, chunk_offset
);
4458 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4459 if (ret
&& ret
!= -ENOSPC
)
4463 } while (key
.offset
-- > 0);
4465 if (failed
&& !retried
) {
4469 } else if (failed
&& retried
) {
4474 /* Shrinking succeeded, else we would be at "done". */
4475 trans
= btrfs_start_transaction(root
, 0);
4476 if (IS_ERR(trans
)) {
4477 ret
= PTR_ERR(trans
);
4481 mutex_lock(&fs_info
->chunk_mutex
);
4484 * We checked in the above loop all device extents that were already in
4485 * the device tree. However before we have updated the device's
4486 * total_bytes to the new size, we might have had chunk allocations that
4487 * have not complete yet (new block groups attached to transaction
4488 * handles), and therefore their device extents were not yet in the
4489 * device tree and we missed them in the loop above. So if we have any
4490 * pending chunk using a device extent that overlaps the device range
4491 * that we can not use anymore, commit the current transaction and
4492 * repeat the search on the device tree - this way we guarantee we will
4493 * not have chunks using device extents that end beyond 'new_size'.
4495 if (!checked_pending_chunks
) {
4496 u64 start
= new_size
;
4497 u64 len
= old_size
- new_size
;
4499 if (contains_pending_extent(trans
->transaction
, device
,
4501 mutex_unlock(&fs_info
->chunk_mutex
);
4502 checked_pending_chunks
= true;
4505 ret
= btrfs_commit_transaction(trans
);
4512 btrfs_device_set_disk_total_bytes(device
, new_size
);
4513 if (list_empty(&device
->resized_list
))
4514 list_add_tail(&device
->resized_list
,
4515 &fs_info
->fs_devices
->resized_devices
);
4517 WARN_ON(diff
> old_total
);
4518 btrfs_set_super_total_bytes(super_copy
,
4519 round_down(old_total
- diff
, fs_info
->sectorsize
));
4520 mutex_unlock(&fs_info
->chunk_mutex
);
4522 /* Now btrfs_update_device() will change the on-disk size. */
4523 ret
= btrfs_update_device(trans
, device
);
4524 btrfs_end_transaction(trans
);
4526 btrfs_free_path(path
);
4528 mutex_lock(&fs_info
->chunk_mutex
);
4529 btrfs_device_set_total_bytes(device
, old_size
);
4530 if (device
->writeable
)
4531 device
->fs_devices
->total_rw_bytes
+= diff
;
4532 atomic64_add(diff
, &fs_info
->free_chunk_space
);
4533 mutex_unlock(&fs_info
->chunk_mutex
);
4538 static int btrfs_add_system_chunk(struct btrfs_fs_info
*fs_info
,
4539 struct btrfs_key
*key
,
4540 struct btrfs_chunk
*chunk
, int item_size
)
4542 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
4543 struct btrfs_disk_key disk_key
;
4547 mutex_lock(&fs_info
->chunk_mutex
);
4548 array_size
= btrfs_super_sys_array_size(super_copy
);
4549 if (array_size
+ item_size
+ sizeof(disk_key
)
4550 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4551 mutex_unlock(&fs_info
->chunk_mutex
);
4555 ptr
= super_copy
->sys_chunk_array
+ array_size
;
4556 btrfs_cpu_key_to_disk(&disk_key
, key
);
4557 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
4558 ptr
+= sizeof(disk_key
);
4559 memcpy(ptr
, chunk
, item_size
);
4560 item_size
+= sizeof(disk_key
);
4561 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
4562 mutex_unlock(&fs_info
->chunk_mutex
);
4568 * sort the devices in descending order by max_avail, total_avail
4570 static int btrfs_cmp_device_info(const void *a
, const void *b
)
4572 const struct btrfs_device_info
*di_a
= a
;
4573 const struct btrfs_device_info
*di_b
= b
;
4575 if (di_a
->max_avail
> di_b
->max_avail
)
4577 if (di_a
->max_avail
< di_b
->max_avail
)
4579 if (di_a
->total_avail
> di_b
->total_avail
)
4581 if (di_a
->total_avail
< di_b
->total_avail
)
4586 static void check_raid56_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
4588 if (!(type
& BTRFS_BLOCK_GROUP_RAID56_MASK
))
4591 btrfs_set_fs_incompat(info
, RAID56
);
4594 #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info) \
4595 - sizeof(struct btrfs_chunk)) \
4596 / sizeof(struct btrfs_stripe) + 1)
4598 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4599 - 2 * sizeof(struct btrfs_disk_key) \
4600 - 2 * sizeof(struct btrfs_chunk)) \
4601 / sizeof(struct btrfs_stripe) + 1)
4603 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4604 u64 start
, u64 type
)
4606 struct btrfs_fs_info
*info
= trans
->fs_info
;
4607 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
4608 struct btrfs_device
*device
;
4609 struct map_lookup
*map
= NULL
;
4610 struct extent_map_tree
*em_tree
;
4611 struct extent_map
*em
;
4612 struct btrfs_device_info
*devices_info
= NULL
;
4614 int num_stripes
; /* total number of stripes to allocate */
4615 int data_stripes
; /* number of stripes that count for
4617 int sub_stripes
; /* sub_stripes info for map */
4618 int dev_stripes
; /* stripes per dev */
4619 int devs_max
; /* max devs to use */
4620 int devs_min
; /* min devs needed */
4621 int devs_increment
; /* ndevs has to be a multiple of this */
4622 int ncopies
; /* how many copies to data has */
4624 u64 max_stripe_size
;
4633 BUG_ON(!alloc_profile_is_valid(type
, 0));
4635 if (list_empty(&fs_devices
->alloc_list
))
4638 index
= __get_raid_index(type
);
4640 sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
4641 dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
4642 devs_max
= btrfs_raid_array
[index
].devs_max
;
4643 devs_min
= btrfs_raid_array
[index
].devs_min
;
4644 devs_increment
= btrfs_raid_array
[index
].devs_increment
;
4645 ncopies
= btrfs_raid_array
[index
].ncopies
;
4647 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
4648 max_stripe_size
= SZ_1G
;
4649 max_chunk_size
= 10 * max_stripe_size
;
4651 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4652 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
4653 /* for larger filesystems, use larger metadata chunks */
4654 if (fs_devices
->total_rw_bytes
> 50ULL * SZ_1G
)
4655 max_stripe_size
= SZ_1G
;
4657 max_stripe_size
= SZ_256M
;
4658 max_chunk_size
= max_stripe_size
;
4660 devs_max
= BTRFS_MAX_DEVS(info
->chunk_root
);
4661 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4662 max_stripe_size
= SZ_32M
;
4663 max_chunk_size
= 2 * max_stripe_size
;
4665 devs_max
= BTRFS_MAX_DEVS_SYS_CHUNK
;
4667 btrfs_err(info
, "invalid chunk type 0x%llx requested",
4672 /* we don't want a chunk larger than 10% of writeable space */
4673 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
4676 devices_info
= kcalloc(fs_devices
->rw_devices
, sizeof(*devices_info
),
4682 * in the first pass through the devices list, we gather information
4683 * about the available holes on each device.
4686 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
4690 if (!device
->writeable
) {
4692 "BTRFS: read-only device in alloc_list\n");
4696 if (!device
->in_fs_metadata
||
4697 device
->is_tgtdev_for_dev_replace
)
4700 if (device
->total_bytes
> device
->bytes_used
)
4701 total_avail
= device
->total_bytes
- device
->bytes_used
;
4705 /* If there is no space on this device, skip it. */
4706 if (total_avail
== 0)
4709 ret
= find_free_dev_extent(trans
, device
,
4710 max_stripe_size
* dev_stripes
,
4711 &dev_offset
, &max_avail
);
4712 if (ret
&& ret
!= -ENOSPC
)
4716 max_avail
= max_stripe_size
* dev_stripes
;
4718 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
)
4721 if (ndevs
== fs_devices
->rw_devices
) {
4722 WARN(1, "%s: found more than %llu devices\n",
4723 __func__
, fs_devices
->rw_devices
);
4726 devices_info
[ndevs
].dev_offset
= dev_offset
;
4727 devices_info
[ndevs
].max_avail
= max_avail
;
4728 devices_info
[ndevs
].total_avail
= total_avail
;
4729 devices_info
[ndevs
].dev
= device
;
4734 * now sort the devices by hole size / available space
4736 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
4737 btrfs_cmp_device_info
, NULL
);
4739 /* round down to number of usable stripes */
4740 ndevs
= round_down(ndevs
, devs_increment
);
4742 if (ndevs
< devs_increment
* sub_stripes
|| ndevs
< devs_min
) {
4747 ndevs
= min(ndevs
, devs_max
);
4750 * The primary goal is to maximize the number of stripes, so use as
4751 * many devices as possible, even if the stripes are not maximum sized.
4753 * The DUP profile stores more than one stripe per device, the
4754 * max_avail is the total size so we have to adjust.
