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>
18 #include "transaction.h"
21 #include "rcu-string.h"
22 #include "dev-replace.h"
24 #include "tree-checker.h"
25 #include "space-info.h"
26 #include "block-group.h"
30 #include "accessors.h"
31 #include "uuid-tree.h"
33 #include "relocation.h"
36 #include "raid-stripe-tree.h"
38 #define BTRFS_BLOCK_GROUP_STRIPE_MASK (BTRFS_BLOCK_GROUP_RAID0 | \
39 BTRFS_BLOCK_GROUP_RAID10 | \
40 BTRFS_BLOCK_GROUP_RAID56_MASK)
42 struct btrfs_io_geometry {
48 u64 raid56_full_stripe_start;
53 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
54 [BTRFS_RAID_RAID10] = {
57 .devs_max = 0, /* 0 == as many as possible */
59 .tolerated_failures = 1,
63 .raid_name = "raid10",
64 .bg_flag = BTRFS_BLOCK_GROUP_RAID10,
65 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
67 [BTRFS_RAID_RAID1] = {
72 .tolerated_failures = 1,
77 .bg_flag = BTRFS_BLOCK_GROUP_RAID1,
78 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
80 [BTRFS_RAID_RAID1C3] = {
85 .tolerated_failures = 2,
89 .raid_name = "raid1c3",
90 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3,
91 .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET,
93 [BTRFS_RAID_RAID1C4] = {
98 .tolerated_failures = 3,
102 .raid_name = "raid1c4",
103 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4,
104 .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET,
111 .tolerated_failures = 0,
116 .bg_flag = BTRFS_BLOCK_GROUP_DUP,
119 [BTRFS_RAID_RAID0] = {
124 .tolerated_failures = 0,
128 .raid_name = "raid0",
129 .bg_flag = BTRFS_BLOCK_GROUP_RAID0,
132 [BTRFS_RAID_SINGLE] = {
137 .tolerated_failures = 0,
141 .raid_name = "single",
145 [BTRFS_RAID_RAID5] = {
150 .tolerated_failures = 1,
154 .raid_name = "raid5",
155 .bg_flag = BTRFS_BLOCK_GROUP_RAID5,
156 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
158 [BTRFS_RAID_RAID6] = {
163 .tolerated_failures = 2,
167 .raid_name = "raid6",
168 .bg_flag = BTRFS_BLOCK_GROUP_RAID6,
169 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
174 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which
175 * can be used as index to access btrfs_raid_array[].
177 enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags)
179 const u64 profile = (flags & BTRFS_BLOCK_GROUP_PROFILE_MASK);
182 return BTRFS_RAID_SINGLE;
184 return BTRFS_BG_FLAG_TO_INDEX(profile);
187 const char *btrfs_bg_type_to_raid_name(u64 flags)
189 const int index = btrfs_bg_flags_to_raid_index(flags);
191 if (index >= BTRFS_NR_RAID_TYPES)
194 return btrfs_raid_array[index].raid_name;
197 int btrfs_nr_parity_stripes(u64 type)
199 enum btrfs_raid_types index = btrfs_bg_flags_to_raid_index(type);
201 return btrfs_raid_array[index].nparity;
205 * Fill @buf with textual description of @bg_flags, no more than @size_buf
206 * bytes including terminating null byte.
208 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf)
213 u64 flags = bg_flags;
214 u32 size_bp = size_buf;
221 #define DESCRIBE_FLAG(flag, desc) \
223 if (flags & (flag)) { \
224 ret = snprintf(bp, size_bp, "%s|", (desc)); \
225 if (ret < 0 || ret >= size_bp) \
233 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data");
234 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system");
235 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata");
237 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single");
238 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
239 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag,
240 btrfs_raid_array[i].raid_name);
244 ret = snprintf(bp, size_bp, "0x%llx|", flags);
248 if (size_bp < size_buf)
249 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */
252 * The text is trimmed, it's up to the caller to provide sufficiently
258 static int init_first_rw_device(struct btrfs_trans_handle *trans);
259 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
260 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
266 * There are several mutexes that protect manipulation of devices and low-level
267 * structures like chunks but not block groups, extents or files
269 * uuid_mutex (global lock)
270 * ------------------------
271 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
272 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
273 * device) or requested by the device= mount option
275 * the mutex can be very coarse and can cover long-running operations
277 * protects: updates to fs_devices counters like missing devices, rw devices,
278 * seeding, structure cloning, opening/closing devices at mount/umount time
280 * global::fs_devs - add, remove, updates to the global list
282 * does not protect: manipulation of the fs_devices::devices list in general
283 * but in mount context it could be used to exclude list modifications by eg.
286 * btrfs_device::name - renames (write side), read is RCU
288 * fs_devices::device_list_mutex (per-fs, with RCU)
289 * ------------------------------------------------
290 * protects updates to fs_devices::devices, ie. adding and deleting
292 * simple list traversal with read-only actions can be done with RCU protection
294 * may be used to exclude some operations from running concurrently without any
295 * modifications to the list (see write_all_supers)
297 * Is not required at mount and close times, because our device list is
298 * protected by the uuid_mutex at that point.
302 * protects balance structures (status, state) and context accessed from
303 * several places (internally, ioctl)
307 * protects chunks, adding or removing during allocation, trim or when a new
308 * device is added/removed. Additionally it also protects post_commit_list of
309 * individual devices, since they can be added to the transaction's
310 * post_commit_list only with chunk_mutex held.
314 * a big lock that is held by the cleaner thread and prevents running subvolume
315 * cleaning together with relocation or delayed iputs
327 * Exclusive operations
328 * ====================
330 * Maintains the exclusivity of the following operations that apply to the
331 * whole filesystem and cannot run in parallel.
336 * - Device replace (*)
339 * The device operations (as above) can be in one of the following states:
345 * Only device operations marked with (*) can go into the Paused state for the
348 * - ioctl (only Balance can be Paused through ioctl)
349 * - filesystem remounted as read-only
350 * - filesystem unmounted and mounted as read-only
351 * - system power-cycle and filesystem mounted as read-only
352 * - filesystem or device errors leading to forced read-only
354 * The status of exclusive operation is set and cleared atomically.
355 * During the course of Paused state, fs_info::exclusive_operation remains set.
356 * A device operation in Paused or Running state can be canceled or resumed
357 * either by ioctl (Balance only) or when remounted as read-write.
358 * The exclusive status is cleared when the device operation is canceled or
362 DEFINE_MUTEX(uuid_mutex);
363 static LIST_HEAD(fs_uuids);
364 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void)
370 * Allocate new btrfs_fs_devices structure identified by a fsid.
372 * @fsid: if not NULL, copy the UUID to fs_devices::fsid and to
373 * fs_devices::metadata_fsid
375 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
376 * The returned struct is not linked onto any lists and can be destroyed with
377 * kfree() right away.
379 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
381 struct btrfs_fs_devices *fs_devs;
383 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
385 return ERR_PTR(-ENOMEM);
387 mutex_init(&fs_devs->device_list_mutex);
389 INIT_LIST_HEAD(&fs_devs->devices);
390 INIT_LIST_HEAD(&fs_devs->alloc_list);
391 INIT_LIST_HEAD(&fs_devs->fs_list);
392 INIT_LIST_HEAD(&fs_devs->seed_list);
395 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
396 memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE);
402 static void btrfs_free_device(struct btrfs_device *device)
404 WARN_ON(!list_empty(&device->post_commit_list));
405 rcu_string_free(device->name);
406 extent_io_tree_release(&device->alloc_state);
407 btrfs_destroy_dev_zone_info(device);
411 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
413 struct btrfs_device *device;
415 WARN_ON(fs_devices->opened);
416 while (!list_empty(&fs_devices->devices)) {
417 device = list_entry(fs_devices->devices.next,
418 struct btrfs_device, dev_list);
419 list_del(&device->dev_list);
420 btrfs_free_device(device);
425 void __exit btrfs_cleanup_fs_uuids(void)
427 struct btrfs_fs_devices *fs_devices;
429 while (!list_empty(&fs_uuids)) {
430 fs_devices = list_entry(fs_uuids.next,
431 struct btrfs_fs_devices, fs_list);
432 list_del(&fs_devices->fs_list);
433 free_fs_devices(fs_devices);
437 static bool match_fsid_fs_devices(const struct btrfs_fs_devices *fs_devices,
438 const u8 *fsid, const u8 *metadata_fsid)
440 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) != 0)
446 if (memcmp(metadata_fsid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE) != 0)
452 static noinline struct btrfs_fs_devices *find_fsid(
453 const u8 *fsid, const u8 *metadata_fsid)
455 struct btrfs_fs_devices *fs_devices;
459 /* Handle non-split brain cases */
460 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
461 if (match_fsid_fs_devices(fs_devices, fsid, metadata_fsid))
468 btrfs_get_bdev_and_sb(const char *device_path, blk_mode_t flags, void *holder,
469 int flush, struct file **bdev_file,
470 struct btrfs_super_block **disk_super)
472 struct block_device *bdev;
475 *bdev_file = bdev_file_open_by_path(device_path, flags, holder, NULL);
477 if (IS_ERR(*bdev_file)) {
478 ret = PTR_ERR(*bdev_file);
481 bdev = file_bdev(*bdev_file);
485 ret = set_blocksize(bdev, BTRFS_BDEV_BLOCKSIZE);
490 invalidate_bdev(bdev);
491 *disk_super = btrfs_read_dev_super(bdev);
492 if (IS_ERR(*disk_super)) {
493 ret = PTR_ERR(*disk_super);
506 * Search and remove all stale devices (which are not mounted). When both
507 * inputs are NULL, it will search and release all stale devices.
509 * @devt: Optional. When provided will it release all unmounted devices
510 * matching this devt only.
511 * @skip_device: Optional. Will skip this device when searching for the stale
514 * Return: 0 for success or if @devt is 0.
515 * -EBUSY if @devt is a mounted device.
516 * -ENOENT if @devt does not match any device in the list.
518 static int btrfs_free_stale_devices(dev_t devt, struct btrfs_device *skip_device)
520 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
521 struct btrfs_device *device, *tmp_device;
525 lockdep_assert_held(&uuid_mutex);
527 /* Return good status if there is no instance of devt. */
529 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
531 mutex_lock(&fs_devices->device_list_mutex);
532 list_for_each_entry_safe(device, tmp_device,
533 &fs_devices->devices, dev_list) {
534 if (skip_device && skip_device == device)
536 if (devt && devt != device->devt)
538 if (fs_devices->opened) {
544 /* delete the stale device */
545 fs_devices->num_devices--;
546 list_del(&device->dev_list);
547 btrfs_free_device(device);
551 mutex_unlock(&fs_devices->device_list_mutex);
553 if (fs_devices->num_devices == 0) {
554 btrfs_sysfs_remove_fsid(fs_devices);
555 list_del(&fs_devices->fs_list);
556 free_fs_devices(fs_devices);
560 /* If there is at least one freed device return 0. */
567 static struct btrfs_fs_devices *find_fsid_by_device(
568 struct btrfs_super_block *disk_super,
569 dev_t devt, bool *same_fsid_diff_dev)
571 struct btrfs_fs_devices *fsid_fs_devices;
572 struct btrfs_fs_devices *devt_fs_devices;
573 const bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
574 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
575 bool found_by_devt = false;
577 /* Find the fs_device by the usual method, if found use it. */
578 fsid_fs_devices = find_fsid(disk_super->fsid,
579 has_metadata_uuid ? disk_super->metadata_uuid : NULL);
581 /* The temp_fsid feature is supported only with single device filesystem. */
582 if (btrfs_super_num_devices(disk_super) != 1)
583 return fsid_fs_devices;
586 * A seed device is an integral component of the sprout device, which
587 * functions as a multi-device filesystem. So, temp-fsid feature is
590 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING)
591 return fsid_fs_devices;
593 /* Try to find a fs_devices by matching devt. */
594 list_for_each_entry(devt_fs_devices, &fs_uuids, fs_list) {
595 struct btrfs_device *device;
597 list_for_each_entry(device, &devt_fs_devices->devices, dev_list) {
598 if (device->devt == devt) {
599 found_by_devt = true;
608 /* Existing device. */
609 if (fsid_fs_devices == NULL) {
610 if (devt_fs_devices->opened == 0) {
614 /* temp_fsid is mounting a subvol. */
615 return devt_fs_devices;
618 /* Regular or temp_fsid device mounting a subvol. */
619 return devt_fs_devices;
623 if (fsid_fs_devices == NULL) {
626 /* sb::fsid is already used create a new temp_fsid. */
627 *same_fsid_diff_dev = true;
636 * This is only used on mount, and we are protected from competing things
637 * messing with our fs_devices by the uuid_mutex, thus we do not need the
638 * fs_devices->device_list_mutex here.
640 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
641 struct btrfs_device *device, blk_mode_t flags,
644 struct file *bdev_file;
645 struct btrfs_super_block *disk_super;
654 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
655 &bdev_file, &disk_super);
659 devid = btrfs_stack_device_id(&disk_super->dev_item);
660 if (devid != device->devid)
661 goto error_free_page;
663 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
664 goto error_free_page;
666 device->generation = btrfs_super_generation(disk_super);
668 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
669 if (btrfs_super_incompat_flags(disk_super) &
670 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
672 "BTRFS: Invalid seeding and uuid-changed device detected\n");
673 goto error_free_page;
676 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
677 fs_devices->seeding = true;
679 if (bdev_read_only(file_bdev(bdev_file)))
680 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
682 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
685 if (!bdev_nonrot(file_bdev(bdev_file)))
686 fs_devices->rotating = true;
688 if (bdev_max_discard_sectors(file_bdev(bdev_file)))
689 fs_devices->discardable = true;
691 device->bdev_file = bdev_file;
692 device->bdev = file_bdev(bdev_file);
693 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
695 fs_devices->open_devices++;
696 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
697 device->devid != BTRFS_DEV_REPLACE_DEVID) {
698 fs_devices->rw_devices++;
699 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
701 btrfs_release_disk_super(disk_super);
706 btrfs_release_disk_super(disk_super);
712 u8 *btrfs_sb_fsid_ptr(struct btrfs_super_block *sb)
714 bool has_metadata_uuid = (btrfs_super_incompat_flags(sb) &
715 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
717 return has_metadata_uuid ? sb->metadata_uuid : sb->fsid;
721 * Add new device to list of registered devices
724 * device pointer which was just added or updated when successful
725 * error pointer when failed
727 static noinline struct btrfs_device *device_list_add(const char *path,
728 struct btrfs_super_block *disk_super,
729 bool *new_device_added)
731 struct btrfs_device *device;
732 struct btrfs_fs_devices *fs_devices = NULL;
733 struct rcu_string *name;
734 u64 found_transid = btrfs_super_generation(disk_super);
735 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
738 bool same_fsid_diff_dev = false;
739 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
740 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
742 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
744 "device %s has incomplete metadata_uuid change, please use btrfstune to complete",
746 return ERR_PTR(-EAGAIN);
749 error = lookup_bdev(path, &path_devt);
751 btrfs_err(NULL, "failed to lookup block device for path %s: %d",
753 return ERR_PTR(error);
756 fs_devices = find_fsid_by_device(disk_super, path_devt, &same_fsid_diff_dev);
759 fs_devices = alloc_fs_devices(disk_super->fsid);
760 if (IS_ERR(fs_devices))
761 return ERR_CAST(fs_devices);
763 if (has_metadata_uuid)
764 memcpy(fs_devices->metadata_uuid,
765 disk_super->metadata_uuid, BTRFS_FSID_SIZE);
767 if (same_fsid_diff_dev) {
768 generate_random_uuid(fs_devices->fsid);
769 fs_devices->temp_fsid = true;
770 pr_info("BTRFS: device %s (%d:%d) using temp-fsid %pU\n",
771 path, MAJOR(path_devt), MINOR(path_devt),
775 mutex_lock(&fs_devices->device_list_mutex);
776 list_add(&fs_devices->fs_list, &fs_uuids);
780 struct btrfs_dev_lookup_args args = {
782 .uuid = disk_super->dev_item.uuid,
785 mutex_lock(&fs_devices->device_list_mutex);
786 device = btrfs_find_device(fs_devices, &args);
788 if (found_transid > fs_devices->latest_generation) {
789 memcpy(fs_devices->fsid, disk_super->fsid,
791 memcpy(fs_devices->metadata_uuid,
792 btrfs_sb_fsid_ptr(disk_super), BTRFS_FSID_SIZE);
797 unsigned int nofs_flag;
799 if (fs_devices->opened) {
801 "device %s (%d:%d) belongs to fsid %pU, and the fs is already mounted, scanned by %s (%d)",
802 path, MAJOR(path_devt), MINOR(path_devt),
803 fs_devices->fsid, current->comm,
804 task_pid_nr(current));
805 mutex_unlock(&fs_devices->device_list_mutex);
806 return ERR_PTR(-EBUSY);
809 nofs_flag = memalloc_nofs_save();
810 device = btrfs_alloc_device(NULL, &devid,
811 disk_super->dev_item.uuid, path);
812 memalloc_nofs_restore(nofs_flag);
813 if (IS_ERR(device)) {
814 mutex_unlock(&fs_devices->device_list_mutex);
815 /* we can safely leave the fs_devices entry around */
819 device->devt = path_devt;
821 list_add_rcu(&device->dev_list, &fs_devices->devices);
822 fs_devices->num_devices++;
824 device->fs_devices = fs_devices;
825 *new_device_added = true;
827 if (disk_super->label[0])
829 "BTRFS: device label %s devid %llu transid %llu %s (%d:%d) scanned by %s (%d)\n",
830 disk_super->label, devid, found_transid, path,
831 MAJOR(path_devt), MINOR(path_devt),
832 current->comm, task_pid_nr(current));
835 "BTRFS: device fsid %pU devid %llu transid %llu %s (%d:%d) scanned by %s (%d)\n",
836 disk_super->fsid, devid, found_transid, path,
837 MAJOR(path_devt), MINOR(path_devt),
838 current->comm, task_pid_nr(current));
840 } else if (!device->name || strcmp(device->name->str, path)) {
842 * When FS is already mounted.
843 * 1. If you are here and if the device->name is NULL that
844 * means this device was missing at time of FS mount.
845 * 2. If you are here and if the device->name is different
846 * from 'path' that means either
847 * a. The same device disappeared and reappeared with
849 * b. The missing-disk-which-was-replaced, has
852 * We must allow 1 and 2a above. But 2b would be a spurious
855 * Further in case of 1 and 2a above, the disk at 'path'
856 * would have missed some transaction when it was away and
857 * in case of 2a the stale bdev has to be updated as well.
858 * 2b must not be allowed at all time.
862 * For now, we do allow update to btrfs_fs_device through the
863 * btrfs dev scan cli after FS has been mounted. We're still
864 * tracking a problem where systems fail mount by subvolume id
865 * when we reject replacement on a mounted FS.
867 if (!fs_devices->opened && found_transid < device->generation) {
869 * That is if the FS is _not_ mounted and if you
870 * are here, that means there is more than one
871 * disk with same uuid and devid.We keep the one
872 * with larger generation number or the last-in if
873 * generation are equal.
875 mutex_unlock(&fs_devices->device_list_mutex);
877 "device %s already registered with a higher generation, found %llu expect %llu",
878 path, found_transid, device->generation);
879 return ERR_PTR(-EEXIST);
883 * We are going to replace the device path for a given devid,
884 * make sure it's the same device if the device is mounted
886 * NOTE: the device->fs_info may not be reliable here so pass
887 * in a NULL to message helpers instead. This avoids a possible
888 * use-after-free when the fs_info and fs_info->sb are already
892 if (device->devt != path_devt) {
893 mutex_unlock(&fs_devices->device_list_mutex);
894 btrfs_warn_in_rcu(NULL,
895 "duplicate device %s devid %llu generation %llu scanned by %s (%d)",
896 path, devid, found_transid,
898 task_pid_nr(current));
899 return ERR_PTR(-EEXIST);
901 btrfs_info_in_rcu(NULL,
902 "devid %llu device path %s changed to %s scanned by %s (%d)",
903 devid, btrfs_dev_name(device),
905 task_pid_nr(current));
908 name = rcu_string_strdup(path, GFP_NOFS);
910 mutex_unlock(&fs_devices->device_list_mutex);
911 return ERR_PTR(-ENOMEM);
913 rcu_string_free(device->name);
914 rcu_assign_pointer(device->name, name);
915 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
916 fs_devices->missing_devices--;
917 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
919 device->devt = path_devt;
923 * Unmount does not free the btrfs_device struct but would zero
924 * generation along with most of the other members. So just update
925 * it back. We need it to pick the disk with largest generation
928 if (!fs_devices->opened) {
929 device->generation = found_transid;
930 fs_devices->latest_generation = max_t(u64, found_transid,
931 fs_devices->latest_generation);
934 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
936 mutex_unlock(&fs_devices->device_list_mutex);
940 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
942 struct btrfs_fs_devices *fs_devices;
943 struct btrfs_device *device;
944 struct btrfs_device *orig_dev;
947 lockdep_assert_held(&uuid_mutex);
949 fs_devices = alloc_fs_devices(orig->fsid);
950 if (IS_ERR(fs_devices))
953 fs_devices->total_devices = orig->total_devices;
955 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
956 const char *dev_path = NULL;
959 * This is ok to do without RCU read locked because we hold the
960 * uuid mutex so nothing we touch in here is going to disappear.
963 dev_path = orig_dev->name->str;
965 device = btrfs_alloc_device(NULL, &orig_dev->devid,
966 orig_dev->uuid, dev_path);
967 if (IS_ERR(device)) {
968 ret = PTR_ERR(device);
972 if (orig_dev->zone_info) {
973 struct btrfs_zoned_device_info *zone_info;
975 zone_info = btrfs_clone_dev_zone_info(orig_dev);
977 btrfs_free_device(device);
981 device->zone_info = zone_info;
984 list_add(&device->dev_list, &fs_devices->devices);
985 device->fs_devices = fs_devices;
986 fs_devices->num_devices++;
990 free_fs_devices(fs_devices);
994 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices,
995 struct btrfs_device **latest_dev)
997 struct btrfs_device *device, *next;
999 /* This is the initialized path, it is safe to release the devices. */
1000 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1001 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) {
1002 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1003 &device->dev_state) &&
1004 !test_bit(BTRFS_DEV_STATE_MISSING,
1005 &device->dev_state) &&
1007 device->generation > (*latest_dev)->generation)) {
1008 *latest_dev = device;
1014 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID,
1015 * in btrfs_init_dev_replace() so just continue.
1017 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1020 if (device->bdev_file) {
1021 fput(device->bdev_file);
1022 device->bdev = NULL;
1023 device->bdev_file = NULL;
1024 fs_devices->open_devices--;
1026 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1027 list_del_init(&device->dev_alloc_list);
1028 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1029 fs_devices->rw_devices--;
1031 list_del_init(&device->dev_list);
1032 fs_devices->num_devices--;
1033 btrfs_free_device(device);
1039 * After we have read the system tree and know devids belonging to this
1040 * filesystem, remove the device which does not belong there.
1042 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices)
1044 struct btrfs_device *latest_dev = NULL;
1045 struct btrfs_fs_devices *seed_dev;
1047 mutex_lock(&uuid_mutex);
1048 __btrfs_free_extra_devids(fs_devices, &latest_dev);
1050 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list)
1051 __btrfs_free_extra_devids(seed_dev, &latest_dev);
1053 fs_devices->latest_dev = latest_dev;
1055 mutex_unlock(&uuid_mutex);
1058 static void btrfs_close_bdev(struct btrfs_device *device)
1063 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1064 sync_blockdev(device->bdev);
1065 invalidate_bdev(device->bdev);
1068 fput(device->bdev_file);
1071 static void btrfs_close_one_device(struct btrfs_device *device)
1073 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1075 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1076 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1077 list_del_init(&device->dev_alloc_list);
1078 fs_devices->rw_devices--;
1081 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1082 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
1084 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
1085 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
1086 fs_devices->missing_devices--;
1089 btrfs_close_bdev(device);
1091 fs_devices->open_devices--;
1092 device->bdev = NULL;
1094 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1095 btrfs_destroy_dev_zone_info(device);
1097 device->fs_info = NULL;
1098 atomic_set(&device->dev_stats_ccnt, 0);
1099 extent_io_tree_release(&device->alloc_state);
1102 * Reset the flush error record. We might have a transient flush error
1103 * in this mount, and if so we aborted the current transaction and set
1104 * the fs to an error state, guaranteeing no super blocks can be further
1105 * committed. However that error might be transient and if we unmount the
1106 * filesystem and mount it again, we should allow the mount to succeed
1107 * (btrfs_check_rw_degradable() should not fail) - if after mounting the
1108 * filesystem again we still get flush errors, then we will again abort
1109 * any transaction and set the error state, guaranteeing no commits of
1110 * unsafe super blocks.
