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>
9 #include <linux/slab.h>
10 #include <linux/blkdev.h>
11 #include <linux/ratelimit.h>
12 #include <linux/kthread.h>
13 #include <linux/raid/pq.h>
14 #include <linux/semaphore.h>
15 #include <linux/uuid.h>
16 #include <linux/list_sort.h>
19 #include "extent_map.h"
21 #include "transaction.h"
22 #include "print-tree.h"
25 #include "async-thread.h"
26 #include "check-integrity.h"
27 #include "rcu-string.h"
28 #include "dev-replace.h"
30 #include "tree-checker.h"
31 #include "space-info.h"
32 #include "block-group.h"
36 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
37 [BTRFS_RAID_RAID10] = {
40 .devs_max = 0, /* 0 == as many as possible */
42 .tolerated_failures = 1,
46 .raid_name = "raid10",
47 .bg_flag = BTRFS_BLOCK_GROUP_RAID10,
48 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
50 [BTRFS_RAID_RAID1] = {
55 .tolerated_failures = 1,
60 .bg_flag = BTRFS_BLOCK_GROUP_RAID1,
61 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
63 [BTRFS_RAID_RAID1C3] = {
68 .tolerated_failures = 2,
72 .raid_name = "raid1c3",
73 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3,
74 .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET,
76 [BTRFS_RAID_RAID1C4] = {
81 .tolerated_failures = 3,
85 .raid_name = "raid1c4",
86 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4,
87 .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET,
94 .tolerated_failures = 0,
99 .bg_flag = BTRFS_BLOCK_GROUP_DUP,
102 [BTRFS_RAID_RAID0] = {
107 .tolerated_failures = 0,
111 .raid_name = "raid0",
112 .bg_flag = BTRFS_BLOCK_GROUP_RAID0,
115 [BTRFS_RAID_SINGLE] = {
120 .tolerated_failures = 0,
124 .raid_name = "single",
128 [BTRFS_RAID_RAID5] = {
133 .tolerated_failures = 1,
137 .raid_name = "raid5",
138 .bg_flag = BTRFS_BLOCK_GROUP_RAID5,
139 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
141 [BTRFS_RAID_RAID6] = {
146 .tolerated_failures = 2,
150 .raid_name = "raid6",
151 .bg_flag = BTRFS_BLOCK_GROUP_RAID6,
152 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
157 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which
158 * can be used as index to access btrfs_raid_array[].
160 enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags)
162 if (flags & BTRFS_BLOCK_GROUP_RAID10)
163 return BTRFS_RAID_RAID10;
164 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
165 return BTRFS_RAID_RAID1;
166 else if (flags & BTRFS_BLOCK_GROUP_RAID1C3)
167 return BTRFS_RAID_RAID1C3;
168 else if (flags & BTRFS_BLOCK_GROUP_RAID1C4)
169 return BTRFS_RAID_RAID1C4;
170 else if (flags & BTRFS_BLOCK_GROUP_DUP)
171 return BTRFS_RAID_DUP;
172 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
173 return BTRFS_RAID_RAID0;
174 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
175 return BTRFS_RAID_RAID5;
176 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
177 return BTRFS_RAID_RAID6;
179 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
182 const char *btrfs_bg_type_to_raid_name(u64 flags)
184 const int index = btrfs_bg_flags_to_raid_index(flags);
186 if (index >= BTRFS_NR_RAID_TYPES)
189 return btrfs_raid_array[index].raid_name;
193 * Fill @buf with textual description of @bg_flags, no more than @size_buf
194 * bytes including terminating null byte.
196 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf)
201 u64 flags = bg_flags;
202 u32 size_bp = size_buf;
209 #define DESCRIBE_FLAG(flag, desc) \
211 if (flags & (flag)) { \
212 ret = snprintf(bp, size_bp, "%s|", (desc)); \
213 if (ret < 0 || ret >= size_bp) \
221 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data");
222 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system");
223 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata");
225 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single");
226 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
227 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag,
228 btrfs_raid_array[i].raid_name);
232 ret = snprintf(bp, size_bp, "0x%llx|", flags);
236 if (size_bp < size_buf)
237 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */
240 * The text is trimmed, it's up to the caller to provide sufficiently
246 static int init_first_rw_device(struct btrfs_trans_handle *trans);
247 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
248 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
249 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
250 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
251 enum btrfs_map_op op,
252 u64 logical, u64 *length,
253 struct btrfs_bio **bbio_ret,
254 int mirror_num, int need_raid_map);
260 * There are several mutexes that protect manipulation of devices and low-level
261 * structures like chunks but not block groups, extents or files
263 * uuid_mutex (global lock)
264 * ------------------------
265 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
266 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
267 * device) or requested by the device= mount option
269 * the mutex can be very coarse and can cover long-running operations
271 * protects: updates to fs_devices counters like missing devices, rw devices,
272 * seeding, structure cloning, opening/closing devices at mount/umount time
274 * global::fs_devs - add, remove, updates to the global list
276 * does not protect: manipulation of the fs_devices::devices list in general
277 * but in mount context it could be used to exclude list modifications by eg.
280 * btrfs_device::name - renames (write side), read is RCU
282 * fs_devices::device_list_mutex (per-fs, with RCU)
283 * ------------------------------------------------
284 * protects updates to fs_devices::devices, ie. adding and deleting
286 * simple list traversal with read-only actions can be done with RCU protection
288 * may be used to exclude some operations from running concurrently without any
289 * modifications to the list (see write_all_supers)
291 * Is not required at mount and close times, because our device list is
292 * protected by the uuid_mutex at that point.
296 * protects balance structures (status, state) and context accessed from
297 * several places (internally, ioctl)
301 * protects chunks, adding or removing during allocation, trim or when a new
302 * device is added/removed. Additionally it also protects post_commit_list of
303 * individual devices, since they can be added to the transaction's
304 * post_commit_list only with chunk_mutex held.
308 * a big lock that is held by the cleaner thread and prevents running subvolume
309 * cleaning together with relocation or delayed iputs
321 * Exclusive operations
322 * ====================
324 * Maintains the exclusivity of the following operations that apply to the
325 * whole filesystem and cannot run in parallel.
330 * - Device replace (*)
333 * The device operations (as above) can be in one of the following states:
339 * Only device operations marked with (*) can go into the Paused state for the
342 * - ioctl (only Balance can be Paused through ioctl)
343 * - filesystem remounted as read-only
344 * - filesystem unmounted and mounted as read-only
345 * - system power-cycle and filesystem mounted as read-only
346 * - filesystem or device errors leading to forced read-only
348 * The status of exclusive operation is set and cleared atomically.
349 * During the course of Paused state, fs_info::exclusive_operation remains set.
350 * A device operation in Paused or Running state can be canceled or resumed
351 * either by ioctl (Balance only) or when remounted as read-write.
352 * The exclusive status is cleared when the device operation is canceled or
356 DEFINE_MUTEX(uuid_mutex);
357 static LIST_HEAD(fs_uuids);
358 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void)
364 * alloc_fs_devices - allocate struct btrfs_fs_devices
365 * @fsid: if not NULL, copy the UUID to fs_devices::fsid
366 * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid
368 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
369 * The returned struct is not linked onto any lists and can be destroyed with
370 * kfree() right away.
372 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid,
373 const u8 *metadata_fsid)
375 struct btrfs_fs_devices *fs_devs;
377 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
379 return ERR_PTR(-ENOMEM);
381 mutex_init(&fs_devs->device_list_mutex);
383 INIT_LIST_HEAD(&fs_devs->devices);
384 INIT_LIST_HEAD(&fs_devs->alloc_list);
385 INIT_LIST_HEAD(&fs_devs->fs_list);
386 INIT_LIST_HEAD(&fs_devs->seed_list);
388 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
391 memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE);
393 memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE);
398 void btrfs_free_device(struct btrfs_device *device)
400 WARN_ON(!list_empty(&device->post_commit_list));
401 rcu_string_free(device->name);
402 extent_io_tree_release(&device->alloc_state);
403 bio_put(device->flush_bio);
404 btrfs_destroy_dev_zone_info(device);
408 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
410 struct btrfs_device *device;
411 WARN_ON(fs_devices->opened);
412 while (!list_empty(&fs_devices->devices)) {
413 device = list_entry(fs_devices->devices.next,
414 struct btrfs_device, dev_list);
415 list_del(&device->dev_list);
416 btrfs_free_device(device);
421 void __exit btrfs_cleanup_fs_uuids(void)
423 struct btrfs_fs_devices *fs_devices;
425 while (!list_empty(&fs_uuids)) {
426 fs_devices = list_entry(fs_uuids.next,
427 struct btrfs_fs_devices, fs_list);
428 list_del(&fs_devices->fs_list);
429 free_fs_devices(fs_devices);
433 static noinline struct btrfs_fs_devices *find_fsid(
434 const u8 *fsid, const u8 *metadata_fsid)
436 struct btrfs_fs_devices *fs_devices;
440 /* Handle non-split brain cases */
441 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
443 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0
444 && memcmp(metadata_fsid, fs_devices->metadata_uuid,
445 BTRFS_FSID_SIZE) == 0)
448 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
455 static struct btrfs_fs_devices *find_fsid_with_metadata_uuid(
456 struct btrfs_super_block *disk_super)
459 struct btrfs_fs_devices *fs_devices;
462 * Handle scanned device having completed its fsid change but
463 * belonging to a fs_devices that was created by first scanning
464 * a device which didn't have its fsid/metadata_uuid changed
465 * at all and the CHANGING_FSID_V2 flag set.
467 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
468 if (fs_devices->fsid_change &&
469 memcmp(disk_super->metadata_uuid, fs_devices->fsid,
470 BTRFS_FSID_SIZE) == 0 &&
471 memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
472 BTRFS_FSID_SIZE) == 0) {
477 * Handle scanned device having completed its fsid change but
478 * belonging to a fs_devices that was created by a device that
479 * has an outdated pair of fsid/metadata_uuid and
480 * CHANGING_FSID_V2 flag set.
482 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
483 if (fs_devices->fsid_change &&
484 memcmp(fs_devices->metadata_uuid,
485 fs_devices->fsid, BTRFS_FSID_SIZE) != 0 &&
486 memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid,
487 BTRFS_FSID_SIZE) == 0) {
492 return find_fsid(disk_super->fsid, disk_super->metadata_uuid);
497 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
498 int flush, struct block_device **bdev,
499 struct btrfs_super_block **disk_super)
503 *bdev = blkdev_get_by_path(device_path, flags, holder);
506 ret = PTR_ERR(*bdev);
511 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
512 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
514 blkdev_put(*bdev, flags);
517 invalidate_bdev(*bdev);
518 *disk_super = btrfs_read_dev_super(*bdev);
519 if (IS_ERR(*disk_super)) {
520 ret = PTR_ERR(*disk_super);
521 blkdev_put(*bdev, flags);
532 static bool device_path_matched(const char *path, struct btrfs_device *device)
537 found = strcmp(rcu_str_deref(device->name), path);
544 * Search and remove all stale (devices which are not mounted) devices.
545 * When both inputs are NULL, it will search and release all stale devices.
546 * path: Optional. When provided will it release all unmounted devices
547 * matching this path only.
548 * skip_dev: Optional. Will skip this device when searching for the stale
550 * Return: 0 for success or if @path is NULL.
551 * -EBUSY if @path is a mounted device.
552 * -ENOENT if @path does not match any device in the list.
554 static int btrfs_free_stale_devices(const char *path,
555 struct btrfs_device *skip_device)
557 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
558 struct btrfs_device *device, *tmp_device;
564 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
566 mutex_lock(&fs_devices->device_list_mutex);
567 list_for_each_entry_safe(device, tmp_device,
568 &fs_devices->devices, dev_list) {
569 if (skip_device && skip_device == device)
571 if (path && !device->name)
573 if (path && !device_path_matched(path, device))
575 if (fs_devices->opened) {
576 /* for an already deleted device return 0 */
577 if (path && ret != 0)
582 /* delete the stale device */
583 fs_devices->num_devices--;
584 list_del(&device->dev_list);
585 btrfs_free_device(device);
589 mutex_unlock(&fs_devices->device_list_mutex);
591 if (fs_devices->num_devices == 0) {
592 btrfs_sysfs_remove_fsid(fs_devices);
593 list_del(&fs_devices->fs_list);
594 free_fs_devices(fs_devices);
602 * This is only used on mount, and we are protected from competing things
603 * messing with our fs_devices by the uuid_mutex, thus we do not need the
604 * fs_devices->device_list_mutex here.
606 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
607 struct btrfs_device *device, fmode_t flags,
610 struct request_queue *q;
611 struct block_device *bdev;
612 struct btrfs_super_block *disk_super;
621 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
626 devid = btrfs_stack_device_id(&disk_super->dev_item);
627 if (devid != device->devid)
628 goto error_free_page;
630 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
631 goto error_free_page;
633 device->generation = btrfs_super_generation(disk_super);
635 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
636 if (btrfs_super_incompat_flags(disk_super) &
637 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
639 "BTRFS: Invalid seeding and uuid-changed device detected\n");
640 goto error_free_page;
643 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
644 fs_devices->seeding = true;
646 if (bdev_read_only(bdev))
647 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
649 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
652 q = bdev_get_queue(bdev);
653 if (!blk_queue_nonrot(q))
654 fs_devices->rotating = true;
657 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
658 device->mode = flags;
660 fs_devices->open_devices++;
661 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
662 device->devid != BTRFS_DEV_REPLACE_DEVID) {
663 fs_devices->rw_devices++;
664 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
666 btrfs_release_disk_super(disk_super);
671 btrfs_release_disk_super(disk_super);
672 blkdev_put(bdev, flags);
678 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices
679 * being created with a disk that has already completed its fsid change. Such
680 * disk can belong to an fs which has its FSID changed or to one which doesn't.
681 * Handle both cases here.
683 static struct btrfs_fs_devices *find_fsid_inprogress(
684 struct btrfs_super_block *disk_super)
686 struct btrfs_fs_devices *fs_devices;
688 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
689 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
690 BTRFS_FSID_SIZE) != 0 &&
691 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
692 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) {
697 return find_fsid(disk_super->fsid, NULL);
701 static struct btrfs_fs_devices *find_fsid_changed(
702 struct btrfs_super_block *disk_super)
704 struct btrfs_fs_devices *fs_devices;
707 * Handles the case where scanned device is part of an fs that had
708 * multiple successful changes of FSID but currently device didn't
709 * observe it. Meaning our fsid will be different than theirs. We need
710 * to handle two subcases :
711 * 1 - The fs still continues to have different METADATA/FSID uuids.
712 * 2 - The fs is switched back to its original FSID (METADATA/FSID
715 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
717 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
718 BTRFS_FSID_SIZE) != 0 &&
719 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid,
720 BTRFS_FSID_SIZE) == 0 &&
721 memcmp(fs_devices->fsid, disk_super->fsid,
722 BTRFS_FSID_SIZE) != 0)
725 /* Unchanged UUIDs */
726 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
727 BTRFS_FSID_SIZE) == 0 &&
728 memcmp(fs_devices->fsid, disk_super->metadata_uuid,
729 BTRFS_FSID_SIZE) == 0)
736 static struct btrfs_fs_devices *find_fsid_reverted_metadata(
737 struct btrfs_super_block *disk_super)
739 struct btrfs_fs_devices *fs_devices;
742 * Handle the case where the scanned device is part of an fs whose last
743 * metadata UUID change reverted it to the original FSID. At the same
744 * time * fs_devices was first created by another constitutent device
745 * which didn't fully observe the operation. This results in an
746 * btrfs_fs_devices created with metadata/fsid different AND
747 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the
748 * fs_devices equal to the FSID of the disk.
750 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
751 if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
752 BTRFS_FSID_SIZE) != 0 &&
753 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
754 BTRFS_FSID_SIZE) == 0 &&
755 fs_devices->fsid_change)
762 * Add new device to list of registered devices
765 * device pointer which was just added or updated when successful
766 * error pointer when failed
768 static noinline struct btrfs_device *device_list_add(const char *path,
769 struct btrfs_super_block *disk_super,
770 bool *new_device_added)
772 struct btrfs_device *device;
773 struct btrfs_fs_devices *fs_devices = NULL;
774 struct rcu_string *name;
775 u64 found_transid = btrfs_super_generation(disk_super);
776 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
777 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
778 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
779 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) &
780 BTRFS_SUPER_FLAG_CHANGING_FSID_V2);
782 if (fsid_change_in_progress) {
783 if (!has_metadata_uuid)
784 fs_devices = find_fsid_inprogress(disk_super);
786 fs_devices = find_fsid_changed(disk_super);
787 } else if (has_metadata_uuid) {
788 fs_devices = find_fsid_with_metadata_uuid(disk_super);
790 fs_devices = find_fsid_reverted_metadata(disk_super);
792 fs_devices = find_fsid(disk_super->fsid, NULL);
797 if (has_metadata_uuid)
798 fs_devices = alloc_fs_devices(disk_super->fsid,
799 disk_super->metadata_uuid);
801 fs_devices = alloc_fs_devices(disk_super->fsid, NULL);
803 if (IS_ERR(fs_devices))
804 return ERR_CAST(fs_devices);
806 fs_devices->fsid_change = fsid_change_in_progress;
808 mutex_lock(&fs_devices->device_list_mutex);
809 list_add(&fs_devices->fs_list, &fs_uuids);
813 mutex_lock(&fs_devices->device_list_mutex);
814 device = btrfs_find_device(fs_devices, devid,
815 disk_super->dev_item.uuid, NULL);
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 if (fs_devices->opened) {
841 mutex_unlock(&fs_devices->device_list_mutex);
842 return ERR_PTR(-EBUSY);
845 device = btrfs_alloc_device(NULL, &devid,
846 disk_super->dev_item.uuid);
847 if (IS_ERR(device)) {
848 mutex_unlock(&fs_devices->device_list_mutex);
849 /* we can safely leave the fs_devices entry around */
853 name = rcu_string_strdup(path, GFP_NOFS);
855 btrfs_free_device(device);
856 mutex_unlock(&fs_devices->device_list_mutex);
857 return ERR_PTR(-ENOMEM);
859 rcu_assign_pointer(device->name, name);
861 list_add_rcu(&device->dev_list, &fs_devices->devices);
862 fs_devices->num_devices++;
864 device->fs_devices = fs_devices;
865 *new_device_added = true;
867 if (disk_super->label[0])
869 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n",
870 disk_super->label, devid, found_transid, path,
871 current->comm, task_pid_nr(current));
874 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n",
875 disk_super->fsid, devid, found_transid, path,
876 current->comm, task_pid_nr(current));
878 } else if (!device->name || strcmp(device->name->str, path)) {
880 * When FS is already mounted.
881 * 1. If you are here and if the device->name is NULL that
882 * means this device was missing at time of FS mount.
883 * 2. If you are here and if the device->name is different
884 * from 'path' that means either
885 * a. The same device disappeared and reappeared with
887 * b. The missing-disk-which-was-replaced, has
890 * We must allow 1 and 2a above. But 2b would be a spurious
893 * Further in case of 1 and 2a above, the disk at 'path'
894 * would have missed some transaction when it was away and
895 * in case of 2a the stale bdev has to be updated as well.
896 * 2b must not be allowed at all time.
900 * For now, we do allow update to btrfs_fs_device through the
901 * btrfs dev scan cli after FS has been mounted. We're still
902 * tracking a problem where systems fail mount by subvolume id
903 * when we reject replacement on a mounted FS.
905 if (!fs_devices->opened && found_transid < device->generation) {
907 * That is if the FS is _not_ mounted and if you
908 * are here, that means there is more than one
909 * disk with same uuid and devid.We keep the one
910 * with larger generation number or the last-in if
911 * generation are equal.
913 mutex_unlock(&fs_devices->device_list_mutex);
914 return ERR_PTR(-EEXIST);
918 * We are going to replace the device path for a given devid,
919 * make sure it's the same device if the device is mounted
925 error = lookup_bdev(path, &path_dev);
927 mutex_unlock(&fs_devices->device_list_mutex);
928 return ERR_PTR(error);
931 if (device->bdev->bd_dev != path_dev) {
932 mutex_unlock(&fs_devices->device_list_mutex);
934 * device->fs_info may not be reliable here, so
935 * pass in a NULL instead. This avoids a
936 * possible use-after-free when the fs_info and
937 * fs_info->sb are already torn down.
939 btrfs_warn_in_rcu(NULL,
940 "duplicate device %s devid %llu generation %llu scanned by %s (%d)",
941 path, devid, found_transid,
943 task_pid_nr(current));
944 return ERR_PTR(-EEXIST);
946 btrfs_info_in_rcu(device->fs_info,
947 "devid %llu device path %s changed to %s scanned by %s (%d)",
948 devid, rcu_str_deref(device->name),
950 task_pid_nr(current));
953 name = rcu_string_strdup(path, GFP_NOFS);
955 mutex_unlock(&fs_devices->device_list_mutex);
956 return ERR_PTR(-ENOMEM);
958 rcu_string_free(device->name);
959 rcu_assign_pointer(device->name, name);
960 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
961 fs_devices->missing_devices--;
962 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
967 * Unmount does not free the btrfs_device struct but would zero
968 * generation along with most of the other members. So just update
969 * it back. We need it to pick the disk with largest generation
972 if (!fs_devices->opened) {
973 device->generation = found_transid;
974 fs_devices->latest_generation = max_t(u64, found_transid,
975 fs_devices->latest_generation);
978 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
980 mutex_unlock(&fs_devices->device_list_mutex);
984 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
986 struct btrfs_fs_devices *fs_devices;
987 struct btrfs_device *device;
988 struct btrfs_device *orig_dev;
991 fs_devices = alloc_fs_devices(orig->fsid, NULL);
992 if (IS_ERR(fs_devices))
995 mutex_lock(&orig->device_list_mutex);
996 fs_devices->total_devices = orig->total_devices;
998 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
999 struct rcu_string *name;
1001 device = btrfs_alloc_device(NULL, &orig_dev->devid,
1003 if (IS_ERR(device)) {
1004 ret = PTR_ERR(device);
1009 * This is ok to do without rcu read locked because we hold the
1010 * uuid mutex so nothing we touch in here is going to disappear.
1012 if (orig_dev->name) {
1013 name = rcu_string_strdup(orig_dev->name->str,
1016 btrfs_free_device(device);
1020 rcu_assign_pointer(device->name, name);
1023 list_add(&device->dev_list, &fs_devices->devices);
1024 device->fs_devices = fs_devices;
1025 fs_devices->num_devices++;
1027 mutex_unlock(&orig->device_list_mutex);
1030 mutex_unlock(&orig->device_list_mutex);
1031 free_fs_devices(fs_devices);
1032 return ERR_PTR(ret);
1035 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices,
1036 struct btrfs_device **latest_dev)
1038 struct btrfs_device *device, *next;
1040 /* This is the initialized path, it is safe to release the devices. */
1041 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1042 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) {
1043 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1044 &device->dev_state) &&
1045 !test_bit(BTRFS_DEV_STATE_MISSING,
1046 &device->dev_state) &&
1048 device->generation > (*latest_dev)->generation)) {
1049 *latest_dev = device;
1055 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID,
1056 * in btrfs_init_dev_replace() so just continue.
