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"
35 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
36 [BTRFS_RAID_RAID10] = {
39 .devs_max = 0, /* 0 == as many as possible */
41 .tolerated_failures = 1,
45 .raid_name = "raid10",
46 .bg_flag = BTRFS_BLOCK_GROUP_RAID10,
47 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
49 [BTRFS_RAID_RAID1] = {
54 .tolerated_failures = 1,
59 .bg_flag = BTRFS_BLOCK_GROUP_RAID1,
60 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
62 [BTRFS_RAID_RAID1C3] = {
67 .tolerated_failures = 2,
71 .raid_name = "raid1c3",
72 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3,
73 .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET,
75 [BTRFS_RAID_RAID1C4] = {
80 .tolerated_failures = 3,
84 .raid_name = "raid1c4",
85 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4,
86 .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET,
93 .tolerated_failures = 0,
98 .bg_flag = BTRFS_BLOCK_GROUP_DUP,
101 [BTRFS_RAID_RAID0] = {
106 .tolerated_failures = 0,
110 .raid_name = "raid0",
111 .bg_flag = BTRFS_BLOCK_GROUP_RAID0,
114 [BTRFS_RAID_SINGLE] = {
119 .tolerated_failures = 0,
123 .raid_name = "single",
127 [BTRFS_RAID_RAID5] = {
132 .tolerated_failures = 1,
136 .raid_name = "raid5",
137 .bg_flag = BTRFS_BLOCK_GROUP_RAID5,
138 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
140 [BTRFS_RAID_RAID6] = {
145 .tolerated_failures = 2,
149 .raid_name = "raid6",
150 .bg_flag = BTRFS_BLOCK_GROUP_RAID6,
151 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
155 const char *btrfs_bg_type_to_raid_name(u64 flags)
157 const int index = btrfs_bg_flags_to_raid_index(flags);
159 if (index >= BTRFS_NR_RAID_TYPES)
162 return btrfs_raid_array[index].raid_name;
166 * Fill @buf with textual description of @bg_flags, no more than @size_buf
167 * bytes including terminating null byte.
169 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf)
174 u64 flags = bg_flags;
175 u32 size_bp = size_buf;
182 #define DESCRIBE_FLAG(flag, desc) \
184 if (flags & (flag)) { \
185 ret = snprintf(bp, size_bp, "%s|", (desc)); \
186 if (ret < 0 || ret >= size_bp) \
194 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data");
195 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system");
196 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata");
198 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single");
199 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
200 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag,
201 btrfs_raid_array[i].raid_name);
205 ret = snprintf(bp, size_bp, "0x%llx|", flags);
209 if (size_bp < size_buf)
210 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */
213 * The text is trimmed, it's up to the caller to provide sufficiently
219 static int init_first_rw_device(struct btrfs_trans_handle *trans);
220 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
221 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
222 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
223 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
224 enum btrfs_map_op op,
225 u64 logical, u64 *length,
226 struct btrfs_bio **bbio_ret,
227 int mirror_num, int need_raid_map);
233 * There are several mutexes that protect manipulation of devices and low-level
234 * structures like chunks but not block groups, extents or files
236 * uuid_mutex (global lock)
237 * ------------------------
238 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
239 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
240 * device) or requested by the device= mount option
242 * the mutex can be very coarse and can cover long-running operations
244 * protects: updates to fs_devices counters like missing devices, rw devices,
245 * seeding, structure cloning, opening/closing devices at mount/umount time
247 * global::fs_devs - add, remove, updates to the global list
249 * does not protect: manipulation of the fs_devices::devices list in general
250 * but in mount context it could be used to exclude list modifications by eg.
253 * btrfs_device::name - renames (write side), read is RCU
255 * fs_devices::device_list_mutex (per-fs, with RCU)
256 * ------------------------------------------------
257 * protects updates to fs_devices::devices, ie. adding and deleting
259 * simple list traversal with read-only actions can be done with RCU protection
261 * may be used to exclude some operations from running concurrently without any
262 * modifications to the list (see write_all_supers)
264 * Is not required at mount and close times, because our device list is
265 * protected by the uuid_mutex at that point.
269 * protects balance structures (status, state) and context accessed from
270 * several places (internally, ioctl)
274 * protects chunks, adding or removing during allocation, trim or when a new
275 * device is added/removed. Additionally it also protects post_commit_list of
276 * individual devices, since they can be added to the transaction's
277 * post_commit_list only with chunk_mutex held.
281 * a big lock that is held by the cleaner thread and prevents running subvolume
282 * cleaning together with relocation or delayed iputs
294 * Exclusive operations, BTRFS_FS_EXCL_OP
295 * ======================================
297 * Maintains the exclusivity of the following operations that apply to the
298 * whole filesystem and cannot run in parallel.
303 * - Device replace (*)
306 * The device operations (as above) can be in one of the following states:
312 * Only device operations marked with (*) can go into the Paused state for the
315 * - ioctl (only Balance can be Paused through ioctl)
316 * - filesystem remounted as read-only
317 * - filesystem unmounted and mounted as read-only
318 * - system power-cycle and filesystem mounted as read-only
319 * - filesystem or device errors leading to forced read-only
321 * BTRFS_FS_EXCL_OP flag is set and cleared using atomic operations.
322 * During the course of Paused state, the BTRFS_FS_EXCL_OP remains set.
323 * A device operation in Paused or Running state can be canceled or resumed
324 * either by ioctl (Balance only) or when remounted as read-write.
325 * BTRFS_FS_EXCL_OP flag is cleared when the device operation is canceled or
329 DEFINE_MUTEX(uuid_mutex);
330 static LIST_HEAD(fs_uuids);
331 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void)
337 * alloc_fs_devices - allocate struct btrfs_fs_devices
338 * @fsid: if not NULL, copy the UUID to fs_devices::fsid
339 * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid
341 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
342 * The returned struct is not linked onto any lists and can be destroyed with
343 * kfree() right away.
345 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid,
346 const u8 *metadata_fsid)
348 struct btrfs_fs_devices *fs_devs;
350 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
352 return ERR_PTR(-ENOMEM);
354 mutex_init(&fs_devs->device_list_mutex);
356 INIT_LIST_HEAD(&fs_devs->devices);
357 INIT_LIST_HEAD(&fs_devs->alloc_list);
358 INIT_LIST_HEAD(&fs_devs->fs_list);
360 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
363 memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE);
365 memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE);
370 void btrfs_free_device(struct btrfs_device *device)
372 WARN_ON(!list_empty(&device->post_commit_list));
373 rcu_string_free(device->name);
374 extent_io_tree_release(&device->alloc_state);
375 bio_put(device->flush_bio);
379 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
381 struct btrfs_device *device;
382 WARN_ON(fs_devices->opened);
383 while (!list_empty(&fs_devices->devices)) {
384 device = list_entry(fs_devices->devices.next,
385 struct btrfs_device, dev_list);
386 list_del(&device->dev_list);
387 btrfs_free_device(device);
392 void __exit btrfs_cleanup_fs_uuids(void)
394 struct btrfs_fs_devices *fs_devices;
396 while (!list_empty(&fs_uuids)) {
397 fs_devices = list_entry(fs_uuids.next,
398 struct btrfs_fs_devices, fs_list);
399 list_del(&fs_devices->fs_list);
400 free_fs_devices(fs_devices);
405 * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
406 * Returned struct is not linked onto any lists and must be destroyed using
409 static struct btrfs_device *__alloc_device(void)
411 struct btrfs_device *dev;
413 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
415 return ERR_PTR(-ENOMEM);
418 * Preallocate a bio that's always going to be used for flushing device
419 * barriers and matches the device lifespan
421 dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
422 if (!dev->flush_bio) {
424 return ERR_PTR(-ENOMEM);
427 INIT_LIST_HEAD(&dev->dev_list);
428 INIT_LIST_HEAD(&dev->dev_alloc_list);
429 INIT_LIST_HEAD(&dev->post_commit_list);
431 atomic_set(&dev->reada_in_flight, 0);
432 atomic_set(&dev->dev_stats_ccnt, 0);
433 btrfs_device_data_ordered_init(dev);
434 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
435 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
436 extent_io_tree_init(NULL, &dev->alloc_state, 0, NULL);
441 static noinline struct btrfs_fs_devices *find_fsid(
442 const u8 *fsid, const u8 *metadata_fsid)
444 struct btrfs_fs_devices *fs_devices;
448 /* Handle non-split brain cases */
449 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
451 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0
452 && memcmp(metadata_fsid, fs_devices->metadata_uuid,
453 BTRFS_FSID_SIZE) == 0)
456 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
463 static struct btrfs_fs_devices *find_fsid_with_metadata_uuid(
464 struct btrfs_super_block *disk_super)
467 struct btrfs_fs_devices *fs_devices;
470 * Handle scanned device having completed its fsid change but
471 * belonging to a fs_devices that was created by first scanning
472 * a device which didn't have its fsid/metadata_uuid changed
473 * at all and the CHANGING_FSID_V2 flag set.
475 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
476 if (fs_devices->fsid_change &&
477 memcmp(disk_super->metadata_uuid, fs_devices->fsid,
478 BTRFS_FSID_SIZE) == 0 &&
479 memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
480 BTRFS_FSID_SIZE) == 0) {
485 * Handle scanned device having completed its fsid change but
486 * belonging to a fs_devices that was created by a device that
487 * has an outdated pair of fsid/metadata_uuid and
488 * CHANGING_FSID_V2 flag set.
490 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
491 if (fs_devices->fsid_change &&
492 memcmp(fs_devices->metadata_uuid,
493 fs_devices->fsid, BTRFS_FSID_SIZE) != 0 &&
494 memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid,
495 BTRFS_FSID_SIZE) == 0) {
500 return find_fsid(disk_super->fsid, disk_super->metadata_uuid);
505 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
506 int flush, struct block_device **bdev,
507 struct btrfs_super_block **disk_super)
511 *bdev = blkdev_get_by_path(device_path, flags, holder);
514 ret = PTR_ERR(*bdev);
519 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
520 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
522 blkdev_put(*bdev, flags);
525 invalidate_bdev(*bdev);
526 *disk_super = btrfs_read_dev_super(*bdev);
527 if (IS_ERR(*disk_super)) {
528 ret = PTR_ERR(*disk_super);
529 blkdev_put(*bdev, flags);
540 static bool device_path_matched(const char *path, struct btrfs_device *device)
545 found = strcmp(rcu_str_deref(device->name), path);
552 * Search and remove all stale (devices which are not mounted) devices.
553 * When both inputs are NULL, it will search and release all stale devices.
554 * path: Optional. When provided will it release all unmounted devices
555 * matching this path only.
556 * skip_dev: Optional. Will skip this device when searching for the stale
558 * Return: 0 for success or if @path is NULL.
559 * -EBUSY if @path is a mounted device.
560 * -ENOENT if @path does not match any device in the list.
562 static int btrfs_free_stale_devices(const char *path,
563 struct btrfs_device *skip_device)
565 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
566 struct btrfs_device *device, *tmp_device;
572 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
574 mutex_lock(&fs_devices->device_list_mutex);
575 list_for_each_entry_safe(device, tmp_device,
576 &fs_devices->devices, dev_list) {
577 if (skip_device && skip_device == device)
579 if (path && !device->name)
581 if (path && !device_path_matched(path, device))
583 if (fs_devices->opened) {
584 /* for an already deleted device return 0 */
585 if (path && ret != 0)
590 /* delete the stale device */
591 fs_devices->num_devices--;
592 list_del(&device->dev_list);
593 btrfs_free_device(device);
596 if (fs_devices->num_devices == 0)
599 mutex_unlock(&fs_devices->device_list_mutex);
601 if (fs_devices->num_devices == 0) {
602 btrfs_sysfs_remove_fsid(fs_devices);
603 list_del(&fs_devices->fs_list);
604 free_fs_devices(fs_devices);
612 * This is only used on mount, and we are protected from competing things
613 * messing with our fs_devices by the uuid_mutex, thus we do not need the
614 * fs_devices->device_list_mutex here.
616 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
617 struct btrfs_device *device, fmode_t flags,
620 struct request_queue *q;
621 struct block_device *bdev;
622 struct btrfs_super_block *disk_super;
631 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
636 devid = btrfs_stack_device_id(&disk_super->dev_item);
637 if (devid != device->devid)
638 goto error_free_page;
640 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
641 goto error_free_page;
643 device->generation = btrfs_super_generation(disk_super);
645 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
646 if (btrfs_super_incompat_flags(disk_super) &
647 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
649 "BTRFS: Invalid seeding and uuid-changed device detected\n");
650 goto error_free_page;
653 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
654 fs_devices->seeding = true;
656 if (bdev_read_only(bdev))
657 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
659 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
662 q = bdev_get_queue(bdev);
663 if (!blk_queue_nonrot(q))
664 fs_devices->rotating = true;
667 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
668 device->mode = flags;
670 fs_devices->open_devices++;
671 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
672 device->devid != BTRFS_DEV_REPLACE_DEVID) {
673 fs_devices->rw_devices++;
674 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
676 btrfs_release_disk_super(disk_super);
681 btrfs_release_disk_super(disk_super);
682 blkdev_put(bdev, flags);
688 * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices
689 * being created with a disk that has already completed its fsid change. Such
690 * disk can belong to an fs which has its FSID changed or to one which doesn't.
691 * Handle both cases here.
693 static struct btrfs_fs_devices *find_fsid_inprogress(
694 struct btrfs_super_block *disk_super)
696 struct btrfs_fs_devices *fs_devices;
698 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
699 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
700 BTRFS_FSID_SIZE) != 0 &&
701 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
702 BTRFS_FSID_SIZE) == 0 && !fs_devices->fsid_change) {
707 return find_fsid(disk_super->fsid, NULL);
711 static struct btrfs_fs_devices *find_fsid_changed(
712 struct btrfs_super_block *disk_super)
714 struct btrfs_fs_devices *fs_devices;
717 * Handles the case where scanned device is part of an fs that had
718 * multiple successful changes of FSID but curently device didn't
719 * observe it. Meaning our fsid will be different than theirs. We need
720 * to handle two subcases :
721 * 1 - The fs still continues to have different METADATA/FSID uuids.
722 * 2 - The fs is switched back to its original FSID (METADATA/FSID
725 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
727 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
728 BTRFS_FSID_SIZE) != 0 &&
729 memcmp(fs_devices->metadata_uuid, disk_super->metadata_uuid,
730 BTRFS_FSID_SIZE) == 0 &&
731 memcmp(fs_devices->fsid, disk_super->fsid,
732 BTRFS_FSID_SIZE) != 0)
735 /* Unchanged UUIDs */
736 if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
737 BTRFS_FSID_SIZE) == 0 &&
738 memcmp(fs_devices->fsid, disk_super->metadata_uuid,
739 BTRFS_FSID_SIZE) == 0)
746 static struct btrfs_fs_devices *find_fsid_reverted_metadata(
747 struct btrfs_super_block *disk_super)
749 struct btrfs_fs_devices *fs_devices;
752 * Handle the case where the scanned device is part of an fs whose last
753 * metadata UUID change reverted it to the original FSID. At the same
754 * time * fs_devices was first created by another constitutent device
755 * which didn't fully observe the operation. This results in an
756 * btrfs_fs_devices created with metadata/fsid different AND
757 * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the
758 * fs_devices equal to the FSID of the disk.
760 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
761 if (memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
762 BTRFS_FSID_SIZE) != 0 &&
763 memcmp(fs_devices->metadata_uuid, disk_super->fsid,
764 BTRFS_FSID_SIZE) == 0 &&
765 fs_devices->fsid_change)
772 * Add new device to list of registered devices
775 * device pointer which was just added or updated when successful
776 * error pointer when failed
778 static noinline struct btrfs_device *device_list_add(const char *path,
779 struct btrfs_super_block *disk_super,
780 bool *new_device_added)
782 struct btrfs_device *device;
783 struct btrfs_fs_devices *fs_devices = NULL;
784 struct rcu_string *name;
785 u64 found_transid = btrfs_super_generation(disk_super);
786 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
787 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
788 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
789 bool fsid_change_in_progress = (btrfs_super_flags(disk_super) &
790 BTRFS_SUPER_FLAG_CHANGING_FSID_V2);
792 if (fsid_change_in_progress) {
793 if (!has_metadata_uuid)
794 fs_devices = find_fsid_inprogress(disk_super);
796 fs_devices = find_fsid_changed(disk_super);
797 } else if (has_metadata_uuid) {
798 fs_devices = find_fsid_with_metadata_uuid(disk_super);
800 fs_devices = find_fsid_reverted_metadata(disk_super);
802 fs_devices = find_fsid(disk_super->fsid, NULL);
807 if (has_metadata_uuid)
808 fs_devices = alloc_fs_devices(disk_super->fsid,
809 disk_super->metadata_uuid);
811 fs_devices = alloc_fs_devices(disk_super->fsid, NULL);
813 if (IS_ERR(fs_devices))
814 return ERR_CAST(fs_devices);
816 fs_devices->fsid_change = fsid_change_in_progress;
818 mutex_lock(&fs_devices->device_list_mutex);
819 list_add(&fs_devices->fs_list, &fs_uuids);
823 mutex_lock(&fs_devices->device_list_mutex);
824 device = btrfs_find_device(fs_devices, devid,
825 disk_super->dev_item.uuid, NULL, false);
828 * If this disk has been pulled into an fs devices created by
829 * a device which had the CHANGING_FSID_V2 flag then replace the
830 * metadata_uuid/fsid values of the fs_devices.
832 if (fs_devices->fsid_change &&
833 found_transid > fs_devices->latest_generation) {
834 memcpy(fs_devices->fsid, disk_super->fsid,
837 if (has_metadata_uuid)
838 memcpy(fs_devices->metadata_uuid,
839 disk_super->metadata_uuid,
842 memcpy(fs_devices->metadata_uuid,
843 disk_super->fsid, BTRFS_FSID_SIZE);
845 fs_devices->fsid_change = false;
850 if (fs_devices->opened) {
851 mutex_unlock(&fs_devices->device_list_mutex);
852 return ERR_PTR(-EBUSY);
855 device = btrfs_alloc_device(NULL, &devid,
856 disk_super->dev_item.uuid);
857 if (IS_ERR(device)) {
858 mutex_unlock(&fs_devices->device_list_mutex);
859 /* we can safely leave the fs_devices entry around */
863 name = rcu_string_strdup(path, GFP_NOFS);
865 btrfs_free_device(device);
866 mutex_unlock(&fs_devices->device_list_mutex);
867 return ERR_PTR(-ENOMEM);
869 rcu_assign_pointer(device->name, name);
871 list_add_rcu(&device->dev_list, &fs_devices->devices);
872 fs_devices->num_devices++;
874 device->fs_devices = fs_devices;
875 *new_device_added = true;
877 if (disk_super->label[0])
879 "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n",
880 disk_super->label, devid, found_transid, path,
881 current->comm, task_pid_nr(current));
884 "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n",
885 disk_super->fsid, devid, found_transid, path,
886 current->comm, task_pid_nr(current));
888 } else if (!device->name || strcmp(device->name->str, path)) {
890 * When FS is already mounted.
891 * 1. If you are here and if the device->name is NULL that
892 * means this device was missing at time of FS mount.
893 * 2. If you are here and if the device->name is different
894 * from 'path' that means either
895 * a. The same device disappeared and reappeared with
897 * b. The missing-disk-which-was-replaced, has
900 * We must allow 1 and 2a above. But 2b would be a spurious
903 * Further in case of 1 and 2a above, the disk at 'path'
904 * would have missed some transaction when it was away and
905 * in case of 2a the stale bdev has to be updated as well.
906 * 2b must not be allowed at all time.
910 * For now, we do allow update to btrfs_fs_device through the
911 * btrfs dev scan cli after FS has been mounted. We're still
912 * tracking a problem where systems fail mount by subvolume id
913 * when we reject replacement on a mounted FS.
915 if (!fs_devices->opened && found_transid < device->generation) {
917 * That is if the FS is _not_ mounted and if you
918 * are here, that means there is more than one
919 * disk with same uuid and devid.We keep the one
920 * with larger generation number or the last-in if
921 * generation are equal.
923 mutex_unlock(&fs_devices->device_list_mutex);
924 return ERR_PTR(-EEXIST);
928 * We are going to replace the device path for a given devid,
929 * make sure it's the same device if the device is mounted
932 struct block_device *path_bdev;
934 path_bdev = lookup_bdev(path);
935 if (IS_ERR(path_bdev)) {
936 mutex_unlock(&fs_devices->device_list_mutex);
937 return ERR_CAST(path_bdev);
940 if (device->bdev != path_bdev) {
942 mutex_unlock(&fs_devices->device_list_mutex);
943 btrfs_warn_in_rcu(device->fs_info,
944 "duplicate device fsid:devid for %pU:%llu old:%s new:%s",
945 disk_super->fsid, devid,
946 rcu_str_deref(device->name), path);
947 return ERR_PTR(-EEXIST);
950 btrfs_info_in_rcu(device->fs_info,
951 "device fsid %pU devid %llu moved old:%s new:%s",
952 disk_super->fsid, devid,
953 rcu_str_deref(device->name), path);
956 name = rcu_string_strdup(path, GFP_NOFS);
958 mutex_unlock(&fs_devices->device_list_mutex);
959 return ERR_PTR(-ENOMEM);
961 rcu_string_free(device->name);
962 rcu_assign_pointer(device->name, name);
963 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
964 fs_devices->missing_devices--;
965 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
970 * Unmount does not free the btrfs_device struct but would zero
971 * generation along with most of the other members. So just update
972 * it back. We need it to pick the disk with largest generation
975 if (!fs_devices->opened) {
976 device->generation = found_transid;
977 fs_devices->latest_generation = max_t(u64, found_transid,
978 fs_devices->latest_generation);
981 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
983 mutex_unlock(&fs_devices->device_list_mutex);
987 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
989 struct btrfs_fs_devices *fs_devices;
990 struct btrfs_device *device;
991 struct btrfs_device *orig_dev;
994 fs_devices = alloc_fs_devices(orig->fsid, NULL);
995 if (IS_ERR(fs_devices))
998 mutex_lock(&orig->device_list_mutex);
999 fs_devices->total_devices = orig->total_devices;
1001 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
1002 struct rcu_string *name;
1004 device = btrfs_alloc_device(NULL, &orig_dev->devid,
1006 if (IS_ERR(device)) {
1007 ret = PTR_ERR(device);
1012 * This is ok to do without rcu read locked because we hold the
1013 * uuid mutex so nothing we touch in here is going to disappear.
