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>
9 #include <linux/slab.h>
10 #include <linux/blkdev.h>
11 #include <linux/ratelimit.h>
12 #include <linux/kthread.h>
13 #include <linux/raid/pq.h>
14 #include <linux/semaphore.h>
15 #include <linux/uuid.h>
16 #include <linux/list_sort.h>
19 #include "extent_map.h"
21 #include "transaction.h"
22 #include "print-tree.h"
25 #include "async-thread.h"
26 #include "check-integrity.h"
27 #include "rcu-string.h"
28 #include "dev-replace.h"
30 #include "tree-checker.h"
31 #include "space-info.h"
32 #include "block-group.h"
36 const struct btrfs_raid_attr btrfs_raid_array
[BTRFS_NR_RAID_TYPES
] = {
37 [BTRFS_RAID_RAID10
] = {
40 .devs_max
= 0, /* 0 == as many as possible */
42 .tolerated_failures
= 1,
46 .raid_name
= "raid10",
47 .bg_flag
= BTRFS_BLOCK_GROUP_RAID10
,
48 .mindev_error
= BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET
,
50 [BTRFS_RAID_RAID1
] = {
55 .tolerated_failures
= 1,
60 .bg_flag
= BTRFS_BLOCK_GROUP_RAID1
,
61 .mindev_error
= BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET
,
63 [BTRFS_RAID_RAID1C3
] = {
68 .tolerated_failures
= 2,
72 .raid_name
= "raid1c3",
73 .bg_flag
= BTRFS_BLOCK_GROUP_RAID1C3
,
74 .mindev_error
= BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET
,
76 [BTRFS_RAID_RAID1C4
] = {
81 .tolerated_failures
= 3,
85 .raid_name
= "raid1c4",
86 .bg_flag
= BTRFS_BLOCK_GROUP_RAID1C4
,
87 .mindev_error
= BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET
,
94 .tolerated_failures
= 0,
99 .bg_flag
= BTRFS_BLOCK_GROUP_DUP
,
102 [BTRFS_RAID_RAID0
] = {
107 .tolerated_failures
= 0,
111 .raid_name
= "raid0",
112 .bg_flag
= BTRFS_BLOCK_GROUP_RAID0
,
115 [BTRFS_RAID_SINGLE
] = {
120 .tolerated_failures
= 0,
124 .raid_name
= "single",
128 [BTRFS_RAID_RAID5
] = {
133 .tolerated_failures
= 1,
137 .raid_name
= "raid5",
138 .bg_flag
= BTRFS_BLOCK_GROUP_RAID5
,
139 .mindev_error
= BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET
,
141 [BTRFS_RAID_RAID6
] = {
146 .tolerated_failures
= 2,
150 .raid_name
= "raid6",
151 .bg_flag
= BTRFS_BLOCK_GROUP_RAID6
,
152 .mindev_error
= BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET
,
156 const char *btrfs_bg_type_to_raid_name(u64 flags
)
158 const int index
= btrfs_bg_flags_to_raid_index(flags
);
160 if (index
>= BTRFS_NR_RAID_TYPES
)
163 return btrfs_raid_array
[index
].raid_name
;
167 * Fill @buf with textual description of @bg_flags, no more than @size_buf
168 * bytes including terminating null byte.
170 void btrfs_describe_block_groups(u64 bg_flags
, char *buf
, u32 size_buf
)
175 u64 flags
= bg_flags
;
176 u32 size_bp
= size_buf
;
183 #define DESCRIBE_FLAG(flag, desc) \
185 if (flags & (flag)) { \
186 ret = snprintf(bp, size_bp, "%s|", (desc)); \
187 if (ret < 0 || ret >= size_bp) \
195 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA
, "data");
196 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM
, "system");
197 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA
, "metadata");
199 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE
, "single");
200 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
201 DESCRIBE_FLAG(btrfs_raid_array
[i
].bg_flag
,
202 btrfs_raid_array
[i
].raid_name
);
206 ret
= snprintf(bp
, size_bp
, "0x%llx|", flags
);
210 if (size_bp
< size_buf
)
211 buf
[size_buf
- size_bp
- 1] = '\0'; /* remove last | */
214 * The text is trimmed, it's up to the caller to provide sufficiently
220 static int init_first_rw_device(struct btrfs_trans_handle
*trans
);
221 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info
*fs_info
);
222 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
);
223 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*device
);
224 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
,
225 enum btrfs_map_op op
,
226 u64 logical
, u64
*length
,
227 struct btrfs_bio
**bbio_ret
,
228 int mirror_num
, int need_raid_map
);
234 * There are several mutexes that protect manipulation of devices and low-level
235 * structures like chunks but not block groups, extents or files
237 * uuid_mutex (global lock)
238 * ------------------------
239 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
240 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
241 * device) or requested by the device= mount option
243 * the mutex can be very coarse and can cover long-running operations
245 * protects: updates to fs_devices counters like missing devices, rw devices,
246 * seeding, structure cloning, opening/closing devices at mount/umount time
248 * global::fs_devs - add, remove, updates to the global list
250 * does not protect: manipulation of the fs_devices::devices list in general
251 * but in mount context it could be used to exclude list modifications by eg.
254 * btrfs_device::name - renames (write side), read is RCU
256 * fs_devices::device_list_mutex (per-fs, with RCU)
257 * ------------------------------------------------
258 * protects updates to fs_devices::devices, ie. adding and deleting
260 * simple list traversal with read-only actions can be done with RCU protection
262 * may be used to exclude some operations from running concurrently without any
263 * modifications to the list (see write_all_supers)
265 * Is not required at mount and close times, because our device list is
266 * protected by the uuid_mutex at that point.
270 * protects balance structures (status, state) and context accessed from
271 * several places (internally, ioctl)
275 * protects chunks, adding or removing during allocation, trim or when a new
276 * device is added/removed. Additionally it also protects post_commit_list of
277 * individual devices, since they can be added to the transaction's
278 * post_commit_list only with chunk_mutex held.
282 * a big lock that is held by the cleaner thread and prevents running subvolume
283 * cleaning together with relocation or delayed iputs
295 * Exclusive operations
296 * ====================
298 * Maintains the exclusivity of the following operations that apply to the
299 * whole filesystem and cannot run in parallel.
304 * - Device replace (*)
307 * The device operations (as above) can be in one of the following states:
313 * Only device operations marked with (*) can go into the Paused state for the
316 * - ioctl (only Balance can be Paused through ioctl)
317 * - filesystem remounted as read-only
318 * - filesystem unmounted and mounted as read-only
319 * - system power-cycle and filesystem mounted as read-only
320 * - filesystem or device errors leading to forced read-only
322 * The status of exclusive operation is set and cleared atomically.
323 * During the course of Paused state, fs_info::exclusive_operation remains set.
324 * A device operation in Paused or Running state can be canceled or resumed
325 * either by ioctl (Balance only) or when remounted as read-write.
326 * The exclusive status is cleared when the device operation is canceled or
330 DEFINE_MUTEX(uuid_mutex
);
331 static LIST_HEAD(fs_uuids
);
332 struct list_head
* __attribute_const__
btrfs_get_fs_uuids(void)
338 * alloc_fs_devices - allocate struct btrfs_fs_devices
339 * @fsid: if not NULL, copy the UUID to fs_devices::fsid
340 * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid
342 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
343 * The returned struct is not linked onto any lists and can be destroyed with
344 * kfree() right away.
346 static struct btrfs_fs_devices
*alloc_fs_devices(const u8
*fsid
,
347 const u8
*metadata_fsid
)
349 struct btrfs_fs_devices
*fs_devs
;
351 fs_devs
= kzalloc(sizeof(*fs_devs
), GFP_KERNEL
);
353 return ERR_PTR(-ENOMEM
);
355 mutex_init(&fs_devs
->device_list_mutex
);
357 INIT_LIST_HEAD(&fs_devs
->devices
);
358 INIT_LIST_HEAD(&fs_devs
->alloc_list
);
359 INIT_LIST_HEAD(&fs_devs
->fs_list
);
360 INIT_LIST_HEAD(&fs_devs
->seed_list
);
362 memcpy(fs_devs
->fsid
, fsid
, BTRFS_FSID_SIZE
);
365 memcpy(fs_devs
->metadata_uuid
, metadata_fsid
, BTRFS_FSID_SIZE
);
367 memcpy(fs_devs
->metadata_uuid
, fsid
, BTRFS_FSID_SIZE
);
372 void btrfs_free_device(struct btrfs_device
*device
)
374 WARN_ON(!list_empty(&device
->post_commit_list
));
375 rcu_string_free(device
->name
);
376 extent_io_tree_release(&device
->alloc_state
);
377 bio_put(device
->flush_bio
);
378 btrfs_destroy_dev_zone_info(device
);
382 static void free_fs_devices(struct btrfs_fs_devices
*fs_devices
)
384 struct btrfs_device
*device
;
385 WARN_ON(fs_devices
->opened
);
386 while (!list_empty(&fs_devices
->devices
)) {
387 device
= list_entry(fs_devices
->devices
.next
,
388 struct btrfs_device
, dev_list
);
389 list_del(&device
->dev_list
);
390 btrfs_free_device(device
);
395 void __exit
btrfs_cleanup_fs_uuids(void)
397 struct btrfs_fs_devices
*fs_devices
;
399 while (!list_empty(&fs_uuids
)) {
400 fs_devices
= list_entry(fs_uuids
.next
,
401 struct btrfs_fs_devices
, fs_list
);
402 list_del(&fs_devices
->fs_list
);
403 free_fs_devices(fs_devices
);
408 * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
409 * Returned struct is not linked onto any lists and must be destroyed using
412 static struct btrfs_device
*__alloc_device(struct btrfs_fs_info
*fs_info
)
414 struct btrfs_device
*dev
;
416 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
418 return ERR_PTR(-ENOMEM
);
421 * Preallocate a bio that's always going to be used for flushing device
422 * barriers and matches the device lifespan
424 dev
->flush_bio
= bio_kmalloc(GFP_KERNEL
, 0);
425 if (!dev
->flush_bio
) {
427 return ERR_PTR(-ENOMEM
);
430 INIT_LIST_HEAD(&dev
->dev_list
);
431 INIT_LIST_HEAD(&dev
->dev_alloc_list
);
432 INIT_LIST_HEAD(&dev
->post_commit_list
);
434 atomic_set(&dev
->reada_in_flight
, 0);
435 atomic_set(&dev
->dev_stats_ccnt
, 0);
436 btrfs_device_data_ordered_init(dev
);
437 INIT_RADIX_TREE(&dev
->reada_zones
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
438 INIT_RADIX_TREE(&dev
->reada_extents
, GFP_NOFS
& ~__GFP_DIRECT_RECLAIM
);
439 extent_io_tree_init(fs_info
, &dev
->alloc_state
,
440 IO_TREE_DEVICE_ALLOC_STATE
, NULL
);
445 static noinline
struct btrfs_fs_devices
*find_fsid(
446 const u8
*fsid
, const u8
*metadata_fsid
)
448 struct btrfs_fs_devices
*fs_devices
;
452 /* Handle non-split brain cases */
453 list_for_each_entry(fs_devices
, &fs_uuids
, fs_list
) {
455 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0
456 && memcmp(metadata_fsid
, fs_devices
->metadata_uuid
,
457 BTRFS_FSID_SIZE
) == 0)
460 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
467 static struct btrfs_fs_devices
*find_fsid_with_metadata_uuid(
468 struct btrfs_super_block
*disk_super
)
471 struct btrfs_fs_devices
*fs_devices
;
474 * Handle scanned device having completed its fsid change but
475 * belonging to a fs_devices that was created by first scanning
476 * a device which didn't have its fsid/metadata_uuid changed
477 * at all and the CHANGING_FSID_V2 flag set.
479 list_for_each_entry(fs_devices
, &fs_uuids
, fs_list
) {
480 if (fs_devices
->fsid_change
&&
481 memcmp(disk_super
->metadata_uuid
, fs_devices
->fsid
,
482 BTRFS_FSID_SIZE
) == 0 &&
483 memcmp(fs_devices
->fsid
, fs_devices
->metadata_uuid
,
484 BTRFS_FSID_SIZE
) == 0) {
489 * Handle scanned device having completed its fsid change but
490 * belonging to a fs_devices that was created by a device that
491 * has an outdated pair of fsid/metadata_uuid and
492 * CHANGING_FSID_V2 flag set.
494 list_for_each_entry(fs_devices
, &fs_uuids
, fs_list
) {
495 if (fs_devices
->fsid_change
&&
496 memcmp(fs_devices
->metadata_uuid
,
497 fs_devices
->fsid
, BTRFS_FSID_SIZE
) != 0 &&
498 memcmp(disk_super
->metadata_uuid
, fs_devices
->metadata_uuid
,
499 BTRFS_FSID_SIZE
) == 0) {
504 return find_fsid(disk_super
->fsid
, disk_super
->metadata_uuid
);
509 btrfs_get_bdev_and_sb(const char *device_path
, fmode_t flags
, void *holder
,
510 int flush
, struct block_device
**bdev
,
511 struct btrfs_super_block
**disk_super
)
515 *bdev
= blkdev_get_by_path(device_path
, flags
, holder
);
518 ret
= PTR_ERR(*bdev
);
523 filemap_write_and_wait((*bdev
)->bd_inode
->i_mapping
);
524 ret
= set_blocksize(*bdev
, BTRFS_BDEV_BLOCKSIZE
);
526 blkdev_put(*bdev
, flags
);
529 invalidate_bdev(*bdev
);
530 *disk_super
= btrfs_read_dev_super(*bdev
);
531 if (IS_ERR(*disk_super
)) {
532 ret
= PTR_ERR(*disk_super
);
533 blkdev_put(*bdev
, flags
);
544 static bool device_path_matched(const char *path
, struct btrfs_device
*device
)
549 found
= strcmp(rcu_str_deref(device
->name
), path
);
556 * Search and remove all stale (devices which are not mounted) devices.
557 * When both inputs are NULL, it will search and release all stale devices.
558 * path: Optional. When provided will it release all unmounted devices
559 * matching this path only.
560 * skip_dev: Optional. Will skip this device when searching for the stale
562 * Return: 0 for success or if @path is NULL.
563 * -EBUSY if @path is a mounted device.
564 * -ENOENT if @path does not match any device in the list.
566 static int btrfs_free_stale_devices(const char *path
,
567 struct btrfs_device
*skip_device
)
569 struct btrfs_fs_devices
*fs_devices
, *tmp_fs_devices
;
570 struct btrfs_device
*device
, *tmp_device
;
576 list_for_each_entry_safe(fs_devices
, tmp_fs_devices
, &fs_uuids
, fs_list
) {
578 mutex_lock(&fs_devices
->device_list_mutex
);
579 list_for_each_entry_safe(device
, tmp_device
,
580 &fs_devices
->devices
, dev_list
) {
581 if (skip_device
&& skip_device
== device
)
583 if (path
&& !device
->name
)
585 if (path
&& !device_path_matched(path
, device
))
587 if (fs_devices
->opened
) {
588 /* for an already deleted device return 0 */
589 if (path
&& ret
!= 0)
594 /* delete the stale device */
595 fs_devices
->num_devices
--;
596 list_del(&device
->dev_list
);
597 btrfs_free_device(device
);
601 mutex_unlock(&fs_devices
->device_list_mutex
);
603 if (fs_devices
->num_devices
== 0) {
604 btrfs_sysfs_remove_fsid(fs_devices
);
605 list_del(&fs_devices
->fs_list
);
606 free_fs_devices(fs_devices
);
614 * This is only used on mount, and we are protected from competing things
615 * messing with our fs_devices by the uuid_mutex, thus we do not need the
616 * fs_devices->device_list_mutex here.
618 static int btrfs_open_one_device(struct btrfs_fs_devices
*fs_devices
,
619 struct btrfs_device
*device
, fmode_t flags
,
622 struct request_queue
*q
;
623 struct block_device
*bdev
;
624 struct btrfs_super_block
*disk_super
;
633 ret
= btrfs_get_bdev_and_sb(device
->name
->str
, flags
, holder
, 1,
638 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
639 if (devid
!= device
->devid
)
640 goto error_free_page
;
642 if (memcmp(device
->uuid
, disk_super
->dev_item
.uuid
, BTRFS_UUID_SIZE
))
643 goto error_free_page
;
645 device
->generation
= btrfs_super_generation(disk_super
);
647 if (btrfs_super_flags(disk_super
) & BTRFS_SUPER_FLAG_SEEDING
) {
648 if (btrfs_super_incompat_flags(disk_super
) &
649 BTRFS_FEATURE_INCOMPAT_METADATA_UUID
) {
651 "BTRFS: Invalid seeding and uuid-changed device detected\n");
652 goto error_free_page
;
655 clear_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
);
656 fs_devices
->seeding
= true;
658 if (bdev_read_only(bdev
))
659 clear_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
);
661 set_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
);
664 q
= bdev_get_queue(bdev
);
665 if (!blk_queue_nonrot(q
))
666 fs_devices
->rotating
= true;
669 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &device
->dev_state
);
670 device
->mode
= flags
;
672 fs_devices
->open_devices
++;
673 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
) &&
674 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
675 fs_devices
->rw_devices
++;
676 list_add_tail(&device
->dev_alloc_list
, &fs_devices
->alloc_list
);
678 btrfs_release_disk_super(disk_super
);
683 btrfs_release_disk_super(disk_super
);
684 blkdev_put(bdev
, flags
);
690 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices
691 * being created with a disk that has already completed its fsid change. Such
692 * disk can belong to an fs which has its FSID changed or to one which doesn't.
693 * Handle both cases here.
695 static struct btrfs_fs_devices
*find_fsid_inprogress(
696 struct btrfs_super_block
*disk_super
)
698 struct btrfs_fs_devices
*fs_devices
;
700 list_for_each_entry(fs_devices
, &fs_uuids
, fs_list
) {
701 if (memcmp(fs_devices
->metadata_uuid
, fs_devices
->fsid
,
702 BTRFS_FSID_SIZE
) != 0 &&
703 memcmp(fs_devices
->metadata_uuid
, disk_super
->fsid
,
704 BTRFS_FSID_SIZE
) == 0 && !fs_devices
->fsid_change
) {
709 return find_fsid(disk_super
->fsid
, NULL
);
713 static struct btrfs_fs_devices
*find_fsid_changed(
714 struct btrfs_super_block
*disk_super
)
716 struct btrfs_fs_devices
*fs_devices
;
719 * Handles the case where scanned device is part of an fs that had
720 * multiple successful changes of FSID but curently device didn't
721 * observe it. Meaning our fsid will be different than theirs. We need
722 * to handle two subcases :
723 * 1 - The fs still continues to have different METADATA/FSID uuids.
724 * 2 - The fs is switched back to its original FSID (METADATA/FSID
727 list_for_each_entry(fs_devices
, &fs_uuids
, fs_list
) {
729 if (memcmp(fs_devices
->metadata_uuid
, fs_devices
->fsid
,
730 BTRFS_FSID_SIZE
) != 0 &&
731 memcmp(fs_devices
->metadata_uuid
, disk_super
->metadata_uuid
,
732 BTRFS_FSID_SIZE
) == 0 &&
733 memcmp(fs_devices
->fsid
, disk_super
->fsid
,
734 BTRFS_FSID_SIZE
) != 0)
737 /* Unchanged UUIDs */
738 if (memcmp(fs_devices
->metadata_uuid
, fs_devices
->fsid
,
739 BTRFS_FSID_SIZE
) == 0 &&
740 memcmp(fs_devices
->fsid
, disk_super
->metadata_uuid
,
741 BTRFS_FSID_SIZE
) == 0)
748 static struct btrfs_fs_devices
*find_fsid_reverted_metadata(
749 struct btrfs_super_block
*disk_super
)
751 struct btrfs_fs_devices
*fs_devices
;
754 * Handle the case where the scanned device is part of an fs whose last
755 * metadata UUID change reverted it to the original FSID. At the same
756 * time * fs_devices was first created by another constitutent device
757 * which didn't fully observe the operation. This results in an
758 * btrfs_fs_devices created with metadata/fsid different AND
759 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the
760 * fs_devices equal to the FSID of the disk.
762 list_for_each_entry(fs_devices
, &fs_uuids
, fs_list
) {
763 if (memcmp(fs_devices
->fsid
, fs_devices
->metadata_uuid
,
764 BTRFS_FSID_SIZE
) != 0 &&
765 memcmp(fs_devices
->metadata_uuid
, disk_super
->fsid
,
766 BTRFS_FSID_SIZE
) == 0 &&
767 fs_devices
->fsid_change
)
774 * Add new device to list of registered devices
777 * device pointer which was just added or updated when successful
778 * error pointer when failed
780 static noinline
struct btrfs_device
*device_list_add(const char *path
,
781 struct btrfs_super_block
*disk_super
,
782 bool *new_device_added
)
784 struct btrfs_device
*device
;
785 struct btrfs_fs_devices
*fs_devices
= NULL
;
786 struct rcu_string
*name
;
787 u64 found_transid
= btrfs_super_generation(disk_super
);
788 u64 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
789 bool has_metadata_uuid
= (btrfs_super_incompat_flags(disk_super
) &
790 BTRFS_FEATURE_INCOMPAT_METADATA_UUID
);
791 bool fsid_change_in_progress
= (btrfs_super_flags(disk_super
) &
792 BTRFS_SUPER_FLAG_CHANGING_FSID_V2
);
794 if (fsid_change_in_progress
) {
795 if (!has_metadata_uuid
)
796 fs_devices
= find_fsid_inprogress(disk_super
);
798 fs_devices
= find_fsid_changed(disk_super
);
799 } else if (has_metadata_uuid
) {
800 fs_devices
= find_fsid_with_metadata_uuid(disk_super
);
802 fs_devices
= find_fsid_reverted_metadata(disk_super
);
804 fs_devices
= find_fsid(disk_super
->fsid
, NULL
);
809 if (has_metadata_uuid
)
810 fs_devices
= alloc_fs_devices(disk_super
->fsid
,
811 disk_super
->metadata_uuid
);
813 fs_devices
= alloc_fs_devices(disk_super
->fsid
, NULL
);
815 if (IS_ERR(fs_devices
))
816 return ERR_CAST(fs_devices
);
818 fs_devices
->fsid_change
= fsid_change_in_progress
;
820 mutex_lock(&fs_devices
->device_list_mutex
);
821 list_add(&fs_devices
->fs_list
, &fs_uuids
);
825 mutex_lock(&fs_devices
->device_list_mutex
);
826 device
= btrfs_find_device(fs_devices
, devid
,
827 disk_super
->dev_item
.uuid
, NULL
);
830 * If this disk has been pulled into an fs devices created by
831 * a device which had the CHANGING_FSID_V2 flag then replace the
832 * metadata_uuid/fsid values of the fs_devices.
834 if (fs_devices
->fsid_change
&&
835 found_transid
> fs_devices
->latest_generation
) {
836 memcpy(fs_devices
->fsid
, disk_super
->fsid
,
839 if (has_metadata_uuid
)
840 memcpy(fs_devices
->metadata_uuid
,
841 disk_super
->metadata_uuid
,
844 memcpy(fs_devices
->metadata_uuid
,
845 disk_super
->fsid
, BTRFS_FSID_SIZE
);
847 fs_devices
->fsid_change
= false;
852 if (fs_devices
->opened
) {
853 mutex_unlock(&fs_devices
->device_list_mutex
);
854 return ERR_PTR(-EBUSY
);
857 device
= btrfs_alloc_device(NULL
, &devid
,
858 disk_super
->dev_item
.uuid
);
859 if (IS_ERR(device
)) {
860 mutex_unlock(&fs_devices
->device_list_mutex
);
861 /* we can safely leave the fs_devices entry around */
865 name
= rcu_string_strdup(path
, GFP_NOFS
);
867 btrfs_free_device(device
);
868 mutex_unlock(&fs_devices
->device_list_mutex
);
869 return ERR_PTR(-ENOMEM
);
871 rcu_assign_pointer(device
->name
, name
);
873 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
874 fs_devices
->num_devices
++;
876 device
->fs_devices
= fs_devices
;
877 *new_device_added
= true;
879 if (disk_super
->label
[0])
881 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n",
882 disk_super
->label
, devid
, found_transid
, path
,
883 current
->comm
, task_pid_nr(current
));
886 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n",
887 disk_super
->fsid
, devid
, found_transid
, path
,
888 current
->comm
, task_pid_nr(current
));
890 } else if (!device
->name
|| strcmp(device
->name
->str
, path
)) {
892 * When FS is already mounted.
893 * 1. If you are here and if the device->name is NULL that
894 * means this device was missing at time of FS mount.
895 * 2. If you are here and if the device->name is different
896 * from 'path' that means either
897 * a. The same device disappeared and reappeared with
899 * b. The missing-disk-which-was-replaced, has
902 * We must allow 1 and 2a above. But 2b would be a spurious
905 * Further in case of 1 and 2a above, the disk at 'path'
906 * would have missed some transaction when it was away and
907 * in case of 2a the stale bdev has to be updated as well.
908 * 2b must not be allowed at all time.
912 * For now, we do allow update to btrfs_fs_device through the
913 * btrfs dev scan cli after FS has been mounted. We're still
914 * tracking a problem where systems fail mount by subvolume id
915 * when we reject replacement on a mounted FS.
917 if (!fs_devices
->opened
&& found_transid
< device
->generation
) {
919 * That is if the FS is _not_ mounted and if you
920 * are here, that means there is more than one
921 * disk with same uuid and devid.We keep the one
922 * with larger generation number or the last-in if
923 * generation are equal.
925 mutex_unlock(&fs_devices
->device_list_mutex
);
926 return ERR_PTR(-EEXIST
);
930 * We are going to replace the device path for a given devid,
931 * make sure it's the same device if the device is mounted
937 error
= lookup_bdev(path
, &path_dev
);
939 mutex_unlock(&fs_devices
->device_list_mutex
);
940 return ERR_PTR(error
);
943 if (device
->bdev
->bd_dev
!= path_dev
) {
944 mutex_unlock(&fs_devices
->device_list_mutex
);
946 * device->fs_info may not be reliable here, so
947 * pass in a NULL instead. This avoids a
948 * possible use-after-free when the fs_info and
949 * fs_info->sb are already torn down.
951 btrfs_warn_in_rcu(NULL
,
952 "duplicate device %s devid %llu generation %llu scanned by %s (%d)",
953 path
, devid
, found_transid
,
955 task_pid_nr(current
));
956 return ERR_PTR(-EEXIST
);
958 btrfs_info_in_rcu(device
->fs_info
,
959 "devid %llu device path %s changed to %s scanned by %s (%d)",
960 devid
, rcu_str_deref(device
->name
),
962 task_pid_nr(current
));
965 name
= rcu_string_strdup(path
, GFP_NOFS
);
967 mutex_unlock(&fs_devices
->device_list_mutex
);
968 return ERR_PTR(-ENOMEM
);
970 rcu_string_free(device
->name
);
971 rcu_assign_pointer(device
->name
, name
);
972 if (test_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
)) {
973 fs_devices
->missing_devices
--;
974 clear_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
);
979 * Unmount does not free the btrfs_device struct but would zero
980 * generation along with most of the other members. So just update
981 * it back. We need it to pick the disk with largest generation
984 if (!fs_devices
->opened
) {
985 device
->generation
= found_transid
;
986 fs_devices
->latest_generation
= max_t(u64
, found_transid
,
987 fs_devices
->latest_generation
);
990 fs_devices
->total_devices
= btrfs_super_num_devices(disk_super
);
992 mutex_unlock(&fs_devices
->device_list_mutex
);
996 static struct btrfs_fs_devices
*clone_fs_devices(struct btrfs_fs_devices
*orig
)
998 struct btrfs_fs_devices
*fs_devices
;
999 struct btrfs_device
*device
;
1000 struct btrfs_device
*orig_dev
;
1003 fs_devices
= alloc_fs_devices(orig
->fsid
, NULL
);
1004 if (IS_ERR(fs_devices
))
1007 mutex_lock(&orig
->device_list_mutex
);
1008 fs_devices
->total_devices
= orig
->total_devices
;
1010 list_for_each_entry(orig_dev
, &orig
->devices
, dev_list
) {
1011 struct rcu_string
*name
;
1013 device
= btrfs_alloc_device(NULL
, &orig_dev
->devid
,
1015 if (IS_ERR(device
)) {
1016 ret
= PTR_ERR(device
);
1021 * This is ok to do without rcu read locked because we hold the
1022 * uuid mutex so nothing we touch in here is going to disappear.
1024 if (orig_dev
->name
) {
1025 name
= rcu_string_strdup(orig_dev
->name
->str
,
1028 btrfs_free_device(device
);
1032 rcu_assign_pointer(device
->name
, name
);
1035 list_add(&device
->dev_list
, &fs_devices
->devices
);
1036 device
->fs_devices
= fs_devices
;
1037 fs_devices
->num_devices
++;
1039 mutex_unlock(&orig
->device_list_mutex
);
1042 mutex_unlock(&orig
->device_list_mutex
);
1043 free_fs_devices(fs_devices
);
1044 return ERR_PTR(ret
);
1047 static void __btrfs_free_extra_devids(struct btrfs_fs_devices
*fs_devices
,
1048 struct btrfs_device
**latest_dev
)
1050 struct btrfs_device
*device
, *next
;
1052 /* This is the initialized path, it is safe to release the devices. */
1053 list_for_each_entry_safe(device
, next
, &fs_devices
->devices
, dev_list
) {
1054 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &device
->dev_state
)) {
1055 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT
,
1056 &device
->dev_state
) &&
1057 !test_bit(BTRFS_DEV_STATE_MISSING
,
1058 &device
->dev_state
) &&
1060 device
->generation
> (*latest_dev
)->generation
)) {
1061 *latest_dev
= device
;
1067 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID,
1068 * in btrfs_init_dev_replace() so just continue.
