1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/sched/mm.h>
8 #include <linux/slab.h>
9 #include <linux/ratelimit.h>
10 #include <linux/kthread.h>
11 #include <linux/semaphore.h>
12 #include <linux/uuid.h>
13 #include <linux/list_sort.h>
14 #include <linux/namei.h>
17 #include "extent_map.h"
19 #include "transaction.h"
20 #include "print-tree.h"
23 #include "rcu-string.h"
24 #include "dev-replace.h"
26 #include "tree-checker.h"
27 #include "space-info.h"
28 #include "block-group.h"
32 #include "accessors.h"
33 #include "uuid-tree.h"
35 #include "relocation.h"
39 #define BTRFS_BLOCK_GROUP_STRIPE_MASK (BTRFS_BLOCK_GROUP_RAID0 | \
40 BTRFS_BLOCK_GROUP_RAID10 | \
41 BTRFS_BLOCK_GROUP_RAID56_MASK)
43 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
44 [BTRFS_RAID_RAID10] = {
47 .devs_max = 0, /* 0 == as many as possible */
49 .tolerated_failures = 1,
53 .raid_name = "raid10",
54 .bg_flag = BTRFS_BLOCK_GROUP_RAID10,
55 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
57 [BTRFS_RAID_RAID1] = {
62 .tolerated_failures = 1,
67 .bg_flag = BTRFS_BLOCK_GROUP_RAID1,
68 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
70 [BTRFS_RAID_RAID1C3] = {
75 .tolerated_failures = 2,
79 .raid_name = "raid1c3",
80 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3,
81 .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET,
83 [BTRFS_RAID_RAID1C4] = {
88 .tolerated_failures = 3,
92 .raid_name = "raid1c4",
93 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4,
94 .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET,
101 .tolerated_failures = 0,
106 .bg_flag = BTRFS_BLOCK_GROUP_DUP,
109 [BTRFS_RAID_RAID0] = {
114 .tolerated_failures = 0,
118 .raid_name = "raid0",
119 .bg_flag = BTRFS_BLOCK_GROUP_RAID0,
122 [BTRFS_RAID_SINGLE] = {
127 .tolerated_failures = 0,
131 .raid_name = "single",
135 [BTRFS_RAID_RAID5] = {
140 .tolerated_failures = 1,
144 .raid_name = "raid5",
145 .bg_flag = BTRFS_BLOCK_GROUP_RAID5,
146 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
148 [BTRFS_RAID_RAID6] = {
153 .tolerated_failures = 2,
157 .raid_name = "raid6",
158 .bg_flag = BTRFS_BLOCK_GROUP_RAID6,
159 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
164 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which
165 * can be used as index to access btrfs_raid_array[].
167 enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags)
169 const u64 profile = (flags & BTRFS_BLOCK_GROUP_PROFILE_MASK);
172 return BTRFS_RAID_SINGLE;
174 return BTRFS_BG_FLAG_TO_INDEX(profile);
177 const char *btrfs_bg_type_to_raid_name(u64 flags)
179 const int index = btrfs_bg_flags_to_raid_index(flags);
181 if (index >= BTRFS_NR_RAID_TYPES)
184 return btrfs_raid_array[index].raid_name;
187 int btrfs_nr_parity_stripes(u64 type)
189 enum btrfs_raid_types index = btrfs_bg_flags_to_raid_index(type);
191 return btrfs_raid_array[index].nparity;
195 * Fill @buf with textual description of @bg_flags, no more than @size_buf
196 * bytes including terminating null byte.
198 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf)
203 u64 flags = bg_flags;
204 u32 size_bp = size_buf;
211 #define DESCRIBE_FLAG(flag, desc) \
213 if (flags & (flag)) { \
214 ret = snprintf(bp, size_bp, "%s|", (desc)); \
215 if (ret < 0 || ret >= size_bp) \
223 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data");
224 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system");
225 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata");
227 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single");
228 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
229 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag,
230 btrfs_raid_array[i].raid_name);
234 ret = snprintf(bp, size_bp, "0x%llx|", flags);
238 if (size_bp < size_buf)
239 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */
242 * The text is trimmed, it's up to the caller to provide sufficiently
248 static int init_first_rw_device(struct btrfs_trans_handle *trans);
249 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
250 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
256 * There are several mutexes that protect manipulation of devices and low-level
257 * structures like chunks but not block groups, extents or files
259 * uuid_mutex (global lock)
260 * ------------------------
261 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
262 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
263 * device) or requested by the device= mount option
265 * the mutex can be very coarse and can cover long-running operations
267 * protects: updates to fs_devices counters like missing devices, rw devices,
268 * seeding, structure cloning, opening/closing devices at mount/umount time
270 * global::fs_devs - add, remove, updates to the global list
272 * does not protect: manipulation of the fs_devices::devices list in general
273 * but in mount context it could be used to exclude list modifications by eg.
276 * btrfs_device::name - renames (write side), read is RCU
278 * fs_devices::device_list_mutex (per-fs, with RCU)
279 * ------------------------------------------------
280 * protects updates to fs_devices::devices, ie. adding and deleting
282 * simple list traversal with read-only actions can be done with RCU protection
284 * may be used to exclude some operations from running concurrently without any
285 * modifications to the list (see write_all_supers)
287 * Is not required at mount and close times, because our device list is
288 * protected by the uuid_mutex at that point.
292 * protects balance structures (status, state) and context accessed from
293 * several places (internally, ioctl)
297 * protects chunks, adding or removing during allocation, trim or when a new
298 * device is added/removed. Additionally it also protects post_commit_list of
299 * individual devices, since they can be added to the transaction's
300 * post_commit_list only with chunk_mutex held.
304 * a big lock that is held by the cleaner thread and prevents running subvolume
305 * cleaning together with relocation or delayed iputs
317 * Exclusive operations
318 * ====================
320 * Maintains the exclusivity of the following operations that apply to the
321 * whole filesystem and cannot run in parallel.
326 * - Device replace (*)
329 * The device operations (as above) can be in one of the following states:
335 * Only device operations marked with (*) can go into the Paused state for the
338 * - ioctl (only Balance can be Paused through ioctl)
339 * - filesystem remounted as read-only
340 * - filesystem unmounted and mounted as read-only
341 * - system power-cycle and filesystem mounted as read-only
342 * - filesystem or device errors leading to forced read-only
344 * The status of exclusive operation is set and cleared atomically.
345 * During the course of Paused state, fs_info::exclusive_operation remains set.
346 * A device operation in Paused or Running state can be canceled or resumed
347 * either by ioctl (Balance only) or when remounted as read-write.
348 * The exclusive status is cleared when the device operation is canceled or
352 DEFINE_MUTEX(uuid_mutex);
353 static LIST_HEAD(fs_uuids);
354 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void)
360 * alloc_fs_devices - allocate struct btrfs_fs_devices
361 * @fsid: if not NULL, copy the UUID to fs_devices::fsid
362 * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid
364 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
365 * The returned struct is not linked onto any lists and can be destroyed with
366 * kfree() right away.
368 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid,
369 const u8 *metadata_fsid)
371 struct btrfs_fs_devices *fs_devs;
373 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
375 return ERR_PTR(-ENOMEM);
377 mutex_init(&fs_devs->device_list_mutex);
379 INIT_LIST_HEAD(&fs_devs->devices);
380 INIT_LIST_HEAD(&fs_devs->alloc_list);
381 INIT_LIST_HEAD(&fs_devs->fs_list);
382 INIT_LIST_HEAD(&fs_devs->seed_list);
384 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
387 memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE);
389 memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE);
394 void btrfs_free_device(struct btrfs_device *device)
396 WARN_ON(!list_empty(&device->post_commit_list));
397 rcu_string_free(device->name);
398 extent_io_tree_release(&device->alloc_state);
399 btrfs_destroy_dev_zone_info(device);
403 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
405 struct btrfs_device *device;
407 WARN_ON(fs_devices->opened);
408 while (!list_empty(&fs_devices->devices)) {
409 device = list_entry(fs_devices->devices.next,
410 struct btrfs_device, dev_list);
411 list_del(&device->dev_list);
412 btrfs_free_device(device);
417 void __exit btrfs_cleanup_fs_uuids(void)
419 struct btrfs_fs_devices *fs_devices;
421 while (!list_empty(&fs_uuids)) {
422 fs_devices = list_entry(fs_uuids.next,
423 struct btrfs_fs_devices, fs_list);
424 list_del(&fs_devices->fs_list);
425 free_fs_devices(fs_devices);
429 static noinline struct btrfs_fs_devices *find_fsid(
430 const u8 *fsid, const u8 *metadata_fsid)
432 struct btrfs_fs_devices *fs_devices;
436 /* Handle non-split brain cases */
437 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
439 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0
440 && memcmp(metadata_fsid, fs_devices->metadata_uuid,
441 BTRFS_FSID_SIZE) == 0)
444 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
451 static struct btrfs_fs_devices *find_fsid_with_metadata_uuid(
452 struct btrfs_super_block *disk_super)
455 struct btrfs_fs_devices *fs_devices;
458 * Handle scanned device having completed its fsid change but
459 * belonging to a fs_devices that was created by first scanning
460 * a device which didn't have its fsid/metadata_uuid changed
461 * at all and the CHANGING_FSID_V2 flag set.
463 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
464 if (fs_devices->fsid_change &&
465 memcmp(disk_super->metadata_uuid, fs_devices->fsid,
466 BTRFS_FSID_SIZE) == 0 &&
467 memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
468 BTRFS_FSID_SIZE) == 0) {
473 * Handle scanned device having completed its fsid change but
474 * belonging to a fs_devices that was created by a device that
475 * has an outdated pair of fsid/metadata_uuid and
476 * CHANGING_FSID_V2 flag set.
478 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
479 if (fs_devices->fsid_change &&
480 memcmp(fs_devices->metadata_uuid,
481 fs_devices->fsid, BTRFS_FSID_SIZE) != 0 &&
482 memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid,
483 BTRFS_FSID_SIZE) == 0) {
488 return find_fsid(disk_super->fsid, disk_super->metadata_uuid);
493 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
494 int flush, struct block_device **bdev,
495 struct btrfs_super_block **disk_super)
499 *bdev = blkdev_get_by_path(device_path, flags, holder, NULL);
502 ret = PTR_ERR(*bdev);
507 sync_blockdev(*bdev);
508 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
510 blkdev_put(*bdev, flags);
513 invalidate_bdev(*bdev);
514 *disk_super = btrfs_read_dev_super(*bdev);
515 if (IS_ERR(*disk_super)) {
516 ret = PTR_ERR(*disk_super);
517 blkdev_put(*bdev, flags);
529 * Search and remove all stale devices (which are not mounted). When both
530 * inputs are NULL, it will search and release all stale devices.
532 * @devt: Optional. When provided will it release all unmounted devices
533 * matching this devt only.
534 * @skip_device: Optional. Will skip this device when searching for the stale
537 * Return: 0 for success or if @devt is 0.
538 * -EBUSY if @devt is a mounted device.
539 * -ENOENT if @devt does not match any device in the list.
541 static int btrfs_free_stale_devices(dev_t devt, struct btrfs_device *skip_device)
543 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
544 struct btrfs_device *device, *tmp_device;
547 lockdep_assert_held(&uuid_mutex);
552 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
554 mutex_lock(&fs_devices->device_list_mutex);
555 list_for_each_entry_safe(device, tmp_device,
556 &fs_devices->devices, dev_list) {
557 if (skip_device && skip_device == device)
559 if (devt && devt != device->devt)
561 if (fs_devices->opened) {
562 /* for an already deleted device return 0 */
563 if (devt && ret != 0)
568 /* delete the stale device */
569 fs_devices->num_devices--;
570 list_del(&device->dev_list);
571 btrfs_free_device(device);
575 mutex_unlock(&fs_devices->device_list_mutex);
577 if (fs_devices->num_devices == 0) {
578 btrfs_sysfs_remove_fsid(fs_devices);
579 list_del(&fs_devices->fs_list);
580 free_fs_devices(fs_devices);
588 * This is only used on mount, and we are protected from competing things
589 * messing with our fs_devices by the uuid_mutex, thus we do not need the
590 * fs_devices->device_list_mutex here.
592 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
593 struct btrfs_device *device, fmode_t flags,
596 struct block_device *bdev;
597 struct btrfs_super_block *disk_super;
606 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
611 devid = btrfs_stack_device_id(&disk_super->dev_item);
612 if (devid != device->devid)
613 goto error_free_page;
615 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
616 goto error_free_page;
618 device->generation = btrfs_super_generation(disk_super);
620 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
621 if (btrfs_super_incompat_flags(disk_super) &
622 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
624 "BTRFS: Invalid seeding and uuid-changed device detected\n");
625 goto error_free_page;
628 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
629 fs_devices->seeding = true;
631 if (bdev_read_only(bdev))
632 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
634 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
637 if (!bdev_nonrot(bdev))
638 fs_devices->rotating = true;
640 if (bdev_max_discard_sectors(bdev))
641 fs_devices->discardable = true;
644 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
645 device->mode = flags;
647 fs_devices->open_devices++;
648 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
649 device->devid != BTRFS_DEV_REPLACE_DEVID) {
650 fs_devices->rw_devices++;
651 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
653 btrfs_release_disk_super(disk_super);
658 btrfs_release_disk_super(disk_super);
659 blkdev_put(bdev, flags);
665 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices
666 * being created with a disk that has already completed its fsid change. Such
667 * disk can belong to an fs which has its FSID changed or to one which doesn't.
668 * Handle both cases here.
670 static struct btrfs_fs_devices *find_fsid_inprogress(
671 struct btrfs_super_block *disk_super)
673 struct btrfs_fs_devices *fs_devices;
675 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
676 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
677 BTRFS_FSID_SIZE) != 0 &&
678 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
679 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) {
684 return find_fsid(disk_super->fsid, NULL);
688 static struct btrfs_fs_devices *find_fsid_changed(
689 struct btrfs_super_block *disk_super)
691 struct btrfs_fs_devices *fs_devices;
694 * Handles the case where scanned device is part of an fs that had
695 * multiple successful changes of FSID but currently device didn't
696 * observe it. Meaning our fsid will be different than theirs. We need
697 * to handle two subcases :
698 * 1 - The fs still continues to have different METADATA/FSID uuids.
699 * 2 - The fs is switched back to its original FSID (METADATA/FSID
702 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
704 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
705 BTRFS_FSID_SIZE) != 0 &&
706 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid,
707 BTRFS_FSID_SIZE) == 0 &&
708 memcmp(fs_devices->fsid, disk_super->fsid,
709 BTRFS_FSID_SIZE) != 0)
712 /* Unchanged UUIDs */
713 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
714 BTRFS_FSID_SIZE) == 0 &&
715 memcmp(fs_devices->fsid, disk_super->metadata_uuid,
716 BTRFS_FSID_SIZE) == 0)
723 static struct btrfs_fs_devices *find_fsid_reverted_metadata(
724 struct btrfs_super_block *disk_super)
726 struct btrfs_fs_devices *fs_devices;
729 * Handle the case where the scanned device is part of an fs whose last
730 * metadata UUID change reverted it to the original FSID. At the same
731 * time fs_devices was first created by another constituent device
732 * which didn't fully observe the operation. This results in an
733 * btrfs_fs_devices created with metadata/fsid different AND
734 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the
735 * fs_devices equal to the FSID of the disk.
737 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
738 if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
739 BTRFS_FSID_SIZE) != 0 &&
740 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
741 BTRFS_FSID_SIZE) == 0 &&
742 fs_devices->fsid_change)
749 * Add new device to list of registered devices
752 * device pointer which was just added or updated when successful
753 * error pointer when failed
755 static noinline struct btrfs_device *device_list_add(const char *path,
756 struct btrfs_super_block *disk_super,
757 bool *new_device_added)
759 struct btrfs_device *device;
760 struct btrfs_fs_devices *fs_devices = NULL;
761 struct rcu_string *name;
762 u64 found_transid = btrfs_super_generation(disk_super);
763 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
766 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
767 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
768 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) &
769 BTRFS_SUPER_FLAG_CHANGING_FSID_V2);
771 error = lookup_bdev(path, &path_devt);
773 btrfs_err(NULL, "failed to lookup block device for path %s: %d",
775 return ERR_PTR(error);
778 if (fsid_change_in_progress) {
779 if (!has_metadata_uuid)
780 fs_devices = find_fsid_inprogress(disk_super);
782 fs_devices = find_fsid_changed(disk_super);
783 } else if (has_metadata_uuid) {
784 fs_devices = find_fsid_with_metadata_uuid(disk_super);
786 fs_devices = find_fsid_reverted_metadata(disk_super);
788 fs_devices = find_fsid(disk_super->fsid, NULL);
793 if (has_metadata_uuid)
794 fs_devices = alloc_fs_devices(disk_super->fsid,
795 disk_super->metadata_uuid);
797 fs_devices = alloc_fs_devices(disk_super->fsid, NULL);
799 if (IS_ERR(fs_devices))
800 return ERR_CAST(fs_devices);
802 fs_devices->fsid_change = fsid_change_in_progress;
804 mutex_lock(&fs_devices->device_list_mutex);
805 list_add(&fs_devices->fs_list, &fs_uuids);
809 struct btrfs_dev_lookup_args args = {
811 .uuid = disk_super->dev_item.uuid,
814 mutex_lock(&fs_devices->device_list_mutex);
815 device = btrfs_find_device(fs_devices, &args);
818 * If this disk has been pulled into an fs devices created by
819 * a device which had the CHANGING_FSID_V2 flag then replace the
820 * metadata_uuid/fsid values of the fs_devices.
822 if (fs_devices->fsid_change &&
823 found_transid > fs_devices->latest_generation) {
824 memcpy(fs_devices->fsid, disk_super->fsid,
827 if (has_metadata_uuid)
828 memcpy(fs_devices->metadata_uuid,
829 disk_super->metadata_uuid,
832 memcpy(fs_devices->metadata_uuid,
833 disk_super->fsid, BTRFS_FSID_SIZE);
835 fs_devices->fsid_change = false;
840 unsigned int nofs_flag;
842 if (fs_devices->opened) {
844 "device %s belongs to fsid %pU, and the fs is already mounted",
845 path, fs_devices->fsid);
846 mutex_unlock(&fs_devices->device_list_mutex);
847 return ERR_PTR(-EBUSY);
850 nofs_flag = memalloc_nofs_save();
851 device = btrfs_alloc_device(NULL, &devid,
852 disk_super->dev_item.uuid, path);
853 memalloc_nofs_restore(nofs_flag);
854 if (IS_ERR(device)) {
855 mutex_unlock(&fs_devices->device_list_mutex);
856 /* we can safely leave the fs_devices entry around */
860 device->devt = path_devt;
862 list_add_rcu(&device->dev_list, &fs_devices->devices);
863 fs_devices->num_devices++;
865 device->fs_devices = fs_devices;
866 *new_device_added = true;
868 if (disk_super->label[0])
870 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n",
871 disk_super->label, devid, found_transid, path,
872 current->comm, task_pid_nr(current));
875 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n",
876 disk_super->fsid, devid, found_transid, path,
877 current->comm, task_pid_nr(current));
879 } else if (!device->name || strcmp(device->name->str, path)) {
881 * When FS is already mounted.
882 * 1. If you are here and if the device->name is NULL that
883 * means this device was missing at time of FS mount.
884 * 2. If you are here and if the device->name is different
885 * from 'path' that means either
886 * a. The same device disappeared and reappeared with
888 * b. The missing-disk-which-was-replaced, has
891 * We must allow 1 and 2a above. But 2b would be a spurious
894 * Further in case of 1 and 2a above, the disk at 'path'
895 * would have missed some transaction when it was away and
896 * in case of 2a the stale bdev has to be updated as well.
897 * 2b must not be allowed at all time.
901 * For now, we do allow update to btrfs_fs_device through the
902 * btrfs dev scan cli after FS has been mounted. We're still
903 * tracking a problem where systems fail mount by subvolume id
904 * when we reject replacement on a mounted FS.
906 if (!fs_devices->opened && found_transid < device->generation) {
908 * That is if the FS is _not_ mounted and if you
909 * are here, that means there is more than one
910 * disk with same uuid and devid.We keep the one
911 * with larger generation number or the last-in if
912 * generation are equal.
914 mutex_unlock(&fs_devices->device_list_mutex);
916 "device %s already registered with a higher generation, found %llu expect %llu",
917 path, found_transid, device->generation);
918 return ERR_PTR(-EEXIST);
922 * We are going to replace the device path for a given devid,
923 * make sure it's the same device if the device is mounted
925 * NOTE: the device->fs_info may not be reliable here so pass
926 * in a NULL to message helpers instead. This avoids a possible
927 * use-after-free when the fs_info and fs_info->sb are already
931 if (device->devt != path_devt) {
932 mutex_unlock(&fs_devices->device_list_mutex);
933 btrfs_warn_in_rcu(NULL,
934 "duplicate device %s devid %llu generation %llu scanned by %s (%d)",
935 path, devid, found_transid,
937 task_pid_nr(current));
938 return ERR_PTR(-EEXIST);
940 btrfs_info_in_rcu(NULL,
941 "devid %llu device path %s changed to %s scanned by %s (%d)",
942 devid, btrfs_dev_name(device),
944 task_pid_nr(current));
947 name = rcu_string_strdup(path, GFP_NOFS);
949 mutex_unlock(&fs_devices->device_list_mutex);
950 return ERR_PTR(-ENOMEM);
952 rcu_string_free(device->name);
953 rcu_assign_pointer(device->name, name);
954 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
955 fs_devices->missing_devices--;
956 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
958 device->devt = path_devt;
962 * Unmount does not free the btrfs_device struct but would zero
963 * generation along with most of the other members. So just update
964 * it back. We need it to pick the disk with largest generation
967 if (!fs_devices->opened) {
968 device->generation = found_transid;
969 fs_devices->latest_generation = max_t(u64, found_transid,
970 fs_devices->latest_generation);
973 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
975 mutex_unlock(&fs_devices->device_list_mutex);
979 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
981 struct btrfs_fs_devices *fs_devices;
982 struct btrfs_device *device;
983 struct btrfs_device *orig_dev;
986 lockdep_assert_held(&uuid_mutex);
988 fs_devices = alloc_fs_devices(orig->fsid, NULL);
989 if (IS_ERR(fs_devices))
992 fs_devices->total_devices = orig->total_devices;
994 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
995 const char *dev_path = NULL;
998 * This is ok to do without RCU read locked because we hold the
999 * uuid mutex so nothing we touch in here is going to disappear.
1002 dev_path = orig_dev->name->str;
1004 device = btrfs_alloc_device(NULL, &orig_dev->devid,
1005 orig_dev->uuid, dev_path);
1006 if (IS_ERR(device)) {
1007 ret = PTR_ERR(device);
1011 if (orig_dev->zone_info) {
1012 struct btrfs_zoned_device_info *zone_info;
1014 zone_info = btrfs_clone_dev_zone_info(orig_dev);
1016 btrfs_free_device(device);
1020 device->zone_info = zone_info;
1023 list_add(&device->dev_list, &fs_devices->devices);
1024 device->fs_devices = fs_devices;
1025 fs_devices->num_devices++;
1029 free_fs_devices(fs_devices);
1030 return ERR_PTR(ret);
1033 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices,
1034 struct btrfs_device **latest_dev)
1036 struct btrfs_device *device, *next;
1038 /* This is the initialized path, it is safe to release the devices. */
1039 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1040 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) {
1041 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1042 &device->dev_state) &&
1043 !test_bit(BTRFS_DEV_STATE_MISSING,
1044 &device->dev_state) &&
1046 device->generation > (*latest_dev)->generation)) {
1047 *latest_dev = device;
1053 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID,
1054 * in btrfs_init_dev_replace() so just continue.
