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 (device->devid == BTRFS_DEV_REPLACE_DEVID)
1122 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
1124 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1125 fs_devices->missing_devices--;
1127 btrfs_close_bdev(device);
1129 fs_devices->open_devices--;
1130 device->bdev = NULL;
1132 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1133 btrfs_destroy_dev_zone_info(device);
1135 device->fs_info = NULL;
1136 atomic_set(&device->dev_stats_ccnt, 0);
1137 extent_io_tree_release(&device->alloc_state);
1139 /* Verify the device is back in a pristine state */
1140 ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1141 ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1142 ASSERT(list_empty(&device->dev_alloc_list));
1143 ASSERT(list_empty(&device->post_commit_list));
1144 ASSERT(atomic_read(&device->reada_in_flight) == 0);
1147 static void close_fs_devices(struct btrfs_fs_devices *fs_devices)
1149 struct btrfs_device *device, *tmp;
1151 lockdep_assert_held(&uuid_mutex);
1153 if (--fs_devices->opened > 0)
1156 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list)
1157 btrfs_close_one_device(device);
1159 WARN_ON(fs_devices->open_devices);
1160 WARN_ON(fs_devices->rw_devices);
1161 fs_devices->opened = 0;
1162 fs_devices->seeding = false;
1163 fs_devices->fs_info = NULL;
1166 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1169 struct btrfs_fs_devices *tmp;
1171 mutex_lock(&uuid_mutex);
1172 close_fs_devices(fs_devices);
1173 if (!fs_devices->opened)
1174 list_splice_init(&fs_devices->seed_list, &list);
1176 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {
1177 close_fs_devices(fs_devices);
1178 list_del(&fs_devices->seed_list);
1179 free_fs_devices(fs_devices);
1181 mutex_unlock(&uuid_mutex);
1184 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1185 fmode_t flags, void *holder)
1187 struct btrfs_device *device;
1188 struct btrfs_device *latest_dev = NULL;
1189 struct btrfs_device *tmp_device;
1191 flags |= FMODE_EXCL;
1193 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,
1197 ret = btrfs_open_one_device(fs_devices, device, flags, holder);
1199 (!latest_dev || device->generation > latest_dev->generation)) {
1200 latest_dev = device;
1201 } else if (ret == -ENODATA) {
1202 fs_devices->num_devices--;
1203 list_del(&device->dev_list);
1204 btrfs_free_device(device);
1207 if (fs_devices->open_devices == 0)
1210 fs_devices->opened = 1;
1211 fs_devices->latest_bdev = latest_dev->bdev;
1212 fs_devices->total_rw_bytes = 0;
1213 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1214 fs_devices->read_policy = BTRFS_READ_POLICY_PID;
1219 static int devid_cmp(void *priv, const struct list_head *a,
1220 const struct list_head *b)
1222 const struct btrfs_device *dev1, *dev2;
1224 dev1 = list_entry(a, struct btrfs_device, dev_list);
1225 dev2 = list_entry(b, struct btrfs_device, dev_list);
1227 if (dev1->devid < dev2->devid)
1229 else if (dev1->devid > dev2->devid)
1234 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1235 fmode_t flags, void *holder)
1239 lockdep_assert_held(&uuid_mutex);
1241 * The device_list_mutex cannot be taken here in case opening the
1242 * underlying device takes further locks like open_mutex.
1244 * We also don't need the lock here as this is called during mount and
1245 * exclusion is provided by uuid_mutex
1248 if (fs_devices->opened) {
1249 fs_devices->opened++;
1252 list_sort(NULL, &fs_devices->devices, devid_cmp);
1253 ret = open_fs_devices(fs_devices, flags, holder);
1259 void btrfs_release_disk_super(struct btrfs_super_block *super)
1261 struct page *page = virt_to_page(super);
1266 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1267 u64 bytenr, u64 bytenr_orig)
1269 struct btrfs_super_block *disk_super;
1274 /* make sure our super fits in the device */
1275 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1276 return ERR_PTR(-EINVAL);
1278 /* make sure our super fits in the page */
1279 if (sizeof(*disk_super) > PAGE_SIZE)
1280 return ERR_PTR(-EINVAL);
1282 /* make sure our super doesn't straddle pages on disk */
1283 index = bytenr >> PAGE_SHIFT;
1284 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1285 return ERR_PTR(-EINVAL);
1287 /* pull in the page with our super */
1288 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1291 return ERR_CAST(page);
1293 p = page_address(page);
1295 /* align our pointer to the offset of the super block */
1296 disk_super = p + offset_in_page(bytenr);
1298 if (btrfs_super_bytenr(disk_super) != bytenr_orig ||
1299 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1300 btrfs_release_disk_super(p);
1301 return ERR_PTR(-EINVAL);
1304 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1305 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1310 int btrfs_forget_devices(const char *path)
1314 mutex_lock(&uuid_mutex);
1315 ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL);
1316 mutex_unlock(&uuid_mutex);
1322 * Look for a btrfs signature on a device. This may be called out of the mount path
1323 * and we are not allowed to call set_blocksize during the scan. The superblock
1324 * is read via pagecache
1326 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1329 struct btrfs_super_block *disk_super;
1330 bool new_device_added = false;
1331 struct btrfs_device *device = NULL;
1332 struct block_device *bdev;
1333 u64 bytenr, bytenr_orig;
1336 lockdep_assert_held(&uuid_mutex);
1339 * we would like to check all the supers, but that would make
1340 * a btrfs mount succeed after a mkfs from a different FS.
1341 * So, we need to add a special mount option to scan for
1342 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1344 flags |= FMODE_EXCL;
1346 bdev = blkdev_get_by_path(path, flags, holder);
1348 return ERR_CAST(bdev);
1350 bytenr_orig = btrfs_sb_offset(0);
1351 ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr);
1353 return ERR_PTR(ret);
1355 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig);
1356 if (IS_ERR(disk_super)) {
1357 device = ERR_CAST(disk_super);
1358 goto error_bdev_put;
1361 device = device_list_add(path, disk_super, &new_device_added);
1362 if (!IS_ERR(device)) {
1363 if (new_device_added)
1364 btrfs_free_stale_devices(path, device);
1367 btrfs_release_disk_super(disk_super);
1370 blkdev_put(bdev, flags);
1376 * Try to find a chunk that intersects [start, start + len] range and when one
1377 * such is found, record the end of it in *start
1379 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1382 u64 physical_start, physical_end;
1384 lockdep_assert_held(&device->fs_info->chunk_mutex);
1386 if (!find_first_extent_bit(&device->alloc_state, *start,
1387 &physical_start, &physical_end,
1388 CHUNK_ALLOCATED, NULL)) {
1390 if (in_range(physical_start, *start, len) ||
1391 in_range(*start, physical_start,
1392 physical_end - physical_start)) {
1393 *start = physical_end + 1;
1400 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start)
1402 switch (device->fs_devices->chunk_alloc_policy) {
1403 case BTRFS_CHUNK_ALLOC_REGULAR:
1405 * We don't want to overwrite the superblock on the drive nor
1406 * any area used by the boot loader (grub for example), so we
1407 * make sure to start at an offset of at least 1MB.
1409 return max_t(u64, start, SZ_1M);
1410 case BTRFS_CHUNK_ALLOC_ZONED:
1412 * We don't care about the starting region like regular
1413 * allocator, because we anyway use/reserve the first two zones
1414 * for superblock logging.
1416 return ALIGN(start, device->zone_info->zone_size);
1422 static bool dev_extent_hole_check_zoned(struct btrfs_device *device,
1423 u64 *hole_start, u64 *hole_size,
1426 u64 zone_size = device->zone_info->zone_size;
1429 bool changed = false;
1431 ASSERT(IS_ALIGNED(*hole_start, zone_size));
1433 while (*hole_size > 0) {
1434 pos = btrfs_find_allocatable_zones(device, *hole_start,
1435 *hole_start + *hole_size,
1437 if (pos != *hole_start) {
1438 *hole_size = *hole_start + *hole_size - pos;
1441 if (*hole_size < num_bytes)
1445 ret = btrfs_ensure_empty_zones(device, pos, num_bytes);
1447 /* Range is ensured to be empty */
1451 /* Given hole range was invalid (outside of device) */
1452 if (ret == -ERANGE) {
1453 *hole_start += *hole_size;
1458 *hole_start += zone_size;
1459 *hole_size -= zone_size;
1467 * dev_extent_hole_check - check if specified hole is suitable for allocation
1468 * @device: the device which we have the hole
1469 * @hole_start: starting position of the hole
1470 * @hole_size: the size of the hole
1471 * @num_bytes: the size of the free space that we need
1473 * This function may modify @hole_start and @hole_size to reflect the suitable
1474 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1476 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1477 u64 *hole_size, u64 num_bytes)
1479 bool changed = false;
1480 u64 hole_end = *hole_start + *hole_size;
1484 * Check before we set max_hole_start, otherwise we could end up
1485 * sending back this offset anyway.
1487 if (contains_pending_extent(device, hole_start, *hole_size)) {
1488 if (hole_end >= *hole_start)
1489 *hole_size = hole_end - *hole_start;
1495 switch (device->fs_devices->chunk_alloc_policy) {
1496 case BTRFS_CHUNK_ALLOC_REGULAR:
1497 /* No extra check */
1499 case BTRFS_CHUNK_ALLOC_ZONED:
1500 if (dev_extent_hole_check_zoned(device, hole_start,
1501 hole_size, num_bytes)) {
1504 * The changed hole can contain pending extent.
1505 * Loop again to check that.
1521 * find_free_dev_extent_start - find free space in the specified device
1522 * @device: the device which we search the free space in
1523 * @num_bytes: the size of the free space that we need
1524 * @search_start: the position from which to begin the search
1525 * @start: store the start of the free space.
1526 * @len: the size of the free space. that we find, or the size
1527 * of the max free space if we don't find suitable free space
1529 * this uses a pretty simple search, the expectation is that it is
1530 * called very infrequently and that a given device has a small number
1533 * @start is used to store the start of the free space if we find. But if we
1534 * don't find suitable free space, it will be used to store the start position
1535 * of the max free space.
1537 * @len is used to store the size of the free space that we find.
1538 * But if we don't find suitable free space, it is used to store the size of
1539 * the max free space.
1541 * NOTE: This function will search *commit* root of device tree, and does extra
1542 * check to ensure dev extents are not double allocated.
1543 * This makes the function safe to allocate dev extents but may not report
1544 * correct usable device space, as device extent freed in current transaction
1545 * is not reported as available.
1547 static int find_free_dev_extent_start(struct btrfs_device *device,
1548 u64 num_bytes, u64 search_start, u64 *start,
1551 struct btrfs_fs_info *fs_info = device->fs_info;
1552 struct btrfs_root *root = fs_info->dev_root;
1553 struct btrfs_key key;
1554 struct btrfs_dev_extent *dev_extent;
1555 struct btrfs_path *path;
1560 u64 search_end = device->total_bytes;
1563 struct extent_buffer *l;
1565 search_start = dev_extent_search_start(device, search_start);
1567 WARN_ON(device->zone_info &&
1568 !IS_ALIGNED(num_bytes, device->zone_info->zone_size));
1570 path = btrfs_alloc_path();
1574 max_hole_start = search_start;
1578 if (search_start >= search_end ||
1579 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1584 path->reada = READA_FORWARD;
1585 path->search_commit_root = 1;
1586 path->skip_locking = 1;
1588 key.objectid = device->devid;
1589 key.offset = search_start;
1590 key.type = BTRFS_DEV_EXTENT_KEY;
1592 ret = btrfs_search_backwards(root, &key, path);
1598 slot = path->slots[0];
1599 if (slot >= btrfs_header_nritems(l)) {
1600 ret = btrfs_next_leaf(root, path);
1608 btrfs_item_key_to_cpu(l, &key, slot);
1610 if (key.objectid < device->devid)
1613 if (key.objectid > device->devid)
1616 if (key.type != BTRFS_DEV_EXTENT_KEY)
1619 if (key.offset > search_start) {
1620 hole_size = key.offset - search_start;
1621 dev_extent_hole_check(device, &search_start, &hole_size,
1624 if (hole_size > max_hole_size) {
1625 max_hole_start = search_start;
1626 max_hole_size = hole_size;
1630 * If this free space is greater than which we need,
1631 * it must be the max free space that we have found
1632 * until now, so max_hole_start must point to the start
1633 * of this free space and the length of this free space
1634 * is stored in max_hole_size. Thus, we return
1635 * max_hole_start and max_hole_size and go back to the
1638 if (hole_size >= num_bytes) {
1644 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1645 extent_end = key.offset + btrfs_dev_extent_length(l,
1647 if (extent_end > search_start)
1648 search_start = extent_end;
1655 * At this point, search_start should be the end of
1656 * allocated dev extents, and when shrinking the device,
1657 * search_end may be smaller than search_start.
1659 if (search_end > search_start) {
1660 hole_size = search_end - search_start;
1661 if (dev_extent_hole_check(device, &search_start, &hole_size,
1663 btrfs_release_path(path);
1667 if (hole_size > max_hole_size) {
1668 max_hole_start = search_start;
1669 max_hole_size = hole_size;
1674 if (max_hole_size < num_bytes)
1680 btrfs_free_path(path);
1681 *start = max_hole_start;
1683 *len = max_hole_size;
1687 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1688 u64 *start, u64 *len)
1690 /* FIXME use last free of some kind */
1691 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1694 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1695 struct btrfs_device *device,
1696 u64 start, u64 *dev_extent_len)
1698 struct btrfs_fs_info *fs_info = device->fs_info;
1699 struct btrfs_root *root = fs_info->dev_root;
1701 struct btrfs_path *path;
1702 struct btrfs_key key;
1703 struct btrfs_key found_key;
1704 struct extent_buffer *leaf = NULL;
1705 struct btrfs_dev_extent *extent = NULL;
1707 path = btrfs_alloc_path();
1711 key.objectid = device->devid;
1713 key.type = BTRFS_DEV_EXTENT_KEY;
1715 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1717 ret = btrfs_previous_item(root, path, key.objectid,
1718 BTRFS_DEV_EXTENT_KEY);
1721 leaf = path->nodes[0];
1722 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1723 extent = btrfs_item_ptr(leaf, path->slots[0],
1724 struct btrfs_dev_extent);
1725 BUG_ON(found_key.offset > start || found_key.offset +
1726 btrfs_dev_extent_length(leaf, extent) < start);
1728 btrfs_release_path(path);
1730 } else if (ret == 0) {
1731 leaf = path->nodes[0];
1732 extent = btrfs_item_ptr(leaf, path->slots[0],
1733 struct btrfs_dev_extent);
1738 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1740 ret = btrfs_del_item(trans, root, path);
1742 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1744 btrfs_free_path(path);
1748 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1750 struct extent_map_tree *em_tree;
1751 struct extent_map *em;
1755 em_tree = &fs_info->mapping_tree;
1756 read_lock(&em_tree->lock);
1757 n = rb_last(&em_tree->map.rb_root);
1759 em = rb_entry(n, struct extent_map, rb_node);
1760 ret = em->start + em->len;
1762 read_unlock(&em_tree->lock);
1767 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1771 struct btrfs_key key;
1772 struct btrfs_key found_key;
1773 struct btrfs_path *path;
1775 path = btrfs_alloc_path();
1779 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1780 key.type = BTRFS_DEV_ITEM_KEY;
1781 key.offset = (u64)-1;
1783 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1789 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1794 ret = btrfs_previous_item(fs_info->chunk_root, path,
1795 BTRFS_DEV_ITEMS_OBJECTID,
1796 BTRFS_DEV_ITEM_KEY);
1800 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1802 *devid_ret = found_key.offset + 1;
1806 btrfs_free_path(path);
1811 * the device information is stored in the chunk root
1812 * the btrfs_device struct should be fully filled in
1814 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1815 struct btrfs_device *device)
1818 struct btrfs_path *path;
1819 struct btrfs_dev_item *dev_item;
1820 struct extent_buffer *leaf;
1821 struct btrfs_key key;
1824 path = btrfs_alloc_path();
1828 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1829 key.type = BTRFS_DEV_ITEM_KEY;
1830 key.offset = device->devid;
1832 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1833 &key, sizeof(*dev_item));
1837 leaf = path->nodes[0];
1838 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1840 btrfs_set_device_id(leaf, dev_item, device->devid);
1841 btrfs_set_device_generation(leaf, dev_item, 0);
1842 btrfs_set_device_type(leaf, dev_item, device->type);
1843 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1844 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1845 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1846 btrfs_set_device_total_bytes(leaf, dev_item,
1847 btrfs_device_get_disk_total_bytes(device));
1848 btrfs_set_device_bytes_used(leaf, dev_item,
1849 btrfs_device_get_bytes_used(device));
1850 btrfs_set_device_group(leaf, dev_item, 0);
1851 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1852 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1853 btrfs_set_device_start_offset(leaf, dev_item, 0);
1855 ptr = btrfs_device_uuid(dev_item);
1856 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1857 ptr = btrfs_device_fsid(dev_item);
1858 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1859 ptr, BTRFS_FSID_SIZE);
1860 btrfs_mark_buffer_dirty(leaf);
1864 btrfs_free_path(path);
1869 * Function to update ctime/mtime for a given device path.
1870 * Mainly used for ctime/mtime based probe like libblkid.
1872 static void update_dev_time(const char *path_name)
1876 filp = filp_open(path_name, O_RDWR, 0);
1879 file_update_time(filp);
1880 filp_close(filp, NULL);
1883 static int btrfs_rm_dev_item(struct btrfs_device *device)
1885 struct btrfs_root *root = device->fs_info->chunk_root;
1887 struct btrfs_path *path;
1888 struct btrfs_key key;
1889 struct btrfs_trans_handle *trans;
1891 path = btrfs_alloc_path();
1895 trans = btrfs_start_transaction(root, 0);
1896 if (IS_ERR(trans)) {
1897 btrfs_free_path(path);
1898 return PTR_ERR(trans);
1900 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1901 key.type = BTRFS_DEV_ITEM_KEY;
1902 key.offset = device->devid;
1904 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1908 btrfs_abort_transaction(trans, ret);
1909 btrfs_end_transaction(trans);
1913 ret = btrfs_del_item(trans, root, path);
1915 btrfs_abort_transaction(trans, ret);
1916 btrfs_end_transaction(trans);
1920 btrfs_free_path(path);
1922 ret = btrfs_commit_transaction(trans);
1927 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1928 * filesystem. It's up to the caller to adjust that number regarding eg. device
1931 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1939 seq = read_seqbegin(&fs_info->profiles_lock);
1941 all_avail = fs_info->avail_data_alloc_bits |
1942 fs_info->avail_system_alloc_bits |
1943 fs_info->avail_metadata_alloc_bits;
1944 } while (read_seqretry(&fs_info->profiles_lock, seq));
1946 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1947 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1950 if (num_devices < btrfs_raid_array[i].devs_min)
1951 return btrfs_raid_array[i].mindev_error;
1957 static struct btrfs_device * btrfs_find_next_active_device(
1958 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1960 struct btrfs_device *next_device;
1962 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1963 if (next_device != device &&
1964 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1965 && next_device->bdev)
1973 * Helper function to check if the given device is part of s_bdev / latest_bdev
1974 * and replace it with the provided or the next active device, in the context
1975 * where this function called, there should be always be another device (or
1976 * this_dev) which is active.
1978 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
1979 struct btrfs_device *next_device)
1981 struct btrfs_fs_info *fs_info = device->fs_info;
1984 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1986 ASSERT(next_device);
1988 if (fs_info->sb->s_bdev &&
1989 (fs_info->sb->s_bdev == device->bdev))
1990 fs_info->sb->s_bdev = next_device->bdev;
1992 if (fs_info->fs_devices->latest_bdev == device->bdev)
1993 fs_info->fs_devices->latest_bdev = next_device->bdev;
1997 * Return btrfs_fs_devices::num_devices excluding the device that's being
1998 * currently replaced.
