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)
1916 ret = kern_path(device_path, LOOKUP_FOLLOW, &path);
1920 inode_update_time(d_inode(path.dentry), S_MTIME | S_CTIME | S_VERSION);
1924 static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans,
1925 struct btrfs_device *device)
1927 struct btrfs_root *root = device->fs_info->chunk_root;
1929 struct btrfs_path *path;
1930 struct btrfs_key key;
1932 path = btrfs_alloc_path();
1936 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1937 key.type = BTRFS_DEV_ITEM_KEY;
1938 key.offset = device->devid;
1940 btrfs_reserve_chunk_metadata(trans, false);
1941 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1942 btrfs_trans_release_chunk_metadata(trans);
1949 ret = btrfs_del_item(trans, root, path);
1951 btrfs_free_path(path);
1956 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1957 * filesystem. It's up to the caller to adjust that number regarding eg. device
1960 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1968 seq = read_seqbegin(&fs_info->profiles_lock);
1970 all_avail = fs_info->avail_data_alloc_bits |
1971 fs_info->avail_system_alloc_bits |
1972 fs_info->avail_metadata_alloc_bits;
1973 } while (read_seqretry(&fs_info->profiles_lock, seq));
1975 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1976 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1979 if (num_devices < btrfs_raid_array[i].devs_min)
1980 return btrfs_raid_array[i].mindev_error;
1986 static struct btrfs_device * btrfs_find_next_active_device(
1987 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1989 struct btrfs_device *next_device;
1991 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1992 if (next_device != device &&
1993 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1994 && next_device->bdev)
2002 * Helper function to check if the given device is part of s_bdev / latest_dev
2003 * and replace it with the provided or the next active device, in the context
2004 * where this function called, there should be always be another device (or
2005 * this_dev) which is active.
2007 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
2008 struct btrfs_device *next_device)
2010 struct btrfs_fs_info *fs_info = device->fs_info;
2013 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
2015 ASSERT(next_device);
2017 if (fs_info->sb->s_bdev &&
2018 (fs_info->sb->s_bdev == device->bdev))
2019 fs_info->sb->s_bdev = next_device->bdev;
2021 if (fs_info->fs_devices->latest_dev->bdev == device->bdev)
2022 fs_info->fs_devices->latest_dev = next_device;
2026 * Return btrfs_fs_devices::num_devices excluding the device that's being
2027 * currently replaced.
2029 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2031 u64 num_devices = fs_info->fs_devices->num_devices;
2033 down_read(&fs_info->dev_replace.rwsem);
2034 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2035 ASSERT(num_devices > 1);
2038 up_read(&fs_info->dev_replace.rwsem);
2043 static void btrfs_scratch_superblock(struct btrfs_fs_info *fs_info,
2044 struct block_device *bdev, int copy_num)
2046 struct btrfs_super_block *disk_super;
2047 const size_t len = sizeof(disk_super->magic);
2048 const u64 bytenr = btrfs_sb_offset(copy_num);
2051 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr);
2052 if (IS_ERR(disk_super))
2055 memset(&disk_super->magic, 0, len);
2056 folio_mark_dirty(virt_to_folio(disk_super));
2057 btrfs_release_disk_super(disk_super);
2059 ret = sync_blockdev_range(bdev, bytenr, bytenr + len - 1);
2061 btrfs_warn(fs_info, "error clearing superblock number %d (%d)",
2065 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
2066 struct block_device *bdev,
2067 const char *device_path)
2074 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2075 if (bdev_is_zoned(bdev))
2076 btrfs_reset_sb_log_zones(bdev, copy_num);
2078 btrfs_scratch_superblock(fs_info, bdev, copy_num);
2081 /* Notify udev that device has changed */
2082 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2084 /* Update ctime/mtime for device path for libblkid */
2085 update_dev_time(device_path);
2088 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
2089 struct btrfs_dev_lookup_args *args,
2090 struct block_device **bdev, void **holder)
2092 struct btrfs_trans_handle *trans;
2093 struct btrfs_device *device;
2094 struct btrfs_fs_devices *cur_devices;
2095 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2099 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2100 btrfs_err(fs_info, "device remove not supported on extent tree v2 yet");
2105 * The device list in fs_devices is accessed without locks (neither
2106 * uuid_mutex nor device_list_mutex) as it won't change on a mounted
2107 * filesystem and another device rm cannot run.
2109 num_devices = btrfs_num_devices(fs_info);
2111 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2115 device = btrfs_find_device(fs_info->fs_devices, args);
2118 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2124 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2125 btrfs_warn_in_rcu(fs_info,
2126 "cannot remove device %s (devid %llu) due to active swapfile",
2127 btrfs_dev_name(device), device->devid);
2131 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
2132 return BTRFS_ERROR_DEV_TGT_REPLACE;
2134 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2135 fs_info->fs_devices->rw_devices == 1)
2136 return BTRFS_ERROR_DEV_ONLY_WRITABLE;
2138 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2139 mutex_lock(&fs_info->chunk_mutex);
2140 list_del_init(&device->dev_alloc_list);
2141 device->fs_devices->rw_devices--;
2142 mutex_unlock(&fs_info->chunk_mutex);
2145 ret = btrfs_shrink_device(device, 0);
2149 trans = btrfs_start_transaction(fs_info->chunk_root, 0);
2150 if (IS_ERR(trans)) {
2151 ret = PTR_ERR(trans);
2155 ret = btrfs_rm_dev_item(trans, device);
2157 /* Any error in dev item removal is critical */
2159 "failed to remove device item for devid %llu: %d",
2160 device->devid, ret);
2161 btrfs_abort_transaction(trans, ret);
2162 btrfs_end_transaction(trans);
2166 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2167 btrfs_scrub_cancel_dev(device);
2170 * the device list mutex makes sure that we don't change
2171 * the device list while someone else is writing out all
2172 * the device supers. Whoever is writing all supers, should
2173 * lock the device list mutex before getting the number of
2174 * devices in the super block (super_copy). Conversely,
2175 * whoever updates the number of devices in the super block
2176 * (super_copy) should hold the device list mutex.
2180 * In normal cases the cur_devices == fs_devices. But in case
2181 * of deleting a seed device, the cur_devices should point to
2182 * its own fs_devices listed under the fs_devices->seed_list.
2184 cur_devices = device->fs_devices;
2185 mutex_lock(&fs_devices->device_list_mutex);
2186 list_del_rcu(&device->dev_list);
2188 cur_devices->num_devices--;
2189 cur_devices->total_devices--;
2190 /* Update total_devices of the parent fs_devices if it's seed */
2191 if (cur_devices != fs_devices)
2192 fs_devices->total_devices--;
2194 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2195 cur_devices->missing_devices--;
2197 btrfs_assign_next_active_device(device, NULL);
2200 cur_devices->open_devices--;
2201 /* remove sysfs entry */
2202 btrfs_sysfs_remove_device(device);
2205 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2206 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2207 mutex_unlock(&fs_devices->device_list_mutex);
2210 * At this point, the device is zero sized and detached from the
2211 * devices list. All that's left is to zero out the old supers and
2214 * We cannot call btrfs_close_bdev() here because we're holding the sb
2215 * write lock, and blkdev_put() will pull in the ->open_mutex on the
2216 * block device and it's dependencies. Instead just flush the device
2217 * and let the caller do the final blkdev_put.
2219 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2220 btrfs_scratch_superblocks(fs_info, device->bdev,
2223 sync_blockdev(device->bdev);
2224 invalidate_bdev(device->bdev);
2228 *bdev = device->bdev;
2229 *holder = device->holder;
2231 btrfs_free_device(device);
2234 * This can happen if cur_devices is the private seed devices list. We
2235 * cannot call close_fs_devices() here because it expects the uuid_mutex
2236 * to be held, but in fact we don't need that for the private
2237 * seed_devices, we can simply decrement cur_devices->opened and then
2238 * remove it from our list and free the fs_devices.
2240 if (cur_devices->num_devices == 0) {
2241 list_del_init(&cur_devices->seed_list);
2242 ASSERT(cur_devices->opened == 1);
2243 cur_devices->opened--;
2244 free_fs_devices(cur_devices);
2247 ret = btrfs_commit_transaction(trans);
2252 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2253 mutex_lock(&fs_info->chunk_mutex);
2254 list_add(&device->dev_alloc_list,
2255 &fs_devices->alloc_list);
2256 device->fs_devices->rw_devices++;
2257 mutex_unlock(&fs_info->chunk_mutex);
2262 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2264 struct btrfs_fs_devices *fs_devices;
2266 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2269 * in case of fs with no seed, srcdev->fs_devices will point
2270 * to fs_devices of fs_info. However when the dev being replaced is
2271 * a seed dev it will point to the seed's local fs_devices. In short
2272 * srcdev will have its correct fs_devices in both the cases.
2274 fs_devices = srcdev->fs_devices;
2276 list_del_rcu(&srcdev->dev_list);
2277 list_del(&srcdev->dev_alloc_list);
2278 fs_devices->num_devices--;
2279 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2280 fs_devices->missing_devices--;
2282 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2283 fs_devices->rw_devices--;
2286 fs_devices->open_devices--;
2289 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2291 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2293 mutex_lock(&uuid_mutex);
2295 btrfs_close_bdev(srcdev);
2297 btrfs_free_device(srcdev);
2299 /* if this is no devs we rather delete the fs_devices */
2300 if (!fs_devices->num_devices) {
2302 * On a mounted FS, num_devices can't be zero unless it's a
2303 * seed. In case of a seed device being replaced, the replace
2304 * target added to the sprout FS, so there will be no more
2305 * device left under the seed FS.
2307 ASSERT(fs_devices->seeding);
2309 list_del_init(&fs_devices->seed_list);
2310 close_fs_devices(fs_devices);
2311 free_fs_devices(fs_devices);
2313 mutex_unlock(&uuid_mutex);
2316 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2318 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2320 mutex_lock(&fs_devices->device_list_mutex);
2322 btrfs_sysfs_remove_device(tgtdev);
2325 fs_devices->open_devices--;
2327 fs_devices->num_devices--;
2329 btrfs_assign_next_active_device(tgtdev, NULL);
2331 list_del_rcu(&tgtdev->dev_list);
2333 mutex_unlock(&fs_devices->device_list_mutex);
2335 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2338 btrfs_close_bdev(tgtdev);
2340 btrfs_free_device(tgtdev);
2344 * Populate args from device at path.
2346 * @fs_info: the filesystem
2347 * @args: the args to populate
2348 * @path: the path to the device
2350 * This will read the super block of the device at @path and populate @args with
2351 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to
2352 * lookup a device to operate on, but need to do it before we take any locks.
2353 * This properly handles the special case of "missing" that a user may pass in,
2354 * and does some basic sanity checks. The caller must make sure that @path is
2355 * properly NUL terminated before calling in, and must call
2356 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and
2359 * Return: 0 for success, -errno for failure
2361 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,
2362 struct btrfs_dev_lookup_args *args,
2365 struct btrfs_super_block *disk_super;
2366 struct block_device *bdev;
2369 if (!path || !path[0])
2371 if (!strcmp(path, "missing")) {
2372 args->missing = true;
2376 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL);
2377 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL);
2378 if (!args->uuid || !args->fsid) {
2379 btrfs_put_dev_args_from_path(args);
2383 ret = btrfs_get_bdev_and_sb(path, BLK_OPEN_READ, NULL, 0,
2384 &bdev, &disk_super);
2386 btrfs_put_dev_args_from_path(args);
2390 args->devid = btrfs_stack_device_id(&disk_super->dev_item);
2391 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE);
2392 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2393 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE);
2395 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
2396 btrfs_release_disk_super(disk_super);
2397 blkdev_put(bdev, NULL);
2402 * Only use this jointly with btrfs_get_dev_args_from_path() because we will
2403 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables
2404 * that don't need to be freed.
2406 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args)
2414 struct btrfs_device *btrfs_find_device_by_devspec(
2415 struct btrfs_fs_info *fs_info, u64 devid,
2416 const char *device_path)
2418 BTRFS_DEV_LOOKUP_ARGS(args);
2419 struct btrfs_device *device;
2424 device = btrfs_find_device(fs_info->fs_devices, &args);
2426 return ERR_PTR(-ENOENT);
2430 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path);
2432 return ERR_PTR(ret);
2433 device = btrfs_find_device(fs_info->fs_devices, &args);
2434 btrfs_put_dev_args_from_path(&args);
2436 return ERR_PTR(-ENOENT);
2440 static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info)
2442 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2443 struct btrfs_fs_devices *old_devices;
2444 struct btrfs_fs_devices *seed_devices;
2446 lockdep_assert_held(&uuid_mutex);
2447 if (!fs_devices->seeding)
2448 return ERR_PTR(-EINVAL);
2451 * Private copy of the seed devices, anchored at
2452 * fs_info->fs_devices->seed_list
2454 seed_devices = alloc_fs_devices(NULL, NULL);
2455 if (IS_ERR(seed_devices))
2456 return seed_devices;
2459 * It's necessary to retain a copy of the original seed fs_devices in
2460 * fs_uuids so that filesystems which have been seeded can successfully
2461 * reference the seed device from open_seed_devices. This also supports
2464 old_devices = clone_fs_devices(fs_devices);
2465 if (IS_ERR(old_devices)) {
2466 kfree(seed_devices);
2470 list_add(&old_devices->fs_list, &fs_uuids);
2472 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2473 seed_devices->opened = 1;
2474 INIT_LIST_HEAD(&seed_devices->devices);
2475 INIT_LIST_HEAD(&seed_devices->alloc_list);
2476 mutex_init(&seed_devices->device_list_mutex);
2478 return seed_devices;
2482 * Splice seed devices into the sprout fs_devices.
2483 * Generate a new fsid for the sprouted read-write filesystem.
2485 static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info,
2486 struct btrfs_fs_devices *seed_devices)
2488 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2489 struct btrfs_super_block *disk_super = fs_info->super_copy;
2490 struct btrfs_device *device;
2494 * We are updating the fsid, the thread leading to device_list_add()
2495 * could race, so uuid_mutex is needed.
2497 lockdep_assert_held(&uuid_mutex);
2500 * The threads listed below may traverse dev_list but can do that without
2501 * device_list_mutex:
2502 * - All device ops and balance - as we are in btrfs_exclop_start.
2503 * - Various dev_list readers - are using RCU.
2504 * - btrfs_ioctl_fitrim() - is using RCU.
2506 * For-read threads as below are using device_list_mutex:
2507 * - Readonly scrub btrfs_scrub_dev()
2508 * - Readonly scrub btrfs_scrub_progress()
2509 * - btrfs_get_dev_stats()
2511 lockdep_assert_held(&fs_devices->device_list_mutex);
2513 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2515 list_for_each_entry(device, &seed_devices->devices, dev_list)
2516 device->fs_devices = seed_devices;
2518 fs_devices->seeding = false;
2519 fs_devices->num_devices = 0;
2520 fs_devices->open_devices = 0;
2521 fs_devices->missing_devices = 0;
2522 fs_devices->rotating = false;
2523 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2525 generate_random_uuid(fs_devices->fsid);
2526 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2527 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2529 super_flags = btrfs_super_flags(disk_super) &
2530 ~BTRFS_SUPER_FLAG_SEEDING;
2531 btrfs_set_super_flags(disk_super, super_flags);
2535 * Store the expected generation for seed devices in device items.
