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);
1307 * Look for a btrfs signature on a device. This may be called out of the mount path
1308 * and we are not allowed to call set_blocksize during the scan. The superblock
1309 * is read via pagecache.
1311 * With @mount_arg_dev it's a scan during mount time that will always register
1312 * the device or return an error. Multi-device and seeding devices are registered
1315 struct btrfs_device *btrfs_scan_one_device(const char *path, blk_mode_t flags,
1318 struct btrfs_super_block *disk_super;
1319 bool new_device_added = false;
1320 struct btrfs_device *device = NULL;
1321 struct file *bdev_file;
1322 u64 bytenr, bytenr_orig;
1325 lockdep_assert_held(&uuid_mutex);
1328 * we would like to check all the supers, but that would make
1329 * a btrfs mount succeed after a mkfs from a different FS.
1330 * So, we need to add a special mount option to scan for
1331 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1335 * Avoid an exclusive open here, as the systemd-udev may initiate the
1336 * device scan which may race with the user's mount or mkfs command,
1337 * resulting in failure.
1338 * Since the device scan is solely for reading purposes, there is no
1339 * need for an exclusive open. Additionally, the devices are read again
1340 * during the mount process. It is ok to get some inconsistent
1341 * values temporarily, as the device paths of the fsid are the only
1342 * required information for assembling the volume.
1344 bdev_file = bdev_file_open_by_path(path, flags, NULL, NULL);
1345 if (IS_ERR(bdev_file))
1346 return ERR_CAST(bdev_file);
1348 bytenr_orig = btrfs_sb_offset(0);
1349 ret = btrfs_sb_log_location_bdev(file_bdev(bdev_file), 0, READ, &bytenr);
1351 device = ERR_PTR(ret);
1352 goto error_bdev_put;
1355 disk_super = btrfs_read_disk_super(file_bdev(bdev_file), bytenr,
1357 if (IS_ERR(disk_super)) {
1358 device = ERR_CAST(disk_super);
1359 goto error_bdev_put;
1362 if (!mount_arg_dev && btrfs_super_num_devices(disk_super) == 1 &&
1363 !(btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING)) {
1366 ret = lookup_bdev(path, &devt);
1368 btrfs_warn(NULL, "lookup bdev failed for path %s: %d",
1371 btrfs_free_stale_devices(devt, NULL);
1373 pr_debug("BTRFS: skip registering single non-seed device %s (%d:%d)\n",
1374 path, MAJOR(devt), MINOR(devt));
1376 goto free_disk_super;
1379 device = device_list_add(path, disk_super, &new_device_added);
1380 if (!IS_ERR(device) && new_device_added)
1381 btrfs_free_stale_devices(device->devt, device);
1384 btrfs_release_disk_super(disk_super);
1393 * Try to find a chunk that intersects [start, start + len] range and when one
1394 * such is found, record the end of it in *start
1396 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1399 u64 physical_start, physical_end;
1401 lockdep_assert_held(&device->fs_info->chunk_mutex);
1403 if (find_first_extent_bit(&device->alloc_state, *start,
1404 &physical_start, &physical_end,
1405 CHUNK_ALLOCATED, NULL)) {
1407 if (in_range(physical_start, *start, len) ||
1408 in_range(*start, physical_start,
1409 physical_end + 1 - physical_start)) {
1410 *start = physical_end + 1;
1417 static u64 dev_extent_search_start(struct btrfs_device *device)
1419 switch (device->fs_devices->chunk_alloc_policy) {
1420 case BTRFS_CHUNK_ALLOC_REGULAR:
1421 return BTRFS_DEVICE_RANGE_RESERVED;
1422 case BTRFS_CHUNK_ALLOC_ZONED:
1424 * We don't care about the starting region like regular
1425 * allocator, because we anyway use/reserve the first two zones
1426 * for superblock logging.
1434 static bool dev_extent_hole_check_zoned(struct btrfs_device *device,
1435 u64 *hole_start, u64 *hole_size,
1438 u64 zone_size = device->zone_info->zone_size;
1441 bool changed = false;
1443 ASSERT(IS_ALIGNED(*hole_start, zone_size));
1445 while (*hole_size > 0) {
1446 pos = btrfs_find_allocatable_zones(device, *hole_start,
1447 *hole_start + *hole_size,
1449 if (pos != *hole_start) {
1450 *hole_size = *hole_start + *hole_size - pos;
1453 if (*hole_size < num_bytes)
1457 ret = btrfs_ensure_empty_zones(device, pos, num_bytes);
1459 /* Range is ensured to be empty */
1463 /* Given hole range was invalid (outside of device) */
1464 if (ret == -ERANGE) {
1465 *hole_start += *hole_size;
1470 *hole_start += zone_size;
1471 *hole_size -= zone_size;
1479 * Check if specified hole is suitable for allocation.
1481 * @device: the device which we have the hole
1482 * @hole_start: starting position of the hole
1483 * @hole_size: the size of the hole
1484 * @num_bytes: the size of the free space that we need
1486 * This function may modify @hole_start and @hole_size to reflect the suitable
1487 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1489 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1490 u64 *hole_size, u64 num_bytes)
1492 bool changed = false;
1493 u64 hole_end = *hole_start + *hole_size;
1497 * Check before we set max_hole_start, otherwise we could end up
1498 * sending back this offset anyway.
1500 if (contains_pending_extent(device, hole_start, *hole_size)) {
1501 if (hole_end >= *hole_start)
1502 *hole_size = hole_end - *hole_start;
1508 switch (device->fs_devices->chunk_alloc_policy) {
1509 case BTRFS_CHUNK_ALLOC_REGULAR:
1510 /* No extra check */
1512 case BTRFS_CHUNK_ALLOC_ZONED:
1513 if (dev_extent_hole_check_zoned(device, hole_start,
1514 hole_size, num_bytes)) {
1517 * The changed hole can contain pending extent.
1518 * Loop again to check that.
1534 * Find free space in the specified device.
1536 * @device: the device which we search the free space in
1537 * @num_bytes: the size of the free space that we need
1538 * @search_start: the position from which to begin the search
1539 * @start: store the start of the free space.
1540 * @len: the size of the free space. that we find, or the size
1541 * of the max free space if we don't find suitable free space
1543 * This does a pretty simple search, the expectation is that it is called very
1544 * infrequently and that a given device has a small number of extents.
1546 * @start is used to store the start of the free space if we find. But if we
1547 * don't find suitable free space, it will be used to store the start position
1548 * of the max free space.
1550 * @len is used to store the size of the free space that we find.
1551 * But if we don't find suitable free space, it is used to store the size of
1552 * the max free space.
1554 * NOTE: This function will search *commit* root of device tree, and does extra
1555 * check to ensure dev extents are not double allocated.
1556 * This makes the function safe to allocate dev extents but may not report
1557 * correct usable device space, as device extent freed in current transaction
1558 * is not reported as available.
1560 static int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1561 u64 *start, u64 *len)
1563 struct btrfs_fs_info *fs_info = device->fs_info;
1564 struct btrfs_root *root = fs_info->dev_root;
1565 struct btrfs_key key;
1566 struct btrfs_dev_extent *dev_extent;
1567 struct btrfs_path *path;
1571 u64 max_hole_size = 0;
1573 u64 search_end = device->total_bytes;
1576 struct extent_buffer *l;
1578 search_start = dev_extent_search_start(device);
1579 max_hole_start = search_start;
1581 WARN_ON(device->zone_info &&
1582 !IS_ALIGNED(num_bytes, device->zone_info->zone_size));
1584 path = btrfs_alloc_path();
1590 if (search_start >= search_end ||
1591 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1596 path->reada = READA_FORWARD;
1597 path->search_commit_root = 1;
1598 path->skip_locking = 1;
1600 key.objectid = device->devid;
1601 key.offset = search_start;
1602 key.type = BTRFS_DEV_EXTENT_KEY;
1604 ret = btrfs_search_backwards(root, &key, path);
1608 while (search_start < search_end) {
1610 slot = path->slots[0];
1611 if (slot >= btrfs_header_nritems(l)) {
1612 ret = btrfs_next_leaf(root, path);
1620 btrfs_item_key_to_cpu(l, &key, slot);
1622 if (key.objectid < device->devid)
1625 if (key.objectid > device->devid)
1628 if (key.type != BTRFS_DEV_EXTENT_KEY)
1631 if (key.offset > search_end)
1634 if (key.offset > search_start) {
1635 hole_size = key.offset - search_start;
1636 dev_extent_hole_check(device, &search_start, &hole_size,
1639 if (hole_size > max_hole_size) {
1640 max_hole_start = search_start;
1641 max_hole_size = hole_size;
1645 * If this free space is greater than which we need,
1646 * it must be the max free space that we have found
1647 * until now, so max_hole_start must point to the start
1648 * of this free space and the length of this free space
1649 * is stored in max_hole_size. Thus, we return
1650 * max_hole_start and max_hole_size and go back to the
1653 if (hole_size >= num_bytes) {
1659 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1660 extent_end = key.offset + btrfs_dev_extent_length(l,
1662 if (extent_end > search_start)
1663 search_start = extent_end;
1670 * At this point, search_start should be the end of
1671 * allocated dev extents, and when shrinking the device,
1672 * search_end may be smaller than search_start.
1674 if (search_end > search_start) {
1675 hole_size = search_end - search_start;
1676 if (dev_extent_hole_check(device, &search_start, &hole_size,
1678 btrfs_release_path(path);
1682 if (hole_size > max_hole_size) {
1683 max_hole_start = search_start;
1684 max_hole_size = hole_size;
1689 if (max_hole_size < num_bytes)
1694 ASSERT(max_hole_start + max_hole_size <= search_end);
1696 btrfs_free_path(path);
1697 *start = max_hole_start;
1699 *len = max_hole_size;
1703 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1704 struct btrfs_device *device,
1705 u64 start, u64 *dev_extent_len)
1707 struct btrfs_fs_info *fs_info = device->fs_info;
1708 struct btrfs_root *root = fs_info->dev_root;
1710 struct btrfs_path *path;
1711 struct btrfs_key key;
1712 struct btrfs_key found_key;
1713 struct extent_buffer *leaf = NULL;
1714 struct btrfs_dev_extent *extent = NULL;
1716 path = btrfs_alloc_path();
1720 key.objectid = device->devid;
1722 key.type = BTRFS_DEV_EXTENT_KEY;
1724 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1726 ret = btrfs_previous_item(root, path, key.objectid,
1727 BTRFS_DEV_EXTENT_KEY);
1730 leaf = path->nodes[0];
1731 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1732 extent = btrfs_item_ptr(leaf, path->slots[0],
1733 struct btrfs_dev_extent);
1734 BUG_ON(found_key.offset > start || found_key.offset +
1735 btrfs_dev_extent_length(leaf, extent) < start);
1737 btrfs_release_path(path);
1739 } else if (ret == 0) {
1740 leaf = path->nodes[0];
1741 extent = btrfs_item_ptr(leaf, path->slots[0],
1742 struct btrfs_dev_extent);
1747 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1749 ret = btrfs_del_item(trans, root, path);
1751 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1753 btrfs_free_path(path);
1757 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1762 read_lock(&fs_info->mapping_tree_lock);
1763 n = rb_last(&fs_info->mapping_tree.rb_root);
1765 struct btrfs_chunk_map *map;
1767 map = rb_entry(n, struct btrfs_chunk_map, rb_node);
1768 ret = map->start + map->chunk_len;
1770 read_unlock(&fs_info->mapping_tree_lock);
1775 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1779 struct btrfs_key key;
1780 struct btrfs_key found_key;
1781 struct btrfs_path *path;
1783 path = btrfs_alloc_path();
1787 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1788 key.type = BTRFS_DEV_ITEM_KEY;
1789 key.offset = (u64)-1;
1791 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1797 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1802 ret = btrfs_previous_item(fs_info->chunk_root, path,
1803 BTRFS_DEV_ITEMS_OBJECTID,
1804 BTRFS_DEV_ITEM_KEY);
1808 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1810 *devid_ret = found_key.offset + 1;
1814 btrfs_free_path(path);
1819 * the device information is stored in the chunk root
1820 * the btrfs_device struct should be fully filled in
1822 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1823 struct btrfs_device *device)
1826 struct btrfs_path *path;
1827 struct btrfs_dev_item *dev_item;
1828 struct extent_buffer *leaf;
1829 struct btrfs_key key;
1832 path = btrfs_alloc_path();
1836 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1837 key.type = BTRFS_DEV_ITEM_KEY;
1838 key.offset = device->devid;
1840 btrfs_reserve_chunk_metadata(trans, true);
1841 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1842 &key, sizeof(*dev_item));
1843 btrfs_trans_release_chunk_metadata(trans);
1847 leaf = path->nodes[0];
1848 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1850 btrfs_set_device_id(leaf, dev_item, device->devid);
1851 btrfs_set_device_generation(leaf, dev_item, 0);
1852 btrfs_set_device_type(leaf, dev_item, device->type);
1853 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1854 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1855 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1856 btrfs_set_device_total_bytes(leaf, dev_item,
1857 btrfs_device_get_disk_total_bytes(device));
1858 btrfs_set_device_bytes_used(leaf, dev_item,
1859 btrfs_device_get_bytes_used(device));
1860 btrfs_set_device_group(leaf, dev_item, 0);
1861 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1862 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1863 btrfs_set_device_start_offset(leaf, dev_item, 0);
1865 ptr = btrfs_device_uuid(dev_item);
1866 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1867 ptr = btrfs_device_fsid(dev_item);
1868 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1869 ptr, BTRFS_FSID_SIZE);
1870 btrfs_mark_buffer_dirty(trans, leaf);
1874 btrfs_free_path(path);
1879 * Function to update ctime/mtime for a given device path.
1880 * Mainly used for ctime/mtime based probe like libblkid.
1882 * We don't care about errors here, this is just to be kind to userspace.
1884 static void update_dev_time(const char *device_path)
1889 ret = kern_path(device_path, LOOKUP_FOLLOW, &path);
1893 inode_update_time(d_inode(path.dentry), S_MTIME | S_CTIME | S_VERSION);
1897 static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans,
1898 struct btrfs_device *device)
1900 struct btrfs_root *root = device->fs_info->chunk_root;
1902 struct btrfs_path *path;
1903 struct btrfs_key key;
1905 path = btrfs_alloc_path();
1909 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1910 key.type = BTRFS_DEV_ITEM_KEY;
1911 key.offset = device->devid;
1913 btrfs_reserve_chunk_metadata(trans, false);
1914 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1915 btrfs_trans_release_chunk_metadata(trans);
1922 ret = btrfs_del_item(trans, root, path);
1924 btrfs_free_path(path);
1929 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1930 * filesystem. It's up to the caller to adjust that number regarding eg. device
1933 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1941 seq = read_seqbegin(&fs_info->profiles_lock);
1943 all_avail = fs_info->avail_data_alloc_bits |
1944 fs_info->avail_system_alloc_bits |
1945 fs_info->avail_metadata_alloc_bits;
1946 } while (read_seqretry(&fs_info->profiles_lock, seq));
1948 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1949 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1952 if (num_devices < btrfs_raid_array[i].devs_min)
1953 return btrfs_raid_array[i].mindev_error;
1959 static struct btrfs_device * btrfs_find_next_active_device(
1960 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1962 struct btrfs_device *next_device;
1964 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1965 if (next_device != device &&
1966 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1967 && next_device->bdev)
1975 * Helper function to check if the given device is part of s_bdev / latest_dev
1976 * and replace it with the provided or the next active device, in the context
1977 * where this function called, there should be always be another device (or
1978 * this_dev) which is active.
1980 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
1981 struct btrfs_device *next_device)
1983 struct btrfs_fs_info *fs_info = device->fs_info;
1986 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1988 ASSERT(next_device);
1990 if (fs_info->sb->s_bdev &&
1991 (fs_info->sb->s_bdev == device->bdev))
1992 fs_info->sb->s_bdev = next_device->bdev;
1994 if (fs_info->fs_devices->latest_dev->bdev == device->bdev)
1995 fs_info->fs_devices->latest_dev = next_device;
1999 * Return btrfs_fs_devices::num_devices excluding the device that's being
2000 * currently replaced.
2002 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2004 u64 num_devices = fs_info->fs_devices->num_devices;
2006 down_read(&fs_info->dev_replace.rwsem);
2007 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2008 ASSERT(num_devices > 1);
2011 up_read(&fs_info->dev_replace.rwsem);
2016 static void btrfs_scratch_superblock(struct btrfs_fs_info *fs_info,
2017 struct block_device *bdev, int copy_num)
2019 struct btrfs_super_block *disk_super;
2020 const size_t len = sizeof(disk_super->magic);
2021 const u64 bytenr = btrfs_sb_offset(copy_num);
2024 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr);
2025 if (IS_ERR(disk_super))
2028 memset(&disk_super->magic, 0, len);
2029 folio_mark_dirty(virt_to_folio(disk_super));
2030 btrfs_release_disk_super(disk_super);
2032 ret = sync_blockdev_range(bdev, bytenr, bytenr + len - 1);
2034 btrfs_warn(fs_info, "error clearing superblock number %d (%d)",
2038 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, struct btrfs_device *device)
2041 struct block_device *bdev = device->bdev;
2046 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2047 if (bdev_is_zoned(bdev))
2048 btrfs_reset_sb_log_zones(bdev, copy_num);
2050 btrfs_scratch_superblock(fs_info, bdev, copy_num);
2053 /* Notify udev that device has changed */
2054 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2056 /* Update ctime/mtime for device path for libblkid */
2057 update_dev_time(device->name->str);
2060 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
2061 struct btrfs_dev_lookup_args *args,
2062 struct file **bdev_file)
2064 struct btrfs_trans_handle *trans;
2065 struct btrfs_device *device;
2066 struct btrfs_fs_devices *cur_devices;
2067 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2071 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2072 btrfs_err(fs_info, "device remove not supported on extent tree v2 yet");
2077 * The device list in fs_devices is accessed without locks (neither
2078 * uuid_mutex nor device_list_mutex) as it won't change on a mounted
2079 * filesystem and another device rm cannot run.
2081 num_devices = btrfs_num_devices(fs_info);
2083 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2087 device = btrfs_find_device(fs_info->fs_devices, args);
2090 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2096 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2097 btrfs_warn_in_rcu(fs_info,
2098 "cannot remove device %s (devid %llu) due to active swapfile",
2099 btrfs_dev_name(device), device->devid);
2103 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
2104 return BTRFS_ERROR_DEV_TGT_REPLACE;
2106 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2107 fs_info->fs_devices->rw_devices == 1)
2108 return BTRFS_ERROR_DEV_ONLY_WRITABLE;
2110 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2111 mutex_lock(&fs_info->chunk_mutex);
2112 list_del_init(&device->dev_alloc_list);
2113 device->fs_devices->rw_devices--;
2114 mutex_unlock(&fs_info->chunk_mutex);
2117 ret = btrfs_shrink_device(device, 0);
2121 trans = btrfs_start_transaction(fs_info->chunk_root, 0);
2122 if (IS_ERR(trans)) {
2123 ret = PTR_ERR(trans);
2127 ret = btrfs_rm_dev_item(trans, device);
2129 /* Any error in dev item removal is critical */
2131 "failed to remove device item for devid %llu: %d",
2132 device->devid, ret);
2133 btrfs_abort_transaction(trans, ret);
2134 btrfs_end_transaction(trans);
2138 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2139 btrfs_scrub_cancel_dev(device);
2142 * the device list mutex makes sure that we don't change
2143 * the device list while someone else is writing out all
2144 * the device supers. Whoever is writing all supers, should
2145 * lock the device list mutex before getting the number of
2146 * devices in the super block (super_copy). Conversely,
2147 * whoever updates the number of devices in the super block
2148 * (super_copy) should hold the device list mutex.
2152 * In normal cases the cur_devices == fs_devices. But in case
2153 * of deleting a seed device, the cur_devices should point to
2154 * its own fs_devices listed under the fs_devices->seed_list.
2156 cur_devices = device->fs_devices;
2157 mutex_lock(&fs_devices->device_list_mutex);
2158 list_del_rcu(&device->dev_list);
2160 cur_devices->num_devices--;
2161 cur_devices->total_devices--;
2162 /* Update total_devices of the parent fs_devices if it's seed */
2163 if (cur_devices != fs_devices)
2164 fs_devices->total_devices--;
2166 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2167 cur_devices->missing_devices--;
2169 btrfs_assign_next_active_device(device, NULL);
2171 if (device->bdev_file) {
2172 cur_devices->open_devices--;
2173 /* remove sysfs entry */
2174 btrfs_sysfs_remove_device(device);
2177 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2178 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2179 mutex_unlock(&fs_devices->device_list_mutex);
2182 * At this point, the device is zero sized and detached from the
2183 * devices list. All that's left is to zero out the old supers and
2186 * We cannot call btrfs_close_bdev() here because we're holding the sb
2187 * write lock, and fput() on the block device will pull in the
2188 * ->open_mutex on the block device and it's dependencies. Instead
2189 * just flush the device and let the caller do the final bdev_release.