1058 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1062 blkdev_put(device->bdev, device->mode);
1063 device->bdev = NULL;
1064 fs_devices->open_devices--;
1066 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1067 list_del_init(&device->dev_alloc_list);
1068 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1069 fs_devices->rw_devices--;
1071 list_del_init(&device->dev_list);
1072 fs_devices->num_devices--;
1073 btrfs_free_device(device);
1079 * After we have read the system tree and know devids belonging to this
1080 * filesystem, remove the device which does not belong there.
1082 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices)
1084 struct btrfs_device *latest_dev = NULL;
1085 struct btrfs_fs_devices *seed_dev;
1087 mutex_lock(&uuid_mutex);
1088 __btrfs_free_extra_devids(fs_devices, &latest_dev);
1090 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list)
1091 __btrfs_free_extra_devids(seed_dev, &latest_dev);
1093 fs_devices->latest_bdev = latest_dev->bdev;
1095 mutex_unlock(&uuid_mutex);
1098 static void btrfs_close_bdev(struct btrfs_device *device)
1103 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1104 sync_blockdev(device->bdev);
1105 invalidate_bdev(device->bdev);
1108 blkdev_put(device->bdev, device->mode);
1111 static void btrfs_close_one_device(struct btrfs_device *device)
1113 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1115 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1116 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1117 list_del_init(&device->dev_alloc_list);
1118 fs_devices->rw_devices--;
1121 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1122 fs_devices->missing_devices--;
1124 btrfs_close_bdev(device);
1126 fs_devices->open_devices--;
1127 device->bdev = NULL;
1129 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1130 btrfs_destroy_dev_zone_info(device);
1132 device->fs_info = NULL;
1133 atomic_set(&device->dev_stats_ccnt, 0);
1134 extent_io_tree_release(&device->alloc_state);
1136 /* Verify the device is back in a pristine state */
1137 ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1138 ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1139 ASSERT(list_empty(&device->dev_alloc_list));
1140 ASSERT(list_empty(&device->post_commit_list));
1141 ASSERT(atomic_read(&device->reada_in_flight) == 0);
1144 static void close_fs_devices(struct btrfs_fs_devices *fs_devices)
1146 struct btrfs_device *device, *tmp;
1148 lockdep_assert_held(&uuid_mutex);
1150 if (--fs_devices->opened > 0)
1153 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list)
1154 btrfs_close_one_device(device);
1156 WARN_ON(fs_devices->open_devices);
1157 WARN_ON(fs_devices->rw_devices);
1158 fs_devices->opened = 0;
1159 fs_devices->seeding = false;
1160 fs_devices->fs_info = NULL;
1163 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1166 struct btrfs_fs_devices *tmp;
1168 mutex_lock(&uuid_mutex);
1169 close_fs_devices(fs_devices);
1170 if (!fs_devices->opened)
1171 list_splice_init(&fs_devices->seed_list, &list);
1173 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {
1174 close_fs_devices(fs_devices);
1175 list_del(&fs_devices->seed_list);
1176 free_fs_devices(fs_devices);
1178 mutex_unlock(&uuid_mutex);
1181 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1182 fmode_t flags, void *holder)
1184 struct btrfs_device *device;
1185 struct btrfs_device *latest_dev = NULL;
1186 struct btrfs_device *tmp_device;
1188 flags |= FMODE_EXCL;
1190 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,
1194 ret = btrfs_open_one_device(fs_devices, device, flags, holder);
1196 (!latest_dev || device->generation > latest_dev->generation)) {
1197 latest_dev = device;
1198 } else if (ret == -ENODATA) {
1199 fs_devices->num_devices--;
1200 list_del(&device->dev_list);
1201 btrfs_free_device(device);
1204 if (fs_devices->open_devices == 0)
1207 fs_devices->opened = 1;
1208 fs_devices->latest_bdev = latest_dev->bdev;
1209 fs_devices->total_rw_bytes = 0;
1210 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1211 fs_devices->read_policy = BTRFS_READ_POLICY_PID;
1216 static int devid_cmp(void *priv, const struct list_head *a,
1217 const struct list_head *b)
1219 const struct btrfs_device *dev1, *dev2;
1221 dev1 = list_entry(a, struct btrfs_device, dev_list);
1222 dev2 = list_entry(b, struct btrfs_device, dev_list);
1224 if (dev1->devid < dev2->devid)
1226 else if (dev1->devid > dev2->devid)
1231 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1232 fmode_t flags, void *holder)
1236 lockdep_assert_held(&uuid_mutex);
1238 * The device_list_mutex cannot be taken here in case opening the
1239 * underlying device takes further locks like open_mutex.
1241 * We also don't need the lock here as this is called during mount and
1242 * exclusion is provided by uuid_mutex
1245 if (fs_devices->opened) {
1246 fs_devices->opened++;
1249 list_sort(NULL, &fs_devices->devices, devid_cmp);
1250 ret = open_fs_devices(fs_devices, flags, holder);
1256 void btrfs_release_disk_super(struct btrfs_super_block *super)
1258 struct page *page = virt_to_page(super);
1263 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1264 u64 bytenr, u64 bytenr_orig)
1266 struct btrfs_super_block *disk_super;
1271 /* make sure our super fits in the device */
1272 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1273 return ERR_PTR(-EINVAL);
1275 /* make sure our super fits in the page */
1276 if (sizeof(*disk_super) > PAGE_SIZE)
1277 return ERR_PTR(-EINVAL);
1279 /* make sure our super doesn't straddle pages on disk */
1280 index = bytenr >> PAGE_SHIFT;
1281 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1282 return ERR_PTR(-EINVAL);
1284 /* pull in the page with our super */
1285 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1288 return ERR_CAST(page);
1290 p = page_address(page);
1292 /* align our pointer to the offset of the super block */
1293 disk_super = p + offset_in_page(bytenr);
1295 if (btrfs_super_bytenr(disk_super) != bytenr_orig ||
1296 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1297 btrfs_release_disk_super(p);
1298 return ERR_PTR(-EINVAL);
1301 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1302 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1307 int btrfs_forget_devices(const char *path)
1311 mutex_lock(&uuid_mutex);
1312 ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL);
1313 mutex_unlock(&uuid_mutex);
1319 * Look for a btrfs signature on a device. This may be called out of the mount path
1320 * and we are not allowed to call set_blocksize during the scan. The superblock
1321 * is read via pagecache
1323 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1326 struct btrfs_super_block *disk_super;
1327 bool new_device_added = false;
1328 struct btrfs_device *device = NULL;
1329 struct block_device *bdev;
1330 u64 bytenr, bytenr_orig;
1333 lockdep_assert_held(&uuid_mutex);
1336 * we would like to check all the supers, but that would make
1337 * a btrfs mount succeed after a mkfs from a different FS.
1338 * So, we need to add a special mount option to scan for
1339 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1341 flags |= FMODE_EXCL;
1343 bdev = blkdev_get_by_path(path, flags, holder);
1345 return ERR_CAST(bdev);
1347 bytenr_orig = btrfs_sb_offset(0);
1348 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr);
1350 return ERR_PTR(ret);
1352 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig);
1353 if (IS_ERR(disk_super)) {
1354 device = ERR_CAST(disk_super);
1355 goto error_bdev_put;
1358 device = device_list_add(path, disk_super, &new_device_added);
1359 if (!IS_ERR(device)) {
1360 if (new_device_added)
1361 btrfs_free_stale_devices(path, device);
1364 btrfs_release_disk_super(disk_super);
1367 blkdev_put(bdev, flags);
1373 * Try to find a chunk that intersects [start, start + len] range and when one
1374 * such is found, record the end of it in *start
1376 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1379 u64 physical_start, physical_end;
1381 lockdep_assert_held(&device->fs_info->chunk_mutex);
1383 if (!find_first_extent_bit(&device->alloc_state, *start,
1384 &physical_start, &physical_end,
1385 CHUNK_ALLOCATED, NULL)) {
1387 if (in_range(physical_start, *start, len) ||
1388 in_range(*start, physical_start,
1389 physical_end - physical_start)) {
1390 *start = physical_end + 1;
1397 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start)
1399 switch (device->fs_devices->chunk_alloc_policy) {
1400 case BTRFS_CHUNK_ALLOC_REGULAR:
1402 * We don't want to overwrite the superblock on the drive nor
1403 * any area used by the boot loader (grub for example), so we
1404 * make sure to start at an offset of at least 1MB.
1406 return max_t(u64, start, SZ_1M);
1407 case BTRFS_CHUNK_ALLOC_ZONED:
1409 * We don't care about the starting region like regular
1410 * allocator, because we anyway use/reserve the first two zones
1411 * for superblock logging.
1413 return ALIGN(start, device->zone_info->zone_size);
1419 static bool dev_extent_hole_check_zoned(struct btrfs_device *device,
1420 u64 *hole_start, u64 *hole_size,
1423 u64 zone_size = device->zone_info->zone_size;
1426 bool changed = false;
1428 ASSERT(IS_ALIGNED(*hole_start, zone_size));
1430 while (*hole_size > 0) {
1431 pos = btrfs_find_allocatable_zones(device, *hole_start,
1432 *hole_start + *hole_size,
1434 if (pos != *hole_start) {
1435 *hole_size = *hole_start + *hole_size - pos;
1438 if (*hole_size < num_bytes)
1442 ret = btrfs_ensure_empty_zones(device, pos, num_bytes);
1444 /* Range is ensured to be empty */
1448 /* Given hole range was invalid (outside of device) */
1449 if (ret == -ERANGE) {
1450 *hole_start += *hole_size;
1455 *hole_start += zone_size;
1456 *hole_size -= zone_size;
1464 * dev_extent_hole_check - check if specified hole is suitable for allocation
1465 * @device: the device which we have the hole
1466 * @hole_start: starting position of the hole
1467 * @hole_size: the size of the hole
1468 * @num_bytes: the size of the free space that we need
1470 * This function may modify @hole_start and @hole_size to reflect the suitable
1471 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1473 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1474 u64 *hole_size, u64 num_bytes)
1476 bool changed = false;
1477 u64 hole_end = *hole_start + *hole_size;
1481 * Check before we set max_hole_start, otherwise we could end up
1482 * sending back this offset anyway.
1484 if (contains_pending_extent(device, hole_start, *hole_size)) {
1485 if (hole_end >= *hole_start)
1486 *hole_size = hole_end - *hole_start;
1492 switch (device->fs_devices->chunk_alloc_policy) {
1493 case BTRFS_CHUNK_ALLOC_REGULAR:
1494 /* No extra check */
1496 case BTRFS_CHUNK_ALLOC_ZONED:
1497 if (dev_extent_hole_check_zoned(device, hole_start,
1498 hole_size, num_bytes)) {
1501 * The changed hole can contain pending extent.
1502 * Loop again to check that.
1518 * find_free_dev_extent_start - find free space in the specified device
1519 * @device: the device which we search the free space in
1520 * @num_bytes: the size of the free space that we need
1521 * @search_start: the position from which to begin the search
1522 * @start: store the start of the free space.
1523 * @len: the size of the free space. that we find, or the size
1524 * of the max free space if we don't find suitable free space
1526 * this uses a pretty simple search, the expectation is that it is
1527 * called very infrequently and that a given device has a small number
1530 * @start is used to store the start of the free space if we find. But if we
1531 * don't find suitable free space, it will be used to store the start position
1532 * of the max free space.
1534 * @len is used to store the size of the free space that we find.
1535 * But if we don't find suitable free space, it is used to store the size of
1536 * the max free space.
1538 * NOTE: This function will search *commit* root of device tree, and does extra
1539 * check to ensure dev extents are not double allocated.
1540 * This makes the function safe to allocate dev extents but may not report
1541 * correct usable device space, as device extent freed in current transaction
1542 * is not reported as available.
1544 static int find_free_dev_extent_start(struct btrfs_device *device,
1545 u64 num_bytes, u64 search_start, u64 *start,
1548 struct btrfs_fs_info *fs_info = device->fs_info;
1549 struct btrfs_root *root = fs_info->dev_root;
1550 struct btrfs_key key;
1551 struct btrfs_dev_extent *dev_extent;
1552 struct btrfs_path *path;
1557 u64 search_end = device->total_bytes;
1560 struct extent_buffer *l;
1562 search_start = dev_extent_search_start(device, search_start);
1564 WARN_ON(device->zone_info &&
1565 !IS_ALIGNED(num_bytes, device->zone_info->zone_size));
1567 path = btrfs_alloc_path();
1571 max_hole_start = search_start;
1575 if (search_start >= search_end ||
1576 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1581 path->reada = READA_FORWARD;
1582 path->search_commit_root = 1;
1583 path->skip_locking = 1;
1585 key.objectid = device->devid;
1586 key.offset = search_start;
1587 key.type = BTRFS_DEV_EXTENT_KEY;
1589 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1593 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1600 slot = path->slots[0];
1601 if (slot >= btrfs_header_nritems(l)) {
1602 ret = btrfs_next_leaf(root, path);
1610 btrfs_item_key_to_cpu(l, &key, slot);
1612 if (key.objectid < device->devid)
1615 if (key.objectid > device->devid)
1618 if (key.type != BTRFS_DEV_EXTENT_KEY)
1621 if (key.offset > search_start) {
1622 hole_size = key.offset - search_start;
1623 dev_extent_hole_check(device, &search_start, &hole_size,
1626 if (hole_size > max_hole_size) {
1627 max_hole_start = search_start;
1628 max_hole_size = hole_size;
1632 * If this free space is greater than which we need,
1633 * it must be the max free space that we have found
1634 * until now, so max_hole_start must point to the start
1635 * of this free space and the length of this free space
1636 * is stored in max_hole_size. Thus, we return
1637 * max_hole_start and max_hole_size and go back to the
1640 if (hole_size >= num_bytes) {
1646 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1647 extent_end = key.offset + btrfs_dev_extent_length(l,
1649 if (extent_end > search_start)
1650 search_start = extent_end;
1657 * At this point, search_start should be the end of
1658 * allocated dev extents, and when shrinking the device,
1659 * search_end may be smaller than search_start.
1661 if (search_end > search_start) {
1662 hole_size = search_end - search_start;
1663 if (dev_extent_hole_check(device, &search_start, &hole_size,
1665 btrfs_release_path(path);
1669 if (hole_size > max_hole_size) {
1670 max_hole_start = search_start;
1671 max_hole_size = hole_size;
1676 if (max_hole_size < num_bytes)
1682 btrfs_free_path(path);
1683 *start = max_hole_start;
1685 *len = max_hole_size;
1689 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1690 u64 *start, u64 *len)
1692 /* FIXME use last free of some kind */
1693 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1696 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1697 struct btrfs_device *device,
1698 u64 start, u64 *dev_extent_len)
1700 struct btrfs_fs_info *fs_info = device->fs_info;
1701 struct btrfs_root *root = fs_info->dev_root;
1703 struct btrfs_path *path;
1704 struct btrfs_key key;
1705 struct btrfs_key found_key;
1706 struct extent_buffer *leaf = NULL;
1707 struct btrfs_dev_extent *extent = NULL;
1709 path = btrfs_alloc_path();
1713 key.objectid = device->devid;
1715 key.type = BTRFS_DEV_EXTENT_KEY;
1717 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1719 ret = btrfs_previous_item(root, path, key.objectid,
1720 BTRFS_DEV_EXTENT_KEY);
1723 leaf = path->nodes[0];
1724 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1725 extent = btrfs_item_ptr(leaf, path->slots[0],
1726 struct btrfs_dev_extent);
1727 BUG_ON(found_key.offset > start || found_key.offset +
1728 btrfs_dev_extent_length(leaf, extent) < start);
1730 btrfs_release_path(path);
1732 } else if (ret == 0) {
1733 leaf = path->nodes[0];
1734 extent = btrfs_item_ptr(leaf, path->slots[0],
1735 struct btrfs_dev_extent);
1740 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1742 ret = btrfs_del_item(trans, root, path);
1744 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1746 btrfs_free_path(path);
1750 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1752 struct extent_map_tree *em_tree;
1753 struct extent_map *em;
1757 em_tree = &fs_info->mapping_tree;
1758 read_lock(&em_tree->lock);
1759 n = rb_last(&em_tree->map.rb_root);
1761 em = rb_entry(n, struct extent_map, rb_node);
1762 ret = em->start + em->len;
1764 read_unlock(&em_tree->lock);
1769 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1773 struct btrfs_key key;
1774 struct btrfs_key found_key;
1775 struct btrfs_path *path;
1777 path = btrfs_alloc_path();
1781 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1782 key.type = BTRFS_DEV_ITEM_KEY;
1783 key.offset = (u64)-1;
1785 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1791 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1796 ret = btrfs_previous_item(fs_info->chunk_root, path,
1797 BTRFS_DEV_ITEMS_OBJECTID,
1798 BTRFS_DEV_ITEM_KEY);
1802 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1804 *devid_ret = found_key.offset + 1;
1808 btrfs_free_path(path);
1813 * the device information is stored in the chunk root
1814 * the btrfs_device struct should be fully filled in
1816 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1817 struct btrfs_device *device)
1820 struct btrfs_path *path;
1821 struct btrfs_dev_item *dev_item;
1822 struct extent_buffer *leaf;
1823 struct btrfs_key key;
1826 path = btrfs_alloc_path();
1830 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1831 key.type = BTRFS_DEV_ITEM_KEY;
1832 key.offset = device->devid;
1834 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1835 &key, sizeof(*dev_item));
1839 leaf = path->nodes[0];
1840 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1842 btrfs_set_device_id(leaf, dev_item, device->devid);
1843 btrfs_set_device_generation(leaf, dev_item, 0);
1844 btrfs_set_device_type(leaf, dev_item, device->type);
1845 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1846 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1847 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1848 btrfs_set_device_total_bytes(leaf, dev_item,
1849 btrfs_device_get_disk_total_bytes(device));
1850 btrfs_set_device_bytes_used(leaf, dev_item,
1851 btrfs_device_get_bytes_used(device));
1852 btrfs_set_device_group(leaf, dev_item, 0);
1853 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1854 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1855 btrfs_set_device_start_offset(leaf, dev_item, 0);
1857 ptr = btrfs_device_uuid(dev_item);
1858 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1859 ptr = btrfs_device_fsid(dev_item);
1860 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1861 ptr, BTRFS_FSID_SIZE);
1862 btrfs_mark_buffer_dirty(leaf);
1866 btrfs_free_path(path);
1871 * Function to update ctime/mtime for a given device path.
1872 * Mainly used for ctime/mtime based probe like libblkid.
1874 static void update_dev_time(const char *path_name)
1878 filp = filp_open(path_name, O_RDWR, 0);
1881 file_update_time(filp);
1882 filp_close(filp, NULL);
1885 static int btrfs_rm_dev_item(struct btrfs_device *device)
1887 struct btrfs_root *root = device->fs_info->chunk_root;
1889 struct btrfs_path *path;
1890 struct btrfs_key key;
1891 struct btrfs_trans_handle *trans;
1893 path = btrfs_alloc_path();
1897 trans = btrfs_start_transaction(root, 0);
1898 if (IS_ERR(trans)) {
1899 btrfs_free_path(path);
1900 return PTR_ERR(trans);
1902 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1903 key.type = BTRFS_DEV_ITEM_KEY;
1904 key.offset = device->devid;
1906 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1910 btrfs_abort_transaction(trans, ret);
1911 btrfs_end_transaction(trans);
1915 ret = btrfs_del_item(trans, root, path);
1917 btrfs_abort_transaction(trans, ret);
1918 btrfs_end_transaction(trans);
1922 btrfs_free_path(path);
1924 ret = btrfs_commit_transaction(trans);
1929 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1930 * filesystem. It's up to the caller to adjust that number regarding eg. device
1933 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1941 seq = read_seqbegin(&fs_info->profiles_lock);
1943 all_avail = fs_info->avail_data_alloc_bits |
1944 fs_info->avail_system_alloc_bits |
1945 fs_info->avail_metadata_alloc_bits;
1946 } while (read_seqretry(&fs_info->profiles_lock, seq));
1948 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1949 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1952 if (num_devices < btrfs_raid_array[i].devs_min) {
1953 int ret = btrfs_raid_array[i].mindev_error;
1963 static struct btrfs_device * btrfs_find_next_active_device(
1964 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1966 struct btrfs_device *next_device;
1968 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1969 if (next_device != device &&
1970 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1971 && next_device->bdev)
1979 * Helper function to check if the given device is part of s_bdev / latest_bdev
1980 * and replace it with the provided or the next active device, in the context
1981 * where this function called, there should be always be another device (or
1982 * this_dev) which is active.
1984 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
1985 struct btrfs_device *next_device)
1987 struct btrfs_fs_info *fs_info = device->fs_info;
1990 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1992 ASSERT(next_device);
1994 if (fs_info->sb->s_bdev &&
1995 (fs_info->sb->s_bdev == device->bdev))
1996 fs_info->sb->s_bdev = next_device->bdev;
1998 if (fs_info->fs_devices->latest_bdev == device->bdev)
1999 fs_info->fs_devices->latest_bdev = next_device->bdev;
2003 * Return btrfs_fs_devices::num_devices excluding the device that's being
2004 * currently replaced.
2006 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2008 u64 num_devices = fs_info->fs_devices->num_devices;
2010 down_read(&fs_info->dev_replace.rwsem);
2011 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2012 ASSERT(num_devices > 1);
2015 up_read(&fs_info->dev_replace.rwsem);
2020 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
2021 struct block_device *bdev,
2022 const char *device_path)
2024 struct btrfs_super_block *disk_super;
2030 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2034 disk_super = btrfs_read_dev_one_super(bdev, copy_num);
2035 if (IS_ERR(disk_super))
2038 if (bdev_is_zoned(bdev)) {
2039 btrfs_reset_sb_log_zones(bdev, copy_num);
2043 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
2045 page = virt_to_page(disk_super);
2046 set_page_dirty(page);
2048 /* write_on_page() unlocks the page */
2049 ret = write_one_page(page);
2052 "error clearing superblock number %d (%d)",
2054 btrfs_release_disk_super(disk_super);
2058 /* Notify udev that device has changed */
2059 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2061 /* Update ctime/mtime for device path for libblkid */
2062 update_dev_time(device_path);
2065 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
2068 struct btrfs_device *device;
2069 struct btrfs_fs_devices *cur_devices;
2070 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2074 mutex_lock(&uuid_mutex);
2076 num_devices = btrfs_num_devices(fs_info);
2078 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2082 device = btrfs_find_device_by_devspec(fs_info, devid, device_path);
2084 if (IS_ERR(device)) {
2085 if (PTR_ERR(device) == -ENOENT &&
2086 strcmp(device_path, "missing") == 0)
2087 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2089 ret = PTR_ERR(device);
2093 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2094 btrfs_warn_in_rcu(fs_info,
2095 "cannot remove device %s (devid %llu) due to active swapfile",
2096 rcu_str_deref(device->name), device->devid);
2101 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2102 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
2106 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2107 fs_info->fs_devices->rw_devices == 1) {
2108 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
2112 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2113 mutex_lock(&fs_info->chunk_mutex);
2114 list_del_init(&device->dev_alloc_list);
2115 device->fs_devices->rw_devices--;
2116 mutex_unlock(&fs_info->chunk_mutex);
2119 mutex_unlock(&uuid_mutex);
2120 ret = btrfs_shrink_device(device, 0);
2122 btrfs_reada_remove_dev(device);
2123 mutex_lock(&uuid_mutex);
2128 * TODO: the superblock still includes this device in its num_devices
2129 * counter although write_all_supers() is not locked out. This
2130 * could give a filesystem state which requires a degraded mount.