2537 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2539 BTRFS_DEV_LOOKUP_ARGS(args);
2540 struct btrfs_fs_info *fs_info = trans->fs_info;
2541 struct btrfs_root *root = fs_info->chunk_root;
2542 struct btrfs_path *path;
2543 struct extent_buffer *leaf;
2544 struct btrfs_dev_item *dev_item;
2545 struct btrfs_device *device;
2546 struct btrfs_key key;
2547 u8 fs_uuid[BTRFS_FSID_SIZE];
2548 u8 dev_uuid[BTRFS_UUID_SIZE];
2551 path = btrfs_alloc_path();
2555 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2557 key.type = BTRFS_DEV_ITEM_KEY;
2560 btrfs_reserve_chunk_metadata(trans, false);
2561 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2562 btrfs_trans_release_chunk_metadata(trans);
2566 leaf = path->nodes[0];
2568 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2569 ret = btrfs_next_leaf(root, path);
2574 leaf = path->nodes[0];
2575 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2576 btrfs_release_path(path);
2580 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2581 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2582 key.type != BTRFS_DEV_ITEM_KEY)
2585 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2586 struct btrfs_dev_item);
2587 args.devid = btrfs_device_id(leaf, dev_item);
2588 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2590 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2592 args.uuid = dev_uuid;
2593 args.fsid = fs_uuid;
2594 device = btrfs_find_device(fs_info->fs_devices, &args);
2595 BUG_ON(!device); /* Logic error */
2597 if (device->fs_devices->seeding) {
2598 btrfs_set_device_generation(leaf, dev_item,
2599 device->generation);
2600 btrfs_mark_buffer_dirty(leaf);
2608 btrfs_free_path(path);
2612 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2614 struct btrfs_root *root = fs_info->dev_root;
2615 struct btrfs_trans_handle *trans;
2616 struct btrfs_device *device;
2617 struct block_device *bdev;
2618 struct super_block *sb = fs_info->sb;
2619 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2620 struct btrfs_fs_devices *seed_devices = NULL;
2621 u64 orig_super_total_bytes;
2622 u64 orig_super_num_devices;
2624 bool seeding_dev = false;
2625 bool locked = false;
2627 if (sb_rdonly(sb) && !fs_devices->seeding)
2630 bdev = blkdev_get_by_path(device_path, BLK_OPEN_WRITE,
2631 fs_info->bdev_holder, NULL);
2633 return PTR_ERR(bdev);
2635 if (!btrfs_check_device_zone_type(fs_info, bdev)) {
2640 if (fs_devices->seeding) {
2642 down_write(&sb->s_umount);
2643 mutex_lock(&uuid_mutex);
2647 sync_blockdev(bdev);
2650 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2651 if (device->bdev == bdev) {
2659 device = btrfs_alloc_device(fs_info, NULL, NULL, device_path);
2660 if (IS_ERR(device)) {
2661 /* we can safely leave the fs_devices entry around */
2662 ret = PTR_ERR(device);
2666 device->fs_info = fs_info;
2667 device->bdev = bdev;
2668 ret = lookup_bdev(device_path, &device->devt);
2670 goto error_free_device;
2672 ret = btrfs_get_dev_zone_info(device, false);
2674 goto error_free_device;
2676 trans = btrfs_start_transaction(root, 0);
2677 if (IS_ERR(trans)) {
2678 ret = PTR_ERR(trans);
2679 goto error_free_zone;
2682 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2683 device->generation = trans->transid;
2684 device->io_width = fs_info->sectorsize;
2685 device->io_align = fs_info->sectorsize;
2686 device->sector_size = fs_info->sectorsize;
2687 device->total_bytes =
2688 round_down(bdev_nr_bytes(bdev), fs_info->sectorsize);
2689 device->disk_total_bytes = device->total_bytes;
2690 device->commit_total_bytes = device->total_bytes;
2691 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2692 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2693 device->holder = fs_info->bdev_holder;
2694 device->dev_stats_valid = 1;
2695 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2698 btrfs_clear_sb_rdonly(sb);
2700 /* GFP_KERNEL allocation must not be under device_list_mutex */
2701 seed_devices = btrfs_init_sprout(fs_info);
2702 if (IS_ERR(seed_devices)) {
2703 ret = PTR_ERR(seed_devices);
2704 btrfs_abort_transaction(trans, ret);
2709 mutex_lock(&fs_devices->device_list_mutex);
2711 btrfs_setup_sprout(fs_info, seed_devices);
2712 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev,
2716 device->fs_devices = fs_devices;
2718 mutex_lock(&fs_info->chunk_mutex);
2719 list_add_rcu(&device->dev_list, &fs_devices->devices);
2720 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2721 fs_devices->num_devices++;
2722 fs_devices->open_devices++;
2723 fs_devices->rw_devices++;
2724 fs_devices->total_devices++;
2725 fs_devices->total_rw_bytes += device->total_bytes;
2727 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2729 if (!bdev_nonrot(bdev))
2730 fs_devices->rotating = true;
2732 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2733 btrfs_set_super_total_bytes(fs_info->super_copy,
2734 round_down(orig_super_total_bytes + device->total_bytes,
2735 fs_info->sectorsize));
2737 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2738 btrfs_set_super_num_devices(fs_info->super_copy,
2739 orig_super_num_devices + 1);
2742 * we've got more storage, clear any full flags on the space
2745 btrfs_clear_space_info_full(fs_info);
2747 mutex_unlock(&fs_info->chunk_mutex);
2749 /* Add sysfs device entry */
2750 btrfs_sysfs_add_device(device);
2752 mutex_unlock(&fs_devices->device_list_mutex);
2755 mutex_lock(&fs_info->chunk_mutex);
2756 ret = init_first_rw_device(trans);
2757 mutex_unlock(&fs_info->chunk_mutex);
2759 btrfs_abort_transaction(trans, ret);
2764 ret = btrfs_add_dev_item(trans, device);
2766 btrfs_abort_transaction(trans, ret);
2771 ret = btrfs_finish_sprout(trans);
2773 btrfs_abort_transaction(trans, ret);
2778 * fs_devices now represents the newly sprouted filesystem and
2779 * its fsid has been changed by btrfs_sprout_splice().
2781 btrfs_sysfs_update_sprout_fsid(fs_devices);
2784 ret = btrfs_commit_transaction(trans);
2787 mutex_unlock(&uuid_mutex);
2788 up_write(&sb->s_umount);
2791 if (ret) /* transaction commit */
2794 ret = btrfs_relocate_sys_chunks(fs_info);
2796 btrfs_handle_fs_error(fs_info, ret,
2797 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2798 trans = btrfs_attach_transaction(root);
2799 if (IS_ERR(trans)) {
2800 if (PTR_ERR(trans) == -ENOENT)
2802 ret = PTR_ERR(trans);
2806 ret = btrfs_commit_transaction(trans);
2810 * Now that we have written a new super block to this device, check all
2811 * other fs_devices list if device_path alienates any other scanned
2813 * We can ignore the return value as it typically returns -EINVAL and
2814 * only succeeds if the device was an alien.
2816 btrfs_forget_devices(device->devt);
2818 /* Update ctime/mtime for blkid or udev */
2819 update_dev_time(device_path);
2824 btrfs_sysfs_remove_device(device);
2825 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2826 mutex_lock(&fs_info->chunk_mutex);
2827 list_del_rcu(&device->dev_list);
2828 list_del(&device->dev_alloc_list);
2829 fs_info->fs_devices->num_devices--;
2830 fs_info->fs_devices->open_devices--;
2831 fs_info->fs_devices->rw_devices--;
2832 fs_info->fs_devices->total_devices--;
2833 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2834 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2835 btrfs_set_super_total_bytes(fs_info->super_copy,
2836 orig_super_total_bytes);
2837 btrfs_set_super_num_devices(fs_info->super_copy,
2838 orig_super_num_devices);
2839 mutex_unlock(&fs_info->chunk_mutex);
2840 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2843 btrfs_set_sb_rdonly(sb);
2845 btrfs_end_transaction(trans);
2847 btrfs_destroy_dev_zone_info(device);
2849 btrfs_free_device(device);
2851 blkdev_put(bdev, fs_info->bdev_holder);
2853 mutex_unlock(&uuid_mutex);
2854 up_write(&sb->s_umount);
2859 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2860 struct btrfs_device *device)
2863 struct btrfs_path *path;
2864 struct btrfs_root *root = device->fs_info->chunk_root;
2865 struct btrfs_dev_item *dev_item;
2866 struct extent_buffer *leaf;
2867 struct btrfs_key key;
2869 path = btrfs_alloc_path();
2873 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2874 key.type = BTRFS_DEV_ITEM_KEY;
2875 key.offset = device->devid;
2877 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2886 leaf = path->nodes[0];
2887 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2889 btrfs_set_device_id(leaf, dev_item, device->devid);
2890 btrfs_set_device_type(leaf, dev_item, device->type);
2891 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2892 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2893 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2894 btrfs_set_device_total_bytes(leaf, dev_item,
2895 btrfs_device_get_disk_total_bytes(device));
2896 btrfs_set_device_bytes_used(leaf, dev_item,
2897 btrfs_device_get_bytes_used(device));
2898 btrfs_mark_buffer_dirty(leaf);
2901 btrfs_free_path(path);
2905 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2906 struct btrfs_device *device, u64 new_size)
2908 struct btrfs_fs_info *fs_info = device->fs_info;
2909 struct btrfs_super_block *super_copy = fs_info->super_copy;
2914 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2917 new_size = round_down(new_size, fs_info->sectorsize);
2919 mutex_lock(&fs_info->chunk_mutex);
2920 old_total = btrfs_super_total_bytes(super_copy);
2921 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2923 if (new_size <= device->total_bytes ||
2924 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2925 mutex_unlock(&fs_info->chunk_mutex);
2929 btrfs_set_super_total_bytes(super_copy,
2930 round_down(old_total + diff, fs_info->sectorsize));
2931 device->fs_devices->total_rw_bytes += diff;
2933 btrfs_device_set_total_bytes(device, new_size);
2934 btrfs_device_set_disk_total_bytes(device, new_size);
2935 btrfs_clear_space_info_full(device->fs_info);
2936 if (list_empty(&device->post_commit_list))
2937 list_add_tail(&device->post_commit_list,
2938 &trans->transaction->dev_update_list);
2939 mutex_unlock(&fs_info->chunk_mutex);
2941 btrfs_reserve_chunk_metadata(trans, false);
2942 ret = btrfs_update_device(trans, device);
2943 btrfs_trans_release_chunk_metadata(trans);
2948 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2950 struct btrfs_fs_info *fs_info = trans->fs_info;
2951 struct btrfs_root *root = fs_info->chunk_root;
2953 struct btrfs_path *path;
2954 struct btrfs_key key;
2956 path = btrfs_alloc_path();
2960 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2961 key.offset = chunk_offset;
2962 key.type = BTRFS_CHUNK_ITEM_KEY;
2964 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2967 else if (ret > 0) { /* Logic error or corruption */
2968 btrfs_handle_fs_error(fs_info, -ENOENT,
2969 "Failed lookup while freeing chunk.");
2974 ret = btrfs_del_item(trans, root, path);
2976 btrfs_handle_fs_error(fs_info, ret,
2977 "Failed to delete chunk item.");
2979 btrfs_free_path(path);
2983 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2985 struct btrfs_super_block *super_copy = fs_info->super_copy;
2986 struct btrfs_disk_key *disk_key;
2987 struct btrfs_chunk *chunk;
2994 struct btrfs_key key;
2996 lockdep_assert_held(&fs_info->chunk_mutex);
2997 array_size = btrfs_super_sys_array_size(super_copy);
2999 ptr = super_copy->sys_chunk_array;
3002 while (cur < array_size) {
3003 disk_key = (struct btrfs_disk_key *)ptr;
3004 btrfs_disk_key_to_cpu(&key, disk_key);
3006 len = sizeof(*disk_key);
3008 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3009 chunk = (struct btrfs_chunk *)(ptr + len);
3010 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
3011 len += btrfs_chunk_item_size(num_stripes);
3016 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
3017 key.offset == chunk_offset) {
3018 memmove(ptr, ptr + len, array_size - (cur + len));
3020 btrfs_set_super_sys_array_size(super_copy, array_size);
3030 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
3031 * @logical: Logical block offset in bytes.
3032 * @length: Length of extent in bytes.
3034 * Return: Chunk mapping or ERR_PTR.
3036 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
3037 u64 logical, u64 length)
3039 struct extent_map_tree *em_tree;
3040 struct extent_map *em;
3042 em_tree = &fs_info->mapping_tree;
3043 read_lock(&em_tree->lock);
3044 em = lookup_extent_mapping(em_tree, logical, length);
3045 read_unlock(&em_tree->lock);
3048 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
3050 return ERR_PTR(-EINVAL);
3053 if (em->start > logical || em->start + em->len < logical) {
3055 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
3056 logical, length, em->start, em->start + em->len);
3057 free_extent_map(em);
3058 return ERR_PTR(-EINVAL);
3061 /* callers are responsible for dropping em's ref. */
3065 static int remove_chunk_item(struct btrfs_trans_handle *trans,
3066 struct map_lookup *map, u64 chunk_offset)
3071 * Removing chunk items and updating the device items in the chunks btree
3072 * requires holding the chunk_mutex.
3073 * See the comment at btrfs_chunk_alloc() for the details.
3075 lockdep_assert_held(&trans->fs_info->chunk_mutex);
3077 for (i = 0; i < map->num_stripes; i++) {
3080 ret = btrfs_update_device(trans, map->stripes[i].dev);
3085 return btrfs_free_chunk(trans, chunk_offset);
3088 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
3090 struct btrfs_fs_info *fs_info = trans->fs_info;
3091 struct extent_map *em;
3092 struct map_lookup *map;
3093 u64 dev_extent_len = 0;
3095 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3097 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
3100 * This is a logic error, but we don't want to just rely on the
3101 * user having built with ASSERT enabled, so if ASSERT doesn't
3102 * do anything we still error out.
3107 map = em->map_lookup;
3110 * First delete the device extent items from the devices btree.
3111 * We take the device_list_mutex to avoid racing with the finishing phase
3112 * of a device replace operation. See the comment below before acquiring
3113 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
3114 * because that can result in a deadlock when deleting the device extent
3115 * items from the devices btree - COWing an extent buffer from the btree
3116 * may result in allocating a new metadata chunk, which would attempt to
3117 * lock again fs_info->chunk_mutex.
3119 mutex_lock(&fs_devices->device_list_mutex);
3120 for (i = 0; i < map->num_stripes; i++) {
3121 struct btrfs_device *device = map->stripes[i].dev;
3122 ret = btrfs_free_dev_extent(trans, device,
3123 map->stripes[i].physical,
3126 mutex_unlock(&fs_devices->device_list_mutex);
3127 btrfs_abort_transaction(trans, ret);
3131 if (device->bytes_used > 0) {
3132 mutex_lock(&fs_info->chunk_mutex);
3133 btrfs_device_set_bytes_used(device,
3134 device->bytes_used - dev_extent_len);
3135 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3136 btrfs_clear_space_info_full(fs_info);
3137 mutex_unlock(&fs_info->chunk_mutex);
3140 mutex_unlock(&fs_devices->device_list_mutex);
3143 * We acquire fs_info->chunk_mutex for 2 reasons:
3145 * 1) Just like with the first phase of the chunk allocation, we must
3146 * reserve system space, do all chunk btree updates and deletions, and
3147 * update the system chunk array in the superblock while holding this
3148 * mutex. This is for similar reasons as explained on the comment at
3149 * the top of btrfs_chunk_alloc();
3151 * 2) Prevent races with the final phase of a device replace operation
3152 * that replaces the device object associated with the map's stripes,
3153 * because the device object's id can change at any time during that
3154 * final phase of the device replace operation
3155 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3156 * replaced device and then see it with an ID of
3157 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3158 * the device item, which does not exists on the chunk btree.
3159 * The finishing phase of device replace acquires both the
3160 * device_list_mutex and the chunk_mutex, in that order, so we are
3161 * safe by just acquiring the chunk_mutex.
3163 trans->removing_chunk = true;
3164 mutex_lock(&fs_info->chunk_mutex);
3166 check_system_chunk(trans, map->type);
3168 ret = remove_chunk_item(trans, map, chunk_offset);
3170 * Normally we should not get -ENOSPC since we reserved space before
3171 * through the call to check_system_chunk().
3173 * Despite our system space_info having enough free space, we may not
3174 * be able to allocate extents from its block groups, because all have
3175 * an incompatible profile, which will force us to allocate a new system
3176 * block group with the right profile, or right after we called
3177 * check_system_space() above, a scrub turned the only system block group
3178 * with enough free space into RO mode.
3179 * This is explained with more detail at do_chunk_alloc().
3181 * So if we get -ENOSPC, allocate a new system chunk and retry once.
3183 if (ret == -ENOSPC) {
3184 const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
3185 struct btrfs_block_group *sys_bg;
3187 sys_bg = btrfs_create_chunk(trans, sys_flags);
3188 if (IS_ERR(sys_bg)) {
3189 ret = PTR_ERR(sys_bg);
3190 btrfs_abort_transaction(trans, ret);
3194 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3196 btrfs_abort_transaction(trans, ret);
3200 ret = remove_chunk_item(trans, map, chunk_offset);
3202 btrfs_abort_transaction(trans, ret);
3206 btrfs_abort_transaction(trans, ret);
3210 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3212 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3213 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3215 btrfs_abort_transaction(trans, ret);
3220 mutex_unlock(&fs_info->chunk_mutex);
3221 trans->removing_chunk = false;
3224 * We are done with chunk btree updates and deletions, so release the
3225 * system space we previously reserved (with check_system_chunk()).
3227 btrfs_trans_release_chunk_metadata(trans);
3229 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3231 btrfs_abort_transaction(trans, ret);
3236 if (trans->removing_chunk) {
3237 mutex_unlock(&fs_info->chunk_mutex);
3238 trans->removing_chunk = false;
3241 free_extent_map(em);
3245 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3247 struct btrfs_root *root = fs_info->chunk_root;
3248 struct btrfs_trans_handle *trans;
3249 struct btrfs_block_group *block_group;
3253 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3255 "relocate: not supported on extent tree v2 yet");
3260 * Prevent races with automatic removal of unused block groups.
3261 * After we relocate and before we remove the chunk with offset
3262 * chunk_offset, automatic removal of the block group can kick in,
3263 * resulting in a failure when calling btrfs_remove_chunk() below.
3265 * Make sure to acquire this mutex before doing a tree search (dev
3266 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3267 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3268 * we release the path used to search the chunk/dev tree and before
3269 * the current task acquires this mutex and calls us.
3271 lockdep_assert_held(&fs_info->reclaim_bgs_lock);
3273 /* step one, relocate all the extents inside this chunk */
3274 btrfs_scrub_pause(fs_info);
3275 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3276 btrfs_scrub_continue(fs_info);
3279 * If we had a transaction abort, stop all running scrubs.
3280 * See transaction.c:cleanup_transaction() why we do it here.
3282 if (BTRFS_FS_ERROR(fs_info))
3283 btrfs_scrub_cancel(fs_info);
3287 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3290 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3291 length = block_group->length;
3292 btrfs_put_block_group(block_group);
3295 * On a zoned file system, discard the whole block group, this will
3296 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3297 * resetting the zone fails, don't treat it as a fatal problem from the
3298 * filesystem's point of view.
3300 if (btrfs_is_zoned(fs_info)) {
3301 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
3304 "failed to reset zone %llu after relocation",
3308 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3310 if (IS_ERR(trans)) {
3311 ret = PTR_ERR(trans);
3312 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3317 * step two, delete the device extents and the
3318 * chunk tree entries
3320 ret = btrfs_remove_chunk(trans, chunk_offset);
3321 btrfs_end_transaction(trans);
3325 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3327 struct btrfs_root *chunk_root = fs_info->chunk_root;
3328 struct btrfs_path *path;
3329 struct extent_buffer *leaf;
3330 struct btrfs_chunk *chunk;
3331 struct btrfs_key key;
3332 struct btrfs_key found_key;
3334 bool retried = false;
3338 path = btrfs_alloc_path();
3343 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3344 key.offset = (u64)-1;
3345 key.type = BTRFS_CHUNK_ITEM_KEY;
3348 mutex_lock(&fs_info->reclaim_bgs_lock);
3349 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3351 mutex_unlock(&fs_info->reclaim_bgs_lock);
3354 BUG_ON(ret == 0); /* Corruption */
3356 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3359 mutex_unlock(&fs_info->reclaim_bgs_lock);
3365 leaf = path->nodes[0];
3366 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3368 chunk = btrfs_item_ptr(leaf, path->slots[0],
3369 struct btrfs_chunk);
3370 chunk_type = btrfs_chunk_type(leaf, chunk);
3371 btrfs_release_path(path);
3373 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3374 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3380 mutex_unlock(&fs_info->reclaim_bgs_lock);
3382 if (found_key.offset == 0)
3384 key.offset = found_key.offset - 1;
3387 if (failed && !retried) {
3391 } else if (WARN_ON(failed && retried)) {
3395 btrfs_free_path(path);
3400 * return 1 : allocate a data chunk successfully,
3401 * return <0: errors during allocating a data chunk,
3402 * return 0 : no need to allocate a data chunk.