2191 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2192 btrfs_scratch_superblocks(fs_info, device);
2194 sync_blockdev(device->bdev);
2195 invalidate_bdev(device->bdev);
2199 *bdev_file = device->bdev_file;
2201 btrfs_free_device(device);
2204 * This can happen if cur_devices is the private seed devices list. We
2205 * cannot call close_fs_devices() here because it expects the uuid_mutex
2206 * to be held, but in fact we don't need that for the private
2207 * seed_devices, we can simply decrement cur_devices->opened and then
2208 * remove it from our list and free the fs_devices.
2210 if (cur_devices->num_devices == 0) {
2211 list_del_init(&cur_devices->seed_list);
2212 ASSERT(cur_devices->opened == 1);
2213 cur_devices->opened--;
2214 free_fs_devices(cur_devices);
2217 ret = btrfs_commit_transaction(trans);
2222 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2223 mutex_lock(&fs_info->chunk_mutex);
2224 list_add(&device->dev_alloc_list,
2225 &fs_devices->alloc_list);
2226 device->fs_devices->rw_devices++;
2227 mutex_unlock(&fs_info->chunk_mutex);
2232 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2234 struct btrfs_fs_devices *fs_devices;
2236 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2239 * in case of fs with no seed, srcdev->fs_devices will point
2240 * to fs_devices of fs_info. However when the dev being replaced is
2241 * a seed dev it will point to the seed's local fs_devices. In short
2242 * srcdev will have its correct fs_devices in both the cases.
2244 fs_devices = srcdev->fs_devices;
2246 list_del_rcu(&srcdev->dev_list);
2247 list_del(&srcdev->dev_alloc_list);
2248 fs_devices->num_devices--;
2249 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2250 fs_devices->missing_devices--;
2252 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2253 fs_devices->rw_devices--;
2256 fs_devices->open_devices--;
2259 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2261 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2263 mutex_lock(&uuid_mutex);
2265 btrfs_close_bdev(srcdev);
2267 btrfs_free_device(srcdev);
2269 /* if this is no devs we rather delete the fs_devices */
2270 if (!fs_devices->num_devices) {
2272 * On a mounted FS, num_devices can't be zero unless it's a
2273 * seed. In case of a seed device being replaced, the replace
2274 * target added to the sprout FS, so there will be no more
2275 * device left under the seed FS.
2277 ASSERT(fs_devices->seeding);
2279 list_del_init(&fs_devices->seed_list);
2280 close_fs_devices(fs_devices);
2281 free_fs_devices(fs_devices);
2283 mutex_unlock(&uuid_mutex);
2286 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2288 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2290 mutex_lock(&fs_devices->device_list_mutex);
2292 btrfs_sysfs_remove_device(tgtdev);
2295 fs_devices->open_devices--;
2297 fs_devices->num_devices--;
2299 btrfs_assign_next_active_device(tgtdev, NULL);
2301 list_del_rcu(&tgtdev->dev_list);
2303 mutex_unlock(&fs_devices->device_list_mutex);
2305 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev);
2307 btrfs_close_bdev(tgtdev);
2309 btrfs_free_device(tgtdev);
2313 * Populate args from device at path.
2315 * @fs_info: the filesystem
2316 * @args: the args to populate
2317 * @path: the path to the device
2319 * This will read the super block of the device at @path and populate @args with
2320 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to
2321 * lookup a device to operate on, but need to do it before we take any locks.
2322 * This properly handles the special case of "missing" that a user may pass in,
2323 * and does some basic sanity checks. The caller must make sure that @path is
2324 * properly NUL terminated before calling in, and must call
2325 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and
2328 * Return: 0 for success, -errno for failure
2330 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,
2331 struct btrfs_dev_lookup_args *args,
2334 struct btrfs_super_block *disk_super;
2335 struct file *bdev_file;
2338 if (!path || !path[0])
2340 if (!strcmp(path, "missing")) {
2341 args->missing = true;
2345 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL);
2346 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL);
2347 if (!args->uuid || !args->fsid) {
2348 btrfs_put_dev_args_from_path(args);
2352 ret = btrfs_get_bdev_and_sb(path, BLK_OPEN_READ, NULL, 0,
2353 &bdev_file, &disk_super);
2355 btrfs_put_dev_args_from_path(args);
2359 args->devid = btrfs_stack_device_id(&disk_super->dev_item);
2360 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE);
2361 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2362 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE);
2364 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
2365 btrfs_release_disk_super(disk_super);
2371 * Only use this jointly with btrfs_get_dev_args_from_path() because we will
2372 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables
2373 * that don't need to be freed.
2375 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args)
2383 struct btrfs_device *btrfs_find_device_by_devspec(
2384 struct btrfs_fs_info *fs_info, u64 devid,
2385 const char *device_path)
2387 BTRFS_DEV_LOOKUP_ARGS(args);
2388 struct btrfs_device *device;
2393 device = btrfs_find_device(fs_info->fs_devices, &args);
2395 return ERR_PTR(-ENOENT);
2399 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path);
2401 return ERR_PTR(ret);
2402 device = btrfs_find_device(fs_info->fs_devices, &args);
2403 btrfs_put_dev_args_from_path(&args);
2405 return ERR_PTR(-ENOENT);
2409 static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info)
2411 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2412 struct btrfs_fs_devices *old_devices;
2413 struct btrfs_fs_devices *seed_devices;
2415 lockdep_assert_held(&uuid_mutex);
2416 if (!fs_devices->seeding)
2417 return ERR_PTR(-EINVAL);
2420 * Private copy of the seed devices, anchored at
2421 * fs_info->fs_devices->seed_list
2423 seed_devices = alloc_fs_devices(NULL);
2424 if (IS_ERR(seed_devices))
2425 return seed_devices;
2428 * It's necessary to retain a copy of the original seed fs_devices in
2429 * fs_uuids so that filesystems which have been seeded can successfully
2430 * reference the seed device from open_seed_devices. This also supports
2433 old_devices = clone_fs_devices(fs_devices);
2434 if (IS_ERR(old_devices)) {
2435 kfree(seed_devices);
2439 list_add(&old_devices->fs_list, &fs_uuids);
2441 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2442 seed_devices->opened = 1;
2443 INIT_LIST_HEAD(&seed_devices->devices);
2444 INIT_LIST_HEAD(&seed_devices->alloc_list);
2445 mutex_init(&seed_devices->device_list_mutex);
2447 return seed_devices;
2451 * Splice seed devices into the sprout fs_devices.
2452 * Generate a new fsid for the sprouted read-write filesystem.
2454 static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info,
2455 struct btrfs_fs_devices *seed_devices)
2457 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2458 struct btrfs_super_block *disk_super = fs_info->super_copy;
2459 struct btrfs_device *device;
2463 * We are updating the fsid, the thread leading to device_list_add()
2464 * could race, so uuid_mutex is needed.
2466 lockdep_assert_held(&uuid_mutex);
2469 * The threads listed below may traverse dev_list but can do that without
2470 * device_list_mutex:
2471 * - All device ops and balance - as we are in btrfs_exclop_start.
2472 * - Various dev_list readers - are using RCU.
2473 * - btrfs_ioctl_fitrim() - is using RCU.
2475 * For-read threads as below are using device_list_mutex:
2476 * - Readonly scrub btrfs_scrub_dev()
2477 * - Readonly scrub btrfs_scrub_progress()
2478 * - btrfs_get_dev_stats()
2480 lockdep_assert_held(&fs_devices->device_list_mutex);
2482 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2484 list_for_each_entry(device, &seed_devices->devices, dev_list)
2485 device->fs_devices = seed_devices;
2487 fs_devices->seeding = false;
2488 fs_devices->num_devices = 0;
2489 fs_devices->open_devices = 0;
2490 fs_devices->missing_devices = 0;
2491 fs_devices->rotating = false;
2492 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2494 generate_random_uuid(fs_devices->fsid);
2495 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2496 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2498 super_flags = btrfs_super_flags(disk_super) &
2499 ~BTRFS_SUPER_FLAG_SEEDING;
2500 btrfs_set_super_flags(disk_super, super_flags);
2504 * Store the expected generation for seed devices in device items.
2506 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2508 BTRFS_DEV_LOOKUP_ARGS(args);
2509 struct btrfs_fs_info *fs_info = trans->fs_info;
2510 struct btrfs_root *root = fs_info->chunk_root;
2511 struct btrfs_path *path;
2512 struct extent_buffer *leaf;
2513 struct btrfs_dev_item *dev_item;
2514 struct btrfs_device *device;
2515 struct btrfs_key key;
2516 u8 fs_uuid[BTRFS_FSID_SIZE];
2517 u8 dev_uuid[BTRFS_UUID_SIZE];
2520 path = btrfs_alloc_path();
2524 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2526 key.type = BTRFS_DEV_ITEM_KEY;
2529 btrfs_reserve_chunk_metadata(trans, false);
2530 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2531 btrfs_trans_release_chunk_metadata(trans);
2535 leaf = path->nodes[0];
2537 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2538 ret = btrfs_next_leaf(root, path);
2543 leaf = path->nodes[0];
2544 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2545 btrfs_release_path(path);
2549 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2550 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2551 key.type != BTRFS_DEV_ITEM_KEY)
2554 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2555 struct btrfs_dev_item);
2556 args.devid = btrfs_device_id(leaf, dev_item);
2557 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2559 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2561 args.uuid = dev_uuid;
2562 args.fsid = fs_uuid;
2563 device = btrfs_find_device(fs_info->fs_devices, &args);
2564 BUG_ON(!device); /* Logic error */
2566 if (device->fs_devices->seeding) {
2567 btrfs_set_device_generation(leaf, dev_item,
2568 device->generation);
2569 btrfs_mark_buffer_dirty(trans, leaf);
2577 btrfs_free_path(path);
2581 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2583 struct btrfs_root *root = fs_info->dev_root;
2584 struct btrfs_trans_handle *trans;
2585 struct btrfs_device *device;
2586 struct file *bdev_file;
2587 struct super_block *sb = fs_info->sb;
2588 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2589 struct btrfs_fs_devices *seed_devices = NULL;
2590 u64 orig_super_total_bytes;
2591 u64 orig_super_num_devices;
2593 bool seeding_dev = false;
2594 bool locked = false;
2596 if (sb_rdonly(sb) && !fs_devices->seeding)
2599 bdev_file = bdev_file_open_by_path(device_path, BLK_OPEN_WRITE,
2600 fs_info->bdev_holder, NULL);
2601 if (IS_ERR(bdev_file))
2602 return PTR_ERR(bdev_file);
2604 if (!btrfs_check_device_zone_type(fs_info, file_bdev(bdev_file))) {
2609 if (fs_devices->seeding) {
2611 down_write(&sb->s_umount);
2612 mutex_lock(&uuid_mutex);
2616 sync_blockdev(file_bdev(bdev_file));
2619 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2620 if (device->bdev == file_bdev(bdev_file)) {
2628 device = btrfs_alloc_device(fs_info, NULL, NULL, device_path);
2629 if (IS_ERR(device)) {
2630 /* we can safely leave the fs_devices entry around */
2631 ret = PTR_ERR(device);
2635 device->fs_info = fs_info;
2636 device->bdev_file = bdev_file;
2637 device->bdev = file_bdev(bdev_file);
2638 ret = lookup_bdev(device_path, &device->devt);
2640 goto error_free_device;
2642 ret = btrfs_get_dev_zone_info(device, false);
2644 goto error_free_device;
2646 trans = btrfs_start_transaction(root, 0);
2647 if (IS_ERR(trans)) {
2648 ret = PTR_ERR(trans);
2649 goto error_free_zone;
2652 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2653 device->generation = trans->transid;
2654 device->io_width = fs_info->sectorsize;
2655 device->io_align = fs_info->sectorsize;
2656 device->sector_size = fs_info->sectorsize;
2657 device->total_bytes =
2658 round_down(bdev_nr_bytes(device->bdev), fs_info->sectorsize);
2659 device->disk_total_bytes = device->total_bytes;
2660 device->commit_total_bytes = device->total_bytes;
2661 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2662 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2663 device->dev_stats_valid = 1;
2664 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2667 btrfs_clear_sb_rdonly(sb);
2669 /* GFP_KERNEL allocation must not be under device_list_mutex */
2670 seed_devices = btrfs_init_sprout(fs_info);
2671 if (IS_ERR(seed_devices)) {
2672 ret = PTR_ERR(seed_devices);
2673 btrfs_abort_transaction(trans, ret);
2678 mutex_lock(&fs_devices->device_list_mutex);
2680 btrfs_setup_sprout(fs_info, seed_devices);
2681 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev,
2685 device->fs_devices = fs_devices;
2687 mutex_lock(&fs_info->chunk_mutex);
2688 list_add_rcu(&device->dev_list, &fs_devices->devices);
2689 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2690 fs_devices->num_devices++;
2691 fs_devices->open_devices++;
2692 fs_devices->rw_devices++;
2693 fs_devices->total_devices++;
2694 fs_devices->total_rw_bytes += device->total_bytes;
2696 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2698 if (!bdev_nonrot(device->bdev))
2699 fs_devices->rotating = true;
2701 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2702 btrfs_set_super_total_bytes(fs_info->super_copy,
2703 round_down(orig_super_total_bytes + device->total_bytes,
2704 fs_info->sectorsize));
2706 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2707 btrfs_set_super_num_devices(fs_info->super_copy,
2708 orig_super_num_devices + 1);
2711 * we've got more storage, clear any full flags on the space
2714 btrfs_clear_space_info_full(fs_info);
2716 mutex_unlock(&fs_info->chunk_mutex);
2718 /* Add sysfs device entry */
2719 btrfs_sysfs_add_device(device);
2721 mutex_unlock(&fs_devices->device_list_mutex);
2724 mutex_lock(&fs_info->chunk_mutex);
2725 ret = init_first_rw_device(trans);
2726 mutex_unlock(&fs_info->chunk_mutex);
2728 btrfs_abort_transaction(trans, ret);
2733 ret = btrfs_add_dev_item(trans, device);
2735 btrfs_abort_transaction(trans, ret);
2740 ret = btrfs_finish_sprout(trans);
2742 btrfs_abort_transaction(trans, ret);
2747 * fs_devices now represents the newly sprouted filesystem and
2748 * its fsid has been changed by btrfs_sprout_splice().
2750 btrfs_sysfs_update_sprout_fsid(fs_devices);
2753 ret = btrfs_commit_transaction(trans);
2756 mutex_unlock(&uuid_mutex);
2757 up_write(&sb->s_umount);
2760 if (ret) /* transaction commit */
2763 ret = btrfs_relocate_sys_chunks(fs_info);
2765 btrfs_handle_fs_error(fs_info, ret,
2766 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2767 trans = btrfs_attach_transaction(root);
2768 if (IS_ERR(trans)) {
2769 if (PTR_ERR(trans) == -ENOENT)
2771 ret = PTR_ERR(trans);
2775 ret = btrfs_commit_transaction(trans);
2779 * Now that we have written a new super block to this device, check all
2780 * other fs_devices list if device_path alienates any other scanned
2782 * We can ignore the return value as it typically returns -EINVAL and
2783 * only succeeds if the device was an alien.
2785 btrfs_forget_devices(device->devt);
2787 /* Update ctime/mtime for blkid or udev */
2788 update_dev_time(device_path);
2793 btrfs_sysfs_remove_device(device);
2794 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2795 mutex_lock(&fs_info->chunk_mutex);
2796 list_del_rcu(&device->dev_list);
2797 list_del(&device->dev_alloc_list);
2798 fs_info->fs_devices->num_devices--;
2799 fs_info->fs_devices->open_devices--;
2800 fs_info->fs_devices->rw_devices--;
2801 fs_info->fs_devices->total_devices--;
2802 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2803 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2804 btrfs_set_super_total_bytes(fs_info->super_copy,
2805 orig_super_total_bytes);
2806 btrfs_set_super_num_devices(fs_info->super_copy,
2807 orig_super_num_devices);
2808 mutex_unlock(&fs_info->chunk_mutex);
2809 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2812 btrfs_set_sb_rdonly(sb);
2814 btrfs_end_transaction(trans);
2816 btrfs_destroy_dev_zone_info(device);
2818 btrfs_free_device(device);
2822 mutex_unlock(&uuid_mutex);
2823 up_write(&sb->s_umount);
2828 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2829 struct btrfs_device *device)
2832 struct btrfs_path *path;
2833 struct btrfs_root *root = device->fs_info->chunk_root;
2834 struct btrfs_dev_item *dev_item;
2835 struct extent_buffer *leaf;
2836 struct btrfs_key key;
2838 path = btrfs_alloc_path();
2842 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2843 key.type = BTRFS_DEV_ITEM_KEY;
2844 key.offset = device->devid;
2846 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2855 leaf = path->nodes[0];
2856 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2858 btrfs_set_device_id(leaf, dev_item, device->devid);
2859 btrfs_set_device_type(leaf, dev_item, device->type);
2860 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2861 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2862 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2863 btrfs_set_device_total_bytes(leaf, dev_item,
2864 btrfs_device_get_disk_total_bytes(device));
2865 btrfs_set_device_bytes_used(leaf, dev_item,
2866 btrfs_device_get_bytes_used(device));
2867 btrfs_mark_buffer_dirty(trans, leaf);
2870 btrfs_free_path(path);
2874 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2875 struct btrfs_device *device, u64 new_size)
2877 struct btrfs_fs_info *fs_info = device->fs_info;
2878 struct btrfs_super_block *super_copy = fs_info->super_copy;
2883 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2886 new_size = round_down(new_size, fs_info->sectorsize);
2888 mutex_lock(&fs_info->chunk_mutex);
2889 old_total = btrfs_super_total_bytes(super_copy);
2890 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2892 if (new_size <= device->total_bytes ||
2893 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2894 mutex_unlock(&fs_info->chunk_mutex);
2898 btrfs_set_super_total_bytes(super_copy,
2899 round_down(old_total + diff, fs_info->sectorsize));
2900 device->fs_devices->total_rw_bytes += diff;
2901 atomic64_add(diff, &fs_info->free_chunk_space);
2903 btrfs_device_set_total_bytes(device, new_size);
2904 btrfs_device_set_disk_total_bytes(device, new_size);
2905 btrfs_clear_space_info_full(device->fs_info);
2906 if (list_empty(&device->post_commit_list))
2907 list_add_tail(&device->post_commit_list,
2908 &trans->transaction->dev_update_list);
2909 mutex_unlock(&fs_info->chunk_mutex);
2911 btrfs_reserve_chunk_metadata(trans, false);
2912 ret = btrfs_update_device(trans, device);
2913 btrfs_trans_release_chunk_metadata(trans);
2918 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2920 struct btrfs_fs_info *fs_info = trans->fs_info;
2921 struct btrfs_root *root = fs_info->chunk_root;
2923 struct btrfs_path *path;
2924 struct btrfs_key key;
2926 path = btrfs_alloc_path();
2930 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2931 key.offset = chunk_offset;
2932 key.type = BTRFS_CHUNK_ITEM_KEY;
2934 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2937 else if (ret > 0) { /* Logic error or corruption */
2938 btrfs_handle_fs_error(fs_info, -ENOENT,
2939 "Failed lookup while freeing chunk.");
2944 ret = btrfs_del_item(trans, root, path);
2946 btrfs_handle_fs_error(fs_info, ret,
2947 "Failed to delete chunk item.");
2949 btrfs_free_path(path);
2953 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2955 struct btrfs_super_block *super_copy = fs_info->super_copy;
2956 struct btrfs_disk_key *disk_key;
2957 struct btrfs_chunk *chunk;
2964 struct btrfs_key key;
2966 lockdep_assert_held(&fs_info->chunk_mutex);
2967 array_size = btrfs_super_sys_array_size(super_copy);
2969 ptr = super_copy->sys_chunk_array;
2972 while (cur < array_size) {
2973 disk_key = (struct btrfs_disk_key *)ptr;
2974 btrfs_disk_key_to_cpu(&key, disk_key);
2976 len = sizeof(*disk_key);
2978 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2979 chunk = (struct btrfs_chunk *)(ptr + len);
2980 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2981 len += btrfs_chunk_item_size(num_stripes);
2986 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2987 key.offset == chunk_offset) {
2988 memmove(ptr, ptr + len, array_size - (cur + len));
2990 btrfs_set_super_sys_array_size(super_copy, array_size);
2999 struct btrfs_chunk_map *btrfs_find_chunk_map_nolock(struct btrfs_fs_info *fs_info,
3000 u64 logical, u64 length)
3002 struct rb_node *node = fs_info->mapping_tree.rb_root.rb_node;
3003 struct rb_node *prev = NULL;
3004 struct rb_node *orig_prev;
3005 struct btrfs_chunk_map *map;
3006 struct btrfs_chunk_map *prev_map = NULL;
3009 map = rb_entry(node, struct btrfs_chunk_map, rb_node);
3013 if (logical < map->start) {
3014 node = node->rb_left;
3015 } else if (logical >= map->start + map->chunk_len) {
3016 node = node->rb_right;
3018 refcount_inc(&map->refs);
3027 while (prev && logical >= prev_map->start + prev_map->chunk_len) {
3028 prev = rb_next(prev);
3029 prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node);
3034 prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node);
3035 while (prev && logical < prev_map->start) {
3036 prev = rb_prev(prev);
3037 prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node);
3042 u64 end = logical + length;
3045 * Caller can pass a U64_MAX length when it wants to get any
3046 * chunk starting at an offset of 'logical' or higher, so deal
3047 * with underflow by resetting the end offset to U64_MAX.
