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 if (device->devt != device->bdev->bd_dev) {
697 "device %s maj:min changed from %d:%d to %d:%d",
698 device->name->str, MAJOR(device->devt),
699 MINOR(device->devt), MAJOR(device->bdev->bd_dev),
700 MINOR(device->bdev->bd_dev));
702 device->devt = device->bdev->bd_dev;
705 fs_devices->open_devices++;
706 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
707 device->devid != BTRFS_DEV_REPLACE_DEVID) {
708 fs_devices->rw_devices++;
709 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
711 btrfs_release_disk_super(disk_super);
716 btrfs_release_disk_super(disk_super);
722 u8 *btrfs_sb_fsid_ptr(struct btrfs_super_block *sb)
724 bool has_metadata_uuid = (btrfs_super_incompat_flags(sb) &
725 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
727 return has_metadata_uuid ? sb->metadata_uuid : sb->fsid;
731 * Add new device to list of registered devices
734 * device pointer which was just added or updated when successful
735 * error pointer when failed
737 static noinline struct btrfs_device *device_list_add(const char *path,
738 struct btrfs_super_block *disk_super,
739 bool *new_device_added)
741 struct btrfs_device *device;
742 struct btrfs_fs_devices *fs_devices = NULL;
743 struct rcu_string *name;
744 u64 found_transid = btrfs_super_generation(disk_super);
745 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
748 bool same_fsid_diff_dev = false;
749 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
750 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
752 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
754 "device %s has incomplete metadata_uuid change, please use btrfstune to complete",
756 return ERR_PTR(-EAGAIN);
759 error = lookup_bdev(path, &path_devt);
761 btrfs_err(NULL, "failed to lookup block device for path %s: %d",
763 return ERR_PTR(error);
766 fs_devices = find_fsid_by_device(disk_super, path_devt, &same_fsid_diff_dev);
769 fs_devices = alloc_fs_devices(disk_super->fsid);
770 if (IS_ERR(fs_devices))
771 return ERR_CAST(fs_devices);
773 if (has_metadata_uuid)
774 memcpy(fs_devices->metadata_uuid,
775 disk_super->metadata_uuid, BTRFS_FSID_SIZE);
777 if (same_fsid_diff_dev) {
778 generate_random_uuid(fs_devices->fsid);
779 fs_devices->temp_fsid = true;
780 pr_info("BTRFS: device %s (%d:%d) using temp-fsid %pU\n",
781 path, MAJOR(path_devt), MINOR(path_devt),
785 mutex_lock(&fs_devices->device_list_mutex);
786 list_add(&fs_devices->fs_list, &fs_uuids);
790 struct btrfs_dev_lookup_args args = {
792 .uuid = disk_super->dev_item.uuid,
795 mutex_lock(&fs_devices->device_list_mutex);
796 device = btrfs_find_device(fs_devices, &args);
798 if (found_transid > fs_devices->latest_generation) {
799 memcpy(fs_devices->fsid, disk_super->fsid,
801 memcpy(fs_devices->metadata_uuid,
802 btrfs_sb_fsid_ptr(disk_super), BTRFS_FSID_SIZE);
807 unsigned int nofs_flag;
809 if (fs_devices->opened) {
811 "device %s (%d:%d) belongs to fsid %pU, and the fs is already mounted, scanned by %s (%d)",
812 path, MAJOR(path_devt), MINOR(path_devt),
813 fs_devices->fsid, current->comm,
814 task_pid_nr(current));
815 mutex_unlock(&fs_devices->device_list_mutex);
816 return ERR_PTR(-EBUSY);
819 nofs_flag = memalloc_nofs_save();
820 device = btrfs_alloc_device(NULL, &devid,
821 disk_super->dev_item.uuid, path);
822 memalloc_nofs_restore(nofs_flag);
823 if (IS_ERR(device)) {
824 mutex_unlock(&fs_devices->device_list_mutex);
825 /* we can safely leave the fs_devices entry around */
829 device->devt = path_devt;
831 list_add_rcu(&device->dev_list, &fs_devices->devices);
832 fs_devices->num_devices++;
834 device->fs_devices = fs_devices;
835 *new_device_added = true;
837 if (disk_super->label[0])
839 "BTRFS: device label %s devid %llu transid %llu %s (%d:%d) scanned by %s (%d)\n",
840 disk_super->label, devid, found_transid, path,
841 MAJOR(path_devt), MINOR(path_devt),
842 current->comm, task_pid_nr(current));
845 "BTRFS: device fsid %pU devid %llu transid %llu %s (%d:%d) scanned by %s (%d)\n",
846 disk_super->fsid, devid, found_transid, path,
847 MAJOR(path_devt), MINOR(path_devt),
848 current->comm, task_pid_nr(current));
850 } else if (!device->name || strcmp(device->name->str, path)) {
852 * When FS is already mounted.
853 * 1. If you are here and if the device->name is NULL that
854 * means this device was missing at time of FS mount.
855 * 2. If you are here and if the device->name is different
856 * from 'path' that means either
857 * a. The same device disappeared and reappeared with
859 * b. The missing-disk-which-was-replaced, has
862 * We must allow 1 and 2a above. But 2b would be a spurious
865 * Further in case of 1 and 2a above, the disk at 'path'
866 * would have missed some transaction when it was away and
867 * in case of 2a the stale bdev has to be updated as well.
868 * 2b must not be allowed at all time.
872 * For now, we do allow update to btrfs_fs_device through the
873 * btrfs dev scan cli after FS has been mounted. We're still
874 * tracking a problem where systems fail mount by subvolume id
875 * when we reject replacement on a mounted FS.
877 if (!fs_devices->opened && found_transid < device->generation) {
879 * That is if the FS is _not_ mounted and if you
880 * are here, that means there is more than one
881 * disk with same uuid and devid.We keep the one
882 * with larger generation number or the last-in if
883 * generation are equal.
885 mutex_unlock(&fs_devices->device_list_mutex);
887 "device %s already registered with a higher generation, found %llu expect %llu",
888 path, found_transid, device->generation);
889 return ERR_PTR(-EEXIST);
893 * We are going to replace the device path for a given devid,
894 * make sure it's the same device if the device is mounted
896 * NOTE: the device->fs_info may not be reliable here so pass
897 * in a NULL to message helpers instead. This avoids a possible
898 * use-after-free when the fs_info and fs_info->sb are already
902 if (device->devt != path_devt) {
903 mutex_unlock(&fs_devices->device_list_mutex);
904 btrfs_warn_in_rcu(NULL,
905 "duplicate device %s devid %llu generation %llu scanned by %s (%d)",
906 path, devid, found_transid,
908 task_pid_nr(current));
909 return ERR_PTR(-EEXIST);
911 btrfs_info_in_rcu(NULL,
912 "devid %llu device path %s changed to %s scanned by %s (%d)",
913 devid, btrfs_dev_name(device),
915 task_pid_nr(current));
918 name = rcu_string_strdup(path, GFP_NOFS);
920 mutex_unlock(&fs_devices->device_list_mutex);
921 return ERR_PTR(-ENOMEM);
923 rcu_string_free(device->name);
924 rcu_assign_pointer(device->name, name);
925 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
926 fs_devices->missing_devices--;
927 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
929 device->devt = path_devt;
933 * Unmount does not free the btrfs_device struct but would zero
934 * generation along with most of the other members. So just update
935 * it back. We need it to pick the disk with largest generation
938 if (!fs_devices->opened) {
939 device->generation = found_transid;
940 fs_devices->latest_generation = max_t(u64, found_transid,
941 fs_devices->latest_generation);
944 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
946 mutex_unlock(&fs_devices->device_list_mutex);
950 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
952 struct btrfs_fs_devices *fs_devices;
953 struct btrfs_device *device;
954 struct btrfs_device *orig_dev;
957 lockdep_assert_held(&uuid_mutex);
959 fs_devices = alloc_fs_devices(orig->fsid);
960 if (IS_ERR(fs_devices))
963 fs_devices->total_devices = orig->total_devices;
965 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
966 const char *dev_path = NULL;
969 * This is ok to do without RCU read locked because we hold the
970 * uuid mutex so nothing we touch in here is going to disappear.
973 dev_path = orig_dev->name->str;
975 device = btrfs_alloc_device(NULL, &orig_dev->devid,
976 orig_dev->uuid, dev_path);
977 if (IS_ERR(device)) {
978 ret = PTR_ERR(device);
982 if (orig_dev->zone_info) {
983 struct btrfs_zoned_device_info *zone_info;
985 zone_info = btrfs_clone_dev_zone_info(orig_dev);
987 btrfs_free_device(device);
991 device->zone_info = zone_info;
994 list_add(&device->dev_list, &fs_devices->devices);
995 device->fs_devices = fs_devices;
996 fs_devices->num_devices++;
1000 free_fs_devices(fs_devices);
1001 return ERR_PTR(ret);
1004 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices,
1005 struct btrfs_device **latest_dev)
1007 struct btrfs_device *device, *next;
1009 /* This is the initialized path, it is safe to release the devices. */
1010 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1011 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) {
1012 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1013 &device->dev_state) &&
1014 !test_bit(BTRFS_DEV_STATE_MISSING,
1015 &device->dev_state) &&
1017 device->generation > (*latest_dev)->generation)) {
1018 *latest_dev = device;
1024 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID,
1025 * in btrfs_init_dev_replace() so just continue.
1027 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1030 if (device->bdev_file) {
1031 fput(device->bdev_file);
1032 device->bdev = NULL;
1033 device->bdev_file = NULL;
1034 fs_devices->open_devices--;
1036 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1037 list_del_init(&device->dev_alloc_list);
1038 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1039 fs_devices->rw_devices--;
1041 list_del_init(&device->dev_list);
1042 fs_devices->num_devices--;
1043 btrfs_free_device(device);
1049 * After we have read the system tree and know devids belonging to this
1050 * filesystem, remove the device which does not belong there.
1052 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices)
1054 struct btrfs_device *latest_dev = NULL;
1055 struct btrfs_fs_devices *seed_dev;
1057 mutex_lock(&uuid_mutex);
1058 __btrfs_free_extra_devids(fs_devices, &latest_dev);
1060 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list)
1061 __btrfs_free_extra_devids(seed_dev, &latest_dev);
1063 fs_devices->latest_dev = latest_dev;
1065 mutex_unlock(&uuid_mutex);
1068 static void btrfs_close_bdev(struct btrfs_device *device)
1073 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1074 sync_blockdev(device->bdev);
1075 invalidate_bdev(device->bdev);
1078 fput(device->bdev_file);
1081 static void btrfs_close_one_device(struct btrfs_device *device)
1083 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1085 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1086 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1087 list_del_init(&device->dev_alloc_list);
1088 fs_devices->rw_devices--;
1091 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1092 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
1094 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
1095 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
1096 fs_devices->missing_devices--;
1099 btrfs_close_bdev(device);
1101 fs_devices->open_devices--;
1102 device->bdev = NULL;
1104 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1105 btrfs_destroy_dev_zone_info(device);
1107 device->fs_info = NULL;
1108 atomic_set(&device->dev_stats_ccnt, 0);
1109 extent_io_tree_release(&device->alloc_state);
1112 * Reset the flush error record. We might have a transient flush error
1113 * in this mount, and if so we aborted the current transaction and set
1114 * the fs to an error state, guaranteeing no super blocks can be further
1115 * committed. However that error might be transient and if we unmount the
1116 * filesystem and mount it again, we should allow the mount to succeed
1117 * (btrfs_check_rw_degradable() should not fail) - if after mounting the
1118 * filesystem again we still get flush errors, then we will again abort
1119 * any transaction and set the error state, guaranteeing no commits of
1120 * unsafe super blocks.
1122 device->last_flush_error = 0;
1124 /* Verify the device is back in a pristine state */
1125 WARN_ON(test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1126 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1127 WARN_ON(!list_empty(&device->dev_alloc_list));
1128 WARN_ON(!list_empty(&device->post_commit_list));
1131 static void close_fs_devices(struct btrfs_fs_devices *fs_devices)
1133 struct btrfs_device *device, *tmp;
1135 lockdep_assert_held(&uuid_mutex);
1137 if (--fs_devices->opened > 0)
1140 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list)
1141 btrfs_close_one_device(device);
1143 WARN_ON(fs_devices->open_devices);
1144 WARN_ON(fs_devices->rw_devices);
1145 fs_devices->opened = 0;
1146 fs_devices->seeding = false;
1147 fs_devices->fs_info = NULL;
1150 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1153 struct btrfs_fs_devices *tmp;
1155 mutex_lock(&uuid_mutex);
1156 close_fs_devices(fs_devices);
1157 if (!fs_devices->opened) {
1158 list_splice_init(&fs_devices->seed_list, &list);
1161 * If the struct btrfs_fs_devices is not assembled with any
1162 * other device, it can be re-initialized during the next mount
1163 * without the needing device-scan step. Therefore, it can be
1166 if (fs_devices->num_devices == 1) {
1167 list_del(&fs_devices->fs_list);
1168 free_fs_devices(fs_devices);
1173 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {
1174 close_fs_devices(fs_devices);
1175 list_del(&fs_devices->seed_list);
1176 free_fs_devices(fs_devices);
1178 mutex_unlock(&uuid_mutex);
1181 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1182 blk_mode_t flags, void *holder)
1184 struct btrfs_device *device;
1185 struct btrfs_device *latest_dev = NULL;
1186 struct btrfs_device *tmp_device;
1189 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,
1193 ret2 = btrfs_open_one_device(fs_devices, device, flags, holder);
1195 (!latest_dev || device->generation > latest_dev->generation)) {
1196 latest_dev = device;
1197 } else if (ret2 == -ENODATA) {
1198 fs_devices->num_devices--;
1199 list_del(&device->dev_list);
1200 btrfs_free_device(device);
1202 if (ret == 0 && ret2 != 0)
1206 if (fs_devices->open_devices == 0) {
1212 fs_devices->opened = 1;
1213 fs_devices->latest_dev = latest_dev;
1214 fs_devices->total_rw_bytes = 0;
1215 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1216 fs_devices->read_policy = BTRFS_READ_POLICY_PID;
1221 static int devid_cmp(void *priv, const struct list_head *a,
1222 const struct list_head *b)
1224 const struct btrfs_device *dev1, *dev2;
1226 dev1 = list_entry(a, struct btrfs_device, dev_list);
1227 dev2 = list_entry(b, struct btrfs_device, dev_list);
1229 if (dev1->devid < dev2->devid)
1231 else if (dev1->devid > dev2->devid)
1236 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1237 blk_mode_t flags, void *holder)
1241 lockdep_assert_held(&uuid_mutex);
1243 * The device_list_mutex cannot be taken here in case opening the
1244 * underlying device takes further locks like open_mutex.
1246 * We also don't need the lock here as this is called during mount and
1247 * exclusion is provided by uuid_mutex
1250 if (fs_devices->opened) {
1251 fs_devices->opened++;
1254 list_sort(NULL, &fs_devices->devices, devid_cmp);
1255 ret = open_fs_devices(fs_devices, flags, holder);
1261 void btrfs_release_disk_super(struct btrfs_super_block *super)
1263 struct page *page = virt_to_page(super);
1268 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1269 u64 bytenr, u64 bytenr_orig)
1271 struct btrfs_super_block *disk_super;
1276 /* make sure our super fits in the device */
1277 if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev))
1278 return ERR_PTR(-EINVAL);
1280 /* make sure our super fits in the page */
1281 if (sizeof(*disk_super) > PAGE_SIZE)
1282 return ERR_PTR(-EINVAL);
1284 /* make sure our super doesn't straddle pages on disk */
1285 index = bytenr >> PAGE_SHIFT;
1286 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1287 return ERR_PTR(-EINVAL);
1289 /* pull in the page with our super */
1290 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1293 return ERR_CAST(page);
1295 p = page_address(page);
1297 /* align our pointer to the offset of the super block */
1298 disk_super = p + offset_in_page(bytenr);
1300 if (btrfs_super_bytenr(disk_super) != bytenr_orig ||
1301 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1302 btrfs_release_disk_super(p);
1303 return ERR_PTR(-EINVAL);
1306 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1307 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1312 int btrfs_forget_devices(dev_t devt)
1316 mutex_lock(&uuid_mutex);
1317 ret = btrfs_free_stale_devices(devt, NULL);
1318 mutex_unlock(&uuid_mutex);
1323 static bool btrfs_skip_registration(struct btrfs_super_block *disk_super,
1324 const char *path, dev_t devt,
1327 struct btrfs_fs_devices *fs_devices;
1330 * Do not skip device registration for mounted devices with matching
1331 * maj:min but different paths. Booting without initrd relies on
1332 * /dev/root initially, later replaced with the actual root device.
1333 * A successful scan ensures grub2-probe selects the correct device.
1335 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
1336 struct btrfs_device *device;
1338 mutex_lock(&fs_devices->device_list_mutex);
1340 if (!fs_devices->opened) {
1341 mutex_unlock(&fs_devices->device_list_mutex);
1345 list_for_each_entry(device, &fs_devices->devices, dev_list) {
1346 if (device->bdev && (device->bdev->bd_dev == devt) &&
1347 strcmp(device->name->str, path) != 0) {
1348 mutex_unlock(&fs_devices->device_list_mutex);
1350 /* Do not skip registration. */
1354 mutex_unlock(&fs_devices->device_list_mutex);
1357 if (!mount_arg_dev && btrfs_super_num_devices(disk_super) == 1 &&
1358 !(btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING))
1365 * Look for a btrfs signature on a device. This may be called out of the mount path
1366 * and we are not allowed to call set_blocksize during the scan. The superblock
1367 * is read via pagecache.
1369 * With @mount_arg_dev it's a scan during mount time that will always register
1370 * the device or return an error. Multi-device and seeding devices are registered
1373 struct btrfs_device *btrfs_scan_one_device(const char *path, blk_mode_t flags,
1376 struct btrfs_super_block *disk_super;
1377 bool new_device_added = false;
1378 struct btrfs_device *device = NULL;
1379 struct file *bdev_file;
1380 u64 bytenr, bytenr_orig;
1384 lockdep_assert_held(&uuid_mutex);
1387 * we would like to check all the supers, but that would make
1388 * a btrfs mount succeed after a mkfs from a different FS.
1389 * So, we need to add a special mount option to scan for
1390 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1394 * Avoid an exclusive open here, as the systemd-udev may initiate the
1395 * device scan which may race with the user's mount or mkfs command,
1396 * resulting in failure.
1397 * Since the device scan is solely for reading purposes, there is no
1398 * need for an exclusive open. Additionally, the devices are read again
1399 * during the mount process. It is ok to get some inconsistent
1400 * values temporarily, as the device paths of the fsid are the only
1401 * required information for assembling the volume.
1403 bdev_file = bdev_file_open_by_path(path, flags, NULL, NULL);
1404 if (IS_ERR(bdev_file))
1405 return ERR_CAST(bdev_file);
1407 bytenr_orig = btrfs_sb_offset(0);
1408 ret = btrfs_sb_log_location_bdev(file_bdev(bdev_file), 0, READ, &bytenr);
1410 device = ERR_PTR(ret);
1411 goto error_bdev_put;
1414 disk_super = btrfs_read_disk_super(file_bdev(bdev_file), bytenr,
1416 if (IS_ERR(disk_super)) {
1417 device = ERR_CAST(disk_super);
1418 goto error_bdev_put;
1421 devt = file_bdev(bdev_file)->bd_dev;
1422 if (btrfs_skip_registration(disk_super, path, devt, mount_arg_dev)) {
1423 pr_debug("BTRFS: skip registering single non-seed device %s (%d:%d)\n",
1424 path, MAJOR(devt), MINOR(devt));
1426 btrfs_free_stale_devices(devt, NULL);
1429 goto free_disk_super;
1432 device = device_list_add(path, disk_super, &new_device_added);
1433 if (!IS_ERR(device) && new_device_added)
1434 btrfs_free_stale_devices(device->devt, device);
1437 btrfs_release_disk_super(disk_super);
1446 * Try to find a chunk that intersects [start, start + len] range and when one
1447 * such is found, record the end of it in *start
1449 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1452 u64 physical_start, physical_end;
1454 lockdep_assert_held(&device->fs_info->chunk_mutex);
1456 if (find_first_extent_bit(&device->alloc_state, *start,
1457 &physical_start, &physical_end,
1458 CHUNK_ALLOCATED, NULL)) {
1460 if (in_range(physical_start, *start, len) ||
1461 in_range(*start, physical_start,
1462 physical_end + 1 - physical_start)) {
1463 *start = physical_end + 1;
1470 static u64 dev_extent_search_start(struct btrfs_device *device)
1472 switch (device->fs_devices->chunk_alloc_policy) {
1473 case BTRFS_CHUNK_ALLOC_REGULAR:
1474 return BTRFS_DEVICE_RANGE_RESERVED;
1475 case BTRFS_CHUNK_ALLOC_ZONED:
1477 * We don't care about the starting region like regular
1478 * allocator, because we anyway use/reserve the first two zones
1479 * for superblock logging.
1487 static bool dev_extent_hole_check_zoned(struct btrfs_device *device,
1488 u64 *hole_start, u64 *hole_size,
1491 u64 zone_size = device->zone_info->zone_size;
1494 bool changed = false;
1496 ASSERT(IS_ALIGNED(*hole_start, zone_size));
1498 while (*hole_size > 0) {
1499 pos = btrfs_find_allocatable_zones(device, *hole_start,
1500 *hole_start + *hole_size,
1502 if (pos != *hole_start) {
1503 *hole_size = *hole_start + *hole_size - pos;
1506 if (*hole_size < num_bytes)
1510 ret = btrfs_ensure_empty_zones(device, pos, num_bytes);
1512 /* Range is ensured to be empty */
1516 /* Given hole range was invalid (outside of device) */
1517 if (ret == -ERANGE) {
1518 *hole_start += *hole_size;
1523 *hole_start += zone_size;
1524 *hole_size -= zone_size;
1532 * Check if specified hole is suitable for allocation.
1534 * @device: the device which we have the hole
1535 * @hole_start: starting position of the hole
1536 * @hole_size: the size of the hole
1537 * @num_bytes: the size of the free space that we need
1539 * This function may modify @hole_start and @hole_size to reflect the suitable
1540 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1542 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1543 u64 *hole_size, u64 num_bytes)
1545 bool changed = false;
1546 u64 hole_end = *hole_start + *hole_size;
1550 * Check before we set max_hole_start, otherwise we could end up
1551 * sending back this offset anyway.
1553 if (contains_pending_extent(device, hole_start, *hole_size)) {
1554 if (hole_end >= *hole_start)
1555 *hole_size = hole_end - *hole_start;
1561 switch (device->fs_devices->chunk_alloc_policy) {
1562 case BTRFS_CHUNK_ALLOC_REGULAR:
1563 /* No extra check */
1565 case BTRFS_CHUNK_ALLOC_ZONED:
1566 if (dev_extent_hole_check_zoned(device, hole_start,
1567 hole_size, num_bytes)) {
1570 * The changed hole can contain pending extent.
1571 * Loop again to check that.
1587 * Find free space in the specified device.
1589 * @device: the device which we search the free space in
1590 * @num_bytes: the size of the free space that we need
1591 * @search_start: the position from which to begin the search
1592 * @start: store the start of the free space.
1593 * @len: the size of the free space. that we find, or the size
1594 * of the max free space if we don't find suitable free space
1596 * This does a pretty simple search, the expectation is that it is called very
1597 * infrequently and that a given device has a small number of extents.
