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);
685 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices
686 * being created with a disk that has already completed its fsid change. Such
687 * disk can belong to an fs which has its FSID changed or to one which doesn't.
688 * Handle both cases here.
690 static struct btrfs_fs_devices *find_fsid_inprogress(
691 struct btrfs_super_block *disk_super)
693 struct btrfs_fs_devices *fs_devices;
695 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
696 if (fs_devices->fsid_change)
699 if (check_fsid_changed(fs_devices, disk_super->fsid))
703 return find_fsid(disk_super->fsid, NULL);
706 static struct btrfs_fs_devices *find_fsid_changed(
707 struct btrfs_super_block *disk_super)
709 struct btrfs_fs_devices *fs_devices;
712 * Handles the case where scanned device is part of an fs that had
713 * multiple successful changes of FSID but currently device didn't
714 * observe it. Meaning our fsid will be different than theirs. We need
715 * to handle two subcases :
716 * 1 - The fs still continues to have different METADATA/FSID uuids.
717 * 2 - The fs is switched back to its original FSID (METADATA/FSID
720 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
722 if (check_fsid_changed(fs_devices, disk_super->metadata_uuid) &&
723 memcmp(fs_devices->fsid, disk_super->fsid,
724 BTRFS_FSID_SIZE) != 0)
727 /* Unchanged UUIDs */
728 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
729 BTRFS_FSID_SIZE) == 0 &&
730 memcmp(fs_devices->fsid, disk_super->metadata_uuid,
731 BTRFS_FSID_SIZE) == 0)
738 static struct btrfs_fs_devices *find_fsid_reverted_metadata(
739 struct btrfs_super_block *disk_super)
741 struct btrfs_fs_devices *fs_devices;
744 * Handle the case where the scanned device is part of an fs whose last
745 * metadata UUID change reverted it to the original FSID. At the same
746 * time fs_devices was first created by another constituent device
747 * which didn't fully observe the operation. This results in an
748 * btrfs_fs_devices created with metadata/fsid different AND
749 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the
750 * fs_devices equal to the FSID of the disk.
752 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
753 if (!fs_devices->fsid_change)
756 if (check_fsid_changed(fs_devices, disk_super->fsid))
763 * Add new device to list of registered devices
766 * device pointer which was just added or updated when successful
767 * error pointer when failed
769 static noinline struct btrfs_device *device_list_add(const char *path,
770 struct btrfs_super_block *disk_super,
771 bool *new_device_added)
773 struct btrfs_device *device;
774 struct btrfs_fs_devices *fs_devices = NULL;
775 struct rcu_string *name;
776 u64 found_transid = btrfs_super_generation(disk_super);
777 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
780 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
781 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
782 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) &
783 BTRFS_SUPER_FLAG_CHANGING_FSID_V2);
785 error = lookup_bdev(path, &path_devt);
787 btrfs_err(NULL, "failed to lookup block device for path %s: %d",
789 return ERR_PTR(error);
792 if (fsid_change_in_progress) {
793 if (!has_metadata_uuid)
794 fs_devices = find_fsid_inprogress(disk_super);
796 fs_devices = find_fsid_changed(disk_super);
797 } else if (has_metadata_uuid) {
798 fs_devices = find_fsid_with_metadata_uuid(disk_super);
800 fs_devices = find_fsid_reverted_metadata(disk_super);
802 fs_devices = find_fsid(disk_super->fsid, NULL);
807 fs_devices = alloc_fs_devices(disk_super->fsid,
808 has_metadata_uuid ? disk_super->metadata_uuid : NULL);
809 if (IS_ERR(fs_devices))
810 return ERR_CAST(fs_devices);
812 fs_devices->fsid_change = fsid_change_in_progress;
814 mutex_lock(&fs_devices->device_list_mutex);
815 list_add(&fs_devices->fs_list, &fs_uuids);
819 struct btrfs_dev_lookup_args args = {
821 .uuid = disk_super->dev_item.uuid,
824 mutex_lock(&fs_devices->device_list_mutex);
825 device = btrfs_find_device(fs_devices, &args);
828 * If this disk has been pulled into an fs devices created by
829 * a device which had the CHANGING_FSID_V2 flag then replace the
830 * metadata_uuid/fsid values of the fs_devices.
832 if (fs_devices->fsid_change &&
833 found_transid > fs_devices->latest_generation) {
834 memcpy(fs_devices->fsid, disk_super->fsid,
837 if (has_metadata_uuid)
838 memcpy(fs_devices->metadata_uuid,
839 disk_super->metadata_uuid,
842 memcpy(fs_devices->metadata_uuid,
843 disk_super->fsid, BTRFS_FSID_SIZE);
845 fs_devices->fsid_change = false;
850 unsigned int nofs_flag;
852 if (fs_devices->opened) {
854 "device %s belongs to fsid %pU, and the fs is already mounted",
855 path, fs_devices->fsid);
856 mutex_unlock(&fs_devices->device_list_mutex);
857 return ERR_PTR(-EBUSY);
860 nofs_flag = memalloc_nofs_save();
861 device = btrfs_alloc_device(NULL, &devid,
862 disk_super->dev_item.uuid, path);
863 memalloc_nofs_restore(nofs_flag);
864 if (IS_ERR(device)) {
865 mutex_unlock(&fs_devices->device_list_mutex);
866 /* we can safely leave the fs_devices entry around */
870 device->devt = path_devt;
872 list_add_rcu(&device->dev_list, &fs_devices->devices);
873 fs_devices->num_devices++;
875 device->fs_devices = fs_devices;
876 *new_device_added = true;
878 if (disk_super->label[0])
880 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n",
881 disk_super->label, devid, found_transid, path,
882 current->comm, task_pid_nr(current));
885 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n",
886 disk_super->fsid, devid, found_transid, path,
887 current->comm, task_pid_nr(current));
889 } else if (!device->name || strcmp(device->name->str, path)) {
891 * When FS is already mounted.
892 * 1. If you are here and if the device->name is NULL that
893 * means this device was missing at time of FS mount.
894 * 2. If you are here and if the device->name is different
895 * from 'path' that means either
896 * a. The same device disappeared and reappeared with
898 * b. The missing-disk-which-was-replaced, has
901 * We must allow 1 and 2a above. But 2b would be a spurious
904 * Further in case of 1 and 2a above, the disk at 'path'
905 * would have missed some transaction when it was away and
906 * in case of 2a the stale bdev has to be updated as well.
907 * 2b must not be allowed at all time.
911 * For now, we do allow update to btrfs_fs_device through the
912 * btrfs dev scan cli after FS has been mounted. We're still
913 * tracking a problem where systems fail mount by subvolume id
914 * when we reject replacement on a mounted FS.
916 if (!fs_devices->opened && found_transid < device->generation) {
918 * That is if the FS is _not_ mounted and if you
919 * are here, that means there is more than one
920 * disk with same uuid and devid.We keep the one
921 * with larger generation number or the last-in if
922 * generation are equal.
924 mutex_unlock(&fs_devices->device_list_mutex);
926 "device %s already registered with a higher generation, found %llu expect %llu",
927 path, found_transid, device->generation);
928 return ERR_PTR(-EEXIST);
932 * We are going to replace the device path for a given devid,
933 * make sure it's the same device if the device is mounted
935 * NOTE: the device->fs_info may not be reliable here so pass
936 * in a NULL to message helpers instead. This avoids a possible
937 * use-after-free when the fs_info and fs_info->sb are already
941 if (device->devt != path_devt) {
942 mutex_unlock(&fs_devices->device_list_mutex);
943 btrfs_warn_in_rcu(NULL,
944 "duplicate device %s devid %llu generation %llu scanned by %s (%d)",
945 path, devid, found_transid,
947 task_pid_nr(current));
948 return ERR_PTR(-EEXIST);
950 btrfs_info_in_rcu(NULL,
951 "devid %llu device path %s changed to %s scanned by %s (%d)",
952 devid, btrfs_dev_name(device),
954 task_pid_nr(current));
957 name = rcu_string_strdup(path, GFP_NOFS);
959 mutex_unlock(&fs_devices->device_list_mutex);
960 return ERR_PTR(-ENOMEM);
962 rcu_string_free(device->name);
963 rcu_assign_pointer(device->name, name);
964 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
965 fs_devices->missing_devices--;
966 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
968 device->devt = path_devt;
972 * Unmount does not free the btrfs_device struct but would zero
973 * generation along with most of the other members. So just update
974 * it back. We need it to pick the disk with largest generation
977 if (!fs_devices->opened) {
978 device->generation = found_transid;
979 fs_devices->latest_generation = max_t(u64, found_transid,
980 fs_devices->latest_generation);
983 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
985 mutex_unlock(&fs_devices->device_list_mutex);
989 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
991 struct btrfs_fs_devices *fs_devices;
992 struct btrfs_device *device;
993 struct btrfs_device *orig_dev;
996 lockdep_assert_held(&uuid_mutex);
998 fs_devices = alloc_fs_devices(orig->fsid, NULL);
999 if (IS_ERR(fs_devices))
1002 fs_devices->total_devices = orig->total_devices;
1004 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
1005 const char *dev_path = NULL;
1008 * This is ok to do without RCU read locked because we hold the
1009 * uuid mutex so nothing we touch in here is going to disappear.
1012 dev_path = orig_dev->name->str;
1014 device = btrfs_alloc_device(NULL, &orig_dev->devid,
1015 orig_dev->uuid, dev_path);
1016 if (IS_ERR(device)) {
1017 ret = PTR_ERR(device);
1021 if (orig_dev->zone_info) {
1022 struct btrfs_zoned_device_info *zone_info;
1024 zone_info = btrfs_clone_dev_zone_info(orig_dev);
1026 btrfs_free_device(device);
1030 device->zone_info = zone_info;
1033 list_add(&device->dev_list, &fs_devices->devices);
1034 device->fs_devices = fs_devices;
1035 fs_devices->num_devices++;
1039 free_fs_devices(fs_devices);
1040 return ERR_PTR(ret);
1043 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices,
1044 struct btrfs_device **latest_dev)
1046 struct btrfs_device *device, *next;
1048 /* This is the initialized path, it is safe to release the devices. */
1049 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1050 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) {
1051 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1052 &device->dev_state) &&
1053 !test_bit(BTRFS_DEV_STATE_MISSING,
1054 &device->dev_state) &&
1056 device->generation > (*latest_dev)->generation)) {
1057 *latest_dev = device;
1063 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID,
1064 * in btrfs_init_dev_replace() so just continue.
1066 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1070 blkdev_put(device->bdev, device->holder);
1071 device->bdev = NULL;
1072 fs_devices->open_devices--;
1074 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1075 list_del_init(&device->dev_alloc_list);
1076 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1077 fs_devices->rw_devices--;
1079 list_del_init(&device->dev_list);
1080 fs_devices->num_devices--;
1081 btrfs_free_device(device);
1087 * After we have read the system tree and know devids belonging to this
1088 * filesystem, remove the device which does not belong there.
1090 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices)
1092 struct btrfs_device *latest_dev = NULL;
1093 struct btrfs_fs_devices *seed_dev;
1095 mutex_lock(&uuid_mutex);
1096 __btrfs_free_extra_devids(fs_devices, &latest_dev);
1098 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list)
1099 __btrfs_free_extra_devids(seed_dev, &latest_dev);
1101 fs_devices->latest_dev = latest_dev;
1103 mutex_unlock(&uuid_mutex);
1106 static void btrfs_close_bdev(struct btrfs_device *device)
1111 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1112 sync_blockdev(device->bdev);
1113 invalidate_bdev(device->bdev);
1116 blkdev_put(device->bdev, device->holder);
1119 static void btrfs_close_one_device(struct btrfs_device *device)
1121 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1123 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1124 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1125 list_del_init(&device->dev_alloc_list);
1126 fs_devices->rw_devices--;
1129 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1130 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
1132 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
1133 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
1134 fs_devices->missing_devices--;
1137 btrfs_close_bdev(device);
1139 fs_devices->open_devices--;
1140 device->bdev = NULL;
1142 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1143 btrfs_destroy_dev_zone_info(device);
1145 device->fs_info = NULL;
1146 atomic_set(&device->dev_stats_ccnt, 0);
1147 extent_io_tree_release(&device->alloc_state);
1150 * Reset the flush error record. We might have a transient flush error
1151 * in this mount, and if so we aborted the current transaction and set
1152 * the fs to an error state, guaranteeing no super blocks can be further
1153 * committed. However that error might be transient and if we unmount the
1154 * filesystem and mount it again, we should allow the mount to succeed
1155 * (btrfs_check_rw_degradable() should not fail) - if after mounting the
1156 * filesystem again we still get flush errors, then we will again abort
1157 * any transaction and set the error state, guaranteeing no commits of
1158 * unsafe super blocks.
1160 device->last_flush_error = 0;
1162 /* Verify the device is back in a pristine state */
1163 WARN_ON(test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1164 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1165 WARN_ON(!list_empty(&device->dev_alloc_list));
1166 WARN_ON(!list_empty(&device->post_commit_list));
1169 static void close_fs_devices(struct btrfs_fs_devices *fs_devices)
1171 struct btrfs_device *device, *tmp;
1173 lockdep_assert_held(&uuid_mutex);
1175 if (--fs_devices->opened > 0)
1178 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list)
1179 btrfs_close_one_device(device);
1181 WARN_ON(fs_devices->open_devices);
1182 WARN_ON(fs_devices->rw_devices);
1183 fs_devices->opened = 0;
1184 fs_devices->seeding = false;
1185 fs_devices->fs_info = NULL;
1188 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1191 struct btrfs_fs_devices *tmp;
1193 mutex_lock(&uuid_mutex);
1194 close_fs_devices(fs_devices);
1195 if (!fs_devices->opened) {
1196 list_splice_init(&fs_devices->seed_list, &list);
1199 * If the struct btrfs_fs_devices is not assembled with any
1200 * other device, it can be re-initialized during the next mount
1201 * without the needing device-scan step. Therefore, it can be
1204 if (fs_devices->num_devices == 1) {
1205 list_del(&fs_devices->fs_list);
1206 free_fs_devices(fs_devices);
1211 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {
1212 close_fs_devices(fs_devices);
1213 list_del(&fs_devices->seed_list);
1214 free_fs_devices(fs_devices);
1216 mutex_unlock(&uuid_mutex);
1219 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1220 blk_mode_t flags, void *holder)
1222 struct btrfs_device *device;
1223 struct btrfs_device *latest_dev = NULL;
1224 struct btrfs_device *tmp_device;
1226 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,
1230 ret = btrfs_open_one_device(fs_devices, device, flags, holder);
1232 (!latest_dev || device->generation > latest_dev->generation)) {
1233 latest_dev = device;
1234 } else if (ret == -ENODATA) {
1235 fs_devices->num_devices--;
1236 list_del(&device->dev_list);
1237 btrfs_free_device(device);
1240 if (fs_devices->open_devices == 0)
1243 fs_devices->opened = 1;
1244 fs_devices->latest_dev = latest_dev;
1245 fs_devices->total_rw_bytes = 0;
1246 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1247 fs_devices->read_policy = BTRFS_READ_POLICY_PID;
1252 static int devid_cmp(void *priv, const struct list_head *a,
1253 const struct list_head *b)
1255 const struct btrfs_device *dev1, *dev2;
1257 dev1 = list_entry(a, struct btrfs_device, dev_list);
1258 dev2 = list_entry(b, struct btrfs_device, dev_list);
1260 if (dev1->devid < dev2->devid)
1262 else if (dev1->devid > dev2->devid)
1267 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1268 blk_mode_t flags, void *holder)
1272 lockdep_assert_held(&uuid_mutex);
1274 * The device_list_mutex cannot be taken here in case opening the
1275 * underlying device takes further locks like open_mutex.
1277 * We also don't need the lock here as this is called during mount and
1278 * exclusion is provided by uuid_mutex
1281 if (fs_devices->opened) {
1282 fs_devices->opened++;
1285 list_sort(NULL, &fs_devices->devices, devid_cmp);
1286 ret = open_fs_devices(fs_devices, flags, holder);
1292 void btrfs_release_disk_super(struct btrfs_super_block *super)
1294 struct page *page = virt_to_page(super);
1299 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1300 u64 bytenr, u64 bytenr_orig)
1302 struct btrfs_super_block *disk_super;
1307 /* make sure our super fits in the device */
1308 if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev))
1309 return ERR_PTR(-EINVAL);
1311 /* make sure our super fits in the page */
1312 if (sizeof(*disk_super) > PAGE_SIZE)
1313 return ERR_PTR(-EINVAL);
1315 /* make sure our super doesn't straddle pages on disk */
1316 index = bytenr >> PAGE_SHIFT;
1317 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1318 return ERR_PTR(-EINVAL);
1320 /* pull in the page with our super */
1321 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1324 return ERR_CAST(page);
1326 p = page_address(page);
1328 /* align our pointer to the offset of the super block */
1329 disk_super = p + offset_in_page(bytenr);
1331 if (btrfs_super_bytenr(disk_super) != bytenr_orig ||
1332 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1333 btrfs_release_disk_super(p);
1334 return ERR_PTR(-EINVAL);
1337 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1338 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1343 int btrfs_forget_devices(dev_t devt)
1347 mutex_lock(&uuid_mutex);
1348 ret = btrfs_free_stale_devices(devt, NULL);
1349 mutex_unlock(&uuid_mutex);
1355 * Look for a btrfs signature on a device. This may be called out of the mount path
1356 * and we are not allowed to call set_blocksize during the scan. The superblock
1357 * is read via pagecache
1359 struct btrfs_device *btrfs_scan_one_device(const char *path, blk_mode_t flags)
1361 struct btrfs_super_block *disk_super;
1362 bool new_device_added = false;
1363 struct btrfs_device *device = NULL;
1364 struct block_device *bdev;
1365 u64 bytenr, bytenr_orig;
1368 lockdep_assert_held(&uuid_mutex);
1371 * we would like to check all the supers, but that would make
1372 * a btrfs mount succeed after a mkfs from a different FS.
1373 * So, we need to add a special mount option to scan for
1374 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1378 * Avoid an exclusive open here, as the systemd-udev may initiate the
1379 * device scan which may race with the user's mount or mkfs command,
1380 * resulting in failure.
1381 * Since the device scan is solely for reading purposes, there is no
1382 * need for an exclusive open. Additionally, the devices are read again
1383 * during the mount process. It is ok to get some inconsistent
1384 * values temporarily, as the device paths of the fsid are the only
1385 * required information for assembling the volume.
1387 bdev = blkdev_get_by_path(path, flags, NULL, NULL);
1389 return ERR_CAST(bdev);
1391 bytenr_orig = btrfs_sb_offset(0);
1392 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr);
1394 device = ERR_PTR(ret);
1395 goto error_bdev_put;
1398 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig);
1399 if (IS_ERR(disk_super)) {
1400 device = ERR_CAST(disk_super);
1401 goto error_bdev_put;
1404 device = device_list_add(path, disk_super, &new_device_added);
1405 if (!IS_ERR(device) && new_device_added)
1406 btrfs_free_stale_devices(device->devt, device);
1408 btrfs_release_disk_super(disk_super);
1411 blkdev_put(bdev, NULL);
1417 * Try to find a chunk that intersects [start, start + len] range and when one
1418 * such is found, record the end of it in *start
1420 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1423 u64 physical_start, physical_end;
1425 lockdep_assert_held(&device->fs_info->chunk_mutex);
1427 if (!find_first_extent_bit(&device->alloc_state, *start,
1428 &physical_start, &physical_end,
1429 CHUNK_ALLOCATED, NULL)) {
1431 if (in_range(physical_start, *start, len) ||
1432 in_range(*start, physical_start,
1433 physical_end - physical_start)) {
1434 *start = physical_end + 1;
1441 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start)
1443 switch (device->fs_devices->chunk_alloc_policy) {
1444 case BTRFS_CHUNK_ALLOC_REGULAR:
1445 return max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED);
1446 case BTRFS_CHUNK_ALLOC_ZONED:
1448 * We don't care about the starting region like regular
1449 * allocator, because we anyway use/reserve the first two zones
1450 * for superblock logging.
1452 return ALIGN(start, device->zone_info->zone_size);
1458 static bool dev_extent_hole_check_zoned(struct btrfs_device *device,
1459 u64 *hole_start, u64 *hole_size,
1462 u64 zone_size = device->zone_info->zone_size;
1465 bool changed = false;
1467 ASSERT(IS_ALIGNED(*hole_start, zone_size));
1469 while (*hole_size > 0) {
1470 pos = btrfs_find_allocatable_zones(device, *hole_start,
1471 *hole_start + *hole_size,
1473 if (pos != *hole_start) {
1474 *hole_size = *hole_start + *hole_size - pos;
1477 if (*hole_size < num_bytes)
1481 ret = btrfs_ensure_empty_zones(device, pos, num_bytes);
1483 /* Range is ensured to be empty */
1487 /* Given hole range was invalid (outside of device) */
1488 if (ret == -ERANGE) {
1489 *hole_start += *hole_size;
1494 *hole_start += zone_size;
1495 *hole_size -= zone_size;
1503 * Check if specified hole is suitable for allocation.
1505 * @device: the device which we have the hole
1506 * @hole_start: starting position of the hole
1507 * @hole_size: the size of the hole
1508 * @num_bytes: the size of the free space that we need
1510 * This function may modify @hole_start and @hole_size to reflect the suitable
1511 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1513 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1514 u64 *hole_size, u64 num_bytes)
1516 bool changed = false;
1517 u64 hole_end = *hole_start + *hole_size;
1521 * Check before we set max_hole_start, otherwise we could end up
1522 * sending back this offset anyway.