2000 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2002 u64 num_devices = fs_info->fs_devices->num_devices;
2004 down_read(&fs_info->dev_replace.rwsem);
2005 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2006 ASSERT(num_devices > 1);
2009 up_read(&fs_info->dev_replace.rwsem);
2014 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
2015 struct block_device *bdev,
2016 const char *device_path)
2018 struct btrfs_super_block *disk_super;
2024 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2028 disk_super = btrfs_read_dev_one_super(bdev, copy_num);
2029 if (IS_ERR(disk_super))
2032 if (bdev_is_zoned(bdev)) {
2033 btrfs_reset_sb_log_zones(bdev, copy_num);
2037 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
2039 page = virt_to_page(disk_super);
2040 set_page_dirty(page);
2042 /* write_on_page() unlocks the page */
2043 ret = write_one_page(page);
2046 "error clearing superblock number %d (%d)",
2048 btrfs_release_disk_super(disk_super);
2052 /* Notify udev that device has changed */
2053 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2055 /* Update ctime/mtime for device path for libblkid */
2056 update_dev_time(device_path);
2059 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
2062 struct btrfs_device *device;
2063 struct btrfs_fs_devices *cur_devices;
2064 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2068 mutex_lock(&uuid_mutex);
2070 num_devices = btrfs_num_devices(fs_info);
2072 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2076 device = btrfs_find_device_by_devspec(fs_info, devid, device_path);
2078 if (IS_ERR(device)) {
2079 if (PTR_ERR(device) == -ENOENT &&
2080 device_path && strcmp(device_path, "missing") == 0)
2081 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2083 ret = PTR_ERR(device);
2087 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2088 btrfs_warn_in_rcu(fs_info,
2089 "cannot remove device %s (devid %llu) due to active swapfile",
2090 rcu_str_deref(device->name), device->devid);
2095 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2096 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
2100 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2101 fs_info->fs_devices->rw_devices == 1) {
2102 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
2106 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2107 mutex_lock(&fs_info->chunk_mutex);
2108 list_del_init(&device->dev_alloc_list);
2109 device->fs_devices->rw_devices--;
2110 mutex_unlock(&fs_info->chunk_mutex);
2113 mutex_unlock(&uuid_mutex);
2114 ret = btrfs_shrink_device(device, 0);
2116 btrfs_reada_remove_dev(device);
2117 mutex_lock(&uuid_mutex);
2122 * TODO: the superblock still includes this device in its num_devices
2123 * counter although write_all_supers() is not locked out. This
2124 * could give a filesystem state which requires a degraded mount.
2126 ret = btrfs_rm_dev_item(device);
2130 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2131 btrfs_scrub_cancel_dev(device);
2134 * the device list mutex makes sure that we don't change
2135 * the device list while someone else is writing out all
2136 * the device supers. Whoever is writing all supers, should
2137 * lock the device list mutex before getting the number of
2138 * devices in the super block (super_copy). Conversely,
2139 * whoever updates the number of devices in the super block
2140 * (super_copy) should hold the device list mutex.
2144 * In normal cases the cur_devices == fs_devices. But in case
2145 * of deleting a seed device, the cur_devices should point to
2146 * its own fs_devices listed under the fs_devices->seed.
2148 cur_devices = device->fs_devices;
2149 mutex_lock(&fs_devices->device_list_mutex);
2150 list_del_rcu(&device->dev_list);
2152 cur_devices->num_devices--;
2153 cur_devices->total_devices--;
2154 /* Update total_devices of the parent fs_devices if it's seed */
2155 if (cur_devices != fs_devices)
2156 fs_devices->total_devices--;
2158 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2159 cur_devices->missing_devices--;
2161 btrfs_assign_next_active_device(device, NULL);
2164 cur_devices->open_devices--;
2165 /* remove sysfs entry */
2166 btrfs_sysfs_remove_device(device);
2169 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2170 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2171 mutex_unlock(&fs_devices->device_list_mutex);
2174 * at this point, the device is zero sized and detached from
2175 * the devices list. All that's left is to zero out the old
2176 * supers and free the device.
2178 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2179 btrfs_scratch_superblocks(fs_info, device->bdev,
2182 btrfs_close_bdev(device);
2184 btrfs_free_device(device);
2186 if (cur_devices->open_devices == 0) {
2187 list_del_init(&cur_devices->seed_list);
2188 close_fs_devices(cur_devices);
2189 free_fs_devices(cur_devices);
2193 mutex_unlock(&uuid_mutex);
2197 btrfs_reada_undo_remove_dev(device);
2198 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2199 mutex_lock(&fs_info->chunk_mutex);
2200 list_add(&device->dev_alloc_list,
2201 &fs_devices->alloc_list);
2202 device->fs_devices->rw_devices++;
2203 mutex_unlock(&fs_info->chunk_mutex);
2208 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2210 struct btrfs_fs_devices *fs_devices;
2212 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2215 * in case of fs with no seed, srcdev->fs_devices will point
2216 * to fs_devices of fs_info. However when the dev being replaced is
2217 * a seed dev it will point to the seed's local fs_devices. In short
2218 * srcdev will have its correct fs_devices in both the cases.
2220 fs_devices = srcdev->fs_devices;
2222 list_del_rcu(&srcdev->dev_list);
2223 list_del(&srcdev->dev_alloc_list);
2224 fs_devices->num_devices--;
2225 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2226 fs_devices->missing_devices--;
2228 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2229 fs_devices->rw_devices--;
2232 fs_devices->open_devices--;
2235 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2237 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2239 mutex_lock(&uuid_mutex);
2241 btrfs_close_bdev(srcdev);
2243 btrfs_free_device(srcdev);
2245 /* if this is no devs we rather delete the fs_devices */
2246 if (!fs_devices->num_devices) {
2248 * On a mounted FS, num_devices can't be zero unless it's a
2249 * seed. In case of a seed device being replaced, the replace
2250 * target added to the sprout FS, so there will be no more
2251 * device left under the seed FS.
2253 ASSERT(fs_devices->seeding);
2255 list_del_init(&fs_devices->seed_list);
2256 close_fs_devices(fs_devices);
2257 free_fs_devices(fs_devices);
2259 mutex_unlock(&uuid_mutex);
2262 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2264 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2266 mutex_lock(&fs_devices->device_list_mutex);
2268 btrfs_sysfs_remove_device(tgtdev);
2271 fs_devices->open_devices--;
2273 fs_devices->num_devices--;
2275 btrfs_assign_next_active_device(tgtdev, NULL);
2277 list_del_rcu(&tgtdev->dev_list);
2279 mutex_unlock(&fs_devices->device_list_mutex);
2282 * The update_dev_time() with in btrfs_scratch_superblocks()
2283 * may lead to a call to btrfs_show_devname() which will try
2284 * to hold device_list_mutex. And here this device
2285 * is already out of device list, so we don't have to hold
2286 * the device_list_mutex lock.
2288 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2291 btrfs_close_bdev(tgtdev);
2293 btrfs_free_device(tgtdev);
2296 static struct btrfs_device *btrfs_find_device_by_path(
2297 struct btrfs_fs_info *fs_info, const char *device_path)
2300 struct btrfs_super_block *disk_super;
2303 struct block_device *bdev;
2304 struct btrfs_device *device;
2306 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2307 fs_info->bdev_holder, 0, &bdev, &disk_super);
2309 return ERR_PTR(ret);
2311 devid = btrfs_stack_device_id(&disk_super->dev_item);
2312 dev_uuid = disk_super->dev_item.uuid;
2313 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2314 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2315 disk_super->metadata_uuid);
2317 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2320 btrfs_release_disk_super(disk_super);
2322 device = ERR_PTR(-ENOENT);
2323 blkdev_put(bdev, FMODE_READ);
2328 * Lookup a device given by device id, or the path if the id is 0.
2330 struct btrfs_device *btrfs_find_device_by_devspec(
2331 struct btrfs_fs_info *fs_info, u64 devid,
2332 const char *device_path)
2334 struct btrfs_device *device;
2337 device = btrfs_find_device(fs_info->fs_devices, devid, NULL,
2340 return ERR_PTR(-ENOENT);
2344 if (!device_path || !device_path[0])
2345 return ERR_PTR(-EINVAL);
2347 if (strcmp(device_path, "missing") == 0) {
2348 /* Find first missing device */
2349 list_for_each_entry(device, &fs_info->fs_devices->devices,
2351 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2352 &device->dev_state) && !device->bdev)
2355 return ERR_PTR(-ENOENT);
2358 return btrfs_find_device_by_path(fs_info, device_path);
2362 * does all the dirty work required for changing file system's UUID.
2364 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2366 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2367 struct btrfs_fs_devices *old_devices;
2368 struct btrfs_fs_devices *seed_devices;
2369 struct btrfs_super_block *disk_super = fs_info->super_copy;
2370 struct btrfs_device *device;
2373 lockdep_assert_held(&uuid_mutex);
2374 if (!fs_devices->seeding)
2378 * Private copy of the seed devices, anchored at
2379 * fs_info->fs_devices->seed_list
2381 seed_devices = alloc_fs_devices(NULL, NULL);
2382 if (IS_ERR(seed_devices))
2383 return PTR_ERR(seed_devices);
2386 * It's necessary to retain a copy of the original seed fs_devices in
2387 * fs_uuids so that filesystems which have been seeded can successfully
2388 * reference the seed device from open_seed_devices. This also supports
2391 old_devices = clone_fs_devices(fs_devices);
2392 if (IS_ERR(old_devices)) {
2393 kfree(seed_devices);
2394 return PTR_ERR(old_devices);
2397 list_add(&old_devices->fs_list, &fs_uuids);
2399 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2400 seed_devices->opened = 1;
2401 INIT_LIST_HEAD(&seed_devices->devices);
2402 INIT_LIST_HEAD(&seed_devices->alloc_list);
2403 mutex_init(&seed_devices->device_list_mutex);
2405 mutex_lock(&fs_devices->device_list_mutex);
2406 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2408 list_for_each_entry(device, &seed_devices->devices, dev_list)
2409 device->fs_devices = seed_devices;
2411 fs_devices->seeding = false;
2412 fs_devices->num_devices = 0;
2413 fs_devices->open_devices = 0;
2414 fs_devices->missing_devices = 0;
2415 fs_devices->rotating = false;
2416 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2418 generate_random_uuid(fs_devices->fsid);
2419 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2420 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2421 mutex_unlock(&fs_devices->device_list_mutex);
2423 super_flags = btrfs_super_flags(disk_super) &
2424 ~BTRFS_SUPER_FLAG_SEEDING;
2425 btrfs_set_super_flags(disk_super, super_flags);
2431 * Store the expected generation for seed devices in device items.
2433 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2435 struct btrfs_fs_info *fs_info = trans->fs_info;
2436 struct btrfs_root *root = fs_info->chunk_root;
2437 struct btrfs_path *path;
2438 struct extent_buffer *leaf;
2439 struct btrfs_dev_item *dev_item;
2440 struct btrfs_device *device;
2441 struct btrfs_key key;
2442 u8 fs_uuid[BTRFS_FSID_SIZE];
2443 u8 dev_uuid[BTRFS_UUID_SIZE];
2447 path = btrfs_alloc_path();
2451 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2453 key.type = BTRFS_DEV_ITEM_KEY;
2456 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2460 leaf = path->nodes[0];
2462 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2463 ret = btrfs_next_leaf(root, path);
2468 leaf = path->nodes[0];
2469 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2470 btrfs_release_path(path);
2474 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2475 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2476 key.type != BTRFS_DEV_ITEM_KEY)
2479 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2480 struct btrfs_dev_item);
2481 devid = btrfs_device_id(leaf, dev_item);
2482 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2484 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2486 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2488 BUG_ON(!device); /* Logic error */
2490 if (device->fs_devices->seeding) {
2491 btrfs_set_device_generation(leaf, dev_item,
2492 device->generation);
2493 btrfs_mark_buffer_dirty(leaf);
2501 btrfs_free_path(path);
2505 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2507 struct btrfs_root *root = fs_info->dev_root;
2508 struct request_queue *q;
2509 struct btrfs_trans_handle *trans;
2510 struct btrfs_device *device;
2511 struct block_device *bdev;
2512 struct super_block *sb = fs_info->sb;
2513 struct rcu_string *name;
2514 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2515 u64 orig_super_total_bytes;
2516 u64 orig_super_num_devices;
2517 int seeding_dev = 0;
2519 bool locked = false;
2521 if (sb_rdonly(sb) && !fs_devices->seeding)
2524 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2525 fs_info->bdev_holder);
2527 return PTR_ERR(bdev);
2529 if (!btrfs_check_device_zone_type(fs_info, bdev)) {
2534 if (fs_devices->seeding) {
2536 down_write(&sb->s_umount);
2537 mutex_lock(&uuid_mutex);
2541 sync_blockdev(bdev);
2544 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2545 if (device->bdev == bdev) {
2553 device = btrfs_alloc_device(fs_info, NULL, NULL);
2554 if (IS_ERR(device)) {
2555 /* we can safely leave the fs_devices entry around */
2556 ret = PTR_ERR(device);
2560 name = rcu_string_strdup(device_path, GFP_KERNEL);
2563 goto error_free_device;
2565 rcu_assign_pointer(device->name, name);
2567 device->fs_info = fs_info;
2568 device->bdev = bdev;
2570 ret = btrfs_get_dev_zone_info(device);
2572 goto error_free_device;
2574 trans = btrfs_start_transaction(root, 0);
2575 if (IS_ERR(trans)) {
2576 ret = PTR_ERR(trans);
2577 goto error_free_zone;
2580 q = bdev_get_queue(bdev);
2581 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2582 device->generation = trans->transid;
2583 device->io_width = fs_info->sectorsize;
2584 device->io_align = fs_info->sectorsize;
2585 device->sector_size = fs_info->sectorsize;
2586 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2587 fs_info->sectorsize);
2588 device->disk_total_bytes = device->total_bytes;
2589 device->commit_total_bytes = device->total_bytes;
2590 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2591 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2592 device->mode = FMODE_EXCL;
2593 device->dev_stats_valid = 1;
2594 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2597 btrfs_clear_sb_rdonly(sb);
2598 ret = btrfs_prepare_sprout(fs_info);
2600 btrfs_abort_transaction(trans, ret);
2605 device->fs_devices = fs_devices;
2607 mutex_lock(&fs_devices->device_list_mutex);
2608 mutex_lock(&fs_info->chunk_mutex);
2609 list_add_rcu(&device->dev_list, &fs_devices->devices);
2610 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2611 fs_devices->num_devices++;
2612 fs_devices->open_devices++;
2613 fs_devices->rw_devices++;
2614 fs_devices->total_devices++;
2615 fs_devices->total_rw_bytes += device->total_bytes;
2617 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2619 if (!blk_queue_nonrot(q))
2620 fs_devices->rotating = true;
2622 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2623 btrfs_set_super_total_bytes(fs_info->super_copy,
2624 round_down(orig_super_total_bytes + device->total_bytes,
2625 fs_info->sectorsize));
2627 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2628 btrfs_set_super_num_devices(fs_info->super_copy,
2629 orig_super_num_devices + 1);
2632 * we've got more storage, clear any full flags on the space
2635 btrfs_clear_space_info_full(fs_info);
2637 mutex_unlock(&fs_info->chunk_mutex);
2639 /* Add sysfs device entry */
2640 btrfs_sysfs_add_device(device);
2642 mutex_unlock(&fs_devices->device_list_mutex);
2645 mutex_lock(&fs_info->chunk_mutex);
2646 ret = init_first_rw_device(trans);
2647 mutex_unlock(&fs_info->chunk_mutex);
2649 btrfs_abort_transaction(trans, ret);
2654 ret = btrfs_add_dev_item(trans, device);
2656 btrfs_abort_transaction(trans, ret);
2661 ret = btrfs_finish_sprout(trans);
2663 btrfs_abort_transaction(trans, ret);
2668 * fs_devices now represents the newly sprouted filesystem and
2669 * its fsid has been changed by btrfs_prepare_sprout
2671 btrfs_sysfs_update_sprout_fsid(fs_devices);
2674 ret = btrfs_commit_transaction(trans);
2677 mutex_unlock(&uuid_mutex);
2678 up_write(&sb->s_umount);
2681 if (ret) /* transaction commit */
2684 ret = btrfs_relocate_sys_chunks(fs_info);
2686 btrfs_handle_fs_error(fs_info, ret,
2687 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2688 trans = btrfs_attach_transaction(root);
2689 if (IS_ERR(trans)) {
2690 if (PTR_ERR(trans) == -ENOENT)
2692 ret = PTR_ERR(trans);
2696 ret = btrfs_commit_transaction(trans);
2700 * Now that we have written a new super block to this device, check all
2701 * other fs_devices list if device_path alienates any other scanned
2703 * We can ignore the return value as it typically returns -EINVAL and
2704 * only succeeds if the device was an alien.
2706 btrfs_forget_devices(device_path);
2708 /* Update ctime/mtime for blkid or udev */
2709 update_dev_time(device_path);
2714 btrfs_sysfs_remove_device(device);
2715 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2716 mutex_lock(&fs_info->chunk_mutex);
2717 list_del_rcu(&device->dev_list);
2718 list_del(&device->dev_alloc_list);
2719 fs_info->fs_devices->num_devices--;
2720 fs_info->fs_devices->open_devices--;
2721 fs_info->fs_devices->rw_devices--;
2722 fs_info->fs_devices->total_devices--;
2723 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2724 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2725 btrfs_set_super_total_bytes(fs_info->super_copy,
2726 orig_super_total_bytes);
2727 btrfs_set_super_num_devices(fs_info->super_copy,
2728 orig_super_num_devices);
2729 mutex_unlock(&fs_info->chunk_mutex);
2730 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2733 btrfs_set_sb_rdonly(sb);
2735 btrfs_end_transaction(trans);
2737 btrfs_destroy_dev_zone_info(device);
2739 btrfs_free_device(device);
2741 blkdev_put(bdev, FMODE_EXCL);
2743 mutex_unlock(&uuid_mutex);
2744 up_write(&sb->s_umount);
2749 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2750 struct btrfs_device *device)
2753 struct btrfs_path *path;
2754 struct btrfs_root *root = device->fs_info->chunk_root;
2755 struct btrfs_dev_item *dev_item;
2756 struct extent_buffer *leaf;
2757 struct btrfs_key key;
2759 path = btrfs_alloc_path();
2763 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2764 key.type = BTRFS_DEV_ITEM_KEY;
2765 key.offset = device->devid;
2767 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2776 leaf = path->nodes[0];
2777 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2779 btrfs_set_device_id(leaf, dev_item, device->devid);
2780 btrfs_set_device_type(leaf, dev_item, device->type);
2781 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2782 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2783 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2784 btrfs_set_device_total_bytes(leaf, dev_item,
2785 btrfs_device_get_disk_total_bytes(device));
2786 btrfs_set_device_bytes_used(leaf, dev_item,
2787 btrfs_device_get_bytes_used(device));
2788 btrfs_mark_buffer_dirty(leaf);
2791 btrfs_free_path(path);
2795 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2796 struct btrfs_device *device, u64 new_size)
2798 struct btrfs_fs_info *fs_info = device->fs_info;
2799 struct btrfs_super_block *super_copy = fs_info->super_copy;
2803 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2806 new_size = round_down(new_size, fs_info->sectorsize);
2808 mutex_lock(&fs_info->chunk_mutex);
2809 old_total = btrfs_super_total_bytes(super_copy);
2810 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2812 if (new_size <= device->total_bytes ||
2813 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2814 mutex_unlock(&fs_info->chunk_mutex);
2818 btrfs_set_super_total_bytes(super_copy,
2819 round_down(old_total + diff, fs_info->sectorsize));
2820 device->fs_devices->total_rw_bytes += diff;
2822 btrfs_device_set_total_bytes(device, new_size);
2823 btrfs_device_set_disk_total_bytes(device, new_size);
2824 btrfs_clear_space_info_full(device->fs_info);
2825 if (list_empty(&device->post_commit_list))
2826 list_add_tail(&device->post_commit_list,
2827 &trans->transaction->dev_update_list);
2828 mutex_unlock(&fs_info->chunk_mutex);
2830 return btrfs_update_device(trans, device);
2833 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2835 struct btrfs_fs_info *fs_info = trans->fs_info;
2836 struct btrfs_root *root = fs_info->chunk_root;
2838 struct btrfs_path *path;
2839 struct btrfs_key key;
2841 path = btrfs_alloc_path();
2845 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2846 key.offset = chunk_offset;
2847 key.type = BTRFS_CHUNK_ITEM_KEY;
2849 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2852 else if (ret > 0) { /* Logic error or corruption */
2853 btrfs_handle_fs_error(fs_info, -ENOENT,
2854 "Failed lookup while freeing chunk.");
2859 ret = btrfs_del_item(trans, root, path);
2861 btrfs_handle_fs_error(fs_info, ret,
2862 "Failed to delete chunk item.");
2864 btrfs_free_path(path);
2868 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2870 struct btrfs_super_block *super_copy = fs_info->super_copy;
2871 struct btrfs_disk_key *disk_key;
2872 struct btrfs_chunk *chunk;
2879 struct btrfs_key key;
2881 lockdep_assert_held(&fs_info->chunk_mutex);
2882 array_size = btrfs_super_sys_array_size(super_copy);
2884 ptr = super_copy->sys_chunk_array;
2887 while (cur < array_size) {
2888 disk_key = (struct btrfs_disk_key *)ptr;
2889 btrfs_disk_key_to_cpu(&key, disk_key);
2891 len = sizeof(*disk_key);
2893 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2894 chunk = (struct btrfs_chunk *)(ptr + len);
2895 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2896 len += btrfs_chunk_item_size(num_stripes);
2901 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2902 key.offset == chunk_offset) {
2903 memmove(ptr, ptr + len, array_size - (cur + len));
2905 btrfs_set_super_sys_array_size(super_copy, array_size);
2915 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
2916 * @logical: Logical block offset in bytes.
