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 extent_io_tree_release(&device->alloc_state);
399 btrfs_destroy_dev_zone_info(device);
403 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
405 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 constitutent 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 ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1153 ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1154 ASSERT(list_empty(&device->dev_alloc_list));
1155 ASSERT(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);
1187 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {
1188 close_fs_devices(fs_devices);
1189 list_del(&fs_devices->seed_list);
1190 free_fs_devices(fs_devices);
1192 mutex_unlock(&uuid_mutex);
1195 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1196 fmode_t flags, void *holder)
1198 struct btrfs_device *device;
1199 struct btrfs_device *latest_dev = NULL;
1200 struct btrfs_device *tmp_device;
1202 flags |= FMODE_EXCL;
1204 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,
1208 ret = btrfs_open_one_device(fs_devices, device, flags, holder);
1210 (!latest_dev || device->generation > latest_dev->generation)) {
1211 latest_dev = device;
1212 } else if (ret == -ENODATA) {
1213 fs_devices->num_devices--;
1214 list_del(&device->dev_list);
1215 btrfs_free_device(device);
1218 if (fs_devices->open_devices == 0)
1221 fs_devices->opened = 1;
1222 fs_devices->latest_dev = latest_dev;
1223 fs_devices->total_rw_bytes = 0;
1224 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1225 fs_devices->read_policy = BTRFS_READ_POLICY_PID;
1230 static int devid_cmp(void *priv, const struct list_head *a,
1231 const struct list_head *b)
1233 const struct btrfs_device *dev1, *dev2;
1235 dev1 = list_entry(a, struct btrfs_device, dev_list);
1236 dev2 = list_entry(b, struct btrfs_device, dev_list);
1238 if (dev1->devid < dev2->devid)
1240 else if (dev1->devid > dev2->devid)
1245 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1246 fmode_t flags, void *holder)
1250 lockdep_assert_held(&uuid_mutex);
1252 * The device_list_mutex cannot be taken here in case opening the
1253 * underlying device takes further locks like open_mutex.
1255 * We also don't need the lock here as this is called during mount and
1256 * exclusion is provided by uuid_mutex
1259 if (fs_devices->opened) {
1260 fs_devices->opened++;
1263 list_sort(NULL, &fs_devices->devices, devid_cmp);
1264 ret = open_fs_devices(fs_devices, flags, holder);
1270 void btrfs_release_disk_super(struct btrfs_super_block *super)
1272 struct page *page = virt_to_page(super);
1277 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1278 u64 bytenr, u64 bytenr_orig)
1280 struct btrfs_super_block *disk_super;
1285 /* make sure our super fits in the device */
1286 if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev))
1287 return ERR_PTR(-EINVAL);
1289 /* make sure our super fits in the page */
1290 if (sizeof(*disk_super) > PAGE_SIZE)
1291 return ERR_PTR(-EINVAL);
1293 /* make sure our super doesn't straddle pages on disk */
1294 index = bytenr >> PAGE_SHIFT;
1295 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1296 return ERR_PTR(-EINVAL);
1298 /* pull in the page with our super */
1299 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1302 return ERR_CAST(page);
1304 p = page_address(page);
1306 /* align our pointer to the offset of the super block */
1307 disk_super = p + offset_in_page(bytenr);
1309 if (btrfs_super_bytenr(disk_super) != bytenr_orig ||
1310 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1311 btrfs_release_disk_super(p);
1312 return ERR_PTR(-EINVAL);
1315 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1316 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1321 int btrfs_forget_devices(dev_t devt)
1325 mutex_lock(&uuid_mutex);
1326 ret = btrfs_free_stale_devices(devt, NULL);
1327 mutex_unlock(&uuid_mutex);
1333 * Look for a btrfs signature on a device. This may be called out of the mount path
1334 * and we are not allowed to call set_blocksize during the scan. The superblock
1335 * is read via pagecache
1337 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1340 struct btrfs_super_block *disk_super;
1341 bool new_device_added = false;
1342 struct btrfs_device *device = NULL;
1343 struct block_device *bdev;
1344 u64 bytenr, bytenr_orig;
1347 lockdep_assert_held(&uuid_mutex);
1350 * we would like to check all the supers, but that would make
1351 * a btrfs mount succeed after a mkfs from a different FS.
1352 * So, we need to add a special mount option to scan for
1353 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1355 flags |= FMODE_EXCL;
1357 bdev = blkdev_get_by_path(path, flags, holder);
1359 return ERR_CAST(bdev);
1361 bytenr_orig = btrfs_sb_offset(0);
1362 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr);
1364 device = ERR_PTR(ret);
1365 goto error_bdev_put;
1368 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig);
1369 if (IS_ERR(disk_super)) {
1370 device = ERR_CAST(disk_super);
1371 goto error_bdev_put;
1374 device = device_list_add(path, disk_super, &new_device_added);
1375 if (!IS_ERR(device) && new_device_added)
1376 btrfs_free_stale_devices(device->devt, device);
1378 btrfs_release_disk_super(disk_super);
1381 blkdev_put(bdev, flags);
1387 * Try to find a chunk that intersects [start, start + len] range and when one
1388 * such is found, record the end of it in *start
1390 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1393 u64 physical_start, physical_end;
1395 lockdep_assert_held(&device->fs_info->chunk_mutex);
1397 if (!find_first_extent_bit(&device->alloc_state, *start,
1398 &physical_start, &physical_end,
1399 CHUNK_ALLOCATED, NULL)) {
1401 if (in_range(physical_start, *start, len) ||
1402 in_range(*start, physical_start,
1403 physical_end - physical_start)) {
1404 *start = physical_end + 1;
1411 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start)
1413 switch (device->fs_devices->chunk_alloc_policy) {
1414 case BTRFS_CHUNK_ALLOC_REGULAR:
1415 return max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED);
1416 case BTRFS_CHUNK_ALLOC_ZONED:
1418 * We don't care about the starting region like regular
1419 * allocator, because we anyway use/reserve the first two zones
1420 * for superblock logging.
1422 return ALIGN(start, device->zone_info->zone_size);
1428 static bool dev_extent_hole_check_zoned(struct btrfs_device *device,
1429 u64 *hole_start, u64 *hole_size,
1432 u64 zone_size = device->zone_info->zone_size;
1435 bool changed = false;
1437 ASSERT(IS_ALIGNED(*hole_start, zone_size));
1439 while (*hole_size > 0) {
1440 pos = btrfs_find_allocatable_zones(device, *hole_start,
1441 *hole_start + *hole_size,
1443 if (pos != *hole_start) {
1444 *hole_size = *hole_start + *hole_size - pos;
1447 if (*hole_size < num_bytes)
1451 ret = btrfs_ensure_empty_zones(device, pos, num_bytes);
1453 /* Range is ensured to be empty */
1457 /* Given hole range was invalid (outside of device) */
1458 if (ret == -ERANGE) {
1459 *hole_start += *hole_size;
1464 *hole_start += zone_size;
1465 *hole_size -= zone_size;
1473 * Check if specified hole is suitable for allocation.
1475 * @device: the device which we have the hole
1476 * @hole_start: starting position of the hole
1477 * @hole_size: the size of the hole
1478 * @num_bytes: the size of the free space that we need
1480 * This function may modify @hole_start and @hole_size to reflect the suitable
1481 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1483 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1484 u64 *hole_size, u64 num_bytes)
1486 bool changed = false;
1487 u64 hole_end = *hole_start + *hole_size;
1491 * Check before we set max_hole_start, otherwise we could end up
1492 * sending back this offset anyway.
1494 if (contains_pending_extent(device, hole_start, *hole_size)) {
1495 if (hole_end >= *hole_start)
1496 *hole_size = hole_end - *hole_start;
1502 switch (device->fs_devices->chunk_alloc_policy) {
1503 case BTRFS_CHUNK_ALLOC_REGULAR:
1504 /* No extra check */
1506 case BTRFS_CHUNK_ALLOC_ZONED:
1507 if (dev_extent_hole_check_zoned(device, hole_start,
1508 hole_size, num_bytes)) {
1511 * The changed hole can contain pending extent.
1512 * Loop again to check that.
1528 * Find free space in the specified device.
1530 * @device: the device which we search the free space in
1531 * @num_bytes: the size of the free space that we need
1532 * @search_start: the position from which to begin the search
1533 * @start: store the start of the free space.
1534 * @len: the size of the free space. that we find, or the size
1535 * of the max free space if we don't find suitable free space
1537 * This does a pretty simple search, the expectation is that it is called very
1538 * infrequently and that a given device has a small number of extents.
1540 * @start is used to store the start of the free space if we find. But if we
1541 * don't find suitable free space, it will be used to store the start position
1542 * of the max free space.
1544 * @len is used to store the size of the free space that we find.
1545 * But if we don't find suitable free space, it is used to store the size of
1546 * the max free space.
1548 * NOTE: This function will search *commit* root of device tree, and does extra
1549 * check to ensure dev extents are not double allocated.
1550 * This makes the function safe to allocate dev extents but may not report
1551 * correct usable device space, as device extent freed in current transaction
1552 * is not reported as available.
1554 static int find_free_dev_extent_start(struct btrfs_device *device,
1555 u64 num_bytes, u64 search_start, u64 *start,
1558 struct btrfs_fs_info *fs_info = device->fs_info;
1559 struct btrfs_root *root = fs_info->dev_root;
1560 struct btrfs_key key;
1561 struct btrfs_dev_extent *dev_extent;
1562 struct btrfs_path *path;
1567 u64 search_end = device->total_bytes;
1570 struct extent_buffer *l;
1572 search_start = dev_extent_search_start(device, search_start);
1574 WARN_ON(device->zone_info &&
1575 !IS_ALIGNED(num_bytes, device->zone_info->zone_size));
1577 path = btrfs_alloc_path();
1581 max_hole_start = search_start;
1585 if (search_start >= search_end ||
1586 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1591 path->reada = READA_FORWARD;
1592 path->search_commit_root = 1;
1593 path->skip_locking = 1;
1595 key.objectid = device->devid;
1596 key.offset = search_start;
1597 key.type = BTRFS_DEV_EXTENT_KEY;
1599 ret = btrfs_search_backwards(root, &key, path);
1605 slot = path->slots[0];
1606 if (slot >= btrfs_header_nritems(l)) {
1607 ret = btrfs_next_leaf(root, path);
1615 btrfs_item_key_to_cpu(l, &key, slot);
1617 if (key.objectid < device->devid)
1620 if (key.objectid > device->devid)
1623 if (key.type != BTRFS_DEV_EXTENT_KEY)
1626 if (key.offset > search_start) {
1627 hole_size = key.offset - search_start;
1628 dev_extent_hole_check(device, &search_start, &hole_size,
1631 if (hole_size > max_hole_size) {
1632 max_hole_start = search_start;
1633 max_hole_size = hole_size;
1637 * If this free space is greater than which we need,
1638 * it must be the max free space that we have found
1639 * until now, so max_hole_start must point to the start
1640 * of this free space and the length of this free space
1641 * is stored in max_hole_size. Thus, we return
1642 * max_hole_start and max_hole_size and go back to the
1645 if (hole_size >= num_bytes) {
1651 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1652 extent_end = key.offset + btrfs_dev_extent_length(l,
1654 if (extent_end > search_start)
1655 search_start = extent_end;
1662 * At this point, search_start should be the end of
1663 * allocated dev extents, and when shrinking the device,
1664 * search_end may be smaller than search_start.
1666 if (search_end > search_start) {
1667 hole_size = search_end - search_start;
1668 if (dev_extent_hole_check(device, &search_start, &hole_size,
1670 btrfs_release_path(path);
1674 if (hole_size > max_hole_size) {
1675 max_hole_start = search_start;
1676 max_hole_size = hole_size;
1681 if (max_hole_size < num_bytes)
1687 btrfs_free_path(path);
1688 *start = max_hole_start;
1690 *len = max_hole_size;
1694 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1695 u64 *start, u64 *len)
1697 /* FIXME use last free of some kind */
1698 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1701 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1702 struct btrfs_device *device,
1703 u64 start, u64 *dev_extent_len)
1705 struct btrfs_fs_info *fs_info = device->fs_info;
1706 struct btrfs_root *root = fs_info->dev_root;
1708 struct btrfs_path *path;
1709 struct btrfs_key key;
1710 struct btrfs_key found_key;
1711 struct extent_buffer *leaf = NULL;
1712 struct btrfs_dev_extent *extent = NULL;
1714 path = btrfs_alloc_path();
1718 key.objectid = device->devid;
1720 key.type = BTRFS_DEV_EXTENT_KEY;
1722 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1724 ret = btrfs_previous_item(root, path, key.objectid,
1725 BTRFS_DEV_EXTENT_KEY);
1728 leaf = path->nodes[0];
1729 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1730 extent = btrfs_item_ptr(leaf, path->slots[0],
1731 struct btrfs_dev_extent);
1732 BUG_ON(found_key.offset > start || found_key.offset +
1733 btrfs_dev_extent_length(leaf, extent) < start);
1735 btrfs_release_path(path);
1737 } else if (ret == 0) {
1738 leaf = path->nodes[0];
1739 extent = btrfs_item_ptr(leaf, path->slots[0],
1740 struct btrfs_dev_extent);
1745 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1747 ret = btrfs_del_item(trans, root, path);
1749 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1751 btrfs_free_path(path);
1755 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1757 struct extent_map_tree *em_tree;
1758 struct extent_map *em;
1762 em_tree = &fs_info->mapping_tree;
1763 read_lock(&em_tree->lock);
1764 n = rb_last(&em_tree->map.rb_root);
1766 em = rb_entry(n, struct extent_map, rb_node);
1767 ret = em->start + em->len;
1769 read_unlock(&em_tree->lock);
1774 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1778 struct btrfs_key key;
1779 struct btrfs_key found_key;
1780 struct btrfs_path *path;
1782 path = btrfs_alloc_path();
1786 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1787 key.type = BTRFS_DEV_ITEM_KEY;
1788 key.offset = (u64)-1;
1790 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1796 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1801 ret = btrfs_previous_item(fs_info->chunk_root, path,
1802 BTRFS_DEV_ITEMS_OBJECTID,
1803 BTRFS_DEV_ITEM_KEY);
1807 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1809 *devid_ret = found_key.offset + 1;
1813 btrfs_free_path(path);
1818 * the device information is stored in the chunk root
1819 * the btrfs_device struct should be fully filled in
1821 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1822 struct btrfs_device *device)
1825 struct btrfs_path *path;
1826 struct btrfs_dev_item *dev_item;
1827 struct extent_buffer *leaf;
1828 struct btrfs_key key;
1831 path = btrfs_alloc_path();
1835 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1836 key.type = BTRFS_DEV_ITEM_KEY;
1837 key.offset = device->devid;
1839 btrfs_reserve_chunk_metadata(trans, true);
1840 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1841 &key, sizeof(*dev_item));
1842 btrfs_trans_release_chunk_metadata(trans);
1846 leaf = path->nodes[0];
1847 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1849 btrfs_set_device_id(leaf, dev_item, device->devid);
1850 btrfs_set_device_generation(leaf, dev_item, 0);
1851 btrfs_set_device_type(leaf, dev_item, device->type);
1852 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1853 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1854 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1855 btrfs_set_device_total_bytes(leaf, dev_item,
1856 btrfs_device_get_disk_total_bytes(device));
1857 btrfs_set_device_bytes_used(leaf, dev_item,
1858 btrfs_device_get_bytes_used(device));
1859 btrfs_set_device_group(leaf, dev_item, 0);
1860 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1861 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1862 btrfs_set_device_start_offset(leaf, dev_item, 0);
1864 ptr = btrfs_device_uuid(dev_item);
1865 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1866 ptr = btrfs_device_fsid(dev_item);
1867 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1868 ptr, BTRFS_FSID_SIZE);
1869 btrfs_mark_buffer_dirty(leaf);
1873 btrfs_free_path(path);
1878 * Function to update ctime/mtime for a given device path.
1879 * Mainly used for ctime/mtime based probe like libblkid.
1881 * We don't care about errors here, this is just to be kind to userspace.
1883 static void update_dev_time(const char *device_path)
1886 struct timespec64 now;
1889 ret = kern_path(device_path, LOOKUP_FOLLOW, &path);
1893 now = current_time(d_inode(path.dentry));
1894 inode_update_time(d_inode(path.dentry), &now, S_MTIME | S_CTIME);
1898 static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans,
1899 struct btrfs_device *device)
1901 struct btrfs_root *root = device->fs_info->chunk_root;
1903 struct btrfs_path *path;
1904 struct btrfs_key key;
1906 path = btrfs_alloc_path();
1910 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1911 key.type = BTRFS_DEV_ITEM_KEY;
1912 key.offset = device->devid;
1914 btrfs_reserve_chunk_metadata(trans, false);
1915 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1916 btrfs_trans_release_chunk_metadata(trans);
1923 ret = btrfs_del_item(trans, root, path);
1925 btrfs_free_path(path);
1930 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1931 * filesystem. It's up to the caller to adjust that number regarding eg. device
1934 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1942 seq = read_seqbegin(&fs_info->profiles_lock);
1944 all_avail = fs_info->avail_data_alloc_bits |
1945 fs_info->avail_system_alloc_bits |
1946 fs_info->avail_metadata_alloc_bits;
1947 } while (read_seqretry(&fs_info->profiles_lock, seq));
1949 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1950 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1953 if (num_devices < btrfs_raid_array[i].devs_min)
1954 return btrfs_raid_array[i].mindev_error;
1960 static struct btrfs_device * btrfs_find_next_active_device(
1961 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1963 struct btrfs_device *next_device;
1965 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1966 if (next_device != device &&
1967 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1968 && next_device->bdev)
1976 * Helper function to check if the given device is part of s_bdev / latest_dev
1977 * and replace it with the provided or the next active device, in the context
1978 * where this function called, there should be always be another device (or
1979 * this_dev) which is active.
1981 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
1982 struct btrfs_device *next_device)
1984 struct btrfs_fs_info *fs_info = device->fs_info;
1987 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1989 ASSERT(next_device);
1991 if (fs_info->sb->s_bdev &&
1992 (fs_info->sb->s_bdev == device->bdev))
1993 fs_info->sb->s_bdev = next_device->bdev;
1995 if (fs_info->fs_devices->latest_dev->bdev == device->bdev)
1996 fs_info->fs_devices->latest_dev = next_device;
2000 * Return btrfs_fs_devices::num_devices excluding the device that's being
2001 * currently replaced.
2003 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2005 u64 num_devices = fs_info->fs_devices->num_devices;
2007 down_read(&fs_info->dev_replace.rwsem);
2008 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2009 ASSERT(num_devices > 1);
2012 up_read(&fs_info->dev_replace.rwsem);
2017 static void btrfs_scratch_superblock(struct btrfs_fs_info *fs_info,
2018 struct block_device *bdev, int copy_num)
2020 struct btrfs_super_block *disk_super;
2024 disk_super = btrfs_read_dev_one_super(bdev, copy_num, false);
2025 if (IS_ERR(disk_super))
2028 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
2029 page = virt_to_page(disk_super);
2030 set_page_dirty(page);
2032 /* write_on_page() unlocks the page */
2033 ret = write_one_page(page);
2035 btrfs_warn(fs_info, "error clearing superblock number %d (%d)",
2037 btrfs_release_disk_super(disk_super);
2040 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
2041 struct block_device *bdev,
2042 const char *device_path)
2049 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2050 if (bdev_is_zoned(bdev))
2051 btrfs_reset_sb_log_zones(bdev, copy_num);
2053 btrfs_scratch_superblock(fs_info, bdev, copy_num);
2056 /* Notify udev that device has changed */
2057 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2059 /* Update ctime/mtime for device path for libblkid */
2060 update_dev_time(device_path);
2063 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
2064 struct btrfs_dev_lookup_args *args,
2065 struct block_device **bdev, fmode_t *mode)
2067 struct btrfs_trans_handle *trans;
2068 struct btrfs_device *device;
2069 struct btrfs_fs_devices *cur_devices;
2070 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2074 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2075 btrfs_err(fs_info, "device remove not supported on extent tree v2 yet");
2080 * The device list in fs_devices is accessed without locks (neither
2081 * uuid_mutex nor device_list_mutex) as it won't change on a mounted
2082 * filesystem and another device rm cannot run.
