1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
8 #include <linux/slab.h>
9 #include <linux/buffer_head.h>
10 #include <linux/blkdev.h>
11 #include <linux/ratelimit.h>
12 #include <linux/kthread.h>
13 #include <linux/raid/pq.h>
14 #include <linux/semaphore.h>
15 #include <linux/uuid.h>
16 #include <linux/list_sort.h>
18 #include "extent_map.h"
20 #include "transaction.h"
21 #include "print-tree.h"
24 #include "async-thread.h"
25 #include "check-integrity.h"
26 #include "rcu-string.h"
28 #include "dev-replace.h"
31 const struct btrfs_raid_attr btrfs_raid_array
[BTRFS_NR_RAID_TYPES
] = {
32 [BTRFS_RAID_RAID10
] = {
35 .devs_max
= 0, /* 0 == as many as possible */
37 .tolerated_failures
= 1,
40 .raid_name
= "raid10",
41 .bg_flag
= BTRFS_BLOCK_GROUP_RAID10
,
42 .mindev_error
= BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET
,
44 [BTRFS_RAID_RAID1
] = {
49 .tolerated_failures
= 1,
53 .bg_flag
= BTRFS_BLOCK_GROUP_RAID1
,
54 .mindev_error
= BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET
,
61 .tolerated_failures
= 0,
65 .bg_flag
= BTRFS_BLOCK_GROUP_DUP
,
68 [BTRFS_RAID_RAID0
] = {
73 .tolerated_failures
= 0,
77 .bg_flag
= BTRFS_BLOCK_GROUP_RAID0
,
80 [BTRFS_RAID_SINGLE
] = {
85 .tolerated_failures
= 0,
88 .raid_name
= "single",
92 [BTRFS_RAID_RAID5
] = {
97 .tolerated_failures
= 1,
100 .raid_name
= "raid5",
101 .bg_flag
= BTRFS_BLOCK_GROUP_RAID5
,
102 .mindev_error
= BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET
,
104 [BTRFS_RAID_RAID6
] = {
109 .tolerated_failures
= 2,
112 .raid_name
= "raid6",
113 .bg_flag
= BTRFS_BLOCK_GROUP_RAID6
,
114 .mindev_error
= BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET
,
118 const char *get_raid_name(enum btrfs_raid_types type
)
120 if (type
>= BTRFS_NR_RAID_TYPES
)
123 return btrfs_raid_array
[type
].raid_name
;
126 static int init_first_rw_device(struct btrfs_trans_handle
*trans
,
127 struct btrfs_fs_info
*fs_info
);
128 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info
*fs_info
);
129 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
);
130 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
);
131 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*device
);
132 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
,
133 enum btrfs_map_op op
,
134 u64 logical
, u64
*length
,
135 struct btrfs_bio
**bbio_ret
,
136 int mirror_num
, int need_raid_map
);
142 * There are several mutexes that protect manipulation of devices and low-level
143 * structures like chunks but not block groups, extents or files
145 * uuid_mutex (global lock)
146 * ------------------------
147 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
148 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
149 * device) or requested by the device= mount option
151 * the mutex can be very coarse and can cover long-running operations
153 * protects: updates to fs_devices counters like missing devices, rw devices,
154 * seeding, structure cloning, openning/closing devices at mount/umount time
156 * global::fs_devs - add, remove, updates to the global list
158 * does not protect: manipulation of the fs_devices::devices list!
160 * btrfs_device::name - renames (write side), read is RCU
162 * fs_devices::device_list_mutex (per-fs, with RCU)
163 * ------------------------------------------------
164 * protects updates to fs_devices::devices, ie. adding and deleting
166 * simple list traversal with read-only actions can be done with RCU protection
168 * may be used to exclude some operations from running concurrently without any
169 * modifications to the list (see write_all_supers)
173 * protects balance structures (status, state) and context accessed from
174 * several places (internally, ioctl)
178 * protects chunks, adding or removing during allocation, trim or when a new
179 * device is added/removed
183 * a big lock that is held by the cleaner thread and prevents running subvolume
184 * cleaning together with relocation or delayed iputs
197 * Exclusive operations, BTRFS_FS_EXCL_OP
198 * ======================================
200 * Maintains the exclusivity of the following operations that apply to the
201 * whole filesystem and cannot run in parallel.
206 * - Device replace (*)
209 * The device operations (as above) can be in one of the following states:
215 * Only device operations marked with (*) can go into the Paused state for the
218 * - ioctl (only Balance can be Paused through ioctl)
219 * - filesystem remounted as read-only
220 * - filesystem unmounted and mounted as read-only
221 * - system power-cycle and filesystem mounted as read-only
222 * - filesystem or device errors leading to forced read-only
224 * BTRFS_FS_EXCL_OP flag is set and cleared using atomic operations.
225 * During the course of Paused state, the BTRFS_FS_EXCL_OP remains set.
226 * A device operation in Paused or Running state can be canceled or resumed
227 * either by ioctl (Balance only) or when remounted as read-write.
228 * BTRFS_FS_EXCL_OP flag is cleared when the device operation is canceled or
232 DEFINE_MUTEX(uuid_mutex
);
233 static LIST_HEAD(fs_uuids
);
234 struct list_head
*btrfs_get_fs_uuids(void)
240 * alloc_fs_devices - allocate struct btrfs_fs_devices
241 * @fsid: if not NULL, copy the uuid to fs_devices::fsid
243 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
244 * The returned struct is not linked onto any lists and can be destroyed with
245 * kfree() right away.
247 static struct btrfs_fs_devices
*alloc_fs_devices(const u8
*fsid
)
249 struct btrfs_fs_devices
*fs_devs
;
251 fs_devs
= kzalloc(sizeof(*fs_devs
), GFP_KERNEL
);
253 return ERR_PTR(-ENOMEM
);
255 mutex_init(&fs_devs
->device_list_mutex
);
257 INIT_LIST_HEAD(&fs_devs
->devices
);
258 INIT_LIST_HEAD(&fs_devs
->resized_devices
);
259 INIT_LIST_HEAD(&fs_devs
->alloc_list
);
260 INIT_LIST_HEAD(&fs_devs
->fs_list
);
262 memcpy(fs_devs
->fsid
, fsid
, BTRFS_FSID_SIZE
);
267 void btrfs_free_device(struct btrfs_device
*device
)
269 rcu_string_free(device
->name
);
270 bio_put(device
->flush_bio
);
274 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
276 struct btrfs_device
*device
;
277 WARN_ON(fs_devices
->opened
);
278 while (!list_empty(&fs_devices
->devices
)) {
279 device
= list_entry(fs_devices
->devices
.next
,
280 struct btrfs_device
, dev_list
);
281 list_del(&device
->dev_list
);
282 btrfs_free_device(device
);
287 static void btrfs_kobject_uevent(struct block_device
*bdev
,
288 enum kobject_action action
)
292 ret
= kobject_uevent(&disk_to_dev(bdev
->bd_disk
)->kobj
, action
);
294 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
296 kobject_name(&disk_to_dev(bdev
->bd_disk
)->kobj
),
297 &disk_to_dev(bdev
->bd_disk
)->kobj
);
300 void __exit
btrfs_cleanup_fs_uuids(void)
302 struct btrfs_fs_devices
*fs_devices
;
304 while (!list_empty(&fs_uuids
)) {
305 fs_devices
= list_entry(fs_uuids
.next
,
306 struct btrfs_fs_devices
, fs_list
);
307 list_del(&fs_devices
->fs_list
);
308 free_fs_devices(fs_devices
);
313 * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
314 * Returned struct is not linked onto any lists and must be destroyed using
317 static struct btrfs_device
*__alloc_device(void)
319 struct btrfs_device
*dev
;
321 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
323 return ERR_PTR(-ENOMEM
);
326 * Preallocate a bio that's always going to be used for flushing device
327 * barriers and matches the device lifespan
329 dev
->flush_bio
= bio_alloc_bioset(GFP_KERNEL
, 0, NULL
);
330 if (!dev
->flush_bio
) {
332 return ERR_PTR(-ENOMEM
);
335 INIT_LIST_HEAD(&dev
->dev_list
);
336 INIT_LIST_HEAD(&dev
->dev_alloc_list
);
337 INIT_LIST_HEAD(&dev
->resized_list
);
339 spin_lock_init(&dev
->io_lock
);
341 atomic_set(&dev
->reada_in_flight
, 0);
342 atomic_set(&dev
->dev_stats_ccnt
, 0);
343 btrfs_device_data_ordered_init(dev
);
344 INIT_RADIX_TREE(&dev
->reada_zones
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
345 INIT_RADIX_TREE(&dev
->reada_extents
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
351 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
354 * If devid and uuid are both specified, the match must be exact, otherwise
355 * only devid is used.
357 static struct btrfs_device
*find_device(struct btrfs_fs_devices
*fs_devices
,
358 u64 devid
, const u8
*uuid
)
360 struct btrfs_device
*dev
;
362 list_for_each_entry(dev
, &fs_devices
->devices
, dev_list
) {
363 if (dev
->devid
== devid
&&
364 (!uuid
|| !memcmp(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
))) {
371 static noinline
struct btrfs_fs_devices
*find_fsid(u8
*fsid
)
373 struct btrfs_fs_devices
*fs_devices
;
375 list_for_each_entry(fs_devices
, &fs_uuids
, fs_list
) {
376 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
383 btrfs_get_bdev_and_sb(const char *device_path
, fmode_t flags
, void *holder
,
384 int flush
, struct block_device
**bdev
,
385 struct buffer_head
**bh
)
389 *bdev
= blkdev_get_by_path(device_path
, flags
, holder
);
392 ret
= PTR_ERR(*bdev
);
397 filemap_write_and_wait((*bdev
)->bd_inode
->i_mapping
);
398 ret
= set_blocksize(*bdev
, BTRFS_BDEV_BLOCKSIZE
);
400 blkdev_put(*bdev
, flags
);
403 invalidate_bdev(*bdev
);
404 *bh
= btrfs_read_dev_super(*bdev
);
407 blkdev_put(*bdev
, flags
);
419 static void requeue_list(struct btrfs_pending_bios
*pending_bios
,
420 struct bio
*head
, struct bio
*tail
)
423 struct bio
*old_head
;
425 old_head
= pending_bios
->head
;
426 pending_bios
->head
= head
;
427 if (pending_bios
->tail
)
428 tail
->bi_next
= old_head
;
430 pending_bios
->tail
= tail
;
434 * we try to collect pending bios for a device so we don't get a large
435 * number of procs sending bios down to the same device. This greatly
436 * improves the schedulers ability to collect and merge the bios.
438 * But, it also turns into a long list of bios to process and that is sure
439 * to eventually make the worker thread block. The solution here is to
440 * make some progress and then put this work struct back at the end of
441 * the list if the block device is congested. This way, multiple devices
442 * can make progress from a single worker thread.
444 static noinline
void run_scheduled_bios(struct btrfs_device
*device
)
446 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
448 struct backing_dev_info
*bdi
;
449 struct btrfs_pending_bios
*pending_bios
;
453 unsigned long num_run
;
454 unsigned long batch_run
= 0;
455 unsigned long last_waited
= 0;
457 int sync_pending
= 0;
458 struct blk_plug plug
;
461 * this function runs all the bios we've collected for
462 * a particular device. We don't want to wander off to
463 * another device without first sending all of these down.
464 * So, setup a plug here and finish it off before we return
466 blk_start_plug(&plug
);
468 bdi
= device
->bdev
->bd_bdi
;
471 spin_lock(&device
->io_lock
);
476 /* take all the bios off the list at once and process them
477 * later on (without the lock held). But, remember the
478 * tail and other pointers so the bios can be properly reinserted
479 * into the list if we hit congestion
481 if (!force_reg
&& device
->pending_sync_bios
.head
) {
482 pending_bios
= &device
->pending_sync_bios
;
485 pending_bios
= &device
->pending_bios
;
489 pending
= pending_bios
->head
;
490 tail
= pending_bios
->tail
;
491 WARN_ON(pending
&& !tail
);
494 * if pending was null this time around, no bios need processing
495 * at all and we can stop. Otherwise it'll loop back up again
496 * and do an additional check so no bios are missed.
498 * device->running_pending is used to synchronize with the
501 if (device
->pending_sync_bios
.head
== NULL
&&
502 device
->pending_bios
.head
== NULL
) {
504 device
->running_pending
= 0;
507 device
->running_pending
= 1;
510 pending_bios
->head
= NULL
;
511 pending_bios
->tail
= NULL
;
513 spin_unlock(&device
->io_lock
);
518 /* we want to work on both lists, but do more bios on the
519 * sync list than the regular list
522 pending_bios
!= &device
->pending_sync_bios
&&
523 device
->pending_sync_bios
.head
) ||
524 (num_run
> 64 && pending_bios
== &device
->pending_sync_bios
&&
525 device
->pending_bios
.head
)) {
526 spin_lock(&device
->io_lock
);
527 requeue_list(pending_bios
, pending
, tail
);
532 pending
= pending
->bi_next
;
535 BUG_ON(atomic_read(&cur
->__bi_cnt
) == 0);
538 * if we're doing the sync list, record that our
539 * plug has some sync requests on it
541 * If we're doing the regular list and there are
542 * sync requests sitting around, unplug before
545 if (pending_bios
== &device
->pending_sync_bios
) {
547 } else if (sync_pending
) {
548 blk_finish_plug(&plug
);
549 blk_start_plug(&plug
);
553 btrfsic_submit_bio(cur
);
560 * we made progress, there is more work to do and the bdi
561 * is now congested. Back off and let other work structs
564 if (pending
&& bdi_write_congested(bdi
) && batch_run
> 8 &&
565 fs_info
->fs_devices
->open_devices
> 1) {
566 struct io_context
*ioc
;
568 ioc
= current
->io_context
;
571 * the main goal here is that we don't want to
572 * block if we're going to be able to submit
573 * more requests without blocking.
575 * This code does two great things, it pokes into
576 * the elevator code from a filesystem _and_
577 * it makes assumptions about how batching works.
579 if (ioc
&& ioc
->nr_batch_requests
> 0 &&
580 time_before(jiffies
, ioc
->last_waited
+ HZ
/50UL) &&
582 ioc
->last_waited
== last_waited
)) {
584 * we want to go through our batch of
585 * requests and stop. So, we copy out
586 * the ioc->last_waited time and test
587 * against it before looping
589 last_waited
= ioc
->last_waited
;
593 spin_lock(&device
->io_lock
);
594 requeue_list(pending_bios
, pending
, tail
);
595 device
->running_pending
= 1;
597 spin_unlock(&device
->io_lock
);
598 btrfs_queue_work(fs_info
->submit_workers
,
608 spin_lock(&device
->io_lock
);
609 if (device
->pending_bios
.head
|| device
->pending_sync_bios
.head
)
611 spin_unlock(&device
->io_lock
);
614 blk_finish_plug(&plug
);
617 static void pending_bios_fn(struct btrfs_work
*work
)
619 struct btrfs_device
*device
;
621 device
= container_of(work
, struct btrfs_device
, work
);
622 run_scheduled_bios(device
);
626 * Search and remove all stale (devices which are not mounted) devices.
627 * When both inputs are NULL, it will search and release all stale devices.
628 * path: Optional. When provided will it release all unmounted devices
629 * matching this path only.
630 * skip_dev: Optional. Will skip this device when searching for the stale
633 static void btrfs_free_stale_devices(const char *path
,
634 struct btrfs_device
*skip_device
)
636 struct btrfs_fs_devices
*fs_devices
, *tmp_fs_devices
;
637 struct btrfs_device
*device
, *tmp_device
;
639 list_for_each_entry_safe(fs_devices
, tmp_fs_devices
, &fs_uuids
, fs_list
) {
640 mutex_lock(&fs_devices
->device_list_mutex
);
641 if (fs_devices
->opened
) {
642 mutex_unlock(&fs_devices
->device_list_mutex
);
646 list_for_each_entry_safe(device
, tmp_device
,
647 &fs_devices
->devices
, dev_list
) {
650 if (skip_device
&& skip_device
== device
)
652 if (path
&& !device
->name
)
657 not_found
= strcmp(rcu_str_deref(device
->name
),
663 /* delete the stale device */
664 fs_devices
->num_devices
--;
665 list_del(&device
->dev_list
);
666 btrfs_free_device(device
);
668 if (fs_devices
->num_devices
== 0)
671 mutex_unlock(&fs_devices
->device_list_mutex
);
672 if (fs_devices
->num_devices
== 0) {
673 btrfs_sysfs_remove_fsid(fs_devices
);
674 list_del(&fs_devices
->fs_list
);
675 free_fs_devices(fs_devices
);
680 static int btrfs_open_one_device(struct btrfs_fs_devices
*fs_devices
,
681 struct btrfs_device
*device
, fmode_t flags
,
684 struct request_queue
*q
;
685 struct block_device
*bdev
;
686 struct buffer_head
*bh
;
687 struct btrfs_super_block
*disk_super
;
696 ret
= btrfs_get_bdev_and_sb(device
->name
->str
, flags
, holder
, 1,
701 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
702 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
703 if (devid
!= device
->devid
)
706 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
, BTRFS_UUID_SIZE
))
709 device
->generation
= btrfs_super_generation(disk_super
);
711 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
712 clear_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
);
713 fs_devices
->seeding
= 1;
715 if (bdev_read_only(bdev
))
716 clear_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
);
718 set_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
);
721 q
= bdev_get_queue(bdev
);
722 if (!blk_queue_nonrot(q
))
723 fs_devices
->rotating
= 1;
726 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &device
->dev_state
);
727 device
->mode
= flags
;
729 fs_devices
->open_devices
++;
730 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
) &&
731 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
732 fs_devices
->rw_devices
++;
733 list_add_tail(&device
->dev_alloc_list
, &fs_devices
->alloc_list
);
741 blkdev_put(bdev
, flags
);
747 * Add new device to list of registered devices
750 * device pointer which was just added or updated when successful
751 * error pointer when failed
753 static noinline
struct btrfs_device
*device_list_add(const char *path
,
754 struct btrfs_super_block
*disk_super
,
755 bool *new_device_added
)
757 struct btrfs_device
*device
;
758 struct btrfs_fs_devices
*fs_devices
;
759 struct rcu_string
*name
;
760 u64 found_transid
= btrfs_super_generation(disk_super
);
761 u64 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
763 fs_devices
= find_fsid(disk_super
->fsid
);
765 fs_devices
= alloc_fs_devices(disk_super
->fsid
);
766 if (IS_ERR(fs_devices
))
767 return ERR_CAST(fs_devices
);
769 mutex_lock(&fs_devices
->device_list_mutex
);
770 list_add(&fs_devices
->fs_list
, &fs_uuids
);
774 mutex_lock(&fs_devices
->device_list_mutex
);
775 device
= find_device(fs_devices
, devid
,
776 disk_super
->dev_item
.uuid
);
780 if (fs_devices
->opened
) {
781 mutex_unlock(&fs_devices
->device_list_mutex
);
782 return ERR_PTR(-EBUSY
);
785 device
= btrfs_alloc_device(NULL
, &devid
,
786 disk_super
->dev_item
.uuid
);
787 if (IS_ERR(device
)) {
788 mutex_unlock(&fs_devices
->device_list_mutex
);
789 /* we can safely leave the fs_devices entry around */
793 name
= rcu_string_strdup(path
, GFP_NOFS
);
795 btrfs_free_device(device
);
796 mutex_unlock(&fs_devices
->device_list_mutex
);
797 return ERR_PTR(-ENOMEM
);
799 rcu_assign_pointer(device
->name
, name
);
801 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
802 fs_devices
->num_devices
++;
804 device
->fs_devices
= fs_devices
;
805 *new_device_added
= true;
807 if (disk_super
->label
[0])
808 pr_info("BTRFS: device label %s devid %llu transid %llu %s\n",
809 disk_super
->label
, devid
, found_transid
, path
);
811 pr_info("BTRFS: device fsid %pU devid %llu transid %llu %s\n",
812 disk_super
->fsid
, devid
, found_transid
, path
);
814 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
816 * When FS is already mounted.
817 * 1. If you are here and if the device->name is NULL that
818 * means this device was missing at time of FS mount.
819 * 2. If you are here and if the device->name is different
820 * from 'path' that means either
821 * a. The same device disappeared and reappeared with
823 * b. The missing-disk-which-was-replaced, has
826 * We must allow 1 and 2a above. But 2b would be a spurious
829 * Further in case of 1 and 2a above, the disk at 'path'
830 * would have missed some transaction when it was away and
831 * in case of 2a the stale bdev has to be updated as well.
832 * 2b must not be allowed at all time.
836 * For now, we do allow update to btrfs_fs_device through the
837 * btrfs dev scan cli after FS has been mounted. We're still
838 * tracking a problem where systems fail mount by subvolume id
839 * when we reject replacement on a mounted FS.
841 if (!fs_devices
->opened
&& found_transid
< device
->generation
) {
843 * That is if the FS is _not_ mounted and if you
844 * are here, that means there is more than one
845 * disk with same uuid and devid.We keep the one
846 * with larger generation number or the last-in if
847 * generation are equal.
849 mutex_unlock(&fs_devices
->device_list_mutex
);
850 return ERR_PTR(-EEXIST
);
853 name
= rcu_string_strdup(path
, GFP_NOFS
);
855 mutex_unlock(&fs_devices
->device_list_mutex
);
856 return ERR_PTR(-ENOMEM
);
858 rcu_string_free(device
->name
);
859 rcu_assign_pointer(device
->name
, name
);
860 if (test_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
)) {
861 fs_devices
->missing_devices
--;
862 clear_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
);
867 * Unmount does not free the btrfs_device struct but would zero
868 * generation along with most of the other members. So just update
869 * it back. We need it to pick the disk with largest generation
872 if (!fs_devices
->opened
)
873 device
->generation
= found_transid
;
875 fs_devices
->total_devices
= btrfs_super_num_devices(disk_super
);
877 mutex_unlock(&fs_devices
->device_list_mutex
);
881 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
883 struct btrfs_fs_devices
*fs_devices
;
884 struct btrfs_device
*device
;
885 struct btrfs_device
*orig_dev
;
887 fs_devices
= alloc_fs_devices(orig
->fsid
);
888 if (IS_ERR(fs_devices
))
891 mutex_lock(&orig
->device_list_mutex
);
892 fs_devices
->total_devices
= orig
->total_devices
;
894 /* We have held the volume lock, it is safe to get the devices. */
895 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
896 struct rcu_string
*name
;
898 device
= btrfs_alloc_device(NULL
, &orig_dev
->devid
,
904 * This is ok to do without rcu read locked because we hold the
905 * uuid mutex so nothing we touch in here is going to disappear.
907 if (orig_dev
->name
) {
908 name
= rcu_string_strdup(orig_dev
->name
->str
,
911 btrfs_free_device(device
);
914 rcu_assign_pointer(device
->name
, name
);
917 list_add(&device
->dev_list
, &fs_devices
->devices
);
918 device
->fs_devices
= fs_devices
;
919 fs_devices
->num_devices
++;
921 mutex_unlock(&orig
->device_list_mutex
);
924 mutex_unlock(&orig
->device_list_mutex
);
925 free_fs_devices(fs_devices
);
926 return ERR_PTR(-ENOMEM
);
930 * After we have read the system tree and know devids belonging to
931 * this filesystem, remove the device which does not belong there.
933 void btrfs_free_extra_devids(struct btrfs_fs_devices
*fs_devices
, int step
)
935 struct btrfs_device
*device
, *next
;
936 struct btrfs_device
*latest_dev
= NULL
;
938 mutex_lock(&uuid_mutex
);
940 /* This is the initialized path, it is safe to release the devices. */
941 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
942 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
,
943 &device
->dev_state
)) {
944 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT
,
945 &device
->dev_state
) &&
947 device
->generation
> latest_dev
->generation
)) {
953 if (device
->devid
== BTRFS_DEV_REPLACE_DEVID
) {
955 * In the first step, keep the device which has
956 * the correct fsid and the devid that is used
957 * for the dev_replace procedure.
958 * In the second step, the dev_replace state is
959 * read from the device tree and it is known
960 * whether the procedure is really active or
961 * not, which means whether this device is
962 * used or whether it should be removed.