1056 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1060 blkdev_put(device->bdev, device->mode);
1061 device->bdev = NULL;
1062 fs_devices->open_devices--;
1064 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1065 list_del_init(&device->dev_alloc_list);
1066 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1067 fs_devices->rw_devices--;
1069 list_del_init(&device->dev_list);
1070 fs_devices->num_devices--;
1071 btrfs_free_device(device);
1077 * After we have read the system tree and know devids belonging to this
1078 * filesystem, remove the device which does not belong there.
1080 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices)
1082 struct btrfs_device *latest_dev = NULL;
1083 struct btrfs_fs_devices *seed_dev;
1085 mutex_lock(&uuid_mutex);
1086 __btrfs_free_extra_devids(fs_devices, &latest_dev);
1088 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list)
1089 __btrfs_free_extra_devids(seed_dev, &latest_dev);
1091 fs_devices->latest_dev = latest_dev;
1093 mutex_unlock(&uuid_mutex);
1096 static void btrfs_close_bdev(struct btrfs_device *device)
1101 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1102 sync_blockdev(device->bdev);
1103 invalidate_bdev(device->bdev);
1106 blkdev_put(device->bdev, device->mode);
1109 static void btrfs_close_one_device(struct btrfs_device *device)
1111 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1113 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1114 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1115 list_del_init(&device->dev_alloc_list);
1116 fs_devices->rw_devices--;
1119 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1120 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
1122 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
1123 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
1124 fs_devices->missing_devices--;
1127 btrfs_close_bdev(device);
1129 fs_devices->open_devices--;
1130 device->bdev = NULL;
1132 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1133 btrfs_destroy_dev_zone_info(device);
1135 device->fs_info = NULL;
1136 atomic_set(&device->dev_stats_ccnt, 0);
1137 extent_io_tree_release(&device->alloc_state);
1140 * Reset the flush error record. We might have a transient flush error
1141 * in this mount, and if so we aborted the current transaction and set
1142 * the fs to an error state, guaranteeing no super blocks can be further
1143 * committed. However that error might be transient and if we unmount the
1144 * filesystem and mount it again, we should allow the mount to succeed
1145 * (btrfs_check_rw_degradable() should not fail) - if after mounting the
1146 * filesystem again we still get flush errors, then we will again abort
1147 * any transaction and set the error state, guaranteeing no commits of
1148 * unsafe super blocks.
1150 device->last_flush_error = 0;
1152 /* Verify the device is back in a pristine state */
1153 WARN_ON(test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1154 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1155 WARN_ON(!list_empty(&device->dev_alloc_list));
1156 WARN_ON(!list_empty(&device->post_commit_list));
1159 static void close_fs_devices(struct btrfs_fs_devices *fs_devices)
1161 struct btrfs_device *device, *tmp;
1163 lockdep_assert_held(&uuid_mutex);
1165 if (--fs_devices->opened > 0)
1168 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list)
1169 btrfs_close_one_device(device);
1171 WARN_ON(fs_devices->open_devices);
1172 WARN_ON(fs_devices->rw_devices);
1173 fs_devices->opened = 0;
1174 fs_devices->seeding = false;
1175 fs_devices->fs_info = NULL;
1178 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1181 struct btrfs_fs_devices *tmp;
1183 mutex_lock(&uuid_mutex);
1184 close_fs_devices(fs_devices);
1185 if (!fs_devices->opened) {
1186 list_splice_init(&fs_devices->seed_list, &list);
1189 * If the struct btrfs_fs_devices is not assembled with any
1190 * other device, it can be re-initialized during the next mount
1191 * without the needing device-scan step. Therefore, it can be
1194 if (fs_devices->num_devices == 1) {
1195 list_del(&fs_devices->fs_list);
1196 free_fs_devices(fs_devices);
1201 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {
1202 close_fs_devices(fs_devices);
1203 list_del(&fs_devices->seed_list);
1204 free_fs_devices(fs_devices);
1206 mutex_unlock(&uuid_mutex);
1209 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1210 fmode_t flags, void *holder)
1212 struct btrfs_device *device;
1213 struct btrfs_device *latest_dev = NULL;
1214 struct btrfs_device *tmp_device;
1216 flags |= FMODE_EXCL;
1218 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,
1222 ret = btrfs_open_one_device(fs_devices, device, flags, holder);
1224 (!latest_dev || device->generation > latest_dev->generation)) {
1225 latest_dev = device;
1226 } else if (ret == -ENODATA) {
1227 fs_devices->num_devices--;
1228 list_del(&device->dev_list);
1229 btrfs_free_device(device);
1232 if (fs_devices->open_devices == 0)
1235 fs_devices->opened = 1;
1236 fs_devices->latest_dev = latest_dev;
1237 fs_devices->total_rw_bytes = 0;
1238 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1239 fs_devices->read_policy = BTRFS_READ_POLICY_PID;
1244 static int devid_cmp(void *priv, const struct list_head *a,
1245 const struct list_head *b)
1247 const struct btrfs_device *dev1, *dev2;
1249 dev1 = list_entry(a, struct btrfs_device, dev_list);
1250 dev2 = list_entry(b, struct btrfs_device, dev_list);
1252 if (dev1->devid < dev2->devid)
1254 else if (dev1->devid > dev2->devid)
1259 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1260 fmode_t flags, void *holder)
1264 lockdep_assert_held(&uuid_mutex);
1266 * The device_list_mutex cannot be taken here in case opening the
1267 * underlying device takes further locks like open_mutex.
1269 * We also don't need the lock here as this is called during mount and
1270 * exclusion is provided by uuid_mutex
1273 if (fs_devices->opened) {
1274 fs_devices->opened++;
1277 list_sort(NULL, &fs_devices->devices, devid_cmp);
1278 ret = open_fs_devices(fs_devices, flags, holder);
1284 void btrfs_release_disk_super(struct btrfs_super_block *super)
1286 struct page *page = virt_to_page(super);
1291 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1292 u64 bytenr, u64 bytenr_orig)
1294 struct btrfs_super_block *disk_super;
1299 /* make sure our super fits in the device */
1300 if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev))
1301 return ERR_PTR(-EINVAL);
1303 /* make sure our super fits in the page */
1304 if (sizeof(*disk_super) > PAGE_SIZE)
1305 return ERR_PTR(-EINVAL);
1307 /* make sure our super doesn't straddle pages on disk */
1308 index = bytenr >> PAGE_SHIFT;
1309 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1310 return ERR_PTR(-EINVAL);
1312 /* pull in the page with our super */
1313 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1316 return ERR_CAST(page);
1318 p = page_address(page);
1320 /* align our pointer to the offset of the super block */
1321 disk_super = p + offset_in_page(bytenr);
1323 if (btrfs_super_bytenr(disk_super) != bytenr_orig ||
1324 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1325 btrfs_release_disk_super(p);
1326 return ERR_PTR(-EINVAL);
1329 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1330 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1335 int btrfs_forget_devices(dev_t devt)
1339 mutex_lock(&uuid_mutex);
1340 ret = btrfs_free_stale_devices(devt, NULL);
1341 mutex_unlock(&uuid_mutex);
1347 * Look for a btrfs signature on a device. This may be called out of the mount path
1348 * and we are not allowed to call set_blocksize during the scan. The superblock
1349 * is read via pagecache
1351 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1354 struct btrfs_super_block *disk_super;
1355 bool new_device_added = false;
1356 struct btrfs_device *device = NULL;
1357 struct block_device *bdev;
1358 u64 bytenr, bytenr_orig;
1361 lockdep_assert_held(&uuid_mutex);
1364 * we would like to check all the supers, but that would make
1365 * a btrfs mount succeed after a mkfs from a different FS.
1366 * So, we need to add a special mount option to scan for
1367 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1371 * Avoid using flag |= FMODE_EXCL here, as the systemd-udev may
1372 * initiate the device scan which may race with the user's mount
1373 * or mkfs command, resulting in failure.
1374 * Since the device scan is solely for reading purposes, there is
1375 * no need for FMODE_EXCL. Additionally, the devices are read again
1376 * during the mount process. It is ok to get some inconsistent
1377 * values temporarily, as the device paths of the fsid are the only
1378 * required information for assembling the volume.
1380 bdev = blkdev_get_by_path(path, flags, holder, NULL);
1382 return ERR_CAST(bdev);
1384 bytenr_orig = btrfs_sb_offset(0);
1385 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr);
1387 device = ERR_PTR(ret);
1388 goto error_bdev_put;
1391 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig);
1392 if (IS_ERR(disk_super)) {
1393 device = ERR_CAST(disk_super);
1394 goto error_bdev_put;
1397 device = device_list_add(path, disk_super, &new_device_added);
1398 if (!IS_ERR(device) && new_device_added)
1399 btrfs_free_stale_devices(device->devt, device);
1401 btrfs_release_disk_super(disk_super);
1404 blkdev_put(bdev, flags);
1410 * Try to find a chunk that intersects [start, start + len] range and when one
1411 * such is found, record the end of it in *start
1413 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1416 u64 physical_start, physical_end;
1418 lockdep_assert_held(&device->fs_info->chunk_mutex);
1420 if (!find_first_extent_bit(&device->alloc_state, *start,
1421 &physical_start, &physical_end,
1422 CHUNK_ALLOCATED, NULL)) {
1424 if (in_range(physical_start, *start, len) ||
1425 in_range(*start, physical_start,
1426 physical_end - physical_start)) {
1427 *start = physical_end + 1;
1434 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start)
1436 switch (device->fs_devices->chunk_alloc_policy) {
1437 case BTRFS_CHUNK_ALLOC_REGULAR:
1438 return max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED);
1439 case BTRFS_CHUNK_ALLOC_ZONED:
1441 * We don't care about the starting region like regular
1442 * allocator, because we anyway use/reserve the first two zones
1443 * for superblock logging.
1445 return ALIGN(start, device->zone_info->zone_size);
1451 static bool dev_extent_hole_check_zoned(struct btrfs_device *device,
1452 u64 *hole_start, u64 *hole_size,
1455 u64 zone_size = device->zone_info->zone_size;
1458 bool changed = false;
1460 ASSERT(IS_ALIGNED(*hole_start, zone_size));
1462 while (*hole_size > 0) {
1463 pos = btrfs_find_allocatable_zones(device, *hole_start,
1464 *hole_start + *hole_size,
1466 if (pos != *hole_start) {
1467 *hole_size = *hole_start + *hole_size - pos;
1470 if (*hole_size < num_bytes)
1474 ret = btrfs_ensure_empty_zones(device, pos, num_bytes);
1476 /* Range is ensured to be empty */
1480 /* Given hole range was invalid (outside of device) */
1481 if (ret == -ERANGE) {
1482 *hole_start += *hole_size;
1487 *hole_start += zone_size;
1488 *hole_size -= zone_size;
1496 * Check if specified hole is suitable for allocation.
1498 * @device: the device which we have the hole
1499 * @hole_start: starting position of the hole
1500 * @hole_size: the size of the hole
1501 * @num_bytes: the size of the free space that we need
1503 * This function may modify @hole_start and @hole_size to reflect the suitable
1504 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1506 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1507 u64 *hole_size, u64 num_bytes)
1509 bool changed = false;
1510 u64 hole_end = *hole_start + *hole_size;
1514 * Check before we set max_hole_start, otherwise we could end up
1515 * sending back this offset anyway.
1517 if (contains_pending_extent(device, hole_start, *hole_size)) {
1518 if (hole_end >= *hole_start)
1519 *hole_size = hole_end - *hole_start;
1525 switch (device->fs_devices->chunk_alloc_policy) {
1526 case BTRFS_CHUNK_ALLOC_REGULAR:
1527 /* No extra check */
1529 case BTRFS_CHUNK_ALLOC_ZONED:
1530 if (dev_extent_hole_check_zoned(device, hole_start,
1531 hole_size, num_bytes)) {
1534 * The changed hole can contain pending extent.
1535 * Loop again to check that.
1551 * Find free space in the specified device.
1553 * @device: the device which we search the free space in
1554 * @num_bytes: the size of the free space that we need
1555 * @search_start: the position from which to begin the search
1556 * @start: store the start of the free space.
1557 * @len: the size of the free space. that we find, or the size
1558 * of the max free space if we don't find suitable free space
1560 * This does a pretty simple search, the expectation is that it is called very
1561 * infrequently and that a given device has a small number of extents.
1563 * @start is used to store the start of the free space if we find. But if we
1564 * don't find suitable free space, it will be used to store the start position
1565 * of the max free space.
1567 * @len is used to store the size of the free space that we find.
1568 * But if we don't find suitable free space, it is used to store the size of
1569 * the max free space.
1571 * NOTE: This function will search *commit* root of device tree, and does extra
1572 * check to ensure dev extents are not double allocated.
1573 * This makes the function safe to allocate dev extents but may not report
1574 * correct usable device space, as device extent freed in current transaction
1575 * is not reported as available.
1577 static int find_free_dev_extent_start(struct btrfs_device *device,
1578 u64 num_bytes, u64 search_start, u64 *start,
1581 struct btrfs_fs_info *fs_info = device->fs_info;
1582 struct btrfs_root *root = fs_info->dev_root;
1583 struct btrfs_key key;
1584 struct btrfs_dev_extent *dev_extent;
1585 struct btrfs_path *path;
1590 u64 search_end = device->total_bytes;
1593 struct extent_buffer *l;
1595 search_start = dev_extent_search_start(device, search_start);
1597 WARN_ON(device->zone_info &&
1598 !IS_ALIGNED(num_bytes, device->zone_info->zone_size));
1600 path = btrfs_alloc_path();
1604 max_hole_start = search_start;
1608 if (search_start >= search_end ||
1609 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1614 path->reada = READA_FORWARD;
1615 path->search_commit_root = 1;
1616 path->skip_locking = 1;
1618 key.objectid = device->devid;
1619 key.offset = search_start;
1620 key.type = BTRFS_DEV_EXTENT_KEY;
1622 ret = btrfs_search_backwards(root, &key, path);
1626 while (search_start < search_end) {
1628 slot = path->slots[0];
1629 if (slot >= btrfs_header_nritems(l)) {
1630 ret = btrfs_next_leaf(root, path);
1638 btrfs_item_key_to_cpu(l, &key, slot);
1640 if (key.objectid < device->devid)
1643 if (key.objectid > device->devid)
1646 if (key.type != BTRFS_DEV_EXTENT_KEY)
1649 if (key.offset > search_end)
1652 if (key.offset > search_start) {
1653 hole_size = key.offset - search_start;
1654 dev_extent_hole_check(device, &search_start, &hole_size,
1657 if (hole_size > max_hole_size) {
1658 max_hole_start = search_start;
1659 max_hole_size = hole_size;
1663 * If this free space is greater than which we need,
1664 * it must be the max free space that we have found
1665 * until now, so max_hole_start must point to the start
1666 * of this free space and the length of this free space
1667 * is stored in max_hole_size. Thus, we return
1668 * max_hole_start and max_hole_size and go back to the
1671 if (hole_size >= num_bytes) {
1677 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1678 extent_end = key.offset + btrfs_dev_extent_length(l,
1680 if (extent_end > search_start)
1681 search_start = extent_end;
1688 * At this point, search_start should be the end of
1689 * allocated dev extents, and when shrinking the device,
1690 * search_end may be smaller than search_start.
1692 if (search_end > search_start) {
1693 hole_size = search_end - search_start;
1694 if (dev_extent_hole_check(device, &search_start, &hole_size,
1696 btrfs_release_path(path);
1700 if (hole_size > max_hole_size) {
1701 max_hole_start = search_start;
1702 max_hole_size = hole_size;
1707 if (max_hole_size < num_bytes)
1712 ASSERT(max_hole_start + max_hole_size <= search_end);
1714 btrfs_free_path(path);
1715 *start = max_hole_start;
1717 *len = max_hole_size;
1721 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1722 u64 *start, u64 *len)
1724 /* FIXME use last free of some kind */
1725 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1728 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1729 struct btrfs_device *device,
1730 u64 start, u64 *dev_extent_len)
1732 struct btrfs_fs_info *fs_info = device->fs_info;
1733 struct btrfs_root *root = fs_info->dev_root;
1735 struct btrfs_path *path;
1736 struct btrfs_key key;
1737 struct btrfs_key found_key;
1738 struct extent_buffer *leaf = NULL;
1739 struct btrfs_dev_extent *extent = NULL;
1741 path = btrfs_alloc_path();
1745 key.objectid = device->devid;
1747 key.type = BTRFS_DEV_EXTENT_KEY;
1749 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1751 ret = btrfs_previous_item(root, path, key.objectid,
1752 BTRFS_DEV_EXTENT_KEY);
1755 leaf = path->nodes[0];
1756 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1757 extent = btrfs_item_ptr(leaf, path->slots[0],
1758 struct btrfs_dev_extent);
1759 BUG_ON(found_key.offset > start || found_key.offset +
1760 btrfs_dev_extent_length(leaf, extent) < start);
1762 btrfs_release_path(path);
1764 } else if (ret == 0) {
1765 leaf = path->nodes[0];
1766 extent = btrfs_item_ptr(leaf, path->slots[0],
1767 struct btrfs_dev_extent);
1772 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1774 ret = btrfs_del_item(trans, root, path);
1776 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1778 btrfs_free_path(path);
1782 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1784 struct extent_map_tree *em_tree;
1785 struct extent_map *em;
1789 em_tree = &fs_info->mapping_tree;
1790 read_lock(&em_tree->lock);
1791 n = rb_last(&em_tree->map.rb_root);
1793 em = rb_entry(n, struct extent_map, rb_node);
1794 ret = em->start + em->len;
1796 read_unlock(&em_tree->lock);
1801 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1805 struct btrfs_key key;
1806 struct btrfs_key found_key;
1807 struct btrfs_path *path;
1809 path = btrfs_alloc_path();
1813 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1814 key.type = BTRFS_DEV_ITEM_KEY;
1815 key.offset = (u64)-1;
1817 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1823 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1828 ret = btrfs_previous_item(fs_info->chunk_root, path,
1829 BTRFS_DEV_ITEMS_OBJECTID,
1830 BTRFS_DEV_ITEM_KEY);
1834 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1836 *devid_ret = found_key.offset + 1;
1840 btrfs_free_path(path);
1845 * the device information is stored in the chunk root
1846 * the btrfs_device struct should be fully filled in
1848 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1849 struct btrfs_device *device)
1852 struct btrfs_path *path;
1853 struct btrfs_dev_item *dev_item;
1854 struct extent_buffer *leaf;
1855 struct btrfs_key key;
1858 path = btrfs_alloc_path();
1862 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1863 key.type = BTRFS_DEV_ITEM_KEY;
1864 key.offset = device->devid;
1866 btrfs_reserve_chunk_metadata(trans, true);
1867 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1868 &key, sizeof(*dev_item));
1869 btrfs_trans_release_chunk_metadata(trans);
1873 leaf = path->nodes[0];
1874 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1876 btrfs_set_device_id(leaf, dev_item, device->devid);
1877 btrfs_set_device_generation(leaf, dev_item, 0);
1878 btrfs_set_device_type(leaf, dev_item, device->type);
1879 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1880 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1881 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1882 btrfs_set_device_total_bytes(leaf, dev_item,
1883 btrfs_device_get_disk_total_bytes(device));
1884 btrfs_set_device_bytes_used(leaf, dev_item,
1885 btrfs_device_get_bytes_used(device));
1886 btrfs_set_device_group(leaf, dev_item, 0);
1887 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1888 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1889 btrfs_set_device_start_offset(leaf, dev_item, 0);
1891 ptr = btrfs_device_uuid(dev_item);
1892 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1893 ptr = btrfs_device_fsid(dev_item);
1894 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1895 ptr, BTRFS_FSID_SIZE);
1896 btrfs_mark_buffer_dirty(leaf);
1900 btrfs_free_path(path);
1905 * Function to update ctime/mtime for a given device path.
1906 * Mainly used for ctime/mtime based probe like libblkid.
1908 * We don't care about errors here, this is just to be kind to userspace.
1910 static void update_dev_time(const char *device_path)
1913 struct timespec64 now;
1916 ret = kern_path(device_path, LOOKUP_FOLLOW, &path);
1920 now = current_time(d_inode(path.dentry));
1921 inode_update_time(d_inode(path.dentry), &now, S_MTIME | S_CTIME);
1925 static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans,
1926 struct btrfs_device *device)
1928 struct btrfs_root *root = device->fs_info->chunk_root;
1930 struct btrfs_path *path;
1931 struct btrfs_key key;
1933 path = btrfs_alloc_path();
1937 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1938 key.type = BTRFS_DEV_ITEM_KEY;
1939 key.offset = device->devid;
1941 btrfs_reserve_chunk_metadata(trans, false);
1942 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1943 btrfs_trans_release_chunk_metadata(trans);
1950 ret = btrfs_del_item(trans, root, path);
1952 btrfs_free_path(path);
1957 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1958 * filesystem. It's up to the caller to adjust that number regarding eg. device
1961 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1969 seq = read_seqbegin(&fs_info->profiles_lock);
1971 all_avail = fs_info->avail_data_alloc_bits |
1972 fs_info->avail_system_alloc_bits |
1973 fs_info->avail_metadata_alloc_bits;
1974 } while (read_seqretry(&fs_info->profiles_lock, seq));
1976 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1977 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1980 if (num_devices < btrfs_raid_array[i].devs_min)
1981 return btrfs_raid_array[i].mindev_error;
1987 static struct btrfs_device * btrfs_find_next_active_device(
1988 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1990 struct btrfs_device *next_device;
1992 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1993 if (next_device != device &&
1994 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1995 && next_device->bdev)
2003 * Helper function to check if the given device is part of s_bdev / latest_dev
2004 * and replace it with the provided or the next active device, in the context
2005 * where this function called, there should be always be another device (or
2006 * this_dev) which is active.
2008 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
2009 struct btrfs_device *next_device)
2011 struct btrfs_fs_info *fs_info = device->fs_info;
2014 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
2016 ASSERT(next_device);
2018 if (fs_info->sb->s_bdev &&
2019 (fs_info->sb->s_bdev == device->bdev))
2020 fs_info->sb->s_bdev = next_device->bdev;
2022 if (fs_info->fs_devices->latest_dev->bdev == device->bdev)
2023 fs_info->fs_devices->latest_dev = next_device;
2027 * Return btrfs_fs_devices::num_devices excluding the device that's being
2028 * currently replaced.