4756 stripe_size
= div_u64(devices_info
[ndevs
- 1].max_avail
, dev_stripes
);
4757 num_stripes
= ndevs
* dev_stripes
;
4760 * this will have to be fixed for RAID1 and RAID10 over
4763 data_stripes
= num_stripes
/ ncopies
;
4765 if (type
& BTRFS_BLOCK_GROUP_RAID5
)
4766 data_stripes
= num_stripes
- 1;
4768 if (type
& BTRFS_BLOCK_GROUP_RAID6
)
4769 data_stripes
= num_stripes
- 2;
4772 * Use the number of data stripes to figure out how big this chunk
4773 * is really going to be in terms of logical address space,
4774 * and compare that answer with the max chunk size
4776 if (stripe_size
* data_stripes
> max_chunk_size
) {
4777 u64 mask
= (1ULL << 24) - 1;
4779 stripe_size
= div_u64(max_chunk_size
, data_stripes
);
4781 /* bump the answer up to a 16MB boundary */
4782 stripe_size
= (stripe_size
+ mask
) & ~mask
;
4784 /* but don't go higher than the limits we found
4785 * while searching for free extents
4787 if (stripe_size
> devices_info
[ndevs
-1].max_avail
)
4788 stripe_size
= devices_info
[ndevs
-1].max_avail
;
4791 /* align to BTRFS_STRIPE_LEN */
4792 stripe_size
= round_down(stripe_size
, BTRFS_STRIPE_LEN
);
4794 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
4799 map
->num_stripes
= num_stripes
;
4801 for (i
= 0; i
< ndevs
; ++i
) {
4802 for (j
= 0; j
< dev_stripes
; ++j
) {
4803 int s
= i
* dev_stripes
+ j
;
4804 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
4805 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
4809 map
->stripe_len
= BTRFS_STRIPE_LEN
;
4810 map
->io_align
= BTRFS_STRIPE_LEN
;
4811 map
->io_width
= BTRFS_STRIPE_LEN
;
4813 map
->sub_stripes
= sub_stripes
;
4815 num_bytes
= stripe_size
* data_stripes
;
4817 trace_btrfs_chunk_alloc(info
, map
, start
, num_bytes
);
4819 em
= alloc_extent_map();
4825 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
4826 em
->map_lookup
= map
;
4828 em
->len
= num_bytes
;
4829 em
->block_start
= 0;
4830 em
->block_len
= em
->len
;
4831 em
->orig_block_len
= stripe_size
;
4833 em_tree
= &info
->mapping_tree
.map_tree
;
4834 write_lock(&em_tree
->lock
);
4835 ret
= add_extent_mapping(em_tree
, em
, 0);
4837 write_unlock(&em_tree
->lock
);
4838 free_extent_map(em
);
4842 list_add_tail(&em
->list
, &trans
->transaction
->pending_chunks
);
4843 refcount_inc(&em
->refs
);
4844 write_unlock(&em_tree
->lock
);
4846 ret
= btrfs_make_block_group(trans
, info
, 0, type
, start
, num_bytes
);
4848 goto error_del_extent
;
4850 for (i
= 0; i
< map
->num_stripes
; i
++) {
4851 num_bytes
= map
->stripes
[i
].dev
->bytes_used
+ stripe_size
;
4852 btrfs_device_set_bytes_used(map
->stripes
[i
].dev
, num_bytes
);
4855 atomic64_sub(stripe_size
* map
->num_stripes
, &info
->free_chunk_space
);
4857 free_extent_map(em
);
4858 check_raid56_incompat_flag(info
, type
);
4860 kfree(devices_info
);
4864 write_lock(&em_tree
->lock
);
4865 remove_extent_mapping(em_tree
, em
);
4866 write_unlock(&em_tree
->lock
);
4868 /* One for our allocation */
4869 free_extent_map(em
);
4870 /* One for the tree reference */
4871 free_extent_map(em
);
4872 /* One for the pending_chunks list reference */
4873 free_extent_map(em
);
4875 kfree(devices_info
);
4879 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
4880 struct btrfs_fs_info
*fs_info
,
4881 u64 chunk_offset
, u64 chunk_size
)
4883 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
4884 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
4885 struct btrfs_key key
;
4886 struct btrfs_device
*device
;
4887 struct btrfs_chunk
*chunk
;
4888 struct btrfs_stripe
*stripe
;
4889 struct extent_map
*em
;
4890 struct map_lookup
*map
;
4897 em
= get_chunk_map(fs_info
, chunk_offset
, chunk_size
);
4901 map
= em
->map_lookup
;
4902 item_size
= btrfs_chunk_item_size(map
->num_stripes
);
4903 stripe_size
= em
->orig_block_len
;
4905 chunk
= kzalloc(item_size
, GFP_NOFS
);
4912 * Take the device list mutex to prevent races with the final phase of
4913 * a device replace operation that replaces the device object associated
4914 * with the map's stripes, because the device object's id can change
4915 * at any time during that final phase of the device replace operation
4916 * (dev-replace.c:btrfs_dev_replace_finishing()).
4918 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
4919 for (i
= 0; i
< map
->num_stripes
; i
++) {
4920 device
= map
->stripes
[i
].dev
;
4921 dev_offset
= map
->stripes
[i
].physical
;
4923 ret
= btrfs_update_device(trans
, device
);
4926 ret
= btrfs_alloc_dev_extent(trans
, device
, chunk_offset
,
4927 dev_offset
, stripe_size
);
4932 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
4936 stripe
= &chunk
->stripe
;
4937 for (i
= 0; i
< map
->num_stripes
; i
++) {
4938 device
= map
->stripes
[i
].dev
;
4939 dev_offset
= map
->stripes
[i
].physical
;
4941 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
4942 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
4943 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
4946 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
4948 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
4949 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
4950 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
4951 btrfs_set_stack_chunk_type(chunk
, map
->type
);
4952 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
4953 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
4954 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
4955 btrfs_set_stack_chunk_sector_size(chunk
, fs_info
->sectorsize
);
4956 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
4958 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
4959 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
4960 key
.offset
= chunk_offset
;
4962 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
4963 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4965 * TODO: Cleanup of inserted chunk root in case of
4968 ret
= btrfs_add_system_chunk(fs_info
, &key
, chunk
, item_size
);
4973 free_extent_map(em
);
4978 * Chunk allocation falls into two parts. The first part does works
4979 * that make the new allocated chunk useable, but not do any operation
4980 * that modifies the chunk tree. The second part does the works that
4981 * require modifying the chunk tree. This division is important for the
4982 * bootstrap process of adding storage to a seed btrfs.
4984 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4985 struct btrfs_fs_info
*fs_info
, u64 type
)
4989 ASSERT(mutex_is_locked(&fs_info
->chunk_mutex
));
4990 chunk_offset
= find_next_chunk(fs_info
);
4991 return __btrfs_alloc_chunk(trans
, chunk_offset
, type
);
4994 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
4995 struct btrfs_fs_info
*fs_info
)
4998 u64 sys_chunk_offset
;
5002 chunk_offset
= find_next_chunk(fs_info
);
5003 alloc_profile
= btrfs_metadata_alloc_profile(fs_info
);
5004 ret
= __btrfs_alloc_chunk(trans
, chunk_offset
, alloc_profile
);
5008 sys_chunk_offset
= find_next_chunk(fs_info
);
5009 alloc_profile
= btrfs_system_alloc_profile(fs_info
);
5010 ret
= __btrfs_alloc_chunk(trans
, sys_chunk_offset
, alloc_profile
);
5014 static inline int btrfs_chunk_max_errors(struct map_lookup
*map
)
5018 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
5019 BTRFS_BLOCK_GROUP_RAID10
|
5020 BTRFS_BLOCK_GROUP_RAID5
|
5021 BTRFS_BLOCK_GROUP_DUP
)) {
5023 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
) {
5032 int btrfs_chunk_readonly(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
5034 struct extent_map
*em
;
5035 struct map_lookup
*map
;
5040 em
= get_chunk_map(fs_info
, chunk_offset
, 1);
5044 map
= em
->map_lookup
;
5045 for (i
= 0; i
< map
->num_stripes
; i
++) {
5046 if (map
->stripes
[i
].dev
->missing
) {
5051 if (!map
->stripes
[i
].dev
->writeable
) {
5058 * If the number of missing devices is larger than max errors,
5059 * we can not write the data into that chunk successfully, so
5062 if (miss_ndevs
> btrfs_chunk_max_errors(map
))
5065 free_extent_map(em
);
5069 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
5071 extent_map_tree_init(&tree
->map_tree
);
5074 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
5076 struct extent_map
*em
;
5079 write_lock(&tree
->map_tree
.lock
);
5080 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
5082 remove_extent_mapping(&tree
->map_tree
, em
);
5083 write_unlock(&tree
->map_tree
.lock
);
5087 free_extent_map(em
);
5088 /* once for the tree */
5089 free_extent_map(em
);
5093 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
5095 struct extent_map
*em
;
5096 struct map_lookup
*map
;
5099 em
= get_chunk_map(fs_info
, logical
, len
);
5102 * We could return errors for these cases, but that could get
5103 * ugly and we'd probably do the same thing which is just not do
5104 * anything else and exit, so return 1 so the callers don't try
5105 * to use other copies.
5109 map
= em
->map_lookup
;
5110 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
5111 ret
= map
->num_stripes
;
5112 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5113 ret
= map
->sub_stripes
;
5114 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID5
)
5116 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5118 * There could be two corrupted data stripes, we need
5119 * to loop retry in order to rebuild the correct data.
5121 * Fail a stripe at a time on every retry except the
5122 * stripe under reconstruction.