964 if (step
== 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT
,
965 &device
->dev_state
)) {
970 blkdev_put(device
->bdev
, device
->mode
);
972 fs_devices
->open_devices
--;
974 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
)) {
975 list_del_init(&device
->dev_alloc_list
);
976 clear_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
);
977 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT
,
979 fs_devices
->rw_devices
--;
981 list_del_init(&device
->dev_list
);
982 fs_devices
->num_devices
--;
983 btrfs_free_device(device
);
986 if (fs_devices
->seed
) {
987 fs_devices
= fs_devices
->seed
;
991 fs_devices
->latest_bdev
= latest_dev
->bdev
;
993 mutex_unlock(&uuid_mutex
);
996 static void free_device_rcu(struct rcu_head
*head
)
998 struct btrfs_device
*device
;
1000 device
= container_of(head
, struct btrfs_device
, rcu
);
1001 btrfs_free_device(device
);
1004 static void btrfs_close_bdev(struct btrfs_device
*device
)
1009 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
)) {
1010 sync_blockdev(device
->bdev
);
1011 invalidate_bdev(device
->bdev
);
1014 blkdev_put(device
->bdev
, device
->mode
);
1017 static void btrfs_close_one_device(struct btrfs_device
*device
)
1019 struct btrfs_fs_devices
*fs_devices
= device
->fs_devices
;
1020 struct btrfs_device
*new_device
;
1021 struct rcu_string
*name
;
1024 fs_devices
->open_devices
--;
1026 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
) &&
1027 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
1028 list_del_init(&device
->dev_alloc_list
);
1029 fs_devices
->rw_devices
--;
1032 if (test_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
))
1033 fs_devices
->missing_devices
--;
1035 btrfs_close_bdev(device
);
1037 new_device
= btrfs_alloc_device(NULL
, &device
->devid
,
1039 BUG_ON(IS_ERR(new_device
)); /* -ENOMEM */
1041 /* Safe because we are under uuid_mutex */
1043 name
= rcu_string_strdup(device
->name
->str
, GFP_NOFS
);
1044 BUG_ON(!name
); /* -ENOMEM */
1045 rcu_assign_pointer(new_device
->name
, name
);
1048 list_replace_rcu(&device
->dev_list
, &new_device
->dev_list
);
1049 new_device
->fs_devices
= device
->fs_devices
;
1051 call_rcu(&device
->rcu
, free_device_rcu
);
1054 static int close_fs_devices(struct btrfs_fs_devices
*fs_devices
)
1056 struct btrfs_device
*device
, *tmp
;
1058 if (--fs_devices
->opened
> 0)
1061 mutex_lock(&fs_devices
->device_list_mutex
);
1062 list_for_each_entry_safe(device
, tmp
, &fs_devices
->devices
, dev_list
) {
1063 btrfs_close_one_device(device
);
1065 mutex_unlock(&fs_devices
->device_list_mutex
);
1067 WARN_ON(fs_devices
->open_devices
);
1068 WARN_ON(fs_devices
->rw_devices
);
1069 fs_devices
->opened
= 0;
1070 fs_devices
->seeding
= 0;
1075 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
1077 struct btrfs_fs_devices
*seed_devices
= NULL
;
1080 mutex_lock(&uuid_mutex
);
1081 ret
= close_fs_devices(fs_devices
);
1082 if (!fs_devices
->opened
) {
1083 seed_devices
= fs_devices
->seed
;
1084 fs_devices
->seed
= NULL
;
1086 mutex_unlock(&uuid_mutex
);
1088 while (seed_devices
) {
1089 fs_devices
= seed_devices
;
1090 seed_devices
= fs_devices
->seed
;
1091 close_fs_devices(fs_devices
);
1092 free_fs_devices(fs_devices
);
1097 static int open_fs_devices(struct btrfs_fs_devices
*fs_devices
,
1098 fmode_t flags
, void *holder
)
1100 struct btrfs_device
*device
;
1101 struct btrfs_device
*latest_dev
= NULL
;
1104 flags
|= FMODE_EXCL
;
1106 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
1107 /* Just open everything we can; ignore failures here */
1108 if (btrfs_open_one_device(fs_devices
, device
, flags
, holder
))
1112 device
->generation
> latest_dev
->generation
)
1113 latest_dev
= device
;
1115 if (fs_devices
->open_devices
== 0) {
1119 fs_devices
->opened
= 1;
1120 fs_devices
->latest_bdev
= latest_dev
->bdev
;
1121 fs_devices
->total_rw_bytes
= 0;
1126 static int devid_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
1128 struct btrfs_device
*dev1
, *dev2
;
1130 dev1
= list_entry(a
, struct btrfs_device
, dev_list
);
1131 dev2
= list_entry(b
, struct btrfs_device
, dev_list
);
1133 if (dev1
->devid
< dev2
->devid
)
1135 else if (dev1
->devid
> dev2
->devid
)
1140 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
1141 fmode_t flags
, void *holder
)
1145 lockdep_assert_held(&uuid_mutex
);
1147 mutex_lock(&fs_devices
->device_list_mutex
);
1148 if (fs_devices
->opened
) {
1149 fs_devices
->opened
++;
1152 list_sort(NULL
, &fs_devices
->devices
, devid_cmp
);
1153 ret
= open_fs_devices(fs_devices
, flags
, holder
);
1155 mutex_unlock(&fs_devices
->device_list_mutex
);
1160 static void btrfs_release_disk_super(struct page
*page
)
1166 static int btrfs_read_disk_super(struct block_device
*bdev
, u64 bytenr
,
1168 struct btrfs_super_block
**disk_super
)
1173 /* make sure our super fits in the device */
1174 if (bytenr
+ PAGE_SIZE
>= i_size_read(bdev
->bd_inode
))
1177 /* make sure our super fits in the page */
1178 if (sizeof(**disk_super
) > PAGE_SIZE
)
1181 /* make sure our super doesn't straddle pages on disk */
1182 index
= bytenr
>> PAGE_SHIFT
;
1183 if ((bytenr
+ sizeof(**disk_super
) - 1) >> PAGE_SHIFT
!= index
)
1186 /* pull in the page with our super */
1187 *page
= read_cache_page_gfp(bdev
->bd_inode
->i_mapping
,
1190 if (IS_ERR_OR_NULL(*page
))
1195 /* align our pointer to the offset of the super block */
1196 *disk_super
= p
+ (bytenr
& ~PAGE_MASK
);
1198 if (btrfs_super_bytenr(*disk_super
) != bytenr
||
1199 btrfs_super_magic(*disk_super
) != BTRFS_MAGIC
) {
1200 btrfs_release_disk_super(*page
);
1204 if ((*disk_super
)->label
[0] &&
1205 (*disk_super
)->label
[BTRFS_LABEL_SIZE
- 1])
1206 (*disk_super
)->label
[BTRFS_LABEL_SIZE
- 1] = '\0';
1212 * Look for a btrfs signature on a device. This may be called out of the mount path
1213 * and we are not allowed to call set_blocksize during the scan. The superblock
1214 * is read via pagecache
1216 struct btrfs_device
*btrfs_scan_one_device(const char *path
, fmode_t flags
,
1219 struct btrfs_super_block
*disk_super
;
1220 bool new_device_added
= false;
1221 struct btrfs_device
*device
= NULL
;
1222 struct block_device
*bdev
;
1226 lockdep_assert_held(&uuid_mutex
);
1229 * we would like to check all the supers, but that would make
1230 * a btrfs mount succeed after a mkfs from a different FS.
1231 * So, we need to add a special mount option to scan for
1232 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1234 bytenr
= btrfs_sb_offset(0);
1235 flags
|= FMODE_EXCL
;
1237 bdev
= blkdev_get_by_path(path
, flags
, holder
);
1239 return ERR_CAST(bdev
);
1241 if (btrfs_read_disk_super(bdev
, bytenr
, &page
, &disk_super
)) {
1242 device
= ERR_PTR(-EINVAL
);
1243 goto error_bdev_put
;
1246 device
= device_list_add(path
, disk_super
, &new_device_added
);
1247 if (!IS_ERR(device
)) {
1248 if (new_device_added
)
1249 btrfs_free_stale_devices(path
, device
);
1252 btrfs_release_disk_super(page
);
1255 blkdev_put(bdev
, flags
);
1260 static int contains_pending_extent(struct btrfs_transaction
*transaction
,
1261 struct btrfs_device
*device
,
1262 u64
*start
, u64 len
)
1264 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1265 struct extent_map
*em
;
1266 struct list_head
*search_list
= &fs_info
->pinned_chunks
;
1268 u64 physical_start
= *start
;
1271 search_list
= &transaction
->pending_chunks
;
1273 list_for_each_entry(em
, search_list
, list
) {
1274 struct map_lookup
*map
;
1277 map
= em
->map_lookup
;
1278 for (i
= 0; i
< map
->num_stripes
; i
++) {
1281 if (map
->stripes
[i
].dev
!= device
)
1283 if (map
->stripes
[i
].physical
>= physical_start
+ len
||
1284 map
->stripes
[i
].physical
+ em
->orig_block_len
<=
1288 * Make sure that while processing the pinned list we do
1289 * not override our *start with a lower value, because
1290 * we can have pinned chunks that fall within this
1291 * device hole and that have lower physical addresses
1292 * than the pending chunks we processed before. If we
1293 * do not take this special care we can end up getting
1294 * 2 pending chunks that start at the same physical
1295 * device offsets because the end offset of a pinned
1296 * chunk can be equal to the start offset of some
1299 end
= map
->stripes
[i
].physical
+ em
->orig_block_len
;
1306 if (search_list
!= &fs_info
->pinned_chunks
) {
1307 search_list
= &fs_info
->pinned_chunks
;
1316 * find_free_dev_extent_start - find free space in the specified device
1317 * @device: the device which we search the free space in
1318 * @num_bytes: the size of the free space that we need
1319 * @search_start: the position from which to begin the search
1320 * @start: store the start of the free space.
1321 * @len: the size of the free space. that we find, or the size
1322 * of the max free space if we don't find suitable free space
1324 * this uses a pretty simple search, the expectation is that it is
1325 * called very infrequently and that a given device has a small number
1328 * @start is used to store the start of the free space if we find. But if we
1329 * don't find suitable free space, it will be used to store the start position
1330 * of the max free space.
1332 * @len is used to store the size of the free space that we find.
1333 * But if we don't find suitable free space, it is used to store the size of
1334 * the max free space.
1336 int find_free_dev_extent_start(struct btrfs_transaction
*transaction
,
1337 struct btrfs_device
*device
, u64 num_bytes
,
1338 u64 search_start
, u64
*start
, u64
*len
)
1340 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1341 struct btrfs_root
*root
= fs_info
->dev_root
;
1342 struct btrfs_key key
;
1343 struct btrfs_dev_extent
*dev_extent
;
1344 struct btrfs_path
*path
;
1349 u64 search_end
= device
->total_bytes
;
1352 struct extent_buffer
*l
;
1355 * We don't want to overwrite the superblock on the drive nor any area
1356 * used by the boot loader (grub for example), so we make sure to start
1357 * at an offset of at least 1MB.
1359 search_start
= max_t(u64
, search_start
, SZ_1M
);
1361 path
= btrfs_alloc_path();
1365 max_hole_start
= search_start
;
1369 if (search_start
>= search_end
||
1370 test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
)) {
1375 path
->reada
= READA_FORWARD
;
1376 path
->search_commit_root
= 1;
1377 path
->skip_locking
= 1;
1379 key
.objectid
= device
->devid
;
1380 key
.offset
= search_start
;
1381 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1383 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1387 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1394 slot
= path
->slots
[0];
1395 if (slot
>= btrfs_header_nritems(l
)) {
1396 ret
= btrfs_next_leaf(root
, path
);
1404 btrfs_item_key_to_cpu(l
, &key
, slot
);
1406 if (key
.objectid
< device
->devid
)
1409 if (key
.objectid
> device
->devid
)
1412 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1415 if (key
.offset
> search_start
) {
1416 hole_size
= key
.offset
- search_start
;
1419 * Have to check before we set max_hole_start, otherwise
1420 * we could end up sending back this offset anyway.
1422 if (contains_pending_extent(transaction
, device
,
1425 if (key
.offset
>= search_start
) {
1426 hole_size
= key
.offset
- search_start
;
1433 if (hole_size
> max_hole_size
) {
1434 max_hole_start
= search_start
;
1435 max_hole_size
= hole_size
;
1439 * If this free space is greater than which we need,
1440 * it must be the max free space that we have found
1441 * until now, so max_hole_start must point to the start
1442 * of this free space and the length of this free space
1443 * is stored in max_hole_size. Thus, we return
1444 * max_hole_start and max_hole_size and go back to the
1447 if (hole_size
>= num_bytes
) {
1453 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1454 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1456 if (extent_end
> search_start
)
1457 search_start
= extent_end
;
1464 * At this point, search_start should be the end of
1465 * allocated dev extents, and when shrinking the device,
1466 * search_end may be smaller than search_start.
1468 if (search_end
> search_start
) {
1469 hole_size
= search_end
- search_start
;
1471 if (contains_pending_extent(transaction
, device
, &search_start
,
1473 btrfs_release_path(path
);
1477 if (hole_size
> max_hole_size
) {
1478 max_hole_start
= search_start
;
1479 max_hole_size
= hole_size
;
1484 if (max_hole_size
< num_bytes
)
1490 btrfs_free_path(path
);
1491 *start
= max_hole_start
;
1493 *len
= max_hole_size
;
1497 int find_free_dev_extent(struct btrfs_trans_handle
*trans
,
1498 struct btrfs_device
*device
, u64 num_bytes
,
1499 u64
*start
, u64
*len
)
1501 /* FIXME use last free of some kind */
1502 return find_free_dev_extent_start(trans
->transaction
, device
,
1503 num_bytes
, 0, start
, len
);
1506 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1507 struct btrfs_device
*device
,
1508 u64 start
, u64
*dev_extent_len
)
1510 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1511 struct btrfs_root
*root
= fs_info
->dev_root
;
1513 struct btrfs_path
*path
;
1514 struct btrfs_key key
;
1515 struct btrfs_key found_key
;
1516 struct extent_buffer
*leaf
= NULL
;
1517 struct btrfs_dev_extent
*extent
= NULL
;
1519 path
= btrfs_alloc_path();
1523 key
.objectid
= device
->devid
;
1525 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1527 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1529 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1530 BTRFS_DEV_EXTENT_KEY
);
1533 leaf
= path
->nodes
[0];
1534 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1535 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1536 struct btrfs_dev_extent
);
1537 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1538 btrfs_dev_extent_length(leaf
, extent
) < start
);
1540 btrfs_release_path(path
);
1542 } else if (ret
== 0) {
1543 leaf
= path
->nodes
[0];
1544 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1545 struct btrfs_dev_extent
);
1547 btrfs_handle_fs_error(fs_info
, ret
, "Slot search failed");
1551 *dev_extent_len
= btrfs_dev_extent_length(leaf
, extent
);
1553 ret
= btrfs_del_item(trans
, root
, path
);
1555 btrfs_handle_fs_error(fs_info
, ret
,
1556 "Failed to remove dev extent item");
1558 set_bit(BTRFS_TRANS_HAVE_FREE_BGS
, &trans
->transaction
->flags
);
1561 btrfs_free_path(path
);
1565 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1566 struct btrfs_device
*device
,
1567 u64 chunk_offset
, u64 start
, u64 num_bytes
)
1570 struct btrfs_path
*path
;
1571 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1572 struct btrfs_root
*root
= fs_info
->dev_root
;
1573 struct btrfs_dev_extent
*extent
;
1574 struct extent_buffer
*leaf
;
1575 struct btrfs_key key
;
1577 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &device
->dev_state
));
1578 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
));
1579 path
= btrfs_alloc_path();
1583 key
.objectid
= device
->devid
;
1585 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1586 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1591 leaf
= path
->nodes
[0];
1592 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1593 struct btrfs_dev_extent
);
1594 btrfs_set_dev_extent_chunk_tree(leaf
, extent
,
1595 BTRFS_CHUNK_TREE_OBJECTID
);
1596 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
,
1597 BTRFS_FIRST_CHUNK_TREE_OBJECTID
);
1598 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1600 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1601 btrfs_mark_buffer_dirty(leaf
);
1603 btrfs_free_path(path
);
1607 static u64
find_next_chunk(struct btrfs_fs_info
*fs_info
)
1609 struct extent_map_tree
*em_tree
;
1610 struct extent_map
*em
;
1614 em_tree
= &fs_info
->mapping_tree
.map_tree
;
1615 read_lock(&em_tree
->lock
);
1616 n
= rb_last(&em_tree
->map
);
1618 em
= rb_entry(n
, struct extent_map
, rb_node
);
1619 ret
= em
->start
+ em
->len
;
1621 read_unlock(&em_tree
->lock
);
1626 static noinline
int find_next_devid(struct btrfs_fs_info
*fs_info
,
1630 struct btrfs_key key
;
1631 struct btrfs_key found_key
;
1632 struct btrfs_path
*path
;
1634 path
= btrfs_alloc_path();
1638 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1639 key
.type
= BTRFS_DEV_ITEM_KEY
;
1640 key
.offset
= (u64
)-1;
1642 ret
= btrfs_search_slot(NULL
, fs_info
->chunk_root
, &key
, path
, 0, 0);
1646 BUG_ON(ret
== 0); /* Corruption */
1648 ret
= btrfs_previous_item(fs_info
->chunk_root
, path
,
1649 BTRFS_DEV_ITEMS_OBJECTID
,
1650 BTRFS_DEV_ITEM_KEY
);
1654 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1656 *devid_ret
= found_key
.offset
+ 1;
1660 btrfs_free_path(path
);
1665 * the device information is stored in the chunk root
1666 * the btrfs_device struct should be fully filled in
1668 static int btrfs_add_dev_item(struct btrfs_trans_handle
*trans
,
1669 struct btrfs_device
*device
)
1672 struct btrfs_path
*path
;
1673 struct btrfs_dev_item
*dev_item
;
1674 struct extent_buffer
*leaf
;
1675 struct btrfs_key key
;
1678 path
= btrfs_alloc_path();
1682 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1683 key
.type
= BTRFS_DEV_ITEM_KEY
;
1684 key
.offset
= device
->devid
;
1686 ret
= btrfs_insert_empty_item(trans
, trans
->fs_info
->chunk_root
, path
,
1687 &key
, sizeof(*dev_item
));
1691 leaf
= path
->nodes
[0];
1692 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1694 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1695 btrfs_set_device_generation(leaf
, dev_item
, 0);
1696 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1697 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1698 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1699 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1700 btrfs_set_device_total_bytes(leaf
, dev_item
,
1701 btrfs_device_get_disk_total_bytes(device
));
1702 btrfs_set_device_bytes_used(leaf
, dev_item
,
1703 btrfs_device_get_bytes_used(device
));
1704 btrfs_set_device_group(leaf
, dev_item
, 0);
1705 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1706 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1707 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1709 ptr
= btrfs_device_uuid(dev_item
);
1710 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1711 ptr
= btrfs_device_fsid(dev_item
);
1712 write_extent_buffer(leaf
, trans
->fs_info
->fsid
, ptr
, BTRFS_FSID_SIZE
);
1713 btrfs_mark_buffer_dirty(leaf
);
1717 btrfs_free_path(path
);
1722 * Function to update ctime/mtime for a given device path.
1723 * Mainly used for ctime/mtime based probe like libblkid.
1725 static void update_dev_time(const char *path_name
)
1729 filp
= filp_open(path_name
, O_RDWR
, 0);
1732 file_update_time(filp
);
1733 filp_close(filp
, NULL
);
1736 static int btrfs_rm_dev_item(struct btrfs_fs_info
*fs_info
,
1737 struct btrfs_device
*device
)
1739 struct btrfs_root
*root
= fs_info
->chunk_root
;
1741 struct btrfs_path
*path
;
1742 struct btrfs_key key
;
1743 struct btrfs_trans_handle
*trans
;
1745 path
= btrfs_alloc_path();
1749 trans
= btrfs_start_transaction(root
, 0);
1750 if (IS_ERR(trans
)) {
1751 btrfs_free_path(path
);
1752 return PTR_ERR(trans
);
1754 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1755 key
.type
= BTRFS_DEV_ITEM_KEY
;
1756 key
.offset
= device
->devid
;
1758 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1762 btrfs_abort_transaction(trans
, ret
);
1763 btrfs_end_transaction(trans
);
1767 ret
= btrfs_del_item(trans
, root
, path
);
1769 btrfs_abort_transaction(trans
, ret
);
1770 btrfs_end_transaction(trans
);
1774 btrfs_free_path(path
);
1776 ret
= btrfs_commit_transaction(trans
);
1781 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1782 * filesystem. It's up to the caller to adjust that number regarding eg. device
1785 static int btrfs_check_raid_min_devices(struct btrfs_fs_info
*fs_info
,
1793 seq
= read_seqbegin(&fs_info
->profiles_lock
);
1795 all_avail
= fs_info
->avail_data_alloc_bits
|
1796 fs_info
->avail_system_alloc_bits
|
1797 fs_info
->avail_metadata_alloc_bits
;
1798 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
1800 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
1801 if (!(all_avail
& btrfs_raid_array
[i
].bg_flag
))
1804 if (num_devices
< btrfs_raid_array
[i
].devs_min
) {
1805 int ret
= btrfs_raid_array
[i
].mindev_error
;
1815 static struct btrfs_device
* btrfs_find_next_active_device(
1816 struct btrfs_fs_devices
*fs_devs
, struct btrfs_device
*device
)
1818 struct btrfs_device
*next_device
;
1820 list_for_each_entry(next_device
, &fs_devs
->devices
, dev_list
) {
1821 if (next_device
!= device
&&
1822 !test_bit(BTRFS_DEV_STATE_MISSING
, &next_device
->dev_state
)
1823 && next_device
->bdev
)
1831 * Helper function to check if the given device is part of s_bdev / latest_bdev
1832 * and replace it with the provided or the next active device, in the context
1833 * where this function called, there should be always be another device (or
1834 * this_dev) which is active.
1836 void btrfs_assign_next_active_device(struct btrfs_device
*device
,
1837 struct btrfs_device
*this_dev
)
1839 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1840 struct btrfs_device
*next_device
;
1843 next_device
= this_dev
;
1845 next_device
= btrfs_find_next_active_device(fs_info
->fs_devices
,
1847 ASSERT(next_device
);
1849 if (fs_info
->sb
->s_bdev
&&
1850 (fs_info
->sb
->s_bdev
== device
->bdev
))
1851 fs_info
->sb
->s_bdev
= next_device
->bdev
;
1853 if (fs_info
->fs_devices
->latest_bdev
== device
->bdev
)
1854 fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
1857 int btrfs_rm_device(struct btrfs_fs_info
*fs_info
, const char *device_path
,
1860 struct btrfs_device
*device
;
1861 struct btrfs_fs_devices
*cur_devices
;
1862 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
1866 mutex_lock(&uuid_mutex
);
1868 num_devices
= fs_devices
->num_devices
;
1869 btrfs_dev_replace_read_lock(&fs_info
->dev_replace
);
1870 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
1871 WARN_ON(num_devices
< 1);
1874 btrfs_dev_replace_read_unlock(&fs_info
->dev_replace
);
1876 ret
= btrfs_check_raid_min_devices(fs_info
, num_devices
- 1);
1880 ret
= btrfs_find_device_by_devspec(fs_info
, devid
, device_path
,
1885 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
)) {
1886 ret
= BTRFS_ERROR_DEV_TGT_REPLACE
;
1890 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
) &&
1891 fs_info
->fs_devices
->rw_devices
== 1) {
1892 ret
= BTRFS_ERROR_DEV_ONLY_WRITABLE
;
1896 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
)) {
1897 mutex_lock(&fs_info
->chunk_mutex
);
1898 list_del_init(&device
->dev_alloc_list
);
1899 device
->fs_devices
->rw_devices
--;
1900 mutex_unlock(&fs_info
->chunk_mutex
);
1903 mutex_unlock(&uuid_mutex
);
1904 ret
= btrfs_shrink_device(device
, 0);
1905 mutex_lock(&uuid_mutex
);
1910 * TODO: the superblock still includes this device in its num_devices
1911 * counter although write_all_supers() is not locked out. This
1912 * could give a filesystem state which requires a degraded mount.
1914 ret
= btrfs_rm_dev_item(fs_info
, device
);
1918 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &device
->dev_state
);
1919 btrfs_scrub_cancel_dev(fs_info
, device
);
1922 * the device list mutex makes sure that we don't change
1923 * the device list while someone else is writing out all
1924 * the device supers. Whoever is writing all supers, should
1925 * lock the device list mutex before getting the number of
1926 * devices in the super block (super_copy). Conversely,
1927 * whoever updates the number of devices in the super block
1928 * (super_copy) should hold the device list mutex.
1932 * In normal cases the cur_devices == fs_devices. But in case
1933 * of deleting a seed device, the cur_devices should point to
1934 * its own fs_devices listed under the fs_devices->seed.
1936 cur_devices
= device
->fs_devices
;
1937 mutex_lock(&fs_devices
->device_list_mutex
);
1938 list_del_rcu(&device
->dev_list
);
1940 cur_devices
->num_devices
--;
1941 cur_devices
->total_devices
--;
1942 /* Update total_devices of the parent fs_devices if it's seed */
1943 if (cur_devices
!= fs_devices
)
1944 fs_devices
->total_devices
--;
1946 if (test_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
))
1947 cur_devices
->missing_devices
--;
1949 btrfs_assign_next_active_device(device
, NULL
);
1952 cur_devices
->open_devices
--;
1953 /* remove sysfs entry */
1954 btrfs_sysfs_rm_device_link(fs_devices
, device
);
1957 num_devices
= btrfs_super_num_devices(fs_info
->super_copy
) - 1;
1958 btrfs_set_super_num_devices(fs_info
->super_copy
, num_devices
);
1959 mutex_unlock(&fs_devices
->device_list_mutex
);
1962 * at this point, the device is zero sized and detached from
1963 * the devices list. All that's left is to zero out the old
1964 * supers and free the device.