2084 num_devices = btrfs_num_devices(fs_info);
2086 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2090 device = btrfs_find_device(fs_info->fs_devices, args);
2093 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2099 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2100 btrfs_warn_in_rcu(fs_info,
2101 "cannot remove device %s (devid %llu) due to active swapfile",
2102 btrfs_dev_name(device), device->devid);
2106 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
2107 return BTRFS_ERROR_DEV_TGT_REPLACE;
2109 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2110 fs_info->fs_devices->rw_devices == 1)
2111 return BTRFS_ERROR_DEV_ONLY_WRITABLE;
2113 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2114 mutex_lock(&fs_info->chunk_mutex);
2115 list_del_init(&device->dev_alloc_list);
2116 device->fs_devices->rw_devices--;
2117 mutex_unlock(&fs_info->chunk_mutex);
2120 ret = btrfs_shrink_device(device, 0);
2124 trans = btrfs_start_transaction(fs_info->chunk_root, 0);
2125 if (IS_ERR(trans)) {
2126 ret = PTR_ERR(trans);
2130 ret = btrfs_rm_dev_item(trans, device);
2132 /* Any error in dev item removal is critical */
2134 "failed to remove device item for devid %llu: %d",
2135 device->devid, ret);
2136 btrfs_abort_transaction(trans, ret);
2137 btrfs_end_transaction(trans);
2141 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2142 btrfs_scrub_cancel_dev(device);
2145 * the device list mutex makes sure that we don't change
2146 * the device list while someone else is writing out all
2147 * the device supers. Whoever is writing all supers, should
2148 * lock the device list mutex before getting the number of
2149 * devices in the super block (super_copy). Conversely,
2150 * whoever updates the number of devices in the super block
2151 * (super_copy) should hold the device list mutex.
2155 * In normal cases the cur_devices == fs_devices. But in case
2156 * of deleting a seed device, the cur_devices should point to
2157 * its own fs_devices listed under the fs_devices->seed_list.
2159 cur_devices = device->fs_devices;
2160 mutex_lock(&fs_devices->device_list_mutex);
2161 list_del_rcu(&device->dev_list);
2163 cur_devices->num_devices--;
2164 cur_devices->total_devices--;
2165 /* Update total_devices of the parent fs_devices if it's seed */
2166 if (cur_devices != fs_devices)
2167 fs_devices->total_devices--;
2169 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2170 cur_devices->missing_devices--;
2172 btrfs_assign_next_active_device(device, NULL);
2175 cur_devices->open_devices--;
2176 /* remove sysfs entry */
2177 btrfs_sysfs_remove_device(device);
2180 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2181 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2182 mutex_unlock(&fs_devices->device_list_mutex);
2185 * At this point, the device is zero sized and detached from the
2186 * devices list. All that's left is to zero out the old supers and
2189 * We cannot call btrfs_close_bdev() here because we're holding the sb
2190 * write lock, and blkdev_put() will pull in the ->open_mutex on the
2191 * block device and it's dependencies. Instead just flush the device
2192 * and let the caller do the final blkdev_put.
2194 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2195 btrfs_scratch_superblocks(fs_info, device->bdev,
2198 sync_blockdev(device->bdev);
2199 invalidate_bdev(device->bdev);
2203 *bdev = device->bdev;
2204 *mode = device->mode;
2206 btrfs_free_device(device);
2209 * This can happen if cur_devices is the private seed devices list. We
2210 * cannot call close_fs_devices() here because it expects the uuid_mutex
2211 * to be held, but in fact we don't need that for the private
2212 * seed_devices, we can simply decrement cur_devices->opened and then
2213 * remove it from our list and free the fs_devices.
2215 if (cur_devices->num_devices == 0) {
2216 list_del_init(&cur_devices->seed_list);
2217 ASSERT(cur_devices->opened == 1);
2218 cur_devices->opened--;
2219 free_fs_devices(cur_devices);
2222 ret = btrfs_commit_transaction(trans);
2227 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2228 mutex_lock(&fs_info->chunk_mutex);
2229 list_add(&device->dev_alloc_list,
2230 &fs_devices->alloc_list);
2231 device->fs_devices->rw_devices++;
2232 mutex_unlock(&fs_info->chunk_mutex);
2237 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2239 struct btrfs_fs_devices *fs_devices;
2241 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2244 * in case of fs with no seed, srcdev->fs_devices will point
2245 * to fs_devices of fs_info. However when the dev being replaced is
2246 * a seed dev it will point to the seed's local fs_devices. In short
2247 * srcdev will have its correct fs_devices in both the cases.
2249 fs_devices = srcdev->fs_devices;
2251 list_del_rcu(&srcdev->dev_list);
2252 list_del(&srcdev->dev_alloc_list);
2253 fs_devices->num_devices--;
2254 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2255 fs_devices->missing_devices--;
2257 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2258 fs_devices->rw_devices--;
2261 fs_devices->open_devices--;
2264 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2266 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2268 mutex_lock(&uuid_mutex);
2270 btrfs_close_bdev(srcdev);
2272 btrfs_free_device(srcdev);
2274 /* if this is no devs we rather delete the fs_devices */
2275 if (!fs_devices->num_devices) {
2277 * On a mounted FS, num_devices can't be zero unless it's a
2278 * seed. In case of a seed device being replaced, the replace
2279 * target added to the sprout FS, so there will be no more
2280 * device left under the seed FS.
2282 ASSERT(fs_devices->seeding);
2284 list_del_init(&fs_devices->seed_list);
2285 close_fs_devices(fs_devices);
2286 free_fs_devices(fs_devices);
2288 mutex_unlock(&uuid_mutex);
2291 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2293 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2295 mutex_lock(&fs_devices->device_list_mutex);
2297 btrfs_sysfs_remove_device(tgtdev);
2300 fs_devices->open_devices--;
2302 fs_devices->num_devices--;
2304 btrfs_assign_next_active_device(tgtdev, NULL);
2306 list_del_rcu(&tgtdev->dev_list);
2308 mutex_unlock(&fs_devices->device_list_mutex);
2310 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2313 btrfs_close_bdev(tgtdev);
2315 btrfs_free_device(tgtdev);
2319 * Populate args from device at path.
2321 * @fs_info: the filesystem
2322 * @args: the args to populate
2323 * @path: the path to the device
2325 * This will read the super block of the device at @path and populate @args with
2326 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to
2327 * lookup a device to operate on, but need to do it before we take any locks.
2328 * This properly handles the special case of "missing" that a user may pass in,
2329 * and does some basic sanity checks. The caller must make sure that @path is
2330 * properly NUL terminated before calling in, and must call
2331 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and
2334 * Return: 0 for success, -errno for failure
2336 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,
2337 struct btrfs_dev_lookup_args *args,
2340 struct btrfs_super_block *disk_super;
2341 struct block_device *bdev;
2344 if (!path || !path[0])
2346 if (!strcmp(path, "missing")) {
2347 args->missing = true;
2351 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL);
2352 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL);
2353 if (!args->uuid || !args->fsid) {
2354 btrfs_put_dev_args_from_path(args);
2358 ret = btrfs_get_bdev_and_sb(path, FMODE_READ, fs_info->bdev_holder, 0,
2359 &bdev, &disk_super);
2361 btrfs_put_dev_args_from_path(args);
2365 args->devid = btrfs_stack_device_id(&disk_super->dev_item);
2366 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE);
2367 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2368 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE);
2370 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
2371 btrfs_release_disk_super(disk_super);
2372 blkdev_put(bdev, FMODE_READ);
2377 * Only use this jointly with btrfs_get_dev_args_from_path() because we will
2378 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables
2379 * that don't need to be freed.
2381 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args)
2389 struct btrfs_device *btrfs_find_device_by_devspec(
2390 struct btrfs_fs_info *fs_info, u64 devid,
2391 const char *device_path)
2393 BTRFS_DEV_LOOKUP_ARGS(args);
2394 struct btrfs_device *device;
2399 device = btrfs_find_device(fs_info->fs_devices, &args);
2401 return ERR_PTR(-ENOENT);
2405 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path);
2407 return ERR_PTR(ret);
2408 device = btrfs_find_device(fs_info->fs_devices, &args);
2409 btrfs_put_dev_args_from_path(&args);
2411 return ERR_PTR(-ENOENT);
2415 static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info)
2417 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2418 struct btrfs_fs_devices *old_devices;
2419 struct btrfs_fs_devices *seed_devices;
2421 lockdep_assert_held(&uuid_mutex);
2422 if (!fs_devices->seeding)
2423 return ERR_PTR(-EINVAL);
2426 * Private copy of the seed devices, anchored at
2427 * fs_info->fs_devices->seed_list
2429 seed_devices = alloc_fs_devices(NULL, NULL);
2430 if (IS_ERR(seed_devices))
2431 return seed_devices;
2434 * It's necessary to retain a copy of the original seed fs_devices in
2435 * fs_uuids so that filesystems which have been seeded can successfully
2436 * reference the seed device from open_seed_devices. This also supports
2439 old_devices = clone_fs_devices(fs_devices);
2440 if (IS_ERR(old_devices)) {
2441 kfree(seed_devices);
2445 list_add(&old_devices->fs_list, &fs_uuids);
2447 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2448 seed_devices->opened = 1;
2449 INIT_LIST_HEAD(&seed_devices->devices);
2450 INIT_LIST_HEAD(&seed_devices->alloc_list);
2451 mutex_init(&seed_devices->device_list_mutex);
2453 return seed_devices;
2457 * Splice seed devices into the sprout fs_devices.
2458 * Generate a new fsid for the sprouted read-write filesystem.
2460 static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info,
2461 struct btrfs_fs_devices *seed_devices)
2463 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2464 struct btrfs_super_block *disk_super = fs_info->super_copy;
2465 struct btrfs_device *device;
2469 * We are updating the fsid, the thread leading to device_list_add()
2470 * could race, so uuid_mutex is needed.
2472 lockdep_assert_held(&uuid_mutex);
2475 * The threads listed below may traverse dev_list but can do that without
2476 * device_list_mutex:
2477 * - All device ops and balance - as we are in btrfs_exclop_start.
2478 * - Various dev_list readers - are using RCU.
2479 * - btrfs_ioctl_fitrim() - is using RCU.
2481 * For-read threads as below are using device_list_mutex:
2482 * - Readonly scrub btrfs_scrub_dev()
2483 * - Readonly scrub btrfs_scrub_progress()
2484 * - btrfs_get_dev_stats()
2486 lockdep_assert_held(&fs_devices->device_list_mutex);
2488 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2490 list_for_each_entry(device, &seed_devices->devices, dev_list)
2491 device->fs_devices = seed_devices;
2493 fs_devices->seeding = false;
2494 fs_devices->num_devices = 0;
2495 fs_devices->open_devices = 0;
2496 fs_devices->missing_devices = 0;
2497 fs_devices->rotating = false;
2498 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2500 generate_random_uuid(fs_devices->fsid);
2501 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2502 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2504 super_flags = btrfs_super_flags(disk_super) &
2505 ~BTRFS_SUPER_FLAG_SEEDING;
2506 btrfs_set_super_flags(disk_super, super_flags);
2510 * Store the expected generation for seed devices in device items.
2512 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2514 BTRFS_DEV_LOOKUP_ARGS(args);
2515 struct btrfs_fs_info *fs_info = trans->fs_info;
2516 struct btrfs_root *root = fs_info->chunk_root;
2517 struct btrfs_path *path;
2518 struct extent_buffer *leaf;
2519 struct btrfs_dev_item *dev_item;
2520 struct btrfs_device *device;
2521 struct btrfs_key key;
2522 u8 fs_uuid[BTRFS_FSID_SIZE];
2523 u8 dev_uuid[BTRFS_UUID_SIZE];
2526 path = btrfs_alloc_path();
2530 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2532 key.type = BTRFS_DEV_ITEM_KEY;
2535 btrfs_reserve_chunk_metadata(trans, false);
2536 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2537 btrfs_trans_release_chunk_metadata(trans);
2541 leaf = path->nodes[0];
2543 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2544 ret = btrfs_next_leaf(root, path);
2549 leaf = path->nodes[0];
2550 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2551 btrfs_release_path(path);
2555 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2556 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2557 key.type != BTRFS_DEV_ITEM_KEY)
2560 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2561 struct btrfs_dev_item);
2562 args.devid = btrfs_device_id(leaf, dev_item);
2563 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2565 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2567 args.uuid = dev_uuid;
2568 args.fsid = fs_uuid;
2569 device = btrfs_find_device(fs_info->fs_devices, &args);
2570 BUG_ON(!device); /* Logic error */
2572 if (device->fs_devices->seeding) {
2573 btrfs_set_device_generation(leaf, dev_item,
2574 device->generation);
2575 btrfs_mark_buffer_dirty(leaf);
2583 btrfs_free_path(path);
2587 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2589 struct btrfs_root *root = fs_info->dev_root;
2590 struct btrfs_trans_handle *trans;
2591 struct btrfs_device *device;
2592 struct block_device *bdev;
2593 struct super_block *sb = fs_info->sb;
2594 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2595 struct btrfs_fs_devices *seed_devices;
2596 u64 orig_super_total_bytes;
2597 u64 orig_super_num_devices;
2599 bool seeding_dev = false;
2600 bool locked = false;
2602 if (sb_rdonly(sb) && !fs_devices->seeding)
2605 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2606 fs_info->bdev_holder);
2608 return PTR_ERR(bdev);
2610 if (!btrfs_check_device_zone_type(fs_info, bdev)) {
2615 if (fs_devices->seeding) {
2617 down_write(&sb->s_umount);
2618 mutex_lock(&uuid_mutex);
2622 sync_blockdev(bdev);
2625 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2626 if (device->bdev == bdev) {
2634 device = btrfs_alloc_device(fs_info, NULL, NULL, device_path);
2635 if (IS_ERR(device)) {
2636 /* we can safely leave the fs_devices entry around */
2637 ret = PTR_ERR(device);
2641 device->fs_info = fs_info;
2642 device->bdev = bdev;
2643 ret = lookup_bdev(device_path, &device->devt);
2645 goto error_free_device;
2647 ret = btrfs_get_dev_zone_info(device, false);
2649 goto error_free_device;
2651 trans = btrfs_start_transaction(root, 0);
2652 if (IS_ERR(trans)) {
2653 ret = PTR_ERR(trans);
2654 goto error_free_zone;
2657 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2658 device->generation = trans->transid;
2659 device->io_width = fs_info->sectorsize;
2660 device->io_align = fs_info->sectorsize;
2661 device->sector_size = fs_info->sectorsize;
2662 device->total_bytes =
2663 round_down(bdev_nr_bytes(bdev), fs_info->sectorsize);
2664 device->disk_total_bytes = device->total_bytes;
2665 device->commit_total_bytes = device->total_bytes;
2666 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2667 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2668 device->mode = FMODE_EXCL;
2669 device->dev_stats_valid = 1;
2670 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2673 btrfs_clear_sb_rdonly(sb);
2675 /* GFP_KERNEL allocation must not be under device_list_mutex */
2676 seed_devices = btrfs_init_sprout(fs_info);
2677 if (IS_ERR(seed_devices)) {
2678 ret = PTR_ERR(seed_devices);
2679 btrfs_abort_transaction(trans, ret);
2684 mutex_lock(&fs_devices->device_list_mutex);
2686 btrfs_setup_sprout(fs_info, seed_devices);
2687 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev,
2691 device->fs_devices = fs_devices;
2693 mutex_lock(&fs_info->chunk_mutex);
2694 list_add_rcu(&device->dev_list, &fs_devices->devices);
2695 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2696 fs_devices->num_devices++;
2697 fs_devices->open_devices++;
2698 fs_devices->rw_devices++;
2699 fs_devices->total_devices++;
2700 fs_devices->total_rw_bytes += device->total_bytes;
2702 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2704 if (!bdev_nonrot(bdev))
2705 fs_devices->rotating = true;
2707 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2708 btrfs_set_super_total_bytes(fs_info->super_copy,
2709 round_down(orig_super_total_bytes + device->total_bytes,
2710 fs_info->sectorsize));
2712 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2713 btrfs_set_super_num_devices(fs_info->super_copy,
2714 orig_super_num_devices + 1);
2717 * we've got more storage, clear any full flags on the space
2720 btrfs_clear_space_info_full(fs_info);
2722 mutex_unlock(&fs_info->chunk_mutex);
2724 /* Add sysfs device entry */
2725 btrfs_sysfs_add_device(device);
2727 mutex_unlock(&fs_devices->device_list_mutex);
2730 mutex_lock(&fs_info->chunk_mutex);
2731 ret = init_first_rw_device(trans);
2732 mutex_unlock(&fs_info->chunk_mutex);
2734 btrfs_abort_transaction(trans, ret);
2739 ret = btrfs_add_dev_item(trans, device);
2741 btrfs_abort_transaction(trans, ret);
2746 ret = btrfs_finish_sprout(trans);
2748 btrfs_abort_transaction(trans, ret);
2753 * fs_devices now represents the newly sprouted filesystem and
2754 * its fsid has been changed by btrfs_sprout_splice().
2756 btrfs_sysfs_update_sprout_fsid(fs_devices);
2759 ret = btrfs_commit_transaction(trans);
2762 mutex_unlock(&uuid_mutex);
2763 up_write(&sb->s_umount);
2766 if (ret) /* transaction commit */
2769 ret = btrfs_relocate_sys_chunks(fs_info);
2771 btrfs_handle_fs_error(fs_info, ret,
2772 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2773 trans = btrfs_attach_transaction(root);
2774 if (IS_ERR(trans)) {
2775 if (PTR_ERR(trans) == -ENOENT)
2777 ret = PTR_ERR(trans);
2781 ret = btrfs_commit_transaction(trans);
2785 * Now that we have written a new super block to this device, check all
2786 * other fs_devices list if device_path alienates any other scanned
2788 * We can ignore the return value as it typically returns -EINVAL and
2789 * only succeeds if the device was an alien.