5124 ret
= map
->num_stripes
;
5127 free_extent_map(em
);
5129 btrfs_dev_replace_lock(&fs_info
->dev_replace
, 0);
5130 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
) &&
5131 fs_info
->dev_replace
.tgtdev
)
5133 btrfs_dev_replace_unlock(&fs_info
->dev_replace
, 0);
5138 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info
*fs_info
,
5141 struct extent_map
*em
;
5142 struct map_lookup
*map
;
5143 unsigned long len
= fs_info
->sectorsize
;
5145 em
= get_chunk_map(fs_info
, logical
, len
);
5147 if (!WARN_ON(IS_ERR(em
))) {
5148 map
= em
->map_lookup
;
5149 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5150 len
= map
->stripe_len
* nr_data_stripes(map
);
5151 free_extent_map(em
);
5156 int btrfs_is_parity_mirror(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
5158 struct extent_map
*em
;
5159 struct map_lookup
*map
;
5162 em
= get_chunk_map(fs_info
, logical
, len
);
5164 if(!WARN_ON(IS_ERR(em
))) {
5165 map
= em
->map_lookup
;
5166 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5168 free_extent_map(em
);
5173 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
5174 struct map_lookup
*map
, int first
, int num
,
5175 int optimal
, int dev_replace_is_ongoing
)
5179 struct btrfs_device
*srcdev
;
5181 if (dev_replace_is_ongoing
&&
5182 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
5183 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
5184 srcdev
= fs_info
->dev_replace
.srcdev
;
5189 * try to avoid the drive that is the source drive for a
5190 * dev-replace procedure, only choose it if no other non-missing
5191 * mirror is available
5193 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
5194 if (map
->stripes
[optimal
].dev
->bdev
&&
5195 (tolerance
|| map
->stripes
[optimal
].dev
!= srcdev
))
5197 for (i
= first
; i
< first
+ num
; i
++) {
5198 if (map
->stripes
[i
].dev
->bdev
&&
5199 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
5204 /* we couldn't find one that doesn't fail. Just return something
5205 * and the io error handling code will clean up eventually
5210 static inline int parity_smaller(u64 a
, u64 b
)
5215 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5216 static void sort_parity_stripes(struct btrfs_bio
*bbio
, int num_stripes
)
5218 struct btrfs_bio_stripe s
;
5225 for (i
= 0; i
< num_stripes
- 1; i
++) {
5226 if (parity_smaller(bbio
->raid_map
[i
],
5227 bbio
->raid_map
[i
+1])) {
5228 s
= bbio
->stripes
[i
];
5229 l
= bbio
->raid_map
[i
];
5230 bbio
->stripes
[i
] = bbio
->stripes
[i
+1];
5231 bbio
->raid_map
[i
] = bbio
->raid_map
[i
+1];
5232 bbio
->stripes
[i
+1] = s
;
5233 bbio
->raid_map
[i
+1] = l
;
5241 static struct btrfs_bio
*alloc_btrfs_bio(int total_stripes
, int real_stripes
)
5243 struct btrfs_bio
*bbio
= kzalloc(
5244 /* the size of the btrfs_bio */
5245 sizeof(struct btrfs_bio
) +
5246 /* plus the variable array for the stripes */
5247 sizeof(struct btrfs_bio_stripe
) * (total_stripes
) +
5248 /* plus the variable array for the tgt dev */
5249 sizeof(int) * (real_stripes
) +
5251 * plus the raid_map, which includes both the tgt dev
5254 sizeof(u64
) * (total_stripes
),
5255 GFP_NOFS
|__GFP_NOFAIL
);
5257 atomic_set(&bbio
->error
, 0);
5258 refcount_set(&bbio
->refs
, 1);
5263 void btrfs_get_bbio(struct btrfs_bio
*bbio
)
5265 WARN_ON(!refcount_read(&bbio
->refs
));
5266 refcount_inc(&bbio
->refs
);
5269 void btrfs_put_bbio(struct btrfs_bio
*bbio
)
5273 if (refcount_dec_and_test(&bbio
->refs
))
5277 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5279 * Please note that, discard won't be sent to target device of device
5282 static int __btrfs_map_block_for_discard(struct btrfs_fs_info
*fs_info
,
5283 u64 logical
, u64 length
,
5284 struct btrfs_bio
**bbio_ret
)
5286 struct extent_map
*em
;
5287 struct map_lookup
*map
;
5288 struct btrfs_bio
*bbio
;
5292 u64 stripe_end_offset
;
5299 u32 sub_stripes
= 0;
5300 u64 stripes_per_dev
= 0;
5301 u32 remaining_stripes
= 0;
5302 u32 last_stripe
= 0;
5306 /* discard always return a bbio */
5309 em
= get_chunk_map(fs_info
, logical
, length
);
5313 map
= em
->map_lookup
;
5314 /* we don't discard raid56 yet */
5315 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5320 offset
= logical
- em
->start
;
5321 length
= min_t(u64
, em
->len
- offset
, length
);
5323 stripe_len
= map
->stripe_len
;
5325 * stripe_nr counts the total number of stripes we have to stride
5326 * to get to this block
5328 stripe_nr
= div64_u64(offset
, stripe_len
);
5330 /* stripe_offset is the offset of this block in its stripe */
5331 stripe_offset
= offset
- stripe_nr
* stripe_len
;
5333 stripe_nr_end
= round_up(offset
+ length
, map
->stripe_len
);
5334 stripe_nr_end
= div64_u64(stripe_nr_end
, map
->stripe_len
);
5335 stripe_cnt
= stripe_nr_end
- stripe_nr
;
5336 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
5339 * after this, stripe_nr is the number of stripes on this
5340 * device we have to walk to find the data, and stripe_index is
5341 * the number of our device in the stripe array
5345 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5346 BTRFS_BLOCK_GROUP_RAID10
)) {
5347 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5350 sub_stripes
= map
->sub_stripes
;
5352 factor
= map
->num_stripes
/ sub_stripes
;
5353 num_stripes
= min_t(u64
, map
->num_stripes
,
5354 sub_stripes
* stripe_cnt
);
5355 stripe_nr
= div_u64_rem(stripe_nr
, factor
, &stripe_index
);
5356 stripe_index
*= sub_stripes
;
5357 stripes_per_dev
= div_u64_rem(stripe_cnt
, factor
,
5358 &remaining_stripes
);
5359 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
5360 last_stripe
*= sub_stripes
;
5361 } else if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
5362 BTRFS_BLOCK_GROUP_DUP
)) {
5363 num_stripes
= map
->num_stripes
;
5365 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5369 bbio
= alloc_btrfs_bio(num_stripes
, 0);
5375 for (i
= 0; i
< num_stripes
; i
++) {
5376 bbio
->stripes
[i
].physical
=
5377 map
->stripes
[stripe_index
].physical
+
5378 stripe_offset
+ stripe_nr
* map
->stripe_len
;
5379 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
5381 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5382 BTRFS_BLOCK_GROUP_RAID10
)) {
5383 bbio
->stripes
[i
].length
= stripes_per_dev
*
5386 if (i
/ sub_stripes
< remaining_stripes
)
5387 bbio
->stripes
[i
].length
+=
5391 * Special for the first stripe and
5394 * |-------|...|-------|
5398 if (i
< sub_stripes
)
5399 bbio
->stripes
[i
].length
-=
5402 if (stripe_index
>= last_stripe
&&
5403 stripe_index
<= (last_stripe
+
5405 bbio
->stripes
[i
].length
-=
5408 if (i
== sub_stripes
- 1)
5411 bbio
->stripes
[i
].length
= length
;
5415 if (stripe_index
== map
->num_stripes
) {
5422 bbio
->map_type
= map
->type
;
5423 bbio
->num_stripes
= num_stripes
;
5425 free_extent_map(em
);
5430 * In dev-replace case, for repair case (that's the only case where the mirror
5431 * is selected explicitly when calling btrfs_map_block), blocks left of the
5432 * left cursor can also be read from the target drive.
5434 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5436 * For READ, it also needs to be supported using the same mirror number.
5438 * If the requested block is not left of the left cursor, EIO is returned. This
5439 * can happen because btrfs_num_copies() returns one more in the dev-replace
5442 static int get_extra_mirror_from_replace(struct btrfs_fs_info
*fs_info
,
5443 u64 logical
, u64 length
,
5444 u64 srcdev_devid
, int *mirror_num
,
5447 struct btrfs_bio
*bbio
= NULL
;
5449 int index_srcdev
= 0;
5451 u64 physical_of_found
= 0;
5455 ret
= __btrfs_map_block(fs_info
, BTRFS_MAP_GET_READ_MIRRORS
,
5456 logical
, &length
, &bbio
, 0, 0);
5458 ASSERT(bbio
== NULL
);
5462 num_stripes
= bbio
->num_stripes
;
5463 if (*mirror_num
> num_stripes
) {
5465 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5466 * that means that the requested area is not left of the left
5469 btrfs_put_bbio(bbio
);
5474 * process the rest of the function using the mirror_num of the source
5475 * drive. Therefore look it up first. At the end, patch the device
5476 * pointer to the one of the target drive.
5478 for (i
= 0; i
< num_stripes
; i
++) {
5479 if (bbio
->stripes
[i
].dev
->devid
!= srcdev_devid
)
5483 * In case of DUP, in order to keep it simple, only add the
5484 * mirror with the lowest physical address
5487 physical_of_found
<= bbio
->stripes
[i
].physical
)
5492 physical_of_found
= bbio
->stripes
[i
].physical
;
5495 btrfs_put_bbio(bbio
);
5501 *mirror_num
= index_srcdev
+ 1;
5502 *physical
= physical_of_found
;
5506 static void handle_ops_on_dev_replace(enum btrfs_map_op op
,
5507 struct btrfs_bio
**bbio_ret
,
5508 struct btrfs_dev_replace
*dev_replace
,
5509 int *num_stripes_ret
, int *max_errors_ret
)
5511 struct btrfs_bio
*bbio
= *bbio_ret
;
5512 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5513 int tgtdev_indexes
= 0;
5514 int num_stripes
= *num_stripes_ret
;
5515 int max_errors
= *max_errors_ret
;
5518 if (op
== BTRFS_MAP_WRITE
) {
5519 int index_where_to_add
;
5522 * duplicate the write operations while the dev replace
5523 * procedure is running. Since the copying of the old disk to
5524 * the new disk takes place at run time while the filesystem is
5525 * mounted writable, the regular write operations to the old
5526 * disk have to be duplicated to go to the new disk as well.
5528 * Note that device->missing is handled by the caller, and that
5529 * the write to the old disk is already set up in the stripes
5532 index_where_to_add
= num_stripes
;
5533 for (i
= 0; i
< num_stripes
; i
++) {
5534 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5535 /* write to new disk, too */
5536 struct btrfs_bio_stripe
*new =
5537 bbio
->stripes
+ index_where_to_add
;
5538 struct btrfs_bio_stripe
*old
=
5541 new->physical
= old
->physical
;
5542 new->length
= old
->length
;
5543 new->dev
= dev_replace
->tgtdev
;
5544 bbio
->tgtdev_map
[i
] = index_where_to_add
;
5545 index_where_to_add
++;
5550 num_stripes
= index_where_to_add
;
5551 } else if (op
== BTRFS_MAP_GET_READ_MIRRORS
) {
5552 int index_srcdev
= 0;
5554 u64 physical_of_found
= 0;
5557 * During the dev-replace procedure, the target drive can also
5558 * be used to read data in case it is needed to repair a corrupt
5559 * block elsewhere. This is possible if the requested area is
5560 * left of the left cursor. In this area, the target drive is a
5561 * full copy of the source drive.