1966 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
))
1967 btrfs_scratch_superblocks(device
->bdev
, device
->name
->str
);
1969 btrfs_close_bdev(device
);
1970 call_rcu(&device
->rcu
, free_device_rcu
);
1972 if (cur_devices
->open_devices
== 0) {
1973 while (fs_devices
) {
1974 if (fs_devices
->seed
== cur_devices
) {
1975 fs_devices
->seed
= cur_devices
->seed
;
1978 fs_devices
= fs_devices
->seed
;
1980 cur_devices
->seed
= NULL
;
1981 close_fs_devices(cur_devices
);
1982 free_fs_devices(cur_devices
);
1986 mutex_unlock(&uuid_mutex
);
1990 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
)) {
1991 mutex_lock(&fs_info
->chunk_mutex
);
1992 list_add(&device
->dev_alloc_list
,
1993 &fs_devices
->alloc_list
);
1994 device
->fs_devices
->rw_devices
++;
1995 mutex_unlock(&fs_info
->chunk_mutex
);
2000 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device
*srcdev
)
2002 struct btrfs_fs_devices
*fs_devices
;
2004 lockdep_assert_held(&srcdev
->fs_info
->fs_devices
->device_list_mutex
);
2007 * in case of fs with no seed, srcdev->fs_devices will point
2008 * to fs_devices of fs_info. However when the dev being replaced is
2009 * a seed dev it will point to the seed's local fs_devices. In short
2010 * srcdev will have its correct fs_devices in both the cases.
2012 fs_devices
= srcdev
->fs_devices
;
2014 list_del_rcu(&srcdev
->dev_list
);
2015 list_del(&srcdev
->dev_alloc_list
);
2016 fs_devices
->num_devices
--;
2017 if (test_bit(BTRFS_DEV_STATE_MISSING
, &srcdev
->dev_state
))
2018 fs_devices
->missing_devices
--;
2020 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &srcdev
->dev_state
))
2021 fs_devices
->rw_devices
--;
2024 fs_devices
->open_devices
--;
2027 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info
*fs_info
,
2028 struct btrfs_device
*srcdev
)
2030 struct btrfs_fs_devices
*fs_devices
= srcdev
->fs_devices
;
2032 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &srcdev
->dev_state
)) {
2033 /* zero out the old super if it is writable */
2034 btrfs_scratch_superblocks(srcdev
->bdev
, srcdev
->name
->str
);
2037 btrfs_close_bdev(srcdev
);
2038 call_rcu(&srcdev
->rcu
, free_device_rcu
);
2040 /* if this is no devs we rather delete the fs_devices */
2041 if (!fs_devices
->num_devices
) {
2042 struct btrfs_fs_devices
*tmp_fs_devices
;
2045 * On a mounted FS, num_devices can't be zero unless it's a
2046 * seed. In case of a seed device being replaced, the replace
2047 * target added to the sprout FS, so there will be no more
2048 * device left under the seed FS.
2050 ASSERT(fs_devices
->seeding
);
2052 tmp_fs_devices
= fs_info
->fs_devices
;
2053 while (tmp_fs_devices
) {
2054 if (tmp_fs_devices
->seed
== fs_devices
) {
2055 tmp_fs_devices
->seed
= fs_devices
->seed
;
2058 tmp_fs_devices
= tmp_fs_devices
->seed
;
2060 fs_devices
->seed
= NULL
;
2061 close_fs_devices(fs_devices
);
2062 free_fs_devices(fs_devices
);
2066 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device
*tgtdev
)
2068 struct btrfs_fs_devices
*fs_devices
= tgtdev
->fs_info
->fs_devices
;
2071 mutex_lock(&fs_devices
->device_list_mutex
);
2073 btrfs_sysfs_rm_device_link(fs_devices
, tgtdev
);
2076 fs_devices
->open_devices
--;
2078 fs_devices
->num_devices
--;
2080 btrfs_assign_next_active_device(tgtdev
, NULL
);
2082 list_del_rcu(&tgtdev
->dev_list
);
2084 mutex_unlock(&fs_devices
->device_list_mutex
);
2087 * The update_dev_time() with in btrfs_scratch_superblocks()
2088 * may lead to a call to btrfs_show_devname() which will try
2089 * to hold device_list_mutex. And here this device
2090 * is already out of device list, so we don't have to hold
2091 * the device_list_mutex lock.
2093 btrfs_scratch_superblocks(tgtdev
->bdev
, tgtdev
->name
->str
);
2095 btrfs_close_bdev(tgtdev
);
2096 call_rcu(&tgtdev
->rcu
, free_device_rcu
);
2099 static int btrfs_find_device_by_path(struct btrfs_fs_info
*fs_info
,
2100 const char *device_path
,
2101 struct btrfs_device
**device
)
2104 struct btrfs_super_block
*disk_super
;
2107 struct block_device
*bdev
;
2108 struct buffer_head
*bh
;
2111 ret
= btrfs_get_bdev_and_sb(device_path
, FMODE_READ
,
2112 fs_info
->bdev_holder
, 0, &bdev
, &bh
);
2115 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
2116 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
2117 dev_uuid
= disk_super
->dev_item
.uuid
;
2118 *device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, disk_super
->fsid
);
2122 blkdev_put(bdev
, FMODE_READ
);
2126 int btrfs_find_device_missing_or_by_path(struct btrfs_fs_info
*fs_info
,
2127 const char *device_path
,
2128 struct btrfs_device
**device
)
2131 if (strcmp(device_path
, "missing") == 0) {
2132 struct list_head
*devices
;
2133 struct btrfs_device
*tmp
;
2135 devices
= &fs_info
->fs_devices
->devices
;
2136 list_for_each_entry(tmp
, devices
, dev_list
) {
2137 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
,
2138 &tmp
->dev_state
) && !tmp
->bdev
) {
2145 return BTRFS_ERROR_DEV_MISSING_NOT_FOUND
;
2149 return btrfs_find_device_by_path(fs_info
, device_path
, device
);
2154 * Lookup a device given by device id, or the path if the id is 0.
2156 int btrfs_find_device_by_devspec(struct btrfs_fs_info
*fs_info
, u64 devid
,
2157 const char *devpath
,
2158 struct btrfs_device
**device
)
2164 *device
= btrfs_find_device(fs_info
, devid
, NULL
, NULL
);
2168 if (!devpath
|| !devpath
[0])
2171 ret
= btrfs_find_device_missing_or_by_path(fs_info
, devpath
,
2178 * does all the dirty work required for changing file system's UUID.
2180 static int btrfs_prepare_sprout(struct btrfs_fs_info
*fs_info
)
2182 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2183 struct btrfs_fs_devices
*old_devices
;
2184 struct btrfs_fs_devices
*seed_devices
;
2185 struct btrfs_super_block
*disk_super
= fs_info
->super_copy
;
2186 struct btrfs_device
*device
;
2189 lockdep_assert_held(&uuid_mutex
);
2190 if (!fs_devices
->seeding
)
2193 seed_devices
= alloc_fs_devices(NULL
);
2194 if (IS_ERR(seed_devices
))
2195 return PTR_ERR(seed_devices
);
2197 old_devices
= clone_fs_devices(fs_devices
);
2198 if (IS_ERR(old_devices
)) {
2199 kfree(seed_devices
);
2200 return PTR_ERR(old_devices
);
2203 list_add(&old_devices
->fs_list
, &fs_uuids
);
2205 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
2206 seed_devices
->opened
= 1;
2207 INIT_LIST_HEAD(&seed_devices
->devices
);
2208 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
2209 mutex_init(&seed_devices
->device_list_mutex
);
2211 mutex_lock(&fs_devices
->device_list_mutex
);
2212 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
2214 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
)
2215 device
->fs_devices
= seed_devices
;
2217 mutex_lock(&fs_info
->chunk_mutex
);
2218 list_splice_init(&fs_devices
->alloc_list
, &seed_devices
->alloc_list
);
2219 mutex_unlock(&fs_info
->chunk_mutex
);
2221 fs_devices
->seeding
= 0;
2222 fs_devices
->num_devices
= 0;
2223 fs_devices
->open_devices
= 0;
2224 fs_devices
->missing_devices
= 0;
2225 fs_devices
->rotating
= 0;
2226 fs_devices
->seed
= seed_devices
;
2228 generate_random_uuid(fs_devices
->fsid
);
2229 memcpy(fs_info
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2230 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2231 mutex_unlock(&fs_devices
->device_list_mutex
);
2233 super_flags
= btrfs_super_flags(disk_super
) &
2234 ~BTRFS_SUPER_FLAG_SEEDING
;
2235 btrfs_set_super_flags(disk_super
, super_flags
);
2241 * Store the expected generation for seed devices in device items.
2243 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
,
2244 struct btrfs_fs_info
*fs_info
)
2246 struct btrfs_root
*root
= fs_info
->chunk_root
;
2247 struct btrfs_path
*path
;
2248 struct extent_buffer
*leaf
;
2249 struct btrfs_dev_item
*dev_item
;
2250 struct btrfs_device
*device
;
2251 struct btrfs_key key
;
2252 u8 fs_uuid
[BTRFS_FSID_SIZE
];
2253 u8 dev_uuid
[BTRFS_UUID_SIZE
];
2257 path
= btrfs_alloc_path();
2261 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2263 key
.type
= BTRFS_DEV_ITEM_KEY
;
2266 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2270 leaf
= path
->nodes
[0];
2272 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
2273 ret
= btrfs_next_leaf(root
, path
);
2278 leaf
= path
->nodes
[0];
2279 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2280 btrfs_release_path(path
);
2284 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2285 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
2286 key
.type
!= BTRFS_DEV_ITEM_KEY
)
2289 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2290 struct btrfs_dev_item
);
2291 devid
= btrfs_device_id(leaf
, dev_item
);
2292 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
2294 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
2296 device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, fs_uuid
);
2297 BUG_ON(!device
); /* Logic error */
2299 if (device
->fs_devices
->seeding
) {
2300 btrfs_set_device_generation(leaf
, dev_item
,
2301 device
->generation
);
2302 btrfs_mark_buffer_dirty(leaf
);
2310 btrfs_free_path(path
);
2314 int btrfs_init_new_device(struct btrfs_fs_info
*fs_info
, const char *device_path
)
2316 struct btrfs_root
*root
= fs_info
->dev_root
;
2317 struct request_queue
*q
;
2318 struct btrfs_trans_handle
*trans
;
2319 struct btrfs_device
*device
;
2320 struct block_device
*bdev
;
2321 struct super_block
*sb
= fs_info
->sb
;
2322 struct rcu_string
*name
;
2323 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2324 u64 orig_super_total_bytes
;
2325 u64 orig_super_num_devices
;
2326 int seeding_dev
= 0;
2328 bool unlocked
= false;
2330 if (sb_rdonly(sb
) && !fs_devices
->seeding
)
2333 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2334 fs_info
->bdev_holder
);
2336 return PTR_ERR(bdev
);
2338 if (fs_devices
->seeding
) {
2340 down_write(&sb
->s_umount
);
2341 mutex_lock(&uuid_mutex
);
2344 filemap_write_and_wait(bdev
->bd_inode
->i_mapping
);
2346 mutex_lock(&fs_devices
->device_list_mutex
);
2347 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
2348 if (device
->bdev
== bdev
) {
2351 &fs_devices
->device_list_mutex
);
2355 mutex_unlock(&fs_devices
->device_list_mutex
);
2357 device
= btrfs_alloc_device(fs_info
, NULL
, NULL
);
2358 if (IS_ERR(device
)) {
2359 /* we can safely leave the fs_devices entry around */
2360 ret
= PTR_ERR(device
);
2364 name
= rcu_string_strdup(device_path
, GFP_KERNEL
);
2367 goto error_free_device
;
2369 rcu_assign_pointer(device
->name
, name
);
2371 trans
= btrfs_start_transaction(root
, 0);
2372 if (IS_ERR(trans
)) {
2373 ret
= PTR_ERR(trans
);
2374 goto error_free_device
;
2377 q
= bdev_get_queue(bdev
);
2378 set_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
);
2379 device
->generation
= trans
->transid
;
2380 device
->io_width
= fs_info
->sectorsize
;
2381 device
->io_align
= fs_info
->sectorsize
;
2382 device
->sector_size
= fs_info
->sectorsize
;
2383 device
->total_bytes
= round_down(i_size_read(bdev
->bd_inode
),
2384 fs_info
->sectorsize
);
2385 device
->disk_total_bytes
= device
->total_bytes
;
2386 device
->commit_total_bytes
= device
->total_bytes
;
2387 device
->fs_info
= fs_info
;
2388 device
->bdev
= bdev
;
2389 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &device
->dev_state
);
2390 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
);
2391 device
->mode
= FMODE_EXCL
;
2392 device
->dev_stats_valid
= 1;
2393 set_blocksize(device
->bdev
, BTRFS_BDEV_BLOCKSIZE
);
2396 sb
->s_flags
&= ~SB_RDONLY
;
2397 ret
= btrfs_prepare_sprout(fs_info
);
2399 btrfs_abort_transaction(trans
, ret
);
2404 device
->fs_devices
= fs_devices
;
2406 mutex_lock(&fs_devices
->device_list_mutex
);
2407 mutex_lock(&fs_info
->chunk_mutex
);
2408 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
2409 list_add(&device
->dev_alloc_list
, &fs_devices
->alloc_list
);
2410 fs_devices
->num_devices
++;
2411 fs_devices
->open_devices
++;
2412 fs_devices
->rw_devices
++;
2413 fs_devices
->total_devices
++;
2414 fs_devices
->total_rw_bytes
+= device
->total_bytes
;
2416 atomic64_add(device
->total_bytes
, &fs_info
->free_chunk_space
);
2418 if (!blk_queue_nonrot(q
))
2419 fs_devices
->rotating
= 1;
2421 orig_super_total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
2422 btrfs_set_super_total_bytes(fs_info
->super_copy
,
2423 round_down(orig_super_total_bytes
+ device
->total_bytes
,
2424 fs_info
->sectorsize
));
2426 orig_super_num_devices
= btrfs_super_num_devices(fs_info
->super_copy
);
2427 btrfs_set_super_num_devices(fs_info
->super_copy
,
2428 orig_super_num_devices
+ 1);
2430 /* add sysfs device entry */
2431 btrfs_sysfs_add_device_link(fs_devices
, device
);
2434 * we've got more storage, clear any full flags on the space
2437 btrfs_clear_space_info_full(fs_info
);
2439 mutex_unlock(&fs_info
->chunk_mutex
);
2440 mutex_unlock(&fs_devices
->device_list_mutex
);
2443 mutex_lock(&fs_info
->chunk_mutex
);
2444 ret
= init_first_rw_device(trans
, fs_info
);
2445 mutex_unlock(&fs_info
->chunk_mutex
);
2447 btrfs_abort_transaction(trans
, ret
);
2452 ret
= btrfs_add_dev_item(trans
, device
);
2454 btrfs_abort_transaction(trans
, ret
);
2459 char fsid_buf
[BTRFS_UUID_UNPARSED_SIZE
];
2461 ret
= btrfs_finish_sprout(trans
, fs_info
);
2463 btrfs_abort_transaction(trans
, ret
);
2467 /* Sprouting would change fsid of the mounted root,
2468 * so rename the fsid on the sysfs
2470 snprintf(fsid_buf
, BTRFS_UUID_UNPARSED_SIZE
, "%pU",
2472 if (kobject_rename(&fs_devices
->fsid_kobj
, fsid_buf
))
2474 "sysfs: failed to create fsid for sprout");
2477 ret
= btrfs_commit_transaction(trans
);
2480 mutex_unlock(&uuid_mutex
);
2481 up_write(&sb
->s_umount
);
2484 if (ret
) /* transaction commit */
2487 ret
= btrfs_relocate_sys_chunks(fs_info
);
2489 btrfs_handle_fs_error(fs_info
, ret
,
2490 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2491 trans
= btrfs_attach_transaction(root
);
2492 if (IS_ERR(trans
)) {
2493 if (PTR_ERR(trans
) == -ENOENT
)
2495 ret
= PTR_ERR(trans
);
2499 ret
= btrfs_commit_transaction(trans
);
2502 /* Update ctime/mtime for libblkid */
2503 update_dev_time(device_path
);
2507 btrfs_sysfs_rm_device_link(fs_devices
, device
);
2508 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2509 mutex_lock(&fs_info
->chunk_mutex
);
2510 list_del_rcu(&device
->dev_list
);
2511 list_del(&device
->dev_alloc_list
);
2512 fs_info
->fs_devices
->num_devices
--;
2513 fs_info
->fs_devices
->open_devices
--;
2514 fs_info
->fs_devices
->rw_devices
--;
2515 fs_info
->fs_devices
->total_devices
--;
2516 fs_info
->fs_devices
->total_rw_bytes
-= device
->total_bytes
;
2517 atomic64_sub(device
->total_bytes
, &fs_info
->free_chunk_space
);
2518 btrfs_set_super_total_bytes(fs_info
->super_copy
,
2519 orig_super_total_bytes
);
2520 btrfs_set_super_num_devices(fs_info
->super_copy
,
2521 orig_super_num_devices
);
2522 mutex_unlock(&fs_info
->chunk_mutex
);
2523 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2526 sb
->s_flags
|= SB_RDONLY
;
2528 btrfs_end_transaction(trans
);
2530 btrfs_free_device(device
);
2532 blkdev_put(bdev
, FMODE_EXCL
);
2533 if (seeding_dev
&& !unlocked
) {
2534 mutex_unlock(&uuid_mutex
);
2535 up_write(&sb
->s_umount
);
2540 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2541 struct btrfs_device
*device
)
2544 struct btrfs_path
*path
;
2545 struct btrfs_root
*root
= device
->fs_info
->chunk_root
;
2546 struct btrfs_dev_item
*dev_item
;
2547 struct extent_buffer
*leaf
;
2548 struct btrfs_key key
;
2550 path
= btrfs_alloc_path();
2554 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2555 key
.type
= BTRFS_DEV_ITEM_KEY
;
2556 key
.offset
= device
->devid
;
2558 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2567 leaf
= path
->nodes
[0];
2568 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2570 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2571 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2572 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2573 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2574 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2575 btrfs_set_device_total_bytes(leaf
, dev_item
,
2576 btrfs_device_get_disk_total_bytes(device
));
2577 btrfs_set_device_bytes_used(leaf
, dev_item
,
2578 btrfs_device_get_bytes_used(device
));
2579 btrfs_mark_buffer_dirty(leaf
);
2582 btrfs_free_path(path
);
2586 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2587 struct btrfs_device
*device
, u64 new_size
)
2589 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
2590 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
2591 struct btrfs_fs_devices
*fs_devices
;
2595 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
))
2598 new_size
= round_down(new_size
, fs_info
->sectorsize
);
2600 mutex_lock(&fs_info
->chunk_mutex
);
2601 old_total
= btrfs_super_total_bytes(super_copy
);
2602 diff
= round_down(new_size
- device
->total_bytes
, fs_info
->sectorsize
);
2604 if (new_size
<= device
->total_bytes
||
2605 test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
)) {
2606 mutex_unlock(&fs_info
->chunk_mutex
);
2610 fs_devices
= fs_info
->fs_devices
;
2612 btrfs_set_super_total_bytes(super_copy
,
2613 round_down(old_total
+ diff
, fs_info
->sectorsize
));
2614 device
->fs_devices
->total_rw_bytes
+= diff
;
2616 btrfs_device_set_total_bytes(device
, new_size
);
2617 btrfs_device_set_disk_total_bytes(device
, new_size
);
2618 btrfs_clear_space_info_full(device
->fs_info
);
2619 if (list_empty(&device
->resized_list
))
2620 list_add_tail(&device
->resized_list
,
2621 &fs_devices
->resized_devices
);
2622 mutex_unlock(&fs_info
->chunk_mutex
);
2624 return btrfs_update_device(trans
, device
);
2627 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
, u64 chunk_offset
)
2629 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2630 struct btrfs_root
*root
= fs_info
->chunk_root
;
2632 struct btrfs_path
*path
;
2633 struct btrfs_key key
;
2635 path
= btrfs_alloc_path();
2639 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2640 key
.offset
= chunk_offset
;
2641 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2643 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2646 else if (ret
> 0) { /* Logic error or corruption */
2647 btrfs_handle_fs_error(fs_info
, -ENOENT
,
2648 "Failed lookup while freeing chunk.");
2653 ret
= btrfs_del_item(trans
, root
, path
);
2655 btrfs_handle_fs_error(fs_info
, ret
,
2656 "Failed to delete chunk item.");
2658 btrfs_free_path(path
);
2662 static int btrfs_del_sys_chunk(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2664 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
2665 struct btrfs_disk_key
*disk_key
;
2666 struct btrfs_chunk
*chunk
;
2673 struct btrfs_key key
;
2675 mutex_lock(&fs_info
->chunk_mutex
);
2676 array_size
= btrfs_super_sys_array_size(super_copy
);
2678 ptr
= super_copy
->sys_chunk_array
;
2681 while (cur
< array_size
) {
2682 disk_key
= (struct btrfs_disk_key
*)ptr
;
2683 btrfs_disk_key_to_cpu(&key
, disk_key
);
2685 len
= sizeof(*disk_key
);
2687 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2688 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2689 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2690 len
+= btrfs_chunk_item_size(num_stripes
);
2695 if (key
.objectid
== BTRFS_FIRST_CHUNK_TREE_OBJECTID
&&
2696 key
.offset
== chunk_offset
) {
2697 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2699 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2705 mutex_unlock(&fs_info
->chunk_mutex
);
2709 static struct extent_map
*get_chunk_map(struct btrfs_fs_info
*fs_info
,
2710 u64 logical
, u64 length
)
2712 struct extent_map_tree
*em_tree
;
2713 struct extent_map
*em
;
2715 em_tree
= &fs_info
->mapping_tree
.map_tree
;
2716 read_lock(&em_tree
->lock
);
2717 em
= lookup_extent_mapping(em_tree
, logical
, length
);
2718 read_unlock(&em_tree
->lock
);
2721 btrfs_crit(fs_info
, "unable to find logical %llu length %llu",
2723 return ERR_PTR(-EINVAL
);
2726 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
2728 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2729 logical
, length
, em
->start
, em
->start
+ em
->len
);
2730 free_extent_map(em
);
2731 return ERR_PTR(-EINVAL
);
2734 /* callers are responsible for dropping em's ref. */
2738 int btrfs_remove_chunk(struct btrfs_trans_handle
*trans
, u64 chunk_offset
)
2740 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2741 struct extent_map
*em
;
2742 struct map_lookup
*map
;
2743 u64 dev_extent_len
= 0;
2745 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2747 em
= get_chunk_map(fs_info
, chunk_offset
, 1);
2750 * This is a logic error, but we don't want to just rely on the
2751 * user having built with ASSERT enabled, so if ASSERT doesn't
2752 * do anything we still error out.
2757 map
= em
->map_lookup
;
2758 mutex_lock(&fs_info
->chunk_mutex
);
2759 check_system_chunk(trans
, map
->type
);
2760 mutex_unlock(&fs_info
->chunk_mutex
);
2763 * Take the device list mutex to prevent races with the final phase of
2764 * a device replace operation that replaces the device object associated
2765 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2767 mutex_lock(&fs_devices
->device_list_mutex
);
2768 for (i
= 0; i
< map
->num_stripes
; i
++) {
2769 struct btrfs_device
*device
= map
->stripes
[i
].dev
;
2770 ret
= btrfs_free_dev_extent(trans
, device
,
2771 map
->stripes
[i
].physical
,
2774 mutex_unlock(&fs_devices
->device_list_mutex
);
2775 btrfs_abort_transaction(trans
, ret
);
2779 if (device
->bytes_used
> 0) {
2780 mutex_lock(&fs_info
->chunk_mutex
);
2781 btrfs_device_set_bytes_used(device
,
2782 device
->bytes_used
- dev_extent_len
);
2783 atomic64_add(dev_extent_len
, &fs_info
->free_chunk_space
);
2784 btrfs_clear_space_info_full(fs_info
);
2785 mutex_unlock(&fs_info
->chunk_mutex
);
2788 if (map
->stripes
[i
].dev
) {
2789 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
2791 mutex_unlock(&fs_devices
->device_list_mutex
);
2792 btrfs_abort_transaction(trans
, ret
);
2797 mutex_unlock(&fs_devices
->device_list_mutex
);
2799 ret
= btrfs_free_chunk(trans
, chunk_offset
);
2801 btrfs_abort_transaction(trans
, ret
);
2805 trace_btrfs_chunk_free(fs_info
, map
, chunk_offset
, em
->len
);
2807 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2808 ret
= btrfs_del_sys_chunk(fs_info
, chunk_offset
);
2810 btrfs_abort_transaction(trans
, ret
);
2815 ret
= btrfs_remove_block_group(trans
, chunk_offset
, em
);
2817 btrfs_abort_transaction(trans
, ret
);
2823 free_extent_map(em
);
2827 static int btrfs_relocate_chunk(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2829 struct btrfs_root
*root
= fs_info
->chunk_root
;
2830 struct btrfs_trans_handle
*trans
;
2834 * Prevent races with automatic removal of unused block groups.
