1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/sched/mm.h>
8 #include <linux/slab.h>
9 #include <linux/ratelimit.h>
10 #include <linux/kthread.h>
11 #include <linux/semaphore.h>
12 #include <linux/uuid.h>
13 #include <linux/list_sort.h>
14 #include <linux/namei.h>
17 #include "extent_map.h"
19 #include "transaction.h"
20 #include "print-tree.h"
23 #include "rcu-string.h"
24 #include "dev-replace.h"
26 #include "tree-checker.h"
27 #include "space-info.h"
28 #include "block-group.h"
32 #include "accessors.h"
33 #include "uuid-tree.h"
35 #include "relocation.h"
39 #define BTRFS_BLOCK_GROUP_STRIPE_MASK (BTRFS_BLOCK_GROUP_RAID0 | \
40 BTRFS_BLOCK_GROUP_RAID10 | \
41 BTRFS_BLOCK_GROUP_RAID56_MASK)
43 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
44 [BTRFS_RAID_RAID10] = {
47 .devs_max = 0, /* 0 == as many as possible */
49 .tolerated_failures = 1,
53 .raid_name = "raid10",
54 .bg_flag = BTRFS_BLOCK_GROUP_RAID10,
55 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
57 [BTRFS_RAID_RAID1] = {
62 .tolerated_failures = 1,
67 .bg_flag = BTRFS_BLOCK_GROUP_RAID1,
68 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
70 [BTRFS_RAID_RAID1C3] = {
75 .tolerated_failures = 2,
79 .raid_name = "raid1c3",
80 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3,
81 .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET,
83 [BTRFS_RAID_RAID1C4] = {
88 .tolerated_failures = 3,
92 .raid_name = "raid1c4",
93 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4,
94 .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET,
101 .tolerated_failures = 0,
106 .bg_flag = BTRFS_BLOCK_GROUP_DUP,
109 [BTRFS_RAID_RAID0] = {
114 .tolerated_failures = 0,
118 .raid_name = "raid0",
119 .bg_flag = BTRFS_BLOCK_GROUP_RAID0,
122 [BTRFS_RAID_SINGLE] = {
127 .tolerated_failures = 0,
131 .raid_name = "single",
135 [BTRFS_RAID_RAID5] = {
140 .tolerated_failures = 1,
144 .raid_name = "raid5",
145 .bg_flag = BTRFS_BLOCK_GROUP_RAID5,
146 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
148 [BTRFS_RAID_RAID6] = {
153 .tolerated_failures = 2,
157 .raid_name = "raid6",
158 .bg_flag = BTRFS_BLOCK_GROUP_RAID6,
159 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
164 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which
165 * can be used as index to access btrfs_raid_array[].
167 enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags)
169 const u64 profile = (flags & BTRFS_BLOCK_GROUP_PROFILE_MASK);
172 return BTRFS_RAID_SINGLE;
174 return BTRFS_BG_FLAG_TO_INDEX(profile);
177 const char *btrfs_bg_type_to_raid_name(u64 flags)
179 const int index = btrfs_bg_flags_to_raid_index(flags);
181 if (index >= BTRFS_NR_RAID_TYPES)
184 return btrfs_raid_array[index].raid_name;
187 int btrfs_nr_parity_stripes(u64 type)
189 enum btrfs_raid_types index = btrfs_bg_flags_to_raid_index(type);
191 return btrfs_raid_array[index].nparity;
195 * Fill @buf with textual description of @bg_flags, no more than @size_buf
196 * bytes including terminating null byte.
198 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf)
203 u64 flags = bg_flags;
204 u32 size_bp = size_buf;
211 #define DESCRIBE_FLAG(flag, desc) \
213 if (flags & (flag)) { \
214 ret = snprintf(bp, size_bp, "%s|", (desc)); \
215 if (ret < 0 || ret >= size_bp) \
223 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data");
224 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system");
225 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata");
227 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single");
228 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
229 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag,
230 btrfs_raid_array[i].raid_name);
234 ret = snprintf(bp, size_bp, "0x%llx|", flags);
238 if (size_bp < size_buf)
239 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */
242 * The text is trimmed, it's up to the caller to provide sufficiently
248 static int init_first_rw_device(struct btrfs_trans_handle *trans);
249 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
250 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
256 * There are several mutexes that protect manipulation of devices and low-level
257 * structures like chunks but not block groups, extents or files
259 * uuid_mutex (global lock)
260 * ------------------------
261 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
262 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
263 * device) or requested by the device= mount option
265 * the mutex can be very coarse and can cover long-running operations
267 * protects: updates to fs_devices counters like missing devices, rw devices,
268 * seeding, structure cloning, opening/closing devices at mount/umount time
270 * global::fs_devs - add, remove, updates to the global list
272 * does not protect: manipulation of the fs_devices::devices list in general
273 * but in mount context it could be used to exclude list modifications by eg.
276 * btrfs_device::name - renames (write side), read is RCU
278 * fs_devices::device_list_mutex (per-fs, with RCU)
279 * ------------------------------------------------
280 * protects updates to fs_devices::devices, ie. adding and deleting
282 * simple list traversal with read-only actions can be done with RCU protection
284 * may be used to exclude some operations from running concurrently without any
285 * modifications to the list (see write_all_supers)
287 * Is not required at mount and close times, because our device list is
288 * protected by the uuid_mutex at that point.
292 * protects balance structures (status, state) and context accessed from
293 * several places (internally, ioctl)
297 * protects chunks, adding or removing during allocation, trim or when a new
298 * device is added/removed. Additionally it also protects post_commit_list of
299 * individual devices, since they can be added to the transaction's
300 * post_commit_list only with chunk_mutex held.
304 * a big lock that is held by the cleaner thread and prevents running subvolume
305 * cleaning together with relocation or delayed iputs
317 * Exclusive operations
318 * ====================
320 * Maintains the exclusivity of the following operations that apply to the
321 * whole filesystem and cannot run in parallel.
326 * - Device replace (*)
329 * The device operations (as above) can be in one of the following states:
335 * Only device operations marked with (*) can go into the Paused state for the
338 * - ioctl (only Balance can be Paused through ioctl)
339 * - filesystem remounted as read-only
340 * - filesystem unmounted and mounted as read-only
341 * - system power-cycle and filesystem mounted as read-only
342 * - filesystem or device errors leading to forced read-only
344 * The status of exclusive operation is set and cleared atomically.
345 * During the course of Paused state, fs_info::exclusive_operation remains set.
346 * A device operation in Paused or Running state can be canceled or resumed
347 * either by ioctl (Balance only) or when remounted as read-write.
348 * The exclusive status is cleared when the device operation is canceled or
352 DEFINE_MUTEX(uuid_mutex);
353 static LIST_HEAD(fs_uuids);
354 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void)
360 * alloc_fs_devices - allocate struct btrfs_fs_devices
361 * @fsid: if not NULL, copy the UUID to fs_devices::fsid
362 * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid
364 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
365 * The returned struct is not linked onto any lists and can be destroyed with
366 * kfree() right away.
368 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid,
369 const u8 *metadata_fsid)
371 struct btrfs_fs_devices *fs_devs;
373 ASSERT(fsid || !metadata_fsid);
375 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
377 return ERR_PTR(-ENOMEM);
379 mutex_init(&fs_devs->device_list_mutex);
381 INIT_LIST_HEAD(&fs_devs->devices);
382 INIT_LIST_HEAD(&fs_devs->alloc_list);
383 INIT_LIST_HEAD(&fs_devs->fs_list);
384 INIT_LIST_HEAD(&fs_devs->seed_list);
387 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
388 memcpy(fs_devs->metadata_uuid,
389 metadata_fsid ?: fsid, BTRFS_FSID_SIZE);
395 static void btrfs_free_device(struct btrfs_device *device)
397 WARN_ON(!list_empty(&device->post_commit_list));
398 rcu_string_free(device->name);
399 extent_io_tree_release(&device->alloc_state);
400 btrfs_destroy_dev_zone_info(device);
404 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
406 struct btrfs_device *device;
408 WARN_ON(fs_devices->opened);
409 while (!list_empty(&fs_devices->devices)) {
410 device = list_entry(fs_devices->devices.next,
411 struct btrfs_device, dev_list);
412 list_del(&device->dev_list);
413 btrfs_free_device(device);
418 void __exit btrfs_cleanup_fs_uuids(void)
420 struct btrfs_fs_devices *fs_devices;
422 while (!list_empty(&fs_uuids)) {
423 fs_devices = list_entry(fs_uuids.next,
424 struct btrfs_fs_devices, fs_list);
425 list_del(&fs_devices->fs_list);
426 free_fs_devices(fs_devices);
430 static bool match_fsid_fs_devices(const struct btrfs_fs_devices *fs_devices,
431 const u8 *fsid, const u8 *metadata_fsid)
433 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) != 0)
439 if (memcmp(metadata_fsid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE) != 0)
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) {
454 if (match_fsid_fs_devices(fs_devices, fsid, metadata_fsid))
461 * First check if the metadata_uuid is different from the fsid in the given
462 * fs_devices. Then check if the given fsid is the same as the metadata_uuid
463 * in the fs_devices. If it is, return true; otherwise, return false.
465 static inline bool check_fsid_changed(const struct btrfs_fs_devices *fs_devices,
468 return memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
469 BTRFS_FSID_SIZE) != 0 &&
470 memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE) == 0;
473 static struct btrfs_fs_devices *find_fsid_with_metadata_uuid(
474 struct btrfs_super_block *disk_super)
477 struct btrfs_fs_devices *fs_devices;
480 * Handle scanned device having completed its fsid change but
481 * belonging to a fs_devices that was created by first scanning
482 * a device which didn't have its fsid/metadata_uuid changed
483 * at all and the CHANGING_FSID_V2 flag set.
485 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
486 if (!fs_devices->fsid_change)
489 if (match_fsid_fs_devices(fs_devices, disk_super->metadata_uuid,
495 * Handle scanned device having completed its fsid change but
496 * belonging to a fs_devices that was created by a device that
497 * has an outdated pair of fsid/metadata_uuid and
498 * CHANGING_FSID_V2 flag set.
500 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
501 if (!fs_devices->fsid_change)
504 if (check_fsid_changed(fs_devices, disk_super->metadata_uuid))
508 return find_fsid(disk_super->fsid, disk_super->metadata_uuid);
513 btrfs_get_bdev_and_sb(const char *device_path, blk_mode_t flags, void *holder,
514 int flush, struct block_device **bdev,
515 struct btrfs_super_block **disk_super)
519 *bdev = blkdev_get_by_path(device_path, flags, holder, NULL);
522 ret = PTR_ERR(*bdev);
527 sync_blockdev(*bdev);
528 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
530 blkdev_put(*bdev, holder);
533 invalidate_bdev(*bdev);
534 *disk_super = btrfs_read_dev_super(*bdev);
535 if (IS_ERR(*disk_super)) {
536 ret = PTR_ERR(*disk_super);
537 blkdev_put(*bdev, holder);
549 * Search and remove all stale devices (which are not mounted). When both
550 * inputs are NULL, it will search and release all stale devices.
552 * @devt: Optional. When provided will it release all unmounted devices
553 * matching this devt only.
554 * @skip_device: Optional. Will skip this device when searching for the stale
557 * Return: 0 for success or if @devt is 0.
558 * -EBUSY if @devt is a mounted device.
559 * -ENOENT if @devt does not match any device in the list.
561 static int btrfs_free_stale_devices(dev_t devt, struct btrfs_device *skip_device)
563 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
564 struct btrfs_device *device, *tmp_device;
567 lockdep_assert_held(&uuid_mutex);
572 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
574 mutex_lock(&fs_devices->device_list_mutex);
575 list_for_each_entry_safe(device, tmp_device,
576 &fs_devices->devices, dev_list) {
577 if (skip_device && skip_device == device)
579 if (devt && devt != device->devt)
581 if (fs_devices->opened) {
582 /* for an already deleted device return 0 */
583 if (devt && ret != 0)
588 /* delete the stale device */
589 fs_devices->num_devices--;
590 list_del(&device->dev_list);
591 btrfs_free_device(device);
595 mutex_unlock(&fs_devices->device_list_mutex);
597 if (fs_devices->num_devices == 0) {
598 btrfs_sysfs_remove_fsid(fs_devices);
599 list_del(&fs_devices->fs_list);
600 free_fs_devices(fs_devices);
608 * This is only used on mount, and we are protected from competing things
609 * messing with our fs_devices by the uuid_mutex, thus we do not need the
610 * fs_devices->device_list_mutex here.
612 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
613 struct btrfs_device *device, blk_mode_t flags,
616 struct block_device *bdev;
617 struct btrfs_super_block *disk_super;
626 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
631 devid = btrfs_stack_device_id(&disk_super->dev_item);
632 if (devid != device->devid)
633 goto error_free_page;
635 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
636 goto error_free_page;
638 device->generation = btrfs_super_generation(disk_super);
640 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
641 if (btrfs_super_incompat_flags(disk_super) &
642 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
644 "BTRFS: Invalid seeding and uuid-changed device detected\n");
645 goto error_free_page;
648 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
649 fs_devices->seeding = true;
651 if (bdev_read_only(bdev))
652 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
654 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
657 if (!bdev_nonrot(bdev))
658 fs_devices->rotating = true;
660 if (bdev_max_discard_sectors(bdev))
661 fs_devices->discardable = true;
664 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
665 device->holder = holder;
667 fs_devices->open_devices++;
668 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
669 device->devid != BTRFS_DEV_REPLACE_DEVID) {
670 fs_devices->rw_devices++;
671 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
673 btrfs_release_disk_super(disk_super);
678 btrfs_release_disk_super(disk_super);
679 blkdev_put(bdev, holder);
684 u8 *btrfs_sb_fsid_ptr(struct btrfs_super_block *sb)
686 bool has_metadata_uuid = (btrfs_super_incompat_flags(sb) &
687 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
689 return has_metadata_uuid ? sb->metadata_uuid : sb->fsid;
693 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices
694 * being created with a disk that has already completed its fsid change. Such
695 * disk can belong to an fs which has its FSID changed or to one which doesn't.
696 * Handle both cases here.
698 static struct btrfs_fs_devices *find_fsid_inprogress(
699 struct btrfs_super_block *disk_super)
701 struct btrfs_fs_devices *fs_devices;
703 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
704 if (fs_devices->fsid_change)
707 if (check_fsid_changed(fs_devices, disk_super->fsid))
711 return find_fsid(disk_super->fsid, NULL);
714 static struct btrfs_fs_devices *find_fsid_changed(
715 struct btrfs_super_block *disk_super)
717 struct btrfs_fs_devices *fs_devices;
720 * Handles the case where scanned device is part of an fs that had
721 * multiple successful changes of FSID but currently device didn't
722 * observe it. Meaning our fsid will be different than theirs. We need
723 * to handle two subcases :
724 * 1 - The fs still continues to have different METADATA/FSID uuids.
725 * 2 - The fs is switched back to its original FSID (METADATA/FSID
728 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
730 if (check_fsid_changed(fs_devices, disk_super->metadata_uuid) &&
731 memcmp(fs_devices->fsid, disk_super->fsid,
732 BTRFS_FSID_SIZE) != 0)
735 /* Unchanged UUIDs */
736 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
737 BTRFS_FSID_SIZE) == 0 &&
738 memcmp(fs_devices->fsid, disk_super->metadata_uuid,
739 BTRFS_FSID_SIZE) == 0)
746 static struct btrfs_fs_devices *find_fsid_reverted_metadata(
747 struct btrfs_super_block *disk_super)
749 struct btrfs_fs_devices *fs_devices;
752 * Handle the case where the scanned device is part of an fs whose last
753 * metadata UUID change reverted it to the original FSID. At the same
754 * time fs_devices was first created by another constituent device
755 * which didn't fully observe the operation. This results in an
756 * btrfs_fs_devices created with metadata/fsid different AND
757 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the
758 * fs_devices equal to the FSID of the disk.
760 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
761 if (!fs_devices->fsid_change)
764 if (check_fsid_changed(fs_devices, disk_super->fsid))
771 * Add new device to list of registered devices
774 * device pointer which was just added or updated when successful
775 * error pointer when failed
777 static noinline struct btrfs_device *device_list_add(const char *path,
778 struct btrfs_super_block *disk_super,
779 bool *new_device_added)
781 struct btrfs_device *device;
782 struct btrfs_fs_devices *fs_devices = NULL;
783 struct rcu_string *name;
784 u64 found_transid = btrfs_super_generation(disk_super);
785 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
788 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
789 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
790 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) &
791 BTRFS_SUPER_FLAG_CHANGING_FSID_V2);
793 error = lookup_bdev(path, &path_devt);
795 btrfs_err(NULL, "failed to lookup block device for path %s: %d",
797 return ERR_PTR(error);
800 if (fsid_change_in_progress) {
801 if (!has_metadata_uuid)
802 fs_devices = find_fsid_inprogress(disk_super);
804 fs_devices = find_fsid_changed(disk_super);
805 } else if (has_metadata_uuid) {
806 fs_devices = find_fsid_with_metadata_uuid(disk_super);
808 fs_devices = find_fsid_reverted_metadata(disk_super);
810 fs_devices = find_fsid(disk_super->fsid, NULL);
815 fs_devices = alloc_fs_devices(disk_super->fsid,
816 has_metadata_uuid ? disk_super->metadata_uuid : NULL);
817 if (IS_ERR(fs_devices))
818 return ERR_CAST(fs_devices);
820 fs_devices->fsid_change = fsid_change_in_progress;
822 mutex_lock(&fs_devices->device_list_mutex);
823 list_add(&fs_devices->fs_list, &fs_uuids);
827 struct btrfs_dev_lookup_args args = {
829 .uuid = disk_super->dev_item.uuid,
832 mutex_lock(&fs_devices->device_list_mutex);
833 device = btrfs_find_device(fs_devices, &args);
836 * If this disk has been pulled into an fs devices created by
837 * a device which had the CHANGING_FSID_V2 flag then replace the
838 * metadata_uuid/fsid values of the fs_devices.
840 if (fs_devices->fsid_change &&
841 found_transid > fs_devices->latest_generation) {
842 memcpy(fs_devices->fsid, disk_super->fsid,
844 memcpy(fs_devices->metadata_uuid,
845 btrfs_sb_fsid_ptr(disk_super), BTRFS_FSID_SIZE);
846 fs_devices->fsid_change = false;
851 unsigned int nofs_flag;
853 if (fs_devices->opened) {
855 "device %s belongs to fsid %pU, and the fs is already mounted, scanned by %s (%d)",
856 path, fs_devices->fsid, current->comm,
857 task_pid_nr(current));
858 mutex_unlock(&fs_devices->device_list_mutex);
859 return ERR_PTR(-EBUSY);
862 nofs_flag = memalloc_nofs_save();
863 device = btrfs_alloc_device(NULL, &devid,
864 disk_super->dev_item.uuid, path);
865 memalloc_nofs_restore(nofs_flag);
866 if (IS_ERR(device)) {
867 mutex_unlock(&fs_devices->device_list_mutex);
868 /* we can safely leave the fs_devices entry around */
872 device->devt = path_devt;
874 list_add_rcu(&device->dev_list, &fs_devices->devices);
875 fs_devices->num_devices++;
877 device->fs_devices = fs_devices;
878 *new_device_added = true;
880 if (disk_super->label[0])
882 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n",
883 disk_super->label, devid, found_transid, path,
884 current->comm, task_pid_nr(current));
887 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n",
888 disk_super->fsid, devid, found_transid, path,
889 current->comm, task_pid_nr(current));
891 } else if (!device->name || strcmp(device->name->str, path)) {
893 * When FS is already mounted.
894 * 1. If you are here and if the device->name is NULL that
895 * means this device was missing at time of FS mount.
896 * 2. If you are here and if the device->name is different
897 * from 'path' that means either
898 * a. The same device disappeared and reappeared with
900 * b. The missing-disk-which-was-replaced, has
903 * We must allow 1 and 2a above. But 2b would be a spurious
906 * Further in case of 1 and 2a above, the disk at 'path'
907 * would have missed some transaction when it was away and
908 * in case of 2a the stale bdev has to be updated as well.
909 * 2b must not be allowed at all time.
913 * For now, we do allow update to btrfs_fs_device through the
914 * btrfs dev scan cli after FS has been mounted. We're still
915 * tracking a problem where systems fail mount by subvolume id
916 * when we reject replacement on a mounted FS.
918 if (!fs_devices->opened && found_transid < device->generation) {
920 * That is if the FS is _not_ mounted and if you
921 * are here, that means there is more than one
922 * disk with same uuid and devid.We keep the one
923 * with larger generation number or the last-in if
924 * generation are equal.
926 mutex_unlock(&fs_devices->device_list_mutex);
928 "device %s already registered with a higher generation, found %llu expect %llu",
929 path, found_transid, device->generation);
930 return ERR_PTR(-EEXIST);
934 * We are going to replace the device path for a given devid,
935 * make sure it's the same device if the device is mounted
937 * NOTE: the device->fs_info may not be reliable here so pass
938 * in a NULL to message helpers instead. This avoids a possible
939 * use-after-free when the fs_info and fs_info->sb are already
943 if (device->devt != path_devt) {
944 mutex_unlock(&fs_devices->device_list_mutex);
945 btrfs_warn_in_rcu(NULL,
946 "duplicate device %s devid %llu generation %llu scanned by %s (%d)",
947 path, devid, found_transid,
949 task_pid_nr(current));
950 return ERR_PTR(-EEXIST);
952 btrfs_info_in_rcu(NULL,
953 "devid %llu device path %s changed to %s scanned by %s (%d)",
954 devid, btrfs_dev_name(device),
956 task_pid_nr(current));
959 name = rcu_string_strdup(path, GFP_NOFS);
961 mutex_unlock(&fs_devices->device_list_mutex);
962 return ERR_PTR(-ENOMEM);
964 rcu_string_free(device->name);
965 rcu_assign_pointer(device->name, name);
966 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
967 fs_devices->missing_devices--;
968 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
970 device->devt = path_devt;
974 * Unmount does not free the btrfs_device struct but would zero
975 * generation along with most of the other members. So just update
976 * it back. We need it to pick the disk with largest generation
979 if (!fs_devices->opened) {
980 device->generation = found_transid;
981 fs_devices->latest_generation = max_t(u64, found_transid,
982 fs_devices->latest_generation);
985 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
987 mutex_unlock(&fs_devices->device_list_mutex);
991 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
993 struct btrfs_fs_devices *fs_devices;
994 struct btrfs_device *device;
995 struct btrfs_device *orig_dev;
998 lockdep_assert_held(&uuid_mutex);
1000 fs_devices = alloc_fs_devices(orig->fsid, NULL);
1001 if (IS_ERR(fs_devices))
1004 fs_devices->total_devices = orig->total_devices;
1006 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
1007 const char *dev_path = NULL;
1010 * This is ok to do without RCU read locked because we hold the
1011 * uuid mutex so nothing we touch in here is going to disappear.
