1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/sched/mm.h>
8 #include <linux/slab.h>
9 #include <linux/ratelimit.h>
10 #include <linux/kthread.h>
11 #include <linux/semaphore.h>
12 #include <linux/uuid.h>
13 #include <linux/list_sort.h>
14 #include <linux/namei.h>
17 #include "extent_map.h"
19 #include "transaction.h"
20 #include "print-tree.h"
23 #include "rcu-string.h"
24 #include "dev-replace.h"
26 #include "tree-checker.h"
27 #include "space-info.h"
28 #include "block-group.h"
32 #include "accessors.h"
33 #include "uuid-tree.h"
35 #include "relocation.h"
38 #include "raid-stripe-tree.h"
40 #define BTRFS_BLOCK_GROUP_STRIPE_MASK (BTRFS_BLOCK_GROUP_RAID0 | \
41 BTRFS_BLOCK_GROUP_RAID10 | \
42 BTRFS_BLOCK_GROUP_RAID56_MASK)
44 const struct btrfs_raid_attr btrfs_raid_array
[BTRFS_NR_RAID_TYPES
] = {
45 [BTRFS_RAID_RAID10
] = {
48 .devs_max
= 0, /* 0 == as many as possible */
50 .tolerated_failures
= 1,
54 .raid_name
= "raid10",
55 .bg_flag
= BTRFS_BLOCK_GROUP_RAID10
,
56 .mindev_error
= BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET
,
58 [BTRFS_RAID_RAID1
] = {
63 .tolerated_failures
= 1,
68 .bg_flag
= BTRFS_BLOCK_GROUP_RAID1
,
69 .mindev_error
= BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET
,
71 [BTRFS_RAID_RAID1C3
] = {
76 .tolerated_failures
= 2,
80 .raid_name
= "raid1c3",
81 .bg_flag
= BTRFS_BLOCK_GROUP_RAID1C3
,
82 .mindev_error
= BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET
,
84 [BTRFS_RAID_RAID1C4
] = {
89 .tolerated_failures
= 3,
93 .raid_name
= "raid1c4",
94 .bg_flag
= BTRFS_BLOCK_GROUP_RAID1C4
,
95 .mindev_error
= BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET
,
102 .tolerated_failures
= 0,
107 .bg_flag
= BTRFS_BLOCK_GROUP_DUP
,
110 [BTRFS_RAID_RAID0
] = {
115 .tolerated_failures
= 0,
119 .raid_name
= "raid0",
120 .bg_flag
= BTRFS_BLOCK_GROUP_RAID0
,
123 [BTRFS_RAID_SINGLE
] = {
128 .tolerated_failures
= 0,
132 .raid_name
= "single",
136 [BTRFS_RAID_RAID5
] = {
141 .tolerated_failures
= 1,
145 .raid_name
= "raid5",
146 .bg_flag
= BTRFS_BLOCK_GROUP_RAID5
,
147 .mindev_error
= BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET
,
149 [BTRFS_RAID_RAID6
] = {
154 .tolerated_failures
= 2,
158 .raid_name
= "raid6",
159 .bg_flag
= BTRFS_BLOCK_GROUP_RAID6
,
160 .mindev_error
= BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET
,
165 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which
166 * can be used as index to access btrfs_raid_array[].
168 enum btrfs_raid_types __attribute_const__
btrfs_bg_flags_to_raid_index(u64 flags
)
170 const u64 profile
= (flags
& BTRFS_BLOCK_GROUP_PROFILE_MASK
);
173 return BTRFS_RAID_SINGLE
;
175 return BTRFS_BG_FLAG_TO_INDEX(profile
);
178 const char *btrfs_bg_type_to_raid_name(u64 flags
)
180 const int index
= btrfs_bg_flags_to_raid_index(flags
);
182 if (index
>= BTRFS_NR_RAID_TYPES
)
185 return btrfs_raid_array
[index
].raid_name
;
188 int btrfs_nr_parity_stripes(u64 type
)
190 enum btrfs_raid_types index
= btrfs_bg_flags_to_raid_index(type
);
192 return btrfs_raid_array
[index
].nparity
;
196 * Fill @buf with textual description of @bg_flags, no more than @size_buf
197 * bytes including terminating null byte.
199 void btrfs_describe_block_groups(u64 bg_flags
, char *buf
, u32 size_buf
)
204 u64 flags
= bg_flags
;
205 u32 size_bp
= size_buf
;
212 #define DESCRIBE_FLAG(flag, desc) \
214 if (flags & (flag)) { \
215 ret = snprintf(bp, size_bp, "%s|", (desc)); \
216 if (ret < 0 || ret >= size_bp) \
224 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA
, "data");
225 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM
, "system");
226 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA
, "metadata");
228 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE
, "single");
229 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
230 DESCRIBE_FLAG(btrfs_raid_array
[i
].bg_flag
,
231 btrfs_raid_array
[i
].raid_name
);
235 ret
= snprintf(bp
, size_bp
, "0x%llx|", flags
);
239 if (size_bp
< size_buf
)
240 buf
[size_buf
- size_bp
- 1] = '\0'; /* remove last | */
243 * The text is trimmed, it's up to the caller to provide sufficiently
249 static int init_first_rw_device(struct btrfs_trans_handle
*trans
);
250 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info
*fs_info
);
251 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*device
);
257 * There are several mutexes that protect manipulation of devices and low-level
258 * structures like chunks but not block groups, extents or files
260 * uuid_mutex (global lock)
261 * ------------------------
262 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
263 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
264 * device) or requested by the device= mount option
266 * the mutex can be very coarse and can cover long-running operations
268 * protects: updates to fs_devices counters like missing devices, rw devices,
269 * seeding, structure cloning, opening/closing devices at mount/umount time
271 * global::fs_devs - add, remove, updates to the global list
273 * does not protect: manipulation of the fs_devices::devices list in general
274 * but in mount context it could be used to exclude list modifications by eg.
277 * btrfs_device::name - renames (write side), read is RCU
279 * fs_devices::device_list_mutex (per-fs, with RCU)
280 * ------------------------------------------------
281 * protects updates to fs_devices::devices, ie. adding and deleting
283 * simple list traversal with read-only actions can be done with RCU protection
285 * may be used to exclude some operations from running concurrently without any
286 * modifications to the list (see write_all_supers)
288 * Is not required at mount and close times, because our device list is
289 * protected by the uuid_mutex at that point.
293 * protects balance structures (status, state) and context accessed from
294 * several places (internally, ioctl)
298 * protects chunks, adding or removing during allocation, trim or when a new
299 * device is added/removed. Additionally it also protects post_commit_list of
300 * individual devices, since they can be added to the transaction's
301 * post_commit_list only with chunk_mutex held.
305 * a big lock that is held by the cleaner thread and prevents running subvolume
306 * cleaning together with relocation or delayed iputs
318 * Exclusive operations
319 * ====================
321 * Maintains the exclusivity of the following operations that apply to the
322 * whole filesystem and cannot run in parallel.
327 * - Device replace (*)
330 * The device operations (as above) can be in one of the following states:
336 * Only device operations marked with (*) can go into the Paused state for the
339 * - ioctl (only Balance can be Paused through ioctl)
340 * - filesystem remounted as read-only
341 * - filesystem unmounted and mounted as read-only
342 * - system power-cycle and filesystem mounted as read-only
343 * - filesystem or device errors leading to forced read-only
345 * The status of exclusive operation is set and cleared atomically.
346 * During the course of Paused state, fs_info::exclusive_operation remains set.
347 * A device operation in Paused or Running state can be canceled or resumed
348 * either by ioctl (Balance only) or when remounted as read-write.
349 * The exclusive status is cleared when the device operation is canceled or
353 DEFINE_MUTEX(uuid_mutex
);
354 static LIST_HEAD(fs_uuids
);
355 struct list_head
* __attribute_const__
btrfs_get_fs_uuids(void)
361 * Allocate new btrfs_fs_devices structure identified by a fsid.
363 * @fsid: if not NULL, copy the UUID to fs_devices::fsid and to
364 * fs_devices::metadata_fsid
366 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
367 * The returned struct is not linked onto any lists and can be destroyed with
368 * kfree() right away.
370 static struct btrfs_fs_devices
*alloc_fs_devices(const u8
*fsid
)
372 struct btrfs_fs_devices
*fs_devs
;
374 fs_devs
= kzalloc(sizeof(*fs_devs
), GFP_KERNEL
);
376 return ERR_PTR(-ENOMEM
);
378 mutex_init(&fs_devs
->device_list_mutex
);
380 INIT_LIST_HEAD(&fs_devs
->devices
);
381 INIT_LIST_HEAD(&fs_devs
->alloc_list
);
382 INIT_LIST_HEAD(&fs_devs
->fs_list
);
383 INIT_LIST_HEAD(&fs_devs
->seed_list
);
386 memcpy(fs_devs
->fsid
, fsid
, BTRFS_FSID_SIZE
);
387 memcpy(fs_devs
->metadata_uuid
, fsid
, BTRFS_FSID_SIZE
);
393 static void btrfs_free_device(struct btrfs_device
*device
)
395 WARN_ON(!list_empty(&device
->post_commit_list
));
396 rcu_string_free(device
->name
);
397 extent_io_tree_release(&device
->alloc_state
);
398 btrfs_destroy_dev_zone_info(device
);
402 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
404 struct btrfs_device
*device
;
406 WARN_ON(fs_devices
->opened
);
407 while (!list_empty(&fs_devices
->devices
)) {
408 device
= list_entry(fs_devices
->devices
.next
,
409 struct btrfs_device
, dev_list
);
410 list_del(&device
->dev_list
);
411 btrfs_free_device(device
);
416 void __exit
btrfs_cleanup_fs_uuids(void)
418 struct btrfs_fs_devices
*fs_devices
;
420 while (!list_empty(&fs_uuids
)) {
421 fs_devices
= list_entry(fs_uuids
.next
,
422 struct btrfs_fs_devices
, fs_list
);
423 list_del(&fs_devices
->fs_list
);
424 free_fs_devices(fs_devices
);
428 static bool match_fsid_fs_devices(const struct btrfs_fs_devices
*fs_devices
,
429 const u8
*fsid
, const u8
*metadata_fsid
)
431 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) != 0)
437 if (memcmp(metadata_fsid
, fs_devices
->metadata_uuid
, BTRFS_FSID_SIZE
) != 0)
443 static noinline
struct btrfs_fs_devices
*find_fsid(
444 const u8
*fsid
, const u8
*metadata_fsid
)
446 struct btrfs_fs_devices
*fs_devices
;
450 /* Handle non-split brain cases */
451 list_for_each_entry(fs_devices
, &fs_uuids
, fs_list
) {
452 if (match_fsid_fs_devices(fs_devices
, fsid
, metadata_fsid
))
459 btrfs_get_bdev_and_sb(const char *device_path
, blk_mode_t flags
, void *holder
,
460 int flush
, struct bdev_handle
**bdev_handle
,
461 struct btrfs_super_block
**disk_super
)
463 struct block_device
*bdev
;
466 *bdev_handle
= bdev_open_by_path(device_path
, flags
, holder
, NULL
);
468 if (IS_ERR(*bdev_handle
)) {
469 ret
= PTR_ERR(*bdev_handle
);
472 bdev
= (*bdev_handle
)->bdev
;
476 ret
= set_blocksize(bdev
, BTRFS_BDEV_BLOCKSIZE
);
478 bdev_release(*bdev_handle
);
481 invalidate_bdev(bdev
);
482 *disk_super
= btrfs_read_dev_super(bdev
);
483 if (IS_ERR(*disk_super
)) {
484 ret
= PTR_ERR(*disk_super
);
485 bdev_release(*bdev_handle
);
497 * Search and remove all stale devices (which are not mounted). When both
498 * inputs are NULL, it will search and release all stale devices.
500 * @devt: Optional. When provided will it release all unmounted devices
501 * matching this devt only.
502 * @skip_device: Optional. Will skip this device when searching for the stale
505 * Return: 0 for success or if @devt is 0.
506 * -EBUSY if @devt is a mounted device.
507 * -ENOENT if @devt does not match any device in the list.
509 static int btrfs_free_stale_devices(dev_t devt
, struct btrfs_device
*skip_device
)
511 struct btrfs_fs_devices
*fs_devices
, *tmp_fs_devices
;
512 struct btrfs_device
*device
, *tmp_device
;
516 lockdep_assert_held(&uuid_mutex
);
518 /* Return good status if there is no instance of devt. */
520 list_for_each_entry_safe(fs_devices
, tmp_fs_devices
, &fs_uuids
, fs_list
) {
522 mutex_lock(&fs_devices
->device_list_mutex
);
523 list_for_each_entry_safe(device
, tmp_device
,
524 &fs_devices
->devices
, dev_list
) {
525 if (skip_device
&& skip_device
== device
)
527 if (devt
&& devt
!= device
->devt
)
529 if (fs_devices
->opened
) {
535 /* delete the stale device */
536 fs_devices
->num_devices
--;
537 list_del(&device
->dev_list
);
538 btrfs_free_device(device
);
542 mutex_unlock(&fs_devices
->device_list_mutex
);
544 if (fs_devices
->num_devices
== 0) {
545 btrfs_sysfs_remove_fsid(fs_devices
);
546 list_del(&fs_devices
->fs_list
);
547 free_fs_devices(fs_devices
);
551 /* If there is at least one freed device return 0. */
558 static struct btrfs_fs_devices
*find_fsid_by_device(
559 struct btrfs_super_block
*disk_super
,
560 dev_t devt
, bool *same_fsid_diff_dev
)
562 struct btrfs_fs_devices
*fsid_fs_devices
;
563 struct btrfs_fs_devices
*devt_fs_devices
;
564 const bool has_metadata_uuid
= (btrfs_super_incompat_flags(disk_super
) &
565 BTRFS_FEATURE_INCOMPAT_METADATA_UUID
);
566 bool found_by_devt
= false;
568 /* Find the fs_device by the usual method, if found use it. */
569 fsid_fs_devices
= find_fsid(disk_super
->fsid
,
570 has_metadata_uuid
? disk_super
->metadata_uuid
: NULL
);
572 /* The temp_fsid feature is supported only with single device filesystem. */
573 if (btrfs_super_num_devices(disk_super
) != 1)
574 return fsid_fs_devices
;
577 * A seed device is an integral component of the sprout device, which
578 * functions as a multi-device filesystem. So, temp-fsid feature is
581 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
)
582 return fsid_fs_devices
;
584 /* Try to find a fs_devices by matching devt. */
585 list_for_each_entry(devt_fs_devices
, &fs_uuids
, fs_list
) {
586 struct btrfs_device
*device
;
588 list_for_each_entry(device
, &devt_fs_devices
->devices
, dev_list
) {
589 if (device
->devt
== devt
) {
590 found_by_devt
= true;
599 /* Existing device. */
600 if (fsid_fs_devices
== NULL
) {
601 if (devt_fs_devices
->opened
== 0) {
605 /* temp_fsid is mounting a subvol. */
606 return devt_fs_devices
;
609 /* Regular or temp_fsid device mounting a subvol. */
610 return devt_fs_devices
;
614 if (fsid_fs_devices
== NULL
) {
617 /* sb::fsid is already used create a new temp_fsid. */
618 *same_fsid_diff_dev
= true;
627 * This is only used on mount, and we are protected from competing things
628 * messing with our fs_devices by the uuid_mutex, thus we do not need the
629 * fs_devices->device_list_mutex here.
631 static int btrfs_open_one_device(struct btrfs_fs_devices
*fs_devices
,
632 struct btrfs_device
*device
, blk_mode_t flags
,
635 struct bdev_handle
*bdev_handle
;
636 struct btrfs_super_block
*disk_super
;
645 ret
= btrfs_get_bdev_and_sb(device
->name
->str
, flags
, holder
, 1,
646 &bdev_handle
, &disk_super
);
650 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
651 if (devid
!= device
->devid
)
652 goto error_free_page
;
654 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
, BTRFS_UUID_SIZE
))
655 goto error_free_page
;
657 device
->generation
= btrfs_super_generation(disk_super
);
659 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
660 if (btrfs_super_incompat_flags(disk_super
) &
661 BTRFS_FEATURE_INCOMPAT_METADATA_UUID
) {
663 "BTRFS: Invalid seeding and uuid-changed device detected\n");
664 goto error_free_page
;
667 clear_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
);
668 fs_devices
->seeding
= true;
670 if (bdev_read_only(bdev_handle
->bdev
))
671 clear_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
);
673 set_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
);
676 if (!bdev_nonrot(bdev_handle
->bdev
))
677 fs_devices
->rotating
= true;
679 if (bdev_max_discard_sectors(bdev_handle
->bdev
))
680 fs_devices
->discardable
= true;
682 device
->bdev_handle
= bdev_handle
;
683 device
->bdev
= bdev_handle
->bdev
;
684 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &device
->dev_state
);
686 fs_devices
->open_devices
++;
687 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
) &&
688 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
689 fs_devices
->rw_devices
++;
690 list_add_tail(&device
->dev_alloc_list
, &fs_devices
->alloc_list
);
692 btrfs_release_disk_super(disk_super
);
697 btrfs_release_disk_super(disk_super
);
698 bdev_release(bdev_handle
);
703 u8
*btrfs_sb_fsid_ptr(struct btrfs_super_block
*sb
)
705 bool has_metadata_uuid
= (btrfs_super_incompat_flags(sb
) &
706 BTRFS_FEATURE_INCOMPAT_METADATA_UUID
);
708 return has_metadata_uuid
? sb
->metadata_uuid
: sb
->fsid
;
712 * Add new device to list of registered devices
715 * device pointer which was just added or updated when successful
716 * error pointer when failed
718 static noinline
struct btrfs_device
*device_list_add(const char *path
,
719 struct btrfs_super_block
*disk_super
,
720 bool *new_device_added
)
722 struct btrfs_device
*device
;
723 struct btrfs_fs_devices
*fs_devices
= NULL
;
724 struct rcu_string
*name
;
725 u64 found_transid
= btrfs_super_generation(disk_super
);
726 u64 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
729 bool same_fsid_diff_dev
= false;
730 bool has_metadata_uuid
= (btrfs_super_incompat_flags(disk_super
) &
731 BTRFS_FEATURE_INCOMPAT_METADATA_UUID
);
733 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_CHANGING_FSID_V2
) {
735 "device %s has incomplete metadata_uuid change, please use btrfstune to complete",
737 return ERR_PTR(-EAGAIN
);
740 error
= lookup_bdev(path
, &path_devt
);
742 btrfs_err(NULL
, "failed to lookup block device for path %s: %d",
744 return ERR_PTR(error
);
747 fs_devices
= find_fsid_by_device(disk_super
, path_devt
, &same_fsid_diff_dev
);
750 fs_devices
= alloc_fs_devices(disk_super
->fsid
);
751 if (IS_ERR(fs_devices
))
752 return ERR_CAST(fs_devices
);
754 if (has_metadata_uuid
)
755 memcpy(fs_devices
->metadata_uuid
,
756 disk_super
->metadata_uuid
, BTRFS_FSID_SIZE
);
758 if (same_fsid_diff_dev
) {
759 generate_random_uuid(fs_devices
->fsid
);
760 fs_devices
->temp_fsid
= true;
761 pr_info("BTRFS: device %s using temp-fsid %pU\n",
762 path
, fs_devices
->fsid
);
765 mutex_lock(&fs_devices
->device_list_mutex
);
766 list_add(&fs_devices
->fs_list
, &fs_uuids
);
770 struct btrfs_dev_lookup_args args
= {
772 .uuid
= disk_super
->dev_item
.uuid
,
775 mutex_lock(&fs_devices
->device_list_mutex
);
776 device
= btrfs_find_device(fs_devices
, &args
);
778 if (found_transid
> fs_devices
->latest_generation
) {
779 memcpy(fs_devices
->fsid
, disk_super
->fsid
,
781 memcpy(fs_devices
->metadata_uuid
,
782 btrfs_sb_fsid_ptr(disk_super
), BTRFS_FSID_SIZE
);
787 unsigned int nofs_flag
;
789 if (fs_devices
->opened
) {
791 "device %s belongs to fsid %pU, and the fs is already mounted, scanned by %s (%d)",
792 path
, fs_devices
->fsid
, current
->comm
,
793 task_pid_nr(current
));
794 mutex_unlock(&fs_devices
->device_list_mutex
);
795 return ERR_PTR(-EBUSY
);
798 nofs_flag
= memalloc_nofs_save();
799 device
= btrfs_alloc_device(NULL
, &devid
,
800 disk_super
->dev_item
.uuid
, path
);
801 memalloc_nofs_restore(nofs_flag
);
802 if (IS_ERR(device
)) {
803 mutex_unlock(&fs_devices
->device_list_mutex
);
804 /* we can safely leave the fs_devices entry around */
808 device
->devt
= path_devt
;
810 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
811 fs_devices
->num_devices
++;
813 device
->fs_devices
= fs_devices
;
814 *new_device_added
= true;
816 if (disk_super
->label
[0])
818 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n",
819 disk_super
->label
, devid
, found_transid
, path
,
820 current
->comm
, task_pid_nr(current
));
823 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n",
824 disk_super
->fsid
, devid
, found_transid
, path
,
825 current
->comm
, task_pid_nr(current
));
827 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
829 * When FS is already mounted.
830 * 1. If you are here and if the device->name is NULL that
831 * means this device was missing at time of FS mount.
832 * 2. If you are here and if the device->name is different
833 * from 'path' that means either
834 * a. The same device disappeared and reappeared with
836 * b. The missing-disk-which-was-replaced, has
839 * We must allow 1 and 2a above. But 2b would be a spurious
842 * Further in case of 1 and 2a above, the disk at 'path'
843 * would have missed some transaction when it was away and
844 * in case of 2a the stale bdev has to be updated as well.
845 * 2b must not be allowed at all time.
849 * For now, we do allow update to btrfs_fs_device through the
850 * btrfs dev scan cli after FS has been mounted. We're still
851 * tracking a problem where systems fail mount by subvolume id
852 * when we reject replacement on a mounted FS.
854 if (!fs_devices
->opened
&& found_transid
< device
->generation
) {
856 * That is if the FS is _not_ mounted and if you
857 * are here, that means there is more than one
858 * disk with same uuid and devid.We keep the one
859 * with larger generation number or the last-in if
860 * generation are equal.
862 mutex_unlock(&fs_devices
->device_list_mutex
);
864 "device %s already registered with a higher generation, found %llu expect %llu",
865 path
, found_transid
, device
->generation
);
866 return ERR_PTR(-EEXIST
);
870 * We are going to replace the device path for a given devid,
871 * make sure it's the same device if the device is mounted
873 * NOTE: the device->fs_info may not be reliable here so pass
874 * in a NULL to message helpers instead. This avoids a possible
875 * use-after-free when the fs_info and fs_info->sb are already
879 if (device
->devt
!= path_devt
) {
880 mutex_unlock(&fs_devices
->device_list_mutex
);
881 btrfs_warn_in_rcu(NULL
,
882 "duplicate device %s devid %llu generation %llu scanned by %s (%d)",
883 path
, devid
, found_transid
,
885 task_pid_nr(current
));
886 return ERR_PTR(-EEXIST
);
888 btrfs_info_in_rcu(NULL
,
889 "devid %llu device path %s changed to %s scanned by %s (%d)",
890 devid
, btrfs_dev_name(device
),
892 task_pid_nr(current
));
895 name
= rcu_string_strdup(path
, GFP_NOFS
);
897 mutex_unlock(&fs_devices
->device_list_mutex
);
898 return ERR_PTR(-ENOMEM
);
900 rcu_string_free(device
->name
);
901 rcu_assign_pointer(device
->name
, name
);
902 if (test_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
)) {
903 fs_devices
->missing_devices
--;
904 clear_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
);
906 device
->devt
= path_devt
;
910 * Unmount does not free the btrfs_device struct but would zero
911 * generation along with most of the other members. So just update
912 * it back. We need it to pick the disk with largest generation
915 if (!fs_devices
->opened
) {
916 device
->generation
= found_transid
;
917 fs_devices
->latest_generation
= max_t(u64
, found_transid
,
918 fs_devices
->latest_generation
);
921 fs_devices
->total_devices
= btrfs_super_num_devices(disk_super
);
923 mutex_unlock(&fs_devices
->device_list_mutex
);
927 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
929 struct btrfs_fs_devices
*fs_devices
;
930 struct btrfs_device
*device
;
931 struct btrfs_device
*orig_dev
;
934 lockdep_assert_held(&uuid_mutex
);
936 fs_devices
= alloc_fs_devices(orig
->fsid
);
937 if (IS_ERR(fs_devices
))
940 fs_devices
->total_devices
= orig
->total_devices
;
942 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
943 const char *dev_path
= NULL
;
946 * This is ok to do without RCU read locked because we hold the
947 * uuid mutex so nothing we touch in here is going to disappear.
950 dev_path
= orig_dev
->name
->str
;
952 device
= btrfs_alloc_device(NULL
, &orig_dev
->devid
,
953 orig_dev
->uuid
, dev_path
);
954 if (IS_ERR(device
)) {
955 ret
= PTR_ERR(device
);
959 if (orig_dev
->zone_info
) {
960 struct btrfs_zoned_device_info
*zone_info
;
962 zone_info
= btrfs_clone_dev_zone_info(orig_dev
);
964 btrfs_free_device(device
);
968 device
->zone_info
= zone_info
;
971 list_add(&device
->dev_list
, &fs_devices
->devices
);
972 device
->fs_devices
= fs_devices
;
973 fs_devices
->num_devices
++;
977 free_fs_devices(fs_devices
);
981 static void __btrfs_free_extra_devids(struct btrfs_fs_devices
*fs_devices
,
982 struct btrfs_device
**latest_dev
)
984 struct btrfs_device
*device
, *next
;
986 /* This is the initialized path, it is safe to release the devices. */
987 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
988 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &device
->dev_state
)) {
989 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT
,
990 &device
->dev_state
) &&
991 !test_bit(BTRFS_DEV_STATE_MISSING
,
992 &device
->dev_state
) &&
994 device
->generation
> (*latest_dev
)->generation
)) {
995 *latest_dev
= device
;
1001 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID,
1002 * in btrfs_init_dev_replace() so just continue.
1004 if (device
->devid
== BTRFS_DEV_REPLACE_DEVID
)
1007 if (device
->bdev_handle
) {
1008 bdev_release(device
->bdev_handle
);
1009 device
->bdev
= NULL
;
1010 device
->bdev_handle
= NULL
;
1011 fs_devices
->open_devices
--;
1013 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
)) {
1014 list_del_init(&device
->dev_alloc_list
);
1015 clear_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
);
1016 fs_devices
->rw_devices
--;
1018 list_del_init(&device
->dev_list
);
1019 fs_devices
->num_devices
--;
1020 btrfs_free_device(device
);
1026 * After we have read the system tree and know devids belonging to this
1027 * filesystem, remove the device which does not belong there.
1029 void btrfs_free_extra_devids(struct btrfs_fs_devices
*fs_devices
)
1031 struct btrfs_device
*latest_dev
= NULL
;
1032 struct btrfs_fs_devices
*seed_dev
;
1034 mutex_lock(&uuid_mutex
);
1035 __btrfs_free_extra_devids(fs_devices
, &latest_dev
);
1037 list_for_each_entry(seed_dev
, &fs_devices
->seed_list
, seed_list
)
1038 __btrfs_free_extra_devids(seed_dev
, &latest_dev
);
1040 fs_devices
->latest_dev
= latest_dev
;
1042 mutex_unlock(&uuid_mutex
);
1045 static void btrfs_close_bdev(struct btrfs_device
*device
)
1050 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
)) {
1051 sync_blockdev(device
->bdev
);
1052 invalidate_bdev(device
->bdev
);
1055 bdev_release(device
->bdev_handle
);
1058 static void btrfs_close_one_device(struct btrfs_device
*device
)
1060 struct btrfs_fs_devices
*fs_devices
= device
->fs_devices
;
1062 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
) &&
1063 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
1064 list_del_init(&device
->dev_alloc_list
);
1065 fs_devices
->rw_devices
--;
1068 if (device
->devid
== BTRFS_DEV_REPLACE_DEVID
)
1069 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
);
1071 if (test_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
)) {
1072 clear_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
);
1073 fs_devices
->missing_devices
--;
1076 btrfs_close_bdev(device
);
1078 fs_devices
->open_devices
--;
1079 device
->bdev
= NULL
;
1081 clear_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
);
1082 btrfs_destroy_dev_zone_info(device
);
1084 device
->fs_info
= NULL
;
1085 atomic_set(&device
->dev_stats_ccnt
, 0);
1086 extent_io_tree_release(&device
->alloc_state
);
1089 * Reset the flush error record. We might have a transient flush error
1090 * in this mount, and if so we aborted the current transaction and set
1091 * the fs to an error state, guaranteeing no super blocks can be further
1092 * committed. However that error might be transient and if we unmount the
1093 * filesystem and mount it again, we should allow the mount to succeed
1094 * (btrfs_check_rw_degradable() should not fail) - if after mounting the
1095 * filesystem again we still get flush errors, then we will again abort
1096 * any transaction and set the error state, guaranteeing no commits of
1097 * unsafe super blocks.
