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
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 * The status of exclusive operation is set and cleared atomically.
322 * During the course of Paused state, fs_info::exclusive_operation 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 * The exclusive status 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);
359 INIT_LIST_HEAD(&fs_devs->seed_list);
361 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
364 memcpy(fs_devs->metadata_uuid, metadata_fsid, BTRFS_FSID_SIZE);
366 memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE);
371 void btrfs_free_device(struct btrfs_device *device)
373 WARN_ON(!list_empty(&device->post_commit_list));
374 rcu_string_free(device->name);
375 extent_io_tree_release(&device->alloc_state);
376 bio_put(device->flush_bio);
380 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
382 struct btrfs_device *device;
383 WARN_ON(fs_devices->opened);
384 while (!list_empty(&fs_devices->devices)) {
385 device = list_entry(fs_devices->devices.next,
386 struct btrfs_device, dev_list);
387 list_del(&device->dev_list);
388 btrfs_free_device(device);
393 void __exit btrfs_cleanup_fs_uuids(void)
395 struct btrfs_fs_devices *fs_devices;
397 while (!list_empty(&fs_uuids)) {
398 fs_devices = list_entry(fs_uuids.next,
399 struct btrfs_fs_devices, fs_list);
400 list_del(&fs_devices->fs_list);
401 free_fs_devices(fs_devices);
406 * Returns a pointer to a new btrfs_device on success; ERR_PTR() on error.
407 * Returned struct is not linked onto any lists and must be destroyed using
410 static struct btrfs_device *__alloc_device(struct btrfs_fs_info *fs_info)
412 struct btrfs_device *dev;
414 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
416 return ERR_PTR(-ENOMEM);
419 * Preallocate a bio that's always going to be used for flushing device
420 * barriers and matches the device lifespan
422 dev->flush_bio = bio_alloc_bioset(GFP_KERNEL, 0, NULL);
423 if (!dev->flush_bio) {
425 return ERR_PTR(-ENOMEM);
428 INIT_LIST_HEAD(&dev->dev_list);
429 INIT_LIST_HEAD(&dev->dev_alloc_list);
430 INIT_LIST_HEAD(&dev->post_commit_list);
432 atomic_set(&dev->reada_in_flight, 0);
433 atomic_set(&dev->dev_stats_ccnt, 0);
434 btrfs_device_data_ordered_init(dev, fs_info);
435 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
436 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
437 extent_io_tree_init(fs_info, &dev->alloc_state,
438 IO_TREE_DEVICE_ALLOC_STATE, NULL);
443 static noinline struct btrfs_fs_devices *find_fsid(
444 const u8 *fsid, const u8 *metadata_fsid)
446 struct btrfs_fs_devices *fs_devices;
450 /* Handle non-split brain cases */
451 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
453 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0
454 && memcmp(metadata_fsid, fs_devices->metadata_uuid,
455 BTRFS_FSID_SIZE) == 0)
458 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
465 static struct btrfs_fs_devices *find_fsid_with_metadata_uuid(
466 struct btrfs_super_block *disk_super)
469 struct btrfs_fs_devices *fs_devices;
472 * Handle scanned device having completed its fsid change but
473 * belonging to a fs_devices that was created by first scanning
474 * a device which didn't have its fsid/metadata_uuid changed
475 * at all and the CHANGING_FSID_V2 flag set.
477 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
478 if (fs_devices->fsid_change &&
479 memcmp(disk_super->metadata_uuid, fs_devices->fsid,
480 BTRFS_FSID_SIZE) == 0 &&
481 memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
482 BTRFS_FSID_SIZE) == 0) {
487 * Handle scanned device having completed its fsid change but
488 * belonging to a fs_devices that was created by a device that
489 * has an outdated pair of fsid/metadata_uuid and
490 * CHANGING_FSID_V2 flag set.
492 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
493 if (fs_devices->fsid_change &&
494 memcmp(fs_devices->metadata_uuid,
495 fs_devices->fsid, BTRFS_FSID_SIZE) != 0 &&
496 memcmp(disk_super->metadata_uuid, fs_devices->metadata_uuid,
497 BTRFS_FSID_SIZE) == 0) {
502 return find_fsid(disk_super->fsid, disk_super->metadata_uuid);
507 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
508 int flush, struct block_device **bdev,
509 struct btrfs_super_block **disk_super)
513 *bdev = blkdev_get_by_path(device_path, flags, holder);
516 ret = PTR_ERR(*bdev);
521 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
522 ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
524 blkdev_put(*bdev, flags);
527 invalidate_bdev(*bdev);
528 *disk_super = btrfs_read_dev_super(*bdev);
529 if (IS_ERR(*disk_super)) {
530 ret = PTR_ERR(*disk_super);
531 blkdev_put(*bdev, flags);
542 static bool device_path_matched(const char *path, struct btrfs_device *device)
547 found = strcmp(rcu_str_deref(device->name), path);
554 * Search and remove all stale (devices which are not mounted) devices.
555 * When both inputs are NULL, it will search and release all stale devices.
556 * path: Optional. When provided will it release all unmounted devices
557 * matching this path only.
558 * skip_dev: Optional. Will skip this device when searching for the stale
560 * Return: 0 for success or if @path is NULL.
561 * -EBUSY if @path is a mounted device.
562 * -ENOENT if @path does not match any device in the list.
564 static int btrfs_free_stale_devices(const char *path,
565 struct btrfs_device *skip_device)
567 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
568 struct btrfs_device *device, *tmp_device;
574 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
576 mutex_lock(&fs_devices->device_list_mutex);
577 list_for_each_entry_safe(device, tmp_device,
578 &fs_devices->devices, dev_list) {
579 if (skip_device && skip_device == device)
581 if (path && !device->name)
583 if (path && !device_path_matched(path, device))
585 if (fs_devices->opened) {
586 /* for an already deleted device return 0 */
587 if (path && ret != 0)
592 /* delete the stale device */
593 fs_devices->num_devices--;
594 list_del(&device->dev_list);
595 btrfs_free_device(device);
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 %s devid %llu generation %llu scanned by %s (%d)",
945 path, devid, found_transid,
947 task_pid_nr(current));
948 return ERR_PTR(-EEXIST);
951 btrfs_info_in_rcu(device->fs_info,
952 "devid %llu device path %s changed to %s scanned by %s (%d)",
953 devid, rcu_str_deref(device->name),
955 task_pid_nr(current));
958 name = rcu_string_strdup(path, GFP_NOFS);
960 mutex_unlock(&fs_devices->device_list_mutex);
961 return ERR_PTR(-ENOMEM);
963 rcu_string_free(device->name);
964 rcu_assign_pointer(device->name, name);
965 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
966 fs_devices->missing_devices--;
967 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
972 * Unmount does not free the btrfs_device struct but would zero
973 * generation along with most of the other members. So just update
974 * it back. We need it to pick the disk with largest generation
977 if (!fs_devices->opened) {
978 device->generation = found_transid;
979 fs_devices->latest_generation = max_t(u64, found_transid,
980 fs_devices->latest_generation);
983 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
985 mutex_unlock(&fs_devices->device_list_mutex);
989 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
991 struct btrfs_fs_devices *fs_devices;
992 struct btrfs_device *device;
993 struct btrfs_device *orig_dev;
996 fs_devices = alloc_fs_devices(orig->fsid, NULL);
997 if (IS_ERR(fs_devices))
1000 mutex_lock(&orig->device_list_mutex);
1001 fs_devices->total_devices = orig->total_devices;
1003 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
1004 struct rcu_string *name;
1006 device = btrfs_alloc_device(NULL, &orig_dev->devid,
1008 if (IS_ERR(device)) {
1009 ret = PTR_ERR(device);
1014 * This is ok to do without rcu read locked because we hold the
1015 * uuid mutex so nothing we touch in here is going to disappear.
1017 if (orig_dev->name) {
1018 name = rcu_string_strdup(orig_dev->name->str,
1021 btrfs_free_device(device);
1025 rcu_assign_pointer(device->name, name);
1028 list_add(&device->dev_list, &fs_devices->devices);
1029 device->fs_devices = fs_devices;
1030 fs_devices->num_devices++;
1032 mutex_unlock(&orig->device_list_mutex);
1035 mutex_unlock(&orig->device_list_mutex);
1036 free_fs_devices(fs_devices);
1037 return ERR_PTR(ret);
1040 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices,
1041 int step, struct btrfs_device **latest_dev)
1043 struct btrfs_device *device, *next;
1045 /* This is the initialized path, it is safe to release the devices. */
1046 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1047 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) {
1048 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1049 &device->dev_state) &&
1050 !test_bit(BTRFS_DEV_STATE_MISSING,
1051 &device->dev_state) &&
1053 device->generation > (*latest_dev)->generation)) {
1054 *latest_dev = device;
1059 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
1061 * In the first step, keep the device which has
1062 * the correct fsid and the devid that is used
1063 * for the dev_replace procedure.
1064 * In the second step, the dev_replace state is
1065 * read from the device tree and it is known
1066 * whether the procedure is really active or
1067 * not, which means whether this device is
1068 * used or whether it should be removed.
1070 if (step == 0 || test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1071 &device->dev_state)) {
1076 blkdev_put(device->bdev, device->mode);
1077 device->bdev = NULL;
1078 fs_devices->open_devices--;
1080 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1081 list_del_init(&device->dev_alloc_list);
1082 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1083 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1084 &device->dev_state))
1085 fs_devices->rw_devices--;
1087 list_del_init(&device->dev_list);
1088 fs_devices->num_devices--;
1089 btrfs_free_device(device);
1095 * After we have read the system tree and know devids belonging to this
1096 * filesystem, remove the device which does not belong there.
1098 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, int step)
1100 struct btrfs_device *latest_dev = NULL;
1101 struct btrfs_fs_devices *seed_dev;
1103 mutex_lock(&uuid_mutex);
1104 __btrfs_free_extra_devids(fs_devices, step, &latest_dev);
1106 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list)
1107 __btrfs_free_extra_devids(seed_dev, step, &latest_dev);
1109 fs_devices->latest_bdev = latest_dev->bdev;
1111 mutex_unlock(&uuid_mutex);
1114 static void btrfs_close_bdev(struct btrfs_device *device)
1119 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1120 sync_blockdev(device->bdev);
1121 invalidate_bdev(device->bdev);
1124 blkdev_put(device->bdev, device->mode);
1127 static void btrfs_close_one_device(struct btrfs_device *device)
1129 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1131 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1132 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1133 list_del_init(&device->dev_alloc_list);
1134 fs_devices->rw_devices--;
1137 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
1138 fs_devices->missing_devices--;
1140 btrfs_close_bdev(device);
1142 fs_devices->open_devices--;
1143 device->bdev = NULL;
1145 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1147 device->fs_info = NULL;
1148 atomic_set(&device->dev_stats_ccnt, 0);
1149 extent_io_tree_release(&device->alloc_state);
1151 /* Verify the device is back in a pristine state */
1152 ASSERT(!test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1153 ASSERT(!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1154 ASSERT(list_empty(&device->dev_alloc_list));
1155 ASSERT(list_empty(&device->post_commit_list));
1156 ASSERT(atomic_read(&device->reada_in_flight) == 0);
1159 static void close_fs_devices(struct btrfs_fs_devices *fs_devices)
1161 struct btrfs_device *device, *tmp;
1163 lockdep_assert_held(&uuid_mutex);
1165 if (--fs_devices->opened > 0)
1168 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list)
1169 btrfs_close_one_device(device);
1171 WARN_ON(fs_devices->open_devices);
1172 WARN_ON(fs_devices->rw_devices);
1173 fs_devices->opened = 0;
1174 fs_devices->seeding = false;
1175 fs_devices->fs_info = NULL;
1178 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1181 struct btrfs_fs_devices *tmp;
1183 mutex_lock(&uuid_mutex);
1184 close_fs_devices(fs_devices);
1185 if (!fs_devices->opened)
1186 list_splice_init(&fs_devices->seed_list, &list);
1188 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {
1189 close_fs_devices(fs_devices);
1190 list_del(&fs_devices->seed_list);
1191 free_fs_devices(fs_devices);
1193 mutex_unlock(&uuid_mutex);
1196 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1197 fmode_t flags, void *holder)
1199 struct btrfs_device *device;
1200 struct btrfs_device *latest_dev = NULL;
1201 struct btrfs_device *tmp_device;
1203 flags |= FMODE_EXCL;
1205 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,
1209 ret = btrfs_open_one_device(fs_devices, device, flags, holder);
1211 (!latest_dev || device->generation > latest_dev->generation)) {
1212 latest_dev = device;
1213 } else if (ret == -ENODATA) {
1214 fs_devices->num_devices--;
1215 list_del(&device->dev_list);
1216 btrfs_free_device(device);
1219 if (fs_devices->open_devices == 0)
1222 fs_devices->opened = 1;
1223 fs_devices->latest_bdev = latest_dev->bdev;
1224 fs_devices->total_rw_bytes = 0;
1225 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1230 static int devid_cmp(void *priv, struct list_head *a, struct list_head *b)
1232 struct btrfs_device *dev1, *dev2;
1234 dev1 = list_entry(a, struct btrfs_device, dev_list);
1235 dev2 = list_entry(b, struct btrfs_device, dev_list);
1237 if (dev1->devid < dev2->devid)
1239 else if (dev1->devid > dev2->devid)
1244 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1245 fmode_t flags, void *holder)
1249 lockdep_assert_held(&uuid_mutex);
1251 * The device_list_mutex cannot be taken here in case opening the
1252 * underlying device takes further locks like bd_mutex.
1254 * We also don't need the lock here as this is called during mount and
1255 * exclusion is provided by uuid_mutex
1258 if (fs_devices->opened) {
1259 fs_devices->opened++;
1262 list_sort(NULL, &fs_devices->devices, devid_cmp);
1263 ret = open_fs_devices(fs_devices, flags, holder);
1269 void btrfs_release_disk_super(struct btrfs_super_block *super)
1271 struct page *page = virt_to_page(super);
1276 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1279 struct btrfs_super_block *disk_super;
1284 /* make sure our super fits in the device */
1285 if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode))
1286 return ERR_PTR(-EINVAL);
1288 /* make sure our super fits in the page */
1289 if (sizeof(*disk_super) > PAGE_SIZE)
1290 return ERR_PTR(-EINVAL);
1292 /* make sure our super doesn't straddle pages on disk */
1293 index = bytenr >> PAGE_SHIFT;
1294 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1295 return ERR_PTR(-EINVAL);
1297 /* pull in the page with our super */
1298 page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1301 return ERR_CAST(page);
1303 p = page_address(page);
1305 /* align our pointer to the offset of the super block */
1306 disk_super = p + offset_in_page(bytenr);
1308 if (btrfs_super_bytenr(disk_super) != bytenr ||
1309 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1310 btrfs_release_disk_super(p);
1311 return ERR_PTR(-EINVAL);
1314 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1315 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1320 int btrfs_forget_devices(const char *path)
1324 mutex_lock(&uuid_mutex);
1325 ret = btrfs_free_stale_devices(strlen(path) ? path : NULL, NULL);
1326 mutex_unlock(&uuid_mutex);
1332 * Look for a btrfs signature on a device. This may be called out of the mount path
1333 * and we are not allowed to call set_blocksize during the scan. The superblock
1334 * is read via pagecache
1336 struct btrfs_device *btrfs_scan_one_device(const char *path, fmode_t flags,
1339 struct btrfs_super_block *disk_super;
1340 bool new_device_added = false;
1341 struct btrfs_device *device = NULL;
1342 struct block_device *bdev;
1345 lockdep_assert_held(&uuid_mutex);
1348 * we would like to check all the supers, but that would make
1349 * a btrfs mount succeed after a mkfs from a different FS.
1350 * So, we need to add a special mount option to scan for
1351 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1353 bytenr = btrfs_sb_offset(0);
1354 flags |= FMODE_EXCL;
1356 bdev = blkdev_get_by_path(path, flags, holder);
1358 return ERR_CAST(bdev);
1360 disk_super = btrfs_read_disk_super(bdev, bytenr);
1361 if (IS_ERR(disk_super)) {
1362 device = ERR_CAST(disk_super);
1363 goto error_bdev_put;
1366 device = device_list_add(path, disk_super, &new_device_added);
1367 if (!IS_ERR(device)) {
1368 if (new_device_added)
1369 btrfs_free_stale_devices(path, device);
1372 btrfs_release_disk_super(disk_super);
1375 blkdev_put(bdev, flags);
1381 * Try to find a chunk that intersects [start, start + len] range and when one
1382 * such is found, record the end of it in *start
1384 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1387 u64 physical_start, physical_end;
1389 lockdep_assert_held(&device->fs_info->chunk_mutex);
1391 if (!find_first_extent_bit(&device->alloc_state, *start,
1392 &physical_start, &physical_end,
1393 CHUNK_ALLOCATED, NULL)) {
1395 if (in_range(physical_start, *start, len) ||
1396 in_range(*start, physical_start,
1397 physical_end - physical_start)) {
1398 *start = physical_end + 1;
1405 static u64 dev_extent_search_start(struct btrfs_device *device, u64 start)
1407 switch (device->fs_devices->chunk_alloc_policy) {
1408 case BTRFS_CHUNK_ALLOC_REGULAR:
1410 * We don't want to overwrite the superblock on the drive nor
1411 * any area used by the boot loader (grub for example), so we
1412 * make sure to start at an offset of at least 1MB.
1414 return max_t(u64, start, SZ_1M);
1421 * dev_extent_hole_check - check if specified hole is suitable for allocation
1422 * @device: the device which we have the hole
1423 * @hole_start: starting position of the hole
1424 * @hole_size: the size of the hole
1425 * @num_bytes: the size of the free space that we need
1427 * This function may modify @hole_start and @hole_end to reflect the suitable
1428 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1430 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1431 u64 *hole_size, u64 num_bytes)
1433 bool changed = false;
1434 u64 hole_end = *hole_start + *hole_size;
1437 * Check before we set max_hole_start, otherwise we could end up
1438 * sending back this offset anyway.
1440 if (contains_pending_extent(device, hole_start, *hole_size)) {
1441 if (hole_end >= *hole_start)
1442 *hole_size = hole_end - *hole_start;
1448 switch (device->fs_devices->chunk_alloc_policy) {
1449 case BTRFS_CHUNK_ALLOC_REGULAR:
1450 /* No extra check */
1460 * find_free_dev_extent_start - find free space in the specified device
1461 * @device: the device which we search the free space in
1462 * @num_bytes: the size of the free space that we need
1463 * @search_start: the position from which to begin the search
1464 * @start: store the start of the free space.
1465 * @len: the size of the free space. that we find, or the size
1466 * of the max free space if we don't find suitable free space
1468 * this uses a pretty simple search, the expectation is that it is
1469 * called very infrequently and that a given device has a small number
1472 * @start is used to store the start of the free space if we find. But if we
1473 * don't find suitable free space, it will be used to store the start position
1474 * of the max free space.
1476 * @len is used to store the size of the free space that we find.
1477 * But if we don't find suitable free space, it is used to store the size of
1478 * the max free space.
1480 * NOTE: This function will search *commit* root of device tree, and does extra
1481 * check to ensure dev extents are not double allocated.
1482 * This makes the function safe to allocate dev extents but may not report
1483 * correct usable device space, as device extent freed in current transaction
1484 * is not reported as avaiable.