2132 ret = btrfs_rm_dev_item(device);
2136 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2137 btrfs_scrub_cancel_dev(device);
2140 * the device list mutex makes sure that we don't change
2141 * the device list while someone else is writing out all
2142 * the device supers. Whoever is writing all supers, should
2143 * lock the device list mutex before getting the number of
2144 * devices in the super block (super_copy). Conversely,
2145 * whoever updates the number of devices in the super block
2146 * (super_copy) should hold the device list mutex.
2150 * In normal cases the cur_devices == fs_devices. But in case
2151 * of deleting a seed device, the cur_devices should point to
2152 * its own fs_devices listed under the fs_devices->seed.
2154 cur_devices = device->fs_devices;
2155 mutex_lock(&fs_devices->device_list_mutex);
2156 list_del_rcu(&device->dev_list);
2158 cur_devices->num_devices--;
2159 cur_devices->total_devices--;
2160 /* Update total_devices of the parent fs_devices if it's seed */
2161 if (cur_devices != fs_devices)
2162 fs_devices->total_devices--;
2164 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2165 cur_devices->missing_devices--;
2167 btrfs_assign_next_active_device(device, NULL);
2170 cur_devices->open_devices--;
2171 /* remove sysfs entry */
2172 btrfs_sysfs_remove_device(device);
2175 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2176 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2177 mutex_unlock(&fs_devices->device_list_mutex);
2180 * at this point, the device is zero sized and detached from
2181 * the devices list. All that's left is to zero out the old
2182 * supers and free the device.
2184 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2185 btrfs_scratch_superblocks(fs_info, device->bdev,
2188 btrfs_close_bdev(device);
2190 btrfs_free_device(device);
2192 if (cur_devices->open_devices == 0) {
2193 list_del_init(&cur_devices->seed_list);
2194 close_fs_devices(cur_devices);
2195 free_fs_devices(cur_devices);
2199 mutex_unlock(&uuid_mutex);
2203 btrfs_reada_undo_remove_dev(device);
2204 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2205 mutex_lock(&fs_info->chunk_mutex);
2206 list_add(&device->dev_alloc_list,
2207 &fs_devices->alloc_list);
2208 device->fs_devices->rw_devices++;
2209 mutex_unlock(&fs_info->chunk_mutex);
2214 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2216 struct btrfs_fs_devices *fs_devices;
2218 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2221 * in case of fs with no seed, srcdev->fs_devices will point
2222 * to fs_devices of fs_info. However when the dev being replaced is
2223 * a seed dev it will point to the seed's local fs_devices. In short
2224 * srcdev will have its correct fs_devices in both the cases.
2226 fs_devices = srcdev->fs_devices;
2228 list_del_rcu(&srcdev->dev_list);
2229 list_del(&srcdev->dev_alloc_list);
2230 fs_devices->num_devices--;
2231 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2232 fs_devices->missing_devices--;
2234 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2235 fs_devices->rw_devices--;
2238 fs_devices->open_devices--;
2241 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2243 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2245 mutex_lock(&uuid_mutex);
2247 btrfs_close_bdev(srcdev);
2249 btrfs_free_device(srcdev);
2251 /* if this is no devs we rather delete the fs_devices */
2252 if (!fs_devices->num_devices) {
2254 * On a mounted FS, num_devices can't be zero unless it's a
2255 * seed. In case of a seed device being replaced, the replace
2256 * target added to the sprout FS, so there will be no more
2257 * device left under the seed FS.
2259 ASSERT(fs_devices->seeding);
2261 list_del_init(&fs_devices->seed_list);
2262 close_fs_devices(fs_devices);
2263 free_fs_devices(fs_devices);
2265 mutex_unlock(&uuid_mutex);
2268 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2270 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2272 mutex_lock(&fs_devices->device_list_mutex);
2274 btrfs_sysfs_remove_device(tgtdev);
2277 fs_devices->open_devices--;
2279 fs_devices->num_devices--;
2281 btrfs_assign_next_active_device(tgtdev, NULL);
2283 list_del_rcu(&tgtdev->dev_list);
2285 mutex_unlock(&fs_devices->device_list_mutex);
2288 * The update_dev_time() with in btrfs_scratch_superblocks()
2289 * may lead to a call to btrfs_show_devname() which will try
2290 * to hold device_list_mutex. And here this device
2291 * is already out of device list, so we don't have to hold
2292 * the device_list_mutex lock.
2294 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2297 btrfs_close_bdev(tgtdev);
2299 btrfs_free_device(tgtdev);
2302 static struct btrfs_device *btrfs_find_device_by_path(
2303 struct btrfs_fs_info *fs_info, const char *device_path)
2306 struct btrfs_super_block *disk_super;
2309 struct block_device *bdev;
2310 struct btrfs_device *device;
2312 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2313 fs_info->bdev_holder, 0, &bdev, &disk_super);
2315 return ERR_PTR(ret);
2317 devid = btrfs_stack_device_id(&disk_super->dev_item);
2318 dev_uuid = disk_super->dev_item.uuid;
2319 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2320 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2321 disk_super->metadata_uuid);
2323 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2326 btrfs_release_disk_super(disk_super);
2328 device = ERR_PTR(-ENOENT);
2329 blkdev_put(bdev, FMODE_READ);
2334 * Lookup a device given by device id, or the path if the id is 0.
2336 struct btrfs_device *btrfs_find_device_by_devspec(
2337 struct btrfs_fs_info *fs_info, u64 devid,
2338 const char *device_path)
2340 struct btrfs_device *device;
2343 device = btrfs_find_device(fs_info->fs_devices, devid, NULL,
2346 return ERR_PTR(-ENOENT);
2350 if (!device_path || !device_path[0])
2351 return ERR_PTR(-EINVAL);
2353 if (strcmp(device_path, "missing") == 0) {
2354 /* Find first missing device */
2355 list_for_each_entry(device, &fs_info->fs_devices->devices,
2357 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2358 &device->dev_state) && !device->bdev)
2361 return ERR_PTR(-ENOENT);
2364 return btrfs_find_device_by_path(fs_info, device_path);
2368 * does all the dirty work required for changing file system's UUID.
2370 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2372 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2373 struct btrfs_fs_devices *old_devices;
2374 struct btrfs_fs_devices *seed_devices;
2375 struct btrfs_super_block *disk_super = fs_info->super_copy;
2376 struct btrfs_device *device;
2379 lockdep_assert_held(&uuid_mutex);
2380 if (!fs_devices->seeding)
2384 * Private copy of the seed devices, anchored at
2385 * fs_info->fs_devices->seed_list
2387 seed_devices = alloc_fs_devices(NULL, NULL);
2388 if (IS_ERR(seed_devices))
2389 return PTR_ERR(seed_devices);
2392 * It's necessary to retain a copy of the original seed fs_devices in
2393 * fs_uuids so that filesystems which have been seeded can successfully
2394 * reference the seed device from open_seed_devices. This also supports
2397 old_devices = clone_fs_devices(fs_devices);
2398 if (IS_ERR(old_devices)) {
2399 kfree(seed_devices);
2400 return PTR_ERR(old_devices);
2403 list_add(&old_devices->fs_list, &fs_uuids);
2405 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2406 seed_devices->opened = 1;
2407 INIT_LIST_HEAD(&seed_devices->devices);
2408 INIT_LIST_HEAD(&seed_devices->alloc_list);
2409 mutex_init(&seed_devices->device_list_mutex);
2411 mutex_lock(&fs_devices->device_list_mutex);
2412 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2414 list_for_each_entry(device, &seed_devices->devices, dev_list)
2415 device->fs_devices = seed_devices;
2417 fs_devices->seeding = false;
2418 fs_devices->num_devices = 0;
2419 fs_devices->open_devices = 0;
2420 fs_devices->missing_devices = 0;
2421 fs_devices->rotating = false;
2422 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2424 generate_random_uuid(fs_devices->fsid);
2425 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2426 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2427 mutex_unlock(&fs_devices->device_list_mutex);
2429 super_flags = btrfs_super_flags(disk_super) &
2430 ~BTRFS_SUPER_FLAG_SEEDING;
2431 btrfs_set_super_flags(disk_super, super_flags);
2437 * Store the expected generation for seed devices in device items.
2439 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2441 struct btrfs_fs_info *fs_info = trans->fs_info;
2442 struct btrfs_root *root = fs_info->chunk_root;
2443 struct btrfs_path *path;
2444 struct extent_buffer *leaf;
2445 struct btrfs_dev_item *dev_item;
2446 struct btrfs_device *device;
2447 struct btrfs_key key;
2448 u8 fs_uuid[BTRFS_FSID_SIZE];
2449 u8 dev_uuid[BTRFS_UUID_SIZE];
2453 path = btrfs_alloc_path();
2457 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2459 key.type = BTRFS_DEV_ITEM_KEY;
2462 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2466 leaf = path->nodes[0];
2468 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2469 ret = btrfs_next_leaf(root, path);
2474 leaf = path->nodes[0];
2475 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2476 btrfs_release_path(path);
2480 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2481 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2482 key.type != BTRFS_DEV_ITEM_KEY)
2485 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2486 struct btrfs_dev_item);
2487 devid = btrfs_device_id(leaf, dev_item);
2488 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2490 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2492 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2494 BUG_ON(!device); /* Logic error */
2496 if (device->fs_devices->seeding) {
2497 btrfs_set_device_generation(leaf, dev_item,
2498 device->generation);
2499 btrfs_mark_buffer_dirty(leaf);
2507 btrfs_free_path(path);
2511 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2513 struct btrfs_root *root = fs_info->dev_root;
2514 struct request_queue *q;
2515 struct btrfs_trans_handle *trans;
2516 struct btrfs_device *device;
2517 struct block_device *bdev;
2518 struct super_block *sb = fs_info->sb;
2519 struct rcu_string *name;
2520 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2521 u64 orig_super_total_bytes;
2522 u64 orig_super_num_devices;
2523 int seeding_dev = 0;
2525 bool locked = false;
2527 if (sb_rdonly(sb) && !fs_devices->seeding)
2530 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2531 fs_info->bdev_holder);
2533 return PTR_ERR(bdev);
2535 if (!btrfs_check_device_zone_type(fs_info, bdev)) {
2540 if (fs_devices->seeding) {
2542 down_write(&sb->s_umount);
2543 mutex_lock(&uuid_mutex);
2547 sync_blockdev(bdev);
2550 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2551 if (device->bdev == bdev) {
2559 device = btrfs_alloc_device(fs_info, NULL, NULL);
2560 if (IS_ERR(device)) {
2561 /* we can safely leave the fs_devices entry around */
2562 ret = PTR_ERR(device);
2566 name = rcu_string_strdup(device_path, GFP_KERNEL);
2569 goto error_free_device;
2571 rcu_assign_pointer(device->name, name);
2573 device->fs_info = fs_info;
2574 device->bdev = bdev;
2576 ret = btrfs_get_dev_zone_info(device);
2578 goto error_free_device;
2580 trans = btrfs_start_transaction(root, 0);
2581 if (IS_ERR(trans)) {
2582 ret = PTR_ERR(trans);
2583 goto error_free_zone;
2586 q = bdev_get_queue(bdev);
2587 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2588 device->generation = trans->transid;
2589 device->io_width = fs_info->sectorsize;
2590 device->io_align = fs_info->sectorsize;
2591 device->sector_size = fs_info->sectorsize;
2592 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2593 fs_info->sectorsize);
2594 device->disk_total_bytes = device->total_bytes;
2595 device->commit_total_bytes = device->total_bytes;
2596 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2597 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2598 device->mode = FMODE_EXCL;
2599 device->dev_stats_valid = 1;
2600 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2603 btrfs_clear_sb_rdonly(sb);
2604 ret = btrfs_prepare_sprout(fs_info);
2606 btrfs_abort_transaction(trans, ret);
2611 device->fs_devices = fs_devices;
2613 mutex_lock(&fs_devices->device_list_mutex);
2614 mutex_lock(&fs_info->chunk_mutex);
2615 list_add_rcu(&device->dev_list, &fs_devices->devices);
2616 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2617 fs_devices->num_devices++;
2618 fs_devices->open_devices++;
2619 fs_devices->rw_devices++;
2620 fs_devices->total_devices++;
2621 fs_devices->total_rw_bytes += device->total_bytes;
2623 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2625 if (!blk_queue_nonrot(q))
2626 fs_devices->rotating = true;
2628 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2629 btrfs_set_super_total_bytes(fs_info->super_copy,
2630 round_down(orig_super_total_bytes + device->total_bytes,
2631 fs_info->sectorsize));
2633 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2634 btrfs_set_super_num_devices(fs_info->super_copy,
2635 orig_super_num_devices + 1);
2638 * we've got more storage, clear any full flags on the space
2641 btrfs_clear_space_info_full(fs_info);
2643 mutex_unlock(&fs_info->chunk_mutex);
2645 /* Add sysfs device entry */
2646 btrfs_sysfs_add_device(device);
2648 mutex_unlock(&fs_devices->device_list_mutex);
2651 mutex_lock(&fs_info->chunk_mutex);
2652 ret = init_first_rw_device(trans);
2653 mutex_unlock(&fs_info->chunk_mutex);
2655 btrfs_abort_transaction(trans, ret);
2660 ret = btrfs_add_dev_item(trans, device);
2662 btrfs_abort_transaction(trans, ret);
2667 ret = btrfs_finish_sprout(trans);
2669 btrfs_abort_transaction(trans, ret);
2674 * fs_devices now represents the newly sprouted filesystem and
2675 * its fsid has been changed by btrfs_prepare_sprout
2677 btrfs_sysfs_update_sprout_fsid(fs_devices);
2680 ret = btrfs_commit_transaction(trans);
2683 mutex_unlock(&uuid_mutex);
2684 up_write(&sb->s_umount);
2687 if (ret) /* transaction commit */
2690 ret = btrfs_relocate_sys_chunks(fs_info);
2692 btrfs_handle_fs_error(fs_info, ret,
2693 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2694 trans = btrfs_attach_transaction(root);
2695 if (IS_ERR(trans)) {
2696 if (PTR_ERR(trans) == -ENOENT)
2698 ret = PTR_ERR(trans);
2702 ret = btrfs_commit_transaction(trans);
2706 * Now that we have written a new super block to this device, check all
2707 * other fs_devices list if device_path alienates any other scanned
2709 * We can ignore the return value as it typically returns -EINVAL and
2710 * only succeeds if the device was an alien.
2712 btrfs_forget_devices(device_path);
2714 /* Update ctime/mtime for blkid or udev */
2715 update_dev_time(device_path);
2720 btrfs_sysfs_remove_device(device);
2721 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2722 mutex_lock(&fs_info->chunk_mutex);
2723 list_del_rcu(&device->dev_list);
2724 list_del(&device->dev_alloc_list);
2725 fs_info->fs_devices->num_devices--;
2726 fs_info->fs_devices->open_devices--;
2727 fs_info->fs_devices->rw_devices--;
2728 fs_info->fs_devices->total_devices--;
2729 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2730 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2731 btrfs_set_super_total_bytes(fs_info->super_copy,
2732 orig_super_total_bytes);
2733 btrfs_set_super_num_devices(fs_info->super_copy,
2734 orig_super_num_devices);
2735 mutex_unlock(&fs_info->chunk_mutex);
2736 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2739 btrfs_set_sb_rdonly(sb);
2741 btrfs_end_transaction(trans);
2743 btrfs_destroy_dev_zone_info(device);
2745 btrfs_free_device(device);
2747 blkdev_put(bdev, FMODE_EXCL);
2749 mutex_unlock(&uuid_mutex);
2750 up_write(&sb->s_umount);
2755 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2756 struct btrfs_device *device)
2759 struct btrfs_path *path;
2760 struct btrfs_root *root = device->fs_info->chunk_root;
2761 struct btrfs_dev_item *dev_item;
2762 struct extent_buffer *leaf;
2763 struct btrfs_key key;
2765 path = btrfs_alloc_path();
2769 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2770 key.type = BTRFS_DEV_ITEM_KEY;
2771 key.offset = device->devid;
2773 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2782 leaf = path->nodes[0];
2783 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2785 btrfs_set_device_id(leaf, dev_item, device->devid);
2786 btrfs_set_device_type(leaf, dev_item, device->type);
2787 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2788 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2789 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2790 btrfs_set_device_total_bytes(leaf, dev_item,
2791 btrfs_device_get_disk_total_bytes(device));
2792 btrfs_set_device_bytes_used(leaf, dev_item,
2793 btrfs_device_get_bytes_used(device));
2794 btrfs_mark_buffer_dirty(leaf);
2797 btrfs_free_path(path);
2801 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2802 struct btrfs_device *device, u64 new_size)
2804 struct btrfs_fs_info *fs_info = device->fs_info;
2805 struct btrfs_super_block *super_copy = fs_info->super_copy;
2809 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2812 new_size = round_down(new_size, fs_info->sectorsize);
2814 mutex_lock(&fs_info->chunk_mutex);
2815 old_total = btrfs_super_total_bytes(super_copy);
2816 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2818 if (new_size <= device->total_bytes ||
2819 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2820 mutex_unlock(&fs_info->chunk_mutex);
2824 btrfs_set_super_total_bytes(super_copy,
2825 round_down(old_total + diff, fs_info->sectorsize));
2826 device->fs_devices->total_rw_bytes += diff;
2828 btrfs_device_set_total_bytes(device, new_size);
2829 btrfs_device_set_disk_total_bytes(device, new_size);
2830 btrfs_clear_space_info_full(device->fs_info);
2831 if (list_empty(&device->post_commit_list))
2832 list_add_tail(&device->post_commit_list,
2833 &trans->transaction->dev_update_list);
2834 mutex_unlock(&fs_info->chunk_mutex);
2836 return btrfs_update_device(trans, device);
2839 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2841 struct btrfs_fs_info *fs_info = trans->fs_info;
2842 struct btrfs_root *root = fs_info->chunk_root;
2844 struct btrfs_path *path;
2845 struct btrfs_key key;
2847 path = btrfs_alloc_path();
2851 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2852 key.offset = chunk_offset;
2853 key.type = BTRFS_CHUNK_ITEM_KEY;
2855 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2858 else if (ret > 0) { /* Logic error or corruption */
2859 btrfs_handle_fs_error(fs_info, -ENOENT,
2860 "Failed lookup while freeing chunk.");
2865 ret = btrfs_del_item(trans, root, path);
2867 btrfs_handle_fs_error(fs_info, ret,
2868 "Failed to delete chunk item.");
2870 btrfs_free_path(path);
2874 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2876 struct btrfs_super_block *super_copy = fs_info->super_copy;
2877 struct btrfs_disk_key *disk_key;
2878 struct btrfs_chunk *chunk;
2885 struct btrfs_key key;
2887 lockdep_assert_held(&fs_info->chunk_mutex);
2888 array_size = btrfs_super_sys_array_size(super_copy);
2890 ptr = super_copy->sys_chunk_array;
2893 while (cur < array_size) {
2894 disk_key = (struct btrfs_disk_key *)ptr;
2895 btrfs_disk_key_to_cpu(&key, disk_key);
2897 len = sizeof(*disk_key);
2899 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2900 chunk = (struct btrfs_chunk *)(ptr + len);
2901 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2902 len += btrfs_chunk_item_size(num_stripes);
2907 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2908 key.offset == chunk_offset) {
2909 memmove(ptr, ptr + len, array_size - (cur + len));
2911 btrfs_set_super_sys_array_size(super_copy, array_size);
2921 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
2922 * @logical: Logical block offset in bytes.
2923 * @length: Length of extent in bytes.
2925 * Return: Chunk mapping or ERR_PTR.
2927 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
2928 u64 logical, u64 length)
2930 struct extent_map_tree *em_tree;
2931 struct extent_map *em;
2933 em_tree = &fs_info->mapping_tree;
2934 read_lock(&em_tree->lock);
2935 em = lookup_extent_mapping(em_tree, logical, length);
2936 read_unlock(&em_tree->lock);
2939 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2941 return ERR_PTR(-EINVAL);
2944 if (em->start > logical || em->start + em->len < logical) {
2946 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2947 logical, length, em->start, em->start + em->len);
2948 free_extent_map(em);
2949 return ERR_PTR(-EINVAL);
2952 /* callers are responsible for dropping em's ref. */
2956 static int remove_chunk_item(struct btrfs_trans_handle *trans,
2957 struct map_lookup *map, u64 chunk_offset)
2962 * Removing chunk items and updating the device items in the chunks btree
2963 * requires holding the chunk_mutex.
2964 * See the comment at btrfs_chunk_alloc() for the details.
2966 lockdep_assert_held(&trans->fs_info->chunk_mutex);
2968 for (i = 0; i < map->num_stripes; i++) {
2971 ret = btrfs_update_device(trans, map->stripes[i].dev);
2976 return btrfs_free_chunk(trans, chunk_offset);
2979 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2981 struct btrfs_fs_info *fs_info = trans->fs_info;
2982 struct extent_map *em;
2983 struct map_lookup *map;
2984 u64 dev_extent_len = 0;
2986 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2988 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
2991 * This is a logic error, but we don't want to just rely on the
2992 * user having built with ASSERT enabled, so if ASSERT doesn't
2993 * do anything we still error out.
2998 map = em->map_lookup;
3001 * First delete the device extent items from the devices btree.
3002 * We take the device_list_mutex to avoid racing with the finishing phase
3003 * of a device replace operation. See the comment below before acquiring
3004 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
3005 * because that can result in a deadlock when deleting the device extent
3006 * items from the devices btree - COWing an extent buffer from the btree
3007 * may result in allocating a new metadata chunk, which would attempt to
3008 * lock again fs_info->chunk_mutex.
3010 mutex_lock(&fs_devices->device_list_mutex);
3011 for (i = 0; i < map->num_stripes; i++) {
3012 struct btrfs_device *device = map->stripes[i].dev;
3013 ret = btrfs_free_dev_extent(trans, device,
3014 map->stripes[i].physical,
3017 mutex_unlock(&fs_devices->device_list_mutex);
3018 btrfs_abort_transaction(trans, ret);
3022 if (device->bytes_used > 0) {
3023 mutex_lock(&fs_info->chunk_mutex);
3024 btrfs_device_set_bytes_used(device,
3025 device->bytes_used - dev_extent_len);
3026 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3027 btrfs_clear_space_info_full(fs_info);
3028 mutex_unlock(&fs_info->chunk_mutex);
3031 mutex_unlock(&fs_devices->device_list_mutex);
3034 * We acquire fs_info->chunk_mutex for 2 reasons:
3036 * 1) Just like with the first phase of the chunk allocation, we must
3037 * reserve system space, do all chunk btree updates and deletions, and
3038 * update the system chunk array in the superblock while holding this
3039 * mutex. This is for similar reasons as explained on the comment at
3040 * the top of btrfs_chunk_alloc();
3042 * 2) Prevent races with the final phase of a device replace operation
3043 * that replaces the device object associated with the map's stripes,
3044 * because the device object's id can change at any time during that
3045 * final phase of the device replace operation
3046 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3047 * replaced device and then see it with an ID of
3048 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3049 * the device item, which does not exists on the chunk btree.