1099 device
->last_flush_error
= 0;
1101 /* Verify the device is back in a pristine state */
1102 WARN_ON(test_bit(BTRFS_DEV_STATE_FLUSH_SENT
, &device
->dev_state
));
1103 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
));
1104 WARN_ON(!list_empty(&device
->dev_alloc_list
));
1105 WARN_ON(!list_empty(&device
->post_commit_list
));
1108 static void close_fs_devices(struct btrfs_fs_devices
*fs_devices
)
1110 struct btrfs_device
*device
, *tmp
;
1112 lockdep_assert_held(&uuid_mutex
);
1114 if (--fs_devices
->opened
> 0)
1117 list_for_each_entry_safe(device
, tmp
, &fs_devices
->devices
, dev_list
)
1118 btrfs_close_one_device(device
);
1120 WARN_ON(fs_devices
->open_devices
);
1121 WARN_ON(fs_devices
->rw_devices
);
1122 fs_devices
->opened
= 0;
1123 fs_devices
->seeding
= false;
1124 fs_devices
->fs_info
= NULL
;
1127 void btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
1130 struct btrfs_fs_devices
*tmp
;
1132 mutex_lock(&uuid_mutex
);
1133 close_fs_devices(fs_devices
);
1134 if (!fs_devices
->opened
) {
1135 list_splice_init(&fs_devices
->seed_list
, &list
);
1138 * If the struct btrfs_fs_devices is not assembled with any
1139 * other device, it can be re-initialized during the next mount
1140 * without the needing device-scan step. Therefore, it can be
1143 if (fs_devices
->num_devices
== 1) {
1144 list_del(&fs_devices
->fs_list
);
1145 free_fs_devices(fs_devices
);
1150 list_for_each_entry_safe(fs_devices
, tmp
, &list
, seed_list
) {
1151 close_fs_devices(fs_devices
);
1152 list_del(&fs_devices
->seed_list
);
1153 free_fs_devices(fs_devices
);
1155 mutex_unlock(&uuid_mutex
);
1158 static int open_fs_devices(struct btrfs_fs_devices
*fs_devices
,
1159 blk_mode_t flags
, void *holder
)
1161 struct btrfs_device
*device
;
1162 struct btrfs_device
*latest_dev
= NULL
;
1163 struct btrfs_device
*tmp_device
;
1165 list_for_each_entry_safe(device
, tmp_device
, &fs_devices
->devices
,
1169 ret
= btrfs_open_one_device(fs_devices
, device
, flags
, holder
);
1171 (!latest_dev
|| device
->generation
> latest_dev
->generation
)) {
1172 latest_dev
= device
;
1173 } else if (ret
== -ENODATA
) {
1174 fs_devices
->num_devices
--;
1175 list_del(&device
->dev_list
);
1176 btrfs_free_device(device
);
1179 if (fs_devices
->open_devices
== 0)
1182 fs_devices
->opened
= 1;
1183 fs_devices
->latest_dev
= latest_dev
;
1184 fs_devices
->total_rw_bytes
= 0;
1185 fs_devices
->chunk_alloc_policy
= BTRFS_CHUNK_ALLOC_REGULAR
;
1186 fs_devices
->read_policy
= BTRFS_READ_POLICY_PID
;
1191 static int devid_cmp(void *priv
, const struct list_head
*a
,
1192 const struct list_head
*b
)
1194 const struct btrfs_device
*dev1
, *dev2
;
1196 dev1
= list_entry(a
, struct btrfs_device
, dev_list
);
1197 dev2
= list_entry(b
, struct btrfs_device
, dev_list
);
1199 if (dev1
->devid
< dev2
->devid
)
1201 else if (dev1
->devid
> dev2
->devid
)
1206 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
1207 blk_mode_t flags
, void *holder
)
1211 lockdep_assert_held(&uuid_mutex
);
1213 * The device_list_mutex cannot be taken here in case opening the
1214 * underlying device takes further locks like open_mutex.
1216 * We also don't need the lock here as this is called during mount and
1217 * exclusion is provided by uuid_mutex
1220 if (fs_devices
->opened
) {
1221 fs_devices
->opened
++;
1224 list_sort(NULL
, &fs_devices
->devices
, devid_cmp
);
1225 ret
= open_fs_devices(fs_devices
, flags
, holder
);
1231 void btrfs_release_disk_super(struct btrfs_super_block
*super
)
1233 struct page
*page
= virt_to_page(super
);
1238 static struct btrfs_super_block
*btrfs_read_disk_super(struct block_device
*bdev
,
1239 u64 bytenr
, u64 bytenr_orig
)
1241 struct btrfs_super_block
*disk_super
;
1246 /* make sure our super fits in the device */
1247 if (bytenr
+ PAGE_SIZE
>= bdev_nr_bytes(bdev
))
1248 return ERR_PTR(-EINVAL
);
1250 /* make sure our super fits in the page */
1251 if (sizeof(*disk_super
) > PAGE_SIZE
)
1252 return ERR_PTR(-EINVAL
);
1254 /* make sure our super doesn't straddle pages on disk */
1255 index
= bytenr
>> PAGE_SHIFT
;
1256 if ((bytenr
+ sizeof(*disk_super
) - 1) >> PAGE_SHIFT
!= index
)
1257 return ERR_PTR(-EINVAL
);
1259 /* pull in the page with our super */
1260 page
= read_cache_page_gfp(bdev
->bd_inode
->i_mapping
, index
, GFP_KERNEL
);
1263 return ERR_CAST(page
);
1265 p
= page_address(page
);
1267 /* align our pointer to the offset of the super block */
1268 disk_super
= p
+ offset_in_page(bytenr
);
1270 if (btrfs_super_bytenr(disk_super
) != bytenr_orig
||
1271 btrfs_super_magic(disk_super
) != BTRFS_MAGIC
) {
1272 btrfs_release_disk_super(p
);
1273 return ERR_PTR(-EINVAL
);
1276 if (disk_super
->label
[0] && disk_super
->label
[BTRFS_LABEL_SIZE
- 1])
1277 disk_super
->label
[BTRFS_LABEL_SIZE
- 1] = 0;
1282 int btrfs_forget_devices(dev_t devt
)
1286 mutex_lock(&uuid_mutex
);
1287 ret
= btrfs_free_stale_devices(devt
, NULL
);
1288 mutex_unlock(&uuid_mutex
);
1294 * Look for a btrfs signature on a device. This may be called out of the mount path
1295 * and we are not allowed to call set_blocksize during the scan. The superblock
1296 * is read via pagecache.
1298 * With @mount_arg_dev it's a scan during mount time that will always register
1299 * the device or return an error. Multi-device and seeding devices are registered
1302 struct btrfs_device
*btrfs_scan_one_device(const char *path
, blk_mode_t flags
,
1305 struct btrfs_super_block
*disk_super
;
1306 bool new_device_added
= false;
1307 struct btrfs_device
*device
= NULL
;
1308 struct bdev_handle
*bdev_handle
;
1309 u64 bytenr
, bytenr_orig
;
1312 lockdep_assert_held(&uuid_mutex
);
1315 * we would like to check all the supers, but that would make
1316 * a btrfs mount succeed after a mkfs from a different FS.
1317 * So, we need to add a special mount option to scan for
1318 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1322 * Avoid an exclusive open here, as the systemd-udev may initiate the
1323 * device scan which may race with the user's mount or mkfs command,
1324 * resulting in failure.
1325 * Since the device scan is solely for reading purposes, there is no
1326 * need for an exclusive open. Additionally, the devices are read again
1327 * during the mount process. It is ok to get some inconsistent
1328 * values temporarily, as the device paths of the fsid are the only
1329 * required information for assembling the volume.
1331 bdev_handle
= bdev_open_by_path(path
, flags
, NULL
, NULL
);
1332 if (IS_ERR(bdev_handle
))
1333 return ERR_CAST(bdev_handle
);
1335 bytenr_orig
= btrfs_sb_offset(0);
1336 ret
= btrfs_sb_log_location_bdev(bdev_handle
->bdev
, 0, READ
, &bytenr
);
1338 device
= ERR_PTR(ret
);
1339 goto error_bdev_put
;
1342 disk_super
= btrfs_read_disk_super(bdev_handle
->bdev
, bytenr
,
1344 if (IS_ERR(disk_super
)) {
1345 device
= ERR_CAST(disk_super
);
1346 goto error_bdev_put
;
1349 if (!mount_arg_dev
&& btrfs_super_num_devices(disk_super
) == 1 &&
1350 !(btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
)) {
1353 ret
= lookup_bdev(path
, &devt
);
1355 btrfs_warn(NULL
, "lookup bdev failed for path %s: %d",
1358 btrfs_free_stale_devices(devt
, NULL
);
1360 pr_debug("BTRFS: skip registering single non-seed device %s\n", path
);
1362 goto free_disk_super
;
1365 device
= device_list_add(path
, disk_super
, &new_device_added
);
1366 if (!IS_ERR(device
) && new_device_added
)
1367 btrfs_free_stale_devices(device
->devt
, device
);
1370 btrfs_release_disk_super(disk_super
);
1373 bdev_release(bdev_handle
);
1379 * Try to find a chunk that intersects [start, start + len] range and when one
1380 * such is found, record the end of it in *start
1382 static bool contains_pending_extent(struct btrfs_device
*device
, u64
*start
,
1385 u64 physical_start
, physical_end
;
1387 lockdep_assert_held(&device
->fs_info
->chunk_mutex
);
1389 if (find_first_extent_bit(&device
->alloc_state
, *start
,
1390 &physical_start
, &physical_end
,
1391 CHUNK_ALLOCATED
, NULL
)) {
1393 if (in_range(physical_start
, *start
, len
) ||
1394 in_range(*start
, physical_start
,
1395 physical_end
- physical_start
)) {
1396 *start
= physical_end
+ 1;
1403 static u64
dev_extent_search_start(struct btrfs_device
*device
)
1405 switch (device
->fs_devices
->chunk_alloc_policy
) {
1406 case BTRFS_CHUNK_ALLOC_REGULAR
:
1407 return BTRFS_DEVICE_RANGE_RESERVED
;
1408 case BTRFS_CHUNK_ALLOC_ZONED
:
1410 * We don't care about the starting region like regular
1411 * allocator, because we anyway use/reserve the first two zones
1412 * for superblock logging.
1420 static bool dev_extent_hole_check_zoned(struct btrfs_device
*device
,
1421 u64
*hole_start
, u64
*hole_size
,
1424 u64 zone_size
= device
->zone_info
->zone_size
;
1427 bool changed
= false;
1429 ASSERT(IS_ALIGNED(*hole_start
, zone_size
));
1431 while (*hole_size
> 0) {
1432 pos
= btrfs_find_allocatable_zones(device
, *hole_start
,
1433 *hole_start
+ *hole_size
,
1435 if (pos
!= *hole_start
) {
1436 *hole_size
= *hole_start
+ *hole_size
- pos
;
1439 if (*hole_size
< num_bytes
)
1443 ret
= btrfs_ensure_empty_zones(device
, pos
, num_bytes
);
1445 /* Range is ensured to be empty */
1449 /* Given hole range was invalid (outside of device) */
1450 if (ret
== -ERANGE
) {
1451 *hole_start
+= *hole_size
;
1456 *hole_start
+= zone_size
;
1457 *hole_size
-= zone_size
;
1465 * Check if specified hole is suitable for allocation.
1467 * @device: the device which we have the hole
1468 * @hole_start: starting position of the hole
1469 * @hole_size: the size of the hole
1470 * @num_bytes: the size of the free space that we need
1472 * This function may modify @hole_start and @hole_size to reflect the suitable
1473 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1475 static bool dev_extent_hole_check(struct btrfs_device
*device
, u64
*hole_start
,
1476 u64
*hole_size
, u64 num_bytes
)
1478 bool changed
= false;
1479 u64 hole_end
= *hole_start
+ *hole_size
;
1483 * Check before we set max_hole_start, otherwise we could end up
1484 * sending back this offset anyway.
1486 if (contains_pending_extent(device
, hole_start
, *hole_size
)) {
1487 if (hole_end
>= *hole_start
)
1488 *hole_size
= hole_end
- *hole_start
;
1494 switch (device
->fs_devices
->chunk_alloc_policy
) {
1495 case BTRFS_CHUNK_ALLOC_REGULAR
:
1496 /* No extra check */
1498 case BTRFS_CHUNK_ALLOC_ZONED
:
1499 if (dev_extent_hole_check_zoned(device
, hole_start
,
1500 hole_size
, num_bytes
)) {
1503 * The changed hole can contain pending extent.
1504 * Loop again to check that.
1520 * Find free space in the specified device.
1522 * @device: the device which we search the free space in
1523 * @num_bytes: the size of the free space that we need
1524 * @search_start: the position from which to begin the search
1525 * @start: store the start of the free space.
1526 * @len: the size of the free space. that we find, or the size
1527 * of the max free space if we don't find suitable free space
1529 * This does a pretty simple search, the expectation is that it is called very
1530 * infrequently and that a given device has a small number of extents.
1532 * @start is used to store the start of the free space if we find. But if we
1533 * don't find suitable free space, it will be used to store the start position
1534 * of the max free space.
1536 * @len is used to store the size of the free space that we find.
1537 * But if we don't find suitable free space, it is used to store the size of
1538 * the max free space.
1540 * NOTE: This function will search *commit* root of device tree, and does extra
1541 * check to ensure dev extents are not double allocated.
1542 * This makes the function safe to allocate dev extents but may not report
1543 * correct usable device space, as device extent freed in current transaction
1544 * is not reported as available.
1546 static int find_free_dev_extent(struct btrfs_device
*device
, u64 num_bytes
,
1547 u64
*start
, u64
*len
)
1549 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1550 struct btrfs_root
*root
= fs_info
->dev_root
;
1551 struct btrfs_key key
;
1552 struct btrfs_dev_extent
*dev_extent
;
1553 struct btrfs_path
*path
;
1557 u64 max_hole_size
= 0;
1559 u64 search_end
= device
->total_bytes
;
1562 struct extent_buffer
*l
;
1564 search_start
= dev_extent_search_start(device
);
1565 max_hole_start
= search_start
;
1567 WARN_ON(device
->zone_info
&&
1568 !IS_ALIGNED(num_bytes
, device
->zone_info
->zone_size
));
1570 path
= btrfs_alloc_path();
1576 if (search_start
>= search_end
||
1577 test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
)) {
1582 path
->reada
= READA_FORWARD
;
1583 path
->search_commit_root
= 1;
1584 path
->skip_locking
= 1;
1586 key
.objectid
= device
->devid
;
1587 key
.offset
= search_start
;
1588 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1590 ret
= btrfs_search_backwards(root
, &key
, path
);
1594 while (search_start
< search_end
) {
1596 slot
= path
->slots
[0];
1597 if (slot
>= btrfs_header_nritems(l
)) {
1598 ret
= btrfs_next_leaf(root
, path
);
1606 btrfs_item_key_to_cpu(l
, &key
, slot
);
1608 if (key
.objectid
< device
->devid
)
1611 if (key
.objectid
> device
->devid
)
1614 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1617 if (key
.offset
> search_end
)
1620 if (key
.offset
> search_start
) {
1621 hole_size
= key
.offset
- search_start
;
1622 dev_extent_hole_check(device
, &search_start
, &hole_size
,
1625 if (hole_size
> max_hole_size
) {
1626 max_hole_start
= search_start
;
1627 max_hole_size
= hole_size
;
1631 * If this free space is greater than which we need,
1632 * it must be the max free space that we have found
1633 * until now, so max_hole_start must point to the start
1634 * of this free space and the length of this free space
1635 * is stored in max_hole_size. Thus, we return
1636 * max_hole_start and max_hole_size and go back to the
1639 if (hole_size
>= num_bytes
) {
1645 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1646 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1648 if (extent_end
> search_start
)
1649 search_start
= extent_end
;
1656 * At this point, search_start should be the end of
1657 * allocated dev extents, and when shrinking the device,
1658 * search_end may be smaller than search_start.
1660 if (search_end
> search_start
) {
1661 hole_size
= search_end
- search_start
;
1662 if (dev_extent_hole_check(device
, &search_start
, &hole_size
,
1664 btrfs_release_path(path
);
1668 if (hole_size
> max_hole_size
) {
1669 max_hole_start
= search_start
;
1670 max_hole_size
= hole_size
;
1675 if (max_hole_size
< num_bytes
)
1680 ASSERT(max_hole_start
+ max_hole_size
<= search_end
);
1682 btrfs_free_path(path
);
1683 *start
= max_hole_start
;
1685 *len
= max_hole_size
;
1689 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1690 struct btrfs_device
*device
,
1691 u64 start
, u64
*dev_extent_len
)
1693 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1694 struct btrfs_root
*root
= fs_info
->dev_root
;
1696 struct btrfs_path
*path
;
1697 struct btrfs_key key
;
1698 struct btrfs_key found_key
;
1699 struct extent_buffer
*leaf
= NULL
;
1700 struct btrfs_dev_extent
*extent
= NULL
;
1702 path
= btrfs_alloc_path();
1706 key
.objectid
= device
->devid
;
1708 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1710 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1712 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1713 BTRFS_DEV_EXTENT_KEY
);
1716 leaf
= path
->nodes
[0];
1717 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1718 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1719 struct btrfs_dev_extent
);
1720 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1721 btrfs_dev_extent_length(leaf
, extent
) < start
);
1723 btrfs_release_path(path
);
1725 } else if (ret
== 0) {
1726 leaf
= path
->nodes
[0];
1727 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1728 struct btrfs_dev_extent
);
1733 *dev_extent_len
= btrfs_dev_extent_length(leaf
, extent
);
1735 ret
= btrfs_del_item(trans
, root
, path
);
1737 set_bit(BTRFS_TRANS_HAVE_FREE_BGS
, &trans
->transaction
->flags
);
1739 btrfs_free_path(path
);
1743 static u64
find_next_chunk(struct btrfs_fs_info
*fs_info
)
1745 struct extent_map_tree
*em_tree
;
1746 struct extent_map
*em
;
1750 em_tree
= &fs_info
->mapping_tree
;
1751 read_lock(&em_tree
->lock
);
1752 n
= rb_last(&em_tree
->map
.rb_root
);
1754 em
= rb_entry(n
, struct extent_map
, rb_node
);
1755 ret
= em
->start
+ em
->len
;
1757 read_unlock(&em_tree
->lock
);
1762 static noinline
int find_next_devid(struct btrfs_fs_info
*fs_info
,
1766 struct btrfs_key key
;
1767 struct btrfs_key found_key
;
1768 struct btrfs_path
*path
;
1770 path
= btrfs_alloc_path();
1774 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1775 key
.type
= BTRFS_DEV_ITEM_KEY
;
1776 key
.offset
= (u64
)-1;
1778 ret
= btrfs_search_slot(NULL
, fs_info
->chunk_root
, &key
, path
, 0, 0);
1784 btrfs_err(fs_info
, "corrupted chunk tree devid -1 matched");
1789 ret
= btrfs_previous_item(fs_info
->chunk_root
, path
,
1790 BTRFS_DEV_ITEMS_OBJECTID
,
1791 BTRFS_DEV_ITEM_KEY
);
1795 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1797 *devid_ret
= found_key
.offset
+ 1;
1801 btrfs_free_path(path
);
1806 * the device information is stored in the chunk root
1807 * the btrfs_device struct should be fully filled in
1809 static int btrfs_add_dev_item(struct btrfs_trans_handle
*trans
,
1810 struct btrfs_device
*device
)
1813 struct btrfs_path
*path
;
1814 struct btrfs_dev_item
*dev_item
;
1815 struct extent_buffer
*leaf
;
1816 struct btrfs_key key
;
1819 path
= btrfs_alloc_path();
1823 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1824 key
.type
= BTRFS_DEV_ITEM_KEY
;
1825 key
.offset
= device
->devid
;
1827 btrfs_reserve_chunk_metadata(trans
, true);
1828 ret
= btrfs_insert_empty_item(trans
, trans
->fs_info
->chunk_root
, path
,
1829 &key
, sizeof(*dev_item
));
1830 btrfs_trans_release_chunk_metadata(trans
);
1834 leaf
= path
->nodes
[0];
1835 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1837 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1838 btrfs_set_device_generation(leaf
, dev_item
, 0);
1839 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1840 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1841 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1842 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1843 btrfs_set_device_total_bytes(leaf
, dev_item
,
1844 btrfs_device_get_disk_total_bytes(device
));
1845 btrfs_set_device_bytes_used(leaf
, dev_item
,
1846 btrfs_device_get_bytes_used(device
));
1847 btrfs_set_device_group(leaf
, dev_item
, 0);
1848 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1849 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1850 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1852 ptr
= btrfs_device_uuid(dev_item
);
1853 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1854 ptr
= btrfs_device_fsid(dev_item
);
1855 write_extent_buffer(leaf
, trans
->fs_info
->fs_devices
->metadata_uuid
,
1856 ptr
, BTRFS_FSID_SIZE
);
1857 btrfs_mark_buffer_dirty(trans
, leaf
);
1861 btrfs_free_path(path
);
1866 * Function to update ctime/mtime for a given device path.
1867 * Mainly used for ctime/mtime based probe like libblkid.
1869 * We don't care about errors here, this is just to be kind to userspace.
1871 static void update_dev_time(const char *device_path
)
1876 ret
= kern_path(device_path
, LOOKUP_FOLLOW
, &path
);
1880 inode_update_time(d_inode(path
.dentry
), S_MTIME
| S_CTIME
| S_VERSION
);
1884 static int btrfs_rm_dev_item(struct btrfs_trans_handle
*trans
,
1885 struct btrfs_device
*device
)
1887 struct btrfs_root
*root
= device
->fs_info
->chunk_root
;
1889 struct btrfs_path
*path
;
1890 struct btrfs_key key
;
1892 path
= btrfs_alloc_path();
1896 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1897 key
.type
= BTRFS_DEV_ITEM_KEY
;
1898 key
.offset
= device
->devid
;
1900 btrfs_reserve_chunk_metadata(trans
, false);
1901 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1902 btrfs_trans_release_chunk_metadata(trans
);
1909 ret
= btrfs_del_item(trans
, root
, path
);
1911 btrfs_free_path(path
);
1916 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1917 * filesystem. It's up to the caller to adjust that number regarding eg. device
1920 static int btrfs_check_raid_min_devices(struct btrfs_fs_info
*fs_info
,
1928 seq
= read_seqbegin(&fs_info
->profiles_lock
);
1930 all_avail
= fs_info
->avail_data_alloc_bits
|
1931 fs_info
->avail_system_alloc_bits
|
1932 fs_info
->avail_metadata_alloc_bits
;
1933 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
1935 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
1936 if (!(all_avail
& btrfs_raid_array
[i
].bg_flag
))
1939 if (num_devices
< btrfs_raid_array
[i
].devs_min
)
1940 return btrfs_raid_array
[i
].mindev_error
;
1946 static struct btrfs_device
* btrfs_find_next_active_device(
1947 struct btrfs_fs_devices
*fs_devs
, struct btrfs_device
*device
)
1949 struct btrfs_device
*next_device
;
1951 list_for_each_entry(next_device
, &fs_devs
->devices
, dev_list
) {
1952 if (next_device
!= device
&&
1953 !test_bit(BTRFS_DEV_STATE_MISSING
, &next_device
->dev_state
)
1954 && next_device
->bdev
)
1962 * Helper function to check if the given device is part of s_bdev / latest_dev
1963 * and replace it with the provided or the next active device, in the context
1964 * where this function called, there should be always be another device (or
1965 * this_dev) which is active.
1967 void __cold
btrfs_assign_next_active_device(struct btrfs_device
*device
,
1968 struct btrfs_device
*next_device
)
1970 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1973 next_device
= btrfs_find_next_active_device(fs_info
->fs_devices
,
1975 ASSERT(next_device
);
1977 if (fs_info
->sb
->s_bdev
&&
1978 (fs_info
->sb
->s_bdev
== device
->bdev
))
1979 fs_info
->sb
->s_bdev
= next_device
->bdev
;
1981 if (fs_info
->fs_devices
->latest_dev
->bdev
== device
->bdev
)
1982 fs_info
->fs_devices
->latest_dev
= next_device
;
1986 * Return btrfs_fs_devices::num_devices excluding the device that's being
1987 * currently replaced.
1989 static u64
btrfs_num_devices(struct btrfs_fs_info
*fs_info
)
1991 u64 num_devices
= fs_info
->fs_devices
->num_devices
;
1993 down_read(&fs_info
->dev_replace
.rwsem
);
1994 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
1995 ASSERT(num_devices
> 1);
1998 up_read(&fs_info
->dev_replace
.rwsem
);
2003 static void btrfs_scratch_superblock(struct btrfs_fs_info
*fs_info
,
2004 struct block_device
*bdev
, int copy_num
)
2006 struct btrfs_super_block
*disk_super
;
2007 const size_t len
= sizeof(disk_super
->magic
);
2008 const u64 bytenr
= btrfs_sb_offset(copy_num
);
2011 disk_super
= btrfs_read_disk_super(bdev
, bytenr
, bytenr
);
2012 if (IS_ERR(disk_super
))
2015 memset(&disk_super
->magic
, 0, len
);
2016 folio_mark_dirty(virt_to_folio(disk_super
));
2017 btrfs_release_disk_super(disk_super
);
2019 ret
= sync_blockdev_range(bdev
, bytenr
, bytenr
+ len
- 1);
2021 btrfs_warn(fs_info
, "error clearing superblock number %d (%d)",
2025 void btrfs_scratch_superblocks(struct btrfs_fs_info
*fs_info
,
2026 struct block_device
*bdev
,
2027 const char *device_path
)
2034 for (copy_num
= 0; copy_num
< BTRFS_SUPER_MIRROR_MAX
; copy_num
++) {
2035 if (bdev_is_zoned(bdev
))
2036 btrfs_reset_sb_log_zones(bdev
, copy_num
);
2038 btrfs_scratch_superblock(fs_info
, bdev
, copy_num
);
2041 /* Notify udev that device has changed */
2042 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
2044 /* Update ctime/mtime for device path for libblkid */
2045 update_dev_time(device_path
);
2048 int btrfs_rm_device(struct btrfs_fs_info
*fs_info
,
2049 struct btrfs_dev_lookup_args
*args
,
2050 struct bdev_handle
**bdev_handle
)
2052 struct btrfs_trans_handle
*trans
;
2053 struct btrfs_device
*device
;
2054 struct btrfs_fs_devices
*cur_devices
;
2055 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2059 if (btrfs_fs_incompat(fs_info
, EXTENT_TREE_V2
)) {
2060 btrfs_err(fs_info
, "device remove not supported on extent tree v2 yet");
2065 * The device list in fs_devices is accessed without locks (neither
2066 * uuid_mutex nor device_list_mutex) as it won't change on a mounted
2067 * filesystem and another device rm cannot run.