3404 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3407 struct btrfs_block_group *cache;
3411 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3413 chunk_type = cache->flags;
3414 btrfs_put_block_group(cache);
3416 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3419 spin_lock(&fs_info->data_sinfo->lock);
3420 bytes_used = fs_info->data_sinfo->bytes_used;
3421 spin_unlock(&fs_info->data_sinfo->lock);
3424 struct btrfs_trans_handle *trans;
3427 trans = btrfs_join_transaction(fs_info->tree_root);
3429 return PTR_ERR(trans);
3431 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3432 btrfs_end_transaction(trans);
3441 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3442 struct btrfs_balance_control *bctl)
3444 struct btrfs_root *root = fs_info->tree_root;
3445 struct btrfs_trans_handle *trans;
3446 struct btrfs_balance_item *item;
3447 struct btrfs_disk_balance_args disk_bargs;
3448 struct btrfs_path *path;
3449 struct extent_buffer *leaf;
3450 struct btrfs_key key;
3453 path = btrfs_alloc_path();
3457 trans = btrfs_start_transaction(root, 0);
3458 if (IS_ERR(trans)) {
3459 btrfs_free_path(path);
3460 return PTR_ERR(trans);
3463 key.objectid = BTRFS_BALANCE_OBJECTID;
3464 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3467 ret = btrfs_insert_empty_item(trans, root, path, &key,
3472 leaf = path->nodes[0];
3473 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3475 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3477 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3478 btrfs_set_balance_data(leaf, item, &disk_bargs);
3479 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3480 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3481 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3482 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3484 btrfs_set_balance_flags(leaf, item, bctl->flags);
3486 btrfs_mark_buffer_dirty(leaf);
3488 btrfs_free_path(path);
3489 err = btrfs_commit_transaction(trans);
3495 static int del_balance_item(struct btrfs_fs_info *fs_info)
3497 struct btrfs_root *root = fs_info->tree_root;
3498 struct btrfs_trans_handle *trans;
3499 struct btrfs_path *path;
3500 struct btrfs_key key;
3503 path = btrfs_alloc_path();
3507 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3508 if (IS_ERR(trans)) {
3509 btrfs_free_path(path);
3510 return PTR_ERR(trans);
3513 key.objectid = BTRFS_BALANCE_OBJECTID;
3514 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3517 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3525 ret = btrfs_del_item(trans, root, path);
3527 btrfs_free_path(path);
3528 err = btrfs_commit_transaction(trans);
3535 * This is a heuristic used to reduce the number of chunks balanced on
3536 * resume after balance was interrupted.
3538 static void update_balance_args(struct btrfs_balance_control *bctl)
3541 * Turn on soft mode for chunk types that were being converted.
3543 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3544 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3545 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3546 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3547 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3548 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3551 * Turn on usage filter if is not already used. The idea is
3552 * that chunks that we have already balanced should be
3553 * reasonably full. Don't do it for chunks that are being
3554 * converted - that will keep us from relocating unconverted
3555 * (albeit full) chunks.
3557 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3558 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3559 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3560 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3561 bctl->data.usage = 90;
3563 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3564 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3565 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3566 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3567 bctl->sys.usage = 90;
3569 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3570 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3571 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3572 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3573 bctl->meta.usage = 90;
3578 * Clear the balance status in fs_info and delete the balance item from disk.
3580 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3582 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3585 BUG_ON(!fs_info->balance_ctl);
3587 spin_lock(&fs_info->balance_lock);
3588 fs_info->balance_ctl = NULL;
3589 spin_unlock(&fs_info->balance_lock);
3592 ret = del_balance_item(fs_info);
3594 btrfs_handle_fs_error(fs_info, ret, NULL);
3598 * Balance filters. Return 1 if chunk should be filtered out
3599 * (should not be balanced).
3601 static int chunk_profiles_filter(u64 chunk_type,
3602 struct btrfs_balance_args *bargs)
3604 chunk_type = chunk_to_extended(chunk_type) &
3605 BTRFS_EXTENDED_PROFILE_MASK;
3607 if (bargs->profiles & chunk_type)
3613 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3614 struct btrfs_balance_args *bargs)
3616 struct btrfs_block_group *cache;
3618 u64 user_thresh_min;
3619 u64 user_thresh_max;
3622 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3623 chunk_used = cache->used;
3625 if (bargs->usage_min == 0)
3626 user_thresh_min = 0;
3628 user_thresh_min = mult_perc(cache->length, bargs->usage_min);
3630 if (bargs->usage_max == 0)
3631 user_thresh_max = 1;
3632 else if (bargs->usage_max > 100)
3633 user_thresh_max = cache->length;
3635 user_thresh_max = mult_perc(cache->length, bargs->usage_max);
3637 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3640 btrfs_put_block_group(cache);
3644 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3645 u64 chunk_offset, struct btrfs_balance_args *bargs)
3647 struct btrfs_block_group *cache;
3648 u64 chunk_used, user_thresh;
3651 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3652 chunk_used = cache->used;
3654 if (bargs->usage_min == 0)
3656 else if (bargs->usage > 100)
3657 user_thresh = cache->length;
3659 user_thresh = mult_perc(cache->length, bargs->usage);
3661 if (chunk_used < user_thresh)
3664 btrfs_put_block_group(cache);
3668 static int chunk_devid_filter(struct extent_buffer *leaf,
3669 struct btrfs_chunk *chunk,
3670 struct btrfs_balance_args *bargs)
3672 struct btrfs_stripe *stripe;
3673 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3676 for (i = 0; i < num_stripes; i++) {
3677 stripe = btrfs_stripe_nr(chunk, i);
3678 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3685 static u64 calc_data_stripes(u64 type, int num_stripes)
3687 const int index = btrfs_bg_flags_to_raid_index(type);
3688 const int ncopies = btrfs_raid_array[index].ncopies;
3689 const int nparity = btrfs_raid_array[index].nparity;
3691 return (num_stripes - nparity) / ncopies;
3694 /* [pstart, pend) */
3695 static int chunk_drange_filter(struct extent_buffer *leaf,
3696 struct btrfs_chunk *chunk,
3697 struct btrfs_balance_args *bargs)
3699 struct btrfs_stripe *stripe;
3700 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3707 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3710 type = btrfs_chunk_type(leaf, chunk);
3711 factor = calc_data_stripes(type, num_stripes);
3713 for (i = 0; i < num_stripes; i++) {
3714 stripe = btrfs_stripe_nr(chunk, i);
3715 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3718 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3719 stripe_length = btrfs_chunk_length(leaf, chunk);
3720 stripe_length = div_u64(stripe_length, factor);
3722 if (stripe_offset < bargs->pend &&
3723 stripe_offset + stripe_length > bargs->pstart)
3730 /* [vstart, vend) */
3731 static int chunk_vrange_filter(struct extent_buffer *leaf,
3732 struct btrfs_chunk *chunk,
3734 struct btrfs_balance_args *bargs)
3736 if (chunk_offset < bargs->vend &&
3737 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3738 /* at least part of the chunk is inside this vrange */
3744 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3745 struct btrfs_chunk *chunk,
3746 struct btrfs_balance_args *bargs)
3748 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3750 if (bargs->stripes_min <= num_stripes
3751 && num_stripes <= bargs->stripes_max)
3757 static int chunk_soft_convert_filter(u64 chunk_type,
3758 struct btrfs_balance_args *bargs)
3760 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3763 chunk_type = chunk_to_extended(chunk_type) &
3764 BTRFS_EXTENDED_PROFILE_MASK;
3766 if (bargs->target == chunk_type)
3772 static int should_balance_chunk(struct extent_buffer *leaf,
3773 struct btrfs_chunk *chunk, u64 chunk_offset)
3775 struct btrfs_fs_info *fs_info = leaf->fs_info;
3776 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3777 struct btrfs_balance_args *bargs = NULL;
3778 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3781 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3782 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3786 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3787 bargs = &bctl->data;
3788 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3790 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3791 bargs = &bctl->meta;
3793 /* profiles filter */
3794 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3795 chunk_profiles_filter(chunk_type, bargs)) {
3800 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3801 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3803 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3804 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3809 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3810 chunk_devid_filter(leaf, chunk, bargs)) {
3814 /* drange filter, makes sense only with devid filter */
3815 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3816 chunk_drange_filter(leaf, chunk, bargs)) {
3821 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3822 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3826 /* stripes filter */
3827 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3828 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3832 /* soft profile changing mode */
3833 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3834 chunk_soft_convert_filter(chunk_type, bargs)) {
3839 * limited by count, must be the last filter
3841 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3842 if (bargs->limit == 0)
3846 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3848 * Same logic as the 'limit' filter; the minimum cannot be
3849 * determined here because we do not have the global information
3850 * about the count of all chunks that satisfy the filters.
3852 if (bargs->limit_max == 0)
3861 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3863 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3864 struct btrfs_root *chunk_root = fs_info->chunk_root;
3866 struct btrfs_chunk *chunk;
3867 struct btrfs_path *path = NULL;
3868 struct btrfs_key key;
3869 struct btrfs_key found_key;
3870 struct extent_buffer *leaf;
3873 int enospc_errors = 0;
3874 bool counting = true;
3875 /* The single value limit and min/max limits use the same bytes in the */
3876 u64 limit_data = bctl->data.limit;
3877 u64 limit_meta = bctl->meta.limit;
3878 u64 limit_sys = bctl->sys.limit;
3882 int chunk_reserved = 0;
3884 path = btrfs_alloc_path();
3890 /* zero out stat counters */
3891 spin_lock(&fs_info->balance_lock);
3892 memset(&bctl->stat, 0, sizeof(bctl->stat));
3893 spin_unlock(&fs_info->balance_lock);
3897 * The single value limit and min/max limits use the same bytes
3900 bctl->data.limit = limit_data;
3901 bctl->meta.limit = limit_meta;
3902 bctl->sys.limit = limit_sys;
3904 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3905 key.offset = (u64)-1;
3906 key.type = BTRFS_CHUNK_ITEM_KEY;
3909 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3910 atomic_read(&fs_info->balance_cancel_req)) {
3915 mutex_lock(&fs_info->reclaim_bgs_lock);
3916 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3918 mutex_unlock(&fs_info->reclaim_bgs_lock);
3923 * this shouldn't happen, it means the last relocate
3927 BUG(); /* FIXME break ? */
3929 ret = btrfs_previous_item(chunk_root, path, 0,
3930 BTRFS_CHUNK_ITEM_KEY);
3932 mutex_unlock(&fs_info->reclaim_bgs_lock);
3937 leaf = path->nodes[0];
3938 slot = path->slots[0];
3939 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3941 if (found_key.objectid != key.objectid) {
3942 mutex_unlock(&fs_info->reclaim_bgs_lock);
3946 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3947 chunk_type = btrfs_chunk_type(leaf, chunk);
3950 spin_lock(&fs_info->balance_lock);
3951 bctl->stat.considered++;
3952 spin_unlock(&fs_info->balance_lock);
3955 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3957 btrfs_release_path(path);
3959 mutex_unlock(&fs_info->reclaim_bgs_lock);
3964 mutex_unlock(&fs_info->reclaim_bgs_lock);
3965 spin_lock(&fs_info->balance_lock);
3966 bctl->stat.expected++;
3967 spin_unlock(&fs_info->balance_lock);
3969 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3971 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3973 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3980 * Apply limit_min filter, no need to check if the LIMITS
3981 * filter is used, limit_min is 0 by default
3983 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3984 count_data < bctl->data.limit_min)
3985 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3986 count_meta < bctl->meta.limit_min)
3987 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3988 count_sys < bctl->sys.limit_min)) {
3989 mutex_unlock(&fs_info->reclaim_bgs_lock);
3993 if (!chunk_reserved) {
3995 * We may be relocating the only data chunk we have,
3996 * which could potentially end up with losing data's
3997 * raid profile, so lets allocate an empty one in
4000 ret = btrfs_may_alloc_data_chunk(fs_info,
4003 mutex_unlock(&fs_info->reclaim_bgs_lock);
4005 } else if (ret == 1) {
4010 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
4011 mutex_unlock(&fs_info->reclaim_bgs_lock);
4012 if (ret == -ENOSPC) {
4014 } else if (ret == -ETXTBSY) {
4016 "skipping relocation of block group %llu due to active swapfile",
4022 spin_lock(&fs_info->balance_lock);
4023 bctl->stat.completed++;
4024 spin_unlock(&fs_info->balance_lock);
4027 if (found_key.offset == 0)
4029 key.offset = found_key.offset - 1;
4033 btrfs_release_path(path);
4038 btrfs_free_path(path);
4039 if (enospc_errors) {
4040 btrfs_info(fs_info, "%d enospc errors during balance",
4050 * See if a given profile is valid and reduced.
4052 * @flags: profile to validate
4053 * @extended: if true @flags is treated as an extended profile
4055 static int alloc_profile_is_valid(u64 flags, int extended)
4057 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
4058 BTRFS_BLOCK_GROUP_PROFILE_MASK);
4060 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
4062 /* 1) check that all other bits are zeroed */
4066 /* 2) see if profile is reduced */
4068 return !extended; /* "0" is valid for usual profiles */
4070 return has_single_bit_set(flags);
4074 * Validate target profile against allowed profiles and return true if it's OK.
4075 * Otherwise print the error message and return false.
4077 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
4078 const struct btrfs_balance_args *bargs,
4079 u64 allowed, const char *type)
4081 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
4084 /* Profile is valid and does not have bits outside of the allowed set */
4085 if (alloc_profile_is_valid(bargs->target, 1) &&
4086 (bargs->target & ~allowed) == 0)
4089 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
4090 type, btrfs_bg_type_to_raid_name(bargs->target));
4095 * Fill @buf with textual description of balance filter flags @bargs, up to
4096 * @size_buf including the terminating null. The output may be trimmed if it
4097 * does not fit into the provided buffer.
4099 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
4103 u32 size_bp = size_buf;
4105 u64 flags = bargs->flags;
4106 char tmp_buf[128] = {'\0'};
4111 #define CHECK_APPEND_NOARG(a) \
4113 ret = snprintf(bp, size_bp, (a)); \
4114 if (ret < 0 || ret >= size_bp) \
4115 goto out_overflow; \
4120 #define CHECK_APPEND_1ARG(a, v1) \
4122 ret = snprintf(bp, size_bp, (a), (v1)); \
4123 if (ret < 0 || ret >= size_bp) \
4124 goto out_overflow; \
4129 #define CHECK_APPEND_2ARG(a, v1, v2) \
4131 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4132 if (ret < 0 || ret >= size_bp) \
4133 goto out_overflow; \
4138 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
4139 CHECK_APPEND_1ARG("convert=%s,",
4140 btrfs_bg_type_to_raid_name(bargs->target));
4142 if (flags & BTRFS_BALANCE_ARGS_SOFT)
4143 CHECK_APPEND_NOARG("soft,");
4145 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4146 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4148 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4151 if (flags & BTRFS_BALANCE_ARGS_USAGE)
4152 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4154 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4155 CHECK_APPEND_2ARG("usage=%u..%u,",
4156 bargs->usage_min, bargs->usage_max);
4158 if (flags & BTRFS_BALANCE_ARGS_DEVID)
4159 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4161 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4162 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4163 bargs->pstart, bargs->pend);
4165 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4166 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4167 bargs->vstart, bargs->vend);
4169 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4170 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4172 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4173 CHECK_APPEND_2ARG("limit=%u..%u,",
4174 bargs->limit_min, bargs->limit_max);
4176 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4177 CHECK_APPEND_2ARG("stripes=%u..%u,",
4178 bargs->stripes_min, bargs->stripes_max);
4180 #undef CHECK_APPEND_2ARG
4181 #undef CHECK_APPEND_1ARG
4182 #undef CHECK_APPEND_NOARG
4186 if (size_bp < size_buf)
4187 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4192 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4194 u32 size_buf = 1024;
4195 char tmp_buf[192] = {'\0'};
4198 u32 size_bp = size_buf;
4200 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4202 buf = kzalloc(size_buf, GFP_KERNEL);
4208 #define CHECK_APPEND_1ARG(a, v1) \
4210 ret = snprintf(bp, size_bp, (a), (v1)); \
4211 if (ret < 0 || ret >= size_bp) \
4212 goto out_overflow; \
4217 if (bctl->flags & BTRFS_BALANCE_FORCE)
4218 CHECK_APPEND_1ARG("%s", "-f ");
4220 if (bctl->flags & BTRFS_BALANCE_DATA) {
4221 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4222 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4225 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4226 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4227 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4230 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4231 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4232 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4235 #undef CHECK_APPEND_1ARG
4239 if (size_bp < size_buf)
4240 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4241 btrfs_info(fs_info, "balance: %s %s",
4242 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4243 "resume" : "start", buf);
4249 * Should be called with balance mutexe held
4251 int btrfs_balance(struct btrfs_fs_info *fs_info,
4252 struct btrfs_balance_control *bctl,
4253 struct btrfs_ioctl_balance_args *bargs)
4255 u64 meta_target, data_target;
4261 bool reducing_redundancy;
4262 bool paused = false;
4265 if (btrfs_fs_closing(fs_info) ||
4266 atomic_read(&fs_info->balance_pause_req) ||
4267 btrfs_should_cancel_balance(fs_info)) {
4272 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4273 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4277 * In case of mixed groups both data and meta should be picked,
4278 * and identical options should be given for both of them.