5563 for (i
= 0; i
< num_stripes
; i
++) {
5564 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5566 * In case of DUP, in order to keep it simple,
5567 * only add the mirror with the lowest physical
5571 physical_of_found
<=
5572 bbio
->stripes
[i
].physical
)
5576 physical_of_found
= bbio
->stripes
[i
].physical
;
5580 struct btrfs_bio_stripe
*tgtdev_stripe
=
5581 bbio
->stripes
+ num_stripes
;
5583 tgtdev_stripe
->physical
= physical_of_found
;
5584 tgtdev_stripe
->length
=
5585 bbio
->stripes
[index_srcdev
].length
;
5586 tgtdev_stripe
->dev
= dev_replace
->tgtdev
;
5587 bbio
->tgtdev_map
[index_srcdev
] = num_stripes
;
5594 *num_stripes_ret
= num_stripes
;
5595 *max_errors_ret
= max_errors
;
5596 bbio
->num_tgtdevs
= tgtdev_indexes
;
5600 static bool need_full_stripe(enum btrfs_map_op op
)
5602 return (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
);
5605 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
,
5606 enum btrfs_map_op op
,
5607 u64 logical
, u64
*length
,
5608 struct btrfs_bio
**bbio_ret
,
5609 int mirror_num
, int need_raid_map
)
5611 struct extent_map
*em
;
5612 struct map_lookup
*map
;
5622 int tgtdev_indexes
= 0;
5623 struct btrfs_bio
*bbio
= NULL
;
5624 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
5625 int dev_replace_is_ongoing
= 0;
5626 int num_alloc_stripes
;
5627 int patch_the_first_stripe_for_dev_replace
= 0;
5628 u64 physical_to_patch_in_first_stripe
= 0;
5629 u64 raid56_full_stripe_start
= (u64
)-1;
5631 if (op
== BTRFS_MAP_DISCARD
)
5632 return __btrfs_map_block_for_discard(fs_info
, logical
,
5635 em
= get_chunk_map(fs_info
, logical
, *length
);
5639 map
= em
->map_lookup
;
5640 offset
= logical
- em
->start
;
5642 stripe_len
= map
->stripe_len
;
5645 * stripe_nr counts the total number of stripes we have to stride
5646 * to get to this block
5648 stripe_nr
= div64_u64(stripe_nr
, stripe_len
);
5650 stripe_offset
= stripe_nr
* stripe_len
;
5651 if (offset
< stripe_offset
) {
5653 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5654 stripe_offset
, offset
, em
->start
, logical
,
5656 free_extent_map(em
);
5660 /* stripe_offset is the offset of this block in its stripe*/
5661 stripe_offset
= offset
- stripe_offset
;
5663 /* if we're here for raid56, we need to know the stripe aligned start */
5664 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5665 unsigned long full_stripe_len
= stripe_len
* nr_data_stripes(map
);
5666 raid56_full_stripe_start
= offset
;
5668 /* allow a write of a full stripe, but make sure we don't
5669 * allow straddling of stripes
5671 raid56_full_stripe_start
= div64_u64(raid56_full_stripe_start
,
5673 raid56_full_stripe_start
*= full_stripe_len
;
5676 if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
5678 /* For writes to RAID[56], allow a full stripeset across all disks.
5679 For other RAID types and for RAID[56] reads, just allow a single
5680 stripe (on a single disk). */
5681 if ((map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
5682 (op
== BTRFS_MAP_WRITE
)) {
5683 max_len
= stripe_len
* nr_data_stripes(map
) -
5684 (offset
- raid56_full_stripe_start
);
5686 /* we limit the length of each bio to what fits in a stripe */
5687 max_len
= stripe_len
- stripe_offset
;
5689 *length
= min_t(u64
, em
->len
- offset
, max_len
);
5691 *length
= em
->len
- offset
;
5694 /* This is for when we're called from btrfs_merge_bio_hook() and all
5695 it cares about is the length */
5699 btrfs_dev_replace_lock(dev_replace
, 0);
5700 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
5701 if (!dev_replace_is_ongoing
)
5702 btrfs_dev_replace_unlock(dev_replace
, 0);
5704 btrfs_dev_replace_set_lock_blocking(dev_replace
);
5706 if (dev_replace_is_ongoing
&& mirror_num
== map
->num_stripes
+ 1 &&
5707 !need_full_stripe(op
) && dev_replace
->tgtdev
!= NULL
) {
5708 ret
= get_extra_mirror_from_replace(fs_info
, logical
, *length
,
5709 dev_replace
->srcdev
->devid
,
5711 &physical_to_patch_in_first_stripe
);
5715 patch_the_first_stripe_for_dev_replace
= 1;
5716 } else if (mirror_num
> map
->num_stripes
) {
5722 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5723 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5725 if (!need_full_stripe(op
))
5727 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
5728 if (need_full_stripe(op
))
5729 num_stripes
= map
->num_stripes
;
5730 else if (mirror_num
)
5731 stripe_index
= mirror_num
- 1;
5733 stripe_index
= find_live_mirror(fs_info
, map
, 0,
5735 current
->pid
% map
->num_stripes
,
5736 dev_replace_is_ongoing
);
5737 mirror_num
= stripe_index
+ 1;
5740 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
5741 if (need_full_stripe(op
)) {
5742 num_stripes
= map
->num_stripes
;
5743 } else if (mirror_num
) {
5744 stripe_index
= mirror_num
- 1;
5749 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5750 u32 factor
= map
->num_stripes
/ map
->sub_stripes
;
5752 stripe_nr
= div_u64_rem(stripe_nr
, factor
, &stripe_index
);
5753 stripe_index
*= map
->sub_stripes
;
5755 if (need_full_stripe(op
))
5756 num_stripes
= map
->sub_stripes
;
5757 else if (mirror_num
)
5758 stripe_index
+= mirror_num
- 1;
5760 int old_stripe_index
= stripe_index
;
5761 stripe_index
= find_live_mirror(fs_info
, map
,
5763 map
->sub_stripes
, stripe_index
+
5764 current
->pid
% map
->sub_stripes
,
5765 dev_replace_is_ongoing
);
5766 mirror_num
= stripe_index
- old_stripe_index
+ 1;
5769 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5770 if (need_raid_map
&& (need_full_stripe(op
) || mirror_num
> 1)) {
5771 /* push stripe_nr back to the start of the full stripe */
5772 stripe_nr
= div64_u64(raid56_full_stripe_start
,
5773 stripe_len
* nr_data_stripes(map
));
5775 /* RAID[56] write or recovery. Return all stripes */
5776 num_stripes
= map
->num_stripes
;
5777 max_errors
= nr_parity_stripes(map
);
5779 *length
= map
->stripe_len
;
5784 * Mirror #0 or #1 means the original data block.
5785 * Mirror #2 is RAID5 parity block.
5786 * Mirror #3 is RAID6 Q block.
5788 stripe_nr
= div_u64_rem(stripe_nr
,
5789 nr_data_stripes(map
), &stripe_index
);
5791 stripe_index
= nr_data_stripes(map
) +
5794 /* We distribute the parity blocks across stripes */
5795 div_u64_rem(stripe_nr
+ stripe_index
, map
->num_stripes
,
5797 if (!need_full_stripe(op
) && mirror_num
<= 1)
5802 * after this, stripe_nr is the number of stripes on this
5803 * device we have to walk to find the data, and stripe_index is
5804 * the number of our device in the stripe array
5806 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5808 mirror_num
= stripe_index
+ 1;
5810 if (stripe_index
>= map
->num_stripes
) {
5812 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5813 stripe_index
, map
->num_stripes
);
5818 num_alloc_stripes
= num_stripes
;
5819 if (dev_replace_is_ongoing
&& dev_replace
->tgtdev
!= NULL
) {
5820 if (op
== BTRFS_MAP_WRITE
)
5821 num_alloc_stripes
<<= 1;
5822 if (op
== BTRFS_MAP_GET_READ_MIRRORS
)
5823 num_alloc_stripes
++;
5824 tgtdev_indexes
= num_stripes
;
5827 bbio
= alloc_btrfs_bio(num_alloc_stripes
, tgtdev_indexes
);
5832 if (dev_replace_is_ongoing
&& dev_replace
->tgtdev
!= NULL
)
5833 bbio
->tgtdev_map
= (int *)(bbio
->stripes
+ num_alloc_stripes
);
5835 /* build raid_map */
5836 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
&& need_raid_map
&&
5837 (need_full_stripe(op
) || mirror_num
> 1)) {
5841 bbio
->raid_map
= (u64
*)((void *)bbio
->stripes
+
5842 sizeof(struct btrfs_bio_stripe
) *
5844 sizeof(int) * tgtdev_indexes
);
5846 /* Work out the disk rotation on this stripe-set */
5847 div_u64_rem(stripe_nr
, num_stripes
, &rot
);
5849 /* Fill in the logical address of each stripe */
5850 tmp
= stripe_nr
* nr_data_stripes(map
);
5851 for (i
= 0; i
< nr_data_stripes(map
); i
++)
5852 bbio
->raid_map
[(i
+rot
) % num_stripes
] =
5853 em
->start
+ (tmp
+ i
) * map
->stripe_len
;
5855 bbio
->raid_map
[(i
+rot
) % map
->num_stripes
] = RAID5_P_STRIPE
;
5856 if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5857 bbio
->raid_map
[(i
+rot
+1) % num_stripes
] =
5862 for (i
= 0; i
< num_stripes
; i
++) {
5863 bbio
->stripes
[i
].physical
=
5864 map
->stripes
[stripe_index
].physical
+
5866 stripe_nr
* map
->stripe_len
;
5867 bbio
->stripes
[i
].dev
=
5868 map
->stripes
[stripe_index
].dev
;
5872 if (need_full_stripe(op
))
5873 max_errors
= btrfs_chunk_max_errors(map
);
5876 sort_parity_stripes(bbio
, num_stripes
);
5878 if (dev_replace_is_ongoing
&& dev_replace
->tgtdev
!= NULL
&&
5879 need_full_stripe(op
)) {
5880 handle_ops_on_dev_replace(op
, &bbio
, dev_replace
, &num_stripes
,
5885 bbio
->map_type
= map
->type
;
5886 bbio
->num_stripes
= num_stripes
;
5887 bbio
->max_errors
= max_errors
;
5888 bbio
->mirror_num
= mirror_num
;
5891 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5892 * mirror_num == num_stripes + 1 && dev_replace target drive is
5893 * available as a mirror
5895 if (patch_the_first_stripe_for_dev_replace
&& num_stripes
> 0) {
5896 WARN_ON(num_stripes
> 1);
5897 bbio
->stripes
[0].dev
= dev_replace
->tgtdev
;
5898 bbio
->stripes
[0].physical
= physical_to_patch_in_first_stripe
;
5899 bbio
->mirror_num
= map
->num_stripes
+ 1;
5902 if (dev_replace_is_ongoing
) {
5903 btrfs_dev_replace_clear_lock_blocking(dev_replace
);
5904 btrfs_dev_replace_unlock(dev_replace
, 0);
5906 free_extent_map(em
);
5910 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, enum btrfs_map_op op
,
5911 u64 logical
, u64
*length
,
5912 struct btrfs_bio
**bbio_ret
, int mirror_num
)
5914 return __btrfs_map_block(fs_info
, op
, logical
, length
, bbio_ret
,
5918 /* For Scrub/replace */
5919 int btrfs_map_sblock(struct btrfs_fs_info
*fs_info
, enum btrfs_map_op op
,
5920 u64 logical
, u64
*length
,
5921 struct btrfs_bio
**bbio_ret
)
5923 return __btrfs_map_block(fs_info
, op
, logical
, length
, bbio_ret
, 0, 1);
5926 int btrfs_rmap_block(struct btrfs_fs_info
*fs_info
,
5927 u64 chunk_start
, u64 physical
, u64 devid
,
5928 u64
**logical
, int *naddrs
, int *stripe_len
)
5930 struct extent_map
*em
;
5931 struct map_lookup
*map
;
5939 em
= get_chunk_map(fs_info
, chunk_start
, 1);
5943 map
= em
->map_lookup
;
5945 rmap_len
= map
->stripe_len
;
5947 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5948 length
= div_u64(length
, map
->num_stripes
/ map
->sub_stripes
);
5949 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5950 length
= div_u64(length
, map
->num_stripes
);
5951 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5952 length
= div_u64(length
, nr_data_stripes(map
));
5953 rmap_len
= map
->stripe_len
* nr_data_stripes(map
);
5956 buf
= kcalloc(map
->num_stripes
, sizeof(u64
), GFP_NOFS
);
5957 BUG_ON(!buf
); /* -ENOMEM */
5959 for (i
= 0; i
< map
->num_stripes
; i
++) {
5960 if (devid
&& map
->stripes
[i
].dev
->devid
!= devid
)
5962 if (map
->stripes
[i
].physical
> physical
||
5963 map
->stripes
[i
].physical
+ length
<= physical
)
5966 stripe_nr
= physical
- map
->stripes
[i
].physical
;
5967 stripe_nr
= div64_u64(stripe_nr
, map
->stripe_len
);
5969 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5970 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5971 stripe_nr
= div_u64(stripe_nr
, map
->sub_stripes
);
5972 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5973 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5974 } /* else if RAID[56], multiply by nr_data_stripes().