1070 if (device
->devid
== BTRFS_DEV_REPLACE_DEVID
)
1074 blkdev_put(device
->bdev
, device
->mode
);
1075 device
->bdev
= NULL
;
1076 fs_devices
->open_devices
--;
1078 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
)) {
1079 list_del_init(&device
->dev_alloc_list
);
1080 clear_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
);
1082 list_del_init(&device
->dev_list
);
1083 fs_devices
->num_devices
--;
1084 btrfs_free_device(device
);
1090 * After we have read the system tree and know devids belonging to this
1091 * filesystem, remove the device which does not belong there.
1093 void btrfs_free_extra_devids(struct btrfs_fs_devices
*fs_devices
)
1095 struct btrfs_device
*latest_dev
= NULL
;
1096 struct btrfs_fs_devices
*seed_dev
;
1098 mutex_lock(&uuid_mutex
);
1099 __btrfs_free_extra_devids(fs_devices
, &latest_dev
);
1101 list_for_each_entry(seed_dev
, &fs_devices
->seed_list
, seed_list
)
1102 __btrfs_free_extra_devids(seed_dev
, &latest_dev
);
1104 fs_devices
->latest_bdev
= latest_dev
->bdev
;
1106 mutex_unlock(&uuid_mutex
);
1109 static void btrfs_close_bdev(struct btrfs_device
*device
)
1114 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
)) {
1115 sync_blockdev(device
->bdev
);
1116 invalidate_bdev(device
->bdev
);
1119 blkdev_put(device
->bdev
, device
->mode
);
1122 static void btrfs_close_one_device(struct btrfs_device
*device
)
1124 struct btrfs_fs_devices
*fs_devices
= device
->fs_devices
;
1126 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
) &&
1127 device
->devid
!= BTRFS_DEV_REPLACE_DEVID
) {
1128 list_del_init(&device
->dev_alloc_list
);
1129 fs_devices
->rw_devices
--;
1132 if (test_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
))
1133 fs_devices
->missing_devices
--;
1135 btrfs_close_bdev(device
);
1137 fs_devices
->open_devices
--;
1138 device
->bdev
= NULL
;
1140 clear_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
);
1141 btrfs_destroy_dev_zone_info(device
);
1143 device
->fs_info
= NULL
;
1144 atomic_set(&device
->dev_stats_ccnt
, 0);
1145 extent_io_tree_release(&device
->alloc_state
);
1147 /* Verify the device is back in a pristine state */
1148 ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT
, &device
->dev_state
));
1149 ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
));
1150 ASSERT(list_empty(&device
->dev_alloc_list
));
1151 ASSERT(list_empty(&device
->post_commit_list
));
1152 ASSERT(atomic_read(&device
->reada_in_flight
) == 0);
1155 static void close_fs_devices(struct btrfs_fs_devices
*fs_devices
)
1157 struct btrfs_device
*device
, *tmp
;
1159 lockdep_assert_held(&uuid_mutex
);
1161 if (--fs_devices
->opened
> 0)
1164 list_for_each_entry_safe(device
, tmp
, &fs_devices
->devices
, dev_list
)
1165 btrfs_close_one_device(device
);
1167 WARN_ON(fs_devices
->open_devices
);
1168 WARN_ON(fs_devices
->rw_devices
);
1169 fs_devices
->opened
= 0;
1170 fs_devices
->seeding
= false;
1171 fs_devices
->fs_info
= NULL
;
1174 void btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
1177 struct btrfs_fs_devices
*tmp
;
1179 mutex_lock(&uuid_mutex
);
1180 close_fs_devices(fs_devices
);
1181 if (!fs_devices
->opened
)
1182 list_splice_init(&fs_devices
->seed_list
, &list
);
1184 list_for_each_entry_safe(fs_devices
, tmp
, &list
, seed_list
) {
1185 close_fs_devices(fs_devices
);
1186 list_del(&fs_devices
->seed_list
);
1187 free_fs_devices(fs_devices
);
1189 mutex_unlock(&uuid_mutex
);
1192 static int open_fs_devices(struct btrfs_fs_devices
*fs_devices
,
1193 fmode_t flags
, void *holder
)
1195 struct btrfs_device
*device
;
1196 struct btrfs_device
*latest_dev
= NULL
;
1197 struct btrfs_device
*tmp_device
;
1199 flags
|= FMODE_EXCL
;
1201 list_for_each_entry_safe(device
, tmp_device
, &fs_devices
->devices
,
1205 ret
= btrfs_open_one_device(fs_devices
, device
, flags
, holder
);
1207 (!latest_dev
|| device
->generation
> latest_dev
->generation
)) {
1208 latest_dev
= device
;
1209 } else if (ret
== -ENODATA
) {
1210 fs_devices
->num_devices
--;
1211 list_del(&device
->dev_list
);
1212 btrfs_free_device(device
);
1215 if (fs_devices
->open_devices
== 0)
1218 fs_devices
->opened
= 1;
1219 fs_devices
->latest_bdev
= latest_dev
->bdev
;
1220 fs_devices
->total_rw_bytes
= 0;
1221 fs_devices
->chunk_alloc_policy
= BTRFS_CHUNK_ALLOC_REGULAR
;
1222 fs_devices
->read_policy
= BTRFS_READ_POLICY_PID
;
1227 static int devid_cmp(void *priv
, const struct list_head
*a
,
1228 const struct list_head
*b
)
1230 struct btrfs_device
*dev1
, *dev2
;
1232 dev1
= list_entry(a
, struct btrfs_device
, dev_list
);
1233 dev2
= list_entry(b
, struct btrfs_device
, dev_list
);
1235 if (dev1
->devid
< dev2
->devid
)
1237 else if (dev1
->devid
> dev2
->devid
)
1242 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
1243 fmode_t flags
, void *holder
)
1247 lockdep_assert_held(&uuid_mutex
);
1249 * The device_list_mutex cannot be taken here in case opening the
1250 * underlying device takes further locks like bd_mutex.
1252 * We also don't need the lock here as this is called during mount and
1253 * exclusion is provided by uuid_mutex
1256 if (fs_devices
->opened
) {
1257 fs_devices
->opened
++;
1260 list_sort(NULL
, &fs_devices
->devices
, devid_cmp
);
1261 ret
= open_fs_devices(fs_devices
, flags
, holder
);
1267 void btrfs_release_disk_super(struct btrfs_super_block
*super
)
1269 struct page
*page
= virt_to_page(super
);
1274 static struct btrfs_super_block
*btrfs_read_disk_super(struct block_device
*bdev
,
1275 u64 bytenr
, u64 bytenr_orig
)
1277 struct btrfs_super_block
*disk_super
;
1282 /* make sure our super fits in the device */
1283 if (bytenr
+ PAGE_SIZE
>= i_size_read(bdev
->bd_inode
))
1284 return ERR_PTR(-EINVAL
);
1286 /* make sure our super fits in the page */
1287 if (sizeof(*disk_super
) > PAGE_SIZE
)
1288 return ERR_PTR(-EINVAL
);
1290 /* make sure our super doesn't straddle pages on disk */
1291 index
= bytenr
>> PAGE_SHIFT
;
1292 if ((bytenr
+ sizeof(*disk_super
) - 1) >> PAGE_SHIFT
!= index
)
1293 return ERR_PTR(-EINVAL
);
1295 /* pull in the page with our super */
1296 page
= read_cache_page_gfp(bdev
->bd_inode
->i_mapping
, index
, GFP_KERNEL
);
1299 return ERR_CAST(page
);
1301 p
= page_address(page
);
1303 /* align our pointer to the offset of the super block */
1304 disk_super
= p
+ offset_in_page(bytenr
);
1306 if (btrfs_super_bytenr(disk_super
) != bytenr_orig
||
1307 btrfs_super_magic(disk_super
) != BTRFS_MAGIC
) {
1308 btrfs_release_disk_super(p
);
1309 return ERR_PTR(-EINVAL
);
1312 if (disk_super
->label
[0] && disk_super
->label
[BTRFS_LABEL_SIZE
- 1])
1313 disk_super
->label
[BTRFS_LABEL_SIZE
- 1] = 0;
1318 int btrfs_forget_devices(const char *path
)
1322 mutex_lock(&uuid_mutex
);
1323 ret
= btrfs_free_stale_devices(strlen(path
) ? path
: NULL
, NULL
);
1324 mutex_unlock(&uuid_mutex
);
1330 * Look for a btrfs signature on a device. This may be called out of the mount path
1331 * and we are not allowed to call set_blocksize during the scan. The superblock
1332 * is read via pagecache
1334 struct btrfs_device
*btrfs_scan_one_device(const char *path
, fmode_t flags
,
1337 struct btrfs_super_block
*disk_super
;
1338 bool new_device_added
= false;
1339 struct btrfs_device
*device
= NULL
;
1340 struct block_device
*bdev
;
1341 u64 bytenr
, bytenr_orig
;
1344 lockdep_assert_held(&uuid_mutex
);
1347 * we would like to check all the supers, but that would make
1348 * a btrfs mount succeed after a mkfs from a different FS.
1349 * So, we need to add a special mount option to scan for
1350 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1352 flags
|= FMODE_EXCL
;
1354 bdev
= blkdev_get_by_path(path
, flags
, holder
);
1356 return ERR_CAST(bdev
);
1358 bytenr_orig
= btrfs_sb_offset(0);
1359 ret
= btrfs_sb_log_location_bdev(bdev
, 0, READ
, &bytenr
);
1361 return ERR_PTR(ret
);
1363 disk_super
= btrfs_read_disk_super(bdev
, bytenr
, bytenr_orig
);
1364 if (IS_ERR(disk_super
)) {
1365 device
= ERR_CAST(disk_super
);
1366 goto error_bdev_put
;
1369 device
= device_list_add(path
, disk_super
, &new_device_added
);
1370 if (!IS_ERR(device
)) {
1371 if (new_device_added
)
1372 btrfs_free_stale_devices(path
, device
);
1375 btrfs_release_disk_super(disk_super
);
1378 blkdev_put(bdev
, flags
);
1384 * Try to find a chunk that intersects [start, start + len] range and when one
1385 * such is found, record the end of it in *start
1387 static bool contains_pending_extent(struct btrfs_device
*device
, u64
*start
,
1390 u64 physical_start
, physical_end
;
1392 lockdep_assert_held(&device
->fs_info
->chunk_mutex
);
1394 if (!find_first_extent_bit(&device
->alloc_state
, *start
,
1395 &physical_start
, &physical_end
,
1396 CHUNK_ALLOCATED
, NULL
)) {
1398 if (in_range(physical_start
, *start
, len
) ||
1399 in_range(*start
, physical_start
,
1400 physical_end
- physical_start
)) {
1401 *start
= physical_end
+ 1;
1408 static u64
dev_extent_search_start(struct btrfs_device
*device
, u64 start
)
1410 switch (device
->fs_devices
->chunk_alloc_policy
) {
1411 case BTRFS_CHUNK_ALLOC_REGULAR
:
1413 * We don't want to overwrite the superblock on the drive nor
1414 * any area used by the boot loader (grub for example), so we
1415 * make sure to start at an offset of at least 1MB.
1417 return max_t(u64
, start
, SZ_1M
);
1418 case BTRFS_CHUNK_ALLOC_ZONED
:
1420 * We don't care about the starting region like regular
1421 * allocator, because we anyway use/reserve the first two zones
1422 * for superblock logging.
1424 return ALIGN(start
, device
->zone_info
->zone_size
);
1430 static bool dev_extent_hole_check_zoned(struct btrfs_device
*device
,
1431 u64
*hole_start
, u64
*hole_size
,
1434 u64 zone_size
= device
->zone_info
->zone_size
;
1437 bool changed
= false;
1439 ASSERT(IS_ALIGNED(*hole_start
, zone_size
));
1441 while (*hole_size
> 0) {
1442 pos
= btrfs_find_allocatable_zones(device
, *hole_start
,
1443 *hole_start
+ *hole_size
,
1445 if (pos
!= *hole_start
) {
1446 *hole_size
= *hole_start
+ *hole_size
- pos
;
1449 if (*hole_size
< num_bytes
)
1453 ret
= btrfs_ensure_empty_zones(device
, pos
, num_bytes
);
1455 /* Range is ensured to be empty */
1459 /* Given hole range was invalid (outside of device) */
1460 if (ret
== -ERANGE
) {
1461 *hole_start
+= *hole_size
;
1466 *hole_start
+= zone_size
;
1467 *hole_size
-= zone_size
;
1475 * dev_extent_hole_check - check if specified hole is suitable for allocation
1476 * @device: the device which we have the hole
1477 * @hole_start: starting position of the hole
1478 * @hole_size: the size of the hole
1479 * @num_bytes: the size of the free space that we need
1481 * This function may modify @hole_start and @hole_size to reflect the suitable
1482 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1484 static bool dev_extent_hole_check(struct btrfs_device
*device
, u64
*hole_start
,
1485 u64
*hole_size
, u64 num_bytes
)
1487 bool changed
= false;
1488 u64 hole_end
= *hole_start
+ *hole_size
;
1492 * Check before we set max_hole_start, otherwise we could end up
1493 * sending back this offset anyway.
1495 if (contains_pending_extent(device
, hole_start
, *hole_size
)) {
1496 if (hole_end
>= *hole_start
)
1497 *hole_size
= hole_end
- *hole_start
;
1503 switch (device
->fs_devices
->chunk_alloc_policy
) {
1504 case BTRFS_CHUNK_ALLOC_REGULAR
:
1505 /* No extra check */
1507 case BTRFS_CHUNK_ALLOC_ZONED
:
1508 if (dev_extent_hole_check_zoned(device
, hole_start
,
1509 hole_size
, num_bytes
)) {
1512 * The changed hole can contain pending extent.
1513 * Loop again to check that.
1529 * find_free_dev_extent_start - find free space in the specified device
1530 * @device: the device which we search the free space in
1531 * @num_bytes: the size of the free space that we need
1532 * @search_start: the position from which to begin the search
1533 * @start: store the start of the free space.
1534 * @len: the size of the free space. that we find, or the size
1535 * of the max free space if we don't find suitable free space
1537 * this uses a pretty simple search, the expectation is that it is
1538 * called very infrequently and that a given device has a small number
1541 * @start is used to store the start of the free space if we find. But if we
1542 * don't find suitable free space, it will be used to store the start position
1543 * of the max free space.
1545 * @len is used to store the size of the free space that we find.
1546 * But if we don't find suitable free space, it is used to store the size of
1547 * the max free space.
1549 * NOTE: This function will search *commit* root of device tree, and does extra
1550 * check to ensure dev extents are not double allocated.
1551 * This makes the function safe to allocate dev extents but may not report
1552 * correct usable device space, as device extent freed in current transaction
1553 * is not reported as avaiable.
1555 static int find_free_dev_extent_start(struct btrfs_device
*device
,
1556 u64 num_bytes
, u64 search_start
, u64
*start
,
1559 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1560 struct btrfs_root
*root
= fs_info
->dev_root
;
1561 struct btrfs_key key
;
1562 struct btrfs_dev_extent
*dev_extent
;
1563 struct btrfs_path
*path
;
1568 u64 search_end
= device
->total_bytes
;
1571 struct extent_buffer
*l
;
1573 search_start
= dev_extent_search_start(device
, search_start
);
1575 WARN_ON(device
->zone_info
&&
1576 !IS_ALIGNED(num_bytes
, device
->zone_info
->zone_size
));
1578 path
= btrfs_alloc_path();
1582 max_hole_start
= search_start
;
1586 if (search_start
>= search_end
||
1587 test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
)) {
1592 path
->reada
= READA_FORWARD
;
1593 path
->search_commit_root
= 1;
1594 path
->skip_locking
= 1;
1596 key
.objectid
= device
->devid
;
1597 key
.offset
= search_start
;
1598 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1600 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1604 ret
= btrfs_previous_item(root
, path
, key
.objectid
, key
.type
);
1611 slot
= path
->slots
[0];
1612 if (slot
>= btrfs_header_nritems(l
)) {
1613 ret
= btrfs_next_leaf(root
, path
);
1621 btrfs_item_key_to_cpu(l
, &key
, slot
);
1623 if (key
.objectid
< device
->devid
)
1626 if (key
.objectid
> device
->devid
)
1629 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
1632 if (key
.offset
> search_start
) {
1633 hole_size
= key
.offset
- search_start
;
1634 dev_extent_hole_check(device
, &search_start
, &hole_size
,
1637 if (hole_size
> max_hole_size
) {
1638 max_hole_start
= search_start
;
1639 max_hole_size
= hole_size
;
1643 * If this free space is greater than which we need,
1644 * it must be the max free space that we have found
1645 * until now, so max_hole_start must point to the start
1646 * of this free space and the length of this free space
1647 * is stored in max_hole_size. Thus, we return
1648 * max_hole_start and max_hole_size and go back to the
1651 if (hole_size
>= num_bytes
) {
1657 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
1658 extent_end
= key
.offset
+ btrfs_dev_extent_length(l
,
1660 if (extent_end
> search_start
)
1661 search_start
= extent_end
;
1668 * At this point, search_start should be the end of
1669 * allocated dev extents, and when shrinking the device,
1670 * search_end may be smaller than search_start.
1672 if (search_end
> search_start
) {
1673 hole_size
= search_end
- search_start
;
1674 if (dev_extent_hole_check(device
, &search_start
, &hole_size
,
1676 btrfs_release_path(path
);
1680 if (hole_size
> max_hole_size
) {
1681 max_hole_start
= search_start
;
1682 max_hole_size
= hole_size
;
1687 if (max_hole_size
< num_bytes
)
1693 btrfs_free_path(path
);
1694 *start
= max_hole_start
;
1696 *len
= max_hole_size
;
1700 int find_free_dev_extent(struct btrfs_device
*device
, u64 num_bytes
,
1701 u64
*start
, u64
*len
)
1703 /* FIXME use last free of some kind */
1704 return find_free_dev_extent_start(device
, num_bytes
, 0, start
, len
);
1707 static int btrfs_free_dev_extent(struct btrfs_trans_handle
*trans
,
1708 struct btrfs_device
*device
,
1709 u64 start
, u64
*dev_extent_len
)
1711 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1712 struct btrfs_root
*root
= fs_info
->dev_root
;
1714 struct btrfs_path
*path
;
1715 struct btrfs_key key
;
1716 struct btrfs_key found_key
;
1717 struct extent_buffer
*leaf
= NULL
;
1718 struct btrfs_dev_extent
*extent
= NULL
;
1720 path
= btrfs_alloc_path();
1724 key
.objectid
= device
->devid
;
1726 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1728 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1730 ret
= btrfs_previous_item(root
, path
, key
.objectid
,
1731 BTRFS_DEV_EXTENT_KEY
);
1734 leaf
= path
->nodes
[0];
1735 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1736 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1737 struct btrfs_dev_extent
);
1738 BUG_ON(found_key
.offset
> start
|| found_key
.offset
+
1739 btrfs_dev_extent_length(leaf
, extent
) < start
);
1741 btrfs_release_path(path
);
1743 } else if (ret
== 0) {
1744 leaf
= path
->nodes
[0];
1745 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1746 struct btrfs_dev_extent
);
1748 btrfs_handle_fs_error(fs_info
, ret
, "Slot search failed");
1752 *dev_extent_len
= btrfs_dev_extent_length(leaf
, extent
);
1754 ret
= btrfs_del_item(trans
, root
, path
);
1756 btrfs_handle_fs_error(fs_info
, ret
,
1757 "Failed to remove dev extent item");
1759 set_bit(BTRFS_TRANS_HAVE_FREE_BGS
, &trans
->transaction
->flags
);
1762 btrfs_free_path(path
);
1766 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
1767 struct btrfs_device
*device
,
1768 u64 chunk_offset
, u64 start
, u64 num_bytes
)
1771 struct btrfs_path
*path
;
1772 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
1773 struct btrfs_root
*root
= fs_info
->dev_root
;
1774 struct btrfs_dev_extent
*extent
;
1775 struct extent_buffer
*leaf
;
1776 struct btrfs_key key
;
1778 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &device
->dev_state
));
1779 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
));
1780 path
= btrfs_alloc_path();
1784 key
.objectid
= device
->devid
;
1786 key
.type
= BTRFS_DEV_EXTENT_KEY
;
1787 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1792 leaf
= path
->nodes
[0];
1793 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
1794 struct btrfs_dev_extent
);
1795 btrfs_set_dev_extent_chunk_tree(leaf
, extent
,
1796 BTRFS_CHUNK_TREE_OBJECTID
);
1797 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
,
1798 BTRFS_FIRST_CHUNK_TREE_OBJECTID
);
1799 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
1801 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
1802 btrfs_mark_buffer_dirty(leaf
);
1804 btrfs_free_path(path
);
1808 static u64
find_next_chunk(struct btrfs_fs_info
*fs_info
)
1810 struct extent_map_tree
*em_tree
;
1811 struct extent_map
*em
;
1815 em_tree
= &fs_info
->mapping_tree
;
1816 read_lock(&em_tree
->lock
);
1817 n
= rb_last(&em_tree
->map
.rb_root
);
1819 em
= rb_entry(n
, struct extent_map
, rb_node
);
1820 ret
= em
->start
+ em
->len
;
1822 read_unlock(&em_tree
->lock
);
1827 static noinline
int find_next_devid(struct btrfs_fs_info
*fs_info
,
1831 struct btrfs_key key
;
1832 struct btrfs_key found_key
;
1833 struct btrfs_path
*path
;
1835 path
= btrfs_alloc_path();
1839 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1840 key
.type
= BTRFS_DEV_ITEM_KEY
;
1841 key
.offset
= (u64
)-1;
1843 ret
= btrfs_search_slot(NULL
, fs_info
->chunk_root
, &key
, path
, 0, 0);
1849 btrfs_err(fs_info
, "corrupted chunk tree devid -1 matched");
1854 ret
= btrfs_previous_item(fs_info
->chunk_root
, path
,
1855 BTRFS_DEV_ITEMS_OBJECTID
,
1856 BTRFS_DEV_ITEM_KEY
);
1860 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
1862 *devid_ret
= found_key
.offset
+ 1;
1866 btrfs_free_path(path
);
1871 * the device information is stored in the chunk root
1872 * the btrfs_device struct should be fully filled in
1874 static int btrfs_add_dev_item(struct btrfs_trans_handle
*trans
,
1875 struct btrfs_device
*device
)
1878 struct btrfs_path
*path
;
1879 struct btrfs_dev_item
*dev_item
;
1880 struct extent_buffer
*leaf
;
1881 struct btrfs_key key
;
1884 path
= btrfs_alloc_path();
1888 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1889 key
.type
= BTRFS_DEV_ITEM_KEY
;
1890 key
.offset
= device
->devid
;
1892 ret
= btrfs_insert_empty_item(trans
, trans
->fs_info
->chunk_root
, path
,
1893 &key
, sizeof(*dev_item
));
1897 leaf
= path
->nodes
[0];
1898 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
1900 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
1901 btrfs_set_device_generation(leaf
, dev_item
, 0);
1902 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
1903 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
1904 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
1905 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
1906 btrfs_set_device_total_bytes(leaf
, dev_item
,
1907 btrfs_device_get_disk_total_bytes(device
));
1908 btrfs_set_device_bytes_used(leaf
, dev_item
,
1909 btrfs_device_get_bytes_used(device
));
1910 btrfs_set_device_group(leaf
, dev_item
, 0);
1911 btrfs_set_device_seek_speed(leaf
, dev_item
, 0);
1912 btrfs_set_device_bandwidth(leaf
, dev_item
, 0);
1913 btrfs_set_device_start_offset(leaf
, dev_item
, 0);
1915 ptr
= btrfs_device_uuid(dev_item
);
1916 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1917 ptr
= btrfs_device_fsid(dev_item
);
1918 write_extent_buffer(leaf
, trans
->fs_info
->fs_devices
->metadata_uuid
,
1919 ptr
, BTRFS_FSID_SIZE
);
1920 btrfs_mark_buffer_dirty(leaf
);
1924 btrfs_free_path(path
);
1929 * Function to update ctime/mtime for a given device path.
1930 * Mainly used for ctime/mtime based probe like libblkid.
1932 static void update_dev_time(const char *path_name
)
1936 filp
= filp_open(path_name
, O_RDWR
, 0);
1939 file_update_time(filp
);
1940 filp_close(filp
, NULL
);
1943 static int btrfs_rm_dev_item(struct btrfs_device
*device
)
1945 struct btrfs_root
*root
= device
->fs_info
->chunk_root
;
1947 struct btrfs_path
*path
;
1948 struct btrfs_key key
;
1949 struct btrfs_trans_handle
*trans
;
1951 path
= btrfs_alloc_path();
1955 trans
= btrfs_start_transaction(root
, 0);
1956 if (IS_ERR(trans
)) {
1957 btrfs_free_path(path
);
1958 return PTR_ERR(trans
);
1960 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1961 key
.type
= BTRFS_DEV_ITEM_KEY
;
1962 key
.offset
= device
->devid
;
1964 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1968 btrfs_abort_transaction(trans
, ret
);
1969 btrfs_end_transaction(trans
);
1973 ret
= btrfs_del_item(trans
, root
, path
);
1975 btrfs_abort_transaction(trans
, ret
);
1976 btrfs_end_transaction(trans
);
1980 btrfs_free_path(path
);
1982 ret
= btrfs_commit_transaction(trans
);
1987 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1988 * filesystem. It's up to the caller to adjust that number regarding eg. device
1991 static int btrfs_check_raid_min_devices(struct btrfs_fs_info
*fs_info
,
1999 seq
= read_seqbegin(&fs_info
->profiles_lock
);
2001 all_avail
= fs_info
->avail_data_alloc_bits
|
2002 fs_info
->avail_system_alloc_bits
|
2003 fs_info
->avail_metadata_alloc_bits
;
2004 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
2006 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++) {
2007 if (!(all_avail
& btrfs_raid_array
[i
].bg_flag
))
2010 if (num_devices
< btrfs_raid_array
[i
].devs_min
) {
2011 int ret
= btrfs_raid_array
[i
].mindev_error
;
2021 static struct btrfs_device
* btrfs_find_next_active_device(
2022 struct btrfs_fs_devices
*fs_devs
, struct btrfs_device
*device
)
2024 struct btrfs_device
*next_device
;
2026 list_for_each_entry(next_device
, &fs_devs
->devices
, dev_list
) {
2027 if (next_device
!= device
&&
2028 !test_bit(BTRFS_DEV_STATE_MISSING
, &next_device
->dev_state
)
2029 && next_device
->bdev
)
2037 * Helper function to check if the given device is part of s_bdev / latest_bdev
2038 * and replace it with the provided or the next active device, in the context
2039 * where this function called, there should be always be another device (or
2040 * this_dev) which is active.
2042 void __cold
btrfs_assign_next_active_device(struct btrfs_device
*device
,
2043 struct btrfs_device
*next_device
)
2045 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
2048 next_device
= btrfs_find_next_active_device(fs_info
->fs_devices
,
2050 ASSERT(next_device
);
2052 if (fs_info
->sb
->s_bdev
&&
2053 (fs_info
->sb
->s_bdev
== device
->bdev
))
2054 fs_info
->sb
->s_bdev
= next_device
->bdev
;
2056 if (fs_info
->fs_devices
->latest_bdev
== device
->bdev
)
2057 fs_info
->fs_devices
->latest_bdev
= next_device
->bdev
;
2061 * Return btrfs_fs_devices::num_devices excluding the device that's being
2062 * currently replaced.
2064 static u64
btrfs_num_devices(struct btrfs_fs_info
*fs_info
)
2066 u64 num_devices
= fs_info
->fs_devices
->num_devices
;
2068 down_read(&fs_info
->dev_replace
.rwsem
);
2069 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
)) {
2070 ASSERT(num_devices
> 1);
2073 up_read(&fs_info
->dev_replace
.rwsem
);
2078 void btrfs_scratch_superblocks(struct btrfs_fs_info
*fs_info
,
2079 struct block_device
*bdev
,
2080 const char *device_path
)
2082 struct btrfs_super_block
*disk_super
;
2088 for (copy_num
= 0; copy_num
< BTRFS_SUPER_MIRROR_MAX
; copy_num
++) {
2092 disk_super
= btrfs_read_dev_one_super(bdev
, copy_num
);
2093 if (IS_ERR(disk_super
))
2096 if (bdev_is_zoned(bdev
)) {
2097 btrfs_reset_sb_log_zones(bdev
, copy_num
);
2101 memset(&disk_super
->magic
, 0, sizeof(disk_super
->magic
));
2103 page
= virt_to_page(disk_super
);
2104 set_page_dirty(page
);
2106 /* write_on_page() unlocks the page */
2107 ret
= write_one_page(page
);
2110 "error clearing superblock number %d (%d)",
2112 btrfs_release_disk_super(disk_super
);
2116 /* Notify udev that device has changed */
2117 btrfs_kobject_uevent(bdev
, KOBJ_CHANGE
);
2119 /* Update ctime/mtime for device path for libblkid */
2120 update_dev_time(device_path
);
2123 int btrfs_rm_device(struct btrfs_fs_info
*fs_info
, const char *device_path
,
2126 struct btrfs_device
*device
;
2127 struct btrfs_fs_devices
*cur_devices
;
2128 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2132 mutex_lock(&uuid_mutex
);
2134 num_devices
= btrfs_num_devices(fs_info
);
2136 ret
= btrfs_check_raid_min_devices(fs_info
, num_devices
- 1);
2140 device
= btrfs_find_device_by_devspec(fs_info
, devid
, device_path
);
2142 if (IS_ERR(device
)) {
2143 if (PTR_ERR(device
) == -ENOENT
&&
2144 strcmp(device_path
, "missing") == 0)
2145 ret
= BTRFS_ERROR_DEV_MISSING_NOT_FOUND
;
2147 ret
= PTR_ERR(device
);
2151 if (btrfs_pinned_by_swapfile(fs_info
, device
)) {
2152 btrfs_warn_in_rcu(fs_info
,
2153 "cannot remove device %s (devid %llu) due to active swapfile",
2154 rcu_str_deref(device
->name
), device
->devid
);
2159 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
)) {
2160 ret
= BTRFS_ERROR_DEV_TGT_REPLACE
;
2164 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
) &&
2165 fs_info
->fs_devices
->rw_devices
== 1) {
2166 ret
= BTRFS_ERROR_DEV_ONLY_WRITABLE
;
2170 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
)) {
2171 mutex_lock(&fs_info
->chunk_mutex
);
2172 list_del_init(&device
->dev_alloc_list
);
2173 device
->fs_devices
->rw_devices
--;
2174 mutex_unlock(&fs_info
->chunk_mutex
);
2177 mutex_unlock(&uuid_mutex
);
2178 ret
= btrfs_shrink_device(device
, 0);
2180 btrfs_reada_remove_dev(device
);
2181 mutex_lock(&uuid_mutex
);
2186 * TODO: the superblock still includes this device in its num_devices
2187 * counter although write_all_supers() is not locked out. This
2188 * could give a filesystem state which requires a degraded mount.