1015 if (orig_dev->name) {
1016 name = rcu_string_strdup(orig_dev->name->str,
1019 btrfs_free_device(device);
1023 rcu_assign_pointer(device->name, name);
1026 list_add(&device->dev_list, &fs_devices->devices);
1027 device->fs_devices = fs_devices;
1028 fs_devices->num_devices++;
1030 mutex_unlock(&orig->device_list_mutex);
1033 mutex_unlock(&orig->device_list_mutex);
1034 free_fs_devices(fs_devices);
1035 return ERR_PTR(ret);
1039 * After we have read the system tree and know devids belonging to
1040 * this filesystem, remove the device which does not belong there.
1042 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step)
1044 struct btrfs_device *device, *next;
1045 struct btrfs_device *latest_dev = NULL;
1047 mutex_lock(&uuid_mutex);
1049 /* This is the initialized path, it is safe to release the devices. */
1050 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1051 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
1052 &device->dev_state)) {
1053 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1054 &device->dev_state) &&
1055 !test_bit(BTRFS_DEV_STATE_MISSING,
1056 &device->dev_state) &&
1058 device->generation > latest_dev->generation)) {
1059 latest_dev = device;
1064 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
1066 * In the first step, keep the device which has
1067 * the correct fsid and the devid that is used
1068 * for the dev_replace procedure.
1069 * In the second step, the dev_replace state is
1070 * read from the device tree and it is known
1071 * whether the procedure is really active or
1072 * not, which means whether this device is
1073 * used or whether it should be removed.
1075 if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1076 &device->dev_state)) {
1081 blkdev_put(device->bdev, device->mode);
1082 device->bdev = NULL;
1083 fs_devices->open_devices--;
1085 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1086 list_del_init(&device->dev_alloc_list);
1087 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1088 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1089 &device->dev_state))
1090 fs_devices->rw_devices--;
1092 list_del_init(&device->dev_list);
1093 fs_devices->num_devices--;
1094 btrfs_free_device(device);
1097 if (fs_devices->seed) {
1098 fs_devices = fs_devices->seed;
1102 fs_devices->latest_bdev = latest_dev->bdev;
1104 mutex_unlock(&uuid_mutex);
1107 static void btrfs_close_bdev(struct btrfs_device *device)
1112 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1113 sync_blockdev(device->bdev);
1114 invalidate_bdev(device->bdev);
1117 blkdev_put(device->bdev, device->mode);
1120 static void btrfs_close_one_device(struct btrfs_device *device)
1122 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1124 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1125 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1126 list_del_init(&device->dev_alloc_list);
1127 fs_devices->rw_devices--;
1130 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1131 fs_devices->missing_devices--;
1133 btrfs_close_bdev(device);
1135 fs_devices->open_devices--;
1136 device->bdev = NULL;
1138 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1140 device->fs_info = NULL;
1141 atomic_set(&device->dev_stats_ccnt, 0);
1142 extent_io_tree_release(&device->alloc_state);
1144 /* Verify the device is back in a pristine state */
1145 ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1146 ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1147 ASSERT(list_empty(&device->dev_alloc_list));
1148 ASSERT(list_empty(&device->post_commit_list));
1149 ASSERT(atomic_read(&device->reada_in_flight) == 0);
1152 static int close_fs_devices(struct btrfs_fs_devices *fs_devices)
1154 struct btrfs_device *device, *tmp;
1156 if (--fs_devices->opened > 0)
1159 mutex_lock(&fs_devices->device_list_mutex);
1160 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
1161 btrfs_close_one_device(device);
1163 mutex_unlock(&fs_devices->device_list_mutex);
1165 WARN_ON(fs_devices->open_devices);
1166 WARN_ON(fs_devices->rw_devices);
1167 fs_devices->opened = 0;
1168 fs_devices->seeding = false;
1173 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1175 struct btrfs_fs_devices *seed_devices = NULL;
1178 mutex_lock(&uuid_mutex);
1179 ret = close_fs_devices(fs_devices);
1180 if (!fs_devices->opened) {
1181 seed_devices = fs_devices->seed;
1182 fs_devices->seed = NULL;
1184 mutex_unlock(&uuid_mutex);
1186 while (seed_devices) {
1187 fs_devices = seed_devices;
1188 seed_devices = fs_devices->seed;
1189 close_fs_devices(fs_devices);
1190 free_fs_devices(fs_devices);
1195 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1196 fmode_t flags, void *holder)
1198 struct btrfs_device *device;
1199 struct btrfs_device *latest_dev = NULL;
1201 flags |= FMODE_EXCL;
1203 list_for_each_entry(device, &fs_devices->devices, dev_list) {
1204 /* Just open everything we can; ignore failures here */
1205 if (btrfs_open_one_device(fs_devices, device, flags, holder))
1209 device->generation > latest_dev->generation)
1210 latest_dev = device;
1212 if (fs_devices->open_devices == 0)
1215 fs_devices->opened = 1;
1216 fs_devices->latest_bdev = latest_dev->bdev;
1217 fs_devices->total_rw_bytes = 0;
1218 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1223 static int devid_cmp(void *priv, struct list_head *a, struct list_head *b)
1225 struct btrfs_device *dev1, *dev2;
1227 dev1 = list_entry(a, struct btrfs_device, dev_list);
1228 dev2 = list_entry(b, struct btrfs_device, dev_list);
1230 if (dev1->devid < dev2->devid)
1232 else if (dev1->devid > dev2->devid)
1237 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1238 fmode_t flags, void *holder)
1242 lockdep_assert_held(&uuid_mutex);
1244 * The device_list_mutex cannot be taken here in case opening the
1245 * underlying device takes further locks like bd_mutex.
1247 * We also don't need the lock here as this is called during mount and
1248 * exclusion is provided by uuid_mutex
1251 if (fs_devices->opened) {
1252 fs_devices->opened++;
1255 list_sort(NULL, &fs_devices->devices, devid_cmp);
1256 ret = open_fs_devices(fs_devices, flags, holder);
1262 void btrfs_release_disk_super(struct btrfs_super_block *super)
1264 struct page *page = virt_to_page(super);
1269 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1272 struct btrfs_super_block *disk_super;
1277 /* make sure our super fits in the device */
1278 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1279 return ERR_PTR(-EINVAL);
1281 /* make sure our super fits in the page */
1282 if (sizeof(*disk_super) > PAGE_SIZE)
1283 return ERR_PTR(-EINVAL);
1285 /* make sure our super doesn't straddle pages on disk */
1286 index = bytenr >> PAGE_SHIFT;
1287 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1288 return ERR_PTR(-EINVAL);
1290 /* pull in the page with our super */
1291 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1294 return ERR_CAST(page);
1296 p = page_address(page);
1298 /* align our pointer to the offset of the super block */
1299 disk_super = p + offset_in_page(bytenr);
1301 if (btrfs_super_bytenr(disk_super) != bytenr ||
1302 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1303 btrfs_release_disk_super(p);
1304 return ERR_PTR(-EINVAL);
1307 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1308 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1313 int btrfs_forget_devices(const char *path)
1317 mutex_lock(&uuid_mutex);
1318 ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL);
1319 mutex_unlock(&uuid_mutex);
1325 * Look for a btrfs signature on a device. This may be called out of the mount path
1326 * and we are not allowed to call set_blocksize during the scan. The superblock
1327 * is read via pagecache
1329 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1332 struct btrfs_super_block *disk_super;
1333 bool new_device_added = false;
1334 struct btrfs_device *device = NULL;
1335 struct block_device *bdev;
1338 lockdep_assert_held(&uuid_mutex);
1341 * we would like to check all the supers, but that would make
1342 * a btrfs mount succeed after a mkfs from a different FS.
1343 * So, we need to add a special mount option to scan for
1344 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1346 bytenr = btrfs_sb_offset(0);
1347 flags |= FMODE_EXCL;
1349 bdev = blkdev_get_by_path(path, flags, holder);
1351 return ERR_CAST(bdev);
1353 disk_super = btrfs_read_disk_super(bdev, bytenr);
1354 if (IS_ERR(disk_super)) {
1355 device = ERR_CAST(disk_super);
1356 goto error_bdev_put;
1359 device = device_list_add(path, disk_super, &new_device_added);
1360 if (!IS_ERR(device)) {
1361 if (new_device_added)
1362 btrfs_free_stale_devices(path, device);
1365 btrfs_release_disk_super(disk_super);
1368 blkdev_put(bdev, flags);
1374 * Try to find a chunk that intersects [start, start + len] range and when one
1375 * such is found, record the end of it in *start
1377 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1380 u64 physical_start, physical_end;
1382 lockdep_assert_held(&device->fs_info->chunk_mutex);
1384 if (!find_first_extent_bit(&device->alloc_state, *start,
1385 &physical_start, &physical_end,
1386 CHUNK_ALLOCATED, NULL)) {
1388 if (in_range(physical_start, *start, len) ||
1389 in_range(*start, physical_start,
1390 physical_end - physical_start)) {
1391 *start = physical_end + 1;
1398 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start)
1400 switch (device->fs_devices->chunk_alloc_policy) {
1401 case BTRFS_CHUNK_ALLOC_REGULAR:
1403 * We don't want to overwrite the superblock on the drive nor
1404 * any area used by the boot loader (grub for example), so we
1405 * make sure to start at an offset of at least 1MB.
1407 return max_t(u64, start, SZ_1M);
1414 * dev_extent_hole_check - check if specified hole is suitable for allocation
1415 * @device: the device which we have the hole
1416 * @hole_start: starting position of the hole
1417 * @hole_size: the size of the hole
1418 * @num_bytes: the size of the free space that we need
1420 * This function may modify @hole_start and @hole_end to reflect the suitable
1421 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1423 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1424 u64 *hole_size, u64 num_bytes)
1426 bool changed = false;
1427 u64 hole_end = *hole_start + *hole_size;
1430 * Check before we set max_hole_start, otherwise we could end up
1431 * sending back this offset anyway.
1433 if (contains_pending_extent(device, hole_start, *hole_size)) {
1434 if (hole_end >= *hole_start)
1435 *hole_size = hole_end - *hole_start;
1441 switch (device->fs_devices->chunk_alloc_policy) {
1442 case BTRFS_CHUNK_ALLOC_REGULAR:
1443 /* No extra check */
1453 * find_free_dev_extent_start - find free space in the specified device
1454 * @device: the device which we search the free space in
1455 * @num_bytes: the size of the free space that we need
1456 * @search_start: the position from which to begin the search
1457 * @start: store the start of the free space.
1458 * @len: the size of the free space. that we find, or the size
1459 * of the max free space if we don't find suitable free space
1461 * this uses a pretty simple search, the expectation is that it is
1462 * called very infrequently and that a given device has a small number
1465 * @start is used to store the start of the free space if we find. But if we
1466 * don't find suitable free space, it will be used to store the start position
1467 * of the max free space.
1469 * @len is used to store the size of the free space that we find.
1470 * But if we don't find suitable free space, it is used to store the size of
1471 * the max free space.
1473 * NOTE: This function will search *commit* root of device tree, and does extra
1474 * check to ensure dev extents are not double allocated.
1475 * This makes the function safe to allocate dev extents but may not report
1476 * correct usable device space, as device extent freed in current transaction
1477 * is not reported as avaiable.
1479 static int find_free_dev_extent_start(struct btrfs_device *device,
1480 u64 num_bytes, u64 search_start, u64 *start,
1483 struct btrfs_fs_info *fs_info = device->fs_info;
1484 struct btrfs_root *root = fs_info->dev_root;
1485 struct btrfs_key key;
1486 struct btrfs_dev_extent *dev_extent;
1487 struct btrfs_path *path;
1492 u64 search_end = device->total_bytes;
1495 struct extent_buffer *l;
1497 search_start = dev_extent_search_start(device, search_start);
1499 path = btrfs_alloc_path();
1503 max_hole_start = search_start;
1507 if (search_start >= search_end ||
1508 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1513 path->reada = READA_FORWARD;
1514 path->search_commit_root = 1;
1515 path->skip_locking = 1;
1517 key.objectid = device->devid;
1518 key.offset = search_start;
1519 key.type = BTRFS_DEV_EXTENT_KEY;
1521 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1525 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1532 slot = path->slots[0];
1533 if (slot >= btrfs_header_nritems(l)) {
1534 ret = btrfs_next_leaf(root, path);
1542 btrfs_item_key_to_cpu(l, &key, slot);
1544 if (key.objectid < device->devid)
1547 if (key.objectid > device->devid)
1550 if (key.type != BTRFS_DEV_EXTENT_KEY)
1553 if (key.offset > search_start) {
1554 hole_size = key.offset - search_start;
1555 dev_extent_hole_check(device, &search_start, &hole_size,
1558 if (hole_size > max_hole_size) {
1559 max_hole_start = search_start;
1560 max_hole_size = hole_size;
1564 * If this free space is greater than which we need,
1565 * it must be the max free space that we have found
1566 * until now, so max_hole_start must point to the start
1567 * of this free space and the length of this free space
1568 * is stored in max_hole_size. Thus, we return
1569 * max_hole_start and max_hole_size and go back to the
1572 if (hole_size >= num_bytes) {
1578 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1579 extent_end = key.offset + btrfs_dev_extent_length(l,
1581 if (extent_end > search_start)
1582 search_start = extent_end;
1589 * At this point, search_start should be the end of
1590 * allocated dev extents, and when shrinking the device,
1591 * search_end may be smaller than search_start.
1593 if (search_end > search_start) {
1594 hole_size = search_end - search_start;
1595 if (dev_extent_hole_check(device, &search_start, &hole_size,
1597 btrfs_release_path(path);
1601 if (hole_size > max_hole_size) {
1602 max_hole_start = search_start;
1603 max_hole_size = hole_size;
1608 if (max_hole_size < num_bytes)
1614 btrfs_free_path(path);
1615 *start = max_hole_start;
1617 *len = max_hole_size;
1621 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1622 u64 *start, u64 *len)
1624 /* FIXME use last free of some kind */
1625 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1628 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1629 struct btrfs_device *device,
1630 u64 start, u64 *dev_extent_len)
1632 struct btrfs_fs_info *fs_info = device->fs_info;
1633 struct btrfs_root *root = fs_info->dev_root;
1635 struct btrfs_path *path;
1636 struct btrfs_key key;
1637 struct btrfs_key found_key;
1638 struct extent_buffer *leaf = NULL;
1639 struct btrfs_dev_extent *extent = NULL;
1641 path = btrfs_alloc_path();
1645 key.objectid = device->devid;
1647 key.type = BTRFS_DEV_EXTENT_KEY;
1649 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1651 ret = btrfs_previous_item(root, path, key.objectid,
1652 BTRFS_DEV_EXTENT_KEY);
1655 leaf = path->nodes[0];
1656 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1657 extent = btrfs_item_ptr(leaf, path->slots[0],
1658 struct btrfs_dev_extent);
1659 BUG_ON(found_key.offset > start || found_key.offset +
1660 btrfs_dev_extent_length(leaf, extent) < start);
1662 btrfs_release_path(path);
1664 } else if (ret == 0) {
1665 leaf = path->nodes[0];
1666 extent = btrfs_item_ptr(leaf, path->slots[0],
1667 struct btrfs_dev_extent);
1669 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1673 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1675 ret = btrfs_del_item(trans, root, path);
1677 btrfs_handle_fs_error(fs_info, ret,
1678 "Failed to remove dev extent item");
1680 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1683 btrfs_free_path(path);
1687 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1688 struct btrfs_device *device,
1689 u64 chunk_offset, u64 start, u64 num_bytes)
1692 struct btrfs_path *path;
1693 struct btrfs_fs_info *fs_info = device->fs_info;
1694 struct btrfs_root *root = fs_info->dev_root;
1695 struct btrfs_dev_extent *extent;
1696 struct extent_buffer *leaf;
1697 struct btrfs_key key;
1699 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1700 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1701 path = btrfs_alloc_path();
1705 key.objectid = device->devid;
1707 key.type = BTRFS_DEV_EXTENT_KEY;
1708 ret = btrfs_insert_empty_item(trans, root, path, &key,
1713 leaf = path->nodes[0];
1714 extent = btrfs_item_ptr(leaf, path->slots[0],
1715 struct btrfs_dev_extent);
1716 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1717 BTRFS_CHUNK_TREE_OBJECTID);
1718 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1719 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1720 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1722 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1723 btrfs_mark_buffer_dirty(leaf);
1725 btrfs_free_path(path);
1729 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1731 struct extent_map_tree *em_tree;
1732 struct extent_map *em;
1736 em_tree = &fs_info->mapping_tree;
1737 read_lock(&em_tree->lock);
1738 n = rb_last(&em_tree->map.rb_root);
1740 em = rb_entry(n, struct extent_map, rb_node);
1741 ret = em->start + em->len;
1743 read_unlock(&em_tree->lock);
1748 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1752 struct btrfs_key key;
1753 struct btrfs_key found_key;
1754 struct btrfs_path *path;
1756 path = btrfs_alloc_path();
1760 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1761 key.type = BTRFS_DEV_ITEM_KEY;
1762 key.offset = (u64)-1;
1764 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1770 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1775 ret = btrfs_previous_item(fs_info->chunk_root, path,
1776 BTRFS_DEV_ITEMS_OBJECTID,
1777 BTRFS_DEV_ITEM_KEY);
1781 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1783 *devid_ret = found_key.offset + 1;
1787 btrfs_free_path(path);
1792 * the device information is stored in the chunk root
1793 * the btrfs_device struct should be fully filled in
1795 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1796 struct btrfs_device *device)
1799 struct btrfs_path *path;
1800 struct btrfs_dev_item *dev_item;
1801 struct extent_buffer *leaf;
1802 struct btrfs_key key;
1805 path = btrfs_alloc_path();
1809 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1810 key.type = BTRFS_DEV_ITEM_KEY;
1811 key.offset = device->devid;
1813 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1814 &key, sizeof(*dev_item));
1818 leaf = path->nodes[0];
1819 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1821 btrfs_set_device_id(leaf, dev_item, device->devid);
1822 btrfs_set_device_generation(leaf, dev_item, 0);
1823 btrfs_set_device_type(leaf, dev_item, device->type);
1824 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1825 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1826 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1827 btrfs_set_device_total_bytes(leaf, dev_item,
1828 btrfs_device_get_disk_total_bytes(device));
1829 btrfs_set_device_bytes_used(leaf, dev_item,
1830 btrfs_device_get_bytes_used(device));
1831 btrfs_set_device_group(leaf, dev_item, 0);
1832 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1833 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1834 btrfs_set_device_start_offset(leaf, dev_item, 0);
1836 ptr = btrfs_device_uuid(dev_item);
1837 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1838 ptr = btrfs_device_fsid(dev_item);
1839 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1840 ptr, BTRFS_FSID_SIZE);
1841 btrfs_mark_buffer_dirty(leaf);
1845 btrfs_free_path(path);
1850 * Function to update ctime/mtime for a given device path.
1851 * Mainly used for ctime/mtime based probe like libblkid.
1853 static void update_dev_time(const char *path_name)
1857 filp = filp_open(path_name, O_RDWR, 0);
1860 file_update_time(filp);
1861 filp_close(filp, NULL);
1864 static int btrfs_rm_dev_item(struct btrfs_device *device)
1866 struct btrfs_root *root = device->fs_info->chunk_root;
1868 struct btrfs_path *path;
1869 struct btrfs_key key;
1870 struct btrfs_trans_handle *trans;
1872 path = btrfs_alloc_path();
1876 trans = btrfs_start_transaction(root, 0);
1877 if (IS_ERR(trans)) {
1878 btrfs_free_path(path);
1879 return PTR_ERR(trans);
1881 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1882 key.type = BTRFS_DEV_ITEM_KEY;
1883 key.offset = device->devid;
1885 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1889 btrfs_abort_transaction(trans, ret);
1890 btrfs_end_transaction(trans);
1894 ret = btrfs_del_item(trans, root, path);
1896 btrfs_abort_transaction(trans, ret);
1897 btrfs_end_transaction(trans);
1901 btrfs_free_path(path);
1903 ret = btrfs_commit_transaction(trans);
1908 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1909 * filesystem. It's up to the caller to adjust that number regarding eg. device
1912 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1920 seq = read_seqbegin(&fs_info->profiles_lock);
1922 all_avail = fs_info->avail_data_alloc_bits |
1923 fs_info->avail_system_alloc_bits |
1924 fs_info->avail_metadata_alloc_bits;
1925 } while (read_seqretry(&fs_info->profiles_lock, seq));
1927 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1928 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1931 if (num_devices < btrfs_raid_array[i].devs_min) {
1932 int ret = btrfs_raid_array[i].mindev_error;
1942 static struct btrfs_device * btrfs_find_next_active_device(
1943 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1945 struct btrfs_device *next_device;
1947 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1948 if (next_device != device &&
1949 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1950 && next_device->bdev)
1958 * Helper function to check if the given device is part of s_bdev / latest_bdev
1959 * and replace it with the provided or the next active device, in the context
1960 * where this function called, there should be always be another device (or
1961 * this_dev) which is active.