3052 if (end > prev_map->start &&
3053 logical < prev_map->start + prev_map->chunk_len) {
3054 refcount_inc(&prev_map->refs);
3062 struct btrfs_chunk_map *btrfs_find_chunk_map(struct btrfs_fs_info *fs_info,
3063 u64 logical, u64 length)
3065 struct btrfs_chunk_map *map;
3067 read_lock(&fs_info->mapping_tree_lock);
3068 map = btrfs_find_chunk_map_nolock(fs_info, logical, length);
3069 read_unlock(&fs_info->mapping_tree_lock);
3075 * Find the mapping containing the given logical extent.
3077 * @logical: Logical block offset in bytes.
3078 * @length: Length of extent in bytes.
3080 * Return: Chunk mapping or ERR_PTR.
3082 struct btrfs_chunk_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
3083 u64 logical, u64 length)
3085 struct btrfs_chunk_map *map;
3087 map = btrfs_find_chunk_map(fs_info, logical, length);
3089 if (unlikely(!map)) {
3091 "unable to find chunk map for logical %llu length %llu",
3093 return ERR_PTR(-EINVAL);
3096 if (unlikely(map->start > logical || map->start + map->chunk_len <= logical)) {
3098 "found a bad chunk map, wanted %llu-%llu, found %llu-%llu",
3099 logical, logical + length, map->start,
3100 map->start + map->chunk_len);
3101 btrfs_free_chunk_map(map);
3102 return ERR_PTR(-EINVAL);
3105 /* Callers are responsible for dropping the reference. */
3109 static int remove_chunk_item(struct btrfs_trans_handle *trans,
3110 struct btrfs_chunk_map *map, u64 chunk_offset)
3115 * Removing chunk items and updating the device items in the chunks btree
3116 * requires holding the chunk_mutex.
3117 * See the comment at btrfs_chunk_alloc() for the details.
3119 lockdep_assert_held(&trans->fs_info->chunk_mutex);
3121 for (i = 0; i < map->num_stripes; i++) {
3124 ret = btrfs_update_device(trans, map->stripes[i].dev);
3129 return btrfs_free_chunk(trans, chunk_offset);
3132 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
3134 struct btrfs_fs_info *fs_info = trans->fs_info;
3135 struct btrfs_chunk_map *map;
3136 u64 dev_extent_len = 0;
3138 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3140 map = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
3143 * This is a logic error, but we don't want to just rely on the
3144 * user having built with ASSERT enabled, so if ASSERT doesn't
3145 * do anything we still error out.
3148 return PTR_ERR(map);
3152 * First delete the device extent items from the devices btree.
3153 * We take the device_list_mutex to avoid racing with the finishing phase
3154 * of a device replace operation. See the comment below before acquiring
3155 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
3156 * because that can result in a deadlock when deleting the device extent
3157 * items from the devices btree - COWing an extent buffer from the btree
3158 * may result in allocating a new metadata chunk, which would attempt to
3159 * lock again fs_info->chunk_mutex.
3161 mutex_lock(&fs_devices->device_list_mutex);
3162 for (i = 0; i < map->num_stripes; i++) {
3163 struct btrfs_device *device = map->stripes[i].dev;
3164 ret = btrfs_free_dev_extent(trans, device,
3165 map->stripes[i].physical,
3168 mutex_unlock(&fs_devices->device_list_mutex);
3169 btrfs_abort_transaction(trans, ret);
3173 if (device->bytes_used > 0) {
3174 mutex_lock(&fs_info->chunk_mutex);
3175 btrfs_device_set_bytes_used(device,
3176 device->bytes_used - dev_extent_len);
3177 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3178 btrfs_clear_space_info_full(fs_info);
3179 mutex_unlock(&fs_info->chunk_mutex);
3182 mutex_unlock(&fs_devices->device_list_mutex);
3185 * We acquire fs_info->chunk_mutex for 2 reasons:
3187 * 1) Just like with the first phase of the chunk allocation, we must
3188 * reserve system space, do all chunk btree updates and deletions, and
3189 * update the system chunk array in the superblock while holding this
3190 * mutex. This is for similar reasons as explained on the comment at
3191 * the top of btrfs_chunk_alloc();
3193 * 2) Prevent races with the final phase of a device replace operation
3194 * that replaces the device object associated with the map's stripes,
3195 * because the device object's id can change at any time during that
3196 * final phase of the device replace operation
3197 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3198 * replaced device and then see it with an ID of
3199 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3200 * the device item, which does not exists on the chunk btree.
3201 * The finishing phase of device replace acquires both the
3202 * device_list_mutex and the chunk_mutex, in that order, so we are
3203 * safe by just acquiring the chunk_mutex.
3205 trans->removing_chunk = true;
3206 mutex_lock(&fs_info->chunk_mutex);
3208 check_system_chunk(trans, map->type);
3210 ret = remove_chunk_item(trans, map, chunk_offset);
3212 * Normally we should not get -ENOSPC since we reserved space before
3213 * through the call to check_system_chunk().
3215 * Despite our system space_info having enough free space, we may not
3216 * be able to allocate extents from its block groups, because all have
3217 * an incompatible profile, which will force us to allocate a new system
3218 * block group with the right profile, or right after we called
3219 * check_system_space() above, a scrub turned the only system block group
3220 * with enough free space into RO mode.
3221 * This is explained with more detail at do_chunk_alloc().
3223 * So if we get -ENOSPC, allocate a new system chunk and retry once.
3225 if (ret == -ENOSPC) {
3226 const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
3227 struct btrfs_block_group *sys_bg;
3229 sys_bg = btrfs_create_chunk(trans, sys_flags);
3230 if (IS_ERR(sys_bg)) {
3231 ret = PTR_ERR(sys_bg);
3232 btrfs_abort_transaction(trans, ret);
3236 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3238 btrfs_abort_transaction(trans, ret);
3242 ret = remove_chunk_item(trans, map, chunk_offset);
3244 btrfs_abort_transaction(trans, ret);
3248 btrfs_abort_transaction(trans, ret);
3252 trace_btrfs_chunk_free(fs_info, map, chunk_offset, map->chunk_len);
3254 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3255 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3257 btrfs_abort_transaction(trans, ret);
3262 mutex_unlock(&fs_info->chunk_mutex);
3263 trans->removing_chunk = false;
3266 * We are done with chunk btree updates and deletions, so release the
3267 * system space we previously reserved (with check_system_chunk()).
3269 btrfs_trans_release_chunk_metadata(trans);
3271 ret = btrfs_remove_block_group(trans, map);
3273 btrfs_abort_transaction(trans, ret);
3278 if (trans->removing_chunk) {
3279 mutex_unlock(&fs_info->chunk_mutex);
3280 trans->removing_chunk = false;
3283 btrfs_free_chunk_map(map);
3287 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3289 struct btrfs_root *root = fs_info->chunk_root;
3290 struct btrfs_trans_handle *trans;
3291 struct btrfs_block_group *block_group;
3295 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3297 "relocate: not supported on extent tree v2 yet");
3302 * Prevent races with automatic removal of unused block groups.
3303 * After we relocate and before we remove the chunk with offset
3304 * chunk_offset, automatic removal of the block group can kick in,
3305 * resulting in a failure when calling btrfs_remove_chunk() below.
3307 * Make sure to acquire this mutex before doing a tree search (dev
3308 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3309 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3310 * we release the path used to search the chunk/dev tree and before
3311 * the current task acquires this mutex and calls us.
3313 lockdep_assert_held(&fs_info->reclaim_bgs_lock);
3315 /* step one, relocate all the extents inside this chunk */
3316 btrfs_scrub_pause(fs_info);
3317 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3318 btrfs_scrub_continue(fs_info);
3321 * If we had a transaction abort, stop all running scrubs.
3322 * See transaction.c:cleanup_transaction() why we do it here.
3324 if (BTRFS_FS_ERROR(fs_info))
3325 btrfs_scrub_cancel(fs_info);
3329 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3332 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3333 length = block_group->length;
3334 btrfs_put_block_group(block_group);
3337 * On a zoned file system, discard the whole block group, this will
3338 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3339 * resetting the zone fails, don't treat it as a fatal problem from the
3340 * filesystem's point of view.
3342 if (btrfs_is_zoned(fs_info)) {
3343 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
3346 "failed to reset zone %llu after relocation",
3350 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3352 if (IS_ERR(trans)) {
3353 ret = PTR_ERR(trans);
3354 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3359 * step two, delete the device extents and the
3360 * chunk tree entries
3362 ret = btrfs_remove_chunk(trans, chunk_offset);
3363 btrfs_end_transaction(trans);
3367 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3369 struct btrfs_root *chunk_root = fs_info->chunk_root;
3370 struct btrfs_path *path;
3371 struct extent_buffer *leaf;
3372 struct btrfs_chunk *chunk;
3373 struct btrfs_key key;
3374 struct btrfs_key found_key;
3376 bool retried = false;
3380 path = btrfs_alloc_path();
3385 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3386 key.offset = (u64)-1;
3387 key.type = BTRFS_CHUNK_ITEM_KEY;
3390 mutex_lock(&fs_info->reclaim_bgs_lock);
3391 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3393 mutex_unlock(&fs_info->reclaim_bgs_lock);
3398 * On the first search we would find chunk tree with
3399 * offset -1, which is not possible. On subsequent
3400 * loops this would find an existing item on an invalid
3401 * offset (one less than the previous one, wrong
3402 * alignment and size).
3408 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3411 mutex_unlock(&fs_info->reclaim_bgs_lock);
3417 leaf = path->nodes[0];
3418 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3420 chunk = btrfs_item_ptr(leaf, path->slots[0],
3421 struct btrfs_chunk);
3422 chunk_type = btrfs_chunk_type(leaf, chunk);
3423 btrfs_release_path(path);
3425 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3426 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3432 mutex_unlock(&fs_info->reclaim_bgs_lock);
3434 if (found_key.offset == 0)
3436 key.offset = found_key.offset - 1;
3439 if (failed && !retried) {
3443 } else if (WARN_ON(failed && retried)) {
3447 btrfs_free_path(path);
3452 * return 1 : allocate a data chunk successfully,
3453 * return <0: errors during allocating a data chunk,
3454 * return 0 : no need to allocate a data chunk.
3456 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3459 struct btrfs_block_group *cache;
3463 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3465 chunk_type = cache->flags;
3466 btrfs_put_block_group(cache);
3468 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3471 spin_lock(&fs_info->data_sinfo->lock);
3472 bytes_used = fs_info->data_sinfo->bytes_used;
3473 spin_unlock(&fs_info->data_sinfo->lock);
3476 struct btrfs_trans_handle *trans;
3479 trans = btrfs_join_transaction(fs_info->tree_root);
3481 return PTR_ERR(trans);
3483 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3484 btrfs_end_transaction(trans);
3493 static void btrfs_disk_balance_args_to_cpu(struct btrfs_balance_args *cpu,
3494 const struct btrfs_disk_balance_args *disk)
3496 memset(cpu, 0, sizeof(*cpu));
3498 cpu->profiles = le64_to_cpu(disk->profiles);
3499 cpu->usage = le64_to_cpu(disk->usage);
3500 cpu->devid = le64_to_cpu(disk->devid);
3501 cpu->pstart = le64_to_cpu(disk->pstart);
3502 cpu->pend = le64_to_cpu(disk->pend);
3503 cpu->vstart = le64_to_cpu(disk->vstart);
3504 cpu->vend = le64_to_cpu(disk->vend);
3505 cpu->target = le64_to_cpu(disk->target);
3506 cpu->flags = le64_to_cpu(disk->flags);
3507 cpu->limit = le64_to_cpu(disk->limit);
3508 cpu->stripes_min = le32_to_cpu(disk->stripes_min);
3509 cpu->stripes_max = le32_to_cpu(disk->stripes_max);
3512 static void btrfs_cpu_balance_args_to_disk(struct btrfs_disk_balance_args *disk,
3513 const struct btrfs_balance_args *cpu)
3515 memset(disk, 0, sizeof(*disk));
3517 disk->profiles = cpu_to_le64(cpu->profiles);
3518 disk->usage = cpu_to_le64(cpu->usage);
3519 disk->devid = cpu_to_le64(cpu->devid);
3520 disk->pstart = cpu_to_le64(cpu->pstart);
3521 disk->pend = cpu_to_le64(cpu->pend);
3522 disk->vstart = cpu_to_le64(cpu->vstart);
3523 disk->vend = cpu_to_le64(cpu->vend);
3524 disk->target = cpu_to_le64(cpu->target);
3525 disk->flags = cpu_to_le64(cpu->flags);
3526 disk->limit = cpu_to_le64(cpu->limit);
3527 disk->stripes_min = cpu_to_le32(cpu->stripes_min);
3528 disk->stripes_max = cpu_to_le32(cpu->stripes_max);
3531 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3532 struct btrfs_balance_control *bctl)
3534 struct btrfs_root *root = fs_info->tree_root;
3535 struct btrfs_trans_handle *trans;
3536 struct btrfs_balance_item *item;
3537 struct btrfs_disk_balance_args disk_bargs;
3538 struct btrfs_path *path;
3539 struct extent_buffer *leaf;
3540 struct btrfs_key key;
3543 path = btrfs_alloc_path();
3547 trans = btrfs_start_transaction(root, 0);
3548 if (IS_ERR(trans)) {
3549 btrfs_free_path(path);
3550 return PTR_ERR(trans);
3553 key.objectid = BTRFS_BALANCE_OBJECTID;
3554 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3557 ret = btrfs_insert_empty_item(trans, root, path, &key,
3562 leaf = path->nodes[0];
3563 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3565 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3567 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3568 btrfs_set_balance_data(leaf, item, &disk_bargs);
3569 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3570 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3571 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3572 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3574 btrfs_set_balance_flags(leaf, item, bctl->flags);
3576 btrfs_mark_buffer_dirty(trans, leaf);
3578 btrfs_free_path(path);
3579 err = btrfs_commit_transaction(trans);
3585 static int del_balance_item(struct btrfs_fs_info *fs_info)
3587 struct btrfs_root *root = fs_info->tree_root;
3588 struct btrfs_trans_handle *trans;
3589 struct btrfs_path *path;
3590 struct btrfs_key key;
3593 path = btrfs_alloc_path();
3597 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3598 if (IS_ERR(trans)) {
3599 btrfs_free_path(path);
3600 return PTR_ERR(trans);
3603 key.objectid = BTRFS_BALANCE_OBJECTID;
3604 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3607 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3615 ret = btrfs_del_item(trans, root, path);
3617 btrfs_free_path(path);
3618 err = btrfs_commit_transaction(trans);
3625 * This is a heuristic used to reduce the number of chunks balanced on
3626 * resume after balance was interrupted.
3628 static void update_balance_args(struct btrfs_balance_control *bctl)
3631 * Turn on soft mode for chunk types that were being converted.
3633 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3634 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3635 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3636 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3637 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3638 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3641 * Turn on usage filter if is not already used. The idea is
3642 * that chunks that we have already balanced should be
3643 * reasonably full. Don't do it for chunks that are being
3644 * converted - that will keep us from relocating unconverted
3645 * (albeit full) chunks.
3647 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3648 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3649 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3650 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3651 bctl->data.usage = 90;
3653 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3654 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3655 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3656 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3657 bctl->sys.usage = 90;
3659 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3660 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3661 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3662 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3663 bctl->meta.usage = 90;
3668 * Clear the balance status in fs_info and delete the balance item from disk.
3670 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3672 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3675 ASSERT(fs_info->balance_ctl);
3677 spin_lock(&fs_info->balance_lock);
3678 fs_info->balance_ctl = NULL;
3679 spin_unlock(&fs_info->balance_lock);
3682 ret = del_balance_item(fs_info);
3684 btrfs_handle_fs_error(fs_info, ret, NULL);
3688 * Balance filters. Return 1 if chunk should be filtered out
3689 * (should not be balanced).
3691 static int chunk_profiles_filter(u64 chunk_type,
3692 struct btrfs_balance_args *bargs)
3694 chunk_type = chunk_to_extended(chunk_type) &
3695 BTRFS_EXTENDED_PROFILE_MASK;
3697 if (bargs->profiles & chunk_type)
3703 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3704 struct btrfs_balance_args *bargs)
3706 struct btrfs_block_group *cache;
3708 u64 user_thresh_min;
3709 u64 user_thresh_max;
3712 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3713 chunk_used = cache->used;
3715 if (bargs->usage_min == 0)
3716 user_thresh_min = 0;
3718 user_thresh_min = mult_perc(cache->length, bargs->usage_min);
3720 if (bargs->usage_max == 0)
3721 user_thresh_max = 1;
3722 else if (bargs->usage_max > 100)
3723 user_thresh_max = cache->length;
3725 user_thresh_max = mult_perc(cache->length, bargs->usage_max);
3727 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3730 btrfs_put_block_group(cache);
3734 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3735 u64 chunk_offset, struct btrfs_balance_args *bargs)
3737 struct btrfs_block_group *cache;
3738 u64 chunk_used, user_thresh;
3741 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3742 chunk_used = cache->used;
3744 if (bargs->usage_min == 0)
3746 else if (bargs->usage > 100)
3747 user_thresh = cache->length;
3749 user_thresh = mult_perc(cache->length, bargs->usage);
3751 if (chunk_used < user_thresh)
3754 btrfs_put_block_group(cache);
3758 static int chunk_devid_filter(struct extent_buffer *leaf,
3759 struct btrfs_chunk *chunk,
3760 struct btrfs_balance_args *bargs)
3762 struct btrfs_stripe *stripe;
3763 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3766 for (i = 0; i < num_stripes; i++) {
3767 stripe = btrfs_stripe_nr(chunk, i);
3768 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3775 static u64 calc_data_stripes(u64 type, int num_stripes)
3777 const int index = btrfs_bg_flags_to_raid_index(type);
3778 const int ncopies = btrfs_raid_array[index].ncopies;
3779 const int nparity = btrfs_raid_array[index].nparity;
3781 return (num_stripes - nparity) / ncopies;
3784 /* [pstart, pend) */
3785 static int chunk_drange_filter(struct extent_buffer *leaf,
3786 struct btrfs_chunk *chunk,
3787 struct btrfs_balance_args *bargs)
3789 struct btrfs_stripe *stripe;
3790 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3797 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3800 type = btrfs_chunk_type(leaf, chunk);
3801 factor = calc_data_stripes(type, num_stripes);
3803 for (i = 0; i < num_stripes; i++) {
3804 stripe = btrfs_stripe_nr(chunk, i);
3805 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3808 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3809 stripe_length = btrfs_chunk_length(leaf, chunk);
3810 stripe_length = div_u64(stripe_length, factor);
3812 if (stripe_offset < bargs->pend &&
3813 stripe_offset + stripe_length > bargs->pstart)
3820 /* [vstart, vend) */
3821 static int chunk_vrange_filter(struct extent_buffer *leaf,
3822 struct btrfs_chunk *chunk,
3824 struct btrfs_balance_args *bargs)
3826 if (chunk_offset < bargs->vend &&
3827 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3828 /* at least part of the chunk is inside this vrange */
3834 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3835 struct btrfs_chunk *chunk,
3836 struct btrfs_balance_args *bargs)
3838 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3840 if (bargs->stripes_min <= num_stripes
3841 && num_stripes <= bargs->stripes_max)
3847 static int chunk_soft_convert_filter(u64 chunk_type,
3848 struct btrfs_balance_args *bargs)
3850 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3853 chunk_type = chunk_to_extended(chunk_type) &
3854 BTRFS_EXTENDED_PROFILE_MASK;
3856 if (bargs->target == chunk_type)
3862 static int should_balance_chunk(struct extent_buffer *leaf,
3863 struct btrfs_chunk *chunk, u64 chunk_offset)
3865 struct btrfs_fs_info *fs_info = leaf->fs_info;
3866 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3867 struct btrfs_balance_args *bargs = NULL;
3868 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3871 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3872 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3876 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3877 bargs = &bctl->data;
3878 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3880 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3881 bargs = &bctl->meta;
3883 /* profiles filter */
3884 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3885 chunk_profiles_filter(chunk_type, bargs)) {
3890 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3891 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3893 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3894 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3899 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3900 chunk_devid_filter(leaf, chunk, bargs)) {
3904 /* drange filter, makes sense only with devid filter */
3905 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3906 chunk_drange_filter(leaf, chunk, bargs)) {
3911 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3912 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3916 /* stripes filter */
3917 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3918 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3922 /* soft profile changing mode */
3923 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3924 chunk_soft_convert_filter(chunk_type, bargs)) {
3929 * limited by count, must be the last filter
3931 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3932 if (bargs->limit == 0)
3936 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3938 * Same logic as the 'limit' filter; the minimum cannot be
3939 * determined here because we do not have the global information
3940 * about the count of all chunks that satisfy the filters.