2190 ret
= btrfs_rm_dev_item(device
);
2194 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &device
->dev_state
);
2195 btrfs_scrub_cancel_dev(device
);
2198 * the device list mutex makes sure that we don't change
2199 * the device list while someone else is writing out all
2200 * the device supers. Whoever is writing all supers, should
2201 * lock the device list mutex before getting the number of
2202 * devices in the super block (super_copy). Conversely,
2203 * whoever updates the number of devices in the super block
2204 * (super_copy) should hold the device list mutex.
2208 * In normal cases the cur_devices == fs_devices. But in case
2209 * of deleting a seed device, the cur_devices should point to
2210 * its own fs_devices listed under the fs_devices->seed.
2212 cur_devices
= device
->fs_devices
;
2213 mutex_lock(&fs_devices
->device_list_mutex
);
2214 list_del_rcu(&device
->dev_list
);
2216 cur_devices
->num_devices
--;
2217 cur_devices
->total_devices
--;
2218 /* Update total_devices of the parent fs_devices if it's seed */
2219 if (cur_devices
!= fs_devices
)
2220 fs_devices
->total_devices
--;
2222 if (test_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
))
2223 cur_devices
->missing_devices
--;
2225 btrfs_assign_next_active_device(device
, NULL
);
2228 cur_devices
->open_devices
--;
2229 /* remove sysfs entry */
2230 btrfs_sysfs_remove_device(device
);
2233 num_devices
= btrfs_super_num_devices(fs_info
->super_copy
) - 1;
2234 btrfs_set_super_num_devices(fs_info
->super_copy
, num_devices
);
2235 mutex_unlock(&fs_devices
->device_list_mutex
);
2238 * at this point, the device is zero sized and detached from
2239 * the devices list. All that's left is to zero out the old
2240 * supers and free the device.
2242 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
))
2243 btrfs_scratch_superblocks(fs_info
, device
->bdev
,
2246 btrfs_close_bdev(device
);
2248 btrfs_free_device(device
);
2250 if (cur_devices
->open_devices
== 0) {
2251 list_del_init(&cur_devices
->seed_list
);
2252 close_fs_devices(cur_devices
);
2253 free_fs_devices(cur_devices
);
2257 mutex_unlock(&uuid_mutex
);
2261 btrfs_reada_undo_remove_dev(device
);
2262 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
)) {
2263 mutex_lock(&fs_info
->chunk_mutex
);
2264 list_add(&device
->dev_alloc_list
,
2265 &fs_devices
->alloc_list
);
2266 device
->fs_devices
->rw_devices
++;
2267 mutex_unlock(&fs_info
->chunk_mutex
);
2272 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device
*srcdev
)
2274 struct btrfs_fs_devices
*fs_devices
;
2276 lockdep_assert_held(&srcdev
->fs_info
->fs_devices
->device_list_mutex
);
2279 * in case of fs with no seed, srcdev->fs_devices will point
2280 * to fs_devices of fs_info. However when the dev being replaced is
2281 * a seed dev it will point to the seed's local fs_devices. In short
2282 * srcdev will have its correct fs_devices in both the cases.
2284 fs_devices
= srcdev
->fs_devices
;
2286 list_del_rcu(&srcdev
->dev_list
);
2287 list_del(&srcdev
->dev_alloc_list
);
2288 fs_devices
->num_devices
--;
2289 if (test_bit(BTRFS_DEV_STATE_MISSING
, &srcdev
->dev_state
))
2290 fs_devices
->missing_devices
--;
2292 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &srcdev
->dev_state
))
2293 fs_devices
->rw_devices
--;
2296 fs_devices
->open_devices
--;
2299 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device
*srcdev
)
2301 struct btrfs_fs_devices
*fs_devices
= srcdev
->fs_devices
;
2303 mutex_lock(&uuid_mutex
);
2305 btrfs_close_bdev(srcdev
);
2307 btrfs_free_device(srcdev
);
2309 /* if this is no devs we rather delete the fs_devices */
2310 if (!fs_devices
->num_devices
) {
2312 * On a mounted FS, num_devices can't be zero unless it's a
2313 * seed. In case of a seed device being replaced, the replace
2314 * target added to the sprout FS, so there will be no more
2315 * device left under the seed FS.
2317 ASSERT(fs_devices
->seeding
);
2319 list_del_init(&fs_devices
->seed_list
);
2320 close_fs_devices(fs_devices
);
2321 free_fs_devices(fs_devices
);
2323 mutex_unlock(&uuid_mutex
);
2326 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device
*tgtdev
)
2328 struct btrfs_fs_devices
*fs_devices
= tgtdev
->fs_info
->fs_devices
;
2330 mutex_lock(&fs_devices
->device_list_mutex
);
2332 btrfs_sysfs_remove_device(tgtdev
);
2335 fs_devices
->open_devices
--;
2337 fs_devices
->num_devices
--;
2339 btrfs_assign_next_active_device(tgtdev
, NULL
);
2341 list_del_rcu(&tgtdev
->dev_list
);
2343 mutex_unlock(&fs_devices
->device_list_mutex
);
2346 * The update_dev_time() with in btrfs_scratch_superblocks()
2347 * may lead to a call to btrfs_show_devname() which will try
2348 * to hold device_list_mutex. And here this device
2349 * is already out of device list, so we don't have to hold
2350 * the device_list_mutex lock.
2352 btrfs_scratch_superblocks(tgtdev
->fs_info
, tgtdev
->bdev
,
2355 btrfs_close_bdev(tgtdev
);
2357 btrfs_free_device(tgtdev
);
2360 static struct btrfs_device
*btrfs_find_device_by_path(
2361 struct btrfs_fs_info
*fs_info
, const char *device_path
)
2364 struct btrfs_super_block
*disk_super
;
2367 struct block_device
*bdev
;
2368 struct btrfs_device
*device
;
2370 ret
= btrfs_get_bdev_and_sb(device_path
, FMODE_READ
,
2371 fs_info
->bdev_holder
, 0, &bdev
, &disk_super
);
2373 return ERR_PTR(ret
);
2375 devid
= btrfs_stack_device_id(&disk_super
->dev_item
);
2376 dev_uuid
= disk_super
->dev_item
.uuid
;
2377 if (btrfs_fs_incompat(fs_info
, METADATA_UUID
))
2378 device
= btrfs_find_device(fs_info
->fs_devices
, devid
, dev_uuid
,
2379 disk_super
->metadata_uuid
);
2381 device
= btrfs_find_device(fs_info
->fs_devices
, devid
, dev_uuid
,
2384 btrfs_release_disk_super(disk_super
);
2386 device
= ERR_PTR(-ENOENT
);
2387 blkdev_put(bdev
, FMODE_READ
);
2392 * Lookup a device given by device id, or the path if the id is 0.
2394 struct btrfs_device
*btrfs_find_device_by_devspec(
2395 struct btrfs_fs_info
*fs_info
, u64 devid
,
2396 const char *device_path
)
2398 struct btrfs_device
*device
;
2401 device
= btrfs_find_device(fs_info
->fs_devices
, devid
, NULL
,
2404 return ERR_PTR(-ENOENT
);
2408 if (!device_path
|| !device_path
[0])
2409 return ERR_PTR(-EINVAL
);
2411 if (strcmp(device_path
, "missing") == 0) {
2412 /* Find first missing device */
2413 list_for_each_entry(device
, &fs_info
->fs_devices
->devices
,
2415 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
,
2416 &device
->dev_state
) && !device
->bdev
)
2419 return ERR_PTR(-ENOENT
);
2422 return btrfs_find_device_by_path(fs_info
, device_path
);
2426 * does all the dirty work required for changing file system's UUID.
2428 static int btrfs_prepare_sprout(struct btrfs_fs_info
*fs_info
)
2430 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2431 struct btrfs_fs_devices
*old_devices
;
2432 struct btrfs_fs_devices
*seed_devices
;
2433 struct btrfs_super_block
*disk_super
= fs_info
->super_copy
;
2434 struct btrfs_device
*device
;
2437 lockdep_assert_held(&uuid_mutex
);
2438 if (!fs_devices
->seeding
)
2442 * Private copy of the seed devices, anchored at
2443 * fs_info->fs_devices->seed_list
2445 seed_devices
= alloc_fs_devices(NULL
, NULL
);
2446 if (IS_ERR(seed_devices
))
2447 return PTR_ERR(seed_devices
);
2450 * It's necessary to retain a copy of the original seed fs_devices in
2451 * fs_uuids so that filesystems which have been seeded can successfully
2452 * reference the seed device from open_seed_devices. This also supports
2455 old_devices
= clone_fs_devices(fs_devices
);
2456 if (IS_ERR(old_devices
)) {
2457 kfree(seed_devices
);
2458 return PTR_ERR(old_devices
);
2461 list_add(&old_devices
->fs_list
, &fs_uuids
);
2463 memcpy(seed_devices
, fs_devices
, sizeof(*seed_devices
));
2464 seed_devices
->opened
= 1;
2465 INIT_LIST_HEAD(&seed_devices
->devices
);
2466 INIT_LIST_HEAD(&seed_devices
->alloc_list
);
2467 mutex_init(&seed_devices
->device_list_mutex
);
2469 mutex_lock(&fs_devices
->device_list_mutex
);
2470 list_splice_init_rcu(&fs_devices
->devices
, &seed_devices
->devices
,
2472 list_for_each_entry(device
, &seed_devices
->devices
, dev_list
)
2473 device
->fs_devices
= seed_devices
;
2475 fs_devices
->seeding
= false;
2476 fs_devices
->num_devices
= 0;
2477 fs_devices
->open_devices
= 0;
2478 fs_devices
->missing_devices
= 0;
2479 fs_devices
->rotating
= false;
2480 list_add(&seed_devices
->seed_list
, &fs_devices
->seed_list
);
2482 generate_random_uuid(fs_devices
->fsid
);
2483 memcpy(fs_devices
->metadata_uuid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2484 memcpy(disk_super
->fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
);
2485 mutex_unlock(&fs_devices
->device_list_mutex
);
2487 super_flags
= btrfs_super_flags(disk_super
) &
2488 ~BTRFS_SUPER_FLAG_SEEDING
;
2489 btrfs_set_super_flags(disk_super
, super_flags
);
2495 * Store the expected generation for seed devices in device items.
2497 static int btrfs_finish_sprout(struct btrfs_trans_handle
*trans
)
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
];
2511 path
= btrfs_alloc_path();
2515 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2517 key
.type
= BTRFS_DEV_ITEM_KEY
;
2520 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2524 leaf
= path
->nodes
[0];
2526 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
2527 ret
= btrfs_next_leaf(root
, path
);
2532 leaf
= path
->nodes
[0];
2533 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2534 btrfs_release_path(path
);
2538 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2539 if (key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
||
2540 key
.type
!= BTRFS_DEV_ITEM_KEY
)
2543 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2544 struct btrfs_dev_item
);
2545 devid
= btrfs_device_id(leaf
, dev_item
);
2546 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
2548 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
2550 device
= btrfs_find_device(fs_info
->fs_devices
, devid
, dev_uuid
,
2552 BUG_ON(!device
); /* Logic error */
2554 if (device
->fs_devices
->seeding
) {
2555 btrfs_set_device_generation(leaf
, dev_item
,
2556 device
->generation
);
2557 btrfs_mark_buffer_dirty(leaf
);
2565 btrfs_free_path(path
);
2569 int btrfs_init_new_device(struct btrfs_fs_info
*fs_info
, const char *device_path
)
2571 struct btrfs_root
*root
= fs_info
->dev_root
;
2572 struct request_queue
*q
;
2573 struct btrfs_trans_handle
*trans
;
2574 struct btrfs_device
*device
;
2575 struct block_device
*bdev
;
2576 struct super_block
*sb
= fs_info
->sb
;
2577 struct rcu_string
*name
;
2578 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
2579 u64 orig_super_total_bytes
;
2580 u64 orig_super_num_devices
;
2581 int seeding_dev
= 0;
2583 bool locked
= false;
2585 if (sb_rdonly(sb
) && !fs_devices
->seeding
)
2588 bdev
= blkdev_get_by_path(device_path
, FMODE_WRITE
| FMODE_EXCL
,
2589 fs_info
->bdev_holder
);
2591 return PTR_ERR(bdev
);
2593 if (!btrfs_check_device_zone_type(fs_info
, bdev
)) {
2598 if (fs_devices
->seeding
) {
2600 down_write(&sb
->s_umount
);
2601 mutex_lock(&uuid_mutex
);
2605 sync_blockdev(bdev
);
2608 list_for_each_entry_rcu(device
, &fs_devices
->devices
, dev_list
) {
2609 if (device
->bdev
== bdev
) {
2617 device
= btrfs_alloc_device(fs_info
, NULL
, NULL
);
2618 if (IS_ERR(device
)) {
2619 /* we can safely leave the fs_devices entry around */
2620 ret
= PTR_ERR(device
);
2624 name
= rcu_string_strdup(device_path
, GFP_KERNEL
);
2627 goto error_free_device
;
2629 rcu_assign_pointer(device
->name
, name
);
2631 device
->fs_info
= fs_info
;
2632 device
->bdev
= bdev
;
2634 ret
= btrfs_get_dev_zone_info(device
);
2636 goto error_free_device
;
2638 trans
= btrfs_start_transaction(root
, 0);
2639 if (IS_ERR(trans
)) {
2640 ret
= PTR_ERR(trans
);
2641 goto error_free_zone
;
2644 q
= bdev_get_queue(bdev
);
2645 set_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
);
2646 device
->generation
= trans
->transid
;
2647 device
->io_width
= fs_info
->sectorsize
;
2648 device
->io_align
= fs_info
->sectorsize
;
2649 device
->sector_size
= fs_info
->sectorsize
;
2650 device
->total_bytes
= round_down(i_size_read(bdev
->bd_inode
),
2651 fs_info
->sectorsize
);
2652 device
->disk_total_bytes
= device
->total_bytes
;
2653 device
->commit_total_bytes
= device
->total_bytes
;
2654 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &device
->dev_state
);
2655 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
);
2656 device
->mode
= FMODE_EXCL
;
2657 device
->dev_stats_valid
= 1;
2658 set_blocksize(device
->bdev
, BTRFS_BDEV_BLOCKSIZE
);
2661 btrfs_clear_sb_rdonly(sb
);
2662 ret
= btrfs_prepare_sprout(fs_info
);
2664 btrfs_abort_transaction(trans
, ret
);
2669 device
->fs_devices
= fs_devices
;
2671 mutex_lock(&fs_devices
->device_list_mutex
);
2672 mutex_lock(&fs_info
->chunk_mutex
);
2673 list_add_rcu(&device
->dev_list
, &fs_devices
->devices
);
2674 list_add(&device
->dev_alloc_list
, &fs_devices
->alloc_list
);
2675 fs_devices
->num_devices
++;
2676 fs_devices
->open_devices
++;
2677 fs_devices
->rw_devices
++;
2678 fs_devices
->total_devices
++;
2679 fs_devices
->total_rw_bytes
+= device
->total_bytes
;
2681 atomic64_add(device
->total_bytes
, &fs_info
->free_chunk_space
);
2683 if (!blk_queue_nonrot(q
))
2684 fs_devices
->rotating
= true;
2686 orig_super_total_bytes
= btrfs_super_total_bytes(fs_info
->super_copy
);
2687 btrfs_set_super_total_bytes(fs_info
->super_copy
,
2688 round_down(orig_super_total_bytes
+ device
->total_bytes
,
2689 fs_info
->sectorsize
));
2691 orig_super_num_devices
= btrfs_super_num_devices(fs_info
->super_copy
);
2692 btrfs_set_super_num_devices(fs_info
->super_copy
,
2693 orig_super_num_devices
+ 1);
2696 * we've got more storage, clear any full flags on the space
2699 btrfs_clear_space_info_full(fs_info
);
2701 mutex_unlock(&fs_info
->chunk_mutex
);
2703 /* Add sysfs device entry */
2704 btrfs_sysfs_add_device(device
);
2706 mutex_unlock(&fs_devices
->device_list_mutex
);
2709 mutex_lock(&fs_info
->chunk_mutex
);
2710 ret
= init_first_rw_device(trans
);
2711 mutex_unlock(&fs_info
->chunk_mutex
);
2713 btrfs_abort_transaction(trans
, ret
);
2718 ret
= btrfs_add_dev_item(trans
, device
);
2720 btrfs_abort_transaction(trans
, ret
);
2725 ret
= btrfs_finish_sprout(trans
);
2727 btrfs_abort_transaction(trans
, ret
);
2732 * fs_devices now represents the newly sprouted filesystem and
2733 * its fsid has been changed by btrfs_prepare_sprout
2735 btrfs_sysfs_update_sprout_fsid(fs_devices
);
2738 ret
= btrfs_commit_transaction(trans
);
2741 mutex_unlock(&uuid_mutex
);
2742 up_write(&sb
->s_umount
);
2745 if (ret
) /* transaction commit */
2748 ret
= btrfs_relocate_sys_chunks(fs_info
);
2750 btrfs_handle_fs_error(fs_info
, ret
,
2751 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2752 trans
= btrfs_attach_transaction(root
);
2753 if (IS_ERR(trans
)) {
2754 if (PTR_ERR(trans
) == -ENOENT
)
2756 ret
= PTR_ERR(trans
);
2760 ret
= btrfs_commit_transaction(trans
);
2764 * Now that we have written a new super block to this device, check all
2765 * other fs_devices list if device_path alienates any other scanned
2767 * We can ignore the return value as it typically returns -EINVAL and
2768 * only succeeds if the device was an alien.
2770 btrfs_forget_devices(device_path
);
2772 /* Update ctime/mtime for blkid or udev */
2773 update_dev_time(device_path
);
2778 btrfs_sysfs_remove_device(device
);
2779 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
2780 mutex_lock(&fs_info
->chunk_mutex
);
2781 list_del_rcu(&device
->dev_list
);
2782 list_del(&device
->dev_alloc_list
);
2783 fs_info
->fs_devices
->num_devices
--;
2784 fs_info
->fs_devices
->open_devices
--;
2785 fs_info
->fs_devices
->rw_devices
--;
2786 fs_info
->fs_devices
->total_devices
--;
2787 fs_info
->fs_devices
->total_rw_bytes
-= device
->total_bytes
;
2788 atomic64_sub(device
->total_bytes
, &fs_info
->free_chunk_space
);
2789 btrfs_set_super_total_bytes(fs_info
->super_copy
,
2790 orig_super_total_bytes
);
2791 btrfs_set_super_num_devices(fs_info
->super_copy
,
2792 orig_super_num_devices
);
2793 mutex_unlock(&fs_info
->chunk_mutex
);
2794 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
2797 btrfs_set_sb_rdonly(sb
);
2799 btrfs_end_transaction(trans
);
2801 btrfs_destroy_dev_zone_info(device
);
2803 btrfs_free_device(device
);
2805 blkdev_put(bdev
, FMODE_EXCL
);
2807 mutex_unlock(&uuid_mutex
);
2808 up_write(&sb
->s_umount
);
2813 static noinline
int btrfs_update_device(struct btrfs_trans_handle
*trans
,
2814 struct btrfs_device
*device
)
2817 struct btrfs_path
*path
;
2818 struct btrfs_root
*root
= device
->fs_info
->chunk_root
;
2819 struct btrfs_dev_item
*dev_item
;
2820 struct extent_buffer
*leaf
;
2821 struct btrfs_key key
;
2823 path
= btrfs_alloc_path();
2827 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
2828 key
.type
= BTRFS_DEV_ITEM_KEY
;
2829 key
.offset
= device
->devid
;
2831 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2840 leaf
= path
->nodes
[0];
2841 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
2843 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
2844 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
2845 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
2846 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
2847 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
2848 btrfs_set_device_total_bytes(leaf
, dev_item
,
2849 btrfs_device_get_disk_total_bytes(device
));
2850 btrfs_set_device_bytes_used(leaf
, dev_item
,
2851 btrfs_device_get_bytes_used(device
));
2852 btrfs_mark_buffer_dirty(leaf
);
2855 btrfs_free_path(path
);
2859 int btrfs_grow_device(struct btrfs_trans_handle
*trans
,
2860 struct btrfs_device
*device
, u64 new_size
)
2862 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
2863 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
2867 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
))
2870 new_size
= round_down(new_size
, fs_info
->sectorsize
);
2872 mutex_lock(&fs_info
->chunk_mutex
);
2873 old_total
= btrfs_super_total_bytes(super_copy
);
2874 diff
= round_down(new_size
- device
->total_bytes
, fs_info
->sectorsize
);
2876 if (new_size
<= device
->total_bytes
||
2877 test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
)) {
2878 mutex_unlock(&fs_info
->chunk_mutex
);
2882 btrfs_set_super_total_bytes(super_copy
,
2883 round_down(old_total
+ diff
, fs_info
->sectorsize
));
2884 device
->fs_devices
->total_rw_bytes
+= diff
;
2886 btrfs_device_set_total_bytes(device
, new_size
);
2887 btrfs_device_set_disk_total_bytes(device
, new_size
);
2888 btrfs_clear_space_info_full(device
->fs_info
);
2889 if (list_empty(&device
->post_commit_list
))
2890 list_add_tail(&device
->post_commit_list
,
2891 &trans
->transaction
->dev_update_list
);
2892 mutex_unlock(&fs_info
->chunk_mutex
);
2894 return btrfs_update_device(trans
, device
);
2897 static int btrfs_free_chunk(struct btrfs_trans_handle
*trans
, u64 chunk_offset
)
2899 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
2900 struct btrfs_root
*root
= fs_info
->chunk_root
;
2902 struct btrfs_path
*path
;
2903 struct btrfs_key key
;
2905 path
= btrfs_alloc_path();
2909 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
2910 key
.offset
= chunk_offset
;
2911 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
2913 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2916 else if (ret
> 0) { /* Logic error or corruption */
2917 btrfs_handle_fs_error(fs_info
, -ENOENT
,
2918 "Failed lookup while freeing chunk.");
2923 ret
= btrfs_del_item(trans
, root
, path
);
2925 btrfs_handle_fs_error(fs_info
, ret
,
2926 "Failed to delete chunk item.");
2928 btrfs_free_path(path
);
2932 static int btrfs_del_sys_chunk(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
2934 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
2935 struct btrfs_disk_key
*disk_key
;
2936 struct btrfs_chunk
*chunk
;
2943 struct btrfs_key key
;
2945 mutex_lock(&fs_info
->chunk_mutex
);
2946 array_size
= btrfs_super_sys_array_size(super_copy
);
2948 ptr
= super_copy
->sys_chunk_array
;
2951 while (cur
< array_size
) {
2952 disk_key
= (struct btrfs_disk_key
*)ptr
;
2953 btrfs_disk_key_to_cpu(&key
, disk_key
);
2955 len
= sizeof(*disk_key
);
2957 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
2958 chunk
= (struct btrfs_chunk
*)(ptr
+ len
);
2959 num_stripes
= btrfs_stack_chunk_num_stripes(chunk
);
2960 len
+= btrfs_chunk_item_size(num_stripes
);
2965 if (key
.objectid
== BTRFS_FIRST_CHUNK_TREE_OBJECTID
&&
2966 key
.offset
== chunk_offset
) {
2967 memmove(ptr
, ptr
+ len
, array_size
- (cur
+ len
));
2969 btrfs_set_super_sys_array_size(super_copy
, array_size
);
2975 mutex_unlock(&fs_info
->chunk_mutex
);
2980 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
2981 * @logical: Logical block offset in bytes.
2982 * @length: Length of extent in bytes.
2984 * Return: Chunk mapping or ERR_PTR.
2986 struct extent_map
*btrfs_get_chunk_map(struct btrfs_fs_info
*fs_info
,
2987 u64 logical
, u64 length
)
2989 struct extent_map_tree
*em_tree
;
2990 struct extent_map
*em
;
2992 em_tree
= &fs_info
->mapping_tree
;
2993 read_lock(&em_tree
->lock
);
2994 em
= lookup_extent_mapping(em_tree
, logical
, length
);
2995 read_unlock(&em_tree
->lock
);
2998 btrfs_crit(fs_info
, "unable to find logical %llu length %llu",
3000 return ERR_PTR(-EINVAL
);
3003 if (em
->start
> logical
|| em
->start
+ em
->len
< logical
) {
3005 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
3006 logical
, length
, em
->start
, em
->start
+ em
->len
);
3007 free_extent_map(em
);
3008 return ERR_PTR(-EINVAL
);
3011 /* callers are responsible for dropping em's ref. */
3015 int btrfs_remove_chunk(struct btrfs_trans_handle
*trans
, u64 chunk_offset
)
3017 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
3018 struct extent_map
*em
;
3019 struct map_lookup
*map
;
3020 u64 dev_extent_len
= 0;
3022 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
3024 em
= btrfs_get_chunk_map(fs_info
, chunk_offset
, 1);
3027 * This is a logic error, but we don't want to just rely on the
3028 * user having built with ASSERT enabled, so if ASSERT doesn't
3029 * do anything we still error out.
3034 map
= em
->map_lookup
;
3035 mutex_lock(&fs_info
->chunk_mutex
);
3036 check_system_chunk(trans
, map
->type
);
3037 mutex_unlock(&fs_info
->chunk_mutex
);
3040 * Take the device list mutex to prevent races with the final phase of
3041 * a device replace operation that replaces the device object associated
3042 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
3044 mutex_lock(&fs_devices
->device_list_mutex
);
3045 for (i
= 0; i
< map
->num_stripes
; i
++) {
3046 struct btrfs_device
*device
= map
->stripes
[i
].dev
;
3047 ret
= btrfs_free_dev_extent(trans
, device
,
3048 map
->stripes
[i
].physical
,
3051 mutex_unlock(&fs_devices
->device_list_mutex
);
3052 btrfs_abort_transaction(trans
, ret
);
3056 if (device
->bytes_used
> 0) {
3057 mutex_lock(&fs_info
->chunk_mutex
);
3058 btrfs_device_set_bytes_used(device
,
3059 device
->bytes_used
- dev_extent_len
);
3060 atomic64_add(dev_extent_len
, &fs_info
->free_chunk_space
);
3061 btrfs_clear_space_info_full(fs_info
);
3062 mutex_unlock(&fs_info
->chunk_mutex
);
3065 ret
= btrfs_update_device(trans
, device
);
3067 mutex_unlock(&fs_devices
->device_list_mutex
);
3068 btrfs_abort_transaction(trans
, ret
);
3072 mutex_unlock(&fs_devices
->device_list_mutex
);
3074 ret
= btrfs_free_chunk(trans
, chunk_offset
);
3076 btrfs_abort_transaction(trans
, ret
);
3080 trace_btrfs_chunk_free(fs_info
, map
, chunk_offset
, em
->len
);
3082 if (map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3083 ret
= btrfs_del_sys_chunk(fs_info
, chunk_offset
);
3085 btrfs_abort_transaction(trans
, ret
);
3090 ret
= btrfs_remove_block_group(trans
, chunk_offset
, em
);
3092 btrfs_abort_transaction(trans
, ret
);
3098 free_extent_map(em
);
3102 int btrfs_relocate_chunk(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
3104 struct btrfs_root
*root
= fs_info
->chunk_root
;
3105 struct btrfs_trans_handle
*trans
;
3106 struct btrfs_block_group
*block_group
;
3111 * Prevent races with automatic removal of unused block groups.
3112 * After we relocate and before we remove the chunk with offset
3113 * chunk_offset, automatic removal of the block group can kick in,
3114 * resulting in a failure when calling btrfs_remove_chunk() below.
3116 * Make sure to acquire this mutex before doing a tree search (dev
3117 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3118 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3119 * we release the path used to search the chunk/dev tree and before
3120 * the current task acquires this mutex and calls us.
3122 lockdep_assert_held(&fs_info
->reclaim_bgs_lock
);
3124 /* step one, relocate all the extents inside this chunk */
3125 btrfs_scrub_pause(fs_info
);
3126 ret
= btrfs_relocate_block_group(fs_info
, chunk_offset
);
3127 btrfs_scrub_continue(fs_info
);
3131 block_group
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3134 btrfs_discard_cancel_work(&fs_info
->discard_ctl
, block_group
);
3135 length
= block_group
->length
;
3136 btrfs_put_block_group(block_group
);
3139 * On a zoned file system, discard the whole block group, this will
3140 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3141 * resetting the zone fails, don't treat it as a fatal problem from the
3142 * filesystem's point of view.