1014 dev_path = orig_dev->name->str;
1016 device = btrfs_alloc_device(NULL, &orig_dev->devid,
1017 orig_dev->uuid, dev_path);
1018 if (IS_ERR(device)) {
1019 ret = PTR_ERR(device);
1023 if (orig_dev->zone_info) {
1024 struct btrfs_zoned_device_info *zone_info;
1026 zone_info = btrfs_clone_dev_zone_info(orig_dev);
1028 btrfs_free_device(device);
1032 device->zone_info = zone_info;
1035 list_add(&device->dev_list, &fs_devices->devices);
1036 device->fs_devices = fs_devices;
1037 fs_devices->num_devices++;
1041 free_fs_devices(fs_devices);
1042 return ERR_PTR(ret);
1045 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices,
1046 struct btrfs_device **latest_dev)
1048 struct btrfs_device *device, *next;
1050 /* This is the initialized path, it is safe to release the devices. */
1051 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1052 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) {
1053 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1054 &device->dev_state) &&
1055 !test_bit(BTRFS_DEV_STATE_MISSING,
1056 &device->dev_state) &&
1058 device->generation > (*latest_dev)->generation)) {
1059 *latest_dev = device;
1065 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID,
1066 * in btrfs_init_dev_replace() so just continue.
1068 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1072 blkdev_put(device->bdev, device->holder);
1073 device->bdev = NULL;
1074 fs_devices->open_devices--;
1076 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1077 list_del_init(&device->dev_alloc_list);
1078 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1079 fs_devices->rw_devices--;
1081 list_del_init(&device->dev_list);
1082 fs_devices->num_devices--;
1083 btrfs_free_device(device);
1089 * After we have read the system tree and know devids belonging to this
1090 * filesystem, remove the device which does not belong there.
1092 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices)
1094 struct btrfs_device *latest_dev = NULL;
1095 struct btrfs_fs_devices *seed_dev;
1097 mutex_lock(&uuid_mutex);
1098 __btrfs_free_extra_devids(fs_devices, &latest_dev);
1100 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list)
1101 __btrfs_free_extra_devids(seed_dev, &latest_dev);
1103 fs_devices->latest_dev = latest_dev;
1105 mutex_unlock(&uuid_mutex);
1108 static void btrfs_close_bdev(struct btrfs_device *device)
1113 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1114 sync_blockdev(device->bdev);
1115 invalidate_bdev(device->bdev);
1118 blkdev_put(device->bdev, device->holder);
1121 static void btrfs_close_one_device(struct btrfs_device *device)
1123 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1125 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1126 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1127 list_del_init(&device->dev_alloc_list);
1128 fs_devices->rw_devices--;
1131 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1132 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
1134 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
1135 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
1136 fs_devices->missing_devices--;
1139 btrfs_close_bdev(device);
1141 fs_devices->open_devices--;
1142 device->bdev = NULL;
1144 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1145 btrfs_destroy_dev_zone_info(device);
1147 device->fs_info = NULL;
1148 atomic_set(&device->dev_stats_ccnt, 0);
1149 extent_io_tree_release(&device->alloc_state);
1152 * Reset the flush error record. We might have a transient flush error
1153 * in this mount, and if so we aborted the current transaction and set
1154 * the fs to an error state, guaranteeing no super blocks can be further
1155 * committed. However that error might be transient and if we unmount the
1156 * filesystem and mount it again, we should allow the mount to succeed
1157 * (btrfs_check_rw_degradable() should not fail) - if after mounting the
1158 * filesystem again we still get flush errors, then we will again abort
1159 * any transaction and set the error state, guaranteeing no commits of
1160 * unsafe super blocks.
1162 device->last_flush_error = 0;
1164 /* Verify the device is back in a pristine state */
1165 WARN_ON(test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1166 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1167 WARN_ON(!list_empty(&device->dev_alloc_list));
1168 WARN_ON(!list_empty(&device->post_commit_list));
1171 static void close_fs_devices(struct btrfs_fs_devices *fs_devices)
1173 struct btrfs_device *device, *tmp;
1175 lockdep_assert_held(&uuid_mutex);
1177 if (--fs_devices->opened > 0)
1180 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list)
1181 btrfs_close_one_device(device);
1183 WARN_ON(fs_devices->open_devices);
1184 WARN_ON(fs_devices->rw_devices);
1185 fs_devices->opened = 0;
1186 fs_devices->seeding = false;
1187 fs_devices->fs_info = NULL;
1190 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1193 struct btrfs_fs_devices *tmp;
1195 mutex_lock(&uuid_mutex);
1196 close_fs_devices(fs_devices);
1197 if (!fs_devices->opened) {
1198 list_splice_init(&fs_devices->seed_list, &list);
1201 * If the struct btrfs_fs_devices is not assembled with any
1202 * other device, it can be re-initialized during the next mount
1203 * without the needing device-scan step. Therefore, it can be
1206 if (fs_devices->num_devices == 1) {
1207 list_del(&fs_devices->fs_list);
1208 free_fs_devices(fs_devices);
1213 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {
1214 close_fs_devices(fs_devices);
1215 list_del(&fs_devices->seed_list);
1216 free_fs_devices(fs_devices);
1218 mutex_unlock(&uuid_mutex);
1221 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1222 blk_mode_t flags, void *holder)
1224 struct btrfs_device *device;
1225 struct btrfs_device *latest_dev = NULL;
1226 struct btrfs_device *tmp_device;
1228 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,
1232 ret = btrfs_open_one_device(fs_devices, device, flags, holder);
1234 (!latest_dev || device->generation > latest_dev->generation)) {
1235 latest_dev = device;
1236 } else if (ret == -ENODATA) {
1237 fs_devices->num_devices--;
1238 list_del(&device->dev_list);
1239 btrfs_free_device(device);
1242 if (fs_devices->open_devices == 0)
1245 fs_devices->opened = 1;
1246 fs_devices->latest_dev = latest_dev;
1247 fs_devices->total_rw_bytes = 0;
1248 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1249 fs_devices->read_policy = BTRFS_READ_POLICY_PID;
1254 static int devid_cmp(void *priv, const struct list_head *a,
1255 const struct list_head *b)
1257 const struct btrfs_device *dev1, *dev2;
1259 dev1 = list_entry(a, struct btrfs_device, dev_list);
1260 dev2 = list_entry(b, struct btrfs_device, dev_list);
1262 if (dev1->devid < dev2->devid)
1264 else if (dev1->devid > dev2->devid)
1269 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1270 blk_mode_t flags, void *holder)
1274 lockdep_assert_held(&uuid_mutex);
1276 * The device_list_mutex cannot be taken here in case opening the
1277 * underlying device takes further locks like open_mutex.
1279 * We also don't need the lock here as this is called during mount and
1280 * exclusion is provided by uuid_mutex
1283 if (fs_devices->opened) {
1284 fs_devices->opened++;
1287 list_sort(NULL, &fs_devices->devices, devid_cmp);
1288 ret = open_fs_devices(fs_devices, flags, holder);
1294 void btrfs_release_disk_super(struct btrfs_super_block *super)
1296 struct page *page = virt_to_page(super);
1301 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1302 u64 bytenr, u64 bytenr_orig)
1304 struct btrfs_super_block *disk_super;
1309 /* make sure our super fits in the device */
1310 if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev))
1311 return ERR_PTR(-EINVAL);
1313 /* make sure our super fits in the page */
1314 if (sizeof(*disk_super) > PAGE_SIZE)
1315 return ERR_PTR(-EINVAL);
1317 /* make sure our super doesn't straddle pages on disk */
1318 index = bytenr >> PAGE_SHIFT;
1319 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1320 return ERR_PTR(-EINVAL);
1322 /* pull in the page with our super */
1323 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1326 return ERR_CAST(page);
1328 p = page_address(page);
1330 /* align our pointer to the offset of the super block */
1331 disk_super = p + offset_in_page(bytenr);
1333 if (btrfs_super_bytenr(disk_super) != bytenr_orig ||
1334 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1335 btrfs_release_disk_super(p);
1336 return ERR_PTR(-EINVAL);
1339 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1340 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1345 int btrfs_forget_devices(dev_t devt)
1349 mutex_lock(&uuid_mutex);
1350 ret = btrfs_free_stale_devices(devt, NULL);
1351 mutex_unlock(&uuid_mutex);
1357 * Look for a btrfs signature on a device. This may be called out of the mount path
1358 * and we are not allowed to call set_blocksize during the scan. The superblock
1359 * is read via pagecache
1361 struct btrfs_device *btrfs_scan_one_device(const char *path, blk_mode_t flags)
1363 struct btrfs_super_block *disk_super;
1364 bool new_device_added = false;
1365 struct btrfs_device *device = NULL;
1366 struct block_device *bdev;
1367 u64 bytenr, bytenr_orig;
1370 lockdep_assert_held(&uuid_mutex);
1373 * we would like to check all the supers, but that would make
1374 * a btrfs mount succeed after a mkfs from a different FS.
1375 * So, we need to add a special mount option to scan for
1376 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1380 * Avoid an exclusive open here, as the systemd-udev may initiate the
1381 * device scan which may race with the user's mount or mkfs command,
1382 * resulting in failure.
1383 * Since the device scan is solely for reading purposes, there is no
1384 * need for an exclusive open. Additionally, the devices are read again
1385 * during the mount process. It is ok to get some inconsistent
1386 * values temporarily, as the device paths of the fsid are the only
1387 * required information for assembling the volume.
1389 bdev = blkdev_get_by_path(path, flags, NULL, NULL);
1391 return ERR_CAST(bdev);
1393 bytenr_orig = btrfs_sb_offset(0);
1394 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr);
1396 device = ERR_PTR(ret);
1397 goto error_bdev_put;
1400 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig);
1401 if (IS_ERR(disk_super)) {
1402 device = ERR_CAST(disk_super);
1403 goto error_bdev_put;
1406 device = device_list_add(path, disk_super, &new_device_added);
1407 if (!IS_ERR(device) && new_device_added)
1408 btrfs_free_stale_devices(device->devt, device);
1410 btrfs_release_disk_super(disk_super);
1413 blkdev_put(bdev, NULL);
1419 * Try to find a chunk that intersects [start, start + len] range and when one
1420 * such is found, record the end of it in *start
1422 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1425 u64 physical_start, physical_end;
1427 lockdep_assert_held(&device->fs_info->chunk_mutex);
1429 if (find_first_extent_bit(&device->alloc_state, *start,
1430 &physical_start, &physical_end,
1431 CHUNK_ALLOCATED, NULL)) {
1433 if (in_range(physical_start, *start, len) ||
1434 in_range(*start, physical_start,
1435 physical_end - physical_start)) {
1436 *start = physical_end + 1;
1443 static u64 dev_extent_search_start(struct btrfs_device *device)
1445 switch (device->fs_devices->chunk_alloc_policy) {
1446 case BTRFS_CHUNK_ALLOC_REGULAR:
1447 return BTRFS_DEVICE_RANGE_RESERVED;
1448 case BTRFS_CHUNK_ALLOC_ZONED:
1450 * We don't care about the starting region like regular
1451 * allocator, because we anyway use/reserve the first two zones
1452 * for superblock logging.
1460 static bool dev_extent_hole_check_zoned(struct btrfs_device *device,
1461 u64 *hole_start, u64 *hole_size,
1464 u64 zone_size = device->zone_info->zone_size;
1467 bool changed = false;
1469 ASSERT(IS_ALIGNED(*hole_start, zone_size));
1471 while (*hole_size > 0) {
1472 pos = btrfs_find_allocatable_zones(device, *hole_start,
1473 *hole_start + *hole_size,
1475 if (pos != *hole_start) {
1476 *hole_size = *hole_start + *hole_size - pos;
1479 if (*hole_size < num_bytes)
1483 ret = btrfs_ensure_empty_zones(device, pos, num_bytes);
1485 /* Range is ensured to be empty */
1489 /* Given hole range was invalid (outside of device) */
1490 if (ret == -ERANGE) {
1491 *hole_start += *hole_size;
1496 *hole_start += zone_size;
1497 *hole_size -= zone_size;
1505 * Check if specified hole is suitable for allocation.
1507 * @device: the device which we have the hole
1508 * @hole_start: starting position of the hole
1509 * @hole_size: the size of the hole
1510 * @num_bytes: the size of the free space that we need
1512 * This function may modify @hole_start and @hole_size to reflect the suitable
1513 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1515 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1516 u64 *hole_size, u64 num_bytes)
1518 bool changed = false;
1519 u64 hole_end = *hole_start + *hole_size;
1523 * Check before we set max_hole_start, otherwise we could end up
1524 * sending back this offset anyway.
1526 if (contains_pending_extent(device, hole_start, *hole_size)) {
1527 if (hole_end >= *hole_start)
1528 *hole_size = hole_end - *hole_start;
1534 switch (device->fs_devices->chunk_alloc_policy) {
1535 case BTRFS_CHUNK_ALLOC_REGULAR:
1536 /* No extra check */
1538 case BTRFS_CHUNK_ALLOC_ZONED:
1539 if (dev_extent_hole_check_zoned(device, hole_start,
1540 hole_size, num_bytes)) {
1543 * The changed hole can contain pending extent.
1544 * Loop again to check that.
1560 * Find free space in the specified device.
1562 * @device: the device which we search the free space in
1563 * @num_bytes: the size of the free space that we need
1564 * @search_start: the position from which to begin the search
1565 * @start: store the start of the free space.
1566 * @len: the size of the free space. that we find, or the size
1567 * of the max free space if we don't find suitable free space
1569 * This does a pretty simple search, the expectation is that it is called very
1570 * infrequently and that a given device has a small number of extents.
1572 * @start is used to store the start of the free space if we find. But if we
1573 * don't find suitable free space, it will be used to store the start position
1574 * of the max free space.
1576 * @len is used to store the size of the free space that we find.
1577 * But if we don't find suitable free space, it is used to store the size of
1578 * the max free space.
1580 * NOTE: This function will search *commit* root of device tree, and does extra
1581 * check to ensure dev extents are not double allocated.
1582 * This makes the function safe to allocate dev extents but may not report
1583 * correct usable device space, as device extent freed in current transaction
1584 * is not reported as available.
1586 static int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1587 u64 *start, u64 *len)
1589 struct btrfs_fs_info *fs_info = device->fs_info;
1590 struct btrfs_root *root = fs_info->dev_root;
1591 struct btrfs_key key;
1592 struct btrfs_dev_extent *dev_extent;
1593 struct btrfs_path *path;
1597 u64 max_hole_size = 0;
1599 u64 search_end = device->total_bytes;
1602 struct extent_buffer *l;
1604 search_start = dev_extent_search_start(device);
1605 max_hole_start = search_start;
1607 WARN_ON(device->zone_info &&
1608 !IS_ALIGNED(num_bytes, device->zone_info->zone_size));
1610 path = btrfs_alloc_path();
1616 if (search_start >= search_end ||
1617 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1622 path->reada = READA_FORWARD;
1623 path->search_commit_root = 1;
1624 path->skip_locking = 1;
1626 key.objectid = device->devid;
1627 key.offset = search_start;
1628 key.type = BTRFS_DEV_EXTENT_KEY;
1630 ret = btrfs_search_backwards(root, &key, path);
1634 while (search_start < search_end) {
1636 slot = path->slots[0];
1637 if (slot >= btrfs_header_nritems(l)) {
1638 ret = btrfs_next_leaf(root, path);
1646 btrfs_item_key_to_cpu(l, &key, slot);
1648 if (key.objectid < device->devid)
1651 if (key.objectid > device->devid)
1654 if (key.type != BTRFS_DEV_EXTENT_KEY)
1657 if (key.offset > search_end)
1660 if (key.offset > search_start) {
1661 hole_size = key.offset - search_start;
1662 dev_extent_hole_check(device, &search_start, &hole_size,
1665 if (hole_size > max_hole_size) {
1666 max_hole_start = search_start;
1667 max_hole_size = hole_size;
1671 * If this free space is greater than which we need,
1672 * it must be the max free space that we have found
1673 * until now, so max_hole_start must point to the start
1674 * of this free space and the length of this free space
1675 * is stored in max_hole_size. Thus, we return
1676 * max_hole_start and max_hole_size and go back to the
1679 if (hole_size >= num_bytes) {
1685 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1686 extent_end = key.offset + btrfs_dev_extent_length(l,
1688 if (extent_end > search_start)
1689 search_start = extent_end;
1696 * At this point, search_start should be the end of
1697 * allocated dev extents, and when shrinking the device,
1698 * search_end may be smaller than search_start.
1700 if (search_end > search_start) {
1701 hole_size = search_end - search_start;
1702 if (dev_extent_hole_check(device, &search_start, &hole_size,
1704 btrfs_release_path(path);
1708 if (hole_size > max_hole_size) {
1709 max_hole_start = search_start;
1710 max_hole_size = hole_size;
1715 if (max_hole_size < num_bytes)
1720 ASSERT(max_hole_start + max_hole_size <= search_end);
1722 btrfs_free_path(path);
1723 *start = max_hole_start;
1725 *len = max_hole_size;
1729 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1730 struct btrfs_device *device,
1731 u64 start, u64 *dev_extent_len)
1733 struct btrfs_fs_info *fs_info = device->fs_info;
1734 struct btrfs_root *root = fs_info->dev_root;
1736 struct btrfs_path *path;
1737 struct btrfs_key key;
1738 struct btrfs_key found_key;
1739 struct extent_buffer *leaf = NULL;
1740 struct btrfs_dev_extent *extent = NULL;
1742 path = btrfs_alloc_path();
1746 key.objectid = device->devid;
1748 key.type = BTRFS_DEV_EXTENT_KEY;
1750 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1752 ret = btrfs_previous_item(root, path, key.objectid,
1753 BTRFS_DEV_EXTENT_KEY);
1756 leaf = path->nodes[0];
1757 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1758 extent = btrfs_item_ptr(leaf, path->slots[0],
1759 struct btrfs_dev_extent);
1760 BUG_ON(found_key.offset > start || found_key.offset +
1761 btrfs_dev_extent_length(leaf, extent) < start);
1763 btrfs_release_path(path);
1765 } else if (ret == 0) {
1766 leaf = path->nodes[0];
1767 extent = btrfs_item_ptr(leaf, path->slots[0],
1768 struct btrfs_dev_extent);
1773 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1775 ret = btrfs_del_item(trans, root, path);
1777 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1779 btrfs_free_path(path);
1783 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1785 struct extent_map_tree *em_tree;
1786 struct extent_map *em;
1790 em_tree = &fs_info->mapping_tree;
1791 read_lock(&em_tree->lock);
1792 n = rb_last(&em_tree->map.rb_root);
1794 em = rb_entry(n, struct extent_map, rb_node);
1795 ret = em->start + em->len;
1797 read_unlock(&em_tree->lock);
1802 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1806 struct btrfs_key key;
1807 struct btrfs_key found_key;
1808 struct btrfs_path *path;
1810 path = btrfs_alloc_path();
1814 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1815 key.type = BTRFS_DEV_ITEM_KEY;
1816 key.offset = (u64)-1;
1818 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1824 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1829 ret = btrfs_previous_item(fs_info->chunk_root, path,
1830 BTRFS_DEV_ITEMS_OBJECTID,
1831 BTRFS_DEV_ITEM_KEY);
1835 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1837 *devid_ret = found_key.offset + 1;
1841 btrfs_free_path(path);
1846 * the device information is stored in the chunk root
1847 * the btrfs_device struct should be fully filled in
1849 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1850 struct btrfs_device *device)
1853 struct btrfs_path *path;
1854 struct btrfs_dev_item *dev_item;
1855 struct extent_buffer *leaf;
1856 struct btrfs_key key;
1859 path = btrfs_alloc_path();
1863 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1864 key.type = BTRFS_DEV_ITEM_KEY;
1865 key.offset = device->devid;
1867 btrfs_reserve_chunk_metadata(trans, true);
1868 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1869 &key, sizeof(*dev_item));
1870 btrfs_trans_release_chunk_metadata(trans);
1874 leaf = path->nodes[0];
1875 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1877 btrfs_set_device_id(leaf, dev_item, device->devid);
1878 btrfs_set_device_generation(leaf, dev_item, 0);
1879 btrfs_set_device_type(leaf, dev_item, device->type);
1880 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1881 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1882 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1883 btrfs_set_device_total_bytes(leaf, dev_item,
1884 btrfs_device_get_disk_total_bytes(device));
1885 btrfs_set_device_bytes_used(leaf, dev_item,
1886 btrfs_device_get_bytes_used(device));
1887 btrfs_set_device_group(leaf, dev_item, 0);
1888 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1889 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1890 btrfs_set_device_start_offset(leaf, dev_item, 0);
1892 ptr = btrfs_device_uuid(dev_item);
1893 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1894 ptr = btrfs_device_fsid(dev_item);
1895 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1896 ptr, BTRFS_FSID_SIZE);
1897 btrfs_mark_buffer_dirty(leaf);
1901 btrfs_free_path(path);
1906 * Function to update ctime/mtime for a given device path.
1907 * Mainly used for ctime/mtime based probe like libblkid.
1909 * We don't care about errors here, this is just to be kind to userspace.
1911 static void update_dev_time(const char *device_path)
1914 struct timespec64 now;
1917 ret = kern_path(device_path, LOOKUP_FOLLOW, &path);
1921 now = current_time(d_inode(path.dentry));
1922 inode_update_time(d_inode(path.dentry), &now, S_MTIME | S_CTIME | S_VERSION);
1926 static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans,
1927 struct btrfs_device *device)
1929 struct btrfs_root *root = device->fs_info->chunk_root;
1931 struct btrfs_path *path;
1932 struct btrfs_key key;
1934 path = btrfs_alloc_path();
1938 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1939 key.type = BTRFS_DEV_ITEM_KEY;
1940 key.offset = device->devid;
1942 btrfs_reserve_chunk_metadata(trans, false);
1943 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1944 btrfs_trans_release_chunk_metadata(trans);
1951 ret = btrfs_del_item(trans, root, path);
1953 btrfs_free_path(path);
1958 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1959 * filesystem. It's up to the caller to adjust that number regarding eg. device
1962 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1970 seq = read_seqbegin(&fs_info->profiles_lock);
1972 all_avail = fs_info->avail_data_alloc_bits |
1973 fs_info->avail_system_alloc_bits |
1974 fs_info->avail_metadata_alloc_bits;
1975 } while (read_seqretry(&fs_info->profiles_lock, seq));
1977 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1978 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1981 if (num_devices < btrfs_raid_array[i].devs_min)
1982 return btrfs_raid_array[i].mindev_error;
1988 static struct btrfs_device * btrfs_find_next_active_device(
1989 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1991 struct btrfs_device *next_device;
1993 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1994 if (next_device != device &&
1995 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1996 && next_device->bdev)
2004 * Helper function to check if the given device is part of s_bdev / latest_dev
2005 * and replace it with the provided or the next active device, in the context
2006 * where this function called, there should be always be another device (or
2007 * this_dev) which is active.
2009 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
2010 struct btrfs_device *next_device)
2012 struct btrfs_fs_info *fs_info = device->fs_info;
2015 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
2017 ASSERT(next_device);
2019 if (fs_info->sb->s_bdev &&
2020 (fs_info->sb->s_bdev == device->bdev))
2021 fs_info->sb->s_bdev = next_device->bdev;
2023 if (fs_info->fs_devices->latest_dev->bdev == device->bdev)
2024 fs_info->fs_devices->latest_dev = next_device;
2028 * Return btrfs_fs_devices::num_devices excluding the device that's being
2029 * currently replaced.