3942 if (bargs->limit_max == 0)
3951 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3953 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3954 struct btrfs_root *chunk_root = fs_info->chunk_root;
3956 struct btrfs_chunk *chunk;
3957 struct btrfs_path *path = NULL;
3958 struct btrfs_key key;
3959 struct btrfs_key found_key;
3960 struct extent_buffer *leaf;
3963 int enospc_errors = 0;
3964 bool counting = true;
3965 /* The single value limit and min/max limits use the same bytes in the */
3966 u64 limit_data = bctl->data.limit;
3967 u64 limit_meta = bctl->meta.limit;
3968 u64 limit_sys = bctl->sys.limit;
3972 int chunk_reserved = 0;
3974 path = btrfs_alloc_path();
3980 /* zero out stat counters */
3981 spin_lock(&fs_info->balance_lock);
3982 memset(&bctl->stat, 0, sizeof(bctl->stat));
3983 spin_unlock(&fs_info->balance_lock);
3987 * The single value limit and min/max limits use the same bytes
3990 bctl->data.limit = limit_data;
3991 bctl->meta.limit = limit_meta;
3992 bctl->sys.limit = limit_sys;
3994 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3995 key.offset = (u64)-1;
3996 key.type = BTRFS_CHUNK_ITEM_KEY;
3999 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
4000 atomic_read(&fs_info->balance_cancel_req)) {
4005 mutex_lock(&fs_info->reclaim_bgs_lock);
4006 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
4008 mutex_unlock(&fs_info->reclaim_bgs_lock);
4013 * this shouldn't happen, it means the last relocate
4017 BUG(); /* FIXME break ? */
4019 ret = btrfs_previous_item(chunk_root, path, 0,
4020 BTRFS_CHUNK_ITEM_KEY);
4022 mutex_unlock(&fs_info->reclaim_bgs_lock);
4027 leaf = path->nodes[0];
4028 slot = path->slots[0];
4029 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4031 if (found_key.objectid != key.objectid) {
4032 mutex_unlock(&fs_info->reclaim_bgs_lock);
4036 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
4037 chunk_type = btrfs_chunk_type(leaf, chunk);
4040 spin_lock(&fs_info->balance_lock);
4041 bctl->stat.considered++;
4042 spin_unlock(&fs_info->balance_lock);
4045 ret = should_balance_chunk(leaf, chunk, found_key.offset);
4047 btrfs_release_path(path);
4049 mutex_unlock(&fs_info->reclaim_bgs_lock);
4054 mutex_unlock(&fs_info->reclaim_bgs_lock);
4055 spin_lock(&fs_info->balance_lock);
4056 bctl->stat.expected++;
4057 spin_unlock(&fs_info->balance_lock);
4059 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
4061 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
4063 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
4070 * Apply limit_min filter, no need to check if the LIMITS
4071 * filter is used, limit_min is 0 by default
4073 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
4074 count_data < bctl->data.limit_min)
4075 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
4076 count_meta < bctl->meta.limit_min)
4077 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
4078 count_sys < bctl->sys.limit_min)) {
4079 mutex_unlock(&fs_info->reclaim_bgs_lock);
4083 if (!chunk_reserved) {
4085 * We may be relocating the only data chunk we have,
4086 * which could potentially end up with losing data's
4087 * raid profile, so lets allocate an empty one in
4090 ret = btrfs_may_alloc_data_chunk(fs_info,
4093 mutex_unlock(&fs_info->reclaim_bgs_lock);
4095 } else if (ret == 1) {
4100 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
4101 mutex_unlock(&fs_info->reclaim_bgs_lock);
4102 if (ret == -ENOSPC) {
4104 } else if (ret == -ETXTBSY) {
4106 "skipping relocation of block group %llu due to active swapfile",
4112 spin_lock(&fs_info->balance_lock);
4113 bctl->stat.completed++;
4114 spin_unlock(&fs_info->balance_lock);
4117 if (found_key.offset == 0)
4119 key.offset = found_key.offset - 1;
4123 btrfs_release_path(path);
4128 btrfs_free_path(path);
4129 if (enospc_errors) {
4130 btrfs_info(fs_info, "%d enospc errors during balance",
4140 * See if a given profile is valid and reduced.
4142 * @flags: profile to validate
4143 * @extended: if true @flags is treated as an extended profile
4145 static int alloc_profile_is_valid(u64 flags, int extended)
4147 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
4148 BTRFS_BLOCK_GROUP_PROFILE_MASK);
4150 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
4152 /* 1) check that all other bits are zeroed */
4156 /* 2) see if profile is reduced */
4158 return !extended; /* "0" is valid for usual profiles */
4160 return has_single_bit_set(flags);
4164 * Validate target profile against allowed profiles and return true if it's OK.
4165 * Otherwise print the error message and return false.
4167 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
4168 const struct btrfs_balance_args *bargs,
4169 u64 allowed, const char *type)
4171 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
4174 /* Profile is valid and does not have bits outside of the allowed set */
4175 if (alloc_profile_is_valid(bargs->target, 1) &&
4176 (bargs->target & ~allowed) == 0)
4179 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
4180 type, btrfs_bg_type_to_raid_name(bargs->target));
4185 * Fill @buf with textual description of balance filter flags @bargs, up to
4186 * @size_buf including the terminating null. The output may be trimmed if it
4187 * does not fit into the provided buffer.
4189 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
4193 u32 size_bp = size_buf;
4195 u64 flags = bargs->flags;
4196 char tmp_buf[128] = {'\0'};
4201 #define CHECK_APPEND_NOARG(a) \
4203 ret = snprintf(bp, size_bp, (a)); \
4204 if (ret < 0 || ret >= size_bp) \
4205 goto out_overflow; \
4210 #define CHECK_APPEND_1ARG(a, v1) \
4212 ret = snprintf(bp, size_bp, (a), (v1)); \
4213 if (ret < 0 || ret >= size_bp) \
4214 goto out_overflow; \
4219 #define CHECK_APPEND_2ARG(a, v1, v2) \
4221 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4222 if (ret < 0 || ret >= size_bp) \
4223 goto out_overflow; \
4228 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
4229 CHECK_APPEND_1ARG("convert=%s,",
4230 btrfs_bg_type_to_raid_name(bargs->target));
4232 if (flags & BTRFS_BALANCE_ARGS_SOFT)
4233 CHECK_APPEND_NOARG("soft,");
4235 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4236 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4238 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4241 if (flags & BTRFS_BALANCE_ARGS_USAGE)
4242 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4244 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4245 CHECK_APPEND_2ARG("usage=%u..%u,",
4246 bargs->usage_min, bargs->usage_max);
4248 if (flags & BTRFS_BALANCE_ARGS_DEVID)
4249 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4251 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4252 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4253 bargs->pstart, bargs->pend);
4255 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4256 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4257 bargs->vstart, bargs->vend);
4259 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4260 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4262 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4263 CHECK_APPEND_2ARG("limit=%u..%u,",
4264 bargs->limit_min, bargs->limit_max);
4266 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4267 CHECK_APPEND_2ARG("stripes=%u..%u,",
4268 bargs->stripes_min, bargs->stripes_max);
4270 #undef CHECK_APPEND_2ARG
4271 #undef CHECK_APPEND_1ARG
4272 #undef CHECK_APPEND_NOARG
4276 if (size_bp < size_buf)
4277 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4282 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4284 u32 size_buf = 1024;
4285 char tmp_buf[192] = {'\0'};
4288 u32 size_bp = size_buf;
4290 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4292 buf = kzalloc(size_buf, GFP_KERNEL);
4298 #define CHECK_APPEND_1ARG(a, v1) \
4300 ret = snprintf(bp, size_bp, (a), (v1)); \
4301 if (ret < 0 || ret >= size_bp) \
4302 goto out_overflow; \
4307 if (bctl->flags & BTRFS_BALANCE_FORCE)
4308 CHECK_APPEND_1ARG("%s", "-f ");
4310 if (bctl->flags & BTRFS_BALANCE_DATA) {
4311 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4312 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4315 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4316 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4317 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4320 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4321 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4322 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4325 #undef CHECK_APPEND_1ARG
4329 if (size_bp < size_buf)
4330 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4331 btrfs_info(fs_info, "balance: %s %s",
4332 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4333 "resume" : "start", buf);
4339 * Should be called with balance mutexe held
4341 int btrfs_balance(struct btrfs_fs_info *fs_info,
4342 struct btrfs_balance_control *bctl,
4343 struct btrfs_ioctl_balance_args *bargs)
4345 u64 meta_target, data_target;
4351 bool reducing_redundancy;
4352 bool paused = false;
4355 if (btrfs_fs_closing(fs_info) ||
4356 atomic_read(&fs_info->balance_pause_req) ||
4357 btrfs_should_cancel_balance(fs_info)) {
4362 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4363 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4367 * In case of mixed groups both data and meta should be picked,
4368 * and identical options should be given for both of them.
4370 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4371 if (mixed && (bctl->flags & allowed)) {
4372 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4373 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4374 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4376 "balance: mixed groups data and metadata options must be the same");
4383 * rw_devices will not change at the moment, device add/delete/replace
4386 num_devices = fs_info->fs_devices->rw_devices;
4389 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4390 * special bit for it, to make it easier to distinguish. Thus we need
4391 * to set it manually, or balance would refuse the profile.
4393 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4394 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4395 if (num_devices >= btrfs_raid_array[i].devs_min)
4396 allowed |= btrfs_raid_array[i].bg_flag;
4398 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4399 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4400 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4406 * Allow to reduce metadata or system integrity only if force set for
4407 * profiles with redundancy (copies, parity)
4410 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4411 if (btrfs_raid_array[i].ncopies >= 2 ||
4412 btrfs_raid_array[i].tolerated_failures >= 1)
4413 allowed |= btrfs_raid_array[i].bg_flag;
4416 seq = read_seqbegin(&fs_info->profiles_lock);
4418 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4419 (fs_info->avail_system_alloc_bits & allowed) &&
4420 !(bctl->sys.target & allowed)) ||
4421 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4422 (fs_info->avail_metadata_alloc_bits & allowed) &&
4423 !(bctl->meta.target & allowed)))
4424 reducing_redundancy = true;
4426 reducing_redundancy = false;
4428 /* if we're not converting, the target field is uninitialized */
4429 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4430 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4431 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4432 bctl->data.target : fs_info->avail_data_alloc_bits;
4433 } while (read_seqretry(&fs_info->profiles_lock, seq));
4435 if (reducing_redundancy) {
4436 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4438 "balance: force reducing metadata redundancy");
4441 "balance: reduces metadata redundancy, use --force if you want this");
4447 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4448 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4450 "balance: metadata profile %s has lower redundancy than data profile %s",
4451 btrfs_bg_type_to_raid_name(meta_target),
4452 btrfs_bg_type_to_raid_name(data_target));
4455 ret = insert_balance_item(fs_info, bctl);
4456 if (ret && ret != -EEXIST)
4459 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4460 BUG_ON(ret == -EEXIST);
4461 BUG_ON(fs_info->balance_ctl);
4462 spin_lock(&fs_info->balance_lock);
4463 fs_info->balance_ctl = bctl;
4464 spin_unlock(&fs_info->balance_lock);
4466 BUG_ON(ret != -EEXIST);
4467 spin_lock(&fs_info->balance_lock);
4468 update_balance_args(bctl);
4469 spin_unlock(&fs_info->balance_lock);
4472 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4473 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4474 describe_balance_start_or_resume(fs_info);
4475 mutex_unlock(&fs_info->balance_mutex);
4477 ret = __btrfs_balance(fs_info);
4479 mutex_lock(&fs_info->balance_mutex);
4480 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) {
4481 btrfs_info(fs_info, "balance: paused");
4482 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
4486 * Balance can be canceled by:
4488 * - Regular cancel request
4489 * Then ret == -ECANCELED and balance_cancel_req > 0
4491 * - Fatal signal to "btrfs" process
4492 * Either the signal caught by wait_reserve_ticket() and callers
4493 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4495 * Either way, in this case balance_cancel_req = 0, and
4496 * ret == -EINTR or ret == -ECANCELED.
4498 * So here we only check the return value to catch canceled balance.
4500 else if (ret == -ECANCELED || ret == -EINTR)
4501 btrfs_info(fs_info, "balance: canceled");
4503 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4505 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4508 memset(bargs, 0, sizeof(*bargs));
4509 btrfs_update_ioctl_balance_args(fs_info, bargs);
4512 /* We didn't pause, we can clean everything up. */
4514 reset_balance_state(fs_info);
4515 btrfs_exclop_finish(fs_info);
4518 wake_up(&fs_info->balance_wait_q);
4522 if (bctl->flags & BTRFS_BALANCE_RESUME)
4523 reset_balance_state(fs_info);
4526 btrfs_exclop_finish(fs_info);
4531 static int balance_kthread(void *data)
4533 struct btrfs_fs_info *fs_info = data;
4536 sb_start_write(fs_info->sb);
4537 mutex_lock(&fs_info->balance_mutex);
4538 if (fs_info->balance_ctl)
4539 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4540 mutex_unlock(&fs_info->balance_mutex);
4541 sb_end_write(fs_info->sb);
4546 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4548 struct task_struct *tsk;
4550 mutex_lock(&fs_info->balance_mutex);
4551 if (!fs_info->balance_ctl) {
4552 mutex_unlock(&fs_info->balance_mutex);
4555 mutex_unlock(&fs_info->balance_mutex);
4557 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4558 btrfs_info(fs_info, "balance: resume skipped");
4562 spin_lock(&fs_info->super_lock);
4563 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
4564 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;
4565 spin_unlock(&fs_info->super_lock);
4567 * A ro->rw remount sequence should continue with the paused balance
4568 * regardless of who pauses it, system or the user as of now, so set
4571 spin_lock(&fs_info->balance_lock);
4572 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4573 spin_unlock(&fs_info->balance_lock);
4575 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4576 return PTR_ERR_OR_ZERO(tsk);
4579 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4581 struct btrfs_balance_control *bctl;
4582 struct btrfs_balance_item *item;
4583 struct btrfs_disk_balance_args disk_bargs;
4584 struct btrfs_path *path;
4585 struct extent_buffer *leaf;
4586 struct btrfs_key key;
4589 path = btrfs_alloc_path();
4593 key.objectid = BTRFS_BALANCE_OBJECTID;
4594 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4597 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4600 if (ret > 0) { /* ret = -ENOENT; */
4605 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4611 leaf = path->nodes[0];
4612 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4614 bctl->flags = btrfs_balance_flags(leaf, item);
4615 bctl->flags |= BTRFS_BALANCE_RESUME;
4617 btrfs_balance_data(leaf, item, &disk_bargs);
4618 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4619 btrfs_balance_meta(leaf, item, &disk_bargs);
4620 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4621 btrfs_balance_sys(leaf, item, &disk_bargs);
4622 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4625 * This should never happen, as the paused balance state is recovered
4626 * during mount without any chance of other exclusive ops to collide.
4628 * This gives the exclusive op status to balance and keeps in paused
4629 * state until user intervention (cancel or umount). If the ownership
4630 * cannot be assigned, show a message but do not fail. The balance
4631 * is in a paused state and must have fs_info::balance_ctl properly
4634 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED))
4636 "balance: cannot set exclusive op status, resume manually");
4638 btrfs_release_path(path);
4640 mutex_lock(&fs_info->balance_mutex);
4641 BUG_ON(fs_info->balance_ctl);
4642 spin_lock(&fs_info->balance_lock);
4643 fs_info->balance_ctl = bctl;
4644 spin_unlock(&fs_info->balance_lock);
4645 mutex_unlock(&fs_info->balance_mutex);
4647 btrfs_free_path(path);
4651 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4655 mutex_lock(&fs_info->balance_mutex);
4656 if (!fs_info->balance_ctl) {
4657 mutex_unlock(&fs_info->balance_mutex);
4661 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4662 atomic_inc(&fs_info->balance_pause_req);
4663 mutex_unlock(&fs_info->balance_mutex);
4665 wait_event(fs_info->balance_wait_q,
4666 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4668 mutex_lock(&fs_info->balance_mutex);
4669 /* we are good with balance_ctl ripped off from under us */
4670 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4671 atomic_dec(&fs_info->balance_pause_req);
4676 mutex_unlock(&fs_info->balance_mutex);
4680 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4682 mutex_lock(&fs_info->balance_mutex);
4683 if (!fs_info->balance_ctl) {
4684 mutex_unlock(&fs_info->balance_mutex);
4689 * A paused balance with the item stored on disk can be resumed at
4690 * mount time if the mount is read-write. Otherwise it's still paused
4691 * and we must not allow cancelling as it deletes the item.
4693 if (sb_rdonly(fs_info->sb)) {
4694 mutex_unlock(&fs_info->balance_mutex);
4698 atomic_inc(&fs_info->balance_cancel_req);
4700 * if we are running just wait and return, balance item is
4701 * deleted in btrfs_balance in this case
4703 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4704 mutex_unlock(&fs_info->balance_mutex);
4705 wait_event(fs_info->balance_wait_q,
4706 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4707 mutex_lock(&fs_info->balance_mutex);
4709 mutex_unlock(&fs_info->balance_mutex);
4711 * Lock released to allow other waiters to continue, we'll
4712 * reexamine the status again.
4714 mutex_lock(&fs_info->balance_mutex);
4716 if (fs_info->balance_ctl) {
4717 reset_balance_state(fs_info);
4718 btrfs_exclop_finish(fs_info);
4719 btrfs_info(fs_info, "balance: canceled");
4723 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4724 atomic_dec(&fs_info->balance_cancel_req);
4725 mutex_unlock(&fs_info->balance_mutex);
4729 int btrfs_uuid_scan_kthread(void *data)
4731 struct btrfs_fs_info *fs_info = data;
4732 struct btrfs_root *root = fs_info->tree_root;
4733 struct btrfs_key key;
4734 struct btrfs_path *path = NULL;
4736 struct extent_buffer *eb;
4738 struct btrfs_root_item root_item;
4740 struct btrfs_trans_handle *trans = NULL;
4741 bool closing = false;
4743 path = btrfs_alloc_path();
4750 key.type = BTRFS_ROOT_ITEM_KEY;
4754 if (btrfs_fs_closing(fs_info)) {
4758 ret = btrfs_search_forward(root, &key, path,
4759 BTRFS_OLDEST_GENERATION);
4766 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4767 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4768 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4769 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4772 eb = path->nodes[0];
4773 slot = path->slots[0];
4774 item_size = btrfs_item_size(eb, slot);
4775 if (item_size < sizeof(root_item))
4778 read_extent_buffer(eb, &root_item,
4779 btrfs_item_ptr_offset(eb, slot),
4780 (int)sizeof(root_item));
4781 if (btrfs_root_refs(&root_item) == 0)
4784 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4785 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4789 btrfs_release_path(path);
4791 * 1 - subvol uuid item
4792 * 1 - received_subvol uuid item
4794 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4795 if (IS_ERR(trans)) {
4796 ret = PTR_ERR(trans);
4804 btrfs_release_path(path);
4805 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4806 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4807 BTRFS_UUID_KEY_SUBVOL,
4810 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4816 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4817 ret = btrfs_uuid_tree_add(trans,
4818 root_item.received_uuid,
4819 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4822 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4829 btrfs_release_path(path);
4831 ret = btrfs_end_transaction(trans);
4837 if (key.offset < (u64)-1) {
4839 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4841 key.type = BTRFS_ROOT_ITEM_KEY;
4842 } else if (key.objectid < (u64)-1) {
4844 key.type = BTRFS_ROOT_ITEM_KEY;
4853 btrfs_free_path(path);
4854 if (trans && !IS_ERR(trans))
4855 btrfs_end_transaction(trans);
4857 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4859 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4860 up(&fs_info->uuid_tree_rescan_sem);
4864 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4866 struct btrfs_trans_handle *trans;
4867 struct btrfs_root *tree_root = fs_info->tree_root;
4868 struct btrfs_root *uuid_root;
4869 struct task_struct *task;
4876 trans = btrfs_start_transaction(tree_root, 2);
4878 return PTR_ERR(trans);
4880 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4881 if (IS_ERR(uuid_root)) {
4882 ret = PTR_ERR(uuid_root);
4883 btrfs_abort_transaction(trans, ret);
4884 btrfs_end_transaction(trans);
4888 fs_info->uuid_root = uuid_root;
4890 ret = btrfs_commit_transaction(trans);
4894 down(&fs_info->uuid_tree_rescan_sem);
4895 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4897 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4898 btrfs_warn(fs_info, "failed to start uuid_scan task");
4899 up(&fs_info->uuid_tree_rescan_sem);
4900 return PTR_ERR(task);
4907 * shrinking a device means finding all of the device extents past
4908 * the new size, and then following the back refs to the chunks.
4909 * The chunk relocation code actually frees the device extent
4911 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4913 struct btrfs_fs_info *fs_info = device->fs_info;
4914 struct btrfs_root *root = fs_info->dev_root;
4915 struct btrfs_trans_handle *trans;
4916 struct btrfs_dev_extent *dev_extent = NULL;
4917 struct btrfs_path *path;
4923 bool retried = false;
4924 struct extent_buffer *l;
4925 struct btrfs_key key;
4926 struct btrfs_super_block *super_copy = fs_info->super_copy;
4927 u64 old_total = btrfs_super_total_bytes(super_copy);
4928 u64 old_size = btrfs_device_get_total_bytes(device);
4933 new_size = round_down(new_size, fs_info->sectorsize);
4935 diff = round_down(old_size - new_size, fs_info->sectorsize);
4937 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4940 path = btrfs_alloc_path();
4944 path->reada = READA_BACK;
4946 trans = btrfs_start_transaction(root, 0);
4947 if (IS_ERR(trans)) {
4948 btrfs_free_path(path);
4949 return PTR_ERR(trans);
4952 mutex_lock(&fs_info->chunk_mutex);
4954 btrfs_device_set_total_bytes(device, new_size);
4955 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4956 device->fs_devices->total_rw_bytes -= diff;
4959 * The new free_chunk_space is new_size - used, so we have to
4960 * subtract the delta of the old free_chunk_space which included
4961 * old_size - used. If used > new_size then just subtract this
4962 * entire device's free space.