1486 static int find_free_dev_extent_start(struct btrfs_device *device,
1487 u64 num_bytes, u64 search_start, u64 *start,
1490 struct btrfs_fs_info *fs_info = device->fs_info;
1491 struct btrfs_root *root = fs_info->dev_root;
1492 struct btrfs_key key;
1493 struct btrfs_dev_extent *dev_extent;
1494 struct btrfs_path *path;
1499 u64 search_end = device->total_bytes;
1502 struct extent_buffer *l;
1504 search_start = dev_extent_search_start(device, search_start);
1506 path = btrfs_alloc_path();
1510 max_hole_start = search_start;
1514 if (search_start >= search_end ||
1515 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1520 path->reada = READA_FORWARD;
1521 path->search_commit_root = 1;
1522 path->skip_locking = 1;
1524 key.objectid = device->devid;
1525 key.offset = search_start;
1526 key.type = BTRFS_DEV_EXTENT_KEY;
1528 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1532 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1539 slot = path->slots[0];
1540 if (slot >= btrfs_header_nritems(l)) {
1541 ret = btrfs_next_leaf(root, path);
1549 btrfs_item_key_to_cpu(l, &key, slot);
1551 if (key.objectid < device->devid)
1554 if (key.objectid > device->devid)
1557 if (key.type != BTRFS_DEV_EXTENT_KEY)
1560 if (key.offset > search_start) {
1561 hole_size = key.offset - search_start;
1562 dev_extent_hole_check(device, &search_start, &hole_size,
1565 if (hole_size > max_hole_size) {
1566 max_hole_start = search_start;
1567 max_hole_size = hole_size;
1571 * If this free space is greater than which we need,
1572 * it must be the max free space that we have found
1573 * until now, so max_hole_start must point to the start
1574 * of this free space and the length of this free space
1575 * is stored in max_hole_size. Thus, we return
1576 * max_hole_start and max_hole_size and go back to the
1579 if (hole_size >= num_bytes) {
1585 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1586 extent_end = key.offset + btrfs_dev_extent_length(l,
1588 if (extent_end > search_start)
1589 search_start = extent_end;
1596 * At this point, search_start should be the end of
1597 * allocated dev extents, and when shrinking the device,
1598 * search_end may be smaller than search_start.
1600 if (search_end > search_start) {
1601 hole_size = search_end - search_start;
1602 if (dev_extent_hole_check(device, &search_start, &hole_size,
1604 btrfs_release_path(path);
1608 if (hole_size > max_hole_size) {
1609 max_hole_start = search_start;
1610 max_hole_size = hole_size;
1615 if (max_hole_size < num_bytes)
1621 btrfs_free_path(path);
1622 *start = max_hole_start;
1624 *len = max_hole_size;
1628 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1629 u64 *start, u64 *len)
1631 /* FIXME use last free of some kind */
1632 return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1635 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1636 struct btrfs_device *device,
1637 u64 start, u64 *dev_extent_len)
1639 struct btrfs_fs_info *fs_info = device->fs_info;
1640 struct btrfs_root *root = fs_info->dev_root;
1642 struct btrfs_path *path;
1643 struct btrfs_key key;
1644 struct btrfs_key found_key;
1645 struct extent_buffer *leaf = NULL;
1646 struct btrfs_dev_extent *extent = NULL;
1648 path = btrfs_alloc_path();
1652 key.objectid = device->devid;
1654 key.type = BTRFS_DEV_EXTENT_KEY;
1656 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1658 ret = btrfs_previous_item(root, path, key.objectid,
1659 BTRFS_DEV_EXTENT_KEY);
1662 leaf = path->nodes[0];
1663 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1664 extent = btrfs_item_ptr(leaf, path->slots[0],
1665 struct btrfs_dev_extent);
1666 BUG_ON(found_key.offset > start || found_key.offset +
1667 btrfs_dev_extent_length(leaf, extent) < start);
1669 btrfs_release_path(path);
1671 } else if (ret == 0) {
1672 leaf = path->nodes[0];
1673 extent = btrfs_item_ptr(leaf, path->slots[0],
1674 struct btrfs_dev_extent);
1676 btrfs_handle_fs_error(fs_info, ret, "Slot search failed");
1680 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1682 ret = btrfs_del_item(trans, root, path);
1684 btrfs_handle_fs_error(fs_info, ret,
1685 "Failed to remove dev extent item");
1687 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1690 btrfs_free_path(path);
1694 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1695 struct btrfs_device *device,
1696 u64 chunk_offset, u64 start, u64 num_bytes)
1699 struct btrfs_path *path;
1700 struct btrfs_fs_info *fs_info = device->fs_info;
1701 struct btrfs_root *root = fs_info->dev_root;
1702 struct btrfs_dev_extent *extent;
1703 struct extent_buffer *leaf;
1704 struct btrfs_key key;
1706 WARN_ON(!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state));
1707 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1708 path = btrfs_alloc_path();
1712 key.objectid = device->devid;
1714 key.type = BTRFS_DEV_EXTENT_KEY;
1715 ret = btrfs_insert_empty_item(trans, root, path, &key,
1720 leaf = path->nodes[0];
1721 extent = btrfs_item_ptr(leaf, path->slots[0],
1722 struct btrfs_dev_extent);
1723 btrfs_set_dev_extent_chunk_tree(leaf, extent,
1724 BTRFS_CHUNK_TREE_OBJECTID);
1725 btrfs_set_dev_extent_chunk_objectid(leaf, extent,
1726 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1727 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1729 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1730 btrfs_mark_buffer_dirty(leaf);
1732 btrfs_free_path(path);
1736 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1738 struct extent_map_tree *em_tree;
1739 struct extent_map *em;
1743 em_tree = &fs_info->mapping_tree;
1744 read_lock(&em_tree->lock);
1745 n = rb_last(&em_tree->map.rb_root);
1747 em = rb_entry(n, struct extent_map, rb_node);
1748 ret = em->start + em->len;
1750 read_unlock(&em_tree->lock);
1755 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1759 struct btrfs_key key;
1760 struct btrfs_key found_key;
1761 struct btrfs_path *path;
1763 path = btrfs_alloc_path();
1767 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1768 key.type = BTRFS_DEV_ITEM_KEY;
1769 key.offset = (u64)-1;
1771 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1777 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1782 ret = btrfs_previous_item(fs_info->chunk_root, path,
1783 BTRFS_DEV_ITEMS_OBJECTID,
1784 BTRFS_DEV_ITEM_KEY);
1788 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1790 *devid_ret = found_key.offset + 1;
1794 btrfs_free_path(path);
1799 * the device information is stored in the chunk root
1800 * the btrfs_device struct should be fully filled in
1802 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1803 struct btrfs_device *device)
1806 struct btrfs_path *path;
1807 struct btrfs_dev_item *dev_item;
1808 struct extent_buffer *leaf;
1809 struct btrfs_key key;
1812 path = btrfs_alloc_path();
1816 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1817 key.type = BTRFS_DEV_ITEM_KEY;
1818 key.offset = device->devid;
1820 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1821 &key, sizeof(*dev_item));
1825 leaf = path->nodes[0];
1826 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1828 btrfs_set_device_id(leaf, dev_item, device->devid);
1829 btrfs_set_device_generation(leaf, dev_item, 0);
1830 btrfs_set_device_type(leaf, dev_item, device->type);
1831 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1832 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1833 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1834 btrfs_set_device_total_bytes(leaf, dev_item,
1835 btrfs_device_get_disk_total_bytes(device));
1836 btrfs_set_device_bytes_used(leaf, dev_item,
1837 btrfs_device_get_bytes_used(device));
1838 btrfs_set_device_group(leaf, dev_item, 0);
1839 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1840 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1841 btrfs_set_device_start_offset(leaf, dev_item, 0);
1843 ptr = btrfs_device_uuid(dev_item);
1844 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1845 ptr = btrfs_device_fsid(dev_item);
1846 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1847 ptr, BTRFS_FSID_SIZE);
1848 btrfs_mark_buffer_dirty(leaf);
1852 btrfs_free_path(path);
1857 * Function to update ctime/mtime for a given device path.
1858 * Mainly used for ctime/mtime based probe like libblkid.
1860 static void update_dev_time(const char *path_name)
1864 filp = filp_open(path_name, O_RDWR, 0);
1867 file_update_time(filp);
1868 filp_close(filp, NULL);
1871 static int btrfs_rm_dev_item(struct btrfs_device *device)
1873 struct btrfs_root *root = device->fs_info->chunk_root;
1875 struct btrfs_path *path;
1876 struct btrfs_key key;
1877 struct btrfs_trans_handle *trans;
1879 path = btrfs_alloc_path();
1883 trans = btrfs_start_transaction(root, 0);
1884 if (IS_ERR(trans)) {
1885 btrfs_free_path(path);
1886 return PTR_ERR(trans);
1888 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1889 key.type = BTRFS_DEV_ITEM_KEY;
1890 key.offset = device->devid;
1892 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1896 btrfs_abort_transaction(trans, ret);
1897 btrfs_end_transaction(trans);
1901 ret = btrfs_del_item(trans, root, path);
1903 btrfs_abort_transaction(trans, ret);
1904 btrfs_end_transaction(trans);
1908 btrfs_free_path(path);
1910 ret = btrfs_commit_transaction(trans);
1915 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1916 * filesystem. It's up to the caller to adjust that number regarding eg. device
1919 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1927 seq = read_seqbegin(&fs_info->profiles_lock);
1929 all_avail = fs_info->avail_data_alloc_bits |
1930 fs_info->avail_system_alloc_bits |
1931 fs_info->avail_metadata_alloc_bits;
1932 } while (read_seqretry(&fs_info->profiles_lock, seq));
1934 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1935 if (!(all_avail & btrfs_raid_array[i].bg_flag))
1938 if (num_devices < btrfs_raid_array[i].devs_min) {
1939 int ret = btrfs_raid_array[i].mindev_error;
1949 static struct btrfs_device * btrfs_find_next_active_device(
1950 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1952 struct btrfs_device *next_device;
1954 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
1955 if (next_device != device &&
1956 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
1957 && next_device->bdev)
1965 * Helper function to check if the given device is part of s_bdev / latest_bdev
1966 * and replace it with the provided or the next active device, in the context
1967 * where this function called, there should be always be another device (or
1968 * this_dev) which is active.
1970 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
1971 struct btrfs_device *next_device)
1973 struct btrfs_fs_info *fs_info = device->fs_info;
1976 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
1978 ASSERT(next_device);
1980 if (fs_info->sb->s_bdev &&
1981 (fs_info->sb->s_bdev == device->bdev))
1982 fs_info->sb->s_bdev = next_device->bdev;
1984 if (fs_info->fs_devices->latest_bdev == device->bdev)
1985 fs_info->fs_devices->latest_bdev = next_device->bdev;
1989 * Return btrfs_fs_devices::num_devices excluding the device that's being
1990 * currently replaced.
1992 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
1994 u64 num_devices = fs_info->fs_devices->num_devices;
1996 down_read(&fs_info->dev_replace.rwsem);
1997 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
1998 ASSERT(num_devices > 1);
2001 up_read(&fs_info->dev_replace.rwsem);
2006 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
2007 struct block_device *bdev,
2008 const char *device_path)
2010 struct btrfs_super_block *disk_super;
2016 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2020 disk_super = btrfs_read_dev_one_super(bdev, copy_num);
2021 if (IS_ERR(disk_super))
2024 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
2026 page = virt_to_page(disk_super);
2027 set_page_dirty(page);
2029 /* write_on_page() unlocks the page */
2030 ret = write_one_page(page);
2033 "error clearing superblock number %d (%d)",
2035 btrfs_release_disk_super(disk_super);
2039 /* Notify udev that device has changed */
2040 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2042 /* Update ctime/mtime for device path for libblkid */
2043 update_dev_time(device_path);
2046 int btrfs_rm_device(struct btrfs_fs_info *fs_info, const char *device_path,
2049 struct btrfs_device *device;
2050 struct btrfs_fs_devices *cur_devices;
2051 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2055 mutex_lock(&uuid_mutex);
2057 num_devices = btrfs_num_devices(fs_info);
2059 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2063 device = btrfs_find_device_by_devspec(fs_info, devid, device_path);
2065 if (IS_ERR(device)) {
2066 if (PTR_ERR(device) == -ENOENT &&
2067 strcmp(device_path, "missing") == 0)
2068 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2070 ret = PTR_ERR(device);
2074 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2075 btrfs_warn_in_rcu(fs_info,
2076 "cannot remove device %s (devid %llu) due to active swapfile",
2077 rcu_str_deref(device->name), device->devid);
2082 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2083 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
2087 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2088 fs_info->fs_devices->rw_devices == 1) {
2089 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
2093 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2094 mutex_lock(&fs_info->chunk_mutex);
2095 list_del_init(&device->dev_alloc_list);
2096 device->fs_devices->rw_devices--;
2097 mutex_unlock(&fs_info->chunk_mutex);
2100 mutex_unlock(&uuid_mutex);
2101 ret = btrfs_shrink_device(device, 0);
2103 btrfs_reada_remove_dev(device);
2104 mutex_lock(&uuid_mutex);
2109 * TODO: the superblock still includes this device in its num_devices
2110 * counter although write_all_supers() is not locked out. This
2111 * could give a filesystem state which requires a degraded mount.
2113 ret = btrfs_rm_dev_item(device);
2117 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2118 btrfs_scrub_cancel_dev(device);
2121 * the device list mutex makes sure that we don't change
2122 * the device list while someone else is writing out all
2123 * the device supers. Whoever is writing all supers, should
2124 * lock the device list mutex before getting the number of
2125 * devices in the super block (super_copy). Conversely,
2126 * whoever updates the number of devices in the super block
2127 * (super_copy) should hold the device list mutex.
2131 * In normal cases the cur_devices == fs_devices. But in case
2132 * of deleting a seed device, the cur_devices should point to
2133 * its own fs_devices listed under the fs_devices->seed.
2135 cur_devices = device->fs_devices;
2136 mutex_lock(&fs_devices->device_list_mutex);
2137 list_del_rcu(&device->dev_list);
2139 cur_devices->num_devices--;
2140 cur_devices->total_devices--;
2141 /* Update total_devices of the parent fs_devices if it's seed */
2142 if (cur_devices != fs_devices)
2143 fs_devices->total_devices--;
2145 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2146 cur_devices->missing_devices--;
2148 btrfs_assign_next_active_device(device, NULL);
2151 cur_devices->open_devices--;
2152 /* remove sysfs entry */
2153 btrfs_sysfs_remove_device(device);
2156 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2157 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2158 mutex_unlock(&fs_devices->device_list_mutex);
2161 * at this point, the device is zero sized and detached from
2162 * the devices list. All that's left is to zero out the old
2163 * supers and free the device.
2165 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2166 btrfs_scratch_superblocks(fs_info, device->bdev,
2169 btrfs_close_bdev(device);
2171 btrfs_free_device(device);
2173 if (cur_devices->open_devices == 0) {
2174 list_del_init(&cur_devices->seed_list);
2175 close_fs_devices(cur_devices);
2176 free_fs_devices(cur_devices);
2180 mutex_unlock(&uuid_mutex);
2184 btrfs_reada_undo_remove_dev(device);
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_devices *fs_devices = srcdev->fs_devices;
2226 mutex_lock(&uuid_mutex);
2228 btrfs_close_bdev(srcdev);
2230 btrfs_free_device(srcdev);
2232 /* if this is no devs we rather delete the fs_devices */
2233 if (!fs_devices->num_devices) {
2235 * On a mounted FS, num_devices can't be zero unless it's a
2236 * seed. In case of a seed device being replaced, the replace
2237 * target added to the sprout FS, so there will be no more
2238 * device left under the seed FS.
2240 ASSERT(fs_devices->seeding);
2242 list_del_init(&fs_devices->seed_list);
2243 close_fs_devices(fs_devices);
2244 free_fs_devices(fs_devices);
2246 mutex_unlock(&uuid_mutex);
2249 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2251 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2253 mutex_lock(&fs_devices->device_list_mutex);
2255 btrfs_sysfs_remove_device(tgtdev);
2258 fs_devices->open_devices--;
2260 fs_devices->num_devices--;
2262 btrfs_assign_next_active_device(tgtdev, NULL);
2264 list_del_rcu(&tgtdev->dev_list);
2266 mutex_unlock(&fs_devices->device_list_mutex);
2269 * The update_dev_time() with in btrfs_scratch_superblocks()
2270 * may lead to a call to btrfs_show_devname() which will try
2271 * to hold device_list_mutex. And here this device
2272 * is already out of device list, so we don't have to hold
2273 * the device_list_mutex lock.
2275 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2278 btrfs_close_bdev(tgtdev);
2280 btrfs_free_device(tgtdev);
2283 static struct btrfs_device *btrfs_find_device_by_path(
2284 struct btrfs_fs_info *fs_info, const char *device_path)
2287 struct btrfs_super_block *disk_super;
2290 struct block_device *bdev;
2291 struct btrfs_device *device;
2293 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
2294 fs_info->bdev_holder, 0, &bdev, &disk_super);
2296 return ERR_PTR(ret);
2298 devid = btrfs_stack_device_id(&disk_super->dev_item);
2299 dev_uuid = disk_super->dev_item.uuid;
2300 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2301 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2302 disk_super->metadata_uuid, true);
2304 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2305 disk_super->fsid, true);
2307 btrfs_release_disk_super(disk_super);
2309 device = ERR_PTR(-ENOENT);
2310 blkdev_put(bdev, FMODE_READ);
2315 * Lookup a device given by device id, or the path if the id is 0.
2317 struct btrfs_device *btrfs_find_device_by_devspec(
2318 struct btrfs_fs_info *fs_info, u64 devid,
2319 const char *device_path)
2321 struct btrfs_device *device;
2324 device = btrfs_find_device(fs_info->fs_devices, devid, NULL,
2327 return ERR_PTR(-ENOENT);
2331 if (!device_path || !device_path[0])
2332 return ERR_PTR(-EINVAL);
2334 if (strcmp(device_path, "missing") == 0) {
2335 /* Find first missing device */
2336 list_for_each_entry(device, &fs_info->fs_devices->devices,
2338 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
2339 &device->dev_state) && !device->bdev)
2342 return ERR_PTR(-ENOENT);
2345 return btrfs_find_device_by_path(fs_info, device_path);
2349 * does all the dirty work required for changing file system's UUID.
2351 static int btrfs_prepare_sprout(struct btrfs_fs_info *fs_info)
2353 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2354 struct btrfs_fs_devices *old_devices;
2355 struct btrfs_fs_devices *seed_devices;
2356 struct btrfs_super_block *disk_super = fs_info->super_copy;
2357 struct btrfs_device *device;
2360 lockdep_assert_held(&uuid_mutex);
2361 if (!fs_devices->seeding)
2365 * Private copy of the seed devices, anchored at
2366 * fs_info->fs_devices->seed_list
2368 seed_devices = alloc_fs_devices(NULL, NULL);
2369 if (IS_ERR(seed_devices))
2370 return PTR_ERR(seed_devices);
2373 * It's necessary to retain a copy of the original seed fs_devices in
2374 * fs_uuids so that filesystems which have been seeded can successfully
2375 * reference the seed device from open_seed_devices. This also supports
2378 old_devices = clone_fs_devices(fs_devices);
2379 if (IS_ERR(old_devices)) {
2380 kfree(seed_devices);
2381 return PTR_ERR(old_devices);
2384 list_add(&old_devices->fs_list, &fs_uuids);
2386 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2387 seed_devices->opened = 1;
2388 INIT_LIST_HEAD(&seed_devices->devices);
2389 INIT_LIST_HEAD(&seed_devices->alloc_list);
2390 mutex_init(&seed_devices->device_list_mutex);
2392 mutex_lock(&fs_devices->device_list_mutex);
2393 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2395 list_for_each_entry(device, &seed_devices->devices, dev_list)
2396 device->fs_devices = seed_devices;
2398 fs_devices->seeding = false;
2399 fs_devices->num_devices = 0;
2400 fs_devices->open_devices = 0;
2401 fs_devices->missing_devices = 0;
2402 fs_devices->rotating = false;
2403 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2405 generate_random_uuid(fs_devices->fsid);
2406 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2407 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2408 mutex_unlock(&fs_devices->device_list_mutex);
2410 super_flags = btrfs_super_flags(disk_super) &
2411 ~BTRFS_SUPER_FLAG_SEEDING;
2412 btrfs_set_super_flags(disk_super, super_flags);
2418 * Store the expected generation for seed devices in device items.