2069 num_devices
= btrfs_num_devices(fs_info
);
2071 ret
= btrfs_check_raid_min_devices(fs_info
, num_devices
- 1);
2075 device
= btrfs_find_device(fs_info
->fs_devices
, args
);
2078 ret
= BTRFS_ERROR_DEV_MISSING_NOT_FOUND
;
2084 if (btrfs_pinned_by_swapfile(fs_info
, device
)) {
2085 btrfs_warn_in_rcu(fs_info
,
2086 "cannot remove device %s (devid %llu) due to active swapfile",
2087 btrfs_dev_name(device
), device
->devid
);
2091 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
))
2092 return BTRFS_ERROR_DEV_TGT_REPLACE
;
2094 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
) &&
2095 fs_info
->fs_devices
->rw_devices
== 1)
2096 return BTRFS_ERROR_DEV_ONLY_WRITABLE
;
2098 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
)) {
2099 mutex_lock(&fs_info
->chunk_mutex
);
2100 list_del_init(&device
->dev_alloc_list
);
2101 device
->fs_devices
->rw_devices
--;
2102 mutex_unlock(&fs_info
->chunk_mutex
);
2105 ret
= btrfs_shrink_device(device
, 0);
2109 trans
= btrfs_start_transaction(fs_info
->chunk_root
, 0);
2110 if (IS_ERR(trans
)) {
2111 ret
= PTR_ERR(trans
);
2115 ret
= btrfs_rm_dev_item(trans
, device
);
2117 /* Any error in dev item removal is critical */
2119 "failed to remove device item for devid %llu: %d",
2120 device
->devid
, ret
);
2121 btrfs_abort_transaction(trans
, ret
);
2122 btrfs_end_transaction(trans
);
2126 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &device
->dev_state
);
2127 btrfs_scrub_cancel_dev(device
);
2130 * the device list mutex makes sure that we don't change
2131 * the device list while someone else is writing out all
2132 * the device supers. Whoever is writing all supers, should
2133 * lock the device list mutex before getting the number of
2134 * devices in the super block (super_copy). Conversely,
2135 * whoever updates the number of devices in the super block
2136 * (super_copy) should hold the device list mutex.
2140 * In normal cases the cur_devices == fs_devices. But in case
2141 * of deleting a seed device, the cur_devices should point to
2142 * its own fs_devices listed under the fs_devices->seed_list.
2144 cur_devices
= device
->fs_devices
;
2145 mutex_lock(&fs_devices
->device_list_mutex
);
2146 list_del_rcu(&device
->dev_list
);
2148 cur_devices
->num_devices
--;
2149 cur_devices
->total_devices
--;
2150 /* Update total_devices of the parent fs_devices if it's seed */
2151 if (cur_devices
!= fs_devices
)
2152 fs_devices
->total_devices
--;
2154 if (test_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
))
2155 cur_devices
->missing_devices
--;
2157 btrfs_assign_next_active_device(device
, NULL
);
2159 if (device
->bdev_handle
) {
2160 cur_devices
->open_devices
--;
2161 /* remove sysfs entry */
2162 btrfs_sysfs_remove_device(device
);
2165 num_devices
= btrfs_super_num_devices(fs_info
->super_copy
) - 1;
2166 btrfs_set_super_num_devices(fs_info
->super_copy
, num_devices
);
2167 mutex_unlock(&fs_devices
->device_list_mutex
);
2170 * At this point, the device is zero sized and detached from the
2171 * devices list. All that's left is to zero out the old supers and
2174 * We cannot call btrfs_close_bdev() here because we're holding the sb
2175 * write lock, and bdev_release() will pull in the ->open_mutex on
2176 * the block device and it's dependencies. Instead just flush the
2177 * device and let the caller do the final bdev_release.
2179 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
)) {
2180 btrfs_scratch_superblocks(fs_info
, device
->bdev
,
2183 sync_blockdev(device
->bdev
);
2184 invalidate_bdev(device
->bdev
);
2188 *bdev_handle
= device
->bdev_handle
;
2190 btrfs_free_device(device
);
2193 * This can happen if cur_devices is the private seed devices list. We
2194 * cannot call close_fs_devices() here because it expects the uuid_mutex
2195 * to be held, but in fact we don't need that for the private
2196 * seed_devices, we can simply decrement cur_devices->opened and then
2197 * remove it from our list and free the fs_devices.
2199 if (cur_devices
->num_devices
== 0) {
2200 list_del_init(&cur_devices
->seed_list
);
2201 ASSERT(cur_devices
->opened
== 1);
2202 cur_devices
->opened
--;
2203 free_fs_devices(cur_devices
);
2206 ret
= btrfs_commit_transaction(trans
);
2211 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
)) {
2212 mutex_lock(&fs_info
->chunk_mutex
);
2213 list_add(&device
->dev_alloc_list
,
2214 &fs_devices
->alloc_list
);
2215 device
->fs_devices
->rw_devices
++;
2216 mutex_unlock(&fs_info
->chunk_mutex
);
2221 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device
*srcdev
)
2223 struct btrfs_fs_devices
*fs_devices
;
2225 lockdep_assert_held(&srcdev
->fs_info
->fs_devices
->device_list_mutex
);
2228 * in case of fs with no seed, srcdev->fs_devices will point
2229 * to fs_devices of fs_info. However when the dev being replaced is
2230 * a seed dev it will point to the seed's local fs_devices. In short
2231 * srcdev will have its correct fs_devices in both the cases.
2233 fs_devices
= srcdev
->fs_devices
;
2235 list_del_rcu(&srcdev
->dev_list
);
2236 list_del(&srcdev
->dev_alloc_list
);
2237 fs_devices
->num_devices
--;
2238 if (test_bit(BTRFS_DEV_STATE_MISSING
, &srcdev
->dev_state
))
2239 fs_devices
->missing_devices
--;
2241 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &srcdev
->dev_state
))
2242 fs_devices
->rw_devices
--;
2245 fs_devices
->open_devices
--;
2248 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device
*srcdev
)
2250 struct btrfs_fs_devices
*fs_devices
= srcdev
->fs_devices
;
2252 mutex_lock(&uuid_mutex
);
2254 btrfs_close_bdev(srcdev
);
2256 btrfs_free_device(srcdev
);
2258 /* if this is no devs we rather delete the fs_devices */
2259 if (!fs_devices
->num_devices
) {
2261 * On a mounted FS, num_devices can't be zero unless it's a
2262 * seed. In case of a seed device being replaced, the replace
2263 * target added to the sprout FS, so there will be no more
2264 * device left under the seed FS.
2266 ASSERT(fs_devices
->seeding
);
2268 list_del_init(&fs_devices
->seed_list
);
2269 close_fs_devices(fs_devices
);
2270 free_fs_devices(fs_devices
);
2272 mutex_unlock(&uuid_mutex
);
2275 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device
*tgtdev
)
2277 struct btrfs_fs_devices
*fs_devices
= tgtdev
->fs_info
->fs_devices
;
2279 mutex_lock(&fs_devices
->device_list_mutex
);
2281 btrfs_sysfs_remove_device(tgtdev
);
2284 fs_devices
->open_devices
--;
2286 fs_devices
->num_devices
--;
2288 btrfs_assign_next_active_device(tgtdev
, NULL
);
2290 list_del_rcu(&tgtdev
->dev_list
);
2292 mutex_unlock(&fs_devices
->device_list_mutex
);
2294 btrfs_scratch_superblocks(tgtdev
->fs_info
, tgtdev
->bdev
,
2297 btrfs_close_bdev(tgtdev
);
2299 btrfs_free_device(tgtdev
);
2303 * Populate args from device at path.
2305 * @fs_info: the filesystem
2306 * @args: the args to populate
2307 * @path: the path to the device
2309 * This will read the super block of the device at @path and populate @args with
2310 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to
2311 * lookup a device to operate on, but need to do it before we take any locks.
2312 * This properly handles the special case of "missing" that a user may pass in,
2313 * and does some basic sanity checks. The caller must make sure that @path is
2314 * properly NUL terminated before calling in, and must call
2315 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and
2318 * Return: 0 for success, -errno for failure
2320 int btrfs_get_dev_args_from_path(struct btrfs_fs_info
*fs_info
,
2321 struct btrfs_dev_lookup_args
*args
,
2324 struct btrfs_super_block
*disk_super
;
2325 struct bdev_handle
*bdev_handle
;
2328 if (!path
|| !path
[0])
2330 if (!strcmp(path
, "missing")) {
2331 args
->missing
= true;
2335 args
->uuid
= kzalloc(BTRFS_UUID_SIZE
, GFP_KERNEL
);
2336 args
->fsid
= kzalloc(BTRFS_FSID_SIZE
, GFP_KERNEL
);
2337 if (!args
->uuid
|| !args
->fsid
) {
2338 btrfs_put_dev_args_from_path(args
);
2342 ret
= btrfs_get_bdev_and_sb(path
, BLK_OPEN_READ
, NULL
, 0,
2343 &bdev_handle
, &disk_super
);
2345 btrfs_put_dev_args_from_path(args
);
2349 args
->devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
2350 memcpy(args
->uuid
, disk_super
->dev_item
.uuid
, BTRFS_UUID_SIZE
);
2351 if (btrfs_fs_incompat(fs_info
, METADATA_UUID
))
2352 memcpy(args
->fsid
, disk_super
->metadata_uuid
, BTRFS_FSID_SIZE
);
2354 memcpy(args
->fsid
, disk_super
->fsid
, BTRFS_FSID_SIZE
);
2355 btrfs_release_disk_super(disk_super
);
2356 bdev_release(bdev_handle
);
2361 * Only use this jointly with btrfs_get_dev_args_from_path() because we will
2362 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables
2363 * that don't need to be freed.
2365 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args
*args
)
2373 struct btrfs_device
*btrfs_find_device_by_devspec(
2374 struct btrfs_fs_info
*fs_info
, u64 devid
,
2375 const char *device_path
)
2377 BTRFS_DEV_LOOKUP_ARGS(args
);
2378 struct btrfs_device
*device
;
2383 device
= btrfs_find_device(fs_info
->fs_devices
, &args
);
2385 return ERR_PTR(-ENOENT
);
2389 ret
= btrfs_get_dev_args_from_path(fs_info
, &args
, device_path
);
2391 return ERR_PTR(ret
);
2392 device
= btrfs_find_device(fs_info
->fs_devices
, &args
);
2393 btrfs_put_dev_args_from_path(&args
);
2395 return ERR_PTR(-ENOENT
);
2399 static struct btrfs_fs_devices
*btrfs_init_sprout(struct btrfs_fs_info
*fs_info
)
2401 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2402 struct btrfs_fs_devices
*old_devices
;
2403 struct btrfs_fs_devices
*seed_devices
;
2405 lockdep_assert_held(&uuid_mutex
);
2406 if (!fs_devices
->seeding
)
2407 return ERR_PTR(-EINVAL
);
2410 * Private copy of the seed devices, anchored at
2411 * fs_info->fs_devices->seed_list
2413 seed_devices
= alloc_fs_devices(NULL
);
2414 if (IS_ERR(seed_devices
))
2415 return seed_devices
;
2418 * It's necessary to retain a copy of the original seed fs_devices in
2419 * fs_uuids so that filesystems which have been seeded can successfully
2420 * reference the seed device from open_seed_devices. This also supports
2423 old_devices
= clone_fs_devices(fs_devices
);
2424 if (IS_ERR(old_devices
)) {
2425 kfree(seed_devices
);
2429 list_add(&old_devices
->fs_list
, &fs_uuids
);
2431 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
2432 seed_devices
->opened
= 1;
2433 INIT_LIST_HEAD(&seed_devices
->devices
);
2434 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
2435 mutex_init(&seed_devices
->device_list_mutex
);
2437 return seed_devices
;
2441 * Splice seed devices into the sprout fs_devices.
2442 * Generate a new fsid for the sprouted read-write filesystem.
2444 static void btrfs_setup_sprout(struct btrfs_fs_info
*fs_info
,
2445 struct btrfs_fs_devices
*seed_devices
)
2447 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2448 struct btrfs_super_block
*disk_super
= fs_info
->super_copy
;
2449 struct btrfs_device
*device
;
2453 * We are updating the fsid, the thread leading to device_list_add()
2454 * could race, so uuid_mutex is needed.
2456 lockdep_assert_held(&uuid_mutex
);
2459 * The threads listed below may traverse dev_list but can do that without
2460 * device_list_mutex:
2461 * - All device ops and balance - as we are in btrfs_exclop_start.
2462 * - Various dev_list readers - are using RCU.
2463 * - btrfs_ioctl_fitrim() - is using RCU.
2465 * For-read threads as below are using device_list_mutex:
2466 * - Readonly scrub btrfs_scrub_dev()
2467 * - Readonly scrub btrfs_scrub_progress()
2468 * - btrfs_get_dev_stats()
2470 lockdep_assert_held(&fs_devices
->device_list_mutex
);
2472 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
2474 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
)
2475 device
->fs_devices
= seed_devices
;
2477 fs_devices
->seeding
= false;
2478 fs_devices
->num_devices
= 0;
2479 fs_devices
->open_devices
= 0;
2480 fs_devices
->missing_devices
= 0;
2481 fs_devices
->rotating
= false;
2482 list_add(&seed_devices
->seed_list
, &fs_devices
->seed_list
);
2484 generate_random_uuid(fs_devices
->fsid
);
2485 memcpy(fs_devices
->metadata_uuid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2486 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2488 super_flags
= btrfs_super_flags(disk_super
) &
2489 ~BTRFS_SUPER_FLAG_SEEDING
;
2490 btrfs_set_super_flags(disk_super
, super_flags
);
2494 * Store the expected generation for seed devices in device items.
2496 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
)
2498 BTRFS_DEV_LOOKUP_ARGS(args
);
2499 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2500 struct btrfs_root
*root
= fs_info
->chunk_root
;
2501 struct btrfs_path
*path
;
2502 struct extent_buffer
*leaf
;
2503 struct btrfs_dev_item
*dev_item
;
2504 struct btrfs_device
*device
;
2505 struct btrfs_key key
;
2506 u8 fs_uuid
[BTRFS_FSID_SIZE
];
2507 u8 dev_uuid
[BTRFS_UUID_SIZE
];
2510 path
= btrfs_alloc_path();
2514 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2516 key
.type
= BTRFS_DEV_ITEM_KEY
;
2519 btrfs_reserve_chunk_metadata(trans
, false);
2520 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2521 btrfs_trans_release_chunk_metadata(trans
);
2525 leaf
= path
->nodes
[0];
2527 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
2528 ret
= btrfs_next_leaf(root
, path
);
2533 leaf
= path
->nodes
[0];
2534 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2535 btrfs_release_path(path
);
2539 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2540 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
2541 key
.type
!= BTRFS_DEV_ITEM_KEY
)
2544 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2545 struct btrfs_dev_item
);
2546 args
.devid
= btrfs_device_id(leaf
, dev_item
);
2547 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
2549 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
2551 args
.uuid
= dev_uuid
;
2552 args
.fsid
= fs_uuid
;
2553 device
= btrfs_find_device(fs_info
->fs_devices
, &args
);
2554 BUG_ON(!device
); /* Logic error */
2556 if (device
->fs_devices
->seeding
) {
2557 btrfs_set_device_generation(leaf
, dev_item
,
2558 device
->generation
);
2559 btrfs_mark_buffer_dirty(trans
, leaf
);
2567 btrfs_free_path(path
);
2571 int btrfs_init_new_device(struct btrfs_fs_info
*fs_info
, const char *device_path
)
2573 struct btrfs_root
*root
= fs_info
->dev_root
;
2574 struct btrfs_trans_handle
*trans
;
2575 struct btrfs_device
*device
;
2576 struct bdev_handle
*bdev_handle
;
2577 struct super_block
*sb
= fs_info
->sb
;
2578 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2579 struct btrfs_fs_devices
*seed_devices
= NULL
;
2580 u64 orig_super_total_bytes
;
2581 u64 orig_super_num_devices
;
2583 bool seeding_dev
= false;
2584 bool locked
= false;
2586 if (sb_rdonly(sb
) && !fs_devices
->seeding
)
2589 bdev_handle
= bdev_open_by_path(device_path
, BLK_OPEN_WRITE
,
2590 fs_info
->bdev_holder
, NULL
);
2591 if (IS_ERR(bdev_handle
))
2592 return PTR_ERR(bdev_handle
);
2594 if (!btrfs_check_device_zone_type(fs_info
, bdev_handle
->bdev
)) {
2599 if (fs_devices
->seeding
) {
2601 down_write(&sb
->s_umount
);
2602 mutex_lock(&uuid_mutex
);
2606 sync_blockdev(bdev_handle
->bdev
);
2609 list_for_each_entry_rcu(device
, &fs_devices
->devices
, dev_list
) {
2610 if (device
->bdev
== bdev_handle
->bdev
) {
2618 device
= btrfs_alloc_device(fs_info
, NULL
, NULL
, device_path
);
2619 if (IS_ERR(device
)) {
2620 /* we can safely leave the fs_devices entry around */
2621 ret
= PTR_ERR(device
);
2625 device
->fs_info
= fs_info
;
2626 device
->bdev_handle
= bdev_handle
;
2627 device
->bdev
= bdev_handle
->bdev
;
2628 ret
= lookup_bdev(device_path
, &device
->devt
);
2630 goto error_free_device
;
2632 ret
= btrfs_get_dev_zone_info(device
, false);
2634 goto error_free_device
;
2636 trans
= btrfs_start_transaction(root
, 0);
2637 if (IS_ERR(trans
)) {
2638 ret
= PTR_ERR(trans
);
2639 goto error_free_zone
;
2642 set_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
);
2643 device
->generation
= trans
->transid
;
2644 device
->io_width
= fs_info
->sectorsize
;
2645 device
->io_align
= fs_info
->sectorsize
;
2646 device
->sector_size
= fs_info
->sectorsize
;
2647 device
->total_bytes
=
2648 round_down(bdev_nr_bytes(device
->bdev
), fs_info
->sectorsize
);
2649 device
->disk_total_bytes
= device
->total_bytes
;
2650 device
->commit_total_bytes
= device
->total_bytes
;
2651 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &device
->dev_state
);
2652 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
);
2653 device
->dev_stats_valid
= 1;
2654 set_blocksize(device
->bdev
, BTRFS_BDEV_BLOCKSIZE
);
2657 btrfs_clear_sb_rdonly(sb
);
2659 /* GFP_KERNEL allocation must not be under device_list_mutex */
2660 seed_devices
= btrfs_init_sprout(fs_info
);
2661 if (IS_ERR(seed_devices
)) {
2662 ret
= PTR_ERR(seed_devices
);
2663 btrfs_abort_transaction(trans
, ret
);
2668 mutex_lock(&fs_devices
->device_list_mutex
);
2670 btrfs_setup_sprout(fs_info
, seed_devices
);
2671 btrfs_assign_next_active_device(fs_info
->fs_devices
->latest_dev
,
2675 device
->fs_devices
= fs_devices
;
2677 mutex_lock(&fs_info
->chunk_mutex
);
2678 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
2679 list_add(&device
->dev_alloc_list
, &fs_devices
->alloc_list
);
2680 fs_devices
->num_devices
++;
2681 fs_devices
->open_devices
++;
2682 fs_devices
->rw_devices
++;
2683 fs_devices
->total_devices
++;
2684 fs_devices
->total_rw_bytes
+= device
->total_bytes
;
2686 atomic64_add(device
->total_bytes
, &fs_info
->free_chunk_space
);
2688 if (!bdev_nonrot(device
->bdev
))
2689 fs_devices
->rotating
= true;
2691 orig_super_total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
2692 btrfs_set_super_total_bytes(fs_info
->super_copy
,
2693 round_down(orig_super_total_bytes
+ device
->total_bytes
,
2694 fs_info
->sectorsize
));
2696 orig_super_num_devices
= btrfs_super_num_devices(fs_info
->super_copy
);
2697 btrfs_set_super_num_devices(fs_info
->super_copy
,
2698 orig_super_num_devices
+ 1);
2701 * we've got more storage, clear any full flags on the space
2704 btrfs_clear_space_info_full(fs_info
);
2706 mutex_unlock(&fs_info
->chunk_mutex
);
2708 /* Add sysfs device entry */
2709 btrfs_sysfs_add_device(device
);
2711 mutex_unlock(&fs_devices
->device_list_mutex
);
2714 mutex_lock(&fs_info
->chunk_mutex
);
2715 ret
= init_first_rw_device(trans
);
2716 mutex_unlock(&fs_info
->chunk_mutex
);
2718 btrfs_abort_transaction(trans
, ret
);
2723 ret
= btrfs_add_dev_item(trans
, device
);
2725 btrfs_abort_transaction(trans
, ret
);
2730 ret
= btrfs_finish_sprout(trans
);
2732 btrfs_abort_transaction(trans
, ret
);
2737 * fs_devices now represents the newly sprouted filesystem and
2738 * its fsid has been changed by btrfs_sprout_splice().
2740 btrfs_sysfs_update_sprout_fsid(fs_devices
);
2743 ret
= btrfs_commit_transaction(trans
);
2746 mutex_unlock(&uuid_mutex
);
2747 up_write(&sb
->s_umount
);
2750 if (ret
) /* transaction commit */
2753 ret
= btrfs_relocate_sys_chunks(fs_info
);
2755 btrfs_handle_fs_error(fs_info
, ret
,
2756 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2757 trans
= btrfs_attach_transaction(root
);
2758 if (IS_ERR(trans
)) {
2759 if (PTR_ERR(trans
) == -ENOENT
)
2761 ret
= PTR_ERR(trans
);
2765 ret
= btrfs_commit_transaction(trans
);
2769 * Now that we have written a new super block to this device, check all
2770 * other fs_devices list if device_path alienates any other scanned
2772 * We can ignore the return value as it typically returns -EINVAL and
2773 * only succeeds if the device was an alien.
2775 btrfs_forget_devices(device
->devt
);
2777 /* Update ctime/mtime for blkid or udev */
2778 update_dev_time(device_path
);
2783 btrfs_sysfs_remove_device(device
);
2784 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2785 mutex_lock(&fs_info
->chunk_mutex
);
2786 list_del_rcu(&device
->dev_list
);
2787 list_del(&device
->dev_alloc_list
);
2788 fs_info
->fs_devices
->num_devices
--;
2789 fs_info
->fs_devices
->open_devices
--;
2790 fs_info
->fs_devices
->rw_devices
--;
2791 fs_info
->fs_devices
->total_devices
--;
2792 fs_info
->fs_devices
->total_rw_bytes
-= device
->total_bytes
;
2793 atomic64_sub(device
->total_bytes
, &fs_info
->free_chunk_space
);
2794 btrfs_set_super_total_bytes(fs_info
->super_copy
,
2795 orig_super_total_bytes
);
2796 btrfs_set_super_num_devices(fs_info
->super_copy
,
2797 orig_super_num_devices
);
2798 mutex_unlock(&fs_info
->chunk_mutex
);
2799 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2802 btrfs_set_sb_rdonly(sb
);
2804 btrfs_end_transaction(trans
);
2806 btrfs_destroy_dev_zone_info(device
);
2808 btrfs_free_device(device
);
2810 bdev_release(bdev_handle
);
2812 mutex_unlock(&uuid_mutex
);
2813 up_write(&sb
->s_umount
);
2818 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2819 struct btrfs_device
*device
)
2822 struct btrfs_path
*path
;
2823 struct btrfs_root
*root
= device
->fs_info
->chunk_root
;
2824 struct btrfs_dev_item
*dev_item
;
2825 struct extent_buffer
*leaf
;
2826 struct btrfs_key key
;
2828 path
= btrfs_alloc_path();
2832 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2833 key
.type
= BTRFS_DEV_ITEM_KEY
;
2834 key
.offset
= device
->devid
;
2836 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2845 leaf
= path
->nodes
[0];
2846 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2848 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2849 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2850 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2851 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2852 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2853 btrfs_set_device_total_bytes(leaf
, dev_item
,
2854 btrfs_device_get_disk_total_bytes(device
));
2855 btrfs_set_device_bytes_used(leaf
, dev_item
,
2856 btrfs_device_get_bytes_used(device
));
2857 btrfs_mark_buffer_dirty(trans
, leaf
);
2860 btrfs_free_path(path
);
2864 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2865 struct btrfs_device
*device
, u64 new_size
)
2867 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
2868 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
2873 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
))
2876 new_size
= round_down(new_size
, fs_info
->sectorsize
);
2878 mutex_lock(&fs_info
->chunk_mutex
);
2879 old_total
= btrfs_super_total_bytes(super_copy
);
2880 diff
= round_down(new_size
- device
->total_bytes
, fs_info
->sectorsize
);
2882 if (new_size
<= device
->total_bytes
||
2883 test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
)) {
2884 mutex_unlock(&fs_info
->chunk_mutex
);
2888 btrfs_set_super_total_bytes(super_copy
,
2889 round_down(old_total
+ diff
, fs_info
->sectorsize
));
2890 device
->fs_devices
->total_rw_bytes
+= diff
;
2891 atomic64_add(diff
, &fs_info
->free_chunk_space
);
2893 btrfs_device_set_total_bytes(device
, new_size
);
2894 btrfs_device_set_disk_total_bytes(device
, new_size
);
2895 btrfs_clear_space_info_full(device
->fs_info
);
2896 if (list_empty(&device
->post_commit_list
))
2897 list_add_tail(&device
->post_commit_list
,
2898 &trans
->transaction
->dev_update_list
);
2899 mutex_unlock(&fs_info
->chunk_mutex
);
2901 btrfs_reserve_chunk_metadata(trans
, false);
2902 ret
= btrfs_update_device(trans
, device
);
2903 btrfs_trans_release_chunk_metadata(trans
);
2908 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
, u64 chunk_offset
)
2910 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2911 struct btrfs_root
*root
= fs_info
->chunk_root
;
2913 struct btrfs_path
*path
;
2914 struct btrfs_key key
;
2916 path
= btrfs_alloc_path();
2920 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2921 key
.offset
= chunk_offset
;
2922 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2924 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2927 else if (ret
> 0) { /* Logic error or corruption */
2928 btrfs_handle_fs_error(fs_info
, -ENOENT
,
2929 "Failed lookup while freeing chunk.");
2934 ret
= btrfs_del_item(trans
, root
, path
);
2936 btrfs_handle_fs_error(fs_info
, ret
,
2937 "Failed to delete chunk item.");
2939 btrfs_free_path(path
);
2943 static int btrfs_del_sys_chunk(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2945 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
2946 struct btrfs_disk_key
*disk_key
;
2947 struct btrfs_chunk
*chunk
;
2954 struct btrfs_key key
;
2956 lockdep_assert_held(&fs_info
->chunk_mutex
);
2957 array_size
= btrfs_super_sys_array_size(super_copy
);
2959 ptr
= super_copy
->sys_chunk_array
;
2962 while (cur
< array_size
) {
2963 disk_key
= (struct btrfs_disk_key
*)ptr
;
2964 btrfs_disk_key_to_cpu(&key
, disk_key
);
2966 len
= sizeof(*disk_key
);
2968 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2969 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2970 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2971 len
+= btrfs_chunk_item_size(num_stripes
);
2976 if (key
.objectid
== BTRFS_FIRST_CHUNK_TREE_OBJECTID
&&
2977 key
.offset
== chunk_offset
) {
2978 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2980 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2990 * Find the mapping containing the given logical extent.
2992 * @logical: Logical block offset in bytes.
2993 * @length: Length of extent in bytes.
2995 * Return: Chunk mapping or ERR_PTR.
2997 struct extent_map
*btrfs_get_chunk_map(struct btrfs_fs_info
*fs_info
,
2998 u64 logical
, u64 length
)
3000 struct extent_map_tree
*em_tree
;
3001 struct extent_map
*em
;
3003 em_tree
= &fs_info
->mapping_tree
;
3004 read_lock(&em_tree
->lock
);
3005 em
= lookup_extent_mapping(em_tree
, logical
, length
);
3006 read_unlock(&em_tree
->lock
);
3010 "unable to find chunk map for logical %llu length %llu",
3012 return ERR_PTR(-EINVAL
);
3015 if (em
->start
> logical
|| em
->start
+ em
->len
<= logical
) {
3017 "found a bad chunk map, wanted %llu-%llu, found %llu-%llu",
3018 logical
, logical
+ length
, em
->start
, em
->start
+ em
->len
);
3019 free_extent_map(em
);
3020 return ERR_PTR(-EINVAL
);
3023 /* callers are responsible for dropping em's ref. */
3027 static int remove_chunk_item(struct btrfs_trans_handle
*trans
,
3028 struct map_lookup
*map
, u64 chunk_offset
)
3033 * Removing chunk items and updating the device items in the chunks btree
3034 * requires holding the chunk_mutex.
3035 * See the comment at btrfs_chunk_alloc() for the details.