2031 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2033 u64 num_devices = fs_info->fs_devices->num_devices;
2035 down_read(&fs_info->dev_replace.rwsem);
2036 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2037 ASSERT(num_devices > 1);
2040 up_read(&fs_info->dev_replace.rwsem);
2045 static void btrfs_scratch_superblock(struct btrfs_fs_info *fs_info,
2046 struct block_device *bdev, int copy_num)
2048 struct btrfs_super_block *disk_super;
2049 const size_t len = sizeof(disk_super->magic);
2050 const u64 bytenr = btrfs_sb_offset(copy_num);
2053 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr);
2054 if (IS_ERR(disk_super))
2057 memset(&disk_super->magic, 0, len);
2058 folio_mark_dirty(virt_to_folio(disk_super));
2059 btrfs_release_disk_super(disk_super);
2061 ret = sync_blockdev_range(bdev, bytenr, bytenr + len - 1);
2063 btrfs_warn(fs_info, "error clearing superblock number %d (%d)",
2067 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
2068 struct block_device *bdev,
2069 const char *device_path)
2076 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2077 if (bdev_is_zoned(bdev))
2078 btrfs_reset_sb_log_zones(bdev, copy_num);
2080 btrfs_scratch_superblock(fs_info, bdev, copy_num);
2083 /* Notify udev that device has changed */
2084 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2086 /* Update ctime/mtime for device path for libblkid */
2087 update_dev_time(device_path);
2090 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
2091 struct btrfs_dev_lookup_args *args,
2092 struct block_device **bdev, void **holder)
2094 struct btrfs_trans_handle *trans;
2095 struct btrfs_device *device;
2096 struct btrfs_fs_devices *cur_devices;
2097 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2101 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2102 btrfs_err(fs_info, "device remove not supported on extent tree v2 yet");
2107 * The device list in fs_devices is accessed without locks (neither
2108 * uuid_mutex nor device_list_mutex) as it won't change on a mounted
2109 * filesystem and another device rm cannot run.
2111 num_devices = btrfs_num_devices(fs_info);
2113 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2117 device = btrfs_find_device(fs_info->fs_devices, args);
2120 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2126 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2127 btrfs_warn_in_rcu(fs_info,
2128 "cannot remove device %s (devid %llu) due to active swapfile",
2129 btrfs_dev_name(device), device->devid);
2133 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
2134 return BTRFS_ERROR_DEV_TGT_REPLACE;
2136 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2137 fs_info->fs_devices->rw_devices == 1)
2138 return BTRFS_ERROR_DEV_ONLY_WRITABLE;
2140 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2141 mutex_lock(&fs_info->chunk_mutex);
2142 list_del_init(&device->dev_alloc_list);
2143 device->fs_devices->rw_devices--;
2144 mutex_unlock(&fs_info->chunk_mutex);
2147 ret = btrfs_shrink_device(device, 0);
2151 trans = btrfs_start_transaction(fs_info->chunk_root, 0);
2152 if (IS_ERR(trans)) {
2153 ret = PTR_ERR(trans);
2157 ret = btrfs_rm_dev_item(trans, device);
2159 /* Any error in dev item removal is critical */
2161 "failed to remove device item for devid %llu: %d",
2162 device->devid, ret);
2163 btrfs_abort_transaction(trans, ret);
2164 btrfs_end_transaction(trans);
2168 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2169 btrfs_scrub_cancel_dev(device);
2172 * the device list mutex makes sure that we don't change
2173 * the device list while someone else is writing out all
2174 * the device supers. Whoever is writing all supers, should
2175 * lock the device list mutex before getting the number of
2176 * devices in the super block (super_copy). Conversely,
2177 * whoever updates the number of devices in the super block
2178 * (super_copy) should hold the device list mutex.
2182 * In normal cases the cur_devices == fs_devices. But in case
2183 * of deleting a seed device, the cur_devices should point to
2184 * its own fs_devices listed under the fs_devices->seed_list.
2186 cur_devices = device->fs_devices;
2187 mutex_lock(&fs_devices->device_list_mutex);
2188 list_del_rcu(&device->dev_list);
2190 cur_devices->num_devices--;
2191 cur_devices->total_devices--;
2192 /* Update total_devices of the parent fs_devices if it's seed */
2193 if (cur_devices != fs_devices)
2194 fs_devices->total_devices--;
2196 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2197 cur_devices->missing_devices--;
2199 btrfs_assign_next_active_device(device, NULL);
2202 cur_devices->open_devices--;
2203 /* remove sysfs entry */
2204 btrfs_sysfs_remove_device(device);
2207 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2208 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2209 mutex_unlock(&fs_devices->device_list_mutex);
2212 * At this point, the device is zero sized and detached from the
2213 * devices list. All that's left is to zero out the old supers and
2216 * We cannot call btrfs_close_bdev() here because we're holding the sb
2217 * write lock, and blkdev_put() will pull in the ->open_mutex on the
2218 * block device and it's dependencies. Instead just flush the device
2219 * and let the caller do the final blkdev_put.
2221 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2222 btrfs_scratch_superblocks(fs_info, device->bdev,
2225 sync_blockdev(device->bdev);
2226 invalidate_bdev(device->bdev);
2230 *bdev = device->bdev;
2231 *holder = device->holder;
2233 btrfs_free_device(device);
2236 * This can happen if cur_devices is the private seed devices list. We
2237 * cannot call close_fs_devices() here because it expects the uuid_mutex
2238 * to be held, but in fact we don't need that for the private
2239 * seed_devices, we can simply decrement cur_devices->opened and then
2240 * remove it from our list and free the fs_devices.
2242 if (cur_devices->num_devices == 0) {
2243 list_del_init(&cur_devices->seed_list);
2244 ASSERT(cur_devices->opened == 1);
2245 cur_devices->opened--;
2246 free_fs_devices(cur_devices);
2249 ret = btrfs_commit_transaction(trans);
2254 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2255 mutex_lock(&fs_info->chunk_mutex);
2256 list_add(&device->dev_alloc_list,
2257 &fs_devices->alloc_list);
2258 device->fs_devices->rw_devices++;
2259 mutex_unlock(&fs_info->chunk_mutex);
2264 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2266 struct btrfs_fs_devices *fs_devices;
2268 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2271 * in case of fs with no seed, srcdev->fs_devices will point
2272 * to fs_devices of fs_info. However when the dev being replaced is
2273 * a seed dev it will point to the seed's local fs_devices. In short
2274 * srcdev will have its correct fs_devices in both the cases.
2276 fs_devices = srcdev->fs_devices;
2278 list_del_rcu(&srcdev->dev_list);
2279 list_del(&srcdev->dev_alloc_list);
2280 fs_devices->num_devices--;
2281 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2282 fs_devices->missing_devices--;
2284 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2285 fs_devices->rw_devices--;
2288 fs_devices->open_devices--;
2291 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2293 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2295 mutex_lock(&uuid_mutex);
2297 btrfs_close_bdev(srcdev);
2299 btrfs_free_device(srcdev);
2301 /* if this is no devs we rather delete the fs_devices */
2302 if (!fs_devices->num_devices) {
2304 * On a mounted FS, num_devices can't be zero unless it's a
2305 * seed. In case of a seed device being replaced, the replace
2306 * target added to the sprout FS, so there will be no more
2307 * device left under the seed FS.
2309 ASSERT(fs_devices->seeding);
2311 list_del_init(&fs_devices->seed_list);
2312 close_fs_devices(fs_devices);
2313 free_fs_devices(fs_devices);
2315 mutex_unlock(&uuid_mutex);
2318 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2320 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2322 mutex_lock(&fs_devices->device_list_mutex);
2324 btrfs_sysfs_remove_device(tgtdev);
2327 fs_devices->open_devices--;
2329 fs_devices->num_devices--;
2331 btrfs_assign_next_active_device(tgtdev, NULL);
2333 list_del_rcu(&tgtdev->dev_list);
2335 mutex_unlock(&fs_devices->device_list_mutex);
2337 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2340 btrfs_close_bdev(tgtdev);
2342 btrfs_free_device(tgtdev);
2346 * Populate args from device at path.
2348 * @fs_info: the filesystem
2349 * @args: the args to populate
2350 * @path: the path to the device
2352 * This will read the super block of the device at @path and populate @args with
2353 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to
2354 * lookup a device to operate on, but need to do it before we take any locks.
2355 * This properly handles the special case of "missing" that a user may pass in,
2356 * and does some basic sanity checks. The caller must make sure that @path is
2357 * properly NUL terminated before calling in, and must call
2358 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and
2361 * Return: 0 for success, -errno for failure
2363 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,
2364 struct btrfs_dev_lookup_args *args,
2367 struct btrfs_super_block *disk_super;
2368 struct block_device *bdev;
2371 if (!path || !path[0])
2373 if (!strcmp(path, "missing")) {
2374 args->missing = true;
2378 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL);
2379 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL);
2380 if (!args->uuid || !args->fsid) {
2381 btrfs_put_dev_args_from_path(args);
2385 ret = btrfs_get_bdev_and_sb(path, BLK_OPEN_READ, NULL, 0,
2386 &bdev, &disk_super);
2388 btrfs_put_dev_args_from_path(args);
2392 args->devid = btrfs_stack_device_id(&disk_super->dev_item);
2393 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE);
2394 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2395 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE);
2397 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
2398 btrfs_release_disk_super(disk_super);
2399 blkdev_put(bdev, NULL);
2404 * Only use this jointly with btrfs_get_dev_args_from_path() because we will
2405 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables
2406 * that don't need to be freed.
2408 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args)
2416 struct btrfs_device *btrfs_find_device_by_devspec(
2417 struct btrfs_fs_info *fs_info, u64 devid,
2418 const char *device_path)
2420 BTRFS_DEV_LOOKUP_ARGS(args);
2421 struct btrfs_device *device;
2426 device = btrfs_find_device(fs_info->fs_devices, &args);
2428 return ERR_PTR(-ENOENT);
2432 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path);
2434 return ERR_PTR(ret);
2435 device = btrfs_find_device(fs_info->fs_devices, &args);
2436 btrfs_put_dev_args_from_path(&args);
2438 return ERR_PTR(-ENOENT);
2442 static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info)
2444 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2445 struct btrfs_fs_devices *old_devices;
2446 struct btrfs_fs_devices *seed_devices;
2448 lockdep_assert_held(&uuid_mutex);
2449 if (!fs_devices->seeding)
2450 return ERR_PTR(-EINVAL);
2453 * Private copy of the seed devices, anchored at
2454 * fs_info->fs_devices->seed_list
2456 seed_devices = alloc_fs_devices(NULL, NULL);
2457 if (IS_ERR(seed_devices))
2458 return seed_devices;
2461 * It's necessary to retain a copy of the original seed fs_devices in
2462 * fs_uuids so that filesystems which have been seeded can successfully
2463 * reference the seed device from open_seed_devices. This also supports
2466 old_devices = clone_fs_devices(fs_devices);
2467 if (IS_ERR(old_devices)) {
2468 kfree(seed_devices);
2472 list_add(&old_devices->fs_list, &fs_uuids);
2474 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2475 seed_devices->opened = 1;
2476 INIT_LIST_HEAD(&seed_devices->devices);
2477 INIT_LIST_HEAD(&seed_devices->alloc_list);
2478 mutex_init(&seed_devices->device_list_mutex);
2480 return seed_devices;
2484 * Splice seed devices into the sprout fs_devices.
2485 * Generate a new fsid for the sprouted read-write filesystem.
2487 static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info,
2488 struct btrfs_fs_devices *seed_devices)
2490 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2491 struct btrfs_super_block *disk_super = fs_info->super_copy;
2492 struct btrfs_device *device;
2496 * We are updating the fsid, the thread leading to device_list_add()
2497 * could race, so uuid_mutex is needed.
2499 lockdep_assert_held(&uuid_mutex);
2502 * The threads listed below may traverse dev_list but can do that without
2503 * device_list_mutex:
2504 * - All device ops and balance - as we are in btrfs_exclop_start.
2505 * - Various dev_list readers - are using RCU.
2506 * - btrfs_ioctl_fitrim() - is using RCU.
2508 * For-read threads as below are using device_list_mutex:
2509 * - Readonly scrub btrfs_scrub_dev()
2510 * - Readonly scrub btrfs_scrub_progress()
2511 * - btrfs_get_dev_stats()
2513 lockdep_assert_held(&fs_devices->device_list_mutex);
2515 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2517 list_for_each_entry(device, &seed_devices->devices, dev_list)
2518 device->fs_devices = seed_devices;
2520 fs_devices->seeding = false;
2521 fs_devices->num_devices = 0;
2522 fs_devices->open_devices = 0;
2523 fs_devices->missing_devices = 0;
2524 fs_devices->rotating = false;
2525 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2527 generate_random_uuid(fs_devices->fsid);
2528 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2529 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2531 super_flags = btrfs_super_flags(disk_super) &
2532 ~BTRFS_SUPER_FLAG_SEEDING;
2533 btrfs_set_super_flags(disk_super, super_flags);
2537 * Store the expected generation for seed devices in device items.
2539 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2541 BTRFS_DEV_LOOKUP_ARGS(args);
2542 struct btrfs_fs_info *fs_info = trans->fs_info;
2543 struct btrfs_root *root = fs_info->chunk_root;
2544 struct btrfs_path *path;
2545 struct extent_buffer *leaf;
2546 struct btrfs_dev_item *dev_item;
2547 struct btrfs_device *device;
2548 struct btrfs_key key;
2549 u8 fs_uuid[BTRFS_FSID_SIZE];
2550 u8 dev_uuid[BTRFS_UUID_SIZE];
2553 path = btrfs_alloc_path();
2557 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2559 key.type = BTRFS_DEV_ITEM_KEY;
2562 btrfs_reserve_chunk_metadata(trans, false);
2563 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2564 btrfs_trans_release_chunk_metadata(trans);
2568 leaf = path->nodes[0];
2570 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2571 ret = btrfs_next_leaf(root, path);
2576 leaf = path->nodes[0];
2577 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2578 btrfs_release_path(path);
2582 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2583 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2584 key.type != BTRFS_DEV_ITEM_KEY)
2587 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2588 struct btrfs_dev_item);
2589 args.devid = btrfs_device_id(leaf, dev_item);
2590 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2592 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2594 args.uuid = dev_uuid;
2595 args.fsid = fs_uuid;
2596 device = btrfs_find_device(fs_info->fs_devices, &args);
2597 BUG_ON(!device); /* Logic error */
2599 if (device->fs_devices->seeding) {
2600 btrfs_set_device_generation(leaf, dev_item,
2601 device->generation);
2602 btrfs_mark_buffer_dirty(leaf);
2610 btrfs_free_path(path);
2614 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2616 struct btrfs_root *root = fs_info->dev_root;
2617 struct btrfs_trans_handle *trans;
2618 struct btrfs_device *device;
2619 struct block_device *bdev;
2620 struct super_block *sb = fs_info->sb;
2621 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2622 struct btrfs_fs_devices *seed_devices = NULL;
2623 u64 orig_super_total_bytes;
2624 u64 orig_super_num_devices;
2626 bool seeding_dev = false;
2627 bool locked = false;
2629 if (sb_rdonly(sb) && !fs_devices->seeding)
2632 bdev = blkdev_get_by_path(device_path, BLK_OPEN_WRITE,
2633 fs_info->bdev_holder, NULL);
2635 return PTR_ERR(bdev);
2637 if (!btrfs_check_device_zone_type(fs_info, bdev)) {
2642 if (fs_devices->seeding) {
2644 down_write(&sb->s_umount);
2645 mutex_lock(&uuid_mutex);
2649 sync_blockdev(bdev);
2652 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2653 if (device->bdev == bdev) {
2661 device = btrfs_alloc_device(fs_info, NULL, NULL, device_path);
2662 if (IS_ERR(device)) {
2663 /* we can safely leave the fs_devices entry around */
2664 ret = PTR_ERR(device);
2668 device->fs_info = fs_info;
2669 device->bdev = bdev;
2670 ret = lookup_bdev(device_path, &device->devt);
2672 goto error_free_device;
2674 ret = btrfs_get_dev_zone_info(device, false);
2676 goto error_free_device;
2678 trans = btrfs_start_transaction(root, 0);
2679 if (IS_ERR(trans)) {
2680 ret = PTR_ERR(trans);
2681 goto error_free_zone;
2684 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2685 device->generation = trans->transid;
2686 device->io_width = fs_info->sectorsize;
2687 device->io_align = fs_info->sectorsize;
2688 device->sector_size = fs_info->sectorsize;
2689 device->total_bytes =
2690 round_down(bdev_nr_bytes(bdev), fs_info->sectorsize);
2691 device->disk_total_bytes = device->total_bytes;
2692 device->commit_total_bytes = device->total_bytes;
2693 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2694 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2695 device->holder = fs_info->bdev_holder;
2696 device->dev_stats_valid = 1;
2697 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2700 btrfs_clear_sb_rdonly(sb);
2702 /* GFP_KERNEL allocation must not be under device_list_mutex */
2703 seed_devices = btrfs_init_sprout(fs_info);
2704 if (IS_ERR(seed_devices)) {
2705 ret = PTR_ERR(seed_devices);
2706 btrfs_abort_transaction(trans, ret);
2711 mutex_lock(&fs_devices->device_list_mutex);
2713 btrfs_setup_sprout(fs_info, seed_devices);
2714 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev,
2718 device->fs_devices = fs_devices;
2720 mutex_lock(&fs_info->chunk_mutex);
2721 list_add_rcu(&device->dev_list, &fs_devices->devices);
2722 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2723 fs_devices->num_devices++;
2724 fs_devices->open_devices++;
2725 fs_devices->rw_devices++;
2726 fs_devices->total_devices++;
2727 fs_devices->total_rw_bytes += device->total_bytes;
2729 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2731 if (!bdev_nonrot(bdev))
2732 fs_devices->rotating = true;
2734 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2735 btrfs_set_super_total_bytes(fs_info->super_copy,
2736 round_down(orig_super_total_bytes + device->total_bytes,
2737 fs_info->sectorsize));
2739 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2740 btrfs_set_super_num_devices(fs_info->super_copy,
2741 orig_super_num_devices + 1);
2744 * we've got more storage, clear any full flags on the space
2747 btrfs_clear_space_info_full(fs_info);
2749 mutex_unlock(&fs_info->chunk_mutex);
2751 /* Add sysfs device entry */
2752 btrfs_sysfs_add_device(device);
2754 mutex_unlock(&fs_devices->device_list_mutex);
2757 mutex_lock(&fs_info->chunk_mutex);
2758 ret = init_first_rw_device(trans);
2759 mutex_unlock(&fs_info->chunk_mutex);
2761 btrfs_abort_transaction(trans, ret);
2766 ret = btrfs_add_dev_item(trans, device);
2768 btrfs_abort_transaction(trans, ret);
2773 ret = btrfs_finish_sprout(trans);
2775 btrfs_abort_transaction(trans, ret);
2780 * fs_devices now represents the newly sprouted filesystem and
2781 * its fsid has been changed by btrfs_sprout_splice().
2783 btrfs_sysfs_update_sprout_fsid(fs_devices);
2786 ret = btrfs_commit_transaction(trans);
2789 mutex_unlock(&uuid_mutex);
2790 up_write(&sb->s_umount);
2793 if (ret) /* transaction commit */
2796 ret = btrfs_relocate_sys_chunks(fs_info);
2798 btrfs_handle_fs_error(fs_info, ret,
2799 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2800 trans = btrfs_attach_transaction(root);
2801 if (IS_ERR(trans)) {
2802 if (PTR_ERR(trans) == -ENOENT)
2804 ret = PTR_ERR(trans);
2808 ret = btrfs_commit_transaction(trans);
2812 * Now that we have written a new super block to this device, check all
2813 * other fs_devices list if device_path alienates any other scanned
2815 * We can ignore the return value as it typically returns -EINVAL and
2816 * only succeeds if the device was an alien.
2818 btrfs_forget_devices(device->devt);
2820 /* Update ctime/mtime for blkid or udev */
2821 update_dev_time(device_path);
2826 btrfs_sysfs_remove_device(device);
2827 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2828 mutex_lock(&fs_info->chunk_mutex);
2829 list_del_rcu(&device->dev_list);
2830 list_del(&device->dev_alloc_list);
2831 fs_info->fs_devices->num_devices--;
2832 fs_info->fs_devices->open_devices--;
2833 fs_info->fs_devices->rw_devices--;
2834 fs_info->fs_devices->total_devices--;
2835 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2836 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2837 btrfs_set_super_total_bytes(fs_info->super_copy,
2838 orig_super_total_bytes);
2839 btrfs_set_super_num_devices(fs_info->super_copy,
2840 orig_super_num_devices);
2841 mutex_unlock(&fs_info->chunk_mutex);
2842 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2845 btrfs_set_sb_rdonly(sb);
2847 btrfs_end_transaction(trans);
2849 btrfs_destroy_dev_zone_info(device);
2851 btrfs_free_device(device);
2853 blkdev_put(bdev, fs_info->bdev_holder);
2855 mutex_unlock(&uuid_mutex);
2856 up_write(&sb->s_umount);
2861 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2862 struct btrfs_device *device)
2865 struct btrfs_path *path;
2866 struct btrfs_root *root = device->fs_info->chunk_root;
2867 struct btrfs_dev_item *dev_item;
2868 struct extent_buffer *leaf;
2869 struct btrfs_key key;
2871 path = btrfs_alloc_path();
2875 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2876 key.type = BTRFS_DEV_ITEM_KEY;
2877 key.offset = device->devid;
2879 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2888 leaf = path->nodes[0];
2889 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2891 btrfs_set_device_id(leaf, dev_item, device->devid);
2892 btrfs_set_device_type(leaf, dev_item, device->type);
2893 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2894 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2895 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2896 btrfs_set_device_total_bytes(leaf, dev_item,
2897 btrfs_device_get_disk_total_bytes(device));
2898 btrfs_set_device_bytes_used(leaf, dev_item,
2899 btrfs_device_get_bytes_used(device));
2900 btrfs_mark_buffer_dirty(leaf);
2903 btrfs_free_path(path);
2907 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2908 struct btrfs_device *device, u64 new_size)
2910 struct btrfs_fs_info *fs_info = device->fs_info;
2911 struct btrfs_super_block *super_copy = fs_info->super_copy;
2916 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2919 new_size = round_down(new_size, fs_info->sectorsize);
2921 mutex_lock(&fs_info->chunk_mutex);
2922 old_total = btrfs_super_total_bytes(super_copy);
2923 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2925 if (new_size <= device->total_bytes ||
2926 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2927 mutex_unlock(&fs_info->chunk_mutex);
2931 btrfs_set_super_total_bytes(super_copy,
2932 round_down(old_total + diff, fs_info->sectorsize));
2933 device->fs_devices->total_rw_bytes += diff;
2935 btrfs_device_set_total_bytes(device, new_size);
2936 btrfs_device_set_disk_total_bytes(device, new_size);
2937 btrfs_clear_space_info_full(device->fs_info);
2938 if (list_empty(&device->post_commit_list))
2939 list_add_tail(&device->post_commit_list,
2940 &trans->transaction->dev_update_list);
2941 mutex_unlock(&fs_info->chunk_mutex);
2943 btrfs_reserve_chunk_metadata(trans, false);
2944 ret = btrfs_update_device(trans, device);
2945 btrfs_trans_release_chunk_metadata(trans);
2950 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2952 struct btrfs_fs_info *fs_info = trans->fs_info;
2953 struct btrfs_root *root = fs_info->chunk_root;
2955 struct btrfs_path *path;
2956 struct btrfs_key key;
2958 path = btrfs_alloc_path();
2962 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2963 key.offset = chunk_offset;
2964 key.type = BTRFS_CHUNK_ITEM_KEY;
2966 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2969 else if (ret > 0) { /* Logic error or corruption */
2970 btrfs_handle_fs_error(fs_info, -ENOENT,
2971 "Failed lookup while freeing chunk.");
2976 ret = btrfs_del_item(trans, root, path);
2978 btrfs_handle_fs_error(fs_info, ret,
2979 "Failed to delete chunk item.");
2981 btrfs_free_path(path);
2985 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2987 struct btrfs_super_block *super_copy = fs_info->super_copy;
2988 struct btrfs_disk_key *disk_key;
2989 struct btrfs_chunk *chunk;
2996 struct btrfs_key key;
2998 lockdep_assert_held(&fs_info->chunk_mutex);
2999 array_size = btrfs_super_sys_array_size(super_copy);
3001 ptr = super_copy->sys_chunk_array;
3004 while (cur < array_size) {
3005 disk_key = (struct btrfs_disk_key *)ptr;
3006 btrfs_disk_key_to_cpu(&key, disk_key);
3008 len = sizeof(*disk_key);
3010 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3011 chunk = (struct btrfs_chunk *)(ptr + len);
3012 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
3013 len += btrfs_chunk_item_size(num_stripes);
3018 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
3019 key.offset == chunk_offset) {
3020 memmove(ptr, ptr + len, array_size - (cur + len));
3022 btrfs_set_super_sys_array_size(super_copy, array_size);
3032 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
3033 * @logical: Logical block offset in bytes.
