1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/sched/mm.h>
8 #include <linux/slab.h>
9 #include <linux/ratelimit.h>
10 #include <linux/kthread.h>
11 #include <linux/semaphore.h>
12 #include <linux/uuid.h>
13 #include <linux/list_sort.h>
14 #include <linux/namei.h>
17 #include "extent_map.h"
19 #include "transaction.h"
20 #include "print-tree.h"
23 #include "rcu-string.h"
24 #include "dev-replace.h"
26 #include "tree-checker.h"
27 #include "space-info.h"
28 #include "block-group.h"
32 #include "accessors.h"
33 #include "uuid-tree.h"
35 #include "relocation.h"
38 #include "raid-stripe-tree.h"
40 #define BTRFS_BLOCK_GROUP_STRIPE_MASK (BTRFS_BLOCK_GROUP_RAID0 | \
41 BTRFS_BLOCK_GROUP_RAID10 | \
42 BTRFS_BLOCK_GROUP_RAID56_MASK)
44 struct btrfs_io_geometry {
50 u64 raid56_full_stripe_start;
55 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
56 [BTRFS_RAID_RAID10] = {
59 .devs_max = 0, /* 0 == as many as possible */
61 .tolerated_failures = 1,
65 .raid_name = "raid10",
66 .bg_flag = BTRFS_BLOCK_GROUP_RAID10,
67 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
69 [BTRFS_RAID_RAID1] = {
74 .tolerated_failures = 1,
79 .bg_flag = BTRFS_BLOCK_GROUP_RAID1,
80 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
82 [BTRFS_RAID_RAID1C3] = {
87 .tolerated_failures = 2,
91 .raid_name = "raid1c3",
92 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3,
93 .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET,
95 [BTRFS_RAID_RAID1C4] = {
100 .tolerated_failures = 3,
104 .raid_name = "raid1c4",
105 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4,
106 .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET,
113 .tolerated_failures = 0,
118 .bg_flag = BTRFS_BLOCK_GROUP_DUP,
121 [BTRFS_RAID_RAID0] = {
126 .tolerated_failures = 0,
130 .raid_name = "raid0",
131 .bg_flag = BTRFS_BLOCK_GROUP_RAID0,
134 [BTRFS_RAID_SINGLE] = {
139 .tolerated_failures = 0,
143 .raid_name = "single",
147 [BTRFS_RAID_RAID5] = {
152 .tolerated_failures = 1,
156 .raid_name = "raid5",
157 .bg_flag = BTRFS_BLOCK_GROUP_RAID5,
158 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
160 [BTRFS_RAID_RAID6] = {
165 .tolerated_failures = 2,
169 .raid_name = "raid6",
170 .bg_flag = BTRFS_BLOCK_GROUP_RAID6,
171 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
176 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which
177 * can be used as index to access btrfs_raid_array[].
179 enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags)
181 const u64 profile = (flags & BTRFS_BLOCK_GROUP_PROFILE_MASK);
184 return BTRFS_RAID_SINGLE;
186 return BTRFS_BG_FLAG_TO_INDEX(profile);
189 const char *btrfs_bg_type_to_raid_name(u64 flags)
191 const int index = btrfs_bg_flags_to_raid_index(flags);
193 if (index >= BTRFS_NR_RAID_TYPES)
196 return btrfs_raid_array[index].raid_name;
199 int btrfs_nr_parity_stripes(u64 type)
201 enum btrfs_raid_types index = btrfs_bg_flags_to_raid_index(type);
203 return btrfs_raid_array[index].nparity;
207 * Fill @buf with textual description of @bg_flags, no more than @size_buf
208 * bytes including terminating null byte.
210 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf)
215 u64 flags = bg_flags;
216 u32 size_bp = size_buf;
223 #define DESCRIBE_FLAG(flag, desc) \
225 if (flags & (flag)) { \
226 ret = snprintf(bp, size_bp, "%s|", (desc)); \
227 if (ret < 0 || ret >= size_bp) \
235 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data");
236 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system");
237 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata");
239 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single");
240 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
241 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag,
242 btrfs_raid_array[i].raid_name);
246 ret = snprintf(bp, size_bp, "0x%llx|", flags);
250 if (size_bp < size_buf)
251 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */
254 * The text is trimmed, it's up to the caller to provide sufficiently
260 static int init_first_rw_device(struct btrfs_trans_handle *trans);
261 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
262 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
268 * There are several mutexes that protect manipulation of devices and low-level
269 * structures like chunks but not block groups, extents or files
271 * uuid_mutex (global lock)
272 * ------------------------
273 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
274 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
275 * device) or requested by the device= mount option
277 * the mutex can be very coarse and can cover long-running operations
279 * protects: updates to fs_devices counters like missing devices, rw devices,
280 * seeding, structure cloning, opening/closing devices at mount/umount time
282 * global::fs_devs - add, remove, updates to the global list
284 * does not protect: manipulation of the fs_devices::devices list in general
285 * but in mount context it could be used to exclude list modifications by eg.
288 * btrfs_device::name - renames (write side), read is RCU
290 * fs_devices::device_list_mutex (per-fs, with RCU)
291 * ------------------------------------------------
292 * protects updates to fs_devices::devices, ie. adding and deleting
294 * simple list traversal with read-only actions can be done with RCU protection
296 * may be used to exclude some operations from running concurrently without any
297 * modifications to the list (see write_all_supers)
299 * Is not required at mount and close times, because our device list is
300 * protected by the uuid_mutex at that point.
304 * protects balance structures (status, state) and context accessed from
305 * several places (internally, ioctl)
309 * protects chunks, adding or removing during allocation, trim or when a new
310 * device is added/removed. Additionally it also protects post_commit_list of
311 * individual devices, since they can be added to the transaction's
312 * post_commit_list only with chunk_mutex held.
316 * a big lock that is held by the cleaner thread and prevents running subvolume
317 * cleaning together with relocation or delayed iputs
329 * Exclusive operations
330 * ====================
332 * Maintains the exclusivity of the following operations that apply to the
333 * whole filesystem and cannot run in parallel.
338 * - Device replace (*)
341 * The device operations (as above) can be in one of the following states:
347 * Only device operations marked with (*) can go into the Paused state for the
350 * - ioctl (only Balance can be Paused through ioctl)
351 * - filesystem remounted as read-only
352 * - filesystem unmounted and mounted as read-only
353 * - system power-cycle and filesystem mounted as read-only
354 * - filesystem or device errors leading to forced read-only
356 * The status of exclusive operation is set and cleared atomically.
357 * During the course of Paused state, fs_info::exclusive_operation remains set.
358 * A device operation in Paused or Running state can be canceled or resumed
359 * either by ioctl (Balance only) or when remounted as read-write.
360 * The exclusive status is cleared when the device operation is canceled or
364 DEFINE_MUTEX(uuid_mutex);
365 static LIST_HEAD(fs_uuids);
366 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void)
372 * Allocate new btrfs_fs_devices structure identified by a fsid.
374 * @fsid: if not NULL, copy the UUID to fs_devices::fsid and to
375 * fs_devices::metadata_fsid
377 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
378 * The returned struct is not linked onto any lists and can be destroyed with
379 * kfree() right away.
381 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
383 struct btrfs_fs_devices *fs_devs;
385 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
387 return ERR_PTR(-ENOMEM);
389 mutex_init(&fs_devs->device_list_mutex);
391 INIT_LIST_HEAD(&fs_devs->devices);
392 INIT_LIST_HEAD(&fs_devs->alloc_list);
393 INIT_LIST_HEAD(&fs_devs->fs_list);
394 INIT_LIST_HEAD(&fs_devs->seed_list);
397 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
398 memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE);
404 static void btrfs_free_device(struct btrfs_device *device)
406 WARN_ON(!list_empty(&device->post_commit_list));
407 rcu_string_free(device->name);
408 extent_io_tree_release(&device->alloc_state);
409 btrfs_destroy_dev_zone_info(device);
413 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
415 struct btrfs_device *device;
417 WARN_ON(fs_devices->opened);
418 while (!list_empty(&fs_devices->devices)) {
419 device = list_entry(fs_devices->devices.next,
420 struct btrfs_device, dev_list);
421 list_del(&device->dev_list);
422 btrfs_free_device(device);
427 void __exit btrfs_cleanup_fs_uuids(void)
429 struct btrfs_fs_devices *fs_devices;
431 while (!list_empty(&fs_uuids)) {
432 fs_devices = list_entry(fs_uuids.next,
433 struct btrfs_fs_devices, fs_list);
434 list_del(&fs_devices->fs_list);
435 free_fs_devices(fs_devices);
439 static bool match_fsid_fs_devices(const struct btrfs_fs_devices *fs_devices,
440 const u8 *fsid, const u8 *metadata_fsid)
442 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) != 0)
448 if (memcmp(metadata_fsid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE) != 0)
454 static noinline struct btrfs_fs_devices *find_fsid(
455 const u8 *fsid, const u8 *metadata_fsid)
457 struct btrfs_fs_devices *fs_devices;
461 /* Handle non-split brain cases */
462 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
463 if (match_fsid_fs_devices(fs_devices, fsid, metadata_fsid))
470 btrfs_get_bdev_and_sb(const char *device_path, blk_mode_t flags, void *holder,
471 int flush, struct bdev_handle **bdev_handle,
472 struct btrfs_super_block **disk_super)
474 struct block_device *bdev;
477 *bdev_handle = bdev_open_by_path(device_path, flags, holder, NULL);
479 if (IS_ERR(*bdev_handle)) {
480 ret = PTR_ERR(*bdev_handle);
483 bdev = (*bdev_handle)->bdev;
487 ret = set_blocksize(bdev, BTRFS_BDEV_BLOCKSIZE);
489 bdev_release(*bdev_handle);
492 invalidate_bdev(bdev);
493 *disk_super = btrfs_read_dev_super(bdev);
494 if (IS_ERR(*disk_super)) {
495 ret = PTR_ERR(*disk_super);
496 bdev_release(*bdev_handle);
508 * Search and remove all stale devices (which are not mounted). When both
509 * inputs are NULL, it will search and release all stale devices.
511 * @devt: Optional. When provided will it release all unmounted devices
512 * matching this devt only.
513 * @skip_device: Optional. Will skip this device when searching for the stale
516 * Return: 0 for success or if @devt is 0.
517 * -EBUSY if @devt is a mounted device.
518 * -ENOENT if @devt does not match any device in the list.
520 static int btrfs_free_stale_devices(dev_t devt, struct btrfs_device *skip_device)
522 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
523 struct btrfs_device *device, *tmp_device;
527 lockdep_assert_held(&uuid_mutex);
529 /* Return good status if there is no instance of devt. */
531 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
533 mutex_lock(&fs_devices->device_list_mutex);
534 list_for_each_entry_safe(device, tmp_device,
535 &fs_devices->devices, dev_list) {
536 if (skip_device && skip_device == device)
538 if (devt && devt != device->devt)
540 if (fs_devices->opened) {
546 /* delete the stale device */
547 fs_devices->num_devices--;
548 list_del(&device->dev_list);
549 btrfs_free_device(device);
553 mutex_unlock(&fs_devices->device_list_mutex);
555 if (fs_devices->num_devices == 0) {
556 btrfs_sysfs_remove_fsid(fs_devices);
557 list_del(&fs_devices->fs_list);
558 free_fs_devices(fs_devices);
562 /* If there is at least one freed device return 0. */
569 static struct btrfs_fs_devices *find_fsid_by_device(
570 struct btrfs_super_block *disk_super,
571 dev_t devt, bool *same_fsid_diff_dev)
573 struct btrfs_fs_devices *fsid_fs_devices;
574 struct btrfs_fs_devices *devt_fs_devices;
575 const bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
576 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
577 bool found_by_devt = false;
579 /* Find the fs_device by the usual method, if found use it. */
580 fsid_fs_devices = find_fsid(disk_super->fsid,
581 has_metadata_uuid ? disk_super->metadata_uuid : NULL);
583 /* The temp_fsid feature is supported only with single device filesystem. */
584 if (btrfs_super_num_devices(disk_super) != 1)
585 return fsid_fs_devices;
588 * A seed device is an integral component of the sprout device, which
589 * functions as a multi-device filesystem. So, temp-fsid feature is
592 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING)
593 return fsid_fs_devices;
595 /* Try to find a fs_devices by matching devt. */
596 list_for_each_entry(devt_fs_devices, &fs_uuids, fs_list) {
597 struct btrfs_device *device;
599 list_for_each_entry(device, &devt_fs_devices->devices, dev_list) {
600 if (device->devt == devt) {
601 found_by_devt = true;
610 /* Existing device. */
611 if (fsid_fs_devices == NULL) {
612 if (devt_fs_devices->opened == 0) {
616 /* temp_fsid is mounting a subvol. */
617 return devt_fs_devices;
620 /* Regular or temp_fsid device mounting a subvol. */
621 return devt_fs_devices;
625 if (fsid_fs_devices == NULL) {
628 /* sb::fsid is already used create a new temp_fsid. */
629 *same_fsid_diff_dev = true;
638 * This is only used on mount, and we are protected from competing things
639 * messing with our fs_devices by the uuid_mutex, thus we do not need the
640 * fs_devices->device_list_mutex here.
642 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
643 struct btrfs_device *device, blk_mode_t flags,
646 struct bdev_handle *bdev_handle;
647 struct btrfs_super_block *disk_super;
656 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
657 &bdev_handle, &disk_super);
661 devid = btrfs_stack_device_id(&disk_super->dev_item);
662 if (devid != device->devid)
663 goto error_free_page;
665 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
666 goto error_free_page;
668 device->generation = btrfs_super_generation(disk_super);
670 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
671 if (btrfs_super_incompat_flags(disk_super) &
672 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
674 "BTRFS: Invalid seeding and uuid-changed device detected\n");
675 goto error_free_page;
678 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
679 fs_devices->seeding = true;
681 if (bdev_read_only(bdev_handle->bdev))
682 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
684 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
687 if (!bdev_nonrot(bdev_handle->bdev))
688 fs_devices->rotating = true;
690 if (bdev_max_discard_sectors(bdev_handle->bdev))
691 fs_devices->discardable = true;
693 device->bdev_handle = bdev_handle;
694 device->bdev = bdev_handle->bdev;
695 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
697 fs_devices->open_devices++;
698 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
699 device->devid != BTRFS_DEV_REPLACE_DEVID) {
700 fs_devices->rw_devices++;
701 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
703 btrfs_release_disk_super(disk_super);
708 btrfs_release_disk_super(disk_super);
709 bdev_release(bdev_handle);
714 u8 *btrfs_sb_fsid_ptr(struct btrfs_super_block *sb)
716 bool has_metadata_uuid = (btrfs_super_incompat_flags(sb) &
717 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
719 return has_metadata_uuid ? sb->metadata_uuid : sb->fsid;
723 * Add new device to list of registered devices
726 * device pointer which was just added or updated when successful
727 * error pointer when failed
729 static noinline struct btrfs_device *device_list_add(const char *path,
730 struct btrfs_super_block *disk_super,
731 bool *new_device_added)
733 struct btrfs_device *device;
734 struct btrfs_fs_devices *fs_devices = NULL;
735 struct rcu_string *name;
736 u64 found_transid = btrfs_super_generation(disk_super);
737 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
740 bool same_fsid_diff_dev = false;
741 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
742 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
744 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
746 "device %s has incomplete metadata_uuid change, please use btrfstune to complete",
748 return ERR_PTR(-EAGAIN);
751 error = lookup_bdev(path, &path_devt);
753 btrfs_err(NULL, "failed to lookup block device for path %s: %d",
755 return ERR_PTR(error);
758 fs_devices = find_fsid_by_device(disk_super, path_devt, &same_fsid_diff_dev);
761 fs_devices = alloc_fs_devices(disk_super->fsid);
762 if (IS_ERR(fs_devices))
763 return ERR_CAST(fs_devices);
765 if (has_metadata_uuid)
766 memcpy(fs_devices->metadata_uuid,
767 disk_super->metadata_uuid, BTRFS_FSID_SIZE);
769 if (same_fsid_diff_dev) {
770 generate_random_uuid(fs_devices->fsid);
771 fs_devices->temp_fsid = true;
772 pr_info("BTRFS: device %s using temp-fsid %pU\n",
773 path, fs_devices->fsid);
776 mutex_lock(&fs_devices->device_list_mutex);
777 list_add(&fs_devices->fs_list, &fs_uuids);
781 struct btrfs_dev_lookup_args args = {
783 .uuid = disk_super->dev_item.uuid,
786 mutex_lock(&fs_devices->device_list_mutex);
787 device = btrfs_find_device(fs_devices, &args);
789 if (found_transid > fs_devices->latest_generation) {
790 memcpy(fs_devices->fsid, disk_super->fsid,
792 memcpy(fs_devices->metadata_uuid,
793 btrfs_sb_fsid_ptr(disk_super), BTRFS_FSID_SIZE);
798 unsigned int nofs_flag;
800 if (fs_devices->opened) {
802 "device %s belongs to fsid %pU, and the fs is already mounted, scanned by %s (%d)",
803 path, fs_devices->fsid, current->comm,
804 task_pid_nr(current));
805 mutex_unlock(&fs_devices->device_list_mutex);
806 return ERR_PTR(-EBUSY);
809 nofs_flag = memalloc_nofs_save();
810 device = btrfs_alloc_device(NULL, &devid,
811 disk_super->dev_item.uuid, path);
812 memalloc_nofs_restore(nofs_flag);
813 if (IS_ERR(device)) {
814 mutex_unlock(&fs_devices->device_list_mutex);
815 /* we can safely leave the fs_devices entry around */
819 device->devt = path_devt;
821 list_add_rcu(&device->dev_list, &fs_devices->devices);
822 fs_devices->num_devices++;
824 device->fs_devices = fs_devices;
825 *new_device_added = true;
827 if (disk_super->label[0])
829 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n",
830 disk_super->label, devid, found_transid, path,
831 current->comm, task_pid_nr(current));
834 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n",
835 disk_super->fsid, devid, found_transid, path,
836 current->comm, task_pid_nr(current));
838 } else if (!device->name || strcmp(device->name->str, path)) {
840 * When FS is already mounted.
841 * 1. If you are here and if the device->name is NULL that
842 * means this device was missing at time of FS mount.
843 * 2. If you are here and if the device->name is different
844 * from 'path' that means either
845 * a. The same device disappeared and reappeared with
847 * b. The missing-disk-which-was-replaced, has
850 * We must allow 1 and 2a above. But 2b would be a spurious
853 * Further in case of 1 and 2a above, the disk at 'path'
854 * would have missed some transaction when it was away and
855 * in case of 2a the stale bdev has to be updated as well.
856 * 2b must not be allowed at all time.
860 * For now, we do allow update to btrfs_fs_device through the
861 * btrfs dev scan cli after FS has been mounted. We're still
862 * tracking a problem where systems fail mount by subvolume id
863 * when we reject replacement on a mounted FS.
865 if (!fs_devices->opened && found_transid < device->generation) {
867 * That is if the FS is _not_ mounted and if you
868 * are here, that means there is more than one
869 * disk with same uuid and devid.We keep the one
870 * with larger generation number or the last-in if
871 * generation are equal.
873 mutex_unlock(&fs_devices->device_list_mutex);
875 "device %s already registered with a higher generation, found %llu expect %llu",
876 path, found_transid, device->generation);
877 return ERR_PTR(-EEXIST);
881 * We are going to replace the device path for a given devid,
882 * make sure it's the same device if the device is mounted
884 * NOTE: the device->fs_info may not be reliable here so pass
885 * in a NULL to message helpers instead. This avoids a possible
886 * use-after-free when the fs_info and fs_info->sb are already
890 if (device->devt != path_devt) {
891 mutex_unlock(&fs_devices->device_list_mutex);
892 btrfs_warn_in_rcu(NULL,
893 "duplicate device %s devid %llu generation %llu scanned by %s (%d)",
894 path, devid, found_transid,
896 task_pid_nr(current));
897 return ERR_PTR(-EEXIST);
899 btrfs_info_in_rcu(NULL,
900 "devid %llu device path %s changed to %s scanned by %s (%d)",
901 devid, btrfs_dev_name(device),
903 task_pid_nr(current));
906 name = rcu_string_strdup(path, GFP_NOFS);
908 mutex_unlock(&fs_devices->device_list_mutex);
909 return ERR_PTR(-ENOMEM);
911 rcu_string_free(device->name);
912 rcu_assign_pointer(device->name, name);
913 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
914 fs_devices->missing_devices--;
915 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
917 device->devt = path_devt;
921 * Unmount does not free the btrfs_device struct but would zero
922 * generation along with most of the other members. So just update
923 * it back. We need it to pick the disk with largest generation
926 if (!fs_devices->opened) {
927 device->generation = found_transid;
928 fs_devices->latest_generation = max_t(u64, found_transid,
929 fs_devices->latest_generation);
932 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
934 mutex_unlock(&fs_devices->device_list_mutex);
938 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
940 struct btrfs_fs_devices *fs_devices;
941 struct btrfs_device *device;
942 struct btrfs_device *orig_dev;
945 lockdep_assert_held(&uuid_mutex);
947 fs_devices = alloc_fs_devices(orig->fsid);
948 if (IS_ERR(fs_devices))
951 fs_devices->total_devices = orig->total_devices;
953 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
954 const char *dev_path = NULL;
957 * This is ok to do without RCU read locked because we hold the
958 * uuid mutex so nothing we touch in here is going to disappear.
961 dev_path = orig_dev->name->str;
963 device = btrfs_alloc_device(NULL, &orig_dev->devid,
964 orig_dev->uuid, dev_path);
965 if (IS_ERR(device)) {
966 ret = PTR_ERR(device);
970 if (orig_dev->zone_info) {
971 struct btrfs_zoned_device_info *zone_info;
973 zone_info = btrfs_clone_dev_zone_info(orig_dev);
975 btrfs_free_device(device);
979 device->zone_info = zone_info;
982 list_add(&device->dev_list, &fs_devices->devices);
983 device->fs_devices = fs_devices;
984 fs_devices->num_devices++;
988 free_fs_devices(fs_devices);
992 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices,
993 struct btrfs_device **latest_dev)
995 struct btrfs_device *device, *next;
997 /* This is the initialized path, it is safe to release the devices. */
998 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
999 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) {
1000 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1001 &device->dev_state) &&
1002 !test_bit(BTRFS_DEV_STATE_MISSING,
1003 &device->dev_state) &&
1005 device->generation > (*latest_dev)->generation)) {
1006 *latest_dev = device;
1012 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID,
1013 * in btrfs_init_dev_replace() so just continue.
1015 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1018 if (device->bdev_handle) {
1019 bdev_release(device->bdev_handle);
1020 device->bdev = NULL;
1021 device->bdev_handle = NULL;
1022 fs_devices->open_devices--;
1024 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1025 list_del_init(&device->dev_alloc_list);
1026 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1027 fs_devices->rw_devices--;
1029 list_del_init(&device->dev_list);
1030 fs_devices->num_devices--;
1031 btrfs_free_device(device);
1037 * After we have read the system tree and know devids belonging to this
1038 * filesystem, remove the device which does not belong there.
1040 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices)
1042 struct btrfs_device *latest_dev = NULL;
1043 struct btrfs_fs_devices *seed_dev;
1045 mutex_lock(&uuid_mutex);
1046 __btrfs_free_extra_devids(fs_devices, &latest_dev);
1048 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list)
1049 __btrfs_free_extra_devids(seed_dev, &latest_dev);
1051 fs_devices->latest_dev = latest_dev;
1053 mutex_unlock(&uuid_mutex);
1056 static void btrfs_close_bdev(struct btrfs_device *device)
1061 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1062 sync_blockdev(device->bdev);
1063 invalidate_bdev(device->bdev);
1066 bdev_release(device->bdev_handle);
1069 static void btrfs_close_one_device(struct btrfs_device *device)
1071 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1073 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1074 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1075 list_del_init(&device->dev_alloc_list);
1076 fs_devices->rw_devices--;
1079 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1080 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
1082 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
1083 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
1084 fs_devices->missing_devices--;
1087 btrfs_close_bdev(device);
1089 fs_devices->open_devices--;
1090 device->bdev = NULL;
1092 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1093 btrfs_destroy_dev_zone_info(device);
1095 device->fs_info = NULL;
1096 atomic_set(&device->dev_stats_ccnt, 0);
1097 extent_io_tree_release(&device->alloc_state);
1100 * Reset the flush error record. We might have a transient flush error
1101 * in this mount, and if so we aborted the current transaction and set
1102 * the fs to an error state, guaranteeing no super blocks can be further
1103 * committed. However that error might be transient and if we unmount the
1104 * filesystem and mount it again, we should allow the mount to succeed
1105 * (btrfs_check_rw_degradable() should not fail) - if after mounting the
1106 * filesystem again we still get flush errors, then we will again abort
1107 * any transaction and set the error state, guaranteeing no commits of
1108 * unsafe super blocks.