1599 * @start is used to store the start of the free space if we find. But if we
1600 * don't find suitable free space, it will be used to store the start position
1601 * of the max free space.
1603 * @len is used to store the size of the free space that we find.
1604 * But if we don't find suitable free space, it is used to store the size of
1605 * the max free space.
1607 * NOTE: This function will search *commit* root of device tree, and does extra
1608 * check to ensure dev extents are not double allocated.
1609 * This makes the function safe to allocate dev extents but may not report
1610 * correct usable device space, as device extent freed in current transaction
1611 * is not reported as available.
1613 static int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1614 u64 *start, u64 *len)
1616 struct btrfs_fs_info *fs_info = device->fs_info;
1617 struct btrfs_root *root = fs_info->dev_root;
1618 struct btrfs_key key;
1619 struct btrfs_dev_extent *dev_extent;
1620 struct btrfs_path *path;
1624 u64 max_hole_size = 0;
1626 u64 search_end = device->total_bytes;
1629 struct extent_buffer *l;
1631 search_start = dev_extent_search_start(device);
1632 max_hole_start = search_start;
1634 WARN_ON(device->zone_info &&
1635 !IS_ALIGNED(num_bytes, device->zone_info->zone_size));
1637 path = btrfs_alloc_path();
1643 if (search_start >= search_end ||
1644 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1649 path->reada = READA_FORWARD;
1650 path->search_commit_root = 1;
1651 path->skip_locking = 1;
1653 key.objectid = device->devid;
1654 key.offset = search_start;
1655 key.type = BTRFS_DEV_EXTENT_KEY;
1657 ret = btrfs_search_backwards(root, &key, path);
1661 while (search_start < search_end) {
1663 slot = path->slots[0];
1664 if (slot >= btrfs_header_nritems(l)) {
1665 ret = btrfs_next_leaf(root, path);
1673 btrfs_item_key_to_cpu(l, &key, slot);
1675 if (key.objectid < device->devid)
1678 if (key.objectid > device->devid)
1681 if (key.type != BTRFS_DEV_EXTENT_KEY)
1684 if (key.offset > search_end)
1687 if (key.offset > search_start) {
1688 hole_size = key.offset - search_start;
1689 dev_extent_hole_check(device, &search_start, &hole_size,
1692 if (hole_size > max_hole_size) {
1693 max_hole_start = search_start;
1694 max_hole_size = hole_size;
1698 * If this free space is greater than which we need,
1699 * it must be the max free space that we have found
1700 * until now, so max_hole_start must point to the start
1701 * of this free space and the length of this free space
1702 * is stored in max_hole_size. Thus, we return
1703 * max_hole_start and max_hole_size and go back to the
1706 if (hole_size >= num_bytes) {
1712 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1713 extent_end = key.offset + btrfs_dev_extent_length(l,
1715 if (extent_end > search_start)
1716 search_start = extent_end;
1723 * At this point, search_start should be the end of
1724 * allocated dev extents, and when shrinking the device,
1725 * search_end may be smaller than search_start.
1727 if (search_end > search_start) {
1728 hole_size = search_end - search_start;
1729 if (dev_extent_hole_check(device, &search_start, &hole_size,
1731 btrfs_release_path(path);
1735 if (hole_size > max_hole_size) {
1736 max_hole_start = search_start;
1737 max_hole_size = hole_size;
1742 if (max_hole_size < num_bytes)
1747 ASSERT(max_hole_start + max_hole_size <= search_end);
1749 btrfs_free_path(path);
1750 *start = max_hole_start;
1752 *len = max_hole_size;
1756 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1757 struct btrfs_device *device,
1758 u64 start, u64 *dev_extent_len)
1760 struct btrfs_fs_info *fs_info = device->fs_info;
1761 struct btrfs_root *root = fs_info->dev_root;
1763 struct btrfs_path *path;
1764 struct btrfs_key key;
1765 struct btrfs_key found_key;
1766 struct extent_buffer *leaf = NULL;
1767 struct btrfs_dev_extent *extent = NULL;
1769 path = btrfs_alloc_path();
1773 key.objectid = device->devid;
1775 key.type = BTRFS_DEV_EXTENT_KEY;
1777 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1779 ret = btrfs_previous_item(root, path, key.objectid,
1780 BTRFS_DEV_EXTENT_KEY);
1783 leaf = path->nodes[0];
1784 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1785 extent = btrfs_item_ptr(leaf, path->slots[0],
1786 struct btrfs_dev_extent);
1787 BUG_ON(found_key.offset > start || found_key.offset +
1788 btrfs_dev_extent_length(leaf, extent) < start);
1790 btrfs_release_path(path);
1792 } else if (ret == 0) {
1793 leaf = path->nodes[0];
1794 extent = btrfs_item_ptr(leaf, path->slots[0],
1795 struct btrfs_dev_extent);
1800 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1802 ret = btrfs_del_item(trans, root, path);
1804 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1806 btrfs_free_path(path);
1810 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1815 read_lock(&fs_info->mapping_tree_lock);
1816 n = rb_last(&fs_info->mapping_tree.rb_root);
1818 struct btrfs_chunk_map *map;
1820 map = rb_entry(n, struct btrfs_chunk_map, rb_node);
1821 ret = map->start + map->chunk_len;
1823 read_unlock(&fs_info->mapping_tree_lock);
1828 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1832 struct btrfs_key key;
1833 struct btrfs_key found_key;
1834 struct btrfs_path *path;
1836 path = btrfs_alloc_path();
1840 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1841 key.type = BTRFS_DEV_ITEM_KEY;
1842 key.offset = (u64)-1;
1844 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1850 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1855 ret = btrfs_previous_item(fs_info->chunk_root, path,
1856 BTRFS_DEV_ITEMS_OBJECTID,
1857 BTRFS_DEV_ITEM_KEY);
1861 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1863 *devid_ret = found_key.offset + 1;
1867 btrfs_free_path(path);
1872 * the device information is stored in the chunk root
1873 * the btrfs_device struct should be fully filled in
1875 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1876 struct btrfs_device *device)
1879 struct btrfs_path *path;
1880 struct btrfs_dev_item *dev_item;
1881 struct extent_buffer *leaf;
1882 struct btrfs_key key;
1885 path = btrfs_alloc_path();
1889 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1890 key.type = BTRFS_DEV_ITEM_KEY;
1891 key.offset = device->devid;
1893 btrfs_reserve_chunk_metadata(trans, true);
1894 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1895 &key, sizeof(*dev_item));
1896 btrfs_trans_release_chunk_metadata(trans);
1900 leaf = path->nodes[0];
1901 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1903 btrfs_set_device_id(leaf, dev_item, device->devid);
1904 btrfs_set_device_generation(leaf, dev_item, 0);
1905 btrfs_set_device_type(leaf, dev_item, device->type);
1906 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1907 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1908 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1909 btrfs_set_device_total_bytes(leaf, dev_item,
1910 btrfs_device_get_disk_total_bytes(device));
1911 btrfs_set_device_bytes_used(leaf, dev_item,
1912 btrfs_device_get_bytes_used(device));
1913 btrfs_set_device_group(leaf, dev_item, 0);
1914 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1915 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1916 btrfs_set_device_start_offset(leaf, dev_item, 0);
1918 ptr = btrfs_device_uuid(dev_item);
1919 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1920 ptr = btrfs_device_fsid(dev_item);
1921 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1922 ptr, BTRFS_FSID_SIZE);
1923 btrfs_mark_buffer_dirty(trans, leaf);
1927 btrfs_free_path(path);
1932 * Function to update ctime/mtime for a given device path.
1933 * Mainly used for ctime/mtime based probe like libblkid.
1935 * We don't care about errors here, this is just to be kind to userspace.
1937 static void update_dev_time(const char *device_path)
1942 ret = kern_path(device_path, LOOKUP_FOLLOW, &path);
1946 inode_update_time(d_inode(path.dentry), S_MTIME | S_CTIME | S_VERSION);
1950 static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans,
1951 struct btrfs_device *device)
1953 struct btrfs_root *root = device->fs_info->chunk_root;
1955 struct btrfs_path *path;
1956 struct btrfs_key key;
1958 path = btrfs_alloc_path();
1962 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1963 key.type = BTRFS_DEV_ITEM_KEY;
1964 key.offset = device->devid;
1966 btrfs_reserve_chunk_metadata(trans, false);
1967 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1968 btrfs_trans_release_chunk_metadata(trans);
1975 ret = btrfs_del_item(trans, root, path);
1977 btrfs_free_path(path);
1982 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1983 * filesystem. It's up to the caller to adjust that number regarding eg. device
1986 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1994 seq = read_seqbegin(&fs_info->profiles_lock);
1996 all_avail = fs_info->avail_data_alloc_bits |
1997 fs_info->avail_system_alloc_bits |
1998 fs_info->avail_metadata_alloc_bits;
1999 } while (read_seqretry(&fs_info->profiles_lock, seq));
2001 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2002 if (!(all_avail & btrfs_raid_array[i].bg_flag))
2005 if (num_devices < btrfs_raid_array[i].devs_min)
2006 return btrfs_raid_array[i].mindev_error;
2012 static struct btrfs_device * btrfs_find_next_active_device(
2013 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
2015 struct btrfs_device *next_device;
2017 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
2018 if (next_device != device &&
2019 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
2020 && next_device->bdev)
2028 * Helper function to check if the given device is part of s_bdev / latest_dev
2029 * and replace it with the provided or the next active device, in the context
2030 * where this function called, there should be always be another device (or
2031 * this_dev) which is active.
2033 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
2034 struct btrfs_device *next_device)
2036 struct btrfs_fs_info *fs_info = device->fs_info;
2039 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
2041 ASSERT(next_device);
2043 if (fs_info->sb->s_bdev &&
2044 (fs_info->sb->s_bdev == device->bdev))
2045 fs_info->sb->s_bdev = next_device->bdev;
2047 if (fs_info->fs_devices->latest_dev->bdev == device->bdev)
2048 fs_info->fs_devices->latest_dev = next_device;
2052 * Return btrfs_fs_devices::num_devices excluding the device that's being
2053 * currently replaced.
2055 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2057 u64 num_devices = fs_info->fs_devices->num_devices;
2059 down_read(&fs_info->dev_replace.rwsem);
2060 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2061 ASSERT(num_devices > 1);
2064 up_read(&fs_info->dev_replace.rwsem);
2069 static void btrfs_scratch_superblock(struct btrfs_fs_info *fs_info,
2070 struct block_device *bdev, int copy_num)
2072 struct btrfs_super_block *disk_super;
2073 const size_t len = sizeof(disk_super->magic);
2074 const u64 bytenr = btrfs_sb_offset(copy_num);
2077 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr);
2078 if (IS_ERR(disk_super))
2081 memset(&disk_super->magic, 0, len);
2082 folio_mark_dirty(virt_to_folio(disk_super));
2083 btrfs_release_disk_super(disk_super);
2085 ret = sync_blockdev_range(bdev, bytenr, bytenr + len - 1);
2087 btrfs_warn(fs_info, "error clearing superblock number %d (%d)",
2091 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, struct btrfs_device *device)
2094 struct block_device *bdev = device->bdev;
2099 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2100 if (bdev_is_zoned(bdev))
2101 btrfs_reset_sb_log_zones(bdev, copy_num);
2103 btrfs_scratch_superblock(fs_info, bdev, copy_num);
2106 /* Notify udev that device has changed */
2107 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2109 /* Update ctime/mtime for device path for libblkid */
2110 update_dev_time(device->name->str);
2113 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
2114 struct btrfs_dev_lookup_args *args,
2115 struct file **bdev_file)
2117 struct btrfs_trans_handle *trans;
2118 struct btrfs_device *device;
2119 struct btrfs_fs_devices *cur_devices;
2120 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2124 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2125 btrfs_err(fs_info, "device remove not supported on extent tree v2 yet");
2130 * The device list in fs_devices is accessed without locks (neither
2131 * uuid_mutex nor device_list_mutex) as it won't change on a mounted
2132 * filesystem and another device rm cannot run.
2134 num_devices = btrfs_num_devices(fs_info);
2136 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2140 device = btrfs_find_device(fs_info->fs_devices, args);
2143 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2149 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2150 btrfs_warn_in_rcu(fs_info,
2151 "cannot remove device %s (devid %llu) due to active swapfile",
2152 btrfs_dev_name(device), device->devid);
2156 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
2157 return BTRFS_ERROR_DEV_TGT_REPLACE;
2159 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2160 fs_info->fs_devices->rw_devices == 1)
2161 return BTRFS_ERROR_DEV_ONLY_WRITABLE;
2163 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2164 mutex_lock(&fs_info->chunk_mutex);
2165 list_del_init(&device->dev_alloc_list);
2166 device->fs_devices->rw_devices--;
2167 mutex_unlock(&fs_info->chunk_mutex);
2170 ret = btrfs_shrink_device(device, 0);
2174 trans = btrfs_start_transaction(fs_info->chunk_root, 0);
2175 if (IS_ERR(trans)) {
2176 ret = PTR_ERR(trans);
2180 ret = btrfs_rm_dev_item(trans, device);
2182 /* Any error in dev item removal is critical */
2184 "failed to remove device item for devid %llu: %d",
2185 device->devid, ret);
2186 btrfs_abort_transaction(trans, ret);
2187 btrfs_end_transaction(trans);
2191 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2192 btrfs_scrub_cancel_dev(device);
2195 * the device list mutex makes sure that we don't change
2196 * the device list while someone else is writing out all
2197 * the device supers. Whoever is writing all supers, should
2198 * lock the device list mutex before getting the number of
2199 * devices in the super block (super_copy). Conversely,
2200 * whoever updates the number of devices in the super block
2201 * (super_copy) should hold the device list mutex.
2205 * In normal cases the cur_devices == fs_devices. But in case
2206 * of deleting a seed device, the cur_devices should point to
2207 * its own fs_devices listed under the fs_devices->seed_list.
2209 cur_devices = device->fs_devices;
2210 mutex_lock(&fs_devices->device_list_mutex);
2211 list_del_rcu(&device->dev_list);
2213 cur_devices->num_devices--;
2214 cur_devices->total_devices--;
2215 /* Update total_devices of the parent fs_devices if it's seed */
2216 if (cur_devices != fs_devices)
2217 fs_devices->total_devices--;
2219 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2220 cur_devices->missing_devices--;
2222 btrfs_assign_next_active_device(device, NULL);
2224 if (device->bdev_file) {
2225 cur_devices->open_devices--;
2226 /* remove sysfs entry */
2227 btrfs_sysfs_remove_device(device);
2230 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2231 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2232 mutex_unlock(&fs_devices->device_list_mutex);
2235 * At this point, the device is zero sized and detached from the
2236 * devices list. All that's left is to zero out the old supers and
2239 * We cannot call btrfs_close_bdev() here because we're holding the sb
2240 * write lock, and fput() on the block device will pull in the
2241 * ->open_mutex on the block device and it's dependencies. Instead
2242 * just flush the device and let the caller do the final bdev_release.
2244 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2245 btrfs_scratch_superblocks(fs_info, device);
2247 sync_blockdev(device->bdev);
2248 invalidate_bdev(device->bdev);
2252 *bdev_file = device->bdev_file;
2254 btrfs_free_device(device);
2257 * This can happen if cur_devices is the private seed devices list. We
2258 * cannot call close_fs_devices() here because it expects the uuid_mutex
2259 * to be held, but in fact we don't need that for the private
2260 * seed_devices, we can simply decrement cur_devices->opened and then
2261 * remove it from our list and free the fs_devices.
2263 if (cur_devices->num_devices == 0) {
2264 list_del_init(&cur_devices->seed_list);
2265 ASSERT(cur_devices->opened == 1);
2266 cur_devices->opened--;
2267 free_fs_devices(cur_devices);
2270 ret = btrfs_commit_transaction(trans);
2275 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2276 mutex_lock(&fs_info->chunk_mutex);
2277 list_add(&device->dev_alloc_list,
2278 &fs_devices->alloc_list);
2279 device->fs_devices->rw_devices++;
2280 mutex_unlock(&fs_info->chunk_mutex);
2285 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2287 struct btrfs_fs_devices *fs_devices;
2289 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2292 * in case of fs with no seed, srcdev->fs_devices will point
2293 * to fs_devices of fs_info. However when the dev being replaced is
2294 * a seed dev it will point to the seed's local fs_devices. In short
2295 * srcdev will have its correct fs_devices in both the cases.
2297 fs_devices = srcdev->fs_devices;
2299 list_del_rcu(&srcdev->dev_list);
2300 list_del(&srcdev->dev_alloc_list);
2301 fs_devices->num_devices--;
2302 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2303 fs_devices->missing_devices--;
2305 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2306 fs_devices->rw_devices--;
2309 fs_devices->open_devices--;
2312 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2314 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2316 mutex_lock(&uuid_mutex);
2318 btrfs_close_bdev(srcdev);
2320 btrfs_free_device(srcdev);
2322 /* if this is no devs we rather delete the fs_devices */
2323 if (!fs_devices->num_devices) {
2325 * On a mounted FS, num_devices can't be zero unless it's a
2326 * seed. In case of a seed device being replaced, the replace
2327 * target added to the sprout FS, so there will be no more
2328 * device left under the seed FS.
2330 ASSERT(fs_devices->seeding);
2332 list_del_init(&fs_devices->seed_list);
2333 close_fs_devices(fs_devices);
2334 free_fs_devices(fs_devices);
2336 mutex_unlock(&uuid_mutex);
2339 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2341 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2343 mutex_lock(&fs_devices->device_list_mutex);
2345 btrfs_sysfs_remove_device(tgtdev);
2348 fs_devices->open_devices--;
2350 fs_devices->num_devices--;
2352 btrfs_assign_next_active_device(tgtdev, NULL);
2354 list_del_rcu(&tgtdev->dev_list);
2356 mutex_unlock(&fs_devices->device_list_mutex);
2358 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev);
2360 btrfs_close_bdev(tgtdev);
2362 btrfs_free_device(tgtdev);
2366 * Populate args from device at path.
2368 * @fs_info: the filesystem
2369 * @args: the args to populate
2370 * @path: the path to the device
2372 * This will read the super block of the device at @path and populate @args with
2373 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to
2374 * lookup a device to operate on, but need to do it before we take any locks.
2375 * This properly handles the special case of "missing" that a user may pass in,
2376 * and does some basic sanity checks. The caller must make sure that @path is
2377 * properly NUL terminated before calling in, and must call
2378 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and
2381 * Return: 0 for success, -errno for failure
2383 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,
2384 struct btrfs_dev_lookup_args *args,
2387 struct btrfs_super_block *disk_super;
2388 struct file *bdev_file;
2391 if (!path || !path[0])
2393 if (!strcmp(path, "missing")) {
2394 args->missing = true;
2398 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL);
2399 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL);
2400 if (!args->uuid || !args->fsid) {
2401 btrfs_put_dev_args_from_path(args);
2405 ret = btrfs_get_bdev_and_sb(path, BLK_OPEN_READ, NULL, 0,
2406 &bdev_file, &disk_super);
2408 btrfs_put_dev_args_from_path(args);
2412 args->devid = btrfs_stack_device_id(&disk_super->dev_item);
2413 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE);
2414 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2415 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE);
2417 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
2418 btrfs_release_disk_super(disk_super);
2424 * Only use this jointly with btrfs_get_dev_args_from_path() because we will
2425 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables
2426 * that don't need to be freed.
2428 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args)
2436 struct btrfs_device *btrfs_find_device_by_devspec(
2437 struct btrfs_fs_info *fs_info, u64 devid,
2438 const char *device_path)
2440 BTRFS_DEV_LOOKUP_ARGS(args);
2441 struct btrfs_device *device;
2446 device = btrfs_find_device(fs_info->fs_devices, &args);
2448 return ERR_PTR(-ENOENT);
2452 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path);
2454 return ERR_PTR(ret);
2455 device = btrfs_find_device(fs_info->fs_devices, &args);
2456 btrfs_put_dev_args_from_path(&args);
2458 return ERR_PTR(-ENOENT);
2462 static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info)
2464 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2465 struct btrfs_fs_devices *old_devices;
2466 struct btrfs_fs_devices *seed_devices;
2468 lockdep_assert_held(&uuid_mutex);
2469 if (!fs_devices->seeding)
2470 return ERR_PTR(-EINVAL);
2473 * Private copy of the seed devices, anchored at
2474 * fs_info->fs_devices->seed_list
2476 seed_devices = alloc_fs_devices(NULL);
2477 if (IS_ERR(seed_devices))
2478 return seed_devices;
2481 * It's necessary to retain a copy of the original seed fs_devices in
2482 * fs_uuids so that filesystems which have been seeded can successfully
2483 * reference the seed device from open_seed_devices. This also supports
2486 old_devices = clone_fs_devices(fs_devices);
2487 if (IS_ERR(old_devices)) {
2488 kfree(seed_devices);
2492 list_add(&old_devices->fs_list, &fs_uuids);
2494 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2495 seed_devices->opened = 1;
2496 INIT_LIST_HEAD(&seed_devices->devices);
2497 INIT_LIST_HEAD(&seed_devices->alloc_list);
2498 mutex_init(&seed_devices->device_list_mutex);
2500 return seed_devices;
2504 * Splice seed devices into the sprout fs_devices.
2505 * Generate a new fsid for the sprouted read-write filesystem.
2507 static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info,
2508 struct btrfs_fs_devices *seed_devices)
2510 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2511 struct btrfs_super_block *disk_super = fs_info->super_copy;
2512 struct btrfs_device *device;
2516 * We are updating the fsid, the thread leading to device_list_add()
2517 * could race, so uuid_mutex is needed.
2519 lockdep_assert_held(&uuid_mutex);
2522 * The threads listed below may traverse dev_list but can do that without
2523 * device_list_mutex:
2524 * - All device ops and balance - as we are in btrfs_exclop_start.
2525 * - Various dev_list readers - are using RCU.
2526 * - btrfs_ioctl_fitrim() - is using RCU.
2528 * For-read threads as below are using device_list_mutex:
2529 * - Readonly scrub btrfs_scrub_dev()
2530 * - Readonly scrub btrfs_scrub_progress()
2531 * - btrfs_get_dev_stats()
2533 lockdep_assert_held(&fs_devices->device_list_mutex);
2535 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2537 list_for_each_entry(device, &seed_devices->devices, dev_list)
2538 device->fs_devices = seed_devices;
2540 fs_devices->seeding = false;
2541 fs_devices->num_devices = 0;
2542 fs_devices->open_devices = 0;
2543 fs_devices->missing_devices = 0;
2544 fs_devices->rotating = false;
2545 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2547 generate_random_uuid(fs_devices->fsid);
2548 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2549 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2551 super_flags = btrfs_super_flags(disk_super) &
2552 ~BTRFS_SUPER_FLAG_SEEDING;
2553 btrfs_set_super_flags(disk_super, super_flags);
2557 * Store the expected generation for seed devices in device items.