2420 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2422 struct btrfs_fs_info *fs_info = trans->fs_info;
2423 struct btrfs_root *root = fs_info->chunk_root;
2424 struct btrfs_path *path;
2425 struct extent_buffer *leaf;
2426 struct btrfs_dev_item *dev_item;
2427 struct btrfs_device *device;
2428 struct btrfs_key key;
2429 u8 fs_uuid[BTRFS_FSID_SIZE];
2430 u8 dev_uuid[BTRFS_UUID_SIZE];
2434 path = btrfs_alloc_path();
2438 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2440 key.type = BTRFS_DEV_ITEM_KEY;
2443 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2447 leaf = path->nodes[0];
2449 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2450 ret = btrfs_next_leaf(root, path);
2455 leaf = path->nodes[0];
2456 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2457 btrfs_release_path(path);
2461 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2462 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2463 key.type != BTRFS_DEV_ITEM_KEY)
2466 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2467 struct btrfs_dev_item);
2468 devid = btrfs_device_id(leaf, dev_item);
2469 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2471 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2473 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
2475 BUG_ON(!device); /* Logic error */
2477 if (device->fs_devices->seeding) {
2478 btrfs_set_device_generation(leaf, dev_item,
2479 device->generation);
2480 btrfs_mark_buffer_dirty(leaf);
2488 btrfs_free_path(path);
2492 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2494 struct btrfs_root *root = fs_info->dev_root;
2495 struct request_queue *q;
2496 struct btrfs_trans_handle *trans;
2497 struct btrfs_device *device;
2498 struct block_device *bdev;
2499 struct super_block *sb = fs_info->sb;
2500 struct rcu_string *name;
2501 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2502 u64 orig_super_total_bytes;
2503 u64 orig_super_num_devices;
2504 int seeding_dev = 0;
2506 bool locked = false;
2508 if (sb_rdonly(sb) && !fs_devices->seeding)
2511 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2512 fs_info->bdev_holder);
2514 return PTR_ERR(bdev);
2516 if (fs_devices->seeding) {
2518 down_write(&sb->s_umount);
2519 mutex_lock(&uuid_mutex);
2523 sync_blockdev(bdev);
2526 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2527 if (device->bdev == bdev) {
2535 device = btrfs_alloc_device(fs_info, NULL, NULL);
2536 if (IS_ERR(device)) {
2537 /* we can safely leave the fs_devices entry around */
2538 ret = PTR_ERR(device);
2542 name = rcu_string_strdup(device_path, GFP_KERNEL);
2545 goto error_free_device;
2547 rcu_assign_pointer(device->name, name);
2549 trans = btrfs_start_transaction(root, 0);
2550 if (IS_ERR(trans)) {
2551 ret = PTR_ERR(trans);
2552 goto error_free_device;
2555 q = bdev_get_queue(bdev);
2556 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2557 device->generation = trans->transid;
2558 device->io_width = fs_info->sectorsize;
2559 device->io_align = fs_info->sectorsize;
2560 device->sector_size = fs_info->sectorsize;
2561 device->total_bytes = round_down(i_size_read(bdev->bd_inode),
2562 fs_info->sectorsize);
2563 device->disk_total_bytes = device->total_bytes;
2564 device->commit_total_bytes = device->total_bytes;
2565 device->fs_info = fs_info;
2566 device->bdev = bdev;
2567 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2568 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2569 device->mode = FMODE_EXCL;
2570 device->dev_stats_valid = 1;
2571 set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2574 sb->s_flags &= ~SB_RDONLY;
2575 ret = btrfs_prepare_sprout(fs_info);
2577 btrfs_abort_transaction(trans, ret);
2582 device->fs_devices = fs_devices;
2584 mutex_lock(&fs_devices->device_list_mutex);
2585 mutex_lock(&fs_info->chunk_mutex);
2586 list_add_rcu(&device->dev_list, &fs_devices->devices);
2587 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2588 fs_devices->num_devices++;
2589 fs_devices->open_devices++;
2590 fs_devices->rw_devices++;
2591 fs_devices->total_devices++;
2592 fs_devices->total_rw_bytes += device->total_bytes;
2594 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2596 if (!blk_queue_nonrot(q))
2597 fs_devices->rotating = true;
2599 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2600 btrfs_set_super_total_bytes(fs_info->super_copy,
2601 round_down(orig_super_total_bytes + device->total_bytes,
2602 fs_info->sectorsize));
2604 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2605 btrfs_set_super_num_devices(fs_info->super_copy,
2606 orig_super_num_devices + 1);
2609 * we've got more storage, clear any full flags on the space
2612 btrfs_clear_space_info_full(fs_info);
2614 mutex_unlock(&fs_info->chunk_mutex);
2616 /* Add sysfs device entry */
2617 btrfs_sysfs_add_device(device);
2619 mutex_unlock(&fs_devices->device_list_mutex);
2622 mutex_lock(&fs_info->chunk_mutex);
2623 ret = init_first_rw_device(trans);
2624 mutex_unlock(&fs_info->chunk_mutex);
2626 btrfs_abort_transaction(trans, ret);
2631 ret = btrfs_add_dev_item(trans, device);
2633 btrfs_abort_transaction(trans, ret);
2638 ret = btrfs_finish_sprout(trans);
2640 btrfs_abort_transaction(trans, ret);
2645 * fs_devices now represents the newly sprouted filesystem and
2646 * its fsid has been changed by btrfs_prepare_sprout
2648 btrfs_sysfs_update_sprout_fsid(fs_devices);
2651 ret = btrfs_commit_transaction(trans);
2654 mutex_unlock(&uuid_mutex);
2655 up_write(&sb->s_umount);
2658 if (ret) /* transaction commit */
2661 ret = btrfs_relocate_sys_chunks(fs_info);
2663 btrfs_handle_fs_error(fs_info, ret,
2664 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2665 trans = btrfs_attach_transaction(root);
2666 if (IS_ERR(trans)) {
2667 if (PTR_ERR(trans) == -ENOENT)
2669 ret = PTR_ERR(trans);
2673 ret = btrfs_commit_transaction(trans);
2677 * Now that we have written a new super block to this device, check all
2678 * other fs_devices list if device_path alienates any other scanned
2680 * We can ignore the return value as it typically returns -EINVAL and
2681 * only succeeds if the device was an alien.
2683 btrfs_forget_devices(device_path);
2685 /* Update ctime/mtime for blkid or udev */
2686 update_dev_time(device_path);
2691 btrfs_sysfs_remove_device(device);
2692 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2693 mutex_lock(&fs_info->chunk_mutex);
2694 list_del_rcu(&device->dev_list);
2695 list_del(&device->dev_alloc_list);
2696 fs_info->fs_devices->num_devices--;
2697 fs_info->fs_devices->open_devices--;
2698 fs_info->fs_devices->rw_devices--;
2699 fs_info->fs_devices->total_devices--;
2700 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2701 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2702 btrfs_set_super_total_bytes(fs_info->super_copy,
2703 orig_super_total_bytes);
2704 btrfs_set_super_num_devices(fs_info->super_copy,
2705 orig_super_num_devices);
2706 mutex_unlock(&fs_info->chunk_mutex);
2707 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2710 sb->s_flags |= SB_RDONLY;
2712 btrfs_end_transaction(trans);
2714 btrfs_free_device(device);
2716 blkdev_put(bdev, FMODE_EXCL);
2718 mutex_unlock(&uuid_mutex);
2719 up_write(&sb->s_umount);
2724 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2725 struct btrfs_device *device)
2728 struct btrfs_path *path;
2729 struct btrfs_root *root = device->fs_info->chunk_root;
2730 struct btrfs_dev_item *dev_item;
2731 struct extent_buffer *leaf;
2732 struct btrfs_key key;
2734 path = btrfs_alloc_path();
2738 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2739 key.type = BTRFS_DEV_ITEM_KEY;
2740 key.offset = device->devid;
2742 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2751 leaf = path->nodes[0];
2752 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2754 btrfs_set_device_id(leaf, dev_item, device->devid);
2755 btrfs_set_device_type(leaf, dev_item, device->type);
2756 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2757 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2758 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2759 btrfs_set_device_total_bytes(leaf, dev_item,
2760 btrfs_device_get_disk_total_bytes(device));
2761 btrfs_set_device_bytes_used(leaf, dev_item,
2762 btrfs_device_get_bytes_used(device));
2763 btrfs_mark_buffer_dirty(leaf);
2766 btrfs_free_path(path);
2770 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2771 struct btrfs_device *device, u64 new_size)
2773 struct btrfs_fs_info *fs_info = device->fs_info;
2774 struct btrfs_super_block *super_copy = fs_info->super_copy;
2778 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2781 new_size = round_down(new_size, fs_info->sectorsize);
2783 mutex_lock(&fs_info->chunk_mutex);
2784 old_total = btrfs_super_total_bytes(super_copy);
2785 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2787 if (new_size <= device->total_bytes ||
2788 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2789 mutex_unlock(&fs_info->chunk_mutex);
2793 btrfs_set_super_total_bytes(super_copy,
2794 round_down(old_total + diff, fs_info->sectorsize));
2795 device->fs_devices->total_rw_bytes += diff;
2797 btrfs_device_set_total_bytes(device, new_size);
2798 btrfs_device_set_disk_total_bytes(device, new_size);
2799 btrfs_clear_space_info_full(device->fs_info);
2800 if (list_empty(&device->post_commit_list))
2801 list_add_tail(&device->post_commit_list,
2802 &trans->transaction->dev_update_list);
2803 mutex_unlock(&fs_info->chunk_mutex);
2805 return btrfs_update_device(trans, device);
2808 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2810 struct btrfs_fs_info *fs_info = trans->fs_info;
2811 struct btrfs_root *root = fs_info->chunk_root;
2813 struct btrfs_path *path;
2814 struct btrfs_key key;
2816 path = btrfs_alloc_path();
2820 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2821 key.offset = chunk_offset;
2822 key.type = BTRFS_CHUNK_ITEM_KEY;
2824 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2827 else if (ret > 0) { /* Logic error or corruption */
2828 btrfs_handle_fs_error(fs_info, -ENOENT,
2829 "Failed lookup while freeing chunk.");
2834 ret = btrfs_del_item(trans, root, path);
2836 btrfs_handle_fs_error(fs_info, ret,
2837 "Failed to delete chunk item.");
2839 btrfs_free_path(path);
2843 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2845 struct btrfs_super_block *super_copy = fs_info->super_copy;
2846 struct btrfs_disk_key *disk_key;
2847 struct btrfs_chunk *chunk;
2854 struct btrfs_key key;
2856 mutex_lock(&fs_info->chunk_mutex);
2857 array_size = btrfs_super_sys_array_size(super_copy);
2859 ptr = super_copy->sys_chunk_array;
2862 while (cur < array_size) {
2863 disk_key = (struct btrfs_disk_key *)ptr;
2864 btrfs_disk_key_to_cpu(&key, disk_key);
2866 len = sizeof(*disk_key);
2868 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2869 chunk = (struct btrfs_chunk *)(ptr + len);
2870 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2871 len += btrfs_chunk_item_size(num_stripes);
2876 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
2877 key.offset == chunk_offset) {
2878 memmove(ptr, ptr + len, array_size - (cur + len));
2880 btrfs_set_super_sys_array_size(super_copy, array_size);
2886 mutex_unlock(&fs_info->chunk_mutex);
2891 * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
2892 * @logical: Logical block offset in bytes.
2893 * @length: Length of extent in bytes.
2895 * Return: Chunk mapping or ERR_PTR.
2897 struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
2898 u64 logical, u64 length)
2900 struct extent_map_tree *em_tree;
2901 struct extent_map *em;
2903 em_tree = &fs_info->mapping_tree;
2904 read_lock(&em_tree->lock);
2905 em = lookup_extent_mapping(em_tree, logical, length);
2906 read_unlock(&em_tree->lock);
2909 btrfs_crit(fs_info, "unable to find logical %llu length %llu",
2911 return ERR_PTR(-EINVAL);
2914 if (em->start > logical || em->start + em->len < logical) {
2916 "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
2917 logical, length, em->start, em->start + em->len);
2918 free_extent_map(em);
2919 return ERR_PTR(-EINVAL);
2922 /* callers are responsible for dropping em's ref. */
2926 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2928 struct btrfs_fs_info *fs_info = trans->fs_info;
2929 struct extent_map *em;
2930 struct map_lookup *map;
2931 u64 dev_extent_len = 0;
2933 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2935 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
2938 * This is a logic error, but we don't want to just rely on the
2939 * user having built with ASSERT enabled, so if ASSERT doesn't
2940 * do anything we still error out.
2945 map = em->map_lookup;
2946 mutex_lock(&fs_info->chunk_mutex);
2947 check_system_chunk(trans, map->type);
2948 mutex_unlock(&fs_info->chunk_mutex);
2951 * Take the device list mutex to prevent races with the final phase of
2952 * a device replace operation that replaces the device object associated
2953 * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()).
2955 mutex_lock(&fs_devices->device_list_mutex);
2956 for (i = 0; i < map->num_stripes; i++) {
2957 struct btrfs_device *device = map->stripes[i].dev;
2958 ret = btrfs_free_dev_extent(trans, device,
2959 map->stripes[i].physical,
2962 mutex_unlock(&fs_devices->device_list_mutex);
2963 btrfs_abort_transaction(trans, ret);
2967 if (device->bytes_used > 0) {
2968 mutex_lock(&fs_info->chunk_mutex);
2969 btrfs_device_set_bytes_used(device,
2970 device->bytes_used - dev_extent_len);
2971 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
2972 btrfs_clear_space_info_full(fs_info);
2973 mutex_unlock(&fs_info->chunk_mutex);
2976 ret = btrfs_update_device(trans, device);
2978 mutex_unlock(&fs_devices->device_list_mutex);
2979 btrfs_abort_transaction(trans, ret);
2983 mutex_unlock(&fs_devices->device_list_mutex);
2985 ret = btrfs_free_chunk(trans, chunk_offset);
2987 btrfs_abort_transaction(trans, ret);
2991 trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
2993 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2994 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
2996 btrfs_abort_transaction(trans, ret);
3001 ret = btrfs_remove_block_group(trans, chunk_offset, em);
3003 btrfs_abort_transaction(trans, ret);
3009 free_extent_map(em);
3013 static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3015 struct btrfs_root *root = fs_info->chunk_root;
3016 struct btrfs_trans_handle *trans;
3017 struct btrfs_block_group *block_group;
3021 * Prevent races with automatic removal of unused block groups.
3022 * After we relocate and before we remove the chunk with offset
3023 * chunk_offset, automatic removal of the block group can kick in,
3024 * resulting in a failure when calling btrfs_remove_chunk() below.
3026 * Make sure to acquire this mutex before doing a tree search (dev
3027 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3028 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3029 * we release the path used to search the chunk/dev tree and before
3030 * the current task acquires this mutex and calls us.
3032 lockdep_assert_held(&fs_info->delete_unused_bgs_mutex);
3034 /* step one, relocate all the extents inside this chunk */
3035 btrfs_scrub_pause(fs_info);
3036 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3037 btrfs_scrub_continue(fs_info);
3041 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3044 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3045 btrfs_put_block_group(block_group);
3047 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3049 if (IS_ERR(trans)) {
3050 ret = PTR_ERR(trans);
3051 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3056 * step two, delete the device extents and the
3057 * chunk tree entries
3059 ret = btrfs_remove_chunk(trans, chunk_offset);
3060 btrfs_end_transaction(trans);
3064 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3066 struct btrfs_root *chunk_root = fs_info->chunk_root;
3067 struct btrfs_path *path;
3068 struct extent_buffer *leaf;
3069 struct btrfs_chunk *chunk;
3070 struct btrfs_key key;
3071 struct btrfs_key found_key;
3073 bool retried = false;
3077 path = btrfs_alloc_path();
3082 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3083 key.offset = (u64)-1;
3084 key.type = BTRFS_CHUNK_ITEM_KEY;
3087 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3088 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3090 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3093 BUG_ON(ret == 0); /* Corruption */
3095 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3098 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3104 leaf = path->nodes[0];
3105 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3107 chunk = btrfs_item_ptr(leaf, path->slots[0],
3108 struct btrfs_chunk);
3109 chunk_type = btrfs_chunk_type(leaf, chunk);
3110 btrfs_release_path(path);
3112 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3113 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3119 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3121 if (found_key.offset == 0)
3123 key.offset = found_key.offset - 1;
3126 if (failed && !retried) {
3130 } else if (WARN_ON(failed && retried)) {
3134 btrfs_free_path(path);
3139 * return 1 : allocate a data chunk successfully,
3140 * return <0: errors during allocating a data chunk,
3141 * return 0 : no need to allocate a data chunk.
3143 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3146 struct btrfs_block_group *cache;
3150 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3152 chunk_type = cache->flags;
3153 btrfs_put_block_group(cache);
3155 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3158 spin_lock(&fs_info->data_sinfo->lock);
3159 bytes_used = fs_info->data_sinfo->bytes_used;
3160 spin_unlock(&fs_info->data_sinfo->lock);
3163 struct btrfs_trans_handle *trans;
3166 trans = btrfs_join_transaction(fs_info->tree_root);
3168 return PTR_ERR(trans);
3170 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3171 btrfs_end_transaction(trans);
3180 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3181 struct btrfs_balance_control *bctl)
3183 struct btrfs_root *root = fs_info->tree_root;
3184 struct btrfs_trans_handle *trans;
3185 struct btrfs_balance_item *item;
3186 struct btrfs_disk_balance_args disk_bargs;
3187 struct btrfs_path *path;
3188 struct extent_buffer *leaf;
3189 struct btrfs_key key;
3192 path = btrfs_alloc_path();
3196 trans = btrfs_start_transaction(root, 0);
3197 if (IS_ERR(trans)) {
3198 btrfs_free_path(path);
3199 return PTR_ERR(trans);
3202 key.objectid = BTRFS_BALANCE_OBJECTID;
3203 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3206 ret = btrfs_insert_empty_item(trans, root, path, &key,
3211 leaf = path->nodes[0];
3212 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3214 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3216 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3217 btrfs_set_balance_data(leaf, item, &disk_bargs);
3218 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3219 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3220 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3221 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3223 btrfs_set_balance_flags(leaf, item, bctl->flags);
3225 btrfs_mark_buffer_dirty(leaf);
3227 btrfs_free_path(path);
3228 err = btrfs_commit_transaction(trans);
3234 static int del_balance_item(struct btrfs_fs_info *fs_info)
3236 struct btrfs_root *root = fs_info->tree_root;
3237 struct btrfs_trans_handle *trans;
3238 struct btrfs_path *path;
3239 struct btrfs_key key;
3242 path = btrfs_alloc_path();
3246 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3247 if (IS_ERR(trans)) {
3248 btrfs_free_path(path);
3249 return PTR_ERR(trans);
3252 key.objectid = BTRFS_BALANCE_OBJECTID;
3253 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3256 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3264 ret = btrfs_del_item(trans, root, path);
3266 btrfs_free_path(path);
3267 err = btrfs_commit_transaction(trans);
3274 * This is a heuristic used to reduce the number of chunks balanced on
3275 * resume after balance was interrupted.
3277 static void update_balance_args(struct btrfs_balance_control *bctl)
3280 * Turn on soft mode for chunk types that were being converted.
3282 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3283 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3284 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3285 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3286 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3287 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3290 * Turn on usage filter if is not already used. The idea is
3291 * that chunks that we have already balanced should be
3292 * reasonably full. Don't do it for chunks that are being
3293 * converted - that will keep us from relocating unconverted
3294 * (albeit full) chunks.
3296 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3297 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3298 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3299 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3300 bctl->data.usage = 90;
3302 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3303 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3304 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3305 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3306 bctl->sys.usage = 90;
3308 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3309 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3310 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3311 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3312 bctl->meta.usage = 90;
3317 * Clear the balance status in fs_info and delete the balance item from disk.
3319 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3321 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3324 BUG_ON(!fs_info->balance_ctl);
3326 spin_lock(&fs_info->balance_lock);
3327 fs_info->balance_ctl = NULL;
3328 spin_unlock(&fs_info->balance_lock);
3331 ret = del_balance_item(fs_info);
3333 btrfs_handle_fs_error(fs_info, ret, NULL);
3337 * Balance filters. Return 1 if chunk should be filtered out
3338 * (should not be balanced).