2835 * After we relocate and before we remove the chunk with offset
2836 * chunk_offset, automatic removal of the block group can kick in,
2837 * resulting in a failure when calling btrfs_remove_chunk() below.
2839 * Make sure to acquire this mutex before doing a tree search (dev
2840 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2841 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2842 * we release the path used to search the chunk/dev tree and before
2843 * the current task acquires this mutex and calls us.
2845 lockdep_assert_held(&fs_info
->delete_unused_bgs_mutex
);
2847 ret
= btrfs_can_relocate(fs_info
, chunk_offset
);
2851 /* step one, relocate all the extents inside this chunk */
2852 btrfs_scrub_pause(fs_info
);
2853 ret
= btrfs_relocate_block_group(fs_info
, chunk_offset
);
2854 btrfs_scrub_continue(fs_info
);
2859 * We add the kobjects here (and after forcing data chunk creation)
2860 * since relocation is the only place we'll create chunks of a new
2861 * type at runtime. The only place where we'll remove the last
2862 * chunk of a type is the call immediately below this one. Even
2863 * so, we're protected against races with the cleaner thread since
2864 * we're covered by the delete_unused_bgs_mutex.
2866 btrfs_add_raid_kobjects(fs_info
);
2868 trans
= btrfs_start_trans_remove_block_group(root
->fs_info
,
2870 if (IS_ERR(trans
)) {
2871 ret
= PTR_ERR(trans
);
2872 btrfs_handle_fs_error(root
->fs_info
, ret
, NULL
);
2877 * step two, delete the device extents and the
2878 * chunk tree entries
2880 ret
= btrfs_remove_chunk(trans
, chunk_offset
);
2881 btrfs_end_transaction(trans
);
2885 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info
*fs_info
)
2887 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
2888 struct btrfs_path
*path
;
2889 struct extent_buffer
*leaf
;
2890 struct btrfs_chunk
*chunk
;
2891 struct btrfs_key key
;
2892 struct btrfs_key found_key
;
2894 bool retried
= false;
2898 path
= btrfs_alloc_path();
2903 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2904 key
.offset
= (u64
)-1;
2905 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2908 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
2909 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
2911 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
2914 BUG_ON(ret
== 0); /* Corruption */
2916 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
2919 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
2925 leaf
= path
->nodes
[0];
2926 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2928 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
2929 struct btrfs_chunk
);
2930 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
2931 btrfs_release_path(path
);
2933 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
2934 ret
= btrfs_relocate_chunk(fs_info
, found_key
.offset
);
2940 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
2942 if (found_key
.offset
== 0)
2944 key
.offset
= found_key
.offset
- 1;
2947 if (failed
&& !retried
) {
2951 } else if (WARN_ON(failed
&& retried
)) {
2955 btrfs_free_path(path
);
2960 * return 1 : allocate a data chunk successfully,
2961 * return <0: errors during allocating a data chunk,
2962 * return 0 : no need to allocate a data chunk.
2964 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info
*fs_info
,
2967 struct btrfs_block_group_cache
*cache
;
2971 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
2973 chunk_type
= cache
->flags
;
2974 btrfs_put_block_group(cache
);
2976 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
) {
2977 spin_lock(&fs_info
->data_sinfo
->lock
);
2978 bytes_used
= fs_info
->data_sinfo
->bytes_used
;
2979 spin_unlock(&fs_info
->data_sinfo
->lock
);
2982 struct btrfs_trans_handle
*trans
;
2985 trans
= btrfs_join_transaction(fs_info
->tree_root
);
2987 return PTR_ERR(trans
);
2989 ret
= btrfs_force_chunk_alloc(trans
,
2990 BTRFS_BLOCK_GROUP_DATA
);
2991 btrfs_end_transaction(trans
);
2995 btrfs_add_raid_kobjects(fs_info
);
3003 static int insert_balance_item(struct btrfs_fs_info
*fs_info
,
3004 struct btrfs_balance_control
*bctl
)
3006 struct btrfs_root
*root
= fs_info
->tree_root
;
3007 struct btrfs_trans_handle
*trans
;
3008 struct btrfs_balance_item
*item
;
3009 struct btrfs_disk_balance_args disk_bargs
;
3010 struct btrfs_path
*path
;
3011 struct extent_buffer
*leaf
;
3012 struct btrfs_key key
;
3015 path
= btrfs_alloc_path();
3019 trans
= btrfs_start_transaction(root
, 0);
3020 if (IS_ERR(trans
)) {
3021 btrfs_free_path(path
);
3022 return PTR_ERR(trans
);
3025 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3026 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3029 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
3034 leaf
= path
->nodes
[0];
3035 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3037 memzero_extent_buffer(leaf
, (unsigned long)item
, sizeof(*item
));
3039 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
3040 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
3041 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
3042 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
3043 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
3044 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
3046 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
3048 btrfs_mark_buffer_dirty(leaf
);
3050 btrfs_free_path(path
);
3051 err
= btrfs_commit_transaction(trans
);
3057 static int del_balance_item(struct btrfs_fs_info
*fs_info
)
3059 struct btrfs_root
*root
= fs_info
->tree_root
;
3060 struct btrfs_trans_handle
*trans
;
3061 struct btrfs_path
*path
;
3062 struct btrfs_key key
;
3065 path
= btrfs_alloc_path();
3069 trans
= btrfs_start_transaction(root
, 0);
3070 if (IS_ERR(trans
)) {
3071 btrfs_free_path(path
);
3072 return PTR_ERR(trans
);
3075 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3076 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3079 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3087 ret
= btrfs_del_item(trans
, root
, path
);
3089 btrfs_free_path(path
);
3090 err
= btrfs_commit_transaction(trans
);
3097 * This is a heuristic used to reduce the number of chunks balanced on
3098 * resume after balance was interrupted.
3100 static void update_balance_args(struct btrfs_balance_control
*bctl
)
3103 * Turn on soft mode for chunk types that were being converted.
3105 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3106 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3107 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3108 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3109 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3110 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3113 * Turn on usage filter if is not already used. The idea is
3114 * that chunks that we have already balanced should be
3115 * reasonably full. Don't do it for chunks that are being
3116 * converted - that will keep us from relocating unconverted
3117 * (albeit full) chunks.
3119 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3120 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3121 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3122 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3123 bctl
->data
.usage
= 90;
3125 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3126 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3127 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3128 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3129 bctl
->sys
.usage
= 90;
3131 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3132 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3133 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3134 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3135 bctl
->meta
.usage
= 90;
3140 * Clear the balance status in fs_info and delete the balance item from disk.
3142 static void reset_balance_state(struct btrfs_fs_info
*fs_info
)
3144 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3147 BUG_ON(!fs_info
->balance_ctl
);
3149 spin_lock(&fs_info
->balance_lock
);
3150 fs_info
->balance_ctl
= NULL
;
3151 spin_unlock(&fs_info
->balance_lock
);
3154 ret
= del_balance_item(fs_info
);
3156 btrfs_handle_fs_error(fs_info
, ret
, NULL
);
3160 * Balance filters. Return 1 if chunk should be filtered out
3161 * (should not be balanced).
3163 static int chunk_profiles_filter(u64 chunk_type
,
3164 struct btrfs_balance_args
*bargs
)
3166 chunk_type
= chunk_to_extended(chunk_type
) &
3167 BTRFS_EXTENDED_PROFILE_MASK
;
3169 if (bargs
->profiles
& chunk_type
)
3175 static int chunk_usage_range_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
3176 struct btrfs_balance_args
*bargs
)
3178 struct btrfs_block_group_cache
*cache
;
3180 u64 user_thresh_min
;
3181 u64 user_thresh_max
;
3184 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3185 chunk_used
= btrfs_block_group_used(&cache
->item
);
3187 if (bargs
->usage_min
== 0)
3188 user_thresh_min
= 0;
3190 user_thresh_min
= div_factor_fine(cache
->key
.offset
,
3193 if (bargs
->usage_max
== 0)
3194 user_thresh_max
= 1;
3195 else if (bargs
->usage_max
> 100)
3196 user_thresh_max
= cache
->key
.offset
;
3198 user_thresh_max
= div_factor_fine(cache
->key
.offset
,
3201 if (user_thresh_min
<= chunk_used
&& chunk_used
< user_thresh_max
)
3204 btrfs_put_block_group(cache
);
3208 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
,
3209 u64 chunk_offset
, struct btrfs_balance_args
*bargs
)
3211 struct btrfs_block_group_cache
*cache
;
3212 u64 chunk_used
, user_thresh
;
3215 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3216 chunk_used
= btrfs_block_group_used(&cache
->item
);
3218 if (bargs
->usage_min
== 0)
3220 else if (bargs
->usage
> 100)
3221 user_thresh
= cache
->key
.offset
;
3223 user_thresh
= div_factor_fine(cache
->key
.offset
,
3226 if (chunk_used
< user_thresh
)
3229 btrfs_put_block_group(cache
);
3233 static int chunk_devid_filter(struct extent_buffer
*leaf
,
3234 struct btrfs_chunk
*chunk
,
3235 struct btrfs_balance_args
*bargs
)
3237 struct btrfs_stripe
*stripe
;
3238 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3241 for (i
= 0; i
< num_stripes
; i
++) {
3242 stripe
= btrfs_stripe_nr(chunk
, i
);
3243 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
3250 /* [pstart, pend) */
3251 static int chunk_drange_filter(struct extent_buffer
*leaf
,
3252 struct btrfs_chunk
*chunk
,
3253 struct btrfs_balance_args
*bargs
)
3255 struct btrfs_stripe
*stripe
;
3256 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3262 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
3265 if (btrfs_chunk_type(leaf
, chunk
) & (BTRFS_BLOCK_GROUP_DUP
|
3266 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)) {
3267 factor
= num_stripes
/ 2;
3268 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID5
) {
3269 factor
= num_stripes
- 1;
3270 } else if (btrfs_chunk_type(leaf
, chunk
) & BTRFS_BLOCK_GROUP_RAID6
) {
3271 factor
= num_stripes
- 2;
3273 factor
= num_stripes
;
3276 for (i
= 0; i
< num_stripes
; i
++) {
3277 stripe
= btrfs_stripe_nr(chunk
, i
);
3278 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
3281 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
3282 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
3283 stripe_length
= div_u64(stripe_length
, factor
);
3285 if (stripe_offset
< bargs
->pend
&&
3286 stripe_offset
+ stripe_length
> bargs
->pstart
)
3293 /* [vstart, vend) */
3294 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
3295 struct btrfs_chunk
*chunk
,
3297 struct btrfs_balance_args
*bargs
)
3299 if (chunk_offset
< bargs
->vend
&&
3300 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
3301 /* at least part of the chunk is inside this vrange */
3307 static int chunk_stripes_range_filter(struct extent_buffer
*leaf
,
3308 struct btrfs_chunk
*chunk
,
3309 struct btrfs_balance_args
*bargs
)
3311 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3313 if (bargs
->stripes_min
<= num_stripes
3314 && num_stripes
<= bargs
->stripes_max
)
3320 static int chunk_soft_convert_filter(u64 chunk_type
,
3321 struct btrfs_balance_args
*bargs
)
3323 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
3326 chunk_type
= chunk_to_extended(chunk_type
) &
3327 BTRFS_EXTENDED_PROFILE_MASK
;
3329 if (bargs
->target
== chunk_type
)
3335 static int should_balance_chunk(struct btrfs_fs_info
*fs_info
,
3336 struct extent_buffer
*leaf
,
3337 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
3339 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3340 struct btrfs_balance_args
*bargs
= NULL
;
3341 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3344 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
3345 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
3349 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3350 bargs
= &bctl
->data
;
3351 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3353 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3354 bargs
= &bctl
->meta
;
3356 /* profiles filter */
3357 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
3358 chunk_profiles_filter(chunk_type
, bargs
)) {
3363 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3364 chunk_usage_filter(fs_info
, chunk_offset
, bargs
)) {
3366 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3367 chunk_usage_range_filter(fs_info
, chunk_offset
, bargs
)) {
3372 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
3373 chunk_devid_filter(leaf
, chunk
, bargs
)) {
3377 /* drange filter, makes sense only with devid filter */
3378 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
3379 chunk_drange_filter(leaf
, chunk
, bargs
)) {
3384 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
3385 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3389 /* stripes filter */
3390 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_STRIPES_RANGE
) &&
3391 chunk_stripes_range_filter(leaf
, chunk
, bargs
)) {
3395 /* soft profile changing mode */
3396 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
3397 chunk_soft_convert_filter(chunk_type
, bargs
)) {
3402 * limited by count, must be the last filter
3404 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT
)) {
3405 if (bargs
->limit
== 0)
3409 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT_RANGE
)) {
3411 * Same logic as the 'limit' filter; the minimum cannot be
3412 * determined here because we do not have the global information
3413 * about the count of all chunks that satisfy the filters.
3415 if (bargs
->limit_max
== 0)
3424 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
3426 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3427 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
3428 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
3429 struct list_head
*devices
;
3430 struct btrfs_device
*device
;
3434 struct btrfs_chunk
*chunk
;
3435 struct btrfs_path
*path
= NULL
;
3436 struct btrfs_key key
;
3437 struct btrfs_key found_key
;
3438 struct btrfs_trans_handle
*trans
;
3439 struct extent_buffer
*leaf
;
3442 int enospc_errors
= 0;
3443 bool counting
= true;
3444 /* The single value limit and min/max limits use the same bytes in the */
3445 u64 limit_data
= bctl
->data
.limit
;
3446 u64 limit_meta
= bctl
->meta
.limit
;
3447 u64 limit_sys
= bctl
->sys
.limit
;
3451 int chunk_reserved
= 0;
3453 /* step one make some room on all the devices */
3454 devices
= &fs_info
->fs_devices
->devices
;
3455 list_for_each_entry(device
, devices
, dev_list
) {
3456 old_size
= btrfs_device_get_total_bytes(device
);
3457 size_to_free
= div_factor(old_size
, 1);
3458 size_to_free
= min_t(u64
, size_to_free
, SZ_1M
);
3459 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
) ||
3460 btrfs_device_get_total_bytes(device
) -
3461 btrfs_device_get_bytes_used(device
) > size_to_free
||
3462 test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
))
3465 ret
= btrfs_shrink_device(device
, old_size
- size_to_free
);
3469 /* btrfs_shrink_device never returns ret > 0 */
3474 trans
= btrfs_start_transaction(dev_root
, 0);
3475 if (IS_ERR(trans
)) {
3476 ret
= PTR_ERR(trans
);
3477 btrfs_info_in_rcu(fs_info
,
3478 "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu",
3479 rcu_str_deref(device
->name
), ret
,
3480 old_size
, old_size
- size_to_free
);
3484 ret
= btrfs_grow_device(trans
, device
, old_size
);
3486 btrfs_end_transaction(trans
);
3487 /* btrfs_grow_device never returns ret > 0 */
3489 btrfs_info_in_rcu(fs_info
,
3490 "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu",
3491 rcu_str_deref(device
->name
), ret
,
3492 old_size
, old_size
- size_to_free
);
3496 btrfs_end_transaction(trans
);
3499 /* step two, relocate all the chunks */
3500 path
= btrfs_alloc_path();
3506 /* zero out stat counters */
3507 spin_lock(&fs_info
->balance_lock
);
3508 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
3509 spin_unlock(&fs_info
->balance_lock
);
3513 * The single value limit and min/max limits use the same bytes
3516 bctl
->data
.limit
= limit_data
;
3517 bctl
->meta
.limit
= limit_meta
;
3518 bctl
->sys
.limit
= limit_sys
;
3520 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3521 key
.offset
= (u64
)-1;
3522 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3525 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
3526 atomic_read(&fs_info
->balance_cancel_req
)) {
3531 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
3532 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
3534 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3539 * this shouldn't happen, it means the last relocate
3543 BUG(); /* FIXME break ? */
3545 ret
= btrfs_previous_item(chunk_root
, path
, 0,
3546 BTRFS_CHUNK_ITEM_KEY
);
3548 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3553 leaf
= path
->nodes
[0];
3554 slot
= path
->slots
[0];
3555 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3557 if (found_key
.objectid
!= key
.objectid
) {
3558 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3562 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
3563 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3566 spin_lock(&fs_info
->balance_lock
);
3567 bctl
->stat
.considered
++;
3568 spin_unlock(&fs_info
->balance_lock
);
3571 ret
= should_balance_chunk(fs_info
, leaf
, chunk
,
3574 btrfs_release_path(path
);
3576 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3581 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3582 spin_lock(&fs_info
->balance_lock
);
3583 bctl
->stat
.expected
++;
3584 spin_unlock(&fs_info
->balance_lock
);
3586 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3588 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3590 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3597 * Apply limit_min filter, no need to check if the LIMITS
3598 * filter is used, limit_min is 0 by default
3600 if (((chunk_type
& BTRFS_BLOCK_GROUP_DATA
) &&
3601 count_data
< bctl
->data
.limit_min
)
3602 || ((chunk_type
& BTRFS_BLOCK_GROUP_METADATA
) &&
3603 count_meta
< bctl
->meta
.limit_min
)
3604 || ((chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) &&
3605 count_sys
< bctl
->sys
.limit_min
)) {
3606 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3610 if (!chunk_reserved
) {
3612 * We may be relocating the only data chunk we have,
3613 * which could potentially end up with losing data's
3614 * raid profile, so lets allocate an empty one in
3617 ret
= btrfs_may_alloc_data_chunk(fs_info
,
3620 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3622 } else if (ret
== 1) {
3627 ret
= btrfs_relocate_chunk(fs_info
, found_key
.offset
);
3628 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
3629 if (ret
&& ret
!= -ENOSPC
)
3631 if (ret
== -ENOSPC
) {
3634 spin_lock(&fs_info
->balance_lock
);
3635 bctl
->stat
.completed
++;
3636 spin_unlock(&fs_info
->balance_lock
);
3639 if (found_key
.offset
== 0)
3641 key
.offset
= found_key
.offset
- 1;
3645 btrfs_release_path(path
);
3650 btrfs_free_path(path
);
3651 if (enospc_errors
) {
3652 btrfs_info(fs_info
, "%d enospc errors during balance",
3662 * alloc_profile_is_valid - see if a given profile is valid and reduced
3663 * @flags: profile to validate
3664 * @extended: if true @flags is treated as an extended profile
3666 static int alloc_profile_is_valid(u64 flags
, int extended
)
3668 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
3669 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
3671 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
3673 /* 1) check that all other bits are zeroed */
3677 /* 2) see if profile is reduced */
3679 return !extended
; /* "0" is valid for usual profiles */
3681 /* true if exactly one bit set */
3682 return (flags
& (flags
- 1)) == 0;
3685 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
3687 /* cancel requested || normal exit path */
3688 return atomic_read(&fs_info
->balance_cancel_req
) ||
3689 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
3690 atomic_read(&fs_info
->balance_cancel_req
) == 0);
3693 /* Non-zero return value signifies invalidity */
3694 static inline int validate_convert_profile(struct btrfs_balance_args
*bctl_arg
,
3697 return ((bctl_arg
->flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3698 (!alloc_profile_is_valid(bctl_arg
->target
, 1) ||
3699 (bctl_arg
->target
& ~allowed
)));
3703 * Should be called with balance mutexe held
3705 int btrfs_balance(struct btrfs_fs_info
*fs_info
,
3706 struct btrfs_balance_control
*bctl
,
3707 struct btrfs_ioctl_balance_args
*bargs
)
3709 u64 meta_target
, data_target
;
3716 if (btrfs_fs_closing(fs_info
) ||
3717 atomic_read(&fs_info
->balance_pause_req
) ||
3718 atomic_read(&fs_info
->balance_cancel_req
)) {
3723 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
3724 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
3728 * In case of mixed groups both data and meta should be picked,
3729 * and identical options should be given for both of them.
3731 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
3732 if (mixed
&& (bctl
->flags
& allowed
)) {
3733 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
3734 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
3735 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
3737 "balance: mixed groups data and metadata options must be the same");
3743 num_devices
= fs_info
->fs_devices
->num_devices
;
3744 btrfs_dev_replace_read_lock(&fs_info
->dev_replace
);
3745 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
3746 BUG_ON(num_devices
< 1);
3749 btrfs_dev_replace_read_unlock(&fs_info
->dev_replace
);
3750 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
| BTRFS_BLOCK_GROUP_DUP
;
3751 if (num_devices
> 1)
3752 allowed
|= (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
);
3753 if (num_devices
> 2)
3754 allowed
|= BTRFS_BLOCK_GROUP_RAID5
;
3755 if (num_devices
> 3)
3756 allowed
|= (BTRFS_BLOCK_GROUP_RAID10
|
3757 BTRFS_BLOCK_GROUP_RAID6
);
3758 if (validate_convert_profile(&bctl
->data
, allowed
)) {
3759 int index
= btrfs_bg_flags_to_raid_index(bctl
->data
.target
);
3762 "balance: invalid convert data profile %s",
3763 get_raid_name(index
));
3767 if (validate_convert_profile(&bctl
->meta
, allowed
)) {
3768 int index
= btrfs_bg_flags_to_raid_index(bctl
->meta
.target
);
3771 "balance: invalid convert metadata profile %s",
3772 get_raid_name(index
));
3776 if (validate_convert_profile(&bctl
->sys
, allowed
)) {
3777 int index
= btrfs_bg_flags_to_raid_index(bctl
->sys
.target
);
3780 "balance: invalid convert system profile %s",
3781 get_raid_name(index
));
3786 /* allow to reduce meta or sys integrity only if force set */
3787 allowed
= BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
3788 BTRFS_BLOCK_GROUP_RAID10
|
3789 BTRFS_BLOCK_GROUP_RAID5
|
3790 BTRFS_BLOCK_GROUP_RAID6
;
3792 seq
= read_seqbegin(&fs_info
->profiles_lock
);
3794 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3795 (fs_info
->avail_system_alloc_bits
& allowed
) &&
3796 !(bctl
->sys
.target
& allowed
)) ||
3797 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
3798 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
3799 !(bctl
->meta
.target
& allowed
))) {
3800 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
3802 "balance: force reducing metadata integrity");
3805 "balance: reduces metadata integrity, use --force if you want this");
3810 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
3812 /* if we're not converting, the target field is uninitialized */
3813 meta_target
= (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) ?
3814 bctl
->meta
.target
: fs_info
->avail_metadata_alloc_bits
;
3815 data_target
= (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) ?
3816 bctl
->data
.target
: fs_info
->avail_data_alloc_bits
;
3817 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target
) <
3818 btrfs_get_num_tolerated_disk_barrier_failures(data_target
)) {
3819 int meta_index
= btrfs_bg_flags_to_raid_index(meta_target
);
3820 int data_index
= btrfs_bg_flags_to_raid_index(data_target
);
3823 "balance: metadata profile %s has lower redundancy than data profile %s",
3824 get_raid_name(meta_index
), get_raid_name(data_index
));
3827 ret
= insert_balance_item(fs_info
, bctl
);
3828 if (ret
&& ret
!= -EEXIST
)
3831 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
3832 BUG_ON(ret
== -EEXIST
);
3833 BUG_ON(fs_info
->balance_ctl
);
3834 spin_lock(&fs_info
->balance_lock
);
3835 fs_info
->balance_ctl
= bctl
;
3836 spin_unlock(&fs_info
->balance_lock
);
3838 BUG_ON(ret
!= -EEXIST
);
3839 spin_lock(&fs_info
->balance_lock
);
3840 update_balance_args(bctl
);
3841 spin_unlock(&fs_info
->balance_lock
);
3844 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
));
3845 set_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
);
3846 mutex_unlock(&fs_info
->balance_mutex
);
3848 ret
= __btrfs_balance(fs_info
);
3850 mutex_lock(&fs_info
->balance_mutex
);
3851 clear_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
);
3854 memset(bargs
, 0, sizeof(*bargs
));
3855 btrfs_update_ioctl_balance_args(fs_info
, bargs
);
3858 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
3859 balance_need_close(fs_info
)) {
3860 reset_balance_state(fs_info
);
3861 clear_bit(BTRFS_FS_EXCL_OP
, &fs_info
->flags
);
3864 wake_up(&fs_info
->balance_wait_q
);
3868 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
3869 reset_balance_state(fs_info
);
3872 clear_bit(BTRFS_FS_EXCL_OP
, &fs_info
->flags
);
3877 static int balance_kthread(void *data
)
3879 struct btrfs_fs_info
*fs_info
= data
;
3882 mutex_lock(&fs_info
->balance_mutex
);
3883 if (fs_info
->balance_ctl
) {
3884 btrfs_info(fs_info
, "balance: resuming");
3885 ret
= btrfs_balance(fs_info
, fs_info
->balance_ctl
, NULL
);
3887 mutex_unlock(&fs_info
->balance_mutex
);
3892 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
3894 struct task_struct
*tsk
;
3896 mutex_lock(&fs_info
->balance_mutex
);
3897 if (!fs_info
->balance_ctl
) {
3898 mutex_unlock(&fs_info
->balance_mutex
);
3901 mutex_unlock(&fs_info
->balance_mutex
);
3903 if (btrfs_test_opt(fs_info
, SKIP_BALANCE
)) {
3904 btrfs_info(fs_info
, "balance: resume skipped");
3909 * A ro->rw remount sequence should continue with the paused balance
3910 * regardless of who pauses it, system or the user as of now, so set
3913 spin_lock(&fs_info
->balance_lock
);
3914 fs_info
->balance_ctl
->flags
|= BTRFS_BALANCE_RESUME
;
3915 spin_unlock(&fs_info
->balance_lock
);
3917 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
3918 return PTR_ERR_OR_ZERO(tsk
);
3921 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
3923 struct btrfs_balance_control
*bctl
;
3924 struct btrfs_balance_item
*item
;
3925 struct btrfs_disk_balance_args disk_bargs
;
3926 struct btrfs_path
*path
;
3927 struct extent_buffer
*leaf
;
3928 struct btrfs_key key
;
3931 path
= btrfs_alloc_path();
3935 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3936 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3939 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
3942 if (ret
> 0) { /* ret = -ENOENT; */
3947 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
3953 leaf
= path
->nodes
[0];
3954 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3956 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
3957 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
3959 btrfs_balance_data(leaf
, item
, &disk_bargs
);
3960 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
3961 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
3962 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
3963 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
3964 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
3967 * This should never happen, as the paused balance state is recovered
3968 * during mount without any chance of other exclusive ops to collide.