2791 btrfs_forget_devices(device->devt);
2793 /* Update ctime/mtime for blkid or udev */
2794 update_dev_time(device_path);
2799 btrfs_sysfs_remove_device(device);
2800 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2801 mutex_lock(&fs_info->chunk_mutex);
2802 list_del_rcu(&device->dev_list);
2803 list_del(&device->dev_alloc_list);
2804 fs_info->fs_devices->num_devices--;
2805 fs_info->fs_devices->open_devices--;
2806 fs_info->fs_devices->rw_devices--;
2807 fs_info->fs_devices->total_devices--;
2808 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2809 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2810 btrfs_set_super_total_bytes(fs_info->super_copy,
2811 orig_super_total_bytes);
2812 btrfs_set_super_num_devices(fs_info->super_copy,
2813 orig_super_num_devices);
2814 mutex_unlock(&fs_info->chunk_mutex);
2815 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2818 btrfs_set_sb_rdonly(sb);
2820 btrfs_end_transaction(trans);
2822 btrfs_destroy_dev_zone_info(device);
2824 btrfs_free_device(device);
2826 blkdev_put(bdev, FMODE_EXCL);
2828 mutex_unlock(&uuid_mutex);
2829 up_write(&sb->s_umount);
2834 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2835 struct btrfs_device *device)
2838 struct btrfs_path *path;
2839 struct btrfs_root *root = device->fs_info->chunk_root;
2840 struct btrfs_dev_item *dev_item;
2841 struct extent_buffer *leaf;
2842 struct btrfs_key key;
2844 path = btrfs_alloc_path();
2848 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2849 key.type = BTRFS_DEV_ITEM_KEY;
2850 key.offset = device->devid;
2852 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2861 leaf = path->nodes[0];
2862 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2864 btrfs_set_device_id(leaf, dev_item, device->devid);
2865 btrfs_set_device_type(leaf, dev_item, device->type);
2866 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2867 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2868 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2869 btrfs_set_device_total_bytes(leaf, dev_item,
2870 btrfs_device_get_disk_total_bytes(device));
2871 btrfs_set_device_bytes_used(leaf, dev_item,
2872 btrfs_device_get_bytes_used(device));
2873 btrfs_mark_buffer_dirty(leaf);
2876 btrfs_free_path(path);
2880 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2881 struct btrfs_device *device, u64 new_size)
2883 struct btrfs_fs_info *fs_info = device->fs_info;
2884 struct btrfs_super_block *super_copy = fs_info->super_copy;
2889 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2892 new_size = round_down(new_size, fs_info->sectorsize);
2894 mutex_lock(&fs_info->chunk_mutex);
2895 old_total = btrfs_super_total_bytes(super_copy);
2896 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2898 if (new_size <= device->total_bytes ||
2899 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2900 mutex_unlock(&fs_info->chunk_mutex);
2904 btrfs_set_super_total_bytes(super_copy,
2905 round_down(old_total + diff, fs_info->sectorsize));
2906 device->fs_devices->total_rw_bytes += diff;
2908 btrfs_device_set_total_bytes(device, new_size);
2909 btrfs_device_set_disk_total_bytes(device, new_size);
2910 btrfs_clear_space_info_full(device->fs_info);
2911 if (list_empty(&device->post_commit_list))
2912 list_add_tail(&device->post_commit_list,
2913 &trans->transaction->dev_update_list);
2914 mutex_unlock(&fs_info->chunk_mutex);
2916 btrfs_reserve_chunk_metadata(trans, false);
2917 ret = btrfs_update_device(trans, device);
2918 btrfs_trans_release_chunk_metadata(trans);
2923 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2925 struct btrfs_fs_info *fs_info = trans->fs_info;
2926 struct btrfs_root *root = fs_info->chunk_root;
2928 struct btrfs_path *path;
2929 struct btrfs_key key;
2931 path = btrfs_alloc_path();
2935 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2936 key.offset = chunk_offset;
2937 key.type = BTRFS_CHUNK_ITEM_KEY;
2939 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2942 else if (ret > 0) { /* Logic error or corruption */
2943 btrfs_handle_fs_error(fs_info, -ENOENT,
2944 "Failed lookup while freeing chunk.");
2949 ret = btrfs_del_item(trans, root, path);
2951 btrfs_handle_fs_error(fs_info, ret,
2952 "Failed to delete chunk item.");
2954 btrfs_free_path(path);
2958 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2960 struct btrfs_super_block *super_copy = fs_info->super_copy;
2961 struct btrfs_disk_key *disk_key;
2962 struct btrfs_chunk *chunk;
2969 struct btrfs_key key;
2971 lockdep_assert_held(&fs_info->chunk_mutex);
2972 array_size = btrfs_super_sys_array_size(super_copy);
2974 ptr = super_copy->sys_chunk_array;
2977 while (cur < array_size) {
2978 disk_key = (struct btrfs_disk_key *)ptr;
2979 btrfs_disk_key_to_cpu(&key, disk_key);
2981 len = sizeof(*disk_key);
2983 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2984 chunk = (struct btrfs_chunk *)(ptr + len);
2985 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2986 len += btrfs_chunk_item_size(num_stripes);
2991 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2992 key.offset == chunk_offset) {
2993 memmove(ptr, ptr + len, array_size - (cur + len));
2995 btrfs_set_super_sys_array_size(super_copy, array_size);
3005 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
3006 * @logical: Logical block offset in bytes.
3007 * @length: Length of extent in bytes.
3009 * Return: Chunk mapping or ERR_PTR.
3011 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
3012 u64 logical, u64 length)
3014 struct extent_map_tree *em_tree;
3015 struct extent_map *em;
3017 em_tree = &fs_info->mapping_tree;
3018 read_lock(&em_tree->lock);
3019 em = lookup_extent_mapping(em_tree, logical, length);
3020 read_unlock(&em_tree->lock);
3023 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
3025 return ERR_PTR(-EINVAL);
3028 if (em->start > logical || em->start + em->len < logical) {
3030 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
3031 logical, length, em->start, em->start + em->len);
3032 free_extent_map(em);
3033 return ERR_PTR(-EINVAL);
3036 /* callers are responsible for dropping em's ref. */
3040 static int remove_chunk_item(struct btrfs_trans_handle *trans,
3041 struct map_lookup *map, u64 chunk_offset)
3046 * Removing chunk items and updating the device items in the chunks btree
3047 * requires holding the chunk_mutex.
3048 * See the comment at btrfs_chunk_alloc() for the details.
3050 lockdep_assert_held(&trans->fs_info->chunk_mutex);
3052 for (i = 0; i < map->num_stripes; i++) {
3055 ret = btrfs_update_device(trans, map->stripes[i].dev);
3060 return btrfs_free_chunk(trans, chunk_offset);
3063 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
3065 struct btrfs_fs_info *fs_info = trans->fs_info;
3066 struct extent_map *em;
3067 struct map_lookup *map;
3068 u64 dev_extent_len = 0;
3070 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3072 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
3075 * This is a logic error, but we don't want to just rely on the
3076 * user having built with ASSERT enabled, so if ASSERT doesn't
3077 * do anything we still error out.
3082 map = em->map_lookup;
3085 * First delete the device extent items from the devices btree.
3086 * We take the device_list_mutex to avoid racing with the finishing phase
3087 * of a device replace operation. See the comment below before acquiring
3088 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
3089 * because that can result in a deadlock when deleting the device extent
3090 * items from the devices btree - COWing an extent buffer from the btree
3091 * may result in allocating a new metadata chunk, which would attempt to
3092 * lock again fs_info->chunk_mutex.
3094 mutex_lock(&fs_devices->device_list_mutex);
3095 for (i = 0; i < map->num_stripes; i++) {
3096 struct btrfs_device *device = map->stripes[i].dev;
3097 ret = btrfs_free_dev_extent(trans, device,
3098 map->stripes[i].physical,
3101 mutex_unlock(&fs_devices->device_list_mutex);
3102 btrfs_abort_transaction(trans, ret);
3106 if (device->bytes_used > 0) {
3107 mutex_lock(&fs_info->chunk_mutex);
3108 btrfs_device_set_bytes_used(device,
3109 device->bytes_used - dev_extent_len);
3110 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3111 btrfs_clear_space_info_full(fs_info);
3112 mutex_unlock(&fs_info->chunk_mutex);
3115 mutex_unlock(&fs_devices->device_list_mutex);
3118 * We acquire fs_info->chunk_mutex for 2 reasons:
3120 * 1) Just like with the first phase of the chunk allocation, we must
3121 * reserve system space, do all chunk btree updates and deletions, and
3122 * update the system chunk array in the superblock while holding this
3123 * mutex. This is for similar reasons as explained on the comment at
3124 * the top of btrfs_chunk_alloc();
3126 * 2) Prevent races with the final phase of a device replace operation
3127 * that replaces the device object associated with the map's stripes,
3128 * because the device object's id can change at any time during that
3129 * final phase of the device replace operation
3130 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3131 * replaced device and then see it with an ID of
3132 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3133 * the device item, which does not exists on the chunk btree.
3134 * The finishing phase of device replace acquires both the
3135 * device_list_mutex and the chunk_mutex, in that order, so we are
3136 * safe by just acquiring the chunk_mutex.
3138 trans->removing_chunk = true;
3139 mutex_lock(&fs_info->chunk_mutex);
3141 check_system_chunk(trans, map->type);
3143 ret = remove_chunk_item(trans, map, chunk_offset);
3145 * Normally we should not get -ENOSPC since we reserved space before
3146 * through the call to check_system_chunk().
3148 * Despite our system space_info having enough free space, we may not
3149 * be able to allocate extents from its block groups, because all have
3150 * an incompatible profile, which will force us to allocate a new system
3151 * block group with the right profile, or right after we called
3152 * check_system_space() above, a scrub turned the only system block group
3153 * with enough free space into RO mode.
3154 * This is explained with more detail at do_chunk_alloc().
3156 * So if we get -ENOSPC, allocate a new system chunk and retry once.
3158 if (ret == -ENOSPC) {
3159 const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
3160 struct btrfs_block_group *sys_bg;
3162 sys_bg = btrfs_create_chunk(trans, sys_flags);
3163 if (IS_ERR(sys_bg)) {
3164 ret = PTR_ERR(sys_bg);
3165 btrfs_abort_transaction(trans, ret);
3169 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3171 btrfs_abort_transaction(trans, ret);
3175 ret = remove_chunk_item(trans, map, chunk_offset);
3177 btrfs_abort_transaction(trans, ret);
3181 btrfs_abort_transaction(trans, ret);
3185 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3187 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3188 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3190 btrfs_abort_transaction(trans, ret);
3195 mutex_unlock(&fs_info->chunk_mutex);
3196 trans->removing_chunk = false;
3199 * We are done with chunk btree updates and deletions, so release the
3200 * system space we previously reserved (with check_system_chunk()).
3202 btrfs_trans_release_chunk_metadata(trans);
3204 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3206 btrfs_abort_transaction(trans, ret);
3211 if (trans->removing_chunk) {
3212 mutex_unlock(&fs_info->chunk_mutex);
3213 trans->removing_chunk = false;
3216 free_extent_map(em);
3220 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3222 struct btrfs_root *root = fs_info->chunk_root;
3223 struct btrfs_trans_handle *trans;
3224 struct btrfs_block_group *block_group;
3228 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3230 "relocate: not supported on extent tree v2 yet");
3235 * Prevent races with automatic removal of unused block groups.
3236 * After we relocate and before we remove the chunk with offset
3237 * chunk_offset, automatic removal of the block group can kick in,
3238 * resulting in a failure when calling btrfs_remove_chunk() below.
3240 * Make sure to acquire this mutex before doing a tree search (dev
3241 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3242 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3243 * we release the path used to search the chunk/dev tree and before
3244 * the current task acquires this mutex and calls us.
3246 lockdep_assert_held(&fs_info->reclaim_bgs_lock);
3248 /* step one, relocate all the extents inside this chunk */
3249 btrfs_scrub_pause(fs_info);
3250 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3251 btrfs_scrub_continue(fs_info);
3255 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3258 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3259 length = block_group->length;
3260 btrfs_put_block_group(block_group);
3263 * On a zoned file system, discard the whole block group, this will
3264 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3265 * resetting the zone fails, don't treat it as a fatal problem from the
3266 * filesystem's point of view.
3268 if (btrfs_is_zoned(fs_info)) {
3269 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
3272 "failed to reset zone %llu after relocation",
3276 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3278 if (IS_ERR(trans)) {
3279 ret = PTR_ERR(trans);
3280 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3285 * step two, delete the device extents and the
3286 * chunk tree entries
3288 ret = btrfs_remove_chunk(trans, chunk_offset);
3289 btrfs_end_transaction(trans);
3293 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3295 struct btrfs_root *chunk_root = fs_info->chunk_root;
3296 struct btrfs_path *path;
3297 struct extent_buffer *leaf;
3298 struct btrfs_chunk *chunk;
3299 struct btrfs_key key;
3300 struct btrfs_key found_key;
3302 bool retried = false;
3306 path = btrfs_alloc_path();
3311 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3312 key.offset = (u64)-1;
3313 key.type = BTRFS_CHUNK_ITEM_KEY;
3316 mutex_lock(&fs_info->reclaim_bgs_lock);
3317 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3319 mutex_unlock(&fs_info->reclaim_bgs_lock);
3322 BUG_ON(ret == 0); /* Corruption */
3324 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3327 mutex_unlock(&fs_info->reclaim_bgs_lock);
3333 leaf = path->nodes[0];
3334 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3336 chunk = btrfs_item_ptr(leaf, path->slots[0],
3337 struct btrfs_chunk);
3338 chunk_type = btrfs_chunk_type(leaf, chunk);
3339 btrfs_release_path(path);
3341 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3342 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3348 mutex_unlock(&fs_info->reclaim_bgs_lock);
3350 if (found_key.offset == 0)
3352 key.offset = found_key.offset - 1;
3355 if (failed && !retried) {
3359 } else if (WARN_ON(failed && retried)) {
3363 btrfs_free_path(path);
3368 * return 1 : allocate a data chunk successfully,
3369 * return <0: errors during allocating a data chunk,
3370 * return 0 : no need to allocate a data chunk.
3372 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3375 struct btrfs_block_group *cache;
3379 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3381 chunk_type = cache->flags;
3382 btrfs_put_block_group(cache);
3384 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3387 spin_lock(&fs_info->data_sinfo->lock);
3388 bytes_used = fs_info->data_sinfo->bytes_used;
3389 spin_unlock(&fs_info->data_sinfo->lock);
3392 struct btrfs_trans_handle *trans;
3395 trans = btrfs_join_transaction(fs_info->tree_root);
3397 return PTR_ERR(trans);
3399 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3400 btrfs_end_transaction(trans);
3409 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3410 struct btrfs_balance_control *bctl)
3412 struct btrfs_root *root = fs_info->tree_root;
3413 struct btrfs_trans_handle *trans;
3414 struct btrfs_balance_item *item;
3415 struct btrfs_disk_balance_args disk_bargs;
3416 struct btrfs_path *path;
3417 struct extent_buffer *leaf;
3418 struct btrfs_key key;
3421 path = btrfs_alloc_path();
3425 trans = btrfs_start_transaction(root, 0);
3426 if (IS_ERR(trans)) {
3427 btrfs_free_path(path);
3428 return PTR_ERR(trans);
3431 key.objectid = BTRFS_BALANCE_OBJECTID;
3432 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3435 ret = btrfs_insert_empty_item(trans, root, path, &key,
3440 leaf = path->nodes[0];
3441 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3443 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3445 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3446 btrfs_set_balance_data(leaf, item, &disk_bargs);
3447 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3448 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3449 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3450 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3452 btrfs_set_balance_flags(leaf, item, bctl->flags);
3454 btrfs_mark_buffer_dirty(leaf);
3456 btrfs_free_path(path);
3457 err = btrfs_commit_transaction(trans);
3463 static int del_balance_item(struct btrfs_fs_info *fs_info)
3465 struct btrfs_root *root = fs_info->tree_root;
3466 struct btrfs_trans_handle *trans;
3467 struct btrfs_path *path;
3468 struct btrfs_key key;
3471 path = btrfs_alloc_path();
3475 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3476 if (IS_ERR(trans)) {
3477 btrfs_free_path(path);
3478 return PTR_ERR(trans);
3481 key.objectid = BTRFS_BALANCE_OBJECTID;
3482 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3485 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3493 ret = btrfs_del_item(trans, root, path);
3495 btrfs_free_path(path);
3496 err = btrfs_commit_transaction(trans);
3503 * This is a heuristic used to reduce the number of chunks balanced on
3504 * resume after balance was interrupted.
3506 static void update_balance_args(struct btrfs_balance_control *bctl)
3509 * Turn on soft mode for chunk types that were being converted.
3511 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3512 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3513 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3514 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3515 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3516 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3519 * Turn on usage filter if is not already used. The idea is
3520 * that chunks that we have already balanced should be
3521 * reasonably full. Don't do it for chunks that are being
3522 * converted - that will keep us from relocating unconverted
3523 * (albeit full) chunks.
3525 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3526 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3527 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3528 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3529 bctl->data.usage = 90;
3531 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3532 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3533 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3534 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3535 bctl->sys.usage = 90;
3537 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3538 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3539 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3540 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3541 bctl->meta.usage = 90;
3546 * Clear the balance status in fs_info and delete the balance item from disk.
3548 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3550 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3553 BUG_ON(!fs_info->balance_ctl);
3555 spin_lock(&fs_info->balance_lock);
3556 fs_info->balance_ctl = NULL;
3557 spin_unlock(&fs_info->balance_lock);
3560 ret = del_balance_item(fs_info);
3562 btrfs_handle_fs_error(fs_info, ret, NULL);
3566 * Balance filters. Return 1 if chunk should be filtered out
3567 * (should not be balanced).
3569 static int chunk_profiles_filter(u64 chunk_type,
3570 struct btrfs_balance_args *bargs)
3572 chunk_type = chunk_to_extended(chunk_type) &
3573 BTRFS_EXTENDED_PROFILE_MASK;
3575 if (bargs->profiles & chunk_type)
3581 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3582 struct btrfs_balance_args *bargs)
3584 struct btrfs_block_group *cache;
3586 u64 user_thresh_min;
3587 u64 user_thresh_max;
3590 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3591 chunk_used = cache->used;
3593 if (bargs->usage_min == 0)
3594 user_thresh_min = 0;
3596 user_thresh_min = mult_perc(cache->length, bargs->usage_min);
3598 if (bargs->usage_max == 0)
3599 user_thresh_max = 1;
3600 else if (bargs->usage_max > 100)
3601 user_thresh_max = cache->length;
3603 user_thresh_max = mult_perc(cache->length, bargs->usage_max);
3605 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3608 btrfs_put_block_group(cache);
3612 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3613 u64 chunk_offset, struct btrfs_balance_args *bargs)
3615 struct btrfs_block_group *cache;
3616 u64 chunk_used, user_thresh;
3619 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3620 chunk_used = cache->used;
3622 if (bargs->usage_min == 0)
3624 else if (bargs->usage > 100)
3625 user_thresh = cache->length;
3627 user_thresh = mult_perc(cache->length, bargs->usage);
3629 if (chunk_used < user_thresh)
3632 btrfs_put_block_group(cache);
3636 static int chunk_devid_filter(struct extent_buffer *leaf,
3637 struct btrfs_chunk *chunk,
3638 struct btrfs_balance_args *bargs)
3640 struct btrfs_stripe *stripe;
3641 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3644 for (i = 0; i < num_stripes; i++) {
3645 stripe = btrfs_stripe_nr(chunk, i);
3646 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3653 static u64 calc_data_stripes(u64 type, int num_stripes)
3655 const int index = btrfs_bg_flags_to_raid_index(type);
3656 const int ncopies = btrfs_raid_array[index].ncopies;
3657 const int nparity = btrfs_raid_array[index].nparity;
3659 return (num_stripes - nparity) / ncopies;
3662 /* [pstart, pend) */
3663 static int chunk_drange_filter(struct extent_buffer *leaf,
3664 struct btrfs_chunk *chunk,
3665 struct btrfs_balance_args *bargs)
3667 struct btrfs_stripe *stripe;
3668 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3675 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3678 type = btrfs_chunk_type(leaf, chunk);
3679 factor = calc_data_stripes(type, num_stripes);
3681 for (i = 0; i < num_stripes; i++) {
3682 stripe = btrfs_stripe_nr(chunk, i);
3683 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3686 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3687 stripe_length = btrfs_chunk_length(leaf, chunk);
3688 stripe_length = div_u64(stripe_length, factor);
3690 if (stripe_offset < bargs->pend &&
3691 stripe_offset + stripe_length > bargs->pstart)
3698 /* [vstart, vend) */
3699 static int chunk_vrange_filter(struct extent_buffer *leaf,
3700 struct btrfs_chunk *chunk,
3702 struct btrfs_balance_args *bargs)
3704 if (chunk_offset < bargs->vend &&
3705 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3706 /* at least part of the chunk is inside this vrange */
3712 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3713 struct btrfs_chunk *chunk,
3714 struct btrfs_balance_args *bargs)
3716 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3718 if (bargs->stripes_min <= num_stripes
3719 && num_stripes <= bargs->stripes_max)
3725 static int chunk_soft_convert_filter(u64 chunk_type,
3726 struct btrfs_balance_args *bargs)
3728 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3731 chunk_type = chunk_to_extended(chunk_type) &
3732 BTRFS_EXTENDED_PROFILE_MASK;
3734 if (bargs->target == chunk_type)
3740 static int should_balance_chunk(struct extent_buffer *leaf,
3741 struct btrfs_chunk *chunk, u64 chunk_offset)
3743 struct btrfs_fs_info *fs_info = leaf->fs_info;
3744 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3745 struct btrfs_balance_args *bargs = NULL;
3746 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3749 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3750 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3754 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3755 bargs = &bctl->data;
3756 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3758 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3759 bargs = &bctl->meta;
3761 /* profiles filter */
3762 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3763 chunk_profiles_filter(chunk_type, bargs)) {
3768 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3769 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3771 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3772 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3777 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3778 chunk_devid_filter(leaf, chunk, bargs)) {
3782 /* drange filter, makes sense only with devid filter */
3783 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3784 chunk_drange_filter(leaf, chunk, bargs)) {
3789 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3790 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3794 /* stripes filter */
3795 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3796 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3800 /* soft profile changing mode */
3801 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3802 chunk_soft_convert_filter(chunk_type, bargs)) {
3807 * limited by count, must be the last filter
3809 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3810 if (bargs->limit == 0)
3814 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3816 * Same logic as the 'limit' filter; the minimum cannot be
3817 * determined here because we do not have the global information
3818 * about the count of all chunks that satisfy the filters.