3340 static int chunk_profiles_filter(u64 chunk_type,
3341 struct btrfs_balance_args *bargs)
3343 chunk_type = chunk_to_extended(chunk_type) &
3344 BTRFS_EXTENDED_PROFILE_MASK;
3346 if (bargs->profiles & chunk_type)
3352 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3353 struct btrfs_balance_args *bargs)
3355 struct btrfs_block_group *cache;
3357 u64 user_thresh_min;
3358 u64 user_thresh_max;
3361 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3362 chunk_used = cache->used;
3364 if (bargs->usage_min == 0)
3365 user_thresh_min = 0;
3367 user_thresh_min = div_factor_fine(cache->length,
3370 if (bargs->usage_max == 0)
3371 user_thresh_max = 1;
3372 else if (bargs->usage_max > 100)
3373 user_thresh_max = cache->length;
3375 user_thresh_max = div_factor_fine(cache->length,
3378 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3381 btrfs_put_block_group(cache);
3385 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3386 u64 chunk_offset, struct btrfs_balance_args *bargs)
3388 struct btrfs_block_group *cache;
3389 u64 chunk_used, user_thresh;
3392 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3393 chunk_used = cache->used;
3395 if (bargs->usage_min == 0)
3397 else if (bargs->usage > 100)
3398 user_thresh = cache->length;
3400 user_thresh = div_factor_fine(cache->length, bargs->usage);
3402 if (chunk_used < user_thresh)
3405 btrfs_put_block_group(cache);
3409 static int chunk_devid_filter(struct extent_buffer *leaf,
3410 struct btrfs_chunk *chunk,
3411 struct btrfs_balance_args *bargs)
3413 struct btrfs_stripe *stripe;
3414 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3417 for (i = 0; i < num_stripes; i++) {
3418 stripe = btrfs_stripe_nr(chunk, i);
3419 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3426 static u64 calc_data_stripes(u64 type, int num_stripes)
3428 const int index = btrfs_bg_flags_to_raid_index(type);
3429 const int ncopies = btrfs_raid_array[index].ncopies;
3430 const int nparity = btrfs_raid_array[index].nparity;
3433 return num_stripes - nparity;
3435 return num_stripes / ncopies;
3438 /* [pstart, pend) */
3439 static int chunk_drange_filter(struct extent_buffer *leaf,
3440 struct btrfs_chunk *chunk,
3441 struct btrfs_balance_args *bargs)
3443 struct btrfs_stripe *stripe;
3444 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3451 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3454 type = btrfs_chunk_type(leaf, chunk);
3455 factor = calc_data_stripes(type, num_stripes);
3457 for (i = 0; i < num_stripes; i++) {
3458 stripe = btrfs_stripe_nr(chunk, i);
3459 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3462 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3463 stripe_length = btrfs_chunk_length(leaf, chunk);
3464 stripe_length = div_u64(stripe_length, factor);
3466 if (stripe_offset < bargs->pend &&
3467 stripe_offset + stripe_length > bargs->pstart)
3474 /* [vstart, vend) */
3475 static int chunk_vrange_filter(struct extent_buffer *leaf,
3476 struct btrfs_chunk *chunk,
3478 struct btrfs_balance_args *bargs)
3480 if (chunk_offset < bargs->vend &&
3481 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3482 /* at least part of the chunk is inside this vrange */
3488 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3489 struct btrfs_chunk *chunk,
3490 struct btrfs_balance_args *bargs)
3492 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3494 if (bargs->stripes_min <= num_stripes
3495 && num_stripes <= bargs->stripes_max)
3501 static int chunk_soft_convert_filter(u64 chunk_type,
3502 struct btrfs_balance_args *bargs)
3504 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3507 chunk_type = chunk_to_extended(chunk_type) &
3508 BTRFS_EXTENDED_PROFILE_MASK;
3510 if (bargs->target == chunk_type)
3516 static int should_balance_chunk(struct extent_buffer *leaf,
3517 struct btrfs_chunk *chunk, u64 chunk_offset)
3519 struct btrfs_fs_info *fs_info = leaf->fs_info;
3520 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3521 struct btrfs_balance_args *bargs = NULL;
3522 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3525 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3526 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3530 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3531 bargs = &bctl->data;
3532 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3534 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3535 bargs = &bctl->meta;
3537 /* profiles filter */
3538 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3539 chunk_profiles_filter(chunk_type, bargs)) {
3544 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3545 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3547 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3548 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3553 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3554 chunk_devid_filter(leaf, chunk, bargs)) {
3558 /* drange filter, makes sense only with devid filter */
3559 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3560 chunk_drange_filter(leaf, chunk, bargs)) {
3565 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3566 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3570 /* stripes filter */
3571 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3572 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3576 /* soft profile changing mode */
3577 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3578 chunk_soft_convert_filter(chunk_type, bargs)) {
3583 * limited by count, must be the last filter
3585 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3586 if (bargs->limit == 0)
3590 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3592 * Same logic as the 'limit' filter; the minimum cannot be
3593 * determined here because we do not have the global information
3594 * about the count of all chunks that satisfy the filters.
3596 if (bargs->limit_max == 0)
3605 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3607 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3608 struct btrfs_root *chunk_root = fs_info->chunk_root;
3610 struct btrfs_chunk *chunk;
3611 struct btrfs_path *path = NULL;
3612 struct btrfs_key key;
3613 struct btrfs_key found_key;
3614 struct extent_buffer *leaf;
3617 int enospc_errors = 0;
3618 bool counting = true;
3619 /* The single value limit and min/max limits use the same bytes in the */
3620 u64 limit_data = bctl->data.limit;
3621 u64 limit_meta = bctl->meta.limit;
3622 u64 limit_sys = bctl->sys.limit;
3626 int chunk_reserved = 0;
3628 path = btrfs_alloc_path();
3634 /* zero out stat counters */
3635 spin_lock(&fs_info->balance_lock);
3636 memset(&bctl->stat, 0, sizeof(bctl->stat));
3637 spin_unlock(&fs_info->balance_lock);
3641 * The single value limit and min/max limits use the same bytes
3644 bctl->data.limit = limit_data;
3645 bctl->meta.limit = limit_meta;
3646 bctl->sys.limit = limit_sys;
3648 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3649 key.offset = (u64)-1;
3650 key.type = BTRFS_CHUNK_ITEM_KEY;
3653 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3654 atomic_read(&fs_info->balance_cancel_req)) {
3659 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3660 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3662 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3667 * this shouldn't happen, it means the last relocate
3671 BUG(); /* FIXME break ? */
3673 ret = btrfs_previous_item(chunk_root, path, 0,
3674 BTRFS_CHUNK_ITEM_KEY);
3676 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3681 leaf = path->nodes[0];
3682 slot = path->slots[0];
3683 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3685 if (found_key.objectid != key.objectid) {
3686 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3690 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3691 chunk_type = btrfs_chunk_type(leaf, chunk);
3694 spin_lock(&fs_info->balance_lock);
3695 bctl->stat.considered++;
3696 spin_unlock(&fs_info->balance_lock);
3699 ret = should_balance_chunk(leaf, chunk, found_key.offset);
3701 btrfs_release_path(path);
3703 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3708 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3709 spin_lock(&fs_info->balance_lock);
3710 bctl->stat.expected++;
3711 spin_unlock(&fs_info->balance_lock);
3713 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3715 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3717 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3724 * Apply limit_min filter, no need to check if the LIMITS
3725 * filter is used, limit_min is 0 by default
3727 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3728 count_data < bctl->data.limit_min)
3729 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3730 count_meta < bctl->meta.limit_min)
3731 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3732 count_sys < bctl->sys.limit_min)) {
3733 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3737 if (!chunk_reserved) {
3739 * We may be relocating the only data chunk we have,
3740 * which could potentially end up with losing data's
3741 * raid profile, so lets allocate an empty one in
3744 ret = btrfs_may_alloc_data_chunk(fs_info,
3747 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3749 } else if (ret == 1) {
3754 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3755 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3756 if (ret == -ENOSPC) {
3758 } else if (ret == -ETXTBSY) {
3760 "skipping relocation of block group %llu due to active swapfile",
3766 spin_lock(&fs_info->balance_lock);
3767 bctl->stat.completed++;
3768 spin_unlock(&fs_info->balance_lock);
3771 if (found_key.offset == 0)
3773 key.offset = found_key.offset - 1;
3777 btrfs_release_path(path);
3782 btrfs_free_path(path);
3783 if (enospc_errors) {
3784 btrfs_info(fs_info, "%d enospc errors during balance",
3794 * alloc_profile_is_valid - see if a given profile is valid and reduced
3795 * @flags: profile to validate
3796 * @extended: if true @flags is treated as an extended profile
3798 static int alloc_profile_is_valid(u64 flags, int extended)
3800 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3801 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3803 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3805 /* 1) check that all other bits are zeroed */
3809 /* 2) see if profile is reduced */
3811 return !extended; /* "0" is valid for usual profiles */
3813 return has_single_bit_set(flags);
3816 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3818 /* cancel requested || normal exit path */
3819 return atomic_read(&fs_info->balance_cancel_req) ||
3820 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3821 atomic_read(&fs_info->balance_cancel_req) == 0);
3825 * Validate target profile against allowed profiles and return true if it's OK.
3826 * Otherwise print the error message and return false.
3828 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
3829 const struct btrfs_balance_args *bargs,
3830 u64 allowed, const char *type)
3832 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3835 /* Profile is valid and does not have bits outside of the allowed set */
3836 if (alloc_profile_is_valid(bargs->target, 1) &&
3837 (bargs->target & ~allowed) == 0)
3840 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
3841 type, btrfs_bg_type_to_raid_name(bargs->target));
3846 * Fill @buf with textual description of balance filter flags @bargs, up to
3847 * @size_buf including the terminating null. The output may be trimmed if it
3848 * does not fit into the provided buffer.
3850 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
3854 u32 size_bp = size_buf;
3856 u64 flags = bargs->flags;
3857 char tmp_buf[128] = {'\0'};
3862 #define CHECK_APPEND_NOARG(a) \
3864 ret = snprintf(bp, size_bp, (a)); \
3865 if (ret < 0 || ret >= size_bp) \
3866 goto out_overflow; \
3871 #define CHECK_APPEND_1ARG(a, v1) \
3873 ret = snprintf(bp, size_bp, (a), (v1)); \
3874 if (ret < 0 || ret >= size_bp) \
3875 goto out_overflow; \
3880 #define CHECK_APPEND_2ARG(a, v1, v2) \
3882 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
3883 if (ret < 0 || ret >= size_bp) \
3884 goto out_overflow; \
3889 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
3890 CHECK_APPEND_1ARG("convert=%s,",
3891 btrfs_bg_type_to_raid_name(bargs->target));
3893 if (flags & BTRFS_BALANCE_ARGS_SOFT)
3894 CHECK_APPEND_NOARG("soft,");
3896 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
3897 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
3899 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
3902 if (flags & BTRFS_BALANCE_ARGS_USAGE)
3903 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
3905 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
3906 CHECK_APPEND_2ARG("usage=%u..%u,",
3907 bargs->usage_min, bargs->usage_max);
3909 if (flags & BTRFS_BALANCE_ARGS_DEVID)
3910 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
3912 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
3913 CHECK_APPEND_2ARG("drange=%llu..%llu,",
3914 bargs->pstart, bargs->pend);
3916 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
3917 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
3918 bargs->vstart, bargs->vend);
3920 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
3921 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
3923 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
3924 CHECK_APPEND_2ARG("limit=%u..%u,",
3925 bargs->limit_min, bargs->limit_max);
3927 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
3928 CHECK_APPEND_2ARG("stripes=%u..%u,",
3929 bargs->stripes_min, bargs->stripes_max);
3931 #undef CHECK_APPEND_2ARG
3932 #undef CHECK_APPEND_1ARG
3933 #undef CHECK_APPEND_NOARG
3937 if (size_bp < size_buf)
3938 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
3943 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
3945 u32 size_buf = 1024;
3946 char tmp_buf[192] = {'\0'};
3949 u32 size_bp = size_buf;
3951 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3953 buf = kzalloc(size_buf, GFP_KERNEL);
3959 #define CHECK_APPEND_1ARG(a, v1) \
3961 ret = snprintf(bp, size_bp, (a), (v1)); \
3962 if (ret < 0 || ret >= size_bp) \
3963 goto out_overflow; \
3968 if (bctl->flags & BTRFS_BALANCE_FORCE)
3969 CHECK_APPEND_1ARG("%s", "-f ");
3971 if (bctl->flags & BTRFS_BALANCE_DATA) {
3972 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
3973 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
3976 if (bctl->flags & BTRFS_BALANCE_METADATA) {
3977 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
3978 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
3981 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
3982 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
3983 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
3986 #undef CHECK_APPEND_1ARG
3990 if (size_bp < size_buf)
3991 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
3992 btrfs_info(fs_info, "balance: %s %s",
3993 (bctl->flags & BTRFS_BALANCE_RESUME) ?
3994 "resume" : "start", buf);
4000 * Should be called with balance mutexe held
4002 int btrfs_balance(struct btrfs_fs_info *fs_info,
4003 struct btrfs_balance_control *bctl,
4004 struct btrfs_ioctl_balance_args *bargs)
4006 u64 meta_target, data_target;
4012 bool reducing_redundancy;
4015 if (btrfs_fs_closing(fs_info) ||
4016 atomic_read(&fs_info->balance_pause_req) ||
4017 btrfs_should_cancel_balance(fs_info)) {
4022 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4023 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4027 * In case of mixed groups both data and meta should be picked,
4028 * and identical options should be given for both of them.
4030 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4031 if (mixed && (bctl->flags & allowed)) {
4032 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4033 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4034 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4036 "balance: mixed groups data and metadata options must be the same");
4043 * rw_devices will not change at the moment, device add/delete/replace
4046 num_devices = fs_info->fs_devices->rw_devices;
4049 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4050 * special bit for it, to make it easier to distinguish. Thus we need
4051 * to set it manually, or balance would refuse the profile.
4053 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4054 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4055 if (num_devices >= btrfs_raid_array[i].devs_min)
4056 allowed |= btrfs_raid_array[i].bg_flag;
4058 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4059 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4060 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4066 * Allow to reduce metadata or system integrity only if force set for
4067 * profiles with redundancy (copies, parity)
4070 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4071 if (btrfs_raid_array[i].ncopies >= 2 ||
4072 btrfs_raid_array[i].tolerated_failures >= 1)
4073 allowed |= btrfs_raid_array[i].bg_flag;
4076 seq = read_seqbegin(&fs_info->profiles_lock);
4078 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4079 (fs_info->avail_system_alloc_bits & allowed) &&
4080 !(bctl->sys.target & allowed)) ||
4081 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4082 (fs_info->avail_metadata_alloc_bits & allowed) &&
4083 !(bctl->meta.target & allowed)))
4084 reducing_redundancy = true;
4086 reducing_redundancy = false;
4088 /* if we're not converting, the target field is uninitialized */
4089 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4090 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4091 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4092 bctl->data.target : fs_info->avail_data_alloc_bits;
4093 } while (read_seqretry(&fs_info->profiles_lock, seq));
4095 if (reducing_redundancy) {
4096 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4098 "balance: force reducing metadata redundancy");
4101 "balance: reduces metadata redundancy, use --force if you want this");
4107 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4108 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4110 "balance: metadata profile %s has lower redundancy than data profile %s",
4111 btrfs_bg_type_to_raid_name(meta_target),
4112 btrfs_bg_type_to_raid_name(data_target));
4115 if (fs_info->send_in_progress) {
4116 btrfs_warn_rl(fs_info,
4117 "cannot run balance while send operations are in progress (%d in progress)",
4118 fs_info->send_in_progress);
4123 ret = insert_balance_item(fs_info, bctl);
4124 if (ret && ret != -EEXIST)
4127 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4128 BUG_ON(ret == -EEXIST);
4129 BUG_ON(fs_info->balance_ctl);
4130 spin_lock(&fs_info->balance_lock);
4131 fs_info->balance_ctl = bctl;
4132 spin_unlock(&fs_info->balance_lock);
4134 BUG_ON(ret != -EEXIST);
4135 spin_lock(&fs_info->balance_lock);
4136 update_balance_args(bctl);
4137 spin_unlock(&fs_info->balance_lock);
4140 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4141 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4142 describe_balance_start_or_resume(fs_info);
4143 mutex_unlock(&fs_info->balance_mutex);
4145 ret = __btrfs_balance(fs_info);
4147 mutex_lock(&fs_info->balance_mutex);
4148 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req))
4149 btrfs_info(fs_info, "balance: paused");
4151 * Balance can be canceled by:
4153 * - Regular cancel request
4154 * Then ret == -ECANCELED and balance_cancel_req > 0
4156 * - Fatal signal to "btrfs" process
4157 * Either the signal caught by wait_reserve_ticket() and callers
4158 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4160 * Either way, in this case balance_cancel_req = 0, and
4161 * ret == -EINTR or ret == -ECANCELED.
4163 * So here we only check the return value to catch canceled balance.
4165 else if (ret == -ECANCELED || ret == -EINTR)
4166 btrfs_info(fs_info, "balance: canceled");
4168 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4170 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4173 memset(bargs, 0, sizeof(*bargs));
4174 btrfs_update_ioctl_balance_args(fs_info, bargs);
4177 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
4178 balance_need_close(fs_info)) {
4179 reset_balance_state(fs_info);
4180 btrfs_exclop_finish(fs_info);
4183 wake_up(&fs_info->balance_wait_q);
4187 if (bctl->flags & BTRFS_BALANCE_RESUME)
4188 reset_balance_state(fs_info);
4191 btrfs_exclop_finish(fs_info);
4196 static int balance_kthread(void *data)
4198 struct btrfs_fs_info *fs_info = data;
4201 mutex_lock(&fs_info->balance_mutex);
4202 if (fs_info->balance_ctl)
4203 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4204 mutex_unlock(&fs_info->balance_mutex);
4209 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4211 struct task_struct *tsk;
4213 mutex_lock(&fs_info->balance_mutex);
4214 if (!fs_info->balance_ctl) {
4215 mutex_unlock(&fs_info->balance_mutex);
4218 mutex_unlock(&fs_info->balance_mutex);
4220 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4221 btrfs_info(fs_info, "balance: resume skipped");
4226 * A ro->rw remount sequence should continue with the paused balance
4227 * regardless of who pauses it, system or the user as of now, so set
4230 spin_lock(&fs_info->balance_lock);
4231 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4232 spin_unlock(&fs_info->balance_lock);
4234 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4235 return PTR_ERR_OR_ZERO(tsk);
4238 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4240 struct btrfs_balance_control *bctl;
4241 struct btrfs_balance_item *item;
4242 struct btrfs_disk_balance_args disk_bargs;
4243 struct btrfs_path *path;
4244 struct extent_buffer *leaf;
4245 struct btrfs_key key;
4248 path = btrfs_alloc_path();
4252 key.objectid = BTRFS_BALANCE_OBJECTID;
4253 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4256 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4259 if (ret > 0) { /* ret = -ENOENT; */
4264 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4270 leaf = path->nodes[0];
4271 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4273 bctl->flags = btrfs_balance_flags(leaf, item);
4274 bctl->flags |= BTRFS_BALANCE_RESUME;
4276 btrfs_balance_data(leaf, item, &disk_bargs);
4277 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4278 btrfs_balance_meta(leaf, item, &disk_bargs);
4279 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4280 btrfs_balance_sys(leaf, item, &disk_bargs);
4281 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4284 * This should never happen, as the paused balance state is recovered
4285 * during mount without any chance of other exclusive ops to collide.