2559 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2561 BTRFS_DEV_LOOKUP_ARGS(args);
2562 struct btrfs_fs_info *fs_info = trans->fs_info;
2563 struct btrfs_root *root = fs_info->chunk_root;
2564 struct btrfs_path *path;
2565 struct extent_buffer *leaf;
2566 struct btrfs_dev_item *dev_item;
2567 struct btrfs_device *device;
2568 struct btrfs_key key;
2569 u8 fs_uuid[BTRFS_FSID_SIZE];
2570 u8 dev_uuid[BTRFS_UUID_SIZE];
2573 path = btrfs_alloc_path();
2577 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2579 key.type = BTRFS_DEV_ITEM_KEY;
2582 btrfs_reserve_chunk_metadata(trans, false);
2583 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2584 btrfs_trans_release_chunk_metadata(trans);
2588 leaf = path->nodes[0];
2590 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2591 ret = btrfs_next_leaf(root, path);
2596 leaf = path->nodes[0];
2597 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2598 btrfs_release_path(path);
2602 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2603 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2604 key.type != BTRFS_DEV_ITEM_KEY)
2607 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2608 struct btrfs_dev_item);
2609 args.devid = btrfs_device_id(leaf, dev_item);
2610 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2612 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2614 args.uuid = dev_uuid;
2615 args.fsid = fs_uuid;
2616 device = btrfs_find_device(fs_info->fs_devices, &args);
2617 BUG_ON(!device); /* Logic error */
2619 if (device->fs_devices->seeding) {
2620 btrfs_set_device_generation(leaf, dev_item,
2621 device->generation);
2622 btrfs_mark_buffer_dirty(trans, leaf);
2630 btrfs_free_path(path);
2634 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2636 struct btrfs_root *root = fs_info->dev_root;
2637 struct btrfs_trans_handle *trans;
2638 struct btrfs_device *device;
2639 struct file *bdev_file;
2640 struct super_block *sb = fs_info->sb;
2641 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2642 struct btrfs_fs_devices *seed_devices = NULL;
2643 u64 orig_super_total_bytes;
2644 u64 orig_super_num_devices;
2646 bool seeding_dev = false;
2647 bool locked = false;
2649 if (sb_rdonly(sb) && !fs_devices->seeding)
2652 bdev_file = bdev_file_open_by_path(device_path, BLK_OPEN_WRITE,
2653 fs_info->bdev_holder, NULL);
2654 if (IS_ERR(bdev_file))
2655 return PTR_ERR(bdev_file);
2657 if (!btrfs_check_device_zone_type(fs_info, file_bdev(bdev_file))) {
2662 if (fs_devices->seeding) {
2664 down_write(&sb->s_umount);
2665 mutex_lock(&uuid_mutex);
2669 sync_blockdev(file_bdev(bdev_file));
2672 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2673 if (device->bdev == file_bdev(bdev_file)) {
2681 device = btrfs_alloc_device(fs_info, NULL, NULL, device_path);
2682 if (IS_ERR(device)) {
2683 /* we can safely leave the fs_devices entry around */
2684 ret = PTR_ERR(device);
2688 device->fs_info = fs_info;
2689 device->bdev_file = bdev_file;
2690 device->bdev = file_bdev(bdev_file);
2691 ret = lookup_bdev(device_path, &device->devt);
2693 goto error_free_device;
2695 ret = btrfs_get_dev_zone_info(device, false);
2697 goto error_free_device;
2699 trans = btrfs_start_transaction(root, 0);
2700 if (IS_ERR(trans)) {
2701 ret = PTR_ERR(trans);
2702 goto error_free_zone;
2705 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2706 device->generation = trans->transid;
2707 device->io_width = fs_info->sectorsize;
2708 device->io_align = fs_info->sectorsize;
2709 device->sector_size = fs_info->sectorsize;
2710 device->total_bytes =
2711 round_down(bdev_nr_bytes(device->bdev), fs_info->sectorsize);
2712 device->disk_total_bytes = device->total_bytes;
2713 device->commit_total_bytes = device->total_bytes;
2714 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2715 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2716 device->dev_stats_valid = 1;
2717 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2720 btrfs_clear_sb_rdonly(sb);
2722 /* GFP_KERNEL allocation must not be under device_list_mutex */
2723 seed_devices = btrfs_init_sprout(fs_info);
2724 if (IS_ERR(seed_devices)) {
2725 ret = PTR_ERR(seed_devices);
2726 btrfs_abort_transaction(trans, ret);
2731 mutex_lock(&fs_devices->device_list_mutex);
2733 btrfs_setup_sprout(fs_info, seed_devices);
2734 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev,
2738 device->fs_devices = fs_devices;
2740 mutex_lock(&fs_info->chunk_mutex);
2741 list_add_rcu(&device->dev_list, &fs_devices->devices);
2742 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2743 fs_devices->num_devices++;
2744 fs_devices->open_devices++;
2745 fs_devices->rw_devices++;
2746 fs_devices->total_devices++;
2747 fs_devices->total_rw_bytes += device->total_bytes;
2749 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2751 if (!bdev_nonrot(device->bdev))
2752 fs_devices->rotating = true;
2754 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2755 btrfs_set_super_total_bytes(fs_info->super_copy,
2756 round_down(orig_super_total_bytes + device->total_bytes,
2757 fs_info->sectorsize));
2759 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2760 btrfs_set_super_num_devices(fs_info->super_copy,
2761 orig_super_num_devices + 1);
2764 * we've got more storage, clear any full flags on the space
2767 btrfs_clear_space_info_full(fs_info);
2769 mutex_unlock(&fs_info->chunk_mutex);
2771 /* Add sysfs device entry */
2772 btrfs_sysfs_add_device(device);
2774 mutex_unlock(&fs_devices->device_list_mutex);
2777 mutex_lock(&fs_info->chunk_mutex);
2778 ret = init_first_rw_device(trans);
2779 mutex_unlock(&fs_info->chunk_mutex);
2781 btrfs_abort_transaction(trans, ret);
2786 ret = btrfs_add_dev_item(trans, device);
2788 btrfs_abort_transaction(trans, ret);
2793 ret = btrfs_finish_sprout(trans);
2795 btrfs_abort_transaction(trans, ret);
2800 * fs_devices now represents the newly sprouted filesystem and
2801 * its fsid has been changed by btrfs_sprout_splice().
2803 btrfs_sysfs_update_sprout_fsid(fs_devices);
2806 ret = btrfs_commit_transaction(trans);
2809 mutex_unlock(&uuid_mutex);
2810 up_write(&sb->s_umount);
2813 if (ret) /* transaction commit */
2816 ret = btrfs_relocate_sys_chunks(fs_info);
2818 btrfs_handle_fs_error(fs_info, ret,
2819 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2820 trans = btrfs_attach_transaction(root);
2821 if (IS_ERR(trans)) {
2822 if (PTR_ERR(trans) == -ENOENT)
2824 ret = PTR_ERR(trans);
2828 ret = btrfs_commit_transaction(trans);
2832 * Now that we have written a new super block to this device, check all
2833 * other fs_devices list if device_path alienates any other scanned
2835 * We can ignore the return value as it typically returns -EINVAL and
2836 * only succeeds if the device was an alien.
2838 btrfs_forget_devices(device->devt);
2840 /* Update ctime/mtime for blkid or udev */
2841 update_dev_time(device_path);
2846 btrfs_sysfs_remove_device(device);
2847 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2848 mutex_lock(&fs_info->chunk_mutex);
2849 list_del_rcu(&device->dev_list);
2850 list_del(&device->dev_alloc_list);
2851 fs_info->fs_devices->num_devices--;
2852 fs_info->fs_devices->open_devices--;
2853 fs_info->fs_devices->rw_devices--;
2854 fs_info->fs_devices->total_devices--;
2855 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2856 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2857 btrfs_set_super_total_bytes(fs_info->super_copy,
2858 orig_super_total_bytes);
2859 btrfs_set_super_num_devices(fs_info->super_copy,
2860 orig_super_num_devices);
2861 mutex_unlock(&fs_info->chunk_mutex);
2862 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2865 btrfs_set_sb_rdonly(sb);
2867 btrfs_end_transaction(trans);
2869 btrfs_destroy_dev_zone_info(device);
2871 btrfs_free_device(device);
2875 mutex_unlock(&uuid_mutex);
2876 up_write(&sb->s_umount);
2881 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2882 struct btrfs_device *device)
2885 struct btrfs_path *path;
2886 struct btrfs_root *root = device->fs_info->chunk_root;
2887 struct btrfs_dev_item *dev_item;
2888 struct extent_buffer *leaf;
2889 struct btrfs_key key;
2891 path = btrfs_alloc_path();
2895 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2896 key.type = BTRFS_DEV_ITEM_KEY;
2897 key.offset = device->devid;
2899 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2908 leaf = path->nodes[0];
2909 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2911 btrfs_set_device_id(leaf, dev_item, device->devid);
2912 btrfs_set_device_type(leaf, dev_item, device->type);
2913 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2914 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2915 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2916 btrfs_set_device_total_bytes(leaf, dev_item,
2917 btrfs_device_get_disk_total_bytes(device));
2918 btrfs_set_device_bytes_used(leaf, dev_item,
2919 btrfs_device_get_bytes_used(device));
2920 btrfs_mark_buffer_dirty(trans, leaf);
2923 btrfs_free_path(path);
2927 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2928 struct btrfs_device *device, u64 new_size)
2930 struct btrfs_fs_info *fs_info = device->fs_info;
2931 struct btrfs_super_block *super_copy = fs_info->super_copy;
2936 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2939 new_size = round_down(new_size, fs_info->sectorsize);
2941 mutex_lock(&fs_info->chunk_mutex);
2942 old_total = btrfs_super_total_bytes(super_copy);
2943 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2945 if (new_size <= device->total_bytes ||
2946 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2947 mutex_unlock(&fs_info->chunk_mutex);
2951 btrfs_set_super_total_bytes(super_copy,
2952 round_down(old_total + diff, fs_info->sectorsize));
2953 device->fs_devices->total_rw_bytes += diff;
2954 atomic64_add(diff, &fs_info->free_chunk_space);
2956 btrfs_device_set_total_bytes(device, new_size);
2957 btrfs_device_set_disk_total_bytes(device, new_size);
2958 btrfs_clear_space_info_full(device->fs_info);
2959 if (list_empty(&device->post_commit_list))
2960 list_add_tail(&device->post_commit_list,
2961 &trans->transaction->dev_update_list);
2962 mutex_unlock(&fs_info->chunk_mutex);
2964 btrfs_reserve_chunk_metadata(trans, false);
2965 ret = btrfs_update_device(trans, device);
2966 btrfs_trans_release_chunk_metadata(trans);
2971 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2973 struct btrfs_fs_info *fs_info = trans->fs_info;
2974 struct btrfs_root *root = fs_info->chunk_root;
2976 struct btrfs_path *path;
2977 struct btrfs_key key;
2979 path = btrfs_alloc_path();
2983 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2984 key.offset = chunk_offset;
2985 key.type = BTRFS_CHUNK_ITEM_KEY;
2987 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2990 else if (ret > 0) { /* Logic error or corruption */
2991 btrfs_handle_fs_error(fs_info, -ENOENT,
2992 "Failed lookup while freeing chunk.");
2997 ret = btrfs_del_item(trans, root, path);
2999 btrfs_handle_fs_error(fs_info, ret,
3000 "Failed to delete chunk item.");
3002 btrfs_free_path(path);
3006 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3008 struct btrfs_super_block *super_copy = fs_info->super_copy;
3009 struct btrfs_disk_key *disk_key;
3010 struct btrfs_chunk *chunk;
3017 struct btrfs_key key;
3019 lockdep_assert_held(&fs_info->chunk_mutex);
3020 array_size = btrfs_super_sys_array_size(super_copy);
3022 ptr = super_copy->sys_chunk_array;
3025 while (cur < array_size) {
3026 disk_key = (struct btrfs_disk_key *)ptr;
3027 btrfs_disk_key_to_cpu(&key, disk_key);
3029 len = sizeof(*disk_key);
3031 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3032 chunk = (struct btrfs_chunk *)(ptr + len);
3033 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
3034 len += btrfs_chunk_item_size(num_stripes);
3039 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
3040 key.offset == chunk_offset) {
3041 memmove(ptr, ptr + len, array_size - (cur + len));
3043 btrfs_set_super_sys_array_size(super_copy, array_size);
3052 struct btrfs_chunk_map *btrfs_find_chunk_map_nolock(struct btrfs_fs_info *fs_info,
3053 u64 logical, u64 length)
3055 struct rb_node *node = fs_info->mapping_tree.rb_root.rb_node;
3056 struct rb_node *prev = NULL;
3057 struct rb_node *orig_prev;
3058 struct btrfs_chunk_map *map;
3059 struct btrfs_chunk_map *prev_map = NULL;
3062 map = rb_entry(node, struct btrfs_chunk_map, rb_node);
3066 if (logical < map->start) {
3067 node = node->rb_left;
3068 } else if (logical >= map->start + map->chunk_len) {
3069 node = node->rb_right;
3071 refcount_inc(&map->refs);
3080 while (prev && logical >= prev_map->start + prev_map->chunk_len) {
3081 prev = rb_next(prev);
3082 prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node);
3087 prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node);
3088 while (prev && logical < prev_map->start) {
3089 prev = rb_prev(prev);
3090 prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node);
3095 u64 end = logical + length;
3098 * Caller can pass a U64_MAX length when it wants to get any
3099 * chunk starting at an offset of 'logical' or higher, so deal
3100 * with underflow by resetting the end offset to U64_MAX.
3105 if (end > prev_map->start &&
3106 logical < prev_map->start + prev_map->chunk_len) {
3107 refcount_inc(&prev_map->refs);
3115 struct btrfs_chunk_map *btrfs_find_chunk_map(struct btrfs_fs_info *fs_info,
3116 u64 logical, u64 length)
3118 struct btrfs_chunk_map *map;
3120 read_lock(&fs_info->mapping_tree_lock);
3121 map = btrfs_find_chunk_map_nolock(fs_info, logical, length);
3122 read_unlock(&fs_info->mapping_tree_lock);
3128 * Find the mapping containing the given logical extent.
3130 * @logical: Logical block offset in bytes.
3131 * @length: Length of extent in bytes.
3133 * Return: Chunk mapping or ERR_PTR.
3135 struct btrfs_chunk_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
3136 u64 logical, u64 length)
3138 struct btrfs_chunk_map *map;
3140 map = btrfs_find_chunk_map(fs_info, logical, length);
3142 if (unlikely(!map)) {
3144 "unable to find chunk map for logical %llu length %llu",
3146 return ERR_PTR(-EINVAL);
3149 if (unlikely(map->start > logical || map->start + map->chunk_len <= logical)) {
3151 "found a bad chunk map, wanted %llu-%llu, found %llu-%llu",
3152 logical, logical + length, map->start,
3153 map->start + map->chunk_len);
3154 btrfs_free_chunk_map(map);
3155 return ERR_PTR(-EINVAL);
3158 /* Callers are responsible for dropping the reference. */
3162 static int remove_chunk_item(struct btrfs_trans_handle *trans,
3163 struct btrfs_chunk_map *map, u64 chunk_offset)
3168 * Removing chunk items and updating the device items in the chunks btree
3169 * requires holding the chunk_mutex.
3170 * See the comment at btrfs_chunk_alloc() for the details.
3172 lockdep_assert_held(&trans->fs_info->chunk_mutex);
3174 for (i = 0; i < map->num_stripes; i++) {
3177 ret = btrfs_update_device(trans, map->stripes[i].dev);
3182 return btrfs_free_chunk(trans, chunk_offset);
3185 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
3187 struct btrfs_fs_info *fs_info = trans->fs_info;
3188 struct btrfs_chunk_map *map;
3189 u64 dev_extent_len = 0;
3191 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3193 map = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
3196 * This is a logic error, but we don't want to just rely on the
3197 * user having built with ASSERT enabled, so if ASSERT doesn't
3198 * do anything we still error out.
3201 return PTR_ERR(map);
3205 * First delete the device extent items from the devices btree.
3206 * We take the device_list_mutex to avoid racing with the finishing phase
3207 * of a device replace operation. See the comment below before acquiring
3208 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
3209 * because that can result in a deadlock when deleting the device extent
3210 * items from the devices btree - COWing an extent buffer from the btree
3211 * may result in allocating a new metadata chunk, which would attempt to
3212 * lock again fs_info->chunk_mutex.
3214 mutex_lock(&fs_devices->device_list_mutex);
3215 for (i = 0; i < map->num_stripes; i++) {
3216 struct btrfs_device *device = map->stripes[i].dev;
3217 ret = btrfs_free_dev_extent(trans, device,
3218 map->stripes[i].physical,
3221 mutex_unlock(&fs_devices->device_list_mutex);
3222 btrfs_abort_transaction(trans, ret);
3226 if (device->bytes_used > 0) {
3227 mutex_lock(&fs_info->chunk_mutex);
3228 btrfs_device_set_bytes_used(device,
3229 device->bytes_used - dev_extent_len);
3230 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3231 btrfs_clear_space_info_full(fs_info);
3232 mutex_unlock(&fs_info->chunk_mutex);
3235 mutex_unlock(&fs_devices->device_list_mutex);
3238 * We acquire fs_info->chunk_mutex for 2 reasons:
3240 * 1) Just like with the first phase of the chunk allocation, we must
3241 * reserve system space, do all chunk btree updates and deletions, and
3242 * update the system chunk array in the superblock while holding this
3243 * mutex. This is for similar reasons as explained on the comment at
3244 * the top of btrfs_chunk_alloc();
3246 * 2) Prevent races with the final phase of a device replace operation
3247 * that replaces the device object associated with the map's stripes,
3248 * because the device object's id can change at any time during that
3249 * final phase of the device replace operation
3250 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3251 * replaced device and then see it with an ID of
3252 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3253 * the device item, which does not exists on the chunk btree.
3254 * The finishing phase of device replace acquires both the
3255 * device_list_mutex and the chunk_mutex, in that order, so we are
3256 * safe by just acquiring the chunk_mutex.
3258 trans->removing_chunk = true;
3259 mutex_lock(&fs_info->chunk_mutex);
3261 check_system_chunk(trans, map->type);
3263 ret = remove_chunk_item(trans, map, chunk_offset);
3265 * Normally we should not get -ENOSPC since we reserved space before
3266 * through the call to check_system_chunk().
3268 * Despite our system space_info having enough free space, we may not
3269 * be able to allocate extents from its block groups, because all have
3270 * an incompatible profile, which will force us to allocate a new system
3271 * block group with the right profile, or right after we called
3272 * check_system_space() above, a scrub turned the only system block group
3273 * with enough free space into RO mode.
3274 * This is explained with more detail at do_chunk_alloc().
3276 * So if we get -ENOSPC, allocate a new system chunk and retry once.
3278 if (ret == -ENOSPC) {
3279 const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
3280 struct btrfs_block_group *sys_bg;
3282 sys_bg = btrfs_create_chunk(trans, sys_flags);
3283 if (IS_ERR(sys_bg)) {
3284 ret = PTR_ERR(sys_bg);
3285 btrfs_abort_transaction(trans, ret);
3289 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3291 btrfs_abort_transaction(trans, ret);
3295 ret = remove_chunk_item(trans, map, chunk_offset);
3297 btrfs_abort_transaction(trans, ret);
3301 btrfs_abort_transaction(trans, ret);
3305 trace_btrfs_chunk_free(fs_info, map, chunk_offset, map->chunk_len);
3307 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3308 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3310 btrfs_abort_transaction(trans, ret);
3315 mutex_unlock(&fs_info->chunk_mutex);
3316 trans->removing_chunk = false;
3319 * We are done with chunk btree updates and deletions, so release the
3320 * system space we previously reserved (with check_system_chunk()).
3322 btrfs_trans_release_chunk_metadata(trans);
3324 ret = btrfs_remove_block_group(trans, map);
3326 btrfs_abort_transaction(trans, ret);
3331 if (trans->removing_chunk) {
3332 mutex_unlock(&fs_info->chunk_mutex);
3333 trans->removing_chunk = false;
3336 btrfs_free_chunk_map(map);
3340 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3342 struct btrfs_root *root = fs_info->chunk_root;
3343 struct btrfs_trans_handle *trans;
3344 struct btrfs_block_group *block_group;
3348 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3350 "relocate: not supported on extent tree v2 yet");
3355 * Prevent races with automatic removal of unused block groups.
3356 * After we relocate and before we remove the chunk with offset
3357 * chunk_offset, automatic removal of the block group can kick in,
3358 * resulting in a failure when calling btrfs_remove_chunk() below.
3360 * Make sure to acquire this mutex before doing a tree search (dev
3361 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3362 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3363 * we release the path used to search the chunk/dev tree and before
3364 * the current task acquires this mutex and calls us.
3366 lockdep_assert_held(&fs_info->reclaim_bgs_lock);
3368 /* step one, relocate all the extents inside this chunk */
3369 btrfs_scrub_pause(fs_info);
3370 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3371 btrfs_scrub_continue(fs_info);
3374 * If we had a transaction abort, stop all running scrubs.
3375 * See transaction.c:cleanup_transaction() why we do it here.
3377 if (BTRFS_FS_ERROR(fs_info))
3378 btrfs_scrub_cancel(fs_info);
3382 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3385 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3386 length = block_group->length;
3387 btrfs_put_block_group(block_group);
3390 * On a zoned file system, discard the whole block group, this will
3391 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3392 * resetting the zone fails, don't treat it as a fatal problem from the
3393 * filesystem's point of view.
3395 if (btrfs_is_zoned(fs_info)) {
3396 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
3399 "failed to reset zone %llu after relocation",
3403 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3405 if (IS_ERR(trans)) {
3406 ret = PTR_ERR(trans);
3407 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3412 * step two, delete the device extents and the
3413 * chunk tree entries
3415 ret = btrfs_remove_chunk(trans, chunk_offset);
3416 btrfs_end_transaction(trans);
3420 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3422 struct btrfs_root *chunk_root = fs_info->chunk_root;
3423 struct btrfs_path *path;
3424 struct extent_buffer *leaf;
3425 struct btrfs_chunk *chunk;
3426 struct btrfs_key key;
3427 struct btrfs_key found_key;
3429 bool retried = false;
3433 path = btrfs_alloc_path();
3438 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3439 key.offset = (u64)-1;
3440 key.type = BTRFS_CHUNK_ITEM_KEY;
3443 mutex_lock(&fs_info->reclaim_bgs_lock);
3444 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3446 mutex_unlock(&fs_info->reclaim_bgs_lock);
3451 * On the first search we would find chunk tree with
3452 * offset -1, which is not possible. On subsequent
3453 * loops this would find an existing item on an invalid
3454 * offset (one less than the previous one, wrong
3455 * alignment and size).
3461 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3464 mutex_unlock(&fs_info->reclaim_bgs_lock);
3470 leaf = path->nodes[0];
3471 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3473 chunk = btrfs_item_ptr(leaf, path->slots[0],
3474 struct btrfs_chunk);
3475 chunk_type = btrfs_chunk_type(leaf, chunk);
3476 btrfs_release_path(path);
3478 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3479 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3485 mutex_unlock(&fs_info->reclaim_bgs_lock);
3487 if (found_key.offset == 0)
3489 key.offset = found_key.offset - 1;
3492 if (failed && !retried) {
3496 } else if (WARN_ON(failed && retried)) {
3500 btrfs_free_path(path);
3505 * return 1 : allocate a data chunk successfully,
3506 * return <0: errors during allocating a data chunk,
3507 * return 0 : no need to allocate a data chunk.