1112 device->last_flush_error = 0;
1114 /* Verify the device is back in a pristine state */
1115 WARN_ON(test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1116 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1117 WARN_ON(!list_empty(&device->dev_alloc_list));
1118 WARN_ON(!list_empty(&device->post_commit_list));
1121 static void close_fs_devices(struct btrfs_fs_devices *fs_devices)
1123 struct btrfs_device *device, *tmp;
1125 lockdep_assert_held(&uuid_mutex);
1127 if (--fs_devices->opened > 0)
1130 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list)
1131 btrfs_close_one_device(device);
1133 WARN_ON(fs_devices->open_devices);
1134 WARN_ON(fs_devices->rw_devices);
1135 fs_devices->opened = 0;
1136 fs_devices->seeding = false;
1137 fs_devices->fs_info = NULL;
1140 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1143 struct btrfs_fs_devices *tmp;
1145 mutex_lock(&uuid_mutex);
1146 close_fs_devices(fs_devices);
1147 if (!fs_devices->opened) {
1148 list_splice_init(&fs_devices->seed_list, &list);
1151 * If the struct btrfs_fs_devices is not assembled with any
1152 * other device, it can be re-initialized during the next mount
1153 * without the needing device-scan step. Therefore, it can be
1156 if (fs_devices->num_devices == 1) {
1157 list_del(&fs_devices->fs_list);
1158 free_fs_devices(fs_devices);
1163 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {
1164 close_fs_devices(fs_devices);
1165 list_del(&fs_devices->seed_list);
1166 free_fs_devices(fs_devices);
1168 mutex_unlock(&uuid_mutex);
1171 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1172 blk_mode_t flags, void *holder)
1174 struct btrfs_device *device;
1175 struct btrfs_device *latest_dev = NULL;
1176 struct btrfs_device *tmp_device;
1178 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,
1182 ret = btrfs_open_one_device(fs_devices, device, flags, holder);
1184 (!latest_dev || device->generation > latest_dev->generation)) {
1185 latest_dev = device;
1186 } else if (ret == -ENODATA) {
1187 fs_devices->num_devices--;
1188 list_del(&device->dev_list);
1189 btrfs_free_device(device);
1192 if (fs_devices->open_devices == 0)
1195 fs_devices->opened = 1;
1196 fs_devices->latest_dev = latest_dev;
1197 fs_devices->total_rw_bytes = 0;
1198 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1199 fs_devices->read_policy = BTRFS_READ_POLICY_PID;
1204 static int devid_cmp(void *priv, const struct list_head *a,
1205 const struct list_head *b)
1207 const struct btrfs_device *dev1, *dev2;
1209 dev1 = list_entry(a, struct btrfs_device, dev_list);
1210 dev2 = list_entry(b, struct btrfs_device, dev_list);
1212 if (dev1->devid < dev2->devid)
1214 else if (dev1->devid > dev2->devid)
1219 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1220 blk_mode_t flags, void *holder)
1224 lockdep_assert_held(&uuid_mutex);
1226 * The device_list_mutex cannot be taken here in case opening the
1227 * underlying device takes further locks like open_mutex.
1229 * We also don't need the lock here as this is called during mount and
1230 * exclusion is provided by uuid_mutex
1233 if (fs_devices->opened) {
1234 fs_devices->opened++;
1237 list_sort(NULL, &fs_devices->devices, devid_cmp);
1238 ret = open_fs_devices(fs_devices, flags, holder);
1244 void btrfs_release_disk_super(struct btrfs_super_block *super)
1246 struct page *page = virt_to_page(super);
1251 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1252 u64 bytenr, u64 bytenr_orig)
1254 struct btrfs_super_block *disk_super;
1259 /* make sure our super fits in the device */
1260 if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev))
1261 return ERR_PTR(-EINVAL);
1263 /* make sure our super fits in the page */
1264 if (sizeof(*disk_super) > PAGE_SIZE)
1265 return ERR_PTR(-EINVAL);
1267 /* make sure our super doesn't straddle pages on disk */
1268 index = bytenr >> PAGE_SHIFT;
1269 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1270 return ERR_PTR(-EINVAL);
1272 /* pull in the page with our super */
1273 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1276 return ERR_CAST(page);
1278 p = page_address(page);
1280 /* align our pointer to the offset of the super block */
1281 disk_super = p + offset_in_page(bytenr);
1283 if (btrfs_super_bytenr(disk_super) != bytenr_orig ||
1284 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1285 btrfs_release_disk_super(p);
1286 return ERR_PTR(-EINVAL);
1289 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1290 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1295 int btrfs_forget_devices(dev_t devt)
1299 mutex_lock(&uuid_mutex);
1300 ret = btrfs_free_stale_devices(devt, NULL);
1301 mutex_unlock(&uuid_mutex);
1306 static bool btrfs_skip_registration(struct btrfs_super_block *disk_super,
1307 const char *path, dev_t devt,
1310 struct btrfs_fs_devices *fs_devices;
1313 * Do not skip device registration for mounted devices with matching
1314 * maj:min but different paths. Booting without initrd relies on
1315 * /dev/root initially, later replaced with the actual root device.
1316 * A successful scan ensures grub2-probe selects the correct device.
1318 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
1319 struct btrfs_device *device;
1321 mutex_lock(&fs_devices->device_list_mutex);
1323 if (!fs_devices->opened) {
1324 mutex_unlock(&fs_devices->device_list_mutex);
1328 list_for_each_entry(device, &fs_devices->devices, dev_list) {
1329 if (device->bdev && (device->bdev->bd_dev == devt) &&
1330 strcmp(device->name->str, path) != 0) {
1331 mutex_unlock(&fs_devices->device_list_mutex);
1333 /* Do not skip registration. */
1337 mutex_unlock(&fs_devices->device_list_mutex);
1340 if (!mount_arg_dev && btrfs_super_num_devices(disk_super) == 1 &&
1341 !(btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING))
1348 * Look for a btrfs signature on a device. This may be called out of the mount path
1349 * and we are not allowed to call set_blocksize during the scan. The superblock
1350 * is read via pagecache.
1352 * With @mount_arg_dev it's a scan during mount time that will always register
1353 * the device or return an error. Multi-device and seeding devices are registered
1356 struct btrfs_device *btrfs_scan_one_device(const char *path, blk_mode_t flags,
1359 struct btrfs_super_block *disk_super;
1360 bool new_device_added = false;
1361 struct btrfs_device *device = NULL;
1362 struct file *bdev_file;
1363 u64 bytenr, bytenr_orig;
1367 lockdep_assert_held(&uuid_mutex);
1370 * we would like to check all the supers, but that would make
1371 * a btrfs mount succeed after a mkfs from a different FS.
1372 * So, we need to add a special mount option to scan for
1373 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1377 * Avoid an exclusive open here, as the systemd-udev may initiate the
1378 * device scan which may race with the user's mount or mkfs command,
1379 * resulting in failure.
1380 * Since the device scan is solely for reading purposes, there is no
1381 * need for an exclusive open. Additionally, the devices are read again
1382 * during the mount process. It is ok to get some inconsistent
1383 * values temporarily, as the device paths of the fsid are the only
1384 * required information for assembling the volume.
1386 bdev_file = bdev_file_open_by_path(path, flags, NULL, NULL);
1387 if (IS_ERR(bdev_file))
1388 return ERR_CAST(bdev_file);
1390 bytenr_orig = btrfs_sb_offset(0);
1391 ret = btrfs_sb_log_location_bdev(file_bdev(bdev_file), 0, READ, &bytenr);
1393 device = ERR_PTR(ret);
1394 goto error_bdev_put;
1397 disk_super = btrfs_read_disk_super(file_bdev(bdev_file), bytenr,
1399 if (IS_ERR(disk_super)) {
1400 device = ERR_CAST(disk_super);
1401 goto error_bdev_put;
1404 devt = file_bdev(bdev_file)->bd_dev;
1405 if (btrfs_skip_registration(disk_super, path, devt, mount_arg_dev)) {
1406 pr_debug("BTRFS: skip registering single non-seed device %s (%d:%d)\n",
1407 path, MAJOR(devt), MINOR(devt));
1409 btrfs_free_stale_devices(devt, NULL);
1412 goto free_disk_super;
1415 device = device_list_add(path, disk_super, &new_device_added);
1416 if (!IS_ERR(device) && new_device_added)
1417 btrfs_free_stale_devices(device->devt, device);
1420 btrfs_release_disk_super(disk_super);
1429 * Try to find a chunk that intersects [start, start + len] range and when one
1430 * such is found, record the end of it in *start
1432 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1435 u64 physical_start, physical_end;
1437 lockdep_assert_held(&device->fs_info->chunk_mutex);
1439 if (find_first_extent_bit(&device->alloc_state, *start,
1440 &physical_start, &physical_end,
1441 CHUNK_ALLOCATED, NULL)) {
1443 if (in_range(physical_start, *start, len) ||
1444 in_range(*start, physical_start,
1445 physical_end + 1 - physical_start)) {
1446 *start = physical_end + 1;
1453 static u64 dev_extent_search_start(struct btrfs_device *device)
1455 switch (device->fs_devices->chunk_alloc_policy) {
1456 case BTRFS_CHUNK_ALLOC_REGULAR:
1457 return BTRFS_DEVICE_RANGE_RESERVED;
1458 case BTRFS_CHUNK_ALLOC_ZONED:
1460 * We don't care about the starting region like regular
1461 * allocator, because we anyway use/reserve the first two zones
1462 * for superblock logging.
1470 static bool dev_extent_hole_check_zoned(struct btrfs_device *device,
1471 u64 *hole_start, u64 *hole_size,
1474 u64 zone_size = device->zone_info->zone_size;
1477 bool changed = false;
1479 ASSERT(IS_ALIGNED(*hole_start, zone_size));
1481 while (*hole_size > 0) {
1482 pos = btrfs_find_allocatable_zones(device, *hole_start,
1483 *hole_start + *hole_size,
1485 if (pos != *hole_start) {
1486 *hole_size = *hole_start + *hole_size - pos;
1489 if (*hole_size < num_bytes)
1493 ret = btrfs_ensure_empty_zones(device, pos, num_bytes);
1495 /* Range is ensured to be empty */
1499 /* Given hole range was invalid (outside of device) */
1500 if (ret == -ERANGE) {
1501 *hole_start += *hole_size;
1506 *hole_start += zone_size;
1507 *hole_size -= zone_size;
1515 * Check if specified hole is suitable for allocation.
1517 * @device: the device which we have the hole
1518 * @hole_start: starting position of the hole
1519 * @hole_size: the size of the hole
1520 * @num_bytes: the size of the free space that we need
1522 * This function may modify @hole_start and @hole_size to reflect the suitable
1523 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1525 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1526 u64 *hole_size, u64 num_bytes)
1528 bool changed = false;
1529 u64 hole_end = *hole_start + *hole_size;
1533 * Check before we set max_hole_start, otherwise we could end up
1534 * sending back this offset anyway.
1536 if (contains_pending_extent(device, hole_start, *hole_size)) {
1537 if (hole_end >= *hole_start)
1538 *hole_size = hole_end - *hole_start;
1544 switch (device->fs_devices->chunk_alloc_policy) {
1545 case BTRFS_CHUNK_ALLOC_REGULAR:
1546 /* No extra check */
1548 case BTRFS_CHUNK_ALLOC_ZONED:
1549 if (dev_extent_hole_check_zoned(device, hole_start,
1550 hole_size, num_bytes)) {
1553 * The changed hole can contain pending extent.
1554 * Loop again to check that.
1570 * Find free space in the specified device.
1572 * @device: the device which we search the free space in
1573 * @num_bytes: the size of the free space that we need
1574 * @search_start: the position from which to begin the search
1575 * @start: store the start of the free space.
1576 * @len: the size of the free space. that we find, or the size
1577 * of the max free space if we don't find suitable free space
1579 * This does a pretty simple search, the expectation is that it is called very
1580 * infrequently and that a given device has a small number of extents.
1582 * @start is used to store the start of the free space if we find. But if we
1583 * don't find suitable free space, it will be used to store the start position
1584 * of the max free space.
1586 * @len is used to store the size of the free space that we find.
1587 * But if we don't find suitable free space, it is used to store the size of
1588 * the max free space.
1590 * NOTE: This function will search *commit* root of device tree, and does extra
1591 * check to ensure dev extents are not double allocated.
1592 * This makes the function safe to allocate dev extents but may not report
1593 * correct usable device space, as device extent freed in current transaction
1594 * is not reported as available.
1596 static int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1597 u64 *start, u64 *len)
1599 struct btrfs_fs_info *fs_info = device->fs_info;
1600 struct btrfs_root *root = fs_info->dev_root;
1601 struct btrfs_key key;
1602 struct btrfs_dev_extent *dev_extent;
1603 struct btrfs_path *path;
1607 u64 max_hole_size = 0;
1609 u64 search_end = device->total_bytes;
1612 struct extent_buffer *l;
1614 search_start = dev_extent_search_start(device);
1615 max_hole_start = search_start;
1617 WARN_ON(device->zone_info &&
1618 !IS_ALIGNED(num_bytes, device->zone_info->zone_size));
1620 path = btrfs_alloc_path();
1626 if (search_start >= search_end ||
1627 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1632 path->reada = READA_FORWARD;
1633 path->search_commit_root = 1;
1634 path->skip_locking = 1;
1636 key.objectid = device->devid;
1637 key.offset = search_start;
1638 key.type = BTRFS_DEV_EXTENT_KEY;
1640 ret = btrfs_search_backwards(root, &key, path);
1644 while (search_start < search_end) {
1646 slot = path->slots[0];
1647 if (slot >= btrfs_header_nritems(l)) {
1648 ret = btrfs_next_leaf(root, path);
1656 btrfs_item_key_to_cpu(l, &key, slot);
1658 if (key.objectid < device->devid)
1661 if (key.objectid > device->devid)
1664 if (key.type != BTRFS_DEV_EXTENT_KEY)
1667 if (key.offset > search_end)
1670 if (key.offset > search_start) {
1671 hole_size = key.offset - search_start;
1672 dev_extent_hole_check(device, &search_start, &hole_size,
1675 if (hole_size > max_hole_size) {
1676 max_hole_start = search_start;
1677 max_hole_size = hole_size;
1681 * If this free space is greater than which we need,
1682 * it must be the max free space that we have found
1683 * until now, so max_hole_start must point to the start
1684 * of this free space and the length of this free space
1685 * is stored in max_hole_size. Thus, we return
1686 * max_hole_start and max_hole_size and go back to the
1689 if (hole_size >= num_bytes) {
1695 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1696 extent_end = key.offset + btrfs_dev_extent_length(l,
1698 if (extent_end > search_start)
1699 search_start = extent_end;
1706 * At this point, search_start should be the end of
1707 * allocated dev extents, and when shrinking the device,
1708 * search_end may be smaller than search_start.
1710 if (search_end > search_start) {
1711 hole_size = search_end - search_start;
1712 if (dev_extent_hole_check(device, &search_start, &hole_size,
1714 btrfs_release_path(path);
1718 if (hole_size > max_hole_size) {
1719 max_hole_start = search_start;
1720 max_hole_size = hole_size;
1725 if (max_hole_size < num_bytes)
1730 ASSERT(max_hole_start + max_hole_size <= search_end);
1732 btrfs_free_path(path);
1733 *start = max_hole_start;
1735 *len = max_hole_size;
1739 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1740 struct btrfs_device *device,
1741 u64 start, u64 *dev_extent_len)
1743 struct btrfs_fs_info *fs_info = device->fs_info;
1744 struct btrfs_root *root = fs_info->dev_root;
1746 struct btrfs_path *path;
1747 struct btrfs_key key;
1748 struct btrfs_key found_key;
1749 struct extent_buffer *leaf = NULL;
1750 struct btrfs_dev_extent *extent = NULL;
1752 path = btrfs_alloc_path();
1756 key.objectid = device->devid;
1758 key.type = BTRFS_DEV_EXTENT_KEY;
1760 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1762 ret = btrfs_previous_item(root, path, key.objectid,
1763 BTRFS_DEV_EXTENT_KEY);
1766 leaf = path->nodes[0];
1767 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1768 extent = btrfs_item_ptr(leaf, path->slots[0],
1769 struct btrfs_dev_extent);
1770 BUG_ON(found_key.offset > start || found_key.offset +
1771 btrfs_dev_extent_length(leaf, extent) < start);
1773 btrfs_release_path(path);
1775 } else if (ret == 0) {
1776 leaf = path->nodes[0];
1777 extent = btrfs_item_ptr(leaf, path->slots[0],
1778 struct btrfs_dev_extent);
1783 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1785 ret = btrfs_del_item(trans, root, path);
1787 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1789 btrfs_free_path(path);
1793 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1798 read_lock(&fs_info->mapping_tree_lock);
1799 n = rb_last(&fs_info->mapping_tree.rb_root);
1801 struct btrfs_chunk_map *map;
1803 map = rb_entry(n, struct btrfs_chunk_map, rb_node);
1804 ret = map->start + map->chunk_len;
1806 read_unlock(&fs_info->mapping_tree_lock);
1811 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1815 struct btrfs_key key;
1816 struct btrfs_key found_key;
1817 struct btrfs_path *path;
1819 path = btrfs_alloc_path();
1823 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1824 key.type = BTRFS_DEV_ITEM_KEY;
1825 key.offset = (u64)-1;
1827 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1833 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1838 ret = btrfs_previous_item(fs_info->chunk_root, path,
1839 BTRFS_DEV_ITEMS_OBJECTID,
1840 BTRFS_DEV_ITEM_KEY);
1844 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1846 *devid_ret = found_key.offset + 1;
1850 btrfs_free_path(path);
1855 * the device information is stored in the chunk root
1856 * the btrfs_device struct should be fully filled in
1858 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1859 struct btrfs_device *device)
1862 struct btrfs_path *path;
1863 struct btrfs_dev_item *dev_item;
1864 struct extent_buffer *leaf;
1865 struct btrfs_key key;
1868 path = btrfs_alloc_path();
1872 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1873 key.type = BTRFS_DEV_ITEM_KEY;
1874 key.offset = device->devid;
1876 btrfs_reserve_chunk_metadata(trans, true);
1877 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1878 &key, sizeof(*dev_item));
1879 btrfs_trans_release_chunk_metadata(trans);
1883 leaf = path->nodes[0];
1884 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1886 btrfs_set_device_id(leaf, dev_item, device->devid);
1887 btrfs_set_device_generation(leaf, dev_item, 0);
1888 btrfs_set_device_type(leaf, dev_item, device->type);
1889 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1890 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1891 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1892 btrfs_set_device_total_bytes(leaf, dev_item,
1893 btrfs_device_get_disk_total_bytes(device));
1894 btrfs_set_device_bytes_used(leaf, dev_item,
1895 btrfs_device_get_bytes_used(device));
1896 btrfs_set_device_group(leaf, dev_item, 0);
1897 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1898 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1899 btrfs_set_device_start_offset(leaf, dev_item, 0);
1901 ptr = btrfs_device_uuid(dev_item);
1902 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1903 ptr = btrfs_device_fsid(dev_item);
1904 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1905 ptr, BTRFS_FSID_SIZE);
1906 btrfs_mark_buffer_dirty(trans, leaf);
1910 btrfs_free_path(path);
1915 * Function to update ctime/mtime for a given device path.
1916 * Mainly used for ctime/mtime based probe like libblkid.
1918 * We don't care about errors here, this is just to be kind to userspace.
1920 static void update_dev_time(const char *device_path)
1925 ret = kern_path(device_path, LOOKUP_FOLLOW, &path);
1929 inode_update_time(d_inode(path.dentry), S_MTIME | S_CTIME | S_VERSION);
1933 static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans,
1934 struct btrfs_device *device)
1936 struct btrfs_root *root = device->fs_info->chunk_root;
1938 struct btrfs_path *path;
1939 struct btrfs_key key;
1941 path = btrfs_alloc_path();
1945 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1946 key.type = BTRFS_DEV_ITEM_KEY;
1947 key.offset = device->devid;
1949 btrfs_reserve_chunk_metadata(trans, false);
1950 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1951 btrfs_trans_release_chunk_metadata(trans);
1958 ret = btrfs_del_item(trans, root, path);
1960 btrfs_free_path(path);
1965 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1966 * filesystem. It's up to the caller to adjust that number regarding eg. device
1969 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1977 seq = read_seqbegin(&fs_info->profiles_lock);
1979 all_avail = fs_info->avail_data_alloc_bits |
1980 fs_info->avail_system_alloc_bits |
1981 fs_info->avail_metadata_alloc_bits;
1982 } while (read_seqretry(&fs_info->profiles_lock, seq));
1984 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1985 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1988 if (num_devices < btrfs_raid_array[i].devs_min)
1989 return btrfs_raid_array[i].mindev_error;
1995 static struct btrfs_device * btrfs_find_next_active_device(
1996 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1998 struct btrfs_device *next_device;
2000 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
2001 if (next_device != device &&
2002 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
2003 && next_device->bdev)
2011 * Helper function to check if the given device is part of s_bdev / latest_dev
2012 * and replace it with the provided or the next active device, in the context
2013 * where this function called, there should be always be another device (or
2014 * this_dev) which is active.
2016 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
2017 struct btrfs_device *next_device)
2019 struct btrfs_fs_info *fs_info = device->fs_info;
2022 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
2024 ASSERT(next_device);
2026 if (fs_info->sb->s_bdev &&
2027 (fs_info->sb->s_bdev == device->bdev))
2028 fs_info->sb->s_bdev = next_device->bdev;
2030 if (fs_info->fs_devices->latest_dev->bdev == device->bdev)
2031 fs_info->fs_devices->latest_dev = next_device;
2035 * Return btrfs_fs_devices::num_devices excluding the device that's being
2036 * currently replaced.
2038 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2040 u64 num_devices = fs_info->fs_devices->num_devices;
2042 down_read(&fs_info->dev_replace.rwsem);
2043 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2044 ASSERT(num_devices > 1);
2047 up_read(&fs_info->dev_replace.rwsem);
2052 static void btrfs_scratch_superblock(struct btrfs_fs_info *fs_info,
2053 struct block_device *bdev, int copy_num)
2055 struct btrfs_super_block *disk_super;
2056 const size_t len = sizeof(disk_super->magic);
2057 const u64 bytenr = btrfs_sb_offset(copy_num);
2060 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr);
2061 if (IS_ERR(disk_super))
2064 memset(&disk_super->magic, 0, len);
2065 folio_mark_dirty(virt_to_folio(disk_super));
2066 btrfs_release_disk_super(disk_super);
2068 ret = sync_blockdev_range(bdev, bytenr, bytenr + len - 1);
2070 btrfs_warn(fs_info, "error clearing superblock number %d (%d)",
2074 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, struct btrfs_device *device)
2077 struct block_device *bdev = device->bdev;
2082 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2083 if (bdev_is_zoned(bdev))
2084 btrfs_reset_sb_log_zones(bdev, copy_num);
2086 btrfs_scratch_superblock(fs_info, bdev, copy_num);
2089 /* Notify udev that device has changed */
2090 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2092 /* Update ctime/mtime for device path for libblkid */
2093 update_dev_time(device->name->str);
2096 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
2097 struct btrfs_dev_lookup_args *args,
2098 struct file **bdev_file)
2100 struct btrfs_trans_handle *trans;
2101 struct btrfs_device *device;
2102 struct btrfs_fs_devices *cur_devices;
2103 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2107 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2108 btrfs_err(fs_info, "device remove not supported on extent tree v2 yet");
2113 * The device list in fs_devices is accessed without locks (neither
2114 * uuid_mutex nor device_list_mutex) as it won't change on a mounted
2115 * filesystem and another device rm cannot run.
2117 num_devices = btrfs_num_devices(fs_info);
2119 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2123 device = btrfs_find_device(fs_info->fs_devices, args);
2126 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2132 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2133 btrfs_warn_in_rcu(fs_info,
2134 "cannot remove device %s (devid %llu) due to active swapfile",
2135 btrfs_dev_name(device), device->devid);
2139 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
2140 return BTRFS_ERROR_DEV_TGT_REPLACE;
2142 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2143 fs_info->fs_devices->rw_devices == 1)
2144 return BTRFS_ERROR_DEV_ONLY_WRITABLE;
2146 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2147 mutex_lock(&fs_info->chunk_mutex);
2148 list_del_init(&device->dev_alloc_list);
2149 device->fs_devices->rw_devices--;
2150 mutex_unlock(&fs_info->chunk_mutex);
2153 ret = btrfs_shrink_device(device, 0);
2157 trans = btrfs_start_transaction(fs_info->chunk_root, 0);
2158 if (IS_ERR(trans)) {
2159 ret = PTR_ERR(trans);
2163 ret = btrfs_rm_dev_item(trans, device);
2165 /* Any error in dev item removal is critical */
2167 "failed to remove device item for devid %llu: %d",
2168 device->devid, ret);
2169 btrfs_abort_transaction(trans, ret);
2170 btrfs_end_transaction(trans);
2174 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2175 btrfs_scrub_cancel_dev(device);
2178 * the device list mutex makes sure that we don't change
2179 * the device list while someone else is writing out all
2180 * the device supers. Whoever is writing all supers, should
2181 * lock the device list mutex before getting the number of
2182 * devices in the super block (super_copy). Conversely,
2183 * whoever updates the number of devices in the super block
2184 * (super_copy) should hold the device list mutex.