1110 device->last_flush_error = 0;
1112 /* Verify the device is back in a pristine state */
1113 WARN_ON(test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1114 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1115 WARN_ON(!list_empty(&device->dev_alloc_list));
1116 WARN_ON(!list_empty(&device->post_commit_list));
1119 static void close_fs_devices(struct btrfs_fs_devices *fs_devices)
1121 struct btrfs_device *device, *tmp;
1123 lockdep_assert_held(&uuid_mutex);
1125 if (--fs_devices->opened > 0)
1128 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list)
1129 btrfs_close_one_device(device);
1131 WARN_ON(fs_devices->open_devices);
1132 WARN_ON(fs_devices->rw_devices);
1133 fs_devices->opened = 0;
1134 fs_devices->seeding = false;
1135 fs_devices->fs_info = NULL;
1138 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1141 struct btrfs_fs_devices *tmp;
1143 mutex_lock(&uuid_mutex);
1144 close_fs_devices(fs_devices);
1145 if (!fs_devices->opened) {
1146 list_splice_init(&fs_devices->seed_list, &list);
1149 * If the struct btrfs_fs_devices is not assembled with any
1150 * other device, it can be re-initialized during the next mount
1151 * without the needing device-scan step. Therefore, it can be
1154 if (fs_devices->num_devices == 1) {
1155 list_del(&fs_devices->fs_list);
1156 free_fs_devices(fs_devices);
1161 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {
1162 close_fs_devices(fs_devices);
1163 list_del(&fs_devices->seed_list);
1164 free_fs_devices(fs_devices);
1166 mutex_unlock(&uuid_mutex);
1169 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1170 blk_mode_t flags, void *holder)
1172 struct btrfs_device *device;
1173 struct btrfs_device *latest_dev = NULL;
1174 struct btrfs_device *tmp_device;
1176 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,
1180 ret = btrfs_open_one_device(fs_devices, device, flags, holder);
1182 (!latest_dev || device->generation > latest_dev->generation)) {
1183 latest_dev = device;
1184 } else if (ret == -ENODATA) {
1185 fs_devices->num_devices--;
1186 list_del(&device->dev_list);
1187 btrfs_free_device(device);
1190 if (fs_devices->open_devices == 0)
1193 fs_devices->opened = 1;
1194 fs_devices->latest_dev = latest_dev;
1195 fs_devices->total_rw_bytes = 0;
1196 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1197 fs_devices->read_policy = BTRFS_READ_POLICY_PID;
1202 static int devid_cmp(void *priv, const struct list_head *a,
1203 const struct list_head *b)
1205 const struct btrfs_device *dev1, *dev2;
1207 dev1 = list_entry(a, struct btrfs_device, dev_list);
1208 dev2 = list_entry(b, struct btrfs_device, dev_list);
1210 if (dev1->devid < dev2->devid)
1212 else if (dev1->devid > dev2->devid)
1217 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1218 blk_mode_t flags, void *holder)
1222 lockdep_assert_held(&uuid_mutex);
1224 * The device_list_mutex cannot be taken here in case opening the
1225 * underlying device takes further locks like open_mutex.
1227 * We also don't need the lock here as this is called during mount and
1228 * exclusion is provided by uuid_mutex
1231 if (fs_devices->opened) {
1232 fs_devices->opened++;
1235 list_sort(NULL, &fs_devices->devices, devid_cmp);
1236 ret = open_fs_devices(fs_devices, flags, holder);
1242 void btrfs_release_disk_super(struct btrfs_super_block *super)
1244 struct page *page = virt_to_page(super);
1249 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1250 u64 bytenr, u64 bytenr_orig)
1252 struct btrfs_super_block *disk_super;
1257 /* make sure our super fits in the device */
1258 if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev))
1259 return ERR_PTR(-EINVAL);
1261 /* make sure our super fits in the page */
1262 if (sizeof(*disk_super) > PAGE_SIZE)
1263 return ERR_PTR(-EINVAL);
1265 /* make sure our super doesn't straddle pages on disk */
1266 index = bytenr >> PAGE_SHIFT;
1267 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1268 return ERR_PTR(-EINVAL);
1270 /* pull in the page with our super */
1271 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1274 return ERR_CAST(page);
1276 p = page_address(page);
1278 /* align our pointer to the offset of the super block */
1279 disk_super = p + offset_in_page(bytenr);
1281 if (btrfs_super_bytenr(disk_super) != bytenr_orig ||
1282 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1283 btrfs_release_disk_super(p);
1284 return ERR_PTR(-EINVAL);
1287 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1288 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1293 int btrfs_forget_devices(dev_t devt)
1297 mutex_lock(&uuid_mutex);
1298 ret = btrfs_free_stale_devices(devt, NULL);
1299 mutex_unlock(&uuid_mutex);
1305 * Look for a btrfs signature on a device. This may be called out of the mount path
1306 * and we are not allowed to call set_blocksize during the scan. The superblock
1307 * is read via pagecache.
1309 * With @mount_arg_dev it's a scan during mount time that will always register
1310 * the device or return an error. Multi-device and seeding devices are registered
1313 struct btrfs_device *btrfs_scan_one_device(const char *path, blk_mode_t flags,
1316 struct btrfs_super_block *disk_super;
1317 bool new_device_added = false;
1318 struct btrfs_device *device = NULL;
1319 struct bdev_handle *bdev_handle;
1320 u64 bytenr, bytenr_orig;
1323 lockdep_assert_held(&uuid_mutex);
1326 * we would like to check all the supers, but that would make
1327 * a btrfs mount succeed after a mkfs from a different FS.
1328 * So, we need to add a special mount option to scan for
1329 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1333 * Avoid an exclusive open here, as the systemd-udev may initiate the
1334 * device scan which may race with the user's mount or mkfs command,
1335 * resulting in failure.
1336 * Since the device scan is solely for reading purposes, there is no
1337 * need for an exclusive open. Additionally, the devices are read again
1338 * during the mount process. It is ok to get some inconsistent
1339 * values temporarily, as the device paths of the fsid are the only
1340 * required information for assembling the volume.
1342 bdev_handle = bdev_open_by_path(path, flags, NULL, NULL);
1343 if (IS_ERR(bdev_handle))
1344 return ERR_CAST(bdev_handle);
1346 bytenr_orig = btrfs_sb_offset(0);
1347 ret = btrfs_sb_log_location_bdev(bdev_handle->bdev, 0, READ, &bytenr);
1349 device = ERR_PTR(ret);
1350 goto error_bdev_put;
1353 disk_super = btrfs_read_disk_super(bdev_handle->bdev, bytenr,
1355 if (IS_ERR(disk_super)) {
1356 device = ERR_CAST(disk_super);
1357 goto error_bdev_put;
1360 if (!mount_arg_dev && btrfs_super_num_devices(disk_super) == 1 &&
1361 !(btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING)) {
1364 ret = lookup_bdev(path, &devt);
1366 btrfs_warn(NULL, "lookup bdev failed for path %s: %d",
1369 btrfs_free_stale_devices(devt, NULL);
1371 pr_debug("BTRFS: skip registering single non-seed device %s\n", path);
1373 goto free_disk_super;
1376 device = device_list_add(path, disk_super, &new_device_added);
1377 if (!IS_ERR(device) && new_device_added)
1378 btrfs_free_stale_devices(device->devt, device);
1381 btrfs_release_disk_super(disk_super);
1384 bdev_release(bdev_handle);
1390 * Try to find a chunk that intersects [start, start + len] range and when one
1391 * such is found, record the end of it in *start
1393 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1396 u64 physical_start, physical_end;
1398 lockdep_assert_held(&device->fs_info->chunk_mutex);
1400 if (find_first_extent_bit(&device->alloc_state, *start,
1401 &physical_start, &physical_end,
1402 CHUNK_ALLOCATED, NULL)) {
1404 if (in_range(physical_start, *start, len) ||
1405 in_range(*start, physical_start,
1406 physical_end - physical_start)) {
1407 *start = physical_end + 1;
1414 static u64 dev_extent_search_start(struct btrfs_device *device)
1416 switch (device->fs_devices->chunk_alloc_policy) {
1417 case BTRFS_CHUNK_ALLOC_REGULAR:
1418 return BTRFS_DEVICE_RANGE_RESERVED;
1419 case BTRFS_CHUNK_ALLOC_ZONED:
1421 * We don't care about the starting region like regular
1422 * allocator, because we anyway use/reserve the first two zones
1423 * for superblock logging.
1431 static bool dev_extent_hole_check_zoned(struct btrfs_device *device,
1432 u64 *hole_start, u64 *hole_size,
1435 u64 zone_size = device->zone_info->zone_size;
1438 bool changed = false;
1440 ASSERT(IS_ALIGNED(*hole_start, zone_size));
1442 while (*hole_size > 0) {
1443 pos = btrfs_find_allocatable_zones(device, *hole_start,
1444 *hole_start + *hole_size,
1446 if (pos != *hole_start) {
1447 *hole_size = *hole_start + *hole_size - pos;
1450 if (*hole_size < num_bytes)
1454 ret = btrfs_ensure_empty_zones(device, pos, num_bytes);
1456 /* Range is ensured to be empty */
1460 /* Given hole range was invalid (outside of device) */
1461 if (ret == -ERANGE) {
1462 *hole_start += *hole_size;
1467 *hole_start += zone_size;
1468 *hole_size -= zone_size;
1476 * Check if specified hole is suitable for allocation.
1478 * @device: the device which we have the hole
1479 * @hole_start: starting position of the hole
1480 * @hole_size: the size of the hole
1481 * @num_bytes: the size of the free space that we need
1483 * This function may modify @hole_start and @hole_size to reflect the suitable
1484 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1486 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1487 u64 *hole_size, u64 num_bytes)
1489 bool changed = false;
1490 u64 hole_end = *hole_start + *hole_size;
1494 * Check before we set max_hole_start, otherwise we could end up
1495 * sending back this offset anyway.
1497 if (contains_pending_extent(device, hole_start, *hole_size)) {
1498 if (hole_end >= *hole_start)
1499 *hole_size = hole_end - *hole_start;
1505 switch (device->fs_devices->chunk_alloc_policy) {
1506 case BTRFS_CHUNK_ALLOC_REGULAR:
1507 /* No extra check */
1509 case BTRFS_CHUNK_ALLOC_ZONED:
1510 if (dev_extent_hole_check_zoned(device, hole_start,
1511 hole_size, num_bytes)) {
1514 * The changed hole can contain pending extent.
1515 * Loop again to check that.
1531 * Find free space in the specified device.
1533 * @device: the device which we search the free space in
1534 * @num_bytes: the size of the free space that we need
1535 * @search_start: the position from which to begin the search
1536 * @start: store the start of the free space.
1537 * @len: the size of the free space. that we find, or the size
1538 * of the max free space if we don't find suitable free space
1540 * This does a pretty simple search, the expectation is that it is called very
1541 * infrequently and that a given device has a small number of extents.
1543 * @start is used to store the start of the free space if we find. But if we
1544 * don't find suitable free space, it will be used to store the start position
1545 * of the max free space.
1547 * @len is used to store the size of the free space that we find.
1548 * But if we don't find suitable free space, it is used to store the size of
1549 * the max free space.
1551 * NOTE: This function will search *commit* root of device tree, and does extra
1552 * check to ensure dev extents are not double allocated.
1553 * This makes the function safe to allocate dev extents but may not report
1554 * correct usable device space, as device extent freed in current transaction
1555 * is not reported as available.
1557 static int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1558 u64 *start, u64 *len)
1560 struct btrfs_fs_info *fs_info = device->fs_info;
1561 struct btrfs_root *root = fs_info->dev_root;
1562 struct btrfs_key key;
1563 struct btrfs_dev_extent *dev_extent;
1564 struct btrfs_path *path;
1568 u64 max_hole_size = 0;
1570 u64 search_end = device->total_bytes;
1573 struct extent_buffer *l;
1575 search_start = dev_extent_search_start(device);
1576 max_hole_start = search_start;
1578 WARN_ON(device->zone_info &&
1579 !IS_ALIGNED(num_bytes, device->zone_info->zone_size));
1581 path = btrfs_alloc_path();
1587 if (search_start >= search_end ||
1588 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1593 path->reada = READA_FORWARD;
1594 path->search_commit_root = 1;
1595 path->skip_locking = 1;
1597 key.objectid = device->devid;
1598 key.offset = search_start;
1599 key.type = BTRFS_DEV_EXTENT_KEY;
1601 ret = btrfs_search_backwards(root, &key, path);
1605 while (search_start < search_end) {
1607 slot = path->slots[0];
1608 if (slot >= btrfs_header_nritems(l)) {
1609 ret = btrfs_next_leaf(root, path);
1617 btrfs_item_key_to_cpu(l, &key, slot);
1619 if (key.objectid < device->devid)
1622 if (key.objectid > device->devid)
1625 if (key.type != BTRFS_DEV_EXTENT_KEY)
1628 if (key.offset > search_end)
1631 if (key.offset > search_start) {
1632 hole_size = key.offset - search_start;
1633 dev_extent_hole_check(device, &search_start, &hole_size,
1636 if (hole_size > max_hole_size) {
1637 max_hole_start = search_start;
1638 max_hole_size = hole_size;
1642 * If this free space is greater than which we need,
1643 * it must be the max free space that we have found
1644 * until now, so max_hole_start must point to the start
1645 * of this free space and the length of this free space
1646 * is stored in max_hole_size. Thus, we return
1647 * max_hole_start and max_hole_size and go back to the
1650 if (hole_size >= num_bytes) {
1656 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1657 extent_end = key.offset + btrfs_dev_extent_length(l,
1659 if (extent_end > search_start)
1660 search_start = extent_end;
1667 * At this point, search_start should be the end of
1668 * allocated dev extents, and when shrinking the device,
1669 * search_end may be smaller than search_start.
1671 if (search_end > search_start) {
1672 hole_size = search_end - search_start;
1673 if (dev_extent_hole_check(device, &search_start, &hole_size,
1675 btrfs_release_path(path);
1679 if (hole_size > max_hole_size) {
1680 max_hole_start = search_start;
1681 max_hole_size = hole_size;
1686 if (max_hole_size < num_bytes)
1691 ASSERT(max_hole_start + max_hole_size <= search_end);
1693 btrfs_free_path(path);
1694 *start = max_hole_start;
1696 *len = max_hole_size;
1700 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1701 struct btrfs_device *device,
1702 u64 start, u64 *dev_extent_len)
1704 struct btrfs_fs_info *fs_info = device->fs_info;
1705 struct btrfs_root *root = fs_info->dev_root;
1707 struct btrfs_path *path;
1708 struct btrfs_key key;
1709 struct btrfs_key found_key;
1710 struct extent_buffer *leaf = NULL;
1711 struct btrfs_dev_extent *extent = NULL;
1713 path = btrfs_alloc_path();
1717 key.objectid = device->devid;
1719 key.type = BTRFS_DEV_EXTENT_KEY;
1721 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1723 ret = btrfs_previous_item(root, path, key.objectid,
1724 BTRFS_DEV_EXTENT_KEY);
1727 leaf = path->nodes[0];
1728 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1729 extent = btrfs_item_ptr(leaf, path->slots[0],
1730 struct btrfs_dev_extent);
1731 BUG_ON(found_key.offset > start || found_key.offset +
1732 btrfs_dev_extent_length(leaf, extent) < start);
1734 btrfs_release_path(path);
1736 } else if (ret == 0) {
1737 leaf = path->nodes[0];
1738 extent = btrfs_item_ptr(leaf, path->slots[0],
1739 struct btrfs_dev_extent);
1744 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1746 ret = btrfs_del_item(trans, root, path);
1748 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1750 btrfs_free_path(path);
1754 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1759 read_lock(&fs_info->mapping_tree_lock);
1760 n = rb_last(&fs_info->mapping_tree.rb_root);
1762 struct btrfs_chunk_map *map;
1764 map = rb_entry(n, struct btrfs_chunk_map, rb_node);
1765 ret = map->start + map->chunk_len;
1767 read_unlock(&fs_info->mapping_tree_lock);
1772 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1776 struct btrfs_key key;
1777 struct btrfs_key found_key;
1778 struct btrfs_path *path;
1780 path = btrfs_alloc_path();
1784 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1785 key.type = BTRFS_DEV_ITEM_KEY;
1786 key.offset = (u64)-1;
1788 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1794 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1799 ret = btrfs_previous_item(fs_info->chunk_root, path,
1800 BTRFS_DEV_ITEMS_OBJECTID,
1801 BTRFS_DEV_ITEM_KEY);
1805 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1807 *devid_ret = found_key.offset + 1;
1811 btrfs_free_path(path);
1816 * the device information is stored in the chunk root
1817 * the btrfs_device struct should be fully filled in
1819 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1820 struct btrfs_device *device)
1823 struct btrfs_path *path;
1824 struct btrfs_dev_item *dev_item;
1825 struct extent_buffer *leaf;
1826 struct btrfs_key key;
1829 path = btrfs_alloc_path();
1833 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1834 key.type = BTRFS_DEV_ITEM_KEY;
1835 key.offset = device->devid;
1837 btrfs_reserve_chunk_metadata(trans, true);
1838 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1839 &key, sizeof(*dev_item));
1840 btrfs_trans_release_chunk_metadata(trans);
1844 leaf = path->nodes[0];
1845 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1847 btrfs_set_device_id(leaf, dev_item, device->devid);
1848 btrfs_set_device_generation(leaf, dev_item, 0);
1849 btrfs_set_device_type(leaf, dev_item, device->type);
1850 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1851 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1852 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1853 btrfs_set_device_total_bytes(leaf, dev_item,
1854 btrfs_device_get_disk_total_bytes(device));
1855 btrfs_set_device_bytes_used(leaf, dev_item,
1856 btrfs_device_get_bytes_used(device));
1857 btrfs_set_device_group(leaf, dev_item, 0);
1858 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1859 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1860 btrfs_set_device_start_offset(leaf, dev_item, 0);
1862 ptr = btrfs_device_uuid(dev_item);
1863 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1864 ptr = btrfs_device_fsid(dev_item);
1865 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1866 ptr, BTRFS_FSID_SIZE);
1867 btrfs_mark_buffer_dirty(trans, leaf);
1871 btrfs_free_path(path);
1876 * Function to update ctime/mtime for a given device path.
1877 * Mainly used for ctime/mtime based probe like libblkid.
1879 * We don't care about errors here, this is just to be kind to userspace.
1881 static void update_dev_time(const char *device_path)
1886 ret = kern_path(device_path, LOOKUP_FOLLOW, &path);
1890 inode_update_time(d_inode(path.dentry), S_MTIME | S_CTIME | S_VERSION);
1894 static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans,
1895 struct btrfs_device *device)
1897 struct btrfs_root *root = device->fs_info->chunk_root;
1899 struct btrfs_path *path;
1900 struct btrfs_key key;
1902 path = btrfs_alloc_path();
1906 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1907 key.type = BTRFS_DEV_ITEM_KEY;
1908 key.offset = device->devid;
1910 btrfs_reserve_chunk_metadata(trans, false);
1911 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1912 btrfs_trans_release_chunk_metadata(trans);
1919 ret = btrfs_del_item(trans, root, path);
1921 btrfs_free_path(path);
1926 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1927 * filesystem. It's up to the caller to adjust that number regarding eg. device
1930 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1938 seq = read_seqbegin(&fs_info->profiles_lock);
1940 all_avail = fs_info->avail_data_alloc_bits |
1941 fs_info->avail_system_alloc_bits |
1942 fs_info->avail_metadata_alloc_bits;
1943 } while (read_seqretry(&fs_info->profiles_lock, seq));
1945 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1946 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1949 if (num_devices < btrfs_raid_array[i].devs_min)
1950 return btrfs_raid_array[i].mindev_error;
1956 static struct btrfs_device * btrfs_find_next_active_device(
1957 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1959 struct btrfs_device *next_device;
1961 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1962 if (next_device != device &&
1963 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1964 && next_device->bdev)
1972 * Helper function to check if the given device is part of s_bdev / latest_dev
1973 * and replace it with the provided or the next active device, in the context
1974 * where this function called, there should be always be another device (or
1975 * this_dev) which is active.
1977 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
1978 struct btrfs_device *next_device)
1980 struct btrfs_fs_info *fs_info = device->fs_info;
1983 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1985 ASSERT(next_device);
1987 if (fs_info->sb->s_bdev &&
1988 (fs_info->sb->s_bdev == device->bdev))
1989 fs_info->sb->s_bdev = next_device->bdev;
1991 if (fs_info->fs_devices->latest_dev->bdev == device->bdev)
1992 fs_info->fs_devices->latest_dev = next_device;
1996 * Return btrfs_fs_devices::num_devices excluding the device that's being
1997 * currently replaced.
1999 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2001 u64 num_devices = fs_info->fs_devices->num_devices;
2003 down_read(&fs_info->dev_replace.rwsem);
2004 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2005 ASSERT(num_devices > 1);
2008 up_read(&fs_info->dev_replace.rwsem);
2013 static void btrfs_scratch_superblock(struct btrfs_fs_info *fs_info,
2014 struct block_device *bdev, int copy_num)
2016 struct btrfs_super_block *disk_super;
2017 const size_t len = sizeof(disk_super->magic);
2018 const u64 bytenr = btrfs_sb_offset(copy_num);
2021 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr);
2022 if (IS_ERR(disk_super))
2025 memset(&disk_super->magic, 0, len);
2026 folio_mark_dirty(virt_to_folio(disk_super));
2027 btrfs_release_disk_super(disk_super);
2029 ret = sync_blockdev_range(bdev, bytenr, bytenr + len - 1);
2031 btrfs_warn(fs_info, "error clearing superblock number %d (%d)",
2035 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
2036 struct block_device *bdev,
2037 const char *device_path)
2044 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2045 if (bdev_is_zoned(bdev))
2046 btrfs_reset_sb_log_zones(bdev, copy_num);
2048 btrfs_scratch_superblock(fs_info, bdev, copy_num);
2051 /* Notify udev that device has changed */
2052 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2054 /* Update ctime/mtime for device path for libblkid */
2055 update_dev_time(device_path);
2058 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
2059 struct btrfs_dev_lookup_args *args,
2060 struct bdev_handle **bdev_handle)
2062 struct btrfs_trans_handle *trans;
2063 struct btrfs_device *device;
2064 struct btrfs_fs_devices *cur_devices;
2065 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2069 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2070 btrfs_err(fs_info, "device remove not supported on extent tree v2 yet");
2075 * The device list in fs_devices is accessed without locks (neither
2076 * uuid_mutex nor device_list_mutex) as it won't change on a mounted
2077 * filesystem and another device rm cannot run.
2079 num_devices = btrfs_num_devices(fs_info);
2081 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2085 device = btrfs_find_device(fs_info->fs_devices, args);
2088 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2094 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2095 btrfs_warn_in_rcu(fs_info,
2096 "cannot remove device %s (devid %llu) due to active swapfile",
2097 btrfs_dev_name(device), device->devid);
2101 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
2102 return BTRFS_ERROR_DEV_TGT_REPLACE;
2104 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2105 fs_info->fs_devices->rw_devices == 1)
2106 return BTRFS_ERROR_DEV_ONLY_WRITABLE;
2108 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2109 mutex_lock(&fs_info->chunk_mutex);
2110 list_del_init(&device->dev_alloc_list);
2111 device->fs_devices->rw_devices--;
2112 mutex_unlock(&fs_info->chunk_mutex);
2115 ret = btrfs_shrink_device(device, 0);
2119 trans = btrfs_start_transaction(fs_info->chunk_root, 0);
2120 if (IS_ERR(trans)) {
2121 ret = PTR_ERR(trans);
2125 ret = btrfs_rm_dev_item(trans, device);
2127 /* Any error in dev item removal is critical */
2129 "failed to remove device item for devid %llu: %d",
2130 device->devid, ret);
2131 btrfs_abort_transaction(trans, ret);
2132 btrfs_end_transaction(trans);
2136 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2137 btrfs_scrub_cancel_dev(device);
2140 * the device list mutex makes sure that we don't change
2141 * the device list while someone else is writing out all
2142 * the device supers. Whoever is writing all supers, should
2143 * lock the device list mutex before getting the number of
2144 * devices in the super block (super_copy). Conversely,
2145 * whoever updates the number of devices in the super block
2146 * (super_copy) should hold the device list mutex.