3509 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3512 struct btrfs_block_group *cache;
3516 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3518 chunk_type = cache->flags;
3519 btrfs_put_block_group(cache);
3521 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3524 spin_lock(&fs_info->data_sinfo->lock);
3525 bytes_used = fs_info->data_sinfo->bytes_used;
3526 spin_unlock(&fs_info->data_sinfo->lock);
3529 struct btrfs_trans_handle *trans;
3532 trans = btrfs_join_transaction(fs_info->tree_root);
3534 return PTR_ERR(trans);
3536 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3537 btrfs_end_transaction(trans);
3546 static void btrfs_disk_balance_args_to_cpu(struct btrfs_balance_args *cpu,
3547 const struct btrfs_disk_balance_args *disk)
3549 memset(cpu, 0, sizeof(*cpu));
3551 cpu->profiles = le64_to_cpu(disk->profiles);
3552 cpu->usage = le64_to_cpu(disk->usage);
3553 cpu->devid = le64_to_cpu(disk->devid);
3554 cpu->pstart = le64_to_cpu(disk->pstart);
3555 cpu->pend = le64_to_cpu(disk->pend);
3556 cpu->vstart = le64_to_cpu(disk->vstart);
3557 cpu->vend = le64_to_cpu(disk->vend);
3558 cpu->target = le64_to_cpu(disk->target);
3559 cpu->flags = le64_to_cpu(disk->flags);
3560 cpu->limit = le64_to_cpu(disk->limit);
3561 cpu->stripes_min = le32_to_cpu(disk->stripes_min);
3562 cpu->stripes_max = le32_to_cpu(disk->stripes_max);
3565 static void btrfs_cpu_balance_args_to_disk(struct btrfs_disk_balance_args *disk,
3566 const struct btrfs_balance_args *cpu)
3568 memset(disk, 0, sizeof(*disk));
3570 disk->profiles = cpu_to_le64(cpu->profiles);
3571 disk->usage = cpu_to_le64(cpu->usage);
3572 disk->devid = cpu_to_le64(cpu->devid);
3573 disk->pstart = cpu_to_le64(cpu->pstart);
3574 disk->pend = cpu_to_le64(cpu->pend);
3575 disk->vstart = cpu_to_le64(cpu->vstart);
3576 disk->vend = cpu_to_le64(cpu->vend);
3577 disk->target = cpu_to_le64(cpu->target);
3578 disk->flags = cpu_to_le64(cpu->flags);
3579 disk->limit = cpu_to_le64(cpu->limit);
3580 disk->stripes_min = cpu_to_le32(cpu->stripes_min);
3581 disk->stripes_max = cpu_to_le32(cpu->stripes_max);
3584 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3585 struct btrfs_balance_control *bctl)
3587 struct btrfs_root *root = fs_info->tree_root;
3588 struct btrfs_trans_handle *trans;
3589 struct btrfs_balance_item *item;
3590 struct btrfs_disk_balance_args disk_bargs;
3591 struct btrfs_path *path;
3592 struct extent_buffer *leaf;
3593 struct btrfs_key key;
3596 path = btrfs_alloc_path();
3600 trans = btrfs_start_transaction(root, 0);
3601 if (IS_ERR(trans)) {
3602 btrfs_free_path(path);
3603 return PTR_ERR(trans);
3606 key.objectid = BTRFS_BALANCE_OBJECTID;
3607 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3610 ret = btrfs_insert_empty_item(trans, root, path, &key,
3615 leaf = path->nodes[0];
3616 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3618 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3620 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3621 btrfs_set_balance_data(leaf, item, &disk_bargs);
3622 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3623 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3624 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3625 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3627 btrfs_set_balance_flags(leaf, item, bctl->flags);
3629 btrfs_mark_buffer_dirty(trans, leaf);
3631 btrfs_free_path(path);
3632 err = btrfs_commit_transaction(trans);
3638 static int del_balance_item(struct btrfs_fs_info *fs_info)
3640 struct btrfs_root *root = fs_info->tree_root;
3641 struct btrfs_trans_handle *trans;
3642 struct btrfs_path *path;
3643 struct btrfs_key key;
3646 path = btrfs_alloc_path();
3650 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3651 if (IS_ERR(trans)) {
3652 btrfs_free_path(path);
3653 return PTR_ERR(trans);
3656 key.objectid = BTRFS_BALANCE_OBJECTID;
3657 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3660 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3668 ret = btrfs_del_item(trans, root, path);
3670 btrfs_free_path(path);
3671 err = btrfs_commit_transaction(trans);
3678 * This is a heuristic used to reduce the number of chunks balanced on
3679 * resume after balance was interrupted.
3681 static void update_balance_args(struct btrfs_balance_control *bctl)
3684 * Turn on soft mode for chunk types that were being converted.
3686 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3687 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3688 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3689 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3690 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3691 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3694 * Turn on usage filter if is not already used. The idea is
3695 * that chunks that we have already balanced should be
3696 * reasonably full. Don't do it for chunks that are being
3697 * converted - that will keep us from relocating unconverted
3698 * (albeit full) chunks.
3700 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3701 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3702 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3703 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3704 bctl->data.usage = 90;
3706 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3707 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3708 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3709 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3710 bctl->sys.usage = 90;
3712 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3713 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3714 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3715 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3716 bctl->meta.usage = 90;
3721 * Clear the balance status in fs_info and delete the balance item from disk.
3723 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3725 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3728 ASSERT(fs_info->balance_ctl);
3730 spin_lock(&fs_info->balance_lock);
3731 fs_info->balance_ctl = NULL;
3732 spin_unlock(&fs_info->balance_lock);
3735 ret = del_balance_item(fs_info);
3737 btrfs_handle_fs_error(fs_info, ret, NULL);
3741 * Balance filters. Return 1 if chunk should be filtered out
3742 * (should not be balanced).
3744 static int chunk_profiles_filter(u64 chunk_type,
3745 struct btrfs_balance_args *bargs)
3747 chunk_type = chunk_to_extended(chunk_type) &
3748 BTRFS_EXTENDED_PROFILE_MASK;
3750 if (bargs->profiles & chunk_type)
3756 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3757 struct btrfs_balance_args *bargs)
3759 struct btrfs_block_group *cache;
3761 u64 user_thresh_min;
3762 u64 user_thresh_max;
3765 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3766 chunk_used = cache->used;
3768 if (bargs->usage_min == 0)
3769 user_thresh_min = 0;
3771 user_thresh_min = mult_perc(cache->length, bargs->usage_min);
3773 if (bargs->usage_max == 0)
3774 user_thresh_max = 1;
3775 else if (bargs->usage_max > 100)
3776 user_thresh_max = cache->length;
3778 user_thresh_max = mult_perc(cache->length, bargs->usage_max);
3780 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3783 btrfs_put_block_group(cache);
3787 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3788 u64 chunk_offset, struct btrfs_balance_args *bargs)
3790 struct btrfs_block_group *cache;
3791 u64 chunk_used, user_thresh;
3794 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3795 chunk_used = cache->used;
3797 if (bargs->usage_min == 0)
3799 else if (bargs->usage > 100)
3800 user_thresh = cache->length;
3802 user_thresh = mult_perc(cache->length, bargs->usage);
3804 if (chunk_used < user_thresh)
3807 btrfs_put_block_group(cache);
3811 static int chunk_devid_filter(struct extent_buffer *leaf,
3812 struct btrfs_chunk *chunk,
3813 struct btrfs_balance_args *bargs)
3815 struct btrfs_stripe *stripe;
3816 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3819 for (i = 0; i < num_stripes; i++) {
3820 stripe = btrfs_stripe_nr(chunk, i);
3821 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3828 static u64 calc_data_stripes(u64 type, int num_stripes)
3830 const int index = btrfs_bg_flags_to_raid_index(type);
3831 const int ncopies = btrfs_raid_array[index].ncopies;
3832 const int nparity = btrfs_raid_array[index].nparity;
3834 return (num_stripes - nparity) / ncopies;
3837 /* [pstart, pend) */
3838 static int chunk_drange_filter(struct extent_buffer *leaf,
3839 struct btrfs_chunk *chunk,
3840 struct btrfs_balance_args *bargs)
3842 struct btrfs_stripe *stripe;
3843 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3850 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3853 type = btrfs_chunk_type(leaf, chunk);
3854 factor = calc_data_stripes(type, num_stripes);
3856 for (i = 0; i < num_stripes; i++) {
3857 stripe = btrfs_stripe_nr(chunk, i);
3858 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3861 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3862 stripe_length = btrfs_chunk_length(leaf, chunk);
3863 stripe_length = div_u64(stripe_length, factor);
3865 if (stripe_offset < bargs->pend &&
3866 stripe_offset + stripe_length > bargs->pstart)
3873 /* [vstart, vend) */
3874 static int chunk_vrange_filter(struct extent_buffer *leaf,
3875 struct btrfs_chunk *chunk,
3877 struct btrfs_balance_args *bargs)
3879 if (chunk_offset < bargs->vend &&
3880 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3881 /* at least part of the chunk is inside this vrange */
3887 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3888 struct btrfs_chunk *chunk,
3889 struct btrfs_balance_args *bargs)
3891 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3893 if (bargs->stripes_min <= num_stripes
3894 && num_stripes <= bargs->stripes_max)
3900 static int chunk_soft_convert_filter(u64 chunk_type,
3901 struct btrfs_balance_args *bargs)
3903 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3906 chunk_type = chunk_to_extended(chunk_type) &
3907 BTRFS_EXTENDED_PROFILE_MASK;
3909 if (bargs->target == chunk_type)
3915 static int should_balance_chunk(struct extent_buffer *leaf,
3916 struct btrfs_chunk *chunk, u64 chunk_offset)
3918 struct btrfs_fs_info *fs_info = leaf->fs_info;
3919 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3920 struct btrfs_balance_args *bargs = NULL;
3921 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3924 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3925 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3929 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3930 bargs = &bctl->data;
3931 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3933 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3934 bargs = &bctl->meta;
3936 /* profiles filter */
3937 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3938 chunk_profiles_filter(chunk_type, bargs)) {
3943 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3944 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3946 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3947 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3952 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3953 chunk_devid_filter(leaf, chunk, bargs)) {
3957 /* drange filter, makes sense only with devid filter */
3958 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3959 chunk_drange_filter(leaf, chunk, bargs)) {
3964 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3965 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3969 /* stripes filter */
3970 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3971 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3975 /* soft profile changing mode */
3976 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3977 chunk_soft_convert_filter(chunk_type, bargs)) {
3982 * limited by count, must be the last filter
3984 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3985 if (bargs->limit == 0)
3989 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3991 * Same logic as the 'limit' filter; the minimum cannot be
3992 * determined here because we do not have the global information
3993 * about the count of all chunks that satisfy the filters.
3995 if (bargs->limit_max == 0)
4004 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
4006 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4007 struct btrfs_root *chunk_root = fs_info->chunk_root;
4009 struct btrfs_chunk *chunk;
4010 struct btrfs_path *path = NULL;
4011 struct btrfs_key key;
4012 struct btrfs_key found_key;
4013 struct extent_buffer *leaf;
4016 int enospc_errors = 0;
4017 bool counting = true;
4018 /* The single value limit and min/max limits use the same bytes in the */
4019 u64 limit_data = bctl->data.limit;
4020 u64 limit_meta = bctl->meta.limit;
4021 u64 limit_sys = bctl->sys.limit;
4025 int chunk_reserved = 0;
4027 path = btrfs_alloc_path();
4033 /* zero out stat counters */
4034 spin_lock(&fs_info->balance_lock);
4035 memset(&bctl->stat, 0, sizeof(bctl->stat));
4036 spin_unlock(&fs_info->balance_lock);
4040 * The single value limit and min/max limits use the same bytes
4043 bctl->data.limit = limit_data;
4044 bctl->meta.limit = limit_meta;
4045 bctl->sys.limit = limit_sys;
4047 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4048 key.offset = (u64)-1;
4049 key.type = BTRFS_CHUNK_ITEM_KEY;
4052 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
4053 atomic_read(&fs_info->balance_cancel_req)) {
4058 mutex_lock(&fs_info->reclaim_bgs_lock);
4059 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
4061 mutex_unlock(&fs_info->reclaim_bgs_lock);
4066 * this shouldn't happen, it means the last relocate
4070 BUG(); /* FIXME break ? */
4072 ret = btrfs_previous_item(chunk_root, path, 0,
4073 BTRFS_CHUNK_ITEM_KEY);
4075 mutex_unlock(&fs_info->reclaim_bgs_lock);
4080 leaf = path->nodes[0];
4081 slot = path->slots[0];
4082 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4084 if (found_key.objectid != key.objectid) {
4085 mutex_unlock(&fs_info->reclaim_bgs_lock);
4089 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
4090 chunk_type = btrfs_chunk_type(leaf, chunk);
4093 spin_lock(&fs_info->balance_lock);
4094 bctl->stat.considered++;
4095 spin_unlock(&fs_info->balance_lock);
4098 ret = should_balance_chunk(leaf, chunk, found_key.offset);
4100 btrfs_release_path(path);
4102 mutex_unlock(&fs_info->reclaim_bgs_lock);
4107 mutex_unlock(&fs_info->reclaim_bgs_lock);
4108 spin_lock(&fs_info->balance_lock);
4109 bctl->stat.expected++;
4110 spin_unlock(&fs_info->balance_lock);
4112 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
4114 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
4116 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
4123 * Apply limit_min filter, no need to check if the LIMITS
4124 * filter is used, limit_min is 0 by default
4126 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
4127 count_data < bctl->data.limit_min)
4128 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
4129 count_meta < bctl->meta.limit_min)
4130 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
4131 count_sys < bctl->sys.limit_min)) {
4132 mutex_unlock(&fs_info->reclaim_bgs_lock);
4136 if (!chunk_reserved) {
4138 * We may be relocating the only data chunk we have,
4139 * which could potentially end up with losing data's
4140 * raid profile, so lets allocate an empty one in
4143 ret = btrfs_may_alloc_data_chunk(fs_info,
4146 mutex_unlock(&fs_info->reclaim_bgs_lock);
4148 } else if (ret == 1) {
4153 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
4154 mutex_unlock(&fs_info->reclaim_bgs_lock);
4155 if (ret == -ENOSPC) {
4157 } else if (ret == -ETXTBSY) {
4159 "skipping relocation of block group %llu due to active swapfile",
4165 spin_lock(&fs_info->balance_lock);
4166 bctl->stat.completed++;
4167 spin_unlock(&fs_info->balance_lock);
4170 if (found_key.offset == 0)
4172 key.offset = found_key.offset - 1;
4176 btrfs_release_path(path);
4181 btrfs_free_path(path);
4182 if (enospc_errors) {
4183 btrfs_info(fs_info, "%d enospc errors during balance",
4193 * See if a given profile is valid and reduced.
4195 * @flags: profile to validate
4196 * @extended: if true @flags is treated as an extended profile
4198 static int alloc_profile_is_valid(u64 flags, int extended)
4200 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
4201 BTRFS_BLOCK_GROUP_PROFILE_MASK);
4203 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
4205 /* 1) check that all other bits are zeroed */
4209 /* 2) see if profile is reduced */
4211 return !extended; /* "0" is valid for usual profiles */
4213 return has_single_bit_set(flags);
4217 * Validate target profile against allowed profiles and return true if it's OK.
4218 * Otherwise print the error message and return false.
4220 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
4221 const struct btrfs_balance_args *bargs,
4222 u64 allowed, const char *type)
4224 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
4227 /* Profile is valid and does not have bits outside of the allowed set */
4228 if (alloc_profile_is_valid(bargs->target, 1) &&
4229 (bargs->target & ~allowed) == 0)
4232 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
4233 type, btrfs_bg_type_to_raid_name(bargs->target));
4238 * Fill @buf with textual description of balance filter flags @bargs, up to
4239 * @size_buf including the terminating null. The output may be trimmed if it
4240 * does not fit into the provided buffer.
4242 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
4246 u32 size_bp = size_buf;
4248 u64 flags = bargs->flags;
4249 char tmp_buf[128] = {'\0'};
4254 #define CHECK_APPEND_NOARG(a) \
4256 ret = snprintf(bp, size_bp, (a)); \
4257 if (ret < 0 || ret >= size_bp) \
4258 goto out_overflow; \
4263 #define CHECK_APPEND_1ARG(a, v1) \
4265 ret = snprintf(bp, size_bp, (a), (v1)); \
4266 if (ret < 0 || ret >= size_bp) \
4267 goto out_overflow; \
4272 #define CHECK_APPEND_2ARG(a, v1, v2) \
4274 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4275 if (ret < 0 || ret >= size_bp) \
4276 goto out_overflow; \
4281 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
4282 CHECK_APPEND_1ARG("convert=%s,",
4283 btrfs_bg_type_to_raid_name(bargs->target));
4285 if (flags & BTRFS_BALANCE_ARGS_SOFT)
4286 CHECK_APPEND_NOARG("soft,");
4288 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4289 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4291 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4294 if (flags & BTRFS_BALANCE_ARGS_USAGE)
4295 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4297 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4298 CHECK_APPEND_2ARG("usage=%u..%u,",
4299 bargs->usage_min, bargs->usage_max);
4301 if (flags & BTRFS_BALANCE_ARGS_DEVID)
4302 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4304 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4305 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4306 bargs->pstart, bargs->pend);
4308 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4309 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4310 bargs->vstart, bargs->vend);
4312 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4313 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4315 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4316 CHECK_APPEND_2ARG("limit=%u..%u,",
4317 bargs->limit_min, bargs->limit_max);
4319 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4320 CHECK_APPEND_2ARG("stripes=%u..%u,",
4321 bargs->stripes_min, bargs->stripes_max);
4323 #undef CHECK_APPEND_2ARG
4324 #undef CHECK_APPEND_1ARG
4325 #undef CHECK_APPEND_NOARG
4329 if (size_bp < size_buf)
4330 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4335 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4337 u32 size_buf = 1024;
4338 char tmp_buf[192] = {'\0'};
4341 u32 size_bp = size_buf;
4343 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4345 buf = kzalloc(size_buf, GFP_KERNEL);
4351 #define CHECK_APPEND_1ARG(a, v1) \
4353 ret = snprintf(bp, size_bp, (a), (v1)); \
4354 if (ret < 0 || ret >= size_bp) \
4355 goto out_overflow; \
4360 if (bctl->flags & BTRFS_BALANCE_FORCE)
4361 CHECK_APPEND_1ARG("%s", "-f ");
4363 if (bctl->flags & BTRFS_BALANCE_DATA) {
4364 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4365 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4368 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4369 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4370 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4373 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4374 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4375 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4378 #undef CHECK_APPEND_1ARG
4382 if (size_bp < size_buf)
4383 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4384 btrfs_info(fs_info, "balance: %s %s",
4385 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4386 "resume" : "start", buf);
4392 * Should be called with balance mutexe held
4394 int btrfs_balance(struct btrfs_fs_info *fs_info,
4395 struct btrfs_balance_control *bctl,
4396 struct btrfs_ioctl_balance_args *bargs)
4398 u64 meta_target, data_target;
4404 bool reducing_redundancy;
4405 bool paused = false;
4408 if (btrfs_fs_closing(fs_info) ||
4409 atomic_read(&fs_info->balance_pause_req) ||
4410 btrfs_should_cancel_balance(fs_info)) {
4415 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4416 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4420 * In case of mixed groups both data and meta should be picked,
4421 * and identical options should be given for both of them.
4423 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4424 if (mixed && (bctl->flags & allowed)) {
4425 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4426 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4427 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4429 "balance: mixed groups data and metadata options must be the same");
4436 * rw_devices will not change at the moment, device add/delete/replace
4439 num_devices = fs_info->fs_devices->rw_devices;
4442 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4443 * special bit for it, to make it easier to distinguish. Thus we need
4444 * to set it manually, or balance would refuse the profile.
4446 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4447 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4448 if (num_devices >= btrfs_raid_array[i].devs_min)
4449 allowed |= btrfs_raid_array[i].bg_flag;
4451 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4452 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4453 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4459 * Allow to reduce metadata or system integrity only if force set for
4460 * profiles with redundancy (copies, parity)
4463 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4464 if (btrfs_raid_array[i].ncopies >= 2 ||
4465 btrfs_raid_array[i].tolerated_failures >= 1)
4466 allowed |= btrfs_raid_array[i].bg_flag;
4469 seq = read_seqbegin(&fs_info->profiles_lock);
4471 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4472 (fs_info->avail_system_alloc_bits & allowed) &&
4473 !(bctl->sys.target & allowed)) ||
4474 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4475 (fs_info->avail_metadata_alloc_bits & allowed) &&
4476 !(bctl->meta.target & allowed)))
4477 reducing_redundancy = true;
4479 reducing_redundancy = false;
4481 /* if we're not converting, the target field is uninitialized */
4482 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4483 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4484 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4485 bctl->data.target : fs_info->avail_data_alloc_bits;
4486 } while (read_seqretry(&fs_info->profiles_lock, seq));
4488 if (reducing_redundancy) {
4489 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4491 "balance: force reducing metadata redundancy");
4494 "balance: reduces metadata redundancy, use --force if you want this");
4500 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4501 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4503 "balance: metadata profile %s has lower redundancy than data profile %s",
4504 btrfs_bg_type_to_raid_name(meta_target),
4505 btrfs_bg_type_to_raid_name(data_target));
4508 ret = insert_balance_item(fs_info, bctl);
4509 if (ret && ret != -EEXIST)
4512 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4513 BUG_ON(ret == -EEXIST);
4514 BUG_ON(fs_info->balance_ctl);
4515 spin_lock(&fs_info->balance_lock);
4516 fs_info->balance_ctl = bctl;
4517 spin_unlock(&fs_info->balance_lock);
4519 BUG_ON(ret != -EEXIST);
4520 spin_lock(&fs_info->balance_lock);
4521 update_balance_args(bctl);
4522 spin_unlock(&fs_info->balance_lock);
4525 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4526 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4527 describe_balance_start_or_resume(fs_info);
4528 mutex_unlock(&fs_info->balance_mutex);
4530 ret = __btrfs_balance(fs_info);
4532 mutex_lock(&fs_info->balance_mutex);
4533 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) {
4534 btrfs_info(fs_info, "balance: paused");
4535 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
4539 * Balance can be canceled by:
4541 * - Regular cancel request
4542 * Then ret == -ECANCELED and balance_cancel_req > 0
4544 * - Fatal signal to "btrfs" process
4545 * Either the signal caught by wait_reserve_ticket() and callers
4546 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4548 * Either way, in this case balance_cancel_req = 0, and
4549 * ret == -EINTR or ret == -ECANCELED.
4551 * So here we only check the return value to catch canceled balance.