4964 if (device->bytes_used < new_size)
4965 free_diff = (old_size - device->bytes_used) -
4966 (new_size - device->bytes_used);
4968 free_diff = old_size - device->bytes_used;
4969 atomic64_sub(free_diff, &fs_info->free_chunk_space);
4973 * Once the device's size has been set to the new size, ensure all
4974 * in-memory chunks are synced to disk so that the loop below sees them
4975 * and relocates them accordingly.
4977 if (contains_pending_extent(device, &start, diff)) {
4978 mutex_unlock(&fs_info->chunk_mutex);
4979 ret = btrfs_commit_transaction(trans);
4983 mutex_unlock(&fs_info->chunk_mutex);
4984 btrfs_end_transaction(trans);
4988 key.objectid = device->devid;
4989 key.offset = (u64)-1;
4990 key.type = BTRFS_DEV_EXTENT_KEY;
4993 mutex_lock(&fs_info->reclaim_bgs_lock);
4994 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4996 mutex_unlock(&fs_info->reclaim_bgs_lock);
5000 ret = btrfs_previous_item(root, path, 0, key.type);
5002 mutex_unlock(&fs_info->reclaim_bgs_lock);
5006 btrfs_release_path(path);
5011 slot = path->slots[0];
5012 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
5014 if (key.objectid != device->devid) {
5015 mutex_unlock(&fs_info->reclaim_bgs_lock);
5016 btrfs_release_path(path);
5020 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
5021 length = btrfs_dev_extent_length(l, dev_extent);
5023 if (key.offset + length <= new_size) {
5024 mutex_unlock(&fs_info->reclaim_bgs_lock);
5025 btrfs_release_path(path);
5029 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
5030 btrfs_release_path(path);
5033 * We may be relocating the only data chunk we have,
5034 * which could potentially end up with losing data's
5035 * raid profile, so lets allocate an empty one in
5038 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
5040 mutex_unlock(&fs_info->reclaim_bgs_lock);
5044 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
5045 mutex_unlock(&fs_info->reclaim_bgs_lock);
5046 if (ret == -ENOSPC) {
5049 if (ret == -ETXTBSY) {
5051 "could not shrink block group %llu due to active swapfile",
5056 } while (key.offset-- > 0);
5058 if (failed && !retried) {
5062 } else if (failed && retried) {
5067 /* Shrinking succeeded, else we would be at "done". */
5068 trans = btrfs_start_transaction(root, 0);
5069 if (IS_ERR(trans)) {
5070 ret = PTR_ERR(trans);
5074 mutex_lock(&fs_info->chunk_mutex);
5075 /* Clear all state bits beyond the shrunk device size */
5076 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
5079 btrfs_device_set_disk_total_bytes(device, new_size);
5080 if (list_empty(&device->post_commit_list))
5081 list_add_tail(&device->post_commit_list,
5082 &trans->transaction->dev_update_list);
5084 WARN_ON(diff > old_total);
5085 btrfs_set_super_total_bytes(super_copy,
5086 round_down(old_total - diff, fs_info->sectorsize));
5087 mutex_unlock(&fs_info->chunk_mutex);
5089 btrfs_reserve_chunk_metadata(trans, false);
5090 /* Now btrfs_update_device() will change the on-disk size. */
5091 ret = btrfs_update_device(trans, device);
5092 btrfs_trans_release_chunk_metadata(trans);
5094 btrfs_abort_transaction(trans, ret);
5095 btrfs_end_transaction(trans);
5097 ret = btrfs_commit_transaction(trans);
5100 btrfs_free_path(path);
5102 mutex_lock(&fs_info->chunk_mutex);
5103 btrfs_device_set_total_bytes(device, old_size);
5104 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5105 device->fs_devices->total_rw_bytes += diff;
5106 atomic64_add(free_diff, &fs_info->free_chunk_space);
5108 mutex_unlock(&fs_info->chunk_mutex);
5113 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
5114 struct btrfs_key *key,
5115 struct btrfs_chunk *chunk, int item_size)
5117 struct btrfs_super_block *super_copy = fs_info->super_copy;
5118 struct btrfs_disk_key disk_key;
5122 lockdep_assert_held(&fs_info->chunk_mutex);
5124 array_size = btrfs_super_sys_array_size(super_copy);
5125 if (array_size + item_size + sizeof(disk_key)
5126 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
5129 ptr = super_copy->sys_chunk_array + array_size;
5130 btrfs_cpu_key_to_disk(&disk_key, key);
5131 memcpy(ptr, &disk_key, sizeof(disk_key));
5132 ptr += sizeof(disk_key);
5133 memcpy(ptr, chunk, item_size);
5134 item_size += sizeof(disk_key);
5135 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
5141 * sort the devices in descending order by max_avail, total_avail
5143 static int btrfs_cmp_device_info(const void *a, const void *b)
5145 const struct btrfs_device_info *di_a = a;
5146 const struct btrfs_device_info *di_b = b;
5148 if (di_a->max_avail > di_b->max_avail)
5150 if (di_a->max_avail < di_b->max_avail)
5152 if (di_a->total_avail > di_b->total_avail)
5154 if (di_a->total_avail < di_b->total_avail)
5159 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
5161 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5164 btrfs_set_fs_incompat(info, RAID56);
5167 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
5169 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
5172 btrfs_set_fs_incompat(info, RAID1C34);
5176 * Structure used internally for btrfs_create_chunk() function.
5177 * Wraps needed parameters.
5179 struct alloc_chunk_ctl {
5182 /* Total number of stripes to allocate */
5184 /* sub_stripes info for map */
5186 /* Stripes per device */
5188 /* Maximum number of devices to use */
5190 /* Minimum number of devices to use */
5192 /* ndevs has to be a multiple of this */
5194 /* Number of copies */
5196 /* Number of stripes worth of bytes to store parity information */
5198 u64 max_stripe_size;
5206 static void init_alloc_chunk_ctl_policy_regular(
5207 struct btrfs_fs_devices *fs_devices,
5208 struct alloc_chunk_ctl *ctl)
5210 struct btrfs_space_info *space_info;
5212 space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type);
5215 ctl->max_chunk_size = READ_ONCE(space_info->chunk_size);
5216 ctl->max_stripe_size = min_t(u64, ctl->max_chunk_size, SZ_1G);
5218 if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM)
5219 ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK);
5221 /* We don't want a chunk larger than 10% of writable space */
5222 ctl->max_chunk_size = min(mult_perc(fs_devices->total_rw_bytes, 10),
5223 ctl->max_chunk_size);
5224 ctl->dev_extent_min = btrfs_stripe_nr_to_offset(ctl->dev_stripes);
5227 static void init_alloc_chunk_ctl_policy_zoned(
5228 struct btrfs_fs_devices *fs_devices,
5229 struct alloc_chunk_ctl *ctl)
5231 u64 zone_size = fs_devices->fs_info->zone_size;
5233 int min_num_stripes = ctl->devs_min * ctl->dev_stripes;
5234 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;
5235 u64 min_chunk_size = min_data_stripes * zone_size;
5236 u64 type = ctl->type;
5238 ctl->max_stripe_size = zone_size;
5239 if (type & BTRFS_BLOCK_GROUP_DATA) {
5240 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,
5242 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5243 ctl->max_chunk_size = ctl->max_stripe_size;
5244 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5245 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5246 ctl->devs_max = min_t(int, ctl->devs_max,
5247 BTRFS_MAX_DEVS_SYS_CHUNK);
5252 /* We don't want a chunk larger than 10% of writable space */
5253 limit = max(round_down(mult_perc(fs_devices->total_rw_bytes, 10),
5256 ctl->max_chunk_size = min(limit, ctl->max_chunk_size);
5257 ctl->dev_extent_min = zone_size * ctl->dev_stripes;
5260 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
5261 struct alloc_chunk_ctl *ctl)
5263 int index = btrfs_bg_flags_to_raid_index(ctl->type);
5265 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
5266 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
5267 ctl->devs_max = btrfs_raid_array[index].devs_max;
5269 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
5270 ctl->devs_min = btrfs_raid_array[index].devs_min;
5271 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
5272 ctl->ncopies = btrfs_raid_array[index].ncopies;
5273 ctl->nparity = btrfs_raid_array[index].nparity;
5276 switch (fs_devices->chunk_alloc_policy) {
5277 case BTRFS_CHUNK_ALLOC_REGULAR:
5278 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
5280 case BTRFS_CHUNK_ALLOC_ZONED:
5281 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);
5288 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
5289 struct alloc_chunk_ctl *ctl,
5290 struct btrfs_device_info *devices_info)
5292 struct btrfs_fs_info *info = fs_devices->fs_info;
5293 struct btrfs_device *device;
5295 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
5302 * in the first pass through the devices list, we gather information
5303 * about the available holes on each device.
5305 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5306 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5308 "BTRFS: read-only device in alloc_list\n");
5312 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5313 &device->dev_state) ||
5314 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5317 if (device->total_bytes > device->bytes_used)
5318 total_avail = device->total_bytes - device->bytes_used;
5322 /* If there is no space on this device, skip it. */
5323 if (total_avail < ctl->dev_extent_min)
5326 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
5328 if (ret && ret != -ENOSPC)
5332 max_avail = dev_extent_want;
5334 if (max_avail < ctl->dev_extent_min) {
5335 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5337 "%s: devid %llu has no free space, have=%llu want=%llu",
5338 __func__, device->devid, max_avail,
5339 ctl->dev_extent_min);
5343 if (ndevs == fs_devices->rw_devices) {
5344 WARN(1, "%s: found more than %llu devices\n",
5345 __func__, fs_devices->rw_devices);
5348 devices_info[ndevs].dev_offset = dev_offset;
5349 devices_info[ndevs].max_avail = max_avail;
5350 devices_info[ndevs].total_avail = total_avail;
5351 devices_info[ndevs].dev = device;
5357 * now sort the devices by hole size / available space
5359 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5360 btrfs_cmp_device_info, NULL);
5365 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5366 struct btrfs_device_info *devices_info)
5368 /* Number of stripes that count for block group size */
5372 * The primary goal is to maximize the number of stripes, so use as
5373 * many devices as possible, even if the stripes are not maximum sized.
5375 * The DUP profile stores more than one stripe per device, the
5376 * max_avail is the total size so we have to adjust.
5378 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5380 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5382 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5383 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5386 * Use the number of data stripes to figure out how big this chunk is
5387 * really going to be in terms of logical address space, and compare
5388 * that answer with the max chunk size. If it's higher, we try to
5389 * reduce stripe_size.
5391 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5393 * Reduce stripe_size, round it up to a 16MB boundary again and
5394 * then use it, unless it ends up being even bigger than the
5395 * previous value we had already.
5397 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5398 data_stripes), SZ_16M),
5402 /* Stripe size should not go beyond 1G. */
5403 ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G);
5405 /* Align to BTRFS_STRIPE_LEN */
5406 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5407 ctl->chunk_size = ctl->stripe_size * data_stripes;
5412 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,
5413 struct btrfs_device_info *devices_info)
5415 u64 zone_size = devices_info[0].dev->zone_info->zone_size;
5416 /* Number of stripes that count for block group size */
5420 * It should hold because:
5421 * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5423 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);
5425 ctl->stripe_size = zone_size;
5426 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5427 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5429 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5430 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5431 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,
5432 ctl->stripe_size) + ctl->nparity,
5434 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5435 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5436 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);
5439 ctl->chunk_size = ctl->stripe_size * data_stripes;
5444 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5445 struct alloc_chunk_ctl *ctl,
5446 struct btrfs_device_info *devices_info)
5448 struct btrfs_fs_info *info = fs_devices->fs_info;
5451 * Round down to number of usable stripes, devs_increment can be any
5452 * number so we can't use round_down() that requires power of 2, while
5453 * rounddown is safe.
5455 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5457 if (ctl->ndevs < ctl->devs_min) {
5458 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5460 "%s: not enough devices with free space: have=%d minimum required=%d",
5461 __func__, ctl->ndevs, ctl->devs_min);
5466 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5468 switch (fs_devices->chunk_alloc_policy) {
5469 case BTRFS_CHUNK_ALLOC_REGULAR:
5470 return decide_stripe_size_regular(ctl, devices_info);
5471 case BTRFS_CHUNK_ALLOC_ZONED:
5472 return decide_stripe_size_zoned(ctl, devices_info);
5478 static void chunk_map_device_set_bits(struct btrfs_chunk_map *map, unsigned int bits)
5480 for (int i = 0; i < map->num_stripes; i++) {
5481 struct btrfs_io_stripe *stripe = &map->stripes[i];
5482 struct btrfs_device *device = stripe->dev;
5484 set_extent_bit(&device->alloc_state, stripe->physical,
5485 stripe->physical + map->stripe_size - 1,
5486 bits | EXTENT_NOWAIT, NULL);
5490 static void chunk_map_device_clear_bits(struct btrfs_chunk_map *map, unsigned int bits)
5492 for (int i = 0; i < map->num_stripes; i++) {
5493 struct btrfs_io_stripe *stripe = &map->stripes[i];
5494 struct btrfs_device *device = stripe->dev;
5496 __clear_extent_bit(&device->alloc_state, stripe->physical,
5497 stripe->physical + map->stripe_size - 1,
5498 bits | EXTENT_NOWAIT,
5503 void btrfs_remove_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map)
5505 write_lock(&fs_info->mapping_tree_lock);
5506 rb_erase_cached(&map->rb_node, &fs_info->mapping_tree);
5507 RB_CLEAR_NODE(&map->rb_node);
5508 chunk_map_device_clear_bits(map, CHUNK_ALLOCATED);
5509 write_unlock(&fs_info->mapping_tree_lock);
5511 /* Once for the tree reference. */
5512 btrfs_free_chunk_map(map);
5516 int btrfs_add_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map)
5519 struct rb_node *parent = NULL;
5520 bool leftmost = true;
5522 write_lock(&fs_info->mapping_tree_lock);
5523 p = &fs_info->mapping_tree.rb_root.rb_node;
5525 struct btrfs_chunk_map *entry;
5528 entry = rb_entry(parent, struct btrfs_chunk_map, rb_node);
5530 if (map->start < entry->start) {
5532 } else if (map->start > entry->start) {
5533 p = &(*p)->rb_right;
5536 write_unlock(&fs_info->mapping_tree_lock);
5540 rb_link_node(&map->rb_node, parent, p);
5541 rb_insert_color_cached(&map->rb_node, &fs_info->mapping_tree, leftmost);
5542 chunk_map_device_set_bits(map, CHUNK_ALLOCATED);
5543 chunk_map_device_clear_bits(map, CHUNK_TRIMMED);
5544 write_unlock(&fs_info->mapping_tree_lock);
5550 struct btrfs_chunk_map *btrfs_alloc_chunk_map(int num_stripes, gfp_t gfp)
5552 struct btrfs_chunk_map *map;
5554 map = kmalloc(btrfs_chunk_map_size(num_stripes), gfp);
5558 refcount_set(&map->refs, 1);
5559 RB_CLEAR_NODE(&map->rb_node);
5564 struct btrfs_chunk_map *btrfs_clone_chunk_map(struct btrfs_chunk_map *map, gfp_t gfp)
5566 const int size = btrfs_chunk_map_size(map->num_stripes);
5567 struct btrfs_chunk_map *clone;
5569 clone = kmemdup(map, size, gfp);
5573 refcount_set(&clone->refs, 1);
5574 RB_CLEAR_NODE(&clone->rb_node);
5579 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
5580 struct alloc_chunk_ctl *ctl,
5581 struct btrfs_device_info *devices_info)
5583 struct btrfs_fs_info *info = trans->fs_info;
5584 struct btrfs_chunk_map *map;
5585 struct btrfs_block_group *block_group;
5586 u64 start = ctl->start;
5587 u64 type = ctl->type;
5592 map = btrfs_alloc_chunk_map(ctl->num_stripes, GFP_NOFS);
5594 return ERR_PTR(-ENOMEM);
5597 map->chunk_len = ctl->chunk_size;
5598 map->stripe_size = ctl->stripe_size;
5600 map->io_align = BTRFS_STRIPE_LEN;
5601 map->io_width = BTRFS_STRIPE_LEN;
5602 map->sub_stripes = ctl->sub_stripes;
5603 map->num_stripes = ctl->num_stripes;
5605 for (i = 0; i < ctl->ndevs; ++i) {
5606 for (j = 0; j < ctl->dev_stripes; ++j) {
5607 int s = i * ctl->dev_stripes + j;
5608 map->stripes[s].dev = devices_info[i].dev;
5609 map->stripes[s].physical = devices_info[i].dev_offset +
5610 j * ctl->stripe_size;
5614 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5616 ret = btrfs_add_chunk_map(info, map);
5618 btrfs_free_chunk_map(map);
5619 return ERR_PTR(ret);
5622 block_group = btrfs_make_block_group(trans, type, start, ctl->chunk_size);
5623 if (IS_ERR(block_group)) {
5624 btrfs_remove_chunk_map(info, map);
5628 for (int i = 0; i < map->num_stripes; i++) {
5629 struct btrfs_device *dev = map->stripes[i].dev;
5631 btrfs_device_set_bytes_used(dev,
5632 dev->bytes_used + ctl->stripe_size);
5633 if (list_empty(&dev->post_commit_list))
5634 list_add_tail(&dev->post_commit_list,
5635 &trans->transaction->dev_update_list);
5638 atomic64_sub(ctl->stripe_size * map->num_stripes,
5639 &info->free_chunk_space);
5641 check_raid56_incompat_flag(info, type);
5642 check_raid1c34_incompat_flag(info, type);
5647 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
5650 struct btrfs_fs_info *info = trans->fs_info;
5651 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5652 struct btrfs_device_info *devices_info = NULL;
5653 struct alloc_chunk_ctl ctl;
5654 struct btrfs_block_group *block_group;
5657 lockdep_assert_held(&info->chunk_mutex);
5659 if (!alloc_profile_is_valid(type, 0)) {
5661 return ERR_PTR(-EINVAL);
5664 if (list_empty(&fs_devices->alloc_list)) {
5665 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5666 btrfs_debug(info, "%s: no writable device", __func__);
5667 return ERR_PTR(-ENOSPC);
5670 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5671 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5673 return ERR_PTR(-EINVAL);
5676 ctl.start = find_next_chunk(info);
5678 init_alloc_chunk_ctl(fs_devices, &ctl);
5680 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5683 return ERR_PTR(-ENOMEM);
5685 ret = gather_device_info(fs_devices, &ctl, devices_info);
5687 block_group = ERR_PTR(ret);
5691 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5693 block_group = ERR_PTR(ret);
5697 block_group = create_chunk(trans, &ctl, devices_info);
5700 kfree(devices_info);
5705 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5706 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5709 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5712 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
5713 struct btrfs_block_group *bg)
5715 struct btrfs_fs_info *fs_info = trans->fs_info;
5716 struct btrfs_root *chunk_root = fs_info->chunk_root;
5717 struct btrfs_key key;
5718 struct btrfs_chunk *chunk;
5719 struct btrfs_stripe *stripe;
5720 struct btrfs_chunk_map *map;
5726 * We take the chunk_mutex for 2 reasons:
5728 * 1) Updates and insertions in the chunk btree must be done while holding
5729 * the chunk_mutex, as well as updating the system chunk array in the
5730 * superblock. See the comment on top of btrfs_chunk_alloc() for the
5733 * 2) To prevent races with the final phase of a device replace operation
5734 * that replaces the device object associated with the map's stripes,
5735 * because the device object's id can change at any time during that
5736 * final phase of the device replace operation
5737 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5738 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5739 * which would cause a failure when updating the device item, which does
5740 * not exists, or persisting a stripe of the chunk item with such ID.
5741 * Here we can't use the device_list_mutex because our caller already
5742 * has locked the chunk_mutex, and the final phase of device replace
5743 * acquires both mutexes - first the device_list_mutex and then the
5744 * chunk_mutex. Using any of those two mutexes protects us from a
5745 * concurrent device replace.