2150 * In normal cases the cur_devices == fs_devices. But in case
2151 * of deleting a seed device, the cur_devices should point to
2152 * its own fs_devices listed under the fs_devices->seed_list.
2154 cur_devices = device->fs_devices;
2155 mutex_lock(&fs_devices->device_list_mutex);
2156 list_del_rcu(&device->dev_list);
2158 cur_devices->num_devices--;
2159 cur_devices->total_devices--;
2160 /* Update total_devices of the parent fs_devices if it's seed */
2161 if (cur_devices != fs_devices)
2162 fs_devices->total_devices--;
2164 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2165 cur_devices->missing_devices--;
2167 btrfs_assign_next_active_device(device, NULL);
2169 if (device->bdev_handle) {
2170 cur_devices->open_devices--;
2171 /* remove sysfs entry */
2172 btrfs_sysfs_remove_device(device);
2175 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2176 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2177 mutex_unlock(&fs_devices->device_list_mutex);
2180 * At this point, the device is zero sized and detached from the
2181 * devices list. All that's left is to zero out the old supers and
2184 * We cannot call btrfs_close_bdev() here because we're holding the sb
2185 * write lock, and bdev_release() will pull in the ->open_mutex on
2186 * the block device and it's dependencies. Instead just flush the
2187 * device and let the caller do the final bdev_release.
2189 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2190 btrfs_scratch_superblocks(fs_info, device->bdev,
2193 sync_blockdev(device->bdev);
2194 invalidate_bdev(device->bdev);
2198 *bdev_handle = device->bdev_handle;
2200 btrfs_free_device(device);
2203 * This can happen if cur_devices is the private seed devices list. We
2204 * cannot call close_fs_devices() here because it expects the uuid_mutex
2205 * to be held, but in fact we don't need that for the private
2206 * seed_devices, we can simply decrement cur_devices->opened and then
2207 * remove it from our list and free the fs_devices.
2209 if (cur_devices->num_devices == 0) {
2210 list_del_init(&cur_devices->seed_list);
2211 ASSERT(cur_devices->opened == 1);
2212 cur_devices->opened--;
2213 free_fs_devices(cur_devices);
2216 ret = btrfs_commit_transaction(trans);
2221 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2222 mutex_lock(&fs_info->chunk_mutex);
2223 list_add(&device->dev_alloc_list,
2224 &fs_devices->alloc_list);
2225 device->fs_devices->rw_devices++;
2226 mutex_unlock(&fs_info->chunk_mutex);
2231 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2233 struct btrfs_fs_devices *fs_devices;
2235 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2238 * in case of fs with no seed, srcdev->fs_devices will point
2239 * to fs_devices of fs_info. However when the dev being replaced is
2240 * a seed dev it will point to the seed's local fs_devices. In short
2241 * srcdev will have its correct fs_devices in both the cases.
2243 fs_devices = srcdev->fs_devices;
2245 list_del_rcu(&srcdev->dev_list);
2246 list_del(&srcdev->dev_alloc_list);
2247 fs_devices->num_devices--;
2248 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2249 fs_devices->missing_devices--;
2251 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2252 fs_devices->rw_devices--;
2255 fs_devices->open_devices--;
2258 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2260 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2262 mutex_lock(&uuid_mutex);
2264 btrfs_close_bdev(srcdev);
2266 btrfs_free_device(srcdev);
2268 /* if this is no devs we rather delete the fs_devices */
2269 if (!fs_devices->num_devices) {
2271 * On a mounted FS, num_devices can't be zero unless it's a
2272 * seed. In case of a seed device being replaced, the replace
2273 * target added to the sprout FS, so there will be no more
2274 * device left under the seed FS.
2276 ASSERT(fs_devices->seeding);
2278 list_del_init(&fs_devices->seed_list);
2279 close_fs_devices(fs_devices);
2280 free_fs_devices(fs_devices);
2282 mutex_unlock(&uuid_mutex);
2285 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2287 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2289 mutex_lock(&fs_devices->device_list_mutex);
2291 btrfs_sysfs_remove_device(tgtdev);
2294 fs_devices->open_devices--;
2296 fs_devices->num_devices--;
2298 btrfs_assign_next_active_device(tgtdev, NULL);
2300 list_del_rcu(&tgtdev->dev_list);
2302 mutex_unlock(&fs_devices->device_list_mutex);
2304 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2307 btrfs_close_bdev(tgtdev);
2309 btrfs_free_device(tgtdev);
2313 * Populate args from device at path.
2315 * @fs_info: the filesystem
2316 * @args: the args to populate
2317 * @path: the path to the device
2319 * This will read the super block of the device at @path and populate @args with
2320 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to
2321 * lookup a device to operate on, but need to do it before we take any locks.
2322 * This properly handles the special case of "missing" that a user may pass in,
2323 * and does some basic sanity checks. The caller must make sure that @path is
2324 * properly NUL terminated before calling in, and must call
2325 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and
2328 * Return: 0 for success, -errno for failure
2330 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,
2331 struct btrfs_dev_lookup_args *args,
2334 struct btrfs_super_block *disk_super;
2335 struct bdev_handle *bdev_handle;
2338 if (!path || !path[0])
2340 if (!strcmp(path, "missing")) {
2341 args->missing = true;
2345 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL);
2346 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL);
2347 if (!args->uuid || !args->fsid) {
2348 btrfs_put_dev_args_from_path(args);
2352 ret = btrfs_get_bdev_and_sb(path, BLK_OPEN_READ, NULL, 0,
2353 &bdev_handle, &disk_super);
2355 btrfs_put_dev_args_from_path(args);
2359 args->devid = btrfs_stack_device_id(&disk_super->dev_item);
2360 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE);
2361 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2362 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE);
2364 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
2365 btrfs_release_disk_super(disk_super);
2366 bdev_release(bdev_handle);
2371 * Only use this jointly with btrfs_get_dev_args_from_path() because we will
2372 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables
2373 * that don't need to be freed.
2375 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args)
2383 struct btrfs_device *btrfs_find_device_by_devspec(
2384 struct btrfs_fs_info *fs_info, u64 devid,
2385 const char *device_path)
2387 BTRFS_DEV_LOOKUP_ARGS(args);
2388 struct btrfs_device *device;
2393 device = btrfs_find_device(fs_info->fs_devices, &args);
2395 return ERR_PTR(-ENOENT);
2399 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path);
2401 return ERR_PTR(ret);
2402 device = btrfs_find_device(fs_info->fs_devices, &args);
2403 btrfs_put_dev_args_from_path(&args);
2405 return ERR_PTR(-ENOENT);
2409 static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info)
2411 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2412 struct btrfs_fs_devices *old_devices;
2413 struct btrfs_fs_devices *seed_devices;
2415 lockdep_assert_held(&uuid_mutex);
2416 if (!fs_devices->seeding)
2417 return ERR_PTR(-EINVAL);
2420 * Private copy of the seed devices, anchored at
2421 * fs_info->fs_devices->seed_list
2423 seed_devices = alloc_fs_devices(NULL);
2424 if (IS_ERR(seed_devices))
2425 return seed_devices;
2428 * It's necessary to retain a copy of the original seed fs_devices in
2429 * fs_uuids so that filesystems which have been seeded can successfully
2430 * reference the seed device from open_seed_devices. This also supports
2433 old_devices = clone_fs_devices(fs_devices);
2434 if (IS_ERR(old_devices)) {
2435 kfree(seed_devices);
2439 list_add(&old_devices->fs_list, &fs_uuids);
2441 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2442 seed_devices->opened = 1;
2443 INIT_LIST_HEAD(&seed_devices->devices);
2444 INIT_LIST_HEAD(&seed_devices->alloc_list);
2445 mutex_init(&seed_devices->device_list_mutex);
2447 return seed_devices;
2451 * Splice seed devices into the sprout fs_devices.
2452 * Generate a new fsid for the sprouted read-write filesystem.
2454 static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info,
2455 struct btrfs_fs_devices *seed_devices)
2457 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2458 struct btrfs_super_block *disk_super = fs_info->super_copy;
2459 struct btrfs_device *device;
2463 * We are updating the fsid, the thread leading to device_list_add()
2464 * could race, so uuid_mutex is needed.
2466 lockdep_assert_held(&uuid_mutex);
2469 * The threads listed below may traverse dev_list but can do that without
2470 * device_list_mutex:
2471 * - All device ops and balance - as we are in btrfs_exclop_start.
2472 * - Various dev_list readers - are using RCU.
2473 * - btrfs_ioctl_fitrim() - is using RCU.
2475 * For-read threads as below are using device_list_mutex:
2476 * - Readonly scrub btrfs_scrub_dev()
2477 * - Readonly scrub btrfs_scrub_progress()
2478 * - btrfs_get_dev_stats()
2480 lockdep_assert_held(&fs_devices->device_list_mutex);
2482 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2484 list_for_each_entry(device, &seed_devices->devices, dev_list)
2485 device->fs_devices = seed_devices;
2487 fs_devices->seeding = false;
2488 fs_devices->num_devices = 0;
2489 fs_devices->open_devices = 0;
2490 fs_devices->missing_devices = 0;
2491 fs_devices->rotating = false;
2492 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2494 generate_random_uuid(fs_devices->fsid);
2495 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2496 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2498 super_flags = btrfs_super_flags(disk_super) &
2499 ~BTRFS_SUPER_FLAG_SEEDING;
2500 btrfs_set_super_flags(disk_super, super_flags);
2504 * Store the expected generation for seed devices in device items.
2506 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2508 BTRFS_DEV_LOOKUP_ARGS(args);
2509 struct btrfs_fs_info *fs_info = trans->fs_info;
2510 struct btrfs_root *root = fs_info->chunk_root;
2511 struct btrfs_path *path;
2512 struct extent_buffer *leaf;
2513 struct btrfs_dev_item *dev_item;
2514 struct btrfs_device *device;
2515 struct btrfs_key key;
2516 u8 fs_uuid[BTRFS_FSID_SIZE];
2517 u8 dev_uuid[BTRFS_UUID_SIZE];
2520 path = btrfs_alloc_path();
2524 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2526 key.type = BTRFS_DEV_ITEM_KEY;
2529 btrfs_reserve_chunk_metadata(trans, false);
2530 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2531 btrfs_trans_release_chunk_metadata(trans);
2535 leaf = path->nodes[0];
2537 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2538 ret = btrfs_next_leaf(root, path);
2543 leaf = path->nodes[0];
2544 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2545 btrfs_release_path(path);
2549 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2550 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2551 key.type != BTRFS_DEV_ITEM_KEY)
2554 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2555 struct btrfs_dev_item);
2556 args.devid = btrfs_device_id(leaf, dev_item);
2557 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2559 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2561 args.uuid = dev_uuid;
2562 args.fsid = fs_uuid;
2563 device = btrfs_find_device(fs_info->fs_devices, &args);
2564 BUG_ON(!device); /* Logic error */
2566 if (device->fs_devices->seeding) {
2567 btrfs_set_device_generation(leaf, dev_item,
2568 device->generation);
2569 btrfs_mark_buffer_dirty(trans, leaf);
2577 btrfs_free_path(path);
2581 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2583 struct btrfs_root *root = fs_info->dev_root;
2584 struct btrfs_trans_handle *trans;
2585 struct btrfs_device *device;
2586 struct bdev_handle *bdev_handle;
2587 struct super_block *sb = fs_info->sb;
2588 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2589 struct btrfs_fs_devices *seed_devices = NULL;
2590 u64 orig_super_total_bytes;
2591 u64 orig_super_num_devices;
2593 bool seeding_dev = false;
2594 bool locked = false;
2596 if (sb_rdonly(sb) && !fs_devices->seeding)
2599 bdev_handle = bdev_open_by_path(device_path, BLK_OPEN_WRITE,
2600 fs_info->bdev_holder, NULL);
2601 if (IS_ERR(bdev_handle))
2602 return PTR_ERR(bdev_handle);
2604 if (!btrfs_check_device_zone_type(fs_info, bdev_handle->bdev)) {
2609 if (fs_devices->seeding) {
2611 down_write(&sb->s_umount);
2612 mutex_lock(&uuid_mutex);
2616 sync_blockdev(bdev_handle->bdev);
2619 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2620 if (device->bdev == bdev_handle->bdev) {
2628 device = btrfs_alloc_device(fs_info, NULL, NULL, device_path);
2629 if (IS_ERR(device)) {
2630 /* we can safely leave the fs_devices entry around */
2631 ret = PTR_ERR(device);
2635 device->fs_info = fs_info;
2636 device->bdev_handle = bdev_handle;
2637 device->bdev = bdev_handle->bdev;
2638 ret = lookup_bdev(device_path, &device->devt);
2640 goto error_free_device;
2642 ret = btrfs_get_dev_zone_info(device, false);
2644 goto error_free_device;
2646 trans = btrfs_start_transaction(root, 0);
2647 if (IS_ERR(trans)) {
2648 ret = PTR_ERR(trans);
2649 goto error_free_zone;
2652 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2653 device->generation = trans->transid;
2654 device->io_width = fs_info->sectorsize;
2655 device->io_align = fs_info->sectorsize;
2656 device->sector_size = fs_info->sectorsize;
2657 device->total_bytes =
2658 round_down(bdev_nr_bytes(device->bdev), fs_info->sectorsize);
2659 device->disk_total_bytes = device->total_bytes;
2660 device->commit_total_bytes = device->total_bytes;
2661 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2662 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2663 device->dev_stats_valid = 1;
2664 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2667 btrfs_clear_sb_rdonly(sb);
2669 /* GFP_KERNEL allocation must not be under device_list_mutex */
2670 seed_devices = btrfs_init_sprout(fs_info);
2671 if (IS_ERR(seed_devices)) {
2672 ret = PTR_ERR(seed_devices);
2673 btrfs_abort_transaction(trans, ret);
2678 mutex_lock(&fs_devices->device_list_mutex);
2680 btrfs_setup_sprout(fs_info, seed_devices);
2681 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev,
2685 device->fs_devices = fs_devices;
2687 mutex_lock(&fs_info->chunk_mutex);
2688 list_add_rcu(&device->dev_list, &fs_devices->devices);
2689 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2690 fs_devices->num_devices++;
2691 fs_devices->open_devices++;
2692 fs_devices->rw_devices++;
2693 fs_devices->total_devices++;
2694 fs_devices->total_rw_bytes += device->total_bytes;
2696 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2698 if (!bdev_nonrot(device->bdev))
2699 fs_devices->rotating = true;
2701 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2702 btrfs_set_super_total_bytes(fs_info->super_copy,
2703 round_down(orig_super_total_bytes + device->total_bytes,
2704 fs_info->sectorsize));
2706 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2707 btrfs_set_super_num_devices(fs_info->super_copy,
2708 orig_super_num_devices + 1);
2711 * we've got more storage, clear any full flags on the space
2714 btrfs_clear_space_info_full(fs_info);
2716 mutex_unlock(&fs_info->chunk_mutex);
2718 /* Add sysfs device entry */
2719 btrfs_sysfs_add_device(device);
2721 mutex_unlock(&fs_devices->device_list_mutex);
2724 mutex_lock(&fs_info->chunk_mutex);
2725 ret = init_first_rw_device(trans);
2726 mutex_unlock(&fs_info->chunk_mutex);
2728 btrfs_abort_transaction(trans, ret);
2733 ret = btrfs_add_dev_item(trans, device);
2735 btrfs_abort_transaction(trans, ret);
2740 ret = btrfs_finish_sprout(trans);
2742 btrfs_abort_transaction(trans, ret);
2747 * fs_devices now represents the newly sprouted filesystem and
2748 * its fsid has been changed by btrfs_sprout_splice().
2750 btrfs_sysfs_update_sprout_fsid(fs_devices);
2753 ret = btrfs_commit_transaction(trans);
2756 mutex_unlock(&uuid_mutex);
2757 up_write(&sb->s_umount);
2760 if (ret) /* transaction commit */
2763 ret = btrfs_relocate_sys_chunks(fs_info);
2765 btrfs_handle_fs_error(fs_info, ret,
2766 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2767 trans = btrfs_attach_transaction(root);
2768 if (IS_ERR(trans)) {
2769 if (PTR_ERR(trans) == -ENOENT)
2771 ret = PTR_ERR(trans);
2775 ret = btrfs_commit_transaction(trans);
2779 * Now that we have written a new super block to this device, check all
2780 * other fs_devices list if device_path alienates any other scanned
2782 * We can ignore the return value as it typically returns -EINVAL and
2783 * only succeeds if the device was an alien.
2785 btrfs_forget_devices(device->devt);
2787 /* Update ctime/mtime for blkid or udev */
2788 update_dev_time(device_path);
2793 btrfs_sysfs_remove_device(device);
2794 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2795 mutex_lock(&fs_info->chunk_mutex);
2796 list_del_rcu(&device->dev_list);
2797 list_del(&device->dev_alloc_list);
2798 fs_info->fs_devices->num_devices--;
2799 fs_info->fs_devices->open_devices--;
2800 fs_info->fs_devices->rw_devices--;
2801 fs_info->fs_devices->total_devices--;
2802 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2803 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2804 btrfs_set_super_total_bytes(fs_info->super_copy,
2805 orig_super_total_bytes);
2806 btrfs_set_super_num_devices(fs_info->super_copy,
2807 orig_super_num_devices);
2808 mutex_unlock(&fs_info->chunk_mutex);
2809 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2812 btrfs_set_sb_rdonly(sb);
2814 btrfs_end_transaction(trans);
2816 btrfs_destroy_dev_zone_info(device);
2818 btrfs_free_device(device);
2820 bdev_release(bdev_handle);
2822 mutex_unlock(&uuid_mutex);
2823 up_write(&sb->s_umount);
2828 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2829 struct btrfs_device *device)
2832 struct btrfs_path *path;
2833 struct btrfs_root *root = device->fs_info->chunk_root;
2834 struct btrfs_dev_item *dev_item;
2835 struct extent_buffer *leaf;
2836 struct btrfs_key key;
2838 path = btrfs_alloc_path();
2842 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2843 key.type = BTRFS_DEV_ITEM_KEY;
2844 key.offset = device->devid;
2846 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2855 leaf = path->nodes[0];
2856 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2858 btrfs_set_device_id(leaf, dev_item, device->devid);
2859 btrfs_set_device_type(leaf, dev_item, device->type);
2860 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2861 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2862 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2863 btrfs_set_device_total_bytes(leaf, dev_item,
2864 btrfs_device_get_disk_total_bytes(device));
2865 btrfs_set_device_bytes_used(leaf, dev_item,
2866 btrfs_device_get_bytes_used(device));
2867 btrfs_mark_buffer_dirty(trans, leaf);
2870 btrfs_free_path(path);
2874 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2875 struct btrfs_device *device, u64 new_size)
2877 struct btrfs_fs_info *fs_info = device->fs_info;
2878 struct btrfs_super_block *super_copy = fs_info->super_copy;
2883 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2886 new_size = round_down(new_size, fs_info->sectorsize);
2888 mutex_lock(&fs_info->chunk_mutex);
2889 old_total = btrfs_super_total_bytes(super_copy);
2890 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2892 if (new_size <= device->total_bytes ||
2893 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2894 mutex_unlock(&fs_info->chunk_mutex);
2898 btrfs_set_super_total_bytes(super_copy,
2899 round_down(old_total + diff, fs_info->sectorsize));
2900 device->fs_devices->total_rw_bytes += diff;
2901 atomic64_add(diff, &fs_info->free_chunk_space);
2903 btrfs_device_set_total_bytes(device, new_size);
2904 btrfs_device_set_disk_total_bytes(device, new_size);
2905 btrfs_clear_space_info_full(device->fs_info);
2906 if (list_empty(&device->post_commit_list))
2907 list_add_tail(&device->post_commit_list,
2908 &trans->transaction->dev_update_list);
2909 mutex_unlock(&fs_info->chunk_mutex);
2911 btrfs_reserve_chunk_metadata(trans, false);
2912 ret = btrfs_update_device(trans, device);
2913 btrfs_trans_release_chunk_metadata(trans);
2918 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2920 struct btrfs_fs_info *fs_info = trans->fs_info;
2921 struct btrfs_root *root = fs_info->chunk_root;
2923 struct btrfs_path *path;
2924 struct btrfs_key key;
2926 path = btrfs_alloc_path();
2930 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2931 key.offset = chunk_offset;
2932 key.type = BTRFS_CHUNK_ITEM_KEY;
2934 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2937 else if (ret > 0) { /* Logic error or corruption */
2938 btrfs_handle_fs_error(fs_info, -ENOENT,
2939 "Failed lookup while freeing chunk.");
2944 ret = btrfs_del_item(trans, root, path);
2946 btrfs_handle_fs_error(fs_info, ret,
2947 "Failed to delete chunk item.");
2949 btrfs_free_path(path);
2953 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2955 struct btrfs_super_block *super_copy = fs_info->super_copy;
2956 struct btrfs_disk_key *disk_key;
2957 struct btrfs_chunk *chunk;
2964 struct btrfs_key key;
2966 lockdep_assert_held(&fs_info->chunk_mutex);
2967 array_size = btrfs_super_sys_array_size(super_copy);
2969 ptr = super_copy->sys_chunk_array;
2972 while (cur < array_size) {
2973 disk_key = (struct btrfs_disk_key *)ptr;
2974 btrfs_disk_key_to_cpu(&key, disk_key);
2976 len = sizeof(*disk_key);
2978 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2979 chunk = (struct btrfs_chunk *)(ptr + len);
2980 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2981 len += btrfs_chunk_item_size(num_stripes);
2986 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2987 key.offset == chunk_offset) {
2988 memmove(ptr, ptr + len, array_size - (cur + len));
2990 btrfs_set_super_sys_array_size(super_copy, array_size);
2999 struct btrfs_chunk_map *btrfs_find_chunk_map_nolock(struct btrfs_fs_info *fs_info,
3000 u64 logical, u64 length)
3002 struct rb_node *node = fs_info->mapping_tree.rb_root.rb_node;
3003 struct rb_node *prev = NULL;
3004 struct rb_node *orig_prev;
3005 struct btrfs_chunk_map *map;
3006 struct btrfs_chunk_map *prev_map = NULL;
3009 map = rb_entry(node, struct btrfs_chunk_map, rb_node);
3013 if (logical < map->start) {
3014 node = node->rb_left;
3015 } else if (logical >= map->start + map->chunk_len) {
3016 node = node->rb_right;
3018 refcount_inc(&map->refs);
3027 while (prev && logical >= prev_map->start + prev_map->chunk_len) {
3028 prev = rb_next(prev);
3029 prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node);
3034 prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node);
3035 while (prev && logical < prev_map->start) {
3036 prev = rb_prev(prev);
3037 prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node);
3042 u64 end = logical + length;
3045 * Caller can pass a U64_MAX length when it wants to get any
3046 * chunk starting at an offset of 'logical' or higher, so deal
3047 * with underflow by resetting the end offset to U64_MAX.
3052 if (end > prev_map->start &&
3053 logical < prev_map->start + prev_map->chunk_len) {
3054 refcount_inc(&prev_map->refs);
3062 struct btrfs_chunk_map *btrfs_find_chunk_map(struct btrfs_fs_info *fs_info,
3063 u64 logical, u64 length)
3065 struct btrfs_chunk_map *map;
3067 read_lock(&fs_info->mapping_tree_lock);
3068 map = btrfs_find_chunk_map_nolock(fs_info, logical, length);
3069 read_unlock(&fs_info->mapping_tree_lock);
3075 * Find the mapping containing the given logical extent.
3077 * @logical: Logical block offset in bytes.
3078 * @length: Length of extent in bytes.
3080 * Return: Chunk mapping or ERR_PTR.
3082 struct btrfs_chunk_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
3083 u64 logical, u64 length)
3085 struct btrfs_chunk_map *map;
3087 map = btrfs_find_chunk_map(fs_info, logical, length);
3089 if (unlikely(!map)) {
3090 read_unlock(&fs_info->mapping_tree_lock);
3092 "unable to find chunk map for logical %llu length %llu",
3094 return ERR_PTR(-EINVAL);
3097 if (unlikely(map->start > logical || map->start + map->chunk_len <= logical)) {
3098 read_unlock(&fs_info->mapping_tree_lock);
3100 "found a bad chunk map, wanted %llu-%llu, found %llu-%llu",
3101 logical, logical + length, map->start,
3102 map->start + map->chunk_len);
3103 btrfs_free_chunk_map(map);
3104 return ERR_PTR(-EINVAL);
3107 /* Callers are responsible for dropping the reference. */
3111 static int remove_chunk_item(struct btrfs_trans_handle *trans,
3112 struct btrfs_chunk_map *map, u64 chunk_offset)
3117 * Removing chunk items and updating the device items in the chunks btree
3118 * requires holding the chunk_mutex.