3050 * The finishing phase of device replace acquires both the
3051 * device_list_mutex and the chunk_mutex, in that order, so we are
3052 * safe by just acquiring the chunk_mutex.
3054 trans->removing_chunk = true;
3055 mutex_lock(&fs_info->chunk_mutex);
3057 check_system_chunk(trans, map->type);
3059 ret = remove_chunk_item(trans, map, chunk_offset);
3061 * Normally we should not get -ENOSPC since we reserved space before
3062 * through the call to check_system_chunk().
3064 * Despite our system space_info having enough free space, we may not
3065 * be able to allocate extents from its block groups, because all have
3066 * an incompatible profile, which will force us to allocate a new system
3067 * block group with the right profile, or right after we called
3068 * check_system_space() above, a scrub turned the only system block group
3069 * with enough free space into RO mode.
3070 * This is explained with more detail at do_chunk_alloc().
3072 * So if we get -ENOSPC, allocate a new system chunk and retry once.
3074 if (ret == -ENOSPC) {
3075 const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
3076 struct btrfs_block_group *sys_bg;
3078 sys_bg = btrfs_alloc_chunk(trans, sys_flags);
3079 if (IS_ERR(sys_bg)) {
3080 ret = PTR_ERR(sys_bg);
3081 btrfs_abort_transaction(trans, ret);
3085 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3087 btrfs_abort_transaction(trans, ret);
3091 ret = remove_chunk_item(trans, map, chunk_offset);
3093 btrfs_abort_transaction(trans, ret);
3097 btrfs_abort_transaction(trans, ret);
3101 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3103 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3104 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3106 btrfs_abort_transaction(trans, ret);
3111 mutex_unlock(&fs_info->chunk_mutex);
3112 trans->removing_chunk = false;
3115 * We are done with chunk btree updates and deletions, so release the
3116 * system space we previously reserved (with check_system_chunk()).
3118 btrfs_trans_release_chunk_metadata(trans);
3120 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3122 btrfs_abort_transaction(trans, ret);
3127 if (trans->removing_chunk) {
3128 mutex_unlock(&fs_info->chunk_mutex);
3129 trans->removing_chunk = false;
3132 free_extent_map(em);
3136 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3138 struct btrfs_root *root = fs_info->chunk_root;
3139 struct btrfs_trans_handle *trans;
3140 struct btrfs_block_group *block_group;
3145 * Prevent races with automatic removal of unused block groups.
3146 * After we relocate and before we remove the chunk with offset
3147 * chunk_offset, automatic removal of the block group can kick in,
3148 * resulting in a failure when calling btrfs_remove_chunk() below.
3150 * Make sure to acquire this mutex before doing a tree search (dev
3151 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3152 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3153 * we release the path used to search the chunk/dev tree and before
3154 * the current task acquires this mutex and calls us.
3156 lockdep_assert_held(&fs_info->reclaim_bgs_lock);
3158 /* step one, relocate all the extents inside this chunk */
3159 btrfs_scrub_pause(fs_info);
3160 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3161 btrfs_scrub_continue(fs_info);
3165 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3168 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3169 length = block_group->length;
3170 btrfs_put_block_group(block_group);
3173 * On a zoned file system, discard the whole block group, this will
3174 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3175 * resetting the zone fails, don't treat it as a fatal problem from the
3176 * filesystem's point of view.
3178 if (btrfs_is_zoned(fs_info)) {
3179 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
3182 "failed to reset zone %llu after relocation",
3186 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3188 if (IS_ERR(trans)) {
3189 ret = PTR_ERR(trans);
3190 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3195 * step two, delete the device extents and the
3196 * chunk tree entries
3198 ret = btrfs_remove_chunk(trans, chunk_offset);
3199 btrfs_end_transaction(trans);
3203 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3205 struct btrfs_root *chunk_root = fs_info->chunk_root;
3206 struct btrfs_path *path;
3207 struct extent_buffer *leaf;
3208 struct btrfs_chunk *chunk;
3209 struct btrfs_key key;
3210 struct btrfs_key found_key;
3212 bool retried = false;
3216 path = btrfs_alloc_path();
3221 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3222 key.offset = (u64)-1;
3223 key.type = BTRFS_CHUNK_ITEM_KEY;
3226 mutex_lock(&fs_info->reclaim_bgs_lock);
3227 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3229 mutex_unlock(&fs_info->reclaim_bgs_lock);
3232 BUG_ON(ret == 0); /* Corruption */
3234 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3237 mutex_unlock(&fs_info->reclaim_bgs_lock);
3243 leaf = path->nodes[0];
3244 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3246 chunk = btrfs_item_ptr(leaf, path->slots[0],
3247 struct btrfs_chunk);
3248 chunk_type = btrfs_chunk_type(leaf, chunk);
3249 btrfs_release_path(path);
3251 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3252 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3258 mutex_unlock(&fs_info->reclaim_bgs_lock);
3260 if (found_key.offset == 0)
3262 key.offset = found_key.offset - 1;
3265 if (failed && !retried) {
3269 } else if (WARN_ON(failed && retried)) {
3273 btrfs_free_path(path);
3278 * return 1 : allocate a data chunk successfully,
3279 * return <0: errors during allocating a data chunk,
3280 * return 0 : no need to allocate a data chunk.
3282 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3285 struct btrfs_block_group *cache;
3289 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3291 chunk_type = cache->flags;
3292 btrfs_put_block_group(cache);
3294 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3297 spin_lock(&fs_info->data_sinfo->lock);
3298 bytes_used = fs_info->data_sinfo->bytes_used;
3299 spin_unlock(&fs_info->data_sinfo->lock);
3302 struct btrfs_trans_handle *trans;
3305 trans = btrfs_join_transaction(fs_info->tree_root);
3307 return PTR_ERR(trans);
3309 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3310 btrfs_end_transaction(trans);
3319 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3320 struct btrfs_balance_control *bctl)
3322 struct btrfs_root *root = fs_info->tree_root;
3323 struct btrfs_trans_handle *trans;
3324 struct btrfs_balance_item *item;
3325 struct btrfs_disk_balance_args disk_bargs;
3326 struct btrfs_path *path;
3327 struct extent_buffer *leaf;
3328 struct btrfs_key key;
3331 path = btrfs_alloc_path();
3335 trans = btrfs_start_transaction(root, 0);
3336 if (IS_ERR(trans)) {
3337 btrfs_free_path(path);
3338 return PTR_ERR(trans);
3341 key.objectid = BTRFS_BALANCE_OBJECTID;
3342 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3345 ret = btrfs_insert_empty_item(trans, root, path, &key,
3350 leaf = path->nodes[0];
3351 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3353 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3355 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3356 btrfs_set_balance_data(leaf, item, &disk_bargs);
3357 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3358 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3359 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3360 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3362 btrfs_set_balance_flags(leaf, item, bctl->flags);
3364 btrfs_mark_buffer_dirty(leaf);
3366 btrfs_free_path(path);
3367 err = btrfs_commit_transaction(trans);
3373 static int del_balance_item(struct btrfs_fs_info *fs_info)
3375 struct btrfs_root *root = fs_info->tree_root;
3376 struct btrfs_trans_handle *trans;
3377 struct btrfs_path *path;
3378 struct btrfs_key key;
3381 path = btrfs_alloc_path();
3385 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3386 if (IS_ERR(trans)) {
3387 btrfs_free_path(path);
3388 return PTR_ERR(trans);
3391 key.objectid = BTRFS_BALANCE_OBJECTID;
3392 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3395 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3403 ret = btrfs_del_item(trans, root, path);
3405 btrfs_free_path(path);
3406 err = btrfs_commit_transaction(trans);
3413 * This is a heuristic used to reduce the number of chunks balanced on
3414 * resume after balance was interrupted.
3416 static void update_balance_args(struct btrfs_balance_control *bctl)
3419 * Turn on soft mode for chunk types that were being converted.
3421 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3422 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3423 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3424 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3425 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3426 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3429 * Turn on usage filter if is not already used. The idea is
3430 * that chunks that we have already balanced should be
3431 * reasonably full. Don't do it for chunks that are being
3432 * converted - that will keep us from relocating unconverted
3433 * (albeit full) chunks.
3435 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3436 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3437 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3438 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3439 bctl->data.usage = 90;
3441 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3442 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3443 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3444 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3445 bctl->sys.usage = 90;
3447 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3448 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3449 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3450 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3451 bctl->meta.usage = 90;
3456 * Clear the balance status in fs_info and delete the balance item from disk.
3458 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3460 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3463 BUG_ON(!fs_info->balance_ctl);
3465 spin_lock(&fs_info->balance_lock);
3466 fs_info->balance_ctl = NULL;
3467 spin_unlock(&fs_info->balance_lock);
3470 ret = del_balance_item(fs_info);
3472 btrfs_handle_fs_error(fs_info, ret, NULL);
3476 * Balance filters. Return 1 if chunk should be filtered out
3477 * (should not be balanced).
3479 static int chunk_profiles_filter(u64 chunk_type,
3480 struct btrfs_balance_args *bargs)
3482 chunk_type = chunk_to_extended(chunk_type) &
3483 BTRFS_EXTENDED_PROFILE_MASK;
3485 if (bargs->profiles & chunk_type)
3491 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3492 struct btrfs_balance_args *bargs)
3494 struct btrfs_block_group *cache;
3496 u64 user_thresh_min;
3497 u64 user_thresh_max;
3500 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3501 chunk_used = cache->used;
3503 if (bargs->usage_min == 0)
3504 user_thresh_min = 0;
3506 user_thresh_min = div_factor_fine(cache->length,
3509 if (bargs->usage_max == 0)
3510 user_thresh_max = 1;
3511 else if (bargs->usage_max > 100)
3512 user_thresh_max = cache->length;
3514 user_thresh_max = div_factor_fine(cache->length,
3517 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3520 btrfs_put_block_group(cache);
3524 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3525 u64 chunk_offset, struct btrfs_balance_args *bargs)
3527 struct btrfs_block_group *cache;
3528 u64 chunk_used, user_thresh;
3531 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3532 chunk_used = cache->used;
3534 if (bargs->usage_min == 0)
3536 else if (bargs->usage > 100)
3537 user_thresh = cache->length;
3539 user_thresh = div_factor_fine(cache->length, bargs->usage);
3541 if (chunk_used < user_thresh)
3544 btrfs_put_block_group(cache);
3548 static int chunk_devid_filter(struct extent_buffer *leaf,
3549 struct btrfs_chunk *chunk,
3550 struct btrfs_balance_args *bargs)
3552 struct btrfs_stripe *stripe;
3553 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3556 for (i = 0; i < num_stripes; i++) {
3557 stripe = btrfs_stripe_nr(chunk, i);
3558 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3565 static u64 calc_data_stripes(u64 type, int num_stripes)
3567 const int index = btrfs_bg_flags_to_raid_index(type);
3568 const int ncopies = btrfs_raid_array[index].ncopies;
3569 const int nparity = btrfs_raid_array[index].nparity;
3571 return (num_stripes - nparity) / ncopies;
3574 /* [pstart, pend) */
3575 static int chunk_drange_filter(struct extent_buffer *leaf,
3576 struct btrfs_chunk *chunk,
3577 struct btrfs_balance_args *bargs)
3579 struct btrfs_stripe *stripe;
3580 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3587 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3590 type = btrfs_chunk_type(leaf, chunk);
3591 factor = calc_data_stripes(type, num_stripes);
3593 for (i = 0; i < num_stripes; i++) {
3594 stripe = btrfs_stripe_nr(chunk, i);
3595 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3598 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3599 stripe_length = btrfs_chunk_length(leaf, chunk);
3600 stripe_length = div_u64(stripe_length, factor);
3602 if (stripe_offset < bargs->pend &&
3603 stripe_offset + stripe_length > bargs->pstart)
3610 /* [vstart, vend) */
3611 static int chunk_vrange_filter(struct extent_buffer *leaf,
3612 struct btrfs_chunk *chunk,
3614 struct btrfs_balance_args *bargs)
3616 if (chunk_offset < bargs->vend &&
3617 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3618 /* at least part of the chunk is inside this vrange */
3624 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3625 struct btrfs_chunk *chunk,
3626 struct btrfs_balance_args *bargs)
3628 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3630 if (bargs->stripes_min <= num_stripes
3631 && num_stripes <= bargs->stripes_max)
3637 static int chunk_soft_convert_filter(u64 chunk_type,
3638 struct btrfs_balance_args *bargs)
3640 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3643 chunk_type = chunk_to_extended(chunk_type) &
3644 BTRFS_EXTENDED_PROFILE_MASK;
3646 if (bargs->target == chunk_type)
3652 static int should_balance_chunk(struct extent_buffer *leaf,
3653 struct btrfs_chunk *chunk, u64 chunk_offset)
3655 struct btrfs_fs_info *fs_info = leaf->fs_info;
3656 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3657 struct btrfs_balance_args *bargs = NULL;
3658 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3661 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3662 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3666 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3667 bargs = &bctl->data;
3668 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3670 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3671 bargs = &bctl->meta;
3673 /* profiles filter */
3674 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3675 chunk_profiles_filter(chunk_type, bargs)) {
3680 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3681 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3683 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3684 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3689 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3690 chunk_devid_filter(leaf, chunk, bargs)) {
3694 /* drange filter, makes sense only with devid filter */
3695 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3696 chunk_drange_filter(leaf, chunk, bargs)) {
3701 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3702 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3706 /* stripes filter */
3707 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3708 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3712 /* soft profile changing mode */
3713 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3714 chunk_soft_convert_filter(chunk_type, bargs)) {
3719 * limited by count, must be the last filter
3721 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3722 if (bargs->limit == 0)
3726 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3728 * Same logic as the 'limit' filter; the minimum cannot be
3729 * determined here because we do not have the global information
3730 * about the count of all chunks that satisfy the filters.
3732 if (bargs->limit_max == 0)
3741 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3743 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3744 struct btrfs_root *chunk_root = fs_info->chunk_root;
3746 struct btrfs_chunk *chunk;
3747 struct btrfs_path *path = NULL;
3748 struct btrfs_key key;
3749 struct btrfs_key found_key;
3750 struct extent_buffer *leaf;
3753 int enospc_errors = 0;
3754 bool counting = true;
3755 /* The single value limit and min/max limits use the same bytes in the */
3756 u64 limit_data = bctl->data.limit;
3757 u64 limit_meta = bctl->meta.limit;
3758 u64 limit_sys = bctl->sys.limit;
3762 int chunk_reserved = 0;
3764 path = btrfs_alloc_path();
3770 /* zero out stat counters */
3771 spin_lock(&fs_info->balance_lock);
3772 memset(&bctl->stat, 0, sizeof(bctl->stat));
3773 spin_unlock(&fs_info->balance_lock);
3777 * The single value limit and min/max limits use the same bytes
3780 bctl->data.limit = limit_data;
3781 bctl->meta.limit = limit_meta;
3782 bctl->sys.limit = limit_sys;
3784 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3785 key.offset = (u64)-1;
3786 key.type = BTRFS_CHUNK_ITEM_KEY;
3789 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3790 atomic_read(&fs_info->balance_cancel_req)) {
3795 mutex_lock(&fs_info->reclaim_bgs_lock);
3796 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3798 mutex_unlock(&fs_info->reclaim_bgs_lock);
3803 * this shouldn't happen, it means the last relocate
3807 BUG(); /* FIXME break ? */
3809 ret = btrfs_previous_item(chunk_root, path, 0,
3810 BTRFS_CHUNK_ITEM_KEY);
3812 mutex_unlock(&fs_info->reclaim_bgs_lock);
3817 leaf = path->nodes[0];
3818 slot = path->slots[0];
3819 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3821 if (found_key.objectid != key.objectid) {
3822 mutex_unlock(&fs_info->reclaim_bgs_lock);
3826 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3827 chunk_type = btrfs_chunk_type(leaf, chunk);
3830 spin_lock(&fs_info->balance_lock);
3831 bctl->stat.considered++;
3832 spin_unlock(&fs_info->balance_lock);
3835 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3837 btrfs_release_path(path);
3839 mutex_unlock(&fs_info->reclaim_bgs_lock);
3844 mutex_unlock(&fs_info->reclaim_bgs_lock);
3845 spin_lock(&fs_info->balance_lock);
3846 bctl->stat.expected++;
3847 spin_unlock(&fs_info->balance_lock);
3849 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3851 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3853 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3860 * Apply limit_min filter, no need to check if the LIMITS
3861 * filter is used, limit_min is 0 by default
3863 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3864 count_data < bctl->data.limit_min)
3865 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3866 count_meta < bctl->meta.limit_min)
3867 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3868 count_sys < bctl->sys.limit_min)) {
3869 mutex_unlock(&fs_info->reclaim_bgs_lock);
3873 if (!chunk_reserved) {
3875 * We may be relocating the only data chunk we have,
3876 * which could potentially end up with losing data's
3877 * raid profile, so lets allocate an empty one in
3880 ret = btrfs_may_alloc_data_chunk(fs_info,
3883 mutex_unlock(&fs_info->reclaim_bgs_lock);
3885 } else if (ret == 1) {
3890 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3891 mutex_unlock(&fs_info->reclaim_bgs_lock);
3892 if (ret == -ENOSPC) {
3894 } else if (ret == -ETXTBSY) {
3896 "skipping relocation of block group %llu due to active swapfile",
3902 spin_lock(&fs_info->balance_lock);
3903 bctl->stat.completed++;
3904 spin_unlock(&fs_info->balance_lock);
3907 if (found_key.offset == 0)
3909 key.offset = found_key.offset - 1;
3913 btrfs_release_path(path);
3918 btrfs_free_path(path);
3919 if (enospc_errors) {
3920 btrfs_info(fs_info, "%d enospc errors during balance",
3930 * alloc_profile_is_valid - see if a given profile is valid and reduced
3931 * @flags: profile to validate
3932 * @extended: if true @flags is treated as an extended profile
3934 static int alloc_profile_is_valid(u64 flags, int extended)
3936 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3937 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3939 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3941 /* 1) check that all other bits are zeroed */
3945 /* 2) see if profile is reduced */
3947 return !extended; /* "0" is valid for usual profiles */
3949 return has_single_bit_set(flags);
3952 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3954 /* cancel requested || normal exit path */
3955 return atomic_read(&fs_info->balance_cancel_req) ||
3956 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3957 atomic_read(&fs_info->balance_cancel_req) == 0);
3961 * Validate target profile against allowed profiles and return true if it's OK.
3962 * Otherwise print the error message and return false.
3964 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
3965 const struct btrfs_balance_args *bargs,
3966 u64 allowed, const char *type)
3968 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3971 /* Profile is valid and does not have bits outside of the allowed set */
3972 if (alloc_profile_is_valid(bargs->target, 1) &&
3973 (bargs->target & ~allowed) == 0)
3976 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
3977 type, btrfs_bg_type_to_raid_name(bargs->target));
3982 * Fill @buf with textual description of balance filter flags @bargs, up to
3983 * @size_buf including the terminating null. The output may be trimmed if it
3984 * does not fit into the provided buffer.
3986 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
3990 u32 size_bp = size_buf;
3992 u64 flags = bargs->flags;
3993 char tmp_buf[128] = {'\0'};
3998 #define CHECK_APPEND_NOARG(a) \
4000 ret = snprintf(bp, size_bp, (a)); \
4001 if (ret < 0 || ret >= size_bp) \
4002 goto out_overflow; \
4007 #define CHECK_APPEND_1ARG(a, v1) \
4009 ret = snprintf(bp, size_bp, (a), (v1)); \
4010 if (ret < 0 || ret >= size_bp) \
4011 goto out_overflow; \
4016 #define CHECK_APPEND_2ARG(a, v1, v2) \
4018 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4019 if (ret < 0 || ret >= size_bp) \
4020 goto out_overflow; \
4025 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
4026 CHECK_APPEND_1ARG("convert=%s,",
4027 btrfs_bg_type_to_raid_name(bargs->target));
4029 if (flags & BTRFS_BALANCE_ARGS_SOFT)
4030 CHECK_APPEND_NOARG("soft,");
4032 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4033 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4035 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4038 if (flags & BTRFS_BALANCE_ARGS_USAGE)
4039 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4041 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4042 CHECK_APPEND_2ARG("usage=%u..%u,",
4043 bargs->usage_min, bargs->usage_max);
4045 if (flags & BTRFS_BALANCE_ARGS_DEVID)
4046 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4048 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4049 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4050 bargs->pstart, bargs->pend);
4052 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4053 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4054 bargs->vstart, bargs->vend);
4056 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4057 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4059 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4060 CHECK_APPEND_2ARG("limit=%u..%u,",
4061 bargs->limit_min, bargs->limit_max);
4063 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4064 CHECK_APPEND_2ARG("stripes=%u..%u,",
4065 bargs->stripes_min, bargs->stripes_max);
4067 #undef CHECK_APPEND_2ARG
4068 #undef CHECK_APPEND_1ARG
4069 #undef CHECK_APPEND_NOARG
4073 if (size_bp < size_buf)
4074 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4079 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4081 u32 size_buf = 1024;
4082 char tmp_buf[192] = {'\0'};
4085 u32 size_bp = size_buf;
4087 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4089 buf = kzalloc(size_buf, GFP_KERNEL);
4095 #define CHECK_APPEND_1ARG(a, v1) \
4097 ret = snprintf(bp, size_bp, (a), (v1)); \
4098 if (ret < 0 || ret >= size_bp) \
4099 goto out_overflow; \
4104 if (bctl->flags & BTRFS_BALANCE_FORCE)
4105 CHECK_APPEND_1ARG("%s", "-f ");
4107 if (bctl->flags & BTRFS_BALANCE_DATA) {
4108 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4109 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4112 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4113 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4114 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4117 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4118 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4119 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4122 #undef CHECK_APPEND_1ARG
4126 if (size_bp < size_buf)
4127 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4128 btrfs_info(fs_info, "balance: %s %s",
4129 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4130 "resume" : "start", buf);
4136 * Should be called with balance mutexe held
4138 int btrfs_balance(struct btrfs_fs_info *fs_info,
4139 struct btrfs_balance_control *bctl,
4140 struct btrfs_ioctl_balance_args *bargs)
4142 u64 meta_target, data_target;
4148 bool reducing_redundancy;
4151 if (btrfs_fs_closing(fs_info) ||
4152 atomic_read(&fs_info->balance_pause_req) ||
4153 btrfs_should_cancel_balance(fs_info)) {
4158 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4159 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4163 * In case of mixed groups both data and meta should be picked,
4164 * and identical options should be given for both of them.