3820 if (bargs->limit_max == 0)
3829 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3831 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3832 struct btrfs_root *chunk_root = fs_info->chunk_root;
3834 struct btrfs_chunk *chunk;
3835 struct btrfs_path *path = NULL;
3836 struct btrfs_key key;
3837 struct btrfs_key found_key;
3838 struct extent_buffer *leaf;
3841 int enospc_errors = 0;
3842 bool counting = true;
3843 /* The single value limit and min/max limits use the same bytes in the */
3844 u64 limit_data = bctl->data.limit;
3845 u64 limit_meta = bctl->meta.limit;
3846 u64 limit_sys = bctl->sys.limit;
3850 int chunk_reserved = 0;
3852 path = btrfs_alloc_path();
3858 /* zero out stat counters */
3859 spin_lock(&fs_info->balance_lock);
3860 memset(&bctl->stat, 0, sizeof(bctl->stat));
3861 spin_unlock(&fs_info->balance_lock);
3865 * The single value limit and min/max limits use the same bytes
3868 bctl->data.limit = limit_data;
3869 bctl->meta.limit = limit_meta;
3870 bctl->sys.limit = limit_sys;
3872 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3873 key.offset = (u64)-1;
3874 key.type = BTRFS_CHUNK_ITEM_KEY;
3877 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3878 atomic_read(&fs_info->balance_cancel_req)) {
3883 mutex_lock(&fs_info->reclaim_bgs_lock);
3884 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3886 mutex_unlock(&fs_info->reclaim_bgs_lock);
3891 * this shouldn't happen, it means the last relocate
3895 BUG(); /* FIXME break ? */
3897 ret = btrfs_previous_item(chunk_root, path, 0,
3898 BTRFS_CHUNK_ITEM_KEY);
3900 mutex_unlock(&fs_info->reclaim_bgs_lock);
3905 leaf = path->nodes[0];
3906 slot = path->slots[0];
3907 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3909 if (found_key.objectid != key.objectid) {
3910 mutex_unlock(&fs_info->reclaim_bgs_lock);
3914 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3915 chunk_type = btrfs_chunk_type(leaf, chunk);
3918 spin_lock(&fs_info->balance_lock);
3919 bctl->stat.considered++;
3920 spin_unlock(&fs_info->balance_lock);
3923 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3925 btrfs_release_path(path);
3927 mutex_unlock(&fs_info->reclaim_bgs_lock);
3932 mutex_unlock(&fs_info->reclaim_bgs_lock);
3933 spin_lock(&fs_info->balance_lock);
3934 bctl->stat.expected++;
3935 spin_unlock(&fs_info->balance_lock);
3937 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3939 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3941 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3948 * Apply limit_min filter, no need to check if the LIMITS
3949 * filter is used, limit_min is 0 by default
3951 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3952 count_data < bctl->data.limit_min)
3953 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3954 count_meta < bctl->meta.limit_min)
3955 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3956 count_sys < bctl->sys.limit_min)) {
3957 mutex_unlock(&fs_info->reclaim_bgs_lock);
3961 if (!chunk_reserved) {
3963 * We may be relocating the only data chunk we have,
3964 * which could potentially end up with losing data's
3965 * raid profile, so lets allocate an empty one in
3968 ret = btrfs_may_alloc_data_chunk(fs_info,
3971 mutex_unlock(&fs_info->reclaim_bgs_lock);
3973 } else if (ret == 1) {
3978 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3979 mutex_unlock(&fs_info->reclaim_bgs_lock);
3980 if (ret == -ENOSPC) {
3982 } else if (ret == -ETXTBSY) {
3984 "skipping relocation of block group %llu due to active swapfile",
3990 spin_lock(&fs_info->balance_lock);
3991 bctl->stat.completed++;
3992 spin_unlock(&fs_info->balance_lock);
3995 if (found_key.offset == 0)
3997 key.offset = found_key.offset - 1;
4001 btrfs_release_path(path);
4006 btrfs_free_path(path);
4007 if (enospc_errors) {
4008 btrfs_info(fs_info, "%d enospc errors during balance",
4018 * See if a given profile is valid and reduced.
4020 * @flags: profile to validate
4021 * @extended: if true @flags is treated as an extended profile
4023 static int alloc_profile_is_valid(u64 flags, int extended)
4025 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
4026 BTRFS_BLOCK_GROUP_PROFILE_MASK);
4028 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
4030 /* 1) check that all other bits are zeroed */
4034 /* 2) see if profile is reduced */
4036 return !extended; /* "0" is valid for usual profiles */
4038 return has_single_bit_set(flags);
4041 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
4043 /* cancel requested || normal exit path */
4044 return atomic_read(&fs_info->balance_cancel_req) ||
4045 (atomic_read(&fs_info->balance_pause_req) == 0 &&
4046 atomic_read(&fs_info->balance_cancel_req) == 0);
4050 * Validate target profile against allowed profiles and return true if it's OK.
4051 * Otherwise print the error message and return false.
4053 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
4054 const struct btrfs_balance_args *bargs,
4055 u64 allowed, const char *type)
4057 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
4060 /* Profile is valid and does not have bits outside of the allowed set */
4061 if (alloc_profile_is_valid(bargs->target, 1) &&
4062 (bargs->target & ~allowed) == 0)
4065 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
4066 type, btrfs_bg_type_to_raid_name(bargs->target));
4071 * Fill @buf with textual description of balance filter flags @bargs, up to
4072 * @size_buf including the terminating null. The output may be trimmed if it
4073 * does not fit into the provided buffer.
4075 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
4079 u32 size_bp = size_buf;
4081 u64 flags = bargs->flags;
4082 char tmp_buf[128] = {'\0'};
4087 #define CHECK_APPEND_NOARG(a) \
4089 ret = snprintf(bp, size_bp, (a)); \
4090 if (ret < 0 || ret >= size_bp) \
4091 goto out_overflow; \
4096 #define CHECK_APPEND_1ARG(a, v1) \
4098 ret = snprintf(bp, size_bp, (a), (v1)); \
4099 if (ret < 0 || ret >= size_bp) \
4100 goto out_overflow; \
4105 #define CHECK_APPEND_2ARG(a, v1, v2) \
4107 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4108 if (ret < 0 || ret >= size_bp) \
4109 goto out_overflow; \
4114 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
4115 CHECK_APPEND_1ARG("convert=%s,",
4116 btrfs_bg_type_to_raid_name(bargs->target));
4118 if (flags & BTRFS_BALANCE_ARGS_SOFT)
4119 CHECK_APPEND_NOARG("soft,");
4121 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4122 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4124 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4127 if (flags & BTRFS_BALANCE_ARGS_USAGE)
4128 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4130 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4131 CHECK_APPEND_2ARG("usage=%u..%u,",
4132 bargs->usage_min, bargs->usage_max);
4134 if (flags & BTRFS_BALANCE_ARGS_DEVID)
4135 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4137 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4138 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4139 bargs->pstart, bargs->pend);
4141 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4142 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4143 bargs->vstart, bargs->vend);
4145 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4146 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4148 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4149 CHECK_APPEND_2ARG("limit=%u..%u,",
4150 bargs->limit_min, bargs->limit_max);
4152 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4153 CHECK_APPEND_2ARG("stripes=%u..%u,",
4154 bargs->stripes_min, bargs->stripes_max);
4156 #undef CHECK_APPEND_2ARG
4157 #undef CHECK_APPEND_1ARG
4158 #undef CHECK_APPEND_NOARG
4162 if (size_bp < size_buf)
4163 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4168 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4170 u32 size_buf = 1024;
4171 char tmp_buf[192] = {'\0'};
4174 u32 size_bp = size_buf;
4176 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4178 buf = kzalloc(size_buf, GFP_KERNEL);
4184 #define CHECK_APPEND_1ARG(a, v1) \
4186 ret = snprintf(bp, size_bp, (a), (v1)); \
4187 if (ret < 0 || ret >= size_bp) \
4188 goto out_overflow; \
4193 if (bctl->flags & BTRFS_BALANCE_FORCE)
4194 CHECK_APPEND_1ARG("%s", "-f ");
4196 if (bctl->flags & BTRFS_BALANCE_DATA) {
4197 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4198 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4201 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4202 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4203 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4206 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4207 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4208 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4211 #undef CHECK_APPEND_1ARG
4215 if (size_bp < size_buf)
4216 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4217 btrfs_info(fs_info, "balance: %s %s",
4218 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4219 "resume" : "start", buf);
4225 * Should be called with balance mutexe held
4227 int btrfs_balance(struct btrfs_fs_info *fs_info,
4228 struct btrfs_balance_control *bctl,
4229 struct btrfs_ioctl_balance_args *bargs)
4231 u64 meta_target, data_target;
4237 bool reducing_redundancy;
4240 if (btrfs_fs_closing(fs_info) ||
4241 atomic_read(&fs_info->balance_pause_req) ||
4242 btrfs_should_cancel_balance(fs_info)) {
4247 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4248 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4252 * In case of mixed groups both data and meta should be picked,
4253 * and identical options should be given for both of them.
4255 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4256 if (mixed && (bctl->flags & allowed)) {
4257 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4258 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4259 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4261 "balance: mixed groups data and metadata options must be the same");
4268 * rw_devices will not change at the moment, device add/delete/replace
4271 num_devices = fs_info->fs_devices->rw_devices;
4274 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4275 * special bit for it, to make it easier to distinguish. Thus we need
4276 * to set it manually, or balance would refuse the profile.
4278 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4279 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4280 if (num_devices >= btrfs_raid_array[i].devs_min)
4281 allowed |= btrfs_raid_array[i].bg_flag;
4283 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4284 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4285 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4291 * Allow to reduce metadata or system integrity only if force set for
4292 * profiles with redundancy (copies, parity)
4295 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4296 if (btrfs_raid_array[i].ncopies >= 2 ||
4297 btrfs_raid_array[i].tolerated_failures >= 1)
4298 allowed |= btrfs_raid_array[i].bg_flag;
4301 seq = read_seqbegin(&fs_info->profiles_lock);
4303 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4304 (fs_info->avail_system_alloc_bits & allowed) &&
4305 !(bctl->sys.target & allowed)) ||
4306 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4307 (fs_info->avail_metadata_alloc_bits & allowed) &&
4308 !(bctl->meta.target & allowed)))
4309 reducing_redundancy = true;
4311 reducing_redundancy = false;
4313 /* if we're not converting, the target field is uninitialized */
4314 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4315 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4316 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4317 bctl->data.target : fs_info->avail_data_alloc_bits;
4318 } while (read_seqretry(&fs_info->profiles_lock, seq));
4320 if (reducing_redundancy) {
4321 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4323 "balance: force reducing metadata redundancy");
4326 "balance: reduces metadata redundancy, use --force if you want this");
4332 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4333 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4335 "balance: metadata profile %s has lower redundancy than data profile %s",
4336 btrfs_bg_type_to_raid_name(meta_target),
4337 btrfs_bg_type_to_raid_name(data_target));
4340 ret = insert_balance_item(fs_info, bctl);
4341 if (ret && ret != -EEXIST)
4344 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4345 BUG_ON(ret == -EEXIST);
4346 BUG_ON(fs_info->balance_ctl);
4347 spin_lock(&fs_info->balance_lock);
4348 fs_info->balance_ctl = bctl;
4349 spin_unlock(&fs_info->balance_lock);
4351 BUG_ON(ret != -EEXIST);
4352 spin_lock(&fs_info->balance_lock);
4353 update_balance_args(bctl);
4354 spin_unlock(&fs_info->balance_lock);
4357 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4358 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4359 describe_balance_start_or_resume(fs_info);
4360 mutex_unlock(&fs_info->balance_mutex);
4362 ret = __btrfs_balance(fs_info);
4364 mutex_lock(&fs_info->balance_mutex);
4365 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) {
4366 btrfs_info(fs_info, "balance: paused");
4367 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
4370 * Balance can be canceled by:
4372 * - Regular cancel request
4373 * Then ret == -ECANCELED and balance_cancel_req > 0
4375 * - Fatal signal to "btrfs" process
4376 * Either the signal caught by wait_reserve_ticket() and callers
4377 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4379 * Either way, in this case balance_cancel_req = 0, and
4380 * ret == -EINTR or ret == -ECANCELED.
4382 * So here we only check the return value to catch canceled balance.
4384 else if (ret == -ECANCELED || ret == -EINTR)
4385 btrfs_info(fs_info, "balance: canceled");
4387 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4389 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4392 memset(bargs, 0, sizeof(*bargs));
4393 btrfs_update_ioctl_balance_args(fs_info, bargs);
4396 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4397 balance_need_close(fs_info)) {
4398 reset_balance_state(fs_info);
4399 btrfs_exclop_finish(fs_info);
4402 wake_up(&fs_info->balance_wait_q);
4406 if (bctl->flags & BTRFS_BALANCE_RESUME)
4407 reset_balance_state(fs_info);
4410 btrfs_exclop_finish(fs_info);
4415 static int balance_kthread(void *data)
4417 struct btrfs_fs_info *fs_info = data;
4420 sb_start_write(fs_info->sb);
4421 mutex_lock(&fs_info->balance_mutex);
4422 if (fs_info->balance_ctl)
4423 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4424 mutex_unlock(&fs_info->balance_mutex);
4425 sb_end_write(fs_info->sb);
4430 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4432 struct task_struct *tsk;
4434 mutex_lock(&fs_info->balance_mutex);
4435 if (!fs_info->balance_ctl) {
4436 mutex_unlock(&fs_info->balance_mutex);
4439 mutex_unlock(&fs_info->balance_mutex);
4441 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4442 btrfs_info(fs_info, "balance: resume skipped");
4446 spin_lock(&fs_info->super_lock);
4447 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
4448 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;
4449 spin_unlock(&fs_info->super_lock);
4451 * A ro->rw remount sequence should continue with the paused balance
4452 * regardless of who pauses it, system or the user as of now, so set
4455 spin_lock(&fs_info->balance_lock);
4456 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4457 spin_unlock(&fs_info->balance_lock);
4459 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4460 return PTR_ERR_OR_ZERO(tsk);
4463 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4465 struct btrfs_balance_control *bctl;
4466 struct btrfs_balance_item *item;
4467 struct btrfs_disk_balance_args disk_bargs;
4468 struct btrfs_path *path;
4469 struct extent_buffer *leaf;
4470 struct btrfs_key key;
4473 path = btrfs_alloc_path();
4477 key.objectid = BTRFS_BALANCE_OBJECTID;
4478 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4481 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4484 if (ret > 0) { /* ret = -ENOENT; */
4489 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4495 leaf = path->nodes[0];
4496 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4498 bctl->flags = btrfs_balance_flags(leaf, item);
4499 bctl->flags |= BTRFS_BALANCE_RESUME;
4501 btrfs_balance_data(leaf, item, &disk_bargs);
4502 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4503 btrfs_balance_meta(leaf, item, &disk_bargs);
4504 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4505 btrfs_balance_sys(leaf, item, &disk_bargs);
4506 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4509 * This should never happen, as the paused balance state is recovered
4510 * during mount without any chance of other exclusive ops to collide.
4512 * This gives the exclusive op status to balance and keeps in paused
4513 * state until user intervention (cancel or umount). If the ownership
4514 * cannot be assigned, show a message but do not fail. The balance
4515 * is in a paused state and must have fs_info::balance_ctl properly
4518 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED))
4520 "balance: cannot set exclusive op status, resume manually");
4522 btrfs_release_path(path);
4524 mutex_lock(&fs_info->balance_mutex);
4525 BUG_ON(fs_info->balance_ctl);
4526 spin_lock(&fs_info->balance_lock);
4527 fs_info->balance_ctl = bctl;
4528 spin_unlock(&fs_info->balance_lock);
4529 mutex_unlock(&fs_info->balance_mutex);
4531 btrfs_free_path(path);
4535 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4539 mutex_lock(&fs_info->balance_mutex);
4540 if (!fs_info->balance_ctl) {
4541 mutex_unlock(&fs_info->balance_mutex);
4545 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4546 atomic_inc(&fs_info->balance_pause_req);
4547 mutex_unlock(&fs_info->balance_mutex);
4549 wait_event(fs_info->balance_wait_q,
4550 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4552 mutex_lock(&fs_info->balance_mutex);
4553 /* we are good with balance_ctl ripped off from under us */
4554 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4555 atomic_dec(&fs_info->balance_pause_req);
4560 mutex_unlock(&fs_info->balance_mutex);
4564 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4566 mutex_lock(&fs_info->balance_mutex);
4567 if (!fs_info->balance_ctl) {
4568 mutex_unlock(&fs_info->balance_mutex);
4573 * A paused balance with the item stored on disk can be resumed at
4574 * mount time if the mount is read-write. Otherwise it's still paused
4575 * and we must not allow cancelling as it deletes the item.
4577 if (sb_rdonly(fs_info->sb)) {
4578 mutex_unlock(&fs_info->balance_mutex);
4582 atomic_inc(&fs_info->balance_cancel_req);
4584 * if we are running just wait and return, balance item is
4585 * deleted in btrfs_balance in this case
4587 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4588 mutex_unlock(&fs_info->balance_mutex);
4589 wait_event(fs_info->balance_wait_q,
4590 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4591 mutex_lock(&fs_info->balance_mutex);
4593 mutex_unlock(&fs_info->balance_mutex);
4595 * Lock released to allow other waiters to continue, we'll
4596 * reexamine the status again.
4598 mutex_lock(&fs_info->balance_mutex);
4600 if (fs_info->balance_ctl) {
4601 reset_balance_state(fs_info);
4602 btrfs_exclop_finish(fs_info);
4603 btrfs_info(fs_info, "balance: canceled");
4607 BUG_ON(fs_info->balance_ctl ||
4608 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4609 atomic_dec(&fs_info->balance_cancel_req);
4610 mutex_unlock(&fs_info->balance_mutex);
4614 int btrfs_uuid_scan_kthread(void *data)
4616 struct btrfs_fs_info *fs_info = data;
4617 struct btrfs_root *root = fs_info->tree_root;
4618 struct btrfs_key key;
4619 struct btrfs_path *path = NULL;
4621 struct extent_buffer *eb;
4623 struct btrfs_root_item root_item;
4625 struct btrfs_trans_handle *trans = NULL;
4626 bool closing = false;
4628 path = btrfs_alloc_path();
4635 key.type = BTRFS_ROOT_ITEM_KEY;
4639 if (btrfs_fs_closing(fs_info)) {
4643 ret = btrfs_search_forward(root, &key, path,
4644 BTRFS_OLDEST_GENERATION);
4651 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4652 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4653 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4654 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4657 eb = path->nodes[0];
4658 slot = path->slots[0];
4659 item_size = btrfs_item_size(eb, slot);
4660 if (item_size < sizeof(root_item))
4663 read_extent_buffer(eb, &root_item,
4664 btrfs_item_ptr_offset(eb, slot),
4665 (int)sizeof(root_item));
4666 if (btrfs_root_refs(&root_item) == 0)
4669 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4670 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4674 btrfs_release_path(path);
4676 * 1 - subvol uuid item
4677 * 1 - received_subvol uuid item
4679 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4680 if (IS_ERR(trans)) {
4681 ret = PTR_ERR(trans);
4689 btrfs_release_path(path);
4690 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4691 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4692 BTRFS_UUID_KEY_SUBVOL,
4695 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4701 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4702 ret = btrfs_uuid_tree_add(trans,
4703 root_item.received_uuid,
4704 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4707 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4714 btrfs_release_path(path);
4716 ret = btrfs_end_transaction(trans);
4722 if (key.offset < (u64)-1) {
4724 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4726 key.type = BTRFS_ROOT_ITEM_KEY;
4727 } else if (key.objectid < (u64)-1) {
4729 key.type = BTRFS_ROOT_ITEM_KEY;
4738 btrfs_free_path(path);
4739 if (trans && !IS_ERR(trans))
4740 btrfs_end_transaction(trans);
4742 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4744 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4745 up(&fs_info->uuid_tree_rescan_sem);
4749 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4751 struct btrfs_trans_handle *trans;
4752 struct btrfs_root *tree_root = fs_info->tree_root;
4753 struct btrfs_root *uuid_root;
4754 struct task_struct *task;
4761 trans = btrfs_start_transaction(tree_root, 2);
4763 return PTR_ERR(trans);
4765 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4766 if (IS_ERR(uuid_root)) {
4767 ret = PTR_ERR(uuid_root);
4768 btrfs_abort_transaction(trans, ret);
4769 btrfs_end_transaction(trans);
4773 fs_info->uuid_root = uuid_root;
4775 ret = btrfs_commit_transaction(trans);
4779 down(&fs_info->uuid_tree_rescan_sem);
4780 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4782 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4783 btrfs_warn(fs_info, "failed to start uuid_scan task");
4784 up(&fs_info->uuid_tree_rescan_sem);
4785 return PTR_ERR(task);
4792 * shrinking a device means finding all of the device extents past
4793 * the new size, and then following the back refs to the chunks.