3119 * See the comment at btrfs_chunk_alloc() for the details.
3121 lockdep_assert_held(&trans->fs_info->chunk_mutex);
3123 for (i = 0; i < map->num_stripes; i++) {
3126 ret = btrfs_update_device(trans, map->stripes[i].dev);
3131 return btrfs_free_chunk(trans, chunk_offset);
3134 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
3136 struct btrfs_fs_info *fs_info = trans->fs_info;
3137 struct btrfs_chunk_map *map;
3138 u64 dev_extent_len = 0;
3140 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3142 map = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
3145 * This is a logic error, but we don't want to just rely on the
3146 * user having built with ASSERT enabled, so if ASSERT doesn't
3147 * do anything we still error out.
3150 return PTR_ERR(map);
3154 * First delete the device extent items from the devices btree.
3155 * We take the device_list_mutex to avoid racing with the finishing phase
3156 * of a device replace operation. See the comment below before acquiring
3157 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
3158 * because that can result in a deadlock when deleting the device extent
3159 * items from the devices btree - COWing an extent buffer from the btree
3160 * may result in allocating a new metadata chunk, which would attempt to
3161 * lock again fs_info->chunk_mutex.
3163 mutex_lock(&fs_devices->device_list_mutex);
3164 for (i = 0; i < map->num_stripes; i++) {
3165 struct btrfs_device *device = map->stripes[i].dev;
3166 ret = btrfs_free_dev_extent(trans, device,
3167 map->stripes[i].physical,
3170 mutex_unlock(&fs_devices->device_list_mutex);
3171 btrfs_abort_transaction(trans, ret);
3175 if (device->bytes_used > 0) {
3176 mutex_lock(&fs_info->chunk_mutex);
3177 btrfs_device_set_bytes_used(device,
3178 device->bytes_used - dev_extent_len);
3179 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3180 btrfs_clear_space_info_full(fs_info);
3181 mutex_unlock(&fs_info->chunk_mutex);
3184 mutex_unlock(&fs_devices->device_list_mutex);
3187 * We acquire fs_info->chunk_mutex for 2 reasons:
3189 * 1) Just like with the first phase of the chunk allocation, we must
3190 * reserve system space, do all chunk btree updates and deletions, and
3191 * update the system chunk array in the superblock while holding this
3192 * mutex. This is for similar reasons as explained on the comment at
3193 * the top of btrfs_chunk_alloc();
3195 * 2) Prevent races with the final phase of a device replace operation
3196 * that replaces the device object associated with the map's stripes,
3197 * because the device object's id can change at any time during that
3198 * final phase of the device replace operation
3199 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3200 * replaced device and then see it with an ID of
3201 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3202 * the device item, which does not exists on the chunk btree.
3203 * The finishing phase of device replace acquires both the
3204 * device_list_mutex and the chunk_mutex, in that order, so we are
3205 * safe by just acquiring the chunk_mutex.
3207 trans->removing_chunk = true;
3208 mutex_lock(&fs_info->chunk_mutex);
3210 check_system_chunk(trans, map->type);
3212 ret = remove_chunk_item(trans, map, chunk_offset);
3214 * Normally we should not get -ENOSPC since we reserved space before
3215 * through the call to check_system_chunk().
3217 * Despite our system space_info having enough free space, we may not
3218 * be able to allocate extents from its block groups, because all have
3219 * an incompatible profile, which will force us to allocate a new system
3220 * block group with the right profile, or right after we called
3221 * check_system_space() above, a scrub turned the only system block group
3222 * with enough free space into RO mode.
3223 * This is explained with more detail at do_chunk_alloc().
3225 * So if we get -ENOSPC, allocate a new system chunk and retry once.
3227 if (ret == -ENOSPC) {
3228 const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
3229 struct btrfs_block_group *sys_bg;
3231 sys_bg = btrfs_create_chunk(trans, sys_flags);
3232 if (IS_ERR(sys_bg)) {
3233 ret = PTR_ERR(sys_bg);
3234 btrfs_abort_transaction(trans, ret);
3238 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3240 btrfs_abort_transaction(trans, ret);
3244 ret = remove_chunk_item(trans, map, chunk_offset);
3246 btrfs_abort_transaction(trans, ret);
3250 btrfs_abort_transaction(trans, ret);
3254 trace_btrfs_chunk_free(fs_info, map, chunk_offset, map->chunk_len);
3256 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3257 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3259 btrfs_abort_transaction(trans, ret);
3264 mutex_unlock(&fs_info->chunk_mutex);
3265 trans->removing_chunk = false;
3268 * We are done with chunk btree updates and deletions, so release the
3269 * system space we previously reserved (with check_system_chunk()).
3271 btrfs_trans_release_chunk_metadata(trans);
3273 ret = btrfs_remove_block_group(trans, map);
3275 btrfs_abort_transaction(trans, ret);
3280 if (trans->removing_chunk) {
3281 mutex_unlock(&fs_info->chunk_mutex);
3282 trans->removing_chunk = false;
3285 btrfs_free_chunk_map(map);
3289 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3291 struct btrfs_root *root = fs_info->chunk_root;
3292 struct btrfs_trans_handle *trans;
3293 struct btrfs_block_group *block_group;
3297 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3299 "relocate: not supported on extent tree v2 yet");
3304 * Prevent races with automatic removal of unused block groups.
3305 * After we relocate and before we remove the chunk with offset
3306 * chunk_offset, automatic removal of the block group can kick in,
3307 * resulting in a failure when calling btrfs_remove_chunk() below.
3309 * Make sure to acquire this mutex before doing a tree search (dev
3310 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3311 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3312 * we release the path used to search the chunk/dev tree and before
3313 * the current task acquires this mutex and calls us.
3315 lockdep_assert_held(&fs_info->reclaim_bgs_lock);
3317 /* step one, relocate all the extents inside this chunk */
3318 btrfs_scrub_pause(fs_info);
3319 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3320 btrfs_scrub_continue(fs_info);
3323 * If we had a transaction abort, stop all running scrubs.
3324 * See transaction.c:cleanup_transaction() why we do it here.
3326 if (BTRFS_FS_ERROR(fs_info))
3327 btrfs_scrub_cancel(fs_info);
3331 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3334 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3335 length = block_group->length;
3336 btrfs_put_block_group(block_group);
3339 * On a zoned file system, discard the whole block group, this will
3340 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3341 * resetting the zone fails, don't treat it as a fatal problem from the
3342 * filesystem's point of view.
3344 if (btrfs_is_zoned(fs_info)) {
3345 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
3348 "failed to reset zone %llu after relocation",
3352 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3354 if (IS_ERR(trans)) {
3355 ret = PTR_ERR(trans);
3356 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3361 * step two, delete the device extents and the
3362 * chunk tree entries
3364 ret = btrfs_remove_chunk(trans, chunk_offset);
3365 btrfs_end_transaction(trans);
3369 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3371 struct btrfs_root *chunk_root = fs_info->chunk_root;
3372 struct btrfs_path *path;
3373 struct extent_buffer *leaf;
3374 struct btrfs_chunk *chunk;
3375 struct btrfs_key key;
3376 struct btrfs_key found_key;
3378 bool retried = false;
3382 path = btrfs_alloc_path();
3387 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3388 key.offset = (u64)-1;
3389 key.type = BTRFS_CHUNK_ITEM_KEY;
3392 mutex_lock(&fs_info->reclaim_bgs_lock);
3393 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3395 mutex_unlock(&fs_info->reclaim_bgs_lock);
3398 BUG_ON(ret == 0); /* Corruption */
3400 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3403 mutex_unlock(&fs_info->reclaim_bgs_lock);
3409 leaf = path->nodes[0];
3410 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3412 chunk = btrfs_item_ptr(leaf, path->slots[0],
3413 struct btrfs_chunk);
3414 chunk_type = btrfs_chunk_type(leaf, chunk);
3415 btrfs_release_path(path);
3417 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3418 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3424 mutex_unlock(&fs_info->reclaim_bgs_lock);
3426 if (found_key.offset == 0)
3428 key.offset = found_key.offset - 1;
3431 if (failed && !retried) {
3435 } else if (WARN_ON(failed && retried)) {
3439 btrfs_free_path(path);
3444 * return 1 : allocate a data chunk successfully,
3445 * return <0: errors during allocating a data chunk,
3446 * return 0 : no need to allocate a data chunk.
3448 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3451 struct btrfs_block_group *cache;
3455 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3457 chunk_type = cache->flags;
3458 btrfs_put_block_group(cache);
3460 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3463 spin_lock(&fs_info->data_sinfo->lock);
3464 bytes_used = fs_info->data_sinfo->bytes_used;
3465 spin_unlock(&fs_info->data_sinfo->lock);
3468 struct btrfs_trans_handle *trans;
3471 trans = btrfs_join_transaction(fs_info->tree_root);
3473 return PTR_ERR(trans);
3475 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3476 btrfs_end_transaction(trans);
3485 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3486 struct btrfs_balance_control *bctl)
3488 struct btrfs_root *root = fs_info->tree_root;
3489 struct btrfs_trans_handle *trans;
3490 struct btrfs_balance_item *item;
3491 struct btrfs_disk_balance_args disk_bargs;
3492 struct btrfs_path *path;
3493 struct extent_buffer *leaf;
3494 struct btrfs_key key;
3497 path = btrfs_alloc_path();
3501 trans = btrfs_start_transaction(root, 0);
3502 if (IS_ERR(trans)) {
3503 btrfs_free_path(path);
3504 return PTR_ERR(trans);
3507 key.objectid = BTRFS_BALANCE_OBJECTID;
3508 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3511 ret = btrfs_insert_empty_item(trans, root, path, &key,
3516 leaf = path->nodes[0];
3517 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3519 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3521 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3522 btrfs_set_balance_data(leaf, item, &disk_bargs);
3523 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3524 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3525 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3526 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3528 btrfs_set_balance_flags(leaf, item, bctl->flags);
3530 btrfs_mark_buffer_dirty(trans, leaf);
3532 btrfs_free_path(path);
3533 err = btrfs_commit_transaction(trans);
3539 static int del_balance_item(struct btrfs_fs_info *fs_info)
3541 struct btrfs_root *root = fs_info->tree_root;
3542 struct btrfs_trans_handle *trans;
3543 struct btrfs_path *path;
3544 struct btrfs_key key;
3547 path = btrfs_alloc_path();
3551 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3552 if (IS_ERR(trans)) {
3553 btrfs_free_path(path);
3554 return PTR_ERR(trans);
3557 key.objectid = BTRFS_BALANCE_OBJECTID;
3558 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3561 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3569 ret = btrfs_del_item(trans, root, path);
3571 btrfs_free_path(path);
3572 err = btrfs_commit_transaction(trans);
3579 * This is a heuristic used to reduce the number of chunks balanced on
3580 * resume after balance was interrupted.
3582 static void update_balance_args(struct btrfs_balance_control *bctl)
3585 * Turn on soft mode for chunk types that were being converted.
3587 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3588 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3589 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3590 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3591 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3592 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3595 * Turn on usage filter if is not already used. The idea is
3596 * that chunks that we have already balanced should be
3597 * reasonably full. Don't do it for chunks that are being
3598 * converted - that will keep us from relocating unconverted
3599 * (albeit full) chunks.
3601 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3602 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3603 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3604 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3605 bctl->data.usage = 90;
3607 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3608 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3609 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3610 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3611 bctl->sys.usage = 90;
3613 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3614 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3615 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3616 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3617 bctl->meta.usage = 90;
3622 * Clear the balance status in fs_info and delete the balance item from disk.
3624 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3626 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3629 BUG_ON(!fs_info->balance_ctl);
3631 spin_lock(&fs_info->balance_lock);
3632 fs_info->balance_ctl = NULL;
3633 spin_unlock(&fs_info->balance_lock);
3636 ret = del_balance_item(fs_info);
3638 btrfs_handle_fs_error(fs_info, ret, NULL);
3642 * Balance filters. Return 1 if chunk should be filtered out
3643 * (should not be balanced).
3645 static int chunk_profiles_filter(u64 chunk_type,
3646 struct btrfs_balance_args *bargs)
3648 chunk_type = chunk_to_extended(chunk_type) &
3649 BTRFS_EXTENDED_PROFILE_MASK;
3651 if (bargs->profiles & chunk_type)
3657 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3658 struct btrfs_balance_args *bargs)
3660 struct btrfs_block_group *cache;
3662 u64 user_thresh_min;
3663 u64 user_thresh_max;
3666 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3667 chunk_used = cache->used;
3669 if (bargs->usage_min == 0)
3670 user_thresh_min = 0;
3672 user_thresh_min = mult_perc(cache->length, bargs->usage_min);
3674 if (bargs->usage_max == 0)
3675 user_thresh_max = 1;
3676 else if (bargs->usage_max > 100)
3677 user_thresh_max = cache->length;
3679 user_thresh_max = mult_perc(cache->length, bargs->usage_max);
3681 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3684 btrfs_put_block_group(cache);
3688 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3689 u64 chunk_offset, struct btrfs_balance_args *bargs)
3691 struct btrfs_block_group *cache;
3692 u64 chunk_used, user_thresh;
3695 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3696 chunk_used = cache->used;
3698 if (bargs->usage_min == 0)
3700 else if (bargs->usage > 100)
3701 user_thresh = cache->length;
3703 user_thresh = mult_perc(cache->length, bargs->usage);
3705 if (chunk_used < user_thresh)
3708 btrfs_put_block_group(cache);
3712 static int chunk_devid_filter(struct extent_buffer *leaf,
3713 struct btrfs_chunk *chunk,
3714 struct btrfs_balance_args *bargs)
3716 struct btrfs_stripe *stripe;
3717 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3720 for (i = 0; i < num_stripes; i++) {
3721 stripe = btrfs_stripe_nr(chunk, i);
3722 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3729 static u64 calc_data_stripes(u64 type, int num_stripes)
3731 const int index = btrfs_bg_flags_to_raid_index(type);
3732 const int ncopies = btrfs_raid_array[index].ncopies;
3733 const int nparity = btrfs_raid_array[index].nparity;
3735 return (num_stripes - nparity) / ncopies;
3738 /* [pstart, pend) */
3739 static int chunk_drange_filter(struct extent_buffer *leaf,
3740 struct btrfs_chunk *chunk,
3741 struct btrfs_balance_args *bargs)
3743 struct btrfs_stripe *stripe;
3744 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3751 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3754 type = btrfs_chunk_type(leaf, chunk);
3755 factor = calc_data_stripes(type, num_stripes);
3757 for (i = 0; i < num_stripes; i++) {
3758 stripe = btrfs_stripe_nr(chunk, i);
3759 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3762 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3763 stripe_length = btrfs_chunk_length(leaf, chunk);
3764 stripe_length = div_u64(stripe_length, factor);
3766 if (stripe_offset < bargs->pend &&
3767 stripe_offset + stripe_length > bargs->pstart)
3774 /* [vstart, vend) */
3775 static int chunk_vrange_filter(struct extent_buffer *leaf,
3776 struct btrfs_chunk *chunk,
3778 struct btrfs_balance_args *bargs)
3780 if (chunk_offset < bargs->vend &&
3781 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3782 /* at least part of the chunk is inside this vrange */
3788 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3789 struct btrfs_chunk *chunk,
3790 struct btrfs_balance_args *bargs)
3792 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3794 if (bargs->stripes_min <= num_stripes
3795 && num_stripes <= bargs->stripes_max)
3801 static int chunk_soft_convert_filter(u64 chunk_type,
3802 struct btrfs_balance_args *bargs)
3804 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3807 chunk_type = chunk_to_extended(chunk_type) &
3808 BTRFS_EXTENDED_PROFILE_MASK;
3810 if (bargs->target == chunk_type)
3816 static int should_balance_chunk(struct extent_buffer *leaf,
3817 struct btrfs_chunk *chunk, u64 chunk_offset)
3819 struct btrfs_fs_info *fs_info = leaf->fs_info;
3820 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3821 struct btrfs_balance_args *bargs = NULL;
3822 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3825 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3826 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3830 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3831 bargs = &bctl->data;
3832 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3834 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3835 bargs = &bctl->meta;
3837 /* profiles filter */
3838 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3839 chunk_profiles_filter(chunk_type, bargs)) {
3844 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3845 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3847 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3848 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3853 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3854 chunk_devid_filter(leaf, chunk, bargs)) {
3858 /* drange filter, makes sense only with devid filter */
3859 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3860 chunk_drange_filter(leaf, chunk, bargs)) {
3865 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3866 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3870 /* stripes filter */
3871 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3872 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3876 /* soft profile changing mode */
3877 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3878 chunk_soft_convert_filter(chunk_type, bargs)) {
3883 * limited by count, must be the last filter
3885 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3886 if (bargs->limit == 0)
3890 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3892 * Same logic as the 'limit' filter; the minimum cannot be
3893 * determined here because we do not have the global information
3894 * about the count of all chunks that satisfy the filters.
3896 if (bargs->limit_max == 0)
3905 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3907 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3908 struct btrfs_root *chunk_root = fs_info->chunk_root;
3910 struct btrfs_chunk *chunk;
3911 struct btrfs_path *path = NULL;
3912 struct btrfs_key key;
3913 struct btrfs_key found_key;
3914 struct extent_buffer *leaf;
3917 int enospc_errors = 0;
3918 bool counting = true;
3919 /* The single value limit and min/max limits use the same bytes in the */
3920 u64 limit_data = bctl->data.limit;
3921 u64 limit_meta = bctl->meta.limit;
3922 u64 limit_sys = bctl->sys.limit;
3926 int chunk_reserved = 0;
3928 path = btrfs_alloc_path();
3934 /* zero out stat counters */
3935 spin_lock(&fs_info->balance_lock);
3936 memset(&bctl->stat, 0, sizeof(bctl->stat));
3937 spin_unlock(&fs_info->balance_lock);
3941 * The single value limit and min/max limits use the same bytes
3944 bctl->data.limit = limit_data;
3945 bctl->meta.limit = limit_meta;
3946 bctl->sys.limit = limit_sys;
3948 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3949 key.offset = (u64)-1;
3950 key.type = BTRFS_CHUNK_ITEM_KEY;
3953 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3954 atomic_read(&fs_info->balance_cancel_req)) {
3959 mutex_lock(&fs_info->reclaim_bgs_lock);
3960 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3962 mutex_unlock(&fs_info->reclaim_bgs_lock);
3967 * this shouldn't happen, it means the last relocate
3971 BUG(); /* FIXME break ? */
3973 ret = btrfs_previous_item(chunk_root, path, 0,
3974 BTRFS_CHUNK_ITEM_KEY);
3976 mutex_unlock(&fs_info->reclaim_bgs_lock);
3981 leaf = path->nodes[0];
3982 slot = path->slots[0];
3983 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3985 if (found_key.objectid != key.objectid) {
3986 mutex_unlock(&fs_info->reclaim_bgs_lock);
3990 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3991 chunk_type = btrfs_chunk_type(leaf, chunk);
3994 spin_lock(&fs_info->balance_lock);
3995 bctl->stat.considered++;
3996 spin_unlock(&fs_info->balance_lock);
3999 ret = should_balance_chunk(leaf, chunk, found_key.offset);
4001 btrfs_release_path(path);
4003 mutex_unlock(&fs_info->reclaim_bgs_lock);
4008 mutex_unlock(&fs_info->reclaim_bgs_lock);
4009 spin_lock(&fs_info->balance_lock);
4010 bctl->stat.expected++;
4011 spin_unlock(&fs_info->balance_lock);
4013 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
4015 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
4017 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
4024 * Apply limit_min filter, no need to check if the LIMITS
4025 * filter is used, limit_min is 0 by default
4027 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
4028 count_data < bctl->data.limit_min)
4029 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
4030 count_meta < bctl->meta.limit_min)
4031 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
4032 count_sys < bctl->sys.limit_min)) {
4033 mutex_unlock(&fs_info->reclaim_bgs_lock);
4037 if (!chunk_reserved) {
4039 * We may be relocating the only data chunk we have,
4040 * which could potentially end up with losing data's
4041 * raid profile, so lets allocate an empty one in
4044 ret = btrfs_may_alloc_data_chunk(fs_info,
4047 mutex_unlock(&fs_info->reclaim_bgs_lock);
4049 } else if (ret == 1) {
4054 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
4055 mutex_unlock(&fs_info->reclaim_bgs_lock);
4056 if (ret == -ENOSPC) {
4058 } else if (ret == -ETXTBSY) {
4060 "skipping relocation of block group %llu due to active swapfile",
4066 spin_lock(&fs_info->balance_lock);
4067 bctl->stat.completed++;
4068 spin_unlock(&fs_info->balance_lock);
4071 if (found_key.offset == 0)
4073 key.offset = found_key.offset - 1;
4077 btrfs_release_path(path);
4082 btrfs_free_path(path);
4083 if (enospc_errors) {
4084 btrfs_info(fs_info, "%d enospc errors during balance",
4094 * See if a given profile is valid and reduced.
4096 * @flags: profile to validate
4097 * @extended: if true @flags is treated as an extended profile
4099 static int alloc_profile_is_valid(u64 flags, int extended)
4101 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
4102 BTRFS_BLOCK_GROUP_PROFILE_MASK);
4104 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
4106 /* 1) check that all other bits are zeroed */
4110 /* 2) see if profile is reduced */
4112 return !extended; /* "0" is valid for usual profiles */
4114 return has_single_bit_set(flags);
4118 * Validate target profile against allowed profiles and return true if it's OK.
4119 * Otherwise print the error message and return false.
4121 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
4122 const struct btrfs_balance_args *bargs,
4123 u64 allowed, const char *type)
4125 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
4128 /* Profile is valid and does not have bits outside of the allowed set */
4129 if (alloc_profile_is_valid(bargs->target, 1) &&
4130 (bargs->target & ~allowed) == 0)
4133 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
4134 type, btrfs_bg_type_to_raid_name(bargs->target));
4139 * Fill @buf with textual description of balance filter flags @bargs, up to
4140 * @size_buf including the terminating null. The output may be trimmed if it
4141 * does not fit into the provided buffer.
4143 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
4147 u32 size_bp = size_buf;
4149 u64 flags = bargs->flags;
4150 char tmp_buf[128] = {'\0'};
4155 #define CHECK_APPEND_NOARG(a) \
4157 ret = snprintf(bp, size_bp, (a)); \
4158 if (ret < 0 || ret >= size_bp) \
4159 goto out_overflow; \
4164 #define CHECK_APPEND_1ARG(a, v1) \
4166 ret = snprintf(bp, size_bp, (a), (v1)); \
4167 if (ret < 0 || ret >= size_bp) \
4168 goto out_overflow; \
4173 #define CHECK_APPEND_2ARG(a, v1, v2) \
4175 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4176 if (ret < 0 || ret >= size_bp) \
4177 goto out_overflow; \
4182 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
4183 CHECK_APPEND_1ARG("convert=%s,",
4184 btrfs_bg_type_to_raid_name(bargs->target));
4186 if (flags & BTRFS_BALANCE_ARGS_SOFT)
4187 CHECK_APPEND_NOARG("soft,");
4189 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4190 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4192 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4195 if (flags & BTRFS_BALANCE_ARGS_USAGE)
4196 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4198 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4199 CHECK_APPEND_2ARG("usage=%u..%u,",
4200 bargs->usage_min, bargs->usage_max);
4202 if (flags & BTRFS_BALANCE_ARGS_DEVID)
4203 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4205 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4206 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4207 bargs->pstart, bargs->pend);
4209 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4210 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4211 bargs->vstart, bargs->vend);
4213 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4214 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4216 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4217 CHECK_APPEND_2ARG("limit=%u..%u,",
4218 bargs->limit_min, bargs->limit_max);
4220 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4221 CHECK_APPEND_2ARG("stripes=%u..%u,",
4222 bargs->stripes_min, bargs->stripes_max);
4224 #undef CHECK_APPEND_2ARG
4225 #undef CHECK_APPEND_1ARG
4226 #undef CHECK_APPEND_NOARG
4230 if (size_bp < size_buf)
4231 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4236 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4238 u32 size_buf = 1024;
4239 char tmp_buf[192] = {'\0'};
4242 u32 size_bp = size_buf;
4244 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4246 buf = kzalloc(size_buf, GFP_KERNEL);
4252 #define CHECK_APPEND_1ARG(a, v1) \
4254 ret = snprintf(bp, size_bp, (a), (v1)); \
4255 if (ret < 0 || ret >= size_bp) \
4256 goto out_overflow; \
4261 if (bctl->flags & BTRFS_BALANCE_FORCE)
4262 CHECK_APPEND_1ARG("%s", "-f ");
4264 if (bctl->flags & BTRFS_BALANCE_DATA) {
4265 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4266 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4269 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4270 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4271 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4274 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4275 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4276 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4279 #undef CHECK_APPEND_1ARG
4283 if (size_bp < size_buf)
4284 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4285 btrfs_info(fs_info, "balance: %s %s",
4286 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4287 "resume" : "start", buf);
4293 * Should be called with balance mutexe held
4295 int btrfs_balance(struct btrfs_fs_info *fs_info,
4296 struct btrfs_balance_control *bctl,
4297 struct btrfs_ioctl_balance_args *bargs)
4299 u64 meta_target, data_target;
4305 bool reducing_redundancy;
4306 bool paused = false;
4309 if (btrfs_fs_closing(fs_info) ||
4310 atomic_read(&fs_info->balance_pause_req) ||
4311 btrfs_should_cancel_balance(fs_info)) {
4316 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4317 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4321 * In case of mixed groups both data and meta should be picked,
4322 * and identical options should be given for both of them.