3034 * @length: Length of extent in bytes.
3036 * Return: Chunk mapping or ERR_PTR.
3038 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
3039 u64 logical, u64 length)
3041 struct extent_map_tree *em_tree;
3042 struct extent_map *em;
3044 em_tree = &fs_info->mapping_tree;
3045 read_lock(&em_tree->lock);
3046 em = lookup_extent_mapping(em_tree, logical, length);
3047 read_unlock(&em_tree->lock);
3050 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
3052 return ERR_PTR(-EINVAL);
3055 if (em->start > logical || em->start + em->len < logical) {
3057 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
3058 logical, length, em->start, em->start + em->len);
3059 free_extent_map(em);
3060 return ERR_PTR(-EINVAL);
3063 /* callers are responsible for dropping em's ref. */
3067 static int remove_chunk_item(struct btrfs_trans_handle *trans,
3068 struct map_lookup *map, u64 chunk_offset)
3073 * Removing chunk items and updating the device items in the chunks btree
3074 * requires holding the chunk_mutex.
3075 * See the comment at btrfs_chunk_alloc() for the details.
3077 lockdep_assert_held(&trans->fs_info->chunk_mutex);
3079 for (i = 0; i < map->num_stripes; i++) {
3082 ret = btrfs_update_device(trans, map->stripes[i].dev);
3087 return btrfs_free_chunk(trans, chunk_offset);
3090 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
3092 struct btrfs_fs_info *fs_info = trans->fs_info;
3093 struct extent_map *em;
3094 struct map_lookup *map;
3095 u64 dev_extent_len = 0;
3097 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3099 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
3102 * This is a logic error, but we don't want to just rely on the
3103 * user having built with ASSERT enabled, so if ASSERT doesn't
3104 * do anything we still error out.
3109 map = em->map_lookup;
3112 * First delete the device extent items from the devices btree.
3113 * We take the device_list_mutex to avoid racing with the finishing phase
3114 * of a device replace operation. See the comment below before acquiring
3115 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
3116 * because that can result in a deadlock when deleting the device extent
3117 * items from the devices btree - COWing an extent buffer from the btree
3118 * may result in allocating a new metadata chunk, which would attempt to
3119 * lock again fs_info->chunk_mutex.
3121 mutex_lock(&fs_devices->device_list_mutex);
3122 for (i = 0; i < map->num_stripes; i++) {
3123 struct btrfs_device *device = map->stripes[i].dev;
3124 ret = btrfs_free_dev_extent(trans, device,
3125 map->stripes[i].physical,
3128 mutex_unlock(&fs_devices->device_list_mutex);
3129 btrfs_abort_transaction(trans, ret);
3133 if (device->bytes_used > 0) {
3134 mutex_lock(&fs_info->chunk_mutex);
3135 btrfs_device_set_bytes_used(device,
3136 device->bytes_used - dev_extent_len);
3137 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3138 btrfs_clear_space_info_full(fs_info);
3139 mutex_unlock(&fs_info->chunk_mutex);
3142 mutex_unlock(&fs_devices->device_list_mutex);
3145 * We acquire fs_info->chunk_mutex for 2 reasons:
3147 * 1) Just like with the first phase of the chunk allocation, we must
3148 * reserve system space, do all chunk btree updates and deletions, and
3149 * update the system chunk array in the superblock while holding this
3150 * mutex. This is for similar reasons as explained on the comment at
3151 * the top of btrfs_chunk_alloc();
3153 * 2) Prevent races with the final phase of a device replace operation
3154 * that replaces the device object associated with the map's stripes,
3155 * because the device object's id can change at any time during that
3156 * final phase of the device replace operation
3157 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3158 * replaced device and then see it with an ID of
3159 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3160 * the device item, which does not exists on the chunk btree.
3161 * The finishing phase of device replace acquires both the
3162 * device_list_mutex and the chunk_mutex, in that order, so we are
3163 * safe by just acquiring the chunk_mutex.
3165 trans->removing_chunk = true;
3166 mutex_lock(&fs_info->chunk_mutex);
3168 check_system_chunk(trans, map->type);
3170 ret = remove_chunk_item(trans, map, chunk_offset);
3172 * Normally we should not get -ENOSPC since we reserved space before
3173 * through the call to check_system_chunk().
3175 * Despite our system space_info having enough free space, we may not
3176 * be able to allocate extents from its block groups, because all have
3177 * an incompatible profile, which will force us to allocate a new system
3178 * block group with the right profile, or right after we called
3179 * check_system_space() above, a scrub turned the only system block group
3180 * with enough free space into RO mode.
3181 * This is explained with more detail at do_chunk_alloc().
3183 * So if we get -ENOSPC, allocate a new system chunk and retry once.
3185 if (ret == -ENOSPC) {
3186 const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
3187 struct btrfs_block_group *sys_bg;
3189 sys_bg = btrfs_create_chunk(trans, sys_flags);
3190 if (IS_ERR(sys_bg)) {
3191 ret = PTR_ERR(sys_bg);
3192 btrfs_abort_transaction(trans, ret);
3196 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3198 btrfs_abort_transaction(trans, ret);
3202 ret = remove_chunk_item(trans, map, chunk_offset);
3204 btrfs_abort_transaction(trans, ret);
3208 btrfs_abort_transaction(trans, ret);
3212 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3214 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3215 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3217 btrfs_abort_transaction(trans, ret);
3222 mutex_unlock(&fs_info->chunk_mutex);
3223 trans->removing_chunk = false;
3226 * We are done with chunk btree updates and deletions, so release the
3227 * system space we previously reserved (with check_system_chunk()).
3229 btrfs_trans_release_chunk_metadata(trans);
3231 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3233 btrfs_abort_transaction(trans, ret);
3238 if (trans->removing_chunk) {
3239 mutex_unlock(&fs_info->chunk_mutex);
3240 trans->removing_chunk = false;
3243 free_extent_map(em);
3247 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3249 struct btrfs_root *root = fs_info->chunk_root;
3250 struct btrfs_trans_handle *trans;
3251 struct btrfs_block_group *block_group;
3255 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3257 "relocate: not supported on extent tree v2 yet");
3262 * Prevent races with automatic removal of unused block groups.
3263 * After we relocate and before we remove the chunk with offset
3264 * chunk_offset, automatic removal of the block group can kick in,
3265 * resulting in a failure when calling btrfs_remove_chunk() below.
3267 * Make sure to acquire this mutex before doing a tree search (dev
3268 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3269 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3270 * we release the path used to search the chunk/dev tree and before
3271 * the current task acquires this mutex and calls us.
3273 lockdep_assert_held(&fs_info->reclaim_bgs_lock);
3275 /* step one, relocate all the extents inside this chunk */
3276 btrfs_scrub_pause(fs_info);
3277 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3278 btrfs_scrub_continue(fs_info);
3281 * If we had a transaction abort, stop all running scrubs.
3282 * See transaction.c:cleanup_transaction() why we do it here.
3284 if (BTRFS_FS_ERROR(fs_info))
3285 btrfs_scrub_cancel(fs_info);
3289 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3292 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3293 length = block_group->length;
3294 btrfs_put_block_group(block_group);
3297 * On a zoned file system, discard the whole block group, this will
3298 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3299 * resetting the zone fails, don't treat it as a fatal problem from the
3300 * filesystem's point of view.
3302 if (btrfs_is_zoned(fs_info)) {
3303 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
3306 "failed to reset zone %llu after relocation",
3310 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3312 if (IS_ERR(trans)) {
3313 ret = PTR_ERR(trans);
3314 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3319 * step two, delete the device extents and the
3320 * chunk tree entries
3322 ret = btrfs_remove_chunk(trans, chunk_offset);
3323 btrfs_end_transaction(trans);
3327 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3329 struct btrfs_root *chunk_root = fs_info->chunk_root;
3330 struct btrfs_path *path;
3331 struct extent_buffer *leaf;
3332 struct btrfs_chunk *chunk;
3333 struct btrfs_key key;
3334 struct btrfs_key found_key;
3336 bool retried = false;
3340 path = btrfs_alloc_path();
3345 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3346 key.offset = (u64)-1;
3347 key.type = BTRFS_CHUNK_ITEM_KEY;
3350 mutex_lock(&fs_info->reclaim_bgs_lock);
3351 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3353 mutex_unlock(&fs_info->reclaim_bgs_lock);
3356 BUG_ON(ret == 0); /* Corruption */
3358 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3361 mutex_unlock(&fs_info->reclaim_bgs_lock);
3367 leaf = path->nodes[0];
3368 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3370 chunk = btrfs_item_ptr(leaf, path->slots[0],
3371 struct btrfs_chunk);
3372 chunk_type = btrfs_chunk_type(leaf, chunk);
3373 btrfs_release_path(path);
3375 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3376 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3382 mutex_unlock(&fs_info->reclaim_bgs_lock);
3384 if (found_key.offset == 0)
3386 key.offset = found_key.offset - 1;
3389 if (failed && !retried) {
3393 } else if (WARN_ON(failed && retried)) {
3397 btrfs_free_path(path);
3402 * return 1 : allocate a data chunk successfully,
3403 * return <0: errors during allocating a data chunk,
3404 * return 0 : no need to allocate a data chunk.
3406 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3409 struct btrfs_block_group *cache;
3413 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3415 chunk_type = cache->flags;
3416 btrfs_put_block_group(cache);
3418 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3421 spin_lock(&fs_info->data_sinfo->lock);
3422 bytes_used = fs_info->data_sinfo->bytes_used;
3423 spin_unlock(&fs_info->data_sinfo->lock);
3426 struct btrfs_trans_handle *trans;
3429 trans = btrfs_join_transaction(fs_info->tree_root);
3431 return PTR_ERR(trans);
3433 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3434 btrfs_end_transaction(trans);
3443 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3444 struct btrfs_balance_control *bctl)
3446 struct btrfs_root *root = fs_info->tree_root;
3447 struct btrfs_trans_handle *trans;
3448 struct btrfs_balance_item *item;
3449 struct btrfs_disk_balance_args disk_bargs;
3450 struct btrfs_path *path;
3451 struct extent_buffer *leaf;
3452 struct btrfs_key key;
3455 path = btrfs_alloc_path();
3459 trans = btrfs_start_transaction(root, 0);
3460 if (IS_ERR(trans)) {
3461 btrfs_free_path(path);
3462 return PTR_ERR(trans);
3465 key.objectid = BTRFS_BALANCE_OBJECTID;
3466 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3469 ret = btrfs_insert_empty_item(trans, root, path, &key,
3474 leaf = path->nodes[0];
3475 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3477 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3479 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3480 btrfs_set_balance_data(leaf, item, &disk_bargs);
3481 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3482 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3483 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3484 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3486 btrfs_set_balance_flags(leaf, item, bctl->flags);
3488 btrfs_mark_buffer_dirty(leaf);
3490 btrfs_free_path(path);
3491 err = btrfs_commit_transaction(trans);
3497 static int del_balance_item(struct btrfs_fs_info *fs_info)
3499 struct btrfs_root *root = fs_info->tree_root;
3500 struct btrfs_trans_handle *trans;
3501 struct btrfs_path *path;
3502 struct btrfs_key key;
3505 path = btrfs_alloc_path();
3509 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3510 if (IS_ERR(trans)) {
3511 btrfs_free_path(path);
3512 return PTR_ERR(trans);
3515 key.objectid = BTRFS_BALANCE_OBJECTID;
3516 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3519 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3527 ret = btrfs_del_item(trans, root, path);
3529 btrfs_free_path(path);
3530 err = btrfs_commit_transaction(trans);
3537 * This is a heuristic used to reduce the number of chunks balanced on
3538 * resume after balance was interrupted.
3540 static void update_balance_args(struct btrfs_balance_control *bctl)
3543 * Turn on soft mode for chunk types that were being converted.
3545 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3546 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3547 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3548 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3549 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3550 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3553 * Turn on usage filter if is not already used. The idea is
3554 * that chunks that we have already balanced should be
3555 * reasonably full. Don't do it for chunks that are being
3556 * converted - that will keep us from relocating unconverted
3557 * (albeit full) chunks.
3559 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3560 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3561 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3562 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3563 bctl->data.usage = 90;
3565 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3566 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3567 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3568 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3569 bctl->sys.usage = 90;
3571 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3572 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3573 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3574 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3575 bctl->meta.usage = 90;
3580 * Clear the balance status in fs_info and delete the balance item from disk.
3582 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3584 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3587 BUG_ON(!fs_info->balance_ctl);
3589 spin_lock(&fs_info->balance_lock);
3590 fs_info->balance_ctl = NULL;
3591 spin_unlock(&fs_info->balance_lock);
3594 ret = del_balance_item(fs_info);
3596 btrfs_handle_fs_error(fs_info, ret, NULL);
3600 * Balance filters. Return 1 if chunk should be filtered out
3601 * (should not be balanced).
3603 static int chunk_profiles_filter(u64 chunk_type,
3604 struct btrfs_balance_args *bargs)
3606 chunk_type = chunk_to_extended(chunk_type) &
3607 BTRFS_EXTENDED_PROFILE_MASK;
3609 if (bargs->profiles & chunk_type)
3615 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3616 struct btrfs_balance_args *bargs)
3618 struct btrfs_block_group *cache;
3620 u64 user_thresh_min;
3621 u64 user_thresh_max;
3624 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3625 chunk_used = cache->used;
3627 if (bargs->usage_min == 0)
3628 user_thresh_min = 0;
3630 user_thresh_min = mult_perc(cache->length, bargs->usage_min);
3632 if (bargs->usage_max == 0)
3633 user_thresh_max = 1;
3634 else if (bargs->usage_max > 100)
3635 user_thresh_max = cache->length;
3637 user_thresh_max = mult_perc(cache->length, bargs->usage_max);
3639 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3642 btrfs_put_block_group(cache);
3646 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3647 u64 chunk_offset, struct btrfs_balance_args *bargs)
3649 struct btrfs_block_group *cache;
3650 u64 chunk_used, user_thresh;
3653 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3654 chunk_used = cache->used;
3656 if (bargs->usage_min == 0)
3658 else if (bargs->usage > 100)
3659 user_thresh = cache->length;
3661 user_thresh = mult_perc(cache->length, bargs->usage);
3663 if (chunk_used < user_thresh)
3666 btrfs_put_block_group(cache);
3670 static int chunk_devid_filter(struct extent_buffer *leaf,
3671 struct btrfs_chunk *chunk,
3672 struct btrfs_balance_args *bargs)
3674 struct btrfs_stripe *stripe;
3675 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3678 for (i = 0; i < num_stripes; i++) {
3679 stripe = btrfs_stripe_nr(chunk, i);
3680 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3687 static u64 calc_data_stripes(u64 type, int num_stripes)
3689 const int index = btrfs_bg_flags_to_raid_index(type);
3690 const int ncopies = btrfs_raid_array[index].ncopies;
3691 const int nparity = btrfs_raid_array[index].nparity;
3693 return (num_stripes - nparity) / ncopies;
3696 /* [pstart, pend) */
3697 static int chunk_drange_filter(struct extent_buffer *leaf,
3698 struct btrfs_chunk *chunk,
3699 struct btrfs_balance_args *bargs)
3701 struct btrfs_stripe *stripe;
3702 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3709 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3712 type = btrfs_chunk_type(leaf, chunk);
3713 factor = calc_data_stripes(type, num_stripes);
3715 for (i = 0; i < num_stripes; i++) {
3716 stripe = btrfs_stripe_nr(chunk, i);
3717 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3720 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3721 stripe_length = btrfs_chunk_length(leaf, chunk);
3722 stripe_length = div_u64(stripe_length, factor);
3724 if (stripe_offset < bargs->pend &&
3725 stripe_offset + stripe_length > bargs->pstart)
3732 /* [vstart, vend) */
3733 static int chunk_vrange_filter(struct extent_buffer *leaf,
3734 struct btrfs_chunk *chunk,
3736 struct btrfs_balance_args *bargs)
3738 if (chunk_offset < bargs->vend &&
3739 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3740 /* at least part of the chunk is inside this vrange */
3746 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3747 struct btrfs_chunk *chunk,
3748 struct btrfs_balance_args *bargs)
3750 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3752 if (bargs->stripes_min <= num_stripes
3753 && num_stripes <= bargs->stripes_max)
3759 static int chunk_soft_convert_filter(u64 chunk_type,
3760 struct btrfs_balance_args *bargs)
3762 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3765 chunk_type = chunk_to_extended(chunk_type) &
3766 BTRFS_EXTENDED_PROFILE_MASK;
3768 if (bargs->target == chunk_type)
3774 static int should_balance_chunk(struct extent_buffer *leaf,
3775 struct btrfs_chunk *chunk, u64 chunk_offset)
3777 struct btrfs_fs_info *fs_info = leaf->fs_info;
3778 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3779 struct btrfs_balance_args *bargs = NULL;
3780 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3783 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3784 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3788 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3789 bargs = &bctl->data;
3790 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3792 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3793 bargs = &bctl->meta;
3795 /* profiles filter */
3796 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3797 chunk_profiles_filter(chunk_type, bargs)) {
3802 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3803 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3805 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3806 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3811 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3812 chunk_devid_filter(leaf, chunk, bargs)) {
3816 /* drange filter, makes sense only with devid filter */
3817 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3818 chunk_drange_filter(leaf, chunk, bargs)) {
3823 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3824 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3828 /* stripes filter */
3829 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3830 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3834 /* soft profile changing mode */
3835 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3836 chunk_soft_convert_filter(chunk_type, bargs)) {
3841 * limited by count, must be the last filter
3843 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3844 if (bargs->limit == 0)
3848 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3850 * Same logic as the 'limit' filter; the minimum cannot be
3851 * determined here because we do not have the global information
3852 * about the count of all chunks that satisfy the filters.
3854 if (bargs->limit_max == 0)
3863 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3865 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3866 struct btrfs_root *chunk_root = fs_info->chunk_root;
3868 struct btrfs_chunk *chunk;
3869 struct btrfs_path *path = NULL;
3870 struct btrfs_key key;
3871 struct btrfs_key found_key;
3872 struct extent_buffer *leaf;
3875 int enospc_errors = 0;
3876 bool counting = true;
3877 /* The single value limit and min/max limits use the same bytes in the */
3878 u64 limit_data = bctl->data.limit;
3879 u64 limit_meta = bctl->meta.limit;
3880 u64 limit_sys = bctl->sys.limit;
3884 int chunk_reserved = 0;
3886 path = btrfs_alloc_path();
3892 /* zero out stat counters */
3893 spin_lock(&fs_info->balance_lock);
3894 memset(&bctl->stat, 0, sizeof(bctl->stat));
3895 spin_unlock(&fs_info->balance_lock);
3899 * The single value limit and min/max limits use the same bytes
3902 bctl->data.limit = limit_data;
3903 bctl->meta.limit = limit_meta;
3904 bctl->sys.limit = limit_sys;
3906 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3907 key.offset = (u64)-1;
3908 key.type = BTRFS_CHUNK_ITEM_KEY;
3911 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3912 atomic_read(&fs_info->balance_cancel_req)) {
3917 mutex_lock(&fs_info->reclaim_bgs_lock);
3918 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3920 mutex_unlock(&fs_info->reclaim_bgs_lock);
3925 * this shouldn't happen, it means the last relocate
3929 BUG(); /* FIXME break ? */
3931 ret = btrfs_previous_item(chunk_root, path, 0,
3932 BTRFS_CHUNK_ITEM_KEY);
3934 mutex_unlock(&fs_info->reclaim_bgs_lock);
3939 leaf = path->nodes[0];
3940 slot = path->slots[0];
3941 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3943 if (found_key.objectid != key.objectid) {
3944 mutex_unlock(&fs_info->reclaim_bgs_lock);
3948 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3949 chunk_type = btrfs_chunk_type(leaf, chunk);
3952 spin_lock(&fs_info->balance_lock);
3953 bctl->stat.considered++;
3954 spin_unlock(&fs_info->balance_lock);
3957 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3959 btrfs_release_path(path);
3961 mutex_unlock(&fs_info->reclaim_bgs_lock);
3966 mutex_unlock(&fs_info->reclaim_bgs_lock);
3967 spin_lock(&fs_info->balance_lock);
3968 bctl->stat.expected++;
3969 spin_unlock(&fs_info->balance_lock);
3971 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3973 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3975 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3982 * Apply limit_min filter, no need to check if the LIMITS
3983 * filter is used, limit_min is 0 by default
3985 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3986 count_data < bctl->data.limit_min)
3987 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3988 count_meta < bctl->meta.limit_min)
3989 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3990 count_sys < bctl->sys.limit_min)) {
3991 mutex_unlock(&fs_info->reclaim_bgs_lock);
3995 if (!chunk_reserved) {
3997 * We may be relocating the only data chunk we have,
3998 * which could potentially end up with losing data's
3999 * raid profile, so lets allocate an empty one in
4002 ret = btrfs_may_alloc_data_chunk(fs_info,
4005 mutex_unlock(&fs_info->reclaim_bgs_lock);
4007 } else if (ret == 1) {
4012 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
4013 mutex_unlock(&fs_info->reclaim_bgs_lock);
4014 if (ret == -ENOSPC) {
4016 } else if (ret == -ETXTBSY) {
4018 "skipping relocation of block group %llu due to active swapfile",
4024 spin_lock(&fs_info->balance_lock);
4025 bctl->stat.completed++;
4026 spin_unlock(&fs_info->balance_lock);
4029 if (found_key.offset == 0)
4031 key.offset = found_key.offset - 1;
4035 btrfs_release_path(path);
4040 btrfs_free_path(path);
4041 if (enospc_errors) {
4042 btrfs_info(fs_info, "%d enospc errors during balance",
4052 * See if a given profile is valid and reduced.
4054 * @flags: profile to validate
4055 * @extended: if true @flags is treated as an extended profile
4057 static int alloc_profile_is_valid(u64 flags, int extended)
4059 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
4060 BTRFS_BLOCK_GROUP_PROFILE_MASK);
4062 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
4064 /* 1) check that all other bits are zeroed */
4068 /* 2) see if profile is reduced */
4070 return !extended; /* "0" is valid for usual profiles */
4072 return has_single_bit_set(flags);
4076 * Validate target profile against allowed profiles and return true if it's OK.
4077 * Otherwise print the error message and return false.