3970 * This gives the exclusive op status to balance and keeps in paused
3971 * state until user intervention (cancel or umount). If the ownership
3972 * cannot be assigned, show a message but do not fail. The balance
3973 * is in a paused state and must have fs_info::balance_ctl properly
3976 if (test_and_set_bit(BTRFS_FS_EXCL_OP
, &fs_info
->flags
))
3978 "balance: cannot set exclusive op status, resume manually");
3980 mutex_lock(&fs_info
->balance_mutex
);
3981 BUG_ON(fs_info
->balance_ctl
);
3982 spin_lock(&fs_info
->balance_lock
);
3983 fs_info
->balance_ctl
= bctl
;
3984 spin_unlock(&fs_info
->balance_lock
);
3985 mutex_unlock(&fs_info
->balance_mutex
);
3987 btrfs_free_path(path
);
3991 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
3995 mutex_lock(&fs_info
->balance_mutex
);
3996 if (!fs_info
->balance_ctl
) {
3997 mutex_unlock(&fs_info
->balance_mutex
);
4001 if (test_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
)) {
4002 atomic_inc(&fs_info
->balance_pause_req
);
4003 mutex_unlock(&fs_info
->balance_mutex
);
4005 wait_event(fs_info
->balance_wait_q
,
4006 !test_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
));
4008 mutex_lock(&fs_info
->balance_mutex
);
4009 /* we are good with balance_ctl ripped off from under us */
4010 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
));
4011 atomic_dec(&fs_info
->balance_pause_req
);
4016 mutex_unlock(&fs_info
->balance_mutex
);
4020 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
4022 mutex_lock(&fs_info
->balance_mutex
);
4023 if (!fs_info
->balance_ctl
) {
4024 mutex_unlock(&fs_info
->balance_mutex
);
4029 * A paused balance with the item stored on disk can be resumed at
4030 * mount time if the mount is read-write. Otherwise it's still paused
4031 * and we must not allow cancelling as it deletes the item.
4033 if (sb_rdonly(fs_info
->sb
)) {
4034 mutex_unlock(&fs_info
->balance_mutex
);
4038 atomic_inc(&fs_info
->balance_cancel_req
);
4040 * if we are running just wait and return, balance item is
4041 * deleted in btrfs_balance in this case
4043 if (test_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
)) {
4044 mutex_unlock(&fs_info
->balance_mutex
);
4045 wait_event(fs_info
->balance_wait_q
,
4046 !test_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
));
4047 mutex_lock(&fs_info
->balance_mutex
);
4049 mutex_unlock(&fs_info
->balance_mutex
);
4051 * Lock released to allow other waiters to continue, we'll
4052 * reexamine the status again.
4054 mutex_lock(&fs_info
->balance_mutex
);
4056 if (fs_info
->balance_ctl
) {
4057 reset_balance_state(fs_info
);
4058 clear_bit(BTRFS_FS_EXCL_OP
, &fs_info
->flags
);
4059 btrfs_info(fs_info
, "balance: canceled");
4063 BUG_ON(fs_info
->balance_ctl
||
4064 test_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
));
4065 atomic_dec(&fs_info
->balance_cancel_req
);
4066 mutex_unlock(&fs_info
->balance_mutex
);
4070 static int btrfs_uuid_scan_kthread(void *data
)
4072 struct btrfs_fs_info
*fs_info
= data
;
4073 struct btrfs_root
*root
= fs_info
->tree_root
;
4074 struct btrfs_key key
;
4075 struct btrfs_path
*path
= NULL
;
4077 struct extent_buffer
*eb
;
4079 struct btrfs_root_item root_item
;
4081 struct btrfs_trans_handle
*trans
= NULL
;
4083 path
= btrfs_alloc_path();
4090 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4094 ret
= btrfs_search_forward(root
, &key
, path
,
4095 BTRFS_OLDEST_GENERATION
);
4102 if (key
.type
!= BTRFS_ROOT_ITEM_KEY
||
4103 (key
.objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
4104 key
.objectid
!= BTRFS_FS_TREE_OBJECTID
) ||
4105 key
.objectid
> BTRFS_LAST_FREE_OBJECTID
)
4108 eb
= path
->nodes
[0];
4109 slot
= path
->slots
[0];
4110 item_size
= btrfs_item_size_nr(eb
, slot
);
4111 if (item_size
< sizeof(root_item
))
4114 read_extent_buffer(eb
, &root_item
,
4115 btrfs_item_ptr_offset(eb
, slot
),
4116 (int)sizeof(root_item
));
4117 if (btrfs_root_refs(&root_item
) == 0)
4120 if (!btrfs_is_empty_uuid(root_item
.uuid
) ||
4121 !btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4125 btrfs_release_path(path
);
4127 * 1 - subvol uuid item
4128 * 1 - received_subvol uuid item
4130 trans
= btrfs_start_transaction(fs_info
->uuid_root
, 2);
4131 if (IS_ERR(trans
)) {
4132 ret
= PTR_ERR(trans
);
4140 if (!btrfs_is_empty_uuid(root_item
.uuid
)) {
4141 ret
= btrfs_uuid_tree_add(trans
, root_item
.uuid
,
4142 BTRFS_UUID_KEY_SUBVOL
,
4145 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4151 if (!btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4152 ret
= btrfs_uuid_tree_add(trans
,
4153 root_item
.received_uuid
,
4154 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
4157 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4165 ret
= btrfs_end_transaction(trans
);
4171 btrfs_release_path(path
);
4172 if (key
.offset
< (u64
)-1) {
4174 } else if (key
.type
< BTRFS_ROOT_ITEM_KEY
) {
4176 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4177 } else if (key
.objectid
< (u64
)-1) {
4179 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4188 btrfs_free_path(path
);
4189 if (trans
&& !IS_ERR(trans
))
4190 btrfs_end_transaction(trans
);
4192 btrfs_warn(fs_info
, "btrfs_uuid_scan_kthread failed %d", ret
);
4194 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN
, &fs_info
->flags
);
4195 up(&fs_info
->uuid_tree_rescan_sem
);
4200 * Callback for btrfs_uuid_tree_iterate().
4202 * 0 check succeeded, the entry is not outdated.
4203 * < 0 if an error occurred.
4204 * > 0 if the check failed, which means the caller shall remove the entry.
4206 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info
*fs_info
,
4207 u8
*uuid
, u8 type
, u64 subid
)
4209 struct btrfs_key key
;
4211 struct btrfs_root
*subvol_root
;
4213 if (type
!= BTRFS_UUID_KEY_SUBVOL
&&
4214 type
!= BTRFS_UUID_KEY_RECEIVED_SUBVOL
)
4217 key
.objectid
= subid
;
4218 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4219 key
.offset
= (u64
)-1;
4220 subvol_root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
4221 if (IS_ERR(subvol_root
)) {
4222 ret
= PTR_ERR(subvol_root
);
4229 case BTRFS_UUID_KEY_SUBVOL
:
4230 if (memcmp(uuid
, subvol_root
->root_item
.uuid
, BTRFS_UUID_SIZE
))
4233 case BTRFS_UUID_KEY_RECEIVED_SUBVOL
:
4234 if (memcmp(uuid
, subvol_root
->root_item
.received_uuid
,
4244 static int btrfs_uuid_rescan_kthread(void *data
)
4246 struct btrfs_fs_info
*fs_info
= (struct btrfs_fs_info
*)data
;
4250 * 1st step is to iterate through the existing UUID tree and
4251 * to delete all entries that contain outdated data.
4252 * 2nd step is to add all missing entries to the UUID tree.
4254 ret
= btrfs_uuid_tree_iterate(fs_info
, btrfs_check_uuid_tree_entry
);
4256 btrfs_warn(fs_info
, "iterating uuid_tree failed %d", ret
);
4257 up(&fs_info
->uuid_tree_rescan_sem
);
4260 return btrfs_uuid_scan_kthread(data
);
4263 int btrfs_create_uuid_tree(struct btrfs_fs_info
*fs_info
)
4265 struct btrfs_trans_handle
*trans
;
4266 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
4267 struct btrfs_root
*uuid_root
;
4268 struct task_struct
*task
;
4275 trans
= btrfs_start_transaction(tree_root
, 2);
4277 return PTR_ERR(trans
);
4279 uuid_root
= btrfs_create_tree(trans
, fs_info
,
4280 BTRFS_UUID_TREE_OBJECTID
);
4281 if (IS_ERR(uuid_root
)) {
4282 ret
= PTR_ERR(uuid_root
);
4283 btrfs_abort_transaction(trans
, ret
);
4284 btrfs_end_transaction(trans
);
4288 fs_info
->uuid_root
= uuid_root
;
4290 ret
= btrfs_commit_transaction(trans
);
4294 down(&fs_info
->uuid_tree_rescan_sem
);
4295 task
= kthread_run(btrfs_uuid_scan_kthread
, fs_info
, "btrfs-uuid");
4297 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4298 btrfs_warn(fs_info
, "failed to start uuid_scan task");
4299 up(&fs_info
->uuid_tree_rescan_sem
);
4300 return PTR_ERR(task
);
4306 int btrfs_check_uuid_tree(struct btrfs_fs_info
*fs_info
)
4308 struct task_struct
*task
;
4310 down(&fs_info
->uuid_tree_rescan_sem
);
4311 task
= kthread_run(btrfs_uuid_rescan_kthread
, fs_info
, "btrfs-uuid");
4313 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4314 btrfs_warn(fs_info
, "failed to start uuid_rescan task");
4315 up(&fs_info
->uuid_tree_rescan_sem
);
4316 return PTR_ERR(task
);
4323 * shrinking a device means finding all of the device extents past
4324 * the new size, and then following the back refs to the chunks.
4325 * The chunk relocation code actually frees the device extent
4327 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
4329 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
4330 struct btrfs_root
*root
= fs_info
->dev_root
;
4331 struct btrfs_trans_handle
*trans
;
4332 struct btrfs_dev_extent
*dev_extent
= NULL
;
4333 struct btrfs_path
*path
;
4339 bool retried
= false;
4340 bool checked_pending_chunks
= false;
4341 struct extent_buffer
*l
;
4342 struct btrfs_key key
;
4343 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
4344 u64 old_total
= btrfs_super_total_bytes(super_copy
);
4345 u64 old_size
= btrfs_device_get_total_bytes(device
);
4348 new_size
= round_down(new_size
, fs_info
->sectorsize
);
4349 diff
= round_down(old_size
- new_size
, fs_info
->sectorsize
);
4351 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
))
4354 path
= btrfs_alloc_path();
4358 path
->reada
= READA_BACK
;
4360 mutex_lock(&fs_info
->chunk_mutex
);
4362 btrfs_device_set_total_bytes(device
, new_size
);
4363 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
)) {
4364 device
->fs_devices
->total_rw_bytes
-= diff
;
4365 atomic64_sub(diff
, &fs_info
->free_chunk_space
);
4367 mutex_unlock(&fs_info
->chunk_mutex
);
4370 key
.objectid
= device
->devid
;
4371 key
.offset
= (u64
)-1;
4372 key
.type
= BTRFS_DEV_EXTENT_KEY
;
4375 mutex_lock(&fs_info
->delete_unused_bgs_mutex
);
4376 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4378 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4382 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
4384 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4389 btrfs_release_path(path
);
4394 slot
= path
->slots
[0];
4395 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
4397 if (key
.objectid
!= device
->devid
) {
4398 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4399 btrfs_release_path(path
);
4403 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
4404 length
= btrfs_dev_extent_length(l
, dev_extent
);
4406 if (key
.offset
+ length
<= new_size
) {
4407 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4408 btrfs_release_path(path
);
4412 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
4413 btrfs_release_path(path
);
4416 * We may be relocating the only data chunk we have,
4417 * which could potentially end up with losing data's
4418 * raid profile, so lets allocate an empty one in
4421 ret
= btrfs_may_alloc_data_chunk(fs_info
, chunk_offset
);
4423 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4427 ret
= btrfs_relocate_chunk(fs_info
, chunk_offset
);
4428 mutex_unlock(&fs_info
->delete_unused_bgs_mutex
);
4429 if (ret
&& ret
!= -ENOSPC
)
4433 } while (key
.offset
-- > 0);
4435 if (failed
&& !retried
) {
4439 } else if (failed
&& retried
) {
4444 /* Shrinking succeeded, else we would be at "done". */
4445 trans
= btrfs_start_transaction(root
, 0);
4446 if (IS_ERR(trans
)) {
4447 ret
= PTR_ERR(trans
);
4451 mutex_lock(&fs_info
->chunk_mutex
);
4454 * We checked in the above loop all device extents that were already in
4455 * the device tree. However before we have updated the device's
4456 * total_bytes to the new size, we might have had chunk allocations that
4457 * have not complete yet (new block groups attached to transaction
4458 * handles), and therefore their device extents were not yet in the
4459 * device tree and we missed them in the loop above. So if we have any
4460 * pending chunk using a device extent that overlaps the device range
4461 * that we can not use anymore, commit the current transaction and
4462 * repeat the search on the device tree - this way we guarantee we will
4463 * not have chunks using device extents that end beyond 'new_size'.
4465 if (!checked_pending_chunks
) {
4466 u64 start
= new_size
;
4467 u64 len
= old_size
- new_size
;
4469 if (contains_pending_extent(trans
->transaction
, device
,
4471 mutex_unlock(&fs_info
->chunk_mutex
);
4472 checked_pending_chunks
= true;
4475 ret
= btrfs_commit_transaction(trans
);
4482 btrfs_device_set_disk_total_bytes(device
, new_size
);
4483 if (list_empty(&device
->resized_list
))
4484 list_add_tail(&device
->resized_list
,
4485 &fs_info
->fs_devices
->resized_devices
);
4487 WARN_ON(diff
> old_total
);
4488 btrfs_set_super_total_bytes(super_copy
,
4489 round_down(old_total
- diff
, fs_info
->sectorsize
));
4490 mutex_unlock(&fs_info
->chunk_mutex
);
4492 /* Now btrfs_update_device() will change the on-disk size. */
4493 ret
= btrfs_update_device(trans
, device
);
4495 btrfs_abort_transaction(trans
, ret
);
4496 btrfs_end_transaction(trans
);
4498 ret
= btrfs_commit_transaction(trans
);
4501 btrfs_free_path(path
);
4503 mutex_lock(&fs_info
->chunk_mutex
);
4504 btrfs_device_set_total_bytes(device
, old_size
);
4505 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
))
4506 device
->fs_devices
->total_rw_bytes
+= diff
;
4507 atomic64_add(diff
, &fs_info
->free_chunk_space
);
4508 mutex_unlock(&fs_info
->chunk_mutex
);
4513 static int btrfs_add_system_chunk(struct btrfs_fs_info
*fs_info
,
4514 struct btrfs_key
*key
,
4515 struct btrfs_chunk
*chunk
, int item_size
)
4517 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
4518 struct btrfs_disk_key disk_key
;
4522 mutex_lock(&fs_info
->chunk_mutex
);
4523 array_size
= btrfs_super_sys_array_size(super_copy
);
4524 if (array_size
+ item_size
+ sizeof(disk_key
)
4525 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4526 mutex_unlock(&fs_info
->chunk_mutex
);
4530 ptr
= super_copy
->sys_chunk_array
+ array_size
;
4531 btrfs_cpu_key_to_disk(&disk_key
, key
);
4532 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
4533 ptr
+= sizeof(disk_key
);
4534 memcpy(ptr
, chunk
, item_size
);
4535 item_size
+= sizeof(disk_key
);
4536 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
4537 mutex_unlock(&fs_info
->chunk_mutex
);
4543 * sort the devices in descending order by max_avail, total_avail
4545 static int btrfs_cmp_device_info(const void *a
, const void *b
)
4547 const struct btrfs_device_info
*di_a
= a
;
4548 const struct btrfs_device_info
*di_b
= b
;
4550 if (di_a
->max_avail
> di_b
->max_avail
)
4552 if (di_a
->max_avail
< di_b
->max_avail
)
4554 if (di_a
->total_avail
> di_b
->total_avail
)
4556 if (di_a
->total_avail
< di_b
->total_avail
)
4561 static void check_raid56_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
4563 if (!(type
& BTRFS_BLOCK_GROUP_RAID56_MASK
))
4566 btrfs_set_fs_incompat(info
, RAID56
);
4569 #define BTRFS_MAX_DEVS(info) ((BTRFS_MAX_ITEM_SIZE(info) \
4570 - sizeof(struct btrfs_chunk)) \
4571 / sizeof(struct btrfs_stripe) + 1)
4573 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4574 - 2 * sizeof(struct btrfs_disk_key) \
4575 - 2 * sizeof(struct btrfs_chunk)) \
4576 / sizeof(struct btrfs_stripe) + 1)
4578 static int __btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
4579 u64 start
, u64 type
)
4581 struct btrfs_fs_info
*info
= trans
->fs_info
;
4582 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
4583 struct btrfs_device
*device
;
4584 struct map_lookup
*map
= NULL
;
4585 struct extent_map_tree
*em_tree
;
4586 struct extent_map
*em
;
4587 struct btrfs_device_info
*devices_info
= NULL
;
4589 int num_stripes
; /* total number of stripes to allocate */
4590 int data_stripes
; /* number of stripes that count for
4592 int sub_stripes
; /* sub_stripes info for map */
4593 int dev_stripes
; /* stripes per dev */
4594 int devs_max
; /* max devs to use */
4595 int devs_min
; /* min devs needed */
4596 int devs_increment
; /* ndevs has to be a multiple of this */
4597 int ncopies
; /* how many copies to data has */
4599 u64 max_stripe_size
;
4608 BUG_ON(!alloc_profile_is_valid(type
, 0));
4610 if (list_empty(&fs_devices
->alloc_list
)) {
4611 if (btrfs_test_opt(info
, ENOSPC_DEBUG
))
4612 btrfs_debug(info
, "%s: no writable device", __func__
);
4616 index
= btrfs_bg_flags_to_raid_index(type
);
4618 sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
4619 dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
4620 devs_max
= btrfs_raid_array
[index
].devs_max
;
4621 devs_min
= btrfs_raid_array
[index
].devs_min
;
4622 devs_increment
= btrfs_raid_array
[index
].devs_increment
;
4623 ncopies
= btrfs_raid_array
[index
].ncopies
;
4625 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
4626 max_stripe_size
= SZ_1G
;
4627 max_chunk_size
= BTRFS_MAX_DATA_CHUNK_SIZE
;
4629 devs_max
= BTRFS_MAX_DEVS(info
);
4630 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
4631 /* for larger filesystems, use larger metadata chunks */
4632 if (fs_devices
->total_rw_bytes
> 50ULL * SZ_1G
)
4633 max_stripe_size
= SZ_1G
;
4635 max_stripe_size
= SZ_256M
;
4636 max_chunk_size
= max_stripe_size
;
4638 devs_max
= BTRFS_MAX_DEVS(info
);
4639 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4640 max_stripe_size
= SZ_32M
;
4641 max_chunk_size
= 2 * max_stripe_size
;
4643 devs_max
= BTRFS_MAX_DEVS_SYS_CHUNK
;
4645 btrfs_err(info
, "invalid chunk type 0x%llx requested",
4650 /* we don't want a chunk larger than 10% of writeable space */
4651 max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
4654 devices_info
= kcalloc(fs_devices
->rw_devices
, sizeof(*devices_info
),
4660 * in the first pass through the devices list, we gather information
4661 * about the available holes on each device.
4664 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
4668 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
)) {
4670 "BTRFS: read-only device in alloc_list\n");
4674 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
,
4675 &device
->dev_state
) ||
4676 test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
))
4679 if (device
->total_bytes
> device
->bytes_used
)
4680 total_avail
= device
->total_bytes
- device
->bytes_used
;
4684 /* If there is no space on this device, skip it. */
4685 if (total_avail
== 0)
4688 ret
= find_free_dev_extent(trans
, device
,
4689 max_stripe_size
* dev_stripes
,
4690 &dev_offset
, &max_avail
);
4691 if (ret
&& ret
!= -ENOSPC
)
4695 max_avail
= max_stripe_size
* dev_stripes
;
4697 if (max_avail
< BTRFS_STRIPE_LEN
* dev_stripes
) {
4698 if (btrfs_test_opt(info
, ENOSPC_DEBUG
))
4700 "%s: devid %llu has no free space, have=%llu want=%u",
4701 __func__
, device
->devid
, max_avail
,
4702 BTRFS_STRIPE_LEN
* dev_stripes
);
4706 if (ndevs
== fs_devices
->rw_devices
) {
4707 WARN(1, "%s: found more than %llu devices\n",
4708 __func__
, fs_devices
->rw_devices
);
4711 devices_info
[ndevs
].dev_offset
= dev_offset
;
4712 devices_info
[ndevs
].max_avail
= max_avail
;
4713 devices_info
[ndevs
].total_avail
= total_avail
;
4714 devices_info
[ndevs
].dev
= device
;
4719 * now sort the devices by hole size / available space
4721 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
4722 btrfs_cmp_device_info
, NULL
);
4724 /* round down to number of usable stripes */
4725 ndevs
= round_down(ndevs
, devs_increment
);
4727 if (ndevs
< devs_min
) {
4729 if (btrfs_test_opt(info
, ENOSPC_DEBUG
)) {
4731 "%s: not enough devices with free space: have=%d minimum required=%d",
4732 __func__
, ndevs
, devs_min
);
4737 ndevs
= min(ndevs
, devs_max
);
4740 * The primary goal is to maximize the number of stripes, so use as
4741 * many devices as possible, even if the stripes are not maximum sized.
4743 * The DUP profile stores more than one stripe per device, the
4744 * max_avail is the total size so we have to adjust.
