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 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
375 return ERR_PTR(-ENOMEM);
377 mutex_init(&fs_devs->device_list_mutex);
379 INIT_LIST_HEAD(&fs_devs->devices);
380 INIT_LIST_HEAD(&fs_devs->alloc_list);
381 INIT_LIST_HEAD(&fs_devs->fs_list);
382 INIT_LIST_HEAD(&fs_devs->seed_list);
384 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
387 memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE);
389 memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE);
394 void btrfs_free_device(struct btrfs_device *device)
396 WARN_ON(!list_empty(&device->post_commit_list));
397 rcu_string_free(device->name);
398 btrfs_destroy_dev_zone_info(device);
402 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
404 struct btrfs_device *device;
406 WARN_ON(fs_devices->opened);
407 while (!list_empty(&fs_devices->devices)) {
408 device = list_entry(fs_devices->devices.next,
409 struct btrfs_device, dev_list);
410 list_del(&device->dev_list);
411 btrfs_free_device(device);
416 void __exit btrfs_cleanup_fs_uuids(void)
418 struct btrfs_fs_devices *fs_devices;
420 while (!list_empty(&fs_uuids)) {
421 fs_devices = list_entry(fs_uuids.next,
422 struct btrfs_fs_devices, fs_list);
423 list_del(&fs_devices->fs_list);
424 free_fs_devices(fs_devices);
428 static noinline struct btrfs_fs_devices *find_fsid(
429 const u8 *fsid, const u8 *metadata_fsid)
431 struct btrfs_fs_devices *fs_devices;
435 /* Handle non-split brain cases */
436 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
438 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0
439 && memcmp(metadata_fsid, fs_devices->metadata_uuid,
440 BTRFS_FSID_SIZE) == 0)
443 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
450 static struct btrfs_fs_devices *find_fsid_with_metadata_uuid(
451 struct btrfs_super_block *disk_super)
454 struct btrfs_fs_devices *fs_devices;
457 * Handle scanned device having completed its fsid change but
458 * belonging to a fs_devices that was created by first scanning
459 * a device which didn't have its fsid/metadata_uuid changed
460 * at all and the CHANGING_FSID_V2 flag set.
462 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
463 if (fs_devices->fsid_change &&
464 memcmp(disk_super->metadata_uuid, fs_devices->fsid,
465 BTRFS_FSID_SIZE) == 0 &&
466 memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
467 BTRFS_FSID_SIZE) == 0) {
472 * Handle scanned device having completed its fsid change but
473 * belonging to a fs_devices that was created by a device that
474 * has an outdated pair of fsid/metadata_uuid and
475 * CHANGING_FSID_V2 flag set.
477 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
478 if (fs_devices->fsid_change &&
479 memcmp(fs_devices->metadata_uuid,
480 fs_devices->fsid, BTRFS_FSID_SIZE) != 0 &&
481 memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid,
482 BTRFS_FSID_SIZE) == 0) {
487 return find_fsid(disk_super->fsid, disk_super->metadata_uuid);
492 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
493 int flush, struct block_device **bdev,
494 struct btrfs_super_block **disk_super)
498 *bdev = blkdev_get_by_path(device_path, flags, holder);
501 ret = PTR_ERR(*bdev);
506 sync_blockdev(*bdev);
507 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
509 blkdev_put(*bdev, flags);
512 invalidate_bdev(*bdev);
513 *disk_super = btrfs_read_dev_super(*bdev);
514 if (IS_ERR(*disk_super)) {
515 ret = PTR_ERR(*disk_super);
516 blkdev_put(*bdev, flags);
528 * Search and remove all stale devices (which are not mounted). When both
529 * inputs are NULL, it will search and release all stale devices.
531 * @devt: Optional. When provided will it release all unmounted devices
532 * matching this devt only.
533 * @skip_device: Optional. Will skip this device when searching for the stale
536 * Return: 0 for success or if @devt is 0.
537 * -EBUSY if @devt is a mounted device.
538 * -ENOENT if @devt does not match any device in the list.
540 static int btrfs_free_stale_devices(dev_t devt, struct btrfs_device *skip_device)
542 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
543 struct btrfs_device *device, *tmp_device;
546 lockdep_assert_held(&uuid_mutex);
551 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
553 mutex_lock(&fs_devices->device_list_mutex);
554 list_for_each_entry_safe(device, tmp_device,
555 &fs_devices->devices, dev_list) {
556 if (skip_device && skip_device == device)
558 if (devt && devt != device->devt)
560 if (fs_devices->opened) {
561 /* for an already deleted device return 0 */
562 if (devt && ret != 0)
567 /* delete the stale device */
568 fs_devices->num_devices--;
569 list_del(&device->dev_list);
570 btrfs_free_device(device);
574 mutex_unlock(&fs_devices->device_list_mutex);
576 if (fs_devices->num_devices == 0) {
577 btrfs_sysfs_remove_fsid(fs_devices);
578 list_del(&fs_devices->fs_list);
579 free_fs_devices(fs_devices);
587 * This is only used on mount, and we are protected from competing things
588 * messing with our fs_devices by the uuid_mutex, thus we do not need the
589 * fs_devices->device_list_mutex here.
591 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
592 struct btrfs_device *device, fmode_t flags,
595 struct block_device *bdev;
596 struct btrfs_super_block *disk_super;
605 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
610 devid = btrfs_stack_device_id(&disk_super->dev_item);
611 if (devid != device->devid)
612 goto error_free_page;
614 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
615 goto error_free_page;
617 device->generation = btrfs_super_generation(disk_super);
619 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
620 if (btrfs_super_incompat_flags(disk_super) &
621 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
623 "BTRFS: Invalid seeding and uuid-changed device detected\n");
624 goto error_free_page;
627 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
628 fs_devices->seeding = true;
630 if (bdev_read_only(bdev))
631 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
633 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
636 if (!bdev_nonrot(bdev))
637 fs_devices->rotating = true;
639 if (bdev_max_discard_sectors(bdev))
640 fs_devices->discardable = true;
643 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
644 device->mode = flags;
646 fs_devices->open_devices++;
647 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
648 device->devid != BTRFS_DEV_REPLACE_DEVID) {
649 fs_devices->rw_devices++;
650 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
652 btrfs_release_disk_super(disk_super);
657 btrfs_release_disk_super(disk_super);
658 blkdev_put(bdev, flags);
664 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices
665 * being created with a disk that has already completed its fsid change. Such
666 * disk can belong to an fs which has its FSID changed or to one which doesn't.
667 * Handle both cases here.
669 static struct btrfs_fs_devices *find_fsid_inprogress(
670 struct btrfs_super_block *disk_super)
672 struct btrfs_fs_devices *fs_devices;
674 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
675 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
676 BTRFS_FSID_SIZE) != 0 &&
677 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
678 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) {
683 return find_fsid(disk_super->fsid, NULL);
687 static struct btrfs_fs_devices *find_fsid_changed(
688 struct btrfs_super_block *disk_super)
690 struct btrfs_fs_devices *fs_devices;
693 * Handles the case where scanned device is part of an fs that had
694 * multiple successful changes of FSID but currently device didn't
695 * observe it. Meaning our fsid will be different than theirs. We need
696 * to handle two subcases :
697 * 1 - The fs still continues to have different METADATA/FSID uuids.
698 * 2 - The fs is switched back to its original FSID (METADATA/FSID
701 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
703 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
704 BTRFS_FSID_SIZE) != 0 &&
705 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid,
706 BTRFS_FSID_SIZE) == 0 &&
707 memcmp(fs_devices->fsid, disk_super->fsid,
708 BTRFS_FSID_SIZE) != 0)
711 /* Unchanged UUIDs */
712 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
713 BTRFS_FSID_SIZE) == 0 &&
714 memcmp(fs_devices->fsid, disk_super->metadata_uuid,
715 BTRFS_FSID_SIZE) == 0)
722 static struct btrfs_fs_devices *find_fsid_reverted_metadata(
723 struct btrfs_super_block *disk_super)
725 struct btrfs_fs_devices *fs_devices;
728 * Handle the case where the scanned device is part of an fs whose last
729 * metadata UUID change reverted it to the original FSID. At the same
730 * time fs_devices was first created by another constituent device
731 * which didn't fully observe the operation. This results in an
732 * btrfs_fs_devices created with metadata/fsid different AND
733 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the
734 * fs_devices equal to the FSID of the disk.
736 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
737 if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
738 BTRFS_FSID_SIZE) != 0 &&
739 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
740 BTRFS_FSID_SIZE) == 0 &&
741 fs_devices->fsid_change)
748 * Add new device to list of registered devices
751 * device pointer which was just added or updated when successful
752 * error pointer when failed
754 static noinline struct btrfs_device *device_list_add(const char *path,
755 struct btrfs_super_block *disk_super,
756 bool *new_device_added)
758 struct btrfs_device *device;
759 struct btrfs_fs_devices *fs_devices = NULL;
760 struct rcu_string *name;
761 u64 found_transid = btrfs_super_generation(disk_super);
762 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
765 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
766 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
767 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) &
768 BTRFS_SUPER_FLAG_CHANGING_FSID_V2);
770 error = lookup_bdev(path, &path_devt);
772 btrfs_err(NULL, "failed to lookup block device for path %s: %d",
774 return ERR_PTR(error);
777 if (fsid_change_in_progress) {
778 if (!has_metadata_uuid)
779 fs_devices = find_fsid_inprogress(disk_super);
781 fs_devices = find_fsid_changed(disk_super);
782 } else if (has_metadata_uuid) {
783 fs_devices = find_fsid_with_metadata_uuid(disk_super);
785 fs_devices = find_fsid_reverted_metadata(disk_super);
787 fs_devices = find_fsid(disk_super->fsid, NULL);
792 if (has_metadata_uuid)
793 fs_devices = alloc_fs_devices(disk_super->fsid,
794 disk_super->metadata_uuid);
796 fs_devices = alloc_fs_devices(disk_super->fsid, NULL);
798 if (IS_ERR(fs_devices))
799 return ERR_CAST(fs_devices);
801 fs_devices->fsid_change = fsid_change_in_progress;
803 mutex_lock(&fs_devices->device_list_mutex);
804 list_add(&fs_devices->fs_list, &fs_uuids);
808 struct btrfs_dev_lookup_args args = {
810 .uuid = disk_super->dev_item.uuid,
813 mutex_lock(&fs_devices->device_list_mutex);
814 device = btrfs_find_device(fs_devices, &args);
817 * If this disk has been pulled into an fs devices created by
818 * a device which had the CHANGING_FSID_V2 flag then replace the
819 * metadata_uuid/fsid values of the fs_devices.
821 if (fs_devices->fsid_change &&
822 found_transid > fs_devices->latest_generation) {
823 memcpy(fs_devices->fsid, disk_super->fsid,
826 if (has_metadata_uuid)
827 memcpy(fs_devices->metadata_uuid,
828 disk_super->metadata_uuid,
831 memcpy(fs_devices->metadata_uuid,
832 disk_super->fsid, BTRFS_FSID_SIZE);
834 fs_devices->fsid_change = false;
839 unsigned int nofs_flag;
841 if (fs_devices->opened) {
843 "device %s belongs to fsid %pU, and the fs is already mounted",
844 path, fs_devices->fsid);
845 mutex_unlock(&fs_devices->device_list_mutex);
846 return ERR_PTR(-EBUSY);
849 nofs_flag = memalloc_nofs_save();
850 device = btrfs_alloc_device(NULL, &devid,
851 disk_super->dev_item.uuid, path);
852 memalloc_nofs_restore(nofs_flag);
853 if (IS_ERR(device)) {
854 mutex_unlock(&fs_devices->device_list_mutex);
855 /* we can safely leave the fs_devices entry around */
859 device->devt = path_devt;
861 list_add_rcu(&device->dev_list, &fs_devices->devices);
862 fs_devices->num_devices++;
864 device->fs_devices = fs_devices;
865 *new_device_added = true;
867 if (disk_super->label[0])
869 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n",
870 disk_super->label, devid, found_transid, path,
871 current->comm, task_pid_nr(current));
874 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n",
875 disk_super->fsid, devid, found_transid, path,
876 current->comm, task_pid_nr(current));
878 } else if (!device->name || strcmp(device->name->str, path)) {
880 * When FS is already mounted.
881 * 1. If you are here and if the device->name is NULL that
882 * means this device was missing at time of FS mount.
883 * 2. If you are here and if the device->name is different
884 * from 'path' that means either
885 * a. The same device disappeared and reappeared with
887 * b. The missing-disk-which-was-replaced, has
890 * We must allow 1 and 2a above. But 2b would be a spurious
893 * Further in case of 1 and 2a above, the disk at 'path'
894 * would have missed some transaction when it was away and
895 * in case of 2a the stale bdev has to be updated as well.
896 * 2b must not be allowed at all time.
900 * For now, we do allow update to btrfs_fs_device through the
901 * btrfs dev scan cli after FS has been mounted. We're still
902 * tracking a problem where systems fail mount by subvolume id
903 * when we reject replacement on a mounted FS.
905 if (!fs_devices->opened && found_transid < device->generation) {
907 * That is if the FS is _not_ mounted and if you
908 * are here, that means there is more than one
909 * disk with same uuid and devid.We keep the one
910 * with larger generation number or the last-in if
911 * generation are equal.
913 mutex_unlock(&fs_devices->device_list_mutex);
915 "device %s already registered with a higher generation, found %llu expect %llu",
916 path, found_transid, device->generation);
917 return ERR_PTR(-EEXIST);
921 * We are going to replace the device path for a given devid,
922 * make sure it's the same device if the device is mounted
924 * NOTE: the device->fs_info may not be reliable here so pass
925 * in a NULL to message helpers instead. This avoids a possible
926 * use-after-free when the fs_info and fs_info->sb are already
930 if (device->devt != path_devt) {
931 mutex_unlock(&fs_devices->device_list_mutex);
932 btrfs_warn_in_rcu(NULL,
933 "duplicate device %s devid %llu generation %llu scanned by %s (%d)",
934 path, devid, found_transid,
936 task_pid_nr(current));
937 return ERR_PTR(-EEXIST);
939 btrfs_info_in_rcu(NULL,
940 "devid %llu device path %s changed to %s scanned by %s (%d)",
941 devid, btrfs_dev_name(device),
943 task_pid_nr(current));
946 name = rcu_string_strdup(path, GFP_NOFS);
948 mutex_unlock(&fs_devices->device_list_mutex);
949 return ERR_PTR(-ENOMEM);
951 rcu_string_free(device->name);
952 rcu_assign_pointer(device->name, name);
953 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
954 fs_devices->missing_devices--;
955 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
957 device->devt = path_devt;
961 * Unmount does not free the btrfs_device struct but would zero
962 * generation along with most of the other members. So just update
963 * it back. We need it to pick the disk with largest generation
966 if (!fs_devices->opened) {
967 device->generation = found_transid;
968 fs_devices->latest_generation = max_t(u64, found_transid,
969 fs_devices->latest_generation);
972 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
974 mutex_unlock(&fs_devices->device_list_mutex);
978 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
980 struct btrfs_fs_devices *fs_devices;
981 struct btrfs_device *device;
982 struct btrfs_device *orig_dev;
985 lockdep_assert_held(&uuid_mutex);
987 fs_devices = alloc_fs_devices(orig->fsid, NULL);
988 if (IS_ERR(fs_devices))
991 fs_devices->total_devices = orig->total_devices;
993 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
994 const char *dev_path = NULL;
997 * This is ok to do without RCU read locked because we hold the
998 * uuid mutex so nothing we touch in here is going to disappear.
1001 dev_path = orig_dev->name->str;
1003 device = btrfs_alloc_device(NULL, &orig_dev->devid,
1004 orig_dev->uuid, dev_path);
1005 if (IS_ERR(device)) {
1006 ret = PTR_ERR(device);
1010 if (orig_dev->zone_info) {
1011 struct btrfs_zoned_device_info *zone_info;
1013 zone_info = btrfs_clone_dev_zone_info(orig_dev);
1015 btrfs_free_device(device);
1019 device->zone_info = zone_info;
1022 list_add(&device->dev_list, &fs_devices->devices);
1023 device->fs_devices = fs_devices;
1024 fs_devices->num_devices++;
1028 free_fs_devices(fs_devices);
1029 return ERR_PTR(ret);
1032 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices,
1033 struct btrfs_device **latest_dev)
1035 struct btrfs_device *device, *next;
1037 /* This is the initialized path, it is safe to release the devices. */
1038 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1039 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) {
1040 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1041 &device->dev_state) &&
1042 !test_bit(BTRFS_DEV_STATE_MISSING,
1043 &device->dev_state) &&
1045 device->generation > (*latest_dev)->generation)) {
1046 *latest_dev = device;
1052 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID,
1053 * in btrfs_init_dev_replace() so just continue.
1055 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1059 blkdev_put(device->bdev, device->mode);
1060 device->bdev = NULL;
1061 fs_devices->open_devices--;
1063 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1064 list_del_init(&device->dev_alloc_list);
1065 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1066 fs_devices->rw_devices--;
1068 list_del_init(&device->dev_list);
1069 fs_devices->num_devices--;
1070 btrfs_free_device(device);
1076 * After we have read the system tree and know devids belonging to this
1077 * filesystem, remove the device which does not belong there.
1079 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices)
1081 struct btrfs_device *latest_dev = NULL;
1082 struct btrfs_fs_devices *seed_dev;
1084 mutex_lock(&uuid_mutex);
1085 __btrfs_free_extra_devids(fs_devices, &latest_dev);
1087 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list)
1088 __btrfs_free_extra_devids(seed_dev, &latest_dev);
1090 fs_devices->latest_dev = latest_dev;
1092 mutex_unlock(&uuid_mutex);
1095 static void btrfs_close_bdev(struct btrfs_device *device)
1100 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1101 sync_blockdev(device->bdev);
1102 invalidate_bdev(device->bdev);
1105 blkdev_put(device->bdev, device->mode);
1108 static void btrfs_close_one_device(struct btrfs_device *device)
1110 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1112 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1113 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1114 list_del_init(&device->dev_alloc_list);
1115 fs_devices->rw_devices--;
1118 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1119 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
1121 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
1122 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
1123 fs_devices->missing_devices--;
1126 btrfs_close_bdev(device);
1128 fs_devices->open_devices--;
1129 device->bdev = NULL;
1131 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1132 btrfs_destroy_dev_zone_info(device);
1134 device->fs_info = NULL;
1135 atomic_set(&device->dev_stats_ccnt, 0);
1136 extent_io_tree_release(&device->alloc_state);
1139 * Reset the flush error record. We might have a transient flush error
1140 * in this mount, and if so we aborted the current transaction and set
1141 * the fs to an error state, guaranteeing no super blocks can be further
1142 * committed. However that error might be transient and if we unmount the
1143 * filesystem and mount it again, we should allow the mount to succeed
1144 * (btrfs_check_rw_degradable() should not fail) - if after mounting the
1145 * filesystem again we still get flush errors, then we will again abort
1146 * any transaction and set the error state, guaranteeing no commits of
1147 * unsafe super blocks.
1149 device->last_flush_error = 0;
1151 /* Verify the device is back in a pristine state */
1152 WARN_ON(test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1153 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1154 WARN_ON(!list_empty(&device->dev_alloc_list));
1155 WARN_ON(!list_empty(&device->post_commit_list));
1158 static void close_fs_devices(struct btrfs_fs_devices *fs_devices)
1160 struct btrfs_device *device, *tmp;
1162 lockdep_assert_held(&uuid_mutex);
1164 if (--fs_devices->opened > 0)
1167 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list)
1168 btrfs_close_one_device(device);
1170 WARN_ON(fs_devices->open_devices);
1171 WARN_ON(fs_devices->rw_devices);
1172 fs_devices->opened = 0;
1173 fs_devices->seeding = false;
1174 fs_devices->fs_info = NULL;
1177 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1180 struct btrfs_fs_devices *tmp;
1182 mutex_lock(&uuid_mutex);
1183 close_fs_devices(fs_devices);
1184 if (!fs_devices->opened) {
1185 list_splice_init(&fs_devices->seed_list, &list);
1188 * If the struct btrfs_fs_devices is not assembled with any
1189 * other device, it can be re-initialized during the next mount
1190 * without the needing device-scan step. Therefore, it can be
1193 if (fs_devices->num_devices == 1) {
1194 list_del(&fs_devices->fs_list);
1195 free_fs_devices(fs_devices);
1200 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {
1201 close_fs_devices(fs_devices);
1202 list_del(&fs_devices->seed_list);
1203 free_fs_devices(fs_devices);
1205 mutex_unlock(&uuid_mutex);
1208 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1209 fmode_t flags, void *holder)
1211 struct btrfs_device *device;
1212 struct btrfs_device *latest_dev = NULL;
1213 struct btrfs_device *tmp_device;
1215 flags |= FMODE_EXCL;
1217 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,
1221 ret = btrfs_open_one_device(fs_devices, device, flags, holder);
1223 (!latest_dev || device->generation > latest_dev->generation)) {
1224 latest_dev = device;
1225 } else if (ret == -ENODATA) {
1226 fs_devices->num_devices--;
1227 list_del(&device->dev_list);
1228 btrfs_free_device(device);
1231 if (fs_devices->open_devices == 0)
1234 fs_devices->opened = 1;
1235 fs_devices->latest_dev = latest_dev;
1236 fs_devices->total_rw_bytes = 0;
1237 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1238 fs_devices->read_policy = BTRFS_READ_POLICY_PID;
1243 static int devid_cmp(void *priv, const struct list_head *a,
1244 const struct list_head *b)
1246 const struct btrfs_device *dev1, *dev2;
1248 dev1 = list_entry(a, struct btrfs_device, dev_list);
1249 dev2 = list_entry(b, struct btrfs_device, dev_list);
1251 if (dev1->devid < dev2->devid)
1253 else if (dev1->devid > dev2->devid)
1258 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1259 fmode_t flags, void *holder)
1263 lockdep_assert_held(&uuid_mutex);
1265 * The device_list_mutex cannot be taken here in case opening the
1266 * underlying device takes further locks like open_mutex.