3037 lockdep_assert_held(&trans
->fs_info
->chunk_mutex
);
3039 for (i
= 0; i
< map
->num_stripes
; i
++) {
3042 ret
= btrfs_update_device(trans
, map
->stripes
[i
].dev
);
3047 return btrfs_free_chunk(trans
, chunk_offset
);
3050 int btrfs_remove_chunk(struct btrfs_trans_handle
*trans
, u64 chunk_offset
)
3052 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
3053 struct extent_map
*em
;
3054 struct map_lookup
*map
;
3055 u64 dev_extent_len
= 0;
3057 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
3059 em
= btrfs_get_chunk_map(fs_info
, chunk_offset
, 1);
3062 * This is a logic error, but we don't want to just rely on the
3063 * user having built with ASSERT enabled, so if ASSERT doesn't
3064 * do anything we still error out.
3069 map
= em
->map_lookup
;
3072 * First delete the device extent items from the devices btree.
3073 * We take the device_list_mutex to avoid racing with the finishing phase
3074 * of a device replace operation. See the comment below before acquiring
3075 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
3076 * because that can result in a deadlock when deleting the device extent
3077 * items from the devices btree - COWing an extent buffer from the btree
3078 * may result in allocating a new metadata chunk, which would attempt to
3079 * lock again fs_info->chunk_mutex.
3081 mutex_lock(&fs_devices
->device_list_mutex
);
3082 for (i
= 0; i
< map
->num_stripes
; i
++) {
3083 struct btrfs_device
*device
= map
->stripes
[i
].dev
;
3084 ret
= btrfs_free_dev_extent(trans
, device
,
3085 map
->stripes
[i
].physical
,
3088 mutex_unlock(&fs_devices
->device_list_mutex
);
3089 btrfs_abort_transaction(trans
, ret
);
3093 if (device
->bytes_used
> 0) {
3094 mutex_lock(&fs_info
->chunk_mutex
);
3095 btrfs_device_set_bytes_used(device
,
3096 device
->bytes_used
- dev_extent_len
);
3097 atomic64_add(dev_extent_len
, &fs_info
->free_chunk_space
);
3098 btrfs_clear_space_info_full(fs_info
);
3099 mutex_unlock(&fs_info
->chunk_mutex
);
3102 mutex_unlock(&fs_devices
->device_list_mutex
);
3105 * We acquire fs_info->chunk_mutex for 2 reasons:
3107 * 1) Just like with the first phase of the chunk allocation, we must
3108 * reserve system space, do all chunk btree updates and deletions, and
3109 * update the system chunk array in the superblock while holding this
3110 * mutex. This is for similar reasons as explained on the comment at
3111 * the top of btrfs_chunk_alloc();
3113 * 2) Prevent races with the final phase of a device replace operation
3114 * that replaces the device object associated with the map's stripes,
3115 * because the device object's id can change at any time during that
3116 * final phase of the device replace operation
3117 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3118 * replaced device and then see it with an ID of
3119 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3120 * the device item, which does not exists on the chunk btree.
3121 * The finishing phase of device replace acquires both the
3122 * device_list_mutex and the chunk_mutex, in that order, so we are
3123 * safe by just acquiring the chunk_mutex.
3125 trans
->removing_chunk
= true;
3126 mutex_lock(&fs_info
->chunk_mutex
);
3128 check_system_chunk(trans
, map
->type
);
3130 ret
= remove_chunk_item(trans
, map
, chunk_offset
);
3132 * Normally we should not get -ENOSPC since we reserved space before
3133 * through the call to check_system_chunk().
3135 * Despite our system space_info having enough free space, we may not
3136 * be able to allocate extents from its block groups, because all have
3137 * an incompatible profile, which will force us to allocate a new system
3138 * block group with the right profile, or right after we called
3139 * check_system_space() above, a scrub turned the only system block group
3140 * with enough free space into RO mode.
3141 * This is explained with more detail at do_chunk_alloc().
3143 * So if we get -ENOSPC, allocate a new system chunk and retry once.
3145 if (ret
== -ENOSPC
) {
3146 const u64 sys_flags
= btrfs_system_alloc_profile(fs_info
);
3147 struct btrfs_block_group
*sys_bg
;
3149 sys_bg
= btrfs_create_chunk(trans
, sys_flags
);
3150 if (IS_ERR(sys_bg
)) {
3151 ret
= PTR_ERR(sys_bg
);
3152 btrfs_abort_transaction(trans
, ret
);
3156 ret
= btrfs_chunk_alloc_add_chunk_item(trans
, sys_bg
);
3158 btrfs_abort_transaction(trans
, ret
);
3162 ret
= remove_chunk_item(trans
, map
, chunk_offset
);
3164 btrfs_abort_transaction(trans
, ret
);
3168 btrfs_abort_transaction(trans
, ret
);
3172 trace_btrfs_chunk_free(fs_info
, map
, chunk_offset
, em
->len
);
3174 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3175 ret
= btrfs_del_sys_chunk(fs_info
, chunk_offset
);
3177 btrfs_abort_transaction(trans
, ret
);
3182 mutex_unlock(&fs_info
->chunk_mutex
);
3183 trans
->removing_chunk
= false;
3186 * We are done with chunk btree updates and deletions, so release the
3187 * system space we previously reserved (with check_system_chunk()).
3189 btrfs_trans_release_chunk_metadata(trans
);
3191 ret
= btrfs_remove_block_group(trans
, chunk_offset
, em
);
3193 btrfs_abort_transaction(trans
, ret
);
3198 if (trans
->removing_chunk
) {
3199 mutex_unlock(&fs_info
->chunk_mutex
);
3200 trans
->removing_chunk
= false;
3203 free_extent_map(em
);
3207 int btrfs_relocate_chunk(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
3209 struct btrfs_root
*root
= fs_info
->chunk_root
;
3210 struct btrfs_trans_handle
*trans
;
3211 struct btrfs_block_group
*block_group
;
3215 if (btrfs_fs_incompat(fs_info
, EXTENT_TREE_V2
)) {
3217 "relocate: not supported on extent tree v2 yet");
3222 * Prevent races with automatic removal of unused block groups.
3223 * After we relocate and before we remove the chunk with offset
3224 * chunk_offset, automatic removal of the block group can kick in,
3225 * resulting in a failure when calling btrfs_remove_chunk() below.
3227 * Make sure to acquire this mutex before doing a tree search (dev
3228 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3229 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3230 * we release the path used to search the chunk/dev tree and before
3231 * the current task acquires this mutex and calls us.
3233 lockdep_assert_held(&fs_info
->reclaim_bgs_lock
);
3235 /* step one, relocate all the extents inside this chunk */
3236 btrfs_scrub_pause(fs_info
);
3237 ret
= btrfs_relocate_block_group(fs_info
, chunk_offset
);
3238 btrfs_scrub_continue(fs_info
);
3241 * If we had a transaction abort, stop all running scrubs.
3242 * See transaction.c:cleanup_transaction() why we do it here.
3244 if (BTRFS_FS_ERROR(fs_info
))
3245 btrfs_scrub_cancel(fs_info
);
3249 block_group
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3252 btrfs_discard_cancel_work(&fs_info
->discard_ctl
, block_group
);
3253 length
= block_group
->length
;
3254 btrfs_put_block_group(block_group
);
3257 * On a zoned file system, discard the whole block group, this will
3258 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3259 * resetting the zone fails, don't treat it as a fatal problem from the
3260 * filesystem's point of view.
3262 if (btrfs_is_zoned(fs_info
)) {
3263 ret
= btrfs_discard_extent(fs_info
, chunk_offset
, length
, NULL
);
3266 "failed to reset zone %llu after relocation",
3270 trans
= btrfs_start_trans_remove_block_group(root
->fs_info
,
3272 if (IS_ERR(trans
)) {
3273 ret
= PTR_ERR(trans
);
3274 btrfs_handle_fs_error(root
->fs_info
, ret
, NULL
);
3279 * step two, delete the device extents and the
3280 * chunk tree entries
3282 ret
= btrfs_remove_chunk(trans
, chunk_offset
);
3283 btrfs_end_transaction(trans
);
3287 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info
*fs_info
)
3289 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
3290 struct btrfs_path
*path
;
3291 struct extent_buffer
*leaf
;
3292 struct btrfs_chunk
*chunk
;
3293 struct btrfs_key key
;
3294 struct btrfs_key found_key
;
3296 bool retried
= false;
3300 path
= btrfs_alloc_path();
3305 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3306 key
.offset
= (u64
)-1;
3307 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3310 mutex_lock(&fs_info
->reclaim_bgs_lock
);
3311 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
3313 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
3316 BUG_ON(ret
== 0); /* Corruption */
3318 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
3321 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
3327 leaf
= path
->nodes
[0];
3328 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3330 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
3331 struct btrfs_chunk
);
3332 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3333 btrfs_release_path(path
);
3335 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3336 ret
= btrfs_relocate_chunk(fs_info
, found_key
.offset
);
3342 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
3344 if (found_key
.offset
== 0)
3346 key
.offset
= found_key
.offset
- 1;
3349 if (failed
&& !retried
) {
3353 } else if (WARN_ON(failed
&& retried
)) {
3357 btrfs_free_path(path
);
3362 * return 1 : allocate a data chunk successfully,
3363 * return <0: errors during allocating a data chunk,
3364 * return 0 : no need to allocate a data chunk.
3366 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info
*fs_info
,
3369 struct btrfs_block_group
*cache
;
3373 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3375 chunk_type
= cache
->flags
;
3376 btrfs_put_block_group(cache
);
3378 if (!(chunk_type
& BTRFS_BLOCK_GROUP_DATA
))
3381 spin_lock(&fs_info
->data_sinfo
->lock
);
3382 bytes_used
= fs_info
->data_sinfo
->bytes_used
;
3383 spin_unlock(&fs_info
->data_sinfo
->lock
);
3386 struct btrfs_trans_handle
*trans
;
3389 trans
= btrfs_join_transaction(fs_info
->tree_root
);
3391 return PTR_ERR(trans
);
3393 ret
= btrfs_force_chunk_alloc(trans
, BTRFS_BLOCK_GROUP_DATA
);
3394 btrfs_end_transaction(trans
);
3403 static int insert_balance_item(struct btrfs_fs_info
*fs_info
,
3404 struct btrfs_balance_control
*bctl
)
3406 struct btrfs_root
*root
= fs_info
->tree_root
;
3407 struct btrfs_trans_handle
*trans
;
3408 struct btrfs_balance_item
*item
;
3409 struct btrfs_disk_balance_args disk_bargs
;
3410 struct btrfs_path
*path
;
3411 struct extent_buffer
*leaf
;
3412 struct btrfs_key key
;
3415 path
= btrfs_alloc_path();
3419 trans
= btrfs_start_transaction(root
, 0);
3420 if (IS_ERR(trans
)) {
3421 btrfs_free_path(path
);
3422 return PTR_ERR(trans
);
3425 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3426 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3429 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
3434 leaf
= path
->nodes
[0];
3435 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3437 memzero_extent_buffer(leaf
, (unsigned long)item
, sizeof(*item
));
3439 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
3440 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
3441 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
3442 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
3443 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
3444 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
3446 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
3448 btrfs_mark_buffer_dirty(trans
, leaf
);
3450 btrfs_free_path(path
);
3451 err
= btrfs_commit_transaction(trans
);
3457 static int del_balance_item(struct btrfs_fs_info
*fs_info
)
3459 struct btrfs_root
*root
= fs_info
->tree_root
;
3460 struct btrfs_trans_handle
*trans
;
3461 struct btrfs_path
*path
;
3462 struct btrfs_key key
;
3465 path
= btrfs_alloc_path();
3469 trans
= btrfs_start_transaction_fallback_global_rsv(root
, 0);
3470 if (IS_ERR(trans
)) {
3471 btrfs_free_path(path
);
3472 return PTR_ERR(trans
);
3475 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3476 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3479 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3487 ret
= btrfs_del_item(trans
, root
, path
);
3489 btrfs_free_path(path
);
3490 err
= btrfs_commit_transaction(trans
);
3497 * This is a heuristic used to reduce the number of chunks balanced on
3498 * resume after balance was interrupted.
3500 static void update_balance_args(struct btrfs_balance_control
*bctl
)
3503 * Turn on soft mode for chunk types that were being converted.
3505 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3506 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3507 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3508 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3509 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3510 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3513 * Turn on usage filter if is not already used. The idea is
3514 * that chunks that we have already balanced should be
3515 * reasonably full. Don't do it for chunks that are being
3516 * converted - that will keep us from relocating unconverted
3517 * (albeit full) chunks.
3519 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3520 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3521 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3522 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3523 bctl
->data
.usage
= 90;
3525 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3526 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3527 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3528 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3529 bctl
->sys
.usage
= 90;
3531 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3532 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3533 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3534 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3535 bctl
->meta
.usage
= 90;
3540 * Clear the balance status in fs_info and delete the balance item from disk.
3542 static void reset_balance_state(struct btrfs_fs_info
*fs_info
)
3544 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3547 BUG_ON(!fs_info
->balance_ctl
);
3549 spin_lock(&fs_info
->balance_lock
);
3550 fs_info
->balance_ctl
= NULL
;
3551 spin_unlock(&fs_info
->balance_lock
);
3554 ret
= del_balance_item(fs_info
);
3556 btrfs_handle_fs_error(fs_info
, ret
, NULL
);
3560 * Balance filters. Return 1 if chunk should be filtered out
3561 * (should not be balanced).
3563 static int chunk_profiles_filter(u64 chunk_type
,
3564 struct btrfs_balance_args
*bargs
)
3566 chunk_type
= chunk_to_extended(chunk_type
) &
3567 BTRFS_EXTENDED_PROFILE_MASK
;
3569 if (bargs
->profiles
& chunk_type
)
3575 static int chunk_usage_range_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
3576 struct btrfs_balance_args
*bargs
)
3578 struct btrfs_block_group
*cache
;
3580 u64 user_thresh_min
;
3581 u64 user_thresh_max
;
3584 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3585 chunk_used
= cache
->used
;
3587 if (bargs
->usage_min
== 0)
3588 user_thresh_min
= 0;
3590 user_thresh_min
= mult_perc(cache
->length
, bargs
->usage_min
);
3592 if (bargs
->usage_max
== 0)
3593 user_thresh_max
= 1;
3594 else if (bargs
->usage_max
> 100)
3595 user_thresh_max
= cache
->length
;
3597 user_thresh_max
= mult_perc(cache
->length
, bargs
->usage_max
);
3599 if (user_thresh_min
<= chunk_used
&& chunk_used
< user_thresh_max
)
3602 btrfs_put_block_group(cache
);
3606 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
,
3607 u64 chunk_offset
, struct btrfs_balance_args
*bargs
)
3609 struct btrfs_block_group
*cache
;
3610 u64 chunk_used
, user_thresh
;
3613 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3614 chunk_used
= cache
->used
;
3616 if (bargs
->usage_min
== 0)
3618 else if (bargs
->usage
> 100)
3619 user_thresh
= cache
->length
;
3621 user_thresh
= mult_perc(cache
->length
, bargs
->usage
);
3623 if (chunk_used
< user_thresh
)
3626 btrfs_put_block_group(cache
);
3630 static int chunk_devid_filter(struct extent_buffer
*leaf
,
3631 struct btrfs_chunk
*chunk
,
3632 struct btrfs_balance_args
*bargs
)
3634 struct btrfs_stripe
*stripe
;
3635 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3638 for (i
= 0; i
< num_stripes
; i
++) {
3639 stripe
= btrfs_stripe_nr(chunk
, i
);
3640 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
3647 static u64
calc_data_stripes(u64 type
, int num_stripes
)
3649 const int index
= btrfs_bg_flags_to_raid_index(type
);
3650 const int ncopies
= btrfs_raid_array
[index
].ncopies
;
3651 const int nparity
= btrfs_raid_array
[index
].nparity
;
3653 return (num_stripes
- nparity
) / ncopies
;
3656 /* [pstart, pend) */
3657 static int chunk_drange_filter(struct extent_buffer
*leaf
,
3658 struct btrfs_chunk
*chunk
,
3659 struct btrfs_balance_args
*bargs
)
3661 struct btrfs_stripe
*stripe
;
3662 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3669 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
3672 type
= btrfs_chunk_type(leaf
, chunk
);
3673 factor
= calc_data_stripes(type
, num_stripes
);
3675 for (i
= 0; i
< num_stripes
; i
++) {
3676 stripe
= btrfs_stripe_nr(chunk
, i
);
3677 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
3680 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
3681 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
3682 stripe_length
= div_u64(stripe_length
, factor
);
3684 if (stripe_offset
< bargs
->pend
&&
3685 stripe_offset
+ stripe_length
> bargs
->pstart
)
3692 /* [vstart, vend) */
3693 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
3694 struct btrfs_chunk
*chunk
,
3696 struct btrfs_balance_args
*bargs
)
3698 if (chunk_offset
< bargs
->vend
&&
3699 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
3700 /* at least part of the chunk is inside this vrange */
3706 static int chunk_stripes_range_filter(struct extent_buffer
*leaf
,
3707 struct btrfs_chunk
*chunk
,
3708 struct btrfs_balance_args
*bargs
)
3710 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3712 if (bargs
->stripes_min
<= num_stripes
3713 && num_stripes
<= bargs
->stripes_max
)
3719 static int chunk_soft_convert_filter(u64 chunk_type
,
3720 struct btrfs_balance_args
*bargs
)
3722 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
3725 chunk_type
= chunk_to_extended(chunk_type
) &
3726 BTRFS_EXTENDED_PROFILE_MASK
;
3728 if (bargs
->target
== chunk_type
)
3734 static int should_balance_chunk(struct extent_buffer
*leaf
,
3735 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
3737 struct btrfs_fs_info
*fs_info
= leaf
->fs_info
;
3738 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3739 struct btrfs_balance_args
*bargs
= NULL
;
3740 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3743 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
3744 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
3748 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3749 bargs
= &bctl
->data
;
3750 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3752 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3753 bargs
= &bctl
->meta
;
3755 /* profiles filter */
3756 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
3757 chunk_profiles_filter(chunk_type
, bargs
)) {
3762 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3763 chunk_usage_filter(fs_info
, chunk_offset
, bargs
)) {
3765 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3766 chunk_usage_range_filter(fs_info
, chunk_offset
, bargs
)) {
3771 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
3772 chunk_devid_filter(leaf
, chunk
, bargs
)) {
3776 /* drange filter, makes sense only with devid filter */
3777 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
3778 chunk_drange_filter(leaf
, chunk
, bargs
)) {
3783 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
3784 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3788 /* stripes filter */
3789 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_STRIPES_RANGE
) &&
3790 chunk_stripes_range_filter(leaf
, chunk
, bargs
)) {
3794 /* soft profile changing mode */
3795 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
3796 chunk_soft_convert_filter(chunk_type
, bargs
)) {
3801 * limited by count, must be the last filter
3803 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT
)) {
3804 if (bargs
->limit
== 0)
3808 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT_RANGE
)) {
3810 * Same logic as the 'limit' filter; the minimum cannot be
3811 * determined here because we do not have the global information
3812 * about the count of all chunks that satisfy the filters.
3814 if (bargs
->limit_max
== 0)
3823 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
3825 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3826 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
3828 struct btrfs_chunk
*chunk
;
3829 struct btrfs_path
*path
= NULL
;
3830 struct btrfs_key key
;
3831 struct btrfs_key found_key
;
3832 struct extent_buffer
*leaf
;
3835 int enospc_errors
= 0;
3836 bool counting
= true;
3837 /* The single value limit and min/max limits use the same bytes in the */
3838 u64 limit_data
= bctl
->data
.limit
;
3839 u64 limit_meta
= bctl
->meta
.limit
;
3840 u64 limit_sys
= bctl
->sys
.limit
;
3844 int chunk_reserved
= 0;
3846 path
= btrfs_alloc_path();
3852 /* zero out stat counters */
3853 spin_lock(&fs_info
->balance_lock
);
3854 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
3855 spin_unlock(&fs_info
->balance_lock
);
3859 * The single value limit and min/max limits use the same bytes
3862 bctl
->data
.limit
= limit_data
;
3863 bctl
->meta
.limit
= limit_meta
;
3864 bctl
->sys
.limit
= limit_sys
;
3866 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3867 key
.offset
= (u64
)-1;
3868 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3871 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
3872 atomic_read(&fs_info
->balance_cancel_req
)) {
3877 mutex_lock(&fs_info
->reclaim_bgs_lock
);
3878 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
3880 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
3885 * this shouldn't happen, it means the last relocate
3889 BUG(); /* FIXME break ? */
3891 ret
= btrfs_previous_item(chunk_root
, path
, 0,
3892 BTRFS_CHUNK_ITEM_KEY
);
3894 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
3899 leaf
= path
->nodes
[0];
3900 slot
= path
->slots
[0];
3901 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3903 if (found_key
.objectid
!= key
.objectid
) {
3904 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
3908 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
3909 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3912 spin_lock(&fs_info
->balance_lock
);
3913 bctl
->stat
.considered
++;
3914 spin_unlock(&fs_info
->balance_lock
);
3917 ret
= should_balance_chunk(leaf
, chunk
, found_key
.offset
);
3919 btrfs_release_path(path
);
3921 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
3926 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
3927 spin_lock(&fs_info
->balance_lock
);
3928 bctl
->stat
.expected
++;
3929 spin_unlock(&fs_info
->balance_lock
);
3931 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3933 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3935 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3942 * Apply limit_min filter, no need to check if the LIMITS
3943 * filter is used, limit_min is 0 by default
3945 if (((chunk_type
& BTRFS_BLOCK_GROUP_DATA
) &&
3946 count_data
< bctl
->data
.limit_min
)
3947 || ((chunk_type
& BTRFS_BLOCK_GROUP_METADATA
) &&
3948 count_meta
< bctl
->meta
.limit_min
)
3949 || ((chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) &&
3950 count_sys
< bctl
->sys
.limit_min
)) {
3951 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
3955 if (!chunk_reserved
) {
3957 * We may be relocating the only data chunk we have,
3958 * which could potentially end up with losing data's
3959 * raid profile, so lets allocate an empty one in
3962 ret
= btrfs_may_alloc_data_chunk(fs_info
,
3965 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
3967 } else if (ret
== 1) {
3972 ret
= btrfs_relocate_chunk(fs_info
, found_key
.offset
);
3973 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
3974 if (ret
== -ENOSPC
) {
3976 } else if (ret
== -ETXTBSY
) {
3978 "skipping relocation of block group %llu due to active swapfile",
3984 spin_lock(&fs_info
->balance_lock
);
3985 bctl
->stat
.completed
++;
3986 spin_unlock(&fs_info
->balance_lock
);
3989 if (found_key
.offset
== 0)
3991 key
.offset
= found_key
.offset
- 1;
3995 btrfs_release_path(path
);
4000 btrfs_free_path(path
);
4001 if (enospc_errors
) {
4002 btrfs_info(fs_info
, "%d enospc errors during balance",
4012 * See if a given profile is valid and reduced.
4014 * @flags: profile to validate
4015 * @extended: if true @flags is treated as an extended profile
4017 static int alloc_profile_is_valid(u64 flags
, int extended
)
4019 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
4020 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
4022 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
4024 /* 1) check that all other bits are zeroed */
4028 /* 2) see if profile is reduced */
4030 return !extended
; /* "0" is valid for usual profiles */
4032 return has_single_bit_set(flags
);
4036 * Validate target profile against allowed profiles and return true if it's OK.
4037 * Otherwise print the error message and return false.
4039 static inline int validate_convert_profile(struct btrfs_fs_info
*fs_info
,
4040 const struct btrfs_balance_args
*bargs
,
4041 u64 allowed
, const char *type
)
4043 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
4046 /* Profile is valid and does not have bits outside of the allowed set */
4047 if (alloc_profile_is_valid(bargs
->target
, 1) &&
4048 (bargs
->target
& ~allowed
) == 0)
4051 btrfs_err(fs_info
, "balance: invalid convert %s profile %s",
4052 type
, btrfs_bg_type_to_raid_name(bargs
->target
));
4057 * Fill @buf with textual description of balance filter flags @bargs, up to
4058 * @size_buf including the terminating null. The output may be trimmed if it
4059 * does not fit into the provided buffer.
4061 static void describe_balance_args(struct btrfs_balance_args
*bargs
, char *buf
,
4065 u32 size_bp
= size_buf
;
4067 u64 flags
= bargs
->flags
;
4068 char tmp_buf
[128] = {'\0'};
4073 #define CHECK_APPEND_NOARG(a) \
4075 ret = snprintf(bp, size_bp, (a)); \
4076 if (ret < 0 || ret >= size_bp) \
4077 goto out_overflow; \
4082 #define CHECK_APPEND_1ARG(a, v1) \
4084 ret = snprintf(bp, size_bp, (a), (v1)); \
4085 if (ret < 0 || ret >= size_bp) \
4086 goto out_overflow; \
4091 #define CHECK_APPEND_2ARG(a, v1, v2) \
4093 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4094 if (ret < 0 || ret >= size_bp) \
4095 goto out_overflow; \
4100 if (flags
& BTRFS_BALANCE_ARGS_CONVERT
)
4101 CHECK_APPEND_1ARG("convert=%s,",
4102 btrfs_bg_type_to_raid_name(bargs
->target
));
4104 if (flags
& BTRFS_BALANCE_ARGS_SOFT
)
4105 CHECK_APPEND_NOARG("soft,");
4107 if (flags
& BTRFS_BALANCE_ARGS_PROFILES
) {
4108 btrfs_describe_block_groups(bargs
->profiles
, tmp_buf
,
4110 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf
);
4113 if (flags
& BTRFS_BALANCE_ARGS_USAGE
)
4114 CHECK_APPEND_1ARG("usage=%llu,", bargs
->usage
);
4116 if (flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
)
4117 CHECK_APPEND_2ARG("usage=%u..%u,",
4118 bargs
->usage_min
, bargs
->usage_max
);
4120 if (flags
& BTRFS_BALANCE_ARGS_DEVID
)
4121 CHECK_APPEND_1ARG("devid=%llu,", bargs
->devid
);
4123 if (flags
& BTRFS_BALANCE_ARGS_DRANGE
)
4124 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4125 bargs
->pstart
, bargs
->pend
);
4127 if (flags
& BTRFS_BALANCE_ARGS_VRANGE
)
4128 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4129 bargs
->vstart
, bargs
->vend
);
4131 if (flags
& BTRFS_BALANCE_ARGS_LIMIT
)
4132 CHECK_APPEND_1ARG("limit=%llu,", bargs
->limit
);
4134 if (flags
& BTRFS_BALANCE_ARGS_LIMIT_RANGE
)
4135 CHECK_APPEND_2ARG("limit=%u..%u,",
4136 bargs
->limit_min
, bargs
->limit_max
);
4138 if (flags
& BTRFS_BALANCE_ARGS_STRIPES_RANGE
)
4139 CHECK_APPEND_2ARG("stripes=%u..%u,",
4140 bargs
->stripes_min
, bargs
->stripes_max
);
4142 #undef CHECK_APPEND_2ARG
4143 #undef CHECK_APPEND_1ARG
4144 #undef CHECK_APPEND_NOARG
4148 if (size_bp
< size_buf
)
4149 buf
[size_buf
- size_bp
- 1] = '\0'; /* remove last , */
4154 static void describe_balance_start_or_resume(struct btrfs_fs_info
*fs_info
)
4156 u32 size_buf
= 1024;
4157 char tmp_buf
[192] = {'\0'};
4160 u32 size_bp
= size_buf
;
4162 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
4164 buf
= kzalloc(size_buf
, GFP_KERNEL
);
4170 #define CHECK_APPEND_1ARG(a, v1) \
4172 ret = snprintf(bp, size_bp, (a), (v1)); \
4173 if (ret < 0 || ret >= size_bp) \
4174 goto out_overflow; \
4179 if (bctl
->flags
& BTRFS_BALANCE_FORCE
)
4180 CHECK_APPEND_1ARG("%s", "-f ");
4182 if (bctl
->flags
& BTRFS_BALANCE_DATA
) {
4183 describe_balance_args(&bctl
->data
, tmp_buf
, sizeof(tmp_buf
));
4184 CHECK_APPEND_1ARG("-d%s ", tmp_buf
);
4187 if (bctl
->flags
& BTRFS_BALANCE_METADATA
) {
4188 describe_balance_args(&bctl
->meta
, tmp_buf
, sizeof(tmp_buf
));
4189 CHECK_APPEND_1ARG("-m%s ", tmp_buf
);
4192 if (bctl
->flags
& BTRFS_BALANCE_SYSTEM
) {
4193 describe_balance_args(&bctl
->sys
, tmp_buf
, sizeof(tmp_buf
));
4194 CHECK_APPEND_1ARG("-s%s ", tmp_buf
);
4197 #undef CHECK_APPEND_1ARG
4201 if (size_bp
< size_buf
)
4202 buf
[size_buf
- size_bp
- 1] = '\0'; /* remove last " " */
4203 btrfs_info(fs_info
, "balance: %s %s",
4204 (bctl
->flags
& BTRFS_BALANCE_RESUME
) ?