4324 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4325 if (mixed && (bctl->flags & allowed)) {
4326 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4327 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4328 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4330 "balance: mixed groups data and metadata options must be the same");
4337 * rw_devices will not change at the moment, device add/delete/replace
4340 num_devices = fs_info->fs_devices->rw_devices;
4343 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4344 * special bit for it, to make it easier to distinguish. Thus we need
4345 * to set it manually, or balance would refuse the profile.
4347 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4348 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4349 if (num_devices >= btrfs_raid_array[i].devs_min)
4350 allowed |= btrfs_raid_array[i].bg_flag;
4352 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4353 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4354 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4360 * Allow to reduce metadata or system integrity only if force set for
4361 * profiles with redundancy (copies, parity)
4364 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4365 if (btrfs_raid_array[i].ncopies >= 2 ||
4366 btrfs_raid_array[i].tolerated_failures >= 1)
4367 allowed |= btrfs_raid_array[i].bg_flag;
4370 seq = read_seqbegin(&fs_info->profiles_lock);
4372 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4373 (fs_info->avail_system_alloc_bits & allowed) &&
4374 !(bctl->sys.target & allowed)) ||
4375 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4376 (fs_info->avail_metadata_alloc_bits & allowed) &&
4377 !(bctl->meta.target & allowed)))
4378 reducing_redundancy = true;
4380 reducing_redundancy = false;
4382 /* if we're not converting, the target field is uninitialized */
4383 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4384 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4385 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4386 bctl->data.target : fs_info->avail_data_alloc_bits;
4387 } while (read_seqretry(&fs_info->profiles_lock, seq));
4389 if (reducing_redundancy) {
4390 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4392 "balance: force reducing metadata redundancy");
4395 "balance: reduces metadata redundancy, use --force if you want this");
4401 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4402 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4404 "balance: metadata profile %s has lower redundancy than data profile %s",
4405 btrfs_bg_type_to_raid_name(meta_target),
4406 btrfs_bg_type_to_raid_name(data_target));
4409 ret = insert_balance_item(fs_info, bctl);
4410 if (ret && ret != -EEXIST)
4413 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4414 BUG_ON(ret == -EEXIST);
4415 BUG_ON(fs_info->balance_ctl);
4416 spin_lock(&fs_info->balance_lock);
4417 fs_info->balance_ctl = bctl;
4418 spin_unlock(&fs_info->balance_lock);
4420 BUG_ON(ret != -EEXIST);
4421 spin_lock(&fs_info->balance_lock);
4422 update_balance_args(bctl);
4423 spin_unlock(&fs_info->balance_lock);
4426 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4427 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4428 describe_balance_start_or_resume(fs_info);
4429 mutex_unlock(&fs_info->balance_mutex);
4431 ret = __btrfs_balance(fs_info);
4433 mutex_lock(&fs_info->balance_mutex);
4434 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) {
4435 btrfs_info(fs_info, "balance: paused");
4436 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
4440 * Balance can be canceled by:
4442 * - Regular cancel request
4443 * Then ret == -ECANCELED and balance_cancel_req > 0
4445 * - Fatal signal to "btrfs" process
4446 * Either the signal caught by wait_reserve_ticket() and callers
4447 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4449 * Either way, in this case balance_cancel_req = 0, and
4450 * ret == -EINTR or ret == -ECANCELED.
4452 * So here we only check the return value to catch canceled balance.
4454 else if (ret == -ECANCELED || ret == -EINTR)
4455 btrfs_info(fs_info, "balance: canceled");
4457 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4459 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4462 memset(bargs, 0, sizeof(*bargs));
4463 btrfs_update_ioctl_balance_args(fs_info, bargs);
4466 /* We didn't pause, we can clean everything up. */
4468 reset_balance_state(fs_info);
4469 btrfs_exclop_finish(fs_info);
4472 wake_up(&fs_info->balance_wait_q);
4476 if (bctl->flags & BTRFS_BALANCE_RESUME)
4477 reset_balance_state(fs_info);
4480 btrfs_exclop_finish(fs_info);
4485 static int balance_kthread(void *data)
4487 struct btrfs_fs_info *fs_info = data;
4490 sb_start_write(fs_info->sb);
4491 mutex_lock(&fs_info->balance_mutex);
4492 if (fs_info->balance_ctl)
4493 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4494 mutex_unlock(&fs_info->balance_mutex);
4495 sb_end_write(fs_info->sb);
4500 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4502 struct task_struct *tsk;
4504 mutex_lock(&fs_info->balance_mutex);
4505 if (!fs_info->balance_ctl) {
4506 mutex_unlock(&fs_info->balance_mutex);
4509 mutex_unlock(&fs_info->balance_mutex);
4511 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4512 btrfs_info(fs_info, "balance: resume skipped");
4516 spin_lock(&fs_info->super_lock);
4517 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
4518 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;
4519 spin_unlock(&fs_info->super_lock);
4521 * A ro->rw remount sequence should continue with the paused balance
4522 * regardless of who pauses it, system or the user as of now, so set
4525 spin_lock(&fs_info->balance_lock);
4526 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4527 spin_unlock(&fs_info->balance_lock);
4529 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4530 return PTR_ERR_OR_ZERO(tsk);
4533 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4535 struct btrfs_balance_control *bctl;
4536 struct btrfs_balance_item *item;
4537 struct btrfs_disk_balance_args disk_bargs;
4538 struct btrfs_path *path;
4539 struct extent_buffer *leaf;
4540 struct btrfs_key key;
4543 path = btrfs_alloc_path();
4547 key.objectid = BTRFS_BALANCE_OBJECTID;
4548 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4551 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4554 if (ret > 0) { /* ret = -ENOENT; */
4559 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4565 leaf = path->nodes[0];
4566 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4568 bctl->flags = btrfs_balance_flags(leaf, item);
4569 bctl->flags |= BTRFS_BALANCE_RESUME;
4571 btrfs_balance_data(leaf, item, &disk_bargs);
4572 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4573 btrfs_balance_meta(leaf, item, &disk_bargs);
4574 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4575 btrfs_balance_sys(leaf, item, &disk_bargs);
4576 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4579 * This should never happen, as the paused balance state is recovered
4580 * during mount without any chance of other exclusive ops to collide.
4582 * This gives the exclusive op status to balance and keeps in paused
4583 * state until user intervention (cancel or umount). If the ownership
4584 * cannot be assigned, show a message but do not fail. The balance
4585 * is in a paused state and must have fs_info::balance_ctl properly
4588 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED))
4590 "balance: cannot set exclusive op status, resume manually");
4592 btrfs_release_path(path);
4594 mutex_lock(&fs_info->balance_mutex);
4595 BUG_ON(fs_info->balance_ctl);
4596 spin_lock(&fs_info->balance_lock);
4597 fs_info->balance_ctl = bctl;
4598 spin_unlock(&fs_info->balance_lock);
4599 mutex_unlock(&fs_info->balance_mutex);
4601 btrfs_free_path(path);
4605 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4609 mutex_lock(&fs_info->balance_mutex);
4610 if (!fs_info->balance_ctl) {
4611 mutex_unlock(&fs_info->balance_mutex);
4615 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4616 atomic_inc(&fs_info->balance_pause_req);
4617 mutex_unlock(&fs_info->balance_mutex);
4619 wait_event(fs_info->balance_wait_q,
4620 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4622 mutex_lock(&fs_info->balance_mutex);
4623 /* we are good with balance_ctl ripped off from under us */
4624 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4625 atomic_dec(&fs_info->balance_pause_req);
4630 mutex_unlock(&fs_info->balance_mutex);
4634 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4636 mutex_lock(&fs_info->balance_mutex);
4637 if (!fs_info->balance_ctl) {
4638 mutex_unlock(&fs_info->balance_mutex);
4643 * A paused balance with the item stored on disk can be resumed at
4644 * mount time if the mount is read-write. Otherwise it's still paused
4645 * and we must not allow cancelling as it deletes the item.
4647 if (sb_rdonly(fs_info->sb)) {
4648 mutex_unlock(&fs_info->balance_mutex);
4652 atomic_inc(&fs_info->balance_cancel_req);
4654 * if we are running just wait and return, balance item is
4655 * deleted in btrfs_balance in this case
4657 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4658 mutex_unlock(&fs_info->balance_mutex);
4659 wait_event(fs_info->balance_wait_q,
4660 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4661 mutex_lock(&fs_info->balance_mutex);
4663 mutex_unlock(&fs_info->balance_mutex);
4665 * Lock released to allow other waiters to continue, we'll
4666 * reexamine the status again.
4668 mutex_lock(&fs_info->balance_mutex);
4670 if (fs_info->balance_ctl) {
4671 reset_balance_state(fs_info);
4672 btrfs_exclop_finish(fs_info);
4673 btrfs_info(fs_info, "balance: canceled");
4677 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4678 atomic_dec(&fs_info->balance_cancel_req);
4679 mutex_unlock(&fs_info->balance_mutex);
4683 int btrfs_uuid_scan_kthread(void *data)
4685 struct btrfs_fs_info *fs_info = data;
4686 struct btrfs_root *root = fs_info->tree_root;
4687 struct btrfs_key key;
4688 struct btrfs_path *path = NULL;
4690 struct extent_buffer *eb;
4692 struct btrfs_root_item root_item;
4694 struct btrfs_trans_handle *trans = NULL;
4695 bool closing = false;
4697 path = btrfs_alloc_path();
4704 key.type = BTRFS_ROOT_ITEM_KEY;
4708 if (btrfs_fs_closing(fs_info)) {
4712 ret = btrfs_search_forward(root, &key, path,
4713 BTRFS_OLDEST_GENERATION);
4720 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4721 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4722 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4723 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4726 eb = path->nodes[0];
4727 slot = path->slots[0];
4728 item_size = btrfs_item_size(eb, slot);
4729 if (item_size < sizeof(root_item))
4732 read_extent_buffer(eb, &root_item,
4733 btrfs_item_ptr_offset(eb, slot),
4734 (int)sizeof(root_item));
4735 if (btrfs_root_refs(&root_item) == 0)
4738 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4739 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4743 btrfs_release_path(path);
4745 * 1 - subvol uuid item
4746 * 1 - received_subvol uuid item
4748 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4749 if (IS_ERR(trans)) {
4750 ret = PTR_ERR(trans);
4758 btrfs_release_path(path);
4759 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4760 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4761 BTRFS_UUID_KEY_SUBVOL,
4764 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4770 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4771 ret = btrfs_uuid_tree_add(trans,
4772 root_item.received_uuid,
4773 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4776 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4783 btrfs_release_path(path);
4785 ret = btrfs_end_transaction(trans);
4791 if (key.offset < (u64)-1) {
4793 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4795 key.type = BTRFS_ROOT_ITEM_KEY;
4796 } else if (key.objectid < (u64)-1) {
4798 key.type = BTRFS_ROOT_ITEM_KEY;
4807 btrfs_free_path(path);
4808 if (trans && !IS_ERR(trans))
4809 btrfs_end_transaction(trans);
4811 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4813 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4814 up(&fs_info->uuid_tree_rescan_sem);
4818 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4820 struct btrfs_trans_handle *trans;
4821 struct btrfs_root *tree_root = fs_info->tree_root;
4822 struct btrfs_root *uuid_root;
4823 struct task_struct *task;
4830 trans = btrfs_start_transaction(tree_root, 2);
4832 return PTR_ERR(trans);
4834 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4835 if (IS_ERR(uuid_root)) {
4836 ret = PTR_ERR(uuid_root);
4837 btrfs_abort_transaction(trans, ret);
4838 btrfs_end_transaction(trans);
4842 fs_info->uuid_root = uuid_root;
4844 ret = btrfs_commit_transaction(trans);
4848 down(&fs_info->uuid_tree_rescan_sem);
4849 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4851 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4852 btrfs_warn(fs_info, "failed to start uuid_scan task");
4853 up(&fs_info->uuid_tree_rescan_sem);
4854 return PTR_ERR(task);
4861 * shrinking a device means finding all of the device extents past
4862 * the new size, and then following the back refs to the chunks.
4863 * The chunk relocation code actually frees the device extent
4865 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4867 struct btrfs_fs_info *fs_info = device->fs_info;
4868 struct btrfs_root *root = fs_info->dev_root;
4869 struct btrfs_trans_handle *trans;
4870 struct btrfs_dev_extent *dev_extent = NULL;
4871 struct btrfs_path *path;
4877 bool retried = false;
4878 struct extent_buffer *l;
4879 struct btrfs_key key;
4880 struct btrfs_super_block *super_copy = fs_info->super_copy;
4881 u64 old_total = btrfs_super_total_bytes(super_copy);
4882 u64 old_size = btrfs_device_get_total_bytes(device);
4887 new_size = round_down(new_size, fs_info->sectorsize);
4889 diff = round_down(old_size - new_size, fs_info->sectorsize);
4891 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4894 path = btrfs_alloc_path();
4898 path->reada = READA_BACK;
4900 trans = btrfs_start_transaction(root, 0);
4901 if (IS_ERR(trans)) {
4902 btrfs_free_path(path);
4903 return PTR_ERR(trans);
4906 mutex_lock(&fs_info->chunk_mutex);
4908 btrfs_device_set_total_bytes(device, new_size);
4909 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4910 device->fs_devices->total_rw_bytes -= diff;
4913 * The new free_chunk_space is new_size - used, so we have to
4914 * subtract the delta of the old free_chunk_space which included
4915 * old_size - used. If used > new_size then just subtract this
4916 * entire device's free space.
4918 if (device->bytes_used < new_size)
4919 free_diff = (old_size - device->bytes_used) -
4920 (new_size - device->bytes_used);
4922 free_diff = old_size - device->bytes_used;
4923 atomic64_sub(free_diff, &fs_info->free_chunk_space);
4927 * Once the device's size has been set to the new size, ensure all
4928 * in-memory chunks are synced to disk so that the loop below sees them
4929 * and relocates them accordingly.
4931 if (contains_pending_extent(device, &start, diff)) {
4932 mutex_unlock(&fs_info->chunk_mutex);
4933 ret = btrfs_commit_transaction(trans);
4937 mutex_unlock(&fs_info->chunk_mutex);
4938 btrfs_end_transaction(trans);
4942 key.objectid = device->devid;
4943 key.offset = (u64)-1;
4944 key.type = BTRFS_DEV_EXTENT_KEY;
4947 mutex_lock(&fs_info->reclaim_bgs_lock);
4948 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4950 mutex_unlock(&fs_info->reclaim_bgs_lock);
4954 ret = btrfs_previous_item(root, path, 0, key.type);
4956 mutex_unlock(&fs_info->reclaim_bgs_lock);
4960 btrfs_release_path(path);
4965 slot = path->slots[0];
4966 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4968 if (key.objectid != device->devid) {
4969 mutex_unlock(&fs_info->reclaim_bgs_lock);
4970 btrfs_release_path(path);
4974 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4975 length = btrfs_dev_extent_length(l, dev_extent);
4977 if (key.offset + length <= new_size) {
4978 mutex_unlock(&fs_info->reclaim_bgs_lock);
4979 btrfs_release_path(path);
4983 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4984 btrfs_release_path(path);
4987 * We may be relocating the only data chunk we have,
4988 * which could potentially end up with losing data's
4989 * raid profile, so lets allocate an empty one in
4992 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4994 mutex_unlock(&fs_info->reclaim_bgs_lock);
4998 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4999 mutex_unlock(&fs_info->reclaim_bgs_lock);
5000 if (ret == -ENOSPC) {
5003 if (ret == -ETXTBSY) {
5005 "could not shrink block group %llu due to active swapfile",
5010 } while (key.offset-- > 0);
5012 if (failed && !retried) {
5016 } else if (failed && retried) {
5021 /* Shrinking succeeded, else we would be at "done". */
5022 trans = btrfs_start_transaction(root, 0);
5023 if (IS_ERR(trans)) {
5024 ret = PTR_ERR(trans);
5028 mutex_lock(&fs_info->chunk_mutex);
5029 /* Clear all state bits beyond the shrunk device size */
5030 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
5033 btrfs_device_set_disk_total_bytes(device, new_size);
5034 if (list_empty(&device->post_commit_list))
5035 list_add_tail(&device->post_commit_list,
5036 &trans->transaction->dev_update_list);
5038 WARN_ON(diff > old_total);
5039 btrfs_set_super_total_bytes(super_copy,
5040 round_down(old_total - diff, fs_info->sectorsize));
5041 mutex_unlock(&fs_info->chunk_mutex);
5043 btrfs_reserve_chunk_metadata(trans, false);
5044 /* Now btrfs_update_device() will change the on-disk size. */
5045 ret = btrfs_update_device(trans, device);
5046 btrfs_trans_release_chunk_metadata(trans);
5048 btrfs_abort_transaction(trans, ret);
5049 btrfs_end_transaction(trans);
5051 ret = btrfs_commit_transaction(trans);
5054 btrfs_free_path(path);
5056 mutex_lock(&fs_info->chunk_mutex);
5057 btrfs_device_set_total_bytes(device, old_size);
5058 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5059 device->fs_devices->total_rw_bytes += diff;
5060 atomic64_add(free_diff, &fs_info->free_chunk_space);
5062 mutex_unlock(&fs_info->chunk_mutex);
5067 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
5068 struct btrfs_key *key,
5069 struct btrfs_chunk *chunk, int item_size)
5071 struct btrfs_super_block *super_copy = fs_info->super_copy;
5072 struct btrfs_disk_key disk_key;
5076 lockdep_assert_held(&fs_info->chunk_mutex);
5078 array_size = btrfs_super_sys_array_size(super_copy);
5079 if (array_size + item_size + sizeof(disk_key)
5080 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
5083 ptr = super_copy->sys_chunk_array + array_size;
5084 btrfs_cpu_key_to_disk(&disk_key, key);
5085 memcpy(ptr, &disk_key, sizeof(disk_key));
5086 ptr += sizeof(disk_key);
5087 memcpy(ptr, chunk, item_size);
5088 item_size += sizeof(disk_key);
5089 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
5095 * sort the devices in descending order by max_avail, total_avail
5097 static int btrfs_cmp_device_info(const void *a, const void *b)
5099 const struct btrfs_device_info *di_a = a;
5100 const struct btrfs_device_info *di_b = b;
5102 if (di_a->max_avail > di_b->max_avail)
5104 if (di_a->max_avail < di_b->max_avail)
5106 if (di_a->total_avail > di_b->total_avail)
5108 if (di_a->total_avail < di_b->total_avail)
5113 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
5115 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5118 btrfs_set_fs_incompat(info, RAID56);
5121 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
5123 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
5126 btrfs_set_fs_incompat(info, RAID1C34);
5130 * Structure used internally for btrfs_create_chunk() function.
5131 * Wraps needed parameters.
5133 struct alloc_chunk_ctl {
5136 /* Total number of stripes to allocate */
5138 /* sub_stripes info for map */
5140 /* Stripes per device */
5142 /* Maximum number of devices to use */
5144 /* Minimum number of devices to use */
5146 /* ndevs has to be a multiple of this */
5148 /* Number of copies */
5150 /* Number of stripes worth of bytes to store parity information */
5152 u64 max_stripe_size;
5160 static void init_alloc_chunk_ctl_policy_regular(
5161 struct btrfs_fs_devices *fs_devices,
5162 struct alloc_chunk_ctl *ctl)
5164 struct btrfs_space_info *space_info;
5166 space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type);
5169 ctl->max_chunk_size = READ_ONCE(space_info->chunk_size);
5170 ctl->max_stripe_size = min_t(u64, ctl->max_chunk_size, SZ_1G);
5172 if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM)
5173 ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK);
5175 /* We don't want a chunk larger than 10% of writable space */
5176 ctl->max_chunk_size = min(mult_perc(fs_devices->total_rw_bytes, 10),
5177 ctl->max_chunk_size);
5178 ctl->dev_extent_min = btrfs_stripe_nr_to_offset(ctl->dev_stripes);
5181 static void init_alloc_chunk_ctl_policy_zoned(
5182 struct btrfs_fs_devices *fs_devices,
5183 struct alloc_chunk_ctl *ctl)
5185 u64 zone_size = fs_devices->fs_info->zone_size;
5187 int min_num_stripes = ctl->devs_min * ctl->dev_stripes;
5188 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;
5189 u64 min_chunk_size = min_data_stripes * zone_size;
5190 u64 type = ctl->type;
5192 ctl->max_stripe_size = zone_size;
5193 if (type & BTRFS_BLOCK_GROUP_DATA) {
5194 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,
5196 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5197 ctl->max_chunk_size = ctl->max_stripe_size;
5198 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5199 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5200 ctl->devs_max = min_t(int, ctl->devs_max,
5201 BTRFS_MAX_DEVS_SYS_CHUNK);
5206 /* We don't want a chunk larger than 10% of writable space */
5207 limit = max(round_down(mult_perc(fs_devices->total_rw_bytes, 10),
5210 ctl->max_chunk_size = min(limit, ctl->max_chunk_size);
5211 ctl->dev_extent_min = zone_size * ctl->dev_stripes;
5214 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
5215 struct alloc_chunk_ctl *ctl)
5217 int index = btrfs_bg_flags_to_raid_index(ctl->type);
5219 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
5220 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
5221 ctl->devs_max = btrfs_raid_array[index].devs_max;
5223 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
5224 ctl->devs_min = btrfs_raid_array[index].devs_min;
5225 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
5226 ctl->ncopies = btrfs_raid_array[index].ncopies;
5227 ctl->nparity = btrfs_raid_array[index].nparity;
5230 switch (fs_devices->chunk_alloc_policy) {
5231 case BTRFS_CHUNK_ALLOC_REGULAR:
5232 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
5234 case BTRFS_CHUNK_ALLOC_ZONED:
5235 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);
5242 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
5243 struct alloc_chunk_ctl *ctl,
5244 struct btrfs_device_info *devices_info)
5246 struct btrfs_fs_info *info = fs_devices->fs_info;
5247 struct btrfs_device *device;
5249 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
5256 * in the first pass through the devices list, we gather information
5257 * about the available holes on each device.
5259 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5260 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5262 "BTRFS: read-only device in alloc_list\n");
5266 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5267 &device->dev_state) ||
5268 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5271 if (device->total_bytes > device->bytes_used)
5272 total_avail = device->total_bytes - device->bytes_used;
5276 /* If there is no space on this device, skip it. */
5277 if (total_avail < ctl->dev_extent_min)
5280 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
5282 if (ret && ret != -ENOSPC)
5286 max_avail = dev_extent_want;
5288 if (max_avail < ctl->dev_extent_min) {
5289 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5291 "%s: devid %llu has no free space, have=%llu want=%llu",
5292 __func__, device->devid, max_avail,
5293 ctl->dev_extent_min);
5297 if (ndevs == fs_devices->rw_devices) {
5298 WARN(1, "%s: found more than %llu devices\n",
5299 __func__, fs_devices->rw_devices);
5302 devices_info[ndevs].dev_offset = dev_offset;
5303 devices_info[ndevs].max_avail = max_avail;
5304 devices_info[ndevs].total_avail = total_avail;
5305 devices_info[ndevs].dev = device;
5311 * now sort the devices by hole size / available space
5313 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5314 btrfs_cmp_device_info, NULL);
5319 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5320 struct btrfs_device_info *devices_info)
5322 /* Number of stripes that count for block group size */
5326 * The primary goal is to maximize the number of stripes, so use as
5327 * many devices as possible, even if the stripes are not maximum sized.