1963 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
1964 struct btrfs_device *this_dev)
1966 struct btrfs_fs_info *fs_info = device->fs_info;
1967 struct btrfs_device *next_device;
1970 next_device = this_dev;
1972 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1974 ASSERT(next_device);
1976 if (fs_info->sb->s_bdev &&
1977 (fs_info->sb->s_bdev == device->bdev))
1978 fs_info->sb->s_bdev = next_device->bdev;
1980 if (fs_info->fs_devices->latest_bdev == device->bdev)
1981 fs_info->fs_devices->latest_bdev = next_device->bdev;
1985 * Return btrfs_fs_devices::num_devices excluding the device that's being
1986 * currently replaced.
1988 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
1990 u64 num_devices = fs_info->fs_devices->num_devices;
1992 down_read(&fs_info->dev_replace.rwsem);
1993 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1994 ASSERT(num_devices > 1);
1997 up_read(&fs_info->dev_replace.rwsem);
2002 static void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
2003 struct block_device *bdev,
2004 const char *device_path)
2006 struct btrfs_super_block *disk_super;
2012 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2016 disk_super = btrfs_read_dev_one_super(bdev, copy_num);
2017 if (IS_ERR(disk_super))
2020 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
2022 page = virt_to_page(disk_super);
2023 set_page_dirty(page);
2025 /* write_on_page() unlocks the page */
2026 ret = write_one_page(page);
2029 "error clearing superblock number %d (%d)",
2031 btrfs_release_disk_super(disk_super);
2035 /* Notify udev that device has changed */
2036 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2038 /* Update ctime/mtime for device path for libblkid */
2039 update_dev_time(device_path);
2042 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
2045 struct btrfs_device *device;
2046 struct btrfs_fs_devices *cur_devices;
2047 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2051 mutex_lock(&uuid_mutex);
2053 num_devices = btrfs_num_devices(fs_info);
2055 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2059 device = btrfs_find_device_by_devspec(fs_info, devid, device_path);
2061 if (IS_ERR(device)) {
2062 if (PTR_ERR(device) == -ENOENT &&
2063 strcmp(device_path, "missing") == 0)
2064 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2066 ret = PTR_ERR(device);
2070 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2071 btrfs_warn_in_rcu(fs_info,
2072 "cannot remove device %s (devid %llu) due to active swapfile",
2073 rcu_str_deref(device->name), device->devid);
2078 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2079 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
2083 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2084 fs_info->fs_devices->rw_devices == 1) {
2085 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
2089 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2090 mutex_lock(&fs_info->chunk_mutex);
2091 list_del_init(&device->dev_alloc_list);
2092 device->fs_devices->rw_devices--;
2093 mutex_unlock(&fs_info->chunk_mutex);
2096 mutex_unlock(&uuid_mutex);
2097 ret = btrfs_shrink_device(device, 0);
2098 mutex_lock(&uuid_mutex);
2103 * TODO: the superblock still includes this device in its num_devices
2104 * counter although write_all_supers() is not locked out. This
2105 * could give a filesystem state which requires a degraded mount.
2107 ret = btrfs_rm_dev_item(device);
2111 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2112 btrfs_scrub_cancel_dev(device);
2115 * the device list mutex makes sure that we don't change
2116 * the device list while someone else is writing out all
2117 * the device supers. Whoever is writing all supers, should
2118 * lock the device list mutex before getting the number of
2119 * devices in the super block (super_copy). Conversely,
2120 * whoever updates the number of devices in the super block
2121 * (super_copy) should hold the device list mutex.
2125 * In normal cases the cur_devices == fs_devices. But in case
2126 * of deleting a seed device, the cur_devices should point to
2127 * its own fs_devices listed under the fs_devices->seed.
2129 cur_devices = device->fs_devices;
2130 mutex_lock(&fs_devices->device_list_mutex);
2131 list_del_rcu(&device->dev_list);
2133 cur_devices->num_devices--;
2134 cur_devices->total_devices--;
2135 /* Update total_devices of the parent fs_devices if it's seed */
2136 if (cur_devices != fs_devices)
2137 fs_devices->total_devices--;
2139 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2140 cur_devices->missing_devices--;
2142 btrfs_assign_next_active_device(device, NULL);
2145 cur_devices->open_devices--;
2146 /* remove sysfs entry */
2147 btrfs_sysfs_remove_devices_dir(fs_devices, device);
2150 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2151 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2152 mutex_unlock(&fs_devices->device_list_mutex);
2155 * at this point, the device is zero sized and detached from
2156 * the devices list. All that's left is to zero out the old
2157 * supers and free the device.
2159 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2160 btrfs_scratch_superblocks(fs_info, device->bdev,
2163 btrfs_close_bdev(device);
2165 btrfs_free_device(device);
2167 if (cur_devices->open_devices == 0) {
2168 while (fs_devices) {
2169 if (fs_devices->seed == cur_devices) {
2170 fs_devices->seed = cur_devices->seed;
2173 fs_devices = fs_devices->seed;
2175 cur_devices->seed = NULL;
2176 close_fs_devices(cur_devices);
2177 free_fs_devices(cur_devices);
2181 mutex_unlock(&uuid_mutex);
2185 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2186 mutex_lock(&fs_info->chunk_mutex);
2187 list_add(&device->dev_alloc_list,
2188 &fs_devices->alloc_list);
2189 device->fs_devices->rw_devices++;
2190 mutex_unlock(&fs_info->chunk_mutex);
2195 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2197 struct btrfs_fs_devices *fs_devices;
2199 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2202 * in case of fs with no seed, srcdev->fs_devices will point
2203 * to fs_devices of fs_info. However when the dev being replaced is
2204 * a seed dev it will point to the seed's local fs_devices. In short
2205 * srcdev will have its correct fs_devices in both the cases.
2207 fs_devices = srcdev->fs_devices;
2209 list_del_rcu(&srcdev->dev_list);
2210 list_del(&srcdev->dev_alloc_list);
2211 fs_devices->num_devices--;
2212 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2213 fs_devices->missing_devices--;
2215 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2216 fs_devices->rw_devices--;
2219 fs_devices->open_devices--;
2222 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2224 struct btrfs_fs_info *fs_info = srcdev->fs_info;
2225 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2227 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) {
2228 /* zero out the old super if it is writable */
2229 btrfs_scratch_superblocks(fs_info, srcdev->bdev,
2233 btrfs_close_bdev(srcdev);
2235 btrfs_free_device(srcdev);
2237 /* if this is no devs we rather delete the fs_devices */
2238 if (!fs_devices->num_devices) {
2239 struct btrfs_fs_devices *tmp_fs_devices;
2242 * On a mounted FS, num_devices can't be zero unless it's a
2243 * seed. In case of a seed device being replaced, the replace
2244 * target added to the sprout FS, so there will be no more
2245 * device left under the seed FS.
2247 ASSERT(fs_devices->seeding);
2249 tmp_fs_devices = fs_info->fs_devices;
2250 while (tmp_fs_devices) {
2251 if (tmp_fs_devices->seed == fs_devices) {
2252 tmp_fs_devices->seed = fs_devices->seed;
2255 tmp_fs_devices = tmp_fs_devices->seed;
2257 fs_devices->seed = NULL;
2258 close_fs_devices(fs_devices);
2259 free_fs_devices(fs_devices);
2263 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2265 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2267 mutex_lock(&fs_devices->device_list_mutex);
2269 btrfs_sysfs_remove_devices_dir(fs_devices, tgtdev);
2272 fs_devices->open_devices--;
2274 fs_devices->num_devices--;
2276 btrfs_assign_next_active_device(tgtdev, NULL);
2278 list_del_rcu(&tgtdev->dev_list);
2280 mutex_unlock(&fs_devices->device_list_mutex);
2283 * The update_dev_time() with in btrfs_scratch_superblocks()
2284 * may lead to a call to btrfs_show_devname() which will try
2285 * to hold device_list_mutex. And here this device
2286 * is already out of device list, so we don't have to hold
2287 * the device_list_mutex lock.
2289 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2292 btrfs_close_bdev(tgtdev);
2294 btrfs_free_device(tgtdev);
2297 static struct btrfs_device *btrfs_find_device_by_path(
2298 struct btrfs_fs_info *fs_info, const char *device_path)
2301 struct btrfs_super_block *disk_super;
2304 struct block_device *bdev;
2305 struct btrfs_device *device;
2307 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2308 fs_info->bdev_holder, 0, &bdev, &disk_super);
2310 return ERR_PTR(ret);
2312 devid = btrfs_stack_device_id(&disk_super->dev_item);
2313 dev_uuid = disk_super->dev_item.uuid;
2314 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2315 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2316 disk_super->metadata_uuid, true);
2318 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2319 disk_super->fsid, true);
2321 btrfs_release_disk_super(disk_super);
2323 device = ERR_PTR(-ENOENT);
2324 blkdev_put(bdev, FMODE_READ);
2329 * Lookup a device given by device id, or the path if the id is 0.
2331 struct btrfs_device *btrfs_find_device_by_devspec(
2332 struct btrfs_fs_info *fs_info, u64 devid,
2333 const char *device_path)
2335 struct btrfs_device *device;
2338 device = btrfs_find_device(fs_info->fs_devices, devid, NULL,
2341 return ERR_PTR(-ENOENT);
2345 if (!device_path || !device_path[0])
2346 return ERR_PTR(-EINVAL);
2348 if (strcmp(device_path, "missing") == 0) {
2349 /* Find first missing device */
2350 list_for_each_entry(device, &fs_info->fs_devices->devices,
2352 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2353 &device->dev_state) && !device->bdev)
2356 return ERR_PTR(-ENOENT);
2359 return btrfs_find_device_by_path(fs_info, device_path);
2363 * does all the dirty work required for changing file system's UUID.
2365 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2367 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2368 struct btrfs_fs_devices *old_devices;
2369 struct btrfs_fs_devices *seed_devices;
2370 struct btrfs_super_block *disk_super = fs_info->super_copy;
2371 struct btrfs_device *device;
2374 lockdep_assert_held(&uuid_mutex);
2375 if (!fs_devices->seeding)
2378 seed_devices = alloc_fs_devices(NULL, NULL);
2379 if (IS_ERR(seed_devices))
2380 return PTR_ERR(seed_devices);
2382 old_devices = clone_fs_devices(fs_devices);
2383 if (IS_ERR(old_devices)) {
2384 kfree(seed_devices);
2385 return PTR_ERR(old_devices);
2388 list_add(&old_devices->fs_list, &fs_uuids);
2390 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2391 seed_devices->opened = 1;
2392 INIT_LIST_HEAD(&seed_devices->devices);
2393 INIT_LIST_HEAD(&seed_devices->alloc_list);
2394 mutex_init(&seed_devices->device_list_mutex);
2396 mutex_lock(&fs_devices->device_list_mutex);
2397 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2399 list_for_each_entry(device, &seed_devices->devices, dev_list)
2400 device->fs_devices = seed_devices;
2402 mutex_lock(&fs_info->chunk_mutex);
2403 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2404 mutex_unlock(&fs_info->chunk_mutex);
2406 fs_devices->seeding = false;
2407 fs_devices->num_devices = 0;
2408 fs_devices->open_devices = 0;
2409 fs_devices->missing_devices = 0;
2410 fs_devices->rotating = false;
2411 fs_devices->seed = seed_devices;
2413 generate_random_uuid(fs_devices->fsid);
2414 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2415 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2416 mutex_unlock(&fs_devices->device_list_mutex);
2418 super_flags = btrfs_super_flags(disk_super) &
2419 ~BTRFS_SUPER_FLAG_SEEDING;
2420 btrfs_set_super_flags(disk_super, super_flags);
2426 * Store the expected generation for seed devices in device items.
2428 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2430 struct btrfs_fs_info *fs_info = trans->fs_info;
2431 struct btrfs_root *root = fs_info->chunk_root;
2432 struct btrfs_path *path;
2433 struct extent_buffer *leaf;
2434 struct btrfs_dev_item *dev_item;
2435 struct btrfs_device *device;
2436 struct btrfs_key key;
2437 u8 fs_uuid[BTRFS_FSID_SIZE];
2438 u8 dev_uuid[BTRFS_UUID_SIZE];
2442 path = btrfs_alloc_path();
2446 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2448 key.type = BTRFS_DEV_ITEM_KEY;
2451 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2455 leaf = path->nodes[0];
2457 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2458 ret = btrfs_next_leaf(root, path);
2463 leaf = path->nodes[0];
2464 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2465 btrfs_release_path(path);
2469 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2470 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2471 key.type != BTRFS_DEV_ITEM_KEY)
2474 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2475 struct btrfs_dev_item);
2476 devid = btrfs_device_id(leaf, dev_item);
2477 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2479 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2481 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2483 BUG_ON(!device); /* Logic error */
2485 if (device->fs_devices->seeding) {
2486 btrfs_set_device_generation(leaf, dev_item,
2487 device->generation);
2488 btrfs_mark_buffer_dirty(leaf);
2496 btrfs_free_path(path);
2500 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2502 struct btrfs_root *root = fs_info->dev_root;
2503 struct request_queue *q;
2504 struct btrfs_trans_handle *trans;
2505 struct btrfs_device *device;
2506 struct block_device *bdev;
2507 struct super_block *sb = fs_info->sb;
2508 struct rcu_string *name;
2509 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2510 u64 orig_super_total_bytes;
2511 u64 orig_super_num_devices;
2512 int seeding_dev = 0;
2514 bool unlocked = false;
2516 if (sb_rdonly(sb) && !fs_devices->seeding)
2519 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2520 fs_info->bdev_holder);
2522 return PTR_ERR(bdev);
2524 if (fs_devices->seeding) {
2526 down_write(&sb->s_umount);
2527 mutex_lock(&uuid_mutex);
2530 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2532 mutex_lock(&fs_devices->device_list_mutex);
2533 list_for_each_entry(device, &fs_devices->devices, dev_list) {
2534 if (device->bdev == bdev) {
2537 &fs_devices->device_list_mutex);
2541 mutex_unlock(&fs_devices->device_list_mutex);
2543 device = btrfs_alloc_device(fs_info, NULL, NULL);
2544 if (IS_ERR(device)) {
2545 /* we can safely leave the fs_devices entry around */
2546 ret = PTR_ERR(device);
2550 name = rcu_string_strdup(device_path, GFP_KERNEL);
2553 goto error_free_device;
2555 rcu_assign_pointer(device->name, name);
2557 trans = btrfs_start_transaction(root, 0);
2558 if (IS_ERR(trans)) {
2559 ret = PTR_ERR(trans);
2560 goto error_free_device;
2563 q = bdev_get_queue(bdev);
2564 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2565 device->generation = trans->transid;
2566 device->io_width = fs_info->sectorsize;
2567 device->io_align = fs_info->sectorsize;
2568 device->sector_size = fs_info->sectorsize;
2569 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2570 fs_info->sectorsize);
2571 device->disk_total_bytes = device->total_bytes;
2572 device->commit_total_bytes = device->total_bytes;
2573 device->fs_info = fs_info;
2574 device->bdev = bdev;
2575 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2576 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2577 device->mode = FMODE_EXCL;
2578 device->dev_stats_valid = 1;
2579 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2582 sb->s_flags &= ~SB_RDONLY;
2583 ret = btrfs_prepare_sprout(fs_info);
2585 btrfs_abort_transaction(trans, ret);
2590 device->fs_devices = fs_devices;
2592 mutex_lock(&fs_devices->device_list_mutex);
2593 mutex_lock(&fs_info->chunk_mutex);
2594 list_add_rcu(&device->dev_list, &fs_devices->devices);
2595 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2596 fs_devices->num_devices++;
2597 fs_devices->open_devices++;
2598 fs_devices->rw_devices++;
2599 fs_devices->total_devices++;
2600 fs_devices->total_rw_bytes += device->total_bytes;
2602 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2604 if (!blk_queue_nonrot(q))
2605 fs_devices->rotating = true;
2607 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2608 btrfs_set_super_total_bytes(fs_info->super_copy,
2609 round_down(orig_super_total_bytes + device->total_bytes,
2610 fs_info->sectorsize));
2612 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2613 btrfs_set_super_num_devices(fs_info->super_copy,
2614 orig_super_num_devices + 1);
2616 /* add sysfs device entry */
2617 btrfs_sysfs_add_devices_dir(fs_devices, device);
2620 * we've got more storage, clear any full flags on the space
2623 btrfs_clear_space_info_full(fs_info);
2625 mutex_unlock(&fs_info->chunk_mutex);
2626 mutex_unlock(&fs_devices->device_list_mutex);
2629 mutex_lock(&fs_info->chunk_mutex);
2630 ret = init_first_rw_device(trans);
2631 mutex_unlock(&fs_info->chunk_mutex);
2633 btrfs_abort_transaction(trans, ret);
2638 ret = btrfs_add_dev_item(trans, device);
2640 btrfs_abort_transaction(trans, ret);
2645 ret = btrfs_finish_sprout(trans);
2647 btrfs_abort_transaction(trans, ret);
2651 btrfs_sysfs_update_sprout_fsid(fs_devices,
2652 fs_info->fs_devices->fsid);
2655 ret = btrfs_commit_transaction(trans);
2658 mutex_unlock(&uuid_mutex);
2659 up_write(&sb->s_umount);
2662 if (ret) /* transaction commit */
2665 ret = btrfs_relocate_sys_chunks(fs_info);
2667 btrfs_handle_fs_error(fs_info, ret,
2668 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2669 trans = btrfs_attach_transaction(root);
2670 if (IS_ERR(trans)) {
2671 if (PTR_ERR(trans) == -ENOENT)
2673 ret = PTR_ERR(trans);
2677 ret = btrfs_commit_transaction(trans);
2681 * Now that we have written a new super block to this device, check all
2682 * other fs_devices list if device_path alienates any other scanned
2684 * We can ignore the return value as it typically returns -EINVAL and
2685 * only succeeds if the device was an alien.
2687 btrfs_forget_devices(device_path);
2689 /* Update ctime/mtime for blkid or udev */
2690 update_dev_time(device_path);
2695 btrfs_sysfs_remove_devices_dir(fs_devices, device);
2696 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2697 mutex_lock(&fs_info->chunk_mutex);
2698 list_del_rcu(&device->dev_list);
2699 list_del(&device->dev_alloc_list);
2700 fs_info->fs_devices->num_devices--;
2701 fs_info->fs_devices->open_devices--;
2702 fs_info->fs_devices->rw_devices--;
2703 fs_info->fs_devices->total_devices--;
2704 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2705 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2706 btrfs_set_super_total_bytes(fs_info->super_copy,
2707 orig_super_total_bytes);
2708 btrfs_set_super_num_devices(fs_info->super_copy,
2709 orig_super_num_devices);
2710 mutex_unlock(&fs_info->chunk_mutex);
2711 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2714 sb->s_flags |= SB_RDONLY;
2716 btrfs_end_transaction(trans);
2718 btrfs_free_device(device);
2720 blkdev_put(bdev, FMODE_EXCL);
2721 if (seeding_dev && !unlocked) {
2722 mutex_unlock(&uuid_mutex);
2723 up_write(&sb->s_umount);
2728 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2729 struct btrfs_device *device)
2732 struct btrfs_path *path;
2733 struct btrfs_root *root = device->fs_info->chunk_root;
2734 struct btrfs_dev_item *dev_item;
2735 struct extent_buffer *leaf;
2736 struct btrfs_key key;
2738 path = btrfs_alloc_path();
2742 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2743 key.type = BTRFS_DEV_ITEM_KEY;
2744 key.offset = device->devid;
2746 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2755 leaf = path->nodes[0];
2756 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2758 btrfs_set_device_id(leaf, dev_item, device->devid);
2759 btrfs_set_device_type(leaf, dev_item, device->type);
2760 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2761 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2762 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2763 btrfs_set_device_total_bytes(leaf, dev_item,
2764 btrfs_device_get_disk_total_bytes(device));
2765 btrfs_set_device_bytes_used(leaf, dev_item,
2766 btrfs_device_get_bytes_used(device));
2767 btrfs_mark_buffer_dirty(leaf);
2770 btrfs_free_path(path);
2774 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2775 struct btrfs_device *device, u64 new_size)
2777 struct btrfs_fs_info *fs_info = device->fs_info;
2778 struct btrfs_super_block *super_copy = fs_info->super_copy;
2782 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2785 new_size = round_down(new_size, fs_info->sectorsize);
2787 mutex_lock(&fs_info->chunk_mutex);
2788 old_total = btrfs_super_total_bytes(super_copy);
2789 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2791 if (new_size <= device->total_bytes ||
2792 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2793 mutex_unlock(&fs_info->chunk_mutex);
2797 btrfs_set_super_total_bytes(super_copy,
2798 round_down(old_total + diff, fs_info->sectorsize));
2799 device->fs_devices->total_rw_bytes += diff;
2801 btrfs_device_set_total_bytes(device, new_size);
2802 btrfs_device_set_disk_total_bytes(device, new_size);
2803 btrfs_clear_space_info_full(device->fs_info);
2804 if (list_empty(&device->post_commit_list))
2805 list_add_tail(&device->post_commit_list,
2806 &trans->transaction->dev_update_list);
2807 mutex_unlock(&fs_info->chunk_mutex);
2809 return btrfs_update_device(trans, device);
2812 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2814 struct btrfs_fs_info *fs_info = trans->fs_info;
2815 struct btrfs_root *root = fs_info->chunk_root;
2817 struct btrfs_path *path;
2818 struct btrfs_key key;
2820 path = btrfs_alloc_path();
2824 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2825 key.offset = chunk_offset;
2826 key.type = BTRFS_CHUNK_ITEM_KEY;
2828 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2831 else if (ret > 0) { /* Logic error or corruption */
2832 btrfs_handle_fs_error(fs_info, -ENOENT,
2833 "Failed lookup while freeing chunk.");
2838 ret = btrfs_del_item(trans, root, path);
2840 btrfs_handle_fs_error(fs_info, ret,
2841 "Failed to delete chunk item.");
2843 btrfs_free_path(path);
2847 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2849 struct btrfs_super_block *super_copy = fs_info->super_copy;
2850 struct btrfs_disk_key *disk_key;
2851 struct btrfs_chunk *chunk;
2858 struct btrfs_key key;
2860 mutex_lock(&fs_info->chunk_mutex);
2861 array_size = btrfs_super_sys_array_size(super_copy);
2863 ptr = super_copy->sys_chunk_array;
2866 while (cur < array_size) {
2867 disk_key = (struct btrfs_disk_key *)ptr;
2868 btrfs_disk_key_to_cpu(&key, disk_key);
2870 len = sizeof(*disk_key);
2872 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2873 chunk = (struct btrfs_chunk *)(ptr + len);
2874 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2875 len += btrfs_chunk_item_size(num_stripes);
2880 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2881 key.offset == chunk_offset) {
2882 memmove(ptr, ptr + len, array_size - (cur + len));
2884 btrfs_set_super_sys_array_size(super_copy, array_size);
2890 mutex_unlock(&fs_info->chunk_mutex);
2895 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
2896 * @logical: Logical block offset in bytes.