4205 "resume" : "start", buf
);
4211 * Should be called with balance mutexe held
4213 int btrfs_balance(struct btrfs_fs_info
*fs_info
,
4214 struct btrfs_balance_control
*bctl
,
4215 struct btrfs_ioctl_balance_args
*bargs
)
4217 u64 meta_target
, data_target
;
4223 bool reducing_redundancy
;
4224 bool paused
= false;
4227 if (btrfs_fs_closing(fs_info
) ||
4228 atomic_read(&fs_info
->balance_pause_req
) ||
4229 btrfs_should_cancel_balance(fs_info
)) {
4234 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
4235 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
4239 * In case of mixed groups both data and meta should be picked,
4240 * and identical options should be given for both of them.
4242 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
4243 if (mixed
&& (bctl
->flags
& allowed
)) {
4244 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
4245 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
4246 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
4248 "balance: mixed groups data and metadata options must be the same");
4255 * rw_devices will not change at the moment, device add/delete/replace
4258 num_devices
= fs_info
->fs_devices
->rw_devices
;
4261 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4262 * special bit for it, to make it easier to distinguish. Thus we need
4263 * to set it manually, or balance would refuse the profile.
4265 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
4266 for (i
= 0; i
< ARRAY_SIZE(btrfs_raid_array
); i
++)
4267 if (num_devices
>= btrfs_raid_array
[i
].devs_min
)
4268 allowed
|= btrfs_raid_array
[i
].bg_flag
;
4270 if (!validate_convert_profile(fs_info
, &bctl
->data
, allowed
, "data") ||
4271 !validate_convert_profile(fs_info
, &bctl
->meta
, allowed
, "metadata") ||
4272 !validate_convert_profile(fs_info
, &bctl
->sys
, allowed
, "system")) {
4278 * Allow to reduce metadata or system integrity only if force set for
4279 * profiles with redundancy (copies, parity)
4282 for (i
= 0; i
< ARRAY_SIZE(btrfs_raid_array
); i
++) {
4283 if (btrfs_raid_array
[i
].ncopies
>= 2 ||
4284 btrfs_raid_array
[i
].tolerated_failures
>= 1)
4285 allowed
|= btrfs_raid_array
[i
].bg_flag
;
4288 seq
= read_seqbegin(&fs_info
->profiles_lock
);
4290 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
4291 (fs_info
->avail_system_alloc_bits
& allowed
) &&
4292 !(bctl
->sys
.target
& allowed
)) ||
4293 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
4294 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
4295 !(bctl
->meta
.target
& allowed
)))
4296 reducing_redundancy
= true;
4298 reducing_redundancy
= false;
4300 /* if we're not converting, the target field is uninitialized */
4301 meta_target
= (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) ?
4302 bctl
->meta
.target
: fs_info
->avail_metadata_alloc_bits
;
4303 data_target
= (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) ?
4304 bctl
->data
.target
: fs_info
->avail_data_alloc_bits
;
4305 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
4307 if (reducing_redundancy
) {
4308 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
4310 "balance: force reducing metadata redundancy");
4313 "balance: reduces metadata redundancy, use --force if you want this");
4319 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target
) <
4320 btrfs_get_num_tolerated_disk_barrier_failures(data_target
)) {
4322 "balance: metadata profile %s has lower redundancy than data profile %s",
4323 btrfs_bg_type_to_raid_name(meta_target
),
4324 btrfs_bg_type_to_raid_name(data_target
));
4327 ret
= insert_balance_item(fs_info
, bctl
);
4328 if (ret
&& ret
!= -EEXIST
)
4331 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
4332 BUG_ON(ret
== -EEXIST
);
4333 BUG_ON(fs_info
->balance_ctl
);
4334 spin_lock(&fs_info
->balance_lock
);
4335 fs_info
->balance_ctl
= bctl
;
4336 spin_unlock(&fs_info
->balance_lock
);
4338 BUG_ON(ret
!= -EEXIST
);
4339 spin_lock(&fs_info
->balance_lock
);
4340 update_balance_args(bctl
);
4341 spin_unlock(&fs_info
->balance_lock
);
4344 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
));
4345 set_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
);
4346 describe_balance_start_or_resume(fs_info
);
4347 mutex_unlock(&fs_info
->balance_mutex
);
4349 ret
= __btrfs_balance(fs_info
);
4351 mutex_lock(&fs_info
->balance_mutex
);
4352 if (ret
== -ECANCELED
&& atomic_read(&fs_info
->balance_pause_req
)) {
4353 btrfs_info(fs_info
, "balance: paused");
4354 btrfs_exclop_balance(fs_info
, BTRFS_EXCLOP_BALANCE_PAUSED
);
4358 * Balance can be canceled by:
4360 * - Regular cancel request
4361 * Then ret == -ECANCELED and balance_cancel_req > 0
4363 * - Fatal signal to "btrfs" process
4364 * Either the signal caught by wait_reserve_ticket() and callers
4365 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4367 * Either way, in this case balance_cancel_req = 0, and
4368 * ret == -EINTR or ret == -ECANCELED.
4370 * So here we only check the return value to catch canceled balance.
4372 else if (ret
== -ECANCELED
|| ret
== -EINTR
)
4373 btrfs_info(fs_info
, "balance: canceled");
4375 btrfs_info(fs_info
, "balance: ended with status: %d", ret
);
4377 clear_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
);
4380 memset(bargs
, 0, sizeof(*bargs
));
4381 btrfs_update_ioctl_balance_args(fs_info
, bargs
);
4384 /* We didn't pause, we can clean everything up. */
4386 reset_balance_state(fs_info
);
4387 btrfs_exclop_finish(fs_info
);
4390 wake_up(&fs_info
->balance_wait_q
);
4394 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
4395 reset_balance_state(fs_info
);
4398 btrfs_exclop_finish(fs_info
);
4403 static int balance_kthread(void *data
)
4405 struct btrfs_fs_info
*fs_info
= data
;
4408 sb_start_write(fs_info
->sb
);
4409 mutex_lock(&fs_info
->balance_mutex
);
4410 if (fs_info
->balance_ctl
)
4411 ret
= btrfs_balance(fs_info
, fs_info
->balance_ctl
, NULL
);
4412 mutex_unlock(&fs_info
->balance_mutex
);
4413 sb_end_write(fs_info
->sb
);
4418 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
4420 struct task_struct
*tsk
;
4422 mutex_lock(&fs_info
->balance_mutex
);
4423 if (!fs_info
->balance_ctl
) {
4424 mutex_unlock(&fs_info
->balance_mutex
);
4427 mutex_unlock(&fs_info
->balance_mutex
);
4429 if (btrfs_test_opt(fs_info
, SKIP_BALANCE
)) {
4430 btrfs_info(fs_info
, "balance: resume skipped");
4434 spin_lock(&fs_info
->super_lock
);
4435 ASSERT(fs_info
->exclusive_operation
== BTRFS_EXCLOP_BALANCE_PAUSED
);
4436 fs_info
->exclusive_operation
= BTRFS_EXCLOP_BALANCE
;
4437 spin_unlock(&fs_info
->super_lock
);
4439 * A ro->rw remount sequence should continue with the paused balance
4440 * regardless of who pauses it, system or the user as of now, so set
4443 spin_lock(&fs_info
->balance_lock
);
4444 fs_info
->balance_ctl
->flags
|= BTRFS_BALANCE_RESUME
;
4445 spin_unlock(&fs_info
->balance_lock
);
4447 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
4448 return PTR_ERR_OR_ZERO(tsk
);
4451 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
4453 struct btrfs_balance_control
*bctl
;
4454 struct btrfs_balance_item
*item
;
4455 struct btrfs_disk_balance_args disk_bargs
;
4456 struct btrfs_path
*path
;
4457 struct extent_buffer
*leaf
;
4458 struct btrfs_key key
;
4461 path
= btrfs_alloc_path();
4465 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
4466 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
4469 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
4472 if (ret
> 0) { /* ret = -ENOENT; */
4477 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
4483 leaf
= path
->nodes
[0];
4484 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
4486 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
4487 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
4489 btrfs_balance_data(leaf
, item
, &disk_bargs
);
4490 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
4491 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
4492 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
4493 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
4494 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
4497 * This should never happen, as the paused balance state is recovered
4498 * during mount without any chance of other exclusive ops to collide.
4500 * This gives the exclusive op status to balance and keeps in paused
4501 * state until user intervention (cancel or umount). If the ownership
4502 * cannot be assigned, show a message but do not fail. The balance
4503 * is in a paused state and must have fs_info::balance_ctl properly
4506 if (!btrfs_exclop_start(fs_info
, BTRFS_EXCLOP_BALANCE_PAUSED
))
4508 "balance: cannot set exclusive op status, resume manually");
4510 btrfs_release_path(path
);
4512 mutex_lock(&fs_info
->balance_mutex
);
4513 BUG_ON(fs_info
->balance_ctl
);
4514 spin_lock(&fs_info
->balance_lock
);
4515 fs_info
->balance_ctl
= bctl
;
4516 spin_unlock(&fs_info
->balance_lock
);
4517 mutex_unlock(&fs_info
->balance_mutex
);
4519 btrfs_free_path(path
);
4523 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
4527 mutex_lock(&fs_info
->balance_mutex
);
4528 if (!fs_info
->balance_ctl
) {
4529 mutex_unlock(&fs_info
->balance_mutex
);
4533 if (test_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
)) {
4534 atomic_inc(&fs_info
->balance_pause_req
);
4535 mutex_unlock(&fs_info
->balance_mutex
);
4537 wait_event(fs_info
->balance_wait_q
,
4538 !test_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
));
4540 mutex_lock(&fs_info
->balance_mutex
);
4541 /* we are good with balance_ctl ripped off from under us */
4542 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
));
4543 atomic_dec(&fs_info
->balance_pause_req
);
4548 mutex_unlock(&fs_info
->balance_mutex
);
4552 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
4554 mutex_lock(&fs_info
->balance_mutex
);
4555 if (!fs_info
->balance_ctl
) {
4556 mutex_unlock(&fs_info
->balance_mutex
);
4561 * A paused balance with the item stored on disk can be resumed at
4562 * mount time if the mount is read-write. Otherwise it's still paused
4563 * and we must not allow cancelling as it deletes the item.
4565 if (sb_rdonly(fs_info
->sb
)) {
4566 mutex_unlock(&fs_info
->balance_mutex
);
4570 atomic_inc(&fs_info
->balance_cancel_req
);
4572 * if we are running just wait and return, balance item is
4573 * deleted in btrfs_balance in this case
4575 if (test_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
)) {
4576 mutex_unlock(&fs_info
->balance_mutex
);
4577 wait_event(fs_info
->balance_wait_q
,
4578 !test_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
));
4579 mutex_lock(&fs_info
->balance_mutex
);
4581 mutex_unlock(&fs_info
->balance_mutex
);
4583 * Lock released to allow other waiters to continue, we'll
4584 * reexamine the status again.
4586 mutex_lock(&fs_info
->balance_mutex
);
4588 if (fs_info
->balance_ctl
) {
4589 reset_balance_state(fs_info
);
4590 btrfs_exclop_finish(fs_info
);
4591 btrfs_info(fs_info
, "balance: canceled");
4595 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
));
4596 atomic_dec(&fs_info
->balance_cancel_req
);
4597 mutex_unlock(&fs_info
->balance_mutex
);
4601 int btrfs_uuid_scan_kthread(void *data
)
4603 struct btrfs_fs_info
*fs_info
= data
;
4604 struct btrfs_root
*root
= fs_info
->tree_root
;
4605 struct btrfs_key key
;
4606 struct btrfs_path
*path
= NULL
;
4608 struct extent_buffer
*eb
;
4610 struct btrfs_root_item root_item
;
4612 struct btrfs_trans_handle
*trans
= NULL
;
4613 bool closing
= false;
4615 path
= btrfs_alloc_path();
4622 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4626 if (btrfs_fs_closing(fs_info
)) {
4630 ret
= btrfs_search_forward(root
, &key
, path
,
4631 BTRFS_OLDEST_GENERATION
);
4638 if (key
.type
!= BTRFS_ROOT_ITEM_KEY
||
4639 (key
.objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
4640 key
.objectid
!= BTRFS_FS_TREE_OBJECTID
) ||
4641 key
.objectid
> BTRFS_LAST_FREE_OBJECTID
)
4644 eb
= path
->nodes
[0];
4645 slot
= path
->slots
[0];
4646 item_size
= btrfs_item_size(eb
, slot
);
4647 if (item_size
< sizeof(root_item
))
4650 read_extent_buffer(eb
, &root_item
,
4651 btrfs_item_ptr_offset(eb
, slot
),
4652 (int)sizeof(root_item
));
4653 if (btrfs_root_refs(&root_item
) == 0)
4656 if (!btrfs_is_empty_uuid(root_item
.uuid
) ||
4657 !btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4661 btrfs_release_path(path
);
4663 * 1 - subvol uuid item
4664 * 1 - received_subvol uuid item
4666 trans
= btrfs_start_transaction(fs_info
->uuid_root
, 2);
4667 if (IS_ERR(trans
)) {
4668 ret
= PTR_ERR(trans
);
4676 btrfs_release_path(path
);
4677 if (!btrfs_is_empty_uuid(root_item
.uuid
)) {
4678 ret
= btrfs_uuid_tree_add(trans
, root_item
.uuid
,
4679 BTRFS_UUID_KEY_SUBVOL
,
4682 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4688 if (!btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4689 ret
= btrfs_uuid_tree_add(trans
,
4690 root_item
.received_uuid
,
4691 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
4694 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4701 btrfs_release_path(path
);
4703 ret
= btrfs_end_transaction(trans
);
4709 if (key
.offset
< (u64
)-1) {
4711 } else if (key
.type
< BTRFS_ROOT_ITEM_KEY
) {
4713 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4714 } else if (key
.objectid
< (u64
)-1) {
4716 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4725 btrfs_free_path(path
);
4726 if (trans
&& !IS_ERR(trans
))
4727 btrfs_end_transaction(trans
);
4729 btrfs_warn(fs_info
, "btrfs_uuid_scan_kthread failed %d", ret
);
4731 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN
, &fs_info
->flags
);
4732 up(&fs_info
->uuid_tree_rescan_sem
);
4736 int btrfs_create_uuid_tree(struct btrfs_fs_info
*fs_info
)
4738 struct btrfs_trans_handle
*trans
;
4739 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
4740 struct btrfs_root
*uuid_root
;
4741 struct task_struct
*task
;
4748 trans
= btrfs_start_transaction(tree_root
, 2);
4750 return PTR_ERR(trans
);
4752 uuid_root
= btrfs_create_tree(trans
, BTRFS_UUID_TREE_OBJECTID
);
4753 if (IS_ERR(uuid_root
)) {
4754 ret
= PTR_ERR(uuid_root
);
4755 btrfs_abort_transaction(trans
, ret
);
4756 btrfs_end_transaction(trans
);
4760 fs_info
->uuid_root
= uuid_root
;
4762 ret
= btrfs_commit_transaction(trans
);
4766 down(&fs_info
->uuid_tree_rescan_sem
);
4767 task
= kthread_run(btrfs_uuid_scan_kthread
, fs_info
, "btrfs-uuid");
4769 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4770 btrfs_warn(fs_info
, "failed to start uuid_scan task");
4771 up(&fs_info
->uuid_tree_rescan_sem
);
4772 return PTR_ERR(task
);
4779 * shrinking a device means finding all of the device extents past
4780 * the new size, and then following the back refs to the chunks.
4781 * The chunk relocation code actually frees the device extent
4783 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
4785 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
4786 struct btrfs_root
*root
= fs_info
->dev_root
;
4787 struct btrfs_trans_handle
*trans
;
4788 struct btrfs_dev_extent
*dev_extent
= NULL
;
4789 struct btrfs_path
*path
;
4795 bool retried
= false;
4796 struct extent_buffer
*l
;
4797 struct btrfs_key key
;
4798 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
4799 u64 old_total
= btrfs_super_total_bytes(super_copy
);
4800 u64 old_size
= btrfs_device_get_total_bytes(device
);
4805 new_size
= round_down(new_size
, fs_info
->sectorsize
);
4807 diff
= round_down(old_size
- new_size
, fs_info
->sectorsize
);
4809 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
))
4812 path
= btrfs_alloc_path();
4816 path
->reada
= READA_BACK
;
4818 trans
= btrfs_start_transaction(root
, 0);
4819 if (IS_ERR(trans
)) {
4820 btrfs_free_path(path
);
4821 return PTR_ERR(trans
);
4824 mutex_lock(&fs_info
->chunk_mutex
);
4826 btrfs_device_set_total_bytes(device
, new_size
);
4827 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
)) {
4828 device
->fs_devices
->total_rw_bytes
-= diff
;
4831 * The new free_chunk_space is new_size - used, so we have to
4832 * subtract the delta of the old free_chunk_space which included
4833 * old_size - used. If used > new_size then just subtract this
4834 * entire device's free space.
4836 if (device
->bytes_used
< new_size
)
4837 free_diff
= (old_size
- device
->bytes_used
) -
4838 (new_size
- device
->bytes_used
);
4840 free_diff
= old_size
- device
->bytes_used
;
4841 atomic64_sub(free_diff
, &fs_info
->free_chunk_space
);
4845 * Once the device's size has been set to the new size, ensure all
4846 * in-memory chunks are synced to disk so that the loop below sees them
4847 * and relocates them accordingly.
4849 if (contains_pending_extent(device
, &start
, diff
)) {
4850 mutex_unlock(&fs_info
->chunk_mutex
);
4851 ret
= btrfs_commit_transaction(trans
);
4855 mutex_unlock(&fs_info
->chunk_mutex
);
4856 btrfs_end_transaction(trans
);
4860 key
.objectid
= device
->devid
;
4861 key
.offset
= (u64
)-1;
4862 key
.type
= BTRFS_DEV_EXTENT_KEY
;
4865 mutex_lock(&fs_info
->reclaim_bgs_lock
);
4866 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4868 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
4872 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
4874 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
4878 btrfs_release_path(path
);
4883 slot
= path
->slots
[0];
4884 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
4886 if (key
.objectid
!= device
->devid
) {
4887 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
4888 btrfs_release_path(path
);
4892 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
4893 length
= btrfs_dev_extent_length(l
, dev_extent
);
4895 if (key
.offset
+ length
<= new_size
) {
4896 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
4897 btrfs_release_path(path
);
4901 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
4902 btrfs_release_path(path
);
4905 * We may be relocating the only data chunk we have,
4906 * which could potentially end up with losing data's
4907 * raid profile, so lets allocate an empty one in
4910 ret
= btrfs_may_alloc_data_chunk(fs_info
, chunk_offset
);
4912 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
4916 ret
= btrfs_relocate_chunk(fs_info
, chunk_offset
);
4917 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
4918 if (ret
== -ENOSPC
) {
4921 if (ret
== -ETXTBSY
) {
4923 "could not shrink block group %llu due to active swapfile",
4928 } while (key
.offset
-- > 0);
4930 if (failed
&& !retried
) {
4934 } else if (failed
&& retried
) {
4939 /* Shrinking succeeded, else we would be at "done". */
4940 trans
= btrfs_start_transaction(root
, 0);
4941 if (IS_ERR(trans
)) {
4942 ret
= PTR_ERR(trans
);
4946 mutex_lock(&fs_info
->chunk_mutex
);
4947 /* Clear all state bits beyond the shrunk device size */
4948 clear_extent_bits(&device
->alloc_state
, new_size
, (u64
)-1,
4951 btrfs_device_set_disk_total_bytes(device
, new_size
);
4952 if (list_empty(&device
->post_commit_list
))
4953 list_add_tail(&device
->post_commit_list
,
4954 &trans
->transaction
->dev_update_list
);
4956 WARN_ON(diff
> old_total
);
4957 btrfs_set_super_total_bytes(super_copy
,
4958 round_down(old_total
- diff
, fs_info
->sectorsize
));
4959 mutex_unlock(&fs_info
->chunk_mutex
);
4961 btrfs_reserve_chunk_metadata(trans
, false);
4962 /* Now btrfs_update_device() will change the on-disk size. */
4963 ret
= btrfs_update_device(trans
, device
);
4964 btrfs_trans_release_chunk_metadata(trans
);
4966 btrfs_abort_transaction(trans
, ret
);
4967 btrfs_end_transaction(trans
);
4969 ret
= btrfs_commit_transaction(trans
);
4972 btrfs_free_path(path
);
4974 mutex_lock(&fs_info
->chunk_mutex
);
4975 btrfs_device_set_total_bytes(device
, old_size
);
4976 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
)) {
4977 device
->fs_devices
->total_rw_bytes
+= diff
;
4978 atomic64_add(free_diff
, &fs_info
->free_chunk_space
);
4980 mutex_unlock(&fs_info
->chunk_mutex
);
4985 static int btrfs_add_system_chunk(struct btrfs_fs_info
*fs_info
,
4986 struct btrfs_key
*key
,
4987 struct btrfs_chunk
*chunk
, int item_size
)
4989 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
4990 struct btrfs_disk_key disk_key
;
4994 lockdep_assert_held(&fs_info
->chunk_mutex
);
4996 array_size
= btrfs_super_sys_array_size(super_copy
);
4997 if (array_size
+ item_size
+ sizeof(disk_key
)
4998 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
)
5001 ptr
= super_copy
->sys_chunk_array
+ array_size
;
5002 btrfs_cpu_key_to_disk(&disk_key
, key
);
5003 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
5004 ptr
+= sizeof(disk_key
);
5005 memcpy(ptr
, chunk
, item_size
);
5006 item_size
+= sizeof(disk_key
);
5007 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
5013 * sort the devices in descending order by max_avail, total_avail
5015 static int btrfs_cmp_device_info(const void *a
, const void *b
)
5017 const struct btrfs_device_info
*di_a
= a
;
5018 const struct btrfs_device_info
*di_b
= b
;
5020 if (di_a
->max_avail
> di_b
->max_avail
)
5022 if (di_a
->max_avail
< di_b
->max_avail
)
5024 if (di_a
->total_avail
> di_b
->total_avail
)
5026 if (di_a
->total_avail
< di_b
->total_avail
)
5031 static void check_raid56_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
5033 if (!(type
& BTRFS_BLOCK_GROUP_RAID56_MASK
))
5036 btrfs_set_fs_incompat(info
, RAID56
);
5039 static void check_raid1c34_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
5041 if (!(type
& (BTRFS_BLOCK_GROUP_RAID1C3
| BTRFS_BLOCK_GROUP_RAID1C4
)))
5044 btrfs_set_fs_incompat(info
, RAID1C34
);
5048 * Structure used internally for btrfs_create_chunk() function.
5049 * Wraps needed parameters.
5051 struct alloc_chunk_ctl
{
5054 /* Total number of stripes to allocate */
5056 /* sub_stripes info for map */
5058 /* Stripes per device */
5060 /* Maximum number of devices to use */
5062 /* Minimum number of devices to use */
5064 /* ndevs has to be a multiple of this */
5066 /* Number of copies */
5068 /* Number of stripes worth of bytes to store parity information */
5070 u64 max_stripe_size
;
5078 static void init_alloc_chunk_ctl_policy_regular(
5079 struct btrfs_fs_devices
*fs_devices
,
5080 struct alloc_chunk_ctl
*ctl
)
5082 struct btrfs_space_info
*space_info
;
5084 space_info
= btrfs_find_space_info(fs_devices
->fs_info
, ctl
->type
);
5087 ctl
->max_chunk_size
= READ_ONCE(space_info
->chunk_size
);
5088 ctl
->max_stripe_size
= min_t(u64
, ctl
->max_chunk_size
, SZ_1G
);
5090 if (ctl
->type
& BTRFS_BLOCK_GROUP_SYSTEM
)
5091 ctl
->devs_max
= min_t(int, ctl
->devs_max
, BTRFS_MAX_DEVS_SYS_CHUNK
);
5093 /* We don't want a chunk larger than 10% of writable space */
5094 ctl
->max_chunk_size
= min(mult_perc(fs_devices
->total_rw_bytes
, 10),
5095 ctl
->max_chunk_size
);
5096 ctl
->dev_extent_min
= btrfs_stripe_nr_to_offset(ctl
->dev_stripes
);
5099 static void init_alloc_chunk_ctl_policy_zoned(
5100 struct btrfs_fs_devices
*fs_devices
,
5101 struct alloc_chunk_ctl
*ctl
)
5103 u64 zone_size
= fs_devices
->fs_info
->zone_size
;
5105 int min_num_stripes
= ctl
->devs_min
* ctl
->dev_stripes
;
5106 int min_data_stripes
= (min_num_stripes
- ctl
->nparity
) / ctl
->ncopies
;
5107 u64 min_chunk_size
= min_data_stripes
* zone_size
;
5108 u64 type
= ctl
->type
;
5110 ctl
->max_stripe_size
= zone_size
;
5111 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
5112 ctl
->max_chunk_size
= round_down(BTRFS_MAX_DATA_CHUNK_SIZE
,
5114 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
5115 ctl
->max_chunk_size
= ctl
->max_stripe_size
;
5116 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
5117 ctl
->max_chunk_size
= 2 * ctl
->max_stripe_size
;
5118 ctl
->devs_max
= min_t(int, ctl
->devs_max
,
5119 BTRFS_MAX_DEVS_SYS_CHUNK
);
5124 /* We don't want a chunk larger than 10% of writable space */
5125 limit
= max(round_down(mult_perc(fs_devices
->total_rw_bytes
, 10),
5128 ctl
->max_chunk_size
= min(limit
, ctl
->max_chunk_size
);
5129 ctl
->dev_extent_min
= zone_size
* ctl
->dev_stripes
;
5132 static void init_alloc_chunk_ctl(struct btrfs_fs_devices
*fs_devices
,
5133 struct alloc_chunk_ctl
*ctl
)
5135 int index
= btrfs_bg_flags_to_raid_index(ctl
->type
);
5137 ctl
->sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
5138 ctl
->dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
5139 ctl
->devs_max
= btrfs_raid_array
[index
].devs_max
;
5141 ctl
->devs_max
= BTRFS_MAX_DEVS(fs_devices
->fs_info
);
5142 ctl
->devs_min
= btrfs_raid_array
[index
].devs_min
;
5143 ctl
->devs_increment
= btrfs_raid_array
[index
].devs_increment
;
5144 ctl
->ncopies
= btrfs_raid_array
[index
].ncopies
;
5145 ctl
->nparity
= btrfs_raid_array
[index
].nparity
;
5148 switch (fs_devices
->chunk_alloc_policy
) {
5149 case BTRFS_CHUNK_ALLOC_REGULAR
:
5150 init_alloc_chunk_ctl_policy_regular(fs_devices
, ctl
);
5152 case BTRFS_CHUNK_ALLOC_ZONED
:
5153 init_alloc_chunk_ctl_policy_zoned(fs_devices
, ctl
);
5160 static int gather_device_info(struct btrfs_fs_devices
*fs_devices
,
5161 struct alloc_chunk_ctl
*ctl
,
5162 struct btrfs_device_info
*devices_info
)
5164 struct btrfs_fs_info
*info
= fs_devices
->fs_info
;
5165 struct btrfs_device
*device
;
5167 u64 dev_extent_want
= ctl
->max_stripe_size
* ctl
->dev_stripes
;
5174 * in the first pass through the devices list, we gather information
5175 * about the available holes on each device.