2917 * @length: Length of extent in bytes.
2919 * Return: Chunk mapping or ERR_PTR.
2921 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
2922 u64 logical, u64 length)
2924 struct extent_map_tree *em_tree;
2925 struct extent_map *em;
2927 em_tree = &fs_info->mapping_tree;
2928 read_lock(&em_tree->lock);
2929 em = lookup_extent_mapping(em_tree, logical, length);
2930 read_unlock(&em_tree->lock);
2933 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2935 return ERR_PTR(-EINVAL);
2938 if (em->start > logical || em->start + em->len < logical) {
2940 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2941 logical, length, em->start, em->start + em->len);
2942 free_extent_map(em);
2943 return ERR_PTR(-EINVAL);
2946 /* callers are responsible for dropping em's ref. */
2950 static int remove_chunk_item(struct btrfs_trans_handle *trans,
2951 struct map_lookup *map, u64 chunk_offset)
2956 * Removing chunk items and updating the device items in the chunks btree
2957 * requires holding the chunk_mutex.
2958 * See the comment at btrfs_chunk_alloc() for the details.
2960 lockdep_assert_held(&trans->fs_info->chunk_mutex);
2962 for (i = 0; i < map->num_stripes; i++) {
2965 ret = btrfs_update_device(trans, map->stripes[i].dev);
2970 return btrfs_free_chunk(trans, chunk_offset);
2973 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2975 struct btrfs_fs_info *fs_info = trans->fs_info;
2976 struct extent_map *em;
2977 struct map_lookup *map;
2978 u64 dev_extent_len = 0;
2980 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2982 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
2985 * This is a logic error, but we don't want to just rely on the
2986 * user having built with ASSERT enabled, so if ASSERT doesn't
2987 * do anything we still error out.
2992 map = em->map_lookup;
2995 * First delete the device extent items from the devices btree.
2996 * We take the device_list_mutex to avoid racing with the finishing phase
2997 * of a device replace operation. See the comment below before acquiring
2998 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
2999 * because that can result in a deadlock when deleting the device extent
3000 * items from the devices btree - COWing an extent buffer from the btree
3001 * may result in allocating a new metadata chunk, which would attempt to
3002 * lock again fs_info->chunk_mutex.
3004 mutex_lock(&fs_devices->device_list_mutex);
3005 for (i = 0; i < map->num_stripes; i++) {
3006 struct btrfs_device *device = map->stripes[i].dev;
3007 ret = btrfs_free_dev_extent(trans, device,
3008 map->stripes[i].physical,
3011 mutex_unlock(&fs_devices->device_list_mutex);
3012 btrfs_abort_transaction(trans, ret);
3016 if (device->bytes_used > 0) {
3017 mutex_lock(&fs_info->chunk_mutex);
3018 btrfs_device_set_bytes_used(device,
3019 device->bytes_used - dev_extent_len);
3020 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3021 btrfs_clear_space_info_full(fs_info);
3022 mutex_unlock(&fs_info->chunk_mutex);
3025 mutex_unlock(&fs_devices->device_list_mutex);
3028 * We acquire fs_info->chunk_mutex for 2 reasons:
3030 * 1) Just like with the first phase of the chunk allocation, we must
3031 * reserve system space, do all chunk btree updates and deletions, and
3032 * update the system chunk array in the superblock while holding this
3033 * mutex. This is for similar reasons as explained on the comment at
3034 * the top of btrfs_chunk_alloc();
3036 * 2) Prevent races with the final phase of a device replace operation
3037 * that replaces the device object associated with the map's stripes,
3038 * because the device object's id can change at any time during that
3039 * final phase of the device replace operation
3040 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3041 * replaced device and then see it with an ID of
3042 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3043 * the device item, which does not exists on the chunk btree.
3044 * The finishing phase of device replace acquires both the
3045 * device_list_mutex and the chunk_mutex, in that order, so we are
3046 * safe by just acquiring the chunk_mutex.
3048 trans->removing_chunk = true;
3049 mutex_lock(&fs_info->chunk_mutex);
3051 check_system_chunk(trans, map->type);
3053 ret = remove_chunk_item(trans, map, chunk_offset);
3055 * Normally we should not get -ENOSPC since we reserved space before
3056 * through the call to check_system_chunk().
3058 * Despite our system space_info having enough free space, we may not
3059 * be able to allocate extents from its block groups, because all have
3060 * an incompatible profile, which will force us to allocate a new system
3061 * block group with the right profile, or right after we called
3062 * check_system_space() above, a scrub turned the only system block group
3063 * with enough free space into RO mode.
3064 * This is explained with more detail at do_chunk_alloc().
3066 * So if we get -ENOSPC, allocate a new system chunk and retry once.
3068 if (ret == -ENOSPC) {
3069 const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
3070 struct btrfs_block_group *sys_bg;
3072 sys_bg = btrfs_alloc_chunk(trans, sys_flags);
3073 if (IS_ERR(sys_bg)) {
3074 ret = PTR_ERR(sys_bg);
3075 btrfs_abort_transaction(trans, ret);
3079 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3081 btrfs_abort_transaction(trans, ret);
3085 ret = remove_chunk_item(trans, map, chunk_offset);
3087 btrfs_abort_transaction(trans, ret);
3091 btrfs_abort_transaction(trans, ret);
3095 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3097 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3098 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3100 btrfs_abort_transaction(trans, ret);
3105 mutex_unlock(&fs_info->chunk_mutex);
3106 trans->removing_chunk = false;
3109 * We are done with chunk btree updates and deletions, so release the
3110 * system space we previously reserved (with check_system_chunk()).
3112 btrfs_trans_release_chunk_metadata(trans);
3114 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3116 btrfs_abort_transaction(trans, ret);
3121 if (trans->removing_chunk) {
3122 mutex_unlock(&fs_info->chunk_mutex);
3123 trans->removing_chunk = false;
3126 free_extent_map(em);
3130 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3132 struct btrfs_root *root = fs_info->chunk_root;
3133 struct btrfs_trans_handle *trans;
3134 struct btrfs_block_group *block_group;
3139 * Prevent races with automatic removal of unused block groups.
3140 * After we relocate and before we remove the chunk with offset
3141 * chunk_offset, automatic removal of the block group can kick in,
3142 * resulting in a failure when calling btrfs_remove_chunk() below.
3144 * Make sure to acquire this mutex before doing a tree search (dev
3145 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3146 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3147 * we release the path used to search the chunk/dev tree and before
3148 * the current task acquires this mutex and calls us.
3150 lockdep_assert_held(&fs_info->reclaim_bgs_lock);
3152 /* step one, relocate all the extents inside this chunk */
3153 btrfs_scrub_pause(fs_info);
3154 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3155 btrfs_scrub_continue(fs_info);
3159 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3162 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3163 length = block_group->length;
3164 btrfs_put_block_group(block_group);
3167 * On a zoned file system, discard the whole block group, this will
3168 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3169 * resetting the zone fails, don't treat it as a fatal problem from the
3170 * filesystem's point of view.
3172 if (btrfs_is_zoned(fs_info)) {
3173 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
3176 "failed to reset zone %llu after relocation",
3180 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3182 if (IS_ERR(trans)) {
3183 ret = PTR_ERR(trans);
3184 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3189 * step two, delete the device extents and the
3190 * chunk tree entries
3192 ret = btrfs_remove_chunk(trans, chunk_offset);
3193 btrfs_end_transaction(trans);
3197 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3199 struct btrfs_root *chunk_root = fs_info->chunk_root;
3200 struct btrfs_path *path;
3201 struct extent_buffer *leaf;
3202 struct btrfs_chunk *chunk;
3203 struct btrfs_key key;
3204 struct btrfs_key found_key;
3206 bool retried = false;
3210 path = btrfs_alloc_path();
3215 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3216 key.offset = (u64)-1;
3217 key.type = BTRFS_CHUNK_ITEM_KEY;
3220 mutex_lock(&fs_info->reclaim_bgs_lock);
3221 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3223 mutex_unlock(&fs_info->reclaim_bgs_lock);
3226 BUG_ON(ret == 0); /* Corruption */
3228 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3231 mutex_unlock(&fs_info->reclaim_bgs_lock);
3237 leaf = path->nodes[0];
3238 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3240 chunk = btrfs_item_ptr(leaf, path->slots[0],
3241 struct btrfs_chunk);
3242 chunk_type = btrfs_chunk_type(leaf, chunk);
3243 btrfs_release_path(path);
3245 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3246 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3252 mutex_unlock(&fs_info->reclaim_bgs_lock);
3254 if (found_key.offset == 0)
3256 key.offset = found_key.offset - 1;
3259 if (failed && !retried) {
3263 } else if (WARN_ON(failed && retried)) {
3267 btrfs_free_path(path);
3272 * return 1 : allocate a data chunk successfully,
3273 * return <0: errors during allocating a data chunk,
3274 * return 0 : no need to allocate a data chunk.
3276 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3279 struct btrfs_block_group *cache;
3283 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3285 chunk_type = cache->flags;
3286 btrfs_put_block_group(cache);
3288 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3291 spin_lock(&fs_info->data_sinfo->lock);
3292 bytes_used = fs_info->data_sinfo->bytes_used;
3293 spin_unlock(&fs_info->data_sinfo->lock);
3296 struct btrfs_trans_handle *trans;
3299 trans = btrfs_join_transaction(fs_info->tree_root);
3301 return PTR_ERR(trans);
3303 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3304 btrfs_end_transaction(trans);
3313 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3314 struct btrfs_balance_control *bctl)
3316 struct btrfs_root *root = fs_info->tree_root;
3317 struct btrfs_trans_handle *trans;
3318 struct btrfs_balance_item *item;
3319 struct btrfs_disk_balance_args disk_bargs;
3320 struct btrfs_path *path;
3321 struct extent_buffer *leaf;
3322 struct btrfs_key key;
3325 path = btrfs_alloc_path();
3329 trans = btrfs_start_transaction(root, 0);
3330 if (IS_ERR(trans)) {
3331 btrfs_free_path(path);
3332 return PTR_ERR(trans);
3335 key.objectid = BTRFS_BALANCE_OBJECTID;
3336 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3339 ret = btrfs_insert_empty_item(trans, root, path, &key,
3344 leaf = path->nodes[0];
3345 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3347 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3349 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3350 btrfs_set_balance_data(leaf, item, &disk_bargs);
3351 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3352 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3353 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3354 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3356 btrfs_set_balance_flags(leaf, item, bctl->flags);
3358 btrfs_mark_buffer_dirty(leaf);
3360 btrfs_free_path(path);
3361 err = btrfs_commit_transaction(trans);
3367 static int del_balance_item(struct btrfs_fs_info *fs_info)
3369 struct btrfs_root *root = fs_info->tree_root;
3370 struct btrfs_trans_handle *trans;
3371 struct btrfs_path *path;
3372 struct btrfs_key key;
3375 path = btrfs_alloc_path();
3379 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3380 if (IS_ERR(trans)) {
3381 btrfs_free_path(path);
3382 return PTR_ERR(trans);
3385 key.objectid = BTRFS_BALANCE_OBJECTID;
3386 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3389 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3397 ret = btrfs_del_item(trans, root, path);
3399 btrfs_free_path(path);
3400 err = btrfs_commit_transaction(trans);
3407 * This is a heuristic used to reduce the number of chunks balanced on
3408 * resume after balance was interrupted.
3410 static void update_balance_args(struct btrfs_balance_control *bctl)
3413 * Turn on soft mode for chunk types that were being converted.
3415 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3416 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3417 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3418 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3419 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3420 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3423 * Turn on usage filter if is not already used. The idea is
3424 * that chunks that we have already balanced should be
3425 * reasonably full. Don't do it for chunks that are being
3426 * converted - that will keep us from relocating unconverted
3427 * (albeit full) chunks.
3429 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3430 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3431 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3432 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3433 bctl->data.usage = 90;
3435 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3436 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3437 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3438 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3439 bctl->sys.usage = 90;
3441 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3442 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3443 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3444 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3445 bctl->meta.usage = 90;
3450 * Clear the balance status in fs_info and delete the balance item from disk.
3452 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3454 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3457 BUG_ON(!fs_info->balance_ctl);
3459 spin_lock(&fs_info->balance_lock);
3460 fs_info->balance_ctl = NULL;
3461 spin_unlock(&fs_info->balance_lock);
3464 ret = del_balance_item(fs_info);
3466 btrfs_handle_fs_error(fs_info, ret, NULL);
3470 * Balance filters. Return 1 if chunk should be filtered out
3471 * (should not be balanced).
3473 static int chunk_profiles_filter(u64 chunk_type,
3474 struct btrfs_balance_args *bargs)
3476 chunk_type = chunk_to_extended(chunk_type) &
3477 BTRFS_EXTENDED_PROFILE_MASK;
3479 if (bargs->profiles & chunk_type)
3485 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3486 struct btrfs_balance_args *bargs)
3488 struct btrfs_block_group *cache;
3490 u64 user_thresh_min;
3491 u64 user_thresh_max;
3494 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3495 chunk_used = cache->used;
3497 if (bargs->usage_min == 0)
3498 user_thresh_min = 0;
3500 user_thresh_min = div_factor_fine(cache->length,
3503 if (bargs->usage_max == 0)
3504 user_thresh_max = 1;
3505 else if (bargs->usage_max > 100)
3506 user_thresh_max = cache->length;
3508 user_thresh_max = div_factor_fine(cache->length,
3511 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3514 btrfs_put_block_group(cache);
3518 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3519 u64 chunk_offset, struct btrfs_balance_args *bargs)
3521 struct btrfs_block_group *cache;
3522 u64 chunk_used, user_thresh;
3525 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3526 chunk_used = cache->used;
3528 if (bargs->usage_min == 0)
3530 else if (bargs->usage > 100)
3531 user_thresh = cache->length;
3533 user_thresh = div_factor_fine(cache->length, bargs->usage);
3535 if (chunk_used < user_thresh)
3538 btrfs_put_block_group(cache);
3542 static int chunk_devid_filter(struct extent_buffer *leaf,
3543 struct btrfs_chunk *chunk,
3544 struct btrfs_balance_args *bargs)
3546 struct btrfs_stripe *stripe;
3547 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3550 for (i = 0; i < num_stripes; i++) {
3551 stripe = btrfs_stripe_nr(chunk, i);
3552 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3559 static u64 calc_data_stripes(u64 type, int num_stripes)
3561 const int index = btrfs_bg_flags_to_raid_index(type);
3562 const int ncopies = btrfs_raid_array[index].ncopies;
3563 const int nparity = btrfs_raid_array[index].nparity;
3565 return (num_stripes - nparity) / ncopies;
3568 /* [pstart, pend) */
3569 static int chunk_drange_filter(struct extent_buffer *leaf,
3570 struct btrfs_chunk *chunk,
3571 struct btrfs_balance_args *bargs)
3573 struct btrfs_stripe *stripe;
3574 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3581 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3584 type = btrfs_chunk_type(leaf, chunk);
3585 factor = calc_data_stripes(type, num_stripes);
3587 for (i = 0; i < num_stripes; i++) {
3588 stripe = btrfs_stripe_nr(chunk, i);
3589 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3592 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3593 stripe_length = btrfs_chunk_length(leaf, chunk);
3594 stripe_length = div_u64(stripe_length, factor);
3596 if (stripe_offset < bargs->pend &&
3597 stripe_offset + stripe_length > bargs->pstart)
3604 /* [vstart, vend) */
3605 static int chunk_vrange_filter(struct extent_buffer *leaf,
3606 struct btrfs_chunk *chunk,
3608 struct btrfs_balance_args *bargs)
3610 if (chunk_offset < bargs->vend &&
3611 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3612 /* at least part of the chunk is inside this vrange */
3618 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3619 struct btrfs_chunk *chunk,
3620 struct btrfs_balance_args *bargs)
3622 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3624 if (bargs->stripes_min <= num_stripes
3625 && num_stripes <= bargs->stripes_max)
3631 static int chunk_soft_convert_filter(u64 chunk_type,
3632 struct btrfs_balance_args *bargs)
3634 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3637 chunk_type = chunk_to_extended(chunk_type) &
3638 BTRFS_EXTENDED_PROFILE_MASK;
3640 if (bargs->target == chunk_type)
3646 static int should_balance_chunk(struct extent_buffer *leaf,
3647 struct btrfs_chunk *chunk, u64 chunk_offset)
3649 struct btrfs_fs_info *fs_info = leaf->fs_info;
3650 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3651 struct btrfs_balance_args *bargs = NULL;
3652 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3655 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3656 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3660 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3661 bargs = &bctl->data;
3662 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3664 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3665 bargs = &bctl->meta;
3667 /* profiles filter */
3668 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3669 chunk_profiles_filter(chunk_type, bargs)) {
3674 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3675 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3677 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3678 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3683 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3684 chunk_devid_filter(leaf, chunk, bargs)) {
3688 /* drange filter, makes sense only with devid filter */
3689 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3690 chunk_drange_filter(leaf, chunk, bargs)) {
3695 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3696 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3700 /* stripes filter */
3701 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3702 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3706 /* soft profile changing mode */
3707 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3708 chunk_soft_convert_filter(chunk_type, bargs)) {
3713 * limited by count, must be the last filter
3715 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3716 if (bargs->limit == 0)
3720 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3722 * Same logic as the 'limit' filter; the minimum cannot be
3723 * determined here because we do not have the global information
3724 * about the count of all chunks that satisfy the filters.