2897 * @length: Length of extent in bytes.
2899 * Return: Chunk mapping or ERR_PTR.
2901 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
2902 u64 logical, u64 length)
2904 struct extent_map_tree *em_tree;
2905 struct extent_map *em;
2907 em_tree = &fs_info->mapping_tree;
2908 read_lock(&em_tree->lock);
2909 em = lookup_extent_mapping(em_tree, logical, length);
2910 read_unlock(&em_tree->lock);
2913 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2915 return ERR_PTR(-EINVAL);
2918 if (em->start > logical || em->start + em->len < logical) {
2920 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2921 logical, length, em->start, em->start + em->len);
2922 free_extent_map(em);
2923 return ERR_PTR(-EINVAL);
2926 /* callers are responsible for dropping em's ref. */
2930 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2932 struct btrfs_fs_info *fs_info = trans->fs_info;
2933 struct extent_map *em;
2934 struct map_lookup *map;
2935 u64 dev_extent_len = 0;
2937 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2939 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
2942 * This is a logic error, but we don't want to just rely on the
2943 * user having built with ASSERT enabled, so if ASSERT doesn't
2944 * do anything we still error out.
2949 map = em->map_lookup;
2950 mutex_lock(&fs_info->chunk_mutex);
2951 check_system_chunk(trans, map->type);
2952 mutex_unlock(&fs_info->chunk_mutex);
2955 * Take the device list mutex to prevent races with the final phase of
2956 * a device replace operation that replaces the device object associated
2957 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2959 mutex_lock(&fs_devices->device_list_mutex);
2960 for (i = 0; i < map->num_stripes; i++) {
2961 struct btrfs_device *device = map->stripes[i].dev;
2962 ret = btrfs_free_dev_extent(trans, device,
2963 map->stripes[i].physical,
2966 mutex_unlock(&fs_devices->device_list_mutex);
2967 btrfs_abort_transaction(trans, ret);
2971 if (device->bytes_used > 0) {
2972 mutex_lock(&fs_info->chunk_mutex);
2973 btrfs_device_set_bytes_used(device,
2974 device->bytes_used - dev_extent_len);
2975 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2976 btrfs_clear_space_info_full(fs_info);
2977 mutex_unlock(&fs_info->chunk_mutex);
2980 ret = btrfs_update_device(trans, device);
2982 mutex_unlock(&fs_devices->device_list_mutex);
2983 btrfs_abort_transaction(trans, ret);
2987 mutex_unlock(&fs_devices->device_list_mutex);
2989 ret = btrfs_free_chunk(trans, chunk_offset);
2991 btrfs_abort_transaction(trans, ret);
2995 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2997 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2998 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3000 btrfs_abort_transaction(trans, ret);
3005 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3007 btrfs_abort_transaction(trans, ret);
3013 free_extent_map(em);
3017 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3019 struct btrfs_root *root = fs_info->chunk_root;
3020 struct btrfs_trans_handle *trans;
3021 struct btrfs_block_group *block_group;
3025 * Prevent races with automatic removal of unused block groups.
3026 * After we relocate and before we remove the chunk with offset
3027 * chunk_offset, automatic removal of the block group can kick in,
3028 * resulting in a failure when calling btrfs_remove_chunk() below.
3030 * Make sure to acquire this mutex before doing a tree search (dev
3031 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3032 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3033 * we release the path used to search the chunk/dev tree and before
3034 * the current task acquires this mutex and calls us.
3036 lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
3038 /* step one, relocate all the extents inside this chunk */
3039 btrfs_scrub_pause(fs_info);
3040 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3041 btrfs_scrub_continue(fs_info);
3045 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3048 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3049 btrfs_put_block_group(block_group);
3051 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3053 if (IS_ERR(trans)) {
3054 ret = PTR_ERR(trans);
3055 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3060 * step two, delete the device extents and the
3061 * chunk tree entries
3063 ret = btrfs_remove_chunk(trans, chunk_offset);
3064 btrfs_end_transaction(trans);
3068 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3070 struct btrfs_root *chunk_root = fs_info->chunk_root;
3071 struct btrfs_path *path;
3072 struct extent_buffer *leaf;
3073 struct btrfs_chunk *chunk;
3074 struct btrfs_key key;
3075 struct btrfs_key found_key;
3077 bool retried = false;
3081 path = btrfs_alloc_path();
3086 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3087 key.offset = (u64)-1;
3088 key.type = BTRFS_CHUNK_ITEM_KEY;
3091 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3092 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3094 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3097 BUG_ON(ret == 0); /* Corruption */
3099 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3102 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3108 leaf = path->nodes[0];
3109 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3111 chunk = btrfs_item_ptr(leaf, path->slots[0],
3112 struct btrfs_chunk);
3113 chunk_type = btrfs_chunk_type(leaf, chunk);
3114 btrfs_release_path(path);
3116 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3117 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3123 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3125 if (found_key.offset == 0)
3127 key.offset = found_key.offset - 1;
3130 if (failed && !retried) {
3134 } else if (WARN_ON(failed && retried)) {
3138 btrfs_free_path(path);
3143 * return 1 : allocate a data chunk successfully,
3144 * return <0: errors during allocating a data chunk,
3145 * return 0 : no need to allocate a data chunk.
3147 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3150 struct btrfs_block_group *cache;
3154 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3156 chunk_type = cache->flags;
3157 btrfs_put_block_group(cache);
3159 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3162 spin_lock(&fs_info->data_sinfo->lock);
3163 bytes_used = fs_info->data_sinfo->bytes_used;
3164 spin_unlock(&fs_info->data_sinfo->lock);
3167 struct btrfs_trans_handle *trans;
3170 trans = btrfs_join_transaction(fs_info->tree_root);
3172 return PTR_ERR(trans);
3174 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3175 btrfs_end_transaction(trans);
3184 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3185 struct btrfs_balance_control *bctl)
3187 struct btrfs_root *root = fs_info->tree_root;
3188 struct btrfs_trans_handle *trans;
3189 struct btrfs_balance_item *item;
3190 struct btrfs_disk_balance_args disk_bargs;
3191 struct btrfs_path *path;
3192 struct extent_buffer *leaf;
3193 struct btrfs_key key;
3196 path = btrfs_alloc_path();
3200 trans = btrfs_start_transaction(root, 0);
3201 if (IS_ERR(trans)) {
3202 btrfs_free_path(path);
3203 return PTR_ERR(trans);
3206 key.objectid = BTRFS_BALANCE_OBJECTID;
3207 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3210 ret = btrfs_insert_empty_item(trans, root, path, &key,
3215 leaf = path->nodes[0];
3216 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3218 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3220 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3221 btrfs_set_balance_data(leaf, item, &disk_bargs);
3222 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3223 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3224 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3225 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3227 btrfs_set_balance_flags(leaf, item, bctl->flags);
3229 btrfs_mark_buffer_dirty(leaf);
3231 btrfs_free_path(path);
3232 err = btrfs_commit_transaction(trans);
3238 static int del_balance_item(struct btrfs_fs_info *fs_info)
3240 struct btrfs_root *root = fs_info->tree_root;
3241 struct btrfs_trans_handle *trans;
3242 struct btrfs_path *path;
3243 struct btrfs_key key;
3246 path = btrfs_alloc_path();
3250 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3251 if (IS_ERR(trans)) {
3252 btrfs_free_path(path);
3253 return PTR_ERR(trans);
3256 key.objectid = BTRFS_BALANCE_OBJECTID;
3257 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3260 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3268 ret = btrfs_del_item(trans, root, path);
3270 btrfs_free_path(path);
3271 err = btrfs_commit_transaction(trans);
3278 * This is a heuristic used to reduce the number of chunks balanced on
3279 * resume after balance was interrupted.
3281 static void update_balance_args(struct btrfs_balance_control *bctl)
3284 * Turn on soft mode for chunk types that were being converted.
3286 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3287 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3288 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3289 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3290 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3291 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3294 * Turn on usage filter if is not already used. The idea is
3295 * that chunks that we have already balanced should be
3296 * reasonably full. Don't do it for chunks that are being
3297 * converted - that will keep us from relocating unconverted
3298 * (albeit full) chunks.
3300 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3301 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3302 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3303 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3304 bctl->data.usage = 90;
3306 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3307 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3308 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3309 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3310 bctl->sys.usage = 90;
3312 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3313 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3314 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3315 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3316 bctl->meta.usage = 90;
3321 * Clear the balance status in fs_info and delete the balance item from disk.
3323 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3325 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3328 BUG_ON(!fs_info->balance_ctl);
3330 spin_lock(&fs_info->balance_lock);
3331 fs_info->balance_ctl = NULL;
3332 spin_unlock(&fs_info->balance_lock);
3335 ret = del_balance_item(fs_info);
3337 btrfs_handle_fs_error(fs_info, ret, NULL);
3341 * Balance filters. Return 1 if chunk should be filtered out
3342 * (should not be balanced).
3344 static int chunk_profiles_filter(u64 chunk_type,
3345 struct btrfs_balance_args *bargs)
3347 chunk_type = chunk_to_extended(chunk_type) &
3348 BTRFS_EXTENDED_PROFILE_MASK;
3350 if (bargs->profiles & chunk_type)
3356 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3357 struct btrfs_balance_args *bargs)
3359 struct btrfs_block_group *cache;
3361 u64 user_thresh_min;
3362 u64 user_thresh_max;
3365 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3366 chunk_used = cache->used;
3368 if (bargs->usage_min == 0)
3369 user_thresh_min = 0;
3371 user_thresh_min = div_factor_fine(cache->length,
3374 if (bargs->usage_max == 0)
3375 user_thresh_max = 1;
3376 else if (bargs->usage_max > 100)
3377 user_thresh_max = cache->length;
3379 user_thresh_max = div_factor_fine(cache->length,
3382 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3385 btrfs_put_block_group(cache);
3389 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3390 u64 chunk_offset, struct btrfs_balance_args *bargs)
3392 struct btrfs_block_group *cache;
3393 u64 chunk_used, user_thresh;
3396 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3397 chunk_used = cache->used;
3399 if (bargs->usage_min == 0)
3401 else if (bargs->usage > 100)
3402 user_thresh = cache->length;
3404 user_thresh = div_factor_fine(cache->length, bargs->usage);
3406 if (chunk_used < user_thresh)
3409 btrfs_put_block_group(cache);
3413 static int chunk_devid_filter(struct extent_buffer *leaf,
3414 struct btrfs_chunk *chunk,
3415 struct btrfs_balance_args *bargs)
3417 struct btrfs_stripe *stripe;
3418 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3421 for (i = 0; i < num_stripes; i++) {
3422 stripe = btrfs_stripe_nr(chunk, i);
3423 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3430 static u64 calc_data_stripes(u64 type, int num_stripes)
3432 const int index = btrfs_bg_flags_to_raid_index(type);
3433 const int ncopies = btrfs_raid_array[index].ncopies;
3434 const int nparity = btrfs_raid_array[index].nparity;
3437 return num_stripes - nparity;
3439 return num_stripes / ncopies;
3442 /* [pstart, pend) */
3443 static int chunk_drange_filter(struct extent_buffer *leaf,
3444 struct btrfs_chunk *chunk,
3445 struct btrfs_balance_args *bargs)
3447 struct btrfs_stripe *stripe;
3448 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3455 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3458 type = btrfs_chunk_type(leaf, chunk);
3459 factor = calc_data_stripes(type, num_stripes);
3461 for (i = 0; i < num_stripes; i++) {
3462 stripe = btrfs_stripe_nr(chunk, i);
3463 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3466 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3467 stripe_length = btrfs_chunk_length(leaf, chunk);
3468 stripe_length = div_u64(stripe_length, factor);
3470 if (stripe_offset < bargs->pend &&
3471 stripe_offset + stripe_length > bargs->pstart)
3478 /* [vstart, vend) */
3479 static int chunk_vrange_filter(struct extent_buffer *leaf,
3480 struct btrfs_chunk *chunk,
3482 struct btrfs_balance_args *bargs)
3484 if (chunk_offset < bargs->vend &&
3485 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3486 /* at least part of the chunk is inside this vrange */
3492 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3493 struct btrfs_chunk *chunk,
3494 struct btrfs_balance_args *bargs)
3496 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3498 if (bargs->stripes_min <= num_stripes
3499 && num_stripes <= bargs->stripes_max)
3505 static int chunk_soft_convert_filter(u64 chunk_type,
3506 struct btrfs_balance_args *bargs)
3508 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3511 chunk_type = chunk_to_extended(chunk_type) &
3512 BTRFS_EXTENDED_PROFILE_MASK;
3514 if (bargs->target == chunk_type)
3520 static int should_balance_chunk(struct extent_buffer *leaf,
3521 struct btrfs_chunk *chunk, u64 chunk_offset)
3523 struct btrfs_fs_info *fs_info = leaf->fs_info;
3524 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3525 struct btrfs_balance_args *bargs = NULL;
3526 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3529 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3530 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3534 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3535 bargs = &bctl->data;
3536 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3538 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3539 bargs = &bctl->meta;
3541 /* profiles filter */
3542 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3543 chunk_profiles_filter(chunk_type, bargs)) {
3548 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3549 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3551 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3552 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3557 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3558 chunk_devid_filter(leaf, chunk, bargs)) {
3562 /* drange filter, makes sense only with devid filter */
3563 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3564 chunk_drange_filter(leaf, chunk, bargs)) {
3569 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3570 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3574 /* stripes filter */
3575 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3576 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3580 /* soft profile changing mode */
3581 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3582 chunk_soft_convert_filter(chunk_type, bargs)) {
3587 * limited by count, must be the last filter
3589 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3590 if (bargs->limit == 0)
3594 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3596 * Same logic as the 'limit' filter; the minimum cannot be
3597 * determined here because we do not have the global information
3598 * about the count of all chunks that satisfy the filters.
3600 if (bargs->limit_max == 0)
3609 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3611 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3612 struct btrfs_root *chunk_root = fs_info->chunk_root;
3614 struct btrfs_chunk *chunk;
3615 struct btrfs_path *path = NULL;
3616 struct btrfs_key key;
3617 struct btrfs_key found_key;
3618 struct extent_buffer *leaf;
3621 int enospc_errors = 0;
3622 bool counting = true;
3623 /* The single value limit and min/max limits use the same bytes in the */
3624 u64 limit_data = bctl->data.limit;
3625 u64 limit_meta = bctl->meta.limit;
3626 u64 limit_sys = bctl->sys.limit;
3630 int chunk_reserved = 0;
3632 path = btrfs_alloc_path();
3638 /* zero out stat counters */
3639 spin_lock(&fs_info->balance_lock);
3640 memset(&bctl->stat, 0, sizeof(bctl->stat));
3641 spin_unlock(&fs_info->balance_lock);
3645 * The single value limit and min/max limits use the same bytes
3648 bctl->data.limit = limit_data;
3649 bctl->meta.limit = limit_meta;
3650 bctl->sys.limit = limit_sys;
3652 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3653 key.offset = (u64)-1;
3654 key.type = BTRFS_CHUNK_ITEM_KEY;
3657 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3658 atomic_read(&fs_info->balance_cancel_req)) {
3663 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3664 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3666 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3671 * this shouldn't happen, it means the last relocate
3675 BUG(); /* FIXME break ? */
3677 ret = btrfs_previous_item(chunk_root, path, 0,
3678 BTRFS_CHUNK_ITEM_KEY);
3680 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3685 leaf = path->nodes[0];
3686 slot = path->slots[0];
3687 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3689 if (found_key.objectid != key.objectid) {
3690 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3694 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3695 chunk_type = btrfs_chunk_type(leaf, chunk);
3698 spin_lock(&fs_info->balance_lock);
3699 bctl->stat.considered++;
3700 spin_unlock(&fs_info->balance_lock);
3703 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3705 btrfs_release_path(path);
3707 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3712 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3713 spin_lock(&fs_info->balance_lock);
3714 bctl->stat.expected++;
3715 spin_unlock(&fs_info->balance_lock);
3717 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3719 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3721 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3728 * Apply limit_min filter, no need to check if the LIMITS
3729 * filter is used, limit_min is 0 by default
3731 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3732 count_data < bctl->data.limit_min)
3733 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3734 count_meta < bctl->meta.limit_min)
3735 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3736 count_sys < bctl->sys.limit_min)) {
3737 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3741 if (!chunk_reserved) {
3743 * We may be relocating the only data chunk we have,
3744 * which could potentially end up with losing data's
3745 * raid profile, so lets allocate an empty one in
3748 ret = btrfs_may_alloc_data_chunk(fs_info,
3751 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3753 } else if (ret == 1) {
3758 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3759 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3760 if (ret == -ENOSPC) {
3762 } else if (ret == -ETXTBSY) {
3764 "skipping relocation of block group %llu due to active swapfile",
3770 spin_lock(&fs_info->balance_lock);
3771 bctl->stat.completed++;
3772 spin_unlock(&fs_info->balance_lock);
3775 if (found_key.offset == 0)
3777 key.offset = found_key.offset - 1;
3781 btrfs_release_path(path);
3786 btrfs_free_path(path);
3787 if (enospc_errors) {
3788 btrfs_info(fs_info, "%d enospc errors during balance",
3798 * alloc_profile_is_valid - see if a given profile is valid and reduced
3799 * @flags: profile to validate
3800 * @extended: if true @flags is treated as an extended profile
3802 static int alloc_profile_is_valid(u64 flags, int extended)
3804 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3805 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3807 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3809 /* 1) check that all other bits are zeroed */
3813 /* 2) see if profile is reduced */
3815 return !extended; /* "0" is valid for usual profiles */
3817 return has_single_bit_set(flags);
3820 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3822 /* cancel requested || normal exit path */
3823 return atomic_read(&fs_info->balance_cancel_req) ||
3824 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3825 atomic_read(&fs_info->balance_cancel_req) == 0);
3829 * Validate target profile against allowed profiles and return true if it's OK.
3830 * Otherwise print the error message and return false.
3832 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
3833 const struct btrfs_balance_args *bargs,
3834 u64 allowed, const char *type)
3836 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3839 /* Profile is valid and does not have bits outside of the allowed set */
3840 if (alloc_profile_is_valid(bargs->target, 1) &&
3841 (bargs->target & ~allowed) == 0)
3844 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
3845 type, btrfs_bg_type_to_raid_name(bargs->target));
3850 * Fill @buf with textual description of balance filter flags @bargs, up to
3851 * @size_buf including the terminating null. The output may be trimmed if it
3852 * does not fit into the provided buffer.