5747 lockdep_assert_held(&fs_info->chunk_mutex);
5749 map = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
5752 btrfs_abort_transaction(trans, ret);
5756 item_size = btrfs_chunk_item_size(map->num_stripes);
5758 chunk = kzalloc(item_size, GFP_NOFS);
5761 btrfs_abort_transaction(trans, ret);
5765 for (i = 0; i < map->num_stripes; i++) {
5766 struct btrfs_device *device = map->stripes[i].dev;
5768 ret = btrfs_update_device(trans, device);
5773 stripe = &chunk->stripe;
5774 for (i = 0; i < map->num_stripes; i++) {
5775 struct btrfs_device *device = map->stripes[i].dev;
5776 const u64 dev_offset = map->stripes[i].physical;
5778 btrfs_set_stack_stripe_devid(stripe, device->devid);
5779 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5780 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5784 btrfs_set_stack_chunk_length(chunk, bg->length);
5785 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
5786 btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
5787 btrfs_set_stack_chunk_type(chunk, map->type);
5788 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5789 btrfs_set_stack_chunk_io_align(chunk, BTRFS_STRIPE_LEN);
5790 btrfs_set_stack_chunk_io_width(chunk, BTRFS_STRIPE_LEN);
5791 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5792 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5794 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5795 key.type = BTRFS_CHUNK_ITEM_KEY;
5796 key.offset = bg->start;
5798 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5802 set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags);
5804 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5805 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5812 btrfs_free_chunk_map(map);
5816 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5818 struct btrfs_fs_info *fs_info = trans->fs_info;
5820 struct btrfs_block_group *meta_bg;
5821 struct btrfs_block_group *sys_bg;
5824 * When adding a new device for sprouting, the seed device is read-only
5825 * so we must first allocate a metadata and a system chunk. But before
5826 * adding the block group items to the extent, device and chunk btrees,
5829 * 1) Create both chunks without doing any changes to the btrees, as
5830 * otherwise we would get -ENOSPC since the block groups from the
5831 * seed device are read-only;
5833 * 2) Add the device item for the new sprout device - finishing the setup
5834 * of a new block group requires updating the device item in the chunk
5835 * btree, so it must exist when we attempt to do it. The previous step
5836 * ensures this does not fail with -ENOSPC.
5838 * After that we can add the block group items to their btrees:
5839 * update existing device item in the chunk btree, add a new block group
5840 * item to the extent btree, add a new chunk item to the chunk btree and
5841 * finally add the new device extent items to the devices btree.
5844 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5845 meta_bg = btrfs_create_chunk(trans, alloc_profile);
5846 if (IS_ERR(meta_bg))
5847 return PTR_ERR(meta_bg);
5849 alloc_profile = btrfs_system_alloc_profile(fs_info);
5850 sys_bg = btrfs_create_chunk(trans, alloc_profile);
5852 return PTR_ERR(sys_bg);
5857 static inline int btrfs_chunk_max_errors(struct btrfs_chunk_map *map)
5859 const int index = btrfs_bg_flags_to_raid_index(map->type);
5861 return btrfs_raid_array[index].tolerated_failures;
5864 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5866 struct btrfs_chunk_map *map;
5871 map = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5875 for (i = 0; i < map->num_stripes; i++) {
5876 if (test_bit(BTRFS_DEV_STATE_MISSING,
5877 &map->stripes[i].dev->dev_state)) {
5881 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5882 &map->stripes[i].dev->dev_state)) {
5889 * If the number of missing devices is larger than max errors, we can
5890 * not write the data into that chunk successfully.
5892 if (miss_ndevs > btrfs_chunk_max_errors(map))
5895 btrfs_free_chunk_map(map);
5899 void btrfs_mapping_tree_free(struct btrfs_fs_info *fs_info)
5901 write_lock(&fs_info->mapping_tree_lock);
5902 while (!RB_EMPTY_ROOT(&fs_info->mapping_tree.rb_root)) {
5903 struct btrfs_chunk_map *map;
5904 struct rb_node *node;
5906 node = rb_first_cached(&fs_info->mapping_tree);
5907 map = rb_entry(node, struct btrfs_chunk_map, rb_node);
5908 rb_erase_cached(&map->rb_node, &fs_info->mapping_tree);
5909 RB_CLEAR_NODE(&map->rb_node);
5910 chunk_map_device_clear_bits(map, CHUNK_ALLOCATED);
5911 /* Once for the tree ref. */
5912 btrfs_free_chunk_map(map);
5913 cond_resched_rwlock_write(&fs_info->mapping_tree_lock);
5915 write_unlock(&fs_info->mapping_tree_lock);
5918 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5920 struct btrfs_chunk_map *map;
5921 enum btrfs_raid_types index;
5924 map = btrfs_get_chunk_map(fs_info, logical, len);
5927 * We could return errors for these cases, but that could get
5928 * ugly and we'd probably do the same thing which is just not do
5929 * anything else and exit, so return 1 so the callers don't try
5930 * to use other copies.
5934 index = btrfs_bg_flags_to_raid_index(map->type);
5936 /* Non-RAID56, use their ncopies from btrfs_raid_array. */
5937 if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5938 ret = btrfs_raid_array[index].ncopies;
5939 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5941 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5943 * There could be two corrupted data stripes, we need
5944 * to loop retry in order to rebuild the correct data.
5946 * Fail a stripe at a time on every retry except the
5947 * stripe under reconstruction.
5949 ret = map->num_stripes;
5950 btrfs_free_chunk_map(map);
5954 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5957 struct btrfs_chunk_map *map;
5958 unsigned long len = fs_info->sectorsize;
5960 if (!btrfs_fs_incompat(fs_info, RAID56))
5963 map = btrfs_get_chunk_map(fs_info, logical, len);
5965 if (!WARN_ON(IS_ERR(map))) {
5966 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5967 len = btrfs_stripe_nr_to_offset(nr_data_stripes(map));
5968 btrfs_free_chunk_map(map);
5973 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5975 struct btrfs_chunk_map *map;
5978 if (!btrfs_fs_incompat(fs_info, RAID56))
5981 map = btrfs_get_chunk_map(fs_info, logical, len);
5983 if (!WARN_ON(IS_ERR(map))) {
5984 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5986 btrfs_free_chunk_map(map);
5991 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5992 struct btrfs_chunk_map *map, int first,
5993 int dev_replace_is_ongoing)
5995 const enum btrfs_read_policy policy = READ_ONCE(fs_info->fs_devices->read_policy);
5998 int preferred_mirror;
6000 struct btrfs_device *srcdev;
6003 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
6005 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
6006 num_stripes = map->sub_stripes;
6008 num_stripes = map->num_stripes;
6012 /* Shouldn't happen, just warn and use pid instead of failing */
6013 btrfs_warn_rl(fs_info, "unknown read_policy type %u, reset to pid",
6015 WRITE_ONCE(fs_info->fs_devices->read_policy, BTRFS_READ_POLICY_PID);
6017 case BTRFS_READ_POLICY_PID:
6018 preferred_mirror = first + (current->pid % num_stripes);
6022 if (dev_replace_is_ongoing &&
6023 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
6024 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
6025 srcdev = fs_info->dev_replace.srcdev;
6030 * try to avoid the drive that is the source drive for a
6031 * dev-replace procedure, only choose it if no other non-missing
6032 * mirror is available
6034 for (tolerance = 0; tolerance < 2; tolerance++) {
6035 if (map->stripes[preferred_mirror].dev->bdev &&
6036 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
6037 return preferred_mirror;
6038 for (i = first; i < first + num_stripes; i++) {
6039 if (map->stripes[i].dev->bdev &&
6040 (tolerance || map->stripes[i].dev != srcdev))
6045 /* we couldn't find one that doesn't fail. Just return something
6046 * and the io error handling code will clean up eventually
6048 return preferred_mirror;
6051 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info,
6055 struct btrfs_io_context *bioc;
6058 /* The size of btrfs_io_context */
6059 sizeof(struct btrfs_io_context) +
6060 /* Plus the variable array for the stripes */
6061 sizeof(struct btrfs_io_stripe) * (total_stripes),
6067 refcount_set(&bioc->refs, 1);
6069 bioc->fs_info = fs_info;
6070 bioc->replace_stripe_src = -1;
6071 bioc->full_stripe_logical = (u64)-1;
6072 bioc->logical = logical;
6077 void btrfs_get_bioc(struct btrfs_io_context *bioc)
6079 WARN_ON(!refcount_read(&bioc->refs));
6080 refcount_inc(&bioc->refs);
6083 void btrfs_put_bioc(struct btrfs_io_context *bioc)
6087 if (refcount_dec_and_test(&bioc->refs))
6092 * Please note that, discard won't be sent to target device of device
6095 struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info,
6096 u64 logical, u64 *length_ret,
6099 struct btrfs_chunk_map *map;
6100 struct btrfs_discard_stripe *stripes;
6101 u64 length = *length_ret;
6106 u64 stripe_end_offset;
6110 u32 sub_stripes = 0;
6111 u32 stripes_per_dev = 0;
6112 u32 remaining_stripes = 0;
6113 u32 last_stripe = 0;
6117 map = btrfs_get_chunk_map(fs_info, logical, length);
6119 return ERR_CAST(map);
6121 /* we don't discard raid56 yet */
6122 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6127 offset = logical - map->start;
6128 length = min_t(u64, map->start + map->chunk_len - logical, length);
6129 *length_ret = length;
6132 * stripe_nr counts the total number of stripes we have to stride
6133 * to get to this block
6135 stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
6137 /* stripe_offset is the offset of this block in its stripe */
6138 stripe_offset = offset - btrfs_stripe_nr_to_offset(stripe_nr);
6140 stripe_nr_end = round_up(offset + length, BTRFS_STRIPE_LEN) >>
6141 BTRFS_STRIPE_LEN_SHIFT;
6142 stripe_cnt = stripe_nr_end - stripe_nr;
6143 stripe_end_offset = btrfs_stripe_nr_to_offset(stripe_nr_end) -
6146 * after this, stripe_nr is the number of stripes on this
6147 * device we have to walk to find the data, and stripe_index is
6148 * the number of our device in the stripe array
6152 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6153 BTRFS_BLOCK_GROUP_RAID10)) {
6154 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6157 sub_stripes = map->sub_stripes;
6159 factor = map->num_stripes / sub_stripes;
6160 *num_stripes = min_t(u64, map->num_stripes,
6161 sub_stripes * stripe_cnt);
6162 stripe_index = stripe_nr % factor;
6163 stripe_nr /= factor;
6164 stripe_index *= sub_stripes;
6166 remaining_stripes = stripe_cnt % factor;
6167 stripes_per_dev = stripe_cnt / factor;
6168 last_stripe = ((stripe_nr_end - 1) % factor) * sub_stripes;
6169 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
6170 BTRFS_BLOCK_GROUP_DUP)) {
6171 *num_stripes = map->num_stripes;
6173 stripe_index = stripe_nr % map->num_stripes;
6174 stripe_nr /= map->num_stripes;
6177 stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS);
6183 for (i = 0; i < *num_stripes; i++) {
6184 stripes[i].physical =
6185 map->stripes[stripe_index].physical +
6186 stripe_offset + btrfs_stripe_nr_to_offset(stripe_nr);
6187 stripes[i].dev = map->stripes[stripe_index].dev;
6189 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6190 BTRFS_BLOCK_GROUP_RAID10)) {
6191 stripes[i].length = btrfs_stripe_nr_to_offset(stripes_per_dev);
6193 if (i / sub_stripes < remaining_stripes)
6194 stripes[i].length += BTRFS_STRIPE_LEN;
6197 * Special for the first stripe and
6200 * |-------|...|-------|
6204 if (i < sub_stripes)
6205 stripes[i].length -= stripe_offset;
6207 if (stripe_index >= last_stripe &&
6208 stripe_index <= (last_stripe +
6210 stripes[i].length -= stripe_end_offset;
6212 if (i == sub_stripes - 1)
6215 stripes[i].length = length;
6219 if (stripe_index == map->num_stripes) {
6225 btrfs_free_chunk_map(map);
6228 btrfs_free_chunk_map(map);
6229 return ERR_PTR(ret);
6232 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)
6234 struct btrfs_block_group *cache;
6237 /* Non zoned filesystem does not use "to_copy" flag */
6238 if (!btrfs_is_zoned(fs_info))
6241 cache = btrfs_lookup_block_group(fs_info, logical);
6243 ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags);
6245 btrfs_put_block_group(cache);
6249 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
6250 struct btrfs_io_context *bioc,
6251 struct btrfs_dev_replace *dev_replace,
6253 int *num_stripes_ret, int *max_errors_ret)
6255 u64 srcdev_devid = dev_replace->srcdev->devid;
6257 * At this stage, num_stripes is still the real number of stripes,
6258 * excluding the duplicated stripes.
6260 int num_stripes = *num_stripes_ret;
6261 int nr_extra_stripes = 0;
6262 int max_errors = *max_errors_ret;
6266 * A block group which has "to_copy" set will eventually be copied by
6267 * the dev-replace process. We can avoid cloning IO here.
6269 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))
6273 * Duplicate the write operations while the dev-replace procedure is
6274 * running. Since the copying of the old disk to the new disk takes
6275 * place at run time while the filesystem is mounted writable, the
6276 * regular write operations to the old disk have to be duplicated to go
6277 * to the new disk as well.
6279 * Note that device->missing is handled by the caller, and that the
6280 * write to the old disk is already set up in the stripes array.
6282 for (i = 0; i < num_stripes; i++) {
6283 struct btrfs_io_stripe *old = &bioc->stripes[i];
6284 struct btrfs_io_stripe *new = &bioc->stripes[num_stripes + nr_extra_stripes];
6286 if (old->dev->devid != srcdev_devid)
6289 new->physical = old->physical;
6290 new->dev = dev_replace->tgtdev;
6291 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK)
6292 bioc->replace_stripe_src = i;
6296 /* We can only have at most 2 extra nr_stripes (for DUP). */
6297 ASSERT(nr_extra_stripes <= 2);
6299 * For GET_READ_MIRRORS, we can only return at most 1 extra stripe for
6301 * If we have 2 extra stripes, only choose the one with smaller physical.
6303 if (op == BTRFS_MAP_GET_READ_MIRRORS && nr_extra_stripes == 2) {
6304 struct btrfs_io_stripe *first = &bioc->stripes[num_stripes];
6305 struct btrfs_io_stripe *second = &bioc->stripes[num_stripes + 1];
6307 /* Only DUP can have two extra stripes. */
6308 ASSERT(bioc->map_type & BTRFS_BLOCK_GROUP_DUP);
6311 * Swap the last stripe stripes and reduce @nr_extra_stripes.
6312 * The extra stripe would still be there, but won't be accessed.
6314 if (first->physical > second->physical) {
6315 swap(second->physical, first->physical);
6316 swap(second->dev, first->dev);
6321 *num_stripes_ret = num_stripes + nr_extra_stripes;
6322 *max_errors_ret = max_errors + nr_extra_stripes;
6323 bioc->replace_nr_stripes = nr_extra_stripes;
6326 static u64 btrfs_max_io_len(struct btrfs_chunk_map *map, u64 offset,
6327 struct btrfs_io_geometry *io_geom)
6330 * Stripe_nr is the stripe where this block falls. stripe_offset is
6331 * the offset of this block in its stripe.
6333 io_geom->stripe_offset = offset & BTRFS_STRIPE_LEN_MASK;
6334 io_geom->stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
6335 ASSERT(io_geom->stripe_offset < U32_MAX);
6337 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6338 unsigned long full_stripe_len =
6339 btrfs_stripe_nr_to_offset(nr_data_stripes(map));
6342 * For full stripe start, we use previously calculated
6343 * @stripe_nr. Align it to nr_data_stripes, then multiply with
6346 * By this we can avoid u64 division completely. And we have
6347 * to go rounddown(), not round_down(), as nr_data_stripes is
6348 * not ensured to be power of 2.
6350 io_geom->raid56_full_stripe_start = btrfs_stripe_nr_to_offset(
6351 rounddown(io_geom->stripe_nr, nr_data_stripes(map)));
6353 ASSERT(io_geom->raid56_full_stripe_start + full_stripe_len > offset);
6354 ASSERT(io_geom->raid56_full_stripe_start <= offset);
6356 * For writes to RAID56, allow to write a full stripe set, but
6357 * no straddling of stripe sets.
6359 if (io_geom->op == BTRFS_MAP_WRITE)
6360 return full_stripe_len - (offset - io_geom->raid56_full_stripe_start);
6364 * For other RAID types and for RAID56 reads, allow a single stripe (on
6367 if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK)
6368 return BTRFS_STRIPE_LEN - io_geom->stripe_offset;
6372 static int set_io_stripe(struct btrfs_fs_info *fs_info, u64 logical,
6373 u64 *length, struct btrfs_io_stripe *dst,
6374 struct btrfs_chunk_map *map,
6375 struct btrfs_io_geometry *io_geom)
6377 dst->dev = map->stripes[io_geom->stripe_index].dev;
6379 if (io_geom->op == BTRFS_MAP_READ &&
6380 btrfs_need_stripe_tree_update(fs_info, map->type))
6381 return btrfs_get_raid_extent_offset(fs_info, logical, length,
6383 io_geom->stripe_index, dst);
6385 dst->physical = map->stripes[io_geom->stripe_index].physical +
6386 io_geom->stripe_offset +
6387 btrfs_stripe_nr_to_offset(io_geom->stripe_nr);
6391 static bool is_single_device_io(struct btrfs_fs_info *fs_info,
6392 const struct btrfs_io_stripe *smap,
6393 const struct btrfs_chunk_map *map,
6394 int num_alloc_stripes,
6395 enum btrfs_map_op op, int mirror_num)
6400 if (num_alloc_stripes != 1)
6403 if (btrfs_need_stripe_tree_update(fs_info, map->type) && op != BTRFS_MAP_READ)
6406 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1)
6412 static void map_blocks_raid0(const struct btrfs_chunk_map *map,
6413 struct btrfs_io_geometry *io_geom)
6415 io_geom->stripe_index = io_geom->stripe_nr % map->num_stripes;
6416 io_geom->stripe_nr /= map->num_stripes;
6417 if (io_geom->op == BTRFS_MAP_READ)
6418 io_geom->mirror_num = 1;
6421 static void map_blocks_raid1(struct btrfs_fs_info *fs_info,
6422 struct btrfs_chunk_map *map,
6423 struct btrfs_io_geometry *io_geom,
6424 bool dev_replace_is_ongoing)
6426 if (io_geom->op != BTRFS_MAP_READ) {
6427 io_geom->num_stripes = map->num_stripes;
6431 if (io_geom->mirror_num) {
6432 io_geom->stripe_index = io_geom->mirror_num - 1;
6436 io_geom->stripe_index = find_live_mirror(fs_info, map, 0,
6437 dev_replace_is_ongoing);
6438 io_geom->mirror_num = io_geom->stripe_index + 1;
6441 static void map_blocks_dup(const struct btrfs_chunk_map *map,
6442 struct btrfs_io_geometry *io_geom)
6444 if (io_geom->op != BTRFS_MAP_READ) {
6445 io_geom->num_stripes = map->num_stripes;
6449 if (io_geom->mirror_num) {
6450 io_geom->stripe_index = io_geom->mirror_num - 1;
6454 io_geom->mirror_num = 1;
6457 static void map_blocks_raid10(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 u32 factor = map->num_stripes / map->sub_stripes;
6463 int old_stripe_index;
6465 io_geom->stripe_index = (io_geom->stripe_nr % factor) * map->sub_stripes;
6466 io_geom->stripe_nr /= factor;
6468 if (io_geom->op != BTRFS_MAP_READ) {
6469 io_geom->num_stripes = map->sub_stripes;
6473 if (io_geom->mirror_num) {
6474 io_geom->stripe_index += io_geom->mirror_num - 1;
6478 old_stripe_index = io_geom->stripe_index;
6479 io_geom->stripe_index = find_live_mirror(fs_info, map,
6480 io_geom->stripe_index,
6481 dev_replace_is_ongoing);
6482 io_geom->mirror_num = io_geom->stripe_index - old_stripe_index + 1;
6485 static void map_blocks_raid56_write(struct btrfs_chunk_map *map,
6486 struct btrfs_io_geometry *io_geom,
6487 u64 logical, u64 *length)
6489 int data_stripes = nr_data_stripes(map);
6492 * Needs full stripe mapping.
6494 * Push stripe_nr back to the start of the full stripe For those cases
6495 * needing a full stripe, @stripe_nr is the full stripe number.
6497 * Originally we go raid56_full_stripe_start / full_stripe_len, but
6498 * that can be expensive. Here we just divide @stripe_nr with
6501 io_geom->stripe_nr /= data_stripes;
6503 /* RAID[56] write or recovery. Return all stripes */
6504 io_geom->num_stripes = map->num_stripes;
6505 io_geom->max_errors = btrfs_chunk_max_errors(map);
6507 /* Return the length to the full stripe end. */
6508 *length = min(logical + *length,
6509 io_geom->raid56_full_stripe_start + map->start +
6510 btrfs_stripe_nr_to_offset(data_stripes)) -
6512 io_geom->stripe_index = 0;
6513 io_geom->stripe_offset = 0;
6516 static void map_blocks_raid56_read(struct btrfs_chunk_map *map,
6517 struct btrfs_io_geometry *io_geom)
6519 int data_stripes = nr_data_stripes(map);
6521 ASSERT(io_geom->mirror_num <= 1);
6522 /* Just grab the data stripe directly. */
6523 io_geom->stripe_index = io_geom->stripe_nr % data_stripes;
6524 io_geom->stripe_nr /= data_stripes;
6526 /* We distribute the parity blocks across stripes. */
6527 io_geom->stripe_index =
6528 (io_geom->stripe_nr + io_geom->stripe_index) % map->num_stripes;
6530 if (io_geom->op == BTRFS_MAP_READ && io_geom->mirror_num < 1)
6531 io_geom->mirror_num = 1;
6534 static void map_blocks_single(const struct btrfs_chunk_map *map,
6535 struct btrfs_io_geometry *io_geom)
6537 io_geom->stripe_index = io_geom->stripe_nr % map->num_stripes;
6538 io_geom->stripe_nr /= map->num_stripes;
6539 io_geom->mirror_num = io_geom->stripe_index + 1;
6543 * Map one logical range to one or more physical ranges.