1268 * We also don't need the lock here as this is called during mount and
1269 * exclusion is provided by uuid_mutex
1272 if (fs_devices->opened) {
1273 fs_devices->opened++;
1276 list_sort(NULL, &fs_devices->devices, devid_cmp);
1277 ret = open_fs_devices(fs_devices, flags, holder);
1283 void btrfs_release_disk_super(struct btrfs_super_block *super)
1285 struct page *page = virt_to_page(super);
1290 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1291 u64 bytenr, u64 bytenr_orig)
1293 struct btrfs_super_block *disk_super;
1298 /* make sure our super fits in the device */
1299 if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev))
1300 return ERR_PTR(-EINVAL);
1302 /* make sure our super fits in the page */
1303 if (sizeof(*disk_super) > PAGE_SIZE)
1304 return ERR_PTR(-EINVAL);
1306 /* make sure our super doesn't straddle pages on disk */
1307 index = bytenr >> PAGE_SHIFT;
1308 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1309 return ERR_PTR(-EINVAL);
1311 /* pull in the page with our super */
1312 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1315 return ERR_CAST(page);
1317 p = page_address(page);
1319 /* align our pointer to the offset of the super block */
1320 disk_super = p + offset_in_page(bytenr);
1322 if (btrfs_super_bytenr(disk_super) != bytenr_orig ||
1323 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1324 btrfs_release_disk_super(p);
1325 return ERR_PTR(-EINVAL);
1328 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1329 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1334 int btrfs_forget_devices(dev_t devt)
1338 mutex_lock(&uuid_mutex);
1339 ret = btrfs_free_stale_devices(devt, NULL);
1340 mutex_unlock(&uuid_mutex);
1346 * Look for a btrfs signature on a device. This may be called out of the mount path
1347 * and we are not allowed to call set_blocksize during the scan. The superblock
1348 * is read via pagecache
1350 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1353 struct btrfs_super_block *disk_super;
1354 bool new_device_added = false;
1355 struct btrfs_device *device = NULL;
1356 struct block_device *bdev;
1357 u64 bytenr, bytenr_orig;
1360 lockdep_assert_held(&uuid_mutex);
1363 * we would like to check all the supers, but that would make
1364 * a btrfs mount succeed after a mkfs from a different FS.
1365 * So, we need to add a special mount option to scan for
1366 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1370 * Avoid using flag |= FMODE_EXCL here, as the systemd-udev may
1371 * initiate the device scan which may race with the user's mount
1372 * or mkfs command, resulting in failure.
1373 * Since the device scan is solely for reading purposes, there is
1374 * no need for FMODE_EXCL. Additionally, the devices are read again
1375 * during the mount process. It is ok to get some inconsistent
1376 * values temporarily, as the device paths of the fsid are the only
1377 * required information for assembling the volume.
1379 bdev = blkdev_get_by_path(path, flags, holder);
1381 return ERR_CAST(bdev);
1383 bytenr_orig = btrfs_sb_offset(0);
1384 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr);
1386 device = ERR_PTR(ret);
1387 goto error_bdev_put;
1390 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig);
1391 if (IS_ERR(disk_super)) {
1392 device = ERR_CAST(disk_super);
1393 goto error_bdev_put;
1396 device = device_list_add(path, disk_super, &new_device_added);
1397 if (!IS_ERR(device) && new_device_added)
1398 btrfs_free_stale_devices(device->devt, device);
1400 btrfs_release_disk_super(disk_super);
1403 blkdev_put(bdev, flags);
1409 * Try to find a chunk that intersects [start, start + len] range and when one
1410 * such is found, record the end of it in *start
1412 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1415 u64 physical_start, physical_end;
1417 lockdep_assert_held(&device->fs_info->chunk_mutex);
1419 if (!find_first_extent_bit(&device->alloc_state, *start,
1420 &physical_start, &physical_end,
1421 CHUNK_ALLOCATED, NULL)) {
1423 if (in_range(physical_start, *start, len) ||
1424 in_range(*start, physical_start,
1425 physical_end - physical_start)) {
1426 *start = physical_end + 1;
1433 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start)
1435 switch (device->fs_devices->chunk_alloc_policy) {
1436 case BTRFS_CHUNK_ALLOC_REGULAR:
1437 return max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED);
1438 case BTRFS_CHUNK_ALLOC_ZONED:
1440 * We don't care about the starting region like regular
1441 * allocator, because we anyway use/reserve the first two zones
1442 * for superblock logging.
1444 return ALIGN(start, device->zone_info->zone_size);
1450 static bool dev_extent_hole_check_zoned(struct btrfs_device *device,
1451 u64 *hole_start, u64 *hole_size,
1454 u64 zone_size = device->zone_info->zone_size;
1457 bool changed = false;
1459 ASSERT(IS_ALIGNED(*hole_start, zone_size));
1461 while (*hole_size > 0) {
1462 pos = btrfs_find_allocatable_zones(device, *hole_start,
1463 *hole_start + *hole_size,
1465 if (pos != *hole_start) {
1466 *hole_size = *hole_start + *hole_size - pos;
1469 if (*hole_size < num_bytes)
1473 ret = btrfs_ensure_empty_zones(device, pos, num_bytes);
1475 /* Range is ensured to be empty */
1479 /* Given hole range was invalid (outside of device) */
1480 if (ret == -ERANGE) {
1481 *hole_start += *hole_size;
1486 *hole_start += zone_size;
1487 *hole_size -= zone_size;
1495 * Check if specified hole is suitable for allocation.
1497 * @device: the device which we have the hole
1498 * @hole_start: starting position of the hole
1499 * @hole_size: the size of the hole
1500 * @num_bytes: the size of the free space that we need
1502 * This function may modify @hole_start and @hole_size to reflect the suitable
1503 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1505 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1506 u64 *hole_size, u64 num_bytes)
1508 bool changed = false;
1509 u64 hole_end = *hole_start + *hole_size;
1513 * Check before we set max_hole_start, otherwise we could end up
1514 * sending back this offset anyway.
1516 if (contains_pending_extent(device, hole_start, *hole_size)) {
1517 if (hole_end >= *hole_start)
1518 *hole_size = hole_end - *hole_start;
1524 switch (device->fs_devices->chunk_alloc_policy) {
1525 case BTRFS_CHUNK_ALLOC_REGULAR:
1526 /* No extra check */
1528 case BTRFS_CHUNK_ALLOC_ZONED:
1529 if (dev_extent_hole_check_zoned(device, hole_start,
1530 hole_size, num_bytes)) {
1533 * The changed hole can contain pending extent.
1534 * Loop again to check that.
1550 * Find free space in the specified device.
1552 * @device: the device which we search the free space in
1553 * @num_bytes: the size of the free space that we need
1554 * @search_start: the position from which to begin the search
1555 * @start: store the start of the free space.
1556 * @len: the size of the free space. that we find, or the size
1557 * of the max free space if we don't find suitable free space
1559 * This does a pretty simple search, the expectation is that it is called very
1560 * infrequently and that a given device has a small number of extents.
1562 * @start is used to store the start of the free space if we find. But if we
1563 * don't find suitable free space, it will be used to store the start position
1564 * of the max free space.
1566 * @len is used to store the size of the free space that we find.
1567 * But if we don't find suitable free space, it is used to store the size of
1568 * the max free space.
1570 * NOTE: This function will search *commit* root of device tree, and does extra
1571 * check to ensure dev extents are not double allocated.
1572 * This makes the function safe to allocate dev extents but may not report
1573 * correct usable device space, as device extent freed in current transaction
1574 * is not reported as available.
1576 static int find_free_dev_extent_start(struct btrfs_device *device,
1577 u64 num_bytes, u64 search_start, u64 *start,
1580 struct btrfs_fs_info *fs_info = device->fs_info;
1581 struct btrfs_root *root = fs_info->dev_root;
1582 struct btrfs_key key;
1583 struct btrfs_dev_extent *dev_extent;
1584 struct btrfs_path *path;
1589 u64 search_end = device->total_bytes;
1592 struct extent_buffer *l;
1594 search_start = dev_extent_search_start(device, search_start);
1596 WARN_ON(device->zone_info &&
1597 !IS_ALIGNED(num_bytes, device->zone_info->zone_size));
1599 path = btrfs_alloc_path();
1603 max_hole_start = search_start;
1607 if (search_start >= search_end ||
1608 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1613 path->reada = READA_FORWARD;
1614 path->search_commit_root = 1;
1615 path->skip_locking = 1;
1617 key.objectid = device->devid;
1618 key.offset = search_start;
1619 key.type = BTRFS_DEV_EXTENT_KEY;
1621 ret = btrfs_search_backwards(root, &key, path);
1625 while (search_start < search_end) {
1627 slot = path->slots[0];
1628 if (slot >= btrfs_header_nritems(l)) {
1629 ret = btrfs_next_leaf(root, path);
1637 btrfs_item_key_to_cpu(l, &key, slot);
1639 if (key.objectid < device->devid)
1642 if (key.objectid > device->devid)
1645 if (key.type != BTRFS_DEV_EXTENT_KEY)
1648 if (key.offset > search_end)
1651 if (key.offset > search_start) {
1652 hole_size = key.offset - search_start;
1653 dev_extent_hole_check(device, &search_start, &hole_size,
1656 if (hole_size > max_hole_size) {
1657 max_hole_start = search_start;
1658 max_hole_size = hole_size;
1662 * If this free space is greater than which we need,
1663 * it must be the max free space that we have found
1664 * until now, so max_hole_start must point to the start
1665 * of this free space and the length of this free space
1666 * is stored in max_hole_size. Thus, we return
1667 * max_hole_start and max_hole_size and go back to the
1670 if (hole_size >= num_bytes) {
1676 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1677 extent_end = key.offset + btrfs_dev_extent_length(l,
1679 if (extent_end > search_start)
1680 search_start = extent_end;
1687 * At this point, search_start should be the end of
1688 * allocated dev extents, and when shrinking the device,
1689 * search_end may be smaller than search_start.
1691 if (search_end > search_start) {
1692 hole_size = search_end - search_start;
1693 if (dev_extent_hole_check(device, &search_start, &hole_size,
1695 btrfs_release_path(path);
1699 if (hole_size > max_hole_size) {
1700 max_hole_start = search_start;
1701 max_hole_size = hole_size;
1706 if (max_hole_size < num_bytes)
1711 ASSERT(max_hole_start + max_hole_size <= search_end);
1713 btrfs_free_path(path);
1714 *start = max_hole_start;
1716 *len = max_hole_size;
1720 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1721 u64 *start, u64 *len)
1723 /* FIXME use last free of some kind */
1724 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1727 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1728 struct btrfs_device *device,
1729 u64 start, u64 *dev_extent_len)
1731 struct btrfs_fs_info *fs_info = device->fs_info;
1732 struct btrfs_root *root = fs_info->dev_root;
1734 struct btrfs_path *path;
1735 struct btrfs_key key;
1736 struct btrfs_key found_key;
1737 struct extent_buffer *leaf = NULL;
1738 struct btrfs_dev_extent *extent = NULL;
1740 path = btrfs_alloc_path();
1744 key.objectid = device->devid;
1746 key.type = BTRFS_DEV_EXTENT_KEY;
1748 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1750 ret = btrfs_previous_item(root, path, key.objectid,
1751 BTRFS_DEV_EXTENT_KEY);
1754 leaf = path->nodes[0];
1755 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1756 extent = btrfs_item_ptr(leaf, path->slots[0],
1757 struct btrfs_dev_extent);
1758 BUG_ON(found_key.offset > start || found_key.offset +
1759 btrfs_dev_extent_length(leaf, extent) < start);
1761 btrfs_release_path(path);
1763 } else if (ret == 0) {
1764 leaf = path->nodes[0];
1765 extent = btrfs_item_ptr(leaf, path->slots[0],
1766 struct btrfs_dev_extent);
1771 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1773 ret = btrfs_del_item(trans, root, path);
1775 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1777 btrfs_free_path(path);
1781 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1783 struct extent_map_tree *em_tree;
1784 struct extent_map *em;
1788 em_tree = &fs_info->mapping_tree;
1789 read_lock(&em_tree->lock);
1790 n = rb_last(&em_tree->map.rb_root);
1792 em = rb_entry(n, struct extent_map, rb_node);
1793 ret = em->start + em->len;
1795 read_unlock(&em_tree->lock);
1800 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1804 struct btrfs_key key;
1805 struct btrfs_key found_key;
1806 struct btrfs_path *path;
1808 path = btrfs_alloc_path();
1812 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1813 key.type = BTRFS_DEV_ITEM_KEY;
1814 key.offset = (u64)-1;
1816 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1822 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1827 ret = btrfs_previous_item(fs_info->chunk_root, path,
1828 BTRFS_DEV_ITEMS_OBJECTID,
1829 BTRFS_DEV_ITEM_KEY);
1833 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1835 *devid_ret = found_key.offset + 1;
1839 btrfs_free_path(path);
1844 * the device information is stored in the chunk root
1845 * the btrfs_device struct should be fully filled in
1847 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1848 struct btrfs_device *device)
1851 struct btrfs_path *path;
1852 struct btrfs_dev_item *dev_item;
1853 struct extent_buffer *leaf;
1854 struct btrfs_key key;
1857 path = btrfs_alloc_path();
1861 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1862 key.type = BTRFS_DEV_ITEM_KEY;
1863 key.offset = device->devid;
1865 btrfs_reserve_chunk_metadata(trans, true);
1866 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1867 &key, sizeof(*dev_item));
1868 btrfs_trans_release_chunk_metadata(trans);
1872 leaf = path->nodes[0];
1873 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1875 btrfs_set_device_id(leaf, dev_item, device->devid);
1876 btrfs_set_device_generation(leaf, dev_item, 0);
1877 btrfs_set_device_type(leaf, dev_item, device->type);
1878 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1879 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1880 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1881 btrfs_set_device_total_bytes(leaf, dev_item,
1882 btrfs_device_get_disk_total_bytes(device));
1883 btrfs_set_device_bytes_used(leaf, dev_item,
1884 btrfs_device_get_bytes_used(device));
1885 btrfs_set_device_group(leaf, dev_item, 0);
1886 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1887 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1888 btrfs_set_device_start_offset(leaf, dev_item, 0);
1890 ptr = btrfs_device_uuid(dev_item);
1891 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1892 ptr = btrfs_device_fsid(dev_item);
1893 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1894 ptr, BTRFS_FSID_SIZE);
1895 btrfs_mark_buffer_dirty(leaf);
1899 btrfs_free_path(path);
1904 * Function to update ctime/mtime for a given device path.
1905 * Mainly used for ctime/mtime based probe like libblkid.
1907 * We don't care about errors here, this is just to be kind to userspace.
1909 static void update_dev_time(const char *device_path)
1912 struct timespec64 now;
1915 ret = kern_path(device_path, LOOKUP_FOLLOW, &path);
1919 now = current_time(d_inode(path.dentry));
1920 inode_update_time(d_inode(path.dentry), &now, S_MTIME | S_CTIME);
1924 static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans,
1925 struct btrfs_device *device)
1927 struct btrfs_root *root = device->fs_info->chunk_root;
1929 struct btrfs_path *path;
1930 struct btrfs_key key;
1932 path = btrfs_alloc_path();
1936 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1937 key.type = BTRFS_DEV_ITEM_KEY;
1938 key.offset = device->devid;
1940 btrfs_reserve_chunk_metadata(trans, false);
1941 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1942 btrfs_trans_release_chunk_metadata(trans);
1949 ret = btrfs_del_item(trans, root, path);
1951 btrfs_free_path(path);
1956 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1957 * filesystem. It's up to the caller to adjust that number regarding eg. device
1960 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1968 seq = read_seqbegin(&fs_info->profiles_lock);
1970 all_avail = fs_info->avail_data_alloc_bits |
1971 fs_info->avail_system_alloc_bits |
1972 fs_info->avail_metadata_alloc_bits;
1973 } while (read_seqretry(&fs_info->profiles_lock, seq));
1975 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1976 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1979 if (num_devices < btrfs_raid_array[i].devs_min)
1980 return btrfs_raid_array[i].mindev_error;
1986 static struct btrfs_device * btrfs_find_next_active_device(
1987 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1989 struct btrfs_device *next_device;
1991 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1992 if (next_device != device &&
1993 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1994 && next_device->bdev)
2002 * Helper function to check if the given device is part of s_bdev / latest_dev
2003 * and replace it with the provided or the next active device, in the context
2004 * where this function called, there should be always be another device (or
2005 * this_dev) which is active.
2007 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
2008 struct btrfs_device *next_device)
2010 struct btrfs_fs_info *fs_info = device->fs_info;
2013 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
2015 ASSERT(next_device);
2017 if (fs_info->sb->s_bdev &&
2018 (fs_info->sb->s_bdev == device->bdev))
2019 fs_info->sb->s_bdev = next_device->bdev;
2021 if (fs_info->fs_devices->latest_dev->bdev == device->bdev)
2022 fs_info->fs_devices->latest_dev = next_device;
2026 * Return btrfs_fs_devices::num_devices excluding the device that's being
2027 * currently replaced.
2029 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2031 u64 num_devices = fs_info->fs_devices->num_devices;
2033 down_read(&fs_info->dev_replace.rwsem);
2034 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2035 ASSERT(num_devices > 1);
2038 up_read(&fs_info->dev_replace.rwsem);
2043 static void btrfs_scratch_superblock(struct btrfs_fs_info *fs_info,
2044 struct block_device *bdev, int copy_num)
2046 struct btrfs_super_block *disk_super;
2047 const size_t len = sizeof(disk_super->magic);
2048 const u64 bytenr = btrfs_sb_offset(copy_num);
2051 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr);
2052 if (IS_ERR(disk_super))
2055 memset(&disk_super->magic, 0, len);
2056 folio_mark_dirty(virt_to_folio(disk_super));
2057 btrfs_release_disk_super(disk_super);
2059 ret = sync_blockdev_range(bdev, bytenr, bytenr + len - 1);
2061 btrfs_warn(fs_info, "error clearing superblock number %d (%d)",
2065 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
2066 struct block_device *bdev,
2067 const char *device_path)
2074 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2075 if (bdev_is_zoned(bdev))
2076 btrfs_reset_sb_log_zones(bdev, copy_num);
2078 btrfs_scratch_superblock(fs_info, bdev, copy_num);
2081 /* Notify udev that device has changed */
2082 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2084 /* Update ctime/mtime for device path for libblkid */
2085 update_dev_time(device_path);
2088 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
2089 struct btrfs_dev_lookup_args *args,
2090 struct block_device **bdev, fmode_t *mode)
2092 struct btrfs_trans_handle *trans;
2093 struct btrfs_device *device;
2094 struct btrfs_fs_devices *cur_devices;
2095 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2099 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2100 btrfs_err(fs_info, "device remove not supported on extent tree v2 yet");
2105 * The device list in fs_devices is accessed without locks (neither
2106 * uuid_mutex nor device_list_mutex) as it won't change on a mounted
2107 * filesystem and another device rm cannot run.
2109 num_devices = btrfs_num_devices(fs_info);
2111 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2115 device = btrfs_find_device(fs_info->fs_devices, args);
2118 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2124 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2125 btrfs_warn_in_rcu(fs_info,
2126 "cannot remove device %s (devid %llu) due to active swapfile",
2127 btrfs_dev_name(device), device->devid);
2131 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
2132 return BTRFS_ERROR_DEV_TGT_REPLACE;
2134 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2135 fs_info->fs_devices->rw_devices == 1)
2136 return BTRFS_ERROR_DEV_ONLY_WRITABLE;
2138 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2139 mutex_lock(&fs_info->chunk_mutex);
2140 list_del_init(&device->dev_alloc_list);
2141 device->fs_devices->rw_devices--;
2142 mutex_unlock(&fs_info->chunk_mutex);
2145 ret = btrfs_shrink_device(device, 0);
2149 trans = btrfs_start_transaction(fs_info->chunk_root, 0);
2150 if (IS_ERR(trans)) {
2151 ret = PTR_ERR(trans);
2155 ret = btrfs_rm_dev_item(trans, device);
2157 /* Any error in dev item removal is critical */
2159 "failed to remove device item for devid %llu: %d",
2160 device->devid, ret);
2161 btrfs_abort_transaction(trans, ret);
2162 btrfs_end_transaction(trans);
2166 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2167 btrfs_scrub_cancel_dev(device);
2170 * the device list mutex makes sure that we don't change
2171 * the device list while someone else is writing out all
2172 * the device supers. Whoever is writing all supers, should
2173 * lock the device list mutex before getting the number of
2174 * devices in the super block (super_copy). Conversely,
2175 * whoever updates the number of devices in the super block
2176 * (super_copy) should hold the device list mutex.
2180 * In normal cases the cur_devices == fs_devices. But in case
2181 * of deleting a seed device, the cur_devices should point to
2182 * its own fs_devices listed under the fs_devices->seed_list.
2184 cur_devices = device->fs_devices;
2185 mutex_lock(&fs_devices->device_list_mutex);
2186 list_del_rcu(&device->dev_list);
2188 cur_devices->num_devices--;
2189 cur_devices->total_devices--;
2190 /* Update total_devices of the parent fs_devices if it's seed */
2191 if (cur_devices != fs_devices)
2192 fs_devices->total_devices--;
2194 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2195 cur_devices->missing_devices--;
2197 btrfs_assign_next_active_device(device, NULL);
2200 cur_devices->open_devices--;
2201 /* remove sysfs entry */
2202 btrfs_sysfs_remove_device(device);
2205 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2206 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2207 mutex_unlock(&fs_devices->device_list_mutex);
2210 * At this point, the device is zero sized and detached from the
2211 * devices list. All that's left is to zero out the old supers and
2214 * We cannot call btrfs_close_bdev() here because we're holding the sb
2215 * write lock, and blkdev_put() will pull in the ->open_mutex on the
2216 * block device and it's dependencies. Instead just flush the device
2217 * and let the caller do the final blkdev_put.
2219 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2220 btrfs_scratch_superblocks(fs_info, device->bdev,
2223 sync_blockdev(device->bdev);
2224 invalidate_bdev(device->bdev);
2228 *bdev = device->bdev;
2229 *mode = device->mode;
2231 btrfs_free_device(device);
2234 * This can happen if cur_devices is the private seed devices list. We
2235 * cannot call close_fs_devices() here because it expects the uuid_mutex
2236 * to be held, but in fact we don't need that for the private
2237 * seed_devices, we can simply decrement cur_devices->opened and then
2238 * remove it from our list and free the fs_devices.