5975 * Alternatively, just use rmap_len below instead of
5976 * map->stripe_len */
5978 bytenr
= chunk_start
+ stripe_nr
* rmap_len
;
5979 WARN_ON(nr
>= map
->num_stripes
);
5980 for (j
= 0; j
< nr
; j
++) {
5981 if (buf
[j
] == bytenr
)
5985 WARN_ON(nr
>= map
->num_stripes
);
5992 *stripe_len
= rmap_len
;
5994 free_extent_map(em
);
5998 static inline void btrfs_end_bbio(struct btrfs_bio
*bbio
, struct bio
*bio
)
6000 bio
->bi_private
= bbio
->private;
6001 bio
->bi_end_io
= bbio
->end_io
;
6004 btrfs_put_bbio(bbio
);
6007 static void btrfs_end_bio(struct bio
*bio
)
6009 struct btrfs_bio
*bbio
= bio
->bi_private
;
6010 int is_orig_bio
= 0;
6012 if (bio
->bi_status
) {
6013 atomic_inc(&bbio
->error
);
6014 if (bio
->bi_status
== BLK_STS_IOERR
||
6015 bio
->bi_status
== BLK_STS_TARGET
) {
6016 unsigned int stripe_index
=
6017 btrfs_io_bio(bio
)->stripe_index
;
6018 struct btrfs_device
*dev
;
6020 BUG_ON(stripe_index
>= bbio
->num_stripes
);
6021 dev
= bbio
->stripes
[stripe_index
].dev
;
6023 if (bio_op(bio
) == REQ_OP_WRITE
)
6024 btrfs_dev_stat_inc(dev
,
6025 BTRFS_DEV_STAT_WRITE_ERRS
);
6027 btrfs_dev_stat_inc(dev
,
6028 BTRFS_DEV_STAT_READ_ERRS
);
6029 if (bio
->bi_opf
& REQ_PREFLUSH
)
6030 btrfs_dev_stat_inc(dev
,
6031 BTRFS_DEV_STAT_FLUSH_ERRS
);
6032 btrfs_dev_stat_print_on_error(dev
);
6037 if (bio
== bbio
->orig_bio
)
6040 btrfs_bio_counter_dec(bbio
->fs_info
);
6042 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
6045 bio
= bbio
->orig_bio
;
6048 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
6049 /* only send an error to the higher layers if it is
6050 * beyond the tolerance of the btrfs bio
6052 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
6053 bio
->bi_status
= BLK_STS_IOERR
;
6056 * this bio is actually up to date, we didn't
6057 * go over the max number of errors
6059 bio
->bi_status
= BLK_STS_OK
;
6062 btrfs_end_bbio(bbio
, bio
);
6063 } else if (!is_orig_bio
) {
6069 * see run_scheduled_bios for a description of why bios are collected for
6072 * This will add one bio to the pending list for a device and make sure
6073 * the work struct is scheduled.
6075 static noinline
void btrfs_schedule_bio(struct btrfs_device
*device
,
6078 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
6079 int should_queue
= 1;
6080 struct btrfs_pending_bios
*pending_bios
;
6082 if (device
->missing
|| !device
->bdev
) {
6087 /* don't bother with additional async steps for reads, right now */
6088 if (bio_op(bio
) == REQ_OP_READ
) {
6090 btrfsic_submit_bio(bio
);
6095 WARN_ON(bio
->bi_next
);
6096 bio
->bi_next
= NULL
;
6098 spin_lock(&device
->io_lock
);
6099 if (op_is_sync(bio
->bi_opf
))
6100 pending_bios
= &device
->pending_sync_bios
;
6102 pending_bios
= &device
->pending_bios
;
6104 if (pending_bios
->tail
)
6105 pending_bios
->tail
->bi_next
= bio
;
6107 pending_bios
->tail
= bio
;
6108 if (!pending_bios
->head
)
6109 pending_bios
->head
= bio
;
6110 if (device
->running_pending
)
6113 spin_unlock(&device
->io_lock
);
6116 btrfs_queue_work(fs_info
->submit_workers
, &device
->work
);
6119 static void submit_stripe_bio(struct btrfs_bio
*bbio
, struct bio
*bio
,
6120 u64 physical
, int dev_nr
, int async
)
6122 struct btrfs_device
*dev
= bbio
->stripes
[dev_nr
].dev
;
6123 struct btrfs_fs_info
*fs_info
= bbio
->fs_info
;
6125 bio
->bi_private
= bbio
;
6126 btrfs_io_bio(bio
)->stripe_index
= dev_nr
;
6127 bio
->bi_end_io
= btrfs_end_bio
;
6128 bio
->bi_iter
.bi_sector
= physical
>> 9;
6131 struct rcu_string
*name
;
6134 name
= rcu_dereference(dev
->name
);
6135 btrfs_debug(fs_info
,
6136 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6137 bio_op(bio
), bio
->bi_opf
,
6138 (u64
)bio
->bi_iter
.bi_sector
,
6139 (u_long
)dev
->bdev
->bd_dev
, name
->str
, dev
->devid
,
6140 bio
->bi_iter
.bi_size
);
6144 bio_set_dev(bio
, dev
->bdev
);
6146 btrfs_bio_counter_inc_noblocked(fs_info
);
6149 btrfs_schedule_bio(dev
, bio
);
6151 btrfsic_submit_bio(bio
);
6154 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
6156 atomic_inc(&bbio
->error
);
6157 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
6158 /* Should be the original bio. */
6159 WARN_ON(bio
!= bbio
->orig_bio
);
6161 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
6162 bio
->bi_iter
.bi_sector
= logical
>> 9;
6163 if (atomic_read(&bbio
->error
) > bbio
->max_errors
)
6164 bio
->bi_status
= BLK_STS_IOERR
;
6166 bio
->bi_status
= BLK_STS_OK
;
6167 btrfs_end_bbio(bbio
, bio
);
6171 blk_status_t
btrfs_map_bio(struct btrfs_fs_info
*fs_info
, struct bio
*bio
,
6172 int mirror_num
, int async_submit
)
6174 struct btrfs_device
*dev
;
6175 struct bio
*first_bio
= bio
;
6176 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
6182 struct btrfs_bio
*bbio
= NULL
;
6184 length
= bio
->bi_iter
.bi_size
;
6185 map_length
= length
;
6187 btrfs_bio_counter_inc_blocked(fs_info
);
6188 ret
= __btrfs_map_block(fs_info
, btrfs_op(bio
), logical
,
6189 &map_length
, &bbio
, mirror_num
, 1);
6191 btrfs_bio_counter_dec(fs_info
);
6192 return errno_to_blk_status(ret
);
6195 total_devs
= bbio
->num_stripes
;
6196 bbio
->orig_bio
= first_bio
;
6197 bbio
->private = first_bio
->bi_private
;
6198 bbio
->end_io
= first_bio
->bi_end_io
;
6199 bbio
->fs_info
= fs_info
;
6200 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
6202 if ((bbio
->map_type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
6203 ((bio_op(bio
) == REQ_OP_WRITE
) || (mirror_num
> 1))) {
6204 /* In this case, map_length has been set to the length of
6205 a single stripe; not the whole write */
6206 if (bio_op(bio
) == REQ_OP_WRITE
) {
6207 ret
= raid56_parity_write(fs_info
, bio
, bbio
,
6210 ret
= raid56_parity_recover(fs_info
, bio
, bbio
,
6211 map_length
, mirror_num
, 1);
6214 btrfs_bio_counter_dec(fs_info
);
6215 return errno_to_blk_status(ret
);
6218 if (map_length
< length
) {
6220 "mapping failed logical %llu bio len %llu len %llu",
6221 logical
, length
, map_length
);
6225 for (dev_nr
= 0; dev_nr
< total_devs
; dev_nr
++) {
6226 dev
= bbio
->stripes
[dev_nr
].dev
;
6227 if (!dev
|| !dev
->bdev
||
6228 (bio_op(first_bio
) == REQ_OP_WRITE
&& !dev
->writeable
)) {
6229 bbio_error(bbio
, first_bio
, logical
);
6233 if (dev_nr
< total_devs
- 1)
6234 bio
= btrfs_bio_clone(first_bio
);
6238 submit_stripe_bio(bbio
, bio
, bbio
->stripes
[dev_nr
].physical
,
6239 dev_nr
, async_submit
);
6241 btrfs_bio_counter_dec(fs_info
);
6245 struct btrfs_device
*btrfs_find_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
6248 struct btrfs_device
*device
;
6249 struct btrfs_fs_devices
*cur_devices
;
6251 cur_devices
= fs_info
->fs_devices
;
6252 while (cur_devices
) {
6254 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_FSID_SIZE
)) {
6255 device
= find_device(cur_devices
, devid
, uuid
);
6259 cur_devices
= cur_devices
->seed
;
6264 static struct btrfs_device
*add_missing_dev(struct btrfs_fs_devices
*fs_devices
,
6265 u64 devid
, u8
*dev_uuid
)
6267 struct btrfs_device
*device
;
6269 device
= btrfs_alloc_device(NULL
, &devid
, dev_uuid
);
6273 list_add(&device
->dev_list
, &fs_devices
->devices
);
6274 device
->fs_devices
= fs_devices
;
6275 fs_devices
->num_devices
++;
6277 device
->missing
= 1;
6278 fs_devices
->missing_devices
++;
6284 * btrfs_alloc_device - allocate struct btrfs_device
6285 * @fs_info: used only for generating a new devid, can be NULL if
6286 * devid is provided (i.e. @devid != NULL).