3144 if (btrfs_is_zoned(fs_info
)) {
3145 ret
= btrfs_discard_extent(fs_info
, chunk_offset
, length
, NULL
);
3148 "failed to reset zone %llu after relocation",
3152 trans
= btrfs_start_trans_remove_block_group(root
->fs_info
,
3154 if (IS_ERR(trans
)) {
3155 ret
= PTR_ERR(trans
);
3156 btrfs_handle_fs_error(root
->fs_info
, ret
, NULL
);
3161 * step two, delete the device extents and the
3162 * chunk tree entries
3164 ret
= btrfs_remove_chunk(trans
, chunk_offset
);
3165 btrfs_end_transaction(trans
);
3169 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info
*fs_info
)
3171 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
3172 struct btrfs_path
*path
;
3173 struct extent_buffer
*leaf
;
3174 struct btrfs_chunk
*chunk
;
3175 struct btrfs_key key
;
3176 struct btrfs_key found_key
;
3178 bool retried
= false;
3182 path
= btrfs_alloc_path();
3187 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3188 key
.offset
= (u64
)-1;
3189 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3192 mutex_lock(&fs_info
->reclaim_bgs_lock
);
3193 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
3195 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
3198 BUG_ON(ret
== 0); /* Corruption */
3200 ret
= btrfs_previous_item(chunk_root
, path
, key
.objectid
,
3203 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
3209 leaf
= path
->nodes
[0];
3210 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3212 chunk
= btrfs_item_ptr(leaf
, path
->slots
[0],
3213 struct btrfs_chunk
);
3214 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3215 btrfs_release_path(path
);
3217 if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
3218 ret
= btrfs_relocate_chunk(fs_info
, found_key
.offset
);
3224 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
3226 if (found_key
.offset
== 0)
3228 key
.offset
= found_key
.offset
- 1;
3231 if (failed
&& !retried
) {
3235 } else if (WARN_ON(failed
&& retried
)) {
3239 btrfs_free_path(path
);
3244 * return 1 : allocate a data chunk successfully,
3245 * return <0: errors during allocating a data chunk,
3246 * return 0 : no need to allocate a data chunk.
3248 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info
*fs_info
,
3251 struct btrfs_block_group
*cache
;
3255 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3257 chunk_type
= cache
->flags
;
3258 btrfs_put_block_group(cache
);
3260 if (!(chunk_type
& BTRFS_BLOCK_GROUP_DATA
))
3263 spin_lock(&fs_info
->data_sinfo
->lock
);
3264 bytes_used
= fs_info
->data_sinfo
->bytes_used
;
3265 spin_unlock(&fs_info
->data_sinfo
->lock
);
3268 struct btrfs_trans_handle
*trans
;
3271 trans
= btrfs_join_transaction(fs_info
->tree_root
);
3273 return PTR_ERR(trans
);
3275 ret
= btrfs_force_chunk_alloc(trans
, BTRFS_BLOCK_GROUP_DATA
);
3276 btrfs_end_transaction(trans
);
3285 static int insert_balance_item(struct btrfs_fs_info
*fs_info
,
3286 struct btrfs_balance_control
*bctl
)
3288 struct btrfs_root
*root
= fs_info
->tree_root
;
3289 struct btrfs_trans_handle
*trans
;
3290 struct btrfs_balance_item
*item
;
3291 struct btrfs_disk_balance_args disk_bargs
;
3292 struct btrfs_path
*path
;
3293 struct extent_buffer
*leaf
;
3294 struct btrfs_key key
;
3297 path
= btrfs_alloc_path();
3301 trans
= btrfs_start_transaction(root
, 0);
3302 if (IS_ERR(trans
)) {
3303 btrfs_free_path(path
);
3304 return PTR_ERR(trans
);
3307 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3308 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3311 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
3316 leaf
= path
->nodes
[0];
3317 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
3319 memzero_extent_buffer(leaf
, (unsigned long)item
, sizeof(*item
));
3321 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->data
);
3322 btrfs_set_balance_data(leaf
, item
, &disk_bargs
);
3323 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->meta
);
3324 btrfs_set_balance_meta(leaf
, item
, &disk_bargs
);
3325 btrfs_cpu_balance_args_to_disk(&disk_bargs
, &bctl
->sys
);
3326 btrfs_set_balance_sys(leaf
, item
, &disk_bargs
);
3328 btrfs_set_balance_flags(leaf
, item
, bctl
->flags
);
3330 btrfs_mark_buffer_dirty(leaf
);
3332 btrfs_free_path(path
);
3333 err
= btrfs_commit_transaction(trans
);
3339 static int del_balance_item(struct btrfs_fs_info
*fs_info
)
3341 struct btrfs_root
*root
= fs_info
->tree_root
;
3342 struct btrfs_trans_handle
*trans
;
3343 struct btrfs_path
*path
;
3344 struct btrfs_key key
;
3347 path
= btrfs_alloc_path();
3351 trans
= btrfs_start_transaction_fallback_global_rsv(root
, 0);
3352 if (IS_ERR(trans
)) {
3353 btrfs_free_path(path
);
3354 return PTR_ERR(trans
);
3357 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
3358 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
3361 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3369 ret
= btrfs_del_item(trans
, root
, path
);
3371 btrfs_free_path(path
);
3372 err
= btrfs_commit_transaction(trans
);
3379 * This is a heuristic used to reduce the number of chunks balanced on
3380 * resume after balance was interrupted.
3382 static void update_balance_args(struct btrfs_balance_control
*bctl
)
3385 * Turn on soft mode for chunk types that were being converted.
3387 if (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3388 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3389 if (bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3390 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3391 if (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3392 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_SOFT
;
3395 * Turn on usage filter if is not already used. The idea is
3396 * that chunks that we have already balanced should be
3397 * reasonably full. Don't do it for chunks that are being
3398 * converted - that will keep us from relocating unconverted
3399 * (albeit full) chunks.
3401 if (!(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3402 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3403 !(bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3404 bctl
->data
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3405 bctl
->data
.usage
= 90;
3407 if (!(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3408 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3409 !(bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3410 bctl
->sys
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3411 bctl
->sys
.usage
= 90;
3413 if (!(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3414 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3415 !(bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
)) {
3416 bctl
->meta
.flags
|= BTRFS_BALANCE_ARGS_USAGE
;
3417 bctl
->meta
.usage
= 90;
3422 * Clear the balance status in fs_info and delete the balance item from disk.
3424 static void reset_balance_state(struct btrfs_fs_info
*fs_info
)
3426 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3429 BUG_ON(!fs_info
->balance_ctl
);
3431 spin_lock(&fs_info
->balance_lock
);
3432 fs_info
->balance_ctl
= NULL
;
3433 spin_unlock(&fs_info
->balance_lock
);
3436 ret
= del_balance_item(fs_info
);
3438 btrfs_handle_fs_error(fs_info
, ret
, NULL
);
3442 * Balance filters. Return 1 if chunk should be filtered out
3443 * (should not be balanced).
3445 static int chunk_profiles_filter(u64 chunk_type
,
3446 struct btrfs_balance_args
*bargs
)
3448 chunk_type
= chunk_to_extended(chunk_type
) &
3449 BTRFS_EXTENDED_PROFILE_MASK
;
3451 if (bargs
->profiles
& chunk_type
)
3457 static int chunk_usage_range_filter(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
,
3458 struct btrfs_balance_args
*bargs
)
3460 struct btrfs_block_group
*cache
;
3462 u64 user_thresh_min
;
3463 u64 user_thresh_max
;
3466 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3467 chunk_used
= cache
->used
;
3469 if (bargs
->usage_min
== 0)
3470 user_thresh_min
= 0;
3472 user_thresh_min
= div_factor_fine(cache
->length
,
3475 if (bargs
->usage_max
== 0)
3476 user_thresh_max
= 1;
3477 else if (bargs
->usage_max
> 100)
3478 user_thresh_max
= cache
->length
;
3480 user_thresh_max
= div_factor_fine(cache
->length
,
3483 if (user_thresh_min
<= chunk_used
&& chunk_used
< user_thresh_max
)
3486 btrfs_put_block_group(cache
);
3490 static int chunk_usage_filter(struct btrfs_fs_info
*fs_info
,
3491 u64 chunk_offset
, struct btrfs_balance_args
*bargs
)
3493 struct btrfs_block_group
*cache
;
3494 u64 chunk_used
, user_thresh
;
3497 cache
= btrfs_lookup_block_group(fs_info
, chunk_offset
);
3498 chunk_used
= cache
->used
;
3500 if (bargs
->usage_min
== 0)
3502 else if (bargs
->usage
> 100)
3503 user_thresh
= cache
->length
;
3505 user_thresh
= div_factor_fine(cache
->length
, bargs
->usage
);
3507 if (chunk_used
< user_thresh
)
3510 btrfs_put_block_group(cache
);
3514 static int chunk_devid_filter(struct extent_buffer
*leaf
,
3515 struct btrfs_chunk
*chunk
,
3516 struct btrfs_balance_args
*bargs
)
3518 struct btrfs_stripe
*stripe
;
3519 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3522 for (i
= 0; i
< num_stripes
; i
++) {
3523 stripe
= btrfs_stripe_nr(chunk
, i
);
3524 if (btrfs_stripe_devid(leaf
, stripe
) == bargs
->devid
)
3531 static u64
calc_data_stripes(u64 type
, int num_stripes
)
3533 const int index
= btrfs_bg_flags_to_raid_index(type
);
3534 const int ncopies
= btrfs_raid_array
[index
].ncopies
;
3535 const int nparity
= btrfs_raid_array
[index
].nparity
;
3538 return num_stripes
- nparity
;
3540 return num_stripes
/ ncopies
;
3543 /* [pstart, pend) */
3544 static int chunk_drange_filter(struct extent_buffer
*leaf
,
3545 struct btrfs_chunk
*chunk
,
3546 struct btrfs_balance_args
*bargs
)
3548 struct btrfs_stripe
*stripe
;
3549 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3556 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
))
3559 type
= btrfs_chunk_type(leaf
, chunk
);
3560 factor
= calc_data_stripes(type
, num_stripes
);
3562 for (i
= 0; i
< num_stripes
; i
++) {
3563 stripe
= btrfs_stripe_nr(chunk
, i
);
3564 if (btrfs_stripe_devid(leaf
, stripe
) != bargs
->devid
)
3567 stripe_offset
= btrfs_stripe_offset(leaf
, stripe
);
3568 stripe_length
= btrfs_chunk_length(leaf
, chunk
);
3569 stripe_length
= div_u64(stripe_length
, factor
);
3571 if (stripe_offset
< bargs
->pend
&&
3572 stripe_offset
+ stripe_length
> bargs
->pstart
)
3579 /* [vstart, vend) */
3580 static int chunk_vrange_filter(struct extent_buffer
*leaf
,
3581 struct btrfs_chunk
*chunk
,
3583 struct btrfs_balance_args
*bargs
)
3585 if (chunk_offset
< bargs
->vend
&&
3586 chunk_offset
+ btrfs_chunk_length(leaf
, chunk
) > bargs
->vstart
)
3587 /* at least part of the chunk is inside this vrange */
3593 static int chunk_stripes_range_filter(struct extent_buffer
*leaf
,
3594 struct btrfs_chunk
*chunk
,
3595 struct btrfs_balance_args
*bargs
)
3597 int num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
3599 if (bargs
->stripes_min
<= num_stripes
3600 && num_stripes
<= bargs
->stripes_max
)
3606 static int chunk_soft_convert_filter(u64 chunk_type
,
3607 struct btrfs_balance_args
*bargs
)
3609 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
3612 chunk_type
= chunk_to_extended(chunk_type
) &
3613 BTRFS_EXTENDED_PROFILE_MASK
;
3615 if (bargs
->target
== chunk_type
)
3621 static int should_balance_chunk(struct extent_buffer
*leaf
,
3622 struct btrfs_chunk
*chunk
, u64 chunk_offset
)
3624 struct btrfs_fs_info
*fs_info
= leaf
->fs_info
;
3625 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3626 struct btrfs_balance_args
*bargs
= NULL
;
3627 u64 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3630 if (!((chunk_type
& BTRFS_BLOCK_GROUP_TYPE_MASK
) &
3631 (bctl
->flags
& BTRFS_BALANCE_TYPE_MASK
))) {
3635 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3636 bargs
= &bctl
->data
;
3637 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3639 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3640 bargs
= &bctl
->meta
;
3642 /* profiles filter */
3643 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_PROFILES
) &&
3644 chunk_profiles_filter(chunk_type
, bargs
)) {
3649 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE
) &&
3650 chunk_usage_filter(fs_info
, chunk_offset
, bargs
)) {
3652 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
) &&
3653 chunk_usage_range_filter(fs_info
, chunk_offset
, bargs
)) {
3658 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DEVID
) &&
3659 chunk_devid_filter(leaf
, chunk
, bargs
)) {
3663 /* drange filter, makes sense only with devid filter */
3664 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_DRANGE
) &&
3665 chunk_drange_filter(leaf
, chunk
, bargs
)) {
3670 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_VRANGE
) &&
3671 chunk_vrange_filter(leaf
, chunk
, chunk_offset
, bargs
)) {
3675 /* stripes filter */
3676 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_STRIPES_RANGE
) &&
3677 chunk_stripes_range_filter(leaf
, chunk
, bargs
)) {
3681 /* soft profile changing mode */
3682 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_SOFT
) &&
3683 chunk_soft_convert_filter(chunk_type
, bargs
)) {
3688 * limited by count, must be the last filter
3690 if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT
)) {
3691 if (bargs
->limit
== 0)
3695 } else if ((bargs
->flags
& BTRFS_BALANCE_ARGS_LIMIT_RANGE
)) {
3697 * Same logic as the 'limit' filter; the minimum cannot be
3698 * determined here because we do not have the global information
3699 * about the count of all chunks that satisfy the filters.
3701 if (bargs
->limit_max
== 0)
3710 static int __btrfs_balance(struct btrfs_fs_info
*fs_info
)
3712 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
3713 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
3715 struct btrfs_chunk
*chunk
;
3716 struct btrfs_path
*path
= NULL
;
3717 struct btrfs_key key
;
3718 struct btrfs_key found_key
;
3719 struct extent_buffer
*leaf
;
3722 int enospc_errors
= 0;
3723 bool counting
= true;
3724 /* The single value limit and min/max limits use the same bytes in the */
3725 u64 limit_data
= bctl
->data
.limit
;
3726 u64 limit_meta
= bctl
->meta
.limit
;
3727 u64 limit_sys
= bctl
->sys
.limit
;
3731 int chunk_reserved
= 0;
3733 path
= btrfs_alloc_path();
3739 /* zero out stat counters */
3740 spin_lock(&fs_info
->balance_lock
);
3741 memset(&bctl
->stat
, 0, sizeof(bctl
->stat
));
3742 spin_unlock(&fs_info
->balance_lock
);
3746 * The single value limit and min/max limits use the same bytes
3749 bctl
->data
.limit
= limit_data
;
3750 bctl
->meta
.limit
= limit_meta
;
3751 bctl
->sys
.limit
= limit_sys
;
3753 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
3754 key
.offset
= (u64
)-1;
3755 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
3758 if ((!counting
&& atomic_read(&fs_info
->balance_pause_req
)) ||
3759 atomic_read(&fs_info
->balance_cancel_req
)) {
3764 mutex_lock(&fs_info
->reclaim_bgs_lock
);
3765 ret
= btrfs_search_slot(NULL
, chunk_root
, &key
, path
, 0, 0);
3767 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
3772 * this shouldn't happen, it means the last relocate
3776 BUG(); /* FIXME break ? */
3778 ret
= btrfs_previous_item(chunk_root
, path
, 0,
3779 BTRFS_CHUNK_ITEM_KEY
);
3781 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
3786 leaf
= path
->nodes
[0];
3787 slot
= path
->slots
[0];
3788 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3790 if (found_key
.objectid
!= key
.objectid
) {
3791 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
3795 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
3796 chunk_type
= btrfs_chunk_type(leaf
, chunk
);
3799 spin_lock(&fs_info
->balance_lock
);
3800 bctl
->stat
.considered
++;
3801 spin_unlock(&fs_info
->balance_lock
);
3804 ret
= should_balance_chunk(leaf
, chunk
, found_key
.offset
);
3806 btrfs_release_path(path
);
3808 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
3813 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
3814 spin_lock(&fs_info
->balance_lock
);
3815 bctl
->stat
.expected
++;
3816 spin_unlock(&fs_info
->balance_lock
);
3818 if (chunk_type
& BTRFS_BLOCK_GROUP_DATA
)
3820 else if (chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
)
3822 else if (chunk_type
& BTRFS_BLOCK_GROUP_METADATA
)
3829 * Apply limit_min filter, no need to check if the LIMITS
3830 * filter is used, limit_min is 0 by default
3832 if (((chunk_type
& BTRFS_BLOCK_GROUP_DATA
) &&
3833 count_data
< bctl
->data
.limit_min
)
3834 || ((chunk_type
& BTRFS_BLOCK_GROUP_METADATA
) &&
3835 count_meta
< bctl
->meta
.limit_min
)
3836 || ((chunk_type
& BTRFS_BLOCK_GROUP_SYSTEM
) &&
3837 count_sys
< bctl
->sys
.limit_min
)) {
3838 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
3842 if (!chunk_reserved
) {
3844 * We may be relocating the only data chunk we have,
3845 * which could potentially end up with losing data's
3846 * raid profile, so lets allocate an empty one in
3849 ret
= btrfs_may_alloc_data_chunk(fs_info
,
3852 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
3854 } else if (ret
== 1) {
3859 ret
= btrfs_relocate_chunk(fs_info
, found_key
.offset
);
3860 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
3861 if (ret
== -ENOSPC
) {
3863 } else if (ret
== -ETXTBSY
) {
3865 "skipping relocation of block group %llu due to active swapfile",
3871 spin_lock(&fs_info
->balance_lock
);
3872 bctl
->stat
.completed
++;
3873 spin_unlock(&fs_info
->balance_lock
);
3876 if (found_key
.offset
== 0)
3878 key
.offset
= found_key
.offset
- 1;
3882 btrfs_release_path(path
);
3887 btrfs_free_path(path
);
3888 if (enospc_errors
) {
3889 btrfs_info(fs_info
, "%d enospc errors during balance",
3899 * alloc_profile_is_valid - see if a given profile is valid and reduced
3900 * @flags: profile to validate
3901 * @extended: if true @flags is treated as an extended profile
3903 static int alloc_profile_is_valid(u64 flags
, int extended
)
3905 u64 mask
= (extended
? BTRFS_EXTENDED_PROFILE_MASK
:
3906 BTRFS_BLOCK_GROUP_PROFILE_MASK
);
3908 flags
&= ~BTRFS_BLOCK_GROUP_TYPE_MASK
;
3910 /* 1) check that all other bits are zeroed */
3914 /* 2) see if profile is reduced */
3916 return !extended
; /* "0" is valid for usual profiles */
3918 return has_single_bit_set(flags
);
3921 static inline int balance_need_close(struct btrfs_fs_info
*fs_info
)
3923 /* cancel requested || normal exit path */
3924 return atomic_read(&fs_info
->balance_cancel_req
) ||
3925 (atomic_read(&fs_info
->balance_pause_req
) == 0 &&
3926 atomic_read(&fs_info
->balance_cancel_req
) == 0);
3930 * Validate target profile against allowed profiles and return true if it's OK.
3931 * Otherwise print the error message and return false.
3933 static inline int validate_convert_profile(struct btrfs_fs_info
*fs_info
,
3934 const struct btrfs_balance_args
*bargs
,
3935 u64 allowed
, const char *type
)
3937 if (!(bargs
->flags
& BTRFS_BALANCE_ARGS_CONVERT
))
3940 /* Profile is valid and does not have bits outside of the allowed set */
3941 if (alloc_profile_is_valid(bargs
->target
, 1) &&
3942 (bargs
->target
& ~allowed
) == 0)
3945 btrfs_err(fs_info
, "balance: invalid convert %s profile %s",
3946 type
, btrfs_bg_type_to_raid_name(bargs
->target
));
3951 * Fill @buf with textual description of balance filter flags @bargs, up to
3952 * @size_buf including the terminating null. The output may be trimmed if it
3953 * does not fit into the provided buffer.
3955 static void describe_balance_args(struct btrfs_balance_args
*bargs
, char *buf
,
3959 u32 size_bp
= size_buf
;
3961 u64 flags
= bargs
->flags
;
3962 char tmp_buf
[128] = {'\0'};
3967 #define CHECK_APPEND_NOARG(a) \
3969 ret = snprintf(bp, size_bp, (a)); \
3970 if (ret < 0 || ret >= size_bp) \
3971 goto out_overflow; \
3976 #define CHECK_APPEND_1ARG(a, v1) \
3978 ret = snprintf(bp, size_bp, (a), (v1)); \
3979 if (ret < 0 || ret >= size_bp) \
3980 goto out_overflow; \
3985 #define CHECK_APPEND_2ARG(a, v1, v2) \
3987 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
3988 if (ret < 0 || ret >= size_bp) \
3989 goto out_overflow; \
3994 if (flags
& BTRFS_BALANCE_ARGS_CONVERT
)
3995 CHECK_APPEND_1ARG("convert=%s,",
3996 btrfs_bg_type_to_raid_name(bargs
->target
));
3998 if (flags
& BTRFS_BALANCE_ARGS_SOFT
)
3999 CHECK_APPEND_NOARG("soft,");
4001 if (flags
& BTRFS_BALANCE_ARGS_PROFILES
) {
4002 btrfs_describe_block_groups(bargs
->profiles
, tmp_buf
,
4004 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf
);
4007 if (flags
& BTRFS_BALANCE_ARGS_USAGE
)
4008 CHECK_APPEND_1ARG("usage=%llu,", bargs
->usage
);
4010 if (flags
& BTRFS_BALANCE_ARGS_USAGE_RANGE
)
4011 CHECK_APPEND_2ARG("usage=%u..%u,",
4012 bargs
->usage_min
, bargs
->usage_max
);
4014 if (flags
& BTRFS_BALANCE_ARGS_DEVID
)
4015 CHECK_APPEND_1ARG("devid=%llu,", bargs
->devid
);
4017 if (flags
& BTRFS_BALANCE_ARGS_DRANGE
)
4018 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4019 bargs
->pstart
, bargs
->pend
);
4021 if (flags
& BTRFS_BALANCE_ARGS_VRANGE
)
4022 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4023 bargs
->vstart
, bargs
->vend
);
4025 if (flags
& BTRFS_BALANCE_ARGS_LIMIT
)
4026 CHECK_APPEND_1ARG("limit=%llu,", bargs
->limit
);
4028 if (flags
& BTRFS_BALANCE_ARGS_LIMIT_RANGE
)
4029 CHECK_APPEND_2ARG("limit=%u..%u,",
4030 bargs
->limit_min
, bargs
->limit_max
);
4032 if (flags
& BTRFS_BALANCE_ARGS_STRIPES_RANGE
)
4033 CHECK_APPEND_2ARG("stripes=%u..%u,",
4034 bargs
->stripes_min
, bargs
->stripes_max
);
4036 #undef CHECK_APPEND_2ARG
4037 #undef CHECK_APPEND_1ARG
4038 #undef CHECK_APPEND_NOARG
4042 if (size_bp
< size_buf
)
4043 buf
[size_buf
- size_bp
- 1] = '\0'; /* remove last , */
4048 static void describe_balance_start_or_resume(struct btrfs_fs_info
*fs_info
)
4050 u32 size_buf
= 1024;
4051 char tmp_buf
[192] = {'\0'};
4054 u32 size_bp
= size_buf
;
4056 struct btrfs_balance_control
*bctl
= fs_info
->balance_ctl
;
4058 buf
= kzalloc(size_buf
, GFP_KERNEL
);
4064 #define CHECK_APPEND_1ARG(a, v1) \
4066 ret = snprintf(bp, size_bp, (a), (v1)); \
4067 if (ret < 0 || ret >= size_bp) \
4068 goto out_overflow; \
4073 if (bctl
->flags
& BTRFS_BALANCE_FORCE
)
4074 CHECK_APPEND_1ARG("%s", "-f ");
4076 if (bctl
->flags
& BTRFS_BALANCE_DATA
) {
4077 describe_balance_args(&bctl
->data
, tmp_buf
, sizeof(tmp_buf
));
4078 CHECK_APPEND_1ARG("-d%s ", tmp_buf
);
4081 if (bctl
->flags
& BTRFS_BALANCE_METADATA
) {
4082 describe_balance_args(&bctl
->meta
, tmp_buf
, sizeof(tmp_buf
));
4083 CHECK_APPEND_1ARG("-m%s ", tmp_buf
);
4086 if (bctl
->flags
& BTRFS_BALANCE_SYSTEM
) {
4087 describe_balance_args(&bctl
->sys
, tmp_buf
, sizeof(tmp_buf
));
4088 CHECK_APPEND_1ARG("-s%s ", tmp_buf
);
4091 #undef CHECK_APPEND_1ARG
4095 if (size_bp
< size_buf
)
4096 buf
[size_buf
- size_bp
- 1] = '\0'; /* remove last " " */
4097 btrfs_info(fs_info
, "balance: %s %s",
4098 (bctl
->flags
& BTRFS_BALANCE_RESUME
) ?
4099 "resume" : "start", buf
);
4105 * Should be called with balance mutexe held
4107 int btrfs_balance(struct btrfs_fs_info
*fs_info
,
4108 struct btrfs_balance_control
*bctl
,
4109 struct btrfs_ioctl_balance_args
*bargs
)
4111 u64 meta_target
, data_target
;
4117 bool reducing_redundancy
;
4120 if (btrfs_fs_closing(fs_info
) ||
4121 atomic_read(&fs_info
->balance_pause_req
) ||
4122 btrfs_should_cancel_balance(fs_info
)) {
4127 allowed
= btrfs_super_incompat_flags(fs_info
->super_copy
);
4128 if (allowed
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
4132 * In case of mixed groups both data and meta should be picked,
4133 * and identical options should be given for both of them.
4135 allowed
= BTRFS_BALANCE_DATA
| BTRFS_BALANCE_METADATA
;
4136 if (mixed
&& (bctl
->flags
& allowed
)) {
4137 if (!(bctl
->flags
& BTRFS_BALANCE_DATA
) ||
4138 !(bctl
->flags
& BTRFS_BALANCE_METADATA
) ||
4139 memcmp(&bctl
->data
, &bctl
->meta
, sizeof(bctl
->data
))) {
4141 "balance: mixed groups data and metadata options must be the same");
4148 * rw_devices will not change at the moment, device add/delete/replace
4151 num_devices
= fs_info
->fs_devices
->rw_devices
;
4154 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4155 * special bit for it, to make it easier to distinguish. Thus we need
4156 * to set it manually, or balance would refuse the profile.
4158 allowed
= BTRFS_AVAIL_ALLOC_BIT_SINGLE
;
4159 for (i
= 0; i
< ARRAY_SIZE(btrfs_raid_array
); i
++)
4160 if (num_devices
>= btrfs_raid_array
[i
].devs_min
)
4161 allowed
|= btrfs_raid_array
[i
].bg_flag
;
4163 if (!validate_convert_profile(fs_info
, &bctl
->data
, allowed
, "data") ||
4164 !validate_convert_profile(fs_info
, &bctl
->meta
, allowed
, "metadata") ||
4165 !validate_convert_profile(fs_info
, &bctl
->sys
, allowed
, "system")) {
4171 * Allow to reduce metadata or system integrity only if force set for
4172 * profiles with redundancy (copies, parity)
4175 for (i
= 0; i
< ARRAY_SIZE(btrfs_raid_array
); i
++) {
4176 if (btrfs_raid_array
[i
].ncopies
>= 2 ||
4177 btrfs_raid_array
[i
].tolerated_failures
>= 1)
4178 allowed
|= btrfs_raid_array
[i
].bg_flag
;
4181 seq
= read_seqbegin(&fs_info
->profiles_lock
);
4183 if (((bctl
->sys
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
4184 (fs_info
->avail_system_alloc_bits
& allowed
) &&
4185 !(bctl
->sys
.target
& allowed
)) ||
4186 ((bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) &&
4187 (fs_info
->avail_metadata_alloc_bits
& allowed
) &&
4188 !(bctl
->meta
.target
& allowed
)))
4189 reducing_redundancy
= true;
4191 reducing_redundancy
= false;
4193 /* if we're not converting, the target field is uninitialized */
4194 meta_target
= (bctl
->meta
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) ?
4195 bctl
->meta
.target
: fs_info
->avail_metadata_alloc_bits
;
4196 data_target
= (bctl
->data
.flags
& BTRFS_BALANCE_ARGS_CONVERT
) ?