2240 if (cur_devices->num_devices == 0) {
2241 list_del_init(&cur_devices->seed_list);
2242 ASSERT(cur_devices->opened == 1);
2243 cur_devices->opened--;
2244 free_fs_devices(cur_devices);
2247 ret = btrfs_commit_transaction(trans);
2252 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2253 mutex_lock(&fs_info->chunk_mutex);
2254 list_add(&device->dev_alloc_list,
2255 &fs_devices->alloc_list);
2256 device->fs_devices->rw_devices++;
2257 mutex_unlock(&fs_info->chunk_mutex);
2262 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2264 struct btrfs_fs_devices *fs_devices;
2266 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2269 * in case of fs with no seed, srcdev->fs_devices will point
2270 * to fs_devices of fs_info. However when the dev being replaced is
2271 * a seed dev it will point to the seed's local fs_devices. In short
2272 * srcdev will have its correct fs_devices in both the cases.
2274 fs_devices = srcdev->fs_devices;
2276 list_del_rcu(&srcdev->dev_list);
2277 list_del(&srcdev->dev_alloc_list);
2278 fs_devices->num_devices--;
2279 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2280 fs_devices->missing_devices--;
2282 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2283 fs_devices->rw_devices--;
2286 fs_devices->open_devices--;
2289 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2291 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2293 mutex_lock(&uuid_mutex);
2295 btrfs_close_bdev(srcdev);
2297 btrfs_free_device(srcdev);
2299 /* if this is no devs we rather delete the fs_devices */
2300 if (!fs_devices->num_devices) {
2302 * On a mounted FS, num_devices can't be zero unless it's a
2303 * seed. In case of a seed device being replaced, the replace
2304 * target added to the sprout FS, so there will be no more
2305 * device left under the seed FS.
2307 ASSERT(fs_devices->seeding);
2309 list_del_init(&fs_devices->seed_list);
2310 close_fs_devices(fs_devices);
2311 free_fs_devices(fs_devices);
2313 mutex_unlock(&uuid_mutex);
2316 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2318 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2320 mutex_lock(&fs_devices->device_list_mutex);
2322 btrfs_sysfs_remove_device(tgtdev);
2325 fs_devices->open_devices--;
2327 fs_devices->num_devices--;
2329 btrfs_assign_next_active_device(tgtdev, NULL);
2331 list_del_rcu(&tgtdev->dev_list);
2333 mutex_unlock(&fs_devices->device_list_mutex);
2335 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2338 btrfs_close_bdev(tgtdev);
2340 btrfs_free_device(tgtdev);
2344 * Populate args from device at path.
2346 * @fs_info: the filesystem
2347 * @args: the args to populate
2348 * @path: the path to the device
2350 * This will read the super block of the device at @path and populate @args with
2351 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to
2352 * lookup a device to operate on, but need to do it before we take any locks.
2353 * This properly handles the special case of "missing" that a user may pass in,
2354 * and does some basic sanity checks. The caller must make sure that @path is
2355 * properly NUL terminated before calling in, and must call
2356 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and
2359 * Return: 0 for success, -errno for failure
2361 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,
2362 struct btrfs_dev_lookup_args *args,
2365 struct btrfs_super_block *disk_super;
2366 struct block_device *bdev;
2369 if (!path || !path[0])
2371 if (!strcmp(path, "missing")) {
2372 args->missing = true;
2376 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL);
2377 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL);
2378 if (!args->uuid || !args->fsid) {
2379 btrfs_put_dev_args_from_path(args);
2383 ret = btrfs_get_bdev_and_sb(path, FMODE_READ, fs_info->bdev_holder, 0,
2384 &bdev, &disk_super);
2386 btrfs_put_dev_args_from_path(args);
2390 args->devid = btrfs_stack_device_id(&disk_super->dev_item);
2391 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE);
2392 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2393 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE);
2395 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
2396 btrfs_release_disk_super(disk_super);
2397 blkdev_put(bdev, FMODE_READ);
2402 * Only use this jointly with btrfs_get_dev_args_from_path() because we will
2403 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables
2404 * that don't need to be freed.
2406 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args)
2414 struct btrfs_device *btrfs_find_device_by_devspec(
2415 struct btrfs_fs_info *fs_info, u64 devid,
2416 const char *device_path)
2418 BTRFS_DEV_LOOKUP_ARGS(args);
2419 struct btrfs_device *device;
2424 device = btrfs_find_device(fs_info->fs_devices, &args);
2426 return ERR_PTR(-ENOENT);
2430 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path);
2432 return ERR_PTR(ret);
2433 device = btrfs_find_device(fs_info->fs_devices, &args);
2434 btrfs_put_dev_args_from_path(&args);
2436 return ERR_PTR(-ENOENT);
2440 static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info)
2442 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2443 struct btrfs_fs_devices *old_devices;
2444 struct btrfs_fs_devices *seed_devices;
2446 lockdep_assert_held(&uuid_mutex);
2447 if (!fs_devices->seeding)
2448 return ERR_PTR(-EINVAL);
2451 * Private copy of the seed devices, anchored at
2452 * fs_info->fs_devices->seed_list
2454 seed_devices = alloc_fs_devices(NULL, NULL);
2455 if (IS_ERR(seed_devices))
2456 return seed_devices;
2459 * It's necessary to retain a copy of the original seed fs_devices in
2460 * fs_uuids so that filesystems which have been seeded can successfully
2461 * reference the seed device from open_seed_devices. This also supports
2464 old_devices = clone_fs_devices(fs_devices);
2465 if (IS_ERR(old_devices)) {
2466 kfree(seed_devices);
2470 list_add(&old_devices->fs_list, &fs_uuids);
2472 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2473 seed_devices->opened = 1;
2474 INIT_LIST_HEAD(&seed_devices->devices);
2475 INIT_LIST_HEAD(&seed_devices->alloc_list);
2476 mutex_init(&seed_devices->device_list_mutex);
2478 return seed_devices;
2482 * Splice seed devices into the sprout fs_devices.
2483 * Generate a new fsid for the sprouted read-write filesystem.
2485 static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info,
2486 struct btrfs_fs_devices *seed_devices)
2488 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2489 struct btrfs_super_block *disk_super = fs_info->super_copy;
2490 struct btrfs_device *device;
2494 * We are updating the fsid, the thread leading to device_list_add()
2495 * could race, so uuid_mutex is needed.
2497 lockdep_assert_held(&uuid_mutex);
2500 * The threads listed below may traverse dev_list but can do that without
2501 * device_list_mutex:
2502 * - All device ops and balance - as we are in btrfs_exclop_start.
2503 * - Various dev_list readers - are using RCU.
2504 * - btrfs_ioctl_fitrim() - is using RCU.
2506 * For-read threads as below are using device_list_mutex:
2507 * - Readonly scrub btrfs_scrub_dev()
2508 * - Readonly scrub btrfs_scrub_progress()
2509 * - btrfs_get_dev_stats()
2511 lockdep_assert_held(&fs_devices->device_list_mutex);
2513 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2515 list_for_each_entry(device, &seed_devices->devices, dev_list)
2516 device->fs_devices = seed_devices;
2518 fs_devices->seeding = false;
2519 fs_devices->num_devices = 0;
2520 fs_devices->open_devices = 0;
2521 fs_devices->missing_devices = 0;
2522 fs_devices->rotating = false;
2523 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2525 generate_random_uuid(fs_devices->fsid);
2526 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2527 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2529 super_flags = btrfs_super_flags(disk_super) &
2530 ~BTRFS_SUPER_FLAG_SEEDING;
2531 btrfs_set_super_flags(disk_super, super_flags);
2535 * Store the expected generation for seed devices in device items.
2537 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2539 BTRFS_DEV_LOOKUP_ARGS(args);
2540 struct btrfs_fs_info *fs_info = trans->fs_info;
2541 struct btrfs_root *root = fs_info->chunk_root;
2542 struct btrfs_path *path;
2543 struct extent_buffer *leaf;
2544 struct btrfs_dev_item *dev_item;
2545 struct btrfs_device *device;
2546 struct btrfs_key key;
2547 u8 fs_uuid[BTRFS_FSID_SIZE];
2548 u8 dev_uuid[BTRFS_UUID_SIZE];
2551 path = btrfs_alloc_path();
2555 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2557 key.type = BTRFS_DEV_ITEM_KEY;
2560 btrfs_reserve_chunk_metadata(trans, false);
2561 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2562 btrfs_trans_release_chunk_metadata(trans);
2566 leaf = path->nodes[0];
2568 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2569 ret = btrfs_next_leaf(root, path);
2574 leaf = path->nodes[0];
2575 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2576 btrfs_release_path(path);
2580 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2581 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2582 key.type != BTRFS_DEV_ITEM_KEY)
2585 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2586 struct btrfs_dev_item);
2587 args.devid = btrfs_device_id(leaf, dev_item);
2588 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2590 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2592 args.uuid = dev_uuid;
2593 args.fsid = fs_uuid;
2594 device = btrfs_find_device(fs_info->fs_devices, &args);
2595 BUG_ON(!device); /* Logic error */
2597 if (device->fs_devices->seeding) {
2598 btrfs_set_device_generation(leaf, dev_item,
2599 device->generation);
2600 btrfs_mark_buffer_dirty(leaf);
2608 btrfs_free_path(path);
2612 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2614 struct btrfs_root *root = fs_info->dev_root;
2615 struct btrfs_trans_handle *trans;
2616 struct btrfs_device *device;
2617 struct block_device *bdev;
2618 struct super_block *sb = fs_info->sb;
2619 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2620 struct btrfs_fs_devices *seed_devices = NULL;
2621 u64 orig_super_total_bytes;
2622 u64 orig_super_num_devices;
2624 bool seeding_dev = false;
2625 bool locked = false;
2627 if (sb_rdonly(sb) && !fs_devices->seeding)
2630 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2631 fs_info->bdev_holder);
2633 return PTR_ERR(bdev);
2635 if (!btrfs_check_device_zone_type(fs_info, bdev)) {
2640 if (fs_devices->seeding) {
2642 down_write(&sb->s_umount);
2643 mutex_lock(&uuid_mutex);
2647 sync_blockdev(bdev);
2650 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2651 if (device->bdev == bdev) {
2659 device = btrfs_alloc_device(fs_info, NULL, NULL, device_path);
2660 if (IS_ERR(device)) {
2661 /* we can safely leave the fs_devices entry around */
2662 ret = PTR_ERR(device);
2666 device->fs_info = fs_info;
2667 device->bdev = bdev;
2668 ret = lookup_bdev(device_path, &device->devt);
2670 goto error_free_device;
2672 ret = btrfs_get_dev_zone_info(device, false);
2674 goto error_free_device;
2676 trans = btrfs_start_transaction(root, 0);
2677 if (IS_ERR(trans)) {
2678 ret = PTR_ERR(trans);
2679 goto error_free_zone;
2682 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2683 device->generation = trans->transid;
2684 device->io_width = fs_info->sectorsize;
2685 device->io_align = fs_info->sectorsize;
2686 device->sector_size = fs_info->sectorsize;
2687 device->total_bytes =
2688 round_down(bdev_nr_bytes(bdev), fs_info->sectorsize);
2689 device->disk_total_bytes = device->total_bytes;
2690 device->commit_total_bytes = device->total_bytes;
2691 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2692 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2693 device->mode = FMODE_EXCL;
2694 device->dev_stats_valid = 1;
2695 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2698 btrfs_clear_sb_rdonly(sb);
2700 /* GFP_KERNEL allocation must not be under device_list_mutex */
2701 seed_devices = btrfs_init_sprout(fs_info);
2702 if (IS_ERR(seed_devices)) {
2703 ret = PTR_ERR(seed_devices);
2704 btrfs_abort_transaction(trans, ret);
2709 mutex_lock(&fs_devices->device_list_mutex);
2711 btrfs_setup_sprout(fs_info, seed_devices);
2712 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev,
2716 device->fs_devices = fs_devices;
2718 mutex_lock(&fs_info->chunk_mutex);
2719 list_add_rcu(&device->dev_list, &fs_devices->devices);
2720 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2721 fs_devices->num_devices++;
2722 fs_devices->open_devices++;
2723 fs_devices->rw_devices++;
2724 fs_devices->total_devices++;
2725 fs_devices->total_rw_bytes += device->total_bytes;
2727 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2729 if (!bdev_nonrot(bdev))
2730 fs_devices->rotating = true;
2732 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2733 btrfs_set_super_total_bytes(fs_info->super_copy,
2734 round_down(orig_super_total_bytes + device->total_bytes,
2735 fs_info->sectorsize));
2737 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2738 btrfs_set_super_num_devices(fs_info->super_copy,
2739 orig_super_num_devices + 1);
2742 * we've got more storage, clear any full flags on the space
2745 btrfs_clear_space_info_full(fs_info);
2747 mutex_unlock(&fs_info->chunk_mutex);
2749 /* Add sysfs device entry */
2750 btrfs_sysfs_add_device(device);
2752 mutex_unlock(&fs_devices->device_list_mutex);
2755 mutex_lock(&fs_info->chunk_mutex);
2756 ret = init_first_rw_device(trans);
2757 mutex_unlock(&fs_info->chunk_mutex);
2759 btrfs_abort_transaction(trans, ret);
2764 ret = btrfs_add_dev_item(trans, device);
2766 btrfs_abort_transaction(trans, ret);
2771 ret = btrfs_finish_sprout(trans);
2773 btrfs_abort_transaction(trans, ret);
2778 * fs_devices now represents the newly sprouted filesystem and
2779 * its fsid has been changed by btrfs_sprout_splice().
2781 btrfs_sysfs_update_sprout_fsid(fs_devices);
2784 ret = btrfs_commit_transaction(trans);
2787 mutex_unlock(&uuid_mutex);
2788 up_write(&sb->s_umount);
2791 if (ret) /* transaction commit */
2794 ret = btrfs_relocate_sys_chunks(fs_info);
2796 btrfs_handle_fs_error(fs_info, ret,
2797 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2798 trans = btrfs_attach_transaction(root);
2799 if (IS_ERR(trans)) {
2800 if (PTR_ERR(trans) == -ENOENT)
2802 ret = PTR_ERR(trans);
2806 ret = btrfs_commit_transaction(trans);
2810 * Now that we have written a new super block to this device, check all
2811 * other fs_devices list if device_path alienates any other scanned
2813 * We can ignore the return value as it typically returns -EINVAL and
2814 * only succeeds if the device was an alien.
2816 btrfs_forget_devices(device->devt);
2818 /* Update ctime/mtime for blkid or udev */
2819 update_dev_time(device_path);
2824 btrfs_sysfs_remove_device(device);
2825 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2826 mutex_lock(&fs_info->chunk_mutex);
2827 list_del_rcu(&device->dev_list);
2828 list_del(&device->dev_alloc_list);
2829 fs_info->fs_devices->num_devices--;
2830 fs_info->fs_devices->open_devices--;
2831 fs_info->fs_devices->rw_devices--;
2832 fs_info->fs_devices->total_devices--;
2833 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2834 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2835 btrfs_set_super_total_bytes(fs_info->super_copy,
2836 orig_super_total_bytes);
2837 btrfs_set_super_num_devices(fs_info->super_copy,
2838 orig_super_num_devices);
2839 mutex_unlock(&fs_info->chunk_mutex);
2840 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2843 btrfs_set_sb_rdonly(sb);
2845 btrfs_end_transaction(trans);
2847 btrfs_destroy_dev_zone_info(device);
2849 btrfs_free_device(device);
2851 blkdev_put(bdev, FMODE_EXCL);
2853 mutex_unlock(&uuid_mutex);
2854 up_write(&sb->s_umount);
2859 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2860 struct btrfs_device *device)
2863 struct btrfs_path *path;
2864 struct btrfs_root *root = device->fs_info->chunk_root;
2865 struct btrfs_dev_item *dev_item;
2866 struct extent_buffer *leaf;
2867 struct btrfs_key key;
2869 path = btrfs_alloc_path();
2873 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2874 key.type = BTRFS_DEV_ITEM_KEY;
2875 key.offset = device->devid;
2877 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2886 leaf = path->nodes[0];
2887 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2889 btrfs_set_device_id(leaf, dev_item, device->devid);
2890 btrfs_set_device_type(leaf, dev_item, device->type);
2891 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2892 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2893 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2894 btrfs_set_device_total_bytes(leaf, dev_item,
2895 btrfs_device_get_disk_total_bytes(device));
2896 btrfs_set_device_bytes_used(leaf, dev_item,
2897 btrfs_device_get_bytes_used(device));
2898 btrfs_mark_buffer_dirty(leaf);
2901 btrfs_free_path(path);
2905 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2906 struct btrfs_device *device, u64 new_size)
2908 struct btrfs_fs_info *fs_info = device->fs_info;
2909 struct btrfs_super_block *super_copy = fs_info->super_copy;
2914 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2917 new_size = round_down(new_size, fs_info->sectorsize);
2919 mutex_lock(&fs_info->chunk_mutex);
2920 old_total = btrfs_super_total_bytes(super_copy);
2921 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2923 if (new_size <= device->total_bytes ||
2924 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2925 mutex_unlock(&fs_info->chunk_mutex);
2929 btrfs_set_super_total_bytes(super_copy,
2930 round_down(old_total + diff, fs_info->sectorsize));
2931 device->fs_devices->total_rw_bytes += diff;
2933 btrfs_device_set_total_bytes(device, new_size);
2934 btrfs_device_set_disk_total_bytes(device, new_size);
2935 btrfs_clear_space_info_full(device->fs_info);
2936 if (list_empty(&device->post_commit_list))
2937 list_add_tail(&device->post_commit_list,
2938 &trans->transaction->dev_update_list);
2939 mutex_unlock(&fs_info->chunk_mutex);
2941 btrfs_reserve_chunk_metadata(trans, false);
2942 ret = btrfs_update_device(trans, device);
2943 btrfs_trans_release_chunk_metadata(trans);
2948 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2950 struct btrfs_fs_info *fs_info = trans->fs_info;
2951 struct btrfs_root *root = fs_info->chunk_root;
2953 struct btrfs_path *path;
2954 struct btrfs_key key;
2956 path = btrfs_alloc_path();
2960 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2961 key.offset = chunk_offset;
2962 key.type = BTRFS_CHUNK_ITEM_KEY;
2964 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2967 else if (ret > 0) { /* Logic error or corruption */
2968 btrfs_handle_fs_error(fs_info, -ENOENT,
2969 "Failed lookup while freeing chunk.");
2974 ret = btrfs_del_item(trans, root, path);
2976 btrfs_handle_fs_error(fs_info, ret,
2977 "Failed to delete chunk item.");
2979 btrfs_free_path(path);
2983 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2985 struct btrfs_super_block *super_copy = fs_info->super_copy;
2986 struct btrfs_disk_key *disk_key;
2987 struct btrfs_chunk *chunk;
2994 struct btrfs_key key;
2996 lockdep_assert_held(&fs_info->chunk_mutex);
2997 array_size = btrfs_super_sys_array_size(super_copy);
2999 ptr = super_copy->sys_chunk_array;
3002 while (cur < array_size) {
3003 disk_key = (struct btrfs_disk_key *)ptr;
3004 btrfs_disk_key_to_cpu(&key, disk_key);
3006 len = sizeof(*disk_key);
3008 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3009 chunk = (struct btrfs_chunk *)(ptr + len);
3010 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
3011 len += btrfs_chunk_item_size(num_stripes);
3016 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
3017 key.offset == chunk_offset) {
3018 memmove(ptr, ptr + len, array_size - (cur + len));
3020 btrfs_set_super_sys_array_size(super_copy, array_size);
3030 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
3031 * @logical: Logical block offset in bytes.
3032 * @length: Length of extent in bytes.
3034 * Return: Chunk mapping or ERR_PTR.
3036 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
3037 u64 logical, u64 length)
3039 struct extent_map_tree *em_tree;
3040 struct extent_map *em;
3042 em_tree = &fs_info->mapping_tree;
3043 read_lock(&em_tree->lock);
3044 em = lookup_extent_mapping(em_tree, logical, length);
3045 read_unlock(&em_tree->lock);
3048 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
3050 return ERR_PTR(-EINVAL);
3053 if (em->start > logical || em->start + em->len < logical) {
3055 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
3056 logical, length, em->start, em->start + em->len);
3057 free_extent_map(em);
3058 return ERR_PTR(-EINVAL);
3061 /* callers are responsible for dropping em's ref. */
3065 static int remove_chunk_item(struct btrfs_trans_handle *trans,
3066 struct map_lookup *map, u64 chunk_offset)
3071 * Removing chunk items and updating the device items in the chunks btree
3072 * requires holding the chunk_mutex.
3073 * See the comment at btrfs_chunk_alloc() for the details.
3075 lockdep_assert_held(&trans->fs_info->chunk_mutex);
3077 for (i = 0; i < map->num_stripes; i++) {
3080 ret = btrfs_update_device(trans, map->stripes[i].dev);
3085 return btrfs_free_chunk(trans, chunk_offset);
3088 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
3090 struct btrfs_fs_info *fs_info = trans->fs_info;
3091 struct extent_map *em;
3092 struct map_lookup *map;
3093 u64 dev_extent_len = 0;
3095 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3097 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
3100 * This is a logic error, but we don't want to just rely on the
3101 * user having built with ASSERT enabled, so if ASSERT doesn't
3102 * do anything we still error out.
3107 map = em->map_lookup;
3110 * First delete the device extent items from the devices btree.
3111 * We take the device_list_mutex to avoid racing with the finishing phase
3112 * of a device replace operation. See the comment below before acquiring
3113 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
3114 * because that can result in a deadlock when deleting the device extent
3115 * items from the devices btree - COWing an extent buffer from the btree
3116 * may result in allocating a new metadata chunk, which would attempt to
3117 * lock again fs_info->chunk_mutex.
3119 mutex_lock(&fs_devices->device_list_mutex);
3120 for (i = 0; i < map->num_stripes; i++) {
3121 struct btrfs_device *device = map->stripes[i].dev;
3122 ret = btrfs_free_dev_extent(trans, device,
3123 map->stripes[i].physical,
3126 mutex_unlock(&fs_devices->device_list_mutex);
3127 btrfs_abort_transaction(trans, ret);
3131 if (device->bytes_used > 0) {
3132 mutex_lock(&fs_info->chunk_mutex);
3133 btrfs_device_set_bytes_used(device,
3134 device->bytes_used - dev_extent_len);
3135 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3136 btrfs_clear_space_info_full(fs_info);
3137 mutex_unlock(&fs_info->chunk_mutex);
3140 mutex_unlock(&fs_devices->device_list_mutex);
3143 * We acquire fs_info->chunk_mutex for 2 reasons:
3145 * 1) Just like with the first phase of the chunk allocation, we must
3146 * reserve system space, do all chunk btree updates and deletions, and
3147 * update the system chunk array in the superblock while holding this
3148 * mutex. This is for similar reasons as explained on the comment at
3149 * the top of btrfs_chunk_alloc();
3151 * 2) Prevent races with the final phase of a device replace operation
3152 * that replaces the device object associated with the map's stripes,
3153 * because the device object's id can change at any time during that
3154 * final phase of the device replace operation
3155 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3156 * replaced device and then see it with an ID of
3157 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3158 * the device item, which does not exists on the chunk btree.