4079 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
4080 const struct btrfs_balance_args *bargs,
4081 u64 allowed, const char *type)
4083 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
4086 /* Profile is valid and does not have bits outside of the allowed set */
4087 if (alloc_profile_is_valid(bargs->target, 1) &&
4088 (bargs->target & ~allowed) == 0)
4091 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
4092 type, btrfs_bg_type_to_raid_name(bargs->target));
4097 * Fill @buf with textual description of balance filter flags @bargs, up to
4098 * @size_buf including the terminating null. The output may be trimmed if it
4099 * does not fit into the provided buffer.
4101 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
4105 u32 size_bp = size_buf;
4107 u64 flags = bargs->flags;
4108 char tmp_buf[128] = {'\0'};
4113 #define CHECK_APPEND_NOARG(a) \
4115 ret = snprintf(bp, size_bp, (a)); \
4116 if (ret < 0 || ret >= size_bp) \
4117 goto out_overflow; \
4122 #define CHECK_APPEND_1ARG(a, v1) \
4124 ret = snprintf(bp, size_bp, (a), (v1)); \
4125 if (ret < 0 || ret >= size_bp) \
4126 goto out_overflow; \
4131 #define CHECK_APPEND_2ARG(a, v1, v2) \
4133 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4134 if (ret < 0 || ret >= size_bp) \
4135 goto out_overflow; \
4140 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
4141 CHECK_APPEND_1ARG("convert=%s,",
4142 btrfs_bg_type_to_raid_name(bargs->target));
4144 if (flags & BTRFS_BALANCE_ARGS_SOFT)
4145 CHECK_APPEND_NOARG("soft,");
4147 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4148 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4150 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4153 if (flags & BTRFS_BALANCE_ARGS_USAGE)
4154 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4156 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4157 CHECK_APPEND_2ARG("usage=%u..%u,",
4158 bargs->usage_min, bargs->usage_max);
4160 if (flags & BTRFS_BALANCE_ARGS_DEVID)
4161 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4163 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4164 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4165 bargs->pstart, bargs->pend);
4167 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4168 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4169 bargs->vstart, bargs->vend);
4171 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4172 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4174 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4175 CHECK_APPEND_2ARG("limit=%u..%u,",
4176 bargs->limit_min, bargs->limit_max);
4178 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4179 CHECK_APPEND_2ARG("stripes=%u..%u,",
4180 bargs->stripes_min, bargs->stripes_max);
4182 #undef CHECK_APPEND_2ARG
4183 #undef CHECK_APPEND_1ARG
4184 #undef CHECK_APPEND_NOARG
4188 if (size_bp < size_buf)
4189 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4194 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4196 u32 size_buf = 1024;
4197 char tmp_buf[192] = {'\0'};
4200 u32 size_bp = size_buf;
4202 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4204 buf = kzalloc(size_buf, GFP_KERNEL);
4210 #define CHECK_APPEND_1ARG(a, v1) \
4212 ret = snprintf(bp, size_bp, (a), (v1)); \
4213 if (ret < 0 || ret >= size_bp) \
4214 goto out_overflow; \
4219 if (bctl->flags & BTRFS_BALANCE_FORCE)
4220 CHECK_APPEND_1ARG("%s", "-f ");
4222 if (bctl->flags & BTRFS_BALANCE_DATA) {
4223 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4224 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4227 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4228 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4229 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4232 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4233 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4234 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4237 #undef CHECK_APPEND_1ARG
4241 if (size_bp < size_buf)
4242 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4243 btrfs_info(fs_info, "balance: %s %s",
4244 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4245 "resume" : "start", buf);
4251 * Should be called with balance mutexe held
4253 int btrfs_balance(struct btrfs_fs_info *fs_info,
4254 struct btrfs_balance_control *bctl,
4255 struct btrfs_ioctl_balance_args *bargs)
4257 u64 meta_target, data_target;
4263 bool reducing_redundancy;
4264 bool paused = false;
4267 if (btrfs_fs_closing(fs_info) ||
4268 atomic_read(&fs_info->balance_pause_req) ||
4269 btrfs_should_cancel_balance(fs_info)) {
4274 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4275 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4279 * In case of mixed groups both data and meta should be picked,
4280 * and identical options should be given for both of them.
4282 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4283 if (mixed && (bctl->flags & allowed)) {
4284 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4285 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4286 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4288 "balance: mixed groups data and metadata options must be the same");
4295 * rw_devices will not change at the moment, device add/delete/replace
4298 num_devices = fs_info->fs_devices->rw_devices;
4301 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4302 * special bit for it, to make it easier to distinguish. Thus we need
4303 * to set it manually, or balance would refuse the profile.
4305 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4306 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4307 if (num_devices >= btrfs_raid_array[i].devs_min)
4308 allowed |= btrfs_raid_array[i].bg_flag;
4310 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4311 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4312 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4318 * Allow to reduce metadata or system integrity only if force set for
4319 * profiles with redundancy (copies, parity)
4322 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4323 if (btrfs_raid_array[i].ncopies >= 2 ||
4324 btrfs_raid_array[i].tolerated_failures >= 1)
4325 allowed |= btrfs_raid_array[i].bg_flag;
4328 seq = read_seqbegin(&fs_info->profiles_lock);
4330 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4331 (fs_info->avail_system_alloc_bits & allowed) &&
4332 !(bctl->sys.target & allowed)) ||
4333 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4334 (fs_info->avail_metadata_alloc_bits & allowed) &&
4335 !(bctl->meta.target & allowed)))
4336 reducing_redundancy = true;
4338 reducing_redundancy = false;
4340 /* if we're not converting, the target field is uninitialized */
4341 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4342 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4343 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4344 bctl->data.target : fs_info->avail_data_alloc_bits;
4345 } while (read_seqretry(&fs_info->profiles_lock, seq));
4347 if (reducing_redundancy) {
4348 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4350 "balance: force reducing metadata redundancy");
4353 "balance: reduces metadata redundancy, use --force if you want this");
4359 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4360 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4362 "balance: metadata profile %s has lower redundancy than data profile %s",
4363 btrfs_bg_type_to_raid_name(meta_target),
4364 btrfs_bg_type_to_raid_name(data_target));
4367 ret = insert_balance_item(fs_info, bctl);
4368 if (ret && ret != -EEXIST)
4371 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4372 BUG_ON(ret == -EEXIST);
4373 BUG_ON(fs_info->balance_ctl);
4374 spin_lock(&fs_info->balance_lock);
4375 fs_info->balance_ctl = bctl;
4376 spin_unlock(&fs_info->balance_lock);
4378 BUG_ON(ret != -EEXIST);
4379 spin_lock(&fs_info->balance_lock);
4380 update_balance_args(bctl);
4381 spin_unlock(&fs_info->balance_lock);
4384 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4385 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4386 describe_balance_start_or_resume(fs_info);
4387 mutex_unlock(&fs_info->balance_mutex);
4389 ret = __btrfs_balance(fs_info);
4391 mutex_lock(&fs_info->balance_mutex);
4392 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) {
4393 btrfs_info(fs_info, "balance: paused");
4394 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
4398 * Balance can be canceled by:
4400 * - Regular cancel request
4401 * Then ret == -ECANCELED and balance_cancel_req > 0
4403 * - Fatal signal to "btrfs" process
4404 * Either the signal caught by wait_reserve_ticket() and callers
4405 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4407 * Either way, in this case balance_cancel_req = 0, and
4408 * ret == -EINTR or ret == -ECANCELED.
4410 * So here we only check the return value to catch canceled balance.
4412 else if (ret == -ECANCELED || ret == -EINTR)
4413 btrfs_info(fs_info, "balance: canceled");
4415 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4417 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4420 memset(bargs, 0, sizeof(*bargs));
4421 btrfs_update_ioctl_balance_args(fs_info, bargs);
4424 /* We didn't pause, we can clean everything up. */
4426 reset_balance_state(fs_info);
4427 btrfs_exclop_finish(fs_info);
4430 wake_up(&fs_info->balance_wait_q);
4434 if (bctl->flags & BTRFS_BALANCE_RESUME)
4435 reset_balance_state(fs_info);
4438 btrfs_exclop_finish(fs_info);
4443 static int balance_kthread(void *data)
4445 struct btrfs_fs_info *fs_info = data;
4448 sb_start_write(fs_info->sb);
4449 mutex_lock(&fs_info->balance_mutex);
4450 if (fs_info->balance_ctl)
4451 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4452 mutex_unlock(&fs_info->balance_mutex);
4453 sb_end_write(fs_info->sb);
4458 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4460 struct task_struct *tsk;
4462 mutex_lock(&fs_info->balance_mutex);
4463 if (!fs_info->balance_ctl) {
4464 mutex_unlock(&fs_info->balance_mutex);
4467 mutex_unlock(&fs_info->balance_mutex);
4469 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4470 btrfs_info(fs_info, "balance: resume skipped");
4474 spin_lock(&fs_info->super_lock);
4475 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
4476 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;
4477 spin_unlock(&fs_info->super_lock);
4479 * A ro->rw remount sequence should continue with the paused balance
4480 * regardless of who pauses it, system or the user as of now, so set
4483 spin_lock(&fs_info->balance_lock);
4484 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4485 spin_unlock(&fs_info->balance_lock);
4487 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4488 return PTR_ERR_OR_ZERO(tsk);
4491 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4493 struct btrfs_balance_control *bctl;
4494 struct btrfs_balance_item *item;
4495 struct btrfs_disk_balance_args disk_bargs;
4496 struct btrfs_path *path;
4497 struct extent_buffer *leaf;
4498 struct btrfs_key key;
4501 path = btrfs_alloc_path();
4505 key.objectid = BTRFS_BALANCE_OBJECTID;
4506 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4509 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4512 if (ret > 0) { /* ret = -ENOENT; */
4517 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4523 leaf = path->nodes[0];
4524 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4526 bctl->flags = btrfs_balance_flags(leaf, item);
4527 bctl->flags |= BTRFS_BALANCE_RESUME;
4529 btrfs_balance_data(leaf, item, &disk_bargs);
4530 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4531 btrfs_balance_meta(leaf, item, &disk_bargs);
4532 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4533 btrfs_balance_sys(leaf, item, &disk_bargs);
4534 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4537 * This should never happen, as the paused balance state is recovered
4538 * during mount without any chance of other exclusive ops to collide.
4540 * This gives the exclusive op status to balance and keeps in paused
4541 * state until user intervention (cancel or umount). If the ownership
4542 * cannot be assigned, show a message but do not fail. The balance
4543 * is in a paused state and must have fs_info::balance_ctl properly
4546 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED))
4548 "balance: cannot set exclusive op status, resume manually");
4550 btrfs_release_path(path);
4552 mutex_lock(&fs_info->balance_mutex);
4553 BUG_ON(fs_info->balance_ctl);
4554 spin_lock(&fs_info->balance_lock);
4555 fs_info->balance_ctl = bctl;
4556 spin_unlock(&fs_info->balance_lock);
4557 mutex_unlock(&fs_info->balance_mutex);
4559 btrfs_free_path(path);
4563 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4567 mutex_lock(&fs_info->balance_mutex);
4568 if (!fs_info->balance_ctl) {
4569 mutex_unlock(&fs_info->balance_mutex);
4573 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4574 atomic_inc(&fs_info->balance_pause_req);
4575 mutex_unlock(&fs_info->balance_mutex);
4577 wait_event(fs_info->balance_wait_q,
4578 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4580 mutex_lock(&fs_info->balance_mutex);
4581 /* we are good with balance_ctl ripped off from under us */
4582 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4583 atomic_dec(&fs_info->balance_pause_req);
4588 mutex_unlock(&fs_info->balance_mutex);
4592 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4594 mutex_lock(&fs_info->balance_mutex);
4595 if (!fs_info->balance_ctl) {
4596 mutex_unlock(&fs_info->balance_mutex);
4601 * A paused balance with the item stored on disk can be resumed at
4602 * mount time if the mount is read-write. Otherwise it's still paused
4603 * and we must not allow cancelling as it deletes the item.
4605 if (sb_rdonly(fs_info->sb)) {
4606 mutex_unlock(&fs_info->balance_mutex);
4610 atomic_inc(&fs_info->balance_cancel_req);
4612 * if we are running just wait and return, balance item is
4613 * deleted in btrfs_balance in this case
4615 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4616 mutex_unlock(&fs_info->balance_mutex);
4617 wait_event(fs_info->balance_wait_q,
4618 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4619 mutex_lock(&fs_info->balance_mutex);
4621 mutex_unlock(&fs_info->balance_mutex);
4623 * Lock released to allow other waiters to continue, we'll
4624 * reexamine the status again.
4626 mutex_lock(&fs_info->balance_mutex);
4628 if (fs_info->balance_ctl) {
4629 reset_balance_state(fs_info);
4630 btrfs_exclop_finish(fs_info);
4631 btrfs_info(fs_info, "balance: canceled");
4635 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4636 atomic_dec(&fs_info->balance_cancel_req);
4637 mutex_unlock(&fs_info->balance_mutex);
4641 int btrfs_uuid_scan_kthread(void *data)
4643 struct btrfs_fs_info *fs_info = data;
4644 struct btrfs_root *root = fs_info->tree_root;
4645 struct btrfs_key key;
4646 struct btrfs_path *path = NULL;
4648 struct extent_buffer *eb;
4650 struct btrfs_root_item root_item;
4652 struct btrfs_trans_handle *trans = NULL;
4653 bool closing = false;
4655 path = btrfs_alloc_path();
4662 key.type = BTRFS_ROOT_ITEM_KEY;
4666 if (btrfs_fs_closing(fs_info)) {
4670 ret = btrfs_search_forward(root, &key, path,
4671 BTRFS_OLDEST_GENERATION);
4678 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4679 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4680 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4681 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4684 eb = path->nodes[0];
4685 slot = path->slots[0];
4686 item_size = btrfs_item_size(eb, slot);
4687 if (item_size < sizeof(root_item))
4690 read_extent_buffer(eb, &root_item,
4691 btrfs_item_ptr_offset(eb, slot),
4692 (int)sizeof(root_item));
4693 if (btrfs_root_refs(&root_item) == 0)
4696 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4697 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4701 btrfs_release_path(path);
4703 * 1 - subvol uuid item
4704 * 1 - received_subvol uuid item
4706 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4707 if (IS_ERR(trans)) {
4708 ret = PTR_ERR(trans);
4716 btrfs_release_path(path);
4717 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4718 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4719 BTRFS_UUID_KEY_SUBVOL,
4722 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4728 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4729 ret = btrfs_uuid_tree_add(trans,
4730 root_item.received_uuid,
4731 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4734 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4741 btrfs_release_path(path);
4743 ret = btrfs_end_transaction(trans);
4749 if (key.offset < (u64)-1) {
4751 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4753 key.type = BTRFS_ROOT_ITEM_KEY;
4754 } else if (key.objectid < (u64)-1) {
4756 key.type = BTRFS_ROOT_ITEM_KEY;
4765 btrfs_free_path(path);
4766 if (trans && !IS_ERR(trans))
4767 btrfs_end_transaction(trans);
4769 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4771 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4772 up(&fs_info->uuid_tree_rescan_sem);
4776 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4778 struct btrfs_trans_handle *trans;
4779 struct btrfs_root *tree_root = fs_info->tree_root;
4780 struct btrfs_root *uuid_root;
4781 struct task_struct *task;
4788 trans = btrfs_start_transaction(tree_root, 2);
4790 return PTR_ERR(trans);
4792 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4793 if (IS_ERR(uuid_root)) {
4794 ret = PTR_ERR(uuid_root);
4795 btrfs_abort_transaction(trans, ret);
4796 btrfs_end_transaction(trans);
4800 fs_info->uuid_root = uuid_root;
4802 ret = btrfs_commit_transaction(trans);
4806 down(&fs_info->uuid_tree_rescan_sem);
4807 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4809 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4810 btrfs_warn(fs_info, "failed to start uuid_scan task");
4811 up(&fs_info->uuid_tree_rescan_sem);
4812 return PTR_ERR(task);
4819 * shrinking a device means finding all of the device extents past
4820 * the new size, and then following the back refs to the chunks.
4821 * The chunk relocation code actually frees the device extent
4823 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4825 struct btrfs_fs_info *fs_info = device->fs_info;
4826 struct btrfs_root *root = fs_info->dev_root;
4827 struct btrfs_trans_handle *trans;
4828 struct btrfs_dev_extent *dev_extent = NULL;
4829 struct btrfs_path *path;
4835 bool retried = false;
4836 struct extent_buffer *l;
4837 struct btrfs_key key;
4838 struct btrfs_super_block *super_copy = fs_info->super_copy;
4839 u64 old_total = btrfs_super_total_bytes(super_copy);
4840 u64 old_size = btrfs_device_get_total_bytes(device);
4844 new_size = round_down(new_size, fs_info->sectorsize);
4846 diff = round_down(old_size - new_size, fs_info->sectorsize);
4848 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4851 path = btrfs_alloc_path();
4855 path->reada = READA_BACK;
4857 trans = btrfs_start_transaction(root, 0);
4858 if (IS_ERR(trans)) {
4859 btrfs_free_path(path);
4860 return PTR_ERR(trans);
4863 mutex_lock(&fs_info->chunk_mutex);
4865 btrfs_device_set_total_bytes(device, new_size);
4866 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4867 device->fs_devices->total_rw_bytes -= diff;
4868 atomic64_sub(diff, &fs_info->free_chunk_space);
4872 * Once the device's size has been set to the new size, ensure all
4873 * in-memory chunks are synced to disk so that the loop below sees them
4874 * and relocates them accordingly.
4876 if (contains_pending_extent(device, &start, diff)) {
4877 mutex_unlock(&fs_info->chunk_mutex);
4878 ret = btrfs_commit_transaction(trans);
4882 mutex_unlock(&fs_info->chunk_mutex);
4883 btrfs_end_transaction(trans);
4887 key.objectid = device->devid;
4888 key.offset = (u64)-1;
4889 key.type = BTRFS_DEV_EXTENT_KEY;
4892 mutex_lock(&fs_info->reclaim_bgs_lock);
4893 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4895 mutex_unlock(&fs_info->reclaim_bgs_lock);
4899 ret = btrfs_previous_item(root, path, 0, key.type);
4901 mutex_unlock(&fs_info->reclaim_bgs_lock);
4905 btrfs_release_path(path);
4910 slot = path->slots[0];
4911 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4913 if (key.objectid != device->devid) {
4914 mutex_unlock(&fs_info->reclaim_bgs_lock);
4915 btrfs_release_path(path);
4919 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4920 length = btrfs_dev_extent_length(l, dev_extent);
4922 if (key.offset + length <= new_size) {
4923 mutex_unlock(&fs_info->reclaim_bgs_lock);
4924 btrfs_release_path(path);
4928 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4929 btrfs_release_path(path);
4932 * We may be relocating the only data chunk we have,
4933 * which could potentially end up with losing data's
4934 * raid profile, so lets allocate an empty one in
4937 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4939 mutex_unlock(&fs_info->reclaim_bgs_lock);
4943 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4944 mutex_unlock(&fs_info->reclaim_bgs_lock);
4945 if (ret == -ENOSPC) {
4948 if (ret == -ETXTBSY) {
4950 "could not shrink block group %llu due to active swapfile",
4955 } while (key.offset-- > 0);
4957 if (failed && !retried) {
4961 } else if (failed && retried) {
4966 /* Shrinking succeeded, else we would be at "done". */
4967 trans = btrfs_start_transaction(root, 0);
4968 if (IS_ERR(trans)) {
4969 ret = PTR_ERR(trans);
4973 mutex_lock(&fs_info->chunk_mutex);
4974 /* Clear all state bits beyond the shrunk device size */
4975 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
4978 btrfs_device_set_disk_total_bytes(device, new_size);
4979 if (list_empty(&device->post_commit_list))
4980 list_add_tail(&device->post_commit_list,
4981 &trans->transaction->dev_update_list);
4983 WARN_ON(diff > old_total);
4984 btrfs_set_super_total_bytes(super_copy,
4985 round_down(old_total - diff, fs_info->sectorsize));
4986 mutex_unlock(&fs_info->chunk_mutex);
4988 btrfs_reserve_chunk_metadata(trans, false);
4989 /* Now btrfs_update_device() will change the on-disk size. */
4990 ret = btrfs_update_device(trans, device);
4991 btrfs_trans_release_chunk_metadata(trans);
4993 btrfs_abort_transaction(trans, ret);
4994 btrfs_end_transaction(trans);
4996 ret = btrfs_commit_transaction(trans);
4999 btrfs_free_path(path);
5001 mutex_lock(&fs_info->chunk_mutex);
5002 btrfs_device_set_total_bytes(device, old_size);
5003 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
5004 device->fs_devices->total_rw_bytes += diff;
5005 atomic64_add(diff, &fs_info->free_chunk_space);
5006 mutex_unlock(&fs_info->chunk_mutex);
5011 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
5012 struct btrfs_key *key,
5013 struct btrfs_chunk *chunk, int item_size)
5015 struct btrfs_super_block *super_copy = fs_info->super_copy;
5016 struct btrfs_disk_key disk_key;
5020 lockdep_assert_held(&fs_info->chunk_mutex);
5022 array_size = btrfs_super_sys_array_size(super_copy);
5023 if (array_size + item_size + sizeof(disk_key)
5024 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
5027 ptr = super_copy->sys_chunk_array + array_size;
5028 btrfs_cpu_key_to_disk(&disk_key, key);
5029 memcpy(ptr, &disk_key, sizeof(disk_key));
5030 ptr += sizeof(disk_key);
5031 memcpy(ptr, chunk, item_size);
5032 item_size += sizeof(disk_key);
5033 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
5039 * sort the devices in descending order by max_avail, total_avail
5041 static int btrfs_cmp_device_info(const void *a, const void *b)
5043 const struct btrfs_device_info *di_a = a;
5044 const struct btrfs_device_info *di_b = b;
5046 if (di_a->max_avail > di_b->max_avail)
5048 if (di_a->max_avail < di_b->max_avail)
5050 if (di_a->total_avail > di_b->total_avail)
5052 if (di_a->total_avail < di_b->total_avail)
5057 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
5059 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5062 btrfs_set_fs_incompat(info, RAID56);
5065 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
5067 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
5070 btrfs_set_fs_incompat(info, RAID1C34);
5074 * Structure used internally for btrfs_create_chunk() function.
5075 * Wraps needed parameters.