6545 * @length: (Mandatory) mapped length of this run.
6546 * One logical range can be split into different segments
6547 * due to factors like zones and RAID0/5/6/10 stripe
6550 * @bioc_ret: (Mandatory) returned btrfs_io_context structure.
6551 * which has one or more physical ranges (btrfs_io_stripe)
6553 * Caller should call btrfs_put_bioc() to free it after use.
6555 * @smap: (Optional) single physical range optimization.
6556 * If the map request can be fulfilled by one single
6557 * physical range, and this is parameter is not NULL,
6558 * then @bioc_ret would be NULL, and @smap would be
6561 * @mirror_num_ret: (Mandatory) returned mirror number if the original
6564 * Mirror number 0 means to choose any live mirrors.
6566 * For non-RAID56 profiles, non-zero mirror_num means
6567 * the Nth mirror. (e.g. mirror_num 1 means the first
6570 * For RAID56 profile, mirror 1 means rebuild from P and
6571 * the remaining data stripes.
6573 * For RAID6 profile, mirror > 2 means mark another
6574 * data/P stripe error and rebuild from the remaining
6577 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6578 u64 logical, u64 *length,
6579 struct btrfs_io_context **bioc_ret,
6580 struct btrfs_io_stripe *smap, int *mirror_num_ret)
6582 struct btrfs_chunk_map *map;
6583 struct btrfs_io_geometry io_geom = { 0 };
6588 struct btrfs_io_context *bioc = NULL;
6589 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6590 int dev_replace_is_ongoing = 0;
6591 u16 num_alloc_stripes;
6596 io_geom.mirror_num = (mirror_num_ret ? *mirror_num_ret : 0);
6597 io_geom.num_stripes = 1;
6598 io_geom.stripe_index = 0;
6601 num_copies = btrfs_num_copies(fs_info, logical, fs_info->sectorsize);
6602 if (io_geom.mirror_num > num_copies)
6605 map = btrfs_get_chunk_map(fs_info, logical, *length);
6607 return PTR_ERR(map);
6609 map_offset = logical - map->start;
6610 io_geom.raid56_full_stripe_start = (u64)-1;
6611 max_len = btrfs_max_io_len(map, map_offset, &io_geom);
6612 *length = min_t(u64, map->chunk_len - map_offset, max_len);
6614 down_read(&dev_replace->rwsem);
6615 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6617 * Hold the semaphore for read during the whole operation, write is
6618 * requested at commit time but must wait.
6620 if (!dev_replace_is_ongoing)
6621 up_read(&dev_replace->rwsem);
6623 switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
6624 case BTRFS_BLOCK_GROUP_RAID0:
6625 map_blocks_raid0(map, &io_geom);
6627 case BTRFS_BLOCK_GROUP_RAID1:
6628 case BTRFS_BLOCK_GROUP_RAID1C3:
6629 case BTRFS_BLOCK_GROUP_RAID1C4:
6630 map_blocks_raid1(fs_info, map, &io_geom, dev_replace_is_ongoing);
6632 case BTRFS_BLOCK_GROUP_DUP:
6633 map_blocks_dup(map, &io_geom);
6635 case BTRFS_BLOCK_GROUP_RAID10:
6636 map_blocks_raid10(fs_info, map, &io_geom, dev_replace_is_ongoing);
6638 case BTRFS_BLOCK_GROUP_RAID5:
6639 case BTRFS_BLOCK_GROUP_RAID6:
6640 if (op != BTRFS_MAP_READ || io_geom.mirror_num > 1)
6641 map_blocks_raid56_write(map, &io_geom, logical, length);
6643 map_blocks_raid56_read(map, &io_geom);
6647 * After this, stripe_nr is the number of stripes on this
6648 * device we have to walk to find the data, and stripe_index is
6649 * the number of our device in the stripe array
6651 map_blocks_single(map, &io_geom);
6654 if (io_geom.stripe_index >= map->num_stripes) {
6656 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6657 io_geom.stripe_index, map->num_stripes);
6662 num_alloc_stripes = io_geom.num_stripes;
6663 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6664 op != BTRFS_MAP_READ)
6666 * For replace case, we need to add extra stripes for extra
6667 * duplicated stripes.
6669 * For both WRITE and GET_READ_MIRRORS, we may have at most
6670 * 2 more stripes (DUP types, otherwise 1).
6672 num_alloc_stripes += 2;
6675 * If this I/O maps to a single device, try to return the device and
6676 * physical block information on the stack instead of allocating an
6677 * I/O context structure.
6679 if (is_single_device_io(fs_info, smap, map, num_alloc_stripes, op,
6680 io_geom.mirror_num)) {
6681 ret = set_io_stripe(fs_info, logical, length, smap, map, &io_geom);
6683 *mirror_num_ret = io_geom.mirror_num;
6688 bioc = alloc_btrfs_io_context(fs_info, logical, num_alloc_stripes);
6693 bioc->map_type = map->type;
6696 * For RAID56 full map, we need to make sure the stripes[] follows the
6697 * rule that data stripes are all ordered, then followed with P and Q
6700 * It's still mostly the same as other profiles, just with extra rotation.
6702 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
6703 (op != BTRFS_MAP_READ || io_geom.mirror_num > 1)) {
6705 * For RAID56 @stripe_nr is already the number of full stripes
6706 * before us, which is also the rotation value (needs to modulo
6707 * with num_stripes).
6709 * In this case, we just add @stripe_nr with @i, then do the
6710 * modulo, to reduce one modulo call.
6712 bioc->full_stripe_logical = map->start +
6713 btrfs_stripe_nr_to_offset(io_geom.stripe_nr *
6714 nr_data_stripes(map));
6715 for (int i = 0; i < io_geom.num_stripes; i++) {
6716 struct btrfs_io_stripe *dst = &bioc->stripes[i];
6719 stripe_index = (i + io_geom.stripe_nr) % io_geom.num_stripes;
6720 dst->dev = map->stripes[stripe_index].dev;
6722 map->stripes[stripe_index].physical +
6723 io_geom.stripe_offset +
6724 btrfs_stripe_nr_to_offset(io_geom.stripe_nr);
6728 * For all other non-RAID56 profiles, just copy the target
6729 * stripe into the bioc.
6731 for (i = 0; i < io_geom.num_stripes; i++) {
6732 ret = set_io_stripe(fs_info, logical, length,
6733 &bioc->stripes[i], map, &io_geom);
6736 io_geom.stripe_index++;
6742 btrfs_put_bioc(bioc);
6746 if (op != BTRFS_MAP_READ)
6747 io_geom.max_errors = btrfs_chunk_max_errors(map);
6749 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6750 op != BTRFS_MAP_READ) {
6751 handle_ops_on_dev_replace(op, bioc, dev_replace, logical,
6752 &io_geom.num_stripes, &io_geom.max_errors);
6756 bioc->num_stripes = io_geom.num_stripes;
6757 bioc->max_errors = io_geom.max_errors;
6758 bioc->mirror_num = io_geom.mirror_num;
6761 if (dev_replace_is_ongoing) {
6762 lockdep_assert_held(&dev_replace->rwsem);
6763 /* Unlock and let waiting writers proceed */
6764 up_read(&dev_replace->rwsem);
6766 btrfs_free_chunk_map(map);
6770 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args,
6771 const struct btrfs_fs_devices *fs_devices)
6773 if (args->fsid == NULL)
6775 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0)
6780 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args,
6781 const struct btrfs_device *device)
6783 if (args->missing) {
6784 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) &&
6790 if (device->devid != args->devid)
6792 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0)
6798 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6801 * If devid and uuid are both specified, the match must be exact, otherwise
6802 * only devid is used.
6804 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,
6805 const struct btrfs_dev_lookup_args *args)
6807 struct btrfs_device *device;
6808 struct btrfs_fs_devices *seed_devs;
6810 if (dev_args_match_fs_devices(args, fs_devices)) {
6811 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6812 if (dev_args_match_device(args, device))
6817 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6818 if (!dev_args_match_fs_devices(args, seed_devs))
6820 list_for_each_entry(device, &seed_devs->devices, dev_list) {
6821 if (dev_args_match_device(args, device))
6829 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6830 u64 devid, u8 *dev_uuid)
6832 struct btrfs_device *device;
6833 unsigned int nofs_flag;
6836 * We call this under the chunk_mutex, so we want to use NOFS for this
6837 * allocation, however we don't want to change btrfs_alloc_device() to
6838 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6842 nofs_flag = memalloc_nofs_save();
6843 device = btrfs_alloc_device(NULL, &devid, dev_uuid, NULL);
6844 memalloc_nofs_restore(nofs_flag);
6848 list_add(&device->dev_list, &fs_devices->devices);
6849 device->fs_devices = fs_devices;
6850 fs_devices->num_devices++;
6852 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6853 fs_devices->missing_devices++;
6859 * Allocate new device struct, set up devid and UUID.
6861 * @fs_info: used only for generating a new devid, can be NULL if
6862 * devid is provided (i.e. @devid != NULL).
6863 * @devid: a pointer to devid for this device. If NULL a new devid
6865 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6867 * @path: a pointer to device path if available, NULL otherwise.
6869 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6870 * on error. Returned struct is not linked onto any lists and must be
6871 * destroyed with btrfs_free_device.
6873 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6874 const u64 *devid, const u8 *uuid,
6877 struct btrfs_device *dev;
6880 if (WARN_ON(!devid && !fs_info))
6881 return ERR_PTR(-EINVAL);
6883 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
6885 return ERR_PTR(-ENOMEM);
6887 INIT_LIST_HEAD(&dev->dev_list);
6888 INIT_LIST_HEAD(&dev->dev_alloc_list);
6889 INIT_LIST_HEAD(&dev->post_commit_list);
6891 atomic_set(&dev->dev_stats_ccnt, 0);
6892 btrfs_device_data_ordered_init(dev);
6893 extent_io_tree_init(fs_info, &dev->alloc_state, IO_TREE_DEVICE_ALLOC_STATE);
6900 ret = find_next_devid(fs_info, &tmp);
6902 btrfs_free_device(dev);
6903 return ERR_PTR(ret);
6909 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6911 generate_random_uuid(dev->uuid);
6914 struct rcu_string *name;
6916 name = rcu_string_strdup(path, GFP_KERNEL);
6918 btrfs_free_device(dev);
6919 return ERR_PTR(-ENOMEM);
6921 rcu_assign_pointer(dev->name, name);
6927 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6928 u64 devid, u8 *uuid, bool error)
6931 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6934 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6938 u64 btrfs_calc_stripe_length(const struct btrfs_chunk_map *map)
6940 const int data_stripes = calc_data_stripes(map->type, map->num_stripes);
6942 return div_u64(map->chunk_len, data_stripes);
6945 #if BITS_PER_LONG == 32
6947 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
6948 * can't be accessed on 32bit systems.
6950 * This function do mount time check to reject the fs if it already has
6951 * metadata chunk beyond that limit.
6953 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6954 u64 logical, u64 length, u64 type)
6956 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6959 if (logical + length < MAX_LFS_FILESIZE)
6962 btrfs_err_32bit_limit(fs_info);
6967 * This is to give early warning for any metadata chunk reaching
6968 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
6969 * Although we can still access the metadata, it's not going to be possible
6970 * once the limit is reached.
6972 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6973 u64 logical, u64 length, u64 type)
6975 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6978 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)
6981 btrfs_warn_32bit_limit(fs_info);
6985 static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info,
6986 u64 devid, u8 *uuid)
6988 struct btrfs_device *dev;
6990 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6991 btrfs_report_missing_device(fs_info, devid, uuid, true);
6992 return ERR_PTR(-ENOENT);
6995 dev = add_missing_dev(fs_info->fs_devices, devid, uuid);
6997 btrfs_err(fs_info, "failed to init missing device %llu: %ld",
6998 devid, PTR_ERR(dev));
7001 btrfs_report_missing_device(fs_info, devid, uuid, false);
7006 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
7007 struct btrfs_chunk *chunk)
7009 BTRFS_DEV_LOOKUP_ARGS(args);
7010 struct btrfs_fs_info *fs_info = leaf->fs_info;
7011 struct btrfs_chunk_map *map;
7016 u8 uuid[BTRFS_UUID_SIZE];
7022 logical = key->offset;
7023 length = btrfs_chunk_length(leaf, chunk);
7024 type = btrfs_chunk_type(leaf, chunk);
7025 index = btrfs_bg_flags_to_raid_index(type);
7026 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
7028 #if BITS_PER_LONG == 32
7029 ret = check_32bit_meta_chunk(fs_info, logical, length, type);
7032 warn_32bit_meta_chunk(fs_info, logical, length, type);
7036 * Only need to verify chunk item if we're reading from sys chunk array,
7037 * as chunk item in tree block is already verified by tree-checker.
7039 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
7040 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
7045 map = btrfs_find_chunk_map(fs_info, logical, 1);
7047 /* already mapped? */
7048 if (map && map->start <= logical && map->start + map->chunk_len > logical) {
7049 btrfs_free_chunk_map(map);
7052 btrfs_free_chunk_map(map);
7055 map = btrfs_alloc_chunk_map(num_stripes, GFP_NOFS);
7059 map->start = logical;
7060 map->chunk_len = length;
7061 map->num_stripes = num_stripes;
7062 map->io_width = btrfs_chunk_io_width(leaf, chunk);
7063 map->io_align = btrfs_chunk_io_align(leaf, chunk);
7066 * We can't use the sub_stripes value, as for profiles other than
7067 * RAID10, they may have 0 as sub_stripes for filesystems created by
7068 * older mkfs (<v5.4).
7069 * In that case, it can cause divide-by-zero errors later.
7070 * Since currently sub_stripes is fixed for each profile, let's
7071 * use the trusted value instead.
7073 map->sub_stripes = btrfs_raid_array[index].sub_stripes;
7074 map->verified_stripes = 0;
7075 map->stripe_size = btrfs_calc_stripe_length(map);
7076 for (i = 0; i < num_stripes; i++) {
7077 map->stripes[i].physical =
7078 btrfs_stripe_offset_nr(leaf, chunk, i);
7079 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
7081 read_extent_buffer(leaf, uuid, (unsigned long)
7082 btrfs_stripe_dev_uuid_nr(chunk, i),
7085 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args);
7086 if (!map->stripes[i].dev) {
7087 map->stripes[i].dev = handle_missing_device(fs_info,
7089 if (IS_ERR(map->stripes[i].dev)) {
7090 ret = PTR_ERR(map->stripes[i].dev);
7091 btrfs_free_chunk_map(map);
7096 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
7097 &(map->stripes[i].dev->dev_state));
7100 ret = btrfs_add_chunk_map(fs_info, map);
7103 "failed to add chunk map, start=%llu len=%llu: %d",
7104 map->start, map->chunk_len, ret);
7110 static void fill_device_from_item(struct extent_buffer *leaf,
7111 struct btrfs_dev_item *dev_item,
7112 struct btrfs_device *device)
7116 device->devid = btrfs_device_id(leaf, dev_item);
7117 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
7118 device->total_bytes = device->disk_total_bytes;
7119 device->commit_total_bytes = device->disk_total_bytes;
7120 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
7121 device->commit_bytes_used = device->bytes_used;
7122 device->type = btrfs_device_type(leaf, dev_item);
7123 device->io_align = btrfs_device_io_align(leaf, dev_item);
7124 device->io_width = btrfs_device_io_width(leaf, dev_item);
7125 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
7126 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
7127 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
7129 ptr = btrfs_device_uuid(dev_item);
7130 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
7133 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
7136 struct btrfs_fs_devices *fs_devices;
7139 lockdep_assert_held(&uuid_mutex);
7142 /* This will match only for multi-device seed fs */
7143 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
7144 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
7148 fs_devices = find_fsid(fsid, NULL);
7150 if (!btrfs_test_opt(fs_info, DEGRADED))
7151 return ERR_PTR(-ENOENT);
7153 fs_devices = alloc_fs_devices(fsid);
7154 if (IS_ERR(fs_devices))
7157 fs_devices->seeding = true;
7158 fs_devices->opened = 1;
7163 * Upon first call for a seed fs fsid, just create a private copy of the
7164 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
7166 fs_devices = clone_fs_devices(fs_devices);
7167 if (IS_ERR(fs_devices))
7170 ret = open_fs_devices(fs_devices, BLK_OPEN_READ, fs_info->bdev_holder);
7172 free_fs_devices(fs_devices);
7173 return ERR_PTR(ret);
7176 if (!fs_devices->seeding) {
7177 close_fs_devices(fs_devices);
7178 free_fs_devices(fs_devices);
7179 return ERR_PTR(-EINVAL);
7182 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
7187 static int read_one_dev(struct extent_buffer *leaf,
7188 struct btrfs_dev_item *dev_item)
7190 BTRFS_DEV_LOOKUP_ARGS(args);
7191 struct btrfs_fs_info *fs_info = leaf->fs_info;
7192 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7193 struct btrfs_device *device;
7196 u8 fs_uuid[BTRFS_FSID_SIZE];
7197 u8 dev_uuid[BTRFS_UUID_SIZE];
7199 devid = btrfs_device_id(leaf, dev_item);
7201 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
7203 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
7205 args.uuid = dev_uuid;
7206 args.fsid = fs_uuid;
7208 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
7209 fs_devices = open_seed_devices(fs_info, fs_uuid);
7210 if (IS_ERR(fs_devices))
7211 return PTR_ERR(fs_devices);
7214 device = btrfs_find_device(fs_info->fs_devices, &args);
7216 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7217 btrfs_report_missing_device(fs_info, devid,
7222 device = add_missing_dev(fs_devices, devid, dev_uuid);
7223 if (IS_ERR(device)) {
7225 "failed to add missing dev %llu: %ld",
7226 devid, PTR_ERR(device));
7227 return PTR_ERR(device);
7229 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
7231 if (!device->bdev) {
7232 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7233 btrfs_report_missing_device(fs_info,
7234 devid, dev_uuid, true);
7237 btrfs_report_missing_device(fs_info, devid,
7241 if (!device->bdev &&
7242 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
7244 * this happens when a device that was properly setup
7245 * in the device info lists suddenly goes bad.
7246 * device->bdev is NULL, and so we have to set
7247 * device->missing to one here
7249 device->fs_devices->missing_devices++;
7250 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
7253 /* Move the device to its own fs_devices */
7254 if (device->fs_devices != fs_devices) {
7255 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
7256 &device->dev_state));
7258 list_move(&device->dev_list, &fs_devices->devices);
7259 device->fs_devices->num_devices--;
7260 fs_devices->num_devices++;
7262 device->fs_devices->missing_devices--;
7263 fs_devices->missing_devices++;
7265 device->fs_devices = fs_devices;
7269 if (device->fs_devices != fs_info->fs_devices) {
7270 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
7271 if (device->generation !=
7272 btrfs_device_generation(leaf, dev_item))
7276 fill_device_from_item(leaf, dev_item, device);
7278 u64 max_total_bytes = bdev_nr_bytes(device->bdev);
7280 if (device->total_bytes > max_total_bytes) {
7282 "device total_bytes should be at most %llu but found %llu",
7283 max_total_bytes, device->total_bytes);
7287 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7288 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7289 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7290 device->fs_devices->total_rw_bytes += device->total_bytes;
7291 atomic64_add(device->total_bytes - device->bytes_used,
7292 &fs_info->free_chunk_space);
7298 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7300 struct btrfs_super_block *super_copy = fs_info->super_copy;
7301 struct extent_buffer *sb;
7302 struct btrfs_disk_key *disk_key;
7303 struct btrfs_chunk *chunk;
7305 unsigned long sb_array_offset;
7312 struct btrfs_key key;
7314 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7317 * We allocated a dummy extent, just to use extent buffer accessors.
7318 * There will be unused space after BTRFS_SUPER_INFO_SIZE, but
7319 * that's fine, we will not go beyond system chunk array anyway.