3854 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
3858 u32 size_bp = size_buf;
3860 u64 flags = bargs->flags;
3861 char tmp_buf[128] = {'\0'};
3866 #define CHECK_APPEND_NOARG(a) \
3868 ret = snprintf(bp, size_bp, (a)); \
3869 if (ret < 0 || ret >= size_bp) \
3870 goto out_overflow; \
3875 #define CHECK_APPEND_1ARG(a, v1) \
3877 ret = snprintf(bp, size_bp, (a), (v1)); \
3878 if (ret < 0 || ret >= size_bp) \
3879 goto out_overflow; \
3884 #define CHECK_APPEND_2ARG(a, v1, v2) \
3886 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
3887 if (ret < 0 || ret >= size_bp) \
3888 goto out_overflow; \
3893 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
3894 CHECK_APPEND_1ARG("convert=%s,",
3895 btrfs_bg_type_to_raid_name(bargs->target));
3897 if (flags & BTRFS_BALANCE_ARGS_SOFT)
3898 CHECK_APPEND_NOARG("soft,");
3900 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
3901 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
3903 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
3906 if (flags & BTRFS_BALANCE_ARGS_USAGE)
3907 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
3909 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
3910 CHECK_APPEND_2ARG("usage=%u..%u,",
3911 bargs->usage_min, bargs->usage_max);
3913 if (flags & BTRFS_BALANCE_ARGS_DEVID)
3914 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
3916 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
3917 CHECK_APPEND_2ARG("drange=%llu..%llu,",
3918 bargs->pstart, bargs->pend);
3920 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
3921 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
3922 bargs->vstart, bargs->vend);
3924 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
3925 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
3927 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
3928 CHECK_APPEND_2ARG("limit=%u..%u,",
3929 bargs->limit_min, bargs->limit_max);
3931 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
3932 CHECK_APPEND_2ARG("stripes=%u..%u,",
3933 bargs->stripes_min, bargs->stripes_max);
3935 #undef CHECK_APPEND_2ARG
3936 #undef CHECK_APPEND_1ARG
3937 #undef CHECK_APPEND_NOARG
3941 if (size_bp < size_buf)
3942 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
3947 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
3949 u32 size_buf = 1024;
3950 char tmp_buf[192] = {'\0'};
3953 u32 size_bp = size_buf;
3955 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3957 buf = kzalloc(size_buf, GFP_KERNEL);
3963 #define CHECK_APPEND_1ARG(a, v1) \
3965 ret = snprintf(bp, size_bp, (a), (v1)); \
3966 if (ret < 0 || ret >= size_bp) \
3967 goto out_overflow; \
3972 if (bctl->flags & BTRFS_BALANCE_FORCE)
3973 CHECK_APPEND_1ARG("%s", "-f ");
3975 if (bctl->flags & BTRFS_BALANCE_DATA) {
3976 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
3977 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
3980 if (bctl->flags & BTRFS_BALANCE_METADATA) {
3981 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
3982 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
3985 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
3986 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
3987 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
3990 #undef CHECK_APPEND_1ARG
3994 if (size_bp < size_buf)
3995 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
3996 btrfs_info(fs_info, "balance: %s %s",
3997 (bctl->flags & BTRFS_BALANCE_RESUME) ?
3998 "resume" : "start", buf);
4004 * Should be called with balance mutexe held
4006 int btrfs_balance(struct btrfs_fs_info *fs_info,
4007 struct btrfs_balance_control *bctl,
4008 struct btrfs_ioctl_balance_args *bargs)
4010 u64 meta_target, data_target;
4016 bool reducing_redundancy;
4019 if (btrfs_fs_closing(fs_info) ||
4020 atomic_read(&fs_info->balance_pause_req) ||
4021 btrfs_should_cancel_balance(fs_info)) {
4026 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4027 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4031 * In case of mixed groups both data and meta should be picked,
4032 * and identical options should be given for both of them.
4034 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4035 if (mixed && (bctl->flags & allowed)) {
4036 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4037 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4038 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4040 "balance: mixed groups data and metadata options must be the same");
4047 * rw_devices will not change at the moment, device add/delete/replace
4048 * are excluded by EXCL_OP
4050 num_devices = fs_info->fs_devices->rw_devices;
4053 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4054 * special bit for it, to make it easier to distinguish. Thus we need
4055 * to set it manually, or balance would refuse the profile.
4057 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4058 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4059 if (num_devices >= btrfs_raid_array[i].devs_min)
4060 allowed |= btrfs_raid_array[i].bg_flag;
4062 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4063 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4064 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4070 * Allow to reduce metadata or system integrity only if force set for
4071 * profiles with redundancy (copies, parity)
4074 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4075 if (btrfs_raid_array[i].ncopies >= 2 ||
4076 btrfs_raid_array[i].tolerated_failures >= 1)
4077 allowed |= btrfs_raid_array[i].bg_flag;
4080 seq = read_seqbegin(&fs_info->profiles_lock);
4082 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4083 (fs_info->avail_system_alloc_bits & allowed) &&
4084 !(bctl->sys.target & allowed)) ||
4085 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4086 (fs_info->avail_metadata_alloc_bits & allowed) &&
4087 !(bctl->meta.target & allowed)))
4088 reducing_redundancy = true;
4090 reducing_redundancy = false;
4092 /* if we're not converting, the target field is uninitialized */
4093 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4094 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4095 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4096 bctl->data.target : fs_info->avail_data_alloc_bits;
4097 } while (read_seqretry(&fs_info->profiles_lock, seq));
4099 if (reducing_redundancy) {
4100 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4102 "balance: force reducing metadata redundancy");
4105 "balance: reduces metadata redundancy, use --force if you want this");
4111 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4112 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4114 "balance: metadata profile %s has lower redundancy than data profile %s",
4115 btrfs_bg_type_to_raid_name(meta_target),
4116 btrfs_bg_type_to_raid_name(data_target));
4119 if (fs_info->send_in_progress) {
4120 btrfs_warn_rl(fs_info,
4121 "cannot run balance while send operations are in progress (%d in progress)",
4122 fs_info->send_in_progress);
4127 ret = insert_balance_item(fs_info, bctl);
4128 if (ret && ret != -EEXIST)
4131 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4132 BUG_ON(ret == -EEXIST);
4133 BUG_ON(fs_info->balance_ctl);
4134 spin_lock(&fs_info->balance_lock);
4135 fs_info->balance_ctl = bctl;
4136 spin_unlock(&fs_info->balance_lock);
4138 BUG_ON(ret != -EEXIST);
4139 spin_lock(&fs_info->balance_lock);
4140 update_balance_args(bctl);
4141 spin_unlock(&fs_info->balance_lock);
4144 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4145 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4146 describe_balance_start_or_resume(fs_info);
4147 mutex_unlock(&fs_info->balance_mutex);
4149 ret = __btrfs_balance(fs_info);
4151 mutex_lock(&fs_info->balance_mutex);
4152 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req))
4153 btrfs_info(fs_info, "balance: paused");
4155 * Balance can be canceled by:
4157 * - Regular cancel request
4158 * Then ret == -ECANCELED and balance_cancel_req > 0
4160 * - Fatal signal to "btrfs" process
4161 * Either the signal caught by wait_reserve_ticket() and callers
4162 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4164 * Either way, in this case balance_cancel_req = 0, and
4165 * ret == -EINTR or ret == -ECANCELED.
4167 * So here we only check the return value to catch canceled balance.
4169 else if (ret == -ECANCELED || ret == -EINTR)
4170 btrfs_info(fs_info, "balance: canceled");
4172 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4174 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4177 memset(bargs, 0, sizeof(*bargs));
4178 btrfs_update_ioctl_balance_args(fs_info, bargs);
4181 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4182 balance_need_close(fs_info)) {
4183 reset_balance_state(fs_info);
4184 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4187 wake_up(&fs_info->balance_wait_q);
4191 if (bctl->flags & BTRFS_BALANCE_RESUME)
4192 reset_balance_state(fs_info);
4195 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4200 static int balance_kthread(void *data)
4202 struct btrfs_fs_info *fs_info = data;
4205 mutex_lock(&fs_info->balance_mutex);
4206 if (fs_info->balance_ctl)
4207 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4208 mutex_unlock(&fs_info->balance_mutex);
4213 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4215 struct task_struct *tsk;
4217 mutex_lock(&fs_info->balance_mutex);
4218 if (!fs_info->balance_ctl) {
4219 mutex_unlock(&fs_info->balance_mutex);
4222 mutex_unlock(&fs_info->balance_mutex);
4224 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4225 btrfs_info(fs_info, "balance: resume skipped");
4230 * A ro->rw remount sequence should continue with the paused balance
4231 * regardless of who pauses it, system or the user as of now, so set
4234 spin_lock(&fs_info->balance_lock);
4235 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4236 spin_unlock(&fs_info->balance_lock);
4238 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4239 return PTR_ERR_OR_ZERO(tsk);
4242 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4244 struct btrfs_balance_control *bctl;
4245 struct btrfs_balance_item *item;
4246 struct btrfs_disk_balance_args disk_bargs;
4247 struct btrfs_path *path;
4248 struct extent_buffer *leaf;
4249 struct btrfs_key key;
4252 path = btrfs_alloc_path();
4256 key.objectid = BTRFS_BALANCE_OBJECTID;
4257 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4260 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4263 if (ret > 0) { /* ret = -ENOENT; */
4268 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4274 leaf = path->nodes[0];
4275 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4277 bctl->flags = btrfs_balance_flags(leaf, item);
4278 bctl->flags |= BTRFS_BALANCE_RESUME;
4280 btrfs_balance_data(leaf, item, &disk_bargs);
4281 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4282 btrfs_balance_meta(leaf, item, &disk_bargs);
4283 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4284 btrfs_balance_sys(leaf, item, &disk_bargs);
4285 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4288 * This should never happen, as the paused balance state is recovered
4289 * during mount without any chance of other exclusive ops to collide.
4291 * This gives the exclusive op status to balance and keeps in paused
4292 * state until user intervention (cancel or umount). If the ownership
4293 * cannot be assigned, show a message but do not fail. The balance
4294 * is in a paused state and must have fs_info::balance_ctl properly
4297 if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags))
4299 "balance: cannot set exclusive op status, resume manually");
4301 mutex_lock(&fs_info->balance_mutex);
4302 BUG_ON(fs_info->balance_ctl);
4303 spin_lock(&fs_info->balance_lock);
4304 fs_info->balance_ctl = bctl;
4305 spin_unlock(&fs_info->balance_lock);
4306 mutex_unlock(&fs_info->balance_mutex);
4308 btrfs_free_path(path);
4312 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4316 mutex_lock(&fs_info->balance_mutex);
4317 if (!fs_info->balance_ctl) {
4318 mutex_unlock(&fs_info->balance_mutex);
4322 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4323 atomic_inc(&fs_info->balance_pause_req);
4324 mutex_unlock(&fs_info->balance_mutex);
4326 wait_event(fs_info->balance_wait_q,
4327 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4329 mutex_lock(&fs_info->balance_mutex);
4330 /* we are good with balance_ctl ripped off from under us */
4331 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4332 atomic_dec(&fs_info->balance_pause_req);
4337 mutex_unlock(&fs_info->balance_mutex);
4341 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4343 mutex_lock(&fs_info->balance_mutex);
4344 if (!fs_info->balance_ctl) {
4345 mutex_unlock(&fs_info->balance_mutex);
4350 * A paused balance with the item stored on disk can be resumed at
4351 * mount time if the mount is read-write. Otherwise it's still paused
4352 * and we must not allow cancelling as it deletes the item.
4354 if (sb_rdonly(fs_info->sb)) {
4355 mutex_unlock(&fs_info->balance_mutex);
4359 atomic_inc(&fs_info->balance_cancel_req);
4361 * if we are running just wait and return, balance item is
4362 * deleted in btrfs_balance in this case
4364 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4365 mutex_unlock(&fs_info->balance_mutex);
4366 wait_event(fs_info->balance_wait_q,
4367 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4368 mutex_lock(&fs_info->balance_mutex);
4370 mutex_unlock(&fs_info->balance_mutex);
4372 * Lock released to allow other waiters to continue, we'll
4373 * reexamine the status again.
4375 mutex_lock(&fs_info->balance_mutex);
4377 if (fs_info->balance_ctl) {
4378 reset_balance_state(fs_info);
4379 clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
4380 btrfs_info(fs_info, "balance: canceled");
4384 BUG_ON(fs_info->balance_ctl ||
4385 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4386 atomic_dec(&fs_info->balance_cancel_req);
4387 mutex_unlock(&fs_info->balance_mutex);
4391 int btrfs_uuid_scan_kthread(void *data)
4393 struct btrfs_fs_info *fs_info = data;
4394 struct btrfs_root *root = fs_info->tree_root;
4395 struct btrfs_key key;
4396 struct btrfs_path *path = NULL;
4398 struct extent_buffer *eb;
4400 struct btrfs_root_item root_item;
4402 struct btrfs_trans_handle *trans = NULL;
4403 bool closing = false;
4405 path = btrfs_alloc_path();
4412 key.type = BTRFS_ROOT_ITEM_KEY;
4416 if (btrfs_fs_closing(fs_info)) {
4420 ret = btrfs_search_forward(root, &key, path,
4421 BTRFS_OLDEST_GENERATION);
4428 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4429 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4430 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4431 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4434 eb = path->nodes[0];
4435 slot = path->slots[0];
4436 item_size = btrfs_item_size_nr(eb, slot);
4437 if (item_size < sizeof(root_item))
4440 read_extent_buffer(eb, &root_item,
4441 btrfs_item_ptr_offset(eb, slot),
4442 (int)sizeof(root_item));
4443 if (btrfs_root_refs(&root_item) == 0)
4446 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4447 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4451 btrfs_release_path(path);
4453 * 1 - subvol uuid item
4454 * 1 - received_subvol uuid item
4456 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4457 if (IS_ERR(trans)) {
4458 ret = PTR_ERR(trans);
4466 btrfs_release_path(path);
4467 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4468 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4469 BTRFS_UUID_KEY_SUBVOL,
4472 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4478 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4479 ret = btrfs_uuid_tree_add(trans,
4480 root_item.received_uuid,
4481 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4484 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4491 btrfs_release_path(path);
4493 ret = btrfs_end_transaction(trans);
4499 if (key.offset < (u64)-1) {
4501 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4503 key.type = BTRFS_ROOT_ITEM_KEY;
4504 } else if (key.objectid < (u64)-1) {
4506 key.type = BTRFS_ROOT_ITEM_KEY;
4515 btrfs_free_path(path);
4516 if (trans && !IS_ERR(trans))
4517 btrfs_end_transaction(trans);
4519 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4521 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4522 up(&fs_info->uuid_tree_rescan_sem);
4526 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4528 struct btrfs_trans_handle *trans;
4529 struct btrfs_root *tree_root = fs_info->tree_root;
4530 struct btrfs_root *uuid_root;
4531 struct task_struct *task;
4538 trans = btrfs_start_transaction(tree_root, 2);
4540 return PTR_ERR(trans);
4542 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4543 if (IS_ERR(uuid_root)) {
4544 ret = PTR_ERR(uuid_root);
4545 btrfs_abort_transaction(trans, ret);
4546 btrfs_end_transaction(trans);
4550 fs_info->uuid_root = uuid_root;
4552 ret = btrfs_commit_transaction(trans);
4556 down(&fs_info->uuid_tree_rescan_sem);
4557 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4559 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4560 btrfs_warn(fs_info, "failed to start uuid_scan task");
4561 up(&fs_info->uuid_tree_rescan_sem);
4562 return PTR_ERR(task);
4569 * shrinking a device means finding all of the device extents past
4570 * the new size, and then following the back refs to the chunks.
4571 * The chunk relocation code actually frees the device extent
4573 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4575 struct btrfs_fs_info *fs_info = device->fs_info;
4576 struct btrfs_root *root = fs_info->dev_root;
4577 struct btrfs_trans_handle *trans;
4578 struct btrfs_dev_extent *dev_extent = NULL;
4579 struct btrfs_path *path;
4585 bool retried = false;
4586 struct extent_buffer *l;
4587 struct btrfs_key key;
4588 struct btrfs_super_block *super_copy = fs_info->super_copy;
4589 u64 old_total = btrfs_super_total_bytes(super_copy);
4590 u64 old_size = btrfs_device_get_total_bytes(device);
4594 new_size = round_down(new_size, fs_info->sectorsize);
4596 diff = round_down(old_size - new_size, fs_info->sectorsize);
4598 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4601 path = btrfs_alloc_path();
4605 path->reada = READA_BACK;
4607 trans = btrfs_start_transaction(root, 0);
4608 if (IS_ERR(trans)) {
4609 btrfs_free_path(path);
4610 return PTR_ERR(trans);
4613 mutex_lock(&fs_info->chunk_mutex);
4615 btrfs_device_set_total_bytes(device, new_size);
4616 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4617 device->fs_devices->total_rw_bytes -= diff;
4618 atomic64_sub(diff, &fs_info->free_chunk_space);
4622 * Once the device's size has been set to the new size, ensure all
4623 * in-memory chunks are synced to disk so that the loop below sees them
4624 * and relocates them accordingly.
4626 if (contains_pending_extent(device, &start, diff)) {
4627 mutex_unlock(&fs_info->chunk_mutex);
4628 ret = btrfs_commit_transaction(trans);
4632 mutex_unlock(&fs_info->chunk_mutex);
4633 btrfs_end_transaction(trans);
4637 key.objectid = device->devid;
4638 key.offset = (u64)-1;
4639 key.type = BTRFS_DEV_EXTENT_KEY;
4642 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4643 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4645 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4649 ret = btrfs_previous_item(root, path, 0, key.type);
4651 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4656 btrfs_release_path(path);
4661 slot = path->slots[0];
4662 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4664 if (key.objectid != device->devid) {
4665 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4666 btrfs_release_path(path);
4670 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4671 length = btrfs_dev_extent_length(l, dev_extent);
4673 if (key.offset + length <= new_size) {
4674 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4675 btrfs_release_path(path);
4679 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4680 btrfs_release_path(path);
4683 * We may be relocating the only data chunk we have,
4684 * which could potentially end up with losing data's
4685 * raid profile, so lets allocate an empty one in
4688 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4690 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4694 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4695 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4696 if (ret == -ENOSPC) {
4699 if (ret == -ETXTBSY) {
4701 "could not shrink block group %llu due to active swapfile",
4706 } while (key.offset-- > 0);
4708 if (failed && !retried) {
4712 } else if (failed && retried) {
4717 /* Shrinking succeeded, else we would be at "done". */
4718 trans = btrfs_start_transaction(root, 0);
4719 if (IS_ERR(trans)) {
4720 ret = PTR_ERR(trans);
4724 mutex_lock(&fs_info->chunk_mutex);
4725 /* Clear all state bits beyond the shrunk device size */
4726 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
4729 btrfs_device_set_disk_total_bytes(device, new_size);
4730 if (list_empty(&device->post_commit_list))
4731 list_add_tail(&device->post_commit_list,
4732 &trans->transaction->dev_update_list);
4734 WARN_ON(diff > old_total);
4735 btrfs_set_super_total_bytes(super_copy,
4736 round_down(old_total - diff, fs_info->sectorsize));
4737 mutex_unlock(&fs_info->chunk_mutex);
4739 /* Now btrfs_update_device() will change the on-disk size. */
4740 ret = btrfs_update_device(trans, device);
4742 btrfs_abort_transaction(trans, ret);
4743 btrfs_end_transaction(trans);
4745 ret = btrfs_commit_transaction(trans);
4748 btrfs_free_path(path);
4750 mutex_lock(&fs_info->chunk_mutex);
4751 btrfs_device_set_total_bytes(device, old_size);
4752 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4753 device->fs_devices->total_rw_bytes += diff;
4754 atomic64_add(diff, &fs_info->free_chunk_space);
4755 mutex_unlock(&fs_info->chunk_mutex);
4760 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4761 struct btrfs_key *key,
4762 struct btrfs_chunk *chunk, int item_size)
4764 struct btrfs_super_block *super_copy = fs_info->super_copy;
4765 struct btrfs_disk_key disk_key;
4769 mutex_lock(&fs_info->chunk_mutex);
4770 array_size = btrfs_super_sys_array_size(super_copy);
4771 if (array_size + item_size + sizeof(disk_key)
4772 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4773 mutex_unlock(&fs_info->chunk_mutex);
4777 ptr = super_copy->sys_chunk_array + array_size;
4778 btrfs_cpu_key_to_disk(&disk_key, key);
4779 memcpy(ptr, &disk_key, sizeof(disk_key));
4780 ptr += sizeof(disk_key);
4781 memcpy(ptr, chunk, item_size);
4782 item_size += sizeof(disk_key);
4783 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4784 mutex_unlock(&fs_info->chunk_mutex);
4790 * sort the devices in descending order by max_avail, total_avail
4792 static int btrfs_cmp_device_info(const void *a, const void *b)
4794 const struct btrfs_device_info *di_a = a;
4795 const struct btrfs_device_info *di_b = b;
4797 if (di_a->max_avail > di_b->max_avail)
4799 if (di_a->max_avail < di_b->max_avail)
4801 if (di_a->total_avail > di_b->total_avail)
4803 if (di_a->total_avail < di_b->total_avail)
4808 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4810 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4813 btrfs_set_fs_incompat(info, RAID56);
4816 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
4818 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
4821 btrfs_set_fs_incompat(info, RAID1C34);
4825 * Structure used internally for __btrfs_alloc_chunk() function.
4826 * Wraps needed parameters.