5077 struct alloc_chunk_ctl {
5080 /* Total number of stripes to allocate */
5082 /* sub_stripes info for map */
5084 /* Stripes per device */
5086 /* Maximum number of devices to use */
5088 /* Minimum number of devices to use */
5090 /* ndevs has to be a multiple of this */
5092 /* Number of copies */
5094 /* Number of stripes worth of bytes to store parity information */
5096 u64 max_stripe_size;
5104 static void init_alloc_chunk_ctl_policy_regular(
5105 struct btrfs_fs_devices *fs_devices,
5106 struct alloc_chunk_ctl *ctl)
5108 struct btrfs_space_info *space_info;
5110 space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type);
5113 ctl->max_chunk_size = READ_ONCE(space_info->chunk_size);
5114 ctl->max_stripe_size = ctl->max_chunk_size;
5116 if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM)
5117 ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK);
5119 /* We don't want a chunk larger than 10% of writable space */
5120 ctl->max_chunk_size = min(mult_perc(fs_devices->total_rw_bytes, 10),
5121 ctl->max_chunk_size);
5122 ctl->dev_extent_min = btrfs_stripe_nr_to_offset(ctl->dev_stripes);
5125 static void init_alloc_chunk_ctl_policy_zoned(
5126 struct btrfs_fs_devices *fs_devices,
5127 struct alloc_chunk_ctl *ctl)
5129 u64 zone_size = fs_devices->fs_info->zone_size;
5131 int min_num_stripes = ctl->devs_min * ctl->dev_stripes;
5132 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;
5133 u64 min_chunk_size = min_data_stripes * zone_size;
5134 u64 type = ctl->type;
5136 ctl->max_stripe_size = zone_size;
5137 if (type & BTRFS_BLOCK_GROUP_DATA) {
5138 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,
5140 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5141 ctl->max_chunk_size = ctl->max_stripe_size;
5142 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5143 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5144 ctl->devs_max = min_t(int, ctl->devs_max,
5145 BTRFS_MAX_DEVS_SYS_CHUNK);
5150 /* We don't want a chunk larger than 10% of writable space */
5151 limit = max(round_down(mult_perc(fs_devices->total_rw_bytes, 10),
5154 ctl->max_chunk_size = min(limit, ctl->max_chunk_size);
5155 ctl->dev_extent_min = zone_size * ctl->dev_stripes;
5158 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
5159 struct alloc_chunk_ctl *ctl)
5161 int index = btrfs_bg_flags_to_raid_index(ctl->type);
5163 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
5164 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
5165 ctl->devs_max = btrfs_raid_array[index].devs_max;
5167 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
5168 ctl->devs_min = btrfs_raid_array[index].devs_min;
5169 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
5170 ctl->ncopies = btrfs_raid_array[index].ncopies;
5171 ctl->nparity = btrfs_raid_array[index].nparity;
5174 switch (fs_devices->chunk_alloc_policy) {
5175 case BTRFS_CHUNK_ALLOC_REGULAR:
5176 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
5178 case BTRFS_CHUNK_ALLOC_ZONED:
5179 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);
5186 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
5187 struct alloc_chunk_ctl *ctl,
5188 struct btrfs_device_info *devices_info)
5190 struct btrfs_fs_info *info = fs_devices->fs_info;
5191 struct btrfs_device *device;
5193 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
5200 * in the first pass through the devices list, we gather information
5201 * about the available holes on each device.
5203 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5204 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5206 "BTRFS: read-only device in alloc_list\n");
5210 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5211 &device->dev_state) ||
5212 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5215 if (device->total_bytes > device->bytes_used)
5216 total_avail = device->total_bytes - device->bytes_used;
5220 /* If there is no space on this device, skip it. */
5221 if (total_avail < ctl->dev_extent_min)
5224 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
5226 if (ret && ret != -ENOSPC)
5230 max_avail = dev_extent_want;
5232 if (max_avail < ctl->dev_extent_min) {
5233 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5235 "%s: devid %llu has no free space, have=%llu want=%llu",
5236 __func__, device->devid, max_avail,
5237 ctl->dev_extent_min);
5241 if (ndevs == fs_devices->rw_devices) {
5242 WARN(1, "%s: found more than %llu devices\n",
5243 __func__, fs_devices->rw_devices);
5246 devices_info[ndevs].dev_offset = dev_offset;
5247 devices_info[ndevs].max_avail = max_avail;
5248 devices_info[ndevs].total_avail = total_avail;
5249 devices_info[ndevs].dev = device;
5255 * now sort the devices by hole size / available space
5257 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5258 btrfs_cmp_device_info, NULL);
5263 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5264 struct btrfs_device_info *devices_info)
5266 /* Number of stripes that count for block group size */
5270 * The primary goal is to maximize the number of stripes, so use as
5271 * many devices as possible, even if the stripes are not maximum sized.
5273 * The DUP profile stores more than one stripe per device, the
5274 * max_avail is the total size so we have to adjust.
5276 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5278 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5280 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5281 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5284 * Use the number of data stripes to figure out how big this chunk is
5285 * really going to be in terms of logical address space, and compare
5286 * that answer with the max chunk size. If it's higher, we try to
5287 * reduce stripe_size.
5289 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5291 * Reduce stripe_size, round it up to a 16MB boundary again and
5292 * then use it, unless it ends up being even bigger than the
5293 * previous value we had already.
5295 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5296 data_stripes), SZ_16M),
5300 /* Stripe size should not go beyond 1G. */
5301 ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G);
5303 /* Align to BTRFS_STRIPE_LEN */
5304 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5305 ctl->chunk_size = ctl->stripe_size * data_stripes;
5310 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,
5311 struct btrfs_device_info *devices_info)
5313 u64 zone_size = devices_info[0].dev->zone_info->zone_size;
5314 /* Number of stripes that count for block group size */
5318 * It should hold because:
5319 * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5321 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);
5323 ctl->stripe_size = zone_size;
5324 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5325 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5327 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5328 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5329 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,
5330 ctl->stripe_size) + ctl->nparity,
5332 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5333 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5334 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);
5337 ctl->chunk_size = ctl->stripe_size * data_stripes;
5342 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5343 struct alloc_chunk_ctl *ctl,
5344 struct btrfs_device_info *devices_info)
5346 struct btrfs_fs_info *info = fs_devices->fs_info;
5349 * Round down to number of usable stripes, devs_increment can be any
5350 * number so we can't use round_down() that requires power of 2, while
5351 * rounddown is safe.
5353 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5355 if (ctl->ndevs < ctl->devs_min) {
5356 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5358 "%s: not enough devices with free space: have=%d minimum required=%d",
5359 __func__, ctl->ndevs, ctl->devs_min);
5364 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5366 switch (fs_devices->chunk_alloc_policy) {
5367 case BTRFS_CHUNK_ALLOC_REGULAR:
5368 return decide_stripe_size_regular(ctl, devices_info);
5369 case BTRFS_CHUNK_ALLOC_ZONED:
5370 return decide_stripe_size_zoned(ctl, devices_info);
5376 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
5377 struct alloc_chunk_ctl *ctl,
5378 struct btrfs_device_info *devices_info)
5380 struct btrfs_fs_info *info = trans->fs_info;
5381 struct map_lookup *map = NULL;
5382 struct extent_map_tree *em_tree;
5383 struct btrfs_block_group *block_group;
5384 struct extent_map *em;
5385 u64 start = ctl->start;
5386 u64 type = ctl->type;
5391 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
5393 return ERR_PTR(-ENOMEM);
5394 map->num_stripes = ctl->num_stripes;
5396 for (i = 0; i < ctl->ndevs; ++i) {
5397 for (j = 0; j < ctl->dev_stripes; ++j) {
5398 int s = i * ctl->dev_stripes + j;
5399 map->stripes[s].dev = devices_info[i].dev;
5400 map->stripes[s].physical = devices_info[i].dev_offset +
5401 j * ctl->stripe_size;
5404 map->io_align = BTRFS_STRIPE_LEN;
5405 map->io_width = BTRFS_STRIPE_LEN;
5407 map->sub_stripes = ctl->sub_stripes;
5409 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5411 em = alloc_extent_map();
5414 return ERR_PTR(-ENOMEM);
5416 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5417 em->map_lookup = map;
5419 em->len = ctl->chunk_size;
5420 em->block_start = 0;
5421 em->block_len = em->len;
5422 em->orig_block_len = ctl->stripe_size;
5424 em_tree = &info->mapping_tree;
5425 write_lock(&em_tree->lock);
5426 ret = add_extent_mapping(em_tree, em, 0);
5428 write_unlock(&em_tree->lock);
5429 free_extent_map(em);
5430 return ERR_PTR(ret);
5432 write_unlock(&em_tree->lock);
5434 block_group = btrfs_make_block_group(trans, type, start, ctl->chunk_size);
5435 if (IS_ERR(block_group))
5436 goto error_del_extent;
5438 for (i = 0; i < map->num_stripes; i++) {
5439 struct btrfs_device *dev = map->stripes[i].dev;
5441 btrfs_device_set_bytes_used(dev,
5442 dev->bytes_used + ctl->stripe_size);
5443 if (list_empty(&dev->post_commit_list))
5444 list_add_tail(&dev->post_commit_list,
5445 &trans->transaction->dev_update_list);
5448 atomic64_sub(ctl->stripe_size * map->num_stripes,
5449 &info->free_chunk_space);
5451 free_extent_map(em);
5452 check_raid56_incompat_flag(info, type);
5453 check_raid1c34_incompat_flag(info, type);
5458 write_lock(&em_tree->lock);
5459 remove_extent_mapping(em_tree, em);
5460 write_unlock(&em_tree->lock);
5462 /* One for our allocation */
5463 free_extent_map(em);
5464 /* One for the tree reference */
5465 free_extent_map(em);
5470 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
5473 struct btrfs_fs_info *info = trans->fs_info;
5474 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5475 struct btrfs_device_info *devices_info = NULL;
5476 struct alloc_chunk_ctl ctl;
5477 struct btrfs_block_group *block_group;
5480 lockdep_assert_held(&info->chunk_mutex);
5482 if (!alloc_profile_is_valid(type, 0)) {
5484 return ERR_PTR(-EINVAL);
5487 if (list_empty(&fs_devices->alloc_list)) {
5488 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5489 btrfs_debug(info, "%s: no writable device", __func__);
5490 return ERR_PTR(-ENOSPC);
5493 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5494 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5496 return ERR_PTR(-EINVAL);
5499 ctl.start = find_next_chunk(info);
5501 init_alloc_chunk_ctl(fs_devices, &ctl);
5503 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5506 return ERR_PTR(-ENOMEM);
5508 ret = gather_device_info(fs_devices, &ctl, devices_info);
5510 block_group = ERR_PTR(ret);
5514 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5516 block_group = ERR_PTR(ret);
5520 block_group = create_chunk(trans, &ctl, devices_info);
5523 kfree(devices_info);
5528 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5529 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5532 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5535 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
5536 struct btrfs_block_group *bg)
5538 struct btrfs_fs_info *fs_info = trans->fs_info;
5539 struct btrfs_root *chunk_root = fs_info->chunk_root;
5540 struct btrfs_key key;
5541 struct btrfs_chunk *chunk;
5542 struct btrfs_stripe *stripe;
5543 struct extent_map *em;
5544 struct map_lookup *map;
5550 * We take the chunk_mutex for 2 reasons:
5552 * 1) Updates and insertions in the chunk btree must be done while holding
5553 * the chunk_mutex, as well as updating the system chunk array in the
5554 * superblock. See the comment on top of btrfs_chunk_alloc() for the
5557 * 2) To prevent races with the final phase of a device replace operation
5558 * that replaces the device object associated with the map's stripes,
5559 * because the device object's id can change at any time during that
5560 * final phase of the device replace operation
5561 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5562 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5563 * which would cause a failure when updating the device item, which does
5564 * not exists, or persisting a stripe of the chunk item with such ID.
5565 * Here we can't use the device_list_mutex because our caller already
5566 * has locked the chunk_mutex, and the final phase of device replace
5567 * acquires both mutexes - first the device_list_mutex and then the
5568 * chunk_mutex. Using any of those two mutexes protects us from a
5569 * concurrent device replace.
5571 lockdep_assert_held(&fs_info->chunk_mutex);
5573 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
5576 btrfs_abort_transaction(trans, ret);
5580 map = em->map_lookup;
5581 item_size = btrfs_chunk_item_size(map->num_stripes);
5583 chunk = kzalloc(item_size, GFP_NOFS);
5586 btrfs_abort_transaction(trans, ret);
5590 for (i = 0; i < map->num_stripes; i++) {
5591 struct btrfs_device *device = map->stripes[i].dev;
5593 ret = btrfs_update_device(trans, device);
5598 stripe = &chunk->stripe;
5599 for (i = 0; i < map->num_stripes; i++) {
5600 struct btrfs_device *device = map->stripes[i].dev;
5601 const u64 dev_offset = map->stripes[i].physical;
5603 btrfs_set_stack_stripe_devid(stripe, device->devid);
5604 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5605 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5609 btrfs_set_stack_chunk_length(chunk, bg->length);
5610 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
5611 btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
5612 btrfs_set_stack_chunk_type(chunk, map->type);
5613 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5614 btrfs_set_stack_chunk_io_align(chunk, BTRFS_STRIPE_LEN);
5615 btrfs_set_stack_chunk_io_width(chunk, BTRFS_STRIPE_LEN);
5616 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5617 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5619 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5620 key.type = BTRFS_CHUNK_ITEM_KEY;
5621 key.offset = bg->start;
5623 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5627 set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags);
5629 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5630 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5637 free_extent_map(em);
5641 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5643 struct btrfs_fs_info *fs_info = trans->fs_info;
5645 struct btrfs_block_group *meta_bg;
5646 struct btrfs_block_group *sys_bg;
5649 * When adding a new device for sprouting, the seed device is read-only
5650 * so we must first allocate a metadata and a system chunk. But before
5651 * adding the block group items to the extent, device and chunk btrees,
5654 * 1) Create both chunks without doing any changes to the btrees, as
5655 * otherwise we would get -ENOSPC since the block groups from the
5656 * seed device are read-only;
5658 * 2) Add the device item for the new sprout device - finishing the setup
5659 * of a new block group requires updating the device item in the chunk
5660 * btree, so it must exist when we attempt to do it. The previous step
5661 * ensures this does not fail with -ENOSPC.
5663 * After that we can add the block group items to their btrees:
5664 * update existing device item in the chunk btree, add a new block group
5665 * item to the extent btree, add a new chunk item to the chunk btree and
5666 * finally add the new device extent items to the devices btree.
5669 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5670 meta_bg = btrfs_create_chunk(trans, alloc_profile);
5671 if (IS_ERR(meta_bg))
5672 return PTR_ERR(meta_bg);
5674 alloc_profile = btrfs_system_alloc_profile(fs_info);
5675 sys_bg = btrfs_create_chunk(trans, alloc_profile);
5677 return PTR_ERR(sys_bg);
5682 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5684 const int index = btrfs_bg_flags_to_raid_index(map->type);
5686 return btrfs_raid_array[index].tolerated_failures;
5689 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5691 struct extent_map *em;
5692 struct map_lookup *map;
5697 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5701 map = em->map_lookup;
5702 for (i = 0; i < map->num_stripes; i++) {
5703 if (test_bit(BTRFS_DEV_STATE_MISSING,
5704 &map->stripes[i].dev->dev_state)) {
5708 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5709 &map->stripes[i].dev->dev_state)) {
5716 * If the number of missing devices is larger than max errors, we can
5717 * not write the data into that chunk successfully.
5719 if (miss_ndevs > btrfs_chunk_max_errors(map))
5722 free_extent_map(em);
5726 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5728 struct extent_map *em;
5731 write_lock(&tree->lock);
5732 em = lookup_extent_mapping(tree, 0, (u64)-1);
5734 remove_extent_mapping(tree, em);
5735 write_unlock(&tree->lock);
5739 free_extent_map(em);
5740 /* once for the tree */
5741 free_extent_map(em);
5745 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5747 struct extent_map *em;
5748 struct map_lookup *map;
5749 enum btrfs_raid_types index;
5752 em = btrfs_get_chunk_map(fs_info, logical, len);
5755 * We could return errors for these cases, but that could get
5756 * ugly and we'd probably do the same thing which is just not do
5757 * anything else and exit, so return 1 so the callers don't try
5758 * to use other copies.
5762 map = em->map_lookup;
5763 index = btrfs_bg_flags_to_raid_index(map->type);
5765 /* Non-RAID56, use their ncopies from btrfs_raid_array. */
5766 if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5767 ret = btrfs_raid_array[index].ncopies;
5768 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5770 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5772 * There could be two corrupted data stripes, we need
5773 * to loop retry in order to rebuild the correct data.
5775 * Fail a stripe at a time on every retry except the
5776 * stripe under reconstruction.
5778 ret = map->num_stripes;
5779 free_extent_map(em);
5783 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5786 struct extent_map *em;
5787 struct map_lookup *map;
5788 unsigned long len = fs_info->sectorsize;
5790 if (!btrfs_fs_incompat(fs_info, RAID56))
5793 em = btrfs_get_chunk_map(fs_info, logical, len);
5795 if (!WARN_ON(IS_ERR(em))) {
5796 map = em->map_lookup;
5797 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5798 len = btrfs_stripe_nr_to_offset(nr_data_stripes(map));
5799 free_extent_map(em);
5804 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5806 struct extent_map *em;
5807 struct map_lookup *map;
5810 if (!btrfs_fs_incompat(fs_info, RAID56))
5813 em = btrfs_get_chunk_map(fs_info, logical, len);
5815 if(!WARN_ON(IS_ERR(em))) {
5816 map = em->map_lookup;
5817 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5819 free_extent_map(em);
5824 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5825 struct map_lookup *map, int first,
5826 int dev_replace_is_ongoing)
5830 int preferred_mirror;
5832 struct btrfs_device *srcdev;
5835 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5837 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5838 num_stripes = map->sub_stripes;
5840 num_stripes = map->num_stripes;
5842 switch (fs_info->fs_devices->read_policy) {
5844 /* Shouldn't happen, just warn and use pid instead of failing */
5845 btrfs_warn_rl(fs_info,
5846 "unknown read_policy type %u, reset to pid",
5847 fs_info->fs_devices->read_policy);
5848 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID;
5850 case BTRFS_READ_POLICY_PID:
5851 preferred_mirror = first + (current->pid % num_stripes);
5855 if (dev_replace_is_ongoing &&
5856 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5857 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5858 srcdev = fs_info->dev_replace.srcdev;
5863 * try to avoid the drive that is the source drive for a
5864 * dev-replace procedure, only choose it if no other non-missing
5865 * mirror is available
5867 for (tolerance = 0; tolerance < 2; tolerance++) {
5868 if (map->stripes[preferred_mirror].dev->bdev &&
5869 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5870 return preferred_mirror;
5871 for (i = first; i < first + num_stripes; i++) {
5872 if (map->stripes[i].dev->bdev &&
5873 (tolerance || map->stripes[i].dev != srcdev))
5878 /* we couldn't find one that doesn't fail. Just return something
5879 * and the io error handling code will clean up eventually
5881 return preferred_mirror;
5884 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info,
5887 struct btrfs_io_context *bioc;
5890 /* The size of btrfs_io_context */
5891 sizeof(struct btrfs_io_context) +
5892 /* Plus the variable array for the stripes */
5893 sizeof(struct btrfs_io_stripe) * (total_stripes),
5899 refcount_set(&bioc->refs, 1);
5901 bioc->fs_info = fs_info;
5902 bioc->replace_stripe_src = -1;
5903 bioc->full_stripe_logical = (u64)-1;
5908 void btrfs_get_bioc(struct btrfs_io_context *bioc)
5910 WARN_ON(!refcount_read(&bioc->refs));
5911 refcount_inc(&bioc->refs);
5914 void btrfs_put_bioc(struct btrfs_io_context *bioc)
5918 if (refcount_dec_and_test(&bioc->refs))
5923 * Please note that, discard won't be sent to target device of device
5926 struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info,
5927 u64 logical, u64 *length_ret,
5930 struct extent_map *em;
5931 struct map_lookup *map;
5932 struct btrfs_discard_stripe *stripes;
5933 u64 length = *length_ret;
5938 u64 stripe_end_offset;
5942 u32 sub_stripes = 0;
5943 u32 stripes_per_dev = 0;
5944 u32 remaining_stripes = 0;
5945 u32 last_stripe = 0;
5949 em = btrfs_get_chunk_map(fs_info, logical, length);
5951 return ERR_CAST(em);
5953 map = em->map_lookup;
5955 /* we don't discard raid56 yet */
5956 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5961 offset = logical - em->start;
5962 length = min_t(u64, em->start + em->len - logical, length);
5963 *length_ret = length;
5966 * stripe_nr counts the total number of stripes we have to stride
5967 * to get to this block
5969 stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
5971 /* stripe_offset is the offset of this block in its stripe */
5972 stripe_offset = offset - btrfs_stripe_nr_to_offset(stripe_nr);
5974 stripe_nr_end = round_up(offset + length, BTRFS_STRIPE_LEN) >>
5975 BTRFS_STRIPE_LEN_SHIFT;
5976 stripe_cnt = stripe_nr_end - stripe_nr;
5977 stripe_end_offset = btrfs_stripe_nr_to_offset(stripe_nr_end) -
5980 * after this, stripe_nr is the number of stripes on this
5981 * device we have to walk to find the data, and stripe_index is
5982 * the number of our device in the stripe array
5986 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5987 BTRFS_BLOCK_GROUP_RAID10)) {
5988 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5991 sub_stripes = map->sub_stripes;
5993 factor = map->num_stripes / sub_stripes;
5994 *num_stripes = min_t(u64, map->num_stripes,
5995 sub_stripes * stripe_cnt);
5996 stripe_index = stripe_nr % factor;
5997 stripe_nr /= factor;
5998 stripe_index *= sub_stripes;
6000 remaining_stripes = stripe_cnt % factor;
6001 stripes_per_dev = stripe_cnt / factor;
6002 last_stripe = ((stripe_nr_end - 1) % factor) * sub_stripes;
6003 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
6004 BTRFS_BLOCK_GROUP_DUP)) {
6005 *num_stripes = map->num_stripes;
6007 stripe_index = stripe_nr % map->num_stripes;
6008 stripe_nr /= map->num_stripes;
6011 stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS);
6017 for (i = 0; i < *num_stripes; i++) {
6018 stripes[i].physical =
6019 map->stripes[stripe_index].physical +
6020 stripe_offset + btrfs_stripe_nr_to_offset(stripe_nr);
6021 stripes[i].dev = map->stripes[stripe_index].dev;
6023 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6024 BTRFS_BLOCK_GROUP_RAID10)) {
6025 stripes[i].length = btrfs_stripe_nr_to_offset(stripes_per_dev);
6027 if (i / sub_stripes < remaining_stripes)
6028 stripes[i].length += BTRFS_STRIPE_LEN;
6031 * Special for the first stripe and
6034 * |-------|...|-------|
6038 if (i < sub_stripes)
6039 stripes[i].length -= stripe_offset;
6041 if (stripe_index >= last_stripe &&
6042 stripe_index <= (last_stripe +
6044 stripes[i].length -= stripe_end_offset;
6046 if (i == sub_stripes - 1)
6049 stripes[i].length = length;
6053 if (stripe_index == map->num_stripes) {
6059 free_extent_map(em);
6062 free_extent_map(em);
6063 return ERR_PTR(ret);
6066 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)
6068 struct btrfs_block_group *cache;
6071 /* Non zoned filesystem does not use "to_copy" flag */
6072 if (!btrfs_is_zoned(fs_info))
6075 cache = btrfs_lookup_block_group(fs_info, logical);
6077 ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags);
6079 btrfs_put_block_group(cache);
6083 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
6084 struct btrfs_io_context *bioc,
6085 struct btrfs_dev_replace *dev_replace,
6087 int *num_stripes_ret, int *max_errors_ret)
6089 u64 srcdev_devid = dev_replace->srcdev->devid;
6091 * At this stage, num_stripes is still the real number of stripes,
6092 * excluding the duplicated stripes.
6094 int num_stripes = *num_stripes_ret;
6095 int nr_extra_stripes = 0;
6096 int max_errors = *max_errors_ret;
6100 * A block group which has "to_copy" set will eventually be copied by
6101 * the dev-replace process. We can avoid cloning IO here.
6103 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))
6107 * Duplicate the write operations while the dev-replace procedure is
6108 * running. Since the copying of the old disk to the new disk takes
6109 * place at run time while the filesystem is mounted writable, the
6110 * regular write operations to the old disk have to be duplicated to go
6111 * to the new disk as well.