4746 stripe_size
= div_u64(devices_info
[ndevs
- 1].max_avail
, dev_stripes
);
4747 num_stripes
= ndevs
* dev_stripes
;
4750 * this will have to be fixed for RAID1 and RAID10 over
4753 data_stripes
= num_stripes
/ ncopies
;
4755 if (type
& BTRFS_BLOCK_GROUP_RAID5
)
4756 data_stripes
= num_stripes
- 1;
4758 if (type
& BTRFS_BLOCK_GROUP_RAID6
)
4759 data_stripes
= num_stripes
- 2;
4762 * Use the number of data stripes to figure out how big this chunk
4763 * is really going to be in terms of logical address space,
4764 * and compare that answer with the max chunk size
4766 if (stripe_size
* data_stripes
> max_chunk_size
) {
4767 stripe_size
= div_u64(max_chunk_size
, data_stripes
);
4769 /* bump the answer up to a 16MB boundary */
4770 stripe_size
= round_up(stripe_size
, SZ_16M
);
4773 * But don't go higher than the limits we found while searching
4776 stripe_size
= min(devices_info
[ndevs
- 1].max_avail
,
4780 /* align to BTRFS_STRIPE_LEN */
4781 stripe_size
= round_down(stripe_size
, BTRFS_STRIPE_LEN
);
4783 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
4788 map
->num_stripes
= num_stripes
;
4790 for (i
= 0; i
< ndevs
; ++i
) {
4791 for (j
= 0; j
< dev_stripes
; ++j
) {
4792 int s
= i
* dev_stripes
+ j
;
4793 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
4794 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
4798 map
->stripe_len
= BTRFS_STRIPE_LEN
;
4799 map
->io_align
= BTRFS_STRIPE_LEN
;
4800 map
->io_width
= BTRFS_STRIPE_LEN
;
4802 map
->sub_stripes
= sub_stripes
;
4804 num_bytes
= stripe_size
* data_stripes
;
4806 trace_btrfs_chunk_alloc(info
, map
, start
, num_bytes
);
4808 em
= alloc_extent_map();
4814 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
4815 em
->map_lookup
= map
;
4817 em
->len
= num_bytes
;
4818 em
->block_start
= 0;
4819 em
->block_len
= em
->len
;
4820 em
->orig_block_len
= stripe_size
;
4822 em_tree
= &info
->mapping_tree
.map_tree
;
4823 write_lock(&em_tree
->lock
);
4824 ret
= add_extent_mapping(em_tree
, em
, 0);
4826 write_unlock(&em_tree
->lock
);
4827 free_extent_map(em
);
4831 list_add_tail(&em
->list
, &trans
->transaction
->pending_chunks
);
4832 refcount_inc(&em
->refs
);
4833 write_unlock(&em_tree
->lock
);
4835 ret
= btrfs_make_block_group(trans
, 0, type
, start
, num_bytes
);
4837 goto error_del_extent
;
4839 for (i
= 0; i
< map
->num_stripes
; i
++) {
4840 num_bytes
= map
->stripes
[i
].dev
->bytes_used
+ stripe_size
;
4841 btrfs_device_set_bytes_used(map
->stripes
[i
].dev
, num_bytes
);
4844 atomic64_sub(stripe_size
* map
->num_stripes
, &info
->free_chunk_space
);
4846 free_extent_map(em
);
4847 check_raid56_incompat_flag(info
, type
);
4849 kfree(devices_info
);
4853 write_lock(&em_tree
->lock
);
4854 remove_extent_mapping(em_tree
, em
);
4855 write_unlock(&em_tree
->lock
);
4857 /* One for our allocation */
4858 free_extent_map(em
);
4859 /* One for the tree reference */
4860 free_extent_map(em
);
4861 /* One for the pending_chunks list reference */
4862 free_extent_map(em
);
4864 kfree(devices_info
);
4868 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
4869 u64 chunk_offset
, u64 chunk_size
)
4871 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
4872 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
4873 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
4874 struct btrfs_key key
;
4875 struct btrfs_device
*device
;
4876 struct btrfs_chunk
*chunk
;
4877 struct btrfs_stripe
*stripe
;
4878 struct extent_map
*em
;
4879 struct map_lookup
*map
;
4886 em
= get_chunk_map(fs_info
, chunk_offset
, chunk_size
);
4890 map
= em
->map_lookup
;
4891 item_size
= btrfs_chunk_item_size(map
->num_stripes
);
4892 stripe_size
= em
->orig_block_len
;
4894 chunk
= kzalloc(item_size
, GFP_NOFS
);
4901 * Take the device list mutex to prevent races with the final phase of
4902 * a device replace operation that replaces the device object associated
4903 * with the map's stripes, because the device object's id can change
4904 * at any time during that final phase of the device replace operation
4905 * (dev-replace.c:btrfs_dev_replace_finishing()).
4907 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
4908 for (i
= 0; i
< map
->num_stripes
; i
++) {
4909 device
= map
->stripes
[i
].dev
;
4910 dev_offset
= map
->stripes
[i
].physical
;
4912 ret
= btrfs_update_device(trans
, device
);
4915 ret
= btrfs_alloc_dev_extent(trans
, device
, chunk_offset
,
4916 dev_offset
, stripe_size
);
4921 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
4925 stripe
= &chunk
->stripe
;
4926 for (i
= 0; i
< map
->num_stripes
; i
++) {
4927 device
= map
->stripes
[i
].dev
;
4928 dev_offset
= map
->stripes
[i
].physical
;
4930 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
4931 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
4932 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
4935 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
4937 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
4938 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
4939 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
4940 btrfs_set_stack_chunk_type(chunk
, map
->type
);
4941 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
4942 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
4943 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
4944 btrfs_set_stack_chunk_sector_size(chunk
, fs_info
->sectorsize
);
4945 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
4947 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
4948 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
4949 key
.offset
= chunk_offset
;
4951 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
4952 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4954 * TODO: Cleanup of inserted chunk root in case of
4957 ret
= btrfs_add_system_chunk(fs_info
, &key
, chunk
, item_size
);
4962 free_extent_map(em
);
4967 * Chunk allocation falls into two parts. The first part does works
4968 * that make the new allocated chunk useable, but not do any operation
4969 * that modifies the chunk tree. The second part does the works that
4970 * require modifying the chunk tree. This division is important for the
4971 * bootstrap process of adding storage to a seed btrfs.
4973 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
, u64 type
)
4977 lockdep_assert_held(&trans
->fs_info
->chunk_mutex
);
4978 chunk_offset
= find_next_chunk(trans
->fs_info
);
4979 return __btrfs_alloc_chunk(trans
, chunk_offset
, type
);
4982 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
,
4983 struct btrfs_fs_info
*fs_info
)
4986 u64 sys_chunk_offset
;
4990 chunk_offset
= find_next_chunk(fs_info
);
4991 alloc_profile
= btrfs_metadata_alloc_profile(fs_info
);
4992 ret
= __btrfs_alloc_chunk(trans
, chunk_offset
, alloc_profile
);
4996 sys_chunk_offset
= find_next_chunk(fs_info
);
4997 alloc_profile
= btrfs_system_alloc_profile(fs_info
);
4998 ret
= __btrfs_alloc_chunk(trans
, sys_chunk_offset
, alloc_profile
);
5002 static inline int btrfs_chunk_max_errors(struct map_lookup
*map
)
5006 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
5007 BTRFS_BLOCK_GROUP_RAID10
|
5008 BTRFS_BLOCK_GROUP_RAID5
|
5009 BTRFS_BLOCK_GROUP_DUP
)) {
5011 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
) {
5020 int btrfs_chunk_readonly(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
5022 struct extent_map
*em
;
5023 struct map_lookup
*map
;
5028 em
= get_chunk_map(fs_info
, chunk_offset
, 1);
5032 map
= em
->map_lookup
;
5033 for (i
= 0; i
< map
->num_stripes
; i
++) {
5034 if (test_bit(BTRFS_DEV_STATE_MISSING
,
5035 &map
->stripes
[i
].dev
->dev_state
)) {
5039 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE
,
5040 &map
->stripes
[i
].dev
->dev_state
)) {
5047 * If the number of missing devices is larger than max errors,
5048 * we can not write the data into that chunk successfully, so
5051 if (miss_ndevs
> btrfs_chunk_max_errors(map
))
5054 free_extent_map(em
);
5058 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
5060 extent_map_tree_init(&tree
->map_tree
);
5063 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
5065 struct extent_map
*em
;
5068 write_lock(&tree
->map_tree
.lock
);
5069 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
5071 remove_extent_mapping(&tree
->map_tree
, em
);
5072 write_unlock(&tree
->map_tree
.lock
);
5076 free_extent_map(em
);
5077 /* once for the tree */
5078 free_extent_map(em
);
5082 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
5084 struct extent_map
*em
;
5085 struct map_lookup
*map
;
5088 em
= get_chunk_map(fs_info
, logical
, len
);
5091 * We could return errors for these cases, but that could get
5092 * ugly and we'd probably do the same thing which is just not do
5093 * anything else and exit, so return 1 so the callers don't try
5094 * to use other copies.
5098 map
= em
->map_lookup
;
5099 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
5100 ret
= map
->num_stripes
;
5101 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5102 ret
= map
->sub_stripes
;
5103 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID5
)
5105 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5107 * There could be two corrupted data stripes, we need
5108 * to loop retry in order to rebuild the correct data.
5110 * Fail a stripe at a time on every retry except the
5111 * stripe under reconstruction.
5113 ret
= map
->num_stripes
;
5116 free_extent_map(em
);
5118 btrfs_dev_replace_read_lock(&fs_info
->dev_replace
);
5119 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
) &&
5120 fs_info
->dev_replace
.tgtdev
)
5122 btrfs_dev_replace_read_unlock(&fs_info
->dev_replace
);
5127 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info
*fs_info
,
5130 struct extent_map
*em
;
5131 struct map_lookup
*map
;
5132 unsigned long len
= fs_info
->sectorsize
;
5134 em
= get_chunk_map(fs_info
, logical
, len
);
5136 if (!WARN_ON(IS_ERR(em
))) {
5137 map
= em
->map_lookup
;
5138 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5139 len
= map
->stripe_len
* nr_data_stripes(map
);
5140 free_extent_map(em
);
5145 int btrfs_is_parity_mirror(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
5147 struct extent_map
*em
;
5148 struct map_lookup
*map
;
5151 em
= get_chunk_map(fs_info
, logical
, len
);
5153 if(!WARN_ON(IS_ERR(em
))) {
5154 map
= em
->map_lookup
;
5155 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5157 free_extent_map(em
);
5162 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
5163 struct map_lookup
*map
, int first
,
5164 int dev_replace_is_ongoing
)
5168 int preferred_mirror
;
5170 struct btrfs_device
*srcdev
;
5173 (BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
)));
5175 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5176 num_stripes
= map
->sub_stripes
;
5178 num_stripes
= map
->num_stripes
;
5180 preferred_mirror
= first
+ current
->pid
% num_stripes
;
5182 if (dev_replace_is_ongoing
&&
5183 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
5184 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
5185 srcdev
= fs_info
->dev_replace
.srcdev
;
5190 * try to avoid the drive that is the source drive for a
5191 * dev-replace procedure, only choose it if no other non-missing
5192 * mirror is available
5194 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
5195 if (map
->stripes
[preferred_mirror
].dev
->bdev
&&
5196 (tolerance
|| map
->stripes
[preferred_mirror
].dev
!= srcdev
))
5197 return preferred_mirror
;
5198 for (i
= first
; i
< first
+ num_stripes
; i
++) {
5199 if (map
->stripes
[i
].dev
->bdev
&&
5200 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
5205 /* we couldn't find one that doesn't fail. Just return something
5206 * and the io error handling code will clean up eventually
5208 return preferred_mirror
;
5211 static inline int parity_smaller(u64 a
, u64 b
)
5216 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5217 static void sort_parity_stripes(struct btrfs_bio
*bbio
, int num_stripes
)
5219 struct btrfs_bio_stripe s
;
5226 for (i
= 0; i
< num_stripes
- 1; i
++) {
5227 if (parity_smaller(bbio
->raid_map
[i
],
5228 bbio
->raid_map
[i
+1])) {
5229 s
= bbio
->stripes
[i
];
5230 l
= bbio
->raid_map
[i
];
5231 bbio
->stripes
[i
] = bbio
->stripes
[i
+1];
5232 bbio
->raid_map
[i
] = bbio
->raid_map
[i
+1];
5233 bbio
->stripes
[i
+1] = s
;
5234 bbio
->raid_map
[i
+1] = l
;
5242 static struct btrfs_bio
*alloc_btrfs_bio(int total_stripes
, int real_stripes
)
5244 struct btrfs_bio
*bbio
= kzalloc(
5245 /* the size of the btrfs_bio */
5246 sizeof(struct btrfs_bio
) +
5247 /* plus the variable array for the stripes */
5248 sizeof(struct btrfs_bio_stripe
) * (total_stripes
) +
5249 /* plus the variable array for the tgt dev */
5250 sizeof(int) * (real_stripes
) +
5252 * plus the raid_map, which includes both the tgt dev
5255 sizeof(u64
) * (total_stripes
),
5256 GFP_NOFS
|__GFP_NOFAIL
);
5258 atomic_set(&bbio
->error
, 0);
5259 refcount_set(&bbio
->refs
, 1);
5264 void btrfs_get_bbio(struct btrfs_bio
*bbio
)
5266 WARN_ON(!refcount_read(&bbio
->refs
));
5267 refcount_inc(&bbio
->refs
);
5270 void btrfs_put_bbio(struct btrfs_bio
*bbio
)
5274 if (refcount_dec_and_test(&bbio
->refs
))
5278 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5280 * Please note that, discard won't be sent to target device of device
5283 static int __btrfs_map_block_for_discard(struct btrfs_fs_info
*fs_info
,
5284 u64 logical
, u64 length
,
5285 struct btrfs_bio
**bbio_ret
)
5287 struct extent_map
*em
;
5288 struct map_lookup
*map
;
5289 struct btrfs_bio
*bbio
;
5293 u64 stripe_end_offset
;
5300 u32 sub_stripes
= 0;
5301 u64 stripes_per_dev
= 0;
5302 u32 remaining_stripes
= 0;
5303 u32 last_stripe
= 0;
5307 /* discard always return a bbio */
5310 em
= get_chunk_map(fs_info
, logical
, length
);
5314 map
= em
->map_lookup
;
5315 /* we don't discard raid56 yet */
5316 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5321 offset
= logical
- em
->start
;
5322 length
= min_t(u64
, em
->len
- offset
, length
);
5324 stripe_len
= map
->stripe_len
;
5326 * stripe_nr counts the total number of stripes we have to stride
5327 * to get to this block
5329 stripe_nr
= div64_u64(offset
, stripe_len
);
5331 /* stripe_offset is the offset of this block in its stripe */
5332 stripe_offset
= offset
- stripe_nr
* stripe_len
;
5334 stripe_nr_end
= round_up(offset
+ length
, map
->stripe_len
);
5335 stripe_nr_end
= div64_u64(stripe_nr_end
, map
->stripe_len
);
5336 stripe_cnt
= stripe_nr_end
- stripe_nr
;
5337 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
5340 * after this, stripe_nr is the number of stripes on this
5341 * device we have to walk to find the data, and stripe_index is
5342 * the number of our device in the stripe array
5346 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5347 BTRFS_BLOCK_GROUP_RAID10
)) {
5348 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5351 sub_stripes
= map
->sub_stripes
;
5353 factor
= map
->num_stripes
/ sub_stripes
;
5354 num_stripes
= min_t(u64
, map
->num_stripes
,
5355 sub_stripes
* stripe_cnt
);
5356 stripe_nr
= div_u64_rem(stripe_nr
, factor
, &stripe_index
);
5357 stripe_index
*= sub_stripes
;
5358 stripes_per_dev
= div_u64_rem(stripe_cnt
, factor
,
5359 &remaining_stripes
);
5360 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
5361 last_stripe
*= sub_stripes
;
5362 } else if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
5363 BTRFS_BLOCK_GROUP_DUP
)) {
5364 num_stripes
= map
->num_stripes
;
5366 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5370 bbio
= alloc_btrfs_bio(num_stripes
, 0);
5376 for (i
= 0; i
< num_stripes
; i
++) {
5377 bbio
->stripes
[i
].physical
=
5378 map
->stripes
[stripe_index
].physical
+
5379 stripe_offset
+ stripe_nr
* map
->stripe_len
;
5380 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
5382 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5383 BTRFS_BLOCK_GROUP_RAID10
)) {
5384 bbio
->stripes
[i
].length
= stripes_per_dev
*
5387 if (i
/ sub_stripes
< remaining_stripes
)
5388 bbio
->stripes
[i
].length
+=
5392 * Special for the first stripe and
5395 * |-------|...|-------|
5399 if (i
< sub_stripes
)
5400 bbio
->stripes
[i
].length
-=
5403 if (stripe_index
>= last_stripe
&&
5404 stripe_index
<= (last_stripe
+
5406 bbio
->stripes
[i
].length
-=
5409 if (i
== sub_stripes
- 1)
5412 bbio
->stripes
[i
].length
= length
;
5416 if (stripe_index
== map
->num_stripes
) {
5423 bbio
->map_type
= map
->type
;
5424 bbio
->num_stripes
= num_stripes
;
5426 free_extent_map(em
);
5431 * In dev-replace case, for repair case (that's the only case where the mirror
5432 * is selected explicitly when calling btrfs_map_block), blocks left of the
5433 * left cursor can also be read from the target drive.
5435 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5437 * For READ, it also needs to be supported using the same mirror number.
5439 * If the requested block is not left of the left cursor, EIO is returned. This
5440 * can happen because btrfs_num_copies() returns one more in the dev-replace
5443 static int get_extra_mirror_from_replace(struct btrfs_fs_info
*fs_info
,
5444 u64 logical
, u64 length
,
5445 u64 srcdev_devid
, int *mirror_num
,
5448 struct btrfs_bio
*bbio
= NULL
;
5450 int index_srcdev
= 0;
5452 u64 physical_of_found
= 0;
5456 ret
= __btrfs_map_block(fs_info
, BTRFS_MAP_GET_READ_MIRRORS
,
5457 logical
, &length
, &bbio
, 0, 0);
5459 ASSERT(bbio
== NULL
);
5463 num_stripes
= bbio
->num_stripes
;
5464 if (*mirror_num
> num_stripes
) {
5466 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5467 * that means that the requested area is not left of the left
5470 btrfs_put_bbio(bbio
);
5475 * process the rest of the function using the mirror_num of the source
5476 * drive. Therefore look it up first. At the end, patch the device
5477 * pointer to the one of the target drive.
5479 for (i
= 0; i
< num_stripes
; i
++) {
5480 if (bbio
->stripes
[i
].dev
->devid
!= srcdev_devid
)
5484 * In case of DUP, in order to keep it simple, only add the
5485 * mirror with the lowest physical address
5488 physical_of_found
<= bbio
->stripes
[i
].physical
)
5493 physical_of_found
= bbio
->stripes
[i
].physical
;
5496 btrfs_put_bbio(bbio
);
5502 *mirror_num
= index_srcdev
+ 1;
5503 *physical
= physical_of_found
;
5507 static void handle_ops_on_dev_replace(enum btrfs_map_op op
,
5508 struct btrfs_bio
**bbio_ret
,
5509 struct btrfs_dev_replace
*dev_replace
,
5510 int *num_stripes_ret
, int *max_errors_ret
)
5512 struct btrfs_bio
*bbio
= *bbio_ret
;
5513 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
5514 int tgtdev_indexes
= 0;
5515 int num_stripes
= *num_stripes_ret
;
5516 int max_errors
= *max_errors_ret
;
5519 if (op
== BTRFS_MAP_WRITE
) {
5520 int index_where_to_add
;
5523 * duplicate the write operations while the dev replace
5524 * procedure is running. Since the copying of the old disk to
5525 * the new disk takes place at run time while the filesystem is
5526 * mounted writable, the regular write operations to the old
5527 * disk have to be duplicated to go to the new disk as well.
5529 * Note that device->missing is handled by the caller, and that
5530 * the write to the old disk is already set up in the stripes
5533 index_where_to_add
= num_stripes
;
5534 for (i
= 0; i
< num_stripes
; i
++) {
5535 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5536 /* write to new disk, too */
5537 struct btrfs_bio_stripe
*new =
5538 bbio
->stripes
+ index_where_to_add
;
5539 struct btrfs_bio_stripe
*old
=
5542 new->physical
= old
->physical
;
5543 new->length
= old
->length
;
5544 new->dev
= dev_replace
->tgtdev
;
5545 bbio
->tgtdev_map
[i
] = index_where_to_add
;
5546 index_where_to_add
++;
5551 num_stripes
= index_where_to_add
;
5552 } else if (op
== BTRFS_MAP_GET_READ_MIRRORS
) {
5553 int index_srcdev
= 0;
5555 u64 physical_of_found
= 0;
5558 * During the dev-replace procedure, the target drive can also
5559 * be used to read data in case it is needed to repair a corrupt
5560 * block elsewhere. This is possible if the requested area is
5561 * left of the left cursor. In this area, the target drive is a
5562 * full copy of the source drive.
5564 for (i
= 0; i
< num_stripes
; i
++) {
5565 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
5567 * In case of DUP, in order to keep it simple,
5568 * only add the mirror with the lowest physical
5572 physical_of_found
<=
5573 bbio
->stripes
[i
].physical
)
5577 physical_of_found
= bbio
->stripes
[i
].physical
;
5581 struct btrfs_bio_stripe
*tgtdev_stripe
=
5582 bbio
->stripes
+ num_stripes
;
5584 tgtdev_stripe
->physical
= physical_of_found
;
5585 tgtdev_stripe
->length
=
5586 bbio
->stripes
[index_srcdev
].length
;
5587 tgtdev_stripe
->dev
= dev_replace
->tgtdev
;
5588 bbio
->tgtdev_map
[index_srcdev
] = num_stripes
;
5595 *num_stripes_ret
= num_stripes
;
5596 *max_errors_ret
= max_errors
;
5597 bbio
->num_tgtdevs
= tgtdev_indexes
;
5601 static bool need_full_stripe(enum btrfs_map_op op
)
5603 return (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
);
5606 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
,
5607 enum btrfs_map_op op
,
5608 u64 logical
, u64
*length
,
5609 struct btrfs_bio
**bbio_ret
,
5610 int mirror_num
, int need_raid_map
)
5612 struct extent_map
*em
;
5613 struct map_lookup
*map
;
5623 int tgtdev_indexes
= 0;
5624 struct btrfs_bio
*bbio
= NULL
;
5625 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
5626 int dev_replace_is_ongoing
= 0;
5627 int num_alloc_stripes
;
5628 int patch_the_first_stripe_for_dev_replace
= 0;
5629 u64 physical_to_patch_in_first_stripe
= 0;
5630 u64 raid56_full_stripe_start
= (u64
)-1;
5632 if (op
== BTRFS_MAP_DISCARD
)
5633 return __btrfs_map_block_for_discard(fs_info
, logical
,
5636 em
= get_chunk_map(fs_info
, logical
, *length
);
5640 map
= em
->map_lookup
;
5641 offset
= logical
- em
->start
;
5643 stripe_len
= map
->stripe_len
;
5646 * stripe_nr counts the total number of stripes we have to stride
5647 * to get to this block
5649 stripe_nr
= div64_u64(stripe_nr
, stripe_len
);
5651 stripe_offset
= stripe_nr
* stripe_len
;
5652 if (offset
< stripe_offset
) {
5654 "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu",
5655 stripe_offset
, offset
, em
->start
, logical
,
5657 free_extent_map(em
);
5661 /* stripe_offset is the offset of this block in its stripe*/
5662 stripe_offset
= offset
- stripe_offset
;
5664 /* if we're here for raid56, we need to know the stripe aligned start */
5665 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5666 unsigned long full_stripe_len
= stripe_len
* nr_data_stripes(map
);
5667 raid56_full_stripe_start
= offset
;
5669 /* allow a write of a full stripe, but make sure we don't
5670 * allow straddling of stripes
5672 raid56_full_stripe_start
= div64_u64(raid56_full_stripe_start
,
5674 raid56_full_stripe_start
*= full_stripe_len
;
5677 if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
5679 /* For writes to RAID[56], allow a full stripeset across all disks.
5680 For other RAID types and for RAID[56] reads, just allow a single
5681 stripe (on a single disk). */
5682 if ((map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
5683 (op
== BTRFS_MAP_WRITE
)) {
5684 max_len
= stripe_len
* nr_data_stripes(map
) -
5685 (offset
- raid56_full_stripe_start
);
5687 /* we limit the length of each bio to what fits in a stripe */
5688 max_len
= stripe_len
- stripe_offset
;
5690 *length
= min_t(u64
, em
->len
- offset
, max_len
);
5692 *length
= em
->len
- offset
;
5695 /* This is for when we're called from btrfs_merge_bio_hook() and all
5696 it cares about is the length */
5700 btrfs_dev_replace_read_lock(dev_replace
);
5701 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
5702 if (!dev_replace_is_ongoing
)
5703 btrfs_dev_replace_read_unlock(dev_replace
);
5705 btrfs_dev_replace_set_lock_blocking(dev_replace
);
5707 if (dev_replace_is_ongoing
&& mirror_num
== map
->num_stripes
+ 1 &&
5708 !need_full_stripe(op
) && dev_replace
->tgtdev
!= NULL
) {
5709 ret
= get_extra_mirror_from_replace(fs_info
, logical
, *length
,
5710 dev_replace
->srcdev
->devid
,
5712 &physical_to_patch_in_first_stripe
);
5716 patch_the_first_stripe_for_dev_replace
= 1;
5717 } else if (mirror_num
> map
->num_stripes
) {
5723 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5724 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5726 if (!need_full_stripe(op
))
5728 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
5729 if (need_full_stripe(op
))
5730 num_stripes
= map
->num_stripes
;
5731 else if (mirror_num
)
5732 stripe_index
= mirror_num
- 1;
5734 stripe_index
= find_live_mirror(fs_info
, map
, 0,
5735 dev_replace_is_ongoing
);
5736 mirror_num
= stripe_index
+ 1;
5739 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
5740 if (need_full_stripe(op
)) {
5741 num_stripes
= map
->num_stripes
;
5742 } else if (mirror_num
) {
5743 stripe_index
= mirror_num
- 1;
5748 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5749 u32 factor
= map
->num_stripes
/ map
->sub_stripes
;
5751 stripe_nr
= div_u64_rem(stripe_nr
, factor
, &stripe_index
);
5752 stripe_index
*= map
->sub_stripes
;
5754 if (need_full_stripe(op
))
5755 num_stripes
= map
->sub_stripes
;
5756 else if (mirror_num
)
5757 stripe_index
+= mirror_num
- 1;
5759 int old_stripe_index
= stripe_index
;
5760 stripe_index
= find_live_mirror(fs_info
, map
,
5762 dev_replace_is_ongoing
);
5763 mirror_num
= stripe_index
- old_stripe_index
+ 1;
5766 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5767 if (need_raid_map
&& (need_full_stripe(op
) || mirror_num
> 1)) {
5768 /* push stripe_nr back to the start of the full stripe */
5769 stripe_nr
= div64_u64(raid56_full_stripe_start
,
5770 stripe_len
* nr_data_stripes(map
));
5772 /* RAID[56] write or recovery. Return all stripes */
5773 num_stripes
= map
->num_stripes
;
5774 max_errors
= nr_parity_stripes(map
);
5776 *length
= map
->stripe_len
;
5781 * Mirror #0 or #1 means the original data block.