4828 struct alloc_chunk_ctl {
4831 /* Total number of stripes to allocate */
4833 /* sub_stripes info for map */
4835 /* Stripes per device */
4837 /* Maximum number of devices to use */
4839 /* Minimum number of devices to use */
4841 /* ndevs has to be a multiple of this */
4843 /* Number of copies */
4845 /* Number of stripes worth of bytes to store parity information */
4847 u64 max_stripe_size;
4855 static void init_alloc_chunk_ctl_policy_regular(
4856 struct btrfs_fs_devices *fs_devices,
4857 struct alloc_chunk_ctl *ctl)
4859 u64 type = ctl->type;
4861 if (type & BTRFS_BLOCK_GROUP_DATA) {
4862 ctl->max_stripe_size = SZ_1G;
4863 ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
4864 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4865 /* For larger filesystems, use larger metadata chunks */
4866 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4867 ctl->max_stripe_size = SZ_1G;
4869 ctl->max_stripe_size = SZ_256M;
4870 ctl->max_chunk_size = ctl->max_stripe_size;
4871 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4872 ctl->max_stripe_size = SZ_32M;
4873 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
4874 ctl->devs_max = min_t(int, ctl->devs_max,
4875 BTRFS_MAX_DEVS_SYS_CHUNK);
4880 /* We don't want a chunk larger than 10% of writable space */
4881 ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4882 ctl->max_chunk_size);
4883 ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes;
4886 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
4887 struct alloc_chunk_ctl *ctl)
4889 int index = btrfs_bg_flags_to_raid_index(ctl->type);
4891 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
4892 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
4893 ctl->devs_max = btrfs_raid_array[index].devs_max;
4895 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
4896 ctl->devs_min = btrfs_raid_array[index].devs_min;
4897 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
4898 ctl->ncopies = btrfs_raid_array[index].ncopies;
4899 ctl->nparity = btrfs_raid_array[index].nparity;
4902 switch (fs_devices->chunk_alloc_policy) {
4903 case BTRFS_CHUNK_ALLOC_REGULAR:
4904 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
4911 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
4912 struct alloc_chunk_ctl *ctl,
4913 struct btrfs_device_info *devices_info)
4915 struct btrfs_fs_info *info = fs_devices->fs_info;
4916 struct btrfs_device *device;
4918 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
4925 * in the first pass through the devices list, we gather information
4926 * about the available holes on each device.
4928 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4929 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4931 "BTRFS: read-only device in alloc_list\n");
4935 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
4936 &device->dev_state) ||
4937 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4940 if (device->total_bytes > device->bytes_used)
4941 total_avail = device->total_bytes - device->bytes_used;
4945 /* If there is no space on this device, skip it. */
4946 if (total_avail < ctl->dev_extent_min)
4949 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
4951 if (ret && ret != -ENOSPC)
4955 max_avail = dev_extent_want;
4957 if (max_avail < ctl->dev_extent_min) {
4958 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4960 "%s: devid %llu has no free space, have=%llu want=%llu",
4961 __func__, device->devid, max_avail,
4962 ctl->dev_extent_min);
4966 if (ndevs == fs_devices->rw_devices) {
4967 WARN(1, "%s: found more than %llu devices\n",
4968 __func__, fs_devices->rw_devices);
4971 devices_info[ndevs].dev_offset = dev_offset;
4972 devices_info[ndevs].max_avail = max_avail;
4973 devices_info[ndevs].total_avail = total_avail;
4974 devices_info[ndevs].dev = device;
4980 * now sort the devices by hole size / available space
4982 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4983 btrfs_cmp_device_info, NULL);
4988 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
4989 struct btrfs_device_info *devices_info)
4991 /* Number of stripes that count for block group size */
4995 * The primary goal is to maximize the number of stripes, so use as
4996 * many devices as possible, even if the stripes are not maximum sized.
4998 * The DUP profile stores more than one stripe per device, the
4999 * max_avail is the total size so we have to adjust.
5001 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5003 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5005 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5006 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5009 * Use the number of data stripes to figure out how big this chunk is
5010 * really going to be in terms of logical address space, and compare
5011 * that answer with the max chunk size. If it's higher, we try to
5012 * reduce stripe_size.
5014 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5016 * Reduce stripe_size, round it up to a 16MB boundary again and
5017 * then use it, unless it ends up being even bigger than the
5018 * previous value we had already.
5020 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5021 data_stripes), SZ_16M),
5025 /* Align to BTRFS_STRIPE_LEN */
5026 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5027 ctl->chunk_size = ctl->stripe_size * data_stripes;
5032 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5033 struct alloc_chunk_ctl *ctl,
5034 struct btrfs_device_info *devices_info)
5036 struct btrfs_fs_info *info = fs_devices->fs_info;
5039 * Round down to number of usable stripes, devs_increment can be any
5040 * number so we can't use round_down() that requires power of 2, while
5041 * rounddown is safe.
5043 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5045 if (ctl->ndevs < ctl->devs_min) {
5046 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5048 "%s: not enough devices with free space: have=%d minimum required=%d",
5049 __func__, ctl->ndevs, ctl->devs_min);
5054 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5056 switch (fs_devices->chunk_alloc_policy) {
5057 case BTRFS_CHUNK_ALLOC_REGULAR:
5058 return decide_stripe_size_regular(ctl, devices_info);
5064 static int create_chunk(struct btrfs_trans_handle *trans,
5065 struct alloc_chunk_ctl *ctl,
5066 struct btrfs_device_info *devices_info)
5068 struct btrfs_fs_info *info = trans->fs_info;
5069 struct map_lookup *map = NULL;
5070 struct extent_map_tree *em_tree;
5071 struct extent_map *em;
5072 u64 start = ctl->start;
5073 u64 type = ctl->type;
5078 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
5081 map->num_stripes = ctl->num_stripes;
5083 for (i = 0; i < ctl->ndevs; ++i) {
5084 for (j = 0; j < ctl->dev_stripes; ++j) {
5085 int s = i * ctl->dev_stripes + j;
5086 map->stripes[s].dev = devices_info[i].dev;
5087 map->stripes[s].physical = devices_info[i].dev_offset +
5088 j * ctl->stripe_size;
5091 map->stripe_len = BTRFS_STRIPE_LEN;
5092 map->io_align = BTRFS_STRIPE_LEN;
5093 map->io_width = BTRFS_STRIPE_LEN;
5095 map->sub_stripes = ctl->sub_stripes;
5097 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5099 em = alloc_extent_map();
5104 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5105 em->map_lookup = map;
5107 em->len = ctl->chunk_size;
5108 em->block_start = 0;
5109 em->block_len = em->len;
5110 em->orig_block_len = ctl->stripe_size;
5112 em_tree = &info->mapping_tree;
5113 write_lock(&em_tree->lock);
5114 ret = add_extent_mapping(em_tree, em, 0);
5116 write_unlock(&em_tree->lock);
5117 free_extent_map(em);
5120 write_unlock(&em_tree->lock);
5122 ret = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size);
5124 goto error_del_extent;
5126 for (i = 0; i < map->num_stripes; i++) {
5127 struct btrfs_device *dev = map->stripes[i].dev;
5129 btrfs_device_set_bytes_used(dev,
5130 dev->bytes_used + ctl->stripe_size);
5131 if (list_empty(&dev->post_commit_list))
5132 list_add_tail(&dev->post_commit_list,
5133 &trans->transaction->dev_update_list);
5136 atomic64_sub(ctl->stripe_size * map->num_stripes,
5137 &info->free_chunk_space);
5139 free_extent_map(em);
5140 check_raid56_incompat_flag(info, type);
5141 check_raid1c34_incompat_flag(info, type);
5146 write_lock(&em_tree->lock);
5147 remove_extent_mapping(em_tree, em);
5148 write_unlock(&em_tree->lock);
5150 /* One for our allocation */
5151 free_extent_map(em);
5152 /* One for the tree reference */
5153 free_extent_map(em);
5158 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type)
5160 struct btrfs_fs_info *info = trans->fs_info;
5161 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5162 struct btrfs_device_info *devices_info = NULL;
5163 struct alloc_chunk_ctl ctl;
5166 lockdep_assert_held(&info->chunk_mutex);
5168 if (!alloc_profile_is_valid(type, 0)) {
5173 if (list_empty(&fs_devices->alloc_list)) {
5174 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5175 btrfs_debug(info, "%s: no writable device", __func__);
5179 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5180 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5185 ctl.start = find_next_chunk(info);
5187 init_alloc_chunk_ctl(fs_devices, &ctl);
5189 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5194 ret = gather_device_info(fs_devices, &ctl, devices_info);
5198 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5202 ret = create_chunk(trans, &ctl, devices_info);
5205 kfree(devices_info);
5210 * Chunk allocation falls into two parts. The first part does work
5211 * that makes the new allocated chunk usable, but does not do any operation
5212 * that modifies the chunk tree. The second part does the work that
5213 * requires modifying the chunk tree. This division is important for the
5214 * bootstrap process of adding storage to a seed btrfs.
5216 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
5217 u64 chunk_offset, u64 chunk_size)
5219 struct btrfs_fs_info *fs_info = trans->fs_info;
5220 struct btrfs_root *extent_root = fs_info->extent_root;
5221 struct btrfs_root *chunk_root = fs_info->chunk_root;
5222 struct btrfs_key key;
5223 struct btrfs_device *device;
5224 struct btrfs_chunk *chunk;
5225 struct btrfs_stripe *stripe;
5226 struct extent_map *em;
5227 struct map_lookup *map;
5234 em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size);
5238 map = em->map_lookup;
5239 item_size = btrfs_chunk_item_size(map->num_stripes);
5240 stripe_size = em->orig_block_len;
5242 chunk = kzalloc(item_size, GFP_NOFS);
5249 * Take the device list mutex to prevent races with the final phase of
5250 * a device replace operation that replaces the device object associated
5251 * with the map's stripes, because the device object's id can change
5252 * at any time during that final phase of the device replace operation
5253 * (dev-replace.c:btrfs_dev_replace_finishing()).
5255 mutex_lock(&fs_info->fs_devices->device_list_mutex);
5256 for (i = 0; i < map->num_stripes; i++) {
5257 device = map->stripes[i].dev;
5258 dev_offset = map->stripes[i].physical;
5260 ret = btrfs_update_device(trans, device);
5263 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
5264 dev_offset, stripe_size);
5269 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5273 stripe = &chunk->stripe;
5274 for (i = 0; i < map->num_stripes; i++) {
5275 device = map->stripes[i].dev;
5276 dev_offset = map->stripes[i].physical;
5278 btrfs_set_stack_stripe_devid(stripe, device->devid);
5279 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5280 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5283 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5285 btrfs_set_stack_chunk_length(chunk, chunk_size);
5286 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5287 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5288 btrfs_set_stack_chunk_type(chunk, map->type);
5289 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5290 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5291 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5292 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5293 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5295 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5296 key.type = BTRFS_CHUNK_ITEM_KEY;
5297 key.offset = chunk_offset;
5299 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5300 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5302 * TODO: Cleanup of inserted chunk root in case of
5305 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5310 free_extent_map(em);
5314 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5316 struct btrfs_fs_info *fs_info = trans->fs_info;
5320 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5321 ret = btrfs_alloc_chunk(trans, alloc_profile);
5325 alloc_profile = btrfs_system_alloc_profile(fs_info);
5326 ret = btrfs_alloc_chunk(trans, alloc_profile);
5330 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5332 const int index = btrfs_bg_flags_to_raid_index(map->type);
5334 return btrfs_raid_array[index].tolerated_failures;
5337 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5339 struct extent_map *em;
5340 struct map_lookup *map;
5345 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5349 map = em->map_lookup;
5350 for (i = 0; i < map->num_stripes; i++) {
5351 if (test_bit(BTRFS_DEV_STATE_MISSING,
5352 &map->stripes[i].dev->dev_state)) {
5356 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5357 &map->stripes[i].dev->dev_state)) {
5364 * If the number of missing devices is larger than max errors,
5365 * we can not write the data into that chunk successfully, so
5368 if (miss_ndevs > btrfs_chunk_max_errors(map))
5371 free_extent_map(em);
5375 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5377 struct extent_map *em;
5380 write_lock(&tree->lock);
5381 em = lookup_extent_mapping(tree, 0, (u64)-1);
5383 remove_extent_mapping(tree, em);
5384 write_unlock(&tree->lock);
5388 free_extent_map(em);
5389 /* once for the tree */
5390 free_extent_map(em);
5394 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5396 struct extent_map *em;
5397 struct map_lookup *map;
5400 em = btrfs_get_chunk_map(fs_info, logical, len);
5403 * We could return errors for these cases, but that could get
5404 * ugly and we'd probably do the same thing which is just not do
5405 * anything else and exit, so return 1 so the callers don't try
5406 * to use other copies.
5410 map = em->map_lookup;
5411 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK))
5412 ret = map->num_stripes;
5413 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5414 ret = map->sub_stripes;
5415 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5417 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5419 * There could be two corrupted data stripes, we need
5420 * to loop retry in order to rebuild the correct data.
5422 * Fail a stripe at a time on every retry except the
5423 * stripe under reconstruction.
5425 ret = map->num_stripes;
5428 free_extent_map(em);
5430 down_read(&fs_info->dev_replace.rwsem);
5431 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5432 fs_info->dev_replace.tgtdev)
5434 up_read(&fs_info->dev_replace.rwsem);
5439 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5442 struct extent_map *em;
5443 struct map_lookup *map;
5444 unsigned long len = fs_info->sectorsize;
5446 em = btrfs_get_chunk_map(fs_info, logical, len);
5448 if (!WARN_ON(IS_ERR(em))) {
5449 map = em->map_lookup;
5450 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5451 len = map->stripe_len * nr_data_stripes(map);
5452 free_extent_map(em);
5457 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5459 struct extent_map *em;
5460 struct map_lookup *map;
5463 em = btrfs_get_chunk_map(fs_info, logical, len);
5465 if(!WARN_ON(IS_ERR(em))) {
5466 map = em->map_lookup;
5467 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5469 free_extent_map(em);
5474 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5475 struct map_lookup *map, int first,
5476 int dev_replace_is_ongoing)
5480 int preferred_mirror;
5482 struct btrfs_device *srcdev;
5485 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5487 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5488 num_stripes = map->sub_stripes;
5490 num_stripes = map->num_stripes;
5492 preferred_mirror = first + current->pid % num_stripes;
5494 if (dev_replace_is_ongoing &&
5495 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5496 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5497 srcdev = fs_info->dev_replace.srcdev;
5502 * try to avoid the drive that is the source drive for a
5503 * dev-replace procedure, only choose it if no other non-missing
5504 * mirror is available
5506 for (tolerance = 0; tolerance < 2; tolerance++) {
5507 if (map->stripes[preferred_mirror].dev->bdev &&
5508 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5509 return preferred_mirror;
5510 for (i = first; i < first + num_stripes; i++) {
5511 if (map->stripes[i].dev->bdev &&
5512 (tolerance || map->stripes[i].dev != srcdev))
5517 /* we couldn't find one that doesn't fail. Just return something
5518 * and the io error handling code will clean up eventually
5520 return preferred_mirror;
5523 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5524 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5531 for (i = 0; i < num_stripes - 1; i++) {
5532 /* Swap if parity is on a smaller index */
5533 if (bbio->raid_map[i] > bbio->raid_map[i + 1]) {
5534 swap(bbio->stripes[i], bbio->stripes[i + 1]);
5535 swap(bbio->raid_map[i], bbio->raid_map[i + 1]);
5542 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5544 struct btrfs_bio *bbio = kzalloc(
5545 /* the size of the btrfs_bio */
5546 sizeof(struct btrfs_bio) +
5547 /* plus the variable array for the stripes */
5548 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5549 /* plus the variable array for the tgt dev */
5550 sizeof(int) * (real_stripes) +
5552 * plus the raid_map, which includes both the tgt dev
5555 sizeof(u64) * (total_stripes),
5556 GFP_NOFS|__GFP_NOFAIL);
5558 atomic_set(&bbio->error, 0);
5559 refcount_set(&bbio->refs, 1);
5561 bbio->tgtdev_map = (int *)(bbio->stripes + total_stripes);
5562 bbio->raid_map = (u64 *)(bbio->tgtdev_map + real_stripes);
5567 void btrfs_get_bbio(struct btrfs_bio *bbio)
5569 WARN_ON(!refcount_read(&bbio->refs));
5570 refcount_inc(&bbio->refs);
5573 void btrfs_put_bbio(struct btrfs_bio *bbio)
5577 if (refcount_dec_and_test(&bbio->refs))
5581 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5583 * Please note that, discard won't be sent to target device of device
5586 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5587 u64 logical, u64 *length_ret,
5588 struct btrfs_bio **bbio_ret)
5590 struct extent_map *em;
5591 struct map_lookup *map;
5592 struct btrfs_bio *bbio;
5593 u64 length = *length_ret;
5597 u64 stripe_end_offset;
5604 u32 sub_stripes = 0;
5605 u64 stripes_per_dev = 0;
5606 u32 remaining_stripes = 0;
5607 u32 last_stripe = 0;
5611 /* discard always return a bbio */
5614 em = btrfs_get_chunk_map(fs_info, logical, length);
5618 map = em->map_lookup;
5619 /* we don't discard raid56 yet */
5620 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5625 offset = logical - em->start;
5626 length = min_t(u64, em->start + em->len - logical, length);
5627 *length_ret = length;
5629 stripe_len = map->stripe_len;
5631 * stripe_nr counts the total number of stripes we have to stride
5632 * to get to this block
5634 stripe_nr = div64_u64(offset, stripe_len);
5636 /* stripe_offset is the offset of this block in its stripe */
5637 stripe_offset = offset - stripe_nr * stripe_len;
5639 stripe_nr_end = round_up(offset + length, map->stripe_len);
5640 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5641 stripe_cnt = stripe_nr_end - stripe_nr;
5642 stripe_end_offset = stripe_nr_end * map->stripe_len -
5645 * after this, stripe_nr is the number of stripes on this
5646 * device we have to walk to find the data, and stripe_index is
5647 * the number of our device in the stripe array
5651 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5652 BTRFS_BLOCK_GROUP_RAID10)) {
5653 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5656 sub_stripes = map->sub_stripes;
5658 factor = map->num_stripes / sub_stripes;
5659 num_stripes = min_t(u64, map->num_stripes,
5660 sub_stripes * stripe_cnt);
5661 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5662 stripe_index *= sub_stripes;
5663 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5664 &remaining_stripes);
5665 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5666 last_stripe *= sub_stripes;
5667 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
5668 BTRFS_BLOCK_GROUP_DUP)) {
5669 num_stripes = map->num_stripes;
5671 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5675 bbio = alloc_btrfs_bio(num_stripes, 0);
5681 for (i = 0; i < num_stripes; i++) {
5682 bbio->stripes[i].physical =
5683 map->stripes[stripe_index].physical +
5684 stripe_offset + stripe_nr * map->stripe_len;
5685 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5687 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5688 BTRFS_BLOCK_GROUP_RAID10)) {
5689 bbio->stripes[i].length = stripes_per_dev *
5692 if (i / sub_stripes < remaining_stripes)
5693 bbio->stripes[i].length +=
5697 * Special for the first stripe and
5700 * |-------|...|-------|
5704 if (i < sub_stripes)
5705 bbio->stripes[i].length -=
5708 if (stripe_index >= last_stripe &&
5709 stripe_index <= (last_stripe +
5711 bbio->stripes[i].length -=
5714 if (i == sub_stripes - 1)
5717 bbio->stripes[i].length = length;
5721 if (stripe_index == map->num_stripes) {
5728 bbio->map_type = map->type;
5729 bbio->num_stripes = num_stripes;
5731 free_extent_map(em);
5736 * In dev-replace case, for repair case (that's the only case where the mirror
5737 * is selected explicitly when calling btrfs_map_block), blocks left of the
5738 * left cursor can also be read from the target drive.
5740 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5742 * For READ, it also needs to be supported using the same mirror number.
5744 * If the requested block is not left of the left cursor, EIO is returned. This
5745 * can happen because btrfs_num_copies() returns one more in the dev-replace
5748 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5749 u64 logical, u64 length,
5750 u64 srcdev_devid, int *mirror_num,
5753 struct btrfs_bio *bbio = NULL;
5755 int index_srcdev = 0;
5757 u64 physical_of_found = 0;
5761 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5762 logical, &length, &bbio, 0, 0);
5764 ASSERT(bbio == NULL);
5768 num_stripes = bbio->num_stripes;
5769 if (*mirror_num > num_stripes) {
5771 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5772 * that means that the requested area is not left of the left
5775 btrfs_put_bbio(bbio);
5780 * process the rest of the function using the mirror_num of the source
5781 * drive. Therefore look it up first. At the end, patch the device
5782 * pointer to the one of the target drive.
5784 for (i = 0; i < num_stripes; i++) {
5785 if (bbio->stripes[i].dev->devid != srcdev_devid)
5789 * In case of DUP, in order to keep it simple, only add the
5790 * mirror with the lowest physical address
5793 physical_of_found <= bbio->stripes[i].physical)
5798 physical_of_found = bbio->stripes[i].physical;
5801 btrfs_put_bbio(bbio);
5807 *mirror_num = index_srcdev + 1;
5808 *physical = physical_of_found;
5812 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5813 struct btrfs_bio **bbio_ret,
5814 struct btrfs_dev_replace *dev_replace,
5815 int *num_stripes_ret, int *max_errors_ret)
5817 struct btrfs_bio *bbio = *bbio_ret;
5818 u64 srcdev_devid = dev_replace->srcdev->devid;
5819 int tgtdev_indexes = 0;
5820 int num_stripes = *num_stripes_ret;
5821 int max_errors = *max_errors_ret;
5824 if (op == BTRFS_MAP_WRITE) {
5825 int index_where_to_add;
5828 * duplicate the write operations while the dev replace
5829 * procedure is running. Since the copying of the old disk to
5830 * the new disk takes place at run time while the filesystem is
5831 * mounted writable, the regular write operations to the old
5832 * disk have to be duplicated to go to the new disk as well.