4553 else if (ret == -ECANCELED || ret == -EINTR)
4554 btrfs_info(fs_info, "balance: canceled");
4556 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4558 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4561 memset(bargs, 0, sizeof(*bargs));
4562 btrfs_update_ioctl_balance_args(fs_info, bargs);
4565 /* We didn't pause, we can clean everything up. */
4567 reset_balance_state(fs_info);
4568 btrfs_exclop_finish(fs_info);
4571 wake_up(&fs_info->balance_wait_q);
4575 if (bctl->flags & BTRFS_BALANCE_RESUME)
4576 reset_balance_state(fs_info);
4579 btrfs_exclop_finish(fs_info);
4584 static int balance_kthread(void *data)
4586 struct btrfs_fs_info *fs_info = data;
4589 sb_start_write(fs_info->sb);
4590 mutex_lock(&fs_info->balance_mutex);
4591 if (fs_info->balance_ctl)
4592 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4593 mutex_unlock(&fs_info->balance_mutex);
4594 sb_end_write(fs_info->sb);
4599 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4601 struct task_struct *tsk;
4603 mutex_lock(&fs_info->balance_mutex);
4604 if (!fs_info->balance_ctl) {
4605 mutex_unlock(&fs_info->balance_mutex);
4608 mutex_unlock(&fs_info->balance_mutex);
4610 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4611 btrfs_info(fs_info, "balance: resume skipped");
4615 spin_lock(&fs_info->super_lock);
4616 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
4617 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;
4618 spin_unlock(&fs_info->super_lock);
4620 * A ro->rw remount sequence should continue with the paused balance
4621 * regardless of who pauses it, system or the user as of now, so set
4624 spin_lock(&fs_info->balance_lock);
4625 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4626 spin_unlock(&fs_info->balance_lock);
4628 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4629 return PTR_ERR_OR_ZERO(tsk);
4632 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4634 struct btrfs_balance_control *bctl;
4635 struct btrfs_balance_item *item;
4636 struct btrfs_disk_balance_args disk_bargs;
4637 struct btrfs_path *path;
4638 struct extent_buffer *leaf;
4639 struct btrfs_key key;
4642 path = btrfs_alloc_path();
4646 key.objectid = BTRFS_BALANCE_OBJECTID;
4647 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4650 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4653 if (ret > 0) { /* ret = -ENOENT; */
4658 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4664 leaf = path->nodes[0];
4665 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4667 bctl->flags = btrfs_balance_flags(leaf, item);
4668 bctl->flags |= BTRFS_BALANCE_RESUME;
4670 btrfs_balance_data(leaf, item, &disk_bargs);
4671 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4672 btrfs_balance_meta(leaf, item, &disk_bargs);
4673 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4674 btrfs_balance_sys(leaf, item, &disk_bargs);
4675 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4678 * This should never happen, as the paused balance state is recovered
4679 * during mount without any chance of other exclusive ops to collide.
4681 * This gives the exclusive op status to balance and keeps in paused
4682 * state until user intervention (cancel or umount). If the ownership
4683 * cannot be assigned, show a message but do not fail. The balance
4684 * is in a paused state and must have fs_info::balance_ctl properly
4687 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED))
4689 "balance: cannot set exclusive op status, resume manually");
4691 btrfs_release_path(path);
4693 mutex_lock(&fs_info->balance_mutex);
4694 BUG_ON(fs_info->balance_ctl);
4695 spin_lock(&fs_info->balance_lock);
4696 fs_info->balance_ctl = bctl;
4697 spin_unlock(&fs_info->balance_lock);
4698 mutex_unlock(&fs_info->balance_mutex);
4700 btrfs_free_path(path);
4704 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4708 mutex_lock(&fs_info->balance_mutex);
4709 if (!fs_info->balance_ctl) {
4710 mutex_unlock(&fs_info->balance_mutex);
4714 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4715 atomic_inc(&fs_info->balance_pause_req);
4716 mutex_unlock(&fs_info->balance_mutex);
4718 wait_event(fs_info->balance_wait_q,
4719 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4721 mutex_lock(&fs_info->balance_mutex);
4722 /* we are good with balance_ctl ripped off from under us */
4723 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4724 atomic_dec(&fs_info->balance_pause_req);
4729 mutex_unlock(&fs_info->balance_mutex);
4733 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4735 mutex_lock(&fs_info->balance_mutex);
4736 if (!fs_info->balance_ctl) {
4737 mutex_unlock(&fs_info->balance_mutex);
4742 * A paused balance with the item stored on disk can be resumed at
4743 * mount time if the mount is read-write. Otherwise it's still paused
4744 * and we must not allow cancelling as it deletes the item.
4746 if (sb_rdonly(fs_info->sb)) {
4747 mutex_unlock(&fs_info->balance_mutex);
4751 atomic_inc(&fs_info->balance_cancel_req);
4753 * if we are running just wait and return, balance item is
4754 * deleted in btrfs_balance in this case
4756 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4757 mutex_unlock(&fs_info->balance_mutex);
4758 wait_event(fs_info->balance_wait_q,
4759 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4760 mutex_lock(&fs_info->balance_mutex);
4762 mutex_unlock(&fs_info->balance_mutex);
4764 * Lock released to allow other waiters to continue, we'll
4765 * reexamine the status again.
4767 mutex_lock(&fs_info->balance_mutex);
4769 if (fs_info->balance_ctl) {
4770 reset_balance_state(fs_info);
4771 btrfs_exclop_finish(fs_info);
4772 btrfs_info(fs_info, "balance: canceled");
4776 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4777 atomic_dec(&fs_info->balance_cancel_req);
4778 mutex_unlock(&fs_info->balance_mutex);
4782 int btrfs_uuid_scan_kthread(void *data)
4784 struct btrfs_fs_info *fs_info = data;
4785 struct btrfs_root *root = fs_info->tree_root;
4786 struct btrfs_key key;
4787 struct btrfs_path *path = NULL;
4789 struct extent_buffer *eb;
4791 struct btrfs_root_item root_item;
4793 struct btrfs_trans_handle *trans = NULL;
4794 bool closing = false;
4796 path = btrfs_alloc_path();
4803 key.type = BTRFS_ROOT_ITEM_KEY;
4807 if (btrfs_fs_closing(fs_info)) {
4811 ret = btrfs_search_forward(root, &key, path,
4812 BTRFS_OLDEST_GENERATION);
4819 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4820 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4821 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4822 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4825 eb = path->nodes[0];
4826 slot = path->slots[0];
4827 item_size = btrfs_item_size(eb, slot);
4828 if (item_size < sizeof(root_item))
4831 read_extent_buffer(eb, &root_item,
4832 btrfs_item_ptr_offset(eb, slot),
4833 (int)sizeof(root_item));
4834 if (btrfs_root_refs(&root_item) == 0)
4837 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4838 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4842 btrfs_release_path(path);
4844 * 1 - subvol uuid item
4845 * 1 - received_subvol uuid item
4847 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4848 if (IS_ERR(trans)) {
4849 ret = PTR_ERR(trans);
4857 btrfs_release_path(path);
4858 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4859 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4860 BTRFS_UUID_KEY_SUBVOL,
4863 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4869 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4870 ret = btrfs_uuid_tree_add(trans,
4871 root_item.received_uuid,
4872 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4875 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4882 btrfs_release_path(path);
4884 ret = btrfs_end_transaction(trans);
4890 if (key.offset < (u64)-1) {
4892 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4894 key.type = BTRFS_ROOT_ITEM_KEY;
4895 } else if (key.objectid < (u64)-1) {
4897 key.type = BTRFS_ROOT_ITEM_KEY;
4906 btrfs_free_path(path);
4907 if (trans && !IS_ERR(trans))
4908 btrfs_end_transaction(trans);
4910 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4912 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4913 up(&fs_info->uuid_tree_rescan_sem);
4917 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4919 struct btrfs_trans_handle *trans;
4920 struct btrfs_root *tree_root = fs_info->tree_root;
4921 struct btrfs_root *uuid_root;
4922 struct task_struct *task;
4929 trans = btrfs_start_transaction(tree_root, 2);
4931 return PTR_ERR(trans);
4933 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4934 if (IS_ERR(uuid_root)) {
4935 ret = PTR_ERR(uuid_root);
4936 btrfs_abort_transaction(trans, ret);
4937 btrfs_end_transaction(trans);
4941 fs_info->uuid_root = uuid_root;
4943 ret = btrfs_commit_transaction(trans);
4947 down(&fs_info->uuid_tree_rescan_sem);
4948 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4950 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4951 btrfs_warn(fs_info, "failed to start uuid_scan task");
4952 up(&fs_info->uuid_tree_rescan_sem);
4953 return PTR_ERR(task);
4960 * shrinking a device means finding all of the device extents past
4961 * the new size, and then following the back refs to the chunks.
4962 * The chunk relocation code actually frees the device extent
4964 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4966 struct btrfs_fs_info *fs_info = device->fs_info;
4967 struct btrfs_root *root = fs_info->dev_root;
4968 struct btrfs_trans_handle *trans;
4969 struct btrfs_dev_extent *dev_extent = NULL;
4970 struct btrfs_path *path;
4976 bool retried = false;
4977 struct extent_buffer *l;
4978 struct btrfs_key key;
4979 struct btrfs_super_block *super_copy = fs_info->super_copy;
4980 u64 old_total = btrfs_super_total_bytes(super_copy);
4981 u64 old_size = btrfs_device_get_total_bytes(device);
4986 new_size = round_down(new_size, fs_info->sectorsize);
4988 diff = round_down(old_size - new_size, fs_info->sectorsize);
4990 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4993 path = btrfs_alloc_path();
4997 path->reada = READA_BACK;
4999 trans = btrfs_start_transaction(root, 0);
5000 if (IS_ERR(trans)) {
5001 btrfs_free_path(path);
5002 return PTR_ERR(trans);
5005 mutex_lock(&fs_info->chunk_mutex);
5007 btrfs_device_set_total_bytes(device, new_size);
5008 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5009 device->fs_devices->total_rw_bytes -= diff;
5012 * The new free_chunk_space is new_size - used, so we have to
5013 * subtract the delta of the old free_chunk_space which included
5014 * old_size - used. If used > new_size then just subtract this
5015 * entire device's free space.
5017 if (device->bytes_used < new_size)
5018 free_diff = (old_size - device->bytes_used) -
5019 (new_size - device->bytes_used);
5021 free_diff = old_size - device->bytes_used;
5022 atomic64_sub(free_diff, &fs_info->free_chunk_space);
5026 * Once the device's size has been set to the new size, ensure all
5027 * in-memory chunks are synced to disk so that the loop below sees them
5028 * and relocates them accordingly.
5030 if (contains_pending_extent(device, &start, diff)) {
5031 mutex_unlock(&fs_info->chunk_mutex);
5032 ret = btrfs_commit_transaction(trans);
5036 mutex_unlock(&fs_info->chunk_mutex);
5037 btrfs_end_transaction(trans);
5041 key.objectid = device->devid;
5042 key.offset = (u64)-1;
5043 key.type = BTRFS_DEV_EXTENT_KEY;
5046 mutex_lock(&fs_info->reclaim_bgs_lock);
5047 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5049 mutex_unlock(&fs_info->reclaim_bgs_lock);
5053 ret = btrfs_previous_item(root, path, 0, key.type);
5055 mutex_unlock(&fs_info->reclaim_bgs_lock);
5059 btrfs_release_path(path);
5064 slot = path->slots[0];
5065 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
5067 if (key.objectid != device->devid) {
5068 mutex_unlock(&fs_info->reclaim_bgs_lock);
5069 btrfs_release_path(path);
5073 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
5074 length = btrfs_dev_extent_length(l, dev_extent);
5076 if (key.offset + length <= new_size) {
5077 mutex_unlock(&fs_info->reclaim_bgs_lock);
5078 btrfs_release_path(path);
5082 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
5083 btrfs_release_path(path);
5086 * We may be relocating the only data chunk we have,
5087 * which could potentially end up with losing data's
5088 * raid profile, so lets allocate an empty one in
5091 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
5093 mutex_unlock(&fs_info->reclaim_bgs_lock);
5097 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
5098 mutex_unlock(&fs_info->reclaim_bgs_lock);
5099 if (ret == -ENOSPC) {
5102 if (ret == -ETXTBSY) {
5104 "could not shrink block group %llu due to active swapfile",
5109 } while (key.offset-- > 0);
5111 if (failed && !retried) {
5115 } else if (failed && retried) {
5120 /* Shrinking succeeded, else we would be at "done". */
5121 trans = btrfs_start_transaction(root, 0);
5122 if (IS_ERR(trans)) {
5123 ret = PTR_ERR(trans);
5127 mutex_lock(&fs_info->chunk_mutex);
5128 /* Clear all state bits beyond the shrunk device size */
5129 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
5132 btrfs_device_set_disk_total_bytes(device, new_size);
5133 if (list_empty(&device->post_commit_list))
5134 list_add_tail(&device->post_commit_list,
5135 &trans->transaction->dev_update_list);
5137 WARN_ON(diff > old_total);
5138 btrfs_set_super_total_bytes(super_copy,
5139 round_down(old_total - diff, fs_info->sectorsize));
5140 mutex_unlock(&fs_info->chunk_mutex);
5142 btrfs_reserve_chunk_metadata(trans, false);
5143 /* Now btrfs_update_device() will change the on-disk size. */
5144 ret = btrfs_update_device(trans, device);
5145 btrfs_trans_release_chunk_metadata(trans);
5147 btrfs_abort_transaction(trans, ret);
5148 btrfs_end_transaction(trans);
5150 ret = btrfs_commit_transaction(trans);
5153 btrfs_free_path(path);
5155 mutex_lock(&fs_info->chunk_mutex);
5156 btrfs_device_set_total_bytes(device, old_size);
5157 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5158 device->fs_devices->total_rw_bytes += diff;
5159 atomic64_add(free_diff, &fs_info->free_chunk_space);
5161 mutex_unlock(&fs_info->chunk_mutex);
5166 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
5167 struct btrfs_key *key,
5168 struct btrfs_chunk *chunk, int item_size)
5170 struct btrfs_super_block *super_copy = fs_info->super_copy;
5171 struct btrfs_disk_key disk_key;
5175 lockdep_assert_held(&fs_info->chunk_mutex);
5177 array_size = btrfs_super_sys_array_size(super_copy);
5178 if (array_size + item_size + sizeof(disk_key)
5179 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
5182 ptr = super_copy->sys_chunk_array + array_size;
5183 btrfs_cpu_key_to_disk(&disk_key, key);
5184 memcpy(ptr, &disk_key, sizeof(disk_key));
5185 ptr += sizeof(disk_key);
5186 memcpy(ptr, chunk, item_size);
5187 item_size += sizeof(disk_key);
5188 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
5194 * sort the devices in descending order by max_avail, total_avail
5196 static int btrfs_cmp_device_info(const void *a, const void *b)
5198 const struct btrfs_device_info *di_a = a;
5199 const struct btrfs_device_info *di_b = b;
5201 if (di_a->max_avail > di_b->max_avail)
5203 if (di_a->max_avail < di_b->max_avail)
5205 if (di_a->total_avail > di_b->total_avail)
5207 if (di_a->total_avail < di_b->total_avail)
5212 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
5214 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5217 btrfs_set_fs_incompat(info, RAID56);
5220 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
5222 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
5225 btrfs_set_fs_incompat(info, RAID1C34);
5229 * Structure used internally for btrfs_create_chunk() function.
5230 * Wraps needed parameters.
5232 struct alloc_chunk_ctl {
5235 /* Total number of stripes to allocate */
5237 /* sub_stripes info for map */
5239 /* Stripes per device */
5241 /* Maximum number of devices to use */
5243 /* Minimum number of devices to use */
5245 /* ndevs has to be a multiple of this */
5247 /* Number of copies */
5249 /* Number of stripes worth of bytes to store parity information */
5251 u64 max_stripe_size;
5259 static void init_alloc_chunk_ctl_policy_regular(
5260 struct btrfs_fs_devices *fs_devices,
5261 struct alloc_chunk_ctl *ctl)
5263 struct btrfs_space_info *space_info;
5265 space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type);
5268 ctl->max_chunk_size = READ_ONCE(space_info->chunk_size);
5269 ctl->max_stripe_size = min_t(u64, ctl->max_chunk_size, SZ_1G);
5271 if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM)
5272 ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK);
5274 /* We don't want a chunk larger than 10% of writable space */
5275 ctl->max_chunk_size = min(mult_perc(fs_devices->total_rw_bytes, 10),
5276 ctl->max_chunk_size);
5277 ctl->dev_extent_min = btrfs_stripe_nr_to_offset(ctl->dev_stripes);
5280 static void init_alloc_chunk_ctl_policy_zoned(
5281 struct btrfs_fs_devices *fs_devices,
5282 struct alloc_chunk_ctl *ctl)
5284 u64 zone_size = fs_devices->fs_info->zone_size;
5286 int min_num_stripes = ctl->devs_min * ctl->dev_stripes;
5287 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;
5288 u64 min_chunk_size = min_data_stripes * zone_size;
5289 u64 type = ctl->type;
5291 ctl->max_stripe_size = zone_size;
5292 if (type & BTRFS_BLOCK_GROUP_DATA) {
5293 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,
5295 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5296 ctl->max_chunk_size = ctl->max_stripe_size;
5297 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5298 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5299 ctl->devs_max = min_t(int, ctl->devs_max,
5300 BTRFS_MAX_DEVS_SYS_CHUNK);
5305 /* We don't want a chunk larger than 10% of writable space */
5306 limit = max(round_down(mult_perc(fs_devices->total_rw_bytes, 10),
5309 ctl->max_chunk_size = min(limit, ctl->max_chunk_size);
5310 ctl->dev_extent_min = zone_size * ctl->dev_stripes;
5313 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
5314 struct alloc_chunk_ctl *ctl)
5316 int index = btrfs_bg_flags_to_raid_index(ctl->type);
5318 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
5319 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
5320 ctl->devs_max = btrfs_raid_array[index].devs_max;
5322 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
5323 ctl->devs_min = btrfs_raid_array[index].devs_min;
5324 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
5325 ctl->ncopies = btrfs_raid_array[index].ncopies;
5326 ctl->nparity = btrfs_raid_array[index].nparity;
5329 switch (fs_devices->chunk_alloc_policy) {
5330 case BTRFS_CHUNK_ALLOC_REGULAR:
5331 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
5333 case BTRFS_CHUNK_ALLOC_ZONED:
5334 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);
5341 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
5342 struct alloc_chunk_ctl *ctl,
5343 struct btrfs_device_info *devices_info)
5345 struct btrfs_fs_info *info = fs_devices->fs_info;
5346 struct btrfs_device *device;
5348 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
5355 * in the first pass through the devices list, we gather information
5356 * about the available holes on each device.
5358 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5359 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5361 "BTRFS: read-only device in alloc_list\n");
5365 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5366 &device->dev_state) ||
5367 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5370 if (device->total_bytes > device->bytes_used)
5371 total_avail = device->total_bytes - device->bytes_used;
5375 /* If there is no space on this device, skip it. */
5376 if (total_avail < ctl->dev_extent_min)
5379 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
5381 if (ret && ret != -ENOSPC)
5385 max_avail = dev_extent_want;
5387 if (max_avail < ctl->dev_extent_min) {
5388 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5390 "%s: devid %llu has no free space, have=%llu want=%llu",
5391 __func__, device->devid, max_avail,
5392 ctl->dev_extent_min);
5396 if (ndevs == fs_devices->rw_devices) {
5397 WARN(1, "%s: found more than %llu devices\n",
5398 __func__, fs_devices->rw_devices);
5401 devices_info[ndevs].dev_offset = dev_offset;
5402 devices_info[ndevs].max_avail = max_avail;
5403 devices_info[ndevs].total_avail = total_avail;
5404 devices_info[ndevs].dev = device;
5410 * now sort the devices by hole size / available space
5412 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5413 btrfs_cmp_device_info, NULL);
5418 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5419 struct btrfs_device_info *devices_info)
5421 /* Number of stripes that count for block group size */
5425 * The primary goal is to maximize the number of stripes, so use as
5426 * many devices as possible, even if the stripes are not maximum sized.
5428 * The DUP profile stores more than one stripe per device, the
5429 * max_avail is the total size so we have to adjust.
5431 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5433 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5435 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5436 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5439 * Use the number of data stripes to figure out how big this chunk is
5440 * really going to be in terms of logical address space, and compare
5441 * that answer with the max chunk size. If it's higher, we try to
5442 * reduce stripe_size.
5444 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5446 * Reduce stripe_size, round it up to a 16MB boundary again and
5447 * then use it, unless it ends up being even bigger than the
5448 * previous value we had already.
5450 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5451 data_stripes), SZ_16M),
5455 /* Stripe size should not go beyond 1G. */
5456 ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G);
5458 /* Align to BTRFS_STRIPE_LEN */
5459 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5460 ctl->chunk_size = ctl->stripe_size * data_stripes;
5465 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,
5466 struct btrfs_device_info *devices_info)
5468 u64 zone_size = devices_info[0].dev->zone_info->zone_size;
5469 /* Number of stripes that count for block group size */
5473 * It should hold because:
5474 * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5476 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);
5478 ctl->stripe_size = zone_size;
5479 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5480 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5482 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5483 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5484 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,
5485 ctl->stripe_size) + ctl->nparity,
5487 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5488 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5489 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);
5492 ctl->chunk_size = ctl->stripe_size * data_stripes;
5497 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5498 struct alloc_chunk_ctl *ctl,
5499 struct btrfs_device_info *devices_info)
5501 struct btrfs_fs_info *info = fs_devices->fs_info;
5504 * Round down to number of usable stripes, devs_increment can be any
5505 * number so we can't use round_down() that requires power of 2, while
5506 * rounddown is safe.