3159 * The finishing phase of device replace acquires both the
3160 * device_list_mutex and the chunk_mutex, in that order, so we are
3161 * safe by just acquiring the chunk_mutex.
3163 trans->removing_chunk = true;
3164 mutex_lock(&fs_info->chunk_mutex);
3166 check_system_chunk(trans, map->type);
3168 ret = remove_chunk_item(trans, map, chunk_offset);
3170 * Normally we should not get -ENOSPC since we reserved space before
3171 * through the call to check_system_chunk().
3173 * Despite our system space_info having enough free space, we may not
3174 * be able to allocate extents from its block groups, because all have
3175 * an incompatible profile, which will force us to allocate a new system
3176 * block group with the right profile, or right after we called
3177 * check_system_space() above, a scrub turned the only system block group
3178 * with enough free space into RO mode.
3179 * This is explained with more detail at do_chunk_alloc().
3181 * So if we get -ENOSPC, allocate a new system chunk and retry once.
3183 if (ret == -ENOSPC) {
3184 const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
3185 struct btrfs_block_group *sys_bg;
3187 sys_bg = btrfs_create_chunk(trans, sys_flags);
3188 if (IS_ERR(sys_bg)) {
3189 ret = PTR_ERR(sys_bg);
3190 btrfs_abort_transaction(trans, ret);
3194 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3196 btrfs_abort_transaction(trans, ret);
3200 ret = remove_chunk_item(trans, map, chunk_offset);
3202 btrfs_abort_transaction(trans, ret);
3206 btrfs_abort_transaction(trans, ret);
3210 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3212 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3213 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3215 btrfs_abort_transaction(trans, ret);
3220 mutex_unlock(&fs_info->chunk_mutex);
3221 trans->removing_chunk = false;
3224 * We are done with chunk btree updates and deletions, so release the
3225 * system space we previously reserved (with check_system_chunk()).
3227 btrfs_trans_release_chunk_metadata(trans);
3229 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3231 btrfs_abort_transaction(trans, ret);
3236 if (trans->removing_chunk) {
3237 mutex_unlock(&fs_info->chunk_mutex);
3238 trans->removing_chunk = false;
3241 free_extent_map(em);
3245 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3247 struct btrfs_root *root = fs_info->chunk_root;
3248 struct btrfs_trans_handle *trans;
3249 struct btrfs_block_group *block_group;
3253 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3255 "relocate: not supported on extent tree v2 yet");
3260 * Prevent races with automatic removal of unused block groups.
3261 * After we relocate and before we remove the chunk with offset
3262 * chunk_offset, automatic removal of the block group can kick in,
3263 * resulting in a failure when calling btrfs_remove_chunk() below.
3265 * Make sure to acquire this mutex before doing a tree search (dev
3266 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3267 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3268 * we release the path used to search the chunk/dev tree and before
3269 * the current task acquires this mutex and calls us.
3271 lockdep_assert_held(&fs_info->reclaim_bgs_lock);
3273 /* step one, relocate all the extents inside this chunk */
3274 btrfs_scrub_pause(fs_info);
3275 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3276 btrfs_scrub_continue(fs_info);
3279 * If we had a transaction abort, stop all running scrubs.
3280 * See transaction.c:cleanup_transaction() why we do it here.
3282 if (BTRFS_FS_ERROR(fs_info))
3283 btrfs_scrub_cancel(fs_info);
3287 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3290 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3291 length = block_group->length;
3292 btrfs_put_block_group(block_group);
3295 * On a zoned file system, discard the whole block group, this will
3296 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3297 * resetting the zone fails, don't treat it as a fatal problem from the
3298 * filesystem's point of view.
3300 if (btrfs_is_zoned(fs_info)) {
3301 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
3304 "failed to reset zone %llu after relocation",
3308 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3310 if (IS_ERR(trans)) {
3311 ret = PTR_ERR(trans);
3312 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3317 * step two, delete the device extents and the
3318 * chunk tree entries
3320 ret = btrfs_remove_chunk(trans, chunk_offset);
3321 btrfs_end_transaction(trans);
3325 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3327 struct btrfs_root *chunk_root = fs_info->chunk_root;
3328 struct btrfs_path *path;
3329 struct extent_buffer *leaf;
3330 struct btrfs_chunk *chunk;
3331 struct btrfs_key key;
3332 struct btrfs_key found_key;
3334 bool retried = false;
3338 path = btrfs_alloc_path();
3343 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3344 key.offset = (u64)-1;
3345 key.type = BTRFS_CHUNK_ITEM_KEY;
3348 mutex_lock(&fs_info->reclaim_bgs_lock);
3349 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3351 mutex_unlock(&fs_info->reclaim_bgs_lock);
3354 BUG_ON(ret == 0); /* Corruption */
3356 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3359 mutex_unlock(&fs_info->reclaim_bgs_lock);
3365 leaf = path->nodes[0];
3366 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3368 chunk = btrfs_item_ptr(leaf, path->slots[0],
3369 struct btrfs_chunk);
3370 chunk_type = btrfs_chunk_type(leaf, chunk);
3371 btrfs_release_path(path);
3373 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3374 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3380 mutex_unlock(&fs_info->reclaim_bgs_lock);
3382 if (found_key.offset == 0)
3384 key.offset = found_key.offset - 1;
3387 if (failed && !retried) {
3391 } else if (WARN_ON(failed && retried)) {
3395 btrfs_free_path(path);
3400 * return 1 : allocate a data chunk successfully,
3401 * return <0: errors during allocating a data chunk,
3402 * return 0 : no need to allocate a data chunk.
3404 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3407 struct btrfs_block_group *cache;
3411 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3413 chunk_type = cache->flags;
3414 btrfs_put_block_group(cache);
3416 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3419 spin_lock(&fs_info->data_sinfo->lock);
3420 bytes_used = fs_info->data_sinfo->bytes_used;
3421 spin_unlock(&fs_info->data_sinfo->lock);
3424 struct btrfs_trans_handle *trans;
3427 trans = btrfs_join_transaction(fs_info->tree_root);
3429 return PTR_ERR(trans);
3431 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3432 btrfs_end_transaction(trans);
3441 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3442 struct btrfs_balance_control *bctl)
3444 struct btrfs_root *root = fs_info->tree_root;
3445 struct btrfs_trans_handle *trans;
3446 struct btrfs_balance_item *item;
3447 struct btrfs_disk_balance_args disk_bargs;
3448 struct btrfs_path *path;
3449 struct extent_buffer *leaf;
3450 struct btrfs_key key;
3453 path = btrfs_alloc_path();
3457 trans = btrfs_start_transaction(root, 0);
3458 if (IS_ERR(trans)) {
3459 btrfs_free_path(path);
3460 return PTR_ERR(trans);
3463 key.objectid = BTRFS_BALANCE_OBJECTID;
3464 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3467 ret = btrfs_insert_empty_item(trans, root, path, &key,
3472 leaf = path->nodes[0];
3473 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3475 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3477 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3478 btrfs_set_balance_data(leaf, item, &disk_bargs);
3479 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3480 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3481 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3482 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3484 btrfs_set_balance_flags(leaf, item, bctl->flags);
3486 btrfs_mark_buffer_dirty(leaf);
3488 btrfs_free_path(path);
3489 err = btrfs_commit_transaction(trans);
3495 static int del_balance_item(struct btrfs_fs_info *fs_info)
3497 struct btrfs_root *root = fs_info->tree_root;
3498 struct btrfs_trans_handle *trans;
3499 struct btrfs_path *path;
3500 struct btrfs_key key;
3503 path = btrfs_alloc_path();
3507 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3508 if (IS_ERR(trans)) {
3509 btrfs_free_path(path);
3510 return PTR_ERR(trans);
3513 key.objectid = BTRFS_BALANCE_OBJECTID;
3514 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3517 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3525 ret = btrfs_del_item(trans, root, path);
3527 btrfs_free_path(path);
3528 err = btrfs_commit_transaction(trans);
3535 * This is a heuristic used to reduce the number of chunks balanced on
3536 * resume after balance was interrupted.
3538 static void update_balance_args(struct btrfs_balance_control *bctl)
3541 * Turn on soft mode for chunk types that were being converted.
3543 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3544 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3545 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3546 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3547 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3548 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3551 * Turn on usage filter if is not already used. The idea is
3552 * that chunks that we have already balanced should be
3553 * reasonably full. Don't do it for chunks that are being
3554 * converted - that will keep us from relocating unconverted
3555 * (albeit full) chunks.
3557 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3558 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3559 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3560 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3561 bctl->data.usage = 90;
3563 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3564 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3565 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3566 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3567 bctl->sys.usage = 90;
3569 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3570 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3571 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3572 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3573 bctl->meta.usage = 90;
3578 * Clear the balance status in fs_info and delete the balance item from disk.
3580 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3582 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3585 BUG_ON(!fs_info->balance_ctl);
3587 spin_lock(&fs_info->balance_lock);
3588 fs_info->balance_ctl = NULL;
3589 spin_unlock(&fs_info->balance_lock);
3592 ret = del_balance_item(fs_info);
3594 btrfs_handle_fs_error(fs_info, ret, NULL);
3598 * Balance filters. Return 1 if chunk should be filtered out
3599 * (should not be balanced).
3601 static int chunk_profiles_filter(u64 chunk_type,
3602 struct btrfs_balance_args *bargs)
3604 chunk_type = chunk_to_extended(chunk_type) &
3605 BTRFS_EXTENDED_PROFILE_MASK;
3607 if (bargs->profiles & chunk_type)
3613 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3614 struct btrfs_balance_args *bargs)
3616 struct btrfs_block_group *cache;
3618 u64 user_thresh_min;
3619 u64 user_thresh_max;
3622 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3623 chunk_used = cache->used;
3625 if (bargs->usage_min == 0)
3626 user_thresh_min = 0;
3628 user_thresh_min = mult_perc(cache->length, bargs->usage_min);
3630 if (bargs->usage_max == 0)
3631 user_thresh_max = 1;
3632 else if (bargs->usage_max > 100)
3633 user_thresh_max = cache->length;
3635 user_thresh_max = mult_perc(cache->length, bargs->usage_max);
3637 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3640 btrfs_put_block_group(cache);
3644 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3645 u64 chunk_offset, struct btrfs_balance_args *bargs)
3647 struct btrfs_block_group *cache;
3648 u64 chunk_used, user_thresh;
3651 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3652 chunk_used = cache->used;
3654 if (bargs->usage_min == 0)
3656 else if (bargs->usage > 100)
3657 user_thresh = cache->length;
3659 user_thresh = mult_perc(cache->length, bargs->usage);
3661 if (chunk_used < user_thresh)
3664 btrfs_put_block_group(cache);
3668 static int chunk_devid_filter(struct extent_buffer *leaf,
3669 struct btrfs_chunk *chunk,
3670 struct btrfs_balance_args *bargs)
3672 struct btrfs_stripe *stripe;
3673 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3676 for (i = 0; i < num_stripes; i++) {
3677 stripe = btrfs_stripe_nr(chunk, i);
3678 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3685 static u64 calc_data_stripes(u64 type, int num_stripes)
3687 const int index = btrfs_bg_flags_to_raid_index(type);
3688 const int ncopies = btrfs_raid_array[index].ncopies;
3689 const int nparity = btrfs_raid_array[index].nparity;
3691 return (num_stripes - nparity) / ncopies;
3694 /* [pstart, pend) */
3695 static int chunk_drange_filter(struct extent_buffer *leaf,
3696 struct btrfs_chunk *chunk,
3697 struct btrfs_balance_args *bargs)
3699 struct btrfs_stripe *stripe;
3700 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3707 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3710 type = btrfs_chunk_type(leaf, chunk);
3711 factor = calc_data_stripes(type, num_stripes);
3713 for (i = 0; i < num_stripes; i++) {
3714 stripe = btrfs_stripe_nr(chunk, i);
3715 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3718 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3719 stripe_length = btrfs_chunk_length(leaf, chunk);
3720 stripe_length = div_u64(stripe_length, factor);
3722 if (stripe_offset < bargs->pend &&
3723 stripe_offset + stripe_length > bargs->pstart)
3730 /* [vstart, vend) */
3731 static int chunk_vrange_filter(struct extent_buffer *leaf,
3732 struct btrfs_chunk *chunk,
3734 struct btrfs_balance_args *bargs)
3736 if (chunk_offset < bargs->vend &&
3737 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3738 /* at least part of the chunk is inside this vrange */
3744 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3745 struct btrfs_chunk *chunk,
3746 struct btrfs_balance_args *bargs)
3748 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3750 if (bargs->stripes_min <= num_stripes
3751 && num_stripes <= bargs->stripes_max)
3757 static int chunk_soft_convert_filter(u64 chunk_type,
3758 struct btrfs_balance_args *bargs)
3760 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3763 chunk_type = chunk_to_extended(chunk_type) &
3764 BTRFS_EXTENDED_PROFILE_MASK;
3766 if (bargs->target == chunk_type)
3772 static int should_balance_chunk(struct extent_buffer *leaf,
3773 struct btrfs_chunk *chunk, u64 chunk_offset)
3775 struct btrfs_fs_info *fs_info = leaf->fs_info;
3776 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3777 struct btrfs_balance_args *bargs = NULL;
3778 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3781 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3782 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3786 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3787 bargs = &bctl->data;
3788 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3790 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3791 bargs = &bctl->meta;
3793 /* profiles filter */
3794 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3795 chunk_profiles_filter(chunk_type, bargs)) {
3800 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3801 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3803 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3804 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3809 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3810 chunk_devid_filter(leaf, chunk, bargs)) {
3814 /* drange filter, makes sense only with devid filter */
3815 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3816 chunk_drange_filter(leaf, chunk, bargs)) {
3821 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3822 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3826 /* stripes filter */
3827 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3828 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3832 /* soft profile changing mode */
3833 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3834 chunk_soft_convert_filter(chunk_type, bargs)) {
3839 * limited by count, must be the last filter
3841 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3842 if (bargs->limit == 0)
3846 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3848 * Same logic as the 'limit' filter; the minimum cannot be
3849 * determined here because we do not have the global information
3850 * about the count of all chunks that satisfy the filters.
3852 if (bargs->limit_max == 0)
3861 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3863 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3864 struct btrfs_root *chunk_root = fs_info->chunk_root;
3866 struct btrfs_chunk *chunk;
3867 struct btrfs_path *path = NULL;
3868 struct btrfs_key key;
3869 struct btrfs_key found_key;
3870 struct extent_buffer *leaf;
3873 int enospc_errors = 0;
3874 bool counting = true;
3875 /* The single value limit and min/max limits use the same bytes in the */
3876 u64 limit_data = bctl->data.limit;
3877 u64 limit_meta = bctl->meta.limit;
3878 u64 limit_sys = bctl->sys.limit;
3882 int chunk_reserved = 0;
3884 path = btrfs_alloc_path();
3890 /* zero out stat counters */
3891 spin_lock(&fs_info->balance_lock);
3892 memset(&bctl->stat, 0, sizeof(bctl->stat));
3893 spin_unlock(&fs_info->balance_lock);
3897 * The single value limit and min/max limits use the same bytes
3900 bctl->data.limit = limit_data;
3901 bctl->meta.limit = limit_meta;
3902 bctl->sys.limit = limit_sys;
3904 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3905 key.offset = (u64)-1;
3906 key.type = BTRFS_CHUNK_ITEM_KEY;
3909 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3910 atomic_read(&fs_info->balance_cancel_req)) {
3915 mutex_lock(&fs_info->reclaim_bgs_lock);
3916 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3918 mutex_unlock(&fs_info->reclaim_bgs_lock);
3923 * this shouldn't happen, it means the last relocate
3927 BUG(); /* FIXME break ? */
3929 ret = btrfs_previous_item(chunk_root, path, 0,
3930 BTRFS_CHUNK_ITEM_KEY);
3932 mutex_unlock(&fs_info->reclaim_bgs_lock);
3937 leaf = path->nodes[0];
3938 slot = path->slots[0];
3939 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3941 if (found_key.objectid != key.objectid) {
3942 mutex_unlock(&fs_info->reclaim_bgs_lock);
3946 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3947 chunk_type = btrfs_chunk_type(leaf, chunk);
3950 spin_lock(&fs_info->balance_lock);
3951 bctl->stat.considered++;
3952 spin_unlock(&fs_info->balance_lock);
3955 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3957 btrfs_release_path(path);
3959 mutex_unlock(&fs_info->reclaim_bgs_lock);
3964 mutex_unlock(&fs_info->reclaim_bgs_lock);
3965 spin_lock(&fs_info->balance_lock);
3966 bctl->stat.expected++;
3967 spin_unlock(&fs_info->balance_lock);
3969 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3971 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3973 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3980 * Apply limit_min filter, no need to check if the LIMITS
3981 * filter is used, limit_min is 0 by default
3983 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3984 count_data < bctl->data.limit_min)
3985 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3986 count_meta < bctl->meta.limit_min)
3987 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3988 count_sys < bctl->sys.limit_min)) {
3989 mutex_unlock(&fs_info->reclaim_bgs_lock);
3993 if (!chunk_reserved) {
3995 * We may be relocating the only data chunk we have,
3996 * which could potentially end up with losing data's
3997 * raid profile, so lets allocate an empty one in
4000 ret = btrfs_may_alloc_data_chunk(fs_info,
4003 mutex_unlock(&fs_info->reclaim_bgs_lock);
4005 } else if (ret == 1) {
4010 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
4011 mutex_unlock(&fs_info->reclaim_bgs_lock);
4012 if (ret == -ENOSPC) {
4014 } else if (ret == -ETXTBSY) {
4016 "skipping relocation of block group %llu due to active swapfile",
4022 spin_lock(&fs_info->balance_lock);
4023 bctl->stat.completed++;
4024 spin_unlock(&fs_info->balance_lock);
4027 if (found_key.offset == 0)
4029 key.offset = found_key.offset - 1;
4033 btrfs_release_path(path);
4038 btrfs_free_path(path);
4039 if (enospc_errors) {
4040 btrfs_info(fs_info, "%d enospc errors during balance",
4050 * See if a given profile is valid and reduced.
4052 * @flags: profile to validate
4053 * @extended: if true @flags is treated as an extended profile
4055 static int alloc_profile_is_valid(u64 flags, int extended)
4057 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
4058 BTRFS_BLOCK_GROUP_PROFILE_MASK);
4060 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
4062 /* 1) check that all other bits are zeroed */
4066 /* 2) see if profile is reduced */
4068 return !extended; /* "0" is valid for usual profiles */
4070 return has_single_bit_set(flags);
4073 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
4075 /* cancel requested || normal exit path */
4076 return atomic_read(&fs_info->balance_cancel_req) ||
4077 (atomic_read(&fs_info->balance_pause_req) == 0 &&
4078 atomic_read(&fs_info->balance_cancel_req) == 0);
4082 * Validate target profile against allowed profiles and return true if it's OK.
4083 * Otherwise print the error message and return false.
4085 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
4086 const struct btrfs_balance_args *bargs,
4087 u64 allowed, const char *type)
4089 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
4092 /* Profile is valid and does not have bits outside of the allowed set */
4093 if (alloc_profile_is_valid(bargs->target, 1) &&
4094 (bargs->target & ~allowed) == 0)
4097 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
4098 type, btrfs_bg_type_to_raid_name(bargs->target));
4103 * Fill @buf with textual description of balance filter flags @bargs, up to
4104 * @size_buf including the terminating null. The output may be trimmed if it
4105 * does not fit into the provided buffer.
4107 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
4111 u32 size_bp = size_buf;
4113 u64 flags = bargs->flags;
4114 char tmp_buf[128] = {'\0'};
4119 #define CHECK_APPEND_NOARG(a) \
4121 ret = snprintf(bp, size_bp, (a)); \
4122 if (ret < 0 || ret >= size_bp) \
4123 goto out_overflow; \
4128 #define CHECK_APPEND_1ARG(a, v1) \
4130 ret = snprintf(bp, size_bp, (a), (v1)); \
4131 if (ret < 0 || ret >= size_bp) \
4132 goto out_overflow; \
4137 #define CHECK_APPEND_2ARG(a, v1, v2) \
4139 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4140 if (ret < 0 || ret >= size_bp) \
4141 goto out_overflow; \
4146 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
4147 CHECK_APPEND_1ARG("convert=%s,",
4148 btrfs_bg_type_to_raid_name(bargs->target));
4150 if (flags & BTRFS_BALANCE_ARGS_SOFT)
4151 CHECK_APPEND_NOARG("soft,");
4153 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4154 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4156 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4159 if (flags & BTRFS_BALANCE_ARGS_USAGE)
4160 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4162 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4163 CHECK_APPEND_2ARG("usage=%u..%u,",
4164 bargs->usage_min, bargs->usage_max);
4166 if (flags & BTRFS_BALANCE_ARGS_DEVID)
4167 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4169 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4170 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4171 bargs->pstart, bargs->pend);
4173 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4174 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4175 bargs->vstart, bargs->vend);
4177 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4178 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4180 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4181 CHECK_APPEND_2ARG("limit=%u..%u,",
4182 bargs->limit_min, bargs->limit_max);
4184 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4185 CHECK_APPEND_2ARG("stripes=%u..%u,",
4186 bargs->stripes_min, bargs->stripes_max);
4188 #undef CHECK_APPEND_2ARG
4189 #undef CHECK_APPEND_1ARG
4190 #undef CHECK_APPEND_NOARG
4194 if (size_bp < size_buf)
4195 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4200 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4202 u32 size_buf = 1024;
4203 char tmp_buf[192] = {'\0'};
4206 u32 size_bp = size_buf;
4208 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4210 buf = kzalloc(size_buf, GFP_KERNEL);
4216 #define CHECK_APPEND_1ARG(a, v1) \
4218 ret = snprintf(bp, size_bp, (a), (v1)); \
4219 if (ret < 0 || ret >= size_bp) \
4220 goto out_overflow; \
4225 if (bctl->flags & BTRFS_BALANCE_FORCE)
4226 CHECK_APPEND_1ARG("%s", "-f ");
4228 if (bctl->flags & BTRFS_BALANCE_DATA) {
4229 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4230 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4233 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4234 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4235 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4238 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4239 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4240 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4243 #undef CHECK_APPEND_1ARG
4247 if (size_bp < size_buf)
4248 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4249 btrfs_info(fs_info, "balance: %s %s",
4250 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4251 "resume" : "start", buf);
4257 * Should be called with balance mutexe held
4259 int btrfs_balance(struct btrfs_fs_info *fs_info,
4260 struct btrfs_balance_control *bctl,
4261 struct btrfs_ioctl_balance_args *bargs)
4263 u64 meta_target, data_target;
4269 bool reducing_redundancy;
4272 if (btrfs_fs_closing(fs_info) ||
4273 atomic_read(&fs_info->balance_pause_req) ||
4274 btrfs_should_cancel_balance(fs_info)) {
4279 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4280 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4284 * In case of mixed groups both data and meta should be picked,
4285 * and identical options should be given for both of them.