2030 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2032 u64 num_devices = fs_info->fs_devices->num_devices;
2034 down_read(&fs_info->dev_replace.rwsem);
2035 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2036 ASSERT(num_devices > 1);
2039 up_read(&fs_info->dev_replace.rwsem);
2044 static void btrfs_scratch_superblock(struct btrfs_fs_info *fs_info,
2045 struct block_device *bdev, int copy_num)
2047 struct btrfs_super_block *disk_super;
2048 const size_t len = sizeof(disk_super->magic);
2049 const u64 bytenr = btrfs_sb_offset(copy_num);
2052 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr);
2053 if (IS_ERR(disk_super))
2056 memset(&disk_super->magic, 0, len);
2057 folio_mark_dirty(virt_to_folio(disk_super));
2058 btrfs_release_disk_super(disk_super);
2060 ret = sync_blockdev_range(bdev, bytenr, bytenr + len - 1);
2062 btrfs_warn(fs_info, "error clearing superblock number %d (%d)",
2066 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
2067 struct block_device *bdev,
2068 const char *device_path)
2075 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2076 if (bdev_is_zoned(bdev))
2077 btrfs_reset_sb_log_zones(bdev, copy_num);
2079 btrfs_scratch_superblock(fs_info, bdev, copy_num);
2082 /* Notify udev that device has changed */
2083 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2085 /* Update ctime/mtime for device path for libblkid */
2086 update_dev_time(device_path);
2089 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
2090 struct btrfs_dev_lookup_args *args,
2091 struct block_device **bdev, fmode_t *mode)
2093 struct btrfs_trans_handle *trans;
2094 struct btrfs_device *device;
2095 struct btrfs_fs_devices *cur_devices;
2096 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2100 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2101 btrfs_err(fs_info, "device remove not supported on extent tree v2 yet");
2106 * The device list in fs_devices is accessed without locks (neither
2107 * uuid_mutex nor device_list_mutex) as it won't change on a mounted
2108 * filesystem and another device rm cannot run.
2110 num_devices = btrfs_num_devices(fs_info);
2112 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2116 device = btrfs_find_device(fs_info->fs_devices, args);
2119 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2125 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2126 btrfs_warn_in_rcu(fs_info,
2127 "cannot remove device %s (devid %llu) due to active swapfile",
2128 btrfs_dev_name(device), device->devid);
2132 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
2133 return BTRFS_ERROR_DEV_TGT_REPLACE;
2135 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2136 fs_info->fs_devices->rw_devices == 1)
2137 return BTRFS_ERROR_DEV_ONLY_WRITABLE;
2139 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2140 mutex_lock(&fs_info->chunk_mutex);
2141 list_del_init(&device->dev_alloc_list);
2142 device->fs_devices->rw_devices--;
2143 mutex_unlock(&fs_info->chunk_mutex);
2146 ret = btrfs_shrink_device(device, 0);
2150 trans = btrfs_start_transaction(fs_info->chunk_root, 0);
2151 if (IS_ERR(trans)) {
2152 ret = PTR_ERR(trans);
2156 ret = btrfs_rm_dev_item(trans, device);
2158 /* Any error in dev item removal is critical */
2160 "failed to remove device item for devid %llu: %d",
2161 device->devid, ret);
2162 btrfs_abort_transaction(trans, ret);
2163 btrfs_end_transaction(trans);
2167 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2168 btrfs_scrub_cancel_dev(device);
2171 * the device list mutex makes sure that we don't change
2172 * the device list while someone else is writing out all
2173 * the device supers. Whoever is writing all supers, should
2174 * lock the device list mutex before getting the number of
2175 * devices in the super block (super_copy). Conversely,
2176 * whoever updates the number of devices in the super block
2177 * (super_copy) should hold the device list mutex.
2181 * In normal cases the cur_devices == fs_devices. But in case
2182 * of deleting a seed device, the cur_devices should point to
2183 * its own fs_devices listed under the fs_devices->seed_list.
2185 cur_devices = device->fs_devices;
2186 mutex_lock(&fs_devices->device_list_mutex);
2187 list_del_rcu(&device->dev_list);
2189 cur_devices->num_devices--;
2190 cur_devices->total_devices--;
2191 /* Update total_devices of the parent fs_devices if it's seed */
2192 if (cur_devices != fs_devices)
2193 fs_devices->total_devices--;
2195 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2196 cur_devices->missing_devices--;
2198 btrfs_assign_next_active_device(device, NULL);
2201 cur_devices->open_devices--;
2202 /* remove sysfs entry */
2203 btrfs_sysfs_remove_device(device);
2206 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2207 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2208 mutex_unlock(&fs_devices->device_list_mutex);
2211 * At this point, the device is zero sized and detached from the
2212 * devices list. All that's left is to zero out the old supers and
2215 * We cannot call btrfs_close_bdev() here because we're holding the sb
2216 * write lock, and blkdev_put() will pull in the ->open_mutex on the
2217 * block device and it's dependencies. Instead just flush the device
2218 * and let the caller do the final blkdev_put.
2220 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2221 btrfs_scratch_superblocks(fs_info, device->bdev,
2224 sync_blockdev(device->bdev);
2225 invalidate_bdev(device->bdev);
2229 *bdev = device->bdev;
2230 *mode = device->mode;
2232 btrfs_free_device(device);
2235 * This can happen if cur_devices is the private seed devices list. We
2236 * cannot call close_fs_devices() here because it expects the uuid_mutex
2237 * to be held, but in fact we don't need that for the private
2238 * seed_devices, we can simply decrement cur_devices->opened and then
2239 * remove it from our list and free the fs_devices.
2241 if (cur_devices->num_devices == 0) {
2242 list_del_init(&cur_devices->seed_list);
2243 ASSERT(cur_devices->opened == 1);
2244 cur_devices->opened--;
2245 free_fs_devices(cur_devices);
2248 ret = btrfs_commit_transaction(trans);
2253 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2254 mutex_lock(&fs_info->chunk_mutex);
2255 list_add(&device->dev_alloc_list,
2256 &fs_devices->alloc_list);
2257 device->fs_devices->rw_devices++;
2258 mutex_unlock(&fs_info->chunk_mutex);
2263 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2265 struct btrfs_fs_devices *fs_devices;
2267 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2270 * in case of fs with no seed, srcdev->fs_devices will point
2271 * to fs_devices of fs_info. However when the dev being replaced is
2272 * a seed dev it will point to the seed's local fs_devices. In short
2273 * srcdev will have its correct fs_devices in both the cases.
2275 fs_devices = srcdev->fs_devices;
2277 list_del_rcu(&srcdev->dev_list);
2278 list_del(&srcdev->dev_alloc_list);
2279 fs_devices->num_devices--;
2280 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2281 fs_devices->missing_devices--;
2283 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2284 fs_devices->rw_devices--;
2287 fs_devices->open_devices--;
2290 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2292 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2294 mutex_lock(&uuid_mutex);
2296 btrfs_close_bdev(srcdev);
2298 btrfs_free_device(srcdev);
2300 /* if this is no devs we rather delete the fs_devices */
2301 if (!fs_devices->num_devices) {
2303 * On a mounted FS, num_devices can't be zero unless it's a
2304 * seed. In case of a seed device being replaced, the replace
2305 * target added to the sprout FS, so there will be no more
2306 * device left under the seed FS.
2308 ASSERT(fs_devices->seeding);
2310 list_del_init(&fs_devices->seed_list);
2311 close_fs_devices(fs_devices);
2312 free_fs_devices(fs_devices);
2314 mutex_unlock(&uuid_mutex);
2317 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2319 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2321 mutex_lock(&fs_devices->device_list_mutex);
2323 btrfs_sysfs_remove_device(tgtdev);
2326 fs_devices->open_devices--;
2328 fs_devices->num_devices--;
2330 btrfs_assign_next_active_device(tgtdev, NULL);
2332 list_del_rcu(&tgtdev->dev_list);
2334 mutex_unlock(&fs_devices->device_list_mutex);
2336 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2339 btrfs_close_bdev(tgtdev);
2341 btrfs_free_device(tgtdev);
2345 * Populate args from device at path.
2347 * @fs_info: the filesystem
2348 * @args: the args to populate
2349 * @path: the path to the device
2351 * This will read the super block of the device at @path and populate @args with
2352 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to
2353 * lookup a device to operate on, but need to do it before we take any locks.
2354 * This properly handles the special case of "missing" that a user may pass in,
2355 * and does some basic sanity checks. The caller must make sure that @path is
2356 * properly NUL terminated before calling in, and must call
2357 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and
2360 * Return: 0 for success, -errno for failure
2362 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,
2363 struct btrfs_dev_lookup_args *args,
2366 struct btrfs_super_block *disk_super;
2367 struct block_device *bdev;
2370 if (!path || !path[0])
2372 if (!strcmp(path, "missing")) {
2373 args->missing = true;
2377 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL);
2378 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL);
2379 if (!args->uuid || !args->fsid) {
2380 btrfs_put_dev_args_from_path(args);
2384 ret = btrfs_get_bdev_and_sb(path, FMODE_READ, fs_info->bdev_holder, 0,
2385 &bdev, &disk_super);
2387 btrfs_put_dev_args_from_path(args);
2391 args->devid = btrfs_stack_device_id(&disk_super->dev_item);
2392 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE);
2393 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2394 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE);
2396 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
2397 btrfs_release_disk_super(disk_super);
2398 blkdev_put(bdev, FMODE_READ);
2403 * Only use this jointly with btrfs_get_dev_args_from_path() because we will
2404 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables
2405 * that don't need to be freed.
2407 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args)
2415 struct btrfs_device *btrfs_find_device_by_devspec(
2416 struct btrfs_fs_info *fs_info, u64 devid,
2417 const char *device_path)
2419 BTRFS_DEV_LOOKUP_ARGS(args);
2420 struct btrfs_device *device;
2425 device = btrfs_find_device(fs_info->fs_devices, &args);
2427 return ERR_PTR(-ENOENT);
2431 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path);
2433 return ERR_PTR(ret);
2434 device = btrfs_find_device(fs_info->fs_devices, &args);
2435 btrfs_put_dev_args_from_path(&args);
2437 return ERR_PTR(-ENOENT);
2441 static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info)
2443 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2444 struct btrfs_fs_devices *old_devices;
2445 struct btrfs_fs_devices *seed_devices;
2447 lockdep_assert_held(&uuid_mutex);
2448 if (!fs_devices->seeding)
2449 return ERR_PTR(-EINVAL);
2452 * Private copy of the seed devices, anchored at
2453 * fs_info->fs_devices->seed_list
2455 seed_devices = alloc_fs_devices(NULL, NULL);
2456 if (IS_ERR(seed_devices))
2457 return seed_devices;
2460 * It's necessary to retain a copy of the original seed fs_devices in
2461 * fs_uuids so that filesystems which have been seeded can successfully
2462 * reference the seed device from open_seed_devices. This also supports
2465 old_devices = clone_fs_devices(fs_devices);
2466 if (IS_ERR(old_devices)) {
2467 kfree(seed_devices);
2471 list_add(&old_devices->fs_list, &fs_uuids);
2473 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2474 seed_devices->opened = 1;
2475 INIT_LIST_HEAD(&seed_devices->devices);
2476 INIT_LIST_HEAD(&seed_devices->alloc_list);
2477 mutex_init(&seed_devices->device_list_mutex);
2479 return seed_devices;
2483 * Splice seed devices into the sprout fs_devices.
2484 * Generate a new fsid for the sprouted read-write filesystem.
2486 static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info,
2487 struct btrfs_fs_devices *seed_devices)
2489 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2490 struct btrfs_super_block *disk_super = fs_info->super_copy;
2491 struct btrfs_device *device;
2495 * We are updating the fsid, the thread leading to device_list_add()
2496 * could race, so uuid_mutex is needed.
2498 lockdep_assert_held(&uuid_mutex);
2501 * The threads listed below may traverse dev_list but can do that without
2502 * device_list_mutex:
2503 * - All device ops and balance - as we are in btrfs_exclop_start.
2504 * - Various dev_list readers - are using RCU.
2505 * - btrfs_ioctl_fitrim() - is using RCU.
2507 * For-read threads as below are using device_list_mutex:
2508 * - Readonly scrub btrfs_scrub_dev()
2509 * - Readonly scrub btrfs_scrub_progress()
2510 * - btrfs_get_dev_stats()
2512 lockdep_assert_held(&fs_devices->device_list_mutex);
2514 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2516 list_for_each_entry(device, &seed_devices->devices, dev_list)
2517 device->fs_devices = seed_devices;
2519 fs_devices->seeding = false;
2520 fs_devices->num_devices = 0;
2521 fs_devices->open_devices = 0;
2522 fs_devices->missing_devices = 0;
2523 fs_devices->rotating = false;
2524 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2526 generate_random_uuid(fs_devices->fsid);
2527 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2528 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2530 super_flags = btrfs_super_flags(disk_super) &
2531 ~BTRFS_SUPER_FLAG_SEEDING;
2532 btrfs_set_super_flags(disk_super, super_flags);
2536 * Store the expected generation for seed devices in device items.
2538 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2540 BTRFS_DEV_LOOKUP_ARGS(args);
2541 struct btrfs_fs_info *fs_info = trans->fs_info;
2542 struct btrfs_root *root = fs_info->chunk_root;
2543 struct btrfs_path *path;
2544 struct extent_buffer *leaf;
2545 struct btrfs_dev_item *dev_item;
2546 struct btrfs_device *device;
2547 struct btrfs_key key;
2548 u8 fs_uuid[BTRFS_FSID_SIZE];
2549 u8 dev_uuid[BTRFS_UUID_SIZE];
2552 path = btrfs_alloc_path();
2556 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2558 key.type = BTRFS_DEV_ITEM_KEY;
2561 btrfs_reserve_chunk_metadata(trans, false);
2562 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2563 btrfs_trans_release_chunk_metadata(trans);
2567 leaf = path->nodes[0];
2569 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2570 ret = btrfs_next_leaf(root, path);
2575 leaf = path->nodes[0];
2576 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2577 btrfs_release_path(path);
2581 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2582 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2583 key.type != BTRFS_DEV_ITEM_KEY)
2586 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2587 struct btrfs_dev_item);
2588 args.devid = btrfs_device_id(leaf, dev_item);
2589 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2591 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2593 args.uuid = dev_uuid;
2594 args.fsid = fs_uuid;
2595 device = btrfs_find_device(fs_info->fs_devices, &args);
2596 BUG_ON(!device); /* Logic error */
2598 if (device->fs_devices->seeding) {
2599 btrfs_set_device_generation(leaf, dev_item,
2600 device->generation);
2601 btrfs_mark_buffer_dirty(leaf);
2609 btrfs_free_path(path);
2613 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2615 struct btrfs_root *root = fs_info->dev_root;
2616 struct btrfs_trans_handle *trans;
2617 struct btrfs_device *device;
2618 struct block_device *bdev;
2619 struct super_block *sb = fs_info->sb;
2620 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2621 struct btrfs_fs_devices *seed_devices = NULL;
2622 u64 orig_super_total_bytes;
2623 u64 orig_super_num_devices;
2625 bool seeding_dev = false;
2626 bool locked = false;
2628 if (sb_rdonly(sb) && !fs_devices->seeding)
2631 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2632 fs_info->bdev_holder, NULL);
2634 return PTR_ERR(bdev);
2636 if (!btrfs_check_device_zone_type(fs_info, bdev)) {
2641 if (fs_devices->seeding) {
2643 down_write(&sb->s_umount);
2644 mutex_lock(&uuid_mutex);
2648 sync_blockdev(bdev);
2651 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2652 if (device->bdev == bdev) {
2660 device = btrfs_alloc_device(fs_info, NULL, NULL, device_path);
2661 if (IS_ERR(device)) {
2662 /* we can safely leave the fs_devices entry around */
2663 ret = PTR_ERR(device);
2667 device->fs_info = fs_info;
2668 device->bdev = bdev;
2669 ret = lookup_bdev(device_path, &device->devt);
2671 goto error_free_device;
2673 ret = btrfs_get_dev_zone_info(device, false);
2675 goto error_free_device;
2677 trans = btrfs_start_transaction(root, 0);
2678 if (IS_ERR(trans)) {
2679 ret = PTR_ERR(trans);
2680 goto error_free_zone;
2683 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2684 device->generation = trans->transid;
2685 device->io_width = fs_info->sectorsize;
2686 device->io_align = fs_info->sectorsize;
2687 device->sector_size = fs_info->sectorsize;
2688 device->total_bytes =
2689 round_down(bdev_nr_bytes(bdev), fs_info->sectorsize);
2690 device->disk_total_bytes = device->total_bytes;
2691 device->commit_total_bytes = device->total_bytes;
2692 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2693 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2694 device->mode = FMODE_EXCL;
2695 device->dev_stats_valid = 1;
2696 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2699 btrfs_clear_sb_rdonly(sb);
2701 /* GFP_KERNEL allocation must not be under device_list_mutex */
2702 seed_devices = btrfs_init_sprout(fs_info);
2703 if (IS_ERR(seed_devices)) {
2704 ret = PTR_ERR(seed_devices);
2705 btrfs_abort_transaction(trans, ret);
2710 mutex_lock(&fs_devices->device_list_mutex);
2712 btrfs_setup_sprout(fs_info, seed_devices);
2713 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev,
2717 device->fs_devices = fs_devices;
2719 mutex_lock(&fs_info->chunk_mutex);
2720 list_add_rcu(&device->dev_list, &fs_devices->devices);
2721 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2722 fs_devices->num_devices++;
2723 fs_devices->open_devices++;
2724 fs_devices->rw_devices++;
2725 fs_devices->total_devices++;
2726 fs_devices->total_rw_bytes += device->total_bytes;
2728 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2730 if (!bdev_nonrot(bdev))
2731 fs_devices->rotating = true;
2733 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2734 btrfs_set_super_total_bytes(fs_info->super_copy,
2735 round_down(orig_super_total_bytes + device->total_bytes,
2736 fs_info->sectorsize));
2738 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2739 btrfs_set_super_num_devices(fs_info->super_copy,
2740 orig_super_num_devices + 1);
2743 * we've got more storage, clear any full flags on the space
2746 btrfs_clear_space_info_full(fs_info);
2748 mutex_unlock(&fs_info->chunk_mutex);
2750 /* Add sysfs device entry */
2751 btrfs_sysfs_add_device(device);
2753 mutex_unlock(&fs_devices->device_list_mutex);
2756 mutex_lock(&fs_info->chunk_mutex);
2757 ret = init_first_rw_device(trans);
2758 mutex_unlock(&fs_info->chunk_mutex);
2760 btrfs_abort_transaction(trans, ret);
2765 ret = btrfs_add_dev_item(trans, device);
2767 btrfs_abort_transaction(trans, ret);
2772 ret = btrfs_finish_sprout(trans);
2774 btrfs_abort_transaction(trans, ret);
2779 * fs_devices now represents the newly sprouted filesystem and
2780 * its fsid has been changed by btrfs_sprout_splice().
2782 btrfs_sysfs_update_sprout_fsid(fs_devices);
2785 ret = btrfs_commit_transaction(trans);
2788 mutex_unlock(&uuid_mutex);
2789 up_write(&sb->s_umount);
2792 if (ret) /* transaction commit */
2795 ret = btrfs_relocate_sys_chunks(fs_info);
2797 btrfs_handle_fs_error(fs_info, ret,
2798 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2799 trans = btrfs_attach_transaction(root);
2800 if (IS_ERR(trans)) {
2801 if (PTR_ERR(trans) == -ENOENT)
2803 ret = PTR_ERR(trans);
2807 ret = btrfs_commit_transaction(trans);
2811 * Now that we have written a new super block to this device, check all
2812 * other fs_devices list if device_path alienates any other scanned
2814 * We can ignore the return value as it typically returns -EINVAL and
2815 * only succeeds if the device was an alien.
2817 btrfs_forget_devices(device->devt);
2819 /* Update ctime/mtime for blkid or udev */
2820 update_dev_time(device_path);
2825 btrfs_sysfs_remove_device(device);
2826 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2827 mutex_lock(&fs_info->chunk_mutex);
2828 list_del_rcu(&device->dev_list);
2829 list_del(&device->dev_alloc_list);
2830 fs_info->fs_devices->num_devices--;
2831 fs_info->fs_devices->open_devices--;
2832 fs_info->fs_devices->rw_devices--;
2833 fs_info->fs_devices->total_devices--;
2834 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2835 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2836 btrfs_set_super_total_bytes(fs_info->super_copy,
2837 orig_super_total_bytes);
2838 btrfs_set_super_num_devices(fs_info->super_copy,
2839 orig_super_num_devices);
2840 mutex_unlock(&fs_info->chunk_mutex);
2841 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2844 btrfs_set_sb_rdonly(sb);
2846 btrfs_end_transaction(trans);
2848 btrfs_destroy_dev_zone_info(device);
2850 btrfs_free_device(device);
2852 blkdev_put(bdev, FMODE_EXCL);
2854 mutex_unlock(&uuid_mutex);
2855 up_write(&sb->s_umount);
2860 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2861 struct btrfs_device *device)
2864 struct btrfs_path *path;
2865 struct btrfs_root *root = device->fs_info->chunk_root;
2866 struct btrfs_dev_item *dev_item;
2867 struct extent_buffer *leaf;
2868 struct btrfs_key key;
2870 path = btrfs_alloc_path();
2874 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2875 key.type = BTRFS_DEV_ITEM_KEY;
2876 key.offset = device->devid;
2878 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2887 leaf = path->nodes[0];
2888 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2890 btrfs_set_device_id(leaf, dev_item, device->devid);
2891 btrfs_set_device_type(leaf, dev_item, device->type);
2892 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2893 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2894 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2895 btrfs_set_device_total_bytes(leaf, dev_item,
2896 btrfs_device_get_disk_total_bytes(device));
2897 btrfs_set_device_bytes_used(leaf, dev_item,
2898 btrfs_device_get_bytes_used(device));
2899 btrfs_mark_buffer_dirty(leaf);
2902 btrfs_free_path(path);
2906 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2907 struct btrfs_device *device, u64 new_size)
2909 struct btrfs_fs_info *fs_info = device->fs_info;
2910 struct btrfs_super_block *super_copy = fs_info->super_copy;
2915 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2918 new_size = round_down(new_size, fs_info->sectorsize);
2920 mutex_lock(&fs_info->chunk_mutex);
2921 old_total = btrfs_super_total_bytes(super_copy);
2922 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2924 if (new_size <= device->total_bytes ||
2925 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2926 mutex_unlock(&fs_info->chunk_mutex);
2930 btrfs_set_super_total_bytes(super_copy,
2931 round_down(old_total + diff, fs_info->sectorsize));
2932 device->fs_devices->total_rw_bytes += diff;
2934 btrfs_device_set_total_bytes(device, new_size);
2935 btrfs_device_set_disk_total_bytes(device, new_size);
2936 btrfs_clear_space_info_full(device->fs_info);
2937 if (list_empty(&device->post_commit_list))
2938 list_add_tail(&device->post_commit_list,
2939 &trans->transaction->dev_update_list);
2940 mutex_unlock(&fs_info->chunk_mutex);
2942 btrfs_reserve_chunk_metadata(trans, false);
2943 ret = btrfs_update_device(trans, device);
2944 btrfs_trans_release_chunk_metadata(trans);
2949 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2951 struct btrfs_fs_info *fs_info = trans->fs_info;
2952 struct btrfs_root *root = fs_info->chunk_root;
2954 struct btrfs_path *path;
2955 struct btrfs_key key;
2957 path = btrfs_alloc_path();
2961 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2962 key.offset = chunk_offset;
2963 key.type = BTRFS_CHUNK_ITEM_KEY;
2965 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2968 else if (ret > 0) { /* Logic error or corruption */
2969 btrfs_handle_fs_error(fs_info, -ENOENT,
2970 "Failed lookup while freeing chunk.");
2975 ret = btrfs_del_item(trans, root, path);
2977 btrfs_handle_fs_error(fs_info, ret,
2978 "Failed to delete chunk item.");
2980 btrfs_free_path(path);
2984 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2986 struct btrfs_super_block *super_copy = fs_info->super_copy;
2987 struct btrfs_disk_key *disk_key;
2988 struct btrfs_chunk *chunk;
2995 struct btrfs_key key;
2997 lockdep_assert_held(&fs_info->chunk_mutex);
2998 array_size = btrfs_super_sys_array_size(super_copy);
3000 ptr = super_copy->sys_chunk_array;
3003 while (cur < array_size) {
3004 disk_key = (struct btrfs_disk_key *)ptr;
3005 btrfs_disk_key_to_cpu(&key, disk_key);
3007 len = sizeof(*disk_key);
3009 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3010 chunk = (struct btrfs_chunk *)(ptr + len);
3011 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
3012 len += btrfs_chunk_item_size(num_stripes);
3017 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
3018 key.offset == chunk_offset) {
3019 memmove(ptr, ptr + len, array_size - (cur + len));
3021 btrfs_set_super_sys_array_size(super_copy, array_size);
3031 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
3032 * @logical: Logical block offset in bytes.