4287 * This gives the exclusive op status to balance and keeps in paused
4288 * state until user intervention (cancel or umount). If the ownership
4289 * cannot be assigned, show a message but do not fail. The balance
4290 * is in a paused state and must have fs_info::balance_ctl properly
4293 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE))
4295 "balance: cannot set exclusive op status, resume manually");
4297 mutex_lock(&fs_info->balance_mutex);
4298 BUG_ON(fs_info->balance_ctl);
4299 spin_lock(&fs_info->balance_lock);
4300 fs_info->balance_ctl = bctl;
4301 spin_unlock(&fs_info->balance_lock);
4302 mutex_unlock(&fs_info->balance_mutex);
4304 btrfs_free_path(path);
4308 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4312 mutex_lock(&fs_info->balance_mutex);
4313 if (!fs_info->balance_ctl) {
4314 mutex_unlock(&fs_info->balance_mutex);
4318 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4319 atomic_inc(&fs_info->balance_pause_req);
4320 mutex_unlock(&fs_info->balance_mutex);
4322 wait_event(fs_info->balance_wait_q,
4323 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4325 mutex_lock(&fs_info->balance_mutex);
4326 /* we are good with balance_ctl ripped off from under us */
4327 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4328 atomic_dec(&fs_info->balance_pause_req);
4333 mutex_unlock(&fs_info->balance_mutex);
4337 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4339 mutex_lock(&fs_info->balance_mutex);
4340 if (!fs_info->balance_ctl) {
4341 mutex_unlock(&fs_info->balance_mutex);
4346 * A paused balance with the item stored on disk can be resumed at
4347 * mount time if the mount is read-write. Otherwise it's still paused
4348 * and we must not allow cancelling as it deletes the item.
4350 if (sb_rdonly(fs_info->sb)) {
4351 mutex_unlock(&fs_info->balance_mutex);
4355 atomic_inc(&fs_info->balance_cancel_req);
4357 * if we are running just wait and return, balance item is
4358 * deleted in btrfs_balance in this case
4360 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4361 mutex_unlock(&fs_info->balance_mutex);
4362 wait_event(fs_info->balance_wait_q,
4363 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4364 mutex_lock(&fs_info->balance_mutex);
4366 mutex_unlock(&fs_info->balance_mutex);
4368 * Lock released to allow other waiters to continue, we'll
4369 * reexamine the status again.
4371 mutex_lock(&fs_info->balance_mutex);
4373 if (fs_info->balance_ctl) {
4374 reset_balance_state(fs_info);
4375 btrfs_exclop_finish(fs_info);
4376 btrfs_info(fs_info, "balance: canceled");
4380 BUG_ON(fs_info->balance_ctl ||
4381 test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4382 atomic_dec(&fs_info->balance_cancel_req);
4383 mutex_unlock(&fs_info->balance_mutex);
4387 int btrfs_uuid_scan_kthread(void *data)
4389 struct btrfs_fs_info *fs_info = data;
4390 struct btrfs_root *root = fs_info->tree_root;
4391 struct btrfs_key key;
4392 struct btrfs_path *path = NULL;
4394 struct extent_buffer *eb;
4396 struct btrfs_root_item root_item;
4398 struct btrfs_trans_handle *trans = NULL;
4399 bool closing = false;
4401 path = btrfs_alloc_path();
4408 key.type = BTRFS_ROOT_ITEM_KEY;
4412 if (btrfs_fs_closing(fs_info)) {
4416 ret = btrfs_search_forward(root, &key, path,
4417 BTRFS_OLDEST_GENERATION);
4424 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4425 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4426 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4427 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4430 eb = path->nodes[0];
4431 slot = path->slots[0];
4432 item_size = btrfs_item_size_nr(eb, slot);
4433 if (item_size < sizeof(root_item))
4436 read_extent_buffer(eb, &root_item,
4437 btrfs_item_ptr_offset(eb, slot),
4438 (int)sizeof(root_item));
4439 if (btrfs_root_refs(&root_item) == 0)
4442 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4443 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4447 btrfs_release_path(path);
4449 * 1 - subvol uuid item
4450 * 1 - received_subvol uuid item
4452 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4453 if (IS_ERR(trans)) {
4454 ret = PTR_ERR(trans);
4462 btrfs_release_path(path);
4463 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4464 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4465 BTRFS_UUID_KEY_SUBVOL,
4468 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4474 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4475 ret = btrfs_uuid_tree_add(trans,
4476 root_item.received_uuid,
4477 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4480 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4487 btrfs_release_path(path);
4489 ret = btrfs_end_transaction(trans);
4495 if (key.offset < (u64)-1) {
4497 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4499 key.type = BTRFS_ROOT_ITEM_KEY;
4500 } else if (key.objectid < (u64)-1) {
4502 key.type = BTRFS_ROOT_ITEM_KEY;
4511 btrfs_free_path(path);
4512 if (trans && !IS_ERR(trans))
4513 btrfs_end_transaction(trans);
4515 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4517 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4518 up(&fs_info->uuid_tree_rescan_sem);
4522 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4524 struct btrfs_trans_handle *trans;
4525 struct btrfs_root *tree_root = fs_info->tree_root;
4526 struct btrfs_root *uuid_root;
4527 struct task_struct *task;
4534 trans = btrfs_start_transaction(tree_root, 2);
4536 return PTR_ERR(trans);
4538 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4539 if (IS_ERR(uuid_root)) {
4540 ret = PTR_ERR(uuid_root);
4541 btrfs_abort_transaction(trans, ret);
4542 btrfs_end_transaction(trans);
4546 fs_info->uuid_root = uuid_root;
4548 ret = btrfs_commit_transaction(trans);
4552 down(&fs_info->uuid_tree_rescan_sem);
4553 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4555 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4556 btrfs_warn(fs_info, "failed to start uuid_scan task");
4557 up(&fs_info->uuid_tree_rescan_sem);
4558 return PTR_ERR(task);
4565 * shrinking a device means finding all of the device extents past
4566 * the new size, and then following the back refs to the chunks.
4567 * The chunk relocation code actually frees the device extent
4569 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4571 struct btrfs_fs_info *fs_info = device->fs_info;
4572 struct btrfs_root *root = fs_info->dev_root;
4573 struct btrfs_trans_handle *trans;
4574 struct btrfs_dev_extent *dev_extent = NULL;
4575 struct btrfs_path *path;
4581 bool retried = false;
4582 struct extent_buffer *l;
4583 struct btrfs_key key;
4584 struct btrfs_super_block *super_copy = fs_info->super_copy;
4585 u64 old_total = btrfs_super_total_bytes(super_copy);
4586 u64 old_size = btrfs_device_get_total_bytes(device);
4590 new_size = round_down(new_size, fs_info->sectorsize);
4592 diff = round_down(old_size - new_size, fs_info->sectorsize);
4594 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4597 path = btrfs_alloc_path();
4601 path->reada = READA_BACK;
4603 trans = btrfs_start_transaction(root, 0);
4604 if (IS_ERR(trans)) {
4605 btrfs_free_path(path);
4606 return PTR_ERR(trans);
4609 mutex_lock(&fs_info->chunk_mutex);
4611 btrfs_device_set_total_bytes(device, new_size);
4612 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4613 device->fs_devices->total_rw_bytes -= diff;
4614 atomic64_sub(diff, &fs_info->free_chunk_space);
4618 * Once the device's size has been set to the new size, ensure all
4619 * in-memory chunks are synced to disk so that the loop below sees them
4620 * and relocates them accordingly.
4622 if (contains_pending_extent(device, &start, diff)) {
4623 mutex_unlock(&fs_info->chunk_mutex);
4624 ret = btrfs_commit_transaction(trans);
4628 mutex_unlock(&fs_info->chunk_mutex);
4629 btrfs_end_transaction(trans);
4633 key.objectid = device->devid;
4634 key.offset = (u64)-1;
4635 key.type = BTRFS_DEV_EXTENT_KEY;
4638 mutex_lock(&fs_info->delete_unused_bgs_mutex);
4639 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4641 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4645 ret = btrfs_previous_item(root, path, 0, key.type);
4647 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4652 btrfs_release_path(path);
4657 slot = path->slots[0];
4658 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4660 if (key.objectid != device->devid) {
4661 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4662 btrfs_release_path(path);
4666 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4667 length = btrfs_dev_extent_length(l, dev_extent);
4669 if (key.offset + length <= new_size) {
4670 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4671 btrfs_release_path(path);
4675 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4676 btrfs_release_path(path);
4679 * We may be relocating the only data chunk we have,
4680 * which could potentially end up with losing data's
4681 * raid profile, so lets allocate an empty one in
4684 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4686 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4690 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4691 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
4692 if (ret == -ENOSPC) {
4695 if (ret == -ETXTBSY) {
4697 "could not shrink block group %llu due to active swapfile",
4702 } while (key.offset-- > 0);
4704 if (failed && !retried) {
4708 } else if (failed && retried) {
4713 /* Shrinking succeeded, else we would be at "done". */
4714 trans = btrfs_start_transaction(root, 0);
4715 if (IS_ERR(trans)) {
4716 ret = PTR_ERR(trans);
4720 mutex_lock(&fs_info->chunk_mutex);
4721 /* Clear all state bits beyond the shrunk device size */
4722 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
4725 btrfs_device_set_disk_total_bytes(device, new_size);
4726 if (list_empty(&device->post_commit_list))
4727 list_add_tail(&device->post_commit_list,
4728 &trans->transaction->dev_update_list);
4730 WARN_ON(diff > old_total);
4731 btrfs_set_super_total_bytes(super_copy,
4732 round_down(old_total - diff, fs_info->sectorsize));
4733 mutex_unlock(&fs_info->chunk_mutex);
4735 /* Now btrfs_update_device() will change the on-disk size. */
4736 ret = btrfs_update_device(trans, device);
4738 btrfs_abort_transaction(trans, ret);
4739 btrfs_end_transaction(trans);
4741 ret = btrfs_commit_transaction(trans);
4744 btrfs_free_path(path);
4746 mutex_lock(&fs_info->chunk_mutex);
4747 btrfs_device_set_total_bytes(device, old_size);
4748 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
4749 device->fs_devices->total_rw_bytes += diff;
4750 atomic64_add(diff, &fs_info->free_chunk_space);
4751 mutex_unlock(&fs_info->chunk_mutex);
4756 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
4757 struct btrfs_key *key,
4758 struct btrfs_chunk *chunk, int item_size)
4760 struct btrfs_super_block *super_copy = fs_info->super_copy;
4761 struct btrfs_disk_key disk_key;
4765 mutex_lock(&fs_info->chunk_mutex);
4766 array_size = btrfs_super_sys_array_size(super_copy);
4767 if (array_size + item_size + sizeof(disk_key)
4768 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4769 mutex_unlock(&fs_info->chunk_mutex);
4773 ptr = super_copy->sys_chunk_array + array_size;
4774 btrfs_cpu_key_to_disk(&disk_key, key);
4775 memcpy(ptr, &disk_key, sizeof(disk_key));
4776 ptr += sizeof(disk_key);
4777 memcpy(ptr, chunk, item_size);
4778 item_size += sizeof(disk_key);
4779 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4780 mutex_unlock(&fs_info->chunk_mutex);
4786 * sort the devices in descending order by max_avail, total_avail
4788 static int btrfs_cmp_device_info(const void *a, const void *b)
4790 const struct btrfs_device_info *di_a = a;
4791 const struct btrfs_device_info *di_b = b;
4793 if (di_a->max_avail > di_b->max_avail)
4795 if (di_a->max_avail < di_b->max_avail)
4797 if (di_a->total_avail > di_b->total_avail)
4799 if (di_a->total_avail < di_b->total_avail)
4804 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4806 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4809 btrfs_set_fs_incompat(info, RAID56);
4812 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
4814 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
4817 btrfs_set_fs_incompat(info, RAID1C34);
4821 * Structure used internally for __btrfs_alloc_chunk() function.
4822 * Wraps needed parameters.
4824 struct alloc_chunk_ctl {
4827 /* Total number of stripes to allocate */
4829 /* sub_stripes info for map */
4831 /* Stripes per device */
4833 /* Maximum number of devices to use */
4835 /* Minimum number of devices to use */
4837 /* ndevs has to be a multiple of this */
4839 /* Number of copies */
4841 /* Number of stripes worth of bytes to store parity information */
4843 u64 max_stripe_size;
4851 static void init_alloc_chunk_ctl_policy_regular(
4852 struct btrfs_fs_devices *fs_devices,
4853 struct alloc_chunk_ctl *ctl)
4855 u64 type = ctl->type;
4857 if (type & BTRFS_BLOCK_GROUP_DATA) {
4858 ctl->max_stripe_size = SZ_1G;
4859 ctl->max_chunk_size = BTRFS_MAX_DATA_CHUNK_SIZE;
4860 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4861 /* For larger filesystems, use larger metadata chunks */
4862 if (fs_devices->total_rw_bytes > 50ULL * SZ_1G)
4863 ctl->max_stripe_size = SZ_1G;
4865 ctl->max_stripe_size = SZ_256M;
4866 ctl->max_chunk_size = ctl->max_stripe_size;
4867 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4868 ctl->max_stripe_size = SZ_32M;
4869 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
4870 ctl->devs_max = min_t(int, ctl->devs_max,
4871 BTRFS_MAX_DEVS_SYS_CHUNK);
4876 /* We don't want a chunk larger than 10% of writable space */
4877 ctl->max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4878 ctl->max_chunk_size);
4879 ctl->dev_extent_min = BTRFS_STRIPE_LEN * ctl->dev_stripes;
4882 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
4883 struct alloc_chunk_ctl *ctl)
4885 int index = btrfs_bg_flags_to_raid_index(ctl->type);
4887 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
4888 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
4889 ctl->devs_max = btrfs_raid_array[index].devs_max;
4891 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
4892 ctl->devs_min = btrfs_raid_array[index].devs_min;
4893 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
4894 ctl->ncopies = btrfs_raid_array[index].ncopies;
4895 ctl->nparity = btrfs_raid_array[index].nparity;
4898 switch (fs_devices->chunk_alloc_policy) {
4899 case BTRFS_CHUNK_ALLOC_REGULAR:
4900 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
4907 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
4908 struct alloc_chunk_ctl *ctl,
4909 struct btrfs_device_info *devices_info)
4911 struct btrfs_fs_info *info = fs_devices->fs_info;
4912 struct btrfs_device *device;
4914 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
4921 * in the first pass through the devices list, we gather information
4922 * about the available holes on each device.
4924 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
4925 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4927 "BTRFS: read-only device in alloc_list\n");
4931 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
4932 &device->dev_state) ||
4933 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4936 if (device->total_bytes > device->bytes_used)
4937 total_avail = device->total_bytes - device->bytes_used;
4941 /* If there is no space on this device, skip it. */
4942 if (total_avail < ctl->dev_extent_min)
4945 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
4947 if (ret && ret != -ENOSPC)
4951 max_avail = dev_extent_want;
4953 if (max_avail < ctl->dev_extent_min) {
4954 if (btrfs_test_opt(info, ENOSPC_DEBUG))
4956 "%s: devid %llu has no free space, have=%llu want=%llu",
4957 __func__, device->devid, max_avail,
4958 ctl->dev_extent_min);
4962 if (ndevs == fs_devices->rw_devices) {
4963 WARN(1, "%s: found more than %llu devices\n",
4964 __func__, fs_devices->rw_devices);
4967 devices_info[ndevs].dev_offset = dev_offset;
4968 devices_info[ndevs].max_avail = max_avail;
4969 devices_info[ndevs].total_avail = total_avail;
4970 devices_info[ndevs].dev = device;
4976 * now sort the devices by hole size / available space
4978 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4979 btrfs_cmp_device_info, NULL);
4984 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
4985 struct btrfs_device_info *devices_info)
4987 /* Number of stripes that count for block group size */
4991 * The primary goal is to maximize the number of stripes, so use as
4992 * many devices as possible, even if the stripes are not maximum sized.
4994 * The DUP profile stores more than one stripe per device, the
4995 * max_avail is the total size so we have to adjust.
4997 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
4999 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5001 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5002 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5005 * Use the number of data stripes to figure out how big this chunk is
5006 * really going to be in terms of logical address space, and compare
5007 * that answer with the max chunk size. If it's higher, we try to
5008 * reduce stripe_size.
5010 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5012 * Reduce stripe_size, round it up to a 16MB boundary again and
5013 * then use it, unless it ends up being even bigger than the
5014 * previous value we had already.
5016 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5017 data_stripes), SZ_16M),
5021 /* Align to BTRFS_STRIPE_LEN */
5022 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5023 ctl->chunk_size = ctl->stripe_size * data_stripes;
5028 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5029 struct alloc_chunk_ctl *ctl,
5030 struct btrfs_device_info *devices_info)
5032 struct btrfs_fs_info *info = fs_devices->fs_info;
5035 * Round down to number of usable stripes, devs_increment can be any
5036 * number so we can't use round_down() that requires power of 2, while
5037 * rounddown is safe.
5039 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5041 if (ctl->ndevs < ctl->devs_min) {
5042 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5044 "%s: not enough devices with free space: have=%d minimum required=%d",
5045 __func__, ctl->ndevs, ctl->devs_min);
5050 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5052 switch (fs_devices->chunk_alloc_policy) {
5053 case BTRFS_CHUNK_ALLOC_REGULAR:
5054 return decide_stripe_size_regular(ctl, devices_info);
5060 static int create_chunk(struct btrfs_trans_handle *trans,
5061 struct alloc_chunk_ctl *ctl,
5062 struct btrfs_device_info *devices_info)
5064 struct btrfs_fs_info *info = trans->fs_info;
5065 struct map_lookup *map = NULL;
5066 struct extent_map_tree *em_tree;
5067 struct extent_map *em;
5068 u64 start = ctl->start;
5069 u64 type = ctl->type;
5074 map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
5077 map->num_stripes = ctl->num_stripes;
5079 for (i = 0; i < ctl->ndevs; ++i) {
5080 for (j = 0; j < ctl->dev_stripes; ++j) {
5081 int s = i * ctl->dev_stripes + j;
5082 map->stripes[s].dev = devices_info[i].dev;
5083 map->stripes[s].physical = devices_info[i].dev_offset +
5084 j * ctl->stripe_size;
5087 map->stripe_len = BTRFS_STRIPE_LEN;
5088 map->io_align = BTRFS_STRIPE_LEN;
5089 map->io_width = BTRFS_STRIPE_LEN;
5091 map->sub_stripes = ctl->sub_stripes;
5093 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5095 em = alloc_extent_map();
5100 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5101 em->map_lookup = map;
5103 em->len = ctl->chunk_size;
5104 em->block_start = 0;
5105 em->block_len = em->len;
5106 em->orig_block_len = ctl->stripe_size;
5108 em_tree = &info->mapping_tree;
5109 write_lock(&em_tree->lock);
5110 ret = add_extent_mapping(em_tree, em, 0);
5112 write_unlock(&em_tree->lock);
5113 free_extent_map(em);
5116 write_unlock(&em_tree->lock);
5118 ret = btrfs_make_block_group(trans, 0, type, start, ctl->chunk_size);
5120 goto error_del_extent;
5122 for (i = 0; i < map->num_stripes; i++) {
5123 struct btrfs_device *dev = map->stripes[i].dev;
5125 btrfs_device_set_bytes_used(dev,
5126 dev->bytes_used + ctl->stripe_size);
5127 if (list_empty(&dev->post_commit_list))
5128 list_add_tail(&dev->post_commit_list,
5129 &trans->transaction->dev_update_list);
5132 atomic64_sub(ctl->stripe_size * map->num_stripes,
5133 &info->free_chunk_space);
5135 free_extent_map(em);
5136 check_raid56_incompat_flag(info, type);
5137 check_raid1c34_incompat_flag(info, type);
5142 write_lock(&em_tree->lock);
5143 remove_extent_mapping(em_tree, em);
5144 write_unlock(&em_tree->lock);
5146 /* One for our allocation */
5147 free_extent_map(em);
5148 /* One for the tree reference */
5149 free_extent_map(em);
5154 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, u64 type)
5156 struct btrfs_fs_info *info = trans->fs_info;
5157 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5158 struct btrfs_device_info *devices_info = NULL;
5159 struct alloc_chunk_ctl ctl;
5162 lockdep_assert_held(&info->chunk_mutex);
5164 if (!alloc_profile_is_valid(type, 0)) {
5169 if (list_empty(&fs_devices->alloc_list)) {
5170 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5171 btrfs_debug(info, "%s: no writable device", __func__);
5175 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5176 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5181 ctl.start = find_next_chunk(info);
5183 init_alloc_chunk_ctl(fs_devices, &ctl);
5185 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5190 ret = gather_device_info(fs_devices, &ctl, devices_info);
5194 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5198 ret = create_chunk(trans, &ctl, devices_info);
5201 kfree(devices_info);
5206 * Chunk allocation falls into two parts. The first part does work
5207 * that makes the new allocated chunk usable, but does not do any operation
5208 * that modifies the chunk tree. The second part does the work that
5209 * requires modifying the chunk tree. This division is important for the
5210 * bootstrap process of adding storage to a seed btrfs.