2188 * In normal cases the cur_devices == fs_devices. But in case
2189 * of deleting a seed device, the cur_devices should point to
2190 * its own fs_devices listed under the fs_devices->seed_list.
2192 cur_devices = device->fs_devices;
2193 mutex_lock(&fs_devices->device_list_mutex);
2194 list_del_rcu(&device->dev_list);
2196 cur_devices->num_devices--;
2197 cur_devices->total_devices--;
2198 /* Update total_devices of the parent fs_devices if it's seed */
2199 if (cur_devices != fs_devices)
2200 fs_devices->total_devices--;
2202 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2203 cur_devices->missing_devices--;
2205 btrfs_assign_next_active_device(device, NULL);
2207 if (device->bdev_file) {
2208 cur_devices->open_devices--;
2209 /* remove sysfs entry */
2210 btrfs_sysfs_remove_device(device);
2213 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2214 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2215 mutex_unlock(&fs_devices->device_list_mutex);
2218 * At this point, the device is zero sized and detached from the
2219 * devices list. All that's left is to zero out the old supers and
2222 * We cannot call btrfs_close_bdev() here because we're holding the sb
2223 * write lock, and fput() on the block device will pull in the
2224 * ->open_mutex on the block device and it's dependencies. Instead
2225 * just flush the device and let the caller do the final bdev_release.
2227 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2228 btrfs_scratch_superblocks(fs_info, device);
2230 sync_blockdev(device->bdev);
2231 invalidate_bdev(device->bdev);
2235 *bdev_file = device->bdev_file;
2237 btrfs_free_device(device);
2240 * This can happen if cur_devices is the private seed devices list. We
2241 * cannot call close_fs_devices() here because it expects the uuid_mutex
2242 * to be held, but in fact we don't need that for the private
2243 * seed_devices, we can simply decrement cur_devices->opened and then
2244 * remove it from our list and free the fs_devices.
2246 if (cur_devices->num_devices == 0) {
2247 list_del_init(&cur_devices->seed_list);
2248 ASSERT(cur_devices->opened == 1);
2249 cur_devices->opened--;
2250 free_fs_devices(cur_devices);
2253 ret = btrfs_commit_transaction(trans);
2258 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2259 mutex_lock(&fs_info->chunk_mutex);
2260 list_add(&device->dev_alloc_list,
2261 &fs_devices->alloc_list);
2262 device->fs_devices->rw_devices++;
2263 mutex_unlock(&fs_info->chunk_mutex);
2268 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2270 struct btrfs_fs_devices *fs_devices;
2272 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2275 * in case of fs with no seed, srcdev->fs_devices will point
2276 * to fs_devices of fs_info. However when the dev being replaced is
2277 * a seed dev it will point to the seed's local fs_devices. In short
2278 * srcdev will have its correct fs_devices in both the cases.
2280 fs_devices = srcdev->fs_devices;
2282 list_del_rcu(&srcdev->dev_list);
2283 list_del(&srcdev->dev_alloc_list);
2284 fs_devices->num_devices--;
2285 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2286 fs_devices->missing_devices--;
2288 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2289 fs_devices->rw_devices--;
2292 fs_devices->open_devices--;
2295 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2297 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2299 mutex_lock(&uuid_mutex);
2301 btrfs_close_bdev(srcdev);
2303 btrfs_free_device(srcdev);
2305 /* if this is no devs we rather delete the fs_devices */
2306 if (!fs_devices->num_devices) {
2308 * On a mounted FS, num_devices can't be zero unless it's a
2309 * seed. In case of a seed device being replaced, the replace
2310 * target added to the sprout FS, so there will be no more
2311 * device left under the seed FS.
2313 ASSERT(fs_devices->seeding);
2315 list_del_init(&fs_devices->seed_list);
2316 close_fs_devices(fs_devices);
2317 free_fs_devices(fs_devices);
2319 mutex_unlock(&uuid_mutex);
2322 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2324 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2326 mutex_lock(&fs_devices->device_list_mutex);
2328 btrfs_sysfs_remove_device(tgtdev);
2331 fs_devices->open_devices--;
2333 fs_devices->num_devices--;
2335 btrfs_assign_next_active_device(tgtdev, NULL);
2337 list_del_rcu(&tgtdev->dev_list);
2339 mutex_unlock(&fs_devices->device_list_mutex);
2341 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev);
2343 btrfs_close_bdev(tgtdev);
2345 btrfs_free_device(tgtdev);
2349 * Populate args from device at path.
2351 * @fs_info: the filesystem
2352 * @args: the args to populate
2353 * @path: the path to the device
2355 * This will read the super block of the device at @path and populate @args with
2356 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to
2357 * lookup a device to operate on, but need to do it before we take any locks.
2358 * This properly handles the special case of "missing" that a user may pass in,
2359 * and does some basic sanity checks. The caller must make sure that @path is
2360 * properly NUL terminated before calling in, and must call
2361 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and
2364 * Return: 0 for success, -errno for failure
2366 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,
2367 struct btrfs_dev_lookup_args *args,
2370 struct btrfs_super_block *disk_super;
2371 struct file *bdev_file;
2374 if (!path || !path[0])
2376 if (!strcmp(path, "missing")) {
2377 args->missing = true;
2381 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL);
2382 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL);
2383 if (!args->uuid || !args->fsid) {
2384 btrfs_put_dev_args_from_path(args);
2388 ret = btrfs_get_bdev_and_sb(path, BLK_OPEN_READ, NULL, 0,
2389 &bdev_file, &disk_super);
2391 btrfs_put_dev_args_from_path(args);
2395 args->devid = btrfs_stack_device_id(&disk_super->dev_item);
2396 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE);
2397 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2398 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE);
2400 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
2401 btrfs_release_disk_super(disk_super);
2407 * Only use this jointly with btrfs_get_dev_args_from_path() because we will
2408 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables
2409 * that don't need to be freed.
2411 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args)
2419 struct btrfs_device *btrfs_find_device_by_devspec(
2420 struct btrfs_fs_info *fs_info, u64 devid,
2421 const char *device_path)
2423 BTRFS_DEV_LOOKUP_ARGS(args);
2424 struct btrfs_device *device;
2429 device = btrfs_find_device(fs_info->fs_devices, &args);
2431 return ERR_PTR(-ENOENT);
2435 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path);
2437 return ERR_PTR(ret);
2438 device = btrfs_find_device(fs_info->fs_devices, &args);
2439 btrfs_put_dev_args_from_path(&args);
2441 return ERR_PTR(-ENOENT);
2445 static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info)
2447 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2448 struct btrfs_fs_devices *old_devices;
2449 struct btrfs_fs_devices *seed_devices;
2451 lockdep_assert_held(&uuid_mutex);
2452 if (!fs_devices->seeding)
2453 return ERR_PTR(-EINVAL);
2456 * Private copy of the seed devices, anchored at
2457 * fs_info->fs_devices->seed_list
2459 seed_devices = alloc_fs_devices(NULL);
2460 if (IS_ERR(seed_devices))
2461 return seed_devices;
2464 * It's necessary to retain a copy of the original seed fs_devices in
2465 * fs_uuids so that filesystems which have been seeded can successfully
2466 * reference the seed device from open_seed_devices. This also supports
2469 old_devices = clone_fs_devices(fs_devices);
2470 if (IS_ERR(old_devices)) {
2471 kfree(seed_devices);
2475 list_add(&old_devices->fs_list, &fs_uuids);
2477 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2478 seed_devices->opened = 1;
2479 INIT_LIST_HEAD(&seed_devices->devices);
2480 INIT_LIST_HEAD(&seed_devices->alloc_list);
2481 mutex_init(&seed_devices->device_list_mutex);
2483 return seed_devices;
2487 * Splice seed devices into the sprout fs_devices.
2488 * Generate a new fsid for the sprouted read-write filesystem.
2490 static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info,
2491 struct btrfs_fs_devices *seed_devices)
2493 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2494 struct btrfs_super_block *disk_super = fs_info->super_copy;
2495 struct btrfs_device *device;
2499 * We are updating the fsid, the thread leading to device_list_add()
2500 * could race, so uuid_mutex is needed.
2502 lockdep_assert_held(&uuid_mutex);
2505 * The threads listed below may traverse dev_list but can do that without
2506 * device_list_mutex:
2507 * - All device ops and balance - as we are in btrfs_exclop_start.
2508 * - Various dev_list readers - are using RCU.
2509 * - btrfs_ioctl_fitrim() - is using RCU.
2511 * For-read threads as below are using device_list_mutex:
2512 * - Readonly scrub btrfs_scrub_dev()
2513 * - Readonly scrub btrfs_scrub_progress()
2514 * - btrfs_get_dev_stats()
2516 lockdep_assert_held(&fs_devices->device_list_mutex);
2518 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2520 list_for_each_entry(device, &seed_devices->devices, dev_list)
2521 device->fs_devices = seed_devices;
2523 fs_devices->seeding = false;
2524 fs_devices->num_devices = 0;
2525 fs_devices->open_devices = 0;
2526 fs_devices->missing_devices = 0;
2527 fs_devices->rotating = false;
2528 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2530 generate_random_uuid(fs_devices->fsid);
2531 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2532 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2534 super_flags = btrfs_super_flags(disk_super) &
2535 ~BTRFS_SUPER_FLAG_SEEDING;
2536 btrfs_set_super_flags(disk_super, super_flags);
2540 * Store the expected generation for seed devices in device items.
2542 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2544 BTRFS_DEV_LOOKUP_ARGS(args);
2545 struct btrfs_fs_info *fs_info = trans->fs_info;
2546 struct btrfs_root *root = fs_info->chunk_root;
2547 struct btrfs_path *path;
2548 struct extent_buffer *leaf;
2549 struct btrfs_dev_item *dev_item;
2550 struct btrfs_device *device;
2551 struct btrfs_key key;
2552 u8 fs_uuid[BTRFS_FSID_SIZE];
2553 u8 dev_uuid[BTRFS_UUID_SIZE];
2556 path = btrfs_alloc_path();
2560 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2562 key.type = BTRFS_DEV_ITEM_KEY;
2565 btrfs_reserve_chunk_metadata(trans, false);
2566 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2567 btrfs_trans_release_chunk_metadata(trans);
2571 leaf = path->nodes[0];
2573 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2574 ret = btrfs_next_leaf(root, path);
2579 leaf = path->nodes[0];
2580 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2581 btrfs_release_path(path);
2585 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2586 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2587 key.type != BTRFS_DEV_ITEM_KEY)
2590 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2591 struct btrfs_dev_item);
2592 args.devid = btrfs_device_id(leaf, dev_item);
2593 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2595 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2597 args.uuid = dev_uuid;
2598 args.fsid = fs_uuid;
2599 device = btrfs_find_device(fs_info->fs_devices, &args);
2600 BUG_ON(!device); /* Logic error */
2602 if (device->fs_devices->seeding) {
2603 btrfs_set_device_generation(leaf, dev_item,
2604 device->generation);
2605 btrfs_mark_buffer_dirty(trans, leaf);
2613 btrfs_free_path(path);
2617 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2619 struct btrfs_root *root = fs_info->dev_root;
2620 struct btrfs_trans_handle *trans;
2621 struct btrfs_device *device;
2622 struct file *bdev_file;
2623 struct super_block *sb = fs_info->sb;
2624 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2625 struct btrfs_fs_devices *seed_devices = NULL;
2626 u64 orig_super_total_bytes;
2627 u64 orig_super_num_devices;
2629 bool seeding_dev = false;
2630 bool locked = false;
2632 if (sb_rdonly(sb) && !fs_devices->seeding)
2635 bdev_file = bdev_file_open_by_path(device_path, BLK_OPEN_WRITE,
2636 fs_info->bdev_holder, NULL);
2637 if (IS_ERR(bdev_file))
2638 return PTR_ERR(bdev_file);
2640 if (!btrfs_check_device_zone_type(fs_info, file_bdev(bdev_file))) {
2645 if (fs_devices->seeding) {
2647 down_write(&sb->s_umount);
2648 mutex_lock(&uuid_mutex);
2652 sync_blockdev(file_bdev(bdev_file));
2655 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2656 if (device->bdev == file_bdev(bdev_file)) {
2664 device = btrfs_alloc_device(fs_info, NULL, NULL, device_path);
2665 if (IS_ERR(device)) {
2666 /* we can safely leave the fs_devices entry around */
2667 ret = PTR_ERR(device);
2671 device->fs_info = fs_info;
2672 device->bdev_file = bdev_file;
2673 device->bdev = file_bdev(bdev_file);
2674 ret = lookup_bdev(device_path, &device->devt);
2676 goto error_free_device;
2678 ret = btrfs_get_dev_zone_info(device, false);
2680 goto error_free_device;
2682 trans = btrfs_start_transaction(root, 0);
2683 if (IS_ERR(trans)) {
2684 ret = PTR_ERR(trans);
2685 goto error_free_zone;
2688 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2689 device->generation = trans->transid;
2690 device->io_width = fs_info->sectorsize;
2691 device->io_align = fs_info->sectorsize;
2692 device->sector_size = fs_info->sectorsize;
2693 device->total_bytes =
2694 round_down(bdev_nr_bytes(device->bdev), fs_info->sectorsize);
2695 device->disk_total_bytes = device->total_bytes;
2696 device->commit_total_bytes = device->total_bytes;
2697 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2698 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2699 device->dev_stats_valid = 1;
2700 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2703 btrfs_clear_sb_rdonly(sb);
2705 /* GFP_KERNEL allocation must not be under device_list_mutex */
2706 seed_devices = btrfs_init_sprout(fs_info);
2707 if (IS_ERR(seed_devices)) {
2708 ret = PTR_ERR(seed_devices);
2709 btrfs_abort_transaction(trans, ret);
2714 mutex_lock(&fs_devices->device_list_mutex);
2716 btrfs_setup_sprout(fs_info, seed_devices);
2717 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev,
2721 device->fs_devices = fs_devices;
2723 mutex_lock(&fs_info->chunk_mutex);
2724 list_add_rcu(&device->dev_list, &fs_devices->devices);
2725 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2726 fs_devices->num_devices++;
2727 fs_devices->open_devices++;
2728 fs_devices->rw_devices++;
2729 fs_devices->total_devices++;
2730 fs_devices->total_rw_bytes += device->total_bytes;
2732 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2734 if (!bdev_nonrot(device->bdev))
2735 fs_devices->rotating = true;
2737 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2738 btrfs_set_super_total_bytes(fs_info->super_copy,
2739 round_down(orig_super_total_bytes + device->total_bytes,
2740 fs_info->sectorsize));
2742 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2743 btrfs_set_super_num_devices(fs_info->super_copy,
2744 orig_super_num_devices + 1);
2747 * we've got more storage, clear any full flags on the space
2750 btrfs_clear_space_info_full(fs_info);
2752 mutex_unlock(&fs_info->chunk_mutex);
2754 /* Add sysfs device entry */
2755 btrfs_sysfs_add_device(device);
2757 mutex_unlock(&fs_devices->device_list_mutex);
2760 mutex_lock(&fs_info->chunk_mutex);
2761 ret = init_first_rw_device(trans);
2762 mutex_unlock(&fs_info->chunk_mutex);
2764 btrfs_abort_transaction(trans, ret);
2769 ret = btrfs_add_dev_item(trans, device);
2771 btrfs_abort_transaction(trans, ret);
2776 ret = btrfs_finish_sprout(trans);
2778 btrfs_abort_transaction(trans, ret);
2783 * fs_devices now represents the newly sprouted filesystem and
2784 * its fsid has been changed by btrfs_sprout_splice().
2786 btrfs_sysfs_update_sprout_fsid(fs_devices);
2789 ret = btrfs_commit_transaction(trans);
2792 mutex_unlock(&uuid_mutex);
2793 up_write(&sb->s_umount);
2796 if (ret) /* transaction commit */
2799 ret = btrfs_relocate_sys_chunks(fs_info);
2801 btrfs_handle_fs_error(fs_info, ret,
2802 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2803 trans = btrfs_attach_transaction(root);
2804 if (IS_ERR(trans)) {
2805 if (PTR_ERR(trans) == -ENOENT)
2807 ret = PTR_ERR(trans);
2811 ret = btrfs_commit_transaction(trans);
2815 * Now that we have written a new super block to this device, check all
2816 * other fs_devices list if device_path alienates any other scanned
2818 * We can ignore the return value as it typically returns -EINVAL and
2819 * only succeeds if the device was an alien.
2821 btrfs_forget_devices(device->devt);
2823 /* Update ctime/mtime for blkid or udev */
2824 update_dev_time(device_path);
2829 btrfs_sysfs_remove_device(device);
2830 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2831 mutex_lock(&fs_info->chunk_mutex);
2832 list_del_rcu(&device->dev_list);
2833 list_del(&device->dev_alloc_list);
2834 fs_info->fs_devices->num_devices--;
2835 fs_info->fs_devices->open_devices--;
2836 fs_info->fs_devices->rw_devices--;
2837 fs_info->fs_devices->total_devices--;
2838 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2839 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2840 btrfs_set_super_total_bytes(fs_info->super_copy,
2841 orig_super_total_bytes);
2842 btrfs_set_super_num_devices(fs_info->super_copy,
2843 orig_super_num_devices);
2844 mutex_unlock(&fs_info->chunk_mutex);
2845 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2848 btrfs_set_sb_rdonly(sb);
2850 btrfs_end_transaction(trans);
2852 btrfs_destroy_dev_zone_info(device);
2854 btrfs_free_device(device);
2858 mutex_unlock(&uuid_mutex);
2859 up_write(&sb->s_umount);
2864 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2865 struct btrfs_device *device)
2868 struct btrfs_path *path;
2869 struct btrfs_root *root = device->fs_info->chunk_root;
2870 struct btrfs_dev_item *dev_item;
2871 struct extent_buffer *leaf;
2872 struct btrfs_key key;
2874 path = btrfs_alloc_path();
2878 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2879 key.type = BTRFS_DEV_ITEM_KEY;
2880 key.offset = device->devid;
2882 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2891 leaf = path->nodes[0];
2892 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2894 btrfs_set_device_id(leaf, dev_item, device->devid);
2895 btrfs_set_device_type(leaf, dev_item, device->type);
2896 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2897 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2898 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2899 btrfs_set_device_total_bytes(leaf, dev_item,
2900 btrfs_device_get_disk_total_bytes(device));
2901 btrfs_set_device_bytes_used(leaf, dev_item,
2902 btrfs_device_get_bytes_used(device));
2903 btrfs_mark_buffer_dirty(trans, leaf);
2906 btrfs_free_path(path);
2910 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2911 struct btrfs_device *device, u64 new_size)
2913 struct btrfs_fs_info *fs_info = device->fs_info;
2914 struct btrfs_super_block *super_copy = fs_info->super_copy;
2919 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2922 new_size = round_down(new_size, fs_info->sectorsize);
2924 mutex_lock(&fs_info->chunk_mutex);
2925 old_total = btrfs_super_total_bytes(super_copy);
2926 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2928 if (new_size <= device->total_bytes ||
2929 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2930 mutex_unlock(&fs_info->chunk_mutex);
2934 btrfs_set_super_total_bytes(super_copy,
2935 round_down(old_total + diff, fs_info->sectorsize));
2936 device->fs_devices->total_rw_bytes += diff;
2937 atomic64_add(diff, &fs_info->free_chunk_space);
2939 btrfs_device_set_total_bytes(device, new_size);
2940 btrfs_device_set_disk_total_bytes(device, new_size);
2941 btrfs_clear_space_info_full(device->fs_info);
2942 if (list_empty(&device->post_commit_list))
2943 list_add_tail(&device->post_commit_list,
2944 &trans->transaction->dev_update_list);
2945 mutex_unlock(&fs_info->chunk_mutex);
2947 btrfs_reserve_chunk_metadata(trans, false);
2948 ret = btrfs_update_device(trans, device);
2949 btrfs_trans_release_chunk_metadata(trans);
2954 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2956 struct btrfs_fs_info *fs_info = trans->fs_info;
2957 struct btrfs_root *root = fs_info->chunk_root;
2959 struct btrfs_path *path;
2960 struct btrfs_key key;
2962 path = btrfs_alloc_path();
2966 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2967 key.offset = chunk_offset;
2968 key.type = BTRFS_CHUNK_ITEM_KEY;
2970 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2973 else if (ret > 0) { /* Logic error or corruption */
2974 btrfs_handle_fs_error(fs_info, -ENOENT,
2975 "Failed lookup while freeing chunk.");
2980 ret = btrfs_del_item(trans, root, path);
2982 btrfs_handle_fs_error(fs_info, ret,
2983 "Failed to delete chunk item.");
2985 btrfs_free_path(path);
2989 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2991 struct btrfs_super_block *super_copy = fs_info->super_copy;
2992 struct btrfs_disk_key *disk_key;
2993 struct btrfs_chunk *chunk;
3000 struct btrfs_key key;
3002 lockdep_assert_held(&fs_info->chunk_mutex);
3003 array_size = btrfs_super_sys_array_size(super_copy);
3005 ptr = super_copy->sys_chunk_array;
3008 while (cur < array_size) {
3009 disk_key = (struct btrfs_disk_key *)ptr;
3010 btrfs_disk_key_to_cpu(&key, disk_key);
3012 len = sizeof(*disk_key);
3014 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3015 chunk = (struct btrfs_chunk *)(ptr + len);
3016 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
3017 len += btrfs_chunk_item_size(num_stripes);
3022 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
3023 key.offset == chunk_offset) {
3024 memmove(ptr, ptr + len, array_size - (cur + len));
3026 btrfs_set_super_sys_array_size(super_copy, array_size);
3035 struct btrfs_chunk_map *btrfs_find_chunk_map_nolock(struct btrfs_fs_info *fs_info,
3036 u64 logical, u64 length)
3038 struct rb_node *node = fs_info->mapping_tree.rb_root.rb_node;
3039 struct rb_node *prev = NULL;
3040 struct rb_node *orig_prev;
3041 struct btrfs_chunk_map *map;
3042 struct btrfs_chunk_map *prev_map = NULL;
3045 map = rb_entry(node, struct btrfs_chunk_map, rb_node);
3049 if (logical < map->start) {
3050 node = node->rb_left;
3051 } else if (logical >= map->start + map->chunk_len) {
3052 node = node->rb_right;
3054 refcount_inc(&map->refs);
3063 while (prev && logical >= prev_map->start + prev_map->chunk_len) {
3064 prev = rb_next(prev);
3065 prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node);
3070 prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node);
3071 while (prev && logical < prev_map->start) {
3072 prev = rb_prev(prev);
3073 prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node);
3078 u64 end = logical + length;
3081 * Caller can pass a U64_MAX length when it wants to get any
3082 * chunk starting at an offset of 'logical' or higher, so deal
3083 * with underflow by resetting the end offset to U64_MAX.
3088 if (end > prev_map->start &&
3089 logical < prev_map->start + prev_map->chunk_len) {
3090 refcount_inc(&prev_map->refs);
3098 struct btrfs_chunk_map *btrfs_find_chunk_map(struct btrfs_fs_info *fs_info,
3099 u64 logical, u64 length)
3101 struct btrfs_chunk_map *map;
3103 read_lock(&fs_info->mapping_tree_lock);
3104 map = btrfs_find_chunk_map_nolock(fs_info, logical, length);
3105 read_unlock(&fs_info->mapping_tree_lock);
3111 * Find the mapping containing the given logical extent.
3113 * @logical: Logical block offset in bytes.
3114 * @length: Length of extent in bytes.
3116 * Return: Chunk mapping or ERR_PTR.
3118 struct btrfs_chunk_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
3119 u64 logical, u64 length)
3121 struct btrfs_chunk_map *map;
3123 map = btrfs_find_chunk_map(fs_info, logical, length);
3125 if (unlikely(!map)) {
3127 "unable to find chunk map for logical %llu length %llu",
3129 return ERR_PTR(-EINVAL);
3132 if (unlikely(map->start > logical || map->start + map->chunk_len <= logical)) {
3134 "found a bad chunk map, wanted %llu-%llu, found %llu-%llu",
3135 logical, logical + length, map->start,
3136 map->start + map->chunk_len);
3137 btrfs_free_chunk_map(map);
3138 return ERR_PTR(-EINVAL);
3141 /* Callers are responsible for dropping the reference. */
3145 static int remove_chunk_item(struct btrfs_trans_handle *trans,
3146 struct btrfs_chunk_map *map, u64 chunk_offset)
3151 * Removing chunk items and updating the device items in the chunks btree
3152 * requires holding the chunk_mutex.
3153 * See the comment at btrfs_chunk_alloc() for the details.