6113 * Note that device->missing is handled by the caller, and that the
6114 * write to the old disk is already set up in the stripes array.
6116 for (i = 0; i < num_stripes; i++) {
6117 struct btrfs_io_stripe *old = &bioc->stripes[i];
6118 struct btrfs_io_stripe *new = &bioc->stripes[num_stripes + nr_extra_stripes];
6120 if (old->dev->devid != srcdev_devid)
6123 new->physical = old->physical;
6124 new->dev = dev_replace->tgtdev;
6125 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK)
6126 bioc->replace_stripe_src = i;
6130 /* We can only have at most 2 extra nr_stripes (for DUP). */
6131 ASSERT(nr_extra_stripes <= 2);
6133 * For GET_READ_MIRRORS, we can only return at most 1 extra stripe for
6135 * If we have 2 extra stripes, only choose the one with smaller physical.
6137 if (op == BTRFS_MAP_GET_READ_MIRRORS && nr_extra_stripes == 2) {
6138 struct btrfs_io_stripe *first = &bioc->stripes[num_stripes];
6139 struct btrfs_io_stripe *second = &bioc->stripes[num_stripes + 1];
6141 /* Only DUP can have two extra stripes. */
6142 ASSERT(bioc->map_type & BTRFS_BLOCK_GROUP_DUP);
6145 * Swap the last stripe stripes and reduce @nr_extra_stripes.
6146 * The extra stripe would still be there, but won't be accessed.
6148 if (first->physical > second->physical) {
6149 swap(second->physical, first->physical);
6150 swap(second->dev, first->dev);
6155 *num_stripes_ret = num_stripes + nr_extra_stripes;
6156 *max_errors_ret = max_errors + nr_extra_stripes;
6157 bioc->replace_nr_stripes = nr_extra_stripes;
6160 static u64 btrfs_max_io_len(struct map_lookup *map, enum btrfs_map_op op,
6161 u64 offset, u32 *stripe_nr, u64 *stripe_offset,
6162 u64 *full_stripe_start)
6165 * Stripe_nr is the stripe where this block falls. stripe_offset is
6166 * the offset of this block in its stripe.
6168 *stripe_offset = offset & BTRFS_STRIPE_LEN_MASK;
6169 *stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
6170 ASSERT(*stripe_offset < U32_MAX);
6172 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6173 unsigned long full_stripe_len =
6174 btrfs_stripe_nr_to_offset(nr_data_stripes(map));
6177 * For full stripe start, we use previously calculated
6178 * @stripe_nr. Align it to nr_data_stripes, then multiply with
6181 * By this we can avoid u64 division completely. And we have
6182 * to go rounddown(), not round_down(), as nr_data_stripes is
6183 * not ensured to be power of 2.
6185 *full_stripe_start =
6186 btrfs_stripe_nr_to_offset(
6187 rounddown(*stripe_nr, nr_data_stripes(map)));
6189 ASSERT(*full_stripe_start + full_stripe_len > offset);
6190 ASSERT(*full_stripe_start <= offset);
6192 * For writes to RAID56, allow to write a full stripe set, but
6193 * no straddling of stripe sets.
6195 if (op == BTRFS_MAP_WRITE)
6196 return full_stripe_len - (offset - *full_stripe_start);
6200 * For other RAID types and for RAID56 reads, allow a single stripe (on
6203 if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK)
6204 return BTRFS_STRIPE_LEN - *stripe_offset;
6208 static void set_io_stripe(struct btrfs_io_stripe *dst, const struct map_lookup *map,
6209 u32 stripe_index, u64 stripe_offset, u32 stripe_nr)
6211 dst->dev = map->stripes[stripe_index].dev;
6212 dst->physical = map->stripes[stripe_index].physical +
6213 stripe_offset + btrfs_stripe_nr_to_offset(stripe_nr);
6217 * Map one logical range to one or more physical ranges.
6219 * @length: (Mandatory) mapped length of this run.
6220 * One logical range can be split into different segments
6221 * due to factors like zones and RAID0/5/6/10 stripe
6224 * @bioc_ret: (Mandatory) returned btrfs_io_context structure.
6225 * which has one or more physical ranges (btrfs_io_stripe)
6227 * Caller should call btrfs_put_bioc() to free it after use.
6229 * @smap: (Optional) single physical range optimization.
6230 * If the map request can be fulfilled by one single
6231 * physical range, and this is parameter is not NULL,
6232 * then @bioc_ret would be NULL, and @smap would be
6235 * @mirror_num_ret: (Mandatory) returned mirror number if the original
6238 * Mirror number 0 means to choose any live mirrors.
6240 * For non-RAID56 profiles, non-zero mirror_num means
6241 * the Nth mirror. (e.g. mirror_num 1 means the first
6244 * For RAID56 profile, mirror 1 means rebuild from P and
6245 * the remaining data stripes.
6247 * For RAID6 profile, mirror > 2 means mark another
6248 * data/P stripe error and rebuild from the remaining
6251 * @need_raid_map: (Used only for integrity checker) whether the map wants
6252 * a full stripe map (including all data and P/Q stripes)
6253 * for RAID56. Should always be 1 except integrity checker.
6255 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6256 u64 logical, u64 *length,
6257 struct btrfs_io_context **bioc_ret,
6258 struct btrfs_io_stripe *smap, int *mirror_num_ret,
6261 struct extent_map *em;
6262 struct map_lookup *map;
6270 int mirror_num = (mirror_num_ret ? *mirror_num_ret : 0);
6274 struct btrfs_io_context *bioc = NULL;
6275 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6276 int dev_replace_is_ongoing = 0;
6277 u16 num_alloc_stripes;
6278 u64 raid56_full_stripe_start = (u64)-1;
6283 num_copies = btrfs_num_copies(fs_info, logical, fs_info->sectorsize);
6284 if (mirror_num > num_copies)
6287 em = btrfs_get_chunk_map(fs_info, logical, *length);
6291 map = em->map_lookup;
6292 data_stripes = nr_data_stripes(map);
6294 map_offset = logical - em->start;
6295 max_len = btrfs_max_io_len(map, op, map_offset, &stripe_nr,
6296 &stripe_offset, &raid56_full_stripe_start);
6297 *length = min_t(u64, em->len - map_offset, max_len);
6299 down_read(&dev_replace->rwsem);
6300 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6302 * Hold the semaphore for read during the whole operation, write is
6303 * requested at commit time but must wait.
6305 if (!dev_replace_is_ongoing)
6306 up_read(&dev_replace->rwsem);
6310 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6311 stripe_index = stripe_nr % map->num_stripes;
6312 stripe_nr /= map->num_stripes;
6313 if (op == BTRFS_MAP_READ)
6315 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6316 if (op != BTRFS_MAP_READ) {
6317 num_stripes = map->num_stripes;
6318 } else if (mirror_num) {
6319 stripe_index = mirror_num - 1;
6321 stripe_index = find_live_mirror(fs_info, map, 0,
6322 dev_replace_is_ongoing);
6323 mirror_num = stripe_index + 1;
6326 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6327 if (op != BTRFS_MAP_READ) {
6328 num_stripes = map->num_stripes;
6329 } else if (mirror_num) {
6330 stripe_index = mirror_num - 1;
6335 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6336 u32 factor = map->num_stripes / map->sub_stripes;
6338 stripe_index = (stripe_nr % factor) * map->sub_stripes;
6339 stripe_nr /= factor;
6341 if (op != BTRFS_MAP_READ)
6342 num_stripes = map->sub_stripes;
6343 else if (mirror_num)
6344 stripe_index += mirror_num - 1;
6346 int old_stripe_index = stripe_index;
6347 stripe_index = find_live_mirror(fs_info, map,
6349 dev_replace_is_ongoing);
6350 mirror_num = stripe_index - old_stripe_index + 1;
6353 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6354 if (need_raid_map && (op != BTRFS_MAP_READ || mirror_num > 1)) {
6356 * Push stripe_nr back to the start of the full stripe
6357 * For those cases needing a full stripe, @stripe_nr
6358 * is the full stripe number.
6360 * Originally we go raid56_full_stripe_start / full_stripe_len,
6361 * but that can be expensive. Here we just divide
6362 * @stripe_nr with @data_stripes.
6364 stripe_nr /= data_stripes;
6366 /* RAID[56] write or recovery. Return all stripes */
6367 num_stripes = map->num_stripes;
6368 max_errors = btrfs_chunk_max_errors(map);
6370 /* Return the length to the full stripe end */
6371 *length = min(logical + *length,
6372 raid56_full_stripe_start + em->start +
6373 btrfs_stripe_nr_to_offset(data_stripes)) -
6379 * Mirror #0 or #1 means the original data block.
6380 * Mirror #2 is RAID5 parity block.
6381 * Mirror #3 is RAID6 Q block.
6383 stripe_index = stripe_nr % data_stripes;
6384 stripe_nr /= data_stripes;
6386 stripe_index = data_stripes + mirror_num - 2;
6388 /* We distribute the parity blocks across stripes */
6389 stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
6390 if (op == BTRFS_MAP_READ && mirror_num <= 1)
6395 * After this, stripe_nr is the number of stripes on this
6396 * device we have to walk to find the data, and stripe_index is
6397 * the number of our device in the stripe array
6399 stripe_index = stripe_nr % map->num_stripes;
6400 stripe_nr /= map->num_stripes;
6401 mirror_num = stripe_index + 1;
6403 if (stripe_index >= map->num_stripes) {
6405 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6406 stripe_index, map->num_stripes);
6411 num_alloc_stripes = num_stripes;
6412 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6413 op != BTRFS_MAP_READ)
6415 * For replace case, we need to add extra stripes for extra
6416 * duplicated stripes.
6418 * For both WRITE and GET_READ_MIRRORS, we may have at most
6419 * 2 more stripes (DUP types, otherwise 1).
6421 num_alloc_stripes += 2;
6424 * If this I/O maps to a single device, try to return the device and
6425 * physical block information on the stack instead of allocating an
6426 * I/O context structure.
6428 if (smap && num_alloc_stripes == 1 &&
6429 !((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1)) {
6430 set_io_stripe(smap, map, stripe_index, stripe_offset, stripe_nr);
6432 *mirror_num_ret = mirror_num;
6438 bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes);
6443 bioc->map_type = map->type;
6446 * For RAID56 full map, we need to make sure the stripes[] follows the
6447 * rule that data stripes are all ordered, then followed with P and Q
6450 * It's still mostly the same as other profiles, just with extra rotation.
6452 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6453 (op != BTRFS_MAP_READ || mirror_num > 1)) {
6455 * For RAID56 @stripe_nr is already the number of full stripes
6456 * before us, which is also the rotation value (needs to modulo
6457 * with num_stripes).
6459 * In this case, we just add @stripe_nr with @i, then do the
6460 * modulo, to reduce one modulo call.
6462 bioc->full_stripe_logical = em->start +
6463 btrfs_stripe_nr_to_offset(stripe_nr * data_stripes);
6464 for (i = 0; i < num_stripes; i++)
6465 set_io_stripe(&bioc->stripes[i], map,
6466 (i + stripe_nr) % num_stripes,
6467 stripe_offset, stripe_nr);
6470 * For all other non-RAID56 profiles, just copy the target
6471 * stripe into the bioc.
6473 for (i = 0; i < num_stripes; i++) {
6474 set_io_stripe(&bioc->stripes[i], map, stripe_index,
6475 stripe_offset, stripe_nr);
6480 if (op != BTRFS_MAP_READ)
6481 max_errors = btrfs_chunk_max_errors(map);
6483 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6484 op != BTRFS_MAP_READ) {
6485 handle_ops_on_dev_replace(op, bioc, dev_replace, logical,
6486 &num_stripes, &max_errors);
6490 bioc->num_stripes = num_stripes;
6491 bioc->max_errors = max_errors;
6492 bioc->mirror_num = mirror_num;
6495 if (dev_replace_is_ongoing) {
6496 lockdep_assert_held(&dev_replace->rwsem);
6497 /* Unlock and let waiting writers proceed */
6498 up_read(&dev_replace->rwsem);
6500 free_extent_map(em);
6504 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args,
6505 const struct btrfs_fs_devices *fs_devices)
6507 if (args->fsid == NULL)
6509 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0)
6514 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args,
6515 const struct btrfs_device *device)
6517 if (args->missing) {
6518 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) &&
6524 if (device->devid != args->devid)
6526 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0)
6532 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6535 * If devid and uuid are both specified, the match must be exact, otherwise
6536 * only devid is used.
6538 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,
6539 const struct btrfs_dev_lookup_args *args)
6541 struct btrfs_device *device;
6542 struct btrfs_fs_devices *seed_devs;
6544 if (dev_args_match_fs_devices(args, fs_devices)) {
6545 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6546 if (dev_args_match_device(args, device))
6551 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6552 if (!dev_args_match_fs_devices(args, seed_devs))
6554 list_for_each_entry(device, &seed_devs->devices, dev_list) {
6555 if (dev_args_match_device(args, device))
6563 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6564 u64 devid, u8 *dev_uuid)
6566 struct btrfs_device *device;
6567 unsigned int nofs_flag;
6570 * We call this under the chunk_mutex, so we want to use NOFS for this
6571 * allocation, however we don't want to change btrfs_alloc_device() to
6572 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6576 nofs_flag = memalloc_nofs_save();
6577 device = btrfs_alloc_device(NULL, &devid, dev_uuid, NULL);
6578 memalloc_nofs_restore(nofs_flag);
6582 list_add(&device->dev_list, &fs_devices->devices);
6583 device->fs_devices = fs_devices;
6584 fs_devices->num_devices++;
6586 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6587 fs_devices->missing_devices++;
6593 * Allocate new device struct, set up devid and UUID.
6595 * @fs_info: used only for generating a new devid, can be NULL if
6596 * devid is provided (i.e. @devid != NULL).
6597 * @devid: a pointer to devid for this device. If NULL a new devid
6599 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6601 * @path: a pointer to device path if available, NULL otherwise.
6603 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6604 * on error. Returned struct is not linked onto any lists and must be
6605 * destroyed with btrfs_free_device.
6607 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6608 const u64 *devid, const u8 *uuid,
6611 struct btrfs_device *dev;
6614 if (WARN_ON(!devid && !fs_info))
6615 return ERR_PTR(-EINVAL);
6617 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
6619 return ERR_PTR(-ENOMEM);
6621 INIT_LIST_HEAD(&dev->dev_list);
6622 INIT_LIST_HEAD(&dev->dev_alloc_list);
6623 INIT_LIST_HEAD(&dev->post_commit_list);
6625 atomic_set(&dev->dev_stats_ccnt, 0);
6626 btrfs_device_data_ordered_init(dev);
6627 extent_io_tree_init(fs_info, &dev->alloc_state, IO_TREE_DEVICE_ALLOC_STATE);
6634 ret = find_next_devid(fs_info, &tmp);
6636 btrfs_free_device(dev);
6637 return ERR_PTR(ret);
6643 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6645 generate_random_uuid(dev->uuid);
6648 struct rcu_string *name;
6650 name = rcu_string_strdup(path, GFP_KERNEL);
6652 btrfs_free_device(dev);
6653 return ERR_PTR(-ENOMEM);
6655 rcu_assign_pointer(dev->name, name);
6661 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6662 u64 devid, u8 *uuid, bool error)
6665 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6668 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6672 u64 btrfs_calc_stripe_length(const struct extent_map *em)
6674 const struct map_lookup *map = em->map_lookup;
6675 const int data_stripes = calc_data_stripes(map->type, map->num_stripes);
6677 return div_u64(em->len, data_stripes);
6680 #if BITS_PER_LONG == 32
6682 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
6683 * can't be accessed on 32bit systems.
6685 * This function do mount time check to reject the fs if it already has
6686 * metadata chunk beyond that limit.
6688 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6689 u64 logical, u64 length, u64 type)
6691 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6694 if (logical + length < MAX_LFS_FILESIZE)
6697 btrfs_err_32bit_limit(fs_info);
6702 * This is to give early warning for any metadata chunk reaching
6703 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
6704 * Although we can still access the metadata, it's not going to be possible
6705 * once the limit is reached.
6707 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6708 u64 logical, u64 length, u64 type)
6710 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6713 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)
6716 btrfs_warn_32bit_limit(fs_info);
6720 static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info,
6721 u64 devid, u8 *uuid)
6723 struct btrfs_device *dev;
6725 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6726 btrfs_report_missing_device(fs_info, devid, uuid, true);
6727 return ERR_PTR(-ENOENT);
6730 dev = add_missing_dev(fs_info->fs_devices, devid, uuid);
6732 btrfs_err(fs_info, "failed to init missing device %llu: %ld",
6733 devid, PTR_ERR(dev));
6736 btrfs_report_missing_device(fs_info, devid, uuid, false);
6741 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6742 struct btrfs_chunk *chunk)
6744 BTRFS_DEV_LOOKUP_ARGS(args);
6745 struct btrfs_fs_info *fs_info = leaf->fs_info;
6746 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6747 struct map_lookup *map;
6748 struct extent_map *em;
6753 u8 uuid[BTRFS_UUID_SIZE];
6759 logical = key->offset;
6760 length = btrfs_chunk_length(leaf, chunk);
6761 type = btrfs_chunk_type(leaf, chunk);
6762 index = btrfs_bg_flags_to_raid_index(type);
6763 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6765 #if BITS_PER_LONG == 32
6766 ret = check_32bit_meta_chunk(fs_info, logical, length, type);
6769 warn_32bit_meta_chunk(fs_info, logical, length, type);
6773 * Only need to verify chunk item if we're reading from sys chunk array,
6774 * as chunk item in tree block is already verified by tree-checker.
6776 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6777 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6782 read_lock(&map_tree->lock);
6783 em = lookup_extent_mapping(map_tree, logical, 1);
6784 read_unlock(&map_tree->lock);
6786 /* already mapped? */
6787 if (em && em->start <= logical && em->start + em->len > logical) {
6788 free_extent_map(em);
6791 free_extent_map(em);
6794 em = alloc_extent_map();
6797 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6799 free_extent_map(em);
6803 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6804 em->map_lookup = map;
6805 em->start = logical;
6808 em->block_start = 0;
6809 em->block_len = em->len;
6811 map->num_stripes = num_stripes;
6812 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6813 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6816 * We can't use the sub_stripes value, as for profiles other than
6817 * RAID10, they may have 0 as sub_stripes for filesystems created by
6818 * older mkfs (<v5.4).
6819 * In that case, it can cause divide-by-zero errors later.
6820 * Since currently sub_stripes is fixed for each profile, let's
6821 * use the trusted value instead.
6823 map->sub_stripes = btrfs_raid_array[index].sub_stripes;
6824 map->verified_stripes = 0;
6825 em->orig_block_len = btrfs_calc_stripe_length(em);
6826 for (i = 0; i < num_stripes; i++) {
6827 map->stripes[i].physical =
6828 btrfs_stripe_offset_nr(leaf, chunk, i);
6829 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6831 read_extent_buffer(leaf, uuid, (unsigned long)
6832 btrfs_stripe_dev_uuid_nr(chunk, i),
6835 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args);
6836 if (!map->stripes[i].dev) {
6837 map->stripes[i].dev = handle_missing_device(fs_info,
6839 if (IS_ERR(map->stripes[i].dev)) {
6840 ret = PTR_ERR(map->stripes[i].dev);
6841 free_extent_map(em);
6846 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6847 &(map->stripes[i].dev->dev_state));
6850 write_lock(&map_tree->lock);
6851 ret = add_extent_mapping(map_tree, em, 0);
6852 write_unlock(&map_tree->lock);
6855 "failed to add chunk map, start=%llu len=%llu: %d",
6856 em->start, em->len, ret);
6858 free_extent_map(em);
6863 static void fill_device_from_item(struct extent_buffer *leaf,
6864 struct btrfs_dev_item *dev_item,
6865 struct btrfs_device *device)
6869 device->devid = btrfs_device_id(leaf, dev_item);
6870 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6871 device->total_bytes = device->disk_total_bytes;
6872 device->commit_total_bytes = device->disk_total_bytes;
6873 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6874 device->commit_bytes_used = device->bytes_used;
6875 device->type = btrfs_device_type(leaf, dev_item);
6876 device->io_align = btrfs_device_io_align(leaf, dev_item);
6877 device->io_width = btrfs_device_io_width(leaf, dev_item);
6878 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6879 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6880 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6882 ptr = btrfs_device_uuid(dev_item);
6883 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6886 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6889 struct btrfs_fs_devices *fs_devices;
6892 lockdep_assert_held(&uuid_mutex);
6895 /* This will match only for multi-device seed fs */
6896 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
6897 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6901 fs_devices = find_fsid(fsid, NULL);
6903 if (!btrfs_test_opt(fs_info, DEGRADED))
6904 return ERR_PTR(-ENOENT);
6906 fs_devices = alloc_fs_devices(fsid, NULL);
6907 if (IS_ERR(fs_devices))
6910 fs_devices->seeding = true;
6911 fs_devices->opened = 1;
6916 * Upon first call for a seed fs fsid, just create a private copy of the
6917 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
6919 fs_devices = clone_fs_devices(fs_devices);
6920 if (IS_ERR(fs_devices))
6923 ret = open_fs_devices(fs_devices, BLK_OPEN_READ, fs_info->bdev_holder);
6925 free_fs_devices(fs_devices);
6926 return ERR_PTR(ret);
6929 if (!fs_devices->seeding) {
6930 close_fs_devices(fs_devices);
6931 free_fs_devices(fs_devices);
6932 return ERR_PTR(-EINVAL);
6935 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
6940 static int read_one_dev(struct extent_buffer *leaf,
6941 struct btrfs_dev_item *dev_item)
6943 BTRFS_DEV_LOOKUP_ARGS(args);
6944 struct btrfs_fs_info *fs_info = leaf->fs_info;
6945 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6946 struct btrfs_device *device;
6949 u8 fs_uuid[BTRFS_FSID_SIZE];
6950 u8 dev_uuid[BTRFS_UUID_SIZE];
6952 devid = btrfs_device_id(leaf, dev_item);
6954 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6956 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6958 args.uuid = dev_uuid;
6959 args.fsid = fs_uuid;
6961 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
6962 fs_devices = open_seed_devices(fs_info, fs_uuid);
6963 if (IS_ERR(fs_devices))
6964 return PTR_ERR(fs_devices);
6967 device = btrfs_find_device(fs_info->fs_devices, &args);
6969 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6970 btrfs_report_missing_device(fs_info, devid,
6975 device = add_missing_dev(fs_devices, devid, dev_uuid);
6976 if (IS_ERR(device)) {
6978 "failed to add missing dev %llu: %ld",
6979 devid, PTR_ERR(device));
6980 return PTR_ERR(device);
6982 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6984 if (!device->bdev) {
6985 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6986 btrfs_report_missing_device(fs_info,
6987 devid, dev_uuid, true);
6990 btrfs_report_missing_device(fs_info, devid,
6994 if (!device->bdev &&
6995 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6997 * this happens when a device that was properly setup
6998 * in the device info lists suddenly goes bad.