4287 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4288 if (mixed && (bctl->flags & allowed)) {
4289 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4290 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4291 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4293 "balance: mixed groups data and metadata options must be the same");
4300 * rw_devices will not change at the moment, device add/delete/replace
4303 num_devices = fs_info->fs_devices->rw_devices;
4306 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4307 * special bit for it, to make it easier to distinguish. Thus we need
4308 * to set it manually, or balance would refuse the profile.
4310 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4311 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4312 if (num_devices >= btrfs_raid_array[i].devs_min)
4313 allowed |= btrfs_raid_array[i].bg_flag;
4315 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4316 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4317 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4323 * Allow to reduce metadata or system integrity only if force set for
4324 * profiles with redundancy (copies, parity)
4327 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4328 if (btrfs_raid_array[i].ncopies >= 2 ||
4329 btrfs_raid_array[i].tolerated_failures >= 1)
4330 allowed |= btrfs_raid_array[i].bg_flag;
4333 seq = read_seqbegin(&fs_info->profiles_lock);
4335 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4336 (fs_info->avail_system_alloc_bits & allowed) &&
4337 !(bctl->sys.target & allowed)) ||
4338 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4339 (fs_info->avail_metadata_alloc_bits & allowed) &&
4340 !(bctl->meta.target & allowed)))
4341 reducing_redundancy = true;
4343 reducing_redundancy = false;
4345 /* if we're not converting, the target field is uninitialized */
4346 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4347 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4348 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4349 bctl->data.target : fs_info->avail_data_alloc_bits;
4350 } while (read_seqretry(&fs_info->profiles_lock, seq));
4352 if (reducing_redundancy) {
4353 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4355 "balance: force reducing metadata redundancy");
4358 "balance: reduces metadata redundancy, use --force if you want this");
4364 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4365 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4367 "balance: metadata profile %s has lower redundancy than data profile %s",
4368 btrfs_bg_type_to_raid_name(meta_target),
4369 btrfs_bg_type_to_raid_name(data_target));
4372 ret = insert_balance_item(fs_info, bctl);
4373 if (ret && ret != -EEXIST)
4376 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4377 BUG_ON(ret == -EEXIST);
4378 BUG_ON(fs_info->balance_ctl);
4379 spin_lock(&fs_info->balance_lock);
4380 fs_info->balance_ctl = bctl;
4381 spin_unlock(&fs_info->balance_lock);
4383 BUG_ON(ret != -EEXIST);
4384 spin_lock(&fs_info->balance_lock);
4385 update_balance_args(bctl);
4386 spin_unlock(&fs_info->balance_lock);
4389 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4390 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4391 describe_balance_start_or_resume(fs_info);
4392 mutex_unlock(&fs_info->balance_mutex);
4394 ret = __btrfs_balance(fs_info);
4396 mutex_lock(&fs_info->balance_mutex);
4397 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) {
4398 btrfs_info(fs_info, "balance: paused");
4399 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
4402 * Balance can be canceled by:
4404 * - Regular cancel request
4405 * Then ret == -ECANCELED and balance_cancel_req > 0
4407 * - Fatal signal to "btrfs" process
4408 * Either the signal caught by wait_reserve_ticket() and callers
4409 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4411 * Either way, in this case balance_cancel_req = 0, and
4412 * ret == -EINTR or ret == -ECANCELED.
4414 * So here we only check the return value to catch canceled balance.
4416 else if (ret == -ECANCELED || ret == -EINTR)
4417 btrfs_info(fs_info, "balance: canceled");
4419 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4421 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4424 memset(bargs, 0, sizeof(*bargs));
4425 btrfs_update_ioctl_balance_args(fs_info, bargs);
4428 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4429 balance_need_close(fs_info)) {
4430 reset_balance_state(fs_info);
4431 btrfs_exclop_finish(fs_info);
4434 wake_up(&fs_info->balance_wait_q);
4438 if (bctl->flags & BTRFS_BALANCE_RESUME)
4439 reset_balance_state(fs_info);
4442 btrfs_exclop_finish(fs_info);
4447 static int balance_kthread(void *data)
4449 struct btrfs_fs_info *fs_info = data;
4452 sb_start_write(fs_info->sb);
4453 mutex_lock(&fs_info->balance_mutex);
4454 if (fs_info->balance_ctl)
4455 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4456 mutex_unlock(&fs_info->balance_mutex);
4457 sb_end_write(fs_info->sb);
4462 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4464 struct task_struct *tsk;
4466 mutex_lock(&fs_info->balance_mutex);
4467 if (!fs_info->balance_ctl) {
4468 mutex_unlock(&fs_info->balance_mutex);
4471 mutex_unlock(&fs_info->balance_mutex);
4473 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4474 btrfs_info(fs_info, "balance: resume skipped");
4478 spin_lock(&fs_info->super_lock);
4479 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
4480 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;
4481 spin_unlock(&fs_info->super_lock);
4483 * A ro->rw remount sequence should continue with the paused balance
4484 * regardless of who pauses it, system or the user as of now, so set
4487 spin_lock(&fs_info->balance_lock);
4488 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4489 spin_unlock(&fs_info->balance_lock);
4491 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4492 return PTR_ERR_OR_ZERO(tsk);
4495 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4497 struct btrfs_balance_control *bctl;
4498 struct btrfs_balance_item *item;
4499 struct btrfs_disk_balance_args disk_bargs;
4500 struct btrfs_path *path;
4501 struct extent_buffer *leaf;
4502 struct btrfs_key key;
4505 path = btrfs_alloc_path();
4509 key.objectid = BTRFS_BALANCE_OBJECTID;
4510 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4513 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4516 if (ret > 0) { /* ret = -ENOENT; */
4521 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4527 leaf = path->nodes[0];
4528 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4530 bctl->flags = btrfs_balance_flags(leaf, item);
4531 bctl->flags |= BTRFS_BALANCE_RESUME;
4533 btrfs_balance_data(leaf, item, &disk_bargs);
4534 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4535 btrfs_balance_meta(leaf, item, &disk_bargs);
4536 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4537 btrfs_balance_sys(leaf, item, &disk_bargs);
4538 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4541 * This should never happen, as the paused balance state is recovered
4542 * during mount without any chance of other exclusive ops to collide.
4544 * This gives the exclusive op status to balance and keeps in paused
4545 * state until user intervention (cancel or umount). If the ownership
4546 * cannot be assigned, show a message but do not fail. The balance
4547 * is in a paused state and must have fs_info::balance_ctl properly
4550 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED))
4552 "balance: cannot set exclusive op status, resume manually");
4554 btrfs_release_path(path);
4556 mutex_lock(&fs_info->balance_mutex);
4557 BUG_ON(fs_info->balance_ctl);
4558 spin_lock(&fs_info->balance_lock);
4559 fs_info->balance_ctl = bctl;
4560 spin_unlock(&fs_info->balance_lock);
4561 mutex_unlock(&fs_info->balance_mutex);
4563 btrfs_free_path(path);
4567 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4571 mutex_lock(&fs_info->balance_mutex);
4572 if (!fs_info->balance_ctl) {
4573 mutex_unlock(&fs_info->balance_mutex);
4577 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4578 atomic_inc(&fs_info->balance_pause_req);
4579 mutex_unlock(&fs_info->balance_mutex);
4581 wait_event(fs_info->balance_wait_q,
4582 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4584 mutex_lock(&fs_info->balance_mutex);
4585 /* we are good with balance_ctl ripped off from under us */
4586 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4587 atomic_dec(&fs_info->balance_pause_req);
4592 mutex_unlock(&fs_info->balance_mutex);
4596 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4598 mutex_lock(&fs_info->balance_mutex);
4599 if (!fs_info->balance_ctl) {
4600 mutex_unlock(&fs_info->balance_mutex);
4605 * A paused balance with the item stored on disk can be resumed at
4606 * mount time if the mount is read-write. Otherwise it's still paused
4607 * and we must not allow cancelling as it deletes the item.
4609 if (sb_rdonly(fs_info->sb)) {
4610 mutex_unlock(&fs_info->balance_mutex);
4614 atomic_inc(&fs_info->balance_cancel_req);
4616 * if we are running just wait and return, balance item is
4617 * deleted in btrfs_balance in this case
4619 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4620 mutex_unlock(&fs_info->balance_mutex);
4621 wait_event(fs_info->balance_wait_q,
4622 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4623 mutex_lock(&fs_info->balance_mutex);
4625 mutex_unlock(&fs_info->balance_mutex);
4627 * Lock released to allow other waiters to continue, we'll
4628 * reexamine the status again.
4630 mutex_lock(&fs_info->balance_mutex);
4632 if (fs_info->balance_ctl) {
4633 reset_balance_state(fs_info);
4634 btrfs_exclop_finish(fs_info);
4635 btrfs_info(fs_info, "balance: canceled");
4639 BUG_ON(fs_info->balance_ctl ||
4640 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4641 atomic_dec(&fs_info->balance_cancel_req);
4642 mutex_unlock(&fs_info->balance_mutex);
4646 int btrfs_uuid_scan_kthread(void *data)
4648 struct btrfs_fs_info *fs_info = data;
4649 struct btrfs_root *root = fs_info->tree_root;
4650 struct btrfs_key key;
4651 struct btrfs_path *path = NULL;
4653 struct extent_buffer *eb;
4655 struct btrfs_root_item root_item;
4657 struct btrfs_trans_handle *trans = NULL;
4658 bool closing = false;
4660 path = btrfs_alloc_path();
4667 key.type = BTRFS_ROOT_ITEM_KEY;
4671 if (btrfs_fs_closing(fs_info)) {
4675 ret = btrfs_search_forward(root, &key, path,
4676 BTRFS_OLDEST_GENERATION);
4683 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4684 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4685 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4686 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4689 eb = path->nodes[0];
4690 slot = path->slots[0];
4691 item_size = btrfs_item_size(eb, slot);
4692 if (item_size < sizeof(root_item))
4695 read_extent_buffer(eb, &root_item,
4696 btrfs_item_ptr_offset(eb, slot),
4697 (int)sizeof(root_item));
4698 if (btrfs_root_refs(&root_item) == 0)
4701 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4702 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4706 btrfs_release_path(path);
4708 * 1 - subvol uuid item
4709 * 1 - received_subvol uuid item
4711 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4712 if (IS_ERR(trans)) {
4713 ret = PTR_ERR(trans);
4721 btrfs_release_path(path);
4722 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4723 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4724 BTRFS_UUID_KEY_SUBVOL,
4727 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4733 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4734 ret = btrfs_uuid_tree_add(trans,
4735 root_item.received_uuid,
4736 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4739 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4746 btrfs_release_path(path);
4748 ret = btrfs_end_transaction(trans);
4754 if (key.offset < (u64)-1) {
4756 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4758 key.type = BTRFS_ROOT_ITEM_KEY;
4759 } else if (key.objectid < (u64)-1) {
4761 key.type = BTRFS_ROOT_ITEM_KEY;
4770 btrfs_free_path(path);
4771 if (trans && !IS_ERR(trans))
4772 btrfs_end_transaction(trans);
4774 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4776 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4777 up(&fs_info->uuid_tree_rescan_sem);
4781 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4783 struct btrfs_trans_handle *trans;
4784 struct btrfs_root *tree_root = fs_info->tree_root;
4785 struct btrfs_root *uuid_root;
4786 struct task_struct *task;
4793 trans = btrfs_start_transaction(tree_root, 2);
4795 return PTR_ERR(trans);
4797 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4798 if (IS_ERR(uuid_root)) {
4799 ret = PTR_ERR(uuid_root);
4800 btrfs_abort_transaction(trans, ret);
4801 btrfs_end_transaction(trans);
4805 fs_info->uuid_root = uuid_root;
4807 ret = btrfs_commit_transaction(trans);
4811 down(&fs_info->uuid_tree_rescan_sem);
4812 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4814 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4815 btrfs_warn(fs_info, "failed to start uuid_scan task");
4816 up(&fs_info->uuid_tree_rescan_sem);
4817 return PTR_ERR(task);
4824 * shrinking a device means finding all of the device extents past
4825 * the new size, and then following the back refs to the chunks.
4826 * The chunk relocation code actually frees the device extent
4828 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4830 struct btrfs_fs_info *fs_info = device->fs_info;
4831 struct btrfs_root *root = fs_info->dev_root;
4832 struct btrfs_trans_handle *trans;
4833 struct btrfs_dev_extent *dev_extent = NULL;
4834 struct btrfs_path *path;
4840 bool retried = false;
4841 struct extent_buffer *l;
4842 struct btrfs_key key;
4843 struct btrfs_super_block *super_copy = fs_info->super_copy;
4844 u64 old_total = btrfs_super_total_bytes(super_copy);
4845 u64 old_size = btrfs_device_get_total_bytes(device);
4849 new_size = round_down(new_size, fs_info->sectorsize);
4851 diff = round_down(old_size - new_size, fs_info->sectorsize);
4853 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4856 path = btrfs_alloc_path();
4860 path->reada = READA_BACK;
4862 trans = btrfs_start_transaction(root, 0);
4863 if (IS_ERR(trans)) {
4864 btrfs_free_path(path);
4865 return PTR_ERR(trans);
4868 mutex_lock(&fs_info->chunk_mutex);
4870 btrfs_device_set_total_bytes(device, new_size);
4871 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4872 device->fs_devices->total_rw_bytes -= diff;
4873 atomic64_sub(diff, &fs_info->free_chunk_space);
4877 * Once the device's size has been set to the new size, ensure all
4878 * in-memory chunks are synced to disk so that the loop below sees them
4879 * and relocates them accordingly.
4881 if (contains_pending_extent(device, &start, diff)) {
4882 mutex_unlock(&fs_info->chunk_mutex);
4883 ret = btrfs_commit_transaction(trans);
4887 mutex_unlock(&fs_info->chunk_mutex);
4888 btrfs_end_transaction(trans);
4892 key.objectid = device->devid;
4893 key.offset = (u64)-1;
4894 key.type = BTRFS_DEV_EXTENT_KEY;
4897 mutex_lock(&fs_info->reclaim_bgs_lock);
4898 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4900 mutex_unlock(&fs_info->reclaim_bgs_lock);
4904 ret = btrfs_previous_item(root, path, 0, key.type);
4906 mutex_unlock(&fs_info->reclaim_bgs_lock);
4910 btrfs_release_path(path);
4915 slot = path->slots[0];
4916 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4918 if (key.objectid != device->devid) {
4919 mutex_unlock(&fs_info->reclaim_bgs_lock);
4920 btrfs_release_path(path);
4924 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4925 length = btrfs_dev_extent_length(l, dev_extent);
4927 if (key.offset + length <= new_size) {
4928 mutex_unlock(&fs_info->reclaim_bgs_lock);
4929 btrfs_release_path(path);
4933 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4934 btrfs_release_path(path);
4937 * We may be relocating the only data chunk we have,
4938 * which could potentially end up with losing data's
4939 * raid profile, so lets allocate an empty one in
4942 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4944 mutex_unlock(&fs_info->reclaim_bgs_lock);
4948 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4949 mutex_unlock(&fs_info->reclaim_bgs_lock);
4950 if (ret == -ENOSPC) {
4953 if (ret == -ETXTBSY) {
4955 "could not shrink block group %llu due to active swapfile",
4960 } while (key.offset-- > 0);
4962 if (failed && !retried) {
4966 } else if (failed && retried) {
4971 /* Shrinking succeeded, else we would be at "done". */
4972 trans = btrfs_start_transaction(root, 0);
4973 if (IS_ERR(trans)) {
4974 ret = PTR_ERR(trans);
4978 mutex_lock(&fs_info->chunk_mutex);
4979 /* Clear all state bits beyond the shrunk device size */
4980 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
4983 btrfs_device_set_disk_total_bytes(device, new_size);
4984 if (list_empty(&device->post_commit_list))
4985 list_add_tail(&device->post_commit_list,
4986 &trans->transaction->dev_update_list);
4988 WARN_ON(diff > old_total);
4989 btrfs_set_super_total_bytes(super_copy,
4990 round_down(old_total - diff, fs_info->sectorsize));
4991 mutex_unlock(&fs_info->chunk_mutex);
4993 btrfs_reserve_chunk_metadata(trans, false);
4994 /* Now btrfs_update_device() will change the on-disk size. */
4995 ret = btrfs_update_device(trans, device);
4996 btrfs_trans_release_chunk_metadata(trans);
4998 btrfs_abort_transaction(trans, ret);
4999 btrfs_end_transaction(trans);
5001 ret = btrfs_commit_transaction(trans);
5004 btrfs_free_path(path);
5006 mutex_lock(&fs_info->chunk_mutex);
5007 btrfs_device_set_total_bytes(device, old_size);
5008 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
5009 device->fs_devices->total_rw_bytes += diff;
5010 atomic64_add(diff, &fs_info->free_chunk_space);
5011 mutex_unlock(&fs_info->chunk_mutex);
5016 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
5017 struct btrfs_key *key,
5018 struct btrfs_chunk *chunk, int item_size)
5020 struct btrfs_super_block *super_copy = fs_info->super_copy;
5021 struct btrfs_disk_key disk_key;
5025 lockdep_assert_held(&fs_info->chunk_mutex);
5027 array_size = btrfs_super_sys_array_size(super_copy);
5028 if (array_size + item_size + sizeof(disk_key)
5029 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
5032 ptr = super_copy->sys_chunk_array + array_size;
5033 btrfs_cpu_key_to_disk(&disk_key, key);
5034 memcpy(ptr, &disk_key, sizeof(disk_key));
5035 ptr += sizeof(disk_key);
5036 memcpy(ptr, chunk, item_size);
5037 item_size += sizeof(disk_key);
5038 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
5044 * sort the devices in descending order by max_avail, total_avail
5046 static int btrfs_cmp_device_info(const void *a, const void *b)
5048 const struct btrfs_device_info *di_a = a;
5049 const struct btrfs_device_info *di_b = b;
5051 if (di_a->max_avail > di_b->max_avail)
5053 if (di_a->max_avail < di_b->max_avail)
5055 if (di_a->total_avail > di_b->total_avail)
5057 if (di_a->total_avail < di_b->total_avail)
5062 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
5064 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5067 btrfs_set_fs_incompat(info, RAID56);
5070 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
5072 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
5075 btrfs_set_fs_incompat(info, RAID1C34);
5079 * Structure used internally for btrfs_create_chunk() function.
5080 * Wraps needed parameters.
5082 struct alloc_chunk_ctl {
5085 /* Total number of stripes to allocate */
5087 /* sub_stripes info for map */
5089 /* Stripes per device */
5091 /* Maximum number of devices to use */
5093 /* Minimum number of devices to use */
5095 /* ndevs has to be a multiple of this */
5097 /* Number of copies */
5099 /* Number of stripes worth of bytes to store parity information */
5101 u64 max_stripe_size;
5109 static void init_alloc_chunk_ctl_policy_regular(
5110 struct btrfs_fs_devices *fs_devices,
5111 struct alloc_chunk_ctl *ctl)
5113 struct btrfs_space_info *space_info;
5115 space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type);
5118 ctl->max_chunk_size = READ_ONCE(space_info->chunk_size);
5119 ctl->max_stripe_size = ctl->max_chunk_size;
5121 if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM)
5122 ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK);
5124 /* We don't want a chunk larger than 10% of writable space */
5125 ctl->max_chunk_size = min(mult_perc(fs_devices->total_rw_bytes, 10),
5126 ctl->max_chunk_size);
5127 ctl->dev_extent_min = ctl->dev_stripes << BTRFS_STRIPE_LEN_SHIFT;
5130 static void init_alloc_chunk_ctl_policy_zoned(
5131 struct btrfs_fs_devices *fs_devices,
5132 struct alloc_chunk_ctl *ctl)
5134 u64 zone_size = fs_devices->fs_info->zone_size;
5136 int min_num_stripes = ctl->devs_min * ctl->dev_stripes;
5137 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;
5138 u64 min_chunk_size = min_data_stripes * zone_size;
5139 u64 type = ctl->type;
5141 ctl->max_stripe_size = zone_size;
5142 if (type & BTRFS_BLOCK_GROUP_DATA) {
5143 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,
5145 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5146 ctl->max_chunk_size = ctl->max_stripe_size;
5147 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5148 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5149 ctl->devs_max = min_t(int, ctl->devs_max,
5150 BTRFS_MAX_DEVS_SYS_CHUNK);
5155 /* We don't want a chunk larger than 10% of writable space */
5156 limit = max(round_down(mult_perc(fs_devices->total_rw_bytes, 10),
5159 ctl->max_chunk_size = min(limit, ctl->max_chunk_size);
5160 ctl->dev_extent_min = zone_size * ctl->dev_stripes;
5163 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
5164 struct alloc_chunk_ctl *ctl)
5166 int index = btrfs_bg_flags_to_raid_index(ctl->type);
5168 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
5169 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
5170 ctl->devs_max = btrfs_raid_array[index].devs_max;
5172 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
5173 ctl->devs_min = btrfs_raid_array[index].devs_min;
5174 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
5175 ctl->ncopies = btrfs_raid_array[index].ncopies;
5176 ctl->nparity = btrfs_raid_array[index].nparity;
5179 switch (fs_devices->chunk_alloc_policy) {
5180 case BTRFS_CHUNK_ALLOC_REGULAR:
5181 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
5183 case BTRFS_CHUNK_ALLOC_ZONED:
5184 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);
5191 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
5192 struct alloc_chunk_ctl *ctl,
5193 struct btrfs_device_info *devices_info)
5195 struct btrfs_fs_info *info = fs_devices->fs_info;
5196 struct btrfs_device *device;
5198 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
5205 * in the first pass through the devices list, we gather information
5206 * about the available holes on each device.