5177 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
5178 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
)) {
5180 "BTRFS: read-only device in alloc_list\n");
5184 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
,
5185 &device
->dev_state
) ||
5186 test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
))
5189 if (device
->total_bytes
> device
->bytes_used
)
5190 total_avail
= device
->total_bytes
- device
->bytes_used
;
5194 /* If there is no space on this device, skip it. */
5195 if (total_avail
< ctl
->dev_extent_min
)
5198 ret
= find_free_dev_extent(device
, dev_extent_want
, &dev_offset
,
5200 if (ret
&& ret
!= -ENOSPC
)
5204 max_avail
= dev_extent_want
;
5206 if (max_avail
< ctl
->dev_extent_min
) {
5207 if (btrfs_test_opt(info
, ENOSPC_DEBUG
))
5209 "%s: devid %llu has no free space, have=%llu want=%llu",
5210 __func__
, device
->devid
, max_avail
,
5211 ctl
->dev_extent_min
);
5215 if (ndevs
== fs_devices
->rw_devices
) {
5216 WARN(1, "%s: found more than %llu devices\n",
5217 __func__
, fs_devices
->rw_devices
);
5220 devices_info
[ndevs
].dev_offset
= dev_offset
;
5221 devices_info
[ndevs
].max_avail
= max_avail
;
5222 devices_info
[ndevs
].total_avail
= total_avail
;
5223 devices_info
[ndevs
].dev
= device
;
5229 * now sort the devices by hole size / available space
5231 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
5232 btrfs_cmp_device_info
, NULL
);
5237 static int decide_stripe_size_regular(struct alloc_chunk_ctl
*ctl
,
5238 struct btrfs_device_info
*devices_info
)
5240 /* Number of stripes that count for block group size */
5244 * The primary goal is to maximize the number of stripes, so use as
5245 * many devices as possible, even if the stripes are not maximum sized.
5247 * The DUP profile stores more than one stripe per device, the
5248 * max_avail is the total size so we have to adjust.
5250 ctl
->stripe_size
= div_u64(devices_info
[ctl
->ndevs
- 1].max_avail
,
5252 ctl
->num_stripes
= ctl
->ndevs
* ctl
->dev_stripes
;
5254 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5255 data_stripes
= (ctl
->num_stripes
- ctl
->nparity
) / ctl
->ncopies
;
5258 * Use the number of data stripes to figure out how big this chunk is
5259 * really going to be in terms of logical address space, and compare
5260 * that answer with the max chunk size. If it's higher, we try to
5261 * reduce stripe_size.
5263 if (ctl
->stripe_size
* data_stripes
> ctl
->max_chunk_size
) {
5265 * Reduce stripe_size, round it up to a 16MB boundary again and
5266 * then use it, unless it ends up being even bigger than the
5267 * previous value we had already.
5269 ctl
->stripe_size
= min(round_up(div_u64(ctl
->max_chunk_size
,
5270 data_stripes
), SZ_16M
),
5274 /* Stripe size should not go beyond 1G. */
5275 ctl
->stripe_size
= min_t(u64
, ctl
->stripe_size
, SZ_1G
);
5277 /* Align to BTRFS_STRIPE_LEN */
5278 ctl
->stripe_size
= round_down(ctl
->stripe_size
, BTRFS_STRIPE_LEN
);
5279 ctl
->chunk_size
= ctl
->stripe_size
* data_stripes
;
5284 static int decide_stripe_size_zoned(struct alloc_chunk_ctl
*ctl
,
5285 struct btrfs_device_info
*devices_info
)
5287 u64 zone_size
= devices_info
[0].dev
->zone_info
->zone_size
;
5288 /* Number of stripes that count for block group size */
5292 * It should hold because:
5293 * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5295 ASSERT(devices_info
[ctl
->ndevs
- 1].max_avail
== ctl
->dev_extent_min
);
5297 ctl
->stripe_size
= zone_size
;
5298 ctl
->num_stripes
= ctl
->ndevs
* ctl
->dev_stripes
;
5299 data_stripes
= (ctl
->num_stripes
- ctl
->nparity
) / ctl
->ncopies
;
5301 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5302 if (ctl
->stripe_size
* data_stripes
> ctl
->max_chunk_size
) {
5303 ctl
->ndevs
= div_u64(div_u64(ctl
->max_chunk_size
* ctl
->ncopies
,
5304 ctl
->stripe_size
) + ctl
->nparity
,
5306 ctl
->num_stripes
= ctl
->ndevs
* ctl
->dev_stripes
;
5307 data_stripes
= (ctl
->num_stripes
- ctl
->nparity
) / ctl
->ncopies
;
5308 ASSERT(ctl
->stripe_size
* data_stripes
<= ctl
->max_chunk_size
);
5311 ctl
->chunk_size
= ctl
->stripe_size
* data_stripes
;
5316 static int decide_stripe_size(struct btrfs_fs_devices
*fs_devices
,
5317 struct alloc_chunk_ctl
*ctl
,
5318 struct btrfs_device_info
*devices_info
)
5320 struct btrfs_fs_info
*info
= fs_devices
->fs_info
;
5323 * Round down to number of usable stripes, devs_increment can be any
5324 * number so we can't use round_down() that requires power of 2, while
5325 * rounddown is safe.
5327 ctl
->ndevs
= rounddown(ctl
->ndevs
, ctl
->devs_increment
);
5329 if (ctl
->ndevs
< ctl
->devs_min
) {
5330 if (btrfs_test_opt(info
, ENOSPC_DEBUG
)) {
5332 "%s: not enough devices with free space: have=%d minimum required=%d",
5333 __func__
, ctl
->ndevs
, ctl
->devs_min
);
5338 ctl
->ndevs
= min(ctl
->ndevs
, ctl
->devs_max
);
5340 switch (fs_devices
->chunk_alloc_policy
) {
5341 case BTRFS_CHUNK_ALLOC_REGULAR
:
5342 return decide_stripe_size_regular(ctl
, devices_info
);
5343 case BTRFS_CHUNK_ALLOC_ZONED
:
5344 return decide_stripe_size_zoned(ctl
, devices_info
);
5350 static struct btrfs_block_group
*create_chunk(struct btrfs_trans_handle
*trans
,
5351 struct alloc_chunk_ctl
*ctl
,
5352 struct btrfs_device_info
*devices_info
)
5354 struct btrfs_fs_info
*info
= trans
->fs_info
;
5355 struct map_lookup
*map
= NULL
;
5356 struct extent_map_tree
*em_tree
;
5357 struct btrfs_block_group
*block_group
;
5358 struct extent_map
*em
;
5359 u64 start
= ctl
->start
;
5360 u64 type
= ctl
->type
;
5365 map
= kmalloc(map_lookup_size(ctl
->num_stripes
), GFP_NOFS
);
5367 return ERR_PTR(-ENOMEM
);
5368 map
->num_stripes
= ctl
->num_stripes
;
5370 for (i
= 0; i
< ctl
->ndevs
; ++i
) {
5371 for (j
= 0; j
< ctl
->dev_stripes
; ++j
) {
5372 int s
= i
* ctl
->dev_stripes
+ j
;
5373 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
5374 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
5375 j
* ctl
->stripe_size
;
5378 map
->io_align
= BTRFS_STRIPE_LEN
;
5379 map
->io_width
= BTRFS_STRIPE_LEN
;
5381 map
->sub_stripes
= ctl
->sub_stripes
;
5383 trace_btrfs_chunk_alloc(info
, map
, start
, ctl
->chunk_size
);
5385 em
= alloc_extent_map();
5388 return ERR_PTR(-ENOMEM
);
5390 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
5391 em
->map_lookup
= map
;
5393 em
->len
= ctl
->chunk_size
;
5394 em
->block_start
= 0;
5395 em
->block_len
= em
->len
;
5396 em
->orig_block_len
= ctl
->stripe_size
;
5398 em_tree
= &info
->mapping_tree
;
5399 write_lock(&em_tree
->lock
);
5400 ret
= add_extent_mapping(em_tree
, em
, 0);
5402 write_unlock(&em_tree
->lock
);
5403 free_extent_map(em
);
5404 return ERR_PTR(ret
);
5406 write_unlock(&em_tree
->lock
);
5408 block_group
= btrfs_make_block_group(trans
, type
, start
, ctl
->chunk_size
);
5409 if (IS_ERR(block_group
))
5410 goto error_del_extent
;
5412 for (i
= 0; i
< map
->num_stripes
; i
++) {
5413 struct btrfs_device
*dev
= map
->stripes
[i
].dev
;
5415 btrfs_device_set_bytes_used(dev
,
5416 dev
->bytes_used
+ ctl
->stripe_size
);
5417 if (list_empty(&dev
->post_commit_list
))
5418 list_add_tail(&dev
->post_commit_list
,
5419 &trans
->transaction
->dev_update_list
);
5422 atomic64_sub(ctl
->stripe_size
* map
->num_stripes
,
5423 &info
->free_chunk_space
);
5425 free_extent_map(em
);
5426 check_raid56_incompat_flag(info
, type
);
5427 check_raid1c34_incompat_flag(info
, type
);
5432 write_lock(&em_tree
->lock
);
5433 remove_extent_mapping(em_tree
, em
);
5434 write_unlock(&em_tree
->lock
);
5436 /* One for our allocation */
5437 free_extent_map(em
);
5438 /* One for the tree reference */
5439 free_extent_map(em
);
5444 struct btrfs_block_group
*btrfs_create_chunk(struct btrfs_trans_handle
*trans
,
5447 struct btrfs_fs_info
*info
= trans
->fs_info
;
5448 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
5449 struct btrfs_device_info
*devices_info
= NULL
;
5450 struct alloc_chunk_ctl ctl
;
5451 struct btrfs_block_group
*block_group
;
5454 lockdep_assert_held(&info
->chunk_mutex
);
5456 if (!alloc_profile_is_valid(type
, 0)) {
5458 return ERR_PTR(-EINVAL
);
5461 if (list_empty(&fs_devices
->alloc_list
)) {
5462 if (btrfs_test_opt(info
, ENOSPC_DEBUG
))
5463 btrfs_debug(info
, "%s: no writable device", __func__
);
5464 return ERR_PTR(-ENOSPC
);
5467 if (!(type
& BTRFS_BLOCK_GROUP_TYPE_MASK
)) {
5468 btrfs_err(info
, "invalid chunk type 0x%llx requested", type
);
5470 return ERR_PTR(-EINVAL
);
5473 ctl
.start
= find_next_chunk(info
);
5475 init_alloc_chunk_ctl(fs_devices
, &ctl
);
5477 devices_info
= kcalloc(fs_devices
->rw_devices
, sizeof(*devices_info
),
5480 return ERR_PTR(-ENOMEM
);
5482 ret
= gather_device_info(fs_devices
, &ctl
, devices_info
);
5484 block_group
= ERR_PTR(ret
);
5488 ret
= decide_stripe_size(fs_devices
, &ctl
, devices_info
);
5490 block_group
= ERR_PTR(ret
);
5494 block_group
= create_chunk(trans
, &ctl
, devices_info
);
5497 kfree(devices_info
);
5502 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5503 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5506 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5509 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle
*trans
,
5510 struct btrfs_block_group
*bg
)
5512 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
5513 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
5514 struct btrfs_key key
;
5515 struct btrfs_chunk
*chunk
;
5516 struct btrfs_stripe
*stripe
;
5517 struct extent_map
*em
;
5518 struct map_lookup
*map
;
5524 * We take the chunk_mutex for 2 reasons:
5526 * 1) Updates and insertions in the chunk btree must be done while holding
5527 * the chunk_mutex, as well as updating the system chunk array in the
5528 * superblock. See the comment on top of btrfs_chunk_alloc() for the
5531 * 2) To prevent races with the final phase of a device replace operation
5532 * that replaces the device object associated with the map's stripes,
5533 * because the device object's id can change at any time during that
5534 * final phase of the device replace operation
5535 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5536 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5537 * which would cause a failure when updating the device item, which does
5538 * not exists, or persisting a stripe of the chunk item with such ID.
5539 * Here we can't use the device_list_mutex because our caller already
5540 * has locked the chunk_mutex, and the final phase of device replace
5541 * acquires both mutexes - first the device_list_mutex and then the
5542 * chunk_mutex. Using any of those two mutexes protects us from a
5543 * concurrent device replace.
5545 lockdep_assert_held(&fs_info
->chunk_mutex
);
5547 em
= btrfs_get_chunk_map(fs_info
, bg
->start
, bg
->length
);
5550 btrfs_abort_transaction(trans
, ret
);
5554 map
= em
->map_lookup
;
5555 item_size
= btrfs_chunk_item_size(map
->num_stripes
);
5557 chunk
= kzalloc(item_size
, GFP_NOFS
);
5560 btrfs_abort_transaction(trans
, ret
);
5564 for (i
= 0; i
< map
->num_stripes
; i
++) {
5565 struct btrfs_device
*device
= map
->stripes
[i
].dev
;
5567 ret
= btrfs_update_device(trans
, device
);
5572 stripe
= &chunk
->stripe
;
5573 for (i
= 0; i
< map
->num_stripes
; i
++) {
5574 struct btrfs_device
*device
= map
->stripes
[i
].dev
;
5575 const u64 dev_offset
= map
->stripes
[i
].physical
;
5577 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
5578 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
5579 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
5583 btrfs_set_stack_chunk_length(chunk
, bg
->length
);
5584 btrfs_set_stack_chunk_owner(chunk
, BTRFS_EXTENT_TREE_OBJECTID
);
5585 btrfs_set_stack_chunk_stripe_len(chunk
, BTRFS_STRIPE_LEN
);
5586 btrfs_set_stack_chunk_type(chunk
, map
->type
);
5587 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
5588 btrfs_set_stack_chunk_io_align(chunk
, BTRFS_STRIPE_LEN
);
5589 btrfs_set_stack_chunk_io_width(chunk
, BTRFS_STRIPE_LEN
);
5590 btrfs_set_stack_chunk_sector_size(chunk
, fs_info
->sectorsize
);
5591 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
5593 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
5594 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
5595 key
.offset
= bg
->start
;
5597 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
5601 set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED
, &bg
->runtime_flags
);
5603 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
5604 ret
= btrfs_add_system_chunk(fs_info
, &key
, chunk
, item_size
);
5611 free_extent_map(em
);
5615 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
)
5617 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
5619 struct btrfs_block_group
*meta_bg
;
5620 struct btrfs_block_group
*sys_bg
;
5623 * When adding a new device for sprouting, the seed device is read-only
5624 * so we must first allocate a metadata and a system chunk. But before
5625 * adding the block group items to the extent, device and chunk btrees,
5628 * 1) Create both chunks without doing any changes to the btrees, as
5629 * otherwise we would get -ENOSPC since the block groups from the
5630 * seed device are read-only;
5632 * 2) Add the device item for the new sprout device - finishing the setup
5633 * of a new block group requires updating the device item in the chunk
5634 * btree, so it must exist when we attempt to do it. The previous step
5635 * ensures this does not fail with -ENOSPC.
5637 * After that we can add the block group items to their btrees:
5638 * update existing device item in the chunk btree, add a new block group
5639 * item to the extent btree, add a new chunk item to the chunk btree and
5640 * finally add the new device extent items to the devices btree.
5643 alloc_profile
= btrfs_metadata_alloc_profile(fs_info
);
5644 meta_bg
= btrfs_create_chunk(trans
, alloc_profile
);
5645 if (IS_ERR(meta_bg
))
5646 return PTR_ERR(meta_bg
);
5648 alloc_profile
= btrfs_system_alloc_profile(fs_info
);
5649 sys_bg
= btrfs_create_chunk(trans
, alloc_profile
);
5651 return PTR_ERR(sys_bg
);
5656 static inline int btrfs_chunk_max_errors(struct map_lookup
*map
)
5658 const int index
= btrfs_bg_flags_to_raid_index(map
->type
);
5660 return btrfs_raid_array
[index
].tolerated_failures
;
5663 bool btrfs_chunk_writeable(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
5665 struct extent_map
*em
;
5666 struct map_lookup
*map
;
5671 em
= btrfs_get_chunk_map(fs_info
, chunk_offset
, 1);
5675 map
= em
->map_lookup
;
5676 for (i
= 0; i
< map
->num_stripes
; i
++) {
5677 if (test_bit(BTRFS_DEV_STATE_MISSING
,
5678 &map
->stripes
[i
].dev
->dev_state
)) {
5682 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE
,
5683 &map
->stripes
[i
].dev
->dev_state
)) {
5690 * If the number of missing devices is larger than max errors, we can
5691 * not write the data into that chunk successfully.
5693 if (miss_ndevs
> btrfs_chunk_max_errors(map
))
5696 free_extent_map(em
);
5700 void btrfs_mapping_tree_free(struct extent_map_tree
*tree
)
5702 struct extent_map
*em
;
5705 write_lock(&tree
->lock
);
5706 em
= lookup_extent_mapping(tree
, 0, (u64
)-1);
5708 remove_extent_mapping(tree
, em
);
5709 write_unlock(&tree
->lock
);
5713 free_extent_map(em
);
5714 /* once for the tree */
5715 free_extent_map(em
);
5719 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
5721 struct extent_map
*em
;
5722 struct map_lookup
*map
;
5723 enum btrfs_raid_types index
;
5726 em
= btrfs_get_chunk_map(fs_info
, logical
, len
);
5729 * We could return errors for these cases, but that could get
5730 * ugly and we'd probably do the same thing which is just not do
5731 * anything else and exit, so return 1 so the callers don't try
5732 * to use other copies.
5736 map
= em
->map_lookup
;
5737 index
= btrfs_bg_flags_to_raid_index(map
->type
);
5739 /* Non-RAID56, use their ncopies from btrfs_raid_array. */
5740 if (!(map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
))
5741 ret
= btrfs_raid_array
[index
].ncopies
;
5742 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID5
)
5744 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5746 * There could be two corrupted data stripes, we need
5747 * to loop retry in order to rebuild the correct data.
5749 * Fail a stripe at a time on every retry except the
5750 * stripe under reconstruction.
5752 ret
= map
->num_stripes
;
5753 free_extent_map(em
);
5757 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info
*fs_info
,
5760 struct extent_map
*em
;
5761 struct map_lookup
*map
;
5762 unsigned long len
= fs_info
->sectorsize
;
5764 if (!btrfs_fs_incompat(fs_info
, RAID56
))
5767 em
= btrfs_get_chunk_map(fs_info
, logical
, len
);
5769 if (!WARN_ON(IS_ERR(em
))) {
5770 map
= em
->map_lookup
;
5771 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5772 len
= btrfs_stripe_nr_to_offset(nr_data_stripes(map
));
5773 free_extent_map(em
);
5778 int btrfs_is_parity_mirror(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
5780 struct extent_map
*em
;
5781 struct map_lookup
*map
;
5784 if (!btrfs_fs_incompat(fs_info
, RAID56
))
5787 em
= btrfs_get_chunk_map(fs_info
, logical
, len
);
5789 if(!WARN_ON(IS_ERR(em
))) {
5790 map
= em
->map_lookup
;
5791 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5793 free_extent_map(em
);
5798 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
5799 struct map_lookup
*map
, int first
,
5800 int dev_replace_is_ongoing
)
5804 int preferred_mirror
;
5806 struct btrfs_device
*srcdev
;
5809 (BTRFS_BLOCK_GROUP_RAID1_MASK
| BTRFS_BLOCK_GROUP_RAID10
)));
5811 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5812 num_stripes
= map
->sub_stripes
;
5814 num_stripes
= map
->num_stripes
;
5816 switch (fs_info
->fs_devices
->read_policy
) {
5818 /* Shouldn't happen, just warn and use pid instead of failing */
5819 btrfs_warn_rl(fs_info
,
5820 "unknown read_policy type %u, reset to pid",
5821 fs_info
->fs_devices
->read_policy
);
5822 fs_info
->fs_devices
->read_policy
= BTRFS_READ_POLICY_PID
;
5824 case BTRFS_READ_POLICY_PID
:
5825 preferred_mirror
= first
+ (current
->pid
% num_stripes
);
5829 if (dev_replace_is_ongoing
&&
5830 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
5831 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
5832 srcdev
= fs_info
->dev_replace
.srcdev
;
5837 * try to avoid the drive that is the source drive for a
5838 * dev-replace procedure, only choose it if no other non-missing
5839 * mirror is available
5841 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
5842 if (map
->stripes
[preferred_mirror
].dev
->bdev
&&
5843 (tolerance
|| map
->stripes
[preferred_mirror
].dev
!= srcdev
))
5844 return preferred_mirror
;
5845 for (i
= first
; i
< first
+ num_stripes
; i
++) {
5846 if (map
->stripes
[i
].dev
->bdev
&&
5847 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
5852 /* we couldn't find one that doesn't fail. Just return something
5853 * and the io error handling code will clean up eventually
5855 return preferred_mirror
;
5858 static struct btrfs_io_context
*alloc_btrfs_io_context(struct btrfs_fs_info
*fs_info
,
5862 struct btrfs_io_context
*bioc
;
5865 /* The size of btrfs_io_context */
5866 sizeof(struct btrfs_io_context
) +
5867 /* Plus the variable array for the stripes */
5868 sizeof(struct btrfs_io_stripe
) * (total_stripes
),
5874 refcount_set(&bioc
->refs
, 1);
5876 bioc
->fs_info
= fs_info
;
5877 bioc
->replace_stripe_src
= -1;
5878 bioc
->full_stripe_logical
= (u64
)-1;
5879 bioc
->logical
= logical
;
5884 void btrfs_get_bioc(struct btrfs_io_context
*bioc
)
5886 WARN_ON(!refcount_read(&bioc
->refs
));
5887 refcount_inc(&bioc
->refs
);
5890 void btrfs_put_bioc(struct btrfs_io_context
*bioc
)
5894 if (refcount_dec_and_test(&bioc
->refs
))
5899 * Please note that, discard won't be sent to target device of device
5902 struct btrfs_discard_stripe
*btrfs_map_discard(struct btrfs_fs_info
*fs_info
,
5903 u64 logical
, u64
*length_ret
,
5906 struct extent_map
*em
;
5907 struct map_lookup
*map
;
5908 struct btrfs_discard_stripe
*stripes
;
5909 u64 length
= *length_ret
;
5914 u64 stripe_end_offset
;
5918 u32 sub_stripes
= 0;
5919 u32 stripes_per_dev
= 0;
5920 u32 remaining_stripes
= 0;
5921 u32 last_stripe
= 0;
5925 em
= btrfs_get_chunk_map(fs_info
, logical
, length
);
5927 return ERR_CAST(em
);
5929 map
= em
->map_lookup
;
5931 /* we don't discard raid56 yet */
5932 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5937 offset
= logical
- em
->start
;
5938 length
= min_t(u64
, em
->start
+ em
->len
- logical
, length
);
5939 *length_ret
= length
;
5942 * stripe_nr counts the total number of stripes we have to stride
5943 * to get to this block
5945 stripe_nr
= offset
>> BTRFS_STRIPE_LEN_SHIFT
;
5947 /* stripe_offset is the offset of this block in its stripe */
5948 stripe_offset
= offset
- btrfs_stripe_nr_to_offset(stripe_nr
);
5950 stripe_nr_end
= round_up(offset
+ length
, BTRFS_STRIPE_LEN
) >>
5951 BTRFS_STRIPE_LEN_SHIFT
;
5952 stripe_cnt
= stripe_nr_end
- stripe_nr
;
5953 stripe_end_offset
= btrfs_stripe_nr_to_offset(stripe_nr_end
) -
5956 * after this, stripe_nr is the number of stripes on this
5957 * device we have to walk to find the data, and stripe_index is
5958 * the number of our device in the stripe array
5962 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5963 BTRFS_BLOCK_GROUP_RAID10
)) {
5964 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5967 sub_stripes
= map
->sub_stripes
;
5969 factor
= map
->num_stripes
/ sub_stripes
;
5970 *num_stripes
= min_t(u64
, map
->num_stripes
,
5971 sub_stripes
* stripe_cnt
);
5972 stripe_index
= stripe_nr
% factor
;
5973 stripe_nr
/= factor
;
5974 stripe_index
*= sub_stripes
;
5976 remaining_stripes
= stripe_cnt
% factor
;
5977 stripes_per_dev
= stripe_cnt
/ factor
;
5978 last_stripe
= ((stripe_nr_end
- 1) % factor
) * sub_stripes
;
5979 } else if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1_MASK
|
5980 BTRFS_BLOCK_GROUP_DUP
)) {
5981 *num_stripes
= map
->num_stripes
;
5983 stripe_index
= stripe_nr
% map
->num_stripes
;
5984 stripe_nr
/= map
->num_stripes
;
5987 stripes
= kcalloc(*num_stripes
, sizeof(*stripes
), GFP_NOFS
);
5993 for (i
= 0; i
< *num_stripes
; i
++) {
5994 stripes
[i
].physical
=
5995 map
->stripes
[stripe_index
].physical
+
5996 stripe_offset
+ btrfs_stripe_nr_to_offset(stripe_nr
);
5997 stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
5999 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
6000 BTRFS_BLOCK_GROUP_RAID10
)) {
6001 stripes
[i
].length
= btrfs_stripe_nr_to_offset(stripes_per_dev
);
6003 if (i
/ sub_stripes
< remaining_stripes
)
6004 stripes
[i
].length
+= BTRFS_STRIPE_LEN
;
6007 * Special for the first stripe and
6010 * |-------|...|-------|
6014 if (i
< sub_stripes
)
6015 stripes
[i
].length
-= stripe_offset
;
6017 if (stripe_index
>= last_stripe
&&
6018 stripe_index
<= (last_stripe
+
6020 stripes
[i
].length
-= stripe_end_offset
;
6022 if (i
== sub_stripes
- 1)
6025 stripes
[i
].length
= length
;
6029 if (stripe_index
== map
->num_stripes
) {
6035 free_extent_map(em
);
6038 free_extent_map(em
);
6039 return ERR_PTR(ret
);
6042 static bool is_block_group_to_copy(struct btrfs_fs_info
*fs_info
, u64 logical
)
6044 struct btrfs_block_group
*cache
;
6047 /* Non zoned filesystem does not use "to_copy" flag */
6048 if (!btrfs_is_zoned(fs_info
))
6051 cache
= btrfs_lookup_block_group(fs_info
, logical
);
6053 ret
= test_bit(BLOCK_GROUP_FLAG_TO_COPY
, &cache
->runtime_flags
);
6055 btrfs_put_block_group(cache
);
6059 static void handle_ops_on_dev_replace(enum btrfs_map_op op
,
6060 struct btrfs_io_context
*bioc
,
6061 struct btrfs_dev_replace
*dev_replace
,
6063 int *num_stripes_ret
, int *max_errors_ret
)
6065 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
6067 * At this stage, num_stripes is still the real number of stripes,
6068 * excluding the duplicated stripes.
6070 int num_stripes
= *num_stripes_ret
;
6071 int nr_extra_stripes
= 0;
6072 int max_errors
= *max_errors_ret
;
6076 * A block group which has "to_copy" set will eventually be copied by
6077 * the dev-replace process. We can avoid cloning IO here.
6079 if (is_block_group_to_copy(dev_replace
->srcdev
->fs_info
, logical
))
6083 * Duplicate the write operations while the dev-replace procedure is
6084 * running. Since the copying of the old disk to the new disk takes
6085 * place at run time while the filesystem is mounted writable, the
6086 * regular write operations to the old disk have to be duplicated to go
6087 * to the new disk as well.
6089 * Note that device->missing is handled by the caller, and that the
6090 * write to the old disk is already set up in the stripes array.
6092 for (i
= 0; i
< num_stripes
; i
++) {
6093 struct btrfs_io_stripe
*old
= &bioc
->stripes
[i
];
6094 struct btrfs_io_stripe
*new = &bioc
->stripes
[num_stripes
+ nr_extra_stripes
];
6096 if (old
->dev
->devid
!= srcdev_devid
)
6099 new->physical
= old
->physical
;
6100 new->dev
= dev_replace
->tgtdev
;
6101 if (bioc
->map_type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
6102 bioc
->replace_stripe_src
= i
;
6106 /* We can only have at most 2 extra nr_stripes (for DUP). */
6107 ASSERT(nr_extra_stripes
<= 2);
6109 * For GET_READ_MIRRORS, we can only return at most 1 extra stripe for
6111 * If we have 2 extra stripes, only choose the one with smaller physical.
6113 if (op
== BTRFS_MAP_GET_READ_MIRRORS
&& nr_extra_stripes
== 2) {
6114 struct btrfs_io_stripe
*first
= &bioc
->stripes
[num_stripes
];
6115 struct btrfs_io_stripe
*second
= &bioc
->stripes
[num_stripes
+ 1];
6117 /* Only DUP can have two extra stripes. */
6118 ASSERT(bioc
->map_type
& BTRFS_BLOCK_GROUP_DUP
);
6121 * Swap the last stripe stripes and reduce @nr_extra_stripes.