3726 if (bargs->limit_max == 0)
3735 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3737 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3738 struct btrfs_root *chunk_root = fs_info->chunk_root;
3740 struct btrfs_chunk *chunk;
3741 struct btrfs_path *path = NULL;
3742 struct btrfs_key key;
3743 struct btrfs_key found_key;
3744 struct extent_buffer *leaf;
3747 int enospc_errors = 0;
3748 bool counting = true;
3749 /* The single value limit and min/max limits use the same bytes in the */
3750 u64 limit_data = bctl->data.limit;
3751 u64 limit_meta = bctl->meta.limit;
3752 u64 limit_sys = bctl->sys.limit;
3756 int chunk_reserved = 0;
3758 path = btrfs_alloc_path();
3764 /* zero out stat counters */
3765 spin_lock(&fs_info->balance_lock);
3766 memset(&bctl->stat, 0, sizeof(bctl->stat));
3767 spin_unlock(&fs_info->balance_lock);
3771 * The single value limit and min/max limits use the same bytes
3774 bctl->data.limit = limit_data;
3775 bctl->meta.limit = limit_meta;
3776 bctl->sys.limit = limit_sys;
3778 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3779 key.offset = (u64)-1;
3780 key.type = BTRFS_CHUNK_ITEM_KEY;
3783 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3784 atomic_read(&fs_info->balance_cancel_req)) {
3789 mutex_lock(&fs_info->reclaim_bgs_lock);
3790 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3792 mutex_unlock(&fs_info->reclaim_bgs_lock);
3797 * this shouldn't happen, it means the last relocate
3801 BUG(); /* FIXME break ? */
3803 ret = btrfs_previous_item(chunk_root, path, 0,
3804 BTRFS_CHUNK_ITEM_KEY);
3806 mutex_unlock(&fs_info->reclaim_bgs_lock);
3811 leaf = path->nodes[0];
3812 slot = path->slots[0];
3813 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3815 if (found_key.objectid != key.objectid) {
3816 mutex_unlock(&fs_info->reclaim_bgs_lock);
3820 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3821 chunk_type = btrfs_chunk_type(leaf, chunk);
3824 spin_lock(&fs_info->balance_lock);
3825 bctl->stat.considered++;
3826 spin_unlock(&fs_info->balance_lock);
3829 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3831 btrfs_release_path(path);
3833 mutex_unlock(&fs_info->reclaim_bgs_lock);
3838 mutex_unlock(&fs_info->reclaim_bgs_lock);
3839 spin_lock(&fs_info->balance_lock);
3840 bctl->stat.expected++;
3841 spin_unlock(&fs_info->balance_lock);
3843 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3845 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3847 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3854 * Apply limit_min filter, no need to check if the LIMITS
3855 * filter is used, limit_min is 0 by default
3857 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3858 count_data < bctl->data.limit_min)
3859 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3860 count_meta < bctl->meta.limit_min)
3861 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3862 count_sys < bctl->sys.limit_min)) {
3863 mutex_unlock(&fs_info->reclaim_bgs_lock);
3867 if (!chunk_reserved) {
3869 * We may be relocating the only data chunk we have,
3870 * which could potentially end up with losing data's
3871 * raid profile, so lets allocate an empty one in
3874 ret = btrfs_may_alloc_data_chunk(fs_info,
3877 mutex_unlock(&fs_info->reclaim_bgs_lock);
3879 } else if (ret == 1) {
3884 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3885 mutex_unlock(&fs_info->reclaim_bgs_lock);
3886 if (ret == -ENOSPC) {
3888 } else if (ret == -ETXTBSY) {
3890 "skipping relocation of block group %llu due to active swapfile",
3896 spin_lock(&fs_info->balance_lock);
3897 bctl->stat.completed++;
3898 spin_unlock(&fs_info->balance_lock);
3901 if (found_key.offset == 0)
3903 key.offset = found_key.offset - 1;
3907 btrfs_release_path(path);
3912 btrfs_free_path(path);
3913 if (enospc_errors) {
3914 btrfs_info(fs_info, "%d enospc errors during balance",
3924 * alloc_profile_is_valid - see if a given profile is valid and reduced
3925 * @flags: profile to validate
3926 * @extended: if true @flags is treated as an extended profile
3928 static int alloc_profile_is_valid(u64 flags, int extended)
3930 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3931 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3933 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3935 /* 1) check that all other bits are zeroed */
3939 /* 2) see if profile is reduced */
3941 return !extended; /* "0" is valid for usual profiles */
3943 return has_single_bit_set(flags);
3946 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3948 /* cancel requested || normal exit path */
3949 return atomic_read(&fs_info->balance_cancel_req) ||
3950 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3951 atomic_read(&fs_info->balance_cancel_req) == 0);
3955 * Validate target profile against allowed profiles and return true if it's OK.
3956 * Otherwise print the error message and return false.
3958 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
3959 const struct btrfs_balance_args *bargs,
3960 u64 allowed, const char *type)
3962 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3965 if (fs_info->sectorsize < PAGE_SIZE &&
3966 bargs->target & BTRFS_BLOCK_GROUP_RAID56_MASK) {
3968 "RAID56 is not yet supported for sectorsize %u with page size %lu",
3969 fs_info->sectorsize, PAGE_SIZE);
3972 /* Profile is valid and does not have bits outside of the allowed set */
3973 if (alloc_profile_is_valid(bargs->target, 1) &&
3974 (bargs->target & ~allowed) == 0)
3977 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
3978 type, btrfs_bg_type_to_raid_name(bargs->target));
3983 * Fill @buf with textual description of balance filter flags @bargs, up to
3984 * @size_buf including the terminating null. The output may be trimmed if it
3985 * does not fit into the provided buffer.
3987 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
3991 u32 size_bp = size_buf;
3993 u64 flags = bargs->flags;
3994 char tmp_buf[128] = {'\0'};
3999 #define CHECK_APPEND_NOARG(a) \
4001 ret = snprintf(bp, size_bp, (a)); \
4002 if (ret < 0 || ret >= size_bp) \
4003 goto out_overflow; \
4008 #define CHECK_APPEND_1ARG(a, v1) \
4010 ret = snprintf(bp, size_bp, (a), (v1)); \
4011 if (ret < 0 || ret >= size_bp) \
4012 goto out_overflow; \
4017 #define CHECK_APPEND_2ARG(a, v1, v2) \
4019 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4020 if (ret < 0 || ret >= size_bp) \
4021 goto out_overflow; \
4026 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
4027 CHECK_APPEND_1ARG("convert=%s,",
4028 btrfs_bg_type_to_raid_name(bargs->target));
4030 if (flags & BTRFS_BALANCE_ARGS_SOFT)
4031 CHECK_APPEND_NOARG("soft,");
4033 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4034 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4036 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4039 if (flags & BTRFS_BALANCE_ARGS_USAGE)
4040 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4042 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4043 CHECK_APPEND_2ARG("usage=%u..%u,",
4044 bargs->usage_min, bargs->usage_max);
4046 if (flags & BTRFS_BALANCE_ARGS_DEVID)
4047 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4049 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4050 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4051 bargs->pstart, bargs->pend);
4053 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4054 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4055 bargs->vstart, bargs->vend);
4057 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4058 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4060 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4061 CHECK_APPEND_2ARG("limit=%u..%u,",
4062 bargs->limit_min, bargs->limit_max);
4064 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4065 CHECK_APPEND_2ARG("stripes=%u..%u,",
4066 bargs->stripes_min, bargs->stripes_max);
4068 #undef CHECK_APPEND_2ARG
4069 #undef CHECK_APPEND_1ARG
4070 #undef CHECK_APPEND_NOARG
4074 if (size_bp < size_buf)
4075 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4080 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4082 u32 size_buf = 1024;
4083 char tmp_buf[192] = {'\0'};
4086 u32 size_bp = size_buf;
4088 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4090 buf = kzalloc(size_buf, GFP_KERNEL);
4096 #define CHECK_APPEND_1ARG(a, v1) \
4098 ret = snprintf(bp, size_bp, (a), (v1)); \
4099 if (ret < 0 || ret >= size_bp) \
4100 goto out_overflow; \
4105 if (bctl->flags & BTRFS_BALANCE_FORCE)
4106 CHECK_APPEND_1ARG("%s", "-f ");
4108 if (bctl->flags & BTRFS_BALANCE_DATA) {
4109 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4110 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4113 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4114 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4115 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4118 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4119 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4120 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4123 #undef CHECK_APPEND_1ARG
4127 if (size_bp < size_buf)
4128 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4129 btrfs_info(fs_info, "balance: %s %s",
4130 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4131 "resume" : "start", buf);
4137 * Should be called with balance mutexe held
4139 int btrfs_balance(struct btrfs_fs_info *fs_info,
4140 struct btrfs_balance_control *bctl,
4141 struct btrfs_ioctl_balance_args *bargs)
4143 u64 meta_target, data_target;
4149 bool reducing_redundancy;
4152 if (btrfs_fs_closing(fs_info) ||
4153 atomic_read(&fs_info->balance_pause_req) ||
4154 btrfs_should_cancel_balance(fs_info)) {
4159 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4160 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4164 * In case of mixed groups both data and meta should be picked,
4165 * and identical options should be given for both of them.
4167 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4168 if (mixed && (bctl->flags & allowed)) {
4169 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4170 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4171 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4173 "balance: mixed groups data and metadata options must be the same");
4180 * rw_devices will not change at the moment, device add/delete/replace
4183 num_devices = fs_info->fs_devices->rw_devices;
4186 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4187 * special bit for it, to make it easier to distinguish. Thus we need
4188 * to set it manually, or balance would refuse the profile.
4190 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4191 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4192 if (num_devices >= btrfs_raid_array[i].devs_min)
4193 allowed |= btrfs_raid_array[i].bg_flag;
4195 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4196 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4197 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4203 * Allow to reduce metadata or system integrity only if force set for
4204 * profiles with redundancy (copies, parity)
4207 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4208 if (btrfs_raid_array[i].ncopies >= 2 ||
4209 btrfs_raid_array[i].tolerated_failures >= 1)
4210 allowed |= btrfs_raid_array[i].bg_flag;
4213 seq = read_seqbegin(&fs_info->profiles_lock);
4215 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4216 (fs_info->avail_system_alloc_bits & allowed) &&
4217 !(bctl->sys.target & allowed)) ||
4218 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4219 (fs_info->avail_metadata_alloc_bits & allowed) &&
4220 !(bctl->meta.target & allowed)))
4221 reducing_redundancy = true;
4223 reducing_redundancy = false;
4225 /* if we're not converting, the target field is uninitialized */
4226 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4227 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4228 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4229 bctl->data.target : fs_info->avail_data_alloc_bits;
4230 } while (read_seqretry(&fs_info->profiles_lock, seq));
4232 if (reducing_redundancy) {
4233 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4235 "balance: force reducing metadata redundancy");
4238 "balance: reduces metadata redundancy, use --force if you want this");
4244 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4245 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4247 "balance: metadata profile %s has lower redundancy than data profile %s",
4248 btrfs_bg_type_to_raid_name(meta_target),
4249 btrfs_bg_type_to_raid_name(data_target));
4252 ret = insert_balance_item(fs_info, bctl);
4253 if (ret && ret != -EEXIST)
4256 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4257 BUG_ON(ret == -EEXIST);
4258 BUG_ON(fs_info->balance_ctl);
4259 spin_lock(&fs_info->balance_lock);
4260 fs_info->balance_ctl = bctl;
4261 spin_unlock(&fs_info->balance_lock);
4263 BUG_ON(ret != -EEXIST);
4264 spin_lock(&fs_info->balance_lock);
4265 update_balance_args(bctl);
4266 spin_unlock(&fs_info->balance_lock);
4269 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4270 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4271 describe_balance_start_or_resume(fs_info);
4272 mutex_unlock(&fs_info->balance_mutex);
4274 ret = __btrfs_balance(fs_info);
4276 mutex_lock(&fs_info->balance_mutex);
4277 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req))
4278 btrfs_info(fs_info, "balance: paused");
4280 * Balance can be canceled by:
4282 * - Regular cancel request
4283 * Then ret == -ECANCELED and balance_cancel_req > 0
4285 * - Fatal signal to "btrfs" process
4286 * Either the signal caught by wait_reserve_ticket() and callers
4287 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4289 * Either way, in this case balance_cancel_req = 0, and
4290 * ret == -EINTR or ret == -ECANCELED.
4292 * So here we only check the return value to catch canceled balance.
4294 else if (ret == -ECANCELED || ret == -EINTR)
4295 btrfs_info(fs_info, "balance: canceled");
4297 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4299 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4302 memset(bargs, 0, sizeof(*bargs));
4303 btrfs_update_ioctl_balance_args(fs_info, bargs);
4306 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4307 balance_need_close(fs_info)) {
4308 reset_balance_state(fs_info);
4309 btrfs_exclop_finish(fs_info);
4312 wake_up(&fs_info->balance_wait_q);
4316 if (bctl->flags & BTRFS_BALANCE_RESUME)
4317 reset_balance_state(fs_info);
4320 btrfs_exclop_finish(fs_info);
4325 static int balance_kthread(void *data)
4327 struct btrfs_fs_info *fs_info = data;
4330 mutex_lock(&fs_info->balance_mutex);
4331 if (fs_info->balance_ctl)
4332 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4333 mutex_unlock(&fs_info->balance_mutex);
4338 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4340 struct task_struct *tsk;
4342 mutex_lock(&fs_info->balance_mutex);
4343 if (!fs_info->balance_ctl) {
4344 mutex_unlock(&fs_info->balance_mutex);
4347 mutex_unlock(&fs_info->balance_mutex);
4349 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4350 btrfs_info(fs_info, "balance: resume skipped");
4355 * A ro->rw remount sequence should continue with the paused balance
4356 * regardless of who pauses it, system or the user as of now, so set
4359 spin_lock(&fs_info->balance_lock);
4360 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4361 spin_unlock(&fs_info->balance_lock);
4363 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4364 return PTR_ERR_OR_ZERO(tsk);
4367 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4369 struct btrfs_balance_control *bctl;
4370 struct btrfs_balance_item *item;
4371 struct btrfs_disk_balance_args disk_bargs;
4372 struct btrfs_path *path;
4373 struct extent_buffer *leaf;
4374 struct btrfs_key key;
4377 path = btrfs_alloc_path();
4381 key.objectid = BTRFS_BALANCE_OBJECTID;
4382 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4385 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4388 if (ret > 0) { /* ret = -ENOENT; */
4393 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4399 leaf = path->nodes[0];
4400 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4402 bctl->flags = btrfs_balance_flags(leaf, item);
4403 bctl->flags |= BTRFS_BALANCE_RESUME;
4405 btrfs_balance_data(leaf, item, &disk_bargs);
4406 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4407 btrfs_balance_meta(leaf, item, &disk_bargs);
4408 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4409 btrfs_balance_sys(leaf, item, &disk_bargs);
4410 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4413 * This should never happen, as the paused balance state is recovered
4414 * during mount without any chance of other exclusive ops to collide.
4416 * This gives the exclusive op status to balance and keeps in paused
4417 * state until user intervention (cancel or umount). If the ownership
4418 * cannot be assigned, show a message but do not fail. The balance
4419 * is in a paused state and must have fs_info::balance_ctl properly
4422 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE))
4424 "balance: cannot set exclusive op status, resume manually");
4426 btrfs_release_path(path);
4428 mutex_lock(&fs_info->balance_mutex);
4429 BUG_ON(fs_info->balance_ctl);
4430 spin_lock(&fs_info->balance_lock);
4431 fs_info->balance_ctl = bctl;
4432 spin_unlock(&fs_info->balance_lock);
4433 mutex_unlock(&fs_info->balance_mutex);
4435 btrfs_free_path(path);
4439 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4443 mutex_lock(&fs_info->balance_mutex);
4444 if (!fs_info->balance_ctl) {
4445 mutex_unlock(&fs_info->balance_mutex);
4449 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4450 atomic_inc(&fs_info->balance_pause_req);
4451 mutex_unlock(&fs_info->balance_mutex);
4453 wait_event(fs_info->balance_wait_q,
4454 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4456 mutex_lock(&fs_info->balance_mutex);
4457 /* we are good with balance_ctl ripped off from under us */
4458 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4459 atomic_dec(&fs_info->balance_pause_req);
4464 mutex_unlock(&fs_info->balance_mutex);
4468 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4470 mutex_lock(&fs_info->balance_mutex);
4471 if (!fs_info->balance_ctl) {
4472 mutex_unlock(&fs_info->balance_mutex);
4477 * A paused balance with the item stored on disk can be resumed at
4478 * mount time if the mount is read-write. Otherwise it's still paused
4479 * and we must not allow cancelling as it deletes the item.
4481 if (sb_rdonly(fs_info->sb)) {
4482 mutex_unlock(&fs_info->balance_mutex);
4486 atomic_inc(&fs_info->balance_cancel_req);
4488 * if we are running just wait and return, balance item is
4489 * deleted in btrfs_balance in this case
4491 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4492 mutex_unlock(&fs_info->balance_mutex);
4493 wait_event(fs_info->balance_wait_q,
4494 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4495 mutex_lock(&fs_info->balance_mutex);
4497 mutex_unlock(&fs_info->balance_mutex);
4499 * Lock released to allow other waiters to continue, we'll
4500 * reexamine the status again.
4502 mutex_lock(&fs_info->balance_mutex);
4504 if (fs_info->balance_ctl) {
4505 reset_balance_state(fs_info);
4506 btrfs_exclop_finish(fs_info);
4507 btrfs_info(fs_info, "balance: canceled");
4511 BUG_ON(fs_info->balance_ctl ||
4512 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4513 atomic_dec(&fs_info->balance_cancel_req);
4514 mutex_unlock(&fs_info->balance_mutex);
4518 int btrfs_uuid_scan_kthread(void *data)
4520 struct btrfs_fs_info *fs_info = data;
4521 struct btrfs_root *root = fs_info->tree_root;
4522 struct btrfs_key key;
4523 struct btrfs_path *path = NULL;
4525 struct extent_buffer *eb;
4527 struct btrfs_root_item root_item;
4529 struct btrfs_trans_handle *trans = NULL;
4530 bool closing = false;
4532 path = btrfs_alloc_path();
4539 key.type = BTRFS_ROOT_ITEM_KEY;
4543 if (btrfs_fs_closing(fs_info)) {
4547 ret = btrfs_search_forward(root, &key, path,
4548 BTRFS_OLDEST_GENERATION);
4555 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4556 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4557 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4558 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4561 eb = path->nodes[0];
4562 slot = path->slots[0];
4563 item_size = btrfs_item_size_nr(eb, slot);
4564 if (item_size < sizeof(root_item))
4567 read_extent_buffer(eb, &root_item,
4568 btrfs_item_ptr_offset(eb, slot),
4569 (int)sizeof(root_item));
4570 if (btrfs_root_refs(&root_item) == 0)
4573 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4574 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4578 btrfs_release_path(path);
4580 * 1 - subvol uuid item
4581 * 1 - received_subvol uuid item
4583 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4584 if (IS_ERR(trans)) {
4585 ret = PTR_ERR(trans);
4593 btrfs_release_path(path);
4594 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4595 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4596 BTRFS_UUID_KEY_SUBVOL,
4599 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4605 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4606 ret = btrfs_uuid_tree_add(trans,
4607 root_item.received_uuid,
4608 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4611 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4618 btrfs_release_path(path);
4620 ret = btrfs_end_transaction(trans);
4626 if (key.offset < (u64)-1) {
4628 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4630 key.type = BTRFS_ROOT_ITEM_KEY;
4631 } else if (key.objectid < (u64)-1) {
4633 key.type = BTRFS_ROOT_ITEM_KEY;
4642 btrfs_free_path(path);
4643 if (trans && !IS_ERR(trans))
4644 btrfs_end_transaction(trans);
4646 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4648 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4649 up(&fs_info->uuid_tree_rescan_sem);
4653 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4655 struct btrfs_trans_handle *trans;
4656 struct btrfs_root *tree_root = fs_info->tree_root;
4657 struct btrfs_root *uuid_root;
4658 struct task_struct *task;
4665 trans = btrfs_start_transaction(tree_root, 2);
4667 return PTR_ERR(trans);
4669 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4670 if (IS_ERR(uuid_root)) {
4671 ret = PTR_ERR(uuid_root);
4672 btrfs_abort_transaction(trans, ret);
4673 btrfs_end_transaction(trans);
4677 fs_info->uuid_root = uuid_root;
4679 ret = btrfs_commit_transaction(trans);
4683 down(&fs_info->uuid_tree_rescan_sem);
4684 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4686 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4687 btrfs_warn(fs_info, "failed to start uuid_scan task");
4688 up(&fs_info->uuid_tree_rescan_sem);
4689 return PTR_ERR(task);
4696 * shrinking a device means finding all of the device extents past
4697 * the new size, and then following the back refs to the chunks.