5208 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5209 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5211 "BTRFS: read-only device in alloc_list\n");
5215 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5216 &device->dev_state) ||
5217 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5220 if (device->total_bytes > device->bytes_used)
5221 total_avail = device->total_bytes - device->bytes_used;
5225 /* If there is no space on this device, skip it. */
5226 if (total_avail < ctl->dev_extent_min)
5229 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
5231 if (ret && ret != -ENOSPC)
5235 max_avail = dev_extent_want;
5237 if (max_avail < ctl->dev_extent_min) {
5238 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5240 "%s: devid %llu has no free space, have=%llu want=%llu",
5241 __func__, device->devid, max_avail,
5242 ctl->dev_extent_min);
5246 if (ndevs == fs_devices->rw_devices) {
5247 WARN(1, "%s: found more than %llu devices\n",
5248 __func__, fs_devices->rw_devices);
5251 devices_info[ndevs].dev_offset = dev_offset;
5252 devices_info[ndevs].max_avail = max_avail;
5253 devices_info[ndevs].total_avail = total_avail;
5254 devices_info[ndevs].dev = device;
5260 * now sort the devices by hole size / available space
5262 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5263 btrfs_cmp_device_info, NULL);
5268 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5269 struct btrfs_device_info *devices_info)
5271 /* Number of stripes that count for block group size */
5275 * The primary goal is to maximize the number of stripes, so use as
5276 * many devices as possible, even if the stripes are not maximum sized.
5278 * The DUP profile stores more than one stripe per device, the
5279 * max_avail is the total size so we have to adjust.
5281 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5283 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5285 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5286 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5289 * Use the number of data stripes to figure out how big this chunk is
5290 * really going to be in terms of logical address space, and compare
5291 * that answer with the max chunk size. If it's higher, we try to
5292 * reduce stripe_size.
5294 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5296 * Reduce stripe_size, round it up to a 16MB boundary again and
5297 * then use it, unless it ends up being even bigger than the
5298 * previous value we had already.
5300 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5301 data_stripes), SZ_16M),
5305 /* Stripe size should not go beyond 1G. */
5306 ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G);
5308 /* Align to BTRFS_STRIPE_LEN */
5309 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5310 ctl->chunk_size = ctl->stripe_size * data_stripes;
5315 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,
5316 struct btrfs_device_info *devices_info)
5318 u64 zone_size = devices_info[0].dev->zone_info->zone_size;
5319 /* Number of stripes that count for block group size */
5323 * It should hold because:
5324 * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5326 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);
5328 ctl->stripe_size = zone_size;
5329 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5330 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5332 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5333 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5334 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,
5335 ctl->stripe_size) + ctl->nparity,
5337 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5338 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5339 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);
5342 ctl->chunk_size = ctl->stripe_size * data_stripes;
5347 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5348 struct alloc_chunk_ctl *ctl,
5349 struct btrfs_device_info *devices_info)
5351 struct btrfs_fs_info *info = fs_devices->fs_info;
5354 * Round down to number of usable stripes, devs_increment can be any
5355 * number so we can't use round_down() that requires power of 2, while
5356 * rounddown is safe.
5358 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5360 if (ctl->ndevs < ctl->devs_min) {
5361 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5363 "%s: not enough devices with free space: have=%d minimum required=%d",
5364 __func__, ctl->ndevs, ctl->devs_min);
5369 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5371 switch (fs_devices->chunk_alloc_policy) {
5372 case BTRFS_CHUNK_ALLOC_REGULAR:
5373 return decide_stripe_size_regular(ctl, devices_info);
5374 case BTRFS_CHUNK_ALLOC_ZONED:
5375 return decide_stripe_size_zoned(ctl, devices_info);
5381 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
5382 struct alloc_chunk_ctl *ctl,
5383 struct btrfs_device_info *devices_info)
5385 struct btrfs_fs_info *info = trans->fs_info;
5386 struct map_lookup *map = NULL;
5387 struct extent_map_tree *em_tree;
5388 struct btrfs_block_group *block_group;
5389 struct extent_map *em;
5390 u64 start = ctl->start;
5391 u64 type = ctl->type;
5396 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
5398 return ERR_PTR(-ENOMEM);
5399 map->num_stripes = ctl->num_stripes;
5401 for (i = 0; i < ctl->ndevs; ++i) {
5402 for (j = 0; j < ctl->dev_stripes; ++j) {
5403 int s = i * ctl->dev_stripes + j;
5404 map->stripes[s].dev = devices_info[i].dev;
5405 map->stripes[s].physical = devices_info[i].dev_offset +
5406 j * ctl->stripe_size;
5409 map->io_align = BTRFS_STRIPE_LEN;
5410 map->io_width = BTRFS_STRIPE_LEN;
5412 map->sub_stripes = ctl->sub_stripes;
5414 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5416 em = alloc_extent_map();
5419 return ERR_PTR(-ENOMEM);
5421 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5422 em->map_lookup = map;
5424 em->len = ctl->chunk_size;
5425 em->block_start = 0;
5426 em->block_len = em->len;
5427 em->orig_block_len = ctl->stripe_size;
5429 em_tree = &info->mapping_tree;
5430 write_lock(&em_tree->lock);
5431 ret = add_extent_mapping(em_tree, em, 0);
5433 write_unlock(&em_tree->lock);
5434 free_extent_map(em);
5435 return ERR_PTR(ret);
5437 write_unlock(&em_tree->lock);
5439 block_group = btrfs_make_block_group(trans, type, start, ctl->chunk_size);
5440 if (IS_ERR(block_group))
5441 goto error_del_extent;
5443 for (i = 0; i < map->num_stripes; i++) {
5444 struct btrfs_device *dev = map->stripes[i].dev;
5446 btrfs_device_set_bytes_used(dev,
5447 dev->bytes_used + ctl->stripe_size);
5448 if (list_empty(&dev->post_commit_list))
5449 list_add_tail(&dev->post_commit_list,
5450 &trans->transaction->dev_update_list);
5453 atomic64_sub(ctl->stripe_size * map->num_stripes,
5454 &info->free_chunk_space);
5456 free_extent_map(em);
5457 check_raid56_incompat_flag(info, type);
5458 check_raid1c34_incompat_flag(info, type);
5463 write_lock(&em_tree->lock);
5464 remove_extent_mapping(em_tree, em);
5465 write_unlock(&em_tree->lock);
5467 /* One for our allocation */
5468 free_extent_map(em);
5469 /* One for the tree reference */
5470 free_extent_map(em);
5475 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
5478 struct btrfs_fs_info *info = trans->fs_info;
5479 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5480 struct btrfs_device_info *devices_info = NULL;
5481 struct alloc_chunk_ctl ctl;
5482 struct btrfs_block_group *block_group;
5485 lockdep_assert_held(&info->chunk_mutex);
5487 if (!alloc_profile_is_valid(type, 0)) {
5489 return ERR_PTR(-EINVAL);
5492 if (list_empty(&fs_devices->alloc_list)) {
5493 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5494 btrfs_debug(info, "%s: no writable device", __func__);
5495 return ERR_PTR(-ENOSPC);
5498 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5499 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5501 return ERR_PTR(-EINVAL);
5504 ctl.start = find_next_chunk(info);
5506 init_alloc_chunk_ctl(fs_devices, &ctl);
5508 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5511 return ERR_PTR(-ENOMEM);
5513 ret = gather_device_info(fs_devices, &ctl, devices_info);
5515 block_group = ERR_PTR(ret);
5519 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5521 block_group = ERR_PTR(ret);
5525 block_group = create_chunk(trans, &ctl, devices_info);
5528 kfree(devices_info);
5533 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5534 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5537 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5540 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
5541 struct btrfs_block_group *bg)
5543 struct btrfs_fs_info *fs_info = trans->fs_info;
5544 struct btrfs_root *chunk_root = fs_info->chunk_root;
5545 struct btrfs_key key;
5546 struct btrfs_chunk *chunk;
5547 struct btrfs_stripe *stripe;
5548 struct extent_map *em;
5549 struct map_lookup *map;
5555 * We take the chunk_mutex for 2 reasons:
5557 * 1) Updates and insertions in the chunk btree must be done while holding
5558 * the chunk_mutex, as well as updating the system chunk array in the
5559 * superblock. See the comment on top of btrfs_chunk_alloc() for the
5562 * 2) To prevent races with the final phase of a device replace operation
5563 * that replaces the device object associated with the map's stripes,
5564 * because the device object's id can change at any time during that
5565 * final phase of the device replace operation
5566 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5567 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5568 * which would cause a failure when updating the device item, which does
5569 * not exists, or persisting a stripe of the chunk item with such ID.
5570 * Here we can't use the device_list_mutex because our caller already
5571 * has locked the chunk_mutex, and the final phase of device replace
5572 * acquires both mutexes - first the device_list_mutex and then the
5573 * chunk_mutex. Using any of those two mutexes protects us from a
5574 * concurrent device replace.
5576 lockdep_assert_held(&fs_info->chunk_mutex);
5578 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
5581 btrfs_abort_transaction(trans, ret);
5585 map = em->map_lookup;
5586 item_size = btrfs_chunk_item_size(map->num_stripes);
5588 chunk = kzalloc(item_size, GFP_NOFS);
5591 btrfs_abort_transaction(trans, ret);
5595 for (i = 0; i < map->num_stripes; i++) {
5596 struct btrfs_device *device = map->stripes[i].dev;
5598 ret = btrfs_update_device(trans, device);
5603 stripe = &chunk->stripe;
5604 for (i = 0; i < map->num_stripes; i++) {
5605 struct btrfs_device *device = map->stripes[i].dev;
5606 const u64 dev_offset = map->stripes[i].physical;
5608 btrfs_set_stack_stripe_devid(stripe, device->devid);
5609 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5610 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5614 btrfs_set_stack_chunk_length(chunk, bg->length);
5615 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
5616 btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
5617 btrfs_set_stack_chunk_type(chunk, map->type);
5618 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5619 btrfs_set_stack_chunk_io_align(chunk, BTRFS_STRIPE_LEN);
5620 btrfs_set_stack_chunk_io_width(chunk, BTRFS_STRIPE_LEN);
5621 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5622 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5624 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5625 key.type = BTRFS_CHUNK_ITEM_KEY;
5626 key.offset = bg->start;
5628 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5632 set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags);
5634 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5635 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5642 free_extent_map(em);
5646 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5648 struct btrfs_fs_info *fs_info = trans->fs_info;
5650 struct btrfs_block_group *meta_bg;
5651 struct btrfs_block_group *sys_bg;
5654 * When adding a new device for sprouting, the seed device is read-only
5655 * so we must first allocate a metadata and a system chunk. But before
5656 * adding the block group items to the extent, device and chunk btrees,
5659 * 1) Create both chunks without doing any changes to the btrees, as
5660 * otherwise we would get -ENOSPC since the block groups from the
5661 * seed device are read-only;
5663 * 2) Add the device item for the new sprout device - finishing the setup
5664 * of a new block group requires updating the device item in the chunk
5665 * btree, so it must exist when we attempt to do it. The previous step
5666 * ensures this does not fail with -ENOSPC.
5668 * After that we can add the block group items to their btrees:
5669 * update existing device item in the chunk btree, add a new block group
5670 * item to the extent btree, add a new chunk item to the chunk btree and
5671 * finally add the new device extent items to the devices btree.
5674 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5675 meta_bg = btrfs_create_chunk(trans, alloc_profile);
5676 if (IS_ERR(meta_bg))
5677 return PTR_ERR(meta_bg);
5679 alloc_profile = btrfs_system_alloc_profile(fs_info);
5680 sys_bg = btrfs_create_chunk(trans, alloc_profile);
5682 return PTR_ERR(sys_bg);
5687 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5689 const int index = btrfs_bg_flags_to_raid_index(map->type);
5691 return btrfs_raid_array[index].tolerated_failures;
5694 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5696 struct extent_map *em;
5697 struct map_lookup *map;
5702 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5706 map = em->map_lookup;
5707 for (i = 0; i < map->num_stripes; i++) {
5708 if (test_bit(BTRFS_DEV_STATE_MISSING,
5709 &map->stripes[i].dev->dev_state)) {
5713 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5714 &map->stripes[i].dev->dev_state)) {
5721 * If the number of missing devices is larger than max errors, we can
5722 * not write the data into that chunk successfully.
5724 if (miss_ndevs > btrfs_chunk_max_errors(map))
5727 free_extent_map(em);
5731 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5733 struct extent_map *em;
5736 write_lock(&tree->lock);
5737 em = lookup_extent_mapping(tree, 0, (u64)-1);
5739 remove_extent_mapping(tree, em);
5740 write_unlock(&tree->lock);
5744 free_extent_map(em);
5745 /* once for the tree */
5746 free_extent_map(em);
5750 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5752 struct extent_map *em;
5753 struct map_lookup *map;
5754 enum btrfs_raid_types index;
5757 em = btrfs_get_chunk_map(fs_info, logical, len);
5760 * We could return errors for these cases, but that could get
5761 * ugly and we'd probably do the same thing which is just not do
5762 * anything else and exit, so return 1 so the callers don't try
5763 * to use other copies.
5767 map = em->map_lookup;
5768 index = btrfs_bg_flags_to_raid_index(map->type);
5770 /* Non-RAID56, use their ncopies from btrfs_raid_array. */
5771 if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5772 ret = btrfs_raid_array[index].ncopies;
5773 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5775 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5777 * There could be two corrupted data stripes, we need
5778 * to loop retry in order to rebuild the correct data.
5780 * Fail a stripe at a time on every retry except the
5781 * stripe under reconstruction.
5783 ret = map->num_stripes;
5784 free_extent_map(em);
5788 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5791 struct extent_map *em;
5792 struct map_lookup *map;
5793 unsigned long len = fs_info->sectorsize;
5795 if (!btrfs_fs_incompat(fs_info, RAID56))
5798 em = btrfs_get_chunk_map(fs_info, logical, len);
5800 if (!WARN_ON(IS_ERR(em))) {
5801 map = em->map_lookup;
5802 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5803 len = nr_data_stripes(map) << BTRFS_STRIPE_LEN_SHIFT;
5804 free_extent_map(em);
5809 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5811 struct extent_map *em;
5812 struct map_lookup *map;
5815 if (!btrfs_fs_incompat(fs_info, RAID56))
5818 em = btrfs_get_chunk_map(fs_info, logical, len);
5820 if(!WARN_ON(IS_ERR(em))) {
5821 map = em->map_lookup;
5822 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5824 free_extent_map(em);
5829 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5830 struct map_lookup *map, int first,
5831 int dev_replace_is_ongoing)
5835 int preferred_mirror;
5837 struct btrfs_device *srcdev;
5840 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5842 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5843 num_stripes = map->sub_stripes;
5845 num_stripes = map->num_stripes;
5847 switch (fs_info->fs_devices->read_policy) {
5849 /* Shouldn't happen, just warn and use pid instead of failing */
5850 btrfs_warn_rl(fs_info,
5851 "unknown read_policy type %u, reset to pid",
5852 fs_info->fs_devices->read_policy);
5853 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID;
5855 case BTRFS_READ_POLICY_PID:
5856 preferred_mirror = first + (current->pid % num_stripes);
5860 if (dev_replace_is_ongoing &&
5861 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5862 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5863 srcdev = fs_info->dev_replace.srcdev;
5868 * try to avoid the drive that is the source drive for a
5869 * dev-replace procedure, only choose it if no other non-missing
5870 * mirror is available
5872 for (tolerance = 0; tolerance < 2; tolerance++) {
5873 if (map->stripes[preferred_mirror].dev->bdev &&
5874 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5875 return preferred_mirror;
5876 for (i = first; i < first + num_stripes; i++) {
5877 if (map->stripes[i].dev->bdev &&
5878 (tolerance || map->stripes[i].dev != srcdev))
5883 /* we couldn't find one that doesn't fail. Just return something
5884 * and the io error handling code will clean up eventually
5886 return preferred_mirror;
5889 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info,
5892 struct btrfs_io_context *bioc;
5895 /* The size of btrfs_io_context */
5896 sizeof(struct btrfs_io_context) +
5897 /* Plus the variable array for the stripes */
5898 sizeof(struct btrfs_io_stripe) * (total_stripes),
5904 refcount_set(&bioc->refs, 1);
5906 bioc->fs_info = fs_info;
5907 bioc->replace_stripe_src = -1;
5908 bioc->full_stripe_logical = (u64)-1;
5913 void btrfs_get_bioc(struct btrfs_io_context *bioc)
5915 WARN_ON(!refcount_read(&bioc->refs));
5916 refcount_inc(&bioc->refs);
5919 void btrfs_put_bioc(struct btrfs_io_context *bioc)
5923 if (refcount_dec_and_test(&bioc->refs))
5928 * Please note that, discard won't be sent to target device of device
5931 struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info,
5932 u64 logical, u64 *length_ret,
5935 struct extent_map *em;
5936 struct map_lookup *map;
5937 struct btrfs_discard_stripe *stripes;
5938 u64 length = *length_ret;
5943 u64 stripe_end_offset;
5947 u32 sub_stripes = 0;
5948 u32 stripes_per_dev = 0;
5949 u32 remaining_stripes = 0;
5950 u32 last_stripe = 0;
5954 em = btrfs_get_chunk_map(fs_info, logical, length);
5956 return ERR_CAST(em);
5958 map = em->map_lookup;
5960 /* we don't discard raid56 yet */
5961 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5966 offset = logical - em->start;
5967 length = min_t(u64, em->start + em->len - logical, length);
5968 *length_ret = length;
5971 * stripe_nr counts the total number of stripes we have to stride
5972 * to get to this block
5974 stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
5976 /* stripe_offset is the offset of this block in its stripe */
5977 stripe_offset = offset - (stripe_nr << BTRFS_STRIPE_LEN_SHIFT);
5979 stripe_nr_end = round_up(offset + length, BTRFS_STRIPE_LEN) >>
5980 BTRFS_STRIPE_LEN_SHIFT;
5981 stripe_cnt = stripe_nr_end - stripe_nr;
5982 stripe_end_offset = (stripe_nr_end << BTRFS_STRIPE_LEN_SHIFT) -
5985 * after this, stripe_nr is the number of stripes on this
5986 * device we have to walk to find the data, and stripe_index is
5987 * the number of our device in the stripe array
5991 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5992 BTRFS_BLOCK_GROUP_RAID10)) {
5993 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5996 sub_stripes = map->sub_stripes;
5998 factor = map->num_stripes / sub_stripes;
5999 *num_stripes = min_t(u64, map->num_stripes,
6000 sub_stripes * stripe_cnt);
6001 stripe_index = stripe_nr % factor;
6002 stripe_nr /= factor;
6003 stripe_index *= sub_stripes;
6005 remaining_stripes = stripe_cnt % factor;
6006 stripes_per_dev = stripe_cnt / factor;
6007 last_stripe = ((stripe_nr_end - 1) % factor) * sub_stripes;
6008 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
6009 BTRFS_BLOCK_GROUP_DUP)) {
6010 *num_stripes = map->num_stripes;
6012 stripe_index = stripe_nr % map->num_stripes;
6013 stripe_nr /= map->num_stripes;
6016 stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS);
6022 for (i = 0; i < *num_stripes; i++) {
6023 stripes[i].physical =
6024 map->stripes[stripe_index].physical +
6025 stripe_offset + (stripe_nr << BTRFS_STRIPE_LEN_SHIFT);
6026 stripes[i].dev = map->stripes[stripe_index].dev;
6028 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6029 BTRFS_BLOCK_GROUP_RAID10)) {
6030 stripes[i].length = stripes_per_dev << BTRFS_STRIPE_LEN_SHIFT;
6032 if (i / sub_stripes < remaining_stripes)
6033 stripes[i].length += BTRFS_STRIPE_LEN;
6036 * Special for the first stripe and
6039 * |-------|...|-------|
6043 if (i < sub_stripes)
6044 stripes[i].length -= stripe_offset;
6046 if (stripe_index >= last_stripe &&
6047 stripe_index <= (last_stripe +
6049 stripes[i].length -= stripe_end_offset;
6051 if (i == sub_stripes - 1)
6054 stripes[i].length = length;
6058 if (stripe_index == map->num_stripes) {
6064 free_extent_map(em);
6067 free_extent_map(em);
6068 return ERR_PTR(ret);
6071 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)
6073 struct btrfs_block_group *cache;
6076 /* Non zoned filesystem does not use "to_copy" flag */
6077 if (!btrfs_is_zoned(fs_info))
6080 cache = btrfs_lookup_block_group(fs_info, logical);
6082 ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags);
6084 btrfs_put_block_group(cache);
6088 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
6089 struct btrfs_io_context *bioc,
6090 struct btrfs_dev_replace *dev_replace,
6092 int *num_stripes_ret, int *max_errors_ret)
6094 u64 srcdev_devid = dev_replace->srcdev->devid;
6096 * At this stage, num_stripes is still the real number of stripes,
6097 * excluding the duplicated stripes.
6099 int num_stripes = *num_stripes_ret;
6100 int nr_extra_stripes = 0;
6101 int max_errors = *max_errors_ret;
6105 * A block group which has "to_copy" set will eventually be copied by
6106 * the dev-replace process. We can avoid cloning IO here.
6108 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))
6112 * Duplicate the write operations while the dev-replace procedure is
6113 * running. Since the copying of the old disk to the new disk takes
6114 * place at run time while the filesystem is mounted writable, the
6115 * regular write operations to the old disk have to be duplicated to go
6116 * to the new disk as well.
6118 * Note that device->missing is handled by the caller, and that the
6119 * write to the old disk is already set up in the stripes array.
6121 for (i = 0; i < num_stripes; i++) {
6122 struct btrfs_io_stripe *old = &bioc->stripes[i];
6123 struct btrfs_io_stripe *new = &bioc->stripes[num_stripes + nr_extra_stripes];
6125 if (old->dev->devid != srcdev_devid)
6128 new->physical = old->physical;
6129 new->dev = dev_replace->tgtdev;
6130 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK)
6131 bioc->replace_stripe_src = i;
6135 /* We can only have at most 2 extra nr_stripes (for DUP). */
6136 ASSERT(nr_extra_stripes <= 2);
6138 * For GET_READ_MIRRORS, we can only return at most 1 extra stripe for
6140 * If we have 2 extra stripes, only choose the one with smaller physical.