3033 * @length: Length of extent in bytes.
3035 * Return: Chunk mapping or ERR_PTR.
3037 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
3038 u64 logical, u64 length)
3040 struct extent_map_tree *em_tree;
3041 struct extent_map *em;
3043 em_tree = &fs_info->mapping_tree;
3044 read_lock(&em_tree->lock);
3045 em = lookup_extent_mapping(em_tree, logical, length);
3046 read_unlock(&em_tree->lock);
3049 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
3051 return ERR_PTR(-EINVAL);
3054 if (em->start > logical || em->start + em->len < logical) {
3056 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
3057 logical, length, em->start, em->start + em->len);
3058 free_extent_map(em);
3059 return ERR_PTR(-EINVAL);
3062 /* callers are responsible for dropping em's ref. */
3066 static int remove_chunk_item(struct btrfs_trans_handle *trans,
3067 struct map_lookup *map, u64 chunk_offset)
3072 * Removing chunk items and updating the device items in the chunks btree
3073 * requires holding the chunk_mutex.
3074 * See the comment at btrfs_chunk_alloc() for the details.
3076 lockdep_assert_held(&trans->fs_info->chunk_mutex);
3078 for (i = 0; i < map->num_stripes; i++) {
3081 ret = btrfs_update_device(trans, map->stripes[i].dev);
3086 return btrfs_free_chunk(trans, chunk_offset);
3089 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
3091 struct btrfs_fs_info *fs_info = trans->fs_info;
3092 struct extent_map *em;
3093 struct map_lookup *map;
3094 u64 dev_extent_len = 0;
3096 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3098 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
3101 * This is a logic error, but we don't want to just rely on the
3102 * user having built with ASSERT enabled, so if ASSERT doesn't
3103 * do anything we still error out.
3108 map = em->map_lookup;
3111 * First delete the device extent items from the devices btree.
3112 * We take the device_list_mutex to avoid racing with the finishing phase
3113 * of a device replace operation. See the comment below before acquiring
3114 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
3115 * because that can result in a deadlock when deleting the device extent
3116 * items from the devices btree - COWing an extent buffer from the btree
3117 * may result in allocating a new metadata chunk, which would attempt to
3118 * lock again fs_info->chunk_mutex.
3120 mutex_lock(&fs_devices->device_list_mutex);
3121 for (i = 0; i < map->num_stripes; i++) {
3122 struct btrfs_device *device = map->stripes[i].dev;
3123 ret = btrfs_free_dev_extent(trans, device,
3124 map->stripes[i].physical,
3127 mutex_unlock(&fs_devices->device_list_mutex);
3128 btrfs_abort_transaction(trans, ret);
3132 if (device->bytes_used > 0) {
3133 mutex_lock(&fs_info->chunk_mutex);
3134 btrfs_device_set_bytes_used(device,
3135 device->bytes_used - dev_extent_len);
3136 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3137 btrfs_clear_space_info_full(fs_info);
3138 mutex_unlock(&fs_info->chunk_mutex);
3141 mutex_unlock(&fs_devices->device_list_mutex);
3144 * We acquire fs_info->chunk_mutex for 2 reasons:
3146 * 1) Just like with the first phase of the chunk allocation, we must
3147 * reserve system space, do all chunk btree updates and deletions, and
3148 * update the system chunk array in the superblock while holding this
3149 * mutex. This is for similar reasons as explained on the comment at
3150 * the top of btrfs_chunk_alloc();
3152 * 2) Prevent races with the final phase of a device replace operation
3153 * that replaces the device object associated with the map's stripes,
3154 * because the device object's id can change at any time during that
3155 * final phase of the device replace operation
3156 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3157 * replaced device and then see it with an ID of
3158 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3159 * the device item, which does not exists on the chunk btree.
3160 * The finishing phase of device replace acquires both the
3161 * device_list_mutex and the chunk_mutex, in that order, so we are
3162 * safe by just acquiring the chunk_mutex.
3164 trans->removing_chunk = true;
3165 mutex_lock(&fs_info->chunk_mutex);
3167 check_system_chunk(trans, map->type);
3169 ret = remove_chunk_item(trans, map, chunk_offset);
3171 * Normally we should not get -ENOSPC since we reserved space before
3172 * through the call to check_system_chunk().
3174 * Despite our system space_info having enough free space, we may not
3175 * be able to allocate extents from its block groups, because all have
3176 * an incompatible profile, which will force us to allocate a new system
3177 * block group with the right profile, or right after we called
3178 * check_system_space() above, a scrub turned the only system block group
3179 * with enough free space into RO mode.
3180 * This is explained with more detail at do_chunk_alloc().
3182 * So if we get -ENOSPC, allocate a new system chunk and retry once.
3184 if (ret == -ENOSPC) {
3185 const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
3186 struct btrfs_block_group *sys_bg;
3188 sys_bg = btrfs_create_chunk(trans, sys_flags);
3189 if (IS_ERR(sys_bg)) {
3190 ret = PTR_ERR(sys_bg);
3191 btrfs_abort_transaction(trans, ret);
3195 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3197 btrfs_abort_transaction(trans, ret);
3201 ret = remove_chunk_item(trans, map, chunk_offset);
3203 btrfs_abort_transaction(trans, ret);
3207 btrfs_abort_transaction(trans, ret);
3211 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3213 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3214 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3216 btrfs_abort_transaction(trans, ret);
3221 mutex_unlock(&fs_info->chunk_mutex);
3222 trans->removing_chunk = false;
3225 * We are done with chunk btree updates and deletions, so release the
3226 * system space we previously reserved (with check_system_chunk()).
3228 btrfs_trans_release_chunk_metadata(trans);
3230 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3232 btrfs_abort_transaction(trans, ret);
3237 if (trans->removing_chunk) {
3238 mutex_unlock(&fs_info->chunk_mutex);
3239 trans->removing_chunk = false;
3242 free_extent_map(em);
3246 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3248 struct btrfs_root *root = fs_info->chunk_root;
3249 struct btrfs_trans_handle *trans;
3250 struct btrfs_block_group *block_group;
3254 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3256 "relocate: not supported on extent tree v2 yet");
3261 * Prevent races with automatic removal of unused block groups.
3262 * After we relocate and before we remove the chunk with offset
3263 * chunk_offset, automatic removal of the block group can kick in,
3264 * resulting in a failure when calling btrfs_remove_chunk() below.
3266 * Make sure to acquire this mutex before doing a tree search (dev
3267 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3268 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3269 * we release the path used to search the chunk/dev tree and before
3270 * the current task acquires this mutex and calls us.
3272 lockdep_assert_held(&fs_info->reclaim_bgs_lock);
3274 /* step one, relocate all the extents inside this chunk */
3275 btrfs_scrub_pause(fs_info);
3276 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3277 btrfs_scrub_continue(fs_info);
3280 * If we had a transaction abort, stop all running scrubs.
3281 * See transaction.c:cleanup_transaction() why we do it here.
3283 if (BTRFS_FS_ERROR(fs_info))
3284 btrfs_scrub_cancel(fs_info);
3288 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3291 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3292 length = block_group->length;
3293 btrfs_put_block_group(block_group);
3296 * On a zoned file system, discard the whole block group, this will
3297 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3298 * resetting the zone fails, don't treat it as a fatal problem from the
3299 * filesystem's point of view.
3301 if (btrfs_is_zoned(fs_info)) {
3302 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
3305 "failed to reset zone %llu after relocation",
3309 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3311 if (IS_ERR(trans)) {
3312 ret = PTR_ERR(trans);
3313 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3318 * step two, delete the device extents and the
3319 * chunk tree entries
3321 ret = btrfs_remove_chunk(trans, chunk_offset);
3322 btrfs_end_transaction(trans);
3326 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3328 struct btrfs_root *chunk_root = fs_info->chunk_root;
3329 struct btrfs_path *path;
3330 struct extent_buffer *leaf;
3331 struct btrfs_chunk *chunk;
3332 struct btrfs_key key;
3333 struct btrfs_key found_key;
3335 bool retried = false;
3339 path = btrfs_alloc_path();
3344 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3345 key.offset = (u64)-1;
3346 key.type = BTRFS_CHUNK_ITEM_KEY;
3349 mutex_lock(&fs_info->reclaim_bgs_lock);
3350 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3352 mutex_unlock(&fs_info->reclaim_bgs_lock);
3355 BUG_ON(ret == 0); /* Corruption */
3357 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3360 mutex_unlock(&fs_info->reclaim_bgs_lock);
3366 leaf = path->nodes[0];
3367 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3369 chunk = btrfs_item_ptr(leaf, path->slots[0],
3370 struct btrfs_chunk);
3371 chunk_type = btrfs_chunk_type(leaf, chunk);
3372 btrfs_release_path(path);
3374 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3375 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3381 mutex_unlock(&fs_info->reclaim_bgs_lock);
3383 if (found_key.offset == 0)
3385 key.offset = found_key.offset - 1;
3388 if (failed && !retried) {
3392 } else if (WARN_ON(failed && retried)) {
3396 btrfs_free_path(path);
3401 * return 1 : allocate a data chunk successfully,
3402 * return <0: errors during allocating a data chunk,
3403 * return 0 : no need to allocate a data chunk.
3405 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3408 struct btrfs_block_group *cache;
3412 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3414 chunk_type = cache->flags;
3415 btrfs_put_block_group(cache);
3417 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3420 spin_lock(&fs_info->data_sinfo->lock);
3421 bytes_used = fs_info->data_sinfo->bytes_used;
3422 spin_unlock(&fs_info->data_sinfo->lock);
3425 struct btrfs_trans_handle *trans;
3428 trans = btrfs_join_transaction(fs_info->tree_root);
3430 return PTR_ERR(trans);
3432 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3433 btrfs_end_transaction(trans);
3442 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3443 struct btrfs_balance_control *bctl)
3445 struct btrfs_root *root = fs_info->tree_root;
3446 struct btrfs_trans_handle *trans;
3447 struct btrfs_balance_item *item;
3448 struct btrfs_disk_balance_args disk_bargs;
3449 struct btrfs_path *path;
3450 struct extent_buffer *leaf;
3451 struct btrfs_key key;
3454 path = btrfs_alloc_path();
3458 trans = btrfs_start_transaction(root, 0);
3459 if (IS_ERR(trans)) {
3460 btrfs_free_path(path);
3461 return PTR_ERR(trans);
3464 key.objectid = BTRFS_BALANCE_OBJECTID;
3465 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3468 ret = btrfs_insert_empty_item(trans, root, path, &key,
3473 leaf = path->nodes[0];
3474 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3476 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3478 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3479 btrfs_set_balance_data(leaf, item, &disk_bargs);
3480 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3481 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3482 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3483 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3485 btrfs_set_balance_flags(leaf, item, bctl->flags);
3487 btrfs_mark_buffer_dirty(leaf);
3489 btrfs_free_path(path);
3490 err = btrfs_commit_transaction(trans);
3496 static int del_balance_item(struct btrfs_fs_info *fs_info)
3498 struct btrfs_root *root = fs_info->tree_root;
3499 struct btrfs_trans_handle *trans;
3500 struct btrfs_path *path;
3501 struct btrfs_key key;
3504 path = btrfs_alloc_path();
3508 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3509 if (IS_ERR(trans)) {
3510 btrfs_free_path(path);
3511 return PTR_ERR(trans);
3514 key.objectid = BTRFS_BALANCE_OBJECTID;
3515 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3518 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3526 ret = btrfs_del_item(trans, root, path);
3528 btrfs_free_path(path);
3529 err = btrfs_commit_transaction(trans);
3536 * This is a heuristic used to reduce the number of chunks balanced on
3537 * resume after balance was interrupted.
3539 static void update_balance_args(struct btrfs_balance_control *bctl)
3542 * Turn on soft mode for chunk types that were being converted.
3544 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3545 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3546 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3547 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3548 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3549 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3552 * Turn on usage filter if is not already used. The idea is
3553 * that chunks that we have already balanced should be
3554 * reasonably full. Don't do it for chunks that are being
3555 * converted - that will keep us from relocating unconverted
3556 * (albeit full) chunks.
3558 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3559 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3560 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3561 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3562 bctl->data.usage = 90;
3564 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3565 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3566 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3567 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3568 bctl->sys.usage = 90;
3570 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3571 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3572 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3573 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3574 bctl->meta.usage = 90;
3579 * Clear the balance status in fs_info and delete the balance item from disk.
3581 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3583 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3586 BUG_ON(!fs_info->balance_ctl);
3588 spin_lock(&fs_info->balance_lock);
3589 fs_info->balance_ctl = NULL;
3590 spin_unlock(&fs_info->balance_lock);
3593 ret = del_balance_item(fs_info);
3595 btrfs_handle_fs_error(fs_info, ret, NULL);
3599 * Balance filters. Return 1 if chunk should be filtered out
3600 * (should not be balanced).
3602 static int chunk_profiles_filter(u64 chunk_type,
3603 struct btrfs_balance_args *bargs)
3605 chunk_type = chunk_to_extended(chunk_type) &
3606 BTRFS_EXTENDED_PROFILE_MASK;
3608 if (bargs->profiles & chunk_type)
3614 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3615 struct btrfs_balance_args *bargs)
3617 struct btrfs_block_group *cache;
3619 u64 user_thresh_min;
3620 u64 user_thresh_max;
3623 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3624 chunk_used = cache->used;
3626 if (bargs->usage_min == 0)
3627 user_thresh_min = 0;
3629 user_thresh_min = mult_perc(cache->length, bargs->usage_min);
3631 if (bargs->usage_max == 0)
3632 user_thresh_max = 1;
3633 else if (bargs->usage_max > 100)
3634 user_thresh_max = cache->length;
3636 user_thresh_max = mult_perc(cache->length, bargs->usage_max);
3638 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3641 btrfs_put_block_group(cache);
3645 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3646 u64 chunk_offset, struct btrfs_balance_args *bargs)
3648 struct btrfs_block_group *cache;
3649 u64 chunk_used, user_thresh;
3652 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3653 chunk_used = cache->used;
3655 if (bargs->usage_min == 0)
3657 else if (bargs->usage > 100)
3658 user_thresh = cache->length;
3660 user_thresh = mult_perc(cache->length, bargs->usage);
3662 if (chunk_used < user_thresh)
3665 btrfs_put_block_group(cache);
3669 static int chunk_devid_filter(struct extent_buffer *leaf,
3670 struct btrfs_chunk *chunk,
3671 struct btrfs_balance_args *bargs)
3673 struct btrfs_stripe *stripe;
3674 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3677 for (i = 0; i < num_stripes; i++) {
3678 stripe = btrfs_stripe_nr(chunk, i);
3679 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3686 static u64 calc_data_stripes(u64 type, int num_stripes)
3688 const int index = btrfs_bg_flags_to_raid_index(type);
3689 const int ncopies = btrfs_raid_array[index].ncopies;
3690 const int nparity = btrfs_raid_array[index].nparity;
3692 return (num_stripes - nparity) / ncopies;
3695 /* [pstart, pend) */
3696 static int chunk_drange_filter(struct extent_buffer *leaf,
3697 struct btrfs_chunk *chunk,
3698 struct btrfs_balance_args *bargs)
3700 struct btrfs_stripe *stripe;
3701 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3708 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3711 type = btrfs_chunk_type(leaf, chunk);
3712 factor = calc_data_stripes(type, num_stripes);
3714 for (i = 0; i < num_stripes; i++) {
3715 stripe = btrfs_stripe_nr(chunk, i);
3716 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3719 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3720 stripe_length = btrfs_chunk_length(leaf, chunk);
3721 stripe_length = div_u64(stripe_length, factor);
3723 if (stripe_offset < bargs->pend &&
3724 stripe_offset + stripe_length > bargs->pstart)
3731 /* [vstart, vend) */
3732 static int chunk_vrange_filter(struct extent_buffer *leaf,
3733 struct btrfs_chunk *chunk,
3735 struct btrfs_balance_args *bargs)
3737 if (chunk_offset < bargs->vend &&
3738 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3739 /* at least part of the chunk is inside this vrange */
3745 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3746 struct btrfs_chunk *chunk,
3747 struct btrfs_balance_args *bargs)
3749 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3751 if (bargs->stripes_min <= num_stripes
3752 && num_stripes <= bargs->stripes_max)
3758 static int chunk_soft_convert_filter(u64 chunk_type,
3759 struct btrfs_balance_args *bargs)
3761 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3764 chunk_type = chunk_to_extended(chunk_type) &
3765 BTRFS_EXTENDED_PROFILE_MASK;
3767 if (bargs->target == chunk_type)
3773 static int should_balance_chunk(struct extent_buffer *leaf,
3774 struct btrfs_chunk *chunk, u64 chunk_offset)
3776 struct btrfs_fs_info *fs_info = leaf->fs_info;
3777 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3778 struct btrfs_balance_args *bargs = NULL;
3779 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3782 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3783 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3787 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3788 bargs = &bctl->data;
3789 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3791 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3792 bargs = &bctl->meta;
3794 /* profiles filter */
3795 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3796 chunk_profiles_filter(chunk_type, bargs)) {
3801 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3802 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3804 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3805 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3810 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3811 chunk_devid_filter(leaf, chunk, bargs)) {
3815 /* drange filter, makes sense only with devid filter */
3816 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3817 chunk_drange_filter(leaf, chunk, bargs)) {
3822 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3823 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3827 /* stripes filter */
3828 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3829 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3833 /* soft profile changing mode */
3834 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3835 chunk_soft_convert_filter(chunk_type, bargs)) {
3840 * limited by count, must be the last filter
3842 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3843 if (bargs->limit == 0)
3847 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3849 * Same logic as the 'limit' filter; the minimum cannot be
3850 * determined here because we do not have the global information
3851 * about the count of all chunks that satisfy the filters.
3853 if (bargs->limit_max == 0)
3862 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3864 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3865 struct btrfs_root *chunk_root = fs_info->chunk_root;
3867 struct btrfs_chunk *chunk;
3868 struct btrfs_path *path = NULL;
3869 struct btrfs_key key;
3870 struct btrfs_key found_key;
3871 struct extent_buffer *leaf;
3874 int enospc_errors = 0;
3875 bool counting = true;
3876 /* The single value limit and min/max limits use the same bytes in the */
3877 u64 limit_data = bctl->data.limit;
3878 u64 limit_meta = bctl->meta.limit;
3879 u64 limit_sys = bctl->sys.limit;
3883 int chunk_reserved = 0;
3885 path = btrfs_alloc_path();
3891 /* zero out stat counters */
3892 spin_lock(&fs_info->balance_lock);
3893 memset(&bctl->stat, 0, sizeof(bctl->stat));
3894 spin_unlock(&fs_info->balance_lock);
3898 * The single value limit and min/max limits use the same bytes
3901 bctl->data.limit = limit_data;
3902 bctl->meta.limit = limit_meta;
3903 bctl->sys.limit = limit_sys;
3905 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3906 key.offset = (u64)-1;
3907 key.type = BTRFS_CHUNK_ITEM_KEY;
3910 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3911 atomic_read(&fs_info->balance_cancel_req)) {
3916 mutex_lock(&fs_info->reclaim_bgs_lock);
3917 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3919 mutex_unlock(&fs_info->reclaim_bgs_lock);
3924 * this shouldn't happen, it means the last relocate
3928 BUG(); /* FIXME break ? */
3930 ret = btrfs_previous_item(chunk_root, path, 0,
3931 BTRFS_CHUNK_ITEM_KEY);
3933 mutex_unlock(&fs_info->reclaim_bgs_lock);
3938 leaf = path->nodes[0];
3939 slot = path->slots[0];
3940 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3942 if (found_key.objectid != key.objectid) {
3943 mutex_unlock(&fs_info->reclaim_bgs_lock);
3947 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3948 chunk_type = btrfs_chunk_type(leaf, chunk);
3951 spin_lock(&fs_info->balance_lock);
3952 bctl->stat.considered++;
3953 spin_unlock(&fs_info->balance_lock);
3956 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3958 btrfs_release_path(path);
3960 mutex_unlock(&fs_info->reclaim_bgs_lock);
3965 mutex_unlock(&fs_info->reclaim_bgs_lock);
3966 spin_lock(&fs_info->balance_lock);
3967 bctl->stat.expected++;
3968 spin_unlock(&fs_info->balance_lock);
3970 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3972 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3974 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3981 * Apply limit_min filter, no need to check if the LIMITS
3982 * filter is used, limit_min is 0 by default
3984 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3985 count_data < bctl->data.limit_min)
3986 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3987 count_meta < bctl->meta.limit_min)
3988 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3989 count_sys < bctl->sys.limit_min)) {
3990 mutex_unlock(&fs_info->reclaim_bgs_lock);
3994 if (!chunk_reserved) {
3996 * We may be relocating the only data chunk we have,
3997 * which could potentially end up with losing data's
3998 * raid profile, so lets allocate an empty one in
4001 ret = btrfs_may_alloc_data_chunk(fs_info,
4004 mutex_unlock(&fs_info->reclaim_bgs_lock);
4006 } else if (ret == 1) {
4011 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
4012 mutex_unlock(&fs_info->reclaim_bgs_lock);
4013 if (ret == -ENOSPC) {
4015 } else if (ret == -ETXTBSY) {
4017 "skipping relocation of block group %llu due to active swapfile",
4023 spin_lock(&fs_info->balance_lock);
4024 bctl->stat.completed++;
4025 spin_unlock(&fs_info->balance_lock);
4028 if (found_key.offset == 0)
4030 key.offset = found_key.offset - 1;
4034 btrfs_release_path(path);
4039 btrfs_free_path(path);
4040 if (enospc_errors) {
4041 btrfs_info(fs_info, "%d enospc errors during balance",
4051 * See if a given profile is valid and reduced.
4053 * @flags: profile to validate
4054 * @extended: if true @flags is treated as an extended profile
4056 static int alloc_profile_is_valid(u64 flags, int extended)
4058 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
4059 BTRFS_BLOCK_GROUP_PROFILE_MASK);
4061 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
4063 /* 1) check that all other bits are zeroed */
4067 /* 2) see if profile is reduced */
4069 return !extended; /* "0" is valid for usual profiles */
4071 return has_single_bit_set(flags);
4074 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
4076 /* cancel requested || normal exit path */
4077 return atomic_read(&fs_info->balance_cancel_req) ||
4078 (atomic_read(&fs_info->balance_pause_req) == 0 &&
4079 atomic_read(&fs_info->balance_cancel_req) == 0);
4083 * Validate target profile against allowed profiles and return true if it's OK.