3155 lockdep_assert_held(&trans->fs_info->chunk_mutex);
3157 for (i = 0; i < map->num_stripes; i++) {
3160 ret = btrfs_update_device(trans, map->stripes[i].dev);
3165 return btrfs_free_chunk(trans, chunk_offset);
3168 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
3170 struct btrfs_fs_info *fs_info = trans->fs_info;
3171 struct btrfs_chunk_map *map;
3172 u64 dev_extent_len = 0;
3174 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3176 map = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
3179 * This is a logic error, but we don't want to just rely on the
3180 * user having built with ASSERT enabled, so if ASSERT doesn't
3181 * do anything we still error out.
3184 return PTR_ERR(map);
3188 * First delete the device extent items from the devices btree.
3189 * We take the device_list_mutex to avoid racing with the finishing phase
3190 * of a device replace operation. See the comment below before acquiring
3191 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
3192 * because that can result in a deadlock when deleting the device extent
3193 * items from the devices btree - COWing an extent buffer from the btree
3194 * may result in allocating a new metadata chunk, which would attempt to
3195 * lock again fs_info->chunk_mutex.
3197 mutex_lock(&fs_devices->device_list_mutex);
3198 for (i = 0; i < map->num_stripes; i++) {
3199 struct btrfs_device *device = map->stripes[i].dev;
3200 ret = btrfs_free_dev_extent(trans, device,
3201 map->stripes[i].physical,
3204 mutex_unlock(&fs_devices->device_list_mutex);
3205 btrfs_abort_transaction(trans, ret);
3209 if (device->bytes_used > 0) {
3210 mutex_lock(&fs_info->chunk_mutex);
3211 btrfs_device_set_bytes_used(device,
3212 device->bytes_used - dev_extent_len);
3213 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3214 btrfs_clear_space_info_full(fs_info);
3215 mutex_unlock(&fs_info->chunk_mutex);
3218 mutex_unlock(&fs_devices->device_list_mutex);
3221 * We acquire fs_info->chunk_mutex for 2 reasons:
3223 * 1) Just like with the first phase of the chunk allocation, we must
3224 * reserve system space, do all chunk btree updates and deletions, and
3225 * update the system chunk array in the superblock while holding this
3226 * mutex. This is for similar reasons as explained on the comment at
3227 * the top of btrfs_chunk_alloc();
3229 * 2) Prevent races with the final phase of a device replace operation
3230 * that replaces the device object associated with the map's stripes,
3231 * because the device object's id can change at any time during that
3232 * final phase of the device replace operation
3233 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3234 * replaced device and then see it with an ID of
3235 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3236 * the device item, which does not exists on the chunk btree.
3237 * The finishing phase of device replace acquires both the
3238 * device_list_mutex and the chunk_mutex, in that order, so we are
3239 * safe by just acquiring the chunk_mutex.
3241 trans->removing_chunk = true;
3242 mutex_lock(&fs_info->chunk_mutex);
3244 check_system_chunk(trans, map->type);
3246 ret = remove_chunk_item(trans, map, chunk_offset);
3248 * Normally we should not get -ENOSPC since we reserved space before
3249 * through the call to check_system_chunk().
3251 * Despite our system space_info having enough free space, we may not
3252 * be able to allocate extents from its block groups, because all have
3253 * an incompatible profile, which will force us to allocate a new system
3254 * block group with the right profile, or right after we called
3255 * check_system_space() above, a scrub turned the only system block group
3256 * with enough free space into RO mode.
3257 * This is explained with more detail at do_chunk_alloc().
3259 * So if we get -ENOSPC, allocate a new system chunk and retry once.
3261 if (ret == -ENOSPC) {
3262 const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
3263 struct btrfs_block_group *sys_bg;
3265 sys_bg = btrfs_create_chunk(trans, sys_flags);
3266 if (IS_ERR(sys_bg)) {
3267 ret = PTR_ERR(sys_bg);
3268 btrfs_abort_transaction(trans, ret);
3272 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3274 btrfs_abort_transaction(trans, ret);
3278 ret = remove_chunk_item(trans, map, chunk_offset);
3280 btrfs_abort_transaction(trans, ret);
3284 btrfs_abort_transaction(trans, ret);
3288 trace_btrfs_chunk_free(fs_info, map, chunk_offset, map->chunk_len);
3290 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3291 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3293 btrfs_abort_transaction(trans, ret);
3298 mutex_unlock(&fs_info->chunk_mutex);
3299 trans->removing_chunk = false;
3302 * We are done with chunk btree updates and deletions, so release the
3303 * system space we previously reserved (with check_system_chunk()).
3305 btrfs_trans_release_chunk_metadata(trans);
3307 ret = btrfs_remove_block_group(trans, map);
3309 btrfs_abort_transaction(trans, ret);
3314 if (trans->removing_chunk) {
3315 mutex_unlock(&fs_info->chunk_mutex);
3316 trans->removing_chunk = false;
3319 btrfs_free_chunk_map(map);
3323 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3325 struct btrfs_root *root = fs_info->chunk_root;
3326 struct btrfs_trans_handle *trans;
3327 struct btrfs_block_group *block_group;
3331 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3333 "relocate: not supported on extent tree v2 yet");
3338 * Prevent races with automatic removal of unused block groups.
3339 * After we relocate and before we remove the chunk with offset
3340 * chunk_offset, automatic removal of the block group can kick in,
3341 * resulting in a failure when calling btrfs_remove_chunk() below.
3343 * Make sure to acquire this mutex before doing a tree search (dev
3344 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3345 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3346 * we release the path used to search the chunk/dev tree and before
3347 * the current task acquires this mutex and calls us.
3349 lockdep_assert_held(&fs_info->reclaim_bgs_lock);
3351 /* step one, relocate all the extents inside this chunk */
3352 btrfs_scrub_pause(fs_info);
3353 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3354 btrfs_scrub_continue(fs_info);
3357 * If we had a transaction abort, stop all running scrubs.
3358 * See transaction.c:cleanup_transaction() why we do it here.
3360 if (BTRFS_FS_ERROR(fs_info))
3361 btrfs_scrub_cancel(fs_info);
3365 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3368 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3369 length = block_group->length;
3370 btrfs_put_block_group(block_group);
3373 * On a zoned file system, discard the whole block group, this will
3374 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3375 * resetting the zone fails, don't treat it as a fatal problem from the
3376 * filesystem's point of view.
3378 if (btrfs_is_zoned(fs_info)) {
3379 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
3382 "failed to reset zone %llu after relocation",
3386 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3388 if (IS_ERR(trans)) {
3389 ret = PTR_ERR(trans);
3390 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3395 * step two, delete the device extents and the
3396 * chunk tree entries
3398 ret = btrfs_remove_chunk(trans, chunk_offset);
3399 btrfs_end_transaction(trans);
3403 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3405 struct btrfs_root *chunk_root = fs_info->chunk_root;
3406 struct btrfs_path *path;
3407 struct extent_buffer *leaf;
3408 struct btrfs_chunk *chunk;
3409 struct btrfs_key key;
3410 struct btrfs_key found_key;
3412 bool retried = false;
3416 path = btrfs_alloc_path();
3421 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3422 key.offset = (u64)-1;
3423 key.type = BTRFS_CHUNK_ITEM_KEY;
3426 mutex_lock(&fs_info->reclaim_bgs_lock);
3427 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3429 mutex_unlock(&fs_info->reclaim_bgs_lock);
3434 * On the first search we would find chunk tree with
3435 * offset -1, which is not possible. On subsequent
3436 * loops this would find an existing item on an invalid
3437 * offset (one less than the previous one, wrong
3438 * alignment and size).
3444 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3447 mutex_unlock(&fs_info->reclaim_bgs_lock);
3453 leaf = path->nodes[0];
3454 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3456 chunk = btrfs_item_ptr(leaf, path->slots[0],
3457 struct btrfs_chunk);
3458 chunk_type = btrfs_chunk_type(leaf, chunk);
3459 btrfs_release_path(path);
3461 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3462 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3468 mutex_unlock(&fs_info->reclaim_bgs_lock);
3470 if (found_key.offset == 0)
3472 key.offset = found_key.offset - 1;
3475 if (failed && !retried) {
3479 } else if (WARN_ON(failed && retried)) {
3483 btrfs_free_path(path);
3488 * return 1 : allocate a data chunk successfully,
3489 * return <0: errors during allocating a data chunk,
3490 * return 0 : no need to allocate a data chunk.
3492 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3495 struct btrfs_block_group *cache;
3499 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3501 chunk_type = cache->flags;
3502 btrfs_put_block_group(cache);
3504 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3507 spin_lock(&fs_info->data_sinfo->lock);
3508 bytes_used = fs_info->data_sinfo->bytes_used;
3509 spin_unlock(&fs_info->data_sinfo->lock);
3512 struct btrfs_trans_handle *trans;
3515 trans = btrfs_join_transaction(fs_info->tree_root);
3517 return PTR_ERR(trans);
3519 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3520 btrfs_end_transaction(trans);
3529 static void btrfs_disk_balance_args_to_cpu(struct btrfs_balance_args *cpu,
3530 const struct btrfs_disk_balance_args *disk)
3532 memset(cpu, 0, sizeof(*cpu));
3534 cpu->profiles = le64_to_cpu(disk->profiles);
3535 cpu->usage = le64_to_cpu(disk->usage);
3536 cpu->devid = le64_to_cpu(disk->devid);
3537 cpu->pstart = le64_to_cpu(disk->pstart);
3538 cpu->pend = le64_to_cpu(disk->pend);
3539 cpu->vstart = le64_to_cpu(disk->vstart);
3540 cpu->vend = le64_to_cpu(disk->vend);
3541 cpu->target = le64_to_cpu(disk->target);
3542 cpu->flags = le64_to_cpu(disk->flags);
3543 cpu->limit = le64_to_cpu(disk->limit);
3544 cpu->stripes_min = le32_to_cpu(disk->stripes_min);
3545 cpu->stripes_max = le32_to_cpu(disk->stripes_max);
3548 static void btrfs_cpu_balance_args_to_disk(struct btrfs_disk_balance_args *disk,
3549 const struct btrfs_balance_args *cpu)
3551 memset(disk, 0, sizeof(*disk));
3553 disk->profiles = cpu_to_le64(cpu->profiles);
3554 disk->usage = cpu_to_le64(cpu->usage);
3555 disk->devid = cpu_to_le64(cpu->devid);
3556 disk->pstart = cpu_to_le64(cpu->pstart);
3557 disk->pend = cpu_to_le64(cpu->pend);
3558 disk->vstart = cpu_to_le64(cpu->vstart);
3559 disk->vend = cpu_to_le64(cpu->vend);
3560 disk->target = cpu_to_le64(cpu->target);
3561 disk->flags = cpu_to_le64(cpu->flags);
3562 disk->limit = cpu_to_le64(cpu->limit);
3563 disk->stripes_min = cpu_to_le32(cpu->stripes_min);
3564 disk->stripes_max = cpu_to_le32(cpu->stripes_max);
3567 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3568 struct btrfs_balance_control *bctl)
3570 struct btrfs_root *root = fs_info->tree_root;
3571 struct btrfs_trans_handle *trans;
3572 struct btrfs_balance_item *item;
3573 struct btrfs_disk_balance_args disk_bargs;
3574 struct btrfs_path *path;
3575 struct extent_buffer *leaf;
3576 struct btrfs_key key;
3579 path = btrfs_alloc_path();
3583 trans = btrfs_start_transaction(root, 0);
3584 if (IS_ERR(trans)) {
3585 btrfs_free_path(path);
3586 return PTR_ERR(trans);
3589 key.objectid = BTRFS_BALANCE_OBJECTID;
3590 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3593 ret = btrfs_insert_empty_item(trans, root, path, &key,
3598 leaf = path->nodes[0];
3599 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3601 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3603 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3604 btrfs_set_balance_data(leaf, item, &disk_bargs);
3605 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3606 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3607 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3608 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3610 btrfs_set_balance_flags(leaf, item, bctl->flags);
3612 btrfs_mark_buffer_dirty(trans, leaf);
3614 btrfs_free_path(path);
3615 err = btrfs_commit_transaction(trans);
3621 static int del_balance_item(struct btrfs_fs_info *fs_info)
3623 struct btrfs_root *root = fs_info->tree_root;
3624 struct btrfs_trans_handle *trans;
3625 struct btrfs_path *path;
3626 struct btrfs_key key;
3629 path = btrfs_alloc_path();
3633 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3634 if (IS_ERR(trans)) {
3635 btrfs_free_path(path);
3636 return PTR_ERR(trans);
3639 key.objectid = BTRFS_BALANCE_OBJECTID;
3640 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3643 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3651 ret = btrfs_del_item(trans, root, path);
3653 btrfs_free_path(path);
3654 err = btrfs_commit_transaction(trans);
3661 * This is a heuristic used to reduce the number of chunks balanced on
3662 * resume after balance was interrupted.
3664 static void update_balance_args(struct btrfs_balance_control *bctl)
3667 * Turn on soft mode for chunk types that were being converted.
3669 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3670 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3671 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3672 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3673 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3674 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3677 * Turn on usage filter if is not already used. The idea is
3678 * that chunks that we have already balanced should be
3679 * reasonably full. Don't do it for chunks that are being
3680 * converted - that will keep us from relocating unconverted
3681 * (albeit full) chunks.
3683 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3684 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3685 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3686 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3687 bctl->data.usage = 90;
3689 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3690 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3691 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3692 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3693 bctl->sys.usage = 90;
3695 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3696 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3697 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3698 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3699 bctl->meta.usage = 90;
3704 * Clear the balance status in fs_info and delete the balance item from disk.
3706 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3708 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3711 ASSERT(fs_info->balance_ctl);
3713 spin_lock(&fs_info->balance_lock);
3714 fs_info->balance_ctl = NULL;
3715 spin_unlock(&fs_info->balance_lock);
3718 ret = del_balance_item(fs_info);
3720 btrfs_handle_fs_error(fs_info, ret, NULL);
3724 * Balance filters. Return 1 if chunk should be filtered out
3725 * (should not be balanced).
3727 static int chunk_profiles_filter(u64 chunk_type,
3728 struct btrfs_balance_args *bargs)
3730 chunk_type = chunk_to_extended(chunk_type) &
3731 BTRFS_EXTENDED_PROFILE_MASK;
3733 if (bargs->profiles & chunk_type)
3739 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3740 struct btrfs_balance_args *bargs)
3742 struct btrfs_block_group *cache;
3744 u64 user_thresh_min;
3745 u64 user_thresh_max;
3748 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3749 chunk_used = cache->used;
3751 if (bargs->usage_min == 0)
3752 user_thresh_min = 0;
3754 user_thresh_min = mult_perc(cache->length, bargs->usage_min);
3756 if (bargs->usage_max == 0)
3757 user_thresh_max = 1;
3758 else if (bargs->usage_max > 100)
3759 user_thresh_max = cache->length;
3761 user_thresh_max = mult_perc(cache->length, bargs->usage_max);
3763 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3766 btrfs_put_block_group(cache);
3770 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3771 u64 chunk_offset, struct btrfs_balance_args *bargs)
3773 struct btrfs_block_group *cache;
3774 u64 chunk_used, user_thresh;
3777 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3778 chunk_used = cache->used;
3780 if (bargs->usage_min == 0)
3782 else if (bargs->usage > 100)
3783 user_thresh = cache->length;
3785 user_thresh = mult_perc(cache->length, bargs->usage);
3787 if (chunk_used < user_thresh)
3790 btrfs_put_block_group(cache);
3794 static int chunk_devid_filter(struct extent_buffer *leaf,
3795 struct btrfs_chunk *chunk,
3796 struct btrfs_balance_args *bargs)
3798 struct btrfs_stripe *stripe;
3799 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3802 for (i = 0; i < num_stripes; i++) {
3803 stripe = btrfs_stripe_nr(chunk, i);
3804 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3811 static u64 calc_data_stripes(u64 type, int num_stripes)
3813 const int index = btrfs_bg_flags_to_raid_index(type);
3814 const int ncopies = btrfs_raid_array[index].ncopies;
3815 const int nparity = btrfs_raid_array[index].nparity;
3817 return (num_stripes - nparity) / ncopies;
3820 /* [pstart, pend) */
3821 static int chunk_drange_filter(struct extent_buffer *leaf,
3822 struct btrfs_chunk *chunk,
3823 struct btrfs_balance_args *bargs)
3825 struct btrfs_stripe *stripe;
3826 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3833 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3836 type = btrfs_chunk_type(leaf, chunk);
3837 factor = calc_data_stripes(type, num_stripes);
3839 for (i = 0; i < num_stripes; i++) {
3840 stripe = btrfs_stripe_nr(chunk, i);
3841 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3844 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3845 stripe_length = btrfs_chunk_length(leaf, chunk);
3846 stripe_length = div_u64(stripe_length, factor);
3848 if (stripe_offset < bargs->pend &&
3849 stripe_offset + stripe_length > bargs->pstart)
3856 /* [vstart, vend) */
3857 static int chunk_vrange_filter(struct extent_buffer *leaf,
3858 struct btrfs_chunk *chunk,
3860 struct btrfs_balance_args *bargs)
3862 if (chunk_offset < bargs->vend &&
3863 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3864 /* at least part of the chunk is inside this vrange */
3870 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3871 struct btrfs_chunk *chunk,
3872 struct btrfs_balance_args *bargs)
3874 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3876 if (bargs->stripes_min <= num_stripes
3877 && num_stripes <= bargs->stripes_max)
3883 static int chunk_soft_convert_filter(u64 chunk_type,
3884 struct btrfs_balance_args *bargs)
3886 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3889 chunk_type = chunk_to_extended(chunk_type) &
3890 BTRFS_EXTENDED_PROFILE_MASK;
3892 if (bargs->target == chunk_type)
3898 static int should_balance_chunk(struct extent_buffer *leaf,
3899 struct btrfs_chunk *chunk, u64 chunk_offset)
3901 struct btrfs_fs_info *fs_info = leaf->fs_info;
3902 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3903 struct btrfs_balance_args *bargs = NULL;
3904 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3907 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3908 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3912 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3913 bargs = &bctl->data;
3914 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3916 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3917 bargs = &bctl->meta;
3919 /* profiles filter */
3920 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3921 chunk_profiles_filter(chunk_type, bargs)) {
3926 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3927 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3929 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3930 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3935 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3936 chunk_devid_filter(leaf, chunk, bargs)) {
3940 /* drange filter, makes sense only with devid filter */
3941 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3942 chunk_drange_filter(leaf, chunk, bargs)) {
3947 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3948 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3952 /* stripes filter */
3953 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3954 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3958 /* soft profile changing mode */
3959 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3960 chunk_soft_convert_filter(chunk_type, bargs)) {
3965 * limited by count, must be the last filter
3967 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3968 if (bargs->limit == 0)
3972 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3974 * Same logic as the 'limit' filter; the minimum cannot be
3975 * determined here because we do not have the global information
3976 * about the count of all chunks that satisfy the filters.
3978 if (bargs->limit_max == 0)
3987 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3989 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3990 struct btrfs_root *chunk_root = fs_info->chunk_root;
3992 struct btrfs_chunk *chunk;
3993 struct btrfs_path *path = NULL;
3994 struct btrfs_key key;
3995 struct btrfs_key found_key;
3996 struct extent_buffer *leaf;
3999 int enospc_errors = 0;
4000 bool counting = true;
4001 /* The single value limit and min/max limits use the same bytes in the */
4002 u64 limit_data = bctl->data.limit;
4003 u64 limit_meta = bctl->meta.limit;
4004 u64 limit_sys = bctl->sys.limit;
4008 int chunk_reserved = 0;
4010 path = btrfs_alloc_path();
4016 /* zero out stat counters */
4017 spin_lock(&fs_info->balance_lock);
4018 memset(&bctl->stat, 0, sizeof(bctl->stat));
4019 spin_unlock(&fs_info->balance_lock);
4023 * The single value limit and min/max limits use the same bytes
4026 bctl->data.limit = limit_data;
4027 bctl->meta.limit = limit_meta;
4028 bctl->sys.limit = limit_sys;
4030 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4031 key.offset = (u64)-1;
4032 key.type = BTRFS_CHUNK_ITEM_KEY;
4035 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
4036 atomic_read(&fs_info->balance_cancel_req)) {
4041 mutex_lock(&fs_info->reclaim_bgs_lock);
4042 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
4044 mutex_unlock(&fs_info->reclaim_bgs_lock);
4049 * this shouldn't happen, it means the last relocate
4053 BUG(); /* FIXME break ? */
4055 ret = btrfs_previous_item(chunk_root, path, 0,
4056 BTRFS_CHUNK_ITEM_KEY);
4058 mutex_unlock(&fs_info->reclaim_bgs_lock);
4063 leaf = path->nodes[0];
4064 slot = path->slots[0];
4065 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4067 if (found_key.objectid != key.objectid) {
4068 mutex_unlock(&fs_info->reclaim_bgs_lock);
4072 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
4073 chunk_type = btrfs_chunk_type(leaf, chunk);
4076 spin_lock(&fs_info->balance_lock);
4077 bctl->stat.considered++;
4078 spin_unlock(&fs_info->balance_lock);
4081 ret = should_balance_chunk(leaf, chunk, found_key.offset);
4083 btrfs_release_path(path);
4085 mutex_unlock(&fs_info->reclaim_bgs_lock);
4090 mutex_unlock(&fs_info->reclaim_bgs_lock);
4091 spin_lock(&fs_info->balance_lock);
4092 bctl->stat.expected++;
4093 spin_unlock(&fs_info->balance_lock);
4095 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
4097 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
4099 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
4106 * Apply limit_min filter, no need to check if the LIMITS
4107 * filter is used, limit_min is 0 by default
4109 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
4110 count_data < bctl->data.limit_min)
4111 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
4112 count_meta < bctl->meta.limit_min)
4113 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
4114 count_sys < bctl->sys.limit_min)) {
4115 mutex_unlock(&fs_info->reclaim_bgs_lock);
4119 if (!chunk_reserved) {
4121 * We may be relocating the only data chunk we have,
4122 * which could potentially end up with losing data's
4123 * raid profile, so lets allocate an empty one in
4126 ret = btrfs_may_alloc_data_chunk(fs_info,
4129 mutex_unlock(&fs_info->reclaim_bgs_lock);
4131 } else if (ret == 1) {
4136 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
4137 mutex_unlock(&fs_info->reclaim_bgs_lock);
4138 if (ret == -ENOSPC) {
4140 } else if (ret == -ETXTBSY) {
4142 "skipping relocation of block group %llu due to active swapfile",
4148 spin_lock(&fs_info->balance_lock);
4149 bctl->stat.completed++;
4150 spin_unlock(&fs_info->balance_lock);
4153 if (found_key.offset == 0)
4155 key.offset = found_key.offset - 1;
4159 btrfs_release_path(path);
4164 btrfs_free_path(path);
4165 if (enospc_errors) {
4166 btrfs_info(fs_info, "%d enospc errors during balance",
4176 * See if a given profile is valid and reduced.
4178 * @flags: profile to validate
4179 * @extended: if true @flags is treated as an extended profile
4181 static int alloc_profile_is_valid(u64 flags, int extended)
4183 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
4184 BTRFS_BLOCK_GROUP_PROFILE_MASK);
4186 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
4188 /* 1) check that all other bits are zeroed */
4192 /* 2) see if profile is reduced */
4194 return !extended; /* "0" is valid for usual profiles */
4196 return has_single_bit_set(flags);
4200 * Validate target profile against allowed profiles and return true if it's OK.
4201 * Otherwise print the error message and return false.
4203 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
4204 const struct btrfs_balance_args *bargs,
4205 u64 allowed, const char *type)
4207 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
4210 /* Profile is valid and does not have bits outside of the allowed set */
4211 if (alloc_profile_is_valid(bargs->target, 1) &&
4212 (bargs->target & ~allowed) == 0)
4215 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
4216 type, btrfs_bg_type_to_raid_name(bargs->target));
4221 * Fill @buf with textual description of balance filter flags @bargs, up to
4222 * @size_buf including the terminating null. The output may be trimmed if it
4223 * does not fit into the provided buffer.