1524 if (contains_pending_extent(device, hole_start, *hole_size)) {
1525 if (hole_end >= *hole_start)
1526 *hole_size = hole_end - *hole_start;
1532 switch (device->fs_devices->chunk_alloc_policy) {
1533 case BTRFS_CHUNK_ALLOC_REGULAR:
1534 /* No extra check */
1536 case BTRFS_CHUNK_ALLOC_ZONED:
1537 if (dev_extent_hole_check_zoned(device, hole_start,
1538 hole_size, num_bytes)) {
1541 * The changed hole can contain pending extent.
1542 * Loop again to check that.
1558 * Find free space in the specified device.
1560 * @device: the device which we search the free space in
1561 * @num_bytes: the size of the free space that we need
1562 * @search_start: the position from which to begin the search
1563 * @start: store the start of the free space.
1564 * @len: the size of the free space. that we find, or the size
1565 * of the max free space if we don't find suitable free space
1567 * This does a pretty simple search, the expectation is that it is called very
1568 * infrequently and that a given device has a small number of extents.
1570 * @start is used to store the start of the free space if we find. But if we
1571 * don't find suitable free space, it will be used to store the start position
1572 * of the max free space.
1574 * @len is used to store the size of the free space that we find.
1575 * But if we don't find suitable free space, it is used to store the size of
1576 * the max free space.
1578 * NOTE: This function will search *commit* root of device tree, and does extra
1579 * check to ensure dev extents are not double allocated.
1580 * This makes the function safe to allocate dev extents but may not report
1581 * correct usable device space, as device extent freed in current transaction
1582 * is not reported as available.
1584 static int find_free_dev_extent_start(struct btrfs_device *device,
1585 u64 num_bytes, u64 search_start, u64 *start,
1588 struct btrfs_fs_info *fs_info = device->fs_info;
1589 struct btrfs_root *root = fs_info->dev_root;
1590 struct btrfs_key key;
1591 struct btrfs_dev_extent *dev_extent;
1592 struct btrfs_path *path;
1597 u64 search_end = device->total_bytes;
1600 struct extent_buffer *l;
1602 search_start = dev_extent_search_start(device, search_start);
1604 WARN_ON(device->zone_info &&
1605 !IS_ALIGNED(num_bytes, device->zone_info->zone_size));
1607 path = btrfs_alloc_path();
1611 max_hole_start = search_start;
1615 if (search_start >= search_end ||
1616 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1621 path->reada = READA_FORWARD;
1622 path->search_commit_root = 1;
1623 path->skip_locking = 1;
1625 key.objectid = device->devid;
1626 key.offset = search_start;
1627 key.type = BTRFS_DEV_EXTENT_KEY;
1629 ret = btrfs_search_backwards(root, &key, path);
1633 while (search_start < search_end) {
1635 slot = path->slots[0];
1636 if (slot >= btrfs_header_nritems(l)) {
1637 ret = btrfs_next_leaf(root, path);
1645 btrfs_item_key_to_cpu(l, &key, slot);
1647 if (key.objectid < device->devid)
1650 if (key.objectid > device->devid)
1653 if (key.type != BTRFS_DEV_EXTENT_KEY)
1656 if (key.offset > search_end)
1659 if (key.offset > search_start) {
1660 hole_size = key.offset - search_start;
1661 dev_extent_hole_check(device, &search_start, &hole_size,
1664 if (hole_size > max_hole_size) {
1665 max_hole_start = search_start;
1666 max_hole_size = hole_size;
1670 * If this free space is greater than which we need,
1671 * it must be the max free space that we have found
1672 * until now, so max_hole_start must point to the start
1673 * of this free space and the length of this free space
1674 * is stored in max_hole_size. Thus, we return
1675 * max_hole_start and max_hole_size and go back to the
1678 if (hole_size >= num_bytes) {
1684 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1685 extent_end = key.offset + btrfs_dev_extent_length(l,
1687 if (extent_end > search_start)
1688 search_start = extent_end;
1695 * At this point, search_start should be the end of
1696 * allocated dev extents, and when shrinking the device,
1697 * search_end may be smaller than search_start.
1699 if (search_end > search_start) {
1700 hole_size = search_end - search_start;
1701 if (dev_extent_hole_check(device, &search_start, &hole_size,
1703 btrfs_release_path(path);
1707 if (hole_size > max_hole_size) {
1708 max_hole_start = search_start;
1709 max_hole_size = hole_size;
1714 if (max_hole_size < num_bytes)
1719 ASSERT(max_hole_start + max_hole_size <= search_end);
1721 btrfs_free_path(path);
1722 *start = max_hole_start;
1724 *len = max_hole_size;
1728 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1729 u64 *start, u64 *len)
1731 /* FIXME use last free of some kind */
1732 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1735 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1736 struct btrfs_device *device,
1737 u64 start, u64 *dev_extent_len)
1739 struct btrfs_fs_info *fs_info = device->fs_info;
1740 struct btrfs_root *root = fs_info->dev_root;
1742 struct btrfs_path *path;
1743 struct btrfs_key key;
1744 struct btrfs_key found_key;
1745 struct extent_buffer *leaf = NULL;
1746 struct btrfs_dev_extent *extent = NULL;
1748 path = btrfs_alloc_path();
1752 key.objectid = device->devid;
1754 key.type = BTRFS_DEV_EXTENT_KEY;
1756 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1758 ret = btrfs_previous_item(root, path, key.objectid,
1759 BTRFS_DEV_EXTENT_KEY);
1762 leaf = path->nodes[0];
1763 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1764 extent = btrfs_item_ptr(leaf, path->slots[0],
1765 struct btrfs_dev_extent);
1766 BUG_ON(found_key.offset > start || found_key.offset +
1767 btrfs_dev_extent_length(leaf, extent) < start);
1769 btrfs_release_path(path);
1771 } else if (ret == 0) {
1772 leaf = path->nodes[0];
1773 extent = btrfs_item_ptr(leaf, path->slots[0],
1774 struct btrfs_dev_extent);
1779 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1781 ret = btrfs_del_item(trans, root, path);
1783 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1785 btrfs_free_path(path);
1789 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1791 struct extent_map_tree *em_tree;
1792 struct extent_map *em;
1796 em_tree = &fs_info->mapping_tree;
1797 read_lock(&em_tree->lock);
1798 n = rb_last(&em_tree->map.rb_root);
1800 em = rb_entry(n, struct extent_map, rb_node);
1801 ret = em->start + em->len;
1803 read_unlock(&em_tree->lock);
1808 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1812 struct btrfs_key key;
1813 struct btrfs_key found_key;
1814 struct btrfs_path *path;
1816 path = btrfs_alloc_path();
1820 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1821 key.type = BTRFS_DEV_ITEM_KEY;
1822 key.offset = (u64)-1;
1824 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1830 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1835 ret = btrfs_previous_item(fs_info->chunk_root, path,
1836 BTRFS_DEV_ITEMS_OBJECTID,
1837 BTRFS_DEV_ITEM_KEY);
1841 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1843 *devid_ret = found_key.offset + 1;
1847 btrfs_free_path(path);
1852 * the device information is stored in the chunk root
1853 * the btrfs_device struct should be fully filled in
1855 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1856 struct btrfs_device *device)
1859 struct btrfs_path *path;
1860 struct btrfs_dev_item *dev_item;
1861 struct extent_buffer *leaf;
1862 struct btrfs_key key;
1865 path = btrfs_alloc_path();
1869 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1870 key.type = BTRFS_DEV_ITEM_KEY;
1871 key.offset = device->devid;
1873 btrfs_reserve_chunk_metadata(trans, true);
1874 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1875 &key, sizeof(*dev_item));
1876 btrfs_trans_release_chunk_metadata(trans);
1880 leaf = path->nodes[0];
1881 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1883 btrfs_set_device_id(leaf, dev_item, device->devid);
1884 btrfs_set_device_generation(leaf, dev_item, 0);
1885 btrfs_set_device_type(leaf, dev_item, device->type);
1886 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1887 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1888 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1889 btrfs_set_device_total_bytes(leaf, dev_item,
1890 btrfs_device_get_disk_total_bytes(device));
1891 btrfs_set_device_bytes_used(leaf, dev_item,
1892 btrfs_device_get_bytes_used(device));
1893 btrfs_set_device_group(leaf, dev_item, 0);
1894 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1895 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1896 btrfs_set_device_start_offset(leaf, dev_item, 0);
1898 ptr = btrfs_device_uuid(dev_item);
1899 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1900 ptr = btrfs_device_fsid(dev_item);
1901 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1902 ptr, BTRFS_FSID_SIZE);
1903 btrfs_mark_buffer_dirty(leaf);
1907 btrfs_free_path(path);
1912 * Function to update ctime/mtime for a given device path.
1913 * Mainly used for ctime/mtime based probe like libblkid.
1915 * We don't care about errors here, this is just to be kind to userspace.
1917 static void update_dev_time(const char *device_path)
1920 struct timespec64 now;
1923 ret = kern_path(device_path, LOOKUP_FOLLOW, &path);
1927 now = current_time(d_inode(path.dentry));
1928 inode_update_time(d_inode(path.dentry), &now, S_MTIME | S_CTIME | S_VERSION);
1932 static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans,
1933 struct btrfs_device *device)
1935 struct btrfs_root *root = device->fs_info->chunk_root;
1937 struct btrfs_path *path;
1938 struct btrfs_key key;
1940 path = btrfs_alloc_path();
1944 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1945 key.type = BTRFS_DEV_ITEM_KEY;
1946 key.offset = device->devid;
1948 btrfs_reserve_chunk_metadata(trans, false);
1949 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1950 btrfs_trans_release_chunk_metadata(trans);
1957 ret = btrfs_del_item(trans, root, path);
1959 btrfs_free_path(path);
1964 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1965 * filesystem. It's up to the caller to adjust that number regarding eg. device
1968 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1976 seq = read_seqbegin(&fs_info->profiles_lock);
1978 all_avail = fs_info->avail_data_alloc_bits |
1979 fs_info->avail_system_alloc_bits |
1980 fs_info->avail_metadata_alloc_bits;
1981 } while (read_seqretry(&fs_info->profiles_lock, seq));
1983 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1984 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1987 if (num_devices < btrfs_raid_array[i].devs_min)
1988 return btrfs_raid_array[i].mindev_error;
1994 static struct btrfs_device * btrfs_find_next_active_device(
1995 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1997 struct btrfs_device *next_device;
1999 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
2000 if (next_device != device &&
2001 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
2002 && next_device->bdev)
2010 * Helper function to check if the given device is part of s_bdev / latest_dev
2011 * and replace it with the provided or the next active device, in the context
2012 * where this function called, there should be always be another device (or
2013 * this_dev) which is active.
2015 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
2016 struct btrfs_device *next_device)
2018 struct btrfs_fs_info *fs_info = device->fs_info;
2021 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
2023 ASSERT(next_device);
2025 if (fs_info->sb->s_bdev &&
2026 (fs_info->sb->s_bdev == device->bdev))
2027 fs_info->sb->s_bdev = next_device->bdev;
2029 if (fs_info->fs_devices->latest_dev->bdev == device->bdev)
2030 fs_info->fs_devices->latest_dev = next_device;
2034 * Return btrfs_fs_devices::num_devices excluding the device that's being
2035 * currently replaced.
2037 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2039 u64 num_devices = fs_info->fs_devices->num_devices;
2041 down_read(&fs_info->dev_replace.rwsem);
2042 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2043 ASSERT(num_devices > 1);
2046 up_read(&fs_info->dev_replace.rwsem);
2051 static void btrfs_scratch_superblock(struct btrfs_fs_info *fs_info,
2052 struct block_device *bdev, int copy_num)
2054 struct btrfs_super_block *disk_super;
2055 const size_t len = sizeof(disk_super->magic);
2056 const u64 bytenr = btrfs_sb_offset(copy_num);
2059 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr);
2060 if (IS_ERR(disk_super))
2063 memset(&disk_super->magic, 0, len);
2064 folio_mark_dirty(virt_to_folio(disk_super));
2065 btrfs_release_disk_super(disk_super);
2067 ret = sync_blockdev_range(bdev, bytenr, bytenr + len - 1);
2069 btrfs_warn(fs_info, "error clearing superblock number %d (%d)",
2073 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
2074 struct block_device *bdev,
2075 const char *device_path)
2082 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2083 if (bdev_is_zoned(bdev))
2084 btrfs_reset_sb_log_zones(bdev, copy_num);
2086 btrfs_scratch_superblock(fs_info, bdev, copy_num);
2089 /* Notify udev that device has changed */
2090 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2092 /* Update ctime/mtime for device path for libblkid */
2093 update_dev_time(device_path);
2096 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
2097 struct btrfs_dev_lookup_args *args,
2098 struct block_device **bdev, void **holder)
2100 struct btrfs_trans_handle *trans;
2101 struct btrfs_device *device;
2102 struct btrfs_fs_devices *cur_devices;
2103 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2107 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2108 btrfs_err(fs_info, "device remove not supported on extent tree v2 yet");
2113 * The device list in fs_devices is accessed without locks (neither
2114 * uuid_mutex nor device_list_mutex) as it won't change on a mounted
2115 * filesystem and another device rm cannot run.
2117 num_devices = btrfs_num_devices(fs_info);
2119 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2123 device = btrfs_find_device(fs_info->fs_devices, args);
2126 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2132 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2133 btrfs_warn_in_rcu(fs_info,
2134 "cannot remove device %s (devid %llu) due to active swapfile",
2135 btrfs_dev_name(device), device->devid);
2139 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
2140 return BTRFS_ERROR_DEV_TGT_REPLACE;
2142 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2143 fs_info->fs_devices->rw_devices == 1)
2144 return BTRFS_ERROR_DEV_ONLY_WRITABLE;
2146 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2147 mutex_lock(&fs_info->chunk_mutex);
2148 list_del_init(&device->dev_alloc_list);
2149 device->fs_devices->rw_devices--;
2150 mutex_unlock(&fs_info->chunk_mutex);
2153 ret = btrfs_shrink_device(device, 0);
2157 trans = btrfs_start_transaction(fs_info->chunk_root, 0);
2158 if (IS_ERR(trans)) {
2159 ret = PTR_ERR(trans);
2163 ret = btrfs_rm_dev_item(trans, device);
2165 /* Any error in dev item removal is critical */
2167 "failed to remove device item for devid %llu: %d",
2168 device->devid, ret);
2169 btrfs_abort_transaction(trans, ret);
2170 btrfs_end_transaction(trans);
2174 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2175 btrfs_scrub_cancel_dev(device);
2178 * the device list mutex makes sure that we don't change
2179 * the device list while someone else is writing out all
2180 * the device supers. Whoever is writing all supers, should
2181 * lock the device list mutex before getting the number of
2182 * devices in the super block (super_copy). Conversely,
2183 * whoever updates the number of devices in the super block
2184 * (super_copy) should hold the device list mutex.
2188 * In normal cases the cur_devices == fs_devices. But in case
2189 * of deleting a seed device, the cur_devices should point to
2190 * its own fs_devices listed under the fs_devices->seed_list.
2192 cur_devices = device->fs_devices;
2193 mutex_lock(&fs_devices->device_list_mutex);
2194 list_del_rcu(&device->dev_list);
2196 cur_devices->num_devices--;
2197 cur_devices->total_devices--;
2198 /* Update total_devices of the parent fs_devices if it's seed */
2199 if (cur_devices != fs_devices)
2200 fs_devices->total_devices--;
2202 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2203 cur_devices->missing_devices--;
2205 btrfs_assign_next_active_device(device, NULL);
2208 cur_devices->open_devices--;
2209 /* remove sysfs entry */
2210 btrfs_sysfs_remove_device(device);
2213 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2214 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2215 mutex_unlock(&fs_devices->device_list_mutex);
2218 * At this point, the device is zero sized and detached from the
2219 * devices list. All that's left is to zero out the old supers and
2222 * We cannot call btrfs_close_bdev() here because we're holding the sb
2223 * write lock, and blkdev_put() will pull in the ->open_mutex on the
2224 * block device and it's dependencies. Instead just flush the device
2225 * and let the caller do the final blkdev_put.
2227 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2228 btrfs_scratch_superblocks(fs_info, device->bdev,
2231 sync_blockdev(device->bdev);
2232 invalidate_bdev(device->bdev);
2236 *bdev = device->bdev;
2237 *holder = device->holder;
2239 btrfs_free_device(device);
2242 * This can happen if cur_devices is the private seed devices list. We
2243 * cannot call close_fs_devices() here because it expects the uuid_mutex
2244 * to be held, but in fact we don't need that for the private
2245 * seed_devices, we can simply decrement cur_devices->opened and then
2246 * remove it from our list and free the fs_devices.
2248 if (cur_devices->num_devices == 0) {
2249 list_del_init(&cur_devices->seed_list);
2250 ASSERT(cur_devices->opened == 1);
2251 cur_devices->opened--;
2252 free_fs_devices(cur_devices);
2255 ret = btrfs_commit_transaction(trans);
2260 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2261 mutex_lock(&fs_info->chunk_mutex);
2262 list_add(&device->dev_alloc_list,
2263 &fs_devices->alloc_list);
2264 device->fs_devices->rw_devices++;
2265 mutex_unlock(&fs_info->chunk_mutex);
2270 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2272 struct btrfs_fs_devices *fs_devices;
2274 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2277 * in case of fs with no seed, srcdev->fs_devices will point
2278 * to fs_devices of fs_info. However when the dev being replaced is
2279 * a seed dev it will point to the seed's local fs_devices. In short
2280 * srcdev will have its correct fs_devices in both the cases.
2282 fs_devices = srcdev->fs_devices;
2284 list_del_rcu(&srcdev->dev_list);
2285 list_del(&srcdev->dev_alloc_list);
2286 fs_devices->num_devices--;
2287 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2288 fs_devices->missing_devices--;
2290 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2291 fs_devices->rw_devices--;
2294 fs_devices->open_devices--;
2297 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2299 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2301 mutex_lock(&uuid_mutex);
2303 btrfs_close_bdev(srcdev);
2305 btrfs_free_device(srcdev);
2307 /* if this is no devs we rather delete the fs_devices */
2308 if (!fs_devices->num_devices) {
2310 * On a mounted FS, num_devices can't be zero unless it's a
2311 * seed. In case of a seed device being replaced, the replace
2312 * target added to the sprout FS, so there will be no more
2313 * device left under the seed FS.
2315 ASSERT(fs_devices->seeding);
2317 list_del_init(&fs_devices->seed_list);
2318 close_fs_devices(fs_devices);
2319 free_fs_devices(fs_devices);
2321 mutex_unlock(&uuid_mutex);
2324 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2326 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2328 mutex_lock(&fs_devices->device_list_mutex);
2330 btrfs_sysfs_remove_device(tgtdev);
2333 fs_devices->open_devices--;
2335 fs_devices->num_devices--;
2337 btrfs_assign_next_active_device(tgtdev, NULL);
2339 list_del_rcu(&tgtdev->dev_list);
2341 mutex_unlock(&fs_devices->device_list_mutex);
2343 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2346 btrfs_close_bdev(tgtdev);
2348 btrfs_free_device(tgtdev);
2352 * Populate args from device at path.
2354 * @fs_info: the filesystem
2355 * @args: the args to populate
2356 * @path: the path to the device
2358 * This will read the super block of the device at @path and populate @args with
2359 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to
2360 * lookup a device to operate on, but need to do it before we take any locks.
2361 * This properly handles the special case of "missing" that a user may pass in,
2362 * and does some basic sanity checks. The caller must make sure that @path is
2363 * properly NUL terminated before calling in, and must call
2364 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and
2367 * Return: 0 for success, -errno for failure
2369 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,
2370 struct btrfs_dev_lookup_args *args,
2373 struct btrfs_super_block *disk_super;
2374 struct block_device *bdev;
2377 if (!path || !path[0])
2379 if (!strcmp(path, "missing")) {
2380 args->missing = true;
2384 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL);
2385 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL);
2386 if (!args->uuid || !args->fsid) {
2387 btrfs_put_dev_args_from_path(args);
2391 ret = btrfs_get_bdev_and_sb(path, BLK_OPEN_READ, NULL, 0,
2392 &bdev, &disk_super);
2394 btrfs_put_dev_args_from_path(args);
2398 args->devid = btrfs_stack_device_id(&disk_super->dev_item);
2399 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE);
2400 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2401 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE);
2403 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
2404 btrfs_release_disk_super(disk_super);
2405 blkdev_put(bdev, NULL);
2410 * Only use this jointly with btrfs_get_dev_args_from_path() because we will
2411 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables
2412 * that don't need to be freed.
2414 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args)
2422 struct btrfs_device *btrfs_find_device_by_devspec(
2423 struct btrfs_fs_info *fs_info, u64 devid,
2424 const char *device_path)
2426 BTRFS_DEV_LOOKUP_ARGS(args);
2427 struct btrfs_device *device;
2432 device = btrfs_find_device(fs_info->fs_devices, &args);
2434 return ERR_PTR(-ENOENT);
2438 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path);
2440 return ERR_PTR(ret);
2441 device = btrfs_find_device(fs_info->fs_devices, &args);
2442 btrfs_put_dev_args_from_path(&args);
2444 return ERR_PTR(-ENOENT);
2448 static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info)
2450 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2451 struct btrfs_fs_devices *old_devices;
2452 struct btrfs_fs_devices *seed_devices;
2454 lockdep_assert_held(&uuid_mutex);
2455 if (!fs_devices->seeding)
2456 return ERR_PTR(-EINVAL);
2459 * Private copy of the seed devices, anchored at
2460 * fs_info->fs_devices->seed_list
2462 seed_devices = alloc_fs_devices(NULL, NULL);
2463 if (IS_ERR(seed_devices))
2464 return seed_devices;
2467 * It's necessary to retain a copy of the original seed fs_devices in
2468 * fs_uuids so that filesystems which have been seeded can successfully
2469 * reference the seed device from open_seed_devices. This also supports
2472 old_devices = clone_fs_devices(fs_devices);
2473 if (IS_ERR(old_devices)) {
2474 kfree(seed_devices);
2478 list_add(&old_devices->fs_list, &fs_uuids);
2480 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2481 seed_devices->opened = 1;
2482 INIT_LIST_HEAD(&seed_devices->devices);
2483 INIT_LIST_HEAD(&seed_devices->alloc_list);
2484 mutex_init(&seed_devices->device_list_mutex);
2486 return seed_devices;
2490 * Splice seed devices into the sprout fs_devices.