5329 * The DUP profile stores more than one stripe per device, the
5330 * max_avail is the total size so we have to adjust.
5332 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5334 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5336 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5337 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5340 * Use the number of data stripes to figure out how big this chunk is
5341 * really going to be in terms of logical address space, and compare
5342 * that answer with the max chunk size. If it's higher, we try to
5343 * reduce stripe_size.
5345 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5347 * Reduce stripe_size, round it up to a 16MB boundary again and
5348 * then use it, unless it ends up being even bigger than the
5349 * previous value we had already.
5351 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5352 data_stripes), SZ_16M),
5356 /* Stripe size should not go beyond 1G. */
5357 ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G);
5359 /* Align to BTRFS_STRIPE_LEN */
5360 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5361 ctl->chunk_size = ctl->stripe_size * data_stripes;
5366 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,
5367 struct btrfs_device_info *devices_info)
5369 u64 zone_size = devices_info[0].dev->zone_info->zone_size;
5370 /* Number of stripes that count for block group size */
5374 * It should hold because:
5375 * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5377 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);
5379 ctl->stripe_size = zone_size;
5380 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5381 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5383 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5384 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5385 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,
5386 ctl->stripe_size) + ctl->nparity,
5388 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5389 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5390 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);
5393 ctl->chunk_size = ctl->stripe_size * data_stripes;
5398 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5399 struct alloc_chunk_ctl *ctl,
5400 struct btrfs_device_info *devices_info)
5402 struct btrfs_fs_info *info = fs_devices->fs_info;
5405 * Round down to number of usable stripes, devs_increment can be any
5406 * number so we can't use round_down() that requires power of 2, while
5407 * rounddown is safe.
5409 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5411 if (ctl->ndevs < ctl->devs_min) {
5412 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5414 "%s: not enough devices with free space: have=%d minimum required=%d",
5415 __func__, ctl->ndevs, ctl->devs_min);
5420 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5422 switch (fs_devices->chunk_alloc_policy) {
5423 case BTRFS_CHUNK_ALLOC_REGULAR:
5424 return decide_stripe_size_regular(ctl, devices_info);
5425 case BTRFS_CHUNK_ALLOC_ZONED:
5426 return decide_stripe_size_zoned(ctl, devices_info);
5432 static void chunk_map_device_set_bits(struct btrfs_chunk_map *map, unsigned int bits)
5434 for (int i = 0; i < map->num_stripes; i++) {
5435 struct btrfs_io_stripe *stripe = &map->stripes[i];
5436 struct btrfs_device *device = stripe->dev;
5438 set_extent_bit(&device->alloc_state, stripe->physical,
5439 stripe->physical + map->stripe_size - 1,
5440 bits | EXTENT_NOWAIT, NULL);
5444 static void chunk_map_device_clear_bits(struct btrfs_chunk_map *map, unsigned int bits)
5446 for (int i = 0; i < map->num_stripes; i++) {
5447 struct btrfs_io_stripe *stripe = &map->stripes[i];
5448 struct btrfs_device *device = stripe->dev;
5450 __clear_extent_bit(&device->alloc_state, stripe->physical,
5451 stripe->physical + map->stripe_size - 1,
5452 bits | EXTENT_NOWAIT,
5457 void btrfs_remove_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map)
5459 write_lock(&fs_info->mapping_tree_lock);
5460 rb_erase_cached(&map->rb_node, &fs_info->mapping_tree);
5461 RB_CLEAR_NODE(&map->rb_node);
5462 chunk_map_device_clear_bits(map, CHUNK_ALLOCATED);
5463 write_unlock(&fs_info->mapping_tree_lock);
5465 /* Once for the tree reference. */
5466 btrfs_free_chunk_map(map);
5470 int btrfs_add_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map)
5473 struct rb_node *parent = NULL;
5474 bool leftmost = true;
5476 write_lock(&fs_info->mapping_tree_lock);
5477 p = &fs_info->mapping_tree.rb_root.rb_node;
5479 struct btrfs_chunk_map *entry;
5482 entry = rb_entry(parent, struct btrfs_chunk_map, rb_node);
5484 if (map->start < entry->start) {
5486 } else if (map->start > entry->start) {
5487 p = &(*p)->rb_right;
5490 write_unlock(&fs_info->mapping_tree_lock);
5494 rb_link_node(&map->rb_node, parent, p);
5495 rb_insert_color_cached(&map->rb_node, &fs_info->mapping_tree, leftmost);
5496 chunk_map_device_set_bits(map, CHUNK_ALLOCATED);
5497 chunk_map_device_clear_bits(map, CHUNK_TRIMMED);
5498 write_unlock(&fs_info->mapping_tree_lock);
5504 struct btrfs_chunk_map *btrfs_alloc_chunk_map(int num_stripes, gfp_t gfp)
5506 struct btrfs_chunk_map *map;
5508 map = kmalloc(btrfs_chunk_map_size(num_stripes), gfp);
5512 refcount_set(&map->refs, 1);
5513 RB_CLEAR_NODE(&map->rb_node);
5518 struct btrfs_chunk_map *btrfs_clone_chunk_map(struct btrfs_chunk_map *map, gfp_t gfp)
5520 const int size = btrfs_chunk_map_size(map->num_stripes);
5521 struct btrfs_chunk_map *clone;
5523 clone = kmemdup(map, size, gfp);
5527 refcount_set(&clone->refs, 1);
5528 RB_CLEAR_NODE(&clone->rb_node);
5533 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
5534 struct alloc_chunk_ctl *ctl,
5535 struct btrfs_device_info *devices_info)
5537 struct btrfs_fs_info *info = trans->fs_info;
5538 struct btrfs_chunk_map *map;
5539 struct btrfs_block_group *block_group;
5540 u64 start = ctl->start;
5541 u64 type = ctl->type;
5546 map = btrfs_alloc_chunk_map(ctl->num_stripes, GFP_NOFS);
5548 return ERR_PTR(-ENOMEM);
5551 map->chunk_len = ctl->chunk_size;
5552 map->stripe_size = ctl->stripe_size;
5554 map->io_align = BTRFS_STRIPE_LEN;
5555 map->io_width = BTRFS_STRIPE_LEN;
5556 map->sub_stripes = ctl->sub_stripes;
5557 map->num_stripes = ctl->num_stripes;
5559 for (i = 0; i < ctl->ndevs; ++i) {
5560 for (j = 0; j < ctl->dev_stripes; ++j) {
5561 int s = i * ctl->dev_stripes + j;
5562 map->stripes[s].dev = devices_info[i].dev;
5563 map->stripes[s].physical = devices_info[i].dev_offset +
5564 j * ctl->stripe_size;
5568 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5570 ret = btrfs_add_chunk_map(info, map);
5572 btrfs_free_chunk_map(map);
5573 return ERR_PTR(ret);
5576 block_group = btrfs_make_block_group(trans, type, start, ctl->chunk_size);
5577 if (IS_ERR(block_group)) {
5578 btrfs_remove_chunk_map(info, map);
5582 for (int i = 0; i < map->num_stripes; i++) {
5583 struct btrfs_device *dev = map->stripes[i].dev;
5585 btrfs_device_set_bytes_used(dev,
5586 dev->bytes_used + ctl->stripe_size);
5587 if (list_empty(&dev->post_commit_list))
5588 list_add_tail(&dev->post_commit_list,
5589 &trans->transaction->dev_update_list);
5592 atomic64_sub(ctl->stripe_size * map->num_stripes,
5593 &info->free_chunk_space);
5595 check_raid56_incompat_flag(info, type);
5596 check_raid1c34_incompat_flag(info, type);
5601 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
5604 struct btrfs_fs_info *info = trans->fs_info;
5605 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5606 struct btrfs_device_info *devices_info = NULL;
5607 struct alloc_chunk_ctl ctl;
5608 struct btrfs_block_group *block_group;
5611 lockdep_assert_held(&info->chunk_mutex);
5613 if (!alloc_profile_is_valid(type, 0)) {
5615 return ERR_PTR(-EINVAL);
5618 if (list_empty(&fs_devices->alloc_list)) {
5619 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5620 btrfs_debug(info, "%s: no writable device", __func__);
5621 return ERR_PTR(-ENOSPC);
5624 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5625 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5627 return ERR_PTR(-EINVAL);
5630 ctl.start = find_next_chunk(info);
5632 init_alloc_chunk_ctl(fs_devices, &ctl);
5634 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5637 return ERR_PTR(-ENOMEM);
5639 ret = gather_device_info(fs_devices, &ctl, devices_info);
5641 block_group = ERR_PTR(ret);
5645 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5647 block_group = ERR_PTR(ret);
5651 block_group = create_chunk(trans, &ctl, devices_info);
5654 kfree(devices_info);
5659 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5660 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5663 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5666 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
5667 struct btrfs_block_group *bg)
5669 struct btrfs_fs_info *fs_info = trans->fs_info;
5670 struct btrfs_root *chunk_root = fs_info->chunk_root;
5671 struct btrfs_key key;
5672 struct btrfs_chunk *chunk;
5673 struct btrfs_stripe *stripe;
5674 struct btrfs_chunk_map *map;
5680 * We take the chunk_mutex for 2 reasons:
5682 * 1) Updates and insertions in the chunk btree must be done while holding
5683 * the chunk_mutex, as well as updating the system chunk array in the
5684 * superblock. See the comment on top of btrfs_chunk_alloc() for the
5687 * 2) To prevent races with the final phase of a device replace operation
5688 * that replaces the device object associated with the map's stripes,
5689 * because the device object's id can change at any time during that
5690 * final phase of the device replace operation
5691 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5692 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5693 * which would cause a failure when updating the device item, which does
5694 * not exists, or persisting a stripe of the chunk item with such ID.
5695 * Here we can't use the device_list_mutex because our caller already
5696 * has locked the chunk_mutex, and the final phase of device replace
5697 * acquires both mutexes - first the device_list_mutex and then the
5698 * chunk_mutex. Using any of those two mutexes protects us from a
5699 * concurrent device replace.
5701 lockdep_assert_held(&fs_info->chunk_mutex);
5703 map = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
5706 btrfs_abort_transaction(trans, ret);
5710 item_size = btrfs_chunk_item_size(map->num_stripes);
5712 chunk = kzalloc(item_size, GFP_NOFS);
5715 btrfs_abort_transaction(trans, ret);
5719 for (i = 0; i < map->num_stripes; i++) {
5720 struct btrfs_device *device = map->stripes[i].dev;
5722 ret = btrfs_update_device(trans, device);
5727 stripe = &chunk->stripe;
5728 for (i = 0; i < map->num_stripes; i++) {
5729 struct btrfs_device *device = map->stripes[i].dev;
5730 const u64 dev_offset = map->stripes[i].physical;
5732 btrfs_set_stack_stripe_devid(stripe, device->devid);
5733 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5734 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5738 btrfs_set_stack_chunk_length(chunk, bg->length);
5739 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
5740 btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
5741 btrfs_set_stack_chunk_type(chunk, map->type);
5742 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5743 btrfs_set_stack_chunk_io_align(chunk, BTRFS_STRIPE_LEN);
5744 btrfs_set_stack_chunk_io_width(chunk, BTRFS_STRIPE_LEN);
5745 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5746 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5748 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5749 key.type = BTRFS_CHUNK_ITEM_KEY;
5750 key.offset = bg->start;
5752 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5756 set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags);
5758 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5759 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5766 btrfs_free_chunk_map(map);
5770 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5772 struct btrfs_fs_info *fs_info = trans->fs_info;
5774 struct btrfs_block_group *meta_bg;
5775 struct btrfs_block_group *sys_bg;
5778 * When adding a new device for sprouting, the seed device is read-only
5779 * so we must first allocate a metadata and a system chunk. But before
5780 * adding the block group items to the extent, device and chunk btrees,
5783 * 1) Create both chunks without doing any changes to the btrees, as
5784 * otherwise we would get -ENOSPC since the block groups from the
5785 * seed device are read-only;
5787 * 2) Add the device item for the new sprout device - finishing the setup
5788 * of a new block group requires updating the device item in the chunk
5789 * btree, so it must exist when we attempt to do it. The previous step
5790 * ensures this does not fail with -ENOSPC.
5792 * After that we can add the block group items to their btrees:
5793 * update existing device item in the chunk btree, add a new block group
5794 * item to the extent btree, add a new chunk item to the chunk btree and
5795 * finally add the new device extent items to the devices btree.
5798 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5799 meta_bg = btrfs_create_chunk(trans, alloc_profile);
5800 if (IS_ERR(meta_bg))
5801 return PTR_ERR(meta_bg);
5803 alloc_profile = btrfs_system_alloc_profile(fs_info);
5804 sys_bg = btrfs_create_chunk(trans, alloc_profile);
5806 return PTR_ERR(sys_bg);
5811 static inline int btrfs_chunk_max_errors(struct btrfs_chunk_map *map)
5813 const int index = btrfs_bg_flags_to_raid_index(map->type);
5815 return btrfs_raid_array[index].tolerated_failures;
5818 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5820 struct btrfs_chunk_map *map;
5825 map = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5829 for (i = 0; i < map->num_stripes; i++) {
5830 if (test_bit(BTRFS_DEV_STATE_MISSING,
5831 &map->stripes[i].dev->dev_state)) {
5835 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5836 &map->stripes[i].dev->dev_state)) {
5843 * If the number of missing devices is larger than max errors, we can
5844 * not write the data into that chunk successfully.
5846 if (miss_ndevs > btrfs_chunk_max_errors(map))
5849 btrfs_free_chunk_map(map);
5853 void btrfs_mapping_tree_free(struct btrfs_fs_info *fs_info)
5855 write_lock(&fs_info->mapping_tree_lock);
5856 while (!RB_EMPTY_ROOT(&fs_info->mapping_tree.rb_root)) {
5857 struct btrfs_chunk_map *map;
5858 struct rb_node *node;
5860 node = rb_first_cached(&fs_info->mapping_tree);
5861 map = rb_entry(node, struct btrfs_chunk_map, rb_node);
5862 rb_erase_cached(&map->rb_node, &fs_info->mapping_tree);
5863 RB_CLEAR_NODE(&map->rb_node);
5864 chunk_map_device_clear_bits(map, CHUNK_ALLOCATED);
5865 /* Once for the tree ref. */
5866 btrfs_free_chunk_map(map);
5867 cond_resched_rwlock_write(&fs_info->mapping_tree_lock);
5869 write_unlock(&fs_info->mapping_tree_lock);
5872 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5874 struct btrfs_chunk_map *map;
5875 enum btrfs_raid_types index;
5878 map = btrfs_get_chunk_map(fs_info, logical, len);
5881 * We could return errors for these cases, but that could get
5882 * ugly and we'd probably do the same thing which is just not do
5883 * anything else and exit, so return 1 so the callers don't try
5884 * to use other copies.
5888 index = btrfs_bg_flags_to_raid_index(map->type);
5890 /* Non-RAID56, use their ncopies from btrfs_raid_array. */
5891 if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5892 ret = btrfs_raid_array[index].ncopies;
5893 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5895 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5897 * There could be two corrupted data stripes, we need
5898 * to loop retry in order to rebuild the correct data.
5900 * Fail a stripe at a time on every retry except the
5901 * stripe under reconstruction.
5903 ret = map->num_stripes;
5904 btrfs_free_chunk_map(map);
5908 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5911 struct btrfs_chunk_map *map;
5912 unsigned long len = fs_info->sectorsize;
5914 if (!btrfs_fs_incompat(fs_info, RAID56))
5917 map = btrfs_get_chunk_map(fs_info, logical, len);
5919 if (!WARN_ON(IS_ERR(map))) {
5920 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5921 len = btrfs_stripe_nr_to_offset(nr_data_stripes(map));
5922 btrfs_free_chunk_map(map);
5927 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5929 struct btrfs_chunk_map *map;
5932 if (!btrfs_fs_incompat(fs_info, RAID56))
5935 map = btrfs_get_chunk_map(fs_info, logical, len);
5937 if (!WARN_ON(IS_ERR(map))) {
5938 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5940 btrfs_free_chunk_map(map);
5945 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5946 struct btrfs_chunk_map *map, int first,
5947 int dev_replace_is_ongoing)
5951 int preferred_mirror;
5953 struct btrfs_device *srcdev;
5956 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5958 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5959 num_stripes = map->sub_stripes;
5961 num_stripes = map->num_stripes;
5963 switch (fs_info->fs_devices->read_policy) {
5965 /* Shouldn't happen, just warn and use pid instead of failing */
5966 btrfs_warn_rl(fs_info,
5967 "unknown read_policy type %u, reset to pid",
5968 fs_info->fs_devices->read_policy);
5969 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID;
5971 case BTRFS_READ_POLICY_PID:
5972 preferred_mirror = first + (current->pid % num_stripes);
5976 if (dev_replace_is_ongoing &&
5977 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5978 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5979 srcdev = fs_info->dev_replace.srcdev;
5984 * try to avoid the drive that is the source drive for a
5985 * dev-replace procedure, only choose it if no other non-missing
5986 * mirror is available
5988 for (tolerance = 0; tolerance < 2; tolerance++) {
5989 if (map->stripes[preferred_mirror].dev->bdev &&
5990 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5991 return preferred_mirror;
5992 for (i = first; i < first + num_stripes; i++) {
5993 if (map->stripes[i].dev->bdev &&
5994 (tolerance || map->stripes[i].dev != srcdev))
5999 /* we couldn't find one that doesn't fail. Just return something
6000 * and the io error handling code will clean up eventually
6002 return preferred_mirror;
6005 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info,
6009 struct btrfs_io_context *bioc;
6012 /* The size of btrfs_io_context */
6013 sizeof(struct btrfs_io_context) +
6014 /* Plus the variable array for the stripes */
6015 sizeof(struct btrfs_io_stripe) * (total_stripes),
6021 refcount_set(&bioc->refs, 1);
6023 bioc->fs_info = fs_info;
6024 bioc->replace_stripe_src = -1;
6025 bioc->full_stripe_logical = (u64)-1;
6026 bioc->logical = logical;
6031 void btrfs_get_bioc(struct btrfs_io_context *bioc)
6033 WARN_ON(!refcount_read(&bioc->refs));
6034 refcount_inc(&bioc->refs);
6037 void btrfs_put_bioc(struct btrfs_io_context *bioc)
6041 if (refcount_dec_and_test(&bioc->refs))
6046 * Please note that, discard won't be sent to target device of device
6049 struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info,
6050 u64 logical, u64 *length_ret,
6053 struct btrfs_chunk_map *map;
6054 struct btrfs_discard_stripe *stripes;
6055 u64 length = *length_ret;
6060 u64 stripe_end_offset;
6064 u32 sub_stripes = 0;
6065 u32 stripes_per_dev = 0;
6066 u32 remaining_stripes = 0;
6067 u32 last_stripe = 0;
6071 map = btrfs_get_chunk_map(fs_info, logical, length);
6073 return ERR_CAST(map);
6075 /* we don't discard raid56 yet */
6076 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6081 offset = logical - map->start;
6082 length = min_t(u64, map->start + map->chunk_len - logical, length);
6083 *length_ret = length;
6086 * stripe_nr counts the total number of stripes we have to stride
6087 * to get to this block
6089 stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
6091 /* stripe_offset is the offset of this block in its stripe */
6092 stripe_offset = offset - btrfs_stripe_nr_to_offset(stripe_nr);
6094 stripe_nr_end = round_up(offset + length, BTRFS_STRIPE_LEN) >>
6095 BTRFS_STRIPE_LEN_SHIFT;
6096 stripe_cnt = stripe_nr_end - stripe_nr;
6097 stripe_end_offset = btrfs_stripe_nr_to_offset(stripe_nr_end) -
6100 * after this, stripe_nr is the number of stripes on this
6101 * device we have to walk to find the data, and stripe_index is
6102 * the number of our device in the stripe array
6106 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6107 BTRFS_BLOCK_GROUP_RAID10)) {
6108 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6111 sub_stripes = map->sub_stripes;
6113 factor = map->num_stripes / sub_stripes;
6114 *num_stripes = min_t(u64, map->num_stripes,
6115 sub_stripes * stripe_cnt);
6116 stripe_index = stripe_nr % factor;
6117 stripe_nr /= factor;
6118 stripe_index *= sub_stripes;
6120 remaining_stripes = stripe_cnt % factor;
6121 stripes_per_dev = stripe_cnt / factor;
6122 last_stripe = ((stripe_nr_end - 1) % factor) * sub_stripes;
6123 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
6124 BTRFS_BLOCK_GROUP_DUP)) {
6125 *num_stripes = map->num_stripes;
6127 stripe_index = stripe_nr % map->num_stripes;
6128 stripe_nr /= map->num_stripes;
6131 stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS);
6137 for (i = 0; i < *num_stripes; i++) {
6138 stripes[i].physical =
6139 map->stripes[stripe_index].physical +
6140 stripe_offset + btrfs_stripe_nr_to_offset(stripe_nr);
6141 stripes[i].dev = map->stripes[stripe_index].dev;
6143 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6144 BTRFS_BLOCK_GROUP_RAID10)) {
6145 stripes[i].length = btrfs_stripe_nr_to_offset(stripes_per_dev);
6147 if (i / sub_stripes < remaining_stripes)
6148 stripes[i].length += BTRFS_STRIPE_LEN;
6151 * Special for the first stripe and
6154 * |-------|...|-------|
6158 if (i < sub_stripes)
6159 stripes[i].length -= stripe_offset;
6161 if (stripe_index >= last_stripe &&
6162 stripe_index <= (last_stripe +
6164 stripes[i].length -= stripe_end_offset;
6166 if (i == sub_stripes - 1)
6169 stripes[i].length = length;
6173 if (stripe_index == map->num_stripes) {
6179 btrfs_free_chunk_map(map);
6182 btrfs_free_chunk_map(map);
6183 return ERR_PTR(ret);
6186 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)
6188 struct btrfs_block_group *cache;
6191 /* Non zoned filesystem does not use "to_copy" flag */
6192 if (!btrfs_is_zoned(fs_info))
6195 cache = btrfs_lookup_block_group(fs_info, logical);
6197 ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags);
6199 btrfs_put_block_group(cache);
6203 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
6204 struct btrfs_io_context *bioc,
6205 struct btrfs_dev_replace *dev_replace,
6207 int *num_stripes_ret, int *max_errors_ret)
6209 u64 srcdev_devid = dev_replace->srcdev->devid;
6211 * At this stage, num_stripes is still the real number of stripes,
6212 * excluding the duplicated stripes.
6214 int num_stripes = *num_stripes_ret;
6215 int nr_extra_stripes = 0;
6216 int max_errors = *max_errors_ret;
6220 * A block group which has "to_copy" set will eventually be copied by
6221 * the dev-replace process. We can avoid cloning IO here.
6223 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))
6227 * Duplicate the write operations while the dev-replace procedure is
6228 * running. Since the copying of the old disk to the new disk takes
6229 * place at run time while the filesystem is mounted writable, the
6230 * regular write operations to the old disk have to be duplicated to go
6231 * to the new disk as well.
6233 * Note that device->missing is handled by the caller, and that the
6234 * write to the old disk is already set up in the stripes array.
6236 for (i = 0; i < num_stripes; i++) {
6237 struct btrfs_io_stripe *old = &bioc->stripes[i];
6238 struct btrfs_io_stripe *new = &bioc->stripes[num_stripes + nr_extra_stripes];
6240 if (old->dev->devid != srcdev_devid)
6243 new->physical = old->physical;
6244 new->dev = dev_replace->tgtdev;
6245 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK)
6246 bioc->replace_stripe_src = i;
6250 /* We can only have at most 2 extra nr_stripes (for DUP). */
6251 ASSERT(nr_extra_stripes <= 2);
6253 * For GET_READ_MIRRORS, we can only return at most 1 extra stripe for
6255 * If we have 2 extra stripes, only choose the one with smaller physical.
6257 if (op == BTRFS_MAP_GET_READ_MIRRORS && nr_extra_stripes == 2) {
6258 struct btrfs_io_stripe *first = &bioc->stripes[num_stripes];
6259 struct btrfs_io_stripe *second = &bioc->stripes[num_stripes + 1];
6261 /* Only DUP can have two extra stripes. */
6262 ASSERT(bioc->map_type & BTRFS_BLOCK_GROUP_DUP);
6265 * Swap the last stripe stripes and reduce @nr_extra_stripes.