4794 * The chunk relocation code actually frees the device extent
4796 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4798 struct btrfs_fs_info *fs_info = device->fs_info;
4799 struct btrfs_root *root = fs_info->dev_root;
4800 struct btrfs_trans_handle *trans;
4801 struct btrfs_dev_extent *dev_extent = NULL;
4802 struct btrfs_path *path;
4808 bool retried = false;
4809 struct extent_buffer *l;
4810 struct btrfs_key key;
4811 struct btrfs_super_block *super_copy = fs_info->super_copy;
4812 u64 old_total = btrfs_super_total_bytes(super_copy);
4813 u64 old_size = btrfs_device_get_total_bytes(device);
4817 new_size = round_down(new_size, fs_info->sectorsize);
4819 diff = round_down(old_size - new_size, fs_info->sectorsize);
4821 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4824 path = btrfs_alloc_path();
4828 path->reada = READA_BACK;
4830 trans = btrfs_start_transaction(root, 0);
4831 if (IS_ERR(trans)) {
4832 btrfs_free_path(path);
4833 return PTR_ERR(trans);
4836 mutex_lock(&fs_info->chunk_mutex);
4838 btrfs_device_set_total_bytes(device, new_size);
4839 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4840 device->fs_devices->total_rw_bytes -= diff;
4841 atomic64_sub(diff, &fs_info->free_chunk_space);
4845 * Once the device's size has been set to the new size, ensure all
4846 * in-memory chunks are synced to disk so that the loop below sees them
4847 * and relocates them accordingly.
4849 if (contains_pending_extent(device, &start, diff)) {
4850 mutex_unlock(&fs_info->chunk_mutex);
4851 ret = btrfs_commit_transaction(trans);
4855 mutex_unlock(&fs_info->chunk_mutex);
4856 btrfs_end_transaction(trans);
4860 key.objectid = device->devid;
4861 key.offset = (u64)-1;
4862 key.type = BTRFS_DEV_EXTENT_KEY;
4865 mutex_lock(&fs_info->reclaim_bgs_lock);
4866 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4868 mutex_unlock(&fs_info->reclaim_bgs_lock);
4872 ret = btrfs_previous_item(root, path, 0, key.type);
4874 mutex_unlock(&fs_info->reclaim_bgs_lock);
4878 btrfs_release_path(path);
4883 slot = path->slots[0];
4884 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4886 if (key.objectid != device->devid) {
4887 mutex_unlock(&fs_info->reclaim_bgs_lock);
4888 btrfs_release_path(path);
4892 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4893 length = btrfs_dev_extent_length(l, dev_extent);
4895 if (key.offset + length <= new_size) {
4896 mutex_unlock(&fs_info->reclaim_bgs_lock);
4897 btrfs_release_path(path);
4901 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4902 btrfs_release_path(path);
4905 * We may be relocating the only data chunk we have,
4906 * which could potentially end up with losing data's
4907 * raid profile, so lets allocate an empty one in
4910 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4912 mutex_unlock(&fs_info->reclaim_bgs_lock);
4916 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4917 mutex_unlock(&fs_info->reclaim_bgs_lock);
4918 if (ret == -ENOSPC) {
4921 if (ret == -ETXTBSY) {
4923 "could not shrink block group %llu due to active swapfile",
4928 } while (key.offset-- > 0);
4930 if (failed && !retried) {
4934 } else if (failed && retried) {
4939 /* Shrinking succeeded, else we would be at "done". */
4940 trans = btrfs_start_transaction(root, 0);
4941 if (IS_ERR(trans)) {
4942 ret = PTR_ERR(trans);
4946 mutex_lock(&fs_info->chunk_mutex);
4947 /* Clear all state bits beyond the shrunk device size */
4948 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
4951 btrfs_device_set_disk_total_bytes(device, new_size);
4952 if (list_empty(&device->post_commit_list))
4953 list_add_tail(&device->post_commit_list,
4954 &trans->transaction->dev_update_list);
4956 WARN_ON(diff > old_total);
4957 btrfs_set_super_total_bytes(super_copy,
4958 round_down(old_total - diff, fs_info->sectorsize));
4959 mutex_unlock(&fs_info->chunk_mutex);
4961 btrfs_reserve_chunk_metadata(trans, false);
4962 /* Now btrfs_update_device() will change the on-disk size. */
4963 ret = btrfs_update_device(trans, device);
4964 btrfs_trans_release_chunk_metadata(trans);
4966 btrfs_abort_transaction(trans, ret);
4967 btrfs_end_transaction(trans);
4969 ret = btrfs_commit_transaction(trans);
4972 btrfs_free_path(path);
4974 mutex_lock(&fs_info->chunk_mutex);
4975 btrfs_device_set_total_bytes(device, old_size);
4976 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4977 device->fs_devices->total_rw_bytes += diff;
4978 atomic64_add(diff, &fs_info->free_chunk_space);
4979 mutex_unlock(&fs_info->chunk_mutex);
4984 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4985 struct btrfs_key *key,
4986 struct btrfs_chunk *chunk, int item_size)
4988 struct btrfs_super_block *super_copy = fs_info->super_copy;
4989 struct btrfs_disk_key disk_key;
4993 lockdep_assert_held(&fs_info->chunk_mutex);
4995 array_size = btrfs_super_sys_array_size(super_copy);
4996 if (array_size + item_size + sizeof(disk_key)
4997 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
5000 ptr = super_copy->sys_chunk_array + array_size;
5001 btrfs_cpu_key_to_disk(&disk_key, key);
5002 memcpy(ptr, &disk_key, sizeof(disk_key));
5003 ptr += sizeof(disk_key);
5004 memcpy(ptr, chunk, item_size);
5005 item_size += sizeof(disk_key);
5006 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
5012 * sort the devices in descending order by max_avail, total_avail
5014 static int btrfs_cmp_device_info(const void *a, const void *b)
5016 const struct btrfs_device_info *di_a = a;
5017 const struct btrfs_device_info *di_b = b;
5019 if (di_a->max_avail > di_b->max_avail)
5021 if (di_a->max_avail < di_b->max_avail)
5023 if (di_a->total_avail > di_b->total_avail)
5025 if (di_a->total_avail < di_b->total_avail)
5030 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
5032 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5035 btrfs_set_fs_incompat(info, RAID56);
5038 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
5040 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
5043 btrfs_set_fs_incompat(info, RAID1C34);
5047 * Structure used internally for btrfs_create_chunk() function.
5048 * Wraps needed parameters.
5050 struct alloc_chunk_ctl {
5053 /* Total number of stripes to allocate */
5055 /* sub_stripes info for map */
5057 /* Stripes per device */
5059 /* Maximum number of devices to use */
5061 /* Minimum number of devices to use */
5063 /* ndevs has to be a multiple of this */
5065 /* Number of copies */
5067 /* Number of stripes worth of bytes to store parity information */
5069 u64 max_stripe_size;
5077 static void init_alloc_chunk_ctl_policy_regular(
5078 struct btrfs_fs_devices *fs_devices,
5079 struct alloc_chunk_ctl *ctl)
5081 struct btrfs_space_info *space_info;
5083 space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type);
5086 ctl->max_chunk_size = READ_ONCE(space_info->chunk_size);
5087 ctl->max_stripe_size = ctl->max_chunk_size;
5089 if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM)
5090 ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK);
5092 /* We don't want a chunk larger than 10% of writable space */
5093 ctl->max_chunk_size = min(mult_perc(fs_devices->total_rw_bytes, 10),
5094 ctl->max_chunk_size);
5095 ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes;
5098 static void init_alloc_chunk_ctl_policy_zoned(
5099 struct btrfs_fs_devices *fs_devices,
5100 struct alloc_chunk_ctl *ctl)
5102 u64 zone_size = fs_devices->fs_info->zone_size;
5104 int min_num_stripes = ctl->devs_min * ctl->dev_stripes;
5105 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;
5106 u64 min_chunk_size = min_data_stripes * zone_size;
5107 u64 type = ctl->type;
5109 ctl->max_stripe_size = zone_size;
5110 if (type & BTRFS_BLOCK_GROUP_DATA) {
5111 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,
5113 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5114 ctl->max_chunk_size = ctl->max_stripe_size;
5115 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5116 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5117 ctl->devs_max = min_t(int, ctl->devs_max,
5118 BTRFS_MAX_DEVS_SYS_CHUNK);
5123 /* We don't want a chunk larger than 10% of writable space */
5124 limit = max(round_down(mult_perc(fs_devices->total_rw_bytes, 10),
5127 ctl->max_chunk_size = min(limit, ctl->max_chunk_size);
5128 ctl->dev_extent_min = zone_size * ctl->dev_stripes;
5131 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
5132 struct alloc_chunk_ctl *ctl)
5134 int index = btrfs_bg_flags_to_raid_index(ctl->type);
5136 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
5137 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
5138 ctl->devs_max = btrfs_raid_array[index].devs_max;
5140 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
5141 ctl->devs_min = btrfs_raid_array[index].devs_min;
5142 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
5143 ctl->ncopies = btrfs_raid_array[index].ncopies;
5144 ctl->nparity = btrfs_raid_array[index].nparity;
5147 switch (fs_devices->chunk_alloc_policy) {
5148 case BTRFS_CHUNK_ALLOC_REGULAR:
5149 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
5151 case BTRFS_CHUNK_ALLOC_ZONED:
5152 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);
5159 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
5160 struct alloc_chunk_ctl *ctl,
5161 struct btrfs_device_info *devices_info)
5163 struct btrfs_fs_info *info = fs_devices->fs_info;
5164 struct btrfs_device *device;
5166 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
5173 * in the first pass through the devices list, we gather information
5174 * about the available holes on each device.
5176 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5177 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5179 "BTRFS: read-only device in alloc_list\n");
5183 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5184 &device->dev_state) ||
5185 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5188 if (device->total_bytes > device->bytes_used)
5189 total_avail = device->total_bytes - device->bytes_used;
5193 /* If there is no space on this device, skip it. */
5194 if (total_avail < ctl->dev_extent_min)
5197 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
5199 if (ret && ret != -ENOSPC)
5203 max_avail = dev_extent_want;
5205 if (max_avail < ctl->dev_extent_min) {
5206 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5208 "%s: devid %llu has no free space, have=%llu want=%llu",
5209 __func__, device->devid, max_avail,
5210 ctl->dev_extent_min);
5214 if (ndevs == fs_devices->rw_devices) {
5215 WARN(1, "%s: found more than %llu devices\n",
5216 __func__, fs_devices->rw_devices);
5219 devices_info[ndevs].dev_offset = dev_offset;
5220 devices_info[ndevs].max_avail = max_avail;
5221 devices_info[ndevs].total_avail = total_avail;
5222 devices_info[ndevs].dev = device;
5228 * now sort the devices by hole size / available space
5230 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5231 btrfs_cmp_device_info, NULL);
5236 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5237 struct btrfs_device_info *devices_info)
5239 /* Number of stripes that count for block group size */
5243 * The primary goal is to maximize the number of stripes, so use as
5244 * many devices as possible, even if the stripes are not maximum sized.
5246 * The DUP profile stores more than one stripe per device, the
5247 * max_avail is the total size so we have to adjust.
5249 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5251 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5253 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5254 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5257 * Use the number of data stripes to figure out how big this chunk is
5258 * really going to be in terms of logical address space, and compare
5259 * that answer with the max chunk size. If it's higher, we try to
5260 * reduce stripe_size.
5262 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5264 * Reduce stripe_size, round it up to a 16MB boundary again and
5265 * then use it, unless it ends up being even bigger than the
5266 * previous value we had already.
5268 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5269 data_stripes), SZ_16M),
5273 /* Stripe size should not go beyond 1G. */
5274 ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G);
5276 /* Align to BTRFS_STRIPE_LEN */
5277 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5278 ctl->chunk_size = ctl->stripe_size * data_stripes;
5283 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,
5284 struct btrfs_device_info *devices_info)
5286 u64 zone_size = devices_info[0].dev->zone_info->zone_size;
5287 /* Number of stripes that count for block group size */
5291 * It should hold because:
5292 * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5294 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);
5296 ctl->stripe_size = zone_size;
5297 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5298 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5300 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5301 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5302 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,
5303 ctl->stripe_size) + ctl->nparity,
5305 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5306 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5307 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);
5310 ctl->chunk_size = ctl->stripe_size * data_stripes;
5315 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5316 struct alloc_chunk_ctl *ctl,
5317 struct btrfs_device_info *devices_info)
5319 struct btrfs_fs_info *info = fs_devices->fs_info;
5322 * Round down to number of usable stripes, devs_increment can be any
5323 * number so we can't use round_down() that requires power of 2, while
5324 * rounddown is safe.
5326 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5328 if (ctl->ndevs < ctl->devs_min) {
5329 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5331 "%s: not enough devices with free space: have=%d minimum required=%d",
5332 __func__, ctl->ndevs, ctl->devs_min);
5337 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5339 switch (fs_devices->chunk_alloc_policy) {
5340 case BTRFS_CHUNK_ALLOC_REGULAR:
5341 return decide_stripe_size_regular(ctl, devices_info);
5342 case BTRFS_CHUNK_ALLOC_ZONED:
5343 return decide_stripe_size_zoned(ctl, devices_info);
5349 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
5350 struct alloc_chunk_ctl *ctl,
5351 struct btrfs_device_info *devices_info)
5353 struct btrfs_fs_info *info = trans->fs_info;
5354 struct map_lookup *map = NULL;
5355 struct extent_map_tree *em_tree;
5356 struct btrfs_block_group *block_group;
5357 struct extent_map *em;
5358 u64 start = ctl->start;
5359 u64 type = ctl->type;
5364 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
5366 return ERR_PTR(-ENOMEM);
5367 map->num_stripes = ctl->num_stripes;
5369 for (i = 0; i < ctl->ndevs; ++i) {
5370 for (j = 0; j < ctl->dev_stripes; ++j) {
5371 int s = i * ctl->dev_stripes + j;
5372 map->stripes[s].dev = devices_info[i].dev;
5373 map->stripes[s].physical = devices_info[i].dev_offset +
5374 j * ctl->stripe_size;
5377 map->stripe_len = BTRFS_STRIPE_LEN;
5378 map->io_align = BTRFS_STRIPE_LEN;
5379 map->io_width = BTRFS_STRIPE_LEN;
5381 map->sub_stripes = ctl->sub_stripes;
5383 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5385 em = alloc_extent_map();
5388 return ERR_PTR(-ENOMEM);
5390 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5391 em->map_lookup = map;
5393 em->len = ctl->chunk_size;
5394 em->block_start = 0;
5395 em->block_len = em->len;
5396 em->orig_block_len = ctl->stripe_size;
5398 em_tree = &info->mapping_tree;
5399 write_lock(&em_tree->lock);
5400 ret = add_extent_mapping(em_tree, em, 0);
5402 write_unlock(&em_tree->lock);
5403 free_extent_map(em);
5404 return ERR_PTR(ret);
5406 write_unlock(&em_tree->lock);
5408 block_group = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size);
5409 if (IS_ERR(block_group))
5410 goto error_del_extent;
5412 for (i = 0; i < map->num_stripes; i++) {
5413 struct btrfs_device *dev = map->stripes[i].dev;
5415 btrfs_device_set_bytes_used(dev,
5416 dev->bytes_used + ctl->stripe_size);
5417 if (list_empty(&dev->post_commit_list))
5418 list_add_tail(&dev->post_commit_list,
5419 &trans->transaction->dev_update_list);
5422 atomic64_sub(ctl->stripe_size * map->num_stripes,
5423 &info->free_chunk_space);
5425 free_extent_map(em);
5426 check_raid56_incompat_flag(info, type);
5427 check_raid1c34_incompat_flag(info, type);
5432 write_lock(&em_tree->lock);
5433 remove_extent_mapping(em_tree, em);
5434 write_unlock(&em_tree->lock);
5436 /* One for our allocation */
5437 free_extent_map(em);
5438 /* One for the tree reference */
5439 free_extent_map(em);
5444 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
5447 struct btrfs_fs_info *info = trans->fs_info;
5448 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5449 struct btrfs_device_info *devices_info = NULL;
5450 struct alloc_chunk_ctl ctl;
5451 struct btrfs_block_group *block_group;
5454 lockdep_assert_held(&info->chunk_mutex);
5456 if (!alloc_profile_is_valid(type, 0)) {
5458 return ERR_PTR(-EINVAL);
5461 if (list_empty(&fs_devices->alloc_list)) {
5462 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5463 btrfs_debug(info, "%s: no writable device", __func__);
5464 return ERR_PTR(-ENOSPC);
5467 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5468 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5470 return ERR_PTR(-EINVAL);
5473 ctl.start = find_next_chunk(info);
5475 init_alloc_chunk_ctl(fs_devices, &ctl);
5477 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5480 return ERR_PTR(-ENOMEM);
5482 ret = gather_device_info(fs_devices, &ctl, devices_info);
5484 block_group = ERR_PTR(ret);
5488 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5490 block_group = ERR_PTR(ret);
5494 block_group = create_chunk(trans, &ctl, devices_info);
5497 kfree(devices_info);
5502 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5503 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5506 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5509 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
5510 struct btrfs_block_group *bg)
5512 struct btrfs_fs_info *fs_info = trans->fs_info;
5513 struct btrfs_root *chunk_root = fs_info->chunk_root;
5514 struct btrfs_key key;
5515 struct btrfs_chunk *chunk;
5516 struct btrfs_stripe *stripe;
5517 struct extent_map *em;
5518 struct map_lookup *map;
5524 * We take the chunk_mutex for 2 reasons:
5526 * 1) Updates and insertions in the chunk btree must be done while holding
5527 * the chunk_mutex, as well as updating the system chunk array in the
5528 * superblock. See the comment on top of btrfs_chunk_alloc() for the
5531 * 2) To prevent races with the final phase of a device replace operation
5532 * that replaces the device object associated with the map's stripes,
5533 * because the device object's id can change at any time during that
5534 * final phase of the device replace operation
5535 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5536 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5537 * which would cause a failure when updating the device item, which does
5538 * not exists, or persisting a stripe of the chunk item with such ID.
5539 * Here we can't use the device_list_mutex because our caller already
5540 * has locked the chunk_mutex, and the final phase of device replace
5541 * acquires both mutexes - first the device_list_mutex and then the
5542 * chunk_mutex. Using any of those two mutexes protects us from a
5543 * concurrent device replace.