5212 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
5213 u64 chunk_offset, u64 chunk_size)
5215 struct btrfs_fs_info *fs_info = trans->fs_info;
5216 struct btrfs_root *extent_root = fs_info->extent_root;
5217 struct btrfs_root *chunk_root = fs_info->chunk_root;
5218 struct btrfs_key key;
5219 struct btrfs_device *device;
5220 struct btrfs_chunk *chunk;
5221 struct btrfs_stripe *stripe;
5222 struct extent_map *em;
5223 struct map_lookup *map;
5230 em = btrfs_get_chunk_map(fs_info, chunk_offset, chunk_size);
5234 map = em->map_lookup;
5235 item_size = btrfs_chunk_item_size(map->num_stripes);
5236 stripe_size = em->orig_block_len;
5238 chunk = kzalloc(item_size, GFP_NOFS);
5245 * Take the device list mutex to prevent races with the final phase of
5246 * a device replace operation that replaces the device object associated
5247 * with the map's stripes, because the device object's id can change
5248 * at any time during that final phase of the device replace operation
5249 * (dev-replace.c:btrfs_dev_replace_finishing()).
5251 mutex_lock(&fs_info->fs_devices->device_list_mutex);
5252 for (i = 0; i < map->num_stripes; i++) {
5253 device = map->stripes[i].dev;
5254 dev_offset = map->stripes[i].physical;
5256 ret = btrfs_update_device(trans, device);
5259 ret = btrfs_alloc_dev_extent(trans, device, chunk_offset,
5260 dev_offset, stripe_size);
5265 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5269 stripe = &chunk->stripe;
5270 for (i = 0; i < map->num_stripes; i++) {
5271 device = map->stripes[i].dev;
5272 dev_offset = map->stripes[i].physical;
5274 btrfs_set_stack_stripe_devid(stripe, device->devid);
5275 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5276 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5279 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
5281 btrfs_set_stack_chunk_length(chunk, chunk_size);
5282 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
5283 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
5284 btrfs_set_stack_chunk_type(chunk, map->type);
5285 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5286 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
5287 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
5288 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5289 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5291 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5292 key.type = BTRFS_CHUNK_ITEM_KEY;
5293 key.offset = chunk_offset;
5295 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5296 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5298 * TODO: Cleanup of inserted chunk root in case of
5301 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5306 free_extent_map(em);
5310 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5312 struct btrfs_fs_info *fs_info = trans->fs_info;
5316 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5317 ret = btrfs_alloc_chunk(trans, alloc_profile);
5321 alloc_profile = btrfs_system_alloc_profile(fs_info);
5322 ret = btrfs_alloc_chunk(trans, alloc_profile);
5326 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5328 const int index = btrfs_bg_flags_to_raid_index(map->type);
5330 return btrfs_raid_array[index].tolerated_failures;
5333 int btrfs_chunk_readonly(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5335 struct extent_map *em;
5336 struct map_lookup *map;
5341 em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5345 map = em->map_lookup;
5346 for (i = 0; i < map->num_stripes; i++) {
5347 if (test_bit(BTRFS_DEV_STATE_MISSING,
5348 &map->stripes[i].dev->dev_state)) {
5352 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5353 &map->stripes[i].dev->dev_state)) {
5360 * If the number of missing devices is larger than max errors,
5361 * we can not write the data into that chunk successfully, so
5364 if (miss_ndevs > btrfs_chunk_max_errors(map))
5367 free_extent_map(em);
5371 void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5373 struct extent_map *em;
5376 write_lock(&tree->lock);
5377 em = lookup_extent_mapping(tree, 0, (u64)-1);
5379 remove_extent_mapping(tree, em);
5380 write_unlock(&tree->lock);
5384 free_extent_map(em);
5385 /* once for the tree */
5386 free_extent_map(em);
5390 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5392 struct extent_map *em;
5393 struct map_lookup *map;
5396 em = btrfs_get_chunk_map(fs_info, logical, len);
5399 * We could return errors for these cases, but that could get
5400 * ugly and we'd probably do the same thing which is just not do
5401 * anything else and exit, so return 1 so the callers don't try
5402 * to use other copies.
5406 map = em->map_lookup;
5407 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1_MASK))
5408 ret = map->num_stripes;
5409 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5410 ret = map->sub_stripes;
5411 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5413 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5415 * There could be two corrupted data stripes, we need
5416 * to loop retry in order to rebuild the correct data.
5418 * Fail a stripe at a time on every retry except the
5419 * stripe under reconstruction.
5421 ret = map->num_stripes;
5424 free_extent_map(em);
5426 down_read(&fs_info->dev_replace.rwsem);
5427 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace) &&
5428 fs_info->dev_replace.tgtdev)
5430 up_read(&fs_info->dev_replace.rwsem);
5435 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5438 struct extent_map *em;
5439 struct map_lookup *map;
5440 unsigned long len = fs_info->sectorsize;
5442 em = btrfs_get_chunk_map(fs_info, logical, len);
5444 if (!WARN_ON(IS_ERR(em))) {
5445 map = em->map_lookup;
5446 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5447 len = map->stripe_len * nr_data_stripes(map);
5448 free_extent_map(em);
5453 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5455 struct extent_map *em;
5456 struct map_lookup *map;
5459 em = btrfs_get_chunk_map(fs_info, logical, len);
5461 if(!WARN_ON(IS_ERR(em))) {
5462 map = em->map_lookup;
5463 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5465 free_extent_map(em);
5470 static int find_live_mirror(struct btrfs_fs_info *fs_info,
5471 struct map_lookup *map, int first,
5472 int dev_replace_is_ongoing)
5476 int preferred_mirror;
5478 struct btrfs_device *srcdev;
5481 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5483 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5484 num_stripes = map->sub_stripes;
5486 num_stripes = map->num_stripes;
5488 preferred_mirror = first + current->pid % num_stripes;
5490 if (dev_replace_is_ongoing &&
5491 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5492 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5493 srcdev = fs_info->dev_replace.srcdev;
5498 * try to avoid the drive that is the source drive for a
5499 * dev-replace procedure, only choose it if no other non-missing
5500 * mirror is available
5502 for (tolerance = 0; tolerance < 2; tolerance++) {
5503 if (map->stripes[preferred_mirror].dev->bdev &&
5504 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5505 return preferred_mirror;
5506 for (i = first; i < first + num_stripes; i++) {
5507 if (map->stripes[i].dev->bdev &&
5508 (tolerance || map->stripes[i].dev != srcdev))
5513 /* we couldn't find one that doesn't fail. Just return something
5514 * and the io error handling code will clean up eventually
5516 return preferred_mirror;
5519 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5520 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5527 for (i = 0; i < num_stripes - 1; i++) {
5528 /* Swap if parity is on a smaller index */
5529 if (bbio->raid_map[i] > bbio->raid_map[i + 1]) {
5530 swap(bbio->stripes[i], bbio->stripes[i + 1]);
5531 swap(bbio->raid_map[i], bbio->raid_map[i + 1]);
5538 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5540 struct btrfs_bio *bbio = kzalloc(
5541 /* the size of the btrfs_bio */
5542 sizeof(struct btrfs_bio) +
5543 /* plus the variable array for the stripes */
5544 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5545 /* plus the variable array for the tgt dev */
5546 sizeof(int) * (real_stripes) +
5548 * plus the raid_map, which includes both the tgt dev
5551 sizeof(u64) * (total_stripes),
5552 GFP_NOFS|__GFP_NOFAIL);
5554 atomic_set(&bbio->error, 0);
5555 refcount_set(&bbio->refs, 1);
5557 bbio->tgtdev_map = (int *)(bbio->stripes + total_stripes);
5558 bbio->raid_map = (u64 *)(bbio->tgtdev_map + real_stripes);
5563 void btrfs_get_bbio(struct btrfs_bio *bbio)
5565 WARN_ON(!refcount_read(&bbio->refs));
5566 refcount_inc(&bbio->refs);
5569 void btrfs_put_bbio(struct btrfs_bio *bbio)
5573 if (refcount_dec_and_test(&bbio->refs))
5577 /* can REQ_OP_DISCARD be sent with other REQ like REQ_OP_WRITE? */
5579 * Please note that, discard won't be sent to target device of device
5582 static int __btrfs_map_block_for_discard(struct btrfs_fs_info *fs_info,
5583 u64 logical, u64 *length_ret,
5584 struct btrfs_bio **bbio_ret)
5586 struct extent_map *em;
5587 struct map_lookup *map;
5588 struct btrfs_bio *bbio;
5589 u64 length = *length_ret;
5593 u64 stripe_end_offset;
5600 u32 sub_stripes = 0;
5601 u64 stripes_per_dev = 0;
5602 u32 remaining_stripes = 0;
5603 u32 last_stripe = 0;
5607 /* discard always return a bbio */
5610 em = btrfs_get_chunk_map(fs_info, logical, length);
5614 map = em->map_lookup;
5615 /* we don't discard raid56 yet */
5616 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5621 offset = logical - em->start;
5622 length = min_t(u64, em->start + em->len - logical, length);
5623 *length_ret = length;
5625 stripe_len = map->stripe_len;
5627 * stripe_nr counts the total number of stripes we have to stride
5628 * to get to this block
5630 stripe_nr = div64_u64(offset, stripe_len);
5632 /* stripe_offset is the offset of this block in its stripe */
5633 stripe_offset = offset - stripe_nr * stripe_len;
5635 stripe_nr_end = round_up(offset + length, map->stripe_len);
5636 stripe_nr_end = div64_u64(stripe_nr_end, map->stripe_len);
5637 stripe_cnt = stripe_nr_end - stripe_nr;
5638 stripe_end_offset = stripe_nr_end * map->stripe_len -
5641 * after this, stripe_nr is the number of stripes on this
5642 * device we have to walk to find the data, and stripe_index is
5643 * the number of our device in the stripe array
5647 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5648 BTRFS_BLOCK_GROUP_RAID10)) {
5649 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5652 sub_stripes = map->sub_stripes;
5654 factor = map->num_stripes / sub_stripes;
5655 num_stripes = min_t(u64, map->num_stripes,
5656 sub_stripes * stripe_cnt);
5657 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5658 stripe_index *= sub_stripes;
5659 stripes_per_dev = div_u64_rem(stripe_cnt, factor,
5660 &remaining_stripes);
5661 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5662 last_stripe *= sub_stripes;
5663 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
5664 BTRFS_BLOCK_GROUP_DUP)) {
5665 num_stripes = map->num_stripes;
5667 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5671 bbio = alloc_btrfs_bio(num_stripes, 0);
5677 for (i = 0; i < num_stripes; i++) {
5678 bbio->stripes[i].physical =
5679 map->stripes[stripe_index].physical +
5680 stripe_offset + stripe_nr * map->stripe_len;
5681 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5683 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5684 BTRFS_BLOCK_GROUP_RAID10)) {
5685 bbio->stripes[i].length = stripes_per_dev *
5688 if (i / sub_stripes < remaining_stripes)
5689 bbio->stripes[i].length +=
5693 * Special for the first stripe and
5696 * |-------|...|-------|
5700 if (i < sub_stripes)
5701 bbio->stripes[i].length -=
5704 if (stripe_index >= last_stripe &&
5705 stripe_index <= (last_stripe +
5707 bbio->stripes[i].length -=
5710 if (i == sub_stripes - 1)
5713 bbio->stripes[i].length = length;
5717 if (stripe_index == map->num_stripes) {
5724 bbio->map_type = map->type;
5725 bbio->num_stripes = num_stripes;
5727 free_extent_map(em);
5732 * In dev-replace case, for repair case (that's the only case where the mirror
5733 * is selected explicitly when calling btrfs_map_block), blocks left of the
5734 * left cursor can also be read from the target drive.
5736 * For REQ_GET_READ_MIRRORS, the target drive is added as the last one to the
5738 * For READ, it also needs to be supported using the same mirror number.
5740 * If the requested block is not left of the left cursor, EIO is returned. This
5741 * can happen because btrfs_num_copies() returns one more in the dev-replace
5744 static int get_extra_mirror_from_replace(struct btrfs_fs_info *fs_info,
5745 u64 logical, u64 length,
5746 u64 srcdev_devid, int *mirror_num,
5749 struct btrfs_bio *bbio = NULL;
5751 int index_srcdev = 0;
5753 u64 physical_of_found = 0;
5757 ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS,
5758 logical, &length, &bbio, 0, 0);
5760 ASSERT(bbio == NULL);
5764 num_stripes = bbio->num_stripes;
5765 if (*mirror_num > num_stripes) {
5767 * BTRFS_MAP_GET_READ_MIRRORS does not contain this mirror,
5768 * that means that the requested area is not left of the left
5771 btrfs_put_bbio(bbio);
5776 * process the rest of the function using the mirror_num of the source
5777 * drive. Therefore look it up first. At the end, patch the device
5778 * pointer to the one of the target drive.
5780 for (i = 0; i < num_stripes; i++) {
5781 if (bbio->stripes[i].dev->devid != srcdev_devid)
5785 * In case of DUP, in order to keep it simple, only add the
5786 * mirror with the lowest physical address
5789 physical_of_found <= bbio->stripes[i].physical)
5794 physical_of_found = bbio->stripes[i].physical;
5797 btrfs_put_bbio(bbio);
5803 *mirror_num = index_srcdev + 1;
5804 *physical = physical_of_found;
5808 static void handle_ops_on_dev_replace(enum btrfs_map_op op,
5809 struct btrfs_bio **bbio_ret,
5810 struct btrfs_dev_replace *dev_replace,
5811 int *num_stripes_ret, int *max_errors_ret)
5813 struct btrfs_bio *bbio = *bbio_ret;
5814 u64 srcdev_devid = dev_replace->srcdev->devid;
5815 int tgtdev_indexes = 0;
5816 int num_stripes = *num_stripes_ret;
5817 int max_errors = *max_errors_ret;
5820 if (op == BTRFS_MAP_WRITE) {
5821 int index_where_to_add;
5824 * duplicate the write operations while the dev replace
5825 * procedure is running. Since the copying of the old disk to
5826 * the new disk takes place at run time while the filesystem is
5827 * mounted writable, the regular write operations to the old
5828 * disk have to be duplicated to go to the new disk as well.
5830 * Note that device->missing is handled by the caller, and that
5831 * the write to the old disk is already set up in the stripes
5834 index_where_to_add = num_stripes;
5835 for (i = 0; i < num_stripes; i++) {
5836 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5837 /* write to new disk, too */
5838 struct btrfs_bio_stripe *new =
5839 bbio->stripes + index_where_to_add;
5840 struct btrfs_bio_stripe *old =
5843 new->physical = old->physical;
5844 new->length = old->length;
5845 new->dev = dev_replace->tgtdev;
5846 bbio->tgtdev_map[i] = index_where_to_add;
5847 index_where_to_add++;
5852 num_stripes = index_where_to_add;
5853 } else if (op == BTRFS_MAP_GET_READ_MIRRORS) {
5854 int index_srcdev = 0;
5856 u64 physical_of_found = 0;
5859 * During the dev-replace procedure, the target drive can also
5860 * be used to read data in case it is needed to repair a corrupt
5861 * block elsewhere. This is possible if the requested area is
5862 * left of the left cursor. In this area, the target drive is a
5863 * full copy of the source drive.
5865 for (i = 0; i < num_stripes; i++) {
5866 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5868 * In case of DUP, in order to keep it simple,
5869 * only add the mirror with the lowest physical
5873 physical_of_found <=
5874 bbio->stripes[i].physical)
5878 physical_of_found = bbio->stripes[i].physical;
5882 struct btrfs_bio_stripe *tgtdev_stripe =
5883 bbio->stripes + num_stripes;
5885 tgtdev_stripe->physical = physical_of_found;
5886 tgtdev_stripe->length =
5887 bbio->stripes[index_srcdev].length;
5888 tgtdev_stripe->dev = dev_replace->tgtdev;
5889 bbio->tgtdev_map[index_srcdev] = num_stripes;
5896 *num_stripes_ret = num_stripes;
5897 *max_errors_ret = max_errors;
5898 bbio->num_tgtdevs = tgtdev_indexes;
5902 static bool need_full_stripe(enum btrfs_map_op op)
5904 return (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS);
5908 * btrfs_get_io_geometry - calculates the geomery of a particular (address, len)
5909 * tuple. This information is used to calculate how big a
5910 * particular bio can get before it straddles a stripe.
5912 * @fs_info - the filesystem
5913 * @logical - address that we want to figure out the geometry of
5914 * @len - the length of IO we are going to perform, starting at @logical
5915 * @op - type of operation - write or read
5916 * @io_geom - pointer used to return values
5918 * Returns < 0 in case a chunk for the given logical address cannot be found,
5919 * usually shouldn't happen unless @logical is corrupted, 0 otherwise.
5921 int btrfs_get_io_geometry(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
5922 u64 logical, u64 len, struct btrfs_io_geometry *io_geom)
5924 struct extent_map *em;
5925 struct map_lookup *map;
5930 u64 raid56_full_stripe_start = (u64)-1;
5934 ASSERT(op != BTRFS_MAP_DISCARD);
5936 em = btrfs_get_chunk_map(fs_info, logical, len);
5940 map = em->map_lookup;
5941 /* Offset of this logical address in the chunk */
5942 offset = logical - em->start;
5943 /* Len of a stripe in a chunk */
5944 stripe_len = map->stripe_len;
5945 /* Stripe wher this block falls in */
5946 stripe_nr = div64_u64(offset, stripe_len);
5947 /* Offset of stripe in the chunk */
5948 stripe_offset = stripe_nr * stripe_len;
5949 if (offset < stripe_offset) {
5951 "stripe math has gone wrong, stripe_offset=%llu offset=%llu start=%llu logical=%llu stripe_len=%llu",
5952 stripe_offset, offset, em->start, logical, stripe_len);
5957 /* stripe_offset is the offset of this block in its stripe */
5958 stripe_offset = offset - stripe_offset;
5959 data_stripes = nr_data_stripes(map);
5961 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5962 u64 max_len = stripe_len - stripe_offset;
5965 * In case of raid56, we need to know the stripe aligned start
5967 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5968 unsigned long full_stripe_len = stripe_len * data_stripes;
5969 raid56_full_stripe_start = offset;
5972 * Allow a write of a full stripe, but make sure we
5973 * don't allow straddling of stripes
5975 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5977 raid56_full_stripe_start *= full_stripe_len;
5980 * For writes to RAID[56], allow a full stripeset across
5981 * all disks. For other RAID types and for RAID[56]
5982 * reads, just allow a single stripe (on a single disk).
5984 if (op == BTRFS_MAP_WRITE) {
5985 max_len = stripe_len * data_stripes -
5986 (offset - raid56_full_stripe_start);
5989 len = min_t(u64, em->len - offset, max_len);
5991 len = em->len - offset;
5995 io_geom->offset = offset;
5996 io_geom->stripe_len = stripe_len;
5997 io_geom->stripe_nr = stripe_nr;
5998 io_geom->stripe_offset = stripe_offset;
5999 io_geom->raid56_stripe_offset = raid56_full_stripe_start;
6003 free_extent_map(em);
6007 static int __btrfs_map_block(struct btrfs_fs_info *fs_info,
6008 enum btrfs_map_op op,
6009 u64 logical, u64 *length,
6010 struct btrfs_bio **bbio_ret,
6011 int mirror_num, int need_raid_map)
6013 struct extent_map *em;
6014 struct map_lookup *map;
6024 int tgtdev_indexes = 0;
6025 struct btrfs_bio *bbio = NULL;
6026 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6027 int dev_replace_is_ongoing = 0;
6028 int num_alloc_stripes;
6029 int patch_the_first_stripe_for_dev_replace = 0;
6030 u64 physical_to_patch_in_first_stripe = 0;
6031 u64 raid56_full_stripe_start = (u64)-1;
6032 struct btrfs_io_geometry geom;
6035 ASSERT(op != BTRFS_MAP_DISCARD);
6037 ret = btrfs_get_io_geometry(fs_info, op, logical, *length, &geom);
6041 em = btrfs_get_chunk_map(fs_info, logical, *length);
6042 ASSERT(!IS_ERR(em));
6043 map = em->map_lookup;
6046 stripe_len = geom.stripe_len;
6047 stripe_nr = geom.stripe_nr;
6048 stripe_offset = geom.stripe_offset;
6049 raid56_full_stripe_start = geom.raid56_stripe_offset;
6050 data_stripes = nr_data_stripes(map);
6052 down_read(&dev_replace->rwsem);
6053 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6055 * Hold the semaphore for read during the whole operation, write is
6056 * requested at commit time but must wait.