4698 * The chunk relocation code actually frees the device extent
4700 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4702 struct btrfs_fs_info *fs_info = device->fs_info;
4703 struct btrfs_root *root = fs_info->dev_root;
4704 struct btrfs_trans_handle *trans;
4705 struct btrfs_dev_extent *dev_extent = NULL;
4706 struct btrfs_path *path;
4712 bool retried = false;
4713 struct extent_buffer *l;
4714 struct btrfs_key key;
4715 struct btrfs_super_block *super_copy = fs_info->super_copy;
4716 u64 old_total = btrfs_super_total_bytes(super_copy);
4717 u64 old_size = btrfs_device_get_total_bytes(device);
4721 new_size = round_down(new_size, fs_info->sectorsize);
4723 diff = round_down(old_size - new_size, fs_info->sectorsize);
4725 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4728 path = btrfs_alloc_path();
4732 path->reada = READA_BACK;
4734 trans = btrfs_start_transaction(root, 0);
4735 if (IS_ERR(trans)) {
4736 btrfs_free_path(path);
4737 return PTR_ERR(trans);
4740 mutex_lock(&fs_info->chunk_mutex);
4742 btrfs_device_set_total_bytes(device, new_size);
4743 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4744 device->fs_devices->total_rw_bytes -= diff;
4745 atomic64_sub(diff, &fs_info->free_chunk_space);
4749 * Once the device's size has been set to the new size, ensure all
4750 * in-memory chunks are synced to disk so that the loop below sees them
4751 * and relocates them accordingly.
4753 if (contains_pending_extent(device, &start, diff)) {
4754 mutex_unlock(&fs_info->chunk_mutex);
4755 ret = btrfs_commit_transaction(trans);
4759 mutex_unlock(&fs_info->chunk_mutex);
4760 btrfs_end_transaction(trans);
4764 key.objectid = device->devid;
4765 key.offset = (u64)-1;
4766 key.type = BTRFS_DEV_EXTENT_KEY;
4769 mutex_lock(&fs_info->reclaim_bgs_lock);
4770 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4772 mutex_unlock(&fs_info->reclaim_bgs_lock);
4776 ret = btrfs_previous_item(root, path, 0, key.type);
4778 mutex_unlock(&fs_info->reclaim_bgs_lock);
4782 btrfs_release_path(path);
4787 slot = path->slots[0];
4788 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4790 if (key.objectid != device->devid) {
4791 mutex_unlock(&fs_info->reclaim_bgs_lock);
4792 btrfs_release_path(path);
4796 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4797 length = btrfs_dev_extent_length(l, dev_extent);
4799 if (key.offset + length <= new_size) {
4800 mutex_unlock(&fs_info->reclaim_bgs_lock);
4801 btrfs_release_path(path);
4805 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4806 btrfs_release_path(path);
4809 * We may be relocating the only data chunk we have,
4810 * which could potentially end up with losing data's
4811 * raid profile, so lets allocate an empty one in
4814 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4816 mutex_unlock(&fs_info->reclaim_bgs_lock);
4820 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4821 mutex_unlock(&fs_info->reclaim_bgs_lock);
4822 if (ret == -ENOSPC) {
4825 if (ret == -ETXTBSY) {
4827 "could not shrink block group %llu due to active swapfile",
4832 } while (key.offset-- > 0);
4834 if (failed && !retried) {
4838 } else if (failed && retried) {
4843 /* Shrinking succeeded, else we would be at "done". */
4844 trans = btrfs_start_transaction(root, 0);
4845 if (IS_ERR(trans)) {
4846 ret = PTR_ERR(trans);
4850 mutex_lock(&fs_info->chunk_mutex);
4851 /* Clear all state bits beyond the shrunk device size */
4852 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
4855 btrfs_device_set_disk_total_bytes(device, new_size);
4856 if (list_empty(&device->post_commit_list))
4857 list_add_tail(&device->post_commit_list,
4858 &trans->transaction->dev_update_list);
4860 WARN_ON(diff > old_total);
4861 btrfs_set_super_total_bytes(super_copy,
4862 round_down(old_total - diff, fs_info->sectorsize));
4863 mutex_unlock(&fs_info->chunk_mutex);
4865 /* Now btrfs_update_device() will change the on-disk size. */
4866 ret = btrfs_update_device(trans, device);
4868 btrfs_abort_transaction(trans, ret);
4869 btrfs_end_transaction(trans);
4871 ret = btrfs_commit_transaction(trans);
4874 btrfs_free_path(path);
4876 mutex_lock(&fs_info->chunk_mutex);
4877 btrfs_device_set_total_bytes(device, old_size);
4878 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4879 device->fs_devices->total_rw_bytes += diff;
4880 atomic64_add(diff, &fs_info->free_chunk_space);
4881 mutex_unlock(&fs_info->chunk_mutex);
4886 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4887 struct btrfs_key *key,
4888 struct btrfs_chunk *chunk, int item_size)
4890 struct btrfs_super_block *super_copy = fs_info->super_copy;
4891 struct btrfs_disk_key disk_key;
4895 lockdep_assert_held(&fs_info->chunk_mutex);
4897 array_size = btrfs_super_sys_array_size(super_copy);
4898 if (array_size + item_size + sizeof(disk_key)
4899 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
4902 ptr = super_copy->sys_chunk_array + array_size;
4903 btrfs_cpu_key_to_disk(&disk_key, key);
4904 memcpy(ptr, &disk_key, sizeof(disk_key));
4905 ptr += sizeof(disk_key);
4906 memcpy(ptr, chunk, item_size);
4907 item_size += sizeof(disk_key);
4908 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4914 * sort the devices in descending order by max_avail, total_avail
4916 static int btrfs_cmp_device_info(const void *a, const void *b)
4918 const struct btrfs_device_info *di_a = a;
4919 const struct btrfs_device_info *di_b = b;
4921 if (di_a->max_avail > di_b->max_avail)
4923 if (di_a->max_avail < di_b->max_avail)
4925 if (di_a->total_avail > di_b->total_avail)
4927 if (di_a->total_avail < di_b->total_avail)
4932 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4934 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4937 btrfs_set_fs_incompat(info, RAID56);
4940 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
4942 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
4945 btrfs_set_fs_incompat(info, RAID1C34);
4949 * Structure used internally for __btrfs_alloc_chunk() function.
4950 * Wraps needed parameters.
4952 struct alloc_chunk_ctl {
4955 /* Total number of stripes to allocate */
4957 /* sub_stripes info for map */
4959 /* Stripes per device */
4961 /* Maximum number of devices to use */
4963 /* Minimum number of devices to use */
4965 /* ndevs has to be a multiple of this */
4967 /* Number of copies */
4969 /* Number of stripes worth of bytes to store parity information */
4971 u64 max_stripe_size;
4979 static void init_alloc_chunk_ctl_policy_regular(
4980 struct btrfs_fs_devices *fs_devices,
4981 struct alloc_chunk_ctl *ctl)
4983 u64 type = ctl->type;
4985 if (type & BTRFS_BLOCK_GROUP_DATA) {
4986 ctl->max_stripe_size = SZ_1G;
4987 ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
4988 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4989 /* For larger filesystems, use larger metadata chunks */
4990 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4991 ctl->max_stripe_size = SZ_1G;
4993 ctl->max_stripe_size = SZ_256M;
4994 ctl->max_chunk_size = ctl->max_stripe_size;
4995 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4996 ctl->max_stripe_size = SZ_32M;
4997 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
4998 ctl->devs_max = min_t(int, ctl->devs_max,
4999 BTRFS_MAX_DEVS_SYS_CHUNK);
5004 /* We don't want a chunk larger than 10% of writable space */
5005 ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
5006 ctl->max_chunk_size);
5007 ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes;
5010 static void init_alloc_chunk_ctl_policy_zoned(
5011 struct btrfs_fs_devices *fs_devices,
5012 struct alloc_chunk_ctl *ctl)
5014 u64 zone_size = fs_devices->fs_info->zone_size;
5016 int min_num_stripes = ctl->devs_min * ctl->dev_stripes;
5017 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;
5018 u64 min_chunk_size = min_data_stripes * zone_size;
5019 u64 type = ctl->type;
5021 ctl->max_stripe_size = zone_size;
5022 if (type & BTRFS_BLOCK_GROUP_DATA) {
5023 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,
5025 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5026 ctl->max_chunk_size = ctl->max_stripe_size;
5027 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5028 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5029 ctl->devs_max = min_t(int, ctl->devs_max,
5030 BTRFS_MAX_DEVS_SYS_CHUNK);
5035 /* We don't want a chunk larger than 10% of writable space */
5036 limit = max(round_down(div_factor(fs_devices->total_rw_bytes, 1),
5039 ctl->max_chunk_size = min(limit, ctl->max_chunk_size);
5040 ctl->dev_extent_min = zone_size * ctl->dev_stripes;
5043 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
5044 struct alloc_chunk_ctl *ctl)
5046 int index = btrfs_bg_flags_to_raid_index(ctl->type);
5048 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
5049 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
5050 ctl->devs_max = btrfs_raid_array[index].devs_max;
5052 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
5053 ctl->devs_min = btrfs_raid_array[index].devs_min;
5054 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
5055 ctl->ncopies = btrfs_raid_array[index].ncopies;
5056 ctl->nparity = btrfs_raid_array[index].nparity;
5059 switch (fs_devices->chunk_alloc_policy) {
5060 case BTRFS_CHUNK_ALLOC_REGULAR:
5061 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
5063 case BTRFS_CHUNK_ALLOC_ZONED:
5064 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);
5071 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
5072 struct alloc_chunk_ctl *ctl,
5073 struct btrfs_device_info *devices_info)
5075 struct btrfs_fs_info *info = fs_devices->fs_info;
5076 struct btrfs_device *device;
5078 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
5085 * in the first pass through the devices list, we gather information
5086 * about the available holes on each device.
5088 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5089 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5091 "BTRFS: read-only device in alloc_list\n");
5095 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5096 &device->dev_state) ||
5097 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5100 if (device->total_bytes > device->bytes_used)
5101 total_avail = device->total_bytes - device->bytes_used;
5105 /* If there is no space on this device, skip it. */
5106 if (total_avail < ctl->dev_extent_min)
5109 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
5111 if (ret && ret != -ENOSPC)
5115 max_avail = dev_extent_want;
5117 if (max_avail < ctl->dev_extent_min) {
5118 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5120 "%s: devid %llu has no free space, have=%llu want=%llu",
5121 __func__, device->devid, max_avail,
5122 ctl->dev_extent_min);
5126 if (ndevs == fs_devices->rw_devices) {
5127 WARN(1, "%s: found more than %llu devices\n",
5128 __func__, fs_devices->rw_devices);
5131 devices_info[ndevs].dev_offset = dev_offset;
5132 devices_info[ndevs].max_avail = max_avail;
5133 devices_info[ndevs].total_avail = total_avail;
5134 devices_info[ndevs].dev = device;
5140 * now sort the devices by hole size / available space
5142 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5143 btrfs_cmp_device_info, NULL);
5148 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5149 struct btrfs_device_info *devices_info)
5151 /* Number of stripes that count for block group size */
5155 * The primary goal is to maximize the number of stripes, so use as
5156 * many devices as possible, even if the stripes are not maximum sized.
5158 * The DUP profile stores more than one stripe per device, the
5159 * max_avail is the total size so we have to adjust.
5161 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5163 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5165 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5166 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5169 * Use the number of data stripes to figure out how big this chunk is
5170 * really going to be in terms of logical address space, and compare
5171 * that answer with the max chunk size. If it's higher, we try to
5172 * reduce stripe_size.
5174 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5176 * Reduce stripe_size, round it up to a 16MB boundary again and
5177 * then use it, unless it ends up being even bigger than the
5178 * previous value we had already.
5180 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5181 data_stripes), SZ_16M),
5185 /* Align to BTRFS_STRIPE_LEN */
5186 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5187 ctl->chunk_size = ctl->stripe_size * data_stripes;
5192 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,
5193 struct btrfs_device_info *devices_info)
5195 u64 zone_size = devices_info[0].dev->zone_info->zone_size;
5196 /* Number of stripes that count for block group size */
5200 * It should hold because:
5201 * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5203 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);
5205 ctl->stripe_size = zone_size;
5206 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5207 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5209 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5210 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5211 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,
5212 ctl->stripe_size) + ctl->nparity,
5214 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5215 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5216 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);
5219 ctl->chunk_size = ctl->stripe_size * data_stripes;
5224 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5225 struct alloc_chunk_ctl *ctl,
5226 struct btrfs_device_info *devices_info)
5228 struct btrfs_fs_info *info = fs_devices->fs_info;
5231 * Round down to number of usable stripes, devs_increment can be any
5232 * number so we can't use round_down() that requires power of 2, while
5233 * rounddown is safe.
5235 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5237 if (ctl->ndevs < ctl->devs_min) {
5238 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5240 "%s: not enough devices with free space: have=%d minimum required=%d",
5241 __func__, ctl->ndevs, ctl->devs_min);
5246 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5248 switch (fs_devices->chunk_alloc_policy) {
5249 case BTRFS_CHUNK_ALLOC_REGULAR:
5250 return decide_stripe_size_regular(ctl, devices_info);
5251 case BTRFS_CHUNK_ALLOC_ZONED:
5252 return decide_stripe_size_zoned(ctl, devices_info);
5258 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
5259 struct alloc_chunk_ctl *ctl,
5260 struct btrfs_device_info *devices_info)
5262 struct btrfs_fs_info *info = trans->fs_info;
5263 struct map_lookup *map = NULL;
5264 struct extent_map_tree *em_tree;
5265 struct btrfs_block_group *block_group;
5266 struct extent_map *em;
5267 u64 start = ctl->start;
5268 u64 type = ctl->type;
5273 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
5275 return ERR_PTR(-ENOMEM);
5276 map->num_stripes = ctl->num_stripes;
5278 for (i = 0; i < ctl->ndevs; ++i) {
5279 for (j = 0; j < ctl->dev_stripes; ++j) {
5280 int s = i * ctl->dev_stripes + j;
5281 map->stripes[s].dev = devices_info[i].dev;
5282 map->stripes[s].physical = devices_info[i].dev_offset +
5283 j * ctl->stripe_size;
5286 map->stripe_len = BTRFS_STRIPE_LEN;
5287 map->io_align = BTRFS_STRIPE_LEN;
5288 map->io_width = BTRFS_STRIPE_LEN;
5290 map->sub_stripes = ctl->sub_stripes;
5292 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5294 em = alloc_extent_map();
5297 return ERR_PTR(-ENOMEM);
5299 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5300 em->map_lookup = map;
5302 em->len = ctl->chunk_size;
5303 em->block_start = 0;
5304 em->block_len = em->len;
5305 em->orig_block_len = ctl->stripe_size;
5307 em_tree = &info->mapping_tree;
5308 write_lock(&em_tree->lock);
5309 ret = add_extent_mapping(em_tree, em, 0);
5311 write_unlock(&em_tree->lock);
5312 free_extent_map(em);
5313 return ERR_PTR(ret);
5315 write_unlock(&em_tree->lock);
5317 block_group = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size);
5318 if (IS_ERR(block_group))
5319 goto error_del_extent;
5321 for (i = 0; i < map->num_stripes; i++) {
5322 struct btrfs_device *dev = map->stripes[i].dev;
5324 btrfs_device_set_bytes_used(dev,
5325 dev->bytes_used + ctl->stripe_size);
5326 if (list_empty(&dev->post_commit_list))
5327 list_add_tail(&dev->post_commit_list,
5328 &trans->transaction->dev_update_list);
5331 atomic64_sub(ctl->stripe_size * map->num_stripes,
5332 &info->free_chunk_space);
5334 free_extent_map(em);
5335 check_raid56_incompat_flag(info, type);
5336 check_raid1c34_incompat_flag(info, type);
5341 write_lock(&em_tree->lock);
5342 remove_extent_mapping(em_tree, em);
5343 write_unlock(&em_tree->lock);
5345 /* One for our allocation */
5346 free_extent_map(em);
5347 /* One for the tree reference */
5348 free_extent_map(em);
5353 struct btrfs_block_group *btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
5356 struct btrfs_fs_info *info = trans->fs_info;
5357 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5358 struct btrfs_device_info *devices_info = NULL;
5359 struct alloc_chunk_ctl ctl;
5360 struct btrfs_block_group *block_group;
5363 lockdep_assert_held(&info->chunk_mutex);
5365 if (!alloc_profile_is_valid(type, 0)) {
5367 return ERR_PTR(-EINVAL);
5370 if (list_empty(&fs_devices->alloc_list)) {
5371 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5372 btrfs_debug(info, "%s: no writable device", __func__);
5373 return ERR_PTR(-ENOSPC);
5376 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5377 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5379 return ERR_PTR(-EINVAL);
5382 ctl.start = find_next_chunk(info);
5384 init_alloc_chunk_ctl(fs_devices, &ctl);
5386 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5389 return ERR_PTR(-ENOMEM);
5391 ret = gather_device_info(fs_devices, &ctl, devices_info);
5393 block_group = ERR_PTR(ret);
5397 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5399 block_group = ERR_PTR(ret);
5403 block_group = create_chunk(trans, &ctl, devices_info);
5406 kfree(devices_info);
5411 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5412 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5415 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5418 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
5419 struct btrfs_block_group *bg)
5421 struct btrfs_fs_info *fs_info = trans->fs_info;
5422 struct btrfs_root *extent_root = fs_info->extent_root;
5423 struct btrfs_root *chunk_root = fs_info->chunk_root;
5424 struct btrfs_key key;
5425 struct btrfs_chunk *chunk;
5426 struct btrfs_stripe *stripe;
5427 struct extent_map *em;
5428 struct map_lookup *map;
5434 * We take the chunk_mutex for 2 reasons:
5436 * 1) Updates and insertions in the chunk btree must be done while holding
5437 * the chunk_mutex, as well as updating the system chunk array in the
5438 * superblock. See the comment on top of btrfs_chunk_alloc() for the
5441 * 2) To prevent races with the final phase of a device replace operation
5442 * that replaces the device object associated with the map's stripes,
5443 * because the device object's id can change at any time during that
5444 * final phase of the device replace operation
5445 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5446 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5447 * which would cause a failure when updating the device item, which does
5448 * not exists, or persisting a stripe of the chunk item with such ID.
5449 * Here we can't use the device_list_mutex because our caller already
5450 * has locked the chunk_mutex, and the final phase of device replace
5451 * acquires both mutexes - first the device_list_mutex and then the
5452 * chunk_mutex. Using any of those two mutexes protects us from a
5453 * concurrent device replace.
5455 lockdep_assert_held(&fs_info->chunk_mutex);
5457 em = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
5460 btrfs_abort_transaction(trans, ret);
5464 map = em->map_lookup;
5465 item_size = btrfs_chunk_item_size(map->num_stripes);
5467 chunk = kzalloc(item_size, GFP_NOFS);
5470 btrfs_abort_transaction(trans, ret);
5474 for (i = 0; i < map->num_stripes; i++) {
5475 struct btrfs_device *device = map->stripes[i].dev;
5477 ret = btrfs_update_device(trans, device);
5482 stripe = &chunk->stripe;
5483 for (i = 0; i < map->num_stripes; i++) {
5484 struct btrfs_device *device = map->stripes[i].dev;
5485 const u64 dev_offset = map->stripes[i].physical;
5487 btrfs_set_stack_stripe_devid(stripe, device->devid);
5488 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5489 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5493 btrfs_set_stack_chunk_length(chunk, bg->length);
5494 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5495 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5496 btrfs_set_stack_chunk_type(chunk, map->type);
5497 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5498 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5499 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5500 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5501 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5503 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5504 key.type = BTRFS_CHUNK_ITEM_KEY;
5505 key.offset = bg->start;
5507 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5511 bg->chunk_item_inserted = 1;
5513 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5514 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5521 free_extent_map(em);
5525 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5527 struct btrfs_fs_info *fs_info = trans->fs_info;
5529 struct btrfs_block_group *meta_bg;
5530 struct btrfs_block_group *sys_bg;
5533 * When adding a new device for sprouting, the seed device is read-only
5534 * so we must first allocate a metadata and a system chunk. But before
5535 * adding the block group items to the extent, device and chunk btrees,
5538 * 1) Create both chunks without doing any changes to the btrees, as
5539 * otherwise we would get -ENOSPC since the block groups from the
5540 * seed device are read-only;
5542 * 2) Add the device item for the new sprout device - finishing the setup
5543 * of a new block group requires updating the device item in the chunk
5544 * btree, so it must exist when we attempt to do it. The previous step
5545 * ensures this does not fail with -ENOSPC.