5545 lockdep_assert_held(&fs_info->chunk_mutex);
5547 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
5550 btrfs_abort_transaction(trans, ret);
5554 map = em->map_lookup;
5555 item_size = btrfs_chunk_item_size(map->num_stripes);
5557 chunk = kzalloc(item_size, GFP_NOFS);
5560 btrfs_abort_transaction(trans, ret);
5564 for (i = 0; i < map->num_stripes; i++) {
5565 struct btrfs_device *device = map->stripes[i].dev;
5567 ret = btrfs_update_device(trans, device);
5572 stripe = &chunk->stripe;
5573 for (i = 0; i < map->num_stripes; i++) {
5574 struct btrfs_device *device = map->stripes[i].dev;
5575 const u64 dev_offset = map->stripes[i].physical;
5577 btrfs_set_stack_stripe_devid(stripe, device->devid);
5578 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5579 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5583 btrfs_set_stack_chunk_length(chunk, bg->length);
5584 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
5585 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5586 btrfs_set_stack_chunk_type(chunk, map->type);
5587 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5588 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5589 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5590 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5591 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5593 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5594 key.type = BTRFS_CHUNK_ITEM_KEY;
5595 key.offset = bg->start;
5597 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5601 set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags);
5603 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5604 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5611 free_extent_map(em);
5615 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5617 struct btrfs_fs_info *fs_info = trans->fs_info;
5619 struct btrfs_block_group *meta_bg;
5620 struct btrfs_block_group *sys_bg;
5623 * When adding a new device for sprouting, the seed device is read-only
5624 * so we must first allocate a metadata and a system chunk. But before
5625 * adding the block group items to the extent, device and chunk btrees,
5628 * 1) Create both chunks without doing any changes to the btrees, as
5629 * otherwise we would get -ENOSPC since the block groups from the
5630 * seed device are read-only;
5632 * 2) Add the device item for the new sprout device - finishing the setup
5633 * of a new block group requires updating the device item in the chunk
5634 * btree, so it must exist when we attempt to do it. The previous step
5635 * ensures this does not fail with -ENOSPC.
5637 * After that we can add the block group items to their btrees:
5638 * update existing device item in the chunk btree, add a new block group
5639 * item to the extent btree, add a new chunk item to the chunk btree and
5640 * finally add the new device extent items to the devices btree.
5643 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5644 meta_bg = btrfs_create_chunk(trans, alloc_profile);
5645 if (IS_ERR(meta_bg))
5646 return PTR_ERR(meta_bg);
5648 alloc_profile = btrfs_system_alloc_profile(fs_info);
5649 sys_bg = btrfs_create_chunk(trans, alloc_profile);
5651 return PTR_ERR(sys_bg);
5656 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5658 const int index = btrfs_bg_flags_to_raid_index(map->type);
5660 return btrfs_raid_array[index].tolerated_failures;
5663 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5665 struct extent_map *em;
5666 struct map_lookup *map;
5671 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5675 map = em->map_lookup;
5676 for (i = 0; i < map->num_stripes; i++) {
5677 if (test_bit(BTRFS_DEV_STATE_MISSING,
5678 &map->stripes[i].dev->dev_state)) {
5682 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5683 &map->stripes[i].dev->dev_state)) {
5690 * If the number of missing devices is larger than max errors, we can
5691 * not write the data into that chunk successfully.
5693 if (miss_ndevs > btrfs_chunk_max_errors(map))
5696 free_extent_map(em);
5700 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5702 struct extent_map *em;
5705 write_lock(&tree->lock);
5706 em = lookup_extent_mapping(tree, 0, (u64)-1);
5708 remove_extent_mapping(tree, em);
5709 write_unlock(&tree->lock);
5713 free_extent_map(em);
5714 /* once for the tree */
5715 free_extent_map(em);
5719 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5721 struct extent_map *em;
5722 struct map_lookup *map;
5723 enum btrfs_raid_types index;
5726 em = btrfs_get_chunk_map(fs_info, logical, len);
5729 * We could return errors for these cases, but that could get
5730 * ugly and we'd probably do the same thing which is just not do
5731 * anything else and exit, so return 1 so the callers don't try
5732 * to use other copies.
5736 map = em->map_lookup;
5737 index = btrfs_bg_flags_to_raid_index(map->type);
5739 /* Non-RAID56, use their ncopies from btrfs_raid_array. */
5740 if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5741 ret = btrfs_raid_array[index].ncopies;
5742 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5744 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5746 * There could be two corrupted data stripes, we need
5747 * to loop retry in order to rebuild the correct data.
5749 * Fail a stripe at a time on every retry except the
5750 * stripe under reconstruction.
5752 ret = map->num_stripes;
5753 free_extent_map(em);
5755 down_read(&fs_info->dev_replace.rwsem);
5756 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5757 fs_info->dev_replace.tgtdev)
5759 up_read(&fs_info->dev_replace.rwsem);
5764 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5767 struct extent_map *em;
5768 struct map_lookup *map;
5769 unsigned long len = fs_info->sectorsize;
5771 if (!btrfs_fs_incompat(fs_info, RAID56))
5774 em = btrfs_get_chunk_map(fs_info, logical, len);
5776 if (!WARN_ON(IS_ERR(em))) {
5777 map = em->map_lookup;
5778 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5779 len = map->stripe_len * nr_data_stripes(map);
5780 free_extent_map(em);
5785 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5787 struct extent_map *em;
5788 struct map_lookup *map;
5791 if (!btrfs_fs_incompat(fs_info, RAID56))
5794 em = btrfs_get_chunk_map(fs_info, logical, len);
5796 if(!WARN_ON(IS_ERR(em))) {
5797 map = em->map_lookup;
5798 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5800 free_extent_map(em);
5805 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5806 struct map_lookup *map, int first,
5807 int dev_replace_is_ongoing)
5811 int preferred_mirror;
5813 struct btrfs_device *srcdev;
5816 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5818 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5819 num_stripes = map->sub_stripes;
5821 num_stripes = map->num_stripes;
5823 switch (fs_info->fs_devices->read_policy) {
5825 /* Shouldn't happen, just warn and use pid instead of failing */
5826 btrfs_warn_rl(fs_info,
5827 "unknown read_policy type %u, reset to pid",
5828 fs_info->fs_devices->read_policy);
5829 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID;
5831 case BTRFS_READ_POLICY_PID:
5832 preferred_mirror = first + (current->pid % num_stripes);
5836 if (dev_replace_is_ongoing &&
5837 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5838 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5839 srcdev = fs_info->dev_replace.srcdev;
5844 * try to avoid the drive that is the source drive for a
5845 * dev-replace procedure, only choose it if no other non-missing
5846 * mirror is available
5848 for (tolerance = 0; tolerance < 2; tolerance++) {
5849 if (map->stripes[preferred_mirror].dev->bdev &&
5850 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5851 return preferred_mirror;
5852 for (i = first; i < first + num_stripes; i++) {
5853 if (map->stripes[i].dev->bdev &&
5854 (tolerance || map->stripes[i].dev != srcdev))
5859 /* we couldn't find one that doesn't fail. Just return something
5860 * and the io error handling code will clean up eventually
5862 return preferred_mirror;
5865 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5866 static void sort_parity_stripes(struct btrfs_io_context *bioc, int num_stripes)
5873 for (i = 0; i < num_stripes - 1; i++) {
5874 /* Swap if parity is on a smaller index */
5875 if (bioc->raid_map[i] > bioc->raid_map[i + 1]) {
5876 swap(bioc->stripes[i], bioc->stripes[i + 1]);
5877 swap(bioc->raid_map[i], bioc->raid_map[i + 1]);
5884 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info,
5888 struct btrfs_io_context *bioc = kzalloc(
5889 /* The size of btrfs_io_context */
5890 sizeof(struct btrfs_io_context) +
5891 /* Plus the variable array for the stripes */
5892 sizeof(struct btrfs_io_stripe) * (total_stripes) +
5893 /* Plus the variable array for the tgt dev */
5894 sizeof(int) * (real_stripes) +
5896 * Plus the raid_map, which includes both the tgt dev
5899 sizeof(u64) * (total_stripes),
5905 refcount_set(&bioc->refs, 1);
5907 bioc->fs_info = fs_info;
5908 bioc->tgtdev_map = (int *)(bioc->stripes + total_stripes);
5909 bioc->raid_map = (u64 *)(bioc->tgtdev_map + real_stripes);
5914 void btrfs_get_bioc(struct btrfs_io_context *bioc)
5916 WARN_ON(!refcount_read(&bioc->refs));
5917 refcount_inc(&bioc->refs);
5920 void btrfs_put_bioc(struct btrfs_io_context *bioc)
5924 if (refcount_dec_and_test(&bioc->refs))
5929 * Please note that, discard won't be sent to target device of device
5932 struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info,
5933 u64 logical, u64 *length_ret,
5936 struct extent_map *em;
5937 struct map_lookup *map;
5938 struct btrfs_discard_stripe *stripes;
5939 u64 length = *length_ret;
5943 u64 stripe_end_offset;
5949 u32 sub_stripes = 0;
5950 u64 stripes_per_dev = 0;
5951 u32 remaining_stripes = 0;
5952 u32 last_stripe = 0;
5956 em = btrfs_get_chunk_map(fs_info, logical, length);
5958 return ERR_CAST(em);
5960 map = em->map_lookup;
5962 /* we don't discard raid56 yet */
5963 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5968 offset = logical - em->start;
5969 length = min_t(u64, em->start + em->len - logical, length);
5970 *length_ret = length;
5972 stripe_len = map->stripe_len;
5974 * stripe_nr counts the total number of stripes we have to stride
5975 * to get to this block
5977 stripe_nr = div64_u64(offset, stripe_len);
5979 /* stripe_offset is the offset of this block in its stripe */
5980 stripe_offset = offset - stripe_nr * stripe_len;
5982 stripe_nr_end = round_up(offset + length, map->stripe_len);
5983 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5984 stripe_cnt = stripe_nr_end - stripe_nr;
5985 stripe_end_offset = stripe_nr_end * map->stripe_len -
5988 * after this, stripe_nr is the number of stripes on this
5989 * device we have to walk to find the data, and stripe_index is
5990 * the number of our device in the stripe array
5994 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5995 BTRFS_BLOCK_GROUP_RAID10)) {
5996 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5999 sub_stripes = map->sub_stripes;
6001 factor = map->num_stripes / sub_stripes;
6002 *num_stripes = min_t(u64, map->num_stripes,
6003 sub_stripes * stripe_cnt);
6004 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6005 stripe_index *= sub_stripes;
6006 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
6007 &remaining_stripes);
6008 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
6009 last_stripe *= sub_stripes;
6010 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
6011 BTRFS_BLOCK_GROUP_DUP)) {
6012 *num_stripes = map->num_stripes;
6014 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6018 stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS);
6024 for (i = 0; i < *num_stripes; i++) {
6025 stripes[i].physical =
6026 map->stripes[stripe_index].physical +
6027 stripe_offset + stripe_nr * map->stripe_len;
6028 stripes[i].dev = map->stripes[stripe_index].dev;
6030 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6031 BTRFS_BLOCK_GROUP_RAID10)) {
6032 stripes[i].length = stripes_per_dev * map->stripe_len;
6034 if (i / sub_stripes < remaining_stripes)
6035 stripes[i].length += map->stripe_len;
6038 * Special for the first stripe and
6041 * |-------|...|-------|
6045 if (i < sub_stripes)
6046 stripes[i].length -= stripe_offset;
6048 if (stripe_index >= last_stripe &&
6049 stripe_index <= (last_stripe +
6051 stripes[i].length -= stripe_end_offset;
6053 if (i == sub_stripes - 1)
6056 stripes[i].length = length;
6060 if (stripe_index == map->num_stripes) {
6066 free_extent_map(em);
6069 free_extent_map(em);
6070 return ERR_PTR(ret);
6074 * In dev-replace case, for repair case (that's the only case where the mirror
6075 * is selected explicitly when calling btrfs_map_block), blocks left of the
6076 * left cursor can also be read from the target drive.
6078 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
6080 * For READ, it also needs to be supported using the same mirror number.
6082 * If the requested block is not left of the left cursor, EIO is returned. This
6083 * can happen because btrfs_num_copies() returns one more in the dev-replace
6086 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
6087 u64 logical, u64 length,
6088 u64 srcdev_devid, int *mirror_num,
6091 struct btrfs_io_context *bioc = NULL;
6093 int index_srcdev = 0;
6095 u64 physical_of_found = 0;
6099 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
6100 logical, &length, &bioc, NULL, NULL, 0);
6102 ASSERT(bioc == NULL);
6106 num_stripes = bioc->num_stripes;
6107 if (*mirror_num > num_stripes) {
6109 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
6110 * that means that the requested area is not left of the left
6113 btrfs_put_bioc(bioc);
6118 * process the rest of the function using the mirror_num of the source
6119 * drive. Therefore look it up first. At the end, patch the device
6120 * pointer to the one of the target drive.
6122 for (i = 0; i < num_stripes; i++) {
6123 if (bioc->stripes[i].dev->devid != srcdev_devid)
6127 * In case of DUP, in order to keep it simple, only add the
6128 * mirror with the lowest physical address
6131 physical_of_found <= bioc->stripes[i].physical)
6136 physical_of_found = bioc->stripes[i].physical;
6139 btrfs_put_bioc(bioc);
6145 *mirror_num = index_srcdev + 1;
6146 *physical = physical_of_found;
6150 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)
6152 struct btrfs_block_group *cache;
6155 /* Non zoned filesystem does not use "to_copy" flag */
6156 if (!btrfs_is_zoned(fs_info))
6159 cache = btrfs_lookup_block_group(fs_info, logical);
6161 ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags);
6163 btrfs_put_block_group(cache);
6167 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
6168 struct btrfs_io_context **bioc_ret,
6169 struct btrfs_dev_replace *dev_replace,
6171 int *num_stripes_ret, int *max_errors_ret)
6173 struct btrfs_io_context *bioc = *bioc_ret;
6174 u64 srcdev_devid = dev_replace->srcdev->devid;
6175 int tgtdev_indexes = 0;
6176 int num_stripes = *num_stripes_ret;
6177 int max_errors = *max_errors_ret;
6180 if (op == BTRFS_MAP_WRITE) {
6181 int index_where_to_add;
6184 * A block group which have "to_copy" set will eventually
6185 * copied by dev-replace process. We can avoid cloning IO here.
6187 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))
6191 * duplicate the write operations while the dev replace
6192 * procedure is running. Since the copying of the old disk to
6193 * the new disk takes place at run time while the filesystem is
6194 * mounted writable, the regular write operations to the old
6195 * disk have to be duplicated to go to the new disk as well.
6197 * Note that device->missing is handled by the caller, and that
6198 * the write to the old disk is already set up in the stripes
6201 index_where_to_add = num_stripes;
6202 for (i = 0; i < num_stripes; i++) {
6203 if (bioc->stripes[i].dev->devid == srcdev_devid) {
6204 /* write to new disk, too */
6205 struct btrfs_io_stripe *new =
6206 bioc->stripes + index_where_to_add;
6207 struct btrfs_io_stripe *old =
6210 new->physical = old->physical;
6211 new->dev = dev_replace->tgtdev;
6212 bioc->tgtdev_map[i] = index_where_to_add;
6213 index_where_to_add++;
6218 num_stripes = index_where_to_add;
6219 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
6220 int index_srcdev = 0;
6222 u64 physical_of_found = 0;
6225 * During the dev-replace procedure, the target drive can also
6226 * be used to read data in case it is needed to repair a corrupt
6227 * block elsewhere. This is possible if the requested area is
6228 * left of the left cursor. In this area, the target drive is a
6229 * full copy of the source drive.
6231 for (i = 0; i < num_stripes; i++) {
6232 if (bioc->stripes[i].dev->devid == srcdev_devid) {
6234 * In case of DUP, in order to keep it simple,
6235 * only add the mirror with the lowest physical
6239 physical_of_found <= bioc->stripes[i].physical)
6243 physical_of_found = bioc->stripes[i].physical;
6247 struct btrfs_io_stripe *tgtdev_stripe =
6248 bioc->stripes + num_stripes;
6250 tgtdev_stripe->physical = physical_of_found;
6251 tgtdev_stripe->dev = dev_replace->tgtdev;
6252 bioc->tgtdev_map[index_srcdev] = num_stripes;
6259 *num_stripes_ret = num_stripes;
6260 *max_errors_ret = max_errors;
6261 bioc->num_tgtdevs = tgtdev_indexes;
6265 static bool need_full_stripe(enum btrfs_map_op op)
6267 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
6271 * Calculate the geometry of a particular (address, len) tuple. This
6272 * information is used to calculate how big a particular bio can get before it
6273 * straddles a stripe.
6275 * @fs_info: the filesystem
6276 * @em: mapping containing the logical extent
6277 * @op: type of operation - write or read
6278 * @logical: address that we want to figure out the geometry of
6279 * @io_geom: pointer used to return values
6281 * Returns < 0 in case a chunk for the given logical address cannot be found,
6282 * usually shouldn't happen unless @logical is corrupted, 0 otherwise.
6284 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *em,
6285 enum btrfs_map_op op, u64 logical,
6286 struct btrfs_io_geometry *io_geom)
6288 struct map_lookup *map;
6294 u64 raid56_full_stripe_start = (u64)-1;
6297 ASSERT(op != BTRFS_MAP_DISCARD);
6299 map = em->map_lookup;
6300 /* Offset of this logical address in the chunk */
6301 offset = logical - em->start;
6302 /* Len of a stripe in a chunk */
6303 stripe_len = map->stripe_len;
6305 * Stripe_nr is where this block falls in
6306 * stripe_offset is the offset of this block in its stripe.
6308 stripe_nr = div64_u64_rem(offset, stripe_len, &stripe_offset);
6309 ASSERT(stripe_offset < U32_MAX);
6311 data_stripes = nr_data_stripes(map);
6313 /* Only stripe based profiles needs to check against stripe length. */
6314 if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK) {
6315 u64 max_len = stripe_len - stripe_offset;
6318 * In case of raid56, we need to know the stripe aligned start
6320 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6321 unsigned long full_stripe_len = stripe_len * data_stripes;
6322 raid56_full_stripe_start = offset;
6325 * Allow a write of a full stripe, but make sure we
6326 * don't allow straddling of stripes
6328 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
6330 raid56_full_stripe_start *= full_stripe_len;
6333 * For writes to RAID[56], allow a full stripeset across
6334 * all disks. For other RAID types and for RAID[56]
6335 * reads, just allow a single stripe (on a single disk).
6337 if (op == BTRFS_MAP_WRITE) {
6338 max_len = stripe_len * data_stripes -
6339 (offset - raid56_full_stripe_start);
6342 len = min_t(u64, em->len - offset, max_len);
6344 len = em->len - offset;
6348 io_geom->offset = offset;
6349 io_geom->stripe_len = stripe_len;
6350 io_geom->stripe_nr = stripe_nr;
6351 io_geom->stripe_offset = stripe_offset;
6352 io_geom->raid56_stripe_offset = raid56_full_stripe_start;
6357 static void set_io_stripe(struct btrfs_io_stripe *dst, const struct map_lookup *map,
6358 u32 stripe_index, u64 stripe_offset, u64 stripe_nr)
6360 dst->dev = map->stripes[stripe_index].dev;
6361 dst->physical = map->stripes[stripe_index].physical +
6362 stripe_offset + stripe_nr * map->stripe_len;
6365 int __btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6366 u64 logical, u64 *length,
6367 struct btrfs_io_context **bioc_ret,
6368 struct btrfs_io_stripe *smap, int *mirror_num_ret,
6371 struct extent_map *em;
6372 struct map_lookup *map;
6380 int mirror_num = (mirror_num_ret ? *mirror_num_ret : 0);
6383 int tgtdev_indexes = 0;
6384 struct btrfs_io_context *bioc = NULL;
6385 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6386 int dev_replace_is_ongoing = 0;
6387 int num_alloc_stripes;
6388 int patch_the_first_stripe_for_dev_replace = 0;
6389 u64 physical_to_patch_in_first_stripe = 0;
6390 u64 raid56_full_stripe_start = (u64)-1;
6391 struct btrfs_io_geometry geom;
6394 ASSERT(op != BTRFS_MAP_DISCARD);
6396 em = btrfs_get_chunk_map(fs_info, logical, *length);
6397 ASSERT(!IS_ERR(em));
6399 ret = btrfs_get_io_geometry(fs_info, em, op, logical, &geom);
6403 map = em->map_lookup;
6406 stripe_len = geom.stripe_len;
6407 stripe_nr = geom.stripe_nr;
6408 stripe_offset = geom.stripe_offset;
6409 raid56_full_stripe_start = geom.raid56_stripe_offset;
6410 data_stripes = nr_data_stripes(map);
6412 down_read(&dev_replace->rwsem);
6413 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6415 * Hold the semaphore for read during the whole operation, write is
6416 * requested at commit time but must wait.