7321 sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET);
7324 set_extent_buffer_uptodate(sb);
7326 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7327 array_size = btrfs_super_sys_array_size(super_copy);
7329 array_ptr = super_copy->sys_chunk_array;
7330 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7333 while (cur_offset < array_size) {
7334 disk_key = (struct btrfs_disk_key *)array_ptr;
7335 len = sizeof(*disk_key);
7336 if (cur_offset + len > array_size)
7337 goto out_short_read;
7339 btrfs_disk_key_to_cpu(&key, disk_key);
7342 sb_array_offset += len;
7345 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7347 "unexpected item type %u in sys_array at offset %u",
7348 (u32)key.type, cur_offset);
7353 chunk = (struct btrfs_chunk *)sb_array_offset;
7355 * At least one btrfs_chunk with one stripe must be present,
7356 * exact stripe count check comes afterwards
7358 len = btrfs_chunk_item_size(1);
7359 if (cur_offset + len > array_size)
7360 goto out_short_read;
7362 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7365 "invalid number of stripes %u in sys_array at offset %u",
7366 num_stripes, cur_offset);
7371 type = btrfs_chunk_type(sb, chunk);
7372 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7374 "invalid chunk type %llu in sys_array at offset %u",
7380 len = btrfs_chunk_item_size(num_stripes);
7381 if (cur_offset + len > array_size)
7382 goto out_short_read;
7384 ret = read_one_chunk(&key, sb, chunk);
7389 sb_array_offset += len;
7392 clear_extent_buffer_uptodate(sb);
7393 free_extent_buffer_stale(sb);
7397 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7399 clear_extent_buffer_uptodate(sb);
7400 free_extent_buffer_stale(sb);
7405 * Check if all chunks in the fs are OK for read-write degraded mount
7407 * If the @failing_dev is specified, it's accounted as missing.
7409 * Return true if all chunks meet the minimal RW mount requirements.
7410 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7412 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7413 struct btrfs_device *failing_dev)
7415 struct btrfs_chunk_map *map;
7419 map = btrfs_find_chunk_map(fs_info, 0, U64_MAX);
7420 /* No chunk at all? Return false anyway */
7431 btrfs_get_num_tolerated_disk_barrier_failures(
7433 for (i = 0; i < map->num_stripes; i++) {
7434 struct btrfs_device *dev = map->stripes[i].dev;
7436 if (!dev || !dev->bdev ||
7437 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7438 dev->last_flush_error)
7440 else if (failing_dev && failing_dev == dev)
7443 if (missing > max_tolerated) {
7446 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7447 map->start, missing, max_tolerated);
7448 btrfs_free_chunk_map(map);
7452 next_start = map->start + map->chunk_len;
7453 btrfs_free_chunk_map(map);
7455 map = btrfs_find_chunk_map(fs_info, next_start, U64_MAX - next_start);
7461 static void readahead_tree_node_children(struct extent_buffer *node)
7464 const int nr_items = btrfs_header_nritems(node);
7466 for (i = 0; i < nr_items; i++)
7467 btrfs_readahead_node_child(node, i);
7470 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7472 struct btrfs_root *root = fs_info->chunk_root;
7473 struct btrfs_path *path;
7474 struct extent_buffer *leaf;
7475 struct btrfs_key key;
7476 struct btrfs_key found_key;
7481 u64 last_ra_node = 0;
7483 path = btrfs_alloc_path();
7488 * uuid_mutex is needed only if we are mounting a sprout FS
7489 * otherwise we don't need it.
7491 mutex_lock(&uuid_mutex);
7494 * It is possible for mount and umount to race in such a way that
7495 * we execute this code path, but open_fs_devices failed to clear
7496 * total_rw_bytes. We certainly want it cleared before reading the
7497 * device items, so clear it here.
7499 fs_info->fs_devices->total_rw_bytes = 0;
7502 * Lockdep complains about possible circular locking dependency between
7503 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores
7504 * used for freeze procection of a fs (struct super_block.s_writers),
7505 * which we take when starting a transaction, and extent buffers of the
7506 * chunk tree if we call read_one_dev() while holding a lock on an
7507 * extent buffer of the chunk tree. Since we are mounting the filesystem
7508 * and at this point there can't be any concurrent task modifying the
7509 * chunk tree, to keep it simple, just skip locking on the chunk tree.
7511 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
7512 path->skip_locking = 1;
7515 * Read all device items, and then all the chunk items. All
7516 * device items are found before any chunk item (their object id
7517 * is smaller than the lowest possible object id for a chunk
7518 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7520 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7523 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) {
7524 struct extent_buffer *node = path->nodes[1];
7526 leaf = path->nodes[0];
7527 slot = path->slots[0];
7530 if (last_ra_node != node->start) {
7531 readahead_tree_node_children(node);
7532 last_ra_node = node->start;
7535 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7536 struct btrfs_dev_item *dev_item;
7537 dev_item = btrfs_item_ptr(leaf, slot,
7538 struct btrfs_dev_item);
7539 ret = read_one_dev(leaf, dev_item);
7543 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7544 struct btrfs_chunk *chunk;
7547 * We are only called at mount time, so no need to take
7548 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7549 * we always lock first fs_info->chunk_mutex before
7550 * acquiring any locks on the chunk tree. This is a
7551 * requirement for chunk allocation, see the comment on
7552 * top of btrfs_chunk_alloc() for details.
7554 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7555 ret = read_one_chunk(&found_key, leaf, chunk);
7560 /* Catch error found during iteration */
7567 * After loading chunk tree, we've got all device information,
7568 * do another round of validation checks.
7570 if (total_dev != fs_info->fs_devices->total_devices) {
7572 "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit",
7573 btrfs_super_num_devices(fs_info->super_copy),
7575 fs_info->fs_devices->total_devices = total_dev;
7576 btrfs_set_super_num_devices(fs_info->super_copy, total_dev);
7578 if (btrfs_super_total_bytes(fs_info->super_copy) <
7579 fs_info->fs_devices->total_rw_bytes) {
7581 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7582 btrfs_super_total_bytes(fs_info->super_copy),
7583 fs_info->fs_devices->total_rw_bytes);
7589 mutex_unlock(&uuid_mutex);
7591 btrfs_free_path(path);
7595 int btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7597 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7598 struct btrfs_device *device;
7601 fs_devices->fs_info = fs_info;
7603 mutex_lock(&fs_devices->device_list_mutex);
7604 list_for_each_entry(device, &fs_devices->devices, dev_list)
7605 device->fs_info = fs_info;
7607 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7608 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7609 device->fs_info = fs_info;
7610 ret = btrfs_get_dev_zone_info(device, false);
7615 seed_devs->fs_info = fs_info;
7617 mutex_unlock(&fs_devices->device_list_mutex);
7622 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7623 const struct btrfs_dev_stats_item *ptr,
7628 read_extent_buffer(eb, &val,
7629 offsetof(struct btrfs_dev_stats_item, values) +
7630 ((unsigned long)ptr) + (index * sizeof(u64)),
7635 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7636 struct btrfs_dev_stats_item *ptr,
7639 write_extent_buffer(eb, &val,
7640 offsetof(struct btrfs_dev_stats_item, values) +
7641 ((unsigned long)ptr) + (index * sizeof(u64)),
7645 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7646 struct btrfs_path *path)
7648 struct btrfs_dev_stats_item *ptr;
7649 struct extent_buffer *eb;
7650 struct btrfs_key key;
7654 if (!device->fs_info->dev_root)
7657 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7658 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7659 key.offset = device->devid;
7660 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7662 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7663 btrfs_dev_stat_set(device, i, 0);
7664 device->dev_stats_valid = 1;
7665 btrfs_release_path(path);
7666 return ret < 0 ? ret : 0;
7668 slot = path->slots[0];
7669 eb = path->nodes[0];
7670 item_size = btrfs_item_size(eb, slot);
7672 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7674 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7675 if (item_size >= (1 + i) * sizeof(__le64))
7676 btrfs_dev_stat_set(device, i,
7677 btrfs_dev_stats_value(eb, ptr, i));
7679 btrfs_dev_stat_set(device, i, 0);
7682 device->dev_stats_valid = 1;
7683 btrfs_dev_stat_print_on_load(device);
7684 btrfs_release_path(path);
7689 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7691 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7692 struct btrfs_device *device;
7693 struct btrfs_path *path = NULL;
7696 path = btrfs_alloc_path();
7700 mutex_lock(&fs_devices->device_list_mutex);
7701 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7702 ret = btrfs_device_init_dev_stats(device, path);
7706 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7707 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7708 ret = btrfs_device_init_dev_stats(device, path);
7714 mutex_unlock(&fs_devices->device_list_mutex);
7716 btrfs_free_path(path);
7720 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7721 struct btrfs_device *device)
7723 struct btrfs_fs_info *fs_info = trans->fs_info;
7724 struct btrfs_root *dev_root = fs_info->dev_root;
7725 struct btrfs_path *path;
7726 struct btrfs_key key;
7727 struct extent_buffer *eb;
7728 struct btrfs_dev_stats_item *ptr;
7732 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7733 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7734 key.offset = device->devid;
7736 path = btrfs_alloc_path();
7739 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7741 btrfs_warn_in_rcu(fs_info,
7742 "error %d while searching for dev_stats item for device %s",
7743 ret, btrfs_dev_name(device));
7748 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7749 /* need to delete old one and insert a new one */
7750 ret = btrfs_del_item(trans, dev_root, path);
7752 btrfs_warn_in_rcu(fs_info,
7753 "delete too small dev_stats item for device %s failed %d",
7754 btrfs_dev_name(device), ret);
7761 /* need to insert a new item */
7762 btrfs_release_path(path);
7763 ret = btrfs_insert_empty_item(trans, dev_root, path,
7764 &key, sizeof(*ptr));
7766 btrfs_warn_in_rcu(fs_info,
7767 "insert dev_stats item for device %s failed %d",
7768 btrfs_dev_name(device), ret);
7773 eb = path->nodes[0];
7774 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7775 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7776 btrfs_set_dev_stats_value(eb, ptr, i,
7777 btrfs_dev_stat_read(device, i));
7778 btrfs_mark_buffer_dirty(trans, eb);
7781 btrfs_free_path(path);
7786 * called from commit_transaction. Writes all changed device stats to disk.
7788 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7790 struct btrfs_fs_info *fs_info = trans->fs_info;
7791 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7792 struct btrfs_device *device;
7796 mutex_lock(&fs_devices->device_list_mutex);
7797 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7798 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7799 if (!device->dev_stats_valid || stats_cnt == 0)
7804 * There is a LOAD-LOAD control dependency between the value of
7805 * dev_stats_ccnt and updating the on-disk values which requires
7806 * reading the in-memory counters. Such control dependencies
7807 * require explicit read memory barriers.
7809 * This memory barriers pairs with smp_mb__before_atomic in
7810 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7811 * barrier implied by atomic_xchg in
7812 * btrfs_dev_stats_read_and_reset
7816 ret = update_dev_stat_item(trans, device);
7818 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7820 mutex_unlock(&fs_devices->device_list_mutex);
7825 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7827 btrfs_dev_stat_inc(dev, index);
7829 if (!dev->dev_stats_valid)
7831 btrfs_err_rl_in_rcu(dev->fs_info,
7832 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7833 btrfs_dev_name(dev),
7834 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7835 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7836 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7837 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7838 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7841 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7845 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7846 if (btrfs_dev_stat_read(dev, i) != 0)
7848 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7849 return; /* all values == 0, suppress message */
7851 btrfs_info_in_rcu(dev->fs_info,
7852 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7853 btrfs_dev_name(dev),
7854 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7855 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7856 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7857 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7858 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7861 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7862 struct btrfs_ioctl_get_dev_stats *stats)
7864 BTRFS_DEV_LOOKUP_ARGS(args);
7865 struct btrfs_device *dev;
7866 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7869 mutex_lock(&fs_devices->device_list_mutex);
7870 args.devid = stats->devid;
7871 dev = btrfs_find_device(fs_info->fs_devices, &args);
7872 mutex_unlock(&fs_devices->device_list_mutex);
7875 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7877 } else if (!dev->dev_stats_valid) {
7878 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7880 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7881 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7882 if (stats->nr_items > i)
7884 btrfs_dev_stat_read_and_reset(dev, i);
7886 btrfs_dev_stat_set(dev, i, 0);
7888 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7889 current->comm, task_pid_nr(current));
7891 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7892 if (stats->nr_items > i)
7893 stats->values[i] = btrfs_dev_stat_read(dev, i);
7895 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7896 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7901 * Update the size and bytes used for each device where it changed. This is
7902 * delayed since we would otherwise get errors while writing out the
7905 * Must be invoked during transaction commit.
7907 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7909 struct btrfs_device *curr, *next;
7911 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7913 if (list_empty(&trans->dev_update_list))
7917 * We don't need the device_list_mutex here. This list is owned by the
7918 * transaction and the transaction must complete before the device is
7921 mutex_lock(&trans->fs_info->chunk_mutex);
7922 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7924 list_del_init(&curr->post_commit_list);
7925 curr->commit_total_bytes = curr->disk_total_bytes;
7926 curr->commit_bytes_used = curr->bytes_used;
7928 mutex_unlock(&trans->fs_info->chunk_mutex);
7932 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7934 int btrfs_bg_type_to_factor(u64 flags)
7936 const int index = btrfs_bg_flags_to_raid_index(flags);
7938 return btrfs_raid_array[index].ncopies;
7943 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7944 u64 chunk_offset, u64 devid,
7945 u64 physical_offset, u64 physical_len)
7947 struct btrfs_dev_lookup_args args = { .devid = devid };
7948 struct btrfs_chunk_map *map;
7949 struct btrfs_device *dev;
7955 map = btrfs_find_chunk_map(fs_info, chunk_offset, 1);
7958 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7959 physical_offset, devid);
7964 stripe_len = btrfs_calc_stripe_length(map);
7965 if (physical_len != stripe_len) {
7967 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7968 physical_offset, devid, map->start, physical_len,
7975 * Very old mkfs.btrfs (before v4.1) will not respect the reserved
7976 * space. Although kernel can handle it without problem, better to warn
7979 if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED)
7981 "devid %llu physical %llu len %llu inside the reserved space",
7982 devid, physical_offset, physical_len);
7984 for (i = 0; i < map->num_stripes; i++) {
7985 if (map->stripes[i].dev->devid == devid &&
7986 map->stripes[i].physical == physical_offset) {
7988 if (map->verified_stripes >= map->num_stripes) {
7990 "too many dev extents for chunk %llu found",
7995 map->verified_stripes++;
8001 "dev extent physical offset %llu devid %llu has no corresponding chunk",
8002 physical_offset, devid);
8006 /* Make sure no dev extent is beyond device boundary */
8007 dev = btrfs_find_device(fs_info->fs_devices, &args);
8009 btrfs_err(fs_info, "failed to find devid %llu", devid);
8014 if (physical_offset + physical_len > dev->disk_total_bytes) {
8016 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
8017 devid, physical_offset, physical_len,
8018 dev->disk_total_bytes);
8023 if (dev->zone_info) {
8024 u64 zone_size = dev->zone_info->zone_size;
8026 if (!IS_ALIGNED(physical_offset, zone_size) ||
8027 !IS_ALIGNED(physical_len, zone_size)) {
8029 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
8030 devid, physical_offset, physical_len);
8037 btrfs_free_chunk_map(map);
8041 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
8043 struct rb_node *node;
8046 read_lock(&fs_info->mapping_tree_lock);
8047 for (node = rb_first_cached(&fs_info->mapping_tree); node; node = rb_next(node)) {
8048 struct btrfs_chunk_map *map;
8050 map = rb_entry(node, struct btrfs_chunk_map, rb_node);
8051 if (map->num_stripes != map->verified_stripes) {
8053 "chunk %llu has missing dev extent, have %d expect %d",
8054 map->start, map->verified_stripes, map->num_stripes);
8060 read_unlock(&fs_info->mapping_tree_lock);
8065 * Ensure that all dev extents are mapped to correct chunk, otherwise
8066 * later chunk allocation/free would cause unexpected behavior.
8068 * NOTE: This will iterate through the whole device tree, which should be of
8069 * the same size level as the chunk tree. This slightly increases mount time.
8071 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
8073 struct btrfs_path *path;
8074 struct btrfs_root *root = fs_info->dev_root;
8075 struct btrfs_key key;
8077 u64 prev_dev_ext_end = 0;
8081 * We don't have a dev_root because we mounted with ignorebadroots and
8082 * failed to load the root, so we want to skip the verification in this
8085 * However if the dev root is fine, but the tree itself is corrupted
8086 * we'd still fail to mount. This verification is only to make sure
8087 * writes can happen safely, so instead just bypass this check
8088 * completely in the case of IGNOREBADROOTS.
8090 if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
8094 key.type = BTRFS_DEV_EXTENT_KEY;
8097 path = btrfs_alloc_path();
8101 path->reada = READA_FORWARD;
8102 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8106 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
8107 ret = btrfs_next_leaf(root, path);
8110 /* No dev extents at all? Not good */
8117 struct extent_buffer *leaf = path->nodes[0];
8118 struct btrfs_dev_extent *dext;
8119 int slot = path->slots[0];
8121 u64 physical_offset;
8125 btrfs_item_key_to_cpu(leaf, &key, slot);
8126 if (key.type != BTRFS_DEV_EXTENT_KEY)
8128 devid = key.objectid;
8129 physical_offset = key.offset;
8131 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
8132 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
8133 physical_len = btrfs_dev_extent_length(leaf, dext);
8135 /* Check if this dev extent overlaps with the previous one */
8136 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
8138 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
8139 devid, physical_offset, prev_dev_ext_end);
8144 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
8145 physical_offset, physical_len);
8149 prev_dev_ext_end = physical_offset + physical_len;
8151 ret = btrfs_next_item(root, path);
8160 /* Ensure all chunks have corresponding dev extents */
8161 ret = verify_chunk_dev_extent_mapping(fs_info);
8163 btrfs_free_path(path);
8168 * Check whether the given block group or device is pinned by any inode being
8169 * used as a swapfile.
8171 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
8173 struct btrfs_swapfile_pin *sp;
8174 struct rb_node *node;
8176 spin_lock(&fs_info->swapfile_pins_lock);
8177 node = fs_info->swapfile_pins.rb_node;
8179 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
8181 node = node->rb_left;
8182 else if (ptr > sp->ptr)
8183 node = node->rb_right;
8187 spin_unlock(&fs_info->swapfile_pins_lock);
8188 return node != NULL;
8191 static int relocating_repair_kthread(void *data)
8193 struct btrfs_block_group *cache = data;
8194 struct btrfs_fs_info *fs_info = cache->fs_info;
8198 target = cache->start;
8199 btrfs_put_block_group(cache);
8201 sb_start_write(fs_info->sb);
8202 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
8204 "zoned: skip relocating block group %llu to repair: EBUSY",
8206 sb_end_write(fs_info->sb);
8210 mutex_lock(&fs_info->reclaim_bgs_lock);
8212 /* Ensure block group still exists */
8213 cache = btrfs_lookup_block_group(fs_info, target);
8217 if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags))
8220 ret = btrfs_may_alloc_data_chunk(fs_info, target);
8225 "zoned: relocating block group %llu to repair IO failure",
8227 ret = btrfs_relocate_chunk(fs_info, target);
8231 btrfs_put_block_group(cache);
8232 mutex_unlock(&fs_info->reclaim_bgs_lock);
8233 btrfs_exclop_finish(fs_info);
8234 sb_end_write(fs_info->sb);
8239 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical)
8241 struct btrfs_block_group *cache;
8243 if (!btrfs_is_zoned(fs_info))
8246 /* Do not attempt to repair in degraded state */
8247 if (btrfs_test_opt(fs_info, DEGRADED))
8250 cache = btrfs_lookup_block_group(fs_info, logical);
8254 if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) {
8255 btrfs_put_block_group(cache);
8259 kthread_run(relocating_repair_kthread, cache,
8260 "btrfs-relocating-repair");
8265 static void map_raid56_repair_block(struct btrfs_io_context *bioc,
8266 struct btrfs_io_stripe *smap,
8269 int data_stripes = nr_bioc_data_stripes(bioc);
8272 for (i = 0; i < data_stripes; i++) {
8273 u64 stripe_start = bioc->full_stripe_logical +
8274 btrfs_stripe_nr_to_offset(i);
8276 if (logical >= stripe_start &&
8277 logical < stripe_start + BTRFS_STRIPE_LEN)
8280 ASSERT(i < data_stripes);
8281 smap->dev = bioc->stripes[i].dev;
8282 smap->physical = bioc->stripes[i].physical +
8283 ((logical - bioc->full_stripe_logical) &
8284 BTRFS_STRIPE_LEN_MASK);
8288 * Map a repair write into a single device.
8290 * A repair write is triggered by read time repair or scrub, which would only
8291 * update the contents of a single device.
8292 * Not update any other mirrors nor go through RMW path.
8294 * Callers should ensure:
8296 * - Call btrfs_bio_counter_inc_blocked() first
8297 * - The range does not cross stripe boundary
8298 * - Has a valid @mirror_num passed in.
8300 int btrfs_map_repair_block(struct btrfs_fs_info *fs_info,
8301 struct btrfs_io_stripe *smap, u64 logical,
8302 u32 length, int mirror_num)
8304 struct btrfs_io_context *bioc = NULL;
8305 u64 map_length = length;
8306 int mirror_ret = mirror_num;
8309 ASSERT(mirror_num > 0);
8311 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, &map_length,
8312 &bioc, smap, &mirror_ret);
8316 /* The map range should not cross stripe boundary. */
8317 ASSERT(map_length >= length);
8319 /* Already mapped to single stripe. */
8323 /* Map the RAID56 multi-stripe writes to a single one. */
8324 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
8325 map_raid56_repair_block(bioc, smap, logical);
8329 ASSERT(mirror_num <= bioc->num_stripes);
8330 smap->dev = bioc->stripes[mirror_num - 1].dev;
8331 smap->physical = bioc->stripes[mirror_num - 1].physical;
8333 btrfs_put_bioc(bioc);