5547 * After that we can add the block group items to their btrees:
5548 * update existing device item in the chunk btree, add a new block group
5549 * item to the extent btree, add a new chunk item to the chunk btree and
5550 * finally add the new device extent items to the devices btree.
5553 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5554 meta_bg = btrfs_alloc_chunk(trans, alloc_profile);
5555 if (IS_ERR(meta_bg))
5556 return PTR_ERR(meta_bg);
5558 alloc_profile = btrfs_system_alloc_profile(fs_info);
5559 sys_bg = btrfs_alloc_chunk(trans, alloc_profile);
5561 return PTR_ERR(sys_bg);
5566 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5568 const int index = btrfs_bg_flags_to_raid_index(map->type);
5570 return btrfs_raid_array[index].tolerated_failures;
5573 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5575 struct extent_map *em;
5576 struct map_lookup *map;
5581 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5585 map = em->map_lookup;
5586 for (i = 0; i < map->num_stripes; i++) {
5587 if (test_bit(BTRFS_DEV_STATE_MISSING,
5588 &map->stripes[i].dev->dev_state)) {
5592 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5593 &map->stripes[i].dev->dev_state)) {
5600 * If the number of missing devices is larger than max errors,
5601 * we can not write the data into that chunk successfully, so
5604 if (miss_ndevs > btrfs_chunk_max_errors(map))
5607 free_extent_map(em);
5611 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5613 struct extent_map *em;
5616 write_lock(&tree->lock);
5617 em = lookup_extent_mapping(tree, 0, (u64)-1);
5619 remove_extent_mapping(tree, em);
5620 write_unlock(&tree->lock);
5624 free_extent_map(em);
5625 /* once for the tree */
5626 free_extent_map(em);
5630 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5632 struct extent_map *em;
5633 struct map_lookup *map;
5636 em = btrfs_get_chunk_map(fs_info, logical, len);
5639 * We could return errors for these cases, but that could get
5640 * ugly and we'd probably do the same thing which is just not do
5641 * anything else and exit, so return 1 so the callers don't try
5642 * to use other copies.
5646 map = em->map_lookup;
5647 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK))
5648 ret = map->num_stripes;
5649 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5650 ret = map->sub_stripes;
5651 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5653 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5655 * There could be two corrupted data stripes, we need
5656 * to loop retry in order to rebuild the correct data.
5658 * Fail a stripe at a time on every retry except the
5659 * stripe under reconstruction.
5661 ret = map->num_stripes;
5664 free_extent_map(em);
5666 down_read(&fs_info->dev_replace.rwsem);
5667 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5668 fs_info->dev_replace.tgtdev)
5670 up_read(&fs_info->dev_replace.rwsem);
5675 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5678 struct extent_map *em;
5679 struct map_lookup *map;
5680 unsigned long len = fs_info->sectorsize;
5682 em = btrfs_get_chunk_map(fs_info, logical, len);
5684 if (!WARN_ON(IS_ERR(em))) {
5685 map = em->map_lookup;
5686 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5687 len = map->stripe_len * nr_data_stripes(map);
5688 free_extent_map(em);
5693 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5695 struct extent_map *em;
5696 struct map_lookup *map;
5699 em = btrfs_get_chunk_map(fs_info, logical, len);
5701 if(!WARN_ON(IS_ERR(em))) {
5702 map = em->map_lookup;
5703 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5705 free_extent_map(em);
5710 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5711 struct map_lookup *map, int first,
5712 int dev_replace_is_ongoing)
5716 int preferred_mirror;
5718 struct btrfs_device *srcdev;
5721 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5723 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5724 num_stripes = map->sub_stripes;
5726 num_stripes = map->num_stripes;
5728 switch (fs_info->fs_devices->read_policy) {
5730 /* Shouldn't happen, just warn and use pid instead of failing */
5731 btrfs_warn_rl(fs_info,
5732 "unknown read_policy type %u, reset to pid",
5733 fs_info->fs_devices->read_policy);
5734 fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID;
5736 case BTRFS_READ_POLICY_PID:
5737 preferred_mirror = first + (current->pid % num_stripes);
5741 if (dev_replace_is_ongoing &&
5742 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5743 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5744 srcdev = fs_info->dev_replace.srcdev;
5749 * try to avoid the drive that is the source drive for a
5750 * dev-replace procedure, only choose it if no other non-missing
5751 * mirror is available
5753 for (tolerance = 0; tolerance < 2; tolerance++) {
5754 if (map->stripes[preferred_mirror].dev->bdev &&
5755 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5756 return preferred_mirror;
5757 for (i = first; i < first + num_stripes; i++) {
5758 if (map->stripes[i].dev->bdev &&
5759 (tolerance || map->stripes[i].dev != srcdev))
5764 /* we couldn't find one that doesn't fail. Just return something
5765 * and the io error handling code will clean up eventually
5767 return preferred_mirror;
5770 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5771 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5778 for (i = 0; i < num_stripes - 1; i++) {
5779 /* Swap if parity is on a smaller index */
5780 if (bbio->raid_map[i] > bbio->raid_map[i + 1]) {
5781 swap(bbio->stripes[i], bbio->stripes[i + 1]);
5782 swap(bbio->raid_map[i], bbio->raid_map[i + 1]);
5789 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5791 struct btrfs_bio *bbio = kzalloc(
5792 /* the size of the btrfs_bio */
5793 sizeof(struct btrfs_bio) +
5794 /* plus the variable array for the stripes */
5795 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5796 /* plus the variable array for the tgt dev */
5797 sizeof(int) * (real_stripes) +
5799 * plus the raid_map, which includes both the tgt dev
5802 sizeof(u64) * (total_stripes),
5803 GFP_NOFS|__GFP_NOFAIL);
5805 atomic_set(&bbio->error, 0);
5806 refcount_set(&bbio->refs, 1);
5808 bbio->tgtdev_map = (int *)(bbio->stripes + total_stripes);
5809 bbio->raid_map = (u64 *)(bbio->tgtdev_map + real_stripes);
5814 void btrfs_get_bbio(struct btrfs_bio *bbio)
5816 WARN_ON(!refcount_read(&bbio->refs));
5817 refcount_inc(&bbio->refs);
5820 void btrfs_put_bbio(struct btrfs_bio *bbio)
5824 if (refcount_dec_and_test(&bbio->refs))
5828 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5830 * Please note that, discard won't be sent to target device of device
5833 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5834 u64 logical, u64 *length_ret,
5835 struct btrfs_bio **bbio_ret)
5837 struct extent_map *em;
5838 struct map_lookup *map;
5839 struct btrfs_bio *bbio;
5840 u64 length = *length_ret;
5844 u64 stripe_end_offset;
5851 u32 sub_stripes = 0;
5852 u64 stripes_per_dev = 0;
5853 u32 remaining_stripes = 0;
5854 u32 last_stripe = 0;
5858 /* discard always return a bbio */
5861 em = btrfs_get_chunk_map(fs_info, logical, length);
5865 map = em->map_lookup;
5866 /* we don't discard raid56 yet */
5867 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5872 offset = logical - em->start;
5873 length = min_t(u64, em->start + em->len - logical, length);
5874 *length_ret = length;
5876 stripe_len = map->stripe_len;
5878 * stripe_nr counts the total number of stripes we have to stride
5879 * to get to this block
5881 stripe_nr = div64_u64(offset, stripe_len);
5883 /* stripe_offset is the offset of this block in its stripe */
5884 stripe_offset = offset - stripe_nr * stripe_len;
5886 stripe_nr_end = round_up(offset + length, map->stripe_len);
5887 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5888 stripe_cnt = stripe_nr_end - stripe_nr;
5889 stripe_end_offset = stripe_nr_end * map->stripe_len -
5892 * after this, stripe_nr is the number of stripes on this
5893 * device we have to walk to find the data, and stripe_index is
5894 * the number of our device in the stripe array
5898 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5899 BTRFS_BLOCK_GROUP_RAID10)) {
5900 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5903 sub_stripes = map->sub_stripes;
5905 factor = map->num_stripes / sub_stripes;
5906 num_stripes = min_t(u64, map->num_stripes,
5907 sub_stripes * stripe_cnt);
5908 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5909 stripe_index *= sub_stripes;
5910 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5911 &remaining_stripes);
5912 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5913 last_stripe *= sub_stripes;
5914 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
5915 BTRFS_BLOCK_GROUP_DUP)) {
5916 num_stripes = map->num_stripes;
5918 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5922 bbio = alloc_btrfs_bio(num_stripes, 0);
5928 for (i = 0; i < num_stripes; i++) {
5929 bbio->stripes[i].physical =
5930 map->stripes[stripe_index].physical +
5931 stripe_offset + stripe_nr * map->stripe_len;
5932 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5934 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5935 BTRFS_BLOCK_GROUP_RAID10)) {
5936 bbio->stripes[i].length = stripes_per_dev *
5939 if (i / sub_stripes < remaining_stripes)
5940 bbio->stripes[i].length +=
5944 * Special for the first stripe and
5947 * |-------|...|-------|
5951 if (i < sub_stripes)
5952 bbio->stripes[i].length -=
5955 if (stripe_index >= last_stripe &&
5956 stripe_index <= (last_stripe +
5958 bbio->stripes[i].length -=
5961 if (i == sub_stripes - 1)
5964 bbio->stripes[i].length = length;
5968 if (stripe_index == map->num_stripes) {
5975 bbio->map_type = map->type;
5976 bbio->num_stripes = num_stripes;
5978 free_extent_map(em);
5983 * In dev-replace case, for repair case (that's the only case where the mirror
5984 * is selected explicitly when calling btrfs_map_block), blocks left of the
5985 * left cursor can also be read from the target drive.
5987 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5989 * For READ, it also needs to be supported using the same mirror number.
5991 * If the requested block is not left of the left cursor, EIO is returned. This
5992 * can happen because btrfs_num_copies() returns one more in the dev-replace
5995 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5996 u64 logical, u64 length,
5997 u64 srcdev_devid, int *mirror_num,
6000 struct btrfs_bio *bbio = NULL;
6002 int index_srcdev = 0;
6004 u64 physical_of_found = 0;
6008 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
6009 logical, &length, &bbio, 0, 0);
6011 ASSERT(bbio == NULL);
6015 num_stripes = bbio->num_stripes;
6016 if (*mirror_num > num_stripes) {
6018 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
6019 * that means that the requested area is not left of the left
6022 btrfs_put_bbio(bbio);
6027 * process the rest of the function using the mirror_num of the source
6028 * drive. Therefore look it up first. At the end, patch the device
6029 * pointer to the one of the target drive.
6031 for (i = 0; i < num_stripes; i++) {
6032 if (bbio->stripes[i].dev->devid != srcdev_devid)
6036 * In case of DUP, in order to keep it simple, only add the
6037 * mirror with the lowest physical address
6040 physical_of_found <= bbio->stripes[i].physical)
6045 physical_of_found = bbio->stripes[i].physical;
6048 btrfs_put_bbio(bbio);
6054 *mirror_num = index_srcdev + 1;
6055 *physical = physical_of_found;
6059 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)
6061 struct btrfs_block_group *cache;
6064 /* Non zoned filesystem does not use "to_copy" flag */
6065 if (!btrfs_is_zoned(fs_info))
6068 cache = btrfs_lookup_block_group(fs_info, logical);
6070 spin_lock(&cache->lock);
6071 ret = cache->to_copy;
6072 spin_unlock(&cache->lock);
6074 btrfs_put_block_group(cache);
6078 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
6079 struct btrfs_bio **bbio_ret,
6080 struct btrfs_dev_replace *dev_replace,
6082 int *num_stripes_ret, int *max_errors_ret)
6084 struct btrfs_bio *bbio = *bbio_ret;
6085 u64 srcdev_devid = dev_replace->srcdev->devid;
6086 int tgtdev_indexes = 0;
6087 int num_stripes = *num_stripes_ret;
6088 int max_errors = *max_errors_ret;
6091 if (op == BTRFS_MAP_WRITE) {
6092 int index_where_to_add;
6095 * A block group which have "to_copy" set will eventually
6096 * copied by dev-replace process. We can avoid cloning IO here.
6098 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))
6102 * duplicate the write operations while the dev replace
6103 * procedure is running. Since the copying of the old disk to
6104 * the new disk takes place at run time while the filesystem is
6105 * mounted writable, the regular write operations to the old
6106 * disk have to be duplicated to go to the new disk as well.
6108 * Note that device->missing is handled by the caller, and that
6109 * the write to the old disk is already set up in the stripes
6112 index_where_to_add = num_stripes;
6113 for (i = 0; i < num_stripes; i++) {
6114 if (bbio->stripes[i].dev->devid == srcdev_devid) {
6115 /* write to new disk, too */
6116 struct btrfs_bio_stripe *new =
6117 bbio->stripes + index_where_to_add;
6118 struct btrfs_bio_stripe *old =
6121 new->physical = old->physical;
6122 new->length = old->length;
6123 new->dev = dev_replace->tgtdev;
6124 bbio->tgtdev_map[i] = index_where_to_add;
6125 index_where_to_add++;
6130 num_stripes = index_where_to_add;
6131 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
6132 int index_srcdev = 0;
6134 u64 physical_of_found = 0;
6137 * During the dev-replace procedure, the target drive can also
6138 * be used to read data in case it is needed to repair a corrupt
6139 * block elsewhere. This is possible if the requested area is
6140 * left of the left cursor. In this area, the target drive is a
6141 * full copy of the source drive.
6143 for (i = 0; i < num_stripes; i++) {
6144 if (bbio->stripes[i].dev->devid == srcdev_devid) {
6146 * In case of DUP, in order to keep it simple,
6147 * only add the mirror with the lowest physical
6151 physical_of_found <=
6152 bbio->stripes[i].physical)
6156 physical_of_found = bbio->stripes[i].physical;
6160 struct btrfs_bio_stripe *tgtdev_stripe =
6161 bbio->stripes + num_stripes;
6163 tgtdev_stripe->physical = physical_of_found;
6164 tgtdev_stripe->length =
6165 bbio->stripes[index_srcdev].length;
6166 tgtdev_stripe->dev = dev_replace->tgtdev;
6167 bbio->tgtdev_map[index_srcdev] = num_stripes;
6174 *num_stripes_ret = num_stripes;
6175 *max_errors_ret = max_errors;
6176 bbio->num_tgtdevs = tgtdev_indexes;
6180 static bool need_full_stripe(enum btrfs_map_op op)
6182 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
6186 * Calculate the geometry of a particular (address, len) tuple. This
6187 * information is used to calculate how big a particular bio can get before it
6188 * straddles a stripe.
6190 * @fs_info: the filesystem
6191 * @em: mapping containing the logical extent
6192 * @op: type of operation - write or read
6193 * @logical: address that we want to figure out the geometry of
6194 * @io_geom: pointer used to return values
6196 * Returns < 0 in case a chunk for the given logical address cannot be found,
6197 * usually shouldn't happen unless @logical is corrupted, 0 otherwise.
6199 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, struct extent_map *em,
6200 enum btrfs_map_op op, u64 logical,
6201 struct btrfs_io_geometry *io_geom)
6203 struct map_lookup *map;
6209 u64 raid56_full_stripe_start = (u64)-1;
6212 ASSERT(op != BTRFS_MAP_DISCARD);
6214 map = em->map_lookup;
6215 /* Offset of this logical address in the chunk */
6216 offset = logical - em->start;
6217 /* Len of a stripe in a chunk */
6218 stripe_len = map->stripe_len;
6219 /* Stripe where this block falls in */
6220 stripe_nr = div64_u64(offset, stripe_len);
6221 /* Offset of stripe in the chunk */
6222 stripe_offset = stripe_nr * stripe_len;
6223 if (offset < stripe_offset) {
6225 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu",
6226 stripe_offset, offset, em->start, logical, stripe_len);
6230 /* stripe_offset is the offset of this block in its stripe */
6231 stripe_offset = offset - stripe_offset;
6232 data_stripes = nr_data_stripes(map);
6234 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
6235 u64 max_len = stripe_len - stripe_offset;
6238 * In case of raid56, we need to know the stripe aligned start
6240 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6241 unsigned long full_stripe_len = stripe_len * data_stripes;
6242 raid56_full_stripe_start = offset;
6245 * Allow a write of a full stripe, but make sure we
6246 * don't allow straddling of stripes
6248 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
6250 raid56_full_stripe_start *= full_stripe_len;
6253 * For writes to RAID[56], allow a full stripeset across
6254 * all disks. For other RAID types and for RAID[56]
6255 * reads, just allow a single stripe (on a single disk).
6257 if (op == BTRFS_MAP_WRITE) {
6258 max_len = stripe_len * data_stripes -
6259 (offset - raid56_full_stripe_start);
6262 len = min_t(u64, em->len - offset, max_len);
6264 len = em->len - offset;
6268 io_geom->offset = offset;
6269 io_geom->stripe_len = stripe_len;
6270 io_geom->stripe_nr = stripe_nr;
6271 io_geom->stripe_offset = stripe_offset;
6272 io_geom->raid56_stripe_offset = raid56_full_stripe_start;
6277 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
6278 enum btrfs_map_op op,
6279 u64 logical, u64 *length,
6280 struct btrfs_bio **bbio_ret,
6281 int mirror_num, int need_raid_map)
6283 struct extent_map *em;
6284 struct map_lookup *map;
6294 int tgtdev_indexes = 0;
6295 struct btrfs_bio *bbio = NULL;
6296 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6297 int dev_replace_is_ongoing = 0;
6298 int num_alloc_stripes;
6299 int patch_the_first_stripe_for_dev_replace = 0;
6300 u64 physical_to_patch_in_first_stripe = 0;
6301 u64 raid56_full_stripe_start = (u64)-1;
6302 struct btrfs_io_geometry geom;
6305 ASSERT(op != BTRFS_MAP_DISCARD);
6307 em = btrfs_get_chunk_map(fs_info, logical, *length);
6308 ASSERT(!IS_ERR(em));
6310 ret = btrfs_get_io_geometry(fs_info, em, op, logical, &geom);
6314 map = em->map_lookup;
6317 stripe_len = geom.stripe_len;
6318 stripe_nr = geom.stripe_nr;
6319 stripe_offset = geom.stripe_offset;
6320 raid56_full_stripe_start = geom.raid56_stripe_offset;
6321 data_stripes = nr_data_stripes(map);
6323 down_read(&dev_replace->rwsem);
6324 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6326 * Hold the semaphore for read during the whole operation, write is
6327 * requested at commit time but must wait.
6329 if (!dev_replace_is_ongoing)
6330 up_read(&dev_replace->rwsem);
6332 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
6333 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
6334 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
6335 dev_replace->srcdev->devid,
6337 &physical_to_patch_in_first_stripe);
6341 patch_the_first_stripe_for_dev_replace = 1;
6342 } else if (mirror_num > map->num_stripes) {
6348 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6349 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6351 if (!need_full_stripe(op))
6353 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6354 if (need_full_stripe(op))
6355 num_stripes = map->num_stripes;
6356 else if (mirror_num)
6357 stripe_index = mirror_num - 1;
6359 stripe_index = find_live_mirror(fs_info, map, 0,
6360 dev_replace_is_ongoing);
6361 mirror_num = stripe_index + 1;
6364 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6365 if (need_full_stripe(op)) {
6366 num_stripes = map->num_stripes;
6367 } else if (mirror_num) {
6368 stripe_index = mirror_num - 1;
6373 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6374 u32 factor = map->num_stripes / map->sub_stripes;
6376 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6377 stripe_index *= map->sub_stripes;
6379 if (need_full_stripe(op))
6380 num_stripes = map->sub_stripes;
6381 else if (mirror_num)
6382 stripe_index += mirror_num - 1;
6384 int old_stripe_index = stripe_index;
6385 stripe_index = find_live_mirror(fs_info, map,
6387 dev_replace_is_ongoing);
6388 mirror_num = stripe_index - old_stripe_index + 1;
6391 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6392 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6393 /* push stripe_nr back to the start of the full stripe */
6394 stripe_nr = div64_u64(raid56_full_stripe_start,
6395 stripe_len * data_stripes);
6397 /* RAID[56] write or recovery. Return all stripes */
6398 num_stripes = map->num_stripes;
6399 max_errors = nr_parity_stripes(map);
6401 *length = map->stripe_len;
6406 * Mirror #0 or #1 means the original data block.