4280 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4281 if (mixed && (bctl->flags & allowed)) {
4282 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4283 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4284 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4286 "balance: mixed groups data and metadata options must be the same");
4293 * rw_devices will not change at the moment, device add/delete/replace
4296 num_devices = fs_info->fs_devices->rw_devices;
4299 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4300 * special bit for it, to make it easier to distinguish. Thus we need
4301 * to set it manually, or balance would refuse the profile.
4303 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4304 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4305 if (num_devices >= btrfs_raid_array[i].devs_min)
4306 allowed |= btrfs_raid_array[i].bg_flag;
4308 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4309 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4310 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4316 * Allow to reduce metadata or system integrity only if force set for
4317 * profiles with redundancy (copies, parity)
4320 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4321 if (btrfs_raid_array[i].ncopies >= 2 ||
4322 btrfs_raid_array[i].tolerated_failures >= 1)
4323 allowed |= btrfs_raid_array[i].bg_flag;
4326 seq = read_seqbegin(&fs_info->profiles_lock);
4328 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4329 (fs_info->avail_system_alloc_bits & allowed) &&
4330 !(bctl->sys.target & allowed)) ||
4331 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4332 (fs_info->avail_metadata_alloc_bits & allowed) &&
4333 !(bctl->meta.target & allowed)))
4334 reducing_redundancy = true;
4336 reducing_redundancy = false;
4338 /* if we're not converting, the target field is uninitialized */
4339 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4340 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4341 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4342 bctl->data.target : fs_info->avail_data_alloc_bits;
4343 } while (read_seqretry(&fs_info->profiles_lock, seq));
4345 if (reducing_redundancy) {
4346 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4348 "balance: force reducing metadata redundancy");
4351 "balance: reduces metadata redundancy, use --force if you want this");
4357 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4358 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4360 "balance: metadata profile %s has lower redundancy than data profile %s",
4361 btrfs_bg_type_to_raid_name(meta_target),
4362 btrfs_bg_type_to_raid_name(data_target));
4365 ret = insert_balance_item(fs_info, bctl);
4366 if (ret && ret != -EEXIST)
4369 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4370 BUG_ON(ret == -EEXIST);
4371 BUG_ON(fs_info->balance_ctl);
4372 spin_lock(&fs_info->balance_lock);
4373 fs_info->balance_ctl = bctl;
4374 spin_unlock(&fs_info->balance_lock);
4376 BUG_ON(ret != -EEXIST);
4377 spin_lock(&fs_info->balance_lock);
4378 update_balance_args(bctl);
4379 spin_unlock(&fs_info->balance_lock);
4382 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4383 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4384 describe_balance_start_or_resume(fs_info);
4385 mutex_unlock(&fs_info->balance_mutex);
4387 ret = __btrfs_balance(fs_info);
4389 mutex_lock(&fs_info->balance_mutex);
4390 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) {
4391 btrfs_info(fs_info, "balance: paused");
4392 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
4396 * Balance can be canceled by:
4398 * - Regular cancel request
4399 * Then ret == -ECANCELED and balance_cancel_req > 0
4401 * - Fatal signal to "btrfs" process
4402 * Either the signal caught by wait_reserve_ticket() and callers
4403 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4405 * Either way, in this case balance_cancel_req = 0, and
4406 * ret == -EINTR or ret == -ECANCELED.
4408 * So here we only check the return value to catch canceled balance.
4410 else if (ret == -ECANCELED || ret == -EINTR)
4411 btrfs_info(fs_info, "balance: canceled");
4413 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4415 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4418 memset(bargs, 0, sizeof(*bargs));
4419 btrfs_update_ioctl_balance_args(fs_info, bargs);
4422 /* We didn't pause, we can clean everything up. */
4424 reset_balance_state(fs_info);
4425 btrfs_exclop_finish(fs_info);
4428 wake_up(&fs_info->balance_wait_q);
4432 if (bctl->flags & BTRFS_BALANCE_RESUME)
4433 reset_balance_state(fs_info);
4436 btrfs_exclop_finish(fs_info);
4441 static int balance_kthread(void *data)
4443 struct btrfs_fs_info *fs_info = data;
4446 sb_start_write(fs_info->sb);
4447 mutex_lock(&fs_info->balance_mutex);
4448 if (fs_info->balance_ctl)
4449 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4450 mutex_unlock(&fs_info->balance_mutex);
4451 sb_end_write(fs_info->sb);
4456 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4458 struct task_struct *tsk;
4460 mutex_lock(&fs_info->balance_mutex);
4461 if (!fs_info->balance_ctl) {
4462 mutex_unlock(&fs_info->balance_mutex);
4465 mutex_unlock(&fs_info->balance_mutex);
4467 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4468 btrfs_info(fs_info, "balance: resume skipped");
4472 spin_lock(&fs_info->super_lock);
4473 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
4474 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;
4475 spin_unlock(&fs_info->super_lock);
4477 * A ro->rw remount sequence should continue with the paused balance
4478 * regardless of who pauses it, system or the user as of now, so set
4481 spin_lock(&fs_info->balance_lock);
4482 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4483 spin_unlock(&fs_info->balance_lock);
4485 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4486 return PTR_ERR_OR_ZERO(tsk);
4489 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4491 struct btrfs_balance_control *bctl;
4492 struct btrfs_balance_item *item;
4493 struct btrfs_disk_balance_args disk_bargs;
4494 struct btrfs_path *path;
4495 struct extent_buffer *leaf;
4496 struct btrfs_key key;
4499 path = btrfs_alloc_path();
4503 key.objectid = BTRFS_BALANCE_OBJECTID;
4504 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4507 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4510 if (ret > 0) { /* ret = -ENOENT; */
4515 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4521 leaf = path->nodes[0];
4522 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4524 bctl->flags = btrfs_balance_flags(leaf, item);
4525 bctl->flags |= BTRFS_BALANCE_RESUME;
4527 btrfs_balance_data(leaf, item, &disk_bargs);
4528 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4529 btrfs_balance_meta(leaf, item, &disk_bargs);
4530 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4531 btrfs_balance_sys(leaf, item, &disk_bargs);
4532 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4535 * This should never happen, as the paused balance state is recovered
4536 * during mount without any chance of other exclusive ops to collide.
4538 * This gives the exclusive op status to balance and keeps in paused
4539 * state until user intervention (cancel or umount). If the ownership
4540 * cannot be assigned, show a message but do not fail. The balance
4541 * is in a paused state and must have fs_info::balance_ctl properly
4544 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED))
4546 "balance: cannot set exclusive op status, resume manually");
4548 btrfs_release_path(path);
4550 mutex_lock(&fs_info->balance_mutex);
4551 BUG_ON(fs_info->balance_ctl);
4552 spin_lock(&fs_info->balance_lock);
4553 fs_info->balance_ctl = bctl;
4554 spin_unlock(&fs_info->balance_lock);
4555 mutex_unlock(&fs_info->balance_mutex);
4557 btrfs_free_path(path);
4561 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4565 mutex_lock(&fs_info->balance_mutex);
4566 if (!fs_info->balance_ctl) {
4567 mutex_unlock(&fs_info->balance_mutex);
4571 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4572 atomic_inc(&fs_info->balance_pause_req);
4573 mutex_unlock(&fs_info->balance_mutex);
4575 wait_event(fs_info->balance_wait_q,
4576 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4578 mutex_lock(&fs_info->balance_mutex);
4579 /* we are good with balance_ctl ripped off from under us */
4580 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4581 atomic_dec(&fs_info->balance_pause_req);
4586 mutex_unlock(&fs_info->balance_mutex);
4590 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4592 mutex_lock(&fs_info->balance_mutex);
4593 if (!fs_info->balance_ctl) {
4594 mutex_unlock(&fs_info->balance_mutex);
4599 * A paused balance with the item stored on disk can be resumed at
4600 * mount time if the mount is read-write. Otherwise it's still paused
4601 * and we must not allow cancelling as it deletes the item.
4603 if (sb_rdonly(fs_info->sb)) {
4604 mutex_unlock(&fs_info->balance_mutex);
4608 atomic_inc(&fs_info->balance_cancel_req);
4610 * if we are running just wait and return, balance item is
4611 * deleted in btrfs_balance in this case
4613 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4614 mutex_unlock(&fs_info->balance_mutex);
4615 wait_event(fs_info->balance_wait_q,
4616 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4617 mutex_lock(&fs_info->balance_mutex);
4619 mutex_unlock(&fs_info->balance_mutex);
4621 * Lock released to allow other waiters to continue, we'll
4622 * reexamine the status again.
4624 mutex_lock(&fs_info->balance_mutex);
4626 if (fs_info->balance_ctl) {
4627 reset_balance_state(fs_info);
4628 btrfs_exclop_finish(fs_info);
4629 btrfs_info(fs_info, "balance: canceled");
4633 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4634 atomic_dec(&fs_info->balance_cancel_req);
4635 mutex_unlock(&fs_info->balance_mutex);
4639 int btrfs_uuid_scan_kthread(void *data)
4641 struct btrfs_fs_info *fs_info = data;
4642 struct btrfs_root *root = fs_info->tree_root;
4643 struct btrfs_key key;
4644 struct btrfs_path *path = NULL;
4646 struct extent_buffer *eb;
4648 struct btrfs_root_item root_item;
4650 struct btrfs_trans_handle *trans = NULL;
4651 bool closing = false;
4653 path = btrfs_alloc_path();
4660 key.type = BTRFS_ROOT_ITEM_KEY;
4664 if (btrfs_fs_closing(fs_info)) {
4668 ret = btrfs_search_forward(root, &key, path,
4669 BTRFS_OLDEST_GENERATION);
4676 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4677 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4678 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4679 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4682 eb = path->nodes[0];
4683 slot = path->slots[0];
4684 item_size = btrfs_item_size(eb, slot);
4685 if (item_size < sizeof(root_item))
4688 read_extent_buffer(eb, &root_item,
4689 btrfs_item_ptr_offset(eb, slot),
4690 (int)sizeof(root_item));
4691 if (btrfs_root_refs(&root_item) == 0)
4694 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4695 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4699 btrfs_release_path(path);
4701 * 1 - subvol uuid item
4702 * 1 - received_subvol uuid item
4704 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4705 if (IS_ERR(trans)) {
4706 ret = PTR_ERR(trans);
4714 btrfs_release_path(path);
4715 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4716 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4717 BTRFS_UUID_KEY_SUBVOL,
4720 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4726 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4727 ret = btrfs_uuid_tree_add(trans,
4728 root_item.received_uuid,
4729 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4732 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4739 btrfs_release_path(path);
4741 ret = btrfs_end_transaction(trans);
4747 if (key.offset < (u64)-1) {
4749 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4751 key.type = BTRFS_ROOT_ITEM_KEY;
4752 } else if (key.objectid < (u64)-1) {
4754 key.type = BTRFS_ROOT_ITEM_KEY;
4763 btrfs_free_path(path);
4764 if (trans && !IS_ERR(trans))
4765 btrfs_end_transaction(trans);
4767 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4769 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4770 up(&fs_info->uuid_tree_rescan_sem);
4774 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4776 struct btrfs_trans_handle *trans;
4777 struct btrfs_root *tree_root = fs_info->tree_root;
4778 struct btrfs_root *uuid_root;
4779 struct task_struct *task;
4786 trans = btrfs_start_transaction(tree_root, 2);
4788 return PTR_ERR(trans);
4790 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4791 if (IS_ERR(uuid_root)) {
4792 ret = PTR_ERR(uuid_root);
4793 btrfs_abort_transaction(trans, ret);
4794 btrfs_end_transaction(trans);
4798 fs_info->uuid_root = uuid_root;
4800 ret = btrfs_commit_transaction(trans);
4804 down(&fs_info->uuid_tree_rescan_sem);
4805 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4807 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4808 btrfs_warn(fs_info, "failed to start uuid_scan task");
4809 up(&fs_info->uuid_tree_rescan_sem);
4810 return PTR_ERR(task);
4817 * shrinking a device means finding all of the device extents past
4818 * the new size, and then following the back refs to the chunks.
4819 * The chunk relocation code actually frees the device extent
4821 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4823 struct btrfs_fs_info *fs_info = device->fs_info;
4824 struct btrfs_root *root = fs_info->dev_root;
4825 struct btrfs_trans_handle *trans;
4826 struct btrfs_dev_extent *dev_extent = NULL;
4827 struct btrfs_path *path;
4833 bool retried = false;
4834 struct extent_buffer *l;
4835 struct btrfs_key key;
4836 struct btrfs_super_block *super_copy = fs_info->super_copy;
4837 u64 old_total = btrfs_super_total_bytes(super_copy);
4838 u64 old_size = btrfs_device_get_total_bytes(device);
4842 new_size = round_down(new_size, fs_info->sectorsize);
4844 diff = round_down(old_size - new_size, fs_info->sectorsize);
4846 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4849 path = btrfs_alloc_path();
4853 path->reada = READA_BACK;
4855 trans = btrfs_start_transaction(root, 0);
4856 if (IS_ERR(trans)) {
4857 btrfs_free_path(path);
4858 return PTR_ERR(trans);
4861 mutex_lock(&fs_info->chunk_mutex);
4863 btrfs_device_set_total_bytes(device, new_size);
4864 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4865 device->fs_devices->total_rw_bytes -= diff;
4866 atomic64_sub(diff, &fs_info->free_chunk_space);
4870 * Once the device's size has been set to the new size, ensure all
4871 * in-memory chunks are synced to disk so that the loop below sees them
4872 * and relocates them accordingly.
4874 if (contains_pending_extent(device, &start, diff)) {
4875 mutex_unlock(&fs_info->chunk_mutex);
4876 ret = btrfs_commit_transaction(trans);
4880 mutex_unlock(&fs_info->chunk_mutex);
4881 btrfs_end_transaction(trans);
4885 key.objectid = device->devid;
4886 key.offset = (u64)-1;
4887 key.type = BTRFS_DEV_EXTENT_KEY;
4890 mutex_lock(&fs_info->reclaim_bgs_lock);
4891 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4893 mutex_unlock(&fs_info->reclaim_bgs_lock);
4897 ret = btrfs_previous_item(root, path, 0, key.type);
4899 mutex_unlock(&fs_info->reclaim_bgs_lock);
4903 btrfs_release_path(path);
4908 slot = path->slots[0];
4909 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4911 if (key.objectid != device->devid) {
4912 mutex_unlock(&fs_info->reclaim_bgs_lock);
4913 btrfs_release_path(path);
4917 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4918 length = btrfs_dev_extent_length(l, dev_extent);
4920 if (key.offset + length <= new_size) {
4921 mutex_unlock(&fs_info->reclaim_bgs_lock);
4922 btrfs_release_path(path);
4926 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4927 btrfs_release_path(path);
4930 * We may be relocating the only data chunk we have,
4931 * which could potentially end up with losing data's
4932 * raid profile, so lets allocate an empty one in
4935 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4937 mutex_unlock(&fs_info->reclaim_bgs_lock);
4941 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4942 mutex_unlock(&fs_info->reclaim_bgs_lock);
4943 if (ret == -ENOSPC) {
4946 if (ret == -ETXTBSY) {
4948 "could not shrink block group %llu due to active swapfile",
4953 } while (key.offset-- > 0);
4955 if (failed && !retried) {
4959 } else if (failed && retried) {
4964 /* Shrinking succeeded, else we would be at "done". */
4965 trans = btrfs_start_transaction(root, 0);
4966 if (IS_ERR(trans)) {
4967 ret = PTR_ERR(trans);
4971 mutex_lock(&fs_info->chunk_mutex);
4972 /* Clear all state bits beyond the shrunk device size */
4973 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
4976 btrfs_device_set_disk_total_bytes(device, new_size);
4977 if (list_empty(&device->post_commit_list))
4978 list_add_tail(&device->post_commit_list,
4979 &trans->transaction->dev_update_list);
4981 WARN_ON(diff > old_total);
4982 btrfs_set_super_total_bytes(super_copy,
4983 round_down(old_total - diff, fs_info->sectorsize));
4984 mutex_unlock(&fs_info->chunk_mutex);
4986 btrfs_reserve_chunk_metadata(trans, false);
4987 /* Now btrfs_update_device() will change the on-disk size. */
4988 ret = btrfs_update_device(trans, device);
4989 btrfs_trans_release_chunk_metadata(trans);
4991 btrfs_abort_transaction(trans, ret);
4992 btrfs_end_transaction(trans);
4994 ret = btrfs_commit_transaction(trans);
4997 btrfs_free_path(path);
4999 mutex_lock(&fs_info->chunk_mutex);
5000 btrfs_device_set_total_bytes(device, old_size);
5001 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
5002 device->fs_devices->total_rw_bytes += diff;
5003 atomic64_add(diff, &fs_info->free_chunk_space);
5004 mutex_unlock(&fs_info->chunk_mutex);
5009 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
5010 struct btrfs_key *key,
5011 struct btrfs_chunk *chunk, int item_size)
5013 struct btrfs_super_block *super_copy = fs_info->super_copy;
5014 struct btrfs_disk_key disk_key;
5018 lockdep_assert_held(&fs_info->chunk_mutex);
5020 array_size = btrfs_super_sys_array_size(super_copy);
5021 if (array_size + item_size + sizeof(disk_key)
5022 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
5025 ptr = super_copy->sys_chunk_array + array_size;
5026 btrfs_cpu_key_to_disk(&disk_key, key);
5027 memcpy(ptr, &disk_key, sizeof(disk_key));
5028 ptr += sizeof(disk_key);
5029 memcpy(ptr, chunk, item_size);
5030 item_size += sizeof(disk_key);
5031 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
5037 * sort the devices in descending order by max_avail, total_avail
5039 static int btrfs_cmp_device_info(const void *a, const void *b)
5041 const struct btrfs_device_info *di_a = a;
5042 const struct btrfs_device_info *di_b = b;
5044 if (di_a->max_avail > di_b->max_avail)
5046 if (di_a->max_avail < di_b->max_avail)
5048 if (di_a->total_avail > di_b->total_avail)
5050 if (di_a->total_avail < di_b->total_avail)
5055 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
5057 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5060 btrfs_set_fs_incompat(info, RAID56);
5063 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
5065 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
5068 btrfs_set_fs_incompat(info, RAID1C34);
5072 * Structure used internally for btrfs_create_chunk() function.