4166 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4167 if (mixed && (bctl->flags & allowed)) {
4168 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4169 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4170 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4172 "balance: mixed groups data and metadata options must be the same");
4179 * rw_devices will not change at the moment, device add/delete/replace
4182 num_devices = fs_info->fs_devices->rw_devices;
4185 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4186 * special bit for it, to make it easier to distinguish. Thus we need
4187 * to set it manually, or balance would refuse the profile.
4189 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4190 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4191 if (num_devices >= btrfs_raid_array[i].devs_min)
4192 allowed |= btrfs_raid_array[i].bg_flag;
4194 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4195 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4196 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4202 * Allow to reduce metadata or system integrity only if force set for
4203 * profiles with redundancy (copies, parity)
4206 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4207 if (btrfs_raid_array[i].ncopies >= 2 ||
4208 btrfs_raid_array[i].tolerated_failures >= 1)
4209 allowed |= btrfs_raid_array[i].bg_flag;
4212 seq = read_seqbegin(&fs_info->profiles_lock);
4214 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4215 (fs_info->avail_system_alloc_bits & allowed) &&
4216 !(bctl->sys.target & allowed)) ||
4217 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4218 (fs_info->avail_metadata_alloc_bits & allowed) &&
4219 !(bctl->meta.target & allowed)))
4220 reducing_redundancy = true;
4222 reducing_redundancy = false;
4224 /* if we're not converting, the target field is uninitialized */
4225 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4226 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4227 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4228 bctl->data.target : fs_info->avail_data_alloc_bits;
4229 } while (read_seqretry(&fs_info->profiles_lock, seq));
4231 if (reducing_redundancy) {
4232 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4234 "balance: force reducing metadata redundancy");
4237 "balance: reduces metadata redundancy, use --force if you want this");
4243 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4244 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4246 "balance: metadata profile %s has lower redundancy than data profile %s",
4247 btrfs_bg_type_to_raid_name(meta_target),
4248 btrfs_bg_type_to_raid_name(data_target));
4251 ret = insert_balance_item(fs_info, bctl);
4252 if (ret && ret != -EEXIST)
4255 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4256 BUG_ON(ret == -EEXIST);
4257 BUG_ON(fs_info->balance_ctl);
4258 spin_lock(&fs_info->balance_lock);
4259 fs_info->balance_ctl = bctl;
4260 spin_unlock(&fs_info->balance_lock);
4262 BUG_ON(ret != -EEXIST);
4263 spin_lock(&fs_info->balance_lock);
4264 update_balance_args(bctl);
4265 spin_unlock(&fs_info->balance_lock);
4268 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4269 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4270 describe_balance_start_or_resume(fs_info);
4271 mutex_unlock(&fs_info->balance_mutex);
4273 ret = __btrfs_balance(fs_info);
4275 mutex_lock(&fs_info->balance_mutex);
4276 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req))
4277 btrfs_info(fs_info, "balance: paused");
4279 * Balance can be canceled by:
4281 * - Regular cancel request
4282 * Then ret == -ECANCELED and balance_cancel_req > 0
4284 * - Fatal signal to "btrfs" process
4285 * Either the signal caught by wait_reserve_ticket() and callers
4286 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4288 * Either way, in this case balance_cancel_req = 0, and
4289 * ret == -EINTR or ret == -ECANCELED.
4291 * So here we only check the return value to catch canceled balance.
4293 else if (ret == -ECANCELED || ret == -EINTR)
4294 btrfs_info(fs_info, "balance: canceled");
4296 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4298 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4301 memset(bargs, 0, sizeof(*bargs));
4302 btrfs_update_ioctl_balance_args(fs_info, bargs);
4305 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4306 balance_need_close(fs_info)) {
4307 reset_balance_state(fs_info);
4308 btrfs_exclop_finish(fs_info);
4311 wake_up(&fs_info->balance_wait_q);
4315 if (bctl->flags & BTRFS_BALANCE_RESUME)
4316 reset_balance_state(fs_info);
4319 btrfs_exclop_finish(fs_info);
4324 static int balance_kthread(void *data)
4326 struct btrfs_fs_info *fs_info = data;
4329 mutex_lock(&fs_info->balance_mutex);
4330 if (fs_info->balance_ctl)
4331 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4332 mutex_unlock(&fs_info->balance_mutex);
4337 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4339 struct task_struct *tsk;
4341 mutex_lock(&fs_info->balance_mutex);
4342 if (!fs_info->balance_ctl) {
4343 mutex_unlock(&fs_info->balance_mutex);
4346 mutex_unlock(&fs_info->balance_mutex);
4348 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4349 btrfs_info(fs_info, "balance: resume skipped");
4354 * A ro->rw remount sequence should continue with the paused balance
4355 * regardless of who pauses it, system or the user as of now, so set
4358 spin_lock(&fs_info->balance_lock);
4359 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4360 spin_unlock(&fs_info->balance_lock);
4362 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4363 return PTR_ERR_OR_ZERO(tsk);
4366 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4368 struct btrfs_balance_control *bctl;
4369 struct btrfs_balance_item *item;
4370 struct btrfs_disk_balance_args disk_bargs;
4371 struct btrfs_path *path;
4372 struct extent_buffer *leaf;
4373 struct btrfs_key key;
4376 path = btrfs_alloc_path();
4380 key.objectid = BTRFS_BALANCE_OBJECTID;
4381 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4384 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4387 if (ret > 0) { /* ret = -ENOENT; */
4392 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4398 leaf = path->nodes[0];
4399 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4401 bctl->flags = btrfs_balance_flags(leaf, item);
4402 bctl->flags |= BTRFS_BALANCE_RESUME;
4404 btrfs_balance_data(leaf, item, &disk_bargs);
4405 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4406 btrfs_balance_meta(leaf, item, &disk_bargs);
4407 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4408 btrfs_balance_sys(leaf, item, &disk_bargs);
4409 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4412 * This should never happen, as the paused balance state is recovered
4413 * during mount without any chance of other exclusive ops to collide.
4415 * This gives the exclusive op status to balance and keeps in paused
4416 * state until user intervention (cancel or umount). If the ownership
4417 * cannot be assigned, show a message but do not fail. The balance
4418 * is in a paused state and must have fs_info::balance_ctl properly
4421 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE))
4423 "balance: cannot set exclusive op status, resume manually");
4425 btrfs_release_path(path);
4427 mutex_lock(&fs_info->balance_mutex);
4428 BUG_ON(fs_info->balance_ctl);
4429 spin_lock(&fs_info->balance_lock);
4430 fs_info->balance_ctl = bctl;
4431 spin_unlock(&fs_info->balance_lock);
4432 mutex_unlock(&fs_info->balance_mutex);
4434 btrfs_free_path(path);
4438 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4442 mutex_lock(&fs_info->balance_mutex);
4443 if (!fs_info->balance_ctl) {
4444 mutex_unlock(&fs_info->balance_mutex);
4448 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4449 atomic_inc(&fs_info->balance_pause_req);
4450 mutex_unlock(&fs_info->balance_mutex);
4452 wait_event(fs_info->balance_wait_q,
4453 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4455 mutex_lock(&fs_info->balance_mutex);
4456 /* we are good with balance_ctl ripped off from under us */
4457 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4458 atomic_dec(&fs_info->balance_pause_req);
4463 mutex_unlock(&fs_info->balance_mutex);
4467 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4469 mutex_lock(&fs_info->balance_mutex);
4470 if (!fs_info->balance_ctl) {
4471 mutex_unlock(&fs_info->balance_mutex);
4476 * A paused balance with the item stored on disk can be resumed at
4477 * mount time if the mount is read-write. Otherwise it's still paused
4478 * and we must not allow cancelling as it deletes the item.
4480 if (sb_rdonly(fs_info->sb)) {
4481 mutex_unlock(&fs_info->balance_mutex);
4485 atomic_inc(&fs_info->balance_cancel_req);
4487 * if we are running just wait and return, balance item is
4488 * deleted in btrfs_balance in this case
4490 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4491 mutex_unlock(&fs_info->balance_mutex);
4492 wait_event(fs_info->balance_wait_q,
4493 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4494 mutex_lock(&fs_info->balance_mutex);
4496 mutex_unlock(&fs_info->balance_mutex);
4498 * Lock released to allow other waiters to continue, we'll
4499 * reexamine the status again.
4501 mutex_lock(&fs_info->balance_mutex);
4503 if (fs_info->balance_ctl) {
4504 reset_balance_state(fs_info);
4505 btrfs_exclop_finish(fs_info);
4506 btrfs_info(fs_info, "balance: canceled");
4510 BUG_ON(fs_info->balance_ctl ||
4511 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4512 atomic_dec(&fs_info->balance_cancel_req);
4513 mutex_unlock(&fs_info->balance_mutex);
4517 int btrfs_uuid_scan_kthread(void *data)
4519 struct btrfs_fs_info *fs_info = data;
4520 struct btrfs_root *root = fs_info->tree_root;
4521 struct btrfs_key key;
4522 struct btrfs_path *path = NULL;
4524 struct extent_buffer *eb;
4526 struct btrfs_root_item root_item;
4528 struct btrfs_trans_handle *trans = NULL;
4529 bool closing = false;
4531 path = btrfs_alloc_path();
4538 key.type = BTRFS_ROOT_ITEM_KEY;
4542 if (btrfs_fs_closing(fs_info)) {
4546 ret = btrfs_search_forward(root, &key, path,
4547 BTRFS_OLDEST_GENERATION);
4554 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4555 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4556 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4557 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4560 eb = path->nodes[0];
4561 slot = path->slots[0];
4562 item_size = btrfs_item_size_nr(eb, slot);
4563 if (item_size < sizeof(root_item))
4566 read_extent_buffer(eb, &root_item,
4567 btrfs_item_ptr_offset(eb, slot),
4568 (int)sizeof(root_item));
4569 if (btrfs_root_refs(&root_item) == 0)
4572 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4573 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4577 btrfs_release_path(path);
4579 * 1 - subvol uuid item
4580 * 1 - received_subvol uuid item
4582 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4583 if (IS_ERR(trans)) {
4584 ret = PTR_ERR(trans);
4592 btrfs_release_path(path);
4593 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4594 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4595 BTRFS_UUID_KEY_SUBVOL,
4598 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4604 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4605 ret = btrfs_uuid_tree_add(trans,
4606 root_item.received_uuid,
4607 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4610 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4617 btrfs_release_path(path);
4619 ret = btrfs_end_transaction(trans);
4625 if (key.offset < (u64)-1) {
4627 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4629 key.type = BTRFS_ROOT_ITEM_KEY;
4630 } else if (key.objectid < (u64)-1) {
4632 key.type = BTRFS_ROOT_ITEM_KEY;
4641 btrfs_free_path(path);
4642 if (trans && !IS_ERR(trans))
4643 btrfs_end_transaction(trans);
4645 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4647 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4648 up(&fs_info->uuid_tree_rescan_sem);
4652 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4654 struct btrfs_trans_handle *trans;
4655 struct btrfs_root *tree_root = fs_info->tree_root;
4656 struct btrfs_root *uuid_root;
4657 struct task_struct *task;
4664 trans = btrfs_start_transaction(tree_root, 2);
4666 return PTR_ERR(trans);
4668 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4669 if (IS_ERR(uuid_root)) {
4670 ret = PTR_ERR(uuid_root);
4671 btrfs_abort_transaction(trans, ret);
4672 btrfs_end_transaction(trans);
4676 fs_info->uuid_root = uuid_root;
4678 ret = btrfs_commit_transaction(trans);
4682 down(&fs_info->uuid_tree_rescan_sem);
4683 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4685 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4686 btrfs_warn(fs_info, "failed to start uuid_scan task");
4687 up(&fs_info->uuid_tree_rescan_sem);
4688 return PTR_ERR(task);
4695 * shrinking a device means finding all of the device extents past
4696 * the new size, and then following the back refs to the chunks.
4697 * The chunk relocation code actually frees the device extent
4699 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4701 struct btrfs_fs_info *fs_info = device->fs_info;
4702 struct btrfs_root *root = fs_info->dev_root;
4703 struct btrfs_trans_handle *trans;
4704 struct btrfs_dev_extent *dev_extent = NULL;
4705 struct btrfs_path *path;
4711 bool retried = false;
4712 struct extent_buffer *l;
4713 struct btrfs_key key;
4714 struct btrfs_super_block *super_copy = fs_info->super_copy;
4715 u64 old_total = btrfs_super_total_bytes(super_copy);
4716 u64 old_size = btrfs_device_get_total_bytes(device);
4720 new_size = round_down(new_size, fs_info->sectorsize);
4722 diff = round_down(old_size - new_size, fs_info->sectorsize);
4724 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4727 path = btrfs_alloc_path();
4731 path->reada = READA_BACK;
4733 trans = btrfs_start_transaction(root, 0);
4734 if (IS_ERR(trans)) {
4735 btrfs_free_path(path);
4736 return PTR_ERR(trans);
4739 mutex_lock(&fs_info->chunk_mutex);
4741 btrfs_device_set_total_bytes(device, new_size);
4742 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4743 device->fs_devices->total_rw_bytes -= diff;
4744 atomic64_sub(diff, &fs_info->free_chunk_space);
4748 * Once the device's size has been set to the new size, ensure all
4749 * in-memory chunks are synced to disk so that the loop below sees them
4750 * and relocates them accordingly.
4752 if (contains_pending_extent(device, &start, diff)) {
4753 mutex_unlock(&fs_info->chunk_mutex);
4754 ret = btrfs_commit_transaction(trans);
4758 mutex_unlock(&fs_info->chunk_mutex);
4759 btrfs_end_transaction(trans);
4763 key.objectid = device->devid;
4764 key.offset = (u64)-1;
4765 key.type = BTRFS_DEV_EXTENT_KEY;
4768 mutex_lock(&fs_info->reclaim_bgs_lock);
4769 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4771 mutex_unlock(&fs_info->reclaim_bgs_lock);
4775 ret = btrfs_previous_item(root, path, 0, key.type);
4777 mutex_unlock(&fs_info->reclaim_bgs_lock);
4781 btrfs_release_path(path);
4786 slot = path->slots[0];
4787 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4789 if (key.objectid != device->devid) {
4790 mutex_unlock(&fs_info->reclaim_bgs_lock);
4791 btrfs_release_path(path);
4795 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4796 length = btrfs_dev_extent_length(l, dev_extent);
4798 if (key.offset + length <= new_size) {
4799 mutex_unlock(&fs_info->reclaim_bgs_lock);
4800 btrfs_release_path(path);
4804 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4805 btrfs_release_path(path);
4808 * We may be relocating the only data chunk we have,
4809 * which could potentially end up with losing data's
4810 * raid profile, so lets allocate an empty one in
4813 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4815 mutex_unlock(&fs_info->reclaim_bgs_lock);
4819 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4820 mutex_unlock(&fs_info->reclaim_bgs_lock);
4821 if (ret == -ENOSPC) {
4824 if (ret == -ETXTBSY) {
4826 "could not shrink block group %llu due to active swapfile",
4831 } while (key.offset-- > 0);
4833 if (failed && !retried) {
4837 } else if (failed && retried) {
4842 /* Shrinking succeeded, else we would be at "done". */
4843 trans = btrfs_start_transaction(root, 0);
4844 if (IS_ERR(trans)) {
4845 ret = PTR_ERR(trans);
4849 mutex_lock(&fs_info->chunk_mutex);
4850 /* Clear all state bits beyond the shrunk device size */
4851 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
4854 btrfs_device_set_disk_total_bytes(device, new_size);
4855 if (list_empty(&device->post_commit_list))
4856 list_add_tail(&device->post_commit_list,
4857 &trans->transaction->dev_update_list);
4859 WARN_ON(diff > old_total);
4860 btrfs_set_super_total_bytes(super_copy,
4861 round_down(old_total - diff, fs_info->sectorsize));
4862 mutex_unlock(&fs_info->chunk_mutex);
4864 /* Now btrfs_update_device() will change the on-disk size. */
4865 ret = btrfs_update_device(trans, device);
4867 btrfs_abort_transaction(trans, ret);
4868 btrfs_end_transaction(trans);
4870 ret = btrfs_commit_transaction(trans);
4873 btrfs_free_path(path);
4875 mutex_lock(&fs_info->chunk_mutex);
4876 btrfs_device_set_total_bytes(device, old_size);
4877 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4878 device->fs_devices->total_rw_bytes += diff;
4879 atomic64_add(diff, &fs_info->free_chunk_space);
4880 mutex_unlock(&fs_info->chunk_mutex);
4885 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4886 struct btrfs_key *key,
4887 struct btrfs_chunk *chunk, int item_size)
4889 struct btrfs_super_block *super_copy = fs_info->super_copy;
4890 struct btrfs_disk_key disk_key;
4894 lockdep_assert_held(&fs_info->chunk_mutex);
4896 array_size = btrfs_super_sys_array_size(super_copy);
4897 if (array_size + item_size + sizeof(disk_key)
4898 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
4901 ptr = super_copy->sys_chunk_array + array_size;
4902 btrfs_cpu_key_to_disk(&disk_key, key);
4903 memcpy(ptr, &disk_key, sizeof(disk_key));
4904 ptr += sizeof(disk_key);
4905 memcpy(ptr, chunk, item_size);
4906 item_size += sizeof(disk_key);
4907 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4913 * sort the devices in descending order by max_avail, total_avail
4915 static int btrfs_cmp_device_info(const void *a, const void *b)
4917 const struct btrfs_device_info *di_a = a;
4918 const struct btrfs_device_info *di_b = b;
4920 if (di_a->max_avail > di_b->max_avail)
4922 if (di_a->max_avail < di_b->max_avail)
4924 if (di_a->total_avail > di_b->total_avail)
4926 if (di_a->total_avail < di_b->total_avail)
4931 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4933 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4936 btrfs_set_fs_incompat(info, RAID56);
4939 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
4941 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
4944 btrfs_set_fs_incompat(info, RAID1C34);
4948 * Structure used internally for __btrfs_alloc_chunk() function.
4949 * Wraps needed parameters.
4951 struct alloc_chunk_ctl {
4954 /* Total number of stripes to allocate */
4956 /* sub_stripes info for map */
4958 /* Stripes per device */
4960 /* Maximum number of devices to use */
4962 /* Minimum number of devices to use */
4964 /* ndevs has to be a multiple of this */
4966 /* Number of copies */
4968 /* Number of stripes worth of bytes to store parity information */
4970 u64 max_stripe_size;
4978 static void init_alloc_chunk_ctl_policy_regular(
4979 struct btrfs_fs_devices *fs_devices,
4980 struct alloc_chunk_ctl *ctl)
4982 u64 type = ctl->type;
4984 if (type & BTRFS_BLOCK_GROUP_DATA) {
4985 ctl->max_stripe_size = SZ_1G;
4986 ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
4987 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4988 /* For larger filesystems, use larger metadata chunks */
4989 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4990 ctl->max_stripe_size = SZ_1G;
4992 ctl->max_stripe_size = SZ_256M;
4993 ctl->max_chunk_size = ctl->max_stripe_size;
4994 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4995 ctl->max_stripe_size = SZ_32M;
4996 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
4997 ctl->devs_max = min_t(int, ctl->devs_max,
4998 BTRFS_MAX_DEVS_SYS_CHUNK);
5003 /* We don't want a chunk larger than 10% of writable space */
5004 ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
5005 ctl->max_chunk_size);
5006 ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes;
5009 static void init_alloc_chunk_ctl_policy_zoned(
5010 struct btrfs_fs_devices *fs_devices,
5011 struct alloc_chunk_ctl *ctl)
5013 u64 zone_size = fs_devices->fs_info->zone_size;
5015 int min_num_stripes = ctl->devs_min * ctl->dev_stripes;
5016 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;
5017 u64 min_chunk_size = min_data_stripes * zone_size;
5018 u64 type = ctl->type;
5020 ctl->max_stripe_size = zone_size;
5021 if (type & BTRFS_BLOCK_GROUP_DATA) {
5022 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,
5024 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5025 ctl->max_chunk_size = ctl->max_stripe_size;
5026 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5027 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5028 ctl->devs_max = min_t(int, ctl->devs_max,
5029 BTRFS_MAX_DEVS_SYS_CHUNK);
5034 /* We don't want a chunk larger than 10% of writable space */
5035 limit = max(round_down(div_factor(fs_devices->total_rw_bytes, 1),
5038 ctl->max_chunk_size = min(limit, ctl->max_chunk_size);
5039 ctl->dev_extent_min = zone_size * ctl->dev_stripes;
5042 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
5043 struct alloc_chunk_ctl *ctl)
5045 int index = btrfs_bg_flags_to_raid_index(ctl->type);
5047 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
5048 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
5049 ctl->devs_max = btrfs_raid_array[index].devs_max;
5051 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
5052 ctl->devs_min = btrfs_raid_array[index].devs_min;
5053 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
5054 ctl->ncopies = btrfs_raid_array[index].ncopies;
5055 ctl->nparity = btrfs_raid_array[index].nparity;
5058 switch (fs_devices->chunk_alloc_policy) {
5059 case BTRFS_CHUNK_ALLOC_REGULAR:
5060 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
5062 case BTRFS_CHUNK_ALLOC_ZONED:
5063 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);
5070 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
5071 struct alloc_chunk_ctl *ctl,
5072 struct btrfs_device_info *devices_info)
5074 struct btrfs_fs_info *info = fs_devices->fs_info;
5075 struct btrfs_device *device;
5077 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
5084 * in the first pass through the devices list, we gather information
5085 * about the available holes on each device.
5087 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5088 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5090 "BTRFS: read-only device in alloc_list\n");
5094 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5095 &device->dev_state) ||
5096 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5099 if (device->total_bytes > device->bytes_used)
5100 total_avail = device->total_bytes - device->bytes_used;
5104 /* If there is no space on this device, skip it. */
5105 if (total_avail < ctl->dev_extent_min)
5108 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
5110 if (ret && ret != -ENOSPC)
5114 max_avail = dev_extent_want;
5116 if (max_avail < ctl->dev_extent_min) {
5117 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5119 "%s: devid %llu has no free space, have=%llu want=%llu",
5120 __func__, device->devid, max_avail,
5121 ctl->dev_extent_min);
5125 if (ndevs == fs_devices->rw_devices) {
5126 WARN(1, "%s: found more than %llu devices\n",
5127 __func__, fs_devices->rw_devices);
5130 devices_info[ndevs].dev_offset = dev_offset;
5131 devices_info[ndevs].max_avail = max_avail;
5132 devices_info[ndevs].total_avail = total_avail;
5133 devices_info[ndevs].dev = device;
5139 * now sort the devices by hole size / available space
5141 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5142 btrfs_cmp_device_info, NULL);
5147 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5148 struct btrfs_device_info *devices_info)
5150 /* Number of stripes that count for block group size */
5154 * The primary goal is to maximize the number of stripes, so use as
5155 * many devices as possible, even if the stripes are not maximum sized.
5157 * The DUP profile stores more than one stripe per device, the
5158 * max_avail is the total size so we have to adjust.
5160 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5162 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5164 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5165 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5168 * Use the number of data stripes to figure out how big this chunk is
5169 * really going to be in terms of logical address space, and compare
5170 * that answer with the max chunk size. If it's higher, we try to
5171 * reduce stripe_size.