4225 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
4229 u32 size_bp = size_buf;
4231 u64 flags = bargs->flags;
4232 char tmp_buf[128] = {'\0'};
4237 #define CHECK_APPEND_NOARG(a) \
4239 ret = snprintf(bp, size_bp, (a)); \
4240 if (ret < 0 || ret >= size_bp) \
4241 goto out_overflow; \
4246 #define CHECK_APPEND_1ARG(a, v1) \
4248 ret = snprintf(bp, size_bp, (a), (v1)); \
4249 if (ret < 0 || ret >= size_bp) \
4250 goto out_overflow; \
4255 #define CHECK_APPEND_2ARG(a, v1, v2) \
4257 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4258 if (ret < 0 || ret >= size_bp) \
4259 goto out_overflow; \
4264 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
4265 CHECK_APPEND_1ARG("convert=%s,",
4266 btrfs_bg_type_to_raid_name(bargs->target));
4268 if (flags & BTRFS_BALANCE_ARGS_SOFT)
4269 CHECK_APPEND_NOARG("soft,");
4271 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4272 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4274 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4277 if (flags & BTRFS_BALANCE_ARGS_USAGE)
4278 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4280 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4281 CHECK_APPEND_2ARG("usage=%u..%u,",
4282 bargs->usage_min, bargs->usage_max);
4284 if (flags & BTRFS_BALANCE_ARGS_DEVID)
4285 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4287 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4288 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4289 bargs->pstart, bargs->pend);
4291 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4292 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4293 bargs->vstart, bargs->vend);
4295 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4296 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4298 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4299 CHECK_APPEND_2ARG("limit=%u..%u,",
4300 bargs->limit_min, bargs->limit_max);
4302 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4303 CHECK_APPEND_2ARG("stripes=%u..%u,",
4304 bargs->stripes_min, bargs->stripes_max);
4306 #undef CHECK_APPEND_2ARG
4307 #undef CHECK_APPEND_1ARG
4308 #undef CHECK_APPEND_NOARG
4312 if (size_bp < size_buf)
4313 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4318 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4320 u32 size_buf = 1024;
4321 char tmp_buf[192] = {'\0'};
4324 u32 size_bp = size_buf;
4326 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4328 buf = kzalloc(size_buf, GFP_KERNEL);
4334 #define CHECK_APPEND_1ARG(a, v1) \
4336 ret = snprintf(bp, size_bp, (a), (v1)); \
4337 if (ret < 0 || ret >= size_bp) \
4338 goto out_overflow; \
4343 if (bctl->flags & BTRFS_BALANCE_FORCE)
4344 CHECK_APPEND_1ARG("%s", "-f ");
4346 if (bctl->flags & BTRFS_BALANCE_DATA) {
4347 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4348 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4351 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4352 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4353 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4356 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4357 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4358 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4361 #undef CHECK_APPEND_1ARG
4365 if (size_bp < size_buf)
4366 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4367 btrfs_info(fs_info, "balance: %s %s",
4368 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4369 "resume" : "start", buf);
4375 * Should be called with balance mutexe held
4377 int btrfs_balance(struct btrfs_fs_info *fs_info,
4378 struct btrfs_balance_control *bctl,
4379 struct btrfs_ioctl_balance_args *bargs)
4381 u64 meta_target, data_target;
4387 bool reducing_redundancy;
4388 bool paused = false;
4391 if (btrfs_fs_closing(fs_info) ||
4392 atomic_read(&fs_info->balance_pause_req) ||
4393 btrfs_should_cancel_balance(fs_info)) {
4398 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4399 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4403 * In case of mixed groups both data and meta should be picked,
4404 * and identical options should be given for both of them.
4406 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4407 if (mixed && (bctl->flags & allowed)) {
4408 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4409 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4410 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4412 "balance: mixed groups data and metadata options must be the same");
4419 * rw_devices will not change at the moment, device add/delete/replace
4422 num_devices = fs_info->fs_devices->rw_devices;
4425 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4426 * special bit for it, to make it easier to distinguish. Thus we need
4427 * to set it manually, or balance would refuse the profile.
4429 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4430 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4431 if (num_devices >= btrfs_raid_array[i].devs_min)
4432 allowed |= btrfs_raid_array[i].bg_flag;
4434 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4435 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4436 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4442 * Allow to reduce metadata or system integrity only if force set for
4443 * profiles with redundancy (copies, parity)
4446 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4447 if (btrfs_raid_array[i].ncopies >= 2 ||
4448 btrfs_raid_array[i].tolerated_failures >= 1)
4449 allowed |= btrfs_raid_array[i].bg_flag;
4452 seq = read_seqbegin(&fs_info->profiles_lock);
4454 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4455 (fs_info->avail_system_alloc_bits & allowed) &&
4456 !(bctl->sys.target & allowed)) ||
4457 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4458 (fs_info->avail_metadata_alloc_bits & allowed) &&
4459 !(bctl->meta.target & allowed)))
4460 reducing_redundancy = true;
4462 reducing_redundancy = false;
4464 /* if we're not converting, the target field is uninitialized */
4465 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4466 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4467 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4468 bctl->data.target : fs_info->avail_data_alloc_bits;
4469 } while (read_seqretry(&fs_info->profiles_lock, seq));
4471 if (reducing_redundancy) {
4472 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4474 "balance: force reducing metadata redundancy");
4477 "balance: reduces metadata redundancy, use --force if you want this");
4483 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4484 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4486 "balance: metadata profile %s has lower redundancy than data profile %s",
4487 btrfs_bg_type_to_raid_name(meta_target),
4488 btrfs_bg_type_to_raid_name(data_target));
4491 ret = insert_balance_item(fs_info, bctl);
4492 if (ret && ret != -EEXIST)
4495 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4496 BUG_ON(ret == -EEXIST);
4497 BUG_ON(fs_info->balance_ctl);
4498 spin_lock(&fs_info->balance_lock);
4499 fs_info->balance_ctl = bctl;
4500 spin_unlock(&fs_info->balance_lock);
4502 BUG_ON(ret != -EEXIST);
4503 spin_lock(&fs_info->balance_lock);
4504 update_balance_args(bctl);
4505 spin_unlock(&fs_info->balance_lock);
4508 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4509 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4510 describe_balance_start_or_resume(fs_info);
4511 mutex_unlock(&fs_info->balance_mutex);
4513 ret = __btrfs_balance(fs_info);
4515 mutex_lock(&fs_info->balance_mutex);
4516 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) {
4517 btrfs_info(fs_info, "balance: paused");
4518 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
4522 * Balance can be canceled by:
4524 * - Regular cancel request
4525 * Then ret == -ECANCELED and balance_cancel_req > 0
4527 * - Fatal signal to "btrfs" process
4528 * Either the signal caught by wait_reserve_ticket() and callers
4529 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4531 * Either way, in this case balance_cancel_req = 0, and
4532 * ret == -EINTR or ret == -ECANCELED.
4534 * So here we only check the return value to catch canceled balance.
4536 else if (ret == -ECANCELED || ret == -EINTR)
4537 btrfs_info(fs_info, "balance: canceled");
4539 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4541 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4544 memset(bargs, 0, sizeof(*bargs));
4545 btrfs_update_ioctl_balance_args(fs_info, bargs);
4548 /* We didn't pause, we can clean everything up. */
4550 reset_balance_state(fs_info);
4551 btrfs_exclop_finish(fs_info);
4554 wake_up(&fs_info->balance_wait_q);
4558 if (bctl->flags & BTRFS_BALANCE_RESUME)
4559 reset_balance_state(fs_info);
4562 btrfs_exclop_finish(fs_info);
4567 static int balance_kthread(void *data)
4569 struct btrfs_fs_info *fs_info = data;
4572 sb_start_write(fs_info->sb);
4573 mutex_lock(&fs_info->balance_mutex);
4574 if (fs_info->balance_ctl)
4575 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4576 mutex_unlock(&fs_info->balance_mutex);
4577 sb_end_write(fs_info->sb);
4582 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4584 struct task_struct *tsk;
4586 mutex_lock(&fs_info->balance_mutex);
4587 if (!fs_info->balance_ctl) {
4588 mutex_unlock(&fs_info->balance_mutex);
4591 mutex_unlock(&fs_info->balance_mutex);
4593 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4594 btrfs_info(fs_info, "balance: resume skipped");
4598 spin_lock(&fs_info->super_lock);
4599 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
4600 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;
4601 spin_unlock(&fs_info->super_lock);
4603 * A ro->rw remount sequence should continue with the paused balance
4604 * regardless of who pauses it, system or the user as of now, so set
4607 spin_lock(&fs_info->balance_lock);
4608 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4609 spin_unlock(&fs_info->balance_lock);
4611 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4612 return PTR_ERR_OR_ZERO(tsk);
4615 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4617 struct btrfs_balance_control *bctl;
4618 struct btrfs_balance_item *item;
4619 struct btrfs_disk_balance_args disk_bargs;
4620 struct btrfs_path *path;
4621 struct extent_buffer *leaf;
4622 struct btrfs_key key;
4625 path = btrfs_alloc_path();
4629 key.objectid = BTRFS_BALANCE_OBJECTID;
4630 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4633 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4636 if (ret > 0) { /* ret = -ENOENT; */
4641 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4647 leaf = path->nodes[0];
4648 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4650 bctl->flags = btrfs_balance_flags(leaf, item);
4651 bctl->flags |= BTRFS_BALANCE_RESUME;
4653 btrfs_balance_data(leaf, item, &disk_bargs);
4654 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4655 btrfs_balance_meta(leaf, item, &disk_bargs);
4656 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4657 btrfs_balance_sys(leaf, item, &disk_bargs);
4658 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4661 * This should never happen, as the paused balance state is recovered
4662 * during mount without any chance of other exclusive ops to collide.
4664 * This gives the exclusive op status to balance and keeps in paused
4665 * state until user intervention (cancel or umount). If the ownership
4666 * cannot be assigned, show a message but do not fail. The balance
4667 * is in a paused state and must have fs_info::balance_ctl properly
4670 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED))
4672 "balance: cannot set exclusive op status, resume manually");
4674 btrfs_release_path(path);
4676 mutex_lock(&fs_info->balance_mutex);
4677 BUG_ON(fs_info->balance_ctl);
4678 spin_lock(&fs_info->balance_lock);
4679 fs_info->balance_ctl = bctl;
4680 spin_unlock(&fs_info->balance_lock);
4681 mutex_unlock(&fs_info->balance_mutex);
4683 btrfs_free_path(path);
4687 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4691 mutex_lock(&fs_info->balance_mutex);
4692 if (!fs_info->balance_ctl) {
4693 mutex_unlock(&fs_info->balance_mutex);
4697 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4698 atomic_inc(&fs_info->balance_pause_req);
4699 mutex_unlock(&fs_info->balance_mutex);
4701 wait_event(fs_info->balance_wait_q,
4702 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4704 mutex_lock(&fs_info->balance_mutex);
4705 /* we are good with balance_ctl ripped off from under us */
4706 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4707 atomic_dec(&fs_info->balance_pause_req);
4712 mutex_unlock(&fs_info->balance_mutex);
4716 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4718 mutex_lock(&fs_info->balance_mutex);
4719 if (!fs_info->balance_ctl) {
4720 mutex_unlock(&fs_info->balance_mutex);
4725 * A paused balance with the item stored on disk can be resumed at
4726 * mount time if the mount is read-write. Otherwise it's still paused
4727 * and we must not allow cancelling as it deletes the item.
4729 if (sb_rdonly(fs_info->sb)) {
4730 mutex_unlock(&fs_info->balance_mutex);
4734 atomic_inc(&fs_info->balance_cancel_req);
4736 * if we are running just wait and return, balance item is
4737 * deleted in btrfs_balance in this case
4739 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4740 mutex_unlock(&fs_info->balance_mutex);
4741 wait_event(fs_info->balance_wait_q,
4742 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4743 mutex_lock(&fs_info->balance_mutex);
4745 mutex_unlock(&fs_info->balance_mutex);
4747 * Lock released to allow other waiters to continue, we'll
4748 * reexamine the status again.
4750 mutex_lock(&fs_info->balance_mutex);
4752 if (fs_info->balance_ctl) {
4753 reset_balance_state(fs_info);
4754 btrfs_exclop_finish(fs_info);
4755 btrfs_info(fs_info, "balance: canceled");
4759 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4760 atomic_dec(&fs_info->balance_cancel_req);
4761 mutex_unlock(&fs_info->balance_mutex);
4765 int btrfs_uuid_scan_kthread(void *data)
4767 struct btrfs_fs_info *fs_info = data;
4768 struct btrfs_root *root = fs_info->tree_root;
4769 struct btrfs_key key;
4770 struct btrfs_path *path = NULL;
4772 struct extent_buffer *eb;
4774 struct btrfs_root_item root_item;
4776 struct btrfs_trans_handle *trans = NULL;
4777 bool closing = false;
4779 path = btrfs_alloc_path();
4786 key.type = BTRFS_ROOT_ITEM_KEY;
4790 if (btrfs_fs_closing(fs_info)) {
4794 ret = btrfs_search_forward(root, &key, path,
4795 BTRFS_OLDEST_GENERATION);
4802 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4803 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4804 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4805 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4808 eb = path->nodes[0];
4809 slot = path->slots[0];
4810 item_size = btrfs_item_size(eb, slot);
4811 if (item_size < sizeof(root_item))
4814 read_extent_buffer(eb, &root_item,
4815 btrfs_item_ptr_offset(eb, slot),
4816 (int)sizeof(root_item));
4817 if (btrfs_root_refs(&root_item) == 0)
4820 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4821 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4825 btrfs_release_path(path);
4827 * 1 - subvol uuid item
4828 * 1 - received_subvol uuid item
4830 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4831 if (IS_ERR(trans)) {
4832 ret = PTR_ERR(trans);
4840 btrfs_release_path(path);
4841 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4842 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4843 BTRFS_UUID_KEY_SUBVOL,
4846 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4852 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4853 ret = btrfs_uuid_tree_add(trans,
4854 root_item.received_uuid,
4855 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4858 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4865 btrfs_release_path(path);
4867 ret = btrfs_end_transaction(trans);
4873 if (key.offset < (u64)-1) {
4875 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4877 key.type = BTRFS_ROOT_ITEM_KEY;
4878 } else if (key.objectid < (u64)-1) {
4880 key.type = BTRFS_ROOT_ITEM_KEY;
4889 btrfs_free_path(path);
4890 if (trans && !IS_ERR(trans))
4891 btrfs_end_transaction(trans);
4893 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4895 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4896 up(&fs_info->uuid_tree_rescan_sem);
4900 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4902 struct btrfs_trans_handle *trans;
4903 struct btrfs_root *tree_root = fs_info->tree_root;
4904 struct btrfs_root *uuid_root;
4905 struct task_struct *task;
4912 trans = btrfs_start_transaction(tree_root, 2);
4914 return PTR_ERR(trans);
4916 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4917 if (IS_ERR(uuid_root)) {
4918 ret = PTR_ERR(uuid_root);
4919 btrfs_abort_transaction(trans, ret);
4920 btrfs_end_transaction(trans);
4924 fs_info->uuid_root = uuid_root;
4926 ret = btrfs_commit_transaction(trans);
4930 down(&fs_info->uuid_tree_rescan_sem);
4931 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4933 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4934 btrfs_warn(fs_info, "failed to start uuid_scan task");
4935 up(&fs_info->uuid_tree_rescan_sem);
4936 return PTR_ERR(task);
4943 * shrinking a device means finding all of the device extents past
4944 * the new size, and then following the back refs to the chunks.
4945 * The chunk relocation code actually frees the device extent
4947 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4949 struct btrfs_fs_info *fs_info = device->fs_info;
4950 struct btrfs_root *root = fs_info->dev_root;
4951 struct btrfs_trans_handle *trans;
4952 struct btrfs_dev_extent *dev_extent = NULL;
4953 struct btrfs_path *path;
4959 bool retried = false;
4960 struct extent_buffer *l;
4961 struct btrfs_key key;
4962 struct btrfs_super_block *super_copy = fs_info->super_copy;
4963 u64 old_total = btrfs_super_total_bytes(super_copy);
4964 u64 old_size = btrfs_device_get_total_bytes(device);
4969 new_size = round_down(new_size, fs_info->sectorsize);
4971 diff = round_down(old_size - new_size, fs_info->sectorsize);
4973 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4976 path = btrfs_alloc_path();
4980 path->reada = READA_BACK;
4982 trans = btrfs_start_transaction(root, 0);
4983 if (IS_ERR(trans)) {
4984 btrfs_free_path(path);
4985 return PTR_ERR(trans);
4988 mutex_lock(&fs_info->chunk_mutex);
4990 btrfs_device_set_total_bytes(device, new_size);
4991 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4992 device->fs_devices->total_rw_bytes -= diff;
4995 * The new free_chunk_space is new_size - used, so we have to
4996 * subtract the delta of the old free_chunk_space which included
4997 * old_size - used. If used > new_size then just subtract this
4998 * entire device's free space.
5000 if (device->bytes_used < new_size)
5001 free_diff = (old_size - device->bytes_used) -
5002 (new_size - device->bytes_used);
5004 free_diff = old_size - device->bytes_used;
5005 atomic64_sub(free_diff, &fs_info->free_chunk_space);
5009 * Once the device's size has been set to the new size, ensure all
5010 * in-memory chunks are synced to disk so that the loop below sees them
5011 * and relocates them accordingly.
5013 if (contains_pending_extent(device, &start, diff)) {
5014 mutex_unlock(&fs_info->chunk_mutex);
5015 ret = btrfs_commit_transaction(trans);
5019 mutex_unlock(&fs_info->chunk_mutex);
5020 btrfs_end_transaction(trans);
5024 key.objectid = device->devid;
5025 key.offset = (u64)-1;
5026 key.type = BTRFS_DEV_EXTENT_KEY;
5029 mutex_lock(&fs_info->reclaim_bgs_lock);
5030 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5032 mutex_unlock(&fs_info->reclaim_bgs_lock);
5036 ret = btrfs_previous_item(root, path, 0, key.type);
5038 mutex_unlock(&fs_info->reclaim_bgs_lock);
5042 btrfs_release_path(path);
5047 slot = path->slots[0];
5048 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
5050 if (key.objectid != device->devid) {
5051 mutex_unlock(&fs_info->reclaim_bgs_lock);
5052 btrfs_release_path(path);
5056 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
5057 length = btrfs_dev_extent_length(l, dev_extent);
5059 if (key.offset + length <= new_size) {
5060 mutex_unlock(&fs_info->reclaim_bgs_lock);
5061 btrfs_release_path(path);
5065 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
5066 btrfs_release_path(path);
5069 * We may be relocating the only data chunk we have,
5070 * which could potentially end up with losing data's
5071 * raid profile, so lets allocate an empty one in
5074 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
5076 mutex_unlock(&fs_info->reclaim_bgs_lock);
5080 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
5081 mutex_unlock(&fs_info->reclaim_bgs_lock);
5082 if (ret == -ENOSPC) {
5085 if (ret == -ETXTBSY) {
5087 "could not shrink block group %llu due to active swapfile",
5092 } while (key.offset-- > 0);
5094 if (failed && !retried) {
5098 } else if (failed && retried) {
5103 /* Shrinking succeeded, else we would be at "done". */
5104 trans = btrfs_start_transaction(root, 0);
5105 if (IS_ERR(trans)) {
5106 ret = PTR_ERR(trans);
5110 mutex_lock(&fs_info->chunk_mutex);
5111 /* Clear all state bits beyond the shrunk device size */
5112 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
5115 btrfs_device_set_disk_total_bytes(device, new_size);
5116 if (list_empty(&device->post_commit_list))
5117 list_add_tail(&device->post_commit_list,
5118 &trans->transaction->dev_update_list);
5120 WARN_ON(diff > old_total);
5121 btrfs_set_super_total_bytes(super_copy,
5122 round_down(old_total - diff, fs_info->sectorsize));
5123 mutex_unlock(&fs_info->chunk_mutex);
5125 btrfs_reserve_chunk_metadata(trans, false);
5126 /* Now btrfs_update_device() will change the on-disk size. */
5127 ret = btrfs_update_device(trans, device);
5128 btrfs_trans_release_chunk_metadata(trans);
5130 btrfs_abort_transaction(trans, ret);
5131 btrfs_end_transaction(trans);
5133 ret = btrfs_commit_transaction(trans);
5136 btrfs_free_path(path);
5138 mutex_lock(&fs_info->chunk_mutex);
5139 btrfs_device_set_total_bytes(device, old_size);
5140 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5141 device->fs_devices->total_rw_bytes += diff;
5142 atomic64_add(free_diff, &fs_info->free_chunk_space);
5144 mutex_unlock(&fs_info->chunk_mutex);
5149 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
5150 struct btrfs_key *key,
5151 struct btrfs_chunk *chunk, int item_size)
5153 struct btrfs_super_block *super_copy = fs_info->super_copy;
5154 struct btrfs_disk_key disk_key;
5158 lockdep_assert_held(&fs_info->chunk_mutex);
5160 array_size = btrfs_super_sys_array_size(super_copy);
5161 if (array_size + item_size + sizeof(disk_key)
5162 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
5165 ptr = super_copy->sys_chunk_array + array_size;
5166 btrfs_cpu_key_to_disk(&disk_key, key);
5167 memcpy(ptr, &disk_key, sizeof(disk_key));
5168 ptr += sizeof(disk_key);
5169 memcpy(ptr, chunk, item_size);
5170 item_size += sizeof(disk_key);
5171 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
5177 * sort the devices in descending order by max_avail, total_avail
5179 static int btrfs_cmp_device_info(const void *a, const void *b)
5181 const struct btrfs_device_info *di_a = a;
5182 const struct btrfs_device_info *di_b = b;
5184 if (di_a->max_avail > di_b->max_avail)
5186 if (di_a->max_avail < di_b->max_avail)
5188 if (di_a->total_avail > di_b->total_avail)
5190 if (di_a->total_avail < di_b->total_avail)
5195 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
5197 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5200 btrfs_set_fs_incompat(info, RAID56);
5203 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
5205 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
5208 btrfs_set_fs_incompat(info, RAID1C34);
5212 * Structure used internally for btrfs_create_chunk() function.
5213 * Wraps needed parameters.
5215 struct alloc_chunk_ctl {
5218 /* Total number of stripes to allocate */
5220 /* sub_stripes info for map */
5222 /* Stripes per device */
5224 /* Maximum number of devices to use */
5226 /* Minimum number of devices to use */
5228 /* ndevs has to be a multiple of this */
5230 /* Number of copies */
5232 /* Number of stripes worth of bytes to store parity information */
5234 u64 max_stripe_size;
5242 static void init_alloc_chunk_ctl_policy_regular(
5243 struct btrfs_fs_devices *fs_devices,
5244 struct alloc_chunk_ctl *ctl)
5246 struct btrfs_space_info *space_info;
5248 space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type);
5251 ctl->max_chunk_size = READ_ONCE(space_info->chunk_size);
5252 ctl->max_stripe_size = min_t(u64, ctl->max_chunk_size, SZ_1G);
5254 if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM)
5255 ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK);
5257 /* We don't want a chunk larger than 10% of writable space */
5258 ctl->max_chunk_size = min(mult_perc(fs_devices->total_rw_bytes, 10),
5259 ctl->max_chunk_size);
5260 ctl->dev_extent_min = btrfs_stripe_nr_to_offset(ctl->dev_stripes);
5263 static void init_alloc_chunk_ctl_policy_zoned(
5264 struct btrfs_fs_devices *fs_devices,
5265 struct alloc_chunk_ctl *ctl)
5267 u64 zone_size = fs_devices->fs_info->zone_size;
5269 int min_num_stripes = ctl->devs_min * ctl->dev_stripes;
5270 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;
5271 u64 min_chunk_size = min_data_stripes * zone_size;
5272 u64 type = ctl->type;
5274 ctl->max_stripe_size = zone_size;
5275 if (type & BTRFS_BLOCK_GROUP_DATA) {
5276 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,
5278 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5279 ctl->max_chunk_size = ctl->max_stripe_size;
5280 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5281 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5282 ctl->devs_max = min_t(int, ctl->devs_max,
5283 BTRFS_MAX_DEVS_SYS_CHUNK);
5288 /* We don't want a chunk larger than 10% of writable space */
5289 limit = max(round_down(mult_perc(fs_devices->total_rw_bytes, 10),
5292 ctl->max_chunk_size = min(limit, ctl->max_chunk_size);
5293 ctl->dev_extent_min = zone_size * ctl->dev_stripes;
5296 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
5297 struct alloc_chunk_ctl *ctl)
5299 int index = btrfs_bg_flags_to_raid_index(ctl->type);
5301 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
5302 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
5303 ctl->devs_max = btrfs_raid_array[index].devs_max;
5305 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
5306 ctl->devs_min = btrfs_raid_array[index].devs_min;
5307 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
5308 ctl->ncopies = btrfs_raid_array[index].ncopies;
5309 ctl->nparity = btrfs_raid_array[index].nparity;
5312 switch (fs_devices->chunk_alloc_policy) {
5313 case BTRFS_CHUNK_ALLOC_REGULAR:
5314 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
5316 case BTRFS_CHUNK_ALLOC_ZONED:
5317 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);
5324 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
5325 struct alloc_chunk_ctl *ctl,
5326 struct btrfs_device_info *devices_info)
5328 struct btrfs_fs_info *info = fs_devices->fs_info;
5329 struct btrfs_device *device;
5331 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
5338 * in the first pass through the devices list, we gather information
5339 * about the available holes on each device.