6122 * The extra stripe would still be there, but won't be accessed.
6124 if (first
->physical
> second
->physical
) {
6125 swap(second
->physical
, first
->physical
);
6126 swap(second
->dev
, first
->dev
);
6131 *num_stripes_ret
= num_stripes
+ nr_extra_stripes
;
6132 *max_errors_ret
= max_errors
+ nr_extra_stripes
;
6133 bioc
->replace_nr_stripes
= nr_extra_stripes
;
6136 static u64
btrfs_max_io_len(struct map_lookup
*map
, enum btrfs_map_op op
,
6137 u64 offset
, u32
*stripe_nr
, u64
*stripe_offset
,
6138 u64
*full_stripe_start
)
6141 * Stripe_nr is the stripe where this block falls. stripe_offset is
6142 * the offset of this block in its stripe.
6144 *stripe_offset
= offset
& BTRFS_STRIPE_LEN_MASK
;
6145 *stripe_nr
= offset
>> BTRFS_STRIPE_LEN_SHIFT
;
6146 ASSERT(*stripe_offset
< U32_MAX
);
6148 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
6149 unsigned long full_stripe_len
=
6150 btrfs_stripe_nr_to_offset(nr_data_stripes(map
));
6153 * For full stripe start, we use previously calculated
6154 * @stripe_nr. Align it to nr_data_stripes, then multiply with
6157 * By this we can avoid u64 division completely. And we have
6158 * to go rounddown(), not round_down(), as nr_data_stripes is
6159 * not ensured to be power of 2.
6161 *full_stripe_start
=
6162 btrfs_stripe_nr_to_offset(
6163 rounddown(*stripe_nr
, nr_data_stripes(map
)));
6165 ASSERT(*full_stripe_start
+ full_stripe_len
> offset
);
6166 ASSERT(*full_stripe_start
<= offset
);
6168 * For writes to RAID56, allow to write a full stripe set, but
6169 * no straddling of stripe sets.
6171 if (op
== BTRFS_MAP_WRITE
)
6172 return full_stripe_len
- (offset
- *full_stripe_start
);
6176 * For other RAID types and for RAID56 reads, allow a single stripe (on
6179 if (map
->type
& BTRFS_BLOCK_GROUP_STRIPE_MASK
)
6180 return BTRFS_STRIPE_LEN
- *stripe_offset
;
6184 static int set_io_stripe(struct btrfs_fs_info
*fs_info
, enum btrfs_map_op op
,
6185 u64 logical
, u64
*length
, struct btrfs_io_stripe
*dst
,
6186 struct map_lookup
*map
, u32 stripe_index
,
6187 u64 stripe_offset
, u64 stripe_nr
)
6189 dst
->dev
= map
->stripes
[stripe_index
].dev
;
6191 if (op
== BTRFS_MAP_READ
&& btrfs_need_stripe_tree_update(fs_info
, map
->type
))
6192 return btrfs_get_raid_extent_offset(fs_info
, logical
, length
,
6193 map
->type
, stripe_index
, dst
);
6195 dst
->physical
= map
->stripes
[stripe_index
].physical
+
6196 stripe_offset
+ btrfs_stripe_nr_to_offset(stripe_nr
);
6201 * Map one logical range to one or more physical ranges.
6203 * @length: (Mandatory) mapped length of this run.
6204 * One logical range can be split into different segments
6205 * due to factors like zones and RAID0/5/6/10 stripe
6208 * @bioc_ret: (Mandatory) returned btrfs_io_context structure.
6209 * which has one or more physical ranges (btrfs_io_stripe)
6211 * Caller should call btrfs_put_bioc() to free it after use.
6213 * @smap: (Optional) single physical range optimization.
6214 * If the map request can be fulfilled by one single
6215 * physical range, and this is parameter is not NULL,
6216 * then @bioc_ret would be NULL, and @smap would be
6219 * @mirror_num_ret: (Mandatory) returned mirror number if the original
6222 * Mirror number 0 means to choose any live mirrors.
6224 * For non-RAID56 profiles, non-zero mirror_num means
6225 * the Nth mirror. (e.g. mirror_num 1 means the first
6228 * For RAID56 profile, mirror 1 means rebuild from P and
6229 * the remaining data stripes.
6231 * For RAID6 profile, mirror > 2 means mark another
6232 * data/P stripe error and rebuild from the remaining
6235 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, enum btrfs_map_op op
,
6236 u64 logical
, u64
*length
,
6237 struct btrfs_io_context
**bioc_ret
,
6238 struct btrfs_io_stripe
*smap
, int *mirror_num_ret
)
6240 struct extent_map
*em
;
6241 struct map_lookup
*map
;
6249 int mirror_num
= (mirror_num_ret
? *mirror_num_ret
: 0);
6253 struct btrfs_io_context
*bioc
= NULL
;
6254 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
6255 int dev_replace_is_ongoing
= 0;
6256 u16 num_alloc_stripes
;
6257 u64 raid56_full_stripe_start
= (u64
)-1;
6262 num_copies
= btrfs_num_copies(fs_info
, logical
, fs_info
->sectorsize
);
6263 if (mirror_num
> num_copies
)
6266 em
= btrfs_get_chunk_map(fs_info
, logical
, *length
);
6270 map
= em
->map_lookup
;
6271 data_stripes
= nr_data_stripes(map
);
6273 map_offset
= logical
- em
->start
;
6274 max_len
= btrfs_max_io_len(map
, op
, map_offset
, &stripe_nr
,
6275 &stripe_offset
, &raid56_full_stripe_start
);
6276 *length
= min_t(u64
, em
->len
- map_offset
, max_len
);
6278 down_read(&dev_replace
->rwsem
);
6279 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
6281 * Hold the semaphore for read during the whole operation, write is
6282 * requested at commit time but must wait.
6284 if (!dev_replace_is_ongoing
)
6285 up_read(&dev_replace
->rwsem
);
6289 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
6290 stripe_index
= stripe_nr
% map
->num_stripes
;
6291 stripe_nr
/= map
->num_stripes
;
6292 if (op
== BTRFS_MAP_READ
)
6294 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1_MASK
) {
6295 if (op
!= BTRFS_MAP_READ
) {
6296 num_stripes
= map
->num_stripes
;
6297 } else if (mirror_num
) {
6298 stripe_index
= mirror_num
- 1;
6300 stripe_index
= find_live_mirror(fs_info
, map
, 0,
6301 dev_replace_is_ongoing
);
6302 mirror_num
= stripe_index
+ 1;
6305 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
6306 if (op
!= BTRFS_MAP_READ
) {
6307 num_stripes
= map
->num_stripes
;
6308 } else if (mirror_num
) {
6309 stripe_index
= mirror_num
- 1;
6314 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
6315 u32 factor
= map
->num_stripes
/ map
->sub_stripes
;
6317 stripe_index
= (stripe_nr
% factor
) * map
->sub_stripes
;
6318 stripe_nr
/= factor
;
6320 if (op
!= BTRFS_MAP_READ
)
6321 num_stripes
= map
->sub_stripes
;
6322 else if (mirror_num
)
6323 stripe_index
+= mirror_num
- 1;
6325 int old_stripe_index
= stripe_index
;
6326 stripe_index
= find_live_mirror(fs_info
, map
,
6328 dev_replace_is_ongoing
);
6329 mirror_num
= stripe_index
- old_stripe_index
+ 1;
6332 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
6333 if (op
!= BTRFS_MAP_READ
|| mirror_num
> 1) {
6335 * Needs full stripe mapping.
6337 * Push stripe_nr back to the start of the full stripe
6338 * For those cases needing a full stripe, @stripe_nr
6339 * is the full stripe number.
6341 * Originally we go raid56_full_stripe_start / full_stripe_len,
6342 * but that can be expensive. Here we just divide
6343 * @stripe_nr with @data_stripes.
6345 stripe_nr
/= data_stripes
;
6347 /* RAID[56] write or recovery. Return all stripes */
6348 num_stripes
= map
->num_stripes
;
6349 max_errors
= btrfs_chunk_max_errors(map
);
6351 /* Return the length to the full stripe end */
6352 *length
= min(logical
+ *length
,
6353 raid56_full_stripe_start
+ em
->start
+
6354 btrfs_stripe_nr_to_offset(data_stripes
)) -
6359 ASSERT(mirror_num
<= 1);
6360 /* Just grab the data stripe directly. */
6361 stripe_index
= stripe_nr
% data_stripes
;
6362 stripe_nr
/= data_stripes
;
6364 /* We distribute the parity blocks across stripes */
6365 stripe_index
= (stripe_nr
+ stripe_index
) % map
->num_stripes
;
6366 if (op
== BTRFS_MAP_READ
&& mirror_num
< 1)
6371 * After this, stripe_nr is the number of stripes on this
6372 * device we have to walk to find the data, and stripe_index is
6373 * the number of our device in the stripe array
6375 stripe_index
= stripe_nr
% map
->num_stripes
;
6376 stripe_nr
/= map
->num_stripes
;
6377 mirror_num
= stripe_index
+ 1;
6379 if (stripe_index
>= map
->num_stripes
) {
6381 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6382 stripe_index
, map
->num_stripes
);
6387 num_alloc_stripes
= num_stripes
;
6388 if (dev_replace_is_ongoing
&& dev_replace
->tgtdev
!= NULL
&&
6389 op
!= BTRFS_MAP_READ
)
6391 * For replace case, we need to add extra stripes for extra
6392 * duplicated stripes.
6394 * For both WRITE and GET_READ_MIRRORS, we may have at most
6395 * 2 more stripes (DUP types, otherwise 1).
6397 num_alloc_stripes
+= 2;
6400 * If this I/O maps to a single device, try to return the device and
6401 * physical block information on the stack instead of allocating an
6402 * I/O context structure.
6404 if (smap
&& num_alloc_stripes
== 1 &&
6405 !(btrfs_need_stripe_tree_update(fs_info
, map
->type
) &&
6406 op
!= BTRFS_MAP_READ
) &&
6407 !((map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) && mirror_num
> 1)) {
6408 ret
= set_io_stripe(fs_info
, op
, logical
, length
, smap
, map
,
6409 stripe_index
, stripe_offset
, stripe_nr
);
6411 *mirror_num_ret
= mirror_num
;
6416 bioc
= alloc_btrfs_io_context(fs_info
, logical
, num_alloc_stripes
);
6421 bioc
->map_type
= map
->type
;
6424 * For RAID56 full map, we need to make sure the stripes[] follows the
6425 * rule that data stripes are all ordered, then followed with P and Q
6428 * It's still mostly the same as other profiles, just with extra rotation.
6430 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
&&
6431 (op
!= BTRFS_MAP_READ
|| mirror_num
> 1)) {
6433 * For RAID56 @stripe_nr is already the number of full stripes
6434 * before us, which is also the rotation value (needs to modulo
6435 * with num_stripes).
6437 * In this case, we just add @stripe_nr with @i, then do the
6438 * modulo, to reduce one modulo call.
6440 bioc
->full_stripe_logical
= em
->start
+
6441 btrfs_stripe_nr_to_offset(stripe_nr
* data_stripes
);
6442 for (int i
= 0; i
< num_stripes
; i
++) {
6443 ret
= set_io_stripe(fs_info
, op
, logical
, length
,
6444 &bioc
->stripes
[i
], map
,
6445 (i
+ stripe_nr
) % num_stripes
,
6446 stripe_offset
, stripe_nr
);
6452 * For all other non-RAID56 profiles, just copy the target
6453 * stripe into the bioc.
6455 for (i
= 0; i
< num_stripes
; i
++) {
6456 ret
= set_io_stripe(fs_info
, op
, logical
, length
,
6457 &bioc
->stripes
[i
], map
, stripe_index
,
6458 stripe_offset
, stripe_nr
);
6467 btrfs_put_bioc(bioc
);
6471 if (op
!= BTRFS_MAP_READ
)
6472 max_errors
= btrfs_chunk_max_errors(map
);
6474 if (dev_replace_is_ongoing
&& dev_replace
->tgtdev
!= NULL
&&
6475 op
!= BTRFS_MAP_READ
) {
6476 handle_ops_on_dev_replace(op
, bioc
, dev_replace
, logical
,
6477 &num_stripes
, &max_errors
);
6481 bioc
->num_stripes
= num_stripes
;
6482 bioc
->max_errors
= max_errors
;
6483 bioc
->mirror_num
= mirror_num
;
6486 if (dev_replace_is_ongoing
) {
6487 lockdep_assert_held(&dev_replace
->rwsem
);
6488 /* Unlock and let waiting writers proceed */
6489 up_read(&dev_replace
->rwsem
);
6491 free_extent_map(em
);
6495 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args
*args
,
6496 const struct btrfs_fs_devices
*fs_devices
)
6498 if (args
->fsid
== NULL
)
6500 if (memcmp(fs_devices
->metadata_uuid
, args
->fsid
, BTRFS_FSID_SIZE
) == 0)
6505 static bool dev_args_match_device(const struct btrfs_dev_lookup_args
*args
,
6506 const struct btrfs_device
*device
)
6508 if (args
->missing
) {
6509 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &device
->dev_state
) &&
6515 if (device
->devid
!= args
->devid
)
6517 if (args
->uuid
&& memcmp(device
->uuid
, args
->uuid
, BTRFS_UUID_SIZE
) != 0)
6523 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6526 * If devid and uuid are both specified, the match must be exact, otherwise
6527 * only devid is used.
6529 struct btrfs_device
*btrfs_find_device(const struct btrfs_fs_devices
*fs_devices
,
6530 const struct btrfs_dev_lookup_args
*args
)
6532 struct btrfs_device
*device
;
6533 struct btrfs_fs_devices
*seed_devs
;
6535 if (dev_args_match_fs_devices(args
, fs_devices
)) {
6536 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6537 if (dev_args_match_device(args
, device
))
6542 list_for_each_entry(seed_devs
, &fs_devices
->seed_list
, seed_list
) {
6543 if (!dev_args_match_fs_devices(args
, seed_devs
))
6545 list_for_each_entry(device
, &seed_devs
->devices
, dev_list
) {
6546 if (dev_args_match_device(args
, device
))
6554 static struct btrfs_device
*add_missing_dev(struct btrfs_fs_devices
*fs_devices
,
6555 u64 devid
, u8
*dev_uuid
)
6557 struct btrfs_device
*device
;
6558 unsigned int nofs_flag
;
6561 * We call this under the chunk_mutex, so we want to use NOFS for this
6562 * allocation, however we don't want to change btrfs_alloc_device() to
6563 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6567 nofs_flag
= memalloc_nofs_save();
6568 device
= btrfs_alloc_device(NULL
, &devid
, dev_uuid
, NULL
);
6569 memalloc_nofs_restore(nofs_flag
);
6573 list_add(&device
->dev_list
, &fs_devices
->devices
);
6574 device
->fs_devices
= fs_devices
;
6575 fs_devices
->num_devices
++;
6577 set_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
);
6578 fs_devices
->missing_devices
++;
6584 * Allocate new device struct, set up devid and UUID.
6586 * @fs_info: used only for generating a new devid, can be NULL if
6587 * devid is provided (i.e. @devid != NULL).
6588 * @devid: a pointer to devid for this device. If NULL a new devid
6590 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6592 * @path: a pointer to device path if available, NULL otherwise.
6594 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6595 * on error. Returned struct is not linked onto any lists and must be
6596 * destroyed with btrfs_free_device.
6598 struct btrfs_device
*btrfs_alloc_device(struct btrfs_fs_info
*fs_info
,
6599 const u64
*devid
, const u8
*uuid
,
6602 struct btrfs_device
*dev
;
6605 if (WARN_ON(!devid
&& !fs_info
))
6606 return ERR_PTR(-EINVAL
);
6608 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
6610 return ERR_PTR(-ENOMEM
);
6612 INIT_LIST_HEAD(&dev
->dev_list
);
6613 INIT_LIST_HEAD(&dev
->dev_alloc_list
);
6614 INIT_LIST_HEAD(&dev
->post_commit_list
);
6616 atomic_set(&dev
->dev_stats_ccnt
, 0);
6617 btrfs_device_data_ordered_init(dev
);
6618 extent_io_tree_init(fs_info
, &dev
->alloc_state
, IO_TREE_DEVICE_ALLOC_STATE
);
6625 ret
= find_next_devid(fs_info
, &tmp
);
6627 btrfs_free_device(dev
);
6628 return ERR_PTR(ret
);
6634 memcpy(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
);
6636 generate_random_uuid(dev
->uuid
);
6639 struct rcu_string
*name
;
6641 name
= rcu_string_strdup(path
, GFP_KERNEL
);
6643 btrfs_free_device(dev
);
6644 return ERR_PTR(-ENOMEM
);
6646 rcu_assign_pointer(dev
->name
, name
);
6652 static void btrfs_report_missing_device(struct btrfs_fs_info
*fs_info
,
6653 u64 devid
, u8
*uuid
, bool error
)
6656 btrfs_err_rl(fs_info
, "devid %llu uuid %pU is missing",
6659 btrfs_warn_rl(fs_info
, "devid %llu uuid %pU is missing",
6663 u64
btrfs_calc_stripe_length(const struct extent_map
*em
)
6665 const struct map_lookup
*map
= em
->map_lookup
;
6666 const int data_stripes
= calc_data_stripes(map
->type
, map
->num_stripes
);
6668 return div_u64(em
->len
, data_stripes
);
6671 #if BITS_PER_LONG == 32
6673 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
6674 * can't be accessed on 32bit systems.
6676 * This function do mount time check to reject the fs if it already has
6677 * metadata chunk beyond that limit.
6679 static int check_32bit_meta_chunk(struct btrfs_fs_info
*fs_info
,
6680 u64 logical
, u64 length
, u64 type
)
6682 if (!(type
& BTRFS_BLOCK_GROUP_METADATA
))
6685 if (logical
+ length
< MAX_LFS_FILESIZE
)
6688 btrfs_err_32bit_limit(fs_info
);
6693 * This is to give early warning for any metadata chunk reaching
6694 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
6695 * Although we can still access the metadata, it's not going to be possible
6696 * once the limit is reached.
6698 static void warn_32bit_meta_chunk(struct btrfs_fs_info
*fs_info
,
6699 u64 logical
, u64 length
, u64 type
)
6701 if (!(type
& BTRFS_BLOCK_GROUP_METADATA
))
6704 if (logical
+ length
< BTRFS_32BIT_EARLY_WARN_THRESHOLD
)
6707 btrfs_warn_32bit_limit(fs_info
);
6711 static struct btrfs_device
*handle_missing_device(struct btrfs_fs_info
*fs_info
,
6712 u64 devid
, u8
*uuid
)
6714 struct btrfs_device
*dev
;
6716 if (!btrfs_test_opt(fs_info
, DEGRADED
)) {
6717 btrfs_report_missing_device(fs_info
, devid
, uuid
, true);
6718 return ERR_PTR(-ENOENT
);
6721 dev
= add_missing_dev(fs_info
->fs_devices
, devid
, uuid
);
6723 btrfs_err(fs_info
, "failed to init missing device %llu: %ld",
6724 devid
, PTR_ERR(dev
));
6727 btrfs_report_missing_device(fs_info
, devid
, uuid
, false);
6732 static int read_one_chunk(struct btrfs_key
*key
, struct extent_buffer
*leaf
,
6733 struct btrfs_chunk
*chunk
)
6735 BTRFS_DEV_LOOKUP_ARGS(args
);
6736 struct btrfs_fs_info
*fs_info
= leaf
->fs_info
;
6737 struct extent_map_tree
*map_tree
= &fs_info
->mapping_tree
;
6738 struct map_lookup
*map
;
6739 struct extent_map
*em
;
6744 u8 uuid
[BTRFS_UUID_SIZE
];
6750 logical
= key
->offset
;
6751 length
= btrfs_chunk_length(leaf
, chunk
);
6752 type
= btrfs_chunk_type(leaf
, chunk
);
6753 index
= btrfs_bg_flags_to_raid_index(type
);
6754 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6756 #if BITS_PER_LONG == 32
6757 ret
= check_32bit_meta_chunk(fs_info
, logical
, length
, type
);
6760 warn_32bit_meta_chunk(fs_info
, logical
, length
, type
);
6764 * Only need to verify chunk item if we're reading from sys chunk array,
6765 * as chunk item in tree block is already verified by tree-checker.
6767 if (leaf
->start
== BTRFS_SUPER_INFO_OFFSET
) {
6768 ret
= btrfs_check_chunk_valid(leaf
, chunk
, logical
);
6773 read_lock(&map_tree
->lock
);
6774 em
= lookup_extent_mapping(map_tree
, logical
, 1);
6775 read_unlock(&map_tree
->lock
);
6777 /* already mapped? */
6778 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
6779 free_extent_map(em
);
6782 free_extent_map(em
);
6785 em
= alloc_extent_map();
6788 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
6790 free_extent_map(em
);
6794 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
6795 em
->map_lookup
= map
;
6796 em
->start
= logical
;
6799 em
->block_start
= 0;
6800 em
->block_len
= em
->len
;
6802 map
->num_stripes
= num_stripes
;
6803 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
6804 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
6807 * We can't use the sub_stripes value, as for profiles other than
6808 * RAID10, they may have 0 as sub_stripes for filesystems created by
6809 * older mkfs (<v5.4).
6810 * In that case, it can cause divide-by-zero errors later.
6811 * Since currently sub_stripes is fixed for each profile, let's
6812 * use the trusted value instead.
6814 map
->sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
6815 map
->verified_stripes
= 0;
6816 em
->orig_block_len
= btrfs_calc_stripe_length(em
);
6817 for (i
= 0; i
< num_stripes
; i
++) {
6818 map
->stripes
[i
].physical
=
6819 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
6820 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
6822 read_extent_buffer(leaf
, uuid
, (unsigned long)
6823 btrfs_stripe_dev_uuid_nr(chunk
, i
),
6826 map
->stripes
[i
].dev
= btrfs_find_device(fs_info
->fs_devices
, &args
);
6827 if (!map
->stripes
[i
].dev
) {
6828 map
->stripes
[i
].dev
= handle_missing_device(fs_info
,
6830 if (IS_ERR(map
->stripes
[i
].dev
)) {
6831 ret
= PTR_ERR(map
->stripes
[i
].dev
);
6832 free_extent_map(em
);
6837 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA
,
6838 &(map
->stripes
[i
].dev
->dev_state
));
6841 write_lock(&map_tree
->lock
);
6842 ret
= add_extent_mapping(map_tree
, em
, 0);
6843 write_unlock(&map_tree
->lock
);
6846 "failed to add chunk map, start=%llu len=%llu: %d",
6847 em
->start
, em
->len
, ret
);
6849 free_extent_map(em
);
6854 static void fill_device_from_item(struct extent_buffer
*leaf
,
6855 struct btrfs_dev_item
*dev_item
,
6856 struct btrfs_device
*device
)
6860 device
->devid
= btrfs_device_id(leaf
, dev_item
);
6861 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
6862 device
->total_bytes
= device
->disk_total_bytes
;
6863 device
->commit_total_bytes
= device
->disk_total_bytes
;
6864 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
6865 device
->commit_bytes_used
= device
->bytes_used
;
6866 device
->type
= btrfs_device_type(leaf
, dev_item
);
6867 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
6868 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
6869 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
6870 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
6871 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
);
6873 ptr
= btrfs_device_uuid(dev_item
);
6874 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
6877 static struct btrfs_fs_devices
*open_seed_devices(struct btrfs_fs_info
*fs_info
,
6880 struct btrfs_fs_devices
*fs_devices
;
6883 lockdep_assert_held(&uuid_mutex
);
6886 /* This will match only for multi-device seed fs */
6887 list_for_each_entry(fs_devices
, &fs_info
->fs_devices
->seed_list
, seed_list
)
6888 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_FSID_SIZE
))
6892 fs_devices
= find_fsid(fsid
, NULL
);
6894 if (!btrfs_test_opt(fs_info
, DEGRADED
))
6895 return ERR_PTR(-ENOENT
);
6897 fs_devices
= alloc_fs_devices(fsid
);
6898 if (IS_ERR(fs_devices
))
6901 fs_devices
->seeding
= true;
6902 fs_devices
->opened
= 1;
6907 * Upon first call for a seed fs fsid, just create a private copy of the
6908 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
6910 fs_devices
= clone_fs_devices(fs_devices
);
6911 if (IS_ERR(fs_devices
))
6914 ret
= open_fs_devices(fs_devices
, BLK_OPEN_READ
, fs_info
->bdev_holder
);
6916 free_fs_devices(fs_devices
);
6917 return ERR_PTR(ret
);
6920 if (!fs_devices
->seeding
) {
6921 close_fs_devices(fs_devices
);
6922 free_fs_devices(fs_devices
);
6923 return ERR_PTR(-EINVAL
);
6926 list_add(&fs_devices
->seed_list
, &fs_info
->fs_devices
->seed_list
);
6931 static int read_one_dev(struct extent_buffer
*leaf
,
6932 struct btrfs_dev_item
*dev_item
)
6934 BTRFS_DEV_LOOKUP_ARGS(args
);
6935 struct btrfs_fs_info
*fs_info
= leaf
->fs_info
;
6936 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
6937 struct btrfs_device
*device
;
6940 u8 fs_uuid
[BTRFS_FSID_SIZE
];
6941 u8 dev_uuid
[BTRFS_UUID_SIZE
];
6943 devid
= btrfs_device_id(leaf
, dev_item
);
6945 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
6947 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
6949 args
.uuid
= dev_uuid
;
6950 args
.fsid
= fs_uuid
;
6952 if (memcmp(fs_uuid
, fs_devices
->metadata_uuid
, BTRFS_FSID_SIZE
)) {
6953 fs_devices
= open_seed_devices(fs_info
, fs_uuid
);
6954 if (IS_ERR(fs_devices
))
6955 return PTR_ERR(fs_devices
);
6958 device
= btrfs_find_device(fs_info
->fs_devices
, &args
);
6960 if (!btrfs_test_opt(fs_info
, DEGRADED
)) {
6961 btrfs_report_missing_device(fs_info
, devid
,
6966 device
= add_missing_dev(fs_devices
, devid
, dev_uuid
);
6967 if (IS_ERR(device
)) {
6969 "failed to add missing dev %llu: %ld",
6970 devid
, PTR_ERR(device
));
6971 return PTR_ERR(device
);
6973 btrfs_report_missing_device(fs_info
, devid
, dev_uuid
, false);
6975 if (!device
->bdev
) {
6976 if (!btrfs_test_opt(fs_info
, DEGRADED
)) {
6977 btrfs_report_missing_device(fs_info
,
6978 devid
, dev_uuid
, true);
6981 btrfs_report_missing_device(fs_info
, devid
,
6985 if (!device
->bdev
&&
6986 !test_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
)) {
6988 * this happens when a device that was properly setup
6989 * in the device info lists suddenly goes bad.
6990 * device->bdev is NULL, and so we have to set
6991 * device->missing to one here
6993 device
->fs_devices
->missing_devices
++;
6994 set_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
);
6997 /* Move the device to its own fs_devices */
6998 if (device
->fs_devices
!= fs_devices
) {
6999 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING
,
7000 &device
->dev_state
));
7002 list_move(&device
->dev_list
, &fs_devices
->devices
);
7003 device
->fs_devices
->num_devices
--;
7004 fs_devices
->num_devices
++;
7006 device
->fs_devices
->missing_devices
--;
7007 fs_devices
->missing_devices
++;
7009 device
->fs_devices
= fs_devices
;
7013 if (device
->fs_devices
!= fs_info
->fs_devices
) {
7014 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
));
7015 if (device
->generation
!=
7016 btrfs_device_generation(leaf
, dev_item
))
7020 fill_device_from_item(leaf
, dev_item
, device
);
7022 u64 max_total_bytes
= bdev_nr_bytes(device
->bdev
);
7024 if (device
->total_bytes
> max_total_bytes
) {
7026 "device total_bytes should be at most %llu but found %llu",
7027 max_total_bytes
, device
->total_bytes
);
7031 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &device
->dev_state
);
7032 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
) &&
7033 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
)) {
7034 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
7035 atomic64_add(device
->total_bytes
- device
->bytes_used
,
7036 &fs_info
->free_chunk_space
);
7042 int btrfs_read_sys_array(struct btrfs_fs_info
*fs_info
)
7044 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
7045 struct extent_buffer
*sb
;
7046 struct btrfs_disk_key
*disk_key
;
7047 struct btrfs_chunk
*chunk
;
7049 unsigned long sb_array_offset
;
7056 struct btrfs_key key
;
7058 ASSERT(BTRFS_SUPER_INFO_SIZE
<= fs_info
->nodesize
);
7061 * We allocated a dummy extent, just to use extent buffer accessors.