5173 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5175 * Reduce stripe_size, round it up to a 16MB boundary again and
5176 * then use it, unless it ends up being even bigger than the
5177 * previous value we had already.
5179 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5180 data_stripes), SZ_16M),
5184 /* Align to BTRFS_STRIPE_LEN */
5185 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5186 ctl->chunk_size = ctl->stripe_size * data_stripes;
5191 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,
5192 struct btrfs_device_info *devices_info)
5194 u64 zone_size = devices_info[0].dev->zone_info->zone_size;
5195 /* Number of stripes that count for block group size */
5199 * It should hold because:
5200 * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5202 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);
5204 ctl->stripe_size = zone_size;
5205 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5206 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5208 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5209 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5210 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,
5211 ctl->stripe_size) + ctl->nparity,
5213 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5214 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5215 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);
5218 ctl->chunk_size = ctl->stripe_size * data_stripes;
5223 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5224 struct alloc_chunk_ctl *ctl,
5225 struct btrfs_device_info *devices_info)
5227 struct btrfs_fs_info *info = fs_devices->fs_info;
5230 * Round down to number of usable stripes, devs_increment can be any
5231 * number so we can't use round_down() that requires power of 2, while
5232 * rounddown is safe.
5234 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5236 if (ctl->ndevs < ctl->devs_min) {
5237 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5239 "%s: not enough devices with free space: have=%d minimum required=%d",
5240 __func__, ctl->ndevs, ctl->devs_min);
5245 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5247 switch (fs_devices->chunk_alloc_policy) {
5248 case BTRFS_CHUNK_ALLOC_REGULAR:
5249 return decide_stripe_size_regular(ctl, devices_info);
5250 case BTRFS_CHUNK_ALLOC_ZONED:
5251 return decide_stripe_size_zoned(ctl, devices_info);
5257 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
5258 struct alloc_chunk_ctl *ctl,
5259 struct btrfs_device_info *devices_info)
5261 struct btrfs_fs_info *info = trans->fs_info;
5262 struct map_lookup *map = NULL;
5263 struct extent_map_tree *em_tree;
5264 struct btrfs_block_group *block_group;
5265 struct extent_map *em;
5266 u64 start = ctl->start;
5267 u64 type = ctl->type;
5272 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
5274 return ERR_PTR(-ENOMEM);
5275 map->num_stripes = ctl->num_stripes;
5277 for (i = 0; i < ctl->ndevs; ++i) {
5278 for (j = 0; j < ctl->dev_stripes; ++j) {
5279 int s = i * ctl->dev_stripes + j;
5280 map->stripes[s].dev = devices_info[i].dev;
5281 map->stripes[s].physical = devices_info[i].dev_offset +
5282 j * ctl->stripe_size;
5285 map->stripe_len = BTRFS_STRIPE_LEN;
5286 map->io_align = BTRFS_STRIPE_LEN;
5287 map->io_width = BTRFS_STRIPE_LEN;
5289 map->sub_stripes = ctl->sub_stripes;
5291 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5293 em = alloc_extent_map();
5296 return ERR_PTR(-ENOMEM);
5298 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5299 em->map_lookup = map;
5301 em->len = ctl->chunk_size;
5302 em->block_start = 0;
5303 em->block_len = em->len;
5304 em->orig_block_len = ctl->stripe_size;
5306 em_tree = &info->mapping_tree;
5307 write_lock(&em_tree->lock);
5308 ret = add_extent_mapping(em_tree, em, 0);
5310 write_unlock(&em_tree->lock);
5311 free_extent_map(em);
5312 return ERR_PTR(ret);
5314 write_unlock(&em_tree->lock);
5316 block_group = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size);
5317 if (IS_ERR(block_group))
5318 goto error_del_extent;
5320 for (i = 0; i < map->num_stripes; i++) {
5321 struct btrfs_device *dev = map->stripes[i].dev;
5323 btrfs_device_set_bytes_used(dev,
5324 dev->bytes_used + ctl->stripe_size);
5325 if (list_empty(&dev->post_commit_list))
5326 list_add_tail(&dev->post_commit_list,
5327 &trans->transaction->dev_update_list);
5330 atomic64_sub(ctl->stripe_size * map->num_stripes,
5331 &info->free_chunk_space);
5333 free_extent_map(em);
5334 check_raid56_incompat_flag(info, type);
5335 check_raid1c34_incompat_flag(info, type);
5340 write_lock(&em_tree->lock);
5341 remove_extent_mapping(em_tree, em);
5342 write_unlock(&em_tree->lock);
5344 /* One for our allocation */
5345 free_extent_map(em);
5346 /* One for the tree reference */
5347 free_extent_map(em);
5352 struct btrfs_block_group *btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5355 struct btrfs_fs_info *info = trans->fs_info;
5356 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5357 struct btrfs_device_info *devices_info = NULL;
5358 struct alloc_chunk_ctl ctl;
5359 struct btrfs_block_group *block_group;
5362 lockdep_assert_held(&info->chunk_mutex);
5364 if (!alloc_profile_is_valid(type, 0)) {
5366 return ERR_PTR(-EINVAL);
5369 if (list_empty(&fs_devices->alloc_list)) {
5370 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5371 btrfs_debug(info, "%s: no writable device", __func__);
5372 return ERR_PTR(-ENOSPC);
5375 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5376 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5378 return ERR_PTR(-EINVAL);
5381 ctl.start = find_next_chunk(info);
5383 init_alloc_chunk_ctl(fs_devices, &ctl);
5385 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5388 return ERR_PTR(-ENOMEM);
5390 ret = gather_device_info(fs_devices, &ctl, devices_info);
5392 block_group = ERR_PTR(ret);
5396 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5398 block_group = ERR_PTR(ret);
5402 block_group = create_chunk(trans, &ctl, devices_info);
5405 kfree(devices_info);
5410 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5411 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5414 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5417 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
5418 struct btrfs_block_group *bg)
5420 struct btrfs_fs_info *fs_info = trans->fs_info;
5421 struct btrfs_root *extent_root = fs_info->extent_root;
5422 struct btrfs_root *chunk_root = fs_info->chunk_root;
5423 struct btrfs_key key;
5424 struct btrfs_chunk *chunk;
5425 struct btrfs_stripe *stripe;
5426 struct extent_map *em;
5427 struct map_lookup *map;
5433 * We take the chunk_mutex for 2 reasons:
5435 * 1) Updates and insertions in the chunk btree must be done while holding
5436 * the chunk_mutex, as well as updating the system chunk array in the
5437 * superblock. See the comment on top of btrfs_chunk_alloc() for the
5440 * 2) To prevent races with the final phase of a device replace operation
5441 * that replaces the device object associated with the map's stripes,
5442 * because the device object's id can change at any time during that
5443 * final phase of the device replace operation
5444 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5445 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5446 * which would cause a failure when updating the device item, which does
5447 * not exists, or persisting a stripe of the chunk item with such ID.
5448 * Here we can't use the device_list_mutex because our caller already
5449 * has locked the chunk_mutex, and the final phase of device replace
5450 * acquires both mutexes - first the device_list_mutex and then the
5451 * chunk_mutex. Using any of those two mutexes protects us from a
5452 * concurrent device replace.
5454 lockdep_assert_held(&fs_info->chunk_mutex);
5456 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
5459 btrfs_abort_transaction(trans, ret);
5463 map = em->map_lookup;
5464 item_size = btrfs_chunk_item_size(map->num_stripes);
5466 chunk = kzalloc(item_size, GFP_NOFS);
5469 btrfs_abort_transaction(trans, ret);
5473 for (i = 0; i < map->num_stripes; i++) {
5474 struct btrfs_device *device = map->stripes[i].dev;
5476 ret = btrfs_update_device(trans, device);
5481 stripe = &chunk->stripe;
5482 for (i = 0; i < map->num_stripes; i++) {
5483 struct btrfs_device *device = map->stripes[i].dev;
5484 const u64 dev_offset = map->stripes[i].physical;
5486 btrfs_set_stack_stripe_devid(stripe, device->devid);
5487 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5488 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5492 btrfs_set_stack_chunk_length(chunk, bg->length);
5493 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5494 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5495 btrfs_set_stack_chunk_type(chunk, map->type);
5496 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5497 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5498 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5499 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5500 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5502 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5503 key.type = BTRFS_CHUNK_ITEM_KEY;
5504 key.offset = bg->start;
5506 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5510 bg->chunk_item_inserted = 1;
5512 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5513 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5520 free_extent_map(em);
5524 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5526 struct btrfs_fs_info *fs_info = trans->fs_info;
5528 struct btrfs_block_group *meta_bg;
5529 struct btrfs_block_group *sys_bg;
5532 * When adding a new device for sprouting, the seed device is read-only
5533 * so we must first allocate a metadata and a system chunk. But before
5534 * adding the block group items to the extent, device and chunk btrees,
5537 * 1) Create both chunks without doing any changes to the btrees, as
5538 * otherwise we would get -ENOSPC since the block groups from the
5539 * seed device are read-only;
5541 * 2) Add the device item for the new sprout device - finishing the setup
5542 * of a new block group requires updating the device item in the chunk
5543 * btree, so it must exist when we attempt to do it. The previous step
5544 * ensures this does not fail with -ENOSPC.
5546 * After that we can add the block group items to their btrees:
5547 * update existing device item in the chunk btree, add a new block group
5548 * item to the extent btree, add a new chunk item to the chunk btree and
5549 * finally add the new device extent items to the devices btree.
5552 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5553 meta_bg = btrfs_alloc_chunk(trans, alloc_profile);
5554 if (IS_ERR(meta_bg))
5555 return PTR_ERR(meta_bg);
5557 alloc_profile = btrfs_system_alloc_profile(fs_info);
5558 sys_bg = btrfs_alloc_chunk(trans, alloc_profile);
5560 return PTR_ERR(sys_bg);
5565 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5567 const int index = btrfs_bg_flags_to_raid_index(map->type);
5569 return btrfs_raid_array[index].tolerated_failures;
5572 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5574 struct extent_map *em;
5575 struct map_lookup *map;
5580 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5584 map = em->map_lookup;
5585 for (i = 0; i < map->num_stripes; i++) {
5586 if (test_bit(BTRFS_DEV_STATE_MISSING,
5587 &map->stripes[i].dev->dev_state)) {
5591 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5592 &map->stripes[i].dev->dev_state)) {
5599 * If the number of missing devices is larger than max errors,
5600 * we can not write the data into that chunk successfully, so
5603 if (miss_ndevs > btrfs_chunk_max_errors(map))
5606 free_extent_map(em);
5610 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5612 struct extent_map *em;
5615 write_lock(&tree->lock);
5616 em = lookup_extent_mapping(tree, 0, (u64)-1);
5618 remove_extent_mapping(tree, em);
5619 write_unlock(&tree->lock);
5623 free_extent_map(em);
5624 /* once for the tree */
5625 free_extent_map(em);
5629 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5631 struct extent_map *em;
5632 struct map_lookup *map;
5635 em = btrfs_get_chunk_map(fs_info, logical, len);
5638 * We could return errors for these cases, but that could get
5639 * ugly and we'd probably do the same thing which is just not do
5640 * anything else and exit, so return 1 so the callers don't try
5641 * to use other copies.
5645 map = em->map_lookup;
5646 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK))
5647 ret = map->num_stripes;
5648 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5649 ret = map->sub_stripes;
5650 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5652 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5654 * There could be two corrupted data stripes, we need
5655 * to loop retry in order to rebuild the correct data.
5657 * Fail a stripe at a time on every retry except the
5658 * stripe under reconstruction.
5660 ret = map->num_stripes;
5663 free_extent_map(em);
5665 down_read(&fs_info->dev_replace.rwsem);
5666 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5667 fs_info->dev_replace.tgtdev)
5669 up_read(&fs_info->dev_replace.rwsem);
5674 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5677 struct extent_map *em;
5678 struct map_lookup *map;
5679 unsigned long len = fs_info->sectorsize;
5681 em = btrfs_get_chunk_map(fs_info, logical, len);
5683 if (!WARN_ON(IS_ERR(em))) {
5684 map = em->map_lookup;
5685 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5686 len = map->stripe_len * nr_data_stripes(map);
5687 free_extent_map(em);
5692 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5694 struct extent_map *em;
5695 struct map_lookup *map;
5698 em = btrfs_get_chunk_map(fs_info, logical, len);
5700 if(!WARN_ON(IS_ERR(em))) {
5701 map = em->map_lookup;
5702 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5704 free_extent_map(em);
5709 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5710 struct map_lookup *map, int first,
5711 int dev_replace_is_ongoing)
5715 int preferred_mirror;
5717 struct btrfs_device *srcdev;
5720 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5722 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5723 num_stripes = map->sub_stripes;
5725 num_stripes = map->num_stripes;
5727 switch (fs_info->fs_devices->read_policy) {
5729 /* Shouldn't happen, just warn and use pid instead of failing */
5730 btrfs_warn_rl(fs_info,
5731 "unknown read_policy type %u, reset to pid",
5732 fs_info->fs_devices->read_policy);
5733 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID;
5735 case BTRFS_READ_POLICY_PID:
5736 preferred_mirror = first + (current->pid % num_stripes);
5740 if (dev_replace_is_ongoing &&
5741 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5742 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5743 srcdev = fs_info->dev_replace.srcdev;
5748 * try to avoid the drive that is the source drive for a
5749 * dev-replace procedure, only choose it if no other non-missing
5750 * mirror is available
5752 for (tolerance = 0; tolerance < 2; tolerance++) {
5753 if (map->stripes[preferred_mirror].dev->bdev &&
5754 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5755 return preferred_mirror;
5756 for (i = first; i < first + num_stripes; i++) {
5757 if (map->stripes[i].dev->bdev &&
5758 (tolerance || map->stripes[i].dev != srcdev))
5763 /* we couldn't find one that doesn't fail. Just return something
5764 * and the io error handling code will clean up eventually
5766 return preferred_mirror;
5769 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5770 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5777 for (i = 0; i < num_stripes - 1; i++) {
5778 /* Swap if parity is on a smaller index */
5779 if (bbio->raid_map[i] > bbio->raid_map[i + 1]) {
5780 swap(bbio->stripes[i], bbio->stripes[i + 1]);
5781 swap(bbio->raid_map[i], bbio->raid_map[i + 1]);
5788 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5790 struct btrfs_bio *bbio = kzalloc(
5791 /* the size of the btrfs_bio */
5792 sizeof(struct btrfs_bio) +
5793 /* plus the variable array for the stripes */
5794 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5795 /* plus the variable array for the tgt dev */
5796 sizeof(int) * (real_stripes) +
5798 * plus the raid_map, which includes both the tgt dev
5801 sizeof(u64) * (total_stripes),
5802 GFP_NOFS|__GFP_NOFAIL);
5804 atomic_set(&bbio->error, 0);
5805 refcount_set(&bbio->refs, 1);
5807 bbio->tgtdev_map = (int *)(bbio->stripes + total_stripes);
5808 bbio->raid_map = (u64 *)(bbio->tgtdev_map + real_stripes);
5813 void btrfs_get_bbio(struct btrfs_bio *bbio)
5815 WARN_ON(!refcount_read(&bbio->refs));
5816 refcount_inc(&bbio->refs);
5819 void btrfs_put_bbio(struct btrfs_bio *bbio)
5823 if (refcount_dec_and_test(&bbio->refs))
5827 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5829 * Please note that, discard won't be sent to target device of device
5832 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5833 u64 logical, u64 *length_ret,
5834 struct btrfs_bio **bbio_ret)
5836 struct extent_map *em;
5837 struct map_lookup *map;
5838 struct btrfs_bio *bbio;
5839 u64 length = *length_ret;
5843 u64 stripe_end_offset;
5850 u32 sub_stripes = 0;
5851 u64 stripes_per_dev = 0;
5852 u32 remaining_stripes = 0;
5853 u32 last_stripe = 0;
5857 /* discard always return a bbio */
5860 em = btrfs_get_chunk_map(fs_info, logical, length);
5864 map = em->map_lookup;
5865 /* we don't discard raid56 yet */
5866 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5871 offset = logical - em->start;
5872 length = min_t(u64, em->start + em->len - logical, length);
5873 *length_ret = length;
5875 stripe_len = map->stripe_len;
5877 * stripe_nr counts the total number of stripes we have to stride
5878 * to get to this block
5880 stripe_nr = div64_u64(offset, stripe_len);
5882 /* stripe_offset is the offset of this block in its stripe */
5883 stripe_offset = offset - stripe_nr * stripe_len;
5885 stripe_nr_end = round_up(offset + length, map->stripe_len);
5886 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5887 stripe_cnt = stripe_nr_end - stripe_nr;
5888 stripe_end_offset = stripe_nr_end * map->stripe_len -
5891 * after this, stripe_nr is the number of stripes on this
5892 * device we have to walk to find the data, and stripe_index is
5893 * the number of our device in the stripe array
5897 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5898 BTRFS_BLOCK_GROUP_RAID10)) {
5899 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5902 sub_stripes = map->sub_stripes;
5904 factor = map->num_stripes / sub_stripes;
5905 num_stripes = min_t(u64, map->num_stripes,
5906 sub_stripes * stripe_cnt);
5907 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5908 stripe_index *= sub_stripes;
5909 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5910 &remaining_stripes);
5911 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5912 last_stripe *= sub_stripes;
5913 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
5914 BTRFS_BLOCK_GROUP_DUP)) {
5915 num_stripes = map->num_stripes;
5917 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5921 bbio = alloc_btrfs_bio(num_stripes, 0);
5927 for (i = 0; i < num_stripes; i++) {
5928 bbio->stripes[i].physical =
5929 map->stripes[stripe_index].physical +
5930 stripe_offset + stripe_nr * map->stripe_len;
5931 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5933 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5934 BTRFS_BLOCK_GROUP_RAID10)) {
5935 bbio->stripes[i].length = stripes_per_dev *
5938 if (i / sub_stripes < remaining_stripes)
5939 bbio->stripes[i].length +=
5943 * Special for the first stripe and
5946 * |-------|...|-------|
5950 if (i < sub_stripes)
5951 bbio->stripes[i].length -=
5954 if (stripe_index >= last_stripe &&
5955 stripe_index <= (last_stripe +
5957 bbio->stripes[i].length -=
5960 if (i == sub_stripes - 1)
5963 bbio->stripes[i].length = length;
5967 if (stripe_index == map->num_stripes) {
5974 bbio->map_type = map->type;
5975 bbio->num_stripes = num_stripes;
5977 free_extent_map(em);
5982 * In dev-replace case, for repair case (that's the only case where the mirror
5983 * is selected explicitly when calling btrfs_map_block), blocks left of the
5984 * left cursor can also be read from the target drive.
5986 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5988 * For READ, it also needs to be supported using the same mirror number.
5990 * If the requested block is not left of the left cursor, EIO is returned. This
5991 * can happen because btrfs_num_copies() returns one more in the dev-replace
5994 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5995 u64 logical, u64 length,
5996 u64 srcdev_devid, int *mirror_num,
5999 struct btrfs_bio *bbio = NULL;
6001 int index_srcdev = 0;
6003 u64 physical_of_found = 0;
6007 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
6008 logical, &length, &bbio, 0, 0);
6010 ASSERT(bbio == NULL);
6014 num_stripes = bbio->num_stripes;
6015 if (*mirror_num > num_stripes) {
6017 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
6018 * that means that the requested area is not left of the left
6021 btrfs_put_bbio(bbio);
6026 * process the rest of the function using the mirror_num of the source
6027 * drive. Therefore look it up first. At the end, patch the device
6028 * pointer to the one of the target drive.
6030 for (i = 0; i < num_stripes; i++) {
6031 if (bbio->stripes[i].dev->devid != srcdev_devid)
6035 * In case of DUP, in order to keep it simple, only add the
6036 * mirror with the lowest physical address
6039 physical_of_found <= bbio->stripes[i].physical)
6044 physical_of_found = bbio->stripes[i].physical;
6047 btrfs_put_bbio(bbio);
6053 *mirror_num = index_srcdev + 1;
6054 *physical = physical_of_found;
6058 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)
6060 struct btrfs_block_group *cache;
6063 /* Non zoned filesystem does not use "to_copy" flag */
6064 if (!btrfs_is_zoned(fs_info))
6067 cache = btrfs_lookup_block_group(fs_info, logical);
6069 spin_lock(&cache->lock);
6070 ret = cache->to_copy;
6071 spin_unlock(&cache->lock);
6073 btrfs_put_block_group(cache);
6077 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
6078 struct btrfs_bio **bbio_ret,
6079 struct btrfs_dev_replace *dev_replace,
6081 int *num_stripes_ret, int *max_errors_ret)
6083 struct btrfs_bio *bbio = *bbio_ret;
6084 u64 srcdev_devid = dev_replace->srcdev->devid;
6085 int tgtdev_indexes = 0;
6086 int num_stripes = *num_stripes_ret;
6087 int max_errors = *max_errors_ret;
6090 if (op == BTRFS_MAP_WRITE) {
6091 int index_where_to_add;
6094 * A block group which have "to_copy" set will eventually
6095 * copied by dev-replace process. We can avoid cloning IO here.
6097 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))
6101 * duplicate the write operations while the dev replace
6102 * procedure is running. Since the copying of the old disk to
6103 * the new disk takes place at run time while the filesystem is
6104 * mounted writable, the regular write operations to the old
6105 * disk have to be duplicated to go to the new disk as well.
6107 * Note that device->missing is handled by the caller, and that
6108 * the write to the old disk is already set up in the stripes
6111 index_where_to_add = num_stripes;
6112 for (i = 0; i < num_stripes; i++) {
6113 if (bbio->stripes[i].dev->devid == srcdev_devid) {
6114 /* write to new disk, too */
6115 struct btrfs_bio_stripe *new =
6116 bbio->stripes + index_where_to_add;
6117 struct btrfs_bio_stripe *old =
6120 new->physical = old->physical;
6121 new->length = old->length;
6122 new->dev = dev_replace->tgtdev;
6123 bbio->tgtdev_map[i] = index_where_to_add;
6124 index_where_to_add++;
6129 num_stripes = index_where_to_add;
6130 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
6131 int index_srcdev = 0;
6133 u64 physical_of_found = 0;
6136 * During the dev-replace procedure, the target drive can also
6137 * be used to read data in case it is needed to repair a corrupt
6138 * block elsewhere. This is possible if the requested area is
6139 * left of the left cursor. In this area, the target drive is a
6140 * full copy of the source drive.