6418 if (!dev_replace_is_ongoing)
6419 up_read(&dev_replace->rwsem);
6421 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
6422 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
6423 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
6424 dev_replace->srcdev->devid,
6426 &physical_to_patch_in_first_stripe);
6430 patch_the_first_stripe_for_dev_replace = 1;
6431 } else if (mirror_num > map->num_stripes) {
6437 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6438 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6440 if (!need_full_stripe(op))
6442 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6443 if (need_full_stripe(op))
6444 num_stripes = map->num_stripes;
6445 else if (mirror_num)
6446 stripe_index = mirror_num - 1;
6448 stripe_index = find_live_mirror(fs_info, map, 0,
6449 dev_replace_is_ongoing);
6450 mirror_num = stripe_index + 1;
6453 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6454 if (need_full_stripe(op)) {
6455 num_stripes = map->num_stripes;
6456 } else if (mirror_num) {
6457 stripe_index = mirror_num - 1;
6462 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6463 u32 factor = map->num_stripes / map->sub_stripes;
6465 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6466 stripe_index *= map->sub_stripes;
6468 if (need_full_stripe(op))
6469 num_stripes = map->sub_stripes;
6470 else if (mirror_num)
6471 stripe_index += mirror_num - 1;
6473 int old_stripe_index = stripe_index;
6474 stripe_index = find_live_mirror(fs_info, map,
6476 dev_replace_is_ongoing);
6477 mirror_num = stripe_index - old_stripe_index + 1;
6480 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6481 ASSERT(map->stripe_len == BTRFS_STRIPE_LEN);
6482 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6483 /* push stripe_nr back to the start of the full stripe */
6484 stripe_nr = div64_u64(raid56_full_stripe_start,
6485 stripe_len * data_stripes);
6487 /* RAID[56] write or recovery. Return all stripes */
6488 num_stripes = map->num_stripes;
6489 max_errors = btrfs_chunk_max_errors(map);
6491 /* Return the length to the full stripe end */
6492 *length = min(logical + *length,
6493 raid56_full_stripe_start + em->start +
6494 data_stripes * stripe_len) - logical;
6499 * Mirror #0 or #1 means the original data block.
6500 * Mirror #2 is RAID5 parity block.
6501 * Mirror #3 is RAID6 Q block.
6503 stripe_nr = div_u64_rem(stripe_nr,
6504 data_stripes, &stripe_index);
6506 stripe_index = data_stripes + mirror_num - 2;
6508 /* We distribute the parity blocks across stripes */
6509 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
6511 if (!need_full_stripe(op) && mirror_num <= 1)
6516 * after this, stripe_nr is the number of stripes on this
6517 * device we have to walk to find the data, and stripe_index is
6518 * the number of our device in the stripe array
6520 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6522 mirror_num = stripe_index + 1;
6524 if (stripe_index >= map->num_stripes) {
6526 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6527 stripe_index, map->num_stripes);
6532 num_alloc_stripes = num_stripes;
6533 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
6534 if (op == BTRFS_MAP_WRITE)
6535 num_alloc_stripes <<= 1;
6536 if (op == BTRFS_MAP_GET_READ_MIRRORS)
6537 num_alloc_stripes++;
6538 tgtdev_indexes = num_stripes;
6542 * If this I/O maps to a single device, try to return the device and
6543 * physical block information on the stack instead of allocating an
6544 * I/O context structure.
6546 if (smap && num_alloc_stripes == 1 &&
6547 !((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1) &&
6548 (!need_full_stripe(op) || !dev_replace_is_ongoing ||
6549 !dev_replace->tgtdev)) {
6550 if (patch_the_first_stripe_for_dev_replace) {
6551 smap->dev = dev_replace->tgtdev;
6552 smap->physical = physical_to_patch_in_first_stripe;
6553 *mirror_num_ret = map->num_stripes + 1;
6555 set_io_stripe(smap, map, stripe_index, stripe_offset,
6557 *mirror_num_ret = mirror_num;
6564 bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes, tgtdev_indexes);
6570 for (i = 0; i < num_stripes; i++) {
6571 set_io_stripe(&bioc->stripes[i], map, stripe_index, stripe_offset,
6576 /* Build raid_map */
6577 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6578 (need_full_stripe(op) || mirror_num > 1)) {
6582 /* Work out the disk rotation on this stripe-set */
6583 div_u64_rem(stripe_nr, num_stripes, &rot);
6585 /* Fill in the logical address of each stripe */
6586 tmp = stripe_nr * data_stripes;
6587 for (i = 0; i < data_stripes; i++)
6588 bioc->raid_map[(i + rot) % num_stripes] =
6589 em->start + (tmp + i) * map->stripe_len;
6591 bioc->raid_map[(i + rot) % map->num_stripes] = RAID5_P_STRIPE;
6592 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
6593 bioc->raid_map[(i + rot + 1) % num_stripes] =
6596 sort_parity_stripes(bioc, num_stripes);
6599 if (need_full_stripe(op))
6600 max_errors = btrfs_chunk_max_errors(map);
6602 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6603 need_full_stripe(op)) {
6604 handle_ops_on_dev_replace(op, &bioc, dev_replace, logical,
6605 &num_stripes, &max_errors);
6609 bioc->map_type = map->type;
6610 bioc->num_stripes = num_stripes;
6611 bioc->max_errors = max_errors;
6612 bioc->mirror_num = mirror_num;
6615 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6616 * mirror_num == num_stripes + 1 && dev_replace target drive is
6617 * available as a mirror
6619 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6620 WARN_ON(num_stripes > 1);
6621 bioc->stripes[0].dev = dev_replace->tgtdev;
6622 bioc->stripes[0].physical = physical_to_patch_in_first_stripe;
6623 bioc->mirror_num = map->num_stripes + 1;
6626 if (dev_replace_is_ongoing) {
6627 lockdep_assert_held(&dev_replace->rwsem);
6628 /* Unlock and let waiting writers proceed */
6629 up_read(&dev_replace->rwsem);
6631 free_extent_map(em);
6635 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6636 u64 logical, u64 *length,
6637 struct btrfs_io_context **bioc_ret, int mirror_num)
6639 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret,
6640 NULL, &mirror_num, 0);
6643 /* For Scrub/replace */
6644 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6645 u64 logical, u64 *length,
6646 struct btrfs_io_context **bioc_ret)
6648 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret,
6652 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args,
6653 const struct btrfs_fs_devices *fs_devices)
6655 if (args->fsid == NULL)
6657 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0)
6662 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args,
6663 const struct btrfs_device *device)
6665 if (args->missing) {
6666 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) &&
6672 if (device->devid != args->devid)
6674 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0)
6680 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6683 * If devid and uuid are both specified, the match must be exact, otherwise
6684 * only devid is used.
6686 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,
6687 const struct btrfs_dev_lookup_args *args)
6689 struct btrfs_device *device;
6690 struct btrfs_fs_devices *seed_devs;
6692 if (dev_args_match_fs_devices(args, fs_devices)) {
6693 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6694 if (dev_args_match_device(args, device))
6699 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6700 if (!dev_args_match_fs_devices(args, seed_devs))
6702 list_for_each_entry(device, &seed_devs->devices, dev_list) {
6703 if (dev_args_match_device(args, device))
6711 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6712 u64 devid, u8 *dev_uuid)
6714 struct btrfs_device *device;
6715 unsigned int nofs_flag;
6718 * We call this under the chunk_mutex, so we want to use NOFS for this
6719 * allocation, however we don't want to change btrfs_alloc_device() to
6720 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6724 nofs_flag = memalloc_nofs_save();
6725 device = btrfs_alloc_device(NULL, &devid, dev_uuid, NULL);
6726 memalloc_nofs_restore(nofs_flag);
6730 list_add(&device->dev_list, &fs_devices->devices);
6731 device->fs_devices = fs_devices;
6732 fs_devices->num_devices++;
6734 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6735 fs_devices->missing_devices++;
6741 * Allocate new device struct, set up devid and UUID.
6743 * @fs_info: used only for generating a new devid, can be NULL if
6744 * devid is provided (i.e. @devid != NULL).
6745 * @devid: a pointer to devid for this device. If NULL a new devid
6747 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6749 * @path: a pointer to device path if available, NULL otherwise.
6751 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6752 * on error. Returned struct is not linked onto any lists and must be
6753 * destroyed with btrfs_free_device.
6755 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6756 const u64 *devid, const u8 *uuid,
6759 struct btrfs_device *dev;
6762 if (WARN_ON(!devid && !fs_info))
6763 return ERR_PTR(-EINVAL);
6765 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
6767 return ERR_PTR(-ENOMEM);
6769 INIT_LIST_HEAD(&dev->dev_list);
6770 INIT_LIST_HEAD(&dev->dev_alloc_list);
6771 INIT_LIST_HEAD(&dev->post_commit_list);
6773 atomic_set(&dev->dev_stats_ccnt, 0);
6774 btrfs_device_data_ordered_init(dev);
6775 extent_io_tree_init(fs_info, &dev->alloc_state, IO_TREE_DEVICE_ALLOC_STATE);
6782 ret = find_next_devid(fs_info, &tmp);
6784 btrfs_free_device(dev);
6785 return ERR_PTR(ret);
6791 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6793 generate_random_uuid(dev->uuid);
6796 struct rcu_string *name;
6798 name = rcu_string_strdup(path, GFP_KERNEL);
6800 btrfs_free_device(dev);
6801 return ERR_PTR(-ENOMEM);
6803 rcu_assign_pointer(dev->name, name);
6809 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6810 u64 devid, u8 *uuid, bool error)
6813 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6816 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6820 u64 btrfs_calc_stripe_length(const struct extent_map *em)
6822 const struct map_lookup *map = em->map_lookup;
6823 const int data_stripes = calc_data_stripes(map->type, map->num_stripes);
6825 return div_u64(em->len, data_stripes);
6828 #if BITS_PER_LONG == 32
6830 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
6831 * can't be accessed on 32bit systems.
6833 * This function do mount time check to reject the fs if it already has
6834 * metadata chunk beyond that limit.
6836 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6837 u64 logical, u64 length, u64 type)
6839 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6842 if (logical + length < MAX_LFS_FILESIZE)
6845 btrfs_err_32bit_limit(fs_info);
6850 * This is to give early warning for any metadata chunk reaching
6851 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
6852 * Although we can still access the metadata, it's not going to be possible
6853 * once the limit is reached.
6855 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6856 u64 logical, u64 length, u64 type)
6858 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6861 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)
6864 btrfs_warn_32bit_limit(fs_info);
6868 static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info,
6869 u64 devid, u8 *uuid)
6871 struct btrfs_device *dev;
6873 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6874 btrfs_report_missing_device(fs_info, devid, uuid, true);
6875 return ERR_PTR(-ENOENT);
6878 dev = add_missing_dev(fs_info->fs_devices, devid, uuid);
6880 btrfs_err(fs_info, "failed to init missing device %llu: %ld",
6881 devid, PTR_ERR(dev));
6884 btrfs_report_missing_device(fs_info, devid, uuid, false);
6889 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6890 struct btrfs_chunk *chunk)
6892 BTRFS_DEV_LOOKUP_ARGS(args);
6893 struct btrfs_fs_info *fs_info = leaf->fs_info;
6894 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6895 struct map_lookup *map;
6896 struct extent_map *em;
6901 u8 uuid[BTRFS_UUID_SIZE];
6907 logical = key->offset;
6908 length = btrfs_chunk_length(leaf, chunk);
6909 type = btrfs_chunk_type(leaf, chunk);
6910 index = btrfs_bg_flags_to_raid_index(type);
6911 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6913 #if BITS_PER_LONG == 32
6914 ret = check_32bit_meta_chunk(fs_info, logical, length, type);
6917 warn_32bit_meta_chunk(fs_info, logical, length, type);
6921 * Only need to verify chunk item if we're reading from sys chunk array,
6922 * as chunk item in tree block is already verified by tree-checker.
6924 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6925 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6930 read_lock(&map_tree->lock);
6931 em = lookup_extent_mapping(map_tree, logical, 1);
6932 read_unlock(&map_tree->lock);
6934 /* already mapped? */
6935 if (em && em->start <= logical && em->start + em->len > logical) {
6936 free_extent_map(em);
6939 free_extent_map(em);
6942 em = alloc_extent_map();
6945 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6947 free_extent_map(em);
6951 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6952 em->map_lookup = map;
6953 em->start = logical;
6956 em->block_start = 0;
6957 em->block_len = em->len;
6959 map->num_stripes = num_stripes;
6960 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6961 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6962 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6965 * We can't use the sub_stripes value, as for profiles other than
6966 * RAID10, they may have 0 as sub_stripes for filesystems created by
6967 * older mkfs (<v5.4).
6968 * In that case, it can cause divide-by-zero errors later.
6969 * Since currently sub_stripes is fixed for each profile, let's
6970 * use the trusted value instead.
6972 map->sub_stripes = btrfs_raid_array[index].sub_stripes;
6973 map->verified_stripes = 0;
6974 em->orig_block_len = btrfs_calc_stripe_length(em);
6975 for (i = 0; i < num_stripes; i++) {
6976 map->stripes[i].physical =
6977 btrfs_stripe_offset_nr(leaf, chunk, i);
6978 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6980 read_extent_buffer(leaf, uuid, (unsigned long)
6981 btrfs_stripe_dev_uuid_nr(chunk, i),
6984 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args);
6985 if (!map->stripes[i].dev) {
6986 map->stripes[i].dev = handle_missing_device(fs_info,
6988 if (IS_ERR(map->stripes[i].dev)) {
6989 ret = PTR_ERR(map->stripes[i].dev);
6990 free_extent_map(em);
6995 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6996 &(map->stripes[i].dev->dev_state));
6999 write_lock(&map_tree->lock);
7000 ret = add_extent_mapping(map_tree, em, 0);
7001 write_unlock(&map_tree->lock);
7004 "failed to add chunk map, start=%llu len=%llu: %d",
7005 em->start, em->len, ret);
7007 free_extent_map(em);
7012 static void fill_device_from_item(struct extent_buffer *leaf,
7013 struct btrfs_dev_item *dev_item,
7014 struct btrfs_device *device)
7018 device->devid = btrfs_device_id(leaf, dev_item);
7019 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
7020 device->total_bytes = device->disk_total_bytes;
7021 device->commit_total_bytes = device->disk_total_bytes;
7022 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
7023 device->commit_bytes_used = device->bytes_used;
7024 device->type = btrfs_device_type(leaf, dev_item);
7025 device->io_align = btrfs_device_io_align(leaf, dev_item);
7026 device->io_width = btrfs_device_io_width(leaf, dev_item);
7027 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
7028 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
7029 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
7031 ptr = btrfs_device_uuid(dev_item);
7032 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
7035 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
7038 struct btrfs_fs_devices *fs_devices;
7041 lockdep_assert_held(&uuid_mutex);
7044 /* This will match only for multi-device seed fs */
7045 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
7046 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
7050 fs_devices = find_fsid(fsid, NULL);
7052 if (!btrfs_test_opt(fs_info, DEGRADED))
7053 return ERR_PTR(-ENOENT);
7055 fs_devices = alloc_fs_devices(fsid, NULL);
7056 if (IS_ERR(fs_devices))
7059 fs_devices->seeding = true;
7060 fs_devices->opened = 1;
7065 * Upon first call for a seed fs fsid, just create a private copy of the
7066 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
7068 fs_devices = clone_fs_devices(fs_devices);
7069 if (IS_ERR(fs_devices))
7072 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
7074 free_fs_devices(fs_devices);
7075 return ERR_PTR(ret);
7078 if (!fs_devices->seeding) {
7079 close_fs_devices(fs_devices);
7080 free_fs_devices(fs_devices);
7081 return ERR_PTR(-EINVAL);
7084 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
7089 static int read_one_dev(struct extent_buffer *leaf,
7090 struct btrfs_dev_item *dev_item)
7092 BTRFS_DEV_LOOKUP_ARGS(args);
7093 struct btrfs_fs_info *fs_info = leaf->fs_info;
7094 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7095 struct btrfs_device *device;
7098 u8 fs_uuid[BTRFS_FSID_SIZE];
7099 u8 dev_uuid[BTRFS_UUID_SIZE];
7101 devid = btrfs_device_id(leaf, dev_item);
7103 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
7105 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
7107 args.uuid = dev_uuid;
7108 args.fsid = fs_uuid;
7110 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
7111 fs_devices = open_seed_devices(fs_info, fs_uuid);
7112 if (IS_ERR(fs_devices))
7113 return PTR_ERR(fs_devices);
7116 device = btrfs_find_device(fs_info->fs_devices, &args);
7118 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7119 btrfs_report_missing_device(fs_info, devid,
7124 device = add_missing_dev(fs_devices, devid, dev_uuid);
7125 if (IS_ERR(device)) {
7127 "failed to add missing dev %llu: %ld",
7128 devid, PTR_ERR(device));
7129 return PTR_ERR(device);
7131 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
7133 if (!device->bdev) {
7134 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7135 btrfs_report_missing_device(fs_info,
7136 devid, dev_uuid, true);
7139 btrfs_report_missing_device(fs_info, devid,
7143 if (!device->bdev &&
7144 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
7146 * this happens when a device that was properly setup
7147 * in the device info lists suddenly goes bad.
7148 * device->bdev is NULL, and so we have to set
7149 * device->missing to one here
7151 device->fs_devices->missing_devices++;
7152 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
7155 /* Move the device to its own fs_devices */
7156 if (device->fs_devices != fs_devices) {
7157 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
7158 &device->dev_state));
7160 list_move(&device->dev_list, &fs_devices->devices);
7161 device->fs_devices->num_devices--;
7162 fs_devices->num_devices++;
7164 device->fs_devices->missing_devices--;
7165 fs_devices->missing_devices++;
7167 device->fs_devices = fs_devices;
7171 if (device->fs_devices != fs_info->fs_devices) {
7172 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
7173 if (device->generation !=
7174 btrfs_device_generation(leaf, dev_item))
7178 fill_device_from_item(leaf, dev_item, device);
7180 u64 max_total_bytes = bdev_nr_bytes(device->bdev);
7182 if (device->total_bytes > max_total_bytes) {
7184 "device total_bytes should be at most %llu but found %llu",
7185 max_total_bytes, device->total_bytes);
7189 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7190 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7191 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7192 device->fs_devices->total_rw_bytes += device->total_bytes;
7193 atomic64_add(device->total_bytes - device->bytes_used,
7194 &fs_info->free_chunk_space);
7200 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7202 struct btrfs_super_block *super_copy = fs_info->super_copy;
7203 struct extent_buffer *sb;
7204 struct btrfs_disk_key *disk_key;
7205 struct btrfs_chunk *chunk;
7207 unsigned long sb_array_offset;
7214 struct btrfs_key key;
7216 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7219 * We allocated a dummy extent, just to use extent buffer accessors.
7220 * There will be unused space after BTRFS_SUPER_INFO_SIZE, but
7221 * that's fine, we will not go beyond system chunk array anyway.