2491 * Generate a new fsid for the sprouted read-write filesystem.
2493 static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info,
2494 struct btrfs_fs_devices *seed_devices)
2496 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2497 struct btrfs_super_block *disk_super = fs_info->super_copy;
2498 struct btrfs_device *device;
2502 * We are updating the fsid, the thread leading to device_list_add()
2503 * could race, so uuid_mutex is needed.
2505 lockdep_assert_held(&uuid_mutex);
2508 * The threads listed below may traverse dev_list but can do that without
2509 * device_list_mutex:
2510 * - All device ops and balance - as we are in btrfs_exclop_start.
2511 * - Various dev_list readers - are using RCU.
2512 * - btrfs_ioctl_fitrim() - is using RCU.
2514 * For-read threads as below are using device_list_mutex:
2515 * - Readonly scrub btrfs_scrub_dev()
2516 * - Readonly scrub btrfs_scrub_progress()
2517 * - btrfs_get_dev_stats()
2519 lockdep_assert_held(&fs_devices->device_list_mutex);
2521 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2523 list_for_each_entry(device, &seed_devices->devices, dev_list)
2524 device->fs_devices = seed_devices;
2526 fs_devices->seeding = false;
2527 fs_devices->num_devices = 0;
2528 fs_devices->open_devices = 0;
2529 fs_devices->missing_devices = 0;
2530 fs_devices->rotating = false;
2531 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2533 generate_random_uuid(fs_devices->fsid);
2534 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2535 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2537 super_flags = btrfs_super_flags(disk_super) &
2538 ~BTRFS_SUPER_FLAG_SEEDING;
2539 btrfs_set_super_flags(disk_super, super_flags);
2543 * Store the expected generation for seed devices in device items.
2545 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2547 BTRFS_DEV_LOOKUP_ARGS(args);
2548 struct btrfs_fs_info *fs_info = trans->fs_info;
2549 struct btrfs_root *root = fs_info->chunk_root;
2550 struct btrfs_path *path;
2551 struct extent_buffer *leaf;
2552 struct btrfs_dev_item *dev_item;
2553 struct btrfs_device *device;
2554 struct btrfs_key key;
2555 u8 fs_uuid[BTRFS_FSID_SIZE];
2556 u8 dev_uuid[BTRFS_UUID_SIZE];
2559 path = btrfs_alloc_path();
2563 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2565 key.type = BTRFS_DEV_ITEM_KEY;
2568 btrfs_reserve_chunk_metadata(trans, false);
2569 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2570 btrfs_trans_release_chunk_metadata(trans);
2574 leaf = path->nodes[0];
2576 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2577 ret = btrfs_next_leaf(root, path);
2582 leaf = path->nodes[0];
2583 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2584 btrfs_release_path(path);
2588 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2589 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2590 key.type != BTRFS_DEV_ITEM_KEY)
2593 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2594 struct btrfs_dev_item);
2595 args.devid = btrfs_device_id(leaf, dev_item);
2596 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2598 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2600 args.uuid = dev_uuid;
2601 args.fsid = fs_uuid;
2602 device = btrfs_find_device(fs_info->fs_devices, &args);
2603 BUG_ON(!device); /* Logic error */
2605 if (device->fs_devices->seeding) {
2606 btrfs_set_device_generation(leaf, dev_item,
2607 device->generation);
2608 btrfs_mark_buffer_dirty(leaf);
2616 btrfs_free_path(path);
2620 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2622 struct btrfs_root *root = fs_info->dev_root;
2623 struct btrfs_trans_handle *trans;
2624 struct btrfs_device *device;
2625 struct block_device *bdev;
2626 struct super_block *sb = fs_info->sb;
2627 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2628 struct btrfs_fs_devices *seed_devices = NULL;
2629 u64 orig_super_total_bytes;
2630 u64 orig_super_num_devices;
2632 bool seeding_dev = false;
2633 bool locked = false;
2635 if (sb_rdonly(sb) && !fs_devices->seeding)
2638 bdev = blkdev_get_by_path(device_path, BLK_OPEN_WRITE,
2639 fs_info->bdev_holder, NULL);
2641 return PTR_ERR(bdev);
2643 if (!btrfs_check_device_zone_type(fs_info, bdev)) {
2648 if (fs_devices->seeding) {
2650 down_write(&sb->s_umount);
2651 mutex_lock(&uuid_mutex);
2655 sync_blockdev(bdev);
2658 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2659 if (device->bdev == bdev) {
2667 device = btrfs_alloc_device(fs_info, NULL, NULL, device_path);
2668 if (IS_ERR(device)) {
2669 /* we can safely leave the fs_devices entry around */
2670 ret = PTR_ERR(device);
2674 device->fs_info = fs_info;
2675 device->bdev = bdev;
2676 ret = lookup_bdev(device_path, &device->devt);
2678 goto error_free_device;
2680 ret = btrfs_get_dev_zone_info(device, false);
2682 goto error_free_device;
2684 trans = btrfs_start_transaction(root, 0);
2685 if (IS_ERR(trans)) {
2686 ret = PTR_ERR(trans);
2687 goto error_free_zone;
2690 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2691 device->generation = trans->transid;
2692 device->io_width = fs_info->sectorsize;
2693 device->io_align = fs_info->sectorsize;
2694 device->sector_size = fs_info->sectorsize;
2695 device->total_bytes =
2696 round_down(bdev_nr_bytes(bdev), fs_info->sectorsize);
2697 device->disk_total_bytes = device->total_bytes;
2698 device->commit_total_bytes = device->total_bytes;
2699 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2700 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2701 device->holder = fs_info->bdev_holder;
2702 device->dev_stats_valid = 1;
2703 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2706 btrfs_clear_sb_rdonly(sb);
2708 /* GFP_KERNEL allocation must not be under device_list_mutex */
2709 seed_devices = btrfs_init_sprout(fs_info);
2710 if (IS_ERR(seed_devices)) {
2711 ret = PTR_ERR(seed_devices);
2712 btrfs_abort_transaction(trans, ret);
2717 mutex_lock(&fs_devices->device_list_mutex);
2719 btrfs_setup_sprout(fs_info, seed_devices);
2720 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev,
2724 device->fs_devices = fs_devices;
2726 mutex_lock(&fs_info->chunk_mutex);
2727 list_add_rcu(&device->dev_list, &fs_devices->devices);
2728 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2729 fs_devices->num_devices++;
2730 fs_devices->open_devices++;
2731 fs_devices->rw_devices++;
2732 fs_devices->total_devices++;
2733 fs_devices->total_rw_bytes += device->total_bytes;
2735 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2737 if (!bdev_nonrot(bdev))
2738 fs_devices->rotating = true;
2740 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2741 btrfs_set_super_total_bytes(fs_info->super_copy,
2742 round_down(orig_super_total_bytes + device->total_bytes,
2743 fs_info->sectorsize));
2745 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2746 btrfs_set_super_num_devices(fs_info->super_copy,
2747 orig_super_num_devices + 1);
2750 * we've got more storage, clear any full flags on the space
2753 btrfs_clear_space_info_full(fs_info);
2755 mutex_unlock(&fs_info->chunk_mutex);
2757 /* Add sysfs device entry */
2758 btrfs_sysfs_add_device(device);
2760 mutex_unlock(&fs_devices->device_list_mutex);
2763 mutex_lock(&fs_info->chunk_mutex);
2764 ret = init_first_rw_device(trans);
2765 mutex_unlock(&fs_info->chunk_mutex);
2767 btrfs_abort_transaction(trans, ret);
2772 ret = btrfs_add_dev_item(trans, device);
2774 btrfs_abort_transaction(trans, ret);
2779 ret = btrfs_finish_sprout(trans);
2781 btrfs_abort_transaction(trans, ret);
2786 * fs_devices now represents the newly sprouted filesystem and
2787 * its fsid has been changed by btrfs_sprout_splice().
2789 btrfs_sysfs_update_sprout_fsid(fs_devices);
2792 ret = btrfs_commit_transaction(trans);
2795 mutex_unlock(&uuid_mutex);
2796 up_write(&sb->s_umount);
2799 if (ret) /* transaction commit */
2802 ret = btrfs_relocate_sys_chunks(fs_info);
2804 btrfs_handle_fs_error(fs_info, ret,
2805 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2806 trans = btrfs_attach_transaction(root);
2807 if (IS_ERR(trans)) {
2808 if (PTR_ERR(trans) == -ENOENT)
2810 ret = PTR_ERR(trans);
2814 ret = btrfs_commit_transaction(trans);
2818 * Now that we have written a new super block to this device, check all
2819 * other fs_devices list if device_path alienates any other scanned
2821 * We can ignore the return value as it typically returns -EINVAL and
2822 * only succeeds if the device was an alien.
2824 btrfs_forget_devices(device->devt);
2826 /* Update ctime/mtime for blkid or udev */
2827 update_dev_time(device_path);
2832 btrfs_sysfs_remove_device(device);
2833 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2834 mutex_lock(&fs_info->chunk_mutex);
2835 list_del_rcu(&device->dev_list);
2836 list_del(&device->dev_alloc_list);
2837 fs_info->fs_devices->num_devices--;
2838 fs_info->fs_devices->open_devices--;
2839 fs_info->fs_devices->rw_devices--;
2840 fs_info->fs_devices->total_devices--;
2841 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2842 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2843 btrfs_set_super_total_bytes(fs_info->super_copy,
2844 orig_super_total_bytes);
2845 btrfs_set_super_num_devices(fs_info->super_copy,
2846 orig_super_num_devices);
2847 mutex_unlock(&fs_info->chunk_mutex);
2848 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2851 btrfs_set_sb_rdonly(sb);
2853 btrfs_end_transaction(trans);
2855 btrfs_destroy_dev_zone_info(device);
2857 btrfs_free_device(device);
2859 blkdev_put(bdev, fs_info->bdev_holder);
2861 mutex_unlock(&uuid_mutex);
2862 up_write(&sb->s_umount);
2867 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2868 struct btrfs_device *device)
2871 struct btrfs_path *path;
2872 struct btrfs_root *root = device->fs_info->chunk_root;
2873 struct btrfs_dev_item *dev_item;
2874 struct extent_buffer *leaf;
2875 struct btrfs_key key;
2877 path = btrfs_alloc_path();
2881 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2882 key.type = BTRFS_DEV_ITEM_KEY;
2883 key.offset = device->devid;
2885 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2894 leaf = path->nodes[0];
2895 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2897 btrfs_set_device_id(leaf, dev_item, device->devid);
2898 btrfs_set_device_type(leaf, dev_item, device->type);
2899 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2900 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2901 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2902 btrfs_set_device_total_bytes(leaf, dev_item,
2903 btrfs_device_get_disk_total_bytes(device));
2904 btrfs_set_device_bytes_used(leaf, dev_item,
2905 btrfs_device_get_bytes_used(device));
2906 btrfs_mark_buffer_dirty(leaf);
2909 btrfs_free_path(path);
2913 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2914 struct btrfs_device *device, u64 new_size)
2916 struct btrfs_fs_info *fs_info = device->fs_info;
2917 struct btrfs_super_block *super_copy = fs_info->super_copy;
2922 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2925 new_size = round_down(new_size, fs_info->sectorsize);
2927 mutex_lock(&fs_info->chunk_mutex);
2928 old_total = btrfs_super_total_bytes(super_copy);
2929 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2931 if (new_size <= device->total_bytes ||
2932 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2933 mutex_unlock(&fs_info->chunk_mutex);
2937 btrfs_set_super_total_bytes(super_copy,
2938 round_down(old_total + diff, fs_info->sectorsize));
2939 device->fs_devices->total_rw_bytes += diff;
2941 btrfs_device_set_total_bytes(device, new_size);
2942 btrfs_device_set_disk_total_bytes(device, new_size);
2943 btrfs_clear_space_info_full(device->fs_info);
2944 if (list_empty(&device->post_commit_list))
2945 list_add_tail(&device->post_commit_list,
2946 &trans->transaction->dev_update_list);
2947 mutex_unlock(&fs_info->chunk_mutex);
2949 btrfs_reserve_chunk_metadata(trans, false);
2950 ret = btrfs_update_device(trans, device);
2951 btrfs_trans_release_chunk_metadata(trans);
2956 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2958 struct btrfs_fs_info *fs_info = trans->fs_info;
2959 struct btrfs_root *root = fs_info->chunk_root;
2961 struct btrfs_path *path;
2962 struct btrfs_key key;
2964 path = btrfs_alloc_path();
2968 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2969 key.offset = chunk_offset;
2970 key.type = BTRFS_CHUNK_ITEM_KEY;
2972 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2975 else if (ret > 0) { /* Logic error or corruption */
2976 btrfs_handle_fs_error(fs_info, -ENOENT,
2977 "Failed lookup while freeing chunk.");
2982 ret = btrfs_del_item(trans, root, path);
2984 btrfs_handle_fs_error(fs_info, ret,
2985 "Failed to delete chunk item.");
2987 btrfs_free_path(path);
2991 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2993 struct btrfs_super_block *super_copy = fs_info->super_copy;
2994 struct btrfs_disk_key *disk_key;
2995 struct btrfs_chunk *chunk;
3002 struct btrfs_key key;
3004 lockdep_assert_held(&fs_info->chunk_mutex);
3005 array_size = btrfs_super_sys_array_size(super_copy);
3007 ptr = super_copy->sys_chunk_array;
3010 while (cur < array_size) {
3011 disk_key = (struct btrfs_disk_key *)ptr;
3012 btrfs_disk_key_to_cpu(&key, disk_key);
3014 len = sizeof(*disk_key);
3016 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3017 chunk = (struct btrfs_chunk *)(ptr + len);
3018 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
3019 len += btrfs_chunk_item_size(num_stripes);
3024 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
3025 key.offset == chunk_offset) {
3026 memmove(ptr, ptr + len, array_size - (cur + len));
3028 btrfs_set_super_sys_array_size(super_copy, array_size);
3038 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
3039 * @logical: Logical block offset in bytes.
3040 * @length: Length of extent in bytes.
3042 * Return: Chunk mapping or ERR_PTR.
3044 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
3045 u64 logical, u64 length)
3047 struct extent_map_tree *em_tree;
3048 struct extent_map *em;
3050 em_tree = &fs_info->mapping_tree;
3051 read_lock(&em_tree->lock);
3052 em = lookup_extent_mapping(em_tree, logical, length);
3053 read_unlock(&em_tree->lock);
3056 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
3058 return ERR_PTR(-EINVAL);
3061 if (em->start > logical || em->start + em->len < logical) {
3063 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
3064 logical, length, em->start, em->start + em->len);
3065 free_extent_map(em);
3066 return ERR_PTR(-EINVAL);
3069 /* callers are responsible for dropping em's ref. */
3073 static int remove_chunk_item(struct btrfs_trans_handle *trans,
3074 struct map_lookup *map, u64 chunk_offset)
3079 * Removing chunk items and updating the device items in the chunks btree
3080 * requires holding the chunk_mutex.
3081 * See the comment at btrfs_chunk_alloc() for the details.
3083 lockdep_assert_held(&trans->fs_info->chunk_mutex);
3085 for (i = 0; i < map->num_stripes; i++) {
3088 ret = btrfs_update_device(trans, map->stripes[i].dev);
3093 return btrfs_free_chunk(trans, chunk_offset);
3096 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
3098 struct btrfs_fs_info *fs_info = trans->fs_info;
3099 struct extent_map *em;
3100 struct map_lookup *map;
3101 u64 dev_extent_len = 0;
3103 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3105 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
3108 * This is a logic error, but we don't want to just rely on the
3109 * user having built with ASSERT enabled, so if ASSERT doesn't
3110 * do anything we still error out.
3115 map = em->map_lookup;
3118 * First delete the device extent items from the devices btree.
3119 * We take the device_list_mutex to avoid racing with the finishing phase
3120 * of a device replace operation. See the comment below before acquiring
3121 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
3122 * because that can result in a deadlock when deleting the device extent
3123 * items from the devices btree - COWing an extent buffer from the btree
3124 * may result in allocating a new metadata chunk, which would attempt to
3125 * lock again fs_info->chunk_mutex.
3127 mutex_lock(&fs_devices->device_list_mutex);
3128 for (i = 0; i < map->num_stripes; i++) {
3129 struct btrfs_device *device = map->stripes[i].dev;
3130 ret = btrfs_free_dev_extent(trans, device,
3131 map->stripes[i].physical,
3134 mutex_unlock(&fs_devices->device_list_mutex);
3135 btrfs_abort_transaction(trans, ret);
3139 if (device->bytes_used > 0) {
3140 mutex_lock(&fs_info->chunk_mutex);
3141 btrfs_device_set_bytes_used(device,
3142 device->bytes_used - dev_extent_len);
3143 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3144 btrfs_clear_space_info_full(fs_info);
3145 mutex_unlock(&fs_info->chunk_mutex);
3148 mutex_unlock(&fs_devices->device_list_mutex);
3151 * We acquire fs_info->chunk_mutex for 2 reasons:
3153 * 1) Just like with the first phase of the chunk allocation, we must
3154 * reserve system space, do all chunk btree updates and deletions, and
3155 * update the system chunk array in the superblock while holding this
3156 * mutex. This is for similar reasons as explained on the comment at
3157 * the top of btrfs_chunk_alloc();
3159 * 2) Prevent races with the final phase of a device replace operation
3160 * that replaces the device object associated with the map's stripes,
3161 * because the device object's id can change at any time during that
3162 * final phase of the device replace operation
3163 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3164 * replaced device and then see it with an ID of
3165 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3166 * the device item, which does not exists on the chunk btree.
3167 * The finishing phase of device replace acquires both the
3168 * device_list_mutex and the chunk_mutex, in that order, so we are
3169 * safe by just acquiring the chunk_mutex.
3171 trans->removing_chunk = true;
3172 mutex_lock(&fs_info->chunk_mutex);
3174 check_system_chunk(trans, map->type);
3176 ret = remove_chunk_item(trans, map, chunk_offset);
3178 * Normally we should not get -ENOSPC since we reserved space before
3179 * through the call to check_system_chunk().
3181 * Despite our system space_info having enough free space, we may not
3182 * be able to allocate extents from its block groups, because all have
3183 * an incompatible profile, which will force us to allocate a new system
3184 * block group with the right profile, or right after we called
3185 * check_system_space() above, a scrub turned the only system block group
3186 * with enough free space into RO mode.
3187 * This is explained with more detail at do_chunk_alloc().
3189 * So if we get -ENOSPC, allocate a new system chunk and retry once.
3191 if (ret == -ENOSPC) {
3192 const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
3193 struct btrfs_block_group *sys_bg;
3195 sys_bg = btrfs_create_chunk(trans, sys_flags);
3196 if (IS_ERR(sys_bg)) {
3197 ret = PTR_ERR(sys_bg);
3198 btrfs_abort_transaction(trans, ret);
3202 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3204 btrfs_abort_transaction(trans, ret);
3208 ret = remove_chunk_item(trans, map, chunk_offset);
3210 btrfs_abort_transaction(trans, ret);
3214 btrfs_abort_transaction(trans, ret);
3218 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3220 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3221 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3223 btrfs_abort_transaction(trans, ret);
3228 mutex_unlock(&fs_info->chunk_mutex);
3229 trans->removing_chunk = false;
3232 * We are done with chunk btree updates and deletions, so release the
3233 * system space we previously reserved (with check_system_chunk()).
3235 btrfs_trans_release_chunk_metadata(trans);
3237 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3239 btrfs_abort_transaction(trans, ret);
3244 if (trans->removing_chunk) {
3245 mutex_unlock(&fs_info->chunk_mutex);
3246 trans->removing_chunk = false;
3249 free_extent_map(em);
3253 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3255 struct btrfs_root *root = fs_info->chunk_root;
3256 struct btrfs_trans_handle *trans;
3257 struct btrfs_block_group *block_group;
3261 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3263 "relocate: not supported on extent tree v2 yet");
3268 * Prevent races with automatic removal of unused block groups.
3269 * After we relocate and before we remove the chunk with offset
3270 * chunk_offset, automatic removal of the block group can kick in,
3271 * resulting in a failure when calling btrfs_remove_chunk() below.
3273 * Make sure to acquire this mutex before doing a tree search (dev
3274 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3275 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3276 * we release the path used to search the chunk/dev tree and before
3277 * the current task acquires this mutex and calls us.
3279 lockdep_assert_held(&fs_info->reclaim_bgs_lock);
3281 /* step one, relocate all the extents inside this chunk */
3282 btrfs_scrub_pause(fs_info);
3283 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3284 btrfs_scrub_continue(fs_info);
3287 * If we had a transaction abort, stop all running scrubs.
3288 * See transaction.c:cleanup_transaction() why we do it here.
3290 if (BTRFS_FS_ERROR(fs_info))
3291 btrfs_scrub_cancel(fs_info);
3295 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3298 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3299 length = block_group->length;
3300 btrfs_put_block_group(block_group);
3303 * On a zoned file system, discard the whole block group, this will
3304 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3305 * resetting the zone fails, don't treat it as a fatal problem from the
3306 * filesystem's point of view.