5782 * Mirror #2 is RAID5 parity block.
5783 * Mirror #3 is RAID6 Q block.
5785 stripe_nr
= div_u64_rem(stripe_nr
,
5786 nr_data_stripes(map
), &stripe_index
);
5788 stripe_index
= nr_data_stripes(map
) +
5791 /* We distribute the parity blocks across stripes */
5792 div_u64_rem(stripe_nr
+ stripe_index
, map
->num_stripes
,
5794 if (!need_full_stripe(op
) && mirror_num
<= 1)
5799 * after this, stripe_nr is the number of stripes on this
5800 * device we have to walk to find the data, and stripe_index is
5801 * the number of our device in the stripe array
5803 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5805 mirror_num
= stripe_index
+ 1;
5807 if (stripe_index
>= map
->num_stripes
) {
5809 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
5810 stripe_index
, map
->num_stripes
);
5815 num_alloc_stripes
= num_stripes
;
5816 if (dev_replace_is_ongoing
&& dev_replace
->tgtdev
!= NULL
) {
5817 if (op
== BTRFS_MAP_WRITE
)
5818 num_alloc_stripes
<<= 1;
5819 if (op
== BTRFS_MAP_GET_READ_MIRRORS
)
5820 num_alloc_stripes
++;
5821 tgtdev_indexes
= num_stripes
;
5824 bbio
= alloc_btrfs_bio(num_alloc_stripes
, tgtdev_indexes
);
5829 if (dev_replace_is_ongoing
&& dev_replace
->tgtdev
!= NULL
)
5830 bbio
->tgtdev_map
= (int *)(bbio
->stripes
+ num_alloc_stripes
);
5832 /* build raid_map */
5833 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
&& need_raid_map
&&
5834 (need_full_stripe(op
) || mirror_num
> 1)) {
5838 bbio
->raid_map
= (u64
*)((void *)bbio
->stripes
+
5839 sizeof(struct btrfs_bio_stripe
) *
5841 sizeof(int) * tgtdev_indexes
);
5843 /* Work out the disk rotation on this stripe-set */
5844 div_u64_rem(stripe_nr
, num_stripes
, &rot
);
5846 /* Fill in the logical address of each stripe */
5847 tmp
= stripe_nr
* nr_data_stripes(map
);
5848 for (i
= 0; i
< nr_data_stripes(map
); i
++)
5849 bbio
->raid_map
[(i
+rot
) % num_stripes
] =
5850 em
->start
+ (tmp
+ i
) * map
->stripe_len
;
5852 bbio
->raid_map
[(i
+rot
) % map
->num_stripes
] = RAID5_P_STRIPE
;
5853 if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5854 bbio
->raid_map
[(i
+rot
+1) % num_stripes
] =
5859 for (i
= 0; i
< num_stripes
; i
++) {
5860 bbio
->stripes
[i
].physical
=
5861 map
->stripes
[stripe_index
].physical
+
5863 stripe_nr
* map
->stripe_len
;
5864 bbio
->stripes
[i
].dev
=
5865 map
->stripes
[stripe_index
].dev
;
5869 if (need_full_stripe(op
))
5870 max_errors
= btrfs_chunk_max_errors(map
);
5873 sort_parity_stripes(bbio
, num_stripes
);
5875 if (dev_replace_is_ongoing
&& dev_replace
->tgtdev
!= NULL
&&
5876 need_full_stripe(op
)) {
5877 handle_ops_on_dev_replace(op
, &bbio
, dev_replace
, &num_stripes
,
5882 bbio
->map_type
= map
->type
;
5883 bbio
->num_stripes
= num_stripes
;
5884 bbio
->max_errors
= max_errors
;
5885 bbio
->mirror_num
= mirror_num
;
5888 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5889 * mirror_num == num_stripes + 1 && dev_replace target drive is
5890 * available as a mirror
5892 if (patch_the_first_stripe_for_dev_replace
&& num_stripes
> 0) {
5893 WARN_ON(num_stripes
> 1);
5894 bbio
->stripes
[0].dev
= dev_replace
->tgtdev
;
5895 bbio
->stripes
[0].physical
= physical_to_patch_in_first_stripe
;
5896 bbio
->mirror_num
= map
->num_stripes
+ 1;
5899 if (dev_replace_is_ongoing
) {
5900 btrfs_dev_replace_clear_lock_blocking(dev_replace
);
5901 btrfs_dev_replace_read_unlock(dev_replace
);
5903 free_extent_map(em
);
5907 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, enum btrfs_map_op op
,
5908 u64 logical
, u64
*length
,
5909 struct btrfs_bio
**bbio_ret
, int mirror_num
)
5911 return __btrfs_map_block(fs_info
, op
, logical
, length
, bbio_ret
,
5915 /* For Scrub/replace */
5916 int btrfs_map_sblock(struct btrfs_fs_info
*fs_info
, enum btrfs_map_op op
,
5917 u64 logical
, u64
*length
,
5918 struct btrfs_bio
**bbio_ret
)
5920 return __btrfs_map_block(fs_info
, op
, logical
, length
, bbio_ret
, 0, 1);
5923 int btrfs_rmap_block(struct btrfs_fs_info
*fs_info
, u64 chunk_start
,
5924 u64 physical
, u64
**logical
, int *naddrs
, int *stripe_len
)
5926 struct extent_map
*em
;
5927 struct map_lookup
*map
;
5935 em
= get_chunk_map(fs_info
, chunk_start
, 1);
5939 map
= em
->map_lookup
;
5941 rmap_len
= map
->stripe_len
;
5943 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5944 length
= div_u64(length
, map
->num_stripes
/ map
->sub_stripes
);
5945 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5946 length
= div_u64(length
, map
->num_stripes
);
5947 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5948 length
= div_u64(length
, nr_data_stripes(map
));
5949 rmap_len
= map
->stripe_len
* nr_data_stripes(map
);
5952 buf
= kcalloc(map
->num_stripes
, sizeof(u64
), GFP_NOFS
);
5953 BUG_ON(!buf
); /* -ENOMEM */
5955 for (i
= 0; i
< map
->num_stripes
; i
++) {
5956 if (map
->stripes
[i
].physical
> physical
||
5957 map
->stripes
[i
].physical
+ length
<= physical
)
5960 stripe_nr
= physical
- map
->stripes
[i
].physical
;
5961 stripe_nr
= div64_u64(stripe_nr
, map
->stripe_len
);
5963 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
5964 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5965 stripe_nr
= div_u64(stripe_nr
, map
->sub_stripes
);
5966 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
5967 stripe_nr
= stripe_nr
* map
->num_stripes
+ i
;
5968 } /* else if RAID[56], multiply by nr_data_stripes().
5969 * Alternatively, just use rmap_len below instead of
5970 * map->stripe_len */
5972 bytenr
= chunk_start
+ stripe_nr
* rmap_len
;
5973 WARN_ON(nr
>= map
->num_stripes
);
5974 for (j
= 0; j
< nr
; j
++) {
5975 if (buf
[j
] == bytenr
)
5979 WARN_ON(nr
>= map
->num_stripes
);
5986 *stripe_len
= rmap_len
;
5988 free_extent_map(em
);
5992 static inline void btrfs_end_bbio(struct btrfs_bio
*bbio
, struct bio
*bio
)
5994 bio
->bi_private
= bbio
->private;
5995 bio
->bi_end_io
= bbio
->end_io
;
5998 btrfs_put_bbio(bbio
);
6001 static void btrfs_end_bio(struct bio
*bio
)
6003 struct btrfs_bio
*bbio
= bio
->bi_private
;
6004 int is_orig_bio
= 0;
6006 if (bio
->bi_status
) {
6007 atomic_inc(&bbio
->error
);
6008 if (bio
->bi_status
== BLK_STS_IOERR
||
6009 bio
->bi_status
== BLK_STS_TARGET
) {
6010 unsigned int stripe_index
=
6011 btrfs_io_bio(bio
)->stripe_index
;
6012 struct btrfs_device
*dev
;
6014 BUG_ON(stripe_index
>= bbio
->num_stripes
);
6015 dev
= bbio
->stripes
[stripe_index
].dev
;
6017 if (bio_op(bio
) == REQ_OP_WRITE
)
6018 btrfs_dev_stat_inc_and_print(dev
,
6019 BTRFS_DEV_STAT_WRITE_ERRS
);
6021 btrfs_dev_stat_inc_and_print(dev
,
6022 BTRFS_DEV_STAT_READ_ERRS
);
6023 if (bio
->bi_opf
& REQ_PREFLUSH
)
6024 btrfs_dev_stat_inc_and_print(dev
,
6025 BTRFS_DEV_STAT_FLUSH_ERRS
);
6030 if (bio
== bbio
->orig_bio
)
6033 btrfs_bio_counter_dec(bbio
->fs_info
);
6035 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
6038 bio
= bbio
->orig_bio
;
6041 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
6042 /* only send an error to the higher layers if it is
6043 * beyond the tolerance of the btrfs bio
6045 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
6046 bio
->bi_status
= BLK_STS_IOERR
;
6049 * this bio is actually up to date, we didn't
6050 * go over the max number of errors
6052 bio
->bi_status
= BLK_STS_OK
;
6055 btrfs_end_bbio(bbio
, bio
);
6056 } else if (!is_orig_bio
) {
6062 * see run_scheduled_bios for a description of why bios are collected for
6065 * This will add one bio to the pending list for a device and make sure
6066 * the work struct is scheduled.
6068 static noinline
void btrfs_schedule_bio(struct btrfs_device
*device
,
6071 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
6072 int should_queue
= 1;
6073 struct btrfs_pending_bios
*pending_bios
;
6075 if (test_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
) ||
6081 /* don't bother with additional async steps for reads, right now */
6082 if (bio_op(bio
) == REQ_OP_READ
) {
6083 btrfsic_submit_bio(bio
);
6087 WARN_ON(bio
->bi_next
);
6088 bio
->bi_next
= NULL
;
6090 spin_lock(&device
->io_lock
);
6091 if (op_is_sync(bio
->bi_opf
))
6092 pending_bios
= &device
->pending_sync_bios
;
6094 pending_bios
= &device
->pending_bios
;
6096 if (pending_bios
->tail
)
6097 pending_bios
->tail
->bi_next
= bio
;
6099 pending_bios
->tail
= bio
;
6100 if (!pending_bios
->head
)
6101 pending_bios
->head
= bio
;
6102 if (device
->running_pending
)
6105 spin_unlock(&device
->io_lock
);
6108 btrfs_queue_work(fs_info
->submit_workers
, &device
->work
);
6111 static void submit_stripe_bio(struct btrfs_bio
*bbio
, struct bio
*bio
,
6112 u64 physical
, int dev_nr
, int async
)
6114 struct btrfs_device
*dev
= bbio
->stripes
[dev_nr
].dev
;
6115 struct btrfs_fs_info
*fs_info
= bbio
->fs_info
;
6117 bio
->bi_private
= bbio
;
6118 btrfs_io_bio(bio
)->stripe_index
= dev_nr
;
6119 bio
->bi_end_io
= btrfs_end_bio
;
6120 bio
->bi_iter
.bi_sector
= physical
>> 9;
6121 btrfs_debug_in_rcu(fs_info
,
6122 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6123 bio_op(bio
), bio
->bi_opf
, (u64
)bio
->bi_iter
.bi_sector
,
6124 (u_long
)dev
->bdev
->bd_dev
, rcu_str_deref(dev
->name
), dev
->devid
,
6125 bio
->bi_iter
.bi_size
);
6126 bio_set_dev(bio
, dev
->bdev
);
6128 btrfs_bio_counter_inc_noblocked(fs_info
);
6131 btrfs_schedule_bio(dev
, bio
);
6133 btrfsic_submit_bio(bio
);
6136 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
6138 atomic_inc(&bbio
->error
);
6139 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
6140 /* Should be the original bio. */
6141 WARN_ON(bio
!= bbio
->orig_bio
);
6143 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
6144 bio
->bi_iter
.bi_sector
= logical
>> 9;
6145 if (atomic_read(&bbio
->error
) > bbio
->max_errors
)
6146 bio
->bi_status
= BLK_STS_IOERR
;
6148 bio
->bi_status
= BLK_STS_OK
;
6149 btrfs_end_bbio(bbio
, bio
);
6153 blk_status_t
btrfs_map_bio(struct btrfs_fs_info
*fs_info
, struct bio
*bio
,
6154 int mirror_num
, int async_submit
)
6156 struct btrfs_device
*dev
;
6157 struct bio
*first_bio
= bio
;
6158 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
6164 struct btrfs_bio
*bbio
= NULL
;
6166 length
= bio
->bi_iter
.bi_size
;
6167 map_length
= length
;
6169 btrfs_bio_counter_inc_blocked(fs_info
);
6170 ret
= __btrfs_map_block(fs_info
, btrfs_op(bio
), logical
,
6171 &map_length
, &bbio
, mirror_num
, 1);
6173 btrfs_bio_counter_dec(fs_info
);
6174 return errno_to_blk_status(ret
);
6177 total_devs
= bbio
->num_stripes
;
6178 bbio
->orig_bio
= first_bio
;
6179 bbio
->private = first_bio
->bi_private
;
6180 bbio
->end_io
= first_bio
->bi_end_io
;
6181 bbio
->fs_info
= fs_info
;
6182 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
6184 if ((bbio
->map_type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
6185 ((bio_op(bio
) == REQ_OP_WRITE
) || (mirror_num
> 1))) {
6186 /* In this case, map_length has been set to the length of
6187 a single stripe; not the whole write */
6188 if (bio_op(bio
) == REQ_OP_WRITE
) {
6189 ret
= raid56_parity_write(fs_info
, bio
, bbio
,
6192 ret
= raid56_parity_recover(fs_info
, bio
, bbio
,
6193 map_length
, mirror_num
, 1);
6196 btrfs_bio_counter_dec(fs_info
);
6197 return errno_to_blk_status(ret
);
6200 if (map_length
< length
) {
6202 "mapping failed logical %llu bio len %llu len %llu",
6203 logical
, length
, map_length
);
6207 for (dev_nr
= 0; dev_nr
< total_devs
; dev_nr
++) {
6208 dev
= bbio
->stripes
[dev_nr
].dev
;
6209 if (!dev
|| !dev
->bdev
||
6210 (bio_op(first_bio
) == REQ_OP_WRITE
&&
6211 !test_bit(BTRFS_DEV_STATE_WRITEABLE
, &dev
->dev_state
))) {
6212 bbio_error(bbio
, first_bio
, logical
);
6216 if (dev_nr
< total_devs
- 1)
6217 bio
= btrfs_bio_clone(first_bio
);
6221 submit_stripe_bio(bbio
, bio
, bbio
->stripes
[dev_nr
].physical
,
6222 dev_nr
, async_submit
);
6224 btrfs_bio_counter_dec(fs_info
);
6228 struct btrfs_device
*btrfs_find_device(struct btrfs_fs_info
*fs_info
, u64 devid
,
6231 struct btrfs_device
*device
;
6232 struct btrfs_fs_devices
*cur_devices
;
6234 cur_devices
= fs_info
->fs_devices
;
6235 while (cur_devices
) {
6237 !memcmp(cur_devices
->fsid
, fsid
, BTRFS_FSID_SIZE
)) {
6238 device
= find_device(cur_devices
, devid
, uuid
);
6242 cur_devices
= cur_devices
->seed
;
6247 static struct btrfs_device
*add_missing_dev(struct btrfs_fs_devices
*fs_devices
,
6248 u64 devid
, u8
*dev_uuid
)
6250 struct btrfs_device
*device
;
6252 device
= btrfs_alloc_device(NULL
, &devid
, dev_uuid
);
6256 list_add(&device
->dev_list
, &fs_devices
->devices
);
6257 device
->fs_devices
= fs_devices
;
6258 fs_devices
->num_devices
++;
6260 set_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
);
6261 fs_devices
->missing_devices
++;
6267 * btrfs_alloc_device - allocate struct btrfs_device
6268 * @fs_info: used only for generating a new devid, can be NULL if
6269 * devid is provided (i.e. @devid != NULL).
6270 * @devid: a pointer to devid for this device. If NULL a new devid
6272 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6275 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6276 * on error. Returned struct is not linked onto any lists and must be
6277 * destroyed with btrfs_free_device.
6279 struct btrfs_device
*btrfs_alloc_device(struct btrfs_fs_info
*fs_info
,
6283 struct btrfs_device
*dev
;
6286 if (WARN_ON(!devid
&& !fs_info
))
6287 return ERR_PTR(-EINVAL
);
6289 dev
= __alloc_device();
6298 ret
= find_next_devid(fs_info
, &tmp
);
6300 btrfs_free_device(dev
);
6301 return ERR_PTR(ret
);
6307 memcpy(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
);
6309 generate_random_uuid(dev
->uuid
);
6311 btrfs_init_work(&dev
->work
, btrfs_submit_helper
,
6312 pending_bios_fn
, NULL
, NULL
);
6317 /* Return -EIO if any error, otherwise return 0. */
6318 static int btrfs_check_chunk_valid(struct btrfs_fs_info
*fs_info
,
6319 struct extent_buffer
*leaf
,
6320 struct btrfs_chunk
*chunk
, u64 logical
)
6330 length
= btrfs_chunk_length(leaf
, chunk
);
6331 stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6332 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6333 sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6334 type
= btrfs_chunk_type(leaf
, chunk
);
6337 btrfs_err(fs_info
, "invalid chunk num_stripes: %u",
6341 if (!IS_ALIGNED(logical
, fs_info
->sectorsize
)) {
6342 btrfs_err(fs_info
, "invalid chunk logical %llu", logical
);
6345 if (btrfs_chunk_sector_size(leaf
, chunk
) != fs_info
->sectorsize
) {
6346 btrfs_err(fs_info
, "invalid chunk sectorsize %u",
6347 btrfs_chunk_sector_size(leaf
, chunk
));
6350 if (!length
|| !IS_ALIGNED(length
, fs_info
->sectorsize
)) {
6351 btrfs_err(fs_info
, "invalid chunk length %llu", length
);
6354 if (!is_power_of_2(stripe_len
) || stripe_len
!= BTRFS_STRIPE_LEN
) {
6355 btrfs_err(fs_info
, "invalid chunk stripe length: %llu",
6359 if (~(BTRFS_BLOCK_GROUP_TYPE_MASK
| BTRFS_BLOCK_GROUP_PROFILE_MASK
) &
6361 btrfs_err(fs_info
, "unrecognized chunk type: %llu",
6362 ~(BTRFS_BLOCK_GROUP_TYPE_MASK
|
6363 BTRFS_BLOCK_GROUP_PROFILE_MASK
) &
6364 btrfs_chunk_type(leaf
, chunk
));
6368 if ((type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) == 0) {
6369 btrfs_err(fs_info
, "missing chunk type flag: 0x%llx", type
);
6373 if ((type
& BTRFS_BLOCK_GROUP_SYSTEM
) &&
6374 (type
& (BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
))) {
6376 "system chunk with data or metadata type: 0x%llx", type
);
6380 features
= btrfs_super_incompat_flags(fs_info
->super_copy
);
6381 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
6385 if ((type
& BTRFS_BLOCK_GROUP_METADATA
) &&
6386 (type
& BTRFS_BLOCK_GROUP_DATA
)) {
6388 "mixed chunk type in non-mixed mode: 0x%llx", type
);
6393 if ((type
& BTRFS_BLOCK_GROUP_RAID10
&& sub_stripes
!= 2) ||
6394 (type
& BTRFS_BLOCK_GROUP_RAID1
&& num_stripes
< 1) ||
6395 (type
& BTRFS_BLOCK_GROUP_RAID5
&& num_stripes
< 2) ||
6396 (type
& BTRFS_BLOCK_GROUP_RAID6
&& num_stripes
< 3) ||
6397 (type
& BTRFS_BLOCK_GROUP_DUP
&& num_stripes
> 2) ||
6398 ((type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) == 0 &&
6399 num_stripes
!= 1)) {
6401 "invalid num_stripes:sub_stripes %u:%u for profile %llu",
6402 num_stripes
, sub_stripes
,
6403 type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
);
6410 static void btrfs_report_missing_device(struct btrfs_fs_info
*fs_info
,
6411 u64 devid
, u8
*uuid
, bool error
)
6414 btrfs_err_rl(fs_info
, "devid %llu uuid %pU is missing",
6417 btrfs_warn_rl(fs_info
, "devid %llu uuid %pU is missing",
6421 static int read_one_chunk(struct btrfs_fs_info
*fs_info
, struct btrfs_key
*key
,
6422 struct extent_buffer
*leaf
,
6423 struct btrfs_chunk
*chunk
)
6425 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
6426 struct map_lookup
*map
;
6427 struct extent_map
*em
;
6431 u8 uuid
[BTRFS_UUID_SIZE
];
6436 logical
= key
->offset
;
6437 length
= btrfs_chunk_length(leaf
, chunk
);
6438 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6440 ret
= btrfs_check_chunk_valid(fs_info
, leaf
, chunk
, logical
);
6444 read_lock(&map_tree
->map_tree
.lock
);
6445 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
6446 read_unlock(&map_tree
->map_tree
.lock
);
6448 /* already mapped? */
6449 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
6450 free_extent_map(em
);
6453 free_extent_map(em
);
6456 em
= alloc_extent_map();
6459 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
6461 free_extent_map(em
);
6465 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
6466 em
->map_lookup
= map
;
6467 em
->start
= logical
;
6470 em
->block_start
= 0;
6471 em
->block_len
= em
->len
;
6473 map
->num_stripes
= num_stripes
;
6474 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
6475 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
6476 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6477 map
->type
= btrfs_chunk_type(leaf
, chunk
);
6478 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6479 map
->verified_stripes
= 0;
6480 for (i
= 0; i
< num_stripes
; i
++) {
6481 map
->stripes
[i
].physical
=
6482 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
6483 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
6484 read_extent_buffer(leaf
, uuid
, (unsigned long)
6485 btrfs_stripe_dev_uuid_nr(chunk
, i
),
6487 map
->stripes
[i
].dev
= btrfs_find_device(fs_info
, devid
,
6489 if (!map
->stripes
[i
].dev
&&
6490 !btrfs_test_opt(fs_info
, DEGRADED
)) {
6491 free_extent_map(em
);
6492 btrfs_report_missing_device(fs_info
, devid
, uuid
, true);
6495 if (!map
->stripes
[i
].dev
) {
6496 map
->stripes
[i
].dev
=
6497 add_missing_dev(fs_info
->fs_devices
, devid
,
6499 if (IS_ERR(map
->stripes
[i
].dev
)) {
6500 free_extent_map(em
);
6502 "failed to init missing dev %llu: %ld",
6503 devid
, PTR_ERR(map
->stripes
[i
].dev
));
6504 return PTR_ERR(map
->stripes
[i
].dev
);
6506 btrfs_report_missing_device(fs_info
, devid
, uuid
, false);
6508 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA
,
6509 &(map
->stripes
[i
].dev
->dev_state
));
6513 write_lock(&map_tree
->map_tree
.lock
);
6514 ret
= add_extent_mapping(&map_tree
->map_tree
, em
, 0);
6515 write_unlock(&map_tree
->map_tree
.lock
);
6518 "failed to add chunk map, start=%llu len=%llu: %d",
6519 em
->start
, em
->len
, ret
);
6521 free_extent_map(em
);
6526 static void fill_device_from_item(struct extent_buffer
*leaf
,
6527 struct btrfs_dev_item
*dev_item
,
6528 struct btrfs_device
*device
)
6532 device
->devid
= btrfs_device_id(leaf
, dev_item
);
6533 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
6534 device
->total_bytes
= device
->disk_total_bytes
;
6535 device
->commit_total_bytes
= device
->disk_total_bytes
;
6536 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
6537 device
->commit_bytes_used
= device
->bytes_used
;
6538 device
->type
= btrfs_device_type(leaf
, dev_item
);
6539 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
6540 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
6541 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
6542 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
6543 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
);
6545 ptr
= btrfs_device_uuid(dev_item
);
6546 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
6549 static struct btrfs_fs_devices
*open_seed_devices(struct btrfs_fs_info
*fs_info
,
6552 struct btrfs_fs_devices
*fs_devices
;
6555 lockdep_assert_held(&uuid_mutex
);
6558 fs_devices
= fs_info
->fs_devices
->seed
;
6559 while (fs_devices
) {
6560 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_FSID_SIZE
))
6563 fs_devices
= fs_devices
->seed
;
6566 fs_devices
= find_fsid(fsid
);
6568 if (!btrfs_test_opt(fs_info
, DEGRADED
))
6569 return ERR_PTR(-ENOENT
);
6571 fs_devices
= alloc_fs_devices(fsid
);
6572 if (IS_ERR(fs_devices
))
6575 fs_devices
->seeding
= 1;
6576 fs_devices
->opened
= 1;
6580 fs_devices
= clone_fs_devices(fs_devices
);
6581 if (IS_ERR(fs_devices
))
6584 ret
= open_fs_devices(fs_devices
, FMODE_READ
, fs_info
->bdev_holder
);
6586 free_fs_devices(fs_devices
);
6587 fs_devices
= ERR_PTR(ret
);
6591 if (!fs_devices
->seeding
) {
6592 close_fs_devices(fs_devices
);
6593 free_fs_devices(fs_devices
);
6594 fs_devices
= ERR_PTR(-EINVAL
);
6598 fs_devices
->seed
= fs_info
->fs_devices
->seed
;
6599 fs_info
->fs_devices
->seed
= fs_devices
;
6604 static int read_one_dev(struct btrfs_fs_info
*fs_info
,
6605 struct extent_buffer
*leaf
,
6606 struct btrfs_dev_item
*dev_item
)
6608 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6609 struct btrfs_device
*device
;
6612 u8 fs_uuid
[BTRFS_FSID_SIZE
];
6613 u8 dev_uuid
[BTRFS_UUID_SIZE
];
6615 devid
= btrfs_device_id(leaf
, dev_item
);
6616 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
6618 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
6621 if (memcmp(fs_uuid
, fs_info
->fsid
, BTRFS_FSID_SIZE
)) {
6622 fs_devices
= open_seed_devices(fs_info
, fs_uuid
);
6623 if (IS_ERR(fs_devices
))
6624 return PTR_ERR(fs_devices
);
6627 device
= btrfs_find_device(fs_info
, devid
, dev_uuid
, fs_uuid
);
6629 if (!btrfs_test_opt(fs_info
, DEGRADED
)) {
6630 btrfs_report_missing_device(fs_info
, devid
,
6635 device
= add_missing_dev(fs_devices
, devid
, dev_uuid
);
6636 if (IS_ERR(device
)) {
6638 "failed to add missing dev %llu: %ld",
6639 devid
, PTR_ERR(device
));
6640 return PTR_ERR(device
);
6642 btrfs_report_missing_device(fs_info
, devid
, dev_uuid
, false);
6644 if (!device
->bdev
) {
6645 if (!btrfs_test_opt(fs_info
, DEGRADED
)) {
6646 btrfs_report_missing_device(fs_info
,
6647 devid
, dev_uuid
, true);
6650 btrfs_report_missing_device(fs_info
, devid
,
6654 if (!device
->bdev
&&
6655 !test_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
)) {
6657 * this happens when a device that was properly setup
6658 * in the device info lists suddenly goes bad.