6407 * Mirror #2 is RAID5 parity block.
6408 * Mirror #3 is RAID6 Q block.
6410 stripe_nr = div_u64_rem(stripe_nr,
6411 data_stripes, &stripe_index);
6413 stripe_index = data_stripes + mirror_num - 2;
6415 /* We distribute the parity blocks across stripes */
6416 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
6418 if (!need_full_stripe(op) && mirror_num <= 1)
6423 * after this, stripe_nr is the number of stripes on this
6424 * device we have to walk to find the data, and stripe_index is
6425 * the number of our device in the stripe array
6427 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6429 mirror_num = stripe_index + 1;
6431 if (stripe_index >= map->num_stripes) {
6433 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6434 stripe_index, map->num_stripes);
6439 num_alloc_stripes = num_stripes;
6440 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
6441 if (op == BTRFS_MAP_WRITE)
6442 num_alloc_stripes <<= 1;
6443 if (op == BTRFS_MAP_GET_READ_MIRRORS)
6444 num_alloc_stripes++;
6445 tgtdev_indexes = num_stripes;
6448 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
6454 for (i = 0; i < num_stripes; i++) {
6455 bbio->stripes[i].physical = map->stripes[stripe_index].physical +
6456 stripe_offset + stripe_nr * map->stripe_len;
6457 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
6461 /* build raid_map */
6462 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6463 (need_full_stripe(op) || mirror_num > 1)) {
6467 /* Work out the disk rotation on this stripe-set */
6468 div_u64_rem(stripe_nr, num_stripes, &rot);
6470 /* Fill in the logical address of each stripe */
6471 tmp = stripe_nr * data_stripes;
6472 for (i = 0; i < data_stripes; i++)
6473 bbio->raid_map[(i+rot) % num_stripes] =
6474 em->start + (tmp + i) * map->stripe_len;
6476 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
6477 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
6478 bbio->raid_map[(i+rot+1) % num_stripes] =
6481 sort_parity_stripes(bbio, num_stripes);
6484 if (need_full_stripe(op))
6485 max_errors = btrfs_chunk_max_errors(map);
6487 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6488 need_full_stripe(op)) {
6489 handle_ops_on_dev_replace(op, &bbio, dev_replace, logical,
6490 &num_stripes, &max_errors);
6494 bbio->map_type = map->type;
6495 bbio->num_stripes = num_stripes;
6496 bbio->max_errors = max_errors;
6497 bbio->mirror_num = mirror_num;
6500 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6501 * mirror_num == num_stripes + 1 && dev_replace target drive is
6502 * available as a mirror
6504 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6505 WARN_ON(num_stripes > 1);
6506 bbio->stripes[0].dev = dev_replace->tgtdev;
6507 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
6508 bbio->mirror_num = map->num_stripes + 1;
6511 if (dev_replace_is_ongoing) {
6512 lockdep_assert_held(&dev_replace->rwsem);
6513 /* Unlock and let waiting writers proceed */
6514 up_read(&dev_replace->rwsem);
6516 free_extent_map(em);
6520 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6521 u64 logical, u64 *length,
6522 struct btrfs_bio **bbio_ret, int mirror_num)
6524 if (op == BTRFS_MAP_DISCARD)
6525 return __btrfs_map_block_for_discard(fs_info, logical,
6528 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
6532 /* For Scrub/replace */
6533 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6534 u64 logical, u64 *length,
6535 struct btrfs_bio **bbio_ret)
6537 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
6540 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6542 bio->bi_private = bbio->private;
6543 bio->bi_end_io = bbio->end_io;
6546 btrfs_put_bbio(bbio);
6549 static void btrfs_end_bio(struct bio *bio)
6551 struct btrfs_bio *bbio = bio->bi_private;
6552 int is_orig_bio = 0;
6554 if (bio->bi_status) {
6555 atomic_inc(&bbio->error);
6556 if (bio->bi_status == BLK_STS_IOERR ||
6557 bio->bi_status == BLK_STS_TARGET) {
6558 struct btrfs_device *dev = btrfs_io_bio(bio)->device;
6561 if (btrfs_op(bio) == BTRFS_MAP_WRITE)
6562 btrfs_dev_stat_inc_and_print(dev,
6563 BTRFS_DEV_STAT_WRITE_ERRS);
6564 else if (!(bio->bi_opf & REQ_RAHEAD))
6565 btrfs_dev_stat_inc_and_print(dev,
6566 BTRFS_DEV_STAT_READ_ERRS);
6567 if (bio->bi_opf & REQ_PREFLUSH)
6568 btrfs_dev_stat_inc_and_print(dev,
6569 BTRFS_DEV_STAT_FLUSH_ERRS);
6573 if (bio == bbio->orig_bio)
6576 btrfs_bio_counter_dec(bbio->fs_info);
6578 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6581 bio = bbio->orig_bio;
6584 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6585 /* only send an error to the higher layers if it is
6586 * beyond the tolerance of the btrfs bio
6588 if (atomic_read(&bbio->error) > bbio->max_errors) {
6589 bio->bi_status = BLK_STS_IOERR;
6592 * this bio is actually up to date, we didn't
6593 * go over the max number of errors
6595 bio->bi_status = BLK_STS_OK;
6598 btrfs_end_bbio(bbio, bio);
6599 } else if (!is_orig_bio) {
6604 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6605 u64 physical, struct btrfs_device *dev)
6607 struct btrfs_fs_info *fs_info = bbio->fs_info;
6609 bio->bi_private = bbio;
6610 btrfs_io_bio(bio)->device = dev;
6611 bio->bi_end_io = btrfs_end_bio;
6612 bio->bi_iter.bi_sector = physical >> 9;
6614 * For zone append writing, bi_sector must point the beginning of the
6617 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
6618 if (btrfs_dev_is_sequential(dev, physical)) {
6619 u64 zone_start = round_down(physical, fs_info->zone_size);
6621 bio->bi_iter.bi_sector = zone_start >> SECTOR_SHIFT;
6623 bio->bi_opf &= ~REQ_OP_ZONE_APPEND;
6624 bio->bi_opf |= REQ_OP_WRITE;
6627 btrfs_debug_in_rcu(fs_info,
6628 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6629 bio_op(bio), bio->bi_opf, bio->bi_iter.bi_sector,
6630 (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name),
6631 dev->devid, bio->bi_iter.bi_size);
6632 bio_set_dev(bio, dev->bdev);
6634 btrfs_bio_counter_inc_noblocked(fs_info);
6636 btrfsic_submit_bio(bio);
6639 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6641 atomic_inc(&bbio->error);
6642 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6643 /* Should be the original bio. */
6644 WARN_ON(bio != bbio->orig_bio);
6646 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6647 bio->bi_iter.bi_sector = logical >> 9;
6648 if (atomic_read(&bbio->error) > bbio->max_errors)
6649 bio->bi_status = BLK_STS_IOERR;
6651 bio->bi_status = BLK_STS_OK;
6652 btrfs_end_bbio(bbio, bio);
6656 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6659 struct btrfs_device *dev;
6660 struct bio *first_bio = bio;
6661 u64 logical = bio->bi_iter.bi_sector << 9;
6667 struct btrfs_bio *bbio = NULL;
6669 length = bio->bi_iter.bi_size;
6670 map_length = length;
6672 btrfs_bio_counter_inc_blocked(fs_info);
6673 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6674 &map_length, &bbio, mirror_num, 1);
6676 btrfs_bio_counter_dec(fs_info);
6677 return errno_to_blk_status(ret);
6680 total_devs = bbio->num_stripes;
6681 bbio->orig_bio = first_bio;
6682 bbio->private = first_bio->bi_private;
6683 bbio->end_io = first_bio->bi_end_io;
6684 bbio->fs_info = fs_info;
6685 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6687 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6688 ((btrfs_op(bio) == BTRFS_MAP_WRITE) || (mirror_num > 1))) {
6689 /* In this case, map_length has been set to the length of
6690 a single stripe; not the whole write */
6691 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
6692 ret = raid56_parity_write(fs_info, bio, bbio,
6695 ret = raid56_parity_recover(fs_info, bio, bbio,
6696 map_length, mirror_num, 1);
6699 btrfs_bio_counter_dec(fs_info);
6700 return errno_to_blk_status(ret);
6703 if (map_length < length) {
6705 "mapping failed logical %llu bio len %llu len %llu",
6706 logical, length, map_length);
6710 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6711 dev = bbio->stripes[dev_nr].dev;
6712 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
6714 (btrfs_op(first_bio) == BTRFS_MAP_WRITE &&
6715 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6716 bbio_error(bbio, first_bio, logical);
6720 if (dev_nr < total_devs - 1)
6721 bio = btrfs_bio_clone(first_bio);
6725 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, dev);
6727 btrfs_bio_counter_dec(fs_info);
6732 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6735 * If devid and uuid are both specified, the match must be exact, otherwise
6736 * only devid is used.
6738 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
6739 u64 devid, u8 *uuid, u8 *fsid)
6741 struct btrfs_device *device;
6742 struct btrfs_fs_devices *seed_devs;
6744 if (!fsid || !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6745 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6746 if (device->devid == devid &&
6747 (!uuid || memcmp(device->uuid, uuid,
6748 BTRFS_UUID_SIZE) == 0))
6753 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6755 !memcmp(seed_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6756 list_for_each_entry(device, &seed_devs->devices,
6758 if (device->devid == devid &&
6759 (!uuid || memcmp(device->uuid, uuid,
6760 BTRFS_UUID_SIZE) == 0))
6769 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6770 u64 devid, u8 *dev_uuid)
6772 struct btrfs_device *device;
6773 unsigned int nofs_flag;
6776 * We call this under the chunk_mutex, so we want to use NOFS for this
6777 * allocation, however we don't want to change btrfs_alloc_device() to
6778 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6781 nofs_flag = memalloc_nofs_save();
6782 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6783 memalloc_nofs_restore(nofs_flag);
6787 list_add(&device->dev_list, &fs_devices->devices);
6788 device->fs_devices = fs_devices;
6789 fs_devices->num_devices++;
6791 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6792 fs_devices->missing_devices++;
6798 * btrfs_alloc_device - allocate struct btrfs_device
6799 * @fs_info: used only for generating a new devid, can be NULL if
6800 * devid is provided (i.e. @devid != NULL).
6801 * @devid: a pointer to devid for this device. If NULL a new devid
6803 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6806 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6807 * on error. Returned struct is not linked onto any lists and must be
6808 * destroyed with btrfs_free_device.
6810 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6814 struct btrfs_device *dev;
6817 if (WARN_ON(!devid && !fs_info))
6818 return ERR_PTR(-EINVAL);
6820 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
6822 return ERR_PTR(-ENOMEM);
6825 * Preallocate a bio that's always going to be used for flushing device
6826 * barriers and matches the device lifespan
6828 dev->flush_bio = bio_kmalloc(GFP_KERNEL, 0);
6829 if (!dev->flush_bio) {
6831 return ERR_PTR(-ENOMEM);
6834 INIT_LIST_HEAD(&dev->dev_list);
6835 INIT_LIST_HEAD(&dev->dev_alloc_list);
6836 INIT_LIST_HEAD(&dev->post_commit_list);
6838 atomic_set(&dev->reada_in_flight, 0);
6839 atomic_set(&dev->dev_stats_ccnt, 0);
6840 btrfs_device_data_ordered_init(dev);
6841 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
6842 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
6843 extent_io_tree_init(fs_info, &dev->alloc_state,
6844 IO_TREE_DEVICE_ALLOC_STATE, NULL);
6851 ret = find_next_devid(fs_info, &tmp);
6853 btrfs_free_device(dev);
6854 return ERR_PTR(ret);
6860 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6862 generate_random_uuid(dev->uuid);
6867 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6868 u64 devid, u8 *uuid, bool error)
6871 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6874 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6878 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes)
6880 const int data_stripes = calc_data_stripes(type, num_stripes);
6882 return div_u64(chunk_len, data_stripes);
6885 #if BITS_PER_LONG == 32
6887 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
6888 * can't be accessed on 32bit systems.
6890 * This function do mount time check to reject the fs if it already has
6891 * metadata chunk beyond that limit.
6893 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6894 u64 logical, u64 length, u64 type)
6896 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6899 if (logical + length < MAX_LFS_FILESIZE)
6902 btrfs_err_32bit_limit(fs_info);
6907 * This is to give early warning for any metadata chunk reaching
6908 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
6909 * Although we can still access the metadata, it's not going to be possible
6910 * once the limit is reached.
6912 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6913 u64 logical, u64 length, u64 type)
6915 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6918 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)
6921 btrfs_warn_32bit_limit(fs_info);
6925 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6926 struct btrfs_chunk *chunk)
6928 struct btrfs_fs_info *fs_info = leaf->fs_info;
6929 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6930 struct map_lookup *map;
6931 struct extent_map *em;
6936 u8 uuid[BTRFS_UUID_SIZE];
6941 logical = key->offset;
6942 length = btrfs_chunk_length(leaf, chunk);
6943 type = btrfs_chunk_type(leaf, chunk);
6944 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6946 #if BITS_PER_LONG == 32
6947 ret = check_32bit_meta_chunk(fs_info, logical, length, type);
6950 warn_32bit_meta_chunk(fs_info, logical, length, type);
6954 * Only need to verify chunk item if we're reading from sys chunk array,
6955 * as chunk item in tree block is already verified by tree-checker.
6957 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6958 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6963 read_lock(&map_tree->lock);
6964 em = lookup_extent_mapping(map_tree, logical, 1);
6965 read_unlock(&map_tree->lock);
6967 /* already mapped? */
6968 if (em && em->start <= logical && em->start + em->len > logical) {
6969 free_extent_map(em);
6972 free_extent_map(em);
6975 em = alloc_extent_map();
6978 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6980 free_extent_map(em);
6984 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6985 em->map_lookup = map;
6986 em->start = logical;
6989 em->block_start = 0;
6990 em->block_len = em->len;
6992 map->num_stripes = num_stripes;
6993 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6994 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6995 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6997 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6998 map->verified_stripes = 0;
6999 em->orig_block_len = calc_stripe_length(type, em->len,
7001 for (i = 0; i < num_stripes; i++) {
7002 map->stripes[i].physical =
7003 btrfs_stripe_offset_nr(leaf, chunk, i);
7004 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
7005 read_extent_buffer(leaf, uuid, (unsigned long)
7006 btrfs_stripe_dev_uuid_nr(chunk, i),
7008 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices,
7010 if (!map->stripes[i].dev &&
7011 !btrfs_test_opt(fs_info, DEGRADED)) {
7012 free_extent_map(em);
7013 btrfs_report_missing_device(fs_info, devid, uuid, true);
7016 if (!map->stripes[i].dev) {
7017 map->stripes[i].dev =
7018 add_missing_dev(fs_info->fs_devices, devid,
7020 if (IS_ERR(map->stripes[i].dev)) {
7021 free_extent_map(em);
7023 "failed to init missing dev %llu: %ld",
7024 devid, PTR_ERR(map->stripes[i].dev));
7025 return PTR_ERR(map->stripes[i].dev);
7027 btrfs_report_missing_device(fs_info, devid, uuid, false);
7029 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
7030 &(map->stripes[i].dev->dev_state));
7034 write_lock(&map_tree->lock);
7035 ret = add_extent_mapping(map_tree, em, 0);
7036 write_unlock(&map_tree->lock);
7039 "failed to add chunk map, start=%llu len=%llu: %d",
7040 em->start, em->len, ret);
7042 free_extent_map(em);
7047 static void fill_device_from_item(struct extent_buffer *leaf,
7048 struct btrfs_dev_item *dev_item,
7049 struct btrfs_device *device)
7053 device->devid = btrfs_device_id(leaf, dev_item);
7054 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
7055 device->total_bytes = device->disk_total_bytes;
7056 device->commit_total_bytes = device->disk_total_bytes;
7057 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
7058 device->commit_bytes_used = device->bytes_used;
7059 device->type = btrfs_device_type(leaf, dev_item);
7060 device->io_align = btrfs_device_io_align(leaf, dev_item);
7061 device->io_width = btrfs_device_io_width(leaf, dev_item);
7062 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
7063 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
7064 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
7066 ptr = btrfs_device_uuid(dev_item);
7067 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
7070 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
7073 struct btrfs_fs_devices *fs_devices;
7076 lockdep_assert_held(&uuid_mutex);
7079 /* This will match only for multi-device seed fs */
7080 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
7081 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
7085 fs_devices = find_fsid(fsid, NULL);
7087 if (!btrfs_test_opt(fs_info, DEGRADED))
7088 return ERR_PTR(-ENOENT);
7090 fs_devices = alloc_fs_devices(fsid, NULL);
7091 if (IS_ERR(fs_devices))
7094 fs_devices->seeding = true;
7095 fs_devices->opened = 1;
7100 * Upon first call for a seed fs fsid, just create a private copy of the
7101 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
7103 fs_devices = clone_fs_devices(fs_devices);
7104 if (IS_ERR(fs_devices))
7107 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
7109 free_fs_devices(fs_devices);
7110 return ERR_PTR(ret);
7113 if (!fs_devices->seeding) {
7114 close_fs_devices(fs_devices);
7115 free_fs_devices(fs_devices);
7116 return ERR_PTR(-EINVAL);
7119 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
7124 static int read_one_dev(struct extent_buffer *leaf,
7125 struct btrfs_dev_item *dev_item)
7127 struct btrfs_fs_info *fs_info = leaf->fs_info;
7128 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7129 struct btrfs_device *device;
7132 u8 fs_uuid[BTRFS_FSID_SIZE];
7133 u8 dev_uuid[BTRFS_UUID_SIZE];
7135 devid = btrfs_device_id(leaf, dev_item);
7136 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
7138 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
7141 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
7142 fs_devices = open_seed_devices(fs_info, fs_uuid);
7143 if (IS_ERR(fs_devices))
7144 return PTR_ERR(fs_devices);
7147 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
7150 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7151 btrfs_report_missing_device(fs_info, devid,
7156 device = add_missing_dev(fs_devices, devid, dev_uuid);
7157 if (IS_ERR(device)) {
7159 "failed to add missing dev %llu: %ld",
7160 devid, PTR_ERR(device));
7161 return PTR_ERR(device);
7163 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
7165 if (!device->bdev) {
7166 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7167 btrfs_report_missing_device(fs_info,
7168 devid, dev_uuid, true);
7171 btrfs_report_missing_device(fs_info, devid,
7175 if (!device->bdev &&
7176 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
7178 * this happens when a device that was properly setup
7179 * in the device info lists suddenly goes bad.
7180 * device->bdev is NULL, and so we have to set
7181 * device->missing to one here
7183 device->fs_devices->missing_devices++;
7184 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
7187 /* Move the device to its own fs_devices */
7188 if (device->fs_devices != fs_devices) {
7189 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
7190 &device->dev_state));
7192 list_move(&device->dev_list, &fs_devices->devices);
7193 device->fs_devices->num_devices--;
7194 fs_devices->num_devices++;
7196 device->fs_devices->missing_devices--;
7197 fs_devices->missing_devices++;
7199 device->fs_devices = fs_devices;
7203 if (device->fs_devices != fs_info->fs_devices) {
7204 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
7205 if (device->generation !=
7206 btrfs_device_generation(leaf, dev_item))
7210 fill_device_from_item(leaf, dev_item, device);
7212 u64 max_total_bytes = i_size_read(device->bdev->bd_inode);
7214 if (device->total_bytes > max_total_bytes) {
7216 "device total_bytes should be at most %llu but found %llu",
7217 max_total_bytes, device->total_bytes);
7221 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7222 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7223 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7224 device->fs_devices->total_rw_bytes += device->total_bytes;
7225 atomic64_add(device->total_bytes - device->bytes_used,
7226 &fs_info->free_chunk_space);
7232 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7234 struct btrfs_root *root = fs_info->tree_root;
7235 struct btrfs_super_block *super_copy = fs_info->super_copy;
7236 struct extent_buffer *sb;
7237 struct btrfs_disk_key *disk_key;
7238 struct btrfs_chunk *chunk;
7240 unsigned long sb_array_offset;
7247 struct btrfs_key key;
7249 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7251 * This will create extent buffer of nodesize, superblock size is
7252 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
7253 * overallocate but we can keep it as-is, only the first page is used.