6287 * @devid: a pointer to devid for this device. If NULL a new devid
6289 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6292 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6293 * on error. Returned struct is not linked onto any lists and can be
6294 * destroyed with kfree() right away.
6296 struct btrfs_device
*btrfs_alloc_device(struct btrfs_fs_info
*fs_info
,
6300 struct btrfs_device
*dev
;
6303 if (WARN_ON(!devid
&& !fs_info
))
6304 return ERR_PTR(-EINVAL
);
6306 dev
= __alloc_device();
6315 ret
= find_next_devid(fs_info
, &tmp
);
6317 bio_put(dev
->flush_bio
);
6319 return ERR_PTR(ret
);
6325 memcpy(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
);
6327 generate_random_uuid(dev
->uuid
);
6329 btrfs_init_work(&dev
->work
, btrfs_submit_helper
,
6330 pending_bios_fn
, NULL
, NULL
);
6335 /* Return -EIO if any error, otherwise return 0. */
6336 static int btrfs_check_chunk_valid(struct btrfs_fs_info
*fs_info
,
6337 struct extent_buffer
*leaf
,
6338 struct btrfs_chunk
*chunk
, u64 logical
)
6346 length
= btrfs_chunk_length(leaf
, chunk
);
6347 stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6348 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6349 sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6350 type
= btrfs_chunk_type(leaf
, chunk
);
6353 btrfs_err(fs_info
, "invalid chunk num_stripes: %u",
6357 if (!IS_ALIGNED(logical
, fs_info
->sectorsize
)) {
6358 btrfs_err(fs_info
, "invalid chunk logical %llu", logical
);
6361 if (btrfs_chunk_sector_size(leaf
, chunk
) != fs_info
->sectorsize
) {
6362 btrfs_err(fs_info
, "invalid chunk sectorsize %u",
6363 btrfs_chunk_sector_size(leaf
, chunk
));
6366 if (!length
|| !IS_ALIGNED(length
, fs_info
->sectorsize
)) {
6367 btrfs_err(fs_info
, "invalid chunk length %llu", length
);
6370 if (!is_power_of_2(stripe_len
) || stripe_len
!= BTRFS_STRIPE_LEN
) {
6371 btrfs_err(fs_info
, "invalid chunk stripe length: %llu",
6375 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK
| BTRFS_BLOCK_GROUP_PROFILE_MASK
) &
6377 btrfs_err(fs_info
, "unrecognized chunk type: %llu",
6378 ~(BTRFS_BLOCK_GROUP_TYPE_MASK
|
6379 BTRFS_BLOCK_GROUP_PROFILE_MASK
) &
6380 btrfs_chunk_type(leaf
, chunk
));
6383 if ((type
& BTRFS_BLOCK_GROUP_RAID10
&& sub_stripes
!= 2) ||
6384 (type
& BTRFS_BLOCK_GROUP_RAID1
&& num_stripes
< 1) ||
6385 (type
& BTRFS_BLOCK_GROUP_RAID5
&& num_stripes
< 2) ||
6386 (type
& BTRFS_BLOCK_GROUP_RAID6
&& num_stripes
< 3) ||
6387 (type
& BTRFS_BLOCK_GROUP_DUP
&& num_stripes
> 2) ||
6388 ((type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) == 0 &&
6389 num_stripes
!= 1)) {
6391 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6392 num_stripes
, sub_stripes
,
6393 type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
);
6400 static void btrfs_report_missing_device(struct btrfs_fs_info
*fs_info
,
6401 u64 devid
, u8
*uuid
, bool error
)
6404 btrfs_err_rl(fs_info
, "devid %llu uuid %pU is missing",
6407 btrfs_warn_rl(fs_info
, "devid %llu uuid %pU is missing",
6411 static int read_one_chunk(struct btrfs_fs_info
*fs_info
, struct btrfs_key
*key
,
6412 struct extent_buffer
*leaf
,
6413 struct btrfs_chunk
*chunk
)
6415 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
6416 struct map_lookup
*map
;
6417 struct extent_map
*em
;
6421 u8 uuid
[BTRFS_UUID_SIZE
];
6426 logical
= key
->offset
;
6427 length
= btrfs_chunk_length(leaf
, chunk
);
6428 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6430 ret
= btrfs_check_chunk_valid(fs_info
, leaf
, chunk
, logical
);
6434 read_lock(&map_tree
->map_tree
.lock
);
6435 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
6436 read_unlock(&map_tree
->map_tree
.lock
);
6438 /* already mapped? */
6439 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
6440 free_extent_map(em
);
6443 free_extent_map(em
);
6446 em
= alloc_extent_map();
6449 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
6451 free_extent_map(em
);
6455 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
6456 em
->map_lookup
= map
;
6457 em
->start
= logical
;
6460 em
->block_start
= 0;
6461 em
->block_len
= em
->len
;
6463 map
->num_stripes
= num_stripes
;
6464 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
6465 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
6466 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6467 map
->type
= btrfs_chunk_type(leaf
, chunk
);
6468 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6469 for (i
= 0; i
< num_stripes
; i
++) {
6470 map
->stripes
[i
].physical
=
6471 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
6472 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
6473 read_extent_buffer(leaf
, uuid
, (unsigned long)
6474 btrfs_stripe_dev_uuid_nr(chunk
, i
),
6476 map
->stripes
[i
].dev
= btrfs_find_device(fs_info
, devid
,
6478 if (!map
->stripes
[i
].dev
&&
6479 !btrfs_test_opt(fs_info
, DEGRADED
)) {
6480 free_extent_map(em
);
6481 btrfs_report_missing_device(fs_info
, devid
, uuid
, true);
6484 if (!map
->stripes
[i
].dev
) {
6485 map
->stripes
[i
].dev
=
6486 add_missing_dev(fs_info
->fs_devices
, devid
,
6488 if (IS_ERR(map
->stripes
[i
].dev
)) {
6489 free_extent_map(em
);
6491 "failed to init missing dev %llu: %ld",
6492 devid
, PTR_ERR(map
->stripes
[i
].dev
));
6493 return PTR_ERR(map
->stripes
[i
].dev
);
6495 btrfs_report_missing_device(fs_info
, devid
, uuid
, false);
6497 map
->stripes
[i
].dev
->in_fs_metadata
= 1;
6500 write_lock(&map_tree
->map_tree
.lock
);
6501 ret
= add_extent_mapping(&map_tree
->map_tree
, em
, 0);
6502 write_unlock(&map_tree
->map_tree
.lock
);
6503 BUG_ON(ret
); /* Tree corruption */
6504 free_extent_map(em
);
6509 static void fill_device_from_item(struct extent_buffer
*leaf
,
6510 struct btrfs_dev_item
*dev_item
,
6511 struct btrfs_device
*device
)
6515 device
->devid
= btrfs_device_id(leaf
, dev_item
);
6516 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
6517 device
->total_bytes
= device
->disk_total_bytes
;
6518 device
->commit_total_bytes
= device
->disk_total_bytes
;
6519 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
6520 device
->commit_bytes_used
= device
->bytes_used
;
6521 device
->type
= btrfs_device_type(leaf
, dev_item
);
6522 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
6523 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
6524 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
6525 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
6526 device
->is_tgtdev_for_dev_replace
= 0;
6528 ptr
= btrfs_device_uuid(dev_item
);
6529 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
6532 static struct btrfs_fs_devices
*open_seed_devices(struct btrfs_fs_info
*fs_info
,
6535 struct btrfs_fs_devices
*fs_devices
;
6538 BUG_ON(!mutex_is_locked(&uuid_mutex
));
6541 fs_devices
= fs_info
->fs_devices
->seed
;
6542 while (fs_devices
) {
6543 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_FSID_SIZE
))
6546 fs_devices
= fs_devices
->seed
;
6549 fs_devices
= find_fsid(fsid
);
6551 if (!btrfs_test_opt(fs_info
, DEGRADED
))
6552 return ERR_PTR(-ENOENT
);
6554 fs_devices
= alloc_fs_devices(fsid
);
6555 if (IS_ERR(fs_devices
))
6558 fs_devices
->seeding
= 1;
6559 fs_devices
->opened
= 1;
6563 fs_devices
= clone_fs_devices(fs_devices
);
6564 if (IS_ERR(fs_devices
))
6567 ret
= __btrfs_open_devices(fs_devices
, FMODE_READ
,
6568 fs_info
->bdev_holder
);
6570 free_fs_devices(fs_devices
);
6571 fs_devices
= ERR_PTR(ret
);
6575 if (!fs_devices
->seeding
) {
6576 __btrfs_close_devices(fs_devices
);
6577 free_fs_devices(fs_devices
);
6578 fs_devices
= ERR_PTR(-EINVAL
);
6582 fs_devices
->seed
= fs_info
->fs_devices
->seed
;
6583 fs_info
->fs_devices
->seed
= fs_devices
;
6588 static int read_one_dev(struct btrfs_fs_info
*fs_info
,
6589 struct extent_buffer
*leaf
,
6590 struct btrfs_dev_item
*dev_item
)
6592 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6593 struct btrfs_device
*device
;
6596 u8 fs_uuid
[BTRFS_FSID_SIZE
];
6597 u8 dev_uuid
[BTRFS_UUID_SIZE
];
6599 devid
= btrfs_device_id(leaf
, dev_item
);
6600 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
6602 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
6605 if (memcmp(fs_uuid
, fs_info
->fsid
, BTRFS_FSID_SIZE
)) {
6606 fs_devices
= open_seed_devices(fs_info
, fs_uuid
);
6607 if (IS_ERR(fs_devices
))
6608 return PTR_ERR(fs_devices
);
6611 device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, fs_uuid
);
6613 if (!btrfs_test_opt(fs_info
, DEGRADED
)) {
6614 btrfs_report_missing_device(fs_info
, devid
,
6619 device
= add_missing_dev(fs_devices
, devid
, dev_uuid
);
6620 if (IS_ERR(device
)) {
6622 "failed to add missing dev %llu: %ld",
6623 devid
, PTR_ERR(device
));
6624 return PTR_ERR(device
);
6626 btrfs_report_missing_device(fs_info
, devid
, dev_uuid
, false);
6628 if (!device
->bdev
) {
6629 if (!btrfs_test_opt(fs_info
, DEGRADED
)) {
6630 btrfs_report_missing_device(fs_info
,
6631 devid
, dev_uuid
, true);
6634 btrfs_report_missing_device(fs_info
, devid
,
6638 if(!device
->bdev
&& !device
->missing
) {
6640 * this happens when a device that was properly setup
6641 * in the device info lists suddenly goes bad.