5073 * Wraps needed parameters.
5075 struct alloc_chunk_ctl {
5078 /* Total number of stripes to allocate */
5080 /* sub_stripes info for map */
5082 /* Stripes per device */
5084 /* Maximum number of devices to use */
5086 /* Minimum number of devices to use */
5088 /* ndevs has to be a multiple of this */
5090 /* Number of copies */
5092 /* Number of stripes worth of bytes to store parity information */
5094 u64 max_stripe_size;
5102 static void init_alloc_chunk_ctl_policy_regular(
5103 struct btrfs_fs_devices *fs_devices,
5104 struct alloc_chunk_ctl *ctl)
5106 struct btrfs_space_info *space_info;
5108 space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type);
5111 ctl->max_chunk_size = READ_ONCE(space_info->chunk_size);
5112 ctl->max_stripe_size = ctl->max_chunk_size;
5114 if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM)
5115 ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK);
5117 /* We don't want a chunk larger than 10% of writable space */
5118 ctl->max_chunk_size = min(mult_perc(fs_devices->total_rw_bytes, 10),
5119 ctl->max_chunk_size);
5120 ctl->dev_extent_min = btrfs_stripe_nr_to_offset(ctl->dev_stripes);
5123 static void init_alloc_chunk_ctl_policy_zoned(
5124 struct btrfs_fs_devices *fs_devices,
5125 struct alloc_chunk_ctl *ctl)
5127 u64 zone_size = fs_devices->fs_info->zone_size;
5129 int min_num_stripes = ctl->devs_min * ctl->dev_stripes;
5130 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;
5131 u64 min_chunk_size = min_data_stripes * zone_size;
5132 u64 type = ctl->type;
5134 ctl->max_stripe_size = zone_size;
5135 if (type & BTRFS_BLOCK_GROUP_DATA) {
5136 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,
5138 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5139 ctl->max_chunk_size = ctl->max_stripe_size;
5140 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5141 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5142 ctl->devs_max = min_t(int, ctl->devs_max,
5143 BTRFS_MAX_DEVS_SYS_CHUNK);
5148 /* We don't want a chunk larger than 10% of writable space */
5149 limit = max(round_down(mult_perc(fs_devices->total_rw_bytes, 10),
5152 ctl->max_chunk_size = min(limit, ctl->max_chunk_size);
5153 ctl->dev_extent_min = zone_size * ctl->dev_stripes;
5156 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
5157 struct alloc_chunk_ctl *ctl)
5159 int index = btrfs_bg_flags_to_raid_index(ctl->type);
5161 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
5162 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
5163 ctl->devs_max = btrfs_raid_array[index].devs_max;
5165 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
5166 ctl->devs_min = btrfs_raid_array[index].devs_min;
5167 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
5168 ctl->ncopies = btrfs_raid_array[index].ncopies;
5169 ctl->nparity = btrfs_raid_array[index].nparity;
5172 switch (fs_devices->chunk_alloc_policy) {
5173 case BTRFS_CHUNK_ALLOC_REGULAR:
5174 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
5176 case BTRFS_CHUNK_ALLOC_ZONED:
5177 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);
5184 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
5185 struct alloc_chunk_ctl *ctl,
5186 struct btrfs_device_info *devices_info)
5188 struct btrfs_fs_info *info = fs_devices->fs_info;
5189 struct btrfs_device *device;
5191 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
5198 * in the first pass through the devices list, we gather information
5199 * about the available holes on each device.
5201 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5202 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5204 "BTRFS: read-only device in alloc_list\n");
5208 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5209 &device->dev_state) ||
5210 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5213 if (device->total_bytes > device->bytes_used)
5214 total_avail = device->total_bytes - device->bytes_used;
5218 /* If there is no space on this device, skip it. */
5219 if (total_avail < ctl->dev_extent_min)
5222 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
5224 if (ret && ret != -ENOSPC)
5228 max_avail = dev_extent_want;
5230 if (max_avail < ctl->dev_extent_min) {
5231 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5233 "%s: devid %llu has no free space, have=%llu want=%llu",
5234 __func__, device->devid, max_avail,
5235 ctl->dev_extent_min);
5239 if (ndevs == fs_devices->rw_devices) {
5240 WARN(1, "%s: found more than %llu devices\n",
5241 __func__, fs_devices->rw_devices);
5244 devices_info[ndevs].dev_offset = dev_offset;
5245 devices_info[ndevs].max_avail = max_avail;
5246 devices_info[ndevs].total_avail = total_avail;
5247 devices_info[ndevs].dev = device;
5253 * now sort the devices by hole size / available space
5255 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5256 btrfs_cmp_device_info, NULL);
5261 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5262 struct btrfs_device_info *devices_info)
5264 /* Number of stripes that count for block group size */
5268 * The primary goal is to maximize the number of stripes, so use as
5269 * many devices as possible, even if the stripes are not maximum sized.
5271 * The DUP profile stores more than one stripe per device, the
5272 * max_avail is the total size so we have to adjust.
5274 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5276 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5278 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5279 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5282 * Use the number of data stripes to figure out how big this chunk is
5283 * really going to be in terms of logical address space, and compare
5284 * that answer with the max chunk size. If it's higher, we try to
5285 * reduce stripe_size.
5287 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5289 * Reduce stripe_size, round it up to a 16MB boundary again and
5290 * then use it, unless it ends up being even bigger than the
5291 * previous value we had already.
5293 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5294 data_stripes), SZ_16M),
5298 /* Stripe size should not go beyond 1G. */
5299 ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G);
5301 /* Align to BTRFS_STRIPE_LEN */
5302 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5303 ctl->chunk_size = ctl->stripe_size * data_stripes;
5308 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,
5309 struct btrfs_device_info *devices_info)
5311 u64 zone_size = devices_info[0].dev->zone_info->zone_size;
5312 /* Number of stripes that count for block group size */
5316 * It should hold because:
5317 * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5319 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);
5321 ctl->stripe_size = zone_size;
5322 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5323 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5325 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5326 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5327 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,
5328 ctl->stripe_size) + ctl->nparity,
5330 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5331 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5332 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);
5335 ctl->chunk_size = ctl->stripe_size * data_stripes;
5340 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5341 struct alloc_chunk_ctl *ctl,
5342 struct btrfs_device_info *devices_info)
5344 struct btrfs_fs_info *info = fs_devices->fs_info;
5347 * Round down to number of usable stripes, devs_increment can be any
5348 * number so we can't use round_down() that requires power of 2, while
5349 * rounddown is safe.
5351 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5353 if (ctl->ndevs < ctl->devs_min) {
5354 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5356 "%s: not enough devices with free space: have=%d minimum required=%d",
5357 __func__, ctl->ndevs, ctl->devs_min);
5362 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5364 switch (fs_devices->chunk_alloc_policy) {
5365 case BTRFS_CHUNK_ALLOC_REGULAR:
5366 return decide_stripe_size_regular(ctl, devices_info);
5367 case BTRFS_CHUNK_ALLOC_ZONED:
5368 return decide_stripe_size_zoned(ctl, devices_info);
5374 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
5375 struct alloc_chunk_ctl *ctl,
5376 struct btrfs_device_info *devices_info)
5378 struct btrfs_fs_info *info = trans->fs_info;
5379 struct map_lookup *map = NULL;
5380 struct extent_map_tree *em_tree;
5381 struct btrfs_block_group *block_group;
5382 struct extent_map *em;
5383 u64 start = ctl->start;
5384 u64 type = ctl->type;
5389 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
5391 return ERR_PTR(-ENOMEM);
5392 map->num_stripes = ctl->num_stripes;
5394 for (i = 0; i < ctl->ndevs; ++i) {
5395 for (j = 0; j < ctl->dev_stripes; ++j) {
5396 int s = i * ctl->dev_stripes + j;
5397 map->stripes[s].dev = devices_info[i].dev;
5398 map->stripes[s].physical = devices_info[i].dev_offset +
5399 j * ctl->stripe_size;
5402 map->io_align = BTRFS_STRIPE_LEN;
5403 map->io_width = BTRFS_STRIPE_LEN;
5405 map->sub_stripes = ctl->sub_stripes;
5407 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5409 em = alloc_extent_map();
5412 return ERR_PTR(-ENOMEM);
5414 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5415 em->map_lookup = map;
5417 em->len = ctl->chunk_size;
5418 em->block_start = 0;
5419 em->block_len = em->len;
5420 em->orig_block_len = ctl->stripe_size;
5422 em_tree = &info->mapping_tree;
5423 write_lock(&em_tree->lock);
5424 ret = add_extent_mapping(em_tree, em, 0);
5426 write_unlock(&em_tree->lock);
5427 free_extent_map(em);
5428 return ERR_PTR(ret);
5430 write_unlock(&em_tree->lock);
5432 block_group = btrfs_make_block_group(trans, type, start, ctl->chunk_size);
5433 if (IS_ERR(block_group))
5434 goto error_del_extent;
5436 for (i = 0; i < map->num_stripes; i++) {
5437 struct btrfs_device *dev = map->stripes[i].dev;
5439 btrfs_device_set_bytes_used(dev,
5440 dev->bytes_used + ctl->stripe_size);
5441 if (list_empty(&dev->post_commit_list))
5442 list_add_tail(&dev->post_commit_list,
5443 &trans->transaction->dev_update_list);
5446 atomic64_sub(ctl->stripe_size * map->num_stripes,
5447 &info->free_chunk_space);
5449 free_extent_map(em);
5450 check_raid56_incompat_flag(info, type);
5451 check_raid1c34_incompat_flag(info, type);
5456 write_lock(&em_tree->lock);
5457 remove_extent_mapping(em_tree, em);
5458 write_unlock(&em_tree->lock);
5460 /* One for our allocation */
5461 free_extent_map(em);
5462 /* One for the tree reference */
5463 free_extent_map(em);
5468 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
5471 struct btrfs_fs_info *info = trans->fs_info;
5472 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5473 struct btrfs_device_info *devices_info = NULL;
5474 struct alloc_chunk_ctl ctl;
5475 struct btrfs_block_group *block_group;
5478 lockdep_assert_held(&info->chunk_mutex);
5480 if (!alloc_profile_is_valid(type, 0)) {
5482 return ERR_PTR(-EINVAL);
5485 if (list_empty(&fs_devices->alloc_list)) {
5486 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5487 btrfs_debug(info, "%s: no writable device", __func__);
5488 return ERR_PTR(-ENOSPC);
5491 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5492 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5494 return ERR_PTR(-EINVAL);
5497 ctl.start = find_next_chunk(info);
5499 init_alloc_chunk_ctl(fs_devices, &ctl);
5501 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5504 return ERR_PTR(-ENOMEM);
5506 ret = gather_device_info(fs_devices, &ctl, devices_info);
5508 block_group = ERR_PTR(ret);
5512 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5514 block_group = ERR_PTR(ret);
5518 block_group = create_chunk(trans, &ctl, devices_info);
5521 kfree(devices_info);
5526 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5527 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5530 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5533 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
5534 struct btrfs_block_group *bg)
5536 struct btrfs_fs_info *fs_info = trans->fs_info;
5537 struct btrfs_root *chunk_root = fs_info->chunk_root;
5538 struct btrfs_key key;
5539 struct btrfs_chunk *chunk;
5540 struct btrfs_stripe *stripe;
5541 struct extent_map *em;
5542 struct map_lookup *map;
5548 * We take the chunk_mutex for 2 reasons:
5550 * 1) Updates and insertions in the chunk btree must be done while holding
5551 * the chunk_mutex, as well as updating the system chunk array in the
5552 * superblock. See the comment on top of btrfs_chunk_alloc() for the
5555 * 2) To prevent races with the final phase of a device replace operation
5556 * that replaces the device object associated with the map's stripes,
5557 * because the device object's id can change at any time during that
5558 * final phase of the device replace operation
5559 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5560 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5561 * which would cause a failure when updating the device item, which does
5562 * not exists, or persisting a stripe of the chunk item with such ID.
5563 * Here we can't use the device_list_mutex because our caller already
5564 * has locked the chunk_mutex, and the final phase of device replace
5565 * acquires both mutexes - first the device_list_mutex and then the
5566 * chunk_mutex. Using any of those two mutexes protects us from a
5567 * concurrent device replace.
5569 lockdep_assert_held(&fs_info->chunk_mutex);
5571 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
5574 btrfs_abort_transaction(trans, ret);
5578 map = em->map_lookup;
5579 item_size = btrfs_chunk_item_size(map->num_stripes);
5581 chunk = kzalloc(item_size, GFP_NOFS);
5584 btrfs_abort_transaction(trans, ret);
5588 for (i = 0; i < map->num_stripes; i++) {
5589 struct btrfs_device *device = map->stripes[i].dev;
5591 ret = btrfs_update_device(trans, device);
5596 stripe = &chunk->stripe;
5597 for (i = 0; i < map->num_stripes; i++) {
5598 struct btrfs_device *device = map->stripes[i].dev;
5599 const u64 dev_offset = map->stripes[i].physical;
5601 btrfs_set_stack_stripe_devid(stripe, device->devid);
5602 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5603 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5607 btrfs_set_stack_chunk_length(chunk, bg->length);
5608 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
5609 btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
5610 btrfs_set_stack_chunk_type(chunk, map->type);
5611 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5612 btrfs_set_stack_chunk_io_align(chunk, BTRFS_STRIPE_LEN);
5613 btrfs_set_stack_chunk_io_width(chunk, BTRFS_STRIPE_LEN);
5614 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5615 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5617 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5618 key.type = BTRFS_CHUNK_ITEM_KEY;
5619 key.offset = bg->start;
5621 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5625 set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags);
5627 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5628 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5635 free_extent_map(em);
5639 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5641 struct btrfs_fs_info *fs_info = trans->fs_info;
5643 struct btrfs_block_group *meta_bg;
5644 struct btrfs_block_group *sys_bg;
5647 * When adding a new device for sprouting, the seed device is read-only
5648 * so we must first allocate a metadata and a system chunk. But before
5649 * adding the block group items to the extent, device and chunk btrees,
5652 * 1) Create both chunks without doing any changes to the btrees, as
5653 * otherwise we would get -ENOSPC since the block groups from the
5654 * seed device are read-only;
5656 * 2) Add the device item for the new sprout device - finishing the setup
5657 * of a new block group requires updating the device item in the chunk
5658 * btree, so it must exist when we attempt to do it. The previous step
5659 * ensures this does not fail with -ENOSPC.
5661 * After that we can add the block group items to their btrees:
5662 * update existing device item in the chunk btree, add a new block group
5663 * item to the extent btree, add a new chunk item to the chunk btree and
5664 * finally add the new device extent items to the devices btree.
5667 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5668 meta_bg = btrfs_create_chunk(trans, alloc_profile);
5669 if (IS_ERR(meta_bg))
5670 return PTR_ERR(meta_bg);
5672 alloc_profile = btrfs_system_alloc_profile(fs_info);
5673 sys_bg = btrfs_create_chunk(trans, alloc_profile);
5675 return PTR_ERR(sys_bg);
5680 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5682 const int index = btrfs_bg_flags_to_raid_index(map->type);
5684 return btrfs_raid_array[index].tolerated_failures;
5687 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5689 struct extent_map *em;
5690 struct map_lookup *map;
5695 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5699 map = em->map_lookup;
5700 for (i = 0; i < map->num_stripes; i++) {
5701 if (test_bit(BTRFS_DEV_STATE_MISSING,
5702 &map->stripes[i].dev->dev_state)) {
5706 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5707 &map->stripes[i].dev->dev_state)) {
5714 * If the number of missing devices is larger than max errors, we can
5715 * not write the data into that chunk successfully.
5717 if (miss_ndevs > btrfs_chunk_max_errors(map))
5720 free_extent_map(em);
5724 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5726 struct extent_map *em;
5729 write_lock(&tree->lock);
5730 em = lookup_extent_mapping(tree, 0, (u64)-1);
5732 remove_extent_mapping(tree, em);
5733 write_unlock(&tree->lock);
5737 free_extent_map(em);
5738 /* once for the tree */
5739 free_extent_map(em);
5743 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5745 struct extent_map *em;
5746 struct map_lookup *map;
5747 enum btrfs_raid_types index;
5750 em = btrfs_get_chunk_map(fs_info, logical, len);
5753 * We could return errors for these cases, but that could get
5754 * ugly and we'd probably do the same thing which is just not do
5755 * anything else and exit, so return 1 so the callers don't try
5756 * to use other copies.
5760 map = em->map_lookup;
5761 index = btrfs_bg_flags_to_raid_index(map->type);
5763 /* Non-RAID56, use their ncopies from btrfs_raid_array. */
5764 if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5765 ret = btrfs_raid_array[index].ncopies;
5766 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5768 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5770 * There could be two corrupted data stripes, we need
5771 * to loop retry in order to rebuild the correct data.
5773 * Fail a stripe at a time on every retry except the
5774 * stripe under reconstruction.