5508 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5510 if (ctl->ndevs < ctl->devs_min) {
5511 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5513 "%s: not enough devices with free space: have=%d minimum required=%d",
5514 __func__, ctl->ndevs, ctl->devs_min);
5519 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5521 switch (fs_devices->chunk_alloc_policy) {
5522 case BTRFS_CHUNK_ALLOC_REGULAR:
5523 return decide_stripe_size_regular(ctl, devices_info);
5524 case BTRFS_CHUNK_ALLOC_ZONED:
5525 return decide_stripe_size_zoned(ctl, devices_info);
5531 static void chunk_map_device_set_bits(struct btrfs_chunk_map *map, unsigned int bits)
5533 for (int i = 0; i < map->num_stripes; i++) {
5534 struct btrfs_io_stripe *stripe = &map->stripes[i];
5535 struct btrfs_device *device = stripe->dev;
5537 set_extent_bit(&device->alloc_state, stripe->physical,
5538 stripe->physical + map->stripe_size - 1,
5539 bits | EXTENT_NOWAIT, NULL);
5543 static void chunk_map_device_clear_bits(struct btrfs_chunk_map *map, unsigned int bits)
5545 for (int i = 0; i < map->num_stripes; i++) {
5546 struct btrfs_io_stripe *stripe = &map->stripes[i];
5547 struct btrfs_device *device = stripe->dev;
5549 __clear_extent_bit(&device->alloc_state, stripe->physical,
5550 stripe->physical + map->stripe_size - 1,
5551 bits | EXTENT_NOWAIT,
5556 void btrfs_remove_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map)
5558 write_lock(&fs_info->mapping_tree_lock);
5559 rb_erase_cached(&map->rb_node, &fs_info->mapping_tree);
5560 RB_CLEAR_NODE(&map->rb_node);
5561 chunk_map_device_clear_bits(map, CHUNK_ALLOCATED);
5562 write_unlock(&fs_info->mapping_tree_lock);
5564 /* Once for the tree reference. */
5565 btrfs_free_chunk_map(map);
5569 int btrfs_add_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map)
5572 struct rb_node *parent = NULL;
5573 bool leftmost = true;
5575 write_lock(&fs_info->mapping_tree_lock);
5576 p = &fs_info->mapping_tree.rb_root.rb_node;
5578 struct btrfs_chunk_map *entry;
5581 entry = rb_entry(parent, struct btrfs_chunk_map, rb_node);
5583 if (map->start < entry->start) {
5585 } else if (map->start > entry->start) {
5586 p = &(*p)->rb_right;
5589 write_unlock(&fs_info->mapping_tree_lock);
5593 rb_link_node(&map->rb_node, parent, p);
5594 rb_insert_color_cached(&map->rb_node, &fs_info->mapping_tree, leftmost);
5595 chunk_map_device_set_bits(map, CHUNK_ALLOCATED);
5596 chunk_map_device_clear_bits(map, CHUNK_TRIMMED);
5597 write_unlock(&fs_info->mapping_tree_lock);
5603 struct btrfs_chunk_map *btrfs_alloc_chunk_map(int num_stripes, gfp_t gfp)
5605 struct btrfs_chunk_map *map;
5607 map = kmalloc(btrfs_chunk_map_size(num_stripes), gfp);
5611 refcount_set(&map->refs, 1);
5612 RB_CLEAR_NODE(&map->rb_node);
5617 struct btrfs_chunk_map *btrfs_clone_chunk_map(struct btrfs_chunk_map *map, gfp_t gfp)
5619 const int size = btrfs_chunk_map_size(map->num_stripes);
5620 struct btrfs_chunk_map *clone;
5622 clone = kmemdup(map, size, gfp);
5626 refcount_set(&clone->refs, 1);
5627 RB_CLEAR_NODE(&clone->rb_node);
5632 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
5633 struct alloc_chunk_ctl *ctl,
5634 struct btrfs_device_info *devices_info)
5636 struct btrfs_fs_info *info = trans->fs_info;
5637 struct btrfs_chunk_map *map;
5638 struct btrfs_block_group *block_group;
5639 u64 start = ctl->start;
5640 u64 type = ctl->type;
5645 map = btrfs_alloc_chunk_map(ctl->num_stripes, GFP_NOFS);
5647 return ERR_PTR(-ENOMEM);
5650 map->chunk_len = ctl->chunk_size;
5651 map->stripe_size = ctl->stripe_size;
5653 map->io_align = BTRFS_STRIPE_LEN;
5654 map->io_width = BTRFS_STRIPE_LEN;
5655 map->sub_stripes = ctl->sub_stripes;
5656 map->num_stripes = ctl->num_stripes;
5658 for (i = 0; i < ctl->ndevs; ++i) {
5659 for (j = 0; j < ctl->dev_stripes; ++j) {
5660 int s = i * ctl->dev_stripes + j;
5661 map->stripes[s].dev = devices_info[i].dev;
5662 map->stripes[s].physical = devices_info[i].dev_offset +
5663 j * ctl->stripe_size;
5667 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5669 ret = btrfs_add_chunk_map(info, map);
5671 btrfs_free_chunk_map(map);
5672 return ERR_PTR(ret);
5675 block_group = btrfs_make_block_group(trans, type, start, ctl->chunk_size);
5676 if (IS_ERR(block_group)) {
5677 btrfs_remove_chunk_map(info, map);
5681 for (int i = 0; i < map->num_stripes; i++) {
5682 struct btrfs_device *dev = map->stripes[i].dev;
5684 btrfs_device_set_bytes_used(dev,
5685 dev->bytes_used + ctl->stripe_size);
5686 if (list_empty(&dev->post_commit_list))
5687 list_add_tail(&dev->post_commit_list,
5688 &trans->transaction->dev_update_list);
5691 atomic64_sub(ctl->stripe_size * map->num_stripes,
5692 &info->free_chunk_space);
5694 check_raid56_incompat_flag(info, type);
5695 check_raid1c34_incompat_flag(info, type);
5700 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
5703 struct btrfs_fs_info *info = trans->fs_info;
5704 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5705 struct btrfs_device_info *devices_info = NULL;
5706 struct alloc_chunk_ctl ctl;
5707 struct btrfs_block_group *block_group;
5710 lockdep_assert_held(&info->chunk_mutex);
5712 if (!alloc_profile_is_valid(type, 0)) {
5714 return ERR_PTR(-EINVAL);
5717 if (list_empty(&fs_devices->alloc_list)) {
5718 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5719 btrfs_debug(info, "%s: no writable device", __func__);
5720 return ERR_PTR(-ENOSPC);
5723 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5724 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5726 return ERR_PTR(-EINVAL);
5729 ctl.start = find_next_chunk(info);
5731 init_alloc_chunk_ctl(fs_devices, &ctl);
5733 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5736 return ERR_PTR(-ENOMEM);
5738 ret = gather_device_info(fs_devices, &ctl, devices_info);
5740 block_group = ERR_PTR(ret);
5744 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5746 block_group = ERR_PTR(ret);
5750 block_group = create_chunk(trans, &ctl, devices_info);
5753 kfree(devices_info);
5758 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5759 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5762 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5765 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
5766 struct btrfs_block_group *bg)
5768 struct btrfs_fs_info *fs_info = trans->fs_info;
5769 struct btrfs_root *chunk_root = fs_info->chunk_root;
5770 struct btrfs_key key;
5771 struct btrfs_chunk *chunk;
5772 struct btrfs_stripe *stripe;
5773 struct btrfs_chunk_map *map;
5779 * We take the chunk_mutex for 2 reasons:
5781 * 1) Updates and insertions in the chunk btree must be done while holding
5782 * the chunk_mutex, as well as updating the system chunk array in the
5783 * superblock. See the comment on top of btrfs_chunk_alloc() for the
5786 * 2) To prevent races with the final phase of a device replace operation
5787 * that replaces the device object associated with the map's stripes,
5788 * because the device object's id can change at any time during that
5789 * final phase of the device replace operation
5790 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5791 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5792 * which would cause a failure when updating the device item, which does
5793 * not exists, or persisting a stripe of the chunk item with such ID.
5794 * Here we can't use the device_list_mutex because our caller already
5795 * has locked the chunk_mutex, and the final phase of device replace
5796 * acquires both mutexes - first the device_list_mutex and then the
5797 * chunk_mutex. Using any of those two mutexes protects us from a
5798 * concurrent device replace.
5800 lockdep_assert_held(&fs_info->chunk_mutex);
5802 map = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
5805 btrfs_abort_transaction(trans, ret);
5809 item_size = btrfs_chunk_item_size(map->num_stripes);
5811 chunk = kzalloc(item_size, GFP_NOFS);
5814 btrfs_abort_transaction(trans, ret);
5818 for (i = 0; i < map->num_stripes; i++) {
5819 struct btrfs_device *device = map->stripes[i].dev;
5821 ret = btrfs_update_device(trans, device);
5826 stripe = &chunk->stripe;
5827 for (i = 0; i < map->num_stripes; i++) {
5828 struct btrfs_device *device = map->stripes[i].dev;
5829 const u64 dev_offset = map->stripes[i].physical;
5831 btrfs_set_stack_stripe_devid(stripe, device->devid);
5832 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5833 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5837 btrfs_set_stack_chunk_length(chunk, bg->length);
5838 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
5839 btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
5840 btrfs_set_stack_chunk_type(chunk, map->type);
5841 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5842 btrfs_set_stack_chunk_io_align(chunk, BTRFS_STRIPE_LEN);
5843 btrfs_set_stack_chunk_io_width(chunk, BTRFS_STRIPE_LEN);
5844 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5845 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5847 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5848 key.type = BTRFS_CHUNK_ITEM_KEY;
5849 key.offset = bg->start;
5851 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5855 set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags);
5857 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5858 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5865 btrfs_free_chunk_map(map);
5869 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5871 struct btrfs_fs_info *fs_info = trans->fs_info;
5873 struct btrfs_block_group *meta_bg;
5874 struct btrfs_block_group *sys_bg;
5877 * When adding a new device for sprouting, the seed device is read-only
5878 * so we must first allocate a metadata and a system chunk. But before
5879 * adding the block group items to the extent, device and chunk btrees,
5882 * 1) Create both chunks without doing any changes to the btrees, as
5883 * otherwise we would get -ENOSPC since the block groups from the
5884 * seed device are read-only;
5886 * 2) Add the device item for the new sprout device - finishing the setup
5887 * of a new block group requires updating the device item in the chunk
5888 * btree, so it must exist when we attempt to do it. The previous step
5889 * ensures this does not fail with -ENOSPC.
5891 * After that we can add the block group items to their btrees:
5892 * update existing device item in the chunk btree, add a new block group
5893 * item to the extent btree, add a new chunk item to the chunk btree and
5894 * finally add the new device extent items to the devices btree.
5897 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5898 meta_bg = btrfs_create_chunk(trans, alloc_profile);
5899 if (IS_ERR(meta_bg))
5900 return PTR_ERR(meta_bg);
5902 alloc_profile = btrfs_system_alloc_profile(fs_info);
5903 sys_bg = btrfs_create_chunk(trans, alloc_profile);
5905 return PTR_ERR(sys_bg);
5910 static inline int btrfs_chunk_max_errors(struct btrfs_chunk_map *map)
5912 const int index = btrfs_bg_flags_to_raid_index(map->type);
5914 return btrfs_raid_array[index].tolerated_failures;
5917 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5919 struct btrfs_chunk_map *map;
5924 map = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5928 for (i = 0; i < map->num_stripes; i++) {
5929 if (test_bit(BTRFS_DEV_STATE_MISSING,
5930 &map->stripes[i].dev->dev_state)) {
5934 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5935 &map->stripes[i].dev->dev_state)) {
5942 * If the number of missing devices is larger than max errors, we can
5943 * not write the data into that chunk successfully.
5945 if (miss_ndevs > btrfs_chunk_max_errors(map))
5948 btrfs_free_chunk_map(map);
5952 void btrfs_mapping_tree_free(struct btrfs_fs_info *fs_info)
5954 write_lock(&fs_info->mapping_tree_lock);
5955 while (!RB_EMPTY_ROOT(&fs_info->mapping_tree.rb_root)) {
5956 struct btrfs_chunk_map *map;
5957 struct rb_node *node;
5959 node = rb_first_cached(&fs_info->mapping_tree);
5960 map = rb_entry(node, struct btrfs_chunk_map, rb_node);
5961 rb_erase_cached(&map->rb_node, &fs_info->mapping_tree);
5962 RB_CLEAR_NODE(&map->rb_node);
5963 chunk_map_device_clear_bits(map, CHUNK_ALLOCATED);
5964 /* Once for the tree ref. */
5965 btrfs_free_chunk_map(map);
5966 cond_resched_rwlock_write(&fs_info->mapping_tree_lock);
5968 write_unlock(&fs_info->mapping_tree_lock);
5971 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5973 struct btrfs_chunk_map *map;
5974 enum btrfs_raid_types index;
5977 map = btrfs_get_chunk_map(fs_info, logical, len);
5980 * We could return errors for these cases, but that could get
5981 * ugly and we'd probably do the same thing which is just not do
5982 * anything else and exit, so return 1 so the callers don't try
5983 * to use other copies.
5987 index = btrfs_bg_flags_to_raid_index(map->type);
5989 /* Non-RAID56, use their ncopies from btrfs_raid_array. */
5990 if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5991 ret = btrfs_raid_array[index].ncopies;
5992 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5994 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5996 * There could be two corrupted data stripes, we need
5997 * to loop retry in order to rebuild the correct data.
5999 * Fail a stripe at a time on every retry except the
6000 * stripe under reconstruction.
6002 ret = map->num_stripes;
6003 btrfs_free_chunk_map(map);
6007 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
6010 struct btrfs_chunk_map *map;
6011 unsigned long len = fs_info->sectorsize;
6013 if (!btrfs_fs_incompat(fs_info, RAID56))
6016 map = btrfs_get_chunk_map(fs_info, logical, len);
6018 if (!WARN_ON(IS_ERR(map))) {
6019 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
6020 len = btrfs_stripe_nr_to_offset(nr_data_stripes(map));
6021 btrfs_free_chunk_map(map);
6026 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
6028 struct btrfs_chunk_map *map;
6031 if (!btrfs_fs_incompat(fs_info, RAID56))
6034 map = btrfs_get_chunk_map(fs_info, logical, len);
6036 if (!WARN_ON(IS_ERR(map))) {
6037 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
6039 btrfs_free_chunk_map(map);
6044 static int find_live_mirror(struct btrfs_fs_info *fs_info,
6045 struct btrfs_chunk_map *map, int first,
6046 int dev_replace_is_ongoing)
6048 const enum btrfs_read_policy policy = READ_ONCE(fs_info->fs_devices->read_policy);
6051 int preferred_mirror;
6053 struct btrfs_device *srcdev;
6056 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
6058 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
6059 num_stripes = map->sub_stripes;
6061 num_stripes = map->num_stripes;
6065 /* Shouldn't happen, just warn and use pid instead of failing */
6066 btrfs_warn_rl(fs_info, "unknown read_policy type %u, reset to pid",
6068 WRITE_ONCE(fs_info->fs_devices->read_policy, BTRFS_READ_POLICY_PID);
6070 case BTRFS_READ_POLICY_PID:
6071 preferred_mirror = first + (current->pid % num_stripes);
6075 if (dev_replace_is_ongoing &&
6076 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
6077 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
6078 srcdev = fs_info->dev_replace.srcdev;
6083 * try to avoid the drive that is the source drive for a
6084 * dev-replace procedure, only choose it if no other non-missing
6085 * mirror is available
6087 for (tolerance = 0; tolerance < 2; tolerance++) {
6088 if (map->stripes[preferred_mirror].dev->bdev &&
6089 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
6090 return preferred_mirror;
6091 for (i = first; i < first + num_stripes; i++) {
6092 if (map->stripes[i].dev->bdev &&
6093 (tolerance || map->stripes[i].dev != srcdev))
6098 /* we couldn't find one that doesn't fail. Just return something
6099 * and the io error handling code will clean up eventually
6101 return preferred_mirror;
6104 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info,
6108 struct btrfs_io_context *bioc;
6111 /* The size of btrfs_io_context */
6112 sizeof(struct btrfs_io_context) +
6113 /* Plus the variable array for the stripes */
6114 sizeof(struct btrfs_io_stripe) * (total_stripes),
6120 refcount_set(&bioc->refs, 1);
6122 bioc->fs_info = fs_info;
6123 bioc->replace_stripe_src = -1;
6124 bioc->full_stripe_logical = (u64)-1;
6125 bioc->logical = logical;
6130 void btrfs_get_bioc(struct btrfs_io_context *bioc)
6132 WARN_ON(!refcount_read(&bioc->refs));
6133 refcount_inc(&bioc->refs);
6136 void btrfs_put_bioc(struct btrfs_io_context *bioc)
6140 if (refcount_dec_and_test(&bioc->refs))
6145 * Please note that, discard won't be sent to target device of device
6148 struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info,
6149 u64 logical, u64 *length_ret,
6152 struct btrfs_chunk_map *map;
6153 struct btrfs_discard_stripe *stripes;
6154 u64 length = *length_ret;
6159 u64 stripe_end_offset;
6163 u32 sub_stripes = 0;
6164 u32 stripes_per_dev = 0;
6165 u32 remaining_stripes = 0;
6166 u32 last_stripe = 0;
6170 map = btrfs_get_chunk_map(fs_info, logical, length);
6172 return ERR_CAST(map);
6174 /* we don't discard raid56 yet */
6175 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6180 offset = logical - map->start;
6181 length = min_t(u64, map->start + map->chunk_len - logical, length);
6182 *length_ret = length;
6185 * stripe_nr counts the total number of stripes we have to stride
6186 * to get to this block
6188 stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
6190 /* stripe_offset is the offset of this block in its stripe */
6191 stripe_offset = offset - btrfs_stripe_nr_to_offset(stripe_nr);
6193 stripe_nr_end = round_up(offset + length, BTRFS_STRIPE_LEN) >>
6194 BTRFS_STRIPE_LEN_SHIFT;
6195 stripe_cnt = stripe_nr_end - stripe_nr;
6196 stripe_end_offset = btrfs_stripe_nr_to_offset(stripe_nr_end) -
6199 * after this, stripe_nr is the number of stripes on this
6200 * device we have to walk to find the data, and stripe_index is
6201 * the number of our device in the stripe array
6205 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6206 BTRFS_BLOCK_GROUP_RAID10)) {
6207 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6210 sub_stripes = map->sub_stripes;
6212 factor = map->num_stripes / sub_stripes;
6213 *num_stripes = min_t(u64, map->num_stripes,
6214 sub_stripes * stripe_cnt);
6215 stripe_index = stripe_nr % factor;
6216 stripe_nr /= factor;
6217 stripe_index *= sub_stripes;
6219 remaining_stripes = stripe_cnt % factor;
6220 stripes_per_dev = stripe_cnt / factor;
6221 last_stripe = ((stripe_nr_end - 1) % factor) * sub_stripes;
6222 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
6223 BTRFS_BLOCK_GROUP_DUP)) {
6224 *num_stripes = map->num_stripes;
6226 stripe_index = stripe_nr % map->num_stripes;
6227 stripe_nr /= map->num_stripes;
6230 stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS);
6236 for (i = 0; i < *num_stripes; i++) {
6237 stripes[i].physical =
6238 map->stripes[stripe_index].physical +
6239 stripe_offset + btrfs_stripe_nr_to_offset(stripe_nr);
6240 stripes[i].dev = map->stripes[stripe_index].dev;
6242 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6243 BTRFS_BLOCK_GROUP_RAID10)) {
6244 stripes[i].length = btrfs_stripe_nr_to_offset(stripes_per_dev);
6246 if (i / sub_stripes < remaining_stripes)
6247 stripes[i].length += BTRFS_STRIPE_LEN;
6250 * Special for the first stripe and
6253 * |-------|...|-------|
6257 if (i < sub_stripes)
6258 stripes[i].length -= stripe_offset;
6260 if (stripe_index >= last_stripe &&
6261 stripe_index <= (last_stripe +
6263 stripes[i].length -= stripe_end_offset;
6265 if (i == sub_stripes - 1)
6268 stripes[i].length = length;
6272 if (stripe_index == map->num_stripes) {
6278 btrfs_free_chunk_map(map);
6281 btrfs_free_chunk_map(map);
6282 return ERR_PTR(ret);
6285 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)
6287 struct btrfs_block_group *cache;
6290 /* Non zoned filesystem does not use "to_copy" flag */
6291 if (!btrfs_is_zoned(fs_info))
6294 cache = btrfs_lookup_block_group(fs_info, logical);
6296 ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags);
6298 btrfs_put_block_group(cache);
6302 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
6303 struct btrfs_io_context *bioc,
6304 struct btrfs_dev_replace *dev_replace,
6306 int *num_stripes_ret, int *max_errors_ret)
6308 u64 srcdev_devid = dev_replace->srcdev->devid;
6310 * At this stage, num_stripes is still the real number of stripes,
6311 * excluding the duplicated stripes.
6313 int num_stripes = *num_stripes_ret;
6314 int nr_extra_stripes = 0;
6315 int max_errors = *max_errors_ret;
6319 * A block group which has "to_copy" set will eventually be copied by
6320 * the dev-replace process. We can avoid cloning IO here.
6322 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))
6326 * Duplicate the write operations while the dev-replace procedure is
6327 * running. Since the copying of the old disk to the new disk takes
6328 * place at run time while the filesystem is mounted writable, the
6329 * regular write operations to the old disk have to be duplicated to go
6330 * to the new disk as well.
6332 * Note that device->missing is handled by the caller, and that the
6333 * write to the old disk is already set up in the stripes array.
6335 for (i = 0; i < num_stripes; i++) {
6336 struct btrfs_io_stripe *old = &bioc->stripes[i];
6337 struct btrfs_io_stripe *new = &bioc->stripes[num_stripes + nr_extra_stripes];
6339 if (old->dev->devid != srcdev_devid)
6342 new->physical = old->physical;
6343 new->dev = dev_replace->tgtdev;
6344 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK)
6345 bioc->replace_stripe_src = i;
6349 /* We can only have at most 2 extra nr_stripes (for DUP). */
6350 ASSERT(nr_extra_stripes <= 2);
6352 * For GET_READ_MIRRORS, we can only return at most 1 extra stripe for
6354 * If we have 2 extra stripes, only choose the one with smaller physical.
6356 if (op == BTRFS_MAP_GET_READ_MIRRORS && nr_extra_stripes == 2) {
6357 struct btrfs_io_stripe *first = &bioc->stripes[num_stripes];
6358 struct btrfs_io_stripe *second = &bioc->stripes[num_stripes + 1];
6360 /* Only DUP can have two extra stripes. */
6361 ASSERT(bioc->map_type & BTRFS_BLOCK_GROUP_DUP);
6364 * Swap the last stripe stripes and reduce @nr_extra_stripes.
6365 * The extra stripe would still be there, but won't be accessed.
6367 if (first->physical > second->physical) {
6368 swap(second->physical, first->physical);
6369 swap(second->dev, first->dev);
6374 *num_stripes_ret = num_stripes + nr_extra_stripes;
6375 *max_errors_ret = max_errors + nr_extra_stripes;
6376 bioc->replace_nr_stripes = nr_extra_stripes;
6379 static u64 btrfs_max_io_len(struct btrfs_chunk_map *map, u64 offset,
6380 struct btrfs_io_geometry *io_geom)
6383 * Stripe_nr is the stripe where this block falls. stripe_offset is
6384 * the offset of this block in its stripe.
6386 io_geom->stripe_offset = offset & BTRFS_STRIPE_LEN_MASK;
6387 io_geom->stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
6388 ASSERT(io_geom->stripe_offset < U32_MAX);
6390 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6391 unsigned long full_stripe_len =
6392 btrfs_stripe_nr_to_offset(nr_data_stripes(map));
6395 * For full stripe start, we use previously calculated
6396 * @stripe_nr. Align it to nr_data_stripes, then multiply with
6399 * By this we can avoid u64 division completely. And we have
6400 * to go rounddown(), not round_down(), as nr_data_stripes is
6401 * not ensured to be power of 2.
6403 io_geom->raid56_full_stripe_start = btrfs_stripe_nr_to_offset(
6404 rounddown(io_geom->stripe_nr, nr_data_stripes(map)));
6406 ASSERT(io_geom->raid56_full_stripe_start + full_stripe_len > offset);
6407 ASSERT(io_geom->raid56_full_stripe_start <= offset);
6409 * For writes to RAID56, allow to write a full stripe set, but
6410 * no straddling of stripe sets.