6142 for (i = 0; i < num_stripes; i++) {
6143 if (bbio->stripes[i].dev->devid == srcdev_devid) {
6145 * In case of DUP, in order to keep it simple,
6146 * only add the mirror with the lowest physical
6150 physical_of_found <=
6151 bbio->stripes[i].physical)
6155 physical_of_found = bbio->stripes[i].physical;
6159 struct btrfs_bio_stripe *tgtdev_stripe =
6160 bbio->stripes + num_stripes;
6162 tgtdev_stripe->physical = physical_of_found;
6163 tgtdev_stripe->length =
6164 bbio->stripes[index_srcdev].length;
6165 tgtdev_stripe->dev = dev_replace->tgtdev;
6166 bbio->tgtdev_map[index_srcdev] = num_stripes;
6173 *num_stripes_ret = num_stripes;
6174 *max_errors_ret = max_errors;
6175 bbio->num_tgtdevs = tgtdev_indexes;
6179 static bool need_full_stripe(enum btrfs_map_op op)
6181 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
6185 * Calculate the geometry of a particular (address, len) tuple. This
6186 * information is used to calculate how big a particular bio can get before it
6187 * straddles a stripe.
6189 * @fs_info: the filesystem
6190 * @em: mapping containing the logical extent
6191 * @op: type of operation - write or read
6192 * @logical: address that we want to figure out the geometry of
6193 * @io_geom: pointer used to return values
6195 * Returns < 0 in case a chunk for the given logical address cannot be found,
6196 * usually shouldn't happen unless @logical is corrupted, 0 otherwise.
6198 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *em,
6199 enum btrfs_map_op op, u64 logical,
6200 struct btrfs_io_geometry *io_geom)
6202 struct map_lookup *map;
6208 u64 raid56_full_stripe_start = (u64)-1;
6211 ASSERT(op != BTRFS_MAP_DISCARD);
6213 map = em->map_lookup;
6214 /* Offset of this logical address in the chunk */
6215 offset = logical - em->start;
6216 /* Len of a stripe in a chunk */
6217 stripe_len = map->stripe_len;
6218 /* Stripe where this block falls in */
6219 stripe_nr = div64_u64(offset, stripe_len);
6220 /* Offset of stripe in the chunk */
6221 stripe_offset = stripe_nr * stripe_len;
6222 if (offset < stripe_offset) {
6224 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu",
6225 stripe_offset, offset, em->start, logical, stripe_len);
6229 /* stripe_offset is the offset of this block in its stripe */
6230 stripe_offset = offset - stripe_offset;
6231 data_stripes = nr_data_stripes(map);
6233 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
6234 u64 max_len = stripe_len - stripe_offset;
6237 * In case of raid56, we need to know the stripe aligned start
6239 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6240 unsigned long full_stripe_len = stripe_len * data_stripes;
6241 raid56_full_stripe_start = offset;
6244 * Allow a write of a full stripe, but make sure we
6245 * don't allow straddling of stripes
6247 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
6249 raid56_full_stripe_start *= full_stripe_len;
6252 * For writes to RAID[56], allow a full stripeset across
6253 * all disks. For other RAID types and for RAID[56]
6254 * reads, just allow a single stripe (on a single disk).
6256 if (op == BTRFS_MAP_WRITE) {
6257 max_len = stripe_len * data_stripes -
6258 (offset - raid56_full_stripe_start);
6261 len = min_t(u64, em->len - offset, max_len);
6263 len = em->len - offset;
6267 io_geom->offset = offset;
6268 io_geom->stripe_len = stripe_len;
6269 io_geom->stripe_nr = stripe_nr;
6270 io_geom->stripe_offset = stripe_offset;
6271 io_geom->raid56_stripe_offset = raid56_full_stripe_start;
6276 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
6277 enum btrfs_map_op op,
6278 u64 logical, u64 *length,
6279 struct btrfs_bio **bbio_ret,
6280 int mirror_num, int need_raid_map)
6282 struct extent_map *em;
6283 struct map_lookup *map;
6293 int tgtdev_indexes = 0;
6294 struct btrfs_bio *bbio = NULL;
6295 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6296 int dev_replace_is_ongoing = 0;
6297 int num_alloc_stripes;
6298 int patch_the_first_stripe_for_dev_replace = 0;
6299 u64 physical_to_patch_in_first_stripe = 0;
6300 u64 raid56_full_stripe_start = (u64)-1;
6301 struct btrfs_io_geometry geom;
6304 ASSERT(op != BTRFS_MAP_DISCARD);
6306 em = btrfs_get_chunk_map(fs_info, logical, *length);
6307 ASSERT(!IS_ERR(em));
6309 ret = btrfs_get_io_geometry(fs_info, em, op, logical, &geom);
6313 map = em->map_lookup;
6316 stripe_len = geom.stripe_len;
6317 stripe_nr = geom.stripe_nr;
6318 stripe_offset = geom.stripe_offset;
6319 raid56_full_stripe_start = geom.raid56_stripe_offset;
6320 data_stripes = nr_data_stripes(map);
6322 down_read(&dev_replace->rwsem);
6323 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6325 * Hold the semaphore for read during the whole operation, write is
6326 * requested at commit time but must wait.
6328 if (!dev_replace_is_ongoing)
6329 up_read(&dev_replace->rwsem);
6331 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
6332 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
6333 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
6334 dev_replace->srcdev->devid,
6336 &physical_to_patch_in_first_stripe);
6340 patch_the_first_stripe_for_dev_replace = 1;
6341 } else if (mirror_num > map->num_stripes) {
6347 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6348 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6350 if (!need_full_stripe(op))
6352 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6353 if (need_full_stripe(op))
6354 num_stripes = map->num_stripes;
6355 else if (mirror_num)
6356 stripe_index = mirror_num - 1;
6358 stripe_index = find_live_mirror(fs_info, map, 0,
6359 dev_replace_is_ongoing);
6360 mirror_num = stripe_index + 1;
6363 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6364 if (need_full_stripe(op)) {
6365 num_stripes = map->num_stripes;
6366 } else if (mirror_num) {
6367 stripe_index = mirror_num - 1;
6372 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6373 u32 factor = map->num_stripes / map->sub_stripes;
6375 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6376 stripe_index *= map->sub_stripes;
6378 if (need_full_stripe(op))
6379 num_stripes = map->sub_stripes;
6380 else if (mirror_num)
6381 stripe_index += mirror_num - 1;
6383 int old_stripe_index = stripe_index;
6384 stripe_index = find_live_mirror(fs_info, map,
6386 dev_replace_is_ongoing);
6387 mirror_num = stripe_index - old_stripe_index + 1;
6390 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6391 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6392 /* push stripe_nr back to the start of the full stripe */
6393 stripe_nr = div64_u64(raid56_full_stripe_start,
6394 stripe_len * data_stripes);
6396 /* RAID[56] write or recovery. Return all stripes */
6397 num_stripes = map->num_stripes;
6398 max_errors = nr_parity_stripes(map);
6400 *length = map->stripe_len;
6405 * Mirror #0 or #1 means the original data block.
6406 * Mirror #2 is RAID5 parity block.
6407 * Mirror #3 is RAID6 Q block.
6409 stripe_nr = div_u64_rem(stripe_nr,
6410 data_stripes, &stripe_index);
6412 stripe_index = data_stripes + mirror_num - 2;
6414 /* We distribute the parity blocks across stripes */
6415 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
6417 if (!need_full_stripe(op) && mirror_num <= 1)
6422 * after this, stripe_nr is the number of stripes on this
6423 * device we have to walk to find the data, and stripe_index is
6424 * the number of our device in the stripe array
6426 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6428 mirror_num = stripe_index + 1;
6430 if (stripe_index >= map->num_stripes) {
6432 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6433 stripe_index, map->num_stripes);
6438 num_alloc_stripes = num_stripes;
6439 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
6440 if (op == BTRFS_MAP_WRITE)
6441 num_alloc_stripes <<= 1;
6442 if (op == BTRFS_MAP_GET_READ_MIRRORS)
6443 num_alloc_stripes++;
6444 tgtdev_indexes = num_stripes;
6447 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
6453 for (i = 0; i < num_stripes; i++) {
6454 bbio->stripes[i].physical = map->stripes[stripe_index].physical +
6455 stripe_offset + stripe_nr * map->stripe_len;
6456 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
6460 /* build raid_map */
6461 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6462 (need_full_stripe(op) || mirror_num > 1)) {
6466 /* Work out the disk rotation on this stripe-set */
6467 div_u64_rem(stripe_nr, num_stripes, &rot);
6469 /* Fill in the logical address of each stripe */
6470 tmp = stripe_nr * data_stripes;
6471 for (i = 0; i < data_stripes; i++)
6472 bbio->raid_map[(i+rot) % num_stripes] =
6473 em->start + (tmp + i) * map->stripe_len;
6475 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
6476 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
6477 bbio->raid_map[(i+rot+1) % num_stripes] =
6480 sort_parity_stripes(bbio, num_stripes);
6483 if (need_full_stripe(op))
6484 max_errors = btrfs_chunk_max_errors(map);
6486 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6487 need_full_stripe(op)) {
6488 handle_ops_on_dev_replace(op, &bbio, dev_replace, logical,
6489 &num_stripes, &max_errors);
6493 bbio->map_type = map->type;
6494 bbio->num_stripes = num_stripes;
6495 bbio->max_errors = max_errors;
6496 bbio->mirror_num = mirror_num;
6499 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6500 * mirror_num == num_stripes + 1 && dev_replace target drive is
6501 * available as a mirror
6503 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6504 WARN_ON(num_stripes > 1);
6505 bbio->stripes[0].dev = dev_replace->tgtdev;
6506 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
6507 bbio->mirror_num = map->num_stripes + 1;
6510 if (dev_replace_is_ongoing) {
6511 lockdep_assert_held(&dev_replace->rwsem);
6512 /* Unlock and let waiting writers proceed */
6513 up_read(&dev_replace->rwsem);
6515 free_extent_map(em);
6519 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6520 u64 logical, u64 *length,
6521 struct btrfs_bio **bbio_ret, int mirror_num)
6523 if (op == BTRFS_MAP_DISCARD)
6524 return __btrfs_map_block_for_discard(fs_info, logical,
6527 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
6531 /* For Scrub/replace */
6532 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6533 u64 logical, u64 *length,
6534 struct btrfs_bio **bbio_ret)
6536 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
6539 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6541 bio->bi_private = bbio->private;
6542 bio->bi_end_io = bbio->end_io;
6545 btrfs_put_bbio(bbio);
6548 static void btrfs_end_bio(struct bio *bio)
6550 struct btrfs_bio *bbio = bio->bi_private;
6551 int is_orig_bio = 0;
6553 if (bio->bi_status) {
6554 atomic_inc(&bbio->error);
6555 if (bio->bi_status == BLK_STS_IOERR ||
6556 bio->bi_status == BLK_STS_TARGET) {
6557 struct btrfs_device *dev = btrfs_io_bio(bio)->device;
6560 if (btrfs_op(bio) == BTRFS_MAP_WRITE)
6561 btrfs_dev_stat_inc_and_print(dev,
6562 BTRFS_DEV_STAT_WRITE_ERRS);
6563 else if (!(bio->bi_opf & REQ_RAHEAD))
6564 btrfs_dev_stat_inc_and_print(dev,
6565 BTRFS_DEV_STAT_READ_ERRS);
6566 if (bio->bi_opf & REQ_PREFLUSH)
6567 btrfs_dev_stat_inc_and_print(dev,
6568 BTRFS_DEV_STAT_FLUSH_ERRS);
6572 if (bio == bbio->orig_bio)
6575 btrfs_bio_counter_dec(bbio->fs_info);
6577 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6580 bio = bbio->orig_bio;
6583 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6584 /* only send an error to the higher layers if it is
6585 * beyond the tolerance of the btrfs bio
6587 if (atomic_read(&bbio->error) > bbio->max_errors) {
6588 bio->bi_status = BLK_STS_IOERR;
6591 * this bio is actually up to date, we didn't
6592 * go over the max number of errors
6594 bio->bi_status = BLK_STS_OK;
6597 btrfs_end_bbio(bbio, bio);
6598 } else if (!is_orig_bio) {
6603 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6604 u64 physical, struct btrfs_device *dev)
6606 struct btrfs_fs_info *fs_info = bbio->fs_info;
6608 bio->bi_private = bbio;
6609 btrfs_io_bio(bio)->device = dev;
6610 bio->bi_end_io = btrfs_end_bio;
6611 bio->bi_iter.bi_sector = physical >> 9;
6613 * For zone append writing, bi_sector must point the beginning of the
6616 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
6617 if (btrfs_dev_is_sequential(dev, physical)) {
6618 u64 zone_start = round_down(physical, fs_info->zone_size);
6620 bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT;
6622 bio->bi_opf &= ~REQ_OP_ZONE_APPEND;
6623 bio->bi_opf |= REQ_OP_WRITE;
6626 btrfs_debug_in_rcu(fs_info,
6627 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6628 bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector,
6629 (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name),
6630 dev->devid, bio->bi_iter.bi_size);
6631 bio_set_dev(bio, dev->bdev);
6633 btrfs_bio_counter_inc_noblocked(fs_info);
6635 btrfsic_submit_bio(bio);
6638 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6640 atomic_inc(&bbio->error);
6641 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6642 /* Should be the original bio. */
6643 WARN_ON(bio != bbio->orig_bio);
6645 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6646 bio->bi_iter.bi_sector = logical >> 9;
6647 if (atomic_read(&bbio->error) > bbio->max_errors)
6648 bio->bi_status = BLK_STS_IOERR;
6650 bio->bi_status = BLK_STS_OK;
6651 btrfs_end_bbio(bbio, bio);
6655 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6658 struct btrfs_device *dev;
6659 struct bio *first_bio = bio;
6660 u64 logical = bio->bi_iter.bi_sector << 9;
6666 struct btrfs_bio *bbio = NULL;
6668 length = bio->bi_iter.bi_size;
6669 map_length = length;
6671 btrfs_bio_counter_inc_blocked(fs_info);
6672 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6673 &map_length, &bbio, mirror_num, 1);
6675 btrfs_bio_counter_dec(fs_info);
6676 return errno_to_blk_status(ret);
6679 total_devs = bbio->num_stripes;
6680 bbio->orig_bio = first_bio;
6681 bbio->private = first_bio->bi_private;
6682 bbio->end_io = first_bio->bi_end_io;
6683 bbio->fs_info = fs_info;
6684 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6686 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6687 ((btrfs_op(bio) == BTRFS_MAP_WRITE) || (mirror_num > 1))) {
6688 /* In this case, map_length has been set to the length of
6689 a single stripe; not the whole write */
6690 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
6691 ret = raid56_parity_write(fs_info, bio, bbio,
6694 ret = raid56_parity_recover(fs_info, bio, bbio,
6695 map_length, mirror_num, 1);
6698 btrfs_bio_counter_dec(fs_info);
6699 return errno_to_blk_status(ret);
6702 if (map_length < length) {
6704 "mapping failed logical %llu bio len %llu len %llu",
6705 logical, length, map_length);
6709 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6710 dev = bbio->stripes[dev_nr].dev;
6711 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
6713 (btrfs_op(first_bio) == BTRFS_MAP_WRITE &&
6714 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6715 bbio_error(bbio, first_bio, logical);
6719 if (dev_nr < total_devs - 1)
6720 bio = btrfs_bio_clone(first_bio);
6724 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, dev);
6726 btrfs_bio_counter_dec(fs_info);
6731 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6734 * If devid and uuid are both specified, the match must be exact, otherwise
6735 * only devid is used.
6737 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
6738 u64 devid, u8 *uuid, u8 *fsid)
6740 struct btrfs_device *device;
6741 struct btrfs_fs_devices *seed_devs;
6743 if (!fsid || !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6744 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6745 if (device->devid == devid &&
6746 (!uuid || memcmp(device->uuid, uuid,
6747 BTRFS_UUID_SIZE) == 0))
6752 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6754 !memcmp(seed_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6755 list_for_each_entry(device, &seed_devs->devices,
6757 if (device->devid == devid &&
6758 (!uuid || memcmp(device->uuid, uuid,
6759 BTRFS_UUID_SIZE) == 0))
6768 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6769 u64 devid, u8 *dev_uuid)
6771 struct btrfs_device *device;
6772 unsigned int nofs_flag;
6775 * We call this under the chunk_mutex, so we want to use NOFS for this
6776 * allocation, however we don't want to change btrfs_alloc_device() to
6777 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6780 nofs_flag = memalloc_nofs_save();
6781 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6782 memalloc_nofs_restore(nofs_flag);
6786 list_add(&device->dev_list, &fs_devices->devices);
6787 device->fs_devices = fs_devices;
6788 fs_devices->num_devices++;
6790 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6791 fs_devices->missing_devices++;
6797 * btrfs_alloc_device - allocate struct btrfs_device
6798 * @fs_info: used only for generating a new devid, can be NULL if
6799 * devid is provided (i.e. @devid != NULL).
6800 * @devid: a pointer to devid for this device. If NULL a new devid
6802 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6805 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6806 * on error. Returned struct is not linked onto any lists and must be
6807 * destroyed with btrfs_free_device.
6809 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6813 struct btrfs_device *dev;
6816 if (WARN_ON(!devid && !fs_info))
6817 return ERR_PTR(-EINVAL);
6819 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
6821 return ERR_PTR(-ENOMEM);
6824 * Preallocate a bio that's always going to be used for flushing device
6825 * barriers and matches the device lifespan
6827 dev->flush_bio = bio_kmalloc(GFP_KERNEL, 0);
6828 if (!dev->flush_bio) {
6830 return ERR_PTR(-ENOMEM);
6833 INIT_LIST_HEAD(&dev->dev_list);
6834 INIT_LIST_HEAD(&dev->dev_alloc_list);
6835 INIT_LIST_HEAD(&dev->post_commit_list);
6837 atomic_set(&dev->reada_in_flight, 0);
6838 atomic_set(&dev->dev_stats_ccnt, 0);
6839 btrfs_device_data_ordered_init(dev);
6840 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
6841 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
6842 extent_io_tree_init(fs_info, &dev->alloc_state,
6843 IO_TREE_DEVICE_ALLOC_STATE, NULL);
6850 ret = find_next_devid(fs_info, &tmp);
6852 btrfs_free_device(dev);
6853 return ERR_PTR(ret);
6859 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6861 generate_random_uuid(dev->uuid);
6866 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6867 u64 devid, u8 *uuid, bool error)
6870 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6873 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6877 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes)
6879 const int data_stripes = calc_data_stripes(type, num_stripes);
6881 return div_u64(chunk_len, data_stripes);
6884 #if BITS_PER_LONG == 32
6886 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
6887 * can't be accessed on 32bit systems.
6889 * This function do mount time check to reject the fs if it already has
6890 * metadata chunk beyond that limit.
6892 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6893 u64 logical, u64 length, u64 type)
6895 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6898 if (logical + length < MAX_LFS_FILESIZE)
6901 btrfs_err_32bit_limit(fs_info);
6906 * This is to give early warning for any metadata chunk reaching
6907 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
6908 * Although we can still access the metadata, it's not going to be possible
6909 * once the limit is reached.
6911 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6912 u64 logical, u64 length, u64 type)
6914 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6917 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)
6920 btrfs_warn_32bit_limit(fs_info);
6924 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6925 struct btrfs_chunk *chunk)
6927 struct btrfs_fs_info *fs_info = leaf->fs_info;
6928 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6929 struct map_lookup *map;
6930 struct extent_map *em;
6935 u8 uuid[BTRFS_UUID_SIZE];
6940 logical = key->offset;
6941 length = btrfs_chunk_length(leaf, chunk);
6942 type = btrfs_chunk_type(leaf, chunk);
6943 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6945 #if BITS_PER_LONG == 32
6946 ret = check_32bit_meta_chunk(fs_info, logical, length, type);
6949 warn_32bit_meta_chunk(fs_info, logical, length, type);
6953 * Only need to verify chunk item if we're reading from sys chunk array,
6954 * as chunk item in tree block is already verified by tree-checker.
6956 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6957 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6962 read_lock(&map_tree->lock);
6963 em = lookup_extent_mapping(map_tree, logical, 1);
6964 read_unlock(&map_tree->lock);
6966 /* already mapped? */
6967 if (em && em->start <= logical && em->start + em->len > logical) {
6968 free_extent_map(em);
6971 free_extent_map(em);
6974 em = alloc_extent_map();
6977 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6979 free_extent_map(em);
6983 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6984 em->map_lookup = map;
6985 em->start = logical;
6988 em->block_start = 0;
6989 em->block_len = em->len;
6991 map->num_stripes = num_stripes;
6992 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6993 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6994 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6996 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6997 map->verified_stripes = 0;
6998 em->orig_block_len = calc_stripe_length(type, em->len,
7000 for (i = 0; i < num_stripes; i++) {
7001 map->stripes[i].physical =
7002 btrfs_stripe_offset_nr(leaf, chunk, i);
7003 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
7004 read_extent_buffer(leaf, uuid, (unsigned long)
7005 btrfs_stripe_dev_uuid_nr(chunk, i),
7007 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices,
7009 if (!map->stripes[i].dev &&
7010 !btrfs_test_opt(fs_info, DEGRADED)) {
7011 free_extent_map(em);
7012 btrfs_report_missing_device(fs_info, devid, uuid, true);
7015 if (!map->stripes[i].dev) {
7016 map->stripes[i].dev =
7017 add_missing_dev(fs_info->fs_devices, devid,
7019 if (IS_ERR(map->stripes[i].dev)) {
7020 free_extent_map(em);
7022 "failed to init missing dev %llu: %ld",
7023 devid, PTR_ERR(map->stripes[i].dev));
7024 return PTR_ERR(map->stripes[i].dev);
7026 btrfs_report_missing_device(fs_info, devid, uuid, false);
7028 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
7029 &(map->stripes[i].dev->dev_state));
7033 write_lock(&map_tree->lock);
7034 ret = add_extent_mapping(map_tree, em, 0);
7035 write_unlock(&map_tree->lock);
7038 "failed to add chunk map, start=%llu len=%llu: %d",
7039 em->start, em->len, ret);
7041 free_extent_map(em);
7046 static void fill_device_from_item(struct extent_buffer *leaf,
7047 struct btrfs_dev_item *dev_item,
7048 struct btrfs_device *device)
7052 device->devid = btrfs_device_id(leaf, dev_item);
7053 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
7054 device->total_bytes = device->disk_total_bytes;
7055 device->commit_total_bytes = device->disk_total_bytes;
7056 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
7057 device->commit_bytes_used = device->bytes_used;
7058 device->type = btrfs_device_type(leaf, dev_item);
7059 device->io_align = btrfs_device_io_align(leaf, dev_item);
7060 device->io_width = btrfs_device_io_width(leaf, dev_item);
7061 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
7062 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
7063 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
7065 ptr = btrfs_device_uuid(dev_item);
7066 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
7069 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
7072 struct btrfs_fs_devices *fs_devices;
7075 lockdep_assert_held(&uuid_mutex);
7078 /* This will match only for multi-device seed fs */
7079 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
7080 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
7084 fs_devices = find_fsid(fsid, NULL);
7086 if (!btrfs_test_opt(fs_info, DEGRADED))
7087 return ERR_PTR(-ENOENT);
7089 fs_devices = alloc_fs_devices(fsid, NULL);
7090 if (IS_ERR(fs_devices))
7093 fs_devices->seeding = true;
7094 fs_devices->opened = 1;
7099 * Upon first call for a seed fs fsid, just create a private copy of the
7100 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
7102 fs_devices = clone_fs_devices(fs_devices);
7103 if (IS_ERR(fs_devices))
7106 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
7108 free_fs_devices(fs_devices);
7109 return ERR_PTR(ret);
7112 if (!fs_devices->seeding) {
7113 close_fs_devices(fs_devices);
7114 free_fs_devices(fs_devices);
7115 return ERR_PTR(-EINVAL);
7118 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
7123 static int read_one_dev(struct extent_buffer *leaf,
7124 struct btrfs_dev_item *dev_item)
7126 struct btrfs_fs_info *fs_info = leaf->fs_info;
7127 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7128 struct btrfs_device *device;
7131 u8 fs_uuid[BTRFS_FSID_SIZE];
7132 u8 dev_uuid[BTRFS_UUID_SIZE];
7134 devid = btrfs_device_id(leaf, dev_item);
7135 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
7137 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
7140 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
7141 fs_devices = open_seed_devices(fs_info, fs_uuid);
7142 if (IS_ERR(fs_devices))
7143 return PTR_ERR(fs_devices);
7146 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
7149 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7150 btrfs_report_missing_device(fs_info, devid,
7155 device = add_missing_dev(fs_devices, devid, dev_uuid);
7156 if (IS_ERR(device)) {
7158 "failed to add missing dev %llu: %ld",
7159 devid, PTR_ERR(device));
7160 return PTR_ERR(device);
7162 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
7164 if (!device->bdev) {
7165 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7166 btrfs_report_missing_device(fs_info,
7167 devid, dev_uuid, true);
7170 btrfs_report_missing_device(fs_info, devid,
7174 if (!device->bdev &&
7175 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
7177 * this happens when a device that was properly setup
7178 * in the device info lists suddenly goes bad.