4197 bctl
->data
.target
: fs_info
->avail_data_alloc_bits
;
4198 } while (read_seqretry(&fs_info
->profiles_lock
, seq
));
4200 if (reducing_redundancy
) {
4201 if (bctl
->flags
& BTRFS_BALANCE_FORCE
) {
4203 "balance: force reducing metadata redundancy");
4206 "balance: reduces metadata redundancy, use --force if you want this");
4212 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target
) <
4213 btrfs_get_num_tolerated_disk_barrier_failures(data_target
)) {
4215 "balance: metadata profile %s has lower redundancy than data profile %s",
4216 btrfs_bg_type_to_raid_name(meta_target
),
4217 btrfs_bg_type_to_raid_name(data_target
));
4220 if (fs_info
->send_in_progress
) {
4221 btrfs_warn_rl(fs_info
,
4222 "cannot run balance while send operations are in progress (%d in progress)",
4223 fs_info
->send_in_progress
);
4228 ret
= insert_balance_item(fs_info
, bctl
);
4229 if (ret
&& ret
!= -EEXIST
)
4232 if (!(bctl
->flags
& BTRFS_BALANCE_RESUME
)) {
4233 BUG_ON(ret
== -EEXIST
);
4234 BUG_ON(fs_info
->balance_ctl
);
4235 spin_lock(&fs_info
->balance_lock
);
4236 fs_info
->balance_ctl
= bctl
;
4237 spin_unlock(&fs_info
->balance_lock
);
4239 BUG_ON(ret
!= -EEXIST
);
4240 spin_lock(&fs_info
->balance_lock
);
4241 update_balance_args(bctl
);
4242 spin_unlock(&fs_info
->balance_lock
);
4245 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
));
4246 set_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
);
4247 describe_balance_start_or_resume(fs_info
);
4248 mutex_unlock(&fs_info
->balance_mutex
);
4250 ret
= __btrfs_balance(fs_info
);
4252 mutex_lock(&fs_info
->balance_mutex
);
4253 if (ret
== -ECANCELED
&& atomic_read(&fs_info
->balance_pause_req
))
4254 btrfs_info(fs_info
, "balance: paused");
4256 * Balance can be canceled by:
4258 * - Regular cancel request
4259 * Then ret == -ECANCELED and balance_cancel_req > 0
4261 * - Fatal signal to "btrfs" process
4262 * Either the signal caught by wait_reserve_ticket() and callers
4263 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4265 * Either way, in this case balance_cancel_req = 0, and
4266 * ret == -EINTR or ret == -ECANCELED.
4268 * So here we only check the return value to catch canceled balance.
4270 else if (ret
== -ECANCELED
|| ret
== -EINTR
)
4271 btrfs_info(fs_info
, "balance: canceled");
4273 btrfs_info(fs_info
, "balance: ended with status: %d", ret
);
4275 clear_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
);
4278 memset(bargs
, 0, sizeof(*bargs
));
4279 btrfs_update_ioctl_balance_args(fs_info
, bargs
);
4282 if ((ret
&& ret
!= -ECANCELED
&& ret
!= -ENOSPC
) ||
4283 balance_need_close(fs_info
)) {
4284 reset_balance_state(fs_info
);
4285 btrfs_exclop_finish(fs_info
);
4288 wake_up(&fs_info
->balance_wait_q
);
4292 if (bctl
->flags
& BTRFS_BALANCE_RESUME
)
4293 reset_balance_state(fs_info
);
4296 btrfs_exclop_finish(fs_info
);
4301 static int balance_kthread(void *data
)
4303 struct btrfs_fs_info
*fs_info
= data
;
4306 mutex_lock(&fs_info
->balance_mutex
);
4307 if (fs_info
->balance_ctl
)
4308 ret
= btrfs_balance(fs_info
, fs_info
->balance_ctl
, NULL
);
4309 mutex_unlock(&fs_info
->balance_mutex
);
4314 int btrfs_resume_balance_async(struct btrfs_fs_info
*fs_info
)
4316 struct task_struct
*tsk
;
4318 mutex_lock(&fs_info
->balance_mutex
);
4319 if (!fs_info
->balance_ctl
) {
4320 mutex_unlock(&fs_info
->balance_mutex
);
4323 mutex_unlock(&fs_info
->balance_mutex
);
4325 if (btrfs_test_opt(fs_info
, SKIP_BALANCE
)) {
4326 btrfs_info(fs_info
, "balance: resume skipped");
4331 * A ro->rw remount sequence should continue with the paused balance
4332 * regardless of who pauses it, system or the user as of now, so set
4335 spin_lock(&fs_info
->balance_lock
);
4336 fs_info
->balance_ctl
->flags
|= BTRFS_BALANCE_RESUME
;
4337 spin_unlock(&fs_info
->balance_lock
);
4339 tsk
= kthread_run(balance_kthread
, fs_info
, "btrfs-balance");
4340 return PTR_ERR_OR_ZERO(tsk
);
4343 int btrfs_recover_balance(struct btrfs_fs_info
*fs_info
)
4345 struct btrfs_balance_control
*bctl
;
4346 struct btrfs_balance_item
*item
;
4347 struct btrfs_disk_balance_args disk_bargs
;
4348 struct btrfs_path
*path
;
4349 struct extent_buffer
*leaf
;
4350 struct btrfs_key key
;
4353 path
= btrfs_alloc_path();
4357 key
.objectid
= BTRFS_BALANCE_OBJECTID
;
4358 key
.type
= BTRFS_TEMPORARY_ITEM_KEY
;
4361 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
4364 if (ret
> 0) { /* ret = -ENOENT; */
4369 bctl
= kzalloc(sizeof(*bctl
), GFP_NOFS
);
4375 leaf
= path
->nodes
[0];
4376 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_balance_item
);
4378 bctl
->flags
= btrfs_balance_flags(leaf
, item
);
4379 bctl
->flags
|= BTRFS_BALANCE_RESUME
;
4381 btrfs_balance_data(leaf
, item
, &disk_bargs
);
4382 btrfs_disk_balance_args_to_cpu(&bctl
->data
, &disk_bargs
);
4383 btrfs_balance_meta(leaf
, item
, &disk_bargs
);
4384 btrfs_disk_balance_args_to_cpu(&bctl
->meta
, &disk_bargs
);
4385 btrfs_balance_sys(leaf
, item
, &disk_bargs
);
4386 btrfs_disk_balance_args_to_cpu(&bctl
->sys
, &disk_bargs
);
4389 * This should never happen, as the paused balance state is recovered
4390 * during mount without any chance of other exclusive ops to collide.
4392 * This gives the exclusive op status to balance and keeps in paused
4393 * state until user intervention (cancel or umount). If the ownership
4394 * cannot be assigned, show a message but do not fail. The balance
4395 * is in a paused state and must have fs_info::balance_ctl properly
4398 if (!btrfs_exclop_start(fs_info
, BTRFS_EXCLOP_BALANCE
))
4400 "balance: cannot set exclusive op status, resume manually");
4402 btrfs_release_path(path
);
4404 mutex_lock(&fs_info
->balance_mutex
);
4405 BUG_ON(fs_info
->balance_ctl
);
4406 spin_lock(&fs_info
->balance_lock
);
4407 fs_info
->balance_ctl
= bctl
;
4408 spin_unlock(&fs_info
->balance_lock
);
4409 mutex_unlock(&fs_info
->balance_mutex
);
4411 btrfs_free_path(path
);
4415 int btrfs_pause_balance(struct btrfs_fs_info
*fs_info
)
4419 mutex_lock(&fs_info
->balance_mutex
);
4420 if (!fs_info
->balance_ctl
) {
4421 mutex_unlock(&fs_info
->balance_mutex
);
4425 if (test_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
)) {
4426 atomic_inc(&fs_info
->balance_pause_req
);
4427 mutex_unlock(&fs_info
->balance_mutex
);
4429 wait_event(fs_info
->balance_wait_q
,
4430 !test_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
));
4432 mutex_lock(&fs_info
->balance_mutex
);
4433 /* we are good with balance_ctl ripped off from under us */
4434 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
));
4435 atomic_dec(&fs_info
->balance_pause_req
);
4440 mutex_unlock(&fs_info
->balance_mutex
);
4444 int btrfs_cancel_balance(struct btrfs_fs_info
*fs_info
)
4446 mutex_lock(&fs_info
->balance_mutex
);
4447 if (!fs_info
->balance_ctl
) {
4448 mutex_unlock(&fs_info
->balance_mutex
);
4453 * A paused balance with the item stored on disk can be resumed at
4454 * mount time if the mount is read-write. Otherwise it's still paused
4455 * and we must not allow cancelling as it deletes the item.
4457 if (sb_rdonly(fs_info
->sb
)) {
4458 mutex_unlock(&fs_info
->balance_mutex
);
4462 atomic_inc(&fs_info
->balance_cancel_req
);
4464 * if we are running just wait and return, balance item is
4465 * deleted in btrfs_balance in this case
4467 if (test_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
)) {
4468 mutex_unlock(&fs_info
->balance_mutex
);
4469 wait_event(fs_info
->balance_wait_q
,
4470 !test_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
));
4471 mutex_lock(&fs_info
->balance_mutex
);
4473 mutex_unlock(&fs_info
->balance_mutex
);
4475 * Lock released to allow other waiters to continue, we'll
4476 * reexamine the status again.
4478 mutex_lock(&fs_info
->balance_mutex
);
4480 if (fs_info
->balance_ctl
) {
4481 reset_balance_state(fs_info
);
4482 btrfs_exclop_finish(fs_info
);
4483 btrfs_info(fs_info
, "balance: canceled");
4487 BUG_ON(fs_info
->balance_ctl
||
4488 test_bit(BTRFS_FS_BALANCE_RUNNING
, &fs_info
->flags
));
4489 atomic_dec(&fs_info
->balance_cancel_req
);
4490 mutex_unlock(&fs_info
->balance_mutex
);
4494 int btrfs_uuid_scan_kthread(void *data
)
4496 struct btrfs_fs_info
*fs_info
= data
;
4497 struct btrfs_root
*root
= fs_info
->tree_root
;
4498 struct btrfs_key key
;
4499 struct btrfs_path
*path
= NULL
;
4501 struct extent_buffer
*eb
;
4503 struct btrfs_root_item root_item
;
4505 struct btrfs_trans_handle
*trans
= NULL
;
4506 bool closing
= false;
4508 path
= btrfs_alloc_path();
4515 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4519 if (btrfs_fs_closing(fs_info
)) {
4523 ret
= btrfs_search_forward(root
, &key
, path
,
4524 BTRFS_OLDEST_GENERATION
);
4531 if (key
.type
!= BTRFS_ROOT_ITEM_KEY
||
4532 (key
.objectid
< BTRFS_FIRST_FREE_OBJECTID
&&
4533 key
.objectid
!= BTRFS_FS_TREE_OBJECTID
) ||
4534 key
.objectid
> BTRFS_LAST_FREE_OBJECTID
)
4537 eb
= path
->nodes
[0];
4538 slot
= path
->slots
[0];
4539 item_size
= btrfs_item_size_nr(eb
, slot
);
4540 if (item_size
< sizeof(root_item
))
4543 read_extent_buffer(eb
, &root_item
,
4544 btrfs_item_ptr_offset(eb
, slot
),
4545 (int)sizeof(root_item
));
4546 if (btrfs_root_refs(&root_item
) == 0)
4549 if (!btrfs_is_empty_uuid(root_item
.uuid
) ||
4550 !btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4554 btrfs_release_path(path
);
4556 * 1 - subvol uuid item
4557 * 1 - received_subvol uuid item
4559 trans
= btrfs_start_transaction(fs_info
->uuid_root
, 2);
4560 if (IS_ERR(trans
)) {
4561 ret
= PTR_ERR(trans
);
4569 btrfs_release_path(path
);
4570 if (!btrfs_is_empty_uuid(root_item
.uuid
)) {
4571 ret
= btrfs_uuid_tree_add(trans
, root_item
.uuid
,
4572 BTRFS_UUID_KEY_SUBVOL
,
4575 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4581 if (!btrfs_is_empty_uuid(root_item
.received_uuid
)) {
4582 ret
= btrfs_uuid_tree_add(trans
,
4583 root_item
.received_uuid
,
4584 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
4587 btrfs_warn(fs_info
, "uuid_tree_add failed %d",
4594 btrfs_release_path(path
);
4596 ret
= btrfs_end_transaction(trans
);
4602 if (key
.offset
< (u64
)-1) {
4604 } else if (key
.type
< BTRFS_ROOT_ITEM_KEY
) {
4606 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4607 } else if (key
.objectid
< (u64
)-1) {
4609 key
.type
= BTRFS_ROOT_ITEM_KEY
;
4618 btrfs_free_path(path
);
4619 if (trans
&& !IS_ERR(trans
))
4620 btrfs_end_transaction(trans
);
4622 btrfs_warn(fs_info
, "btrfs_uuid_scan_kthread failed %d", ret
);
4624 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN
, &fs_info
->flags
);
4625 up(&fs_info
->uuid_tree_rescan_sem
);
4629 int btrfs_create_uuid_tree(struct btrfs_fs_info
*fs_info
)
4631 struct btrfs_trans_handle
*trans
;
4632 struct btrfs_root
*tree_root
= fs_info
->tree_root
;
4633 struct btrfs_root
*uuid_root
;
4634 struct task_struct
*task
;
4641 trans
= btrfs_start_transaction(tree_root
, 2);
4643 return PTR_ERR(trans
);
4645 uuid_root
= btrfs_create_tree(trans
, BTRFS_UUID_TREE_OBJECTID
);
4646 if (IS_ERR(uuid_root
)) {
4647 ret
= PTR_ERR(uuid_root
);
4648 btrfs_abort_transaction(trans
, ret
);
4649 btrfs_end_transaction(trans
);
4653 fs_info
->uuid_root
= uuid_root
;
4655 ret
= btrfs_commit_transaction(trans
);
4659 down(&fs_info
->uuid_tree_rescan_sem
);
4660 task
= kthread_run(btrfs_uuid_scan_kthread
, fs_info
, "btrfs-uuid");
4662 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4663 btrfs_warn(fs_info
, "failed to start uuid_scan task");
4664 up(&fs_info
->uuid_tree_rescan_sem
);
4665 return PTR_ERR(task
);
4672 * shrinking a device means finding all of the device extents past
4673 * the new size, and then following the back refs to the chunks.
4674 * The chunk relocation code actually frees the device extent
4676 int btrfs_shrink_device(struct btrfs_device
*device
, u64 new_size
)
4678 struct btrfs_fs_info
*fs_info
= device
->fs_info
;
4679 struct btrfs_root
*root
= fs_info
->dev_root
;
4680 struct btrfs_trans_handle
*trans
;
4681 struct btrfs_dev_extent
*dev_extent
= NULL
;
4682 struct btrfs_path
*path
;
4688 bool retried
= false;
4689 struct extent_buffer
*l
;
4690 struct btrfs_key key
;
4691 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
4692 u64 old_total
= btrfs_super_total_bytes(super_copy
);
4693 u64 old_size
= btrfs_device_get_total_bytes(device
);
4697 new_size
= round_down(new_size
, fs_info
->sectorsize
);
4699 diff
= round_down(old_size
- new_size
, fs_info
->sectorsize
);
4701 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
))
4704 path
= btrfs_alloc_path();
4708 path
->reada
= READA_BACK
;
4710 trans
= btrfs_start_transaction(root
, 0);
4711 if (IS_ERR(trans
)) {
4712 btrfs_free_path(path
);
4713 return PTR_ERR(trans
);
4716 mutex_lock(&fs_info
->chunk_mutex
);
4718 btrfs_device_set_total_bytes(device
, new_size
);
4719 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
)) {
4720 device
->fs_devices
->total_rw_bytes
-= diff
;
4721 atomic64_sub(diff
, &fs_info
->free_chunk_space
);
4725 * Once the device's size has been set to the new size, ensure all
4726 * in-memory chunks are synced to disk so that the loop below sees them
4727 * and relocates them accordingly.
4729 if (contains_pending_extent(device
, &start
, diff
)) {
4730 mutex_unlock(&fs_info
->chunk_mutex
);
4731 ret
= btrfs_commit_transaction(trans
);
4735 mutex_unlock(&fs_info
->chunk_mutex
);
4736 btrfs_end_transaction(trans
);
4740 key
.objectid
= device
->devid
;
4741 key
.offset
= (u64
)-1;
4742 key
.type
= BTRFS_DEV_EXTENT_KEY
;
4745 mutex_lock(&fs_info
->reclaim_bgs_lock
);
4746 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4748 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
4752 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
4754 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
4758 btrfs_release_path(path
);
4763 slot
= path
->slots
[0];
4764 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
4766 if (key
.objectid
!= device
->devid
) {
4767 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
4768 btrfs_release_path(path
);
4772 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
4773 length
= btrfs_dev_extent_length(l
, dev_extent
);
4775 if (key
.offset
+ length
<= new_size
) {
4776 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
4777 btrfs_release_path(path
);
4781 chunk_offset
= btrfs_dev_extent_chunk_offset(l
, dev_extent
);
4782 btrfs_release_path(path
);
4785 * We may be relocating the only data chunk we have,
4786 * which could potentially end up with losing data's
4787 * raid profile, so lets allocate an empty one in
4790 ret
= btrfs_may_alloc_data_chunk(fs_info
, chunk_offset
);
4792 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
4796 ret
= btrfs_relocate_chunk(fs_info
, chunk_offset
);
4797 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
4798 if (ret
== -ENOSPC
) {
4801 if (ret
== -ETXTBSY
) {
4803 "could not shrink block group %llu due to active swapfile",
4808 } while (key
.offset
-- > 0);
4810 if (failed
&& !retried
) {
4814 } else if (failed
&& retried
) {
4819 /* Shrinking succeeded, else we would be at "done". */
4820 trans
= btrfs_start_transaction(root
, 0);
4821 if (IS_ERR(trans
)) {
4822 ret
= PTR_ERR(trans
);
4826 mutex_lock(&fs_info
->chunk_mutex
);
4827 /* Clear all state bits beyond the shrunk device size */
4828 clear_extent_bits(&device
->alloc_state
, new_size
, (u64
)-1,
4831 btrfs_device_set_disk_total_bytes(device
, new_size
);
4832 if (list_empty(&device
->post_commit_list
))
4833 list_add_tail(&device
->post_commit_list
,
4834 &trans
->transaction
->dev_update_list
);
4836 WARN_ON(diff
> old_total
);
4837 btrfs_set_super_total_bytes(super_copy
,
4838 round_down(old_total
- diff
, fs_info
->sectorsize
));
4839 mutex_unlock(&fs_info
->chunk_mutex
);
4841 /* Now btrfs_update_device() will change the on-disk size. */
4842 ret
= btrfs_update_device(trans
, device
);
4844 btrfs_abort_transaction(trans
, ret
);
4845 btrfs_end_transaction(trans
);
4847 ret
= btrfs_commit_transaction(trans
);
4850 btrfs_free_path(path
);
4852 mutex_lock(&fs_info
->chunk_mutex
);
4853 btrfs_device_set_total_bytes(device
, old_size
);
4854 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
))
4855 device
->fs_devices
->total_rw_bytes
+= diff
;
4856 atomic64_add(diff
, &fs_info
->free_chunk_space
);
4857 mutex_unlock(&fs_info
->chunk_mutex
);
4862 static int btrfs_add_system_chunk(struct btrfs_fs_info
*fs_info
,
4863 struct btrfs_key
*key
,
4864 struct btrfs_chunk
*chunk
, int item_size
)
4866 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
4867 struct btrfs_disk_key disk_key
;
4871 mutex_lock(&fs_info
->chunk_mutex
);
4872 array_size
= btrfs_super_sys_array_size(super_copy
);
4873 if (array_size
+ item_size
+ sizeof(disk_key
)
4874 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
) {
4875 mutex_unlock(&fs_info
->chunk_mutex
);
4879 ptr
= super_copy
->sys_chunk_array
+ array_size
;
4880 btrfs_cpu_key_to_disk(&disk_key
, key
);
4881 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
4882 ptr
+= sizeof(disk_key
);
4883 memcpy(ptr
, chunk
, item_size
);
4884 item_size
+= sizeof(disk_key
);
4885 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
4886 mutex_unlock(&fs_info
->chunk_mutex
);
4892 * sort the devices in descending order by max_avail, total_avail
4894 static int btrfs_cmp_device_info(const void *a
, const void *b
)
4896 const struct btrfs_device_info
*di_a
= a
;
4897 const struct btrfs_device_info
*di_b
= b
;
4899 if (di_a
->max_avail
> di_b
->max_avail
)
4901 if (di_a
->max_avail
< di_b
->max_avail
)
4903 if (di_a
->total_avail
> di_b
->total_avail
)
4905 if (di_a
->total_avail
< di_b
->total_avail
)
4910 static void check_raid56_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
4912 if (!(type
& BTRFS_BLOCK_GROUP_RAID56_MASK
))
4915 btrfs_set_fs_incompat(info
, RAID56
);
4918 static void check_raid1c34_incompat_flag(struct btrfs_fs_info
*info
, u64 type
)
4920 if (!(type
& (BTRFS_BLOCK_GROUP_RAID1C3
| BTRFS_BLOCK_GROUP_RAID1C4
)))
4923 btrfs_set_fs_incompat(info
, RAID1C34
);
4927 * Structure used internally for __btrfs_alloc_chunk() function.
4928 * Wraps needed parameters.
4930 struct alloc_chunk_ctl
{
4933 /* Total number of stripes to allocate */
4935 /* sub_stripes info for map */
4937 /* Stripes per device */
4939 /* Maximum number of devices to use */
4941 /* Minimum number of devices to use */
4943 /* ndevs has to be a multiple of this */
4945 /* Number of copies */
4947 /* Number of stripes worth of bytes to store parity information */
4949 u64 max_stripe_size
;
4957 static void init_alloc_chunk_ctl_policy_regular(
4958 struct btrfs_fs_devices
*fs_devices
,
4959 struct alloc_chunk_ctl
*ctl
)
4961 u64 type
= ctl
->type
;
4963 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
4964 ctl
->max_stripe_size
= SZ_1G
;
4965 ctl
->max_chunk_size
= BTRFS_MAX_DATA_CHUNK_SIZE
;
4966 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
4967 /* For larger filesystems, use larger metadata chunks */
4968 if (fs_devices
->total_rw_bytes
> 50ULL * SZ_1G
)
4969 ctl
->max_stripe_size
= SZ_1G
;
4971 ctl
->max_stripe_size
= SZ_256M
;
4972 ctl
->max_chunk_size
= ctl
->max_stripe_size
;
4973 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
4974 ctl
->max_stripe_size
= SZ_32M
;
4975 ctl
->max_chunk_size
= 2 * ctl
->max_stripe_size
;
4976 ctl
->devs_max
= min_t(int, ctl
->devs_max
,
4977 BTRFS_MAX_DEVS_SYS_CHUNK
);
4982 /* We don't want a chunk larger than 10% of writable space */
4983 ctl
->max_chunk_size
= min(div_factor(fs_devices
->total_rw_bytes
, 1),
4984 ctl
->max_chunk_size
);
4985 ctl
->dev_extent_min
= BTRFS_STRIPE_LEN
* ctl
->dev_stripes
;
4988 static void init_alloc_chunk_ctl_policy_zoned(
4989 struct btrfs_fs_devices
*fs_devices
,
4990 struct alloc_chunk_ctl
*ctl
)
4992 u64 zone_size
= fs_devices
->fs_info
->zone_size
;
4994 int min_num_stripes
= ctl
->devs_min
* ctl
->dev_stripes
;
4995 int min_data_stripes
= (min_num_stripes
- ctl
->nparity
) / ctl
->ncopies
;
4996 u64 min_chunk_size
= min_data_stripes
* zone_size
;
4997 u64 type
= ctl
->type
;
4999 ctl
->max_stripe_size
= zone_size
;
5000 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
5001 ctl
->max_chunk_size
= round_down(BTRFS_MAX_DATA_CHUNK_SIZE
,
5003 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
5004 ctl
->max_chunk_size
= ctl
->max_stripe_size
;
5005 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
5006 ctl
->max_chunk_size
= 2 * ctl
->max_stripe_size
;
5007 ctl
->devs_max
= min_t(int, ctl
->devs_max
,
5008 BTRFS_MAX_DEVS_SYS_CHUNK
);
5013 /* We don't want a chunk larger than 10% of writable space */
5014 limit
= max(round_down(div_factor(fs_devices
->total_rw_bytes
, 1),
5017 ctl
->max_chunk_size
= min(limit
, ctl
->max_chunk_size
);
5018 ctl
->dev_extent_min
= zone_size
* ctl
->dev_stripes
;
5021 static void init_alloc_chunk_ctl(struct btrfs_fs_devices
*fs_devices
,
5022 struct alloc_chunk_ctl
*ctl
)
5024 int index
= btrfs_bg_flags_to_raid_index(ctl
->type
);
5026 ctl
->sub_stripes
= btrfs_raid_array
[index
].sub_stripes
;
5027 ctl
->dev_stripes
= btrfs_raid_array
[index
].dev_stripes
;
5028 ctl
->devs_max
= btrfs_raid_array
[index
].devs_max
;
5030 ctl
->devs_max
= BTRFS_MAX_DEVS(fs_devices
->fs_info
);
5031 ctl
->devs_min
= btrfs_raid_array
[index
].devs_min
;
5032 ctl
->devs_increment
= btrfs_raid_array
[index
].devs_increment
;
5033 ctl
->ncopies
= btrfs_raid_array
[index
].ncopies
;
5034 ctl
->nparity
= btrfs_raid_array
[index
].nparity
;
5037 switch (fs_devices
->chunk_alloc_policy
) {
5038 case BTRFS_CHUNK_ALLOC_REGULAR
:
5039 init_alloc_chunk_ctl_policy_regular(fs_devices
, ctl
);
5041 case BTRFS_CHUNK_ALLOC_ZONED
:
5042 init_alloc_chunk_ctl_policy_zoned(fs_devices
, ctl
);
5049 static int gather_device_info(struct btrfs_fs_devices
*fs_devices
,
5050 struct alloc_chunk_ctl
*ctl
,
5051 struct btrfs_device_info
*devices_info
)
5053 struct btrfs_fs_info
*info
= fs_devices
->fs_info
;
5054 struct btrfs_device
*device
;
5056 u64 dev_extent_want
= ctl
->max_stripe_size
* ctl
->dev_stripes
;
5063 * in the first pass through the devices list, we gather information
5064 * about the available holes on each device.
5066 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
5067 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
)) {
5069 "BTRFS: read-only device in alloc_list\n");
5073 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA
,
5074 &device
->dev_state
) ||
5075 test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
))
5078 if (device
->total_bytes
> device
->bytes_used
)
5079 total_avail
= device
->total_bytes
- device
->bytes_used
;
5083 /* If there is no space on this device, skip it. */
5084 if (total_avail
< ctl
->dev_extent_min
)
5087 ret
= find_free_dev_extent(device
, dev_extent_want
, &dev_offset
,
5089 if (ret
&& ret
!= -ENOSPC
)
5093 max_avail
= dev_extent_want
;
5095 if (max_avail
< ctl
->dev_extent_min
) {
5096 if (btrfs_test_opt(info
, ENOSPC_DEBUG
))
5098 "%s: devid %llu has no free space, have=%llu want=%llu",
5099 __func__
, device
->devid
, max_avail
,
5100 ctl
->dev_extent_min
);
5104 if (ndevs
== fs_devices
->rw_devices
) {
5105 WARN(1, "%s: found more than %llu devices\n",
5106 __func__
, fs_devices
->rw_devices
);
5109 devices_info
[ndevs
].dev_offset
= dev_offset
;
5110 devices_info
[ndevs
].max_avail
= max_avail
;
5111 devices_info
[ndevs
].total_avail
= total_avail
;
5112 devices_info
[ndevs
].dev
= device
;
5118 * now sort the devices by hole size / available space
5120 sort(devices_info
, ndevs
, sizeof(struct btrfs_device_info
),
5121 btrfs_cmp_device_info
, NULL
);
5126 static int decide_stripe_size_regular(struct alloc_chunk_ctl
*ctl
,
5127 struct btrfs_device_info
*devices_info
)
5129 /* Number of stripes that count for block group size */
5133 * The primary goal is to maximize the number of stripes, so use as
5134 * many devices as possible, even if the stripes are not maximum sized.
5136 * The DUP profile stores more than one stripe per device, the
5137 * max_avail is the total size so we have to adjust.
5139 ctl
->stripe_size
= div_u64(devices_info
[ctl
->ndevs
- 1].max_avail
,
5141 ctl
->num_stripes
= ctl
->ndevs
* ctl
->dev_stripes
;
5143 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5144 data_stripes
= (ctl
->num_stripes
- ctl
->nparity
) / ctl
->ncopies
;
5147 * Use the number of data stripes to figure out how big this chunk is
5148 * really going to be in terms of logical address space, and compare
5149 * that answer with the max chunk size. If it's higher, we try to
5150 * reduce stripe_size.
5152 if (ctl
->stripe_size
* data_stripes
> ctl
->max_chunk_size
) {
5154 * Reduce stripe_size, round it up to a 16MB boundary again and
5155 * then use it, unless it ends up being even bigger than the
5156 * previous value we had already.
5158 ctl
->stripe_size
= min(round_up(div_u64(ctl
->max_chunk_size
,
5159 data_stripes
), SZ_16M
),
5163 /* Align to BTRFS_STRIPE_LEN */
5164 ctl
->stripe_size
= round_down(ctl
->stripe_size
, BTRFS_STRIPE_LEN
);
5165 ctl
->chunk_size
= ctl
->stripe_size
* data_stripes
;
5170 static int decide_stripe_size_zoned(struct alloc_chunk_ctl
*ctl
,
5171 struct btrfs_device_info
*devices_info
)
5173 u64 zone_size
= devices_info
[0].dev
->zone_info
->zone_size
;
5174 /* Number of stripes that count for block group size */
5178 * It should hold because:
5179 * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5181 ASSERT(devices_info
[ctl
->ndevs
- 1].max_avail
== ctl
->dev_extent_min
);
5183 ctl
->stripe_size
= zone_size
;
5184 ctl
->num_stripes
= ctl
->ndevs
* ctl
->dev_stripes
;
5185 data_stripes
= (ctl
->num_stripes
- ctl
->nparity
) / ctl
->ncopies
;
5187 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5188 if (ctl
->stripe_size
* data_stripes
> ctl
->max_chunk_size
) {
5189 ctl
->ndevs
= div_u64(div_u64(ctl
->max_chunk_size
* ctl
->ncopies
,
5190 ctl
->stripe_size
) + ctl
->nparity
,
5192 ctl
->num_stripes
= ctl
->ndevs
* ctl
->dev_stripes
;
5193 data_stripes
= (ctl
->num_stripes
- ctl
->nparity
) / ctl
->ncopies
;
5194 ASSERT(ctl
->stripe_size
* data_stripes
<= ctl
->max_chunk_size
);
5197 ctl
->chunk_size
= ctl
->stripe_size
* data_stripes
;
5202 static int decide_stripe_size(struct btrfs_fs_devices
*fs_devices
,
5203 struct alloc_chunk_ctl
*ctl
,
5204 struct btrfs_device_info
*devices_info
)
5206 struct btrfs_fs_info
*info
= fs_devices
->fs_info
;
5209 * Round down to number of usable stripes, devs_increment can be any
5210 * number so we can't use round_down() that requires power of 2, while
5211 * rounddown is safe.