7255 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET,
7256 root->root_key.objectid, 0);
7259 set_extent_buffer_uptodate(sb);
7261 * The sb extent buffer is artificial and just used to read the system array.
7262 * set_extent_buffer_uptodate() call does not properly mark all it's
7263 * pages up-to-date when the page is larger: extent does not cover the
7264 * whole page and consequently check_page_uptodate does not find all
7265 * the page's extents up-to-date (the hole beyond sb),
7266 * write_extent_buffer then triggers a WARN_ON.
7268 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
7269 * but sb spans only this function. Add an explicit SetPageUptodate call
7270 * to silence the warning eg. on PowerPC 64.
7272 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
7273 SetPageUptodate(sb->pages[0]);
7275 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7276 array_size = btrfs_super_sys_array_size(super_copy);
7278 array_ptr = super_copy->sys_chunk_array;
7279 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7282 while (cur_offset < array_size) {
7283 disk_key = (struct btrfs_disk_key *)array_ptr;
7284 len = sizeof(*disk_key);
7285 if (cur_offset + len > array_size)
7286 goto out_short_read;
7288 btrfs_disk_key_to_cpu(&key, disk_key);
7291 sb_array_offset += len;
7294 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7296 "unexpected item type %u in sys_array at offset %u",
7297 (u32)key.type, cur_offset);
7302 chunk = (struct btrfs_chunk *)sb_array_offset;
7304 * At least one btrfs_chunk with one stripe must be present,
7305 * exact stripe count check comes afterwards
7307 len = btrfs_chunk_item_size(1);
7308 if (cur_offset + len > array_size)
7309 goto out_short_read;
7311 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7314 "invalid number of stripes %u in sys_array at offset %u",
7315 num_stripes, cur_offset);
7320 type = btrfs_chunk_type(sb, chunk);
7321 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7323 "invalid chunk type %llu in sys_array at offset %u",
7329 len = btrfs_chunk_item_size(num_stripes);
7330 if (cur_offset + len > array_size)
7331 goto out_short_read;
7333 ret = read_one_chunk(&key, sb, chunk);
7338 sb_array_offset += len;
7341 clear_extent_buffer_uptodate(sb);
7342 free_extent_buffer_stale(sb);
7346 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7348 clear_extent_buffer_uptodate(sb);
7349 free_extent_buffer_stale(sb);
7354 * Check if all chunks in the fs are OK for read-write degraded mount
7356 * If the @failing_dev is specified, it's accounted as missing.
7358 * Return true if all chunks meet the minimal RW mount requirements.
7359 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7361 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7362 struct btrfs_device *failing_dev)
7364 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7365 struct extent_map *em;
7369 read_lock(&map_tree->lock);
7370 em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7371 read_unlock(&map_tree->lock);
7372 /* No chunk at all? Return false anyway */
7378 struct map_lookup *map;
7383 map = em->map_lookup;
7385 btrfs_get_num_tolerated_disk_barrier_failures(
7387 for (i = 0; i < map->num_stripes; i++) {
7388 struct btrfs_device *dev = map->stripes[i].dev;
7390 if (!dev || !dev->bdev ||
7391 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7392 dev->last_flush_error)
7394 else if (failing_dev && failing_dev == dev)
7397 if (missing > max_tolerated) {
7400 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7401 em->start, missing, max_tolerated);
7402 free_extent_map(em);
7406 next_start = extent_map_end(em);
7407 free_extent_map(em);
7409 read_lock(&map_tree->lock);
7410 em = lookup_extent_mapping(map_tree, next_start,
7411 (u64)(-1) - next_start);
7412 read_unlock(&map_tree->lock);
7418 static void readahead_tree_node_children(struct extent_buffer *node)
7421 const int nr_items = btrfs_header_nritems(node);
7423 for (i = 0; i < nr_items; i++)
7424 btrfs_readahead_node_child(node, i);
7427 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7429 struct btrfs_root *root = fs_info->chunk_root;
7430 struct btrfs_path *path;
7431 struct extent_buffer *leaf;
7432 struct btrfs_key key;
7433 struct btrfs_key found_key;
7437 u64 last_ra_node = 0;
7439 path = btrfs_alloc_path();
7444 * uuid_mutex is needed only if we are mounting a sprout FS
7445 * otherwise we don't need it.
7447 mutex_lock(&uuid_mutex);
7450 * It is possible for mount and umount to race in such a way that
7451 * we execute this code path, but open_fs_devices failed to clear
7452 * total_rw_bytes. We certainly want it cleared before reading the
7453 * device items, so clear it here.
7455 fs_info->fs_devices->total_rw_bytes = 0;
7458 * Read all device items, and then all the chunk items. All
7459 * device items are found before any chunk item (their object id
7460 * is smaller than the lowest possible object id for a chunk
7461 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7463 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7466 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7470 struct extent_buffer *node;
7472 leaf = path->nodes[0];
7473 slot = path->slots[0];
7474 if (slot >= btrfs_header_nritems(leaf)) {
7475 ret = btrfs_next_leaf(root, path);
7483 * The nodes on level 1 are not locked but we don't need to do
7484 * that during mount time as nothing else can access the tree
7486 node = path->nodes[1];
7488 if (last_ra_node != node->start) {
7489 readahead_tree_node_children(node);
7490 last_ra_node = node->start;
7493 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7494 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7495 struct btrfs_dev_item *dev_item;
7496 dev_item = btrfs_item_ptr(leaf, slot,
7497 struct btrfs_dev_item);
7498 ret = read_one_dev(leaf, dev_item);
7502 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7503 struct btrfs_chunk *chunk;
7506 * We are only called at mount time, so no need to take
7507 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7508 * we always lock first fs_info->chunk_mutex before
7509 * acquiring any locks on the chunk tree. This is a
7510 * requirement for chunk allocation, see the comment on
7511 * top of btrfs_chunk_alloc() for details.
7513 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
7514 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7515 ret = read_one_chunk(&found_key, leaf, chunk);
7523 * After loading chunk tree, we've got all device information,
7524 * do another round of validation checks.
7526 if (total_dev != fs_info->fs_devices->total_devices) {
7528 "super_num_devices %llu mismatch with num_devices %llu found here",
7529 btrfs_super_num_devices(fs_info->super_copy),
7534 if (btrfs_super_total_bytes(fs_info->super_copy) <
7535 fs_info->fs_devices->total_rw_bytes) {
7537 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7538 btrfs_super_total_bytes(fs_info->super_copy),
7539 fs_info->fs_devices->total_rw_bytes);
7545 mutex_unlock(&uuid_mutex);
7547 btrfs_free_path(path);
7551 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7553 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7554 struct btrfs_device *device;
7556 fs_devices->fs_info = fs_info;
7558 mutex_lock(&fs_devices->device_list_mutex);
7559 list_for_each_entry(device, &fs_devices->devices, dev_list)
7560 device->fs_info = fs_info;
7562 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7563 list_for_each_entry(device, &seed_devs->devices, dev_list)
7564 device->fs_info = fs_info;
7566 seed_devs->fs_info = fs_info;
7568 mutex_unlock(&fs_devices->device_list_mutex);
7571 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7572 const struct btrfs_dev_stats_item *ptr,
7577 read_extent_buffer(eb, &val,
7578 offsetof(struct btrfs_dev_stats_item, values) +
7579 ((unsigned long)ptr) + (index * sizeof(u64)),
7584 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7585 struct btrfs_dev_stats_item *ptr,
7588 write_extent_buffer(eb, &val,
7589 offsetof(struct btrfs_dev_stats_item, values) +
7590 ((unsigned long)ptr) + (index * sizeof(u64)),
7594 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7595 struct btrfs_path *path)
7597 struct btrfs_dev_stats_item *ptr;
7598 struct extent_buffer *eb;
7599 struct btrfs_key key;
7603 if (!device->fs_info->dev_root)
7606 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7607 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7608 key.offset = device->devid;
7609 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7611 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7612 btrfs_dev_stat_set(device, i, 0);
7613 device->dev_stats_valid = 1;
7614 btrfs_release_path(path);
7615 return ret < 0 ? ret : 0;
7617 slot = path->slots[0];
7618 eb = path->nodes[0];
7619 item_size = btrfs_item_size_nr(eb, slot);
7621 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7623 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7624 if (item_size >= (1 + i) * sizeof(__le64))
7625 btrfs_dev_stat_set(device, i,
7626 btrfs_dev_stats_value(eb, ptr, i));
7628 btrfs_dev_stat_set(device, i, 0);
7631 device->dev_stats_valid = 1;
7632 btrfs_dev_stat_print_on_load(device);
7633 btrfs_release_path(path);
7638 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7640 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7641 struct btrfs_device *device;
7642 struct btrfs_path *path = NULL;
7645 path = btrfs_alloc_path();
7649 mutex_lock(&fs_devices->device_list_mutex);
7650 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7651 ret = btrfs_device_init_dev_stats(device, path);
7655 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7656 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7657 ret = btrfs_device_init_dev_stats(device, path);
7663 mutex_unlock(&fs_devices->device_list_mutex);
7665 btrfs_free_path(path);
7669 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7670 struct btrfs_device *device)
7672 struct btrfs_fs_info *fs_info = trans->fs_info;
7673 struct btrfs_root *dev_root = fs_info->dev_root;
7674 struct btrfs_path *path;
7675 struct btrfs_key key;
7676 struct extent_buffer *eb;
7677 struct btrfs_dev_stats_item *ptr;
7681 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7682 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7683 key.offset = device->devid;
7685 path = btrfs_alloc_path();
7688 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7690 btrfs_warn_in_rcu(fs_info,
7691 "error %d while searching for dev_stats item for device %s",
7692 ret, rcu_str_deref(device->name));
7697 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7698 /* need to delete old one and insert a new one */
7699 ret = btrfs_del_item(trans, dev_root, path);
7701 btrfs_warn_in_rcu(fs_info,
7702 "delete too small dev_stats item for device %s failed %d",
7703 rcu_str_deref(device->name), ret);
7710 /* need to insert a new item */
7711 btrfs_release_path(path);
7712 ret = btrfs_insert_empty_item(trans, dev_root, path,
7713 &key, sizeof(*ptr));
7715 btrfs_warn_in_rcu(fs_info,
7716 "insert dev_stats item for device %s failed %d",
7717 rcu_str_deref(device->name), ret);
7722 eb = path->nodes[0];
7723 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7724 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7725 btrfs_set_dev_stats_value(eb, ptr, i,
7726 btrfs_dev_stat_read(device, i));
7727 btrfs_mark_buffer_dirty(eb);
7730 btrfs_free_path(path);
7735 * called from commit_transaction. Writes all changed device stats to disk.
7737 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7739 struct btrfs_fs_info *fs_info = trans->fs_info;
7740 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7741 struct btrfs_device *device;
7745 mutex_lock(&fs_devices->device_list_mutex);
7746 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7747 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7748 if (!device->dev_stats_valid || stats_cnt == 0)
7753 * There is a LOAD-LOAD control dependency between the value of
7754 * dev_stats_ccnt and updating the on-disk values which requires
7755 * reading the in-memory counters. Such control dependencies
7756 * require explicit read memory barriers.
7758 * This memory barriers pairs with smp_mb__before_atomic in
7759 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7760 * barrier implied by atomic_xchg in
7761 * btrfs_dev_stats_read_and_reset
7765 ret = update_dev_stat_item(trans, device);
7767 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7769 mutex_unlock(&fs_devices->device_list_mutex);
7774 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7776 btrfs_dev_stat_inc(dev, index);
7777 btrfs_dev_stat_print_on_error(dev);
7780 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7782 if (!dev->dev_stats_valid)
7784 btrfs_err_rl_in_rcu(dev->fs_info,
7785 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7786 rcu_str_deref(dev->name),
7787 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7788 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7789 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7790 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7791 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7794 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7798 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7799 if (btrfs_dev_stat_read(dev, i) != 0)
7801 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7802 return; /* all values == 0, suppress message */
7804 btrfs_info_in_rcu(dev->fs_info,
7805 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7806 rcu_str_deref(dev->name),
7807 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7808 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7809 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7810 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7811 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7814 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7815 struct btrfs_ioctl_get_dev_stats *stats)
7817 struct btrfs_device *dev;
7818 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7821 mutex_lock(&fs_devices->device_list_mutex);
7822 dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL);
7823 mutex_unlock(&fs_devices->device_list_mutex);
7826 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7828 } else if (!dev->dev_stats_valid) {
7829 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7831 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7832 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7833 if (stats->nr_items > i)
7835 btrfs_dev_stat_read_and_reset(dev, i);
7837 btrfs_dev_stat_set(dev, i, 0);
7839 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7840 current->comm, task_pid_nr(current));
7842 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7843 if (stats->nr_items > i)
7844 stats->values[i] = btrfs_dev_stat_read(dev, i);
7846 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7847 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7852 * Update the size and bytes used for each device where it changed. This is
7853 * delayed since we would otherwise get errors while writing out the
7856 * Must be invoked during transaction commit.
7858 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7860 struct btrfs_device *curr, *next;
7862 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7864 if (list_empty(&trans->dev_update_list))
7868 * We don't need the device_list_mutex here. This list is owned by the
7869 * transaction and the transaction must complete before the device is
7872 mutex_lock(&trans->fs_info->chunk_mutex);
7873 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7875 list_del_init(&curr->post_commit_list);
7876 curr->commit_total_bytes = curr->disk_total_bytes;
7877 curr->commit_bytes_used = curr->bytes_used;
7879 mutex_unlock(&trans->fs_info->chunk_mutex);
7883 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7885 int btrfs_bg_type_to_factor(u64 flags)
7887 const int index = btrfs_bg_flags_to_raid_index(flags);
7889 return btrfs_raid_array[index].ncopies;
7894 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7895 u64 chunk_offset, u64 devid,
7896 u64 physical_offset, u64 physical_len)
7898 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7899 struct extent_map *em;
7900 struct map_lookup *map;
7901 struct btrfs_device *dev;
7907 read_lock(&em_tree->lock);
7908 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7909 read_unlock(&em_tree->lock);
7913 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7914 physical_offset, devid);
7919 map = em->map_lookup;
7920 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes);
7921 if (physical_len != stripe_len) {
7923 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7924 physical_offset, devid, em->start, physical_len,
7930 for (i = 0; i < map->num_stripes; i++) {
7931 if (map->stripes[i].dev->devid == devid &&
7932 map->stripes[i].physical == physical_offset) {
7934 if (map->verified_stripes >= map->num_stripes) {
7936 "too many dev extents for chunk %llu found",
7941 map->verified_stripes++;
7947 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7948 physical_offset, devid);
7952 /* Make sure no dev extent is beyond device boundary */
7953 dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL);
7955 btrfs_err(fs_info, "failed to find devid %llu", devid);
7960 if (physical_offset + physical_len > dev->disk_total_bytes) {
7962 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7963 devid, physical_offset, physical_len,
7964 dev->disk_total_bytes);
7969 if (dev->zone_info) {
7970 u64 zone_size = dev->zone_info->zone_size;
7972 if (!IS_ALIGNED(physical_offset, zone_size) ||
7973 !IS_ALIGNED(physical_len, zone_size)) {
7975 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
7976 devid, physical_offset, physical_len);
7983 free_extent_map(em);
7987 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7989 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7990 struct extent_map *em;
7991 struct rb_node *node;
7994 read_lock(&em_tree->lock);
7995 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
7996 em = rb_entry(node, struct extent_map, rb_node);
7997 if (em->map_lookup->num_stripes !=
7998 em->map_lookup->verified_stripes) {
8000 "chunk %llu has missing dev extent, have %d expect %d",
8001 em->start, em->map_lookup->verified_stripes,
8002 em->map_lookup->num_stripes);
8008 read_unlock(&em_tree->lock);
8013 * Ensure that all dev extents are mapped to correct chunk, otherwise
8014 * later chunk allocation/free would cause unexpected behavior.
8016 * NOTE: This will iterate through the whole device tree, which should be of
8017 * the same size level as the chunk tree. This slightly increases mount time.
8019 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
8021 struct btrfs_path *path;
8022 struct btrfs_root *root = fs_info->dev_root;
8023 struct btrfs_key key;
8025 u64 prev_dev_ext_end = 0;
8029 * We don't have a dev_root because we mounted with ignorebadroots and
8030 * failed to load the root, so we want to skip the verification in this
8033 * However if the dev root is fine, but the tree itself is corrupted
8034 * we'd still fail to mount. This verification is only to make sure
8035 * writes can happen safely, so instead just bypass this check
8036 * completely in the case of IGNOREBADROOTS.
8038 if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
8042 key.type = BTRFS_DEV_EXTENT_KEY;
8045 path = btrfs_alloc_path();
8049 path->reada = READA_FORWARD;
8050 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8054 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
8055 ret = btrfs_next_leaf(root, path);
8058 /* No dev extents at all? Not good */
8065 struct extent_buffer *leaf = path->nodes[0];
8066 struct btrfs_dev_extent *dext;
8067 int slot = path->slots[0];
8069 u64 physical_offset;
8073 btrfs_item_key_to_cpu(leaf, &key, slot);
8074 if (key.type != BTRFS_DEV_EXTENT_KEY)
8076 devid = key.objectid;
8077 physical_offset = key.offset;
8079 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
8080 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
8081 physical_len = btrfs_dev_extent_length(leaf, dext);
8083 /* Check if this dev extent overlaps with the previous one */
8084 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
8086 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
8087 devid, physical_offset, prev_dev_ext_end);
8092 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
8093 physical_offset, physical_len);
8097 prev_dev_ext_end = physical_offset + physical_len;
8099 ret = btrfs_next_item(root, path);
8108 /* Ensure all chunks have corresponding dev extents */
8109 ret = verify_chunk_dev_extent_mapping(fs_info);
8111 btrfs_free_path(path);
8116 * Check whether the given block group or device is pinned by any inode being
8117 * used as a swapfile.
8119 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
8121 struct btrfs_swapfile_pin *sp;
8122 struct rb_node *node;
8124 spin_lock(&fs_info->swapfile_pins_lock);
8125 node = fs_info->swapfile_pins.rb_node;
8127 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
8129 node = node->rb_left;
8130 else if (ptr > sp->ptr)
8131 node = node->rb_right;
8135 spin_unlock(&fs_info->swapfile_pins_lock);
8136 return node != NULL;
8139 static int relocating_repair_kthread(void *data)
8141 struct btrfs_block_group *cache = (struct btrfs_block_group *)data;
8142 struct btrfs_fs_info *fs_info = cache->fs_info;
8146 target = cache->start;
8147 btrfs_put_block_group(cache);
8149 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
8151 "zoned: skip relocating block group %llu to repair: EBUSY",
8156 mutex_lock(&fs_info->reclaim_bgs_lock);
8158 /* Ensure block group still exists */
8159 cache = btrfs_lookup_block_group(fs_info, target);
8163 if (!cache->relocating_repair)
8166 ret = btrfs_may_alloc_data_chunk(fs_info, target);
8171 "zoned: relocating block group %llu to repair IO failure",
8173 ret = btrfs_relocate_chunk(fs_info, target);
8177 btrfs_put_block_group(cache);
8178 mutex_unlock(&fs_info->reclaim_bgs_lock);
8179 btrfs_exclop_finish(fs_info);
8184 int btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical)
8186 struct btrfs_block_group *cache;
8188 /* Do not attempt to repair in degraded state */
8189 if (btrfs_test_opt(fs_info, DEGRADED))
8192 cache = btrfs_lookup_block_group(fs_info, logical);
8196 spin_lock(&cache->lock);
8197 if (cache->relocating_repair) {
8198 spin_unlock(&cache->lock);
8199 btrfs_put_block_group(cache);
8202 cache->relocating_repair = 1;
8203 spin_unlock(&cache->lock);
8205 kthread_run(relocating_repair_kthread, cache,
8206 "btrfs-relocating-repair");