5341 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5342 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5344 "BTRFS: read-only device in alloc_list\n");
5348 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5349 &device->dev_state) ||
5350 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5353 if (device->total_bytes > device->bytes_used)
5354 total_avail = device->total_bytes - device->bytes_used;
5358 /* If there is no space on this device, skip it. */
5359 if (total_avail < ctl->dev_extent_min)
5362 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
5364 if (ret && ret != -ENOSPC)
5368 max_avail = dev_extent_want;
5370 if (max_avail < ctl->dev_extent_min) {
5371 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5373 "%s: devid %llu has no free space, have=%llu want=%llu",
5374 __func__, device->devid, max_avail,
5375 ctl->dev_extent_min);
5379 if (ndevs == fs_devices->rw_devices) {
5380 WARN(1, "%s: found more than %llu devices\n",
5381 __func__, fs_devices->rw_devices);
5384 devices_info[ndevs].dev_offset = dev_offset;
5385 devices_info[ndevs].max_avail = max_avail;
5386 devices_info[ndevs].total_avail = total_avail;
5387 devices_info[ndevs].dev = device;
5393 * now sort the devices by hole size / available space
5395 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5396 btrfs_cmp_device_info, NULL);
5401 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5402 struct btrfs_device_info *devices_info)
5404 /* Number of stripes that count for block group size */
5408 * The primary goal is to maximize the number of stripes, so use as
5409 * many devices as possible, even if the stripes are not maximum sized.
5411 * The DUP profile stores more than one stripe per device, the
5412 * max_avail is the total size so we have to adjust.
5414 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5416 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5418 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5419 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5422 * Use the number of data stripes to figure out how big this chunk is
5423 * really going to be in terms of logical address space, and compare
5424 * that answer with the max chunk size. If it's higher, we try to
5425 * reduce stripe_size.
5427 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5429 * Reduce stripe_size, round it up to a 16MB boundary again and
5430 * then use it, unless it ends up being even bigger than the
5431 * previous value we had already.
5433 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5434 data_stripes), SZ_16M),
5438 /* Stripe size should not go beyond 1G. */
5439 ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G);
5441 /* Align to BTRFS_STRIPE_LEN */
5442 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5443 ctl->chunk_size = ctl->stripe_size * data_stripes;
5448 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,
5449 struct btrfs_device_info *devices_info)
5451 u64 zone_size = devices_info[0].dev->zone_info->zone_size;
5452 /* Number of stripes that count for block group size */
5456 * It should hold because:
5457 * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5459 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);
5461 ctl->stripe_size = zone_size;
5462 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5463 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5465 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5466 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5467 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,
5468 ctl->stripe_size) + ctl->nparity,
5470 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5471 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5472 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);
5475 ctl->chunk_size = ctl->stripe_size * data_stripes;
5480 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5481 struct alloc_chunk_ctl *ctl,
5482 struct btrfs_device_info *devices_info)
5484 struct btrfs_fs_info *info = fs_devices->fs_info;
5487 * Round down to number of usable stripes, devs_increment can be any
5488 * number so we can't use round_down() that requires power of 2, while
5489 * rounddown is safe.
5491 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5493 if (ctl->ndevs < ctl->devs_min) {
5494 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5496 "%s: not enough devices with free space: have=%d minimum required=%d",
5497 __func__, ctl->ndevs, ctl->devs_min);
5502 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5504 switch (fs_devices->chunk_alloc_policy) {
5505 case BTRFS_CHUNK_ALLOC_REGULAR:
5506 return decide_stripe_size_regular(ctl, devices_info);
5507 case BTRFS_CHUNK_ALLOC_ZONED:
5508 return decide_stripe_size_zoned(ctl, devices_info);
5514 static void chunk_map_device_set_bits(struct btrfs_chunk_map *map, unsigned int bits)
5516 for (int i = 0; i < map->num_stripes; i++) {
5517 struct btrfs_io_stripe *stripe = &map->stripes[i];
5518 struct btrfs_device *device = stripe->dev;
5520 set_extent_bit(&device->alloc_state, stripe->physical,
5521 stripe->physical + map->stripe_size - 1,
5522 bits | EXTENT_NOWAIT, NULL);
5526 static void chunk_map_device_clear_bits(struct btrfs_chunk_map *map, unsigned int bits)
5528 for (int i = 0; i < map->num_stripes; i++) {
5529 struct btrfs_io_stripe *stripe = &map->stripes[i];
5530 struct btrfs_device *device = stripe->dev;
5532 __clear_extent_bit(&device->alloc_state, stripe->physical,
5533 stripe->physical + map->stripe_size - 1,
5534 bits | EXTENT_NOWAIT,
5539 void btrfs_remove_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map)
5541 write_lock(&fs_info->mapping_tree_lock);
5542 rb_erase_cached(&map->rb_node, &fs_info->mapping_tree);
5543 RB_CLEAR_NODE(&map->rb_node);
5544 chunk_map_device_clear_bits(map, CHUNK_ALLOCATED);
5545 write_unlock(&fs_info->mapping_tree_lock);
5547 /* Once for the tree reference. */
5548 btrfs_free_chunk_map(map);
5552 int btrfs_add_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map)
5555 struct rb_node *parent = NULL;
5556 bool leftmost = true;
5558 write_lock(&fs_info->mapping_tree_lock);
5559 p = &fs_info->mapping_tree.rb_root.rb_node;
5561 struct btrfs_chunk_map *entry;
5564 entry = rb_entry(parent, struct btrfs_chunk_map, rb_node);
5566 if (map->start < entry->start) {
5568 } else if (map->start > entry->start) {
5569 p = &(*p)->rb_right;
5572 write_unlock(&fs_info->mapping_tree_lock);
5576 rb_link_node(&map->rb_node, parent, p);
5577 rb_insert_color_cached(&map->rb_node, &fs_info->mapping_tree, leftmost);
5578 chunk_map_device_set_bits(map, CHUNK_ALLOCATED);
5579 chunk_map_device_clear_bits(map, CHUNK_TRIMMED);
5580 write_unlock(&fs_info->mapping_tree_lock);
5586 struct btrfs_chunk_map *btrfs_alloc_chunk_map(int num_stripes, gfp_t gfp)
5588 struct btrfs_chunk_map *map;
5590 map = kmalloc(btrfs_chunk_map_size(num_stripes), gfp);
5594 refcount_set(&map->refs, 1);
5595 RB_CLEAR_NODE(&map->rb_node);
5600 struct btrfs_chunk_map *btrfs_clone_chunk_map(struct btrfs_chunk_map *map, gfp_t gfp)
5602 const int size = btrfs_chunk_map_size(map->num_stripes);
5603 struct btrfs_chunk_map *clone;
5605 clone = kmemdup(map, size, gfp);
5609 refcount_set(&clone->refs, 1);
5610 RB_CLEAR_NODE(&clone->rb_node);
5615 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
5616 struct alloc_chunk_ctl *ctl,
5617 struct btrfs_device_info *devices_info)
5619 struct btrfs_fs_info *info = trans->fs_info;
5620 struct btrfs_chunk_map *map;
5621 struct btrfs_block_group *block_group;
5622 u64 start = ctl->start;
5623 u64 type = ctl->type;
5628 map = btrfs_alloc_chunk_map(ctl->num_stripes, GFP_NOFS);
5630 return ERR_PTR(-ENOMEM);
5633 map->chunk_len = ctl->chunk_size;
5634 map->stripe_size = ctl->stripe_size;
5636 map->io_align = BTRFS_STRIPE_LEN;
5637 map->io_width = BTRFS_STRIPE_LEN;
5638 map->sub_stripes = ctl->sub_stripes;
5639 map->num_stripes = ctl->num_stripes;
5641 for (i = 0; i < ctl->ndevs; ++i) {
5642 for (j = 0; j < ctl->dev_stripes; ++j) {
5643 int s = i * ctl->dev_stripes + j;
5644 map->stripes[s].dev = devices_info[i].dev;
5645 map->stripes[s].physical = devices_info[i].dev_offset +
5646 j * ctl->stripe_size;
5650 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5652 ret = btrfs_add_chunk_map(info, map);
5654 btrfs_free_chunk_map(map);
5655 return ERR_PTR(ret);
5658 block_group = btrfs_make_block_group(trans, type, start, ctl->chunk_size);
5659 if (IS_ERR(block_group)) {
5660 btrfs_remove_chunk_map(info, map);
5664 for (int i = 0; i < map->num_stripes; i++) {
5665 struct btrfs_device *dev = map->stripes[i].dev;
5667 btrfs_device_set_bytes_used(dev,
5668 dev->bytes_used + ctl->stripe_size);
5669 if (list_empty(&dev->post_commit_list))
5670 list_add_tail(&dev->post_commit_list,
5671 &trans->transaction->dev_update_list);
5674 atomic64_sub(ctl->stripe_size * map->num_stripes,
5675 &info->free_chunk_space);
5677 check_raid56_incompat_flag(info, type);
5678 check_raid1c34_incompat_flag(info, type);
5683 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
5686 struct btrfs_fs_info *info = trans->fs_info;
5687 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5688 struct btrfs_device_info *devices_info = NULL;
5689 struct alloc_chunk_ctl ctl;
5690 struct btrfs_block_group *block_group;
5693 lockdep_assert_held(&info->chunk_mutex);
5695 if (!alloc_profile_is_valid(type, 0)) {
5697 return ERR_PTR(-EINVAL);
5700 if (list_empty(&fs_devices->alloc_list)) {
5701 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5702 btrfs_debug(info, "%s: no writable device", __func__);
5703 return ERR_PTR(-ENOSPC);
5706 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5707 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5709 return ERR_PTR(-EINVAL);
5712 ctl.start = find_next_chunk(info);
5714 init_alloc_chunk_ctl(fs_devices, &ctl);
5716 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5719 return ERR_PTR(-ENOMEM);
5721 ret = gather_device_info(fs_devices, &ctl, devices_info);
5723 block_group = ERR_PTR(ret);
5727 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5729 block_group = ERR_PTR(ret);
5733 block_group = create_chunk(trans, &ctl, devices_info);
5736 kfree(devices_info);
5741 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5742 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5745 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5748 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
5749 struct btrfs_block_group *bg)
5751 struct btrfs_fs_info *fs_info = trans->fs_info;
5752 struct btrfs_root *chunk_root = fs_info->chunk_root;
5753 struct btrfs_key key;
5754 struct btrfs_chunk *chunk;
5755 struct btrfs_stripe *stripe;
5756 struct btrfs_chunk_map *map;
5762 * We take the chunk_mutex for 2 reasons:
5764 * 1) Updates and insertions in the chunk btree must be done while holding
5765 * the chunk_mutex, as well as updating the system chunk array in the
5766 * superblock. See the comment on top of btrfs_chunk_alloc() for the
5769 * 2) To prevent races with the final phase of a device replace operation
5770 * that replaces the device object associated with the map's stripes,
5771 * because the device object's id can change at any time during that
5772 * final phase of the device replace operation
5773 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5774 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5775 * which would cause a failure when updating the device item, which does
5776 * not exists, or persisting a stripe of the chunk item with such ID.
5777 * Here we can't use the device_list_mutex because our caller already
5778 * has locked the chunk_mutex, and the final phase of device replace
5779 * acquires both mutexes - first the device_list_mutex and then the
5780 * chunk_mutex. Using any of those two mutexes protects us from a
5781 * concurrent device replace.
5783 lockdep_assert_held(&fs_info->chunk_mutex);
5785 map = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
5788 btrfs_abort_transaction(trans, ret);
5792 item_size = btrfs_chunk_item_size(map->num_stripes);
5794 chunk = kzalloc(item_size, GFP_NOFS);
5797 btrfs_abort_transaction(trans, ret);
5801 for (i = 0; i < map->num_stripes; i++) {
5802 struct btrfs_device *device = map->stripes[i].dev;
5804 ret = btrfs_update_device(trans, device);
5809 stripe = &chunk->stripe;
5810 for (i = 0; i < map->num_stripes; i++) {
5811 struct btrfs_device *device = map->stripes[i].dev;
5812 const u64 dev_offset = map->stripes[i].physical;
5814 btrfs_set_stack_stripe_devid(stripe, device->devid);
5815 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5816 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5820 btrfs_set_stack_chunk_length(chunk, bg->length);
5821 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
5822 btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
5823 btrfs_set_stack_chunk_type(chunk, map->type);
5824 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5825 btrfs_set_stack_chunk_io_align(chunk, BTRFS_STRIPE_LEN);
5826 btrfs_set_stack_chunk_io_width(chunk, BTRFS_STRIPE_LEN);
5827 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5828 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5830 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5831 key.type = BTRFS_CHUNK_ITEM_KEY;
5832 key.offset = bg->start;
5834 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5838 set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags);
5840 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5841 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5848 btrfs_free_chunk_map(map);
5852 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5854 struct btrfs_fs_info *fs_info = trans->fs_info;
5856 struct btrfs_block_group *meta_bg;
5857 struct btrfs_block_group *sys_bg;
5860 * When adding a new device for sprouting, the seed device is read-only
5861 * so we must first allocate a metadata and a system chunk. But before
5862 * adding the block group items to the extent, device and chunk btrees,
5865 * 1) Create both chunks without doing any changes to the btrees, as
5866 * otherwise we would get -ENOSPC since the block groups from the
5867 * seed device are read-only;
5869 * 2) Add the device item for the new sprout device - finishing the setup
5870 * of a new block group requires updating the device item in the chunk
5871 * btree, so it must exist when we attempt to do it. The previous step
5872 * ensures this does not fail with -ENOSPC.
5874 * After that we can add the block group items to their btrees:
5875 * update existing device item in the chunk btree, add a new block group
5876 * item to the extent btree, add a new chunk item to the chunk btree and
5877 * finally add the new device extent items to the devices btree.
5880 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5881 meta_bg = btrfs_create_chunk(trans, alloc_profile);
5882 if (IS_ERR(meta_bg))
5883 return PTR_ERR(meta_bg);
5885 alloc_profile = btrfs_system_alloc_profile(fs_info);
5886 sys_bg = btrfs_create_chunk(trans, alloc_profile);
5888 return PTR_ERR(sys_bg);
5893 static inline int btrfs_chunk_max_errors(struct btrfs_chunk_map *map)
5895 const int index = btrfs_bg_flags_to_raid_index(map->type);
5897 return btrfs_raid_array[index].tolerated_failures;
5900 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5902 struct btrfs_chunk_map *map;
5907 map = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5911 for (i = 0; i < map->num_stripes; i++) {
5912 if (test_bit(BTRFS_DEV_STATE_MISSING,
5913 &map->stripes[i].dev->dev_state)) {
5917 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5918 &map->stripes[i].dev->dev_state)) {
5925 * If the number of missing devices is larger than max errors, we can
5926 * not write the data into that chunk successfully.
5928 if (miss_ndevs > btrfs_chunk_max_errors(map))
5931 btrfs_free_chunk_map(map);
5935 void btrfs_mapping_tree_free(struct btrfs_fs_info *fs_info)
5937 write_lock(&fs_info->mapping_tree_lock);
5938 while (!RB_EMPTY_ROOT(&fs_info->mapping_tree.rb_root)) {
5939 struct btrfs_chunk_map *map;
5940 struct rb_node *node;
5942 node = rb_first_cached(&fs_info->mapping_tree);
5943 map = rb_entry(node, struct btrfs_chunk_map, rb_node);
5944 rb_erase_cached(&map->rb_node, &fs_info->mapping_tree);
5945 RB_CLEAR_NODE(&map->rb_node);
5946 chunk_map_device_clear_bits(map, CHUNK_ALLOCATED);
5947 /* Once for the tree ref. */
5948 btrfs_free_chunk_map(map);
5949 cond_resched_rwlock_write(&fs_info->mapping_tree_lock);
5951 write_unlock(&fs_info->mapping_tree_lock);
5954 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5956 struct btrfs_chunk_map *map;
5957 enum btrfs_raid_types index;
5960 map = btrfs_get_chunk_map(fs_info, logical, len);
5963 * We could return errors for these cases, but that could get
5964 * ugly and we'd probably do the same thing which is just not do
5965 * anything else and exit, so return 1 so the callers don't try
5966 * to use other copies.
5970 index = btrfs_bg_flags_to_raid_index(map->type);
5972 /* Non-RAID56, use their ncopies from btrfs_raid_array. */
5973 if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5974 ret = btrfs_raid_array[index].ncopies;
5975 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5977 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5979 * There could be two corrupted data stripes, we need
5980 * to loop retry in order to rebuild the correct data.
5982 * Fail a stripe at a time on every retry except the
5983 * stripe under reconstruction.
5985 ret = map->num_stripes;
5986 btrfs_free_chunk_map(map);
5990 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5993 struct btrfs_chunk_map *map;
5994 unsigned long len = fs_info->sectorsize;
5996 if (!btrfs_fs_incompat(fs_info, RAID56))
5999 map = btrfs_get_chunk_map(fs_info, logical, len);
6001 if (!WARN_ON(IS_ERR(map))) {
6002 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
6003 len = btrfs_stripe_nr_to_offset(nr_data_stripes(map));
6004 btrfs_free_chunk_map(map);
6009 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
6011 struct btrfs_chunk_map *map;
6014 if (!btrfs_fs_incompat(fs_info, RAID56))
6017 map = btrfs_get_chunk_map(fs_info, logical, len);
6019 if (!WARN_ON(IS_ERR(map))) {
6020 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
6022 btrfs_free_chunk_map(map);
6027 static int find_live_mirror(struct btrfs_fs_info *fs_info,
6028 struct btrfs_chunk_map *map, int first,
6029 int dev_replace_is_ongoing)
6031 const enum btrfs_read_policy policy = READ_ONCE(fs_info->fs_devices->read_policy);
6034 int preferred_mirror;
6036 struct btrfs_device *srcdev;
6039 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
6041 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
6042 num_stripes = map->sub_stripes;
6044 num_stripes = map->num_stripes;
6048 /* Shouldn't happen, just warn and use pid instead of failing */
6049 btrfs_warn_rl(fs_info, "unknown read_policy type %u, reset to pid",
6051 WRITE_ONCE(fs_info->fs_devices->read_policy, BTRFS_READ_POLICY_PID);
6053 case BTRFS_READ_POLICY_PID:
6054 preferred_mirror = first + (current->pid % num_stripes);
6058 if (dev_replace_is_ongoing &&
6059 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
6060 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
6061 srcdev = fs_info->dev_replace.srcdev;
6066 * try to avoid the drive that is the source drive for a
6067 * dev-replace procedure, only choose it if no other non-missing
6068 * mirror is available
6070 for (tolerance = 0; tolerance < 2; tolerance++) {
6071 if (map->stripes[preferred_mirror].dev->bdev &&
6072 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
6073 return preferred_mirror;
6074 for (i = first; i < first + num_stripes; i++) {
6075 if (map->stripes[i].dev->bdev &&
6076 (tolerance || map->stripes[i].dev != srcdev))
6081 /* we couldn't find one that doesn't fail. Just return something
6082 * and the io error handling code will clean up eventually
6084 return preferred_mirror;
6087 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info,
6091 struct btrfs_io_context *bioc;
6094 /* The size of btrfs_io_context */
6095 sizeof(struct btrfs_io_context) +
6096 /* Plus the variable array for the stripes */
6097 sizeof(struct btrfs_io_stripe) * (total_stripes),
6103 refcount_set(&bioc->refs, 1);
6105 bioc->fs_info = fs_info;
6106 bioc->replace_stripe_src = -1;
6107 bioc->full_stripe_logical = (u64)-1;
6108 bioc->logical = logical;
6113 void btrfs_get_bioc(struct btrfs_io_context *bioc)
6115 WARN_ON(!refcount_read(&bioc->refs));
6116 refcount_inc(&bioc->refs);
6119 void btrfs_put_bioc(struct btrfs_io_context *bioc)
6123 if (refcount_dec_and_test(&bioc->refs))
6128 * Please note that, discard won't be sent to target device of device
6131 struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info,
6132 u64 logical, u64 *length_ret,
6135 struct btrfs_chunk_map *map;
6136 struct btrfs_discard_stripe *stripes;
6137 u64 length = *length_ret;
6142 u64 stripe_end_offset;
6146 u32 sub_stripes = 0;
6147 u32 stripes_per_dev = 0;
6148 u32 remaining_stripes = 0;
6149 u32 last_stripe = 0;
6153 map = btrfs_get_chunk_map(fs_info, logical, length);
6155 return ERR_CAST(map);
6157 /* we don't discard raid56 yet */
6158 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6163 offset = logical - map->start;
6164 length = min_t(u64, map->start + map->chunk_len - logical, length);
6165 *length_ret = length;
6168 * stripe_nr counts the total number of stripes we have to stride
6169 * to get to this block
6171 stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
6173 /* stripe_offset is the offset of this block in its stripe */
6174 stripe_offset = offset - btrfs_stripe_nr_to_offset(stripe_nr);
6176 stripe_nr_end = round_up(offset + length, BTRFS_STRIPE_LEN) >>
6177 BTRFS_STRIPE_LEN_SHIFT;
6178 stripe_cnt = stripe_nr_end - stripe_nr;
6179 stripe_end_offset = btrfs_stripe_nr_to_offset(stripe_nr_end) -
6182 * after this, stripe_nr is the number of stripes on this
6183 * device we have to walk to find the data, and stripe_index is
6184 * the number of our device in the stripe array
6188 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6189 BTRFS_BLOCK_GROUP_RAID10)) {
6190 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6193 sub_stripes = map->sub_stripes;
6195 factor = map->num_stripes / sub_stripes;
6196 *num_stripes = min_t(u64, map->num_stripes,
6197 sub_stripes * stripe_cnt);
6198 stripe_index = stripe_nr % factor;
6199 stripe_nr /= factor;
6200 stripe_index *= sub_stripes;
6202 remaining_stripes = stripe_cnt % factor;
6203 stripes_per_dev = stripe_cnt / factor;
6204 last_stripe = ((stripe_nr_end - 1) % factor) * sub_stripes;
6205 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
6206 BTRFS_BLOCK_GROUP_DUP)) {
6207 *num_stripes = map->num_stripes;
6209 stripe_index = stripe_nr % map->num_stripes;
6210 stripe_nr /= map->num_stripes;
6213 stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS);
6219 for (i = 0; i < *num_stripes; i++) {
6220 stripes[i].physical =
6221 map->stripes[stripe_index].physical +
6222 stripe_offset + btrfs_stripe_nr_to_offset(stripe_nr);
6223 stripes[i].dev = map->stripes[stripe_index].dev;
6225 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6226 BTRFS_BLOCK_GROUP_RAID10)) {
6227 stripes[i].length = btrfs_stripe_nr_to_offset(stripes_per_dev);
6229 if (i / sub_stripes < remaining_stripes)
6230 stripes[i].length += BTRFS_STRIPE_LEN;
6233 * Special for the first stripe and
6236 * |-------|...|-------|
6240 if (i < sub_stripes)
6241 stripes[i].length -= stripe_offset;
6243 if (stripe_index >= last_stripe &&
6244 stripe_index <= (last_stripe +
6246 stripes[i].length -= stripe_end_offset;
6248 if (i == sub_stripes - 1)
6251 stripes[i].length = length;
6255 if (stripe_index == map->num_stripes) {
6261 btrfs_free_chunk_map(map);
6264 btrfs_free_chunk_map(map);
6265 return ERR_PTR(ret);
6268 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)
6270 struct btrfs_block_group *cache;
6273 /* Non zoned filesystem does not use "to_copy" flag */
6274 if (!btrfs_is_zoned(fs_info))
6277 cache = btrfs_lookup_block_group(fs_info, logical);
6279 ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags);
6281 btrfs_put_block_group(cache);
6285 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
6286 struct btrfs_io_context *bioc,
6287 struct btrfs_dev_replace *dev_replace,
6289 int *num_stripes_ret, int *max_errors_ret)
6291 u64 srcdev_devid = dev_replace->srcdev->devid;
6293 * At this stage, num_stripes is still the real number of stripes,
6294 * excluding the duplicated stripes.
6296 int num_stripes = *num_stripes_ret;
6297 int nr_extra_stripes = 0;
6298 int max_errors = *max_errors_ret;
6302 * A block group which has "to_copy" set will eventually be copied by
6303 * the dev-replace process. We can avoid cloning IO here.
6305 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))
6309 * Duplicate the write operations while the dev-replace procedure is
6310 * running. Since the copying of the old disk to the new disk takes
6311 * place at run time while the filesystem is mounted writable, the
6312 * regular write operations to the old disk have to be duplicated to go
6313 * to the new disk as well.
6315 * Note that device->missing is handled by the caller, and that the
6316 * write to the old disk is already set up in the stripes array.
6318 for (i = 0; i < num_stripes; i++) {
6319 struct btrfs_io_stripe *old = &bioc->stripes[i];
6320 struct btrfs_io_stripe *new = &bioc->stripes[num_stripes + nr_extra_stripes];
6322 if (old->dev->devid != srcdev_devid)
6325 new->physical = old->physical;
6326 new->dev = dev_replace->tgtdev;
6327 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK)
6328 bioc->replace_stripe_src = i;
6332 /* We can only have at most 2 extra nr_stripes (for DUP). */
6333 ASSERT(nr_extra_stripes <= 2);
6335 * For GET_READ_MIRRORS, we can only return at most 1 extra stripe for
6337 * If we have 2 extra stripes, only choose the one with smaller physical.