3308 if (btrfs_is_zoned(fs_info)) {
3309 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
3312 "failed to reset zone %llu after relocation",
3316 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3318 if (IS_ERR(trans)) {
3319 ret = PTR_ERR(trans);
3320 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3325 * step two, delete the device extents and the
3326 * chunk tree entries
3328 ret = btrfs_remove_chunk(trans, chunk_offset);
3329 btrfs_end_transaction(trans);
3333 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3335 struct btrfs_root *chunk_root = fs_info->chunk_root;
3336 struct btrfs_path *path;
3337 struct extent_buffer *leaf;
3338 struct btrfs_chunk *chunk;
3339 struct btrfs_key key;
3340 struct btrfs_key found_key;
3342 bool retried = false;
3346 path = btrfs_alloc_path();
3351 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3352 key.offset = (u64)-1;
3353 key.type = BTRFS_CHUNK_ITEM_KEY;
3356 mutex_lock(&fs_info->reclaim_bgs_lock);
3357 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3359 mutex_unlock(&fs_info->reclaim_bgs_lock);
3362 BUG_ON(ret == 0); /* Corruption */
3364 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3367 mutex_unlock(&fs_info->reclaim_bgs_lock);
3373 leaf = path->nodes[0];
3374 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3376 chunk = btrfs_item_ptr(leaf, path->slots[0],
3377 struct btrfs_chunk);
3378 chunk_type = btrfs_chunk_type(leaf, chunk);
3379 btrfs_release_path(path);
3381 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3382 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3388 mutex_unlock(&fs_info->reclaim_bgs_lock);
3390 if (found_key.offset == 0)
3392 key.offset = found_key.offset - 1;
3395 if (failed && !retried) {
3399 } else if (WARN_ON(failed && retried)) {
3403 btrfs_free_path(path);
3408 * return 1 : allocate a data chunk successfully,
3409 * return <0: errors during allocating a data chunk,
3410 * return 0 : no need to allocate a data chunk.
3412 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3415 struct btrfs_block_group *cache;
3419 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3421 chunk_type = cache->flags;
3422 btrfs_put_block_group(cache);
3424 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3427 spin_lock(&fs_info->data_sinfo->lock);
3428 bytes_used = fs_info->data_sinfo->bytes_used;
3429 spin_unlock(&fs_info->data_sinfo->lock);
3432 struct btrfs_trans_handle *trans;
3435 trans = btrfs_join_transaction(fs_info->tree_root);
3437 return PTR_ERR(trans);
3439 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3440 btrfs_end_transaction(trans);
3449 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3450 struct btrfs_balance_control *bctl)
3452 struct btrfs_root *root = fs_info->tree_root;
3453 struct btrfs_trans_handle *trans;
3454 struct btrfs_balance_item *item;
3455 struct btrfs_disk_balance_args disk_bargs;
3456 struct btrfs_path *path;
3457 struct extent_buffer *leaf;
3458 struct btrfs_key key;
3461 path = btrfs_alloc_path();
3465 trans = btrfs_start_transaction(root, 0);
3466 if (IS_ERR(trans)) {
3467 btrfs_free_path(path);
3468 return PTR_ERR(trans);
3471 key.objectid = BTRFS_BALANCE_OBJECTID;
3472 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3475 ret = btrfs_insert_empty_item(trans, root, path, &key,
3480 leaf = path->nodes[0];
3481 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3483 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3485 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3486 btrfs_set_balance_data(leaf, item, &disk_bargs);
3487 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3488 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3489 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3490 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3492 btrfs_set_balance_flags(leaf, item, bctl->flags);
3494 btrfs_mark_buffer_dirty(leaf);
3496 btrfs_free_path(path);
3497 err = btrfs_commit_transaction(trans);
3503 static int del_balance_item(struct btrfs_fs_info *fs_info)
3505 struct btrfs_root *root = fs_info->tree_root;
3506 struct btrfs_trans_handle *trans;
3507 struct btrfs_path *path;
3508 struct btrfs_key key;
3511 path = btrfs_alloc_path();
3515 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3516 if (IS_ERR(trans)) {
3517 btrfs_free_path(path);
3518 return PTR_ERR(trans);
3521 key.objectid = BTRFS_BALANCE_OBJECTID;
3522 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3525 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3533 ret = btrfs_del_item(trans, root, path);
3535 btrfs_free_path(path);
3536 err = btrfs_commit_transaction(trans);
3543 * This is a heuristic used to reduce the number of chunks balanced on
3544 * resume after balance was interrupted.
3546 static void update_balance_args(struct btrfs_balance_control *bctl)
3549 * Turn on soft mode for chunk types that were being converted.
3551 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3552 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3553 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3554 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3555 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3556 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3559 * Turn on usage filter if is not already used. The idea is
3560 * that chunks that we have already balanced should be
3561 * reasonably full. Don't do it for chunks that are being
3562 * converted - that will keep us from relocating unconverted
3563 * (albeit full) chunks.
3565 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3566 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3567 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3568 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3569 bctl->data.usage = 90;
3571 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3572 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3573 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3574 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3575 bctl->sys.usage = 90;
3577 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3578 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3579 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3580 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3581 bctl->meta.usage = 90;
3586 * Clear the balance status in fs_info and delete the balance item from disk.
3588 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3590 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3593 BUG_ON(!fs_info->balance_ctl);
3595 spin_lock(&fs_info->balance_lock);
3596 fs_info->balance_ctl = NULL;
3597 spin_unlock(&fs_info->balance_lock);
3600 ret = del_balance_item(fs_info);
3602 btrfs_handle_fs_error(fs_info, ret, NULL);
3606 * Balance filters. Return 1 if chunk should be filtered out
3607 * (should not be balanced).
3609 static int chunk_profiles_filter(u64 chunk_type,
3610 struct btrfs_balance_args *bargs)
3612 chunk_type = chunk_to_extended(chunk_type) &
3613 BTRFS_EXTENDED_PROFILE_MASK;
3615 if (bargs->profiles & chunk_type)
3621 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3622 struct btrfs_balance_args *bargs)
3624 struct btrfs_block_group *cache;
3626 u64 user_thresh_min;
3627 u64 user_thresh_max;
3630 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3631 chunk_used = cache->used;
3633 if (bargs->usage_min == 0)
3634 user_thresh_min = 0;
3636 user_thresh_min = mult_perc(cache->length, bargs->usage_min);
3638 if (bargs->usage_max == 0)
3639 user_thresh_max = 1;
3640 else if (bargs->usage_max > 100)
3641 user_thresh_max = cache->length;
3643 user_thresh_max = mult_perc(cache->length, bargs->usage_max);
3645 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3648 btrfs_put_block_group(cache);
3652 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3653 u64 chunk_offset, struct btrfs_balance_args *bargs)
3655 struct btrfs_block_group *cache;
3656 u64 chunk_used, user_thresh;
3659 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3660 chunk_used = cache->used;
3662 if (bargs->usage_min == 0)
3664 else if (bargs->usage > 100)
3665 user_thresh = cache->length;
3667 user_thresh = mult_perc(cache->length, bargs->usage);
3669 if (chunk_used < user_thresh)
3672 btrfs_put_block_group(cache);
3676 static int chunk_devid_filter(struct extent_buffer *leaf,
3677 struct btrfs_chunk *chunk,
3678 struct btrfs_balance_args *bargs)
3680 struct btrfs_stripe *stripe;
3681 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3684 for (i = 0; i < num_stripes; i++) {
3685 stripe = btrfs_stripe_nr(chunk, i);
3686 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3693 static u64 calc_data_stripes(u64 type, int num_stripes)
3695 const int index = btrfs_bg_flags_to_raid_index(type);
3696 const int ncopies = btrfs_raid_array[index].ncopies;
3697 const int nparity = btrfs_raid_array[index].nparity;
3699 return (num_stripes - nparity) / ncopies;
3702 /* [pstart, pend) */
3703 static int chunk_drange_filter(struct extent_buffer *leaf,
3704 struct btrfs_chunk *chunk,
3705 struct btrfs_balance_args *bargs)
3707 struct btrfs_stripe *stripe;
3708 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3715 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3718 type = btrfs_chunk_type(leaf, chunk);
3719 factor = calc_data_stripes(type, num_stripes);
3721 for (i = 0; i < num_stripes; i++) {
3722 stripe = btrfs_stripe_nr(chunk, i);
3723 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3726 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3727 stripe_length = btrfs_chunk_length(leaf, chunk);
3728 stripe_length = div_u64(stripe_length, factor);
3730 if (stripe_offset < bargs->pend &&
3731 stripe_offset + stripe_length > bargs->pstart)
3738 /* [vstart, vend) */
3739 static int chunk_vrange_filter(struct extent_buffer *leaf,
3740 struct btrfs_chunk *chunk,
3742 struct btrfs_balance_args *bargs)
3744 if (chunk_offset < bargs->vend &&
3745 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3746 /* at least part of the chunk is inside this vrange */
3752 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3753 struct btrfs_chunk *chunk,
3754 struct btrfs_balance_args *bargs)
3756 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3758 if (bargs->stripes_min <= num_stripes
3759 && num_stripes <= bargs->stripes_max)
3765 static int chunk_soft_convert_filter(u64 chunk_type,
3766 struct btrfs_balance_args *bargs)
3768 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3771 chunk_type = chunk_to_extended(chunk_type) &
3772 BTRFS_EXTENDED_PROFILE_MASK;
3774 if (bargs->target == chunk_type)
3780 static int should_balance_chunk(struct extent_buffer *leaf,
3781 struct btrfs_chunk *chunk, u64 chunk_offset)
3783 struct btrfs_fs_info *fs_info = leaf->fs_info;
3784 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3785 struct btrfs_balance_args *bargs = NULL;
3786 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3789 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3790 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3794 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3795 bargs = &bctl->data;
3796 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3798 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3799 bargs = &bctl->meta;
3801 /* profiles filter */
3802 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3803 chunk_profiles_filter(chunk_type, bargs)) {
3808 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3809 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3811 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3812 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3817 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3818 chunk_devid_filter(leaf, chunk, bargs)) {
3822 /* drange filter, makes sense only with devid filter */
3823 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3824 chunk_drange_filter(leaf, chunk, bargs)) {
3829 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3830 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3834 /* stripes filter */
3835 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3836 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3840 /* soft profile changing mode */
3841 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3842 chunk_soft_convert_filter(chunk_type, bargs)) {
3847 * limited by count, must be the last filter
3849 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3850 if (bargs->limit == 0)
3854 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3856 * Same logic as the 'limit' filter; the minimum cannot be
3857 * determined here because we do not have the global information
3858 * about the count of all chunks that satisfy the filters.
3860 if (bargs->limit_max == 0)
3869 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3871 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3872 struct btrfs_root *chunk_root = fs_info->chunk_root;
3874 struct btrfs_chunk *chunk;
3875 struct btrfs_path *path = NULL;
3876 struct btrfs_key key;
3877 struct btrfs_key found_key;
3878 struct extent_buffer *leaf;
3881 int enospc_errors = 0;
3882 bool counting = true;
3883 /* The single value limit and min/max limits use the same bytes in the */
3884 u64 limit_data = bctl->data.limit;
3885 u64 limit_meta = bctl->meta.limit;
3886 u64 limit_sys = bctl->sys.limit;
3890 int chunk_reserved = 0;
3892 path = btrfs_alloc_path();
3898 /* zero out stat counters */
3899 spin_lock(&fs_info->balance_lock);
3900 memset(&bctl->stat, 0, sizeof(bctl->stat));
3901 spin_unlock(&fs_info->balance_lock);
3905 * The single value limit and min/max limits use the same bytes
3908 bctl->data.limit = limit_data;
3909 bctl->meta.limit = limit_meta;
3910 bctl->sys.limit = limit_sys;
3912 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3913 key.offset = (u64)-1;
3914 key.type = BTRFS_CHUNK_ITEM_KEY;
3917 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3918 atomic_read(&fs_info->balance_cancel_req)) {
3923 mutex_lock(&fs_info->reclaim_bgs_lock);
3924 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3926 mutex_unlock(&fs_info->reclaim_bgs_lock);
3931 * this shouldn't happen, it means the last relocate
3935 BUG(); /* FIXME break ? */
3937 ret = btrfs_previous_item(chunk_root, path, 0,
3938 BTRFS_CHUNK_ITEM_KEY);
3940 mutex_unlock(&fs_info->reclaim_bgs_lock);
3945 leaf = path->nodes[0];
3946 slot = path->slots[0];
3947 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3949 if (found_key.objectid != key.objectid) {
3950 mutex_unlock(&fs_info->reclaim_bgs_lock);
3954 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3955 chunk_type = btrfs_chunk_type(leaf, chunk);
3958 spin_lock(&fs_info->balance_lock);
3959 bctl->stat.considered++;
3960 spin_unlock(&fs_info->balance_lock);
3963 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3965 btrfs_release_path(path);
3967 mutex_unlock(&fs_info->reclaim_bgs_lock);
3972 mutex_unlock(&fs_info->reclaim_bgs_lock);
3973 spin_lock(&fs_info->balance_lock);
3974 bctl->stat.expected++;
3975 spin_unlock(&fs_info->balance_lock);
3977 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3979 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3981 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3988 * Apply limit_min filter, no need to check if the LIMITS
3989 * filter is used, limit_min is 0 by default
3991 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3992 count_data < bctl->data.limit_min)
3993 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3994 count_meta < bctl->meta.limit_min)
3995 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3996 count_sys < bctl->sys.limit_min)) {
3997 mutex_unlock(&fs_info->reclaim_bgs_lock);
4001 if (!chunk_reserved) {
4003 * We may be relocating the only data chunk we have,
4004 * which could potentially end up with losing data's
4005 * raid profile, so lets allocate an empty one in
4008 ret = btrfs_may_alloc_data_chunk(fs_info,
4011 mutex_unlock(&fs_info->reclaim_bgs_lock);
4013 } else if (ret == 1) {
4018 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
4019 mutex_unlock(&fs_info->reclaim_bgs_lock);
4020 if (ret == -ENOSPC) {
4022 } else if (ret == -ETXTBSY) {
4024 "skipping relocation of block group %llu due to active swapfile",
4030 spin_lock(&fs_info->balance_lock);
4031 bctl->stat.completed++;
4032 spin_unlock(&fs_info->balance_lock);
4035 if (found_key.offset == 0)
4037 key.offset = found_key.offset - 1;
4041 btrfs_release_path(path);
4046 btrfs_free_path(path);
4047 if (enospc_errors) {
4048 btrfs_info(fs_info, "%d enospc errors during balance",
4058 * See if a given profile is valid and reduced.
4060 * @flags: profile to validate
4061 * @extended: if true @flags is treated as an extended profile
4063 static int alloc_profile_is_valid(u64 flags, int extended)
4065 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
4066 BTRFS_BLOCK_GROUP_PROFILE_MASK);
4068 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
4070 /* 1) check that all other bits are zeroed */
4074 /* 2) see if profile is reduced */
4076 return !extended; /* "0" is valid for usual profiles */
4078 return has_single_bit_set(flags);
4081 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
4083 /* cancel requested || normal exit path */
4084 return atomic_read(&fs_info->balance_cancel_req) ||
4085 (atomic_read(&fs_info->balance_pause_req) == 0 &&
4086 atomic_read(&fs_info->balance_cancel_req) == 0);
4090 * Validate target profile against allowed profiles and return true if it's OK.
4091 * Otherwise print the error message and return false.
4093 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
4094 const struct btrfs_balance_args *bargs,
4095 u64 allowed, const char *type)
4097 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
4100 /* Profile is valid and does not have bits outside of the allowed set */
4101 if (alloc_profile_is_valid(bargs->target, 1) &&
4102 (bargs->target & ~allowed) == 0)
4105 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
4106 type, btrfs_bg_type_to_raid_name(bargs->target));
4111 * Fill @buf with textual description of balance filter flags @bargs, up to
4112 * @size_buf including the terminating null. The output may be trimmed if it
4113 * does not fit into the provided buffer.
4115 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
4119 u32 size_bp = size_buf;
4121 u64 flags = bargs->flags;
4122 char tmp_buf[128] = {'\0'};
4127 #define CHECK_APPEND_NOARG(a) \
4129 ret = snprintf(bp, size_bp, (a)); \
4130 if (ret < 0 || ret >= size_bp) \
4131 goto out_overflow; \
4136 #define CHECK_APPEND_1ARG(a, v1) \
4138 ret = snprintf(bp, size_bp, (a), (v1)); \
4139 if (ret < 0 || ret >= size_bp) \
4140 goto out_overflow; \
4145 #define CHECK_APPEND_2ARG(a, v1, v2) \
4147 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4148 if (ret < 0 || ret >= size_bp) \
4149 goto out_overflow; \
4154 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
4155 CHECK_APPEND_1ARG("convert=%s,",
4156 btrfs_bg_type_to_raid_name(bargs->target));
4158 if (flags & BTRFS_BALANCE_ARGS_SOFT)
4159 CHECK_APPEND_NOARG("soft,");
4161 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4162 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4164 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4167 if (flags & BTRFS_BALANCE_ARGS_USAGE)
4168 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4170 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4171 CHECK_APPEND_2ARG("usage=%u..%u,",
4172 bargs->usage_min, bargs->usage_max);
4174 if (flags & BTRFS_BALANCE_ARGS_DEVID)
4175 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4177 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4178 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4179 bargs->pstart, bargs->pend);
4181 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4182 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4183 bargs->vstart, bargs->vend);
4185 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4186 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4188 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4189 CHECK_APPEND_2ARG("limit=%u..%u,",
4190 bargs->limit_min, bargs->limit_max);
4192 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4193 CHECK_APPEND_2ARG("stripes=%u..%u,",
4194 bargs->stripes_min, bargs->stripes_max);
4196 #undef CHECK_APPEND_2ARG
4197 #undef CHECK_APPEND_1ARG
4198 #undef CHECK_APPEND_NOARG
4202 if (size_bp < size_buf)
4203 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4208 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4210 u32 size_buf = 1024;
4211 char tmp_buf[192] = {'\0'};
4214 u32 size_bp = size_buf;
4216 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4218 buf = kzalloc(size_buf, GFP_KERNEL);
4224 #define CHECK_APPEND_1ARG(a, v1) \
4226 ret = snprintf(bp, size_bp, (a), (v1)); \
4227 if (ret < 0 || ret >= size_bp) \
4228 goto out_overflow; \
4233 if (bctl->flags & BTRFS_BALANCE_FORCE)
4234 CHECK_APPEND_1ARG("%s", "-f ");
4236 if (bctl->flags & BTRFS_BALANCE_DATA) {
4237 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4238 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4241 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4242 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4243 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4246 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4247 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4248 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4251 #undef CHECK_APPEND_1ARG
4255 if (size_bp < size_buf)
4256 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4257 btrfs_info(fs_info, "balance: %s %s",
4258 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4259 "resume" : "start", buf);
4265 * Should be called with balance mutexe held
4267 int btrfs_balance(struct btrfs_fs_info *fs_info,
4268 struct btrfs_balance_control *bctl,
4269 struct btrfs_ioctl_balance_args *bargs)
4271 u64 meta_target, data_target;
4277 bool reducing_redundancy;
4280 if (btrfs_fs_closing(fs_info) ||
4281 atomic_read(&fs_info->balance_pause_req) ||
4282 btrfs_should_cancel_balance(fs_info)) {
4287 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4288 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4292 * In case of mixed groups both data and meta should be picked,
4293 * and identical options should be given for both of them.
4295 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4296 if (mixed && (bctl->flags & allowed)) {
4297 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4298 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4299 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4301 "balance: mixed groups data and metadata options must be the same");
4308 * rw_devices will not change at the moment, device add/delete/replace
4311 num_devices = fs_info->fs_devices->rw_devices;
4314 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4315 * special bit for it, to make it easier to distinguish. Thus we need
4316 * to set it manually, or balance would refuse the profile.
4318 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4319 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4320 if (num_devices >= btrfs_raid_array[i].devs_min)
4321 allowed |= btrfs_raid_array[i].bg_flag;
4323 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4324 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4325 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4331 * Allow to reduce metadata or system integrity only if force set for
4332 * profiles with redundancy (copies, parity)
4335 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4336 if (btrfs_raid_array[i].ncopies >= 2 ||
4337 btrfs_raid_array[i].tolerated_failures >= 1)
4338 allowed |= btrfs_raid_array[i].bg_flag;
4341 seq = read_seqbegin(&fs_info->profiles_lock);
4343 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4344 (fs_info->avail_system_alloc_bits & allowed) &&
4345 !(bctl->sys.target & allowed)) ||
4346 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4347 (fs_info->avail_metadata_alloc_bits & allowed) &&
4348 !(bctl->meta.target & allowed)))
4349 reducing_redundancy = true;
4351 reducing_redundancy = false;
4353 /* if we're not converting, the target field is uninitialized */
4354 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4355 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4356 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4357 bctl->data.target : fs_info->avail_data_alloc_bits;
4358 } while (read_seqretry(&fs_info->profiles_lock, seq));
4360 if (reducing_redundancy) {
4361 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4363 "balance: force reducing metadata redundancy");
4366 "balance: reduces metadata redundancy, use --force if you want this");
4372 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4373 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4375 "balance: metadata profile %s has lower redundancy than data profile %s",
4376 btrfs_bg_type_to_raid_name(meta_target),
4377 btrfs_bg_type_to_raid_name(data_target));
4380 ret = insert_balance_item(fs_info, bctl);
4381 if (ret && ret != -EEXIST)
4384 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4385 BUG_ON(ret == -EEXIST);
4386 BUG_ON(fs_info->balance_ctl);
4387 spin_lock(&fs_info->balance_lock);
4388 fs_info->balance_ctl = bctl;
4389 spin_unlock(&fs_info->balance_lock);
4391 BUG_ON(ret != -EEXIST);
4392 spin_lock(&fs_info->balance_lock);
4393 update_balance_args(bctl);
4394 spin_unlock(&fs_info->balance_lock);
4397 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4398 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4399 describe_balance_start_or_resume(fs_info);
4400 mutex_unlock(&fs_info->balance_mutex);
4402 ret = __btrfs_balance(fs_info);
4404 mutex_lock(&fs_info->balance_mutex);
4405 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) {
4406 btrfs_info(fs_info, "balance: paused");
4407 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
4410 * Balance can be canceled by:
4412 * - Regular cancel request
4413 * Then ret == -ECANCELED and balance_cancel_req > 0
4415 * - Fatal signal to "btrfs" process
4416 * Either the signal caught by wait_reserve_ticket() and callers
4417 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4419 * Either way, in this case balance_cancel_req = 0, and
4420 * ret == -EINTR or ret == -ECANCELED.