6142 if (op == BTRFS_MAP_GET_READ_MIRRORS && nr_extra_stripes == 2) {
6143 struct btrfs_io_stripe *first = &bioc->stripes[num_stripes];
6144 struct btrfs_io_stripe *second = &bioc->stripes[num_stripes + 1];
6146 /* Only DUP can have two extra stripes. */
6147 ASSERT(bioc->map_type & BTRFS_BLOCK_GROUP_DUP);
6150 * Swap the last stripe stripes and reduce @nr_extra_stripes.
6151 * The extra stripe would still be there, but won't be accessed.
6153 if (first->physical > second->physical) {
6154 swap(second->physical, first->physical);
6155 swap(second->dev, first->dev);
6160 *num_stripes_ret = num_stripes + nr_extra_stripes;
6161 *max_errors_ret = max_errors + nr_extra_stripes;
6162 bioc->replace_nr_stripes = nr_extra_stripes;
6165 static bool need_full_stripe(enum btrfs_map_op op)
6167 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
6170 static u64 btrfs_max_io_len(struct map_lookup *map, enum btrfs_map_op op,
6171 u64 offset, u32 *stripe_nr, u64 *stripe_offset,
6172 u64 *full_stripe_start)
6174 ASSERT(op != BTRFS_MAP_DISCARD);
6177 * Stripe_nr is the stripe where this block falls. stripe_offset is
6178 * the offset of this block in its stripe.
6180 *stripe_offset = offset & BTRFS_STRIPE_LEN_MASK;
6181 *stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
6182 ASSERT(*stripe_offset < U32_MAX);
6184 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6185 unsigned long full_stripe_len = nr_data_stripes(map) <<
6186 BTRFS_STRIPE_LEN_SHIFT;
6189 * For full stripe start, we use previously calculated
6190 * @stripe_nr. Align it to nr_data_stripes, then multiply with
6193 * By this we can avoid u64 division completely. And we have
6194 * to go rounddown(), not round_down(), as nr_data_stripes is
6195 * not ensured to be power of 2.
6197 *full_stripe_start =
6198 rounddown(*stripe_nr, nr_data_stripes(map)) <<
6199 BTRFS_STRIPE_LEN_SHIFT;
6202 * For writes to RAID56, allow to write a full stripe set, but
6203 * no straddling of stripe sets.
6205 if (op == BTRFS_MAP_WRITE)
6206 return full_stripe_len - (offset - *full_stripe_start);
6210 * For other RAID types and for RAID56 reads, allow a single stripe (on
6213 if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK)
6214 return BTRFS_STRIPE_LEN - *stripe_offset;
6218 static void set_io_stripe(struct btrfs_io_stripe *dst, const struct map_lookup *map,
6219 u32 stripe_index, u64 stripe_offset, u32 stripe_nr)
6221 dst->dev = map->stripes[stripe_index].dev;
6222 dst->physical = map->stripes[stripe_index].physical +
6223 stripe_offset + (stripe_nr << BTRFS_STRIPE_LEN_SHIFT);
6226 int __btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6227 u64 logical, u64 *length,
6228 struct btrfs_io_context **bioc_ret,
6229 struct btrfs_io_stripe *smap, int *mirror_num_ret,
6232 struct extent_map *em;
6233 struct map_lookup *map;
6241 int mirror_num = (mirror_num_ret ? *mirror_num_ret : 0);
6245 struct btrfs_io_context *bioc = NULL;
6246 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6247 int dev_replace_is_ongoing = 0;
6248 u16 num_alloc_stripes;
6249 u64 raid56_full_stripe_start = (u64)-1;
6253 ASSERT(op != BTRFS_MAP_DISCARD);
6255 num_copies = btrfs_num_copies(fs_info, logical, fs_info->sectorsize);
6256 if (mirror_num > num_copies)
6259 em = btrfs_get_chunk_map(fs_info, logical, *length);
6263 map = em->map_lookup;
6264 data_stripes = nr_data_stripes(map);
6266 map_offset = logical - em->start;
6267 max_len = btrfs_max_io_len(map, op, map_offset, &stripe_nr,
6268 &stripe_offset, &raid56_full_stripe_start);
6269 *length = min_t(u64, em->len - map_offset, max_len);
6271 down_read(&dev_replace->rwsem);
6272 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6274 * Hold the semaphore for read during the whole operation, write is
6275 * requested at commit time but must wait.
6277 if (!dev_replace_is_ongoing)
6278 up_read(&dev_replace->rwsem);
6282 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6283 stripe_index = stripe_nr % map->num_stripes;
6284 stripe_nr /= map->num_stripes;
6285 if (!need_full_stripe(op))
6287 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6288 if (need_full_stripe(op))
6289 num_stripes = map->num_stripes;
6290 else if (mirror_num)
6291 stripe_index = mirror_num - 1;
6293 stripe_index = find_live_mirror(fs_info, map, 0,
6294 dev_replace_is_ongoing);
6295 mirror_num = stripe_index + 1;
6298 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6299 if (need_full_stripe(op)) {
6300 num_stripes = map->num_stripes;
6301 } else if (mirror_num) {
6302 stripe_index = mirror_num - 1;
6307 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6308 u32 factor = map->num_stripes / map->sub_stripes;
6310 stripe_index = (stripe_nr % factor) * map->sub_stripes;
6311 stripe_nr /= factor;
6313 if (need_full_stripe(op))
6314 num_stripes = map->sub_stripes;
6315 else if (mirror_num)
6316 stripe_index += mirror_num - 1;
6318 int old_stripe_index = stripe_index;
6319 stripe_index = find_live_mirror(fs_info, map,
6321 dev_replace_is_ongoing);
6322 mirror_num = stripe_index - old_stripe_index + 1;
6325 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6326 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6328 * Push stripe_nr back to the start of the full stripe
6329 * For those cases needing a full stripe, @stripe_nr
6330 * is the full stripe number.
6332 * Originally we go raid56_full_stripe_start / full_stripe_len,
6333 * but that can be expensive. Here we just divide
6334 * @stripe_nr with @data_stripes.
6336 stripe_nr /= data_stripes;
6338 /* RAID[56] write or recovery. Return all stripes */
6339 num_stripes = map->num_stripes;
6340 max_errors = btrfs_chunk_max_errors(map);
6342 /* Return the length to the full stripe end */
6343 *length = min(logical + *length,
6344 raid56_full_stripe_start + em->start +
6345 (data_stripes << BTRFS_STRIPE_LEN_SHIFT)) - logical;
6350 * Mirror #0 or #1 means the original data block.
6351 * Mirror #2 is RAID5 parity block.
6352 * Mirror #3 is RAID6 Q block.
6354 stripe_index = stripe_nr % data_stripes;
6355 stripe_nr /= data_stripes;
6357 stripe_index = data_stripes + mirror_num - 2;
6359 /* We distribute the parity blocks across stripes */
6360 stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
6361 if (!need_full_stripe(op) && mirror_num <= 1)
6366 * After this, stripe_nr is the number of stripes on this
6367 * device we have to walk to find the data, and stripe_index is
6368 * the number of our device in the stripe array
6370 stripe_index = stripe_nr % map->num_stripes;
6371 stripe_nr /= map->num_stripes;
6372 mirror_num = stripe_index + 1;
6374 if (stripe_index >= map->num_stripes) {
6376 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6377 stripe_index, map->num_stripes);
6382 num_alloc_stripes = num_stripes;
6383 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6384 op != BTRFS_MAP_READ)
6386 * For replace case, we need to add extra stripes for extra
6387 * duplicated stripes.
6389 * For both WRITE and GET_READ_MIRRORS, we may have at most
6390 * 2 more stripes (DUP types, otherwise 1).
6392 num_alloc_stripes += 2;
6395 * If this I/O maps to a single device, try to return the device and
6396 * physical block information on the stack instead of allocating an
6397 * I/O context structure.
6399 if (smap && num_alloc_stripes == 1 &&
6400 !((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1) &&
6401 (!need_full_stripe(op) || !dev_replace_is_ongoing ||
6402 !dev_replace->tgtdev)) {
6403 set_io_stripe(smap, map, stripe_index, stripe_offset, stripe_nr);
6404 *mirror_num_ret = mirror_num;
6410 bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes);
6415 bioc->map_type = map->type;
6418 * For RAID56 full map, we need to make sure the stripes[] follows the
6419 * rule that data stripes are all ordered, then followed with P and Q
6422 * It's still mostly the same as other profiles, just with extra rotation.
6424 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6425 (need_full_stripe(op) || mirror_num > 1)) {
6427 * For RAID56 @stripe_nr is already the number of full stripes
6428 * before us, which is also the rotation value (needs to modulo
6429 * with num_stripes).
6431 * In this case, we just add @stripe_nr with @i, then do the
6432 * modulo, to reduce one modulo call.
6434 bioc->full_stripe_logical = em->start +
6435 ((stripe_nr * data_stripes) << BTRFS_STRIPE_LEN_SHIFT);
6436 for (i = 0; i < num_stripes; i++)
6437 set_io_stripe(&bioc->stripes[i], map,
6438 (i + stripe_nr) % num_stripes,
6439 stripe_offset, stripe_nr);
6442 * For all other non-RAID56 profiles, just copy the target
6443 * stripe into the bioc.
6445 for (i = 0; i < num_stripes; i++) {
6446 set_io_stripe(&bioc->stripes[i], map, stripe_index,
6447 stripe_offset, stripe_nr);
6452 if (need_full_stripe(op))
6453 max_errors = btrfs_chunk_max_errors(map);
6455 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6456 need_full_stripe(op)) {
6457 handle_ops_on_dev_replace(op, bioc, dev_replace, logical,
6458 &num_stripes, &max_errors);
6462 bioc->num_stripes = num_stripes;
6463 bioc->max_errors = max_errors;
6464 bioc->mirror_num = mirror_num;
6467 if (dev_replace_is_ongoing) {
6468 lockdep_assert_held(&dev_replace->rwsem);
6469 /* Unlock and let waiting writers proceed */
6470 up_read(&dev_replace->rwsem);
6472 free_extent_map(em);
6476 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6477 u64 logical, u64 *length,
6478 struct btrfs_io_context **bioc_ret, int mirror_num)
6480 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret,
6481 NULL, &mirror_num, 0);
6484 /* For Scrub/replace */
6485 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6486 u64 logical, u64 *length,
6487 struct btrfs_io_context **bioc_ret)
6489 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret,
6493 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args,
6494 const struct btrfs_fs_devices *fs_devices)
6496 if (args->fsid == NULL)
6498 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0)
6503 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args,
6504 const struct btrfs_device *device)
6506 if (args->missing) {
6507 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) &&
6513 if (device->devid != args->devid)
6515 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0)
6521 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6524 * If devid and uuid are both specified, the match must be exact, otherwise
6525 * only devid is used.
6527 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,
6528 const struct btrfs_dev_lookup_args *args)
6530 struct btrfs_device *device;
6531 struct btrfs_fs_devices *seed_devs;
6533 if (dev_args_match_fs_devices(args, fs_devices)) {
6534 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6535 if (dev_args_match_device(args, device))
6540 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6541 if (!dev_args_match_fs_devices(args, seed_devs))
6543 list_for_each_entry(device, &seed_devs->devices, dev_list) {
6544 if (dev_args_match_device(args, device))
6552 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6553 u64 devid, u8 *dev_uuid)
6555 struct btrfs_device *device;
6556 unsigned int nofs_flag;
6559 * We call this under the chunk_mutex, so we want to use NOFS for this
6560 * allocation, however we don't want to change btrfs_alloc_device() to
6561 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6565 nofs_flag = memalloc_nofs_save();
6566 device = btrfs_alloc_device(NULL, &devid, dev_uuid, NULL);
6567 memalloc_nofs_restore(nofs_flag);
6571 list_add(&device->dev_list, &fs_devices->devices);
6572 device->fs_devices = fs_devices;
6573 fs_devices->num_devices++;
6575 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6576 fs_devices->missing_devices++;
6582 * Allocate new device struct, set up devid and UUID.
6584 * @fs_info: used only for generating a new devid, can be NULL if
6585 * devid is provided (i.e. @devid != NULL).
6586 * @devid: a pointer to devid for this device. If NULL a new devid
6588 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6590 * @path: a pointer to device path if available, NULL otherwise.
6592 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6593 * on error. Returned struct is not linked onto any lists and must be
6594 * destroyed with btrfs_free_device.
6596 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6597 const u64 *devid, const u8 *uuid,
6600 struct btrfs_device *dev;
6603 if (WARN_ON(!devid && !fs_info))
6604 return ERR_PTR(-EINVAL);
6606 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
6608 return ERR_PTR(-ENOMEM);
6610 INIT_LIST_HEAD(&dev->dev_list);
6611 INIT_LIST_HEAD(&dev->dev_alloc_list);
6612 INIT_LIST_HEAD(&dev->post_commit_list);
6614 atomic_set(&dev->dev_stats_ccnt, 0);
6615 btrfs_device_data_ordered_init(dev);
6616 extent_io_tree_init(fs_info, &dev->alloc_state, IO_TREE_DEVICE_ALLOC_STATE);
6623 ret = find_next_devid(fs_info, &tmp);
6625 btrfs_free_device(dev);
6626 return ERR_PTR(ret);
6632 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6634 generate_random_uuid(dev->uuid);
6637 struct rcu_string *name;
6639 name = rcu_string_strdup(path, GFP_KERNEL);
6641 btrfs_free_device(dev);
6642 return ERR_PTR(-ENOMEM);
6644 rcu_assign_pointer(dev->name, name);
6650 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6651 u64 devid, u8 *uuid, bool error)
6654 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6657 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6661 u64 btrfs_calc_stripe_length(const struct extent_map *em)
6663 const struct map_lookup *map = em->map_lookup;
6664 const int data_stripes = calc_data_stripes(map->type, map->num_stripes);
6666 return div_u64(em->len, data_stripes);
6669 #if BITS_PER_LONG == 32
6671 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
6672 * can't be accessed on 32bit systems.
6674 * This function do mount time check to reject the fs if it already has
6675 * metadata chunk beyond that limit.
6677 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6678 u64 logical, u64 length, u64 type)
6680 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6683 if (logical + length < MAX_LFS_FILESIZE)
6686 btrfs_err_32bit_limit(fs_info);
6691 * This is to give early warning for any metadata chunk reaching
6692 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
6693 * Although we can still access the metadata, it's not going to be possible
6694 * once the limit is reached.
6696 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6697 u64 logical, u64 length, u64 type)
6699 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6702 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)
6705 btrfs_warn_32bit_limit(fs_info);
6709 static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info,
6710 u64 devid, u8 *uuid)
6712 struct btrfs_device *dev;
6714 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6715 btrfs_report_missing_device(fs_info, devid, uuid, true);
6716 return ERR_PTR(-ENOENT);
6719 dev = add_missing_dev(fs_info->fs_devices, devid, uuid);
6721 btrfs_err(fs_info, "failed to init missing device %llu: %ld",
6722 devid, PTR_ERR(dev));
6725 btrfs_report_missing_device(fs_info, devid, uuid, false);
6730 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6731 struct btrfs_chunk *chunk)
6733 BTRFS_DEV_LOOKUP_ARGS(args);
6734 struct btrfs_fs_info *fs_info = leaf->fs_info;
6735 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6736 struct map_lookup *map;
6737 struct extent_map *em;
6742 u8 uuid[BTRFS_UUID_SIZE];
6748 logical = key->offset;
6749 length = btrfs_chunk_length(leaf, chunk);
6750 type = btrfs_chunk_type(leaf, chunk);
6751 index = btrfs_bg_flags_to_raid_index(type);
6752 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6754 #if BITS_PER_LONG == 32
6755 ret = check_32bit_meta_chunk(fs_info, logical, length, type);
6758 warn_32bit_meta_chunk(fs_info, logical, length, type);
6762 * Only need to verify chunk item if we're reading from sys chunk array,
6763 * as chunk item in tree block is already verified by tree-checker.
6765 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6766 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6771 read_lock(&map_tree->lock);
6772 em = lookup_extent_mapping(map_tree, logical, 1);
6773 read_unlock(&map_tree->lock);
6775 /* already mapped? */
6776 if (em && em->start <= logical && em->start + em->len > logical) {
6777 free_extent_map(em);
6780 free_extent_map(em);
6783 em = alloc_extent_map();
6786 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6788 free_extent_map(em);
6792 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6793 em->map_lookup = map;
6794 em->start = logical;
6797 em->block_start = 0;
6798 em->block_len = em->len;
6800 map->num_stripes = num_stripes;
6801 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6802 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6805 * We can't use the sub_stripes value, as for profiles other than
6806 * RAID10, they may have 0 as sub_stripes for filesystems created by
6807 * older mkfs (<v5.4).
6808 * In that case, it can cause divide-by-zero errors later.
6809 * Since currently sub_stripes is fixed for each profile, let's
6810 * use the trusted value instead.
6812 map->sub_stripes = btrfs_raid_array[index].sub_stripes;
6813 map->verified_stripes = 0;
6814 em->orig_block_len = btrfs_calc_stripe_length(em);
6815 for (i = 0; i < num_stripes; i++) {
6816 map->stripes[i].physical =
6817 btrfs_stripe_offset_nr(leaf, chunk, i);
6818 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6820 read_extent_buffer(leaf, uuid, (unsigned long)
6821 btrfs_stripe_dev_uuid_nr(chunk, i),
6824 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args);
6825 if (!map->stripes[i].dev) {
6826 map->stripes[i].dev = handle_missing_device(fs_info,
6828 if (IS_ERR(map->stripes[i].dev)) {
6829 ret = PTR_ERR(map->stripes[i].dev);
6830 free_extent_map(em);
6835 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6836 &(map->stripes[i].dev->dev_state));
6839 write_lock(&map_tree->lock);
6840 ret = add_extent_mapping(map_tree, em, 0);
6841 write_unlock(&map_tree->lock);
6844 "failed to add chunk map, start=%llu len=%llu: %d",
6845 em->start, em->len, ret);
6847 free_extent_map(em);
6852 static void fill_device_from_item(struct extent_buffer *leaf,
6853 struct btrfs_dev_item *dev_item,
6854 struct btrfs_device *device)
6858 device->devid = btrfs_device_id(leaf, dev_item);
6859 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6860 device->total_bytes = device->disk_total_bytes;
6861 device->commit_total_bytes = device->disk_total_bytes;
6862 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6863 device->commit_bytes_used = device->bytes_used;
6864 device->type = btrfs_device_type(leaf, dev_item);
6865 device->io_align = btrfs_device_io_align(leaf, dev_item);
6866 device->io_width = btrfs_device_io_width(leaf, dev_item);
6867 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6868 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6869 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6871 ptr = btrfs_device_uuid(dev_item);
6872 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6875 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6878 struct btrfs_fs_devices *fs_devices;
6881 lockdep_assert_held(&uuid_mutex);
6884 /* This will match only for multi-device seed fs */
6885 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
6886 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6890 fs_devices = find_fsid(fsid, NULL);
6892 if (!btrfs_test_opt(fs_info, DEGRADED))
6893 return ERR_PTR(-ENOENT);
6895 fs_devices = alloc_fs_devices(fsid, NULL);
6896 if (IS_ERR(fs_devices))
6899 fs_devices->seeding = true;
6900 fs_devices->opened = 1;
6905 * Upon first call for a seed fs fsid, just create a private copy of the
6906 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
6908 fs_devices = clone_fs_devices(fs_devices);
6909 if (IS_ERR(fs_devices))
6912 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
6914 free_fs_devices(fs_devices);
6915 return ERR_PTR(ret);
6918 if (!fs_devices->seeding) {
6919 close_fs_devices(fs_devices);
6920 free_fs_devices(fs_devices);
6921 return ERR_PTR(-EINVAL);
6924 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
6929 static int read_one_dev(struct extent_buffer *leaf,
6930 struct btrfs_dev_item *dev_item)
6932 BTRFS_DEV_LOOKUP_ARGS(args);
6933 struct btrfs_fs_info *fs_info = leaf->fs_info;
6934 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6935 struct btrfs_device *device;
6938 u8 fs_uuid[BTRFS_FSID_SIZE];
6939 u8 dev_uuid[BTRFS_UUID_SIZE];
6941 devid = btrfs_device_id(leaf, dev_item);
6943 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6945 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6947 args.uuid = dev_uuid;
6948 args.fsid = fs_uuid;
6950 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
6951 fs_devices = open_seed_devices(fs_info, fs_uuid);
6952 if (IS_ERR(fs_devices))
6953 return PTR_ERR(fs_devices);
6956 device = btrfs_find_device(fs_info->fs_devices, &args);
6958 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6959 btrfs_report_missing_device(fs_info, devid,
6964 device = add_missing_dev(fs_devices, devid, dev_uuid);
6965 if (IS_ERR(device)) {
6967 "failed to add missing dev %llu: %ld",
6968 devid, PTR_ERR(device));
6969 return PTR_ERR(device);
6971 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6973 if (!device->bdev) {
6974 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6975 btrfs_report_missing_device(fs_info,
6976 devid, dev_uuid, true);
6979 btrfs_report_missing_device(fs_info, devid,
6983 if (!device->bdev &&
6984 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6986 * this happens when a device that was properly setup
6987 * in the device info lists suddenly goes bad.
6988 * device->bdev is NULL, and so we have to set
6989 * device->missing to one here
6991 device->fs_devices->missing_devices++;
6992 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6995 /* Move the device to its own fs_devices */
6996 if (device->fs_devices != fs_devices) {
6997 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6998 &device->dev_state));
7000 list_move(&device->dev_list, &fs_devices->devices);
7001 device->fs_devices->num_devices--;
7002 fs_devices->num_devices++;
7004 device->fs_devices->missing_devices--;
7005 fs_devices->missing_devices++;
7007 device->fs_devices = fs_devices;
7011 if (device->fs_devices != fs_info->fs_devices) {
7012 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
7013 if (device->generation !=
7014 btrfs_device_generation(leaf, dev_item))
7018 fill_device_from_item(leaf, dev_item, device);
7020 u64 max_total_bytes = bdev_nr_bytes(device->bdev);
7022 if (device->total_bytes > max_total_bytes) {
7024 "device total_bytes should be at most %llu but found %llu",
7025 max_total_bytes, device->total_bytes);
7029 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7030 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7031 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7032 device->fs_devices->total_rw_bytes += device->total_bytes;
7033 atomic64_add(device->total_bytes - device->bytes_used,
7034 &fs_info->free_chunk_space);
7040 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7042 struct btrfs_super_block *super_copy = fs_info->super_copy;
7043 struct extent_buffer *sb;
7044 struct btrfs_disk_key *disk_key;
7045 struct btrfs_chunk *chunk;
7047 unsigned long sb_array_offset;
7054 struct btrfs_key key;
7056 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7059 * We allocated a dummy extent, just to use extent buffer accessors.