4084 * Otherwise print the error message and return false.
4086 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
4087 const struct btrfs_balance_args *bargs,
4088 u64 allowed, const char *type)
4090 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
4093 /* Profile is valid and does not have bits outside of the allowed set */
4094 if (alloc_profile_is_valid(bargs->target, 1) &&
4095 (bargs->target & ~allowed) == 0)
4098 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
4099 type, btrfs_bg_type_to_raid_name(bargs->target));
4104 * Fill @buf with textual description of balance filter flags @bargs, up to
4105 * @size_buf including the terminating null. The output may be trimmed if it
4106 * does not fit into the provided buffer.
4108 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
4112 u32 size_bp = size_buf;
4114 u64 flags = bargs->flags;
4115 char tmp_buf[128] = {'\0'};
4120 #define CHECK_APPEND_NOARG(a) \
4122 ret = snprintf(bp, size_bp, (a)); \
4123 if (ret < 0 || ret >= size_bp) \
4124 goto out_overflow; \
4129 #define CHECK_APPEND_1ARG(a, v1) \
4131 ret = snprintf(bp, size_bp, (a), (v1)); \
4132 if (ret < 0 || ret >= size_bp) \
4133 goto out_overflow; \
4138 #define CHECK_APPEND_2ARG(a, v1, v2) \
4140 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4141 if (ret < 0 || ret >= size_bp) \
4142 goto out_overflow; \
4147 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
4148 CHECK_APPEND_1ARG("convert=%s,",
4149 btrfs_bg_type_to_raid_name(bargs->target));
4151 if (flags & BTRFS_BALANCE_ARGS_SOFT)
4152 CHECK_APPEND_NOARG("soft,");
4154 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4155 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4157 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4160 if (flags & BTRFS_BALANCE_ARGS_USAGE)
4161 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4163 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4164 CHECK_APPEND_2ARG("usage=%u..%u,",
4165 bargs->usage_min, bargs->usage_max);
4167 if (flags & BTRFS_BALANCE_ARGS_DEVID)
4168 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4170 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4171 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4172 bargs->pstart, bargs->pend);
4174 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4175 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4176 bargs->vstart, bargs->vend);
4178 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4179 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4181 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4182 CHECK_APPEND_2ARG("limit=%u..%u,",
4183 bargs->limit_min, bargs->limit_max);
4185 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4186 CHECK_APPEND_2ARG("stripes=%u..%u,",
4187 bargs->stripes_min, bargs->stripes_max);
4189 #undef CHECK_APPEND_2ARG
4190 #undef CHECK_APPEND_1ARG
4191 #undef CHECK_APPEND_NOARG
4195 if (size_bp < size_buf)
4196 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4201 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4203 u32 size_buf = 1024;
4204 char tmp_buf[192] = {'\0'};
4207 u32 size_bp = size_buf;
4209 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4211 buf = kzalloc(size_buf, GFP_KERNEL);
4217 #define CHECK_APPEND_1ARG(a, v1) \
4219 ret = snprintf(bp, size_bp, (a), (v1)); \
4220 if (ret < 0 || ret >= size_bp) \
4221 goto out_overflow; \
4226 if (bctl->flags & BTRFS_BALANCE_FORCE)
4227 CHECK_APPEND_1ARG("%s", "-f ");
4229 if (bctl->flags & BTRFS_BALANCE_DATA) {
4230 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4231 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4234 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4235 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4236 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4239 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4240 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4241 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4244 #undef CHECK_APPEND_1ARG
4248 if (size_bp < size_buf)
4249 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4250 btrfs_info(fs_info, "balance: %s %s",
4251 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4252 "resume" : "start", buf);
4258 * Should be called with balance mutexe held
4260 int btrfs_balance(struct btrfs_fs_info *fs_info,
4261 struct btrfs_balance_control *bctl,
4262 struct btrfs_ioctl_balance_args *bargs)
4264 u64 meta_target, data_target;
4270 bool reducing_redundancy;
4273 if (btrfs_fs_closing(fs_info) ||
4274 atomic_read(&fs_info->balance_pause_req) ||
4275 btrfs_should_cancel_balance(fs_info)) {
4280 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4281 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4285 * In case of mixed groups both data and meta should be picked,
4286 * and identical options should be given for both of them.
4288 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4289 if (mixed && (bctl->flags & allowed)) {
4290 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4291 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4292 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4294 "balance: mixed groups data and metadata options must be the same");
4301 * rw_devices will not change at the moment, device add/delete/replace
4304 num_devices = fs_info->fs_devices->rw_devices;
4307 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4308 * special bit for it, to make it easier to distinguish. Thus we need
4309 * to set it manually, or balance would refuse the profile.
4311 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4312 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4313 if (num_devices >= btrfs_raid_array[i].devs_min)
4314 allowed |= btrfs_raid_array[i].bg_flag;
4316 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4317 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4318 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4324 * Allow to reduce metadata or system integrity only if force set for
4325 * profiles with redundancy (copies, parity)
4328 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4329 if (btrfs_raid_array[i].ncopies >= 2 ||
4330 btrfs_raid_array[i].tolerated_failures >= 1)
4331 allowed |= btrfs_raid_array[i].bg_flag;
4334 seq = read_seqbegin(&fs_info->profiles_lock);
4336 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4337 (fs_info->avail_system_alloc_bits & allowed) &&
4338 !(bctl->sys.target & allowed)) ||
4339 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4340 (fs_info->avail_metadata_alloc_bits & allowed) &&
4341 !(bctl->meta.target & allowed)))
4342 reducing_redundancy = true;
4344 reducing_redundancy = false;
4346 /* if we're not converting, the target field is uninitialized */
4347 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4348 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4349 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4350 bctl->data.target : fs_info->avail_data_alloc_bits;
4351 } while (read_seqretry(&fs_info->profiles_lock, seq));
4353 if (reducing_redundancy) {
4354 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4356 "balance: force reducing metadata redundancy");
4359 "balance: reduces metadata redundancy, use --force if you want this");
4365 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4366 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4368 "balance: metadata profile %s has lower redundancy than data profile %s",
4369 btrfs_bg_type_to_raid_name(meta_target),
4370 btrfs_bg_type_to_raid_name(data_target));
4373 ret = insert_balance_item(fs_info, bctl);
4374 if (ret && ret != -EEXIST)
4377 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4378 BUG_ON(ret == -EEXIST);
4379 BUG_ON(fs_info->balance_ctl);
4380 spin_lock(&fs_info->balance_lock);
4381 fs_info->balance_ctl = bctl;
4382 spin_unlock(&fs_info->balance_lock);
4384 BUG_ON(ret != -EEXIST);
4385 spin_lock(&fs_info->balance_lock);
4386 update_balance_args(bctl);
4387 spin_unlock(&fs_info->balance_lock);
4390 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4391 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4392 describe_balance_start_or_resume(fs_info);
4393 mutex_unlock(&fs_info->balance_mutex);
4395 ret = __btrfs_balance(fs_info);
4397 mutex_lock(&fs_info->balance_mutex);
4398 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) {
4399 btrfs_info(fs_info, "balance: paused");
4400 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
4403 * Balance can be canceled by:
4405 * - Regular cancel request
4406 * Then ret == -ECANCELED and balance_cancel_req > 0
4408 * - Fatal signal to "btrfs" process
4409 * Either the signal caught by wait_reserve_ticket() and callers
4410 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4412 * Either way, in this case balance_cancel_req = 0, and
4413 * ret == -EINTR or ret == -ECANCELED.
4415 * So here we only check the return value to catch canceled balance.
4417 else if (ret == -ECANCELED || ret == -EINTR)
4418 btrfs_info(fs_info, "balance: canceled");
4420 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4422 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4425 memset(bargs, 0, sizeof(*bargs));
4426 btrfs_update_ioctl_balance_args(fs_info, bargs);
4429 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4430 balance_need_close(fs_info)) {
4431 reset_balance_state(fs_info);
4432 btrfs_exclop_finish(fs_info);
4435 wake_up(&fs_info->balance_wait_q);
4439 if (bctl->flags & BTRFS_BALANCE_RESUME)
4440 reset_balance_state(fs_info);
4443 btrfs_exclop_finish(fs_info);
4448 static int balance_kthread(void *data)
4450 struct btrfs_fs_info *fs_info = data;
4453 sb_start_write(fs_info->sb);
4454 mutex_lock(&fs_info->balance_mutex);
4455 if (fs_info->balance_ctl)
4456 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4457 mutex_unlock(&fs_info->balance_mutex);
4458 sb_end_write(fs_info->sb);
4463 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4465 struct task_struct *tsk;
4467 mutex_lock(&fs_info->balance_mutex);
4468 if (!fs_info->balance_ctl) {
4469 mutex_unlock(&fs_info->balance_mutex);
4472 mutex_unlock(&fs_info->balance_mutex);
4474 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4475 btrfs_info(fs_info, "balance: resume skipped");
4479 spin_lock(&fs_info->super_lock);
4480 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
4481 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;
4482 spin_unlock(&fs_info->super_lock);
4484 * A ro->rw remount sequence should continue with the paused balance
4485 * regardless of who pauses it, system or the user as of now, so set
4488 spin_lock(&fs_info->balance_lock);
4489 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4490 spin_unlock(&fs_info->balance_lock);
4492 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4493 return PTR_ERR_OR_ZERO(tsk);
4496 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4498 struct btrfs_balance_control *bctl;
4499 struct btrfs_balance_item *item;
4500 struct btrfs_disk_balance_args disk_bargs;
4501 struct btrfs_path *path;
4502 struct extent_buffer *leaf;
4503 struct btrfs_key key;
4506 path = btrfs_alloc_path();
4510 key.objectid = BTRFS_BALANCE_OBJECTID;
4511 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4514 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4517 if (ret > 0) { /* ret = -ENOENT; */
4522 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4528 leaf = path->nodes[0];
4529 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4531 bctl->flags = btrfs_balance_flags(leaf, item);
4532 bctl->flags |= BTRFS_BALANCE_RESUME;
4534 btrfs_balance_data(leaf, item, &disk_bargs);
4535 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4536 btrfs_balance_meta(leaf, item, &disk_bargs);
4537 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4538 btrfs_balance_sys(leaf, item, &disk_bargs);
4539 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4542 * This should never happen, as the paused balance state is recovered
4543 * during mount without any chance of other exclusive ops to collide.
4545 * This gives the exclusive op status to balance and keeps in paused
4546 * state until user intervention (cancel or umount). If the ownership
4547 * cannot be assigned, show a message but do not fail. The balance
4548 * is in a paused state and must have fs_info::balance_ctl properly
4551 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED))
4553 "balance: cannot set exclusive op status, resume manually");
4555 btrfs_release_path(path);
4557 mutex_lock(&fs_info->balance_mutex);
4558 BUG_ON(fs_info->balance_ctl);
4559 spin_lock(&fs_info->balance_lock);
4560 fs_info->balance_ctl = bctl;
4561 spin_unlock(&fs_info->balance_lock);
4562 mutex_unlock(&fs_info->balance_mutex);
4564 btrfs_free_path(path);
4568 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4572 mutex_lock(&fs_info->balance_mutex);
4573 if (!fs_info->balance_ctl) {
4574 mutex_unlock(&fs_info->balance_mutex);
4578 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4579 atomic_inc(&fs_info->balance_pause_req);
4580 mutex_unlock(&fs_info->balance_mutex);
4582 wait_event(fs_info->balance_wait_q,
4583 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4585 mutex_lock(&fs_info->balance_mutex);
4586 /* we are good with balance_ctl ripped off from under us */
4587 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4588 atomic_dec(&fs_info->balance_pause_req);
4593 mutex_unlock(&fs_info->balance_mutex);
4597 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4599 mutex_lock(&fs_info->balance_mutex);
4600 if (!fs_info->balance_ctl) {
4601 mutex_unlock(&fs_info->balance_mutex);
4606 * A paused balance with the item stored on disk can be resumed at
4607 * mount time if the mount is read-write. Otherwise it's still paused
4608 * and we must not allow cancelling as it deletes the item.
4610 if (sb_rdonly(fs_info->sb)) {
4611 mutex_unlock(&fs_info->balance_mutex);
4615 atomic_inc(&fs_info->balance_cancel_req);
4617 * if we are running just wait and return, balance item is
4618 * deleted in btrfs_balance in this case
4620 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4621 mutex_unlock(&fs_info->balance_mutex);
4622 wait_event(fs_info->balance_wait_q,
4623 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4624 mutex_lock(&fs_info->balance_mutex);
4626 mutex_unlock(&fs_info->balance_mutex);
4628 * Lock released to allow other waiters to continue, we'll
4629 * reexamine the status again.
4631 mutex_lock(&fs_info->balance_mutex);
4633 if (fs_info->balance_ctl) {
4634 reset_balance_state(fs_info);
4635 btrfs_exclop_finish(fs_info);
4636 btrfs_info(fs_info, "balance: canceled");
4640 BUG_ON(fs_info->balance_ctl ||
4641 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4642 atomic_dec(&fs_info->balance_cancel_req);
4643 mutex_unlock(&fs_info->balance_mutex);
4647 int btrfs_uuid_scan_kthread(void *data)
4649 struct btrfs_fs_info *fs_info = data;
4650 struct btrfs_root *root = fs_info->tree_root;
4651 struct btrfs_key key;
4652 struct btrfs_path *path = NULL;
4654 struct extent_buffer *eb;
4656 struct btrfs_root_item root_item;
4658 struct btrfs_trans_handle *trans = NULL;
4659 bool closing = false;
4661 path = btrfs_alloc_path();
4668 key.type = BTRFS_ROOT_ITEM_KEY;
4672 if (btrfs_fs_closing(fs_info)) {
4676 ret = btrfs_search_forward(root, &key, path,
4677 BTRFS_OLDEST_GENERATION);
4684 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4685 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4686 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4687 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4690 eb = path->nodes[0];
4691 slot = path->slots[0];
4692 item_size = btrfs_item_size(eb, slot);
4693 if (item_size < sizeof(root_item))
4696 read_extent_buffer(eb, &root_item,
4697 btrfs_item_ptr_offset(eb, slot),
4698 (int)sizeof(root_item));
4699 if (btrfs_root_refs(&root_item) == 0)
4702 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4703 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4707 btrfs_release_path(path);
4709 * 1 - subvol uuid item
4710 * 1 - received_subvol uuid item
4712 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4713 if (IS_ERR(trans)) {
4714 ret = PTR_ERR(trans);
4722 btrfs_release_path(path);
4723 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4724 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4725 BTRFS_UUID_KEY_SUBVOL,
4728 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4734 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4735 ret = btrfs_uuid_tree_add(trans,
4736 root_item.received_uuid,
4737 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4740 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4747 btrfs_release_path(path);
4749 ret = btrfs_end_transaction(trans);
4755 if (key.offset < (u64)-1) {
4757 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4759 key.type = BTRFS_ROOT_ITEM_KEY;
4760 } else if (key.objectid < (u64)-1) {
4762 key.type = BTRFS_ROOT_ITEM_KEY;
4771 btrfs_free_path(path);
4772 if (trans && !IS_ERR(trans))
4773 btrfs_end_transaction(trans);
4775 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4777 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4778 up(&fs_info->uuid_tree_rescan_sem);
4782 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4784 struct btrfs_trans_handle *trans;
4785 struct btrfs_root *tree_root = fs_info->tree_root;
4786 struct btrfs_root *uuid_root;
4787 struct task_struct *task;
4794 trans = btrfs_start_transaction(tree_root, 2);
4796 return PTR_ERR(trans);
4798 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4799 if (IS_ERR(uuid_root)) {
4800 ret = PTR_ERR(uuid_root);
4801 btrfs_abort_transaction(trans, ret);
4802 btrfs_end_transaction(trans);
4806 fs_info->uuid_root = uuid_root;
4808 ret = btrfs_commit_transaction(trans);
4812 down(&fs_info->uuid_tree_rescan_sem);
4813 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4815 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4816 btrfs_warn(fs_info, "failed to start uuid_scan task");
4817 up(&fs_info->uuid_tree_rescan_sem);
4818 return PTR_ERR(task);
4825 * shrinking a device means finding all of the device extents past
4826 * the new size, and then following the back refs to the chunks.
4827 * The chunk relocation code actually frees the device extent
4829 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4831 struct btrfs_fs_info *fs_info = device->fs_info;
4832 struct btrfs_root *root = fs_info->dev_root;
4833 struct btrfs_trans_handle *trans;
4834 struct btrfs_dev_extent *dev_extent = NULL;
4835 struct btrfs_path *path;
4841 bool retried = false;
4842 struct extent_buffer *l;
4843 struct btrfs_key key;
4844 struct btrfs_super_block *super_copy = fs_info->super_copy;
4845 u64 old_total = btrfs_super_total_bytes(super_copy);
4846 u64 old_size = btrfs_device_get_total_bytes(device);
4850 new_size = round_down(new_size, fs_info->sectorsize);
4852 diff = round_down(old_size - new_size, fs_info->sectorsize);
4854 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4857 path = btrfs_alloc_path();
4861 path->reada = READA_BACK;
4863 trans = btrfs_start_transaction(root, 0);
4864 if (IS_ERR(trans)) {
4865 btrfs_free_path(path);
4866 return PTR_ERR(trans);
4869 mutex_lock(&fs_info->chunk_mutex);
4871 btrfs_device_set_total_bytes(device, new_size);
4872 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4873 device->fs_devices->total_rw_bytes -= diff;
4874 atomic64_sub(diff, &fs_info->free_chunk_space);
4878 * Once the device's size has been set to the new size, ensure all
4879 * in-memory chunks are synced to disk so that the loop below sees them
4880 * and relocates them accordingly.
4882 if (contains_pending_extent(device, &start, diff)) {
4883 mutex_unlock(&fs_info->chunk_mutex);
4884 ret = btrfs_commit_transaction(trans);
4888 mutex_unlock(&fs_info->chunk_mutex);
4889 btrfs_end_transaction(trans);
4893 key.objectid = device->devid;
4894 key.offset = (u64)-1;
4895 key.type = BTRFS_DEV_EXTENT_KEY;
4898 mutex_lock(&fs_info->reclaim_bgs_lock);
4899 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4901 mutex_unlock(&fs_info->reclaim_bgs_lock);
4905 ret = btrfs_previous_item(root, path, 0, key.type);
4907 mutex_unlock(&fs_info->reclaim_bgs_lock);
4911 btrfs_release_path(path);
4916 slot = path->slots[0];
4917 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4919 if (key.objectid != device->devid) {
4920 mutex_unlock(&fs_info->reclaim_bgs_lock);
4921 btrfs_release_path(path);
4925 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4926 length = btrfs_dev_extent_length(l, dev_extent);
4928 if (key.offset + length <= new_size) {
4929 mutex_unlock(&fs_info->reclaim_bgs_lock);
4930 btrfs_release_path(path);
4934 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4935 btrfs_release_path(path);
4938 * We may be relocating the only data chunk we have,
4939 * which could potentially end up with losing data's
4940 * raid profile, so lets allocate an empty one in
4943 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4945 mutex_unlock(&fs_info->reclaim_bgs_lock);
4949 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4950 mutex_unlock(&fs_info->reclaim_bgs_lock);
4951 if (ret == -ENOSPC) {
4954 if (ret == -ETXTBSY) {
4956 "could not shrink block group %llu due to active swapfile",
4961 } while (key.offset-- > 0);
4963 if (failed && !retried) {
4967 } else if (failed && retried) {
4972 /* Shrinking succeeded, else we would be at "done". */
4973 trans = btrfs_start_transaction(root, 0);
4974 if (IS_ERR(trans)) {
4975 ret = PTR_ERR(trans);
4979 mutex_lock(&fs_info->chunk_mutex);
4980 /* Clear all state bits beyond the shrunk device size */
4981 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
4984 btrfs_device_set_disk_total_bytes(device, new_size);
4985 if (list_empty(&device->post_commit_list))
4986 list_add_tail(&device->post_commit_list,
4987 &trans->transaction->dev_update_list);
4989 WARN_ON(diff > old_total);
4990 btrfs_set_super_total_bytes(super_copy,
4991 round_down(old_total - diff, fs_info->sectorsize));
4992 mutex_unlock(&fs_info->chunk_mutex);
4994 btrfs_reserve_chunk_metadata(trans, false);
4995 /* Now btrfs_update_device() will change the on-disk size. */
4996 ret = btrfs_update_device(trans, device);
4997 btrfs_trans_release_chunk_metadata(trans);
4999 btrfs_abort_transaction(trans, ret);
5000 btrfs_end_transaction(trans);
5002 ret = btrfs_commit_transaction(trans);
5005 btrfs_free_path(path);
5007 mutex_lock(&fs_info->chunk_mutex);
5008 btrfs_device_set_total_bytes(device, old_size);
5009 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
5010 device->fs_devices->total_rw_bytes += diff;
5011 atomic64_add(diff, &fs_info->free_chunk_space);
5012 mutex_unlock(&fs_info->chunk_mutex);
5017 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
5018 struct btrfs_key *key,
5019 struct btrfs_chunk *chunk, int item_size)
5021 struct btrfs_super_block *super_copy = fs_info->super_copy;
5022 struct btrfs_disk_key disk_key;
5026 lockdep_assert_held(&fs_info->chunk_mutex);
5028 array_size = btrfs_super_sys_array_size(super_copy);
5029 if (array_size + item_size + sizeof(disk_key)
5030 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
5033 ptr = super_copy->sys_chunk_array + array_size;
5034 btrfs_cpu_key_to_disk(&disk_key, key);
5035 memcpy(ptr, &disk_key, sizeof(disk_key));
5036 ptr += sizeof(disk_key);
5037 memcpy(ptr, chunk, item_size);
5038 item_size += sizeof(disk_key);
5039 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
5045 * sort the devices in descending order by max_avail, total_avail
5047 static int btrfs_cmp_device_info(const void *a, const void *b)
5049 const struct btrfs_device_info *di_a = a;
5050 const struct btrfs_device_info *di_b = b;
5052 if (di_a->max_avail > di_b->max_avail)
5054 if (di_a->max_avail < di_b->max_avail)
5056 if (di_a->total_avail > di_b->total_avail)
5058 if (di_a->total_avail < di_b->total_avail)
5063 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
5065 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5068 btrfs_set_fs_incompat(info, RAID56);
5071 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
5073 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
5076 btrfs_set_fs_incompat(info, RAID1C34);
5080 * Structure used internally for btrfs_create_chunk() function.
5081 * Wraps needed parameters.