5834 * Note that device->missing is handled by the caller, and that
5835 * the write to the old disk is already set up in the stripes
5838 index_where_to_add = num_stripes;
5839 for (i = 0; i < num_stripes; i++) {
5840 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5841 /* write to new disk, too */
5842 struct btrfs_bio_stripe *new =
5843 bbio->stripes + index_where_to_add;
5844 struct btrfs_bio_stripe *old =
5847 new->physical = old->physical;
5848 new->length = old->length;
5849 new->dev = dev_replace->tgtdev;
5850 bbio->tgtdev_map[i] = index_where_to_add;
5851 index_where_to_add++;
5856 num_stripes = index_where_to_add;
5857 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5858 int index_srcdev = 0;
5860 u64 physical_of_found = 0;
5863 * During the dev-replace procedure, the target drive can also
5864 * be used to read data in case it is needed to repair a corrupt
5865 * block elsewhere. This is possible if the requested area is
5866 * left of the left cursor. In this area, the target drive is a
5867 * full copy of the source drive.
5869 for (i = 0; i < num_stripes; i++) {
5870 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5872 * In case of DUP, in order to keep it simple,
5873 * only add the mirror with the lowest physical
5877 physical_of_found <=
5878 bbio->stripes[i].physical)
5882 physical_of_found = bbio->stripes[i].physical;
5886 struct btrfs_bio_stripe *tgtdev_stripe =
5887 bbio->stripes + num_stripes;
5889 tgtdev_stripe->physical = physical_of_found;
5890 tgtdev_stripe->length =
5891 bbio->stripes[index_srcdev].length;
5892 tgtdev_stripe->dev = dev_replace->tgtdev;
5893 bbio->tgtdev_map[index_srcdev] = num_stripes;
5900 *num_stripes_ret = num_stripes;
5901 *max_errors_ret = max_errors;
5902 bbio->num_tgtdevs = tgtdev_indexes;
5906 static bool need_full_stripe(enum btrfs_map_op op)
5908 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5912 * btrfs_get_io_geometry - calculates the geomery of a particular (address, len)
5913 * tuple. This information is used to calculate how big a
5914 * particular bio can get before it straddles a stripe.
5916 * @fs_info - the filesystem
5917 * @logical - address that we want to figure out the geometry of
5918 * @len - the length of IO we are going to perform, starting at @logical
5919 * @op - type of operation - write or read
5920 * @io_geom - pointer used to return values
5922 * Returns < 0 in case a chunk for the given logical address cannot be found,
5923 * usually shouldn't happen unless @logical is corrupted, 0 otherwise.
5925 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5926 u64 logical, u64 len, struct btrfs_io_geometry *io_geom)
5928 struct extent_map *em;
5929 struct map_lookup *map;
5934 u64 raid56_full_stripe_start = (u64)-1;
5938 ASSERT(op != BTRFS_MAP_DISCARD);
5940 em = btrfs_get_chunk_map(fs_info, logical, len);
5944 map = em->map_lookup;
5945 /* Offset of this logical address in the chunk */
5946 offset = logical - em->start;
5947 /* Len of a stripe in a chunk */
5948 stripe_len = map->stripe_len;
5949 /* Stripe wher this block falls in */
5950 stripe_nr = div64_u64(offset, stripe_len);
5951 /* Offset of stripe in the chunk */
5952 stripe_offset = stripe_nr * stripe_len;
5953 if (offset < stripe_offset) {
5955 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu",
5956 stripe_offset, offset, em->start, logical, stripe_len);
5961 /* stripe_offset is the offset of this block in its stripe */
5962 stripe_offset = offset - stripe_offset;
5963 data_stripes = nr_data_stripes(map);
5965 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5966 u64 max_len = stripe_len - stripe_offset;
5969 * In case of raid56, we need to know the stripe aligned start
5971 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5972 unsigned long full_stripe_len = stripe_len * data_stripes;
5973 raid56_full_stripe_start = offset;
5976 * Allow a write of a full stripe, but make sure we
5977 * don't allow straddling of stripes
5979 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5981 raid56_full_stripe_start *= full_stripe_len;
5984 * For writes to RAID[56], allow a full stripeset across
5985 * all disks. For other RAID types and for RAID[56]
5986 * reads, just allow a single stripe (on a single disk).
5988 if (op == BTRFS_MAP_WRITE) {
5989 max_len = stripe_len * data_stripes -
5990 (offset - raid56_full_stripe_start);
5993 len = min_t(u64, em->len - offset, max_len);
5995 len = em->len - offset;
5999 io_geom->offset = offset;
6000 io_geom->stripe_len = stripe_len;
6001 io_geom->stripe_nr = stripe_nr;
6002 io_geom->stripe_offset = stripe_offset;
6003 io_geom->raid56_stripe_offset = raid56_full_stripe_start;
6007 free_extent_map(em);
6011 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
6012 enum btrfs_map_op op,
6013 u64 logical, u64 *length,
6014 struct btrfs_bio **bbio_ret,
6015 int mirror_num, int need_raid_map)
6017 struct extent_map *em;
6018 struct map_lookup *map;
6028 int tgtdev_indexes = 0;
6029 struct btrfs_bio *bbio = NULL;
6030 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6031 int dev_replace_is_ongoing = 0;
6032 int num_alloc_stripes;
6033 int patch_the_first_stripe_for_dev_replace = 0;
6034 u64 physical_to_patch_in_first_stripe = 0;
6035 u64 raid56_full_stripe_start = (u64)-1;
6036 struct btrfs_io_geometry geom;
6039 ASSERT(op != BTRFS_MAP_DISCARD);
6041 ret = btrfs_get_io_geometry(fs_info, op, logical, *length, &geom);
6045 em = btrfs_get_chunk_map(fs_info, logical, *length);
6046 ASSERT(!IS_ERR(em));
6047 map = em->map_lookup;
6050 stripe_len = geom.stripe_len;
6051 stripe_nr = geom.stripe_nr;
6052 stripe_offset = geom.stripe_offset;
6053 raid56_full_stripe_start = geom.raid56_stripe_offset;
6054 data_stripes = nr_data_stripes(map);
6056 down_read(&dev_replace->rwsem);
6057 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6059 * Hold the semaphore for read during the whole operation, write is
6060 * requested at commit time but must wait.
6062 if (!dev_replace_is_ongoing)
6063 up_read(&dev_replace->rwsem);
6065 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
6066 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
6067 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
6068 dev_replace->srcdev->devid,
6070 &physical_to_patch_in_first_stripe);
6074 patch_the_first_stripe_for_dev_replace = 1;
6075 } else if (mirror_num > map->num_stripes) {
6081 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6082 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6084 if (!need_full_stripe(op))
6086 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6087 if (need_full_stripe(op))
6088 num_stripes = map->num_stripes;
6089 else if (mirror_num)
6090 stripe_index = mirror_num - 1;
6092 stripe_index = find_live_mirror(fs_info, map, 0,
6093 dev_replace_is_ongoing);
6094 mirror_num = stripe_index + 1;
6097 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6098 if (need_full_stripe(op)) {
6099 num_stripes = map->num_stripes;
6100 } else if (mirror_num) {
6101 stripe_index = mirror_num - 1;
6106 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6107 u32 factor = map->num_stripes / map->sub_stripes;
6109 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6110 stripe_index *= map->sub_stripes;
6112 if (need_full_stripe(op))
6113 num_stripes = map->sub_stripes;
6114 else if (mirror_num)
6115 stripe_index += mirror_num - 1;
6117 int old_stripe_index = stripe_index;
6118 stripe_index = find_live_mirror(fs_info, map,
6120 dev_replace_is_ongoing);
6121 mirror_num = stripe_index - old_stripe_index + 1;
6124 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6125 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6126 /* push stripe_nr back to the start of the full stripe */
6127 stripe_nr = div64_u64(raid56_full_stripe_start,
6128 stripe_len * data_stripes);
6130 /* RAID[56] write or recovery. Return all stripes */
6131 num_stripes = map->num_stripes;
6132 max_errors = nr_parity_stripes(map);
6134 *length = map->stripe_len;
6139 * Mirror #0 or #1 means the original data block.
6140 * Mirror #2 is RAID5 parity block.
6141 * Mirror #3 is RAID6 Q block.
6143 stripe_nr = div_u64_rem(stripe_nr,
6144 data_stripes, &stripe_index);
6146 stripe_index = data_stripes + mirror_num - 2;
6148 /* We distribute the parity blocks across stripes */
6149 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
6151 if (!need_full_stripe(op) && mirror_num <= 1)
6156 * after this, stripe_nr is the number of stripes on this
6157 * device we have to walk to find the data, and stripe_index is
6158 * the number of our device in the stripe array
6160 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6162 mirror_num = stripe_index + 1;
6164 if (stripe_index >= map->num_stripes) {
6166 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6167 stripe_index, map->num_stripes);
6172 num_alloc_stripes = num_stripes;
6173 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
6174 if (op == BTRFS_MAP_WRITE)
6175 num_alloc_stripes <<= 1;
6176 if (op == BTRFS_MAP_GET_READ_MIRRORS)
6177 num_alloc_stripes++;
6178 tgtdev_indexes = num_stripes;
6181 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
6187 for (i = 0; i < num_stripes; i++) {
6188 bbio->stripes[i].physical = map->stripes[stripe_index].physical +
6189 stripe_offset + stripe_nr * map->stripe_len;
6190 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
6194 /* build raid_map */
6195 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6196 (need_full_stripe(op) || mirror_num > 1)) {
6200 /* Work out the disk rotation on this stripe-set */
6201 div_u64_rem(stripe_nr, num_stripes, &rot);
6203 /* Fill in the logical address of each stripe */
6204 tmp = stripe_nr * data_stripes;
6205 for (i = 0; i < data_stripes; i++)
6206 bbio->raid_map[(i+rot) % num_stripes] =
6207 em->start + (tmp + i) * map->stripe_len;
6209 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
6210 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
6211 bbio->raid_map[(i+rot+1) % num_stripes] =
6214 sort_parity_stripes(bbio, num_stripes);
6217 if (need_full_stripe(op))
6218 max_errors = btrfs_chunk_max_errors(map);
6220 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6221 need_full_stripe(op)) {
6222 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
6227 bbio->map_type = map->type;
6228 bbio->num_stripes = num_stripes;
6229 bbio->max_errors = max_errors;
6230 bbio->mirror_num = mirror_num;
6233 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6234 * mirror_num == num_stripes + 1 && dev_replace target drive is
6235 * available as a mirror
6237 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6238 WARN_ON(num_stripes > 1);
6239 bbio->stripes[0].dev = dev_replace->tgtdev;
6240 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
6241 bbio->mirror_num = map->num_stripes + 1;
6244 if (dev_replace_is_ongoing) {
6245 lockdep_assert_held(&dev_replace->rwsem);
6246 /* Unlock and let waiting writers proceed */
6247 up_read(&dev_replace->rwsem);
6249 free_extent_map(em);
6253 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6254 u64 logical, u64 *length,
6255 struct btrfs_bio **bbio_ret, int mirror_num)
6257 if (op == BTRFS_MAP_DISCARD)
6258 return __btrfs_map_block_for_discard(fs_info, logical,
6261 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
6265 /* For Scrub/replace */
6266 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6267 u64 logical, u64 *length,
6268 struct btrfs_bio **bbio_ret)
6270 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
6273 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6275 bio->bi_private = bbio->private;
6276 bio->bi_end_io = bbio->end_io;
6279 btrfs_put_bbio(bbio);
6282 static void btrfs_end_bio(struct bio *bio)
6284 struct btrfs_bio *bbio = bio->bi_private;
6285 int is_orig_bio = 0;
6287 if (bio->bi_status) {
6288 atomic_inc(&bbio->error);
6289 if (bio->bi_status == BLK_STS_IOERR ||
6290 bio->bi_status == BLK_STS_TARGET) {
6291 struct btrfs_device *dev = btrfs_io_bio(bio)->device;
6294 if (bio_op(bio) == REQ_OP_WRITE)
6295 btrfs_dev_stat_inc_and_print(dev,
6296 BTRFS_DEV_STAT_WRITE_ERRS);
6297 else if (!(bio->bi_opf & REQ_RAHEAD))
6298 btrfs_dev_stat_inc_and_print(dev,
6299 BTRFS_DEV_STAT_READ_ERRS);
6300 if (bio->bi_opf & REQ_PREFLUSH)
6301 btrfs_dev_stat_inc_and_print(dev,
6302 BTRFS_DEV_STAT_FLUSH_ERRS);
6306 if (bio == bbio->orig_bio)
6309 btrfs_bio_counter_dec(bbio->fs_info);
6311 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6314 bio = bbio->orig_bio;
6317 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6318 /* only send an error to the higher layers if it is
6319 * beyond the tolerance of the btrfs bio
6321 if (atomic_read(&bbio->error) > bbio->max_errors) {
6322 bio->bi_status = BLK_STS_IOERR;
6325 * this bio is actually up to date, we didn't
6326 * go over the max number of errors
6328 bio->bi_status = BLK_STS_OK;
6331 btrfs_end_bbio(bbio, bio);
6332 } else if (!is_orig_bio) {
6337 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6338 u64 physical, struct btrfs_device *dev)
6340 struct btrfs_fs_info *fs_info = bbio->fs_info;
6342 bio->bi_private = bbio;
6343 btrfs_io_bio(bio)->device = dev;
6344 bio->bi_end_io = btrfs_end_bio;
6345 bio->bi_iter.bi_sector = physical >> 9;
6346 btrfs_debug_in_rcu(fs_info,
6347 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6348 bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector,
6349 (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name),
6350 dev->devid, bio->bi_iter.bi_size);
6351 bio_set_dev(bio, dev->bdev);
6353 btrfs_bio_counter_inc_noblocked(fs_info);
6355 btrfsic_submit_bio(bio);
6358 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6360 atomic_inc(&bbio->error);
6361 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6362 /* Should be the original bio. */
6363 WARN_ON(bio != bbio->orig_bio);
6365 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6366 bio->bi_iter.bi_sector = logical >> 9;
6367 if (atomic_read(&bbio->error) > bbio->max_errors)
6368 bio->bi_status = BLK_STS_IOERR;
6370 bio->bi_status = BLK_STS_OK;
6371 btrfs_end_bbio(bbio, bio);
6375 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6378 struct btrfs_device *dev;
6379 struct bio *first_bio = bio;
6380 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6386 struct btrfs_bio *bbio = NULL;
6388 length = bio->bi_iter.bi_size;
6389 map_length = length;
6391 btrfs_bio_counter_inc_blocked(fs_info);
6392 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6393 &map_length, &bbio, mirror_num, 1);
6395 btrfs_bio_counter_dec(fs_info);
6396 return errno_to_blk_status(ret);
6399 total_devs = bbio->num_stripes;
6400 bbio->orig_bio = first_bio;
6401 bbio->private = first_bio->bi_private;
6402 bbio->end_io = first_bio->bi_end_io;
6403 bbio->fs_info = fs_info;
6404 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6406 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6407 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6408 /* In this case, map_length has been set to the length of
6409 a single stripe; not the whole write */
6410 if (bio_op(bio) == REQ_OP_WRITE) {
6411 ret = raid56_parity_write(fs_info, bio, bbio,
6414 ret = raid56_parity_recover(fs_info, bio, bbio,
6415 map_length, mirror_num, 1);
6418 btrfs_bio_counter_dec(fs_info);
6419 return errno_to_blk_status(ret);
6422 if (map_length < length) {
6424 "mapping failed logical %llu bio len %llu len %llu",
6425 logical, length, map_length);
6429 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6430 dev = bbio->stripes[dev_nr].dev;
6431 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
6433 (bio_op(first_bio) == REQ_OP_WRITE &&
6434 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6435 bbio_error(bbio, first_bio, logical);
6439 if (dev_nr < total_devs - 1)
6440 bio = btrfs_bio_clone(first_bio);
6444 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, dev);
6446 btrfs_bio_counter_dec(fs_info);
6451 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6454 * If devid and uuid are both specified, the match must be exact, otherwise
6455 * only devid is used.
6457 * If @seed is true, traverse through the seed devices.
6459 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
6460 u64 devid, u8 *uuid, u8 *fsid,
6463 struct btrfs_device *device;
6465 while (fs_devices) {
6467 !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6468 list_for_each_entry(device, &fs_devices->devices,
6470 if (device->devid == devid &&
6471 (!uuid || memcmp(device->uuid, uuid,
6472 BTRFS_UUID_SIZE) == 0))
6477 fs_devices = fs_devices->seed;
6484 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6485 u64 devid, u8 *dev_uuid)
6487 struct btrfs_device *device;
6488 unsigned int nofs_flag;
6491 * We call this under the chunk_mutex, so we want to use NOFS for this
6492 * allocation, however we don't want to change btrfs_alloc_device() to
6493 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6496 nofs_flag = memalloc_nofs_save();
6497 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6498 memalloc_nofs_restore(nofs_flag);
6502 list_add(&device->dev_list, &fs_devices->devices);
6503 device->fs_devices = fs_devices;
6504 fs_devices->num_devices++;
6506 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6507 fs_devices->missing_devices++;
6513 * btrfs_alloc_device - allocate struct btrfs_device
6514 * @fs_info: used only for generating a new devid, can be NULL if
6515 * devid is provided (i.e. @devid != NULL).
6516 * @devid: a pointer to devid for this device. If NULL a new devid
6518 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6521 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6522 * on error. Returned struct is not linked onto any lists and must be
6523 * destroyed with btrfs_free_device.
6525 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6529 struct btrfs_device *dev;
6532 if (WARN_ON(!devid && !fs_info))
6533 return ERR_PTR(-EINVAL);
6535 dev = __alloc_device();
6544 ret = find_next_devid(fs_info, &tmp);
6546 btrfs_free_device(dev);
6547 return ERR_PTR(ret);
6553 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6555 generate_random_uuid(dev->uuid);
6560 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6561 u64 devid, u8 *uuid, bool error)
6564 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6567 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6571 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes)
6573 int index = btrfs_bg_flags_to_raid_index(type);
6574 int ncopies = btrfs_raid_array[index].ncopies;
6575 const int nparity = btrfs_raid_array[index].nparity;
6579 data_stripes = num_stripes - nparity;
6581 data_stripes = num_stripes / ncopies;
6583 return div_u64(chunk_len, data_stripes);
6586 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6587 struct btrfs_chunk *chunk)
6589 struct btrfs_fs_info *fs_info = leaf->fs_info;
6590 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6591 struct map_lookup *map;
6592 struct extent_map *em;
6596 u8 uuid[BTRFS_UUID_SIZE];
6601 logical = key->offset;
6602 length = btrfs_chunk_length(leaf, chunk);
6603 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6606 * Only need to verify chunk item if we're reading from sys chunk array,
6607 * as chunk item in tree block is already verified by tree-checker.
6609 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6610 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6615 read_lock(&map_tree->lock);
6616 em = lookup_extent_mapping(map_tree, logical, 1);
6617 read_unlock(&map_tree->lock);
6619 /* already mapped? */
6620 if (em && em->start <= logical && em->start + em->len > logical) {
6621 free_extent_map(em);
6624 free_extent_map(em);
6627 em = alloc_extent_map();
6630 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6632 free_extent_map(em);
6636 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6637 em->map_lookup = map;
6638 em->start = logical;
6641 em->block_start = 0;
6642 em->block_len = em->len;
6644 map->num_stripes = num_stripes;
6645 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6646 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6647 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6648 map->type = btrfs_chunk_type(leaf, chunk);
6649 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6650 map->verified_stripes = 0;
6651 em->orig_block_len = calc_stripe_length(map->type, em->len,
6653 for (i = 0; i < num_stripes; i++) {
6654 map->stripes[i].physical =
6655 btrfs_stripe_offset_nr(leaf, chunk, i);
6656 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6657 read_extent_buffer(leaf, uuid, (unsigned long)
6658 btrfs_stripe_dev_uuid_nr(chunk, i),
6660 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices,
6661 devid, uuid, NULL, true);
6662 if (!map->stripes[i].dev &&
6663 !btrfs_test_opt(fs_info, DEGRADED)) {
6664 free_extent_map(em);
6665 btrfs_report_missing_device(fs_info, devid, uuid, true);
6668 if (!map->stripes[i].dev) {
6669 map->stripes[i].dev =
6670 add_missing_dev(fs_info->fs_devices, devid,
6672 if (IS_ERR(map->stripes[i].dev)) {
6673 free_extent_map(em);
6675 "failed to init missing dev %llu: %ld",
6676 devid, PTR_ERR(map->stripes[i].dev));
6677 return PTR_ERR(map->stripes[i].dev);
6679 btrfs_report_missing_device(fs_info, devid, uuid, false);
6681 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6682 &(map->stripes[i].dev->dev_state));
6686 write_lock(&map_tree->lock);
6687 ret = add_extent_mapping(map_tree, em, 0);
6688 write_unlock(&map_tree->lock);
6691 "failed to add chunk map, start=%llu len=%llu: %d",
6692 em->start, em->len, ret);
6694 free_extent_map(em);
6699 static void fill_device_from_item(struct extent_buffer *leaf,
6700 struct btrfs_dev_item *dev_item,
6701 struct btrfs_device *device)
6705 device->devid = btrfs_device_id(leaf, dev_item);
6706 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6707 device->total_bytes = device->disk_total_bytes;
6708 device->commit_total_bytes = device->disk_total_bytes;
6709 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6710 device->commit_bytes_used = device->bytes_used;
6711 device->type = btrfs_device_type(leaf, dev_item);
6712 device->io_align = btrfs_device_io_align(leaf, dev_item);
6713 device->io_width = btrfs_device_io_width(leaf, dev_item);
6714 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6715 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6716 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6718 ptr = btrfs_device_uuid(dev_item);
6719 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6722 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6725 struct btrfs_fs_devices *fs_devices;
6728 lockdep_assert_held(&uuid_mutex);
6731 fs_devices = fs_info->fs_devices->seed;
6732 while (fs_devices) {
6733 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6736 fs_devices = fs_devices->seed;
6739 fs_devices = find_fsid(fsid, NULL);
6741 if (!btrfs_test_opt(fs_info, DEGRADED))
6742 return ERR_PTR(-ENOENT);
6744 fs_devices = alloc_fs_devices(fsid, NULL);
6745 if (IS_ERR(fs_devices))
6748 fs_devices->seeding = true;
6749 fs_devices->opened = 1;
6753 fs_devices = clone_fs_devices(fs_devices);
6754 if (IS_ERR(fs_devices))
6757 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
6759 free_fs_devices(fs_devices);
6760 fs_devices = ERR_PTR(ret);
6764 if (!fs_devices->seeding) {
6765 close_fs_devices(fs_devices);
6766 free_fs_devices(fs_devices);
6767 fs_devices = ERR_PTR(-EINVAL);
6771 fs_devices->seed = fs_info->fs_devices->seed;
6772 fs_info->fs_devices->seed = fs_devices;
6777 static int read_one_dev(struct extent_buffer *leaf,
6778 struct btrfs_dev_item *dev_item)
6780 struct btrfs_fs_info *fs_info = leaf->fs_info;
6781 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6782 struct btrfs_device *device;
6785 u8 fs_uuid[BTRFS_FSID_SIZE];
6786 u8 dev_uuid[BTRFS_UUID_SIZE];
6788 devid = btrfs_device_id(leaf, dev_item);
6789 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6791 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6794 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
6795 fs_devices = open_seed_devices(fs_info, fs_uuid);
6796 if (IS_ERR(fs_devices))
6797 return PTR_ERR(fs_devices);
6800 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
6803 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6804 btrfs_report_missing_device(fs_info, devid,
6809 device = add_missing_dev(fs_devices, devid, dev_uuid);
6810 if (IS_ERR(device)) {
6812 "failed to add missing dev %llu: %ld",
6813 devid, PTR_ERR(device));
6814 return PTR_ERR(device);
6816 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6818 if (!device->bdev) {
6819 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6820 btrfs_report_missing_device(fs_info,
6821 devid, dev_uuid, true);
6824 btrfs_report_missing_device(fs_info, devid,
6828 if (!device->bdev &&
6829 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6831 * this happens when a device that was properly setup
6832 * in the device info lists suddenly goes bad.