5213 ctl
->ndevs
= rounddown(ctl
->ndevs
, ctl
->devs_increment
);
5215 if (ctl
->ndevs
< ctl
->devs_min
) {
5216 if (btrfs_test_opt(info
, ENOSPC_DEBUG
)) {
5218 "%s: not enough devices with free space: have=%d minimum required=%d",
5219 __func__
, ctl
->ndevs
, ctl
->devs_min
);
5224 ctl
->ndevs
= min(ctl
->ndevs
, ctl
->devs_max
);
5226 switch (fs_devices
->chunk_alloc_policy
) {
5227 case BTRFS_CHUNK_ALLOC_REGULAR
:
5228 return decide_stripe_size_regular(ctl
, devices_info
);
5229 case BTRFS_CHUNK_ALLOC_ZONED
:
5230 return decide_stripe_size_zoned(ctl
, devices_info
);
5236 static int create_chunk(struct btrfs_trans_handle
*trans
,
5237 struct alloc_chunk_ctl
*ctl
,
5238 struct btrfs_device_info
*devices_info
)
5240 struct btrfs_fs_info
*info
= trans
->fs_info
;
5241 struct map_lookup
*map
= NULL
;
5242 struct extent_map_tree
*em_tree
;
5243 struct extent_map
*em
;
5244 u64 start
= ctl
->start
;
5245 u64 type
= ctl
->type
;
5250 map
= kmalloc(map_lookup_size(ctl
->num_stripes
), GFP_NOFS
);
5253 map
->num_stripes
= ctl
->num_stripes
;
5255 for (i
= 0; i
< ctl
->ndevs
; ++i
) {
5256 for (j
= 0; j
< ctl
->dev_stripes
; ++j
) {
5257 int s
= i
* ctl
->dev_stripes
+ j
;
5258 map
->stripes
[s
].dev
= devices_info
[i
].dev
;
5259 map
->stripes
[s
].physical
= devices_info
[i
].dev_offset
+
5260 j
* ctl
->stripe_size
;
5263 map
->stripe_len
= BTRFS_STRIPE_LEN
;
5264 map
->io_align
= BTRFS_STRIPE_LEN
;
5265 map
->io_width
= BTRFS_STRIPE_LEN
;
5267 map
->sub_stripes
= ctl
->sub_stripes
;
5269 trace_btrfs_chunk_alloc(info
, map
, start
, ctl
->chunk_size
);
5271 em
= alloc_extent_map();
5276 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
5277 em
->map_lookup
= map
;
5279 em
->len
= ctl
->chunk_size
;
5280 em
->block_start
= 0;
5281 em
->block_len
= em
->len
;
5282 em
->orig_block_len
= ctl
->stripe_size
;
5284 em_tree
= &info
->mapping_tree
;
5285 write_lock(&em_tree
->lock
);
5286 ret
= add_extent_mapping(em_tree
, em
, 0);
5288 write_unlock(&em_tree
->lock
);
5289 free_extent_map(em
);
5292 write_unlock(&em_tree
->lock
);
5294 ret
= btrfs_make_block_group(trans
, 0, type
, start
, ctl
->chunk_size
);
5296 goto error_del_extent
;
5298 for (i
= 0; i
< map
->num_stripes
; i
++) {
5299 struct btrfs_device
*dev
= map
->stripes
[i
].dev
;
5301 btrfs_device_set_bytes_used(dev
,
5302 dev
->bytes_used
+ ctl
->stripe_size
);
5303 if (list_empty(&dev
->post_commit_list
))
5304 list_add_tail(&dev
->post_commit_list
,
5305 &trans
->transaction
->dev_update_list
);
5308 atomic64_sub(ctl
->stripe_size
* map
->num_stripes
,
5309 &info
->free_chunk_space
);
5311 free_extent_map(em
);
5312 check_raid56_incompat_flag(info
, type
);
5313 check_raid1c34_incompat_flag(info
, type
);
5318 write_lock(&em_tree
->lock
);
5319 remove_extent_mapping(em_tree
, em
);
5320 write_unlock(&em_tree
->lock
);
5322 /* One for our allocation */
5323 free_extent_map(em
);
5324 /* One for the tree reference */
5325 free_extent_map(em
);
5330 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
, u64 type
)
5332 struct btrfs_fs_info
*info
= trans
->fs_info
;
5333 struct btrfs_fs_devices
*fs_devices
= info
->fs_devices
;
5334 struct btrfs_device_info
*devices_info
= NULL
;
5335 struct alloc_chunk_ctl ctl
;
5338 lockdep_assert_held(&info
->chunk_mutex
);
5340 if (!alloc_profile_is_valid(type
, 0)) {
5345 if (list_empty(&fs_devices
->alloc_list
)) {
5346 if (btrfs_test_opt(info
, ENOSPC_DEBUG
))
5347 btrfs_debug(info
, "%s: no writable device", __func__
);
5351 if (!(type
& BTRFS_BLOCK_GROUP_TYPE_MASK
)) {
5352 btrfs_err(info
, "invalid chunk type 0x%llx requested", type
);
5357 ctl
.start
= find_next_chunk(info
);
5359 init_alloc_chunk_ctl(fs_devices
, &ctl
);
5361 devices_info
= kcalloc(fs_devices
->rw_devices
, sizeof(*devices_info
),
5366 ret
= gather_device_info(fs_devices
, &ctl
, devices_info
);
5370 ret
= decide_stripe_size(fs_devices
, &ctl
, devices_info
);
5374 ret
= create_chunk(trans
, &ctl
, devices_info
);
5377 kfree(devices_info
);
5382 * Chunk allocation falls into two parts. The first part does work
5383 * that makes the new allocated chunk usable, but does not do any operation
5384 * that modifies the chunk tree. The second part does the work that
5385 * requires modifying the chunk tree. This division is important for the
5386 * bootstrap process of adding storage to a seed btrfs.
5388 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle
*trans
,
5389 u64 chunk_offset
, u64 chunk_size
)
5391 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
5392 struct btrfs_root
*extent_root
= fs_info
->extent_root
;
5393 struct btrfs_root
*chunk_root
= fs_info
->chunk_root
;
5394 struct btrfs_key key
;
5395 struct btrfs_device
*device
;
5396 struct btrfs_chunk
*chunk
;
5397 struct btrfs_stripe
*stripe
;
5398 struct extent_map
*em
;
5399 struct map_lookup
*map
;
5406 em
= btrfs_get_chunk_map(fs_info
, chunk_offset
, chunk_size
);
5410 map
= em
->map_lookup
;
5411 item_size
= btrfs_chunk_item_size(map
->num_stripes
);
5412 stripe_size
= em
->orig_block_len
;
5414 chunk
= kzalloc(item_size
, GFP_NOFS
);
5421 * Take the device list mutex to prevent races with the final phase of
5422 * a device replace operation that replaces the device object associated
5423 * with the map's stripes, because the device object's id can change
5424 * at any time during that final phase of the device replace operation
5425 * (dev-replace.c:btrfs_dev_replace_finishing()).
5427 mutex_lock(&fs_info
->fs_devices
->device_list_mutex
);
5428 for (i
= 0; i
< map
->num_stripes
; i
++) {
5429 device
= map
->stripes
[i
].dev
;
5430 dev_offset
= map
->stripes
[i
].physical
;
5432 ret
= btrfs_update_device(trans
, device
);
5435 ret
= btrfs_alloc_dev_extent(trans
, device
, chunk_offset
,
5436 dev_offset
, stripe_size
);
5441 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
5445 stripe
= &chunk
->stripe
;
5446 for (i
= 0; i
< map
->num_stripes
; i
++) {
5447 device
= map
->stripes
[i
].dev
;
5448 dev_offset
= map
->stripes
[i
].physical
;
5450 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
5451 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
5452 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
5455 mutex_unlock(&fs_info
->fs_devices
->device_list_mutex
);
5457 btrfs_set_stack_chunk_length(chunk
, chunk_size
);
5458 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
5459 btrfs_set_stack_chunk_stripe_len(chunk
, map
->stripe_len
);
5460 btrfs_set_stack_chunk_type(chunk
, map
->type
);
5461 btrfs_set_stack_chunk_num_stripes(chunk
, map
->num_stripes
);
5462 btrfs_set_stack_chunk_io_align(chunk
, map
->stripe_len
);
5463 btrfs_set_stack_chunk_io_width(chunk
, map
->stripe_len
);
5464 btrfs_set_stack_chunk_sector_size(chunk
, fs_info
->sectorsize
);
5465 btrfs_set_stack_chunk_sub_stripes(chunk
, map
->sub_stripes
);
5467 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
5468 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
5469 key
.offset
= chunk_offset
;
5471 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
, item_size
);
5472 if (ret
== 0 && map
->type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
5474 * TODO: Cleanup of inserted chunk root in case of
5477 ret
= btrfs_add_system_chunk(fs_info
, &key
, chunk
, item_size
);
5482 free_extent_map(em
);
5486 static noinline
int init_first_rw_device(struct btrfs_trans_handle
*trans
)
5488 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
5492 alloc_profile
= btrfs_metadata_alloc_profile(fs_info
);
5493 ret
= btrfs_alloc_chunk(trans
, alloc_profile
);
5497 alloc_profile
= btrfs_system_alloc_profile(fs_info
);
5498 ret
= btrfs_alloc_chunk(trans
, alloc_profile
);
5502 static inline int btrfs_chunk_max_errors(struct map_lookup
*map
)
5504 const int index
= btrfs_bg_flags_to_raid_index(map
->type
);
5506 return btrfs_raid_array
[index
].tolerated_failures
;
5509 int btrfs_chunk_readonly(struct btrfs_fs_info
*fs_info
, u64 chunk_offset
)
5511 struct extent_map
*em
;
5512 struct map_lookup
*map
;
5517 em
= btrfs_get_chunk_map(fs_info
, chunk_offset
, 1);
5521 map
= em
->map_lookup
;
5522 for (i
= 0; i
< map
->num_stripes
; i
++) {
5523 if (test_bit(BTRFS_DEV_STATE_MISSING
,
5524 &map
->stripes
[i
].dev
->dev_state
)) {
5528 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE
,
5529 &map
->stripes
[i
].dev
->dev_state
)) {
5536 * If the number of missing devices is larger than max errors,
5537 * we can not write the data into that chunk successfully, so
5540 if (miss_ndevs
> btrfs_chunk_max_errors(map
))
5543 free_extent_map(em
);
5547 void btrfs_mapping_tree_free(struct extent_map_tree
*tree
)
5549 struct extent_map
*em
;
5552 write_lock(&tree
->lock
);
5553 em
= lookup_extent_mapping(tree
, 0, (u64
)-1);
5555 remove_extent_mapping(tree
, em
);
5556 write_unlock(&tree
->lock
);
5560 free_extent_map(em
);
5561 /* once for the tree */
5562 free_extent_map(em
);
5566 int btrfs_num_copies(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
5568 struct extent_map
*em
;
5569 struct map_lookup
*map
;
5572 em
= btrfs_get_chunk_map(fs_info
, logical
, len
);
5575 * We could return errors for these cases, but that could get
5576 * ugly and we'd probably do the same thing which is just not do
5577 * anything else and exit, so return 1 so the callers don't try
5578 * to use other copies.
5582 map
= em
->map_lookup
;
5583 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1_MASK
))
5584 ret
= map
->num_stripes
;
5585 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5586 ret
= map
->sub_stripes
;
5587 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID5
)
5589 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
5591 * There could be two corrupted data stripes, we need
5592 * to loop retry in order to rebuild the correct data.
5594 * Fail a stripe at a time on every retry except the
5595 * stripe under reconstruction.
5597 ret
= map
->num_stripes
;
5600 free_extent_map(em
);
5602 down_read(&fs_info
->dev_replace
.rwsem
);
5603 if (btrfs_dev_replace_is_ongoing(&fs_info
->dev_replace
) &&
5604 fs_info
->dev_replace
.tgtdev
)
5606 up_read(&fs_info
->dev_replace
.rwsem
);
5611 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info
*fs_info
,
5614 struct extent_map
*em
;
5615 struct map_lookup
*map
;
5616 unsigned long len
= fs_info
->sectorsize
;
5618 em
= btrfs_get_chunk_map(fs_info
, logical
, len
);
5620 if (!WARN_ON(IS_ERR(em
))) {
5621 map
= em
->map_lookup
;
5622 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5623 len
= map
->stripe_len
* nr_data_stripes(map
);
5624 free_extent_map(em
);
5629 int btrfs_is_parity_mirror(struct btrfs_fs_info
*fs_info
, u64 logical
, u64 len
)
5631 struct extent_map
*em
;
5632 struct map_lookup
*map
;
5635 em
= btrfs_get_chunk_map(fs_info
, logical
, len
);
5637 if(!WARN_ON(IS_ERR(em
))) {
5638 map
= em
->map_lookup
;
5639 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
)
5641 free_extent_map(em
);
5646 static int find_live_mirror(struct btrfs_fs_info
*fs_info
,
5647 struct map_lookup
*map
, int first
,
5648 int dev_replace_is_ongoing
)
5652 int preferred_mirror
;
5654 struct btrfs_device
*srcdev
;
5657 (BTRFS_BLOCK_GROUP_RAID1_MASK
| BTRFS_BLOCK_GROUP_RAID10
)));
5659 if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
5660 num_stripes
= map
->sub_stripes
;
5662 num_stripes
= map
->num_stripes
;
5664 switch (fs_info
->fs_devices
->read_policy
) {
5666 /* Shouldn't happen, just warn and use pid instead of failing */
5667 btrfs_warn_rl(fs_info
,
5668 "unknown read_policy type %u, reset to pid",
5669 fs_info
->fs_devices
->read_policy
);
5670 fs_info
->fs_devices
->read_policy
= BTRFS_READ_POLICY_PID
;
5672 case BTRFS_READ_POLICY_PID
:
5673 preferred_mirror
= first
+ (current
->pid
% num_stripes
);
5677 if (dev_replace_is_ongoing
&&
5678 fs_info
->dev_replace
.cont_reading_from_srcdev_mode
==
5679 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID
)
5680 srcdev
= fs_info
->dev_replace
.srcdev
;
5685 * try to avoid the drive that is the source drive for a
5686 * dev-replace procedure, only choose it if no other non-missing
5687 * mirror is available
5689 for (tolerance
= 0; tolerance
< 2; tolerance
++) {
5690 if (map
->stripes
[preferred_mirror
].dev
->bdev
&&
5691 (tolerance
|| map
->stripes
[preferred_mirror
].dev
!= srcdev
))
5692 return preferred_mirror
;
5693 for (i
= first
; i
< first
+ num_stripes
; i
++) {
5694 if (map
->stripes
[i
].dev
->bdev
&&
5695 (tolerance
|| map
->stripes
[i
].dev
!= srcdev
))
5700 /* we couldn't find one that doesn't fail. Just return something
5701 * and the io error handling code will clean up eventually
5703 return preferred_mirror
;
5706 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5707 static void sort_parity_stripes(struct btrfs_bio
*bbio
, int num_stripes
)
5714 for (i
= 0; i
< num_stripes
- 1; i
++) {
5715 /* Swap if parity is on a smaller index */
5716 if (bbio
->raid_map
[i
] > bbio
->raid_map
[i
+ 1]) {
5717 swap(bbio
->stripes
[i
], bbio
->stripes
[i
+ 1]);
5718 swap(bbio
->raid_map
[i
], bbio
->raid_map
[i
+ 1]);
5725 static struct btrfs_bio
*alloc_btrfs_bio(int total_stripes
, int real_stripes
)
5727 struct btrfs_bio
*bbio
= kzalloc(
5728 /* the size of the btrfs_bio */
5729 sizeof(struct btrfs_bio
) +
5730 /* plus the variable array for the stripes */
5731 sizeof(struct btrfs_bio_stripe
) * (total_stripes
) +
5732 /* plus the variable array for the tgt dev */
5733 sizeof(int) * (real_stripes
) +
5735 * plus the raid_map, which includes both the tgt dev
5738 sizeof(u64
) * (total_stripes
),
5739 GFP_NOFS
|__GFP_NOFAIL
);
5741 atomic_set(&bbio
->error
, 0);
5742 refcount_set(&bbio
->refs
, 1);
5744 bbio
->tgtdev_map
= (int *)(bbio
->stripes
+ total_stripes
);
5745 bbio
->raid_map
= (u64
*)(bbio
->tgtdev_map
+ real_stripes
);
5750 void btrfs_get_bbio(struct btrfs_bio
*bbio
)
5752 WARN_ON(!refcount_read(&bbio
->refs
));
5753 refcount_inc(&bbio
->refs
);
5756 void btrfs_put_bbio(struct btrfs_bio
*bbio
)
5760 if (refcount_dec_and_test(&bbio
->refs
))
5764 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5766 * Please note that, discard won't be sent to target device of device
5769 static int __btrfs_map_block_for_discard(struct btrfs_fs_info
*fs_info
,
5770 u64 logical
, u64
*length_ret
,
5771 struct btrfs_bio
**bbio_ret
)
5773 struct extent_map
*em
;
5774 struct map_lookup
*map
;
5775 struct btrfs_bio
*bbio
;
5776 u64 length
= *length_ret
;
5780 u64 stripe_end_offset
;
5787 u32 sub_stripes
= 0;
5788 u64 stripes_per_dev
= 0;
5789 u32 remaining_stripes
= 0;
5790 u32 last_stripe
= 0;
5794 /* discard always return a bbio */
5797 em
= btrfs_get_chunk_map(fs_info
, logical
, length
);
5801 map
= em
->map_lookup
;
5802 /* we don't discard raid56 yet */
5803 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
5808 offset
= logical
- em
->start
;
5809 length
= min_t(u64
, em
->start
+ em
->len
- logical
, length
);
5810 *length_ret
= length
;
5812 stripe_len
= map
->stripe_len
;
5814 * stripe_nr counts the total number of stripes we have to stride
5815 * to get to this block
5817 stripe_nr
= div64_u64(offset
, stripe_len
);
5819 /* stripe_offset is the offset of this block in its stripe */
5820 stripe_offset
= offset
- stripe_nr
* stripe_len
;
5822 stripe_nr_end
= round_up(offset
+ length
, map
->stripe_len
);
5823 stripe_nr_end
= div64_u64(stripe_nr_end
, map
->stripe_len
);
5824 stripe_cnt
= stripe_nr_end
- stripe_nr
;
5825 stripe_end_offset
= stripe_nr_end
* map
->stripe_len
-
5828 * after this, stripe_nr is the number of stripes on this
5829 * device we have to walk to find the data, and stripe_index is
5830 * the number of our device in the stripe array
5834 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5835 BTRFS_BLOCK_GROUP_RAID10
)) {
5836 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
)
5839 sub_stripes
= map
->sub_stripes
;
5841 factor
= map
->num_stripes
/ sub_stripes
;
5842 num_stripes
= min_t(u64
, map
->num_stripes
,
5843 sub_stripes
* stripe_cnt
);
5844 stripe_nr
= div_u64_rem(stripe_nr
, factor
, &stripe_index
);
5845 stripe_index
*= sub_stripes
;
5846 stripes_per_dev
= div_u64_rem(stripe_cnt
, factor
,
5847 &remaining_stripes
);
5848 div_u64_rem(stripe_nr_end
- 1, factor
, &last_stripe
);
5849 last_stripe
*= sub_stripes
;
5850 } else if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1_MASK
|
5851 BTRFS_BLOCK_GROUP_DUP
)) {
5852 num_stripes
= map
->num_stripes
;
5854 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
5858 bbio
= alloc_btrfs_bio(num_stripes
, 0);
5864 for (i
= 0; i
< num_stripes
; i
++) {
5865 bbio
->stripes
[i
].physical
=
5866 map
->stripes
[stripe_index
].physical
+
5867 stripe_offset
+ stripe_nr
* map
->stripe_len
;
5868 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
5870 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
|
5871 BTRFS_BLOCK_GROUP_RAID10
)) {
5872 bbio
->stripes
[i
].length
= stripes_per_dev
*
5875 if (i
/ sub_stripes
< remaining_stripes
)
5876 bbio
->stripes
[i
].length
+=
5880 * Special for the first stripe and
5883 * |-------|...|-------|
5887 if (i
< sub_stripes
)
5888 bbio
->stripes
[i
].length
-=
5891 if (stripe_index
>= last_stripe
&&
5892 stripe_index
<= (last_stripe
+
5894 bbio
->stripes
[i
].length
-=
5897 if (i
== sub_stripes
- 1)
5900 bbio
->stripes
[i
].length
= length
;
5904 if (stripe_index
== map
->num_stripes
) {
5911 bbio
->map_type
= map
->type
;
5912 bbio
->num_stripes
= num_stripes
;
5914 free_extent_map(em
);
5919 * In dev-replace case, for repair case (that's the only case where the mirror
5920 * is selected explicitly when calling btrfs_map_block), blocks left of the
5921 * left cursor can also be read from the target drive.
5923 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5925 * For READ, it also needs to be supported using the same mirror number.
5927 * If the requested block is not left of the left cursor, EIO is returned. This
5928 * can happen because btrfs_num_copies() returns one more in the dev-replace
5931 static int get_extra_mirror_from_replace(struct btrfs_fs_info
*fs_info
,
5932 u64 logical
, u64 length
,
5933 u64 srcdev_devid
, int *mirror_num
,
5936 struct btrfs_bio
*bbio
= NULL
;
5938 int index_srcdev
= 0;
5940 u64 physical_of_found
= 0;
5944 ret
= __btrfs_map_block(fs_info
, BTRFS_MAP_GET_READ_MIRRORS
,
5945 logical
, &length
, &bbio
, 0, 0);
5947 ASSERT(bbio
== NULL
);
5951 num_stripes
= bbio
->num_stripes
;
5952 if (*mirror_num
> num_stripes
) {
5954 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5955 * that means that the requested area is not left of the left
5958 btrfs_put_bbio(bbio
);
5963 * process the rest of the function using the mirror_num of the source
5964 * drive. Therefore look it up first. At the end, patch the device
5965 * pointer to the one of the target drive.
5967 for (i
= 0; i
< num_stripes
; i
++) {
5968 if (bbio
->stripes
[i
].dev
->devid
!= srcdev_devid
)
5972 * In case of DUP, in order to keep it simple, only add the
5973 * mirror with the lowest physical address
5976 physical_of_found
<= bbio
->stripes
[i
].physical
)
5981 physical_of_found
= bbio
->stripes
[i
].physical
;
5984 btrfs_put_bbio(bbio
);
5990 *mirror_num
= index_srcdev
+ 1;
5991 *physical
= physical_of_found
;
5995 static bool is_block_group_to_copy(struct btrfs_fs_info
*fs_info
, u64 logical
)
5997 struct btrfs_block_group
*cache
;
6000 /* Non zoned filesystem does not use "to_copy" flag */
6001 if (!btrfs_is_zoned(fs_info
))
6004 cache
= btrfs_lookup_block_group(fs_info
, logical
);
6006 spin_lock(&cache
->lock
);
6007 ret
= cache
->to_copy
;
6008 spin_unlock(&cache
->lock
);
6010 btrfs_put_block_group(cache
);
6014 static void handle_ops_on_dev_replace(enum btrfs_map_op op
,
6015 struct btrfs_bio
**bbio_ret
,
6016 struct btrfs_dev_replace
*dev_replace
,
6018 int *num_stripes_ret
, int *max_errors_ret
)
6020 struct btrfs_bio
*bbio
= *bbio_ret
;
6021 u64 srcdev_devid
= dev_replace
->srcdev
->devid
;
6022 int tgtdev_indexes
= 0;
6023 int num_stripes
= *num_stripes_ret
;
6024 int max_errors
= *max_errors_ret
;
6027 if (op
== BTRFS_MAP_WRITE
) {
6028 int index_where_to_add
;
6031 * A block group which have "to_copy" set will eventually
6032 * copied by dev-replace process. We can avoid cloning IO here.
6034 if (is_block_group_to_copy(dev_replace
->srcdev
->fs_info
, logical
))
6038 * duplicate the write operations while the dev replace
6039 * procedure is running. Since the copying of the old disk to
6040 * the new disk takes place at run time while the filesystem is
6041 * mounted writable, the regular write operations to the old
6042 * disk have to be duplicated to go to the new disk as well.
6044 * Note that device->missing is handled by the caller, and that
6045 * the write to the old disk is already set up in the stripes
6048 index_where_to_add
= num_stripes
;
6049 for (i
= 0; i
< num_stripes
; i
++) {
6050 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
6051 /* write to new disk, too */
6052 struct btrfs_bio_stripe
*new =
6053 bbio
->stripes
+ index_where_to_add
;
6054 struct btrfs_bio_stripe
*old
=
6057 new->physical
= old
->physical
;
6058 new->length
= old
->length
;
6059 new->dev
= dev_replace
->tgtdev
;
6060 bbio
->tgtdev_map
[i
] = index_where_to_add
;
6061 index_where_to_add
++;
6066 num_stripes
= index_where_to_add
;
6067 } else if (op
== BTRFS_MAP_GET_READ_MIRRORS
) {
6068 int index_srcdev
= 0;
6070 u64 physical_of_found
= 0;
6073 * During the dev-replace procedure, the target drive can also
6074 * be used to read data in case it is needed to repair a corrupt
6075 * block elsewhere. This is possible if the requested area is
6076 * left of the left cursor. In this area, the target drive is a
6077 * full copy of the source drive.
6079 for (i
= 0; i
< num_stripes
; i
++) {
6080 if (bbio
->stripes
[i
].dev
->devid
== srcdev_devid
) {
6082 * In case of DUP, in order to keep it simple,
6083 * only add the mirror with the lowest physical
6087 physical_of_found
<=
6088 bbio
->stripes
[i
].physical
)
6092 physical_of_found
= bbio
->stripes
[i
].physical
;
6096 struct btrfs_bio_stripe
*tgtdev_stripe
=
6097 bbio
->stripes
+ num_stripes
;
6099 tgtdev_stripe
->physical
= physical_of_found
;
6100 tgtdev_stripe
->length
=
6101 bbio
->stripes
[index_srcdev
].length
;
6102 tgtdev_stripe
->dev
= dev_replace
->tgtdev
;
6103 bbio
->tgtdev_map
[index_srcdev
] = num_stripes
;
6110 *num_stripes_ret
= num_stripes
;
6111 *max_errors_ret
= max_errors
;
6112 bbio
->num_tgtdevs
= tgtdev_indexes
;
6116 static bool need_full_stripe(enum btrfs_map_op op
)
6118 return (op
== BTRFS_MAP_WRITE
|| op
== BTRFS_MAP_GET_READ_MIRRORS
);
6122 * Calculate the geometry of a particular (address, len) tuple. This
6123 * information is used to calculate how big a particular bio can get before it
6124 * straddles a stripe.
6126 * @fs_info: the filesystem
6127 * @em: mapping containing the logical extent
6128 * @op: type of operation - write or read
6129 * @logical: address that we want to figure out the geometry of
6130 * @len: the length of IO we are going to perform, starting at @logical
6131 * @io_geom: pointer used to return values
6133 * Returns < 0 in case a chunk for the given logical address cannot be found,
6134 * usually shouldn't happen unless @logical is corrupted, 0 otherwise.
6136 int btrfs_get_io_geometry(struct btrfs_fs_info
*fs_info
, struct extent_map
*em
,
6137 enum btrfs_map_op op
, u64 logical
, u64 len
,
6138 struct btrfs_io_geometry
*io_geom
)
6140 struct map_lookup
*map
;
6145 u64 raid56_full_stripe_start
= (u64
)-1;
6148 ASSERT(op
!= BTRFS_MAP_DISCARD
);
6150 map
= em
->map_lookup
;
6151 /* Offset of this logical address in the chunk */
6152 offset
= logical
- em
->start
;
6153 /* Len of a stripe in a chunk */
6154 stripe_len
= map
->stripe_len
;
6155 /* Stripe wher this block falls in */
6156 stripe_nr
= div64_u64(offset
, stripe_len
);
6157 /* Offset of stripe in the chunk */
6158 stripe_offset
= stripe_nr
* stripe_len
;
6159 if (offset
< stripe_offset
) {
6161 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu",
6162 stripe_offset
, offset
, em
->start
, logical
, stripe_len
);
6166 /* stripe_offset is the offset of this block in its stripe */
6167 stripe_offset
= offset
- stripe_offset
;
6168 data_stripes
= nr_data_stripes(map
);
6170 if (map
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
6171 u64 max_len
= stripe_len
- stripe_offset
;
6174 * In case of raid56, we need to know the stripe aligned start
6176 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
6177 unsigned long full_stripe_len
= stripe_len
* data_stripes
;
6178 raid56_full_stripe_start
= offset
;
6181 * Allow a write of a full stripe, but make sure we
6182 * don't allow straddling of stripes
6184 raid56_full_stripe_start
= div64_u64(raid56_full_stripe_start
,
6186 raid56_full_stripe_start
*= full_stripe_len
;
6189 * For writes to RAID[56], allow a full stripeset across
6190 * all disks. For other RAID types and for RAID[56]
6191 * reads, just allow a single stripe (on a single disk).