5083 struct alloc_chunk_ctl {
5086 /* Total number of stripes to allocate */
5088 /* sub_stripes info for map */
5090 /* Stripes per device */
5092 /* Maximum number of devices to use */
5094 /* Minimum number of devices to use */
5096 /* ndevs has to be a multiple of this */
5098 /* Number of copies */
5100 /* Number of stripes worth of bytes to store parity information */
5102 u64 max_stripe_size;
5110 static void init_alloc_chunk_ctl_policy_regular(
5111 struct btrfs_fs_devices *fs_devices,
5112 struct alloc_chunk_ctl *ctl)
5114 struct btrfs_space_info *space_info;
5116 space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type);
5119 ctl->max_chunk_size = READ_ONCE(space_info->chunk_size);
5120 ctl->max_stripe_size = ctl->max_chunk_size;
5122 if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM)
5123 ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK);
5125 /* We don't want a chunk larger than 10% of writable space */
5126 ctl->max_chunk_size = min(mult_perc(fs_devices->total_rw_bytes, 10),
5127 ctl->max_chunk_size);
5128 ctl->dev_extent_min = ctl->dev_stripes << BTRFS_STRIPE_LEN_SHIFT;
5131 static void init_alloc_chunk_ctl_policy_zoned(
5132 struct btrfs_fs_devices *fs_devices,
5133 struct alloc_chunk_ctl *ctl)
5135 u64 zone_size = fs_devices->fs_info->zone_size;
5137 int min_num_stripes = ctl->devs_min * ctl->dev_stripes;
5138 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;
5139 u64 min_chunk_size = min_data_stripes * zone_size;
5140 u64 type = ctl->type;
5142 ctl->max_stripe_size = zone_size;
5143 if (type & BTRFS_BLOCK_GROUP_DATA) {
5144 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,
5146 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5147 ctl->max_chunk_size = ctl->max_stripe_size;
5148 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5149 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5150 ctl->devs_max = min_t(int, ctl->devs_max,
5151 BTRFS_MAX_DEVS_SYS_CHUNK);
5156 /* We don't want a chunk larger than 10% of writable space */
5157 limit = max(round_down(mult_perc(fs_devices->total_rw_bytes, 10),
5160 ctl->max_chunk_size = min(limit, ctl->max_chunk_size);
5161 ctl->dev_extent_min = zone_size * ctl->dev_stripes;
5164 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
5165 struct alloc_chunk_ctl *ctl)
5167 int index = btrfs_bg_flags_to_raid_index(ctl->type);
5169 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
5170 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
5171 ctl->devs_max = btrfs_raid_array[index].devs_max;
5173 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
5174 ctl->devs_min = btrfs_raid_array[index].devs_min;
5175 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
5176 ctl->ncopies = btrfs_raid_array[index].ncopies;
5177 ctl->nparity = btrfs_raid_array[index].nparity;
5180 switch (fs_devices->chunk_alloc_policy) {
5181 case BTRFS_CHUNK_ALLOC_REGULAR:
5182 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
5184 case BTRFS_CHUNK_ALLOC_ZONED:
5185 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);
5192 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
5193 struct alloc_chunk_ctl *ctl,
5194 struct btrfs_device_info *devices_info)
5196 struct btrfs_fs_info *info = fs_devices->fs_info;
5197 struct btrfs_device *device;
5199 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
5206 * in the first pass through the devices list, we gather information
5207 * about the available holes on each device.
5209 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5210 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5212 "BTRFS: read-only device in alloc_list\n");
5216 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5217 &device->dev_state) ||
5218 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5221 if (device->total_bytes > device->bytes_used)
5222 total_avail = device->total_bytes - device->bytes_used;
5226 /* If there is no space on this device, skip it. */
5227 if (total_avail < ctl->dev_extent_min)
5230 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
5232 if (ret && ret != -ENOSPC)
5236 max_avail = dev_extent_want;
5238 if (max_avail < ctl->dev_extent_min) {
5239 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5241 "%s: devid %llu has no free space, have=%llu want=%llu",
5242 __func__, device->devid, max_avail,
5243 ctl->dev_extent_min);
5247 if (ndevs == fs_devices->rw_devices) {
5248 WARN(1, "%s: found more than %llu devices\n",
5249 __func__, fs_devices->rw_devices);
5252 devices_info[ndevs].dev_offset = dev_offset;
5253 devices_info[ndevs].max_avail = max_avail;
5254 devices_info[ndevs].total_avail = total_avail;
5255 devices_info[ndevs].dev = device;
5261 * now sort the devices by hole size / available space
5263 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5264 btrfs_cmp_device_info, NULL);
5269 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5270 struct btrfs_device_info *devices_info)
5272 /* Number of stripes that count for block group size */
5276 * The primary goal is to maximize the number of stripes, so use as
5277 * many devices as possible, even if the stripes are not maximum sized.
5279 * The DUP profile stores more than one stripe per device, the
5280 * max_avail is the total size so we have to adjust.
5282 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5284 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5286 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5287 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5290 * Use the number of data stripes to figure out how big this chunk is
5291 * really going to be in terms of logical address space, and compare
5292 * that answer with the max chunk size. If it's higher, we try to
5293 * reduce stripe_size.
5295 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5297 * Reduce stripe_size, round it up to a 16MB boundary again and
5298 * then use it, unless it ends up being even bigger than the
5299 * previous value we had already.
5301 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5302 data_stripes), SZ_16M),
5306 /* Stripe size should not go beyond 1G. */
5307 ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G);
5309 /* Align to BTRFS_STRIPE_LEN */
5310 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5311 ctl->chunk_size = ctl->stripe_size * data_stripes;
5316 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,
5317 struct btrfs_device_info *devices_info)
5319 u64 zone_size = devices_info[0].dev->zone_info->zone_size;
5320 /* Number of stripes that count for block group size */
5324 * It should hold because:
5325 * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5327 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);
5329 ctl->stripe_size = zone_size;
5330 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5331 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5333 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5334 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5335 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,
5336 ctl->stripe_size) + ctl->nparity,
5338 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5339 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5340 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);
5343 ctl->chunk_size = ctl->stripe_size * data_stripes;
5348 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5349 struct alloc_chunk_ctl *ctl,
5350 struct btrfs_device_info *devices_info)
5352 struct btrfs_fs_info *info = fs_devices->fs_info;
5355 * Round down to number of usable stripes, devs_increment can be any
5356 * number so we can't use round_down() that requires power of 2, while
5357 * rounddown is safe.
5359 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5361 if (ctl->ndevs < ctl->devs_min) {
5362 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5364 "%s: not enough devices with free space: have=%d minimum required=%d",
5365 __func__, ctl->ndevs, ctl->devs_min);
5370 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5372 switch (fs_devices->chunk_alloc_policy) {
5373 case BTRFS_CHUNK_ALLOC_REGULAR:
5374 return decide_stripe_size_regular(ctl, devices_info);
5375 case BTRFS_CHUNK_ALLOC_ZONED:
5376 return decide_stripe_size_zoned(ctl, devices_info);
5382 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
5383 struct alloc_chunk_ctl *ctl,
5384 struct btrfs_device_info *devices_info)
5386 struct btrfs_fs_info *info = trans->fs_info;
5387 struct map_lookup *map = NULL;
5388 struct extent_map_tree *em_tree;
5389 struct btrfs_block_group *block_group;
5390 struct extent_map *em;
5391 u64 start = ctl->start;
5392 u64 type = ctl->type;
5397 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
5399 return ERR_PTR(-ENOMEM);
5400 map->num_stripes = ctl->num_stripes;
5402 for (i = 0; i < ctl->ndevs; ++i) {
5403 for (j = 0; j < ctl->dev_stripes; ++j) {
5404 int s = i * ctl->dev_stripes + j;
5405 map->stripes[s].dev = devices_info[i].dev;
5406 map->stripes[s].physical = devices_info[i].dev_offset +
5407 j * ctl->stripe_size;
5410 map->io_align = BTRFS_STRIPE_LEN;
5411 map->io_width = BTRFS_STRIPE_LEN;
5413 map->sub_stripes = ctl->sub_stripes;
5415 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5417 em = alloc_extent_map();
5420 return ERR_PTR(-ENOMEM);
5422 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5423 em->map_lookup = map;
5425 em->len = ctl->chunk_size;
5426 em->block_start = 0;
5427 em->block_len = em->len;
5428 em->orig_block_len = ctl->stripe_size;
5430 em_tree = &info->mapping_tree;
5431 write_lock(&em_tree->lock);
5432 ret = add_extent_mapping(em_tree, em, 0);
5434 write_unlock(&em_tree->lock);
5435 free_extent_map(em);
5436 return ERR_PTR(ret);
5438 write_unlock(&em_tree->lock);
5440 block_group = btrfs_make_block_group(trans, type, start, ctl->chunk_size);
5441 if (IS_ERR(block_group))
5442 goto error_del_extent;
5444 for (i = 0; i < map->num_stripes; i++) {
5445 struct btrfs_device *dev = map->stripes[i].dev;
5447 btrfs_device_set_bytes_used(dev,
5448 dev->bytes_used + ctl->stripe_size);
5449 if (list_empty(&dev->post_commit_list))
5450 list_add_tail(&dev->post_commit_list,
5451 &trans->transaction->dev_update_list);
5454 atomic64_sub(ctl->stripe_size * map->num_stripes,
5455 &info->free_chunk_space);
5457 free_extent_map(em);
5458 check_raid56_incompat_flag(info, type);
5459 check_raid1c34_incompat_flag(info, type);
5464 write_lock(&em_tree->lock);
5465 remove_extent_mapping(em_tree, em);
5466 write_unlock(&em_tree->lock);
5468 /* One for our allocation */
5469 free_extent_map(em);
5470 /* One for the tree reference */
5471 free_extent_map(em);
5476 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
5479 struct btrfs_fs_info *info = trans->fs_info;
5480 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5481 struct btrfs_device_info *devices_info = NULL;
5482 struct alloc_chunk_ctl ctl;
5483 struct btrfs_block_group *block_group;
5486 lockdep_assert_held(&info->chunk_mutex);
5488 if (!alloc_profile_is_valid(type, 0)) {
5490 return ERR_PTR(-EINVAL);
5493 if (list_empty(&fs_devices->alloc_list)) {
5494 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5495 btrfs_debug(info, "%s: no writable device", __func__);
5496 return ERR_PTR(-ENOSPC);
5499 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5500 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5502 return ERR_PTR(-EINVAL);
5505 ctl.start = find_next_chunk(info);
5507 init_alloc_chunk_ctl(fs_devices, &ctl);
5509 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5512 return ERR_PTR(-ENOMEM);
5514 ret = gather_device_info(fs_devices, &ctl, devices_info);
5516 block_group = ERR_PTR(ret);
5520 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5522 block_group = ERR_PTR(ret);
5526 block_group = create_chunk(trans, &ctl, devices_info);
5529 kfree(devices_info);
5534 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5535 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5538 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5541 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
5542 struct btrfs_block_group *bg)
5544 struct btrfs_fs_info *fs_info = trans->fs_info;
5545 struct btrfs_root *chunk_root = fs_info->chunk_root;
5546 struct btrfs_key key;
5547 struct btrfs_chunk *chunk;
5548 struct btrfs_stripe *stripe;
5549 struct extent_map *em;
5550 struct map_lookup *map;
5556 * We take the chunk_mutex for 2 reasons:
5558 * 1) Updates and insertions in the chunk btree must be done while holding
5559 * the chunk_mutex, as well as updating the system chunk array in the
5560 * superblock. See the comment on top of btrfs_chunk_alloc() for the
5563 * 2) To prevent races with the final phase of a device replace operation
5564 * that replaces the device object associated with the map's stripes,
5565 * because the device object's id can change at any time during that
5566 * final phase of the device replace operation
5567 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5568 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5569 * which would cause a failure when updating the device item, which does
5570 * not exists, or persisting a stripe of the chunk item with such ID.
5571 * Here we can't use the device_list_mutex because our caller already
5572 * has locked the chunk_mutex, and the final phase of device replace
5573 * acquires both mutexes - first the device_list_mutex and then the
5574 * chunk_mutex. Using any of those two mutexes protects us from a
5575 * concurrent device replace.
5577 lockdep_assert_held(&fs_info->chunk_mutex);
5579 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
5582 btrfs_abort_transaction(trans, ret);
5586 map = em->map_lookup;
5587 item_size = btrfs_chunk_item_size(map->num_stripes);
5589 chunk = kzalloc(item_size, GFP_NOFS);
5592 btrfs_abort_transaction(trans, ret);
5596 for (i = 0; i < map->num_stripes; i++) {
5597 struct btrfs_device *device = map->stripes[i].dev;
5599 ret = btrfs_update_device(trans, device);
5604 stripe = &chunk->stripe;
5605 for (i = 0; i < map->num_stripes; i++) {
5606 struct btrfs_device *device = map->stripes[i].dev;
5607 const u64 dev_offset = map->stripes[i].physical;
5609 btrfs_set_stack_stripe_devid(stripe, device->devid);
5610 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5611 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5615 btrfs_set_stack_chunk_length(chunk, bg->length);
5616 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
5617 btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
5618 btrfs_set_stack_chunk_type(chunk, map->type);
5619 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5620 btrfs_set_stack_chunk_io_align(chunk, BTRFS_STRIPE_LEN);
5621 btrfs_set_stack_chunk_io_width(chunk, BTRFS_STRIPE_LEN);
5622 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5623 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5625 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5626 key.type = BTRFS_CHUNK_ITEM_KEY;
5627 key.offset = bg->start;
5629 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5633 set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags);
5635 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5636 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5643 free_extent_map(em);
5647 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5649 struct btrfs_fs_info *fs_info = trans->fs_info;
5651 struct btrfs_block_group *meta_bg;
5652 struct btrfs_block_group *sys_bg;
5655 * When adding a new device for sprouting, the seed device is read-only
5656 * so we must first allocate a metadata and a system chunk. But before
5657 * adding the block group items to the extent, device and chunk btrees,
5660 * 1) Create both chunks without doing any changes to the btrees, as
5661 * otherwise we would get -ENOSPC since the block groups from the
5662 * seed device are read-only;
5664 * 2) Add the device item for the new sprout device - finishing the setup
5665 * of a new block group requires updating the device item in the chunk
5666 * btree, so it must exist when we attempt to do it. The previous step
5667 * ensures this does not fail with -ENOSPC.
5669 * After that we can add the block group items to their btrees:
5670 * update existing device item in the chunk btree, add a new block group
5671 * item to the extent btree, add a new chunk item to the chunk btree and
5672 * finally add the new device extent items to the devices btree.
5675 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5676 meta_bg = btrfs_create_chunk(trans, alloc_profile);
5677 if (IS_ERR(meta_bg))
5678 return PTR_ERR(meta_bg);
5680 alloc_profile = btrfs_system_alloc_profile(fs_info);
5681 sys_bg = btrfs_create_chunk(trans, alloc_profile);
5683 return PTR_ERR(sys_bg);
5688 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5690 const int index = btrfs_bg_flags_to_raid_index(map->type);
5692 return btrfs_raid_array[index].tolerated_failures;
5695 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5697 struct extent_map *em;
5698 struct map_lookup *map;
5703 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5707 map = em->map_lookup;
5708 for (i = 0; i < map->num_stripes; i++) {
5709 if (test_bit(BTRFS_DEV_STATE_MISSING,
5710 &map->stripes[i].dev->dev_state)) {
5714 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5715 &map->stripes[i].dev->dev_state)) {
5722 * If the number of missing devices is larger than max errors, we can
5723 * not write the data into that chunk successfully.
5725 if (miss_ndevs > btrfs_chunk_max_errors(map))
5728 free_extent_map(em);
5732 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5734 struct extent_map *em;
5737 write_lock(&tree->lock);
5738 em = lookup_extent_mapping(tree, 0, (u64)-1);
5740 remove_extent_mapping(tree, em);
5741 write_unlock(&tree->lock);
5745 free_extent_map(em);
5746 /* once for the tree */
5747 free_extent_map(em);
5751 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5753 struct extent_map *em;
5754 struct map_lookup *map;
5755 enum btrfs_raid_types index;
5758 em = btrfs_get_chunk_map(fs_info, logical, len);
5761 * We could return errors for these cases, but that could get
5762 * ugly and we'd probably do the same thing which is just not do
5763 * anything else and exit, so return 1 so the callers don't try
5764 * to use other copies.
5768 map = em->map_lookup;
5769 index = btrfs_bg_flags_to_raid_index(map->type);
5771 /* Non-RAID56, use their ncopies from btrfs_raid_array. */
5772 if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5773 ret = btrfs_raid_array[index].ncopies;
5774 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5776 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5778 * There could be two corrupted data stripes, we need
5779 * to loop retry in order to rebuild the correct data.
5781 * Fail a stripe at a time on every retry except the
5782 * stripe under reconstruction.
5784 ret = map->num_stripes;
5785 free_extent_map(em);
5789 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5792 struct extent_map *em;
5793 struct map_lookup *map;
5794 unsigned long len = fs_info->sectorsize;
5796 if (!btrfs_fs_incompat(fs_info, RAID56))
5799 em = btrfs_get_chunk_map(fs_info, logical, len);
5801 if (!WARN_ON(IS_ERR(em))) {
5802 map = em->map_lookup;
5803 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5804 len = nr_data_stripes(map) << BTRFS_STRIPE_LEN_SHIFT;
5805 free_extent_map(em);
5810 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5812 struct extent_map *em;
5813 struct map_lookup *map;
5816 if (!btrfs_fs_incompat(fs_info, RAID56))
5819 em = btrfs_get_chunk_map(fs_info, logical, len);
5821 if(!WARN_ON(IS_ERR(em))) {
5822 map = em->map_lookup;
5823 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5825 free_extent_map(em);
5830 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5831 struct map_lookup *map, int first,
5832 int dev_replace_is_ongoing)
5836 int preferred_mirror;
5838 struct btrfs_device *srcdev;
5841 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5843 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5844 num_stripes = map->sub_stripes;
5846 num_stripes = map->num_stripes;
5848 switch (fs_info->fs_devices->read_policy) {
5850 /* Shouldn't happen, just warn and use pid instead of failing */
5851 btrfs_warn_rl(fs_info,
5852 "unknown read_policy type %u, reset to pid",
5853 fs_info->fs_devices->read_policy);
5854 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID;
5856 case BTRFS_READ_POLICY_PID:
5857 preferred_mirror = first + (current->pid % num_stripes);
5861 if (dev_replace_is_ongoing &&
5862 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5863 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5864 srcdev = fs_info->dev_replace.srcdev;
5869 * try to avoid the drive that is the source drive for a
5870 * dev-replace procedure, only choose it if no other non-missing
5871 * mirror is available
5873 for (tolerance = 0; tolerance < 2; tolerance++) {
5874 if (map->stripes[preferred_mirror].dev->bdev &&
5875 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5876 return preferred_mirror;
5877 for (i = first; i < first + num_stripes; i++) {
5878 if (map->stripes[i].dev->bdev &&
5879 (tolerance || map->stripes[i].dev != srcdev))
5884 /* we couldn't find one that doesn't fail. Just return something
5885 * and the io error handling code will clean up eventually
5887 return preferred_mirror;
5890 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info,
5893 struct btrfs_io_context *bioc;
5896 /* The size of btrfs_io_context */
5897 sizeof(struct btrfs_io_context) +
5898 /* Plus the variable array for the stripes */
5899 sizeof(struct btrfs_io_stripe) * (total_stripes),
5905 refcount_set(&bioc->refs, 1);
5907 bioc->fs_info = fs_info;
5908 bioc->replace_stripe_src = -1;
5909 bioc->full_stripe_logical = (u64)-1;
5914 void btrfs_get_bioc(struct btrfs_io_context *bioc)
5916 WARN_ON(!refcount_read(&bioc->refs));
5917 refcount_inc(&bioc->refs);
5920 void btrfs_put_bioc(struct btrfs_io_context *bioc)
5924 if (refcount_dec_and_test(&bioc->refs))
5929 * Please note that, discard won't be sent to target device of device
5932 struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info,
5933 u64 logical, u64 *length_ret,
5936 struct extent_map *em;
5937 struct map_lookup *map;
5938 struct btrfs_discard_stripe *stripes;
5939 u64 length = *length_ret;
5944 u64 stripe_end_offset;
5948 u32 sub_stripes = 0;
5949 u32 stripes_per_dev = 0;
5950 u32 remaining_stripes = 0;
5951 u32 last_stripe = 0;
5955 em = btrfs_get_chunk_map(fs_info, logical, length);
5957 return ERR_CAST(em);
5959 map = em->map_lookup;
5961 /* we don't discard raid56 yet */
5962 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5967 offset = logical - em->start;
5968 length = min_t(u64, em->start + em->len - logical, length);
5969 *length_ret = length;
5972 * stripe_nr counts the total number of stripes we have to stride
5973 * to get to this block
5975 stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
5977 /* stripe_offset is the offset of this block in its stripe */
5978 stripe_offset = offset - (stripe_nr << BTRFS_STRIPE_LEN_SHIFT);
5980 stripe_nr_end = round_up(offset + length, BTRFS_STRIPE_LEN) >>
5981 BTRFS_STRIPE_LEN_SHIFT;
5982 stripe_cnt = stripe_nr_end - stripe_nr;
5983 stripe_end_offset = (stripe_nr_end << BTRFS_STRIPE_LEN_SHIFT) -
5986 * after this, stripe_nr is the number of stripes on this
5987 * device we have to walk to find the data, and stripe_index is
5988 * the number of our device in the stripe array
5992 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5993 BTRFS_BLOCK_GROUP_RAID10)) {
5994 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5997 sub_stripes = map->sub_stripes;
5999 factor = map->num_stripes / sub_stripes;
6000 *num_stripes = min_t(u64, map->num_stripes,
6001 sub_stripes * stripe_cnt);
6002 stripe_index = stripe_nr % factor;
6003 stripe_nr /= factor;
6004 stripe_index *= sub_stripes;
6006 remaining_stripes = stripe_cnt % factor;
6007 stripes_per_dev = stripe_cnt / factor;
6008 last_stripe = ((stripe_nr_end - 1) % factor) * sub_stripes;
6009 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
6010 BTRFS_BLOCK_GROUP_DUP)) {
6011 *num_stripes = map->num_stripes;
6013 stripe_index = stripe_nr % map->num_stripes;
6014 stripe_nr /= map->num_stripes;
6017 stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS);
6023 for (i = 0; i < *num_stripes; i++) {
6024 stripes[i].physical =
6025 map->stripes[stripe_index].physical +
6026 stripe_offset + (stripe_nr << BTRFS_STRIPE_LEN_SHIFT);
6027 stripes[i].dev = map->stripes[stripe_index].dev;
6029 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6030 BTRFS_BLOCK_GROUP_RAID10)) {
6031 stripes[i].length = stripes_per_dev << BTRFS_STRIPE_LEN_SHIFT;
6033 if (i / sub_stripes < remaining_stripes)
6034 stripes[i].length += BTRFS_STRIPE_LEN;
6037 * Special for the first stripe and
6040 * |-------|...|-------|
6044 if (i < sub_stripes)
6045 stripes[i].length -= stripe_offset;
6047 if (stripe_index >= last_stripe &&
6048 stripe_index <= (last_stripe +
6050 stripes[i].length -= stripe_end_offset;
6052 if (i == sub_stripes - 1)
6055 stripes[i].length = length;
6059 if (stripe_index == map->num_stripes) {
6065 free_extent_map(em);
6068 free_extent_map(em);
6069 return ERR_PTR(ret);
6072 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)
6074 struct btrfs_block_group *cache;
6077 /* Non zoned filesystem does not use "to_copy" flag */
6078 if (!btrfs_is_zoned(fs_info))
6081 cache = btrfs_lookup_block_group(fs_info, logical);
6083 ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags);
6085 btrfs_put_block_group(cache);
6089 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
6090 struct btrfs_io_context *bioc,
6091 struct btrfs_dev_replace *dev_replace,
6093 int *num_stripes_ret, int *max_errors_ret)
6095 u64 srcdev_devid = dev_replace->srcdev->devid;
6097 * At this stage, num_stripes is still the real number of stripes,
6098 * excluding the duplicated stripes.