6833 * device->bdev is NULL, and so we have to set
6834 * device->missing to one here
6836 device->fs_devices->missing_devices++;
6837 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6840 /* Move the device to its own fs_devices */
6841 if (device->fs_devices != fs_devices) {
6842 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6843 &device->dev_state));
6845 list_move(&device->dev_list, &fs_devices->devices);
6846 device->fs_devices->num_devices--;
6847 fs_devices->num_devices++;
6849 device->fs_devices->missing_devices--;
6850 fs_devices->missing_devices++;
6852 device->fs_devices = fs_devices;
6856 if (device->fs_devices != fs_info->fs_devices) {
6857 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6858 if (device->generation !=
6859 btrfs_device_generation(leaf, dev_item))
6863 fill_device_from_item(leaf, dev_item, device);
6864 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
6865 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
6866 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
6867 device->fs_devices->total_rw_bytes += device->total_bytes;
6868 atomic64_add(device->total_bytes - device->bytes_used,
6869 &fs_info->free_chunk_space);
6875 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6877 struct btrfs_root *root = fs_info->tree_root;
6878 struct btrfs_super_block *super_copy = fs_info->super_copy;
6879 struct extent_buffer *sb;
6880 struct btrfs_disk_key *disk_key;
6881 struct btrfs_chunk *chunk;
6883 unsigned long sb_array_offset;
6890 struct btrfs_key key;
6892 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6894 * This will create extent buffer of nodesize, superblock size is
6895 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6896 * overallocate but we can keep it as-is, only the first page is used.
6898 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6901 set_extent_buffer_uptodate(sb);
6902 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6904 * The sb extent buffer is artificial and just used to read the system array.
6905 * set_extent_buffer_uptodate() call does not properly mark all it's
6906 * pages up-to-date when the page is larger: extent does not cover the
6907 * whole page and consequently check_page_uptodate does not find all
6908 * the page's extents up-to-date (the hole beyond sb),
6909 * write_extent_buffer then triggers a WARN_ON.
6911 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6912 * but sb spans only this function. Add an explicit SetPageUptodate call
6913 * to silence the warning eg. on PowerPC 64.
6915 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6916 SetPageUptodate(sb->pages[0]);
6918 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6919 array_size = btrfs_super_sys_array_size(super_copy);
6921 array_ptr = super_copy->sys_chunk_array;
6922 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6925 while (cur_offset < array_size) {
6926 disk_key = (struct btrfs_disk_key *)array_ptr;
6927 len = sizeof(*disk_key);
6928 if (cur_offset + len > array_size)
6929 goto out_short_read;
6931 btrfs_disk_key_to_cpu(&key, disk_key);
6934 sb_array_offset += len;
6937 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
6939 "unexpected item type %u in sys_array at offset %u",
6940 (u32)key.type, cur_offset);
6945 chunk = (struct btrfs_chunk *)sb_array_offset;
6947 * At least one btrfs_chunk with one stripe must be present,
6948 * exact stripe count check comes afterwards
6950 len = btrfs_chunk_item_size(1);
6951 if (cur_offset + len > array_size)
6952 goto out_short_read;
6954 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6957 "invalid number of stripes %u in sys_array at offset %u",
6958 num_stripes, cur_offset);
6963 type = btrfs_chunk_type(sb, chunk);
6964 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6966 "invalid chunk type %llu in sys_array at offset %u",
6972 len = btrfs_chunk_item_size(num_stripes);
6973 if (cur_offset + len > array_size)
6974 goto out_short_read;
6976 ret = read_one_chunk(&key, sb, chunk);
6981 sb_array_offset += len;
6984 clear_extent_buffer_uptodate(sb);
6985 free_extent_buffer_stale(sb);
6989 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6991 clear_extent_buffer_uptodate(sb);
6992 free_extent_buffer_stale(sb);
6997 * Check if all chunks in the fs are OK for read-write degraded mount
6999 * If the @failing_dev is specified, it's accounted as missing.
7001 * Return true if all chunks meet the minimal RW mount requirements.
7002 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7004 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7005 struct btrfs_device *failing_dev)
7007 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7008 struct extent_map *em;
7012 read_lock(&map_tree->lock);
7013 em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7014 read_unlock(&map_tree->lock);
7015 /* No chunk at all? Return false anyway */
7021 struct map_lookup *map;
7026 map = em->map_lookup;
7028 btrfs_get_num_tolerated_disk_barrier_failures(
7030 for (i = 0; i < map->num_stripes; i++) {
7031 struct btrfs_device *dev = map->stripes[i].dev;
7033 if (!dev || !dev->bdev ||
7034 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7035 dev->last_flush_error)
7037 else if (failing_dev && failing_dev == dev)
7040 if (missing > max_tolerated) {
7043 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7044 em->start, missing, max_tolerated);
7045 free_extent_map(em);
7049 next_start = extent_map_end(em);
7050 free_extent_map(em);
7052 read_lock(&map_tree->lock);
7053 em = lookup_extent_mapping(map_tree, next_start,
7054 (u64)(-1) - next_start);
7055 read_unlock(&map_tree->lock);
7061 static void readahead_tree_node_children(struct extent_buffer *node)
7064 const int nr_items = btrfs_header_nritems(node);
7066 for (i = 0; i < nr_items; i++) {
7069 start = btrfs_node_blockptr(node, i);
7070 readahead_tree_block(node->fs_info, start);
7074 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7076 struct btrfs_root *root = fs_info->chunk_root;
7077 struct btrfs_path *path;
7078 struct extent_buffer *leaf;
7079 struct btrfs_key key;
7080 struct btrfs_key found_key;
7084 u64 last_ra_node = 0;
7086 path = btrfs_alloc_path();
7091 * uuid_mutex is needed only if we are mounting a sprout FS
7092 * otherwise we don't need it.
7094 mutex_lock(&uuid_mutex);
7097 * It is possible for mount and umount to race in such a way that
7098 * we execute this code path, but open_fs_devices failed to clear
7099 * total_rw_bytes. We certainly want it cleared before reading the
7100 * device items, so clear it here.
7102 fs_info->fs_devices->total_rw_bytes = 0;
7105 * Read all device items, and then all the chunk items. All
7106 * device items are found before any chunk item (their object id
7107 * is smaller than the lowest possible object id for a chunk
7108 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7110 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7113 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7117 struct extent_buffer *node;
7119 leaf = path->nodes[0];
7120 slot = path->slots[0];
7121 if (slot >= btrfs_header_nritems(leaf)) {
7122 ret = btrfs_next_leaf(root, path);
7130 * The nodes on level 1 are not locked but we don't need to do
7131 * that during mount time as nothing else can access the tree
7133 node = path->nodes[1];
7135 if (last_ra_node != node->start) {
7136 readahead_tree_node_children(node);
7137 last_ra_node = node->start;
7140 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7141 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7142 struct btrfs_dev_item *dev_item;
7143 dev_item = btrfs_item_ptr(leaf, slot,
7144 struct btrfs_dev_item);
7145 ret = read_one_dev(leaf, dev_item);
7149 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7150 struct btrfs_chunk *chunk;
7151 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7152 mutex_lock(&fs_info->chunk_mutex);
7153 ret = read_one_chunk(&found_key, leaf, chunk);
7154 mutex_unlock(&fs_info->chunk_mutex);
7162 * After loading chunk tree, we've got all device information,
7163 * do another round of validation checks.
7165 if (total_dev != fs_info->fs_devices->total_devices) {
7167 "super_num_devices %llu mismatch with num_devices %llu found here",
7168 btrfs_super_num_devices(fs_info->super_copy),
7173 if (btrfs_super_total_bytes(fs_info->super_copy) <
7174 fs_info->fs_devices->total_rw_bytes) {
7176 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7177 btrfs_super_total_bytes(fs_info->super_copy),
7178 fs_info->fs_devices->total_rw_bytes);
7184 mutex_unlock(&uuid_mutex);
7186 btrfs_free_path(path);
7190 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7192 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7193 struct btrfs_device *device;
7195 while (fs_devices) {
7196 mutex_lock(&fs_devices->device_list_mutex);
7197 list_for_each_entry(device, &fs_devices->devices, dev_list)
7198 device->fs_info = fs_info;
7199 mutex_unlock(&fs_devices->device_list_mutex);
7201 fs_devices = fs_devices->seed;
7205 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7206 const struct btrfs_dev_stats_item *ptr,
7211 read_extent_buffer(eb, &val,
7212 offsetof(struct btrfs_dev_stats_item, values) +
7213 ((unsigned long)ptr) + (index * sizeof(u64)),
7218 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7219 struct btrfs_dev_stats_item *ptr,
7222 write_extent_buffer(eb, &val,
7223 offsetof(struct btrfs_dev_stats_item, values) +
7224 ((unsigned long)ptr) + (index * sizeof(u64)),
7228 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7230 struct btrfs_key key;
7231 struct btrfs_root *dev_root = fs_info->dev_root;
7232 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7233 struct extent_buffer *eb;
7236 struct btrfs_device *device;
7237 struct btrfs_path *path = NULL;
7240 path = btrfs_alloc_path();
7244 mutex_lock(&fs_devices->device_list_mutex);
7245 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7247 struct btrfs_dev_stats_item *ptr;
7249 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7250 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7251 key.offset = device->devid;
7252 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
7254 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7255 btrfs_dev_stat_set(device, i, 0);
7256 device->dev_stats_valid = 1;
7257 btrfs_release_path(path);
7260 slot = path->slots[0];
7261 eb = path->nodes[0];
7262 item_size = btrfs_item_size_nr(eb, slot);
7264 ptr = btrfs_item_ptr(eb, slot,
7265 struct btrfs_dev_stats_item);
7267 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7268 if (item_size >= (1 + i) * sizeof(__le64))
7269 btrfs_dev_stat_set(device, i,
7270 btrfs_dev_stats_value(eb, ptr, i));
7272 btrfs_dev_stat_set(device, i, 0);
7275 device->dev_stats_valid = 1;
7276 btrfs_dev_stat_print_on_load(device);
7277 btrfs_release_path(path);
7279 mutex_unlock(&fs_devices->device_list_mutex);
7281 btrfs_free_path(path);
7282 return ret < 0 ? ret : 0;
7285 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7286 struct btrfs_device *device)
7288 struct btrfs_fs_info *fs_info = trans->fs_info;
7289 struct btrfs_root *dev_root = fs_info->dev_root;
7290 struct btrfs_path *path;
7291 struct btrfs_key key;
7292 struct extent_buffer *eb;
7293 struct btrfs_dev_stats_item *ptr;
7297 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7298 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7299 key.offset = device->devid;
7301 path = btrfs_alloc_path();
7304 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7306 btrfs_warn_in_rcu(fs_info,
7307 "error %d while searching for dev_stats item for device %s",
7308 ret, rcu_str_deref(device->name));
7313 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7314 /* need to delete old one and insert a new one */
7315 ret = btrfs_del_item(trans, dev_root, path);
7317 btrfs_warn_in_rcu(fs_info,
7318 "delete too small dev_stats item for device %s failed %d",
7319 rcu_str_deref(device->name), ret);
7326 /* need to insert a new item */
7327 btrfs_release_path(path);
7328 ret = btrfs_insert_empty_item(trans, dev_root, path,
7329 &key, sizeof(*ptr));
7331 btrfs_warn_in_rcu(fs_info,
7332 "insert dev_stats item for device %s failed %d",
7333 rcu_str_deref(device->name), ret);
7338 eb = path->nodes[0];
7339 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7340 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7341 btrfs_set_dev_stats_value(eb, ptr, i,
7342 btrfs_dev_stat_read(device, i));
7343 btrfs_mark_buffer_dirty(eb);
7346 btrfs_free_path(path);
7351 * called from commit_transaction. Writes all changed device stats to disk.
7353 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7355 struct btrfs_fs_info *fs_info = trans->fs_info;
7356 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7357 struct btrfs_device *device;
7361 mutex_lock(&fs_devices->device_list_mutex);
7362 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7363 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7364 if (!device->dev_stats_valid || stats_cnt == 0)
7369 * There is a LOAD-LOAD control dependency between the value of
7370 * dev_stats_ccnt and updating the on-disk values which requires
7371 * reading the in-memory counters. Such control dependencies
7372 * require explicit read memory barriers.
7374 * This memory barriers pairs with smp_mb__before_atomic in
7375 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7376 * barrier implied by atomic_xchg in
7377 * btrfs_dev_stats_read_and_reset
7381 ret = update_dev_stat_item(trans, device);
7383 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7385 mutex_unlock(&fs_devices->device_list_mutex);
7390 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7392 btrfs_dev_stat_inc(dev, index);
7393 btrfs_dev_stat_print_on_error(dev);
7396 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7398 if (!dev->dev_stats_valid)
7400 btrfs_err_rl_in_rcu(dev->fs_info,
7401 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7402 rcu_str_deref(dev->name),
7403 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7404 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7405 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7406 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7407 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7410 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7414 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7415 if (btrfs_dev_stat_read(dev, i) != 0)
7417 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7418 return; /* all values == 0, suppress message */
7420 btrfs_info_in_rcu(dev->fs_info,
7421 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7422 rcu_str_deref(dev->name),
7423 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7424 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7425 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7426 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7427 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7430 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7431 struct btrfs_ioctl_get_dev_stats *stats)
7433 struct btrfs_device *dev;
7434 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7437 mutex_lock(&fs_devices->device_list_mutex);
7438 dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL,
7440 mutex_unlock(&fs_devices->device_list_mutex);
7443 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7445 } else if (!dev->dev_stats_valid) {
7446 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7448 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7449 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7450 if (stats->nr_items > i)
7452 btrfs_dev_stat_read_and_reset(dev, i);
7454 btrfs_dev_stat_set(dev, i, 0);
7456 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7457 current->comm, task_pid_nr(current));
7459 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7460 if (stats->nr_items > i)
7461 stats->values[i] = btrfs_dev_stat_read(dev, i);
7463 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7464 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7469 * Update the size and bytes used for each device where it changed. This is
7470 * delayed since we would otherwise get errors while writing out the
7473 * Must be invoked during transaction commit.
7475 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7477 struct btrfs_device *curr, *next;
7479 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7481 if (list_empty(&trans->dev_update_list))
7485 * We don't need the device_list_mutex here. This list is owned by the
7486 * transaction and the transaction must complete before the device is
7489 mutex_lock(&trans->fs_info->chunk_mutex);
7490 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7492 list_del_init(&curr->post_commit_list);
7493 curr->commit_total_bytes = curr->disk_total_bytes;
7494 curr->commit_bytes_used = curr->bytes_used;
7496 mutex_unlock(&trans->fs_info->chunk_mutex);
7499 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
7501 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7502 while (fs_devices) {
7503 fs_devices->fs_info = fs_info;
7504 fs_devices = fs_devices->seed;
7508 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
7510 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7511 while (fs_devices) {
7512 fs_devices->fs_info = NULL;
7513 fs_devices = fs_devices->seed;
7518 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7520 int btrfs_bg_type_to_factor(u64 flags)
7522 const int index = btrfs_bg_flags_to_raid_index(flags);
7524 return btrfs_raid_array[index].ncopies;
7529 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7530 u64 chunk_offset, u64 devid,
7531 u64 physical_offset, u64 physical_len)
7533 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7534 struct extent_map *em;
7535 struct map_lookup *map;
7536 struct btrfs_device *dev;
7542 read_lock(&em_tree->lock);
7543 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7544 read_unlock(&em_tree->lock);
7548 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7549 physical_offset, devid);
7554 map = em->map_lookup;
7555 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes);
7556 if (physical_len != stripe_len) {
7558 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7559 physical_offset, devid, em->start, physical_len,
7565 for (i = 0; i < map->num_stripes; i++) {
7566 if (map->stripes[i].dev->devid == devid &&
7567 map->stripes[i].physical == physical_offset) {
7569 if (map->verified_stripes >= map->num_stripes) {
7571 "too many dev extents for chunk %llu found",
7576 map->verified_stripes++;
7582 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7583 physical_offset, devid);
7587 /* Make sure no dev extent is beyond device bondary */
7588 dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true);
7590 btrfs_err(fs_info, "failed to find devid %llu", devid);
7595 /* It's possible this device is a dummy for seed device */
7596 if (dev->disk_total_bytes == 0) {
7597 dev = btrfs_find_device(fs_info->fs_devices->seed, devid, NULL,
7600 btrfs_err(fs_info, "failed to find seed devid %llu",
7607 if (physical_offset + physical_len > dev->disk_total_bytes) {
7609 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7610 devid, physical_offset, physical_len,
7611 dev->disk_total_bytes);
7616 free_extent_map(em);
7620 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7622 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7623 struct extent_map *em;
7624 struct rb_node *node;
7627 read_lock(&em_tree->lock);
7628 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
7629 em = rb_entry(node, struct extent_map, rb_node);
7630 if (em->map_lookup->num_stripes !=
7631 em->map_lookup->verified_stripes) {
7633 "chunk %llu has missing dev extent, have %d expect %d",
7634 em->start, em->map_lookup->verified_stripes,
7635 em->map_lookup->num_stripes);
7641 read_unlock(&em_tree->lock);
7646 * Ensure that all dev extents are mapped to correct chunk, otherwise
7647 * later chunk allocation/free would cause unexpected behavior.
7649 * NOTE: This will iterate through the whole device tree, which should be of
7650 * the same size level as the chunk tree. This slightly increases mount time.
7652 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
7654 struct btrfs_path *path;
7655 struct btrfs_root *root = fs_info->dev_root;
7656 struct btrfs_key key;
7658 u64 prev_dev_ext_end = 0;
7662 key.type = BTRFS_DEV_EXTENT_KEY;
7665 path = btrfs_alloc_path();
7669 path->reada = READA_FORWARD;
7670 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7674 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7675 ret = btrfs_next_item(root, path);
7678 /* No dev extents at all? Not good */
7685 struct extent_buffer *leaf = path->nodes[0];
7686 struct btrfs_dev_extent *dext;
7687 int slot = path->slots[0];
7689 u64 physical_offset;
7693 btrfs_item_key_to_cpu(leaf, &key, slot);
7694 if (key.type != BTRFS_DEV_EXTENT_KEY)
7696 devid = key.objectid;
7697 physical_offset = key.offset;
7699 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
7700 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
7701 physical_len = btrfs_dev_extent_length(leaf, dext);
7703 /* Check if this dev extent overlaps with the previous one */
7704 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
7706 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
7707 devid, physical_offset, prev_dev_ext_end);
7712 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
7713 physical_offset, physical_len);
7717 prev_dev_ext_end = physical_offset + physical_len;
7719 ret = btrfs_next_item(root, path);
7728 /* Ensure all chunks have corresponding dev extents */
7729 ret = verify_chunk_dev_extent_mapping(fs_info);
7731 btrfs_free_path(path);
7736 * Check whether the given block group or device is pinned by any inode being
7737 * used as a swapfile.
7739 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
7741 struct btrfs_swapfile_pin *sp;
7742 struct rb_node *node;
7744 spin_lock(&fs_info->swapfile_pins_lock);
7745 node = fs_info->swapfile_pins.rb_node;
7747 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
7749 node = node->rb_left;
7750 else if (ptr > sp->ptr)
7751 node = node->rb_right;
7755 spin_unlock(&fs_info->swapfile_pins_lock);
7756 return node != NULL;