6659 * device->bdev is NULL, and so we have to set
6660 * device->missing to one here
6662 device
->fs_devices
->missing_devices
++;
6663 set_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
);
6666 /* Move the device to its own fs_devices */
6667 if (device
->fs_devices
!= fs_devices
) {
6668 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING
,
6669 &device
->dev_state
));
6671 list_move(&device
->dev_list
, &fs_devices
->devices
);
6672 device
->fs_devices
->num_devices
--;
6673 fs_devices
->num_devices
++;
6675 device
->fs_devices
->missing_devices
--;
6676 fs_devices
->missing_devices
++;
6678 device
->fs_devices
= fs_devices
;
6682 if (device
->fs_devices
!= fs_info
->fs_devices
) {
6683 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
));
6684 if (device
->generation
!=
6685 btrfs_device_generation(leaf
, dev_item
))
6689 fill_device_from_item(leaf
, dev_item
, device
);
6690 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &device
->dev_state
);
6691 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
) &&
6692 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
)) {
6693 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
6694 atomic64_add(device
->total_bytes
- device
->bytes_used
,
6695 &fs_info
->free_chunk_space
);
6701 int btrfs_read_sys_array(struct btrfs_fs_info
*fs_info
)
6703 struct btrfs_root
*root
= fs_info
->tree_root
;
6704 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
6705 struct extent_buffer
*sb
;
6706 struct btrfs_disk_key
*disk_key
;
6707 struct btrfs_chunk
*chunk
;
6709 unsigned long sb_array_offset
;
6716 struct btrfs_key key
;
6718 ASSERT(BTRFS_SUPER_INFO_SIZE
<= fs_info
->nodesize
);
6720 * This will create extent buffer of nodesize, superblock size is
6721 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6722 * overallocate but we can keep it as-is, only the first page is used.
6724 sb
= btrfs_find_create_tree_block(fs_info
, BTRFS_SUPER_INFO_OFFSET
);
6727 set_extent_buffer_uptodate(sb
);
6728 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, sb
, 0);
6730 * The sb extent buffer is artificial and just used to read the system array.
6731 * set_extent_buffer_uptodate() call does not properly mark all it's
6732 * pages up-to-date when the page is larger: extent does not cover the
6733 * whole page and consequently check_page_uptodate does not find all
6734 * the page's extents up-to-date (the hole beyond sb),
6735 * write_extent_buffer then triggers a WARN_ON.
6737 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6738 * but sb spans only this function. Add an explicit SetPageUptodate call
6739 * to silence the warning eg. on PowerPC 64.
6741 if (PAGE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
6742 SetPageUptodate(sb
->pages
[0]);
6744 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
6745 array_size
= btrfs_super_sys_array_size(super_copy
);
6747 array_ptr
= super_copy
->sys_chunk_array
;
6748 sb_array_offset
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
6751 while (cur_offset
< array_size
) {
6752 disk_key
= (struct btrfs_disk_key
*)array_ptr
;
6753 len
= sizeof(*disk_key
);
6754 if (cur_offset
+ len
> array_size
)
6755 goto out_short_read
;
6757 btrfs_disk_key_to_cpu(&key
, disk_key
);
6760 sb_array_offset
+= len
;
6763 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6764 chunk
= (struct btrfs_chunk
*)sb_array_offset
;
6766 * At least one btrfs_chunk with one stripe must be
6767 * present, exact stripe count check comes afterwards
6769 len
= btrfs_chunk_item_size(1);
6770 if (cur_offset
+ len
> array_size
)
6771 goto out_short_read
;
6773 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
6776 "invalid number of stripes %u in sys_array at offset %u",
6777 num_stripes
, cur_offset
);
6782 type
= btrfs_chunk_type(sb
, chunk
);
6783 if ((type
& BTRFS_BLOCK_GROUP_SYSTEM
) == 0) {
6785 "invalid chunk type %llu in sys_array at offset %u",
6791 len
= btrfs_chunk_item_size(num_stripes
);
6792 if (cur_offset
+ len
> array_size
)
6793 goto out_short_read
;
6795 ret
= read_one_chunk(fs_info
, &key
, sb
, chunk
);
6800 "unexpected item type %u in sys_array at offset %u",
6801 (u32
)key
.type
, cur_offset
);
6806 sb_array_offset
+= len
;
6809 clear_extent_buffer_uptodate(sb
);
6810 free_extent_buffer_stale(sb
);
6814 btrfs_err(fs_info
, "sys_array too short to read %u bytes at offset %u",
6816 clear_extent_buffer_uptodate(sb
);
6817 free_extent_buffer_stale(sb
);
6822 * Check if all chunks in the fs are OK for read-write degraded mount
6824 * If the @failing_dev is specified, it's accounted as missing.
6826 * Return true if all chunks meet the minimal RW mount requirements.
6827 * Return false if any chunk doesn't meet the minimal RW mount requirements.
6829 bool btrfs_check_rw_degradable(struct btrfs_fs_info
*fs_info
,
6830 struct btrfs_device
*failing_dev
)
6832 struct btrfs_mapping_tree
*map_tree
= &fs_info
->mapping_tree
;
6833 struct extent_map
*em
;
6837 read_lock(&map_tree
->map_tree
.lock
);
6838 em
= lookup_extent_mapping(&map_tree
->map_tree
, 0, (u64
)-1);
6839 read_unlock(&map_tree
->map_tree
.lock
);
6840 /* No chunk at all? Return false anyway */
6846 struct map_lookup
*map
;
6851 map
= em
->map_lookup
;
6853 btrfs_get_num_tolerated_disk_barrier_failures(
6855 for (i
= 0; i
< map
->num_stripes
; i
++) {
6856 struct btrfs_device
*dev
= map
->stripes
[i
].dev
;
6858 if (!dev
|| !dev
->bdev
||
6859 test_bit(BTRFS_DEV_STATE_MISSING
, &dev
->dev_state
) ||
6860 dev
->last_flush_error
)
6862 else if (failing_dev
&& failing_dev
== dev
)
6865 if (missing
> max_tolerated
) {
6868 "chunk %llu missing %d devices, max tolerance is %d for writeable mount",
6869 em
->start
, missing
, max_tolerated
);
6870 free_extent_map(em
);
6874 next_start
= extent_map_end(em
);
6875 free_extent_map(em
);
6877 read_lock(&map_tree
->map_tree
.lock
);
6878 em
= lookup_extent_mapping(&map_tree
->map_tree
, next_start
,
6879 (u64
)(-1) - next_start
);
6880 read_unlock(&map_tree
->map_tree
.lock
);
6886 int btrfs_read_chunk_tree(struct btrfs_fs_info
*fs_info
)
6888 struct btrfs_root
*root
= fs_info
->chunk_root
;
6889 struct btrfs_path
*path
;
6890 struct extent_buffer
*leaf
;
6891 struct btrfs_key key
;
6892 struct btrfs_key found_key
;
6897 path
= btrfs_alloc_path();
6902 * uuid_mutex is needed only if we are mounting a sprout FS
6903 * otherwise we don't need it.
6905 mutex_lock(&uuid_mutex
);
6906 mutex_lock(&fs_info
->chunk_mutex
);
6909 * Read all device items, and then all the chunk items. All
6910 * device items are found before any chunk item (their object id
6911 * is smaller than the lowest possible object id for a chunk
6912 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6914 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
6917 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6921 leaf
= path
->nodes
[0];
6922 slot
= path
->slots
[0];
6923 if (slot
>= btrfs_header_nritems(leaf
)) {
6924 ret
= btrfs_next_leaf(root
, path
);
6931 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6932 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
6933 struct btrfs_dev_item
*dev_item
;
6934 dev_item
= btrfs_item_ptr(leaf
, slot
,
6935 struct btrfs_dev_item
);
6936 ret
= read_one_dev(fs_info
, leaf
, dev_item
);
6940 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
6941 struct btrfs_chunk
*chunk
;
6942 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
6943 ret
= read_one_chunk(fs_info
, &found_key
, leaf
, chunk
);
6951 * After loading chunk tree, we've got all device information,
6952 * do another round of validation checks.
6954 if (total_dev
!= fs_info
->fs_devices
->total_devices
) {
6956 "super_num_devices %llu mismatch with num_devices %llu found here",
6957 btrfs_super_num_devices(fs_info
->super_copy
),
6962 if (btrfs_super_total_bytes(fs_info
->super_copy
) <
6963 fs_info
->fs_devices
->total_rw_bytes
) {
6965 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
6966 btrfs_super_total_bytes(fs_info
->super_copy
),
6967 fs_info
->fs_devices
->total_rw_bytes
);
6973 mutex_unlock(&fs_info
->chunk_mutex
);
6974 mutex_unlock(&uuid_mutex
);
6976 btrfs_free_path(path
);
6980 void btrfs_init_devices_late(struct btrfs_fs_info
*fs_info
)
6982 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6983 struct btrfs_device
*device
;
6985 while (fs_devices
) {
6986 mutex_lock(&fs_devices
->device_list_mutex
);
6987 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
)
6988 device
->fs_info
= fs_info
;
6989 mutex_unlock(&fs_devices
->device_list_mutex
);
6991 fs_devices
= fs_devices
->seed
;
6995 static void __btrfs_reset_dev_stats(struct btrfs_device
*dev
)
6999 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7000 btrfs_dev_stat_reset(dev
, i
);
7003 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
7005 struct btrfs_key key
;
7006 struct btrfs_key found_key
;
7007 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
7008 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7009 struct extent_buffer
*eb
;
7012 struct btrfs_device
*device
;
7013 struct btrfs_path
*path
= NULL
;
7016 path
= btrfs_alloc_path();
7022 mutex_lock(&fs_devices
->device_list_mutex
);
7023 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
7025 struct btrfs_dev_stats_item
*ptr
;
7027 key
.objectid
= BTRFS_DEV_STATS_OBJECTID
;
7028 key
.type
= BTRFS_PERSISTENT_ITEM_KEY
;
7029 key
.offset
= device
->devid
;
7030 ret
= btrfs_search_slot(NULL
, dev_root
, &key
, path
, 0, 0);
7032 __btrfs_reset_dev_stats(device
);
7033 device
->dev_stats_valid
= 1;
7034 btrfs_release_path(path
);
7037 slot
= path
->slots
[0];
7038 eb
= path
->nodes
[0];
7039 btrfs_item_key_to_cpu(eb
, &found_key
, slot
);
7040 item_size
= btrfs_item_size_nr(eb
, slot
);
7042 ptr
= btrfs_item_ptr(eb
, slot
,
7043 struct btrfs_dev_stats_item
);
7045 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
7046 if (item_size
>= (1 + i
) * sizeof(__le64
))
7047 btrfs_dev_stat_set(device
, i
,
7048 btrfs_dev_stats_value(eb
, ptr
, i
));
7050 btrfs_dev_stat_reset(device
, i
);
7053 device
->dev_stats_valid
= 1;
7054 btrfs_dev_stat_print_on_load(device
);
7055 btrfs_release_path(path
);
7057 mutex_unlock(&fs_devices
->device_list_mutex
);
7060 btrfs_free_path(path
);
7061 return ret
< 0 ? ret
: 0;
7064 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
7065 struct btrfs_device
*device
)
7067 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
7068 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
7069 struct btrfs_path
*path
;
7070 struct btrfs_key key
;
7071 struct extent_buffer
*eb
;
7072 struct btrfs_dev_stats_item
*ptr
;
7076 key
.objectid
= BTRFS_DEV_STATS_OBJECTID
;
7077 key
.type
= BTRFS_PERSISTENT_ITEM_KEY
;
7078 key
.offset
= device
->devid
;
7080 path
= btrfs_alloc_path();
7083 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
7085 btrfs_warn_in_rcu(fs_info
,
7086 "error %d while searching for dev_stats item for device %s",
7087 ret
, rcu_str_deref(device
->name
));
7092 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
7093 /* need to delete old one and insert a new one */
7094 ret
= btrfs_del_item(trans
, dev_root
, path
);
7096 btrfs_warn_in_rcu(fs_info
,
7097 "delete too small dev_stats item for device %s failed %d",
7098 rcu_str_deref(device
->name
), ret
);
7105 /* need to insert a new item */
7106 btrfs_release_path(path
);
7107 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
7108 &key
, sizeof(*ptr
));
7110 btrfs_warn_in_rcu(fs_info
,
7111 "insert dev_stats item for device %s failed %d",
7112 rcu_str_deref(device
->name
), ret
);
7117 eb
= path
->nodes
[0];
7118 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
7119 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7120 btrfs_set_dev_stats_value(eb
, ptr
, i
,
7121 btrfs_dev_stat_read(device
, i
));
7122 btrfs_mark_buffer_dirty(eb
);
7125 btrfs_free_path(path
);
7130 * called from commit_transaction. Writes all changed device stats to disk.
7132 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
,
7133 struct btrfs_fs_info
*fs_info
)
7135 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7136 struct btrfs_device
*device
;
7140 mutex_lock(&fs_devices
->device_list_mutex
);
7141 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
7142 stats_cnt
= atomic_read(&device
->dev_stats_ccnt
);
7143 if (!device
->dev_stats_valid
|| stats_cnt
== 0)
7148 * There is a LOAD-LOAD control dependency between the value of
7149 * dev_stats_ccnt and updating the on-disk values which requires
7150 * reading the in-memory counters. Such control dependencies
7151 * require explicit read memory barriers.
7153 * This memory barriers pairs with smp_mb__before_atomic in
7154 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7155 * barrier implied by atomic_xchg in
7156 * btrfs_dev_stats_read_and_reset
7160 ret
= update_dev_stat_item(trans
, device
);
7162 atomic_sub(stats_cnt
, &device
->dev_stats_ccnt
);
7164 mutex_unlock(&fs_devices
->device_list_mutex
);
7169 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
7171 btrfs_dev_stat_inc(dev
, index
);
7172 btrfs_dev_stat_print_on_error(dev
);
7175 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
7177 if (!dev
->dev_stats_valid
)
7179 btrfs_err_rl_in_rcu(dev
->fs_info
,
7180 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7181 rcu_str_deref(dev
->name
),
7182 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
7183 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
7184 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
7185 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
7186 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
7189 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
7193 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7194 if (btrfs_dev_stat_read(dev
, i
) != 0)
7196 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
7197 return; /* all values == 0, suppress message */
7199 btrfs_info_in_rcu(dev
->fs_info
,
7200 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7201 rcu_str_deref(dev
->name
),
7202 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
7203 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
7204 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
7205 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
7206 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
7209 int btrfs_get_dev_stats(struct btrfs_fs_info
*fs_info
,
7210 struct btrfs_ioctl_get_dev_stats
*stats
)
7212 struct btrfs_device
*dev
;
7213 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7216 mutex_lock(&fs_devices
->device_list_mutex
);
7217 dev
= btrfs_find_device(fs_info
, stats
->devid
, NULL
, NULL
);
7218 mutex_unlock(&fs_devices
->device_list_mutex
);
7221 btrfs_warn(fs_info
, "get dev_stats failed, device not found");
7223 } else if (!dev
->dev_stats_valid
) {
7224 btrfs_warn(fs_info
, "get dev_stats failed, not yet valid");
7226 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
7227 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
7228 if (stats
->nr_items
> i
)
7230 btrfs_dev_stat_read_and_reset(dev
, i
);
7232 btrfs_dev_stat_reset(dev
, i
);
7235 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7236 if (stats
->nr_items
> i
)
7237 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
7239 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
7240 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
7244 void btrfs_scratch_superblocks(struct block_device
*bdev
, const char *device_path
)
7246 struct buffer_head
*bh
;
7247 struct btrfs_super_block
*disk_super
;
7253 for (copy_num
= 0; copy_num
< BTRFS_SUPER_MIRROR_MAX
;
7256 if (btrfs_read_dev_one_super(bdev
, copy_num
, &bh
))
7259 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
7261 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
7262 set_buffer_dirty(bh
);
7263 sync_dirty_buffer(bh
);
7267 /* Notify udev that device has changed */
7268 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
7270 /* Update ctime/mtime for device path for libblkid */
7271 update_dev_time(device_path
);
7275 * Update the size of all devices, which is used for writing out the
7278 void btrfs_update_commit_device_size(struct btrfs_fs_info
*fs_info
)
7280 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7281 struct btrfs_device
*curr
, *next
;
7283 if (list_empty(&fs_devices
->resized_devices
))
7286 mutex_lock(&fs_devices
->device_list_mutex
);
7287 mutex_lock(&fs_info
->chunk_mutex
);
7288 list_for_each_entry_safe(curr
, next
, &fs_devices
->resized_devices
,
7290 list_del_init(&curr
->resized_list
);
7291 curr
->commit_total_bytes
= curr
->disk_total_bytes
;
7293 mutex_unlock(&fs_info
->chunk_mutex
);
7294 mutex_unlock(&fs_devices
->device_list_mutex
);
7297 /* Must be invoked during the transaction commit */
7298 void btrfs_update_commit_device_bytes_used(struct btrfs_transaction
*trans
)
7300 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
7301 struct extent_map
*em
;
7302 struct map_lookup
*map
;
7303 struct btrfs_device
*dev
;
7306 if (list_empty(&trans
->pending_chunks
))
7309 /* In order to kick the device replace finish process */
7310 mutex_lock(&fs_info
->chunk_mutex
);
7311 list_for_each_entry(em
, &trans
->pending_chunks
, list
) {
7312 map
= em
->map_lookup
;
7314 for (i
= 0; i
< map
->num_stripes
; i
++) {
7315 dev
= map
->stripes
[i
].dev
;
7316 dev
->commit_bytes_used
= dev
->bytes_used
;
7319 mutex_unlock(&fs_info
->chunk_mutex
);
7322 void btrfs_set_fs_info_ptr(struct btrfs_fs_info
*fs_info
)
7324 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7325 while (fs_devices
) {
7326 fs_devices
->fs_info
= fs_info
;
7327 fs_devices
= fs_devices
->seed
;
7331 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info
*fs_info
)
7333 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7334 while (fs_devices
) {
7335 fs_devices
->fs_info
= NULL
;
7336 fs_devices
= fs_devices
->seed
;
7341 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7343 int btrfs_bg_type_to_factor(u64 flags
)
7345 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
7346 BTRFS_BLOCK_GROUP_RAID10
))
7352 static u64
calc_stripe_length(u64 type
, u64 chunk_len
, int num_stripes
)
7354 int index
= btrfs_bg_flags_to_raid_index(type
);
7355 int ncopies
= btrfs_raid_array
[index
].ncopies
;
7358 switch (type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
7359 case BTRFS_BLOCK_GROUP_RAID5
:
7360 data_stripes
= num_stripes
- 1;
7362 case BTRFS_BLOCK_GROUP_RAID6
:
7363 data_stripes
= num_stripes
- 2;
7366 data_stripes
= num_stripes
/ ncopies
;
7369 return div_u64(chunk_len
, data_stripes
);
7372 static int verify_one_dev_extent(struct btrfs_fs_info
*fs_info
,
7373 u64 chunk_offset
, u64 devid
,
7374 u64 physical_offset
, u64 physical_len
)
7376 struct extent_map_tree
*em_tree
= &fs_info
->mapping_tree
.map_tree
;
7377 struct extent_map
*em
;
7378 struct map_lookup
*map
;
7384 read_lock(&em_tree
->lock
);
7385 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
7386 read_unlock(&em_tree
->lock
);
7390 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7391 physical_offset
, devid
);
7396 map
= em
->map_lookup
;
7397 stripe_len
= calc_stripe_length(map
->type
, em
->len
, map
->num_stripes
);
7398 if (physical_len
!= stripe_len
) {
7400 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7401 physical_offset
, devid
, em
->start
, physical_len
,
7407 for (i
= 0; i
< map
->num_stripes
; i
++) {
7408 if (map
->stripes
[i
].dev
->devid
== devid
&&
7409 map
->stripes
[i
].physical
== physical_offset
) {
7411 if (map
->verified_stripes
>= map
->num_stripes
) {
7413 "too many dev extents for chunk %llu found",
7418 map
->verified_stripes
++;
7424 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7425 physical_offset
, devid
);
7429 free_extent_map(em
);
7433 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info
*fs_info
)
7435 struct extent_map_tree
*em_tree
= &fs_info
->mapping_tree
.map_tree
;
7436 struct extent_map
*em
;
7437 struct rb_node
*node
;
7440 read_lock(&em_tree
->lock
);
7441 for (node
= rb_first(&em_tree
->map
); node
; node
= rb_next(node
)) {
7442 em
= rb_entry(node
, struct extent_map
, rb_node
);
7443 if (em
->map_lookup
->num_stripes
!=
7444 em
->map_lookup
->verified_stripes
) {
7446 "chunk %llu has missing dev extent, have %d expect %d",
7447 em
->start
, em
->map_lookup
->verified_stripes
,
7448 em
->map_lookup
->num_stripes
);
7454 read_unlock(&em_tree
->lock
);
7459 * Ensure that all dev extents are mapped to correct chunk, otherwise
7460 * later chunk allocation/free would cause unexpected behavior.
7462 * NOTE: This will iterate through the whole device tree, which should be of
7463 * the same size level as the chunk tree. This slightly increases mount time.
7465 int btrfs_verify_dev_extents(struct btrfs_fs_info
*fs_info
)
7467 struct btrfs_path
*path
;
7468 struct btrfs_root
*root
= fs_info
->dev_root
;
7469 struct btrfs_key key
;
7473 key
.type
= BTRFS_DEV_EXTENT_KEY
;
7476 path
= btrfs_alloc_path();
7480 path
->reada
= READA_FORWARD
;
7481 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
7485 if (path
->slots
[0] >= btrfs_header_nritems(path
->nodes
[0])) {
7486 ret
= btrfs_next_item(root
, path
);
7489 /* No dev extents at all? Not good */
7496 struct extent_buffer
*leaf
= path
->nodes
[0];
7497 struct btrfs_dev_extent
*dext
;
7498 int slot
= path
->slots
[0];
7500 u64 physical_offset
;
7504 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
7505 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
7507 devid
= key
.objectid
;
7508 physical_offset
= key
.offset
;
7510 dext
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dev_extent
);
7511 chunk_offset
= btrfs_dev_extent_chunk_offset(leaf
, dext
);
7512 physical_len
= btrfs_dev_extent_length(leaf
, dext
);
7514 ret
= verify_one_dev_extent(fs_info
, chunk_offset
, devid
,
7515 physical_offset
, physical_len
);
7518 ret
= btrfs_next_item(root
, path
);
7527 /* Ensure all chunks have corresponding dev extents */
7528 ret
= verify_chunk_dev_extent_mapping(fs_info
);
7530 btrfs_free_path(path
);