6266 * The extra stripe would still be there, but won't be accessed.
6268 if (first->physical > second->physical) {
6269 swap(second->physical, first->physical);
6270 swap(second->dev, first->dev);
6275 *num_stripes_ret = num_stripes + nr_extra_stripes;
6276 *max_errors_ret = max_errors + nr_extra_stripes;
6277 bioc->replace_nr_stripes = nr_extra_stripes;
6280 static u64 btrfs_max_io_len(struct btrfs_chunk_map *map, enum btrfs_map_op op,
6281 u64 offset, u32 *stripe_nr, u64 *stripe_offset,
6282 u64 *full_stripe_start)
6285 * Stripe_nr is the stripe where this block falls. stripe_offset is
6286 * the offset of this block in its stripe.
6288 *stripe_offset = offset & BTRFS_STRIPE_LEN_MASK;
6289 *stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
6290 ASSERT(*stripe_offset < U32_MAX);
6292 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6293 unsigned long full_stripe_len =
6294 btrfs_stripe_nr_to_offset(nr_data_stripes(map));
6297 * For full stripe start, we use previously calculated
6298 * @stripe_nr. Align it to nr_data_stripes, then multiply with
6301 * By this we can avoid u64 division completely. And we have
6302 * to go rounddown(), not round_down(), as nr_data_stripes is
6303 * not ensured to be power of 2.
6305 *full_stripe_start =
6306 btrfs_stripe_nr_to_offset(
6307 rounddown(*stripe_nr, nr_data_stripes(map)));
6309 ASSERT(*full_stripe_start + full_stripe_len > offset);
6310 ASSERT(*full_stripe_start <= offset);
6312 * For writes to RAID56, allow to write a full stripe set, but
6313 * no straddling of stripe sets.
6315 if (op == BTRFS_MAP_WRITE)
6316 return full_stripe_len - (offset - *full_stripe_start);
6320 * For other RAID types and for RAID56 reads, allow a single stripe (on
6323 if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK)
6324 return BTRFS_STRIPE_LEN - *stripe_offset;
6328 static int set_io_stripe(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6329 u64 logical, u64 *length, struct btrfs_io_stripe *dst,
6330 struct btrfs_chunk_map *map, u32 stripe_index,
6331 u64 stripe_offset, u64 stripe_nr)
6333 dst->dev = map->stripes[stripe_index].dev;
6335 if (op == BTRFS_MAP_READ && btrfs_need_stripe_tree_update(fs_info, map->type))
6336 return btrfs_get_raid_extent_offset(fs_info, logical, length,
6337 map->type, stripe_index, dst);
6339 dst->physical = map->stripes[stripe_index].physical +
6340 stripe_offset + btrfs_stripe_nr_to_offset(stripe_nr);
6344 static bool is_single_device_io(struct btrfs_fs_info *fs_info,
6345 const struct btrfs_io_stripe *smap,
6346 const struct btrfs_chunk_map *map,
6347 int num_alloc_stripes,
6348 enum btrfs_map_op op, int mirror_num)
6353 if (num_alloc_stripes != 1)
6356 if (btrfs_need_stripe_tree_update(fs_info, map->type) && op != BTRFS_MAP_READ)
6359 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1)
6365 static void map_blocks_raid0(const struct btrfs_chunk_map *map,
6366 struct btrfs_io_geometry *io_geom)
6368 io_geom->stripe_index = io_geom->stripe_nr % map->num_stripes;
6369 io_geom->stripe_nr /= map->num_stripes;
6370 if (io_geom->op == BTRFS_MAP_READ)
6371 io_geom->mirror_num = 1;
6374 static void map_blocks_raid1(struct btrfs_fs_info *fs_info,
6375 struct btrfs_chunk_map *map,
6376 struct btrfs_io_geometry *io_geom,
6377 bool dev_replace_is_ongoing)
6379 if (io_geom->op != BTRFS_MAP_READ) {
6380 io_geom->num_stripes = map->num_stripes;
6384 if (io_geom->mirror_num) {
6385 io_geom->stripe_index = io_geom->mirror_num - 1;
6389 io_geom->stripe_index = find_live_mirror(fs_info, map, 0,
6390 dev_replace_is_ongoing);
6391 io_geom->mirror_num = io_geom->stripe_index + 1;
6394 static void map_blocks_dup(const struct btrfs_chunk_map *map,
6395 struct btrfs_io_geometry *io_geom)
6397 if (io_geom->op != BTRFS_MAP_READ) {
6398 io_geom->num_stripes = map->num_stripes;
6402 if (io_geom->mirror_num) {
6403 io_geom->stripe_index = io_geom->mirror_num - 1;
6407 io_geom->mirror_num = 1;
6410 static void map_blocks_raid10(struct btrfs_fs_info *fs_info,
6411 struct btrfs_chunk_map *map,
6412 struct btrfs_io_geometry *io_geom,
6413 bool dev_replace_is_ongoing)
6415 u32 factor = map->num_stripes / map->sub_stripes;
6416 int old_stripe_index;
6418 io_geom->stripe_index = (io_geom->stripe_nr % factor) * map->sub_stripes;
6419 io_geom->stripe_nr /= factor;
6421 if (io_geom->op != BTRFS_MAP_READ) {
6422 io_geom->num_stripes = map->sub_stripes;
6426 if (io_geom->mirror_num) {
6427 io_geom->stripe_index += io_geom->mirror_num - 1;
6431 old_stripe_index = io_geom->stripe_index;
6432 io_geom->stripe_index = find_live_mirror(fs_info, map,
6433 io_geom->stripe_index,
6434 dev_replace_is_ongoing);
6435 io_geom->mirror_num = io_geom->stripe_index - old_stripe_index + 1;
6438 static void map_blocks_raid56_write(struct btrfs_chunk_map *map,
6439 struct btrfs_io_geometry *io_geom,
6440 u64 logical, u64 *length)
6442 int data_stripes = nr_data_stripes(map);
6445 * Needs full stripe mapping.
6447 * Push stripe_nr back to the start of the full stripe For those cases
6448 * needing a full stripe, @stripe_nr is the full stripe number.
6450 * Originally we go raid56_full_stripe_start / full_stripe_len, but
6451 * that can be expensive. Here we just divide @stripe_nr with
6454 io_geom->stripe_nr /= data_stripes;
6456 /* RAID[56] write or recovery. Return all stripes */
6457 io_geom->num_stripes = map->num_stripes;
6458 io_geom->max_errors = btrfs_chunk_max_errors(map);
6460 /* Return the length to the full stripe end. */
6461 *length = min(logical + *length,
6462 io_geom->raid56_full_stripe_start + map->start +
6463 btrfs_stripe_nr_to_offset(data_stripes)) -
6465 io_geom->stripe_index = 0;
6466 io_geom->stripe_offset = 0;
6469 static void map_blocks_raid56_read(struct btrfs_chunk_map *map,
6470 struct btrfs_io_geometry *io_geom)
6472 int data_stripes = nr_data_stripes(map);
6474 ASSERT(io_geom->mirror_num <= 1);
6475 /* Just grab the data stripe directly. */
6476 io_geom->stripe_index = io_geom->stripe_nr % data_stripes;
6477 io_geom->stripe_nr /= data_stripes;
6479 /* We distribute the parity blocks across stripes. */
6480 io_geom->stripe_index =
6481 (io_geom->stripe_nr + io_geom->stripe_index) % map->num_stripes;
6483 if (io_geom->op == BTRFS_MAP_READ && io_geom->mirror_num < 1)
6484 io_geom->mirror_num = 1;
6488 * Map one logical range to one or more physical ranges.
6490 * @length: (Mandatory) mapped length of this run.
6491 * One logical range can be split into different segments
6492 * due to factors like zones and RAID0/5/6/10 stripe
6495 * @bioc_ret: (Mandatory) returned btrfs_io_context structure.
6496 * which has one or more physical ranges (btrfs_io_stripe)
6498 * Caller should call btrfs_put_bioc() to free it after use.
6500 * @smap: (Optional) single physical range optimization.
6501 * If the map request can be fulfilled by one single
6502 * physical range, and this is parameter is not NULL,
6503 * then @bioc_ret would be NULL, and @smap would be
6506 * @mirror_num_ret: (Mandatory) returned mirror number if the original
6509 * Mirror number 0 means to choose any live mirrors.
6511 * For non-RAID56 profiles, non-zero mirror_num means
6512 * the Nth mirror. (e.g. mirror_num 1 means the first
6515 * For RAID56 profile, mirror 1 means rebuild from P and
6516 * the remaining data stripes.
6518 * For RAID6 profile, mirror > 2 means mark another
6519 * data/P stripe error and rebuild from the remaining
6522 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6523 u64 logical, u64 *length,
6524 struct btrfs_io_context **bioc_ret,
6525 struct btrfs_io_stripe *smap, int *mirror_num_ret)
6527 struct btrfs_chunk_map *map;
6528 struct btrfs_io_geometry io_geom = { 0 };
6533 struct btrfs_io_context *bioc = NULL;
6534 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6535 int dev_replace_is_ongoing = 0;
6536 u16 num_alloc_stripes;
6541 io_geom.mirror_num = (mirror_num_ret ? *mirror_num_ret : 0);
6542 io_geom.num_stripes = 1;
6543 io_geom.stripe_index = 0;
6546 num_copies = btrfs_num_copies(fs_info, logical, fs_info->sectorsize);
6547 if (io_geom.mirror_num > num_copies)
6550 map = btrfs_get_chunk_map(fs_info, logical, *length);
6552 return PTR_ERR(map);
6554 map_offset = logical - map->start;
6555 io_geom.raid56_full_stripe_start = (u64)-1;
6556 max_len = btrfs_max_io_len(map, io_geom.op, map_offset, &io_geom.stripe_nr,
6557 &io_geom.stripe_offset,
6558 &io_geom.raid56_full_stripe_start);
6559 *length = min_t(u64, map->chunk_len - map_offset, max_len);
6561 down_read(&dev_replace->rwsem);
6562 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6564 * Hold the semaphore for read during the whole operation, write is
6565 * requested at commit time but must wait.
6567 if (!dev_replace_is_ongoing)
6568 up_read(&dev_replace->rwsem);
6570 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6571 map_blocks_raid0(map, &io_geom);
6572 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6573 map_blocks_raid1(fs_info, map, &io_geom, dev_replace_is_ongoing);
6574 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6575 map_blocks_dup(map, &io_geom);
6576 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6577 map_blocks_raid10(fs_info, map, &io_geom, dev_replace_is_ongoing);
6578 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6579 if (op != BTRFS_MAP_READ || io_geom.mirror_num > 1)
6580 map_blocks_raid56_write(map, &io_geom, logical, length);
6582 map_blocks_raid56_read(map, &io_geom);
6585 * After this, stripe_nr is the number of stripes on this
6586 * device we have to walk to find the data, and stripe_index is
6587 * the number of our device in the stripe array
6589 io_geom.stripe_index = io_geom.stripe_nr % map->num_stripes;
6590 io_geom.stripe_nr /= map->num_stripes;
6591 io_geom.mirror_num = io_geom.stripe_index + 1;
6593 if (io_geom.stripe_index >= map->num_stripes) {
6595 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6596 io_geom.stripe_index, map->num_stripes);
6601 num_alloc_stripes = io_geom.num_stripes;
6602 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6603 op != BTRFS_MAP_READ)
6605 * For replace case, we need to add extra stripes for extra
6606 * duplicated stripes.
6608 * For both WRITE and GET_READ_MIRRORS, we may have at most
6609 * 2 more stripes (DUP types, otherwise 1).
6611 num_alloc_stripes += 2;
6614 * If this I/O maps to a single device, try to return the device and
6615 * physical block information on the stack instead of allocating an
6616 * I/O context structure.
6618 if (is_single_device_io(fs_info, smap, map, num_alloc_stripes, op,
6619 io_geom.mirror_num)) {
6620 ret = set_io_stripe(fs_info, op, logical, length, smap, map,
6621 io_geom.stripe_index, io_geom.stripe_offset,
6624 *mirror_num_ret = io_geom.mirror_num;
6629 bioc = alloc_btrfs_io_context(fs_info, logical, num_alloc_stripes);
6634 bioc->map_type = map->type;
6637 * For RAID56 full map, we need to make sure the stripes[] follows the
6638 * rule that data stripes are all ordered, then followed with P and Q
6641 * It's still mostly the same as other profiles, just with extra rotation.
6643 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
6644 (op != BTRFS_MAP_READ || io_geom.mirror_num > 1)) {
6646 * For RAID56 @stripe_nr is already the number of full stripes
6647 * before us, which is also the rotation value (needs to modulo
6648 * with num_stripes).
6650 * In this case, we just add @stripe_nr with @i, then do the
6651 * modulo, to reduce one modulo call.
6653 bioc->full_stripe_logical = map->start +
6654 btrfs_stripe_nr_to_offset(io_geom.stripe_nr *
6655 nr_data_stripes(map));
6656 for (int i = 0; i < io_geom.num_stripes; i++) {
6657 ret = set_io_stripe(fs_info, op, logical, length,
6658 &bioc->stripes[i], map,
6659 (i + io_geom.stripe_nr) % io_geom.num_stripes,
6660 io_geom.stripe_offset,
6667 * For all other non-RAID56 profiles, just copy the target
6668 * stripe into the bioc.
6670 for (i = 0; i < io_geom.num_stripes; i++) {
6671 ret = set_io_stripe(fs_info, op, logical, length,
6672 &bioc->stripes[i], map,
6673 io_geom.stripe_index,
6674 io_geom.stripe_offset,
6678 io_geom.stripe_index++;
6684 btrfs_put_bioc(bioc);
6688 if (op != BTRFS_MAP_READ)
6689 io_geom.max_errors = btrfs_chunk_max_errors(map);
6691 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6692 op != BTRFS_MAP_READ) {
6693 handle_ops_on_dev_replace(op, bioc, dev_replace, logical,
6694 &io_geom.num_stripes, &io_geom.max_errors);
6698 bioc->num_stripes = io_geom.num_stripes;
6699 bioc->max_errors = io_geom.max_errors;
6700 bioc->mirror_num = io_geom.mirror_num;
6703 if (dev_replace_is_ongoing) {
6704 lockdep_assert_held(&dev_replace->rwsem);
6705 /* Unlock and let waiting writers proceed */
6706 up_read(&dev_replace->rwsem);
6708 btrfs_free_chunk_map(map);
6712 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args,
6713 const struct btrfs_fs_devices *fs_devices)
6715 if (args->fsid == NULL)
6717 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0)
6722 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args,
6723 const struct btrfs_device *device)
6725 if (args->missing) {
6726 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) &&
6732 if (device->devid != args->devid)
6734 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0)
6740 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6743 * If devid and uuid are both specified, the match must be exact, otherwise
6744 * only devid is used.
6746 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,
6747 const struct btrfs_dev_lookup_args *args)
6749 struct btrfs_device *device;
6750 struct btrfs_fs_devices *seed_devs;
6752 if (dev_args_match_fs_devices(args, fs_devices)) {
6753 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6754 if (dev_args_match_device(args, device))
6759 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6760 if (!dev_args_match_fs_devices(args, seed_devs))
6762 list_for_each_entry(device, &seed_devs->devices, dev_list) {
6763 if (dev_args_match_device(args, device))
6771 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6772 u64 devid, u8 *dev_uuid)
6774 struct btrfs_device *device;
6775 unsigned int nofs_flag;
6778 * We call this under the chunk_mutex, so we want to use NOFS for this
6779 * allocation, however we don't want to change btrfs_alloc_device() to
6780 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6784 nofs_flag = memalloc_nofs_save();
6785 device = btrfs_alloc_device(NULL, &devid, dev_uuid, NULL);
6786 memalloc_nofs_restore(nofs_flag);
6790 list_add(&device->dev_list, &fs_devices->devices);
6791 device->fs_devices = fs_devices;
6792 fs_devices->num_devices++;
6794 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6795 fs_devices->missing_devices++;
6801 * Allocate new device struct, set up devid and UUID.
6803 * @fs_info: used only for generating a new devid, can be NULL if
6804 * devid is provided (i.e. @devid != NULL).
6805 * @devid: a pointer to devid for this device. If NULL a new devid
6807 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6809 * @path: a pointer to device path if available, NULL otherwise.
6811 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6812 * on error. Returned struct is not linked onto any lists and must be
6813 * destroyed with btrfs_free_device.
6815 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6816 const u64 *devid, const u8 *uuid,
6819 struct btrfs_device *dev;
6822 if (WARN_ON(!devid && !fs_info))
6823 return ERR_PTR(-EINVAL);
6825 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
6827 return ERR_PTR(-ENOMEM);
6829 INIT_LIST_HEAD(&dev->dev_list);
6830 INIT_LIST_HEAD(&dev->dev_alloc_list);
6831 INIT_LIST_HEAD(&dev->post_commit_list);
6833 atomic_set(&dev->dev_stats_ccnt, 0);
6834 btrfs_device_data_ordered_init(dev);
6835 extent_io_tree_init(fs_info, &dev->alloc_state, IO_TREE_DEVICE_ALLOC_STATE);
6842 ret = find_next_devid(fs_info, &tmp);
6844 btrfs_free_device(dev);
6845 return ERR_PTR(ret);
6851 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6853 generate_random_uuid(dev->uuid);
6856 struct rcu_string *name;
6858 name = rcu_string_strdup(path, GFP_KERNEL);
6860 btrfs_free_device(dev);
6861 return ERR_PTR(-ENOMEM);
6863 rcu_assign_pointer(dev->name, name);
6869 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6870 u64 devid, u8 *uuid, bool error)
6873 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6876 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6880 u64 btrfs_calc_stripe_length(const struct btrfs_chunk_map *map)
6882 const int data_stripes = calc_data_stripes(map->type, map->num_stripes);
6884 return div_u64(map->chunk_len, data_stripes);
6887 #if BITS_PER_LONG == 32
6889 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
6890 * can't be accessed on 32bit systems.
6892 * This function do mount time check to reject the fs if it already has
6893 * metadata chunk beyond that limit.
6895 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6896 u64 logical, u64 length, u64 type)
6898 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6901 if (logical + length < MAX_LFS_FILESIZE)
6904 btrfs_err_32bit_limit(fs_info);
6909 * This is to give early warning for any metadata chunk reaching
6910 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
6911 * Although we can still access the metadata, it's not going to be possible
6912 * once the limit is reached.
6914 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6915 u64 logical, u64 length, u64 type)
6917 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6920 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)
6923 btrfs_warn_32bit_limit(fs_info);
6927 static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info,
6928 u64 devid, u8 *uuid)
6930 struct btrfs_device *dev;
6932 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6933 btrfs_report_missing_device(fs_info, devid, uuid, true);
6934 return ERR_PTR(-ENOENT);
6937 dev = add_missing_dev(fs_info->fs_devices, devid, uuid);
6939 btrfs_err(fs_info, "failed to init missing device %llu: %ld",
6940 devid, PTR_ERR(dev));
6943 btrfs_report_missing_device(fs_info, devid, uuid, false);
6948 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6949 struct btrfs_chunk *chunk)
6951 BTRFS_DEV_LOOKUP_ARGS(args);
6952 struct btrfs_fs_info *fs_info = leaf->fs_info;
6953 struct btrfs_chunk_map *map;
6958 u8 uuid[BTRFS_UUID_SIZE];
6964 logical = key->offset;
6965 length = btrfs_chunk_length(leaf, chunk);
6966 type = btrfs_chunk_type(leaf, chunk);
6967 index = btrfs_bg_flags_to_raid_index(type);
6968 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6970 #if BITS_PER_LONG == 32
6971 ret = check_32bit_meta_chunk(fs_info, logical, length, type);
6974 warn_32bit_meta_chunk(fs_info, logical, length, type);
6978 * Only need to verify chunk item if we're reading from sys chunk array,
6979 * as chunk item in tree block is already verified by tree-checker.
6981 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6982 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6987 map = btrfs_find_chunk_map(fs_info, logical, 1);
6989 /* already mapped? */
6990 if (map && map->start <= logical && map->start + map->chunk_len > logical) {
6991 btrfs_free_chunk_map(map);
6994 btrfs_free_chunk_map(map);
6997 map = btrfs_alloc_chunk_map(num_stripes, GFP_NOFS);
7001 map->start = logical;
7002 map->chunk_len = length;
7003 map->num_stripes = num_stripes;
7004 map->io_width = btrfs_chunk_io_width(leaf, chunk);
7005 map->io_align = btrfs_chunk_io_align(leaf, chunk);
7008 * We can't use the sub_stripes value, as for profiles other than
7009 * RAID10, they may have 0 as sub_stripes for filesystems created by
7010 * older mkfs (<v5.4).
7011 * In that case, it can cause divide-by-zero errors later.
7012 * Since currently sub_stripes is fixed for each profile, let's
7013 * use the trusted value instead.
7015 map->sub_stripes = btrfs_raid_array[index].sub_stripes;
7016 map->verified_stripes = 0;
7017 map->stripe_size = btrfs_calc_stripe_length(map);
7018 for (i = 0; i < num_stripes; i++) {
7019 map->stripes[i].physical =
7020 btrfs_stripe_offset_nr(leaf, chunk, i);
7021 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
7023 read_extent_buffer(leaf, uuid, (unsigned long)
7024 btrfs_stripe_dev_uuid_nr(chunk, i),
7027 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args);
7028 if (!map->stripes[i].dev) {
7029 map->stripes[i].dev = handle_missing_device(fs_info,
7031 if (IS_ERR(map->stripes[i].dev)) {
7032 ret = PTR_ERR(map->stripes[i].dev);
7033 btrfs_free_chunk_map(map);
7038 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
7039 &(map->stripes[i].dev->dev_state));
7042 ret = btrfs_add_chunk_map(fs_info, map);
7045 "failed to add chunk map, start=%llu len=%llu: %d",
7046 map->start, map->chunk_len, ret);
7052 static void fill_device_from_item(struct extent_buffer *leaf,
7053 struct btrfs_dev_item *dev_item,
7054 struct btrfs_device *device)
7058 device->devid = btrfs_device_id(leaf, dev_item);
7059 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
7060 device->total_bytes = device->disk_total_bytes;
7061 device->commit_total_bytes = device->disk_total_bytes;
7062 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
7063 device->commit_bytes_used = device->bytes_used;
7064 device->type = btrfs_device_type(leaf, dev_item);
7065 device->io_align = btrfs_device_io_align(leaf, dev_item);
7066 device->io_width = btrfs_device_io_width(leaf, dev_item);
7067 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
7068 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
7069 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
7071 ptr = btrfs_device_uuid(dev_item);
7072 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
7075 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
7078 struct btrfs_fs_devices *fs_devices;
7081 lockdep_assert_held(&uuid_mutex);
7084 /* This will match only for multi-device seed fs */
7085 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
7086 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
7090 fs_devices = find_fsid(fsid, NULL);
7092 if (!btrfs_test_opt(fs_info, DEGRADED))
7093 return ERR_PTR(-ENOENT);
7095 fs_devices = alloc_fs_devices(fsid);
7096 if (IS_ERR(fs_devices))
7099 fs_devices->seeding = true;
7100 fs_devices->opened = 1;
7105 * Upon first call for a seed fs fsid, just create a private copy of the
7106 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
7108 fs_devices = clone_fs_devices(fs_devices);
7109 if (IS_ERR(fs_devices))
7112 ret = open_fs_devices(fs_devices, BLK_OPEN_READ, fs_info->bdev_holder);
7114 free_fs_devices(fs_devices);
7115 return ERR_PTR(ret);
7118 if (!fs_devices->seeding) {
7119 close_fs_devices(fs_devices);
7120 free_fs_devices(fs_devices);
7121 return ERR_PTR(-EINVAL);
7124 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
7129 static int read_one_dev(struct extent_buffer *leaf,
7130 struct btrfs_dev_item *dev_item)
7132 BTRFS_DEV_LOOKUP_ARGS(args);
7133 struct btrfs_fs_info *fs_info = leaf->fs_info;
7134 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7135 struct btrfs_device *device;
7138 u8 fs_uuid[BTRFS_FSID_SIZE];
7139 u8 dev_uuid[BTRFS_UUID_SIZE];
7141 devid = btrfs_device_id(leaf, dev_item);
7143 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
7145 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
7147 args.uuid = dev_uuid;
7148 args.fsid = fs_uuid;
7150 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
7151 fs_devices = open_seed_devices(fs_info, fs_uuid);
7152 if (IS_ERR(fs_devices))
7153 return PTR_ERR(fs_devices);
7156 device = btrfs_find_device(fs_info->fs_devices, &args);
7158 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7159 btrfs_report_missing_device(fs_info, devid,
7164 device = add_missing_dev(fs_devices, devid, dev_uuid);
7165 if (IS_ERR(device)) {
7167 "failed to add missing dev %llu: %ld",
7168 devid, PTR_ERR(device));
7169 return PTR_ERR(device);
7171 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
7173 if (!device->bdev) {
7174 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7175 btrfs_report_missing_device(fs_info,
7176 devid, dev_uuid, true);
7179 btrfs_report_missing_device(fs_info, devid,
7183 if (!device->bdev &&
7184 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
7186 * this happens when a device that was properly setup
7187 * in the device info lists suddenly goes bad.