6058 if (!dev_replace_is_ongoing)
6059 up_read(&dev_replace->rwsem);
6061 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
6062 !need_full_stripe(op) && dev_replace->tgtdev != NULL) {
6063 ret = get_extra_mirror_from_replace(fs_info, logical, *length,
6064 dev_replace->srcdev->devid,
6066 &physical_to_patch_in_first_stripe);
6070 patch_the_first_stripe_for_dev_replace = 1;
6071 } else if (mirror_num > map->num_stripes) {
6077 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6078 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6080 if (!need_full_stripe(op))
6082 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6083 if (need_full_stripe(op))
6084 num_stripes = map->num_stripes;
6085 else if (mirror_num)
6086 stripe_index = mirror_num - 1;
6088 stripe_index = find_live_mirror(fs_info, map, 0,
6089 dev_replace_is_ongoing);
6090 mirror_num = stripe_index + 1;
6093 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6094 if (need_full_stripe(op)) {
6095 num_stripes = map->num_stripes;
6096 } else if (mirror_num) {
6097 stripe_index = mirror_num - 1;
6102 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6103 u32 factor = map->num_stripes / map->sub_stripes;
6105 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
6106 stripe_index *= map->sub_stripes;
6108 if (need_full_stripe(op))
6109 num_stripes = map->sub_stripes;
6110 else if (mirror_num)
6111 stripe_index += mirror_num - 1;
6113 int old_stripe_index = stripe_index;
6114 stripe_index = find_live_mirror(fs_info, map,
6116 dev_replace_is_ongoing);
6117 mirror_num = stripe_index - old_stripe_index + 1;
6120 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6121 if (need_raid_map && (need_full_stripe(op) || mirror_num > 1)) {
6122 /* push stripe_nr back to the start of the full stripe */
6123 stripe_nr = div64_u64(raid56_full_stripe_start,
6124 stripe_len * data_stripes);
6126 /* RAID[56] write or recovery. Return all stripes */
6127 num_stripes = map->num_stripes;
6128 max_errors = nr_parity_stripes(map);
6130 *length = map->stripe_len;
6135 * Mirror #0 or #1 means the original data block.
6136 * Mirror #2 is RAID5 parity block.
6137 * Mirror #3 is RAID6 Q block.
6139 stripe_nr = div_u64_rem(stripe_nr,
6140 data_stripes, &stripe_index);
6142 stripe_index = data_stripes + mirror_num - 2;
6144 /* We distribute the parity blocks across stripes */
6145 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
6147 if (!need_full_stripe(op) && mirror_num <= 1)
6152 * after this, stripe_nr is the number of stripes on this
6153 * device we have to walk to find the data, and stripe_index is
6154 * the number of our device in the stripe array
6156 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
6158 mirror_num = stripe_index + 1;
6160 if (stripe_index >= map->num_stripes) {
6162 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6163 stripe_index, map->num_stripes);
6168 num_alloc_stripes = num_stripes;
6169 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) {
6170 if (op == BTRFS_MAP_WRITE)
6171 num_alloc_stripes <<= 1;
6172 if (op == BTRFS_MAP_GET_READ_MIRRORS)
6173 num_alloc_stripes++;
6174 tgtdev_indexes = num_stripes;
6177 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
6183 for (i = 0; i < num_stripes; i++) {
6184 bbio->stripes[i].physical = map->stripes[stripe_index].physical +
6185 stripe_offset + stripe_nr * map->stripe_len;
6186 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
6190 /* build raid_map */
6191 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6192 (need_full_stripe(op) || mirror_num > 1)) {
6196 /* Work out the disk rotation on this stripe-set */
6197 div_u64_rem(stripe_nr, num_stripes, &rot);
6199 /* Fill in the logical address of each stripe */
6200 tmp = stripe_nr * data_stripes;
6201 for (i = 0; i < data_stripes; i++)
6202 bbio->raid_map[(i+rot) % num_stripes] =
6203 em->start + (tmp + i) * map->stripe_len;
6205 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
6206 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
6207 bbio->raid_map[(i+rot+1) % num_stripes] =
6210 sort_parity_stripes(bbio, num_stripes);
6213 if (need_full_stripe(op))
6214 max_errors = btrfs_chunk_max_errors(map);
6216 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6217 need_full_stripe(op)) {
6218 handle_ops_on_dev_replace(op, &bbio, dev_replace, &num_stripes,
6223 bbio->map_type = map->type;
6224 bbio->num_stripes = num_stripes;
6225 bbio->max_errors = max_errors;
6226 bbio->mirror_num = mirror_num;
6229 * this is the case that REQ_READ && dev_replace_is_ongoing &&
6230 * mirror_num == num_stripes + 1 && dev_replace target drive is
6231 * available as a mirror
6233 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
6234 WARN_ON(num_stripes > 1);
6235 bbio->stripes[0].dev = dev_replace->tgtdev;
6236 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
6237 bbio->mirror_num = map->num_stripes + 1;
6240 if (dev_replace_is_ongoing) {
6241 lockdep_assert_held(&dev_replace->rwsem);
6242 /* Unlock and let waiting writers proceed */
6243 up_read(&dev_replace->rwsem);
6245 free_extent_map(em);
6249 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6250 u64 logical, u64 *length,
6251 struct btrfs_bio **bbio_ret, int mirror_num)
6253 if (op == BTRFS_MAP_DISCARD)
6254 return __btrfs_map_block_for_discard(fs_info, logical,
6257 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret,
6261 /* For Scrub/replace */
6262 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6263 u64 logical, u64 *length,
6264 struct btrfs_bio **bbio_ret)
6266 return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, 0, 1);
6269 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio)
6271 bio->bi_private = bbio->private;
6272 bio->bi_end_io = bbio->end_io;
6275 btrfs_put_bbio(bbio);
6278 static void btrfs_end_bio(struct bio *bio)
6280 struct btrfs_bio *bbio = bio->bi_private;
6281 int is_orig_bio = 0;
6283 if (bio->bi_status) {
6284 atomic_inc(&bbio->error);
6285 if (bio->bi_status == BLK_STS_IOERR ||
6286 bio->bi_status == BLK_STS_TARGET) {
6287 struct btrfs_device *dev = btrfs_io_bio(bio)->device;
6290 if (bio_op(bio) == REQ_OP_WRITE)
6291 btrfs_dev_stat_inc_and_print(dev,
6292 BTRFS_DEV_STAT_WRITE_ERRS);
6293 else if (!(bio->bi_opf & REQ_RAHEAD))
6294 btrfs_dev_stat_inc_and_print(dev,
6295 BTRFS_DEV_STAT_READ_ERRS);
6296 if (bio->bi_opf & REQ_PREFLUSH)
6297 btrfs_dev_stat_inc_and_print(dev,
6298 BTRFS_DEV_STAT_FLUSH_ERRS);
6302 if (bio == bbio->orig_bio)
6305 btrfs_bio_counter_dec(bbio->fs_info);
6307 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6310 bio = bbio->orig_bio;
6313 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6314 /* only send an error to the higher layers if it is
6315 * beyond the tolerance of the btrfs bio
6317 if (atomic_read(&bbio->error) > bbio->max_errors) {
6318 bio->bi_status = BLK_STS_IOERR;
6321 * this bio is actually up to date, we didn't
6322 * go over the max number of errors
6324 bio->bi_status = BLK_STS_OK;
6327 btrfs_end_bbio(bbio, bio);
6328 } else if (!is_orig_bio) {
6333 static void submit_stripe_bio(struct btrfs_bio *bbio, struct bio *bio,
6334 u64 physical, struct btrfs_device *dev)
6336 struct btrfs_fs_info *fs_info = bbio->fs_info;
6338 bio->bi_private = bbio;
6339 btrfs_io_bio(bio)->device = dev;
6340 bio->bi_end_io = btrfs_end_bio;
6341 bio->bi_iter.bi_sector = physical >> 9;
6342 btrfs_debug_in_rcu(fs_info,
6343 "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u",
6344 bio_op(bio), bio->bi_opf, (u64)bio->bi_iter.bi_sector,
6345 (unsigned long)dev->bdev->bd_dev, rcu_str_deref(dev->name),
6346 dev->devid, bio->bi_iter.bi_size);
6347 bio_set_dev(bio, dev->bdev);
6349 btrfs_bio_counter_inc_noblocked(fs_info);
6351 btrfsic_submit_bio(bio);
6354 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
6356 atomic_inc(&bbio->error);
6357 if (atomic_dec_and_test(&bbio->stripes_pending)) {
6358 /* Should be the original bio. */
6359 WARN_ON(bio != bbio->orig_bio);
6361 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
6362 bio->bi_iter.bi_sector = logical >> 9;
6363 if (atomic_read(&bbio->error) > bbio->max_errors)
6364 bio->bi_status = BLK_STS_IOERR;
6366 bio->bi_status = BLK_STS_OK;
6367 btrfs_end_bbio(bbio, bio);
6371 blk_status_t btrfs_map_bio(struct btrfs_fs_info *fs_info, struct bio *bio,
6374 struct btrfs_device *dev;
6375 struct bio *first_bio = bio;
6376 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
6382 struct btrfs_bio *bbio = NULL;
6384 length = bio->bi_iter.bi_size;
6385 map_length = length;
6387 btrfs_bio_counter_inc_blocked(fs_info);
6388 ret = __btrfs_map_block(fs_info, btrfs_op(bio), logical,
6389 &map_length, &bbio, mirror_num, 1);
6391 btrfs_bio_counter_dec(fs_info);
6392 return errno_to_blk_status(ret);
6395 total_devs = bbio->num_stripes;
6396 bbio->orig_bio = first_bio;
6397 bbio->private = first_bio->bi_private;
6398 bbio->end_io = first_bio->bi_end_io;
6399 bbio->fs_info = fs_info;
6400 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6402 if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
6403 ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) {
6404 /* In this case, map_length has been set to the length of
6405 a single stripe; not the whole write */
6406 if (bio_op(bio) == REQ_OP_WRITE) {
6407 ret = raid56_parity_write(fs_info, bio, bbio,
6410 ret = raid56_parity_recover(fs_info, bio, bbio,
6411 map_length, mirror_num, 1);
6414 btrfs_bio_counter_dec(fs_info);
6415 return errno_to_blk_status(ret);
6418 if (map_length < length) {
6420 "mapping failed logical %llu bio len %llu len %llu",
6421 logical, length, map_length);
6425 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6426 dev = bbio->stripes[dev_nr].dev;
6427 if (!dev || !dev->bdev || test_bit(BTRFS_DEV_STATE_MISSING,
6429 (bio_op(first_bio) == REQ_OP_WRITE &&
6430 !test_bit(BTRFS_DEV_STATE_WRITEABLE, &dev->dev_state))) {
6431 bbio_error(bbio, first_bio, logical);
6435 if (dev_nr < total_devs - 1)
6436 bio = btrfs_bio_clone(first_bio);
6440 submit_stripe_bio(bbio, bio, bbio->stripes[dev_nr].physical, dev);
6442 btrfs_bio_counter_dec(fs_info);
6447 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6450 * If devid and uuid are both specified, the match must be exact, otherwise
6451 * only devid is used.
6453 * If @seed is true, traverse through the seed devices.
6455 struct btrfs_device *btrfs_find_device(struct btrfs_fs_devices *fs_devices,
6456 u64 devid, u8 *uuid, u8 *fsid,
6459 struct btrfs_device *device;
6460 struct btrfs_fs_devices *seed_devs;
6462 if (!fsid || !memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6463 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6464 if (device->devid == devid &&
6465 (!uuid || memcmp(device->uuid, uuid,
6466 BTRFS_UUID_SIZE) == 0))
6471 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6473 !memcmp(seed_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE)) {
6474 list_for_each_entry(device, &seed_devs->devices,
6476 if (device->devid == devid &&
6477 (!uuid || memcmp(device->uuid, uuid,
6478 BTRFS_UUID_SIZE) == 0))
6487 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6488 u64 devid, u8 *dev_uuid)
6490 struct btrfs_device *device;
6491 unsigned int nofs_flag;
6494 * We call this under the chunk_mutex, so we want to use NOFS for this
6495 * allocation, however we don't want to change btrfs_alloc_device() to
6496 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6499 nofs_flag = memalloc_nofs_save();
6500 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6501 memalloc_nofs_restore(nofs_flag);
6505 list_add(&device->dev_list, &fs_devices->devices);
6506 device->fs_devices = fs_devices;
6507 fs_devices->num_devices++;
6509 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6510 fs_devices->missing_devices++;
6516 * btrfs_alloc_device - allocate struct btrfs_device
6517 * @fs_info: used only for generating a new devid, can be NULL if
6518 * devid is provided (i.e. @devid != NULL).
6519 * @devid: a pointer to devid for this device. If NULL a new devid
6521 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6524 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6525 * on error. Returned struct is not linked onto any lists and must be
6526 * destroyed with btrfs_free_device.
6528 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6532 struct btrfs_device *dev;
6535 if (WARN_ON(!devid && !fs_info))
6536 return ERR_PTR(-EINVAL);
6538 dev = __alloc_device(fs_info);
6547 ret = find_next_devid(fs_info, &tmp);
6549 btrfs_free_device(dev);
6550 return ERR_PTR(ret);
6556 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6558 generate_random_uuid(dev->uuid);
6563 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6564 u64 devid, u8 *uuid, bool error)
6567 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6570 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6574 static u64 calc_stripe_length(u64 type, u64 chunk_len, int num_stripes)
6576 int index = btrfs_bg_flags_to_raid_index(type);
6577 int ncopies = btrfs_raid_array[index].ncopies;
6578 const int nparity = btrfs_raid_array[index].nparity;
6582 data_stripes = num_stripes - nparity;
6584 data_stripes = num_stripes / ncopies;
6586 return div_u64(chunk_len, data_stripes);
6589 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6590 struct btrfs_chunk *chunk)
6592 struct btrfs_fs_info *fs_info = leaf->fs_info;
6593 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6594 struct map_lookup *map;
6595 struct extent_map *em;
6599 u8 uuid[BTRFS_UUID_SIZE];
6604 logical = key->offset;
6605 length = btrfs_chunk_length(leaf, chunk);
6606 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6609 * Only need to verify chunk item if we're reading from sys chunk array,
6610 * as chunk item in tree block is already verified by tree-checker.
6612 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6613 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6618 read_lock(&map_tree->lock);
6619 em = lookup_extent_mapping(map_tree, logical, 1);
6620 read_unlock(&map_tree->lock);
6622 /* already mapped? */
6623 if (em && em->start <= logical && em->start + em->len > logical) {
6624 free_extent_map(em);
6627 free_extent_map(em);
6630 em = alloc_extent_map();
6633 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6635 free_extent_map(em);
6639 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6640 em->map_lookup = map;
6641 em->start = logical;
6644 em->block_start = 0;
6645 em->block_len = em->len;
6647 map->num_stripes = num_stripes;
6648 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6649 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6650 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6651 map->type = btrfs_chunk_type(leaf, chunk);
6652 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6653 map->verified_stripes = 0;
6654 em->orig_block_len = calc_stripe_length(map->type, em->len,
6656 for (i = 0; i < num_stripes; i++) {
6657 map->stripes[i].physical =
6658 btrfs_stripe_offset_nr(leaf, chunk, i);
6659 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6660 read_extent_buffer(leaf, uuid, (unsigned long)
6661 btrfs_stripe_dev_uuid_nr(chunk, i),
6663 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices,
6664 devid, uuid, NULL, true);
6665 if (!map->stripes[i].dev &&
6666 !btrfs_test_opt(fs_info, DEGRADED)) {
6667 free_extent_map(em);
6668 btrfs_report_missing_device(fs_info, devid, uuid, true);
6671 if (!map->stripes[i].dev) {
6672 map->stripes[i].dev =
6673 add_missing_dev(fs_info->fs_devices, devid,
6675 if (IS_ERR(map->stripes[i].dev)) {
6676 free_extent_map(em);
6678 "failed to init missing dev %llu: %ld",
6679 devid, PTR_ERR(map->stripes[i].dev));
6680 return PTR_ERR(map->stripes[i].dev);
6682 btrfs_report_missing_device(fs_info, devid, uuid, false);
6684 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6685 &(map->stripes[i].dev->dev_state));
6689 write_lock(&map_tree->lock);
6690 ret = add_extent_mapping(map_tree, em, 0);
6691 write_unlock(&map_tree->lock);
6694 "failed to add chunk map, start=%llu len=%llu: %d",
6695 em->start, em->len, ret);
6697 free_extent_map(em);
6702 static void fill_device_from_item(struct extent_buffer *leaf,
6703 struct btrfs_dev_item *dev_item,
6704 struct btrfs_device *device)
6708 device->devid = btrfs_device_id(leaf, dev_item);
6709 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6710 device->total_bytes = device->disk_total_bytes;
6711 device->commit_total_bytes = device->disk_total_bytes;
6712 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6713 device->commit_bytes_used = device->bytes_used;
6714 device->type = btrfs_device_type(leaf, dev_item);
6715 device->io_align = btrfs_device_io_align(leaf, dev_item);
6716 device->io_width = btrfs_device_io_width(leaf, dev_item);
6717 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6718 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6719 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6721 ptr = btrfs_device_uuid(dev_item);
6722 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6725 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6728 struct btrfs_fs_devices *fs_devices;
6731 lockdep_assert_held(&uuid_mutex);
6734 /* This will match only for multi-device seed fs */
6735 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
6736 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6740 fs_devices = find_fsid(fsid, NULL);
6742 if (!btrfs_test_opt(fs_info, DEGRADED))
6743 return ERR_PTR(-ENOENT);
6745 fs_devices = alloc_fs_devices(fsid, NULL);
6746 if (IS_ERR(fs_devices))
6749 fs_devices->seeding = true;
6750 fs_devices->opened = 1;
6755 * Upon first call for a seed fs fsid, just create a private copy of the
6756 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
6758 fs_devices = clone_fs_devices(fs_devices);
6759 if (IS_ERR(fs_devices))
6762 ret = open_fs_devices(fs_devices, FMODE_READ, fs_info->bdev_holder);
6764 free_fs_devices(fs_devices);
6765 return ERR_PTR(ret);
6768 if (!fs_devices->seeding) {
6769 close_fs_devices(fs_devices);
6770 free_fs_devices(fs_devices);
6771 return ERR_PTR(-EINVAL);
6774 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
6779 static int read_one_dev(struct extent_buffer *leaf,
6780 struct btrfs_dev_item *dev_item)
6782 struct btrfs_fs_info *fs_info = leaf->fs_info;
6783 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6784 struct btrfs_device *device;
6787 u8 fs_uuid[BTRFS_FSID_SIZE];
6788 u8 dev_uuid[BTRFS_UUID_SIZE];
6790 devid = btrfs_device_id(leaf, dev_item);
6791 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6793 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6796 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
6797 fs_devices = open_seed_devices(fs_info, fs_uuid);
6798 if (IS_ERR(fs_devices))
6799 return PTR_ERR(fs_devices);
6802 device = btrfs_find_device(fs_info->fs_devices, devid, dev_uuid,
6805 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6806 btrfs_report_missing_device(fs_info, devid,
6811 device = add_missing_dev(fs_devices, devid, dev_uuid);
6812 if (IS_ERR(device)) {
6814 "failed to add missing dev %llu: %ld",
6815 devid, PTR_ERR(device));
6816 return PTR_ERR(device);
6818 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6820 if (!device->bdev) {
6821 if (!btrfs_test_opt(fs_info, DEGRADED)) {
6822 btrfs_report_missing_device(fs_info,
6823 devid, dev_uuid, true);
6826 btrfs_report_missing_device(fs_info, devid,
6830 if (!device->bdev &&
6831 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6833 * this happens when a device that was properly setup
6834 * in the device info lists suddenly goes bad.