5776 ret = map->num_stripes;
5777 free_extent_map(em);
5781 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5784 struct extent_map *em;
5785 struct map_lookup *map;
5786 unsigned long len = fs_info->sectorsize;
5788 if (!btrfs_fs_incompat(fs_info, RAID56))
5791 em = btrfs_get_chunk_map(fs_info, logical, len);
5793 if (!WARN_ON(IS_ERR(em))) {
5794 map = em->map_lookup;
5795 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5796 len = btrfs_stripe_nr_to_offset(nr_data_stripes(map));
5797 free_extent_map(em);
5802 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5804 struct extent_map *em;
5805 struct map_lookup *map;
5808 if (!btrfs_fs_incompat(fs_info, RAID56))
5811 em = btrfs_get_chunk_map(fs_info, logical, len);
5813 if(!WARN_ON(IS_ERR(em))) {
5814 map = em->map_lookup;
5815 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5817 free_extent_map(em);
5822 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5823 struct map_lookup *map, int first,
5824 int dev_replace_is_ongoing)
5828 int preferred_mirror;
5830 struct btrfs_device *srcdev;
5833 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5835 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5836 num_stripes = map->sub_stripes;
5838 num_stripes = map->num_stripes;
5840 switch (fs_info->fs_devices->read_policy) {
5842 /* Shouldn't happen, just warn and use pid instead of failing */
5843 btrfs_warn_rl(fs_info,
5844 "unknown read_policy type %u, reset to pid",
5845 fs_info->fs_devices->read_policy);
5846 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID;
5848 case BTRFS_READ_POLICY_PID:
5849 preferred_mirror = first + (current->pid % num_stripes);
5853 if (dev_replace_is_ongoing &&
5854 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5855 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5856 srcdev = fs_info->dev_replace.srcdev;
5861 * try to avoid the drive that is the source drive for a
5862 * dev-replace procedure, only choose it if no other non-missing
5863 * mirror is available
5865 for (tolerance = 0; tolerance < 2; tolerance++) {
5866 if (map->stripes[preferred_mirror].dev->bdev &&
5867 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5868 return preferred_mirror;
5869 for (i = first; i < first + num_stripes; i++) {
5870 if (map->stripes[i].dev->bdev &&
5871 (tolerance || map->stripes[i].dev != srcdev))
5876 /* we couldn't find one that doesn't fail. Just return something
5877 * and the io error handling code will clean up eventually
5879 return preferred_mirror;
5882 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info,
5885 struct btrfs_io_context *bioc;
5888 /* The size of btrfs_io_context */
5889 sizeof(struct btrfs_io_context) +
5890 /* Plus the variable array for the stripes */
5891 sizeof(struct btrfs_io_stripe) * (total_stripes),
5897 refcount_set(&bioc->refs, 1);
5899 bioc->fs_info = fs_info;
5900 bioc->replace_stripe_src = -1;
5901 bioc->full_stripe_logical = (u64)-1;
5906 void btrfs_get_bioc(struct btrfs_io_context *bioc)
5908 WARN_ON(!refcount_read(&bioc->refs));
5909 refcount_inc(&bioc->refs);
5912 void btrfs_put_bioc(struct btrfs_io_context *bioc)
5916 if (refcount_dec_and_test(&bioc->refs))
5921 * Please note that, discard won't be sent to target device of device
5924 struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info,
5925 u64 logical, u64 *length_ret,
5928 struct extent_map *em;
5929 struct map_lookup *map;
5930 struct btrfs_discard_stripe *stripes;
5931 u64 length = *length_ret;
5936 u64 stripe_end_offset;
5940 u32 sub_stripes = 0;
5941 u32 stripes_per_dev = 0;
5942 u32 remaining_stripes = 0;
5943 u32 last_stripe = 0;
5947 em = btrfs_get_chunk_map(fs_info, logical, length);
5949 return ERR_CAST(em);
5951 map = em->map_lookup;
5953 /* we don't discard raid56 yet */
5954 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5959 offset = logical - em->start;
5960 length = min_t(u64, em->start + em->len - logical, length);
5961 *length_ret = length;
5964 * stripe_nr counts the total number of stripes we have to stride
5965 * to get to this block
5967 stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
5969 /* stripe_offset is the offset of this block in its stripe */
5970 stripe_offset = offset - btrfs_stripe_nr_to_offset(stripe_nr);
5972 stripe_nr_end = round_up(offset + length, BTRFS_STRIPE_LEN) >>
5973 BTRFS_STRIPE_LEN_SHIFT;
5974 stripe_cnt = stripe_nr_end - stripe_nr;
5975 stripe_end_offset = btrfs_stripe_nr_to_offset(stripe_nr_end) -
5978 * after this, stripe_nr is the number of stripes on this
5979 * device we have to walk to find the data, and stripe_index is
5980 * the number of our device in the stripe array
5984 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5985 BTRFS_BLOCK_GROUP_RAID10)) {
5986 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5989 sub_stripes = map->sub_stripes;
5991 factor = map->num_stripes / sub_stripes;
5992 *num_stripes = min_t(u64, map->num_stripes,
5993 sub_stripes * stripe_cnt);
5994 stripe_index = stripe_nr % factor;
5995 stripe_nr /= factor;
5996 stripe_index *= sub_stripes;
5998 remaining_stripes = stripe_cnt % factor;
5999 stripes_per_dev = stripe_cnt / factor;
6000 last_stripe = ((stripe_nr_end - 1) % factor) * sub_stripes;
6001 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
6002 BTRFS_BLOCK_GROUP_DUP)) {
6003 *num_stripes = map->num_stripes;
6005 stripe_index = stripe_nr % map->num_stripes;
6006 stripe_nr /= map->num_stripes;
6009 stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS);
6015 for (i = 0; i < *num_stripes; i++) {
6016 stripes[i].physical =
6017 map->stripes[stripe_index].physical +
6018 stripe_offset + btrfs_stripe_nr_to_offset(stripe_nr);
6019 stripes[i].dev = map->stripes[stripe_index].dev;
6021 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6022 BTRFS_BLOCK_GROUP_RAID10)) {
6023 stripes[i].length = btrfs_stripe_nr_to_offset(stripes_per_dev);
6025 if (i / sub_stripes < remaining_stripes)
6026 stripes[i].length += BTRFS_STRIPE_LEN;
6029 * Special for the first stripe and
6032 * |-------|...|-------|
6036 if (i < sub_stripes)
6037 stripes[i].length -= stripe_offset;
6039 if (stripe_index >= last_stripe &&
6040 stripe_index <= (last_stripe +
6042 stripes[i].length -= stripe_end_offset;
6044 if (i == sub_stripes - 1)
6047 stripes[i].length = length;
6051 if (stripe_index == map->num_stripes) {
6057 free_extent_map(em);
6060 free_extent_map(em);
6061 return ERR_PTR(ret);
6064 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)
6066 struct btrfs_block_group *cache;
6069 /* Non zoned filesystem does not use "to_copy" flag */
6070 if (!btrfs_is_zoned(fs_info))
6073 cache = btrfs_lookup_block_group(fs_info, logical);
6075 ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags);
6077 btrfs_put_block_group(cache);
6081 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
6082 struct btrfs_io_context *bioc,
6083 struct btrfs_dev_replace *dev_replace,
6085 int *num_stripes_ret, int *max_errors_ret)
6087 u64 srcdev_devid = dev_replace->srcdev->devid;
6089 * At this stage, num_stripes is still the real number of stripes,
6090 * excluding the duplicated stripes.
6092 int num_stripes = *num_stripes_ret;
6093 int nr_extra_stripes = 0;
6094 int max_errors = *max_errors_ret;
6098 * A block group which has "to_copy" set will eventually be copied by
6099 * the dev-replace process. We can avoid cloning IO here.
6101 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))
6105 * Duplicate the write operations while the dev-replace procedure is
6106 * running. Since the copying of the old disk to the new disk takes
6107 * place at run time while the filesystem is mounted writable, the
6108 * regular write operations to the old disk have to be duplicated to go
6109 * to the new disk as well.
6111 * Note that device->missing is handled by the caller, and that the
6112 * write to the old disk is already set up in the stripes array.
6114 for (i = 0; i < num_stripes; i++) {
6115 struct btrfs_io_stripe *old = &bioc->stripes[i];
6116 struct btrfs_io_stripe *new = &bioc->stripes[num_stripes + nr_extra_stripes];
6118 if (old->dev->devid != srcdev_devid)
6121 new->physical = old->physical;
6122 new->dev = dev_replace->tgtdev;
6123 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK)
6124 bioc->replace_stripe_src = i;
6128 /* We can only have at most 2 extra nr_stripes (for DUP). */
6129 ASSERT(nr_extra_stripes <= 2);
6131 * For GET_READ_MIRRORS, we can only return at most 1 extra stripe for
6133 * If we have 2 extra stripes, only choose the one with smaller physical.
6135 if (op == BTRFS_MAP_GET_READ_MIRRORS && nr_extra_stripes == 2) {
6136 struct btrfs_io_stripe *first = &bioc->stripes[num_stripes];
6137 struct btrfs_io_stripe *second = &bioc->stripes[num_stripes + 1];
6139 /* Only DUP can have two extra stripes. */
6140 ASSERT(bioc->map_type & BTRFS_BLOCK_GROUP_DUP);
6143 * Swap the last stripe stripes and reduce @nr_extra_stripes.
6144 * The extra stripe would still be there, but won't be accessed.
6146 if (first->physical > second->physical) {
6147 swap(second->physical, first->physical);
6148 swap(second->dev, first->dev);
6153 *num_stripes_ret = num_stripes + nr_extra_stripes;
6154 *max_errors_ret = max_errors + nr_extra_stripes;
6155 bioc->replace_nr_stripes = nr_extra_stripes;
6158 static u64 btrfs_max_io_len(struct map_lookup *map, enum btrfs_map_op op,
6159 u64 offset, u32 *stripe_nr, u64 *stripe_offset,
6160 u64 *full_stripe_start)
6163 * Stripe_nr is the stripe where this block falls. stripe_offset is
6164 * the offset of this block in its stripe.
6166 *stripe_offset = offset & BTRFS_STRIPE_LEN_MASK;
6167 *stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
6168 ASSERT(*stripe_offset < U32_MAX);
6170 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6171 unsigned long full_stripe_len =
6172 btrfs_stripe_nr_to_offset(nr_data_stripes(map));
6175 * For full stripe start, we use previously calculated
6176 * @stripe_nr. Align it to nr_data_stripes, then multiply with
6179 * By this we can avoid u64 division completely. And we have
6180 * to go rounddown(), not round_down(), as nr_data_stripes is
6181 * not ensured to be power of 2.
6183 *full_stripe_start =
6184 btrfs_stripe_nr_to_offset(
6185 rounddown(*stripe_nr, nr_data_stripes(map)));
6187 ASSERT(*full_stripe_start + full_stripe_len > offset);
6188 ASSERT(*full_stripe_start <= offset);
6190 * For writes to RAID56, allow to write a full stripe set, but
6191 * no straddling of stripe sets.
6193 if (op == BTRFS_MAP_WRITE)
6194 return full_stripe_len - (offset - *full_stripe_start);
6198 * For other RAID types and for RAID56 reads, allow a single stripe (on
6201 if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK)
6202 return BTRFS_STRIPE_LEN - *stripe_offset;
6206 static void set_io_stripe(struct btrfs_io_stripe *dst, const struct map_lookup *map,
6207 u32 stripe_index, u64 stripe_offset, u32 stripe_nr)
6209 dst->dev = map->stripes[stripe_index].dev;
6210 dst->physical = map->stripes[stripe_index].physical +
6211 stripe_offset + btrfs_stripe_nr_to_offset(stripe_nr);
6215 * Map one logical range to one or more physical ranges.
6217 * @length: (Mandatory) mapped length of this run.
6218 * One logical range can be split into different segments
6219 * due to factors like zones and RAID0/5/6/10 stripe
6222 * @bioc_ret: (Mandatory) returned btrfs_io_context structure.
6223 * which has one or more physical ranges (btrfs_io_stripe)
6225 * Caller should call btrfs_put_bioc() to free it after use.
6227 * @smap: (Optional) single physical range optimization.
6228 * If the map request can be fulfilled by one single
6229 * physical range, and this is parameter is not NULL,
6230 * then @bioc_ret would be NULL, and @smap would be
6233 * @mirror_num_ret: (Mandatory) returned mirror number if the original
6236 * Mirror number 0 means to choose any live mirrors.
6238 * For non-RAID56 profiles, non-zero mirror_num means
6239 * the Nth mirror. (e.g. mirror_num 1 means the first
6242 * For RAID56 profile, mirror 1 means rebuild from P and
6243 * the remaining data stripes.
6245 * For RAID6 profile, mirror > 2 means mark another
6246 * data/P stripe error and rebuild from the remaining
6249 * @need_raid_map: (Used only for integrity checker) whether the map wants
6250 * a full stripe map (including all data and P/Q stripes)
6251 * for RAID56. Should always be 1 except integrity checker.
6253 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6254 u64 logical, u64 *length,
6255 struct btrfs_io_context **bioc_ret,
6256 struct btrfs_io_stripe *smap, int *mirror_num_ret,
6259 struct extent_map *em;
6260 struct map_lookup *map;
6268 int mirror_num = (mirror_num_ret ? *mirror_num_ret : 0);
6272 struct btrfs_io_context *bioc = NULL;
6273 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6274 int dev_replace_is_ongoing = 0;
6275 u16 num_alloc_stripes;
6276 u64 raid56_full_stripe_start = (u64)-1;
6281 num_copies = btrfs_num_copies(fs_info, logical, fs_info->sectorsize);
6282 if (mirror_num > num_copies)
6285 em = btrfs_get_chunk_map(fs_info, logical, *length);
6289 map = em->map_lookup;
6290 data_stripes = nr_data_stripes(map);
6292 map_offset = logical - em->start;
6293 max_len = btrfs_max_io_len(map, op, map_offset, &stripe_nr,
6294 &stripe_offset, &raid56_full_stripe_start);
6295 *length = min_t(u64, em->len - map_offset, max_len);
6297 down_read(&dev_replace->rwsem);
6298 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6300 * Hold the semaphore for read during the whole operation, write is
6301 * requested at commit time but must wait.
6303 if (!dev_replace_is_ongoing)
6304 up_read(&dev_replace->rwsem);
6308 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6309 stripe_index = stripe_nr % map->num_stripes;
6310 stripe_nr /= map->num_stripes;
6311 if (op == BTRFS_MAP_READ)
6313 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6314 if (op != BTRFS_MAP_READ) {
6315 num_stripes = map->num_stripes;
6316 } else if (mirror_num) {
6317 stripe_index = mirror_num - 1;
6319 stripe_index = find_live_mirror(fs_info, map, 0,
6320 dev_replace_is_ongoing);
6321 mirror_num = stripe_index + 1;
6324 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6325 if (op != BTRFS_MAP_READ) {
6326 num_stripes = map->num_stripes;
6327 } else if (mirror_num) {
6328 stripe_index = mirror_num - 1;
6333 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6334 u32 factor = map->num_stripes / map->sub_stripes;
6336 stripe_index = (stripe_nr % factor) * map->sub_stripes;
6337 stripe_nr /= factor;
6339 if (op != BTRFS_MAP_READ)
6340 num_stripes = map->sub_stripes;
6341 else if (mirror_num)
6342 stripe_index += mirror_num - 1;
6344 int old_stripe_index = stripe_index;
6345 stripe_index = find_live_mirror(fs_info, map,
6347 dev_replace_is_ongoing);
6348 mirror_num = stripe_index - old_stripe_index + 1;
6351 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6352 if (need_raid_map && (op != BTRFS_MAP_READ || mirror_num > 1)) {
6354 * Push stripe_nr back to the start of the full stripe
6355 * For those cases needing a full stripe, @stripe_nr
6356 * is the full stripe number.
6358 * Originally we go raid56_full_stripe_start / full_stripe_len,
6359 * but that can be expensive. Here we just divide
6360 * @stripe_nr with @data_stripes.
6362 stripe_nr /= data_stripes;
6364 /* RAID[56] write or recovery. Return all stripes */
6365 num_stripes = map->num_stripes;
6366 max_errors = btrfs_chunk_max_errors(map);
6368 /* Return the length to the full stripe end */
6369 *length = min(logical + *length,
6370 raid56_full_stripe_start + em->start +
6371 btrfs_stripe_nr_to_offset(data_stripes)) -
6377 * Mirror #0 or #1 means the original data block.
6378 * Mirror #2 is RAID5 parity block.
6379 * Mirror #3 is RAID6 Q block.
6381 stripe_index = stripe_nr % data_stripes;
6382 stripe_nr /= data_stripes;
6384 stripe_index = data_stripes + mirror_num - 2;
6386 /* We distribute the parity blocks across stripes */
6387 stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
6388 if (op == BTRFS_MAP_READ && mirror_num <= 1)
6393 * After this, stripe_nr is the number of stripes on this
6394 * device we have to walk to find the data, and stripe_index is
6395 * the number of our device in the stripe array
6397 stripe_index = stripe_nr % map->num_stripes;
6398 stripe_nr /= map->num_stripes;
6399 mirror_num = stripe_index + 1;
6401 if (stripe_index >= map->num_stripes) {
6403 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6404 stripe_index, map->num_stripes);
6409 num_alloc_stripes = num_stripes;
6410 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6411 op != BTRFS_MAP_READ)
6413 * For replace case, we need to add extra stripes for extra
6414 * duplicated stripes.
6416 * For both WRITE and GET_READ_MIRRORS, we may have at most
6417 * 2 more stripes (DUP types, otherwise 1).
6419 num_alloc_stripes += 2;
6422 * If this I/O maps to a single device, try to return the device and
6423 * physical block information on the stack instead of allocating an
6424 * I/O context structure.
6426 if (smap && num_alloc_stripes == 1 &&
6427 !((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1)) {
6428 set_io_stripe(smap, map, stripe_index, stripe_offset, stripe_nr);
6430 *mirror_num_ret = mirror_num;
6436 bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes);
6441 bioc->map_type = map->type;
6444 * For RAID56 full map, we need to make sure the stripes[] follows the
6445 * rule that data stripes are all ordered, then followed with P and Q
6448 * It's still mostly the same as other profiles, just with extra rotation.
6450 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6451 (op != BTRFS_MAP_READ || mirror_num > 1)) {
6453 * For RAID56 @stripe_nr is already the number of full stripes
6454 * before us, which is also the rotation value (needs to modulo
6455 * with num_stripes).
6457 * In this case, we just add @stripe_nr with @i, then do the
6458 * modulo, to reduce one modulo call.
6460 bioc->full_stripe_logical = em->start +
6461 btrfs_stripe_nr_to_offset(stripe_nr * data_stripes);
6462 for (i = 0; i < num_stripes; i++)
6463 set_io_stripe(&bioc->stripes[i], map,
6464 (i + stripe_nr) % num_stripes,
6465 stripe_offset, stripe_nr);
6468 * For all other non-RAID56 profiles, just copy the target
6469 * stripe into the bioc.