6193 if (op
== BTRFS_MAP_WRITE
) {
6194 max_len
= stripe_len
* data_stripes
-
6195 (offset
- raid56_full_stripe_start
);
6198 len
= min_t(u64
, em
->len
- offset
, max_len
);
6200 len
= em
->len
- offset
;
6204 io_geom
->offset
= offset
;
6205 io_geom
->stripe_len
= stripe_len
;
6206 io_geom
->stripe_nr
= stripe_nr
;
6207 io_geom
->stripe_offset
= stripe_offset
;
6208 io_geom
->raid56_stripe_offset
= raid56_full_stripe_start
;
6213 static int __btrfs_map_block(struct btrfs_fs_info
*fs_info
,
6214 enum btrfs_map_op op
,
6215 u64 logical
, u64
*length
,
6216 struct btrfs_bio
**bbio_ret
,
6217 int mirror_num
, int need_raid_map
)
6219 struct extent_map
*em
;
6220 struct map_lookup
*map
;
6230 int tgtdev_indexes
= 0;
6231 struct btrfs_bio
*bbio
= NULL
;
6232 struct btrfs_dev_replace
*dev_replace
= &fs_info
->dev_replace
;
6233 int dev_replace_is_ongoing
= 0;
6234 int num_alloc_stripes
;
6235 int patch_the_first_stripe_for_dev_replace
= 0;
6236 u64 physical_to_patch_in_first_stripe
= 0;
6237 u64 raid56_full_stripe_start
= (u64
)-1;
6238 struct btrfs_io_geometry geom
;
6241 ASSERT(op
!= BTRFS_MAP_DISCARD
);
6243 em
= btrfs_get_chunk_map(fs_info
, logical
, *length
);
6244 ASSERT(!IS_ERR(em
));
6246 ret
= btrfs_get_io_geometry(fs_info
, em
, op
, logical
, *length
, &geom
);
6250 map
= em
->map_lookup
;
6253 stripe_len
= geom
.stripe_len
;
6254 stripe_nr
= geom
.stripe_nr
;
6255 stripe_offset
= geom
.stripe_offset
;
6256 raid56_full_stripe_start
= geom
.raid56_stripe_offset
;
6257 data_stripes
= nr_data_stripes(map
);
6259 down_read(&dev_replace
->rwsem
);
6260 dev_replace_is_ongoing
= btrfs_dev_replace_is_ongoing(dev_replace
);
6262 * Hold the semaphore for read during the whole operation, write is
6263 * requested at commit time but must wait.
6265 if (!dev_replace_is_ongoing
)
6266 up_read(&dev_replace
->rwsem
);
6268 if (dev_replace_is_ongoing
&& mirror_num
== map
->num_stripes
+ 1 &&
6269 !need_full_stripe(op
) && dev_replace
->tgtdev
!= NULL
) {
6270 ret
= get_extra_mirror_from_replace(fs_info
, logical
, *length
,
6271 dev_replace
->srcdev
->devid
,
6273 &physical_to_patch_in_first_stripe
);
6277 patch_the_first_stripe_for_dev_replace
= 1;
6278 } else if (mirror_num
> map
->num_stripes
) {
6284 if (map
->type
& BTRFS_BLOCK_GROUP_RAID0
) {
6285 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
6287 if (!need_full_stripe(op
))
6289 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID1_MASK
) {
6290 if (need_full_stripe(op
))
6291 num_stripes
= map
->num_stripes
;
6292 else if (mirror_num
)
6293 stripe_index
= mirror_num
- 1;
6295 stripe_index
= find_live_mirror(fs_info
, map
, 0,
6296 dev_replace_is_ongoing
);
6297 mirror_num
= stripe_index
+ 1;
6300 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
6301 if (need_full_stripe(op
)) {
6302 num_stripes
= map
->num_stripes
;
6303 } else if (mirror_num
) {
6304 stripe_index
= mirror_num
- 1;
6309 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
6310 u32 factor
= map
->num_stripes
/ map
->sub_stripes
;
6312 stripe_nr
= div_u64_rem(stripe_nr
, factor
, &stripe_index
);
6313 stripe_index
*= map
->sub_stripes
;
6315 if (need_full_stripe(op
))
6316 num_stripes
= map
->sub_stripes
;
6317 else if (mirror_num
)
6318 stripe_index
+= mirror_num
- 1;
6320 int old_stripe_index
= stripe_index
;
6321 stripe_index
= find_live_mirror(fs_info
, map
,
6323 dev_replace_is_ongoing
);
6324 mirror_num
= stripe_index
- old_stripe_index
+ 1;
6327 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) {
6328 if (need_raid_map
&& (need_full_stripe(op
) || mirror_num
> 1)) {
6329 /* push stripe_nr back to the start of the full stripe */
6330 stripe_nr
= div64_u64(raid56_full_stripe_start
,
6331 stripe_len
* data_stripes
);
6333 /* RAID[56] write or recovery. Return all stripes */
6334 num_stripes
= map
->num_stripes
;
6335 max_errors
= nr_parity_stripes(map
);
6337 *length
= map
->stripe_len
;
6342 * Mirror #0 or #1 means the original data block.
6343 * Mirror #2 is RAID5 parity block.
6344 * Mirror #3 is RAID6 Q block.
6346 stripe_nr
= div_u64_rem(stripe_nr
,
6347 data_stripes
, &stripe_index
);
6349 stripe_index
= data_stripes
+ mirror_num
- 2;
6351 /* We distribute the parity blocks across stripes */
6352 div_u64_rem(stripe_nr
+ stripe_index
, map
->num_stripes
,
6354 if (!need_full_stripe(op
) && mirror_num
<= 1)
6359 * after this, stripe_nr is the number of stripes on this
6360 * device we have to walk to find the data, and stripe_index is
6361 * the number of our device in the stripe array
6363 stripe_nr
= div_u64_rem(stripe_nr
, map
->num_stripes
,
6365 mirror_num
= stripe_index
+ 1;
6367 if (stripe_index
>= map
->num_stripes
) {
6369 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6370 stripe_index
, map
->num_stripes
);
6375 num_alloc_stripes
= num_stripes
;
6376 if (dev_replace_is_ongoing
&& dev_replace
->tgtdev
!= NULL
) {
6377 if (op
== BTRFS_MAP_WRITE
)
6378 num_alloc_stripes
<<= 1;
6379 if (op
== BTRFS_MAP_GET_READ_MIRRORS
)
6380 num_alloc_stripes
++;
6381 tgtdev_indexes
= num_stripes
;
6384 bbio
= alloc_btrfs_bio(num_alloc_stripes
, tgtdev_indexes
);
6390 for (i
= 0; i
< num_stripes
; i
++) {
6391 bbio
->stripes
[i
].physical
= map
->stripes
[stripe_index
].physical
+
6392 stripe_offset
+ stripe_nr
* map
->stripe_len
;
6393 bbio
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
6397 /* build raid_map */
6398 if (map
->type
& BTRFS_BLOCK_GROUP_RAID56_MASK
&& need_raid_map
&&
6399 (need_full_stripe(op
) || mirror_num
> 1)) {
6403 /* Work out the disk rotation on this stripe-set */
6404 div_u64_rem(stripe_nr
, num_stripes
, &rot
);
6406 /* Fill in the logical address of each stripe */
6407 tmp
= stripe_nr
* data_stripes
;
6408 for (i
= 0; i
< data_stripes
; i
++)
6409 bbio
->raid_map
[(i
+rot
) % num_stripes
] =
6410 em
->start
+ (tmp
+ i
) * map
->stripe_len
;
6412 bbio
->raid_map
[(i
+rot
) % map
->num_stripes
] = RAID5_P_STRIPE
;
6413 if (map
->type
& BTRFS_BLOCK_GROUP_RAID6
)
6414 bbio
->raid_map
[(i
+rot
+1) % num_stripes
] =
6417 sort_parity_stripes(bbio
, num_stripes
);
6420 if (need_full_stripe(op
))
6421 max_errors
= btrfs_chunk_max_errors(map
);
6423 if (dev_replace_is_ongoing
&& dev_replace
->tgtdev
!= NULL
&&
6424 need_full_stripe(op
)) {
6425 handle_ops_on_dev_replace(op
, &bbio
, dev_replace
, logical
,
6426 &num_stripes
, &max_errors
);
6430 bbio
->map_type
= map
->type
;
6431 bbio
->num_stripes
= num_stripes
;
6432 bbio
->max_errors
= max_errors
;
6433 bbio
->mirror_num
= mirror_num
;
6436 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6437 * mirror_num == num_stripes + 1 && dev_replace target drive is
6438 * available as a mirror
6440 if (patch_the_first_stripe_for_dev_replace
&& num_stripes
> 0) {
6441 WARN_ON(num_stripes
> 1);
6442 bbio
->stripes
[0].dev
= dev_replace
->tgtdev
;
6443 bbio
->stripes
[0].physical
= physical_to_patch_in_first_stripe
;
6444 bbio
->mirror_num
= map
->num_stripes
+ 1;
6447 if (dev_replace_is_ongoing
) {
6448 lockdep_assert_held(&dev_replace
->rwsem
);
6449 /* Unlock and let waiting writers proceed */
6450 up_read(&dev_replace
->rwsem
);
6452 free_extent_map(em
);
6456 int btrfs_map_block(struct btrfs_fs_info
*fs_info
, enum btrfs_map_op op
,
6457 u64 logical
, u64
*length
,
6458 struct btrfs_bio
**bbio_ret
, int mirror_num
)
6460 if (op
== BTRFS_MAP_DISCARD
)
6461 return __btrfs_map_block_for_discard(fs_info
, logical
,
6464 return __btrfs_map_block(fs_info
, op
, logical
, length
, bbio_ret
,
6468 /* For Scrub/replace */
6469 int btrfs_map_sblock(struct btrfs_fs_info
*fs_info
, enum btrfs_map_op op
,
6470 u64 logical
, u64
*length
,
6471 struct btrfs_bio
**bbio_ret
)
6473 return __btrfs_map_block(fs_info
, op
, logical
, length
, bbio_ret
, 0, 1);
6476 static inline void btrfs_end_bbio(struct btrfs_bio
*bbio
, struct bio
*bio
)
6478 bio
->bi_private
= bbio
->private;
6479 bio
->bi_end_io
= bbio
->end_io
;
6482 btrfs_put_bbio(bbio
);
6485 static void btrfs_end_bio(struct bio
*bio
)
6487 struct btrfs_bio
*bbio
= bio
->bi_private
;
6488 int is_orig_bio
= 0;
6490 if (bio
->bi_status
) {
6491 atomic_inc(&bbio
->error
);
6492 if (bio
->bi_status
== BLK_STS_IOERR
||
6493 bio
->bi_status
== BLK_STS_TARGET
) {
6494 struct btrfs_device
*dev
= btrfs_io_bio(bio
)->device
;
6497 if (btrfs_op(bio
) == BTRFS_MAP_WRITE
)
6498 btrfs_dev_stat_inc_and_print(dev
,
6499 BTRFS_DEV_STAT_WRITE_ERRS
);
6500 else if (!(bio
->bi_opf
& REQ_RAHEAD
))
6501 btrfs_dev_stat_inc_and_print(dev
,
6502 BTRFS_DEV_STAT_READ_ERRS
);
6503 if (bio
->bi_opf
& REQ_PREFLUSH
)
6504 btrfs_dev_stat_inc_and_print(dev
,
6505 BTRFS_DEV_STAT_FLUSH_ERRS
);
6509 if (bio
== bbio
->orig_bio
)
6512 btrfs_bio_counter_dec(bbio
->fs_info
);
6514 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
6517 bio
= bbio
->orig_bio
;
6520 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
6521 /* only send an error to the higher layers if it is
6522 * beyond the tolerance of the btrfs bio
6524 if (atomic_read(&bbio
->error
) > bbio
->max_errors
) {
6525 bio
->bi_status
= BLK_STS_IOERR
;
6528 * this bio is actually up to date, we didn't
6529 * go over the max number of errors
6531 bio
->bi_status
= BLK_STS_OK
;
6534 btrfs_end_bbio(bbio
, bio
);
6535 } else if (!is_orig_bio
) {
6540 static void submit_stripe_bio(struct btrfs_bio
*bbio
, struct bio
*bio
,
6541 u64 physical
, struct btrfs_device
*dev
)
6543 struct btrfs_fs_info
*fs_info
= bbio
->fs_info
;
6545 bio
->bi_private
= bbio
;
6546 btrfs_io_bio(bio
)->device
= dev
;
6547 bio
->bi_end_io
= btrfs_end_bio
;
6548 bio
->bi_iter
.bi_sector
= physical
>> 9;
6550 * For zone append writing, bi_sector must point the beginning of the
6553 if (bio_op(bio
) == REQ_OP_ZONE_APPEND
) {
6554 if (btrfs_dev_is_sequential(dev
, physical
)) {
6555 u64 zone_start
= round_down(physical
, fs_info
->zone_size
);
6557 bio
->bi_iter
.bi_sector
= zone_start
>> SECTOR_SHIFT
;
6559 bio
->bi_opf
&= ~REQ_OP_ZONE_APPEND
;
6560 bio
->bi_opf
|= REQ_OP_WRITE
;
6563 btrfs_debug_in_rcu(fs_info
,
6564 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6565 bio_op(bio
), bio
->bi_opf
, bio
->bi_iter
.bi_sector
,
6566 (unsigned long)dev
->bdev
->bd_dev
, rcu_str_deref(dev
->name
),
6567 dev
->devid
, bio
->bi_iter
.bi_size
);
6568 bio_set_dev(bio
, dev
->bdev
);
6570 btrfs_bio_counter_inc_noblocked(fs_info
);
6572 btrfsic_submit_bio(bio
);
6575 static void bbio_error(struct btrfs_bio
*bbio
, struct bio
*bio
, u64 logical
)
6577 atomic_inc(&bbio
->error
);
6578 if (atomic_dec_and_test(&bbio
->stripes_pending
)) {
6579 /* Should be the original bio. */
6580 WARN_ON(bio
!= bbio
->orig_bio
);
6582 btrfs_io_bio(bio
)->mirror_num
= bbio
->mirror_num
;
6583 bio
->bi_iter
.bi_sector
= logical
>> 9;
6584 if (atomic_read(&bbio
->error
) > bbio
->max_errors
)
6585 bio
->bi_status
= BLK_STS_IOERR
;
6587 bio
->bi_status
= BLK_STS_OK
;
6588 btrfs_end_bbio(bbio
, bio
);
6592 blk_status_t
btrfs_map_bio(struct btrfs_fs_info
*fs_info
, struct bio
*bio
,
6595 struct btrfs_device
*dev
;
6596 struct bio
*first_bio
= bio
;
6597 u64 logical
= bio
->bi_iter
.bi_sector
<< 9;
6603 struct btrfs_bio
*bbio
= NULL
;
6605 length
= bio
->bi_iter
.bi_size
;
6606 map_length
= length
;
6608 btrfs_bio_counter_inc_blocked(fs_info
);
6609 ret
= __btrfs_map_block(fs_info
, btrfs_op(bio
), logical
,
6610 &map_length
, &bbio
, mirror_num
, 1);
6612 btrfs_bio_counter_dec(fs_info
);
6613 return errno_to_blk_status(ret
);
6616 total_devs
= bbio
->num_stripes
;
6617 bbio
->orig_bio
= first_bio
;
6618 bbio
->private = first_bio
->bi_private
;
6619 bbio
->end_io
= first_bio
->bi_end_io
;
6620 bbio
->fs_info
= fs_info
;
6621 atomic_set(&bbio
->stripes_pending
, bbio
->num_stripes
);
6623 if ((bbio
->map_type
& BTRFS_BLOCK_GROUP_RAID56_MASK
) &&
6624 ((btrfs_op(bio
) == BTRFS_MAP_WRITE
) || (mirror_num
> 1))) {
6625 /* In this case, map_length has been set to the length of
6626 a single stripe; not the whole write */
6627 if (btrfs_op(bio
) == BTRFS_MAP_WRITE
) {
6628 ret
= raid56_parity_write(fs_info
, bio
, bbio
,
6631 ret
= raid56_parity_recover(fs_info
, bio
, bbio
,
6632 map_length
, mirror_num
, 1);
6635 btrfs_bio_counter_dec(fs_info
);
6636 return errno_to_blk_status(ret
);
6639 if (map_length
< length
) {
6641 "mapping failed logical %llu bio len %llu len %llu",
6642 logical
, length
, map_length
);
6646 for (dev_nr
= 0; dev_nr
< total_devs
; dev_nr
++) {
6647 dev
= bbio
->stripes
[dev_nr
].dev
;
6648 if (!dev
|| !dev
->bdev
|| test_bit(BTRFS_DEV_STATE_MISSING
,
6650 (btrfs_op(first_bio
) == BTRFS_MAP_WRITE
&&
6651 !test_bit(BTRFS_DEV_STATE_WRITEABLE
, &dev
->dev_state
))) {
6652 bbio_error(bbio
, first_bio
, logical
);
6656 if (dev_nr
< total_devs
- 1)
6657 bio
= btrfs_bio_clone(first_bio
);
6661 submit_stripe_bio(bbio
, bio
, bbio
->stripes
[dev_nr
].physical
, dev
);
6663 btrfs_bio_counter_dec(fs_info
);
6668 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6671 * If devid and uuid are both specified, the match must be exact, otherwise
6672 * only devid is used.
6674 * If @seed is true, traverse through the seed devices.
6676 struct btrfs_device
*btrfs_find_device(struct btrfs_fs_devices
*fs_devices
,
6677 u64 devid
, u8
*uuid
, u8
*fsid
)
6679 struct btrfs_device
*device
;
6680 struct btrfs_fs_devices
*seed_devs
;
6682 if (!fsid
|| !memcmp(fs_devices
->metadata_uuid
, fsid
, BTRFS_FSID_SIZE
)) {
6683 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
6684 if (device
->devid
== devid
&&
6685 (!uuid
|| memcmp(device
->uuid
, uuid
,
6686 BTRFS_UUID_SIZE
) == 0))
6691 list_for_each_entry(seed_devs
, &fs_devices
->seed_list
, seed_list
) {
6693 !memcmp(seed_devs
->metadata_uuid
, fsid
, BTRFS_FSID_SIZE
)) {
6694 list_for_each_entry(device
, &seed_devs
->devices
,
6696 if (device
->devid
== devid
&&
6697 (!uuid
|| memcmp(device
->uuid
, uuid
,
6698 BTRFS_UUID_SIZE
) == 0))
6707 static struct btrfs_device
*add_missing_dev(struct btrfs_fs_devices
*fs_devices
,
6708 u64 devid
, u8
*dev_uuid
)
6710 struct btrfs_device
*device
;
6711 unsigned int nofs_flag
;
6714 * We call this under the chunk_mutex, so we want to use NOFS for this
6715 * allocation, however we don't want to change btrfs_alloc_device() to
6716 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6719 nofs_flag
= memalloc_nofs_save();
6720 device
= btrfs_alloc_device(NULL
, &devid
, dev_uuid
);
6721 memalloc_nofs_restore(nofs_flag
);
6725 list_add(&device
->dev_list
, &fs_devices
->devices
);
6726 device
->fs_devices
= fs_devices
;
6727 fs_devices
->num_devices
++;
6729 set_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
);
6730 fs_devices
->missing_devices
++;
6736 * btrfs_alloc_device - allocate struct btrfs_device
6737 * @fs_info: used only for generating a new devid, can be NULL if
6738 * devid is provided (i.e. @devid != NULL).
6739 * @devid: a pointer to devid for this device. If NULL a new devid
6741 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6744 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6745 * on error. Returned struct is not linked onto any lists and must be
6746 * destroyed with btrfs_free_device.
6748 struct btrfs_device
*btrfs_alloc_device(struct btrfs_fs_info
*fs_info
,
6752 struct btrfs_device
*dev
;
6755 if (WARN_ON(!devid
&& !fs_info
))
6756 return ERR_PTR(-EINVAL
);
6758 dev
= __alloc_device(fs_info
);
6767 ret
= find_next_devid(fs_info
, &tmp
);
6769 btrfs_free_device(dev
);
6770 return ERR_PTR(ret
);
6776 memcpy(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
);
6778 generate_random_uuid(dev
->uuid
);
6783 static void btrfs_report_missing_device(struct btrfs_fs_info
*fs_info
,
6784 u64 devid
, u8
*uuid
, bool error
)
6787 btrfs_err_rl(fs_info
, "devid %llu uuid %pU is missing",
6790 btrfs_warn_rl(fs_info
, "devid %llu uuid %pU is missing",
6794 static u64
calc_stripe_length(u64 type
, u64 chunk_len
, int num_stripes
)
6796 int index
= btrfs_bg_flags_to_raid_index(type
);
6797 int ncopies
= btrfs_raid_array
[index
].ncopies
;
6798 const int nparity
= btrfs_raid_array
[index
].nparity
;
6802 data_stripes
= num_stripes
- nparity
;
6804 data_stripes
= num_stripes
/ ncopies
;
6806 return div_u64(chunk_len
, data_stripes
);
6809 #if BITS_PER_LONG == 32
6811 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
6812 * can't be accessed on 32bit systems.
6814 * This function do mount time check to reject the fs if it already has
6815 * metadata chunk beyond that limit.
6817 static int check_32bit_meta_chunk(struct btrfs_fs_info
*fs_info
,
6818 u64 logical
, u64 length
, u64 type
)
6820 if (!(type
& BTRFS_BLOCK_GROUP_METADATA
))
6823 if (logical
+ length
< MAX_LFS_FILESIZE
)
6826 btrfs_err_32bit_limit(fs_info
);
6831 * This is to give early warning for any metadata chunk reaching
6832 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
6833 * Although we can still access the metadata, it's not going to be possible
6834 * once the limit is reached.
6836 static void warn_32bit_meta_chunk(struct btrfs_fs_info
*fs_info
,
6837 u64 logical
, u64 length
, u64 type
)
6839 if (!(type
& BTRFS_BLOCK_GROUP_METADATA
))
6842 if (logical
+ length
< BTRFS_32BIT_EARLY_WARN_THRESHOLD
)
6845 btrfs_warn_32bit_limit(fs_info
);
6849 static int read_one_chunk(struct btrfs_key
*key
, struct extent_buffer
*leaf
,
6850 struct btrfs_chunk
*chunk
)
6852 struct btrfs_fs_info
*fs_info
= leaf
->fs_info
;
6853 struct extent_map_tree
*map_tree
= &fs_info
->mapping_tree
;
6854 struct map_lookup
*map
;
6855 struct extent_map
*em
;
6860 u8 uuid
[BTRFS_UUID_SIZE
];
6865 logical
= key
->offset
;
6866 length
= btrfs_chunk_length(leaf
, chunk
);
6867 type
= btrfs_chunk_type(leaf
, chunk
);
6868 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
6870 #if BITS_PER_LONG == 32
6871 ret
= check_32bit_meta_chunk(fs_info
, logical
, length
, type
);
6874 warn_32bit_meta_chunk(fs_info
, logical
, length
, type
);
6878 * Only need to verify chunk item if we're reading from sys chunk array,
6879 * as chunk item in tree block is already verified by tree-checker.
6881 if (leaf
->start
== BTRFS_SUPER_INFO_OFFSET
) {
6882 ret
= btrfs_check_chunk_valid(leaf
, chunk
, logical
);
6887 read_lock(&map_tree
->lock
);
6888 em
= lookup_extent_mapping(map_tree
, logical
, 1);
6889 read_unlock(&map_tree
->lock
);
6891 /* already mapped? */
6892 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
6893 free_extent_map(em
);
6896 free_extent_map(em
);
6899 em
= alloc_extent_map();
6902 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
6904 free_extent_map(em
);
6908 set_bit(EXTENT_FLAG_FS_MAPPING
, &em
->flags
);
6909 em
->map_lookup
= map
;
6910 em
->start
= logical
;
6913 em
->block_start
= 0;
6914 em
->block_len
= em
->len
;
6916 map
->num_stripes
= num_stripes
;
6917 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
6918 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
6919 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
6921 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
6922 map
->verified_stripes
= 0;
6923 em
->orig_block_len
= calc_stripe_length(type
, em
->len
,
6925 for (i
= 0; i
< num_stripes
; i
++) {
6926 map
->stripes
[i
].physical
=
6927 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
6928 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
6929 read_extent_buffer(leaf
, uuid
, (unsigned long)
6930 btrfs_stripe_dev_uuid_nr(chunk
, i
),
6932 map
->stripes
[i
].dev
= btrfs_find_device(fs_info
->fs_devices
,
6934 if (!map
->stripes
[i
].dev
&&
6935 !btrfs_test_opt(fs_info
, DEGRADED
)) {
6936 free_extent_map(em
);
6937 btrfs_report_missing_device(fs_info
, devid
, uuid
, true);
6940 if (!map
->stripes
[i
].dev
) {
6941 map
->stripes
[i
].dev
=
6942 add_missing_dev(fs_info
->fs_devices
, devid
,
6944 if (IS_ERR(map
->stripes
[i
].dev
)) {
6945 free_extent_map(em
);
6947 "failed to init missing dev %llu: %ld",
6948 devid
, PTR_ERR(map
->stripes
[i
].dev
));
6949 return PTR_ERR(map
->stripes
[i
].dev
);
6951 btrfs_report_missing_device(fs_info
, devid
, uuid
, false);
6953 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA
,
6954 &(map
->stripes
[i
].dev
->dev_state
));
6958 write_lock(&map_tree
->lock
);
6959 ret
= add_extent_mapping(map_tree
, em
, 0);
6960 write_unlock(&map_tree
->lock
);
6963 "failed to add chunk map, start=%llu len=%llu: %d",
6964 em
->start
, em
->len
, ret
);
6966 free_extent_map(em
);
6971 static void fill_device_from_item(struct extent_buffer
*leaf
,
6972 struct btrfs_dev_item
*dev_item
,
6973 struct btrfs_device
*device
)
6977 device
->devid
= btrfs_device_id(leaf
, dev_item
);
6978 device
->disk_total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
6979 device
->total_bytes
= device
->disk_total_bytes
;
6980 device
->commit_total_bytes
= device
->disk_total_bytes
;
6981 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
6982 device
->commit_bytes_used
= device
->bytes_used
;
6983 device
->type
= btrfs_device_type(leaf
, dev_item
);
6984 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
6985 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
6986 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
6987 WARN_ON(device
->devid
== BTRFS_DEV_REPLACE_DEVID
);
6988 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
);
6990 ptr
= btrfs_device_uuid(dev_item
);
6991 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
6994 static struct btrfs_fs_devices
*open_seed_devices(struct btrfs_fs_info
*fs_info
,
6997 struct btrfs_fs_devices
*fs_devices
;
7000 lockdep_assert_held(&uuid_mutex
);
7003 /* This will match only for multi-device seed fs */
7004 list_for_each_entry(fs_devices
, &fs_info
->fs_devices
->seed_list
, seed_list
)
7005 if (!memcmp(fs_devices
->fsid
, fsid
, BTRFS_FSID_SIZE
))
7009 fs_devices
= find_fsid(fsid
, NULL
);
7011 if (!btrfs_test_opt(fs_info
, DEGRADED
))
7012 return ERR_PTR(-ENOENT
);
7014 fs_devices
= alloc_fs_devices(fsid
, NULL
);
7015 if (IS_ERR(fs_devices
))
7018 fs_devices
->seeding
= true;
7019 fs_devices
->opened
= 1;
7024 * Upon first call for a seed fs fsid, just create a private copy of the
7025 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
7027 fs_devices
= clone_fs_devices(fs_devices
);
7028 if (IS_ERR(fs_devices
))
7031 ret
= open_fs_devices(fs_devices
, FMODE_READ
, fs_info
->bdev_holder
);
7033 free_fs_devices(fs_devices
);
7034 return ERR_PTR(ret
);
7037 if (!fs_devices
->seeding
) {
7038 close_fs_devices(fs_devices
);
7039 free_fs_devices(fs_devices
);
7040 return ERR_PTR(-EINVAL
);
7043 list_add(&fs_devices
->seed_list
, &fs_info
->fs_devices
->seed_list
);
7048 static int read_one_dev(struct extent_buffer
*leaf
,
7049 struct btrfs_dev_item
*dev_item
)
7051 struct btrfs_fs_info
*fs_info
= leaf
->fs_info
;
7052 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7053 struct btrfs_device
*device
;
7056 u8 fs_uuid
[BTRFS_FSID_SIZE
];
7057 u8 dev_uuid
[BTRFS_UUID_SIZE
];
7059 devid
= btrfs_device_id(leaf
, dev_item
);
7060 read_extent_buffer(leaf
, dev_uuid
, btrfs_device_uuid(dev_item
),
7062 read_extent_buffer(leaf
, fs_uuid
, btrfs_device_fsid(dev_item
),
7065 if (memcmp(fs_uuid
, fs_devices
->metadata_uuid
, BTRFS_FSID_SIZE
)) {
7066 fs_devices
= open_seed_devices(fs_info
, fs_uuid
);
7067 if (IS_ERR(fs_devices
))
7068 return PTR_ERR(fs_devices
);
7071 device
= btrfs_find_device(fs_info
->fs_devices
, devid
, dev_uuid
,
7074 if (!btrfs_test_opt(fs_info
, DEGRADED
)) {
7075 btrfs_report_missing_device(fs_info
, devid
,
7080 device
= add_missing_dev(fs_devices
, devid
, dev_uuid
);
7081 if (IS_ERR(device
)) {
7083 "failed to add missing dev %llu: %ld",
7084 devid
, PTR_ERR(device
));
7085 return PTR_ERR(device
);
7087 btrfs_report_missing_device(fs_info
, devid
, dev_uuid
, false);
7089 if (!device
->bdev
) {
7090 if (!btrfs_test_opt(fs_info
, DEGRADED
)) {
7091 btrfs_report_missing_device(fs_info
,
7092 devid
, dev_uuid
, true);
7095 btrfs_report_missing_device(fs_info
, devid
,
7099 if (!device
->bdev
&&
7100 !test_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
)) {
7102 * this happens when a device that was properly setup
7103 * in the device info lists suddenly goes bad.