6835 * device->bdev is NULL, and so we have to set
6836 * device->missing to one here
6838 device->fs_devices->missing_devices++;
6839 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6842 /* Move the device to its own fs_devices */
6843 if (device->fs_devices != fs_devices) {
6844 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6845 &device->dev_state));
6847 list_move(&device->dev_list, &fs_devices->devices);
6848 device->fs_devices->num_devices--;
6849 fs_devices->num_devices++;
6851 device->fs_devices->missing_devices--;
6852 fs_devices->missing_devices++;
6854 device->fs_devices = fs_devices;
6858 if (device->fs_devices != fs_info->fs_devices) {
6859 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6860 if (device->generation !=
6861 btrfs_device_generation(leaf, dev_item))
6865 fill_device_from_item(leaf, dev_item, device);
6866 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
6867 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
6868 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
6869 device->fs_devices->total_rw_bytes += device->total_bytes;
6870 atomic64_add(device->total_bytes - device->bytes_used,
6871 &fs_info->free_chunk_space);
6877 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
6879 struct btrfs_root *root = fs_info->tree_root;
6880 struct btrfs_super_block *super_copy = fs_info->super_copy;
6881 struct extent_buffer *sb;
6882 struct btrfs_disk_key *disk_key;
6883 struct btrfs_chunk *chunk;
6885 unsigned long sb_array_offset;
6892 struct btrfs_key key;
6894 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
6896 * This will create extent buffer of nodesize, superblock size is
6897 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6898 * overallocate but we can keep it as-is, only the first page is used.
6900 sb = btrfs_find_create_tree_block(fs_info, BTRFS_SUPER_INFO_OFFSET);
6903 set_extent_buffer_uptodate(sb);
6904 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6906 * The sb extent buffer is artificial and just used to read the system array.
6907 * set_extent_buffer_uptodate() call does not properly mark all it's
6908 * pages up-to-date when the page is larger: extent does not cover the
6909 * whole page and consequently check_page_uptodate does not find all
6910 * the page's extents up-to-date (the hole beyond sb),
6911 * write_extent_buffer then triggers a WARN_ON.
6913 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6914 * but sb spans only this function. Add an explicit SetPageUptodate call
6915 * to silence the warning eg. on PowerPC 64.
6917 if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE)
6918 SetPageUptodate(sb->pages[0]);
6920 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6921 array_size = btrfs_super_sys_array_size(super_copy);
6923 array_ptr = super_copy->sys_chunk_array;
6924 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6927 while (cur_offset < array_size) {
6928 disk_key = (struct btrfs_disk_key *)array_ptr;
6929 len = sizeof(*disk_key);
6930 if (cur_offset + len > array_size)
6931 goto out_short_read;
6933 btrfs_disk_key_to_cpu(&key, disk_key);
6936 sb_array_offset += len;
6939 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
6941 "unexpected item type %u in sys_array at offset %u",
6942 (u32)key.type, cur_offset);
6947 chunk = (struct btrfs_chunk *)sb_array_offset;
6949 * At least one btrfs_chunk with one stripe must be present,
6950 * exact stripe count check comes afterwards
6952 len = btrfs_chunk_item_size(1);
6953 if (cur_offset + len > array_size)
6954 goto out_short_read;
6956 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6959 "invalid number of stripes %u in sys_array at offset %u",
6960 num_stripes, cur_offset);
6965 type = btrfs_chunk_type(sb, chunk);
6966 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
6968 "invalid chunk type %llu in sys_array at offset %u",
6974 len = btrfs_chunk_item_size(num_stripes);
6975 if (cur_offset + len > array_size)
6976 goto out_short_read;
6978 ret = read_one_chunk(&key, sb, chunk);
6983 sb_array_offset += len;
6986 clear_extent_buffer_uptodate(sb);
6987 free_extent_buffer_stale(sb);
6991 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
6993 clear_extent_buffer_uptodate(sb);
6994 free_extent_buffer_stale(sb);
6999 * Check if all chunks in the fs are OK for read-write degraded mount
7001 * If the @failing_dev is specified, it's accounted as missing.
7003 * Return true if all chunks meet the minimal RW mount requirements.
7004 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7006 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7007 struct btrfs_device *failing_dev)
7009 struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7010 struct extent_map *em;
7014 read_lock(&map_tree->lock);
7015 em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7016 read_unlock(&map_tree->lock);
7017 /* No chunk at all? Return false anyway */
7023 struct map_lookup *map;
7028 map = em->map_lookup;
7030 btrfs_get_num_tolerated_disk_barrier_failures(
7032 for (i = 0; i < map->num_stripes; i++) {
7033 struct btrfs_device *dev = map->stripes[i].dev;
7035 if (!dev || !dev->bdev ||
7036 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7037 dev->last_flush_error)
7039 else if (failing_dev && failing_dev == dev)
7042 if (missing > max_tolerated) {
7045 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7046 em->start, missing, max_tolerated);
7047 free_extent_map(em);
7051 next_start = extent_map_end(em);
7052 free_extent_map(em);
7054 read_lock(&map_tree->lock);
7055 em = lookup_extent_mapping(map_tree, next_start,
7056 (u64)(-1) - next_start);
7057 read_unlock(&map_tree->lock);
7063 static void readahead_tree_node_children(struct extent_buffer *node)
7066 const int nr_items = btrfs_header_nritems(node);
7068 for (i = 0; i < nr_items; i++) {
7071 start = btrfs_node_blockptr(node, i);
7072 readahead_tree_block(node->fs_info, start);
7076 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7078 struct btrfs_root *root = fs_info->chunk_root;
7079 struct btrfs_path *path;
7080 struct extent_buffer *leaf;
7081 struct btrfs_key key;
7082 struct btrfs_key found_key;
7086 u64 last_ra_node = 0;
7088 path = btrfs_alloc_path();
7093 * uuid_mutex is needed only if we are mounting a sprout FS
7094 * otherwise we don't need it.
7096 mutex_lock(&uuid_mutex);
7099 * It is possible for mount and umount to race in such a way that
7100 * we execute this code path, but open_fs_devices failed to clear
7101 * total_rw_bytes. We certainly want it cleared before reading the
7102 * device items, so clear it here.
7104 fs_info->fs_devices->total_rw_bytes = 0;
7107 * Read all device items, and then all the chunk items. All
7108 * device items are found before any chunk item (their object id
7109 * is smaller than the lowest possible object id for a chunk
7110 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7112 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7115 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7119 struct extent_buffer *node;
7121 leaf = path->nodes[0];
7122 slot = path->slots[0];
7123 if (slot >= btrfs_header_nritems(leaf)) {
7124 ret = btrfs_next_leaf(root, path);
7132 * The nodes on level 1 are not locked but we don't need to do
7133 * that during mount time as nothing else can access the tree
7135 node = path->nodes[1];
7137 if (last_ra_node != node->start) {
7138 readahead_tree_node_children(node);
7139 last_ra_node = node->start;
7142 btrfs_item_key_to_cpu(leaf, &found_key, slot);
7143 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7144 struct btrfs_dev_item *dev_item;
7145 dev_item = btrfs_item_ptr(leaf, slot,
7146 struct btrfs_dev_item);
7147 ret = read_one_dev(leaf, dev_item);
7151 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7152 struct btrfs_chunk *chunk;
7153 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7154 mutex_lock(&fs_info->chunk_mutex);
7155 ret = read_one_chunk(&found_key, leaf, chunk);
7156 mutex_unlock(&fs_info->chunk_mutex);
7164 * After loading chunk tree, we've got all device information,
7165 * do another round of validation checks.
7167 if (total_dev != fs_info->fs_devices->total_devices) {
7169 "super_num_devices %llu mismatch with num_devices %llu found here",
7170 btrfs_super_num_devices(fs_info->super_copy),
7175 if (btrfs_super_total_bytes(fs_info->super_copy) <
7176 fs_info->fs_devices->total_rw_bytes) {
7178 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7179 btrfs_super_total_bytes(fs_info->super_copy),
7180 fs_info->fs_devices->total_rw_bytes);
7186 mutex_unlock(&uuid_mutex);
7188 btrfs_free_path(path);
7192 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7194 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7195 struct btrfs_device *device;
7197 fs_devices->fs_info = fs_info;
7199 mutex_lock(&fs_devices->device_list_mutex);
7200 list_for_each_entry(device, &fs_devices->devices, dev_list)
7201 device->fs_info = fs_info;
7203 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7204 list_for_each_entry(device, &seed_devs->devices, dev_list)
7205 device->fs_info = fs_info;
7207 seed_devs->fs_info = fs_info;
7209 mutex_unlock(&fs_devices->device_list_mutex);
7212 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7213 const struct btrfs_dev_stats_item *ptr,
7218 read_extent_buffer(eb, &val,
7219 offsetof(struct btrfs_dev_stats_item, values) +
7220 ((unsigned long)ptr) + (index * sizeof(u64)),
7225 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7226 struct btrfs_dev_stats_item *ptr,
7229 write_extent_buffer(eb, &val,
7230 offsetof(struct btrfs_dev_stats_item, values) +
7231 ((unsigned long)ptr) + (index * sizeof(u64)),
7235 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7236 struct btrfs_path *path)
7238 struct btrfs_dev_stats_item *ptr;
7239 struct extent_buffer *eb;
7240 struct btrfs_key key;
7244 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7245 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7246 key.offset = device->devid;
7247 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7249 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7250 btrfs_dev_stat_set(device, i, 0);
7251 device->dev_stats_valid = 1;
7252 btrfs_release_path(path);
7253 return ret < 0 ? ret : 0;
7255 slot = path->slots[0];
7256 eb = path->nodes[0];
7257 item_size = btrfs_item_size_nr(eb, slot);
7259 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7261 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7262 if (item_size >= (1 + i) * sizeof(__le64))
7263 btrfs_dev_stat_set(device, i,
7264 btrfs_dev_stats_value(eb, ptr, i));
7266 btrfs_dev_stat_set(device, i, 0);
7269 device->dev_stats_valid = 1;
7270 btrfs_dev_stat_print_on_load(device);
7271 btrfs_release_path(path);
7276 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7278 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7279 struct btrfs_device *device;
7280 struct btrfs_path *path = NULL;
7283 path = btrfs_alloc_path();
7287 mutex_lock(&fs_devices->device_list_mutex);
7288 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7289 ret = btrfs_device_init_dev_stats(device, path);
7293 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7294 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7295 ret = btrfs_device_init_dev_stats(device, path);
7301 mutex_unlock(&fs_devices->device_list_mutex);
7303 btrfs_free_path(path);
7307 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7308 struct btrfs_device *device)
7310 struct btrfs_fs_info *fs_info = trans->fs_info;
7311 struct btrfs_root *dev_root = fs_info->dev_root;
7312 struct btrfs_path *path;
7313 struct btrfs_key key;
7314 struct extent_buffer *eb;
7315 struct btrfs_dev_stats_item *ptr;
7319 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7320 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7321 key.offset = device->devid;
7323 path = btrfs_alloc_path();
7326 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7328 btrfs_warn_in_rcu(fs_info,
7329 "error %d while searching for dev_stats item for device %s",
7330 ret, rcu_str_deref(device->name));
7335 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7336 /* need to delete old one and insert a new one */
7337 ret = btrfs_del_item(trans, dev_root, path);
7339 btrfs_warn_in_rcu(fs_info,
7340 "delete too small dev_stats item for device %s failed %d",
7341 rcu_str_deref(device->name), ret);
7348 /* need to insert a new item */
7349 btrfs_release_path(path);
7350 ret = btrfs_insert_empty_item(trans, dev_root, path,
7351 &key, sizeof(*ptr));
7353 btrfs_warn_in_rcu(fs_info,
7354 "insert dev_stats item for device %s failed %d",
7355 rcu_str_deref(device->name), ret);
7360 eb = path->nodes[0];
7361 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7362 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7363 btrfs_set_dev_stats_value(eb, ptr, i,
7364 btrfs_dev_stat_read(device, i));
7365 btrfs_mark_buffer_dirty(eb);
7368 btrfs_free_path(path);
7373 * called from commit_transaction. Writes all changed device stats to disk.
7375 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7377 struct btrfs_fs_info *fs_info = trans->fs_info;
7378 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7379 struct btrfs_device *device;
7383 mutex_lock(&fs_devices->device_list_mutex);
7384 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7385 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7386 if (!device->dev_stats_valid || stats_cnt == 0)
7391 * There is a LOAD-LOAD control dependency between the value of
7392 * dev_stats_ccnt and updating the on-disk values which requires
7393 * reading the in-memory counters. Such control dependencies
7394 * require explicit read memory barriers.
7396 * This memory barriers pairs with smp_mb__before_atomic in
7397 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7398 * barrier implied by atomic_xchg in
7399 * btrfs_dev_stats_read_and_reset
7403 ret = update_dev_stat_item(trans, device);
7405 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7407 mutex_unlock(&fs_devices->device_list_mutex);
7412 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7414 btrfs_dev_stat_inc(dev, index);
7415 btrfs_dev_stat_print_on_error(dev);
7418 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
7420 if (!dev->dev_stats_valid)
7422 btrfs_err_rl_in_rcu(dev->fs_info,
7423 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7424 rcu_str_deref(dev->name),
7425 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7426 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7427 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7428 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7429 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7432 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7436 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7437 if (btrfs_dev_stat_read(dev, i) != 0)
7439 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7440 return; /* all values == 0, suppress message */
7442 btrfs_info_in_rcu(dev->fs_info,
7443 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7444 rcu_str_deref(dev->name),
7445 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7446 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7447 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7448 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7449 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7452 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7453 struct btrfs_ioctl_get_dev_stats *stats)
7455 struct btrfs_device *dev;
7456 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7459 mutex_lock(&fs_devices->device_list_mutex);
7460 dev = btrfs_find_device(fs_info->fs_devices, stats->devid, NULL, NULL,
7462 mutex_unlock(&fs_devices->device_list_mutex);
7465 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7467 } else if (!dev->dev_stats_valid) {
7468 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7470 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7471 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7472 if (stats->nr_items > i)
7474 btrfs_dev_stat_read_and_reset(dev, i);
7476 btrfs_dev_stat_set(dev, i, 0);
7478 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7479 current->comm, task_pid_nr(current));
7481 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7482 if (stats->nr_items > i)
7483 stats->values[i] = btrfs_dev_stat_read(dev, i);
7485 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7486 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7491 * Update the size and bytes used for each device where it changed. This is
7492 * delayed since we would otherwise get errors while writing out the
7495 * Must be invoked during transaction commit.
7497 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7499 struct btrfs_device *curr, *next;
7501 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7503 if (list_empty(&trans->dev_update_list))
7507 * We don't need the device_list_mutex here. This list is owned by the
7508 * transaction and the transaction must complete before the device is
7511 mutex_lock(&trans->fs_info->chunk_mutex);
7512 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7514 list_del_init(&curr->post_commit_list);
7515 curr->commit_total_bytes = curr->disk_total_bytes;
7516 curr->commit_bytes_used = curr->bytes_used;
7518 mutex_unlock(&trans->fs_info->chunk_mutex);
7522 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7524 int btrfs_bg_type_to_factor(u64 flags)
7526 const int index = btrfs_bg_flags_to_raid_index(flags);
7528 return btrfs_raid_array[index].ncopies;
7533 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7534 u64 chunk_offset, u64 devid,
7535 u64 physical_offset, u64 physical_len)
7537 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7538 struct extent_map *em;
7539 struct map_lookup *map;
7540 struct btrfs_device *dev;
7546 read_lock(&em_tree->lock);
7547 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7548 read_unlock(&em_tree->lock);
7552 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7553 physical_offset, devid);
7558 map = em->map_lookup;
7559 stripe_len = calc_stripe_length(map->type, em->len, map->num_stripes);
7560 if (physical_len != stripe_len) {
7562 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7563 physical_offset, devid, em->start, physical_len,
7569 for (i = 0; i < map->num_stripes; i++) {
7570 if (map->stripes[i].dev->devid == devid &&
7571 map->stripes[i].physical == physical_offset) {
7573 if (map->verified_stripes >= map->num_stripes) {
7575 "too many dev extents for chunk %llu found",
7580 map->verified_stripes++;
7586 "dev extent physical offset %llu devid %llu has no corresponding chunk",
7587 physical_offset, devid);
7591 /* Make sure no dev extent is beyond device bondary */
7592 dev = btrfs_find_device(fs_info->fs_devices, devid, NULL, NULL, true);
7594 btrfs_err(fs_info, "failed to find devid %llu", devid);
7599 /* It's possible this device is a dummy for seed device */
7600 if (dev->disk_total_bytes == 0) {
7601 struct btrfs_fs_devices *devs;
7603 devs = list_first_entry(&fs_info->fs_devices->seed_list,
7604 struct btrfs_fs_devices, seed_list);
7605 dev = btrfs_find_device(devs, devid, NULL, NULL, false);
7607 btrfs_err(fs_info, "failed to find seed devid %llu",
7614 if (physical_offset + physical_len > dev->disk_total_bytes) {
7616 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7617 devid, physical_offset, physical_len,
7618 dev->disk_total_bytes);
7623 free_extent_map(em);
7627 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7629 struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7630 struct extent_map *em;
7631 struct rb_node *node;
7634 read_lock(&em_tree->lock);
7635 for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
7636 em = rb_entry(node, struct extent_map, rb_node);
7637 if (em->map_lookup->num_stripes !=
7638 em->map_lookup->verified_stripes) {
7640 "chunk %llu has missing dev extent, have %d expect %d",
7641 em->start, em->map_lookup->verified_stripes,
7642 em->map_lookup->num_stripes);
7648 read_unlock(&em_tree->lock);
7653 * Ensure that all dev extents are mapped to correct chunk, otherwise
7654 * later chunk allocation/free would cause unexpected behavior.
7656 * NOTE: This will iterate through the whole device tree, which should be of
7657 * the same size level as the chunk tree. This slightly increases mount time.
7659 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
7661 struct btrfs_path *path;
7662 struct btrfs_root *root = fs_info->dev_root;
7663 struct btrfs_key key;
7665 u64 prev_dev_ext_end = 0;
7669 key.type = BTRFS_DEV_EXTENT_KEY;
7672 path = btrfs_alloc_path();
7676 path->reada = READA_FORWARD;
7677 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7681 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7682 ret = btrfs_next_item(root, path);
7685 /* No dev extents at all? Not good */
7692 struct extent_buffer *leaf = path->nodes[0];
7693 struct btrfs_dev_extent *dext;
7694 int slot = path->slots[0];
7696 u64 physical_offset;
7700 btrfs_item_key_to_cpu(leaf, &key, slot);
7701 if (key.type != BTRFS_DEV_EXTENT_KEY)
7703 devid = key.objectid;
7704 physical_offset = key.offset;
7706 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
7707 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
7708 physical_len = btrfs_dev_extent_length(leaf, dext);
7710 /* Check if this dev extent overlaps with the previous one */
7711 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
7713 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
7714 devid, physical_offset, prev_dev_ext_end);
7719 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
7720 physical_offset, physical_len);
7724 prev_dev_ext_end = physical_offset + physical_len;
7726 ret = btrfs_next_item(root, path);
7735 /* Ensure all chunks have corresponding dev extents */
7736 ret = verify_chunk_dev_extent_mapping(fs_info);
7738 btrfs_free_path(path);
7743 * Check whether the given block group or device is pinned by any inode being
7744 * used as a swapfile.
7746 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
7748 struct btrfs_swapfile_pin *sp;
7749 struct rb_node *node;
7751 spin_lock(&fs_info->swapfile_pins_lock);
7752 node = fs_info->swapfile_pins.rb_node;
7754 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
7756 node = node->rb_left;
7757 else if (ptr > sp->ptr)
7758 node = node->rb_right;
7762 spin_unlock(&fs_info->swapfile_pins_lock);
7763 return node != NULL;