6100 int num_stripes = *num_stripes_ret;
6101 int nr_extra_stripes = 0;
6102 int max_errors = *max_errors_ret;
6106 * A block group which has "to_copy" set will eventually be copied by
6107 * the dev-replace process. We can avoid cloning IO here.
6109 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))
6113 * Duplicate the write operations while the dev-replace procedure is
6114 * running. Since the copying of the old disk to the new disk takes
6115 * place at run time while the filesystem is mounted writable, the
6116 * regular write operations to the old disk have to be duplicated to go
6117 * to the new disk as well.
6119 * Note that device->missing is handled by the caller, and that the
6120 * write to the old disk is already set up in the stripes array.
6122 for (i = 0; i < num_stripes; i++) {
6123 struct btrfs_io_stripe *old = &bioc->stripes[i];
6124 struct btrfs_io_stripe *new = &bioc->stripes[num_stripes + nr_extra_stripes];
6126 if (old->dev->devid != srcdev_devid)
6129 new->physical = old->physical;
6130 new->dev = dev_replace->tgtdev;
6131 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK)
6132 bioc->replace_stripe_src = i;
6136 /* We can only have at most 2 extra nr_stripes (for DUP). */
6137 ASSERT(nr_extra_stripes <= 2);
6139 * For GET_READ_MIRRORS, we can only return at most 1 extra stripe for
6141 * If we have 2 extra stripes, only choose the one with smaller physical.
6143 if (op == BTRFS_MAP_GET_READ_MIRRORS && nr_extra_stripes == 2) {
6144 struct btrfs_io_stripe *first = &bioc->stripes[num_stripes];
6145 struct btrfs_io_stripe *second = &bioc->stripes[num_stripes + 1];
6147 /* Only DUP can have two extra stripes. */
6148 ASSERT(bioc->map_type & BTRFS_BLOCK_GROUP_DUP);
6151 * Swap the last stripe stripes and reduce @nr_extra_stripes.
6152 * The extra stripe would still be there, but won't be accessed.
6154 if (first->physical > second->physical) {
6155 swap(second->physical, first->physical);
6156 swap(second->dev, first->dev);
6161 *num_stripes_ret = num_stripes + nr_extra_stripes;
6162 *max_errors_ret = max_errors + nr_extra_stripes;
6163 bioc->replace_nr_stripes = nr_extra_stripes;
6166 static bool need_full_stripe(enum btrfs_map_op op)
6168 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
6171 static u64 btrfs_max_io_len(struct map_lookup *map, enum btrfs_map_op op,
6172 u64 offset, u32 *stripe_nr, u64 *stripe_offset,
6173 u64 *full_stripe_start)
6175 ASSERT(op != BTRFS_MAP_DISCARD);
6178 * Stripe_nr is the stripe where this block falls. stripe_offset is
6179 * the offset of this block in its stripe.
6181 *stripe_offset = offset & BTRFS_STRIPE_LEN_MASK;
6182 *stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
6183 ASSERT(*stripe_offset < U32_MAX);
6185 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6186 unsigned long full_stripe_len = nr_data_stripes(map) <<
6187 BTRFS_STRIPE_LEN_SHIFT;
6190 * For full stripe start, we use previously calculated
6191 * @stripe_nr. Align it to nr_data_stripes, then multiply with
6194 * By this we can avoid u64 division completely. And we have
6195 * to go rounddown(), not round_down(), as nr_data_stripes is
6196 * not ensured to be power of 2.
6198 *full_stripe_start =
6199 rounddown(*stripe_nr, nr_data_stripes(map)) <<
6200 BTRFS_STRIPE_LEN_SHIFT;
6203 * For writes to RAID56, allow to write a full stripe set, but
6204 * no straddling of stripe sets.
6206 if (op == BTRFS_MAP_WRITE)
6207 return full_stripe_len - (offset - *full_stripe_start);
6211 * For other RAID types and for RAID56 reads, allow a single stripe (on
6214 if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK)
6215 return BTRFS_STRIPE_LEN - *stripe_offset;
6219 static void set_io_stripe(struct btrfs_io_stripe *dst, const struct map_lookup *map,
6220 u32 stripe_index, u64 stripe_offset, u32 stripe_nr)
6222 dst->dev = map->stripes[stripe_index].dev;
6223 dst->physical = map->stripes[stripe_index].physical +
6224 stripe_offset + (stripe_nr << BTRFS_STRIPE_LEN_SHIFT);
6227 int __btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6228 u64 logical, u64 *length,
6229 struct btrfs_io_context **bioc_ret,
6230 struct btrfs_io_stripe *smap, int *mirror_num_ret,
6233 struct extent_map *em;
6234 struct map_lookup *map;
6242 int mirror_num = (mirror_num_ret ? *mirror_num_ret : 0);
6246 struct btrfs_io_context *bioc = NULL;
6247 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6248 int dev_replace_is_ongoing = 0;
6249 u16 num_alloc_stripes;
6250 u64 raid56_full_stripe_start = (u64)-1;
6254 ASSERT(op != BTRFS_MAP_DISCARD);
6256 num_copies = btrfs_num_copies(fs_info, logical, fs_info->sectorsize);
6257 if (mirror_num > num_copies)
6260 em = btrfs_get_chunk_map(fs_info, logical, *length);
6264 map = em->map_lookup;
6265 data_stripes = nr_data_stripes(map);
6267 map_offset = logical - em->start;
6268 max_len = btrfs_max_io_len(map, op, map_offset, &stripe_nr,
6269 &stripe_offset, &raid56_full_stripe_start);
6270 *length = min_t(u64, em->len - map_offset, max_len);
6272 down_read(&dev_replace->rwsem);
6273 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6275 * Hold the semaphore for read during the whole operation, write is
6276 * requested at commit time but must wait.
6278 if (!dev_replace_is_ongoing)
6279 up_read(&dev_replace->rwsem);
6283 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6284 stripe_index = stripe_nr % map->num_stripes;
6285 stripe_nr /= map->num_stripes;
6286 if (!need_full_stripe(op))
6288 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6289 if (need_full_stripe(op))
6290 num_stripes = map->num_stripes;
6291 else if (mirror_num)
6292 stripe_index = mirror_num - 1;
6294 stripe_index = find_live_mirror(fs_info, map, 0,
6295 dev_replace_is_ongoing);
6296 mirror_num = stripe_index + 1;
6299 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6300 if (need_full_stripe(op)) {
6301 num_stripes = map->num_stripes;
6302 } else if (mirror_num) {
6303 stripe_index = mirror_num - 1;
6308 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6309 u32 factor = map->num_stripes / map->sub_stripes;
6311 stripe_index = (stripe_nr % factor) * map->sub_stripes;
6312 stripe_nr /= factor;
6314 if (need_full_stripe(op))
6315 num_stripes = map->sub_stripes;
6316 else if (mirror_num)
6317 stripe_index += mirror_num - 1;
6319 int old_stripe_index = stripe_index;
6320 stripe_index = find_live_mirror(fs_info, map,
6322 dev_replace_is_ongoing);
6323 mirror_num = stripe_index - old_stripe_index + 1;
6326 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6327 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6329 * Push stripe_nr back to the start of the full stripe
6330 * For those cases needing a full stripe, @stripe_nr
6331 * is the full stripe number.
6333 * Originally we go raid56_full_stripe_start / full_stripe_len,
6334 * but that can be expensive. Here we just divide
6335 * @stripe_nr with @data_stripes.
6337 stripe_nr /= data_stripes;
6339 /* RAID[56] write or recovery. Return all stripes */
6340 num_stripes = map->num_stripes;
6341 max_errors = btrfs_chunk_max_errors(map);
6343 /* Return the length to the full stripe end */
6344 *length = min(logical + *length,
6345 raid56_full_stripe_start + em->start +
6346 (data_stripes << BTRFS_STRIPE_LEN_SHIFT)) - logical;
6351 * Mirror #0 or #1 means the original data block.
6352 * Mirror #2 is RAID5 parity block.
6353 * Mirror #3 is RAID6 Q block.
6355 stripe_index = stripe_nr % data_stripes;
6356 stripe_nr /= data_stripes;
6358 stripe_index = data_stripes + mirror_num - 2;
6360 /* We distribute the parity blocks across stripes */
6361 stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
6362 if (!need_full_stripe(op) && mirror_num <= 1)
6367 * After this, stripe_nr is the number of stripes on this
6368 * device we have to walk to find the data, and stripe_index is
6369 * the number of our device in the stripe array
6371 stripe_index = stripe_nr % map->num_stripes;
6372 stripe_nr /= map->num_stripes;
6373 mirror_num = stripe_index + 1;
6375 if (stripe_index >= map->num_stripes) {
6377 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6378 stripe_index, map->num_stripes);
6383 num_alloc_stripes = num_stripes;
6384 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6385 op != BTRFS_MAP_READ)
6387 * For replace case, we need to add extra stripes for extra
6388 * duplicated stripes.
6390 * For both WRITE and GET_READ_MIRRORS, we may have at most
6391 * 2 more stripes (DUP types, otherwise 1).
6393 num_alloc_stripes += 2;
6396 * If this I/O maps to a single device, try to return the device and
6397 * physical block information on the stack instead of allocating an
6398 * I/O context structure.
6400 if (smap && num_alloc_stripes == 1 &&
6401 !((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1) &&
6402 (!need_full_stripe(op) || !dev_replace_is_ongoing ||
6403 !dev_replace->tgtdev)) {
6404 set_io_stripe(smap, map, stripe_index, stripe_offset, stripe_nr);
6405 *mirror_num_ret = mirror_num;
6411 bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes);
6416 bioc->map_type = map->type;
6419 * For RAID56 full map, we need to make sure the stripes[] follows the
6420 * rule that data stripes are all ordered, then followed with P and Q
6423 * It's still mostly the same as other profiles, just with extra rotation.
6425 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6426 (need_full_stripe(op) || mirror_num > 1)) {
6428 * For RAID56 @stripe_nr is already the number of full stripes
6429 * before us, which is also the rotation value (needs to modulo
6430 * with num_stripes).
6432 * In this case, we just add @stripe_nr with @i, then do the
6433 * modulo, to reduce one modulo call.
6435 bioc->full_stripe_logical = em->start +
6436 ((stripe_nr * data_stripes) << BTRFS_STRIPE_LEN_SHIFT);
6437 for (i = 0; i < num_stripes; i++)
6438 set_io_stripe(&bioc->stripes[i], map,
6439 (i + stripe_nr) % num_stripes,
6440 stripe_offset, stripe_nr);
6443 * For all other non-RAID56 profiles, just copy the target
6444 * stripe into the bioc.
6446 for (i = 0; i < num_stripes; i++) {
6447 set_io_stripe(&bioc->stripes[i], map, stripe_index,
6448 stripe_offset, stripe_nr);
6453 if (need_full_stripe(op))
6454 max_errors = btrfs_chunk_max_errors(map);
6456 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6457 need_full_stripe(op)) {
6458 handle_ops_on_dev_replace(op, bioc, dev_replace, logical,
6459 &num_stripes, &max_errors);
6463 bioc->num_stripes = num_stripes;
6464 bioc->max_errors = max_errors;
6465 bioc->mirror_num = mirror_num;
6468 if (dev_replace_is_ongoing) {
6469 lockdep_assert_held(&dev_replace->rwsem);
6470 /* Unlock and let waiting writers proceed */
6471 up_read(&dev_replace->rwsem);
6473 free_extent_map(em);
6477 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6478 u64 logical, u64 *length,
6479 struct btrfs_io_context **bioc_ret, int mirror_num)
6481 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret,
6482 NULL, &mirror_num, 0);
6485 /* For Scrub/replace */
6486 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6487 u64 logical, u64 *length,
6488 struct btrfs_io_context **bioc_ret)
6490 return __btrfs_map_block(fs_info, op, logical, length, bioc_ret,
6494 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args,
6495 const struct btrfs_fs_devices *fs_devices)
6497 if (args->fsid == NULL)
6499 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0)
6504 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args,
6505 const struct btrfs_device *device)
6507 if (args->missing) {
6508 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) &&
6514 if (device->devid != args->devid)
6516 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0)
6522 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6525 * If devid and uuid are both specified, the match must be exact, otherwise
6526 * only devid is used.
6528 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,
6529 const struct btrfs_dev_lookup_args *args)
6531 struct btrfs_device *device;
6532 struct btrfs_fs_devices *seed_devs;
6534 if (dev_args_match_fs_devices(args, fs_devices)) {
6535 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6536 if (dev_args_match_device(args, device))
6541 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6542 if (!dev_args_match_fs_devices(args, seed_devs))
6544 list_for_each_entry(device, &seed_devs->devices, dev_list) {
6545 if (dev_args_match_device(args, device))
6553 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6554 u64 devid, u8 *dev_uuid)
6556 struct btrfs_device *device;
6557 unsigned int nofs_flag;
6560 * We call this under the chunk_mutex, so we want to use NOFS for this
6561 * allocation, however we don't want to change btrfs_alloc_device() to
6562 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6566 nofs_flag = memalloc_nofs_save();
6567 device = btrfs_alloc_device(NULL, &devid, dev_uuid, NULL);
6568 memalloc_nofs_restore(nofs_flag);
6572 list_add(&device->dev_list, &fs_devices->devices);
6573 device->fs_devices = fs_devices;
6574 fs_devices->num_devices++;
6576 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6577 fs_devices->missing_devices++;
6583 * Allocate new device struct, set up devid and UUID.
6585 * @fs_info: used only for generating a new devid, can be NULL if
6586 * devid is provided (i.e. @devid != NULL).
6587 * @devid: a pointer to devid for this device. If NULL a new devid
6589 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6591 * @path: a pointer to device path if available, NULL otherwise.
6593 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6594 * on error. Returned struct is not linked onto any lists and must be
6595 * destroyed with btrfs_free_device.
6597 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6598 const u64 *devid, const u8 *uuid,
6601 struct btrfs_device *dev;
6604 if (WARN_ON(!devid && !fs_info))
6605 return ERR_PTR(-EINVAL);
6607 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
6609 return ERR_PTR(-ENOMEM);
6611 INIT_LIST_HEAD(&dev->dev_list);
6612 INIT_LIST_HEAD(&dev->dev_alloc_list);
6613 INIT_LIST_HEAD(&dev->post_commit_list);
6615 atomic_set(&dev->dev_stats_ccnt, 0);
6616 btrfs_device_data_ordered_init(dev);
6617 extent_io_tree_init(fs_info, &dev->alloc_state, IO_TREE_DEVICE_ALLOC_STATE);
6624 ret = find_next_devid(fs_info, &tmp);
6626 btrfs_free_device(dev);
6627 return ERR_PTR(ret);
6633 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6635 generate_random_uuid(dev->uuid);
6638 struct rcu_string *name;
6640 name = rcu_string_strdup(path, GFP_KERNEL);
6642 btrfs_free_device(dev);
6643 return ERR_PTR(-ENOMEM);
6645 rcu_assign_pointer(dev->name, name);
6651 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6652 u64 devid, u8 *uuid, bool error)
6655 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6658 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6662 u64 btrfs_calc_stripe_length(const struct extent_map *em)
6664 const struct map_lookup *map = em->map_lookup;
6665 const int data_stripes = calc_data_stripes(map->type, map->num_stripes);
6667 return div_u64(em->len, data_stripes);
6670 #if BITS_PER_LONG == 32
6672 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
6673 * can't be accessed on 32bit systems.
6675 * This function do mount time check to reject the fs if it already has
6676 * metadata chunk beyond that limit.
6678 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6679 u64 logical, u64 length, u64 type)
6681 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6684 if (logical + length < MAX_LFS_FILESIZE)
6687 btrfs_err_32bit_limit(fs_info);
6692 * This is to give early warning for any metadata chunk reaching
6693 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
6694 * Although we can still access the metadata, it's not going to be possible
6695 * once the limit is reached.
6697 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6698 u64 logical, u64 length, u64 type)
6700 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6703 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)
6706 btrfs_warn_32bit_limit(fs_info);
6710 static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info,
6711 u64 devid, u8 *uuid)
6713 struct btrfs_device *dev;
6715 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6716 btrfs_report_missing_device(fs_info, devid, uuid, true);
6717 return ERR_PTR(-ENOENT);
6720 dev = add_missing_dev(fs_info->fs_devices, devid, uuid);
6722 btrfs_err(fs_info, "failed to init missing device %llu: %ld",
6723 devid, PTR_ERR(dev));
6726 btrfs_report_missing_device(fs_info, devid, uuid, false);
6731 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6732 struct btrfs_chunk *chunk)
6734 BTRFS_DEV_LOOKUP_ARGS(args);
6735 struct btrfs_fs_info *fs_info = leaf->fs_info;
6736 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6737 struct map_lookup *map;
6738 struct extent_map *em;
6743 u8 uuid[BTRFS_UUID_SIZE];
6749 logical = key->offset;
6750 length = btrfs_chunk_length(leaf, chunk);
6751 type = btrfs_chunk_type(leaf, chunk);
6752 index = btrfs_bg_flags_to_raid_index(type);
6753 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6755 #if BITS_PER_LONG == 32
6756 ret = check_32bit_meta_chunk(fs_info, logical, length, type);
6759 warn_32bit_meta_chunk(fs_info, logical, length, type);
6763 * Only need to verify chunk item if we're reading from sys chunk array,
6764 * as chunk item in tree block is already verified by tree-checker.
6766 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6767 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6772 read_lock(&map_tree->lock);
6773 em = lookup_extent_mapping(map_tree, logical, 1);
6774 read_unlock(&map_tree->lock);
6776 /* already mapped? */
6777 if (em && em->start <= logical && em->start + em->len > logical) {
6778 free_extent_map(em);
6781 free_extent_map(em);
6784 em = alloc_extent_map();
6787 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6789 free_extent_map(em);
6793 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6794 em->map_lookup = map;
6795 em->start = logical;
6798 em->block_start = 0;
6799 em->block_len = em->len;
6801 map->num_stripes = num_stripes;
6802 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6803 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6806 * We can't use the sub_stripes value, as for profiles other than
6807 * RAID10, they may have 0 as sub_stripes for filesystems created by
6808 * older mkfs (<v5.4).
6809 * In that case, it can cause divide-by-zero errors later.
6810 * Since currently sub_stripes is fixed for each profile, let's
6811 * use the trusted value instead.
6813 map->sub_stripes = btrfs_raid_array[index].sub_stripes;
6814 map->verified_stripes = 0;
6815 em->orig_block_len = btrfs_calc_stripe_length(em);
6816 for (i = 0; i < num_stripes; i++) {
6817 map->stripes[i].physical =
6818 btrfs_stripe_offset_nr(leaf, chunk, i);
6819 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6821 read_extent_buffer(leaf, uuid, (unsigned long)
6822 btrfs_stripe_dev_uuid_nr(chunk, i),
6825 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args);
6826 if (!map->stripes[i].dev) {
6827 map->stripes[i].dev = handle_missing_device(fs_info,
6829 if (IS_ERR(map->stripes[i].dev)) {
6830 ret = PTR_ERR(map->stripes[i].dev);
6831 free_extent_map(em);
6836 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6837 &(map->stripes[i].dev->dev_state));
6840 write_lock(&map_tree->lock);
6841 ret = add_extent_mapping(map_tree, em, 0);
6842 write_unlock(&map_tree->lock);
6845 "failed to add chunk map, start=%llu len=%llu: %d",
6846 em->start, em->len, ret);
6848 free_extent_map(em);
6853 static void fill_device_from_item(struct extent_buffer *leaf,
6854 struct btrfs_dev_item *dev_item,
6855 struct btrfs_device *device)
6859 device->devid = btrfs_device_id(leaf, dev_item);
6860 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6861 device->total_bytes = device->disk_total_bytes;
6862 device->commit_total_bytes = device->disk_total_bytes;
6863 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6864 device->commit_bytes_used = device->bytes_used;
6865 device->type = btrfs_device_type(leaf, dev_item);
6866 device->io_align = btrfs_device_io_align(leaf, dev_item);
6867 device->io_width = btrfs_device_io_width(leaf, dev_item);
6868 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6869 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6870 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6872 ptr = btrfs_device_uuid(dev_item);
6873 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6876 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6879 struct btrfs_fs_devices *fs_devices;
6882 lockdep_assert_held(&uuid_mutex);
6885 /* This will match only for multi-device seed fs */
6886 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
6887 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6891 fs_devices = find_fsid(fsid, NULL);
6893 if (!btrfs_test_opt(fs_info, DEGRADED))
6894 return ERR_PTR(-ENOENT);
6896 fs_devices = alloc_fs_devices(fsid, NULL);
6897 if (IS_ERR(fs_devices))
6900 fs_devices->seeding = true;
6901 fs_devices->opened = 1;
6906 * Upon first call for a seed fs fsid, just create a private copy of the
6907 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
6909 fs_devices = clone_fs_devices(fs_devices);
6910 if (IS_ERR(fs_devices))
6913 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
6915 free_fs_devices(fs_devices);
6916 return ERR_PTR(ret);
6919 if (!fs_devices->seeding) {
6920 close_fs_devices(fs_devices);
6921 free_fs_devices(fs_devices);
6922 return ERR_PTR(-EINVAL);
6925 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
6930 static int read_one_dev(struct extent_buffer *leaf,
6931 struct btrfs_dev_item *dev_item)
6933 BTRFS_DEV_LOOKUP_ARGS(args);
6934 struct btrfs_fs_info *fs_info = leaf->fs_info;
6935 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6936 struct btrfs_device *device;
6939 u8 fs_uuid[BTRFS_FSID_SIZE];
6940 u8 dev_uuid[BTRFS_UUID_SIZE];
6942 devid = btrfs_device_id(leaf, dev_item);
6944 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6946 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6948 args.uuid = dev_uuid;
6949 args.fsid = fs_uuid;
6951 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
6952 fs_devices = open_seed_devices(fs_info, fs_uuid);
6953 if (IS_ERR(fs_devices))
6954 return PTR_ERR(fs_devices);
6957 device = btrfs_find_device(fs_info->fs_devices, &args);
6959 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6960 btrfs_report_missing_device(fs_info, devid,
6965 device = add_missing_dev(fs_devices, devid, dev_uuid);
6966 if (IS_ERR(device)) {
6968 "failed to add missing dev %llu: %ld",
6969 devid, PTR_ERR(device));
6970 return PTR_ERR(device);
6972 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6974 if (!device->bdev) {
6975 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6976 btrfs_report_missing_device(fs_info,
6977 devid, dev_uuid, true);
6980 btrfs_report_missing_device(fs_info, devid,
6984 if (!device->bdev &&
6985 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6987 * this happens when a device that was properly setup
6988 * in the device info lists suddenly goes bad.