7223 sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET);
7226 set_extent_buffer_uptodate(sb);
7228 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7229 array_size = btrfs_super_sys_array_size(super_copy);
7231 array_ptr = super_copy->sys_chunk_array;
7232 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7235 while (cur_offset < array_size) {
7236 disk_key = (struct btrfs_disk_key *)array_ptr;
7237 len = sizeof(*disk_key);
7238 if (cur_offset + len > array_size)
7239 goto out_short_read;
7241 btrfs_disk_key_to_cpu(&key, disk_key);
7244 sb_array_offset += len;
7247 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7249 "unexpected item type %u in sys_array at offset %u",
7250 (u32)key.type, cur_offset);
7255 chunk = (struct btrfs_chunk *)sb_array_offset;
7257 * At least one btrfs_chunk with one stripe must be present,
7258 * exact stripe count check comes afterwards
7260 len = btrfs_chunk_item_size(1);
7261 if (cur_offset + len > array_size)
7262 goto out_short_read;
7264 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7267 "invalid number of stripes %u in sys_array at offset %u",
7268 num_stripes, cur_offset);
7273 type = btrfs_chunk_type(sb, chunk);
7274 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7276 "invalid chunk type %llu in sys_array at offset %u",
7282 len = btrfs_chunk_item_size(num_stripes);
7283 if (cur_offset + len > array_size)
7284 goto out_short_read;
7286 ret = read_one_chunk(&key, sb, chunk);
7291 sb_array_offset += len;
7294 clear_extent_buffer_uptodate(sb);
7295 free_extent_buffer_stale(sb);
7299 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7301 clear_extent_buffer_uptodate(sb);
7302 free_extent_buffer_stale(sb);
7307 * Check if all chunks in the fs are OK for read-write degraded mount
7309 * If the @failing_dev is specified, it's accounted as missing.
7311 * Return true if all chunks meet the minimal RW mount requirements.
7312 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7314 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7315 struct btrfs_device *failing_dev)
7317 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7318 struct extent_map *em;
7322 read_lock(&map_tree->lock);
7323 em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7324 read_unlock(&map_tree->lock);
7325 /* No chunk at all? Return false anyway */
7331 struct map_lookup *map;
7336 map = em->map_lookup;
7338 btrfs_get_num_tolerated_disk_barrier_failures(
7340 for (i = 0; i < map->num_stripes; i++) {
7341 struct btrfs_device *dev = map->stripes[i].dev;
7343 if (!dev || !dev->bdev ||
7344 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7345 dev->last_flush_error)
7347 else if (failing_dev && failing_dev == dev)
7350 if (missing > max_tolerated) {
7353 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7354 em->start, missing, max_tolerated);
7355 free_extent_map(em);
7359 next_start = extent_map_end(em);
7360 free_extent_map(em);
7362 read_lock(&map_tree->lock);
7363 em = lookup_extent_mapping(map_tree, next_start,
7364 (u64)(-1) - next_start);
7365 read_unlock(&map_tree->lock);
7371 static void readahead_tree_node_children(struct extent_buffer *node)
7374 const int nr_items = btrfs_header_nritems(node);
7376 for (i = 0; i < nr_items; i++)
7377 btrfs_readahead_node_child(node, i);
7380 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7382 struct btrfs_root *root = fs_info->chunk_root;
7383 struct btrfs_path *path;
7384 struct extent_buffer *leaf;
7385 struct btrfs_key key;
7386 struct btrfs_key found_key;
7391 u64 last_ra_node = 0;
7393 path = btrfs_alloc_path();
7398 * uuid_mutex is needed only if we are mounting a sprout FS
7399 * otherwise we don't need it.
7401 mutex_lock(&uuid_mutex);
7404 * It is possible for mount and umount to race in such a way that
7405 * we execute this code path, but open_fs_devices failed to clear
7406 * total_rw_bytes. We certainly want it cleared before reading the
7407 * device items, so clear it here.
7409 fs_info->fs_devices->total_rw_bytes = 0;
7412 * Lockdep complains about possible circular locking dependency between
7413 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores
7414 * used for freeze procection of a fs (struct super_block.s_writers),
7415 * which we take when starting a transaction, and extent buffers of the
7416 * chunk tree if we call read_one_dev() while holding a lock on an
7417 * extent buffer of the chunk tree. Since we are mounting the filesystem
7418 * and at this point there can't be any concurrent task modifying the
7419 * chunk tree, to keep it simple, just skip locking on the chunk tree.
7421 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
7422 path->skip_locking = 1;
7425 * Read all device items, and then all the chunk items. All
7426 * device items are found before any chunk item (their object id
7427 * is smaller than the lowest possible object id for a chunk
7428 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7430 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7433 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) {
7434 struct extent_buffer *node = path->nodes[1];
7436 leaf = path->nodes[0];
7437 slot = path->slots[0];
7440 if (last_ra_node != node->start) {
7441 readahead_tree_node_children(node);
7442 last_ra_node = node->start;
7445 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7446 struct btrfs_dev_item *dev_item;
7447 dev_item = btrfs_item_ptr(leaf, slot,
7448 struct btrfs_dev_item);
7449 ret = read_one_dev(leaf, dev_item);
7453 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7454 struct btrfs_chunk *chunk;
7457 * We are only called at mount time, so no need to take
7458 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7459 * we always lock first fs_info->chunk_mutex before
7460 * acquiring any locks on the chunk tree. This is a
7461 * requirement for chunk allocation, see the comment on
7462 * top of btrfs_chunk_alloc() for details.
7464 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7465 ret = read_one_chunk(&found_key, leaf, chunk);
7470 /* Catch error found during iteration */
7477 * After loading chunk tree, we've got all device information,
7478 * do another round of validation checks.
7480 if (total_dev != fs_info->fs_devices->total_devices) {
7482 "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit",
7483 btrfs_super_num_devices(fs_info->super_copy),
7485 fs_info->fs_devices->total_devices = total_dev;
7486 btrfs_set_super_num_devices(fs_info->super_copy, total_dev);
7488 if (btrfs_super_total_bytes(fs_info->super_copy) <
7489 fs_info->fs_devices->total_rw_bytes) {
7491 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7492 btrfs_super_total_bytes(fs_info->super_copy),
7493 fs_info->fs_devices->total_rw_bytes);
7499 mutex_unlock(&uuid_mutex);
7501 btrfs_free_path(path);
7505 int btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7507 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7508 struct btrfs_device *device;
7511 fs_devices->fs_info = fs_info;
7513 mutex_lock(&fs_devices->device_list_mutex);
7514 list_for_each_entry(device, &fs_devices->devices, dev_list)
7515 device->fs_info = fs_info;
7517 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7518 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7519 device->fs_info = fs_info;
7520 ret = btrfs_get_dev_zone_info(device, false);
7525 seed_devs->fs_info = fs_info;
7527 mutex_unlock(&fs_devices->device_list_mutex);
7532 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7533 const struct btrfs_dev_stats_item *ptr,
7538 read_extent_buffer(eb, &val,
7539 offsetof(struct btrfs_dev_stats_item, values) +
7540 ((unsigned long)ptr) + (index * sizeof(u64)),
7545 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7546 struct btrfs_dev_stats_item *ptr,
7549 write_extent_buffer(eb, &val,
7550 offsetof(struct btrfs_dev_stats_item, values) +
7551 ((unsigned long)ptr) + (index * sizeof(u64)),
7555 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7556 struct btrfs_path *path)
7558 struct btrfs_dev_stats_item *ptr;
7559 struct extent_buffer *eb;
7560 struct btrfs_key key;
7564 if (!device->fs_info->dev_root)
7567 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7568 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7569 key.offset = device->devid;
7570 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7572 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7573 btrfs_dev_stat_set(device, i, 0);
7574 device->dev_stats_valid = 1;
7575 btrfs_release_path(path);
7576 return ret < 0 ? ret : 0;
7578 slot = path->slots[0];
7579 eb = path->nodes[0];
7580 item_size = btrfs_item_size(eb, slot);
7582 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7584 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7585 if (item_size >= (1 + i) * sizeof(__le64))
7586 btrfs_dev_stat_set(device, i,
7587 btrfs_dev_stats_value(eb, ptr, i));
7589 btrfs_dev_stat_set(device, i, 0);
7592 device->dev_stats_valid = 1;
7593 btrfs_dev_stat_print_on_load(device);
7594 btrfs_release_path(path);
7599 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7601 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7602 struct btrfs_device *device;
7603 struct btrfs_path *path = NULL;
7606 path = btrfs_alloc_path();
7610 mutex_lock(&fs_devices->device_list_mutex);
7611 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7612 ret = btrfs_device_init_dev_stats(device, path);
7616 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7617 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7618 ret = btrfs_device_init_dev_stats(device, path);
7624 mutex_unlock(&fs_devices->device_list_mutex);
7626 btrfs_free_path(path);
7630 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7631 struct btrfs_device *device)
7633 struct btrfs_fs_info *fs_info = trans->fs_info;
7634 struct btrfs_root *dev_root = fs_info->dev_root;
7635 struct btrfs_path *path;
7636 struct btrfs_key key;
7637 struct extent_buffer *eb;
7638 struct btrfs_dev_stats_item *ptr;
7642 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7643 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7644 key.offset = device->devid;
7646 path = btrfs_alloc_path();
7649 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7651 btrfs_warn_in_rcu(fs_info,
7652 "error %d while searching for dev_stats item for device %s",
7653 ret, btrfs_dev_name(device));
7658 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7659 /* need to delete old one and insert a new one */
7660 ret = btrfs_del_item(trans, dev_root, path);
7662 btrfs_warn_in_rcu(fs_info,
7663 "delete too small dev_stats item for device %s failed %d",
7664 btrfs_dev_name(device), ret);
7671 /* need to insert a new item */
7672 btrfs_release_path(path);
7673 ret = btrfs_insert_empty_item(trans, dev_root, path,
7674 &key, sizeof(*ptr));
7676 btrfs_warn_in_rcu(fs_info,
7677 "insert dev_stats item for device %s failed %d",
7678 btrfs_dev_name(device), ret);
7683 eb = path->nodes[0];
7684 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7685 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7686 btrfs_set_dev_stats_value(eb, ptr, i,
7687 btrfs_dev_stat_read(device, i));
7688 btrfs_mark_buffer_dirty(eb);
7691 btrfs_free_path(path);
7696 * called from commit_transaction. Writes all changed device stats to disk.
7698 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7700 struct btrfs_fs_info *fs_info = trans->fs_info;
7701 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7702 struct btrfs_device *device;
7706 mutex_lock(&fs_devices->device_list_mutex);
7707 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7708 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7709 if (!device->dev_stats_valid || stats_cnt == 0)
7714 * There is a LOAD-LOAD control dependency between the value of
7715 * dev_stats_ccnt and updating the on-disk values which requires
7716 * reading the in-memory counters. Such control dependencies
7717 * require explicit read memory barriers.
7719 * This memory barriers pairs with smp_mb__before_atomic in
7720 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7721 * barrier implied by atomic_xchg in
7722 * btrfs_dev_stats_read_and_reset
7726 ret = update_dev_stat_item(trans, device);
7728 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7730 mutex_unlock(&fs_devices->device_list_mutex);
7735 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7737 btrfs_dev_stat_inc(dev, index);
7739 if (!dev->dev_stats_valid)
7741 btrfs_err_rl_in_rcu(dev->fs_info,
7742 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7743 btrfs_dev_name(dev),
7744 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7745 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7746 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7747 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7748 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7751 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7755 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7756 if (btrfs_dev_stat_read(dev, i) != 0)
7758 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7759 return; /* all values == 0, suppress message */
7761 btrfs_info_in_rcu(dev->fs_info,
7762 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7763 btrfs_dev_name(dev),
7764 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7765 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7766 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7767 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7768 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7771 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7772 struct btrfs_ioctl_get_dev_stats *stats)
7774 BTRFS_DEV_LOOKUP_ARGS(args);
7775 struct btrfs_device *dev;
7776 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7779 mutex_lock(&fs_devices->device_list_mutex);
7780 args.devid = stats->devid;
7781 dev = btrfs_find_device(fs_info->fs_devices, &args);
7782 mutex_unlock(&fs_devices->device_list_mutex);
7785 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7787 } else if (!dev->dev_stats_valid) {
7788 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7790 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7791 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7792 if (stats->nr_items > i)
7794 btrfs_dev_stat_read_and_reset(dev, i);
7796 btrfs_dev_stat_set(dev, i, 0);
7798 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7799 current->comm, task_pid_nr(current));
7801 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7802 if (stats->nr_items > i)
7803 stats->values[i] = btrfs_dev_stat_read(dev, i);
7805 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7806 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7811 * Update the size and bytes used for each device where it changed. This is
7812 * delayed since we would otherwise get errors while writing out the
7815 * Must be invoked during transaction commit.
7817 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7819 struct btrfs_device *curr, *next;
7821 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7823 if (list_empty(&trans->dev_update_list))
7827 * We don't need the device_list_mutex here. This list is owned by the
7828 * transaction and the transaction must complete before the device is
7831 mutex_lock(&trans->fs_info->chunk_mutex);
7832 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7834 list_del_init(&curr->post_commit_list);
7835 curr->commit_total_bytes = curr->disk_total_bytes;
7836 curr->commit_bytes_used = curr->bytes_used;
7838 mutex_unlock(&trans->fs_info->chunk_mutex);
7842 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7844 int btrfs_bg_type_to_factor(u64 flags)
7846 const int index = btrfs_bg_flags_to_raid_index(flags);
7848 return btrfs_raid_array[index].ncopies;
7853 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7854 u64 chunk_offset, u64 devid,
7855 u64 physical_offset, u64 physical_len)
7857 struct btrfs_dev_lookup_args args = { .devid = devid };
7858 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7859 struct extent_map *em;
7860 struct map_lookup *map;
7861 struct btrfs_device *dev;
7867 read_lock(&em_tree->lock);
7868 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7869 read_unlock(&em_tree->lock);
7873 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7874 physical_offset, devid);
7879 map = em->map_lookup;
7880 stripe_len = btrfs_calc_stripe_length(em);
7881 if (physical_len != stripe_len) {
7883 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7884 physical_offset, devid, em->start, physical_len,
7891 * Very old mkfs.btrfs (before v4.1) will not respect the reserved
7892 * space. Although kernel can handle it without problem, better to warn
7895 if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED)
7897 "devid %llu physical %llu len %llu inside the reserved space",
7898 devid, physical_offset, physical_len);
7900 for (i = 0; i < map->num_stripes; i++) {
7901 if (map->stripes[i].dev->devid == devid &&
7902 map->stripes[i].physical == physical_offset) {
7904 if (map->verified_stripes >= map->num_stripes) {
7906 "too many dev extents for chunk %llu found",
7911 map->verified_stripes++;
7917 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7918 physical_offset, devid);
7922 /* Make sure no dev extent is beyond device boundary */
7923 dev = btrfs_find_device(fs_info->fs_devices, &args);
7925 btrfs_err(fs_info, "failed to find devid %llu", devid);
7930 if (physical_offset + physical_len > dev->disk_total_bytes) {
7932 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7933 devid, physical_offset, physical_len,
7934 dev->disk_total_bytes);
7939 if (dev->zone_info) {
7940 u64 zone_size = dev->zone_info->zone_size;
7942 if (!IS_ALIGNED(physical_offset, zone_size) ||
7943 !IS_ALIGNED(physical_len, zone_size)) {
7945 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
7946 devid, physical_offset, physical_len);
7953 free_extent_map(em);
7957 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7959 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7960 struct extent_map *em;
7961 struct rb_node *node;
7964 read_lock(&em_tree->lock);
7965 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
7966 em = rb_entry(node, struct extent_map, rb_node);
7967 if (em->map_lookup->num_stripes !=
7968 em->map_lookup->verified_stripes) {
7970 "chunk %llu has missing dev extent, have %d expect %d",
7971 em->start, em->map_lookup->verified_stripes,
7972 em->map_lookup->num_stripes);
7978 read_unlock(&em_tree->lock);
7983 * Ensure that all dev extents are mapped to correct chunk, otherwise
7984 * later chunk allocation/free would cause unexpected behavior.
7986 * NOTE: This will iterate through the whole device tree, which should be of
7987 * the same size level as the chunk tree. This slightly increases mount time.
7989 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
7991 struct btrfs_path *path;
7992 struct btrfs_root *root = fs_info->dev_root;
7993 struct btrfs_key key;
7995 u64 prev_dev_ext_end = 0;
7999 * We don't have a dev_root because we mounted with ignorebadroots and
8000 * failed to load the root, so we want to skip the verification in this
8003 * However if the dev root is fine, but the tree itself is corrupted
8004 * we'd still fail to mount. This verification is only to make sure
8005 * writes can happen safely, so instead just bypass this check
8006 * completely in the case of IGNOREBADROOTS.
8008 if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
8012 key.type = BTRFS_DEV_EXTENT_KEY;
8015 path = btrfs_alloc_path();
8019 path->reada = READA_FORWARD;
8020 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8024 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
8025 ret = btrfs_next_leaf(root, path);
8028 /* No dev extents at all? Not good */
8035 struct extent_buffer *leaf = path->nodes[0];
8036 struct btrfs_dev_extent *dext;
8037 int slot = path->slots[0];
8039 u64 physical_offset;
8043 btrfs_item_key_to_cpu(leaf, &key, slot);
8044 if (key.type != BTRFS_DEV_EXTENT_KEY)
8046 devid = key.objectid;
8047 physical_offset = key.offset;
8049 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
8050 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
8051 physical_len = btrfs_dev_extent_length(leaf, dext);
8053 /* Check if this dev extent overlaps with the previous one */
8054 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
8056 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
8057 devid, physical_offset, prev_dev_ext_end);
8062 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
8063 physical_offset, physical_len);
8067 prev_dev_ext_end = physical_offset + physical_len;
8069 ret = btrfs_next_item(root, path);
8078 /* Ensure all chunks have corresponding dev extents */
8079 ret = verify_chunk_dev_extent_mapping(fs_info);
8081 btrfs_free_path(path);
8086 * Check whether the given block group or device is pinned by any inode being
8087 * used as a swapfile.
8089 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
8091 struct btrfs_swapfile_pin *sp;
8092 struct rb_node *node;
8094 spin_lock(&fs_info->swapfile_pins_lock);
8095 node = fs_info->swapfile_pins.rb_node;
8097 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
8099 node = node->rb_left;
8100 else if (ptr > sp->ptr)
8101 node = node->rb_right;
8105 spin_unlock(&fs_info->swapfile_pins_lock);
8106 return node != NULL;
8109 static int relocating_repair_kthread(void *data)
8111 struct btrfs_block_group *cache = data;
8112 struct btrfs_fs_info *fs_info = cache->fs_info;
8116 target = cache->start;
8117 btrfs_put_block_group(cache);
8119 sb_start_write(fs_info->sb);
8120 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
8122 "zoned: skip relocating block group %llu to repair: EBUSY",
8124 sb_end_write(fs_info->sb);
8128 mutex_lock(&fs_info->reclaim_bgs_lock);
8130 /* Ensure block group still exists */
8131 cache = btrfs_lookup_block_group(fs_info, target);
8135 if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags))
8138 ret = btrfs_may_alloc_data_chunk(fs_info, target);
8143 "zoned: relocating block group %llu to repair IO failure",
8145 ret = btrfs_relocate_chunk(fs_info, target);
8149 btrfs_put_block_group(cache);
8150 mutex_unlock(&fs_info->reclaim_bgs_lock);
8151 btrfs_exclop_finish(fs_info);
8152 sb_end_write(fs_info->sb);
8157 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical)
8159 struct btrfs_block_group *cache;
8161 if (!btrfs_is_zoned(fs_info))
8164 /* Do not attempt to repair in degraded state */
8165 if (btrfs_test_opt(fs_info, DEGRADED))
8168 cache = btrfs_lookup_block_group(fs_info, logical);
8172 if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) {
8173 btrfs_put_block_group(cache);
8177 kthread_run(relocating_repair_kthread, cache,
8178 "btrfs-relocating-repair");