7188 * device->bdev is NULL, and so we have to set
7189 * device->missing to one here
7191 device->fs_devices->missing_devices++;
7192 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
7195 /* Move the device to its own fs_devices */
7196 if (device->fs_devices != fs_devices) {
7197 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
7198 &device->dev_state));
7200 list_move(&device->dev_list, &fs_devices->devices);
7201 device->fs_devices->num_devices--;
7202 fs_devices->num_devices++;
7204 device->fs_devices->missing_devices--;
7205 fs_devices->missing_devices++;
7207 device->fs_devices = fs_devices;
7211 if (device->fs_devices != fs_info->fs_devices) {
7212 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
7213 if (device->generation !=
7214 btrfs_device_generation(leaf, dev_item))
7218 fill_device_from_item(leaf, dev_item, device);
7220 u64 max_total_bytes = bdev_nr_bytes(device->bdev);
7222 if (device->total_bytes > max_total_bytes) {
7224 "device total_bytes should be at most %llu but found %llu",
7225 max_total_bytes, device->total_bytes);
7229 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7230 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7231 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7232 device->fs_devices->total_rw_bytes += device->total_bytes;
7233 atomic64_add(device->total_bytes - device->bytes_used,
7234 &fs_info->free_chunk_space);
7240 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7242 struct btrfs_super_block *super_copy = fs_info->super_copy;
7243 struct extent_buffer *sb;
7244 struct btrfs_disk_key *disk_key;
7245 struct btrfs_chunk *chunk;
7247 unsigned long sb_array_offset;
7254 struct btrfs_key key;
7256 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7259 * We allocated a dummy extent, just to use extent buffer accessors.
7260 * There will be unused space after BTRFS_SUPER_INFO_SIZE, but
7261 * that's fine, we will not go beyond system chunk array anyway.
7263 sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET);
7266 set_extent_buffer_uptodate(sb);
7268 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7269 array_size = btrfs_super_sys_array_size(super_copy);
7271 array_ptr = super_copy->sys_chunk_array;
7272 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7275 while (cur_offset < array_size) {
7276 disk_key = (struct btrfs_disk_key *)array_ptr;
7277 len = sizeof(*disk_key);
7278 if (cur_offset + len > array_size)
7279 goto out_short_read;
7281 btrfs_disk_key_to_cpu(&key, disk_key);
7284 sb_array_offset += len;
7287 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7289 "unexpected item type %u in sys_array at offset %u",
7290 (u32)key.type, cur_offset);
7295 chunk = (struct btrfs_chunk *)sb_array_offset;
7297 * At least one btrfs_chunk with one stripe must be present,
7298 * exact stripe count check comes afterwards
7300 len = btrfs_chunk_item_size(1);
7301 if (cur_offset + len > array_size)
7302 goto out_short_read;
7304 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7307 "invalid number of stripes %u in sys_array at offset %u",
7308 num_stripes, cur_offset);
7313 type = btrfs_chunk_type(sb, chunk);
7314 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7316 "invalid chunk type %llu in sys_array at offset %u",
7322 len = btrfs_chunk_item_size(num_stripes);
7323 if (cur_offset + len > array_size)
7324 goto out_short_read;
7326 ret = read_one_chunk(&key, sb, chunk);
7331 sb_array_offset += len;
7334 clear_extent_buffer_uptodate(sb);
7335 free_extent_buffer_stale(sb);
7339 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7341 clear_extent_buffer_uptodate(sb);
7342 free_extent_buffer_stale(sb);
7347 * Check if all chunks in the fs are OK for read-write degraded mount
7349 * If the @failing_dev is specified, it's accounted as missing.
7351 * Return true if all chunks meet the minimal RW mount requirements.
7352 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7354 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7355 struct btrfs_device *failing_dev)
7357 struct btrfs_chunk_map *map;
7361 map = btrfs_find_chunk_map(fs_info, 0, U64_MAX);
7362 /* No chunk at all? Return false anyway */
7373 btrfs_get_num_tolerated_disk_barrier_failures(
7375 for (i = 0; i < map->num_stripes; i++) {
7376 struct btrfs_device *dev = map->stripes[i].dev;
7378 if (!dev || !dev->bdev ||
7379 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7380 dev->last_flush_error)
7382 else if (failing_dev && failing_dev == dev)
7385 if (missing > max_tolerated) {
7388 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7389 map->start, missing, max_tolerated);
7390 btrfs_free_chunk_map(map);
7394 next_start = map->start + map->chunk_len;
7395 btrfs_free_chunk_map(map);
7397 map = btrfs_find_chunk_map(fs_info, next_start, U64_MAX - next_start);
7403 static void readahead_tree_node_children(struct extent_buffer *node)
7406 const int nr_items = btrfs_header_nritems(node);
7408 for (i = 0; i < nr_items; i++)
7409 btrfs_readahead_node_child(node, i);
7412 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7414 struct btrfs_root *root = fs_info->chunk_root;
7415 struct btrfs_path *path;
7416 struct extent_buffer *leaf;
7417 struct btrfs_key key;
7418 struct btrfs_key found_key;
7423 u64 last_ra_node = 0;
7425 path = btrfs_alloc_path();
7430 * uuid_mutex is needed only if we are mounting a sprout FS
7431 * otherwise we don't need it.
7433 mutex_lock(&uuid_mutex);
7436 * It is possible for mount and umount to race in such a way that
7437 * we execute this code path, but open_fs_devices failed to clear
7438 * total_rw_bytes. We certainly want it cleared before reading the
7439 * device items, so clear it here.
7441 fs_info->fs_devices->total_rw_bytes = 0;
7444 * Lockdep complains about possible circular locking dependency between
7445 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores
7446 * used for freeze procection of a fs (struct super_block.s_writers),
7447 * which we take when starting a transaction, and extent buffers of the
7448 * chunk tree if we call read_one_dev() while holding a lock on an
7449 * extent buffer of the chunk tree. Since we are mounting the filesystem
7450 * and at this point there can't be any concurrent task modifying the
7451 * chunk tree, to keep it simple, just skip locking on the chunk tree.
7453 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
7454 path->skip_locking = 1;
7457 * Read all device items, and then all the chunk items. All
7458 * device items are found before any chunk item (their object id
7459 * is smaller than the lowest possible object id for a chunk
7460 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7462 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7465 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) {
7466 struct extent_buffer *node = path->nodes[1];
7468 leaf = path->nodes[0];
7469 slot = path->slots[0];
7472 if (last_ra_node != node->start) {
7473 readahead_tree_node_children(node);
7474 last_ra_node = node->start;
7477 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7478 struct btrfs_dev_item *dev_item;
7479 dev_item = btrfs_item_ptr(leaf, slot,
7480 struct btrfs_dev_item);
7481 ret = read_one_dev(leaf, dev_item);
7485 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7486 struct btrfs_chunk *chunk;
7489 * We are only called at mount time, so no need to take
7490 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7491 * we always lock first fs_info->chunk_mutex before
7492 * acquiring any locks on the chunk tree. This is a
7493 * requirement for chunk allocation, see the comment on
7494 * top of btrfs_chunk_alloc() for details.
7496 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7497 ret = read_one_chunk(&found_key, leaf, chunk);
7502 /* Catch error found during iteration */
7509 * After loading chunk tree, we've got all device information,
7510 * do another round of validation checks.
7512 if (total_dev != fs_info->fs_devices->total_devices) {
7514 "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit",
7515 btrfs_super_num_devices(fs_info->super_copy),
7517 fs_info->fs_devices->total_devices = total_dev;
7518 btrfs_set_super_num_devices(fs_info->super_copy, total_dev);
7520 if (btrfs_super_total_bytes(fs_info->super_copy) <
7521 fs_info->fs_devices->total_rw_bytes) {
7523 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7524 btrfs_super_total_bytes(fs_info->super_copy),
7525 fs_info->fs_devices->total_rw_bytes);
7531 mutex_unlock(&uuid_mutex);
7533 btrfs_free_path(path);
7537 int btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7539 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7540 struct btrfs_device *device;
7543 fs_devices->fs_info = fs_info;
7545 mutex_lock(&fs_devices->device_list_mutex);
7546 list_for_each_entry(device, &fs_devices->devices, dev_list)
7547 device->fs_info = fs_info;
7549 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7550 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7551 device->fs_info = fs_info;
7552 ret = btrfs_get_dev_zone_info(device, false);
7557 seed_devs->fs_info = fs_info;
7559 mutex_unlock(&fs_devices->device_list_mutex);
7564 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7565 const struct btrfs_dev_stats_item *ptr,
7570 read_extent_buffer(eb, &val,
7571 offsetof(struct btrfs_dev_stats_item, values) +
7572 ((unsigned long)ptr) + (index * sizeof(u64)),
7577 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7578 struct btrfs_dev_stats_item *ptr,
7581 write_extent_buffer(eb, &val,
7582 offsetof(struct btrfs_dev_stats_item, values) +
7583 ((unsigned long)ptr) + (index * sizeof(u64)),
7587 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7588 struct btrfs_path *path)
7590 struct btrfs_dev_stats_item *ptr;
7591 struct extent_buffer *eb;
7592 struct btrfs_key key;
7596 if (!device->fs_info->dev_root)
7599 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7600 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7601 key.offset = device->devid;
7602 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7604 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7605 btrfs_dev_stat_set(device, i, 0);
7606 device->dev_stats_valid = 1;
7607 btrfs_release_path(path);
7608 return ret < 0 ? ret : 0;
7610 slot = path->slots[0];
7611 eb = path->nodes[0];
7612 item_size = btrfs_item_size(eb, slot);
7614 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7616 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7617 if (item_size >= (1 + i) * sizeof(__le64))
7618 btrfs_dev_stat_set(device, i,
7619 btrfs_dev_stats_value(eb, ptr, i));
7621 btrfs_dev_stat_set(device, i, 0);
7624 device->dev_stats_valid = 1;
7625 btrfs_dev_stat_print_on_load(device);
7626 btrfs_release_path(path);
7631 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7633 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7634 struct btrfs_device *device;
7635 struct btrfs_path *path = NULL;
7638 path = btrfs_alloc_path();
7642 mutex_lock(&fs_devices->device_list_mutex);
7643 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7644 ret = btrfs_device_init_dev_stats(device, path);
7648 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7649 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7650 ret = btrfs_device_init_dev_stats(device, path);
7656 mutex_unlock(&fs_devices->device_list_mutex);
7658 btrfs_free_path(path);
7662 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7663 struct btrfs_device *device)
7665 struct btrfs_fs_info *fs_info = trans->fs_info;
7666 struct btrfs_root *dev_root = fs_info->dev_root;
7667 struct btrfs_path *path;
7668 struct btrfs_key key;
7669 struct extent_buffer *eb;
7670 struct btrfs_dev_stats_item *ptr;
7674 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7675 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7676 key.offset = device->devid;
7678 path = btrfs_alloc_path();
7681 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7683 btrfs_warn_in_rcu(fs_info,
7684 "error %d while searching for dev_stats item for device %s",
7685 ret, btrfs_dev_name(device));
7690 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7691 /* need to delete old one and insert a new one */
7692 ret = btrfs_del_item(trans, dev_root, path);
7694 btrfs_warn_in_rcu(fs_info,
7695 "delete too small dev_stats item for device %s failed %d",
7696 btrfs_dev_name(device), ret);
7703 /* need to insert a new item */
7704 btrfs_release_path(path);
7705 ret = btrfs_insert_empty_item(trans, dev_root, path,
7706 &key, sizeof(*ptr));
7708 btrfs_warn_in_rcu(fs_info,
7709 "insert dev_stats item for device %s failed %d",
7710 btrfs_dev_name(device), ret);
7715 eb = path->nodes[0];
7716 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7717 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7718 btrfs_set_dev_stats_value(eb, ptr, i,
7719 btrfs_dev_stat_read(device, i));
7720 btrfs_mark_buffer_dirty(trans, eb);
7723 btrfs_free_path(path);
7728 * called from commit_transaction. Writes all changed device stats to disk.
7730 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7732 struct btrfs_fs_info *fs_info = trans->fs_info;
7733 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7734 struct btrfs_device *device;
7738 mutex_lock(&fs_devices->device_list_mutex);
7739 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7740 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7741 if (!device->dev_stats_valid || stats_cnt == 0)
7746 * There is a LOAD-LOAD control dependency between the value of
7747 * dev_stats_ccnt and updating the on-disk values which requires
7748 * reading the in-memory counters. Such control dependencies
7749 * require explicit read memory barriers.
7751 * This memory barriers pairs with smp_mb__before_atomic in
7752 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7753 * barrier implied by atomic_xchg in
7754 * btrfs_dev_stats_read_and_reset
7758 ret = update_dev_stat_item(trans, device);
7760 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7762 mutex_unlock(&fs_devices->device_list_mutex);
7767 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7769 btrfs_dev_stat_inc(dev, index);
7771 if (!dev->dev_stats_valid)
7773 btrfs_err_rl_in_rcu(dev->fs_info,
7774 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7775 btrfs_dev_name(dev),
7776 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7777 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7778 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7779 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7780 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7783 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7787 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7788 if (btrfs_dev_stat_read(dev, i) != 0)
7790 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7791 return; /* all values == 0, suppress message */
7793 btrfs_info_in_rcu(dev->fs_info,
7794 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7795 btrfs_dev_name(dev),
7796 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7797 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7798 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7799 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7800 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7803 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7804 struct btrfs_ioctl_get_dev_stats *stats)
7806 BTRFS_DEV_LOOKUP_ARGS(args);
7807 struct btrfs_device *dev;
7808 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7811 mutex_lock(&fs_devices->device_list_mutex);
7812 args.devid = stats->devid;
7813 dev = btrfs_find_device(fs_info->fs_devices, &args);
7814 mutex_unlock(&fs_devices->device_list_mutex);
7817 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7819 } else if (!dev->dev_stats_valid) {
7820 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7822 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7823 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7824 if (stats->nr_items > i)
7826 btrfs_dev_stat_read_and_reset(dev, i);
7828 btrfs_dev_stat_set(dev, i, 0);
7830 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7831 current->comm, task_pid_nr(current));
7833 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7834 if (stats->nr_items > i)
7835 stats->values[i] = btrfs_dev_stat_read(dev, i);
7837 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7838 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7843 * Update the size and bytes used for each device where it changed. This is
7844 * delayed since we would otherwise get errors while writing out the
7847 * Must be invoked during transaction commit.
7849 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7851 struct btrfs_device *curr, *next;
7853 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7855 if (list_empty(&trans->dev_update_list))
7859 * We don't need the device_list_mutex here. This list is owned by the
7860 * transaction and the transaction must complete before the device is
7863 mutex_lock(&trans->fs_info->chunk_mutex);
7864 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7866 list_del_init(&curr->post_commit_list);
7867 curr->commit_total_bytes = curr->disk_total_bytes;
7868 curr->commit_bytes_used = curr->bytes_used;
7870 mutex_unlock(&trans->fs_info->chunk_mutex);
7874 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7876 int btrfs_bg_type_to_factor(u64 flags)
7878 const int index = btrfs_bg_flags_to_raid_index(flags);
7880 return btrfs_raid_array[index].ncopies;
7885 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7886 u64 chunk_offset, u64 devid,
7887 u64 physical_offset, u64 physical_len)
7889 struct btrfs_dev_lookup_args args = { .devid = devid };
7890 struct btrfs_chunk_map *map;
7891 struct btrfs_device *dev;
7897 map = btrfs_find_chunk_map(fs_info, chunk_offset, 1);
7900 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7901 physical_offset, devid);
7906 stripe_len = btrfs_calc_stripe_length(map);
7907 if (physical_len != stripe_len) {
7909 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7910 physical_offset, devid, map->start, physical_len,
7917 * Very old mkfs.btrfs (before v4.1) will not respect the reserved
7918 * space. Although kernel can handle it without problem, better to warn
7921 if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED)
7923 "devid %llu physical %llu len %llu inside the reserved space",
7924 devid, physical_offset, physical_len);
7926 for (i = 0; i < map->num_stripes; i++) {
7927 if (map->stripes[i].dev->devid == devid &&
7928 map->stripes[i].physical == physical_offset) {
7930 if (map->verified_stripes >= map->num_stripes) {
7932 "too many dev extents for chunk %llu found",
7937 map->verified_stripes++;
7943 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7944 physical_offset, devid);
7948 /* Make sure no dev extent is beyond device boundary */
7949 dev = btrfs_find_device(fs_info->fs_devices, &args);
7951 btrfs_err(fs_info, "failed to find devid %llu", devid);
7956 if (physical_offset + physical_len > dev->disk_total_bytes) {
7958 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7959 devid, physical_offset, physical_len,
7960 dev->disk_total_bytes);
7965 if (dev->zone_info) {
7966 u64 zone_size = dev->zone_info->zone_size;
7968 if (!IS_ALIGNED(physical_offset, zone_size) ||
7969 !IS_ALIGNED(physical_len, zone_size)) {
7971 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
7972 devid, physical_offset, physical_len);
7979 btrfs_free_chunk_map(map);
7983 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7985 struct rb_node *node;
7988 read_lock(&fs_info->mapping_tree_lock);
7989 for (node = rb_first_cached(&fs_info->mapping_tree); node; node = rb_next(node)) {
7990 struct btrfs_chunk_map *map;
7992 map = rb_entry(node, struct btrfs_chunk_map, rb_node);
7993 if (map->num_stripes != map->verified_stripes) {
7995 "chunk %llu has missing dev extent, have %d expect %d",
7996 map->start, map->verified_stripes, map->num_stripes);
8002 read_unlock(&fs_info->mapping_tree_lock);
8007 * Ensure that all dev extents are mapped to correct chunk, otherwise
8008 * later chunk allocation/free would cause unexpected behavior.
8010 * NOTE: This will iterate through the whole device tree, which should be of
8011 * the same size level as the chunk tree. This slightly increases mount time.
8013 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
8015 struct btrfs_path *path;
8016 struct btrfs_root *root = fs_info->dev_root;
8017 struct btrfs_key key;
8019 u64 prev_dev_ext_end = 0;
8023 * We don't have a dev_root because we mounted with ignorebadroots and
8024 * failed to load the root, so we want to skip the verification in this
8027 * However if the dev root is fine, but the tree itself is corrupted
8028 * we'd still fail to mount. This verification is only to make sure
8029 * writes can happen safely, so instead just bypass this check
8030 * completely in the case of IGNOREBADROOTS.
8032 if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
8036 key.type = BTRFS_DEV_EXTENT_KEY;
8039 path = btrfs_alloc_path();
8043 path->reada = READA_FORWARD;
8044 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8048 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
8049 ret = btrfs_next_leaf(root, path);
8052 /* No dev extents at all? Not good */
8059 struct extent_buffer *leaf = path->nodes[0];
8060 struct btrfs_dev_extent *dext;
8061 int slot = path->slots[0];
8063 u64 physical_offset;
8067 btrfs_item_key_to_cpu(leaf, &key, slot);
8068 if (key.type != BTRFS_DEV_EXTENT_KEY)
8070 devid = key.objectid;
8071 physical_offset = key.offset;
8073 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
8074 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
8075 physical_len = btrfs_dev_extent_length(leaf, dext);
8077 /* Check if this dev extent overlaps with the previous one */
8078 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
8080 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
8081 devid, physical_offset, prev_dev_ext_end);
8086 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
8087 physical_offset, physical_len);
8091 prev_dev_ext_end = physical_offset + physical_len;
8093 ret = btrfs_next_item(root, path);
8102 /* Ensure all chunks have corresponding dev extents */
8103 ret = verify_chunk_dev_extent_mapping(fs_info);
8105 btrfs_free_path(path);
8110 * Check whether the given block group or device is pinned by any inode being
8111 * used as a swapfile.
8113 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
8115 struct btrfs_swapfile_pin *sp;
8116 struct rb_node *node;
8118 spin_lock(&fs_info->swapfile_pins_lock);
8119 node = fs_info->swapfile_pins.rb_node;
8121 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
8123 node = node->rb_left;
8124 else if (ptr > sp->ptr)
8125 node = node->rb_right;
8129 spin_unlock(&fs_info->swapfile_pins_lock);
8130 return node != NULL;
8133 static int relocating_repair_kthread(void *data)
8135 struct btrfs_block_group *cache = data;
8136 struct btrfs_fs_info *fs_info = cache->fs_info;
8140 target = cache->start;
8141 btrfs_put_block_group(cache);
8143 sb_start_write(fs_info->sb);
8144 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
8146 "zoned: skip relocating block group %llu to repair: EBUSY",
8148 sb_end_write(fs_info->sb);
8152 mutex_lock(&fs_info->reclaim_bgs_lock);
8154 /* Ensure block group still exists */
8155 cache = btrfs_lookup_block_group(fs_info, target);
8159 if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags))
8162 ret = btrfs_may_alloc_data_chunk(fs_info, target);
8167 "zoned: relocating block group %llu to repair IO failure",
8169 ret = btrfs_relocate_chunk(fs_info, target);
8173 btrfs_put_block_group(cache);
8174 mutex_unlock(&fs_info->reclaim_bgs_lock);
8175 btrfs_exclop_finish(fs_info);
8176 sb_end_write(fs_info->sb);
8181 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical)
8183 struct btrfs_block_group *cache;
8185 if (!btrfs_is_zoned(fs_info))
8188 /* Do not attempt to repair in degraded state */
8189 if (btrfs_test_opt(fs_info, DEGRADED))
8192 cache = btrfs_lookup_block_group(fs_info, logical);
8196 if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) {
8197 btrfs_put_block_group(cache);
8201 kthread_run(relocating_repair_kthread, cache,
8202 "btrfs-relocating-repair");
8207 static void map_raid56_repair_block(struct btrfs_io_context *bioc,
8208 struct btrfs_io_stripe *smap,
8211 int data_stripes = nr_bioc_data_stripes(bioc);
8214 for (i = 0; i < data_stripes; i++) {
8215 u64 stripe_start = bioc->full_stripe_logical +
8216 btrfs_stripe_nr_to_offset(i);
8218 if (logical >= stripe_start &&
8219 logical < stripe_start + BTRFS_STRIPE_LEN)
8222 ASSERT(i < data_stripes);
8223 smap->dev = bioc->stripes[i].dev;
8224 smap->physical = bioc->stripes[i].physical +
8225 ((logical - bioc->full_stripe_logical) &
8226 BTRFS_STRIPE_LEN_MASK);
8230 * Map a repair write into a single device.
8232 * A repair write is triggered by read time repair or scrub, which would only
8233 * update the contents of a single device.
8234 * Not update any other mirrors nor go through RMW path.
8236 * Callers should ensure:
8238 * - Call btrfs_bio_counter_inc_blocked() first
8239 * - The range does not cross stripe boundary
8240 * - Has a valid @mirror_num passed in.
8242 int btrfs_map_repair_block(struct btrfs_fs_info *fs_info,
8243 struct btrfs_io_stripe *smap, u64 logical,
8244 u32 length, int mirror_num)
8246 struct btrfs_io_context *bioc = NULL;
8247 u64 map_length = length;
8248 int mirror_ret = mirror_num;
8251 ASSERT(mirror_num > 0);
8253 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, &map_length,
8254 &bioc, smap, &mirror_ret);
8258 /* The map range should not cross stripe boundary. */
8259 ASSERT(map_length >= length);
8261 /* Already mapped to single stripe. */
8265 /* Map the RAID56 multi-stripe writes to a single one. */
8266 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
8267 map_raid56_repair_block(bioc, smap, logical);
8271 ASSERT(mirror_num <= bioc->num_stripes);
8272 smap->dev = bioc->stripes[mirror_num - 1].dev;
8273 smap->physical = bioc->stripes[mirror_num - 1].physical;
8275 btrfs_put_bioc(bioc);