7179 * device->bdev is NULL, and so we have to set
7180 * device->missing to one here
7182 device->fs_devices->missing_devices++;
7183 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
7186 /* Move the device to its own fs_devices */
7187 if (device->fs_devices != fs_devices) {
7188 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
7189 &device->dev_state));
7191 list_move(&device->dev_list, &fs_devices->devices);
7192 device->fs_devices->num_devices--;
7193 fs_devices->num_devices++;
7195 device->fs_devices->missing_devices--;
7196 fs_devices->missing_devices++;
7198 device->fs_devices = fs_devices;
7202 if (device->fs_devices != fs_info->fs_devices) {
7203 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
7204 if (device->generation !=
7205 btrfs_device_generation(leaf, dev_item))
7209 fill_device_from_item(leaf, dev_item, device);
7211 u64 max_total_bytes = i_size_read(device->bdev->bd_inode);
7213 if (device->total_bytes > max_total_bytes) {
7215 "device total_bytes should be at most %llu but found %llu",
7216 max_total_bytes, device->total_bytes);
7220 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7221 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7222 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7223 device->fs_devices->total_rw_bytes += device->total_bytes;
7224 atomic64_add(device->total_bytes - device->bytes_used,
7225 &fs_info->free_chunk_space);
7231 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7233 struct btrfs_root *root = fs_info->tree_root;
7234 struct btrfs_super_block *super_copy = fs_info->super_copy;
7235 struct extent_buffer *sb;
7236 struct btrfs_disk_key *disk_key;
7237 struct btrfs_chunk *chunk;
7239 unsigned long sb_array_offset;
7246 struct btrfs_key key;
7248 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7250 * This will create extent buffer of nodesize, superblock size is
7251 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
7252 * overallocate but we can keep it as-is, only the first page is used.
7254 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET,
7255 root->root_key.objectid, 0);
7258 set_extent_buffer_uptodate(sb);
7260 * The sb extent buffer is artificial and just used to read the system array.
7261 * set_extent_buffer_uptodate() call does not properly mark all it's
7262 * pages up-to-date when the page is larger: extent does not cover the
7263 * whole page and consequently check_page_uptodate does not find all
7264 * the page's extents up-to-date (the hole beyond sb),
7265 * write_extent_buffer then triggers a WARN_ON.
7267 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
7268 * but sb spans only this function. Add an explicit SetPageUptodate call
7269 * to silence the warning eg. on PowerPC 64.
7271 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
7272 SetPageUptodate(sb->pages[0]);
7274 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7275 array_size = btrfs_super_sys_array_size(super_copy);
7277 array_ptr = super_copy->sys_chunk_array;
7278 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7281 while (cur_offset < array_size) {
7282 disk_key = (struct btrfs_disk_key *)array_ptr;
7283 len = sizeof(*disk_key);
7284 if (cur_offset + len > array_size)
7285 goto out_short_read;
7287 btrfs_disk_key_to_cpu(&key, disk_key);
7290 sb_array_offset += len;
7293 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7295 "unexpected item type %u in sys_array at offset %u",
7296 (u32)key.type, cur_offset);
7301 chunk = (struct btrfs_chunk *)sb_array_offset;
7303 * At least one btrfs_chunk with one stripe must be present,
7304 * exact stripe count check comes afterwards
7306 len = btrfs_chunk_item_size(1);
7307 if (cur_offset + len > array_size)
7308 goto out_short_read;
7310 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7313 "invalid number of stripes %u in sys_array at offset %u",
7314 num_stripes, cur_offset);
7319 type = btrfs_chunk_type(sb, chunk);
7320 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7322 "invalid chunk type %llu in sys_array at offset %u",
7328 len = btrfs_chunk_item_size(num_stripes);
7329 if (cur_offset + len > array_size)
7330 goto out_short_read;
7332 ret = read_one_chunk(&key, sb, chunk);
7337 sb_array_offset += len;
7340 clear_extent_buffer_uptodate(sb);
7341 free_extent_buffer_stale(sb);
7345 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7347 clear_extent_buffer_uptodate(sb);
7348 free_extent_buffer_stale(sb);
7353 * Check if all chunks in the fs are OK for read-write degraded mount
7355 * If the @failing_dev is specified, it's accounted as missing.
7357 * Return true if all chunks meet the minimal RW mount requirements.
7358 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7360 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7361 struct btrfs_device *failing_dev)
7363 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7364 struct extent_map *em;
7368 read_lock(&map_tree->lock);
7369 em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7370 read_unlock(&map_tree->lock);
7371 /* No chunk at all? Return false anyway */
7377 struct map_lookup *map;
7382 map = em->map_lookup;
7384 btrfs_get_num_tolerated_disk_barrier_failures(
7386 for (i = 0; i < map->num_stripes; i++) {
7387 struct btrfs_device *dev = map->stripes[i].dev;
7389 if (!dev || !dev->bdev ||
7390 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7391 dev->last_flush_error)
7393 else if (failing_dev && failing_dev == dev)
7396 if (missing > max_tolerated) {
7399 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7400 em->start, missing, max_tolerated);
7401 free_extent_map(em);
7405 next_start = extent_map_end(em);
7406 free_extent_map(em);
7408 read_lock(&map_tree->lock);
7409 em = lookup_extent_mapping(map_tree, next_start,
7410 (u64)(-1) - next_start);
7411 read_unlock(&map_tree->lock);
7417 static void readahead_tree_node_children(struct extent_buffer *node)
7420 const int nr_items = btrfs_header_nritems(node);
7422 for (i = 0; i < nr_items; i++)
7423 btrfs_readahead_node_child(node, i);
7426 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7428 struct btrfs_root *root = fs_info->chunk_root;
7429 struct btrfs_path *path;
7430 struct extent_buffer *leaf;
7431 struct btrfs_key key;
7432 struct btrfs_key found_key;
7436 u64 last_ra_node = 0;
7438 path = btrfs_alloc_path();
7443 * uuid_mutex is needed only if we are mounting a sprout FS
7444 * otherwise we don't need it.
7446 mutex_lock(&uuid_mutex);
7449 * It is possible for mount and umount to race in such a way that
7450 * we execute this code path, but open_fs_devices failed to clear
7451 * total_rw_bytes. We certainly want it cleared before reading the
7452 * device items, so clear it here.
7454 fs_info->fs_devices->total_rw_bytes = 0;
7457 * Read all device items, and then all the chunk items. All
7458 * device items are found before any chunk item (their object id
7459 * is smaller than the lowest possible object id for a chunk
7460 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7462 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7465 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7469 struct extent_buffer *node;
7471 leaf = path->nodes[0];
7472 slot = path->slots[0];
7473 if (slot >= btrfs_header_nritems(leaf)) {
7474 ret = btrfs_next_leaf(root, path);
7482 * The nodes on level 1 are not locked but we don't need to do
7483 * that during mount time as nothing else can access the tree
7485 node = path->nodes[1];
7487 if (last_ra_node != node->start) {
7488 readahead_tree_node_children(node);
7489 last_ra_node = node->start;
7492 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7493 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7494 struct btrfs_dev_item *dev_item;
7495 dev_item = btrfs_item_ptr(leaf, slot,
7496 struct btrfs_dev_item);
7497 ret = read_one_dev(leaf, dev_item);
7501 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7502 struct btrfs_chunk *chunk;
7505 * We are only called at mount time, so no need to take
7506 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7507 * we always lock first fs_info->chunk_mutex before
7508 * acquiring any locks on the chunk tree. This is a
7509 * requirement for chunk allocation, see the comment on
7510 * top of btrfs_chunk_alloc() for details.
7512 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
7513 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7514 ret = read_one_chunk(&found_key, leaf, chunk);
7522 * After loading chunk tree, we've got all device information,
7523 * do another round of validation checks.
7525 if (total_dev != fs_info->fs_devices->total_devices) {
7527 "super_num_devices %llu mismatch with num_devices %llu found here",
7528 btrfs_super_num_devices(fs_info->super_copy),
7533 if (btrfs_super_total_bytes(fs_info->super_copy) <
7534 fs_info->fs_devices->total_rw_bytes) {
7536 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7537 btrfs_super_total_bytes(fs_info->super_copy),
7538 fs_info->fs_devices->total_rw_bytes);
7544 mutex_unlock(&uuid_mutex);
7546 btrfs_free_path(path);
7550 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7552 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7553 struct btrfs_device *device;
7555 fs_devices->fs_info = fs_info;
7557 mutex_lock(&fs_devices->device_list_mutex);
7558 list_for_each_entry(device, &fs_devices->devices, dev_list)
7559 device->fs_info = fs_info;
7561 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7562 list_for_each_entry(device, &seed_devs->devices, dev_list)
7563 device->fs_info = fs_info;
7565 seed_devs->fs_info = fs_info;
7567 mutex_unlock(&fs_devices->device_list_mutex);
7570 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7571 const struct btrfs_dev_stats_item *ptr,
7576 read_extent_buffer(eb, &val,
7577 offsetof(struct btrfs_dev_stats_item, values) +
7578 ((unsigned long)ptr) + (index * sizeof(u64)),
7583 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7584 struct btrfs_dev_stats_item *ptr,
7587 write_extent_buffer(eb, &val,
7588 offsetof(struct btrfs_dev_stats_item, values) +
7589 ((unsigned long)ptr) + (index * sizeof(u64)),
7593 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7594 struct btrfs_path *path)
7596 struct btrfs_dev_stats_item *ptr;
7597 struct extent_buffer *eb;
7598 struct btrfs_key key;
7602 if (!device->fs_info->dev_root)
7605 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7606 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7607 key.offset = device->devid;
7608 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7610 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7611 btrfs_dev_stat_set(device, i, 0);
7612 device->dev_stats_valid = 1;
7613 btrfs_release_path(path);
7614 return ret < 0 ? ret : 0;
7616 slot = path->slots[0];
7617 eb = path->nodes[0];
7618 item_size = btrfs_item_size_nr(eb, slot);
7620 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7622 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7623 if (item_size >= (1 + i) * sizeof(__le64))
7624 btrfs_dev_stat_set(device, i,
7625 btrfs_dev_stats_value(eb, ptr, i));
7627 btrfs_dev_stat_set(device, i, 0);
7630 device->dev_stats_valid = 1;
7631 btrfs_dev_stat_print_on_load(device);
7632 btrfs_release_path(path);
7637 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7639 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7640 struct btrfs_device *device;
7641 struct btrfs_path *path = NULL;
7644 path = btrfs_alloc_path();
7648 mutex_lock(&fs_devices->device_list_mutex);
7649 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7650 ret = btrfs_device_init_dev_stats(device, path);
7654 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7655 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7656 ret = btrfs_device_init_dev_stats(device, path);
7662 mutex_unlock(&fs_devices->device_list_mutex);
7664 btrfs_free_path(path);
7668 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7669 struct btrfs_device *device)
7671 struct btrfs_fs_info *fs_info = trans->fs_info;
7672 struct btrfs_root *dev_root = fs_info->dev_root;
7673 struct btrfs_path *path;
7674 struct btrfs_key key;
7675 struct extent_buffer *eb;
7676 struct btrfs_dev_stats_item *ptr;
7680 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7681 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7682 key.offset = device->devid;
7684 path = btrfs_alloc_path();
7687 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7689 btrfs_warn_in_rcu(fs_info,
7690 "error %d while searching for dev_stats item for device %s",
7691 ret, rcu_str_deref(device->name));
7696 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7697 /* need to delete old one and insert a new one */
7698 ret = btrfs_del_item(trans, dev_root, path);
7700 btrfs_warn_in_rcu(fs_info,
7701 "delete too small dev_stats item for device %s failed %d",
7702 rcu_str_deref(device->name), ret);
7709 /* need to insert a new item */
7710 btrfs_release_path(path);
7711 ret = btrfs_insert_empty_item(trans, dev_root, path,
7712 &key, sizeof(*ptr));
7714 btrfs_warn_in_rcu(fs_info,
7715 "insert dev_stats item for device %s failed %d",
7716 rcu_str_deref(device->name), ret);
7721 eb = path->nodes[0];
7722 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7723 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7724 btrfs_set_dev_stats_value(eb, ptr, i,
7725 btrfs_dev_stat_read(device, i));
7726 btrfs_mark_buffer_dirty(eb);
7729 btrfs_free_path(path);
7734 * called from commit_transaction. Writes all changed device stats to disk.
7736 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7738 struct btrfs_fs_info *fs_info = trans->fs_info;
7739 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7740 struct btrfs_device *device;
7744 mutex_lock(&fs_devices->device_list_mutex);
7745 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7746 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7747 if (!device->dev_stats_valid || stats_cnt == 0)
7752 * There is a LOAD-LOAD control dependency between the value of
7753 * dev_stats_ccnt and updating the on-disk values which requires
7754 * reading the in-memory counters. Such control dependencies
7755 * require explicit read memory barriers.
7757 * This memory barriers pairs with smp_mb__before_atomic in
7758 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7759 * barrier implied by atomic_xchg in
7760 * btrfs_dev_stats_read_and_reset
7764 ret = update_dev_stat_item(trans, device);
7766 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7768 mutex_unlock(&fs_devices->device_list_mutex);
7773 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7775 btrfs_dev_stat_inc(dev, index);
7776 btrfs_dev_stat_print_on_error(dev);
7779 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7781 if (!dev->dev_stats_valid)
7783 btrfs_err_rl_in_rcu(dev->fs_info,
7784 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7785 rcu_str_deref(dev->name),
7786 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7787 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7788 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7789 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7790 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7793 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7797 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7798 if (btrfs_dev_stat_read(dev, i) != 0)
7800 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7801 return; /* all values == 0, suppress message */
7803 btrfs_info_in_rcu(dev->fs_info,
7804 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7805 rcu_str_deref(dev->name),
7806 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7807 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7808 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7809 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7810 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7813 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7814 struct btrfs_ioctl_get_dev_stats *stats)
7816 struct btrfs_device *dev;
7817 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7820 mutex_lock(&fs_devices->device_list_mutex);
7821 dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL);
7822 mutex_unlock(&fs_devices->device_list_mutex);
7825 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7827 } else if (!dev->dev_stats_valid) {
7828 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7830 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7831 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7832 if (stats->nr_items > i)
7834 btrfs_dev_stat_read_and_reset(dev, i);
7836 btrfs_dev_stat_set(dev, i, 0);
7838 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7839 current->comm, task_pid_nr(current));
7841 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7842 if (stats->nr_items > i)
7843 stats->values[i] = btrfs_dev_stat_read(dev, i);
7845 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7846 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7851 * Update the size and bytes used for each device where it changed. This is
7852 * delayed since we would otherwise get errors while writing out the
7855 * Must be invoked during transaction commit.
7857 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7859 struct btrfs_device *curr, *next;
7861 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7863 if (list_empty(&trans->dev_update_list))
7867 * We don't need the device_list_mutex here. This list is owned by the
7868 * transaction and the transaction must complete before the device is
7871 mutex_lock(&trans->fs_info->chunk_mutex);
7872 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7874 list_del_init(&curr->post_commit_list);
7875 curr->commit_total_bytes = curr->disk_total_bytes;
7876 curr->commit_bytes_used = curr->bytes_used;
7878 mutex_unlock(&trans->fs_info->chunk_mutex);
7882 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7884 int btrfs_bg_type_to_factor(u64 flags)
7886 const int index = btrfs_bg_flags_to_raid_index(flags);
7888 return btrfs_raid_array[index].ncopies;
7893 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7894 u64 chunk_offset, u64 devid,
7895 u64 physical_offset, u64 physical_len)
7897 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7898 struct extent_map *em;
7899 struct map_lookup *map;
7900 struct btrfs_device *dev;
7906 read_lock(&em_tree->lock);
7907 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7908 read_unlock(&em_tree->lock);
7912 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7913 physical_offset, devid);
7918 map = em->map_lookup;
7919 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes);
7920 if (physical_len != stripe_len) {
7922 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7923 physical_offset, devid, em->start, physical_len,
7929 for (i = 0; i < map->num_stripes; i++) {
7930 if (map->stripes[i].dev->devid == devid &&
7931 map->stripes[i].physical == physical_offset) {
7933 if (map->verified_stripes >= map->num_stripes) {
7935 "too many dev extents for chunk %llu found",
7940 map->verified_stripes++;
7946 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7947 physical_offset, devid);
7951 /* Make sure no dev extent is beyond device boundary */
7952 dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL);
7954 btrfs_err(fs_info, "failed to find devid %llu", devid);
7959 if (physical_offset + physical_len > dev->disk_total_bytes) {
7961 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7962 devid, physical_offset, physical_len,
7963 dev->disk_total_bytes);
7968 if (dev->zone_info) {
7969 u64 zone_size = dev->zone_info->zone_size;
7971 if (!IS_ALIGNED(physical_offset, zone_size) ||
7972 !IS_ALIGNED(physical_len, zone_size)) {
7974 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
7975 devid, physical_offset, physical_len);
7982 free_extent_map(em);
7986 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7988 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7989 struct extent_map *em;
7990 struct rb_node *node;
7993 read_lock(&em_tree->lock);
7994 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
7995 em = rb_entry(node, struct extent_map, rb_node);
7996 if (em->map_lookup->num_stripes !=
7997 em->map_lookup->verified_stripes) {
7999 "chunk %llu has missing dev extent, have %d expect %d",
8000 em->start, em->map_lookup->verified_stripes,
8001 em->map_lookup->num_stripes);
8007 read_unlock(&em_tree->lock);
8012 * Ensure that all dev extents are mapped to correct chunk, otherwise
8013 * later chunk allocation/free would cause unexpected behavior.
8015 * NOTE: This will iterate through the whole device tree, which should be of
8016 * the same size level as the chunk tree. This slightly increases mount time.
8018 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
8020 struct btrfs_path *path;
8021 struct btrfs_root *root = fs_info->dev_root;
8022 struct btrfs_key key;
8024 u64 prev_dev_ext_end = 0;
8028 * We don't have a dev_root because we mounted with ignorebadroots and
8029 * failed to load the root, so we want to skip the verification in this
8032 * However if the dev root is fine, but the tree itself is corrupted
8033 * we'd still fail to mount. This verification is only to make sure
8034 * writes can happen safely, so instead just bypass this check
8035 * completely in the case of IGNOREBADROOTS.
8037 if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
8041 key.type = BTRFS_DEV_EXTENT_KEY;
8044 path = btrfs_alloc_path();
8048 path->reada = READA_FORWARD;
8049 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8053 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
8054 ret = btrfs_next_leaf(root, path);
8057 /* No dev extents at all? Not good */
8064 struct extent_buffer *leaf = path->nodes[0];
8065 struct btrfs_dev_extent *dext;
8066 int slot = path->slots[0];
8068 u64 physical_offset;
8072 btrfs_item_key_to_cpu(leaf, &key, slot);
8073 if (key.type != BTRFS_DEV_EXTENT_KEY)
8075 devid = key.objectid;
8076 physical_offset = key.offset;
8078 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
8079 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
8080 physical_len = btrfs_dev_extent_length(leaf, dext);
8082 /* Check if this dev extent overlaps with the previous one */
8083 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
8085 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
8086 devid, physical_offset, prev_dev_ext_end);
8091 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
8092 physical_offset, physical_len);
8096 prev_dev_ext_end = physical_offset + physical_len;
8098 ret = btrfs_next_item(root, path);
8107 /* Ensure all chunks have corresponding dev extents */
8108 ret = verify_chunk_dev_extent_mapping(fs_info);
8110 btrfs_free_path(path);
8115 * Check whether the given block group or device is pinned by any inode being
8116 * used as a swapfile.
8118 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
8120 struct btrfs_swapfile_pin *sp;
8121 struct rb_node *node;
8123 spin_lock(&fs_info->swapfile_pins_lock);
8124 node = fs_info->swapfile_pins.rb_node;
8126 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
8128 node = node->rb_left;
8129 else if (ptr > sp->ptr)
8130 node = node->rb_right;
8134 spin_unlock(&fs_info->swapfile_pins_lock);
8135 return node != NULL;
8138 static int relocating_repair_kthread(void *data)
8140 struct btrfs_block_group *cache = (struct btrfs_block_group *)data;
8141 struct btrfs_fs_info *fs_info = cache->fs_info;
8145 target = cache->start;
8146 btrfs_put_block_group(cache);
8148 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
8150 "zoned: skip relocating block group %llu to repair: EBUSY",
8155 mutex_lock(&fs_info->reclaim_bgs_lock);
8157 /* Ensure block group still exists */
8158 cache = btrfs_lookup_block_group(fs_info, target);
8162 if (!cache->relocating_repair)
8165 ret = btrfs_may_alloc_data_chunk(fs_info, target);
8170 "zoned: relocating block group %llu to repair IO failure",
8172 ret = btrfs_relocate_chunk(fs_info, target);
8176 btrfs_put_block_group(cache);
8177 mutex_unlock(&fs_info->reclaim_bgs_lock);
8178 btrfs_exclop_finish(fs_info);
8183 int btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical)
8185 struct btrfs_block_group *cache;
8187 /* Do not attempt to repair in degraded state */
8188 if (btrfs_test_opt(fs_info, DEGRADED))
8191 cache = btrfs_lookup_block_group(fs_info, logical);
8195 spin_lock(&cache->lock);
8196 if (cache->relocating_repair) {
8197 spin_unlock(&cache->lock);
8198 btrfs_put_block_group(cache);
8201 cache->relocating_repair = 1;
8202 spin_unlock(&cache->lock);
8204 kthread_run(relocating_repair_kthread, cache,
8205 "btrfs-relocating-repair");