6642 * device->bdev is NULL, and so we have to set
6643 * device->missing to one here
6645 device
->fs_devices
->missing_devices
++;
6646 device
->missing
= 1;
6649 /* Move the device to its own fs_devices */
6650 if (device
->fs_devices
!= fs_devices
) {
6651 ASSERT(device
->missing
);
6653 list_move(&device
->dev_list
, &fs_devices
->devices
);
6654 device
->fs_devices
->num_devices
--;
6655 fs_devices
->num_devices
++;
6657 device
->fs_devices
->missing_devices
--;
6658 fs_devices
->missing_devices
++;
6660 device
->fs_devices
= fs_devices
;
6664 if (device
->fs_devices
!= fs_info
->fs_devices
) {
6665 BUG_ON(device
->writeable
);
6666 if (device
->generation
!=
6667 btrfs_device_generation(leaf
, dev_item
))
6671 fill_device_from_item(leaf
, dev_item
, device
);
6672 device
->in_fs_metadata
= 1;
6673 if (device
->writeable
&& !device
->is_tgtdev_for_dev_replace
) {
6674 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
6675 atomic64_add(device
->total_bytes
- device
->bytes_used
,
6676 &fs_info
->free_chunk_space
);
6682 int btrfs_read_sys_array(struct btrfs_fs_info
*fs_info
)
6684 struct btrfs_root
*root
= fs_info
->tree_root
;
6685 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
6686 struct extent_buffer
*sb
;
6687 struct btrfs_disk_key
*disk_key
;
6688 struct btrfs_chunk
*chunk
;
6690 unsigned long sb_array_offset
;
6697 struct btrfs_key key
;
6699 ASSERT(BTRFS_SUPER_INFO_SIZE
<= fs_info
->nodesize
);
6701 * This will create extent buffer of nodesize, superblock size is
6702 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6703 * overallocate but we can keep it as-is, only the first page is used.
6705 sb
= btrfs_find_create_tree_block(fs_info
, BTRFS_SUPER_INFO_OFFSET
);
6708 set_extent_buffer_uptodate(sb
);
6709 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
6711 * The sb extent buffer is artificial and just used to read the system array.
6712 * set_extent_buffer_uptodate() call does not properly mark all it's
6713 * pages up-to-date when the page is larger: extent does not cover the
6714 * whole page and consequently check_page_uptodate does not find all
6715 * the page's extents up-to-date (the hole beyond sb),
6716 * write_extent_buffer then triggers a WARN_ON.
6718 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6719 * but sb spans only this function. Add an explicit SetPageUptodate call
6720 * to silence the warning eg. on PowerPC 64.
6722 if (PAGE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
6723 SetPageUptodate(sb
->pages
[0]);
6725 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
6726 array_size
= btrfs_super_sys_array_size(super_copy
);
6728 array_ptr
= super_copy
->sys_chunk_array
;
6729 sb_array_offset
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
6732 while (cur_offset
< array_size
) {
6733 disk_key
= (struct btrfs_disk_key
*)array_ptr
;
6734 len
= sizeof(*disk_key
);
6735 if (cur_offset
+ len
> array_size
)
6736 goto out_short_read
;
6738 btrfs_disk_key_to_cpu(&key
, disk_key
);
6741 sb_array_offset
+= len
;
6744 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6745 chunk
= (struct btrfs_chunk
*)sb_array_offset
;
6747 * At least one btrfs_chunk with one stripe must be
6748 * present, exact stripe count check comes afterwards
6750 len
= btrfs_chunk_item_size(1);
6751 if (cur_offset
+ len
> array_size
)
6752 goto out_short_read
;
6754 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
6757 "invalid number of stripes %u in sys_array at offset %u",
6758 num_stripes
, cur_offset
);
6763 type
= btrfs_chunk_type(sb
, chunk
);
6764 if ((type
& BTRFS_BLOCK_GROUP_SYSTEM
) == 0) {
6766 "invalid chunk type %llu in sys_array at offset %u",
6772 len
= btrfs_chunk_item_size(num_stripes
);
6773 if (cur_offset
+ len
> array_size
)
6774 goto out_short_read
;
6776 ret
= read_one_chunk(fs_info
, &key
, sb
, chunk
);
6781 "unexpected item type %u in sys_array at offset %u",
6782 (u32
)key
.type
, cur_offset
);
6787 sb_array_offset
+= len
;
6790 clear_extent_buffer_uptodate(sb
);
6791 free_extent_buffer_stale(sb
);
6795 btrfs_err(fs_info
, "sys_array too short to read %u bytes at offset %u",
6797 clear_extent_buffer_uptodate(sb
);
6798 free_extent_buffer_stale(sb
);
6803 * Check if all chunks in the fs are OK for read-write degraded mount
6805 * Return true if all chunks meet the minimal RW mount requirements.
6806 * Return false if any chunk doesn't meet the minimal RW mount requirements.
6808 bool btrfs_check_rw_degradable(struct btrfs_fs_info
*fs_info
)
6810 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
6811 struct extent_map
*em
;
6815 read_lock(&map_tree
->map_tree
.lock
);
6816 em
= lookup_extent_mapping(&map_tree
->map_tree
, 0, (u64
)-1);
6817 read_unlock(&map_tree
->map_tree
.lock
);
6818 /* No chunk at all? Return false anyway */
6824 struct map_lookup
*map
;
6829 map
= em
->map_lookup
;
6831 btrfs_get_num_tolerated_disk_barrier_failures(
6833 for (i
= 0; i
< map
->num_stripes
; i
++) {
6834 struct btrfs_device
*dev
= map
->stripes
[i
].dev
;
6836 if (!dev
|| !dev
->bdev
|| dev
->missing
||
6837 dev
->last_flush_error
)
6840 if (missing
> max_tolerated
) {
6842 "chunk %llu missing %d devices, max tolerance is %d for writeable mount",
6843 em
->start
, missing
, max_tolerated
);
6844 free_extent_map(em
);
6848 next_start
= extent_map_end(em
);
6849 free_extent_map(em
);
6851 read_lock(&map_tree
->map_tree
.lock
);
6852 em
= lookup_extent_mapping(&map_tree
->map_tree
, next_start
,
6853 (u64
)(-1) - next_start
);
6854 read_unlock(&map_tree
->map_tree
.lock
);
6860 int btrfs_read_chunk_tree(struct btrfs_fs_info
*fs_info
)
6862 struct btrfs_root
*root
= fs_info
->chunk_root
;
6863 struct btrfs_path
*path
;
6864 struct extent_buffer
*leaf
;
6865 struct btrfs_key key
;
6866 struct btrfs_key found_key
;
6871 path
= btrfs_alloc_path();
6875 mutex_lock(&uuid_mutex
);
6876 mutex_lock(&fs_info
->chunk_mutex
);
6879 * Read all device items, and then all the chunk items. All
6880 * device items are found before any chunk item (their object id
6881 * is smaller than the lowest possible object id for a chunk
6882 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6884 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
6887 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6891 leaf
= path
->nodes
[0];
6892 slot
= path
->slots
[0];
6893 if (slot
>= btrfs_header_nritems(leaf
)) {
6894 ret
= btrfs_next_leaf(root
, path
);
6901 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6902 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
6903 struct btrfs_dev_item
*dev_item
;
6904 dev_item
= btrfs_item_ptr(leaf
, slot
,
6905 struct btrfs_dev_item
);
6906 ret
= read_one_dev(fs_info
, leaf
, dev_item
);
6910 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6911 struct btrfs_chunk
*chunk
;
6912 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
6913 ret
= read_one_chunk(fs_info
, &found_key
, leaf
, chunk
);
6921 * After loading chunk tree, we've got all device information,
6922 * do another round of validation checks.