4422 * So here we only check the return value to catch canceled balance.
4424 else if (ret == -ECANCELED || ret == -EINTR)
4425 btrfs_info(fs_info, "balance: canceled");
4427 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4429 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4432 memset(bargs, 0, sizeof(*bargs));
4433 btrfs_update_ioctl_balance_args(fs_info, bargs);
4436 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4437 balance_need_close(fs_info)) {
4438 reset_balance_state(fs_info);
4439 btrfs_exclop_finish(fs_info);
4442 wake_up(&fs_info->balance_wait_q);
4446 if (bctl->flags & BTRFS_BALANCE_RESUME)
4447 reset_balance_state(fs_info);
4450 btrfs_exclop_finish(fs_info);
4455 static int balance_kthread(void *data)
4457 struct btrfs_fs_info *fs_info = data;
4460 sb_start_write(fs_info->sb);
4461 mutex_lock(&fs_info->balance_mutex);
4462 if (fs_info->balance_ctl)
4463 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4464 mutex_unlock(&fs_info->balance_mutex);
4465 sb_end_write(fs_info->sb);
4470 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4472 struct task_struct *tsk;
4474 mutex_lock(&fs_info->balance_mutex);
4475 if (!fs_info->balance_ctl) {
4476 mutex_unlock(&fs_info->balance_mutex);
4479 mutex_unlock(&fs_info->balance_mutex);
4481 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4482 btrfs_info(fs_info, "balance: resume skipped");
4486 spin_lock(&fs_info->super_lock);
4487 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
4488 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;
4489 spin_unlock(&fs_info->super_lock);
4491 * A ro->rw remount sequence should continue with the paused balance
4492 * regardless of who pauses it, system or the user as of now, so set
4495 spin_lock(&fs_info->balance_lock);
4496 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4497 spin_unlock(&fs_info->balance_lock);
4499 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4500 return PTR_ERR_OR_ZERO(tsk);
4503 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4505 struct btrfs_balance_control *bctl;
4506 struct btrfs_balance_item *item;
4507 struct btrfs_disk_balance_args disk_bargs;
4508 struct btrfs_path *path;
4509 struct extent_buffer *leaf;
4510 struct btrfs_key key;
4513 path = btrfs_alloc_path();
4517 key.objectid = BTRFS_BALANCE_OBJECTID;
4518 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4521 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4524 if (ret > 0) { /* ret = -ENOENT; */
4529 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4535 leaf = path->nodes[0];
4536 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4538 bctl->flags = btrfs_balance_flags(leaf, item);
4539 bctl->flags |= BTRFS_BALANCE_RESUME;
4541 btrfs_balance_data(leaf, item, &disk_bargs);
4542 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4543 btrfs_balance_meta(leaf, item, &disk_bargs);
4544 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4545 btrfs_balance_sys(leaf, item, &disk_bargs);
4546 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4549 * This should never happen, as the paused balance state is recovered
4550 * during mount without any chance of other exclusive ops to collide.
4552 * This gives the exclusive op status to balance and keeps in paused
4553 * state until user intervention (cancel or umount). If the ownership
4554 * cannot be assigned, show a message but do not fail. The balance
4555 * is in a paused state and must have fs_info::balance_ctl properly
4558 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED))
4560 "balance: cannot set exclusive op status, resume manually");
4562 btrfs_release_path(path);
4564 mutex_lock(&fs_info->balance_mutex);
4565 BUG_ON(fs_info->balance_ctl);
4566 spin_lock(&fs_info->balance_lock);
4567 fs_info->balance_ctl = bctl;
4568 spin_unlock(&fs_info->balance_lock);
4569 mutex_unlock(&fs_info->balance_mutex);
4571 btrfs_free_path(path);
4575 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4579 mutex_lock(&fs_info->balance_mutex);
4580 if (!fs_info->balance_ctl) {
4581 mutex_unlock(&fs_info->balance_mutex);
4585 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4586 atomic_inc(&fs_info->balance_pause_req);
4587 mutex_unlock(&fs_info->balance_mutex);
4589 wait_event(fs_info->balance_wait_q,
4590 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4592 mutex_lock(&fs_info->balance_mutex);
4593 /* we are good with balance_ctl ripped off from under us */
4594 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4595 atomic_dec(&fs_info->balance_pause_req);
4600 mutex_unlock(&fs_info->balance_mutex);
4604 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4606 mutex_lock(&fs_info->balance_mutex);
4607 if (!fs_info->balance_ctl) {
4608 mutex_unlock(&fs_info->balance_mutex);
4613 * A paused balance with the item stored on disk can be resumed at
4614 * mount time if the mount is read-write. Otherwise it's still paused
4615 * and we must not allow cancelling as it deletes the item.
4617 if (sb_rdonly(fs_info->sb)) {
4618 mutex_unlock(&fs_info->balance_mutex);
4622 atomic_inc(&fs_info->balance_cancel_req);
4624 * if we are running just wait and return, balance item is
4625 * deleted in btrfs_balance in this case
4627 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4628 mutex_unlock(&fs_info->balance_mutex);
4629 wait_event(fs_info->balance_wait_q,
4630 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4631 mutex_lock(&fs_info->balance_mutex);
4633 mutex_unlock(&fs_info->balance_mutex);
4635 * Lock released to allow other waiters to continue, we'll
4636 * reexamine the status again.
4638 mutex_lock(&fs_info->balance_mutex);
4640 if (fs_info->balance_ctl) {
4641 reset_balance_state(fs_info);
4642 btrfs_exclop_finish(fs_info);
4643 btrfs_info(fs_info, "balance: canceled");
4647 BUG_ON(fs_info->balance_ctl ||
4648 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4649 atomic_dec(&fs_info->balance_cancel_req);
4650 mutex_unlock(&fs_info->balance_mutex);
4654 int btrfs_uuid_scan_kthread(void *data)
4656 struct btrfs_fs_info *fs_info = data;
4657 struct btrfs_root *root = fs_info->tree_root;
4658 struct btrfs_key key;
4659 struct btrfs_path *path = NULL;
4661 struct extent_buffer *eb;
4663 struct btrfs_root_item root_item;
4665 struct btrfs_trans_handle *trans = NULL;
4666 bool closing = false;
4668 path = btrfs_alloc_path();
4675 key.type = BTRFS_ROOT_ITEM_KEY;
4679 if (btrfs_fs_closing(fs_info)) {
4683 ret = btrfs_search_forward(root, &key, path,
4684 BTRFS_OLDEST_GENERATION);
4691 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4692 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4693 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4694 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4697 eb = path->nodes[0];
4698 slot = path->slots[0];
4699 item_size = btrfs_item_size(eb, slot);
4700 if (item_size < sizeof(root_item))
4703 read_extent_buffer(eb, &root_item,
4704 btrfs_item_ptr_offset(eb, slot),
4705 (int)sizeof(root_item));
4706 if (btrfs_root_refs(&root_item) == 0)
4709 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4710 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4714 btrfs_release_path(path);
4716 * 1 - subvol uuid item
4717 * 1 - received_subvol uuid item
4719 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4720 if (IS_ERR(trans)) {
4721 ret = PTR_ERR(trans);
4729 btrfs_release_path(path);
4730 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4731 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4732 BTRFS_UUID_KEY_SUBVOL,
4735 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4741 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4742 ret = btrfs_uuid_tree_add(trans,
4743 root_item.received_uuid,
4744 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4747 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4754 btrfs_release_path(path);
4756 ret = btrfs_end_transaction(trans);
4762 if (key.offset < (u64)-1) {
4764 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4766 key.type = BTRFS_ROOT_ITEM_KEY;
4767 } else if (key.objectid < (u64)-1) {
4769 key.type = BTRFS_ROOT_ITEM_KEY;
4778 btrfs_free_path(path);
4779 if (trans && !IS_ERR(trans))
4780 btrfs_end_transaction(trans);
4782 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4784 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4785 up(&fs_info->uuid_tree_rescan_sem);
4789 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4791 struct btrfs_trans_handle *trans;
4792 struct btrfs_root *tree_root = fs_info->tree_root;
4793 struct btrfs_root *uuid_root;
4794 struct task_struct *task;
4801 trans = btrfs_start_transaction(tree_root, 2);
4803 return PTR_ERR(trans);
4805 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4806 if (IS_ERR(uuid_root)) {
4807 ret = PTR_ERR(uuid_root);
4808 btrfs_abort_transaction(trans, ret);
4809 btrfs_end_transaction(trans);
4813 fs_info->uuid_root = uuid_root;
4815 ret = btrfs_commit_transaction(trans);
4819 down(&fs_info->uuid_tree_rescan_sem);
4820 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4822 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4823 btrfs_warn(fs_info, "failed to start uuid_scan task");
4824 up(&fs_info->uuid_tree_rescan_sem);
4825 return PTR_ERR(task);
4832 * shrinking a device means finding all of the device extents past
4833 * the new size, and then following the back refs to the chunks.
4834 * The chunk relocation code actually frees the device extent
4836 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4838 struct btrfs_fs_info *fs_info = device->fs_info;
4839 struct btrfs_root *root = fs_info->dev_root;
4840 struct btrfs_trans_handle *trans;
4841 struct btrfs_dev_extent *dev_extent = NULL;
4842 struct btrfs_path *path;
4848 bool retried = false;
4849 struct extent_buffer *l;
4850 struct btrfs_key key;
4851 struct btrfs_super_block *super_copy = fs_info->super_copy;
4852 u64 old_total = btrfs_super_total_bytes(super_copy);
4853 u64 old_size = btrfs_device_get_total_bytes(device);
4857 new_size = round_down(new_size, fs_info->sectorsize);
4859 diff = round_down(old_size - new_size, fs_info->sectorsize);
4861 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4864 path = btrfs_alloc_path();
4868 path->reada = READA_BACK;
4870 trans = btrfs_start_transaction(root, 0);
4871 if (IS_ERR(trans)) {
4872 btrfs_free_path(path);
4873 return PTR_ERR(trans);
4876 mutex_lock(&fs_info->chunk_mutex);
4878 btrfs_device_set_total_bytes(device, new_size);
4879 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4880 device->fs_devices->total_rw_bytes -= diff;
4881 atomic64_sub(diff, &fs_info->free_chunk_space);
4885 * Once the device's size has been set to the new size, ensure all
4886 * in-memory chunks are synced to disk so that the loop below sees them
4887 * and relocates them accordingly.
4889 if (contains_pending_extent(device, &start, diff)) {
4890 mutex_unlock(&fs_info->chunk_mutex);
4891 ret = btrfs_commit_transaction(trans);
4895 mutex_unlock(&fs_info->chunk_mutex);
4896 btrfs_end_transaction(trans);
4900 key.objectid = device->devid;
4901 key.offset = (u64)-1;
4902 key.type = BTRFS_DEV_EXTENT_KEY;
4905 mutex_lock(&fs_info->reclaim_bgs_lock);
4906 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4908 mutex_unlock(&fs_info->reclaim_bgs_lock);
4912 ret = btrfs_previous_item(root, path, 0, key.type);
4914 mutex_unlock(&fs_info->reclaim_bgs_lock);
4918 btrfs_release_path(path);
4923 slot = path->slots[0];
4924 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4926 if (key.objectid != device->devid) {
4927 mutex_unlock(&fs_info->reclaim_bgs_lock);
4928 btrfs_release_path(path);
4932 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4933 length = btrfs_dev_extent_length(l, dev_extent);
4935 if (key.offset + length <= new_size) {
4936 mutex_unlock(&fs_info->reclaim_bgs_lock);
4937 btrfs_release_path(path);
4941 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4942 btrfs_release_path(path);
4945 * We may be relocating the only data chunk we have,
4946 * which could potentially end up with losing data's
4947 * raid profile, so lets allocate an empty one in
4950 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4952 mutex_unlock(&fs_info->reclaim_bgs_lock);
4956 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4957 mutex_unlock(&fs_info->reclaim_bgs_lock);
4958 if (ret == -ENOSPC) {
4961 if (ret == -ETXTBSY) {
4963 "could not shrink block group %llu due to active swapfile",
4968 } while (key.offset-- > 0);
4970 if (failed && !retried) {
4974 } else if (failed && retried) {
4979 /* Shrinking succeeded, else we would be at "done". */
4980 trans = btrfs_start_transaction(root, 0);
4981 if (IS_ERR(trans)) {
4982 ret = PTR_ERR(trans);
4986 mutex_lock(&fs_info->chunk_mutex);
4987 /* Clear all state bits beyond the shrunk device size */
4988 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
4991 btrfs_device_set_disk_total_bytes(device, new_size);
4992 if (list_empty(&device->post_commit_list))
4993 list_add_tail(&device->post_commit_list,
4994 &trans->transaction->dev_update_list);
4996 WARN_ON(diff > old_total);
4997 btrfs_set_super_total_bytes(super_copy,
4998 round_down(old_total - diff, fs_info->sectorsize));
4999 mutex_unlock(&fs_info->chunk_mutex);
5001 btrfs_reserve_chunk_metadata(trans, false);
5002 /* Now btrfs_update_device() will change the on-disk size. */
5003 ret = btrfs_update_device(trans, device);
5004 btrfs_trans_release_chunk_metadata(trans);
5006 btrfs_abort_transaction(trans, ret);
5007 btrfs_end_transaction(trans);
5009 ret = btrfs_commit_transaction(trans);
5012 btrfs_free_path(path);
5014 mutex_lock(&fs_info->chunk_mutex);
5015 btrfs_device_set_total_bytes(device, old_size);
5016 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
5017 device->fs_devices->total_rw_bytes += diff;
5018 atomic64_add(diff, &fs_info->free_chunk_space);
5019 mutex_unlock(&fs_info->chunk_mutex);
5024 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
5025 struct btrfs_key *key,
5026 struct btrfs_chunk *chunk, int item_size)
5028 struct btrfs_super_block *super_copy = fs_info->super_copy;
5029 struct btrfs_disk_key disk_key;
5033 lockdep_assert_held(&fs_info->chunk_mutex);
5035 array_size = btrfs_super_sys_array_size(super_copy);
5036 if (array_size + item_size + sizeof(disk_key)
5037 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
5040 ptr = super_copy->sys_chunk_array + array_size;
5041 btrfs_cpu_key_to_disk(&disk_key, key);
5042 memcpy(ptr, &disk_key, sizeof(disk_key));
5043 ptr += sizeof(disk_key);
5044 memcpy(ptr, chunk, item_size);
5045 item_size += sizeof(disk_key);
5046 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
5052 * sort the devices in descending order by max_avail, total_avail
5054 static int btrfs_cmp_device_info(const void *a, const void *b)
5056 const struct btrfs_device_info *di_a = a;
5057 const struct btrfs_device_info *di_b = b;
5059 if (di_a->max_avail > di_b->max_avail)
5061 if (di_a->max_avail < di_b->max_avail)
5063 if (di_a->total_avail > di_b->total_avail)
5065 if (di_a->total_avail < di_b->total_avail)
5070 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
5072 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5075 btrfs_set_fs_incompat(info, RAID56);
5078 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
5080 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
5083 btrfs_set_fs_incompat(info, RAID1C34);
5087 * Structure used internally for btrfs_create_chunk() function.
5088 * Wraps needed parameters.
5090 struct alloc_chunk_ctl {
5093 /* Total number of stripes to allocate */
5095 /* sub_stripes info for map */
5097 /* Stripes per device */
5099 /* Maximum number of devices to use */
5101 /* Minimum number of devices to use */
5103 /* ndevs has to be a multiple of this */
5105 /* Number of copies */
5107 /* Number of stripes worth of bytes to store parity information */
5109 u64 max_stripe_size;
5117 static void init_alloc_chunk_ctl_policy_regular(
5118 struct btrfs_fs_devices *fs_devices,
5119 struct alloc_chunk_ctl *ctl)
5121 struct btrfs_space_info *space_info;
5123 space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type);
5126 ctl->max_chunk_size = READ_ONCE(space_info->chunk_size);
5127 ctl->max_stripe_size = ctl->max_chunk_size;
5129 if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM)
5130 ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK);
5132 /* We don't want a chunk larger than 10% of writable space */
5133 ctl->max_chunk_size = min(mult_perc(fs_devices->total_rw_bytes, 10),
5134 ctl->max_chunk_size);
5135 ctl->dev_extent_min = btrfs_stripe_nr_to_offset(ctl->dev_stripes);
5138 static void init_alloc_chunk_ctl_policy_zoned(
5139 struct btrfs_fs_devices *fs_devices,
5140 struct alloc_chunk_ctl *ctl)
5142 u64 zone_size = fs_devices->fs_info->zone_size;
5144 int min_num_stripes = ctl->devs_min * ctl->dev_stripes;
5145 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;
5146 u64 min_chunk_size = min_data_stripes * zone_size;
5147 u64 type = ctl->type;
5149 ctl->max_stripe_size = zone_size;
5150 if (type & BTRFS_BLOCK_GROUP_DATA) {
5151 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,
5153 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5154 ctl->max_chunk_size = ctl->max_stripe_size;
5155 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5156 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5157 ctl->devs_max = min_t(int, ctl->devs_max,
5158 BTRFS_MAX_DEVS_SYS_CHUNK);
5163 /* We don't want a chunk larger than 10% of writable space */
5164 limit = max(round_down(mult_perc(fs_devices->total_rw_bytes, 10),
5167 ctl->max_chunk_size = min(limit, ctl->max_chunk_size);
5168 ctl->dev_extent_min = zone_size * ctl->dev_stripes;
5171 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
5172 struct alloc_chunk_ctl *ctl)
5174 int index = btrfs_bg_flags_to_raid_index(ctl->type);
5176 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
5177 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
5178 ctl->devs_max = btrfs_raid_array[index].devs_max;
5180 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
5181 ctl->devs_min = btrfs_raid_array[index].devs_min;
5182 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
5183 ctl->ncopies = btrfs_raid_array[index].ncopies;
5184 ctl->nparity = btrfs_raid_array[index].nparity;
5187 switch (fs_devices->chunk_alloc_policy) {
5188 case BTRFS_CHUNK_ALLOC_REGULAR:
5189 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
5191 case BTRFS_CHUNK_ALLOC_ZONED:
5192 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);
5199 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
5200 struct alloc_chunk_ctl *ctl,
5201 struct btrfs_device_info *devices_info)
5203 struct btrfs_fs_info *info = fs_devices->fs_info;
5204 struct btrfs_device *device;
5206 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
5213 * in the first pass through the devices list, we gather information
5214 * about the available holes on each device.
5216 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5217 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5219 "BTRFS: read-only device in alloc_list\n");
5223 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5224 &device->dev_state) ||
5225 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5228 if (device->total_bytes > device->bytes_used)
5229 total_avail = device->total_bytes - device->bytes_used;
5233 /* If there is no space on this device, skip it. */
5234 if (total_avail < ctl->dev_extent_min)
5237 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
5239 if (ret && ret != -ENOSPC)
5243 max_avail = dev_extent_want;
5245 if (max_avail < ctl->dev_extent_min) {
5246 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5248 "%s: devid %llu has no free space, have=%llu want=%llu",
5249 __func__, device->devid, max_avail,
5250 ctl->dev_extent_min);
5254 if (ndevs == fs_devices->rw_devices) {
5255 WARN(1, "%s: found more than %llu devices\n",
5256 __func__, fs_devices->rw_devices);
5259 devices_info[ndevs].dev_offset = dev_offset;
5260 devices_info[ndevs].max_avail = max_avail;
5261 devices_info[ndevs].total_avail = total_avail;
5262 devices_info[ndevs].dev = device;
5268 * now sort the devices by hole size / available space
5270 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5271 btrfs_cmp_device_info, NULL);
5276 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5277 struct btrfs_device_info *devices_info)
5279 /* Number of stripes that count for block group size */
5283 * The primary goal is to maximize the number of stripes, so use as
5284 * many devices as possible, even if the stripes are not maximum sized.
5286 * The DUP profile stores more than one stripe per device, the
5287 * max_avail is the total size so we have to adjust.
5289 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5291 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5293 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5294 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5297 * Use the number of data stripes to figure out how big this chunk is
5298 * really going to be in terms of logical address space, and compare
5299 * that answer with the max chunk size. If it's higher, we try to
5300 * reduce stripe_size.
5302 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5304 * Reduce stripe_size, round it up to a 16MB boundary again and
5305 * then use it, unless it ends up being even bigger than the
5306 * previous value we had already.