6989 * device->bdev is NULL, and so we have to set
6990 * device->missing to one here
6992 device->fs_devices->missing_devices++;
6993 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6996 /* Move the device to its own fs_devices */
6997 if (device->fs_devices != fs_devices) {
6998 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6999 &device->dev_state));
7001 list_move(&device->dev_list, &fs_devices->devices);
7002 device->fs_devices->num_devices--;
7003 fs_devices->num_devices++;
7005 device->fs_devices->missing_devices--;
7006 fs_devices->missing_devices++;
7008 device->fs_devices = fs_devices;
7012 if (device->fs_devices != fs_info->fs_devices) {
7013 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
7014 if (device->generation !=
7015 btrfs_device_generation(leaf, dev_item))
7019 fill_device_from_item(leaf, dev_item, device);
7021 u64 max_total_bytes = bdev_nr_bytes(device->bdev);
7023 if (device->total_bytes > max_total_bytes) {
7025 "device total_bytes should be at most %llu but found %llu",
7026 max_total_bytes, device->total_bytes);
7030 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7031 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7032 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7033 device->fs_devices->total_rw_bytes += device->total_bytes;
7034 atomic64_add(device->total_bytes - device->bytes_used,
7035 &fs_info->free_chunk_space);
7041 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7043 struct btrfs_super_block *super_copy = fs_info->super_copy;
7044 struct extent_buffer *sb;
7045 struct btrfs_disk_key *disk_key;
7046 struct btrfs_chunk *chunk;
7048 unsigned long sb_array_offset;
7055 struct btrfs_key key;
7057 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7060 * We allocated a dummy extent, just to use extent buffer accessors.
7061 * There will be unused space after BTRFS_SUPER_INFO_SIZE, but
7062 * that's fine, we will not go beyond system chunk array anyway.
7064 sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET);
7067 set_extent_buffer_uptodate(sb);
7069 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7070 array_size = btrfs_super_sys_array_size(super_copy);
7072 array_ptr = super_copy->sys_chunk_array;
7073 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7076 while (cur_offset < array_size) {
7077 disk_key = (struct btrfs_disk_key *)array_ptr;
7078 len = sizeof(*disk_key);
7079 if (cur_offset + len > array_size)
7080 goto out_short_read;
7082 btrfs_disk_key_to_cpu(&key, disk_key);
7085 sb_array_offset += len;
7088 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7090 "unexpected item type %u in sys_array at offset %u",
7091 (u32)key.type, cur_offset);
7096 chunk = (struct btrfs_chunk *)sb_array_offset;
7098 * At least one btrfs_chunk with one stripe must be present,
7099 * exact stripe count check comes afterwards
7101 len = btrfs_chunk_item_size(1);
7102 if (cur_offset + len > array_size)
7103 goto out_short_read;
7105 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7108 "invalid number of stripes %u in sys_array at offset %u",
7109 num_stripes, cur_offset);
7114 type = btrfs_chunk_type(sb, chunk);
7115 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7117 "invalid chunk type %llu in sys_array at offset %u",
7123 len = btrfs_chunk_item_size(num_stripes);
7124 if (cur_offset + len > array_size)
7125 goto out_short_read;
7127 ret = read_one_chunk(&key, sb, chunk);
7132 sb_array_offset += len;
7135 clear_extent_buffer_uptodate(sb);
7136 free_extent_buffer_stale(sb);
7140 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7142 clear_extent_buffer_uptodate(sb);
7143 free_extent_buffer_stale(sb);
7148 * Check if all chunks in the fs are OK for read-write degraded mount
7150 * If the @failing_dev is specified, it's accounted as missing.
7152 * Return true if all chunks meet the minimal RW mount requirements.
7153 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7155 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7156 struct btrfs_device *failing_dev)
7158 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7159 struct extent_map *em;
7163 read_lock(&map_tree->lock);
7164 em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7165 read_unlock(&map_tree->lock);
7166 /* No chunk at all? Return false anyway */
7172 struct map_lookup *map;
7177 map = em->map_lookup;
7179 btrfs_get_num_tolerated_disk_barrier_failures(
7181 for (i = 0; i < map->num_stripes; i++) {
7182 struct btrfs_device *dev = map->stripes[i].dev;
7184 if (!dev || !dev->bdev ||
7185 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7186 dev->last_flush_error)
7188 else if (failing_dev && failing_dev == dev)
7191 if (missing > max_tolerated) {
7194 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7195 em->start, missing, max_tolerated);
7196 free_extent_map(em);
7200 next_start = extent_map_end(em);
7201 free_extent_map(em);
7203 read_lock(&map_tree->lock);
7204 em = lookup_extent_mapping(map_tree, next_start,
7205 (u64)(-1) - next_start);
7206 read_unlock(&map_tree->lock);
7212 static void readahead_tree_node_children(struct extent_buffer *node)
7215 const int nr_items = btrfs_header_nritems(node);
7217 for (i = 0; i < nr_items; i++)
7218 btrfs_readahead_node_child(node, i);
7221 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7223 struct btrfs_root *root = fs_info->chunk_root;
7224 struct btrfs_path *path;
7225 struct extent_buffer *leaf;
7226 struct btrfs_key key;
7227 struct btrfs_key found_key;
7232 u64 last_ra_node = 0;
7234 path = btrfs_alloc_path();
7239 * uuid_mutex is needed only if we are mounting a sprout FS
7240 * otherwise we don't need it.
7242 mutex_lock(&uuid_mutex);
7245 * It is possible for mount and umount to race in such a way that
7246 * we execute this code path, but open_fs_devices failed to clear
7247 * total_rw_bytes. We certainly want it cleared before reading the
7248 * device items, so clear it here.
7250 fs_info->fs_devices->total_rw_bytes = 0;
7253 * Lockdep complains about possible circular locking dependency between
7254 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores
7255 * used for freeze procection of a fs (struct super_block.s_writers),
7256 * which we take when starting a transaction, and extent buffers of the
7257 * chunk tree if we call read_one_dev() while holding a lock on an
7258 * extent buffer of the chunk tree. Since we are mounting the filesystem
7259 * and at this point there can't be any concurrent task modifying the
7260 * chunk tree, to keep it simple, just skip locking on the chunk tree.
7262 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
7263 path->skip_locking = 1;
7266 * Read all device items, and then all the chunk items. All
7267 * device items are found before any chunk item (their object id
7268 * is smaller than the lowest possible object id for a chunk
7269 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7271 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7274 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) {
7275 struct extent_buffer *node = path->nodes[1];
7277 leaf = path->nodes[0];
7278 slot = path->slots[0];
7281 if (last_ra_node != node->start) {
7282 readahead_tree_node_children(node);
7283 last_ra_node = node->start;
7286 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7287 struct btrfs_dev_item *dev_item;
7288 dev_item = btrfs_item_ptr(leaf, slot,
7289 struct btrfs_dev_item);
7290 ret = read_one_dev(leaf, dev_item);
7294 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7295 struct btrfs_chunk *chunk;
7298 * We are only called at mount time, so no need to take
7299 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7300 * we always lock first fs_info->chunk_mutex before
7301 * acquiring any locks on the chunk tree. This is a
7302 * requirement for chunk allocation, see the comment on
7303 * top of btrfs_chunk_alloc() for details.
7305 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7306 ret = read_one_chunk(&found_key, leaf, chunk);
7311 /* Catch error found during iteration */
7318 * After loading chunk tree, we've got all device information,
7319 * do another round of validation checks.
7321 if (total_dev != fs_info->fs_devices->total_devices) {
7323 "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit",
7324 btrfs_super_num_devices(fs_info->super_copy),
7326 fs_info->fs_devices->total_devices = total_dev;
7327 btrfs_set_super_num_devices(fs_info->super_copy, total_dev);
7329 if (btrfs_super_total_bytes(fs_info->super_copy) <
7330 fs_info->fs_devices->total_rw_bytes) {
7332 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7333 btrfs_super_total_bytes(fs_info->super_copy),
7334 fs_info->fs_devices->total_rw_bytes);
7340 mutex_unlock(&uuid_mutex);
7342 btrfs_free_path(path);
7346 int btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7348 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7349 struct btrfs_device *device;
7352 fs_devices->fs_info = fs_info;
7354 mutex_lock(&fs_devices->device_list_mutex);
7355 list_for_each_entry(device, &fs_devices->devices, dev_list)
7356 device->fs_info = fs_info;
7358 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7359 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7360 device->fs_info = fs_info;
7361 ret = btrfs_get_dev_zone_info(device, false);
7366 seed_devs->fs_info = fs_info;
7368 mutex_unlock(&fs_devices->device_list_mutex);
7373 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7374 const struct btrfs_dev_stats_item *ptr,
7379 read_extent_buffer(eb, &val,
7380 offsetof(struct btrfs_dev_stats_item, values) +
7381 ((unsigned long)ptr) + (index * sizeof(u64)),
7386 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7387 struct btrfs_dev_stats_item *ptr,
7390 write_extent_buffer(eb, &val,
7391 offsetof(struct btrfs_dev_stats_item, values) +
7392 ((unsigned long)ptr) + (index * sizeof(u64)),
7396 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7397 struct btrfs_path *path)
7399 struct btrfs_dev_stats_item *ptr;
7400 struct extent_buffer *eb;
7401 struct btrfs_key key;
7405 if (!device->fs_info->dev_root)
7408 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7409 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7410 key.offset = device->devid;
7411 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7413 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7414 btrfs_dev_stat_set(device, i, 0);
7415 device->dev_stats_valid = 1;
7416 btrfs_release_path(path);
7417 return ret < 0 ? ret : 0;
7419 slot = path->slots[0];
7420 eb = path->nodes[0];
7421 item_size = btrfs_item_size(eb, slot);
7423 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7425 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7426 if (item_size >= (1 + i) * sizeof(__le64))
7427 btrfs_dev_stat_set(device, i,
7428 btrfs_dev_stats_value(eb, ptr, i));
7430 btrfs_dev_stat_set(device, i, 0);
7433 device->dev_stats_valid = 1;
7434 btrfs_dev_stat_print_on_load(device);
7435 btrfs_release_path(path);
7440 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7442 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7443 struct btrfs_device *device;
7444 struct btrfs_path *path = NULL;
7447 path = btrfs_alloc_path();
7451 mutex_lock(&fs_devices->device_list_mutex);
7452 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7453 ret = btrfs_device_init_dev_stats(device, path);
7457 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7458 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7459 ret = btrfs_device_init_dev_stats(device, path);
7465 mutex_unlock(&fs_devices->device_list_mutex);
7467 btrfs_free_path(path);
7471 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7472 struct btrfs_device *device)
7474 struct btrfs_fs_info *fs_info = trans->fs_info;
7475 struct btrfs_root *dev_root = fs_info->dev_root;
7476 struct btrfs_path *path;
7477 struct btrfs_key key;
7478 struct extent_buffer *eb;
7479 struct btrfs_dev_stats_item *ptr;
7483 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7484 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7485 key.offset = device->devid;
7487 path = btrfs_alloc_path();
7490 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7492 btrfs_warn_in_rcu(fs_info,
7493 "error %d while searching for dev_stats item for device %s",
7494 ret, btrfs_dev_name(device));
7499 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7500 /* need to delete old one and insert a new one */
7501 ret = btrfs_del_item(trans, dev_root, path);
7503 btrfs_warn_in_rcu(fs_info,
7504 "delete too small dev_stats item for device %s failed %d",
7505 btrfs_dev_name(device), ret);
7512 /* need to insert a new item */
7513 btrfs_release_path(path);
7514 ret = btrfs_insert_empty_item(trans, dev_root, path,
7515 &key, sizeof(*ptr));
7517 btrfs_warn_in_rcu(fs_info,
7518 "insert dev_stats item for device %s failed %d",
7519 btrfs_dev_name(device), ret);
7524 eb = path->nodes[0];
7525 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7526 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7527 btrfs_set_dev_stats_value(eb, ptr, i,
7528 btrfs_dev_stat_read(device, i));
7529 btrfs_mark_buffer_dirty(eb);
7532 btrfs_free_path(path);
7537 * called from commit_transaction. Writes all changed device stats to disk.
7539 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7541 struct btrfs_fs_info *fs_info = trans->fs_info;
7542 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7543 struct btrfs_device *device;
7547 mutex_lock(&fs_devices->device_list_mutex);
7548 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7549 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7550 if (!device->dev_stats_valid || stats_cnt == 0)
7555 * There is a LOAD-LOAD control dependency between the value of
7556 * dev_stats_ccnt and updating the on-disk values which requires
7557 * reading the in-memory counters. Such control dependencies
7558 * require explicit read memory barriers.
7560 * This memory barriers pairs with smp_mb__before_atomic in
7561 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7562 * barrier implied by atomic_xchg in
7563 * btrfs_dev_stats_read_and_reset
7567 ret = update_dev_stat_item(trans, device);
7569 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7571 mutex_unlock(&fs_devices->device_list_mutex);
7576 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7578 btrfs_dev_stat_inc(dev, index);
7580 if (!dev->dev_stats_valid)
7582 btrfs_err_rl_in_rcu(dev->fs_info,
7583 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7584 btrfs_dev_name(dev),
7585 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7586 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7587 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7588 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7589 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7592 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7596 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7597 if (btrfs_dev_stat_read(dev, i) != 0)
7599 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7600 return; /* all values == 0, suppress message */
7602 btrfs_info_in_rcu(dev->fs_info,
7603 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7604 btrfs_dev_name(dev),
7605 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7606 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7607 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7608 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7609 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7612 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7613 struct btrfs_ioctl_get_dev_stats *stats)
7615 BTRFS_DEV_LOOKUP_ARGS(args);
7616 struct btrfs_device *dev;
7617 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7620 mutex_lock(&fs_devices->device_list_mutex);
7621 args.devid = stats->devid;
7622 dev = btrfs_find_device(fs_info->fs_devices, &args);
7623 mutex_unlock(&fs_devices->device_list_mutex);
7626 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7628 } else if (!dev->dev_stats_valid) {
7629 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7631 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7632 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7633 if (stats->nr_items > i)
7635 btrfs_dev_stat_read_and_reset(dev, i);
7637 btrfs_dev_stat_set(dev, i, 0);
7639 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7640 current->comm, task_pid_nr(current));
7642 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7643 if (stats->nr_items > i)
7644 stats->values[i] = btrfs_dev_stat_read(dev, i);
7646 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7647 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7652 * Update the size and bytes used for each device where it changed. This is
7653 * delayed since we would otherwise get errors while writing out the
7656 * Must be invoked during transaction commit.
7658 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7660 struct btrfs_device *curr, *next;
7662 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7664 if (list_empty(&trans->dev_update_list))
7668 * We don't need the device_list_mutex here. This list is owned by the
7669 * transaction and the transaction must complete before the device is
7672 mutex_lock(&trans->fs_info->chunk_mutex);
7673 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7675 list_del_init(&curr->post_commit_list);
7676 curr->commit_total_bytes = curr->disk_total_bytes;
7677 curr->commit_bytes_used = curr->bytes_used;
7679 mutex_unlock(&trans->fs_info->chunk_mutex);
7683 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7685 int btrfs_bg_type_to_factor(u64 flags)
7687 const int index = btrfs_bg_flags_to_raid_index(flags);
7689 return btrfs_raid_array[index].ncopies;
7694 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7695 u64 chunk_offset, u64 devid,
7696 u64 physical_offset, u64 physical_len)
7698 struct btrfs_dev_lookup_args args = { .devid = devid };
7699 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7700 struct extent_map *em;
7701 struct map_lookup *map;
7702 struct btrfs_device *dev;
7708 read_lock(&em_tree->lock);
7709 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7710 read_unlock(&em_tree->lock);
7714 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7715 physical_offset, devid);
7720 map = em->map_lookup;
7721 stripe_len = btrfs_calc_stripe_length(em);
7722 if (physical_len != stripe_len) {
7724 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7725 physical_offset, devid, em->start, physical_len,
7732 * Very old mkfs.btrfs (before v4.1) will not respect the reserved
7733 * space. Although kernel can handle it without problem, better to warn
7736 if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED)
7738 "devid %llu physical %llu len %llu inside the reserved space",
7739 devid, physical_offset, physical_len);
7741 for (i = 0; i < map->num_stripes; i++) {
7742 if (map->stripes[i].dev->devid == devid &&
7743 map->stripes[i].physical == physical_offset) {
7745 if (map->verified_stripes >= map->num_stripes) {
7747 "too many dev extents for chunk %llu found",
7752 map->verified_stripes++;
7758 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7759 physical_offset, devid);
7763 /* Make sure no dev extent is beyond device boundary */
7764 dev = btrfs_find_device(fs_info->fs_devices, &args);
7766 btrfs_err(fs_info, "failed to find devid %llu", devid);
7771 if (physical_offset + physical_len > dev->disk_total_bytes) {
7773 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7774 devid, physical_offset, physical_len,
7775 dev->disk_total_bytes);
7780 if (dev->zone_info) {
7781 u64 zone_size = dev->zone_info->zone_size;
7783 if (!IS_ALIGNED(physical_offset, zone_size) ||
7784 !IS_ALIGNED(physical_len, zone_size)) {
7786 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
7787 devid, physical_offset, physical_len);
7794 free_extent_map(em);
7798 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7800 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7801 struct extent_map *em;
7802 struct rb_node *node;
7805 read_lock(&em_tree->lock);
7806 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
7807 em = rb_entry(node, struct extent_map, rb_node);
7808 if (em->map_lookup->num_stripes !=
7809 em->map_lookup->verified_stripes) {
7811 "chunk %llu has missing dev extent, have %d expect %d",
7812 em->start, em->map_lookup->verified_stripes,
7813 em->map_lookup->num_stripes);
7819 read_unlock(&em_tree->lock);
7824 * Ensure that all dev extents are mapped to correct chunk, otherwise
7825 * later chunk allocation/free would cause unexpected behavior.
7827 * NOTE: This will iterate through the whole device tree, which should be of
7828 * the same size level as the chunk tree. This slightly increases mount time.
7830 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
7832 struct btrfs_path *path;
7833 struct btrfs_root *root = fs_info->dev_root;
7834 struct btrfs_key key;
7836 u64 prev_dev_ext_end = 0;
7840 * We don't have a dev_root because we mounted with ignorebadroots and
7841 * failed to load the root, so we want to skip the verification in this
7844 * However if the dev root is fine, but the tree itself is corrupted
7845 * we'd still fail to mount. This verification is only to make sure
7846 * writes can happen safely, so instead just bypass this check
7847 * completely in the case of IGNOREBADROOTS.
7849 if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
7853 key.type = BTRFS_DEV_EXTENT_KEY;
7856 path = btrfs_alloc_path();
7860 path->reada = READA_FORWARD;
7861 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7865 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7866 ret = btrfs_next_leaf(root, path);
7869 /* No dev extents at all? Not good */
7876 struct extent_buffer *leaf = path->nodes[0];
7877 struct btrfs_dev_extent *dext;
7878 int slot = path->slots[0];
7880 u64 physical_offset;
7884 btrfs_item_key_to_cpu(leaf, &key, slot);
7885 if (key.type != BTRFS_DEV_EXTENT_KEY)
7887 devid = key.objectid;
7888 physical_offset = key.offset;
7890 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
7891 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
7892 physical_len = btrfs_dev_extent_length(leaf, dext);
7894 /* Check if this dev extent overlaps with the previous one */
7895 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
7897 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
7898 devid, physical_offset, prev_dev_ext_end);
7903 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
7904 physical_offset, physical_len);
7908 prev_dev_ext_end = physical_offset + physical_len;
7910 ret = btrfs_next_item(root, path);
7919 /* Ensure all chunks have corresponding dev extents */
7920 ret = verify_chunk_dev_extent_mapping(fs_info);
7922 btrfs_free_path(path);
7927 * Check whether the given block group or device is pinned by any inode being
7928 * used as a swapfile.
7930 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
7932 struct btrfs_swapfile_pin *sp;
7933 struct rb_node *node;
7935 spin_lock(&fs_info->swapfile_pins_lock);
7936 node = fs_info->swapfile_pins.rb_node;
7938 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
7940 node = node->rb_left;
7941 else if (ptr > sp->ptr)
7942 node = node->rb_right;
7946 spin_unlock(&fs_info->swapfile_pins_lock);
7947 return node != NULL;
7950 static int relocating_repair_kthread(void *data)
7952 struct btrfs_block_group *cache = data;
7953 struct btrfs_fs_info *fs_info = cache->fs_info;
7957 target = cache->start;
7958 btrfs_put_block_group(cache);
7960 sb_start_write(fs_info->sb);
7961 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
7963 "zoned: skip relocating block group %llu to repair: EBUSY",
7965 sb_end_write(fs_info->sb);
7969 mutex_lock(&fs_info->reclaim_bgs_lock);
7971 /* Ensure block group still exists */
7972 cache = btrfs_lookup_block_group(fs_info, target);
7976 if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags))
7979 ret = btrfs_may_alloc_data_chunk(fs_info, target);
7984 "zoned: relocating block group %llu to repair IO failure",
7986 ret = btrfs_relocate_chunk(fs_info, target);
7990 btrfs_put_block_group(cache);
7991 mutex_unlock(&fs_info->reclaim_bgs_lock);
7992 btrfs_exclop_finish(fs_info);
7993 sb_end_write(fs_info->sb);
7998 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical)
8000 struct btrfs_block_group *cache;
8002 if (!btrfs_is_zoned(fs_info))
8005 /* Do not attempt to repair in degraded state */
8006 if (btrfs_test_opt(fs_info, DEGRADED))
8009 cache = btrfs_lookup_block_group(fs_info, logical);
8013 if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) {
8014 btrfs_put_block_group(cache);
8018 kthread_run(relocating_repair_kthread, cache,
8019 "btrfs-relocating-repair");
8024 static void map_raid56_repair_block(struct btrfs_io_context *bioc,
8025 struct btrfs_io_stripe *smap,
8028 int data_stripes = nr_bioc_data_stripes(bioc);
8031 for (i = 0; i < data_stripes; i++) {
8032 u64 stripe_start = bioc->full_stripe_logical +
8033 (i << BTRFS_STRIPE_LEN_SHIFT);
8035 if (logical >= stripe_start &&
8036 logical < stripe_start + BTRFS_STRIPE_LEN)
8039 ASSERT(i < data_stripes);
8040 smap->dev = bioc->stripes[i].dev;
8041 smap->physical = bioc->stripes[i].physical +
8042 ((logical - bioc->full_stripe_logical) &
8043 BTRFS_STRIPE_LEN_MASK);
8047 * Map a repair write into a single device.
8049 * A repair write is triggered by read time repair or scrub, which would only
8050 * update the contents of a single device.
8051 * Not update any other mirrors nor go through RMW path.
8053 * Callers should ensure:
8055 * - Call btrfs_bio_counter_inc_blocked() first
8056 * - The range does not cross stripe boundary
8057 * - Has a valid @mirror_num passed in.
8059 int btrfs_map_repair_block(struct btrfs_fs_info *fs_info,
8060 struct btrfs_io_stripe *smap, u64 logical,
8061 u32 length, int mirror_num)
8063 struct btrfs_io_context *bioc = NULL;
8064 u64 map_length = length;
8065 int mirror_ret = mirror_num;
8068 ASSERT(mirror_num > 0);
8070 ret = __btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, &map_length,
8071 &bioc, smap, &mirror_ret, true);
8075 /* The map range should not cross stripe boundary. */
8076 ASSERT(map_length >= length);
8078 /* Already mapped to single stripe. */
8082 /* Map the RAID56 multi-stripe writes to a single one. */
8083 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
8084 map_raid56_repair_block(bioc, smap, logical);
8088 ASSERT(mirror_num <= bioc->num_stripes);
8089 smap->dev = bioc->stripes[mirror_num - 1].dev;
8090 smap->physical = bioc->stripes[mirror_num - 1].physical;
8092 btrfs_put_bioc(bioc);