6471 for (i = 0; i < num_stripes; i++) {
6472 set_io_stripe(&bioc->stripes[i], map, stripe_index,
6473 stripe_offset, stripe_nr);
6478 if (op != BTRFS_MAP_READ)
6479 max_errors = btrfs_chunk_max_errors(map);
6481 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6482 op != BTRFS_MAP_READ) {
6483 handle_ops_on_dev_replace(op, bioc, dev_replace, logical,
6484 &num_stripes, &max_errors);
6488 bioc->num_stripes = num_stripes;
6489 bioc->max_errors = max_errors;
6490 bioc->mirror_num = mirror_num;
6493 if (dev_replace_is_ongoing) {
6494 lockdep_assert_held(&dev_replace->rwsem);
6495 /* Unlock and let waiting writers proceed */
6496 up_read(&dev_replace->rwsem);
6498 free_extent_map(em);
6502 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args,
6503 const struct btrfs_fs_devices *fs_devices)
6505 if (args->fsid == NULL)
6507 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0)
6512 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args,
6513 const struct btrfs_device *device)
6515 if (args->missing) {
6516 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) &&
6522 if (device->devid != args->devid)
6524 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0)
6530 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6533 * If devid and uuid are both specified, the match must be exact, otherwise
6534 * only devid is used.
6536 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,
6537 const struct btrfs_dev_lookup_args *args)
6539 struct btrfs_device *device;
6540 struct btrfs_fs_devices *seed_devs;
6542 if (dev_args_match_fs_devices(args, fs_devices)) {
6543 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6544 if (dev_args_match_device(args, device))
6549 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6550 if (!dev_args_match_fs_devices(args, seed_devs))
6552 list_for_each_entry(device, &seed_devs->devices, dev_list) {
6553 if (dev_args_match_device(args, device))
6561 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6562 u64 devid, u8 *dev_uuid)
6564 struct btrfs_device *device;
6565 unsigned int nofs_flag;
6568 * We call this under the chunk_mutex, so we want to use NOFS for this
6569 * allocation, however we don't want to change btrfs_alloc_device() to
6570 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6574 nofs_flag = memalloc_nofs_save();
6575 device = btrfs_alloc_device(NULL, &devid, dev_uuid, NULL);
6576 memalloc_nofs_restore(nofs_flag);
6580 list_add(&device->dev_list, &fs_devices->devices);
6581 device->fs_devices = fs_devices;
6582 fs_devices->num_devices++;
6584 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6585 fs_devices->missing_devices++;
6591 * Allocate new device struct, set up devid and UUID.
6593 * @fs_info: used only for generating a new devid, can be NULL if
6594 * devid is provided (i.e. @devid != NULL).
6595 * @devid: a pointer to devid for this device. If NULL a new devid
6597 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6599 * @path: a pointer to device path if available, NULL otherwise.
6601 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6602 * on error. Returned struct is not linked onto any lists and must be
6603 * destroyed with btrfs_free_device.
6605 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6606 const u64 *devid, const u8 *uuid,
6609 struct btrfs_device *dev;
6612 if (WARN_ON(!devid && !fs_info))
6613 return ERR_PTR(-EINVAL);
6615 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
6617 return ERR_PTR(-ENOMEM);
6619 INIT_LIST_HEAD(&dev->dev_list);
6620 INIT_LIST_HEAD(&dev->dev_alloc_list);
6621 INIT_LIST_HEAD(&dev->post_commit_list);
6623 atomic_set(&dev->dev_stats_ccnt, 0);
6624 btrfs_device_data_ordered_init(dev);
6625 extent_io_tree_init(fs_info, &dev->alloc_state, IO_TREE_DEVICE_ALLOC_STATE);
6632 ret = find_next_devid(fs_info, &tmp);
6634 btrfs_free_device(dev);
6635 return ERR_PTR(ret);
6641 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6643 generate_random_uuid(dev->uuid);
6646 struct rcu_string *name;
6648 name = rcu_string_strdup(path, GFP_KERNEL);
6650 btrfs_free_device(dev);
6651 return ERR_PTR(-ENOMEM);
6653 rcu_assign_pointer(dev->name, name);
6659 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6660 u64 devid, u8 *uuid, bool error)
6663 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6666 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6670 u64 btrfs_calc_stripe_length(const struct extent_map *em)
6672 const struct map_lookup *map = em->map_lookup;
6673 const int data_stripes = calc_data_stripes(map->type, map->num_stripes);
6675 return div_u64(em->len, data_stripes);
6678 #if BITS_PER_LONG == 32
6680 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
6681 * can't be accessed on 32bit systems.
6683 * This function do mount time check to reject the fs if it already has
6684 * metadata chunk beyond that limit.
6686 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6687 u64 logical, u64 length, u64 type)
6689 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6692 if (logical + length < MAX_LFS_FILESIZE)
6695 btrfs_err_32bit_limit(fs_info);
6700 * This is to give early warning for any metadata chunk reaching
6701 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
6702 * Although we can still access the metadata, it's not going to be possible
6703 * once the limit is reached.
6705 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6706 u64 logical, u64 length, u64 type)
6708 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6711 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)
6714 btrfs_warn_32bit_limit(fs_info);
6718 static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info,
6719 u64 devid, u8 *uuid)
6721 struct btrfs_device *dev;
6723 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6724 btrfs_report_missing_device(fs_info, devid, uuid, true);
6725 return ERR_PTR(-ENOENT);
6728 dev = add_missing_dev(fs_info->fs_devices, devid, uuid);
6730 btrfs_err(fs_info, "failed to init missing device %llu: %ld",
6731 devid, PTR_ERR(dev));
6734 btrfs_report_missing_device(fs_info, devid, uuid, false);
6739 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6740 struct btrfs_chunk *chunk)
6742 BTRFS_DEV_LOOKUP_ARGS(args);
6743 struct btrfs_fs_info *fs_info = leaf->fs_info;
6744 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6745 struct map_lookup *map;
6746 struct extent_map *em;
6751 u8 uuid[BTRFS_UUID_SIZE];
6757 logical = key->offset;
6758 length = btrfs_chunk_length(leaf, chunk);
6759 type = btrfs_chunk_type(leaf, chunk);
6760 index = btrfs_bg_flags_to_raid_index(type);
6761 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6763 #if BITS_PER_LONG == 32
6764 ret = check_32bit_meta_chunk(fs_info, logical, length, type);
6767 warn_32bit_meta_chunk(fs_info, logical, length, type);
6771 * Only need to verify chunk item if we're reading from sys chunk array,
6772 * as chunk item in tree block is already verified by tree-checker.
6774 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6775 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6780 read_lock(&map_tree->lock);
6781 em = lookup_extent_mapping(map_tree, logical, 1);
6782 read_unlock(&map_tree->lock);
6784 /* already mapped? */
6785 if (em && em->start <= logical && em->start + em->len > logical) {
6786 free_extent_map(em);
6789 free_extent_map(em);
6792 em = alloc_extent_map();
6795 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6797 free_extent_map(em);
6801 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6802 em->map_lookup = map;
6803 em->start = logical;
6806 em->block_start = 0;
6807 em->block_len = em->len;
6809 map->num_stripes = num_stripes;
6810 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6811 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6814 * We can't use the sub_stripes value, as for profiles other than
6815 * RAID10, they may have 0 as sub_stripes for filesystems created by
6816 * older mkfs (<v5.4).
6817 * In that case, it can cause divide-by-zero errors later.
6818 * Since currently sub_stripes is fixed for each profile, let's
6819 * use the trusted value instead.
6821 map->sub_stripes = btrfs_raid_array[index].sub_stripes;
6822 map->verified_stripes = 0;
6823 em->orig_block_len = btrfs_calc_stripe_length(em);
6824 for (i = 0; i < num_stripes; i++) {
6825 map->stripes[i].physical =
6826 btrfs_stripe_offset_nr(leaf, chunk, i);
6827 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6829 read_extent_buffer(leaf, uuid, (unsigned long)
6830 btrfs_stripe_dev_uuid_nr(chunk, i),
6833 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args);
6834 if (!map->stripes[i].dev) {
6835 map->stripes[i].dev = handle_missing_device(fs_info,
6837 if (IS_ERR(map->stripes[i].dev)) {
6838 ret = PTR_ERR(map->stripes[i].dev);
6839 free_extent_map(em);
6844 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6845 &(map->stripes[i].dev->dev_state));
6848 write_lock(&map_tree->lock);
6849 ret = add_extent_mapping(map_tree, em, 0);
6850 write_unlock(&map_tree->lock);
6853 "failed to add chunk map, start=%llu len=%llu: %d",
6854 em->start, em->len, ret);
6856 free_extent_map(em);
6861 static void fill_device_from_item(struct extent_buffer *leaf,
6862 struct btrfs_dev_item *dev_item,
6863 struct btrfs_device *device)
6867 device->devid = btrfs_device_id(leaf, dev_item);
6868 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6869 device->total_bytes = device->disk_total_bytes;
6870 device->commit_total_bytes = device->disk_total_bytes;
6871 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6872 device->commit_bytes_used = device->bytes_used;
6873 device->type = btrfs_device_type(leaf, dev_item);
6874 device->io_align = btrfs_device_io_align(leaf, dev_item);
6875 device->io_width = btrfs_device_io_width(leaf, dev_item);
6876 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6877 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6878 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6880 ptr = btrfs_device_uuid(dev_item);
6881 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6884 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6887 struct btrfs_fs_devices *fs_devices;
6890 lockdep_assert_held(&uuid_mutex);
6893 /* This will match only for multi-device seed fs */
6894 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
6895 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6899 fs_devices = find_fsid(fsid, NULL);
6901 if (!btrfs_test_opt(fs_info, DEGRADED))
6902 return ERR_PTR(-ENOENT);
6904 fs_devices = alloc_fs_devices(fsid, NULL);
6905 if (IS_ERR(fs_devices))
6908 fs_devices->seeding = true;
6909 fs_devices->opened = 1;
6914 * Upon first call for a seed fs fsid, just create a private copy of the
6915 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
6917 fs_devices = clone_fs_devices(fs_devices);
6918 if (IS_ERR(fs_devices))
6921 ret = open_fs_devices(fs_devices, BLK_OPEN_READ, fs_info->bdev_holder);
6923 free_fs_devices(fs_devices);
6924 return ERR_PTR(ret);
6927 if (!fs_devices->seeding) {
6928 close_fs_devices(fs_devices);
6929 free_fs_devices(fs_devices);
6930 return ERR_PTR(-EINVAL);
6933 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
6938 static int read_one_dev(struct extent_buffer *leaf,
6939 struct btrfs_dev_item *dev_item)
6941 BTRFS_DEV_LOOKUP_ARGS(args);
6942 struct btrfs_fs_info *fs_info = leaf->fs_info;
6943 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6944 struct btrfs_device *device;
6947 u8 fs_uuid[BTRFS_FSID_SIZE];
6948 u8 dev_uuid[BTRFS_UUID_SIZE];
6950 devid = btrfs_device_id(leaf, dev_item);
6952 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6954 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6956 args.uuid = dev_uuid;
6957 args.fsid = fs_uuid;
6959 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
6960 fs_devices = open_seed_devices(fs_info, fs_uuid);
6961 if (IS_ERR(fs_devices))
6962 return PTR_ERR(fs_devices);
6965 device = btrfs_find_device(fs_info->fs_devices, &args);
6967 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6968 btrfs_report_missing_device(fs_info, devid,
6973 device = add_missing_dev(fs_devices, devid, dev_uuid);
6974 if (IS_ERR(device)) {
6976 "failed to add missing dev %llu: %ld",
6977 devid, PTR_ERR(device));
6978 return PTR_ERR(device);
6980 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6982 if (!device->bdev) {
6983 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6984 btrfs_report_missing_device(fs_info,
6985 devid, dev_uuid, true);
6988 btrfs_report_missing_device(fs_info, devid,
6992 if (!device->bdev &&
6993 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6995 * this happens when a device that was properly setup
6996 * in the device info lists suddenly goes bad.
6997 * device->bdev is NULL, and so we have to set
6998 * device->missing to one here
7000 device->fs_devices->missing_devices++;
7001 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
7004 /* Move the device to its own fs_devices */
7005 if (device->fs_devices != fs_devices) {
7006 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
7007 &device->dev_state));
7009 list_move(&device->dev_list, &fs_devices->devices);
7010 device->fs_devices->num_devices--;
7011 fs_devices->num_devices++;
7013 device->fs_devices->missing_devices--;
7014 fs_devices->missing_devices++;
7016 device->fs_devices = fs_devices;
7020 if (device->fs_devices != fs_info->fs_devices) {
7021 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
7022 if (device->generation !=
7023 btrfs_device_generation(leaf, dev_item))
7027 fill_device_from_item(leaf, dev_item, device);
7029 u64 max_total_bytes = bdev_nr_bytes(device->bdev);
7031 if (device->total_bytes > max_total_bytes) {
7033 "device total_bytes should be at most %llu but found %llu",
7034 max_total_bytes, device->total_bytes);
7038 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7039 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7040 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7041 device->fs_devices->total_rw_bytes += device->total_bytes;
7042 atomic64_add(device->total_bytes - device->bytes_used,
7043 &fs_info->free_chunk_space);
7049 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7051 struct btrfs_super_block *super_copy = fs_info->super_copy;
7052 struct extent_buffer *sb;
7053 struct btrfs_disk_key *disk_key;
7054 struct btrfs_chunk *chunk;
7056 unsigned long sb_array_offset;
7063 struct btrfs_key key;
7065 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7068 * We allocated a dummy extent, just to use extent buffer accessors.
7069 * There will be unused space after BTRFS_SUPER_INFO_SIZE, but
7070 * that's fine, we will not go beyond system chunk array anyway.
7072 sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET);
7075 set_extent_buffer_uptodate(sb);
7077 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7078 array_size = btrfs_super_sys_array_size(super_copy);
7080 array_ptr = super_copy->sys_chunk_array;
7081 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7084 while (cur_offset < array_size) {
7085 disk_key = (struct btrfs_disk_key *)array_ptr;
7086 len = sizeof(*disk_key);
7087 if (cur_offset + len > array_size)
7088 goto out_short_read;
7090 btrfs_disk_key_to_cpu(&key, disk_key);
7093 sb_array_offset += len;
7096 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7098 "unexpected item type %u in sys_array at offset %u",
7099 (u32)key.type, cur_offset);
7104 chunk = (struct btrfs_chunk *)sb_array_offset;
7106 * At least one btrfs_chunk with one stripe must be present,
7107 * exact stripe count check comes afterwards
7109 len = btrfs_chunk_item_size(1);
7110 if (cur_offset + len > array_size)
7111 goto out_short_read;
7113 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7116 "invalid number of stripes %u in sys_array at offset %u",
7117 num_stripes, cur_offset);
7122 type = btrfs_chunk_type(sb, chunk);
7123 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7125 "invalid chunk type %llu in sys_array at offset %u",
7131 len = btrfs_chunk_item_size(num_stripes);
7132 if (cur_offset + len > array_size)
7133 goto out_short_read;
7135 ret = read_one_chunk(&key, sb, chunk);
7140 sb_array_offset += len;
7143 clear_extent_buffer_uptodate(sb);
7144 free_extent_buffer_stale(sb);
7148 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7150 clear_extent_buffer_uptodate(sb);
7151 free_extent_buffer_stale(sb);
7156 * Check if all chunks in the fs are OK for read-write degraded mount
7158 * If the @failing_dev is specified, it's accounted as missing.
7160 * Return true if all chunks meet the minimal RW mount requirements.
7161 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7163 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7164 struct btrfs_device *failing_dev)
7166 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7167 struct extent_map *em;
7171 read_lock(&map_tree->lock);
7172 em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7173 read_unlock(&map_tree->lock);
7174 /* No chunk at all? Return false anyway */
7180 struct map_lookup *map;
7185 map = em->map_lookup;
7187 btrfs_get_num_tolerated_disk_barrier_failures(
7189 for (i = 0; i < map->num_stripes; i++) {
7190 struct btrfs_device *dev = map->stripes[i].dev;
7192 if (!dev || !dev->bdev ||
7193 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7194 dev->last_flush_error)
7196 else if (failing_dev && failing_dev == dev)
7199 if (missing > max_tolerated) {
7202 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7203 em->start, missing, max_tolerated);
7204 free_extent_map(em);
7208 next_start = extent_map_end(em);
7209 free_extent_map(em);
7211 read_lock(&map_tree->lock);
7212 em = lookup_extent_mapping(map_tree, next_start,
7213 (u64)(-1) - next_start);
7214 read_unlock(&map_tree->lock);
7220 static void readahead_tree_node_children(struct extent_buffer *node)
7223 const int nr_items = btrfs_header_nritems(node);
7225 for (i = 0; i < nr_items; i++)
7226 btrfs_readahead_node_child(node, i);
7229 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7231 struct btrfs_root *root = fs_info->chunk_root;
7232 struct btrfs_path *path;
7233 struct extent_buffer *leaf;
7234 struct btrfs_key key;
7235 struct btrfs_key found_key;
7240 u64 last_ra_node = 0;
7242 path = btrfs_alloc_path();
7247 * uuid_mutex is needed only if we are mounting a sprout FS
7248 * otherwise we don't need it.
7250 mutex_lock(&uuid_mutex);
7253 * It is possible for mount and umount to race in such a way that
7254 * we execute this code path, but open_fs_devices failed to clear
7255 * total_rw_bytes. We certainly want it cleared before reading the
7256 * device items, so clear it here.
7258 fs_info->fs_devices->total_rw_bytes = 0;
7261 * Lockdep complains about possible circular locking dependency between
7262 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores
7263 * used for freeze procection of a fs (struct super_block.s_writers),
7264 * which we take when starting a transaction, and extent buffers of the
7265 * chunk tree if we call read_one_dev() while holding a lock on an
7266 * extent buffer of the chunk tree. Since we are mounting the filesystem
7267 * and at this point there can't be any concurrent task modifying the
7268 * chunk tree, to keep it simple, just skip locking on the chunk tree.