6412 if (io_geom->op == BTRFS_MAP_WRITE)
6413 return full_stripe_len - (offset - io_geom->raid56_full_stripe_start);
6417 * For other RAID types and for RAID56 reads, allow a single stripe (on
6420 if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK)
6421 return BTRFS_STRIPE_LEN - io_geom->stripe_offset;
6425 static int set_io_stripe(struct btrfs_fs_info *fs_info, u64 logical,
6426 u64 *length, struct btrfs_io_stripe *dst,
6427 struct btrfs_chunk_map *map,
6428 struct btrfs_io_geometry *io_geom)
6430 dst->dev = map->stripes[io_geom->stripe_index].dev;
6432 if (io_geom->op == BTRFS_MAP_READ &&
6433 btrfs_need_stripe_tree_update(fs_info, map->type))
6434 return btrfs_get_raid_extent_offset(fs_info, logical, length,
6436 io_geom->stripe_index, dst);
6438 dst->physical = map->stripes[io_geom->stripe_index].physical +
6439 io_geom->stripe_offset +
6440 btrfs_stripe_nr_to_offset(io_geom->stripe_nr);
6444 static bool is_single_device_io(struct btrfs_fs_info *fs_info,
6445 const struct btrfs_io_stripe *smap,
6446 const struct btrfs_chunk_map *map,
6447 int num_alloc_stripes,
6448 enum btrfs_map_op op, int mirror_num)
6453 if (num_alloc_stripes != 1)
6456 if (btrfs_need_stripe_tree_update(fs_info, map->type) && op != BTRFS_MAP_READ)
6459 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1)
6465 static void map_blocks_raid0(const struct btrfs_chunk_map *map,
6466 struct btrfs_io_geometry *io_geom)
6468 io_geom->stripe_index = io_geom->stripe_nr % map->num_stripes;
6469 io_geom->stripe_nr /= map->num_stripes;
6470 if (io_geom->op == BTRFS_MAP_READ)
6471 io_geom->mirror_num = 1;
6474 static void map_blocks_raid1(struct btrfs_fs_info *fs_info,
6475 struct btrfs_chunk_map *map,
6476 struct btrfs_io_geometry *io_geom,
6477 bool dev_replace_is_ongoing)
6479 if (io_geom->op != BTRFS_MAP_READ) {
6480 io_geom->num_stripes = map->num_stripes;
6484 if (io_geom->mirror_num) {
6485 io_geom->stripe_index = io_geom->mirror_num - 1;
6489 io_geom->stripe_index = find_live_mirror(fs_info, map, 0,
6490 dev_replace_is_ongoing);
6491 io_geom->mirror_num = io_geom->stripe_index + 1;
6494 static void map_blocks_dup(const struct btrfs_chunk_map *map,
6495 struct btrfs_io_geometry *io_geom)
6497 if (io_geom->op != BTRFS_MAP_READ) {
6498 io_geom->num_stripes = map->num_stripes;
6502 if (io_geom->mirror_num) {
6503 io_geom->stripe_index = io_geom->mirror_num - 1;
6507 io_geom->mirror_num = 1;
6510 static void map_blocks_raid10(struct btrfs_fs_info *fs_info,
6511 struct btrfs_chunk_map *map,
6512 struct btrfs_io_geometry *io_geom,
6513 bool dev_replace_is_ongoing)
6515 u32 factor = map->num_stripes / map->sub_stripes;
6516 int old_stripe_index;
6518 io_geom->stripe_index = (io_geom->stripe_nr % factor) * map->sub_stripes;
6519 io_geom->stripe_nr /= factor;
6521 if (io_geom->op != BTRFS_MAP_READ) {
6522 io_geom->num_stripes = map->sub_stripes;
6526 if (io_geom->mirror_num) {
6527 io_geom->stripe_index += io_geom->mirror_num - 1;
6531 old_stripe_index = io_geom->stripe_index;
6532 io_geom->stripe_index = find_live_mirror(fs_info, map,
6533 io_geom->stripe_index,
6534 dev_replace_is_ongoing);
6535 io_geom->mirror_num = io_geom->stripe_index - old_stripe_index + 1;
6538 static void map_blocks_raid56_write(struct btrfs_chunk_map *map,
6539 struct btrfs_io_geometry *io_geom,
6540 u64 logical, u64 *length)
6542 int data_stripes = nr_data_stripes(map);
6545 * Needs full stripe mapping.
6547 * Push stripe_nr back to the start of the full stripe For those cases
6548 * needing a full stripe, @stripe_nr is the full stripe number.
6550 * Originally we go raid56_full_stripe_start / full_stripe_len, but
6551 * that can be expensive. Here we just divide @stripe_nr with
6554 io_geom->stripe_nr /= data_stripes;
6556 /* RAID[56] write or recovery. Return all stripes */
6557 io_geom->num_stripes = map->num_stripes;
6558 io_geom->max_errors = btrfs_chunk_max_errors(map);
6560 /* Return the length to the full stripe end. */
6561 *length = min(logical + *length,
6562 io_geom->raid56_full_stripe_start + map->start +
6563 btrfs_stripe_nr_to_offset(data_stripes)) -
6565 io_geom->stripe_index = 0;
6566 io_geom->stripe_offset = 0;
6569 static void map_blocks_raid56_read(struct btrfs_chunk_map *map,
6570 struct btrfs_io_geometry *io_geom)
6572 int data_stripes = nr_data_stripes(map);
6574 ASSERT(io_geom->mirror_num <= 1);
6575 /* Just grab the data stripe directly. */
6576 io_geom->stripe_index = io_geom->stripe_nr % data_stripes;
6577 io_geom->stripe_nr /= data_stripes;
6579 /* We distribute the parity blocks across stripes. */
6580 io_geom->stripe_index =
6581 (io_geom->stripe_nr + io_geom->stripe_index) % map->num_stripes;
6583 if (io_geom->op == BTRFS_MAP_READ && io_geom->mirror_num < 1)
6584 io_geom->mirror_num = 1;
6587 static void map_blocks_single(const struct btrfs_chunk_map *map,
6588 struct btrfs_io_geometry *io_geom)
6590 io_geom->stripe_index = io_geom->stripe_nr % map->num_stripes;
6591 io_geom->stripe_nr /= map->num_stripes;
6592 io_geom->mirror_num = io_geom->stripe_index + 1;
6596 * Map one logical range to one or more physical ranges.
6598 * @length: (Mandatory) mapped length of this run.
6599 * One logical range can be split into different segments
6600 * due to factors like zones and RAID0/5/6/10 stripe
6603 * @bioc_ret: (Mandatory) returned btrfs_io_context structure.
6604 * which has one or more physical ranges (btrfs_io_stripe)
6606 * Caller should call btrfs_put_bioc() to free it after use.
6608 * @smap: (Optional) single physical range optimization.
6609 * If the map request can be fulfilled by one single
6610 * physical range, and this is parameter is not NULL,
6611 * then @bioc_ret would be NULL, and @smap would be
6614 * @mirror_num_ret: (Mandatory) returned mirror number if the original
6617 * Mirror number 0 means to choose any live mirrors.
6619 * For non-RAID56 profiles, non-zero mirror_num means
6620 * the Nth mirror. (e.g. mirror_num 1 means the first
6623 * For RAID56 profile, mirror 1 means rebuild from P and
6624 * the remaining data stripes.
6626 * For RAID6 profile, mirror > 2 means mark another
6627 * data/P stripe error and rebuild from the remaining
6630 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6631 u64 logical, u64 *length,
6632 struct btrfs_io_context **bioc_ret,
6633 struct btrfs_io_stripe *smap, int *mirror_num_ret)
6635 struct btrfs_chunk_map *map;
6636 struct btrfs_io_geometry io_geom = { 0 };
6641 struct btrfs_io_context *bioc = NULL;
6642 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6643 int dev_replace_is_ongoing = 0;
6644 u16 num_alloc_stripes;
6649 io_geom.mirror_num = (mirror_num_ret ? *mirror_num_ret : 0);
6650 io_geom.num_stripes = 1;
6651 io_geom.stripe_index = 0;
6654 num_copies = btrfs_num_copies(fs_info, logical, fs_info->sectorsize);
6655 if (io_geom.mirror_num > num_copies)
6658 map = btrfs_get_chunk_map(fs_info, logical, *length);
6660 return PTR_ERR(map);
6662 map_offset = logical - map->start;
6663 io_geom.raid56_full_stripe_start = (u64)-1;
6664 max_len = btrfs_max_io_len(map, map_offset, &io_geom);
6665 *length = min_t(u64, map->chunk_len - map_offset, max_len);
6667 down_read(&dev_replace->rwsem);
6668 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6670 * Hold the semaphore for read during the whole operation, write is
6671 * requested at commit time but must wait.
6673 if (!dev_replace_is_ongoing)
6674 up_read(&dev_replace->rwsem);
6676 switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
6677 case BTRFS_BLOCK_GROUP_RAID0:
6678 map_blocks_raid0(map, &io_geom);
6680 case BTRFS_BLOCK_GROUP_RAID1:
6681 case BTRFS_BLOCK_GROUP_RAID1C3:
6682 case BTRFS_BLOCK_GROUP_RAID1C4:
6683 map_blocks_raid1(fs_info, map, &io_geom, dev_replace_is_ongoing);
6685 case BTRFS_BLOCK_GROUP_DUP:
6686 map_blocks_dup(map, &io_geom);
6688 case BTRFS_BLOCK_GROUP_RAID10:
6689 map_blocks_raid10(fs_info, map, &io_geom, dev_replace_is_ongoing);
6691 case BTRFS_BLOCK_GROUP_RAID5:
6692 case BTRFS_BLOCK_GROUP_RAID6:
6693 if (op != BTRFS_MAP_READ || io_geom.mirror_num > 1)
6694 map_blocks_raid56_write(map, &io_geom, logical, length);
6696 map_blocks_raid56_read(map, &io_geom);
6700 * After this, stripe_nr is the number of stripes on this
6701 * device we have to walk to find the data, and stripe_index is
6702 * the number of our device in the stripe array
6704 map_blocks_single(map, &io_geom);
6707 if (io_geom.stripe_index >= map->num_stripes) {
6709 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6710 io_geom.stripe_index, map->num_stripes);
6715 num_alloc_stripes = io_geom.num_stripes;
6716 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6717 op != BTRFS_MAP_READ)
6719 * For replace case, we need to add extra stripes for extra
6720 * duplicated stripes.
6722 * For both WRITE and GET_READ_MIRRORS, we may have at most
6723 * 2 more stripes (DUP types, otherwise 1).
6725 num_alloc_stripes += 2;
6728 * If this I/O maps to a single device, try to return the device and
6729 * physical block information on the stack instead of allocating an
6730 * I/O context structure.
6732 if (is_single_device_io(fs_info, smap, map, num_alloc_stripes, op,
6733 io_geom.mirror_num)) {
6734 ret = set_io_stripe(fs_info, logical, length, smap, map, &io_geom);
6736 *mirror_num_ret = io_geom.mirror_num;
6741 bioc = alloc_btrfs_io_context(fs_info, logical, num_alloc_stripes);
6746 bioc->map_type = map->type;
6749 * For RAID56 full map, we need to make sure the stripes[] follows the
6750 * rule that data stripes are all ordered, then followed with P and Q
6753 * It's still mostly the same as other profiles, just with extra rotation.
6755 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
6756 (op != BTRFS_MAP_READ || io_geom.mirror_num > 1)) {
6758 * For RAID56 @stripe_nr is already the number of full stripes
6759 * before us, which is also the rotation value (needs to modulo
6760 * with num_stripes).
6762 * In this case, we just add @stripe_nr with @i, then do the
6763 * modulo, to reduce one modulo call.
6765 bioc->full_stripe_logical = map->start +
6766 btrfs_stripe_nr_to_offset(io_geom.stripe_nr *
6767 nr_data_stripes(map));
6768 for (int i = 0; i < io_geom.num_stripes; i++) {
6769 struct btrfs_io_stripe *dst = &bioc->stripes[i];
6772 stripe_index = (i + io_geom.stripe_nr) % io_geom.num_stripes;
6773 dst->dev = map->stripes[stripe_index].dev;
6775 map->stripes[stripe_index].physical +
6776 io_geom.stripe_offset +
6777 btrfs_stripe_nr_to_offset(io_geom.stripe_nr);
6781 * For all other non-RAID56 profiles, just copy the target
6782 * stripe into the bioc.
6784 for (i = 0; i < io_geom.num_stripes; i++) {
6785 ret = set_io_stripe(fs_info, logical, length,
6786 &bioc->stripes[i], map, &io_geom);
6789 io_geom.stripe_index++;
6795 btrfs_put_bioc(bioc);
6799 if (op != BTRFS_MAP_READ)
6800 io_geom.max_errors = btrfs_chunk_max_errors(map);
6802 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6803 op != BTRFS_MAP_READ) {
6804 handle_ops_on_dev_replace(op, bioc, dev_replace, logical,
6805 &io_geom.num_stripes, &io_geom.max_errors);
6809 bioc->num_stripes = io_geom.num_stripes;
6810 bioc->max_errors = io_geom.max_errors;
6811 bioc->mirror_num = io_geom.mirror_num;
6814 if (dev_replace_is_ongoing) {
6815 lockdep_assert_held(&dev_replace->rwsem);
6816 /* Unlock and let waiting writers proceed */
6817 up_read(&dev_replace->rwsem);
6819 btrfs_free_chunk_map(map);
6823 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args,
6824 const struct btrfs_fs_devices *fs_devices)
6826 if (args->fsid == NULL)
6828 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0)
6833 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args,
6834 const struct btrfs_device *device)
6836 if (args->missing) {
6837 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) &&
6843 if (device->devid != args->devid)
6845 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0)
6851 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6854 * If devid and uuid are both specified, the match must be exact, otherwise
6855 * only devid is used.
6857 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,
6858 const struct btrfs_dev_lookup_args *args)
6860 struct btrfs_device *device;
6861 struct btrfs_fs_devices *seed_devs;
6863 if (dev_args_match_fs_devices(args, fs_devices)) {
6864 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6865 if (dev_args_match_device(args, device))
6870 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6871 if (!dev_args_match_fs_devices(args, seed_devs))
6873 list_for_each_entry(device, &seed_devs->devices, dev_list) {
6874 if (dev_args_match_device(args, device))
6882 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6883 u64 devid, u8 *dev_uuid)
6885 struct btrfs_device *device;
6886 unsigned int nofs_flag;
6889 * We call this under the chunk_mutex, so we want to use NOFS for this
6890 * allocation, however we don't want to change btrfs_alloc_device() to
6891 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6895 nofs_flag = memalloc_nofs_save();
6896 device = btrfs_alloc_device(NULL, &devid, dev_uuid, NULL);
6897 memalloc_nofs_restore(nofs_flag);
6901 list_add(&device->dev_list, &fs_devices->devices);
6902 device->fs_devices = fs_devices;
6903 fs_devices->num_devices++;
6905 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6906 fs_devices->missing_devices++;
6912 * Allocate new device struct, set up devid and UUID.
6914 * @fs_info: used only for generating a new devid, can be NULL if
6915 * devid is provided (i.e. @devid != NULL).
6916 * @devid: a pointer to devid for this device. If NULL a new devid
6918 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6920 * @path: a pointer to device path if available, NULL otherwise.
6922 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6923 * on error. Returned struct is not linked onto any lists and must be
6924 * destroyed with btrfs_free_device.
6926 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6927 const u64 *devid, const u8 *uuid,
6930 struct btrfs_device *dev;
6933 if (WARN_ON(!devid && !fs_info))
6934 return ERR_PTR(-EINVAL);
6936 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
6938 return ERR_PTR(-ENOMEM);
6940 INIT_LIST_HEAD(&dev->dev_list);
6941 INIT_LIST_HEAD(&dev->dev_alloc_list);
6942 INIT_LIST_HEAD(&dev->post_commit_list);
6944 atomic_set(&dev->dev_stats_ccnt, 0);
6945 btrfs_device_data_ordered_init(dev);
6946 extent_io_tree_init(fs_info, &dev->alloc_state, IO_TREE_DEVICE_ALLOC_STATE);
6953 ret = find_next_devid(fs_info, &tmp);
6955 btrfs_free_device(dev);
6956 return ERR_PTR(ret);
6962 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6964 generate_random_uuid(dev->uuid);
6967 struct rcu_string *name;
6969 name = rcu_string_strdup(path, GFP_KERNEL);
6971 btrfs_free_device(dev);
6972 return ERR_PTR(-ENOMEM);
6974 rcu_assign_pointer(dev->name, name);
6980 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6981 u64 devid, u8 *uuid, bool error)
6984 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6987 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6991 u64 btrfs_calc_stripe_length(const struct btrfs_chunk_map *map)
6993 const int data_stripes = calc_data_stripes(map->type, map->num_stripes);
6995 return div_u64(map->chunk_len, data_stripes);
6998 #if BITS_PER_LONG == 32
7000 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
7001 * can't be accessed on 32bit systems.
7003 * This function do mount time check to reject the fs if it already has
7004 * metadata chunk beyond that limit.
7006 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
7007 u64 logical, u64 length, u64 type)
7009 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
7012 if (logical + length < MAX_LFS_FILESIZE)
7015 btrfs_err_32bit_limit(fs_info);
7020 * This is to give early warning for any metadata chunk reaching
7021 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
7022 * Although we can still access the metadata, it's not going to be possible
7023 * once the limit is reached.
7025 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
7026 u64 logical, u64 length, u64 type)
7028 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
7031 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)
7034 btrfs_warn_32bit_limit(fs_info);
7038 static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info,
7039 u64 devid, u8 *uuid)
7041 struct btrfs_device *dev;
7043 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7044 btrfs_report_missing_device(fs_info, devid, uuid, true);
7045 return ERR_PTR(-ENOENT);
7048 dev = add_missing_dev(fs_info->fs_devices, devid, uuid);
7050 btrfs_err(fs_info, "failed to init missing device %llu: %ld",
7051 devid, PTR_ERR(dev));
7054 btrfs_report_missing_device(fs_info, devid, uuid, false);
7059 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
7060 struct btrfs_chunk *chunk)
7062 BTRFS_DEV_LOOKUP_ARGS(args);
7063 struct btrfs_fs_info *fs_info = leaf->fs_info;
7064 struct btrfs_chunk_map *map;
7069 u8 uuid[BTRFS_UUID_SIZE];
7075 logical = key->offset;
7076 length = btrfs_chunk_length(leaf, chunk);
7077 type = btrfs_chunk_type(leaf, chunk);
7078 index = btrfs_bg_flags_to_raid_index(type);
7079 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
7081 #if BITS_PER_LONG == 32
7082 ret = check_32bit_meta_chunk(fs_info, logical, length, type);
7085 warn_32bit_meta_chunk(fs_info, logical, length, type);
7089 * Only need to verify chunk item if we're reading from sys chunk array,
7090 * as chunk item in tree block is already verified by tree-checker.
7092 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
7093 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
7098 map = btrfs_find_chunk_map(fs_info, logical, 1);
7100 /* already mapped? */
7101 if (map && map->start <= logical && map->start + map->chunk_len > logical) {
7102 btrfs_free_chunk_map(map);
7105 btrfs_free_chunk_map(map);
7108 map = btrfs_alloc_chunk_map(num_stripes, GFP_NOFS);
7112 map->start = logical;
7113 map->chunk_len = length;
7114 map->num_stripes = num_stripes;
7115 map->io_width = btrfs_chunk_io_width(leaf, chunk);
7116 map->io_align = btrfs_chunk_io_align(leaf, chunk);
7119 * We can't use the sub_stripes value, as for profiles other than
7120 * RAID10, they may have 0 as sub_stripes for filesystems created by
7121 * older mkfs (<v5.4).
7122 * In that case, it can cause divide-by-zero errors later.
7123 * Since currently sub_stripes is fixed for each profile, let's
7124 * use the trusted value instead.
7126 map->sub_stripes = btrfs_raid_array[index].sub_stripes;
7127 map->verified_stripes = 0;
7128 map->stripe_size = btrfs_calc_stripe_length(map);
7129 for (i = 0; i < num_stripes; i++) {
7130 map->stripes[i].physical =
7131 btrfs_stripe_offset_nr(leaf, chunk, i);
7132 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
7134 read_extent_buffer(leaf, uuid, (unsigned long)
7135 btrfs_stripe_dev_uuid_nr(chunk, i),
7138 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args);
7139 if (!map->stripes[i].dev) {
7140 map->stripes[i].dev = handle_missing_device(fs_info,
7142 if (IS_ERR(map->stripes[i].dev)) {
7143 ret = PTR_ERR(map->stripes[i].dev);
7144 btrfs_free_chunk_map(map);
7149 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
7150 &(map->stripes[i].dev->dev_state));
7153 ret = btrfs_add_chunk_map(fs_info, map);
7156 "failed to add chunk map, start=%llu len=%llu: %d",
7157 map->start, map->chunk_len, ret);
7163 static void fill_device_from_item(struct extent_buffer *leaf,
7164 struct btrfs_dev_item *dev_item,
7165 struct btrfs_device *device)
7169 device->devid = btrfs_device_id(leaf, dev_item);
7170 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
7171 device->total_bytes = device->disk_total_bytes;
7172 device->commit_total_bytes = device->disk_total_bytes;
7173 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
7174 device->commit_bytes_used = device->bytes_used;
7175 device->type = btrfs_device_type(leaf, dev_item);
7176 device->io_align = btrfs_device_io_align(leaf, dev_item);
7177 device->io_width = btrfs_device_io_width(leaf, dev_item);
7178 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
7179 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
7180 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
7182 ptr = btrfs_device_uuid(dev_item);
7183 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
7186 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
7189 struct btrfs_fs_devices *fs_devices;
7192 lockdep_assert_held(&uuid_mutex);
7195 /* This will match only for multi-device seed fs */
7196 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
7197 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
7201 fs_devices = find_fsid(fsid, NULL);
7203 if (!btrfs_test_opt(fs_info, DEGRADED))
7204 return ERR_PTR(-ENOENT);
7206 fs_devices = alloc_fs_devices(fsid);
7207 if (IS_ERR(fs_devices))
7210 fs_devices->seeding = true;
7211 fs_devices->opened = 1;
7216 * Upon first call for a seed fs fsid, just create a private copy of the
7217 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
7219 fs_devices = clone_fs_devices(fs_devices);
7220 if (IS_ERR(fs_devices))
7223 ret = open_fs_devices(fs_devices, BLK_OPEN_READ, fs_info->bdev_holder);
7225 free_fs_devices(fs_devices);
7226 return ERR_PTR(ret);
7229 if (!fs_devices->seeding) {
7230 close_fs_devices(fs_devices);
7231 free_fs_devices(fs_devices);
7232 return ERR_PTR(-EINVAL);
7235 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
7240 static int read_one_dev(struct extent_buffer *leaf,
7241 struct btrfs_dev_item *dev_item)
7243 BTRFS_DEV_LOOKUP_ARGS(args);
7244 struct btrfs_fs_info *fs_info = leaf->fs_info;
7245 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7246 struct btrfs_device *device;
7249 u8 fs_uuid[BTRFS_FSID_SIZE];
7250 u8 dev_uuid[BTRFS_UUID_SIZE];
7252 devid = btrfs_device_id(leaf, dev_item);
7254 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
7256 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
7258 args.uuid = dev_uuid;
7259 args.fsid = fs_uuid;
7261 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
7262 fs_devices = open_seed_devices(fs_info, fs_uuid);
7263 if (IS_ERR(fs_devices))
7264 return PTR_ERR(fs_devices);
7267 device = btrfs_find_device(fs_info->fs_devices, &args);
7269 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7270 btrfs_report_missing_device(fs_info, devid,
7275 device = add_missing_dev(fs_devices, devid, dev_uuid);
7276 if (IS_ERR(device)) {
7278 "failed to add missing dev %llu: %ld",
7279 devid, PTR_ERR(device));
7280 return PTR_ERR(device);
7282 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
7284 if (!device->bdev) {
7285 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7286 btrfs_report_missing_device(fs_info,
7287 devid, dev_uuid, true);
7290 btrfs_report_missing_device(fs_info, devid,
7294 if (!device->bdev &&
7295 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
7297 * this happens when a device that was properly setup
7298 * in the device info lists suddenly goes bad.