7062 * There will be unused space after BTRFS_SUPER_INFO_SIZE, but
7063 * that's fine, we will not go beyond system chunk array anyway.
7065 sb
= alloc_dummy_extent_buffer(fs_info
, BTRFS_SUPER_INFO_OFFSET
);
7068 set_extent_buffer_uptodate(sb
);
7070 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
7071 array_size
= btrfs_super_sys_array_size(super_copy
);
7073 array_ptr
= super_copy
->sys_chunk_array
;
7074 sb_array_offset
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
7077 while (cur_offset
< array_size
) {
7078 disk_key
= (struct btrfs_disk_key
*)array_ptr
;
7079 len
= sizeof(*disk_key
);
7080 if (cur_offset
+ len
> array_size
)
7081 goto out_short_read
;
7083 btrfs_disk_key_to_cpu(&key
, disk_key
);
7086 sb_array_offset
+= len
;
7089 if (key
.type
!= BTRFS_CHUNK_ITEM_KEY
) {
7091 "unexpected item type %u in sys_array at offset %u",
7092 (u32
)key
.type
, cur_offset
);
7097 chunk
= (struct btrfs_chunk
*)sb_array_offset
;
7099 * At least one btrfs_chunk with one stripe must be present,
7100 * exact stripe count check comes afterwards
7102 len
= btrfs_chunk_item_size(1);
7103 if (cur_offset
+ len
> array_size
)
7104 goto out_short_read
;
7106 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
7109 "invalid number of stripes %u in sys_array at offset %u",
7110 num_stripes
, cur_offset
);
7115 type
= btrfs_chunk_type(sb
, chunk
);
7116 if ((type
& BTRFS_BLOCK_GROUP_SYSTEM
) == 0) {
7118 "invalid chunk type %llu in sys_array at offset %u",
7124 len
= btrfs_chunk_item_size(num_stripes
);
7125 if (cur_offset
+ len
> array_size
)
7126 goto out_short_read
;
7128 ret
= read_one_chunk(&key
, sb
, chunk
);
7133 sb_array_offset
+= len
;
7136 clear_extent_buffer_uptodate(sb
);
7137 free_extent_buffer_stale(sb
);
7141 btrfs_err(fs_info
, "sys_array too short to read %u bytes at offset %u",
7143 clear_extent_buffer_uptodate(sb
);
7144 free_extent_buffer_stale(sb
);
7149 * Check if all chunks in the fs are OK for read-write degraded mount
7151 * If the @failing_dev is specified, it's accounted as missing.
7153 * Return true if all chunks meet the minimal RW mount requirements.
7154 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7156 bool btrfs_check_rw_degradable(struct btrfs_fs_info
*fs_info
,
7157 struct btrfs_device
*failing_dev
)
7159 struct extent_map_tree
*map_tree
= &fs_info
->mapping_tree
;
7160 struct extent_map
*em
;
7164 read_lock(&map_tree
->lock
);
7165 em
= lookup_extent_mapping(map_tree
, 0, (u64
)-1);
7166 read_unlock(&map_tree
->lock
);
7167 /* No chunk at all? Return false anyway */
7173 struct map_lookup
*map
;
7178 map
= em
->map_lookup
;
7180 btrfs_get_num_tolerated_disk_barrier_failures(
7182 for (i
= 0; i
< map
->num_stripes
; i
++) {
7183 struct btrfs_device
*dev
= map
->stripes
[i
].dev
;
7185 if (!dev
|| !dev
->bdev
||
7186 test_bit(BTRFS_DEV_STATE_MISSING
, &dev
->dev_state
) ||
7187 dev
->last_flush_error
)
7189 else if (failing_dev
&& failing_dev
== dev
)
7192 if (missing
> max_tolerated
) {
7195 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7196 em
->start
, missing
, max_tolerated
);
7197 free_extent_map(em
);
7201 next_start
= extent_map_end(em
);
7202 free_extent_map(em
);
7204 read_lock(&map_tree
->lock
);
7205 em
= lookup_extent_mapping(map_tree
, next_start
,
7206 (u64
)(-1) - next_start
);
7207 read_unlock(&map_tree
->lock
);
7213 static void readahead_tree_node_children(struct extent_buffer
*node
)
7216 const int nr_items
= btrfs_header_nritems(node
);
7218 for (i
= 0; i
< nr_items
; i
++)
7219 btrfs_readahead_node_child(node
, i
);
7222 int btrfs_read_chunk_tree(struct btrfs_fs_info
*fs_info
)
7224 struct btrfs_root
*root
= fs_info
->chunk_root
;
7225 struct btrfs_path
*path
;
7226 struct extent_buffer
*leaf
;
7227 struct btrfs_key key
;
7228 struct btrfs_key found_key
;
7233 u64 last_ra_node
= 0;
7235 path
= btrfs_alloc_path();
7240 * uuid_mutex is needed only if we are mounting a sprout FS
7241 * otherwise we don't need it.
7243 mutex_lock(&uuid_mutex
);
7246 * It is possible for mount and umount to race in such a way that
7247 * we execute this code path, but open_fs_devices failed to clear
7248 * total_rw_bytes. We certainly want it cleared before reading the
7249 * device items, so clear it here.
7251 fs_info
->fs_devices
->total_rw_bytes
= 0;
7254 * Lockdep complains about possible circular locking dependency between
7255 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores
7256 * used for freeze procection of a fs (struct super_block.s_writers),
7257 * which we take when starting a transaction, and extent buffers of the
7258 * chunk tree if we call read_one_dev() while holding a lock on an
7259 * extent buffer of the chunk tree. Since we are mounting the filesystem
7260 * and at this point there can't be any concurrent task modifying the
7261 * chunk tree, to keep it simple, just skip locking on the chunk tree.
7263 ASSERT(!test_bit(BTRFS_FS_OPEN
, &fs_info
->flags
));
7264 path
->skip_locking
= 1;
7267 * Read all device items, and then all the chunk items. All
7268 * device items are found before any chunk item (their object id
7269 * is smaller than the lowest possible object id for a chunk
7270 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7272 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
7275 btrfs_for_each_slot(root
, &key
, &found_key
, path
, iter_ret
) {
7276 struct extent_buffer
*node
= path
->nodes
[1];
7278 leaf
= path
->nodes
[0];
7279 slot
= path
->slots
[0];
7282 if (last_ra_node
!= node
->start
) {
7283 readahead_tree_node_children(node
);
7284 last_ra_node
= node
->start
;
7287 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
7288 struct btrfs_dev_item
*dev_item
;
7289 dev_item
= btrfs_item_ptr(leaf
, slot
,
7290 struct btrfs_dev_item
);
7291 ret
= read_one_dev(leaf
, dev_item
);
7295 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
7296 struct btrfs_chunk
*chunk
;
7299 * We are only called at mount time, so no need to take
7300 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7301 * we always lock first fs_info->chunk_mutex before
7302 * acquiring any locks on the chunk tree. This is a
7303 * requirement for chunk allocation, see the comment on
7304 * top of btrfs_chunk_alloc() for details.
7306 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
7307 ret
= read_one_chunk(&found_key
, leaf
, chunk
);
7312 /* Catch error found during iteration */
7319 * After loading chunk tree, we've got all device information,
7320 * do another round of validation checks.
7322 if (total_dev
!= fs_info
->fs_devices
->total_devices
) {
7324 "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit",
7325 btrfs_super_num_devices(fs_info
->super_copy
),
7327 fs_info
->fs_devices
->total_devices
= total_dev
;
7328 btrfs_set_super_num_devices(fs_info
->super_copy
, total_dev
);
7330 if (btrfs_super_total_bytes(fs_info
->super_copy
) <
7331 fs_info
->fs_devices
->total_rw_bytes
) {
7333 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7334 btrfs_super_total_bytes(fs_info
->super_copy
),
7335 fs_info
->fs_devices
->total_rw_bytes
);
7341 mutex_unlock(&uuid_mutex
);
7343 btrfs_free_path(path
);
7347 int btrfs_init_devices_late(struct btrfs_fs_info
*fs_info
)
7349 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
, *seed_devs
;
7350 struct btrfs_device
*device
;
7353 fs_devices
->fs_info
= fs_info
;
7355 mutex_lock(&fs_devices
->device_list_mutex
);
7356 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
)
7357 device
->fs_info
= fs_info
;
7359 list_for_each_entry(seed_devs
, &fs_devices
->seed_list
, seed_list
) {
7360 list_for_each_entry(device
, &seed_devs
->devices
, dev_list
) {
7361 device
->fs_info
= fs_info
;
7362 ret
= btrfs_get_dev_zone_info(device
, false);
7367 seed_devs
->fs_info
= fs_info
;
7369 mutex_unlock(&fs_devices
->device_list_mutex
);
7374 static u64
btrfs_dev_stats_value(const struct extent_buffer
*eb
,
7375 const struct btrfs_dev_stats_item
*ptr
,
7380 read_extent_buffer(eb
, &val
,
7381 offsetof(struct btrfs_dev_stats_item
, values
) +
7382 ((unsigned long)ptr
) + (index
* sizeof(u64
)),
7387 static void btrfs_set_dev_stats_value(struct extent_buffer
*eb
,
7388 struct btrfs_dev_stats_item
*ptr
,
7391 write_extent_buffer(eb
, &val
,
7392 offsetof(struct btrfs_dev_stats_item
, values
) +
7393 ((unsigned long)ptr
) + (index
* sizeof(u64
)),
7397 static int btrfs_device_init_dev_stats(struct btrfs_device
*device
,
7398 struct btrfs_path
*path
)
7400 struct btrfs_dev_stats_item
*ptr
;
7401 struct extent_buffer
*eb
;
7402 struct btrfs_key key
;
7406 if (!device
->fs_info
->dev_root
)
7409 key
.objectid
= BTRFS_DEV_STATS_OBJECTID
;
7410 key
.type
= BTRFS_PERSISTENT_ITEM_KEY
;
7411 key
.offset
= device
->devid
;
7412 ret
= btrfs_search_slot(NULL
, device
->fs_info
->dev_root
, &key
, path
, 0, 0);
7414 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7415 btrfs_dev_stat_set(device
, i
, 0);
7416 device
->dev_stats_valid
= 1;
7417 btrfs_release_path(path
);
7418 return ret
< 0 ? ret
: 0;
7420 slot
= path
->slots
[0];
7421 eb
= path
->nodes
[0];
7422 item_size
= btrfs_item_size(eb
, slot
);
7424 ptr
= btrfs_item_ptr(eb
, slot
, struct btrfs_dev_stats_item
);
7426 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
7427 if (item_size
>= (1 + i
) * sizeof(__le64
))
7428 btrfs_dev_stat_set(device
, i
,
7429 btrfs_dev_stats_value(eb
, ptr
, i
));
7431 btrfs_dev_stat_set(device
, i
, 0);
7434 device
->dev_stats_valid
= 1;
7435 btrfs_dev_stat_print_on_load(device
);
7436 btrfs_release_path(path
);
7441 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
7443 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
, *seed_devs
;
7444 struct btrfs_device
*device
;
7445 struct btrfs_path
*path
= NULL
;
7448 path
= btrfs_alloc_path();
7452 mutex_lock(&fs_devices
->device_list_mutex
);
7453 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
7454 ret
= btrfs_device_init_dev_stats(device
, path
);
7458 list_for_each_entry(seed_devs
, &fs_devices
->seed_list
, seed_list
) {
7459 list_for_each_entry(device
, &seed_devs
->devices
, dev_list
) {
7460 ret
= btrfs_device_init_dev_stats(device
, path
);
7466 mutex_unlock(&fs_devices
->device_list_mutex
);
7468 btrfs_free_path(path
);
7472 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
7473 struct btrfs_device
*device
)
7475 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
7476 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
7477 struct btrfs_path
*path
;
7478 struct btrfs_key key
;
7479 struct extent_buffer
*eb
;
7480 struct btrfs_dev_stats_item
*ptr
;
7484 key
.objectid
= BTRFS_DEV_STATS_OBJECTID
;
7485 key
.type
= BTRFS_PERSISTENT_ITEM_KEY
;
7486 key
.offset
= device
->devid
;
7488 path
= btrfs_alloc_path();
7491 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
7493 btrfs_warn_in_rcu(fs_info
,
7494 "error %d while searching for dev_stats item for device %s",
7495 ret
, btrfs_dev_name(device
));
7500 btrfs_item_size(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
7501 /* need to delete old one and insert a new one */
7502 ret
= btrfs_del_item(trans
, dev_root
, path
);
7504 btrfs_warn_in_rcu(fs_info
,
7505 "delete too small dev_stats item for device %s failed %d",
7506 btrfs_dev_name(device
), ret
);
7513 /* need to insert a new item */
7514 btrfs_release_path(path
);
7515 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
7516 &key
, sizeof(*ptr
));
7518 btrfs_warn_in_rcu(fs_info
,
7519 "insert dev_stats item for device %s failed %d",
7520 btrfs_dev_name(device
), ret
);
7525 eb
= path
->nodes
[0];
7526 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
7527 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7528 btrfs_set_dev_stats_value(eb
, ptr
, i
,
7529 btrfs_dev_stat_read(device
, i
));
7530 btrfs_mark_buffer_dirty(trans
, eb
);
7533 btrfs_free_path(path
);
7538 * called from commit_transaction. Writes all changed device stats to disk.
7540 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
)
7542 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
7543 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7544 struct btrfs_device
*device
;
7548 mutex_lock(&fs_devices
->device_list_mutex
);
7549 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
7550 stats_cnt
= atomic_read(&device
->dev_stats_ccnt
);
7551 if (!device
->dev_stats_valid
|| stats_cnt
== 0)
7556 * There is a LOAD-LOAD control dependency between the value of
7557 * dev_stats_ccnt and updating the on-disk values which requires
7558 * reading the in-memory counters. Such control dependencies
7559 * require explicit read memory barriers.
7561 * This memory barriers pairs with smp_mb__before_atomic in
7562 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7563 * barrier implied by atomic_xchg in
7564 * btrfs_dev_stats_read_and_reset
7568 ret
= update_dev_stat_item(trans
, device
);
7570 atomic_sub(stats_cnt
, &device
->dev_stats_ccnt
);
7572 mutex_unlock(&fs_devices
->device_list_mutex
);
7577 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
7579 btrfs_dev_stat_inc(dev
, index
);
7581 if (!dev
->dev_stats_valid
)
7583 btrfs_err_rl_in_rcu(dev
->fs_info
,
7584 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7585 btrfs_dev_name(dev
),
7586 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
7587 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
7588 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
7589 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
7590 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
7593 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
7597 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7598 if (btrfs_dev_stat_read(dev
, i
) != 0)
7600 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
7601 return; /* all values == 0, suppress message */
7603 btrfs_info_in_rcu(dev
->fs_info
,
7604 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7605 btrfs_dev_name(dev
),
7606 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
7607 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
7608 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
7609 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
7610 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
7613 int btrfs_get_dev_stats(struct btrfs_fs_info
*fs_info
,
7614 struct btrfs_ioctl_get_dev_stats
*stats
)
7616 BTRFS_DEV_LOOKUP_ARGS(args
);
7617 struct btrfs_device
*dev
;
7618 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7621 mutex_lock(&fs_devices
->device_list_mutex
);
7622 args
.devid
= stats
->devid
;
7623 dev
= btrfs_find_device(fs_info
->fs_devices
, &args
);
7624 mutex_unlock(&fs_devices
->device_list_mutex
);
7627 btrfs_warn(fs_info
, "get dev_stats failed, device not found");
7629 } else if (!dev
->dev_stats_valid
) {
7630 btrfs_warn(fs_info
, "get dev_stats failed, not yet valid");
7632 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
7633 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
7634 if (stats
->nr_items
> i
)
7636 btrfs_dev_stat_read_and_reset(dev
, i
);
7638 btrfs_dev_stat_set(dev
, i
, 0);
7640 btrfs_info(fs_info
, "device stats zeroed by %s (%d)",
7641 current
->comm
, task_pid_nr(current
));
7643 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7644 if (stats
->nr_items
> i
)
7645 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
7647 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
7648 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
7653 * Update the size and bytes used for each device where it changed. This is
7654 * delayed since we would otherwise get errors while writing out the
7657 * Must be invoked during transaction commit.
7659 void btrfs_commit_device_sizes(struct btrfs_transaction
*trans
)
7661 struct btrfs_device
*curr
, *next
;
7663 ASSERT(trans
->state
== TRANS_STATE_COMMIT_DOING
);
7665 if (list_empty(&trans
->dev_update_list
))
7669 * We don't need the device_list_mutex here. This list is owned by the
7670 * transaction and the transaction must complete before the device is
7673 mutex_lock(&trans
->fs_info
->chunk_mutex
);
7674 list_for_each_entry_safe(curr
, next
, &trans
->dev_update_list
,
7676 list_del_init(&curr
->post_commit_list
);
7677 curr
->commit_total_bytes
= curr
->disk_total_bytes
;
7678 curr
->commit_bytes_used
= curr
->bytes_used
;
7680 mutex_unlock(&trans
->fs_info
->chunk_mutex
);
7684 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7686 int btrfs_bg_type_to_factor(u64 flags
)
7688 const int index
= btrfs_bg_flags_to_raid_index(flags
);
7690 return btrfs_raid_array
[index
].ncopies
;
7695 static int verify_one_dev_extent(struct btrfs_fs_info
*fs_info
,
7696 u64 chunk_offset
, u64 devid
,
7697 u64 physical_offset
, u64 physical_len
)
7699 struct btrfs_dev_lookup_args args
= { .devid
= devid
};
7700 struct extent_map_tree
*em_tree
= &fs_info
->mapping_tree
;
7701 struct extent_map
*em
;
7702 struct map_lookup
*map
;
7703 struct btrfs_device
*dev
;
7709 read_lock(&em_tree
->lock
);
7710 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
7711 read_unlock(&em_tree
->lock
);
7715 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7716 physical_offset
, devid
);
7721 map
= em
->map_lookup
;
7722 stripe_len
= btrfs_calc_stripe_length(em
);
7723 if (physical_len
!= stripe_len
) {
7725 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7726 physical_offset
, devid
, em
->start
, physical_len
,
7733 * Very old mkfs.btrfs (before v4.1) will not respect the reserved
7734 * space. Although kernel can handle it without problem, better to warn
7737 if (physical_offset
< BTRFS_DEVICE_RANGE_RESERVED
)
7739 "devid %llu physical %llu len %llu inside the reserved space",
7740 devid
, physical_offset
, physical_len
);
7742 for (i
= 0; i
< map
->num_stripes
; i
++) {
7743 if (map
->stripes
[i
].dev
->devid
== devid
&&
7744 map
->stripes
[i
].physical
== physical_offset
) {
7746 if (map
->verified_stripes
>= map
->num_stripes
) {
7748 "too many dev extents for chunk %llu found",
7753 map
->verified_stripes
++;
7759 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7760 physical_offset
, devid
);
7764 /* Make sure no dev extent is beyond device boundary */
7765 dev
= btrfs_find_device(fs_info
->fs_devices
, &args
);
7767 btrfs_err(fs_info
, "failed to find devid %llu", devid
);
7772 if (physical_offset
+ physical_len
> dev
->disk_total_bytes
) {
7774 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7775 devid
, physical_offset
, physical_len
,
7776 dev
->disk_total_bytes
);
7781 if (dev
->zone_info
) {
7782 u64 zone_size
= dev
->zone_info
->zone_size
;
7784 if (!IS_ALIGNED(physical_offset
, zone_size
) ||
7785 !IS_ALIGNED(physical_len
, zone_size
)) {
7787 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
7788 devid
, physical_offset
, physical_len
);
7795 free_extent_map(em
);
7799 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info
*fs_info
)
7801 struct extent_map_tree
*em_tree
= &fs_info
->mapping_tree
;
7802 struct extent_map
*em
;
7803 struct rb_node
*node
;
7806 read_lock(&em_tree
->lock
);
7807 for (node
= rb_first_cached(&em_tree
->map
); node
; node
= rb_next(node
)) {
7808 em
= rb_entry(node
, struct extent_map
, rb_node
);
7809 if (em
->map_lookup
->num_stripes
!=
7810 em
->map_lookup
->verified_stripes
) {
7812 "chunk %llu has missing dev extent, have %d expect %d",
7813 em
->start
, em
->map_lookup
->verified_stripes
,
7814 em
->map_lookup
->num_stripes
);
7820 read_unlock(&em_tree
->lock
);
7825 * Ensure that all dev extents are mapped to correct chunk, otherwise
7826 * later chunk allocation/free would cause unexpected behavior.
7828 * NOTE: This will iterate through the whole device tree, which should be of
7829 * the same size level as the chunk tree. This slightly increases mount time.
7831 int btrfs_verify_dev_extents(struct btrfs_fs_info
*fs_info
)
7833 struct btrfs_path
*path
;
7834 struct btrfs_root
*root
= fs_info
->dev_root
;
7835 struct btrfs_key key
;
7837 u64 prev_dev_ext_end
= 0;
7841 * We don't have a dev_root because we mounted with ignorebadroots and
7842 * failed to load the root, so we want to skip the verification in this
7845 * However if the dev root is fine, but the tree itself is corrupted
7846 * we'd still fail to mount. This verification is only to make sure
7847 * writes can happen safely, so instead just bypass this check
7848 * completely in the case of IGNOREBADROOTS.
7850 if (btrfs_test_opt(fs_info
, IGNOREBADROOTS
))
7854 key
.type
= BTRFS_DEV_EXTENT_KEY
;
7857 path
= btrfs_alloc_path();
7861 path
->reada
= READA_FORWARD
;
7862 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
7866 if (path
->slots
[0] >= btrfs_header_nritems(path
->nodes
[0])) {
7867 ret
= btrfs_next_leaf(root
, path
);
7870 /* No dev extents at all? Not good */
7877 struct extent_buffer
*leaf
= path
->nodes
[0];
7878 struct btrfs_dev_extent
*dext
;
7879 int slot
= path
->slots
[0];
7881 u64 physical_offset
;
7885 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
7886 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
7888 devid
= key
.objectid
;
7889 physical_offset
= key
.offset
;
7891 dext
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dev_extent
);
7892 chunk_offset
= btrfs_dev_extent_chunk_offset(leaf
, dext
);
7893 physical_len
= btrfs_dev_extent_length(leaf
, dext
);
7895 /* Check if this dev extent overlaps with the previous one */
7896 if (devid
== prev_devid
&& physical_offset
< prev_dev_ext_end
) {
7898 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
7899 devid
, physical_offset
, prev_dev_ext_end
);
7904 ret
= verify_one_dev_extent(fs_info
, chunk_offset
, devid
,
7905 physical_offset
, physical_len
);
7909 prev_dev_ext_end
= physical_offset
+ physical_len
;
7911 ret
= btrfs_next_item(root
, path
);
7920 /* Ensure all chunks have corresponding dev extents */
7921 ret
= verify_chunk_dev_extent_mapping(fs_info
);
7923 btrfs_free_path(path
);
7928 * Check whether the given block group or device is pinned by any inode being
7929 * used as a swapfile.
7931 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info
*fs_info
, void *ptr
)
7933 struct btrfs_swapfile_pin
*sp
;
7934 struct rb_node
*node
;
7936 spin_lock(&fs_info
->swapfile_pins_lock
);
7937 node
= fs_info
->swapfile_pins
.rb_node
;
7939 sp
= rb_entry(node
, struct btrfs_swapfile_pin
, node
);
7941 node
= node
->rb_left
;
7942 else if (ptr
> sp
->ptr
)
7943 node
= node
->rb_right
;
7947 spin_unlock(&fs_info
->swapfile_pins_lock
);
7948 return node
!= NULL
;
7951 static int relocating_repair_kthread(void *data
)
7953 struct btrfs_block_group
*cache
= data
;
7954 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
7958 target
= cache
->start
;
7959 btrfs_put_block_group(cache
);
7961 sb_start_write(fs_info
->sb
);
7962 if (!btrfs_exclop_start(fs_info
, BTRFS_EXCLOP_BALANCE
)) {
7964 "zoned: skip relocating block group %llu to repair: EBUSY",
7966 sb_end_write(fs_info
->sb
);
7970 mutex_lock(&fs_info
->reclaim_bgs_lock
);
7972 /* Ensure block group still exists */
7973 cache
= btrfs_lookup_block_group(fs_info
, target
);
7977 if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR
, &cache
->runtime_flags
))
7980 ret
= btrfs_may_alloc_data_chunk(fs_info
, target
);
7985 "zoned: relocating block group %llu to repair IO failure",
7987 ret
= btrfs_relocate_chunk(fs_info
, target
);
7991 btrfs_put_block_group(cache
);
7992 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
7993 btrfs_exclop_finish(fs_info
);
7994 sb_end_write(fs_info
->sb
);
7999 bool btrfs_repair_one_zone(struct btrfs_fs_info
*fs_info
, u64 logical
)
8001 struct btrfs_block_group
*cache
;
8003 if (!btrfs_is_zoned(fs_info
))
8006 /* Do not attempt to repair in degraded state */
8007 if (btrfs_test_opt(fs_info
, DEGRADED
))
8010 cache
= btrfs_lookup_block_group(fs_info
, logical
);
8014 if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR
, &cache
->runtime_flags
)) {
8015 btrfs_put_block_group(cache
);
8019 kthread_run(relocating_repair_kthread
, cache
,
8020 "btrfs-relocating-repair");
8025 static void map_raid56_repair_block(struct btrfs_io_context
*bioc
,
8026 struct btrfs_io_stripe
*smap
,
8029 int data_stripes
= nr_bioc_data_stripes(bioc
);
8032 for (i
= 0; i
< data_stripes
; i
++) {
8033 u64 stripe_start
= bioc
->full_stripe_logical
+
8034 btrfs_stripe_nr_to_offset(i
);
8036 if (logical
>= stripe_start
&&
8037 logical
< stripe_start
+ BTRFS_STRIPE_LEN
)
8040 ASSERT(i
< data_stripes
);
8041 smap
->dev
= bioc
->stripes
[i
].dev
;
8042 smap
->physical
= bioc
->stripes
[i
].physical
+
8043 ((logical
- bioc
->full_stripe_logical
) &
8044 BTRFS_STRIPE_LEN_MASK
);
8048 * Map a repair write into a single device.
8050 * A repair write is triggered by read time repair or scrub, which would only
8051 * update the contents of a single device.
8052 * Not update any other mirrors nor go through RMW path.
8054 * Callers should ensure:
8056 * - Call btrfs_bio_counter_inc_blocked() first
8057 * - The range does not cross stripe boundary
8058 * - Has a valid @mirror_num passed in.
8060 int btrfs_map_repair_block(struct btrfs_fs_info
*fs_info
,
8061 struct btrfs_io_stripe
*smap
, u64 logical
,
8062 u32 length
, int mirror_num
)
8064 struct btrfs_io_context
*bioc
= NULL
;
8065 u64 map_length
= length
;
8066 int mirror_ret
= mirror_num
;
8069 ASSERT(mirror_num
> 0);
8071 ret
= btrfs_map_block(fs_info
, BTRFS_MAP_WRITE
, logical
, &map_length
,
8072 &bioc
, smap
, &mirror_ret
);
8076 /* The map range should not cross stripe boundary. */
8077 ASSERT(map_length
>= length
);
8079 /* Already mapped to single stripe. */
8083 /* Map the RAID56 multi-stripe writes to a single one. */
8084 if (bioc
->map_type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
8085 map_raid56_repair_block(bioc
, smap
, logical
);
8089 ASSERT(mirror_num
<= bioc
->num_stripes
);
8090 smap
->dev
= bioc
->stripes
[mirror_num
- 1].dev
;
8091 smap
->physical
= bioc
->stripes
[mirror_num
- 1].physical
;
8093 btrfs_put_bioc(bioc
);