6339 if (op == BTRFS_MAP_GET_READ_MIRRORS && nr_extra_stripes == 2) {
6340 struct btrfs_io_stripe *first = &bioc->stripes[num_stripes];
6341 struct btrfs_io_stripe *second = &bioc->stripes[num_stripes + 1];
6343 /* Only DUP can have two extra stripes. */
6344 ASSERT(bioc->map_type & BTRFS_BLOCK_GROUP_DUP);
6347 * Swap the last stripe stripes and reduce @nr_extra_stripes.
6348 * The extra stripe would still be there, but won't be accessed.
6350 if (first->physical > second->physical) {
6351 swap(second->physical, first->physical);
6352 swap(second->dev, first->dev);
6357 *num_stripes_ret = num_stripes + nr_extra_stripes;
6358 *max_errors_ret = max_errors + nr_extra_stripes;
6359 bioc->replace_nr_stripes = nr_extra_stripes;
6362 static u64 btrfs_max_io_len(struct btrfs_chunk_map *map, u64 offset,
6363 struct btrfs_io_geometry *io_geom)
6366 * Stripe_nr is the stripe where this block falls. stripe_offset is
6367 * the offset of this block in its stripe.
6369 io_geom->stripe_offset = offset & BTRFS_STRIPE_LEN_MASK;
6370 io_geom->stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
6371 ASSERT(io_geom->stripe_offset < U32_MAX);
6373 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6374 unsigned long full_stripe_len =
6375 btrfs_stripe_nr_to_offset(nr_data_stripes(map));
6378 * For full stripe start, we use previously calculated
6379 * @stripe_nr. Align it to nr_data_stripes, then multiply with
6382 * By this we can avoid u64 division completely. And we have
6383 * to go rounddown(), not round_down(), as nr_data_stripes is
6384 * not ensured to be power of 2.
6386 io_geom->raid56_full_stripe_start = btrfs_stripe_nr_to_offset(
6387 rounddown(io_geom->stripe_nr, nr_data_stripes(map)));
6389 ASSERT(io_geom->raid56_full_stripe_start + full_stripe_len > offset);
6390 ASSERT(io_geom->raid56_full_stripe_start <= offset);
6392 * For writes to RAID56, allow to write a full stripe set, but
6393 * no straddling of stripe sets.
6395 if (io_geom->op == BTRFS_MAP_WRITE)
6396 return full_stripe_len - (offset - io_geom->raid56_full_stripe_start);
6400 * For other RAID types and for RAID56 reads, allow a single stripe (on
6403 if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK)
6404 return BTRFS_STRIPE_LEN - io_geom->stripe_offset;
6408 static int set_io_stripe(struct btrfs_fs_info *fs_info, u64 logical,
6409 u64 *length, struct btrfs_io_stripe *dst,
6410 struct btrfs_chunk_map *map,
6411 struct btrfs_io_geometry *io_geom)
6413 dst->dev = map->stripes[io_geom->stripe_index].dev;
6415 if (io_geom->op == BTRFS_MAP_READ &&
6416 btrfs_need_stripe_tree_update(fs_info, map->type))
6417 return btrfs_get_raid_extent_offset(fs_info, logical, length,
6419 io_geom->stripe_index, dst);
6421 dst->physical = map->stripes[io_geom->stripe_index].physical +
6422 io_geom->stripe_offset +
6423 btrfs_stripe_nr_to_offset(io_geom->stripe_nr);
6427 static bool is_single_device_io(struct btrfs_fs_info *fs_info,
6428 const struct btrfs_io_stripe *smap,
6429 const struct btrfs_chunk_map *map,
6430 int num_alloc_stripes,
6431 enum btrfs_map_op op, int mirror_num)
6436 if (num_alloc_stripes != 1)
6439 if (btrfs_need_stripe_tree_update(fs_info, map->type) && op != BTRFS_MAP_READ)
6442 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1)
6448 static void map_blocks_raid0(const struct btrfs_chunk_map *map,
6449 struct btrfs_io_geometry *io_geom)
6451 io_geom->stripe_index = io_geom->stripe_nr % map->num_stripes;
6452 io_geom->stripe_nr /= map->num_stripes;
6453 if (io_geom->op == BTRFS_MAP_READ)
6454 io_geom->mirror_num = 1;
6457 static void map_blocks_raid1(struct btrfs_fs_info *fs_info,
6458 struct btrfs_chunk_map *map,
6459 struct btrfs_io_geometry *io_geom,
6460 bool dev_replace_is_ongoing)
6462 if (io_geom->op != BTRFS_MAP_READ) {
6463 io_geom->num_stripes = map->num_stripes;
6467 if (io_geom->mirror_num) {
6468 io_geom->stripe_index = io_geom->mirror_num - 1;
6472 io_geom->stripe_index = find_live_mirror(fs_info, map, 0,
6473 dev_replace_is_ongoing);
6474 io_geom->mirror_num = io_geom->stripe_index + 1;
6477 static void map_blocks_dup(const struct btrfs_chunk_map *map,
6478 struct btrfs_io_geometry *io_geom)
6480 if (io_geom->op != BTRFS_MAP_READ) {
6481 io_geom->num_stripes = map->num_stripes;
6485 if (io_geom->mirror_num) {
6486 io_geom->stripe_index = io_geom->mirror_num - 1;
6490 io_geom->mirror_num = 1;
6493 static void map_blocks_raid10(struct btrfs_fs_info *fs_info,
6494 struct btrfs_chunk_map *map,
6495 struct btrfs_io_geometry *io_geom,
6496 bool dev_replace_is_ongoing)
6498 u32 factor = map->num_stripes / map->sub_stripes;
6499 int old_stripe_index;
6501 io_geom->stripe_index = (io_geom->stripe_nr % factor) * map->sub_stripes;
6502 io_geom->stripe_nr /= factor;
6504 if (io_geom->op != BTRFS_MAP_READ) {
6505 io_geom->num_stripes = map->sub_stripes;
6509 if (io_geom->mirror_num) {
6510 io_geom->stripe_index += io_geom->mirror_num - 1;
6514 old_stripe_index = io_geom->stripe_index;
6515 io_geom->stripe_index = find_live_mirror(fs_info, map,
6516 io_geom->stripe_index,
6517 dev_replace_is_ongoing);
6518 io_geom->mirror_num = io_geom->stripe_index - old_stripe_index + 1;
6521 static void map_blocks_raid56_write(struct btrfs_chunk_map *map,
6522 struct btrfs_io_geometry *io_geom,
6523 u64 logical, u64 *length)
6525 int data_stripes = nr_data_stripes(map);
6528 * Needs full stripe mapping.
6530 * Push stripe_nr back to the start of the full stripe For those cases
6531 * needing a full stripe, @stripe_nr is the full stripe number.
6533 * Originally we go raid56_full_stripe_start / full_stripe_len, but
6534 * that can be expensive. Here we just divide @stripe_nr with
6537 io_geom->stripe_nr /= data_stripes;
6539 /* RAID[56] write or recovery. Return all stripes */
6540 io_geom->num_stripes = map->num_stripes;
6541 io_geom->max_errors = btrfs_chunk_max_errors(map);
6543 /* Return the length to the full stripe end. */
6544 *length = min(logical + *length,
6545 io_geom->raid56_full_stripe_start + map->start +
6546 btrfs_stripe_nr_to_offset(data_stripes)) -
6548 io_geom->stripe_index = 0;
6549 io_geom->stripe_offset = 0;
6552 static void map_blocks_raid56_read(struct btrfs_chunk_map *map,
6553 struct btrfs_io_geometry *io_geom)
6555 int data_stripes = nr_data_stripes(map);
6557 ASSERT(io_geom->mirror_num <= 1);
6558 /* Just grab the data stripe directly. */
6559 io_geom->stripe_index = io_geom->stripe_nr % data_stripes;
6560 io_geom->stripe_nr /= data_stripes;
6562 /* We distribute the parity blocks across stripes. */
6563 io_geom->stripe_index =
6564 (io_geom->stripe_nr + io_geom->stripe_index) % map->num_stripes;
6566 if (io_geom->op == BTRFS_MAP_READ && io_geom->mirror_num < 1)
6567 io_geom->mirror_num = 1;
6570 static void map_blocks_single(const struct btrfs_chunk_map *map,
6571 struct btrfs_io_geometry *io_geom)
6573 io_geom->stripe_index = io_geom->stripe_nr % map->num_stripes;
6574 io_geom->stripe_nr /= map->num_stripes;
6575 io_geom->mirror_num = io_geom->stripe_index + 1;
6579 * Map one logical range to one or more physical ranges.
6581 * @length: (Mandatory) mapped length of this run.
6582 * One logical range can be split into different segments
6583 * due to factors like zones and RAID0/5/6/10 stripe
6586 * @bioc_ret: (Mandatory) returned btrfs_io_context structure.
6587 * which has one or more physical ranges (btrfs_io_stripe)
6589 * Caller should call btrfs_put_bioc() to free it after use.
6591 * @smap: (Optional) single physical range optimization.
6592 * If the map request can be fulfilled by one single
6593 * physical range, and this is parameter is not NULL,
6594 * then @bioc_ret would be NULL, and @smap would be
6597 * @mirror_num_ret: (Mandatory) returned mirror number if the original
6600 * Mirror number 0 means to choose any live mirrors.
6602 * For non-RAID56 profiles, non-zero mirror_num means
6603 * the Nth mirror. (e.g. mirror_num 1 means the first
6606 * For RAID56 profile, mirror 1 means rebuild from P and
6607 * the remaining data stripes.
6609 * For RAID6 profile, mirror > 2 means mark another
6610 * data/P stripe error and rebuild from the remaining
6613 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6614 u64 logical, u64 *length,
6615 struct btrfs_io_context **bioc_ret,
6616 struct btrfs_io_stripe *smap, int *mirror_num_ret)
6618 struct btrfs_chunk_map *map;
6619 struct btrfs_io_geometry io_geom = { 0 };
6624 struct btrfs_io_context *bioc = NULL;
6625 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6626 int dev_replace_is_ongoing = 0;
6627 u16 num_alloc_stripes;
6632 io_geom.mirror_num = (mirror_num_ret ? *mirror_num_ret : 0);
6633 io_geom.num_stripes = 1;
6634 io_geom.stripe_index = 0;
6637 num_copies = btrfs_num_copies(fs_info, logical, fs_info->sectorsize);
6638 if (io_geom.mirror_num > num_copies)
6641 map = btrfs_get_chunk_map(fs_info, logical, *length);
6643 return PTR_ERR(map);
6645 map_offset = logical - map->start;
6646 io_geom.raid56_full_stripe_start = (u64)-1;
6647 max_len = btrfs_max_io_len(map, map_offset, &io_geom);
6648 *length = min_t(u64, map->chunk_len - map_offset, max_len);
6650 down_read(&dev_replace->rwsem);
6651 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6653 * Hold the semaphore for read during the whole operation, write is
6654 * requested at commit time but must wait.
6656 if (!dev_replace_is_ongoing)
6657 up_read(&dev_replace->rwsem);
6659 switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
6660 case BTRFS_BLOCK_GROUP_RAID0:
6661 map_blocks_raid0(map, &io_geom);
6663 case BTRFS_BLOCK_GROUP_RAID1:
6664 case BTRFS_BLOCK_GROUP_RAID1C3:
6665 case BTRFS_BLOCK_GROUP_RAID1C4:
6666 map_blocks_raid1(fs_info, map, &io_geom, dev_replace_is_ongoing);
6668 case BTRFS_BLOCK_GROUP_DUP:
6669 map_blocks_dup(map, &io_geom);
6671 case BTRFS_BLOCK_GROUP_RAID10:
6672 map_blocks_raid10(fs_info, map, &io_geom, dev_replace_is_ongoing);
6674 case BTRFS_BLOCK_GROUP_RAID5:
6675 case BTRFS_BLOCK_GROUP_RAID6:
6676 if (op != BTRFS_MAP_READ || io_geom.mirror_num > 1)
6677 map_blocks_raid56_write(map, &io_geom, logical, length);
6679 map_blocks_raid56_read(map, &io_geom);
6683 * After this, stripe_nr is the number of stripes on this
6684 * device we have to walk to find the data, and stripe_index is
6685 * the number of our device in the stripe array
6687 map_blocks_single(map, &io_geom);
6690 if (io_geom.stripe_index >= map->num_stripes) {
6692 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6693 io_geom.stripe_index, map->num_stripes);
6698 num_alloc_stripes = io_geom.num_stripes;
6699 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6700 op != BTRFS_MAP_READ)
6702 * For replace case, we need to add extra stripes for extra
6703 * duplicated stripes.
6705 * For both WRITE and GET_READ_MIRRORS, we may have at most
6706 * 2 more stripes (DUP types, otherwise 1).
6708 num_alloc_stripes += 2;
6711 * If this I/O maps to a single device, try to return the device and
6712 * physical block information on the stack instead of allocating an
6713 * I/O context structure.
6715 if (is_single_device_io(fs_info, smap, map, num_alloc_stripes, op,
6716 io_geom.mirror_num)) {
6717 ret = set_io_stripe(fs_info, logical, length, smap, map, &io_geom);
6719 *mirror_num_ret = io_geom.mirror_num;
6724 bioc = alloc_btrfs_io_context(fs_info, logical, num_alloc_stripes);
6729 bioc->map_type = map->type;
6732 * For RAID56 full map, we need to make sure the stripes[] follows the
6733 * rule that data stripes are all ordered, then followed with P and Q
6736 * It's still mostly the same as other profiles, just with extra rotation.
6738 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
6739 (op != BTRFS_MAP_READ || io_geom.mirror_num > 1)) {
6741 * For RAID56 @stripe_nr is already the number of full stripes
6742 * before us, which is also the rotation value (needs to modulo
6743 * with num_stripes).
6745 * In this case, we just add @stripe_nr with @i, then do the
6746 * modulo, to reduce one modulo call.
6748 bioc->full_stripe_logical = map->start +
6749 btrfs_stripe_nr_to_offset(io_geom.stripe_nr *
6750 nr_data_stripes(map));
6751 for (int i = 0; i < io_geom.num_stripes; i++) {
6752 struct btrfs_io_stripe *dst = &bioc->stripes[i];
6755 stripe_index = (i + io_geom.stripe_nr) % io_geom.num_stripes;
6756 dst->dev = map->stripes[stripe_index].dev;
6758 map->stripes[stripe_index].physical +
6759 io_geom.stripe_offset +
6760 btrfs_stripe_nr_to_offset(io_geom.stripe_nr);
6764 * For all other non-RAID56 profiles, just copy the target
6765 * stripe into the bioc.
6767 for (i = 0; i < io_geom.num_stripes; i++) {
6768 ret = set_io_stripe(fs_info, logical, length,
6769 &bioc->stripes[i], map, &io_geom);
6772 io_geom.stripe_index++;
6778 btrfs_put_bioc(bioc);
6782 if (op != BTRFS_MAP_READ)
6783 io_geom.max_errors = btrfs_chunk_max_errors(map);
6785 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6786 op != BTRFS_MAP_READ) {
6787 handle_ops_on_dev_replace(op, bioc, dev_replace, logical,
6788 &io_geom.num_stripes, &io_geom.max_errors);
6792 bioc->num_stripes = io_geom.num_stripes;
6793 bioc->max_errors = io_geom.max_errors;
6794 bioc->mirror_num = io_geom.mirror_num;
6797 if (dev_replace_is_ongoing) {
6798 lockdep_assert_held(&dev_replace->rwsem);
6799 /* Unlock and let waiting writers proceed */
6800 up_read(&dev_replace->rwsem);
6802 btrfs_free_chunk_map(map);
6806 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args,
6807 const struct btrfs_fs_devices *fs_devices)
6809 if (args->fsid == NULL)
6811 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0)
6816 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args,
6817 const struct btrfs_device *device)
6819 if (args->missing) {
6820 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) &&
6826 if (device->devid != args->devid)
6828 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0)
6834 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6837 * If devid and uuid are both specified, the match must be exact, otherwise
6838 * only devid is used.
6840 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,
6841 const struct btrfs_dev_lookup_args *args)
6843 struct btrfs_device *device;
6844 struct btrfs_fs_devices *seed_devs;
6846 if (dev_args_match_fs_devices(args, fs_devices)) {
6847 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6848 if (dev_args_match_device(args, device))
6853 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6854 if (!dev_args_match_fs_devices(args, seed_devs))
6856 list_for_each_entry(device, &seed_devs->devices, dev_list) {
6857 if (dev_args_match_device(args, device))
6865 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6866 u64 devid, u8 *dev_uuid)
6868 struct btrfs_device *device;
6869 unsigned int nofs_flag;
6872 * We call this under the chunk_mutex, so we want to use NOFS for this
6873 * allocation, however we don't want to change btrfs_alloc_device() to
6874 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6878 nofs_flag = memalloc_nofs_save();
6879 device = btrfs_alloc_device(NULL, &devid, dev_uuid, NULL);
6880 memalloc_nofs_restore(nofs_flag);
6884 list_add(&device->dev_list, &fs_devices->devices);
6885 device->fs_devices = fs_devices;
6886 fs_devices->num_devices++;
6888 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6889 fs_devices->missing_devices++;
6895 * Allocate new device struct, set up devid and UUID.
6897 * @fs_info: used only for generating a new devid, can be NULL if
6898 * devid is provided (i.e. @devid != NULL).
6899 * @devid: a pointer to devid for this device. If NULL a new devid
6901 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6903 * @path: a pointer to device path if available, NULL otherwise.
6905 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6906 * on error. Returned struct is not linked onto any lists and must be
6907 * destroyed with btrfs_free_device.
6909 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6910 const u64 *devid, const u8 *uuid,
6913 struct btrfs_device *dev;
6916 if (WARN_ON(!devid && !fs_info))
6917 return ERR_PTR(-EINVAL);
6919 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
6921 return ERR_PTR(-ENOMEM);
6923 INIT_LIST_HEAD(&dev->dev_list);
6924 INIT_LIST_HEAD(&dev->dev_alloc_list);
6925 INIT_LIST_HEAD(&dev->post_commit_list);
6927 atomic_set(&dev->dev_stats_ccnt, 0);
6928 btrfs_device_data_ordered_init(dev);
6929 extent_io_tree_init(fs_info, &dev->alloc_state, IO_TREE_DEVICE_ALLOC_STATE);
6936 ret = find_next_devid(fs_info, &tmp);
6938 btrfs_free_device(dev);
6939 return ERR_PTR(ret);
6945 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6947 generate_random_uuid(dev->uuid);
6950 struct rcu_string *name;
6952 name = rcu_string_strdup(path, GFP_KERNEL);
6954 btrfs_free_device(dev);
6955 return ERR_PTR(-ENOMEM);
6957 rcu_assign_pointer(dev->name, name);
6963 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6964 u64 devid, u8 *uuid, bool error)
6967 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6970 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6974 u64 btrfs_calc_stripe_length(const struct btrfs_chunk_map *map)
6976 const int data_stripes = calc_data_stripes(map->type, map->num_stripes);
6978 return div_u64(map->chunk_len, data_stripes);
6981 #if BITS_PER_LONG == 32
6983 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
6984 * can't be accessed on 32bit systems.
6986 * This function do mount time check to reject the fs if it already has
6987 * metadata chunk beyond that limit.
6989 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6990 u64 logical, u64 length, u64 type)
6992 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6995 if (logical + length < MAX_LFS_FILESIZE)
6998 btrfs_err_32bit_limit(fs_info);
7003 * This is to give early warning for any metadata chunk reaching
7004 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
7005 * Although we can still access the metadata, it's not going to be possible
7006 * once the limit is reached.
7008 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
7009 u64 logical, u64 length, u64 type)
7011 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
7014 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)
7017 btrfs_warn_32bit_limit(fs_info);
7021 static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info,
7022 u64 devid, u8 *uuid)
7024 struct btrfs_device *dev;
7026 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7027 btrfs_report_missing_device(fs_info, devid, uuid, true);
7028 return ERR_PTR(-ENOENT);
7031 dev = add_missing_dev(fs_info->fs_devices, devid, uuid);
7033 btrfs_err(fs_info, "failed to init missing device %llu: %ld",
7034 devid, PTR_ERR(dev));
7037 btrfs_report_missing_device(fs_info, devid, uuid, false);
7042 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
7043 struct btrfs_chunk *chunk)
7045 BTRFS_DEV_LOOKUP_ARGS(args);
7046 struct btrfs_fs_info *fs_info = leaf->fs_info;
7047 struct btrfs_chunk_map *map;
7052 u8 uuid[BTRFS_UUID_SIZE];
7058 logical = key->offset;
7059 length = btrfs_chunk_length(leaf, chunk);
7060 type = btrfs_chunk_type(leaf, chunk);
7061 index = btrfs_bg_flags_to_raid_index(type);
7062 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
7064 #if BITS_PER_LONG == 32
7065 ret = check_32bit_meta_chunk(fs_info, logical, length, type);
7068 warn_32bit_meta_chunk(fs_info, logical, length, type);
7072 * Only need to verify chunk item if we're reading from sys chunk array,
7073 * as chunk item in tree block is already verified by tree-checker.
7075 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
7076 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
7081 map = btrfs_find_chunk_map(fs_info, logical, 1);
7083 /* already mapped? */
7084 if (map && map->start <= logical && map->start + map->chunk_len > logical) {
7085 btrfs_free_chunk_map(map);
7088 btrfs_free_chunk_map(map);
7091 map = btrfs_alloc_chunk_map(num_stripes, GFP_NOFS);
7095 map->start = logical;
7096 map->chunk_len = length;
7097 map->num_stripes = num_stripes;
7098 map->io_width = btrfs_chunk_io_width(leaf, chunk);
7099 map->io_align = btrfs_chunk_io_align(leaf, chunk);
7102 * We can't use the sub_stripes value, as for profiles other than
7103 * RAID10, they may have 0 as sub_stripes for filesystems created by
7104 * older mkfs (<v5.4).
7105 * In that case, it can cause divide-by-zero errors later.
7106 * Since currently sub_stripes is fixed for each profile, let's
7107 * use the trusted value instead.
7109 map->sub_stripes = btrfs_raid_array[index].sub_stripes;
7110 map->verified_stripes = 0;
7111 map->stripe_size = btrfs_calc_stripe_length(map);
7112 for (i = 0; i < num_stripes; i++) {
7113 map->stripes[i].physical =
7114 btrfs_stripe_offset_nr(leaf, chunk, i);
7115 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
7117 read_extent_buffer(leaf, uuid, (unsigned long)
7118 btrfs_stripe_dev_uuid_nr(chunk, i),
7121 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args);
7122 if (!map->stripes[i].dev) {
7123 map->stripes[i].dev = handle_missing_device(fs_info,
7125 if (IS_ERR(map->stripes[i].dev)) {
7126 ret = PTR_ERR(map->stripes[i].dev);
7127 btrfs_free_chunk_map(map);
7132 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
7133 &(map->stripes[i].dev->dev_state));
7136 ret = btrfs_add_chunk_map(fs_info, map);
7139 "failed to add chunk map, start=%llu len=%llu: %d",
7140 map->start, map->chunk_len, ret);
7146 static void fill_device_from_item(struct extent_buffer *leaf,
7147 struct btrfs_dev_item *dev_item,
7148 struct btrfs_device *device)
7152 device->devid = btrfs_device_id(leaf, dev_item);
7153 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
7154 device->total_bytes = device->disk_total_bytes;
7155 device->commit_total_bytes = device->disk_total_bytes;
7156 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
7157 device->commit_bytes_used = device->bytes_used;
7158 device->type = btrfs_device_type(leaf, dev_item);
7159 device->io_align = btrfs_device_io_align(leaf, dev_item);
7160 device->io_width = btrfs_device_io_width(leaf, dev_item);
7161 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
7162 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
7163 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
7165 ptr = btrfs_device_uuid(dev_item);
7166 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
7169 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
7172 struct btrfs_fs_devices *fs_devices;
7175 lockdep_assert_held(&uuid_mutex);
7178 /* This will match only for multi-device seed fs */
7179 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
7180 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
7184 fs_devices = find_fsid(fsid, NULL);
7186 if (!btrfs_test_opt(fs_info, DEGRADED))
7187 return ERR_PTR(-ENOENT);
7189 fs_devices = alloc_fs_devices(fsid);
7190 if (IS_ERR(fs_devices))
7193 fs_devices->seeding = true;
7194 fs_devices->opened = 1;
7199 * Upon first call for a seed fs fsid, just create a private copy of the
7200 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
7202 fs_devices = clone_fs_devices(fs_devices);
7203 if (IS_ERR(fs_devices))
7206 ret = open_fs_devices(fs_devices, BLK_OPEN_READ, fs_info->bdev_holder);
7208 free_fs_devices(fs_devices);
7209 return ERR_PTR(ret);
7212 if (!fs_devices->seeding) {
7213 close_fs_devices(fs_devices);
7214 free_fs_devices(fs_devices);
7215 return ERR_PTR(-EINVAL);
7218 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
7223 static int read_one_dev(struct extent_buffer *leaf,
7224 struct btrfs_dev_item *dev_item)
7226 BTRFS_DEV_LOOKUP_ARGS(args);
7227 struct btrfs_fs_info *fs_info = leaf->fs_info;
7228 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7229 struct btrfs_device *device;
7232 u8 fs_uuid[BTRFS_FSID_SIZE];
7233 u8 dev_uuid[BTRFS_UUID_SIZE];
7235 devid = btrfs_device_id(leaf, dev_item);
7237 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
7239 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
7241 args.uuid = dev_uuid;
7242 args.fsid = fs_uuid;
7244 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
7245 fs_devices = open_seed_devices(fs_info, fs_uuid);
7246 if (IS_ERR(fs_devices))
7247 return PTR_ERR(fs_devices);
7250 device = btrfs_find_device(fs_info->fs_devices, &args);
7252 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7253 btrfs_report_missing_device(fs_info, devid,
7258 device = add_missing_dev(fs_devices, devid, dev_uuid);
7259 if (IS_ERR(device)) {
7261 "failed to add missing dev %llu: %ld",
7262 devid, PTR_ERR(device));
7263 return PTR_ERR(device);
7265 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
7267 if (!device->bdev) {
7268 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7269 btrfs_report_missing_device(fs_info,
7270 devid, dev_uuid, true);
7273 btrfs_report_missing_device(fs_info, devid,
7277 if (!device->bdev &&
7278 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
7280 * this happens when a device that was properly setup
7281 * in the device info lists suddenly goes bad.