5308 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5309 data_stripes), SZ_16M),
5313 /* Stripe size should not go beyond 1G. */
5314 ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G);
5316 /* Align to BTRFS_STRIPE_LEN */
5317 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5318 ctl->chunk_size = ctl->stripe_size * data_stripes;
5323 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,
5324 struct btrfs_device_info *devices_info)
5326 u64 zone_size = devices_info[0].dev->zone_info->zone_size;
5327 /* Number of stripes that count for block group size */
5331 * It should hold because:
5332 * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5334 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);
5336 ctl->stripe_size = zone_size;
5337 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5338 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5340 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5341 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5342 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,
5343 ctl->stripe_size) + ctl->nparity,
5345 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5346 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5347 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);
5350 ctl->chunk_size = ctl->stripe_size * data_stripes;
5355 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5356 struct alloc_chunk_ctl *ctl,
5357 struct btrfs_device_info *devices_info)
5359 struct btrfs_fs_info *info = fs_devices->fs_info;
5362 * Round down to number of usable stripes, devs_increment can be any
5363 * number so we can't use round_down() that requires power of 2, while
5364 * rounddown is safe.
5366 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5368 if (ctl->ndevs < ctl->devs_min) {
5369 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5371 "%s: not enough devices with free space: have=%d minimum required=%d",
5372 __func__, ctl->ndevs, ctl->devs_min);
5377 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5379 switch (fs_devices->chunk_alloc_policy) {
5380 case BTRFS_CHUNK_ALLOC_REGULAR:
5381 return decide_stripe_size_regular(ctl, devices_info);
5382 case BTRFS_CHUNK_ALLOC_ZONED:
5383 return decide_stripe_size_zoned(ctl, devices_info);
5389 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
5390 struct alloc_chunk_ctl *ctl,
5391 struct btrfs_device_info *devices_info)
5393 struct btrfs_fs_info *info = trans->fs_info;
5394 struct map_lookup *map = NULL;
5395 struct extent_map_tree *em_tree;
5396 struct btrfs_block_group *block_group;
5397 struct extent_map *em;
5398 u64 start = ctl->start;
5399 u64 type = ctl->type;
5404 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
5406 return ERR_PTR(-ENOMEM);
5407 map->num_stripes = ctl->num_stripes;
5409 for (i = 0; i < ctl->ndevs; ++i) {
5410 for (j = 0; j < ctl->dev_stripes; ++j) {
5411 int s = i * ctl->dev_stripes + j;
5412 map->stripes[s].dev = devices_info[i].dev;
5413 map->stripes[s].physical = devices_info[i].dev_offset +
5414 j * ctl->stripe_size;
5417 map->io_align = BTRFS_STRIPE_LEN;
5418 map->io_width = BTRFS_STRIPE_LEN;
5420 map->sub_stripes = ctl->sub_stripes;
5422 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5424 em = alloc_extent_map();
5427 return ERR_PTR(-ENOMEM);
5429 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5430 em->map_lookup = map;
5432 em->len = ctl->chunk_size;
5433 em->block_start = 0;
5434 em->block_len = em->len;
5435 em->orig_block_len = ctl->stripe_size;
5437 em_tree = &info->mapping_tree;
5438 write_lock(&em_tree->lock);
5439 ret = add_extent_mapping(em_tree, em, 0);
5441 write_unlock(&em_tree->lock);
5442 free_extent_map(em);
5443 return ERR_PTR(ret);
5445 write_unlock(&em_tree->lock);
5447 block_group = btrfs_make_block_group(trans, type, start, ctl->chunk_size);
5448 if (IS_ERR(block_group))
5449 goto error_del_extent;
5451 for (i = 0; i < map->num_stripes; i++) {
5452 struct btrfs_device *dev = map->stripes[i].dev;
5454 btrfs_device_set_bytes_used(dev,
5455 dev->bytes_used + ctl->stripe_size);
5456 if (list_empty(&dev->post_commit_list))
5457 list_add_tail(&dev->post_commit_list,
5458 &trans->transaction->dev_update_list);
5461 atomic64_sub(ctl->stripe_size * map->num_stripes,
5462 &info->free_chunk_space);
5464 free_extent_map(em);
5465 check_raid56_incompat_flag(info, type);
5466 check_raid1c34_incompat_flag(info, type);
5471 write_lock(&em_tree->lock);
5472 remove_extent_mapping(em_tree, em);
5473 write_unlock(&em_tree->lock);
5475 /* One for our allocation */
5476 free_extent_map(em);
5477 /* One for the tree reference */
5478 free_extent_map(em);
5483 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
5486 struct btrfs_fs_info *info = trans->fs_info;
5487 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5488 struct btrfs_device_info *devices_info = NULL;
5489 struct alloc_chunk_ctl ctl;
5490 struct btrfs_block_group *block_group;
5493 lockdep_assert_held(&info->chunk_mutex);
5495 if (!alloc_profile_is_valid(type, 0)) {
5497 return ERR_PTR(-EINVAL);
5500 if (list_empty(&fs_devices->alloc_list)) {
5501 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5502 btrfs_debug(info, "%s: no writable device", __func__);
5503 return ERR_PTR(-ENOSPC);
5506 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5507 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5509 return ERR_PTR(-EINVAL);
5512 ctl.start = find_next_chunk(info);
5514 init_alloc_chunk_ctl(fs_devices, &ctl);
5516 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5519 return ERR_PTR(-ENOMEM);
5521 ret = gather_device_info(fs_devices, &ctl, devices_info);
5523 block_group = ERR_PTR(ret);
5527 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5529 block_group = ERR_PTR(ret);
5533 block_group = create_chunk(trans, &ctl, devices_info);
5536 kfree(devices_info);
5541 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5542 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5545 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5548 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
5549 struct btrfs_block_group *bg)
5551 struct btrfs_fs_info *fs_info = trans->fs_info;
5552 struct btrfs_root *chunk_root = fs_info->chunk_root;
5553 struct btrfs_key key;
5554 struct btrfs_chunk *chunk;
5555 struct btrfs_stripe *stripe;
5556 struct extent_map *em;
5557 struct map_lookup *map;
5563 * We take the chunk_mutex for 2 reasons:
5565 * 1) Updates and insertions in the chunk btree must be done while holding
5566 * the chunk_mutex, as well as updating the system chunk array in the
5567 * superblock. See the comment on top of btrfs_chunk_alloc() for the
5570 * 2) To prevent races with the final phase of a device replace operation
5571 * that replaces the device object associated with the map's stripes,
5572 * because the device object's id can change at any time during that
5573 * final phase of the device replace operation
5574 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5575 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5576 * which would cause a failure when updating the device item, which does
5577 * not exists, or persisting a stripe of the chunk item with such ID.
5578 * Here we can't use the device_list_mutex because our caller already
5579 * has locked the chunk_mutex, and the final phase of device replace
5580 * acquires both mutexes - first the device_list_mutex and then the
5581 * chunk_mutex. Using any of those two mutexes protects us from a
5582 * concurrent device replace.
5584 lockdep_assert_held(&fs_info->chunk_mutex);
5586 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
5589 btrfs_abort_transaction(trans, ret);
5593 map = em->map_lookup;
5594 item_size = btrfs_chunk_item_size(map->num_stripes);
5596 chunk = kzalloc(item_size, GFP_NOFS);
5599 btrfs_abort_transaction(trans, ret);
5603 for (i = 0; i < map->num_stripes; i++) {
5604 struct btrfs_device *device = map->stripes[i].dev;
5606 ret = btrfs_update_device(trans, device);
5611 stripe = &chunk->stripe;
5612 for (i = 0; i < map->num_stripes; i++) {
5613 struct btrfs_device *device = map->stripes[i].dev;
5614 const u64 dev_offset = map->stripes[i].physical;
5616 btrfs_set_stack_stripe_devid(stripe, device->devid);
5617 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5618 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5622 btrfs_set_stack_chunk_length(chunk, bg->length);
5623 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
5624 btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
5625 btrfs_set_stack_chunk_type(chunk, map->type);
5626 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5627 btrfs_set_stack_chunk_io_align(chunk, BTRFS_STRIPE_LEN);
5628 btrfs_set_stack_chunk_io_width(chunk, BTRFS_STRIPE_LEN);
5629 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5630 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5632 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5633 key.type = BTRFS_CHUNK_ITEM_KEY;
5634 key.offset = bg->start;
5636 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5640 set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags);
5642 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5643 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5650 free_extent_map(em);
5654 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5656 struct btrfs_fs_info *fs_info = trans->fs_info;
5658 struct btrfs_block_group *meta_bg;
5659 struct btrfs_block_group *sys_bg;
5662 * When adding a new device for sprouting, the seed device is read-only
5663 * so we must first allocate a metadata and a system chunk. But before
5664 * adding the block group items to the extent, device and chunk btrees,
5667 * 1) Create both chunks without doing any changes to the btrees, as
5668 * otherwise we would get -ENOSPC since the block groups from the
5669 * seed device are read-only;
5671 * 2) Add the device item for the new sprout device - finishing the setup
5672 * of a new block group requires updating the device item in the chunk
5673 * btree, so it must exist when we attempt to do it. The previous step
5674 * ensures this does not fail with -ENOSPC.
5676 * After that we can add the block group items to their btrees:
5677 * update existing device item in the chunk btree, add a new block group
5678 * item to the extent btree, add a new chunk item to the chunk btree and
5679 * finally add the new device extent items to the devices btree.
5682 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5683 meta_bg = btrfs_create_chunk(trans, alloc_profile);
5684 if (IS_ERR(meta_bg))
5685 return PTR_ERR(meta_bg);
5687 alloc_profile = btrfs_system_alloc_profile(fs_info);
5688 sys_bg = btrfs_create_chunk(trans, alloc_profile);
5690 return PTR_ERR(sys_bg);
5695 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5697 const int index = btrfs_bg_flags_to_raid_index(map->type);
5699 return btrfs_raid_array[index].tolerated_failures;
5702 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5704 struct extent_map *em;
5705 struct map_lookup *map;
5710 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5714 map = em->map_lookup;
5715 for (i = 0; i < map->num_stripes; i++) {
5716 if (test_bit(BTRFS_DEV_STATE_MISSING,
5717 &map->stripes[i].dev->dev_state)) {
5721 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5722 &map->stripes[i].dev->dev_state)) {
5729 * If the number of missing devices is larger than max errors, we can
5730 * not write the data into that chunk successfully.
5732 if (miss_ndevs > btrfs_chunk_max_errors(map))
5735 free_extent_map(em);
5739 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5741 struct extent_map *em;
5744 write_lock(&tree->lock);
5745 em = lookup_extent_mapping(tree, 0, (u64)-1);
5747 remove_extent_mapping(tree, em);
5748 write_unlock(&tree->lock);
5752 free_extent_map(em);
5753 /* once for the tree */
5754 free_extent_map(em);
5758 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5760 struct extent_map *em;
5761 struct map_lookup *map;
5762 enum btrfs_raid_types index;
5765 em = btrfs_get_chunk_map(fs_info, logical, len);
5768 * We could return errors for these cases, but that could get
5769 * ugly and we'd probably do the same thing which is just not do
5770 * anything else and exit, so return 1 so the callers don't try
5771 * to use other copies.
5775 map = em->map_lookup;
5776 index = btrfs_bg_flags_to_raid_index(map->type);
5778 /* Non-RAID56, use their ncopies from btrfs_raid_array. */
5779 if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5780 ret = btrfs_raid_array[index].ncopies;
5781 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5783 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5785 * There could be two corrupted data stripes, we need
5786 * to loop retry in order to rebuild the correct data.
5788 * Fail a stripe at a time on every retry except the
5789 * stripe under reconstruction.
5791 ret = map->num_stripes;
5792 free_extent_map(em);
5796 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5799 struct extent_map *em;
5800 struct map_lookup *map;
5801 unsigned long len = fs_info->sectorsize;
5803 if (!btrfs_fs_incompat(fs_info, RAID56))
5806 em = btrfs_get_chunk_map(fs_info, logical, len);
5808 if (!WARN_ON(IS_ERR(em))) {
5809 map = em->map_lookup;
5810 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5811 len = btrfs_stripe_nr_to_offset(nr_data_stripes(map));
5812 free_extent_map(em);
5817 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5819 struct extent_map *em;
5820 struct map_lookup *map;
5823 if (!btrfs_fs_incompat(fs_info, RAID56))
5826 em = btrfs_get_chunk_map(fs_info, logical, len);
5828 if(!WARN_ON(IS_ERR(em))) {
5829 map = em->map_lookup;
5830 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5832 free_extent_map(em);
5837 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5838 struct map_lookup *map, int first,
5839 int dev_replace_is_ongoing)
5843 int preferred_mirror;
5845 struct btrfs_device *srcdev;
5848 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5850 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5851 num_stripes = map->sub_stripes;
5853 num_stripes = map->num_stripes;
5855 switch (fs_info->fs_devices->read_policy) {
5857 /* Shouldn't happen, just warn and use pid instead of failing */
5858 btrfs_warn_rl(fs_info,
5859 "unknown read_policy type %u, reset to pid",
5860 fs_info->fs_devices->read_policy);
5861 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID;
5863 case BTRFS_READ_POLICY_PID:
5864 preferred_mirror = first + (current->pid % num_stripes);
5868 if (dev_replace_is_ongoing &&
5869 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5870 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5871 srcdev = fs_info->dev_replace.srcdev;
5876 * try to avoid the drive that is the source drive for a
5877 * dev-replace procedure, only choose it if no other non-missing
5878 * mirror is available
5880 for (tolerance = 0; tolerance < 2; tolerance++) {
5881 if (map->stripes[preferred_mirror].dev->bdev &&
5882 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5883 return preferred_mirror;
5884 for (i = first; i < first + num_stripes; i++) {
5885 if (map->stripes[i].dev->bdev &&
5886 (tolerance || map->stripes[i].dev != srcdev))
5891 /* we couldn't find one that doesn't fail. Just return something
5892 * and the io error handling code will clean up eventually
5894 return preferred_mirror;
5897 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info,
5900 struct btrfs_io_context *bioc;
5903 /* The size of btrfs_io_context */
5904 sizeof(struct btrfs_io_context) +
5905 /* Plus the variable array for the stripes */
5906 sizeof(struct btrfs_io_stripe) * (total_stripes),
5912 refcount_set(&bioc->refs, 1);
5914 bioc->fs_info = fs_info;
5915 bioc->replace_stripe_src = -1;
5916 bioc->full_stripe_logical = (u64)-1;
5921 void btrfs_get_bioc(struct btrfs_io_context *bioc)
5923 WARN_ON(!refcount_read(&bioc->refs));
5924 refcount_inc(&bioc->refs);
5927 void btrfs_put_bioc(struct btrfs_io_context *bioc)
5931 if (refcount_dec_and_test(&bioc->refs))
5936 * Please note that, discard won't be sent to target device of device
5939 struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info,
5940 u64 logical, u64 *length_ret,
5943 struct extent_map *em;
5944 struct map_lookup *map;
5945 struct btrfs_discard_stripe *stripes;
5946 u64 length = *length_ret;
5951 u64 stripe_end_offset;
5955 u32 sub_stripes = 0;
5956 u32 stripes_per_dev = 0;
5957 u32 remaining_stripes = 0;
5958 u32 last_stripe = 0;
5962 em = btrfs_get_chunk_map(fs_info, logical, length);
5964 return ERR_CAST(em);
5966 map = em->map_lookup;
5968 /* we don't discard raid56 yet */
5969 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5974 offset = logical - em->start;
5975 length = min_t(u64, em->start + em->len - logical, length);
5976 *length_ret = length;
5979 * stripe_nr counts the total number of stripes we have to stride
5980 * to get to this block
5982 stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
5984 /* stripe_offset is the offset of this block in its stripe */
5985 stripe_offset = offset - btrfs_stripe_nr_to_offset(stripe_nr);
5987 stripe_nr_end = round_up(offset + length, BTRFS_STRIPE_LEN) >>
5988 BTRFS_STRIPE_LEN_SHIFT;
5989 stripe_cnt = stripe_nr_end - stripe_nr;
5990 stripe_end_offset = btrfs_stripe_nr_to_offset(stripe_nr_end) -
5993 * after this, stripe_nr is the number of stripes on this
5994 * device we have to walk to find the data, and stripe_index is
5995 * the number of our device in the stripe array
5999 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6000 BTRFS_BLOCK_GROUP_RAID10)) {
6001 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6004 sub_stripes = map->sub_stripes;
6006 factor = map->num_stripes / sub_stripes;
6007 *num_stripes = min_t(u64, map->num_stripes,
6008 sub_stripes * stripe_cnt);
6009 stripe_index = stripe_nr % factor;
6010 stripe_nr /= factor;
6011 stripe_index *= sub_stripes;
6013 remaining_stripes = stripe_cnt % factor;
6014 stripes_per_dev = stripe_cnt / factor;
6015 last_stripe = ((stripe_nr_end - 1) % factor) * sub_stripes;
6016 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
6017 BTRFS_BLOCK_GROUP_DUP)) {
6018 *num_stripes = map->num_stripes;
6020 stripe_index = stripe_nr % map->num_stripes;
6021 stripe_nr /= map->num_stripes;
6024 stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS);
6030 for (i = 0; i < *num_stripes; i++) {
6031 stripes[i].physical =
6032 map->stripes[stripe_index].physical +
6033 stripe_offset + btrfs_stripe_nr_to_offset(stripe_nr);
6034 stripes[i].dev = map->stripes[stripe_index].dev;
6036 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6037 BTRFS_BLOCK_GROUP_RAID10)) {
6038 stripes[i].length = btrfs_stripe_nr_to_offset(stripes_per_dev);
6040 if (i / sub_stripes < remaining_stripes)
6041 stripes[i].length += BTRFS_STRIPE_LEN;
6044 * Special for the first stripe and
6047 * |-------|...|-------|
6051 if (i < sub_stripes)
6052 stripes[i].length -= stripe_offset;
6054 if (stripe_index >= last_stripe &&
6055 stripe_index <= (last_stripe +
6057 stripes[i].length -= stripe_end_offset;
6059 if (i == sub_stripes - 1)
6062 stripes[i].length = length;
6066 if (stripe_index == map->num_stripes) {
6072 free_extent_map(em);
6075 free_extent_map(em);
6076 return ERR_PTR(ret);
6079 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)
6081 struct btrfs_block_group *cache;
6084 /* Non zoned filesystem does not use "to_copy" flag */
6085 if (!btrfs_is_zoned(fs_info))
6088 cache = btrfs_lookup_block_group(fs_info, logical);
6090 ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags);
6092 btrfs_put_block_group(cache);
6096 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
6097 struct btrfs_io_context *bioc,
6098 struct btrfs_dev_replace *dev_replace,
6100 int *num_stripes_ret, int *max_errors_ret)
6102 u64 srcdev_devid = dev_replace->srcdev->devid;
6104 * At this stage, num_stripes is still the real number of stripes,
6105 * excluding the duplicated stripes.
6107 int num_stripes = *num_stripes_ret;
6108 int nr_extra_stripes = 0;
6109 int max_errors = *max_errors_ret;
6113 * A block group which has "to_copy" set will eventually be copied by
6114 * the dev-replace process. We can avoid cloning IO here.
6116 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))
6120 * Duplicate the write operations while the dev-replace procedure is
6121 * running. Since the copying of the old disk to the new disk takes
6122 * place at run time while the filesystem is mounted writable, the
6123 * regular write operations to the old disk have to be duplicated to go
6124 * to the new disk as well.
6126 * Note that device->missing is handled by the caller, and that the
6127 * write to the old disk is already set up in the stripes array.
6129 for (i = 0; i < num_stripes; i++) {
6130 struct btrfs_io_stripe *old = &bioc->stripes[i];
6131 struct btrfs_io_stripe *new = &bioc->stripes[num_stripes + nr_extra_stripes];
6133 if (old->dev->devid != srcdev_devid)
6136 new->physical = old->physical;
6137 new->dev = dev_replace->tgtdev;
6138 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK)
6139 bioc->replace_stripe_src = i;
6143 /* We can only have at most 2 extra nr_stripes (for DUP). */
6144 ASSERT(nr_extra_stripes <= 2);
6146 * For GET_READ_MIRRORS, we can only return at most 1 extra stripe for
6148 * If we have 2 extra stripes, only choose the one with smaller physical.
6150 if (op == BTRFS_MAP_GET_READ_MIRRORS && nr_extra_stripes == 2) {
6151 struct btrfs_io_stripe *first = &bioc->stripes[num_stripes];
6152 struct btrfs_io_stripe *second = &bioc->stripes[num_stripes + 1];
6154 /* Only DUP can have two extra stripes. */
6155 ASSERT(bioc->map_type & BTRFS_BLOCK_GROUP_DUP);
6158 * Swap the last stripe stripes and reduce @nr_extra_stripes.
6159 * The extra stripe would still be there, but won't be accessed.
6161 if (first->physical > second->physical) {
6162 swap(second->physical, first->physical);
6163 swap(second->dev, first->dev);
6168 *num_stripes_ret = num_stripes + nr_extra_stripes;
6169 *max_errors_ret = max_errors + nr_extra_stripes;
6170 bioc->replace_nr_stripes = nr_extra_stripes;
6173 static u64 btrfs_max_io_len(struct map_lookup *map, enum btrfs_map_op op,
6174 u64 offset, u32 *stripe_nr, u64 *stripe_offset,
6175 u64 *full_stripe_start)
6178 * Stripe_nr is the stripe where this block falls. stripe_offset is
6179 * the offset of this block in its stripe.
6181 *stripe_offset = offset & BTRFS_STRIPE_LEN_MASK;
6182 *stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
6183 ASSERT(*stripe_offset < U32_MAX);
6185 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6186 unsigned long full_stripe_len =
6187 btrfs_stripe_nr_to_offset(nr_data_stripes(map));
6190 * For full stripe start, we use previously calculated
6191 * @stripe_nr. Align it to nr_data_stripes, then multiply with
6194 * By this we can avoid u64 division completely. And we have
6195 * to go rounddown(), not round_down(), as nr_data_stripes is
6196 * not ensured to be power of 2.