7270 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
7271 path->skip_locking = 1;
7274 * Read all device items, and then all the chunk items. All
7275 * device items are found before any chunk item (their object id
7276 * is smaller than the lowest possible object id for a chunk
7277 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7279 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7282 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) {
7283 struct extent_buffer *node = path->nodes[1];
7285 leaf = path->nodes[0];
7286 slot = path->slots[0];
7289 if (last_ra_node != node->start) {
7290 readahead_tree_node_children(node);
7291 last_ra_node = node->start;
7294 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7295 struct btrfs_dev_item *dev_item;
7296 dev_item = btrfs_item_ptr(leaf, slot,
7297 struct btrfs_dev_item);
7298 ret = read_one_dev(leaf, dev_item);
7302 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7303 struct btrfs_chunk *chunk;
7306 * We are only called at mount time, so no need to take
7307 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7308 * we always lock first fs_info->chunk_mutex before
7309 * acquiring any locks on the chunk tree. This is a
7310 * requirement for chunk allocation, see the comment on
7311 * top of btrfs_chunk_alloc() for details.
7313 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7314 ret = read_one_chunk(&found_key, leaf, chunk);
7319 /* Catch error found during iteration */
7326 * After loading chunk tree, we've got all device information,
7327 * do another round of validation checks.
7329 if (total_dev != fs_info->fs_devices->total_devices) {
7331 "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit",
7332 btrfs_super_num_devices(fs_info->super_copy),
7334 fs_info->fs_devices->total_devices = total_dev;
7335 btrfs_set_super_num_devices(fs_info->super_copy, total_dev);
7337 if (btrfs_super_total_bytes(fs_info->super_copy) <
7338 fs_info->fs_devices->total_rw_bytes) {
7340 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7341 btrfs_super_total_bytes(fs_info->super_copy),
7342 fs_info->fs_devices->total_rw_bytes);
7348 mutex_unlock(&uuid_mutex);
7350 btrfs_free_path(path);
7354 int btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7356 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7357 struct btrfs_device *device;
7360 fs_devices->fs_info = fs_info;
7362 mutex_lock(&fs_devices->device_list_mutex);
7363 list_for_each_entry(device, &fs_devices->devices, dev_list)
7364 device->fs_info = fs_info;
7366 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7367 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7368 device->fs_info = fs_info;
7369 ret = btrfs_get_dev_zone_info(device, false);
7374 seed_devs->fs_info = fs_info;
7376 mutex_unlock(&fs_devices->device_list_mutex);
7381 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7382 const struct btrfs_dev_stats_item *ptr,
7387 read_extent_buffer(eb, &val,
7388 offsetof(struct btrfs_dev_stats_item, values) +
7389 ((unsigned long)ptr) + (index * sizeof(u64)),
7394 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7395 struct btrfs_dev_stats_item *ptr,
7398 write_extent_buffer(eb, &val,
7399 offsetof(struct btrfs_dev_stats_item, values) +
7400 ((unsigned long)ptr) + (index * sizeof(u64)),
7404 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7405 struct btrfs_path *path)
7407 struct btrfs_dev_stats_item *ptr;
7408 struct extent_buffer *eb;
7409 struct btrfs_key key;
7413 if (!device->fs_info->dev_root)
7416 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7417 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7418 key.offset = device->devid;
7419 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7421 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7422 btrfs_dev_stat_set(device, i, 0);
7423 device->dev_stats_valid = 1;
7424 btrfs_release_path(path);
7425 return ret < 0 ? ret : 0;
7427 slot = path->slots[0];
7428 eb = path->nodes[0];
7429 item_size = btrfs_item_size(eb, slot);
7431 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7433 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7434 if (item_size >= (1 + i) * sizeof(__le64))
7435 btrfs_dev_stat_set(device, i,
7436 btrfs_dev_stats_value(eb, ptr, i));
7438 btrfs_dev_stat_set(device, i, 0);
7441 device->dev_stats_valid = 1;
7442 btrfs_dev_stat_print_on_load(device);
7443 btrfs_release_path(path);
7448 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7450 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7451 struct btrfs_device *device;
7452 struct btrfs_path *path = NULL;
7455 path = btrfs_alloc_path();
7459 mutex_lock(&fs_devices->device_list_mutex);
7460 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7461 ret = btrfs_device_init_dev_stats(device, path);
7465 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7466 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7467 ret = btrfs_device_init_dev_stats(device, path);
7473 mutex_unlock(&fs_devices->device_list_mutex);
7475 btrfs_free_path(path);
7479 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7480 struct btrfs_device *device)
7482 struct btrfs_fs_info *fs_info = trans->fs_info;
7483 struct btrfs_root *dev_root = fs_info->dev_root;
7484 struct btrfs_path *path;
7485 struct btrfs_key key;
7486 struct extent_buffer *eb;
7487 struct btrfs_dev_stats_item *ptr;
7491 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7492 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7493 key.offset = device->devid;
7495 path = btrfs_alloc_path();
7498 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7500 btrfs_warn_in_rcu(fs_info,
7501 "error %d while searching for dev_stats item for device %s",
7502 ret, btrfs_dev_name(device));
7507 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7508 /* need to delete old one and insert a new one */
7509 ret = btrfs_del_item(trans, dev_root, path);
7511 btrfs_warn_in_rcu(fs_info,
7512 "delete too small dev_stats item for device %s failed %d",
7513 btrfs_dev_name(device), ret);
7520 /* need to insert a new item */
7521 btrfs_release_path(path);
7522 ret = btrfs_insert_empty_item(trans, dev_root, path,
7523 &key, sizeof(*ptr));
7525 btrfs_warn_in_rcu(fs_info,
7526 "insert dev_stats item for device %s failed %d",
7527 btrfs_dev_name(device), ret);
7532 eb = path->nodes[0];
7533 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7534 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7535 btrfs_set_dev_stats_value(eb, ptr, i,
7536 btrfs_dev_stat_read(device, i));
7537 btrfs_mark_buffer_dirty(eb);
7540 btrfs_free_path(path);
7545 * called from commit_transaction. Writes all changed device stats to disk.
7547 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7549 struct btrfs_fs_info *fs_info = trans->fs_info;
7550 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7551 struct btrfs_device *device;
7555 mutex_lock(&fs_devices->device_list_mutex);
7556 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7557 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7558 if (!device->dev_stats_valid || stats_cnt == 0)
7563 * There is a LOAD-LOAD control dependency between the value of
7564 * dev_stats_ccnt and updating the on-disk values which requires
7565 * reading the in-memory counters. Such control dependencies
7566 * require explicit read memory barriers.
7568 * This memory barriers pairs with smp_mb__before_atomic in
7569 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7570 * barrier implied by atomic_xchg in
7571 * btrfs_dev_stats_read_and_reset
7575 ret = update_dev_stat_item(trans, device);
7577 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7579 mutex_unlock(&fs_devices->device_list_mutex);
7584 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7586 btrfs_dev_stat_inc(dev, index);
7588 if (!dev->dev_stats_valid)
7590 btrfs_err_rl_in_rcu(dev->fs_info,
7591 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7592 btrfs_dev_name(dev),
7593 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7594 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7595 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7596 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7597 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7600 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7604 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7605 if (btrfs_dev_stat_read(dev, i) != 0)
7607 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7608 return; /* all values == 0, suppress message */
7610 btrfs_info_in_rcu(dev->fs_info,
7611 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7612 btrfs_dev_name(dev),
7613 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7614 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7615 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7616 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7617 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7620 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7621 struct btrfs_ioctl_get_dev_stats *stats)
7623 BTRFS_DEV_LOOKUP_ARGS(args);
7624 struct btrfs_device *dev;
7625 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7628 mutex_lock(&fs_devices->device_list_mutex);
7629 args.devid = stats->devid;
7630 dev = btrfs_find_device(fs_info->fs_devices, &args);
7631 mutex_unlock(&fs_devices->device_list_mutex);
7634 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7636 } else if (!dev->dev_stats_valid) {
7637 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7639 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7640 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7641 if (stats->nr_items > i)
7643 btrfs_dev_stat_read_and_reset(dev, i);
7645 btrfs_dev_stat_set(dev, i, 0);
7647 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7648 current->comm, task_pid_nr(current));
7650 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7651 if (stats->nr_items > i)
7652 stats->values[i] = btrfs_dev_stat_read(dev, i);
7654 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7655 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7660 * Update the size and bytes used for each device where it changed. This is
7661 * delayed since we would otherwise get errors while writing out the
7664 * Must be invoked during transaction commit.
7666 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7668 struct btrfs_device *curr, *next;
7670 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7672 if (list_empty(&trans->dev_update_list))
7676 * We don't need the device_list_mutex here. This list is owned by the
7677 * transaction and the transaction must complete before the device is
7680 mutex_lock(&trans->fs_info->chunk_mutex);
7681 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7683 list_del_init(&curr->post_commit_list);
7684 curr->commit_total_bytes = curr->disk_total_bytes;
7685 curr->commit_bytes_used = curr->bytes_used;
7687 mutex_unlock(&trans->fs_info->chunk_mutex);
7691 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7693 int btrfs_bg_type_to_factor(u64 flags)
7695 const int index = btrfs_bg_flags_to_raid_index(flags);
7697 return btrfs_raid_array[index].ncopies;
7702 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7703 u64 chunk_offset, u64 devid,
7704 u64 physical_offset, u64 physical_len)
7706 struct btrfs_dev_lookup_args args = { .devid = devid };
7707 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7708 struct extent_map *em;
7709 struct map_lookup *map;
7710 struct btrfs_device *dev;
7716 read_lock(&em_tree->lock);
7717 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7718 read_unlock(&em_tree->lock);
7722 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7723 physical_offset, devid);
7728 map = em->map_lookup;
7729 stripe_len = btrfs_calc_stripe_length(em);
7730 if (physical_len != stripe_len) {
7732 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7733 physical_offset, devid, em->start, physical_len,
7740 * Very old mkfs.btrfs (before v4.1) will not respect the reserved
7741 * space. Although kernel can handle it without problem, better to warn
7744 if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED)
7746 "devid %llu physical %llu len %llu inside the reserved space",
7747 devid, physical_offset, physical_len);
7749 for (i = 0; i < map->num_stripes; i++) {
7750 if (map->stripes[i].dev->devid == devid &&
7751 map->stripes[i].physical == physical_offset) {
7753 if (map->verified_stripes >= map->num_stripes) {
7755 "too many dev extents for chunk %llu found",
7760 map->verified_stripes++;
7766 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7767 physical_offset, devid);
7771 /* Make sure no dev extent is beyond device boundary */
7772 dev = btrfs_find_device(fs_info->fs_devices, &args);
7774 btrfs_err(fs_info, "failed to find devid %llu", devid);
7779 if (physical_offset + physical_len > dev->disk_total_bytes) {
7781 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7782 devid, physical_offset, physical_len,
7783 dev->disk_total_bytes);
7788 if (dev->zone_info) {
7789 u64 zone_size = dev->zone_info->zone_size;
7791 if (!IS_ALIGNED(physical_offset, zone_size) ||
7792 !IS_ALIGNED(physical_len, zone_size)) {
7794 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
7795 devid, physical_offset, physical_len);
7802 free_extent_map(em);
7806 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7808 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7809 struct extent_map *em;
7810 struct rb_node *node;
7813 read_lock(&em_tree->lock);
7814 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
7815 em = rb_entry(node, struct extent_map, rb_node);
7816 if (em->map_lookup->num_stripes !=
7817 em->map_lookup->verified_stripes) {
7819 "chunk %llu has missing dev extent, have %d expect %d",
7820 em->start, em->map_lookup->verified_stripes,
7821 em->map_lookup->num_stripes);
7827 read_unlock(&em_tree->lock);
7832 * Ensure that all dev extents are mapped to correct chunk, otherwise
7833 * later chunk allocation/free would cause unexpected behavior.
7835 * NOTE: This will iterate through the whole device tree, which should be of
7836 * the same size level as the chunk tree. This slightly increases mount time.
7838 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
7840 struct btrfs_path *path;
7841 struct btrfs_root *root = fs_info->dev_root;
7842 struct btrfs_key key;
7844 u64 prev_dev_ext_end = 0;
7848 * We don't have a dev_root because we mounted with ignorebadroots and
7849 * failed to load the root, so we want to skip the verification in this
7852 * However if the dev root is fine, but the tree itself is corrupted
7853 * we'd still fail to mount. This verification is only to make sure
7854 * writes can happen safely, so instead just bypass this check
7855 * completely in the case of IGNOREBADROOTS.
7857 if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
7861 key.type = BTRFS_DEV_EXTENT_KEY;
7864 path = btrfs_alloc_path();
7868 path->reada = READA_FORWARD;
7869 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7873 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7874 ret = btrfs_next_leaf(root, path);
7877 /* No dev extents at all? Not good */
7884 struct extent_buffer *leaf = path->nodes[0];
7885 struct btrfs_dev_extent *dext;
7886 int slot = path->slots[0];
7888 u64 physical_offset;
7892 btrfs_item_key_to_cpu(leaf, &key, slot);
7893 if (key.type != BTRFS_DEV_EXTENT_KEY)
7895 devid = key.objectid;
7896 physical_offset = key.offset;
7898 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
7899 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
7900 physical_len = btrfs_dev_extent_length(leaf, dext);
7902 /* Check if this dev extent overlaps with the previous one */
7903 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
7905 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
7906 devid, physical_offset, prev_dev_ext_end);
7911 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
7912 physical_offset, physical_len);
7916 prev_dev_ext_end = physical_offset + physical_len;
7918 ret = btrfs_next_item(root, path);
7927 /* Ensure all chunks have corresponding dev extents */
7928 ret = verify_chunk_dev_extent_mapping(fs_info);
7930 btrfs_free_path(path);
7935 * Check whether the given block group or device is pinned by any inode being
7936 * used as a swapfile.
7938 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
7940 struct btrfs_swapfile_pin *sp;
7941 struct rb_node *node;
7943 spin_lock(&fs_info->swapfile_pins_lock);
7944 node = fs_info->swapfile_pins.rb_node;
7946 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
7948 node = node->rb_left;
7949 else if (ptr > sp->ptr)
7950 node = node->rb_right;
7954 spin_unlock(&fs_info->swapfile_pins_lock);
7955 return node != NULL;
7958 static int relocating_repair_kthread(void *data)
7960 struct btrfs_block_group *cache = data;
7961 struct btrfs_fs_info *fs_info = cache->fs_info;
7965 target = cache->start;
7966 btrfs_put_block_group(cache);
7968 sb_start_write(fs_info->sb);
7969 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
7971 "zoned: skip relocating block group %llu to repair: EBUSY",
7973 sb_end_write(fs_info->sb);
7977 mutex_lock(&fs_info->reclaim_bgs_lock);
7979 /* Ensure block group still exists */
7980 cache = btrfs_lookup_block_group(fs_info, target);
7984 if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags))
7987 ret = btrfs_may_alloc_data_chunk(fs_info, target);
7992 "zoned: relocating block group %llu to repair IO failure",
7994 ret = btrfs_relocate_chunk(fs_info, target);
7998 btrfs_put_block_group(cache);
7999 mutex_unlock(&fs_info->reclaim_bgs_lock);
8000 btrfs_exclop_finish(fs_info);
8001 sb_end_write(fs_info->sb);
8006 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical)
8008 struct btrfs_block_group *cache;
8010 if (!btrfs_is_zoned(fs_info))
8013 /* Do not attempt to repair in degraded state */
8014 if (btrfs_test_opt(fs_info, DEGRADED))
8017 cache = btrfs_lookup_block_group(fs_info, logical);
8021 if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) {
8022 btrfs_put_block_group(cache);
8026 kthread_run(relocating_repair_kthread, cache,
8027 "btrfs-relocating-repair");
8032 static void map_raid56_repair_block(struct btrfs_io_context *bioc,
8033 struct btrfs_io_stripe *smap,
8036 int data_stripes = nr_bioc_data_stripes(bioc);
8039 for (i = 0; i < data_stripes; i++) {
8040 u64 stripe_start = bioc->full_stripe_logical +
8041 btrfs_stripe_nr_to_offset(i);
8043 if (logical >= stripe_start &&
8044 logical < stripe_start + BTRFS_STRIPE_LEN)
8047 ASSERT(i < data_stripes);
8048 smap->dev = bioc->stripes[i].dev;
8049 smap->physical = bioc->stripes[i].physical +
8050 ((logical - bioc->full_stripe_logical) &
8051 BTRFS_STRIPE_LEN_MASK);
8055 * Map a repair write into a single device.
8057 * A repair write is triggered by read time repair or scrub, which would only
8058 * update the contents of a single device.
8059 * Not update any other mirrors nor go through RMW path.
8061 * Callers should ensure:
8063 * - Call btrfs_bio_counter_inc_blocked() first
8064 * - The range does not cross stripe boundary
8065 * - Has a valid @mirror_num passed in.
8067 int btrfs_map_repair_block(struct btrfs_fs_info *fs_info,
8068 struct btrfs_io_stripe *smap, u64 logical,
8069 u32 length, int mirror_num)
8071 struct btrfs_io_context *bioc = NULL;
8072 u64 map_length = length;
8073 int mirror_ret = mirror_num;
8076 ASSERT(mirror_num > 0);
8078 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, &map_length,
8079 &bioc, smap, &mirror_ret, true);
8083 /* The map range should not cross stripe boundary. */
8084 ASSERT(map_length >= length);
8086 /* Already mapped to single stripe. */
8090 /* Map the RAID56 multi-stripe writes to a single one. */
8091 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
8092 map_raid56_repair_block(bioc, smap, logical);
8096 ASSERT(mirror_num <= bioc->num_stripes);
8097 smap->dev = bioc->stripes[mirror_num - 1].dev;
8098 smap->physical = bioc->stripes[mirror_num - 1].physical;
8100 btrfs_put_bioc(bioc);