7104 * device->bdev is NULL, and so we have to set
7105 * device->missing to one here
7107 device
->fs_devices
->missing_devices
++;
7108 set_bit(BTRFS_DEV_STATE_MISSING
, &device
->dev_state
);
7111 /* Move the device to its own fs_devices */
7112 if (device
->fs_devices
!= fs_devices
) {
7113 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING
,
7114 &device
->dev_state
));
7116 list_move(&device
->dev_list
, &fs_devices
->devices
);
7117 device
->fs_devices
->num_devices
--;
7118 fs_devices
->num_devices
++;
7120 device
->fs_devices
->missing_devices
--;
7121 fs_devices
->missing_devices
++;
7123 device
->fs_devices
= fs_devices
;
7127 if (device
->fs_devices
!= fs_info
->fs_devices
) {
7128 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
));
7129 if (device
->generation
!=
7130 btrfs_device_generation(leaf
, dev_item
))
7134 fill_device_from_item(leaf
, dev_item
, device
);
7136 u64 max_total_bytes
= i_size_read(device
->bdev
->bd_inode
);
7138 if (device
->total_bytes
> max_total_bytes
) {
7140 "device total_bytes should be at most %llu but found %llu",
7141 max_total_bytes
, device
->total_bytes
);
7145 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA
, &device
->dev_state
);
7146 if (test_bit(BTRFS_DEV_STATE_WRITEABLE
, &device
->dev_state
) &&
7147 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT
, &device
->dev_state
)) {
7148 device
->fs_devices
->total_rw_bytes
+= device
->total_bytes
;
7149 atomic64_add(device
->total_bytes
- device
->bytes_used
,
7150 &fs_info
->free_chunk_space
);
7156 int btrfs_read_sys_array(struct btrfs_fs_info
*fs_info
)
7158 struct btrfs_root
*root
= fs_info
->tree_root
;
7159 struct btrfs_super_block
*super_copy
= fs_info
->super_copy
;
7160 struct extent_buffer
*sb
;
7161 struct btrfs_disk_key
*disk_key
;
7162 struct btrfs_chunk
*chunk
;
7164 unsigned long sb_array_offset
;
7171 struct btrfs_key key
;
7173 ASSERT(BTRFS_SUPER_INFO_SIZE
<= fs_info
->nodesize
);
7175 * This will create extent buffer of nodesize, superblock size is
7176 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
7177 * overallocate but we can keep it as-is, only the first page is used.
7179 sb
= btrfs_find_create_tree_block(fs_info
, BTRFS_SUPER_INFO_OFFSET
,
7180 root
->root_key
.objectid
, 0);
7183 set_extent_buffer_uptodate(sb
);
7185 * The sb extent buffer is artificial and just used to read the system array.
7186 * set_extent_buffer_uptodate() call does not properly mark all it's
7187 * pages up-to-date when the page is larger: extent does not cover the
7188 * whole page and consequently check_page_uptodate does not find all
7189 * the page's extents up-to-date (the hole beyond sb),
7190 * write_extent_buffer then triggers a WARN_ON.
7192 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
7193 * but sb spans only this function. Add an explicit SetPageUptodate call
7194 * to silence the warning eg. on PowerPC 64.
7196 if (PAGE_SIZE
> BTRFS_SUPER_INFO_SIZE
)
7197 SetPageUptodate(sb
->pages
[0]);
7199 write_extent_buffer(sb
, super_copy
, 0, BTRFS_SUPER_INFO_SIZE
);
7200 array_size
= btrfs_super_sys_array_size(super_copy
);
7202 array_ptr
= super_copy
->sys_chunk_array
;
7203 sb_array_offset
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
7206 while (cur_offset
< array_size
) {
7207 disk_key
= (struct btrfs_disk_key
*)array_ptr
;
7208 len
= sizeof(*disk_key
);
7209 if (cur_offset
+ len
> array_size
)
7210 goto out_short_read
;
7212 btrfs_disk_key_to_cpu(&key
, disk_key
);
7215 sb_array_offset
+= len
;
7218 if (key
.type
!= BTRFS_CHUNK_ITEM_KEY
) {
7220 "unexpected item type %u in sys_array at offset %u",
7221 (u32
)key
.type
, cur_offset
);
7226 chunk
= (struct btrfs_chunk
*)sb_array_offset
;
7228 * At least one btrfs_chunk with one stripe must be present,
7229 * exact stripe count check comes afterwards
7231 len
= btrfs_chunk_item_size(1);
7232 if (cur_offset
+ len
> array_size
)
7233 goto out_short_read
;
7235 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
7238 "invalid number of stripes %u in sys_array at offset %u",
7239 num_stripes
, cur_offset
);
7244 type
= btrfs_chunk_type(sb
, chunk
);
7245 if ((type
& BTRFS_BLOCK_GROUP_SYSTEM
) == 0) {
7247 "invalid chunk type %llu in sys_array at offset %u",
7253 len
= btrfs_chunk_item_size(num_stripes
);
7254 if (cur_offset
+ len
> array_size
)
7255 goto out_short_read
;
7257 ret
= read_one_chunk(&key
, sb
, chunk
);
7262 sb_array_offset
+= len
;
7265 clear_extent_buffer_uptodate(sb
);
7266 free_extent_buffer_stale(sb
);
7270 btrfs_err(fs_info
, "sys_array too short to read %u bytes at offset %u",
7272 clear_extent_buffer_uptodate(sb
);
7273 free_extent_buffer_stale(sb
);
7278 * Check if all chunks in the fs are OK for read-write degraded mount
7280 * If the @failing_dev is specified, it's accounted as missing.
7282 * Return true if all chunks meet the minimal RW mount requirements.
7283 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7285 bool btrfs_check_rw_degradable(struct btrfs_fs_info
*fs_info
,
7286 struct btrfs_device
*failing_dev
)
7288 struct extent_map_tree
*map_tree
= &fs_info
->mapping_tree
;
7289 struct extent_map
*em
;
7293 read_lock(&map_tree
->lock
);
7294 em
= lookup_extent_mapping(map_tree
, 0, (u64
)-1);
7295 read_unlock(&map_tree
->lock
);
7296 /* No chunk at all? Return false anyway */
7302 struct map_lookup
*map
;
7307 map
= em
->map_lookup
;
7309 btrfs_get_num_tolerated_disk_barrier_failures(
7311 for (i
= 0; i
< map
->num_stripes
; i
++) {
7312 struct btrfs_device
*dev
= map
->stripes
[i
].dev
;
7314 if (!dev
|| !dev
->bdev
||
7315 test_bit(BTRFS_DEV_STATE_MISSING
, &dev
->dev_state
) ||
7316 dev
->last_flush_error
)
7318 else if (failing_dev
&& failing_dev
== dev
)
7321 if (missing
> max_tolerated
) {
7324 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7325 em
->start
, missing
, max_tolerated
);
7326 free_extent_map(em
);
7330 next_start
= extent_map_end(em
);
7331 free_extent_map(em
);
7333 read_lock(&map_tree
->lock
);
7334 em
= lookup_extent_mapping(map_tree
, next_start
,
7335 (u64
)(-1) - next_start
);
7336 read_unlock(&map_tree
->lock
);
7342 static void readahead_tree_node_children(struct extent_buffer
*node
)
7345 const int nr_items
= btrfs_header_nritems(node
);
7347 for (i
= 0; i
< nr_items
; i
++)
7348 btrfs_readahead_node_child(node
, i
);
7351 int btrfs_read_chunk_tree(struct btrfs_fs_info
*fs_info
)
7353 struct btrfs_root
*root
= fs_info
->chunk_root
;
7354 struct btrfs_path
*path
;
7355 struct extent_buffer
*leaf
;
7356 struct btrfs_key key
;
7357 struct btrfs_key found_key
;
7361 u64 last_ra_node
= 0;
7363 path
= btrfs_alloc_path();
7368 * uuid_mutex is needed only if we are mounting a sprout FS
7369 * otherwise we don't need it.
7371 mutex_lock(&uuid_mutex
);
7374 * It is possible for mount and umount to race in such a way that
7375 * we execute this code path, but open_fs_devices failed to clear
7376 * total_rw_bytes. We certainly want it cleared before reading the
7377 * device items, so clear it here.
7379 fs_info
->fs_devices
->total_rw_bytes
= 0;
7382 * Read all device items, and then all the chunk items. All
7383 * device items are found before any chunk item (their object id
7384 * is smaller than the lowest possible object id for a chunk
7385 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7387 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
7390 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
7394 struct extent_buffer
*node
;
7396 leaf
= path
->nodes
[0];
7397 slot
= path
->slots
[0];
7398 if (slot
>= btrfs_header_nritems(leaf
)) {
7399 ret
= btrfs_next_leaf(root
, path
);
7407 * The nodes on level 1 are not locked but we don't need to do
7408 * that during mount time as nothing else can access the tree
7410 node
= path
->nodes
[1];
7412 if (last_ra_node
!= node
->start
) {
7413 readahead_tree_node_children(node
);
7414 last_ra_node
= node
->start
;
7417 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
7418 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
7419 struct btrfs_dev_item
*dev_item
;
7420 dev_item
= btrfs_item_ptr(leaf
, slot
,
7421 struct btrfs_dev_item
);
7422 ret
= read_one_dev(leaf
, dev_item
);
7426 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
7427 struct btrfs_chunk
*chunk
;
7428 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
7429 mutex_lock(&fs_info
->chunk_mutex
);
7430 ret
= read_one_chunk(&found_key
, leaf
, chunk
);
7431 mutex_unlock(&fs_info
->chunk_mutex
);
7439 * After loading chunk tree, we've got all device information,
7440 * do another round of validation checks.
7442 if (total_dev
!= fs_info
->fs_devices
->total_devices
) {
7444 "super_num_devices %llu mismatch with num_devices %llu found here",
7445 btrfs_super_num_devices(fs_info
->super_copy
),
7450 if (btrfs_super_total_bytes(fs_info
->super_copy
) <
7451 fs_info
->fs_devices
->total_rw_bytes
) {
7453 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7454 btrfs_super_total_bytes(fs_info
->super_copy
),
7455 fs_info
->fs_devices
->total_rw_bytes
);
7461 mutex_unlock(&uuid_mutex
);
7463 btrfs_free_path(path
);
7467 void btrfs_init_devices_late(struct btrfs_fs_info
*fs_info
)
7469 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
, *seed_devs
;
7470 struct btrfs_device
*device
;
7472 fs_devices
->fs_info
= fs_info
;
7474 mutex_lock(&fs_devices
->device_list_mutex
);
7475 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
)
7476 device
->fs_info
= fs_info
;
7478 list_for_each_entry(seed_devs
, &fs_devices
->seed_list
, seed_list
) {
7479 list_for_each_entry(device
, &seed_devs
->devices
, dev_list
)
7480 device
->fs_info
= fs_info
;
7482 seed_devs
->fs_info
= fs_info
;
7484 mutex_unlock(&fs_devices
->device_list_mutex
);
7487 static u64
btrfs_dev_stats_value(const struct extent_buffer
*eb
,
7488 const struct btrfs_dev_stats_item
*ptr
,
7493 read_extent_buffer(eb
, &val
,
7494 offsetof(struct btrfs_dev_stats_item
, values
) +
7495 ((unsigned long)ptr
) + (index
* sizeof(u64
)),
7500 static void btrfs_set_dev_stats_value(struct extent_buffer
*eb
,
7501 struct btrfs_dev_stats_item
*ptr
,
7504 write_extent_buffer(eb
, &val
,
7505 offsetof(struct btrfs_dev_stats_item
, values
) +
7506 ((unsigned long)ptr
) + (index
* sizeof(u64
)),
7510 static int btrfs_device_init_dev_stats(struct btrfs_device
*device
,
7511 struct btrfs_path
*path
)
7513 struct btrfs_dev_stats_item
*ptr
;
7514 struct extent_buffer
*eb
;
7515 struct btrfs_key key
;
7519 if (!device
->fs_info
->dev_root
)
7522 key
.objectid
= BTRFS_DEV_STATS_OBJECTID
;
7523 key
.type
= BTRFS_PERSISTENT_ITEM_KEY
;
7524 key
.offset
= device
->devid
;
7525 ret
= btrfs_search_slot(NULL
, device
->fs_info
->dev_root
, &key
, path
, 0, 0);
7527 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7528 btrfs_dev_stat_set(device
, i
, 0);
7529 device
->dev_stats_valid
= 1;
7530 btrfs_release_path(path
);
7531 return ret
< 0 ? ret
: 0;
7533 slot
= path
->slots
[0];
7534 eb
= path
->nodes
[0];
7535 item_size
= btrfs_item_size_nr(eb
, slot
);
7537 ptr
= btrfs_item_ptr(eb
, slot
, struct btrfs_dev_stats_item
);
7539 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
7540 if (item_size
>= (1 + i
) * sizeof(__le64
))
7541 btrfs_dev_stat_set(device
, i
,
7542 btrfs_dev_stats_value(eb
, ptr
, i
));
7544 btrfs_dev_stat_set(device
, i
, 0);
7547 device
->dev_stats_valid
= 1;
7548 btrfs_dev_stat_print_on_load(device
);
7549 btrfs_release_path(path
);
7554 int btrfs_init_dev_stats(struct btrfs_fs_info
*fs_info
)
7556 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
, *seed_devs
;
7557 struct btrfs_device
*device
;
7558 struct btrfs_path
*path
= NULL
;
7561 path
= btrfs_alloc_path();
7565 mutex_lock(&fs_devices
->device_list_mutex
);
7566 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
7567 ret
= btrfs_device_init_dev_stats(device
, path
);
7571 list_for_each_entry(seed_devs
, &fs_devices
->seed_list
, seed_list
) {
7572 list_for_each_entry(device
, &seed_devs
->devices
, dev_list
) {
7573 ret
= btrfs_device_init_dev_stats(device
, path
);
7579 mutex_unlock(&fs_devices
->device_list_mutex
);
7581 btrfs_free_path(path
);
7585 static int update_dev_stat_item(struct btrfs_trans_handle
*trans
,
7586 struct btrfs_device
*device
)
7588 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
7589 struct btrfs_root
*dev_root
= fs_info
->dev_root
;
7590 struct btrfs_path
*path
;
7591 struct btrfs_key key
;
7592 struct extent_buffer
*eb
;
7593 struct btrfs_dev_stats_item
*ptr
;
7597 key
.objectid
= BTRFS_DEV_STATS_OBJECTID
;
7598 key
.type
= BTRFS_PERSISTENT_ITEM_KEY
;
7599 key
.offset
= device
->devid
;
7601 path
= btrfs_alloc_path();
7604 ret
= btrfs_search_slot(trans
, dev_root
, &key
, path
, -1, 1);
7606 btrfs_warn_in_rcu(fs_info
,
7607 "error %d while searching for dev_stats item for device %s",
7608 ret
, rcu_str_deref(device
->name
));
7613 btrfs_item_size_nr(path
->nodes
[0], path
->slots
[0]) < sizeof(*ptr
)) {
7614 /* need to delete old one and insert a new one */
7615 ret
= btrfs_del_item(trans
, dev_root
, path
);
7617 btrfs_warn_in_rcu(fs_info
,
7618 "delete too small dev_stats item for device %s failed %d",
7619 rcu_str_deref(device
->name
), ret
);
7626 /* need to insert a new item */
7627 btrfs_release_path(path
);
7628 ret
= btrfs_insert_empty_item(trans
, dev_root
, path
,
7629 &key
, sizeof(*ptr
));
7631 btrfs_warn_in_rcu(fs_info
,
7632 "insert dev_stats item for device %s failed %d",
7633 rcu_str_deref(device
->name
), ret
);
7638 eb
= path
->nodes
[0];
7639 ptr
= btrfs_item_ptr(eb
, path
->slots
[0], struct btrfs_dev_stats_item
);
7640 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7641 btrfs_set_dev_stats_value(eb
, ptr
, i
,
7642 btrfs_dev_stat_read(device
, i
));
7643 btrfs_mark_buffer_dirty(eb
);
7646 btrfs_free_path(path
);
7651 * called from commit_transaction. Writes all changed device stats to disk.
7653 int btrfs_run_dev_stats(struct btrfs_trans_handle
*trans
)
7655 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
7656 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7657 struct btrfs_device
*device
;
7661 mutex_lock(&fs_devices
->device_list_mutex
);
7662 list_for_each_entry(device
, &fs_devices
->devices
, dev_list
) {
7663 stats_cnt
= atomic_read(&device
->dev_stats_ccnt
);
7664 if (!device
->dev_stats_valid
|| stats_cnt
== 0)
7669 * There is a LOAD-LOAD control dependency between the value of
7670 * dev_stats_ccnt and updating the on-disk values which requires
7671 * reading the in-memory counters. Such control dependencies
7672 * require explicit read memory barriers.
7674 * This memory barriers pairs with smp_mb__before_atomic in
7675 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7676 * barrier implied by atomic_xchg in
7677 * btrfs_dev_stats_read_and_reset
7681 ret
= update_dev_stat_item(trans
, device
);
7683 atomic_sub(stats_cnt
, &device
->dev_stats_ccnt
);
7685 mutex_unlock(&fs_devices
->device_list_mutex
);
7690 void btrfs_dev_stat_inc_and_print(struct btrfs_device
*dev
, int index
)
7692 btrfs_dev_stat_inc(dev
, index
);
7693 btrfs_dev_stat_print_on_error(dev
);
7696 static void btrfs_dev_stat_print_on_error(struct btrfs_device
*dev
)
7698 if (!dev
->dev_stats_valid
)
7700 btrfs_err_rl_in_rcu(dev
->fs_info
,
7701 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7702 rcu_str_deref(dev
->name
),
7703 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
7704 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
7705 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
7706 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
7707 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
7710 static void btrfs_dev_stat_print_on_load(struct btrfs_device
*dev
)
7714 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7715 if (btrfs_dev_stat_read(dev
, i
) != 0)
7717 if (i
== BTRFS_DEV_STAT_VALUES_MAX
)
7718 return; /* all values == 0, suppress message */
7720 btrfs_info_in_rcu(dev
->fs_info
,
7721 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7722 rcu_str_deref(dev
->name
),
7723 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_WRITE_ERRS
),
7724 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_READ_ERRS
),
7725 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_FLUSH_ERRS
),
7726 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
),
7727 btrfs_dev_stat_read(dev
, BTRFS_DEV_STAT_GENERATION_ERRS
));
7730 int btrfs_get_dev_stats(struct btrfs_fs_info
*fs_info
,
7731 struct btrfs_ioctl_get_dev_stats
*stats
)
7733 struct btrfs_device
*dev
;
7734 struct btrfs_fs_devices
*fs_devices
= fs_info
->fs_devices
;
7737 mutex_lock(&fs_devices
->device_list_mutex
);
7738 dev
= btrfs_find_device(fs_info
->fs_devices
, stats
->devid
, NULL
, NULL
);
7739 mutex_unlock(&fs_devices
->device_list_mutex
);
7742 btrfs_warn(fs_info
, "get dev_stats failed, device not found");
7744 } else if (!dev
->dev_stats_valid
) {
7745 btrfs_warn(fs_info
, "get dev_stats failed, not yet valid");
7747 } else if (stats
->flags
& BTRFS_DEV_STATS_RESET
) {
7748 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++) {
7749 if (stats
->nr_items
> i
)
7751 btrfs_dev_stat_read_and_reset(dev
, i
);
7753 btrfs_dev_stat_set(dev
, i
, 0);
7755 btrfs_info(fs_info
, "device stats zeroed by %s (%d)",
7756 current
->comm
, task_pid_nr(current
));
7758 for (i
= 0; i
< BTRFS_DEV_STAT_VALUES_MAX
; i
++)
7759 if (stats
->nr_items
> i
)
7760 stats
->values
[i
] = btrfs_dev_stat_read(dev
, i
);
7762 if (stats
->nr_items
> BTRFS_DEV_STAT_VALUES_MAX
)
7763 stats
->nr_items
= BTRFS_DEV_STAT_VALUES_MAX
;
7768 * Update the size and bytes used for each device where it changed. This is
7769 * delayed since we would otherwise get errors while writing out the
7772 * Must be invoked during transaction commit.
7774 void btrfs_commit_device_sizes(struct btrfs_transaction
*trans
)
7776 struct btrfs_device
*curr
, *next
;
7778 ASSERT(trans
->state
== TRANS_STATE_COMMIT_DOING
);
7780 if (list_empty(&trans
->dev_update_list
))
7784 * We don't need the device_list_mutex here. This list is owned by the
7785 * transaction and the transaction must complete before the device is
7788 mutex_lock(&trans
->fs_info
->chunk_mutex
);
7789 list_for_each_entry_safe(curr
, next
, &trans
->dev_update_list
,
7791 list_del_init(&curr
->post_commit_list
);
7792 curr
->commit_total_bytes
= curr
->disk_total_bytes
;
7793 curr
->commit_bytes_used
= curr
->bytes_used
;
7795 mutex_unlock(&trans
->fs_info
->chunk_mutex
);
7799 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7801 int btrfs_bg_type_to_factor(u64 flags
)
7803 const int index
= btrfs_bg_flags_to_raid_index(flags
);
7805 return btrfs_raid_array
[index
].ncopies
;
7810 static int verify_one_dev_extent(struct btrfs_fs_info
*fs_info
,
7811 u64 chunk_offset
, u64 devid
,
7812 u64 physical_offset
, u64 physical_len
)
7814 struct extent_map_tree
*em_tree
= &fs_info
->mapping_tree
;
7815 struct extent_map
*em
;
7816 struct map_lookup
*map
;
7817 struct btrfs_device
*dev
;
7823 read_lock(&em_tree
->lock
);
7824 em
= lookup_extent_mapping(em_tree
, chunk_offset
, 1);
7825 read_unlock(&em_tree
->lock
);
7829 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7830 physical_offset
, devid
);
7835 map
= em
->map_lookup
;
7836 stripe_len
= calc_stripe_length(map
->type
, em
->len
, map
->num_stripes
);
7837 if (physical_len
!= stripe_len
) {
7839 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7840 physical_offset
, devid
, em
->start
, physical_len
,
7846 for (i
= 0; i
< map
->num_stripes
; i
++) {
7847 if (map
->stripes
[i
].dev
->devid
== devid
&&
7848 map
->stripes
[i
].physical
== physical_offset
) {
7850 if (map
->verified_stripes
>= map
->num_stripes
) {
7852 "too many dev extents for chunk %llu found",
7857 map
->verified_stripes
++;
7863 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7864 physical_offset
, devid
);
7868 /* Make sure no dev extent is beyond device bondary */
7869 dev
= btrfs_find_device(fs_info
->fs_devices
, devid
, NULL
, NULL
);
7871 btrfs_err(fs_info
, "failed to find devid %llu", devid
);
7876 if (physical_offset
+ physical_len
> dev
->disk_total_bytes
) {
7878 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7879 devid
, physical_offset
, physical_len
,
7880 dev
->disk_total_bytes
);
7885 if (dev
->zone_info
) {
7886 u64 zone_size
= dev
->zone_info
->zone_size
;
7888 if (!IS_ALIGNED(physical_offset
, zone_size
) ||
7889 !IS_ALIGNED(physical_len
, zone_size
)) {
7891 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
7892 devid
, physical_offset
, physical_len
);
7899 free_extent_map(em
);
7903 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info
*fs_info
)
7905 struct extent_map_tree
*em_tree
= &fs_info
->mapping_tree
;
7906 struct extent_map
*em
;
7907 struct rb_node
*node
;
7910 read_lock(&em_tree
->lock
);
7911 for (node
= rb_first_cached(&em_tree
->map
); node
; node
= rb_next(node
)) {
7912 em
= rb_entry(node
, struct extent_map
, rb_node
);
7913 if (em
->map_lookup
->num_stripes
!=
7914 em
->map_lookup
->verified_stripes
) {
7916 "chunk %llu has missing dev extent, have %d expect %d",
7917 em
->start
, em
->map_lookup
->verified_stripes
,
7918 em
->map_lookup
->num_stripes
);
7924 read_unlock(&em_tree
->lock
);
7929 * Ensure that all dev extents are mapped to correct chunk, otherwise
7930 * later chunk allocation/free would cause unexpected behavior.
7932 * NOTE: This will iterate through the whole device tree, which should be of
7933 * the same size level as the chunk tree. This slightly increases mount time.
7935 int btrfs_verify_dev_extents(struct btrfs_fs_info
*fs_info
)
7937 struct btrfs_path
*path
;
7938 struct btrfs_root
*root
= fs_info
->dev_root
;
7939 struct btrfs_key key
;
7941 u64 prev_dev_ext_end
= 0;
7945 * We don't have a dev_root because we mounted with ignorebadroots and
7946 * failed to load the root, so we want to skip the verification in this
7949 * However if the dev root is fine, but the tree itself is corrupted
7950 * we'd still fail to mount. This verification is only to make sure
7951 * writes can happen safely, so instead just bypass this check
7952 * completely in the case of IGNOREBADROOTS.
7954 if (btrfs_test_opt(fs_info
, IGNOREBADROOTS
))
7958 key
.type
= BTRFS_DEV_EXTENT_KEY
;
7961 path
= btrfs_alloc_path();
7965 path
->reada
= READA_FORWARD
;
7966 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
7970 if (path
->slots
[0] >= btrfs_header_nritems(path
->nodes
[0])) {
7971 ret
= btrfs_next_item(root
, path
);
7974 /* No dev extents at all? Not good */
7981 struct extent_buffer
*leaf
= path
->nodes
[0];
7982 struct btrfs_dev_extent
*dext
;
7983 int slot
= path
->slots
[0];
7985 u64 physical_offset
;
7989 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
7990 if (key
.type
!= BTRFS_DEV_EXTENT_KEY
)
7992 devid
= key
.objectid
;
7993 physical_offset
= key
.offset
;
7995 dext
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dev_extent
);
7996 chunk_offset
= btrfs_dev_extent_chunk_offset(leaf
, dext
);
7997 physical_len
= btrfs_dev_extent_length(leaf
, dext
);
7999 /* Check if this dev extent overlaps with the previous one */
8000 if (devid
== prev_devid
&& physical_offset
< prev_dev_ext_end
) {
8002 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
8003 devid
, physical_offset
, prev_dev_ext_end
);
8008 ret
= verify_one_dev_extent(fs_info
, chunk_offset
, devid
,
8009 physical_offset
, physical_len
);
8013 prev_dev_ext_end
= physical_offset
+ physical_len
;
8015 ret
= btrfs_next_item(root
, path
);
8024 /* Ensure all chunks have corresponding dev extents */
8025 ret
= verify_chunk_dev_extent_mapping(fs_info
);
8027 btrfs_free_path(path
);
8032 * Check whether the given block group or device is pinned by any inode being
8033 * used as a swapfile.
8035 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info
*fs_info
, void *ptr
)
8037 struct btrfs_swapfile_pin
*sp
;
8038 struct rb_node
*node
;
8040 spin_lock(&fs_info
->swapfile_pins_lock
);
8041 node
= fs_info
->swapfile_pins
.rb_node
;
8043 sp
= rb_entry(node
, struct btrfs_swapfile_pin
, node
);
8045 node
= node
->rb_left
;
8046 else if (ptr
> sp
->ptr
)
8047 node
= node
->rb_right
;
8051 spin_unlock(&fs_info
->swapfile_pins_lock
);
8052 return node
!= NULL
;
8055 static int relocating_repair_kthread(void *data
)
8057 struct btrfs_block_group
*cache
= (struct btrfs_block_group
*)data
;
8058 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
8062 target
= cache
->start
;
8063 btrfs_put_block_group(cache
);
8065 if (!btrfs_exclop_start(fs_info
, BTRFS_EXCLOP_BALANCE
)) {
8067 "zoned: skip relocating block group %llu to repair: EBUSY",
8072 mutex_lock(&fs_info
->reclaim_bgs_lock
);
8074 /* Ensure block group still exists */
8075 cache
= btrfs_lookup_block_group(fs_info
, target
);
8079 if (!cache
->relocating_repair
)
8082 ret
= btrfs_may_alloc_data_chunk(fs_info
, target
);
8087 "zoned: relocating block group %llu to repair IO failure",
8089 ret
= btrfs_relocate_chunk(fs_info
, target
);
8093 btrfs_put_block_group(cache
);
8094 mutex_unlock(&fs_info
->reclaim_bgs_lock
);
8095 btrfs_exclop_finish(fs_info
);
8100 int btrfs_repair_one_zone(struct btrfs_fs_info
*fs_info
, u64 logical
)
8102 struct btrfs_block_group
*cache
;
8104 /* Do not attempt to repair in degraded state */
8105 if (btrfs_test_opt(fs_info
, DEGRADED
))
8108 cache
= btrfs_lookup_block_group(fs_info
, logical
);
8112 spin_lock(&cache
->lock
);
8113 if (cache
->relocating_repair
) {
8114 spin_unlock(&cache
->lock
);
8115 btrfs_put_block_group(cache
);
8118 cache
->relocating_repair
= 1;
8119 spin_unlock(&cache
->lock
);
8121 kthread_run(relocating_repair_kthread
, cache
,
8122 "btrfs-relocating-repair");