6198 *full_stripe_start =
6199 btrfs_stripe_nr_to_offset(
6200 rounddown(*stripe_nr, nr_data_stripes(map)));
6202 ASSERT(*full_stripe_start + full_stripe_len > offset);
6203 ASSERT(*full_stripe_start <= offset);
6205 * For writes to RAID56, allow to write a full stripe set, but
6206 * no straddling of stripe sets.
6208 if (op == BTRFS_MAP_WRITE)
6209 return full_stripe_len - (offset - *full_stripe_start);
6213 * For other RAID types and for RAID56 reads, allow a single stripe (on
6216 if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK)
6217 return BTRFS_STRIPE_LEN - *stripe_offset;
6221 static void set_io_stripe(struct btrfs_io_stripe *dst, const struct map_lookup *map,
6222 u32 stripe_index, u64 stripe_offset, u32 stripe_nr)
6224 dst->dev = map->stripes[stripe_index].dev;
6225 dst->physical = map->stripes[stripe_index].physical +
6226 stripe_offset + btrfs_stripe_nr_to_offset(stripe_nr);
6229 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6230 u64 logical, u64 *length,
6231 struct btrfs_io_context **bioc_ret,
6232 struct btrfs_io_stripe *smap, int *mirror_num_ret,
6235 struct extent_map *em;
6236 struct map_lookup *map;
6244 int mirror_num = (mirror_num_ret ? *mirror_num_ret : 0);
6248 struct btrfs_io_context *bioc = NULL;
6249 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6250 int dev_replace_is_ongoing = 0;
6251 u16 num_alloc_stripes;
6252 u64 raid56_full_stripe_start = (u64)-1;
6257 num_copies = btrfs_num_copies(fs_info, logical, fs_info->sectorsize);
6258 if (mirror_num > num_copies)
6261 em = btrfs_get_chunk_map(fs_info, logical, *length);
6265 map = em->map_lookup;
6266 data_stripes = nr_data_stripes(map);
6268 map_offset = logical - em->start;
6269 max_len = btrfs_max_io_len(map, op, map_offset, &stripe_nr,
6270 &stripe_offset, &raid56_full_stripe_start);
6271 *length = min_t(u64, em->len - map_offset, max_len);
6273 down_read(&dev_replace->rwsem);
6274 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6276 * Hold the semaphore for read during the whole operation, write is
6277 * requested at commit time but must wait.
6279 if (!dev_replace_is_ongoing)
6280 up_read(&dev_replace->rwsem);
6284 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6285 stripe_index = stripe_nr % map->num_stripes;
6286 stripe_nr /= map->num_stripes;
6287 if (op == BTRFS_MAP_READ)
6289 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6290 if (op != BTRFS_MAP_READ) {
6291 num_stripes = map->num_stripes;
6292 } else if (mirror_num) {
6293 stripe_index = mirror_num - 1;
6295 stripe_index = find_live_mirror(fs_info, map, 0,
6296 dev_replace_is_ongoing);
6297 mirror_num = stripe_index + 1;
6300 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6301 if (op != BTRFS_MAP_READ) {
6302 num_stripes = map->num_stripes;
6303 } else if (mirror_num) {
6304 stripe_index = mirror_num - 1;
6309 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6310 u32 factor = map->num_stripes / map->sub_stripes;
6312 stripe_index = (stripe_nr % factor) * map->sub_stripes;
6313 stripe_nr /= factor;
6315 if (op != BTRFS_MAP_READ)
6316 num_stripes = map->sub_stripes;
6317 else if (mirror_num)
6318 stripe_index += mirror_num - 1;
6320 int old_stripe_index = stripe_index;
6321 stripe_index = find_live_mirror(fs_info, map,
6323 dev_replace_is_ongoing);
6324 mirror_num = stripe_index - old_stripe_index + 1;
6327 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6328 if (need_raid_map && (op != BTRFS_MAP_READ || mirror_num > 1)) {
6330 * Push stripe_nr back to the start of the full stripe
6331 * For those cases needing a full stripe, @stripe_nr
6332 * is the full stripe number.
6334 * Originally we go raid56_full_stripe_start / full_stripe_len,
6335 * but that can be expensive. Here we just divide
6336 * @stripe_nr with @data_stripes.
6338 stripe_nr /= data_stripes;
6340 /* RAID[56] write or recovery. Return all stripes */
6341 num_stripes = map->num_stripes;
6342 max_errors = btrfs_chunk_max_errors(map);
6344 /* Return the length to the full stripe end */
6345 *length = min(logical + *length,
6346 raid56_full_stripe_start + em->start +
6347 btrfs_stripe_nr_to_offset(data_stripes)) -
6353 * Mirror #0 or #1 means the original data block.
6354 * Mirror #2 is RAID5 parity block.
6355 * Mirror #3 is RAID6 Q block.
6357 stripe_index = stripe_nr % data_stripes;
6358 stripe_nr /= data_stripes;
6360 stripe_index = data_stripes + mirror_num - 2;
6362 /* We distribute the parity blocks across stripes */
6363 stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
6364 if (op == BTRFS_MAP_READ && mirror_num <= 1)
6369 * After this, stripe_nr is the number of stripes on this
6370 * device we have to walk to find the data, and stripe_index is
6371 * the number of our device in the stripe array
6373 stripe_index = stripe_nr % map->num_stripes;
6374 stripe_nr /= map->num_stripes;
6375 mirror_num = stripe_index + 1;
6377 if (stripe_index >= map->num_stripes) {
6379 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6380 stripe_index, map->num_stripes);
6385 num_alloc_stripes = num_stripes;
6386 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6387 op != BTRFS_MAP_READ)
6389 * For replace case, we need to add extra stripes for extra
6390 * duplicated stripes.
6392 * For both WRITE and GET_READ_MIRRORS, we may have at most
6393 * 2 more stripes (DUP types, otherwise 1).
6395 num_alloc_stripes += 2;
6398 * If this I/O maps to a single device, try to return the device and
6399 * physical block information on the stack instead of allocating an
6400 * I/O context structure.
6402 if (smap && num_alloc_stripes == 1 &&
6403 !((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1) &&
6404 (op == BTRFS_MAP_READ || !dev_replace_is_ongoing ||
6405 !dev_replace->tgtdev)) {
6406 set_io_stripe(smap, map, stripe_index, stripe_offset, stripe_nr);
6407 *mirror_num_ret = mirror_num;
6413 bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes);
6418 bioc->map_type = map->type;
6421 * For RAID56 full map, we need to make sure the stripes[] follows the
6422 * rule that data stripes are all ordered, then followed with P and Q
6425 * It's still mostly the same as other profiles, just with extra rotation.
6427 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6428 (op != BTRFS_MAP_READ || mirror_num > 1)) {
6430 * For RAID56 @stripe_nr is already the number of full stripes
6431 * before us, which is also the rotation value (needs to modulo
6432 * with num_stripes).
6434 * In this case, we just add @stripe_nr with @i, then do the
6435 * modulo, to reduce one modulo call.
6437 bioc->full_stripe_logical = em->start +
6438 btrfs_stripe_nr_to_offset(stripe_nr * data_stripes);
6439 for (i = 0; i < num_stripes; i++)
6440 set_io_stripe(&bioc->stripes[i], map,
6441 (i + stripe_nr) % num_stripes,
6442 stripe_offset, stripe_nr);
6445 * For all other non-RAID56 profiles, just copy the target
6446 * stripe into the bioc.
6448 for (i = 0; i < num_stripes; i++) {
6449 set_io_stripe(&bioc->stripes[i], map, stripe_index,
6450 stripe_offset, stripe_nr);
6455 if (op != BTRFS_MAP_READ)
6456 max_errors = btrfs_chunk_max_errors(map);
6458 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6459 op != BTRFS_MAP_READ) {
6460 handle_ops_on_dev_replace(op, bioc, dev_replace, logical,
6461 &num_stripes, &max_errors);
6465 bioc->num_stripes = num_stripes;
6466 bioc->max_errors = max_errors;
6467 bioc->mirror_num = mirror_num;
6470 if (dev_replace_is_ongoing) {
6471 lockdep_assert_held(&dev_replace->rwsem);
6472 /* Unlock and let waiting writers proceed */
6473 up_read(&dev_replace->rwsem);
6475 free_extent_map(em);
6479 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args,
6480 const struct btrfs_fs_devices *fs_devices)
6482 if (args->fsid == NULL)
6484 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0)
6489 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args,
6490 const struct btrfs_device *device)
6492 if (args->missing) {
6493 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) &&
6499 if (device->devid != args->devid)
6501 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0)
6507 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6510 * If devid and uuid are both specified, the match must be exact, otherwise
6511 * only devid is used.
6513 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,
6514 const struct btrfs_dev_lookup_args *args)
6516 struct btrfs_device *device;
6517 struct btrfs_fs_devices *seed_devs;
6519 if (dev_args_match_fs_devices(args, fs_devices)) {
6520 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6521 if (dev_args_match_device(args, device))
6526 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6527 if (!dev_args_match_fs_devices(args, seed_devs))
6529 list_for_each_entry(device, &seed_devs->devices, dev_list) {
6530 if (dev_args_match_device(args, device))
6538 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6539 u64 devid, u8 *dev_uuid)
6541 struct btrfs_device *device;
6542 unsigned int nofs_flag;
6545 * We call this under the chunk_mutex, so we want to use NOFS for this
6546 * allocation, however we don't want to change btrfs_alloc_device() to
6547 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6551 nofs_flag = memalloc_nofs_save();
6552 device = btrfs_alloc_device(NULL, &devid, dev_uuid, NULL);
6553 memalloc_nofs_restore(nofs_flag);
6557 list_add(&device->dev_list, &fs_devices->devices);
6558 device->fs_devices = fs_devices;
6559 fs_devices->num_devices++;
6561 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6562 fs_devices->missing_devices++;
6568 * Allocate new device struct, set up devid and UUID.
6570 * @fs_info: used only for generating a new devid, can be NULL if
6571 * devid is provided (i.e. @devid != NULL).
6572 * @devid: a pointer to devid for this device. If NULL a new devid
6574 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6576 * @path: a pointer to device path if available, NULL otherwise.
6578 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6579 * on error. Returned struct is not linked onto any lists and must be
6580 * destroyed with btrfs_free_device.
6582 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6583 const u64 *devid, const u8 *uuid,
6586 struct btrfs_device *dev;
6589 if (WARN_ON(!devid && !fs_info))
6590 return ERR_PTR(-EINVAL);
6592 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
6594 return ERR_PTR(-ENOMEM);
6596 INIT_LIST_HEAD(&dev->dev_list);
6597 INIT_LIST_HEAD(&dev->dev_alloc_list);
6598 INIT_LIST_HEAD(&dev->post_commit_list);
6600 atomic_set(&dev->dev_stats_ccnt, 0);
6601 btrfs_device_data_ordered_init(dev);
6602 extent_io_tree_init(fs_info, &dev->alloc_state, IO_TREE_DEVICE_ALLOC_STATE);
6609 ret = find_next_devid(fs_info, &tmp);
6611 btrfs_free_device(dev);
6612 return ERR_PTR(ret);
6618 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6620 generate_random_uuid(dev->uuid);
6623 struct rcu_string *name;
6625 name = rcu_string_strdup(path, GFP_KERNEL);
6627 btrfs_free_device(dev);
6628 return ERR_PTR(-ENOMEM);
6630 rcu_assign_pointer(dev->name, name);
6636 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6637 u64 devid, u8 *uuid, bool error)
6640 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6643 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6647 u64 btrfs_calc_stripe_length(const struct extent_map *em)
6649 const struct map_lookup *map = em->map_lookup;
6650 const int data_stripes = calc_data_stripes(map->type, map->num_stripes);
6652 return div_u64(em->len, data_stripes);
6655 #if BITS_PER_LONG == 32
6657 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
6658 * can't be accessed on 32bit systems.
6660 * This function do mount time check to reject the fs if it already has
6661 * metadata chunk beyond that limit.
6663 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6664 u64 logical, u64 length, u64 type)
6666 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6669 if (logical + length < MAX_LFS_FILESIZE)
6672 btrfs_err_32bit_limit(fs_info);
6677 * This is to give early warning for any metadata chunk reaching
6678 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
6679 * Although we can still access the metadata, it's not going to be possible
6680 * once the limit is reached.
6682 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6683 u64 logical, u64 length, u64 type)
6685 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6688 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)
6691 btrfs_warn_32bit_limit(fs_info);
6695 static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info,
6696 u64 devid, u8 *uuid)
6698 struct btrfs_device *dev;
6700 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6701 btrfs_report_missing_device(fs_info, devid, uuid, true);
6702 return ERR_PTR(-ENOENT);
6705 dev = add_missing_dev(fs_info->fs_devices, devid, uuid);
6707 btrfs_err(fs_info, "failed to init missing device %llu: %ld",
6708 devid, PTR_ERR(dev));
6711 btrfs_report_missing_device(fs_info, devid, uuid, false);
6716 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6717 struct btrfs_chunk *chunk)
6719 BTRFS_DEV_LOOKUP_ARGS(args);
6720 struct btrfs_fs_info *fs_info = leaf->fs_info;
6721 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6722 struct map_lookup *map;
6723 struct extent_map *em;
6728 u8 uuid[BTRFS_UUID_SIZE];
6734 logical = key->offset;
6735 length = btrfs_chunk_length(leaf, chunk);
6736 type = btrfs_chunk_type(leaf, chunk);
6737 index = btrfs_bg_flags_to_raid_index(type);
6738 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6740 #if BITS_PER_LONG == 32
6741 ret = check_32bit_meta_chunk(fs_info, logical, length, type);
6744 warn_32bit_meta_chunk(fs_info, logical, length, type);
6748 * Only need to verify chunk item if we're reading from sys chunk array,
6749 * as chunk item in tree block is already verified by tree-checker.
6751 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6752 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6757 read_lock(&map_tree->lock);
6758 em = lookup_extent_mapping(map_tree, logical, 1);
6759 read_unlock(&map_tree->lock);
6761 /* already mapped? */
6762 if (em && em->start <= logical && em->start + em->len > logical) {
6763 free_extent_map(em);
6766 free_extent_map(em);
6769 em = alloc_extent_map();
6772 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6774 free_extent_map(em);
6778 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6779 em->map_lookup = map;
6780 em->start = logical;
6783 em->block_start = 0;
6784 em->block_len = em->len;
6786 map->num_stripes = num_stripes;
6787 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6788 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6791 * We can't use the sub_stripes value, as for profiles other than
6792 * RAID10, they may have 0 as sub_stripes for filesystems created by
6793 * older mkfs (<v5.4).
6794 * In that case, it can cause divide-by-zero errors later.
6795 * Since currently sub_stripes is fixed for each profile, let's
6796 * use the trusted value instead.
6798 map->sub_stripes = btrfs_raid_array[index].sub_stripes;
6799 map->verified_stripes = 0;
6800 em->orig_block_len = btrfs_calc_stripe_length(em);
6801 for (i = 0; i < num_stripes; i++) {
6802 map->stripes[i].physical =
6803 btrfs_stripe_offset_nr(leaf, chunk, i);
6804 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6806 read_extent_buffer(leaf, uuid, (unsigned long)
6807 btrfs_stripe_dev_uuid_nr(chunk, i),
6810 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args);
6811 if (!map->stripes[i].dev) {
6812 map->stripes[i].dev = handle_missing_device(fs_info,
6814 if (IS_ERR(map->stripes[i].dev)) {
6815 ret = PTR_ERR(map->stripes[i].dev);
6816 free_extent_map(em);
6821 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6822 &(map->stripes[i].dev->dev_state));
6825 write_lock(&map_tree->lock);
6826 ret = add_extent_mapping(map_tree, em, 0);
6827 write_unlock(&map_tree->lock);
6830 "failed to add chunk map, start=%llu len=%llu: %d",
6831 em->start, em->len, ret);
6833 free_extent_map(em);
6838 static void fill_device_from_item(struct extent_buffer *leaf,
6839 struct btrfs_dev_item *dev_item,
6840 struct btrfs_device *device)
6844 device->devid = btrfs_device_id(leaf, dev_item);
6845 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6846 device->total_bytes = device->disk_total_bytes;
6847 device->commit_total_bytes = device->disk_total_bytes;
6848 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6849 device->commit_bytes_used = device->bytes_used;
6850 device->type = btrfs_device_type(leaf, dev_item);
6851 device->io_align = btrfs_device_io_align(leaf, dev_item);
6852 device->io_width = btrfs_device_io_width(leaf, dev_item);
6853 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6854 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6855 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6857 ptr = btrfs_device_uuid(dev_item);
6858 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6861 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6864 struct btrfs_fs_devices *fs_devices;
6867 lockdep_assert_held(&uuid_mutex);
6870 /* This will match only for multi-device seed fs */
6871 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
6872 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6876 fs_devices = find_fsid(fsid, NULL);
6878 if (!btrfs_test_opt(fs_info, DEGRADED))
6879 return ERR_PTR(-ENOENT);
6881 fs_devices = alloc_fs_devices(fsid, NULL);
6882 if (IS_ERR(fs_devices))
6885 fs_devices->seeding = true;
6886 fs_devices->opened = 1;
6891 * Upon first call for a seed fs fsid, just create a private copy of the
6892 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
6894 fs_devices = clone_fs_devices(fs_devices);
6895 if (IS_ERR(fs_devices))
6898 ret = open_fs_devices(fs_devices, BLK_OPEN_READ, fs_info->bdev_holder);
6900 free_fs_devices(fs_devices);
6901 return ERR_PTR(ret);
6904 if (!fs_devices->seeding) {
6905 close_fs_devices(fs_devices);
6906 free_fs_devices(fs_devices);
6907 return ERR_PTR(-EINVAL);
6910 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
6915 static int read_one_dev(struct extent_buffer *leaf,
6916 struct btrfs_dev_item *dev_item)
6918 BTRFS_DEV_LOOKUP_ARGS(args);
6919 struct btrfs_fs_info *fs_info = leaf->fs_info;
6920 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6921 struct btrfs_device *device;
6924 u8 fs_uuid[BTRFS_FSID_SIZE];
6925 u8 dev_uuid[BTRFS_UUID_SIZE];
6927 devid = btrfs_device_id(leaf, dev_item);
6929 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6931 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6933 args.uuid = dev_uuid;
6934 args.fsid = fs_uuid;
6936 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
6937 fs_devices = open_seed_devices(fs_info, fs_uuid);
6938 if (IS_ERR(fs_devices))
6939 return PTR_ERR(fs_devices);
6942 device = btrfs_find_device(fs_info->fs_devices, &args);
6944 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6945 btrfs_report_missing_device(fs_info, devid,
6950 device = add_missing_dev(fs_devices, devid, dev_uuid);
6951 if (IS_ERR(device)) {
6953 "failed to add missing dev %llu: %ld",
6954 devid, PTR_ERR(device));
6955 return PTR_ERR(device);
6957 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6959 if (!device->bdev) {
6960 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6961 btrfs_report_missing_device(fs_info,
6962 devid, dev_uuid, true);
6965 btrfs_report_missing_device(fs_info, devid,
6969 if (!device->bdev &&
6970 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6972 * this happens when a device that was properly setup
6973 * in the device info lists suddenly goes bad.
6974 * device->bdev is NULL, and so we have to set
6975 * device->missing to one here
6977 device->fs_devices->missing_devices++;
6978 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6981 /* Move the device to its own fs_devices */
6982 if (device->fs_devices != fs_devices) {
6983 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6984 &device->dev_state));
6986 list_move(&device->dev_list, &fs_devices->devices);
6987 device->fs_devices->num_devices--;
6988 fs_devices->num_devices++;
6990 device->fs_devices->missing_devices--;
6991 fs_devices->missing_devices++;
6993 device->fs_devices = fs_devices;
6997 if (device->fs_devices != fs_info->fs_devices) {
6998 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6999 if (device->generation !=
7000 btrfs_device_generation(leaf, dev_item))
7004 fill_device_from_item(leaf, dev_item, device);
7006 u64 max_total_bytes = bdev_nr_bytes(device->bdev);
7008 if (device->total_bytes > max_total_bytes) {
7010 "device total_bytes should be at most %llu but found %llu",
7011 max_total_bytes, device->total_bytes);
7015 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7016 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7017 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7018 device->fs_devices->total_rw_bytes += device->total_bytes;
7019 atomic64_add(device->total_bytes - device->bytes_used,
7020 &fs_info->free_chunk_space);
7026 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7028 struct btrfs_super_block *super_copy = fs_info->super_copy;
7029 struct extent_buffer *sb;
7030 struct btrfs_disk_key *disk_key;
7031 struct btrfs_chunk *chunk;
7033 unsigned long sb_array_offset;
7040 struct btrfs_key key;
7042 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7045 * We allocated a dummy extent, just to use extent buffer accessors.
7046 * There will be unused space after BTRFS_SUPER_INFO_SIZE, but
7047 * that's fine, we will not go beyond system chunk array anyway.