6924 if (total_dev
!= fs_info
->fs_devices
->total_devices
) {
6926 "super_num_devices %llu mismatch with num_devices %llu found here",
6927 btrfs_super_num_devices(fs_info
->super_copy
),
6932 if (btrfs_super_total_bytes(fs_info
->super_copy
) <
6933 fs_info
->fs_devices
->total_rw_bytes
) {
6935 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6936 btrfs_super_total_bytes(fs_info
->super_copy
),
6937 fs_info
->fs_devices
->total_rw_bytes
);
6943 mutex_unlock(&fs_info
->chunk_mutex
);
6944 mutex_unlock(&uuid_mutex
);
6946 btrfs_free_path(path
);
6950 void btrfs_init_devices_late(struct btrfs_fs_info
*fs_info
)
6952 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6953 struct btrfs_device
*device
;
6955 while (fs_devices
) {
6956 mutex_lock(&fs_devices
->device_list_mutex
);
6957 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
)
6958 device
->fs_info
= fs_info
;
6959 mutex_unlock(&fs_devices
->device_list_mutex
);
6961 fs_devices
= fs_devices
->seed
;
6965 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
6969 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
6970 btrfs_dev_stat_reset(dev
, i
);
6973 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
6975 struct btrfs_key key
;
6976 struct btrfs_key found_key
;
6977 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
6978 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6979 struct extent_buffer
*eb
;
6982 struct btrfs_device
*device
;
6983 struct btrfs_path
*path
= NULL
;
6986 path
= btrfs_alloc_path();
6992 mutex_lock(&fs_devices
->device_list_mutex
);
6993 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6995 struct btrfs_dev_stats_item
*ptr
;
6997 key
.objectid
= BTRFS_DEV_STATS_OBJECTID
;
6998 key
.type
= BTRFS_PERSISTENT_ITEM_KEY
;
6999 key
.offset
= device
->devid
;
7000 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
7002 __btrfs_reset_dev_stats(device
);
7003 device
->dev_stats_valid
= 1;
7004 btrfs_release_path(path
);
7007 slot
= path
->slots
[0];
7008 eb
= path
->nodes
[0];
7009 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
7010 item_size
= btrfs_item_size_nr(eb
, slot
);
7012 ptr
= btrfs_item_ptr(eb
, slot
,
7013 struct btrfs_dev_stats_item
);
7015 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
7016 if (item_size
>= (1 + i
) * sizeof(__le64
))
7017 btrfs_dev_stat_set(device
, i
,
7018 btrfs_dev_stats_value(eb
, ptr
, i
));
7020 btrfs_dev_stat_reset(device
, i
);
7023 device
->dev_stats_valid
= 1;
7024 btrfs_dev_stat_print_on_load(device
);
7025 btrfs_release_path(path
);
7027 mutex_unlock(&fs_devices
->device_list_mutex
);
7030 btrfs_free_path(path
);
7031 return ret
< 0 ? ret
: 0;
7034 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
7035 struct btrfs_fs_info
*fs_info
,
7036 struct btrfs_device
*device
)
7038 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
7039 struct btrfs_path
*path
;
7040 struct btrfs_key key
;
7041 struct extent_buffer
*eb
;
7042 struct btrfs_dev_stats_item
*ptr
;
7046 key
.objectid
= BTRFS_DEV_STATS_OBJECTID
;
7047 key
.type
= BTRFS_PERSISTENT_ITEM_KEY
;
7048 key
.offset
= device
->devid
;
7050 path
= btrfs_alloc_path();
7053 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
7055 btrfs_warn_in_rcu(fs_info
,
7056 "error %d while searching for dev_stats item for device %s",
7057 ret
, rcu_str_deref(device
->name
));
7062 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
7063 /* need to delete old one and insert a new one */
7064 ret
= btrfs_del_item(trans
, dev_root
, path
);
7066 btrfs_warn_in_rcu(fs_info
,
7067 "delete too small dev_stats item for device %s failed %d",
7068 rcu_str_deref(device
->name
), ret
);
7075 /* need to insert a new item */
7076 btrfs_release_path(path
);
7077 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
7078 &key
, sizeof(*ptr
));
7080 btrfs_warn_in_rcu(fs_info
,
7081 "insert dev_stats item for device %s failed %d",
7082 rcu_str_deref(device
->name
), ret
);
7087 eb
= path
->nodes
[0];
7088 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
7089 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7090 btrfs_set_dev_stats_value(eb
, ptr
, i
,
7091 btrfs_dev_stat_read(device
, i
));
7092 btrfs_mark_buffer_dirty(eb
);
7095 btrfs_free_path(path
);
7100 * called from commit_transaction. Writes all changed device stats to disk.
7102 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
7103 struct btrfs_fs_info
*fs_info
)
7105 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7106 struct btrfs_device
*device
;
7110 mutex_lock(&fs_devices
->device_list_mutex
);
7111 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
7112 stats_cnt
= atomic_read(&device
->dev_stats_ccnt
);
7113 if (!device
->dev_stats_valid
|| stats_cnt
== 0)
7118 * There is a LOAD-LOAD control dependency between the value of
7119 * dev_stats_ccnt and updating the on-disk values which requires
7120 * reading the in-memory counters. Such control dependencies
7121 * require explicit read memory barriers.
7123 * This memory barriers pairs with smp_mb__before_atomic in
7124 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7125 * barrier implied by atomic_xchg in
7126 * btrfs_dev_stats_read_and_reset
7130 ret
= update_dev_stat_item(trans
, fs_info
, device
);
7132 atomic_sub(stats_cnt
, &device
->dev_stats_ccnt
);
7134 mutex_unlock(&fs_devices
->device_list_mutex
);
7139 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
7141 btrfs_dev_stat_inc(dev
, index
);
7142 btrfs_dev_stat_print_on_error(dev
);
7145 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
7147 if (!dev
->dev_stats_valid
)
7149 btrfs_err_rl_in_rcu(dev
->fs_info
,
7150 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7151 rcu_str_deref(dev
->name
),
7152 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
7153 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
7154 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
7155 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
7156 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
7159 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
7163 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7164 if (btrfs_dev_stat_read(dev
, i
) != 0)
7166 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
7167 return; /* all values == 0, suppress message */
7169 btrfs_info_in_rcu(dev
->fs_info
,
7170 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7171 rcu_str_deref(dev
->name
),
7172 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
7173 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
7174 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
7175 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
7176 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
7179 int btrfs_get_dev_stats(struct btrfs_fs_info
*fs_info
,
7180 struct btrfs_ioctl_get_dev_stats
*stats
)
7182 struct btrfs_device
*dev
;
7183 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7186 mutex_lock(&fs_devices
->device_list_mutex
);
7187 dev
= btrfs_find_device(fs_info
, stats
->devid
, NULL
, NULL
);
7188 mutex_unlock(&fs_devices
->device_list_mutex
);
7191 btrfs_warn(fs_info
, "get dev_stats failed, device not found");
7193 } else if (!dev
->dev_stats_valid
) {
7194 btrfs_warn(fs_info
, "get dev_stats failed, not yet valid");
7196 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
7197 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
7198 if (stats
->nr_items
> i
)
7200 btrfs_dev_stat_read_and_reset(dev
, i
);
7202 btrfs_dev_stat_reset(dev
, i
);
7205 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7206 if (stats
->nr_items
> i
)
7207 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
7209 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
7210 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
7214 void btrfs_scratch_superblocks(struct block_device
*bdev
, const char *device_path
)
7216 struct buffer_head
*bh
;
7217 struct btrfs_super_block
*disk_super
;
7223 for (copy_num
= 0; copy_num
< BTRFS_SUPER_MIRROR_MAX
;
7226 if (btrfs_read_dev_one_super(bdev
, copy_num
, &bh
))
7229 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
7231 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
7232 set_buffer_dirty(bh
);
7233 sync_dirty_buffer(bh
);
7237 /* Notify udev that device has changed */
7238 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
7240 /* Update ctime/mtime for device path for libblkid */
7241 update_dev_time(device_path
);
7245 * Update the size of all devices, which is used for writing out the
7248 void btrfs_update_commit_device_size(struct btrfs_fs_info
*fs_info
)
7250 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7251 struct btrfs_device
*curr
, *next
;
7253 if (list_empty(&fs_devices
->resized_devices
))
7256 mutex_lock(&fs_devices
->device_list_mutex
);
7257 mutex_lock(&fs_info
->chunk_mutex
);
7258 list_for_each_entry_safe(curr
, next
, &fs_devices
->resized_devices
,
7260 list_del_init(&curr
->resized_list
);
7261 curr
->commit_total_bytes
= curr
->disk_total_bytes
;
7263 mutex_unlock(&fs_info
->chunk_mutex
);
7264 mutex_unlock(&fs_devices
->device_list_mutex
);
7267 /* Must be invoked during the transaction commit */
7268 void btrfs_update_commit_device_bytes_used(struct btrfs_fs_info
*fs_info
,
7269 struct btrfs_transaction
*transaction
)
7271 struct extent_map
*em
;
7272 struct map_lookup
*map
;
7273 struct btrfs_device
*dev
;
7276 if (list_empty(&transaction
->pending_chunks
))
7279 /* In order to kick the device replace finish process */
7280 mutex_lock(&fs_info
->chunk_mutex
);
7281 list_for_each_entry(em
, &transaction
->pending_chunks
, list
) {
7282 map
= em
->map_lookup
;
7284 for (i
= 0; i
< map
->num_stripes
; i
++) {
7285 dev
= map
->stripes
[i
].dev
;
7286 dev
->commit_bytes_used
= dev
->bytes_used
;
7289 mutex_unlock(&fs_info
->chunk_mutex
);
7292 void btrfs_set_fs_info_ptr(struct btrfs_fs_info
*fs_info
)
7294 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7295 while (fs_devices
) {
7296 fs_devices
->fs_info
= fs_info
;
7297 fs_devices
= fs_devices
->seed
;
7301 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info
*fs_info
)
7303 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7304 while (fs_devices
) {
7305 fs_devices
->fs_info
= NULL
;
7306 fs_devices
= fs_devices
->seed
;