6999 * device->bdev is NULL, and so we have to set
7000 * device->missing to one here
7002 device->fs_devices->missing_devices++;
7003 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
7006 /* Move the device to its own fs_devices */
7007 if (device->fs_devices != fs_devices) {
7008 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
7009 &device->dev_state));
7011 list_move(&device->dev_list, &fs_devices->devices);
7012 device->fs_devices->num_devices--;
7013 fs_devices->num_devices++;
7015 device->fs_devices->missing_devices--;
7016 fs_devices->missing_devices++;
7018 device->fs_devices = fs_devices;
7022 if (device->fs_devices != fs_info->fs_devices) {
7023 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
7024 if (device->generation !=
7025 btrfs_device_generation(leaf, dev_item))
7029 fill_device_from_item(leaf, dev_item, device);
7031 u64 max_total_bytes = bdev_nr_bytes(device->bdev);
7033 if (device->total_bytes > max_total_bytes) {
7035 "device total_bytes should be at most %llu but found %llu",
7036 max_total_bytes, device->total_bytes);
7040 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7041 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7042 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7043 device->fs_devices->total_rw_bytes += device->total_bytes;
7044 atomic64_add(device->total_bytes - device->bytes_used,
7045 &fs_info->free_chunk_space);
7051 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7053 struct btrfs_super_block *super_copy = fs_info->super_copy;
7054 struct extent_buffer *sb;
7055 struct btrfs_disk_key *disk_key;
7056 struct btrfs_chunk *chunk;
7058 unsigned long sb_array_offset;
7065 struct btrfs_key key;
7067 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7070 * We allocated a dummy extent, just to use extent buffer accessors.
7071 * There will be unused space after BTRFS_SUPER_INFO_SIZE, but
7072 * that's fine, we will not go beyond system chunk array anyway.
7074 sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET);
7077 set_extent_buffer_uptodate(sb);
7079 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7080 array_size = btrfs_super_sys_array_size(super_copy);
7082 array_ptr = super_copy->sys_chunk_array;
7083 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7086 while (cur_offset < array_size) {
7087 disk_key = (struct btrfs_disk_key *)array_ptr;
7088 len = sizeof(*disk_key);
7089 if (cur_offset + len > array_size)
7090 goto out_short_read;
7092 btrfs_disk_key_to_cpu(&key, disk_key);
7095 sb_array_offset += len;
7098 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7100 "unexpected item type %u in sys_array at offset %u",
7101 (u32)key.type, cur_offset);
7106 chunk = (struct btrfs_chunk *)sb_array_offset;
7108 * At least one btrfs_chunk with one stripe must be present,
7109 * exact stripe count check comes afterwards
7111 len = btrfs_chunk_item_size(1);
7112 if (cur_offset + len > array_size)
7113 goto out_short_read;
7115 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7118 "invalid number of stripes %u in sys_array at offset %u",
7119 num_stripes, cur_offset);
7124 type = btrfs_chunk_type(sb, chunk);
7125 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7127 "invalid chunk type %llu in sys_array at offset %u",
7133 len = btrfs_chunk_item_size(num_stripes);
7134 if (cur_offset + len > array_size)
7135 goto out_short_read;
7137 ret = read_one_chunk(&key, sb, chunk);
7142 sb_array_offset += len;
7145 clear_extent_buffer_uptodate(sb);
7146 free_extent_buffer_stale(sb);
7150 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7152 clear_extent_buffer_uptodate(sb);
7153 free_extent_buffer_stale(sb);
7158 * Check if all chunks in the fs are OK for read-write degraded mount
7160 * If the @failing_dev is specified, it's accounted as missing.
7162 * Return true if all chunks meet the minimal RW mount requirements.
7163 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7165 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7166 struct btrfs_device *failing_dev)
7168 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7169 struct extent_map *em;
7173 read_lock(&map_tree->lock);
7174 em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7175 read_unlock(&map_tree->lock);
7176 /* No chunk at all? Return false anyway */
7182 struct map_lookup *map;
7187 map = em->map_lookup;
7189 btrfs_get_num_tolerated_disk_barrier_failures(
7191 for (i = 0; i < map->num_stripes; i++) {
7192 struct btrfs_device *dev = map->stripes[i].dev;
7194 if (!dev || !dev->bdev ||
7195 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7196 dev->last_flush_error)
7198 else if (failing_dev && failing_dev == dev)
7201 if (missing > max_tolerated) {
7204 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7205 em->start, missing, max_tolerated);
7206 free_extent_map(em);
7210 next_start = extent_map_end(em);
7211 free_extent_map(em);
7213 read_lock(&map_tree->lock);
7214 em = lookup_extent_mapping(map_tree, next_start,
7215 (u64)(-1) - next_start);
7216 read_unlock(&map_tree->lock);
7222 static void readahead_tree_node_children(struct extent_buffer *node)
7225 const int nr_items = btrfs_header_nritems(node);
7227 for (i = 0; i < nr_items; i++)
7228 btrfs_readahead_node_child(node, i);
7231 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7233 struct btrfs_root *root = fs_info->chunk_root;
7234 struct btrfs_path *path;
7235 struct extent_buffer *leaf;
7236 struct btrfs_key key;
7237 struct btrfs_key found_key;
7242 u64 last_ra_node = 0;
7244 path = btrfs_alloc_path();
7249 * uuid_mutex is needed only if we are mounting a sprout FS
7250 * otherwise we don't need it.
7252 mutex_lock(&uuid_mutex);
7255 * It is possible for mount and umount to race in such a way that
7256 * we execute this code path, but open_fs_devices failed to clear
7257 * total_rw_bytes. We certainly want it cleared before reading the
7258 * device items, so clear it here.
7260 fs_info->fs_devices->total_rw_bytes = 0;
7263 * Lockdep complains about possible circular locking dependency between
7264 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores
7265 * used for freeze procection of a fs (struct super_block.s_writers),
7266 * which we take when starting a transaction, and extent buffers of the
7267 * chunk tree if we call read_one_dev() while holding a lock on an
7268 * extent buffer of the chunk tree. Since we are mounting the filesystem
7269 * and at this point there can't be any concurrent task modifying the
7270 * chunk tree, to keep it simple, just skip locking on the chunk tree.
7272 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
7273 path->skip_locking = 1;
7276 * Read all device items, and then all the chunk items. All
7277 * device items are found before any chunk item (their object id
7278 * is smaller than the lowest possible object id for a chunk
7279 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7281 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7284 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) {
7285 struct extent_buffer *node = path->nodes[1];
7287 leaf = path->nodes[0];
7288 slot = path->slots[0];
7291 if (last_ra_node != node->start) {
7292 readahead_tree_node_children(node);
7293 last_ra_node = node->start;
7296 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7297 struct btrfs_dev_item *dev_item;
7298 dev_item = btrfs_item_ptr(leaf, slot,
7299 struct btrfs_dev_item);
7300 ret = read_one_dev(leaf, dev_item);
7304 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7305 struct btrfs_chunk *chunk;
7308 * We are only called at mount time, so no need to take
7309 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7310 * we always lock first fs_info->chunk_mutex before
7311 * acquiring any locks on the chunk tree. This is a
7312 * requirement for chunk allocation, see the comment on
7313 * top of btrfs_chunk_alloc() for details.
7315 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7316 ret = read_one_chunk(&found_key, leaf, chunk);
7321 /* Catch error found during iteration */
7328 * After loading chunk tree, we've got all device information,
7329 * do another round of validation checks.
7331 if (total_dev != fs_info->fs_devices->total_devices) {
7333 "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit",
7334 btrfs_super_num_devices(fs_info->super_copy),
7336 fs_info->fs_devices->total_devices = total_dev;
7337 btrfs_set_super_num_devices(fs_info->super_copy, total_dev);
7339 if (btrfs_super_total_bytes(fs_info->super_copy) <
7340 fs_info->fs_devices->total_rw_bytes) {
7342 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7343 btrfs_super_total_bytes(fs_info->super_copy),
7344 fs_info->fs_devices->total_rw_bytes);
7350 mutex_unlock(&uuid_mutex);
7352 btrfs_free_path(path);
7356 int btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7358 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7359 struct btrfs_device *device;
7362 fs_devices->fs_info = fs_info;
7364 mutex_lock(&fs_devices->device_list_mutex);
7365 list_for_each_entry(device, &fs_devices->devices, dev_list)
7366 device->fs_info = fs_info;
7368 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7369 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7370 device->fs_info = fs_info;
7371 ret = btrfs_get_dev_zone_info(device, false);
7376 seed_devs->fs_info = fs_info;
7378 mutex_unlock(&fs_devices->device_list_mutex);
7383 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7384 const struct btrfs_dev_stats_item *ptr,
7389 read_extent_buffer(eb, &val,
7390 offsetof(struct btrfs_dev_stats_item, values) +
7391 ((unsigned long)ptr) + (index * sizeof(u64)),
7396 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7397 struct btrfs_dev_stats_item *ptr,
7400 write_extent_buffer(eb, &val,
7401 offsetof(struct btrfs_dev_stats_item, values) +
7402 ((unsigned long)ptr) + (index * sizeof(u64)),
7406 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7407 struct btrfs_path *path)
7409 struct btrfs_dev_stats_item *ptr;
7410 struct extent_buffer *eb;
7411 struct btrfs_key key;
7415 if (!device->fs_info->dev_root)
7418 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7419 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7420 key.offset = device->devid;
7421 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7423 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7424 btrfs_dev_stat_set(device, i, 0);
7425 device->dev_stats_valid = 1;
7426 btrfs_release_path(path);
7427 return ret < 0 ? ret : 0;
7429 slot = path->slots[0];
7430 eb = path->nodes[0];
7431 item_size = btrfs_item_size(eb, slot);
7433 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7435 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7436 if (item_size >= (1 + i) * sizeof(__le64))
7437 btrfs_dev_stat_set(device, i,
7438 btrfs_dev_stats_value(eb, ptr, i));
7440 btrfs_dev_stat_set(device, i, 0);
7443 device->dev_stats_valid = 1;
7444 btrfs_dev_stat_print_on_load(device);
7445 btrfs_release_path(path);
7450 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7452 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7453 struct btrfs_device *device;
7454 struct btrfs_path *path = NULL;
7457 path = btrfs_alloc_path();
7461 mutex_lock(&fs_devices->device_list_mutex);
7462 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7463 ret = btrfs_device_init_dev_stats(device, path);
7467 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7468 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7469 ret = btrfs_device_init_dev_stats(device, path);
7475 mutex_unlock(&fs_devices->device_list_mutex);
7477 btrfs_free_path(path);
7481 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7482 struct btrfs_device *device)
7484 struct btrfs_fs_info *fs_info = trans->fs_info;
7485 struct btrfs_root *dev_root = fs_info->dev_root;
7486 struct btrfs_path *path;
7487 struct btrfs_key key;
7488 struct extent_buffer *eb;
7489 struct btrfs_dev_stats_item *ptr;
7493 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7494 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7495 key.offset = device->devid;
7497 path = btrfs_alloc_path();
7500 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7502 btrfs_warn_in_rcu(fs_info,
7503 "error %d while searching for dev_stats item for device %s",
7504 ret, btrfs_dev_name(device));
7509 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7510 /* need to delete old one and insert a new one */
7511 ret = btrfs_del_item(trans, dev_root, path);
7513 btrfs_warn_in_rcu(fs_info,
7514 "delete too small dev_stats item for device %s failed %d",
7515 btrfs_dev_name(device), ret);
7522 /* need to insert a new item */
7523 btrfs_release_path(path);
7524 ret = btrfs_insert_empty_item(trans, dev_root, path,
7525 &key, sizeof(*ptr));
7527 btrfs_warn_in_rcu(fs_info,
7528 "insert dev_stats item for device %s failed %d",
7529 btrfs_dev_name(device), ret);
7534 eb = path->nodes[0];
7535 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7536 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7537 btrfs_set_dev_stats_value(eb, ptr, i,
7538 btrfs_dev_stat_read(device, i));
7539 btrfs_mark_buffer_dirty(eb);
7542 btrfs_free_path(path);
7547 * called from commit_transaction. Writes all changed device stats to disk.
7549 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7551 struct btrfs_fs_info *fs_info = trans->fs_info;
7552 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7553 struct btrfs_device *device;
7557 mutex_lock(&fs_devices->device_list_mutex);
7558 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7559 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7560 if (!device->dev_stats_valid || stats_cnt == 0)
7565 * There is a LOAD-LOAD control dependency between the value of
7566 * dev_stats_ccnt and updating the on-disk values which requires
7567 * reading the in-memory counters. Such control dependencies
7568 * require explicit read memory barriers.
7570 * This memory barriers pairs with smp_mb__before_atomic in
7571 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7572 * barrier implied by atomic_xchg in
7573 * btrfs_dev_stats_read_and_reset
7577 ret = update_dev_stat_item(trans, device);
7579 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7581 mutex_unlock(&fs_devices->device_list_mutex);
7586 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7588 btrfs_dev_stat_inc(dev, index);
7590 if (!dev->dev_stats_valid)
7592 btrfs_err_rl_in_rcu(dev->fs_info,
7593 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7594 btrfs_dev_name(dev),
7595 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7596 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7597 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7598 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7599 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7602 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7606 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7607 if (btrfs_dev_stat_read(dev, i) != 0)
7609 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7610 return; /* all values == 0, suppress message */
7612 btrfs_info_in_rcu(dev->fs_info,
7613 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7614 btrfs_dev_name(dev),
7615 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7616 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7617 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7618 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7619 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7622 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7623 struct btrfs_ioctl_get_dev_stats *stats)
7625 BTRFS_DEV_LOOKUP_ARGS(args);
7626 struct btrfs_device *dev;
7627 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7630 mutex_lock(&fs_devices->device_list_mutex);
7631 args.devid = stats->devid;
7632 dev = btrfs_find_device(fs_info->fs_devices, &args);
7633 mutex_unlock(&fs_devices->device_list_mutex);
7636 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7638 } else if (!dev->dev_stats_valid) {
7639 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7641 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7642 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7643 if (stats->nr_items > i)
7645 btrfs_dev_stat_read_and_reset(dev, i);
7647 btrfs_dev_stat_set(dev, i, 0);
7649 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7650 current->comm, task_pid_nr(current));
7652 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7653 if (stats->nr_items > i)
7654 stats->values[i] = btrfs_dev_stat_read(dev, i);
7656 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7657 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7662 * Update the size and bytes used for each device where it changed. This is
7663 * delayed since we would otherwise get errors while writing out the
7666 * Must be invoked during transaction commit.
7668 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7670 struct btrfs_device *curr, *next;
7672 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7674 if (list_empty(&trans->dev_update_list))
7678 * We don't need the device_list_mutex here. This list is owned by the
7679 * transaction and the transaction must complete before the device is
7682 mutex_lock(&trans->fs_info->chunk_mutex);
7683 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7685 list_del_init(&curr->post_commit_list);
7686 curr->commit_total_bytes = curr->disk_total_bytes;
7687 curr->commit_bytes_used = curr->bytes_used;
7689 mutex_unlock(&trans->fs_info->chunk_mutex);
7693 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7695 int btrfs_bg_type_to_factor(u64 flags)
7697 const int index = btrfs_bg_flags_to_raid_index(flags);
7699 return btrfs_raid_array[index].ncopies;
7704 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7705 u64 chunk_offset, u64 devid,
7706 u64 physical_offset, u64 physical_len)
7708 struct btrfs_dev_lookup_args args = { .devid = devid };
7709 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7710 struct extent_map *em;
7711 struct map_lookup *map;
7712 struct btrfs_device *dev;
7718 read_lock(&em_tree->lock);
7719 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7720 read_unlock(&em_tree->lock);
7724 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7725 physical_offset, devid);
7730 map = em->map_lookup;
7731 stripe_len = btrfs_calc_stripe_length(em);
7732 if (physical_len != stripe_len) {
7734 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7735 physical_offset, devid, em->start, physical_len,
7742 * Very old mkfs.btrfs (before v4.1) will not respect the reserved
7743 * space. Although kernel can handle it without problem, better to warn
7746 if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED)
7748 "devid %llu physical %llu len %llu inside the reserved space",
7749 devid, physical_offset, physical_len);
7751 for (i = 0; i < map->num_stripes; i++) {
7752 if (map->stripes[i].dev->devid == devid &&
7753 map->stripes[i].physical == physical_offset) {
7755 if (map->verified_stripes >= map->num_stripes) {
7757 "too many dev extents for chunk %llu found",
7762 map->verified_stripes++;
7768 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7769 physical_offset, devid);
7773 /* Make sure no dev extent is beyond device boundary */
7774 dev = btrfs_find_device(fs_info->fs_devices, &args);
7776 btrfs_err(fs_info, "failed to find devid %llu", devid);
7781 if (physical_offset + physical_len > dev->disk_total_bytes) {
7783 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7784 devid, physical_offset, physical_len,
7785 dev->disk_total_bytes);
7790 if (dev->zone_info) {
7791 u64 zone_size = dev->zone_info->zone_size;
7793 if (!IS_ALIGNED(physical_offset, zone_size) ||
7794 !IS_ALIGNED(physical_len, zone_size)) {
7796 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
7797 devid, physical_offset, physical_len);
7804 free_extent_map(em);
7808 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7810 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7811 struct extent_map *em;
7812 struct rb_node *node;
7815 read_lock(&em_tree->lock);
7816 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
7817 em = rb_entry(node, struct extent_map, rb_node);
7818 if (em->map_lookup->num_stripes !=
7819 em->map_lookup->verified_stripes) {
7821 "chunk %llu has missing dev extent, have %d expect %d",
7822 em->start, em->map_lookup->verified_stripes,
7823 em->map_lookup->num_stripes);
7829 read_unlock(&em_tree->lock);
7834 * Ensure that all dev extents are mapped to correct chunk, otherwise
7835 * later chunk allocation/free would cause unexpected behavior.
7837 * NOTE: This will iterate through the whole device tree, which should be of
7838 * the same size level as the chunk tree. This slightly increases mount time.
7840 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
7842 struct btrfs_path *path;
7843 struct btrfs_root *root = fs_info->dev_root;
7844 struct btrfs_key key;
7846 u64 prev_dev_ext_end = 0;
7850 * We don't have a dev_root because we mounted with ignorebadroots and
7851 * failed to load the root, so we want to skip the verification in this
7854 * However if the dev root is fine, but the tree itself is corrupted
7855 * we'd still fail to mount. This verification is only to make sure
7856 * writes can happen safely, so instead just bypass this check
7857 * completely in the case of IGNOREBADROOTS.
7859 if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
7863 key.type = BTRFS_DEV_EXTENT_KEY;
7866 path = btrfs_alloc_path();
7870 path->reada = READA_FORWARD;
7871 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7875 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7876 ret = btrfs_next_leaf(root, path);
7879 /* No dev extents at all? Not good */
7886 struct extent_buffer *leaf = path->nodes[0];
7887 struct btrfs_dev_extent *dext;
7888 int slot = path->slots[0];
7890 u64 physical_offset;
7894 btrfs_item_key_to_cpu(leaf, &key, slot);
7895 if (key.type != BTRFS_DEV_EXTENT_KEY)
7897 devid = key.objectid;
7898 physical_offset = key.offset;
7900 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
7901 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
7902 physical_len = btrfs_dev_extent_length(leaf, dext);
7904 /* Check if this dev extent overlaps with the previous one */
7905 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
7907 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
7908 devid, physical_offset, prev_dev_ext_end);
7913 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
7914 physical_offset, physical_len);
7918 prev_dev_ext_end = physical_offset + physical_len;
7920 ret = btrfs_next_item(root, path);
7929 /* Ensure all chunks have corresponding dev extents */
7930 ret = verify_chunk_dev_extent_mapping(fs_info);
7932 btrfs_free_path(path);
7937 * Check whether the given block group or device is pinned by any inode being
7938 * used as a swapfile.
7940 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
7942 struct btrfs_swapfile_pin *sp;
7943 struct rb_node *node;
7945 spin_lock(&fs_info->swapfile_pins_lock);
7946 node = fs_info->swapfile_pins.rb_node;
7948 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
7950 node = node->rb_left;
7951 else if (ptr > sp->ptr)
7952 node = node->rb_right;
7956 spin_unlock(&fs_info->swapfile_pins_lock);
7957 return node != NULL;
7960 static int relocating_repair_kthread(void *data)
7962 struct btrfs_block_group *cache = data;
7963 struct btrfs_fs_info *fs_info = cache->fs_info;
7967 target = cache->start;
7968 btrfs_put_block_group(cache);
7970 sb_start_write(fs_info->sb);
7971 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
7973 "zoned: skip relocating block group %llu to repair: EBUSY",
7975 sb_end_write(fs_info->sb);
7979 mutex_lock(&fs_info->reclaim_bgs_lock);
7981 /* Ensure block group still exists */
7982 cache = btrfs_lookup_block_group(fs_info, target);
7986 if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags))
7989 ret = btrfs_may_alloc_data_chunk(fs_info, target);
7994 "zoned: relocating block group %llu to repair IO failure",
7996 ret = btrfs_relocate_chunk(fs_info, target);
8000 btrfs_put_block_group(cache);
8001 mutex_unlock(&fs_info->reclaim_bgs_lock);
8002 btrfs_exclop_finish(fs_info);
8003 sb_end_write(fs_info->sb);
8008 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical)
8010 struct btrfs_block_group *cache;
8012 if (!btrfs_is_zoned(fs_info))
8015 /* Do not attempt to repair in degraded state */
8016 if (btrfs_test_opt(fs_info, DEGRADED))
8019 cache = btrfs_lookup_block_group(fs_info, logical);
8023 if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) {
8024 btrfs_put_block_group(cache);
8028 kthread_run(relocating_repair_kthread, cache,
8029 "btrfs-relocating-repair");
8034 static void map_raid56_repair_block(struct btrfs_io_context *bioc,
8035 struct btrfs_io_stripe *smap,
8038 int data_stripes = nr_bioc_data_stripes(bioc);
8041 for (i = 0; i < data_stripes; i++) {
8042 u64 stripe_start = bioc->full_stripe_logical +
8043 btrfs_stripe_nr_to_offset(i);
8045 if (logical >= stripe_start &&
8046 logical < stripe_start + BTRFS_STRIPE_LEN)
8049 ASSERT(i < data_stripes);
8050 smap->dev = bioc->stripes[i].dev;
8051 smap->physical = bioc->stripes[i].physical +
8052 ((logical - bioc->full_stripe_logical) &
8053 BTRFS_STRIPE_LEN_MASK);
8057 * Map a repair write into a single device.
8059 * A repair write is triggered by read time repair or scrub, which would only
8060 * update the contents of a single device.
8061 * Not update any other mirrors nor go through RMW path.
8063 * Callers should ensure:
8065 * - Call btrfs_bio_counter_inc_blocked() first
8066 * - The range does not cross stripe boundary
8067 * - Has a valid @mirror_num passed in.
8069 int btrfs_map_repair_block(struct btrfs_fs_info *fs_info,
8070 struct btrfs_io_stripe *smap, u64 logical,
8071 u32 length, int mirror_num)
8073 struct btrfs_io_context *bioc = NULL;
8074 u64 map_length = length;
8075 int mirror_ret = mirror_num;
8078 ASSERT(mirror_num > 0);
8080 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, &map_length,
8081 &bioc, smap, &mirror_ret, true);
8085 /* The map range should not cross stripe boundary. */
8086 ASSERT(map_length >= length);
8088 /* Already mapped to single stripe. */
8092 /* Map the RAID56 multi-stripe writes to a single one. */
8093 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
8094 map_raid56_repair_block(bioc, smap, logical);
8098 ASSERT(mirror_num <= bioc->num_stripes);
8099 smap->dev = bioc->stripes[mirror_num - 1].dev;
8100 smap->physical = bioc->stripes[mirror_num - 1].physical;
8102 btrfs_put_bioc(bioc);