7049 sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET);
7052 set_extent_buffer_uptodate(sb);
7054 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7055 array_size = btrfs_super_sys_array_size(super_copy);
7057 array_ptr = super_copy->sys_chunk_array;
7058 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7061 while (cur_offset < array_size) {
7062 disk_key = (struct btrfs_disk_key *)array_ptr;
7063 len = sizeof(*disk_key);
7064 if (cur_offset + len > array_size)
7065 goto out_short_read;
7067 btrfs_disk_key_to_cpu(&key, disk_key);
7070 sb_array_offset += len;
7073 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7075 "unexpected item type %u in sys_array at offset %u",
7076 (u32)key.type, cur_offset);
7081 chunk = (struct btrfs_chunk *)sb_array_offset;
7083 * At least one btrfs_chunk with one stripe must be present,
7084 * exact stripe count check comes afterwards
7086 len = btrfs_chunk_item_size(1);
7087 if (cur_offset + len > array_size)
7088 goto out_short_read;
7090 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7093 "invalid number of stripes %u in sys_array at offset %u",
7094 num_stripes, cur_offset);
7099 type = btrfs_chunk_type(sb, chunk);
7100 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7102 "invalid chunk type %llu in sys_array at offset %u",
7108 len = btrfs_chunk_item_size(num_stripes);
7109 if (cur_offset + len > array_size)
7110 goto out_short_read;
7112 ret = read_one_chunk(&key, sb, chunk);
7117 sb_array_offset += len;
7120 clear_extent_buffer_uptodate(sb);
7121 free_extent_buffer_stale(sb);
7125 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7127 clear_extent_buffer_uptodate(sb);
7128 free_extent_buffer_stale(sb);
7133 * Check if all chunks in the fs are OK for read-write degraded mount
7135 * If the @failing_dev is specified, it's accounted as missing.
7137 * Return true if all chunks meet the minimal RW mount requirements.
7138 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7140 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7141 struct btrfs_device *failing_dev)
7143 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7144 struct extent_map *em;
7148 read_lock(&map_tree->lock);
7149 em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7150 read_unlock(&map_tree->lock);
7151 /* No chunk at all? Return false anyway */
7157 struct map_lookup *map;
7162 map = em->map_lookup;
7164 btrfs_get_num_tolerated_disk_barrier_failures(
7166 for (i = 0; i < map->num_stripes; i++) {
7167 struct btrfs_device *dev = map->stripes[i].dev;
7169 if (!dev || !dev->bdev ||
7170 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7171 dev->last_flush_error)
7173 else if (failing_dev && failing_dev == dev)
7176 if (missing > max_tolerated) {
7179 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7180 em->start, missing, max_tolerated);
7181 free_extent_map(em);
7185 next_start = extent_map_end(em);
7186 free_extent_map(em);
7188 read_lock(&map_tree->lock);
7189 em = lookup_extent_mapping(map_tree, next_start,
7190 (u64)(-1) - next_start);
7191 read_unlock(&map_tree->lock);
7197 static void readahead_tree_node_children(struct extent_buffer *node)
7200 const int nr_items = btrfs_header_nritems(node);
7202 for (i = 0; i < nr_items; i++)
7203 btrfs_readahead_node_child(node, i);
7206 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7208 struct btrfs_root *root = fs_info->chunk_root;
7209 struct btrfs_path *path;
7210 struct extent_buffer *leaf;
7211 struct btrfs_key key;
7212 struct btrfs_key found_key;
7217 u64 last_ra_node = 0;
7219 path = btrfs_alloc_path();
7224 * uuid_mutex is needed only if we are mounting a sprout FS
7225 * otherwise we don't need it.
7227 mutex_lock(&uuid_mutex);
7230 * It is possible for mount and umount to race in such a way that
7231 * we execute this code path, but open_fs_devices failed to clear
7232 * total_rw_bytes. We certainly want it cleared before reading the
7233 * device items, so clear it here.
7235 fs_info->fs_devices->total_rw_bytes = 0;
7238 * Lockdep complains about possible circular locking dependency between
7239 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores
7240 * used for freeze procection of a fs (struct super_block.s_writers),
7241 * which we take when starting a transaction, and extent buffers of the
7242 * chunk tree if we call read_one_dev() while holding a lock on an
7243 * extent buffer of the chunk tree. Since we are mounting the filesystem
7244 * and at this point there can't be any concurrent task modifying the
7245 * chunk tree, to keep it simple, just skip locking on the chunk tree.
7247 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
7248 path->skip_locking = 1;
7251 * Read all device items, and then all the chunk items. All
7252 * device items are found before any chunk item (their object id
7253 * is smaller than the lowest possible object id for a chunk
7254 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7256 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7259 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) {
7260 struct extent_buffer *node = path->nodes[1];
7262 leaf = path->nodes[0];
7263 slot = path->slots[0];
7266 if (last_ra_node != node->start) {
7267 readahead_tree_node_children(node);
7268 last_ra_node = node->start;
7271 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7272 struct btrfs_dev_item *dev_item;
7273 dev_item = btrfs_item_ptr(leaf, slot,
7274 struct btrfs_dev_item);
7275 ret = read_one_dev(leaf, dev_item);
7279 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7280 struct btrfs_chunk *chunk;
7283 * We are only called at mount time, so no need to take
7284 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7285 * we always lock first fs_info->chunk_mutex before
7286 * acquiring any locks on the chunk tree. This is a
7287 * requirement for chunk allocation, see the comment on
7288 * top of btrfs_chunk_alloc() for details.
7290 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7291 ret = read_one_chunk(&found_key, leaf, chunk);
7296 /* Catch error found during iteration */
7303 * After loading chunk tree, we've got all device information,
7304 * do another round of validation checks.
7306 if (total_dev != fs_info->fs_devices->total_devices) {
7308 "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit",
7309 btrfs_super_num_devices(fs_info->super_copy),
7311 fs_info->fs_devices->total_devices = total_dev;
7312 btrfs_set_super_num_devices(fs_info->super_copy, total_dev);
7314 if (btrfs_super_total_bytes(fs_info->super_copy) <
7315 fs_info->fs_devices->total_rw_bytes) {
7317 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7318 btrfs_super_total_bytes(fs_info->super_copy),
7319 fs_info->fs_devices->total_rw_bytes);
7325 mutex_unlock(&uuid_mutex);
7327 btrfs_free_path(path);
7331 int btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7333 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7334 struct btrfs_device *device;
7337 fs_devices->fs_info = fs_info;
7339 mutex_lock(&fs_devices->device_list_mutex);
7340 list_for_each_entry(device, &fs_devices->devices, dev_list)
7341 device->fs_info = fs_info;
7343 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7344 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7345 device->fs_info = fs_info;
7346 ret = btrfs_get_dev_zone_info(device, false);
7351 seed_devs->fs_info = fs_info;
7353 mutex_unlock(&fs_devices->device_list_mutex);
7358 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7359 const struct btrfs_dev_stats_item *ptr,
7364 read_extent_buffer(eb, &val,
7365 offsetof(struct btrfs_dev_stats_item, values) +
7366 ((unsigned long)ptr) + (index * sizeof(u64)),
7371 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7372 struct btrfs_dev_stats_item *ptr,
7375 write_extent_buffer(eb, &val,
7376 offsetof(struct btrfs_dev_stats_item, values) +
7377 ((unsigned long)ptr) + (index * sizeof(u64)),
7381 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7382 struct btrfs_path *path)
7384 struct btrfs_dev_stats_item *ptr;
7385 struct extent_buffer *eb;
7386 struct btrfs_key key;
7390 if (!device->fs_info->dev_root)
7393 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7394 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7395 key.offset = device->devid;
7396 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7398 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7399 btrfs_dev_stat_set(device, i, 0);
7400 device->dev_stats_valid = 1;
7401 btrfs_release_path(path);
7402 return ret < 0 ? ret : 0;
7404 slot = path->slots[0];
7405 eb = path->nodes[0];
7406 item_size = btrfs_item_size(eb, slot);
7408 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7410 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7411 if (item_size >= (1 + i) * sizeof(__le64))
7412 btrfs_dev_stat_set(device, i,
7413 btrfs_dev_stats_value(eb, ptr, i));
7415 btrfs_dev_stat_set(device, i, 0);
7418 device->dev_stats_valid = 1;
7419 btrfs_dev_stat_print_on_load(device);
7420 btrfs_release_path(path);
7425 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7427 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7428 struct btrfs_device *device;
7429 struct btrfs_path *path = NULL;
7432 path = btrfs_alloc_path();
7436 mutex_lock(&fs_devices->device_list_mutex);
7437 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7438 ret = btrfs_device_init_dev_stats(device, path);
7442 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7443 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7444 ret = btrfs_device_init_dev_stats(device, path);
7450 mutex_unlock(&fs_devices->device_list_mutex);
7452 btrfs_free_path(path);
7456 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7457 struct btrfs_device *device)
7459 struct btrfs_fs_info *fs_info = trans->fs_info;
7460 struct btrfs_root *dev_root = fs_info->dev_root;
7461 struct btrfs_path *path;
7462 struct btrfs_key key;
7463 struct extent_buffer *eb;
7464 struct btrfs_dev_stats_item *ptr;
7468 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7469 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7470 key.offset = device->devid;
7472 path = btrfs_alloc_path();
7475 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7477 btrfs_warn_in_rcu(fs_info,
7478 "error %d while searching for dev_stats item for device %s",
7479 ret, btrfs_dev_name(device));
7484 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7485 /* need to delete old one and insert a new one */
7486 ret = btrfs_del_item(trans, dev_root, path);
7488 btrfs_warn_in_rcu(fs_info,
7489 "delete too small dev_stats item for device %s failed %d",
7490 btrfs_dev_name(device), ret);
7497 /* need to insert a new item */
7498 btrfs_release_path(path);
7499 ret = btrfs_insert_empty_item(trans, dev_root, path,
7500 &key, sizeof(*ptr));
7502 btrfs_warn_in_rcu(fs_info,
7503 "insert dev_stats item for device %s failed %d",
7504 btrfs_dev_name(device), ret);
7509 eb = path->nodes[0];
7510 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7511 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7512 btrfs_set_dev_stats_value(eb, ptr, i,
7513 btrfs_dev_stat_read(device, i));
7514 btrfs_mark_buffer_dirty(eb);
7517 btrfs_free_path(path);
7522 * called from commit_transaction. Writes all changed device stats to disk.
7524 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7526 struct btrfs_fs_info *fs_info = trans->fs_info;
7527 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7528 struct btrfs_device *device;
7532 mutex_lock(&fs_devices->device_list_mutex);
7533 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7534 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7535 if (!device->dev_stats_valid || stats_cnt == 0)
7540 * There is a LOAD-LOAD control dependency between the value of
7541 * dev_stats_ccnt and updating the on-disk values which requires
7542 * reading the in-memory counters. Such control dependencies
7543 * require explicit read memory barriers.
7545 * This memory barriers pairs with smp_mb__before_atomic in
7546 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7547 * barrier implied by atomic_xchg in
7548 * btrfs_dev_stats_read_and_reset
7552 ret = update_dev_stat_item(trans, device);
7554 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7556 mutex_unlock(&fs_devices->device_list_mutex);
7561 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7563 btrfs_dev_stat_inc(dev, index);
7565 if (!dev->dev_stats_valid)
7567 btrfs_err_rl_in_rcu(dev->fs_info,
7568 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7569 btrfs_dev_name(dev),
7570 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7571 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7572 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7573 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7574 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7577 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7581 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7582 if (btrfs_dev_stat_read(dev, i) != 0)
7584 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7585 return; /* all values == 0, suppress message */
7587 btrfs_info_in_rcu(dev->fs_info,
7588 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7589 btrfs_dev_name(dev),
7590 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7591 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7592 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7593 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7594 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7597 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7598 struct btrfs_ioctl_get_dev_stats *stats)
7600 BTRFS_DEV_LOOKUP_ARGS(args);
7601 struct btrfs_device *dev;
7602 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7605 mutex_lock(&fs_devices->device_list_mutex);
7606 args.devid = stats->devid;
7607 dev = btrfs_find_device(fs_info->fs_devices, &args);
7608 mutex_unlock(&fs_devices->device_list_mutex);
7611 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7613 } else if (!dev->dev_stats_valid) {
7614 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7616 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7617 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7618 if (stats->nr_items > i)
7620 btrfs_dev_stat_read_and_reset(dev, i);
7622 btrfs_dev_stat_set(dev, i, 0);
7624 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7625 current->comm, task_pid_nr(current));
7627 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7628 if (stats->nr_items > i)
7629 stats->values[i] = btrfs_dev_stat_read(dev, i);
7631 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7632 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7637 * Update the size and bytes used for each device where it changed. This is
7638 * delayed since we would otherwise get errors while writing out the
7641 * Must be invoked during transaction commit.
7643 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7645 struct btrfs_device *curr, *next;
7647 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7649 if (list_empty(&trans->dev_update_list))
7653 * We don't need the device_list_mutex here. This list is owned by the
7654 * transaction and the transaction must complete before the device is
7657 mutex_lock(&trans->fs_info->chunk_mutex);
7658 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7660 list_del_init(&curr->post_commit_list);
7661 curr->commit_total_bytes = curr->disk_total_bytes;
7662 curr->commit_bytes_used = curr->bytes_used;
7664 mutex_unlock(&trans->fs_info->chunk_mutex);
7668 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7670 int btrfs_bg_type_to_factor(u64 flags)
7672 const int index = btrfs_bg_flags_to_raid_index(flags);
7674 return btrfs_raid_array[index].ncopies;
7679 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7680 u64 chunk_offset, u64 devid,
7681 u64 physical_offset, u64 physical_len)
7683 struct btrfs_dev_lookup_args args = { .devid = devid };
7684 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7685 struct extent_map *em;
7686 struct map_lookup *map;
7687 struct btrfs_device *dev;
7693 read_lock(&em_tree->lock);
7694 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7695 read_unlock(&em_tree->lock);
7699 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7700 physical_offset, devid);
7705 map = em->map_lookup;
7706 stripe_len = btrfs_calc_stripe_length(em);
7707 if (physical_len != stripe_len) {
7709 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7710 physical_offset, devid, em->start, physical_len,
7717 * Very old mkfs.btrfs (before v4.1) will not respect the reserved
7718 * space. Although kernel can handle it without problem, better to warn
7721 if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED)
7723 "devid %llu physical %llu len %llu inside the reserved space",
7724 devid, physical_offset, physical_len);
7726 for (i = 0; i < map->num_stripes; i++) {
7727 if (map->stripes[i].dev->devid == devid &&
7728 map->stripes[i].physical == physical_offset) {
7730 if (map->verified_stripes >= map->num_stripes) {
7732 "too many dev extents for chunk %llu found",
7737 map->verified_stripes++;
7743 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7744 physical_offset, devid);
7748 /* Make sure no dev extent is beyond device boundary */
7749 dev = btrfs_find_device(fs_info->fs_devices, &args);
7751 btrfs_err(fs_info, "failed to find devid %llu", devid);
7756 if (physical_offset + physical_len > dev->disk_total_bytes) {
7758 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7759 devid, physical_offset, physical_len,
7760 dev->disk_total_bytes);
7765 if (dev->zone_info) {
7766 u64 zone_size = dev->zone_info->zone_size;
7768 if (!IS_ALIGNED(physical_offset, zone_size) ||
7769 !IS_ALIGNED(physical_len, zone_size)) {
7771 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
7772 devid, physical_offset, physical_len);
7779 free_extent_map(em);
7783 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7785 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7786 struct extent_map *em;
7787 struct rb_node *node;
7790 read_lock(&em_tree->lock);
7791 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
7792 em = rb_entry(node, struct extent_map, rb_node);
7793 if (em->map_lookup->num_stripes !=
7794 em->map_lookup->verified_stripes) {
7796 "chunk %llu has missing dev extent, have %d expect %d",
7797 em->start, em->map_lookup->verified_stripes,
7798 em->map_lookup->num_stripes);
7804 read_unlock(&em_tree->lock);
7809 * Ensure that all dev extents are mapped to correct chunk, otherwise
7810 * later chunk allocation/free would cause unexpected behavior.
7812 * NOTE: This will iterate through the whole device tree, which should be of
7813 * the same size level as the chunk tree. This slightly increases mount time.
7815 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
7817 struct btrfs_path *path;
7818 struct btrfs_root *root = fs_info->dev_root;
7819 struct btrfs_key key;
7821 u64 prev_dev_ext_end = 0;
7825 * We don't have a dev_root because we mounted with ignorebadroots and
7826 * failed to load the root, so we want to skip the verification in this
7829 * However if the dev root is fine, but the tree itself is corrupted
7830 * we'd still fail to mount. This verification is only to make sure
7831 * writes can happen safely, so instead just bypass this check
7832 * completely in the case of IGNOREBADROOTS.
7834 if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
7838 key.type = BTRFS_DEV_EXTENT_KEY;
7841 path = btrfs_alloc_path();
7845 path->reada = READA_FORWARD;
7846 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7850 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7851 ret = btrfs_next_leaf(root, path);
7854 /* No dev extents at all? Not good */
7861 struct extent_buffer *leaf = path->nodes[0];
7862 struct btrfs_dev_extent *dext;
7863 int slot = path->slots[0];
7865 u64 physical_offset;
7869 btrfs_item_key_to_cpu(leaf, &key, slot);
7870 if (key.type != BTRFS_DEV_EXTENT_KEY)
7872 devid = key.objectid;
7873 physical_offset = key.offset;
7875 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
7876 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
7877 physical_len = btrfs_dev_extent_length(leaf, dext);
7879 /* Check if this dev extent overlaps with the previous one */
7880 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
7882 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
7883 devid, physical_offset, prev_dev_ext_end);
7888 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
7889 physical_offset, physical_len);
7893 prev_dev_ext_end = physical_offset + physical_len;
7895 ret = btrfs_next_item(root, path);
7904 /* Ensure all chunks have corresponding dev extents */
7905 ret = verify_chunk_dev_extent_mapping(fs_info);
7907 btrfs_free_path(path);
7912 * Check whether the given block group or device is pinned by any inode being
7913 * used as a swapfile.
7915 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
7917 struct btrfs_swapfile_pin *sp;
7918 struct rb_node *node;
7920 spin_lock(&fs_info->swapfile_pins_lock);
7921 node = fs_info->swapfile_pins.rb_node;
7923 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
7925 node = node->rb_left;
7926 else if (ptr > sp->ptr)
7927 node = node->rb_right;
7931 spin_unlock(&fs_info->swapfile_pins_lock);
7932 return node != NULL;
7935 static int relocating_repair_kthread(void *data)
7937 struct btrfs_block_group *cache = data;
7938 struct btrfs_fs_info *fs_info = cache->fs_info;
7942 target = cache->start;
7943 btrfs_put_block_group(cache);
7945 sb_start_write(fs_info->sb);
7946 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
7948 "zoned: skip relocating block group %llu to repair: EBUSY",
7950 sb_end_write(fs_info->sb);
7954 mutex_lock(&fs_info->reclaim_bgs_lock);
7956 /* Ensure block group still exists */
7957 cache = btrfs_lookup_block_group(fs_info, target);
7961 if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags))
7964 ret = btrfs_may_alloc_data_chunk(fs_info, target);
7969 "zoned: relocating block group %llu to repair IO failure",
7971 ret = btrfs_relocate_chunk(fs_info, target);
7975 btrfs_put_block_group(cache);
7976 mutex_unlock(&fs_info->reclaim_bgs_lock);
7977 btrfs_exclop_finish(fs_info);
7978 sb_end_write(fs_info->sb);
7983 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical)
7985 struct btrfs_block_group *cache;
7987 if (!btrfs_is_zoned(fs_info))
7990 /* Do not attempt to repair in degraded state */
7991 if (btrfs_test_opt(fs_info, DEGRADED))
7994 cache = btrfs_lookup_block_group(fs_info, logical);
7998 if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) {
7999 btrfs_put_block_group(cache);
8003 kthread_run(relocating_repair_kthread, cache,
8004 "btrfs-relocating-repair");
8009 static void map_raid56_repair_block(struct btrfs_io_context *bioc,
8010 struct btrfs_io_stripe *smap,
8013 int data_stripes = nr_bioc_data_stripes(bioc);
8016 for (i = 0; i < data_stripes; i++) {
8017 u64 stripe_start = bioc->full_stripe_logical +
8018 btrfs_stripe_nr_to_offset(i);
8020 if (logical >= stripe_start &&
8021 logical < stripe_start + BTRFS_STRIPE_LEN)
8024 ASSERT(i < data_stripes);
8025 smap->dev = bioc->stripes[i].dev;
8026 smap->physical = bioc->stripes[i].physical +
8027 ((logical - bioc->full_stripe_logical) &
8028 BTRFS_STRIPE_LEN_MASK);
8032 * Map a repair write into a single device.
8034 * A repair write is triggered by read time repair or scrub, which would only
8035 * update the contents of a single device.
8036 * Not update any other mirrors nor go through RMW path.
8038 * Callers should ensure:
8040 * - Call btrfs_bio_counter_inc_blocked() first
8041 * - The range does not cross stripe boundary
8042 * - Has a valid @mirror_num passed in.
8044 int btrfs_map_repair_block(struct btrfs_fs_info *fs_info,
8045 struct btrfs_io_stripe *smap, u64 logical,
8046 u32 length, int mirror_num)
8048 struct btrfs_io_context *bioc = NULL;
8049 u64 map_length = length;
8050 int mirror_ret = mirror_num;
8053 ASSERT(mirror_num > 0);
8055 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, &map_length,
8056 &bioc, smap, &mirror_ret, true);
8060 /* The map range should not cross stripe boundary. */
8061 ASSERT(map_length >= length);
8063 /* Already mapped to single stripe. */
8067 /* Map the RAID56 multi-stripe writes to a single one. */
8068 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
8069 map_raid56_repair_block(bioc, smap, logical);
8073 ASSERT(mirror_num <= bioc->num_stripes);
8074 smap->dev = bioc->stripes[mirror_num - 1].dev;
8075 smap->physical = bioc->stripes[mirror_num - 1].physical;
8077 btrfs_put_bioc(bioc);