7299 * device->bdev is NULL, and so we have to set
7300 * device->missing to one here
7302 device->fs_devices->missing_devices++;
7303 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
7306 /* Move the device to its own fs_devices */
7307 if (device->fs_devices != fs_devices) {
7308 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
7309 &device->dev_state));
7311 list_move(&device->dev_list, &fs_devices->devices);
7312 device->fs_devices->num_devices--;
7313 fs_devices->num_devices++;
7315 device->fs_devices->missing_devices--;
7316 fs_devices->missing_devices++;
7318 device->fs_devices = fs_devices;
7322 if (device->fs_devices != fs_info->fs_devices) {
7323 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
7324 if (device->generation !=
7325 btrfs_device_generation(leaf, dev_item))
7329 fill_device_from_item(leaf, dev_item, device);
7331 u64 max_total_bytes = bdev_nr_bytes(device->bdev);
7333 if (device->total_bytes > max_total_bytes) {
7335 "device total_bytes should be at most %llu but found %llu",
7336 max_total_bytes, device->total_bytes);
7340 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7341 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7342 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7343 device->fs_devices->total_rw_bytes += device->total_bytes;
7344 atomic64_add(device->total_bytes - device->bytes_used,
7345 &fs_info->free_chunk_space);
7351 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7353 struct btrfs_super_block *super_copy = fs_info->super_copy;
7354 struct extent_buffer *sb;
7355 struct btrfs_disk_key *disk_key;
7356 struct btrfs_chunk *chunk;
7358 unsigned long sb_array_offset;
7365 struct btrfs_key key;
7367 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7370 * We allocated a dummy extent, just to use extent buffer accessors.
7371 * There will be unused space after BTRFS_SUPER_INFO_SIZE, but
7372 * that's fine, we will not go beyond system chunk array anyway.
7374 sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET);
7377 set_extent_buffer_uptodate(sb);
7379 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7380 array_size = btrfs_super_sys_array_size(super_copy);
7382 array_ptr = super_copy->sys_chunk_array;
7383 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7386 while (cur_offset < array_size) {
7387 disk_key = (struct btrfs_disk_key *)array_ptr;
7388 len = sizeof(*disk_key);
7389 if (cur_offset + len > array_size)
7390 goto out_short_read;
7392 btrfs_disk_key_to_cpu(&key, disk_key);
7395 sb_array_offset += len;
7398 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7400 "unexpected item type %u in sys_array at offset %u",
7401 (u32)key.type, cur_offset);
7406 chunk = (struct btrfs_chunk *)sb_array_offset;
7408 * At least one btrfs_chunk with one stripe must be present,
7409 * exact stripe count check comes afterwards
7411 len = btrfs_chunk_item_size(1);
7412 if (cur_offset + len > array_size)
7413 goto out_short_read;
7415 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7418 "invalid number of stripes %u in sys_array at offset %u",
7419 num_stripes, cur_offset);
7424 type = btrfs_chunk_type(sb, chunk);
7425 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7427 "invalid chunk type %llu in sys_array at offset %u",
7433 len = btrfs_chunk_item_size(num_stripes);
7434 if (cur_offset + len > array_size)
7435 goto out_short_read;
7437 ret = read_one_chunk(&key, sb, chunk);
7442 sb_array_offset += len;
7445 clear_extent_buffer_uptodate(sb);
7446 free_extent_buffer_stale(sb);
7450 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7452 clear_extent_buffer_uptodate(sb);
7453 free_extent_buffer_stale(sb);
7458 * Check if all chunks in the fs are OK for read-write degraded mount
7460 * If the @failing_dev is specified, it's accounted as missing.
7462 * Return true if all chunks meet the minimal RW mount requirements.
7463 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7465 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7466 struct btrfs_device *failing_dev)
7468 struct btrfs_chunk_map *map;
7472 map = btrfs_find_chunk_map(fs_info, 0, U64_MAX);
7473 /* No chunk at all? Return false anyway */
7484 btrfs_get_num_tolerated_disk_barrier_failures(
7486 for (i = 0; i < map->num_stripes; i++) {
7487 struct btrfs_device *dev = map->stripes[i].dev;
7489 if (!dev || !dev->bdev ||
7490 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7491 dev->last_flush_error)
7493 else if (failing_dev && failing_dev == dev)
7496 if (missing > max_tolerated) {
7499 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7500 map->start, missing, max_tolerated);
7501 btrfs_free_chunk_map(map);
7505 next_start = map->start + map->chunk_len;
7506 btrfs_free_chunk_map(map);
7508 map = btrfs_find_chunk_map(fs_info, next_start, U64_MAX - next_start);
7514 static void readahead_tree_node_children(struct extent_buffer *node)
7517 const int nr_items = btrfs_header_nritems(node);
7519 for (i = 0; i < nr_items; i++)
7520 btrfs_readahead_node_child(node, i);
7523 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7525 struct btrfs_root *root = fs_info->chunk_root;
7526 struct btrfs_path *path;
7527 struct extent_buffer *leaf;
7528 struct btrfs_key key;
7529 struct btrfs_key found_key;
7534 u64 last_ra_node = 0;
7536 path = btrfs_alloc_path();
7541 * uuid_mutex is needed only if we are mounting a sprout FS
7542 * otherwise we don't need it.
7544 mutex_lock(&uuid_mutex);
7547 * It is possible for mount and umount to race in such a way that
7548 * we execute this code path, but open_fs_devices failed to clear
7549 * total_rw_bytes. We certainly want it cleared before reading the
7550 * device items, so clear it here.
7552 fs_info->fs_devices->total_rw_bytes = 0;
7555 * Lockdep complains about possible circular locking dependency between
7556 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores
7557 * used for freeze procection of a fs (struct super_block.s_writers),
7558 * which we take when starting a transaction, and extent buffers of the
7559 * chunk tree if we call read_one_dev() while holding a lock on an
7560 * extent buffer of the chunk tree. Since we are mounting the filesystem
7561 * and at this point there can't be any concurrent task modifying the
7562 * chunk tree, to keep it simple, just skip locking on the chunk tree.
7564 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
7565 path->skip_locking = 1;
7568 * Read all device items, and then all the chunk items. All
7569 * device items are found before any chunk item (their object id
7570 * is smaller than the lowest possible object id for a chunk
7571 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7573 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7576 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) {
7577 struct extent_buffer *node = path->nodes[1];
7579 leaf = path->nodes[0];
7580 slot = path->slots[0];
7583 if (last_ra_node != node->start) {
7584 readahead_tree_node_children(node);
7585 last_ra_node = node->start;
7588 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7589 struct btrfs_dev_item *dev_item;
7590 dev_item = btrfs_item_ptr(leaf, slot,
7591 struct btrfs_dev_item);
7592 ret = read_one_dev(leaf, dev_item);
7596 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7597 struct btrfs_chunk *chunk;
7600 * We are only called at mount time, so no need to take
7601 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7602 * we always lock first fs_info->chunk_mutex before
7603 * acquiring any locks on the chunk tree. This is a
7604 * requirement for chunk allocation, see the comment on
7605 * top of btrfs_chunk_alloc() for details.
7607 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7608 ret = read_one_chunk(&found_key, leaf, chunk);
7613 /* Catch error found during iteration */
7620 * After loading chunk tree, we've got all device information,
7621 * do another round of validation checks.
7623 if (total_dev != fs_info->fs_devices->total_devices) {
7625 "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit",
7626 btrfs_super_num_devices(fs_info->super_copy),
7628 fs_info->fs_devices->total_devices = total_dev;
7629 btrfs_set_super_num_devices(fs_info->super_copy, total_dev);
7631 if (btrfs_super_total_bytes(fs_info->super_copy) <
7632 fs_info->fs_devices->total_rw_bytes) {
7634 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7635 btrfs_super_total_bytes(fs_info->super_copy),
7636 fs_info->fs_devices->total_rw_bytes);
7642 mutex_unlock(&uuid_mutex);
7644 btrfs_free_path(path);
7648 int btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7650 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7651 struct btrfs_device *device;
7654 fs_devices->fs_info = fs_info;
7656 mutex_lock(&fs_devices->device_list_mutex);
7657 list_for_each_entry(device, &fs_devices->devices, dev_list)
7658 device->fs_info = fs_info;
7660 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7661 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7662 device->fs_info = fs_info;
7663 ret = btrfs_get_dev_zone_info(device, false);
7668 seed_devs->fs_info = fs_info;
7670 mutex_unlock(&fs_devices->device_list_mutex);
7675 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7676 const struct btrfs_dev_stats_item *ptr,
7681 read_extent_buffer(eb, &val,
7682 offsetof(struct btrfs_dev_stats_item, values) +
7683 ((unsigned long)ptr) + (index * sizeof(u64)),
7688 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7689 struct btrfs_dev_stats_item *ptr,
7692 write_extent_buffer(eb, &val,
7693 offsetof(struct btrfs_dev_stats_item, values) +
7694 ((unsigned long)ptr) + (index * sizeof(u64)),
7698 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7699 struct btrfs_path *path)
7701 struct btrfs_dev_stats_item *ptr;
7702 struct extent_buffer *eb;
7703 struct btrfs_key key;
7707 if (!device->fs_info->dev_root)
7710 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7711 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7712 key.offset = device->devid;
7713 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7715 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7716 btrfs_dev_stat_set(device, i, 0);
7717 device->dev_stats_valid = 1;
7718 btrfs_release_path(path);
7719 return ret < 0 ? ret : 0;
7721 slot = path->slots[0];
7722 eb = path->nodes[0];
7723 item_size = btrfs_item_size(eb, slot);
7725 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7727 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7728 if (item_size >= (1 + i) * sizeof(__le64))
7729 btrfs_dev_stat_set(device, i,
7730 btrfs_dev_stats_value(eb, ptr, i));
7732 btrfs_dev_stat_set(device, i, 0);
7735 device->dev_stats_valid = 1;
7736 btrfs_dev_stat_print_on_load(device);
7737 btrfs_release_path(path);
7742 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7744 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7745 struct btrfs_device *device;
7746 struct btrfs_path *path = NULL;
7749 path = btrfs_alloc_path();
7753 mutex_lock(&fs_devices->device_list_mutex);
7754 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7755 ret = btrfs_device_init_dev_stats(device, path);
7759 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7760 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7761 ret = btrfs_device_init_dev_stats(device, path);
7767 mutex_unlock(&fs_devices->device_list_mutex);
7769 btrfs_free_path(path);
7773 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7774 struct btrfs_device *device)
7776 struct btrfs_fs_info *fs_info = trans->fs_info;
7777 struct btrfs_root *dev_root = fs_info->dev_root;
7778 struct btrfs_path *path;
7779 struct btrfs_key key;
7780 struct extent_buffer *eb;
7781 struct btrfs_dev_stats_item *ptr;
7785 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7786 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7787 key.offset = device->devid;
7789 path = btrfs_alloc_path();
7792 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7794 btrfs_warn_in_rcu(fs_info,
7795 "error %d while searching for dev_stats item for device %s",
7796 ret, btrfs_dev_name(device));
7801 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7802 /* need to delete old one and insert a new one */
7803 ret = btrfs_del_item(trans, dev_root, path);
7805 btrfs_warn_in_rcu(fs_info,
7806 "delete too small dev_stats item for device %s failed %d",
7807 btrfs_dev_name(device), ret);
7814 /* need to insert a new item */
7815 btrfs_release_path(path);
7816 ret = btrfs_insert_empty_item(trans, dev_root, path,
7817 &key, sizeof(*ptr));
7819 btrfs_warn_in_rcu(fs_info,
7820 "insert dev_stats item for device %s failed %d",
7821 btrfs_dev_name(device), ret);
7826 eb = path->nodes[0];
7827 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7828 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7829 btrfs_set_dev_stats_value(eb, ptr, i,
7830 btrfs_dev_stat_read(device, i));
7831 btrfs_mark_buffer_dirty(trans, eb);
7834 btrfs_free_path(path);
7839 * called from commit_transaction. Writes all changed device stats to disk.
7841 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7843 struct btrfs_fs_info *fs_info = trans->fs_info;
7844 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7845 struct btrfs_device *device;
7849 mutex_lock(&fs_devices->device_list_mutex);
7850 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7851 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7852 if (!device->dev_stats_valid || stats_cnt == 0)
7857 * There is a LOAD-LOAD control dependency between the value of
7858 * dev_stats_ccnt and updating the on-disk values which requires
7859 * reading the in-memory counters. Such control dependencies
7860 * require explicit read memory barriers.
7862 * This memory barriers pairs with smp_mb__before_atomic in
7863 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7864 * barrier implied by atomic_xchg in
7865 * btrfs_dev_stats_read_and_reset
7869 ret = update_dev_stat_item(trans, device);
7871 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7873 mutex_unlock(&fs_devices->device_list_mutex);
7878 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7880 btrfs_dev_stat_inc(dev, index);
7882 if (!dev->dev_stats_valid)
7884 btrfs_err_rl_in_rcu(dev->fs_info,
7885 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7886 btrfs_dev_name(dev),
7887 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7888 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7889 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7890 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7891 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7894 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7898 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7899 if (btrfs_dev_stat_read(dev, i) != 0)
7901 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7902 return; /* all values == 0, suppress message */
7904 btrfs_info_in_rcu(dev->fs_info,
7905 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7906 btrfs_dev_name(dev),
7907 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7908 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7909 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7910 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7911 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7914 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7915 struct btrfs_ioctl_get_dev_stats *stats)
7917 BTRFS_DEV_LOOKUP_ARGS(args);
7918 struct btrfs_device *dev;
7919 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7922 mutex_lock(&fs_devices->device_list_mutex);
7923 args.devid = stats->devid;
7924 dev = btrfs_find_device(fs_info->fs_devices, &args);
7925 mutex_unlock(&fs_devices->device_list_mutex);
7928 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7930 } else if (!dev->dev_stats_valid) {
7931 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7933 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7934 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7935 if (stats->nr_items > i)
7937 btrfs_dev_stat_read_and_reset(dev, i);
7939 btrfs_dev_stat_set(dev, i, 0);
7941 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7942 current->comm, task_pid_nr(current));
7944 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7945 if (stats->nr_items > i)
7946 stats->values[i] = btrfs_dev_stat_read(dev, i);
7948 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7949 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7954 * Update the size and bytes used for each device where it changed. This is
7955 * delayed since we would otherwise get errors while writing out the
7958 * Must be invoked during transaction commit.
7960 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7962 struct btrfs_device *curr, *next;
7964 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7966 if (list_empty(&trans->dev_update_list))
7970 * We don't need the device_list_mutex here. This list is owned by the
7971 * transaction and the transaction must complete before the device is
7974 mutex_lock(&trans->fs_info->chunk_mutex);
7975 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7977 list_del_init(&curr->post_commit_list);
7978 curr->commit_total_bytes = curr->disk_total_bytes;
7979 curr->commit_bytes_used = curr->bytes_used;
7981 mutex_unlock(&trans->fs_info->chunk_mutex);
7985 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7987 int btrfs_bg_type_to_factor(u64 flags)
7989 const int index = btrfs_bg_flags_to_raid_index(flags);
7991 return btrfs_raid_array[index].ncopies;
7996 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7997 u64 chunk_offset, u64 devid,
7998 u64 physical_offset, u64 physical_len)
8000 struct btrfs_dev_lookup_args args = { .devid = devid };
8001 struct btrfs_chunk_map *map;
8002 struct btrfs_device *dev;
8008 map = btrfs_find_chunk_map(fs_info, chunk_offset, 1);
8011 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
8012 physical_offset, devid);
8017 stripe_len = btrfs_calc_stripe_length(map);
8018 if (physical_len != stripe_len) {
8020 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
8021 physical_offset, devid, map->start, physical_len,
8028 * Very old mkfs.btrfs (before v4.1) will not respect the reserved
8029 * space. Although kernel can handle it without problem, better to warn
8032 if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED)
8034 "devid %llu physical %llu len %llu inside the reserved space",
8035 devid, physical_offset, physical_len);
8037 for (i = 0; i < map->num_stripes; i++) {
8038 if (map->stripes[i].dev->devid == devid &&
8039 map->stripes[i].physical == physical_offset) {
8041 if (map->verified_stripes >= map->num_stripes) {
8043 "too many dev extents for chunk %llu found",
8048 map->verified_stripes++;
8054 "dev extent physical offset %llu devid %llu has no corresponding chunk",
8055 physical_offset, devid);
8059 /* Make sure no dev extent is beyond device boundary */
8060 dev = btrfs_find_device(fs_info->fs_devices, &args);
8062 btrfs_err(fs_info, "failed to find devid %llu", devid);
8067 if (physical_offset + physical_len > dev->disk_total_bytes) {
8069 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
8070 devid, physical_offset, physical_len,
8071 dev->disk_total_bytes);
8076 if (dev->zone_info) {
8077 u64 zone_size = dev->zone_info->zone_size;
8079 if (!IS_ALIGNED(physical_offset, zone_size) ||
8080 !IS_ALIGNED(physical_len, zone_size)) {
8082 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
8083 devid, physical_offset, physical_len);
8090 btrfs_free_chunk_map(map);
8094 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
8096 struct rb_node *node;
8099 read_lock(&fs_info->mapping_tree_lock);
8100 for (node = rb_first_cached(&fs_info->mapping_tree); node; node = rb_next(node)) {
8101 struct btrfs_chunk_map *map;
8103 map = rb_entry(node, struct btrfs_chunk_map, rb_node);
8104 if (map->num_stripes != map->verified_stripes) {
8106 "chunk %llu has missing dev extent, have %d expect %d",
8107 map->start, map->verified_stripes, map->num_stripes);
8113 read_unlock(&fs_info->mapping_tree_lock);
8118 * Ensure that all dev extents are mapped to correct chunk, otherwise
8119 * later chunk allocation/free would cause unexpected behavior.
8121 * NOTE: This will iterate through the whole device tree, which should be of
8122 * the same size level as the chunk tree. This slightly increases mount time.
8124 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
8126 struct btrfs_path *path;
8127 struct btrfs_root *root = fs_info->dev_root;
8128 struct btrfs_key key;
8130 u64 prev_dev_ext_end = 0;
8134 * We don't have a dev_root because we mounted with ignorebadroots and
8135 * failed to load the root, so we want to skip the verification in this
8138 * However if the dev root is fine, but the tree itself is corrupted
8139 * we'd still fail to mount. This verification is only to make sure
8140 * writes can happen safely, so instead just bypass this check
8141 * completely in the case of IGNOREBADROOTS.
8143 if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
8147 key.type = BTRFS_DEV_EXTENT_KEY;
8150 path = btrfs_alloc_path();
8154 path->reada = READA_FORWARD;
8155 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8159 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
8160 ret = btrfs_next_leaf(root, path);
8163 /* No dev extents at all? Not good */
8170 struct extent_buffer *leaf = path->nodes[0];
8171 struct btrfs_dev_extent *dext;
8172 int slot = path->slots[0];
8174 u64 physical_offset;
8178 btrfs_item_key_to_cpu(leaf, &key, slot);
8179 if (key.type != BTRFS_DEV_EXTENT_KEY)
8181 devid = key.objectid;
8182 physical_offset = key.offset;
8184 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
8185 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
8186 physical_len = btrfs_dev_extent_length(leaf, dext);
8188 /* Check if this dev extent overlaps with the previous one */
8189 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
8191 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
8192 devid, physical_offset, prev_dev_ext_end);
8197 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
8198 physical_offset, physical_len);
8202 prev_dev_ext_end = physical_offset + physical_len;
8204 ret = btrfs_next_item(root, path);
8213 /* Ensure all chunks have corresponding dev extents */
8214 ret = verify_chunk_dev_extent_mapping(fs_info);
8216 btrfs_free_path(path);
8221 * Check whether the given block group or device is pinned by any inode being
8222 * used as a swapfile.
8224 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
8226 struct btrfs_swapfile_pin *sp;
8227 struct rb_node *node;
8229 spin_lock(&fs_info->swapfile_pins_lock);
8230 node = fs_info->swapfile_pins.rb_node;
8232 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
8234 node = node->rb_left;
8235 else if (ptr > sp->ptr)
8236 node = node->rb_right;
8240 spin_unlock(&fs_info->swapfile_pins_lock);
8241 return node != NULL;
8244 static int relocating_repair_kthread(void *data)
8246 struct btrfs_block_group *cache = data;
8247 struct btrfs_fs_info *fs_info = cache->fs_info;
8251 target = cache->start;
8252 btrfs_put_block_group(cache);
8254 sb_start_write(fs_info->sb);
8255 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
8257 "zoned: skip relocating block group %llu to repair: EBUSY",
8259 sb_end_write(fs_info->sb);
8263 mutex_lock(&fs_info->reclaim_bgs_lock);
8265 /* Ensure block group still exists */
8266 cache = btrfs_lookup_block_group(fs_info, target);
8270 if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags))
8273 ret = btrfs_may_alloc_data_chunk(fs_info, target);
8278 "zoned: relocating block group %llu to repair IO failure",
8280 ret = btrfs_relocate_chunk(fs_info, target);
8284 btrfs_put_block_group(cache);
8285 mutex_unlock(&fs_info->reclaim_bgs_lock);
8286 btrfs_exclop_finish(fs_info);
8287 sb_end_write(fs_info->sb);
8292 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical)
8294 struct btrfs_block_group *cache;
8296 if (!btrfs_is_zoned(fs_info))
8299 /* Do not attempt to repair in degraded state */
8300 if (btrfs_test_opt(fs_info, DEGRADED))
8303 cache = btrfs_lookup_block_group(fs_info, logical);
8307 if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) {
8308 btrfs_put_block_group(cache);
8312 kthread_run(relocating_repair_kthread, cache,
8313 "btrfs-relocating-repair");
8318 static void map_raid56_repair_block(struct btrfs_io_context *bioc,
8319 struct btrfs_io_stripe *smap,
8322 int data_stripes = nr_bioc_data_stripes(bioc);
8325 for (i = 0; i < data_stripes; i++) {
8326 u64 stripe_start = bioc->full_stripe_logical +
8327 btrfs_stripe_nr_to_offset(i);
8329 if (logical >= stripe_start &&
8330 logical < stripe_start + BTRFS_STRIPE_LEN)
8333 ASSERT(i < data_stripes);
8334 smap->dev = bioc->stripes[i].dev;
8335 smap->physical = bioc->stripes[i].physical +
8336 ((logical - bioc->full_stripe_logical) &
8337 BTRFS_STRIPE_LEN_MASK);
8341 * Map a repair write into a single device.
8343 * A repair write is triggered by read time repair or scrub, which would only
8344 * update the contents of a single device.
8345 * Not update any other mirrors nor go through RMW path.
8347 * Callers should ensure:
8349 * - Call btrfs_bio_counter_inc_blocked() first
8350 * - The range does not cross stripe boundary
8351 * - Has a valid @mirror_num passed in.
8353 int btrfs_map_repair_block(struct btrfs_fs_info *fs_info,
8354 struct btrfs_io_stripe *smap, u64 logical,
8355 u32 length, int mirror_num)
8357 struct btrfs_io_context *bioc = NULL;
8358 u64 map_length = length;
8359 int mirror_ret = mirror_num;
8362 ASSERT(mirror_num > 0);
8364 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, &map_length,
8365 &bioc, smap, &mirror_ret);
8369 /* The map range should not cross stripe boundary. */
8370 ASSERT(map_length >= length);
8372 /* Already mapped to single stripe. */
8376 /* Map the RAID56 multi-stripe writes to a single one. */
8377 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
8378 map_raid56_repair_block(bioc, smap, logical);
8382 ASSERT(mirror_num <= bioc->num_stripes);
8383 smap->dev = bioc->stripes[mirror_num - 1].dev;
8384 smap->physical = bioc->stripes[mirror_num - 1].physical;
8386 btrfs_put_bioc(bioc);