7282 * device->bdev is NULL, and so we have to set
7283 * device->missing to one here
7285 device->fs_devices->missing_devices++;
7286 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
7289 /* Move the device to its own fs_devices */
7290 if (device->fs_devices != fs_devices) {
7291 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
7292 &device->dev_state));
7294 list_move(&device->dev_list, &fs_devices->devices);
7295 device->fs_devices->num_devices--;
7296 fs_devices->num_devices++;
7298 device->fs_devices->missing_devices--;
7299 fs_devices->missing_devices++;
7301 device->fs_devices = fs_devices;
7305 if (device->fs_devices != fs_info->fs_devices) {
7306 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
7307 if (device->generation !=
7308 btrfs_device_generation(leaf, dev_item))
7312 fill_device_from_item(leaf, dev_item, device);
7314 u64 max_total_bytes = bdev_nr_bytes(device->bdev);
7316 if (device->total_bytes > max_total_bytes) {
7318 "device total_bytes should be at most %llu but found %llu",
7319 max_total_bytes, device->total_bytes);
7323 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7324 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7325 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7326 device->fs_devices->total_rw_bytes += device->total_bytes;
7327 atomic64_add(device->total_bytes - device->bytes_used,
7328 &fs_info->free_chunk_space);
7334 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7336 struct btrfs_super_block *super_copy = fs_info->super_copy;
7337 struct extent_buffer *sb;
7338 struct btrfs_disk_key *disk_key;
7339 struct btrfs_chunk *chunk;
7341 unsigned long sb_array_offset;
7348 struct btrfs_key key;
7350 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7353 * We allocated a dummy extent, just to use extent buffer accessors.
7354 * There will be unused space after BTRFS_SUPER_INFO_SIZE, but
7355 * that's fine, we will not go beyond system chunk array anyway.
7357 sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET);
7360 set_extent_buffer_uptodate(sb);
7362 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7363 array_size = btrfs_super_sys_array_size(super_copy);
7365 array_ptr = super_copy->sys_chunk_array;
7366 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7369 while (cur_offset < array_size) {
7370 disk_key = (struct btrfs_disk_key *)array_ptr;
7371 len = sizeof(*disk_key);
7372 if (cur_offset + len > array_size)
7373 goto out_short_read;
7375 btrfs_disk_key_to_cpu(&key, disk_key);
7378 sb_array_offset += len;
7381 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7383 "unexpected item type %u in sys_array at offset %u",
7384 (u32)key.type, cur_offset);
7389 chunk = (struct btrfs_chunk *)sb_array_offset;
7391 * At least one btrfs_chunk with one stripe must be present,
7392 * exact stripe count check comes afterwards
7394 len = btrfs_chunk_item_size(1);
7395 if (cur_offset + len > array_size)
7396 goto out_short_read;
7398 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7401 "invalid number of stripes %u in sys_array at offset %u",
7402 num_stripes, cur_offset);
7407 type = btrfs_chunk_type(sb, chunk);
7408 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7410 "invalid chunk type %llu in sys_array at offset %u",
7416 len = btrfs_chunk_item_size(num_stripes);
7417 if (cur_offset + len > array_size)
7418 goto out_short_read;
7420 ret = read_one_chunk(&key, sb, chunk);
7425 sb_array_offset += len;
7428 clear_extent_buffer_uptodate(sb);
7429 free_extent_buffer_stale(sb);
7433 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7435 clear_extent_buffer_uptodate(sb);
7436 free_extent_buffer_stale(sb);
7441 * Check if all chunks in the fs are OK for read-write degraded mount
7443 * If the @failing_dev is specified, it's accounted as missing.
7445 * Return true if all chunks meet the minimal RW mount requirements.
7446 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7448 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7449 struct btrfs_device *failing_dev)
7451 struct btrfs_chunk_map *map;
7455 map = btrfs_find_chunk_map(fs_info, 0, U64_MAX);
7456 /* No chunk at all? Return false anyway */
7467 btrfs_get_num_tolerated_disk_barrier_failures(
7469 for (i = 0; i < map->num_stripes; i++) {
7470 struct btrfs_device *dev = map->stripes[i].dev;
7472 if (!dev || !dev->bdev ||
7473 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7474 dev->last_flush_error)
7476 else if (failing_dev && failing_dev == dev)
7479 if (missing > max_tolerated) {
7482 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7483 map->start, missing, max_tolerated);
7484 btrfs_free_chunk_map(map);
7488 next_start = map->start + map->chunk_len;
7489 btrfs_free_chunk_map(map);
7491 map = btrfs_find_chunk_map(fs_info, next_start, U64_MAX - next_start);
7497 static void readahead_tree_node_children(struct extent_buffer *node)
7500 const int nr_items = btrfs_header_nritems(node);
7502 for (i = 0; i < nr_items; i++)
7503 btrfs_readahead_node_child(node, i);
7506 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7508 struct btrfs_root *root = fs_info->chunk_root;
7509 struct btrfs_path *path;
7510 struct extent_buffer *leaf;
7511 struct btrfs_key key;
7512 struct btrfs_key found_key;
7517 u64 last_ra_node = 0;
7519 path = btrfs_alloc_path();
7524 * uuid_mutex is needed only if we are mounting a sprout FS
7525 * otherwise we don't need it.
7527 mutex_lock(&uuid_mutex);
7530 * It is possible for mount and umount to race in such a way that
7531 * we execute this code path, but open_fs_devices failed to clear
7532 * total_rw_bytes. We certainly want it cleared before reading the
7533 * device items, so clear it here.
7535 fs_info->fs_devices->total_rw_bytes = 0;
7538 * Lockdep complains about possible circular locking dependency between
7539 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores
7540 * used for freeze procection of a fs (struct super_block.s_writers),
7541 * which we take when starting a transaction, and extent buffers of the
7542 * chunk tree if we call read_one_dev() while holding a lock on an
7543 * extent buffer of the chunk tree. Since we are mounting the filesystem
7544 * and at this point there can't be any concurrent task modifying the
7545 * chunk tree, to keep it simple, just skip locking on the chunk tree.
7547 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
7548 path->skip_locking = 1;
7551 * Read all device items, and then all the chunk items. All
7552 * device items are found before any chunk item (their object id
7553 * is smaller than the lowest possible object id for a chunk
7554 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7556 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7559 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) {
7560 struct extent_buffer *node = path->nodes[1];
7562 leaf = path->nodes[0];
7563 slot = path->slots[0];
7566 if (last_ra_node != node->start) {
7567 readahead_tree_node_children(node);
7568 last_ra_node = node->start;
7571 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7572 struct btrfs_dev_item *dev_item;
7573 dev_item = btrfs_item_ptr(leaf, slot,
7574 struct btrfs_dev_item);
7575 ret = read_one_dev(leaf, dev_item);
7579 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7580 struct btrfs_chunk *chunk;
7583 * We are only called at mount time, so no need to take
7584 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7585 * we always lock first fs_info->chunk_mutex before
7586 * acquiring any locks on the chunk tree. This is a
7587 * requirement for chunk allocation, see the comment on
7588 * top of btrfs_chunk_alloc() for details.
7590 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7591 ret = read_one_chunk(&found_key, leaf, chunk);
7596 /* Catch error found during iteration */
7603 * After loading chunk tree, we've got all device information,
7604 * do another round of validation checks.
7606 if (total_dev != fs_info->fs_devices->total_devices) {
7608 "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit",
7609 btrfs_super_num_devices(fs_info->super_copy),
7611 fs_info->fs_devices->total_devices = total_dev;
7612 btrfs_set_super_num_devices(fs_info->super_copy, total_dev);
7614 if (btrfs_super_total_bytes(fs_info->super_copy) <
7615 fs_info->fs_devices->total_rw_bytes) {
7617 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7618 btrfs_super_total_bytes(fs_info->super_copy),
7619 fs_info->fs_devices->total_rw_bytes);
7625 mutex_unlock(&uuid_mutex);
7627 btrfs_free_path(path);
7631 int btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7633 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7634 struct btrfs_device *device;
7637 fs_devices->fs_info = fs_info;
7639 mutex_lock(&fs_devices->device_list_mutex);
7640 list_for_each_entry(device, &fs_devices->devices, dev_list)
7641 device->fs_info = fs_info;
7643 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7644 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7645 device->fs_info = fs_info;
7646 ret = btrfs_get_dev_zone_info(device, false);
7651 seed_devs->fs_info = fs_info;
7653 mutex_unlock(&fs_devices->device_list_mutex);
7658 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7659 const struct btrfs_dev_stats_item *ptr,
7664 read_extent_buffer(eb, &val,
7665 offsetof(struct btrfs_dev_stats_item, values) +
7666 ((unsigned long)ptr) + (index * sizeof(u64)),
7671 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7672 struct btrfs_dev_stats_item *ptr,
7675 write_extent_buffer(eb, &val,
7676 offsetof(struct btrfs_dev_stats_item, values) +
7677 ((unsigned long)ptr) + (index * sizeof(u64)),
7681 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7682 struct btrfs_path *path)
7684 struct btrfs_dev_stats_item *ptr;
7685 struct extent_buffer *eb;
7686 struct btrfs_key key;
7690 if (!device->fs_info->dev_root)
7693 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7694 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7695 key.offset = device->devid;
7696 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7698 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7699 btrfs_dev_stat_set(device, i, 0);
7700 device->dev_stats_valid = 1;
7701 btrfs_release_path(path);
7702 return ret < 0 ? ret : 0;
7704 slot = path->slots[0];
7705 eb = path->nodes[0];
7706 item_size = btrfs_item_size(eb, slot);
7708 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7710 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7711 if (item_size >= (1 + i) * sizeof(__le64))
7712 btrfs_dev_stat_set(device, i,
7713 btrfs_dev_stats_value(eb, ptr, i));
7715 btrfs_dev_stat_set(device, i, 0);
7718 device->dev_stats_valid = 1;
7719 btrfs_dev_stat_print_on_load(device);
7720 btrfs_release_path(path);
7725 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7727 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7728 struct btrfs_device *device;
7729 struct btrfs_path *path = NULL;
7732 path = btrfs_alloc_path();
7736 mutex_lock(&fs_devices->device_list_mutex);
7737 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7738 ret = btrfs_device_init_dev_stats(device, path);
7742 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7743 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7744 ret = btrfs_device_init_dev_stats(device, path);
7750 mutex_unlock(&fs_devices->device_list_mutex);
7752 btrfs_free_path(path);
7756 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7757 struct btrfs_device *device)
7759 struct btrfs_fs_info *fs_info = trans->fs_info;
7760 struct btrfs_root *dev_root = fs_info->dev_root;
7761 struct btrfs_path *path;
7762 struct btrfs_key key;
7763 struct extent_buffer *eb;
7764 struct btrfs_dev_stats_item *ptr;
7768 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7769 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7770 key.offset = device->devid;
7772 path = btrfs_alloc_path();
7775 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7777 btrfs_warn_in_rcu(fs_info,
7778 "error %d while searching for dev_stats item for device %s",
7779 ret, btrfs_dev_name(device));
7784 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7785 /* need to delete old one and insert a new one */
7786 ret = btrfs_del_item(trans, dev_root, path);
7788 btrfs_warn_in_rcu(fs_info,
7789 "delete too small dev_stats item for device %s failed %d",
7790 btrfs_dev_name(device), ret);
7797 /* need to insert a new item */
7798 btrfs_release_path(path);
7799 ret = btrfs_insert_empty_item(trans, dev_root, path,
7800 &key, sizeof(*ptr));
7802 btrfs_warn_in_rcu(fs_info,
7803 "insert dev_stats item for device %s failed %d",
7804 btrfs_dev_name(device), ret);
7809 eb = path->nodes[0];
7810 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7811 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7812 btrfs_set_dev_stats_value(eb, ptr, i,
7813 btrfs_dev_stat_read(device, i));
7814 btrfs_mark_buffer_dirty(trans, eb);
7817 btrfs_free_path(path);
7822 * called from commit_transaction. Writes all changed device stats to disk.
7824 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7826 struct btrfs_fs_info *fs_info = trans->fs_info;
7827 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7828 struct btrfs_device *device;
7832 mutex_lock(&fs_devices->device_list_mutex);
7833 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7834 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7835 if (!device->dev_stats_valid || stats_cnt == 0)
7840 * There is a LOAD-LOAD control dependency between the value of
7841 * dev_stats_ccnt and updating the on-disk values which requires
7842 * reading the in-memory counters. Such control dependencies
7843 * require explicit read memory barriers.
7845 * This memory barriers pairs with smp_mb__before_atomic in
7846 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7847 * barrier implied by atomic_xchg in
7848 * btrfs_dev_stats_read_and_reset
7852 ret = update_dev_stat_item(trans, device);
7854 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7856 mutex_unlock(&fs_devices->device_list_mutex);
7861 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7863 btrfs_dev_stat_inc(dev, index);
7865 if (!dev->dev_stats_valid)
7867 btrfs_err_rl_in_rcu(dev->fs_info,
7868 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7869 btrfs_dev_name(dev),
7870 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7871 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7872 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7873 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7874 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7877 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7881 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7882 if (btrfs_dev_stat_read(dev, i) != 0)
7884 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7885 return; /* all values == 0, suppress message */
7887 btrfs_info_in_rcu(dev->fs_info,
7888 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7889 btrfs_dev_name(dev),
7890 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7891 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7892 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7893 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7894 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7897 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7898 struct btrfs_ioctl_get_dev_stats *stats)
7900 BTRFS_DEV_LOOKUP_ARGS(args);
7901 struct btrfs_device *dev;
7902 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7905 mutex_lock(&fs_devices->device_list_mutex);
7906 args.devid = stats->devid;
7907 dev = btrfs_find_device(fs_info->fs_devices, &args);
7908 mutex_unlock(&fs_devices->device_list_mutex);
7911 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7913 } else if (!dev->dev_stats_valid) {
7914 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7916 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7917 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7918 if (stats->nr_items > i)
7920 btrfs_dev_stat_read_and_reset(dev, i);
7922 btrfs_dev_stat_set(dev, i, 0);
7924 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7925 current->comm, task_pid_nr(current));
7927 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7928 if (stats->nr_items > i)
7929 stats->values[i] = btrfs_dev_stat_read(dev, i);
7931 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7932 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7937 * Update the size and bytes used for each device where it changed. This is
7938 * delayed since we would otherwise get errors while writing out the
7941 * Must be invoked during transaction commit.
7943 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7945 struct btrfs_device *curr, *next;
7947 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7949 if (list_empty(&trans->dev_update_list))
7953 * We don't need the device_list_mutex here. This list is owned by the
7954 * transaction and the transaction must complete before the device is
7957 mutex_lock(&trans->fs_info->chunk_mutex);
7958 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7960 list_del_init(&curr->post_commit_list);
7961 curr->commit_total_bytes = curr->disk_total_bytes;
7962 curr->commit_bytes_used = curr->bytes_used;
7964 mutex_unlock(&trans->fs_info->chunk_mutex);
7968 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7970 int btrfs_bg_type_to_factor(u64 flags)
7972 const int index = btrfs_bg_flags_to_raid_index(flags);
7974 return btrfs_raid_array[index].ncopies;
7979 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7980 u64 chunk_offset, u64 devid,
7981 u64 physical_offset, u64 physical_len)
7983 struct btrfs_dev_lookup_args args = { .devid = devid };
7984 struct btrfs_chunk_map *map;
7985 struct btrfs_device *dev;
7991 map = btrfs_find_chunk_map(fs_info, chunk_offset, 1);
7994 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7995 physical_offset, devid);
8000 stripe_len = btrfs_calc_stripe_length(map);
8001 if (physical_len != stripe_len) {
8003 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
8004 physical_offset, devid, map->start, physical_len,
8011 * Very old mkfs.btrfs (before v4.1) will not respect the reserved
8012 * space. Although kernel can handle it without problem, better to warn
8015 if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED)
8017 "devid %llu physical %llu len %llu inside the reserved space",
8018 devid, physical_offset, physical_len);
8020 for (i = 0; i < map->num_stripes; i++) {
8021 if (map->stripes[i].dev->devid == devid &&
8022 map->stripes[i].physical == physical_offset) {
8024 if (map->verified_stripes >= map->num_stripes) {
8026 "too many dev extents for chunk %llu found",
8031 map->verified_stripes++;
8037 "dev extent physical offset %llu devid %llu has no corresponding chunk",
8038 physical_offset, devid);
8042 /* Make sure no dev extent is beyond device boundary */
8043 dev = btrfs_find_device(fs_info->fs_devices, &args);
8045 btrfs_err(fs_info, "failed to find devid %llu", devid);
8050 if (physical_offset + physical_len > dev->disk_total_bytes) {
8052 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
8053 devid, physical_offset, physical_len,
8054 dev->disk_total_bytes);
8059 if (dev->zone_info) {
8060 u64 zone_size = dev->zone_info->zone_size;
8062 if (!IS_ALIGNED(physical_offset, zone_size) ||
8063 !IS_ALIGNED(physical_len, zone_size)) {
8065 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
8066 devid, physical_offset, physical_len);
8073 btrfs_free_chunk_map(map);
8077 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
8079 struct rb_node *node;
8082 read_lock(&fs_info->mapping_tree_lock);
8083 for (node = rb_first_cached(&fs_info->mapping_tree); node; node = rb_next(node)) {
8084 struct btrfs_chunk_map *map;
8086 map = rb_entry(node, struct btrfs_chunk_map, rb_node);
8087 if (map->num_stripes != map->verified_stripes) {
8089 "chunk %llu has missing dev extent, have %d expect %d",
8090 map->start, map->verified_stripes, map->num_stripes);
8096 read_unlock(&fs_info->mapping_tree_lock);
8101 * Ensure that all dev extents are mapped to correct chunk, otherwise
8102 * later chunk allocation/free would cause unexpected behavior.
8104 * NOTE: This will iterate through the whole device tree, which should be of
8105 * the same size level as the chunk tree. This slightly increases mount time.
8107 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
8109 struct btrfs_path *path;
8110 struct btrfs_root *root = fs_info->dev_root;
8111 struct btrfs_key key;
8113 u64 prev_dev_ext_end = 0;
8117 * We don't have a dev_root because we mounted with ignorebadroots and
8118 * failed to load the root, so we want to skip the verification in this
8121 * However if the dev root is fine, but the tree itself is corrupted
8122 * we'd still fail to mount. This verification is only to make sure
8123 * writes can happen safely, so instead just bypass this check
8124 * completely in the case of IGNOREBADROOTS.
8126 if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
8130 key.type = BTRFS_DEV_EXTENT_KEY;
8133 path = btrfs_alloc_path();
8137 path->reada = READA_FORWARD;
8138 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8142 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
8143 ret = btrfs_next_leaf(root, path);
8146 /* No dev extents at all? Not good */
8153 struct extent_buffer *leaf = path->nodes[0];
8154 struct btrfs_dev_extent *dext;
8155 int slot = path->slots[0];
8157 u64 physical_offset;
8161 btrfs_item_key_to_cpu(leaf, &key, slot);
8162 if (key.type != BTRFS_DEV_EXTENT_KEY)
8164 devid = key.objectid;
8165 physical_offset = key.offset;
8167 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
8168 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
8169 physical_len = btrfs_dev_extent_length(leaf, dext);
8171 /* Check if this dev extent overlaps with the previous one */
8172 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
8174 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
8175 devid, physical_offset, prev_dev_ext_end);
8180 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
8181 physical_offset, physical_len);
8185 prev_dev_ext_end = physical_offset + physical_len;
8187 ret = btrfs_next_item(root, path);
8196 /* Ensure all chunks have corresponding dev extents */
8197 ret = verify_chunk_dev_extent_mapping(fs_info);
8199 btrfs_free_path(path);
8204 * Check whether the given block group or device is pinned by any inode being
8205 * used as a swapfile.
8207 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
8209 struct btrfs_swapfile_pin *sp;
8210 struct rb_node *node;
8212 spin_lock(&fs_info->swapfile_pins_lock);
8213 node = fs_info->swapfile_pins.rb_node;
8215 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
8217 node = node->rb_left;
8218 else if (ptr > sp->ptr)
8219 node = node->rb_right;
8223 spin_unlock(&fs_info->swapfile_pins_lock);
8224 return node != NULL;
8227 static int relocating_repair_kthread(void *data)
8229 struct btrfs_block_group *cache = data;
8230 struct btrfs_fs_info *fs_info = cache->fs_info;
8234 target = cache->start;
8235 btrfs_put_block_group(cache);
8237 sb_start_write(fs_info->sb);
8238 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
8240 "zoned: skip relocating block group %llu to repair: EBUSY",
8242 sb_end_write(fs_info->sb);
8246 mutex_lock(&fs_info->reclaim_bgs_lock);
8248 /* Ensure block group still exists */
8249 cache = btrfs_lookup_block_group(fs_info, target);
8253 if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags))
8256 ret = btrfs_may_alloc_data_chunk(fs_info, target);
8261 "zoned: relocating block group %llu to repair IO failure",
8263 ret = btrfs_relocate_chunk(fs_info, target);
8267 btrfs_put_block_group(cache);
8268 mutex_unlock(&fs_info->reclaim_bgs_lock);
8269 btrfs_exclop_finish(fs_info);
8270 sb_end_write(fs_info->sb);
8275 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical)
8277 struct btrfs_block_group *cache;
8279 if (!btrfs_is_zoned(fs_info))
8282 /* Do not attempt to repair in degraded state */
8283 if (btrfs_test_opt(fs_info, DEGRADED))
8286 cache = btrfs_lookup_block_group(fs_info, logical);
8290 if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) {
8291 btrfs_put_block_group(cache);
8295 kthread_run(relocating_repair_kthread, cache,
8296 "btrfs-relocating-repair");
8301 static void map_raid56_repair_block(struct btrfs_io_context *bioc,
8302 struct btrfs_io_stripe *smap,
8305 int data_stripes = nr_bioc_data_stripes(bioc);
8308 for (i = 0; i < data_stripes; i++) {
8309 u64 stripe_start = bioc->full_stripe_logical +
8310 btrfs_stripe_nr_to_offset(i);
8312 if (logical >= stripe_start &&
8313 logical < stripe_start + BTRFS_STRIPE_LEN)
8316 ASSERT(i < data_stripes);
8317 smap->dev = bioc->stripes[i].dev;
8318 smap->physical = bioc->stripes[i].physical +
8319 ((logical - bioc->full_stripe_logical) &
8320 BTRFS_STRIPE_LEN_MASK);
8324 * Map a repair write into a single device.
8326 * A repair write is triggered by read time repair or scrub, which would only
8327 * update the contents of a single device.
8328 * Not update any other mirrors nor go through RMW path.
8330 * Callers should ensure:
8332 * - Call btrfs_bio_counter_inc_blocked() first
8333 * - The range does not cross stripe boundary
8334 * - Has a valid @mirror_num passed in.
8336 int btrfs_map_repair_block(struct btrfs_fs_info *fs_info,
8337 struct btrfs_io_stripe *smap, u64 logical,
8338 u32 length, int mirror_num)
8340 struct btrfs_io_context *bioc = NULL;
8341 u64 map_length = length;
8342 int mirror_ret = mirror_num;
8345 ASSERT(mirror_num > 0);
8347 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, &map_length,
8348 &bioc, smap, &mirror_ret);
8352 /* The map range should not cross stripe boundary. */
8353 ASSERT(map_length >= length);
8355 /* Already mapped to single stripe. */
8359 /* Map the RAID56 multi-stripe writes to a single one. */
8360 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
8361 map_raid56_repair_block(bioc, smap, logical);
8365 ASSERT(mirror_num <= bioc->num_stripes);
8366 smap->dev = bioc->stripes[mirror_num - 1].dev;
8367 smap->physical = bioc->stripes[mirror_num - 1].physical;
8369 btrfs_put_bioc(bioc);