7060 * There will be unused space after BTRFS_SUPER_INFO_SIZE, but
7061 * that's fine, we will not go beyond system chunk array anyway.
7063 sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET);
7066 set_extent_buffer_uptodate(sb);
7068 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7069 array_size = btrfs_super_sys_array_size(super_copy);
7071 array_ptr = super_copy->sys_chunk_array;
7072 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7075 while (cur_offset < array_size) {
7076 disk_key = (struct btrfs_disk_key *)array_ptr;
7077 len = sizeof(*disk_key);
7078 if (cur_offset + len > array_size)
7079 goto out_short_read;
7081 btrfs_disk_key_to_cpu(&key, disk_key);
7084 sb_array_offset += len;
7087 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7089 "unexpected item type %u in sys_array at offset %u",
7090 (u32)key.type, cur_offset);
7095 chunk = (struct btrfs_chunk *)sb_array_offset;
7097 * At least one btrfs_chunk with one stripe must be present,
7098 * exact stripe count check comes afterwards
7100 len = btrfs_chunk_item_size(1);
7101 if (cur_offset + len > array_size)
7102 goto out_short_read;
7104 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7107 "invalid number of stripes %u in sys_array at offset %u",
7108 num_stripes, cur_offset);
7113 type = btrfs_chunk_type(sb, chunk);
7114 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7116 "invalid chunk type %llu in sys_array at offset %u",
7122 len = btrfs_chunk_item_size(num_stripes);
7123 if (cur_offset + len > array_size)
7124 goto out_short_read;
7126 ret = read_one_chunk(&key, sb, chunk);
7131 sb_array_offset += len;
7134 clear_extent_buffer_uptodate(sb);
7135 free_extent_buffer_stale(sb);
7139 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7141 clear_extent_buffer_uptodate(sb);
7142 free_extent_buffer_stale(sb);
7147 * Check if all chunks in the fs are OK for read-write degraded mount
7149 * If the @failing_dev is specified, it's accounted as missing.
7151 * Return true if all chunks meet the minimal RW mount requirements.
7152 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7154 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7155 struct btrfs_device *failing_dev)
7157 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7158 struct extent_map *em;
7162 read_lock(&map_tree->lock);
7163 em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7164 read_unlock(&map_tree->lock);
7165 /* No chunk at all? Return false anyway */
7171 struct map_lookup *map;
7176 map = em->map_lookup;
7178 btrfs_get_num_tolerated_disk_barrier_failures(
7180 for (i = 0; i < map->num_stripes; i++) {
7181 struct btrfs_device *dev = map->stripes[i].dev;
7183 if (!dev || !dev->bdev ||
7184 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7185 dev->last_flush_error)
7187 else if (failing_dev && failing_dev == dev)
7190 if (missing > max_tolerated) {
7193 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7194 em->start, missing, max_tolerated);
7195 free_extent_map(em);
7199 next_start = extent_map_end(em);
7200 free_extent_map(em);
7202 read_lock(&map_tree->lock);
7203 em = lookup_extent_mapping(map_tree, next_start,
7204 (u64)(-1) - next_start);
7205 read_unlock(&map_tree->lock);
7211 static void readahead_tree_node_children(struct extent_buffer *node)
7214 const int nr_items = btrfs_header_nritems(node);
7216 for (i = 0; i < nr_items; i++)
7217 btrfs_readahead_node_child(node, i);
7220 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7222 struct btrfs_root *root = fs_info->chunk_root;
7223 struct btrfs_path *path;
7224 struct extent_buffer *leaf;
7225 struct btrfs_key key;
7226 struct btrfs_key found_key;
7231 u64 last_ra_node = 0;
7233 path = btrfs_alloc_path();
7238 * uuid_mutex is needed only if we are mounting a sprout FS
7239 * otherwise we don't need it.
7241 mutex_lock(&uuid_mutex);
7244 * It is possible for mount and umount to race in such a way that
7245 * we execute this code path, but open_fs_devices failed to clear
7246 * total_rw_bytes. We certainly want it cleared before reading the
7247 * device items, so clear it here.
7249 fs_info->fs_devices->total_rw_bytes = 0;
7252 * Lockdep complains about possible circular locking dependency between
7253 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores
7254 * used for freeze procection of a fs (struct super_block.s_writers),
7255 * which we take when starting a transaction, and extent buffers of the
7256 * chunk tree if we call read_one_dev() while holding a lock on an
7257 * extent buffer of the chunk tree. Since we are mounting the filesystem
7258 * and at this point there can't be any concurrent task modifying the
7259 * chunk tree, to keep it simple, just skip locking on the chunk tree.
7261 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
7262 path->skip_locking = 1;
7265 * Read all device items, and then all the chunk items. All
7266 * device items are found before any chunk item (their object id
7267 * is smaller than the lowest possible object id for a chunk
7268 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7270 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7273 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) {
7274 struct extent_buffer *node = path->nodes[1];
7276 leaf = path->nodes[0];
7277 slot = path->slots[0];
7280 if (last_ra_node != node->start) {
7281 readahead_tree_node_children(node);
7282 last_ra_node = node->start;
7285 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7286 struct btrfs_dev_item *dev_item;
7287 dev_item = btrfs_item_ptr(leaf, slot,
7288 struct btrfs_dev_item);
7289 ret = read_one_dev(leaf, dev_item);
7293 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7294 struct btrfs_chunk *chunk;
7297 * We are only called at mount time, so no need to take
7298 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7299 * we always lock first fs_info->chunk_mutex before
7300 * acquiring any locks on the chunk tree. This is a
7301 * requirement for chunk allocation, see the comment on
7302 * top of btrfs_chunk_alloc() for details.
7304 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7305 ret = read_one_chunk(&found_key, leaf, chunk);
7310 /* Catch error found during iteration */
7317 * After loading chunk tree, we've got all device information,
7318 * do another round of validation checks.
7320 if (total_dev != fs_info->fs_devices->total_devices) {
7322 "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit",
7323 btrfs_super_num_devices(fs_info->super_copy),
7325 fs_info->fs_devices->total_devices = total_dev;
7326 btrfs_set_super_num_devices(fs_info->super_copy, total_dev);
7328 if (btrfs_super_total_bytes(fs_info->super_copy) <
7329 fs_info->fs_devices->total_rw_bytes) {
7331 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7332 btrfs_super_total_bytes(fs_info->super_copy),
7333 fs_info->fs_devices->total_rw_bytes);
7339 mutex_unlock(&uuid_mutex);
7341 btrfs_free_path(path);
7345 int btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7347 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7348 struct btrfs_device *device;
7351 fs_devices->fs_info = fs_info;
7353 mutex_lock(&fs_devices->device_list_mutex);
7354 list_for_each_entry(device, &fs_devices->devices, dev_list)
7355 device->fs_info = fs_info;
7357 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7358 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7359 device->fs_info = fs_info;
7360 ret = btrfs_get_dev_zone_info(device, false);
7365 seed_devs->fs_info = fs_info;
7367 mutex_unlock(&fs_devices->device_list_mutex);
7372 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7373 const struct btrfs_dev_stats_item *ptr,
7378 read_extent_buffer(eb, &val,
7379 offsetof(struct btrfs_dev_stats_item, values) +
7380 ((unsigned long)ptr) + (index * sizeof(u64)),
7385 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7386 struct btrfs_dev_stats_item *ptr,
7389 write_extent_buffer(eb, &val,
7390 offsetof(struct btrfs_dev_stats_item, values) +
7391 ((unsigned long)ptr) + (index * sizeof(u64)),
7395 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7396 struct btrfs_path *path)
7398 struct btrfs_dev_stats_item *ptr;
7399 struct extent_buffer *eb;
7400 struct btrfs_key key;
7404 if (!device->fs_info->dev_root)
7407 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7408 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7409 key.offset = device->devid;
7410 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7412 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7413 btrfs_dev_stat_set(device, i, 0);
7414 device->dev_stats_valid = 1;
7415 btrfs_release_path(path);
7416 return ret < 0 ? ret : 0;
7418 slot = path->slots[0];
7419 eb = path->nodes[0];
7420 item_size = btrfs_item_size(eb, slot);
7422 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7424 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7425 if (item_size >= (1 + i) * sizeof(__le64))
7426 btrfs_dev_stat_set(device, i,
7427 btrfs_dev_stats_value(eb, ptr, i));
7429 btrfs_dev_stat_set(device, i, 0);
7432 device->dev_stats_valid = 1;
7433 btrfs_dev_stat_print_on_load(device);
7434 btrfs_release_path(path);
7439 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7441 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7442 struct btrfs_device *device;
7443 struct btrfs_path *path = NULL;
7446 path = btrfs_alloc_path();
7450 mutex_lock(&fs_devices->device_list_mutex);
7451 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7452 ret = btrfs_device_init_dev_stats(device, path);
7456 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7457 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7458 ret = btrfs_device_init_dev_stats(device, path);
7464 mutex_unlock(&fs_devices->device_list_mutex);
7466 btrfs_free_path(path);
7470 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7471 struct btrfs_device *device)
7473 struct btrfs_fs_info *fs_info = trans->fs_info;
7474 struct btrfs_root *dev_root = fs_info->dev_root;
7475 struct btrfs_path *path;
7476 struct btrfs_key key;
7477 struct extent_buffer *eb;
7478 struct btrfs_dev_stats_item *ptr;
7482 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7483 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7484 key.offset = device->devid;
7486 path = btrfs_alloc_path();
7489 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7491 btrfs_warn_in_rcu(fs_info,
7492 "error %d while searching for dev_stats item for device %s",
7493 ret, btrfs_dev_name(device));
7498 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7499 /* need to delete old one and insert a new one */
7500 ret = btrfs_del_item(trans, dev_root, path);
7502 btrfs_warn_in_rcu(fs_info,
7503 "delete too small dev_stats item for device %s failed %d",
7504 btrfs_dev_name(device), ret);
7511 /* need to insert a new item */
7512 btrfs_release_path(path);
7513 ret = btrfs_insert_empty_item(trans, dev_root, path,
7514 &key, sizeof(*ptr));
7516 btrfs_warn_in_rcu(fs_info,
7517 "insert dev_stats item for device %s failed %d",
7518 btrfs_dev_name(device), ret);
7523 eb = path->nodes[0];
7524 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7525 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7526 btrfs_set_dev_stats_value(eb, ptr, i,
7527 btrfs_dev_stat_read(device, i));
7528 btrfs_mark_buffer_dirty(eb);
7531 btrfs_free_path(path);
7536 * called from commit_transaction. Writes all changed device stats to disk.
7538 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7540 struct btrfs_fs_info *fs_info = trans->fs_info;
7541 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7542 struct btrfs_device *device;
7546 mutex_lock(&fs_devices->device_list_mutex);
7547 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7548 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7549 if (!device->dev_stats_valid || stats_cnt == 0)
7554 * There is a LOAD-LOAD control dependency between the value of
7555 * dev_stats_ccnt and updating the on-disk values which requires
7556 * reading the in-memory counters. Such control dependencies
7557 * require explicit read memory barriers.
7559 * This memory barriers pairs with smp_mb__before_atomic in
7560 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7561 * barrier implied by atomic_xchg in
7562 * btrfs_dev_stats_read_and_reset
7566 ret = update_dev_stat_item(trans, device);
7568 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7570 mutex_unlock(&fs_devices->device_list_mutex);
7575 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7577 btrfs_dev_stat_inc(dev, index);
7579 if (!dev->dev_stats_valid)
7581 btrfs_err_rl_in_rcu(dev->fs_info,
7582 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7583 btrfs_dev_name(dev),
7584 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7585 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7586 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7587 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7588 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7591 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7595 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7596 if (btrfs_dev_stat_read(dev, i) != 0)
7598 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7599 return; /* all values == 0, suppress message */
7601 btrfs_info_in_rcu(dev->fs_info,
7602 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7603 btrfs_dev_name(dev),
7604 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7605 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7606 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7607 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7608 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7611 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7612 struct btrfs_ioctl_get_dev_stats *stats)
7614 BTRFS_DEV_LOOKUP_ARGS(args);
7615 struct btrfs_device *dev;
7616 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7619 mutex_lock(&fs_devices->device_list_mutex);
7620 args.devid = stats->devid;
7621 dev = btrfs_find_device(fs_info->fs_devices, &args);
7622 mutex_unlock(&fs_devices->device_list_mutex);
7625 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7627 } else if (!dev->dev_stats_valid) {
7628 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7630 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7631 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7632 if (stats->nr_items > i)
7634 btrfs_dev_stat_read_and_reset(dev, i);
7636 btrfs_dev_stat_set(dev, i, 0);
7638 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7639 current->comm, task_pid_nr(current));
7641 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7642 if (stats->nr_items > i)
7643 stats->values[i] = btrfs_dev_stat_read(dev, i);
7645 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7646 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7651 * Update the size and bytes used for each device where it changed. This is
7652 * delayed since we would otherwise get errors while writing out the
7655 * Must be invoked during transaction commit.
7657 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7659 struct btrfs_device *curr, *next;
7661 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7663 if (list_empty(&trans->dev_update_list))
7667 * We don't need the device_list_mutex here. This list is owned by the
7668 * transaction and the transaction must complete before the device is
7671 mutex_lock(&trans->fs_info->chunk_mutex);
7672 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7674 list_del_init(&curr->post_commit_list);
7675 curr->commit_total_bytes = curr->disk_total_bytes;
7676 curr->commit_bytes_used = curr->bytes_used;
7678 mutex_unlock(&trans->fs_info->chunk_mutex);
7682 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7684 int btrfs_bg_type_to_factor(u64 flags)
7686 const int index = btrfs_bg_flags_to_raid_index(flags);
7688 return btrfs_raid_array[index].ncopies;
7693 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7694 u64 chunk_offset, u64 devid,
7695 u64 physical_offset, u64 physical_len)
7697 struct btrfs_dev_lookup_args args = { .devid = devid };
7698 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7699 struct extent_map *em;
7700 struct map_lookup *map;
7701 struct btrfs_device *dev;
7707 read_lock(&em_tree->lock);
7708 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7709 read_unlock(&em_tree->lock);
7713 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7714 physical_offset, devid);
7719 map = em->map_lookup;
7720 stripe_len = btrfs_calc_stripe_length(em);
7721 if (physical_len != stripe_len) {
7723 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7724 physical_offset, devid, em->start, physical_len,
7731 * Very old mkfs.btrfs (before v4.1) will not respect the reserved
7732 * space. Although kernel can handle it without problem, better to warn
7735 if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED)
7737 "devid %llu physical %llu len %llu inside the reserved space",
7738 devid, physical_offset, physical_len);
7740 for (i = 0; i < map->num_stripes; i++) {
7741 if (map->stripes[i].dev->devid == devid &&
7742 map->stripes[i].physical == physical_offset) {
7744 if (map->verified_stripes >= map->num_stripes) {
7746 "too many dev extents for chunk %llu found",
7751 map->verified_stripes++;
7757 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7758 physical_offset, devid);
7762 /* Make sure no dev extent is beyond device boundary */
7763 dev = btrfs_find_device(fs_info->fs_devices, &args);
7765 btrfs_err(fs_info, "failed to find devid %llu", devid);
7770 if (physical_offset + physical_len > dev->disk_total_bytes) {
7772 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7773 devid, physical_offset, physical_len,
7774 dev->disk_total_bytes);
7779 if (dev->zone_info) {
7780 u64 zone_size = dev->zone_info->zone_size;
7782 if (!IS_ALIGNED(physical_offset, zone_size) ||
7783 !IS_ALIGNED(physical_len, zone_size)) {
7785 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
7786 devid, physical_offset, physical_len);
7793 free_extent_map(em);
7797 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7799 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7800 struct extent_map *em;
7801 struct rb_node *node;
7804 read_lock(&em_tree->lock);
7805 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
7806 em = rb_entry(node, struct extent_map, rb_node);
7807 if (em->map_lookup->num_stripes !=
7808 em->map_lookup->verified_stripes) {
7810 "chunk %llu has missing dev extent, have %d expect %d",
7811 em->start, em->map_lookup->verified_stripes,
7812 em->map_lookup->num_stripes);
7818 read_unlock(&em_tree->lock);
7823 * Ensure that all dev extents are mapped to correct chunk, otherwise
7824 * later chunk allocation/free would cause unexpected behavior.
7826 * NOTE: This will iterate through the whole device tree, which should be of
7827 * the same size level as the chunk tree. This slightly increases mount time.
7829 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
7831 struct btrfs_path *path;
7832 struct btrfs_root *root = fs_info->dev_root;
7833 struct btrfs_key key;
7835 u64 prev_dev_ext_end = 0;
7839 * We don't have a dev_root because we mounted with ignorebadroots and
7840 * failed to load the root, so we want to skip the verification in this
7843 * However if the dev root is fine, but the tree itself is corrupted
7844 * we'd still fail to mount. This verification is only to make sure
7845 * writes can happen safely, so instead just bypass this check
7846 * completely in the case of IGNOREBADROOTS.
7848 if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
7852 key.type = BTRFS_DEV_EXTENT_KEY;
7855 path = btrfs_alloc_path();
7859 path->reada = READA_FORWARD;
7860 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7864 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7865 ret = btrfs_next_leaf(root, path);
7868 /* No dev extents at all? Not good */
7875 struct extent_buffer *leaf = path->nodes[0];
7876 struct btrfs_dev_extent *dext;
7877 int slot = path->slots[0];
7879 u64 physical_offset;
7883 btrfs_item_key_to_cpu(leaf, &key, slot);
7884 if (key.type != BTRFS_DEV_EXTENT_KEY)
7886 devid = key.objectid;
7887 physical_offset = key.offset;
7889 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
7890 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
7891 physical_len = btrfs_dev_extent_length(leaf, dext);
7893 /* Check if this dev extent overlaps with the previous one */
7894 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
7896 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
7897 devid, physical_offset, prev_dev_ext_end);
7902 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
7903 physical_offset, physical_len);
7907 prev_dev_ext_end = physical_offset + physical_len;
7909 ret = btrfs_next_item(root, path);
7918 /* Ensure all chunks have corresponding dev extents */
7919 ret = verify_chunk_dev_extent_mapping(fs_info);
7921 btrfs_free_path(path);
7926 * Check whether the given block group or device is pinned by any inode being
7927 * used as a swapfile.
7929 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
7931 struct btrfs_swapfile_pin *sp;
7932 struct rb_node *node;
7934 spin_lock(&fs_info->swapfile_pins_lock);
7935 node = fs_info->swapfile_pins.rb_node;
7937 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
7939 node = node->rb_left;
7940 else if (ptr > sp->ptr)
7941 node = node->rb_right;
7945 spin_unlock(&fs_info->swapfile_pins_lock);
7946 return node != NULL;
7949 static int relocating_repair_kthread(void *data)
7951 struct btrfs_block_group *cache = data;
7952 struct btrfs_fs_info *fs_info = cache->fs_info;
7956 target = cache->start;
7957 btrfs_put_block_group(cache);
7959 sb_start_write(fs_info->sb);
7960 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
7962 "zoned: skip relocating block group %llu to repair: EBUSY",
7964 sb_end_write(fs_info->sb);
7968 mutex_lock(&fs_info->reclaim_bgs_lock);
7970 /* Ensure block group still exists */
7971 cache = btrfs_lookup_block_group(fs_info, target);
7975 if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags))
7978 ret = btrfs_may_alloc_data_chunk(fs_info, target);
7983 "zoned: relocating block group %llu to repair IO failure",
7985 ret = btrfs_relocate_chunk(fs_info, target);
7989 btrfs_put_block_group(cache);
7990 mutex_unlock(&fs_info->reclaim_bgs_lock);
7991 btrfs_exclop_finish(fs_info);
7992 sb_end_write(fs_info->sb);
7997 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical)
7999 struct btrfs_block_group *cache;
8001 if (!btrfs_is_zoned(fs_info))
8004 /* Do not attempt to repair in degraded state */
8005 if (btrfs_test_opt(fs_info, DEGRADED))
8008 cache = btrfs_lookup_block_group(fs_info, logical);
8012 if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) {
8013 btrfs_put_block_group(cache);
8017 kthread_run(relocating_repair_kthread, cache,
8018 "btrfs-relocating-repair");
8023 static void map_raid56_repair_block(struct btrfs_io_context *bioc,
8024 struct btrfs_io_stripe *smap,
8027 int data_stripes = nr_bioc_data_stripes(bioc);
8030 for (i = 0; i < data_stripes; i++) {
8031 u64 stripe_start = bioc->full_stripe_logical +
8032 (i << BTRFS_STRIPE_LEN_SHIFT);
8034 if (logical >= stripe_start &&
8035 logical < stripe_start + BTRFS_STRIPE_LEN)
8038 ASSERT(i < data_stripes);
8039 smap->dev = bioc->stripes[i].dev;
8040 smap->physical = bioc->stripes[i].physical +
8041 ((logical - bioc->full_stripe_logical) &
8042 BTRFS_STRIPE_LEN_MASK);
8046 * Map a repair write into a single device.
8048 * A repair write is triggered by read time repair or scrub, which would only
8049 * update the contents of a single device.
8050 * Not update any other mirrors nor go through RMW path.
8052 * Callers should ensure:
8054 * - Call btrfs_bio_counter_inc_blocked() first
8055 * - The range does not cross stripe boundary
8056 * - Has a valid @mirror_num passed in.
8058 int btrfs_map_repair_block(struct btrfs_fs_info *fs_info,
8059 struct btrfs_io_stripe *smap, u64 logical,
8060 u32 length, int mirror_num)
8062 struct btrfs_io_context *bioc = NULL;
8063 u64 map_length = length;
8064 int mirror_ret = mirror_num;
8067 ASSERT(mirror_num > 0);
8069 ret = __btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, &map_length,
8070 &bioc, smap, &mirror_ret, true);
8074 /* The map range should not cross stripe boundary. */
8075 ASSERT(map_length >= length);
8077 /* Already mapped to single stripe. */
8081 /* Map the RAID56 multi-stripe writes to a single one. */
8082 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
8083 map_raid56_repair_block(bioc, smap, logical);
8087 ASSERT(mirror_num <= bioc->num_stripes);
8088 smap->dev = bioc->stripes[mirror_num - 1].dev;
8089 smap->physical = bioc->stripes[mirror_num - 1].physical;
8091 btrfs_put_bioc(bioc);